Index: projects/runtime-coverage/UPDATING =================================================================== --- projects/runtime-coverage/UPDATING (revision 325498) +++ projects/runtime-coverage/UPDATING (revision 325499) @@ -1,2011 +1,2018 @@ Updating Information for FreeBSD current users. This file is maintained and copyrighted by M. Warner Losh . See end of file for further details. For commonly done items, please see the COMMON ITEMS: section later in the file. These instructions assume that you basically know what you are doing. If not, then please consult the FreeBSD handbook: https://www.freebsd.org/doc/en_US.ISO8859-1/books/handbook/current-stable.html Items affecting the ports and packages system can be found in /usr/ports/UPDATING. Please read that file before running portupgrade. NOTE: FreeBSD has switched from gcc to clang. If you have trouble bootstrapping from older versions of FreeBSD, try WITHOUT_CLANG and WITH_GCC to bootstrap to the tip of head, and then rebuild without this option. The bootstrap process from older version of current across the gcc/clang cutover is a bit fragile. NOTE TO PEOPLE WHO THINK THAT FreeBSD 12.x IS SLOW: FreeBSD 12.x has many debugging features turned on, in both the kernel and userland. These features attempt to detect incorrect use of system primitives, and encourage loud failure through extra sanity checking and fail stop semantics. They also substantially impact system performance. If you want to do performance measurement, benchmarking, and optimization, you'll want to turn them off. This includes various WITNESS- related kernel options, INVARIANTS, malloc debugging flags in userland, and various verbose features in the kernel. Many developers choose to disable these features on build machines to maximize performance. (To completely disable malloc debugging, define MALLOC_PRODUCTION in /etc/make.conf, or to merely disable the most expensive debugging functionality run "ln -s 'abort:false,junk:false' /etc/malloc.conf".) ****************************** SPECIAL WARNING: ****************************** Due to a bug in some versions of clang that's very hard to workaround in the upgrade process, to upgrade to -current you must first upgrade either stable/9 after r286035 or stable/10 after r286033 (including 10.3-RELEASE) or current after r286007 (including stable/11 and 11.0-RELEASE). These revisions post-date the 10.2 and 9.3 releases, so you'll need to take the unusual step of upgrading to the tip of the stable branch before moving to 11 or -current via a source upgrade. stable/11 and 11.0-RELEASE have working newer compiler. This differs from the historical situation where one could upgrade from anywhere on the last couple of stable branches, so be careful. If you're running a hybrid system on 9.x or 10.x with an updated clang compiler or are using an supported external toolchain, the build system will allow the upgrade. Otherwise it will print a reminder. ****************************** SPECIAL WARNING: ****************************** +20171106: + The naive and non-compliant support of posix_fallocate(2) in ZFS + has been removed as of r325320. The system call now returns EINVAL + when used on a ZFS file. Although the new behavior complies with the + standard, some consumers are not prepared to cope with it. + One known victim is lld prior to r325420. + 20171102: Building in a FreeBSD src checkout will automatically create object directories now rather than store files in the current directory if 'make obj' was not ran. Calling 'make obj' is no longer necesarry. This feature can be disabled by setting WITHOUT_AUTO_OBJ=yes in /etc/src-env.conf (not /etc/src.conf), or passing the option in the environment. 20171101: The default MAKEOBJDIR has changed from /usr/obj/ for native builds, and /usr/obj// for cross-builds, to a unified /usr/obj//. This behavior can be changed to the old format by setting WITHOUT_UNIFIED_OBJDIR=yes in /etc/src-env.conf, the environment, or with -DWITHOUT_UNIFIED_OBJDIR when building. The UNIFIED_OBJDIR option is a transitional feature that will be removed for 12.0 release; please migrate to the new format for any tools by looking up the OBJDIR used by 'make -V .OBJDIR' means rather than hardcoding paths. 20171028: The native-xtools target no longer installs the files by default to the OBJDIR. Use the native-xtools-install target with a DESTDIR to install to ${DESTDIR}/${NXTP} where NXTP defaults to /nxb-bin. 20171021: As part of the boot loader infrastructure cleanup, LOADER_*_SUPPORT options are changing from controlling the build if defined / undefined to controlling the build with explicit 'yes' or 'no' values. They will shift to WITH/WITHOUT options to match other options in the system. 20171010: libstand has turned into a private library for sys/boot use only. It is no longer supported as a public interface outside of sys/boot. 20171005: The arm port has split armv6 into armv6 and armv7. armv7 is now a valid TARGET_ARCH/MACHINE_ARCH setting. If you have an armv7 system and are running a kernel from before r324363, you will need to add MACHINE_ARCH=armv7 to 'make buildworld' to do a native build. 20171003: When building multiple kernels using KERNCONF, non-existent KERNCONF files will produce an error and buildkernel will fail. Previously missing KERNCONF files silently failed giving no indication as to why, only to subsequently discover during installkernel that the desired kernel was never built in the first place. 20170912: The default serial number format for CTL LUNs has changed. This will affect users who use /dev/diskid/* device nodes, or whose FibreChannel or iSCSI clients care about their LUNs' serial numbers. Users who require serial number stability should hardcode serial numbers in /etc/ctl.conf . 20170912: For 32-bit arm compiled for hard-float support, soft-floating point binaries now always get their shared libraries from LD_SOFT_LIBRARY_PATH (in the past, this was only used if /usr/libsoft also existed). Only users with a hard-float ld.so, but soft-float everything else should be affected. 20170826: The geli password typed at boot is now hidden. To restore the previous behavior, see geli(8) for configuration options. 20170825: Move PMTUD blackhole counters to TCPSTATS and remove them from bare sysctl values. Minor nit, but requires a rebuild of both world/kernel to complete. 20170814: "make check" behavior (made in ^/head@r295380) has been changed to execute from a limited sandbox, as opposed to executing from ${TESTSDIR}. Behavioral changes: - The "beforecheck" and "aftercheck" targets are now specified. - ${CHECKDIR} (added in commit noted above) has been removed. - Legacy behavior can be enabled by setting WITHOUT_MAKE_CHECK_USE_SANDBOX in src.conf(5) or the environment. If the limited sandbox mode is enabled, "make check" will execute "make distribution", then install, execute the tests, and clean up the sandbox if successful. The "make distribution" and "make install" targets are typically run as root to set appropriate permissions and ownership at installation time. The end-user should set "WITH_INSTALL_AS_USER" in src.conf(5) or the environment if executing "make check" with limited sandbox mode using an unprivileged user. 20170808: Since the switch to GPT disk labels, fsck for UFS/FFS has been unable to automatically find alternate superblocks. As of r322297, the information needed to find alternate superblocks has been moved to the end of the area reserved for the boot block. Filesystems created with a newfs of this vintage or later will create the recovery information. If you have a filesystem created prior to this change and wish to have a recovery block created for your filesystem, you can do so by running fsck in forground mode (i.e., do not use the -p or -y options). As it starts, fsck will ask ``SAVE DATA TO FIND ALTERNATE SUPERBLOCKS'' to which you should answer yes. 20170728: As of r321665, an NFSv4 server configuration that services Kerberos mounts or clients that do not support the uid/gid in owner/owner_group string capability, must explicitly enable the nfsuserd daemon by adding nfsuserd_enable="YES" to the machine's /etc/rc.conf file. 20170722: Clang, llvm, lldb, compiler-rt and libc++ have been upgraded to 5.0.0. Please see the 20141231 entry below for information about prerequisites and upgrading, if you are not already using clang 3.5.0 or higher. 20170701: WITHOUT_RCMDS is now the default. Set WITH_RCMDS if you need the r-commands (rlogin, rsh, etc.) to be built with the base system. 20170625: The FreeBSD/powerpc platform now uses a 64-bit type for time_t. This is a very major ABI incompatible change, so users of FreeBSD/powerpc must be careful when performing source upgrades. It is best to run 'make installworld' from an alternate root system, either a live CD/memory stick, or a temporary root partition. Additionally, all ports must be recompiled. powerpc64 is largely unaffected, except in the case of 32-bit compatibility. All 32-bit binaries will be affected. 20170623: Forward compatibility for the "ino64" project have been committed. This will allow most new binaries to run on older kernels in a limited fashion. This prevents many of the common foot-shooting actions in the upgrade as well as the limited ability to roll back the kernel across the ino64 upgrade. Complicated use cases may not work properly, though enough simpler ones work to allow recovery in most situations. 20170620: Switch back to the BSDL dtc (Device Tree Compiler). Set WITH_GPL_DTC if you require the GPL compiler. 20170618: The internal ABI used for communication between the NFS kernel modules was changed by r320085, so __FreeBSD_version was bumped to ensure all the NFS related modules are updated together. 20170617: The ABI of struct event was changed by extending the data member to 64bit and adding ext fields. For upgrade, same precautions as for the entry 20170523 "ino64" must be followed. 20170531: The GNU roff toolchain has been removed from base. To render manpages which are not supported by mandoc(1), man(1) can fallback on GNU roff from ports (and recommends to install it). To render roff(7) documents, consider using GNU roff from ports or the heirloom doctools roff toolchain from ports via pkg install groff or via pkg install heirloom-doctools. 20170524: The ath(4) and ath_hal(4) modules now build piecemeal to allow for smaller runtime footprint builds. This is useful for embedded systems which only require one chipset support. If you load it as a module, make sure this is in /boot/loader.conf: if_ath_load="YES" This will load the HAL, all chip/RF backends and if_ath_pci. If you have if_ath_pci in /boot/loader.conf, ensure it is after if_ath or it will not load any HAL chipset support. If you want to selectively load things (eg on ye cheape ARM/MIPS platforms where RAM is at a premium) you should: * load ath_hal * load the chip modules in question * load ath_rate, ath_dfs * load ath_main * load if_ath_pci and/or if_ath_ahb depending upon your particular bus bind type - this is where probe/attach is done. For further comments/feedback, poke adrian@ . 20170523: The "ino64" 64-bit inode project has been committed, which extends a number of types to 64 bits. Upgrading in place requires care and adherence to the documented upgrade procedure. If using a custom kernel configuration ensure that the COMPAT_FREEBSD11 option is included (as during the upgrade the system will be running the ino64 kernel with the existing world). For the safest in-place upgrade begin by removing previous build artifacts via "rm -rf /usr/obj/*". Then, carefully follow the full procedure documented below under the heading "To rebuild everything and install it on the current system." Specifically, a reboot is required after installing the new kernel before installing world. 20170424: The NATM framework including the en(4), fatm(4), hatm(4), and patm(4) devices has been removed. Consumers should plan a migration before the end-of-life date for FreeBSD 11. 20170420: GNU diff has been replaced by a BSD licensed diff. Some features of GNU diff has not been implemented, if those are needed a newer version of GNU diff is available via the diffutils package under the gdiff name. 20170413: As of r316810 for ipfilter, keep frags is no longer assumed when keep state is specified in a rule. r316810 aligns ipfilter with documentation in man pages separating keep frags from keep state. This allows keep state to be specified without forcing keep frags and allows keep frags to be specified independently of keep state. To maintain previous behaviour, also specify keep frags with keep state (as documented in ipf.conf.5). 20170407: arm64 builds now use the base system LLD 4.0.0 linker by default, instead of requiring that the aarch64-binutils port or package be installed. To continue using aarch64-binutils, set CROSS_BINUTILS_PREFIX=/usr/local/aarch64-freebsd/bin . 20170405: The UDP optimization in entry 20160818 that added the sysctl net.inet.udp.require_l2_bcast has been reverted. L2 broadcast packets will no longer be treated as L3 broadcast packets. 20170331: Binds and sends to the loopback addresses, IPv6 and IPv4, will now use any explicitly assigned loopback address available in the jail instead of using the first assigned address of the jail. 20170329: The ctl.ko module no longer implements the iSCSI target frontend: cfiscsi.ko does instead. If building cfiscsi.ko as a kernel module, the module can be loaded via one of the following methods: - `cfiscsi_load="YES"` in loader.conf(5). - Add `cfiscsi` to `$kld_list` in rc.conf(5). - ctladm(8)/ctld(8), when compiled with iSCSI support (`WITH_ISCSI=yes` in src.conf(5)) Please see cfiscsi(4) for more details. 20170316: The mmcsd.ko module now additionally depends on geom_flashmap.ko. Also, mmc.ko and mmcsd.ko need to be a matching pair built from the same source (previously, the dependency of mmcsd.ko on mmc.ko was missing, but mmcsd.ko now will refuse to load if it is incompatible with mmc.ko). 20170315: The syntax of ipfw(8) named states was changed to avoid ambiguity. If you have used named states in the firewall rules, you need to modify them after installworld and before rebooting. Now named states must be prefixed with colon. 20170311: The old drm (sys/dev/drm/) drivers for i915 and radeon have been removed as the userland we provide cannot use them. The KMS version (sys/dev/drm2) supports the same hardware. 20170302: Clang, llvm, lldb, compiler-rt and libc++ have been upgraded to 4.0.0. Please see the 20141231 entry below for information about prerequisites and upgrading, if you are not already using clang 3.5.0 or higher. 20170221: The code that provides support for ZFS .zfs/ directory functionality has been reimplemented. It's not possible now to create a snapshot by mkdir under .zfs/snapshot/. That should be the only user visible change. 20170216: EISA bus support has been removed. The WITH_EISA option is no longer valid. 20170215: MCA bus support has been removed. 20170127: The WITH_LLD_AS_LD / WITHOUT_LLD_AS_LD build knobs have been renamed WITH_LLD_IS_LD / WITHOUT_LLD_IS_LD, for consistency with CLANG_IS_CC. 20170112: The EM_MULTIQUEUE kernel configuration option is deprecated now that the em(4) driver conforms to iflib specifications. 20170109: The igb(4), em(4) and lem(4) ethernet drivers are now implemented via IFLIB. If you have a custom kernel configuration that excludes em(4) but you use igb(4), you need to re-add em(4) to your custom configuration. 20161217: Clang, llvm, lldb, compiler-rt and libc++ have been upgraded to 3.9.1. Please see the 20141231 entry below for information about prerequisites and upgrading, if you are not already using clang 3.5.0 or higher. 20161124: Clang, llvm, lldb, compiler-rt and libc++ have been upgraded to 3.9.0. Please see the 20141231 entry below for information about prerequisites and upgrading, if you are not already using clang 3.5.0 or higher. 20161119: The layout of the pmap structure has changed for powerpc to put the pmap statistics at the front for all CPU variations. libkvm(3) and all tools that link against it need to be recompiled. 20161030: isl(4) and cyapa(4) drivers now require a new driver, chromebook_platform(4), to work properly on Chromebook-class hardware. On other types of hardware the drivers may need to be configured using device hints. Please see the corresponding manual pages for details. 20161017: The urtwn(4) driver was merged into rtwn(4) and now consists of rtwn(4) main module + rtwn_usb(4) and rtwn_pci(4) bus-specific parts. Also, firmware for RTL8188CE was renamed due to possible name conflict (rtwnrtl8192cU(B) -> rtwnrtl8192cE(B)) 20161015: GNU rcs has been removed from base. It is available as packages: - rcs: Latest GPLv3 GNU rcs version. - rcs57: Copy of the latest version of GNU rcs (GPLv2) before it was removed from base. 20161008: Use of the cc_cdg, cc_chd, cc_hd, or cc_vegas congestion control modules now requires that the kernel configuration contain the TCP_HHOOK option. (This option is included in the GENERIC kernel.) 20161003: The WITHOUT_ELFCOPY_AS_OBJCOPY src.conf(5) knob has been retired. ELF Tool Chain's elfcopy is always installed as /usr/bin/objcopy. 20160924: Relocatable object files with the extension of .So have been renamed to use an extension of .pico instead. The purpose of this change is to avoid a name clash with shared libraries on case-insensitive file systems. On those file systems, foo.So is the same file as foo.so. 20160918: GNU rcs has been turned off by default. It can (temporarily) be built again by adding WITH_RCS knob in src.conf. Otherwise, GNU rcs is available from packages: - rcs: Latest GPLv3 GNU rcs version. - rcs57: Copy of the latest version of GNU rcs (GPLv2) from base. 20160918: The backup_uses_rcs functionality has been removed from rc.subr. 20160908: The queue(3) debugging macro, QUEUE_MACRO_DEBUG, has been split into two separate components, QUEUE_MACRO_DEBUG_TRACE and QUEUE_MACRO_DEBUG_TRASH. Define both for the original QUEUE_MACRO_DEBUG behavior. 20160824: r304787 changed some ioctl interfaces between the iSCSI userspace programs and the kernel. ctladm, ctld, iscsictl, and iscsid must be rebuilt to work with new kernels. __FreeBSD_version has been bumped to 1200005. 20160818: The UDP receive code has been updated to only treat incoming UDP packets that were addressed to an L2 broadcast address as L3 broadcast packets. It is not expected that this will affect any standards-conforming UDP application. The new behaviour can be disabled by setting the sysctl net.inet.udp.require_l2_bcast to 0. 20160818: Remove the openbsd_poll system call. __FreeBSD_version has been bumped because of this. 20160622: The libc stub for the pipe(2) system call has been replaced with a wrapper that calls the pipe2(2) system call and the pipe(2) system call is now only implemented by the kernels that include "options COMPAT_FREEBSD10" in their config file (this is the default). Users should ensure that this option is enabled in their kernel or upgrade userspace to r302092 before upgrading their kernel. 20160527: CAM will now strip leading spaces from SCSI disks' serial numbers. This will affect users who create UFS filesystems on SCSI disks using those disk's diskid device nodes. For example, if /etc/fstab previously contained a line like "/dev/diskid/DISK-%20%20%20%20%20%20%20ABCDEFG0123456", you should change it to "/dev/diskid/DISK-ABCDEFG0123456". Users of geom transforms like gmirror may also be affected. ZFS users should generally be fine. 20160523: The bitstring(3) API has been updated with new functionality and improved performance. But it is binary-incompatible with the old API. Objects built with the new headers may not be linked against objects built with the old headers. 20160520: The brk and sbrk functions have been removed from libc on arm64. Binutils from ports has been updated to not link to these functions and should be updated to the latest version before installing a new libc. 20160517: The armv6 port now defaults to hard float ABI. Limited support for running both hardfloat and soft float on the same system is available using the libraries installed with -DWITH_LIBSOFT. This has only been tested as an upgrade path for installworld and packages may fail or need manual intervention to run. New packages will be needed. To update an existing self-hosted armv6hf system, you must add TARGET_ARCH=armv6 on the make command line for both the build and the install steps. 20160510: Kernel modules compiled outside of a kernel build now default to installing to /boot/modules instead of /boot/kernel. Many kernel modules built this way (such as those in ports) already overrode KMODDIR explicitly to install into /boot/modules. However, manually building and installing a module from /sys/modules will now install to /boot/modules instead of /boot/kernel. 20160414: The CAM I/O scheduler has been committed to the kernel. There should be no user visible impact. This does enable NCQ Trim on ada SSDs. While the list of known rogues that claim support for this but actually corrupt data is believed to be complete, be on the lookout for data corruption. The known rogue list is believed to be complete: o Crucial MX100, M550 drives with MU01 firmware. o Micron M510 and M550 drives with MU01 firmware. o Micron M500 prior to MU07 firmware o Samsung 830, 840, and 850 all firmwares o FCCT M500 all firmwares Crucial has firmware http://www.crucial.com/usa/en/support-ssd-firmware with working NCQ TRIM. For Micron branded drives, see your sales rep for updated firmware. Black listed drives will work correctly because these drives work correctly so long as no NCQ TRIMs are sent to them. Given this list is the same as found in Linux, it's believed there are no other rogues in the market place. All other models from the above vendors work. To be safe, if you are at all concerned, you can quirk each of your drives to prevent NCQ from being sent by setting: kern.cam.ada.X.quirks="0x2" in loader.conf. If the drive requires the 4k sector quirk, set the quirks entry to 0x3. 20160330: The FAST_DEPEND build option has been removed and its functionality is now the one true way. The old mkdep(1) style of 'make depend' has been removed. See 20160311 for further details. 20160317: Resource range types have grown from unsigned long to uintmax_t. All drivers, and anything using libdevinfo, need to be recompiled. 20160311: WITH_FAST_DEPEND is now enabled by default for in-tree and out-of-tree builds. It no longer runs mkdep(1) during 'make depend', and the 'make depend' stage can safely be skipped now as it is auto ran when building 'make all' and will generate all SRCS and DPSRCS before building anything else. Dependencies are gathered at compile time with -MF flags kept in separate .depend files per object file. Users should run 'make cleandepend' once if using -DNO_CLEAN to clean out older stale .depend files. 20160306: On amd64, clang 3.8.0 can now insert sections of type AMD64_UNWIND into kernel modules. Therefore, if you load any kernel modules at boot time, please install the boot loaders after you install the kernel, but before rebooting, e.g.: make buildworld make kernel KERNCONF=YOUR_KERNEL_HERE make -C sys/boot install Then follow the usual steps, described in the General Notes section, below. 20160305: Clang, llvm, lldb and compiler-rt have been upgraded to 3.8.0. Please see the 20141231 entry below for information about prerequisites and upgrading, if you are not already using clang 3.5.0 or higher. 20160301: The AIO subsystem is now a standard part of the kernel. The VFS_AIO kernel option and aio.ko kernel module have been removed. Due to stability concerns, asynchronous I/O requests are only permitted on sockets and raw disks by default. To enable asynchronous I/O requests on all file types, set the vfs.aio.enable_unsafe sysctl to a non-zero value. 20160226: The ELF object manipulation tool objcopy is now provided by the ELF Tool Chain project rather than by GNU binutils. It should be a drop-in replacement, with the addition of arm64 support. The (temporary) src.conf knob WITHOUT_ELFCOPY_AS_OBJCOPY knob may be set to obtain the GNU version if necessary. 20160129: Building ZFS pools on top of zvols is prohibited by default. That feature has never worked safely; it's always been prone to deadlocks. Using a zvol as the backing store for a VM guest's virtual disk will still work, even if the guest is using ZFS. Legacy behavior can be restored by setting vfs.zfs.vol.recursive=1. 20160119: The NONE and HPN patches has been removed from OpenSSH. They are still available in the security/openssh-portable port. 20160113: With the addition of ypldap(8), a new _ypldap user is now required during installworld. "mergemaster -p" can be used to add the user prior to installworld, as documented in the handbook. 20151216: The tftp loader (pxeboot) now uses the option root-path directive. As a consequence it no longer looks for a pxeboot.4th file on the tftp server. Instead it uses the regular /boot infrastructure as with the other loaders. 20151211: The code to start recording plug and play data into the modules has been committed. While the old tools will properly build a new kernel, a number of warnings about "unknown metadata record 4" will be produced for an older kldxref. To avoid such warnings, make sure to rebuild the kernel toolchain (or world). Make sure that you have r292078 or later when trying to build 292077 or later before rebuilding. 20151207: Debug data files are now built by default with 'make buildworld' and installed with 'make installworld'. This facilitates debugging but requires more disk space both during the build and for the installed world. Debug files may be disabled by setting WITHOUT_DEBUG_FILES=yes in src.conf(5). 20151130: r291527 changed the internal interface between the nfsd.ko and nfscommon.ko modules. As such, they must both be upgraded to-gether. __FreeBSD_version has been bumped because of this. 20151108: Add support for unicode collation strings leads to a change of order of files listed by ls(1) for example. To get back to the old behaviour, set LC_COLLATE environment variable to "C". Databases administrators will need to reindex their databases given collation results will be different. Due to a bug in install(1) it is recommended to remove the ancient locales before running make installworld. rm -rf /usr/share/locale/* 20151030: The OpenSSL has been upgraded to 1.0.2d. Any binaries requiring libcrypto.so.7 or libssl.so.7 must be recompiled. 20151020: Qlogic 24xx/25xx firmware images were updated from 5.5.0 to 7.3.0. Kernel modules isp_2400_multi and isp_2500_multi were removed and should be replaced with isp_2400 and isp_2500 modules respectively. 20151017: The build previously allowed using 'make -n' to not recurse into sub-directories while showing what commands would be executed, and 'make -n -n' to recursively show commands. Now 'make -n' will recurse and 'make -N' will not. 20151012: If you specify SENDMAIL_MC or SENDMAIL_CF in make.conf, mergemaster and etcupdate will now use this file. A custom sendmail.cf is now updated via this mechanism rather than via installworld. If you had excluded sendmail.cf in mergemaster.rc or etcupdate.conf, you may want to remove the exclusion or change it to "always install". /etc/mail/sendmail.cf is now managed the same way regardless of whether SENDMAIL_MC/SENDMAIL_CF is used. If you are not using SENDMAIL_MC/SENDMAIL_CF there should be no change in behavior. 20151011: Compatibility shims for legacy ATA device names have been removed. It includes ATA_STATIC_ID kernel option, kern.cam.ada.legacy_aliases and kern.geom.raid.legacy_aliases loader tunables, kern.devalias.* environment variables, /dev/ad* and /dev/ar* symbolic links. 20151006: Clang, llvm, lldb, compiler-rt and libc++ have been upgraded to 3.7.0. Please see the 20141231 entry below for information about prerequisites and upgrading, if you are not already using clang 3.5.0 or higher. 20150924: Kernel debug files have been moved to /usr/lib/debug/boot/kernel/, and renamed from .symbols to .debug. This reduces the size requirements on the boot partition or file system and provides consistency with userland debug files. When using the supported kernel installation method the /usr/lib/debug/boot/kernel directory will be renamed (to kernel.old) as is done with /boot/kernel. Developers wishing to maintain the historical behavior of installing debug files in /boot/kernel/ can set KERN_DEBUGDIR="" in src.conf(5). 20150827: The wireless drivers had undergone changes that remove the 'parent interface' from the ifconfig -l output. The rc.d network scripts used to check presence of a parent interface in the list, so old scripts would fail to start wireless networking. Thus, etcupdate(3) or mergemaster(8) run is required after kernel update, to update your rc.d scripts in /etc. 20150827: pf no longer supports 'scrub fragment crop' or 'scrub fragment drop-ovl' These configurations are now automatically interpreted as 'scrub fragment reassemble'. 20150817: Kernel-loadable modules for the random(4) device are back. To use them, the kernel must have device random options RANDOM_LOADABLE kldload(8) can then be used to load random_fortuna.ko or random_yarrow.ko. Please note that due to the indirect function calls that the loadable modules need to provide, the build-in variants will be slightly more efficient. The random(4) kernel option RANDOM_DUMMY has been retired due to unpopularity. It was not all that useful anyway. 20150813: The WITHOUT_ELFTOOLCHAIN_TOOLS src.conf(5) knob has been retired. Control over building the ELF Tool Chain tools is now provided by the WITHOUT_TOOLCHAIN knob. 20150810: The polarity of Pulse Per Second (PPS) capture events with the uart(4) driver has been corrected. Prior to this change the PPS "assert" event corresponded to the trailing edge of a positive PPS pulse and the "clear" event was the leading edge of the next pulse. As the width of a PPS pulse in a typical GPS receiver is on the order of 1 millisecond, most users will not notice any significant difference with this change. Anyone who has compensated for the historical polarity reversal by configuring a negative offset equal to the pulse width will need to remove that workaround. 20150809: The default group assigned to /dev/dri entries has been changed from 'wheel' to 'video' with the id of '44'. If you want to have access to the dri devices please add yourself to the video group with: # pw groupmod video -m $USER 20150806: The menu.rc and loader.rc files will now be replaced during upgrades. Please migrate local changes to menu.rc.local and loader.rc.local instead. 20150805: GNU Binutils versions of addr2line, c++filt, nm, readelf, size, strings and strip have been removed. The src.conf(5) knob WITHOUT_ELFTOOLCHAIN_TOOLS no longer provides the binutils tools. 20150728: As ZFS requires more kernel stack pages than is the default on some architectures e.g. i386, it now warns if KSTACK_PAGES is less than ZFS_MIN_KSTACK_PAGES (which is 4 at the time of writing). Please consider using 'options KSTACK_PAGES=X' where X is greater than or equal to ZFS_MIN_KSTACK_PAGES i.e. 4 in such configurations. 20150706: sendmail has been updated to 8.15.2. Starting with FreeBSD 11.0 and sendmail 8.15, sendmail uses uncompressed IPv6 addresses by default, i.e., they will not contain "::". For example, instead of ::1, it will be 0:0:0:0:0:0:0:1. This permits a zero subnet to have a more specific match, such as different map entries for IPv6:0:0 vs IPv6:0. This change requires that configuration data (including maps, files, classes, custom ruleset, etc.) must use the same format, so make certain such configuration data is upgrading. As a very simple check search for patterns like 'IPv6:[0-9a-fA-F:]*::' and 'IPv6::'. To return to the old behavior, set the m4 option confUSE_COMPRESSED_IPV6_ADDRESSES or the cf option UseCompressedIPv6Addresses. 20150630: The default kernel entropy-processing algorithm is now Fortuna, replacing Yarrow. Assuming you have 'device random' in your kernel config file, the configurations allow a kernel option to override this default. You may choose *ONE* of: options RANDOM_YARROW # Legacy /dev/random algorithm. options RANDOM_DUMMY # Blocking-only driver. If you have neither, you get Fortuna. For most people, read no further, Fortuna will give a /dev/random that works like it always used to, and the difference will be irrelevant. If you remove 'device random', you get *NO* kernel-processed entropy at all. This may be acceptable to folks building embedded systems, but has complications. Carry on reading, and it is assumed you know what you need. *PLEASE* read random(4) and random(9) if you are in the habit of tweaking kernel configs, and/or if you are a member of the embedded community, wanting specific and not-usual behaviour from your security subsystems. NOTE!! If you use RANDOM_DUMMY and/or have no 'device random', you will NOT have a functioning /dev/random, and many cryptographic features will not work, including SSH. You may also find strange behaviour from the random(3) set of library functions, in particular sranddev(3), srandomdev(3) and arc4random(3). The reason for this is that the KERN_ARND sysctl only returns entropy if it thinks it has some to share, and with RANDOM_DUMMY or no 'device random' this will never happen. 20150623: An additional fix for the issue described in the 20150614 sendmail entry below has been been committed in revision 284717. 20150616: FreeBSD's old make (fmake) has been removed from the system. It is available as the devel/fmake port or via pkg install fmake. 20150615: The fix for the issue described in the 20150614 sendmail entry below has been been committed in revision 284436. The work around described in that entry is no longer needed unless the default setting is overridden by a confDH_PARAMETERS configuration setting of '5' or pointing to a 512 bit DH parameter file. 20150614: ALLOW_DEPRECATED_ATF_TOOLS/ATFFILE support has been removed from atf.test.mk (included from bsd.test.mk). Please upgrade devel/atf and devel/kyua to version 0.20+ and adjust any calling code to work with Kyuafile and kyua. 20150614: The import of openssl to address the FreeBSD-SA-15:10.openssl security advisory includes a change which rejects handshakes with DH parameters below 768 bits. sendmail releases prior to 8.15.2 (not yet released), defaulted to a 512 bit DH parameter setting for client connections. To work around this interoperability, sendmail can be configured to use a 2048 bit DH parameter by: 1. Edit /etc/mail/`hostname`.mc 2. If a setting for confDH_PARAMETERS does not exist or exists and is set to a string beginning with '5', replace it with '2'. 3. If a setting for confDH_PARAMETERS exists and is set to a file path, create a new file with: openssl dhparam -out /path/to/file 2048 4. Rebuild the .cf file: cd /etc/mail/; make; make install 5. Restart sendmail: cd /etc/mail/; make restart A sendmail patch is coming, at which time this file will be updated. 20150604: Generation of legacy formatted entries have been disabled by default in pwd_mkdb(8), as all base system consumers of the legacy formatted entries were converted to use the new format by default when the new, machine independent format have been added and supported since FreeBSD 5.x. Please see the pwd_mkdb(8) manual page for further details. 20150525: Clang and llvm have been upgraded to 3.6.1 release. Please see the 20141231 entry below for information about prerequisites and upgrading, if you are not already using 3.5.0 or higher. 20150521: TI platform code switched to using vendor DTS files and this update may break existing systems running on Beaglebone, Beaglebone Black, and Pandaboard: - dtb files should be regenerated/reinstalled. Filenames are the same but content is different now - GPIO addressing was changed, now each GPIO bank (32 pins per bank) has its own /dev/gpiocX device, e.g. pin 121 on /dev/gpioc0 in old addressing scheme is now pin 25 on /dev/gpioc3. - Pandaboard: /etc/ttys should be updated, serial console device is now /dev/ttyu2, not /dev/ttyu0 20150501: soelim(1) from gnu/usr.bin/groff has been replaced by usr.bin/soelim. If you need the GNU extension from groff soelim(1), install groff from package: pkg install groff, or via ports: textproc/groff. 20150423: chmod, chflags, chown and chgrp now affect symlinks in -R mode as defined in symlink(7); previously symlinks were silently ignored. 20150415: The const qualifier has been removed from iconv(3) to comply with POSIX. The ports tree is aware of this from r384038 onwards. 20150416: Libraries specified by LIBADD in Makefiles must have a corresponding DPADD_ variable to ensure correct dependencies. This is now enforced in src.libnames.mk. 20150324: From legacy ata(4) driver was removed support for SATA controllers supported by more functional drivers ahci(4), siis(4) and mvs(4). Kernel modules ataahci and ataadaptec were removed completely, replaced by ahci and mvs modules respectively. 20150315: Clang, llvm and lldb have been upgraded to 3.6.0 release. Please see the 20141231 entry below for information about prerequisites and upgrading, if you are not already using 3.5.0 or higher. 20150307: The 32-bit PowerPC kernel has been changed to a position-independent executable. This can only be booted with a version of loader(8) newer than January 31, 2015, so make sure to update both world and kernel before rebooting. 20150217: If you are running a -CURRENT kernel since r273872 (Oct 30th, 2014), but before r278950, the RNG was not seeded properly. Immediately upgrade the kernel to r278950 or later and regenerate any keys (e.g. ssh keys or openssl keys) that were generated w/ a kernel from that range. This does not affect programs that directly used /dev/random or /dev/urandom. All userland uses of arc4random(3) are affected. 20150210: The autofs(4) ABI was changed in order to restore binary compatibility with 10.1-RELEASE. The automountd(8) daemon needs to be rebuilt to work with the new kernel. 20150131: The powerpc64 kernel has been changed to a position-independent executable. This can only be booted with a new version of loader(8), so make sure to update both world and kernel before rebooting. 20150118: Clang and llvm have been upgraded to 3.5.1 release. This is a bugfix only release, no new features have been added. Please see the 20141231 entry below for information about prerequisites and upgrading, if you are not already using 3.5.0. 20150107: ELF tools addr2line, elfcopy (strip), nm, size, and strings are now taken from the ELF Tool Chain project rather than GNU binutils. They should be drop-in replacements, with the addition of arm64 support. The WITHOUT_ELFTOOLCHAIN_TOOLS= knob may be used to obtain the binutils tools, if necessary. See 20150805 for updated information. 20150105: The default Unbound configuration now enables remote control using a local socket. Users who have already enabled the local_unbound service should regenerate their configuration by running "service local_unbound setup" as root. 20150102: The GNU texinfo and GNU info pages have been removed. To be able to view GNU info pages please install texinfo from ports. 20141231: Clang, llvm and lldb have been upgraded to 3.5.0 release. As of this release, a prerequisite for building clang, llvm and lldb is a C++11 capable compiler and C++11 standard library. This means that to be able to successfully build the cross-tools stage of buildworld, with clang as the bootstrap compiler, your system compiler or cross compiler should either be clang 3.3 or later, or gcc 4.8 or later, and your system C++ library should be libc++, or libdstdc++ from gcc 4.8 or later. On any standard FreeBSD 10.x or 11.x installation, where clang and libc++ are on by default (that is, on x86 or arm), this should work out of the box. On 9.x installations where clang is enabled by default, e.g. on x86 and powerpc, libc++ will not be enabled by default, so libc++ should be built (with clang) and installed first. If both clang and libc++ are missing, build clang first, then use it to build libc++. On 8.x and earlier installations, upgrade to 9.x first, and then follow the instructions for 9.x above. Sparc64 and mips users are unaffected, as they still use gcc 4.2.1 by default, and do not build clang. Many embedded systems are resource constrained, and will not be able to build clang in a reasonable time, or in some cases at all. In those cases, cross building bootable systems on amd64 is a workaround. This new version of clang introduces a number of new warnings, of which the following are most likely to appear: -Wabsolute-value This warns in two cases, for both C and C++: * When the code is trying to take the absolute value of an unsigned quantity, which is effectively a no-op, and almost never what was intended. The code should be fixed, if at all possible. If you are sure that the unsigned quantity can be safely cast to signed, without loss of information or undefined behavior, you can add an explicit cast, or disable the warning. * When the code is trying to take an absolute value, but the called abs() variant is for the wrong type, which can lead to truncation. If you want to disable the warning instead of fixing the code, please make sure that truncation will not occur, or it might lead to unwanted side-effects. -Wtautological-undefined-compare and -Wundefined-bool-conversion These warn when C++ code is trying to compare 'this' against NULL, while 'this' should never be NULL in well-defined C++ code. However, there is some legacy (pre C++11) code out there, which actively abuses this feature, which was less strictly defined in previous C++ versions. Squid and openjdk do this, for example. The warning can be turned off for C++98 and earlier, but compiling the code in C++11 mode might result in unexpected behavior; for example, the parts of the program that are unreachable could be optimized away. 20141222: The old NFS client and server (kernel options NFSCLIENT, NFSSERVER) kernel sources have been removed. The .h files remain, since some utilities include them. This will need to be fixed later. If "mount -t oldnfs ..." is attempted, it will fail. If the "-o" option on mountd(8), nfsd(8) or nfsstat(1) is used, the utilities will report errors. 20141121: The handling of LOCAL_LIB_DIRS has been altered to skip addition of directories to top level SUBDIR variable when their parent directory is included in LOCAL_DIRS. Users with build systems with such hierarchies and without SUBDIR entries in the parent directory Makefiles should add them or add the directories to LOCAL_DIRS. 20141109: faith(4) and faithd(8) have been removed from the base system. Faith has been obsolete for a very long time. 20141104: vt(4), the new console driver, is enabled by default. It brings support for Unicode and double-width characters, as well as support for UEFI and integration with the KMS kernel video drivers. You may need to update your console settings in /etc/rc.conf, most probably the keymap. During boot, /etc/rc.d/syscons will indicate what you need to do. vt(4) still has issues and lacks some features compared to syscons(4). See the wiki for up-to-date information: https://wiki.freebsd.org/Newcons If you want to keep using syscons(4), you can do so by adding the following line to /boot/loader.conf: kern.vty=sc 20141102: pjdfstest has been integrated into kyua as an opt-in test suite. Please see share/doc/pjdfstest/README for more details on how to execute it. 20141009: gperf has been removed from the base system for architectures that use clang. Ports that require gperf will obtain it from the devel/gperf port. 20140923: pjdfstest has been moved from tools/regression/pjdfstest to contrib/pjdfstest . 20140922: At svn r271982, The default linux compat kernel ABI has been adjusted to 2.6.18 in support of the linux-c6 compat ports infrastructure update. If you wish to continue using the linux-f10 compat ports, add compat.linux.osrelease=2.6.16 to your local sysctl.conf. Users are encouraged to update their linux-compat packages to linux-c6 during their next update cycle. 20140729: The ofwfb driver, used to provide a graphics console on PowerPC when using vt(4), no longer allows mmap() of all physical memory. This will prevent Xorg on PowerPC with some ATI graphics cards from initializing properly unless x11-servers/xorg-server is updated to 1.12.4_8 or newer. 20140723: The xdev targets have been converted to using TARGET and TARGET_ARCH instead of XDEV and XDEV_ARCH. 20140719: The default unbound configuration has been modified to address issues with reverse lookups on networks that use private address ranges. If you use the local_unbound service, run "service local_unbound setup" as root to regenerate your configuration, then "service local_unbound reload" to load the new configuration. 20140709: The GNU texinfo and GNU info pages are not built and installed anymore, WITH_INFO knob has been added to allow to built and install them again. UPDATE: see 20150102 entry on texinfo's removal 20140708: The GNU readline library is now an INTERNALLIB - that is, it is statically linked into consumers (GDB and variants) in the base system, and the shared library is no longer installed. The devel/readline port is available for third party software that requires readline. 20140702: The Itanium architecture (ia64) has been removed from the list of known architectures. This is the first step in the removal of the architecture. 20140701: Commit r268115 has added NFSv4.1 server support, merged from projects/nfsv4.1-server. Since this includes changes to the internal interfaces between the NFS related modules, a full build of the kernel and modules will be necessary. __FreeBSD_version has been bumped. 20140629: The WITHOUT_VT_SUPPORT kernel config knob has been renamed WITHOUT_VT. (The other _SUPPORT knobs have a consistent meaning which differs from the behaviour controlled by this knob.) 20140619: Maximal length of the serial number in CTL was increased from 16 to 64 chars, that breaks ABI. All CTL-related tools, such as ctladm and ctld, need to be rebuilt to work with a new kernel. 20140606: The libatf-c and libatf-c++ major versions were downgraded to 0 and 1 respectively to match the upstream numbers. They were out of sync because, when they were originally added to FreeBSD, the upstream versions were not respected. These libraries are private and not yet built by default, so renumbering them should be a non-issue. However, unclean source trees will yield broken test programs once the operator executes "make delete-old-libs" after a "make installworld". Additionally, the atf-sh binary was made private by moving it into /usr/libexec/. Already-built shell test programs will keep the path to the old binary so they will break after "make delete-old" is run. If you are using WITH_TESTS=yes (not the default), wipe the object tree and rebuild from scratch to prevent spurious test failures. This is only needed once: the misnumbered libraries and misplaced binaries have been added to OptionalObsoleteFiles.inc so they will be removed during a clean upgrade. 20140512: Clang and llvm have been upgraded to 3.4.1 release. 20140508: We bogusly installed src.opts.mk in /usr/share/mk. This file should be removed to avoid issues in the future (and has been added to ObsoleteFiles.inc). 20140505: /etc/src.conf now affects only builds of the FreeBSD src tree. In the past, it affected all builds that used the bsd.*.mk files. The old behavior was a bug, but people may have relied upon it. To get this behavior back, you can .include /etc/src.conf from /etc/make.conf (which is still global and isn't changed). This also changes the behavior of incremental builds inside the tree of individual directories. Set MAKESYSPATH to ".../share/mk" to do that. Although this has survived make universe and some upgrade scenarios, other upgrade scenarios may have broken. At least one form of temporary breakage was fixed with MAKESYSPATH settings for buildworld as well... In cases where MAKESYSPATH isn't working with this setting, you'll need to set it to the full path to your tree. One side effect of all this cleaning up is that bsd.compiler.mk is no longer implicitly included by bsd.own.mk. If you wish to use COMPILER_TYPE, you must now explicitly include bsd.compiler.mk as well. 20140430: The lindev device has been removed since /dev/full has been made a standard device. __FreeBSD_version has been bumped. 20140424: The knob WITHOUT_VI was added to the base system, which controls building ex(1), vi(1), etc. Older releases of FreeBSD required ex(1) in order to reorder files share/termcap and didn't build ex(1) as a build tool, so building/installing with WITH_VI is highly advised for build hosts for older releases. This issue has been fixed in stable/9 and stable/10 in r277022 and r276991, respectively. 20140418: The YES_HESIOD knob has been removed. It has been obsolete for a decade. Please move to using WITH_HESIOD instead or your builds will silently lack HESIOD. 20140405: The uart(4) driver has been changed with respect to its handling of the low-level console. Previously the uart(4) driver prevented any process from changing the baudrate or the CLOCAL and HUPCL control flags. By removing the restrictions, operators can make changes to the serial console port without having to reboot. However, when getty(8) is started on the serial device that is associated with the low-level console, a misconfigured terminal line in /etc/ttys will now have a real impact. Before upgrading the kernel, make sure that /etc/ttys has the serial console device configured as 3wire without baudrate to preserve the previous behaviour. E.g: ttyu0 "/usr/libexec/getty 3wire" vt100 on secure 20140306: Support for libwrap (TCP wrappers) in rpcbind was disabled by default to improve performance. To re-enable it, if needed, run rpcbind with command line option -W. 20140226: Switched back to the GPL dtc compiler due to updates in the upstream dts files not being supported by the BSDL dtc compiler. You will need to rebuild your kernel toolchain to pick up the new compiler. Core dumps may result while building dtb files during a kernel build if you fail to do so. Set WITHOUT_GPL_DTC if you require the BSDL compiler. 20140216: Clang and llvm have been upgraded to 3.4 release. 20140216: The nve(4) driver has been removed. Please use the nfe(4) driver for NVIDIA nForce MCP Ethernet adapters instead. 20140212: An ABI incompatibility crept into the libc++ 3.4 import in r261283. This could cause certain C++ applications using shared libraries built against the previous version of libc++ to crash. The incompatibility has now been fixed, but any C++ applications or shared libraries built between r261283 and r261801 should be recompiled. 20140204: OpenSSH will now ignore errors caused by kernel lacking of Capsicum capability mode support. Please note that enabling the feature in kernel is still highly recommended. 20140131: OpenSSH is now built with sandbox support, and will use sandbox as the default privilege separation method. This requires Capsicum capability mode support in kernel. 20140128: The libelf and libdwarf libraries have been updated to newer versions from upstream. Shared library version numbers for these two libraries were bumped. Any ports or binaries requiring these two libraries should be recompiled. __FreeBSD_version is bumped to 1100006. 20140110: If a Makefile in a tests/ directory was auto-generating a Kyuafile instead of providing an explicit one, this would prevent such Makefile from providing its own Kyuafile in the future during NO_CLEAN builds. This has been fixed in the Makefiles but manual intervention is needed to clean an objdir if you use NO_CLEAN: # find /usr/obj -name Kyuafile | xargs rm -f 20131213: The behavior of gss_pseudo_random() for the krb5 mechanism has changed, for applications requesting a longer random string than produced by the underlying enctype's pseudo-random() function. In particular, the random string produced from a session key of enctype aes256-cts-hmac-sha1-96 or aes256-cts-hmac-sha1-96 will be different at the 17th octet and later, after this change. The counter used in the PRF+ construction is now encoded as a big-endian integer in accordance with RFC 4402. __FreeBSD_version is bumped to 1100004. 20131108: The WITHOUT_ATF build knob has been removed and its functionality has been subsumed into the more generic WITHOUT_TESTS. If you were using the former to disable the build of the ATF libraries, you should change your settings to use the latter. 20131025: The default version of mtree is nmtree which is obtained from NetBSD. The output is generally the same, but may vary slightly. If you found you need identical output adding "-F freebsd9" to the command line should do the trick. For the time being, the old mtree is available as fmtree. 20131014: libbsdyml has been renamed to libyaml and moved to /usr/lib/private. This will break ports-mgmt/pkg. Rebuild the port, or upgrade to pkg 1.1.4_8 and verify bsdyml not linked in, before running "make delete-old-libs": # make -C /usr/ports/ports-mgmt/pkg build deinstall install clean or # pkg install pkg; ldd /usr/local/sbin/pkg | grep bsdyml 20131010: The stable/10 branch has been created in subversion from head revision r256279. 20131010: The rc.d/jail script has been updated to support jail(8) configuration file. The "jail__*" rc.conf(5) variables for per-jail configuration are automatically converted to /var/run/jail..conf before the jail(8) utility is invoked. This is transparently backward compatible. See below about some incompatibilities and rc.conf(5) manual page for more details. These variables are now deprecated in favor of jail(8) configuration file. One can use "rc.d/jail config " command to generate a jail(8) configuration file in /var/run/jail..conf without running the jail(8) utility. The default pathname of the configuration file is /etc/jail.conf and can be specified by using $jail_conf or $jail__conf variables. Please note that jail_devfs_ruleset accepts an integer at this moment. Please consider to rewrite the ruleset name with an integer. 20130930: BIND has been removed from the base system. If all you need is a local resolver, simply enable and start the local_unbound service instead. Otherwise, several versions of BIND are available in the ports tree. The dns/bind99 port is one example. With this change, nslookup(1) and dig(1) are no longer in the base system. Users should instead use host(1) and drill(1) which are in the base system. Alternatively, nslookup and dig can be obtained by installing the dns/bind-tools port. 20130916: With the addition of unbound(8), a new unbound user is now required during installworld. "mergemaster -p" can be used to add the user prior to installworld, as documented in the handbook. 20130911: OpenSSH is now built with DNSSEC support, and will by default silently trust signed SSHFP records. This can be controlled with the VerifyHostKeyDNS client configuration setting. DNSSEC support can be disabled entirely with the WITHOUT_LDNS option in src.conf. 20130906: The GNU Compiler Collection and C++ standard library (libstdc++) are no longer built by default on platforms where clang is the system compiler. You can enable them with the WITH_GCC and WITH_GNUCXX options in src.conf. 20130905: The PROCDESC kernel option is now part of the GENERIC kernel configuration and is required for the rwhod(8) to work. If you are using custom kernel configuration, you should include 'options PROCDESC'. 20130905: The API and ABI related to the Capsicum framework was modified in backward incompatible way. The userland libraries and programs have to be recompiled to work with the new kernel. This includes the following libraries and programs, but the whole buildworld is advised: libc, libprocstat, dhclient, tcpdump, hastd, hastctl, kdump, procstat, rwho, rwhod, uniq. 20130903: AES-NI intrinsic support has been added to gcc. The AES-NI module has been updated to use this support. A new gcc is required to build the aesni module on both i386 and amd64. 20130821: The PADLOCK_RNG and RDRAND_RNG kernel options are now devices. Thus "device padlock_rng" and "device rdrand_rng" should be used instead of "options PADLOCK_RNG" & "options RDRAND_RNG". 20130813: WITH_ICONV has been split into two feature sets. WITH_ICONV now enables just the iconv* functionality and is now on by default. WITH_LIBICONV_COMPAT enables the libiconv api and link time compatibility. Set WITHOUT_ICONV to build the old way. If you have been using WITH_ICONV before, you will very likely need to turn on WITH_LIBICONV_COMPAT. 20130806: INVARIANTS option now enables DEBUG for code with OpenSolaris and Illumos origin, including ZFS. If you have INVARIANTS in your kernel configuration, then there is no need to set DEBUG or ZFS_DEBUG explicitly. DEBUG used to enable witness(9) tracking of OpenSolaris (mostly ZFS) locks if WITNESS option was set. Because that generated a lot of witness(9) reports and all of them were believed to be false positives, this is no longer done. New option OPENSOLARIS_WITNESS can be used to achieve the previous behavior. 20130806: Timer values in IPv6 data structures now use time_uptime instead of time_second. Although this is not a user-visible functional change, userland utilities which directly use them---ndp(8), rtadvd(8), and rtsold(8) in the base system---need to be updated to r253970 or later. 20130802: find -delete can now delete the pathnames given as arguments, instead of only files found below them or if the pathname did not contain any slashes. Formerly, the following error message would result: find: -delete: : relative path potentially not safe Deleting the pathnames given as arguments can be prevented without error messages using -mindepth 1 or by changing directory and passing "." as argument to find. This works in the old as well as the new version of find. 20130726: Behavior of devfs rules path matching has been changed. Pattern is now always matched against fully qualified devfs path and slash characters must be explicitly matched by slashes in pattern (FNM_PATHNAME). Rulesets involving devfs subdirectories must be reviewed. 20130716: The default ARM ABI has changed to the ARM EABI. The old ABI is incompatible with the ARM EABI and all programs and modules will need to be rebuilt to work with a new kernel. To keep using the old ABI ensure the WITHOUT_ARM_EABI knob is set. NOTE: Support for the old ABI will be removed in the future and users are advised to upgrade. 20130709: pkg_install has been disconnected from the build if you really need it you should add WITH_PKGTOOLS in your src.conf(5). 20130709: Most of network statistics structures were changed to be able keep 64-bits counters. Thus all tools, that work with networking statistics, must be rebuilt (netstat(1), bsnmpd(1), etc.) 20130618: Fix a bug that allowed a tracing process (e.g. gdb) to write to a memory-mapped file in the traced process's address space even if neither the traced process nor the tracing process had write access to that file. 20130615: CVS has been removed from the base system. An exact copy of the code is available from the devel/cvs port. 20130613: Some people report the following error after the switch to bmake: make: illegal option -- J usage: make [-BPSXeiknpqrstv] [-C directory] [-D variable] ... *** [buildworld] Error code 2 this likely due to an old instance of make in ${MAKEPATH} (${MAKEOBJDIRPREFIX}${.CURDIR}/make.${MACHINE}) which src/Makefile will use that blindly, if it exists, so if you see the above error: rm -rf `make -V MAKEPATH` should resolve it. 20130516: Use bmake by default. Whereas before one could choose to build with bmake via -DWITH_BMAKE one must now use -DWITHOUT_BMAKE to use the old make. The goal is to remove these knobs for 10-RELEASE. It is worth noting that bmake (like gmake) treats the command line as the unit of failure, rather than statements within the command line. Thus '(cd some/where && dosomething)' is safer than 'cd some/where; dosomething'. The '()' allows consistent behavior in parallel build. 20130429: Fix a bug that allows NFS clients to issue READDIR on files. 20130426: The WITHOUT_IDEA option has been removed because the IDEA patent expired. 20130426: The sysctl which controls TRIM support under ZFS has been renamed from vfs.zfs.trim_disable -> vfs.zfs.trim.enabled and has been enabled by default. 20130425: The mergemaster command now uses the default MAKEOBJDIRPREFIX rather than creating it's own in the temporary directory in order allow access to bootstrapped versions of tools such as install and mtree. When upgrading from version of FreeBSD where the install command does not support -l, you will need to install a new mergemaster command if mergemaster -p is required. This can be accomplished with the command (cd src/usr.sbin/mergemaster && make install). 20130404: Legacy ATA stack, disabled and replaced by new CAM-based one since FreeBSD 9.0, completely removed from the sources. Kernel modules atadisk and atapi*, user-level tools atacontrol and burncd are removed. Kernel option `options ATA_CAM` is now permanently enabled and removed. 20130319: SOCK_CLOEXEC and SOCK_NONBLOCK flags have been added to socket(2) and socketpair(2). Software, in particular Kerberos, may automatically detect and use these during building. The resulting binaries will not work on older kernels. 20130308: CTL_DISABLE has also been added to the sparc64 GENERIC (for further information, see the respective 20130304 entry). 20130304: Recent commits to callout(9) changed the size of struct callout, so the KBI is probably heavily disturbed. Also, some functions in callout(9)/sleep(9)/sleepqueue(9)/condvar(9) KPIs were replaced by macros. Every kernel module using it won't load, so rebuild is requested. The ctl device has been re-enabled in GENERIC for i386 and amd64, but does not initialize by default (because of the new CTL_DISABLE option) to save memory. To re-enable it, remove the CTL_DISABLE option from the kernel config file or set kern.cam.ctl.disable=0 in /boot/loader.conf. 20130301: The ctl device has been disabled in GENERIC for i386 and amd64. This was done due to the extra memory being allocated at system initialisation time by the ctl driver which was only used if a CAM target device was created. This makes a FreeBSD system unusable on 128MB or less of RAM. 20130208: A new compression method (lz4) has been merged to -HEAD. Please refer to zpool-features(7) for more information. Please refer to the "ZFS notes" section of this file for information on upgrading boot ZFS pools. 20130129: A BSD-licensed patch(1) variant has been added and is installed as bsdpatch, being the GNU version the default patch. To inverse the logic and use the BSD-licensed one as default, while having the GNU version installed as gnupatch, rebuild and install world with the WITH_BSD_PATCH knob set. 20130121: Due to the use of the new -l option to install(1) during build and install, you must take care not to directly set the INSTALL make variable in your /etc/make.conf, /etc/src.conf, or on the command line. If you wish to use the -C flag for all installs you may be able to add INSTALL+=-C to /etc/make.conf or /etc/src.conf. 20130118: The install(1) option -M has changed meaning and now takes an argument that is a file or path to append logs to. In the unlikely event that -M was the last option on the command line and the command line contained at least two files and a target directory the first file will have logs appended to it. The -M option served little practical purpose in the last decade so its use is expected to be extremely rare. 20121223: After switching to Clang as the default compiler some users of ZFS on i386 systems started to experience stack overflow kernel panics. Please consider using 'options KSTACK_PAGES=4' in such configurations. 20121222: GEOM_LABEL now mangles label names read from file system metadata. Mangling affect labels containing spaces, non-printable characters, '%' or '"'. Device names in /etc/fstab and other places may need to be updated. 20121217: By default, only the 10 most recent kernel dumps will be saved. To restore the previous behaviour (no limit on the number of kernel dumps stored in the dump directory) add the following line to /etc/rc.conf: savecore_flags="" 20121201: With the addition of auditdistd(8), a new auditdistd user is now required during installworld. "mergemaster -p" can be used to add the user prior to installworld, as documented in the handbook. 20121117: The sin6_scope_id member variable in struct sockaddr_in6 is now filled by the kernel before passing the structure to the userland via sysctl or routing socket. This means the KAME-specific embedded scope id in sin6_addr.s6_addr[2] is always cleared in userland application. This behavior can be controlled by net.inet6.ip6.deembed_scopeid. __FreeBSD_version is bumped to 1000025. 20121105: On i386 and amd64 systems WITH_CLANG_IS_CC is now the default. This means that the world and kernel will be compiled with clang and that clang will be installed as /usr/bin/cc, /usr/bin/c++, and /usr/bin/cpp. To disable this behavior and revert to building with gcc, compile with WITHOUT_CLANG_IS_CC. Really old versions of current may need to bootstrap WITHOUT_CLANG first if the clang build fails (its compatibility window doesn't extend to the 9 stable branch point). 20121102: The IPFIREWALL_FORWARD kernel option has been removed. Its functionality now turned on by default. 20121023: The ZERO_COPY_SOCKET kernel option has been removed and split into SOCKET_SEND_COW and SOCKET_RECV_PFLIP. NB: SOCKET_SEND_COW uses the VM page based copy-on-write mechanism which is not safe and may result in kernel crashes. NB: The SOCKET_RECV_PFLIP mechanism is useless as no current driver supports disposeable external page sized mbuf storage. Proper replacements for both zero-copy mechanisms are under consideration and will eventually lead to complete removal of the two kernel options. 20121023: The IPv4 network stack has been converted to network byte order. The following modules need to be recompiled together with kernel: carp(4), divert(4), gif(4), siftr(4), gre(4), pf(4), ipfw(4), ng_ipfw(4), stf(4). 20121022: Support for non-MPSAFE filesystems was removed from VFS. The VFS_VERSION was bumped, all filesystem modules shall be recompiled. 20121018: All the non-MPSAFE filesystems have been disconnected from the build. The full list includes: codafs, hpfs, ntfs, nwfs, portalfs, smbfs, xfs. 20121016: The interface cloning API and ABI has changed. The following modules need to be recompiled together with kernel: ipfw(4), pfsync(4), pflog(4), usb(4), wlan(4), stf(4), vlan(4), disc(4), edsc(4), if_bridge(4), gif(4), tap(4), faith(4), epair(4), enc(4), tun(4), if_lagg(4), gre(4). 20121015: The sdhci driver was split in two parts: sdhci (generic SD Host Controller logic) and sdhci_pci (actual hardware driver). No kernel config modifications are required, but if you load sdhc as a module you must switch to sdhci_pci instead. 20121014: Import the FUSE kernel and userland support into base system. 20121013: The GNU sort(1) program has been removed since the BSD-licensed sort(1) has been the default for quite some time and no serious problems have been reported. The corresponding WITH_GNU_SORT knob has also gone. 20121006: The pfil(9) API/ABI for AF_INET family has been changed. Packet filtering modules: pf(4), ipfw(4), ipfilter(4) need to be recompiled with new kernel. 20121001: The net80211(4) ABI has been changed to allow for improved driver PS-POLL and power-save support. All wireless drivers need to be recompiled to work with the new kernel. 20120913: The random(4) support for the VIA hardware random number generator (`PADLOCK') is no longer enabled unconditionally. Add the padlock_rng device in the custom kernel config if needed. The GENERIC kernels on i386 and amd64 do include the device, so the change only affects the custom kernel configurations. 20120908: The pf(4) packet filter ABI has been changed. pfctl(8) and snmp_pf module need to be recompiled to work with new kernel. 20120828: A new ZFS feature flag "com.delphix:empty_bpobj" has been merged to -HEAD. Pools that have empty_bpobj in active state can not be imported read-write with ZFS implementations that do not support this feature. For more information read the zpool-features(5) manual page. 20120727: The sparc64 ZFS loader has been changed to no longer try to auto- detect ZFS providers based on diskN aliases but now requires these to be explicitly listed in the OFW boot-device environment variable. 20120712: The OpenSSL has been upgraded to 1.0.1c. Any binaries requiring libcrypto.so.6 or libssl.so.6 must be recompiled. Also, there are configuration changes. Make sure to merge /etc/ssl/openssl.cnf. 20120712: The following sysctls and tunables have been renamed for consistency with other variables: kern.cam.da.da_send_ordered -> kern.cam.da.send_ordered kern.cam.ada.ada_send_ordered -> kern.cam.ada.send_ordered 20120628: The sort utility has been replaced with BSD sort. For now, GNU sort is also available as "gnusort" or the default can be set back to GNU sort by setting WITH_GNU_SORT. In this case, BSD sort will be installed as "bsdsort". 20120611: A new version of ZFS (pool version 5000) has been merged to -HEAD. Starting with this version the old system of ZFS pool versioning is superseded by "feature flags". This concept enables forward compatibility against certain future changes in functionality of ZFS pools. The first read-only compatible "feature flag" for ZFS pools is named "com.delphix:async_destroy". For more information read the new zpool-features(5) manual page. Please refer to the "ZFS notes" section of this file for information on upgrading boot ZFS pools. 20120417: The malloc(3) implementation embedded in libc now uses sources imported as contrib/jemalloc. The most disruptive API change is to /etc/malloc.conf. If your system has an old-style /etc/malloc.conf, delete it prior to installworld, and optionally re-create it using the new format after rebooting. See malloc.conf(5) for details (specifically the TUNING section and the "opt.*" entries in the MALLCTL NAMESPACE section). 20120328: Big-endian MIPS TARGET_ARCH values no longer end in "eb". mips64eb is now spelled mips64. mipsn32eb is now spelled mipsn32. mipseb is now spelled mips. This is to aid compatibility with third-party software that expects this naming scheme in uname(3). Little-endian settings are unchanged. If you are updating a big-endian mips64 machine from before this change, you may need to set MACHINE_ARCH=mips64 in your environment before the new build system will recognize your machine. 20120306: Disable by default the option VFS_ALLOW_NONMPSAFE for all supported platforms. 20120229: Now unix domain sockets behave "as expected" on nullfs(5). Previously nullfs(5) did not pass through all behaviours to the underlying layer, as a result if we bound to a socket on the lower layer we could connect only to the lower path; if we bound to the upper layer we could connect only to the upper path. The new behavior is one can connect to both the lower and the upper paths regardless what layer path one binds to. 20120211: The getifaddrs upgrade path broken with 20111215 has been restored. If you have upgraded in between 20111215 and 20120209 you need to recompile libc again with your kernel. You still need to recompile world to be able to configure CARP but this restriction already comes from 20111215. 20120114: The set_rcvar() function has been removed from /etc/rc.subr. All base and ports rc.d scripts have been updated, so if you have a port installed with a script in /usr/local/etc/rc.d you can either hand-edit the rcvar= line, or reinstall the port. An easy way to handle the mass-update of /etc/rc.d: rm /etc/rc.d/* && mergemaster -i 20120109: panic(9) now stops other CPUs in the SMP systems, disables interrupts on the current CPU and prevents other threads from running. This behavior can be reverted using the kern.stop_scheduler_on_panic tunable/sysctl. The new behavior can be incompatible with kern.sync_on_panic. 20111215: The carp(4) facility has been changed significantly. Configuration of the CARP protocol via ifconfig(8) has changed, as well as format of CARP events submitted to devd(8) has changed. See manual pages for more information. The arpbalance feature of carp(4) is currently not supported anymore. Size of struct in_aliasreq, struct in6_aliasreq has changed. User utilities using SIOCAIFADDR, SIOCAIFADDR_IN6, e.g. ifconfig(8), need to be recompiled. 20111122: The acpi_wmi(4) status device /dev/wmistat has been renamed to /dev/wmistat0. 20111108: The option VFS_ALLOW_NONMPSAFE option has been added in order to explicitely support non-MPSAFE filesystems. It is on by default for all supported platform at this present time. 20111101: The broken amd(4) driver has been replaced with esp(4) in the amd64, i386 and pc98 GENERIC kernel configuration files. 20110930: sysinstall has been removed 20110923: The stable/9 branch created in subversion. This corresponds to the RELENG_9 branch in CVS. COMMON ITEMS: General Notes ------------- Avoid using make -j when upgrading. While generally safe, there are sometimes problems using -j to upgrade. If your upgrade fails with -j, please try again without -j. From time to time in the past there have been problems using -j with buildworld and/or installworld. This is especially true when upgrading between "distant" versions (eg one that cross a major release boundary or several minor releases, or when several months have passed on the -current branch). Sometimes, obscure build problems are the result of environment poisoning. This can happen because the make utility reads its environment when searching for values for global variables. To run your build attempts in an "environmental clean room", prefix all make commands with 'env -i '. See the env(1) manual page for more details. When upgrading from one major version to another it is generally best to upgrade to the latest code in the currently installed branch first, then do an upgrade to the new branch. This is the best-tested upgrade path, and has the highest probability of being successful. Please try this approach if you encounter problems with a major version upgrade. Since the stable 4.x branch point, one has generally been able to upgade from anywhere in the most recent stable branch to head / current (or even the last couple of stable branches). See the top of this file when there's an exception. When upgrading a live system, having a root shell around before installing anything can help undo problems. Not having a root shell around can lead to problems if pam has changed too much from your starting point to allow continued authentication after the upgrade. This file should be read as a log of events. When a later event changes information of a prior event, the prior event should not be deleted. Instead, a pointer to the entry with the new information should be placed in the old entry. Readers of this file should also sanity check older entries before relying on them blindly. Authors of new entries should write them with this in mind. ZFS notes --------- When upgrading the boot ZFS pool to a new version, always follow these two steps: 1.) recompile and reinstall the ZFS boot loader and boot block (this is part of "make buildworld" and "make installworld") 2.) update the ZFS boot block on your boot drive The following example updates the ZFS boot block on the first partition (freebsd-boot) of a GPT partitioned drive ada0: "gpart bootcode -p /boot/gptzfsboot -i 1 ada0" Non-boot pools do not need these updates. To build a kernel ----------------- If you are updating from a prior version of FreeBSD (even one just a few days old), you should follow this procedure. It is the most failsafe as it uses a /usr/obj tree with a fresh mini-buildworld, make kernel-toolchain make -DALWAYS_CHECK_MAKE buildkernel KERNCONF=YOUR_KERNEL_HERE make -DALWAYS_CHECK_MAKE installkernel KERNCONF=YOUR_KERNEL_HERE To test a kernel once --------------------- If you just want to boot a kernel once (because you are not sure if it works, or if you want to boot a known bad kernel to provide debugging information) run make installkernel KERNCONF=YOUR_KERNEL_HERE KODIR=/boot/testkernel nextboot -k testkernel To rebuild everything and install it on the current system. ----------------------------------------------------------- # Note: sometimes if you are running current you gotta do more than # is listed here if you are upgrading from a really old current. make buildworld make buildkernel KERNCONF=YOUR_KERNEL_HERE make installkernel KERNCONF=YOUR_KERNEL_HERE [1] [3] mergemaster -Fp [5] make installworld mergemaster -Fi [4] make delete-old [6] To cross-install current onto a separate partition -------------------------------------------------- # In this approach we use a separate partition to hold # current's root, 'usr', and 'var' directories. A partition # holding "/", "/usr" and "/var" should be about 2GB in # size. make buildworld make buildkernel KERNCONF=YOUR_KERNEL_HERE make installworld DESTDIR=${CURRENT_ROOT} -DDB_FROM_SRC make distribution DESTDIR=${CURRENT_ROOT} # if newfs'd make installkernel KERNCONF=YOUR_KERNEL_HERE DESTDIR=${CURRENT_ROOT} cp /etc/fstab ${CURRENT_ROOT}/etc/fstab # if newfs'd To upgrade in-place from stable to current ---------------------------------------------- make buildworld [9] make kernel KERNCONF=YOUR_KERNEL_HERE [8] [1] [3] mergemaster -Fp [5] make installworld mergemaster -Fi [4] make delete-old [6] Make sure that you've read the UPDATING file to understand the tweaks to various things you need. At this point in the life cycle of current, things change often and you are on your own to cope. The defaults can also change, so please read ALL of the UPDATING entries. Also, if you are tracking -current, you must be subscribed to freebsd-current@freebsd.org. Make sure that before you update your sources that you have read and understood all the recent messages there. If in doubt, please track -stable which has much fewer pitfalls. [1] If you have third party modules, such as vmware, you should disable them at this point so they don't crash your system on reboot. [3] From the bootblocks, boot -s, and then do fsck -p mount -u / mount -a cd src adjkerntz -i # if CMOS is wall time Also, when doing a major release upgrade, it is required that you boot into single user mode to do the installworld. [4] Note: This step is non-optional. Failure to do this step can result in a significant reduction in the functionality of the system. Attempting to do it by hand is not recommended and those that pursue this avenue should read this file carefully, as well as the archives of freebsd-current and freebsd-hackers mailing lists for potential gotchas. The -U option is also useful to consider. See mergemaster(8) for more information. [5] Usually this step is a noop. However, from time to time you may need to do this if you get unknown user in the following step. It never hurts to do it all the time. You may need to install a new mergemaster (cd src/usr.sbin/mergemaster && make install) after the buildworld before this step if you last updated from current before 20130425 or from -stable before 20130430. [6] This only deletes old files and directories. Old libraries can be deleted by "make delete-old-libs", but you have to make sure that no program is using those libraries anymore. [8] In order to have a kernel that can run the 4.x binaries needed to do an installworld, you must include the COMPAT_FREEBSD4 option in your kernel. Failure to do so may leave you with a system that is hard to boot to recover. A similar kernel option COMPAT_FREEBSD5 is required to run the 5.x binaries on more recent kernels. And so on for COMPAT_FREEBSD6 and COMPAT_FREEBSD7. Make sure that you merge any new devices from GENERIC since the last time you updated your kernel config file. [9] If CPUTYPE is defined in your /etc/make.conf, make sure to use the "?=" instead of the "=" assignment operator, so that buildworld can override the CPUTYPE if it needs to. MAKEOBJDIRPREFIX must be defined in an environment variable, and not on the command line, or in /etc/make.conf. buildworld will warn if it is improperly defined. FORMAT: This file contains a list, in reverse chronological order, of major breakages in tracking -current. It is not guaranteed to be a complete list of such breakages, and only contains entries since September 23, 2011. If you need to see UPDATING entries from before that date, you will need to fetch an UPDATING file from an older FreeBSD release. Copyright information: Copyright 1998-2009 M. Warner Losh. All Rights Reserved. Redistribution, publication, translation and use, with or without modification, in full or in part, in any form or format of this document are permitted without further permission from the author. THIS DOCUMENT IS PROVIDED BY WARNER LOSH ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL WARNER LOSH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. Contact Warner Losh if you have any questions about your use of this document. $FreeBSD$ Index: projects/runtime-coverage/release/release.sh =================================================================== --- projects/runtime-coverage/release/release.sh (revision 325498) +++ projects/runtime-coverage/release/release.sh (revision 325499) @@ -1,428 +1,428 @@ #!