diff --git a/config/kernel-timer.m4 b/config/kernel-timer.m4 index 4dc3f84ed47e..b0e1afa153ab 100644 --- a/config/kernel-timer.m4 +++ b/config/kernel-timer.m4 @@ -1,32 +1,73 @@ +dnl # 4.14-rc3 API change +dnl # https://lwn.net/Articles/735887/ dnl # -dnl # 4.15 API change -dnl # https://lkml.org/lkml/2017/11/25/90 dnl # Check if timer_list.func get passed a timer_list or an unsigned long dnl # (older kernels). Also sanity check the from_timer() and timer_setup() dnl # macros are available as well, since they will be used in the same newer dnl # kernels that support the new timer_list.func signature. dnl # -AC_DEFUN([ZFS_AC_KERNEL_TIMER_FUNCTION_TIMER_LIST], [ - AC_MSG_CHECKING([whether timer_list.function gets a timer_list]) +dnl # Also check for the existance of flags in struct timer_list, they were +dnl # added in 4.1-rc8 via 0eeda71bc30d. + +AC_DEFUN([ZFS_AC_KERNEL_TIMER_SETUP], [ + AC_MSG_CHECKING([whether timer_setup() is available]) tmp_flags="$EXTRA_KCFLAGS" EXTRA_KCFLAGS="-Werror" + ZFS_LINUX_TRY_COMPILE([ #include - void task_expire(struct timer_list *tl) {} + + struct my_task_timer { + struct timer_list timer; + int data; + }; + + void task_expire(struct timer_list *tl) + { + struct my_task_timer *task_timer = from_timer(task_timer, tl, timer); + task_timer->data = 42; + } + ],[ + struct my_task_timer task_timer; + timer_setup(&task_timer.timer, task_expire, 0); ],[ - #ifndef from_timer - #error "No from_timer() macro" - #endif + AC_MSG_RESULT(yes) + AC_DEFINE(HAVE_KERNEL_TIMER_SETUP, 1, + [timer_setup() is available]) + ],[ + AC_MSG_RESULT(no) + ]) - struct timer_list timer; - timer.function = task_expire; - timer_setup(&timer, NULL, 0); + AC_MSG_CHECKING([whether timer function expects timer_list]) + + ZFS_LINUX_TRY_COMPILE([ + #include + void task_expire(struct timer_list *tl) {} + ],[ + struct timer_list tl; + tl.function = task_expire; ],[ AC_MSG_RESULT(yes) AC_DEFINE(HAVE_KERNEL_TIMER_FUNCTION_TIMER_LIST, 1, [timer_list.function gets a timer_list]) ],[ AC_MSG_RESULT(no) ]) + + AC_MSG_CHECKING([whether struct timer_list has flags]) + + ZFS_LINUX_TRY_COMPILE([ + #include + ],[ + struct timer_list tl; + tl.flags = 2; + ],[ + AC_MSG_RESULT(yes) + AC_DEFINE(HAVE_KERNEL_TIMER_LIST_FLAGS, 1, + [struct timer_list has a flags member]) + ],[ + AC_MSG_RESULT(no) + ]) + EXTRA_KCFLAGS="$tmp_flags" ]) diff --git a/config/kernel.m4 b/config/kernel.m4 index 026a5258f9f0..54f39164bb5f 100644 --- a/config/kernel.m4 +++ b/config/kernel.m4 @@ -1,682 +1,682 @@ dnl # dnl # Default ZFS kernel configuration dnl # AC_DEFUN([ZFS_AC_CONFIG_KERNEL], [ ZFS_AC_KERNEL ZFS_AC_QAT ZFS_AC_KERNEL_ACCESS_OK_TYPE ZFS_AC_TEST_MODULE ZFS_AC_KERNEL_MISC_MINOR ZFS_AC_KERNEL_OBJTOOL ZFS_AC_KERNEL_CONFIG ZFS_AC_KERNEL_CTL_NAME ZFS_AC_KERNEL_PDE_DATA ZFS_AC_KERNEL_SET_FS_PWD_WITH_CONST ZFS_AC_KERNEL_2ARGS_VFS_FSYNC ZFS_AC_KERNEL_FS_STRUCT_SPINLOCK ZFS_AC_KERNEL_KUIDGID_T ZFS_AC_KERNEL_FALLOCATE ZFS_AC_KERNEL_2ARGS_ZLIB_DEFLATE_WORKSPACESIZE ZFS_AC_KERNEL_RWSEM_SPINLOCK_IS_RAW ZFS_AC_KERNEL_RWSEM_ACTIVITY ZFS_AC_KERNEL_RWSEM_ATOMIC_LONG_COUNT ZFS_AC_KERNEL_SCHED_RT_HEADER ZFS_AC_KERNEL_SCHED_SIGNAL_HEADER ZFS_AC_KERNEL_IO_SCHEDULE_TIMEOUT ZFS_AC_KERNEL_4ARGS_VFS_GETATTR ZFS_AC_KERNEL_3ARGS_VFS_GETATTR ZFS_AC_KERNEL_2ARGS_VFS_GETATTR ZFS_AC_KERNEL_USLEEP_RANGE ZFS_AC_KERNEL_KMEM_CACHE_ALLOCFLAGS ZFS_AC_KERNEL_KMEM_CACHE_CREATE_USERCOPY ZFS_AC_KERNEL_WAIT_ON_BIT ZFS_AC_KERNEL_WAIT_QUEUE_ENTRY_T ZFS_AC_KERNEL_WAIT_QUEUE_HEAD_ENTRY ZFS_AC_KERNEL_INODE_TIMES ZFS_AC_KERNEL_INODE_LOCK ZFS_AC_KERNEL_GROUP_INFO_GID ZFS_AC_KERNEL_WRITE ZFS_AC_KERNEL_READ - ZFS_AC_KERNEL_TIMER_FUNCTION_TIMER_LIST + ZFS_AC_KERNEL_TIMER_SETUP ZFS_AC_KERNEL_DECLARE_EVENT_CLASS ZFS_AC_KERNEL_CURRENT_BIO_TAIL ZFS_AC_KERNEL_SUPER_USER_NS ZFS_AC_KERNEL_SUBMIT_BIO ZFS_AC_KERNEL_BLOCK_DEVICE_OPERATIONS_CHECK_EVENTS ZFS_AC_KERNEL_BLOCK_DEVICE_OPERATIONS_RELEASE_VOID ZFS_AC_KERNEL_TYPE_FMODE_T ZFS_AC_KERNEL_BLKDEV_GET_BY_PATH ZFS_AC_KERNEL_BLKDEV_REREAD_PART ZFS_AC_KERNEL_OPEN_BDEV_EXCLUSIVE ZFS_AC_KERNEL_LOOKUP_BDEV ZFS_AC_KERNEL_INVALIDATE_BDEV_ARGS ZFS_AC_KERNEL_BDEV_LOGICAL_BLOCK_SIZE ZFS_AC_KERNEL_BDEV_PHYSICAL_BLOCK_SIZE ZFS_AC_KERNEL_BIO_BVEC_ITER ZFS_AC_KERNEL_BIO_FAILFAST_DTD ZFS_AC_KERNEL_BIO_SET_DEV ZFS_AC_KERNEL_REQ_FAILFAST_MASK ZFS_AC_KERNEL_REQ_OP_DISCARD ZFS_AC_KERNEL_REQ_OP_SECURE_ERASE ZFS_AC_KERNEL_REQ_OP_FLUSH ZFS_AC_KERNEL_BIO_BI_OPF ZFS_AC_KERNEL_BIO_END_IO_T_ARGS ZFS_AC_KERNEL_BIO_BI_STATUS ZFS_AC_KERNEL_BIO_RW_BARRIER ZFS_AC_KERNEL_BIO_RW_DISCARD ZFS_AC_KERNEL_BLK_QUEUE_BDI ZFS_AC_KERNEL_BLK_QUEUE_FLAG_CLEAR ZFS_AC_KERNEL_BLK_QUEUE_FLAG_SET ZFS_AC_KERNEL_BLK_QUEUE_FLUSH ZFS_AC_KERNEL_BLK_QUEUE_MAX_HW_SECTORS ZFS_AC_KERNEL_BLK_QUEUE_MAX_SEGMENTS ZFS_AC_KERNEL_BLK_QUEUE_HAVE_BIO_RW_UNPLUG ZFS_AC_KERNEL_BLK_QUEUE_HAVE_BLK_PLUG ZFS_AC_KERNEL_GET_DISK_AND_MODULE ZFS_AC_KERNEL_GET_DISK_RO ZFS_AC_KERNEL_HAVE_BIO_SET_OP_ATTRS ZFS_AC_KERNEL_GENERIC_READLINK_GLOBAL ZFS_AC_KERNEL_DISCARD_GRANULARITY ZFS_AC_KERNEL_CONST_XATTR_HANDLER ZFS_AC_KERNEL_XATTR_HANDLER_NAME ZFS_AC_KERNEL_XATTR_HANDLER_GET ZFS_AC_KERNEL_XATTR_HANDLER_SET ZFS_AC_KERNEL_XATTR_HANDLER_LIST ZFS_AC_KERNEL_INODE_OWNER_OR_CAPABLE ZFS_AC_KERNEL_POSIX_ACL_FROM_XATTR_USERNS ZFS_AC_KERNEL_POSIX_ACL_RELEASE ZFS_AC_KERNEL_SET_CACHED_ACL_USABLE ZFS_AC_KERNEL_POSIX_ACL_CHMOD ZFS_AC_KERNEL_POSIX_ACL_EQUIV_MODE_WANTS_UMODE_T ZFS_AC_KERNEL_POSIX_ACL_VALID_WITH_NS ZFS_AC_KERNEL_INODE_OPERATIONS_PERMISSION ZFS_AC_KERNEL_INODE_OPERATIONS_PERMISSION_WITH_NAMEIDATA ZFS_AC_KERNEL_INODE_OPERATIONS_CHECK_ACL ZFS_AC_KERNEL_INODE_OPERATIONS_CHECK_ACL_WITH_FLAGS ZFS_AC_KERNEL_INODE_OPERATIONS_GET_ACL ZFS_AC_KERNEL_INODE_OPERATIONS_SET_ACL ZFS_AC_KERNEL_INODE_OPERATIONS_GETATTR ZFS_AC_KERNEL_INODE_SET_FLAGS ZFS_AC_KERNEL_INODE_SET_IVERSION ZFS_AC_KERNEL_GET_ACL_HANDLE_CACHE ZFS_AC_KERNEL_SHOW_OPTIONS ZFS_AC_KERNEL_FILE_INODE ZFS_AC_KERNEL_FILE_DENTRY ZFS_AC_KERNEL_FSYNC ZFS_AC_KERNEL_EVICT_INODE ZFS_AC_KERNEL_DIRTY_INODE_WITH_FLAGS ZFS_AC_KERNEL_NR_CACHED_OBJECTS ZFS_AC_KERNEL_FREE_CACHED_OBJECTS ZFS_AC_KERNEL_FALLOCATE ZFS_AC_KERNEL_AIO_FSYNC ZFS_AC_KERNEL_MKDIR_UMODE_T ZFS_AC_KERNEL_LOOKUP_NAMEIDATA ZFS_AC_KERNEL_CREATE_NAMEIDATA ZFS_AC_KERNEL_GET_LINK ZFS_AC_KERNEL_PUT_LINK ZFS_AC_KERNEL_TMPFILE ZFS_AC_KERNEL_TRUNCATE_RANGE ZFS_AC_KERNEL_AUTOMOUNT ZFS_AC_KERNEL_ENCODE_FH_WITH_INODE ZFS_AC_KERNEL_COMMIT_METADATA ZFS_AC_KERNEL_CLEAR_INODE ZFS_AC_KERNEL_SETATTR_PREPARE ZFS_AC_KERNEL_INSERT_INODE_LOCKED ZFS_AC_KERNEL_D_MAKE_ROOT ZFS_AC_KERNEL_D_OBTAIN_ALIAS ZFS_AC_KERNEL_D_PRUNE_ALIASES ZFS_AC_KERNEL_D_SET_D_OP ZFS_AC_KERNEL_D_REVALIDATE_NAMEIDATA ZFS_AC_KERNEL_CONST_DENTRY_OPERATIONS ZFS_AC_KERNEL_TRUNCATE_SETSIZE ZFS_AC_KERNEL_6ARGS_SECURITY_INODE_INIT_SECURITY ZFS_AC_KERNEL_CALLBACK_SECURITY_INODE_INIT_SECURITY ZFS_AC_KERNEL_FST_MOUNT ZFS_AC_KERNEL_SHRINK ZFS_AC_KERNEL_SHRINK_CONTROL_HAS_NID ZFS_AC_KERNEL_SHRINK_CONTROL_STRUCT ZFS_AC_KERNEL_SHRINKER_CALLBACK ZFS_AC_KERNEL_S_INSTANCES_LIST_HEAD ZFS_AC_KERNEL_S_D_OP ZFS_AC_KERNEL_BDI ZFS_AC_KERNEL_SET_NLINK ZFS_AC_KERNEL_ELEVATOR_CHANGE ZFS_AC_KERNEL_5ARG_SGET ZFS_AC_KERNEL_LSEEK_EXECUTE ZFS_AC_KERNEL_VFS_ITERATE ZFS_AC_KERNEL_VFS_RW_ITERATE ZFS_AC_KERNEL_VFS_DIRECT_IO ZFS_AC_KERNEL_GENERIC_WRITE_CHECKS ZFS_AC_KERNEL_KMAP_ATOMIC_ARGS ZFS_AC_KERNEL_FOLLOW_DOWN_ONE ZFS_AC_KERNEL_MAKE_REQUEST_FN ZFS_AC_KERNEL_GENERIC_IO_ACCT_3ARG ZFS_AC_KERNEL_GENERIC_IO_ACCT_4ARG ZFS_AC_KERNEL_FPU ZFS_AC_KERNEL_KUID_HELPERS ZFS_AC_KERNEL_MODULE_PARAM_CALL_CONST ZFS_AC_KERNEL_RENAME_WANTS_FLAGS ZFS_AC_KERNEL_HAVE_GENERIC_SETXATTR ZFS_AC_KERNEL_CURRENT_TIME ZFS_AC_KERNEL_GLOBAL_PAGE_STATE ZFS_AC_KERNEL_ACL_HAS_REFCOUNT ZFS_AC_KERNEL_USERNS_CAPABILITIES ZFS_AC_KERNEL_IN_COMPAT_SYSCALL ZFS_AC_KERNEL_KTIME_GET_COARSE_REAL_TS64 ZFS_AC_KERNEL_TOTALRAM_PAGES_FUNC ZFS_AC_KERNEL_TOTALHIGH_PAGES ZFS_AC_KERNEL_BLK_QUEUE_DISCARD ZFS_AC_KERNEL_BLK_QUEUE_SECURE_ERASE AS_IF([test "$LINUX_OBJ" != "$LINUX"], [ KERNEL_MAKE="$KERNEL_MAKE O=$LINUX_OBJ" ]) AC_SUBST(KERNEL_MAKE) ]) dnl # dnl # Detect name used for Module.symvers file in kernel dnl # AC_DEFUN([ZFS_AC_MODULE_SYMVERS], [ modpost=$LINUX/scripts/Makefile.modpost AC_MSG_CHECKING([kernel file name for module symbols]) AS_IF([test "x$enable_linux_builtin" != xyes -a -f "$modpost"], [ AS_IF([grep -q Modules.symvers $modpost], [ LINUX_SYMBOLS=Modules.symvers ], [ LINUX_SYMBOLS=Module.symvers ]) AS_IF([test ! -f "$LINUX_OBJ/$LINUX_SYMBOLS"], [ AC_MSG_ERROR([ *** Please make sure the kernel devel package for your distribution *** is installed. If you are building with a custom kernel, make sure the *** kernel is configured, built, and the '--with-linux=PATH' configure *** option refers to the location of the kernel source.]) ]) ], [ LINUX_SYMBOLS=NONE ]) AC_MSG_RESULT($LINUX_SYMBOLS) AC_SUBST(LINUX_SYMBOLS) ]) dnl # dnl # Detect the kernel to be built against dnl # AC_DEFUN([ZFS_AC_KERNEL], [ AC_ARG_WITH([linux], AS_HELP_STRING([--with-linux=PATH], [Path to kernel source]), [kernelsrc="$withval"]) AC_ARG_WITH(linux-obj, AS_HELP_STRING([--with-linux-obj=PATH], [Path to kernel build objects]), [kernelbuild="$withval"]) AC_MSG_CHECKING([kernel source directory]) AS_IF([test -z "$kernelsrc"], [ AS_IF([test -e "/lib/modules/$(uname -r)/source"], [ headersdir="/lib/modules/$(uname -r)/source" sourcelink=$(readlink -f "$headersdir") ], [test -e "/lib/modules/$(uname -r)/build"], [ headersdir="/lib/modules/$(uname -r)/build" sourcelink=$(readlink -f "$headersdir") ], [ sourcelink=$(ls -1d /usr/src/kernels/* \ /usr/src/linux-* \ 2>/dev/null | grep -v obj | tail -1) ]) AS_IF([test -n "$sourcelink" && test -e ${sourcelink}], [ kernelsrc=`readlink -f ${sourcelink}` ], [ kernelsrc="[Not found]" ]) ], [ AS_IF([test "$kernelsrc" = "NONE"], [ kernsrcver=NONE ]) withlinux=yes ]) AC_MSG_RESULT([$kernelsrc]) AS_IF([test ! -d "$kernelsrc"], [ AC_MSG_ERROR([ *** Please make sure the kernel devel package for your distribution *** is installed and then try again. If that fails, you can specify the *** location of the kernel source with the '--with-linux=PATH' option.]) ]) AC_MSG_CHECKING([kernel build directory]) AS_IF([test -z "$kernelbuild"], [ AS_IF([test x$withlinux != xyes -a -e "/lib/modules/$(uname -r)/build"], [ kernelbuild=`readlink -f /lib/modules/$(uname -r)/build` ], [test -d ${kernelsrc}-obj/${target_cpu}/${target_cpu}], [ kernelbuild=${kernelsrc}-obj/${target_cpu}/${target_cpu} ], [test -d ${kernelsrc}-obj/${target_cpu}/default], [ kernelbuild=${kernelsrc}-obj/${target_cpu}/default ], [test -d `dirname ${kernelsrc}`/build-${target_cpu}], [ kernelbuild=`dirname ${kernelsrc}`/build-${target_cpu} ], [ kernelbuild=${kernelsrc} ]) ]) AC_MSG_RESULT([$kernelbuild]) AC_MSG_CHECKING([kernel source version]) utsrelease1=$kernelbuild/include/linux/version.h utsrelease2=$kernelbuild/include/linux/utsrelease.h utsrelease3=$kernelbuild/include/generated/utsrelease.h AS_IF([test -r $utsrelease1 && fgrep -q UTS_RELEASE $utsrelease1], [ utsrelease=linux/version.h ], [test -r $utsrelease2 && fgrep -q UTS_RELEASE $utsrelease2], [ utsrelease=linux/utsrelease.h ], [test -r $utsrelease3 && fgrep -q UTS_RELEASE $utsrelease3], [ utsrelease=generated/utsrelease.h ]) AS_IF([test "$utsrelease"], [ kernsrcver=`(echo "#include <$utsrelease>"; echo "kernsrcver=UTS_RELEASE") | ${CPP} -I $kernelbuild/include - | grep "^kernsrcver=" | cut -d \" -f 2` AS_IF([test -z "$kernsrcver"], [ AC_MSG_RESULT([Not found]) AC_MSG_ERROR([*** Cannot determine kernel version.]) ]) ], [ AC_MSG_RESULT([Not found]) if test "x$enable_linux_builtin" != xyes; then AC_MSG_ERROR([*** Cannot find UTS_RELEASE definition.]) else AC_MSG_ERROR([ *** Cannot find UTS_RELEASE definition. *** Please run 'make prepare' inside the kernel source tree.]) fi ]) AC_MSG_RESULT([$kernsrcver]) LINUX=${kernelsrc} LINUX_OBJ=${kernelbuild} LINUX_VERSION=${kernsrcver} AC_SUBST(LINUX) AC_SUBST(LINUX_OBJ) AC_SUBST(LINUX_VERSION) ZFS_AC_MODULE_SYMVERS ]) dnl # dnl # Detect the QAT module to be built against dnl # QAT provides hardware acceleration for data compression: dnl # https://01.org/intel-quickassist-technology dnl # * Download and install QAT driver from the above link dnl # * Start QAT driver in your system: dnl # service qat_service start dnl # * Enable QAT in ZFS, e.g.: dnl # ./configure --with-qat=/QAT1.6 dnl # make dnl # * Set GZIP compression in ZFS dataset: dnl # zfs set compression = gzip dnl # Then the data written to this ZFS pool is compressed dnl # by QAT accelerator automatically, and de-compressed by dnl # QAT when read from the pool. dnl # * Get QAT hardware statistics by: dnl # cat /proc/icp_dh895xcc_dev/qat dnl # * To disable QAT: dnl # insmod zfs.ko zfs_qat_disable=1 dnl # AC_DEFUN([ZFS_AC_QAT], [ AC_ARG_WITH([qat], AS_HELP_STRING([--with-qat=PATH], [Path to qat source]), AS_IF([test "$withval" = "yes"], AC_MSG_ERROR([--with-qat=PATH requires a PATH]), [qatsrc="$withval"])) AC_ARG_WITH([qat-obj], AS_HELP_STRING([--with-qat-obj=PATH], [Path to qat build objects]), [qatbuild="$withval"]) AS_IF([test ! -z "${qatsrc}"], [ AC_MSG_CHECKING([qat source directory]) AC_MSG_RESULT([$qatsrc]) QAT_SRC="${qatsrc}/quickassist" AS_IF([ test ! -e "$QAT_SRC/include/cpa.h"], [ AC_MSG_ERROR([ *** Please make sure the qat driver package is installed *** and specify the location of the qat source with the *** '--with-qat=PATH' option then try again. Failed to *** find cpa.h in: ${QAT_SRC}/include]) ]) ]) AS_IF([test ! -z "${qatsrc}"], [ AC_MSG_CHECKING([qat build directory]) AS_IF([test -z "$qatbuild"], [ qatbuild="${qatsrc}/build" ]) AC_MSG_RESULT([$qatbuild]) QAT_OBJ=${qatbuild} AS_IF([ ! test -e "$QAT_OBJ/icp_qa_al.ko" && ! test -e "$QAT_OBJ/qat_api.ko"], [ AC_MSG_ERROR([ *** Please make sure the qat driver is installed then try again. *** Failed to find icp_qa_al.ko or qat_api.ko in: $QAT_OBJ]) ]) AC_SUBST(QAT_SRC) AC_SUBST(QAT_OBJ) AC_DEFINE(HAVE_QAT, 1, [qat is enabled and existed]) ]) dnl # dnl # Detect the name used for the QAT Module.symvers file. dnl # AS_IF([test ! -z "${qatsrc}"], [ AC_MSG_CHECKING([qat file for module symbols]) QAT_SYMBOLS=$QAT_SRC/lookaside/access_layer/src/Module.symvers AS_IF([test -r $QAT_SYMBOLS], [ AC_MSG_RESULT([$QAT_SYMBOLS]) AC_SUBST(QAT_SYMBOLS) ],[ AC_MSG_ERROR([ *** Please make sure the qat driver is installed then try again. *** Failed to find Module.symvers in: $QAT_SYMBOLS]) ]) ]) ]) ]) dnl # dnl # Basic toolchain sanity check. dnl # AC_DEFUN([ZFS_AC_TEST_MODULE], [ AC_MSG_CHECKING([whether modules can be built]) ZFS_LINUX_TRY_COMPILE([],[],[ AC_MSG_RESULT([yes]) ],[ AC_MSG_RESULT([no]) if test "x$enable_linux_builtin" != xyes; then AC_MSG_ERROR([*** Unable to build an empty module.]) else AC_MSG_ERROR([ *** Unable to build an empty