/bin/sh #- # Copyright (c) 2013-2017 The FreeBSD Foundation # Copyright (c) 2013 Glen Barber # Copyright (c) 2011 Nathan Whitehorn # All rights reserved. # # Portions of this software were developed by Glen Barber # under sponsorship from the FreeBSD Foundation. # # Redistribution and use in source and binary forms, with or without # modification, are permitted provided that the following conditions # are met: # 1. Redistributions of source code must retain the above copyright # notice, this list of conditions and the following disclaimer. # 2. Redistributions in binary form must reproduce the above copyright # notice, this list of conditions and the following disclaimer in the # documentation and/or other materials provided with the distribution. # # THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND # ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE # IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE # ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE # FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL # DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS # OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) # HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT # LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY # OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF # SUCH DAMAGE. # # release.sh: check out source trees, and build release components with # totally clean, fresh trees. # Based on release/generate-release.sh written by Nathan Whitehorn # # $FreeBSD$ # export PATH="/bin:/sbin:/usr/bin:/usr/sbin:/usr/local/bin" VERSION=2 # Prototypes that can be redefined per-chroot or per-target. load_chroot_env() { } load_target_env() { } buildenv_setup() { } usage() { echo "Usage: $0 [-c release.conf]" exit 1 } # env_setup(): Set up the default build environment variables, such as the # CHROOTDIR, VCSCMD, SVNROOT, etc. This is called before the release.conf # file is sourced, if '-c ' is specified. env_setup() { # The directory within which the release will be built. CHROOTDIR="/scratch" RELENGDIR="$(dirname $(realpath ${0}))" # The default version control system command to obtain the sources. for _dir in /usr/bin /usr/local/bin; do for _svn in svn svnlite; do [ -x "${_dir}/${_svn}" ] && VCSCMD="${_dir}/${_svn}" [ ! -z "${VCSCMD}" ] && break 2 done done VCSCMD="${VCSCMD} checkout" # The default svn checkout server, and svn branches for src/, doc/, # and ports/. SVNROOT="svn://svn.FreeBSD.org/" SRCBRANCH="base/head@rHEAD" DOCBRANCH="doc/head@rHEAD" PORTBRANCH="ports/head@rHEAD" # Set for embedded device builds. EMBEDDEDBUILD= # Sometimes one needs to checkout src with --force svn option. # If custom kernel configs copied to src tree before checkout, e.g. SRC_FORCE_CHECKOUT= # The default make.conf and src.conf to use. Set to /dev/null # by default to avoid polluting the chroot(8) environment with # non-default settings. MAKE_CONF="/dev/null" SRC_CONF="/dev/null" # The number of make(1) jobs, defaults to the number of CPUs available # for buildworld, and half of number of CPUs available for buildkernel. WORLD_FLAGS="-j$(sysctl -n hw.ncpu)" KERNEL_FLAGS="-j$(( $(( $(sysctl -n hw.ncpu) + 1 )) / 2))" MAKE_FLAGS="-s" # The name of the kernel to build, defaults to GENERIC. KERNEL="GENERIC" # Set to non-empty value to disable checkout of doc/ and/or ports/. # Disabling ports/ checkout also forces NODOC to be set. NODOC= NOPORTS= # Set to non-empty value to disable distributing source tree. NOSRC= # Set to non-empty value to build dvd1.iso as part of the release. WITH_DVD= WITH_COMPRESSED_IMAGES= # Set to non-empty value to build virtual machine images as part of # the release. WITH_VMIMAGES= WITH_COMPRESSED_VMIMAGES= XZ_THREADS=0 # Set to non-empty value to build virtual machine images for various # cloud providers as part of the release. WITH_CLOUDWARE= return 0 } # env_setup() # env_check(): Perform sanity tests on the build environment, such as ensuring # files/directories exist, as well as adding backwards-compatibility hacks if # necessary. This is called unconditionally, and overrides the defaults set # in env_setup() if '-c ' is specified. env_check() { chroot_build_release_cmd="chroot_build_release" # Fix for backwards-compatibility with release.conf that does not have # the trailing '/'. case ${SVNROOT} in *svn*) SVNROOT="${SVNROOT}/" ;; *) ;; esac # Prefix the branches with the SVNROOT for the full checkout URL. SRCBRANCH="${SVNROOT}${SRCBRANCH}" DOCBRANCH="${SVNROOT}${DOCBRANCH}" PORTBRANCH="${SVNROOT}${PORTBRANCH}" if [ -n "${EMBEDDEDBUILD}" ]; then WITH_DVD= WITH_COMPRESSED_IMAGES= NODOC=yes case ${EMBEDDED_TARGET}:${EMBEDDED_TARGET_ARCH} in arm:armv6|arm:armv7|arm64:aarch64) chroot_build_release_cmd="chroot_arm_build_release" ;; *) ;; esac fi # If PORTS is set and NODOC is unset, force NODOC=yes because the ports # tree is required to build the documentation set. if [ -n "${NOPORTS}" ] && [ -z "${NODOC}" ]; then echo "*** NOTICE: Setting NODOC=1 since ports tree is required" echo " and NOPORTS is set." NODOC=yes fi # If NOSRC, NOPORTS and/or NODOC are unset, they must not pass to make # as variables. The release makefile verifies definedness of the # NOPORTS/NODOC variables instead of their values. SRCDOCPORTS= if [ -n "${NOPORTS}" ]; then SRCDOCPORTS="NOPORTS=yes" fi if [ -n "${NODOC}" ]; then SRCDOCPORTS="${SRCDOCPORTS}${SRCDOCPORTS:+ }NODOC=yes" fi if [ -n "${NOSRC}" ]; then SRCDOCPORTS="${SRCDOCPORTS}${SRCDOCPORTS:+ }NOSRC=yes" fi # The aggregated build-time flags based upon variables defined within # this file, unless overridden by release.conf. In most cases, these # will not need to be changed. CONF_FILES="__MAKE_CONF=${MAKE_CONF} SRCCONF=${SRC_CONF}" if [ -n "${TARGET}" ] && [ -n "${TARGET_ARCH}" ]; then ARCH_FLAGS="TARGET=${TARGET} TARGET_ARCH=${TARGET_ARCH}" else ARCH_FLAGS= fi # Force src checkout if configured FORCE_SRC_KEY= if [ -n "${SRC_FORCE_CHECKOUT}" ]; then FORCE_SRC_KEY="--force" fi if [ -z "${CHROOTDIR}" ]; then echo "Please set CHROOTDIR." exit 1 fi if [ $(id -u) -ne 0 ]; then echo "Needs to be run as root." exit 1 fi CHROOT_MAKEENV="${CHROOT_MAKEENV} \ MAKEOBJDIRPREFIX=${CHROOTDIR}/tmp/obj" CHROOT_WMAKEFLAGS="${MAKE_FLAGS} ${WORLD_FLAGS} ${CONF_FILES}" CHROOT_IMAKEFLAGS="${WORLD_FLAGS} ${CONF_FILES}" CHROOT_DMAKEFLAGS="${WORLD_FLAGS} ${CONF_FILES}" RELEASE_WMAKEFLAGS="${MAKE_FLAGS} ${WORLD_FLAGS} ${ARCH_FLAGS} \ ${CONF_FILES}" RELEASE_KMAKEFLAGS="${MAKE_FLAGS} ${KERNEL_FLAGS} \ KERNCONF=\"${KERNEL}\" ${ARCH_FLAGS} ${CONF_FILES}" RELEASE_RMAKEFLAGS="${ARCH_FLAGS} \ KERNCONF=\"${KERNEL}\" ${CONF_FILES} ${SRCDOCPORTS} \ WITH_DVD=${WITH_DVD} WITH_VMIMAGES=${WITH_VMIMAGES} \ WITH_CLOUDWARE=${WITH_CLOUDWARE} XZ_THREADS=${XZ_THREADS}" return 0 } # env_check() # chroot_setup(): Prepare the build chroot environment for the release build. chroot_setup() { load_chroot_env mkdir -p ${CHROOTDIR}/usr if [ -z "${SRC_UPDATE_SKIP}" ]; then ${VCSCMD} ${FORCE_SRC_KEY} ${SRCBRANCH} ${CHROOTDIR}/usr/src fi if [ -z "${NODOC}" ] && [ -z "${DOC_UPDATE_SKIP}" ]; then ${VCSCMD} ${DOCBRANCH} ${CHROOTDIR}/usr/doc fi if [ -z "${NOPORTS}" ] && [ -z "${PORTS_UPDATE_SKIP}" ]; then ${VCSCMD} ${PORTBRANCH} ${CHROOTDIR}/usr/ports fi if [ -z "${CHROOTBUILD_SKIP}" ]; then cd ${CHROOTDIR}/usr/src env ${CHROOT_MAKEENV} make ${CHROOT_WMAKEFLAGS} buildworld env ${CHROOT_MAKEENV} make ${CHROOT_IMAKEFLAGS} installworld \ DESTDIR=${CHROOTDIR} env ${CHROOT_MAKEENV} make ${CHROOT_DMAKEFLAGS} distribution \ DESTDIR=${CHROOTDIR} fi return 0 } # chroot_setup() # extra_chroot_setup(): Prepare anything additional within the build # necessary for the release build. extra_chroot_setup() { mkdir -p ${CHROOTDIR}/dev mount -t devfs devfs ${CHROOTDIR}/dev [ -e /etc/resolv.conf ] && cp /etc/resolv.conf \ ${CHROOTDIR}/etc/resolv.conf # Run ldconfig(8) in the chroot directory so /var/run/ld-elf*.so.hints # is created. This is needed by ports-mgmt/pkg. eval chroot ${CHROOTDIR} /etc/rc.d/ldconfig forcerestart # If MAKE_CONF and/or SRC_CONF are set and not character devices # (/dev/null), copy them to the chroot. if [ -e ${MAKE_CONF} ] && [ ! -c ${MAKE_CONF} ]; then mkdir -p ${CHROOTDIR}/$(dirname ${MAKE_CONF}) cp ${MAKE_CONF} ${CHROOTDIR}/${MAKE_CONF} fi if [ -e ${SRC_CONF} ] && [ ! -c ${SRC_CONF} ]; then mkdir -p ${CHROOTDIR}/$(dirname ${SRC_CONF}) cp ${SRC_CONF} ${CHROOTDIR}/${SRC_CONF} fi if [ -d ${CHROOTDIR}/usr/ports ]; then # Trick the ports 'run-autotools-fixup' target to do the right # thing. _OSVERSION=$(chroot ${CHROOTDIR} /usr/bin/uname -U) REVISION=$(chroot ${CHROOTDIR} make -C /usr/src/release -V REVISION) BRANCH=$(chroot ${CHROOTDIR} make -C /usr/src/release -V BRANCH) UNAME_r=${REVISION}-${BRANCH} if [ -d ${CHROOTDIR}/usr/doc ] && [ -z "${NODOC}" ]; then PBUILD_FLAGS="OSVERSION=${_OSVERSION} BATCH=yes" PBUILD_FLAGS="${PBUILD_FLAGS} UNAME_r=${UNAME_r}" PBUILD_FLAGS="${PBUILD_FLAGS} OSREL=${REVISION}" chroot ${CHROOTDIR} make -C /usr/ports/textproc/docproj \ ${PBUILD_FLAGS} OPTIONS_UNSET="FOP IGOR" \ FORCE_PKG_REGISTER=1 \ install clean distclean fi fi if [ ! -z "${EMBEDDEDPORTS}" ]; then _OSVERSION=$(chroot ${CHROOTDIR} /usr/bin/uname -U) REVISION=$(chroot ${CHROOTDIR} make -C /usr/src/release -V REVISION) BRANCH=$(chroot ${CHROOTDIR} make -C /usr/src/release -V BRANCH) PBUILD_FLAGS="OSVERSION=${_OSVERSION} BATCH=yes" PBUILD_FLAGS="${PBUILD_FLAGS} UNAME_r=${UNAME_r}" PBUILD_FLAGS="${PBUILD_FLAGS} OSREL=${REVISION}" for _PORT in ${EMBEDDEDPORTS}; do eval chroot ${CHROOTDIR} make -C /usr/ports/${_PORT} \ FORCE_PKG_REGISTER=1 ${PBUILD_FLAGS} install clean distclean done fi buildenv_setup return 0 } # extra_chroot_setup() # chroot_build_target(): Build the userland and kernel for the build target. chroot_build_target() { load_target_env if [ ! -z "${EMBEDDEDBUILD}" ]; then RELEASE_WMAKEFLAGS="${RELEASE_WMAKEFLAGS} \ TARGET=${EMBEDDED_TARGET} \ TARGET_ARCH=${EMBEDDED_TARGET_ARCH}" RELEASE_KMAKEFLAGS="${RELEASE_KMAKEFLAGS} \ TARGET=${EMBEDDED_TARGET} \ TARGET_ARCH=${EMBEDDED_TARGET_ARCH}" fi eval chroot ${CHROOTDIR} make -C /usr/src ${RELEASE_WMAKEFLAGS} buildworld eval chroot ${CHROOTDIR} make -C /usr/src ${RELEASE_KMAKEFLAGS} buildkernel return 0 } # chroot_build_target # chroot_build_release(): Invoke the 'make release' target. chroot_build_release() { load_target_env if [ ! -z "${WITH_VMIMAGES}" ]; then if [ -z "${VMFORMATS}" ]; then VMFORMATS="$(eval chroot ${CHROOTDIR} \ make -C /usr/src/release -V VMFORMATS)" fi if [ -z "${VMSIZE}" ]; then VMSIZE="$(eval chroot ${CHROOTDIR} \ make -C /usr/src/release -V VMSIZE)" fi RELEASE_RMAKEFLAGS="${RELEASE_RMAKEFLAGS} \ VMFORMATS=\"${VMFORMATS}\" VMSIZE=${VMSIZE}" fi eval chroot ${CHROOTDIR} make -C /usr/src/release \ ${RELEASE_RMAKEFLAGS} release eval chroot ${CHROOTDIR} make -C /usr/src/release \ ${RELEASE_RMAKEFLAGS} install DESTDIR=/R \ WITH_COMPRESSED_IMAGES=${WITH_COMPRESSED_IMAGES} \ WITH_COMPRESSED_VMIMAGES=${WITH_COMPRESSED_VMIMAGES} return 0 } # chroot_build_release() # chroot_arm_build_release(): Create arm SD card image. chroot_arm_build_release() { load_target_env - MAKE_FLAGS="${MAKE_FLAGS} TARGET=${EMBEDDED_TARGET}" - MAKE_FLAGS="${MAKE_FLAGS} TARGET_ARCH=${EMBEDDED_TARGET_ARCH}" - eval chroot ${CHROOTDIR} make ${MAKE_FLAGS} -C /usr/src/release obj case ${EMBEDDED_TARGET} in arm|arm64) if [ -e "${RELENGDIR}/tools/arm.subr" ]; then . "${RELENGDIR}/tools/arm.subr" fi ;; *) ;; esac [ ! -z "${RELEASECONF}" ] && . "${RELEASECONF}" - WORLDDIR="$(eval chroot ${CHROOTDIR} make ${MAKE_FLAGS} -C /usr/src/release -V WORLDDIR)" - OBJDIR="$(eval chroot ${CHROOTDIR} make ${MAKE_FLAGS} -C /usr/src/release -V .OBJDIR)" - DESTDIR="${OBJDIR}/${KERNEL}" - IMGBASE="${CHROOTDIR}/${OBJDIR}/${KERNEL}.img" - OSRELEASE="$(eval chroot ${CHROOTDIR} make ${MAKE_FLAGS} -C /usr/src/release \ + export MAKE_FLAGS="${MAKE_FLAGS} TARGET=${EMBEDDED_TARGET}" + export MAKE_FLAGS="${MAKE_FLAGS} TARGET_ARCH=${EMBEDDED_TARGET_ARCH}" + eval chroot ${CHROOTDIR} env WITH_UNIFIED_OBJDIR=1 make ${MAKE_FLAGS} -C /usr/src/release obj + export WORLDDIR="$(eval chroot ${CHROOTDIR} make ${MAKE_FLAGS} -C /usr/src/release -V WORLDDIR)" + export OBJDIR="$(eval chroot ${CHROOTDIR} env WITH_UNIFIED_OBJDIR=1 make ${MAKE_FLAGS} -C /usr/src/release -V .OBJDIR)" + export DESTDIR="${OBJDIR}/${KERNEL}" + export IMGBASE="${CHROOTDIR}/${OBJDIR}/${KERNEL}.img" + export OSRELEASE="$(eval chroot ${CHROOTDIR} make ${MAKE_FLAGS} -C /usr/src/release \ TARGET=${EMBEDDED_TARGET} TARGET_ARCH=${EMBEDDED_TARGET_ARCH} \ -V OSRELEASE)" chroot ${CHROOTDIR} mkdir -p ${DESTDIR} chroot ${CHROOTDIR} truncate -s ${IMAGE_SIZE} ${IMGBASE##${CHROOTDIR}} export mddev=$(chroot ${CHROOTDIR} \ mdconfig -f ${IMGBASE##${CHROOTDIR}} ${MD_ARGS}) arm_create_disk arm_install_base arm_install_uboot mdconfig -d -u ${mddev} chroot ${CHROOTDIR} rmdir ${DESTDIR} mv ${IMGBASE} ${CHROOTDIR}/${OBJDIR}/${OSRELEASE}-${KERNEL}.img chroot ${CHROOTDIR} mkdir -p /R chroot ${CHROOTDIR} cp -p ${OBJDIR}/${OSRELEASE}-${KERNEL}.img \ /R/${OSRELEASE}-${KERNEL}.img chroot ${CHROOTDIR} xz -T ${XZ_THREADS} /R/${OSRELEASE}-${KERNEL}.img cd ${CHROOTDIR}/R && sha512 ${OSRELEASE}* \ > CHECKSUM.SHA512 cd ${CHROOTDIR}/R && sha256 ${OSRELEASE}* \ > CHECKSUM.SHA256 return 0 } # chroot_arm_build_release() # main(): Start here. main() { set -e # Everything must succeed env_setup while getopts c: opt; do case ${opt} in c) RELEASECONF="$(realpath ${OPTARG})" ;; \?) usage ;; esac done shift $(($OPTIND - 1)) if [ ! -z "${RELEASECONF}" ]; then if [ -e "${RELEASECONF}" ]; then . ${RELEASECONF} else echo "Nonexistent configuration file: ${RELEASECONF}" echo "Using default build environment." fi fi env_check trap "umount ${CHROOTDIR}/dev" EXIT # Clean up devfs mount on exit chroot_setup extra_chroot_setup chroot_build_target ${chroot_build_release_cmd} return 0 } # main() main "${@}" Index: projects/runtime-coverage/release/scripts/mm-mtree.sh =================================================================== --- projects/runtime-coverage/release/scripts/mm-mtree.sh (revision 325498) +++ projects/runtime-coverage/release/scripts/mm-mtree.sh (revision 325499) @@ -1,158 +1,154 @@ #!/bin/sh # mergemaster mtree database generator # This script is intended to be used as part of the release building # process to generate the /var/db/mergemaster.mtree file relevant to # the source tree used to create the release so that users can make # use of mergemaster's -U option to update their files after updating # to -stable. # Copyright 2009 Douglas Barton # dougb@FreeBSD.org # $FreeBSD$ PATH=/bin:/usr/bin:/usr/sbin display_usage () { VERSION_NUMBER=`grep "[$]FreeBSD:" $0 | cut -d ' ' -f 4` echo "${0##*/} version ${VERSION_NUMBER}" echo "Usage: ${0##*/} [-m /path] [-t /path] [-A arch] [-F ] [-D /path]" echo "Options:" echo " -m /path/directory Specify location of source to do the make in" echo " -t /path/directory Specify temp root directory" echo " -A architecture Alternative architecture name to pass to make" echo " -F Specify what to put on the make command line" echo ' -D /path/directory Specify the destination directory to install files to' echo '' } # Set the default path for the temporary root environment # TEMPROOT=`TMPDIR=/var/tmp mktemp -d -t temproot` # Assign the location of the mtree database # MTREEDB=${MTREEDB:-/var/db} MTREEFILE="${MTREEDB}/mergemaster.mtree" # Check the command line options # while getopts "m:t:A:F:D:h" COMMAND_LINE_ARGUMENT ; do case "${COMMAND_LINE_ARGUMENT}" in m) SOURCEDIR=${OPTARG} ;; t) TEMPROOT=${OPTARG} ;; A) ARCHSTRING='TARGET_ARCH='${OPTARG} ;; F) MM_MAKE_ARGS="${OPTARG}" ;; D) DESTDIR=${OPTARG} ;; h) display_usage exit 0 ;; *) echo '' display_usage exit 1 ;; esac done # Assign the source directory # SOURCEDIR=${SOURCEDIR:-/usr/src} if [ ! -f ${SOURCEDIR}/Makefile.inc1 -a \ -f ${SOURCEDIR}/../Makefile.inc1 ]; then echo " *** The source directory you specified (${SOURCEDIR})" echo " will be reset to ${SOURCEDIR}/.." echo '' sleep 3 SOURCEDIR=${SOURCEDIR}/.. fi # Setup make to use system files from SOURCEDIR MM_MAKE="make ${ARCHSTRING} ${MM_MAKE_ARGS} -m ${SOURCEDIR}/share/mk" delete_temproot () { rm -rf "${TEMPROOT}" 2>/dev/null chflags -R 0 "${TEMPROOT}" 2>/dev/null rm -rf "${TEMPROOT}" || exit 1 } [ -d "${TEMPROOT}" ] && delete_temproot echo "*** Creating the temporary root environment in ${TEMPROOT}" if mkdir -p "${TEMPROOT}"; then echo " *** ${TEMPROOT} ready for use" fi if [ ! -d "${TEMPROOT}" ]; then echo '' echo " *** FATAL ERROR: Cannot create ${TEMPROOT}" echo '' exit 1 fi echo " *** Creating and populating directory structure in ${TEMPROOT}" echo '' { cd ${SOURCEDIR} || { echo "*** Cannot cd to ${SOURCEDIR}" ; exit 1;} case "${DESTDIR}" in '') ;; *) ${MM_MAKE} DESTDIR=${DESTDIR} distrib-dirs ;; esac - od=${TEMPROOT}/usr/obj ${MM_MAKE} DESTDIR=${TEMPROOT} distrib-dirs && - MAKEOBJDIRPREFIX=$od ${MM_MAKE} kernel-toolchain \ - MK_TOOLCHAIN=no MK_CROSS_COMPILER=no \ - MK_CDDL=no MK_TESTS=no MK_RESCUE=no && - MAKEOBJDIRPREFIX=$od ${MM_MAKE} _obj SUBDIR_OVERRIDE=etc && - MAKEOBJDIRPREFIX=$od ${MM_MAKE} everything SUBDIR_OVERRIDE=etc && - MAKEOBJDIRPREFIX=$od ${MM_MAKE} DESTDIR=${TEMPROOT} distribution;} || + ${MM_MAKE} _obj SUBDIR_OVERRIDE=etc && + ${MM_MAKE} everything SUBDIR_OVERRIDE=etc && + ${MM_MAKE} DESTDIR=${TEMPROOT} distribution;} || { echo ''; echo " *** FATAL ERROR: Cannot 'cd' to ${SOURCEDIR} and install files to"; echo " the temproot environment"; echo ''; exit 1;} # We really don't want to have to deal with files like login.conf.db, pwd.db, # or spwd.db. Instead, we want to compare the text versions, and run *_mkdb. # Prompt the user to do so below, as needed. # rm -f ${TEMPROOT}/etc/*.db ${TEMPROOT}/etc/passwd # We only need to compare things like freebsd.cf once find ${TEMPROOT}/usr/obj -type f -delete 2>/dev/null # Delete stuff we do not need to keep the mtree database small, # and to make the actual comparison faster. find ${TEMPROOT}/usr -type l -delete 2>/dev/null find ${TEMPROOT} -type f -size 0 -delete 2>/dev/null find -d ${TEMPROOT} -type d -empty -delete 2>/dev/null # Build the mtree database in a temporary location. MTREENEW=`mktemp -t mergemaster.mtree` mtree -nci -p ${TEMPROOT} -k size,md5digest > ${MTREENEW} 2>/dev/null if [ -s "${MTREENEW}" ]; then echo "*** Saving mtree database for future upgrades" test -e "${DESTDIR}${MTREEFILE}" && unlink ${DESTDIR}${MTREEFILE} mv ${MTREENEW} ${DESTDIR}${MTREEFILE} fi delete_temproot exit 0 Index: projects/runtime-coverage/share/mk/bsd.incs.mk =================================================================== --- projects/runtime-coverage/share/mk/bsd.incs.mk (revision 325498) +++ projects/runtime-coverage/share/mk/bsd.incs.mk (revision 325499) @@ -1,111 +1,110 @@ # $FreeBSD$ .if !target(____) .error bsd.incs.mk cannot be included directly. .endif .if ${MK_INCLUDES} != "no" INCSGROUPS?= INCS .if defined(NO_ROOT) .if !defined(TAGS) || ! ${TAGS:Mpackage=*} TAGS+= package=${PACKAGE:Uruntime} .endif TAG_ARGS= -T ${TAGS:[*]:S/ /,/g} .endif .if !target(buildincludes) .for group in ${INCSGROUPS} buildincludes: ${${group}} .endfor .endif .if !defined(_SKIP_BUILD) all: buildincludes .endif .if !target(installincludes) .for group in ${INCSGROUPS} .if defined(${group}) && !empty(${group}) ${group}OWN?= ${BINOWN} ${group}GRP?= ${BINGRP} ${group}MODE?= ${NOBINMODE} ${group}DIR?= ${INCLUDEDIR}${PRIVATELIB:D/private/${LIB}} STAGE_SETS+= ${group:C,[/*],_,g} STAGE_DIR.${group:C,[/*],_,g}= ${STAGE_OBJTOP}${${group}DIR} STAGE_SYMLINKS_DIR.${group:C,[/*],_,g}= ${STAGE_OBJTOP} _${group}INCS= .for header in ${${group}} .if defined(${group}OWN_${header:T}) || defined(${group}GRP_${header:T}) || \ defined(${group}MODE_${header:T}) || defined(${group}DIR_${header:T}) || \ defined(${group}NAME_${header:T}) || defined(${group}NAME) ${group}OWN_${header:T}?= ${${group}OWN} ${group}GRP_${header:T}?= ${${group}GRP} ${group}MODE_${header:T}?= ${${group}MODE} ${group}DIR_${header:T}?= ${${group}DIR} .if defined(${group}NAME) ${group}NAME_${header:T}?= ${${group}NAME} .else ${group}NAME_${header:T}?= ${header:T} .endif STAGE_AS_SETS+= ${header:T} STAGE_AS_${header:T}= ${${group}NAME_${header:T}} # XXX {group}OWN,GRP,MODE STAGE_DIR.${header:T}= ${STAGE_OBJTOP}${${group}DIR_${header:T}} stage_as.${header:T}: ${header} stage_includes: stage_as.${header:T} installincludes: _${group}INS_${header:T} _${group}INS_${header:T}: ${header} ${INSTALL} -C -o ${${group}OWN_${.ALLSRC:T}} \ -g ${${group}GRP_${.ALLSRC:T}} -m ${${group}MODE_${.ALLSRC:T}} \ ${.ALLSRC} \ ${DESTDIR}${${group}DIR_${.ALLSRC:T}}/${${group}NAME_${.ALLSRC:T}} .else _${group}INCS+= ${header} .endif .endfor .if !empty(_${group}INCS) stage_files.${group}: ${_${group}INCS} stage_includes: stage_files.${group} installincludes: _${group}INS _${group}INS: ${_${group}INCS} .if defined(${group}NAME) ${INSTALL} ${TAG_ARGS:D${TAG_ARGS},development} -C -o ${${group}OWN} -g ${${group}GRP} -m ${${group}MODE} \ ${.ALLSRC} ${DESTDIR}${${group}DIR}/${${group}NAME} .else ${INSTALL} ${TAG_ARGS:D${TAG_ARGS},development} -C -o ${${group}OWN} -g ${${group}GRP} -m ${${group}MODE} \ ${.ALLSRC} ${DESTDIR}${${group}DIR}/ .endif .endif .endif # defined(${group}) && !empty(${group}) .endfor .if defined(INCSLINKS) && !empty(INCSLINKS) installincludes: .for s t in ${INCSLINKS} - @${ECHO} "${DESTDIR}${t} -> ${s}" ; \ ${INSTALL_SYMLINK} ${TAG_ARGS:D${TAG_ARGS},development} ${s} ${DESTDIR}${t} .endfor .endif .endif # !target(installincludes) realinstall: installincludes .ORDER: beforeinstall installincludes .if ${MK_STAGING} != "no" && !defined(_SKIP_BUILD) .if !defined(NO_STAGE_INCLUDES) STAGE_TARGETS+= stage_includes .if !empty(INCSLINKS) STAGE_TARGETS+= stage_symlinks STAGE_SYMLINKS.INCS= ${INCSLINKS} .endif .endif .endif .endif # ${MK_INCLUDES} != "no" Index: projects/runtime-coverage/share/mk/bsd.links.mk =================================================================== --- projects/runtime-coverage/share/mk/bsd.links.mk (revision 325498) +++ projects/runtime-coverage/share/mk/bsd.links.mk (revision 325499) @@ -1,24 +1,22 @@ # $FreeBSD$ .if !target(____) .error bsd.links.mk cannot be included directly. .endif .if defined(NO_ROOT) .if !defined(TAGS) || ! ${TAGS:Mpackage=*} TAGS+= package=${PACKAGE} .endif TAG_ARGS= -T ${TAGS:[*]:S/ /,/g} .endif afterinstall: _installlinks .ORDER: realinstall _installlinks _installlinks: .for s t in ${LINKS} - @${ECHO} "${t} -> ${s}" ;\ ${INSTALL_LINK} ${TAG_ARGS} ${DESTDIR}${s} ${DESTDIR}${t} .endfor .for s t in ${SYMLINKS} - @${ECHO} "${t} -> ${s}" ;\ ${INSTALL_SYMLINK} ${TAG_ARGS} ${s} ${DESTDIR}${t} .endfor Index: projects/runtime-coverage/sys/boot/Makefile =================================================================== --- projects/runtime-coverage/sys/boot/Makefile (revision 325498) +++ projects/runtime-coverage/sys/boot/Makefile (revision 325499) @@ -1,22 +1,20 @@ # $FreeBSD$ .include SUBDIR+= libsa .if ${MK_FORTH} != "no" # Build the add-in FORTH interpreter. SUBDIR+= ficl SUBDIR+= forth .endif SUBDIR+= man .include -# Pick the machine-dependent subdir based on the target architecture. -ADIR= ${MACHINE:S/powerpc64/powerpc/} -.if exists(${.CURDIR}/${ADIR}/.) -SUBDIR+= ${ADIR} +.if exists(${.CURDIR}/${MACHINE}/.) +SUBDIR+= ${MACHINE} .endif .include Index: projects/runtime-coverage/sys/boot/Makefile.inc =================================================================== --- projects/runtime-coverage/sys/boot/Makefile.inc (revision 325498) +++ projects/runtime-coverage/sys/boot/Makefile.inc (revision 325499) @@ -1,41 +1,39 @@ # $FreeBSD$ -.include - MK_COVERAGE= no .include "defs.mk" .if !defined(__BOOT_MAKEFILE_INC__) __BOOT_MAKEFILE_INC__=${MFILE} CFLAGS+=-I${SASRC} SSP_CFLAGS= .if ${MACHINE_CPUARCH} == "arm" # Do not generate movt/movw, because the relocation fixup for them does not # translate to the -Bsymbolic -pie format required by self_reloc() in loader(8). # Also, the fpu is not available in a standalone environment. .if ${COMPILER_VERSION} < 30800 CFLAGS.clang+= -mllvm -arm-use-movt=0 .else CFLAGS.clang+= -mno-movt .endif CFLAGS.clang+= -mfpu=none .endif # The boot loader build uses dd status=none, where possible, for reproducible # build output (since performance varies from run to run). Trouble is that # option was recently (10.3) added to FreeBSD and is non-standard. Only use it # when this test succeeds rather than require dd to be a bootstrap tool. DD_NOSTATUS!=(dd status=none count=0 2> /dev/null && echo status=none) || true DD=dd ${DD_NOSTATUS} .if ${MK_LOADER_FORCE_LE} != "no" .if ${MACHINE_ARCH} == "powerpc64" CFLAGS+= -mlittle-endian .endif .endif .endif Index: projects/runtime-coverage/sys/boot/Makefile.powerpc =================================================================== --- projects/runtime-coverage/sys/boot/Makefile.powerpc (revision 325498) +++ projects/runtime-coverage/sys/boot/Makefile.powerpc (revision 325499) @@ -1,9 +1,8 @@ # $FreeBSD$ .if ${MK_FDT} != "no" SUBDIR+= fdt .endif -SUBDIR+= libsa32 SUBDIR+= ofw SUBDIR+= uboot Index: projects/runtime-coverage/sys/boot/defs.mk =================================================================== --- projects/runtime-coverage/sys/boot/defs.mk (revision 325498) +++ projects/runtime-coverage/sys/boot/defs.mk (revision 325499) @@ -1,89 +1,148 @@ # $FreeBSD$ .include .if !defined(__BOOT_DEFS_MK__) __BOOT_DEFS_MK__=${MFILE} BOOTSRC= ${SRCTOP}/sys/boot EFISRC= ${BOOTSRC}/efi EFIINC= ${EFISRC}/include EFIINCMD= ${EFIINC}/${MACHINE} FDTSRC= ${BOOTSRC}/fdt FICLSRC= ${BOOTSRC}/ficl LDRSRC= ${BOOTSRC}/common SASRC= ${BOOTSRC}/libsa SYSDIR= ${SRCTOP}/sys UBOOTSRC= ${BOOTSRC}/uboot ZFSSRC= ${BOOTSRC}/zfs BOOTOBJ= ${OBJTOP}/sys/boot # BINDIR is where we install BINDIR?= /boot # NB: The makefiles depend on these being empty when we don't build forth. .if ${MK_FORTH} != "no" LIBFICL= ${BOOTOBJ}/ficl/libficl.a +.if ${MACHINE} == "i386" +LIBFICL32= ${LIBFICL} +.else LIBFICL32= ${BOOTOBJ}/ficl32/libficl.a .endif +.endif LIBSA= ${BOOTOBJ}/libsa/libsa.a .if ${MACHINE} == "i386" LIBSA32= ${LIBSA} .else LIBSA32= ${BOOTOBJ}/libsa32/libsa32.a .endif # Standard options: # Filesystem support .if ${LOADER_CD9660_SUPPORT:Uno} == "yes" CFLAGS+= -DLOADER_CD9660_SUPPORT .endif .if ${LOADER_EXT2FS_SUPPORT:Uno} == "yes" CFLAGS+= -DLOADER_EXT2FS_SUPPORT .endif .if ${LOADER_MSDOS_SUPPORT:Uno} == "yes" CFLAGS+= -DLOADER_MSDOS_SUPPORT .endif .if ${LOADER_NANDFS_SUPPORT:U${MK_NAND}} == "yes" CFLAGS+= -DLOADER_NANDFS_SUPPORT .endif .if ${LOADER_UFS_SUPPORT:Uyes} == "yes" CFLAGS+= -DLOADER_UFS_SUPPORT .endif # Compression .if ${LOADER_GZIP_SUPPORT:Uno} == "yes" CFLAGS+= -DLOADER_GZIP_SUPPORT .endif .if ${LOADER_BZIP2_SUPPORT:Uno} == "yes" CFLAGS+= -DLOADER_BZIP2_SUPPORT .endif # Network related things .if ${LOADER_NET_SUPPORT:Uno} == "yes" CFLAGS+= -DLOADER_NET_SUPPORT .endif .if ${LOADER_NFS_SUPPORT:Uno} == "yes" CFLAGS+= -DLOADER_NFS_SUPPORT .endif .if ${LOADER_TFTP_SUPPORT:Uno} == "yes" CFLAGS+= -DLOADER_TFTP_SUPPORT .endif # Disk and partition support .if ${LOADER_DISK_SUPPORT:Uyes} == "yes" CFLAGS+= -DLOADER_DISK_SUPPORT .if ${LOADER_GPT_SUPPORT:Uyes} == "yes" CFLAGS+= -DLOADER_GPT_SUPPORT .endif .if ${LOADER_MBR_SUPPORT:Uyes} == "yes" CFLAGS+= -DLOADER_MBR_SUPPORT .endif .if ${LOADER_GELI_SUPPORT:Uyes} == "yes" CFLAGS+= -DLOADER_GELI_SUPPORT .endif .endif + +# All PowerPC builds are 32 bit. We have no 64-bit loaders on powerpc +# or powerpc64. +.if ${MACHINE_ARCH} == "powerpc64" +CFLAGS+= -m32 -mcpu=powerpc +.endif + +# For amd64, there's a bit of mixed bag. Some of the tree (i386, lib*32) is +# build 32-bit and some 64-bit (lib*, efi). Centralize all the 32-bit magic here +# and activate it when DO32 is explicitly defined to be 1. +.if ${MACHINE_ARCH} == "amd64" && ${DO32:U0} == 1 +CFLAGS+= -m32 -mcpu=i386 +# LD_FLAGS is passed directly to ${LD}, not via ${CC}: +LD_FLAGS+= -m elf_i386_fbsd +AFLAGS+= --32 +.endif + +_ILINKS=machine +.if ${MACHINE} != ${MACHINE_CPUARCH} && ${MACHINE} != "arm64" +_ILINKS+=${MACHINE_CPUARCH} +.endif +.if ${MACHINE_CPUARCH} == "i386" || ${MACHINE_CPUARCH} == "amd64" +_ILINKS+=x86 +.endif +CLEANFILES+=${_ILINKS} + +all: ${PROG} + +beforedepend: ${_ILINKS} +beforebuild: ${_ILINKS} + +# Ensure that the links exist without depending on it when it exists which +# causes all the modules to be rebuilt when the directory pointed to changes. +.for _link in ${_ILINKS} +.if !exists(${.OBJDIR}/${_link}) +${OBJS}: ${_link} +.endif +.endfor + +.NOPATH: ${_ILINKS} + +${_ILINKS}: + @case ${.TARGET} in \ + machine) \ + if [ ${DO32:U0} -eq 0 ]; then \ + path=${SYSDIR}/${MACHINE}/include ; \ + else \ + path=${SYSDIR}/${MACHINE:C/amd64/i386/}/include ; \ + fi ;; \ + *) \ + path=${SYSDIR}/${.TARGET:T}/include ;; \ + esac ; \ + path=`(cd $$path && /bin/pwd)` ; \ + ${ECHO} ${.TARGET:T} "->" $$path ; \ + ln -fhs $$path ${.TARGET:T} .endif # __BOOT_DEFS_MK__ Index: projects/runtime-coverage/sys/boot/efi/Makefile =================================================================== --- projects/runtime-coverage/sys/boot/efi/Makefile (revision 325498) +++ projects/runtime-coverage/sys/boot/efi/Makefile (revision 325499) @@ -1,23 +1,23 @@ # $FreeBSD$ -.include +.include # In-tree GCC does not support __attribute__((ms_abi)), but gcc newer # than 4.5 supports it. .if ${COMPILER_TYPE} != "gcc" || ${COMPILER_VERSION} >= 40500 .if ${MACHINE_CPUARCH} == "aarch64" || ${MACHINE_CPUARCH} == "arm" .if ${MK_FDT} != "no" SUBDIR+= fdt .endif .endif .if ${MACHINE_CPUARCH} == "aarch64" || \ ${MACHINE_CPUARCH} == "amd64" || \ ${MACHINE_CPUARCH} == "arm" SUBDIR+= libefi loader boot1 .endif .endif # ${COMPILER_TYPE} != "gcc" || ${COMPILER_VERSION} >= 40500 .include Index: projects/runtime-coverage/sys/boot/efi/boot1/Makefile =================================================================== --- projects/runtime-coverage/sys/boot/efi/boot1/Makefile (revision 325498) +++ projects/runtime-coverage/sys/boot/efi/boot1/Makefile (revision 325499) @@ -1,146 +1,131 @@ # $FreeBSD$ MAN= .include MK_SSP= no MK_FORTH= no PROG= boot1.sym INTERNALPROG= WARNS?= 6 # We implement a slightly non-standard %S in that it always takes a # CHAR16 that's common in UEFI-land instead of a wchar_t. This only # seems to matter on arm64 where wchar_t defaults to an int instead # of a short. There's no good cast to use here so just ignore the # warnings for now. CWARNFLAGS.boot1.c+= -Wno-format # Disable warnings that are currently incompatible with the zfs boot code CWARNFLAGS.zfs_module.c += -Wno-array-bounds CWARNFLAGS.zfs_module.c += -Wno-cast-align CWARNFLAGS.zfs_module.c += -Wno-cast-qual CWARNFLAGS.zfs_module.c += -Wno-missing-prototypes CWARNFLAGS.zfs_module.c += -Wno-sign-compare CWARNFLAGS.zfs_module.c += -Wno-unused-parameter CWARNFLAGS.zfs_module.c += -Wno-unused-function # architecture-specific loader code SRCS= boot1.c self_reloc.c start.S ufs_module.c .if ${MK_ZFS} != "no" SRCS+= zfs_module.c CFLAGS+= -I${ZFSSRC} CFLAGS+= -I${SYSDIR}/cddl/boot/zfs CFLAGS+= -DEFI_ZFS_BOOT LIBZFSBOOT= ${BOOTOBJ}/zfs/libzfsboot.a .endif .if ${COMPILER_TYPE} == "gcc" && ${COMPILER_VERSION} > 40201 CWARNFLAGS.self_reloc.c+= -Wno-error=maybe-uninitialized .endif CFLAGS+= -I. CFLAGS+= -I${EFIINC} CFLAGS+= -I${EFIINCMD} CFLAGS+= -I${SYSDIR}/contrib/dev/acpica/include CFLAGS+= -I${SYSDIR} CFLAGS+= -DEFI_UFS_BOOT .ifdef(EFI_DEBUG) CFLAGS+= -DEFI_DEBUG .endif # Always add MI sources and REGULAR efi loader bits .PATH: ${EFISRC}/loader/arch/${MACHINE} .PATH: ${EFISRC}/loader .PATH: ${LDRSRC} CFLAGS+= -I${LDRSRC} FILES= boot1.efi boot1.efifat FILESMODE_boot1.efi= ${BINMODE} LDSCRIPT= ${EFISRC}/loader/arch/${MACHINE}/ldscript.${MACHINE} LDFLAGS+= -Wl,-T${LDSCRIPT},-Bsymbolic,-znotext -shared .if ${MACHINE_CPUARCH} == "aarch64" CFLAGS+= -mgeneral-regs-only .endif .if ${MACHINE_CPUARCH} == "amd64" || ${MACHINE_CPUARCH} == "i386" CFLAGS+= -fPIC LDFLAGS+= -Wl,-znocombreloc .endif LIBEFI= ${BOOTOBJ}/efi/libefi/libefi.a # # Add libstand for the runtime functions used by the compiler - for example # __aeabi_* (arm) or __divdi3 (i386). # as well as required string and memory functions for all platforms. # DPADD+= ${LIBEFI} ${LIBZFSBOOT} ${LIBSA} LDADD+= ${LIBEFI} ${LIBZFSBOOT} ${LIBSA} DPADD+= ${LDSCRIPT} NM?= nm OBJCOPY?= objcopy .if ${MACHINE_CPUARCH} == "amd64" EFI_TARGET= efi-app-x86_64 .elif ${MACHINE_CPUARCH} == "i386" EFI_TARGET= efi-app-ia32 .else EFI_TARGET= binary .endif # Arbitrarily set the PE/COFF header timestamps to 1 Jan 2016 00:00:00 # for build reproducibility. SOURCE_DATE_EPOCH?=1451606400 boot1.efi: ${PROG} if ${NM} ${.ALLSRC} | grep ' U '; then \ echo "Undefined symbols in ${.ALLSRC}"; \ exit 1; \ fi SOURCE_DATE_EPOCH=${SOURCE_DATE_EPOCH} \ ${OBJCOPY} -j .peheader -j .text -j .sdata -j .data \ -j .dynamic -j .dynsym -j .rel.dyn \ -j .rela.dyn -j .reloc -j .eh_frame \ --output-target=${EFI_TARGET} ${.ALLSRC} ${.TARGET} boot1.o: ${SASRC}/ufsread.c # The following inserts our objects into a template FAT file system # created by generate-fat.sh .include "${.CURDIR}/Makefile.fat" boot1.efifat: boot1.efi @set -- `ls -l ${.ALLSRC}`; \ x=$$(($$5-${BOOT1_MAXSIZE})); \ if [ $$x -ge 0 ]; then \ echo "boot1 $$x bytes too large; regenerate FAT templates?" >&2 ;\ exit 1; \ fi echo ${.OBJDIR} xz -d -c ${.CURDIR}/fat-${MACHINE}.tmpl.xz > ${.TARGET} ${DD} if=${.ALLSRC} of=${.TARGET} seek=${BOOT1_OFFSET} conv=notrunc CLEANFILES= boot1.efi boot1.efifat .include - -beforedepend ${OBJS}: machine - -CLEANFILES+= machine - -machine: .NOMETA - ln -sf ${SYSDIR}/${MACHINE}/include machine - -.if ${MACHINE_CPUARCH} == "amd64" || ${MACHINE_CPUARCH} == "i386" -beforedepend ${OBJS}: x86 -CLEANFILES+= x86 - -x86: .NOMETA - ln -sf ${SYSDIR}/x86/include x86 -.endif Index: projects/runtime-coverage/sys/boot/efi/fdt/Makefile =================================================================== --- projects/runtime-coverage/sys/boot/efi/fdt/Makefile (revision 325498) +++ projects/runtime-coverage/sys/boot/efi/fdt/Makefile (revision 325499) @@ -1,37 +1,30 @@ # $FreeBSD$ .include .PATH: ${LDRSRC} LIB= efi_fdt INTERNALLIB= WARNS?= 6 SRCS= efi_fdt.c CFLAGS+= -ffreestanding .if ${MACHINE_CPUARCH} == "aarch64" CFLAGS+= -mgeneral-regs-only .else CFLAGS+= -msoft-float .endif # EFI library headers CFLAGS+= -I${EFISRC}/include CFLAGS+= -I${EFISRC}/include/${MACHINE} # libfdt headers CFLAGS+= -I${FDTSRC} # Pick up the bootstrap header for some interface items CFLAGS+= -I${LDRSRC} -I${SYSDIR} -I. -machine: .NOMETA - ln -sf ${SYSDIR}/${MACHINE}/include machine - -CLEANFILES+= machine - .include - -beforedepend ${OBJS}: machine Index: projects/runtime-coverage/sys/boot/efi/loader/Makefile =================================================================== --- projects/runtime-coverage/sys/boot/efi/loader/Makefile (revision 325498) +++ projects/runtime-coverage/sys/boot/efi/loader/Makefile (revision 325499) @@ -1,150 +1,135 @@ # $FreeBSD$ MAN= .include MK_SSP= no PROG= loader.sym INTERNALPROG= WARNS?= 3 LOADER_NET_SUPPORT?= yes LOADER_MSDOS_SUPPORT?= yes LOADER_UFS_SUPPORT?= yes LOADER_CD9660_SUPPORT?= no LOADER_EXT2FS_SUPPORT?= no # architecture-specific loader code SRCS= autoload.c \ bootinfo.c \ conf.c \ copy.c \ efi_main.c \ framebuffer.c \ main.c \ self_reloc.c \ smbios.c \ vers.c .if ${MK_ZFS} != "no" LIBZFSBOOT= ${BOOTOBJ}/zfs/libzfsboot.a CFLAGS+= -I${ZFSSRC} CFLAGS+= -DEFI_ZFS_BOOT .endif .if ${COMPILER_TYPE} == "gcc" && ${COMPILER_VERSION} > 40201 CWARNFLAGS.self_reloc.c+= -Wno-error=maybe-uninitialized .endif # We implement a slightly non-standard %S in that it always takes a # CHAR16 that's common in UEFI-land instead of a wchar_t. This only # seems to matter on arm64 where wchar_t defaults to an int instead # of a short. There's no good cast to use here so just ignore the # warnings for now. CWARNFLAGS.main.c+= -Wno-format .PATH: ${.CURDIR}/arch/${MACHINE} # For smbios.c .PATH: ${BOOTSRC}/i386/libi386 .include "${.CURDIR}/arch/${MACHINE}/Makefile.inc" CFLAGS+= -I${.CURDIR} CFLAGS+= -I${.CURDIR}/arch/${MACHINE} CFLAGS+= -I${EFISRC}/include CFLAGS+= -I${EFISRC}/include/${MACHINE} CFLAGS+= -I${SYSDIR}/contrib/dev/acpica/include CFLAGS+= -I${SYSDIR} CFLAGS+= -I${BOOTSRC}/i386/libi386 CFLAGS+= -DNO_PCI -DEFI .if !defined(BOOT_HIDE_SERIAL_NUMBERS) # Export serial numbers, UUID, and asset tag from loader. CFLAGS+= -DSMBIOS_SERIAL_NUMBERS .if defined(BOOT_LITTLE_ENDIAN_UUID) # Use little-endian UUID format as defined in SMBIOS 2.6. CFLAGS+= -DSMBIOS_LITTLE_ENDIAN_UUID .elif defined(BOOT_NETWORK_ENDIAN_UUID) # Use network-endian UUID format for backward compatibility. CFLAGS+= -DSMBIOS_NETWORK_ENDIAN_UUID .endif .endif LOADER_FDT_SUPPORT?= no .if ${MK_FDT} != "no" && ${LOADER_FDT_SUPPORT} != "no" CFLAGS+= -I${BOOTSRC}/fdt CFLAGS+= -I${BOOTSRC}/fdt CFLAGS+= -DLOADER_FDT_SUPPORT LIBEFI_FDT= ${BOOTOBJ}/efi/fdt/libefi_fdt.a LIBFDT= ${BOOTOBJ}/fdt/libfdt.a .endif # Include bcache code. HAVE_BCACHE= yes .if defined(EFI_STAGING_SIZE) CFLAGS+= -DEFI_STAGING_SIZE=${EFI_STAGING_SIZE} .endif # Always add MI sources .include "${BOOTSRC}/loader.mk" FILES+= loader.efi FILESMODE_loader.efi= ${BINMODE} LDSCRIPT= ${.CURDIR}/arch/${MACHINE}/ldscript.${MACHINE} LDFLAGS+= -Wl,-T${LDSCRIPT},-Bsymbolic,-znotext -shared CLEANFILES+= loader.efi NEWVERSWHAT= "EFI loader" ${MACHINE} NM?= nm OBJCOPY?= objcopy .if ${MACHINE_CPUARCH} == "amd64" EFI_TARGET= efi-app-x86_64 .elif ${MACHINE_CPUARCH} == "i386" EFI_TARGET= efi-app-ia32 .else EFI_TARGET= binary .endif # Arbitrarily set the PE/COFF header timestamps to 1 Jan 2016 00:00:00 # for build reproducibility. SOURCE_DATE_EPOCH?=1451606400 loader.efi: ${PROG} if ${NM} ${.ALLSRC} | grep ' U '; then \ echo "Undefined symbols in ${.ALLSRC}"; \ exit 1; \ fi SOURCE_DATE_EPOCH=${SOURCE_DATE_EPOCH} \ ${OBJCOPY} -j .peheader -j .text -j .sdata -j .data \ -j .dynamic -j .dynsym -j .rel.dyn \ -j .rela.dyn -j .reloc -j .eh_frame -j set_Xcommand_set \ -j set_Xficl_compile_set \ --output-target=${EFI_TARGET} ${.ALLSRC} ${.TARGET} LIBEFI= ${BOOTOBJ}/efi/libefi/libefi.a DPADD= ${LIBFICL} ${LIBEFI} ${LIBFDT} ${LIBEFI_FDT} ${LIBZFSBOOT} ${LIBSA} \ ${LDSCRIPT} LDADD= ${LIBFICL} ${LIBEFI} ${LIBFDT} ${LIBEFI_FDT} ${LIBZFSBOOT} ${LIBSA} .include - -beforedepend ${OBJS}: machine - -CLEANFILES+= machine - -machine: .NOMETA - ln -sf ${SYSDIR}/${MACHINE}/include machine - -.if ${MACHINE_CPUARCH} == "amd64" || ${MACHINE_CPUARCH} == "i386" -beforedepend ${OBJS}: x86 -CLEANFILES+= x86 - -x86: .NOMETA - ln -sf ${SYSDIR}/x86/include x86 -.endif Index: projects/runtime-coverage/sys/boot/fdt/Makefile =================================================================== --- projects/runtime-coverage/sys/boot/fdt/Makefile (revision 325498) +++ projects/runtime-coverage/sys/boot/fdt/Makefile (revision 325499) @@ -1,32 +1,28 @@ # $FreeBSD$ .include .PATH: ${SYSDIR}/contrib/libfdt/ LIB= fdt INTERNALLIB= # Vendor sources of libfdt. SRCS+= fdt.c fdt_ro.c fdt_wip.c fdt_sw.c fdt_rw.c fdt_strerror.c \ fdt_empty_tree.c # Loader's fdt commands extension sources. SRCS+= fdt_loader_cmd.c fdt_overlay.c CFLAGS+= -I${SYSDIR}/contrib/libfdt/ -I${LDRSRC} CFLAGS+= -ffreestanding .if ${MACHINE_CPUARCH} == "powerpc" || ${MACHINE_CPUARCH} == "arm" || ${MACHINE_CPUARCH} == "mips" CFLAGS+= -msoft-float .endif -.if ${MACHINE_ARCH} == "powerpc64" -CFLAGS+= -m32 -.endif - CFLAGS+= -Wformat -Wall .include .include Index: projects/runtime-coverage/sys/boot/ficl/Makefile =================================================================== --- projects/runtime-coverage/sys/boot/ficl/Makefile (revision 325498) +++ projects/runtime-coverage/sys/boot/ficl/Makefile (revision 325499) @@ -1,38 +1,35 @@ # $FreeBSD$ # .include .include "${BOOTSRC}/ficl.mk" BASE_SRCS= dict.c ficl.c fileaccess.c float.c loader.c math64.c \ prefix.c search.c stack.c tools.c vm.c words.c SRCS= ${BASE_SRCS} sysdep.c softcore.c CLEANFILES= softcore.c testmain testmain.o -.if ${MACHINE_CPUARCH} == "i386" || \ - (${MACHINE_CPUARCH} == "amd64" && defined(FICL32)) -CFLAGS+= -march=i386 -.endif + .if defined(HAVE_PNP) CFLAGS+= -DHAVE_PNP .endif .include .ifmake testmain CFLAGS+= -DTESTMAIN -D_TESTMAIN SRCS+= testmain.c PROG= testmain .include .else LIB= ficl INTERNALLIB= .include .endif # Standard softwords .PATH: ${FICLSRC}/softwords SOFTWORDS= softcore.fr jhlocal.fr marker.fr freebsd.fr ficllocal.fr \ ifbrack.fr softcore.c: ${SOFTWORDS} softcore.awk (cd ${FICLSRC}/softwords; cat ${SOFTWORDS} \ | awk -f softcore.awk -v datestamp="`LC_ALL=C date`") > ${.TARGET} Index: projects/runtime-coverage/sys/boot/ficl.mk =================================================================== --- projects/runtime-coverage/sys/boot/ficl.mk (revision 325498) +++ projects/runtime-coverage/sys/boot/ficl.mk (revision 325499) @@ -1,44 +1,23 @@ # $FreeBSD$ # Common flags to build FICL related files .include "defs.mk" -.if ${MACHINE_CPUARCH} == "amd64" && defined(FICL32) +.if ${MACHINE_CPUARCH} == "amd64" && ${DO32:U0} == 1 FICL_CPUARCH= i386 .elif ${MACHINE_ARCH:Mmips64*} != "" FICL_CPUARCH= mips64 .else FICL_CPUARCH= ${MACHINE_CPUARCH} .endif .PATH: ${FICLSRC} ${FICLSRC}/${FICL_CPUARCH} -.if ${MACHINE_CPUARCH} == "amd64" -.if defined(FICL32) -CFLAGS+= -m32 -I. -.else +.if ${MACHINE_CPUARCH} == "amd64" && ${DO32:U0} == 0 CFLAGS+= -fPIC .endif -.endif -.if ${MACHINE_ARCH} == "powerpc64" -CFLAGS+= -m32 -mcpu=powerpc -I. -.endif - CFLAGS+= -I${FICLSRC} -I${FICLSRC}/${FICL_CPUARCH} -I${LDRSRC} CFLAGS+= -DBOOT_FORTH CFLAGS+= -DBF_DICTSIZE=15000 - -.if ${MACHINE_CPUARCH} == "amd64" && defined(FICL32) -.if !exists(machine) -${SRCS:M*.c:R:S/$/.o/g}: machine - -beforedepend ${OBJS}: machine -.endif - -machine: .NOMETA - ln -sf ${SYSDIR}/i386/include machine - -CLEANFILES+= machine -.endif Index: projects/runtime-coverage/sys/boot/ficl32/Makefile =================================================================== --- projects/runtime-coverage/sys/boot/ficl32/Makefile (revision 325498) +++ projects/runtime-coverage/sys/boot/ficl32/Makefile (revision 325499) @@ -1,5 +1,5 @@ # $FreeBSD$ -FICL32= +DO32=1 .include "${.CURDIR}/../ficl/Makefile" Index: projects/runtime-coverage/sys/boot/geli/Makefile =================================================================== --- projects/runtime-coverage/sys/boot/geli/Makefile (revision 325498) +++ projects/runtime-coverage/sys/boot/geli/Makefile (revision 325499) @@ -1,54 +1,53 @@ # $FreeBSD$ # libgeliboot MAN= -.include -MK_SSP= no .include +MK_SSP= no LIB= geliboot INTERNALLIB= MK_PROFILE= no NO_PIC= .if ${MACHINE_CPUARCH} == "i386" || ${MACHINE_CPUARCH} == "amd64" CFLAGS+= -march=i386 .endif -.if ${MACHINE_ARCH} == "amd64" || ${MACHINE_ARCH} == "powerpc64" +.if ${MACHINE_ARCH} == "amd64" CFLAGS+= -m32 .endif WARNS?= 0 # string functions from libc .PATH: ${SRCTOP}/lib/libc/string SRCS+= bcmp.c bcopy.c bzero.c # Our password input method SRCS+= pwgets.c # sha256 and sha512 from sys/crypto .PATH: ${SYSDIR}/crypto/sha2 CFLAGS+= -DWEAK_REFS SRCS+= sha256c.c sha512c.c # md5 from libmd .PATH: ${SRCTOP}/lib/libmd SRCS+= md5c.c # AES implementation from sys/crypto .PATH: ${SYSDIR}/crypto/rijndael CFLAGS+= -I${SYSDIR} CFLAGS+= -I${LDRSRC} # Remove asserts CFLAGS+= -DNDEBUG SRCS+= rijndael-alg-fst.c rijndael-api-fst.c rijndael-api.c # local GELI Implementation .PATH: ${SYSDIR}/geom/eli CFLAGS+= -D_STAND SRCS+= geliboot_crypto.c g_eli_hmac.c g_eli_key.c g_eli_key_cache.c pkcs5v2.c .include .include Index: projects/runtime-coverage/sys/boot/i386/Makefile =================================================================== --- projects/runtime-coverage/sys/boot/i386/Makefile (revision 325498) +++ projects/runtime-coverage/sys/boot/i386/Makefile (revision 325499) @@ -1,19 +1,19 @@ # $FreeBSD$ -.include +.include SUBDIR= mbr pmbr boot0 boot0sio btx boot2 cdboot gptboot \ libi386 libfirewire loader # special boot programs, 'self-extracting boot2+loader' SUBDIR+= pxeldr .if ${MACHINE_CPUARCH} == "i386" SUBDIR+= kgzldr .endif .if ${MK_ZFS} != "no" SUBDIR+= zfsboot gptzfsboot zfsloader .endif .include Index: projects/runtime-coverage/sys/boot/i386/Makefile.inc =================================================================== --- projects/runtime-coverage/sys/boot/i386/Makefile.inc (revision 325498) +++ projects/runtime-coverage/sys/boot/i386/Makefile.inc (revision 325499) @@ -1,35 +1,31 @@ # Common defines for all of /sys/boot/i386/ # # $FreeBSD$ LOADER_ADDRESS?=0x200000 CFLAGS+= -march=i386 -ffreestanding CFLAGS.gcc+= -mpreferred-stack-boundary=2 CFLAGS+= ${CFLAGS_NO_SIMD} -msoft-float LDFLAGS+= -nostdlib # BTX components BTXDIR= ${BOOTOBJ}/i386/btx BTXLDR= ${BTXDIR}/btxldr/btxldr BTXKERN= ${BTXDIR}/btx/btx BTXCRT= ${BTXDIR}/lib/crt0.o BTXSRC= ${BOOTSRC}/i386/btx BTXLIB= ${BTXSRC}/lib # compact binary with no padding between text, data, bss LDSCRIPT= ${BOOTSRC}/i386/boot.ldscript # LDFLAGS_BIN=-e start -Ttext ${ORG} -Wl,-T,${LDSCRIPT},-S,--oformat,binary # LD_FLAGS_BIN=-static -T ${LDSCRIPT} --gc-sections LDFLAGS_BIN=-e start -Ttext ${ORG} -Wl,-N,-S,--oformat,binary LD_FLAGS_BIN=-static -N --gc-sections .if ${MACHINE_CPUARCH} == "amd64" -CFLAGS+= -m32 -ACFLAGS+= -m32 -# LD_FLAGS is passed directly to ${LD}, not via ${CC}: -LD_FLAGS+= -m elf_i386_fbsd -AFLAGS+= --32 +DO32=1 .endif .include "../Makefile.inc" Index: projects/runtime-coverage/sys/boot/i386/boot2/Makefile =================================================================== --- projects/runtime-coverage/sys/boot/i386/boot2/Makefile (revision 325498) +++ projects/runtime-coverage/sys/boot/i386/boot2/Makefile (revision 325499) @@ -1,106 +1,99 @@ # $FreeBSD$ .include FILES= boot boot1 boot2 NM?= nm # A value of 0x80 enables LBA support. BOOT_BOOT1_FLAGS?= 0x80 BOOT_COMCONSOLE_PORT?= 0x3f8 BOOT_COMCONSOLE_SPEED?= 9600 B2SIOFMT?= 0x3 REL1= 0x700 ORG1= 0x7c00 ORG2= 0x2000 # Decide level of UFS support. BOOT2_UFS?= UFS1_AND_UFS2 #BOOT2_UFS?= UFS2_ONLY #BOOT2_UFS?= UFS1_ONLY CFLAGS+=-fomit-frame-pointer \ -mrtd \ -mregparm=3 \ -D${BOOT2_UFS} \ -DFLAGS=${BOOT_BOOT1_FLAGS} \ -DSIOPRT=${BOOT_COMCONSOLE_PORT} \ -DSIOFMT=${B2SIOFMT} \ -DSIOSPD=${BOOT_COMCONSOLE_SPEED} \ -I${LDRSRC} \ -I${BTXLIB} -I. \ -Wall -Waggregate-return -Wbad-function-cast -Wno-cast-align \ -Wmissing-declarations -Wmissing-prototypes -Wnested-externs \ -Wpointer-arith -Wshadow -Wstrict-prototypes -Wwrite-strings \ -Winline CFLAGS.gcc+= -Os \ -fno-asynchronous-unwind-tables \ --param max-inline-insns-single=100 .if ${COMPILER_TYPE} == "gcc" && ${COMPILER_VERSION} <= 40201 CFLAGS.gcc+= -mno-align-long-strings .endif CFLAGS.clang+= -Oz ${CLANG_OPT_SMALL} LD_FLAGS+=${LD_FLAGS_BIN} CLEANFILES= boot boot: boot1 boot2 cat boot1 boot2 > boot CLEANFILES+= boot1 boot1.out boot1.o boot1: boot1.out ${OBJCOPY} -S -O binary boot1.out ${.TARGET} boot1.out: boot1.o ${LD} ${LD_FLAGS} -e start -Ttext ${ORG1} -o ${.TARGET} boot1.o CLEANFILES+= boot2 boot2.ld boot2.ldr boot2.bin boot2.out boot2.o \ boot2.h sio.o BOOT2SIZE= 7680 boot2: boot2.ld @set -- `ls -l ${.ALLSRC}`; x=$$((${BOOT2SIZE}-$$5)); \ echo "$$x bytes available"; test $$x -ge 0 ${DD} if=${.ALLSRC} of=${.TARGET} obs=${BOOT2SIZE} conv=osync boot2.ld: boot2.ldr boot2.bin ${BTXKERN} btxld -v -E ${ORG2} -f bin -b ${BTXKERN} -l boot2.ldr \ -o ${.TARGET} -P 1 boot2.bin boot2.ldr: ${DD} if=/dev/zero of=${.TARGET} bs=512 count=1 boot2.bin: boot2.out ${OBJCOPY} -S -O binary boot2.out ${.TARGET} boot2.out: ${BTXCRT} boot2.o sio.o ${LD} ${LD_FLAGS} -Ttext ${ORG2} -o ${.TARGET} ${.ALLSRC} SRCS= boot2.c boot2.h boot2.h: boot1.out ${NM} -t d ${.ALLSRC} | awk '/([0-9])+ T xread/ \ { x = $$1 - ORG1; \ printf("#define XREADORG %#x\n", REL1 + x) }' \ ORG1=`printf "%d" ${ORG1}` \ REL1=`printf "%d" ${REL1}` > ${.TARGET} -.if ${MACHINE_CPUARCH} == "amd64" -beforedepend boot2.s: machine -CLEANFILES+= machine -machine: ${SYSDIR}/i386/include .NOMETA - ln -sf ${.ALLSRC} ${.TARGET} -.endif - .include # XXX: clang integrated-as doesn't grok .codeNN directives yet CFLAGS.boot1.S= ${CLANG_NO_IAS} Index: projects/runtime-coverage/sys/boot/i386/gptboot/Makefile =================================================================== --- projects/runtime-coverage/sys/boot/i386/gptboot/Makefile (revision 325498) +++ projects/runtime-coverage/sys/boot/i386/gptboot/Makefile (revision 325499) @@ -1,89 +1,82 @@ # $FreeBSD$ .include .PATH: ${BOOTSRC}/i386/boot2 ${BOOTSRC}/i386/common ${SASRC} FILES= gptboot MAN= gptboot.8 NM?= nm BOOT_COMCONSOLE_PORT?= 0x3f8 BOOT_COMCONSOLE_SPEED?= 9600 B2SIOFMT?= 0x3 REL1= 0x700 ORG1= 0x7c00 ORG2= 0x0 # Decide level of UFS support. GPTBOOT_UFS?= UFS1_AND_UFS2 #GPTBOOT_UFS?= UFS2_ONLY #GPTBOOT_UFS?= UFS1_ONLY CFLAGS+=-DBOOTPROG=\"gptboot\" \ -O1 \ -DGPT \ -D${GPTBOOT_UFS} \ -DSIOPRT=${BOOT_COMCONSOLE_PORT} \ -DSIOFMT=${B2SIOFMT} \ -DSIOSPD=${BOOT_COMCONSOLE_SPEED} \ -I${LDRSRC} \ -I${BOOTSRC}/i386/common \ -I${BTXLIB} -I. \ -I${BOOTSRC}/i386/boot2 \ -I${SYSDIR} \ -Wall -Waggregate-return -Wbad-function-cast -Wno-cast-align \ -Wmissing-declarations -Wmissing-prototypes -Wnested-externs \ -Wpointer-arith -Wshadow -Wstrict-prototypes -Wwrite-strings \ -Winline -Wno-pointer-sign CFLAGS.gcc+= --param max-inline-insns-single=100 .if ${LOADER_GELI_SUPPORT:Uyes} == "yes" CFLAGS+= -DLOADER_GELI_SUPPORT CFLAGS+= -I${BOOTSRC}/geli LIBGELIBOOT= ${BOOTOBJ}/geli/libgeliboot.a .PATH: ${SYSDIR}/opencrypto OPENCRYPTO_XTS= xform_aes_xts.o .endif LD_FLAGS+=${LD_FLAGS_BIN} CLEANFILES= gptboot gptboot: gptldr.bin gptboot.bin ${BTXKERN} btxld -v -E ${ORG2} -f bin -b ${BTXKERN} -l gptldr.bin \ -o ${.TARGET} gptboot.bin CLEANFILES+= gptldr.bin gptldr.out gptldr.o gptldr.bin: gptldr.out ${OBJCOPY} -S -O binary gptldr.out ${.TARGET} gptldr.out: gptldr.o ${LD} ${LD_FLAGS} -e start -Ttext ${ORG1} -o ${.TARGET} gptldr.o CLEANFILES+= gptboot.bin gptboot.out gptboot.o sio.o crc32.o drv.o \ cons.o util.o ${OPENCRYPTO_XTS} gptboot.bin: gptboot.out ${OBJCOPY} -S -O binary gptboot.out ${.TARGET} gptboot.out: ${BTXCRT} gptboot.o sio.o crc32.o drv.o cons.o util.o ${OPENCRYPTO_XTS} ${LD} ${LD_FLAGS} -Ttext ${ORG2} -o ${.TARGET} ${.ALLSRC} ${LIBGELIBOOT} ${LIBSA32} gptboot.o: ${SASRC}/ufsread.c -.if ${MACHINE_CPUARCH} == "amd64" -beforedepend gptboot.o: machine -CLEANFILES+= machine -machine: .NOMETA - ln -sf ${SYSDIR}/i386/include machine -.endif - .include # XXX: clang integrated-as doesn't grok .codeNN directives yet CFLAGS.gptldr.S= ${CLANG_NO_IAS} Index: projects/runtime-coverage/sys/boot/i386/gptzfsboot/Makefile =================================================================== --- projects/runtime-coverage/sys/boot/i386/gptzfsboot/Makefile (revision 325498) +++ projects/runtime-coverage/sys/boot/i386/gptzfsboot/Makefile (revision 325499) @@ -1,101 +1,94 @@ # $FreeBSD$ .include .PATH: ${BOOTSRC}/i386/boot2 ${BOOTSRC}/i386/gptboot \ ${BOOTSRC}/i386/zfsboot ${BOOTSRC}/i386/common \ ${SASRC} FILES= gptzfsboot MAN= gptzfsboot.8 NM?= nm BOOT_COMCONSOLE_PORT?= 0x3f8 BOOT_COMCONSOLE_SPEED?= 9600 B2SIOFMT?= 0x3 REL1= 0x700 ORG1= 0x7c00 ORG2= 0x0 CFLAGS+=-DBOOTPROG=\"gptzfsboot\" \ -O1 \ -DGPT -DZFS -DBOOT2 \ -DSIOPRT=${BOOT_COMCONSOLE_PORT} \ -DSIOFMT=${B2SIOFMT} \ -DSIOSPD=${BOOT_COMCONSOLE_SPEED} \ -I${LDRSRC} \ -I${BOOTSRC}/i386/common \ -I${ZFSSRC} \ -I${SYSDIR}/cddl/boot/zfs \ -I${BOOTSRC}/i386/btx/lib -I. \ -I${BOOTSRC}/i386/boot2 \ -I${SYSDIR} \ -Wall -Waggregate-return -Wbad-function-cast \ -Wmissing-declarations -Wmissing-prototypes -Wnested-externs \ -Wpointer-arith -Wshadow -Wstrict-prototypes -Wwrite-strings \ -Winline -Wno-pointer-sign NO_WCAST_ALIGN= .if ${COMPILER_TYPE} == "clang" || \ (${COMPILER_TYPE} == "gcc" && ${COMPILER_VERSION} > 40201) CFLAGS+= -Wno-tentative-definition-incomplete-type .endif .if ${MACHINE} == "amd64" LIBZFSBOOT=${BOOTOBJ}/zfs32/libzfsboot.a .else LIBZFSBOOT=${BOOTOBJ}/zfs/libzfsboot.a .endif .if ${LOADER_GELI_SUPPORT:Uyes} == "yes" CFLAGS+= -DLOADER_GELI_SUPPORT CFLAGS+= -I${BOOTSRC}/geli LIBGELIBOOT= ${BOOTOBJ}/geli/libgeliboot.a .PATH: ${SYSDIR}/opencrypto OPENCRYPTO_XTS= xform_aes_xts.o .endif CFLAGS.gcc+= --param max-inline-insns-single=100 LD_FLAGS+=${LD_FLAGS_BIN} CLEANFILES= gptzfsboot gptzfsboot: gptldr.bin gptzfsboot.bin ${BTXKERN} btxld -v -E ${ORG2} -f bin -b ${BTXKERN} -l gptldr.bin \ -o ${.TARGET} gptzfsboot.bin CLEANFILES+= gptldr.bin gptldr.out gptldr.o gptldr.bin: gptldr.out ${OBJCOPY} -S -O binary gptldr.out ${.TARGET} gptldr.out: gptldr.o ${LD} ${LD_FLAGS} -e start -Ttext ${ORG1} -o ${.TARGET} gptldr.o CLEANFILES+= gptzfsboot.bin gptzfsboot.out zfsboot.o sio.o cons.o \ drv.o gpt.o util.o ${OPENCRYPTO_XTS} gptzfsboot.bin: gptzfsboot.out ${OBJCOPY} -S -O binary gptzfsboot.out ${.TARGET} gptzfsboot.out: ${BTXCRT} zfsboot.o sio.o gpt.o drv.o cons.o util.o \ ${OPENCRYPTO_XTS} ${LD} ${LD_FLAGS} -Ttext ${ORG2} -o ${.TARGET} ${.ALLSRC} ${LIBGELIBOOT} ${LIBZFSBOOT} ${LIBSA32} zfsboot.o: ${ZFSSRC}/zfsimpl.c -.if ${MACHINE_CPUARCH} == "amd64" -beforedepend zfsboot.o: machine -CLEANFILES+= machine -machine: .NOMETA - ln -sf ${SYSDIR}/i386/include machine -.endif - .include # XXX: clang integrated-as doesn't grok .codeNN directives yet CFLAGS.gptldr.S= ${CLANG_NO_IAS} Index: projects/runtime-coverage/sys/boot/i386/libfirewire/Makefile =================================================================== --- projects/runtime-coverage/sys/boot/i386/libfirewire/Makefile (revision 325498) +++ projects/runtime-coverage/sys/boot/i386/libfirewire/Makefile (revision 325499) @@ -1,31 +1,20 @@ # $FreeBSD$ .include LIB= firewire INTERNALLIB= .PATH: ${SYSDIR}/dev/dcons ${SYSDIR}/dev/firewire SRCS+= firewire.c fwohci.c dconsole.c SRCS+= dcons.c fwcrom.c CFLAGS+= -D_BOOT CFLAGS+= -I${LDRSRC} -I${SYSDIR} -I. CFLAGS+= -I${BTXLIB} CFLAGS+= -I${BOOTSRC}/i386/libi386 CFLAGS+= -Wformat -Wall -.if ${MACHINE_CPUARCH} == "amd64" -CLEANFILES+= machine -machine: .NOMETA - ln -sf ${SYSDIR}/i386/include machine -.endif - .include - -.if ${MACHINE_CPUARCH} == "amd64" -beforedepend ${OBJS}: machine -.endif - Index: projects/runtime-coverage/sys/boot/i386/libi386/Makefile =================================================================== --- projects/runtime-coverage/sys/boot/i386/libi386/Makefile (revision 325498) +++ projects/runtime-coverage/sys/boot/i386/libi386/Makefile (revision 325499) @@ -1,75 +1,65 @@ # $FreeBSD$ .include LIB= i386 INTERNALLIB= SRCS= biosacpi.c bioscd.c biosdisk.c biosmem.c biospnp.c \ biospci.c biossmap.c bootinfo.c bootinfo32.c bootinfo64.c \ comconsole.c devicename.c elf32_freebsd.c \ elf64_freebsd.c multiboot.c multiboot_tramp.S relocater_tramp.S \ i386_copy.c i386_module.c nullconsole.c pxe.c pxetramp.s \ smbios.c time.c vidconsole.c amd64_tramp.S spinconsole.c .PATH: ${ZFSSRC} SRCS+= devicename_stubs.c BOOT_COMCONSOLE_PORT?= 0x3f8 CFLAGS+= -DCOMPORT=${BOOT_COMCONSOLE_PORT} BOOT_COMCONSOLE_SPEED?= 9600 CFLAGS+= -DCOMSPEED=${BOOT_COMCONSOLE_SPEED} .ifdef(BOOT_BIOSDISK_DEBUG) # Make the disk code more talkative CFLAGS+= -DDISK_DEBUG .endif .if ${LOADER_GELI_SUPPORT:Uyes} == "yes" # Decrypt encrypted drives CFLAGS+= -DLOADER_GELI_SUPPORT CFLAGS+= -I${BOOTSRC}/geli .endif .if !defined(BOOT_HIDE_SERIAL_NUMBERS) # Export serial numbers, UUID, and asset tag from loader. CFLAGS+= -DSMBIOS_SERIAL_NUMBERS .if defined(BOOT_LITTLE_ENDIAN_UUID) # Use little-endian UUID format as defined in SMBIOS 2.6. CFLAGS+= -DSMBIOS_LITTLE_ENDIAN_UUID .elif defined(BOOT_NETWORK_ENDIAN_UUID) # Use network-endian UUID format for backward compatibility. CFLAGS+= -DSMBIOS_NETWORK_ENDIAN_UUID .endif .endif # Include simple terminal emulation (cons25-compatible) CFLAGS+= -DTERM_EMU # XXX: make alloca() useable CFLAGS+= -Dalloca=__builtin_alloca CFLAGS+= -I${BOOTSRC}/ficl -I${BOOTSRC}/ficl/i386 \ -I${LDRSRC} -I${BOOTSRC}/i386/common \ -I${BTXLIB} \ -I${SYSDIR}/contrib/dev/acpica/include \ -I${SYSDIR} -I. # Handle FreeBSD specific %b and %D printf format specifiers CFLAGS+= ${FORMAT_EXTENSIONS} -.if ${MACHINE_CPUARCH} == "amd64" -CLEANFILES+= machine -machine: .NOMETA - ln -sf ${SYSDIR}/i386/include machine -.endif - .include # XXX: clang integrated-as doesn't grok .codeNN directives yet CFLAGS.amd64_tramp.S= ${CLANG_NO_IAS} CFLAGS.multiboot_tramp.S= ${CLANG_NO_IAS} - -.if ${MACHINE_CPUARCH} == "amd64" -beforedepend ${OBJS}: machine -.endif Index: projects/runtime-coverage/sys/boot/i386/loader/Makefile =================================================================== --- projects/runtime-coverage/sys/boot/i386/loader/Makefile (revision 325498) +++ projects/runtime-coverage/sys/boot/i386/loader/Makefile (revision 325499) @@ -1,122 +1,111 @@ # $FreeBSD$ .include MK_SSP= no LOADER?= loader PROG= ${LOADER}.sym MAN= INTERNALPROG= NEWVERSWHAT?= "bootstrap loader" x86 VERSION_FILE= ${.CURDIR}/../loader/version LOADER_NET_SUPPORT?= yes LOADER_NFS_SUPPORT?= yes LOADER_TFTP_SUPPORT?= yes LOADER_CD9660_SUPPORT?= no LOADER_EXT2FS_SUPPORT?= no LOADER_MSDOS_SUPPORT?= no LOADER_UFS_SUPPORT?= yes LOADER_GZIP_SUPPORT?= yes LOADER_BZIP2_SUPPORT?= yes # architecture-specific loader code SRCS= main.c conf.c vers.c chain.c # Put LOADER_FIREWIRE_SUPPORT=yes in /etc/make.conf for FireWire/dcons support .if defined(LOADER_FIREWIRE_SUPPORT) CFLAGS+= -DLOADER_FIREWIRE_SUPPORT LIBFIREWIRE= ${BOOTOBJ}/i386/libfirewire/libfirewire.a .endif # Set by zfsloader Makefile .if defined(LOADER_ZFS_SUPPORT) CFLAGS+= -DLOADER_ZFS_SUPPORT .if ${MACHINE} == "amd64" LIBZFSBOOT= ${BOOTOBJ}/zfs32/libzfsboot.a .else LIBZFSBOOT= ${BOOTOBJ}/zfs/libzfsboot.a .endif .endif # Include bcache code. HAVE_BCACHE= yes # Enable PnP and ISA-PnP code. HAVE_PNP= yes HAVE_ISABUS= yes .if ${LOADER_GELI_SUPPORT:Uyes} == "yes" CFLAGS+= -DLOADER_GELI_SUPPORT CFLAGS+= -I${BOOTSRC}/geli LIBGELIBOOT= ${BOOTOBJ}/geli/libgeliboot.a .PATH: ${SYSDIR}/opencrypto SRCS+= xform_aes_xts.c CFLAGS+= -I${SYSDIR} -D_STAND .endif # Always add MI sources .include "${BOOTSRC}/loader.mk" CFLAGS+= -I. CLEANFILES= ${LOADER} ${LOADER}.bin loader.help CFLAGS+= -Wall LDFLAGS+= -static -Ttext 0x0 # i386 standalone support library LIBI386= ${BOOTOBJ}/i386/libi386/libi386.a CFLAGS+= -I${BOOTSRC}/i386 # BTX components CFLAGS+= -I${BTXLIB} # Debug me! #CFLAGS+= -g #LDFLAGS+= -g ${LOADER}: ${LOADER}.bin ${BTXLDR} ${BTXKERN} btxld -v -f aout -e ${LOADER_ADDRESS} -o ${.TARGET} -l ${BTXLDR} \ -b ${BTXKERN} ${LOADER}.bin ${LOADER}.bin: ${LOADER}.sym strip -R .comment -R .note -o ${.TARGET} ${.ALLSRC} loader.help: help.common help.i386 cat ${.ALLSRC} | awk -f ${LDRSRC}/merge_help.awk > ${.TARGET} FILES= ${LOADER} # XXX INSTALLFLAGS_loader= -b FILESMODE_${LOADER}= ${BINMODE} -b .if !defined(LOADER_ONLY) .PATH: ${BOOTSRC}/forth .include "${BOOTSRC}/forth/Makefile.inc" FILES+= pcibios.4th FILES+= loader.rc menu.rc .endif # XXX crt0.o needs to be first for pxeboot(8) to work OBJS= ${BTXCRT} -# XXX not happy with this construct, revisit -.if ${MACHINE_CPUARCH} == "amd64" -FICL= ${LIBFICL32} -.else -FICL= ${LIBFICL} -.endif +DPADD= ${LIBFICL32} ${LIBFIREWIRE} ${LIBZFSBOOT} ${LIBI386} ${LIBGELIBOOT} ${LIBSA32} +LDADD= ${LIBFICL32} ${LIBFIREWIRE} ${LIBZFSBOOT} ${LIBI386} ${LIBGELIBOOT} ${LIBSA32} -DPADD= ${FICL} ${LIBFIREWIRE} ${LIBZFSBOOT} ${LIBI386} ${LIBGELIBOOT} ${LIBSA32} -LDADD= ${FICL} ${LIBFIREWIRE} ${LIBZFSBOOT} ${LIBI386} ${LIBGELIBOOT} ${LIBSA32} - -.include - .if ${MACHINE_CPUARCH} == "amd64" -beforedepend ${OBJS}: machine -CLEANFILES+= machine CFLAGS+= -DLOADER_PREFER_AMD64 -machine: .NOMETA - ln -sf ${SYSDIR}/i386/include machine .endif + +.include Index: projects/runtime-coverage/sys/boot/i386/zfsboot/Makefile =================================================================== --- projects/runtime-coverage/sys/boot/i386/zfsboot/Makefile (revision 325498) +++ projects/runtime-coverage/sys/boot/i386/zfsboot/Makefile (revision 325499) @@ -1,100 +1,93 @@ # $FreeBSD$ LOADER_GELI_SUPPORT=no .include .PATH: ${BOOTSRC}/i386/boot2 ${BOOTSRC}/i386/common ${SASRC} FILES= zfsboot MAN= zfsboot.8 NM?= nm BOOT_COMCONSOLE_PORT?= 0x3f8 BOOT_COMCONSOLE_SPEED?= 9600 B2SIOFMT?= 0x3 REL1= 0x700 ORG1= 0x7c00 ORG2= 0x2000 CFLAGS+=-DBOOTPROG=\"zfsboot\" \ -O1 \ -DZFS -DBOOT2 \ -DSIOPRT=${BOOT_COMCONSOLE_PORT} \ -DSIOFMT=${B2SIOFMT} \ -DSIOSPD=${BOOT_COMCONSOLE_SPEED} \ -I${LDRSRC} \ -I${BOOTSRC}/i386/common \ -I${BOOTSRC}/i386 \ -I${ZFSSRC} \ -I${SYSDIR}/cddl/boot/zfs \ -I${BTXLIB} -I. \ -I${BOOTSRC}/i386/boot2 \ -Wall -Waggregate-return -Wbad-function-cast -Wno-cast-align \ -Wmissing-declarations -Wmissing-prototypes -Wnested-externs \ -Wpointer-arith -Wshadow -Wstrict-prototypes -Wwrite-strings \ -Winline CFLAGS.gcc+= --param max-inline-insns-single=100 .if ${MACHINE} == "amd64" LIBZFSBOOT=${BOOTOBJ}/zfs32/libzfsboot.a .else LIBZFSBOOT=${BOOTOBJ}/zfs/libzfsboot.a .endif LD_FLAGS+=${LD_FLAGS_BIN} CLEANFILES= zfsboot zfsboot: zfsboot1 zfsboot2 cat zfsboot1 zfsboot2 > zfsboot CLEANFILES+= zfsboot1 zfsldr.out zfsldr.o zfsboot1: zfsldr.out ${OBJCOPY} -S -O binary zfsldr.out ${.TARGET} zfsldr.out: zfsldr.o ${LD} ${LD_FLAGS} -e start -Ttext ${ORG1} -o ${.TARGET} zfsldr.o CLEANFILES+= zfsboot2 zfsboot.ld zfsboot.ldr zfsboot.bin zfsboot.out \ zfsboot.o zfsboot.s zfsboot.s.tmp sio.o cons.o drv.o util.o # We currently allow 128k bytes for zfsboot - in practice it could be # any size up to 3.5Mb but keeping it fixed size simplifies zfsldr. # BOOT2SIZE= 131072 zfsboot2: zfsboot.ld @set -- `ls -l ${.ALLSRC}`; x=$$((${BOOT2SIZE}-$$5)); \ echo "$$x bytes available"; test $$x -ge 0 ${DD} if=${.ALLSRC} of=${.TARGET} obs=${BOOT2SIZE} conv=osync zfsboot.ld: zfsboot.ldr zfsboot.bin ${BTXKERN} btxld -v -E ${ORG2} -f bin -b ${BTXKERN} -l zfsboot.ldr \ -o ${.TARGET} -P 1 zfsboot.bin zfsboot.ldr: cp /dev/null ${.TARGET} zfsboot.bin: zfsboot.out ${OBJCOPY} -S -O binary zfsboot.out ${.TARGET} zfsboot.out: ${BTXCRT} zfsboot.o sio.o drv.o cons.o util.o ${LD} ${LD_FLAGS} -Ttext ${ORG2} -o ${.TARGET} ${.ALLSRC} ${LIBZFSBOOT} ${LIBSA32} SRCS= zfsboot.c -.if ${MACHINE_CPUARCH} == "amd64" -beforedepend zfsboot.o: machine -CLEANFILES+= machine -machine: .NOMETA - ln -sf ${SYSDIR}/i386/include machine -.endif - .include # XXX: clang integrated-as doesn't grok .codeNN directives yet CFLAGS.zfsldr.S= ${CLANG_NO_IAS} Index: projects/runtime-coverage/sys/boot/libsa/Makefile =================================================================== --- projects/runtime-coverage/sys/boot/libsa/Makefile (revision 325498) +++ projects/runtime-coverage/sys/boot/libsa/Makefile (revision 325499) @@ -1,161 +1,159 @@ # $FreeBSD$ # Originally from $NetBSD: Makefile,v 1.21 1997/10/26 22:08:38 lukem Exp $ # # Notes: # - We don't use the libc strerror/sys_errlist because the string table is # quite large. # MK_PROFILE= no MK_SSP= no -.include - -.include "../Makefile.inc" +.include INTERNALLIB= LIBSA_CPUARCH?=${MACHINE_CPUARCH} LIBC_SRC= ${SRCTOP}/lib/libc LIB?= sa NO_PIC= WARNS?= 0 CFLAGS+= -I${SASRC} -I${SYSDIR} # standalone components and stuff we have modified locally SRCS+= gzguts.h zutil.h __main.c assert.c bcd.c environment.c getopt.c gets.c \ globals.c pager.c panic.c printf.c strdup.c strerror.c strtol.c strtoul.c \ random.c sbrk.c twiddle.c zalloc.c zalloc_malloc.c # private (pruned) versions of libc string functions SRCS+= strcasecmp.c .PATH: ${LIBC_SRC}/net SRCS+= ntoh.c # string functions from libc .PATH: ${LIBC_SRC}/string SRCS+= bcmp.c bcopy.c bzero.c ffs.c fls.c \ memccpy.c memchr.c memcmp.c memcpy.c memmove.c memset.c \ qdivrem.c strcat.c strchr.c strcmp.c strcpy.c stpcpy.c stpncpy.c \ strcspn.c strlcat.c strlcpy.c strlen.c strncat.c strncmp.c strncpy.c \ strpbrk.c strrchr.c strsep.c strspn.c strstr.c strtok.c swab.c .if ${MACHINE_CPUARCH} == "arm" .PATH: ${LIBC_SRC}/arm/gen # Do not generate movt/movw, because the relocation fixup for them does not # translate to the -Bsymbolic -pie format required by self_reloc() in loader(8). # Also, the fpu is not available in a standalone environment. .if ${COMPILER_VERSION} < 30800 CFLAGS.clang+= -mllvm -arm-use-movt=0 .else CFLAGS.clang+= -mno-movt .endif CFLAGS.clang+= -mfpu=none # Compiler support functions .PATH: ${SRCTOP}/contrib/compiler-rt/lib/builtins/ # __clzsi2 and ctzsi2 for various builtin functions SRCS+= clzsi2.c ctzsi2.c # Divide and modulus functions called by the compiler SRCS+= divmoddi4.c divmodsi4.c divdi3.c divsi3.c moddi3.c modsi3.c SRCS+= udivmoddi4.c udivmodsi4.c udivdi3.c udivsi3.c umoddi3.c umodsi3.c .PATH: ${SRCTOP}/contrib/compiler-rt/lib/builtins/arm/ SRCS+= aeabi_idivmod.S aeabi_ldivmod.S aeabi_uidivmod.S aeabi_uldivmod.S SRCS+= aeabi_memcmp.S aeabi_memcpy.S aeabi_memmove.S aeabi_memset.S .endif .if ${MACHINE_CPUARCH} == "aarch64" || ${MACHINE_CPUARCH} == "riscv" .PATH: ${LIBC_SRC}/${MACHINE_CPUARCH}/gen .endif .if ${MACHINE_CPUARCH} == "powerpc" .PATH: ${LIBC_SRC}/quad SRCS+= ashldi3.c ashrdi3.c SRCS+= syncicache.c .endif # uuid functions from libc .PATH: ${LIBC_SRC}/uuid SRCS+= uuid_create_nil.c uuid_equal.c uuid_from_string.c uuid_is_nil.c uuid_to_string.c # _setjmp/_longjmp .PATH: ${SASRC}/${LIBSA_CPUARCH} SRCS+= _setjmp.S # decompression functionality from libbz2 # NOTE: to actually test this functionality after libbz2 upgrade compile # loader(8) with LOADER_BZIP2_SUPPORT defined .PATH: ${SRCTOP}/contrib/bzip2 CFLAGS+= -DBZ_NO_STDIO -DBZ_NO_COMPRESS SRCS+= libsa_bzlib_private.h .for file in bzlib.c crctable.c decompress.c huffman.c randtable.c SRCS+= _${file} CLEANFILES+= _${file} _${file}: ${file} sed "s|bzlib_private\.h|libsa_bzlib_private.h|" \ ${.ALLSRC} > ${.TARGET} .endfor CLEANFILES+= libsa_bzlib_private.h libsa_bzlib_private.h: bzlib_private.h sed -e 's||"stand.h"|' \ ${.ALLSRC} > ${.TARGET} # decompression functionality from zlib .PATH: ${SRCTOP}/contrib/zlib CFLAGS+=-DHAVE_MEMCPY -I${SRCTOP}/contrib/zlib SRCS+= adler32.c crc32.c libsa_zutil.h libsa_gzguts.h .for file in infback.c inffast.c inflate.c inftrees.c zutil.c SRCS+= _${file} CLEANFILES+= _${file} _${file}: ${file} sed -e "s|zutil\.h|libsa_zutil.h|" \ -e "s|gzguts\.h|libsa_gzguts.h|" \ ${.ALLSRC} > ${.TARGET} .endfor # depend on stand.h being able to be included multiple times .for file in zutil.h gzguts.h CLEANFILES+= libsa_${file} libsa_${file}: ${file} sed -e 's||"stand.h"|' \ -e 's||"stand.h"|' \ -e 's||"stand.h"|' \ -e 's||"stand.h"|' \ -e 's||"stand.h"|' \ ${.ALLSRC} > ${.TARGET} .endfor # io routines SRCS+= closeall.c dev.c ioctl.c nullfs.c stat.c \ fstat.c close.c lseek.c open.c read.c write.c readdir.c # network routines SRCS+= arp.c ether.c ip.c inet_ntoa.c in_cksum.c net.c udp.c netif.c rpc.c # network info services: SRCS+= bootp.c rarp.c bootparam.c # boot filesystems SRCS+= ufs.c nfs.c cd9660.c tftp.c gzipfs.c bzipfs.c SRCS+= dosfs.c ext2fs.c SRCS+= splitfs.c SRCS+= pkgfs.c .if ${MK_NAND} != "no" SRCS+= nandfs.c .endif # explicit_bzero .PATH: ${SYSDIR}/libkern SRCS+= explicit_bzero.c .include .include Index: projects/runtime-coverage/sys/boot/libsa32/Makefile =================================================================== --- projects/runtime-coverage/sys/boot/libsa32/Makefile (revision 325498) +++ projects/runtime-coverage/sys/boot/libsa32/Makefile (revision 325499) @@ -1,24 +1,13 @@ # $FreeBSD$ +DO32=1 + .include LIB=sa32 -.if ${MACHINE_CPUARCH} == "amd64" -LIBSA_CPUARCH=i386 -.else -LIBSA_CPUARCH=${MACHINE_CPUARCH} -.endif +LIBSA_CPUARCH=${MACHINE_CPUARCH:C/amd64/i386/} -.if ${MACHINE_ARCH} == "amd64" || ${MACHINE_ARCH} == "powerpc64" -CFLAGS+= -m32 -I. -.endif +CFLAGS+= -I. .PATH: ${SASRC} .include "${SASRC}/Makefile" - -.if ${MACHINE_CPUARCH} == "amd64" -CLEANFILES+= machine -beforedepend ${OBJS}: machine -machine: .NOMETA - ln -fs ${SYSDIR}/i386/include machine -.endif Index: projects/runtime-coverage/sys/boot/man/Makefile =================================================================== --- projects/runtime-coverage/sys/boot/man/Makefile (revision 325498) +++ projects/runtime-coverage/sys/boot/man/Makefile (revision 325499) @@ -1,10 +1,10 @@ # $FreeBSD$ -.include +.include MAN+= loader.8 .if ${MK_ZFS} != "no" MAN+= zfsloader.8 .endif .include Index: projects/runtime-coverage/sys/boot/ofw/Makefile.inc =================================================================== --- projects/runtime-coverage/sys/boot/ofw/Makefile.inc (revision 325498) +++ projects/runtime-coverage/sys/boot/ofw/Makefile.inc (revision 325499) @@ -1,8 +1,3 @@ # $FreeBSD$ -.if ${MACHINE_ARCH} == "powerpc64" -CFLAGS+= -m32 -mcpu=powerpc -LDFLAGS+= -m elf32ppc_fbsd -.endif - .include "../Makefile.inc" Index: projects/runtime-coverage/sys/boot/ofw/libofw/Makefile =================================================================== --- projects/runtime-coverage/sys/boot/ofw/libofw/Makefile (revision 325498) +++ projects/runtime-coverage/sys/boot/ofw/libofw/Makefile (revision 325499) @@ -1,35 +1,28 @@ # $FreeBSD$ .include LIB= ofw INTERNALLIB= SRCS= devicename.c elf_freebsd.c ofw_console.c ofw_copy.c ofw_disk.c \ ofw_memory.c ofw_module.c ofw_net.c ofw_reboot.c \ ofw_time.c openfirm.c .PATH: ${ZFSSRC} SRCS+= devicename_stubs.c # Pick up the bootstrap header for some interface items CFLAGS+= -I${LDRSRC} -I${SYSDIR} -I. CFLAGS+= -ffreestanding .if ${MACHINE_CPUARCH} == "powerpc" CFLAGS+= -msoft-float SRCS+= ppc64_elf_freebsd.c .endif .ifdef(BOOT_DISK_DEBUG) # Make the disk code more talkative CFLAGS+= -DDISK_DEBUG .endif -machine: .NOMETA - ln -sf ${SYSDIR}/${MACHINE_CPUARCH}/include machine - -CLEANFILES+= machine - .include - -beforedepend ${OBJS}: machine Index: projects/runtime-coverage/sys/boot/powerpc/Makefile.inc =================================================================== --- projects/runtime-coverage/sys/boot/powerpc/Makefile.inc (revision 325498) +++ projects/runtime-coverage/sys/boot/powerpc/Makefile.inc (revision 325499) @@ -1,7 +1,3 @@ # $FreeBSD$ -.if ${MACHINE_ARCH} == "powerpc64" -CFLAGS+= -m32 -mcpu=powerpc -.endif - .include "../Makefile.inc" Index: projects/runtime-coverage/sys/boot/powerpc/kboot/Makefile =================================================================== --- projects/runtime-coverage/sys/boot/powerpc/kboot/Makefile (revision 325498) +++ projects/runtime-coverage/sys/boot/powerpc/kboot/Makefile (revision 325499) @@ -1,66 +1,66 @@ # $FreeBSD$ .include MK_SSP= no MAN= PROG= loader.kboot NEWVERSWHAT= "kboot loader" ${MACHINE_ARCH} INSTALLFLAGS= -b # Architecture-specific loader code SRCS= conf.c metadata.c vers.c main.c ppc64_elf_freebsd.c SRCS+= host_syscall.S hostcons.c hostdisk.c kerneltramp.S kbootfdt.c SRCS+= ucmpdi2.c LOADER_CD9660_SUPPORT?= yes LOADER_MSDOS_SUPPORT?= no LOADER_EXT2FS_SUPPORT?= yes LOADER_UFS_SUPPORT?= yes LOADER_NET_SUPPORT?= yes LOADER_NFS_SUPPORT?= yes LOADER_TFTP_SUPPORT?= no LOADER_GZIP_SUPPORT?= yes LOADER_FDT_SUPPORT= yes LOADER_BZIP2_SUPPORT?= no .if ${LOADER_FDT_SUPPORT} == "yes" CFLAGS+= -I${FDTSRC} CFLAGS+= -I${SYSDIR}/contrib/libfdt CFLAGS+= -DLOADER_FDT_SUPPORT LIBFDT= ${BOOTOBJ}/fdt/libfdt.a .endif CFLAGS+= -mcpu=powerpc64 # Always add MI sources .include "${BOOTSRC}/loader.mk" .PATH: ${SYSDIR}/libkern CFLAGS+= -I${SYSDIR} CFLAGS+= -I. CLEANFILES+= loader.help CFLAGS+= -Wall -ffreestanding -msoft-float -DAIM # load address. set in linker script RELOC?= 0x0 CFLAGS+= -DRELOC=${RELOC} LDFLAGS= -nostdlib -static -T ${.CURDIR}/ldscript.powerpc # 64-bit bridge extensions CFLAGS+= -Wa,-mppc64bridge -DPADD= ${LIBFICL} ${LIBOFW} ${LIBFDT} ${LIBSA32} -LDADD= ${LIBFICL} ${LIBOFW} ${LIBFDT} ${LIBSA32} +DPADD= ${LIBFICL} ${LIBOFW} ${LIBFDT} ${LIBSA} +LDADD= ${LIBFICL} ${LIBOFW} ${LIBFDT} ${LIBSA} loader.help: help.common help.kboot ${FDTSRC}/help.fdt cat ${.ALLSRC} | \ awk -f ${LDRSRC}/merge_help.awk > ${.TARGET} .PATH: ${BOOTSRC}/forth .include "${BOOTSRC}/forth/Makefile.inc" FILES+= loader.rc menu.rc .include Index: projects/runtime-coverage/sys/boot/powerpc/ofw/Makefile =================================================================== --- projects/runtime-coverage/sys/boot/powerpc/ofw/Makefile (revision 325498) +++ projects/runtime-coverage/sys/boot/powerpc/ofw/Makefile (revision 325499) @@ -1,70 +1,70 @@ # $FreeBSD$ .include MK_SSP= no MAN= PROG= loader NEWVERSWHAT= "Open Firmware loader" ${MACHINE_ARCH} INSTALLFLAGS= -b # Architecture-specific loader code SRCS= conf.c metadata.c vers.c start.c SRCS+= ucmpdi2.c LOADER_CD9660_SUPPORT?= yes LOADER_EXT2FS_SUPPORT?= no LOADER_MSDOS_SUPPORT?= no LOADER_UFS_SUPPORT?= yes LOADER_NET_SUPPORT?= yes LOADER_NFS_SUPPORT?= yes LOADER_TFTP_SUPPORT?= no LOADER_GZIP_SUPPORT?= yes LOADER_BZIP2_SUPPORT?= no LOADER_FDT_SUPPORT?= yes .if ${LOADER_FDT_SUPPORT} == "yes" SRCS+= ofwfdt.c CFLAGS+= -I${FDTSRC} CFLAGS+= -I${SYSDIR}/contrib/libfdt CFLAGS+= -DLOADER_FDT_SUPPORT LIBFDT= ${BOOTOBJ}/fdt/libfdt.a .endif # Always add MI sources .include "${BOOTSRC}/loader.mk" .PATH: ${SYSDIR}/libkern CFLAGS+= -I${SYSDIR} CFLAGS+= -I. CLEANFILES+= loader.help CFLAGS+= -ffreestanding -msoft-float # load address. set in linker script RELOC?= 0x1C00000 CFLAGS+= -DRELOC=${RELOC} LDFLAGS= -nostdlib -static -T ${.CURDIR}/ldscript.powerpc # Pull in common loader code .PATH: ${BOOTSRC}/ofw/common .include "${BOOTSRC}/ofw/common/Makefile.inc" # Open Firmware standalone support library LIBOFW= ${BOOTOBJ}/ofw/libofw/libofw.a CFLAGS+= -I${BOOTSRC}/ofw/libofw -DPADD= ${LIBFICL} ${LIBOFW} ${LIBFDT} ${LIBSA32} -LDADD= ${LIBFICL} ${LIBOFW} ${LIBFDT} ${LIBSA32} +DPADD= ${LIBFICL} ${LIBOFW} ${LIBFDT} ${LIBSA} +LDADD= ${LIBFICL} ${LIBOFW} ${LIBFDT} ${LIBSA} loader.help: help.common help.ofw ${FDTSRC}/help.fdt cat ${.ALLSRC} | \ awk -f ${LDRSRC}/merge_help.awk > ${.TARGET} .PATH: ${BOOTSRC}/forth .include "${BOOTSRC}/forth/Makefile.inc" FILES+= loader.rc menu.rc .include Index: projects/runtime-coverage/sys/boot/powerpc/ps3/Makefile =================================================================== --- projects/runtime-coverage/sys/boot/powerpc/ps3/Makefile (revision 325498) +++ projects/runtime-coverage/sys/boot/powerpc/ps3/Makefile (revision 325499) @@ -1,67 +1,67 @@ # $FreeBSD$ .include MK_SSP= no MAN= PROG= loader.ps3 NEWVERSWHAT= "Playstation 3 loader" ${MACHINE_ARCH} INSTALLFLAGS= -b # Architecture-specific loader code SRCS= start.S conf.c metadata.c vers.c main.c devicename.c ppc64_elf_freebsd.c SRCS+= lv1call.S ps3cons.c font.h ps3mmu.c ps3net.c ps3repo.c \ ps3stor.c ps3disk.c ps3cdrom.c SRCS+= ucmpdi2.c LOADER_UFS_SUPPORT?= yes LOADER_CD9660_SUPPORT?= yes LOADER_EXT2FS_SUPPORT?= yes LOADER_NET_SUPPORT?= yes LOADER_NFS_SUPPORT?= yes LOADER_TFTP_SUPPORT?= no LOADER_GZIP_SUPPORT?= yes LOADER_FDT_SUPPORT?= no LOADER_BZIP2_SUPPORT?= no .if ${LOADER_FDT_SUPPORT} == "yes" CFLAGS+= -I${FDTSRC} CFLAGS+= -DLOADER_FDT_SUPPORT LIBFDT= ${BOOTOBJ}/fdt/libfdt.a .endif CFLAGS+= -mcpu=powerpc64 # Always add MI sources .include "${BOOTSRC}/loader.mk" .PATH: ${SYSDIR}/libkern CFLAGS+= -I${SYSDIR} CFLAGS+= -I. CLEANFILES+= loader.help CFLAGS+= -Wall -ffreestanding -msoft-float -DAIM # load address. set in linker script RELOC?= 0x0 CFLAGS+= -DRELOC=${RELOC} LDFLAGS= -nostdlib -static -T ${.CURDIR}/ldscript.powerpc -DPADD= ${LIBFICL} ${LIBOFW} ${LIBSA32} -LDADD= ${LIBFICL} ${LIBOFW} ${LIBSA32} +DPADD= ${LIBFICL} ${LIBOFW} ${LIBSA} +LDADD= ${LIBFICL} ${LIBOFW} ${LIBSA} SC_DFLT_FONT=cp437 font.h: uudecode < /usr/share/syscons/fonts/${SC_DFLT_FONT}-8x16.fnt && file2c 'u_char dflt_font_16[16*256] = {' '};' < ${SC_DFLT_FONT}-8x16 > font.h && uudecode < /usr/share/syscons/fonts/${SC_DFLT_FONT}-8x14.fnt && file2c 'u_char dflt_font_14[14*256] = {' '};' < ${SC_DFLT_FONT}-8x14 >> font.h && uudecode < /usr/share/syscons/fonts/${SC_DFLT_FONT}-8x8.fnt && file2c 'u_char dflt_font_8[8*256] = {' '};' < ${SC_DFLT_FONT}-8x8 >> font.h loader.help: help.common help.ps3 ${FDTSRC}/help.fdt cat ${.ALLSRC} | \ awk -f ${LDRSRC}/merge_help.awk > ${.TARGET} .PATH: ${BOOTSRC}/forth .include "${BOOTSRC}/forth/Makefile.inc" FILES+= loader.rc menu.rc .include Index: projects/runtime-coverage/sys/boot/powerpc/uboot/Makefile =================================================================== --- projects/runtime-coverage/sys/boot/powerpc/uboot/Makefile (revision 325498) +++ projects/runtime-coverage/sys/boot/powerpc/uboot/Makefile (revision 325499) @@ -1,48 +1,48 @@ # $FreeBSD$ .include PROG= ubldr NEWVERSWHAT= "U-Boot loader" ${MACHINE_ARCH} INSTALLFLAGS= -b MAN= # Architecture-specific loader code SRCS= start.S conf.c vers.c SRCS+= ucmpdi2.c LOADER_UFS_SUPPORT?= yes LOADER_CD9660_SUPPORT?= no LOADER_EXT2FS_SUPPORT?= no LOADER_NET_SUPPORT?= yes LOADER_NFS_SUPPORT?= yes LOADER_TFTP_SUPPORT?= no LOADER_GZIP_SUPPORT?= no LOADER_BZIP2_SUPPORT?= no LOADER_FDT_SUPPORT= ${MK_FDT} # Always add MI sources .include "${BOOTSRC}/loader.mk" .PATH: ${SYSDIR}/libkern CFLAGS+= -I${SYSDIR} CFLAGS+= -I. CLEANFILES+= ${PROG}.help CFLAGS+= -ffreestanding LDFLAGS= -nostdlib -static -T ${.CURDIR}/ldscript.powerpc .include "${BOOTSRC}/uboot.mk" -DPADD= ${LIBFICL} ${LIBUBOOT} ${LIBFDT} ${LIBUBOOT_FDT} ${LIBSA32} -LDADD= ${LIBFICL} ${LIBUBOOT} ${LIBFDT} ${LIBUBOOT_FDT} ${LIBSA32} +DPADD= ${LIBFICL} ${LIBUBOOT} ${LIBFDT} ${LIBUBOOT_FDT} ${LIBSA} +LDADD= ${LIBFICL} ${LIBUBOOT} ${LIBFDT} ${LIBUBOOT_FDT} ${LIBSA} loader.help: help.common help.uboot ${BOOTSRC}/fdt/help.fdt cat ${.ALLSRC} | \ awk -f ${LDRSRC}/merge_help.awk > ${.TARGET} .PATH: ${BOOTSRC}/forth FILES= loader.help .include Index: projects/runtime-coverage/sys/boot/uboot/Makefile =================================================================== --- projects/runtime-coverage/sys/boot/uboot/Makefile (revision 325498) +++ projects/runtime-coverage/sys/boot/uboot/Makefile (revision 325499) @@ -1,11 +1,11 @@ # $FreeBSD$ -.include +.include SUBDIR= lib .if ${MK_FDT} != "no" SUBDIR+=fdt .endif .include Index: projects/runtime-coverage/sys/boot/uboot/Makefile.inc =================================================================== --- projects/runtime-coverage/sys/boot/uboot/Makefile.inc (revision 325498) +++ projects/runtime-coverage/sys/boot/uboot/Makefile.inc (revision 325499) @@ -1,8 +1,3 @@ # $FreeBSD$ -.if ${MACHINE_ARCH} == "powerpc64" -CFLAGS+= -m32 -mcpu=powerpc -LDFLAGS+= -m elf32ppc_fbsd -.endif - .include "../Makefile.inc" Index: projects/runtime-coverage/sys/boot/uboot/fdt/Makefile =================================================================== --- projects/runtime-coverage/sys/boot/uboot/fdt/Makefile (revision 325498) +++ projects/runtime-coverage/sys/boot/uboot/fdt/Makefile (revision 325499) @@ -1,32 +1,25 @@ # $FreeBSD$ .include .PATH: ${LDRSRC} LIB= uboot_fdt INTERNALLIB= WARNS?= 2 SRCS= uboot_fdt.c CFLAGS+= -ffreestanding -msoft-float # U-Boot library headers CFLAGS+= -I${UBOOTSRC}/lib # libfdt headers CFLAGS+= -I${FDTSRC} # Pick up the bootstrap header for some interface items CFLAGS+= -I${LDRSRC} -I${SYSDIR} -I. -machine: .NOMETA - ln -sf ${SYSDIR}/${MACHINE_CPUARCH}/include machine - -CLEANFILES+= machine - .include .include - -beforedepend ${OBJS}: machine Index: projects/runtime-coverage/sys/boot/uboot/lib/Makefile =================================================================== --- projects/runtime-coverage/sys/boot/uboot/lib/Makefile (revision 325498) +++ projects/runtime-coverage/sys/boot/uboot/lib/Makefile (revision 325499) @@ -1,49 +1,42 @@ # $FreeBSD$ .include .PATH: ${LDRSRC} LIB= uboot INTERNALLIB= WARNS?