module. *** Please run 'make scripts' inside the kernel source tree.]) fi ]) ]) dnl # dnl # Certain kernel build options are not supported. These must be dnl # detected at configure time and cause a build failure. Otherwise dnl # modules may be successfully built that behave incorrectly. dnl # AC_DEFUN([ZFS_AC_KERNEL_CONFIG], [ AS_IF([test "x$cross_compiling" != xyes], [ AC_RUN_IFELSE([ AC_LANG_PROGRAM([ #include "$LINUX/include/linux/license.h" ], [ return !license_is_gpl_compatible("$ZFS_META_LICENSE"); ]) ], [ AC_DEFINE([ZFS_IS_GPL_COMPATIBLE], [1], [Define to 1 if GPL-only symbols can be used]) ], [ ]) ]) ZFS_AC_KERNEL_CONFIG_THREAD_SIZE ZFS_AC_KERNEL_CONFIG_DEBUG_LOCK_ALLOC ZFS_AC_KERNEL_CONFIG_TRIM_UNUSED_KSYMS ZFS_AC_KERNEL_CONFIG_ZLIB_INFLATE ZFS_AC_KERNEL_CONFIG_ZLIB_DEFLATE ]) dnl # dnl # Check configured THREAD_SIZE dnl # dnl # The stack size will vary by architecture, but as of Linux 3.15 on x86_64 dnl # the default thread stack size was increased to 16K from 8K. Therefore, dnl # on newer kernels and some architectures stack usage optimizations can be dnl # conditionally applied to improve performance without negatively impacting dnl # stability. dnl # AC_DEFUN([ZFS_AC_KERNEL_CONFIG_THREAD_SIZE], [ AC_MSG_CHECKING([whether kernel was built with 16K or larger stacks]) ZFS_LINUX_TRY_COMPILE([ #include ],[ #if (THREAD_SIZE < 16384) #error "THREAD_SIZE is less than 16K" #endif ],[ AC_MSG_RESULT([yes]) AC_DEFINE(HAVE_LARGE_STACKS, 1, [kernel has large stacks]) ],[ AC_MSG_RESULT([no]) ]) ]) dnl # dnl # Check CONFIG_DEBUG_LOCK_ALLOC dnl # dnl # This is typically only set for debug kernels because it comes with dnl # a performance penalty. However, when it is set it maps the non-GPL dnl # symbol mutex_lock() to the GPL-only mutex_lock_nested() symbol. dnl # This will cause a failure at link time which we'd rather know about dnl # at compile time. dnl # dnl # Since we plan to pursue making mutex_lock_nested() a non-GPL symbol dnl # with the upstream community we add a check to detect this case. dnl # AC_DEFUN([ZFS_AC_KERNEL_CONFIG_DEBUG_LOCK_ALLOC], [ ZFS_LINUX_CONFIG([DEBUG_LOCK_ALLOC], [ AC_MSG_CHECKING([whether mutex_lock() is GPL-only]) tmp_flags="$EXTRA_KCFLAGS" ZFS_LINUX_TRY_COMPILE([ #include #include MODULE_LICENSE("$ZFS_META_LICENSE"); ],[ struct mutex lock; mutex_init(&lock); mutex_lock(&lock); mutex_unlock(&lock); ],[ AC_MSG_RESULT(no) ],[ AC_MSG_RESULT(yes) AC_MSG_ERROR([ *** Kernel built with CONFIG_DEBUG_LOCK_ALLOC which is incompatible *** with the CDDL license and will prevent the module linking stage *** from succeeding. You must rebuild your kernel without this *** option enabled.]) ]) EXTRA_KCFLAGS="$tmp_flags" ], []) ]) dnl # dnl # Check CONFIG_TRIM_UNUSED_KSYMS dnl # dnl # Verify the kernel has CONFIG_TRIM_UNUSED_KSYMS disabled. dnl # AC_DEFUN([ZFS_AC_KERNEL_CONFIG_TRIM_UNUSED_KSYMS], [ AC_MSG_CHECKING([whether CONFIG_TRIM_UNUSED_KSYM is disabled]) ZFS_LINUX_TRY_COMPILE([ #if defined(CONFIG_TRIM_UNUSED_KSYMS) #error CONFIG_TRIM_UNUSED_KSYMS not defined #endif ],[ ],[ AC_MSG_RESULT([yes]) ],[ AC_MSG_RESULT([no]) AC_MSG_ERROR([ *** This kernel has unused symbols trimming enabled, please disable. *** Rebuild the kernel with CONFIG_TRIM_UNUSED_KSYMS=n set.]) ]) ]) dnl # dnl # ZFS_LINUX_CONFTEST_H dnl # AC_DEFUN([ZFS_LINUX_CONFTEST_H], [ cat - <<_ACEOF >conftest.h $1 _ACEOF ]) dnl # dnl # ZFS_LINUX_CONFTEST_C dnl # AC_DEFUN([ZFS_LINUX_CONFTEST_C], [ cat confdefs.h - <<_ACEOF >conftest.c $1 _ACEOF ]) dnl # dnl # ZFS_LANG_PROGRAM(C)([PROLOGUE], [BODY]) dnl # m4_define([ZFS_LANG_PROGRAM], [ $1 int main (void) { dnl Do *not* indent the following line: there may be CPP directives. dnl Don't move the `;' right after for the same reason. $2 ; return 0; } ]) dnl # dnl # ZFS_LINUX_COMPILE_IFELSE / like AC_COMPILE_IFELSE dnl # AC_DEFUN([ZFS_LINUX_COMPILE_IFELSE], [ m4_ifvaln([$1], [ZFS_LINUX_CONFTEST_C([$1])]) m4_ifvaln([$6], [ZFS_LINUX_CONFTEST_H([$6])], [ZFS_LINUX_CONFTEST_H([])]) rm -Rf build && mkdir -p build && touch build/conftest.mod.c echo "obj-m := conftest.o" >build/Makefile modpost_flag='' test "x$enable_linux_builtin" = xyes && modpost_flag='modpost=true' # fake modpost stage AS_IF( [AC_TRY_COMMAND(cp conftest.c conftest.h build && make [$2] -C $LINUX_OBJ EXTRA_CFLAGS="-Werror $FRAME_LARGER_THAN $EXTRA_KCFLAGS" $ARCH_UM M=$PWD/build $modpost_flag) >/dev/null && AC_TRY_COMMAND([$3])], [$4], [_AC_MSG_LOG_CONFTEST m4_ifvaln([$5],[$5])] ) rm -Rf build ]) dnl # dnl # ZFS_LINUX_TRY_COMPILE like AC_TRY_COMPILE dnl # AC_DEFUN([ZFS_LINUX_TRY_COMPILE], [ZFS_LINUX_COMPILE_IFELSE( [AC_LANG_SOURCE([ZFS_LANG_PROGRAM([[$1]], [[$2]])])], [modules], [test -s build/conftest.o], [$3], [$4]) ]) dnl # dnl # ZFS_LINUX_CONFIG dnl # AC_DEFUN([ZFS_LINUX_CONFIG], [AC_MSG_CHECKING([whether kernel was built with CONFIG_$1]) ZFS_LINUX_TRY_COMPILE([ #include ],[ #ifndef CONFIG_$1 #error CONFIG_$1 not #defined #endif ],[ AC_MSG_RESULT([yes]) $2 ],[ AC_MSG_RESULT([no]) $3 ]) ]) dnl # dnl # ZFS_CHECK_SYMBOL_EXPORT dnl # check symbol exported or not dnl # AC_DEFUN([ZFS_CHECK_SYMBOL_EXPORT], [ grep -q -E '[[[:space:]]]$1[[[:space:]]]' \ $LINUX_OBJ/$LINUX_SYMBOLS 2>/dev/null rc=$? if test $rc -ne 0; then export=0 for file in $2; do grep -q -E "EXPORT_SYMBOL.*($1)" \ "$LINUX/$file" 2>/dev/null rc=$? if test $rc -eq 0; then export=1 break; fi done if test $export -eq 0; then : $4 else : $3 fi else : $3 fi ]) dnl # dnl # ZFS_LINUX_TRY_COMPILE_SYMBOL dnl # like ZFS_LINUX_TRY_COMPILE, except ZFS_CHECK_SYMBOL_EXPORT dnl # is called if not compiling for builtin dnl # AC_DEFUN([ZFS_LINUX_TRY_COMPILE_SYMBOL], [ ZFS_LINUX_TRY_COMPILE([$1], [$2], [rc=0], [rc=1]) if test $rc -ne 0; then : $6 else if test "x$enable_linux_builtin" != xyes; then ZFS_CHECK_SYMBOL_EXPORT([$3], [$4], [rc=0], [rc=1]) fi if test $rc -ne 0; then : $6 else : $5 fi fi ]) dnl # dnl # ZFS_LINUX_TRY_COMPILE_HEADER dnl # like ZFS_LINUX_TRY_COMPILE, except the contents conftest.h are dnl # provided via the fifth parameter dnl # AC_DEFUN([ZFS_LINUX_TRY_COMPILE_HEADER], [ZFS_LINUX_COMPILE_IFELSE( [AC_LANG_SOURCE([ZFS_LANG_PROGRAM([[$1]], [[$2]])])], [modules], [test -s build/conftest.o], [$3], [$4], [$5]) ]) diff --git a/include/spl/sys/timer.h b/include/spl/sys/timer.h index a6b134570cd8..31d89d3b97d6 100644 --- a/include/spl/sys/timer.h +++ b/include/spl/sys/timer.h @@ -1,75 +1,100 @@ /* * Copyright (C) 2007-2010 Lawrence Livermore National Security, LLC. * Copyright (C) 2007 The Regents of the University of California. * Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER). * Written by Brian Behlendorf . * UCRL-CODE-235197 * * This file is part of the SPL, Solaris Porting Layer. * For details, see . * * The SPL is free software; you can redistribute it and/or modify it * under the terms of the GNU General Public License as published by the * Free Software Foundation; either version 2 of the License, or (at your * option) any later version. * * The SPL is distributed in the hope that it will be useful, but WITHOUT * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License * for more details. * * You should have received a copy of the GNU General Public License along * with the SPL. If not, see . */ #ifndef _SPL_TIMER_H #define _SPL_TIMER_H #include #include #include #include #include #define lbolt ((clock_t)jiffies) #define lbolt64 ((int64_t)get_jiffies_64()) #define ddi_get_lbolt() ((clock_t)jiffies) #define ddi_get_lbolt64() ((int64_t)get_jiffies_64()) #define ddi_time_before(a, b) (typecheck(clock_t, a) && \ typecheck(clock_t, b) && \ ((a) - (b) < 0)) #define ddi_time_after(a, b) ddi_time_before(b, a) #define ddi_time_before_eq(a, b) (!ddi_time_after(a, b)) #define ddi_time_after_eq(a, b) ddi_time_before_eq(b, a) #define ddi_time_before64(a, b) (typecheck(int64_t, a) && \ typecheck(int64_t, b) && \ ((a) - (b) < 0)) #define ddi_time_after64(a, b) ddi_time_before64(b, a) #define ddi_time_before_eq64(a, b) (!ddi_time_after64(a, b)) #define ddi_time_after_eq64(a, b) ddi_time_before_eq64(b, a) #define delay(ticks) schedule_timeout_uninterruptible(ticks) /* usleep_range() introduced in 2.6.36 */ #ifndef HAVE_USLEEP_RANGE static inline void usleep_range(unsigned long min, unsigned long max) { unsigned int min_ms = min / USEC_PER_MSEC; if (min >= MAX_UDELAY_MS) msleep(min_ms); else udelay(min); } #endif /* HAVE_USLEEP_RANGE */ #define SEC_TO_TICK(sec) ((sec) * HZ) #define MSEC_TO_TICK(ms) msecs_to_jiffies(ms) #define USEC_TO_TICK(us) usecs_to_jiffies(us) #define NSEC_TO_TICK(ns) usecs_to_jiffies(ns / NSEC_PER_USEC) +#ifndef from_timer +#define from_timer(var, timer, timer_field) \ + container_of(timer, typeof(*var), timer_field) +#endif + +#ifdef HAVE_KERNEL_TIMER_FUNCTION_TIMER_LIST +typedef struct timer_list *spl_timer_list_t; +#else +typedef unsigned long spl_timer_list_t; +#endif + +#ifndef HAVE_KERNEL_TIMER_SETUP + +static inline void +timer_setup(struct timer_list *timer, void (*func)(spl_timer_list_t), u32 fl) +{ +#ifdef HAVE_KERNEL_TIMER_LIST_FLAGS + (timer)->flags = fl; +#endif + init_timer(timer); + setup_timer(timer, func, (spl_timer_list_t)(timer)); +} + +#endif /* HAVE_KERNEL_TIMER_SETUP */ + #endif /* _SPL_TIMER_H */ diff --git a/module/spl/spl-condvar.c b/module/spl/spl-condvar.c index 1e6e38b7874b..a7a9d1db9a98 100644 --- a/module/spl/spl-condvar.c +++ b/module/spl/spl-condvar.c @@ -1,411 +1,424 @@ /* * Copyright (C) 2007-2010 Lawrence Livermore National Security, LLC. * Copyright (C) 2007 The Regents of the University of California. * Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER). * Written by Brian Behlendorf . * UCRL-CODE-235197 * * This file is part of the SPL, Solaris Porting Layer. * For details, see . * * The SPL is free software; you can redistribute it and/or modify it * under the terms of the GNU General Public License as published by the * Free Software Foundation; either version 2 of the License, or (at your * option) any later version. * * The SPL is distributed in the hope that it will be useful, but WITHOUT * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License * for more details. * * You should have received a copy of the GNU General Public License along * with the SPL. If not, see . * * Solaris Porting Layer (SPL) Credential Implementation. */ #include #include #include #include void __cv_init(kcondvar_t *cvp, char *name, kcv_type_t type, void *arg) { ASSERT(cvp); ASSERT(name == NULL); ASSERT(type == CV_DEFAULT); ASSERT(arg == NULL); cvp->cv_magic = CV_MAGIC; init_waitqueue_head(&cvp->cv_event); init_waitqueue_head(&cvp->cv_destroy); atomic_set(&cvp->cv_waiters, 0); atomic_set(&cvp->cv_refs, 1); cvp->cv_mutex = NULL; } EXPORT_SYMBOL(__cv_init); static int cv_destroy_wakeup(kcondvar_t *cvp) { if (!atomic_read(&cvp->cv_waiters) && !atomic_read(&cvp->cv_refs)) { ASSERT(cvp->cv_mutex == NULL); ASSERT(!waitqueue_active(&cvp->cv_event)); return (1); } return (0); } void __cv_destroy(kcondvar_t *cvp) { ASSERT(cvp); ASSERT(cvp->cv_magic == CV_MAGIC); cvp->cv_magic = CV_DESTROY; atomic_dec(&cvp->cv_refs); /* Block until all waiters are woken and references dropped. */ while (cv_destroy_wakeup(cvp) == 0) wait_event_timeout(cvp->cv_destroy, cv_destroy_wakeup(cvp), 1); ASSERT3P(cvp->cv_mutex, ==, NULL); ASSERT3S(atomic_read(&cvp->cv_refs), ==, 0); ASSERT3S(atomic_read(&cvp->cv_waiters), ==, 0); ASSERT3S(waitqueue_active(&cvp->cv_event), ==, 0); } EXPORT_SYMBOL(__cv_destroy); static void cv_wait_common(kcondvar_t *cvp, kmutex_t *mp, int state, int io) { DEFINE_WAIT(wait); kmutex_t *m; ASSERT(cvp); ASSERT(mp); ASSERT(cvp->cv_magic == CV_MAGIC); ASSERT(mutex_owned(mp)); atomic_inc(&cvp->cv_refs); m = READ_ONCE(cvp->cv_mutex); if (!m) m = xchg(&cvp->cv_mutex, mp); /* Ensure the same mutex is used by all callers */ ASSERT(m == NULL || m == mp); prepare_to_wait_exclusive(&cvp->cv_event, &wait, state); atomic_inc(&cvp->cv_waiters); /* * Mutex should be dropped after prepare_to_wait() this * ensures we're linked in to the waiters list and avoids the * race where 'cvp->cv_waiters > 0' but the list is empty. */ mutex_exit(mp); if (io) io_schedule(); else schedule(); /* No more waiters a different mutex could be used */ if (atomic_dec_and_test(&cvp->cv_waiters)) { /* * This is set without any lock, so it's racy. But this is * just for debug anyway, so make it best-effort */ cvp->cv_mutex = NULL; wake_up(&cvp->cv_destroy); } finish_wait(&cvp->cv_event, &wait); atomic_dec(&cvp->cv_refs); /* * Hold mutex after we release the cvp, otherwise we could dead lock * with a thread holding the mutex and call cv_destroy. */ mutex_enter(mp); } void __cv_wait(kcondvar_t *cvp, kmutex_t *mp) { cv_wait_common(cvp, mp, TASK_UNINTERRUPTIBLE, 0); } EXPORT_SYMBOL(__cv_wait); void __cv_wait_io(kcondvar_t *cvp, kmutex_t *mp) { cv_wait_common(cvp, mp, TASK_UNINTERRUPTIBLE, 1); } EXPORT_SYMBOL(__cv_wait_io); void __cv_wait_sig(kcondvar_t *cvp, kmutex_t *mp) { cv_wait_common(cvp, mp, TASK_INTERRUPTIBLE, 0); } EXPORT_SYMBOL(__cv_wait_sig); #if defined(HAVE_IO_SCHEDULE_TIMEOUT) #define spl_io_schedule_timeout(t) io_schedule_timeout(t) #else + +struct spl_task_timer { + struct timer_list timer; + struct task_struct *task; +}; + static void -__cv_wakeup(unsigned long data) +__cv_wakeup(spl_timer_list_t t) { - wake_up_process((struct task_struct *)data); + struct timer_list *tmr = (struct timer_list *)t; + struct spl_task_timer *task_timer = from_timer(task_timer, tmr, timer); + + wake_up_process(task_timer->task); } static long spl_io_schedule_timeout(long time_left) { long expire_time = jiffies + time_left; - struct timer_list timer; + struct spl_task_timer task_timer; + struct timer_list *timer = &task_timer.timer; + + task_timer.task = current; - init_timer(&timer); - setup_timer(&timer, __cv_wakeup, (unsigned long)current); - timer.expires = expire_time; - add_timer(&timer); + timer_setup(timer, __cv_wakeup, 0); + + timer->expires = expire_time; + add_timer(timer); io_schedule(); - del_timer_sync(&timer); + del_timer_sync(timer); + time_left = expire_time - jiffies; return (time_left < 0 ? 