= 2 SRCS= console.c copy.c devicename.c elf_freebsd.c glue.c SRCS+= module.c net.c reboot.c time.c CFLAGS+= -ffreestanding -msoft-float .if ${LOADER_DISK_SUPPORT:Uyes} == "yes" SRCS+= disk.c .endif .if ${MK_FDT} != "no" LOADER_FDT_SUPPORT= yes .else LOADER_FDT_SUPPORT= no .endif .if ${LOADER_FDT_SUPPORT} == "yes" CFLAGS+= -DLOADER_FDT_SUPPORT -I${FDTSRC} .endif # Pick up FDT includes CFLAGS+= -I${SYSDIR}/contrib/libfdt/ # Pick up the bootstrap header for some interface items CFLAGS+= -I${LDRSRC} -I${SYSDIR} -I. .ifdef(BOOT_DISK_DEBUG) # Make the disk code more talkative CFLAGS+= -DDISK_DEBUG .endif -machine: .NOMETA - ln -sf ${SYSDIR}/${MACHINE_CPUARCH}/include machine - -CLEANFILES+= machine - .include .include - -beforedepend ${OBJS}: machine Index: projects/runtime-coverage/sys/boot/zfs/Makefile =================================================================== --- projects/runtime-coverage/sys/boot/zfs/Makefile (revision 325498) +++ projects/runtime-coverage/sys/boot/zfs/Makefile (revision 325499) @@ -1,42 +1,22 @@ # $FreeBSD$ .include LIB= zfsboot INTERNALLIB= .PATH: ${ZFSSRC} SRCS+= zfs.c skein.c skein_block.c # Do not unroll skein loops, reduce code size CFLAGS+= -DSKEIN_LOOP=111 .PATH: ${SYSDIR}/crypto/skein CFLAGS+= -DBOOTPROG=\"zfsloader\" CFLAGS+= -I${LDRSRC} -I${SYSDIR} -I. CFLAGS+= -I${SYSDIR}/cddl/boot/zfs CFLAGS+= -I${SYSDIR}/crypto/skein -.if ${MACHINE_CPUARCH} == "i386" || \ - (${MACHINE_CPUARCH} == "amd64" && defined(ZFS32)) -CFLAGS+= -march=i386 -.endif -.if ${MACHINE_CPUARCH} == "amd64" && defined(ZFS32) -CFLAGS+= -m32 -.endif - CFLAGS+= -Wformat -Wall -.if ${MACHINE_CPUARCH} == "amd64" && defined(ZFS32) -CLEANFILES+= machine -machine: .NOMETA - ln -sf ${SYSDIR}/i386/include machine -.endif - .include .include - -.if ${MACHINE_CPUARCH} == "amd64" && defined(ZFS32) -.if !exists(machine) -beforedepend ${OBJS}: machine -.endif -.endif Index: projects/runtime-coverage/sys/boot/zfs32/Makefile =================================================================== --- projects/runtime-coverage/sys/boot/zfs32/Makefile (revision 325498) +++ projects/runtime-coverage/sys/boot/zfs32/Makefile (revision 325499) @@ -1,5 +1,5 @@ # $FreeBSD$ -ZFS32=1 +DO32=1 .include "${.CURDIR}/../zfs/Makefile" Index: projects/runtime-coverage/sys/dev/bnxt/if_bnxt.c =================================================================== --- projects/runtime-coverage/sys/dev/bnxt/if_bnxt.c (revision 325498) +++ projects/runtime-coverage/sys/dev/bnxt/if_bnxt.c (revision 325499) @@ -1,2603 +1,2455 @@ /*- * Broadcom NetXtreme-C/E network driver. * * Copyright (c) 2016 Broadcom, All Rights Reserved. * The term Broadcom refers to Broadcom Limited and/or its subsidiaries * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS' * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF * THE POSSIBILITY OF SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "opt_inet.h" #include "opt_inet6.h" #include "opt_rss.h" #include "ifdi_if.h" #include "bnxt.h" #include "bnxt_hwrm.h" #include "bnxt_ioctl.h" #include "bnxt_sysctl.h" #include "hsi_struct_def.h" /* * PCI Device ID Table */ static pci_vendor_info_t bnxt_vendor_info_array[] = { PVID(BROADCOM_VENDOR_ID, BCM57301, "Broadcom BCM57301 NetXtreme-C 10Gb Ethernet Controller"), PVID(BROADCOM_VENDOR_ID, BCM57302, "Broadcom BCM57302 NetXtreme-C 10Gb/25Gb Ethernet Controller"), PVID(BROADCOM_VENDOR_ID, BCM57304, "Broadcom BCM57304 NetXtreme-C 10Gb/25Gb/40Gb/50Gb Ethernet Controller"), PVID(BROADCOM_VENDOR_ID, BCM57311, "Broadcom BCM57311 NetXtreme-C 10Gb Ethernet"), PVID(BROADCOM_VENDOR_ID, BCM57312, "Broadcom BCM57312 NetXtreme-C 10Gb/25Gb Ethernet"), PVID(BROADCOM_VENDOR_ID, BCM57314, "Broadcom BCM57314 NetXtreme-C 10Gb/25Gb/40Gb/50Gb Ethernet"), PVID(BROADCOM_VENDOR_ID, BCM57402, "Broadcom BCM57402 NetXtreme-E 10Gb Ethernet Controller"), PVID(BROADCOM_VENDOR_ID, BCM57402_NPAR, "Broadcom BCM57402 NetXtreme-E Partition"), PVID(BROADCOM_VENDOR_ID, BCM57404, "Broadcom BCM57404 NetXtreme-E 10Gb/25Gb Ethernet Controller"), PVID(BROADCOM_VENDOR_ID, BCM57404_NPAR, "Broadcom BCM57404 NetXtreme-E Partition"), PVID(BROADCOM_VENDOR_ID, BCM57406, "Broadcom BCM57406 NetXtreme-E 10GBase-T Ethernet Controller"), PVID(BROADCOM_VENDOR_ID, BCM57406_NPAR, "Broadcom BCM57406 NetXtreme-E Partition"), PVID(BROADCOM_VENDOR_ID, BCM57407, "Broadcom BCM57407 NetXtreme-E 10GBase-T Ethernet Controller"), PVID(BROADCOM_VENDOR_ID, BCM57407_NPAR, "Broadcom BCM57407 NetXtreme-E Ethernet Partition"), PVID(BROADCOM_VENDOR_ID, BCM57407_SFP, "Broadcom BCM57407 NetXtreme-E 25Gb Ethernet Controller"), PVID(BROADCOM_VENDOR_ID, BCM57412, "Broadcom BCM57412 NetXtreme-E 10Gb Ethernet"), PVID(BROADCOM_VENDOR_ID, BCM57412_NPAR1, "Broadcom BCM57412 NetXtreme-E Ethernet Partition"), PVID(BROADCOM_VENDOR_ID, BCM57412_NPAR2, "Broadcom BCM57412 NetXtreme-E Ethernet Partition"), PVID(BROADCOM_VENDOR_ID, BCM57414, "Broadcom BCM57414 NetXtreme-E 10Gb/25Gb Ethernet"), PVID(BROADCOM_VENDOR_ID, BCM57414_NPAR1, "Broadcom BCM57414 NetXtreme-E Ethernet Partition"), PVID(BROADCOM_VENDOR_ID, BCM57414_NPAR2, "Broadcom BCM57414 NetXtreme-E Ethernet Partition"), PVID(BROADCOM_VENDOR_ID, BCM57416, "Broadcom BCM57416 NetXtreme-E 10GBase-T Ethernet"), PVID(BROADCOM_VENDOR_ID, BCM57416_NPAR1, "Broadcom BCM57416 NetXtreme-E Ethernet Partition"), PVID(BROADCOM_VENDOR_ID, BCM57416_NPAR2, "Broadcom BCM57416 NetXtreme-E Ethernet Partition"), PVID(BROADCOM_VENDOR_ID, BCM57416_SFP, "Broadcom BCM57416 NetXtreme-E 10Gb Ethernet"), PVID(BROADCOM_VENDOR_ID, BCM57417, "Broadcom BCM57417 NetXtreme-E 10GBase-T Ethernet"), PVID(BROADCOM_VENDOR_ID, BCM57417_NPAR1, "Broadcom BCM57417 NetXtreme-E Ethernet Partition"), PVID(BROADCOM_VENDOR_ID, BCM57417_NPAR2, "Broadcom BCM57417 NetXtreme-E Ethernet Partition"), PVID(BROADCOM_VENDOR_ID, BCM57417_SFP, "Broadcom BCM57417 NetXtreme-E 10Gb/25Gb Ethernet"), PVID(BROADCOM_VENDOR_ID, BCM57454, "Broadcom BCM57454 NetXtreme-E 10Gb/25Gb/40Gb/50Gb/100Gb Ethernet"), PVID(BROADCOM_VENDOR_ID, BCM58700, "Broadcom BCM58700 Nitro 1Gb/2.5Gb/10Gb Ethernet"), PVID(BROADCOM_VENDOR_ID, NETXTREME_C_VF1, "Broadcom NetXtreme-C Ethernet Virtual Function"), PVID(BROADCOM_VENDOR_ID, NETXTREME_C_VF2, "Broadcom NetXtreme-C Ethernet Virtual Function"), PVID(BROADCOM_VENDOR_ID, NETXTREME_C_VF3, "Broadcom NetXtreme-C Ethernet Virtual Function"), PVID(BROADCOM_VENDOR_ID, NETXTREME_E_VF1, "Broadcom NetXtreme-E Ethernet Virtual Function"), PVID(BROADCOM_VENDOR_ID, NETXTREME_E_VF2, "Broadcom NetXtreme-E Ethernet Virtual Function"), PVID(BROADCOM_VENDOR_ID, NETXTREME_E_VF3, "Broadcom NetXtreme-E Ethernet Virtual Function"), /* required last entry */ PVID_END }; /* * Function prototypes */ static void *bnxt_register(device_t dev); /* Soft queue setup and teardown */ static int bnxt_tx_queues_alloc(if_ctx_t ctx, caddr_t *vaddrs, uint64_t *paddrs, int ntxqs, int ntxqsets); static int bnxt_rx_queues_alloc(if_ctx_t ctx, caddr_t *vaddrs, uint64_t *paddrs, int nrxqs, int nrxqsets); static void bnxt_queues_free(if_ctx_t ctx); /* Device setup and teardown */ static int bnxt_attach_pre(if_ctx_t ctx); static int bnxt_attach_post(if_ctx_t ctx); static int bnxt_detach(if_ctx_t ctx); /* Device configuration */ static void bnxt_init(if_ctx_t ctx); static void bnxt_stop(if_ctx_t ctx); static void bnxt_multi_set(if_ctx_t ctx); static int bnxt_mtu_set(if_ctx_t ctx, uint32_t mtu); static void bnxt_media_status(if_ctx_t ctx, struct ifmediareq * ifmr); static int bnxt_media_change(if_ctx_t ctx); static int bnxt_promisc_set(if_ctx_t ctx, int flags); static uint64_t bnxt_get_counter(if_ctx_t, ift_counter); static void bnxt_update_admin_status(if_ctx_t ctx); static void bnxt_if_timer(if_ctx_t ctx, uint16_t qid); /* Interrupt enable / disable */ static void bnxt_intr_enable(if_ctx_t ctx); static int bnxt_rx_queue_intr_enable(if_ctx_t ctx, uint16_t qid); static int bnxt_tx_queue_intr_enable(if_ctx_t ctx, uint16_t qid); static void bnxt_disable_intr(if_ctx_t ctx); static int bnxt_msix_intr_assign(if_ctx_t ctx, int msix); /* vlan support */ static void bnxt_vlan_register(if_ctx_t ctx, uint16_t vtag); static void bnxt_vlan_unregister(if_ctx_t ctx, uint16_t vtag); /* ioctl */ static int bnxt_priv_ioctl(if_ctx_t ctx, u_long command, caddr_t data); static int bnxt_shutdown(if_ctx_t ctx); static int bnxt_suspend(if_ctx_t ctx); static int bnxt_resume(if_ctx_t ctx); /* Internal support functions */ static int bnxt_probe_phy(struct bnxt_softc *softc); static void bnxt_add_media_types(struct bnxt_softc *softc); static int bnxt_pci_mapping(struct bnxt_softc *softc); static void bnxt_pci_mapping_free(struct bnxt_softc *softc); static int bnxt_update_link(struct bnxt_softc *softc, bool chng_link_state); static int bnxt_handle_def_cp(void *arg); static int bnxt_handle_rx_cp(void *arg); static void bnxt_clear_ids(struct bnxt_softc *softc); static void inline bnxt_do_enable_intr(struct bnxt_cp_ring *cpr); static void inline bnxt_do_disable_intr(struct bnxt_cp_ring *cpr); static void bnxt_mark_cpr_invalid(struct bnxt_cp_ring *cpr); static void bnxt_def_cp_task(void *context); static void bnxt_handle_async_event(struct bnxt_softc *softc, struct cmpl_base *cmpl); static uint8_t get_phy_type(struct bnxt_softc *softc); static uint64_t bnxt_get_baudrate(struct bnxt_link_info *link); static void bnxt_get_wol_settings(struct bnxt_softc *softc); static int bnxt_wol_config(if_ctx_t ctx); /* * Device Interface Declaration */ static device_method_t bnxt_methods[] = { /* Device interface */ DEVMETHOD(device_register, bnxt_register), DEVMETHOD(device_probe, iflib_device_probe), DEVMETHOD(device_attach, iflib_device_attach), DEVMETHOD(device_detach, iflib_device_detach), DEVMETHOD(device_shutdown, iflib_device_shutdown), DEVMETHOD(device_suspend, iflib_device_suspend), DEVMETHOD(device_resume, iflib_device_resume), DEVMETHOD_END }; static driver_t bnxt_driver = { "bnxt", bnxt_methods, sizeof(struct bnxt_softc), }; devclass_t bnxt_devclass; DRIVER_MODULE(bnxt, pci, bnxt_driver, bnxt_devclass, 0, 0); MODULE_DEPEND(bnxt, pci, 1, 1, 1); MODULE_DEPEND(bnxt, ether, 1, 1, 1); MODULE_DEPEND(bnxt, iflib, 1, 1, 1); IFLIB_PNP_INFO(pci, bnxt, bnxt_vendor_info_array); static device_method_t bnxt_iflib_methods[] = { DEVMETHOD(ifdi_tx_queues_alloc, bnxt_tx_queues_alloc), DEVMETHOD(ifdi_rx_queues_alloc, bnxt_rx_queues_alloc), DEVMETHOD(ifdi_queues_free, bnxt_queues_free), DEVMETHOD(ifdi_attach_pre, bnxt_attach_pre), DEVMETHOD(ifdi_attach_post, bnxt_attach_post), DEVMETHOD(ifdi_detach, bnxt_detach), DEVMETHOD(ifdi_init, bnxt_init), DEVMETHOD(ifdi_stop, bnxt_stop), DEVMETHOD(ifdi_multi_set, bnxt_multi_set), DEVMETHOD(ifdi_mtu_set, bnxt_mtu_set), DEVMETHOD(ifdi_media_status, bnxt_media_status), DEVMETHOD(ifdi_media_change, bnxt_media_change), DEVMETHOD(ifdi_promisc_set, bnxt_promisc_set), DEVMETHOD(ifdi_get_counter, bnxt_get_counter), DEVMETHOD(ifdi_update_admin_status, bnxt_update_admin_status), DEVMETHOD(ifdi_timer, bnxt_if_timer), DEVMETHOD(ifdi_intr_enable, bnxt_intr_enable), DEVMETHOD(ifdi_tx_queue_intr_enable, bnxt_tx_queue_intr_enable), DEVMETHOD(ifdi_rx_queue_intr_enable, bnxt_rx_queue_intr_enable), DEVMETHOD(ifdi_intr_disable, bnxt_disable_intr), DEVMETHOD(ifdi_msix_intr_assign, bnxt_msix_intr_assign), DEVMETHOD(ifdi_vlan_register, bnxt_vlan_register), DEVMETHOD(ifdi_vlan_unregister, bnxt_vlan_unregister), DEVMETHOD(ifdi_priv_ioctl, bnxt_priv_ioctl), DEVMETHOD(ifdi_suspend, bnxt_suspend), DEVMETHOD(ifdi_shutdown, bnxt_shutdown), DEVMETHOD(ifdi_resume, bnxt_resume), DEVMETHOD_END }; static driver_t bnxt_iflib_driver = { "bnxt", bnxt_iflib_methods, sizeof(struct bnxt_softc) }; /* * iflib shared context */ #define BNXT_DRIVER_VERSION "1.0.0.2" char bnxt_driver_version[] = BNXT_DRIVER_VERSION; extern struct if_txrx bnxt_txrx; static struct if_shared_ctx bnxt_sctx_init = { .isc_magic = IFLIB_MAGIC, .isc_driver = &bnxt_iflib_driver, .isc_nfl = 2, // Number of Free Lists .isc_flags = IFLIB_HAS_RXCQ | IFLIB_HAS_TXCQ, .isc_q_align = PAGE_SIZE, .isc_tx_maxsize = BNXT_TSO_SIZE, .isc_tx_maxsegsize = BNXT_TSO_SIZE, .isc_rx_maxsize = BNXT_TSO_SIZE, .isc_rx_maxsegsize = BNXT_TSO_SIZE, // Only use a single segment to avoid page size constraints .isc_rx_nsegments = 1, .isc_ntxqs = 2, .isc_nrxqs = 3, .isc_nrxd_min = {16, 16, 16}, .isc_nrxd_default = {PAGE_SIZE / sizeof(struct cmpl_base) * 8, PAGE_SIZE / sizeof(struct rx_prod_pkt_bd), PAGE_SIZE / sizeof(struct rx_prod_pkt_bd)}, .isc_nrxd_max = {INT32_MAX, INT32_MAX, INT32_MAX}, .isc_ntxd_min = {16, 16, 16}, .isc_ntxd_default = {PAGE_SIZE / sizeof(struct cmpl_base) * 2, PAGE_SIZE / sizeof(struct tx_bd_short)}, .isc_ntxd_max = {INT32_MAX, INT32_MAX, INT32_MAX}, .isc_admin_intrcnt = 1, .isc_vendor_info = bnxt_vendor_info_array, .isc_driver_version = bnxt_driver_version, }; if_shared_ctx_t bnxt_sctx = &bnxt_sctx_init; /* * Device Methods */ static void * bnxt_register(device_t dev) { return bnxt_sctx; } /* * Device Dependent Configuration Functions */ /* Soft queue setup and teardown */ static int bnxt_tx_queues_alloc(if_ctx_t ctx, caddr_t *vaddrs, uint64_t *paddrs, int ntxqs, int ntxqsets) { struct bnxt_softc *softc; int i; int rc; softc = iflib_get_softc(ctx); softc->tx_cp_rings = malloc(sizeof(struct bnxt_cp_ring) * ntxqsets, M_DEVBUF, M_NOWAIT | M_ZERO); if (!softc->tx_cp_rings) { device_printf(iflib_get_dev(ctx), "unable to allocate TX completion rings\n"); rc = ENOMEM; goto cp_alloc_fail; } softc->tx_rings = malloc(sizeof(struct bnxt_ring) * ntxqsets, M_DEVBUF, M_NOWAIT | M_ZERO); if (!softc->tx_rings) { device_printf(iflib_get_dev(ctx), "unable to allocate TX rings\n"); rc = ENOMEM; goto ring_alloc_fail; } rc = iflib_dma_alloc(ctx, sizeof(struct ctx_hw_stats) * ntxqsets, &softc->tx_stats, 0); if (rc) goto dma_alloc_fail; bus_dmamap_sync(softc->tx_stats.idi_tag, softc->tx_stats.idi_map, BUS_DMASYNC_PREREAD); for (i = 0; i < ntxqsets; i++) { /* Set up the completion ring */ softc->tx_cp_rings[i].stats_ctx_id = HWRM_NA_SIGNATURE; softc->tx_cp_rings[i].ring.phys_id = (uint16_t)HWRM_NA_SIGNATURE; softc->tx_cp_rings[i].ring.softc = softc; softc->tx_cp_rings[i].ring.id = (softc->scctx->isc_nrxqsets * 2) + 1 + i; softc->tx_cp_rings[i].ring.doorbell = softc->tx_cp_rings[i].ring.id * 0x80; softc->tx_cp_rings[i].ring.ring_size = softc->scctx->isc_ntxd[0]; softc->tx_cp_rings[i].ring.vaddr = vaddrs[i * ntxqs]; softc->tx_cp_rings[i].ring.paddr = paddrs[i * ntxqs]; /* Set up the TX ring */ softc->tx_rings[i].phys_id = (uint16_t)HWRM_NA_SIGNATURE; softc->tx_rings[i].softc = softc; softc->tx_rings[i].id = (softc->scctx->isc_nrxqsets * 2) + 1 + i; softc->tx_rings[i].doorbell = softc->tx_rings[i].id * 0x80; softc->tx_rings[i].ring_size = softc->scctx->isc_ntxd[1]; softc->tx_rings[i].vaddr = vaddrs[i * ntxqs + 1]; softc->tx_rings[i].paddr = paddrs[i * ntxqs + 1]; bnxt_create_tx_sysctls(softc, i); } softc->ntxqsets = ntxqsets; return rc; dma_alloc_fail: free(softc->tx_rings, M_DEVBUF); ring_alloc_fail: free(softc->tx_cp_rings, M_DEVBUF); cp_alloc_fail: return rc; } static void bnxt_queues_free(if_ctx_t ctx) { struct bnxt_softc *softc = iflib_get_softc(ctx); // Free TX queues iflib_dma_free(&softc->tx_stats); free(softc->tx_rings, M_DEVBUF); softc->tx_rings = NULL; free(softc->tx_cp_rings, M_DEVBUF); softc->tx_cp_rings = NULL; softc->ntxqsets = 0; // Free RX queues iflib_dma_free(&softc->rx_stats); iflib_dma_free(&softc->hw_tx_port_stats); iflib_dma_free(&softc->hw_rx_port_stats); free(softc->grp_info, M_DEVBUF); free(softc->ag_rings, M_DEVBUF); free(softc->rx_rings, M_DEVBUF); free(softc->rx_cp_rings, M_DEVBUF); } static int bnxt_rx_queues_alloc(if_ctx_t ctx, caddr_t *vaddrs, uint64_t *paddrs, int nrxqs, int nrxqsets) { struct bnxt_softc *softc; int i; int rc; softc = iflib_get_softc(ctx); softc->rx_cp_rings = malloc(sizeof(struct bnxt_cp_ring) * nrxqsets, M_DEVBUF, M_NOWAIT | M_ZERO); if (!softc->rx_cp_rings) { device_printf(iflib_get_dev(ctx), "unable to allocate RX completion rings\n"); rc = ENOMEM; goto cp_alloc_fail; } softc->rx_rings = malloc(sizeof(struct bnxt_ring) * nrxqsets, M_DEVBUF, M_NOWAIT | M_ZERO); if (!softc->rx_rings) { device_printf(iflib_get_dev(ctx), "unable to allocate RX rings\n"); rc = ENOMEM; goto ring_alloc_fail; } softc->ag_rings = malloc(sizeof(struct bnxt_ring) * nrxqsets, M_DEVBUF, M_NOWAIT | M_ZERO); if (!softc->ag_rings) { device_printf(iflib_get_dev(ctx), "unable to allocate aggregation rings\n"); rc = ENOMEM; goto ag_alloc_fail; } softc->grp_info = malloc(sizeof(struct bnxt_grp_info) * nrxqsets, M_DEVBUF, M_NOWAIT | M_ZERO); if (!softc->grp_info) { device_printf(iflib_get_dev(ctx), "unable to allocate ring groups\n"); rc = ENOMEM; goto grp_alloc_fail; } rc = iflib_dma_alloc(ctx, sizeof(struct ctx_hw_stats) * nrxqsets, &softc->rx_stats, 0); if (rc) goto hw_stats_alloc_fail; bus_dmamap_sync(softc->rx_stats.idi_tag, softc->rx_stats.idi_map, BUS_DMASYNC_PREREAD); /* * Additional 512 bytes for future expansion. * To prevent corruption when loaded with newer firmwares with added counters. * This can be deleted when there will be no further additions of counters. */ #define BNXT_PORT_STAT_PADDING 512 rc = iflib_dma_alloc(ctx, sizeof(struct rx_port_stats) + BNXT_PORT_STAT_PADDING, &softc->hw_rx_port_stats, 0); if (rc) goto hw_port_rx_stats_alloc_fail; bus_dmamap_sync(softc->hw_rx_port_stats.idi_tag, softc->hw_rx_port_stats.idi_map, BUS_DMASYNC_PREREAD); rc = iflib_dma_alloc(ctx, sizeof(struct tx_port_stats) + BNXT_PORT_STAT_PADDING, &softc->hw_tx_port_stats, 0); if (rc) goto hw_port_tx_stats_alloc_fail; bus_dmamap_sync(softc->hw_tx_port_stats.idi_tag, softc->hw_tx_port_stats.idi_map, BUS_DMASYNC_PREREAD); softc->rx_port_stats = (void *) softc->hw_rx_port_stats.idi_vaddr; softc->tx_port_stats = (void *) softc->hw_tx_port_stats.idi_vaddr; for (i = 0; i < nrxqsets; i++) { /* Allocation the completion ring */ softc->rx_cp_rings[i].stats_ctx_id = HWRM_NA_SIGNATURE; softc->rx_cp_rings[i].ring.phys_id = (uint16_t)HWRM_NA_SIGNATURE; softc->rx_cp_rings[i].ring.softc = softc; softc->rx_cp_rings[i].ring.id = i + 1; softc->rx_cp_rings[i].ring.doorbell = softc->rx_cp_rings[i].ring.id * 0x80; /* * If this ring overflows, RX stops working. */ softc->rx_cp_rings[i].ring.ring_size = softc->scctx->isc_nrxd[0]; softc->rx_cp_rings[i].ring.vaddr = vaddrs[i * nrxqs]; softc->rx_cp_rings[i].ring.paddr = paddrs[i * nrxqs]; /* Allocate the RX ring */ softc->rx_rings[i].phys_id = (uint16_t)HWRM_NA_SIGNATURE; softc->rx_rings[i].softc = softc; softc->rx_rings[i].id = i + 1; softc->rx_rings[i].doorbell = softc->rx_rings[i].id * 0x80; softc->rx_rings[i].ring_size = softc->scctx->isc_nrxd[1]; softc->rx_rings[i].vaddr = vaddrs[i * nrxqs + 1]; softc->rx_rings[i].paddr = paddrs[i * nrxqs + 1]; /* Allocate the TPA start buffer */ softc->rx_rings[i].tpa_start = malloc(sizeof(struct bnxt_full_tpa_start) * (RX_TPA_START_CMPL_AGG_ID_MASK >> RX_TPA_START_CMPL_AGG_ID_SFT), M_DEVBUF, M_NOWAIT | M_ZERO); if (softc->rx_rings[i].tpa_start == NULL) { rc = -ENOMEM; device_printf(softc->dev, "Unable to allocate space for TPA\n"); goto tpa_alloc_fail; } /* Allocate the AG ring */ softc->ag_rings[i].phys_id = (uint16_t)HWRM_NA_SIGNATURE; softc->ag_rings[i].softc = softc; softc->ag_rings[i].id = nrxqsets + i + 1; softc->ag_rings[i].doorbell = softc->ag_rings[i].id * 0x80; softc->ag_rings[i].ring_size = softc->scctx->isc_nrxd[2]; softc->ag_rings[i].vaddr = vaddrs[i * nrxqs + 2]; softc->ag_rings[i].paddr = paddrs[i * nrxqs + 2]; /* Allocate the ring group */ softc->grp_info[i].grp_id = (uint16_t)HWRM_NA_SIGNATURE; softc->grp_info[i].stats_ctx = softc->rx_cp_rings[i].stats_ctx_id; softc->grp_info[i].rx_ring_id = softc->rx_rings[i].phys_id; softc->grp_info[i].ag_ring_id = softc->ag_rings[i].phys_id; softc->grp_info[i].cp_ring_id = softc->rx_cp_rings[i].ring.phys_id; bnxt_create_rx_sysctls(softc, i); } /* * When SR-IOV is enabled, avoid each VF sending PORT_QSTATS * HWRM every sec with which firmware timeouts can happen */ if (BNXT_PF(softc)) bnxt_create_port_stats_sysctls(softc); /* And finally, the VNIC */ softc->vnic_info.id = (uint16_t)HWRM_NA_SIGNATURE; softc->vnic_info.flow_id = (uint16_t)HWRM_NA_SIGNATURE; softc->vnic_info.filter_id = -1; softc->vnic_info.def_ring_grp = (uint16_t)HWRM_NA_SIGNATURE; softc->vnic_info.cos_rule = (uint16_t)HWRM_NA_SIGNATURE; softc->vnic_info.lb_rule = (uint16_t)HWRM_NA_SIGNATURE; softc->vnic_info.rx_mask = HWRM_CFA_L2_SET_RX_MASK_INPUT_MASK_BCAST; softc->vnic_info.mc_list_count = 0; softc->vnic_info.flags = BNXT_VNIC_FLAG_DEFAULT; rc = iflib_dma_alloc(ctx, BNXT_MAX_MC_ADDRS * ETHER_ADDR_LEN, &softc->vnic_info.mc_list, 0); if (rc) goto mc_list_alloc_fail; /* The VNIC RSS Hash Key */ rc = iflib_dma_alloc(ctx, HW_HASH_KEY_SIZE, &softc->vnic_info.rss_hash_key_tbl, 0); if (rc) goto rss_hash_alloc_fail; bus_dmamap_sync(softc->vnic_info.rss_hash_key_tbl.idi_tag, softc->vnic_info.rss_hash_key_tbl.idi_map, BUS_DMASYNC_PREWRITE); memcpy(softc->vnic_info.rss_hash_key_tbl.idi_vaddr, softc->vnic_info.rss_hash_key, HW_HASH_KEY_SIZE); /* Allocate the RSS tables */ rc = iflib_dma_alloc(ctx, HW_HASH_INDEX_SIZE * sizeof(uint16_t), &softc->vnic_info.rss_grp_tbl, 0); if (rc) goto rss_grp_alloc_fail; bus_dmamap_sync(softc->vnic_info.rss_grp_tbl.idi_tag, softc->vnic_info.rss_grp_tbl.idi_map, BUS_DMASYNC_PREWRITE); memset(softc->vnic_info.rss_grp_tbl.idi_vaddr, 0xff, softc->vnic_info.rss_grp_tbl.idi_size); softc->nrxqsets = nrxqsets; return rc; rss_grp_alloc_fail: iflib_dma_free(&softc->vnic_info.rss_hash_key_tbl); rss_hash_alloc_fail: iflib_dma_free(&softc->vnic_info.mc_list); tpa_alloc_fail: mc_list_alloc_fail: for (i = i - 1; i >= 0; i--) free(softc->rx_rings[i].tpa_start, M_DEVBUF); iflib_dma_free(&softc->hw_tx_port_stats); hw_port_tx_stats_alloc_fail: iflib_dma_free(&softc->hw_rx_port_stats); hw_port_rx_stats_alloc_fail: iflib_dma_free(&softc->rx_stats); hw_stats_alloc_fail: free(softc->grp_info, M_DEVBUF); grp_alloc_fail: free(softc->ag_rings, M_DEVBUF); ag_alloc_fail: free(softc->rx_rings, M_DEVBUF); ring_alloc_fail: free(softc->rx_cp_rings, M_DEVBUF); cp_alloc_fail: return rc; } /* Device setup and teardown */ static int bnxt_attach_pre(if_ctx_t ctx) { struct bnxt_softc *softc = iflib_get_softc(ctx); if_softc_ctx_t scctx; int rc = 0; softc->ctx = ctx; softc->dev = iflib_get_dev(ctx); softc->media = iflib_get_media(ctx); softc->scctx = iflib_get_softc_ctx(ctx); softc->sctx = iflib_get_sctx(ctx); scctx = softc->scctx; /* TODO: Better way of detecting NPAR/VF is needed */ switch (pci_get_device(softc->dev)) { case BCM57402_NPAR: case BCM57404_NPAR: case BCM57406_NPAR: case BCM57407_NPAR: case BCM57412_NPAR1: case BCM57412_NPAR2: case BCM57414_NPAR1: case BCM57414_NPAR2: case BCM57416_NPAR1: case BCM57416_NPAR2: softc->flags |= BNXT_FLAG_NPAR; break; case NETXTREME_C_VF1: case NETXTREME_C_VF2: case NETXTREME_C_VF3: case NETXTREME_E_VF1: case NETXTREME_E_VF2: case NETXTREME_E_VF3: softc->flags |= BNXT_FLAG_VF; break; } pci_enable_busmaster(softc->dev); if (bnxt_pci_mapping(softc)) return (ENXIO); /* HWRM setup/init */ BNXT_HWRM_LOCK_INIT(softc, device_get_nameunit(softc->dev)); rc = bnxt_alloc_hwrm_dma_mem(softc); if (rc) goto dma_fail; /* Get firmware version and compare with driver */ softc->ver_info = malloc(sizeof(struct bnxt_ver_info), M_DEVBUF, M_NOWAIT | M_ZERO); if (softc->ver_info == NULL) { rc = ENOMEM; device_printf(softc->dev, "Unable to allocate space for version info\n"); goto ver_alloc_fail; } /* Default minimum required HWRM version */ softc->ver_info->hwrm_min_major = 1; softc->ver_info->hwrm_min_minor = 2; softc->ver_info->hwrm_min_update = 2; rc = bnxt_hwrm_ver_get(softc); if (rc) { device_printf(softc->dev, "attach: hwrm ver get failed\n"); goto ver_fail; } /* Get NVRAM info */ softc->nvm_info = malloc(sizeof(struct bnxt_nvram_info), M_DEVBUF, M_NOWAIT | M_ZERO); if (softc->nvm_info == NULL) { rc = ENOMEM; device_printf(softc->dev, "Unable to allocate space for NVRAM info\n"); goto nvm_alloc_fail; } rc = bnxt_hwrm_nvm_get_dev_info(softc, &softc->nvm_info->mfg_id, &softc->nvm_info->device_id, &softc->nvm_info->sector_size, &softc->nvm_info->size, &softc->nvm_info->reserved_size, &softc->nvm_info->available_size); /* Register the driver with the FW */ rc = bnxt_hwrm_func_drv_rgtr(softc); if (rc) { device_printf(softc->dev, "attach: hwrm drv rgtr failed\n"); goto drv_rgtr_fail; } rc = bnxt_hwrm_func_rgtr_async_events(softc, NULL, 0); if (rc) { device_printf(softc->dev, "attach: hwrm rgtr async evts failed\n"); goto drv_rgtr_fail; } /* Get the HW capabilities */ rc = bnxt_hwrm_func_qcaps(softc); if (rc) goto failed; /* Get the current configuration of this function */ rc = bnxt_hwrm_func_qcfg(softc); if (rc) { device_printf(softc->dev, "attach: hwrm func qcfg failed\n"); goto failed; } iflib_set_mac(ctx, softc->func.mac_addr); scctx->isc_txrx = &bnxt_txrx; scctx->isc_tx_csum_flags = (CSUM_IP | CSUM_TCP | CSUM_UDP | CSUM_TCP_IPV6 | CSUM_UDP_IPV6 | CSUM_TSO); scctx->isc_capenable = /* These are translated to hwassit bits */ IFCAP_TXCSUM | IFCAP_TXCSUM_IPV6 | IFCAP_TSO4 | IFCAP_TSO6 | /* These are checked by iflib */ IFCAP_LRO | IFCAP_VLAN_HWFILTER | /* These are part of the iflib mask */ IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWTSO | /* These likely get lost... */ IFCAP_VLAN_HWCSUM | IFCAP_JUMBO_MTU; if (bnxt_wol_supported(softc)) scctx->isc_capenable |= IFCAP_WOL_MAGIC; /* Get the queue config */ rc = bnxt_hwrm_queue_qportcfg(softc); if (rc) { device_printf(softc->dev, "attach: hwrm qportcfg failed\n"); goto failed; } bnxt_get_wol_settings(softc); /* Now perform a function reset */ rc = bnxt_hwrm_func_reset(softc); bnxt_clear_ids(softc); if (rc) goto failed; /* Now set up iflib sc */ scctx->isc_tx_nsegments = 31, scctx->isc_tx_tso_segments_max = 31; scctx->isc_tx_tso_size_max = BNXT_TSO_SIZE; scctx->isc_tx_tso_segsize_max = BNXT_TSO_SIZE; scctx->isc_vectors = softc->func.max_cp_rings; scctx->isc_txrx = &bnxt_txrx; if (scctx->isc_nrxd[0] < ((scctx->isc_nrxd[1] * 4) + scctx->isc_nrxd[2])) device_printf(softc->dev, "WARNING: nrxd0 (%d) should be at least 4 * nrxd1 (%d) + nrxd2 (%d). Driver may be unstable\n", scctx->isc_nrxd[0], scctx->isc_nrxd[1], scctx->isc_nrxd[2]); if (scctx->isc_ntxd[0] < scctx->isc_ntxd[1] * 2) device_printf(softc->dev, "WARNING: ntxd0 (%d) should be at least 2 * ntxd1 (%d). Driver may be unstable\n", scctx->isc_ntxd[0], scctx->isc_ntxd[1]); scctx->isc_txqsizes[0] = sizeof(struct cmpl_base) * scctx->isc_ntxd[0]; scctx->isc_txqsizes[1] = sizeof(struct tx_bd_short) * scctx->isc_ntxd[1]; scctx->isc_rxqsizes[0] = sizeof(struct cmpl_base) * scctx->isc_nrxd[0]; scctx->isc_rxqsizes[1] = sizeof(struct rx_prod_pkt_bd) * scctx->isc_nrxd[1]; scctx->isc_rxqsizes[2] = sizeof(struct rx_prod_pkt_bd) * scctx->isc_nrxd[2]; scctx->isc_nrxqsets_max = min(pci_msix_count(softc->dev)-1, softc->fn_qcfg.alloc_completion_rings - 1); scctx->isc_nrxqsets_max = min(scctx->isc_nrxqsets_max, softc->fn_qcfg.alloc_rx_rings); scctx->isc_nrxqsets_max = min(scctx->isc_nrxqsets_max, softc->fn_qcfg.alloc_vnics); scctx->isc_ntxqsets_max = min(softc->fn_qcfg.alloc_tx_rings, softc->fn_qcfg.alloc_completion_rings - scctx->isc_nrxqsets_max - 1); scctx->isc_rss_table_size = HW_HASH_INDEX_SIZE; scctx->isc_rss_table_mask = scctx->isc_rss_table_size - 1; /* iflib will map and release this bar */ scctx->isc_msix_bar = pci_msix_table_bar(softc->dev); /* * Default settings for HW LRO (TPA): * Disable HW LRO by default * Can be enabled after taking care of 'packet forwarding' */ softc->hw_lro.enable = 0; softc->hw_lro.is_mode_gro = 0; softc->hw_lro.max_agg_segs = 5; /* 2^5 = 32 segs */ softc->hw_lro.max_aggs = HWRM_VNIC_TPA_CFG_INPUT_MAX_AGGS_MAX; softc->hw_lro.min_agg_len = 512; /* Allocate the default completion ring */ softc->def_cp_ring.stats_ctx_id = HWRM_NA_SIGNATURE; softc->def_cp_ring.ring.phys_id = (uint16_t)HWRM_NA_SIGNATURE; softc->def_cp_ring.ring.softc = softc; softc->def_cp_ring.ring.id = 0; softc->def_cp_ring.ring.doorbell = softc->def_cp_ring.ring.id * 0x80; softc->def_cp_ring.ring.ring_size = PAGE_SIZE / sizeof(struct cmpl_base); rc = iflib_dma_alloc(ctx, sizeof(struct cmpl_base) * softc->def_cp_ring.ring.ring_size, &softc->def_cp_ring_mem, 0); softc->def_cp_ring.ring.vaddr = softc->def_cp_ring_mem.idi_vaddr; softc->def_cp_ring.ring.paddr = softc->def_cp_ring_mem.idi_paddr; iflib_config_gtask_init(ctx, &softc->def_cp_task, bnxt_def_cp_task, "dflt_cp"); rc = bnxt_init_sysctl_ctx(softc); if (rc) goto init_sysctl_failed; rc = bnxt_create_nvram_sysctls(softc->nvm_info); if (rc) goto failed; arc4rand(softc->vnic_info.rss_hash_key, HW_HASH_KEY_SIZE, 0); softc->vnic_info.rss_hash_type = HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_IPV4 | HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_TCP_IPV4 | HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_UDP_IPV4 | HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_IPV6 | HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_TCP_IPV6 | HWRM_VNIC_RSS_CFG_INPUT_HASH_TYPE_UDP_IPV6; rc = bnxt_create_config_sysctls_pre(softc); if (rc) goto failed; rc = bnxt_create_hw_lro_sysctls(softc); if (rc) goto failed; rc = bnxt_create_pause_fc_sysctls(softc); if (rc) goto failed; /* Initialize the vlan list */ SLIST_INIT(&softc->vnic_info.vlan_tags); softc->vnic_info.vlan_tag_list.idi_vaddr = NULL; return (rc); failed: bnxt_free_sysctl_ctx(softc); init_sysctl_failed: bnxt_hwrm_func_drv_unrgtr(softc, false); drv_rgtr_fail: free(softc->nvm_info, M_DEVBUF); nvm_alloc_fail: ver_fail: free(softc->ver_info, M_DEVBUF); ver_alloc_fail: bnxt_free_hwrm_dma_mem(softc); dma_fail: BNXT_HWRM_LOCK_DESTROY(softc); bnxt_pci_mapping_free(softc); pci_disable_busmaster(softc->dev); return (rc); } static int bnxt_attach_post(if_ctx_t ctx) { struct bnxt_softc *softc = iflib_get_softc(ctx); if_t ifp = iflib_get_ifp(ctx); int rc; bnxt_create_config_sysctls_post(softc); /* Update link state etc... */ rc = bnxt_probe_phy(softc); if (rc) goto failed; /* Needs to be done after probing the phy */ bnxt_create_ver_sysctls(softc); bnxt_add_media_types(softc); ifmedia_set(softc->media, IFM_ETHER | IFM_AUTO); softc->scctx->isc_max_frame_size = ifp->if_mtu + ETHER_HDR_LEN + ETHER_CRC_LEN; failed: return rc; } static int bnxt_detach(if_ctx_t ctx) { struct bnxt_softc *softc = iflib_get_softc(ctx); struct bnxt_vlan_tag *tag; struct bnxt_vlan_tag *tmp; int i; bnxt_wol_config(ctx); bnxt_do_disable_intr(&softc->def_cp_ring); bnxt_free_sysctl_ctx(softc); bnxt_hwrm_func_reset(softc); bnxt_clear_ids(softc); iflib_irq_free(ctx, &softc->def_cp_ring.irq); iflib_config_gtask_deinit(&softc->def_cp_task); /* We need to free() these here... */ for (i = softc->nrxqsets-1; i>=0; i--) { iflib_irq_free(ctx, &softc->rx_cp_rings[i].irq); } iflib_dma_free(&softc->vnic_info.mc_list); iflib_dma_free(&softc->vnic_info.rss_hash_key_tbl); iflib_dma_free(&softc->vnic_info.rss_grp_tbl); if (softc->vnic_info.vlan_tag_list.idi_vaddr) iflib_dma_free(&softc->vnic_info.vlan_tag_list); SLIST_FOREACH_SAFE(tag, &softc->vnic_info.vlan_tags, next, tmp) free(tag, M_DEVBUF); iflib_dma_free(&softc->def_cp_ring_mem); for (i = 0; i < softc->nrxqsets; i++) free(softc->rx_rings[i].tpa_start, M_DEVBUF); free(softc->ver_info, M_DEVBUF); free(softc->nvm_info, M_DEVBUF); bnxt_hwrm_func_drv_unrgtr(softc, false); bnxt_free_hwrm_dma_mem(softc); BNXT_HWRM_LOCK_DESTROY(softc); pci_disable_busmaster(softc->dev); bnxt_pci_mapping_free(softc); return 0; } /* Device configuration */ static void bnxt_init(if_ctx_t ctx) { struct bnxt_softc *softc = iflib_get_softc(ctx); struct ifmediareq ifmr; int i, j; int rc; rc = bnxt_hwrm_func_reset(softc); if (rc) return; bnxt_clear_ids(softc); /* Allocate the default completion ring */ softc->def_cp_ring.cons = UINT32_MAX; softc->def_cp_ring.v_bit = 1; bnxt_mark_cpr_invalid(&softc->def_cp_ring); rc = bnxt_hwrm_ring_alloc(softc, HWRM_RING_ALLOC_INPUT_RING_TYPE_L2_CMPL, &softc->def_cp_ring.ring, (uint16_t)HWRM_NA_SIGNATURE, HWRM_NA_SIGNATURE, true); if (rc) goto fail; /* And now set the default CP ring as the async CP ring */ rc = bnxt_cfg_async_cr(softc); if (rc) goto fail; for (i = 0; i < softc->nrxqsets; i++) { /* Allocate the statistics context */ rc = bnxt_hwrm_stat_ctx_alloc(softc, &softc->rx_cp_rings[i], softc->rx_stats.idi_paddr + (sizeof(struct ctx_hw_stats) * i)); if (rc) goto fail; /* Allocate the completion ring */ softc->rx_cp_rings[i].cons = UINT32_MAX; softc->rx_cp_rings[i].v_bit = 1; softc->rx_cp_rings[i].last_idx = UINT32_MAX; bnxt_mark_cpr_invalid(&softc->rx_cp_rings[i]); rc = bnxt_hwrm_ring_alloc(softc, HWRM_RING_ALLOC_INPUT_RING_TYPE_L2_CMPL, &softc->rx_cp_rings[i].ring, (uint16_t)HWRM_NA_SIGNATURE, HWRM_NA_SIGNATURE, true); if (rc) goto fail; /* Allocate the RX ring */ rc = bnxt_hwrm_ring_alloc(softc, HWRM_RING_ALLOC_INPUT_RING_TYPE_RX, &softc->rx_rings[i], (uint16_t)HWRM_NA_SIGNATURE, HWRM_NA_SIGNATURE, false); if (rc) goto fail; BNXT_RX_DB(&softc->rx_rings[i], 0); /* TODO: Cumulus+ doesn't need the double doorbell */ BNXT_RX_DB(&softc->rx_rings[i], 0); /* Allocate the AG ring */ rc = bnxt_hwrm_ring_alloc(softc, HWRM_RING_ALLOC_INPUT_RING_TYPE_RX, &softc->ag_rings[i], (uint16_t)HWRM_NA_SIGNATURE, HWRM_NA_SIGNATURE, false); if (rc) goto fail; BNXT_RX_DB(&softc->rx_rings[i], 0); /* TODO: Cumulus+ doesn't need the double doorbell */ BNXT_RX_DB(&softc->ag_rings[i], 0); /* Allocate the ring group */ softc->grp_info[i].stats_ctx = softc->rx_cp_rings[i].stats_ctx_id; softc->grp_info[i].rx_ring_id = softc->rx_rings[i].phys_id; softc->grp_info[i].ag_ring_id = softc->ag_rings[i].phys_id; softc->grp_info[i].cp_ring_id = softc->rx_cp_rings[i].ring.phys_id; rc = bnxt_hwrm_ring_grp_alloc(softc, &softc->grp_info[i]); if (rc) goto fail; } /* Allocate the VNIC RSS context */ rc = bnxt_hwrm_vnic_ctx_alloc(softc, &softc->vnic_info.rss_id); if (rc) goto fail; /* Allocate the vnic */ softc->vnic_info.def_ring_grp = softc->grp_info[0].grp_id; softc->vnic_info.mru = softc->scctx->isc_max_frame_size; rc = bnxt_hwrm_vnic_alloc(softc, &softc->vnic_info); if (rc) goto fail; rc = bnxt_hwrm_vnic_cfg(softc, &softc->vnic_info); if (rc) goto fail; rc = bnxt_hwrm_set_filter(softc, &softc->vnic_info); if (rc) goto fail; /* Enable RSS on the VNICs */ for (i = 0, j = 0; i < HW_HASH_INDEX_SIZE; i++) { ((uint16_t *) softc->vnic_info.rss_grp_tbl.idi_vaddr)[i] = htole16(softc->grp_info[j].grp_id); if (++j == softc->nrxqsets) j = 0; } rc = bnxt_hwrm_rss_cfg(softc, &softc->vnic_info, softc->vnic_info.rss_hash_type); if (rc) goto fail; rc = bnxt_hwrm_vnic_tpa_cfg(softc); if (rc) goto fail; for (i = 0; i < softc->ntxqsets; i++) { /* Allocate the statistics context */ rc = bnxt_hwrm_stat_ctx_alloc(softc, &softc->tx_cp_rings[i], softc->tx_stats.idi_paddr + (sizeof(struct ctx_hw_stats) * i)); if (rc) goto fail; /* Allocate the completion ring */ softc->tx_cp_rings[i].cons = UINT32_MAX; softc->tx_cp_rings[i].v_bit = 1; bnxt_mark_cpr_invalid(&softc->tx_cp_rings[i]); rc = bnxt_hwrm_ring_alloc(softc, HWRM_RING_ALLOC_INPUT_RING_TYPE_L2_CMPL, &softc->tx_cp_rings[i].ring, (uint16_t)HWRM_NA_SIGNATURE, HWRM_NA_SIGNATURE, false); if (rc) goto fail; /* Allocate the TX ring */ rc = bnxt_hwrm_ring_alloc(softc, HWRM_RING_ALLOC_INPUT_RING_TYPE_TX, &softc->tx_rings[i], softc->tx_cp_rings[i].ring.phys_id, softc->tx_cp_rings[i].stats_ctx_id, false); if (rc) goto fail; BNXT_TX_DB(&softc->tx_rings[i], 0); /* TODO: Cumulus+ doesn't need the double doorbell */ BNXT_TX_DB(&softc->tx_rings[i], 0); } bnxt_do_enable_intr(&softc->def_cp_ring); bnxt_media_status(softc->ctx, &ifmr); bnxt_hwrm_cfa_l2_set_rx_mask(softc, &softc->vnic_info); return; fail: bnxt_hwrm_func_reset(softc); bnxt_clear_ids(softc); return; } static void bnxt_stop(if_ctx_t ctx) { struct bnxt_softc *softc = iflib_get_softc(ctx); bnxt_do_disable_intr(&softc->def_cp_ring); bnxt_hwrm_func_reset(softc); bnxt_clear_ids(softc); return; } static void bnxt_multi_set(if_ctx_t ctx) { struct bnxt_softc *softc = iflib_get_softc(ctx); if_t ifp = iflib_get_ifp(ctx); uint8_t *mta; int cnt, mcnt; mcnt = if_multiaddr_count(ifp, -1); if (mcnt > BNXT_MAX_MC_ADDRS) { softc->vnic_info.rx_mask |= HWRM_CFA_L2_SET_RX_MASK_INPUT_MASK_ALL_MCAST; bnxt_hwrm_cfa_l2_set_rx_mask(softc, &softc->vnic_info); } else { softc->vnic_info.rx_mask &= ~HWRM_CFA_L2_SET_RX_MASK_INPUT_MASK_ALL_MCAST; mta = softc->vnic_info.mc_list.idi_vaddr; bzero(mta, softc->vnic_info.mc_list.idi_size); if_multiaddr_array(ifp, mta, &cnt, mcnt); bus_dmamap_sync(softc->vnic_info.mc_list.idi_tag, softc->vnic_info.mc_list.idi_map, BUS_DMASYNC_PREWRITE); softc->vnic_info.mc_list_count = cnt; softc->vnic_info.rx_mask |= HWRM_CFA_L2_SET_RX_MASK_INPUT_MASK_MCAST; if (bnxt_hwrm_cfa_l2_set_rx_mask(softc, &softc->vnic_info)) device_printf(softc->dev, "set_multi: rx_mask set failed\n"); } } static int bnxt_mtu_set(if_ctx_t ctx, uint32_t mtu) { struct bnxt_softc *softc = iflib_get_softc(ctx); if (mtu > BNXT_MAX_MTU) return EINVAL; softc->scctx->isc_max_frame_size = mtu + ETHER_HDR_LEN + ETHER_CRC_LEN; return 0; } static void bnxt_media_status(if_ctx_t ctx, struct ifmediareq * ifmr) { struct bnxt_softc *softc = iflib_get_softc(ctx); struct bnxt_link_info *link_info = &softc->link_info; - uint8_t phy_type = get_phy_type(softc); + struct ifmedia_entry *next; + uint64_t target_baudrate = bnxt_get_baudrate(link_info); + int active_media = IFM_UNKNOWN; + bnxt_update_link(softc, true); ifmr->ifm_status = IFM_AVALID; ifmr->ifm_active = IFM_ETHER; if (link_info->link_up) ifmr->ifm_status |= IFM_ACTIVE; else ifmr->ifm_status &= ~IFM_ACTIVE; if (link_info->duplex == HWRM_PORT_PHY_QCFG_OUTPUT_DUPLEX_CFG_FULL) ifmr->ifm_active |= IFM_FDX; else ifmr->ifm_active |= IFM_HDX; - switch (link_info->link_speed) { - case HWRM_PORT_PHY_QCFG_OUTPUT_LINK_SPEED_100MB: - ifmr->ifm_active |= IFM_100_T; - break; - case HWRM_PORT_PHY_QCFG_OUTPUT_LINK_SPEED_1GB: - switch (phy_type) { - case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASEKX: - ifmr->ifm_active |= IFM_1000_KX; + /* + * Go through the list of supported media which got prepared + * as part of bnxt_add_media_types() using api ifmedia_add(). + */ + LIST_FOREACH(next, &(iflib_get_media(ctx)->ifm_list), ifm_list) { + if (ifmedia_baudrate(next->ifm_media) == target_baudrate) { + active_media = next->ifm_media; break; - case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASET: - ifmr->ifm_active |= IFM_1000_T; - break; - case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_SGMIIEXTPHY: - ifmr->ifm_active |= IFM_1000_SGMII; - break; - default: - /* - * Workaround: - * Don't return IFM_UNKNOWN until - * Stratus return proper media_type - */ - ifmr->ifm_active |= IFM_1000_KX; - break; } - break; - case HWRM_PORT_PHY_QCFG_OUTPUT_LINK_SPEED_2_5GB: - switch (phy_type) { - case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASEKX: - ifmr->ifm_active |= IFM_2500_KX; - break; - case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASET: - ifmr->ifm_active |= IFM_2500_T; - break; - default: - ifmr->ifm_active |= IFM_UNKNOWN; - break; - } - break; - case HWRM_PORT_PHY_QCFG_OUTPUT_LINK_SPEED_10GB: - switch (phy_type) { - case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASECR: - ifmr->ifm_active |= IFM_10G_CR1; - break; - case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASEKR4: - case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASEKR2: - case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASEKR: - ifmr->ifm_active |= IFM_10G_KR; - break; - case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASELR: - ifmr->ifm_active |= IFM_10G_LR; - break; - case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASESR: - ifmr->ifm_active |= IFM_10G_SR; - break; - case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASEKX: - ifmr->ifm_active |= IFM_10G_KX4; - break; - case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASET: - ifmr->ifm_active |= IFM_10G_T; - break; - default: - /* - * Workaround: - * Don't return IFM_UNKNOWN until - * Stratus return proper media_type - */ - ifmr->ifm_active |= IFM_10G_CR1; - break; - } - break; - case HWRM_PORT_PHY_QCFG_OUTPUT_LINK_SPEED_20GB: - ifmr->ifm_active |= IFM_20G_KR2; - break; - case HWRM_PORT_PHY_QCFG_OUTPUT_LINK_SPEED_25GB: - switch (phy_type) { - case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASECR: - ifmr->ifm_active |= IFM_25G_CR; - break; - case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASEKR4: - case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASEKR2: - case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASEKR: - ifmr->ifm_active |= IFM_25G_KR; - break; - case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASESR: - ifmr->ifm_active |= IFM_25G_SR; - break; - default: - /* - * Workaround: - * Don't return IFM_UNKNOWN until - * Stratus return proper media_type - */ - ifmr->ifm_active |= IFM_25G_CR; - break; - } - break; - case HWRM_PORT_PHY_QCFG_OUTPUT_LINK_SPEED_40GB: - switch (phy_type) { - case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASECR: - ifmr->ifm_active |= IFM_40G_CR4; - break; - case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASEKR4: - case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASEKR2: - case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASEKR: - ifmr->ifm_active |= IFM_40G_KR4; - break; - case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASELR: - ifmr->ifm_active |= IFM_40G_LR4; - break; - case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASESR: - ifmr->ifm_active |= IFM_40G_SR4; - break; - default: - ifmr->ifm_active |= IFM_UNKNOWN; - break; - } - break; - case HWRM_PORT_PHY_QCFG_OUTPUT_LINK_SPEED_50GB: - switch (phy_type) { - case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASECR: - ifmr->ifm_active |= IFM_50G_CR2; - break; - case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASEKR4: - case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASEKR2: - case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASEKR: - ifmr->ifm_active |= IFM_50G_KR2; - break; - default: - /* - * Workaround: - * Don't return IFM_UNKNOWN until - * Stratus return proper media_type - */ - ifmr->ifm_active |= IFM_50G_CR2; - break; - } - break; - case HWRM_PORT_PHY_QCFG_OUTPUT_LINK_SPEED_100GB: - switch (phy_type) { - case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASECR: - ifmr->ifm_active |= IFM_100G_CR4; - break; - case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASEKR4: - case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASEKR2: - case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASEKR: - ifmr->ifm_active |= IFM_100G_KR4; - break; - case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASELR: - ifmr->ifm_active |= IFM_100G_LR4; - break; - case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASESR: - ifmr->ifm_active |= IFM_100G_SR4; - break; - default: - /* - * Workaround: - * Don't return IFM_UNKNOWN until - * Stratus return proper media_type - */ - ifmr->ifm_active |= IFM_100G_CR4; - break; - } - default: - return; } + ifmr->ifm_active |= active_media; if (link_info->flow_ctrl.rx) ifmr->ifm_active |= IFM_ETH_RXPAUSE; if (link_info->flow_ctrl.tx) ifmr->ifm_active |= IFM_ETH_TXPAUSE; bnxt_report_link(softc); return; } static int bnxt_media_change(if_ctx_t ctx) { struct bnxt_softc *softc = iflib_get_softc(ctx); struct ifmedia *ifm = iflib_get_media(ctx); struct ifmediareq ifmr; int rc; if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER) return EINVAL; switch (IFM_SUBTYPE(ifm->ifm_media)) { case IFM_100_T: softc->link_info.autoneg &= ~BNXT_AUTONEG_SPEED; softc->link_info.req_link_speed = HWRM_PORT_PHY_CFG_INPUT_FORCE_LINK_SPEED_100MB; break; case IFM_1000_KX: case IFM_1000_T: case IFM_1000_SGMII: softc->link_info.autoneg &= ~BNXT_AUTONEG_SPEED; softc->link_info.req_link_speed = HWRM_PORT_PHY_CFG_INPUT_FORCE_LINK_SPEED_1GB; break; case IFM_2500_KX: case IFM_2500_T: softc->link_info.autoneg &= ~BNXT_AUTONEG_SPEED; softc->link_info.req_link_speed = HWRM_PORT_PHY_CFG_INPUT_FORCE_LINK_SPEED_2_5GB; break; case IFM_10G_CR1: case IFM_10G_KR: case IFM_10G_LR: case IFM_10G_SR: case IFM_10G_T: softc->link_info.autoneg &= ~BNXT_AUTONEG_SPEED; softc->link_info.req_link_speed = HWRM_PORT_PHY_CFG_INPUT_FORCE_LINK_SPEED_10GB; break; case IFM_20G_KR2: softc->link_info.autoneg &= ~BNXT_AUTONEG_SPEED; softc->link_info.req_link_speed = HWRM_PORT_PHY_CFG_INPUT_FORCE_LINK_SPEED_20GB; break; case IFM_25G_CR: case IFM_25G_KR: case IFM_25G_SR: softc->link_info.autoneg &= ~BNXT_AUTONEG_SPEED; softc->link_info.req_link_speed = HWRM_PORT_PHY_CFG_INPUT_FORCE_LINK_SPEED_25GB; break; case IFM_40G_CR4: case IFM_40G_KR4: case IFM_40G_LR4: case IFM_40G_SR4: softc->link_info.autoneg &= ~BNXT_AUTONEG_SPEED; softc->link_info.req_link_speed = HWRM_PORT_PHY_CFG_INPUT_FORCE_LINK_SPEED_40GB; break; case IFM_50G_CR2: case IFM_50G_KR2: softc->link_info.autoneg &= ~BNXT_AUTONEG_SPEED; softc->link_info.req_link_speed = HWRM_PORT_PHY_CFG_INPUT_FORCE_LINK_SPEED_50GB; break; case IFM_100G_CR4: case IFM_100G_KR4: case IFM_100G_LR4: case IFM_100G_SR4: softc->link_info.autoneg &= ~BNXT_AUTONEG_SPEED; softc->link_info.req_link_speed = HWRM_PORT_PHY_CFG_INPUT_FORCE_LINK_SPEED_100GB; break; default: device_printf(softc->dev, "Unsupported media type! Using auto\n"); /* Fall-through */ case IFM_AUTO: // Auto softc->link_info.autoneg |= BNXT_AUTONEG_SPEED; break; } rc = bnxt_hwrm_set_link_setting(softc, true, true, true); bnxt_media_status(softc->ctx, &ifmr); return rc; } static int bnxt_promisc_set(if_ctx_t ctx, int flags) { struct bnxt_softc *softc = iflib_get_softc(ctx); if_t ifp = iflib_get_ifp(ctx); int rc; if (ifp->if_flags & IFF_ALLMULTI || if_multiaddr_count(ifp, -1) > BNXT_MAX_MC_ADDRS) softc->vnic_info.rx_mask |= HWRM_CFA_L2_SET_RX_MASK_INPUT_MASK_ALL_MCAST; else softc->vnic_info.rx_mask &= ~HWRM_CFA_L2_SET_RX_MASK_INPUT_MASK_ALL_MCAST; if (ifp->if_flags & IFF_PROMISC) softc->vnic_info.rx_mask |= HWRM_CFA_L2_SET_RX_MASK_INPUT_MASK_PROMISCUOUS | HWRM_CFA_L2_SET_RX_MASK_INPUT_MASK_ANYVLAN_NONVLAN; else softc->vnic_info.rx_mask &= ~(HWRM_CFA_L2_SET_RX_MASK_INPUT_MASK_PROMISCUOUS | HWRM_CFA_L2_SET_RX_MASK_INPUT_MASK_ANYVLAN_NONVLAN); rc = bnxt_hwrm_cfa_l2_set_rx_mask(softc, &softc->vnic_info); return rc; } static uint64_t bnxt_get_counter(if_ctx_t ctx, ift_counter cnt) { if_t ifp = iflib_get_ifp(ctx); if (cnt < IFCOUNTERS) return if_get_counter_default(ifp, cnt); return 0; } static void bnxt_update_admin_status(if_ctx_t ctx) { struct bnxt_softc *softc = iflib_get_softc(ctx); /* * When SR-IOV is enabled, avoid each VF sending this HWRM * request every sec with which firmware timeouts can happen */ if (BNXT_PF(softc)) { bnxt_hwrm_port_qstats(softc); } return; } static void bnxt_if_timer(if_ctx_t ctx, uint16_t qid) { struct bnxt_softc *softc = iflib_get_softc(ctx); uint64_t ticks_now = ticks; /* Schedule bnxt_update_admin_status() once per sec */ if (ticks_now - softc->admin_ticks >= hz) { softc->admin_ticks = ticks_now; iflib_admin_intr_deferred(ctx); } return; } static void inline bnxt_do_enable_intr(struct bnxt_cp_ring *cpr) { if (cpr->ring.phys_id != (uint16_t)HWRM_NA_SIGNATURE) { /* First time enabling, do not set index */ if (cpr->cons == UINT32_MAX) BNXT_CP_ENABLE_DB(&cpr->ring); else BNXT_CP_IDX_ENABLE_DB(&cpr->ring, cpr->cons); } } static void inline bnxt_do_disable_intr(struct bnxt_cp_ring *cpr) { if (cpr->ring.phys_id != (uint16_t)HWRM_NA_SIGNATURE) BNXT_CP_DISABLE_DB(&cpr->ring); } /* Enable all interrupts */ static void bnxt_intr_enable(if_ctx_t ctx) { struct bnxt_softc *softc = iflib_get_softc(ctx); int i; bnxt_do_enable_intr(&softc->def_cp_ring); for (i = 0; i < softc->nrxqsets; i++) bnxt_do_enable_intr(&softc->rx_cp_rings[i]); return; } /* Enable interrupt for a single queue */ static int bnxt_tx_queue_intr_enable(if_ctx_t ctx, uint16_t qid) { struct bnxt_softc *softc = iflib_get_softc(ctx); bnxt_do_enable_intr(&softc->tx_cp_rings[qid]); return 0; } static int bnxt_rx_queue_intr_enable(if_ctx_t ctx, uint16_t qid) { struct bnxt_softc *softc = iflib_get_softc(ctx); bnxt_do_enable_intr(&softc->rx_cp_rings[qid]); return 0; } /* Disable all interrupts */ static void bnxt_disable_intr(if_ctx_t ctx) { struct bnxt_softc *softc = iflib_get_softc(ctx); int i; /* * NOTE: These TX interrupts should never get enabled, so don't * update the index */ for (i = 0; i < softc->ntxqsets; i++) bnxt_do_disable_intr(&softc->tx_cp_rings[i]); for (i = 0; i < softc->nrxqsets; i++) bnxt_do_disable_intr(&softc->rx_cp_rings[i]); return; } static int bnxt_msix_intr_assign(if_ctx_t ctx, int msix) { struct bnxt_softc *softc = iflib_get_softc(ctx); int rc; int i; char irq_name[16]; rc = iflib_irq_alloc_generic(ctx, &softc->def_cp_ring.irq, softc->def_cp_ring.ring.id + 1, IFLIB_INTR_ADMIN, bnxt_handle_def_cp, softc, 0, "def_cp"); if (rc) { device_printf(iflib_get_dev(ctx), "Failed to register default completion ring handler\n"); return rc; } for (i=0; iscctx->isc_nrxqsets; i++) { snprintf(irq_name, sizeof(irq_name), "rxq%d", i); rc = iflib_irq_alloc_generic(ctx, &softc->rx_cp_rings[i].irq, softc->rx_cp_rings[i].ring.id + 1, IFLIB_INTR_RX, bnxt_handle_rx_cp, &softc->rx_cp_rings[i], i, irq_name); if (rc) { device_printf(iflib_get_dev(ctx), "Failed to register RX completion ring handler\n"); i--; goto fail; } } for (i=0; iscctx->isc_ntxqsets; i++) iflib_softirq_alloc_generic(ctx, NULL, IFLIB_INTR_TX, NULL, i, "tx_cp"); return rc; fail: for (; i>=0; i--) iflib_irq_free(ctx, &softc->rx_cp_rings[i].irq); iflib_irq_free(ctx, &softc->def_cp_ring.irq); return rc; } /* * We're explicitly allowing duplicates here. They will need to be * removed as many times as they are added. */ static void bnxt_vlan_register(if_ctx_t ctx, uint16_t vtag) { struct bnxt_softc *softc = iflib_get_softc(ctx); struct bnxt_vlan_tag *new_tag; new_tag = malloc(sizeof(struct bnxt_vlan_tag), M_DEVBUF, M_NOWAIT); if (new_tag == NULL) return; new_tag->tag = vtag; new_tag->tpid = 8100; SLIST_INSERT_HEAD(&softc->vnic_info.vlan_tags, new_tag, next); }; static void bnxt_vlan_unregister(if_ctx_t ctx, uint16_t vtag) { struct bnxt_softc *softc = iflib_get_softc(ctx); struct bnxt_vlan_tag *vlan_tag; SLIST_FOREACH(vlan_tag, &softc->vnic_info.vlan_tags, next) { if (vlan_tag->tag == vtag) { SLIST_REMOVE(&softc->vnic_info.vlan_tags, vlan_tag, bnxt_vlan_tag, next); free(vlan_tag, M_DEVBUF); break; } } } static int bnxt_wol_config(if_ctx_t ctx) { struct bnxt_softc *softc = iflib_get_softc(ctx); if_t ifp = iflib_get_ifp(ctx); if (!softc) return -EBUSY; if (!bnxt_wol_supported(softc)) return -ENOTSUP; if (if_getcapabilities(ifp) & IFCAP_WOL_MAGIC) { if (!softc->wol) { if (bnxt_hwrm_alloc_wol_fltr(softc)) return -EBUSY; softc->wol = 1; } } else { if (softc->wol) { if (bnxt_hwrm_free_wol_fltr(softc)) return -EBUSY; softc->wol = 0; } } return 0; } static int bnxt_shutdown(if_ctx_t ctx) { bnxt_wol_config(ctx); return 0; } static int bnxt_suspend(if_ctx_t ctx) { bnxt_wol_config(ctx); return 0; } static int bnxt_resume(if_ctx_t ctx) { struct bnxt_softc *softc = iflib_get_softc(ctx); bnxt_get_wol_settings(softc); return 0; } static int bnxt_priv_ioctl(if_ctx_t ctx, u_long command, caddr_t data) { struct bnxt_softc *softc = iflib_get_softc(ctx); struct ifreq *ifr = (struct ifreq *)data; struct ifreq_buffer *ifbuf = &ifr->ifr_ifru.ifru_buffer; struct bnxt_ioctl_header *ioh = (struct bnxt_ioctl_header *)(ifbuf->buffer); int rc = ENOTSUP; struct bnxt_ioctl_data *iod = NULL; switch (command) { case SIOCGPRIVATE_0: if ((rc = priv_check(curthread, PRIV_DRIVER)) != 0) goto exit; iod = malloc(ifbuf->length, M_DEVBUF, M_NOWAIT | M_ZERO); if (!iod) { rc = ENOMEM; goto exit; } copyin(ioh, iod, ifbuf->length); switch (ioh->type) { case BNXT_HWRM_NVM_FIND_DIR_ENTRY: { struct bnxt_ioctl_hwrm_nvm_find_dir_entry *find = &iod->find; rc = bnxt_hwrm_nvm_find_dir_entry(softc, find->type, &find->ordinal, find->ext, &find->index, find->use_index, find->search_opt, &find->data_length, &find->item_length, &find->fw_ver); if (rc) { iod->hdr.rc = rc; copyout(&iod->hdr.rc, &ioh->rc, sizeof(ioh->rc)); } else { iod->hdr.rc = 0; copyout(iod, ioh, ifbuf->length); } rc = 0; goto exit; } case BNXT_HWRM_NVM_READ: { struct bnxt_ioctl_hwrm_nvm_read *rd = &iod->read; struct iflib_dma_info dma_data; size_t offset; size_t remain; size_t csize; /* * Some HWRM versions can't read more than 0x8000 bytes */ rc = iflib_dma_alloc(softc->ctx, min(rd->length, 0x8000), &dma_data, BUS_DMA_NOWAIT); if (rc) break; for (remain = rd->length, offset = 0; remain && offset < rd->length; offset += 0x8000) { csize = min(remain, 0x8000); rc = bnxt_hwrm_nvm_read(softc, rd->index, rd->offset + offset, csize, &dma_data); if (rc) { iod->hdr.rc = rc; copyout(&iod->hdr.rc, &ioh->rc, sizeof(ioh->rc)); break; } else { copyout(dma_data.idi_vaddr, rd->data + offset, csize); iod->hdr.rc = 0; } remain -= csize; } if (iod->hdr.rc == 0) copyout(iod, ioh, ifbuf->length); iflib_dma_free(&dma_data); rc = 0; goto exit; } case BNXT_HWRM_FW_RESET: { struct bnxt_ioctl_hwrm_fw_reset *rst = &iod->reset; rc = bnxt_hwrm_fw_reset(softc, rst->processor, &rst->selfreset); if (rc) { iod->hdr.rc = rc; copyout(&iod->hdr.rc, &ioh->rc, sizeof(ioh->rc)); } else { iod->hdr.rc = 0; copyout(iod, ioh, ifbuf->length); } rc = 0; goto exit; } case BNXT_HWRM_FW_QSTATUS: { struct bnxt_ioctl_hwrm_fw_qstatus *qstat = &iod->status; rc = bnxt_hwrm_fw_qstatus(softc, qstat->processor, &qstat->selfreset); if (rc) { iod->hdr.rc = rc; copyout(&iod->hdr.rc, &ioh->rc, sizeof(ioh->rc)); } else { iod->hdr.rc = 0; copyout(iod, ioh, ifbuf->length); } rc = 0; goto exit; } case BNXT_HWRM_NVM_WRITE: { struct bnxt_ioctl_hwrm_nvm_write *wr = &iod->write; rc = bnxt_hwrm_nvm_write(softc, wr->data, true, wr->type, wr->ordinal, wr->ext, wr->attr, wr->option, wr->data_length, wr->keep, &wr->item_length, &wr->index); if (rc) { iod->hdr.rc = rc; copyout(&iod->hdr.rc, &ioh->rc, sizeof(ioh->rc)); } else { iod->hdr.rc = 0; copyout(iod, ioh, ifbuf->length); } rc = 0; goto exit; } case BNXT_HWRM_NVM_ERASE_DIR_ENTRY: { struct bnxt_ioctl_hwrm_nvm_erase_dir_entry *erase = &iod->erase; rc = bnxt_hwrm_nvm_erase_dir_entry(softc, erase->index); if (rc) { iod->hdr.rc = rc; copyout(&iod->hdr.rc, &ioh->rc, sizeof(ioh->rc)); } else { iod->hdr.rc = 0; copyout(iod, ioh, ifbuf->length); } rc = 0; goto exit; } case BNXT_HWRM_NVM_GET_DIR_INFO: { struct bnxt_ioctl_hwrm_nvm_get_dir_info *info = &iod->dir_info; rc = bnxt_hwrm_nvm_get_dir_info(softc, &info->entries, &info->entry_length); if (rc) { iod->hdr.rc = rc; copyout(&iod->hdr.rc, &ioh->rc, sizeof(ioh->rc)); } else { iod->hdr.rc = 0; copyout(iod, ioh, ifbuf->length); } rc = 0; goto exit; } case BNXT_HWRM_NVM_GET_DIR_ENTRIES: { struct bnxt_ioctl_hwrm_nvm_get_dir_entries *get = &iod->dir_entries; struct iflib_dma_info dma_data; rc = iflib_dma_alloc(softc->ctx, get->max_size, &dma_data, BUS_DMA_NOWAIT); if (rc) break; rc = bnxt_hwrm_nvm_get_dir_entries(softc, &get->entries, &get->entry_length, &dma_data); if (rc) { iod->hdr.rc = rc; copyout(&iod->hdr.rc, &ioh->rc, sizeof(ioh->rc)); } else { copyout(dma_data.idi_vaddr, get->data, get->entry_length * get->entries); iod->hdr.rc = 0; copyout(iod, ioh, ifbuf->length); } iflib_dma_free(&dma_data); rc = 0; goto exit; } case BNXT_HWRM_NVM_VERIFY_UPDATE: { struct bnxt_ioctl_hwrm_nvm_verify_update *vrfy = &iod->verify; rc = bnxt_hwrm_nvm_verify_update(softc, vrfy->type, vrfy->ordinal, vrfy->ext); if (rc) { iod->hdr.rc = rc; copyout(&iod->hdr.rc, &ioh->rc, sizeof(ioh->rc)); } else { iod->hdr.rc = 0; copyout(iod, ioh, ifbuf->length); } rc = 0; goto exit; } case BNXT_HWRM_NVM_INSTALL_UPDATE: { struct bnxt_ioctl_hwrm_nvm_install_update *inst = &iod->install; rc = bnxt_hwrm_nvm_install_update(softc, inst->install_type, &inst->installed_items, &inst->result, &inst->problem_item, &inst->reset_required); if (rc) { iod->hdr.rc = rc; copyout(&iod->hdr.rc, &ioh->rc, sizeof(ioh->rc)); } else { iod->hdr.rc = 0; copyout(iod, ioh, ifbuf->length); } rc = 0; goto exit; } case BNXT_HWRM_NVM_MODIFY: { struct bnxt_ioctl_hwrm_nvm_modify *mod = &iod->modify; rc = bnxt_hwrm_nvm_modify(softc, mod->index, mod->offset, mod->data, true, mod->length); if (rc) { iod->hdr.rc = rc; copyout(&iod->hdr.rc, &ioh->rc, sizeof(ioh->rc)); } else { iod->hdr.rc = 0; copyout(iod, ioh, ifbuf->length); } rc = 0; goto exit; } case BNXT_HWRM_FW_GET_TIME: { struct bnxt_ioctl_hwrm_fw_get_time *gtm = &iod->get_time; rc = bnxt_hwrm_fw_get_time(softc, >m->year, >m->month, >m->day, >m->hour, >m->minute, >m->second, >m->millisecond, >m->zone); if (rc) { iod->hdr.rc = rc; copyout(&iod->hdr.rc, &ioh->rc, sizeof(ioh->rc)); } else { iod->hdr.rc = 0; copyout(iod, ioh, ifbuf->length); } rc = 0; goto exit; } case BNXT_HWRM_FW_SET_TIME: { struct bnxt_ioctl_hwrm_fw_set_time *stm = &iod->set_time; rc = bnxt_hwrm_fw_set_time(softc, stm->year, stm->month, stm->day, stm->hour, stm->minute, stm->second, stm->millisecond, stm->zone); if (rc) { iod->hdr.rc = rc; copyout(&iod->hdr.rc, &ioh->rc, sizeof(ioh->rc)); } else { iod->hdr.rc = 0; copyout(iod, ioh, ifbuf->length); } rc = 0; goto exit; } } break; } exit: if (iod) free(iod, M_DEVBUF); return rc; } /* * Support functions */ static int bnxt_probe_phy(struct bnxt_softc *softc) { struct bnxt_link_info *link_info = &softc->link_info; int rc = 0; rc = bnxt_update_link(softc, false); if (rc) { device_printf(softc->dev, "Probe phy can't update link (rc: %x)\n", rc); return (rc); } /*initialize the ethool setting copy with NVM settings */ if (link_info->auto_mode != HWRM_PORT_PHY_QCFG_OUTPUT_AUTO_MODE_NONE) link_info->autoneg |= BNXT_AUTONEG_SPEED; link_info->req_duplex = link_info->duplex_setting; if (link_info->autoneg & BNXT_AUTONEG_SPEED) link_info->req_link_speed = link_info->auto_link_speed; else link_info->req_link_speed = link_info->force_link_speed; return (rc); } static void bnxt_add_media_types(struct bnxt_softc *softc) { struct bnxt_link_info *link_info = &softc->link_info; uint16_t supported; uint8_t phy_type = get_phy_type(softc); supported = link_info->support_speeds; /* Auto is always supported */ ifmedia_add(softc->media, IFM_ETHER | IFM_AUTO, 0, NULL); if (softc->flags & BNXT_FLAG_NPAR) return; switch (phy_type) { case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_100G_BASECR4: case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_40G_BASECR4: case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_25G_BASECR_CA_L: case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_25G_BASECR_CA_S: case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_25G_BASECR_CA_N: case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASECR: BNXT_IFMEDIA_ADD(supported, SPEEDS_100GB, IFM_100G_CR4); BNXT_IFMEDIA_ADD(supported, SPEEDS_50GB, IFM_50G_CR2); BNXT_IFMEDIA_ADD(supported, SPEEDS_40GB, IFM_40G_CR4); BNXT_IFMEDIA_ADD(supported, SPEEDS_25GB, IFM_25G_CR); BNXT_IFMEDIA_ADD(supported, SPEEDS_10GB, IFM_10G_CR1); BNXT_IFMEDIA_ADD(supported, SPEEDS_1GB, IFM_1000_T); break; case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_100G_BASELR4: case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_40G_BASELR4: case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASELR: BNXT_IFMEDIA_ADD(supported, SPEEDS_100GB, IFM_100G_LR4); BNXT_IFMEDIA_ADD(supported, SPEEDS_40GB, IFM_40G_LR4); BNXT_IFMEDIA_ADD(supported, SPEEDS_25GB, IFM_25G_LR); BNXT_IFMEDIA_ADD(supported, SPEEDS_10GB, IFM_10G_LR); BNXT_IFMEDIA_ADD(supported, SPEEDS_1GB, IFM_1000_LX); break; case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_100G_BASESR10: case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_100G_BASESR4: case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_40G_BASESR4: case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASESR: case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_40G_BASEER4: case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_100G_BASEER4: case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_25G_BASESR: case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_1G_BASESX: BNXT_IFMEDIA_ADD(supported, SPEEDS_100GB, IFM_100G_SR4); BNXT_IFMEDIA_ADD(supported, SPEEDS_40GB, IFM_40G_SR4); BNXT_IFMEDIA_ADD(supported, SPEEDS_25GB, IFM_25G_SR); BNXT_IFMEDIA_ADD(supported, SPEEDS_10GB, IFM_10G_SR); BNXT_IFMEDIA_ADD(supported, SPEEDS_1GB, IFM_1000_SX); break; case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASEKR4: case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASEKR2: case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASEKR: BNXT_IFMEDIA_ADD(supported, SPEEDS_100GB, IFM_100G_KR4); BNXT_IFMEDIA_ADD(supported, SPEEDS_50GB, IFM_50G_KR2); BNXT_IFMEDIA_ADD(supported, SPEEDS_40GB, IFM_40G_KR4); BNXT_IFMEDIA_ADD(supported, SPEEDS_25GB, IFM_25G_KR); BNXT_IFMEDIA_ADD(supported, SPEEDS_20GB, IFM_20G_KR2); BNXT_IFMEDIA_ADD(supported, SPEEDS_10GB, IFM_10G_KR); BNXT_IFMEDIA_ADD(supported, SPEEDS_1GB, IFM_1000_KX); break; case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_40G_ACTIVE_CABLE: BNXT_IFMEDIA_ADD(supported, SPEEDS_25GB, IFM_25G_ACC); BNXT_IFMEDIA_ADD(supported, SPEEDS_10GB, IFM_10G_AOC); break; case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_1G_BASECX: BNXT_IFMEDIA_ADD(supported, SPEEDS_1GBHD, IFM_1000_CX); break; case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_1G_BASET: case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASET: case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASETE: + BNXT_IFMEDIA_ADD(supported, SPEEDS_10GB, IFM_10G_T); + BNXT_IFMEDIA_ADD(supported, SPEEDS_2_5GB, IFM_2500_T); BNXT_IFMEDIA_ADD(supported, SPEEDS_1GB, IFM_1000_T); BNXT_IFMEDIA_ADD(supported, SPEEDS_100MB, IFM_100_T); BNXT_IFMEDIA_ADD(supported, SPEEDS_10MB, IFM_10_T); break; case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASEKX: BNXT_IFMEDIA_ADD(supported, SPEEDS_10GB, IFM_10G_KR); + BNXT_IFMEDIA_ADD(supported, SPEEDS_2_5GB, IFM_2500_KX); BNXT_IFMEDIA_ADD(supported, SPEEDS_1GB, IFM_1000_KX); break; case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_SGMIIEXTPHY: BNXT_IFMEDIA_ADD(supported, SPEEDS_1GB, IFM_1000_SGMII); break; case HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_UNKNOWN: default: /* Only Autoneg is supported for TYPE_UNKNOWN */ break; } return; } static int bnxt_map_bar(struct bnxt_softc *softc, struct bnxt_bar_info *bar, int bar_num, bool shareable) { uint32_t flag; if (bar->res != NULL) { device_printf(softc->dev, "Bar %d already mapped\n", bar_num); return EDOOFUS; } bar->rid = PCIR_BAR(bar_num); flag = RF_ACTIVE; if (shareable) flag |= RF_SHAREABLE; if ((bar->res = bus_alloc_resource_any(softc->dev, SYS_RES_MEMORY, &bar->rid, flag)) == NULL) { device_printf(softc->dev, "PCI BAR%d mapping failure\n", bar_num); return (ENXIO); } bar->tag = rman_get_bustag(bar->res); bar->handle = rman_get_bushandle(bar->res); bar->size = rman_get_size(bar->res); return 0; } static int bnxt_pci_mapping(struct bnxt_softc *softc) { int rc; rc = bnxt_map_bar(softc, &softc->hwrm_bar, 0, true); if (rc) return rc; rc = bnxt_map_bar(softc, &softc->doorbell_bar, 2, false); return rc; } static void bnxt_pci_mapping_free(struct bnxt_softc *softc) { if (softc->hwrm_bar.res != NULL) bus_release_resource(softc->dev, SYS_RES_MEMORY, softc->hwrm_bar.rid, softc->hwrm_bar.res); softc->hwrm_bar.res = NULL; if (softc->doorbell_bar.res != NULL) bus_release_resource(softc->dev, SYS_RES_MEMORY, softc->doorbell_bar.rid, softc->doorbell_bar.res); softc->doorbell_bar.res = NULL; } static int bnxt_update_link(struct bnxt_softc *softc, bool chng_link_state) { struct bnxt_link_info *link_info = &softc->link_info; uint8_t link_up = link_info->link_up; int rc = 0; rc = bnxt_hwrm_port_phy_qcfg(softc); if (rc) goto exit; /* TODO: need to add more logic to report VF link */ if (chng_link_state) { if (link_info->phy_link_status == HWRM_PORT_PHY_QCFG_OUTPUT_LINK_LINK) link_info->link_up = 1; else link_info->link_up = 0; if (link_up != link_info->link_up) bnxt_report_link(softc); } else { /* always link down if not require to update link state */ link_info->link_up = 0; } exit: return rc; } void bnxt_report_link(struct bnxt_softc *softc) { struct bnxt_link_info *link_info = &softc->link_info; const char *duplex = NULL, *flow_ctrl = NULL; if (link_info->link_up == link_info->last_link_up) { if (!link_info->link_up) return; if ((link_info->duplex == link_info->last_duplex) && (!(BNXT_IS_FLOW_CTRL_CHANGED(link_info)))) return; } if (link_info->link_up) { if (link_info->duplex == HWRM_PORT_PHY_QCFG_OUTPUT_DUPLEX_CFG_FULL) duplex = "full duplex"; else duplex = "half duplex"; if (link_info->flow_ctrl.tx & link_info->flow_ctrl.rx) flow_ctrl = "FC - receive & transmit"; else if (link_info->flow_ctrl.tx) flow_ctrl = "FC - transmit"; else if (link_info->flow_ctrl.rx) flow_ctrl = "FC - receive"; else flow_ctrl = "FC - none"; iflib_link_state_change(softc->ctx, LINK_STATE_UP, IF_Gbps(100)); device_printf(softc->dev, "Link is UP %s, %s - %d Mbps \n", duplex, flow_ctrl, (link_info->link_speed * 100)); } else { iflib_link_state_change(softc->ctx, LINK_STATE_DOWN, bnxt_get_baudrate(&softc->link_info)); device_printf(softc->dev, "Link is Down\n"); } link_info->last_link_up = link_info->link_up; link_info->last_duplex = link_info->duplex; link_info->last_flow_ctrl.tx = link_info->flow_ctrl.tx; link_info->last_flow_ctrl.rx = link_info->flow_ctrl.rx; link_info->last_flow_ctrl.autoneg = link_info->flow_ctrl.autoneg; } static int bnxt_handle_rx_cp(void *arg) { struct bnxt_cp_ring *cpr = arg; /* Disable further interrupts for this queue */ BNXT_CP_DISABLE_DB(&cpr->ring); return FILTER_SCHEDULE_THREAD; } static int bnxt_handle_def_cp(void *arg) { struct bnxt_softc *softc = arg; BNXT_CP_DISABLE_DB(&softc->def_cp_ring.ring); GROUPTASK_ENQUEUE(&softc->def_cp_task); return FILTER_HANDLED; } static void bnxt_clear_ids(struct bnxt_softc *softc) { int i; softc->def_cp_ring.stats_ctx_id = HWRM_NA_SIGNATURE; softc->def_cp_ring.ring.phys_id = (uint16_t)HWRM_NA_SIGNATURE; for (i = 0; i < softc->ntxqsets; i++) { softc->tx_cp_rings[i].stats_ctx_id = HWRM_NA_SIGNATURE; softc->tx_cp_rings[i].ring.phys_id = (uint16_t)HWRM_NA_SIGNATURE; softc->tx_rings[i].phys_id = (uint16_t)HWRM_NA_SIGNATURE; } for (i = 0; i < softc->nrxqsets; i++) { softc->rx_cp_rings[i].stats_ctx_id = HWRM_NA_SIGNATURE; softc->rx_cp_rings[i].ring.phys_id = (uint16_t)HWRM_NA_SIGNATURE; softc->rx_rings[i].phys_id = (uint16_t)HWRM_NA_SIGNATURE; softc->ag_rings[i].phys_id = (uint16_t)HWRM_NA_SIGNATURE; softc->grp_info[i].grp_id = (uint16_t)HWRM_NA_SIGNATURE; } softc->vnic_info.filter_id = -1; softc->vnic_info.id = (uint16_t)HWRM_NA_SIGNATURE; softc->vnic_info.rss_id = (uint16_t)HWRM_NA_SIGNATURE; memset(softc->vnic_info.rss_grp_tbl.idi_vaddr, 0xff, softc->vnic_info.rss_grp_tbl.idi_size); } static void bnxt_mark_cpr_invalid(struct bnxt_cp_ring *cpr) { struct cmpl_base *cmp = (void *)cpr->ring.vaddr; int i; for (i = 0; i < cpr->ring.ring_size; i++) cmp[i].info3_v = !cpr->v_bit; } static void bnxt_handle_async_event(struct bnxt_softc *softc, struct cmpl_base *cmpl) { struct hwrm_async_event_cmpl *ae = (void *)cmpl; uint16_t async_id = le16toh(ae->event_id); struct ifmediareq ifmr; switch (async_id) { case HWRM_ASYNC_EVENT_CMPL_EVENT_ID_LINK_STATUS_CHANGE: case HWRM_ASYNC_EVENT_CMPL_EVENT_ID_LINK_SPEED_CHANGE: case HWRM_ASYNC_EVENT_CMPL_EVENT_ID_LINK_SPEED_CFG_CHANGE: bnxt_media_status(softc->ctx, &ifmr); break; case HWRM_ASYNC_EVENT_CMPL_EVENT_ID_LINK_MTU_CHANGE: case HWRM_ASYNC_EVENT_CMPL_EVENT_ID_DCB_CONFIG_CHANGE: case HWRM_ASYNC_EVENT_CMPL_EVENT_ID_PORT_CONN_NOT_ALLOWED: case HWRM_ASYNC_EVENT_CMPL_EVENT_ID_LINK_SPEED_CFG_NOT_ALLOWED: case HWRM_ASYNC_EVENT_CMPL_EVENT_ID_FUNC_DRVR_UNLOAD: case HWRM_ASYNC_EVENT_CMPL_EVENT_ID_FUNC_DRVR_LOAD: case HWRM_ASYNC_EVENT_CMPL_EVENT_ID_PF_DRVR_UNLOAD: case HWRM_ASYNC_EVENT_CMPL_EVENT_ID_PF_DRVR_LOAD: case HWRM_ASYNC_EVENT_CMPL_EVENT_ID_VF_FLR: case HWRM_ASYNC_EVENT_CMPL_EVENT_ID_VF_MAC_ADDR_CHANGE: case HWRM_ASYNC_EVENT_CMPL_EVENT_ID_PF_VF_COMM_STATUS_CHANGE: case HWRM_ASYNC_EVENT_CMPL_EVENT_ID_VF_CFG_CHANGE: case HWRM_ASYNC_EVENT_CMPL_EVENT_ID_HWRM_ERROR: device_printf(softc->dev, "Unhandled async completion type %u\n", async_id); break; default: device_printf(softc->dev, "Unknown async completion type %u\n", async_id); break; } } static void bnxt_def_cp_task(void *context) { if_ctx_t ctx = context; struct bnxt_softc *softc = iflib_get_softc(ctx); struct bnxt_cp_ring *cpr = &softc->def_cp_ring; /* Handle completions on the default completion ring */ struct cmpl_base *cmpl; uint32_t cons = cpr->cons; bool v_bit = cpr->v_bit; bool last_v_bit; uint32_t last_cons; uint16_t type; for (;;) { last_cons = cons; last_v_bit = v_bit; NEXT_CP_CONS_V(&cpr->ring, cons, v_bit); cmpl = &((struct cmpl_base *)cpr->ring.vaddr)[cons]; if (!CMP_VALID(cmpl, v_bit)) break; type = le16toh(cmpl->type) & CMPL_BASE_TYPE_MASK; switch (type) { case CMPL_BASE_TYPE_HWRM_ASYNC_EVENT: bnxt_handle_async_event(softc, cmpl); break; case CMPL_BASE_TYPE_TX_L2: case CMPL_BASE_TYPE_RX_L2: case CMPL_BASE_TYPE_RX_AGG: case CMPL_BASE_TYPE_RX_TPA_START: case CMPL_BASE_TYPE_RX_TPA_END: case CMPL_BASE_TYPE_STAT_EJECT: case CMPL_BASE_TYPE_HWRM_DONE: case CMPL_BASE_TYPE_HWRM_FWD_REQ: case CMPL_BASE_TYPE_HWRM_FWD_RESP: case CMPL_BASE_TYPE_CQ_NOTIFICATION: case CMPL_BASE_TYPE_SRQ_EVENT: case CMPL_BASE_TYPE_DBQ_EVENT: case CMPL_BASE_TYPE_QP_EVENT: case CMPL_BASE_TYPE_FUNC_EVENT: device_printf(softc->dev, "Unhandled completion type %u\n", type); break; default: device_printf(softc->dev, "Unknown completion type %u\n", type); break; } } cpr->cons = last_cons; cpr->v_bit = last_v_bit; BNXT_CP_IDX_ENABLE_DB(&cpr->ring, cpr->cons); } static uint8_t get_phy_type(struct bnxt_softc *softc) { struct bnxt_link_info *link_info = &softc->link_info; uint8_t phy_type = link_info->phy_type; uint16_t supported; if (phy_type != HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_UNKNOWN) return phy_type; /* Deduce the phy type from the media type and supported speeds */ supported = link_info->support_speeds; if (link_info->media_type == HWRM_PORT_PHY_QCFG_OUTPUT_MEDIA_TYPE_TP) return HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASET; if (link_info->media_type == HWRM_PORT_PHY_QCFG_OUTPUT_MEDIA_TYPE_DAC) { if (supported & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_2_5GB) return HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASEKX; if (supported & HWRM_PORT_PHY_QCFG_OUTPUT_SUPPORT_SPEEDS_20GB) return HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASEKR; return HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASECR; } if (link_info->media_type == HWRM_PORT_PHY_QCFG_OUTPUT_MEDIA_TYPE_FIBRE) return HWRM_PORT_PHY_QCFG_OUTPUT_PHY_TYPE_BASESR; return phy_type; } bool bnxt_check_hwrm_version(struct bnxt_softc *softc) { char buf[16]; sprintf(buf, "%hhu.%hhu.%hhu", softc->ver_info->hwrm_min_major, softc->ver_info->hwrm_min_minor, softc->ver_info->hwrm_min_update); if (softc->ver_info->hwrm_min_major > softc->ver_info->hwrm_if_major) { device_printf(softc->dev, "WARNING: HWRM version %s is too old (older than %s)\n", softc->ver_info->hwrm_if_ver, buf); return false; } else if(softc->ver_info->hwrm_min_major == softc->ver_info->hwrm_if_major) { if (softc->ver_info->hwrm_min_minor > softc->ver_info->hwrm_if_minor) { device_printf(softc->dev, "WARNING: HWRM version %s is too old (older than %s)\n", softc->ver_info->hwrm_if_ver, buf); return false; } else if (softc->ver_info->hwrm_min_minor == softc->ver_info->hwrm_if_minor) { if (softc->ver_info->hwrm_min_update > softc->ver_info->hwrm_if_update) { device_printf(softc->dev, "WARNING: HWRM version %s is too old (older than %s)\n", softc->ver_info->hwrm_if_ver, buf); return false; } } } return true; } static uint64_t bnxt_get_baudrate(struct bnxt_link_info *link) { switch (link->link_speed) { case HWRM_PORT_PHY_QCFG_OUTPUT_LINK_SPEED_100MB: return IF_Mbps(100); case HWRM_PORT_PHY_QCFG_OUTPUT_LINK_SPEED_1GB: return IF_Gbps(1); case HWRM_PORT_PHY_QCFG_OUTPUT_LINK_SPEED_2GB: return IF_Gbps(2); case HWRM_PORT_PHY_QCFG_OUTPUT_LINK_SPEED_2_5GB: return IF_Mbps(2500); case HWRM_PORT_PHY_QCFG_OUTPUT_LINK_SPEED_10GB: return IF_Gbps(10); case HWRM_PORT_PHY_QCFG_OUTPUT_LINK_SPEED_20GB: return IF_Gbps(20); case HWRM_PORT_PHY_QCFG_OUTPUT_LINK_SPEED_25GB: return IF_Gbps(25); case HWRM_PORT_PHY_QCFG_OUTPUT_LINK_SPEED_40GB: return IF_Gbps(40); case HWRM_PORT_PHY_QCFG_OUTPUT_LINK_SPEED_50GB: return IF_Gbps(50); case HWRM_PORT_PHY_QCFG_OUTPUT_LINK_SPEED_100GB: return IF_Gbps(100); case HWRM_PORT_PHY_QCFG_OUTPUT_LINK_SPEED_10MB: return IF_Mbps(10); } return IF_Gbps(100); } static void bnxt_get_wol_settings(struct bnxt_softc *softc) { uint16_t wol_handle = 0; if (!bnxt_wol_supported(softc)) return; do { wol_handle = bnxt_hwrm_get_wol_fltrs(softc, wol_handle); } while (wol_handle && wol_handle != BNXT_NO_MORE_WOL_FILTERS); } Index: projects/runtime-coverage/sys/dev/ixgbe/if_ix.c =================================================================== --- projects/runtime-coverage/sys/dev/ixgbe/if_ix.c (revision 325498) +++ projects/runtime-coverage/sys/dev/ixgbe/if_ix.c (revision 325499) @@ -1,5270 +1,5272 @@ /****************************************************************************** Copyright (c) 2001-2017, Intel Corporation All rights reserved. Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met: 1. Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer. 2. Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. 3. Neither the name of the Intel Corporation nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. ******************************************************************************/ /*$FreeBSD$*/ #ifndef IXGBE_STANDALONE_BUILD #include "opt_inet.h" #include "opt_inet6.h" #include "opt_rss.h" #endif #include "ixgbe.h" /************************************************************************ * Driver version ************************************************************************/ char ixgbe_driver_version[] = "3.2.12-k"; /************************************************************************ * PCI Device ID Table * * Used by probe to select devices to load on * Last field stores an index into ixgbe_strings * Last entry must be all 0s * * { Vendor ID, Device ID, SubVendor ID, SubDevice ID, String Index } ************************************************************************/ static ixgbe_vendor_info_t ixgbe_vendor_info_array[] = { {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82598AF_DUAL_PORT, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82598AF_SINGLE_PORT, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82598EB_CX4, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82598AT, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82598AT2, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82598, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82598_DA_DUAL_PORT, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82598_CX4_DUAL_PORT, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82598EB_XF_LR, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82598EB_SFP_LOM, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82599_KX4, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82599_KX4_MEZZ, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82599_SFP, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82599_XAUI_LOM, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82599_CX4, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82599_T3_LOM, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82599_COMBO_BACKPLANE, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82599_BACKPLANE_FCOE, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82599_SFP_SF2, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82599_SFP_FCOE, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82599EN_SFP, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82599_SFP_SF_QP, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82599_QSFP_SF_QP, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_X540T, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_X540T1, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_X550T, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_X550T1, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_X550EM_X_KR, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_X550EM_X_KX4, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_X550EM_X_10G_T, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_X550EM_X_1G_T, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_X550EM_X_SFP, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_X550EM_A_KR, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_X550EM_A_KR_L, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_X550EM_A_SFP, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_X550EM_A_SFP_N, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_X550EM_A_SGMII, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_X550EM_A_SGMII_L, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_X550EM_A_10G_T, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_X550EM_A_1G_T, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_X550EM_A_1G_T_L, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_X540_BYPASS, 0, 0, 0}, {IXGBE_INTEL_VENDOR_ID, IXGBE_DEV_ID_82599_BYPASS, 0, 0, 0}, /* required last entry */ {0, 0, 0, 0, 0} }; /************************************************************************ * Table of branding strings ************************************************************************/ static char *ixgbe_strings[] = { "Intel(R) PRO/10GbE PCI-Express Network Driver" }; /************************************************************************ * Function prototypes ************************************************************************/ static int ixgbe_probe(device_t); static int ixgbe_attach(device_t); static int ixgbe_detach(device_t); static int ixgbe_shutdown(device_t); static int ixgbe_suspend(device_t); static int ixgbe_resume(device_t); static int ixgbe_ioctl(struct ifnet *, u_long, caddr_t); static void ixgbe_init(void *); static void ixgbe_init_locked(struct adapter *); static void ixgbe_stop(void *); #if __FreeBSD_version >= 1100036 static uint64_t ixgbe_get_counter(struct ifnet *, ift_counter); #endif static void ixgbe_init_device_features(struct adapter *); static void ixgbe_check_fan_failure(struct adapter *, u32, bool); static void ixgbe_add_media_types(struct adapter *); static void ixgbe_media_status(struct ifnet *, struct ifmediareq *); static int ixgbe_media_change(struct ifnet *); static int ixgbe_allocate_pci_resources(struct adapter *); static void ixgbe_get_slot_info(struct adapter *); static int ixgbe_allocate_msix(struct adapter *); static int ixgbe_allocate_legacy(struct adapter *); static int ixgbe_configure_interrupts(struct adapter *); static void ixgbe_free_pci_resources(struct adapter *); static void ixgbe_local_timer(void *); static int ixgbe_setup_interface(device_t, struct adapter *); static void ixgbe_config_gpie(struct adapter *); static void ixgbe_config_dmac(struct adapter *); static void ixgbe_config_delay_values(struct adapter *); static void ixgbe_config_link(struct adapter *); static void ixgbe_check_wol_support(struct adapter *); static int ixgbe_setup_low_power_mode(struct adapter *); static void ixgbe_rearm_queues(struct adapter *, u64); static void ixgbe_initialize_transmit_units(struct adapter *); static void ixgbe_initialize_receive_units(struct adapter *); static void ixgbe_enable_rx_drop(struct adapter *); static void ixgbe_disable_rx_drop(struct adapter *); static void ixgbe_initialize_rss_mapping(struct adapter *); static void ixgbe_enable_intr(struct adapter *); static void ixgbe_disable_intr(struct adapter *); static void ixgbe_update_stats_counters(struct adapter *); static void ixgbe_set_promisc(struct adapter *); static void ixgbe_set_multi(struct adapter *); static void ixgbe_update_link_status(struct adapter *); static void ixgbe_set_ivar(struct adapter *, u8, u8, s8); static void ixgbe_configure_ivars(struct adapter *); static u8 *ixgbe_mc_array_itr(struct ixgbe_hw *, u8 **, u32 *); static void ixgbe_setup_vlan_hw_support(struct adapter *); static void ixgbe_register_vlan(void *, struct ifnet *, u16); static void ixgbe_unregister_vlan(void *, struct ifnet *, u16); static void ixgbe_add_device_sysctls(struct adapter *); static void ixgbe_add_hw_stats(struct adapter *); static int ixgbe_set_flowcntl(struct adapter *, int); static int ixgbe_set_advertise(struct adapter *, int); static int ixgbe_get_advertise(struct adapter *); /* Sysctl handlers */ static void ixgbe_set_sysctl_value(struct adapter *, const char *, const char *, int *, int); static int ixgbe_sysctl_flowcntl(SYSCTL_HANDLER_ARGS); static int ixgbe_sysctl_advertise(SYSCTL_HANDLER_ARGS); static int ixgbe_sysctl_interrupt_rate_handler(SYSCTL_HANDLER_ARGS); static int ixgbe_sysctl_dmac(SYSCTL_HANDLER_ARGS); static int ixgbe_sysctl_phy_temp(SYSCTL_HANDLER_ARGS); static int ixgbe_sysctl_phy_overtemp_occurred(SYSCTL_HANDLER_ARGS); #ifdef IXGBE_DEBUG static int ixgbe_sysctl_power_state(SYSCTL_HANDLER_ARGS); static int ixgbe_sysctl_print_rss_config(SYSCTL_HANDLER_ARGS); #endif static int ixgbe_sysctl_rdh_handler(SYSCTL_HANDLER_ARGS); static int ixgbe_sysctl_rdt_handler(SYSCTL_HANDLER_ARGS); static int ixgbe_sysctl_tdt_handler(SYSCTL_HANDLER_ARGS); static int ixgbe_sysctl_tdh_handler(SYSCTL_HANDLER_ARGS); static int ixgbe_sysctl_eee_state(SYSCTL_HANDLER_ARGS); static int ixgbe_sysctl_wol_enable(SYSCTL_HANDLER_ARGS); static int ixgbe_sysctl_wufc(SYSCTL_HANDLER_ARGS); /* Support for pluggable optic modules */ static bool ixgbe_sfp_probe(struct adapter *); /* Legacy (single vector) interrupt handler */ static void ixgbe_legacy_irq(void *); /* The MSI/MSI-X Interrupt handlers */ static void ixgbe_msix_que(void *); static void ixgbe_msix_link(void *); /* Deferred interrupt tasklets */ static void ixgbe_handle_que(void *, int); static void ixgbe_handle_link(void *, int); static void ixgbe_handle_msf(void *, int); static void ixgbe_handle_mod(void *, int); static void ixgbe_handle_phy(void *, int); /************************************************************************ * FreeBSD Device Interface Entry Points ************************************************************************/ static device_method_t ix_methods[] = { /* Device interface */ DEVMETHOD(device_probe, ixgbe_probe), DEVMETHOD(device_attach, ixgbe_attach), DEVMETHOD(device_detach, ixgbe_detach), DEVMETHOD(device_shutdown, ixgbe_shutdown), DEVMETHOD(device_suspend, ixgbe_suspend), DEVMETHOD(device_resume, ixgbe_resume), #ifdef PCI_IOV DEVMETHOD(pci_iov_init, ixgbe_init_iov), DEVMETHOD(pci_iov_uninit, ixgbe_uninit_iov), DEVMETHOD(pci_iov_add_vf, ixgbe_add_vf), #endif /* PCI_IOV */ DEVMETHOD_END }; static driver_t ix_driver = { "ix", ix_methods, sizeof(struct adapter), }; devclass_t ix_devclass; DRIVER_MODULE(ix, pci, ix_driver, ix_devclass, 0, 0); MODULE_DEPEND(ix, pci, 1, 1, 1); MODULE_DEPEND(ix, ether, 1, 1, 1); #ifdef DEV_NETMAP MODULE_DEPEND(ix, netmap, 1, 1, 1); #endif /* * TUNEABLE PARAMETERS: */ static SYSCTL_NODE(_hw, OID_AUTO, ix, CTLFLAG_RD, 0, "IXGBE driver parameters"); /* * AIM: Adaptive Interrupt Moderation * which means that the interrupt rate * is varied over time based on the * traffic for that interrupt vector */ static int ixgbe_enable_aim = TRUE; SYSCTL_INT(_hw_ix, OID_AUTO, enable_aim, CTLFLAG_RDTUN, &ixgbe_enable_aim, 0, "Enable adaptive interrupt moderation"); static int ixgbe_max_interrupt_rate = (4000000 / IXGBE_LOW_LATENCY); SYSCTL_INT(_hw_ix, OID_AUTO, max_interrupt_rate, CTLFLAG_RDTUN, &ixgbe_max_interrupt_rate, 0, "Maximum interrupts per second"); /* How many packets rxeof tries to clean at a time */ static int ixgbe_rx_process_limit = 256; SYSCTL_INT(_hw_ix, OID_AUTO, rx_process_limit, CTLFLAG_RDTUN, &ixgbe_rx_process_limit, 0, "Maximum number of received packets to process at a time, -1 means unlimited"); /* How many packets txeof tries to clean at a time */ static int ixgbe_tx_process_limit = 256; SYSCTL_INT(_hw_ix, OID_AUTO, tx_process_limit, CTLFLAG_RDTUN, &ixgbe_tx_process_limit, 0, "Maximum number of sent packets to process at a time, -1 means unlimited"); /* Flow control setting, default to full */ static int ixgbe_flow_control = ixgbe_fc_full; SYSCTL_INT(_hw_ix, OID_AUTO, flow_control, CTLFLAG_RDTUN, &ixgbe_flow_control, 0, "Default flow control used for all adapters"); /* Advertise Speed, default to 0 (auto) */ static int ixgbe_advertise_speed = 0; SYSCTL_INT(_hw_ix, OID_AUTO, advertise_speed, CTLFLAG_RDTUN, &ixgbe_advertise_speed, 0, "Default advertised speed for all adapters"); /* * Smart speed setting, default to on * this only works as a compile option * right now as its during attach, set * this to 'ixgbe_smart_speed_off' to * disable. */ static int ixgbe_smart_speed = ixgbe_smart_speed_on; /* * MSI-X should be the default for best performance, * but this allows it to be forced off for testing. */ static int ixgbe_enable_msix = 1; SYSCTL_INT(_hw_ix, OID_AUTO, enable_msix, CTLFLAG_RDTUN, &ixgbe_enable_msix, 0, "Enable MSI-X interrupts"); /* * Number of Queues, can be set to 0, * it then autoconfigures based on the * number of cpus with a max of 8. This * can be overriden manually here. */ static int ixgbe_num_queues = 0; SYSCTL_INT(_hw_ix, OID_AUTO, num_queues, CTLFLAG_RDTUN, &ixgbe_num_queues, 0, "Number of queues to configure, 0 indicates autoconfigure"); /* * Number of TX descriptors per ring, * setting higher than RX as this seems * the better performing choice. */ static int ixgbe_txd = PERFORM_TXD; SYSCTL_INT(_hw_ix, OID_AUTO, txd, CTLFLAG_RDTUN, &ixgbe_txd, 0, "Number of transmit descriptors per queue"); /* Number of RX descriptors per ring */ static int ixgbe_rxd = PERFORM_RXD; SYSCTL_INT(_hw_ix, OID_AUTO, rxd, CTLFLAG_RDTUN, &ixgbe_rxd, 0, "Number of receive descriptors per queue"); /* * Defining this on will allow the use * of unsupported SFP+ modules, note that * doing so you are on your own :) */ static int allow_unsupported_sfp = FALSE; SYSCTL_INT(_hw_ix, OID_AUTO, unsupported_sfp, CTLFLAG_RDTUN, &allow_unsupported_sfp, 0, "Allow unsupported SFP modules...use at your own risk"); /* * Not sure if Flow Director is fully baked, * so we'll default to turning it off. */ static int ixgbe_enable_fdir = 0; SYSCTL_INT(_hw_ix, OID_AUTO, enable_fdir, CTLFLAG_RDTUN, &ixgbe_enable_fdir, 0, "Enable Flow Director"); /* Legacy Transmit (single queue) */ static int ixgbe_enable_legacy_tx = 0; SYSCTL_INT(_hw_ix, OID_AUTO, enable_legacy_tx, CTLFLAG_RDTUN, &ixgbe_enable_legacy_tx, 0, "Enable Legacy TX flow"); /* Receive-Side Scaling */ static int ixgbe_enable_rss = 1; SYSCTL_INT(_hw_ix, OID_AUTO, enable_rss, CTLFLAG_RDTUN, &ixgbe_enable_rss, 0, "Enable Receive-Side Scaling (RSS)"); /* Keep running tab on them for sanity check */ static int ixgbe_total_ports; static int (*ixgbe_start_locked)(struct ifnet *, struct tx_ring *); static int (*ixgbe_ring_empty)(struct ifnet *, struct buf_ring *); MALLOC_DEFINE(M_IXGBE, "ix", "ix driver allocations"); /************************************************************************ * ixgbe_initialize_rss_mapping ************************************************************************/ static void ixgbe_initialize_rss_mapping(struct adapter *adapter) { struct ixgbe_hw *hw = &adapter->hw; u32 reta = 0, mrqc, rss_key[10]; int queue_id, table_size, index_mult; int i, j; u32 rss_hash_config; if (adapter->feat_en & IXGBE_FEATURE_RSS) { /* Fetch the configured RSS key */ rss_getkey((uint8_t *)&rss_key); } else { /* set up random bits */ arc4rand(&rss_key, sizeof(rss_key), 0); } /* Set multiplier for RETA setup and table size based on MAC */ index_mult = 0x1; table_size = 128; switch (adapter->hw.mac.type) { case ixgbe_mac_82598EB: index_mult = 0x11; break; case ixgbe_mac_X550: case ixgbe_mac_X550EM_x: case ixgbe_mac_X550EM_a: table_size = 512; break; default: break; } /* Set up the redirection table */ for (i = 0, j = 0; i < table_size; i++, j++) { if (j == adapter->num_queues) j = 0; if (adapter->feat_en & IXGBE_FEATURE_RSS) { /* * Fetch the RSS bucket id for the given indirection * entry. Cap it at the number of configured buckets * (which is num_queues.) */ queue_id = rss_get_indirection_to_bucket(i); queue_id = queue_id % adapter->num_queues; } else queue_id = (j * index_mult); /* * The low 8 bits are for hash value (n+0); * The next 8 bits are for hash value (n+1), etc. */ reta = reta >> 8; reta = reta | (((uint32_t)queue_id) << 24); if ((i & 3) == 3) { if (i < 128) IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta); else IXGBE_WRITE_REG(hw, IXGBE_ERETA((i >> 2) - 32), reta); reta = 0; } } /* Now fill our hash function seeds */ for (i = 0; i < 10; i++) IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), rss_key[i]); /* Perform hash on these packet types */ if (adapter->feat_en & IXGBE_FEATURE_RSS) rss_hash_config = rss_gethashconfig(); else { /* * Disable UDP - IP fragments aren't currently being handled * and so we end up with a mix of 2-tuple and 4-tuple * traffic. */ rss_hash_config = RSS_HASHTYPE_RSS_IPV4 | RSS_HASHTYPE_RSS_TCP_IPV4 | RSS_HASHTYPE_RSS_IPV6 | RSS_HASHTYPE_RSS_TCP_IPV6 | RSS_HASHTYPE_RSS_IPV6_EX | RSS_HASHTYPE_RSS_TCP_IPV6_EX; } mrqc = IXGBE_MRQC_RSSEN; if (rss_hash_config & RSS_HASHTYPE_RSS_IPV4) mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4; if (rss_hash_config & RSS_HASHTYPE_RSS_TCP_IPV4) mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_TCP; if (rss_hash_config & RSS_HASHTYPE_RSS_IPV6) mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6; if (rss_hash_config & RSS_HASHTYPE_RSS_TCP_IPV6) mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_TCP; if (rss_hash_config & RSS_HASHTYPE_RSS_IPV6_EX) mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_EX; if (rss_hash_config & RSS_HASHTYPE_RSS_TCP_IPV6_EX) mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP; if (rss_hash_config & RSS_HASHTYPE_RSS_UDP_IPV4) mrqc |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP; if (rss_hash_config & RSS_HASHTYPE_RSS_UDP_IPV6) mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP; if (rss_hash_config & RSS_HASHTYPE_RSS_UDP_IPV6_EX) mrqc |= IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP; mrqc |= ixgbe_get_mrqc(adapter->iov_mode); IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc); } /* ixgbe_initialize_rss_mapping */ /************************************************************************ * ixgbe_initialize_receive_units - Setup receive registers and features. ************************************************************************/ #define BSIZEPKT_ROUNDUP ((1<rx_rings; struct ixgbe_hw *hw = &adapter->hw; struct ifnet *ifp = adapter->ifp; int i, j; u32 bufsz, fctrl, srrctl, rxcsum; u32 hlreg; /* * Make sure receives are disabled while * setting up the descriptor ring */ ixgbe_disable_rx(hw); /* Enable broadcasts */ fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL); fctrl |= IXGBE_FCTRL_BAM; if (adapter->hw.mac.type == ixgbe_mac_82598EB) { fctrl |= IXGBE_FCTRL_DPF; fctrl |= IXGBE_FCTRL_PMCF; } IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl); /* Set for Jumbo Frames? */ hlreg = IXGBE_READ_REG(hw, IXGBE_HLREG0); if (ifp->if_mtu > ETHERMTU) hlreg |= IXGBE_HLREG0_JUMBOEN; else hlreg &= ~IXGBE_HLREG0_JUMBOEN; #ifdef DEV_NETMAP /* CRC stripping is conditional in Netmap */ if ((adapter->feat_en & IXGBE_FEATURE_NETMAP) && (ifp->if_capenable & IFCAP_NETMAP) && !ix_crcstrip) hlreg &= ~IXGBE_HLREG0_RXCRCSTRP; else #endif /* DEV_NETMAP */ hlreg |= IXGBE_HLREG0_RXCRCSTRP; IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg); bufsz = (adapter->rx_mbuf_sz + BSIZEPKT_ROUNDUP) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT; for (i = 0; i < adapter->num_queues; i++, rxr++) { u64 rdba = rxr->rxdma.dma_paddr; j = rxr->me; /* Setup the Base and Length of the Rx Descriptor Ring */ IXGBE_WRITE_REG(hw, IXGBE_RDBAL(j), (rdba & 0x00000000ffffffffULL)); IXGBE_WRITE_REG(hw, IXGBE_RDBAH(j), (rdba >> 32)); IXGBE_WRITE_REG(hw, IXGBE_RDLEN(j), adapter->num_rx_desc * sizeof(union ixgbe_adv_rx_desc)); /* Set up the SRRCTL register */ srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(j)); srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK; srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK; srrctl |= bufsz; srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF; /* * Set DROP_EN iff we have no flow control and >1 queue. * Note that srrctl was cleared shortly before during reset, * so we do not need to clear the bit, but do it just in case * this code is moved elsewhere. */ if (adapter->num_queues > 1 && adapter->hw.fc.requested_mode == ixgbe_fc_none) { srrctl |= IXGBE_SRRCTL_DROP_EN; } else { srrctl &= ~IXGBE_SRRCTL_DROP_EN; } IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(j), srrctl); /* Setup the HW Rx Head and Tail Descriptor Pointers */ IXGBE_WRITE_REG(hw, IXGBE_RDH(j), 0); IXGBE_WRITE_REG(hw, IXGBE_RDT(j), 0); /* Set the driver rx tail address */ rxr->tail = IXGBE_RDT(rxr->me); } if (adapter->hw.mac.type != ixgbe_mac_82598EB) { u32 psrtype = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR | IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR; IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(0), psrtype); } rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM); ixgbe_initialize_rss_mapping(adapter); if (adapter->num_queues > 1) { /* RSS and RX IPP Checksum are mutually exclusive */ rxcsum |= IXGBE_RXCSUM_PCSD; } if (ifp->if_capenable & IFCAP_RXCSUM) rxcsum |= IXGBE_RXCSUM_PCSD; /* This is useful for calculating UDP/IP fragment checksums */ if (!(rxcsum & IXGBE_RXCSUM_PCSD)) rxcsum |= IXGBE_RXCSUM_IPPCSE; IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum); return; } /* ixgbe_initialize_receive_units */ /************************************************************************ * ixgbe_initialize_transmit_units - Enable transmit units. ************************************************************************/ static void ixgbe_initialize_transmit_units(struct adapter *adapter) { struct tx_ring *txr = adapter->tx_rings; struct ixgbe_hw *hw = &adapter->hw; /* Setup the Base and Length of the Tx Descriptor Ring */ for (int i = 0; i < adapter->num_queues; i++, txr++) { u64 tdba = txr->txdma.dma_paddr; u32 txctrl = 0; int j = txr->me; IXGBE_WRITE_REG(hw, IXGBE_TDBAL(j), (tdba & 0x00000000ffffffffULL)); IXGBE_WRITE_REG(hw, IXGBE_TDBAH(j), (tdba >> 32)); IXGBE_WRITE_REG(hw, IXGBE_TDLEN(j), adapter->num_tx_desc * sizeof(union ixgbe_adv_tx_desc)); /* Setup the HW Tx Head and Tail descriptor pointers */ IXGBE_WRITE_REG(hw, IXGBE_TDH(j), 0); IXGBE_WRITE_REG(hw, IXGBE_TDT(j), 0); /* Cache the tail address */ txr->tail = IXGBE_TDT(j); /* Disable Head Writeback */ /* * Note: for X550 series devices, these registers are actually * prefixed with TPH_ isntead of DCA_, but the addresses and * fields remain the same. */ switch (hw->mac.type) { case ixgbe_mac_82598EB: txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(j)); break; default: txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL_82599(j)); break; } txctrl &= ~IXGBE_DCA_TXCTRL_DESC_WRO_EN; switch (hw->mac.type) { case ixgbe_mac_82598EB: IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(j), txctrl); break; default: IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL_82599(j), txctrl); break; } } if (hw->mac.type != ixgbe_mac_82598EB) { u32 dmatxctl, rttdcs; dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL); dmatxctl |= IXGBE_DMATXCTL_TE; IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl); /* Disable arbiter to set MTQC */ rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS); rttdcs |= IXGBE_RTTDCS_ARBDIS; IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs); IXGBE_WRITE_REG(hw, IXGBE_MTQC, ixgbe_get_mtqc(adapter->iov_mode)); rttdcs &= ~IXGBE_RTTDCS_ARBDIS; IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs); } return; } /* ixgbe_initialize_transmit_units */ /************************************************************************ * ixgbe_attach - Device initialization routine * * Called when the driver is being loaded. * Identifies the type of hardware, allocates all resources * and initializes the hardware. * * return 0 on success, positive on failure ************************************************************************/ static int ixgbe_attach(device_t dev) { struct adapter *adapter; struct ixgbe_hw *hw; int error = 0; u32 ctrl_ext; INIT_DEBUGOUT("ixgbe_attach: begin"); /* Allocate, clear, and link in our adapter structure */ adapter = device_get_softc(dev); adapter->hw.back = adapter; adapter->dev = dev; hw = &adapter->hw; /* Core Lock Init*/ IXGBE_CORE_LOCK_INIT(adapter, device_get_nameunit(dev)); /* Set up the timer callout */ callout_init_mtx(&adapter->timer, &adapter->core_mtx, 0); /* Determine hardware revision */ hw->vendor_id = pci_get_vendor(dev); hw->device_id = pci_get_device(dev); hw->revision_id = pci_get_revid(dev); hw->subsystem_vendor_id = pci_get_subvendor(dev); hw->subsystem_device_id = pci_get_subdevice(dev); /* * Make sure BUSMASTER is set */ pci_enable_busmaster(dev); /* Do base PCI setup - map BAR0 */ if (ixgbe_allocate_pci_resources(adapter)) { device_printf(dev, "Allocation of PCI resources failed\n"); error = ENXIO; goto err_out; } /* let hardware know driver is loaded */ ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT); ctrl_ext |= IXGBE_CTRL_EXT_DRV_LOAD; IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext); /* * Initialize the shared code */ if (ixgbe_init_shared_code(hw)) { device_printf(dev, "Unable to initialize the shared code\n"); error = ENXIO; goto err_out; } if (hw->mbx.ops.init_params) hw->mbx.ops.init_params(hw); hw->allow_unsupported_sfp = allow_unsupported_sfp; /* Pick up the 82599 settings */ if (hw->mac.type != ixgbe_mac_82598EB) { hw->phy.smart_speed = ixgbe_smart_speed; adapter->num_segs = IXGBE_82599_SCATTER; } else adapter->num_segs = IXGBE_82598_SCATTER; ixgbe_init_device_features(adapter); if (ixgbe_configure_interrupts(adapter)) { error = ENXIO; goto err_out; } /* Allocate multicast array memory. */ adapter->mta = malloc(sizeof(*adapter->mta) * MAX_NUM_MULTICAST_ADDRESSES, M_IXGBE, M_NOWAIT); if (adapter->mta == NULL) { device_printf(dev, "Can not allocate multicast setup array\n"); error = ENOMEM; goto err_out; } /* Enable WoL (if supported) */ ixgbe_check_wol_support(adapter); /* Register for VLAN events */ adapter->vlan_attach = EVENTHANDLER_REGISTER(vlan_config, ixgbe_register_vlan, adapter, EVENTHANDLER_PRI_FIRST); adapter->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig, ixgbe_unregister_vlan, adapter, EVENTHANDLER_PRI_FIRST); /* Verify adapter fan is still functional (if applicable) */ if (adapter->feat_en & IXGBE_FEATURE_FAN_FAIL) { u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP); ixgbe_check_fan_failure(adapter, esdp, FALSE); } /* Ensure SW/FW semaphore is free */ ixgbe_init_swfw_semaphore(hw); /* Enable EEE power saving */ if (adapter->feat_en & IXGBE_FEATURE_EEE) hw->mac.ops.setup_eee(hw, TRUE); /* Set an initial default flow control value */ hw->fc.requested_mode = ixgbe_flow_control; /* Sysctls for limiting the amount of work done in the taskqueues */ ixgbe_set_sysctl_value(adapter, "rx_processing_limit", "max number of rx packets to process", &adapter->rx_process_limit, ixgbe_rx_process_limit); ixgbe_set_sysctl_value(adapter, "tx_processing_limit", "max number of tx packets to process", &adapter->tx_process_limit, ixgbe_tx_process_limit); /* Do descriptor calc and sanity checks */ if (((ixgbe_txd * sizeof(union ixgbe_adv_tx_desc)) % DBA_ALIGN) != 0 || ixgbe_txd < MIN_TXD || ixgbe_txd > MAX_TXD) { device_printf(dev, "TXD config issue, using default!\n"); adapter->num_tx_desc = DEFAULT_TXD; } else adapter->num_tx_desc = ixgbe_txd; /* * With many RX rings it is easy to exceed the * system mbuf allocation. Tuning nmbclusters * can alleviate this. */ if (nmbclusters > 0) { int s; s = (ixgbe_rxd * adapter->num_queues) * ixgbe_total_ports; if (s > nmbclusters) { device_printf(dev, "RX Descriptors exceed system mbuf max, using default instead!\n"); ixgbe_rxd = DEFAULT_RXD; } } if (((ixgbe_rxd * sizeof(union ixgbe_adv_rx_desc)) % DBA_ALIGN) != 0 || ixgbe_rxd < MIN_RXD || ixgbe_rxd > MAX_RXD) { device_printf(dev, "RXD config issue, using default!\n"); adapter->num_rx_desc = DEFAULT_RXD; } else adapter->num_rx_desc = ixgbe_rxd; /* Allocate our TX/RX Queues */ if (ixgbe_allocate_queues(adapter)) { error = ENOMEM; goto err_out; } hw->phy.reset_if_overtemp = TRUE; error = ixgbe_reset_hw(hw); hw->phy.reset_if_overtemp = FALSE; if (error == IXGBE_ERR_SFP_NOT_PRESENT) { /* * No optics in this port, set up * so the timer routine will probe * for later insertion. */ adapter->sfp_probe = TRUE; error = IXGBE_SUCCESS; } else if (error == IXGBE_ERR_SFP_NOT_SUPPORTED) { device_printf(dev, "Unsupported SFP+ module detected!\n"); error = EIO; goto err_late; } else if (error) { device_printf(dev, "Hardware initialization failed\n"); error = EIO; goto err_late; } /* Make sure we have a good EEPROM before we read from it */ if (ixgbe_validate_eeprom_checksum(&adapter->hw, NULL) < 0) { device_printf(dev, "The EEPROM Checksum Is Not Valid\n"); error = EIO; goto err_late; } /* Setup OS specific network interface */ if (ixgbe_setup_interface(dev, adapter) != 0) goto err_late; if (adapter->feat_en & IXGBE_FEATURE_MSIX) error = ixgbe_allocate_msix(adapter); else error = ixgbe_allocate_legacy(adapter); if (error) goto err_late; error = ixgbe_start_hw(hw); switch (error) { case IXGBE_ERR_EEPROM_VERSION: device_printf(dev, "This device is a pre-production adapter/LOM. Please be aware there may be issues associated with your hardware.\nIf you are experiencing problems please contact your Intel or hardware representative who provided you with this hardware.\n"); break; case IXGBE_ERR_SFP_NOT_SUPPORTED: device_printf(dev, "Unsupported SFP+ Module\n"); error = EIO; goto err_late; case IXGBE_ERR_SFP_NOT_PRESENT: device_printf(dev, "No SFP+ Module found\n"); /* falls thru */ default: break; } /* Enable the optics for 82599 SFP+ fiber */ ixgbe_enable_tx_laser(hw); /* Enable power to the phy. */ ixgbe_set_phy_power(hw, TRUE); /* Initialize statistics */ ixgbe_update_stats_counters(adapter); /* Check PCIE slot type/speed/width */ ixgbe_get_slot_info(adapter); /* * Do time init and sysctl init here, but * only on the first port of a bypass adapter. */ ixgbe_bypass_init(adapter); /* Set an initial dmac value */ adapter->dmac = 0; /* Set initial advertised speeds (if applicable) */ adapter->advertise = ixgbe_get_advertise(adapter); if (adapter->feat_cap & IXGBE_FEATURE_SRIOV) ixgbe_define_iov_schemas(dev, &error); /* Add sysctls */ ixgbe_add_device_sysctls(adapter); ixgbe_add_hw_stats(adapter); /* For Netmap */ adapter->init_locked = ixgbe_init_locked; adapter->stop_locked = ixgbe_stop; if (adapter->feat_en & IXGBE_FEATURE_NETMAP) ixgbe_netmap_attach(adapter); INIT_DEBUGOUT("ixgbe_attach: end"); return (0); err_late: ixgbe_free_transmit_structures(adapter); ixgbe_free_receive_structures(adapter); free(adapter->queues, M_DEVBUF); err_out: if (adapter->ifp != NULL) if_free(adapter->ifp); ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT); ctrl_ext &= ~IXGBE_CTRL_EXT_DRV_LOAD; IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT, ctrl_ext); ixgbe_free_pci_resources(adapter); free(adapter->mta, M_IXGBE); IXGBE_CORE_LOCK_DESTROY(adapter); return (error); } /* ixgbe_attach */ /************************************************************************ * ixgbe_check_wol_support * * Checks whether the adapter's ports are capable of * Wake On LAN by reading the adapter's NVM. * * Sets each port's hw->wol_enabled value depending * on the value read here. ************************************************************************/ static void ixgbe_check_wol_support(struct adapter *adapter) { struct ixgbe_hw *hw = &adapter->hw; u16 dev_caps = 0; /* Find out WoL support for port */ adapter->wol_support = hw->wol_enabled = 0; ixgbe_get_device_caps(hw, &dev_caps); if ((dev_caps & IXGBE_DEVICE_CAPS_WOL_PORT0_1) || ((dev_caps & IXGBE_DEVICE_CAPS_WOL_PORT0) && hw->bus.func == 0)) adapter->wol_support = hw->wol_enabled = 1; /* Save initial wake up filter configuration */ adapter->wufc = IXGBE_READ_REG(hw, IXGBE_WUFC); return; } /* ixgbe_check_wol_support */ /************************************************************************ * ixgbe_setup_interface * * Setup networking device structure and register an interface. ************************************************************************/ static int ixgbe_setup_interface(device_t dev, struct adapter *adapter) { struct ifnet *ifp; INIT_DEBUGOUT("ixgbe_setup_interface: begin"); ifp = adapter->ifp = if_alloc(IFT_ETHER); if (ifp == NULL) { device_printf(dev, "can not allocate ifnet structure\n"); return (-1); } if_initname(ifp, device_get_name(dev), device_get_unit(dev)); ifp->if_baudrate = IF_Gbps(10); ifp->if_init = ixgbe_init; ifp->if_softc = adapter; ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST; ifp->if_ioctl = ixgbe_ioctl; #if __FreeBSD_version >= 1100036 if_setgetcounterfn(ifp, ixgbe_get_counter); #endif #if __FreeBSD_version >= 1100045 /* TSO parameters */ ifp->if_hw_tsomax = 65518; ifp->if_hw_tsomaxsegcount = IXGBE_82599_SCATTER; ifp->if_hw_tsomaxsegsize = 2048; #endif if (adapter->feat_en & IXGBE_FEATURE_LEGACY_TX) { ifp->if_start = ixgbe_legacy_start; IFQ_SET_MAXLEN(&ifp->if_snd, adapter->num_tx_desc - 2); ifp->if_snd.ifq_drv_maxlen = adapter->num_tx_desc - 2; IFQ_SET_READY(&ifp->if_snd); ixgbe_start_locked = ixgbe_legacy_start_locked; ixgbe_ring_empty = ixgbe_legacy_ring_empty; } else { ifp->if_transmit = ixgbe_mq_start; ifp->if_qflush = ixgbe_qflush; ixgbe_start_locked = ixgbe_mq_start_locked; ixgbe_ring_empty = drbr_empty; } ether_ifattach(ifp, adapter->hw.mac.addr); adapter->max_frame_size = ifp->if_mtu + ETHER_HDR_LEN + ETHER_CRC_LEN; /* * Tell the upper layer(s) we support long frames. */ ifp->if_hdrlen = sizeof(struct ether_vlan_header); /* Set capability flags */ ifp->if_capabilities |= IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6 | IFCAP_TSO | IFCAP_LRO | IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWTSO | IFCAP_VLAN_HWCSUM | IFCAP_JUMBO_MTU | IFCAP_VLAN_MTU | IFCAP_HWSTATS; /* Enable the above capabilities by default */ ifp->if_capenable = ifp->if_capabilities; /* * Don't turn this on by default, if vlans are * created on another pseudo device (eg. lagg) * then vlan events are not passed thru, breaking * operation, but with HW FILTER off it works. If * using vlans directly on the ixgbe driver you can * enable this and get full hardware tag filtering. */ ifp->if_capabilities |= IFCAP_VLAN_HWFILTER; /* * Specify the media types supported by this adapter and register * callbacks to update media and link information */ ifmedia_init(&adapter->media, IFM_IMASK, ixgbe_media_change, ixgbe_media_status); adapter->phy_layer = ixgbe_get_supported_physical_layer(&adapter->hw); ixgbe_add_media_types(adapter); /* Set autoselect media by default */ ifmedia_set(&adapter->media, IFM_ETHER | IFM_AUTO); return (0); } /* ixgbe_setup_interface */ #if __FreeBSD_version >= 1100036 /************************************************************************ * ixgbe_get_counter ************************************************************************/ static uint64_t ixgbe_get_counter(struct ifnet *ifp, ift_counter cnt) { struct adapter *adapter; struct tx_ring *txr; uint64_t rv; adapter = if_getsoftc(ifp); switch (cnt) { case IFCOUNTER_IPACKETS: return (adapter->ipackets); case IFCOUNTER_OPACKETS: return (adapter->opackets); case IFCOUNTER_IBYTES: return (adapter->ibytes); case IFCOUNTER_OBYTES: return (adapter->obytes); case IFCOUNTER_IMCASTS: return (adapter->imcasts); case IFCOUNTER_OMCASTS: return (adapter->omcasts); case IFCOUNTER_COLLISIONS: return (0); case IFCOUNTER_IQDROPS: return (adapter->iqdrops); case IFCOUNTER_OQDROPS: rv = 0; txr = adapter->tx_rings; for (int i = 0; i < adapter->num_queues; i++, txr++) rv += txr->br->br_drops; return (rv); case IFCOUNTER_IERRORS: return (adapter->ierrors); default: return (if_get_counter_default(ifp, cnt)); } } /* ixgbe_get_counter */ #endif /************************************************************************ * ixgbe_add_media_types ************************************************************************/ static void ixgbe_add_media_types(struct adapter *adapter) { struct ixgbe_hw *hw = &adapter->hw; device_t dev = adapter->dev; u64 layer; layer = adapter->phy_layer; /* Media types with matching FreeBSD media defines */ if (layer & IXGBE_PHYSICAL_LAYER_10GBASE_T) ifmedia_add(&adapter->media, IFM_ETHER | IFM_10G_T, 0, NULL); if (layer & IXGBE_PHYSICAL_LAYER_1000BASE_T) ifmedia_add(&adapter->media, IFM_ETHER | IFM_1000_T, 0, NULL); if (layer & IXGBE_PHYSICAL_LAYER_100BASE_TX) ifmedia_add(&adapter->media, IFM_ETHER | IFM_100_TX, 0, NULL); if (layer & IXGBE_PHYSICAL_LAYER_10BASE_T) ifmedia_add(&adapter->media, IFM_ETHER | IFM_10_T, 0, NULL); if (layer & IXGBE_PHYSICAL_LAYER_SFP_PLUS_CU || layer & IXGBE_PHYSICAL_LAYER_SFP_ACTIVE_DA) ifmedia_add(&adapter->media, IFM_ETHER | IFM_10G_TWINAX, 0, NULL); if (layer & IXGBE_PHYSICAL_LAYER_10GBASE_LR) { ifmedia_add(&adapter->media, IFM_ETHER | IFM_10G_LR, 0, NULL); if (hw->phy.multispeed_fiber) ifmedia_add(&adapter->media, IFM_ETHER | IFM_1000_LX, 0, NULL); } if (layer & IXGBE_PHYSICAL_LAYER_10GBASE_SR) { ifmedia_add(&adapter->media, IFM_ETHER | IFM_10G_SR, 0, NULL); if (hw->phy.multispeed_fiber) ifmedia_add(&adapter->media, IFM_ETHER | IFM_1000_SX, 0, NULL); } else if (layer & IXGBE_PHYSICAL_LAYER_1000BASE_SX) ifmedia_add(&adapter->media, IFM_ETHER | IFM_1000_SX, 0, NULL); if (layer & IXGBE_PHYSICAL_LAYER_10GBASE_CX4) ifmedia_add(&adapter->media, IFM_ETHER | IFM_10G_CX4, 0, NULL); #ifdef IFM_ETH_XTYPE if (layer & IXGBE_PHYSICAL_LAYER_10GBASE_KR) ifmedia_add(&adapter->media, IFM_ETHER | IFM_10G_KR, 0, NULL); if (layer & IXGBE_PHYSICAL_LAYER_10GBASE_KX4) ifmedia_add(&adapter->media, IFM_ETHER | IFM_10G_KX4, 0, NULL); if (layer & IXGBE_PHYSICAL_LAYER_1000BASE_KX) ifmedia_add(&adapter->media, IFM_ETHER | IFM_1000_KX, 0, NULL); if (layer & IXGBE_PHYSICAL_LAYER_2500BASE_KX) ifmedia_add(&adapter->media, IFM_ETHER | IFM_2500_KX, 0, NULL); #else if (layer & IXGBE_PHYSICAL_LAYER_10GBASE_KR) { device_printf(dev, "Media supported: 10GbaseKR\n"); device_printf(dev, "10GbaseKR mapped to 10GbaseSR\n"); ifmedia_add(&adapter->media, IFM_ETHER | IFM_10G_SR, 0, NULL); } if (layer & IXGBE_PHYSICAL_LAYER_10GBASE_KX4) { device_printf(dev, "Media supported: 10GbaseKX4\n"); device_printf(dev, "10GbaseKX4 mapped to 10GbaseCX4\n"); ifmedia_add(&adapter->media, IFM_ETHER | IFM_10G_CX4, 0, NULL); } if (layer & IXGBE_PHYSICAL_LAYER_1000BASE_KX) { device_printf(dev, "Media supported: 1000baseKX\n"); device_printf(dev, "1000baseKX mapped to 1000baseCX\n"); ifmedia_add(&adapter->media, IFM_ETHER | IFM_1000_CX, 0, NULL); } if (layer & IXGBE_PHYSICAL_LAYER_2500BASE_KX) { device_printf(dev, "Media supported: 2500baseKX\n"); device_printf(dev, "2500baseKX mapped to 2500baseSX\n"); ifmedia_add(&adapter->media, IFM_ETHER | IFM_2500_SX, 0, NULL); } #endif if (layer & IXGBE_PHYSICAL_LAYER_1000BASE_BX) device_printf(dev, "Media supported: 1000baseBX\n"); if (hw->device_id == IXGBE_DEV_ID_82598AT) { ifmedia_add(&adapter->media, IFM_ETHER | IFM_1000_T | IFM_FDX, 0, NULL); ifmedia_add(&adapter->media, IFM_ETHER | IFM_1000_T, 0, NULL); } ifmedia_add(&adapter->media, IFM_ETHER | IFM_AUTO, 0, NULL); } /* ixgbe_add_media_types */ /************************************************************************ * ixgbe_is_sfp ************************************************************************/ static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw) { switch (hw->mac.type) { case ixgbe_mac_82598EB: if (hw->phy.type == ixgbe_phy_nl) return TRUE; return FALSE; case ixgbe_mac_82599EB: switch (hw->mac.ops.get_media_type(hw)) { case ixgbe_media_type_fiber: case ixgbe_media_type_fiber_qsfp: return TRUE; default: return FALSE; } case ixgbe_mac_X550EM_x: case ixgbe_mac_X550EM_a: if (hw->mac.ops.get_media_type(hw) == ixgbe_media_type_fiber) return TRUE; return FALSE; default: return FALSE; } } /* ixgbe_is_sfp */ /************************************************************************ * ixgbe_config_link ************************************************************************/ static void ixgbe_config_link(struct adapter *adapter) { struct ixgbe_hw *hw = &adapter->hw; u32 autoneg, err = 0; bool sfp, negotiate; sfp = ixgbe_is_sfp(hw); if (sfp) { if (hw->phy.multispeed_fiber) { hw->mac.ops.setup_sfp(hw); ixgbe_enable_tx_laser(hw); taskqueue_enqueue(adapter->tq, &adapter->msf_task); } else taskqueue_enqueue(adapter->tq, &adapter->mod_task); } else { if (hw->mac.ops.check_link) err = ixgbe_check_link(hw, &adapter->link_speed, &adapter->link_up, FALSE); if (err) goto out; autoneg = hw->phy.autoneg_advertised; if ((!autoneg) && (hw->mac.ops.get_link_capabilities)) err = hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiate); if (err) goto out; if (hw->mac.ops.setup_link) err = hw->mac.ops.setup_link(hw, autoneg, adapter->link_up); } out: return; } /* ixgbe_config_link */ /************************************************************************ * ixgbe_update_stats_counters - Update board statistics counters. ************************************************************************/ static void ixgbe_update_stats_counters(struct adapter *adapter) { struct ixgbe_hw *hw = &adapter->hw; struct ixgbe_hw_stats *stats = &adapter->stats.pf; u32 missed_rx = 0, bprc, lxon, lxoff, total; u64 total_missed_rx = 0; stats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS); stats->illerrc += IXGBE_READ_REG(hw, IXGBE_ILLERRC); stats->errbc += IXGBE_READ_REG(hw, IXGBE_ERRBC); stats->mspdc += IXGBE_READ_REG(hw, IXGBE_MSPDC); stats->mpc[0] += IXGBE_READ_REG(hw, IXGBE_MPC(0)); for (int i = 0; i < 16; i++) { stats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i)); stats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i)); stats->qprdc[i] += IXGBE_READ_REG(hw, IXGBE_QPRDC(i)); } stats->mlfc += IXGBE_READ_REG(hw, IXGBE_MLFC); stats->mrfc += IXGBE_READ_REG(hw, IXGBE_MRFC); stats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC); /* Hardware workaround, gprc counts missed packets */ stats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC); stats->gprc -= missed_rx; if (hw->mac.type != ixgbe_mac_82598EB) { stats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL) + ((u64)IXGBE_READ_REG(hw, IXGBE_GORCH) << 32); stats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL) + ((u64)IXGBE_READ_REG(hw, IXGBE_GOTCH) << 32); stats->tor += IXGBE_READ_REG(hw, IXGBE_TORL) + ((u64)IXGBE_READ_REG(hw, IXGBE_TORH) << 32); stats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT); stats->lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT); } else { stats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC); stats->lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXC); /* 82598 only has a counter in the high register */ stats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH); stats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH); stats->tor += IXGBE_READ_REG(hw, IXGBE_TORH); } /* * Workaround: mprc hardware is incorrectly counting * broadcasts, so for now we subtract those. */ bprc = IXGBE_READ_REG(hw, IXGBE_BPRC); stats->bprc += bprc; stats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC); if (hw->mac.type == ixgbe_mac_82598EB) stats->mprc -= bprc; stats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64); stats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127); stats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255); stats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511); stats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023); stats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522); lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC); stats->lxontxc += lxon; lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC); stats->lxofftxc += lxoff; total = lxon + lxoff; stats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC); stats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC); stats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64); stats->gptc -= total; stats->mptc -= total; stats->ptc64 -= total; stats->gotc -= total * ETHER_MIN_LEN; stats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC); stats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC); stats->roc += IXGBE_READ_REG(hw, IXGBE_ROC); stats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC); stats->mngprc += IXGBE_READ_REG(hw, IXGBE_MNGPRC); stats->mngpdc += IXGBE_READ_REG(hw, IXGBE_MNGPDC); stats->mngptc += IXGBE_READ_REG(hw, IXGBE_MNGPTC); stats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR); stats->tpt += IXGBE_READ_REG(hw, IXGBE_TPT); stats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127); stats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255); stats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511); stats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023); stats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522); stats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC); stats->xec += IXGBE_READ_REG(hw, IXGBE_XEC); stats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC); stats->fclast += IXGBE_READ_REG(hw, IXGBE_FCLAST); /* Only read FCOE on 82599 */ if (hw->mac.type != ixgbe_mac_82598EB) { stats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC); stats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC); stats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC); stats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC); stats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC); } /* Fill out the OS statistics structure */ IXGBE_SET_IPACKETS(adapter, stats->gprc); IXGBE_SET_OPACKETS(adapter, stats->gptc); IXGBE_SET_IBYTES(adapter, stats->gorc); IXGBE_SET_OBYTES(adapter, stats->gotc); IXGBE_SET_IMCASTS(adapter, stats->mprc); IXGBE_SET_OMCASTS(adapter, stats->mptc); IXGBE_SET_COLLISIONS(adapter, 0); IXGBE_SET_IQDROPS(adapter, total_missed_rx); IXGBE_SET_IERRORS(adapter, stats->crcerrs + stats->rlec); } /* ixgbe_update_stats_counters */ /************************************************************************ * ixgbe_add_hw_stats * * Add sysctl variables, one per statistic, to the system. ************************************************************************/ static void ixgbe_add_hw_stats(struct adapter *adapter) { device_t dev = adapter->dev; struct tx_ring *txr = adapter->tx_rings; struct rx_ring *rxr = adapter->rx_rings; struct sysctl_ctx_list *ctx = device_get_sysctl_ctx(dev); struct sysctl_oid *tree = device_get_sysctl_tree(dev); struct sysctl_oid_list *child = SYSCTL_CHILDREN(tree); struct ixgbe_hw_stats *stats = &adapter->stats.pf; struct sysctl_oid *stat_node, *queue_node; struct sysctl_oid_list *stat_list, *queue_list; #define QUEUE_NAME_LEN 32 char namebuf[QUEUE_NAME_LEN]; /* Driver Statistics */ SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "dropped", CTLFLAG_RD, &adapter->dropped_pkts, "Driver dropped packets"); SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "mbuf_defrag_failed", CTLFLAG_RD, &adapter->mbuf_defrag_failed, "m_defrag() failed"); SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "watchdog_events", CTLFLAG_RD, &adapter->watchdog_events, "Watchdog timeouts"); SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "link_irq", CTLFLAG_RD, &adapter->link_irq, "Link MSI-X IRQ Handled"); for (int i = 0; i < adapter->num_queues; i++, txr++) { snprintf(namebuf, QUEUE_NAME_LEN, "queue%d", i); queue_node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf, CTLFLAG_RD, NULL, "Queue Name"); queue_list = SYSCTL_CHILDREN(queue_node); SYSCTL_ADD_PROC(ctx, queue_list, OID_AUTO, "interrupt_rate", CTLTYPE_UINT | CTLFLAG_RW, &adapter->queues[i], sizeof(&adapter->queues[i]), ixgbe_sysctl_interrupt_rate_handler, "IU", "Interrupt Rate"); SYSCTL_ADD_UQUAD(ctx, queue_list, OID_AUTO, "irqs", CTLFLAG_RD, &(adapter->queues[i].irqs), "irqs on this queue"); SYSCTL_ADD_PROC(ctx, queue_list, OID_AUTO, "txd_head", CTLTYPE_UINT | CTLFLAG_RD, txr, sizeof(txr), ixgbe_sysctl_tdh_handler, "IU", "Transmit Descriptor Head"); SYSCTL_ADD_PROC(ctx, queue_list, OID_AUTO, "txd_tail", CTLTYPE_UINT | CTLFLAG_RD, txr, sizeof(txr), ixgbe_sysctl_tdt_handler, "IU", "Transmit Descriptor Tail"); SYSCTL_ADD_UQUAD(ctx, queue_list, OID_AUTO, "tso_tx", CTLFLAG_RD, &txr->tso_tx, "TSO"); SYSCTL_ADD_UQUAD(ctx, queue_list, OID_AUTO, "no_tx_dma_setup", CTLFLAG_RD, &txr->no_tx_dma_setup, "Driver tx dma failure in xmit"); SYSCTL_ADD_UQUAD(ctx, queue_list, OID_AUTO, "no_desc_avail", CTLFLAG_RD, &txr->no_desc_avail, "Queue No Descriptor Available"); SYSCTL_ADD_UQUAD(ctx, queue_list, OID_AUTO, "tx_packets", CTLFLAG_RD, &txr->total_packets, "Queue Packets Transmitted"); SYSCTL_ADD_UQUAD(ctx, queue_list, OID_AUTO, "br_drops", CTLFLAG_RD, &txr->br->br_drops, "Packets dropped in buf_ring"); } for (int i = 0; i < adapter->num_queues; i++, rxr++) { struct lro_ctrl *lro = &rxr->lro; snprintf(namebuf, QUEUE_NAME_LEN, "queue%d", i); queue_node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf, CTLFLAG_RD, NULL, "Queue Name"); queue_list = SYSCTL_CHILDREN(queue_node); SYSCTL_ADD_PROC(ctx, queue_list, OID_AUTO, "rxd_head", CTLTYPE_UINT | CTLFLAG_RD, rxr, sizeof(rxr), ixgbe_sysctl_rdh_handler, "IU", "Receive Descriptor Head"); SYSCTL_ADD_PROC(ctx, queue_list, OID_AUTO, "rxd_tail", CTLTYPE_UINT | CTLFLAG_RD, rxr, sizeof(rxr), ixgbe_sysctl_rdt_handler, "IU", "Receive Descriptor Tail"); SYSCTL_ADD_UQUAD(ctx, queue_list, OID_AUTO, "rx_packets", CTLFLAG_RD, &rxr->rx_packets, "Queue Packets Received"); SYSCTL_ADD_UQUAD(ctx, queue_list, OID_AUTO, "rx_bytes", CTLFLAG_RD, &rxr->rx_bytes, "Queue Bytes Received"); SYSCTL_ADD_UQUAD(ctx, queue_list, OID_AUTO, "rx_copies", CTLFLAG_RD, &rxr->rx_copies, "Copied RX Frames"); SYSCTL_ADD_UQUAD(ctx, queue_list, OID_AUTO, "rx_discarded", CTLFLAG_RD, &rxr->rx_discarded, "Discarded RX packets"); SYSCTL_ADD_U64(ctx, queue_list, OID_AUTO, "lro_queued", CTLFLAG_RD, &lro->lro_queued, 0, "LRO Queued"); SYSCTL_ADD_U64(ctx, queue_list, OID_AUTO, "lro_flushed", CTLFLAG_RD, &lro->lro_flushed, 0, "LRO Flushed"); } /* MAC stats get their own sub node */ stat_node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "mac_stats", CTLFLAG_RD, NULL, "MAC Statistics"); stat_list = SYSCTL_CHILDREN(stat_node); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "crc_errs", CTLFLAG_RD, &stats->crcerrs, "CRC Errors"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "ill_errs", CTLFLAG_RD, &stats->illerrc, "Illegal Byte Errors"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "byte_errs", CTLFLAG_RD, &stats->errbc, "Byte Errors"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "short_discards", CTLFLAG_RD, &stats->mspdc, "MAC Short Packets Discarded"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "local_faults", CTLFLAG_RD, &stats->mlfc, "MAC Local Faults"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "remote_faults", CTLFLAG_RD, &stats->mrfc, "MAC Remote Faults"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "rec_len_errs", CTLFLAG_RD, &stats->rlec, "Receive Length Errors"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "rx_missed_packets", CTLFLAG_RD, &stats->mpc[0], "RX Missed Packet Count"); /* Flow Control stats */ SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "xon_txd", CTLFLAG_RD, &stats->lxontxc, "Link XON Transmitted"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "xon_recvd", CTLFLAG_RD, &stats->lxonrxc, "Link XON Received"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "xoff_txd", CTLFLAG_RD, &stats->lxofftxc, "Link XOFF Transmitted"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "xoff_recvd", CTLFLAG_RD, &stats->lxoffrxc, "Link XOFF Received"); /* Packet Reception Stats */ SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "total_octets_rcvd", CTLFLAG_RD, &stats->tor, "Total Octets Received"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "good_octets_rcvd", CTLFLAG_RD, &stats->gorc, "Good Octets Received"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "total_pkts_rcvd", CTLFLAG_RD, &stats->tpr, "Total Packets Received"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "good_pkts_rcvd", CTLFLAG_RD, &stats->gprc, "Good Packets Received"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "mcast_pkts_rcvd", CTLFLAG_RD, &stats->mprc, "Multicast Packets Received"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "bcast_pkts_rcvd", CTLFLAG_RD, &stats->bprc, "Broadcast Packets Received"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "rx_frames_64", CTLFLAG_RD, &stats->prc64, "64 byte frames received "); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "rx_frames_65_127", CTLFLAG_RD, &stats->prc127, "65-127 byte frames received"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "rx_frames_128_255", CTLFLAG_RD, &stats->prc255, "128-255 byte frames received"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "rx_frames_256_511", CTLFLAG_RD, &stats->prc511, "256-511 byte frames received"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "rx_frames_512_1023", CTLFLAG_RD, &stats->prc1023, "512-1023 byte frames received"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "rx_frames_1024_1522", CTLFLAG_RD, &stats->prc1522, "1023-1522 byte frames received"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "recv_undersized", CTLFLAG_RD, &stats->ruc, "Receive Undersized"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "recv_fragmented", CTLFLAG_RD, &stats->rfc, "Fragmented Packets Received "); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "recv_oversized", CTLFLAG_RD, &stats->roc, "Oversized Packets Received"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "recv_jabberd", CTLFLAG_RD, &stats->rjc, "Received Jabber"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "management_pkts_rcvd", CTLFLAG_RD, &stats->mngprc, "Management Packets Received"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "management_pkts_drpd", CTLFLAG_RD, &stats->mngptc, "Management Packets Dropped"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "checksum_errs", CTLFLAG_RD, &stats->xec, "Checksum Errors"); /* Packet Transmission Stats */ SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "good_octets_txd", CTLFLAG_RD, &stats->gotc, "Good Octets Transmitted"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "total_pkts_txd", CTLFLAG_RD, &stats->tpt, "Total Packets Transmitted"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "good_pkts_txd", CTLFLAG_RD, &stats->gptc, "Good Packets Transmitted"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "bcast_pkts_txd", CTLFLAG_RD, &stats->bptc, "Broadcast Packets Transmitted"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "mcast_pkts_txd", CTLFLAG_RD, &stats->mptc, "Multicast Packets Transmitted"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "management_pkts_txd", CTLFLAG_RD, &stats->mngptc, "Management Packets Transmitted"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "tx_frames_64", CTLFLAG_RD, &stats->ptc64, "64 byte frames transmitted "); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "tx_frames_65_127", CTLFLAG_RD, &stats->ptc127, "65-127 byte frames transmitted"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "tx_frames_128_255", CTLFLAG_RD, &stats->ptc255, "128-255 byte frames transmitted"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "tx_frames_256_511", CTLFLAG_RD, &stats->ptc511, "256-511 byte frames transmitted"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "tx_frames_512_1023", CTLFLAG_RD, &stats->ptc1023, "512-1023 byte frames transmitted"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "tx_frames_1024_1522", CTLFLAG_RD, &stats->ptc1522, "1024-1522 byte frames transmitted"); } /* ixgbe_add_hw_stats */ /************************************************************************ * ixgbe_sysctl_tdh_handler - Transmit Descriptor Head handler function * * Retrieves the TDH value from the hardware ************************************************************************/ static int ixgbe_sysctl_tdh_handler(SYSCTL_HANDLER_ARGS) { struct tx_ring *txr = ((struct tx_ring *)oidp->oid_arg1); int error; unsigned int val; if (!txr) return (0); val = IXGBE_READ_REG(&txr->adapter->hw, IXGBE_TDH(txr->me)); error = sysctl_handle_int(oidp, &val, 0, req); if (error || !req->newptr) return error; return (0); } /* ixgbe_sysctl_tdh_handler */ /************************************************************************ * ixgbe_sysctl_tdt_handler - Transmit Descriptor Tail handler function * * Retrieves the TDT value from the hardware ************************************************************************/ static int ixgbe_sysctl_tdt_handler(SYSCTL_HANDLER_ARGS) { struct tx_ring *txr = ((struct tx_ring *)oidp->oid_arg1); int error; unsigned int val; if (!txr) return (0); val = IXGBE_READ_REG(&txr->adapter->hw, IXGBE_TDT(txr->me)); error = sysctl_handle_int(oidp, &val, 0, req); if (error || !req->newptr) return error; return (0); } /* ixgbe_sysctl_tdt_handler */ /************************************************************************ * ixgbe_sysctl_rdh_handler - Receive Descriptor Head handler function * * Retrieves the RDH value from the hardware ************************************************************************/ static int ixgbe_sysctl_rdh_handler(SYSCTL_HANDLER_ARGS) { struct rx_ring *rxr = ((struct rx_ring *)oidp->oid_arg1); int error; unsigned int val; if (!rxr) return (0); val = IXGBE_READ_REG(&rxr->adapter->hw, IXGBE_RDH(rxr->me)); error = sysctl_handle_int(oidp, &val, 0, req); if (error || !req->newptr) return error; return (0); } /* ixgbe_sysctl_rdh_handler */ /************************************************************************ * ixgbe_sysctl_rdt_handler - Receive Descriptor Tail handler function * * Retrieves the RDT value from the hardware ************************************************************************/ static int ixgbe_sysctl_rdt_handler(SYSCTL_HANDLER_ARGS) { struct rx_ring *rxr = ((struct rx_ring *)oidp->oid_arg1); int error; unsigned int val; if (!rxr) return (0); val = IXGBE_READ_REG(&rxr->adapter->hw, IXGBE_RDT(rxr->me)); error = sysctl_handle_int(oidp, &val, 0, req); if (error || !req->newptr) return error; return (0); } /* ixgbe_sysctl_rdt_handler */ /************************************************************************ * ixgbe_register_vlan * * Run via vlan config EVENT, it enables us to use the * HW Filter table since we can get the vlan id. This * just creates the entry in the soft version of the * VFTA, init will repopulate the real table. ************************************************************************/ static void ixgbe_register_vlan(void *arg, struct ifnet *ifp, u16 vtag) { struct adapter *adapter = ifp->if_softc; u16 index, bit; if (ifp->if_softc != arg) /* Not our event */ return; if ((vtag == 0) || (vtag > 4095)) /* Invalid */ return; IXGBE_CORE_LOCK(adapter); index = (vtag >> 5) & 0x7F; bit = vtag & 0x1F; adapter->shadow_vfta[index] |= (1 << bit); ++adapter->num_vlans; ixgbe_setup_vlan_hw_support(adapter); IXGBE_CORE_UNLOCK(adapter); } /* ixgbe_register_vlan */ /************************************************************************ * ixgbe_unregister_vlan * * Run via vlan unconfig EVENT, remove our entry in the soft vfta. ************************************************************************/ static void ixgbe_unregister_vlan(void *arg, struct ifnet *ifp, u16 vtag) { struct adapter *adapter = ifp->if_softc; u16 index, bit; if (ifp->if_softc != arg) return; if ((vtag == 0) || (vtag > 4095)) /* Invalid */ return; IXGBE_CORE_LOCK(adapter); index = (vtag >> 5) & 0x7F; bit = vtag & 0x1F; adapter->shadow_vfta[index] &= ~(1 << bit); --adapter->num_vlans; /* Re-init to load the changes */ ixgbe_setup_vlan_hw_support(adapter); IXGBE_CORE_UNLOCK(adapter); } /* ixgbe_unregister_vlan */ /************************************************************************ * ixgbe_setup_vlan_hw_support ************************************************************************/ static void ixgbe_setup_vlan_hw_support(struct adapter *adapter) { struct ifnet *ifp = adapter->ifp; struct ixgbe_hw *hw = &adapter->hw; struct rx_ring *rxr; int i; u32 ctrl; /* * We get here thru init_locked, meaning * a soft reset, this has already cleared * the VFTA and other state, so if there * have been no vlan's registered do nothing. */ if (adapter->num_vlans == 0) return; /* Setup the queues for vlans */ - for (i = 0; i < adapter->num_queues; i++) { - rxr = &adapter->rx_rings[i]; - /* On 82599 the VLAN enable is per/queue in RXDCTL */ - if (hw->mac.type != ixgbe_mac_82598EB) { - ctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxr->me)); - ctrl |= IXGBE_RXDCTL_VME; - IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(rxr->me), ctrl); + if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) { + for (i = 0; i < adapter->num_queues; i++) { + rxr = &adapter->rx_rings[i]; + /* On 82599 the VLAN enable is per/queue in RXDCTL */ + if (hw->mac.type != ixgbe_mac_82598EB) { + ctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxr->me)); + ctrl |= IXGBE_RXDCTL_VME; + IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(rxr->me), ctrl); + } + rxr->vtag_strip = TRUE; } - rxr->vtag_strip = TRUE; } if ((ifp->if_capenable & IFCAP_VLAN_HWFILTER) == 0) return; /* * A soft reset zero's out the VFTA, so * we need to repopulate it now. */ for (i = 0; i < IXGBE_VFTA_SIZE; i++) if (adapter->shadow_vfta[i] != 0) IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), adapter->shadow_vfta[i]); ctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL); /* Enable the Filter Table if enabled */ if (ifp->if_capenable & IFCAP_VLAN_HWFILTER) { ctrl &= ~IXGBE_VLNCTRL_CFIEN; ctrl |= IXGBE_VLNCTRL_VFE; } if (hw->mac.type == ixgbe_mac_82598EB) ctrl |= IXGBE_VLNCTRL_VME; IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, ctrl); } /* ixgbe_setup_vlan_hw_support */ /************************************************************************ * ixgbe_get_slot_info * * Get the width and transaction speed of * the slot this adapter is plugged into. ************************************************************************/ static void ixgbe_get_slot_info(struct adapter *adapter) { device_t dev = adapter->dev; struct ixgbe_hw *hw = &adapter->hw; u32 offset; u16 link; int bus_info_valid = TRUE; /* Some devices are behind an internal bridge */ switch (hw->device_id) { case IXGBE_DEV_ID_82599_SFP_SF_QP: case IXGBE_DEV_ID_82599_QSFP_SF_QP: goto get_parent_info; default: break; } ixgbe_get_bus_info(hw); /* * Some devices don't use PCI-E, but there is no need * to display "Unknown" for bus speed and width. */ switch (hw->mac.type) { case ixgbe_mac_X550EM_x: case ixgbe_mac_X550EM_a: return; default: goto display; } get_parent_info: /* * For the Quad port adapter we need to parse back * up the PCI tree to find the speed of the expansion * slot into which this adapter is plugged. A bit more work. */ dev = device_get_parent(device_get_parent(dev)); #ifdef IXGBE_DEBUG device_printf(dev, "parent pcib = %x,%x,%x\n", pci_get_bus(dev), pci_get_slot(dev), pci_get_function(dev)); #endif dev = device_get_parent(device_get_parent(dev)); #ifdef IXGBE_DEBUG device_printf(dev, "slot pcib = %x,%x,%x\n", pci_get_bus(dev), pci_get_slot(dev), pci_get_function(dev)); #endif /* Now get the PCI Express Capabilities offset */ if (pci_find_cap(dev, PCIY_EXPRESS, &offset)) { /* * Hmm...can't get PCI-Express capabilities. * Falling back to default method. */ bus_info_valid = FALSE; ixgbe_get_bus_info(hw); goto display; } /* ...and read the Link Status Register */ link = pci_read_config(dev, offset + PCIER_LINK_STA, 2); ixgbe_set_pci_config_data_generic(hw, link); display: device_printf(dev, "PCI Express Bus: Speed %s %s\n", ((hw->bus.speed == ixgbe_bus_speed_8000) ? "8.0GT/s" : (hw->bus.speed == ixgbe_bus_speed_5000) ? "5.0GT/s" : (hw->bus.speed == ixgbe_bus_speed_2500) ? "2.5GT/s" : "Unknown"), ((hw->bus.width == ixgbe_bus_width_pcie_x8) ? "Width x8" : (hw->bus.width == ixgbe_bus_width_pcie_x4) ? "Width x4" : (hw->bus.width == ixgbe_bus_width_pcie_x1) ? "Width x1" : "Unknown")); if (bus_info_valid) { if ((hw->device_id != IXGBE_DEV_ID_82599_SFP_SF_QP) && ((hw->bus.width <= ixgbe_bus_width_pcie_x4) && (hw->bus.speed == ixgbe_bus_speed_2500))) { device_printf(dev, "PCI-Express bandwidth available for this card\n is not sufficient for optimal performance.\n"); device_printf(dev, "For optimal performance a x8 PCIE, or x4 PCIE Gen2 slot is required.\n"); } if ((hw->device_id == IXGBE_DEV_ID_82599_SFP_SF_QP) && ((hw->bus.width <= ixgbe_bus_width_pcie_x8) && (hw->bus.speed < ixgbe_bus_speed_8000))) { device_printf(dev, "PCI-Express bandwidth available for this card\n is not sufficient for optimal performance.\n"); device_printf(dev, "For optimal performance a x8 PCIE Gen3 slot is required.\n"); } } else device_printf(dev, "Unable to determine slot speed/width. The speed/width reported are that of the internal switch.\n"); return; } /* ixgbe_get_slot_info */ /************************************************************************ * ixgbe_enable_queue - MSI-X Interrupt Handlers and Tasklets ************************************************************************/ static inline void ixgbe_enable_queue(struct adapter *adapter, u32 vector) { struct ixgbe_hw *hw = &adapter->hw; u64 queue = (u64)(1 << vector); u32 mask; if (hw->mac.type == ixgbe_mac_82598EB) { mask = (IXGBE_EIMS_RTX_QUEUE & queue); IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask); } else { mask = (queue & 0xFFFFFFFF); if (mask) IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask); mask = (queue >> 32); if (mask) IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask); } } /* ixgbe_enable_queue */ /************************************************************************ * ixgbe_disable_queue ************************************************************************/ static inline void ixgbe_disable_queue(struct adapter *adapter, u32 vector) { struct ixgbe_hw *hw = &adapter->hw; u64 queue = (u64)(1 << vector); u32 mask; if (hw->mac.type == ixgbe_mac_82598EB) { mask = (IXGBE_EIMS_RTX_QUEUE & queue); IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask); } else { mask = (queue & 0xFFFFFFFF); if (mask) IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask); mask = (queue >> 32); if (mask) IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask); } } /* ixgbe_disable_queue */ /************************************************************************ * ixgbe_msix_que - MSI-X Queue Interrupt Service routine ************************************************************************/ void ixgbe_msix_que(void *arg) { struct ix_queue *que = arg; struct adapter *adapter = que->adapter; struct ifnet *ifp = adapter->ifp; struct tx_ring *txr = que->txr; struct rx_ring *rxr = que->rxr; bool more; u32 newitr = 0; /* Protect against spurious interrupts */ if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) return; ixgbe_disable_queue(adapter, que->msix); ++que->irqs; more = ixgbe_rxeof(que); IXGBE_TX_LOCK(txr); ixgbe_txeof(txr); if (!ixgbe_ring_empty(ifp, txr->br)) ixgbe_start_locked(ifp, txr); IXGBE_TX_UNLOCK(txr); /* Do AIM now? */ if (adapter->enable_aim == FALSE) goto no_calc; /* * Do Adaptive Interrupt Moderation: * - Write out last calculated setting * - Calculate based on average size over * the last interval. */ if (que->eitr_setting) IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(que->msix), que->eitr_setting); que->eitr_setting = 0; /* Idle, do nothing */ if ((txr->bytes == 0) && (rxr->bytes == 0)) goto no_calc; if ((txr->bytes) && (txr->packets)) newitr = txr->bytes/txr->packets; if ((rxr->bytes) && (rxr->packets)) newitr = max(newitr, (rxr->bytes / rxr->packets)); newitr += 24; /* account for hardware frame, crc */ /* set an upper boundary */ newitr = min(newitr, 3000); /* Be nice to the mid range */ if ((newitr > 300) && (newitr < 1200)) newitr = (newitr / 3); else newitr = (newitr / 2); if (adapter->hw.mac.type == ixgbe_mac_82598EB) newitr |= newitr << 16; else newitr |= IXGBE_EITR_CNT_WDIS; /* save for next interrupt */ que->eitr_setting = newitr; /* Reset state */ txr->bytes = 0; txr->packets = 0; rxr->bytes = 0; rxr->packets = 0; no_calc: if (more) taskqueue_enqueue(que->tq, &que->que_task); else ixgbe_enable_queue(adapter, que->msix); return; } /* ixgbe_msix_que */ /************************************************************************ * ixgbe_media_status - Media Ioctl callback * * Called whenever the user queries the status of * the interface using ifconfig. ************************************************************************/ static void ixgbe_media_status(struct ifnet *ifp, struct ifmediareq *ifmr) { struct adapter *adapter = ifp->if_softc; struct ixgbe_hw *hw = &adapter->hw; int layer; INIT_DEBUGOUT("ixgbe_media_status: begin"); IXGBE_CORE_LOCK(adapter); ixgbe_update_link_status(adapter); ifmr->ifm_status = IFM_AVALID; ifmr->ifm_active = IFM_ETHER; if (!adapter->link_active) { IXGBE_CORE_UNLOCK(adapter); return; } ifmr->ifm_status |= IFM_ACTIVE; layer = adapter->phy_layer; if (layer & IXGBE_PHYSICAL_LAYER_10GBASE_T || layer & IXGBE_PHYSICAL_LAYER_1000BASE_T || layer & IXGBE_PHYSICAL_LAYER_100BASE_TX || layer & IXGBE_PHYSICAL_LAYER_10BASE_T) switch (adapter->link_speed) { case IXGBE_LINK_SPEED_10GB_FULL: ifmr->ifm_active |= IFM_10G_T | IFM_FDX; break; case IXGBE_LINK_SPEED_1GB_FULL: ifmr->ifm_active |= IFM_1000_T | IFM_FDX; break; case IXGBE_LINK_SPEED_100_FULL: ifmr->ifm_active |= IFM_100_TX | IFM_FDX; break; case IXGBE_LINK_SPEED_10_FULL: ifmr->ifm_active |= IFM_10_T | IFM_FDX; break; } if (layer & IXGBE_PHYSICAL_LAYER_SFP_PLUS_CU || layer & IXGBE_PHYSICAL_LAYER_SFP_ACTIVE_DA) switch (adapter->link_speed) { case IXGBE_LINK_SPEED_10GB_FULL: ifmr->ifm_active |= IFM_10G_TWINAX | IFM_FDX; break; } if (layer & IXGBE_PHYSICAL_LAYER_10GBASE_LR) switch (adapter->link_speed) { case IXGBE_LINK_SPEED_10GB_FULL: ifmr->ifm_active |= IFM_10G_LR | IFM_FDX; break; case IXGBE_LINK_SPEED_1GB_FULL: ifmr->ifm_active |= IFM_1000_LX | IFM_FDX; break; } if (layer & IXGBE_PHYSICAL_LAYER_10GBASE_LRM) switch (adapter->link_speed) { case IXGBE_LINK_SPEED_10GB_FULL: ifmr->ifm_active |= IFM_10G_LRM | IFM_FDX; break; case IXGBE_LINK_SPEED_1GB_FULL: ifmr->ifm_active |= IFM_1000_LX | IFM_FDX; break; } if (layer & IXGBE_PHYSICAL_LAYER_10GBASE_SR || layer & IXGBE_PHYSICAL_LAYER_1000BASE_SX) switch (adapter->link_speed) { case IXGBE_LINK_SPEED_10GB_FULL: ifmr->ifm_active |= IFM_10G_SR | IFM_FDX; break; case IXGBE_LINK_SPEED_1GB_FULL: ifmr->ifm_active |= IFM_1000_SX | IFM_FDX; break; } if (layer & IXGBE_PHYSICAL_LAYER_10GBASE_CX4) switch (adapter->link_speed) { case IXGBE_LINK_SPEED_10GB_FULL: ifmr->ifm_active |= IFM_10G_CX4 | IFM_FDX; break; } /* * XXX: These need to use the proper media types once * they're added. */ #ifndef IFM_ETH_XTYPE if (layer & IXGBE_PHYSICAL_LAYER_10GBASE_KR) switch (adapter->link_speed) { case IXGBE_LINK_SPEED_10GB_FULL: ifmr->ifm_active |= IFM_10G_SR | IFM_FDX; break; case IXGBE_LINK_SPEED_2_5GB_FULL: ifmr->ifm_active |= IFM_2500_SX | IFM_FDX; break; case IXGBE_LINK_SPEED_1GB_FULL: ifmr->ifm_active |= IFM_1000_CX | IFM_FDX; break; } else if (layer & IXGBE_PHYSICAL_LAYER_10GBASE_KX4 || layer & IXGBE_PHYSICAL_LAYER_2500BASE_KX || layer & IXGBE_PHYSICAL_LAYER_1000BASE_KX) switch (adapter->link_speed) { case IXGBE_LINK_SPEED_10GB_FULL: ifmr->ifm_active |= IFM_10G_CX4 | IFM_FDX; break; case IXGBE_LINK_SPEED_2_5GB_FULL: ifmr->ifm_active |= IFM_2500_SX | IFM_FDX; break; case IXGBE_LINK_SPEED_1GB_FULL: ifmr->ifm_active |= IFM_1000_CX | IFM_FDX; break; } #else if (layer & IXGBE_PHYSICAL_LAYER_10GBASE_KR) switch (adapter->link_speed) { case IXGBE_LINK_SPEED_10GB_FULL: ifmr->ifm_active |= IFM_10G_KR | IFM_FDX; break; case IXGBE_LINK_SPEED_2_5GB_FULL: ifmr->ifm_active |= IFM_2500_KX | IFM_FDX; break; case IXGBE_LINK_SPEED_1GB_FULL: ifmr->ifm_active |= IFM_1000_KX | IFM_FDX; break; } else if (layer & IXGBE_PHYSICAL_LAYER_10GBASE_KX4 || layer & IXGBE_PHYSICAL_LAYER_2500BASE_KX || layer & IXGBE_PHYSICAL_LAYER_1000BASE_KX) switch (adapter->link_speed) { case IXGBE_LINK_SPEED_10GB_FULL: ifmr->ifm_active |= IFM_10G_KX4 | IFM_FDX; break; case IXGBE_LINK_SPEED_2_5GB_FULL: ifmr->ifm_active |= IFM_2500_KX | IFM_FDX; break; case IXGBE_LINK_SPEED_1GB_FULL: ifmr->ifm_active |= IFM_1000_KX | IFM_FDX; break; } #endif /* If nothing is recognized... */ if (IFM_SUBTYPE(ifmr->ifm_active) == 0) ifmr->ifm_active |= IFM_UNKNOWN; #if __FreeBSD_version >= 900025 /* Display current flow control setting used on link */ if (hw->fc.current_mode == ixgbe_fc_rx_pause || hw->fc.current_mode == ixgbe_fc_full) ifmr->ifm_active |= IFM_ETH_RXPAUSE; if (hw->fc.current_mode == ixgbe_fc_tx_pause || hw->fc.current_mode == ixgbe_fc_full) ifmr->ifm_active |= IFM_ETH_TXPAUSE; #endif IXGBE_CORE_UNLOCK(adapter); return; } /* ixgbe_media_status */ /************************************************************************ * ixgbe_media_change - Media Ioctl callback * * Called when the user changes speed/duplex using * media/mediopt option with ifconfig. ************************************************************************/ static int ixgbe_media_change(struct ifnet *ifp) { struct adapter *adapter = ifp->if_softc; struct ifmedia *ifm = &adapter->media; struct ixgbe_hw *hw = &adapter->hw; ixgbe_link_speed speed = 0; INIT_DEBUGOUT("ixgbe_media_change: begin"); if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER) return (EINVAL); if (hw->phy.media_type == ixgbe_media_type_backplane) return (ENODEV); /* * We don't actually need to check against the supported * media types of the adapter; ifmedia will take care of * that for us. */ switch (IFM_SUBTYPE(ifm->ifm_media)) { case IFM_AUTO: case IFM_10G_T: speed |= IXGBE_LINK_SPEED_100_FULL; speed |= IXGBE_LINK_SPEED_1GB_FULL; speed |= IXGBE_LINK_SPEED_10GB_FULL; break; case IFM_10G_LRM: case IFM_10G_LR: #ifndef IFM_ETH_XTYPE case IFM_10G_SR: /* KR, too */ case IFM_10G_CX4: /* KX4 */ #else case IFM_10G_KR: case IFM_10G_KX4: #endif speed |= IXGBE_LINK_SPEED_1GB_FULL; speed |= IXGBE_LINK_SPEED_10GB_FULL; break; #ifndef IFM_ETH_XTYPE case IFM_1000_CX: /* KX */ #else case IFM_1000_KX: #endif case IFM_1000_LX: case IFM_1000_SX: speed |= IXGBE_LINK_SPEED_1GB_FULL; break; case IFM_1000_T: speed |= IXGBE_LINK_SPEED_100_FULL; speed |= IXGBE_LINK_SPEED_1GB_FULL; break; case IFM_10G_TWINAX: speed |= IXGBE_LINK_SPEED_10GB_FULL; break; case IFM_100_TX: speed |= IXGBE_LINK_SPEED_100_FULL; break; case IFM_10_T: speed |= IXGBE_LINK_SPEED_10_FULL; break; default: goto invalid; } hw->mac.autotry_restart = TRUE; hw->mac.ops.setup_link(hw, speed, TRUE); adapter->advertise = ((speed & IXGBE_LINK_SPEED_10GB_FULL) ? 4 : 0) | ((speed & IXGBE_LINK_SPEED_1GB_FULL) ? 2 : 0) | ((speed & IXGBE_LINK_SPEED_100_FULL) ? 1 : 0) | ((speed & IXGBE_LINK_SPEED_10_FULL) ? 8 : 0); return (0); invalid: device_printf(adapter->dev, "Invalid media type!\n"); return (EINVAL); } /* ixgbe_media_change */ /************************************************************************ * ixgbe_set_promisc ************************************************************************/ static void ixgbe_set_promisc(struct adapter *adapter) { struct ifnet *ifp = adapter->ifp; int mcnt = 0; u32 rctl; rctl = IXGBE_READ_REG(&adapter->hw, IXGBE_FCTRL); rctl &= (~IXGBE_FCTRL_UPE); if (ifp->if_flags & IFF_ALLMULTI) mcnt = MAX_NUM_MULTICAST_ADDRESSES; else { struct ifmultiaddr *ifma; #if __FreeBSD_version < 800000 IF_ADDR_LOCK(ifp); #else if_maddr_rlock(ifp); #endif TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { if (ifma->ifma_addr->sa_family != AF_LINK) continue; if (mcnt == MAX_NUM_MULTICAST_ADDRESSES) break; mcnt++; } #if __FreeBSD_version < 800000 IF_ADDR_UNLOCK(ifp); #else if_maddr_runlock(ifp); #endif } if (mcnt < MAX_NUM_MULTICAST_ADDRESSES) rctl &= (~IXGBE_FCTRL_MPE); IXGBE_WRITE_REG(&adapter->hw, IXGBE_FCTRL, rctl); if (ifp->if_flags & IFF_PROMISC) { rctl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE); IXGBE_WRITE_REG(&adapter->hw, IXGBE_FCTRL, rctl); } else if (ifp->if_flags & IFF_ALLMULTI) { rctl |= IXGBE_FCTRL_MPE; rctl &= ~IXGBE_FCTRL_UPE; IXGBE_WRITE_REG(&adapter->hw, IXGBE_FCTRL, rctl); } } /* ixgbe_set_promisc */ /************************************************************************ * ixgbe_msix_link - Link status change ISR (MSI/MSI-X) ************************************************************************/ static void ixgbe_msix_link(void *arg) { struct adapter *adapter = arg; struct ixgbe_hw *hw = &adapter->hw; u32 eicr, eicr_mask; s32 retval; ++adapter->link_irq; /* Pause other interrupts */ IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_OTHER); /* First get the cause */ eicr = IXGBE_READ_REG(hw, IXGBE_EICS); /* Be sure the queue bits are not cleared */ eicr &= ~IXGBE_EICR_RTX_QUEUE; /* Clear interrupt with write */ IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr); /* Link status change */ if (eicr & IXGBE_EICR_LSC) { IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC); taskqueue_enqueue(adapter->tq, &adapter->link_task); } if (adapter->hw.mac.type != ixgbe_mac_82598EB) { if ((adapter->feat_en & IXGBE_FEATURE_FDIR) && (eicr & IXGBE_EICR_FLOW_DIR)) { /* This is probably overkill :) */ if (!atomic_cmpset_int(&adapter->fdir_reinit, 0, 1)) return; /* Disable the interrupt */ IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR); taskqueue_enqueue(adapter->tq, &adapter->fdir_task); } if (eicr & IXGBE_EICR_ECC) { device_printf(adapter->dev, "CRITICAL: ECC ERROR!! Please Reboot!!\n"); IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC); } /* Check for over temp condition */ if (adapter->feat_en & IXGBE_FEATURE_TEMP_SENSOR) { switch (adapter->hw.mac.type) { case ixgbe_mac_X550EM_a: if (!(eicr & IXGBE_EICR_GPI_SDP0_X550EM_a)) break; IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EICR_GPI_SDP0_X550EM_a); IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP0_X550EM_a); retval = hw->phy.ops.check_overtemp(hw); if (retval != IXGBE_ERR_OVERTEMP) break; device_printf(adapter->dev, "CRITICAL: OVER TEMP!! PHY IS SHUT DOWN!!\n"); device_printf(adapter->dev, "System shutdown required!\n"); break; default: if (!(eicr & IXGBE_EICR_TS)) break; retval = hw->phy.ops.check_overtemp(hw); if (retval != IXGBE_ERR_OVERTEMP) break; device_printf(adapter->dev, "CRITICAL: OVER TEMP!! PHY IS SHUT DOWN!!\n"); device_printf(adapter->dev, "System shutdown required!\n"); IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_TS); break; } } /* Check for VF message */ if ((adapter->feat_en & IXGBE_FEATURE_SRIOV) && (eicr & IXGBE_EICR_MAILBOX)) taskqueue_enqueue(adapter->tq, &adapter->mbx_task); } if (ixgbe_is_sfp(hw)) { /* Pluggable optics-related interrupt */ if (hw->mac.type >= ixgbe_mac_X540) eicr_mask = IXGBE_EICR_GPI_SDP0_X540; else eicr_mask = IXGBE_EICR_GPI_SDP2_BY_MAC(hw); if (eicr & eicr_mask) { IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr_mask); taskqueue_enqueue(adapter->tq, &adapter->mod_task); } if ((hw->mac.type == ixgbe_mac_82599EB) && (eicr & IXGBE_EICR_GPI_SDP1_BY_MAC(hw))) { IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1_BY_MAC(hw)); taskqueue_enqueue(adapter->tq, &adapter->msf_task); } } /* Check for fan failure */ if (adapter->feat_en & IXGBE_FEATURE_FAN_FAIL) { ixgbe_check_fan_failure(adapter, eicr, TRUE); IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1_BY_MAC(hw)); } /* External PHY interrupt */ if ((hw->phy.type == ixgbe_phy_x550em_ext_t) && (eicr & IXGBE_EICR_GPI_SDP0_X540)) { IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP0_X540); taskqueue_enqueue(adapter->tq, &adapter->phy_task); } /* Re-enable other interrupts */ IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_OTHER); } /* ixgbe_msix_link */ /************************************************************************ * ixgbe_sysctl_interrupt_rate_handler ************************************************************************/ static int ixgbe_sysctl_interrupt_rate_handler(SYSCTL_HANDLER_ARGS) { struct ix_queue *que = ((struct ix_queue *)oidp->oid_arg1); int error; unsigned int reg, usec, rate; reg = IXGBE_READ_REG(&que->adapter->hw, IXGBE_EITR(que->msix)); usec = ((reg & 0x0FF8) >> 3); if (usec > 0) rate = 500000 / usec; else rate = 0; error = sysctl_handle_int(oidp, &rate, 0, req); if (error || !req->newptr) return error; reg &= ~0xfff; /* default, no limitation */ ixgbe_max_interrupt_rate = 0; if (rate > 0 && rate < 500000) { if (rate < 1000) rate = 1000; ixgbe_max_interrupt_rate = rate; reg |= ((4000000/rate) & 0xff8); } IXGBE_WRITE_REG(&que->adapter->hw, IXGBE_EITR(que->msix), reg); return (0); } /* ixgbe_sysctl_interrupt_rate_handler */ /************************************************************************ * ixgbe_add_device_sysctls ************************************************************************/ static void ixgbe_add_device_sysctls(struct adapter *adapter) { device_t dev = adapter->dev; struct ixgbe_hw *hw = &adapter->hw; struct sysctl_oid_list *child; struct sysctl_ctx_list *ctx; ctx = device_get_sysctl_ctx(dev); child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev)); /* Sysctls for all devices */ SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "fc", CTLTYPE_INT | CTLFLAG_RW, adapter, 0, ixgbe_sysctl_flowcntl, "I", IXGBE_SYSCTL_DESC_SET_FC); adapter->enable_aim = ixgbe_enable_aim; SYSCTL_ADD_INT(ctx, child, OID_AUTO, "enable_aim", CTLFLAG_RW, &adapter->enable_aim, 1, "Interrupt Moderation"); SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "advertise_speed", CTLTYPE_INT | CTLFLAG_RW, adapter, 0, ixgbe_sysctl_advertise, "I", IXGBE_SYSCTL_DESC_ADV_SPEED); #ifdef IXGBE_DEBUG /* testing sysctls (for all devices) */ SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "power_state", CTLTYPE_INT | CTLFLAG_RW, adapter, 0, ixgbe_sysctl_power_state, "I", "PCI Power State"); SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "print_rss_config", CTLTYPE_STRING | CTLFLAG_RD, adapter, 0, ixgbe_sysctl_print_rss_config, "A", "Prints RSS Configuration"); #endif /* for X550 series devices */ if (hw->mac.type >= ixgbe_mac_X550) SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "dmac", CTLTYPE_INT | CTLFLAG_RW, adapter, 0, ixgbe_sysctl_dmac, "I", "DMA Coalesce"); /* for WoL-capable devices */ if (hw->device_id == IXGBE_DEV_ID_X550EM_X_10G_T) { SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "wol_enable", CTLTYPE_INT | CTLFLAG_RW, adapter, 0, ixgbe_sysctl_wol_enable, "I", "Enable/Disable Wake on LAN"); SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "wufc", CTLTYPE_INT | CTLFLAG_RW, adapter, 0, ixgbe_sysctl_wufc, "I", "Enable/Disable Wake Up Filters"); } /* for X552/X557-AT devices */ if (hw->device_id == IXGBE_DEV_ID_X550EM_X_10G_T) { struct sysctl_oid *phy_node; struct sysctl_oid_list *phy_list; phy_node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "phy", CTLFLAG_RD, NULL, "External PHY sysctls"); phy_list = SYSCTL_CHILDREN(phy_node); SYSCTL_ADD_PROC(ctx, phy_list, OID_AUTO, "temp", CTLTYPE_INT | CTLFLAG_RD, adapter, 0, ixgbe_sysctl_phy_temp, "I", "Current External PHY Temperature (Celsius)"); SYSCTL_ADD_PROC(ctx, phy_list, OID_AUTO, "overtemp_occurred", CTLTYPE_INT | CTLFLAG_RD, adapter, 0, ixgbe_sysctl_phy_overtemp_occurred, "I", "External PHY High Temperature Event Occurred"); } if (adapter->feat_cap & IXGBE_FEATURE_EEE) { SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "eee_state", CTLTYPE_INT | CTLFLAG_RW, adapter, 0, ixgbe_sysctl_eee_state, "I", "EEE Power Save State"); } } /* ixgbe_add_device_sysctls */ /************************************************************************ * ixgbe_allocate_pci_resources ************************************************************************/ static int ixgbe_allocate_pci_resources(struct adapter *adapter) { device_t dev = adapter->dev; int rid; rid = PCIR_BAR(0); adapter->pci_mem = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid, RF_ACTIVE); if (!(adapter->pci_mem)) { device_printf(dev, "Unable to allocate bus resource: memory\n"); return (ENXIO); } /* Save bus_space values for READ/WRITE_REG macros */ adapter->osdep.mem_bus_space_tag = rman_get_bustag(adapter->pci_mem); adapter->osdep.mem_bus_space_handle = rman_get_bushandle(adapter->pci_mem); /* Set hw values for shared code */ adapter->hw.hw_addr = (u8 *)&adapter->osdep.mem_bus_space_handle; return (0); } /* ixgbe_allocate_pci_resources */ /************************************************************************ * ixgbe_detach - Device removal routine * * Called when the driver is being removed. * Stops the adapter and deallocates all the resources * that were allocated for driver operation. * * return 0 on success, positive on failure ************************************************************************/ static int ixgbe_detach(device_t dev) { struct adapter *adapter = device_get_softc(dev); struct ix_queue *que = adapter->queues; struct tx_ring *txr = adapter->tx_rings; u32 ctrl_ext; INIT_DEBUGOUT("ixgbe_detach: begin"); /* Make sure VLANS are not using driver */ if (adapter->ifp->if_vlantrunk != NULL) { device_printf(dev, "Vlan in use, detach first\n"); return (EBUSY); } if (ixgbe_pci_iov_detach(dev) != 0) { device_printf(dev, "SR-IOV in use; detach first.\n"); return (EBUSY); } ether_ifdetach(adapter->ifp); /* Stop the adapter */ IXGBE_CORE_LOCK(adapter); ixgbe_setup_low_power_mode(adapter); IXGBE_CORE_UNLOCK(adapter); for (int i = 0; i < adapter->num_queues; i++, que++, txr++) { if (que->tq) { if (!(adapter->feat_en & IXGBE_FEATURE_LEGACY_TX)) taskqueue_drain(que->tq, &txr->txq_task); taskqueue_drain(que->tq, &que->que_task); taskqueue_free(que->tq); } } /* Drain the Link queue */ if (adapter->tq) { taskqueue_drain(adapter->tq, &adapter->link_task); taskqueue_drain(adapter->tq, &adapter->mod_task); taskqueue_drain(adapter->tq, &adapter->msf_task); if (adapter->feat_cap & IXGBE_FEATURE_SRIOV) taskqueue_drain(adapter->tq, &adapter->mbx_task); taskqueue_drain(adapter->tq, &adapter->phy_task); if (adapter->feat_en & IXGBE_FEATURE_FDIR) taskqueue_drain(adapter->tq, &adapter->fdir_task); taskqueue_free(adapter->tq); } /* let hardware know driver is unloading */ ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT); ctrl_ext &= ~IXGBE_CTRL_EXT_DRV_LOAD; IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT, ctrl_ext); /* Unregister VLAN events */ if (adapter->vlan_attach != NULL) EVENTHANDLER_DEREGISTER(vlan_config, adapter->vlan_attach); if (adapter->vlan_detach != NULL) EVENTHANDLER_DEREGISTER(vlan_unconfig, adapter->vlan_detach); callout_drain(&adapter->timer); if (adapter->feat_en & IXGBE_FEATURE_NETMAP) netmap_detach(adapter->ifp); ixgbe_free_pci_resources(adapter); bus_generic_detach(dev); if_free(adapter->ifp); ixgbe_free_transmit_structures(adapter); ixgbe_free_receive_structures(adapter); free(adapter->queues, M_DEVBUF); free(adapter->mta, M_IXGBE); IXGBE_CORE_LOCK_DESTROY(adapter); return (0); } /* ixgbe_detach */ /************************************************************************ * ixgbe_setup_low_power_mode - LPLU/WoL preparation * * Prepare the adapter/port for LPLU and/or WoL ************************************************************************/ static int ixgbe_setup_low_power_mode(struct adapter *adapter) { struct ixgbe_hw *hw = &adapter->hw; device_t dev = adapter->dev; s32 error = 0; mtx_assert(&adapter->core_mtx, MA_OWNED); /* Limit power management flow to X550EM baseT */ if (hw->device_id == IXGBE_DEV_ID_X550EM_X_10G_T && hw->phy.ops.enter_lplu) { /* Turn off support for APM wakeup. (Using ACPI instead) */ IXGBE_WRITE_REG(hw, IXGBE_GRC, IXGBE_READ_REG(hw, IXGBE_GRC) & ~(u32)2); /* * Clear Wake Up Status register to prevent any previous wakeup * events from waking us up immediately after we suspend. */ IXGBE_WRITE_REG(hw, IXGBE_WUS, 0xffffffff); /* * Program the Wakeup Filter Control register with user filter * settings */ IXGBE_WRITE_REG(hw, IXGBE_WUFC, adapter->wufc); /* Enable wakeups and power management in Wakeup Control */ IXGBE_WRITE_REG(hw, IXGBE_WUC, IXGBE_WUC_WKEN | IXGBE_WUC_PME_EN); /* X550EM baseT adapters need a special LPLU flow */ hw->phy.reset_disable = true; ixgbe_stop(adapter); error = hw->phy.ops.enter_lplu(hw); if (error) device_printf(dev, "Error entering LPLU: %d\n", error); hw->phy.reset_disable = false; } else { /* Just stop for other adapters */ ixgbe_stop(adapter); } return error; } /* ixgbe_setup_low_power_mode */ /************************************************************************ * ixgbe_shutdown - Shutdown entry point ************************************************************************/ static int ixgbe_shutdown(device_t dev) { struct adapter *adapter = device_get_softc(dev); int error = 0; INIT_DEBUGOUT("ixgbe_shutdown: begin"); IXGBE_CORE_LOCK(adapter); error = ixgbe_setup_low_power_mode(adapter); IXGBE_CORE_UNLOCK(adapter); return (error); } /* ixgbe_shutdown */ /************************************************************************ * ixgbe_suspend * * From D0 to D3 ************************************************************************/ static int ixgbe_suspend(device_t dev) { struct adapter *adapter = device_get_softc(dev); int error = 0; INIT_DEBUGOUT("ixgbe_suspend: begin"); IXGBE_CORE_LOCK(adapter); error = ixgbe_setup_low_power_mode(adapter); IXGBE_CORE_UNLOCK(adapter); return (error); } /* ixgbe_suspend */ /************************************************************************ * ixgbe_resume * * From D3 to D0 ************************************************************************/ static int ixgbe_resume(device_t dev) { struct adapter *adapter = device_get_softc(dev); struct ifnet *ifp = adapter->ifp; struct ixgbe_hw *hw = &adapter->hw; u32 wus; INIT_DEBUGOUT("ixgbe_resume: begin"); IXGBE_CORE_LOCK(adapter); /* Read & clear WUS register */ wus = IXGBE_READ_REG(hw, IXGBE_WUS); if (wus) device_printf(dev, "Woken up by (WUS): %#010x\n", IXGBE_READ_REG(hw, IXGBE_WUS)); IXGBE_WRITE_REG(hw, IXGBE_WUS, 0xffffffff); /* And clear WUFC until next low-power transition */ IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0); /* * Required after D3->D0 transition; * will re-advertise all previous advertised speeds */ if (ifp->if_flags & IFF_UP) ixgbe_init_locked(adapter); IXGBE_CORE_UNLOCK(adapter); return (0); } /* ixgbe_resume */ /************************************************************************ * ixgbe_set_if_hwassist - Set the various hardware offload abilities. * * Takes the ifnet's if_capenable flags (e.g. set by the user using * ifconfig) and indicates to the OS via the ifnet's if_hwassist * field what mbuf offload flags the driver will understand. ************************************************************************/ static void ixgbe_set_if_hwassist(struct adapter *adapter) { struct ifnet *ifp = adapter->ifp; ifp->if_hwassist = 0; #if __FreeBSD_version >= 1000000 if (ifp->if_capenable & IFCAP_TSO4) ifp->if_hwassist |= CSUM_IP_TSO; if (ifp->if_capenable & IFCAP_TSO6) ifp->if_hwassist |= CSUM_IP6_TSO; if (ifp->if_capenable & IFCAP_TXCSUM) { ifp->if_hwassist |= (CSUM_IP | CSUM_IP_UDP | CSUM_IP_TCP); if (adapter->hw.mac.type != ixgbe_mac_82598EB) ifp->if_hwassist |= CSUM_IP_SCTP; } if (ifp->if_capenable & IFCAP_TXCSUM_IPV6) { ifp->if_hwassist |= (CSUM_IP6_UDP | CSUM_IP6_TCP); if (adapter->hw.mac.type != ixgbe_mac_82598EB) ifp->if_hwassist |= CSUM_IP6_SCTP; } #else if (ifp->if_capenable & IFCAP_TSO) ifp->if_hwassist |= CSUM_TSO; if (ifp->if_capenable & IFCAP_TXCSUM) { ifp->if_hwassist |= (CSUM_TCP | CSUM_UDP); if (adapter->hw.mac.type != ixgbe_mac_82598EB) ifp->if_hwassist |= CSUM_SCTP; } #endif } /* ixgbe_set_if_hwassist */ /************************************************************************ * ixgbe_init_locked - Init entry point * * Used in two ways: It is used by the stack as an init * entry point in network interface structure. It is also * used by the driver as a hw/sw initialization routine to * get to a consistent state. * * return 0 on success, positive on failure ************************************************************************/ void ixgbe_init_locked(struct adapter *adapter) { struct ifnet *ifp = adapter->ifp; device_t dev = adapter->dev; struct ixgbe_hw *hw = &adapter->hw; struct tx_ring *txr; struct rx_ring *rxr; u32 txdctl, mhadd; u32 rxdctl, rxctrl; u32 ctrl_ext; int err = 0; mtx_assert(&adapter->core_mtx, MA_OWNED); INIT_DEBUGOUT("ixgbe_init_locked: begin"); hw->adapter_stopped = FALSE; ixgbe_stop_adapter(hw); callout_stop(&adapter->timer); /* Queue indices may change with IOV mode */ ixgbe_align_all_queue_indices(adapter); /* reprogram the RAR[0] in case user changed it. */ ixgbe_set_rar(hw, 0, hw->mac.addr, adapter->pool, IXGBE_RAH_AV); /* Get the latest mac address, User can use a LAA */ bcopy(IF_LLADDR(ifp), hw->mac.addr, IXGBE_ETH_LENGTH_OF_ADDRESS); ixgbe_set_rar(hw, 0, hw->mac.addr, adapter->pool, 1); hw->addr_ctrl.rar_used_count = 1; /* Set hardware offload abilities from ifnet flags */ ixgbe_set_if_hwassist(adapter); /* Prepare transmit descriptors and buffers */ if (ixgbe_setup_transmit_structures(adapter)) { device_printf(dev, "Could not setup transmit structures\n"); ixgbe_stop(adapter); return; } ixgbe_init_hw(hw); ixgbe_initialize_iov(adapter); ixgbe_initialize_transmit_units(adapter); /* Setup Multicast table */ ixgbe_set_multi(adapter); /* Determine the correct mbuf pool, based on frame size */ if (adapter->max_frame_size <= MCLBYTES) adapter->rx_mbuf_sz = MCLBYTES; else adapter->rx_mbuf_sz = MJUMPAGESIZE; /* Prepare receive descriptors and buffers */ if (ixgbe_setup_receive_structures(adapter)) { device_printf(dev, "Could not setup receive structures\n"); ixgbe_stop(adapter); return; } /* Configure RX settings */ ixgbe_initialize_receive_units(adapter); /* Enable SDP & MSI-X interrupts based on adapter */ ixgbe_config_gpie(adapter); /* Set MTU size */ if (ifp->if_mtu > ETHERMTU) { /* aka IXGBE_MAXFRS on 82599 and newer */ mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD); mhadd &= ~IXGBE_MHADD_MFS_MASK; mhadd |= adapter->max_frame_size << IXGBE_MHADD_MFS_SHIFT; IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd); } /* Now enable all the queues */ for (int i = 0; i < adapter->num_queues; i++) { txr = &adapter->tx_rings[i]; txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(txr->me)); txdctl |= IXGBE_TXDCTL_ENABLE; /* Set WTHRESH to 8, burst writeback */ txdctl |= (8 << 16); /* * When the internal queue falls below PTHRESH (32), * start prefetching as long as there are at least * HTHRESH (1) buffers ready. The values are taken * from the Intel linux driver 3.8.21. * Prefetching enables tx line rate even with 1 queue. */ txdctl |= (32 << 0) | (1 << 8); IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(txr->me), txdctl); } for (int i = 0, j = 0; i < adapter->num_queues; i++) { rxr = &adapter->rx_rings[i]; rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxr->me)); if (hw->mac.type == ixgbe_mac_82598EB) { /* * PTHRESH = 21 * HTHRESH = 4 * WTHRESH = 8 */ rxdctl &= ~0x3FFFFF; rxdctl |= 0x080420; } rxdctl |= IXGBE_RXDCTL_ENABLE; IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(rxr->me), rxdctl); for (; j < 10; j++) { if (IXGBE_READ_REG(hw, IXGBE_RXDCTL(rxr->me)) & IXGBE_RXDCTL_ENABLE) break; else msec_delay(1); } wmb(); /* * In netmap mode, we must preserve the buffers made * available to userspace before the if_init() * (this is true by default on the TX side, because * init makes all buffers available to userspace). * * netmap_reset() and the device specific routines * (e.g. ixgbe_setup_receive_rings()) map these * buffers at the end of the NIC ring, so here we * must set the RDT (tail) register to make sure * they are not overwritten. * * In this driver the NIC ring starts at RDH = 0, * RDT points to the last slot available for reception (?), * so RDT = num_rx_desc - 1 means the whole ring is available. */ #ifdef DEV_NETMAP if ((adapter->feat_en & IXGBE_FEATURE_NETMAP) && (ifp->if_capenable & IFCAP_NETMAP)) { struct netmap_adapter *na = NA(adapter->ifp); struct netmap_kring *kring = &na->rx_rings[i]; int t = na->num_rx_desc - 1 - nm_kr_rxspace(kring); IXGBE_WRITE_REG(hw, IXGBE_RDT(rxr->me), t); } else #endif /* DEV_NETMAP */ IXGBE_WRITE_REG(hw, IXGBE_RDT(rxr->me), adapter->num_rx_desc - 1); } /* Enable Receive engine */ rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL); if (hw->mac.type == ixgbe_mac_82598EB) rxctrl |= IXGBE_RXCTRL_DMBYPS; rxctrl |= IXGBE_RXCTRL_RXEN; ixgbe_enable_rx_dma(hw, rxctrl); callout_reset(&adapter->timer, hz, ixgbe_local_timer, adapter); /* Set up MSI-X routing */ if (adapter->feat_en & IXGBE_FEATURE_MSIX) { ixgbe_configure_ivars(adapter); /* Set up auto-mask */ if (hw->mac.type == ixgbe_mac_82598EB) IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE); else { IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF); IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF); } } else { /* Simple settings for Legacy/MSI */ ixgbe_set_ivar(adapter, 0, 0, 0); ixgbe_set_ivar(adapter, 0, 0, 1); IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE); } ixgbe_init_fdir(adapter); /* * Check on any SFP devices that * need to be kick-started */ if (hw->phy.type == ixgbe_phy_none) { err = hw->phy.ops.identify(hw); if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) { device_printf(dev, "Unsupported SFP+ module type was detected.\n"); return; } } /* Set moderation on the Link interrupt */ IXGBE_WRITE_REG(hw, IXGBE_EITR(adapter->vector), IXGBE_LINK_ITR); /* Config/Enable Link */ ixgbe_config_link(adapter); /* Hardware Packet Buffer & Flow Control setup */ ixgbe_config_delay_values(adapter); /* Initialize the FC settings */ ixgbe_start_hw(hw); /* Set up VLAN support and filter */ ixgbe_setup_vlan_hw_support(adapter); /* Setup DMA Coalescing */ ixgbe_config_dmac(adapter); /* And now turn on interrupts */ ixgbe_enable_intr(adapter); /* Enable the use of the MBX by the VF's */ if (adapter->feat_en & IXGBE_FEATURE_SRIOV) { ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT); ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD; IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext); } /* Now inform the stack we're ready */ ifp->if_drv_flags |= IFF_DRV_RUNNING; return; } /* ixgbe_init_locked */ /************************************************************************ * ixgbe_init ************************************************************************/ static void ixgbe_init(void *arg) { struct adapter *adapter = arg; IXGBE_CORE_LOCK(adapter); ixgbe_init_locked(adapter); IXGBE_CORE_UNLOCK(adapter); return; } /* ixgbe_init */ /************************************************************************ * ixgbe_set_ivar * * Setup the correct IVAR register for a particular MSI-X interrupt * (yes this is all very magic and confusing :) * - entry is the register array entry * - vector is the MSI-X vector for this queue * - type is RX/TX/MISC ************************************************************************/ static void ixgbe_set_ivar(struct adapter *adapter, u8 entry, u8 vector, s8 type) { struct ixgbe_hw *hw = &adapter->hw; u32 ivar, index; vector |= IXGBE_IVAR_ALLOC_VAL; switch (hw->mac.type) { case ixgbe_mac_82598EB: if (type == -1) entry = IXGBE_IVAR_OTHER_CAUSES_INDEX; else entry += (type * 64); index = (entry >> 2) & 0x1F; ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index)); ivar &= ~(0xFF << (8 * (entry & 0x3))); ivar |= (vector << (8 * (entry & 0x3))); IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR(index), ivar); break; case ixgbe_mac_82599EB: case ixgbe_mac_X540: case ixgbe_mac_X550: case ixgbe_mac_X550EM_x: case ixgbe_mac_X550EM_a: if (type == -1) { /* MISC IVAR */ index = (entry & 1) * 8; ivar = IXGBE_READ_REG(hw, IXGBE_IVAR_MISC); ivar &= ~(0xFF << index); ivar |= (vector << index); IXGBE_WRITE_REG(hw, IXGBE_IVAR_MISC, ivar); } else { /* RX/TX IVARS */ index = (16 * (entry & 1)) + (8 * type); ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(entry >> 1)); ivar &= ~(0xFF << index); ivar |= (vector << index); IXGBE_WRITE_REG(hw, IXGBE_IVAR(entry >> 1), ivar); } default: break; } } /* ixgbe_set_ivar */ /************************************************************************ * ixgbe_configure_ivars ************************************************************************/ static void ixgbe_configure_ivars(struct adapter *adapter) { struct ix_queue *que = adapter->queues; u32 newitr; if (ixgbe_max_interrupt_rate > 0) newitr = (4000000 / ixgbe_max_interrupt_rate) & 0x0FF8; else { /* * Disable DMA coalescing if interrupt moderation is * disabled. */ adapter->dmac = 0; newitr = 0; } for (int i = 0; i < adapter->num_queues; i++, que++) { struct rx_ring *rxr = &adapter->rx_rings[i]; struct tx_ring *txr = &adapter->tx_rings[i]; /* First the RX queue entry */ ixgbe_set_ivar(adapter, rxr->me, que->msix, 0); /* ... and the TX */ ixgbe_set_ivar(adapter, txr->me, que->msix, 1); /* Set an Initial EITR value */ IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(que->msix), newitr); } /* For the Link interrupt */ ixgbe_set_ivar(adapter, 1, adapter->vector, -1); } /* ixgbe_configure_ivars */ /************************************************************************ * ixgbe_config_gpie ************************************************************************/ static void ixgbe_config_gpie(struct adapter *adapter) { struct ixgbe_hw *hw = &adapter->hw; u32 gpie; gpie = IXGBE_READ_REG(hw, IXGBE_GPIE); if (adapter->feat_en & IXGBE_FEATURE_MSIX) { /* Enable Enhanced MSI-X mode */ gpie |= IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_EIAME | IXGBE_GPIE_PBA_SUPPORT | IXGBE_GPIE_OCD; } /* Fan Failure Interrupt */ if (adapter->feat_en & IXGBE_FEATURE_FAN_FAIL) gpie |= IXGBE_SDP1_GPIEN; /* Thermal Sensor Interrupt */ if (adapter->feat_en & IXGBE_FEATURE_TEMP_SENSOR) gpie |= IXGBE_SDP0_GPIEN_X540; /* Link detection */ switch (hw->mac.type) { case ixgbe_mac_82599EB: gpie |= IXGBE_SDP1_GPIEN | IXGBE_SDP2_GPIEN; break; case ixgbe_mac_X550EM_x: case ixgbe_mac_X550EM_a: gpie |= IXGBE_SDP0_GPIEN_X540; break; default: break; } IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie); return; } /* ixgbe_config_gpie */ /************************************************************************ * ixgbe_config_delay_values * * Requires adapter->max_frame_size to be set. ************************************************************************/ static void ixgbe_config_delay_values(struct adapter *adapter) { struct ixgbe_hw *hw = &adapter->hw; u32 rxpb, frame, size, tmp; frame = adapter->max_frame_size; /* Calculate High Water */ switch (hw->mac.type) { case ixgbe_mac_X540: case ixgbe_mac_X550: case ixgbe_mac_X550EM_x: case ixgbe_mac_X550EM_a: tmp = IXGBE_DV_X540(frame, frame); break; default: tmp = IXGBE_DV(frame, frame); break; } size = IXGBE_BT2KB(tmp); rxpb = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(0)) >> 10; hw->fc.high_water[0] = rxpb - size; /* Now calculate Low Water */ switch (hw->mac.type) { case ixgbe_mac_X540: case ixgbe_mac_X550: case ixgbe_mac_X550EM_x: case ixgbe_mac_X550EM_a: tmp = IXGBE_LOW_DV_X540(frame); break; default: tmp = IXGBE_LOW_DV(frame); break; } hw->fc.low_water[0] = IXGBE_BT2KB(tmp); hw->fc.pause_time = IXGBE_FC_PAUSE; hw->fc.send_xon = TRUE; } /* ixgbe_config_delay_values */ /************************************************************************ * ixgbe_set_multi - Multicast Update * * Called whenever multicast address list is updated. ************************************************************************/ static void ixgbe_set_multi(struct adapter *adapter) { struct ifmultiaddr *ifma; struct ixgbe_mc_addr *mta; struct ifnet *ifp = adapter->ifp; u8 *update_ptr; int mcnt = 0; u32 fctrl; IOCTL_DEBUGOUT("ixgbe_set_multi: begin"); mta = adapter->mta; bzero(mta, sizeof(*mta) * MAX_NUM_MULTICAST_ADDRESSES); #if __FreeBSD_version < 800000 IF_ADDR_LOCK(ifp); #else if_maddr_rlock(ifp); #endif TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { if (ifma->ifma_addr->sa_family != AF_LINK) continue; if (mcnt == MAX_NUM_MULTICAST_ADDRESSES) break; bcopy(LLADDR((struct sockaddr_dl *) ifma->ifma_addr), mta[mcnt].addr, IXGBE_ETH_LENGTH_OF_ADDRESS); mta[mcnt].vmdq = adapter->pool; mcnt++; } #if __FreeBSD_version < 800000 IF_ADDR_UNLOCK(ifp); #else if_maddr_runlock(ifp); #endif fctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_FCTRL); fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE); if (ifp->if_flags & IFF_PROMISC) fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE); else if (mcnt >= MAX_NUM_MULTICAST_ADDRESSES || ifp->if_flags & IFF_ALLMULTI) { fctrl |= IXGBE_FCTRL_MPE; fctrl &= ~IXGBE_FCTRL_UPE; } else fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE); IXGBE_WRITE_REG(&adapter->hw, IXGBE_FCTRL, fctrl); if (mcnt < MAX_NUM_MULTICAST_ADDRESSES) { update_ptr = (u8 *)mta; ixgbe_update_mc_addr_list(&adapter->hw, update_ptr, mcnt, ixgbe_mc_array_itr, TRUE); } return; } /* ixgbe_set_multi */ /************************************************************************ * ixgbe_mc_array_itr * * An iterator function needed by the multicast shared code. * It feeds the shared code routine the addresses in the * array of ixgbe_set_multi() one by one. ************************************************************************/ static u8 * ixgbe_mc_array_itr(struct ixgbe_hw *hw, u8 **update_ptr, u32 *vmdq) { struct ixgbe_mc_addr *mta; mta = (struct ixgbe_mc_addr *)*update_ptr; *vmdq = mta->vmdq; *update_ptr = (u8*)(mta + 1); return (mta->addr); } /* ixgbe_mc_array_itr */ /************************************************************************ * ixgbe_local_timer - Timer routine * * Checks for link status, updates statistics, * and runs the watchdog check. ************************************************************************/ static void ixgbe_local_timer(void *arg) { struct adapter *adapter = arg; device_t dev = adapter->dev; struct ix_queue *que = adapter->queues; u64 queues = 0; int hung = 0; mtx_assert(&adapter->core_mtx, MA_OWNED); /* Check for pluggable optics */ if (adapter->sfp_probe) if (!ixgbe_sfp_probe(adapter)) goto out; /* Nothing to do */ ixgbe_update_link_status(adapter); ixgbe_update_stats_counters(adapter); /* * Check the TX queues status * - mark hung queues so we don't schedule on them * - watchdog only if all queues show hung */ for (int i = 0; i < adapter->num_queues; i++, que++) { /* Keep track of queues with work for soft irq */ if (que->txr->busy) queues |= ((u64)1 << que->me); /* * Each time txeof runs without cleaning, but there * are uncleaned descriptors it increments busy. If * we get to the MAX we declare it hung. */ if (que->busy == IXGBE_QUEUE_HUNG) { ++hung; /* Mark the queue as inactive */ adapter->active_queues &= ~((u64)1 << que->me); continue; } else { /* Check if we've come back from hung */ if ((adapter->active_queues & ((u64)1 << que->me)) == 0) adapter->active_queues |= ((u64)1 << que->me); } if (que->busy >= IXGBE_MAX_TX_BUSY) { device_printf(dev, "Warning queue %d appears to be hung!\n", i); que->txr->busy = IXGBE_QUEUE_HUNG; ++hung; } } /* Only truly watchdog if all queues show hung */ if (hung == adapter->num_queues) goto watchdog; else if (queues != 0) { /* Force an IRQ on queues with work */ ixgbe_rearm_queues(adapter, queues); } out: callout_reset(&adapter->timer, hz, ixgbe_local_timer, adapter); return; watchdog: device_printf(adapter->dev, "Watchdog timeout -- resetting\n"); adapter->ifp->if_drv_flags &= ~IFF_DRV_RUNNING; adapter->watchdog_events++; ixgbe_init_locked(adapter); } /* ixgbe_local_timer */ /************************************************************************ * ixgbe_sfp_probe * * Determine if a port had optics inserted. ************************************************************************/ static bool ixgbe_sfp_probe(struct adapter *adapter) { struct ixgbe_hw *hw = &adapter->hw; device_t dev = adapter->dev; bool result = FALSE; if ((hw->phy.type == ixgbe_phy_nl) && (hw->phy.sfp_type == ixgbe_sfp_type_not_present)) { s32 ret = hw->phy.ops.identify_sfp(hw); if (ret) goto out; ret = hw->phy.ops.reset(hw); adapter->sfp_probe = FALSE; if (ret == IXGBE_ERR_SFP_NOT_SUPPORTED) { device_printf(dev, "Unsupported SFP+ module detected!"); device_printf(dev, "Reload driver with supported module.\n"); goto out; } else device_printf(dev, "SFP+ module detected!\n"); /* We now have supported optics */ result = TRUE; } out: return (result); } /* ixgbe_sfp_probe */ /************************************************************************ * ixgbe_handle_mod - Tasklet for SFP module interrupts ************************************************************************/ static void ixgbe_handle_mod(void *context, int pending) { struct adapter *adapter = context; struct ixgbe_hw *hw = &adapter->hw; device_t dev = adapter->dev; u32 err, cage_full = 0; if (adapter->hw.need_crosstalk_fix) { switch (hw->mac.type) { case ixgbe_mac_82599EB: cage_full = IXGBE_READ_REG(hw, IXGBE_ESDP) & IXGBE_ESDP_SDP2; break; case ixgbe_mac_X550EM_x: case ixgbe_mac_X550EM_a: cage_full = IXGBE_READ_REG(hw, IXGBE_ESDP) & IXGBE_ESDP_SDP0; break; default: break; } if (!cage_full) return; } err = hw->phy.ops.identify_sfp(hw); if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) { device_printf(dev, "Unsupported SFP+ module type was detected.\n"); return; } err = hw->mac.ops.setup_sfp(hw); if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) { device_printf(dev, "Setup failure - unsupported SFP+ module type.\n"); return; } taskqueue_enqueue(adapter->tq, &adapter->msf_task); } /* ixgbe_handle_mod */ /************************************************************************ * ixgbe_handle_msf - Tasklet for MSF (multispeed fiber) interrupts ************************************************************************/ static void ixgbe_handle_msf(void *context, int pending) { struct adapter *adapter = context; struct ixgbe_hw *hw = &adapter->hw; u32 autoneg; bool negotiate; /* get_supported_phy_layer will call hw->phy.ops.identify_sfp() */ adapter->phy_layer = ixgbe_get_supported_physical_layer(hw); autoneg = hw->phy.autoneg_advertised; if ((!autoneg) && (hw->mac.ops.get_link_capabilities)) hw->mac.ops.get_link_capabilities(hw, &autoneg, &negotiate); if (hw->mac.ops.setup_link) hw->mac.ops.setup_link(hw, autoneg, TRUE); /* Adjust media types shown in ifconfig */ ifmedia_removeall(&adapter->media); ixgbe_add_media_types(adapter); ifmedia_set(&adapter->media, IFM_ETHER | IFM_AUTO); } /* ixgbe_handle_msf */ /************************************************************************ * ixgbe_handle_phy - Tasklet for external PHY interrupts ************************************************************************/ static void ixgbe_handle_phy(void *context, int pending) { struct adapter *adapter = context; struct ixgbe_hw *hw = &adapter->hw; int error; error = hw->phy.ops.handle_lasi(hw); if (error == IXGBE_ERR_OVERTEMP) device_printf(adapter->dev, "CRITICAL: EXTERNAL PHY OVER TEMP!! PHY will downshift to lower power state!\n"); else if (error) device_printf(adapter->dev, "Error handling LASI interrupt: %d\n", error); } /* ixgbe_handle_phy */ /************************************************************************ * ixgbe_stop - Stop the hardware * * Disables all traffic on the adapter by issuing a * global reset on the MAC and deallocates TX/RX buffers. ************************************************************************/ static void ixgbe_stop(void *arg) { struct ifnet *ifp; struct adapter *adapter = arg; struct ixgbe_hw *hw = &adapter->hw; ifp = adapter->ifp; mtx_assert(&adapter->core_mtx, MA_OWNED); INIT_DEBUGOUT("ixgbe_stop: begin\n"); ixgbe_disable_intr(adapter); callout_stop(&adapter->timer); /* Let the stack know...