0 : time_left); } #endif /* * 'expire_time' argument is an absolute wall clock time in jiffies. * Return value is time left (expire_time - now) or -1 if timeout occurred. */ static clock_t __cv_timedwait_common(kcondvar_t *cvp, kmutex_t *mp, clock_t expire_time, int state, int io) { DEFINE_WAIT(wait); kmutex_t *m; clock_t time_left; ASSERT(cvp); ASSERT(mp); ASSERT(cvp->cv_magic == CV_MAGIC); ASSERT(mutex_owned(mp)); /* XXX - Does not handle jiffie wrap properly */ time_left = expire_time - jiffies; if (time_left <= 0) return (-1); atomic_inc(&cvp->cv_refs); m = READ_ONCE(cvp->cv_mutex); if (!m) m = xchg(&cvp->cv_mutex, mp); /* Ensure the same mutex is used by all callers */ ASSERT(m == NULL || m == mp); prepare_to_wait_exclusive(&cvp->cv_event, &wait, state); atomic_inc(&cvp->cv_waiters); /* * Mutex should be dropped after prepare_to_wait() this * ensures we're linked in to the waiters list and avoids the * race where 'cvp->cv_waiters > 0' but the list is empty. */ mutex_exit(mp); if (io) time_left = spl_io_schedule_timeout(time_left); else time_left = schedule_timeout(time_left); /* No more waiters a different mutex could be used */ if (atomic_dec_and_test(&cvp->cv_waiters)) { /* * This is set without any lock, so it's racy. But this is * just for debug anyway, so make it best-effort */ cvp->cv_mutex = NULL; wake_up(&cvp->cv_destroy); } finish_wait(&cvp->cv_event, &wait); atomic_dec(&cvp->cv_refs); /* * Hold mutex after we release the cvp, otherwise we could dead lock * with a thread holding the mutex and call cv_destroy. */ mutex_enter(mp); return (time_left > 0 ? time_left : -1); } clock_t __cv_timedwait(kcondvar_t *cvp, kmutex_t *mp, clock_t exp_time) { return (__cv_timedwait_common(cvp, mp, exp_time, TASK_UNINTERRUPTIBLE, 0)); } EXPORT_SYMBOL(__cv_timedwait); clock_t __cv_timedwait_io(kcondvar_t *cvp, kmutex_t *mp, clock_t exp_time) { return (__cv_timedwait_common(cvp, mp, exp_time, TASK_UNINTERRUPTIBLE, 1)); } EXPORT_SYMBOL(__cv_timedwait_io); clock_t __cv_timedwait_sig(kcondvar_t *cvp, kmutex_t *mp, clock_t exp_time) { return (__cv_timedwait_common(cvp, mp, exp_time, TASK_INTERRUPTIBLE, 0)); } EXPORT_SYMBOL(__cv_timedwait_sig); /* * 'expire_time' argument is an absolute clock time in nanoseconds. * Return value is time left (expire_time - now) or -1 if timeout occurred. */ static clock_t __cv_timedwait_hires(kcondvar_t *cvp, kmutex_t *mp, hrtime_t expire_time, int state) { DEFINE_WAIT(wait); kmutex_t *m; hrtime_t time_left; ktime_t ktime_left; ASSERT(cvp); ASSERT(mp); ASSERT(cvp->cv_magic == CV_MAGIC); ASSERT(mutex_owned(mp)); time_left = expire_time - gethrtime(); if (time_left <= 0) return (-1); atomic_inc(&cvp->cv_refs); m = READ_ONCE(cvp->cv_mutex); if (!m) m = xchg(&cvp->cv_mutex, mp); /* Ensure the same mutex is used by all callers */ ASSERT(m == NULL || m == mp); prepare_to_wait_exclusive(&cvp->cv_event, &wait, state); atomic_inc(&cvp->cv_waiters); /* * Mutex should be dropped after prepare_to_wait() this * ensures we're linked in to the waiters list and avoids the * race where 'cvp->cv_waiters > 0' but the list is empty. */ mutex_exit(mp); /* * Allow a 100 us range to give kernel an opportunity to coalesce * interrupts */ ktime_left = ktime_set(0, time_left); schedule_hrtimeout_range(&ktime_left, 100 * NSEC_PER_USEC, HRTIMER_MODE_REL); /* No more waiters a different mutex could be used */ if (atomic_dec_and_test(&cvp->cv_waiters)) { /* * This is set without any lock, so it's racy. But this is * just for debug anyway, so make it best-effort */ cvp->cv_mutex = NULL; wake_up(&cvp->cv_destroy); } finish_wait(&cvp->cv_event, &wait); atomic_dec(&cvp->cv_refs); mutex_enter(mp); time_left = expire_time - gethrtime(); return (time_left > 0 ? NSEC_TO_TICK(time_left) : -1); } /* * Compatibility wrapper for the cv_timedwait_hires() Illumos interface. */ static clock_t cv_timedwait_hires_common(kcondvar_t *cvp, kmutex_t *mp, hrtime_t tim, hrtime_t res, int flag, int state) { if (res > 1) { /* * Align expiration to the specified resolution. */ if (flag & CALLOUT_FLAG_ROUNDUP) tim += res - 1; tim = (tim / res) * res; } if (!(flag & CALLOUT_FLAG_ABSOLUTE)) tim += gethrtime(); return (__cv_timedwait_hires(cvp, mp, tim, state)); } clock_t cv_timedwait_hires(kcondvar_t *cvp, kmutex_t *mp, hrtime_t tim, hrtime_t res, int flag) { return (cv_timedwait_hires_common(cvp, mp, tim, res, flag, TASK_UNINTERRUPTIBLE)); } EXPORT_SYMBOL(cv_timedwait_hires); clock_t cv_timedwait_sig_hires(kcondvar_t *cvp, kmutex_t *mp, hrtime_t tim, hrtime_t res, int flag) { return (cv_timedwait_hires_common(cvp, mp, tim, res, flag, TASK_INTERRUPTIBLE)); } EXPORT_SYMBOL(cv_timedwait_sig_hires); void __cv_signal(kcondvar_t *cvp) { ASSERT(cvp); ASSERT(cvp->cv_magic == CV_MAGIC); atomic_inc(&cvp->cv_refs); /* * All waiters are added with WQ_FLAG_EXCLUSIVE so only one * waiter will be set runable with each call to wake_up(). * Additionally wake_up() holds a spin_lock assoicated with * the wait queue to ensure we don't race waking up processes. */ if (atomic_read(&cvp->cv_waiters) > 0) wake_up(&cvp->cv_event); atomic_dec(&cvp->cv_refs); } EXPORT_SYMBOL(__cv_signal); void __cv_broadcast(kcondvar_t *cvp) { ASSERT(cvp); ASSERT(cvp->cv_magic == CV_MAGIC); atomic_inc(&cvp->cv_refs); /* * Wake_up_all() will wake up all waiters even those which * have the WQ_FLAG_EXCLUSIVE flag set. */ if (atomic_read(&cvp->cv_waiters) > 0) wake_up_all(&cvp->cv_event); atomic_dec(&cvp->cv_refs); } EXPORT_SYMBOL(__cv_broadcast); diff --git a/module/spl/spl-taskq.c b/module/spl/spl-taskq.c index 7684257be7ad..a39f94e4cc20 100644 --- a/module/spl/spl-taskq.c +++ b/module/spl/spl-taskq.c @@ -1,1306 +1,1290 @@ /* * Copyright (C) 2007-2010 Lawrence Livermore National Security, LLC. * Copyright (C) 2007 The Regents of the University of California. * Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER). * Written by Brian Behlendorf . * UCRL-CODE-235197 * * This file is part of the SPL, Solaris Porting Layer. * For details, see . * * The SPL is free software; you can redistribute it and/or modify it * under the terms of the GNU General Public License as published by the * Free Software Foundation; either version 2 of the License, or (at your * option) any later version. * * The SPL is distributed in the hope that it will be useful, but WITHOUT * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License * for more details. * * You should have received a copy of the GNU General Public License along * with the SPL. If not, see . * * Solaris Porting Layer (SPL) Task Queue Implementation. */ +#include #include #include #include int spl_taskq_thread_bind = 0; module_param(spl_taskq_thread_bind, int, 0644); MODULE_PARM_DESC(spl_taskq_thread_bind, "Bind taskq thread to CPU by default"); int spl_taskq_thread_dynamic = 1; module_param(spl_taskq_thread_dynamic, int, 0644); MODULE_PARM_DESC(spl_taskq_thread_dynamic, "Allow dynamic taskq threads"); int spl_taskq_thread_priority = 1; module_param(spl_taskq_thread_priority, int, 0644); MODULE_PARM_DESC(spl_taskq_thread_priority, "Allow non-default priority for taskq threads"); int spl_taskq_thread_sequential = 4; module_param(spl_taskq_thread_sequential, int, 0644); MODULE_PARM_DESC(spl_taskq_thread_sequential, "Create new taskq threads after N sequential tasks"); /* Global system-wide dynamic task queue available for all consumers */ taskq_t *system_taskq; EXPORT_SYMBOL(system_taskq); /* Global dynamic task queue for long delay */ taskq_t *system_delay_taskq; EXPORT_SYMBOL(system_delay_taskq); /* Private dedicated taskq for creating new taskq threads on demand. */ static taskq_t *dynamic_taskq; static taskq_thread_t *taskq_thread_create(taskq_t *); /* List of all taskqs */ LIST_HEAD(tq_list); struct rw_semaphore tq_list_sem; static uint_t taskq_tsd; static int task_km_flags(uint_t flags) { if (flags & TQ_NOSLEEP) return (KM_NOSLEEP); if (flags & TQ_PUSHPAGE) return (KM_PUSHPAGE); return (KM_SLEEP); } /* * taskq_find_by_name - Find the largest instance number of a named taskq. */ static int taskq_find_by_name(const char *name) { struct list_head *tql; taskq_t *tq; list_for_each_prev(tql, &tq_list) { tq = list_entry(tql, taskq_t, tq_taskqs); if (strcmp(name, tq->tq_name) == 0) return (tq->tq_instance); } return (-1); } /* * NOTE: Must be called with tq->tq_lock held, returns a list_t which * is not attached to the free, work, or pending taskq lists. */ static taskq_ent_t * task_alloc(taskq_t *tq, uint_t flags, unsigned long *irqflags) { taskq_ent_t *t; int count = 0; ASSERT(tq); retry: /* Acquire taskq_ent_t's from free list if available */ if (!list_empty(&tq->tq_free_list) && !