*/ ifp->if_drv_flags &= ~IFF_DRV_RUNNING; ixgbe_reset_hw(hw); hw->adapter_stopped = FALSE; ixgbe_stop_adapter(hw); if (hw->mac.type == ixgbe_mac_82599EB) ixgbe_stop_mac_link_on_d3_82599(hw); /* Turn off the laser - noop with no optics */ ixgbe_disable_tx_laser(hw); /* Update the stack */ adapter->link_up = FALSE; ixgbe_update_link_status(adapter); /* reprogram the RAR[0] in case user changed it. */ ixgbe_set_rar(&adapter->hw, 0, adapter->hw.mac.addr, 0, IXGBE_RAH_AV); return; } /* ixgbe_stop */ /************************************************************************ * ixgbe_update_link_status - Update OS on link state * * Note: Only updates the OS on the cached link state. * The real check of the hardware only happens with * a link interrupt. ************************************************************************/ static void ixgbe_update_link_status(struct adapter *adapter) { struct ifnet *ifp = adapter->ifp; device_t dev = adapter->dev; if (adapter->link_up) { if (adapter->link_active == FALSE) { if (bootverbose) device_printf(dev, "Link is up %d Gbps %s \n", ((adapter->link_speed == 128) ? 10 : 1), "Full Duplex"); adapter->link_active = TRUE; /* Update any Flow Control changes */ ixgbe_fc_enable(&adapter->hw); /* Update DMA coalescing config */ ixgbe_config_dmac(adapter); if_link_state_change(ifp, LINK_STATE_UP); if (adapter->feat_en & IXGBE_FEATURE_SRIOV) ixgbe_ping_all_vfs(adapter); } } else { /* Link down */ if (adapter->link_active == TRUE) { if (bootverbose) device_printf(dev, "Link is Down\n"); if_link_state_change(ifp, LINK_STATE_DOWN); adapter->link_active = FALSE; if (adapter->feat_en & IXGBE_FEATURE_SRIOV) ixgbe_ping_all_vfs(adapter); } } return; } /* ixgbe_update_link_status */ /************************************************************************ * ixgbe_config_dmac - Configure DMA Coalescing ************************************************************************/ static void ixgbe_config_dmac(struct adapter *adapter) { struct ixgbe_hw *hw = &adapter->hw; struct ixgbe_dmac_config *dcfg = &hw->mac.dmac_config; if (hw->mac.type < ixgbe_mac_X550 || !hw->mac.ops.dmac_config) return; if (dcfg->watchdog_timer ^ adapter->dmac || dcfg->link_speed ^ adapter->link_speed) { dcfg->watchdog_timer = adapter->dmac; dcfg->fcoe_en = false; dcfg->link_speed = adapter->link_speed; dcfg->num_tcs = 1; INIT_DEBUGOUT2("dmac settings: watchdog %d, link speed %d\n", dcfg->watchdog_timer, dcfg->link_speed); hw->mac.ops.dmac_config(hw); } } /* ixgbe_config_dmac */ /************************************************************************ * ixgbe_enable_intr ************************************************************************/ static void ixgbe_enable_intr(struct adapter *adapter) { struct ixgbe_hw *hw = &adapter->hw; struct ix_queue *que = adapter->queues; u32 mask, fwsm; mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE); switch (adapter->hw.mac.type) { case ixgbe_mac_82599EB: mask |= IXGBE_EIMS_ECC; /* Temperature sensor on some adapters */ mask |= IXGBE_EIMS_GPI_SDP0; /* SFP+ (RX_LOS_N & MOD_ABS_N) */ mask |= IXGBE_EIMS_GPI_SDP1; mask |= IXGBE_EIMS_GPI_SDP2; break; case ixgbe_mac_X540: /* Detect if Thermal Sensor is enabled */ fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM); if (fwsm & IXGBE_FWSM_TS_ENABLED) mask |= IXGBE_EIMS_TS; mask |= IXGBE_EIMS_ECC; break; case ixgbe_mac_X550: /* MAC thermal sensor is automatically enabled */ mask |= IXGBE_EIMS_TS; mask |= IXGBE_EIMS_ECC; break; case ixgbe_mac_X550EM_x: case ixgbe_mac_X550EM_a: /* Some devices use SDP0 for important information */ if (hw->device_id == IXGBE_DEV_ID_X550EM_X_SFP || hw->device_id == IXGBE_DEV_ID_X550EM_A_SFP || hw->device_id == IXGBE_DEV_ID_X550EM_A_SFP_N || hw->device_id == IXGBE_DEV_ID_X550EM_X_10G_T) mask |= IXGBE_EIMS_GPI_SDP0_BY_MAC(hw); if (hw->phy.type == ixgbe_phy_x550em_ext_t) mask |= IXGBE_EICR_GPI_SDP0_X540; mask |= IXGBE_EIMS_ECC; break; default: break; } /* Enable Fan Failure detection */ if (adapter->feat_en & IXGBE_FEATURE_FAN_FAIL) mask |= IXGBE_EIMS_GPI_SDP1; /* Enable SR-IOV */ if (adapter->feat_en & IXGBE_FEATURE_SRIOV) mask |= IXGBE_EIMS_MAILBOX; /* Enable Flow Director */ if (adapter->feat_en & IXGBE_FEATURE_FDIR) mask |= IXGBE_EIMS_FLOW_DIR; IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask); /* With MSI-X we use auto clear */ if (adapter->msix_mem) { mask = IXGBE_EIMS_ENABLE_MASK; /* Don't autoclear Link */ mask &= ~IXGBE_EIMS_OTHER; mask &= ~IXGBE_EIMS_LSC; if (adapter->feat_cap & IXGBE_FEATURE_SRIOV) mask &= ~IXGBE_EIMS_MAILBOX; IXGBE_WRITE_REG(hw, IXGBE_EIAC, mask); } /* * Now enable all queues, this is done separately to * allow for handling the extended (beyond 32) MSI-X * vectors that can be used by 82599 */ for (int i = 0; i < adapter->num_queues; i++, que++) ixgbe_enable_queue(adapter, que->msix); IXGBE_WRITE_FLUSH(hw); return; } /* ixgbe_enable_intr */ /************************************************************************ * ixgbe_disable_intr ************************************************************************/ static void ixgbe_disable_intr(struct adapter *adapter) { if (adapter->msix_mem) IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, 0); if (adapter->hw.mac.type == ixgbe_mac_82598EB) { IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0); } else { IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000); IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0); IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0); } IXGBE_WRITE_FLUSH(&adapter->hw); return; } /* ixgbe_disable_intr */ /************************************************************************ * ixgbe_legacy_irq - Legacy Interrupt Service routine ************************************************************************/ static void ixgbe_legacy_irq(void *arg) { struct ix_queue *que = arg; struct adapter *adapter = que->adapter; struct ixgbe_hw *hw = &adapter->hw; struct ifnet *ifp = adapter->ifp; struct tx_ring *txr = adapter->tx_rings; bool more = false; u32 eicr, eicr_mask; /* Silicon errata #26 on 82598 */ IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK); eicr = IXGBE_READ_REG(hw, IXGBE_EICR); ++que->irqs; if (eicr == 0) { ixgbe_enable_intr(adapter); return; } if (ifp->if_drv_flags & IFF_DRV_RUNNING) { more = ixgbe_rxeof(que); IXGBE_TX_LOCK(txr); ixgbe_txeof(txr); if (!ixgbe_ring_empty(ifp, txr->br)) ixgbe_start_locked(ifp, txr); IXGBE_TX_UNLOCK(txr); } /* Check for fan failure */ if (adapter->feat_en & IXGBE_FEATURE_FAN_FAIL) { ixgbe_check_fan_failure(adapter, eicr, true); IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EICR_GPI_SDP1_BY_MAC(hw)); } /* Link status change */ if (eicr & IXGBE_EICR_LSC) taskqueue_enqueue(adapter->tq, &adapter->link_task); if (ixgbe_is_sfp(hw)) { /* Pluggable optics-related interrupt */ if (hw->mac.type >= ixgbe_mac_X540) eicr_mask = IXGBE_EICR_GPI_SDP0_X540; else eicr_mask = IXGBE_EICR_GPI_SDP2_BY_MAC(hw); if (eicr & eicr_mask) { IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr_mask); taskqueue_enqueue(adapter->tq, &adapter->mod_task); } if ((hw->mac.type == ixgbe_mac_82599EB) && (eicr & IXGBE_EICR_GPI_SDP1_BY_MAC(hw))) { IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1_BY_MAC(hw)); taskqueue_enqueue(adapter->tq, &adapter->msf_task); } } /* External PHY interrupt */ if ((hw->phy.type == ixgbe_phy_x550em_ext_t) && (eicr & IXGBE_EICR_GPI_SDP0_X540)) taskqueue_enqueue(adapter->tq, &adapter->phy_task); if (more) taskqueue_enqueue(que->tq, &que->que_task); else ixgbe_enable_intr(adapter); return; } /* ixgbe_legacy_irq */ /************************************************************************ * ixgbe_free_pci_resources ************************************************************************/ static void ixgbe_free_pci_resources(struct adapter *adapter) { struct ix_queue *que = adapter->queues; device_t dev = adapter->dev; int rid, memrid; if (adapter->hw.mac.type == ixgbe_mac_82598EB) memrid = PCIR_BAR(MSIX_82598_BAR); else memrid = PCIR_BAR(MSIX_82599_BAR); /* * There is a slight possibility of a failure mode * in attach that will result in entering this function * before interrupt resources have been initialized, and * in that case we do not want to execute the loops below * We can detect this reliably by the state of the adapter * res pointer. */ if (adapter->res == NULL) goto mem; /* * Release all msix queue resources: */ for (int i = 0; i < adapter->num_queues; i++, que++) { rid = que->msix + 1; if (que->tag != NULL) { bus_teardown_intr(dev, que->res, que->tag); que->tag = NULL; } if (que->res != NULL) bus_release_resource(dev, SYS_RES_IRQ, rid, que->res); } if (adapter->tag != NULL) { bus_teardown_intr(dev, adapter->res, adapter->tag); adapter->tag = NULL; } /* Clean the Legacy or Link interrupt last */ if (adapter->res != NULL) bus_release_resource(dev, SYS_RES_IRQ, adapter->link_rid, adapter->res); mem: if ((adapter->feat_en & IXGBE_FEATURE_MSI) || (adapter->feat_en & IXGBE_FEATURE_MSIX)) pci_release_msi(dev); if (adapter->msix_mem != NULL) bus_release_resource(dev, SYS_RES_MEMORY, memrid, adapter->msix_mem); if (adapter->pci_mem != NULL) bus_release_resource(dev, SYS_RES_MEMORY, PCIR_BAR(0), adapter->pci_mem); return; } /* ixgbe_free_pci_resources */ /************************************************************************ * ixgbe_set_sysctl_value ************************************************************************/ static void ixgbe_set_sysctl_value(struct adapter *adapter, const char *name, const char *description, int *limit, int value) { *limit = value; SYSCTL_ADD_INT(device_get_sysctl_ctx(adapter->dev), SYSCTL_CHILDREN(device_get_sysctl_tree(adapter->dev)), OID_AUTO, name, CTLFLAG_RW, limit, value, description); } /* ixgbe_set_sysctl_value */ /************************************************************************ * ixgbe_sysctl_flowcntl * * SYSCTL wrapper around setting Flow Control ************************************************************************/ static int ixgbe_sysctl_flowcntl(SYSCTL_HANDLER_ARGS) { struct adapter *adapter; int error, fc; adapter = (struct adapter *)arg1; fc = adapter->hw.fc.current_mode; error = sysctl_handle_int(oidp, &fc, 0, req); if ((error) || (req->newptr == NULL)) return (error); /* Don't bother if it's not changed */ if (fc == adapter->hw.fc.current_mode) return (0); return ixgbe_set_flowcntl(adapter, fc); } /* ixgbe_sysctl_flowcntl */ /************************************************************************ * ixgbe_set_flowcntl - Set flow control * * Flow control values: * 0 - off * 1 - rx pause * 2 - tx pause * 3 - full ************************************************************************/ static int ixgbe_set_flowcntl(struct adapter *adapter, int fc) { switch (fc) { case ixgbe_fc_rx_pause: case ixgbe_fc_tx_pause: case ixgbe_fc_full: adapter->hw.fc.requested_mode = fc; if (adapter->num_queues > 1) ixgbe_disable_rx_drop(adapter); break; case ixgbe_fc_none: adapter->hw.fc.requested_mode = ixgbe_fc_none; if (adapter->num_queues > 1) ixgbe_enable_rx_drop(adapter); break; default: return (EINVAL); } /* Don't autoneg if forcing a value */ adapter->hw.fc.disable_fc_autoneg = TRUE; ixgbe_fc_enable(&adapter->hw); return (0); } /* ixgbe_set_flowcntl */ /************************************************************************ * ixgbe_enable_rx_drop * * Enable the hardware to drop packets when the buffer is * full. This is useful with multiqueue, so that no single * queue being full stalls the entire RX engine. We only * enable this when Multiqueue is enabled AND Flow Control * is disabled. ************************************************************************/ static void ixgbe_enable_rx_drop(struct adapter *adapter) { struct ixgbe_hw *hw = &adapter->hw; struct rx_ring *rxr; u32 srrctl; for (int i = 0; i < adapter->num_queues; i++) { rxr = &adapter->rx_rings[i]; srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(rxr->me)); srrctl |= IXGBE_SRRCTL_DROP_EN; IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(rxr->me), srrctl); } /* enable drop for each vf */ for (int i = 0; i < adapter->num_vfs; i++) { IXGBE_WRITE_REG(hw, IXGBE_QDE, (IXGBE_QDE_WRITE | (i << IXGBE_QDE_IDX_SHIFT) | IXGBE_QDE_ENABLE)); } } /* ixgbe_enable_rx_drop */ /************************************************************************ * ixgbe_disable_rx_drop ************************************************************************/ static void ixgbe_disable_rx_drop(struct adapter *adapter) { struct ixgbe_hw *hw = &adapter->hw; struct rx_ring *rxr; u32 srrctl; for (int i = 0; i < adapter->num_queues; i++) { rxr = &adapter->rx_rings[i]; srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(rxr->me)); srrctl &= ~IXGBE_SRRCTL_DROP_EN; IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(rxr->me), srrctl); } /* disable drop for each vf */ for (int i = 0; i < adapter->num_vfs; i++) { IXGBE_WRITE_REG(hw, IXGBE_QDE, (IXGBE_QDE_WRITE | (i << IXGBE_QDE_IDX_SHIFT))); } } /* ixgbe_disable_rx_drop */ /************************************************************************ * ixgbe_sysctl_advertise * * SYSCTL wrapper around setting advertised speed ************************************************************************/ static int ixgbe_sysctl_advertise(SYSCTL_HANDLER_ARGS) { struct adapter *adapter; int error, advertise; adapter = (struct adapter *)arg1; advertise = adapter->advertise; error = sysctl_handle_int(oidp, &advertise, 0, req); if ((error) || (req->newptr == NULL)) return (error); return ixgbe_set_advertise(adapter, advertise); } /* ixgbe_sysctl_advertise */ /************************************************************************ * ixgbe_set_advertise - Control advertised link speed * * Flags: * 0x1 - advertise 100 Mb * 0x2 - advertise 1G * 0x4 - advertise 10G * 0x8 - advertise 10 Mb (yes, Mb) ************************************************************************/ static int ixgbe_set_advertise(struct adapter *adapter, int advertise) { device_t dev; struct ixgbe_hw *hw; ixgbe_link_speed speed = 0; ixgbe_link_speed link_caps = 0; s32 err = IXGBE_NOT_IMPLEMENTED; bool negotiate = FALSE; /* Checks to validate new value */ if (adapter->advertise == advertise) /* no change */ return (0); dev = adapter->dev; hw = &adapter->hw; /* No speed changes for backplane media */ if (hw->phy.media_type == ixgbe_media_type_backplane) return (ENODEV); if (!((hw->phy.media_type == ixgbe_media_type_copper) || (hw->phy.multispeed_fiber))) { device_printf(dev, "Advertised speed can only be set on copper or multispeed fiber media types.\n"); return (EINVAL); } if (advertise < 0x1 || advertise > 0xF) { device_printf(dev, "Invalid advertised speed; valid modes are 0x1 through 0xF\n"); return (EINVAL); } if (hw->mac.ops.get_link_capabilities) { err = hw->mac.ops.get_link_capabilities(hw, &link_caps, &negotiate); if (err != IXGBE_SUCCESS) { device_printf(dev, "Unable to determine supported advertise speeds\n"); return (ENODEV); } } /* Set new value and report new advertised mode */ if (advertise & 0x1) { if (!(link_caps & IXGBE_LINK_SPEED_100_FULL)) { device_printf(dev, "Interface does not support 100Mb advertised speed\n"); return (EINVAL); } speed |= IXGBE_LINK_SPEED_100_FULL; } if (advertise & 0x2) { if (!(link_caps & IXGBE_LINK_SPEED_1GB_FULL)) { device_printf(dev, "Interface does not support 1Gb advertised speed\n"); return (EINVAL); } speed |= IXGBE_LINK_SPEED_1GB_FULL; } if (advertise & 0x4) { if (!(link_caps & IXGBE_LINK_SPEED_10GB_FULL)) { device_printf(dev, "Interface does not support 10Gb advertised speed\n"); return (EINVAL); } speed |= IXGBE_LINK_SPEED_10GB_FULL; } if (advertise & 0x8) { if (!(link_caps & IXGBE_LINK_SPEED_10_FULL)) { device_printf(dev, "Interface does not support 10Mb advertised speed\n"); return (EINVAL); } speed |= IXGBE_LINK_SPEED_10_FULL; } hw->mac.autotry_restart = TRUE; hw->mac.ops.setup_link(hw, speed, TRUE); adapter->advertise = advertise; return (0); } /* ixgbe_set_advertise */ /************************************************************************ * ixgbe_get_advertise - Get current advertised speed settings * * Formatted for sysctl usage. * Flags: * 0x1 - advertise 100 Mb * 0x2 - advertise 1G * 0x4 - advertise 10G * 0x8 - advertise 10 Mb (yes, Mb) ************************************************************************/ static int ixgbe_get_advertise(struct adapter *adapter) { struct ixgbe_hw *hw = &adapter->hw; int speed; ixgbe_link_speed link_caps = 0; s32 err; bool negotiate = FALSE; /* * Advertised speed means nothing unless it's copper or * multi-speed fiber */ if (!(hw->phy.media_type == ixgbe_media_type_copper) && !(hw->phy.multispeed_fiber)) return (0); err = hw->mac.ops.get_link_capabilities(hw, &link_caps, &negotiate); if (err != IXGBE_SUCCESS) return (0); speed = ((link_caps & IXGBE_LINK_SPEED_10GB_FULL) ? 4 : 0) | ((link_caps & IXGBE_LINK_SPEED_1GB_FULL) ? 2 : 0) | ((link_caps & IXGBE_LINK_SPEED_100_FULL) ? 1 : 0) | ((link_caps & IXGBE_LINK_SPEED_10_FULL) ? 8 : 0); return speed; } /* ixgbe_get_advertise */ /************************************************************************ * ixgbe_sysctl_dmac - Manage DMA Coalescing * * Control values: * 0/1 - off / on (use default value of 1000) * * Legal timer values are: * 50,100,250,500,1000,2000,5000,10000 * * Turning off interrupt moderation will also turn this off. ************************************************************************/ static int ixgbe_sysctl_dmac(SYSCTL_HANDLER_ARGS) { struct adapter *adapter = (struct adapter *)arg1; struct ifnet *ifp = adapter->ifp; int error; u32 newval; newval = adapter->dmac; error = sysctl_handle_int(oidp, &newval, 0, req); if ((error) || (req->newptr == NULL)) return (error); switch (newval) { case 0: /* Disabled */ adapter->dmac = 0; break; case 1: /* Enable and use default */ adapter->dmac = 1000; break; case 50: case 100: case 250: case 500: case 1000: case 2000: case 5000: case 10000: /* Legal values - allow */ adapter->dmac = newval; break; default: /* Do nothing, illegal value */ return (EINVAL); } /* Re-initialize hardware if it's already running */ if (ifp->if_drv_flags & IFF_DRV_RUNNING) ixgbe_init(adapter); return (0); } /* ixgbe_sysctl_dmac */ #ifdef IXGBE_DEBUG /************************************************************************ * ixgbe_sysctl_power_state * * Sysctl to test power states * Values: * 0 - set device to D0 * 3 - set device to D3 * (none) - get current device power state ************************************************************************/ static int ixgbe_sysctl_power_state(SYSCTL_HANDLER_ARGS) { struct adapter *adapter = (struct adapter *)arg1; device_t dev = adapter->dev; int curr_ps, new_ps, error = 0; curr_ps = new_ps = pci_get_powerstate(dev); error = sysctl_handle_int(oidp, &new_ps, 0, req); if ((error) || (req->newptr == NULL)) return (error); if (new_ps == curr_ps) return (0); if (new_ps == 3 && curr_ps == 0) error = DEVICE_SUSPEND(dev); else if (new_ps == 0 && curr_ps == 3) error = DEVICE_RESUME(dev); else return (EINVAL); device_printf(dev, "New state: %d\n", pci_get_powerstate(dev)); return (error); } /* ixgbe_sysctl_power_state */ #endif /************************************************************************ * ixgbe_sysctl_wol_enable * * Sysctl to enable/disable the WoL capability, * if supported by the adapter. * * Values: * 0 - disabled * 1 - enabled ************************************************************************/ static int ixgbe_sysctl_wol_enable(SYSCTL_HANDLER_ARGS) { struct adapter *adapter = (struct adapter *)arg1; struct ixgbe_hw *hw = &adapter->hw; int new_wol_enabled; int error = 0; new_wol_enabled = hw->wol_enabled; error = sysctl_handle_int(oidp, &new_wol_enabled, 0, req); if ((error) || (req->newptr == NULL)) return (error); new_wol_enabled = !!(new_wol_enabled); if (new_wol_enabled == hw->wol_enabled) return (0); if (new_wol_enabled > 0 && !adapter->wol_support) return (ENODEV); else hw->wol_enabled = new_wol_enabled; return (0); } /* ixgbe_sysctl_wol_enable */ /************************************************************************ * ixgbe_sysctl_wufc - Wake Up Filter Control * * Sysctl to enable/disable the types of packets that the * adapter will wake up on upon receipt. * Flags: * 0x1 - Link Status Change * 0x2 - Magic Packet * 0x4 - Direct Exact * 0x8 - Directed Multicast * 0x10 - Broadcast * 0x20 - ARP/IPv4 Request Packet * 0x40 - Direct IPv4 Packet * 0x80 - Direct IPv6 Packet * * Settings not listed above will cause the sysctl to return an error. ************************************************************************/ static int ixgbe_sysctl_wufc(SYSCTL_HANDLER_ARGS) { struct adapter *adapter = (struct adapter *)arg1; int error = 0; u32 new_wufc; new_wufc = adapter->wufc; error = sysctl_handle_int(oidp, &new_wufc, 0, req); if ((error) || (req->newptr == NULL)) return (error); if (new_wufc == adapter->wufc) return (0); if (new_wufc & 0xffffff00) return (EINVAL); new_wufc &= 0xff; new_wufc |= (0xffffff & adapter->wufc); adapter->wufc = new_wufc; return (0); } /* ixgbe_sysctl_wufc */ #ifdef IXGBE_DEBUG /************************************************************************ * ixgbe_sysctl_print_rss_config ************************************************************************/ static int ixgbe_sysctl_print_rss_config(SYSCTL_HANDLER_ARGS) { struct adapter *adapter = (struct adapter *)arg1; struct ixgbe_hw *hw = &adapter->hw; device_t dev = adapter->dev; struct sbuf *buf; int error = 0, reta_size; u32 reg; buf = sbuf_new_for_sysctl(NULL, NULL, 128, req); if (!buf) { device_printf(dev, "Could not allocate sbuf for output.\n"); return (ENOMEM); } // TODO: use sbufs to make a string to print out /* Set multiplier for RETA setup and table size based on MAC */ switch (adapter->hw.mac.type) { case ixgbe_mac_X550: case ixgbe_mac_X550EM_x: case ixgbe_mac_X550EM_a: reta_size = 128; break; default: reta_size = 32; break; } /* Print out the redirection table */ sbuf_cat(buf, "\n"); for (int i = 0; i < reta_size; i++) { if (i < 32) { reg = IXGBE_READ_REG(hw, IXGBE_RETA(i)); sbuf_printf(buf, "RETA(%2d): 0x%08x\n", i, reg); } else { reg = IXGBE_READ_REG(hw, IXGBE_ERETA(i - 32)); sbuf_printf(buf, "ERETA(%2d): 0x%08x\n", i - 32, reg); } } // TODO: print more config error = sbuf_finish(buf); if (error) device_printf(dev, "Error finishing sbuf: %d\n", error); sbuf_delete(buf); return (0); } /* ixgbe_sysctl_print_rss_config */ #endif /* IXGBE_DEBUG */ /************************************************************************ * ixgbe_sysctl_phy_temp - Retrieve temperature of PHY * * For X552/X557-AT devices using an external PHY ************************************************************************/ static int ixgbe_sysctl_phy_temp(SYSCTL_HANDLER_ARGS) { struct adapter *adapter = (struct adapter *)arg1; struct ixgbe_hw *hw = &adapter->hw; u16 reg; if (hw->device_id != IXGBE_DEV_ID_X550EM_X_10G_T) { device_printf(adapter->dev, "Device has no supported external thermal sensor.\n"); return (ENODEV); } if (hw->phy.ops.read_reg(hw, IXGBE_PHY_CURRENT_TEMP, IXGBE_MDIO_VENDOR_SPECIFIC_1_DEV_TYPE, ®)) { device_printf(adapter->dev, "Error reading from PHY's current temperature register\n"); return (EAGAIN); } /* Shift temp for output */ reg = reg >> 8; return (sysctl_handle_int(oidp, NULL, reg, req)); } /* ixgbe_sysctl_phy_temp */ /************************************************************************ * ixgbe_sysctl_phy_overtemp_occurred * * Reports (directly from the PHY) whether the current PHY * temperature is over the overtemp threshold. ************************************************************************/ static int ixgbe_sysctl_phy_overtemp_occurred(SYSCTL_HANDLER_ARGS) { struct adapter *adapter = (struct adapter *)arg1; struct ixgbe_hw *hw = &adapter->hw; u16 reg; if (hw->device_id != IXGBE_DEV_ID_X550EM_X_10G_T) { device_printf(adapter->dev, "Device has no supported external thermal sensor.\n"); return (ENODEV); } if (hw->phy.ops.read_reg(hw, IXGBE_PHY_OVERTEMP_STATUS, IXGBE_MDIO_VENDOR_SPECIFIC_1_DEV_TYPE, ®)) { device_printf(adapter->dev, "Error reading from PHY's temperature status register\n"); return (EAGAIN); } /* Get occurrence bit */ reg = !!(reg & 0x4000); return (sysctl_handle_int(oidp, 0, reg, req)); } /* ixgbe_sysctl_phy_overtemp_occurred */ /************************************************************************ * ixgbe_sysctl_eee_state * * Sysctl to set EEE power saving feature * Values: * 0 - disable EEE * 1 - enable EEE * (none) - get current device EEE state ************************************************************************/ static int ixgbe_sysctl_eee_state(SYSCTL_HANDLER_ARGS) { struct adapter *adapter = (struct adapter *)arg1; device_t dev = adapter->dev; int curr_eee, new_eee, error = 0; s32 retval; curr_eee = new_eee = !!(adapter->feat_en & IXGBE_FEATURE_EEE); error = sysctl_handle_int(oidp, &new_eee, 0, req); if ((error) || (req->newptr == NULL)) return (error); /* Nothing to do */ if (new_eee == curr_eee) return (0); /* Not supported */ if (!(adapter->feat_cap & IXGBE_FEATURE_EEE)) return (EINVAL); /* Bounds checking */ if ((new_eee < 0) || (new_eee > 1)) return (EINVAL); retval = adapter->hw.mac.ops.setup_eee(&adapter->hw, new_eee); if (retval) { device_printf(dev, "Error in EEE setup: 0x%08X\n", retval); return (EINVAL); } /* Restart auto-neg */ ixgbe_init(adapter); device_printf(dev, "New EEE state: %d\n", new_eee); /* Cache new value */ if (new_eee) adapter->feat_en |= IXGBE_FEATURE_EEE; else adapter->feat_en &= ~IXGBE_FEATURE_EEE; return (error); } /* ixgbe_sysctl_eee_state */ /************************************************************************ * ixgbe_init_device_features ************************************************************************/ static void ixgbe_init_device_features(struct adapter *adapter) { adapter->feat_cap = IXGBE_FEATURE_NETMAP | IXGBE_FEATURE_RSS | IXGBE_FEATURE_MSI | IXGBE_FEATURE_MSIX | IXGBE_FEATURE_LEGACY_IRQ | IXGBE_FEATURE_LEGACY_TX; /* Set capabilities first... */ switch (adapter->hw.mac.type) { case ixgbe_mac_82598EB: if (adapter->hw.device_id == IXGBE_DEV_ID_82598AT) adapter->feat_cap |= IXGBE_FEATURE_FAN_FAIL; break; case ixgbe_mac_X540: adapter->feat_cap |= IXGBE_FEATURE_SRIOV; adapter->feat_cap |= IXGBE_FEATURE_FDIR; if ((adapter->hw.device_id == IXGBE_DEV_ID_X540_BYPASS) && (adapter->hw.bus.func == 0)) adapter->feat_cap |= IXGBE_FEATURE_BYPASS; break; case ixgbe_mac_X550: adapter->feat_cap |= IXGBE_FEATURE_TEMP_SENSOR; adapter->feat_cap |= IXGBE_FEATURE_SRIOV; adapter->feat_cap |= IXGBE_FEATURE_FDIR; break; case ixgbe_mac_X550EM_x: adapter->feat_cap |= IXGBE_FEATURE_SRIOV; adapter->feat_cap |= IXGBE_FEATURE_FDIR; if (adapter->hw.device_id == IXGBE_DEV_ID_X550EM_X_KR) adapter->feat_cap |= IXGBE_FEATURE_EEE; break; case ixgbe_mac_X550EM_a: adapter->feat_cap |= IXGBE_FEATURE_SRIOV; adapter->feat_cap |= IXGBE_FEATURE_FDIR; adapter->feat_cap &= ~IXGBE_FEATURE_LEGACY_IRQ; if ((adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_1G_T) || (adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_1G_T_L)) { adapter->feat_cap |= IXGBE_FEATURE_TEMP_SENSOR; adapter->feat_cap |= IXGBE_FEATURE_EEE; } break; case ixgbe_mac_82599EB: adapter->feat_cap |= IXGBE_FEATURE_SRIOV; adapter->feat_cap |= IXGBE_FEATURE_FDIR; if ((adapter->hw.device_id == IXGBE_DEV_ID_82599_BYPASS) && (adapter->hw.bus.func == 0)) adapter->feat_cap |= IXGBE_FEATURE_BYPASS; if (adapter->hw.device_id == IXGBE_DEV_ID_82599_QSFP_SF_QP) adapter->feat_cap &= ~IXGBE_FEATURE_LEGACY_IRQ; break; default: break; } /* Enabled by default... */ /* Fan failure detection */ if (adapter->feat_cap & IXGBE_FEATURE_FAN_FAIL) adapter->feat_en |= IXGBE_FEATURE_FAN_FAIL; /* Netmap */ if (adapter->feat_cap & IXGBE_FEATURE_NETMAP) adapter->feat_en |= IXGBE_FEATURE_NETMAP; /* EEE */ if (adapter->feat_cap & IXGBE_FEATURE_EEE) adapter->feat_en |= IXGBE_FEATURE_EEE; /* Thermal Sensor */ if (adapter->feat_cap & IXGBE_FEATURE_TEMP_SENSOR) adapter->feat_en |= IXGBE_FEATURE_TEMP_SENSOR; /* Enabled via global sysctl... */ /* Flow Director */ if (ixgbe_enable_fdir) { if (adapter->feat_cap & IXGBE_FEATURE_FDIR) adapter->feat_en |= IXGBE_FEATURE_FDIR; else device_printf(adapter->dev, "Device does not support Flow Director. Leaving disabled."); } /* Legacy (single queue) transmit */ if ((adapter->feat_cap & IXGBE_FEATURE_LEGACY_TX) && ixgbe_enable_legacy_tx) adapter->feat_en |= IXGBE_FEATURE_LEGACY_TX; /* * Message Signal Interrupts - Extended (MSI-X) * Normal MSI is only enabled if MSI-X calls fail. */ if (!ixgbe_enable_msix) adapter->feat_cap &= ~IXGBE_FEATURE_MSIX; /* Receive-Side Scaling (RSS) */ if ((adapter->feat_cap & IXGBE_FEATURE_RSS) && ixgbe_enable_rss) adapter->feat_en |= IXGBE_FEATURE_RSS; /* Disable features with unmet dependencies... */ /* No MSI-X */ if (!(adapter->feat_cap & IXGBE_FEATURE_MSIX)) { adapter->feat_cap &= ~IXGBE_FEATURE_RSS; adapter->feat_cap &= ~IXGBE_FEATURE_SRIOV; adapter->feat_en &= ~IXGBE_FEATURE_RSS; adapter->feat_en &= ~IXGBE_FEATURE_SRIOV; } } /* ixgbe_init_device_features */ /************************************************************************ * ixgbe_probe - Device identification routine * * Determines if the driver should be loaded on * adapter based on its PCI vendor/device ID. * * return BUS_PROBE_DEFAULT on success, positive on failure ************************************************************************/ static int ixgbe_probe(device_t dev) { ixgbe_vendor_info_t *ent; u16 pci_vendor_id = 0; u16 pci_device_id = 0; u16 pci_subvendor_id = 0; u16 pci_subdevice_id = 0; char adapter_name[256]; INIT_DEBUGOUT("ixgbe_probe: begin"); pci_vendor_id = pci_get_vendor(dev); if (pci_vendor_id != IXGBE_INTEL_VENDOR_ID) return (ENXIO); pci_device_id = pci_get_device(dev); pci_subvendor_id = pci_get_subvendor(dev); pci_subdevice_id = pci_get_subdevice(dev); ent = ixgbe_vendor_info_array; while (ent->vendor_id != 0) { if ((pci_vendor_id == ent->vendor_id) && (pci_device_id == ent->device_id) && ((pci_subvendor_id == ent->subvendor_id) || (ent->subvendor_id == 0)) && ((pci_subdevice_id == ent->subdevice_id) || (ent->subdevice_id == 0))) { sprintf(adapter_name, "%s, Version - %s", ixgbe_strings[ent->index], ixgbe_driver_version); device_set_desc_copy(dev, adapter_name); ++ixgbe_total_ports; return (BUS_PROBE_DEFAULT); } ent++; } return (ENXIO); } /* ixgbe_probe */ /************************************************************************ * ixgbe_ioctl - Ioctl entry point * * Called when the user wants to configure the interface. * * return 0 on success, positive on failure ************************************************************************/ static int ixgbe_ioctl(struct ifnet *ifp, u_long command, caddr_t data) { struct adapter *adapter = ifp->if_softc; struct ifreq *ifr = (struct ifreq *) data; #if defined(INET) || defined(INET6) struct ifaddr *ifa = (struct ifaddr *)data; #endif int error = 0; bool avoid_reset = FALSE; switch (command) { case SIOCSIFADDR: #ifdef INET if (ifa->ifa_addr->sa_family == AF_INET) avoid_reset = TRUE; #endif #ifdef INET6 if (ifa->ifa_addr->sa_family == AF_INET6) avoid_reset = TRUE; #endif /* * Calling init results in link renegotiation, * so we avoid doing it when possible. */ if (avoid_reset) { ifp->if_flags |= IFF_UP; if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) ixgbe_init(adapter); #ifdef INET if (!(ifp->if_flags & IFF_NOARP)) arp_ifinit(ifp, ifa); #endif } else error = ether_ioctl(ifp, command, data); break; case SIOCSIFMTU: IOCTL_DEBUGOUT("ioctl: SIOCSIFMTU (Set Interface MTU)"); if (ifr->ifr_mtu > IXGBE_MAX_MTU) { error = EINVAL; } else { IXGBE_CORE_LOCK(adapter); ifp->if_mtu = ifr->ifr_mtu; adapter->max_frame_size = ifp->if_mtu + IXGBE_MTU_HDR; if (ifp->if_drv_flags & IFF_DRV_RUNNING) ixgbe_init_locked(adapter); ixgbe_recalculate_max_frame(adapter); IXGBE_CORE_UNLOCK(adapter); } break; case SIOCSIFFLAGS: IOCTL_DEBUGOUT("ioctl: SIOCSIFFLAGS (Set Interface Flags)"); IXGBE_CORE_LOCK(adapter); if (ifp->if_flags & IFF_UP) { if ((ifp->if_drv_flags & IFF_DRV_RUNNING)) { if ((ifp->if_flags ^ adapter->if_flags) & (IFF_PROMISC | IFF_ALLMULTI)) { ixgbe_set_promisc(adapter); } } else ixgbe_init_locked(adapter); } else if (ifp->if_drv_flags & IFF_DRV_RUNNING) ixgbe_stop(adapter); adapter->if_flags = ifp->if_flags; IXGBE_CORE_UNLOCK(adapter); break; case SIOCADDMULTI: case SIOCDELMULTI: IOCTL_DEBUGOUT("ioctl: SIOC(ADD|DEL)MULTI"); if (ifp->if_drv_flags & IFF_DRV_RUNNING) { IXGBE_CORE_LOCK(adapter); ixgbe_disable_intr(adapter); ixgbe_set_multi(adapter); ixgbe_enable_intr(adapter); IXGBE_CORE_UNLOCK(adapter); } break; case SIOCSIFMEDIA: case SIOCGIFMEDIA: IOCTL_DEBUGOUT("ioctl: SIOCxIFMEDIA (Get/Set Interface Media)"); error = ifmedia_ioctl(ifp, ifr, &adapter->media, command); break; case SIOCSIFCAP: { IOCTL_DEBUGOUT("ioctl: SIOCSIFCAP (Set Capabilities)"); int mask = ifr->ifr_reqcap ^ ifp->if_capenable; if (!mask) break; /* HW cannot turn these on/off separately */ if (mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) { ifp->if_capenable ^= IFCAP_RXCSUM; ifp->if_capenable ^= IFCAP_RXCSUM_IPV6; } if (mask & IFCAP_TXCSUM) ifp->if_capenable ^= IFCAP_TXCSUM; if (mask & IFCAP_TXCSUM_IPV6) ifp->if_capenable ^= IFCAP_TXCSUM_IPV6; if (mask & IFCAP_TSO4) ifp->if_capenable ^= IFCAP_TSO4; if (mask & IFCAP_TSO6) ifp->if_capenable ^= IFCAP_TSO6; if (mask & IFCAP_LRO) ifp->if_capenable ^= IFCAP_LRO; if (mask & IFCAP_VLAN_HWTAGGING) ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING; if (mask & IFCAP_VLAN_HWFILTER) ifp->if_capenable ^= IFCAP_VLAN_HWFILTER; if (mask & IFCAP_VLAN_HWTSO) ifp->if_capenable ^= IFCAP_VLAN_HWTSO; if (ifp->if_drv_flags & IFF_DRV_RUNNING) { IXGBE_CORE_LOCK(adapter); ixgbe_init_locked(adapter); IXGBE_CORE_UNLOCK(adapter); } VLAN_CAPABILITIES(ifp); break; } #if __FreeBSD_version >= 1100036 case SIOCGI2C: { struct ixgbe_hw *hw = &adapter->hw; struct ifi2creq i2c; int i; IOCTL_DEBUGOUT("ioctl: SIOCGI2C (Get I2C Data)"); error = copyin(ifr->ifr_data, &i2c, sizeof(i2c)); if (error != 0) break; if (i2c.dev_addr != 0xA0 && i2c.dev_addr != 0xA2) { error = EINVAL; break; } if (i2c.len > sizeof(i2c.data)) { error = EINVAL; break; } for (i = 0; i < i2c.len; i++) hw->phy.ops.read_i2c_byte(hw, i2c.offset + i, i2c.dev_addr, &i2c.data[i]); error = copyout(&i2c, ifr->ifr_data, sizeof(i2c)); break; } #endif default: IOCTL_DEBUGOUT1("ioctl: UNKNOWN (0x%X)\n", (int)command); error = ether_ioctl(ifp, command, data); break; } return (error); } /* ixgbe_ioctl */ /************************************************************************ * ixgbe_check_fan_failure ************************************************************************/ static void ixgbe_check_fan_failure(struct adapter *adapter, u32 reg, bool in_interrupt) { u32 mask; mask = (in_interrupt) ? IXGBE_EICR_GPI_SDP1_BY_MAC(&adapter->hw) : IXGBE_ESDP_SDP1; if (reg & mask) device_printf(adapter->dev, "\nCRITICAL: FAN FAILURE!! REPLACE IMMEDIATELY!!\n"); } /* ixgbe_check_fan_failure */ /************************************************************************ * ixgbe_handle_que ************************************************************************/ static void ixgbe_handle_que(void *context, int pending) { struct ix_queue *que = context; struct adapter *adapter = que->adapter; struct tx_ring *txr = que->txr; struct ifnet *ifp = adapter->ifp; if (ifp->if_drv_flags & IFF_DRV_RUNNING) { ixgbe_rxeof(que); IXGBE_TX_LOCK(txr); ixgbe_txeof(txr); if (!ixgbe_ring_empty(ifp, txr->br)) ixgbe_start_locked(ifp, txr); IXGBE_TX_UNLOCK(txr); } /* Re-enable this interrupt */ if (que->res != NULL) ixgbe_enable_queue(adapter, que->msix); else ixgbe_enable_intr(adapter); return; } /* ixgbe_handle_que */ /************************************************************************ * ixgbe_allocate_legacy - Setup the Legacy or MSI Interrupt handler ************************************************************************/ static int ixgbe_allocate_legacy(struct adapter *adapter) { device_t dev = adapter->dev; struct ix_queue *que = adapter->queues; struct tx_ring *txr = adapter->tx_rings; int error; /* We allocate a single interrupt resource */ adapter->res = bus_alloc_resource_any(dev, SYS_RES_IRQ, &adapter->link_rid, RF_SHAREABLE | RF_ACTIVE); if (adapter->res == NULL) { device_printf(dev, "Unable to allocate bus resource: interrupt\n"); return (ENXIO); } /* * Try allocating a fast interrupt and the associated deferred * processing contexts. */ if (!(adapter->feat_en & IXGBE_FEATURE_LEGACY_TX)) TASK_INIT(&txr->txq_task, 0, ixgbe_deferred_mq_start, txr); TASK_INIT(&que->que_task, 0, ixgbe_handle_que, que); que->tq = taskqueue_create_fast("ixgbe_que", M_NOWAIT, taskqueue_thread_enqueue, &que->tq); taskqueue_start_threads(&que->tq, 1, PI_NET, "%s ixq", device_get_nameunit(adapter->dev)); /* Tasklets for Link, SFP and Multispeed Fiber */ TASK_INIT(&adapter->link_task, 0, ixgbe_handle_link, adapter); TASK_INIT(&adapter->mod_task, 0, ixgbe_handle_mod, adapter); TASK_INIT(&adapter->msf_task, 0, ixgbe_handle_msf, adapter); TASK_INIT(&adapter->phy_task, 0, ixgbe_handle_phy, adapter); if (adapter->feat_en & IXGBE_FEATURE_FDIR) TASK_INIT(&adapter->fdir_task, 0, ixgbe_reinit_fdir, adapter); adapter->tq = taskqueue_create_fast("ixgbe_link", M_NOWAIT, taskqueue_thread_enqueue, &adapter->tq); taskqueue_start_threads(&adapter->tq, 1, PI_NET, "%s linkq", device_get_nameunit(adapter->dev)); if ((error = bus_setup_intr(dev, adapter->res, INTR_TYPE_NET | INTR_MPSAFE, NULL, ixgbe_legacy_irq, que, &adapter->tag)) != 0) { device_printf(dev, "Failed to register fast interrupt handler: %d\n", error); taskqueue_free(que->tq); taskqueue_free(adapter->tq); que->tq = NULL; adapter->tq = NULL; return (error); } /* For simplicity in the handlers */ adapter->active_queues = IXGBE_EIMS_ENABLE_MASK; return (0); } /* ixgbe_allocate_legacy */ /************************************************************************ * ixgbe_allocate_msix - Setup MSI-X Interrupt resources and handlers ************************************************************************/ static int ixgbe_allocate_msix(struct adapter *adapter) { device_t dev = adapter->dev; struct ix_queue *que = adapter->queues; struct tx_ring *txr = adapter->tx_rings; int error, rid, vector = 0; int cpu_id = 0; unsigned int rss_buckets = 0; cpuset_t cpu_mask; /* * If we're doing RSS, the number of queues needs to * match the number of RSS buckets that are configured. * * + If there's more queues than RSS buckets, we'll end * up with queues that get no traffic. * * + If there's more RSS buckets than queues, we'll end * up having multiple RSS buckets map to the same queue, * so there'll be some contention. */ rss_buckets = rss_getnumbuckets(); if ((adapter->feat_en & IXGBE_FEATURE_RSS) && (adapter->num_queues != rss_buckets)) { device_printf(dev, "%s: number of queues (%d) != number of RSS buckets (%d); performance will be impacted.\n", __func__, adapter->num_queues, rss_buckets); } for (int i = 0; i < adapter->num_queues; i++, vector++, que++, txr++) { rid = vector + 1; que->res = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid, RF_SHAREABLE | RF_ACTIVE); if (que->res == NULL) { device_printf(dev, "Unable to allocate bus resource: que interrupt [%d]\n", vector); return (ENXIO); } /* Set the handler function */ error = bus_setup_intr(dev, que->res, INTR_TYPE_NET | INTR_MPSAFE, NULL, ixgbe_msix_que, que, &que->tag); if (error) { que->res = NULL; device_printf(dev, "Failed to register QUE handler"); return (error); } #if __FreeBSD_version >= 800504 bus_describe_intr(dev, que->res, que->tag, "q%d", i); #endif que->msix = vector; adapter->active_queues |= (u64)(1 << que->msix); if (adapter->feat_en & IXGBE_FEATURE_RSS) { /* * The queue ID is used as the RSS layer bucket ID. * We look up the queue ID -> RSS CPU ID and select * that. */ cpu_id = rss_getcpu(i % rss_buckets); CPU_SETOF(cpu_id, &cpu_mask); } else { /* * Bind the MSI-X vector, and thus the * rings to the corresponding CPU. * * This just happens to match the default RSS * round-robin bucket -> queue -> CPU allocation. */ if (adapter->num_queues > 1) cpu_id = i; } if (adapter->num_queues > 1) bus_bind_intr(dev, que->res, cpu_id); #ifdef IXGBE_DEBUG if (adapter->feat_en & IXGBE_FEATURE_RSS) device_printf(dev, "Bound RSS bucket %d to CPU %d\n", i, cpu_id); else device_printf(dev, "Bound queue %d to cpu %d\n", i, cpu_id); #endif /* IXGBE_DEBUG */ if (!(adapter->feat_en & IXGBE_FEATURE_LEGACY_TX)) TASK_INIT(&txr->txq_task, 0, ixgbe_deferred_mq_start, txr); TASK_INIT(&que->que_task, 0, ixgbe_handle_que, que); que->tq = taskqueue_create_fast("ixgbe_que", M_NOWAIT, taskqueue_thread_enqueue, &que->tq); #if __FreeBSD_version < 1100000 taskqueue_start_threads(&que->tq, 1, PI_NET, "%s:q%d", device_get_nameunit(adapter->dev), i); #else if (adapter->feat_en & IXGBE_FEATURE_RSS) taskqueue_start_threads_cpuset(&que->tq, 1, PI_NET, &cpu_mask, "%s (bucket %d)", device_get_nameunit(adapter->dev), cpu_id); else taskqueue_start_threads_cpuset(&que->tq, 1, PI_NET, NULL, "%s:q%d", device_get_nameunit(adapter->dev), i); #endif } /* and Link */ adapter->link_rid = vector + 1; adapter->res = bus_alloc_resource_any(dev, SYS_RES_IRQ, &adapter->link_rid, RF_SHAREABLE | RF_ACTIVE); if (!adapter->res) { device_printf(dev, "Unable to allocate bus resource: Link interrupt [%d]\n", adapter->link_rid); return (ENXIO); } /* Set the link handler function */ error = bus_setup_intr(dev, adapter->res, INTR_TYPE_NET | INTR_MPSAFE, NULL, ixgbe_msix_link, adapter, &adapter->tag); if (error) { adapter->res = NULL; device_printf(dev, "Failed to register LINK handler"); return (error); } #if __FreeBSD_version >= 800504 bus_describe_intr(dev, adapter->res, adapter->tag, "link"); #endif adapter->vector = vector; /* Tasklets for Link, SFP and Multispeed Fiber */ TASK_INIT(&adapter->link_task, 0, ixgbe_handle_link, adapter); TASK_INIT(&adapter->mod_task, 0, ixgbe_handle_mod, adapter); TASK_INIT(&adapter->msf_task, 0, ixgbe_handle_msf, adapter); if (adapter->feat_cap & IXGBE_FEATURE_SRIOV) TASK_INIT(&adapter->mbx_task, 0, ixgbe_handle_mbx, adapter); TASK_INIT(&adapter->phy_task, 0, ixgbe_handle_phy, adapter); if (adapter->feat_en & IXGBE_FEATURE_FDIR) TASK_INIT(&adapter->fdir_task, 0, ixgbe_reinit_fdir, adapter); adapter->tq = taskqueue_create_fast("ixgbe_link", M_NOWAIT, taskqueue_thread_enqueue, &adapter->tq); taskqueue_start_threads(&adapter->tq, 1, PI_NET, "%s linkq", device_get_nameunit(adapter->dev)); return (0); } /* ixgbe_allocate_msix */ /************************************************************************ * ixgbe_configure_interrupts * * Setup MSI-X, MSI, or legacy interrupts (in that order). * This will also depend on user settings. ************************************************************************/ static int ixgbe_configure_interrupts(struct adapter *adapter) { device_t dev = adapter->dev; int rid, want, queues, msgs; /* Default to 1 queue if MSI-X setup fails */ adapter->num_queues = 1; /* Override by tuneable */ if (!(adapter->feat_cap & IXGBE_FEATURE_MSIX)) goto msi; /* First try MSI-X */ msgs = pci_msix_count(dev); if (msgs == 0) goto msi; rid = PCIR_BAR(MSIX_82598_BAR); adapter->msix_mem = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid, RF_ACTIVE); if (adapter->msix_mem == NULL) { rid += 4; /* 82599 maps in higher BAR */ adapter->msix_mem = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid, RF_ACTIVE); } if (adapter->msix_mem == NULL) { /* May not be enabled */ device_printf(adapter->dev, "Unable to map MSI-X table.\n"); goto msi; } /* Figure out a reasonable auto config value */ queues = min(mp_ncpus, msgs - 1); /* If we're doing RSS, clamp at the number of RSS buckets */ if (adapter->feat_en & IXGBE_FEATURE_RSS) queues = min(queues, rss_getnumbuckets()); if (ixgbe_num_queues > queues) { device_printf(adapter->dev, "ixgbe_num_queues (%d) is too large, using reduced amount (%d).\n", ixgbe_num_queues, queues); ixgbe_num_queues = queues; } if (ixgbe_num_queues != 0) queues = ixgbe_num_queues; /* Set max queues to 8 when autoconfiguring */ else queues = min(queues, 8); /* reflect correct sysctl value */ ixgbe_num_queues = queues; /* * Want one vector (RX/TX pair) per queue * plus an additional for Link. */ want = queues + 1; if (msgs >= want) msgs = want; else { device_printf(adapter->dev, "MSI-X Configuration Problem, %d vectors but %d queues wanted!\n", msgs, want); goto msi; } if ((pci_alloc_msix(dev, &msgs) == 0) && (msgs == want)) { device_printf(adapter->dev, "Using MSI-X interrupts with %d vectors\n", msgs); adapter->num_queues = queues; adapter->feat_en |= IXGBE_FEATURE_MSIX; return (0); } /* * MSI-X allocation failed or provided us with * less vectors than needed. Free MSI-X resources * and we'll try enabling MSI. */ pci_release_msi(dev); msi: /* Without MSI-X, some features are no longer supported */ adapter->feat_cap &= ~IXGBE_FEATURE_RSS; adapter->feat_en &= ~IXGBE_FEATURE_RSS; adapter->feat_cap &= ~IXGBE_FEATURE_SRIOV; adapter->feat_en &= ~IXGBE_FEATURE_SRIOV; if (adapter->msix_mem != NULL) { bus_release_resource(dev, SYS_RES_MEMORY, rid, adapter->msix_mem); adapter->msix_mem = NULL; } msgs = 1; if (pci_alloc_msi(dev, &msgs) == 0) { adapter->feat_en |= IXGBE_FEATURE_MSI; adapter->link_rid = 1; device_printf(adapter->dev, "Using an MSI interrupt\n"); return (0); } if (!(adapter->feat_cap & IXGBE_FEATURE_LEGACY_IRQ)) { device_printf(adapter->dev, "Device does not support legacy interrupts.\n"); return 1; } adapter->feat_en |= IXGBE_FEATURE_LEGACY_IRQ; adapter->link_rid = 0; device_printf(adapter->dev, "Using a Legacy interrupt\n"); return (0); } /* ixgbe_configure_interrupts */ /************************************************************************ * ixgbe_handle_link - Tasklet for MSI-X Link interrupts * * Done outside of interrupt context since the driver might sleep ************************************************************************/ static void ixgbe_handle_link(void *context, int pending) { struct adapter *adapter = context; struct ixgbe_hw *hw = &adapter->hw; ixgbe_check_link(hw, &adapter->link_speed, &adapter->link_up, 0); ixgbe_update_link_status(adapter); /* Re-enable link interrupts */ IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_LSC); } /* ixgbe_handle_link */ /************************************************************************ * ixgbe_rearm_queues ************************************************************************/ static void ixgbe_rearm_queues(struct adapter *adapter, u64 queues) { u32 mask; switch (adapter->hw.mac.type) { case ixgbe_mac_82598EB: mask = (IXGBE_EIMS_RTX_QUEUE & queues); IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask); break; case ixgbe_mac_82599EB: case ixgbe_mac_X540: case ixgbe_mac_X550: case ixgbe_mac_X550EM_x: case ixgbe_mac_X550EM_a: mask = (queues & 0xFFFFFFFF); IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask); mask = (queues >> 32); IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask); break; default: break; } } /* ixgbe_rearm_queues */ Index: projects/runtime-coverage/sys/kern/kern_rwlock.c =================================================================== --- projects/runtime-coverage/sys/kern/kern_rwlock.c (revision 325498) +++ projects/runtime-coverage/sys/kern/kern_rwlock.c (revision 325499) @@ -1,1411 +1,1411 @@ /*- * Copyright (c) 2006 John Baldwin * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /* * Machine independent bits of reader/writer lock implementation. */ #include __FBSDID("$FreeBSD$"); #include "opt_ddb.h" #include "opt_hwpmc_hooks.h" #include "opt_no_adaptive_rwlocks.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #if defined(SMP) && !defined(NO_ADAPTIVE_RWLOCKS) #define ADAPTIVE_RWLOCKS #endif #ifdef HWPMC_HOOKS #include PMC_SOFT_DECLARE( , , lock, failed); #endif /* * Return the rwlock address when the lock cookie address is provided. * This functionality assumes that struct rwlock* have a member named rw_lock. */ #define rwlock2rw(c) (__containerof(c, struct rwlock, rw_lock)) #ifdef DDB #include static void db_show_rwlock(const struct lock_object *lock); #endif static void assert_rw(const struct lock_object *lock, int what); static void lock_rw(struct lock_object *lock, uintptr_t how); #ifdef KDTRACE_HOOKS static int owner_rw(const struct lock_object *lock, struct thread **owner); #endif static uintptr_t unlock_rw(struct lock_object *lock); struct lock_class lock_class_rw = { .lc_name = "rw", .lc_flags = LC_SLEEPLOCK | LC_RECURSABLE | LC_UPGRADABLE, .lc_assert = assert_rw, #ifdef DDB .lc_ddb_show = db_show_rwlock, #endif .lc_lock = lock_rw, .lc_unlock = unlock_rw, #ifdef KDTRACE_HOOKS .lc_owner = owner_rw, #endif }; #ifdef ADAPTIVE_RWLOCKS static int __read_frequently rowner_retries = 10; static int __read_frequently rowner_loops = 10000; static SYSCTL_NODE(_debug, OID_AUTO, rwlock, CTLFLAG_RD, NULL, "rwlock debugging"); SYSCTL_INT(_debug_rwlock, OID_AUTO, retry, CTLFLAG_RW, &rowner_retries, 0, ""); SYSCTL_INT(_debug_rwlock, OID_AUTO, loops, CTLFLAG_RW, &rowner_loops, 0, ""); static struct lock_delay_config __read_frequently rw_delay; SYSCTL_INT(_debug_rwlock, OID_AUTO, delay_base, CTLFLAG_RW, &rw_delay.base, 0, ""); SYSCTL_INT(_debug_rwlock, OID_AUTO, delay_max, CTLFLAG_RW, &rw_delay.max, 0, ""); LOCK_DELAY_SYSINIT_DEFAULT(rw_delay); #endif /* * Return a pointer to the owning thread if the lock is write-locked or * NULL if the lock is unlocked or read-locked. */ #define lv_rw_wowner(v) \ ((v) & RW_LOCK_READ ? NULL : \ (struct thread *)RW_OWNER((v))) #define rw_wowner(rw) lv_rw_wowner(RW_READ_VALUE(rw)) /* * Returns if a write owner is recursed. Write ownership is not assured * here and should be previously checked. */ #define rw_recursed(rw) ((rw)->rw_recurse != 0) /* * Return true if curthread helds the lock. */ #define rw_wlocked(rw) (rw_wowner((rw)) == curthread) /* * Return a pointer to the owning thread for this lock who should receive * any priority lent by threads that block on this lock. Currently this * is identical to rw_wowner(). */ #define rw_owner(rw) rw_wowner(rw) #ifndef INVARIANTS #define __rw_assert(c, what, file, line) #endif void assert_rw(const struct lock_object *lock, int what) { rw_assert((const struct rwlock *)lock, what); } void lock_rw(struct lock_object *lock, uintptr_t how) { struct rwlock *rw; rw = (struct rwlock *)lock; if (how) rw_rlock(rw); else rw_wlock(rw); } uintptr_t unlock_rw(struct lock_object *lock) { struct rwlock *rw; rw = (struct rwlock *)lock; rw_assert(rw, RA_LOCKED | LA_NOTRECURSED); if (rw->rw_lock & RW_LOCK_READ) { rw_runlock(rw); return (1); } else { rw_wunlock(rw); return (0); } } #ifdef KDTRACE_HOOKS int owner_rw(const struct lock_object *lock, struct thread **owner) { const struct rwlock *rw = (const struct rwlock *)lock; uintptr_t x = rw->rw_lock; *owner = rw_wowner(rw); return ((x & RW_LOCK_READ) != 0 ? (RW_READERS(x) != 0) : (*owner != NULL)); } #endif void _rw_init_flags(volatile uintptr_t *c, const char *name, int opts) { struct rwlock *rw; int flags; rw = rwlock2rw(c); MPASS((opts & ~(RW_DUPOK | RW_NOPROFILE | RW_NOWITNESS | RW_QUIET | RW_RECURSE | RW_NEW)) == 0); ASSERT_ATOMIC_LOAD_PTR(rw->rw_lock, ("%s: rw_lock not aligned for %s: %p", __func__, name, &rw->rw_lock)); flags = LO_UPGRADABLE; if (opts & RW_DUPOK) flags |= LO_DUPOK; if (opts & RW_NOPROFILE) flags |= LO_NOPROFILE; if (!(opts & RW_NOWITNESS)) flags |= LO_WITNESS; if (opts & RW_RECURSE) flags |= LO_RECURSABLE; if (opts & RW_QUIET) flags |= LO_QUIET; if (opts & RW_NEW) flags |= LO_NEW; lock_init(&rw->lock_object, &lock_class_rw, name, NULL, flags); rw->rw_lock = RW_UNLOCKED; rw->rw_recurse = 0; } void _rw_destroy(volatile uintptr_t *c) { struct rwlock *rw; rw = rwlock2rw(c); KASSERT(rw->rw_lock == RW_UNLOCKED, ("rw lock %p not unlocked", rw)); KASSERT(rw->rw_recurse == 0, ("rw lock %p still recursed", rw)); rw->rw_lock = RW_DESTROYED; lock_destroy(&rw->lock_object); } void rw_sysinit(void *arg) { struct rw_args *args = arg; rw_init((struct rwlock *)args->ra_rw, args->ra_desc); } void rw_sysinit_flags(void *arg) { struct rw_args_flags *args = arg; rw_init_flags((struct rwlock *)args->ra_rw, args->ra_desc, args->ra_flags); } int _rw_wowned(const volatile uintptr_t *c) { return (rw_wowner(rwlock2rw(c)) == curthread); } void _rw_wlock_cookie(volatile uintptr_t *c, const char *file, int line) { struct rwlock *rw; uintptr_t tid, v; rw = rwlock2rw(c); KASSERT(kdb_active != 0 || SCHEDULER_STOPPED() || !TD_IS_IDLETHREAD(curthread), ("rw_wlock() by idle thread %p on rwlock %s @ %s:%d", curthread, rw->lock_object.lo_name, file, line)); KASSERT(rw->rw_lock != RW_DESTROYED, ("rw_wlock() of destroyed rwlock @ %s:%d", file, line)); WITNESS_CHECKORDER(&rw->lock_object, LOP_NEWORDER | LOP_EXCLUSIVE, file, line, NULL); tid = (uintptr_t)curthread; v = RW_UNLOCKED; if (!_rw_write_lock_fetch(rw, &v, tid)) _rw_wlock_hard(rw, v, tid, file, line); else LOCKSTAT_PROFILE_OBTAIN_RWLOCK_SUCCESS(rw__acquire, rw, 0, 0, file, line, LOCKSTAT_WRITER); LOCK_LOG_LOCK("WLOCK", &rw->lock_object, 0, rw->rw_recurse, file, line); WITNESS_LOCK(&rw->lock_object, LOP_EXCLUSIVE, file, line); TD_LOCKS_INC(curthread); } int __rw_try_wlock(volatile uintptr_t *c, const char *file, int line) { struct rwlock *rw; struct thread *td; uintptr_t tid, v; int rval; bool recursed; td = curthread; tid = (uintptr_t)td; if (SCHEDULER_STOPPED_TD(td)) return (1); rw = rwlock2rw(c); KASSERT(kdb_active != 0 || !TD_IS_IDLETHREAD(td), ("rw_try_wlock() by idle thread %p on rwlock %s @ %s:%d", curthread, rw->lock_object.lo_name, file, line)); KASSERT(rw->rw_lock != RW_DESTROYED, ("rw_try_wlock() of destroyed rwlock @ %s:%d", file, line)); rval = 1; recursed = false; v = RW_UNLOCKED; for (;;) { if (atomic_fcmpset_acq_ptr(&rw->rw_lock, &v, tid)) break; if (v == RW_UNLOCKED) continue; if (v == tid && (rw->lock_object.lo_flags & LO_RECURSABLE)) { rw->rw_recurse++; atomic_set_ptr(&rw->rw_lock, RW_LOCK_WRITER_RECURSED); break; } rval = 0; break; } LOCK_LOG_TRY("WLOCK", &rw->lock_object, 0, rval, file, line); if (rval) { WITNESS_LOCK(&rw->lock_object, LOP_EXCLUSIVE | LOP_TRYLOCK, file, line); if (!recursed) LOCKSTAT_PROFILE_OBTAIN_RWLOCK_SUCCESS(rw__acquire, rw, 0, 0, file, line, LOCKSTAT_WRITER); TD_LOCKS_INC(curthread); } return (rval); } void _rw_wunlock_cookie(volatile uintptr_t *c, const char *file, int line) { struct rwlock *rw; rw = rwlock2rw(c); KASSERT(rw->rw_lock != RW_DESTROYED, ("rw_wunlock() of destroyed rwlock @ %s:%d", file, line)); __rw_assert(c, RA_WLOCKED, file, line); WITNESS_UNLOCK(&rw->lock_object, LOP_EXCLUSIVE, file, line); LOCK_LOG_LOCK("WUNLOCK", &rw->lock_object, 0, rw->rw_recurse, file, line); #ifdef LOCK_PROFILING _rw_wunlock_hard(rw, (uintptr_t)curthread, file, line); #else __rw_wunlock(rw, curthread, file, line); #endif TD_LOCKS_DEC(curthread); } /* * Determines whether a new reader can acquire a lock. Succeeds if the * reader already owns a read lock and the lock is locked for read to * prevent deadlock from reader recursion. Also succeeds if the lock * is unlocked and has no writer waiters or spinners. Failing otherwise * prioritizes writers before readers. */ #define RW_CAN_READ(td, _rw) \ (((_rw) & (RW_LOCK_READ | RW_LOCK_WRITE_WAITERS | RW_LOCK_WRITE_SPINNER)) ==\ RW_LOCK_READ || ((td)->td_rw_rlocks && (_rw) & RW_LOCK_READ)) static bool __always_inline __rw_rlock_try(struct rwlock *rw, struct thread *td, uintptr_t *vp, const char *file, int line) { /* * Handle the easy case. If no other thread has a write * lock, then try to bump up the count of read locks. Note * that we have to preserve the current state of the * RW_LOCK_WRITE_WAITERS flag. If we fail to acquire a * read lock, then rw_lock must have changed, so restart * the loop. Note that this handles the case of a * completely unlocked rwlock since such a lock is encoded * as a read lock with no waiters. */ while (RW_CAN_READ(td, *vp)) { if (atomic_fcmpset_acq_ptr(&rw->rw_lock, vp, *vp + RW_ONE_READER)) { if (LOCK_LOG_TEST(&rw->lock_object, 0)) CTR4(KTR_LOCK, "%s: %p succeed %p -> %p", __func__, rw, (void *)*vp, (void *)(*vp + RW_ONE_READER)); td->td_rw_rlocks++; return (true); } } return (false); } static void __noinline __rw_rlock_hard(volatile uintptr_t *c, struct thread *td, uintptr_t v, const char *file, int line) { struct rwlock *rw; struct turnstile *ts; #ifdef ADAPTIVE_RWLOCKS volatile struct thread *owner; int spintries = 0; int i, n; #endif #ifdef LOCK_PROFILING uint64_t waittime = 0; int contested = 0; #endif #if defined(ADAPTIVE_RWLOCKS) || defined(KDTRACE_HOOKS) struct lock_delay_arg lda; #endif #ifdef KDTRACE_HOOKS uintptr_t state; u_int sleep_cnt = 0; int64_t sleep_time = 0; int64_t all_time = 0; #endif if (SCHEDULER_STOPPED()) return; #if defined(ADAPTIVE_RWLOCKS) lock_delay_arg_init(&lda, &rw_delay); #elif defined(KDTRACE_HOOKS) lock_delay_arg_init(&lda, NULL); #endif rw = rwlock2rw(c); #ifdef KDTRACE_HOOKS all_time -= lockstat_nsecs(&rw->lock_object); #endif #ifdef KDTRACE_HOOKS state = v; #endif for (;;) { if (__rw_rlock_try(rw, td, &v, file, line)) break; #ifdef KDTRACE_HOOKS lda.spin_cnt++; #endif #ifdef HWPMC_HOOKS PMC_SOFT_CALL( , , lock, failed); #endif lock_profile_obtain_lock_failed(&rw->lock_object, &contested, &waittime); #ifdef ADAPTIVE_RWLOCKS /* * If the owner is running on another CPU, spin until * the owner stops running or the state of the lock * changes. */ if ((v & RW_LOCK_READ) == 0) { owner = (struct thread *)RW_OWNER(v); if (TD_IS_RUNNING(owner)) { if (LOCK_LOG_TEST(&rw->lock_object, 0)) CTR3(KTR_LOCK, "%s: spinning on %p held by %p", __func__, rw, owner); KTR_STATE1(KTR_SCHED, "thread", sched_tdname(curthread), "spinning", "lockname:\"%s\"", rw->lock_object.lo_name); do { lock_delay(&lda); v = RW_READ_VALUE(rw); owner = lv_rw_wowner(v); } while (owner != NULL && TD_IS_RUNNING(owner)); KTR_STATE0(KTR_SCHED, "thread", sched_tdname(curthread), "running"); continue; } } else if (spintries < rowner_retries) { spintries++; KTR_STATE1(KTR_SCHED, "thread", sched_tdname(curthread), "spinning", "lockname:\"%s\"", rw->lock_object.lo_name); for (i = 0; i < rowner_loops; i += n) { n = RW_READERS(v); lock_delay_spin(n); v = RW_READ_VALUE(rw); if ((v & RW_LOCK_READ) == 0 || RW_CAN_READ(td, v)) break; } #ifdef KDTRACE_HOOKS lda.spin_cnt += rowner_loops - i; #endif KTR_STATE0(KTR_SCHED, "thread", sched_tdname(curthread), "running"); if (i != rowner_loops) continue; } #endif /* * Okay, now it's the hard case. Some other thread already * has a write lock or there are write waiters present, * acquire the turnstile lock so we can begin the process * of blocking. */ ts = turnstile_trywait(&rw->lock_object); /* * The lock might have been released while we spun, so * recheck its state and restart the loop if needed. */ v = RW_READ_VALUE(rw); if (RW_CAN_READ(td, v)) { turnstile_cancel(ts); continue; } #ifdef ADAPTIVE_RWLOCKS /* * The current lock owner might have started executing * on another CPU (or the lock could have changed * owners) while we were waiting on the turnstile * chain lock. If so, drop the turnstile lock and try * again. */ if ((v & RW_LOCK_READ) == 0) { owner = (struct thread *)RW_OWNER(v); if (TD_IS_RUNNING(owner)) { turnstile_cancel(ts); continue; } } #endif /* * The lock is held in write mode or it already has waiters. */ MPASS(!RW_CAN_READ(td, v)); /* * If the RW_LOCK_READ_WAITERS flag is already set, then * we can go ahead and block. If it is not set then try * to set it. If we fail to set it drop the turnstile * lock and restart the loop. */ if (!(v & RW_LOCK_READ_WAITERS)) { if (!atomic_cmpset_ptr(&rw->rw_lock, v, v | RW_LOCK_READ_WAITERS)) { turnstile_cancel(ts); v = RW_READ_VALUE(rw); continue; } if (LOCK_LOG_TEST(&rw->lock_object, 0)) CTR2(KTR_LOCK, "%s: %p set read waiters flag", __func__, rw); } /* * We were unable to acquire the lock and the read waiters * flag is set, so we must block on the turnstile. */ if (LOCK_LOG_TEST(&rw->lock_object, 0)) CTR2(KTR_LOCK, "%s: %p blocking on turnstile", __func__, rw); #ifdef KDTRACE_HOOKS sleep_time -= lockstat_nsecs(&rw->lock_object); #endif turnstile_wait(ts, rw_owner(rw), TS_SHARED_QUEUE); #ifdef KDTRACE_HOOKS sleep_time += lockstat_nsecs(&rw->lock_object); sleep_cnt++; #endif if (LOCK_LOG_TEST(&rw->lock_object, 0)) CTR2(KTR_LOCK, "%s: %p resuming from turnstile", __func__, rw); v = RW_READ_VALUE(rw); } #ifdef KDTRACE_HOOKS all_time += lockstat_nsecs(&rw->lock_object); if (sleep_time) LOCKSTAT_RECORD4(rw__block, rw, sleep_time, LOCKSTAT_READER, (state & RW_LOCK_READ) == 0, (state & RW_LOCK_READ) == 0 ? 0 : RW_READERS(state)); /* Record only the loops spinning and not sleeping. */ if (lda.spin_cnt > sleep_cnt) LOCKSTAT_RECORD4(rw__spin, rw, all_time - sleep_time, LOCKSTAT_READER, (state & RW_LOCK_READ) == 0, (state & RW_LOCK_READ) == 0 ? 0 : RW_READERS(state)); #endif /* * TODO: acquire "owner of record" here. Here be turnstile dragons * however. turnstiles don't like owners changing between calls to * turnstile_wait() currently. */ LOCKSTAT_PROFILE_OBTAIN_RWLOCK_SUCCESS(rw__acquire, rw, contested, waittime, file, line, LOCKSTAT_READER); } void __rw_rlock(volatile uintptr_t *c, const char *file, int line) { struct rwlock *rw; struct thread *td; uintptr_t v; td = curthread; rw = rwlock2rw(c); KASSERT(kdb_active != 0 || SCHEDULER_STOPPED_TD(td) || !TD_IS_IDLETHREAD(td), ("rw_rlock() by idle thread %p on rwlock %s @ %s:%d", td, rw->lock_object.lo_name, file, line)); KASSERT(rw->rw_lock != RW_DESTROYED, ("rw_rlock() of destroyed rwlock @ %s:%d", file, line)); KASSERT(rw_wowner(rw) != td, ("rw_rlock: wlock already held for %s @ %s:%d", rw->lock_object.lo_name, file, line)); WITNESS_CHECKORDER(&rw->lock_object, LOP_NEWORDER, file, line, NULL); v = RW_READ_VALUE(rw); if (__predict_false(LOCKSTAT_OOL_PROFILE_ENABLED(rw__acquire) || !__rw_rlock_try(rw, td, &v, file, line))) __rw_rlock_hard(c, td, v, file, line); LOCK_LOG_LOCK("RLOCK", &rw->lock_object, 0, 0, file, line); WITNESS_LOCK(&rw->lock_object, 0, file, line); TD_LOCKS_INC(curthread); } int __rw_try_rlock(volatile uintptr_t *c, const char *file, int line) { struct rwlock *rw; uintptr_t x; if (SCHEDULER_STOPPED()) return (1); rw = rwlock2rw(c); KASSERT(kdb_active != 0 || !TD_IS_IDLETHREAD(curthread), ("rw_try_rlock() by idle thread %p on rwlock %s @ %s:%d", curthread, rw->lock_object.lo_name, file, line)); x = rw->rw_lock; for (;;) { KASSERT(rw->rw_lock != RW_DESTROYED, ("rw_try_rlock() of destroyed rwlock @ %s:%d", file, line)); if (!(x & RW_LOCK_READ)) break; if (atomic_fcmpset_acq_ptr(&rw->rw_lock, &x, x + RW_ONE_READER)) { LOCK_LOG_TRY("RLOCK", &rw->lock_object, 0, 1, file, line); WITNESS_LOCK(&rw->lock_object, LOP_TRYLOCK, file, line); LOCKSTAT_PROFILE_OBTAIN_RWLOCK_SUCCESS(rw__acquire, rw, 0, 0, file, line, LOCKSTAT_READER); TD_LOCKS_INC(curthread); curthread->td_rw_rlocks++; return (1); } } LOCK_LOG_TRY("RLOCK", &rw->lock_object, 0, 0, file, line); return (0); } static bool __always_inline __rw_runlock_try(struct rwlock *rw, struct thread *td, uintptr_t *vp) { for (;;) { /* * See if there is more than one read lock held. If so, * just drop one and return. */ if (RW_READERS(*vp) > 1) { if (atomic_fcmpset_rel_ptr(&rw->rw_lock, vp, *vp - RW_ONE_READER)) { if (LOCK_LOG_TEST(&rw->lock_object, 0)) CTR4(KTR_LOCK, "%s: %p succeeded %p -> %p", __func__, rw, (void *)*vp, (void *)(*vp - RW_ONE_READER)); td->td_rw_rlocks--; return (true); } continue; } /* * If there aren't any waiters for a write lock, then try * to drop it quickly. */ if (!(*vp & RW_LOCK_WAITERS)) { MPASS((*vp & ~RW_LOCK_WRITE_SPINNER) == RW_READERS_LOCK(1)); if (atomic_fcmpset_rel_ptr(&rw->rw_lock, vp, RW_UNLOCKED)) { if (LOCK_LOG_TEST(&rw->lock_object, 0)) CTR2(KTR_LOCK, "%s: %p last succeeded", __func__, rw); td->td_rw_rlocks--; return (true); } continue; } break; } return (false); } static void __noinline __rw_runlock_hard(volatile uintptr_t *c, struct thread *td, uintptr_t v, const char *file, int line) { struct rwlock *rw; struct turnstile *ts; uintptr_t x, queue; if (SCHEDULER_STOPPED()) return; rw = rwlock2rw(c); for (;;) { if (__rw_runlock_try(rw, td, &v)) break; /* * Ok, we know we have waiters and we think we are the * last reader, so grab the turnstile lock. */ turnstile_chain_lock(&rw->lock_object); v = rw->rw_lock & (RW_LOCK_WAITERS | RW_LOCK_WRITE_SPINNER); MPASS(v & RW_LOCK_WAITERS); /* * Try to drop our lock leaving the lock in a unlocked * state. * * If you wanted to do explicit lock handoff you'd have to * do it here. You'd also want to use turnstile_signal() * and you'd have to handle the race where a higher * priority thread blocks on the write lock before the * thread you wakeup actually runs and have the new thread * "steal" the lock. For now it's a lot simpler to just * wakeup all of the waiters. * * As above, if we fail, then another thread might have * acquired a read lock, so drop the turnstile lock and * restart. */ x = RW_UNLOCKED; if (v & RW_LOCK_WRITE_WAITERS) { queue = TS_EXCLUSIVE_QUEUE; x |= (v & RW_LOCK_READ_WAITERS); } else queue = TS_SHARED_QUEUE; if (!atomic_cmpset_rel_ptr(&rw->rw_lock, RW_READERS_LOCK(1) | v, x)) { turnstile_chain_unlock(&rw->lock_object); v = RW_READ_VALUE(rw); continue; } if (LOCK_LOG_TEST(&rw->lock_object, 0)) CTR2(KTR_LOCK, "%s: %p last succeeded with waiters", __func__, rw); /* * Ok. The lock is released and all that's left is to * wake up the waiters. Note that the lock might not be * free anymore, but in that case the writers will just * block again if they run before the new lock holder(s) * release the lock. */ ts = turnstile_lookup(&rw->lock_object); MPASS(ts != NULL); turnstile_broadcast(ts, queue); turnstile_unpend(ts, TS_SHARED_LOCK); turnstile_chain_unlock(&rw->lock_object); td->td_rw_rlocks--; break; } LOCKSTAT_PROFILE_RELEASE_RWLOCK(rw__release, rw, LOCKSTAT_READER); } void _rw_runlock_cookie(volatile uintptr_t *c, const char *file, int line) { struct rwlock *rw; struct thread *td; uintptr_t v; rw = rwlock2rw(c); KASSERT(rw->rw_lock != RW_DESTROYED, ("rw_runlock() of destroyed rwlock @ %s:%d", file, line)); __rw_assert(c, RA_RLOCKED, file, line); WITNESS_UNLOCK(&rw->lock_object, 0, file, line); LOCK_LOG_LOCK("RUNLOCK", &rw->lock_object, 0, 0, file, line); td = curthread; v = RW_READ_VALUE(rw); if (__predict_false(LOCKSTAT_OOL_PROFILE_ENABLED(rw__release) || !__rw_runlock_try(rw, td, &v))) __rw_runlock_hard(c, td, v, file, line); TD_LOCKS_DEC(curthread); } /* * This function is called when we are unable to obtain a write lock on the * first try. This means that at least one other thread holds either a * read or write lock. */ void __rw_wlock_hard(volatile uintptr_t *c, uintptr_t v, uintptr_t tid, const char *file, int line) { struct rwlock *rw; struct turnstile *ts; #ifdef ADAPTIVE_RWLOCKS volatile struct thread *owner; int spintries = 0; int i, n; #endif uintptr_t x; #ifdef LOCK_PROFILING uint64_t waittime = 0; int contested = 0; #endif #if defined(ADAPTIVE_RWLOCKS) || defined(KDTRACE_HOOKS) struct lock_delay_arg lda; #endif #ifdef KDTRACE_HOOKS - uintptr_t state; u_int sleep_cnt = 0; int64_t sleep_time = 0; int64_t all_time = 0; #endif #if defined(KDTRACE_HOOKS) || defined(LOCK_PROFILING) + uintptr_t state; int doing_lockprof; #endif if (SCHEDULER_STOPPED()) return; #if defined(ADAPTIVE_RWLOCKS) lock_delay_arg_init(&lda, &rw_delay); #elif defined(KDTRACE_HOOKS) lock_delay_arg_init(&lda, NULL); #endif rw = rwlock2rw(c); if (__predict_false(v == RW_UNLOCKED)) v = RW_READ_VALUE(rw); if (__predict_false(lv_rw_wowner(v) == (struct thread *)tid)) { KASSERT(rw->lock_object.lo_flags & LO_RECURSABLE, ("%s: recursing but non-recursive rw %s @ %s:%d\n", __func__, rw->lock_object.lo_name, file, line)); rw->rw_recurse++; atomic_set_ptr(&rw->rw_lock, RW_LOCK_WRITER_RECURSED); if (LOCK_LOG_TEST(&rw->lock_object, 0)) CTR2(KTR_LOCK, "%s: %p recursing", __func__, rw); return; } if (LOCK_LOG_TEST(&rw->lock_object, 0)) CTR5(KTR_LOCK, "%s: %s contested (lock=%p) at %s:%d", __func__, rw->lock_object.lo_name, (void *)rw->rw_lock, file, line); #ifdef LOCK_PROFILING doing_lockprof = 1; state = v; #elif defined(KDTRACE_HOOKS) doing_lockprof = lockstat_enabled; if (__predict_false(doing_lockprof)) { all_time -= lockstat_nsecs(&rw->lock_object); state = v; } #endif for (;;) { if (v == RW_UNLOCKED) { if (_rw_write_lock_fetch(rw, &v, tid)) break; continue; } #ifdef KDTRACE_HOOKS lda.spin_cnt++; #endif #ifdef HWPMC_HOOKS PMC_SOFT_CALL( , , lock, failed); #endif lock_profile_obtain_lock_failed(&rw->lock_object, &contested, &waittime); #ifdef ADAPTIVE_RWLOCKS /* * If the lock is write locked and the owner is * running on another CPU, spin until the owner stops * running or the state of the lock changes. */ owner = lv_rw_wowner(v); if (!(v & RW_LOCK_READ) && TD_IS_RUNNING(owner)) { if (LOCK_LOG_TEST(&rw->lock_object, 0)) CTR3(KTR_LOCK, "%s: spinning on %p held by %p", __func__, rw, owner); KTR_STATE1(KTR_SCHED, "thread", sched_tdname(curthread), "spinning", "lockname:\"%s\"", rw->lock_object.lo_name); do { lock_delay(&lda); v = RW_READ_VALUE(rw); owner = lv_rw_wowner(v); } while (owner != NULL && TD_IS_RUNNING(owner)); KTR_STATE0(KTR_SCHED, "thread", sched_tdname(curthread), "running"); continue; } if ((v & RW_LOCK_READ) && RW_READERS(v) && spintries < rowner_retries) { if (!(v & RW_LOCK_WRITE_SPINNER)) { if (!atomic_cmpset_ptr(&rw->rw_lock, v, v | RW_LOCK_WRITE_SPINNER)) { v = RW_READ_VALUE(rw); continue; } } spintries++; KTR_STATE1(KTR_SCHED, "thread", sched_tdname(curthread), "spinning", "lockname:\"%s\"", rw->lock_object.lo_name); for (i = 0; i < rowner_loops; i += n) { n = RW_READERS(v); lock_delay_spin(n); v = RW_READ_VALUE(rw); if ((v & RW_LOCK_WRITE_SPINNER) == 0) break; } KTR_STATE0(KTR_SCHED, "thread", sched_tdname(curthread), "running"); #ifdef KDTRACE_HOOKS lda.spin_cnt += rowner_loops - i; #endif if (i != rowner_loops) continue; } #endif ts = turnstile_trywait(&rw->lock_object); v = RW_READ_VALUE(rw); #ifdef ADAPTIVE_RWLOCKS /* * The current lock owner might have started executing * on another CPU (or the lock could have changed * owners) while we were waiting on the turnstile * chain lock. If so, drop the turnstile lock and try * again. */ if (!(v & RW_LOCK_READ)) { owner = (struct thread *)RW_OWNER(v); if (TD_IS_RUNNING(owner)) { turnstile_cancel(ts); continue; } } #endif /* * Check for the waiters flags about this rwlock. * If the lock was released, without maintain any pending * waiters queue, simply try to acquire it. * If a pending waiters queue is present, claim the lock * ownership and maintain the pending queue. */ x = v & (RW_LOCK_WAITERS | RW_LOCK_WRITE_SPINNER); if ((v & ~x) == RW_UNLOCKED) { x &= ~RW_LOCK_WRITE_SPINNER; if (atomic_cmpset_acq_ptr(&rw->rw_lock, v, tid | x)) { if (x) turnstile_claim(ts); else turnstile_cancel(ts); break; } turnstile_cancel(ts); v = RW_READ_VALUE(rw); continue; } /* * If the RW_LOCK_WRITE_WAITERS flag isn't set, then try to * set it. If we fail to set it, then loop back and try * again. */ if (!(v & RW_LOCK_WRITE_WAITERS)) { if (!atomic_cmpset_ptr(&rw->rw_lock, v, v | RW_LOCK_WRITE_WAITERS)) { turnstile_cancel(ts); v = RW_READ_VALUE(rw); continue; } if (LOCK_LOG_TEST(&rw->lock_object, 0)) CTR2(KTR_LOCK, "%s: %p set write waiters flag", __func__, rw); } /* * We were unable to acquire the lock and the write waiters * flag is set, so we must block on the turnstile. */ if (LOCK_LOG_TEST(&rw->lock_object, 0)) CTR2(KTR_LOCK, "%s: %p blocking on turnstile", __func__, rw); #ifdef KDTRACE_HOOKS sleep_time -= lockstat_nsecs(&rw->lock_object); #endif turnstile_wait(ts, rw_owner(rw), TS_EXCLUSIVE_QUEUE); #ifdef KDTRACE_HOOKS sleep_time += lockstat_nsecs(&rw->lock_object); sleep_cnt++; #endif if (LOCK_LOG_TEST(&rw->lock_object, 0)) CTR2(KTR_LOCK, "%s: %p resuming from turnstile", __func__, rw); #ifdef ADAPTIVE_RWLOCKS spintries = 0; #endif v = RW_READ_VALUE(rw); } #if defined(KDTRACE_HOOKS) || defined(LOCK_PROFILING) if (__predict_true(!doing_lockprof)) return; #endif #ifdef KDTRACE_HOOKS all_time += lockstat_nsecs(&rw->lock_object); if (sleep_time) LOCKSTAT_RECORD4(rw__block, rw, sleep_time, LOCKSTAT_WRITER, (state & RW_LOCK_READ) == 0, (state & RW_LOCK_READ) == 0 ? 0 : RW_READERS(state)); /* Record only the loops spinning and not sleeping. */ if (lda.spin_cnt > sleep_cnt) LOCKSTAT_RECORD4(rw__spin, rw, all_time - sleep_time, LOCKSTAT_WRITER, (state & RW_LOCK_READ) == 0, (state & RW_LOCK_READ) == 0 ? 0 : RW_READERS(state)); #endif LOCKSTAT_PROFILE_OBTAIN_RWLOCK_SUCCESS(rw__acquire, rw, contested, waittime, file, line, LOCKSTAT_WRITER); } /* * This function is called if lockstat is active or the first try at releasing * a write lock failed. The latter means that the lock is recursed or one of * the 2 waiter bits must be set indicating that at least one thread is waiting * on this lock. */ void __rw_wunlock_hard(volatile uintptr_t *c, uintptr_t tid, const char *file, int line) { struct rwlock *rw; struct turnstile *ts; uintptr_t v; int queue; if (SCHEDULER_STOPPED()) return; rw = rwlock2rw(c); v = RW_READ_VALUE(rw); if (v & RW_LOCK_WRITER_RECURSED) { if (--(rw->rw_recurse) == 0) atomic_clear_ptr(&rw->rw_lock, RW_LOCK_WRITER_RECURSED); if (LOCK_LOG_TEST(&rw->lock_object, 0)) CTR2(KTR_LOCK, "%s: %p unrecursing", __func__, rw); return; } LOCKSTAT_PROFILE_RELEASE_RWLOCK(rw__release, rw, LOCKSTAT_WRITER); if (v == tid && _rw_write_unlock(rw, tid)) return; KASSERT(rw->rw_lock & (RW_LOCK_READ_WAITERS | RW_LOCK_WRITE_WAITERS), ("%s: neither of the waiter flags are set", __func__)); if (LOCK_LOG_TEST(&rw->lock_object, 0)) CTR2(KTR_LOCK, "%s: %p contested", __func__, rw); turnstile_chain_lock(&rw->lock_object); ts = turnstile_lookup(&rw->lock_object); MPASS(ts != NULL); /* * Use the same algo as sx locks for now. Prefer waking up shared * waiters if we have any over writers. This is probably not ideal. * * 'v' is the value we are going to write back to rw_lock. If we * have waiters on both queues, we need to preserve the state of * the waiter flag for the queue we don't wake up. For now this is * hardcoded for the algorithm mentioned above. * * In the case of both readers and writers waiting we wakeup the * readers but leave the RW_LOCK_WRITE_WAITERS flag set. If a * new writer comes in before a reader it will claim the lock up * above. There is probably a potential priority inversion in * there that could be worked around either by waking both queues * of waiters or doing some complicated lock handoff gymnastics. */ v = RW_UNLOCKED; if (rw->rw_lock & RW_LOCK_WRITE_WAITERS) { queue = TS_EXCLUSIVE_QUEUE; v |= (rw->rw_lock & RW_LOCK_READ_WAITERS); } else queue = TS_SHARED_QUEUE; /* Wake up all waiters for the specific queue. */ if (LOCK_LOG_TEST(&rw->lock_object, 0)) CTR3(KTR_LOCK, "%s: %p waking up %s waiters", __func__, rw, queue == TS_SHARED_QUEUE ? "read" : "write"); turnstile_broadcast(ts, queue); atomic_store_rel_ptr(&rw->rw_lock, v); turnstile_unpend(ts, TS_EXCLUSIVE_LOCK); turnstile_chain_unlock(&rw->lock_object); } /* * Attempt to do a non-blocking upgrade from a read lock to a write * lock. This will only succeed if this thread holds a single read * lock. Returns true if the upgrade succeeded and false otherwise. */ int __rw_try_upgrade(volatile uintptr_t *c, const char *file, int line) { struct rwlock *rw; uintptr_t v, x, tid; struct turnstile *ts; int success; if (SCHEDULER_STOPPED()) return (1); rw = rwlock2rw(c); KASSERT(rw->rw_lock != RW_DESTROYED, ("rw_try_upgrade() of destroyed rwlock @ %s:%d", file, line)); __rw_assert(c, RA_RLOCKED, file, line); /* * Attempt to switch from one reader to a writer. If there * are any write waiters, then we will have to lock the * turnstile first to prevent races with another writer * calling turnstile_wait() before we have claimed this * turnstile. So, do the simple case of no waiters first. */ tid = (uintptr_t)curthread; success = 0; for (;;) { v = rw->rw_lock; if (RW_READERS(v) > 1) break; if (!(v & RW_LOCK_WAITERS)) { success = atomic_cmpset_acq_ptr(&rw->rw_lock, v, tid); if (!success) continue; break; } /* * Ok, we think we have waiters, so lock the turnstile. */ ts = turnstile_trywait(&rw->lock_object); v = rw->rw_lock; if (RW_READERS(v) > 1) { turnstile_cancel(ts); break; } /* * Try to switch from one reader to a writer again. This time * we honor the current state of the waiters flags. * If we obtain the lock with the flags set, then claim * ownership of the turnstile. */ x = rw->rw_lock & RW_LOCK_WAITERS; success = atomic_cmpset_ptr(&rw->rw_lock, v, tid | x); if (success) { if (x) turnstile_claim(ts); else turnstile_cancel(ts); break; } turnstile_cancel(ts); } LOCK_LOG_TRY("WUPGRADE", &rw->lock_object, 0, success, file, line); if (success) { curthread->td_rw_rlocks--; WITNESS_UPGRADE(&rw->lock_object, LOP_EXCLUSIVE | LOP_TRYLOCK, file, line); LOCKSTAT_RECORD0(rw__upgrade, rw); } return (success); } /* * Downgrade a write lock into a single read lock. */ void __rw_downgrade(volatile uintptr_t *c, const char *file, int line) { struct rwlock *rw; struct turnstile *ts; uintptr_t tid, v; int rwait, wwait; if (SCHEDULER_STOPPED()) return; rw = rwlock2rw(c); KASSERT(rw->rw_lock != RW_DESTROYED, ("rw_downgrade() of destroyed rwlock @ %s:%d", file, line)); __rw_assert(c, RA_WLOCKED | RA_NOTRECURSED, file, line); #ifndef INVARIANTS if (rw_recursed(rw)) panic("downgrade of a recursed lock"); #endif WITNESS_DOWNGRADE(&rw->lock_object, 0, file, line); /* * Convert from a writer to a single reader. First we handle * the easy case with no waiters. If there are any waiters, we * lock the turnstile and "disown" the lock. */ tid = (uintptr_t)curthread; if (atomic_cmpset_rel_ptr(&rw->rw_lock, tid, RW_READERS_LOCK(1))) goto out; /* * Ok, we think we have waiters, so lock the turnstile so we can * read the waiter flags without any races. */ turnstile_chain_lock(&rw->lock_object); v = rw->rw_lock & RW_LOCK_WAITERS; rwait = v & RW_LOCK_READ_WAITERS; wwait = v & RW_LOCK_WRITE_WAITERS; MPASS(rwait | wwait); /* * Downgrade from a write lock while preserving waiters flag * and give up ownership of the turnstile. */ ts = turnstile_lookup(&rw->lock_object); MPASS(ts != NULL); if (!wwait) v &= ~RW_LOCK_READ_WAITERS; atomic_store_rel_ptr(&rw->rw_lock, RW_READERS_LOCK(1) | v); /* * Wake other readers if there are no writers pending. Otherwise they * won't be able to acquire the lock anyway. */ if (rwait && !wwait) { turnstile_broadcast(ts, TS_SHARED_QUEUE); turnstile_unpend(ts, TS_EXCLUSIVE_LOCK); } else turnstile_disown(ts); turnstile_chain_unlock(&rw->lock_object); out: curthread->td_rw_rlocks++; LOCK_LOG_LOCK("WDOWNGRADE", &rw->lock_object, 0, 0, file, line); LOCKSTAT_RECORD0(rw__downgrade, rw); } #ifdef INVARIANT_SUPPORT #ifndef INVARIANTS #undef __rw_assert #endif /* * In the non-WITNESS case, rw_assert() can only detect that at least * *some* thread owns an rlock, but it cannot guarantee that *this* * thread owns an rlock. */ void __rw_assert(const volatile uintptr_t *c, int what, const char *file, int line) { const struct rwlock *rw; if (panicstr != NULL) return; rw = rwlock2rw(c); switch (what) { case RA_LOCKED: case RA_LOCKED | RA_RECURSED: case RA_LOCKED | RA_NOTRECURSED: case RA_RLOCKED: case RA_RLOCKED | RA_RECURSED: case RA_RLOCKED | RA_NOTRECURSED: #ifdef WITNESS witness_assert(&rw->lock_object, what, file, line); #else /* * If some other thread has a write lock or we have one * and are asserting a read lock, fail. Also, if no one * has a lock at all, fail. */ if (rw->rw_lock == RW_UNLOCKED || (!(rw->rw_lock & RW_LOCK_READ) && (what & RA_RLOCKED || rw_wowner(rw) != curthread))) panic("Lock %s not %slocked @ %s:%d\n", rw->lock_object.lo_name, (what & RA_RLOCKED) ? "read " : "", file, line); if (!(rw->rw_lock & RW_LOCK_READ) && !(what & RA_RLOCKED)) { if (rw_recursed(rw)) { if (what & RA_NOTRECURSED) panic("Lock %s recursed @ %s:%d\n", rw->lock_object.lo_name, file, line); } else if (what & RA_RECURSED) panic("Lock %s not recursed @ %s:%d\n", rw->lock_object.lo_name, file, line); } #endif break; case RA_WLOCKED: case RA_WLOCKED | RA_RECURSED: case RA_WLOCKED | RA_NOTRECURSED: if (rw_wowner(rw) != curthread) panic("Lock %s not exclusively locked @ %s:%d\n", rw->lock_object.lo_name, file, line); if (rw_recursed(rw)) { if (what & RA_NOTRECURSED) panic("Lock %s recursed @ %s:%d\n", rw->lock_object.lo_name, file, line); } else if (what & RA_RECURSED) panic("Lock %s not recursed @ %s:%d\n", rw->lock_object.lo_name, file, line); break; case RA_UNLOCKED: #ifdef WITNESS witness_assert(&rw->lock_object, what, file, line); #else /* * If we hold a write lock fail. We can't reliably check * to see if we hold a read lock or not. */ if (rw_wowner(rw) == curthread) panic("Lock %s exclusively locked @ %s:%d\n", rw->lock_object.lo_name, file, line); #endif break; default: panic("Unknown rw lock assertion: %d @ %s:%d", what, file, line); } } #endif /* INVARIANT_SUPPORT */ #ifdef DDB void db_show_rwlock(const struct lock_object *lock) { const struct rwlock *rw; struct thread *td; rw = (const struct rwlock *)lock; db_printf(" state: "); if (rw->rw_lock == RW_UNLOCKED) db_printf("UNLOCKED\n"); else if (rw->rw_lock == RW_DESTROYED) { db_printf("DESTROYED\n"); return; } else if (rw->rw_lock & RW_LOCK_READ) db_printf("RLOCK: %ju locks\n", (uintmax_t)(RW_READERS(rw->rw_lock))); else { td = rw_wowner(rw); db_printf("WLOCK: %p (tid %d, pid %d, \"%s\")\n", td, td->td_tid, td->td_proc->p_pid, td->td_name); if (rw_recursed(rw)) db_printf(" recursed: %u\n", rw->rw_recurse); } db_printf(" waiters: "); switch (rw->rw_lock & (RW_LOCK_READ_WAITERS | RW_LOCK_WRITE_WAITERS)) { case RW_LOCK_READ_WAITERS: db_printf("readers\n"); break; case RW_LOCK_WRITE_WAITERS: db_printf("writers\n"); break; case RW_LOCK_READ_WAITERS | RW_LOCK_WRITE_WAITERS: db_printf("readers and writers\n"); break; default: db_printf("none\n"); break; } } #endif Index: projects/runtime-coverage/sys/kern/vfs_cache.c =================================================================== --- projects/runtime-coverage/sys/kern/vfs_cache.c (revision 325498) +++ projects/runtime-coverage/sys/kern/vfs_cache.c (revision 325499) @@ -1,2502 +1,2525 @@ /*- * Copyright (c) 1989, 1993, 1995 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Poul-Henning Kamp of the FreeBSD Project. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)vfs_cache.c 8.5 (Berkeley) 3/22/95 */ #include __FBSDID("$FreeBSD$"); #include "opt_ktrace.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef KTRACE #include #endif #include SDT_PROVIDER_DECLARE(vfs); SDT_PROBE_DEFINE3(vfs, namecache, enter, done, "struct vnode *", "char *", "struct vnode *"); SDT_PROBE_DEFINE2(vfs, namecache, enter_negative, done, "struct vnode *", "char *"); SDT_PROBE_DEFINE1(vfs, namecache, fullpath, entry, "struct vnode *"); SDT_PROBE_DEFINE3(vfs, namecache, fullpath, hit, "struct vnode *", "char *", "struct vnode *"); SDT_PROBE_DEFINE1(vfs, namecache, fullpath, miss, "struct vnode *"); SDT_PROBE_DEFINE3(vfs, namecache, fullpath, return, "int", "struct vnode *", "char *"); SDT_PROBE_DEFINE3(vfs, namecache, lookup, hit, "struct vnode *", "char *", "struct vnode *"); SDT_PROBE_DEFINE2(vfs, namecache, lookup, hit__negative, "struct vnode *", "char *"); SDT_PROBE_DEFINE2(vfs, namecache, lookup, miss, "struct vnode *", "char *"); SDT_PROBE_DEFINE1(vfs, namecache, purge, done, "struct vnode *"); SDT_PROBE_DEFINE1(vfs, namecache, purge_negative, done, "struct vnode *"); SDT_PROBE_DEFINE1(vfs, namecache, purgevfs, done, "struct mount *"); SDT_PROBE_DEFINE3(vfs, namecache, zap, done, "struct vnode *", "char *", "struct vnode *"); SDT_PROBE_DEFINE3(vfs, namecache, zap_negative, done, "struct vnode *", "char *", "int"); SDT_PROBE_DEFINE3(vfs, namecache, shrink_negative, done, "struct vnode *", "char *", "int"); /* * This structure describes the elements in the cache of recent * names looked up by namei. */ struct namecache { LIST_ENTRY(namecache) nc_hash; /* hash chain */ LIST_ENTRY(namecache) nc_src; /* source vnode list */ TAILQ_ENTRY(namecache) nc_dst; /* destination vnode list */ struct vnode *nc_dvp; /* vnode of parent of name */ union { struct vnode *nu_vp; /* vnode the name refers to */ u_int nu_neghits; /* negative entry hits */ } n_un; u_char nc_flag; /* flag bits */ u_char nc_nlen; /* length of name */ char nc_name[0]; /* segment name + nul */ }; /* * struct namecache_ts repeats struct namecache layout up to the * nc_nlen member. * struct namecache_ts is used in place of struct namecache when time(s) need * to be stored. The nc_dotdottime field is used when a cache entry is mapping * both a non-dotdot directory name plus dotdot for the directory's * parent. */ struct namecache_ts { struct timespec nc_time; /* timespec provided by fs */ struct timespec nc_dotdottime; /* dotdot timespec provided by fs */ int nc_ticks; /* ticks value when entry was added */ struct namecache nc_nc; }; #define nc_vp n_un.nu_vp #define nc_neghits n_un.nu_neghits /* * Flags in namecache.nc_flag */ #define NCF_WHITE 0x01 #define NCF_ISDOTDOT 0x02 #define NCF_TS 0x04 #define NCF_DTS 0x08 #define NCF_DVDROP 0x10 #define NCF_NEGATIVE 0x20 #define NCF_HOTNEGATIVE 0x40 /* * Name caching works as follows: * * Names found by directory scans are retained in a cache * for future reference. It is managed LRU, so frequently * used names will hang around. Cache is indexed by hash value * obtained from (vp, name) where vp refers to the directory * containing name. * * If it is a "negative" entry, (i.e. for a name that is known NOT to * exist) the vnode pointer will be NULL. * * Upon reaching the last segment of a path, if the reference * is for DELETE, or NOCACHE is set (rewrite), and the * name is located in the cache, it will be dropped. * * These locks are used (in the order in which they can be taken): * NAME TYPE ROLE * vnodelock mtx vnode lists and v_cache_dd field protection * bucketlock rwlock for access to given set of hash buckets * neglist mtx negative entry LRU management * * Additionally, ncneg_shrink_lock mtx is used to have at most one thread * shrinking the LRU list. * * It is legal to take multiple vnodelock and bucketlock locks. The locking * order is lower address first. Both are recursive. * * "." lookups are lockless. * * ".." and vnode -> name lookups require vnodelock. * * name -> vnode lookup requires the relevant bucketlock to be held for reading. * * Insertions and removals of entries require involved vnodes and bucketlocks * to be write-locked to prevent other threads from seeing the entry. * * Some lookups result in removal of the found entry (e.g. getting rid of a * negative entry with the intent to create a positive one), which poses a * problem when multiple threads reach the state. Similarly, two different * threads can purge two different vnodes and try to remove the same name. * * If the already held vnode lock is lower than the second required lock, we * can just take the other lock. However, in the opposite case, this could * deadlock. As such, this is resolved by trylocking and if that fails unlocking * the first node, locking everything in order and revalidating the state. */ /* * Structures associated with name caching. */ #define NCHHASH(hash) \ (&nchashtbl[(hash) & nchash]) static __read_mostly LIST_HEAD(nchashhead, namecache) *nchashtbl;/* Hash Table */ static u_long __read_mostly nchash; /* size of hash table */ SYSCTL_ULONG(_debug, OID_AUTO, nchash, CTLFLAG_RD, &nchash, 0, "Size of namecache hash table"); static u_long __read_mostly ncnegfactor = 12; /* ratio of negative entries */ SYSCTL_ULONG(_vfs, OID_AUTO, ncnegfactor, CTLFLAG_RW, &ncnegfactor, 0, "Ratio of negative namecache entries"); static u_long __exclusive_cache_line numneg; /* number of negative entries allocated */ SYSCTL_ULONG(_debug, OID_AUTO, numneg, CTLFLAG_RD, &numneg, 0, "Number of negative entries in namecache"); static u_long __exclusive_cache_line numcache;/* number of cache entries allocated */ SYSCTL_ULONG(_debug, OID_AUTO, numcache, CTLFLAG_RD, &numcache, 0, "Number of namecache entries"); static u_long __exclusive_cache_line numcachehv;/* number of cache entries with vnodes held */ SYSCTL_ULONG(_debug, OID_AUTO, numcachehv, CTLFLAG_RD, &numcachehv, 0, "Number of namecache entries with vnodes held"); u_int __read_mostly ncsizefactor = 2; SYSCTL_UINT(_vfs, OID_AUTO, ncsizefactor, CTLFLAG_RW, &ncsizefactor, 0, "Size factor for namecache"); static u_int __read_mostly ncpurgeminvnodes; SYSCTL_UINT(_vfs, OID_AUTO, ncpurgeminvnodes, CTLFLAG_RW, &ncpurgeminvnodes, 0, "Number of vnodes below which purgevfs ignores the request"); static u_int __read_mostly ncneghitsrequeue = 8; SYSCTL_UINT(_vfs, OID_AUTO, ncneghitsrequeue, CTLFLAG_RW, &ncneghitsrequeue, 0, "Number of hits to requeue a negative entry in the LRU list"); struct nchstats nchstats; /* cache effectiveness statistics */ static struct mtx ncneg_shrink_lock; static int shrink_list_turn; struct neglist { struct mtx nl_lock; TAILQ_HEAD(, namecache) nl_list; } __aligned(CACHE_LINE_SIZE); static struct neglist __read_mostly *neglists; static struct neglist ncneg_hot; #define numneglists (ncneghash + 1) static u_int __read_mostly ncneghash; static inline struct neglist * NCP2NEGLIST(struct namecache *ncp) { return (&neglists[(((uintptr_t)(ncp) >> 8) & ncneghash)]); } #define numbucketlocks (ncbuckethash + 1) static u_int __read_mostly ncbuckethash; static struct rwlock_padalign __read_mostly *bucketlocks; #define HASH2BUCKETLOCK(hash) \ ((struct rwlock *)(&bucketlocks[((hash) & ncbuckethash)])) #define numvnodelocks (ncvnodehash + 1) static u_int __read_mostly ncvnodehash; static struct mtx __read_mostly *vnodelocks; static inline struct mtx * VP2VNODELOCK(struct vnode *vp) { return (&vnodelocks[(((uintptr_t)(vp) >> 8) & ncvnodehash)]); } /* * UMA zones for the VFS cache. * * The small cache is used for entries with short names, which are the * most common. The large cache is used for entries which are too big to * fit in the small cache. */ static uma_zone_t __read_mostly cache_zone_small; static uma_zone_t __read_mostly cache_zone_small_ts; static uma_zone_t __read_mostly cache_zone_large; static uma_zone_t __read_mostly cache_zone_large_ts; #define CACHE_PATH_CUTOFF 35 static struct namecache * cache_alloc(int len, int ts) { struct namecache_ts *ncp_ts; struct namecache *ncp; if (__predict_false(ts)) { if (len <= CACHE_PATH_CUTOFF) ncp_ts = uma_zalloc(cache_zone_small_ts, M_WAITOK); else ncp_ts = uma_zalloc(cache_zone_large_ts, M_WAITOK); ncp = &ncp_ts->nc_nc; } else { if (len <= CACHE_PATH_CUTOFF) ncp = uma_zalloc(cache_zone_small, M_WAITOK); else ncp = uma_zalloc(cache_zone_large, M_WAITOK); } return (ncp); } static void cache_free(struct namecache *ncp) { struct namecache_ts *ncp_ts; if (ncp == NULL) return; if ((ncp->nc_flag & NCF_DVDROP) != 0) vdrop(ncp->nc_dvp); if (__predict_false(ncp->nc_flag & NCF_TS)) { ncp_ts = __containerof(ncp, struct namecache_ts, nc_nc); if (ncp->nc_nlen <= CACHE_PATH_CUTOFF) uma_zfree(cache_zone_small_ts, ncp_ts); else uma_zfree(cache_zone_large_ts, ncp_ts); } else { if (ncp->nc_nlen <= CACHE_PATH_CUTOFF) uma_zfree(cache_zone_small, ncp); else uma_zfree(cache_zone_large, ncp); } } static void cache_out_ts(struct namecache *ncp, struct timespec *tsp, int *ticksp) { struct namecache_ts *ncp_ts; KASSERT((ncp->nc_flag & NCF_TS) != 0 || (tsp == NULL && ticksp == NULL), ("No NCF_TS")); if (tsp == NULL && ticksp == NULL) return; ncp_ts = __containerof(ncp, struct namecache_ts, nc_nc); if (tsp != NULL) *tsp = ncp_ts->nc_time; if (ticksp != NULL) *ticksp = ncp_ts->nc_ticks; } static int __read_mostly doingcache = 1; /* 1 => enable the cache */ SYSCTL_INT(_debug, OID_AUTO, vfscache, CTLFLAG_RW, &doingcache, 0, "VFS namecache enabled"); /* Export size information to userland */ SYSCTL_INT(_debug_sizeof, OID_AUTO, namecache, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, sizeof(struct namecache), "sizeof(struct namecache)"); /* * The new name cache statistics */ static SYSCTL_NODE(_vfs, OID_AUTO, cache, CTLFLAG_RW, 0, "Name cache statistics"); #define STATNODE_ULONG(name, descr) \ SYSCTL_ULONG(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, 0, descr); #define STATNODE_COUNTER(name, descr) \ static counter_u64_t __read_mostly name; \ SYSCTL_COUNTER_U64(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, descr); STATNODE_ULONG(numneg, "Number of negative cache entries"); STATNODE_ULONG(numcache, "Number of cache entries"); STATNODE_COUNTER(numcalls, "Number of cache lookups"); STATNODE_COUNTER(dothits, "Number of '.' hits"); STATNODE_COUNTER(dotdothits, "Number of '..' hits"); STATNODE_COUNTER(numchecks, "Number of checks in lookup"); STATNODE_COUNTER(nummiss, "Number of cache misses"); STATNODE_COUNTER(nummisszap, "Number of cache misses we do not want to cache"); STATNODE_COUNTER(numposzaps, "Number of cache hits (positive) we do not want to cache"); STATNODE_COUNTER(numposhits, "Number of cache hits (positive)"); STATNODE_COUNTER(numnegzaps, "Number of cache hits (negative) we do not want to cache"); STATNODE_COUNTER(numneghits, "Number of cache hits (negative)"); /* These count for kern___getcwd(), too. */ STATNODE_COUNTER(numfullpathcalls, "Number of fullpath search calls"); STATNODE_COUNTER(numfullpathfail1, "Number of fullpath search errors (ENOTDIR)"); STATNODE_COUNTER(numfullpathfail2, "Number of fullpath search errors (VOP_VPTOCNP failures)"); STATNODE_COUNTER(numfullpathfail4, "Number of fullpath search errors (ENOMEM)"); STATNODE_COUNTER(numfullpathfound, "Number of successful fullpath calls"); static long zap_and_exit_bucket_fail; STATNODE_ULONG(zap_and_exit_bucket_fail, "Number of times zap_and_exit failed to lock"); static long cache_lock_vnodes_cel_3_failures; STATNODE_ULONG(cache_lock_vnodes_cel_3_failures, "Number of times 3-way vnode locking failed"); static void cache_zap_locked(struct namecache *ncp, bool neg_locked); static int vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir, char *buf, char **retbuf, u_int buflen); static MALLOC_DEFINE(M_VFSCACHE, "vfscache", "VFS name cache entries"); static int cache_yield; SYSCTL_INT(_vfs_cache, OID_AUTO, yield, CTLFLAG_RD, &cache_yield, 0, "Number of times cache called yield"); static void cache_maybe_yield(void) { if (should_yield()) { cache_yield++; kern_yield(PRI_USER); } } static inline void cache_assert_vlp_locked(struct mtx *vlp) { if (vlp != NULL) mtx_assert(vlp, MA_OWNED); } static inline void cache_assert_vnode_locked(struct vnode *vp) { struct mtx *vlp; vlp = VP2VNODELOCK(vp); cache_assert_vlp_locked(vlp); } static uint32_t cache_get_hash(char *name, u_char len, struct vnode *dvp) { uint32_t hash; hash = fnv_32_buf(name, len, FNV1_32_INIT); hash = fnv_32_buf(&dvp, sizeof(dvp), hash); return (hash); } static inline struct rwlock * NCP2BUCKETLOCK(struct namecache *ncp) { uint32_t hash; hash = cache_get_hash(ncp->nc_name, ncp->nc_nlen, ncp->nc_dvp); return (HASH2BUCKETLOCK(hash)); } #ifdef INVARIANTS static void cache_assert_bucket_locked(struct namecache *ncp, int mode) { struct rwlock *blp; blp = NCP2BUCKETLOCK(ncp); rw_assert(blp, mode); } #else #define cache_assert_bucket_locked(x, y) do { } while (0) #endif #define cache_sort(x, y) _cache_sort((void **)(x), (void **)(y)) static void _cache_sort(void **p1, void **p2) { void *tmp; if (*p1 > *p2) { tmp = *p2; *p2 = *p1; *p1 = tmp; } } static void cache_lock_all_buckets(void) { u_int i; for (i = 0; i < numbucketlocks; i++) rw_wlock(&bucketlocks[i]); } static void cache_unlock_all_buckets(void) { u_int i; for (i = 0; i < numbucketlocks; i++) rw_wunlock(&bucketlocks[i]); } static void cache_lock_all_vnodes(void) { u_int i; for (i = 0; i < numvnodelocks; i++) mtx_lock(&vnodelocks[i]); } static void cache_unlock_all_vnodes(void) { u_int i; for (i = 0; i < numvnodelocks; i++) mtx_unlock(&vnodelocks[i]); } static int cache_trylock_vnodes(struct mtx *vlp1, struct mtx *vlp2) { cache_sort(&vlp1, &vlp2); MPASS(vlp2 != NULL); if (vlp1 != NULL) { if (!mtx_trylock(vlp1)) return (EAGAIN); } if (!mtx_trylock(vlp2)) { if (vlp1 != NULL) mtx_unlock(vlp1); return (EAGAIN); } return (0); } static void cache_unlock_vnodes(struct mtx *vlp1, struct mtx *vlp2) { MPASS(vlp1 != NULL || vlp2 != NULL); if (vlp1 != NULL) mtx_unlock(vlp1); if (vlp2 != NULL) mtx_unlock(vlp2); } static int sysctl_nchstats(SYSCTL_HANDLER_ARGS) { struct nchstats snap; if (req->oldptr == NULL) return (SYSCTL_OUT(req, 0, sizeof(snap))); snap = nchstats; snap.ncs_goodhits = counter_u64_fetch(numposhits); snap.ncs_neghits = counter_u64_fetch(numneghits); snap.ncs_badhits = counter_u64_fetch(numposzaps) + counter_u64_fetch(numnegzaps); snap.ncs_miss = counter_u64_fetch(nummisszap) + counter_u64_fetch(nummiss); return (SYSCTL_OUT(req, &snap, sizeof(snap))); } SYSCTL_PROC(_vfs_cache, OID_AUTO, nchstats, CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 0, sysctl_nchstats, "LU", "VFS cache effectiveness statistics"); #ifdef DIAGNOSTIC /* * Grab an atomic snapshot of the name cache hash chain lengths */ static SYSCTL_NODE(_debug, OID_AUTO, hashstat, CTLFLAG_RW, NULL, "hash table stats"); static int sysctl_debug_hashstat_rawnchash(SYSCTL_HANDLER_ARGS) { struct nchashhead *ncpp; struct namecache *ncp; int i, error, n_nchash, *cntbuf; retry: n_nchash = nchash + 1; /* nchash is max index, not count */ if (req->oldptr == NULL) return SYSCTL_OUT(req, 0, n_nchash * sizeof(int)); cntbuf = malloc(n_nchash * sizeof(int), M_TEMP, M_ZERO | M_WAITOK); cache_lock_all_buckets(); if (n_nchash != nchash + 1) { cache_unlock_all_buckets(); free(cntbuf, M_TEMP); goto retry; } /* Scan hash tables counting entries */ for (ncpp = nchashtbl, i = 0; i < n_nchash; ncpp++, i++) LIST_FOREACH(ncp, ncpp, nc_hash) cntbuf[i]++; cache_unlock_all_buckets(); for (error = 0, i = 0; i < n_nchash; i++) if ((error = SYSCTL_OUT(req, &cntbuf[i], sizeof(int))) != 0) break; free(cntbuf, M_TEMP); return (error); } SYSCTL_PROC(_debug_hashstat, OID_AUTO, rawnchash, CTLTYPE_INT|CTLFLAG_RD| CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_rawnchash, "S,int", "nchash chain lengths"); static int sysctl_debug_hashstat_nchash(SYSCTL_HANDLER_ARGS) { int error; struct nchashhead *ncpp; struct namecache *ncp; int n_nchash; int count, maxlength, used, pct; if (!req->oldptr) return SYSCTL_OUT(req, 0, 4 * sizeof(int)); cache_lock_all_buckets(); n_nchash = nchash + 1; /* nchash is max index, not count */ used = 0; maxlength = 0; /* Scan hash tables for applicable entries */ for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) { count = 0; LIST_FOREACH(ncp, ncpp, nc_hash) { count++; } if (count) used++; if (maxlength < count) maxlength = count; } n_nchash = nchash + 1; cache_unlock_all_buckets(); pct = (used * 100) / (n_nchash / 100); error = SYSCTL_OUT(req, &n_nchash, sizeof(n_nchash)); if (error) return (error); error = SYSCTL_OUT(req, &used, sizeof(used)); if (error) return (error); error = SYSCTL_OUT(req, &maxlength, sizeof(maxlength)); if (error) return (error); error = SYSCTL_OUT(req, &pct, sizeof(pct)); if (error) return (error); return (0); } SYSCTL_PROC(_debug_hashstat, OID_AUTO, nchash, CTLTYPE_INT|CTLFLAG_RD| CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_nchash, "I", "nchash statistics (number of total/used buckets, maximum chain length, usage percentage)"); #endif /* * Negative entries management * * A variation of LRU scheme is used. New entries are hashed into one of * numneglists cold lists. Entries get promoted to the hot list on first hit. * Partial LRU for the hot list is maintained by requeueing them every * ncneghitsrequeue hits. * * The shrinker will demote hot list head and evict from the cold list in a * round-robin manner. */ static void cache_negative_hit(struct namecache *ncp) { struct neglist *neglist; u_int hits; MPASS(ncp->nc_flag & NCF_NEGATIVE); hits = atomic_fetchadd_int(&ncp->nc_neghits, 1); if (ncp->nc_flag & NCF_HOTNEGATIVE) { if ((hits % ncneghitsrequeue) != 0) return; mtx_lock(&ncneg_hot.nl_lock); if (ncp->nc_flag & NCF_HOTNEGATIVE) { TAILQ_REMOVE(&ncneg_hot.nl_list, ncp, nc_dst); TAILQ_INSERT_TAIL(&ncneg_hot.nl_list, ncp, nc_dst); mtx_unlock(&ncneg_hot.nl_lock); return; } /* * The shrinker cleared the flag and removed the entry from * the hot list. Put it back. */ } else { mtx_lock(&ncneg_hot.nl_lock); } neglist = NCP2NEGLIST(ncp); mtx_lock(&neglist->nl_lock); if (!(ncp->nc_flag & NCF_HOTNEGATIVE)) { TAILQ_REMOVE(&neglist->nl_list, ncp, nc_dst); TAILQ_INSERT_TAIL(&ncneg_hot.nl_list, ncp, nc_dst); ncp->nc_flag |= NCF_HOTNEGATIVE; } mtx_unlock(&neglist->nl_lock); mtx_unlock(&ncneg_hot.nl_lock); } static void cache_negative_insert(struct namecache *ncp, bool neg_locked) { struct neglist *neglist; MPASS(ncp->nc_flag & NCF_NEGATIVE); cache_assert_bucket_locked(ncp, RA_WLOCKED); neglist = NCP2NEGLIST(ncp); if (!neg_locked) { mtx_lock(&neglist->nl_lock); } else { mtx_assert(&neglist->nl_lock, MA_OWNED); } TAILQ_INSERT_TAIL(&neglist->nl_list, ncp, nc_dst); if (!neg_locked) mtx_unlock(&neglist->nl_lock); atomic_add_rel_long(&numneg, 1); } static void cache_negative_remove(struct namecache *ncp, bool neg_locked) { struct neglist *neglist; bool hot_locked = false; bool list_locked = false; MPASS(ncp->nc_flag & NCF_NEGATIVE); cache_assert_bucket_locked(ncp, RA_WLOCKED); neglist = NCP2NEGLIST(ncp); if (!neg_locked) { if (ncp->nc_flag & NCF_HOTNEGATIVE) { hot_locked = true; mtx_lock(&ncneg_hot.nl_lock); if (!(ncp->nc_flag & NCF_HOTNEGATIVE)) { list_locked = true; mtx_lock(&neglist->nl_lock); } } else { list_locked = true; mtx_lock(&neglist->nl_lock); } } if (ncp->nc_flag & NCF_HOTNEGATIVE) { mtx_assert(&ncneg_hot.nl_lock, MA_OWNED); TAILQ_REMOVE(&ncneg_hot.nl_list, ncp, nc_dst); } else { mtx_assert(&neglist->nl_lock, MA_OWNED); TAILQ_REMOVE(&neglist->nl_list, ncp, nc_dst); } if (list_locked) mtx_unlock(&neglist->nl_lock); if (hot_locked) mtx_unlock(&ncneg_hot.nl_lock); atomic_subtract_rel_long(&numneg, 1); } static void cache_negative_shrink_select(int start, struct namecache **ncpp, struct neglist **neglistpp) { struct neglist *neglist; struct namecache *ncp; int i; *ncpp = ncp = NULL; for (i = start; i < numneglists; i++) { neglist = &neglists[i]; if (TAILQ_FIRST(&neglist->nl_list) == NULL) continue; mtx_lock(&neglist->nl_lock); ncp = TAILQ_FIRST(&neglist->nl_list); if (ncp != NULL) break; mtx_unlock(&neglist->nl_lock); } *neglistpp = neglist; *ncpp = ncp; } static void cache_negative_zap_one(void) { struct namecache *ncp, *ncp2; struct neglist *neglist; struct mtx *dvlp; struct rwlock *blp; if (!mtx_trylock(&ncneg_shrink_lock)) return; mtx_lock(&ncneg_hot.nl_lock); ncp = TAILQ_FIRST(&ncneg_hot.nl_list); if (ncp != NULL) { neglist = NCP2NEGLIST(ncp); mtx_lock(&neglist->nl_lock); TAILQ_REMOVE(&ncneg_hot.nl_list, ncp, nc_dst); TAILQ_INSERT_TAIL(&neglist->nl_list, ncp, nc_dst); ncp->nc_flag &= ~NCF_HOTNEGATIVE; mtx_unlock(&neglist->nl_lock); } cache_negative_shrink_select(shrink_list_turn, &ncp, &neglist); shrink_list_turn++; if (shrink_list_turn == numneglists) shrink_list_turn = 0; if (ncp == NULL && shrink_list_turn == 0) cache_negative_shrink_select(shrink_list_turn, &ncp, &neglist); if (ncp == NULL) { mtx_unlock(&ncneg_hot.nl_lock); goto out; } MPASS(ncp->nc_flag & NCF_NEGATIVE); dvlp = VP2VNODELOCK(ncp->nc_dvp); blp = NCP2BUCKETLOCK(ncp); mtx_unlock(&neglist->nl_lock); mtx_unlock(&ncneg_hot.nl_lock); mtx_lock(dvlp); rw_wlock(blp); mtx_lock(&neglist->nl_lock); ncp2 = TAILQ_FIRST(&neglist->nl_list); if (ncp != ncp2 || dvlp != VP2VNODELOCK(ncp2->nc_dvp) || blp != NCP2BUCKETLOCK(ncp2) || !(ncp2->nc_flag & NCF_NEGATIVE)) { ncp = NULL; goto out_unlock_all; } SDT_PROBE3(vfs, namecache, shrink_negative, done, ncp->nc_dvp, ncp->nc_name, ncp->nc_neghits); cache_zap_locked(ncp, true); out_unlock_all: mtx_unlock(&neglist->nl_lock); rw_wunlock(blp); mtx_unlock(dvlp); out: mtx_unlock(&ncneg_shrink_lock); cache_free(ncp); } /* * cache_zap_locked(): * * Removes a namecache entry from cache, whether it contains an actual * pointer to a vnode or if it is just a negative cache entry. */ static void cache_zap_locked(struct namecache *ncp, bool neg_locked) { if (!(ncp->nc_flag & NCF_NEGATIVE)) cache_assert_vnode_locked(ncp->nc_vp); cache_assert_vnode_locked(ncp->nc_dvp); cache_assert_bucket_locked(ncp, RA_WLOCKED); CTR2(KTR_VFS, "cache_zap(%p) vp %p", ncp, (ncp->nc_flag & NCF_NEGATIVE) ? NULL : ncp->nc_vp); if (!(ncp->nc_flag & NCF_NEGATIVE)) { SDT_PROBE3(vfs, namecache, zap, done, ncp->nc_dvp, ncp->nc_name, ncp->nc_vp); } else { SDT_PROBE3(vfs, namecache, zap_negative, done, ncp->nc_dvp, ncp->nc_name, ncp->nc_neghits); } LIST_REMOVE(ncp, nc_hash); if (!(ncp->nc_flag & NCF_NEGATIVE)) { TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, ncp, nc_dst); if (ncp == ncp->nc_vp->v_cache_dd) ncp->nc_vp->v_cache_dd = NULL; } else { cache_negative_remove(ncp, neg_locked); } if (ncp->nc_flag & NCF_ISDOTDOT) { if (ncp == ncp->nc_dvp->v_cache_dd) ncp->nc_dvp->v_cache_dd = NULL; } else { LIST_REMOVE(ncp, nc_src); if (LIST_EMPTY(&ncp->nc_dvp->v_cache_src)) { ncp->nc_flag |= NCF_DVDROP; atomic_subtract_rel_long(&numcachehv, 1); } } atomic_subtract_rel_long(&numcache, 1); } static void cache_zap_negative_locked_vnode_kl(struct namecache *ncp, struct vnode *vp) { struct rwlock *blp; MPASS(ncp->nc_dvp == vp); MPASS(ncp->nc_flag & NCF_NEGATIVE); cache_assert_vnode_locked(vp); blp = NCP2BUCKETLOCK(ncp); rw_wlock(blp); cache_zap_locked(ncp, false); rw_wunlock(blp); } static bool cache_zap_locked_vnode_kl2(struct namecache *ncp, struct vnode *vp, struct mtx **vlpp) { struct mtx *pvlp, *vlp1, *vlp2, *to_unlock; struct rwlock *blp; MPASS(vp == ncp->nc_dvp || vp == ncp->nc_vp); cache_assert_vnode_locked(vp); if (ncp->nc_flag & NCF_NEGATIVE) { if (*vlpp != NULL) { mtx_unlock(*vlpp); *vlpp = NULL; } cache_zap_negative_locked_vnode_kl(ncp, vp); return (true); } pvlp = VP2VNODELOCK(vp); blp = NCP2BUCKETLOCK(ncp); vlp1 = VP2VNODELOCK(ncp->nc_dvp); vlp2 = VP2VNODELOCK(ncp->nc_vp); if (*vlpp == vlp1 || *vlpp == vlp2) { to_unlock = *vlpp; *vlpp = NULL; } else { if (*vlpp != NULL) { mtx_unlock(*vlpp); *vlpp = NULL; } cache_sort(&vlp1, &vlp2); if (vlp1 == pvlp) { mtx_lock(vlp2); to_unlock = vlp2; } else { if (!mtx_trylock(vlp1)) goto out_relock; to_unlock = vlp1; } } rw_wlock(blp); cache_zap_locked(ncp, false); rw_wunlock(blp); if (to_unlock != NULL) mtx_unlock(to_unlock); return (true); out_relock: mtx_unlock(vlp2); mtx_lock(vlp1); mtx_lock(vlp2); MPASS(*vlpp == NULL); *vlpp = vlp1; return (false); } static int cache_zap_locked_vnode(struct namecache *ncp, struct vnode *vp) { struct mtx *pvlp, *vlp1, *vlp2, *to_unlock; struct rwlock *blp; int error = 0; MPASS(vp == ncp->nc_dvp || vp == ncp->nc_vp); cache_assert_vnode_locked(vp); pvlp = VP2VNODELOCK(vp); if (ncp->nc_flag & NCF_NEGATIVE) { cache_zap_negative_locked_vnode_kl(ncp, vp); goto out; } blp = NCP2BUCKETLOCK(ncp); vlp1 = VP2VNODELOCK(ncp->nc_dvp); vlp2 = VP2VNODELOCK(ncp->nc_vp); cache_sort(&vlp1, &vlp2); if (vlp1 == pvlp) { mtx_lock(vlp2); to_unlock = vlp2; } else { if (!mtx_trylock(vlp1)) { error = EAGAIN; goto out; } to_unlock = vlp1; } rw_wlock(blp); cache_zap_locked(ncp, false); rw_wunlock(blp); mtx_unlock(to_unlock); out: mtx_unlock(pvlp); return (error); } static int +cache_zap_wlocked_bucket(struct namecache *ncp, struct rwlock *blp) +{ + struct mtx *dvlp, *vlp; + + cache_assert_bucket_locked(ncp, RA_WLOCKED); + + dvlp = VP2VNODELOCK(ncp->nc_dvp); + vlp = NULL; + if (!(ncp->nc_flag & NCF_NEGATIVE)) + vlp = VP2VNODELOCK(ncp->nc_vp); + if (cache_trylock_vnodes(dvlp, vlp) == 0) { + cache_zap_locked(ncp, false); + rw_wunlock(blp); + cache_unlock_vnodes(dvlp, vlp); + return (0); + } + + rw_wunlock(blp); + return (EAGAIN); +} + +static int cache_zap_rlocked_bucket(struct namecache *ncp, struct rwlock *blp) { struct mtx *dvlp, *vlp; cache_assert_bucket_locked(ncp, RA_RLOCKED); dvlp = VP2VNODELOCK(ncp->nc_dvp); vlp = NULL; if (!(ncp->nc_flag & NCF_NEGATIVE)) vlp = VP2VNODELOCK(ncp->nc_vp); if (cache_trylock_vnodes(dvlp, vlp) == 0) { rw_runlock(blp); rw_wlock(blp); cache_zap_locked(ncp, false); rw_wunlock(blp); cache_unlock_vnodes(dvlp, vlp); return (0); } rw_runlock(blp); return (EAGAIN); } static int cache_zap_wlocked_bucket_kl(struct namecache *ncp, struct rwlock *blp, struct mtx **vlpp1, struct mtx **vlpp2) { struct mtx *dvlp, *vlp; cache_assert_bucket_locked(ncp, RA_WLOCKED); dvlp = VP2VNODELOCK(ncp->nc_dvp); vlp = NULL; if (!(ncp->nc_flag & NCF_NEGATIVE)) vlp = VP2VNODELOCK(ncp->nc_vp); cache_sort(&dvlp, &vlp); if (*vlpp1 == dvlp && *vlpp2 == vlp) { cache_zap_locked(ncp, false); cache_unlock_vnodes(dvlp, vlp); *vlpp1 = NULL; *vlpp2 = NULL; return (0); } if (*vlpp1 != NULL) mtx_unlock(*vlpp1); if (*vlpp2 != NULL) mtx_unlock(*vlpp2); *vlpp1 = NULL; *vlpp2 = NULL; if (cache_trylock_vnodes(dvlp, vlp) == 0) { cache_zap_locked(ncp, false); cache_unlock_vnodes(dvlp, vlp); return (0); } rw_wunlock(blp); *vlpp1 = dvlp; *vlpp2 = vlp; if (*vlpp1 != NULL) mtx_lock(*vlpp1); mtx_lock(*vlpp2); rw_wlock(blp); return (EAGAIN); } static void cache_lookup_unlock(struct rwlock *blp, struct mtx *vlp) { if (blp != NULL) { rw_runlock(blp); } else { mtx_unlock(vlp); } } static int __noinline cache_lookup_dot(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct timespec *tsp, int *ticksp) { int ltype; *vpp = dvp; CTR2(KTR_VFS, "cache_lookup(%p, %s) found via .", dvp, cnp->cn_nameptr); counter_u64_add(dothits, 1); SDT_PROBE3(vfs, namecache, lookup, hit, dvp, ".", *vpp); if (tsp != NULL) timespecclear(tsp); if (ticksp != NULL) *ticksp = ticks; vrefact(*vpp); /* * When we lookup "." we still can be asked to lock it * differently... */ ltype = cnp->cn_lkflags & LK_TYPE_MASK; if (ltype != VOP_ISLOCKED(*vpp)) { if (ltype == LK_EXCLUSIVE) { vn_lock(*vpp, LK_UPGRADE | LK_RETRY); if ((*vpp)->v_iflag & VI_DOOMED) { /* forced unmount */ vrele(*vpp); *vpp = NULL; return (ENOENT); } } else vn_lock(*vpp, LK_DOWNGRADE | LK_RETRY); } return (-1); } /* * Lookup an entry in the cache * * Lookup is called with dvp pointing to the directory to search, * cnp pointing to the name of the entry being sought. If the lookup * succeeds, the vnode is returned in *vpp, and a status of -1 is * returned. If the lookup determines that the name does not exist * (negative caching), a status of ENOENT is returned. If the lookup * fails, a status of zero is returned. If the directory vnode is * recycled out from under us due to a forced unmount, a status of * ENOENT is returned. * * vpp is locked and ref'd on return. If we're looking up DOTDOT, dvp is * unlocked. If we're looking up . an extra ref is taken, but the lock is * not recursively acquired. */ static __noinline int cache_lookup_nomakeentry(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct timespec *tsp, int *ticksp) { struct namecache *ncp; struct rwlock *blp; struct mtx *dvlp, *dvlp2; uint32_t hash; int error; if (cnp->cn_namelen == 2 && cnp->cn_nameptr[0] == '.' && cnp->cn_nameptr[1] == '.') { counter_u64_add(dotdothits, 1); dvlp = VP2VNODELOCK(dvp); dvlp2 = NULL; mtx_lock(dvlp); retry_dotdot: ncp = dvp->v_cache_dd; if (ncp == NULL) { SDT_PROBE3(vfs, namecache, lookup, miss, dvp, "..", NULL); mtx_unlock(dvlp); return (0); } if ((ncp->nc_flag & NCF_ISDOTDOT) != 0) { if (ncp->nc_dvp != dvp) panic("dvp %p v_cache_dd %p\n", dvp, ncp); if (!cache_zap_locked_vnode_kl2(ncp, dvp, &dvlp2)) goto retry_dotdot; MPASS(dvp->v_cache_dd == NULL); mtx_unlock(dvlp); if (dvlp2 != NULL) mtx_unlock(dvlp2); cache_free(ncp); } else { dvp->v_cache_dd = NULL; mtx_unlock(dvlp); if (dvlp2 != NULL) mtx_unlock(dvlp2); } return (0); } hash = cache_get_hash(cnp->cn_nameptr, cnp->cn_namelen, dvp); blp = HASH2BUCKETLOCK(hash); retry: if (LIST_EMPTY(NCHHASH(hash))) goto out_no_entry; - rw_rlock(blp); + rw_wlock(blp); LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) { counter_u64_add(numchecks, 1); if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen && !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen)) break; } /* We failed to find an entry */ if (ncp == NULL) { - rw_runlock(blp); + rw_wunlock(blp); goto out_no_entry; } counter_u64_add(numposzaps, 1); - error = cache_zap_rlocked_bucket(ncp, blp); + error = cache_zap_wlocked_bucket(ncp, blp); if (error != 0) { zap_and_exit_bucket_fail++; cache_maybe_yield(); goto retry; } cache_free(ncp); return (0); out_no_entry: SDT_PROBE3(vfs, namecache, lookup, miss, dvp, cnp->cn_nameptr, NULL); counter_u64_add(nummisszap, 1); return (0); } int cache_lookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct timespec *tsp, int *ticksp) { struct namecache_ts *ncp_ts; struct namecache *ncp; struct rwlock *blp; struct mtx *dvlp, *dvlp2; uint32_t hash; int error, ltype; if (__predict_false(!doingcache)) { cnp->cn_flags &= ~MAKEENTRY; return (0); } counter_u64_add(numcalls, 1); if (__predict_false(cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.')) return (cache_lookup_dot(dvp, vpp, cnp, tsp, ticksp)); if ((cnp->cn_flags & MAKEENTRY) == 0) return (cache_lookup_nomakeentry(dvp, vpp, cnp, tsp, ticksp)); retry: blp = NULL; error = 0; if (cnp->cn_namelen == 2 && cnp->cn_nameptr[0] == '.' && cnp->cn_nameptr[1] == '.') { counter_u64_add(dotdothits, 1); dvlp = VP2VNODELOCK(dvp); dvlp2 = NULL; mtx_lock(dvlp); ncp = dvp->v_cache_dd; if (ncp == NULL) { SDT_PROBE3(vfs, namecache, lookup, miss, dvp, "..", NULL); mtx_unlock(dvlp); return (0); } if ((ncp->nc_flag & NCF_ISDOTDOT) != 0) { if (ncp->nc_flag & NCF_NEGATIVE) *vpp = NULL; else *vpp = ncp->nc_vp; } else *vpp = ncp->nc_dvp; /* Return failure if negative entry was found. */ if (*vpp == NULL) goto negative_success; CTR3(KTR_VFS, "cache_lookup(%p, %s) found %p via ..", dvp, cnp->cn_nameptr, *vpp); SDT_PROBE3(vfs, namecache, lookup, hit, dvp, "..", *vpp); cache_out_ts(ncp, tsp, ticksp); if ((ncp->nc_flag & (NCF_ISDOTDOT | NCF_DTS)) == NCF_DTS && tsp != NULL) { ncp_ts = __containerof(ncp, struct namecache_ts, nc_nc); *tsp = ncp_ts->nc_dotdottime; } goto success; } hash = cache_get_hash(cnp->cn_nameptr, cnp->cn_namelen, dvp); blp = HASH2BUCKETLOCK(hash); rw_rlock(blp); LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) { counter_u64_add(numchecks, 1); if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen && !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen)) break; } /* We failed to find an entry */ if (ncp == NULL) { rw_runlock(blp); SDT_PROBE3(vfs, namecache, lookup, miss, dvp, cnp->cn_nameptr, NULL); counter_u64_add(nummiss, 1); return (0); } /* We found a "positive" match, return the vnode */ if (!(ncp->nc_flag & NCF_NEGATIVE)) { counter_u64_add(numposhits, 1); *vpp = ncp->nc_vp; CTR4(KTR_VFS, "cache_lookup(%p, %s) found %p via ncp %p", dvp, cnp->cn_nameptr, *vpp, ncp); SDT_PROBE3(vfs, namecache, lookup, hit, dvp, ncp->nc_name, *vpp); cache_out_ts(ncp, tsp, ticksp); goto success; } negative_success: /* We found a negative match, and want to create it, so purge */ if (cnp->cn_nameiop == CREATE) { counter_u64_add(numnegzaps, 1); goto zap_and_exit; } counter_u64_add(numneghits, 1); cache_negative_hit(ncp); if (ncp->nc_flag & NCF_WHITE) cnp->cn_flags |= ISWHITEOUT; SDT_PROBE2(vfs, namecache, lookup, hit__negative, dvp, ncp->nc_name); cache_out_ts(ncp, tsp, ticksp); cache_lookup_unlock(blp, dvlp); return (ENOENT); success: /* * On success we return a locked and ref'd vnode as per the lookup * protocol. */ MPASS(dvp != *vpp); ltype = 0; /* silence gcc warning */ if (cnp->cn_flags & ISDOTDOT) { ltype = VOP_ISLOCKED(dvp); VOP_UNLOCK(dvp, 0); } vhold(*vpp); cache_lookup_unlock(blp, dvlp); error = vget(*vpp, cnp->cn_lkflags | LK_VNHELD, cnp->cn_thread); if (cnp->cn_flags & ISDOTDOT) { vn_lock(dvp, ltype | LK_RETRY); if (dvp->v_iflag & VI_DOOMED) { if (error == 0) vput(*vpp); *vpp = NULL; return (ENOENT); } } if (error) { *vpp = NULL; goto retry; } if ((cnp->cn_flags & ISLASTCN) && (cnp->cn_lkflags & LK_TYPE_MASK) == LK_EXCLUSIVE) { ASSERT_VOP_ELOCKED(*vpp, "cache_lookup"); } return (-1); zap_and_exit: if (blp != NULL) error = cache_zap_rlocked_bucket(ncp, blp); else error = cache_zap_locked_vnode(ncp, dvp); if (error != 0) { zap_and_exit_bucket_fail++; cache_maybe_yield(); goto retry; } cache_free(ncp); return (0); } struct celockstate { struct mtx *vlp[3]; struct rwlock *blp[2]; }; CTASSERT((nitems(((struct celockstate *)0)->vlp) == 3)); CTASSERT((nitems(((struct celockstate *)0)->blp) == 2)); static inline void cache_celockstate_init(struct celockstate *cel) { bzero(cel, sizeof(*cel)); } static void cache_lock_vnodes_cel(struct celockstate *cel, struct vnode *vp, struct vnode *dvp) { struct mtx *vlp1, *vlp2; MPASS(cel->vlp[0] == NULL); MPASS(cel->vlp[1] == NULL); MPASS(cel->vlp[2] == NULL); MPASS(vp != NULL || dvp != NULL); vlp1 = VP2VNODELOCK(vp); vlp2 = VP2VNODELOCK(dvp); cache_sort(&vlp1, &vlp2); if (vlp1 != NULL) { mtx_lock(vlp1); cel->vlp[0] = vlp1; } mtx_lock(vlp2); cel->vlp[1] = vlp2; } static void cache_unlock_vnodes_cel(struct celockstate *cel) { MPASS(cel->vlp[0] != NULL || cel->vlp[1] != NULL); if (cel->vlp[0] != NULL) mtx_unlock(cel->vlp[0]); if (cel->vlp[1] != NULL) mtx_unlock(cel->vlp[1]); if (cel->vlp[2] != NULL) mtx_unlock(cel->vlp[2]); } static bool cache_lock_vnodes_cel_3(struct celockstate *cel, struct vnode *vp) { struct mtx *vlp; bool ret; cache_assert_vlp_locked(cel->vlp[0]); cache_assert_vlp_locked(cel->vlp[1]); MPASS(cel->vlp[2] == NULL); MPASS(vp != NULL); vlp = VP2VNODELOCK(vp); ret = true; if (vlp >= cel->vlp[1]) { mtx_lock(vlp); } else { if (mtx_trylock(vlp)) goto out; cache_lock_vnodes_cel_3_failures++; cache_unlock_vnodes_cel(cel); if (vlp < cel->vlp[0]) { mtx_lock(vlp); mtx_lock(cel->vlp[0]); mtx_lock(cel->vlp[1]); } else { if (cel->vlp[0] != NULL) mtx_lock(cel->vlp[0]); mtx_lock(vlp); mtx_lock(cel->vlp[1]); } ret = false; } out: cel->vlp[2] = vlp; return (ret); } static void cache_lock_buckets_cel(struct celockstate *cel, struct rwlock *blp1, struct rwlock *blp2) { MPASS(cel->blp[0] == NULL); MPASS(cel->blp[1] == NULL); cache_sort(&blp1, &blp2); if (blp1 != NULL) { rw_wlock(blp1); cel->blp[0] = blp1; } rw_wlock(blp2); cel->blp[1] = blp2; } static void cache_unlock_buckets_cel(struct celockstate *cel) { if (cel->blp[0] != NULL) rw_wunlock(cel->blp[0]); rw_wunlock(cel->blp[1]); } /* * Lock part of the cache affected by the insertion. * * This means vnodelocks for dvp, vp and the relevant bucketlock. * However, insertion can result in removal of an old entry. In this * case we have an additional vnode and bucketlock pair to lock. If the * entry is negative, ncelock is locked instead of the vnode. * * That is, in the worst case we have to lock 3 vnodes and 2 bucketlocks, while * preserving the locking order (smaller address first). */ static void cache_enter_lock(struct celockstate *cel, struct vnode *dvp, struct vnode *vp, uint32_t hash) { struct namecache *ncp; struct rwlock *blps[2]; blps[0] = HASH2BUCKETLOCK(hash); for (;;) { blps[1] = NULL; cache_lock_vnodes_cel(cel, dvp, vp); if (vp == NULL || vp->v_type != VDIR) break; ncp = vp->v_cache_dd; if (ncp == NULL) break; if ((ncp->nc_flag & NCF_ISDOTDOT) == 0) break; MPASS(ncp->nc_dvp == vp); blps[1] = NCP2BUCKETLOCK(ncp); if (ncp->nc_flag & NCF_NEGATIVE) break; if (cache_lock_vnodes_cel_3(cel, ncp->nc_vp)) break; /* * All vnodes got re-locked. Re-validate the state and if * nothing changed we are done. Otherwise restart. */ if (ncp == vp->v_cache_dd && (ncp->nc_flag & NCF_ISDOTDOT) != 0 && blps[1] == NCP2BUCKETLOCK(ncp) && VP2VNODELOCK(ncp->nc_vp) == cel->vlp[2]) break; cache_unlock_vnodes_cel(cel); cel->vlp[0] = NULL; cel->vlp[1] = NULL; cel->vlp[2] = NULL; } cache_lock_buckets_cel(cel, blps[0], blps[1]); } static void cache_enter_lock_dd(struct celockstate *cel, struct vnode *dvp, struct vnode *vp, uint32_t hash) { struct namecache *ncp; struct rwlock *blps[2]; blps[0] = HASH2BUCKETLOCK(hash); for (;;) { blps[1] = NULL; cache_lock_vnodes_cel(cel, dvp, vp); ncp = dvp->v_cache_dd; if (ncp == NULL) break; if ((ncp->nc_flag & NCF_ISDOTDOT) == 0) break; MPASS(ncp->nc_dvp == dvp); blps[1] = NCP2BUCKETLOCK(ncp); if (ncp->nc_flag & NCF_NEGATIVE) break; if (cache_lock_vnodes_cel_3(cel, ncp->nc_vp)) break; if (ncp == dvp->v_cache_dd && (ncp->nc_flag & NCF_ISDOTDOT) != 0 && blps[1] == NCP2BUCKETLOCK(ncp) && VP2VNODELOCK(ncp->nc_vp) == cel->vlp[2]) break; cache_unlock_vnodes_cel(cel); cel->vlp[0] = NULL; cel->vlp[1] = NULL; cel->vlp[2] = NULL; } cache_lock_buckets_cel(cel, blps[0], blps[1]); } static void cache_enter_unlock(struct celockstate *cel) { cache_unlock_buckets_cel(cel); cache_unlock_vnodes_cel(cel); } /* * Add an entry to the cache. */ void cache_enter_time(struct vnode *dvp, struct vnode *vp, struct componentname *cnp, struct timespec *tsp, struct timespec *dtsp) { struct celockstate cel; struct namecache *ncp, *n2, *ndd; struct namecache_ts *ncp_ts, *n2_ts; struct nchashhead *ncpp; struct neglist *neglist; uint32_t hash; int flag; int len; bool neg_locked; + int lnumcache; CTR3(KTR_VFS, "cache_enter(%p, %p, %s)", dvp, vp, cnp->cn_nameptr); VNASSERT(vp == NULL || (vp->v_iflag & VI_DOOMED) == 0, vp, ("cache_enter: Adding a doomed vnode")); VNASSERT(dvp == NULL || (dvp->v_iflag & VI_DOOMED) == 0, dvp, ("cache_enter: Doomed vnode used as src")); if (__predict_false(!doingcache)) return; /* * Avoid blowout in namecache entries. */ if (__predict_false(numcache >= desiredvnodes * ncsizefactor)) return; cache_celockstate_init(&cel); ndd = NULL; flag = 0; if (cnp->cn_nameptr[0] == '.') { if (cnp->cn_namelen == 1) return; if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') { len = cnp->cn_namelen; hash = cache_get_hash(cnp->cn_nameptr, len, dvp); cache_enter_lock_dd(&cel, dvp, vp, hash); /* * If dotdot entry already exists, just retarget it * to new parent vnode, otherwise continue with new * namecache entry allocation. */ if ((ncp = dvp->v_cache_dd) != NULL && ncp->nc_flag & NCF_ISDOTDOT) { KASSERT(ncp->nc_dvp == dvp, ("wrong isdotdot parent")); neg_locked = false; if (ncp->nc_flag & NCF_NEGATIVE || vp == NULL) { neglist = NCP2NEGLIST(ncp); mtx_lock(&ncneg_hot.nl_lock); mtx_lock(&neglist->nl_lock); neg_locked = true; } if (!(ncp->nc_flag & NCF_NEGATIVE)) { TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, ncp, nc_dst); } else { cache_negative_remove(ncp, true); } if (vp != NULL) { TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst); ncp->nc_flag &= ~(NCF_NEGATIVE|NCF_HOTNEGATIVE); } else { ncp->nc_flag &= ~(NCF_HOTNEGATIVE); ncp->nc_flag |= NCF_NEGATIVE; cache_negative_insert(ncp, true); } if (neg_locked) { mtx_unlock(&neglist->nl_lock); mtx_unlock(&ncneg_hot.nl_lock); } ncp->nc_vp = vp; cache_enter_unlock(&cel); return; } dvp->v_cache_dd = NULL; cache_enter_unlock(&cel); cache_celockstate_init(&cel); SDT_PROBE3(vfs, namecache, enter, done, dvp, "..", vp); flag = NCF_ISDOTDOT; } } /* * Calculate the hash key and setup as much of the new * namecache entry as possible before acquiring the lock. */ ncp = cache_alloc(cnp->cn_namelen, tsp != NULL); ncp->nc_flag = flag; ncp->nc_vp = vp; if (vp == NULL) ncp->nc_flag |= NCF_NEGATIVE; ncp->nc_dvp = dvp; if (tsp != NULL) { ncp_ts = __containerof(ncp, struct namecache_ts, nc_nc); ncp_ts->nc_time = *tsp; ncp_ts->nc_ticks = ticks; ncp_ts->nc_nc.nc_flag |= NCF_TS; if (dtsp != NULL) { ncp_ts->nc_dotdottime = *dtsp; ncp_ts->nc_nc.nc_flag |= NCF_DTS; } } len = ncp->nc_nlen = cnp->cn_namelen; hash = cache_get_hash(cnp->cn_nameptr, len, dvp); strlcpy(ncp->nc_name, cnp->cn_nameptr, len + 1); cache_enter_lock(&cel, dvp, vp, hash); /* * See if this vnode or negative entry is already in the cache * with this name. This can happen with concurrent lookups of * the same path name. */ ncpp = NCHHASH(hash); LIST_FOREACH(n2, ncpp, nc_hash) { if (n2->nc_dvp == dvp && n2->nc_nlen == cnp->cn_namelen && !bcmp(n2->nc_name, cnp->cn_nameptr, n2->nc_nlen)) { if (tsp != NULL) { KASSERT((n2->nc_flag & NCF_TS) != 0, ("no NCF_TS")); n2_ts = __containerof(n2, struct namecache_ts, nc_nc); n2_ts->nc_time = ncp_ts->nc_time; n2_ts->nc_ticks = ncp_ts->nc_ticks; if (dtsp != NULL) { n2_ts->nc_dotdottime = ncp_ts->nc_dotdottime; if (ncp->nc_flag & NCF_NEGATIVE) mtx_lock(&ncneg_hot.nl_lock); n2_ts->nc_nc.nc_flag |= NCF_DTS; if (ncp->nc_flag & NCF_NEGATIVE) mtx_unlock(&ncneg_hot.nl_lock); } } goto out_unlock_free; } } if (flag == NCF_ISDOTDOT) { /* * See if we are trying to add .. entry, but some other lookup * has populated v_cache_dd pointer already. */ if (dvp->v_cache_dd != NULL) goto out_unlock_free; KASSERT(vp == NULL || vp->v_type == VDIR, ("wrong vnode type %p", vp)); dvp->v_cache_dd = ncp; } - atomic_add_rel_long(&numcache, 1); if (vp != NULL) { if (vp->v_type == VDIR) { if (flag != NCF_ISDOTDOT) { /* * For this case, the cache entry maps both the * directory name in it and the name ".." for the * directory's parent. */ if ((ndd = vp->v_cache_dd) != NULL) { if ((ndd->nc_flag & NCF_ISDOTDOT) != 0) cache_zap_locked(ndd, false); else ndd = NULL; } vp->v_cache_dd = ncp; } } else { vp->v_cache_dd = NULL; } } if (flag != NCF_ISDOTDOT) { if (LIST_EMPTY(&dvp->v_cache_src)) { vhold(dvp); atomic_add_rel_long(&numcachehv, 1); } LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src); } /* * Insert the new namecache entry into the appropriate chain * within the cache entries table. */ LIST_INSERT_HEAD(ncpp, ncp, nc_hash); /* * If the entry is "negative", we place it into the * "negative" cache queue, otherwise, we place it into the * destination vnode's cache entries queue. */ if (vp != NULL) { TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst); SDT_PROBE3(vfs, namecache, enter, done, dvp, ncp->nc_name, vp); } else { if (cnp->cn_flags & ISWHITEOUT) ncp->nc_flag |= NCF_WHITE; cache_negative_insert(ncp, false); SDT_PROBE2(vfs, namecache, enter_negative, done, dvp, ncp->nc_name); } cache_enter_unlock(&cel); - if (numneg * ncnegfactor > numcache) + lnumcache = atomic_fetchadd_long(&numcache, 1) + 1; + if (numneg * ncnegfactor > lnumcache) cache_negative_zap_one(); cache_free(ndd); return; out_unlock_free: cache_enter_unlock(&cel); cache_free(ncp); return; } static u_int cache_roundup_2(u_int val) { u_int res; for (res = 1; res <= val; res <<= 1) continue; return (res); } /* * Name cache initialization, from vfs_init() when we are booting */ static void nchinit(void *dummy __unused) { u_int i; cache_zone_small = uma_zcreate("S VFS Cache", sizeof(struct namecache) + CACHE_PATH_CUTOFF + 1, NULL, NULL, NULL, NULL, UMA_ALIGNOF(struct namecache), UMA_ZONE_ZINIT); cache_zone_small_ts = uma_zcreate("STS VFS Cache", sizeof(struct namecache_ts) + CACHE_PATH_CUTOFF + 1, NULL, NULL, NULL, NULL, UMA_ALIGNOF(struct namecache_ts), UMA_ZONE_ZINIT); cache_zone_large = uma_zcreate("L VFS Cache", sizeof(struct namecache) + NAME_MAX + 1, NULL, NULL, NULL, NULL, UMA_ALIGNOF(struct namecache), UMA_ZONE_ZINIT); cache_zone_large_ts = uma_zcreate("LTS VFS Cache", sizeof(struct namecache_ts) + NAME_MAX + 1, NULL, NULL, NULL, NULL, UMA_ALIGNOF(struct namecache_ts), UMA_ZONE_ZINIT); nchashtbl = hashinit(desiredvnodes * 2, M_VFSCACHE, &nchash); ncbuckethash = cache_roundup_2(mp_ncpus * 64) - 1; if (ncbuckethash > nchash) ncbuckethash = nchash; bucketlocks = malloc(sizeof(*bucketlocks) * numbucketlocks, M_VFSCACHE, M_WAITOK | M_ZERO); for (i = 0; i < numbucketlocks; i++) rw_init_flags(&bucketlocks[i], "ncbuc", RW_DUPOK | RW_RECURSE); ncvnodehash = cache_roundup_2(mp_ncpus * 64) - 1; vnodelocks = malloc(sizeof(*vnodelocks) * numvnodelocks, M_VFSCACHE, M_WAITOK | M_ZERO); for (i = 0; i < numvnodelocks; i++) mtx_init(&vnodelocks[i], "ncvn", NULL, MTX_DUPOK | MTX_RECURSE); ncpurgeminvnodes = numbucketlocks; ncneghash = 3; neglists = malloc(sizeof(*neglists) * numneglists, M_VFSCACHE, M_WAITOK | M_ZERO); for (i = 0; i < numneglists; i++) { mtx_init(&neglists[i].nl_lock, "ncnegl", NULL, MTX_DEF); TAILQ_INIT(&neglists[i].nl_list); } mtx_init(&ncneg_hot.nl_lock, "ncneglh", NULL, MTX_DEF); TAILQ_INIT(&ncneg_hot.nl_list); mtx_init(&ncneg_shrink_lock, "ncnegs", NULL, MTX_DEF); numcalls = counter_u64_alloc(M_WAITOK); dothits = counter_u64_alloc(M_WAITOK); dotdothits = counter_u64_alloc(M_WAITOK); numchecks = counter_u64_alloc(M_WAITOK); nummiss = counter_u64_alloc(M_WAITOK); nummisszap = counter_u64_alloc(M_WAITOK); numposzaps = counter_u64_alloc(M_WAITOK); numposhits = counter_u64_alloc(M_WAITOK); numnegzaps = counter_u64_alloc(M_WAITOK); numneghits = counter_u64_alloc(M_WAITOK); numfullpathcalls = counter_u64_alloc(M_WAITOK); numfullpathfail1 = counter_u64_alloc(M_WAITOK); numfullpathfail2 = counter_u64_alloc(M_WAITOK); numfullpathfail4 = counter_u64_alloc(M_WAITOK); numfullpathfound = counter_u64_alloc(M_WAITOK); } SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL); void cache_changesize(int newmaxvnodes) { struct nchashhead *new_nchashtbl, *old_nchashtbl; u_long new_nchash, old_nchash; struct namecache *ncp; uint32_t hash; int i; newmaxvnodes = cache_roundup_2(newmaxvnodes * 2); if (newmaxvnodes < numbucketlocks) newmaxvnodes = numbucketlocks; new_nchashtbl = hashinit(newmaxvnodes, M_VFSCACHE, &new_nchash); /* If same hash table size, nothing to do */ if (nchash == new_nchash) { free(new_nchashtbl, M_VFSCACHE); return; } /* * Move everything from the old hash table to the new table. * None of the namecache entries in the table can be removed * because to do so, they have to be removed from the hash table. */ cache_lock_all_vnodes(); cache_lock_all_buckets(); old_nchashtbl = nchashtbl; old_nchash = nchash; nchashtbl = new_nchashtbl; nchash = new_nchash; for (i = 0; i <= old_nchash; i++) { while ((ncp = LIST_FIRST(&old_nchashtbl[i])) != NULL) { hash = cache_get_hash(ncp->nc_name, ncp->nc_nlen, ncp->nc_dvp); LIST_REMOVE(ncp, nc_hash); LIST_INSERT_HEAD(NCHHASH(hash), ncp, nc_hash); } } cache_unlock_all_buckets(); cache_unlock_all_vnodes(); free(old_nchashtbl, M_VFSCACHE); } /* * Invalidate all entries to a particular vnode. */ void cache_purge(struct vnode *vp) { TAILQ_HEAD(, namecache) ncps; struct namecache *ncp, *nnp; struct mtx *vlp, *vlp2; CTR1(KTR_VFS, "cache_purge(%p)", vp); SDT_PROBE1(vfs, namecache, purge, done, vp); if (LIST_EMPTY(&vp->v_cache_src) && TAILQ_EMPTY(&vp->v_cache_dst) && vp->v_cache_dd == NULL) return; TAILQ_INIT(&ncps); vlp = VP2VNODELOCK(vp); vlp2 = NULL; mtx_lock(vlp); retry: while (!LIST_EMPTY(&vp->v_cache_src)) { ncp = LIST_FIRST(&vp->v_cache_src); if (!cache_zap_locked_vnode_kl2(ncp, vp, &vlp2)) goto retry; TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst); } while (!TAILQ_EMPTY(&vp->v_cache_dst)) { ncp = TAILQ_FIRST(&vp->v_cache_dst); if (!cache_zap_locked_vnode_kl2(ncp, vp, &vlp2)) goto retry; TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst); } ncp = vp->v_cache_dd; if (ncp != NULL) { KASSERT(ncp->nc_flag & NCF_ISDOTDOT, ("lost dotdot link")); if (!cache_zap_locked_vnode_kl2(ncp, vp, &vlp2)) goto retry; TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst); } KASSERT(vp->v_cache_dd == NULL, ("incomplete purge")); mtx_unlock(vlp); if (vlp2 != NULL) mtx_unlock(vlp2); TAILQ_FOREACH_SAFE(ncp, &ncps, nc_dst, nnp) { cache_free(ncp); } } /* * Invalidate all negative entries for a particular directory vnode. */ void cache_purge_negative(struct vnode *vp) { TAILQ_HEAD(, namecache) ncps; struct namecache *ncp, *nnp; struct mtx *vlp; CTR1(KTR_VFS, "cache_purge_negative(%p)", vp); SDT_PROBE1(vfs, namecache, purge_negative, done, vp); if (LIST_EMPTY(&vp->v_cache_src)) return; TAILQ_INIT(&ncps); vlp = VP2VNODELOCK(vp); mtx_lock(vlp); LIST_FOREACH_SAFE(ncp, &vp->v_cache_src, nc_src, nnp) { if (!(ncp->nc_flag & NCF_NEGATIVE)) continue; cache_zap_negative_locked_vnode_kl(ncp, vp); TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst); } mtx_unlock(vlp); TAILQ_FOREACH_SAFE(ncp, &ncps, nc_dst, nnp) { cache_free(ncp); } } /* * Flush all entries referencing a particular filesystem. */ void cache_purgevfs(struct mount *mp, bool force) { TAILQ_HEAD(, namecache) ncps; struct mtx *vlp1, *vlp2; struct rwlock *blp; struct nchashhead *bucket; struct namecache *ncp, *nnp; u_long i, j, n_nchash; int error; /* Scan hash tables for applicable entries */ SDT_PROBE1(vfs, namecache, purgevfs, done, mp); if (!force && mp->mnt_nvnodelistsize <= ncpurgeminvnodes) return; TAILQ_INIT(&ncps); n_nchash = nchash + 1; vlp1 = vlp2 = NULL; for (i = 0; i < numbucketlocks; i++) { blp = (struct rwlock *)&bucketlocks[i]; rw_wlock(blp); for (j = i; j < n_nchash; j += numbucketlocks) { retry: bucket = &nchashtbl[j]; LIST_FOREACH_SAFE(ncp, bucket, nc_hash, nnp) { cache_assert_bucket_locked(ncp, RA_WLOCKED); if (ncp->nc_dvp->v_mount != mp) continue; error = cache_zap_wlocked_bucket_kl(ncp, blp, &vlp1, &vlp2); if (error != 0) goto retry; TAILQ_INSERT_HEAD(&ncps, ncp, nc_dst); } } rw_wunlock(blp); if (vlp1 == NULL && vlp2 == NULL) cache_maybe_yield(); } if (vlp1 != NULL) mtx_unlock(vlp1); if (vlp2 != NULL) mtx_unlock(vlp2); TAILQ_FOREACH_SAFE(ncp, &ncps, nc_dst, nnp) { cache_free(ncp); } } /* * Perform canonical checks and cache lookup and pass on to filesystem * through the vop_cachedlookup only if needed. */ int vfs_cache_lookup(struct vop_lookup_args *ap) { struct vnode *dvp; int error; struct vnode **vpp = ap->a_vpp; struct componentname *cnp = ap->a_cnp; struct ucred *cred = cnp->cn_cred; int flags = cnp->cn_flags; struct thread *td = cnp->cn_thread; *vpp = NULL; dvp = ap->a_dvp; if (dvp->v_type != VDIR) return (ENOTDIR); if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) return (EROFS); error = VOP_ACCESS(dvp, VEXEC, cred, td); if (error) return (error); error = cache_lookup(dvp, vpp, cnp, NULL, NULL); if (error == 0) return (VOP_CACHEDLOOKUP(dvp, vpp, cnp)); if (error == -1) return (0); return (error); } /* * XXX All of these sysctls would probably be more productive dead. */ static int __read_mostly disablecwd; SYSCTL_INT(_debug, OID_AUTO, disablecwd, CTLFLAG_RW, &disablecwd, 0, "Disable the getcwd syscall"); /* Implementation of the getcwd syscall. */ int sys___getcwd(struct thread *td, struct __getcwd_args *uap) { return (kern___getcwd(td, uap->buf, UIO_USERSPACE, uap->buflen, MAXPATHLEN)); } int kern___getcwd(struct thread *td, char *buf, enum uio_seg bufseg, size_t buflen, size_t path_max) { char *bp, *tmpbuf; struct filedesc *fdp; struct vnode *cdir, *rdir; int error; if (__predict_false(disablecwd)) return (ENODEV); if (__predict_false(buflen < 2)) return (EINVAL); if (buflen > path_max) buflen = path_max; tmpbuf = malloc(buflen, M_TEMP, M_WAITOK); fdp = td->td_proc->p_fd; FILEDESC_SLOCK(fdp); cdir = fdp->fd_cdir; vrefact(cdir); rdir = fdp->fd_rdir; vrefact(rdir); FILEDESC_SUNLOCK(fdp); error = vn_fullpath1(td, cdir, rdir, tmpbuf, &bp, buflen); vrele(rdir); vrele(cdir); if (!error) { if (bufseg == UIO_SYSSPACE) bcopy(bp, buf, strlen(bp) + 1); else error = copyout(bp, buf, strlen(bp) + 1); #ifdef KTRACE if (KTRPOINT(curthread, KTR_NAMEI)) ktrnamei(bp); #endif } free(tmpbuf, M_TEMP); return (error); } /* * Thus begins the fullpath magic. */ static int __read_mostly disablefullpath; SYSCTL_INT(_debug, OID_AUTO, disablefullpath, CTLFLAG_RW, &disablefullpath, 0, "Disable the vn_fullpath function"); /* * Retrieve the full filesystem path that correspond to a vnode from the name * cache (if available) */ int vn_fullpath(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf) { char *buf; struct filedesc *fdp; struct vnode *rdir; int error; if (__predict_false(disablefullpath)) return (ENODEV); if (__predict_false(vn == NULL)) return (EINVAL); buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); fdp = td->td_proc->p_fd; FILEDESC_SLOCK(fdp); rdir = fdp->fd_rdir; vrefact(rdir); FILEDESC_SUNLOCK(fdp); error = vn_fullpath1(td, vn, rdir, buf, retbuf, MAXPATHLEN); vrele(rdir); if (!error) *freebuf = buf; else free(buf, M_TEMP); return (error); } /* * This function is similar to vn_fullpath, but it attempts to lookup the * pathname relative to the global root mount point. This is required for the * auditing sub-system, as audited pathnames must be absolute, relative to the * global root mount point. */ int vn_fullpath_global(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf) { char *buf; int error; if (__predict_false(disablefullpath)) return (ENODEV); if (__predict_false(vn == NULL)) return (EINVAL); buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); error = vn_fullpath1(td, vn, rootvnode, buf, retbuf, MAXPATHLEN); if (!error) *freebuf = buf; else free(buf, M_TEMP); return (error); } int vn_vptocnp(struct vnode **vp, struct ucred *cred, char *buf, u_int *buflen) { struct vnode *dvp; struct namecache *ncp; struct mtx *vlp; int error; vlp = VP2VNODELOCK(*vp); mtx_lock(vlp); TAILQ_FOREACH(ncp, &((*vp)->v_cache_dst), nc_dst) { if ((ncp->nc_flag & NCF_ISDOTDOT) == 0) break; } if (ncp != NULL) { if (*buflen < ncp->nc_nlen) { mtx_unlock(vlp); vrele(*vp); counter_u64_add(numfullpathfail4, 1); error = ENOMEM; SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL); return (error); } *buflen -= ncp->nc_nlen; memcpy(buf + *buflen, ncp->nc_name, ncp->nc_nlen); SDT_PROBE3(vfs, namecache, fullpath, hit, ncp->nc_dvp, ncp->nc_name, vp); dvp = *vp; *vp = ncp->nc_dvp; vref(*vp); mtx_unlock(vlp); vrele(dvp); return (0); } SDT_PROBE1(vfs, namecache, fullpath, miss, vp); mtx_unlock(vlp); vn_lock(*vp, LK_SHARED | LK_RETRY); error = VOP_VPTOCNP(*vp, &dvp, cred, buf, buflen); vput(*vp); if (error) { counter_u64_add(numfullpathfail2, 1); SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL); return (error); } *vp = dvp; if (dvp->v_iflag & VI_DOOMED) { /* forced unmount */ vrele(dvp); error = ENOENT; SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL); return (error); } /* * *vp has its use count incremented still. */ return (0); } /* * The magic behind kern___getcwd() and vn_fullpath(). */ static int vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir, char *buf, char **retbuf, u_int buflen) { int error, slash_prefixed; #ifdef KDTRACE_HOOKS struct vnode *startvp = vp; #endif struct vnode *vp1; buflen--; buf[buflen] = '\0'; error = 0; slash_prefixed = 0; SDT_PROBE1(vfs, namecache, fullpath, entry, vp); counter_u64_add(numfullpathcalls, 1); vref(vp); if (vp->v_type != VDIR) { error = vn_vptocnp(&vp, td->td_ucred, buf, &buflen); if (error) return (error); if (buflen == 0) { vrele(vp); return (ENOMEM); } buf[--buflen] = '/'; slash_prefixed = 1; } while (vp != rdir && vp != rootvnode) { /* * The vp vnode must be already fully constructed, * since it is either found in namecache or obtained * from VOP_VPTOCNP(). We may test for VV_ROOT safely * without obtaining the vnode lock. */ if ((vp->v_vflag & VV_ROOT) != 0) { vn_lock(vp, LK_RETRY | LK_SHARED); /* * With the vnode locked, check for races with * unmount, forced or not. Note that we * already verified that vp is not equal to * the root vnode, which means that * mnt_vnodecovered can be NULL only for the * case of unmount. */ if ((vp->v_iflag & VI_DOOMED) != 0 || (vp1 = vp->v_mount->mnt_vnodecovered) == NULL || vp1->v_mountedhere != vp->v_mount) { vput(vp); error = ENOENT; SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL); break; } vref(vp1); vput(vp); vp = vp1; continue; } if (vp->v_type != VDIR) { vrele(vp); counter_u64_add(numfullpathfail1, 1); error = ENOTDIR; SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL); break; } error = vn_vptocnp(&vp, td->td_ucred, buf, &buflen); if (error) break; if (buflen == 0) { vrele(vp); error = ENOMEM; SDT_PROBE3(vfs, namecache, fullpath, return, error, startvp, NULL); break; } buf[--buflen] = '/'; slash_prefixed = 1; } if (error) return (error); if (!slash_prefixed) { if (buflen == 0) { vrele(vp); counter_u64_add(numfullpathfail4, 1); SDT_PROBE3(vfs, namecache, fullpath, return, ENOMEM, startvp, NULL); return (ENOMEM); } buf[--buflen] = '/'; } counter_u64_add(numfullpathfound, 1); vrele(vp); SDT_PROBE3(vfs, namecache, fullpath, return, 0, startvp, buf + buflen); *retbuf = buf + buflen; return (0); } struct vnode * vn_dir_dd_ino(struct vnode *vp) { struct namecache *ncp; struct vnode *ddvp; struct mtx *vlp; ASSERT_VOP_LOCKED(vp, "vn_dir_dd_ino"); vlp = VP2VNODELOCK(vp); mtx_lock(vlp); TAILQ_FOREACH(ncp, &(vp->v_cache_dst), nc_dst) { if ((ncp->nc_flag & NCF_ISDOTDOT) != 0) continue; ddvp = ncp->nc_dvp; vhold(ddvp); mtx_unlock(vlp); if (vget(ddvp, LK_SHARED | LK_NOWAIT | LK_VNHELD, curthread)) return (NULL); return (ddvp); } mtx_unlock(vlp); return (NULL); } int vn_commname(struct vnode *vp, char *buf, u_int buflen) { struct namecache *ncp; struct mtx *vlp; int l; vlp = VP2VNODELOCK(vp); mtx_lock(vlp); TAILQ_FOREACH(ncp, &vp->v_cache_dst, nc_dst) if ((ncp->nc_flag & NCF_ISDOTDOT) == 0) break; if (ncp == NULL) { mtx_unlock(vlp); return (ENOENT); } l = min(ncp->nc_nlen, buflen - 1); memcpy(buf, ncp->nc_name, l); mtx_unlock(vlp); buf[l] = '\0'; return (0); } /* ABI compat shims for old kernel modules. */ #undef cache_enter void cache_enter(struct vnode *dvp, struct vnode *vp, struct componentname *cnp); void cache_enter(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) { cache_enter_time(dvp, vp, cnp, NULL, NULL); } /* * This function updates path string to vnode's full global path * and checks the size of the new path string against the pathlen argument. * * Requires a locked, referenced vnode. * Vnode is re-locked on success or ENODEV, otherwise unlocked. * * If sysctl debug.disablefullpath is set, ENODEV is returned, * vnode is left locked and path remain untouched. * * If vp is a directory, the call to vn_fullpath_global() always succeeds * because it falls back to the ".." lookup if the namecache lookup fails. */ int vn_path_to_global_path(struct thread *td, struct vnode *vp, char *path, u_int pathlen) { struct nameidata nd; struct vnode *vp1; char *rpath, *fbuf; int error; ASSERT_VOP_ELOCKED(vp, __func__); /* Return ENODEV if sysctl debug.disablefullpath==1 */ if (__predict_false(disablefullpath)) return (ENODEV); /* Construct global filesystem path from vp. */ VOP_UNLOCK(vp, 0); error = vn_fullpath_global(td, vp, &rpath, &fbuf); if (error != 0) { vrele(vp); return (error); } if (strlen(rpath) >= pathlen) { vrele(vp); error = ENAMETOOLONG; goto out; } /* * Re-lookup the vnode by path to detect a possible rename. * As a side effect, the vnode is relocked. * If vnode was renamed, return ENOENT. */ NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_SYSSPACE, path, td); error = namei(&nd); if (error != 0) { vrele(vp); goto out; } NDFREE(&nd, NDF_ONLY_PNBUF); vp1 = nd.ni_vp; vrele(vp); if (vp1 == vp) strcpy(path, rpath); else { vput(vp1); error = ENOENT; } out: free(fbuf, M_TEMP); return (error); } Index: projects/runtime-coverage/sys/net/iflib.c =================================================================== --- projects/runtime-coverage/sys/net/iflib.c (revision 325498) +++ projects/runtime-coverage/sys/net/iflib.c (revision 325499) @@ -1,5713 +1,5775 @@ /*- * Copyright (c) 2014-2017, Matthew Macy * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions are met: * * 1. Redistributions of source code must retain the above copyright notice, * this list of conditions and the following disclaimer. * * 2. Neither the name of Matthew Macy nor the names of its * contributors may be used to endorse or promote products derived from * this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE * POSSIBILITY OF SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_acpi.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include +#include #include #include #include #include #include #include #include #include +#include +#include #include #include #include #include #include #include #include #include #include #include "ifdi_if.h" #if defined(__i386__) || defined(__amd64__) #include #include #include #include #include #include #endif #include /* * enable accounting of every mbuf as it comes in to and goes out of * iflib's software descriptor references */ #define MEMORY_LOGGING 0 /* * Enable mbuf vectors for compressing long mbuf chains */ /* * NB: * - Prefetching in tx cleaning should perhaps be a tunable. The distance ahead * we prefetch needs to be determined by the time spent in m_free vis a vis * the cost of a prefetch. This will of course vary based on the workload: * - NFLX's m_free path is dominated by vm-based M_EXT manipulation which * is quite expensive, thus suggesting very little prefetch. * - small packet forwarding which is just returning a single mbuf to * UMA will typically be very fast vis a vis the cost of a memory * access. */ /* * File organization: * - private structures * - iflib private utility functions * - ifnet functions * - vlan registry and other exported functions * - iflib public core functions * * */ static MALLOC_DEFINE(M_IFLIB, "iflib", "ifnet library"); struct iflib_txq; typedef struct iflib_txq *iflib_txq_t; struct iflib_rxq; typedef struct iflib_rxq *iflib_rxq_t; struct iflib_fl; typedef struct iflib_fl *iflib_fl_t; struct iflib_ctx; static void iru_init(if_rxd_update_t iru, iflib_rxq_t rxq, uint8_t flid); typedef struct iflib_filter_info { driver_filter_t *ifi_filter; void *ifi_filter_arg; struct grouptask *ifi_task; void *ifi_ctx; } *iflib_filter_info_t; struct iflib_ctx { KOBJ_FIELDS; /* * Pointer to hardware driver's softc */ void *ifc_softc; device_t ifc_dev; if_t ifc_ifp; cpuset_t ifc_cpus; if_shared_ctx_t ifc_sctx; struct if_softc_ctx ifc_softc_ctx; struct mtx ifc_mtx; uint16_t ifc_nhwtxqs; uint16_t ifc_nhwrxqs; iflib_txq_t ifc_txqs; iflib_rxq_t ifc_rxqs; uint32_t ifc_if_flags; uint32_t ifc_flags; uint32_t ifc_max_fl_buf_size; int ifc_in_detach; int ifc_link_state; int ifc_link_irq; int ifc_watchdog_events; struct cdev *ifc_led_dev; struct resource *ifc_msix_mem; struct if_irq ifc_legacy_irq; struct grouptask ifc_admin_task; struct grouptask ifc_vflr_task; struct iflib_filter_info ifc_filter_info; struct ifmedia ifc_media; struct sysctl_oid *ifc_sysctl_node; uint16_t ifc_sysctl_ntxqs; uint16_t ifc_sysctl_nrxqs; uint16_t ifc_sysctl_qs_eq_override; uint16_t ifc_sysctl_rx_budget; qidx_t ifc_sysctl_ntxds[8]; qidx_t ifc_sysctl_nrxds[8]; struct if_txrx ifc_txrx; #define isc_txd_encap ifc_txrx.ift_txd_encap #define isc_txd_flush ifc_txrx.ift_txd_flush #define isc_txd_credits_update ifc_txrx.ift_txd_credits_update #define isc_rxd_available ifc_txrx.ift_rxd_available #define isc_rxd_pkt_get ifc_txrx.ift_rxd_pkt_get #define isc_rxd_refill ifc_txrx.ift_rxd_refill #define isc_rxd_flush ifc_txrx.ift_rxd_flush #define isc_rxd_refill ifc_txrx.ift_rxd_refill #define isc_rxd_refill ifc_txrx.ift_rxd_refill #define isc_legacy_intr ifc_txrx.ift_legacy_intr eventhandler_tag ifc_vlan_attach_event; eventhandler_tag ifc_vlan_detach_event; uint8_t ifc_mac[ETHER_ADDR_LEN]; char ifc_mtx_name[16]; }; void * iflib_get_softc(if_ctx_t ctx) { return (ctx->ifc_softc); } device_t iflib_get_dev(if_ctx_t ctx) { return (ctx->ifc_dev); } if_t iflib_get_ifp(if_ctx_t ctx) { return (ctx->ifc_ifp); } struct ifmedia * iflib_get_media(if_ctx_t ctx) { return (&ctx->ifc_media); } void iflib_set_mac(if_ctx_t ctx, uint8_t mac[ETHER_ADDR_LEN]) { bcopy(mac, ctx->ifc_mac, ETHER_ADDR_LEN); } if_softc_ctx_t iflib_get_softc_ctx(if_ctx_t ctx) { return (&ctx->ifc_softc_ctx); } if_shared_ctx_t iflib_get_sctx(if_ctx_t ctx) { return (ctx->ifc_sctx); } #define IP_ALIGNED(m) ((((uintptr_t)(m)->m_data) & 0x3) == 0x2) #define CACHE_PTR_INCREMENT (CACHE_LINE_SIZE/sizeof(void*)) #define CACHE_PTR_NEXT(ptr) ((void *)(((uintptr_t)(ptr)+CACHE_LINE_SIZE-1) & (CACHE_LINE_SIZE-1))) #define LINK_ACTIVE(ctx) ((ctx)->ifc_link_state == LINK_STATE_UP) #define CTX_IS_VF(ctx) ((ctx)->ifc_sctx->isc_flags & IFLIB_IS_VF) #define RX_SW_DESC_MAP_CREATED (1 << 0) #define TX_SW_DESC_MAP_CREATED (1 << 1) #define RX_SW_DESC_INUSE (1 << 3) #define TX_SW_DESC_MAPPED (1 << 4) #define M_TOOBIG M_PROTO1 typedef struct iflib_sw_rx_desc_array { bus_dmamap_t *ifsd_map; /* bus_dma maps for packet */ struct mbuf **ifsd_m; /* pkthdr mbufs */ caddr_t *ifsd_cl; /* direct cluster pointer for rx */ uint8_t *ifsd_flags; } iflib_rxsd_array_t; typedef struct iflib_sw_tx_desc_array { bus_dmamap_t *ifsd_map; /* bus_dma maps for packet */ struct mbuf **ifsd_m; /* pkthdr mbufs */ uint8_t *ifsd_flags; } if_txsd_vec_t; /* magic number that should be high enough for any hardware */ #define IFLIB_MAX_TX_SEGS 128 /* bnxt supports 64 with hardware LRO enabled */ #define IFLIB_MAX_RX_SEGS 64 #define IFLIB_RX_COPY_THRESH 128 #define IFLIB_MAX_RX_REFRESH 32 /* The minimum descriptors per second before we start coalescing */ #define IFLIB_MIN_DESC_SEC 16384 #define IFLIB_DEFAULT_TX_UPDATE_FREQ 16 #define IFLIB_QUEUE_IDLE 0 #define IFLIB_QUEUE_HUNG 1 #define IFLIB_QUEUE_WORKING 2 /* maximum number of txqs that can share an rx interrupt */ #define IFLIB_MAX_TX_SHARED_INTR 4 /* this should really scale with ring size - this is a fairly arbitrary value */ #define TX_BATCH_SIZE 32 #define IFLIB_RESTART_BUDGET 8 #define IFC_LEGACY 0x001 #define IFC_QFLUSH 0x002 #define IFC_MULTISEG 0x004 #define IFC_DMAR 0x008 #define IFC_SC_ALLOCATED 0x010 #define IFC_INIT_DONE 0x020 #define IFC_PREFETCH 0x040 #define IFC_DO_RESET 0x080 #define IFC_CHECK_HUNG 0x100 #define CSUM_OFFLOAD (CSUM_IP_TSO|CSUM_IP6_TSO|CSUM_IP| \ CSUM_IP_UDP|CSUM_IP_TCP|CSUM_IP_SCTP| \ CSUM_IP6_UDP|CSUM_IP6_TCP|CSUM_IP6_SCTP) struct iflib_txq { qidx_t ift_in_use; qidx_t ift_cidx; qidx_t ift_cidx_processed; qidx_t ift_pidx; uint8_t ift_gen; uint8_t ift_br_offset; uint16_t ift_npending; uint16_t ift_db_pending; uint16_t ift_rs_pending; /* implicit pad */ uint8_t ift_txd_size[8]; uint64_t ift_processed; uint64_t ift_cleaned; uint64_t ift_cleaned_prev; #if MEMORY_LOGGING uint64_t ift_enqueued; uint64_t ift_dequeued; #endif uint64_t ift_no_tx_dma_setup; uint64_t ift_no_desc_avail; uint64_t ift_mbuf_defrag_failed; uint64_t ift_mbuf_defrag; uint64_t ift_map_failed; uint64_t ift_txd_encap_efbig; uint64_t ift_pullups; struct mtx ift_mtx; struct mtx ift_db_mtx; /* constant values */ if_ctx_t ift_ctx; struct ifmp_ring *ift_br; struct grouptask ift_task; qidx_t ift_size; uint16_t ift_id; struct callout ift_timer; if_txsd_vec_t ift_sds; uint8_t ift_qstatus; uint8_t ift_closed; uint8_t ift_update_freq; struct iflib_filter_info ift_filter_info; bus_dma_tag_t ift_desc_tag; bus_dma_tag_t ift_tso_desc_tag; iflib_dma_info_t ift_ifdi; #define MTX_NAME_LEN 16 char ift_mtx_name[MTX_NAME_LEN]; char ift_db_mtx_name[MTX_NAME_LEN]; bus_dma_segment_t ift_segs[IFLIB_MAX_TX_SEGS] __aligned(CACHE_LINE_SIZE); #ifdef IFLIB_DIAGNOSTICS uint64_t ift_cpu_exec_count[256]; #endif } __aligned(CACHE_LINE_SIZE); struct iflib_fl { qidx_t ifl_cidx; qidx_t ifl_pidx; qidx_t ifl_credits; uint8_t ifl_gen; uint8_t ifl_rxd_size; #if MEMORY_LOGGING uint64_t ifl_m_enqueued; uint64_t ifl_m_dequeued; uint64_t ifl_cl_enqueued; uint64_t ifl_cl_dequeued; #endif /* implicit pad */ bitstr_t *ifl_rx_bitmap; qidx_t ifl_fragidx; /* constant */ qidx_t ifl_size; uint16_t ifl_buf_size; uint16_t ifl_cltype; uma_zone_t ifl_zone; iflib_rxsd_array_t ifl_sds; iflib_rxq_t ifl_rxq; uint8_t ifl_id; bus_dma_tag_t ifl_desc_tag; iflib_dma_info_t ifl_ifdi; uint64_t ifl_bus_addrs[IFLIB_MAX_RX_REFRESH] __aligned(CACHE_LINE_SIZE); caddr_t ifl_vm_addrs[IFLIB_MAX_RX_REFRESH]; qidx_t ifl_rxd_idxs[IFLIB_MAX_RX_REFRESH]; } __aligned(CACHE_LINE_SIZE); static inline qidx_t get_inuse(int size, qidx_t cidx, qidx_t pidx, uint8_t gen) { qidx_t used; if (pidx > cidx) used = pidx - cidx; else if (pidx < cidx) used = size - cidx + pidx; else if (gen == 0 && pidx == cidx) used = 0; else if (gen == 1 && pidx == cidx) used = size; else panic("bad state"); return (used); } #define TXQ_AVAIL(txq) (txq->ift_size - get_inuse(txq->ift_size, txq->ift_cidx, txq->ift_pidx, txq->ift_gen)) #define IDXDIFF(head, tail, wrap) \ ((head) >= (tail) ? (head) - (tail) : (wrap) - (tail) + (head)) struct iflib_rxq { /* If there is a separate completion queue - * these are the cq cidx and pidx. Otherwise * these are unused. */ qidx_t ifr_size; qidx_t ifr_cq_cidx; qidx_t ifr_cq_pidx; uint8_t ifr_cq_gen; uint8_t ifr_fl_offset; if_ctx_t ifr_ctx; iflib_fl_t ifr_fl; uint64_t ifr_rx_irq; uint16_t ifr_id; uint8_t ifr_lro_enabled; uint8_t ifr_nfl; uint8_t ifr_ntxqirq; uint8_t ifr_txqid[IFLIB_MAX_TX_SHARED_INTR]; struct lro_ctrl ifr_lc; struct grouptask ifr_task; struct iflib_filter_info ifr_filter_info; iflib_dma_info_t ifr_ifdi; /* dynamically allocate if any drivers need a value substantially larger than this */ struct if_rxd_frag ifr_frags[IFLIB_MAX_RX_SEGS] __aligned(CACHE_LINE_SIZE); #ifdef IFLIB_DIAGNOSTICS uint64_t ifr_cpu_exec_count[256]; #endif } __aligned(CACHE_LINE_SIZE); typedef struct if_rxsd { caddr_t *ifsd_cl; struct mbuf **ifsd_m; iflib_fl_t ifsd_fl; qidx_t ifsd_cidx; } *if_rxsd_t; /* multiple of word size */ #ifdef __LP64__ #define PKT_INFO_SIZE 6 #define RXD_INFO_SIZE 5 #define PKT_TYPE uint64_t #else #define PKT_INFO_SIZE 11 #define RXD_INFO_SIZE 8 #define PKT_TYPE uint32_t #endif #define PKT_LOOP_BOUND ((PKT_INFO_SIZE/3)*3) #define RXD_LOOP_BOUND ((RXD_INFO_SIZE/4)*4) typedef struct if_pkt_info_pad { PKT_TYPE pkt_val[PKT_INFO_SIZE]; } *if_pkt_info_pad_t; typedef struct if_rxd_info_pad { PKT_TYPE rxd_val[RXD_INFO_SIZE]; } *if_rxd_info_pad_t; CTASSERT(sizeof(struct if_pkt_info_pad) == sizeof(struct if_pkt_info)); CTASSERT(sizeof(struct if_rxd_info_pad) == sizeof(struct if_rxd_info)); static inline void pkt_info_zero(if_pkt_info_t pi) { if_pkt_info_pad_t pi_pad; pi_pad = (if_pkt_info_pad_t)pi; pi_pad->pkt_val[0] = 0; pi_pad->pkt_val[1] = 0; pi_pad->pkt_val[2] = 0; pi_pad->pkt_val[3] = 0; pi_pad->pkt_val[4] = 0; pi_pad->pkt_val[5] = 0; #ifndef __LP64__ pi_pad->pkt_val[6] = 0; pi_pad->pkt_val[7] = 0; pi_pad->pkt_val[8] = 0; pi_pad->pkt_val[9] = 0; pi_pad->pkt_val[10] = 0; #endif } static inline void rxd_info_zero(if_rxd_info_t ri) { if_rxd_info_pad_t ri_pad; int i; ri_pad = (if_rxd_info_pad_t)ri; for (i = 0; i < RXD_LOOP_BOUND; i += 4) { ri_pad->rxd_val[i] = 0; ri_pad->rxd_val[i+1] = 0; ri_pad->rxd_val[i+2] = 0; ri_pad->rxd_val[i+3] = 0; } #ifdef __LP64__ ri_pad->rxd_val[RXD_INFO_SIZE-1] = 0; #endif } /* * Only allow a single packet to take up most 1/nth of the tx ring */ #define MAX_SINGLE_PACKET_FRACTION 12 #define IF_BAD_DMA (bus_addr_t)-1 #define CTX_ACTIVE(ctx) ((if_getdrvflags((ctx)->ifc_ifp) & IFF_DRV_RUNNING)) #define CTX_LOCK_INIT(_sc, _name) mtx_init(&(_sc)->ifc_mtx, _name, "iflib ctx lock", MTX_DEF) #define CTX_LOCK(ctx) mtx_lock(&(ctx)->ifc_mtx) #define CTX_UNLOCK(ctx) mtx_unlock(&(ctx)->ifc_mtx) #define CTX_LOCK_DESTROY(ctx) mtx_destroy(&(ctx)->ifc_mtx) #define CALLOUT_LOCK(txq) mtx_lock(&txq->ift_mtx) #define CALLOUT_UNLOCK(txq) mtx_unlock(&txq->ift_mtx) /* Our boot-time initialization hook */ static int iflib_module_event_handler(module_t, int, void *); static moduledata_t iflib_moduledata = { "iflib", iflib_module_event_handler, NULL }; DECLARE_MODULE(iflib, iflib_moduledata, SI_SUB_INIT_IF, SI_ORDER_ANY); MODULE_VERSION(iflib, 1); MODULE_DEPEND(iflib, pci, 1, 1, 1); MODULE_DEPEND(iflib, ether, 1, 1, 1); TASKQGROUP_DEFINE(if_io_tqg, mp_ncpus, 1); TASKQGROUP_DEFINE(if_config_tqg, 1, 1); #ifndef IFLIB_DEBUG_COUNTERS #ifdef INVARIANTS #define IFLIB_DEBUG_COUNTERS 1 #else #define IFLIB_DEBUG_COUNTERS 0 #endif /* !INVARIANTS */ #endif static SYSCTL_NODE(_net, OID_AUTO, iflib, CTLFLAG_RD, 0, "iflib driver parameters"); /* * XXX need to ensure that this can't accidentally cause the head to be moved backwards */ static int iflib_min_tx_latency = 0; SYSCTL_INT(_net_iflib, OID_AUTO, min_tx_latency, CTLFLAG_RW, &iflib_min_tx_latency, 0, "minimize transmit latency at the possible expense of throughput"); static int iflib_no_tx_batch = 0; SYSCTL_INT(_net_iflib, OID_AUTO, no_tx_batch, CTLFLAG_RW, &iflib_no_tx_batch, 0, "minimize transmit latency at the possible expense of throughput"); #if IFLIB_DEBUG_COUNTERS static int iflib_tx_seen; static int iflib_tx_sent; static int iflib_tx_encap; static int iflib_rx_allocs; static int iflib_fl_refills; static int iflib_fl_refills_large; static int iflib_tx_frees; SYSCTL_INT(_net_iflib, OID_AUTO, tx_seen, CTLFLAG_RD, &iflib_tx_seen, 0, "# tx mbufs seen"); SYSCTL_INT(_net_iflib, OID_AUTO, tx_sent, CTLFLAG_RD, &iflib_tx_sent, 0, "# tx mbufs sent"); SYSCTL_INT(_net_iflib, OID_AUTO, tx_encap, CTLFLAG_RD, &iflib_tx_encap, 0, "# tx mbufs encapped"); SYSCTL_INT(_net_iflib, OID_AUTO, tx_frees, CTLFLAG_RD, &iflib_tx_frees, 0, "# tx frees"); SYSCTL_INT(_net_iflib, OID_AUTO, rx_allocs, CTLFLAG_RD, &iflib_rx_allocs, 0, "# rx allocations"); SYSCTL_INT(_net_iflib, OID_AUTO, fl_refills, CTLFLAG_RD, &iflib_fl_refills, 0, "# refills"); SYSCTL_INT(_net_iflib, OID_AUTO, fl_refills_large, CTLFLAG_RD, &iflib_fl_refills_large, 0, "# large refills"); static int iflib_txq_drain_flushing; static int iflib_txq_drain_oactive; static int iflib_txq_drain_notready; static int iflib_txq_drain_encapfail; SYSCTL_INT(_net_iflib, OID_AUTO, txq_drain_flushing, CTLFLAG_RD, &iflib_txq_drain_flushing, 0, "# drain flushes"); SYSCTL_INT(_net_iflib, OID_AUTO, txq_drain_oactive, CTLFLAG_RD, &iflib_txq_drain_oactive, 0, "# drain oactives"); SYSCTL_INT(_net_iflib, OID_AUTO, txq_drain_notready, CTLFLAG_RD, &iflib_txq_drain_notready, 0, "# drain notready"); SYSCTL_INT(_net_iflib, OID_AUTO, txq_drain_encapfail, CTLFLAG_RD, &iflib_txq_drain_encapfail, 0, "# drain encap fails"); static int iflib_encap_load_mbuf_fail; static int iflib_encap_txq_avail_fail; static int iflib_encap_txd_encap_fail; SYSCTL_INT(_net_iflib, OID_AUTO, encap_load_mbuf_fail, CTLFLAG_RD, &iflib_encap_load_mbuf_fail, 0, "# busdma load failures"); SYSCTL_INT(_net_iflib, OID_AUTO, encap_txq_avail_fail, CTLFLAG_RD, &iflib_encap_txq_avail_fail, 0, "# txq avail failures"); SYSCTL_INT(_net_iflib, OID_AUTO, encap_txd_encap_fail, CTLFLAG_RD, &iflib_encap_txd_encap_fail, 0, "# driver encap failures"); static int iflib_task_fn_rxs; static int iflib_rx_intr_enables; static int iflib_fast_intrs; static int iflib_intr_link; static int iflib_intr_msix; static int iflib_rx_unavail; static int iflib_rx_ctx_inactive; static int iflib_rx_zero_len; static int iflib_rx_if_input; static int iflib_rx_mbuf_null; static int iflib_rxd_flush; static int iflib_verbose_debug; SYSCTL_INT(_net_iflib, OID_AUTO, intr_link, CTLFLAG_RD, &iflib_intr_link, 0, "# intr link calls"); SYSCTL_INT(_net_iflib, OID_AUTO, intr_msix, CTLFLAG_RD, &iflib_intr_msix, 0, "# intr msix calls"); SYSCTL_INT(_net_iflib, OID_AUTO, task_fn_rx, CTLFLAG_RD, &iflib_task_fn_rxs, 0, "# task_fn_rx calls"); SYSCTL_INT(_net_iflib, OID_AUTO, rx_intr_enables, CTLFLAG_RD, &iflib_rx_intr_enables, 0, "# rx intr enables"); SYSCTL_INT(_net_iflib, OID_AUTO, fast_intrs, CTLFLAG_RD, &iflib_fast_intrs, 0, "# fast_intr calls"); SYSCTL_INT(_net_iflib, OID_AUTO, rx_unavail, CTLFLAG_RD, &iflib_rx_unavail, 0, "# times rxeof called with no available data"); SYSCTL_INT(_net_iflib, OID_AUTO, rx_ctx_inactive, CTLFLAG_RD, &iflib_rx_ctx_inactive, 0, "# times rxeof called with inactive context"); SYSCTL_INT(_net_iflib, OID_AUTO, rx_zero_len, CTLFLAG_RD, &iflib_rx_zero_len, 0, "# times rxeof saw zero len mbuf"); SYSCTL_INT(_net_iflib, OID_AUTO, rx_if_input, CTLFLAG_RD, &iflib_rx_if_input, 0, "# times rxeof called if_input"); SYSCTL_INT(_net_iflib, OID_AUTO, rx_mbuf_null, CTLFLAG_RD, &iflib_rx_mbuf_null, 0, "# times rxeof got null mbuf"); SYSCTL_INT(_net_iflib, OID_AUTO, rxd_flush, CTLFLAG_RD, &iflib_rxd_flush, 0, "# times rxd_flush called"); SYSCTL_INT(_net_iflib, OID_AUTO, verbose_debug, CTLFLAG_RW, &iflib_verbose_debug, 0, "enable verbose debugging"); #define DBG_COUNTER_INC(name) atomic_add_int(&(iflib_ ## name), 1) static void iflib_debug_reset(void) { iflib_tx_seen = iflib_tx_sent = iflib_tx_encap = iflib_rx_allocs = iflib_fl_refills = iflib_fl_refills_large = iflib_tx_frees = iflib_txq_drain_flushing = iflib_txq_drain_oactive = iflib_txq_drain_notready = iflib_txq_drain_encapfail = iflib_encap_load_mbuf_fail = iflib_encap_txq_avail_fail = iflib_encap_txd_encap_fail = iflib_task_fn_rxs = iflib_rx_intr_enables = iflib_fast_intrs = iflib_intr_link = iflib_intr_msix = iflib_rx_unavail = iflib_rx_ctx_inactive = iflib_rx_zero_len = iflib_rx_if_input = iflib_rx_mbuf_null = iflib_rxd_flush = 0; } #else #define DBG_COUNTER_INC(name) static void iflib_debug_reset(void) {} #endif #define IFLIB_DEBUG 0 static void iflib_tx_structures_free(if_ctx_t ctx); static void iflib_rx_structures_free(if_ctx_t ctx); static int iflib_queues_alloc(if_ctx_t ctx); static int iflib_tx_credits_update(if_ctx_t ctx, iflib_txq_t txq); static int iflib_rxd_avail(if_ctx_t ctx, iflib_rxq_t rxq, qidx_t cidx, qidx_t budget); static int iflib_qset_structures_setup(if_ctx_t ctx); static int iflib_msix_init(if_ctx_t ctx); static int iflib_legacy_setup(if_ctx_t ctx, driver_filter_t filter, void *filterarg, int *rid, char *str); static void iflib_txq_check_drain(iflib_txq_t txq, int budget); static uint32_t iflib_txq_can_drain(struct ifmp_ring *); static int iflib_register(if_ctx_t); static void iflib_init_locked(if_ctx_t ctx); static void iflib_add_device_sysctl_pre(if_ctx_t ctx); static void iflib_add_device_sysctl_post(if_ctx_t ctx); static void iflib_ifmp_purge(iflib_txq_t txq); static void _iflib_pre_assert(if_softc_ctx_t scctx); static void iflib_stop(if_ctx_t ctx); static void iflib_if_init_locked(if_ctx_t ctx); #ifndef __NO_STRICT_ALIGNMENT static struct mbuf * iflib_fixup_rx(struct mbuf *m); #endif #ifdef DEV_NETMAP #include #include #include MODULE_DEPEND(iflib, netmap, 1, 1, 1); static int netmap_fl_refill(iflib_rxq_t rxq, struct netmap_kring *kring, uint32_t nm_i, bool init); /* * device-specific sysctl variables: * * iflib_crcstrip: 0: keep CRC in rx frames (default), 1: strip it. * During regular operations the CRC is stripped, but on some * hardware reception of frames not multiple of 64 is slower, * so using crcstrip=0 helps in benchmarks. * * iflib_rx_miss, iflib_rx_miss_bufs: * count packets that might be missed due to lost interrupts. */ SYSCTL_DECL(_dev_netmap); /* * The xl driver by default strips CRCs and we do not override it. */ int iflib_crcstrip = 1; SYSCTL_INT(_dev_netmap, OID_AUTO, iflib_crcstrip, CTLFLAG_RW, &iflib_crcstrip, 1, "strip CRC on rx frames"); int iflib_rx_miss, iflib_rx_miss_bufs; SYSCTL_INT(_dev_netmap, OID_AUTO, iflib_rx_miss, CTLFLAG_RW, &iflib_rx_miss, 0, "potentially missed rx intr"); SYSCTL_INT(_dev_netmap, OID_AUTO, iflib_rx_miss_bufs, CTLFLAG_RW, &iflib_rx_miss_bufs, 0, "potentially missed rx intr bufs"); /* * Register/unregister. We are already under netmap lock. * Only called on the first register or the last unregister. */ static int iflib_netmap_register(struct netmap_adapter *na, int onoff) { struct ifnet *ifp = na->ifp; if_ctx_t ctx = ifp->if_softc; int status; CTX_LOCK(ctx); IFDI_INTR_DISABLE(ctx); /* Tell the stack that the interface is no longer active */ ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE); if (!CTX_IS_VF(ctx)) IFDI_CRCSTRIP_SET(ctx, onoff, iflib_crcstrip); /* enable or disable flags and callbacks in na and ifp */ if (onoff) { nm_set_native_flags(na); } else { nm_clear_native_flags(na); } iflib_stop(ctx); iflib_init_locked(ctx); IFDI_CRCSTRIP_SET(ctx, onoff, iflib_crcstrip); // XXX why twice ? status = ifp->if_drv_flags & IFF_DRV_RUNNING ? 0 : 1; if (status) nm_clear_native_flags(na); CTX_UNLOCK(ctx); return (status); } static int netmap_fl_refill(iflib_rxq_t rxq, struct netmap_kring *kring, uint32_t nm_i, bool init) { struct netmap_adapter *na = kring->na; u_int const lim = kring->nkr_num_slots - 1; u_int head = kring->rhead; struct netmap_ring *ring = kring->ring; bus_dmamap_t *map; struct if_rxd_update iru; if_ctx_t ctx = rxq->ifr_ctx; iflib_fl_t fl = &rxq->ifr_fl[0]; uint32_t refill_pidx, nic_i; if (nm_i == head && __predict_true(!init)) return 0; iru_init(&iru, rxq, 0 /* flid */); map = fl->ifl_sds.ifsd_map; refill_pidx = netmap_idx_k2n(kring, nm_i); /* * IMPORTANT: we must leave one free slot in the ring, * so move head back by one unit */ head = nm_prev(head, lim); while (nm_i != head) { for (int tmp_pidx = 0; tmp_pidx < IFLIB_MAX_RX_REFRESH && nm_i != head; tmp_pidx++) { struct netmap_slot *slot = &ring->slot[nm_i]; void *addr = PNMB(na, slot, &fl->ifl_bus_addrs[tmp_pidx]); uint32_t nic_i_dma = refill_pidx; nic_i = netmap_idx_k2n(kring, nm_i); MPASS(tmp_pidx < IFLIB_MAX_RX_REFRESH); if (addr == NETMAP_BUF_BASE(na)) /* bad buf */ return netmap_ring_reinit(kring); fl->ifl_vm_addrs[tmp_pidx] = addr; if (__predict_false(init) && map) { netmap_load_map(na, fl->ifl_ifdi->idi_tag, map[nic_i], addr); } else if (map && (slot->flags & NS_BUF_CHANGED)) { /* buffer has changed, reload map */ netmap_reload_map(na, fl->ifl_ifdi->idi_tag, map[nic_i], addr); } slot->flags &= ~NS_BUF_CHANGED; nm_i = nm_next(nm_i, lim); fl->ifl_rxd_idxs[tmp_pidx] = nic_i = nm_next(nic_i, lim); if (nm_i != head && tmp_pidx < IFLIB_MAX_RX_REFRESH-1) continue; iru.iru_pidx = refill_pidx; iru.iru_count = tmp_pidx+1; ctx->isc_rxd_refill(ctx->ifc_softc, &iru); refill_pidx = nic_i; if (map == NULL) continue; for (int n = 0; n < iru.iru_count; n++) { bus_dmamap_sync(fl->ifl_ifdi->idi_tag, map[nic_i_dma], BUS_DMASYNC_PREREAD); /* XXX - change this to not use the netmap func*/ nic_i_dma = nm_next(nic_i_dma, lim); } } } kring->nr_hwcur = head; if (map) bus_dmamap_sync(fl->ifl_ifdi->idi_tag, fl->ifl_ifdi->idi_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); ctx->isc_rxd_flush(ctx->ifc_softc, rxq->ifr_id, fl->ifl_id, nic_i); return (0); } /* * Reconcile kernel and user view of the transmit ring. * * All information is in the kring. * Userspace wants to send packets up to the one before kring->rhead, * kernel knows kring->nr_hwcur is the first unsent packet. * * Here we push packets out (as many as possible), and possibly * reclaim buffers from previously completed transmission. * * The caller (netmap) guarantees that there is only one instance * running at any time. Any interference with other driver * methods should be handled by the individual drivers. */ static int iflib_netmap_txsync(struct netmap_kring *kring, int flags) { struct netmap_adapter *na = kring->na; struct ifnet *ifp = na->ifp; struct netmap_ring *ring = kring->ring; u_int nm_i; /* index into the netmap ring */ u_int nic_i; /* index into the NIC ring */ u_int n; u_int const lim = kring->nkr_num_slots - 1; u_int const head = kring->rhead; struct if_pkt_info pi; /* * interrupts on every tx packet are expensive so request * them every half ring, or where NS_REPORT is set */ u_int report_frequency = kring->nkr_num_slots >> 1; /* device-specific */ if_ctx_t ctx = ifp->if_softc; iflib_txq_t txq = &ctx->ifc_txqs[kring->ring_id]; if (txq->ift_sds.ifsd_map) bus_dmamap_sync(txq->ift_desc_tag, txq->ift_ifdi->idi_map, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); /* * First part: process new packets to send. * nm_i is the current index in the netmap ring, * nic_i is the corresponding index in the NIC ring. * * If we have packets to send (nm_i != head) * iterate over the netmap ring, fetch length and update * the corresponding slot in the NIC ring. Some drivers also * need to update the buffer's physical address in the NIC slot * even NS_BUF_CHANGED is not set (PNMB computes the addresses). * * The netmap_reload_map() calls is especially expensive, * even when (as in this case) the tag is 0, so do only * when the buffer has actually changed. * * If possible do not set the report/intr bit on all slots, * but only a few times per ring or when NS_REPORT is set. * * Finally, on 10G and faster drivers, it might be useful * to prefetch the next slot and txr entry. */ nm_i = netmap_idx_n2k(kring, kring->nr_hwcur); pkt_info_zero(&pi); pi.ipi_segs = txq->ift_segs; pi.ipi_qsidx = kring->ring_id; if (nm_i != head) { /* we have new packets to send */ nic_i = netmap_idx_k2n(kring, nm_i); __builtin_prefetch(&ring->slot[nm_i]); __builtin_prefetch(&txq->ift_sds.ifsd_m[nic_i]); if (txq->ift_sds.ifsd_map) __builtin_prefetch(&txq->ift_sds.ifsd_map[nic_i]); for (n = 0; nm_i != head; n++) { struct netmap_slot *slot = &ring->slot[nm_i]; u_int len = slot->len; uint64_t paddr; void *addr = PNMB(na, slot, &paddr); int flags = (slot->flags & NS_REPORT || nic_i == 0 || nic_i == report_frequency) ? IPI_TX_INTR : 0; /* device-specific */ pi.ipi_len = len; pi.ipi_segs[0].ds_addr = paddr; pi.ipi_segs[0].ds_len = len; pi.ipi_nsegs = 1; pi.ipi_ndescs = 0; pi.ipi_pidx = nic_i; pi.ipi_flags = flags; /* Fill the slot in the NIC ring. */ ctx->isc_txd_encap(ctx->ifc_softc, &pi); /* prefetch for next round */ __builtin_prefetch(&ring->slot[nm_i + 1]); __builtin_prefetch(&txq->ift_sds.ifsd_m[nic_i + 1]); if (txq->ift_sds.ifsd_map) { __builtin_prefetch(&txq->ift_sds.ifsd_map[nic_i + 1]); NM_CHECK_ADDR_LEN(na, addr, len); if (slot->flags & NS_BUF_CHANGED) { /* buffer has changed, reload map */ netmap_reload_map(na, txq->ift_desc_tag, txq->ift_sds.ifsd_map[nic_i], addr); } /* make sure changes to the buffer are synced */ bus_dmamap_sync(txq->ift_ifdi->idi_tag, txq->ift_sds.ifsd_map[nic_i], BUS_DMASYNC_PREWRITE); } slot->flags &= ~(NS_REPORT | NS_BUF_CHANGED); nm_i = nm_next(nm_i, lim); nic_i = nm_next(nic_i, lim); } kring->nr_hwcur = head; /* synchronize the NIC ring */ if (txq->ift_sds.ifsd_map) bus_dmamap_sync(txq->ift_desc_tag, txq->ift_ifdi->idi_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); /* (re)start the tx unit up to slot nic_i (excluded) */ ctx->isc_txd_flush(ctx->ifc_softc, txq->ift_id, nic_i); } /* * Second part: reclaim buffers for completed transmissions. */ if (iflib_tx_credits_update(ctx, txq)) { /* some tx completed, increment avail */ nic_i = txq->ift_cidx_processed; kring->nr_hwtail = nm_prev(netmap_idx_n2k(kring, nic_i), lim); } return (0); } /* * Reconcile kernel and user view of the receive ring. * Same as for the txsync, this routine must be efficient. * The caller guarantees a single invocations, but races against * the rest of the driver should be handled here. * * On call, kring->rhead is the first packet that userspace wants * to keep, and kring->rcur is the wakeup point. * The kernel has previously reported packets up to kring->rtail. * * If (flags & NAF_FORCE_READ) also check for incoming packets irrespective * of whether or not we received an interrupt. */ static int iflib_netmap_rxsync(struct netmap_kring *kring, int flags) { struct netmap_adapter *na = kring->na; struct netmap_ring *ring = kring->ring; uint32_t nm_i; /* index into the netmap ring */ uint32_t nic_i; /* index into the NIC ring */ u_int i, n; u_int const lim = kring->nkr_num_slots - 1; u_int const head = netmap_idx_n2k(kring, kring->rhead); int force_update = (flags & NAF_FORCE_READ) || kring->nr_kflags & NKR_PENDINTR; struct if_rxd_info ri; struct ifnet *ifp = na->ifp; if_ctx_t ctx = ifp->if_softc; iflib_rxq_t rxq = &ctx->ifc_rxqs[kring->ring_id]; iflib_fl_t fl = rxq->ifr_fl; if (head > lim) return netmap_ring_reinit(kring); /* XXX check sync modes */ for (i = 0, fl = rxq->ifr_fl; i < rxq->ifr_nfl; i++, fl++) { if (fl->ifl_sds.ifsd_map == NULL) continue; bus_dmamap_sync(rxq->ifr_fl[i].ifl_desc_tag, fl->ifl_ifdi->idi_map, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); } /* * First part: import newly received packets. * * nm_i is the index of the next free slot in the netmap ring, * nic_i is the index of the next received packet in the NIC ring, * and they may differ in case if_init() has been called while * in netmap mode. For the receive ring we have * * nic_i = rxr->next_check; * nm_i = kring->nr_hwtail (previous) * and * nm_i == (nic_i + kring->nkr_hwofs) % ring_size * * rxr->next_check is set to 0 on a ring reinit */ if (netmap_no_pendintr || force_update) { int crclen = iflib_crcstrip ? 0 : 4; int error, avail; uint16_t slot_flags = kring->nkr_slot_flags; for (i = 0; i < rxq->ifr_nfl; i++) { fl = &rxq->ifr_fl[i]; nic_i = fl->ifl_cidx; nm_i = netmap_idx_n2k(kring, nic_i); avail = iflib_rxd_avail(ctx, rxq, nic_i, USHRT_MAX); for (n = 0; avail > 0; n++, avail--) { rxd_info_zero(&ri); ri.iri_frags = rxq->ifr_frags; ri.iri_qsidx = kring->ring_id; ri.iri_ifp = ctx->ifc_ifp; ri.iri_cidx = nic_i; error = ctx->isc_rxd_pkt_get(ctx->ifc_softc, &ri); ring->slot[nm_i].len = error ? 