(flags & TQ_NEW)) { t = list_entry(tq->tq_free_list.next, taskq_ent_t, tqent_list); ASSERT(!(t->tqent_flags & TQENT_FLAG_PREALLOC)); ASSERT(!(t->tqent_flags & TQENT_FLAG_CANCEL)); ASSERT(!timer_pending(&t->tqent_timer)); list_del_init(&t->tqent_list); return (t); } /* Free list is empty and memory allocations are prohibited */ if (flags & TQ_NOALLOC) return (NULL); /* Hit maximum taskq_ent_t pool size */ if (tq->tq_nalloc >= tq->tq_maxalloc) { if (flags & TQ_NOSLEEP) return (NULL); /* * Sleep periodically polling the free list for an available * taskq_ent_t. Dispatching with TQ_SLEEP should always succeed * but we cannot block forever waiting for an taskq_ent_t to * show up in the free list, otherwise a deadlock can happen. * * Therefore, we need to allocate a new task even if the number * of allocated tasks is above tq->tq_maxalloc, but we still * end up delaying the task allocation by one second, thereby * throttling the task dispatch rate. */ spin_unlock_irqrestore(&tq->tq_lock, *irqflags); schedule_timeout(HZ / 100); spin_lock_irqsave_nested(&tq->tq_lock, *irqflags, tq->tq_lock_class); if (count < 100) { count++; goto retry; } } spin_unlock_irqrestore(&tq->tq_lock, *irqflags); t = kmem_alloc(sizeof (taskq_ent_t), task_km_flags(flags)); spin_lock_irqsave_nested(&tq->tq_lock, *irqflags, tq->tq_lock_class); if (t) { taskq_init_ent(t); tq->tq_nalloc++; } return (t); } /* * NOTE: Must be called with tq->tq_lock held, expects the taskq_ent_t * to already be removed from the free, work, or pending taskq lists. */ static void task_free(taskq_t *tq, taskq_ent_t *t) { ASSERT(tq); ASSERT(t); ASSERT(list_empty(&t->tqent_list)); ASSERT(!timer_pending(&t->tqent_timer)); kmem_free(t, sizeof (taskq_ent_t)); tq->tq_nalloc--; } /* * NOTE: Must be called with tq->tq_lock held, either destroys the * taskq_ent_t if too many exist or moves it to the free list for later use. */ static void task_done(taskq_t *tq, taskq_ent_t *t) { ASSERT(tq); ASSERT(t); /* Wake tasks blocked in taskq_wait_id() */ wake_up_all(&t->tqent_waitq); list_del_init(&t->tqent_list); if (tq->tq_nalloc <= tq->tq_minalloc) { t->tqent_id = TASKQID_INVALID; t->tqent_func = NULL; t->tqent_arg = NULL; t->tqent_flags = 0; list_add_tail(&t->tqent_list, &tq->tq_free_list); } else { task_free(tq, t); } } /* * When a delayed task timer expires remove it from the delay list and * add it to the priority list in order for immediate processing. */ static void task_expire_impl(taskq_ent_t *t) { taskq_ent_t *w; taskq_t *tq = t->tqent_taskq; struct list_head *l; unsigned long flags; spin_lock_irqsave_nested(&tq->tq_lock, flags, tq->tq_lock_class); if (t->tqent_flags & TQENT_FLAG_CANCEL) { ASSERT(list_empty(&t->tqent_list)); spin_unlock_irqrestore(&tq->tq_lock, flags); return; } t->tqent_birth = jiffies; /* * The priority list must be maintained in strict task id order * from lowest to highest for lowest_id to be easily calculable. */ list_del(&t->tqent_list); list_for_each_prev(l, &tq->tq_prio_list) { w = list_entry(l, taskq_ent_t, tqent_list); if (w->tqent_id < t->tqent_id) { list_add(&t->tqent_list, l); break; } } if (l == &tq->tq_prio_list) list_add(&t->tqent_list, &tq->tq_prio_list); spin_unlock_irqrestore(&tq->tq_lock, flags); wake_up(&tq->tq_work_waitq); } -#ifdef HAVE_KERNEL_TIMER_FUNCTION_TIMER_LIST static void -task_expire(struct timer_list *tl) +task_expire(spl_timer_list_t tl) { - taskq_ent_t *t = from_timer(t, tl, tqent_timer); + struct timer_list *tmr = (struct timer_list *)tl; + taskq_ent_t *t = from_timer(t, tmr, tqent_timer); task_expire_impl(t); } -#else -static void -task_expire(unsigned long data) -{ - task_expire_impl((taskq_ent_t *)data); -} -#endif /* * Returns the lowest incomplete taskqid_t. The taskqid_t may * be queued on the pending list, on the priority list, on the * delay list, or on the work list currently being handled, but * it is not 100% complete yet. */ static taskqid_t taskq_lowest_id(taskq_t *tq) { taskqid_t lowest_id = tq->tq_next_id; taskq_ent_t *t; taskq_thread_t *tqt; ASSERT(tq); if (!list_empty(&tq->tq_pend_list)) { t = list_entry(tq->tq_pend_list.next, taskq_ent_t, tqent_list); lowest_id = MIN(lowest_id, t->tqent_id); } if (!list_empty(&tq->tq_prio_list)) { t = list_entry(tq->tq_prio_list.next, taskq_ent_t, tqent_list); lowest_id = MIN(lowest_id, t->tqent_id); } if (!list_empty(&tq->tq_delay_list)) { t = list_entry(tq->tq_delay_list.next, taskq_ent_t, tqent_list); lowest_id = MIN(lowest_id, t->tqent_id); } if (!list_empty(&tq->tq_active_list)) { tqt = list_entry(tq->tq_active_list.next, taskq_thread_t, tqt_active_list); ASSERT(tqt->tqt_id != TASKQID_INVALID); lowest_id = MIN(lowest_id, tqt->tqt_id); } return (lowest_id); } /* * Insert a task into a list keeping the list sorted by increasing taskqid. */ static void taskq_insert_in_order(taskq_t *tq, taskq_thread_t *tqt) { taskq_thread_t *w; struct list_head *l; ASSERT(tq); ASSERT(tqt); list_for_each_prev(l, &tq->tq_active_list) { w = list_entry(l, taskq_thread_t, tqt_active_list); if (w->tqt_id < tqt->tqt_id) { list_add(&tqt->tqt_active_list, l); break; } } if (l == &tq->tq_active_list) list_add(&tqt->tqt_active_list, &tq->tq_active_list); } /* * Find and return a task from the given list if it exists. The list * must be in lowest to highest task id order. */ static taskq_ent_t * taskq_find_list(taskq_t *tq, struct list_head *lh, taskqid_t id) { struct list_head *l; taskq_ent_t *t; list_for_each(l, lh) { t = list_entry(l, taskq_ent_t, tqent_list); if (t->tqent_id == id) return (t); if (t->tqent_id > id) break; } return (NULL); } /* * Find an already dispatched task given the task id regardless of what * state it is in. If a task is still pending it will be returned. * If a task is executing, then -EBUSY will be returned instead. * If the task has already been run then NULL is returned. */ static taskq_ent_t * taskq_find(taskq_t *tq, taskqid_t id) { taskq_thread_t *tqt; struct list_head *l; taskq_ent_t *t; t = taskq_find_list(tq, &tq->tq_delay_list, id); if (t) return (t); t = taskq_find_list(tq, &tq->tq_prio_list, id); if (t) return (t); t = taskq_find_list(tq, &tq->tq_pend_list, id); if (t) return (t); list_for_each(l, &tq->tq_active_list) { tqt = list_entry(l, taskq_thread_t, tqt_active_list); if (tqt->tqt_id == id) { /* * Instead of returning tqt_task, we just return a non * NULL value to prevent misuse, since tqt_task only * has two valid fields. */ return (ERR_PTR(-EBUSY)); } } return (NULL); } /* * Theory for the taskq_wait_id(), taskq_wait_outstanding(), and * taskq_wait() functions below. * * Taskq waiting is accomplished by tracking the lowest outstanding task * id and the