0 : ri.iri_len - crclen; ring->slot[nm_i].flags = slot_flags; if (fl->ifl_sds.ifsd_map) bus_dmamap_sync(fl->ifl_ifdi->idi_tag, fl->ifl_sds.ifsd_map[nic_i], BUS_DMASYNC_POSTREAD); nm_i = nm_next(nm_i, lim); nic_i = nm_next(nic_i, lim); } if (n) { /* update the state variables */ if (netmap_no_pendintr && !force_update) { /* diagnostics */ iflib_rx_miss ++; iflib_rx_miss_bufs += n; } fl->ifl_cidx = nic_i; kring->nr_hwtail = netmap_idx_k2n(kring, nm_i); } kring->nr_kflags &= ~NKR_PENDINTR; } } /* * Second part: skip past packets that userspace has released. * (kring->nr_hwcur to head excluded), * and make the buffers available for reception. * As usual nm_i is the index in the netmap ring, * nic_i is the index in the NIC ring, and * nm_i == (nic_i + kring->nkr_hwofs) % ring_size */ /* XXX not sure how this will work with multiple free lists */ nm_i = netmap_idx_n2k(kring, kring->nr_hwcur); return (netmap_fl_refill(rxq, kring, nm_i, false)); } static void iflib_netmap_intr(struct netmap_adapter *na, int onoff) { struct ifnet *ifp = na->ifp; if_ctx_t ctx = ifp->if_softc; CTX_LOCK(ctx); if (onoff) { IFDI_INTR_ENABLE(ctx); } else { IFDI_INTR_DISABLE(ctx); } CTX_UNLOCK(ctx); } static int iflib_netmap_attach(if_ctx_t ctx) { struct netmap_adapter na; if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; bzero(&na, sizeof(na)); na.ifp = ctx->ifc_ifp; na.na_flags = NAF_BDG_MAYSLEEP; MPASS(ctx->ifc_softc_ctx.isc_ntxqsets); MPASS(ctx->ifc_softc_ctx.isc_nrxqsets); na.num_tx_desc = scctx->isc_ntxd[0]; na.num_rx_desc = scctx->isc_nrxd[0]; na.nm_txsync = iflib_netmap_txsync; na.nm_rxsync = iflib_netmap_rxsync; na.nm_register = iflib_netmap_register; na.nm_intr = iflib_netmap_intr; na.num_tx_rings = ctx->ifc_softc_ctx.isc_ntxqsets; na.num_rx_rings = ctx->ifc_softc_ctx.isc_nrxqsets; return (netmap_attach(&na)); } static void iflib_netmap_txq_init(if_ctx_t ctx, iflib_txq_t txq) { struct netmap_adapter *na = NA(ctx->ifc_ifp); struct netmap_slot *slot; slot = netmap_reset(na, NR_TX, txq->ift_id, 0); if (slot == NULL) return; if (txq->ift_sds.ifsd_map == NULL) return; for (int i = 0; i < ctx->ifc_softc_ctx.isc_ntxd[0]; i++) { /* * In netmap mode, set the map for the packet buffer. * NOTE: Some drivers (not this one) also need to set * the physical buffer address in the NIC ring. * netmap_idx_n2k() maps a nic index, i, into the corresponding * netmap slot index, si */ int si = netmap_idx_n2k(&na->tx_rings[txq->ift_id], i); netmap_load_map(na, txq->ift_desc_tag, txq->ift_sds.ifsd_map[i], NMB(na, slot + si)); } } static void iflib_netmap_rxq_init(if_ctx_t ctx, iflib_rxq_t rxq) { struct netmap_adapter *na = NA(ctx->ifc_ifp); struct netmap_kring *kring = &na->rx_rings[rxq->ifr_id]; struct netmap_slot *slot; uint32_t nm_i; slot = netmap_reset(na, NR_RX, rxq->ifr_id, 0); if (slot == NULL) return; nm_i = netmap_idx_n2k(kring, 0); netmap_fl_refill(rxq, kring, nm_i, true); } #define iflib_netmap_detach(ifp) netmap_detach(ifp) #else #define iflib_netmap_txq_init(ctx, txq) #define iflib_netmap_rxq_init(ctx, rxq) #define iflib_netmap_detach(ifp) #define iflib_netmap_attach(ctx) (0) #define netmap_rx_irq(ifp, qid, budget) (0) #define netmap_tx_irq(ifp, qid) do {} while (0) #endif #if defined(__i386__) || defined(__amd64__) static __inline void prefetch(void *x) { __asm volatile("prefetcht0 %0" :: "m" (*(unsigned long *)x)); } static __inline void prefetch2cachelines(void *x) { __asm volatile("prefetcht0 %0" :: "m" (*(unsigned long *)x)); #if (CACHE_LINE_SIZE < 128) __asm volatile("prefetcht0 %0" :: "m" (*(((unsigned long *)x)+CACHE_LINE_SIZE/(sizeof(unsigned long))))); #endif } #else #define prefetch(x) #define prefetch2cachelines(x) #endif static void iru_init(if_rxd_update_t iru, iflib_rxq_t rxq, uint8_t flid) { iflib_fl_t fl; fl = &rxq->ifr_fl[flid]; iru->iru_paddrs = fl->ifl_bus_addrs; iru->iru_vaddrs = &fl->ifl_vm_addrs[0]; iru->iru_idxs = fl->ifl_rxd_idxs; iru->iru_qsidx = rxq->ifr_id; iru->iru_buf_size = fl->ifl_buf_size; iru->iru_flidx = fl->ifl_id; } static void _iflib_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nseg, int err) { if (err) return; *(bus_addr_t *) arg = segs[0].ds_addr; } int iflib_dma_alloc(if_ctx_t ctx, int size, iflib_dma_info_t dma, int mapflags) { int err; if_shared_ctx_t sctx = ctx->ifc_sctx; device_t dev = ctx->ifc_dev; KASSERT(sctx->isc_q_align != 0, ("alignment value not initialized")); err = bus_dma_tag_create(bus_get_dma_tag(dev), /* parent */ sctx->isc_q_align, 0, /* alignment, bounds */ BUS_SPACE_MAXADDR, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ size, /* maxsize */ 1, /* nsegments */ size, /* maxsegsize */ BUS_DMA_ALLOCNOW, /* flags */ NULL, /* lockfunc */ NULL, /* lockarg */ &dma->idi_tag); if (err) { device_printf(dev, "%s: bus_dma_tag_create failed: %d\n", __func__, err); goto fail_0; } err = bus_dmamem_alloc(dma->idi_tag, (void**) &dma->idi_vaddr, BUS_DMA_NOWAIT | BUS_DMA_COHERENT | BUS_DMA_ZERO, &dma->idi_map); if (err) { device_printf(dev, "%s: bus_dmamem_alloc(%ju) failed: %d\n", __func__, (uintmax_t)size, err); goto fail_1; } dma->idi_paddr = IF_BAD_DMA; err = bus_dmamap_load(dma->idi_tag, dma->idi_map, dma->idi_vaddr, size, _iflib_dmamap_cb, &dma->idi_paddr, mapflags | BUS_DMA_NOWAIT); if (err || dma->idi_paddr == IF_BAD_DMA) { device_printf(dev, "%s: bus_dmamap_load failed: %d\n", __func__, err); goto fail_2; } dma->idi_size = size; return (0); fail_2: bus_dmamem_free(dma->idi_tag, dma->idi_vaddr, dma->idi_map); fail_1: bus_dma_tag_destroy(dma->idi_tag); fail_0: dma->idi_tag = NULL; return (err); } int iflib_dma_alloc_multi(if_ctx_t ctx, int *sizes, iflib_dma_info_t *dmalist, int mapflags, int count) { int i, err; iflib_dma_info_t *dmaiter; dmaiter = dmalist; for (i = 0; i < count; i++, dmaiter++) { if ((err = iflib_dma_alloc(ctx, sizes[i], *dmaiter, mapflags)) != 0) break; } if (err) iflib_dma_free_multi(dmalist, i); return (err); } void iflib_dma_free(iflib_dma_info_t dma) { if (dma->idi_tag == NULL) return; if (dma->idi_paddr != IF_BAD_DMA) { bus_dmamap_sync(dma->idi_tag, dma->idi_map, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); bus_dmamap_unload(dma->idi_tag, dma->idi_map); dma->idi_paddr = IF_BAD_DMA; } if (dma->idi_vaddr != NULL) { bus_dmamem_free(dma->idi_tag, dma->idi_vaddr, dma->idi_map); dma->idi_vaddr = NULL; } bus_dma_tag_destroy(dma->idi_tag); dma->idi_tag = NULL; } void iflib_dma_free_multi(iflib_dma_info_t *dmalist, int count) { int i; iflib_dma_info_t *dmaiter = dmalist; for (i = 0; i < count; i++, dmaiter++) iflib_dma_free(*dmaiter); } #ifdef EARLY_AP_STARTUP static const int iflib_started = 1; #else /* * We used to abuse the smp_started flag to decide if the queues have been * fully initialized (by late taskqgroup_adjust() calls in a SYSINIT()). * That gave bad races, since the SYSINIT() runs strictly after smp_started * is set. Run a SYSINIT() strictly after that to just set a usable * completion flag. */ static int iflib_started; static void iflib_record_started(void *arg) { iflib_started = 1; } SYSINIT(iflib_record_started, SI_SUB_SMP + 1, SI_ORDER_FIRST, iflib_record_started, NULL); #endif static int iflib_fast_intr(void *arg) { iflib_filter_info_t info = arg; struct grouptask *gtask = info->ifi_task; if (!iflib_started) return (FILTER_HANDLED); DBG_COUNTER_INC(fast_intrs); if (info->ifi_filter != NULL && info->ifi_filter(info->ifi_filter_arg) == FILTER_HANDLED) return (FILTER_HANDLED); GROUPTASK_ENQUEUE(gtask); return (FILTER_HANDLED); } static int iflib_fast_intr_rxtx(void *arg) { iflib_filter_info_t info = arg; struct grouptask *gtask = info->ifi_task; iflib_rxq_t rxq = (iflib_rxq_t)info->ifi_ctx; if_ctx_t ctx; int i, cidx; if (!iflib_started) return (FILTER_HANDLED); DBG_COUNTER_INC(fast_intrs); if (info->ifi_filter != NULL && info->ifi_filter(info->ifi_filter_arg) == FILTER_HANDLED) return (FILTER_HANDLED); for (i = 0; i < rxq->ifr_ntxqirq; i++) { qidx_t txqid = rxq->ifr_txqid[i]; ctx = rxq->ifr_ctx; if (!ctx->isc_txd_credits_update(ctx->ifc_softc, txqid, false)) { IFDI_TX_QUEUE_INTR_ENABLE(ctx, txqid); continue; } GROUPTASK_ENQUEUE(&ctx->ifc_txqs[txqid].ift_task); } if (ctx->ifc_sctx->isc_flags & IFLIB_HAS_RXCQ) cidx = rxq->ifr_cq_cidx; else cidx = rxq->ifr_fl[0].ifl_cidx; if (iflib_rxd_avail(ctx, rxq, cidx, 1)) GROUPTASK_ENQUEUE(gtask); else IFDI_RX_QUEUE_INTR_ENABLE(ctx, rxq->ifr_id); return (FILTER_HANDLED); } static int iflib_fast_intr_ctx(void *arg) { iflib_filter_info_t info = arg; struct grouptask *gtask = info->ifi_task; if (!iflib_started) return (FILTER_HANDLED); DBG_COUNTER_INC(fast_intrs); if (info->ifi_filter != NULL && info->ifi_filter(info->ifi_filter_arg) == FILTER_HANDLED) return (FILTER_HANDLED); GROUPTASK_ENQUEUE(gtask); return (FILTER_HANDLED); } static int _iflib_irq_alloc(if_ctx_t ctx, if_irq_t irq, int rid, driver_filter_t filter, driver_intr_t handler, void *arg, char *name) { int rc, flags; struct resource *res; void *tag = NULL; device_t dev = ctx->ifc_dev; flags = RF_ACTIVE; if (ctx->ifc_flags & IFC_LEGACY) flags |= RF_SHAREABLE; MPASS(rid < 512); irq->ii_rid = rid; res = bus_alloc_resource_any(dev, SYS_RES_IRQ, &irq->ii_rid, flags); if (res == NULL) { device_printf(dev, "failed to allocate IRQ for rid %d, name %s.\n", rid, name); return (ENOMEM); } irq->ii_res = res; KASSERT(filter == NULL || handler == NULL, ("filter and handler can't both be non-NULL")); rc = bus_setup_intr(dev, res, INTR_MPSAFE | INTR_TYPE_NET, filter, handler, arg, &tag); if (rc != 0) { device_printf(dev, "failed to setup interrupt for rid %d, name %s: %d\n", rid, name ? name : "unknown", rc); return (rc); } else if (name) bus_describe_intr(dev, res, tag, "%s", name); irq->ii_tag = tag; return (0); } /********************************************************************* * * Allocate memory for tx_buffer structures. The tx_buffer stores all * the information needed to transmit a packet on the wire. This is * called only once at attach, setup is done every reset. * **********************************************************************/ static int iflib_txsd_alloc(iflib_txq_t txq) { if_ctx_t ctx = txq->ift_ctx; if_shared_ctx_t sctx = ctx->ifc_sctx; if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; device_t dev = ctx->ifc_dev; int err, nsegments, ntsosegments; nsegments = scctx->isc_tx_nsegments; ntsosegments = scctx->isc_tx_tso_segments_max; MPASS(scctx->isc_ntxd[0] > 0); MPASS(scctx->isc_ntxd[txq->ift_br_offset] > 0); MPASS(nsegments > 0); MPASS(ntsosegments > 0); /* * Setup DMA descriptor areas. */ if ((err = bus_dma_tag_create(bus_get_dma_tag(dev), 1, 0, /* alignment, bounds */ BUS_SPACE_MAXADDR, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ sctx->isc_tx_maxsize, /* maxsize */ nsegments, /* nsegments */ sctx->isc_tx_maxsegsize, /* maxsegsize */ 0, /* flags */ NULL, /* lockfunc */ NULL, /* lockfuncarg */ &txq->ift_desc_tag))) { device_printf(dev,"Unable to allocate TX DMA tag: %d\n", err); device_printf(dev,"maxsize: %ju nsegments: %d maxsegsize: %ju\n", (uintmax_t)sctx->isc_tx_maxsize, nsegments, (uintmax_t)sctx->isc_tx_maxsegsize); goto fail; } if ((err = bus_dma_tag_create(bus_get_dma_tag(dev), 1, 0, /* alignment, bounds */ BUS_SPACE_MAXADDR, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ scctx->isc_tx_tso_size_max, /* maxsize */ ntsosegments, /* nsegments */ scctx->isc_tx_tso_segsize_max, /* maxsegsize */ 0, /* flags */ NULL, /* lockfunc */ NULL, /* lockfuncarg */ &txq->ift_tso_desc_tag))) { device_printf(dev,"Unable to allocate TX TSO DMA tag: %d\n", err); goto fail; } if (!(txq->ift_sds.ifsd_flags = (uint8_t *) malloc(sizeof(uint8_t) * scctx->isc_ntxd[txq->ift_br_offset], M_IFLIB, M_NOWAIT | M_ZERO))) { device_printf(dev, "Unable to allocate tx_buffer memory\n"); err = ENOMEM; goto fail; } if (!(txq->ift_sds.ifsd_m = (struct mbuf **) malloc(sizeof(struct mbuf *) * scctx->isc_ntxd[txq->ift_br_offset], M_IFLIB, M_NOWAIT | M_ZERO))) { device_printf(dev, "Unable to allocate tx_buffer memory\n"); err = ENOMEM; goto fail; } /* Create the descriptor buffer dma maps */ #if defined(ACPI_DMAR) || (! (defined(__i386__) || defined(__amd64__))) if ((ctx->ifc_flags & IFC_DMAR) == 0) return (0); if (!(txq->ift_sds.ifsd_map = (bus_dmamap_t *) malloc(sizeof(bus_dmamap_t) * scctx->isc_ntxd[txq->ift_br_offset], M_IFLIB, M_NOWAIT | M_ZERO))) { device_printf(dev, "Unable to allocate tx_buffer map memory\n"); err = ENOMEM; goto fail; } for (int i = 0; i < scctx->isc_ntxd[txq->ift_br_offset]; i++) { err = bus_dmamap_create(txq->ift_desc_tag, 0, &txq->ift_sds.ifsd_map[i]); if (err != 0) { device_printf(dev, "Unable to create TX DMA map\n"); goto fail; } } #endif return (0); fail: /* We free all, it handles case where we are in the middle */ iflib_tx_structures_free(ctx); return (err); } static void iflib_txsd_destroy(if_ctx_t ctx, iflib_txq_t txq, int i) { bus_dmamap_t map; map = NULL; if (txq->ift_sds.ifsd_map != NULL) map = txq->ift_sds.ifsd_map[i]; if (map != NULL) { bus_dmamap_unload(txq->ift_desc_tag, map); bus_dmamap_destroy(txq->ift_desc_tag, map); txq->ift_sds.ifsd_map[i] = NULL; } } static void iflib_txq_destroy(iflib_txq_t txq) { if_ctx_t ctx = txq->ift_ctx; for (int i = 0; i < txq->ift_size; i++) iflib_txsd_destroy(ctx, txq, i); if (txq->ift_sds.ifsd_map != NULL) { free(txq->ift_sds.ifsd_map, M_IFLIB); txq->ift_sds.ifsd_map = NULL; } if (txq->ift_sds.ifsd_m != NULL) { free(txq->ift_sds.ifsd_m, M_IFLIB); txq->ift_sds.ifsd_m = NULL; } if (txq->ift_sds.ifsd_flags != NULL) { free(txq->ift_sds.ifsd_flags, M_IFLIB); txq->ift_sds.ifsd_flags = NULL; } if (txq->ift_desc_tag != NULL) { bus_dma_tag_destroy(txq->ift_desc_tag); txq->ift_desc_tag = NULL; } if (txq->ift_tso_desc_tag != NULL) { bus_dma_tag_destroy(txq->ift_tso_desc_tag); txq->ift_tso_desc_tag = NULL; } } static void iflib_txsd_free(if_ctx_t ctx, iflib_txq_t txq, int i) { struct mbuf **mp; mp = &txq->ift_sds.ifsd_m[i]; if (*mp == NULL) return; if (txq->ift_sds.ifsd_map != NULL) { bus_dmamap_sync(txq->ift_desc_tag, txq->ift_sds.ifsd_map[i], BUS_DMASYNC_POSTWRITE); bus_dmamap_unload(txq->ift_desc_tag, txq->ift_sds.ifsd_map[i]); } m_free(*mp); DBG_COUNTER_INC(tx_frees); *mp = NULL; } static int iflib_txq_setup(iflib_txq_t txq) { if_ctx_t ctx = txq->ift_ctx; if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; iflib_dma_info_t di; int i; /* Set number of descriptors available */ txq->ift_qstatus = IFLIB_QUEUE_IDLE; /* XXX make configurable */ txq->ift_update_freq = IFLIB_DEFAULT_TX_UPDATE_FREQ; /* Reset indices */ txq->ift_cidx_processed = 0; txq->ift_pidx = txq->ift_cidx = txq->ift_npending = 0; txq->ift_size = scctx->isc_ntxd[txq->ift_br_offset]; for (i = 0, di = txq->ift_ifdi; i < ctx->ifc_nhwtxqs; i++, di++) bzero((void *)di->idi_vaddr, di->idi_size); IFDI_TXQ_SETUP(ctx, txq->ift_id); for (i = 0, di = txq->ift_ifdi; i < ctx->ifc_nhwtxqs; i++, di++) bus_dmamap_sync(di->idi_tag, di->idi_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); return (0); } /********************************************************************* * * Allocate memory for rx_buffer structures. Since we use one * rx_buffer per received packet, the maximum number of rx_buffer's * that we'll need is equal to the number of receive descriptors * that we've allocated. * **********************************************************************/ static int iflib_rxsd_alloc(iflib_rxq_t rxq) { if_ctx_t ctx = rxq->ifr_ctx; if_shared_ctx_t sctx = ctx->ifc_sctx; if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; device_t dev = ctx->ifc_dev; iflib_fl_t fl; int err; MPASS(scctx->isc_nrxd[0] > 0); MPASS(scctx->isc_nrxd[rxq->ifr_fl_offset] > 0); fl = rxq->ifr_fl; for (int i = 0; i < rxq->ifr_nfl; i++, fl++) { fl->ifl_size = scctx->isc_nrxd[rxq->ifr_fl_offset]; /* this isn't necessarily the same */ err = bus_dma_tag_create(bus_get_dma_tag(dev), /* parent */ 1, 0, /* alignment, bounds */ BUS_SPACE_MAXADDR, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ sctx->isc_rx_maxsize, /* maxsize */ sctx->isc_rx_nsegments, /* nsegments */ sctx->isc_rx_maxsegsize, /* maxsegsize */ 0, /* flags */ NULL, /* lockfunc */ NULL, /* lockarg */ &fl->ifl_desc_tag); if (err) { device_printf(dev, "%s: bus_dma_tag_create failed %d\n", __func__, err); goto fail; } if (!(fl->ifl_sds.ifsd_flags = (uint8_t *) malloc(sizeof(uint8_t) * scctx->isc_nrxd[rxq->ifr_fl_offset], M_IFLIB, M_NOWAIT | M_ZERO))) { device_printf(dev, "Unable to allocate tx_buffer memory\n"); err = ENOMEM; goto fail; } if (!(fl->ifl_sds.ifsd_m = (struct mbuf **) malloc(sizeof(struct mbuf *) * scctx->isc_nrxd[rxq->ifr_fl_offset], M_IFLIB, M_NOWAIT | M_ZERO))) { device_printf(dev, "Unable to allocate tx_buffer memory\n"); err = ENOMEM; goto fail; } if (!(fl->ifl_sds.ifsd_cl = (caddr_t *) malloc(sizeof(caddr_t) * scctx->isc_nrxd[rxq->ifr_fl_offset], M_IFLIB, M_NOWAIT | M_ZERO))) { device_printf(dev, "Unable to allocate tx_buffer memory\n"); err = ENOMEM; goto fail; } /* Create the descriptor buffer dma maps */ #if defined(ACPI_DMAR) || (! (defined(__i386__) || defined(__amd64__))) if ((ctx->ifc_flags & IFC_DMAR) == 0) continue; if (!(fl->ifl_sds.ifsd_map = (bus_dmamap_t *) malloc(sizeof(bus_dmamap_t) * scctx->isc_nrxd[rxq->ifr_fl_offset], M_IFLIB, M_NOWAIT | M_ZERO))) { device_printf(dev, "Unable to allocate tx_buffer map memory\n"); err = ENOMEM; goto fail; } for (int i = 0; i < scctx->isc_nrxd[rxq->ifr_fl_offset]; i++) { err = bus_dmamap_create(fl->ifl_desc_tag, 0, &fl->ifl_sds.ifsd_map[i]); if (err != 0) { device_printf(dev, "Unable to create RX buffer DMA map\n"); goto fail; } } #endif } return (0); fail: iflib_rx_structures_free(ctx); return (err); } /* * Internal service routines */ struct rxq_refill_cb_arg { int error; bus_dma_segment_t seg; int nseg; }; static void _rxq_refill_cb(void *arg, bus_dma_segment_t *segs, int nseg, int error) { struct rxq_refill_cb_arg *cb_arg = arg; cb_arg->error = error; cb_arg->seg = segs[0]; cb_arg->nseg = nseg; } #ifdef ACPI_DMAR #define IS_DMAR(ctx) (ctx->ifc_flags & IFC_DMAR) #else #define IS_DMAR(ctx) (0) #endif /** * rxq_refill - refill an rxq free-buffer list * @ctx: the iflib context * @rxq: the free-list to refill * @n: the number of new buffers to allocate * * (Re)populate an rxq free-buffer list with up to @n new packet buffers. * The caller must assure that @n does not exceed the queue's capacity. */ static void _iflib_fl_refill(if_ctx_t ctx, iflib_fl_t fl, int count) { struct mbuf *m; int idx, frag_idx = fl->ifl_fragidx; int pidx = fl->ifl_pidx; caddr_t cl, *sd_cl; struct mbuf **sd_m; uint8_t *sd_flags; struct if_rxd_update iru; bus_dmamap_t *sd_map; int n, i = 0; uint64_t bus_addr; int err; qidx_t credits; sd_m = fl->ifl_sds.ifsd_m; sd_map = fl->ifl_sds.ifsd_map; sd_cl = fl->ifl_sds.ifsd_cl; sd_flags = fl->ifl_sds.ifsd_flags; idx = pidx; credits = fl->ifl_credits; n = count; MPASS(n > 0); MPASS(credits + n <= fl->ifl_size); if (pidx < fl->ifl_cidx) MPASS(pidx + n <= fl->ifl_cidx); if (pidx == fl->ifl_cidx && (credits < fl->ifl_size)) MPASS(fl->ifl_gen == 0); if (pidx > fl->ifl_cidx) MPASS(n <= fl->ifl_size - pidx + fl->ifl_cidx); DBG_COUNTER_INC(fl_refills); if (n > 8) DBG_COUNTER_INC(fl_refills_large); iru_init(&iru, fl->ifl_rxq, fl->ifl_id); while (n--) { /* * We allocate an uninitialized mbuf + cluster, mbuf is * initialized after rx. * * If the cluster is still set then we know a minimum sized packet was received */ bit_ffc_at(fl->ifl_rx_bitmap, frag_idx, fl->ifl_size, &frag_idx); if ((frag_idx < 0) || (frag_idx >= fl->ifl_size)) bit_ffc(fl->ifl_rx_bitmap, fl->ifl_size, &frag_idx); if ((cl = sd_cl[frag_idx]) == NULL) { if ((cl = sd_cl[frag_idx] = m_cljget(NULL, M_NOWAIT, fl->ifl_buf_size)) == NULL) break; #if MEMORY_LOGGING fl->ifl_cl_enqueued++; #endif } if ((m = m_gethdr(M_NOWAIT, MT_NOINIT)) == NULL) { break; } #if MEMORY_LOGGING fl->ifl_m_enqueued++; #endif DBG_COUNTER_INC(rx_allocs); #if defined(__i386__) || defined(__amd64__) if (!IS_DMAR(ctx)) { bus_addr = pmap_kextract((vm_offset_t)cl); } else #endif { struct rxq_refill_cb_arg cb_arg; iflib_rxq_t q; cb_arg.error = 0; q = fl->ifl_rxq; MPASS(sd_map != NULL); MPASS(sd_map[frag_idx] != NULL); err = bus_dmamap_load(fl->ifl_desc_tag, sd_map[frag_idx], cl, fl->ifl_buf_size, _rxq_refill_cb, &cb_arg, 0); bus_dmamap_sync(fl->ifl_desc_tag, sd_map[frag_idx], BUS_DMASYNC_PREREAD); if (err != 0 || cb_arg.error) { /* * !zone_pack ? */ if (fl->ifl_zone == zone_pack) uma_zfree(fl->ifl_zone, cl); m_free(m); n = 0; goto done; } bus_addr = cb_arg.seg.ds_addr; } bit_set(fl->ifl_rx_bitmap, frag_idx); sd_flags[frag_idx] |= RX_SW_DESC_INUSE; MPASS(sd_m[frag_idx] == NULL); sd_cl[frag_idx] = cl; sd_m[frag_idx] = m; fl->ifl_rxd_idxs[i] = frag_idx; fl->ifl_bus_addrs[i] = bus_addr; fl->ifl_vm_addrs[i] = cl; credits++; i++; MPASS(credits <= fl->ifl_size); if (++idx == fl->ifl_size) { fl->ifl_gen = 1; idx = 0; } if (n == 0 || i == IFLIB_MAX_RX_REFRESH) { iru.iru_pidx = pidx; iru.iru_count = i; ctx->isc_rxd_refill(ctx->ifc_softc, &iru); i = 0; pidx = idx; fl->ifl_pidx = idx; fl->ifl_credits = credits; } } done: if (i) { iru.iru_pidx = pidx; iru.iru_count = i; ctx->isc_rxd_refill(ctx->ifc_softc, &iru); fl->ifl_pidx = idx; fl->ifl_credits = credits; } DBG_COUNTER_INC(rxd_flush); if (fl->ifl_pidx == 0) pidx = fl->ifl_size - 1; else pidx = fl->ifl_pidx - 1; if (sd_map) bus_dmamap_sync(fl->ifl_ifdi->idi_tag, fl->ifl_ifdi->idi_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); ctx->isc_rxd_flush(ctx->ifc_softc, fl->ifl_rxq->ifr_id, fl->ifl_id, pidx); fl->ifl_fragidx = frag_idx; } static __inline void __iflib_fl_refill_lt(if_ctx_t ctx, iflib_fl_t fl, int max) { /* we avoid allowing pidx to catch up with cidx as it confuses ixl */ int32_t reclaimable = fl->ifl_size - fl->ifl_credits - 1; #ifdef INVARIANTS int32_t delta = fl->ifl_size - get_inuse(fl->ifl_size, fl->ifl_cidx, fl->ifl_pidx, fl->ifl_gen) - 1; #endif MPASS(fl->ifl_credits <= fl->ifl_size); MPASS(reclaimable == delta); if (reclaimable > 0) _iflib_fl_refill(ctx, fl, min(max, reclaimable)); } static void iflib_fl_bufs_free(iflib_fl_t fl) { iflib_dma_info_t idi = fl->ifl_ifdi; uint32_t i; for (i = 0; i < fl->ifl_size; i++) { struct mbuf **sd_m = &fl->ifl_sds.ifsd_m[i]; uint8_t *sd_flags = &fl->ifl_sds.ifsd_flags[i]; caddr_t *sd_cl = &fl->ifl_sds.ifsd_cl[i]; if (*sd_flags & RX_SW_DESC_INUSE) { if (fl->ifl_sds.ifsd_map != NULL) { bus_dmamap_t sd_map = fl->ifl_sds.ifsd_map[i]; bus_dmamap_unload(fl->ifl_desc_tag, sd_map); bus_dmamap_destroy(fl->ifl_desc_tag, sd_map); } if (*sd_m != NULL) { m_init(*sd_m, M_NOWAIT, MT_DATA, 0); uma_zfree(zone_mbuf, *sd_m); } if (*sd_cl != NULL) uma_zfree(fl->ifl_zone, *sd_cl); *sd_flags = 0; } else { MPASS(*sd_cl == NULL); MPASS(*sd_m == NULL); } #if MEMORY_LOGGING fl->ifl_m_dequeued++; fl->ifl_cl_dequeued++; #endif *sd_cl = NULL; *sd_m = NULL; } #ifdef INVARIANTS for (i = 0; i < fl->ifl_size; i++) { MPASS(fl->ifl_sds.ifsd_flags[i] == 0); MPASS(fl->ifl_sds.ifsd_cl[i] == NULL); MPASS(fl->ifl_sds.ifsd_m[i] == NULL); } #endif /* * Reset free list values */ fl->ifl_credits = fl->ifl_cidx = fl->ifl_pidx = fl->ifl_gen = fl->ifl_fragidx = 0; bzero(idi->idi_vaddr, idi->idi_size); } /********************************************************************* * * Initialize a receive ring and its buffers. * **********************************************************************/ static int iflib_fl_setup(iflib_fl_t fl) { iflib_rxq_t rxq = fl->ifl_rxq; if_ctx_t ctx = rxq->ifr_ctx; if_softc_ctx_t sctx = &ctx->ifc_softc_ctx; bit_nclear(fl->ifl_rx_bitmap, 0, fl->ifl_size); /* ** Free current RX buffer structs and their mbufs */ iflib_fl_bufs_free(fl); /* Now replenish the mbufs */ MPASS(fl->ifl_credits == 0); /* * XXX don't set the max_frame_size to larger * than the hardware can handle */ if (sctx->isc_max_frame_size <= 2048) fl->ifl_buf_size = MCLBYTES; #ifndef CONTIGMALLOC_WORKS else fl->ifl_buf_size = MJUMPAGESIZE; #else else if (sctx->isc_max_frame_size <= 4096) fl->ifl_buf_size = MJUMPAGESIZE; else if (sctx->isc_max_frame_size <= 9216) fl->ifl_buf_size = MJUM9BYTES; else fl->ifl_buf_size = MJUM16BYTES; #endif if (fl->ifl_buf_size > ctx->ifc_max_fl_buf_size) ctx->ifc_max_fl_buf_size = fl->ifl_buf_size; fl->ifl_cltype = m_gettype(fl->ifl_buf_size); fl->ifl_zone = m_getzone(fl->ifl_buf_size); /* avoid pre-allocating zillions of clusters to an idle card * potentially speeding up attach */ _iflib_fl_refill(ctx, fl, min(128, fl->ifl_size)); MPASS(min(128, fl->ifl_size) == fl->ifl_credits); if (min(128, fl->ifl_size) != fl->ifl_credits) return (ENOBUFS); /* * handle failure */ MPASS(rxq != NULL); MPASS(fl->ifl_ifdi != NULL); bus_dmamap_sync(fl->ifl_ifdi->idi_tag, fl->ifl_ifdi->idi_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); return (0); } /********************************************************************* * * Free receive ring data structures * **********************************************************************/ static void iflib_rx_sds_free(iflib_rxq_t rxq) { iflib_fl_t fl; int i; if (rxq->ifr_fl != NULL) { for (i = 0; i < rxq->ifr_nfl; i++) { fl = &rxq->ifr_fl[i]; if (fl->ifl_desc_tag != NULL) { bus_dma_tag_destroy(fl->ifl_desc_tag); fl->ifl_desc_tag = NULL; } free(fl->ifl_sds.ifsd_m, M_IFLIB); free(fl->ifl_sds.ifsd_cl, M_IFLIB); /* XXX destroy maps first */ free(fl->ifl_sds.ifsd_map, M_IFLIB); fl->ifl_sds.ifsd_m = NULL; fl->ifl_sds.ifsd_cl = NULL; fl->ifl_sds.ifsd_map = NULL; } free(rxq->ifr_fl, M_IFLIB); rxq->ifr_fl = NULL; rxq->ifr_cq_gen = rxq->ifr_cq_cidx = rxq->ifr_cq_pidx = 0; } } /* * MI independent logic * */ static void iflib_timer(void *arg) { iflib_txq_t txq = arg; if_ctx_t ctx = txq->ift_ctx; if_softc_ctx_t sctx = &ctx->ifc_softc_ctx; if (!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING)) return; /* ** Check on the state of the TX queue(s), this ** can be done without the lock because its RO ** and the HUNG state will be static if set. */ IFDI_TIMER(ctx, txq->ift_id); if ((txq->ift_qstatus == IFLIB_QUEUE_HUNG) && ((txq->ift_cleaned_prev == txq->ift_cleaned) || (sctx->isc_pause_frames == 0))) goto hung; if (ifmp_ring_is_stalled(txq->ift_br)) txq->ift_qstatus = IFLIB_QUEUE_HUNG; txq->ift_cleaned_prev = txq->ift_cleaned; /* handle any laggards */ if (txq->ift_db_pending) GROUPTASK_ENQUEUE(&txq->ift_task); sctx->isc_pause_frames = 0; if (if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING) callout_reset_on(&txq->ift_timer, hz/2, iflib_timer, txq, txq->ift_timer.c_cpu); return; hung: CTX_LOCK(ctx); if_setdrvflagbits(ctx->ifc_ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING); device_printf(ctx->ifc_dev, "TX(%d) desc avail = %d, pidx = %d\n", txq->ift_id, TXQ_AVAIL(txq), txq->ift_pidx); IFDI_WATCHDOG_RESET(ctx); ctx->ifc_watchdog_events++; ctx->ifc_flags |= IFC_DO_RESET; iflib_admin_intr_deferred(ctx); CTX_UNLOCK(ctx); } static void iflib_init_locked(if_ctx_t ctx) { if_softc_ctx_t sctx = &ctx->ifc_softc_ctx; if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; if_t ifp = ctx->ifc_ifp; iflib_fl_t fl; iflib_txq_t txq; iflib_rxq_t rxq; int i, j, tx_ip_csum_flags, tx_ip6_csum_flags; if_setdrvflagbits(ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING); IFDI_INTR_DISABLE(ctx); tx_ip_csum_flags = scctx->isc_tx_csum_flags & (CSUM_IP | CSUM_TCP | CSUM_UDP | CSUM_SCTP); tx_ip6_csum_flags = scctx->isc_tx_csum_flags & (CSUM_IP6_TCP | CSUM_IP6_UDP | CSUM_IP6_SCTP); /* Set hardware offload abilities */ if_clearhwassist(ifp); if (if_getcapenable(ifp) & IFCAP_TXCSUM) if_sethwassistbits(ifp, tx_ip_csum_flags, 0); if (if_getcapenable(ifp) & IFCAP_TXCSUM_IPV6) if_sethwassistbits(ifp, tx_ip6_csum_flags, 0); if (if_getcapenable(ifp) & IFCAP_TSO4) if_sethwassistbits(ifp, CSUM_IP_TSO, 0); if (if_getcapenable(ifp) & IFCAP_TSO6) if_sethwassistbits(ifp, CSUM_IP6_TSO, 0); for (i = 0, txq = ctx->ifc_txqs; i < sctx->isc_ntxqsets; i++, txq++) { CALLOUT_LOCK(txq); callout_stop(&txq->ift_timer); CALLOUT_UNLOCK(txq); iflib_netmap_txq_init(ctx, txq); } #ifdef INVARIANTS i = if_getdrvflags(ifp); #endif IFDI_INIT(ctx); MPASS(if_getdrvflags(ifp) == i); for (i = 0, rxq = ctx->ifc_rxqs; i < sctx->isc_nrxqsets; i++, rxq++) { /* XXX this should really be done on a per-queue basis */ if (if_getcapenable(ifp) & IFCAP_NETMAP) { MPASS(rxq->ifr_id == i); iflib_netmap_rxq_init(ctx, rxq); continue; } for (j = 0, fl = rxq->ifr_fl; j < rxq->ifr_nfl; j++, fl++) { if (iflib_fl_setup(fl)) { device_printf(ctx->ifc_dev, "freelist setup failed - check cluster settings\n"); goto done; } } } done: if_setdrvflagbits(ctx->ifc_ifp, IFF_DRV_RUNNING, IFF_DRV_OACTIVE); IFDI_INTR_ENABLE(ctx); txq = ctx->ifc_txqs; for (i = 0; i < sctx->isc_ntxqsets; i++, txq++) callout_reset_on(&txq->ift_timer, hz/2, iflib_timer, txq, txq->ift_timer.c_cpu); } static int iflib_media_change(if_t ifp) { if_ctx_t ctx = if_getsoftc(ifp); int err; CTX_LOCK(ctx); if ((err = IFDI_MEDIA_CHANGE(ctx)) == 0) iflib_init_locked(ctx); CTX_UNLOCK(ctx); return (err); } static void iflib_media_status(if_t ifp, struct ifmediareq *ifmr) { if_ctx_t ctx = if_getsoftc(ifp); CTX_LOCK(ctx); IFDI_UPDATE_ADMIN_STATUS(ctx); IFDI_MEDIA_STATUS(ctx, ifmr); CTX_UNLOCK(ctx); } static void iflib_stop(if_ctx_t ctx) { iflib_txq_t txq = ctx->ifc_txqs; iflib_rxq_t rxq = ctx->ifc_rxqs; if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; iflib_dma_info_t di; iflib_fl_t fl; int i, j; /* Tell the stack that the interface is no longer active */ if_setdrvflagbits(ctx->ifc_ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING); IFDI_INTR_DISABLE(ctx); DELAY(1000); IFDI_STOP(ctx); DELAY(1000); iflib_debug_reset(); /* Wait for current tx queue users to exit to disarm watchdog timer. */ for (i = 0; i < scctx->isc_ntxqsets; i++, txq++) { /* make sure all transmitters have completed before proceeding XXX */ /* clean any enqueued buffers */ iflib_ifmp_purge(txq); /* Free any existing tx buffers. */ for (j = 0; j < txq->ift_size; j++) { iflib_txsd_free(ctx, txq, j); } txq->ift_processed = txq->ift_cleaned = txq->ift_cidx_processed = 0; txq->ift_in_use = txq->ift_gen = txq->ift_cidx = txq->ift_pidx = txq->ift_no_desc_avail = 0; txq->ift_closed = txq->ift_mbuf_defrag = txq->ift_mbuf_defrag_failed = 0; txq->ift_no_tx_dma_setup = txq->ift_txd_encap_efbig = txq->ift_map_failed = 0; txq->ift_pullups = 0; ifmp_ring_reset_stats(txq->ift_br); for (j = 0, di = txq->ift_ifdi; j < ctx->ifc_nhwtxqs; j++, di++) bzero((void *)di->idi_vaddr, di->idi_size); } for (i = 0; i < scctx->isc_nrxqsets; i++, rxq++) { /* make sure all transmitters have completed before proceeding XXX */ for (j = 0, di = txq->ift_ifdi; j < ctx->ifc_nhwrxqs; j++, di++) bzero((void *)di->idi_vaddr, di->idi_size); /* also resets the free lists pidx/cidx */ for (j = 0, fl = rxq->ifr_fl; j < rxq->ifr_nfl; j++, fl++) iflib_fl_bufs_free(fl); } } static inline caddr_t calc_next_rxd(iflib_fl_t fl, int cidx) { qidx_t size; int nrxd; caddr_t start, end, cur, next; nrxd = fl->ifl_size; size = fl->ifl_rxd_size; start = fl->ifl_ifdi->idi_vaddr; if (__predict_false(size == 0)) return (start); cur = start + size*cidx; end = start + size*nrxd; next = CACHE_PTR_NEXT(cur); return (next < end ? next : start); } static inline void prefetch_pkts(iflib_fl_t fl, int cidx) { int nextptr; int nrxd = fl->ifl_size; caddr_t next_rxd; nextptr = (cidx + CACHE_PTR_INCREMENT) & (nrxd-1); prefetch(&fl->ifl_sds.ifsd_m[nextptr]); prefetch(&fl->ifl_sds.ifsd_cl[nextptr]); next_rxd = calc_next_rxd(fl, cidx); prefetch(next_rxd); prefetch(fl->ifl_sds.ifsd_m[(cidx + 1) & (nrxd-1)]); prefetch(fl->ifl_sds.ifsd_m[(cidx + 2) & (nrxd-1)]); prefetch(fl->ifl_sds.ifsd_m[(cidx + 3) & (nrxd-1)]); prefetch(fl->ifl_sds.ifsd_m[(cidx + 4) & (nrxd-1)]); prefetch(fl->ifl_sds.ifsd_cl[(cidx + 1) & (nrxd-1)]); prefetch(fl->ifl_sds.ifsd_cl[(cidx + 2) & (nrxd-1)]); prefetch(fl->ifl_sds.ifsd_cl[(cidx + 3) & (nrxd-1)]); prefetch(fl->ifl_sds.ifsd_cl[(cidx + 4) & (nrxd-1)]); } static void rxd_frag_to_sd(iflib_rxq_t rxq, if_rxd_frag_t irf, int unload, if_rxsd_t sd) { int flid, cidx; bus_dmamap_t map; iflib_fl_t fl; iflib_dma_info_t di; int next; map = NULL; flid = irf->irf_flid; cidx = irf->irf_idx; fl = &rxq->ifr_fl[flid]; sd->ifsd_fl = fl; sd->ifsd_cidx = cidx; sd->ifsd_m = &fl->ifl_sds.ifsd_m[cidx]; sd->ifsd_cl = &fl->ifl_sds.ifsd_cl[cidx]; fl->ifl_credits--; #if MEMORY_LOGGING fl->ifl_m_dequeued++; #endif if (rxq->ifr_ctx->ifc_flags & IFC_PREFETCH) prefetch_pkts(fl, cidx); if (fl->ifl_sds.ifsd_map != NULL) { next = (cidx + CACHE_PTR_INCREMENT) & (fl->ifl_size-1); prefetch(&fl->ifl_sds.ifsd_map[next]); map = fl->ifl_sds.ifsd_map[cidx]; di = fl->ifl_ifdi; next = (cidx + CACHE_LINE_SIZE) & (fl->ifl_size-1); prefetch(&fl->ifl_sds.ifsd_flags[next]); bus_dmamap_sync(di->idi_tag, di->idi_map, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); /* not valid assert if bxe really does SGE from non-contiguous elements */ MPASS(fl->ifl_cidx == cidx); if (unload) bus_dmamap_unload(fl->ifl_desc_tag, map); } fl->ifl_cidx = (fl->ifl_cidx + 1) & (fl->ifl_size-1); if (__predict_false(fl->ifl_cidx == 0)) fl->ifl_gen = 0; if (map != NULL) bus_dmamap_sync(fl->ifl_ifdi->idi_tag, fl->ifl_ifdi->idi_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); bit_clear(fl->ifl_rx_bitmap, cidx); } static struct mbuf * assemble_segments(iflib_rxq_t rxq, if_rxd_info_t ri, if_rxsd_t sd) { int i, padlen , flags; struct mbuf *m, *mh, *mt; caddr_t cl; i = 0; mh = NULL; do { rxd_frag_to_sd(rxq, &ri->iri_frags[i], TRUE, sd); MPASS(*sd->ifsd_cl != NULL); MPASS(*sd->ifsd_m != NULL); /* Don't include zero-length frags */ if (ri->iri_frags[i].irf_len == 0) { /* XXX we can save the cluster here, but not the mbuf */ m_init(*sd->ifsd_m, M_NOWAIT, MT_DATA, 0); m_free(*sd->ifsd_m); *sd->ifsd_m = NULL; continue; } m = *sd->ifsd_m; *sd->ifsd_m = NULL; if (mh == NULL) { flags = M_PKTHDR|M_EXT; mh = mt = m; padlen = ri->iri_pad; } else { flags = M_EXT; mt->m_next = m; mt = m; /* assuming padding is only on the first fragment */ padlen = 0; } cl = *sd->ifsd_cl; *sd->ifsd_cl = NULL; /* Can these two be made one ? */ m_init(m, M_NOWAIT, MT_DATA, flags); m_cljset(m, cl, sd->ifsd_fl->ifl_cltype); /* * These must follow m_init and m_cljset */ m->m_data += padlen; ri->iri_len -= padlen; m->m_len = ri->iri_frags[i].irf_len; } while (++i < ri->iri_nfrags); return (mh); } /* * Process one software descriptor */ static struct mbuf * iflib_rxd_pkt_get(iflib_rxq_t rxq, if_rxd_info_t ri) { struct if_rxsd sd; struct mbuf *m; /* should I merge this back in now that the two paths are basically duplicated? */ if (ri->iri_nfrags == 1 && ri->iri_frags[0].irf_len <= IFLIB_RX_COPY_THRESH) { rxd_frag_to_sd(rxq, &ri->iri_frags[0], FALSE, &sd); m = *sd.ifsd_m; *sd.ifsd_m = NULL; m_init(m, M_NOWAIT, MT_DATA, M_PKTHDR); #ifndef __NO_STRICT_ALIGNMENT if (!IP_ALIGNED(m)) m->m_data += 2; #endif memcpy(m->m_data, *sd.ifsd_cl, ri->iri_len); m->m_len = ri->iri_frags[0].irf_len; } else { m = assemble_segments(rxq, ri, &sd); } m->m_pkthdr.len = ri->iri_len; m->m_pkthdr.rcvif = ri->iri_ifp; m->m_flags |= ri->iri_flags; m->m_pkthdr.ether_vtag = ri->iri_vtag; m->m_pkthdr.flowid = ri->iri_flowid; M_HASHTYPE_SET(m, ri->iri_rsstype); m->m_pkthdr.csum_flags = ri->iri_csum_flags; m->m_pkthdr.csum_data = ri->iri_csum_data; return (m); } +#if defined(INET6) || defined(INET) +/* + * Returns true if it's possible this packet could be LROed. + * if it returns false, it is guaranteed that tcp_lro_rx() + * would not return zero. + */ static bool +iflib_check_lro_possible(struct lro_ctrl *lc, struct mbuf *m) +{ + struct ether_header *eh; + uint16_t eh_type; + + eh = mtod(m, struct ether_header *); + eh_type = ntohs(eh->ether_type); + switch (eh_type) { +#if defined(INET6) + case ETHERTYPE_IPV6: + { + CURVNET_SET(lc->ifp->if_vnet); + if (VNET(ip6_forwarding) == 0) { + CURVNET_RESTORE(); + return true; + } + CURVNET_RESTORE(); + break; + } +#endif +#if defined (INET) + case ETHERTYPE_IP: + { + CURVNET_SET(lc->ifp->if_vnet); + if (VNET(ipforwarding) == 0) { + CURVNET_RESTORE(); + return true; + } + CURVNET_RESTORE(); + break; + } +#endif + } + + return false; +} +#endif + +static bool iflib_rxeof(iflib_rxq_t rxq, qidx_t budget) { if_ctx_t ctx = rxq->ifr_ctx; if_shared_ctx_t sctx = ctx->ifc_sctx; if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; int avail, i; qidx_t *cidxp; struct if_rxd_info ri; int err, budget_left, rx_bytes, rx_pkts; iflib_fl_t fl; struct ifnet *ifp; int lro_enabled; + bool lro_possible = false; /* * XXX early demux data packets so that if_input processing only handles * acks in interrupt context */ struct mbuf *m, *mh, *mt, *mf; ifp = ctx->ifc_ifp; mh = mt = NULL; MPASS(budget > 0); rx_pkts = rx_bytes = 0; if (sctx->isc_flags & IFLIB_HAS_RXCQ) cidxp = &rxq->ifr_cq_cidx; else cidxp = &rxq->ifr_fl[0].ifl_cidx; if ((avail = iflib_rxd_avail(ctx, rxq, *cidxp, budget)) == 0) { for (i = 0, fl = &rxq->ifr_fl[0]; i < sctx->isc_nfl; i++, fl++) __iflib_fl_refill_lt(ctx, fl, budget + 8); DBG_COUNTER_INC(rx_unavail); return (false); } for (budget_left = budget; (budget_left > 0) && (avail > 0); budget_left--, avail--) { if (__predict_false(!CTX_ACTIVE(ctx))) { DBG_COUNTER_INC(rx_ctx_inactive); break; } /* * Reset client set fields to their default values */ rxd_info_zero(&ri); ri.iri_qsidx = rxq->ifr_id; ri.iri_cidx = *cidxp; ri.iri_ifp = ifp; ri.iri_frags = rxq->ifr_frags; err = ctx->isc_rxd_pkt_get(ctx->ifc_softc, &ri); if (err) goto err; if (sctx->isc_flags & IFLIB_HAS_RXCQ) { *cidxp = ri.iri_cidx; /* Update our consumer index */ /* XXX NB: shurd - check if this is still safe */ while (rxq->ifr_cq_cidx >= scctx->isc_nrxd[0]) { rxq->ifr_cq_cidx -= scctx->isc_nrxd[0]; rxq->ifr_cq_gen = 0; } /* was this only a completion queue message? */ if (__predict_false(ri.iri_nfrags == 0)) continue; } MPASS(ri.iri_nfrags != 0); MPASS(ri.iri_len != 0); /* will advance the cidx on the corresponding free lists */ m = iflib_rxd_pkt_get(rxq, &ri); if (avail == 0 && budget_left) avail = iflib_rxd_avail(ctx, rxq, *cidxp, budget_left); if (__predict_false(m == NULL)) { DBG_COUNTER_INC(rx_mbuf_null); continue; } /* imm_pkt: -- cxgb */ if (mh == NULL) mh = mt = m; else { mt->m_nextpkt = m; mt = m; } } /* make sure that we can refill faster than drain */ for (i = 0, fl = &rxq->ifr_fl[0]; i < sctx->isc_nfl; i++, fl++) __iflib_fl_refill_lt(ctx, fl, budget + 8); lro_enabled = (if_getcapenable(ifp) & IFCAP_LRO); mt = mf = NULL; while (mh != NULL) { m = mh; - if (mf == NULL) - mf = m; mh = mh->m_nextpkt; m->m_nextpkt = NULL; #ifndef __NO_STRICT_ALIGNMENT if (!IP_ALIGNED(m) && (m = iflib_fixup_rx(m)) == NULL) continue; #endif rx_bytes += m->m_pkthdr.len; rx_pkts++; #if defined(INET6) || defined(INET) - if (lro_enabled && tcp_lro_rx(&rxq->ifr_lc, m, 0) == 0) { - if (mf == m) - mf = NULL; - continue; + if (lro_enabled) { + if (!lro_possible) { + lro_possible = iflib_check_lro_possible(&rxq->ifr_lc, m); + if (lro_possible && mf != NULL) { + ifp->if_input(ifp, mf); + DBG_COUNTER_INC(rx_if_input); + mt = mf = NULL; + } + } + if (lro_possible && tcp_lro_rx(&rxq->ifr_lc, m, 0) == 0) + continue; } #endif + if (lro_possible) { + ifp->if_input(ifp, m); + DBG_COUNTER_INC(rx_if_input); + continue; + } + + if (mf == NULL) + mf = m; if (mt != NULL) mt->m_nextpkt = m; mt = m; } if (mf != NULL) { ifp->if_input(ifp, mf); DBG_COUNTER_INC(rx_if_input); } if_inc_counter(ifp, IFCOUNTER_IBYTES, rx_bytes); if_inc_counter(ifp, IFCOUNTER_IPACKETS, rx_pkts); /* * Flush any outstanding LRO work */ #if defined(INET6) || defined(INET) tcp_lro_flush_all(&rxq->ifr_lc); #endif if (avail) return true; return (iflib_rxd_avail(ctx, rxq, *cidxp, 1)); err: CTX_LOCK(ctx); ctx->ifc_flags |= IFC_DO_RESET; iflib_admin_intr_deferred(ctx); CTX_UNLOCK(ctx); return (false); } #define TXD_NOTIFY_COUNT(txq) (((txq)->ift_size / (txq)->ift_update_freq)-1) static inline qidx_t txq_max_db_deferred(iflib_txq_t txq, qidx_t in_use) { qidx_t notify_count = TXD_NOTIFY_COUNT(txq); qidx_t minthresh = txq->ift_size / 8; if (in_use > 4*minthresh) return (notify_count); if (in_use > 2*minthresh) return (notify_count >> 1); if (in_use > minthresh) return (notify_count >> 3); return (0); } static inline qidx_t txq_max_rs_deferred(iflib_txq_t txq) { qidx_t notify_count = TXD_NOTIFY_COUNT(txq); qidx_t minthresh = txq->ift_size / 8; if (txq->ift_in_use > 4*minthresh) return (notify_count); if (txq->ift_in_use > 2*minthresh) return (notify_count >> 1); if (txq->ift_in_use > minthresh) return (notify_count >> 2); return (2); } #define M_CSUM_FLAGS(m) ((m)->m_pkthdr.csum_flags) #define M_HAS_VLANTAG(m) (m->m_flags & M_VLANTAG) #define TXQ_MAX_DB_DEFERRED(txq, in_use) txq_max_db_deferred((txq), (in_use)) #define TXQ_MAX_RS_DEFERRED(txq) txq_max_rs_deferred(txq) #define TXQ_MAX_DB_CONSUMED(size) (size >> 4) /* forward compatibility for cxgb */ #define FIRST_QSET(ctx) 0 #define NTXQSETS(ctx) ((ctx)->ifc_softc_ctx.isc_ntxqsets) #define NRXQSETS(ctx) ((ctx)->ifc_softc_ctx.isc_nrxqsets) #define QIDX(ctx, m) ((((m)->m_pkthdr.flowid & ctx->ifc_softc_ctx.isc_rss_table_mask) % NTXQSETS(ctx)) + FIRST_QSET(ctx)) #define DESC_RECLAIMABLE(q) ((int)((q)->ift_processed - (q)->ift_cleaned - (q)->ift_ctx->ifc_softc_ctx.isc_tx_nsegments)) /* XXX we should be setting this to something other than zero */ #define RECLAIM_THRESH(ctx) ((ctx)->ifc_sctx->isc_tx_reclaim_thresh) #define MAX_TX_DESC(ctx) ((ctx)->ifc_softc_ctx.isc_tx_tso_segments_max) static inline bool iflib_txd_db_check(if_ctx_t ctx, iflib_txq_t txq, int ring, qidx_t in_use) { qidx_t dbval, max; bool rang; rang = false; max = TXQ_MAX_DB_DEFERRED(txq, in_use); if (ring || txq->ift_db_pending >= max) { dbval = txq->ift_npending ? txq->ift_npending : txq->ift_pidx; ctx->isc_txd_flush(ctx->ifc_softc, txq->ift_id, dbval); txq->ift_db_pending = txq->ift_npending = 0; rang = true; } return (rang); } #ifdef PKT_DEBUG static void print_pkt(if_pkt_info_t pi) { printf("pi len: %d qsidx: %d nsegs: %d ndescs: %d flags: %x pidx: %d\n", pi->ipi_len, pi->ipi_qsidx, pi->ipi_nsegs, pi->ipi_ndescs, pi->ipi_flags, pi->ipi_pidx); printf("pi new_pidx: %d csum_flags: %lx tso_segsz: %d mflags: %x vtag: %d\n", pi->ipi_new_pidx, pi->ipi_csum_flags, pi->ipi_tso_segsz, pi->ipi_mflags, pi->ipi_vtag); printf("pi etype: %d ehdrlen: %d ip_hlen: %d ipproto: %d\n", pi->ipi_etype, pi->ipi_ehdrlen, pi->ipi_ip_hlen, pi->ipi_ipproto); } #endif #define IS_TSO4(pi) ((pi)->ipi_csum_flags & CSUM_IP_TSO) #define IS_TSO6(pi) ((pi)->ipi_csum_flags & CSUM_IP6_TSO) static int iflib_parse_header(iflib_txq_t txq, if_pkt_info_t pi, struct mbuf **mp) { if_shared_ctx_t sctx = txq->ift_ctx->ifc_sctx; struct ether_vlan_header *eh; struct mbuf *m, *n; n = m = *mp; if ((sctx->isc_flags & IFLIB_NEED_SCRATCH) && M_WRITABLE(m) == 0) { if ((m = m_dup(m, M_NOWAIT)) == NULL) { return (ENOMEM); } else { m_freem(*mp); n = *mp = m; } } /* * Determine where frame payload starts. * Jump over vlan headers if already present, * helpful for QinQ too. */ if (__predict_false(m->m_len < sizeof(*eh))) { txq->ift_pullups++; if (__predict_false((m = m_pullup(m, sizeof(*eh))) == NULL)) return (ENOMEM); } eh = mtod(m, struct ether_vlan_header *); if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) { pi->ipi_etype = ntohs(eh->evl_proto); pi->ipi_ehdrlen = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN; } else { pi->ipi_etype = ntohs(eh->evl_encap_proto); pi->ipi_ehdrlen = ETHER_HDR_LEN; } switch (pi->ipi_etype) { #ifdef INET case ETHERTYPE_IP: { struct ip *ip = NULL; struct tcphdr *th = NULL; int minthlen; minthlen = min(m->m_pkthdr.len, pi->ipi_ehdrlen + sizeof(*ip) + sizeof(*th)); if (__predict_false(m->m_len < minthlen)) { /* * if this code bloat is causing too much of a hit * move it to a separate function and mark it noinline */ if (m->m_len == pi->ipi_ehdrlen) { n = m->m_next; MPASS(n); if (n->m_len >= sizeof(*ip)) { ip = (struct ip *)n->m_data; if (n->m_len >= (ip->ip_hl << 2) + sizeof(*th)) th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); } else { txq->ift_pullups++; if (__predict_false((m = m_pullup(m, minthlen)) == NULL)) return (ENOMEM); ip = (struct ip *)(m->m_data + pi->ipi_ehdrlen); } } else { txq->ift_pullups++; if (__predict_false((m = m_pullup(m, minthlen)) == NULL)) return (ENOMEM); ip = (struct ip *)(m->m_data + pi->ipi_ehdrlen); if (m->m_len >= (ip->ip_hl << 2) + sizeof(*th)) th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); } } else { ip = (struct ip *)(m->m_data + pi->ipi_ehdrlen); if (m->m_len >= (ip->ip_hl << 2) + sizeof(*th)) th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); } pi->ipi_ip_hlen = ip->ip_hl << 2; pi->ipi_ipproto = ip->ip_p; pi->ipi_flags |= IPI_TX_IPV4; if ((sctx->isc_flags & IFLIB_NEED_ZERO_CSUM) && (pi->ipi_csum_flags & CSUM_IP)) ip->ip_sum = 0; if (IS_TSO4(pi)) { if (pi->ipi_ipproto == IPPROTO_TCP) { if (__predict_false(th == NULL)) { txq->ift_pullups++; if (__predict_false((m = m_pullup(m, (ip->ip_hl << 2) + sizeof(*th))) == NULL)) return (ENOMEM); th = (struct tcphdr *)((caddr_t)ip + pi->ipi_ip_hlen); } pi->ipi_tcp_hflags = th->th_flags; pi->ipi_tcp_hlen = th->th_off << 2; pi->ipi_tcp_seq = th->th_seq; } if (__predict_false(ip->ip_p != IPPROTO_TCP)) return (ENXIO); th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, htons(IPPROTO_TCP)); pi->ipi_tso_segsz = m->m_pkthdr.tso_segsz; if (sctx->isc_flags & IFLIB_TSO_INIT_IP) { ip->ip_sum = 0; ip->ip_len = htons(pi->ipi_ip_hlen + pi->ipi_tcp_hlen + pi->ipi_tso_segsz); } } break; } #endif #ifdef INET6 case ETHERTYPE_IPV6: { struct ip6_hdr *ip6 = (struct ip6_hdr *)(m->m_data + pi->ipi_ehdrlen); struct tcphdr *th; pi->ipi_ip_hlen = sizeof(struct ip6_hdr); if (__predict_false(m->m_len < pi->ipi_ehdrlen + sizeof(struct ip6_hdr))) { if (__predict_false((m = m_pullup(m, pi->ipi_ehdrlen + sizeof(struct ip6_hdr))) == NULL)) return (ENOMEM); } th = (struct tcphdr *)((caddr_t)ip6 + pi->ipi_ip_hlen); /* XXX-BZ this will go badly in case of ext hdrs. */ pi->ipi_ipproto = ip6->ip6_nxt; pi->ipi_flags |= IPI_TX_IPV6; if (IS_TSO6(pi)) { if (pi->ipi_ipproto == IPPROTO_TCP) { if (__predict_false(m->m_len < pi->ipi_ehdrlen + sizeof(struct ip6_hdr) + sizeof(struct tcphdr))) { if (__predict_false((m = m_pullup(m, pi->ipi_ehdrlen + sizeof(struct ip6_hdr) + sizeof(struct tcphdr))) == NULL)) return (ENOMEM); } pi->ipi_tcp_hflags = th->th_flags; pi->ipi_tcp_hlen = th->th_off << 2; } if (__predict_false(ip6->ip6_nxt != IPPROTO_TCP)) return (ENXIO); /* * The corresponding flag is set by the stack in the IPv4 * TSO case, but not in IPv6 (at least in FreeBSD 10.2). * So, set it here because the rest of the flow requires it. */ pi->ipi_csum_flags |= CSUM_TCP_IPV6; th->th_sum = in6_cksum_pseudo(ip6, 0, IPPROTO_TCP, 0); pi->ipi_tso_segsz = m->m_pkthdr.tso_segsz; } break; } #endif default: pi->ipi_csum_flags &= ~CSUM_OFFLOAD; pi->ipi_ip_hlen = 0; break; } *mp = m; return (0); } static __noinline struct mbuf * collapse_pkthdr(struct mbuf *m0) { struct mbuf *m, *m_next, *tmp; m = m0; m_next = m->m_next; while (m_next != NULL && m_next->m_len == 0) { m = m_next; m->m_next = NULL; m_free(m); m_next = m_next->m_next; } m = m0; m->m_next = m_next; if ((m_next->m_flags & M_EXT) == 0) { m = m_defrag(m, M_NOWAIT); } else { tmp = m_next->m_next; memcpy(m_next, m, MPKTHSIZE); m = m_next; m->m_next = tmp; } return (m); } /* * If dodgy hardware rejects the scatter gather chain we've handed it * we'll need to remove the mbuf chain from ifsg_m[] before we can add the * m_defrag'd mbufs */ static __noinline struct mbuf * iflib_remove_mbuf(iflib_txq_t txq) { int ntxd, i, pidx; struct mbuf *m, *mh, **ifsd_m; pidx = txq->ift_pidx; ifsd_m = txq->ift_sds.ifsd_m; ntxd = txq->ift_size; mh = m = ifsd_m[pidx]; ifsd_m[pidx] = NULL; #if MEMORY_LOGGING txq->ift_dequeued++; #endif i = 1; while (m) { ifsd_m[(pidx + i) & (ntxd -1)] = NULL; #if MEMORY_LOGGING txq->ift_dequeued++; #endif m = m->m_next; i++; } return (mh); } static int iflib_busdma_load_mbuf_sg(iflib_txq_t txq, bus_dma_tag_t tag, bus_dmamap_t map, struct mbuf **m0, bus_dma_segment_t *segs, int *nsegs, int max_segs, int flags) { if_ctx_t ctx; if_shared_ctx_t sctx; if_softc_ctx_t scctx; int i, next, pidx, err, ntxd, count; struct mbuf *m, *tmp, **ifsd_m; m = *m0; /* * Please don't ever do this */ if (__predict_false(m->m_len == 0)) *m0 = m = collapse_pkthdr(m); ctx = txq->ift_ctx; sctx = ctx->ifc_sctx; scctx = &ctx->ifc_softc_ctx; ifsd_m = txq->ift_sds.ifsd_m; ntxd = txq->ift_size; pidx = txq->ift_pidx; if (map != NULL) { uint8_t *ifsd_flags = txq->ift_sds.ifsd_flags; err = bus_dmamap_load_mbuf_sg(tag, map, *m0, segs, nsegs, BUS_DMA_NOWAIT); if (err) return (err); ifsd_flags[pidx] |= TX_SW_DESC_MAPPED; count = 0; m = *m0; do { if (__predict_false(m->m_len <= 0)) { tmp = m; m = m->m_next; tmp->m_next = NULL; m_free(tmp); continue; } m = m->m_next; count++; } while (m != NULL); if (count > *nsegs) { ifsd_m[pidx] = *m0; ifsd_m[pidx]->m_flags |= M_TOOBIG; return (0); } m = *m0; count = 0; do { next = (pidx + count) & (ntxd-1); MPASS(ifsd_m[next] == NULL); ifsd_m[next] = m; count++; tmp = m; m = m->m_next; } while (m != NULL); } else { int buflen, sgsize, maxsegsz, max_sgsize; vm_offset_t vaddr; vm_paddr_t curaddr; count = i = 0; m = *m0; if (m->m_pkthdr.csum_flags & CSUM_TSO) maxsegsz = scctx->isc_tx_tso_segsize_max; else maxsegsz = sctx->isc_tx_maxsegsize; do { if (__predict_false(m->m_len <= 0)) { tmp = m; m = m->m_next; tmp->m_next = NULL; m_free(tmp); continue; } buflen = m->m_len; vaddr = (vm_offset_t)m->m_data; /* * see if we can't be smarter about physically * contiguous mappings */ next = (pidx + count) & (ntxd-1); MPASS(ifsd_m[next] == NULL); #if MEMORY_LOGGING txq->ift_enqueued++; #endif ifsd_m[next] = m; while (buflen > 0) { if (i >= max_segs) goto err; max_sgsize = MIN(buflen, maxsegsz); curaddr = pmap_kextract(vaddr); sgsize = PAGE_SIZE - (curaddr & PAGE_MASK); sgsize = MIN(sgsize, max_sgsize); segs[i].ds_addr = curaddr; segs[i].ds_len = sgsize; vaddr += sgsize; buflen -= sgsize; i++; } count++; tmp = m; m = m->m_next; } while (m != NULL); *nsegs = i; } return (0); err: *m0 = iflib_remove_mbuf(txq); return (EFBIG); } static inline caddr_t calc_next_txd(iflib_txq_t txq, int cidx, uint8_t qid) { qidx_t size; int ntxd; caddr_t start, end, cur, next; ntxd = txq->ift_size; size = txq->ift_txd_size[qid]; start = txq->ift_ifdi[qid].idi_vaddr; if (__predict_false(size == 0)) return (start); cur = start + size*cidx; end = start + size*ntxd; next = CACHE_PTR_NEXT(cur); return (next < end ? next : start); } static int iflib_encap(iflib_txq_t txq, struct mbuf **m_headp) { if_ctx_t ctx; if_shared_ctx_t sctx; if_softc_ctx_t scctx; bus_dma_segment_t *segs; struct mbuf *m_head; void *next_txd; bus_dmamap_t map; struct if_pkt_info pi; int remap = 0; int err, nsegs, ndesc, max_segs, pidx, cidx, next, ntxd; bus_dma_tag_t desc_tag; segs = txq->ift_segs; ctx = txq->ift_ctx; sctx = ctx->ifc_sctx; scctx = &ctx->ifc_softc_ctx; segs = txq->ift_segs; ntxd = txq->ift_size; m_head = *m_headp; map = NULL; /* * If we're doing TSO the next descriptor to clean may be quite far ahead */ cidx = txq->ift_cidx; pidx = txq->ift_pidx; if (ctx->ifc_flags & IFC_PREFETCH) { next = (cidx + CACHE_PTR_INCREMENT) & (ntxd-1); if (!(ctx->ifc_flags & IFLIB_HAS_TXCQ)) { next_txd = calc_next_txd(txq, cidx, 0); prefetch(next_txd); } /* prefetch the next cache line of mbuf pointers and flags */ prefetch(&txq->ift_sds.ifsd_m[next]); if (txq->ift_sds.ifsd_map != NULL) { prefetch(&txq->ift_sds.ifsd_map[next]); next = (cidx + CACHE_LINE_SIZE) & (ntxd-1); prefetch(&txq->ift_sds.ifsd_flags[next]); } } else if (txq->ift_sds.ifsd_map != NULL) map = txq->ift_sds.ifsd_map[pidx]; if (m_head->m_pkthdr.csum_flags & CSUM_TSO) { desc_tag = txq->ift_tso_desc_tag; max_segs = scctx->isc_tx_tso_segments_max; } else { desc_tag = txq->ift_desc_tag; max_segs = scctx->isc_tx_nsegments; } m_head = *m_headp; pkt_info_zero(&pi); pi.ipi_mflags = (m_head->m_flags & (M_VLANTAG|M_BCAST|M_MCAST)); pi.ipi_pidx = pidx; pi.ipi_qsidx = txq->ift_id; pi.ipi_len = m_head->m_pkthdr.len; pi.ipi_csum_flags = m_head->m_pkthdr.csum_flags; pi.ipi_vtag = (m_head->m_flags & M_VLANTAG) ? m_head->m_pkthdr.ether_vtag : 0; /* deliberate bitwise OR to make one condition */ if (__predict_true((pi.ipi_csum_flags | pi.ipi_vtag))) { if (__predict_false((err = iflib_parse_header(txq, &pi, m_headp)) != 0)) return (err); m_head = *m_headp; } retry: err = iflib_busdma_load_mbuf_sg(txq, desc_tag, map, m_headp, segs, &nsegs, max_segs, BUS_DMA_NOWAIT); defrag: if (__predict_false(err)) { switch (err) { case EFBIG: /* try collapse once and defrag once */ if (remap == 0) m_head = m_collapse(*m_headp, M_NOWAIT, max_segs); if (remap == 1) m_head = m_defrag(*m_headp, M_NOWAIT); remap++; if (__predict_false(m_head == NULL)) goto defrag_failed; txq->ift_mbuf_defrag++; *m_headp = m_head; goto retry; break; case ENOMEM: txq->ift_no_tx_dma_setup++; break; default: txq->ift_no_tx_dma_setup++; m_freem(*m_headp); DBG_COUNTER_INC(tx_frees); *m_headp = NULL; break; } txq->ift_map_failed++; DBG_COUNTER_INC(encap_load_mbuf_fail); return (err); } /* * XXX assumes a 1 to 1 relationship between segments and * descriptors - this does not hold true on all drivers, e.g. * cxgb */ if (__predict_false(nsegs + 2 > TXQ_AVAIL(txq))) { txq->ift_no_desc_avail++; if (map != NULL) bus_dmamap_unload(desc_tag, map); DBG_COUNTER_INC(encap_txq_avail_fail); if ((txq->ift_task.gt_task.ta_flags & TASK_ENQUEUED) == 0) GROUPTASK_ENQUEUE(&txq->ift_task); return (ENOBUFS); } /* * On Intel cards we can greatly reduce the number of TX interrupts * we see by only setting report status on every Nth descriptor. * However, this also means that the driver will need to keep track * of the descriptors that RS was set on to check them for the DD bit. */ txq->ift_rs_pending += nsegs + 1; if (txq->ift_rs_pending > TXQ_MAX_RS_DEFERRED(txq) || iflib_no_tx_batch || (TXQ_AVAIL(txq) - nsegs - 1) <= MAX_TX_DESC(ctx)) { pi.ipi_flags |= IPI_TX_INTR; txq->ift_rs_pending = 0; } pi.ipi_segs = segs; pi.ipi_nsegs = nsegs; MPASS(pidx >= 0 && pidx < txq->ift_size); #ifdef PKT_DEBUG print_pkt(&pi); #endif if (map != NULL) bus_dmamap_sync(desc_tag, map, BUS_DMASYNC_PREWRITE); if ((err = ctx->isc_txd_encap(ctx->ifc_softc, &pi)) == 0) { if (map != NULL) bus_dmamap_sync(txq->ift_ifdi->idi_tag, txq->ift_ifdi->idi_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); DBG_COUNTER_INC(tx_encap); MPASS(pi.ipi_new_pidx < txq->ift_size); ndesc = pi.ipi_new_pidx - pi.ipi_pidx; if (pi.ipi_new_pidx < pi.ipi_pidx) { ndesc += txq->ift_size; txq->ift_gen = 1; } /* * drivers can need as many as * two sentinels */ MPASS(ndesc <= pi.ipi_nsegs + 2); MPASS(pi.ipi_new_pidx != pidx); MPASS(ndesc > 0); txq->ift_in_use += ndesc; /* * We update the last software descriptor again here because there may * be a sentinel and/or there may be more mbufs than segments */ txq->ift_pidx = pi.ipi_new_pidx; txq->ift_npending += pi.ipi_ndescs; } else if (__predict_false(err == EFBIG && remap < 2)) { *m_headp = m_head = iflib_remove_mbuf(txq); remap = 1; txq->ift_txd_encap_efbig++; goto defrag; } else DBG_COUNTER_INC(encap_txd_encap_fail); return (err); defrag_failed: txq->ift_mbuf_defrag_failed++; txq->ift_map_failed++; m_freem(*m_headp); DBG_COUNTER_INC(tx_frees); *m_headp = NULL; return (ENOMEM); } static void iflib_tx_desc_free(iflib_txq_t txq, int n) { int hasmap; uint32_t qsize, cidx, mask, gen; struct mbuf *m, **ifsd_m; uint8_t *ifsd_flags; bus_dmamap_t *ifsd_map; bool do_prefetch; cidx = txq->ift_cidx; gen = txq->ift_gen; qsize = txq->ift_size; mask = qsize-1; hasmap = txq->ift_sds.ifsd_map != NULL; ifsd_flags = txq->ift_sds.ifsd_flags; ifsd_m = txq->ift_sds.ifsd_m; ifsd_map = txq->ift_sds.ifsd_map; do_prefetch = (txq->ift_ctx->ifc_flags & IFC_PREFETCH); while (n--) { if (do_prefetch) { prefetch(ifsd_m[(cidx + 3) & mask]); prefetch(ifsd_m[(cidx + 4) & mask]); } if (ifsd_m[cidx] != NULL) { prefetch(&ifsd_m[(cidx + CACHE_PTR_INCREMENT) & mask]); prefetch(&ifsd_flags[(cidx + CACHE_PTR_INCREMENT) & mask]); if (hasmap && (ifsd_flags[cidx] & TX_SW_DESC_MAPPED)) { /* * does it matter if it's not the TSO tag? If so we'll * have to add the type to flags */ bus_dmamap_unload(txq->ift_desc_tag, ifsd_map[cidx]); ifsd_flags[cidx] &= ~TX_SW_DESC_MAPPED; } if ((m = ifsd_m[cidx]) != NULL) { /* XXX we don't support any drivers that batch packets yet */ MPASS(m->m_nextpkt == NULL); /* if the number of clusters exceeds the number of segments * there won't be space on the ring to save a pointer to each * cluster so we simply free the list here */ if (m->m_flags & M_TOOBIG) { m_freem(m); } else { m_free(m); } ifsd_m[cidx] = NULL; #if MEMORY_LOGGING txq->ift_dequeued++; #endif DBG_COUNTER_INC(tx_frees); } } if (__predict_false(++cidx == qsize)) { cidx = 0; gen = 0; } } txq->ift_cidx = cidx; txq->ift_gen = gen; } static __inline int iflib_completed_tx_reclaim(iflib_txq_t txq, int thresh) { int reclaim; if_ctx_t ctx = txq->ift_ctx; KASSERT(thresh >= 0, ("invalid threshold to reclaim")); MPASS(thresh /*+ MAX_TX_DESC(txq->ift_ctx) */ < txq->ift_size); /* * Need a rate-limiting check so that this isn't called every time */ iflib_tx_credits_update(ctx, txq); reclaim = DESC_RECLAIMABLE(txq); if (reclaim <= thresh /* + MAX_TX_DESC(txq->ift_ctx) */) { #ifdef INVARIANTS if (iflib_verbose_debug) { printf("%s processed=%ju cleaned=%ju tx_nsegments=%d reclaim=%d thresh=%d\n", __FUNCTION__, txq->ift_processed, txq->ift_cleaned, txq->ift_ctx->ifc_softc_ctx.isc_tx_nsegments, reclaim, thresh); } #endif return (0); } iflib_tx_desc_free(txq, reclaim); txq->ift_cleaned += reclaim; txq->ift_in_use -= reclaim; return (reclaim); } static struct mbuf ** _ring_peek_one(struct ifmp_ring *r, int cidx, int offset, int remaining) { int next, size; struct mbuf **items; size = r->size; next = (cidx + CACHE_PTR_INCREMENT) & (size-1); items = __DEVOLATILE(struct mbuf **, &r->items[0]); prefetch(items[(cidx + offset) & (size-1)]); if (remaining > 1) { prefetch2cachelines(&items[next]); prefetch2cachelines(items[(cidx + offset + 1) & (size-1)]); prefetch2cachelines(items[(cidx + offset + 2) & (size-1)]); prefetch2cachelines(items[(cidx + offset + 3) & (size-1)]); } return (__DEVOLATILE(struct mbuf **, &r->items[(cidx + offset) & (size-1)])); } static void iflib_txq_check_drain(iflib_txq_t txq, int budget) { ifmp_ring_check_drainage(txq->ift_br, budget); } static uint32_t iflib_txq_can_drain(struct ifmp_ring *r) { iflib_txq_t txq = r->cookie; if_ctx_t ctx = txq->ift_ctx; return ((TXQ_AVAIL(txq) > MAX_TX_DESC(ctx) + 2) || ctx->isc_txd_credits_update(ctx->ifc_softc, txq->ift_id, false)); } static uint32_t iflib_txq_drain(struct ifmp_ring *r, uint32_t cidx, uint32_t pidx) { iflib_txq_t txq = r->cookie; if_ctx_t ctx = txq->ift_ctx; struct ifnet *ifp = ctx->ifc_ifp; struct mbuf **mp, *m; int i, count, consumed, pkt_sent, bytes_sent, mcast_sent, avail; int reclaimed, err, in_use_prev, desc_used; bool do_prefetch, ring, rang; if (__predict_false(!(if_getdrvflags(ifp) & IFF_DRV_RUNNING) || !LINK_ACTIVE(ctx))) { DBG_COUNTER_INC(txq_drain_notready); return (0); } reclaimed = iflib_completed_tx_reclaim(txq, RECLAIM_THRESH(ctx)); rang = iflib_txd_db_check(ctx, txq, reclaimed, txq->ift_in_use); avail = IDXDIFF(pidx, cidx, r->size); if (__predict_false(ctx->ifc_flags & IFC_QFLUSH)) { DBG_COUNTER_INC(txq_drain_flushing); for (i = 0; i < avail; i++) { m_free(r->items[(cidx + i) & (r->size-1)]); r->items[(cidx + i) & (r->size-1)] = NULL; } return (avail); } if (__predict_false(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_OACTIVE)) { txq->ift_qstatus = IFLIB_QUEUE_IDLE; CALLOUT_LOCK(txq); callout_stop(&txq->ift_timer); CALLOUT_UNLOCK(txq); DBG_COUNTER_INC(txq_drain_oactive); return (0); } if (reclaimed) txq->ift_qstatus = IFLIB_QUEUE_IDLE; consumed = mcast_sent = bytes_sent = pkt_sent = 0; count = MIN(avail, TX_BATCH_SIZE); #ifdef INVARIANTS if (iflib_verbose_debug) printf("%s avail=%d ifc_flags=%x txq_avail=%d ", __FUNCTION__, avail, ctx->ifc_flags, TXQ_AVAIL(txq)); #endif do_prefetch = (ctx->ifc_flags & IFC_PREFETCH); avail = TXQ_AVAIL(txq); for (desc_used = i = 0; i < count && avail > MAX_TX_DESC(ctx) + 2; i++) { int pidx_prev, rem = do_prefetch ? count - i : 0; mp = _ring_peek_one(r, cidx, i, rem); MPASS(mp != NULL && *mp != NULL); if (__predict_false(*mp == (struct mbuf *)txq)) { consumed++; reclaimed++; continue; } in_use_prev = txq->ift_in_use; pidx_prev = txq->ift_pidx; err = iflib_encap(txq, mp); if (__predict_false(err)) { DBG_COUNTER_INC(txq_drain_encapfail); /* no room - bail out */ if (err == ENOBUFS) break; consumed++; DBG_COUNTER_INC(txq_drain_encapfail); /* we can't send this packet - skip it */ continue; } consumed++; pkt_sent++; m = *mp; DBG_COUNTER_INC(tx_sent); bytes_sent += m->m_pkthdr.len; mcast_sent += !!(m->m_flags & M_MCAST); avail = TXQ_AVAIL(txq); txq->ift_db_pending += (txq->ift_in_use - in_use_prev); desc_used += (txq->ift_in_use - in_use_prev); ETHER_BPF_MTAP(ifp, m); if (__predict_false(!(ifp->if_drv_flags & IFF_DRV_RUNNING))) break; rang = iflib_txd_db_check(ctx, txq, false, in_use_prev); } /* deliberate use of bitwise or to avoid gratuitous short-circuit */ ring = rang ? false : (iflib_min_tx_latency | err) || (TXQ_AVAIL(txq) < MAX_TX_DESC(ctx)); iflib_txd_db_check(ctx, txq, ring, txq->ift_in_use); if_inc_counter(ifp, IFCOUNTER_OBYTES, bytes_sent); if_inc_counter(ifp, IFCOUNTER_OPACKETS, pkt_sent); if (mcast_sent) if_inc_counter(ifp, IFCOUNTER_OMCASTS, mcast_sent); #ifdef INVARIANTS if (iflib_verbose_debug) printf("consumed=%d\n", consumed); #endif return (consumed); } static uint32_t iflib_txq_drain_always(struct ifmp_ring *r) { return (1); } static uint32_t iflib_txq_drain_free(struct ifmp_ring *r, uint32_t cidx, uint32_t pidx) { int i, avail; struct mbuf **mp; iflib_txq_t txq; txq = r->cookie; txq->ift_qstatus = IFLIB_QUEUE_IDLE; CALLOUT_LOCK(txq); callout_stop(&txq->ift_timer); CALLOUT_UNLOCK(txq); avail = IDXDIFF(pidx, cidx, r->size); for (i = 0; i < avail; i++) { mp = _ring_peek_one(r, cidx, i, avail - i); if (__predict_false(*mp == (struct mbuf *)txq)) continue; m_freem(*mp); } MPASS(ifmp_ring_is_stalled(r) == 0); return (avail); } static void iflib_ifmp_purge(iflib_txq_t txq) { struct ifmp_ring *r; r = txq->ift_br; r->drain = iflib_txq_drain_free; r->can_drain = iflib_txq_drain_always; ifmp_ring_check_drainage(r, r->size); r->drain = iflib_txq_drain; r->can_drain = iflib_txq_can_drain; } static void _task_fn_tx(void *context) { iflib_txq_t txq = context; if_ctx_t ctx = txq->ift_ctx; struct ifnet *ifp = ctx->ifc_ifp; int rc; #ifdef IFLIB_DIAGNOSTICS txq->ift_cpu_exec_count[curcpu]++; #endif if (!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING)) return; if (if_getcapenable(ifp) & IFCAP_NETMAP) { if (ctx->isc_txd_credits_update(ctx->ifc_softc, txq->ift_id, false)) netmap_tx_irq(ifp, txq->ift_id); IFDI_TX_QUEUE_INTR_ENABLE(ctx, txq->ift_id); return; } if (txq->ift_db_pending) ifmp_ring_enqueue(txq->ift_br, (void **)&txq, 1, TX_BATCH_SIZE); ifmp_ring_check_drainage(txq->ift_br, TX_BATCH_SIZE); if (ctx->ifc_flags & IFC_LEGACY) IFDI_INTR_ENABLE(ctx); else { rc = IFDI_TX_QUEUE_INTR_ENABLE(ctx, txq->ift_id); KASSERT(rc != ENOTSUP, ("MSI-X support requires queue_intr_enable, but not implemented in driver")); } } static void _task_fn_rx(void *context) { iflib_rxq_t rxq = context; if_ctx_t ctx = rxq->ifr_ctx; bool more; int rc; uint16_t budget; #ifdef IFLIB_DIAGNOSTICS rxq->ifr_cpu_exec_count[curcpu]++; #endif DBG_COUNTER_INC(task_fn_rxs); if (__predict_false(!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING))) return; more = true; #ifdef DEV_NETMAP if (if_getcapenable(ctx->ifc_ifp) & IFCAP_NETMAP) { u_int work = 0; if (netmap_rx_irq(ctx->ifc_ifp, rxq->ifr_id, &work)) { more = false; } } #endif budget = ctx->ifc_sysctl_rx_budget; if (budget == 0) budget = 16; /* XXX */ if (more == false || (more = iflib_rxeof(rxq, budget)) == false) { if (ctx->ifc_flags & IFC_LEGACY) IFDI_INTR_ENABLE(ctx); else { DBG_COUNTER_INC(rx_intr_enables); rc = IFDI_RX_QUEUE_INTR_ENABLE(ctx, rxq->ifr_id); KASSERT(rc != ENOTSUP, ("MSI-X support requires queue_intr_enable, but not implemented in driver")); } } if (__predict_false(!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING))) return; if (more) GROUPTASK_ENQUEUE(&rxq->ifr_task); } static void _task_fn_admin(void *context) { if_ctx_t ctx = context; if_softc_ctx_t sctx = &ctx->ifc_softc_ctx; iflib_txq_t txq; int i; if (!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING)) { if (!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_OACTIVE)) { return; } } CTX_LOCK(ctx); for (txq = ctx->ifc_txqs, i = 0; i < sctx->isc_ntxqsets; i++, txq++) { CALLOUT_LOCK(txq); callout_stop(&txq->ift_timer); CALLOUT_UNLOCK(txq); } IFDI_UPDATE_ADMIN_STATUS(ctx); for (txq = ctx->ifc_txqs, i = 0; i < sctx->isc_ntxqsets; i++, txq++) callout_reset_on(&txq->ift_timer, hz/2, iflib_timer, txq, txq->ift_timer.c_cpu); IFDI_LINK_INTR_ENABLE(ctx); if (ctx->ifc_flags & IFC_DO_RESET) { ctx->ifc_flags &= ~IFC_DO_RESET; iflib_if_init_locked(ctx); } CTX_UNLOCK(ctx); if (LINK_ACTIVE(ctx) == 0) return; for (txq = ctx->ifc_txqs, i = 0; i < sctx->isc_ntxqsets; i++, txq++) iflib_txq_check_drain(txq, IFLIB_RESTART_BUDGET); } static void _task_fn_iov(void *context) { if_ctx_t ctx = context; if (!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING)) return; CTX_LOCK(ctx); IFDI_VFLR_HANDLE(ctx); CTX_UNLOCK(ctx); } static int iflib_sysctl_int_delay(SYSCTL_HANDLER_ARGS) { int err; if_int_delay_info_t info; if_ctx_t ctx; info = (if_int_delay_info_t)arg1; ctx = info->iidi_ctx; info->iidi_req = req; info->iidi_oidp = oidp; CTX_LOCK(ctx); err = IFDI_SYSCTL_INT_DELAY(ctx, info); CTX_UNLOCK(ctx); return (err); } /********************************************************************* * * IFNET FUNCTIONS * **********************************************************************/ static void iflib_if_init_locked(if_ctx_t ctx) { iflib_stop(ctx); iflib_init_locked(ctx); } static void iflib_if_init(void *arg) { if_ctx_t ctx = arg; CTX_LOCK(ctx); iflib_if_init_locked(ctx); CTX_UNLOCK(ctx); } static int iflib_if_transmit(if_t ifp, struct mbuf *m) { if_ctx_t ctx = if_getsoftc(ifp); iflib_txq_t txq; int err, qidx; if (__predict_false((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || !LINK_ACTIVE(ctx))) { DBG_COUNTER_INC(tx_frees); m_freem(m); return (ENOBUFS); } MPASS(m->m_nextpkt == NULL); qidx = 0; if ((NTXQSETS(ctx) > 1) && M_HASHTYPE_GET(m)) qidx = QIDX(ctx, m); /* * XXX calculate buf_ring based on flowid (divvy up bits?) */ txq = &ctx->ifc_txqs[qidx]; #ifdef DRIVER_BACKPRESSURE if (txq->ift_closed) { while (m != NULL) { next = m->m_nextpkt; m->m_nextpkt = NULL; m_freem(m); m = next; } return (ENOBUFS); } #endif #ifdef notyet qidx = count = 0; mp = marr; next = m; do { count++; next = next->m_nextpkt; } while (next != NULL); if (count > nitems(marr)) if ((mp = malloc(count*sizeof(struct mbuf *), M_IFLIB, M_NOWAIT)) == NULL) { /* XXX check nextpkt */ m_freem(m); /* XXX simplify for now */ DBG_COUNTER_INC(tx_frees); return (ENOBUFS); } for (next = m, i = 0; next != NULL; i++) { mp[i] = next; next = next->m_nextpkt; mp[i]->m_nextpkt = NULL; } #endif DBG_COUNTER_INC(tx_seen); err = ifmp_ring_enqueue(txq->ift_br, (void **)&m, 1, TX_BATCH_SIZE); GROUPTASK_ENQUEUE(&txq->ift_task); if (err) { /* support forthcoming later */ #ifdef DRIVER_BACKPRESSURE txq->ift_closed = TRUE; #endif ifmp_ring_check_drainage(txq->ift_br, TX_BATCH_SIZE); m_freem(m); } return (err); } static void iflib_if_qflush(if_t ifp) { if_ctx_t ctx = if_getsoftc(ifp); iflib_txq_t txq = ctx->ifc_txqs; int i; CTX_LOCK(ctx); ctx->ifc_flags |= IFC_QFLUSH; CTX_UNLOCK(ctx); for (i = 0; i < NTXQSETS(ctx); i++, txq++) while (!(ifmp_ring_is_idle(txq->ift_br) || ifmp_ring_is_stalled(txq->ift_br))) iflib_txq_check_drain(txq, 0); CTX_LOCK(ctx); ctx->ifc_flags &= ~IFC_QFLUSH; CTX_UNLOCK(ctx); if_qflush(ifp); } #define IFCAP_FLAGS (IFCAP_TXCSUM_IPV6 | IFCAP_RXCSUM_IPV6 | IFCAP_HWCSUM | IFCAP_LRO | \ IFCAP_TSO4 | IFCAP_TSO6 | IFCAP_VLAN_HWTAGGING | IFCAP_HWSTATS | \ IFCAP_VLAN_MTU | IFCAP_VLAN_HWFILTER | IFCAP_VLAN_HWTSO) static int iflib_if_ioctl(if_t ifp, u_long command, caddr_t data) { if_ctx_t ctx = if_getsoftc(ifp); struct ifreq *ifr = (struct ifreq *)data; #if defined(INET) || defined(INET6) struct ifaddr *ifa = (struct ifaddr *)data; #endif bool avoid_reset = FALSE; int err = 0, reinit = 0, bits; switch (command) { case SIOCSIFADDR: #ifdef INET if (ifa->ifa_addr->sa_family == AF_INET) avoid_reset = TRUE; #endif #ifdef INET6 if (ifa->ifa_addr->sa_family == AF_INET6) avoid_reset = TRUE; #endif /* ** Calling init results in link renegotiation, ** so we avoid doing it when possible. */ if (avoid_reset) { if_setflagbits(ifp, IFF_UP,0); if (!(if_getdrvflags(ifp)& IFF_DRV_RUNNING)) reinit = 1; #ifdef INET if (!(if_getflags(ifp) & IFF_NOARP)) arp_ifinit(ifp, ifa); #endif } else err = ether_ioctl(ifp, command, data); break; case SIOCSIFMTU: CTX_LOCK(ctx); if (ifr->ifr_mtu == if_getmtu(ifp)) { CTX_UNLOCK(ctx); break; } bits = if_getdrvflags(ifp); /* stop the driver and free any clusters before proceeding */ iflib_stop(ctx); if ((err = IFDI_MTU_SET(ctx, ifr->ifr_mtu)) == 0) { if (ifr->ifr_mtu > ctx->ifc_max_fl_buf_size) ctx->ifc_flags |= IFC_MULTISEG; else ctx->ifc_flags &= ~IFC_MULTISEG; err = if_setmtu(ifp, ifr->ifr_mtu); } iflib_init_locked(ctx); if_setdrvflags(ifp, bits); CTX_UNLOCK(ctx); break; case SIOCSIFFLAGS: CTX_LOCK(ctx); if (if_getflags(ifp) & IFF_UP) { if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) { if ((if_getflags(ifp) ^ ctx->ifc_if_flags) & (IFF_PROMISC | IFF_ALLMULTI)) { err = IFDI_PROMISC_SET(ctx, if_getflags(ifp)); } } else reinit = 1; } else if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) { iflib_stop(ctx); } ctx->ifc_if_flags = if_getflags(ifp); CTX_UNLOCK(ctx); break; case SIOCADDMULTI: case SIOCDELMULTI: if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) { CTX_LOCK(ctx); IFDI_INTR_DISABLE(ctx); IFDI_MULTI_SET(ctx); IFDI_INTR_ENABLE(ctx); CTX_UNLOCK(ctx); } break; case SIOCSIFMEDIA: CTX_LOCK(ctx); IFDI_MEDIA_SET(ctx); CTX_UNLOCK(ctx); /* falls thru */ case SIOCGIFMEDIA: err = ifmedia_ioctl(ifp, ifr, &ctx->ifc_media, command); break; case SIOCGI2C: { struct ifi2creq i2c; err = copyin(ifr->ifr_data, &i2c, sizeof(i2c)); if (err != 0) break; if (i2c.dev_addr != 0xA0 && i2c.dev_addr != 0xA2) { err = EINVAL; break; } if (i2c.len > sizeof(i2c.data)) { err = EINVAL; break; } if ((err = IFDI_I2C_REQ(ctx, &i2c)) == 0) err = copyout(&i2c, ifr->ifr_data, sizeof(i2c)); break; } case SIOCSIFCAP: { int mask, setmask; mask = ifr->ifr_reqcap ^ if_getcapenable(ifp); setmask = 0; #ifdef TCP_OFFLOAD setmask |= mask & (IFCAP_TOE4|IFCAP_TOE6); #endif setmask |= (mask & IFCAP_FLAGS); if (setmask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) setmask |= (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6); if ((mask & IFCAP_WOL) && (if_getcapabilities(ifp) & IFCAP_WOL) != 0) setmask |= (mask & (IFCAP_WOL_MCAST|IFCAP_WOL_MAGIC)); if_vlancap(ifp); /* * want to ensure that traffic has stopped before we change any of the flags */ if (setmask) { CTX_LOCK(ctx); bits = if_getdrvflags(ifp); if (bits & IFF_DRV_RUNNING) iflib_stop(ctx); if_togglecapenable(ifp, setmask); if (bits & IFF_DRV_RUNNING) iflib_init_locked(ctx); if_setdrvflags(ifp, bits); CTX_UNLOCK(ctx); } break; } case SIOCGPRIVATE_0: case SIOCSDRVSPEC: case SIOCGDRVSPEC: CTX_LOCK(ctx); err = IFDI_PRIV_IOCTL(ctx, command, data); CTX_UNLOCK(ctx); break; default: err = ether_ioctl(ifp, command, data); break; } if (reinit) iflib_if_init(ctx); return (err); } static uint64_t iflib_if_get_counter(if_t ifp, ift_counter cnt) { if_ctx_t ctx = if_getsoftc(ifp); return (IFDI_GET_COUNTER(ctx, cnt)); } /********************************************************************* * * OTHER FUNCTIONS EXPORTED TO THE STACK * **********************************************************************/ static void iflib_vlan_register(void *arg, if_t ifp, uint16_t vtag) { if_ctx_t ctx = if_getsoftc(ifp); if ((void *)ctx != arg) return; if ((vtag == 0) || (vtag > 4095)) return; CTX_LOCK(ctx); IFDI_VLAN_REGISTER(ctx, vtag); /* Re-init to load the changes */ if (if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER) iflib_if_init_locked(ctx); CTX_UNLOCK(ctx); } static void iflib_vlan_unregister(void *arg, if_t ifp, uint16_t vtag) { if_ctx_t ctx = if_getsoftc(ifp); if ((void *)ctx != arg) return; if ((vtag == 0) || (vtag > 4095)) return; CTX_LOCK(ctx); IFDI_VLAN_UNREGISTER(ctx, vtag); /* Re-init to load the changes */ if (if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER) iflib_if_init_locked(ctx); CTX_UNLOCK(ctx); } static void iflib_led_func(void *arg, int onoff) { if_ctx_t ctx = arg; CTX_LOCK(ctx); IFDI_LED_FUNC(ctx, onoff); CTX_UNLOCK(ctx); } /********************************************************************* * * BUS FUNCTION DEFINITIONS * **********************************************************************/ int iflib_device_probe(device_t dev) { pci_vendor_info_t *ent; uint16_t pci_vendor_id, pci_device_id; uint16_t pci_subvendor_id, pci_subdevice_id; uint16_t pci_rev_id; if_shared_ctx_t sctx; if ((sctx = DEVICE_REGISTER(dev)) == NULL || sctx->isc_magic != IFLIB_MAGIC) return (ENOTSUP); pci_vendor_id = pci_get_vendor(dev); pci_device_id = pci_get_device(dev); pci_subvendor_id = pci_get_subvendor(dev); pci_subdevice_id = pci_get_subdevice(dev); pci_rev_id = pci_get_revid(dev); if (sctx->isc_parse_devinfo != NULL) sctx->isc_parse_devinfo(&pci_device_id, &pci_subvendor_id, &pci_subdevice_id, &pci_rev_id); ent = sctx->isc_vendor_info; while (ent->pvi_vendor_id != 0) { if (pci_vendor_id != ent->pvi_vendor_id) { ent++; continue; } if ((pci_device_id == ent->pvi_device_id) && ((pci_subvendor_id == ent->pvi_subvendor_id) || (ent->pvi_subvendor_id == 0)) && ((pci_subdevice_id == ent->pvi_subdevice_id) || (ent->pvi_subdevice_id == 0)) && ((pci_rev_id == ent->pvi_rev_id) || (ent->pvi_rev_id == 0))) { device_set_desc_copy(dev, ent->pvi_name); /* this needs to be changed to zero if the bus probing code * ever stops re-probing on best match because the sctx * may have its values over written by register calls * in subsequent probes */ return (BUS_PROBE_DEFAULT); } ent++; } return (ENXIO); } int iflib_device_register(device_t dev, void *sc, if_shared_ctx_t sctx, if_ctx_t *ctxp) { int err, rid, msix, msix_bar; if_ctx_t ctx; if_t ifp; if_softc_ctx_t scctx; int i; uint16_t main_txq; uint16_t main_rxq; ctx = malloc(sizeof(* ctx), M_IFLIB, M_WAITOK|M_ZERO); if (sc == NULL) { sc = malloc(sctx->isc_driver->size, M_IFLIB, M_WAITOK|M_ZERO); device_set_softc(dev, ctx); ctx->ifc_flags |= IFC_SC_ALLOCATED; } ctx->ifc_sctx = sctx; ctx->ifc_dev = dev; ctx->ifc_softc = sc; if ((err = iflib_register(ctx)) != 0) { device_printf(dev, "iflib_register failed %d\n", err); return (err); } iflib_add_device_sysctl_pre(ctx); scctx = &ctx->ifc_softc_ctx; ifp = ctx->ifc_ifp; /* * XXX sanity check that ntxd & nrxd are a power of 2 */ if (ctx->ifc_sysctl_ntxqs != 0) scctx->isc_ntxqsets = ctx->ifc_sysctl_ntxqs; if (ctx->ifc_sysctl_nrxqs != 0) scctx->isc_nrxqsets = ctx->ifc_sysctl_nrxqs; for (i = 0; i < sctx->isc_ntxqs; i++) { if (ctx->ifc_sysctl_ntxds[i] != 0) scctx->isc_ntxd[i] = ctx->ifc_sysctl_ntxds[i]; else scctx->isc_ntxd[i] = sctx->isc_ntxd_default[i]; } for (i = 0; i < sctx->isc_nrxqs; i++) { if (ctx->ifc_sysctl_nrxds[i] != 0) scctx->isc_nrxd[i] = ctx->ifc_sysctl_nrxds[i]; else scctx->isc_nrxd[i] = sctx->isc_nrxd_default[i]; } for (i = 0; i < sctx->isc_nrxqs; i++) { if (scctx->isc_nrxd[i] < sctx->isc_nrxd_min[i]) { device_printf(dev, "nrxd%d: %d less than nrxd_min %d - resetting to min\n", i, scctx->isc_nrxd[i], sctx->isc_nrxd_min[i]); scctx->isc_nrxd[i] = sctx->isc_nrxd_min[i]; } if (scctx->isc_nrxd[i] > sctx->isc_nrxd_max[i]) { device_printf(dev, "nrxd%d: %d greater than nrxd_max %d - resetting to max\n", i, scctx->isc_nrxd[i], sctx->isc_nrxd_max[i]); scctx->isc_nrxd[i] = sctx->isc_nrxd_max[i]; } } for (i = 0; i < sctx->isc_ntxqs; i++) { if (scctx->isc_ntxd[i] < sctx->isc_ntxd_min[i]) { device_printf(dev, "ntxd%d: %d less than ntxd_min %d - resetting to min\n", i, scctx->isc_ntxd[i], sctx->isc_ntxd_min[i]); scctx->isc_ntxd[i] = sctx->isc_ntxd_min[i]; } if (scctx->isc_ntxd[i] > sctx->isc_ntxd_max[i]) { device_printf(dev, "ntxd%d: %d greater than ntxd_max %d - resetting to max\n", i, scctx->isc_ntxd[i], sctx->isc_ntxd_max[i]); scctx->isc_ntxd[i] = sctx->isc_ntxd_max[i]; } } if ((err = IFDI_ATTACH_PRE(ctx)) != 0) { device_printf(dev, "IFDI_ATTACH_PRE failed %d\n", err); return (err); } _iflib_pre_assert(scctx); ctx->ifc_txrx = *scctx->isc_txrx; #ifdef INVARIANTS MPASS(scctx->isc_capenable); if (scctx->isc_capenable & IFCAP_TXCSUM) MPASS(scctx->isc_tx_csum_flags); #endif if_setcapabilities(ifp, scctx->isc_capenable | IFCAP_HWSTATS); if_setcapenable(ifp, scctx->isc_capenable | IFCAP_HWSTATS); if (scctx->isc_ntxqsets == 0 || (scctx->isc_ntxqsets_max && scctx->isc_ntxqsets_max < scctx->isc_ntxqsets)) scctx->isc_ntxqsets = scctx->isc_ntxqsets_max; if (scctx->isc_nrxqsets == 0 || (scctx->isc_nrxqsets_max && scctx->isc_nrxqsets_max < scctx->isc_nrxqsets)) scctx->isc_nrxqsets = scctx->isc_nrxqsets_max; #ifdef ACPI_DMAR if (dmar_get_dma_tag(device_get_parent(dev), dev) != NULL) ctx->ifc_flags |= IFC_DMAR; #elif !(defined(__i386__) || defined(__amd64__)) /* set unconditionally for !x86 */ ctx->ifc_flags |= IFC_DMAR; #endif msix_bar = scctx->isc_msix_bar; main_txq = (sctx->isc_flags & IFLIB_HAS_TXCQ) ? 1 : 0; main_rxq = (sctx->isc_flags & IFLIB_HAS_RXCQ) ? 1 : 0; /* XXX change for per-queue sizes */ device_printf(dev, "using %d tx descriptors and %d rx descriptors\n", scctx->isc_ntxd[main_txq], scctx->isc_nrxd[main_rxq]); for (i = 0; i < sctx->isc_nrxqs; i++) { if (!powerof2(scctx->isc_nrxd[i])) { /* round down instead? */ device_printf(dev, "# rx descriptors must be a power of 2\n"); err = EINVAL; goto fail; } } for (i = 0; i < sctx->isc_ntxqs; i++) { if (!powerof2(scctx->isc_ntxd[i])) { device_printf(dev, "# tx descriptors must be a power of 2"); err = EINVAL; goto fail; } } if (scctx->isc_tx_nsegments > scctx->isc_ntxd[main_txq] / MAX_SINGLE_PACKET_FRACTION) scctx->isc_tx_nsegments = max(1, scctx->isc_ntxd[main_txq] / MAX_SINGLE_PACKET_FRACTION); if (scctx->isc_tx_tso_segments_max > scctx->isc_ntxd[main_txq] / MAX_SINGLE_PACKET_FRACTION) scctx->isc_tx_tso_segments_max = max(1, scctx->isc_ntxd[main_txq] / MAX_SINGLE_PACKET_FRACTION); /* * Protect the stack against modern hardware */ if (scctx->isc_tx_tso_size_max > FREEBSD_TSO_SIZE_MAX) scctx->isc_tx_tso_size_max = FREEBSD_TSO_SIZE_MAX; /* TSO parameters - dig these out of the data sheet - simply correspond to tag setup */ ifp->if_hw_tsomaxsegcount = scctx->isc_tx_tso_segments_max; ifp->if_hw_tsomax = scctx->isc_tx_tso_size_max; ifp->if_hw_tsomaxsegsize = scctx->isc_tx_tso_segsize_max; if (scctx->isc_rss_table_size == 0) scctx->isc_rss_table_size = 64; scctx->isc_rss_table_mask = scctx->isc_rss_table_size-1; GROUPTASK_INIT(&ctx->ifc_admin_task, 0, _task_fn_admin, ctx); /* XXX format name */ taskqgroup_attach(qgroup_if_config_tqg, &ctx->ifc_admin_task, ctx, -1, "admin"); /* ** Now setup MSI or MSI/X, should ** return us the number of supported ** vectors. (Will be 1 for MSI) */ if (sctx->isc_flags & IFLIB_SKIP_MSIX) { msix = scctx->isc_vectors; } else if (scctx->isc_msix_bar != 0) /* * The simple fact that isc_msix_bar is not 0 does not mean we * we have a good value there that is known to work. */ msix = iflib_msix_init(ctx); else { scctx->isc_vectors = 1; scctx->isc_ntxqsets = 1; scctx->isc_nrxqsets = 1; scctx->isc_intr = IFLIB_INTR_LEGACY; msix = 0; } /* Get memory for the station queues */ if ((err = iflib_queues_alloc(ctx))) { device_printf(dev, "Unable to allocate queue memory\n"); goto fail; } if ((err = iflib_qset_structures_setup(ctx))) { device_printf(dev, "qset structure setup failed %d\n", err); goto fail_queues; } /* * Group taskqueues aren't properly set up until SMP is started, * so we disable interrupts until we can handle them post * SI_SUB_SMP. * * XXX: disabling interrupts doesn't actually work, at least for * the non-MSI case. When they occur before SI_SUB_SMP completes, * we do null handling and depend on this not causing too large an * interrupt storm. */ IFDI_INTR_DISABLE(ctx); if (msix > 1 && (err = IFDI_MSIX_INTR_ASSIGN(ctx, msix)) != 0) { device_printf(dev, "IFDI_MSIX_INTR_ASSIGN failed %d\n", err); goto fail_intr_free; } if (msix <= 1) { rid = 0; if (scctx->isc_intr == IFLIB_INTR_MSI) { MPASS(msix == 1); rid = 1; } if ((err = iflib_legacy_setup(ctx, ctx->isc_legacy_intr, ctx->ifc_softc, &rid, "irq0")) != 0) { device_printf(dev, "iflib_legacy_setup failed %d\n", err); goto fail_intr_free; } } ether_ifattach(ctx->ifc_ifp, ctx->ifc_mac); if ((err = IFDI_ATTACH_POST(ctx)) != 0) { device_printf(dev, "IFDI_ATTACH_POST failed %d\n", err); goto fail_detach; } if ((err = iflib_netmap_attach(ctx))) { device_printf(ctx->ifc_dev, "netmap attach failed: %d\n", err); goto fail_detach; } *ctxp = ctx; if_setgetcounterfn(ctx->ifc_ifp, iflib_if_get_counter); iflib_add_device_sysctl_post(ctx); ctx->ifc_flags |= IFC_INIT_DONE; return (0); fail_detach: ether_ifdetach(ctx->ifc_ifp); fail_intr_free: if (scctx->isc_intr == IFLIB_INTR_MSIX || scctx->isc_intr == IFLIB_INTR_MSI) pci_release_msi(ctx->ifc_dev); fail_queues: /* XXX free queues */ fail: IFDI_DETACH(ctx); return (err); } int iflib_device_attach(device_t dev) { if_ctx_t ctx; if_shared_ctx_t sctx; if ((sctx = DEVICE_REGISTER(dev)) == NULL || sctx->isc_magic != IFLIB_MAGIC) return (ENOTSUP); pci_enable_busmaster(dev); return (iflib_device_register(dev, NULL, sctx, &ctx)); } int iflib_device_deregister(if_ctx_t ctx) { if_t ifp = ctx->ifc_ifp; iflib_txq_t txq; iflib_rxq_t rxq; device_t dev = ctx->ifc_dev; int i, j; struct taskqgroup *tqg; iflib_fl_t fl; /* Make sure VLANS are not using driver */ if (if_vlantrunkinuse(ifp)) { device_printf(dev,"Vlan in use, detach first\n"); return (EBUSY); } CTX_LOCK(ctx); ctx->ifc_in_detach = 1; iflib_stop(ctx); CTX_UNLOCK(ctx); /* Unregister VLAN events */ if (ctx->ifc_vlan_attach_event != NULL) EVENTHANDLER_DEREGISTER(vlan_config, ctx->ifc_vlan_attach_event); if (ctx->ifc_vlan_detach_event != NULL) EVENTHANDLER_DEREGISTER(vlan_unconfig, ctx->ifc_vlan_detach_event); iflib_netmap_detach(ifp); ether_ifdetach(ifp); /* ether_ifdetach calls if_qflush - lock must be destroy afterwards*/ CTX_LOCK_DESTROY(ctx); if (ctx->ifc_led_dev != NULL) led_destroy(ctx->ifc_led_dev); /* XXX drain any dependent tasks */ tqg = qgroup_if_io_tqg; for (txq = ctx->ifc_txqs, i = 0; i < NTXQSETS(ctx); i++, txq++) { callout_drain(&txq->ift_timer); if (txq->ift_task.gt_uniq != NULL) taskqgroup_detach(tqg, &txq->ift_task); } for (i = 0, rxq = ctx->ifc_rxqs; i < NRXQSETS(ctx); i++, rxq++) { if (rxq->ifr_task.gt_uniq != NULL) taskqgroup_detach(tqg, &rxq->ifr_task); for (j = 0, fl = rxq->ifr_fl; j < rxq->ifr_nfl; j++, fl++) free(fl->ifl_rx_bitmap, M_IFLIB); } tqg = qgroup_if_config_tqg; if (ctx->ifc_admin_task.gt_uniq != NULL) taskqgroup_detach(tqg, &ctx->ifc_admin_task); if (ctx->ifc_vflr_task.gt_uniq != NULL) taskqgroup_detach(tqg, &ctx->ifc_vflr_task); IFDI_DETACH(ctx); device_set_softc(ctx->ifc_dev, NULL); if (ctx->ifc_softc_ctx.isc_intr != IFLIB_INTR_LEGACY) { pci_release_msi(dev); } if (ctx->ifc_softc_ctx.isc_intr != IFLIB_INTR_MSIX) { iflib_irq_free(ctx, &ctx->ifc_legacy_irq); } if (ctx->ifc_msix_mem != NULL) { bus_release_resource(ctx->ifc_dev, SYS_RES_MEMORY, ctx->ifc_softc_ctx.isc_msix_bar, ctx->ifc_msix_mem); ctx->ifc_msix_mem = NULL; } bus_generic_detach(dev); if_free(ifp); iflib_tx_structures_free(ctx); iflib_rx_structures_free(ctx); if (ctx->ifc_flags & IFC_SC_ALLOCATED) free(ctx->ifc_softc, M_IFLIB); free(ctx, M_IFLIB); return (0); } int iflib_device_detach(device_t dev) { if_ctx_t ctx = device_get_softc(dev); return (iflib_device_deregister(ctx)); } int iflib_device_suspend(device_t dev) { if_ctx_t ctx = device_get_softc(dev); CTX_LOCK(ctx); IFDI_SUSPEND(ctx); CTX_UNLOCK(ctx); return bus_generic_suspend(dev); } int iflib_device_shutdown(device_t dev) { if_ctx_t ctx = device_get_softc(dev); CTX_LOCK(ctx); IFDI_SHUTDOWN(ctx); CTX_UNLOCK(ctx); return bus_generic_suspend(dev); } int iflib_device_resume(device_t dev) { if_ctx_t ctx = device_get_softc(dev); iflib_txq_t txq = ctx->ifc_txqs; CTX_LOCK(ctx); IFDI_RESUME(ctx); iflib_init_locked(ctx); CTX_UNLOCK(ctx); for (int i = 0; i < NTXQSETS(ctx); i++, txq++) iflib_txq_check_drain(txq, IFLIB_RESTART_BUDGET); return (bus_generic_resume(dev)); } int iflib_device_iov_init(device_t dev, uint16_t num_vfs, const nvlist_t *params) { int error; if_ctx_t ctx = device_get_softc(dev); CTX_LOCK(ctx); error = IFDI_IOV_INIT(ctx, num_vfs, params); CTX_UNLOCK(ctx); return (error); } void iflib_device_iov_uninit(device_t dev) { if_ctx_t ctx = device_get_softc(dev); CTX_LOCK(ctx); IFDI_IOV_UNINIT(ctx); CTX_UNLOCK(ctx); } int iflib_device_iov_add_vf(device_t dev, uint16_t vfnum, const nvlist_t *params) { int error; if_ctx_t ctx = device_get_softc(dev); CTX_LOCK(ctx); error = IFDI_IOV_VF_ADD(ctx, vfnum, params); CTX_UNLOCK(ctx); return (error); } /********************************************************************* * * MODULE FUNCTION DEFINITIONS * **********************************************************************/ /* * - Start a fast taskqueue thread for each core * - Start a taskqueue for control operations */ static int iflib_module_init(void) { return (0); } static int iflib_module_event_handler(module_t mod, int what, void *arg) { int err; switch (what) { case MOD_LOAD: if ((err = iflib_module_init()) != 0) return (err); break; case MOD_UNLOAD: return (EBUSY); default: return (EOPNOTSUPP); } return (0); } /********************************************************************* * * PUBLIC FUNCTION DEFINITIONS * ordered as in iflib.h * **********************************************************************/ static void _iflib_assert(if_shared_ctx_t sctx) { MPASS(sctx->isc_tx_maxsize); MPASS(sctx->isc_tx_maxsegsize); MPASS(sctx->isc_rx_maxsize); MPASS(sctx->isc_rx_nsegments); MPASS(sctx->isc_rx_maxsegsize); MPASS(sctx->isc_nrxd_min[0]); MPASS(sctx->isc_nrxd_max[0]); MPASS(sctx->isc_nrxd_default[0]); MPASS(sctx->isc_ntxd_min[0]); MPASS(sctx->isc_ntxd_max[0]); MPASS(sctx->isc_ntxd_default[0]); } static void _iflib_pre_assert(if_softc_ctx_t scctx) { MPASS(scctx->isc_txrx->ift_txd_encap); MPASS(scctx->isc_txrx->ift_txd_flush); MPASS(scctx->isc_txrx->ift_txd_credits_update); MPASS(scctx->isc_txrx->ift_rxd_available); MPASS(scctx->isc_txrx->ift_rxd_pkt_get); MPASS(scctx->isc_txrx->ift_rxd_refill); MPASS(scctx->isc_txrx->ift_rxd_flush); } static int iflib_register(if_ctx_t ctx) { if_shared_ctx_t sctx = ctx->ifc_sctx; driver_t *driver = sctx->isc_driver; device_t dev = ctx->ifc_dev; if_t ifp; _iflib_assert(sctx); CTX_LOCK_INIT(ctx, device_get_nameunit(ctx->ifc_dev)); ifp = ctx->ifc_ifp = if_gethandle(IFT_ETHER); if (ifp == NULL) { device_printf(dev, "can not allocate ifnet structure\n"); return (ENOMEM); } /* * Initialize our context's device specific methods */ kobj_init((kobj_t) ctx, (kobj_class_t) driver); kobj_class_compile((kobj_class_t) driver); driver->refs++; if_initname(ifp, device_get_name(dev), device_get_unit(dev)); if_setsoftc(ifp, ctx); if_setdev(ifp, dev); if_setinitfn(ifp, iflib_if_init); if_setioctlfn(ifp, iflib_if_ioctl); if_settransmitfn(ifp, iflib_if_transmit); if_setqflushfn(ifp, iflib_if_qflush); if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST); ctx->ifc_vlan_attach_event = EVENTHANDLER_REGISTER(vlan_config, iflib_vlan_register, ctx, EVENTHANDLER_PRI_FIRST); ctx->ifc_vlan_detach_event = EVENTHANDLER_REGISTER(vlan_unconfig, iflib_vlan_unregister, ctx, EVENTHANDLER_PRI_FIRST); ifmedia_init(&ctx->ifc_media, IFM_IMASK, iflib_media_change, iflib_media_status); return (0); } static int iflib_queues_alloc(if_ctx_t ctx) { if_shared_ctx_t sctx = ctx->ifc_sctx; if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; device_t dev = ctx->ifc_dev; int nrxqsets = scctx->isc_nrxqsets; int ntxqsets = scctx->isc_ntxqsets; iflib_txq_t txq; iflib_rxq_t rxq; iflib_fl_t fl = NULL; int i, j, cpu, err, txconf, rxconf; iflib_dma_info_t ifdip; uint32_t *rxqsizes = scctx->isc_rxqsizes; uint32_t *txqsizes = scctx->isc_txqsizes; uint8_t nrxqs = sctx->isc_nrxqs; uint8_t ntxqs = sctx->isc_ntxqs; int nfree_lists = sctx->isc_nfl ? sctx->isc_nfl : 1; caddr_t *vaddrs; uint64_t *paddrs; struct ifmp_ring **brscp; KASSERT(ntxqs > 0, ("number of queues per qset must be at least 1")); KASSERT(nrxqs > 0, ("number of queues per qset must be at least 1")); brscp = NULL; txq = NULL; rxq = NULL; /* Allocate the TX ring struct memory */ if (!