next available task id. As tasks are dispatched they are * added to the tail of the pending, priority, or delay lists. As worker * threads become available the tasks are removed from the heads of these * lists and linked to the worker threads. This ensures the lists are * kept sorted by lowest to highest task id. * * Therefore the lowest outstanding task id can be quickly determined by * checking the head item from all of these lists. This value is stored * with the taskq as the lowest id. It only needs to be recalculated when * either the task with the current lowest id completes or is canceled. * * By blocking until the lowest task id exceeds the passed task id the * taskq_wait_outstanding() function can be easily implemented. Similarly, * by blocking until the lowest task id matches the next task id taskq_wait() * can be implemented. * * Callers should be aware that when there are multiple worked threads it * is possible for larger task ids to complete before smaller ones. Also * when the taskq contains delay tasks with small task ids callers may * block for a considerable length of time waiting for them to expire and * execute. */ static int taskq_wait_id_check(taskq_t *tq, taskqid_t id) { int rc; unsigned long flags; spin_lock_irqsave_nested(&tq->tq_lock, flags, tq->tq_lock_class); rc = (taskq_find(tq, id) == NULL); spin_unlock_irqrestore(&tq->tq_lock, flags); return (rc); } /* * The taskq_wait_id() function blocks until the passed task id completes. * This does not guarantee that all lower task ids have completed. */ void taskq_wait_id(taskq_t *tq, taskqid_t id) { wait_event(tq->tq_wait_waitq, taskq_wait_id_check(tq, id)); } EXPORT_SYMBOL(taskq_wait_id); static int taskq_wait_outstanding_check(taskq_t *tq, taskqid_t id) { int rc; unsigned long flags; spin_lock_irqsave_nested(&tq->tq_lock, flags, tq->tq_lock_class); rc = (id < tq->tq_lowest_id); spin_unlock_irqrestore(&tq->tq_lock, flags); return (rc); } /* * The taskq_wait_outstanding() function will block until all tasks with a * lower taskqid than the passed 'id' have been completed. Note that all * task id's are assigned monotonically at dispatch time. Zero may be * passed for the id to indicate all tasks dispatch up to this point, * but not after, should be waited for. */ void taskq_wait_outstanding(taskq_t *tq, taskqid_t id) { id = id ? id : tq->tq_next_id - 1; wait_event(tq->tq_wait_waitq, taskq_wait_outstanding_check(tq, id)); } EXPORT_SYMBOL(taskq_wait_outstanding); static int taskq_wait_check(taskq_t *tq) { int rc; unsigned long flags; spin_lock_irqsave_nested(&tq->tq_lock, flags, tq->tq_lock_class); rc = (tq->tq_lowest_id == tq->tq_next_id); spin_unlock_irqrestore(&tq->tq_lock, flags); return (rc); } /* * The taskq_wait() function will block until the taskq is empty. * This means that if a taskq re-dispatches work to itself taskq_wait() * callers will block indefinitely. */ void taskq_wait(taskq_t *tq) { wait_event(tq->tq_wait_waitq, taskq_wait_check(tq)); } EXPORT_SYMBOL(taskq_wait); int taskq_member(taskq_t *tq, kthread_t *t) { return (tq == (taskq_t *)tsd_get_by_thread(taskq_tsd, t)); } EXPORT_SYMBOL(taskq_member); /* * Cancel an already dispatched task given the task id. Still pending tasks * will be immediately canceled, and if the task is active the function will * block until it completes. Preallocated tasks which are canceled must be * freed by the caller. */ int taskq_cancel_id(taskq_t *tq, taskqid_t id) { taskq_ent_t *t; int rc = ENOENT; unsigned long flags; ASSERT(tq); spin_lock_irqsave_nested(&tq->tq_lock, flags, tq->tq_lock_class); t = taskq_find(tq, id); if (t && t != ERR_PTR(-EBUSY)) { list_del_init(&t->tqent_list); t->tqent_flags |= TQENT_FLAG_CANCEL; /* * When canceling the lowest outstanding task id we * must recalculate the new lowest outstanding id. */ if (tq->tq_lowest_id == t->tqent_id) { tq->tq_lowest_id = taskq_lowest_id(tq); ASSERT3S(tq->tq_lowest_id, >, t->tqent_id); } /* * The task_expire() function takes the tq->tq_lock so drop * drop the lock before synchronously cancelling the timer. */ if (timer_pending(&t->tqent_timer)) { spin_unlock_irqrestore(&tq->tq_lock, flags); del_timer_sync(&t->tqent_timer); spin_lock_irqsave_nested(&tq->tq_lock, flags, tq->tq_lock_class); } if (!(t->tqent_flags & TQENT_FLAG_PREALLOC)) task_done(tq, t); rc = 0; } spin_unlock_irqrestore(&tq->tq_lock, flags); if (t == ERR_PTR(-EBUSY)) { taskq_wait_id(tq, id); rc = EBUSY; } return (rc); } EXPORT_SYMBOL(taskq_cancel_id); static int taskq_thread_spawn(taskq_t *tq); taskqid_t taskq_dispatch(taskq_t *tq, task_func_t func, void *arg, uint_t flags) { taskq_ent_t *t; taskqid_t rc = TASKQID_INVALID; unsigned long irqflags; ASSERT(tq); ASSERT(func); spin_lock_irqsave_nested(&tq->tq_lock, irqflags, tq->tq_lock_class); /* Taskq being destroyed and all tasks drained */ if (!(tq->tq_flags & TASKQ_ACTIVE)) goto out; /* Do not queue the task unless there is idle thread for it */ ASSERT(tq->tq_nactive <= tq->tq_nthreads); if ((flags & TQ_NOQUEUE) && (tq->tq_nactive == tq->tq_nthreads)) { /* Dynamic taskq may be able to spawn another thread */ if (!(tq->tq_flags & TASKQ_DYNAMIC) || taskq_thread_spawn(tq) == 0) goto out; } if ((t = task_alloc(tq, flags, &irqflags)) == NULL) goto out; spin_lock(&t->tqent_lock); /* Queue to the front of the list to enforce TQ_NOQUEUE semantics */ if (flags & TQ_NOQUEUE) list_add(&t->tqent_list, &tq->tq_prio_list); /* Queue to the priority list instead of the pending list */ else if (flags & TQ_FRONT) list_add_tail(&t->tqent_list, &tq->tq_prio_list); else list_add_tail(&t->tqent_list, &tq->tq_pend_list); t->tqent_id = rc = tq->tq_next_id; tq->tq_next_id++; t->tqent_func = func; t->tqent_arg = arg; t->tqent_taskq = tq; -#ifndef HAVE_KERNEL_TIMER_FUNCTION_TIMER_LIST - t->tqent_timer.data = 0; -#endif t->tqent_timer.function = NULL; t->tqent_timer.expires = 0; t->tqent_birth = jiffies; ASSERT(!(t->tqent_flags & TQENT_FLAG_PREALLOC)); spin_unlock(&t->tqent_lock); wake_up(&tq->tq_work_waitq); out: /* Spawn additional taskq threads if required. */ if (!(flags & TQ_NOQUEUE) && tq->tq_nactive == tq->tq_nthreads) (void) taskq_thread_spawn(tq); spin_unlock_irqrestore(&tq->tq_lock, irqflags); return (rc); } EXPORT_SYMBOL(taskq_dispatch); taskqid_t taskq_dispatch_delay(taskq_t *tq, task_func_t func, void *arg, uint_t flags, clock_t expire_time) { taskqid_t rc = TASKQID_INVALID; taskq_ent_t *t; unsigned long irqflags; ASSERT(tq); ASSERT(func); spin_lock_irqsave_nested(&tq->tq_lock, irqflags, tq->tq_lock_class); /* Taskq being destroyed and all tasks drained */ if (!(tq->tq_flags & TASKQ_ACTIVE)) goto out; if ((t = task_alloc(tq, flags, &irqflags)) == NULL) goto out; spin_lock(&t->tqent_lock); /* Queue to the delay list for subsequent execution */ list_add_tail(&t->tqent_list, &tq->tq_delay_list); t->tqent_id = rc = tq->tq_next_id; tq->tq_next_id++; t->tqent_func = func; t->tqent_arg = arg; t->tqent_taskq = tq; -#ifndef HAVE_KERNEL_TIMER_FUNCTION_TIMER_LIST - t->tqent_timer.data = (unsigned long)t; -#endif t->tqent_timer.function = task_expire; t->tqent_timer.expires = (unsigned long)expire_time; add_timer(&t->tqent_timer); ASSERT(!