(txq = (iflib_txq_t) malloc(sizeof(struct iflib_txq) * ntxqsets, M_IFLIB, M_NOWAIT | M_ZERO))) { device_printf(dev, "Unable to allocate TX ring memory\n"); err = ENOMEM; goto fail; } /* Now allocate the RX */ if (!(rxq = (iflib_rxq_t) malloc(sizeof(struct iflib_rxq) * nrxqsets, M_IFLIB, M_NOWAIT | M_ZERO))) { device_printf(dev, "Unable to allocate RX ring memory\n"); err = ENOMEM; goto rx_fail; } ctx->ifc_txqs = txq; ctx->ifc_rxqs = rxq; /* * XXX handle allocation failure */ for (txconf = i = 0, cpu = CPU_FIRST(); i < ntxqsets; i++, txconf++, txq++, cpu = CPU_NEXT(cpu)) { /* Set up some basics */ if ((ifdip = malloc(sizeof(struct iflib_dma_info) * ntxqs, M_IFLIB, M_WAITOK|M_ZERO)) == NULL) { device_printf(dev, "failed to allocate iflib_dma_info\n"); err = ENOMEM; goto err_tx_desc; } txq->ift_ifdi = ifdip; for (j = 0; j < ntxqs; j++, ifdip++) { if (iflib_dma_alloc(ctx, txqsizes[j], ifdip, BUS_DMA_NOWAIT)) { device_printf(dev, "Unable to allocate Descriptor memory\n"); err = ENOMEM; goto err_tx_desc; } txq->ift_txd_size[j] = scctx->isc_txd_size[j]; bzero((void *)ifdip->idi_vaddr, txqsizes[j]); } txq->ift_ctx = ctx; txq->ift_id = i; if (sctx->isc_flags & IFLIB_HAS_TXCQ) { txq->ift_br_offset = 1; } else { txq->ift_br_offset = 0; } /* XXX fix this */ txq->ift_timer.c_cpu = cpu; if (iflib_txsd_alloc(txq)) { device_printf(dev, "Critical Failure setting up TX buffers\n"); err = ENOMEM; goto err_tx_desc; } /* Initialize the TX lock */ snprintf(txq->ift_mtx_name, MTX_NAME_LEN, "%s:tx(%d):callout", device_get_nameunit(dev), txq->ift_id); mtx_init(&txq->ift_mtx, txq->ift_mtx_name, NULL, MTX_DEF); callout_init_mtx(&txq->ift_timer, &txq->ift_mtx, 0); snprintf(txq->ift_db_mtx_name, MTX_NAME_LEN, "%s:tx(%d):db", device_get_nameunit(dev), txq->ift_id); err = ifmp_ring_alloc(&txq->ift_br, 2048, txq, iflib_txq_drain, iflib_txq_can_drain, M_IFLIB, M_WAITOK); if (err) { /* XXX free any allocated rings */ device_printf(dev, "Unable to allocate buf_ring\n"); goto err_tx_desc; } } for (rxconf = i = 0; i < nrxqsets; i++, rxconf++, rxq++) { /* Set up some basics */ if ((ifdip = malloc(sizeof(struct iflib_dma_info) * nrxqs, M_IFLIB, M_WAITOK|M_ZERO)) == NULL) { device_printf(dev, "failed to allocate iflib_dma_info\n"); err = ENOMEM; goto err_tx_desc; } rxq->ifr_ifdi = ifdip; /* XXX this needs to be changed if #rx queues != #tx queues */ rxq->ifr_ntxqirq = 1; rxq->ifr_txqid[0] = i; for (j = 0; j < nrxqs; j++, ifdip++) { if (iflib_dma_alloc(ctx, rxqsizes[j], ifdip, BUS_DMA_NOWAIT)) { device_printf(dev, "Unable to allocate Descriptor memory\n"); err = ENOMEM; goto err_tx_desc; } bzero((void *)ifdip->idi_vaddr, rxqsizes[j]); } rxq->ifr_ctx = ctx; rxq->ifr_id = i; if (sctx->isc_flags & IFLIB_HAS_RXCQ) { rxq->ifr_fl_offset = 1; } else { rxq->ifr_fl_offset = 0; } rxq->ifr_nfl = nfree_lists; if (!(fl = (iflib_fl_t) malloc(sizeof(struct iflib_fl) * nfree_lists, M_IFLIB, M_NOWAIT | M_ZERO))) { device_printf(dev, "Unable to allocate free list memory\n"); err = ENOMEM; goto err_tx_desc; } rxq->ifr_fl = fl; for (j = 0; j < nfree_lists; j++) { fl[j].ifl_rxq = rxq; fl[j].ifl_id = j; fl[j].ifl_ifdi = &rxq->ifr_ifdi[j + rxq->ifr_fl_offset]; fl[j].ifl_rxd_size = scctx->isc_rxd_size[j]; } /* Allocate receive buffers for the ring*/ if (iflib_rxsd_alloc(rxq)) { device_printf(dev, "Critical Failure setting up receive buffers\n"); err = ENOMEM; goto err_rx_desc; } for (j = 0, fl = rxq->ifr_fl; j < rxq->ifr_nfl; j++, fl++) fl->ifl_rx_bitmap = bit_alloc(fl->ifl_size, M_IFLIB, M_WAITOK|M_ZERO); } /* TXQs */ vaddrs = malloc(sizeof(caddr_t)*ntxqsets*ntxqs, M_IFLIB, M_WAITOK); paddrs = malloc(sizeof(uint64_t)*ntxqsets*ntxqs, M_IFLIB, M_WAITOK); for (i = 0; i < ntxqsets; i++) { iflib_dma_info_t di = ctx->ifc_txqs[i].ift_ifdi; for (j = 0; j < ntxqs; j++, di++) { vaddrs[i*ntxqs + j] = di->idi_vaddr; paddrs[i*ntxqs + j] = di->idi_paddr; } } if ((err = IFDI_TX_QUEUES_ALLOC(ctx, vaddrs, paddrs, ntxqs, ntxqsets)) != 0) { device_printf(ctx->ifc_dev, "device queue allocation failed\n"); iflib_tx_structures_free(ctx); free(vaddrs, M_IFLIB); free(paddrs, M_IFLIB); goto err_rx_desc; } free(vaddrs, M_IFLIB); free(paddrs, M_IFLIB); /* RXQs */ vaddrs = malloc(sizeof(caddr_t)*nrxqsets*nrxqs, M_IFLIB, M_WAITOK); paddrs = malloc(sizeof(uint64_t)*nrxqsets*nrxqs, M_IFLIB, M_WAITOK); for (i = 0; i < nrxqsets; i++) { iflib_dma_info_t di = ctx->ifc_rxqs[i].ifr_ifdi; for (j = 0; j < nrxqs; j++, di++) { vaddrs[i*nrxqs + j] = di->idi_vaddr; paddrs[i*nrxqs + j] = di->idi_paddr; } } if ((err = IFDI_RX_QUEUES_ALLOC(ctx, vaddrs, paddrs, nrxqs, nrxqsets)) != 0) { device_printf(ctx->ifc_dev, "device queue allocation failed\n"); iflib_tx_structures_free(ctx); free(vaddrs, M_IFLIB); free(paddrs, M_IFLIB); goto err_rx_desc; } free(vaddrs, M_IFLIB); free(paddrs, M_IFLIB); return (0); /* XXX handle allocation failure changes */ err_rx_desc: err_tx_desc: if (ctx->ifc_rxqs != NULL) free(ctx->ifc_rxqs, M_IFLIB); ctx->ifc_rxqs = NULL; if (ctx->ifc_txqs != NULL) free(ctx->ifc_txqs, M_IFLIB); ctx->ifc_txqs = NULL; rx_fail: if (brscp != NULL) free(brscp, M_IFLIB); if (rxq != NULL) free(rxq, M_IFLIB); if (txq != NULL) free(txq, M_IFLIB); fail: return (err); } static int iflib_tx_structures_setup(if_ctx_t ctx) { iflib_txq_t txq = ctx->ifc_txqs; int i; for (i = 0; i < NTXQSETS(ctx); i++, txq++) iflib_txq_setup(txq); return (0); } static void iflib_tx_structures_free(if_ctx_t ctx) { iflib_txq_t txq = ctx->ifc_txqs; int i, j; for (i = 0; i < NTXQSETS(ctx); i++, txq++) { iflib_txq_destroy(txq); for (j = 0; j < ctx->ifc_nhwtxqs; j++) iflib_dma_free(&txq->ift_ifdi[j]); } free(ctx->ifc_txqs, M_IFLIB); ctx->ifc_txqs = NULL; IFDI_QUEUES_FREE(ctx); } /********************************************************************* * * Initialize all receive rings. * **********************************************************************/ static int iflib_rx_structures_setup(if_ctx_t ctx) { iflib_rxq_t rxq = ctx->ifc_rxqs; int q; #if defined(INET6) || defined(INET) int i, err; #endif for (q = 0; q < ctx->ifc_softc_ctx.isc_nrxqsets; q++, rxq++) { #if defined(INET6) || defined(INET) tcp_lro_free(&rxq->ifr_lc); if ((err = tcp_lro_init_args(&rxq->ifr_lc, ctx->ifc_ifp, TCP_LRO_ENTRIES, min(1024, ctx->ifc_softc_ctx.isc_nrxd[rxq->ifr_fl_offset]))) != 0) { device_printf(ctx->ifc_dev, "LRO Initialization failed!\n"); goto fail; } rxq->ifr_lro_enabled = TRUE; #endif IFDI_RXQ_SETUP(ctx, rxq->ifr_id); } return (0); #if defined(INET6) || defined(INET) fail: /* * Free RX software descriptors allocated so far, we will only handle * the rings that completed, the failing case will have * cleaned up for itself. 'q' failed, so its the terminus. */ rxq = ctx->ifc_rxqs; for (i = 0; i < q; ++i, rxq++) { iflib_rx_sds_free(rxq); rxq->ifr_cq_gen = rxq->ifr_cq_cidx = rxq->ifr_cq_pidx = 0; } return (err); #endif } /********************************************************************* * * Free all receive rings. * **********************************************************************/ static void iflib_rx_structures_free(if_ctx_t ctx) { iflib_rxq_t rxq = ctx->ifc_rxqs; for (int i = 0; i < ctx->ifc_softc_ctx.isc_nrxqsets; i++, rxq++) { iflib_rx_sds_free(rxq); } } static int iflib_qset_structures_setup(if_ctx_t ctx) { int err; if ((err = iflib_tx_structures_setup(ctx)) != 0) return (err); if ((err = iflib_rx_structures_setup(ctx)) != 0) { device_printf(ctx->ifc_dev, "iflib_rx_structures_setup failed: %d\n", err); iflib_tx_structures_free(ctx); iflib_rx_structures_free(ctx); } return (err); } int iflib_irq_alloc(if_ctx_t ctx, if_irq_t irq, int rid, driver_filter_t filter, void *filter_arg, driver_intr_t handler, void *arg, char *name) { return (_iflib_irq_alloc(ctx, irq, rid, filter, handler, arg, name)); } static int find_nth(if_ctx_t ctx, cpuset_t *cpus, int qid) { int i, cpuid, eqid, count; CPU_COPY(&ctx->ifc_cpus, cpus); count = CPU_COUNT(&ctx->ifc_cpus); eqid = qid % count; /* clear up to the qid'th bit */ for (i = 0; i < eqid; i++) { cpuid = CPU_FFS(cpus); MPASS(cpuid != 0); CPU_CLR(cpuid-1, cpus); } cpuid = CPU_FFS(cpus); MPASS(cpuid != 0); return (cpuid-1); } int iflib_irq_alloc_generic(if_ctx_t ctx, if_irq_t irq, int rid, iflib_intr_type_t type, driver_filter_t *filter, void *filter_arg, int qid, char *name) { struct grouptask *gtask; struct taskqgroup *tqg; iflib_filter_info_t info; cpuset_t cpus; gtask_fn_t *fn; int tqrid, err, cpuid; driver_filter_t *intr_fast; void *q; info = &ctx->ifc_filter_info; tqrid = rid; switch (type) { /* XXX merge tx/rx for netmap? */ case IFLIB_INTR_TX: q = &ctx->ifc_txqs[qid]; info = &ctx->ifc_txqs[qid].ift_filter_info; gtask = &ctx->ifc_txqs[qid].ift_task; tqg = qgroup_if_io_tqg; fn = _task_fn_tx; intr_fast = iflib_fast_intr; GROUPTASK_INIT(gtask, 0, fn, q); break; case IFLIB_INTR_RX: q = &ctx->ifc_rxqs[qid]; info = &ctx->ifc_rxqs[qid].ifr_filter_info; gtask = &ctx->ifc_rxqs[qid].ifr_task; tqg = qgroup_if_io_tqg; fn = _task_fn_rx; intr_fast = iflib_fast_intr; GROUPTASK_INIT(gtask, 0, fn, q); break; case IFLIB_INTR_RXTX: q = &ctx->ifc_rxqs[qid]; info = &ctx->ifc_rxqs[qid].ifr_filter_info; gtask = &ctx->ifc_rxqs[qid].ifr_task; tqg = qgroup_if_io_tqg; fn = _task_fn_rx; intr_fast = iflib_fast_intr_rxtx; GROUPTASK_INIT(gtask, 0, fn, q); break; case IFLIB_INTR_ADMIN: q = ctx; tqrid = -1; info = &ctx->ifc_filter_info; gtask = &ctx->ifc_admin_task; tqg = qgroup_if_config_tqg; fn = _task_fn_admin; intr_fast = iflib_fast_intr_ctx; break; default: panic("unknown net intr type"); } info->ifi_filter = filter; info->ifi_filter_arg = filter_arg; info->ifi_task = gtask; info->ifi_ctx = q; err = _iflib_irq_alloc(ctx, irq, rid, intr_fast, NULL, info, name); if (err != 0) { device_printf(ctx->ifc_dev, "_iflib_irq_alloc failed %d\n", err); return (err); } if (type == IFLIB_INTR_ADMIN) return (0); if (tqrid != -1) { cpuid = find_nth(ctx, &cpus, qid); taskqgroup_attach_cpu(tqg, gtask, q, cpuid, rman_get_start(irq->ii_res), name); } else { taskqgroup_attach(tqg, gtask, q, rman_get_start(irq->ii_res), name); } return (0); } void iflib_softirq_alloc_generic(if_ctx_t ctx, if_irq_t irq, iflib_intr_type_t type, void *arg, int qid, char *name) { struct grouptask *gtask; struct taskqgroup *tqg; gtask_fn_t *fn; void *q; int irq_num = -1; switch (type) { case IFLIB_INTR_TX: q = &ctx->ifc_txqs[qid]; gtask = &ctx->ifc_txqs[qid].ift_task; tqg = qgroup_if_io_tqg; fn = _task_fn_tx; if (irq != NULL) irq_num = rman_get_start(irq->ii_res); break; case IFLIB_INTR_RX: q = &ctx->ifc_rxqs[qid]; gtask = &ctx->ifc_rxqs[qid].ifr_task; tqg = qgroup_if_io_tqg; fn = _task_fn_rx; if (irq != NULL) irq_num = rman_get_start(irq->ii_res); break; case IFLIB_INTR_IOV: q = ctx; gtask = &ctx->ifc_vflr_task; tqg = qgroup_if_config_tqg; fn = _task_fn_iov; break; default: panic("unknown net intr type"); } GROUPTASK_INIT(gtask, 0, fn, q); taskqgroup_attach(tqg, gtask, q, irq_num, name); } void iflib_irq_free(if_ctx_t ctx, if_irq_t irq) { if (irq->ii_tag) bus_teardown_intr(ctx->ifc_dev, irq->ii_res, irq->ii_tag); if (irq->ii_res) bus_release_resource(ctx->ifc_dev, SYS_RES_IRQ, irq->ii_rid, irq->ii_res); } static int iflib_legacy_setup(if_ctx_t ctx, driver_filter_t filter, void *filter_arg, int *rid, char *name) { iflib_txq_t txq = ctx->ifc_txqs; iflib_rxq_t rxq = ctx->ifc_rxqs; if_irq_t irq = &ctx->ifc_legacy_irq; iflib_filter_info_t info; struct grouptask *gtask; struct taskqgroup *tqg; gtask_fn_t *fn; int tqrid; void *q; int err; q = &ctx->ifc_rxqs[0]; info = &rxq[0].ifr_filter_info; gtask = &rxq[0].ifr_task; tqg = qgroup_if_io_tqg; tqrid = irq->ii_rid = *rid; fn = _task_fn_rx; ctx->ifc_flags |= IFC_LEGACY; info->ifi_filter = filter; info->ifi_filter_arg = filter_arg; info->ifi_task = gtask; info->ifi_ctx = ctx; /* We allocate a single interrupt resource */ if ((err = _iflib_irq_alloc(ctx, irq, tqrid, iflib_fast_intr_ctx, NULL, info, name)) != 0) return (err); GROUPTASK_INIT(gtask, 0, fn, q); taskqgroup_attach(tqg, gtask, q, tqrid, name); GROUPTASK_INIT(&txq->ift_task, 0, _task_fn_tx, txq); taskqgroup_attach(qgroup_if_io_tqg, &txq->ift_task, txq, tqrid, "tx"); return (0); } void iflib_led_create(if_ctx_t ctx) { ctx->ifc_led_dev = led_create(iflib_led_func, ctx, device_get_nameunit(ctx->ifc_dev)); } void iflib_tx_intr_deferred(if_ctx_t ctx, int txqid) { GROUPTASK_ENQUEUE(&ctx->ifc_txqs[txqid].ift_task); } void iflib_rx_intr_deferred(if_ctx_t ctx, int rxqid) { GROUPTASK_ENQUEUE(&ctx->ifc_rxqs[rxqid].ifr_task); } void iflib_admin_intr_deferred(if_ctx_t ctx) { #ifdef INVARIANTS struct grouptask *gtask; gtask = &ctx->ifc_admin_task; MPASS(gtask != NULL && gtask->gt_taskqueue != NULL); #endif GROUPTASK_ENQUEUE(&ctx->ifc_admin_task); } void iflib_iov_intr_deferred(if_ctx_t ctx) { GROUPTASK_ENQUEUE(&ctx->ifc_vflr_task); } void iflib_io_tqg_attach(struct grouptask *gt, void *uniq, int cpu, char *name) { taskqgroup_attach_cpu(qgroup_if_io_tqg, gt, uniq, cpu, -1, name); } void iflib_config_gtask_init(if_ctx_t ctx, struct grouptask *gtask, gtask_fn_t *fn, char *name) { GROUPTASK_INIT(gtask, 0, fn, ctx); taskqgroup_attach(qgroup_if_config_tqg, gtask, gtask, -1, name); } void iflib_config_gtask_deinit(struct grouptask *gtask) { taskqgroup_detach(qgroup_if_config_tqg, gtask); } void iflib_link_state_change(if_ctx_t ctx, int link_state, uint64_t baudrate) { if_t ifp = ctx->ifc_ifp; iflib_txq_t txq = ctx->ifc_txqs; if_setbaudrate(ifp, baudrate); if (baudrate >= IF_Gbps(10)) ctx->ifc_flags |= IFC_PREFETCH; /* If link down, disable watchdog */ if ((ctx->ifc_link_state == LINK_STATE_UP) && (link_state == LINK_STATE_DOWN)) { for (int i = 0; i < ctx->ifc_softc_ctx.isc_ntxqsets; i++, txq++) txq->ift_qstatus = IFLIB_QUEUE_IDLE; } ctx->ifc_link_state = link_state; if_link_state_change(ifp, link_state); } static int iflib_tx_credits_update(if_ctx_t ctx, iflib_txq_t txq) { int credits; #ifdef INVARIANTS int credits_pre = txq->ift_cidx_processed; #endif if (ctx->isc_txd_credits_update == NULL) return (0); if ((credits = ctx->isc_txd_credits_update(ctx->ifc_softc, txq->ift_id, true)) == 0) return (0); txq->ift_processed += credits; txq->ift_cidx_processed += credits; MPASS(credits_pre + credits == txq->ift_cidx_processed); if (txq->ift_cidx_processed >= txq->ift_size) txq->ift_cidx_processed -= txq->ift_size; return (credits); } static int iflib_rxd_avail(if_ctx_t ctx, iflib_rxq_t rxq, qidx_t cidx, qidx_t budget) { return (ctx->isc_rxd_available(ctx->ifc_softc, rxq->ifr_id, cidx, budget)); } void iflib_add_int_delay_sysctl(if_ctx_t ctx, const char *name, const char *description, if_int_delay_info_t info, int offset, int value) { info->iidi_ctx = ctx; info->iidi_offset = offset; info->iidi_value = value; SYSCTL_ADD_PROC(device_get_sysctl_ctx(ctx->ifc_dev), SYSCTL_CHILDREN(device_get_sysctl_tree(ctx->ifc_dev)), OID_AUTO, name, CTLTYPE_INT|CTLFLAG_RW, info, 0, iflib_sysctl_int_delay, "I", description); } struct mtx * iflib_ctx_lock_get(if_ctx_t ctx) { return (&ctx->ifc_mtx); } static int iflib_msix_init(if_ctx_t ctx) { device_t dev = ctx->ifc_dev; if_shared_ctx_t sctx = ctx->ifc_sctx; if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; int vectors, queues, rx_queues, tx_queues, queuemsgs, msgs; int iflib_num_tx_queues, iflib_num_rx_queues; int err, admincnt, bar; iflib_num_tx_queues = scctx->isc_ntxqsets; iflib_num_rx_queues = scctx->isc_nrxqsets; device_printf(dev, "msix_init qsets capped at %d\n", iflib_num_tx_queues); bar = ctx->ifc_softc_ctx.isc_msix_bar; admincnt = sctx->isc_admin_intrcnt; /* Override by global tuneable */ { int i; size_t len = sizeof(i); err = kernel_sysctlbyname(curthread, "hw.pci.enable_msix", &i, &len, NULL, 0, NULL, 0); if (err == 0) { if (i == 0) goto msi; } else { device_printf(dev, "unable to read hw.pci.enable_msix."); } } /* Override by tuneable */ if (scctx->isc_disable_msix) goto msi; /* ** When used in a virtualized environment ** PCI BUSMASTER capability may not be set ** so explicity set it here and rewrite ** the ENABLE in the MSIX control register ** at this point to cause the host to ** successfully initialize us. */ { int msix_ctrl, rid; pci_enable_busmaster(dev); rid = 0; if (pci_find_cap(dev, PCIY_MSIX, &rid) == 0 && rid != 0) { rid += PCIR_MSIX_CTRL; msix_ctrl = pci_read_config(dev, rid, 2); msix_ctrl |= PCIM_MSIXCTRL_MSIX_ENABLE; pci_write_config(dev, rid, msix_ctrl, 2); } else { device_printf(dev, "PCIY_MSIX capability not found; " "or rid %d == 0.\n", rid); goto msi; } } /* * bar == -1 => "trust me I know what I'm doing" * Some drivers are for hardware that is so shoddily * documented that no one knows which bars are which * so the developer has to map all bars. This hack * allows shoddy garbage to use msix in this framework. */ if (bar != -1) { ctx->ifc_msix_mem = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &bar, RF_ACTIVE); if (ctx->ifc_msix_mem == NULL) { /* May not be enabled */ device_printf(dev, "Unable to map MSIX table \n"); goto msi; } } /* First try MSI/X */ if ((msgs = pci_msix_count(dev)) == 0) { /* system has msix disabled */ device_printf(dev, "System has MSIX disabled \n"); bus_release_resource(dev, SYS_RES_MEMORY, bar, ctx->ifc_msix_mem); ctx->ifc_msix_mem = NULL; goto msi; } #if IFLIB_DEBUG /* use only 1 qset in debug mode */ queuemsgs = min(msgs - admincnt, 1); #else queuemsgs = msgs - admincnt; #endif if (bus_get_cpus(dev, INTR_CPUS, sizeof(ctx->ifc_cpus), &ctx->ifc_cpus) == 0) { #ifdef RSS queues = imin(queuemsgs, rss_getnumbuckets()); #else queues = queuemsgs; #endif queues = imin(CPU_COUNT(&ctx->ifc_cpus), queues); device_printf(dev, "pxm cpus: %d queue msgs: %d admincnt: %d\n", CPU_COUNT(&ctx->ifc_cpus), queuemsgs, admincnt); } else { device_printf(dev, "Unable to fetch CPU list\n"); /* Figure out a reasonable auto config value */ queues = min(queuemsgs, mp_ncpus); } #ifdef RSS /* If we're doing RSS, clamp at the number of RSS buckets */ if (queues > rss_getnumbuckets()) queues = rss_getnumbuckets(); #endif if (iflib_num_rx_queues > 0 && iflib_num_rx_queues < queuemsgs - admincnt) rx_queues = iflib_num_rx_queues; else rx_queues = queues; /* * We want this to be all logical CPUs by default */ if (iflib_num_tx_queues > 0 && iflib_num_tx_queues < queues) tx_queues = iflib_num_tx_queues; else tx_queues = mp_ncpus; if (ctx->ifc_sysctl_qs_eq_override == 0) { #ifdef INVARIANTS if (tx_queues != rx_queues) device_printf(dev, "queue equality override not set, capping rx_queues at %d and tx_queues at %d\n", min(rx_queues, tx_queues), min(rx_queues, tx_queues)); #endif tx_queues = min(rx_queues, tx_queues); rx_queues = min(rx_queues, tx_queues); } device_printf(dev, "using %d rx queues %d tx queues \n", rx_queues, tx_queues); vectors = rx_queues + admincnt; if ((err = pci_alloc_msix(dev, &vectors)) == 0) { device_printf(dev, "Using MSIX interrupts with %d vectors\n", vectors); scctx->isc_vectors = vectors; scctx->isc_nrxqsets = rx_queues; scctx->isc_ntxqsets = tx_queues; scctx->isc_intr = IFLIB_INTR_MSIX; return (vectors); } else { device_printf(dev, "failed to allocate %d msix vectors, err: %d - using MSI\n", vectors, err); } msi: vectors = pci_msi_count(dev); scctx->isc_nrxqsets = 1; scctx->isc_ntxqsets = 1; scctx->isc_vectors = vectors; if (vectors == 1 && pci_alloc_msi(dev, &vectors) == 0) { device_printf(dev,"Using an MSI interrupt\n"); scctx->isc_intr = IFLIB_INTR_MSI; } else { device_printf(dev,"Using a Legacy interrupt\n"); scctx->isc_intr = IFLIB_INTR_LEGACY; } return (vectors); } char * ring_states[] = { "IDLE", "BUSY", "STALLED", "ABDICATED" }; static int mp_ring_state_handler(SYSCTL_HANDLER_ARGS) { int rc; uint16_t *state = ((uint16_t *)oidp->oid_arg1); struct sbuf *sb; char *ring_state = "UNKNOWN"; /* XXX needed ? */ rc = sysctl_wire_old_buffer(req, 0); MPASS(rc == 0); if (rc != 0) return (rc); sb = sbuf_new_for_sysctl(NULL, NULL, 80, req); MPASS(sb != NULL); if (sb == NULL) return (ENOMEM); if (state[3] <= 3) ring_state = ring_states[state[3]]; sbuf_printf(sb, "pidx_head: %04hd pidx_tail: %04hd cidx: %04hd state: %s", state[0], state[1], state[2], ring_state); rc = sbuf_finish(sb); sbuf_delete(sb); return(rc); } enum iflib_ndesc_handler { IFLIB_NTXD_HANDLER, IFLIB_NRXD_HANDLER, }; static int mp_ndesc_handler(SYSCTL_HANDLER_ARGS) { if_ctx_t ctx = (void *)arg1; enum iflib_ndesc_handler type = arg2; char buf[256] = {0}; qidx_t *ndesc; char *p, *next; int nqs, rc, i; MPASS(type == IFLIB_NTXD_HANDLER || type == IFLIB_NRXD_HANDLER); nqs = 8; switch(type) { case IFLIB_NTXD_HANDLER: ndesc = ctx->ifc_sysctl_ntxds; if (ctx->ifc_sctx) nqs = ctx->ifc_sctx->isc_ntxqs; break; case IFLIB_NRXD_HANDLER: ndesc = ctx->ifc_sysctl_nrxds; if (ctx->ifc_sctx) nqs = ctx->ifc_sctx->isc_nrxqs; break; } if (nqs == 0) nqs = 8; for (i=0; i<8; i++) { if (i >= nqs) break; if (i) strcat(buf, ","); sprintf(strchr(buf, 0), "%d", ndesc[i]); } rc = sysctl_handle_string(oidp, buf, sizeof(buf), req); if (rc || req->newptr == NULL) return rc; for (i = 0, next = buf, p = strsep(&next, " ,"); i < 8 && p; i++, p = strsep(&next, " ,")) { ndesc[i] = strtoul(p, NULL, 10); } return(rc); } #define NAME_BUFLEN 32 static void iflib_add_device_sysctl_pre(if_ctx_t ctx) { device_t dev = iflib_get_dev(ctx); struct sysctl_oid_list *child, *oid_list; struct sysctl_ctx_list *ctx_list; struct sysctl_oid *node; ctx_list = device_get_sysctl_ctx(dev); child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev)); ctx->ifc_sysctl_node = node = SYSCTL_ADD_NODE(ctx_list, child, OID_AUTO, "iflib", CTLFLAG_RD, NULL, "IFLIB fields"); oid_list = SYSCTL_CHILDREN(node); SYSCTL_ADD_STRING(ctx_list, oid_list, OID_AUTO, "driver_version", CTLFLAG_RD, ctx->ifc_sctx->isc_driver_version, 0, "driver version"); SYSCTL_ADD_U16(ctx_list, oid_list, OID_AUTO, "override_ntxqs", CTLFLAG_RWTUN, &ctx->ifc_sysctl_ntxqs, 0, "# of txqs to use, 0 => use default #"); SYSCTL_ADD_U16(ctx_list, oid_list, OID_AUTO, "override_nrxqs", CTLFLAG_RWTUN, &ctx->ifc_sysctl_nrxqs, 0, "# of rxqs to use, 0 => use default #"); SYSCTL_ADD_U16(ctx_list, oid_list, OID_AUTO, "override_qs_enable", CTLFLAG_RWTUN, &ctx->ifc_sysctl_qs_eq_override, 0, "permit #txq != #rxq"); SYSCTL_ADD_INT(ctx_list, oid_list, OID_AUTO, "disable_msix", CTLFLAG_RWTUN, &ctx->ifc_softc_ctx.isc_disable_msix, 0, "disable MSIX (default 0)"); SYSCTL_ADD_U16(ctx_list, oid_list, OID_AUTO, "rx_budget", CTLFLAG_RWTUN, &ctx->ifc_sysctl_rx_budget, 0, "set the rx budget"); /* XXX change for per-queue sizes */ SYSCTL_ADD_PROC(ctx_list, oid_list, OID_AUTO, "override_ntxds", CTLTYPE_STRING|CTLFLAG_RWTUN, ctx, IFLIB_NTXD_HANDLER, mp_ndesc_handler, "A", "list of # of tx descriptors to use, 0 = use default #"); SYSCTL_ADD_PROC(ctx_list, oid_list, OID_AUTO, "override_nrxds", CTLTYPE_STRING|CTLFLAG_RWTUN, ctx, IFLIB_NRXD_HANDLER, mp_ndesc_handler, "A", "list of # of rx descriptors to use, 0 = use default #"); } static void iflib_add_device_sysctl_post(if_ctx_t ctx) { if_shared_ctx_t sctx = ctx->ifc_sctx; if_softc_ctx_t scctx = &ctx->ifc_softc_ctx; device_t dev = iflib_get_dev(ctx); struct sysctl_oid_list *child; struct sysctl_ctx_list *ctx_list; iflib_fl_t fl; iflib_txq_t txq; iflib_rxq_t rxq; int i, j; char namebuf[NAME_BUFLEN]; char *qfmt; struct sysctl_oid *queue_node, *fl_node, *node; struct sysctl_oid_list *queue_list, *fl_list; ctx_list = device_get_sysctl_ctx(dev); node = ctx->ifc_sysctl_node; child = SYSCTL_CHILDREN(node); if (scctx->isc_ntxqsets > 100) qfmt = "txq%03d"; else if (scctx->isc_ntxqsets > 10) qfmt = "txq%02d"; else qfmt = "txq%d"; for (i = 0, txq = ctx->ifc_txqs; i < scctx->isc_ntxqsets; i++, txq++) { snprintf(namebuf, NAME_BUFLEN, qfmt, i); queue_node = SYSCTL_ADD_NODE(ctx_list, child, OID_AUTO, namebuf, CTLFLAG_RD, NULL, "Queue Name"); queue_list = SYSCTL_CHILDREN(queue_node); #if MEMORY_LOGGING SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "txq_dequeued", CTLFLAG_RD, &txq->ift_dequeued, "total mbufs freed"); SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "txq_enqueued", CTLFLAG_RD, &txq->ift_enqueued, "total mbufs enqueued"); #endif SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "mbuf_defrag", CTLFLAG_RD, &txq->ift_mbuf_defrag, "# of times m_defrag was called"); SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "m_pullups", CTLFLAG_RD, &txq->ift_pullups, "# of times m_pullup was called"); SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "mbuf_defrag_failed", CTLFLAG_RD, &txq->ift_mbuf_defrag_failed, "# of times m_defrag failed"); SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "no_desc_avail", CTLFLAG_RD, &txq->ift_no_desc_avail, "# of times no descriptors were available"); SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "tx_map_failed", CTLFLAG_RD, &txq->ift_map_failed, "# of times dma map failed"); SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "txd_encap_efbig", CTLFLAG_RD, &txq->ift_txd_encap_efbig, "# of times txd_encap returned EFBIG"); SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "no_tx_dma_setup", CTLFLAG_RD, &txq->ift_no_tx_dma_setup, "# of times map failed for other than EFBIG"); SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "txq_pidx", CTLFLAG_RD, &txq->ift_pidx, 1, "Producer Index"); SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "txq_cidx", CTLFLAG_RD, &txq->ift_cidx, 1, "Consumer Index"); SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "txq_cidx_processed", CTLFLAG_RD, &txq->ift_cidx_processed, 1, "Consumer Index seen by credit update"); SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "txq_in_use", CTLFLAG_RD, &txq->ift_in_use, 1, "descriptors in use"); SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "txq_processed", CTLFLAG_RD, &txq->ift_processed, "descriptors procesed for clean"); SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "txq_cleaned", CTLFLAG_RD, &txq->ift_cleaned, "total cleaned"); SYSCTL_ADD_PROC(ctx_list, queue_list, OID_AUTO, "ring_state", CTLTYPE_STRING | CTLFLAG_RD, __DEVOLATILE(uint64_t *, &txq->ift_br->state), 0, mp_ring_state_handler, "A", "soft ring state"); SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_enqueues", CTLFLAG_RD, &txq->ift_br->enqueues, "# of enqueues to the mp_ring for this queue"); SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_drops", CTLFLAG_RD, &txq->ift_br->drops, "# of drops in the mp_ring for this queue"); SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_starts", CTLFLAG_RD, &txq->ift_br->starts, "# of normal consumer starts in the mp_ring for this queue"); SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_stalls", CTLFLAG_RD, &txq->ift_br->stalls, "# of consumer stalls in the mp_ring for this queue"); SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_restarts", CTLFLAG_RD, &txq->ift_br->restarts, "# of consumer restarts in the mp_ring for this queue"); SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_abdications", CTLFLAG_RD, &txq->ift_br->abdications, "# of consumer abdications in the mp_ring for this queue"); } if (scctx->isc_nrxqsets > 100) qfmt = "rxq%03d"; else if (scctx->isc_nrxqsets > 10) qfmt = "rxq%02d"; else qfmt = "rxq%d"; for (i = 0, rxq = ctx->ifc_rxqs; i < scctx->isc_nrxqsets; i++, rxq++) { snprintf(namebuf, NAME_BUFLEN, qfmt, i); queue_node = SYSCTL_ADD_NODE(ctx_list, child, OID_AUTO, namebuf, CTLFLAG_RD, NULL, "Queue Name"); queue_list = SYSCTL_CHILDREN(queue_node); if (sctx->isc_flags & IFLIB_HAS_RXCQ) { SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "rxq_cq_pidx", CTLFLAG_RD, &rxq->ifr_cq_pidx, 1, "Producer Index"); SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "rxq_cq_cidx", CTLFLAG_RD, &rxq->ifr_cq_cidx, 1, "Consumer Index"); } for (j = 0, fl = rxq->ifr_fl; j < rxq->ifr_nfl; j++, fl++) { snprintf(namebuf, NAME_BUFLEN, "rxq_fl%d", j); fl_node = SYSCTL_ADD_NODE(ctx_list, queue_list, OID_AUTO, namebuf, CTLFLAG_RD, NULL, "freelist Name"); fl_list = SYSCTL_CHILDREN(fl_node); SYSCTL_ADD_U16(ctx_list, fl_list, OID_AUTO, "pidx", CTLFLAG_RD, &fl->ifl_pidx, 1, "Producer Index"); SYSCTL_ADD_U16(ctx_list, fl_list, OID_AUTO, "cidx", CTLFLAG_RD, &fl->ifl_cidx, 1, "Consumer Index"); SYSCTL_ADD_U16(ctx_list, fl_list, OID_AUTO, "credits", CTLFLAG_RD, &fl->ifl_credits, 1, "credits available"); #if MEMORY_LOGGING SYSCTL_ADD_QUAD(ctx_list, fl_list, OID_AUTO, "fl_m_enqueued", CTLFLAG_RD, &fl->ifl_m_enqueued, "mbufs allocated"); SYSCTL_ADD_QUAD(ctx_list, fl_list, OID_AUTO, "fl_m_dequeued", CTLFLAG_RD, &fl->ifl_m_dequeued, "mbufs freed"); SYSCTL_ADD_QUAD(ctx_list, fl_list, OID_AUTO, "fl_cl_enqueued", CTLFLAG_RD, &fl->ifl_cl_enqueued, "clusters allocated"); SYSCTL_ADD_QUAD(ctx_list, fl_list, OID_AUTO, "fl_cl_dequeued", CTLFLAG_RD, &fl->ifl_cl_dequeued, "clusters freed"); #endif } } } #ifndef __NO_STRICT_ALIGNMENT static struct mbuf * iflib_fixup_rx(struct mbuf *m) { struct mbuf *n; if (m->m_len <= (MCLBYTES - ETHER_HDR_LEN)) { bcopy(m->m_data, m->m_data + ETHER_HDR_LEN, m->m_len); m->m_data += ETHER_HDR_LEN; n = m; } else { MGETHDR(n, M_NOWAIT, MT_DATA); if (n == NULL) { m_freem(m); return (NULL); } bcopy(m->m_data, n->m_data, ETHER_HDR_LEN); m->m_data += ETHER_HDR_LEN; m->m_len -= ETHER_HDR_LEN; n->m_len = ETHER_HDR_LEN; M_MOVE_PKTHDR(n, m); n->m_next = m; } return (n); } #endif Index: projects/runtime-coverage/sys/powerpc/mpc85xx/fsl_sata.c =================================================================== --- projects/runtime-coverage/sys/powerpc/mpc85xx/fsl_sata.c (revision 325498) +++ projects/runtime-coverage/sys/powerpc/mpc85xx/fsl_sata.c (revision 325499) @@ -1,1918 +1,1918 @@ /*- * Copyright (c) 2009-2012 Alexander Motin * Copyright (c) 2017 Justin Hibbits * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer, * without modification, immediately at the beginning of the file. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "fsl_sata.h" struct fsl_sata_channel; struct fsl_sata_slot; enum fsl_sata_err_type; struct fsl_sata_cmd_tab; /* local prototypes */ static int fsl_sata_init(device_t dev); static int fsl_sata_deinit(device_t dev); static int fsl_sata_suspend(device_t dev); static int fsl_sata_resume(device_t dev); static void fsl_sata_pm(void *arg); static void fsl_sata_intr(void *arg); static void fsl_sata_intr_main(struct fsl_sata_channel *ch, uint32_t istatus); static void fsl_sata_begin_transaction(struct fsl_sata_channel *ch, union ccb *ccb); static void fsl_sata_dmasetprd(void *arg, bus_dma_segment_t *segs, int nsegs, int error); static void fsl_sata_execute_transaction(struct fsl_sata_slot *slot); static void fsl_sata_timeout(struct fsl_sata_slot *slot); static void fsl_sata_end_transaction(struct fsl_sata_slot *slot, enum fsl_sata_err_type et); static int fsl_sata_setup_fis(struct fsl_sata_channel *ch, struct fsl_sata_cmd_tab *ctp, union ccb *ccb, int tag); static void fsl_sata_dmainit(device_t dev); static void fsl_sata_dmasetupc_cb(void *xsc, bus_dma_segment_t *segs, int nsegs, int error); static void fsl_sata_dmafini(device_t dev); static void fsl_sata_slotsalloc(device_t dev); static void fsl_sata_slotsfree(device_t dev); static void fsl_sata_reset(struct fsl_sata_channel *ch); static void fsl_sata_start(struct fsl_sata_channel *ch); static void fsl_sata_stop(struct fsl_sata_channel *ch); static void fsl_sata_issue_recovery(struct fsl_sata_channel *ch); static void fsl_sata_process_read_log(struct fsl_sata_channel *ch, union ccb *ccb); static void fsl_sata_process_request_sense(struct fsl_sata_channel *ch, union ccb *ccb); static void fsl_sataaction(struct cam_sim *sim, union ccb *ccb); static void fsl_satapoll(struct cam_sim *sim); static MALLOC_DEFINE(M_FSL_SATA, "FSL SATA driver", "FSL SATA driver data buffers"); #define recovery_type spriv_field0 #define RECOVERY_NONE 0 #define RECOVERY_READ_LOG 1 #define RECOVERY_REQUEST_SENSE 2 #define recovery_slot spriv_field1 #define FSL_SATA_P_CQR 0x0 #define FSL_SATA_P_CAR 0x4 #define FSL_SATA_P_CCR 0x10 #define FSL_SATA_P_CER 0x18 #define FSL_SATA_P_DER 0x20 #define FSL_SATA_P_CHBA 0x24 #define FSL_SATA_P_HSTS 0x28 #define FSL_SATA_P_HSTS_HS_ON 0x80000000 #define FSL_SATA_P_HSTS_ME 0x00040000 #define FSL_SATA_P_HSTS_DLM 0x00001000 #define FSL_SATA_P_HSTS_FOT 0x00000200 #define FSL_SATA_P_HSTS_FOR 0x00000100 #define FSL_SATA_P_HSTS_FE 0x00000020 #define FSL_SATA_P_HSTS_PR 0x00000010 #define FSL_SATA_P_HSTS_SNTFU 0x00000004 #define FSL_SATA_P_HSTS_DE 0x00000002 #define FSL_SATA_P_HCTRL 0x2c #define FSL_SATA_P_HCTRL_HC_ON 0x80000000 #define FSL_SATA_P_HCTRL_HC_FORCE_OFF 0x40000000 #define FSL_SATA_P_HCTRL_ENT 0x10000000 #define FSL_SATA_P_HCTRL_SNOOP 0x00000400 #define FSL_SATA_P_HCTRL_PM 0x00000200 #define FSL_SATA_P_HCTRL_FATAL 0x00000020 #define FSL_SATA_P_HCTRL_PHYRDY 0x00000010 #define FSL_SATA_P_HCTRL_SIG 0x00000008 #define FSL_SATA_P_HCTRL_SNTFY 0x00000004 #define FSL_SATA_P_HCTRL_DE 0x00000002 #define FSL_SATA_P_HCTRL_CC 0x00000001 #define FSL_SATA_P_HCTRL_INT_MASK 0x0000003f #define FSL_SATA_P_CQPMP 0x30 #define FSL_SATA_P_SIG 0x34 #define FSL_SATA_P_ICC 0x38 #define FSL_SATA_P_ICC_ITC_M 0x1f000000 #define FSL_SATA_P_ICC_ITC_S 24 #define FSL_SATA_P_ICC_ITTCV_M 0x0007ffff #define FSL_SATA_P_PCC 0x15c #define FSL_SATA_P_PCC_SLUMBER 0x0000000c #define FSL_SATA_P_PCC_PARTIAL 0x0000000a #define FSL_SATA_PCC_LPB_EN 0x0000000e #define FSL_SATA_MAX_SLOTS 16 /* FSL_SATA register defines */ #define FSL_SATA_P_SSTS 0x100 #define FSL_SATA_P_SERR 0x104 #define FSL_SATA_P_SCTL 0x108 #define FSL_SATA_P_SNTF 0x10c /* Pessimistic prognosis on number of required S/G entries */ #define FSL_SATA_SG_ENTRIES 63 /* Command list. 16 commands. First, 1Kbyte aligned. */ #define FSL_SATA_CL_OFFSET 0 #define FSL_SATA_CL_SIZE 16 /* Command tables. Up to 32 commands, Each, 4-byte aligned. */ #define FSL_SATA_CT_OFFSET (FSL_SATA_CL_OFFSET + FSL_SATA_CL_SIZE * FSL_SATA_MAX_SLOTS) #define FSL_SATA_CT_SIZE (96 + FSL_SATA_SG_ENTRIES * 16) /* Total main work area. */ #define FSL_SATA_WORK_SIZE (FSL_SATA_CT_OFFSET + FSL_SATA_CT_SIZE * FSL_SATA_MAX_SLOTS) #define FSL_SATA_MAX_XFER (64 * 1024 * 1024) /* Some convenience macros for getting the CTP and CLP */ #define FSL_SATA_CTP_BUS(ch, slot) \ ((ch->dma.work_bus + FSL_SATA_CT_OFFSET + (FSL_SATA_CT_SIZE * slot->slot))) #define FSL_SATA_PRD_OFFSET(prd) (96 + (prd) * 16) #define FSL_SATA_CTP(ch, slot) \ ((struct fsl_sata_cmd_tab *)(ch->dma.work + FSL_SATA_CT_OFFSET + \ (FSL_SATA_CT_SIZE * slot->slot))) #define FSL_SATA_CLP(ch, slot) \ ((struct fsl_sata_cmd_list *) (ch->dma.work + FSL_SATA_CL_OFFSET + \ (FSL_SATA_CL_SIZE * slot->slot))) struct fsl_sata_dma_prd { uint32_t dba; uint32_t reserved; uint32_t reserved2; uint32_t dwc_flg; /* 0 based */ #define FSL_SATA_PRD_MASK 0x01fffffc /* max 32MB */ #define FSL_SATA_PRD_MAX (FSL_SATA_PRD_MASK + 4) #define FSL_SATA_PRD_SNOOP 0x10000000 #define FSL_SATA_PRD_EXT 0x80000000 } __packed; struct fsl_sata_cmd_tab { uint8_t cfis[32]; uint8_t sfis[32]; uint8_t acmd[16]; uint8_t reserved[16]; struct fsl_sata_dma_prd prd_tab[FSL_SATA_SG_ENTRIES]; #define FSL_SATA_PRD_EXT_INDEX 15 #define FSL_SATA_PRD_MAX_DIRECT 16 } __packed; struct fsl_sata_cmd_list { uint32_t cda; /* word aligned */ uint16_t fis_length; /* length in bytes (aligned to words) */ uint16_t prd_length; /* PRD entries */ uint32_t ttl; uint32_t cmd_flags; #define FSL_SATA_CMD_TAG_MASK 0x001f #define FSL_SATA_CMD_ATAPI 0x0020 #define FSL_SATA_CMD_BIST 0x0040 #define FSL_SATA_CMD_RESET 0x0080 #define FSL_SATA_CMD_QUEUED 0x0100 #define FSL_SATA_CMD_SNOOP 0x0200 #define FSL_SATA_CMD_VBIST 0x0400 #define FSL_SATA_CMD_WRITE 0x0800 } __packed; /* misc defines */ #define ATA_IRQ_RID 0 #define ATA_INTR_FLAGS (INTR_MPSAFE|INTR_TYPE_BIO|INTR_ENTROPY) struct ata_dmaslot { bus_dmamap_t data_map; /* data DMA map */ int nsegs; /* Number of segs loaded */ }; /* structure holding DMA related information */ struct ata_dma { bus_dma_tag_t work_tag; /* workspace DMA tag */ bus_dmamap_t work_map; /* workspace DMA map */ uint8_t *work; /* workspace */ bus_addr_t work_bus; /* bus address of work */ bus_dma_tag_t data_tag; /* data DMA tag */ }; enum fsl_sata_slot_states { FSL_SATA_SLOT_EMPTY, FSL_SATA_SLOT_LOADING, FSL_SATA_SLOT_RUNNING, FSL_SATA_SLOT_EXECUTING }; struct fsl_sata_slot { struct fsl_sata_channel *ch; /* Channel */ uint8_t slot; /* Number of this slot */ enum fsl_sata_slot_states state; /* Slot state */ union ccb *ccb; /* CCB occupying slot */ struct ata_dmaslot dma; /* DMA data of this slot */ struct callout timeout; /* Execution timeout */ uint32_t ttl; }; struct fsl_sata_device { int revision; int mode; u_int bytecount; u_int atapi; u_int tags; u_int caps; }; /* structure describing an ATA channel */ struct fsl_sata_channel { device_t dev; /* Device handle */ int unit; /* Physical channel */ struct resource *r_mem; /* Memory of this channel */ struct resource *r_irq; /* Interrupt of this channel */ void *ih; /* Interrupt handle */ struct ata_dma dma; /* DMA data */ struct cam_sim *sim; struct cam_path *path; uint32_t caps; /* Controller capabilities */ int pm_level; /* power management level */ int devices; /* What is present */ int pm_present; /* PM presence reported */ union ccb *hold[FSL_SATA_MAX_SLOTS]; struct fsl_sata_slot slot[FSL_SATA_MAX_SLOTS]; uint32_t oslots; /* Occupied slots */ uint32_t rslots; /* Running slots */ uint32_t aslots; /* Slots with atomic commands */ uint32_t eslots; /* Slots in error */ uint32_t toslots; /* Slots in timeout */ int lastslot; /* Last used slot */ int taggedtarget; /* Last tagged target */ int numrslots; /* Number of running slots */ int numrslotspd[16];/* Number of running slots per dev */ int numtslots; /* Number of tagged slots */ int numtslotspd[16];/* Number of tagged slots per dev */ int numhslots; /* Number of held slots */ int recoverycmd; /* Our READ LOG active */ int fatalerr; /* Fatal error happend */ int resetting; /* Hard-reset in progress. */ int resetpolldiv; /* Hard-reset poll divider. */ union ccb *frozen; /* Frozen command */ struct callout pm_timer; /* Power management events */ struct callout reset_timer; /* Hard-reset timeout */ struct fsl_sata_device user[16]; /* User-specified settings */ struct fsl_sata_device curr[16]; /* Current settings */ struct mtx_padalign mtx; /* state lock */ STAILQ_HEAD(, ccb_hdr) doneq; /* queue of completed CCBs */ int batch; /* doneq is in use */ }; enum fsl_sata_err_type { FSL_SATA_ERR_NONE, /* No error */ FSL_SATA_ERR_INVALID, /* Error detected by us before submitting. */ FSL_SATA_ERR_INNOCENT, /* Innocent victim. */ FSL_SATA_ERR_TFE, /* Task File Error. */ FSL_SATA_ERR_SATA, /* SATA error. */ FSL_SATA_ERR_TIMEOUT, /* Command execution timeout. */ FSL_SATA_ERR_NCQ, /* NCQ command error. CCB should be put on hold * until READ LOG executed to reveal error. */ }; /* macros to hide busspace uglyness */ #define ATA_INB(res, offset) \ bus_read_1((res), (offset)) #define ATA_INW(res, offset) \ bus_read_2((res), (offset)) #define ATA_INL(res, offset) \ bus_read_4((res), (offset)) #define ATA_INSW(res, offset, addr, count) \ bus_read_multi_2((res), (offset), (addr), (count)) #define ATA_INSW_STRM(res, offset, addr, count) \ bus_read_multi_stream_2((res), (offset), (addr), (count)) #define ATA_INSL(res, offset, addr, count) \ bus_read_multi_4((res), (offset), (addr), (count)) #define ATA_INSL_STRM(res, offset, addr, count) \ bus_read_multi_stream_4((res), (offset), (addr), (count)) #define ATA_OUTB(res, offset, value) \ bus_write_1((res), (offset), (value)) #define ATA_OUTW(res, offset, value) \ bus_write_2((res), (offset), (value)) #define ATA_OUTL(res, offset, value) \ bus_write_4((res), (offset), (value)) #define ATA_OUTSW(res, offset, addr, count) \ bus_write_multi_2((res), (offset), (addr), (count)) #define ATA_OUTSW_STRM(res, offset, addr, count) \ bus_write_multi_stream_2((res), (offset), (addr), (count)) #define ATA_OUTSL(res, offset, addr, count) \ bus_write_multi_4((res), (offset), (addr), (count)) #define ATA_OUTSL_STRM(res, offset, addr, count) \ bus_write_multi_stream_4((res), (offset), (addr), (count)) static int fsl_sata_probe(device_t dev) { if (!ofw_bus_is_compatible(dev, "fsl,pq-sata-v2") && !ofw_bus_is_compatible(dev, "fsl,pq-sata")) return (ENXIO); device_set_desc_copy(dev, "Freescale Integrated SATA Controller"); return (BUS_PROBE_DEFAULT); } static int fsl_sata_attach(device_t dev) { struct fsl_sata_channel *ch = device_get_softc(dev); struct cam_devq *devq; int rid, error, i, sata_rev = 0; ch->dev = dev; ch->unit = (intptr_t)device_get_ivars(dev); mtx_init(&ch->mtx, "FSL SATA channel lock", NULL, MTX_DEF); ch->pm_level = 0; resource_int_value(device_get_name(dev), device_get_unit(dev), "pm_level", &ch->pm_level); STAILQ_INIT(&ch->doneq); if (ch->pm_level > 3) callout_init_mtx(&ch->pm_timer, &ch->mtx, 0); resource_int_value(device_get_name(dev), device_get_unit(dev), "sata_rev", &sata_rev); for (i = 0; i < 16; i++) { ch->user[i].revision = sata_rev; ch->user[i].mode = 0; ch->user[i].bytecount = 8192; ch->user[i].tags = FSL_SATA_MAX_SLOTS; ch->user[i].caps = 0; ch->curr[i] = ch->user[i]; if (ch->pm_level) { ch->user[i].caps = CTS_SATA_CAPS_H_PMREQ | CTS_SATA_CAPS_D_PMREQ; } ch->user[i].caps |= CTS_SATA_CAPS_H_AN; } rid = 0; if (!(ch->r_mem = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid, RF_ACTIVE))) return (ENXIO); rman_set_bustag(ch->r_mem, &bs_le_tag); fsl_sata_dmainit(dev); fsl_sata_slotsalloc(dev); fsl_sata_init(dev); rid = ATA_IRQ_RID; if (!(ch->r_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &rid, RF_SHAREABLE | RF_ACTIVE))) { device_printf(dev, "Unable to map interrupt\n"); error = ENXIO; goto err0; } if ((bus_setup_intr(dev, ch->r_irq, ATA_INTR_FLAGS, NULL, fsl_sata_intr, ch, &ch->ih))) { device_printf(dev, "Unable to setup interrupt\n"); error = ENXIO; goto err1; } mtx_lock(&ch->mtx); /* Create the device queue for our SIM. */ devq = cam_simq_alloc(FSL_SATA_MAX_SLOTS); if (devq == NULL) { device_printf(dev, "Unable to allocate simq\n"); error = ENOMEM; goto err1; } /* Construct SIM entry */ ch->sim = cam_sim_alloc(fsl_sataaction, fsl_satapoll, "fslsata", ch, device_get_unit(dev), (struct mtx *)&ch->mtx, 2, FSL_SATA_MAX_SLOTS, devq); if (ch->sim == NULL) { cam_simq_free(devq); device_printf(dev, "unable to allocate sim\n"); error = ENOMEM; goto err1; } if (xpt_bus_register(ch->sim, dev, 0) != CAM_SUCCESS) { device_printf(dev, "unable to register xpt bus\n"); error = ENXIO; goto err2; } if (xpt_create_path(&ch->path, /*periph*/NULL, cam_sim_path(ch->sim), CAM_TARGET_WILDCARD, CAM_LUN_WILDCARD) != CAM_REQ_CMP) { device_printf(dev, "unable to create path\n"); error = ENXIO; goto err3; } if (ch->pm_level > 3) { callout_reset(&ch->pm_timer, (ch->pm_level == 4) ? hz / 1000 : hz / 8, fsl_sata_pm, ch); } mtx_unlock(&ch->mtx); return (0); err3: xpt_bus_deregister(cam_sim_path(ch->sim)); err2: cam_sim_free(ch->sim, /*free_devq*/TRUE); err1: mtx_unlock(&ch->mtx); bus_release_resource(dev, SYS_RES_IRQ, ATA_IRQ_RID, ch->r_irq); err0: bus_release_resource(dev, SYS_RES_MEMORY, ch->unit, ch->r_mem); mtx_destroy(&ch->mtx); return (error); } static int fsl_sata_detach(device_t dev) { struct fsl_sata_channel *ch = device_get_softc(dev); mtx_lock(&ch->mtx); xpt_async(AC_LOST_DEVICE, ch->path, NULL); xpt_free_path(ch->path); xpt_bus_deregister(cam_sim_path(ch->sim)); cam_sim_free(ch->sim, /*free_devq*/TRUE); mtx_unlock(&ch->mtx); if (ch->pm_level > 3) callout_drain(&ch->pm_timer); bus_teardown_intr(dev, ch->r_irq, ch->ih); bus_release_resource(dev, SYS_RES_IRQ, ATA_IRQ_RID, ch->r_irq); fsl_sata_deinit(dev); fsl_sata_slotsfree(dev); fsl_sata_dmafini(dev); bus_release_resource(dev, SYS_RES_MEMORY, ch->unit, ch->r_mem); mtx_destroy(&ch->mtx); return (0); } static int fsl_sata_wait_register(struct fsl_sata_channel *ch, bus_size_t off, unsigned int mask, unsigned int val, int t) { int timeout = 0; uint32_t rval; while (((rval = ATA_INL(ch->r_mem, off)) & mask) != val) { if (timeout > t) { return (EBUSY); } DELAY(1000); timeout++; } return (0); } static int fsl_sata_init(device_t dev) { struct fsl_sata_channel *ch = device_get_softc(dev); uint64_t work; uint32_t r; /* Disable port interrupts */ r = ATA_INL(ch->r_mem, FSL_SATA_P_HCTRL); r &= ~FSL_SATA_P_HCTRL_HC_ON; r |= FSL_SATA_P_HCTRL_HC_FORCE_OFF; ATA_OUTL(ch->r_mem, FSL_SATA_P_HCTRL, r & ~FSL_SATA_P_HCTRL_INT_MASK); fsl_sata_wait_register(ch, FSL_SATA_P_HSTS, FSL_SATA_P_HSTS_HS_ON, 0, 1000); /* Setup work areas */ work = ch->dma.work_bus + FSL_SATA_CL_OFFSET; ATA_OUTL(ch->r_mem, FSL_SATA_P_CHBA, work); r &= ~FSL_SATA_P_HCTRL_ENT; r &= ~FSL_SATA_P_HCTRL_PM; ATA_OUTL(ch->r_mem, FSL_SATA_P_HCTRL, r); r = ATA_INL(ch->r_mem, FSL_SATA_P_PCC); ATA_OUTL(ch->r_mem, FSL_SATA_P_PCC, r & ~FSL_SATA_PCC_LPB_EN); ATA_OUTL(ch->r_mem, FSL_SATA_P_ICC, (1 << FSL_SATA_P_ICC_ITC_S)); fsl_sata_start(ch); return (0); } static int fsl_sata_deinit(device_t dev) { struct fsl_sata_channel *ch = device_get_softc(dev); uint32_t r; /* Disable port interrupts. */ r = ATA_INL(ch->r_mem, FSL_SATA_P_HCTRL); ATA_OUTL(ch->r_mem, FSL_SATA_P_HCTRL, r & ~FSL_SATA_P_HCTRL_INT_MASK); /* Reset command register. */ fsl_sata_stop(ch); /* Allow everything, including partial and slumber modes. */ ATA_OUTL(ch->r_mem, FSL_SATA_P_SCTL, 0); DELAY(100); /* Disable PHY. */ ATA_OUTL(ch->r_mem, FSL_SATA_P_SCTL, ATA_SC_DET_DISABLE); r = ATA_INL(ch->r_mem, FSL_SATA_P_HCTRL); /* Turn off the controller. */ ATA_OUTL(ch->r_mem, FSL_SATA_P_HCTRL, r & ~FSL_SATA_P_HCTRL_HC_ON); return (0); } static int fsl_sata_suspend(device_t dev) { struct fsl_sata_channel *ch = device_get_softc(dev); mtx_lock(&ch->mtx); xpt_freeze_simq(ch->sim, 1); while (ch->oslots) msleep(ch, &ch->mtx, PRIBIO, "fsl_satasusp", hz/100); fsl_sata_deinit(dev); mtx_unlock(&ch->mtx); return (0); } static int fsl_sata_resume(device_t dev) { struct fsl_sata_channel *ch = device_get_softc(dev); mtx_lock(&ch->mtx); fsl_sata_init(dev); fsl_sata_reset(ch); xpt_release_simq(ch->sim, TRUE); mtx_unlock(&ch->mtx); return (0); } devclass_t fsl_satach_devclass; static device_method_t fsl_satach_methods[] = { DEVMETHOD(device_probe, fsl_sata_probe), DEVMETHOD(device_attach, fsl_sata_attach), DEVMETHOD(device_detach, fsl_sata_detach), DEVMETHOD(device_suspend, fsl_sata_suspend), DEVMETHOD(device_resume, fsl_sata_resume), DEVMETHOD_END }; static driver_t fsl_satach_driver = { "fslsata", fsl_satach_methods, sizeof(struct fsl_sata_channel) }; DRIVER_MODULE(fsl_satach, simplebus, fsl_satach_driver, fsl_satach_devclass, NULL, NULL); struct fsl_sata_dc_cb_args { bus_addr_t maddr; int error; }; static void fsl_sata_dmainit(device_t dev) { struct fsl_sata_channel *ch = device_get_softc(dev); struct fsl_sata_dc_cb_args dcba; /* Command area. */ if (bus_dma_tag_create(bus_get_dma_tag(dev), 1024, 0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL, NULL, FSL_SATA_WORK_SIZE, 1, FSL_SATA_WORK_SIZE, 0, NULL, NULL, &ch->dma.work_tag)) goto error; if (bus_dmamem_alloc(ch->dma.work_tag, (void **)&ch->dma.work, BUS_DMA_ZERO, &ch->dma.work_map)) goto error; if (bus_dmamap_load(ch->dma.work_tag, ch->dma.work_map, ch->dma.work, FSL_SATA_WORK_SIZE, fsl_sata_dmasetupc_cb, &dcba, 0) || dcba.error) { bus_dmamem_free(ch->dma.work_tag, ch->dma.work, ch->dma.work_map); goto error; } ch->dma.work_bus = dcba.maddr; /* Data area. */ if (bus_dma_tag_create(bus_get_dma_tag(dev), 4, 0, BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR, NULL, NULL, FSL_SATA_MAX_XFER, FSL_SATA_SG_ENTRIES - 1, FSL_SATA_PRD_MAX, 0, busdma_lock_mutex, &ch->mtx, &ch->dma.data_tag)) { goto error; } if (bootverbose) device_printf(dev, "work area: %p\n", ch->dma.work); return; error: device_printf(dev, "WARNING - DMA initialization failed\n"); fsl_sata_dmafini(dev); } static void fsl_sata_dmasetupc_cb(void *xsc, bus_dma_segment_t *segs, int nsegs, int error) { struct fsl_sata_dc_cb_args *dcba = (struct fsl_sata_dc_cb_args *)xsc; if (!(dcba->error = error)) dcba->maddr = segs[0].ds_addr; } static void fsl_sata_dmafini(device_t dev) { struct fsl_sata_channel *ch = device_get_softc(dev); if (ch->dma.data_tag) { bus_dma_tag_destroy(ch->dma.data_tag); ch->dma.data_tag = NULL; } if (ch->dma.work_bus) { bus_dmamap_unload(ch->dma.work_tag, ch->dma.work_map); bus_dmamem_free(ch->dma.work_tag, ch->dma.work, ch->dma.work_map); ch->dma.work_bus = 0; ch->dma.work = NULL; } if (ch->dma.work_tag) { bus_dma_tag_destroy(ch->dma.work_tag); ch->dma.work_tag = NULL; } } static void fsl_sata_slotsalloc(device_t dev) { struct fsl_sata_channel *ch = device_get_softc(dev); int i; /* Alloc and setup command/dma slots */ bzero(ch->slot, sizeof(ch->slot)); for (i = 0; i < FSL_SATA_MAX_SLOTS; i++) { struct fsl_sata_slot *slot = &ch->slot[i]; slot->ch = ch; slot->slot = i; slot->state = FSL_SATA_SLOT_EMPTY; slot->ccb = NULL; callout_init_mtx(&slot->timeout, &ch->mtx, 0); if (bus_dmamap_create(ch->dma.data_tag, 0, &slot->dma.data_map)) device_printf(ch->dev, "FAILURE - create data_map\n"); } } static void fsl_sata_slotsfree(device_t dev) { struct fsl_sata_channel *ch = device_get_softc(dev); int i; /* Free all dma slots */ for (i = 0; i < FSL_SATA_MAX_SLOTS; i++) { struct fsl_sata_slot *slot = &ch->slot[i]; callout_drain(&slot->timeout); if (slot->dma.data_map) { bus_dmamap_destroy(ch->dma.data_tag, slot->dma.data_map); slot->dma.data_map = NULL; } } } static int fsl_sata_phy_check_events(struct fsl_sata_channel *ch, u_int32_t serr) { if (((ch->pm_level == 0) && (serr & ATA_SE_PHY_CHANGED)) || ((ch->pm_level != 0) && (serr & ATA_SE_EXCHANGED))) { u_int32_t status = ATA_INL(ch->r_mem, FSL_SATA_P_SSTS); union ccb *ccb; if (bootverbose) { if ((status & ATA_SS_DET_MASK) != ATA_SS_DET_NO_DEVICE) device_printf(ch->dev, "CONNECT requested\n"); else device_printf(ch->dev, "DISCONNECT requested\n"); } /* Issue soft reset */ xpt_async(AC_BUS_RESET, ch->path, NULL); if ((ccb = xpt_alloc_ccb_nowait()) == NULL) return (0); if (xpt_create_path(&ccb->ccb_h.path, NULL, cam_sim_path(ch->sim), CAM_TARGET_WILDCARD, CAM_LUN_WILDCARD) != CAM_REQ_CMP) { xpt_free_ccb(ccb); return (0); } xpt_rescan(ccb); return (1); } return (0); } static void fsl_sata_notify_events(struct fsl_sata_channel *ch, u_int32_t status) { struct cam_path *dpath; int i; ATA_OUTL(ch->r_mem, FSL_SATA_P_SNTF, status); if (bootverbose) device_printf(ch->dev, "SNTF 0x%04x\n", status); for (i = 0; i < 16; i++) { if ((status & (1 << i)) == 0) continue; if (xpt_create_path(&dpath, NULL, xpt_path_path_id(ch->path), i, 0) == CAM_REQ_CMP) { xpt_async(AC_SCSI_AEN, dpath, NULL); xpt_free_path(dpath); } } } static void fsl_sata_done(struct fsl_sata_channel *ch, union ccb *ccb) { mtx_assert(&ch->mtx, MA_OWNED); if ((ccb->ccb_h.func_code & XPT_FC_QUEUED) == 0 || ch->batch == 0) { xpt_done(ccb); return; } STAILQ_INSERT_TAIL(&ch->doneq, &ccb->ccb_h, sim_links.stqe); } static void fsl_sata_intr(void *arg) { struct fsl_sata_channel *ch = (struct fsl_sata_channel *)arg; struct ccb_hdr *ccb_h; uint32_t istatus; STAILQ_HEAD(, ccb_hdr) tmp_doneq = STAILQ_HEAD_INITIALIZER(tmp_doneq); /* Read interrupt statuses. */ istatus = ATA_INL(ch->r_mem, FSL_SATA_P_HSTS) & 0x7ffff; if ((istatus & 0x3f) == 0) return; mtx_lock(&ch->mtx); ch->batch = 1; fsl_sata_intr_main(ch, istatus); ch->batch = 0; /* * Prevent the possibility of issues caused by processing the queue * while unlocked below by moving the contents to a local queue. */ STAILQ_CONCAT(&tmp_doneq, &ch->doneq); mtx_unlock(&ch->mtx); while ((ccb_h = STAILQ_FIRST(&tmp_doneq)) != NULL) { STAILQ_REMOVE_HEAD(&tmp_doneq, sim_links.stqe); xpt_done_direct((union ccb *)ccb_h); } /* Clear interrupt statuses. */ ATA_OUTL(ch->r_mem, FSL_SATA_P_HSTS, istatus & 0x3f); } static void fsl_sata_pm(void *arg) { struct fsl_sata_channel *ch = (struct fsl_sata_channel *)arg; uint32_t work; if (ch->numrslots != 0) return; work = ATA_INL(ch->r_mem, FSL_SATA_P_PCC) & ~FSL_SATA_PCC_LPB_EN; if (ch->pm_level == 4) work |= FSL_SATA_P_PCC_PARTIAL; else work |= FSL_SATA_P_PCC_SLUMBER; ATA_OUTL(ch->r_mem, FSL_SATA_P_PCC, work); } /* XXX: interrupt todo */ static void fsl_sata_intr_main(struct fsl_sata_channel *ch, uint32_t istatus) { uint32_t cer, der, serr = 0, sntf = 0, ok, err; enum fsl_sata_err_type et; int i; /* Complete all successful commands. */ ok = ATA_INL(ch->r_mem, FSL_SATA_P_CCR); if (ch->aslots == 0) for (i = 0; i < FSL_SATA_MAX_SLOTS; i++) { if (((ok >> i) & 1) && ch->slot[i].ccb != NULL) fsl_sata_end_transaction(&ch->slot[i], FSL_SATA_ERR_NONE); } /* Read command statuses. */ if (istatus & FSL_SATA_P_HSTS_SNTFU) sntf = ATA_INL(ch->r_mem, FSL_SATA_P_SNTF); /* XXX: Process PHY events */ serr = ATA_INL(ch->r_mem, FSL_SATA_P_SERR); ATA_OUTL(ch->r_mem, FSL_SATA_P_SERR, serr); if (istatus & (FSL_SATA_P_HSTS_PR)) { if (serr) { fsl_sata_phy_check_events(ch, serr); } } /* Process command errors */ err = (istatus & (FSL_SATA_P_HSTS_FE | FSL_SATA_P_HSTS_DE)); cer = ATA_INL(ch->r_mem, FSL_SATA_P_CER); ATA_OUTL(ch->r_mem, FSL_SATA_P_CER, cer); der = ATA_INL(ch->r_mem, FSL_SATA_P_DER); ATA_OUTL(ch->r_mem, FSL_SATA_P_DER, der); /* On error, complete the rest of commands with error statuses. */ if (err) { if (ch->frozen) { union ccb *fccb = ch->frozen; ch->frozen = NULL; fccb->ccb_h.status = CAM_REQUEUE_REQ | CAM_RELEASE_SIMQ; if (!(fccb->ccb_h.status & CAM_DEV_QFRZN)) { xpt_freeze_devq(fccb->ccb_h.path, 1); fccb->ccb_h.status |= CAM_DEV_QFRZN; } fsl_sata_done(ch, fccb); } for (i = 0; i < FSL_SATA_MAX_SLOTS; i++) { if (ch->slot[i].ccb == NULL) continue; if ((cer & (1 << i)) != 0) et = FSL_SATA_ERR_TFE; else if ((der & (1 << ch->slot[i].ccb->ccb_h.target_id)) != 0) et = FSL_SATA_ERR_SATA; else et = FSL_SATA_ERR_INVALID; fsl_sata_end_transaction(&ch->slot[i], et); } } /* Process NOTIFY events */ if (sntf) fsl_sata_notify_events(ch, sntf); } /* Must be called with channel locked. */ static int fsl_sata_check_collision(struct fsl_sata_channel *ch, union ccb *ccb) { int t = ccb->ccb_h.target_id; if ((ccb->ccb_h.func_code == XPT_ATA_IO) && (ccb->ataio.cmd.flags & CAM_ATAIO_FPDMA)) { /* Tagged command while we have no supported tag free. */ if (((~ch->oslots) & (0xffffffff >> (32 - ch->curr[t].tags))) == 0) return (1); /* Tagged command while untagged are active. */ if (ch->numrslotspd[t] != 0 && ch->numtslotspd[t] == 0) return (1); } else { /* Untagged command while tagged are active. */ if (ch->numrslotspd[t] != 0 && ch->numtslotspd[t] != 0) return (1); } if ((ccb->ccb_h.func_code == XPT_ATA_IO) && (ccb->ataio.cmd.flags & (CAM_ATAIO_CONTROL | CAM_ATAIO_NEEDRESULT))) { /* Atomic command while anything active. */ if (ch->numrslots != 0) return (1); } /* We have some atomic command running. */ if (ch->aslots != 0) return (1); return (0); } /* Must be called with channel locked. */ static void fsl_sata_begin_transaction(struct fsl_sata_channel *ch, union ccb *ccb) { struct fsl_sata_slot *slot; int tag, tags; CAM_DEBUG(ccb->ccb_h.path, CAM_DEBUG_TRACE, ("fsl_sata_begin_transaction func_code=0x%x\n", ccb->ccb_h.func_code)); /* Choose empty slot. */ tags = FSL_SATA_MAX_SLOTS; if ((ccb->ccb_h.func_code == XPT_ATA_IO) && (ccb->ataio.cmd.flags & CAM_ATAIO_FPDMA)) tags = ch->curr[ccb->ccb_h.target_id].tags; if (ch->lastslot + 1 < tags) tag = ffs(~(ch->oslots >> (ch->lastslot + 1))); else tag = 0; if (tag == 0 || tag + ch->lastslot >= tags) tag = ffs(~ch->oslots) - 1; else tag += ch->lastslot; ch->lastslot = tag; /* Occupy chosen slot. */ slot = &ch->slot[tag]; slot->ccb = ccb; slot->ttl = 0; /* Stop PM timer. */ if (ch->numrslots == 0 && ch->pm_level > 3) callout_stop(&ch->pm_timer); /* Update channel stats. */ ch->oslots |= (1 << tag); ch->numrslots++; ch->numrslotspd[ccb->ccb_h.target_id]++; if ((ccb->ccb_h.func_code == XPT_ATA_IO) && (ccb->ataio.cmd.flags & CAM_ATAIO_FPDMA)) { ch->numtslots++; ch->numtslotspd[ccb->ccb_h.target_id]++; ch->taggedtarget = ccb->ccb_h.target_id; } if ((ccb->ccb_h.func_code == XPT_ATA_IO) && (ccb->ataio.cmd.flags & (CAM_ATAIO_CONTROL | CAM_ATAIO_NEEDRESULT))) ch->aslots |= (1 << tag); if ((ccb->ccb_h.flags & CAM_DIR_MASK) != CAM_DIR_NONE) { slot->state = FSL_SATA_SLOT_LOADING; bus_dmamap_load_ccb(ch->dma.data_tag, slot->dma.data_map, ccb, fsl_sata_dmasetprd, slot, 0); } else { slot->dma.nsegs = 0; fsl_sata_execute_transaction(slot); } CAM_DEBUG(ccb->ccb_h.path, CAM_DEBUG_TRACE, ("fsl_sata_begin_transaction exit\n")); } /* Locked by busdma engine. */ static void fsl_sata_dmasetprd(void *arg, bus_dma_segment_t *segs, int nsegs, int error) { struct fsl_sata_slot *slot = arg; struct fsl_sata_channel *ch = slot->ch; struct fsl_sata_cmd_tab *ctp; struct fsl_sata_dma_prd *prd; int i, j, len, extlen; if (error) { device_printf(ch->dev, "DMA load error %d\n", error); fsl_sata_end_transaction(slot, FSL_SATA_ERR_INVALID); return; } KASSERT(nsegs <= FSL_SATA_SG_ENTRIES - 1, ("too many DMA segment entries\n")); /* Get a piece of the workspace for this request */ ctp = FSL_SATA_CTP(ch, slot); /* Fill S/G table */ prd = &ctp->prd_tab[0]; for (i = 0, j = 0; i < nsegs; i++, j++) { if (j == FSL_SATA_PRD_EXT_INDEX && FSL_SATA_PRD_MAX_DIRECT < nsegs) { - prd[j].dba = FSL_SATA_CTP_BUS(ch, slot) + - FSL_SATA_PRD_OFFSET(j+1); + prd[j].dba = htole32(FSL_SATA_CTP_BUS(ch, slot) + + FSL_SATA_PRD_OFFSET(j+1)); j++; extlen = 0; } len = segs[i].ds_len; len = roundup2(len, sizeof(uint32_t)); prd[j].dba = htole32((uint32_t)segs[i].ds_addr); prd[j].dwc_flg = htole32(FSL_SATA_PRD_SNOOP | len); slot->ttl += len; if (j > FSL_SATA_PRD_MAX_DIRECT) extlen += len; } slot->dma.nsegs = j; if (j > FSL_SATA_PRD_MAX_DIRECT) prd[FSL_SATA_PRD_EXT_INDEX].dwc_flg = htole32(FSL_SATA_PRD_SNOOP | FSL_SATA_PRD_EXT | extlen); bus_dmamap_sync(ch->dma.data_tag, slot->dma.data_map, ((slot->ccb->ccb_h.flags & CAM_DIR_IN) ? BUS_DMASYNC_PREREAD : BUS_DMASYNC_PREWRITE)); fsl_sata_execute_transaction(slot); } /* Must be called with channel locked. */ static void fsl_sata_execute_transaction(struct fsl_sata_slot *slot) { struct fsl_sata_channel *ch = slot->ch; struct fsl_sata_cmd_tab *ctp; struct fsl_sata_cmd_list *clp; union ccb *ccb = slot->ccb; int port = ccb->ccb_h.target_id & 0x0f; int fis_size, i, softreset; uint32_t tmp; uint32_t cmd_flags = FSL_SATA_CMD_WRITE | FSL_SATA_CMD_SNOOP; softreset = 0; CAM_DEBUG(ccb->ccb_h.path, CAM_DEBUG_TRACE, ("fsl_sata_execute_transaction func_code=0x%x\n", ccb->ccb_h.func_code)); /* Get a piece of the workspace for this request */ ctp = FSL_SATA_CTP(ch, slot); /* Setup the FIS for this request */ if (!(fis_size = fsl_sata_setup_fis(ch, ctp, ccb, slot->slot))) { device_printf(ch->dev, "Setting up SATA FIS failed\n"); fsl_sata_end_transaction(slot, FSL_SATA_ERR_INVALID); return; } /* Setup the command list entry */ clp = FSL_SATA_CLP(ch, slot); clp->fis_length = htole16(fis_size); clp->prd_length = htole16(slot->dma.nsegs); /* Special handling for Soft Reset command. */ if ((ccb->ccb_h.func_code == XPT_ATA_IO) && (ccb->ataio.cmd.flags & CAM_ATAIO_CONTROL)) { if (ccb->ataio.cmd.control & ATA_A_RESET) { softreset = 1; cmd_flags |= FSL_SATA_CMD_RESET; } else { /* Prepare FIS receive area for check. */ for (i = 0; i < 32; i++) ctp->sfis[i] = 0xff; softreset = 2; } } if (ccb->ataio.cmd.flags & CAM_ATAIO_FPDMA) cmd_flags |= FSL_SATA_CMD_QUEUED; clp->cmd_flags = htole32(cmd_flags | (ccb->ccb_h.func_code == XPT_SCSI_IO ? FSL_SATA_CMD_ATAPI : 0) | slot->slot); clp->ttl = htole32(slot->ttl); clp->cda = htole32(FSL_SATA_CTP_BUS(ch, slot)); bus_dmamap_sync(ch->dma.work_tag, ch->dma.work_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); /* Issue command to the controller. */ slot->state = FSL_SATA_SLOT_RUNNING; ch->rslots |= (1 << slot->slot); ATA_OUTL(ch->r_mem, FSL_SATA_P_CQPMP, port); ATA_OUTL(ch->r_mem, FSL_SATA_P_CQR, (1 << slot->slot)); /* Device reset commands don't interrupt. Poll them. */ if (ccb->ccb_h.func_code == XPT_ATA_IO && (ccb->ataio.cmd.command == ATA_DEVICE_RESET || softreset)) { int count, timeout = ccb->ccb_h.timeout * 100; enum fsl_sata_err_type et = FSL_SATA_ERR_NONE; for (count = 0; count < timeout; count++) { DELAY(10); tmp = 0; if (softreset == 2) { tmp = ATA_INL(ch->r_mem, FSL_SATA_P_SIG); if (tmp != 0 && tmp != 0xffffffff) break; continue; } if ((ATA_INL(ch->r_mem, FSL_SATA_P_CCR) & (1 << slot->slot)) != 0) break; } if (timeout && (count >= timeout)) { device_printf(ch->dev, "Poll timeout on slot %d port %d (round %d)\n", slot->slot, port, softreset); device_printf(ch->dev, "hsts %08x cqr %08x ccr %08x ss %08x " "rs %08x cer %08x der %08x serr %08x car %08x sig %08x\n", ATA_INL(ch->r_mem, FSL_SATA_P_HSTS), ATA_INL(ch->r_mem, FSL_SATA_P_CQR), ATA_INL(ch->r_mem, FSL_SATA_P_CCR), ATA_INL(ch->r_mem, FSL_SATA_P_SSTS), ch->rslots, ATA_INL(ch->r_mem, FSL_SATA_P_CER), ATA_INL(ch->r_mem, FSL_SATA_P_DER), ATA_INL(ch->r_mem, FSL_SATA_P_SERR), ATA_INL(ch->r_mem, FSL_SATA_P_CAR), ATA_INL(ch->r_mem, FSL_SATA_P_SIG)); et = FSL_SATA_ERR_TIMEOUT; } fsl_sata_end_transaction(slot, et); return; } /* Start command execution timeout */ callout_reset_sbt(&slot->timeout, SBT_1MS * ccb->ccb_h.timeout / 2, 0, (timeout_t*)fsl_sata_timeout, slot, 0); return; } /* Must be called with channel locked. */ static void fsl_sata_process_timeout(struct fsl_sata_channel *ch) { int i; mtx_assert(&ch->mtx, MA_OWNED); /* Handle the rest of commands. */ for (i = 0; i < FSL_SATA_MAX_SLOTS; i++) { /* Do we have a running request on slot? */ if (ch->slot[i].state < FSL_SATA_SLOT_RUNNING) continue; fsl_sata_end_transaction(&ch->slot[i], FSL_SATA_ERR_TIMEOUT); } } /* Must be called with channel locked. */ static void fsl_sata_rearm_timeout(struct fsl_sata_channel *ch) { int i; mtx_assert(&ch->mtx, MA_OWNED); for (i = 0; i < FSL_SATA_MAX_SLOTS; i++) { struct fsl_sata_slot *slot = &ch->slot[i]; /* Do we have a running request on slot? */ if (slot->state < FSL_SATA_SLOT_RUNNING) continue; if ((ch->toslots & (1 << i)) == 0) continue; callout_reset_sbt(&slot->timeout, SBT_1MS * slot->ccb->ccb_h.timeout / 2, 0, (timeout_t*)fsl_sata_timeout, slot, 0); } } /* Locked by callout mechanism. */ static void fsl_sata_timeout(struct fsl_sata_slot *slot) { struct fsl_sata_channel *ch = slot->ch; device_t dev = ch->dev; uint32_t sstatus; /* Check for stale timeout. */ if (slot->state < FSL_SATA_SLOT_RUNNING) return; /* Check if slot was not being executed last time we checked. */ if (slot->state < FSL_SATA_SLOT_EXECUTING) { /* Check if slot started executing. */ sstatus = ATA_INL(ch->r_mem, FSL_SATA_P_CAR); if ((sstatus & (1 << slot->slot)) != 0) slot->state = FSL_SATA_SLOT_EXECUTING; callout_reset_sbt(&slot->timeout, SBT_1MS * slot->ccb->ccb_h.timeout / 2, 0, (timeout_t*)fsl_sata_timeout, slot, 0); return; } device_printf(dev, "Timeout on slot %d port %d\n", slot->slot, slot->ccb->ccb_h.target_id & 0x0f); /* Handle frozen command. */ if (ch->frozen) { union ccb *fccb = ch->frozen; ch->frozen = NULL; fccb->ccb_h.status = CAM_REQUEUE_REQ | CAM_RELEASE_SIMQ; if (!(fccb->ccb_h.status & CAM_DEV_QFRZN)) { xpt_freeze_devq(fccb->ccb_h.path, 1); fccb->ccb_h.status |= CAM_DEV_QFRZN; } fsl_sata_done(ch, fccb); } if (ch->toslots == 0) xpt_freeze_simq(ch->sim, 1); ch->toslots |= (1 << slot->slot); if ((ch->rslots & ~ch->toslots) == 0) fsl_sata_process_timeout(ch); else device_printf(dev, " ... waiting for slots %08x\n", ch->rslots & ~ch->toslots); } /* Must be called with channel locked. */ static void fsl_sata_end_transaction(struct fsl_sata_slot *slot, enum fsl_sata_err_type et) { struct fsl_sata_channel *ch = slot->ch; union ccb *ccb = slot->ccb; struct fsl_sata_cmd_list *clp; int lastto; uint32_t sig; bus_dmamap_sync(ch->dma.work_tag, ch->dma.work_map, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); clp = FSL_SATA_CLP(ch, slot); /* Read result registers to the result struct */ if (ccb->ccb_h.func_code == XPT_ATA_IO) { struct ata_res *res = &ccb->ataio.res; if ((et == FSL_SATA_ERR_TFE) || (ccb->ataio.cmd.flags & CAM_ATAIO_NEEDRESULT)) { struct fsl_sata_cmd_tab *ctp = FSL_SATA_CTP(ch, slot); uint8_t *fis = ctp->sfis; res->status = fis[2]; res->error = fis[3]; res->lba_low = fis[4]; res->lba_mid = fis[5]; res->lba_high = fis[6]; res->device = fis[7]; res->lba_low_exp = fis[8]; res->lba_mid_exp = fis[9]; res->lba_high_exp = fis[10]; res->sector_count = fis[12]; res->sector_count_exp = fis[13]; if ((ccb->ataio.cmd.flags & CAM_ATAIO_CONTROL) && (ccb->ataio.cmd.control & ATA_A_RESET) == 0) { sig = ATA_INL(ch->r_mem, FSL_SATA_P_SIG); res->lba_high = sig >> 24; res->lba_mid = sig >> 16; res->lba_low = sig >> 8; res->sector_count = sig; } } else bzero(res, sizeof(*res)); if ((ccb->ataio.cmd.flags & CAM_ATAIO_FPDMA) == 0 && (ccb->ccb_h.flags & CAM_DIR_MASK) != CAM_DIR_NONE) { ccb->ataio.resid = ccb->ataio.dxfer_len - le32toh(clp->ttl); } } else { if ((ccb->ccb_h.flags & CAM_DIR_MASK) != CAM_DIR_NONE) { ccb->csio.resid = ccb->csio.dxfer_len - le32toh(clp->ttl); } } if ((ccb->ccb_h.flags & CAM_DIR_MASK) != CAM_DIR_NONE) { bus_dmamap_sync(ch->dma.data_tag, slot->dma.data_map, (ccb->ccb_h.flags & CAM_DIR_IN) ? BUS_DMASYNC_POSTREAD : BUS_DMASYNC_POSTWRITE); bus_dmamap_unload(ch->dma.data_tag, slot->dma.data_map); } if (et != FSL_SATA_ERR_NONE) ch->eslots |= (1 << slot->slot); /* In case of error, freeze device for proper recovery. */ if ((et != FSL_SATA_ERR_NONE) && (!ch->recoverycmd) && !(ccb->ccb_h.status & CAM_DEV_QFRZN)) { xpt_freeze_devq(ccb->ccb_h.path, 1); ccb->ccb_h.status |= CAM_DEV_QFRZN; } /* Set proper result status. */ ccb->ccb_h.status &= ~CAM_STATUS_MASK; switch (et) { case FSL_SATA_ERR_NONE: ccb->ccb_h.status |= CAM_REQ_CMP; if (ccb->ccb_h.func_code == XPT_SCSI_IO) ccb->csio.scsi_status = SCSI_STATUS_OK; break; case FSL_SATA_ERR_INVALID: ch->fatalerr = 1; ccb->ccb_h.status |= CAM_REQ_INVALID; break; case FSL_SATA_ERR_INNOCENT: ccb->ccb_h.status |= CAM_REQUEUE_REQ; break; case FSL_SATA_ERR_TFE: case FSL_SATA_ERR_NCQ: if (ccb->ccb_h.func_code == XPT_SCSI_IO) { ccb->ccb_h.status |= CAM_SCSI_STATUS_ERROR; ccb->csio.scsi_status = SCSI_STATUS_CHECK_COND; } else { ccb->ccb_h.status |= CAM_ATA_STATUS_ERROR; } break; case FSL_SATA_ERR_SATA: ch->fatalerr = 1; if (!ch->recoverycmd) { xpt_freeze_simq(ch->sim, 1); ccb->ccb_h.status &= ~CAM_STATUS_MASK; ccb->ccb_h.status |= CAM_RELEASE_SIMQ; } ccb->ccb_h.status |= CAM_UNCOR_PARITY; break; case FSL_SATA_ERR_TIMEOUT: if (!ch->recoverycmd) { xpt_freeze_simq(ch->sim, 1); ccb->ccb_h.status &= ~CAM_STATUS_MASK; ccb->ccb_h.status |= CAM_RELEASE_SIMQ; } ccb->ccb_h.status |= CAM_CMD_TIMEOUT; break; default: ch->fatalerr = 1; ccb->ccb_h.status |= CAM_REQ_CMP_ERR; } /* Free slot. */ ch->oslots &= ~(1 << slot->slot); ch->rslots &= ~(1 << slot->slot); ch->aslots &= ~(1 << slot->slot); slot->state = FSL_SATA_SLOT_EMPTY; slot->ccb = NULL; /* Update channel stats. */ ch->numrslots--; ch->numrslotspd[ccb->ccb_h.target_id]--; ATA_OUTL(ch->r_mem, FSL_SATA_P_CCR, 1 << slot->slot); if ((ccb->ccb_h.func_code == XPT_ATA_IO) && (ccb->ataio.cmd.flags & CAM_ATAIO_FPDMA)) { ch->numtslots--; ch->numtslotspd[ccb->ccb_h.target_id]--; } /* Cancel timeout state if request completed normally. */ if (et != FSL_SATA_ERR_TIMEOUT) { lastto = (ch->toslots == (1 << slot->slot)); ch->toslots &= ~(1 << slot->slot); if (lastto) xpt_release_simq(ch->sim, TRUE); } /* If it was first request of reset sequence and there is no error, * proceed to second request. */ if ((ccb->ccb_h.func_code == XPT_ATA_IO) && (ccb->ataio.cmd.flags & CAM_ATAIO_CONTROL) && (ccb->ataio.cmd.control & ATA_A_RESET) && et == FSL_SATA_ERR_NONE) { ccb->ataio.cmd.control &= ~ATA_A_RESET; fsl_sata_begin_transaction(ch, ccb); return; } /* If it was our READ LOG command - process it. */ if (ccb->ccb_h.recovery_type == RECOVERY_READ_LOG) { fsl_sata_process_read_log(ch, ccb); /* If it was our REQUEST SENSE command - process it. */ } else if (ccb->ccb_h.recovery_type == RECOVERY_REQUEST_SENSE) { fsl_sata_process_request_sense(ch, ccb); /* If it was NCQ or ATAPI command error, put result on hold. */ } else if (et == FSL_SATA_ERR_NCQ || ((ccb->ccb_h.status & CAM_STATUS_MASK) == CAM_SCSI_STATUS_ERROR && (ccb->ccb_h.flags & CAM_DIS_AUTOSENSE) == 0)) { ch->hold[slot->slot] = ccb; ch->numhslots++; } else fsl_sata_done(ch, ccb); /* If we have no other active commands, ... */ if (ch->rslots == 0) { /* if there was fatal error - reset port. */ if (ch->toslots != 0 || ch->fatalerr) { fsl_sata_reset(ch); } else { /* if we have slots in error, we can reinit port. */ if (ch->eslots != 0) { fsl_sata_stop(ch); fsl_sata_start(ch); } /* if there commands on hold, we can do READ LOG. */ if (!ch->recoverycmd && ch->numhslots) fsl_sata_issue_recovery(ch); } /* If all the rest of commands are in timeout - give them chance. */ } else if ((ch->rslots & ~ch->toslots) == 0 && et != FSL_SATA_ERR_TIMEOUT) fsl_sata_rearm_timeout(ch); /* Unfreeze frozen command. */ if (ch->frozen && !fsl_sata_check_collision(ch, ch->frozen)) { union ccb *fccb = ch->frozen; ch->frozen = NULL; fsl_sata_begin_transaction(ch, fccb); xpt_release_simq(ch->sim, TRUE); } /* Start PM timer. */ if (ch->numrslots == 0 && ch->pm_level > 3 && (ch->curr[ch->pm_present ? 15 : 0].caps & CTS_SATA_CAPS_D_PMREQ)) { callout_schedule(&ch->pm_timer, (ch->pm_level == 4) ? hz / 1000 : hz / 8); } } static void fsl_sata_issue_recovery(struct fsl_sata_channel *ch) { union ccb *ccb; struct ccb_ataio *ataio; struct ccb_scsiio *csio; int i; /* Find some held command. */ for (i = 0; i < FSL_SATA_MAX_SLOTS; i++) { if (ch->hold[i]) break; } ccb = xpt_alloc_ccb_nowait(); if (ccb == NULL) { device_printf(ch->dev, "Unable to allocate recovery command\n"); completeall: /* We can't do anything -- complete held commands. */ for (i = 0; i < FSL_SATA_MAX_SLOTS; i++) { if (ch->hold[i] == NULL) continue; ch->hold[i]->ccb_h.status &= ~CAM_STATUS_MASK; ch->hold[i]->ccb_h.status |= CAM_RESRC_UNAVAIL; fsl_sata_done(ch, ch->hold[i]); ch->hold[i] = NULL; ch->numhslots--; } fsl_sata_reset(ch); return; } ccb->ccb_h = ch->hold[i]->ccb_h; /* Reuse old header. */ if (ccb->ccb_h.func_code == XPT_ATA_IO) { /* READ LOG */ ccb->ccb_h.recovery_type = RECOVERY_READ_LOG; ccb->ccb_h.func_code = XPT_ATA_IO; ccb->ccb_h.flags = CAM_DIR_IN; ccb->ccb_h.timeout = 1000; /* 1s should be enough. */ ataio = &ccb->ataio; ataio->data_ptr = malloc(512, M_FSL_SATA, M_NOWAIT); if (ataio->data_ptr == NULL) { xpt_free_ccb(ccb); device_printf(ch->dev, "Unable to allocate memory for READ LOG command\n"); goto completeall; } ataio->dxfer_len = 512; bzero(&ataio->cmd, sizeof(ataio->cmd)); ataio->cmd.flags = CAM_ATAIO_48BIT; ataio->cmd.command = 0x2F; /* READ LOG EXT */ ataio->cmd.sector_count = 1; ataio->cmd.sector_count_exp = 0; ataio->cmd.lba_low = 0x10; ataio->cmd.lba_mid = 0; ataio->cmd.lba_mid_exp = 0; } else { /* REQUEST SENSE */ ccb->ccb_h.recovery_type = RECOVERY_REQUEST_SENSE; ccb->ccb_h.recovery_slot = i; ccb->ccb_h.func_code = XPT_SCSI_IO; ccb->ccb_h.flags = CAM_DIR_IN; ccb->ccb_h.status = 0; ccb->ccb_h.timeout = 1000; /* 1s should be enough. */ csio = &ccb->csio; csio->data_ptr = (void *)&ch->hold[i]->csio.sense_data; csio->dxfer_len = ch->hold[i]->csio.sense_len; csio->cdb_len = 6; bzero(&csio->cdb_io, sizeof(csio->cdb_io)); csio->cdb_io.cdb_bytes[0] = 0x03; csio->cdb_io.cdb_bytes[4] = csio->dxfer_len; } /* Freeze SIM while doing recovery. */ ch->recoverycmd = 1; xpt_freeze_simq(ch->sim, 1); fsl_sata_begin_transaction(ch, ccb); } static void fsl_sata_process_read_log(struct fsl_sata_channel *ch, union ccb *ccb) { uint8_t *data; struct ata_res *res; int i; ch->recoverycmd = 0; data = ccb->ataio.data_ptr; if ((ccb->ccb_h.status & CAM_STATUS_MASK) == CAM_REQ_CMP && (data[0] & 0x80) == 0) { for (i = 0; i < FSL_SATA_MAX_SLOTS; i++) { if (!ch->hold[i]) continue; if (ch->hold[i]->ccb_h.func_code != XPT_ATA_IO) continue; if ((data[0] & 0x1F) == i) { res = &ch->hold[i]->ataio.res; res->status = data[2]; res->error = data[3]; res->lba_low = data[4]; res->lba_mid = data[5]; res->lba_high = data[6]; res->device = data[7]; res->lba_low_exp = data[8]; res->lba_mid_exp = data[9]; res->lba_high_exp = data[10]; res->sector_count = data[12]; res->sector_count_exp = data[13]; } else { ch->hold[i]->ccb_h.status &= ~CAM_STATUS_MASK; ch->hold[i]->ccb_h.status |= CAM_REQUEUE_REQ; } fsl_sata_done(ch, ch->hold[i]); ch->hold[i] = NULL; ch->numhslots--; } } else { if ((ccb->ccb_h.status & CAM_STATUS_MASK) != CAM_REQ_CMP) device_printf(ch->dev, "Error while READ LOG EXT\n"); else if ((data[0] & 0x80) == 0) { device_printf(ch->dev, "Non-queued command error in READ LOG EXT\n"); } for (i = 0; i < FSL_SATA_MAX_SLOTS; i++) { if (!ch->hold[i]) continue; if (ch->hold[i]->ccb_h.func_code != XPT_ATA_IO) continue; fsl_sata_done(ch, ch->hold[i]); ch->hold[i] = NULL; ch->numhslots--; } } free(ccb->ataio.data_ptr, M_FSL_SATA); xpt_free_ccb(ccb); xpt_release_simq(ch->sim, TRUE); } static void fsl_sata_process_request_sense(struct fsl_sata_channel *ch, union ccb *ccb) { int i; ch->recoverycmd = 0; i = ccb->ccb_h.recovery_slot; if ((ccb->ccb_h.status & CAM_STATUS_MASK) == CAM_REQ_CMP) { ch->hold[i]->ccb_h.status |= CAM_AUTOSNS_VALID; } else { ch->hold[i]->ccb_h.status &= ~CAM_STATUS_MASK; ch->hold[i]->ccb_h.status |= CAM_AUTOSENSE_FAIL; } fsl_sata_done(ch, ch->hold[i]); ch->hold[i] = NULL; ch->numhslots--; xpt_free_ccb(ccb); xpt_release_simq(ch->sim, TRUE); } static void fsl_sata_start(struct fsl_sata_channel *ch) { u_int32_t cmd; /* Clear SATA error register */ ATA_OUTL(ch->r_mem, FSL_SATA_P_SERR, 0xFFFFFFFF); /* Clear any interrupts pending on this channel */ ATA_OUTL(ch->r_mem, FSL_SATA_P_HSTS, 0x3F); ATA_OUTL(ch->r_mem, FSL_SATA_P_CER, 0xFFFF); ATA_OUTL(ch->r_mem, FSL_SATA_P_DER, 0xFFFF); /* Start operations on this channel */ cmd = ATA_INL(ch->r_mem, FSL_SATA_P_HCTRL); cmd |= FSL_SATA_P_HCTRL_HC_ON | FSL_SATA_P_HCTRL_SNOOP; cmd &= ~FSL_SATA_P_HCTRL_HC_FORCE_OFF; ATA_OUTL(ch->r_mem, FSL_SATA_P_HCTRL, cmd | (ch->pm_present ? FSL_SATA_P_HCTRL_PM : 0)); fsl_sata_wait_register(ch, FSL_SATA_P_HSTS, FSL_SATA_P_HSTS_PR, FSL_SATA_P_HSTS_PR, 500); ATA_OUTL(ch->r_mem, FSL_SATA_P_HSTS, ATA_INL(ch->r_mem, FSL_SATA_P_HSTS) & FSL_SATA_P_HSTS_PR); } static void fsl_sata_stop(struct fsl_sata_channel *ch) { uint32_t cmd; int i; /* Kill all activity on this channel */ cmd = ATA_INL(ch->r_mem, FSL_SATA_P_HCTRL); cmd &= ~FSL_SATA_P_HCTRL_HC_ON; for (i = 0; i < 2; i++) { ATA_OUTL(ch->r_mem, FSL_SATA_P_HCTRL, cmd); if (fsl_sata_wait_register(ch, FSL_SATA_P_HSTS, FSL_SATA_P_HSTS_HS_ON, 0, 500)) { if (i != 0) device_printf(ch->dev, "stopping FSL SATA engine failed\n"); cmd |= FSL_SATA_P_HCTRL_HC_FORCE_OFF; } else break; } ch->eslots = 0; } static void fsl_sata_reset(struct fsl_sata_channel *ch) { uint32_t ctrl; int i; xpt_freeze_simq(ch->sim, 1); if (bootverbose) device_printf(ch->dev, "FSL SATA reset...\n"); /* Requeue freezed command. */ if (ch->frozen) { union ccb *fccb = ch->frozen; ch->frozen = NULL; fccb->ccb_h.status = CAM_REQUEUE_REQ | CAM_RELEASE_SIMQ; if (!(fccb->ccb_h.status & CAM_DEV_QFRZN)) { xpt_freeze_devq(fccb->ccb_h.path, 1); fccb->ccb_h.status |= CAM_DEV_QFRZN; } fsl_sata_done(ch, fccb); } /* Kill the engine and requeue all running commands. */ fsl_sata_stop(ch); DELAY(1000); /* sleep for 1ms */ for (i = 0; i < FSL_SATA_MAX_SLOTS; i++) { /* Do we have a running request on slot? */ if (ch->slot[i].state < FSL_SATA_SLOT_RUNNING) continue; /* XXX; Commands in loading state. */ fsl_sata_end_transaction(&ch->slot[i], FSL_SATA_ERR_INNOCENT); } for (i = 0; i < FSL_SATA_MAX_SLOTS; i++) { if (!ch->hold[i]) continue; fsl_sata_done(ch, ch->hold[i]); ch->hold[i] = NULL; ch->numhslots--; } if (ch->toslots != 0) xpt_release_simq(ch->sim, TRUE); ch->eslots = 0; ch->toslots = 0; ch->fatalerr = 0; /* Tell the XPT about the event */ xpt_async(AC_BUS_RESET, ch->path, NULL); /* Disable port interrupts */ ATA_OUTL(ch->r_mem, FSL_SATA_P_HCTRL, ATA_INL(ch->r_mem, FSL_SATA_P_HCTRL) & ~0x3f); /* Reset and reconnect PHY, */ fsl_sata_start(ch); if (fsl_sata_wait_register(ch, FSL_SATA_P_HSTS, 0x08, 0x08, 500)) { if (bootverbose) device_printf(ch->dev, "FSL SATA reset: device not found\n"); ch->devices = 0; /* Enable wanted port interrupts */ ATA_OUTL(ch->r_mem, FSL_SATA_P_HCTRL, ATA_INL(ch->r_mem, FSL_SATA_P_HCTRL) | FSL_SATA_P_HCTRL_PHYRDY); xpt_release_simq(ch->sim, TRUE); return; } if (bootverbose) device_printf(ch->dev, "FSL SATA reset: device found\n"); ch->devices = 1; /* Enable wanted port interrupts */ ctrl = ATA_INL(ch->r_mem, FSL_SATA_P_HCTRL) & ~0x3f; ATA_OUTL(ch->r_mem, FSL_SATA_P_HCTRL, ctrl | FSL_SATA_P_HCTRL_FATAL | FSL_SATA_P_HCTRL_PHYRDY | FSL_SATA_P_HCTRL_SIG | FSL_SATA_P_HCTRL_SNTFY | FSL_SATA_P_HCTRL_DE | FSL_SATA_P_HCTRL_CC); xpt_release_simq(ch->sim, TRUE); } static int fsl_sata_setup_fis(struct fsl_sata_channel *ch, struct fsl_sata_cmd_tab *ctp, union ccb *ccb, int tag) { uint8_t *fis = &ctp->cfis[0]; bzero(fis, 32); fis[0] = 0x27; /* host to device */ fis[1] = (ccb->ccb_h.target_id & 0x0f); if (ccb->ccb_h.func_code == XPT_SCSI_IO) { fis[1] |= 0x80; fis[2] = ATA_PACKET_CMD; if ((ccb->ccb_h.flags & CAM_DIR_MASK) != CAM_DIR_NONE && ch->curr[ccb->ccb_h.target_id].mode >= ATA_DMA) fis[3] = ATA_F_DMA; else { fis[5] = ccb->csio.dxfer_len; fis[6] = ccb->csio.dxfer_len >> 8; } fis[7] = ATA_D_LBA; fis[15] = ATA_A_4BIT; bcopy((ccb->ccb_h.flags & CAM_CDB_POINTER) ? ccb->csio.cdb_io.cdb_ptr : ccb->csio.cdb_io.cdb_bytes, ctp->acmd, ccb->csio.cdb_len); bzero(ctp->acmd + ccb->csio.cdb_len, 32 - ccb->csio.cdb_len); } else if ((ccb->ataio.cmd.flags & CAM_ATAIO_CONTROL) == 0) { fis[1] |= 0x80; fis[2] = ccb->ataio.cmd.command; fis[3] = ccb->ataio.cmd.features; fis[4] = ccb->ataio.cmd.lba_low; fis[5] = ccb->ataio.cmd.lba_mid; fis[6] = ccb->ataio.cmd.lba_high; fis[7] = ccb->ataio.cmd.device; fis[8] = ccb->ataio.cmd.lba_low_exp; fis[9] = ccb->ataio.cmd.lba_mid_exp; fis[10] = ccb->ataio.cmd.lba_high_exp; fis[11] = ccb->ataio.cmd.features_exp; if (ccb->ataio.cmd.flags & CAM_ATAIO_FPDMA) { fis[12] = tag << 3; fis[13] = 0; } else { fis[12] = ccb->ataio.cmd.sector_count; fis[13] = ccb->ataio.cmd.sector_count_exp; } fis[15] = ATA_A_4BIT; } else { fis[15] = ccb->ataio.cmd.control; } return (20); } static int fsl_sata_check_ids(struct fsl_sata_channel *ch, union ccb *ccb) { if (ccb->ccb_h.target_id > 15) { ccb->ccb_h.status = CAM_TID_INVALID; fsl_sata_done(ch, ccb); return (-1); } if (ccb->ccb_h.target_lun != 0) { ccb->ccb_h.status = CAM_LUN_INVALID; fsl_sata_done(ch, ccb); return (-1); } return (0); } static void fsl_sataaction(struct cam_sim *sim, union ccb *ccb) { struct fsl_sata_channel *ch; CAM_DEBUG(ccb->ccb_h.path, CAM_DEBUG_TRACE, ("fsl_sataaction func_code=0x%x\n", ccb->ccb_h.func_code)); ch = (struct fsl_sata_channel *)cam_sim_softc(sim); switch (ccb->ccb_h.func_code) { /* Common cases first */ case XPT_ATA_IO: /* Execute the requested I/O operation */ case XPT_SCSI_IO: if (fsl_sata_check_ids(ch, ccb)) return; if (ch->devices == 0 || (ch->pm_present == 0 && ccb->ccb_h.target_id > 0 && ccb->ccb_h.target_id < 15)) { ccb->ccb_h.status = CAM_SEL_TIMEOUT; break; } ccb->ccb_h.recovery_type = RECOVERY_NONE; /* Check for command collision. */ if (fsl_sata_check_collision(ch, ccb)) { /* Freeze command. */ ch->frozen = ccb; /* We have only one frozen slot, so freeze simq also. */ xpt_freeze_simq(ch->sim, 1); return; } fsl_sata_begin_transaction(ch, ccb); return; case XPT_ABORT: /* Abort the specified CCB */ /* XXX Implement */ ccb->ccb_h.status = CAM_REQ_INVALID; break; case XPT_SET_TRAN_SETTINGS: { struct ccb_trans_settings *cts = &ccb->cts; struct fsl_sata_device *d; if (fsl_sata_check_ids(ch, ccb)) return; if (cts->type == CTS_TYPE_CURRENT_SETTINGS) d = &ch->curr[ccb->ccb_h.target_id]; else d = &ch->user[ccb->ccb_h.target_id]; if (cts->xport_specific.sata.valid & CTS_SATA_VALID_REVISION) d->revision = cts->xport_specific.sata.revision; if (cts->xport_specific.sata.valid & CTS_SATA_VALID_MODE) d->mode = cts->xport_specific.sata.mode; if (cts->xport_specific.sata.valid & CTS_SATA_VALID_BYTECOUNT) d->bytecount = min(8192, cts->xport_specific.sata.bytecount); if (cts->xport_specific.sata.valid & CTS_SATA_VALID_TAGS) d->tags = min(FSL_SATA_MAX_SLOTS, cts->xport_specific.sata.tags); if (cts->xport_specific.sata.valid & CTS_SATA_VALID_PM) ch->pm_present = cts->xport_specific.sata.pm_present; if (cts->xport_specific.sata.valid & CTS_SATA_VALID_ATAPI) d->atapi = cts->xport_specific.sata.atapi; ccb->ccb_h.status = CAM_REQ_CMP; break; } case XPT_GET_TRAN_SETTINGS: /* Get default/user set transfer settings for the target */ { struct ccb_trans_settings *cts = &ccb->cts; struct fsl_sata_device *d; uint32_t status; if (fsl_sata_check_ids(ch, ccb)) return; if (cts->type == CTS_TYPE_CURRENT_SETTINGS) d = &ch->curr[ccb->ccb_h.target_id]; else d = &ch->user[ccb->ccb_h.target_id]; cts->protocol = PROTO_UNSPECIFIED; cts->protocol_version = PROTO_VERSION_UNSPECIFIED; cts->transport = XPORT_SATA; cts->transport_version = XPORT_VERSION_UNSPECIFIED; cts->proto_specific.valid = 0; cts->xport_specific.sata.valid = 0; if (cts->type == CTS_TYPE_CURRENT_SETTINGS && (ccb->ccb_h.target_id == 15 || (ccb->ccb_h.target_id == 0 && !ch->pm_present))) { status = ATA_INL(ch->r_mem, FSL_SATA_P_SSTS) & ATA_SS_SPD_MASK; if (status & 0x0f0) { cts->xport_specific.sata.revision = (status & 0x0f0) >> 4; cts->xport_specific.sata.valid |= CTS_SATA_VALID_REVISION; } cts->xport_specific.sata.caps = d->caps & CTS_SATA_CAPS_D; if (ch->pm_level) { cts->xport_specific.sata.caps |= CTS_SATA_CAPS_H_PMREQ; } cts->xport_specific.sata.caps |= CTS_SATA_CAPS_H_AN; cts->xport_specific.sata.caps &= ch->user[ccb->ccb_h.target_id].caps; cts->xport_specific.sata.valid |= CTS_SATA_VALID_CAPS; } else { cts->xport_specific.sata.revision = d->revision; cts->xport_specific.sata.valid |= CTS_SATA_VALID_REVISION; cts->xport_specific.sata.caps = d->caps; cts->xport_specific.sata.valid |= CTS_SATA_VALID_CAPS; } cts->xport_specific.sata.mode = d->mode; cts->xport_specific.sata.valid |= CTS_SATA_VALID_MODE; cts->xport_specific.sata.bytecount = d->bytecount; cts->xport_specific.sata.valid |= CTS_SATA_VALID_BYTECOUNT; cts->xport_specific.sata.pm_present = ch->pm_present; cts->xport_specific.sata.valid |= CTS_SATA_VALID_PM; cts->xport_specific.sata.tags = d->tags; cts->xport_specific.sata.valid |= CTS_SATA_VALID_TAGS; cts->xport_specific.sata.atapi = d->atapi; cts->xport_specific.sata.valid |= CTS_SATA_VALID_ATAPI; ccb->ccb_h.status = CAM_REQ_CMP; break; } case XPT_RESET_BUS: /* Reset the specified SCSI bus */ case XPT_RESET_DEV: /* Bus Device Reset the specified SCSI device */ fsl_sata_reset(ch); ccb->ccb_h.status = CAM_REQ_CMP; break; case XPT_TERM_IO: /* Terminate the I/O process */ /* XXX Implement */ ccb->ccb_h.status = CAM_REQ_INVALID; break; case XPT_PATH_INQ: /* Path routing inquiry */ { struct ccb_pathinq *cpi = &ccb->cpi; cpi->version_num = 1; /* XXX??? */ cpi->hba_inquiry = PI_SDTR_ABLE; cpi->hba_inquiry |= PI_TAG_ABLE; #if 0 /* * XXX: CAM tries to reset port 15 if it sees port multiplier * support. Disable it for now. */ cpi->hba_inquiry |= PI_SATAPM; #endif cpi->target_sprt = 0; cpi->hba_misc = PIM_SEQSCAN | PIM_UNMAPPED; cpi->hba_eng_cnt = 0; /* * XXX: This should be 15, since hardware *does* support a port * multiplier. See above. */ cpi->max_target = 0; cpi->max_lun = 0; cpi->initiator_id = 0; cpi->bus_id = cam_sim_bus(sim); cpi->base_transfer_speed = 150000; strncpy(cpi->sim_vid, "FreeBSD", SIM_IDLEN); strncpy(cpi->hba_vid, "FSL SATA", HBA_IDLEN); strncpy(cpi->dev_name, cam_sim_name(sim), DEV_IDLEN); cpi->unit_number = cam_sim_unit(sim); cpi->transport = XPORT_SATA; cpi->transport_version = XPORT_VERSION_UNSPECIFIED; cpi->protocol = PROTO_ATA; cpi->protocol_version = PROTO_VERSION_UNSPECIFIED; cpi->maxio = MAXPHYS; cpi->ccb_h.status = CAM_REQ_CMP; break; } default: ccb->ccb_h.status = CAM_REQ_INVALID; break; } fsl_sata_done(ch, ccb); } static void fsl_satapoll(struct cam_sim *sim) { struct fsl_sata_channel *ch = (struct fsl_sata_channel *)cam_sim_softc(sim); uint32_t istatus; /* Read interrupt statuses and process if any. */ istatus = ATA_INL(ch->r_mem, FSL_SATA_P_HSTS); if (istatus != 0) fsl_sata_intr_main(ch, istatus); } MODULE_VERSION(fsl_sata, 1); MODULE_DEPEND(fsl_sata, cam, 1, 1, 1); Index: projects/runtime-coverage =================================================================== --- projects/runtime-coverage (revision 325498) +++ projects/runtime-coverage (revision 325499) Property changes on: projects/runtime-coverage ___________________________________________________________________ Modified: svn:mergeinfo ## -0,0 +0,1 ## Merged /head:r325454-325498