(t->tqent_flags & TQENT_FLAG_PREALLOC)); spin_unlock(&t->tqent_lock); out: /* Spawn additional taskq threads if required. */ if (tq->tq_nactive == tq->tq_nthreads) (void) taskq_thread_spawn(tq); spin_unlock_irqrestore(&tq->tq_lock, irqflags); return (rc); } EXPORT_SYMBOL(taskq_dispatch_delay); void taskq_dispatch_ent(taskq_t *tq, task_func_t func, void *arg, uint_t flags, taskq_ent_t *t) { unsigned long irqflags; ASSERT(tq); ASSERT(func); spin_lock_irqsave_nested(&tq->tq_lock, irqflags, tq->tq_lock_class); /* Taskq being destroyed and all tasks drained */ if (!(tq->tq_flags & TASKQ_ACTIVE)) { t->tqent_id = TASKQID_INVALID; goto out; } if ((flags & TQ_NOQUEUE) && (tq->tq_nactive == tq->tq_nthreads)) { /* Dynamic taskq may be able to spawn another thread */ if (!(tq->tq_flags & TASKQ_DYNAMIC) || taskq_thread_spawn(tq) == 0) goto out2; flags |= TQ_FRONT; } spin_lock(&t->tqent_lock); /* * Make sure the entry is not on some other taskq; it is important to * ASSERT() under lock */ ASSERT(taskq_empty_ent(t)); /* * Mark it as a prealloc'd task. This is important * to ensure that we don't free it later. */ t->tqent_flags |= TQENT_FLAG_PREALLOC; /* Queue to the priority list instead of the pending list */ if (flags & TQ_FRONT) list_add_tail(&t->tqent_list, &tq->tq_prio_list); else list_add_tail(&t->tqent_list, &tq->tq_pend_list); t->tqent_id = tq->tq_next_id; tq->tq_next_id++; t->tqent_func = func; t->tqent_arg = arg; t->tqent_taskq = tq; t->tqent_birth = jiffies; spin_unlock(&t->tqent_lock); wake_up(&tq->tq_work_waitq); out: /* Spawn additional taskq threads if required. */ if (tq->tq_nactive == tq->tq_nthreads) (void) taskq_thread_spawn(tq); out2: spin_unlock_irqrestore(&tq->tq_lock, irqflags); } EXPORT_SYMBOL(taskq_dispatch_ent); int taskq_empty_ent(taskq_ent_t *t) { return (list_empty(&t->tqent_list)); } EXPORT_SYMBOL(taskq_empty_ent); void taskq_init_ent(taskq_ent_t *t) { spin_lock_init(&t->tqent_lock); init_waitqueue_head(&t->tqent_waitq); -#ifdef HAVE_KERNEL_TIMER_FUNCTION_TIMER_LIST timer_setup(&t->tqent_timer, NULL, 0); -#else - init_timer(&t->tqent_timer); -#endif INIT_LIST_HEAD(&t->tqent_list); t->tqent_id = 0; t->tqent_func = NULL; t->tqent_arg = NULL; t->tqent_flags = 0; t->tqent_taskq = NULL; } EXPORT_SYMBOL(taskq_init_ent); /* * Return the next pending task, preference is given to tasks on the * priority list which were dispatched with TQ_FRONT. */ static taskq_ent_t * taskq_next_ent(taskq_t *tq) { struct list_head *list; if (!list_empty(&tq->tq_prio_list)) list = &tq->tq_prio_list; else if (!list_empty(&tq->tq_pend_list)) list = &tq->tq_pend_list; else return (NULL); return (list_entry(list->next, taskq_ent_t, tqent_list)); } /* * Spawns a new thread for the specified taskq. */ static void taskq_thread_spawn_task(void *arg) { taskq_t *tq = (taskq_t *)arg; unsigned long flags; if (taskq_thread_create(tq) == NULL) { /* restore spawning count if failed */ spin_lock_irqsave_nested(&tq->tq_lock, flags, tq->tq_lock_class); tq->tq_nspawn--; spin_unlock_irqrestore(&tq->tq_lock, flags); } } /* * Spawn addition threads for dynamic taskqs (TASKQ_DYNAMIC) the current * number of threads is insufficient to handle the pending tasks. These * new threads must be created by the dedicated dynamic_taskq to avoid * deadlocks between thread creation and memory reclaim. The system_taskq * which is also a dynamic taskq cannot be safely used for this. */ static int taskq_thread_spawn(taskq_t *tq) { int spawning = 0; if (!(tq->tq_flags & TASKQ_DYNAMIC)) return (0); if ((tq->tq_nthreads + tq->tq_nspawn < tq->tq_maxthreads) && (tq->tq_flags & TASKQ_ACTIVE)) { spawning = (++tq->tq_nspawn); taskq_dispatch(dynamic_taskq, taskq_thread_spawn_task, tq, TQ_NOSLEEP); } return (spawning); } /* * Threads in a dynamic taskq should only exit once it has been completely * drained and no other threads are actively servicing tasks. This prevents * threads from being created and destroyed more than is required. * * The first thread is the thread list is treated as the primary thread. * There is nothing special about the primary thread but in order to avoid * all the taskq pids from changing we opt to make it long running. */ static int taskq_thread_should_stop(taskq_t *tq, taskq_thread_t *tqt) { if (!(tq->tq_flags & TASKQ_DYNAMIC)) return (0); if (list_first_entry(&(tq->tq_thread_list), taskq_thread_t, tqt_thread_list) == tqt) return (0); return ((tq->tq_nspawn == 0) && /* No threads are being spawned */ (tq->tq_nactive == 0) && /* No threads are handling tasks */ (tq->tq_nthreads > 1) && /* More than 1 thread is running */ (!taskq_next_ent(tq)) && /* There are no pending tasks */ (spl_taskq_thread_dynamic)); /* Dynamic taskqs are allowed */ } static int taskq_thread(void *args) { DECLARE_WAITQUEUE(wait, current); sigset_t blocked; taskq_thread_t *tqt = args; taskq_t *tq; taskq_ent_t *t; int seq_tasks = 0; unsigned long flags; taskq_ent_t dup_task = {}; ASSERT(tqt); ASSERT(tqt->tqt_tq); tq = tqt->tqt_tq; current->flags |= PF_NOFREEZE; (void) spl_fstrans_mark(); sigfillset(&blocked); sigprocmask(SIG_BLOCK, &blocked, NULL); flush_signals(current); tsd_set(taskq_tsd, tq); spin_lock_irqsave_nested(&tq->tq_lock, flags, tq->tq_lock_class); /* * If we are dynamically spawned, decrease spawning count. Note that * we could be created during taskq_create, in which case we shouldn't * do the decrement. But it's fine because taskq_create will reset * tq_nspawn later. */ if (tq->tq_flags & TASKQ_DYNAMIC) tq->tq_nspawn--; /* Immediately exit if more threads than allowed were created. */ if (tq->tq_nthreads >= tq->tq_maxthreads) goto error; tq->tq_nthreads++; list_add_tail(&tqt->tqt_thread_list, &tq->tq_thread_list); wake_up(&tq->tq_wait_waitq); set_current_state(TASK_INTERRUPTIBLE); while (!kthread_should_stop()) { if (list_empty(&tq->tq_pend_list) && list_empty(&tq->tq_prio_list)) { if (taskq_thread_should_stop(tq, tqt)) { wake_up_all(&tq->tq_wait_waitq); break; } add_wait_queue_exclusive(&tq->tq_work_waitq, &wait); spin_unlock_irqrestore(&tq->tq_lock, flags); schedule(); seq_tasks = 0; spin_lock_irqsave_nested(&tq->tq_lock, flags, tq->tq_lock_class); remove_wait_queue(&tq->tq_work_waitq, &wait); } else { __set_current_state(TASK_RUNNING); } if ((t = taskq_next_ent(tq)) != NULL) { list_del_init(&t->tqent_list); /* * A TQENT_FLAG_PREALLOC task may be reused or freed * during the task function call. Store tqent_id and * tqent_flags here. * * Also use an on stack taskq_ent_t for tqt_task * assignment in this case. We only populate the two * fields used by the only user in taskq proc file. */ tqt->tqt_id = t->tqent_id; tqt->tqt_flags = t->tqent_flags; if (t->tqent_flags & TQENT_FLAG_PREALLOC) { dup_task.tqent_func = t->tqent_func; dup_task.tqent_arg = t->tqent_arg; t = &dup_task; } tqt->tqt_task = t; taskq_insert_in_order(tq, tqt); tq->tq_nactive++; spin_unlock_irqrestore(&tq->tq_lock, flags); /* Perform the requested task */ t->tqent_func(t->tqent_arg); spin_lock_irqsave_nested(&tq->tq_lock, flags, tq->tq_lock_class); tq->tq_nactive--; list_del_init(&tqt->tqt_active_list); tqt->tqt_task = NULL; /* For prealloc'd tasks, we don't free anything. */ if (!(tqt->tqt_flags & TQENT_FLAG_PREALLOC)) task_done(tq, t); /* * When the current lowest outstanding taskqid is * done calculate the new lowest outstanding id */ if (tq->tq_lowest_id == tqt->tqt_id) { tq->tq_lowest_id = taskq_lowest_id(tq); ASSERT3S(tq->tq_lowest_id, >, tqt->tqt_id); } /* Spawn additional taskq threads if required. */ if ((++seq_tasks) > spl_taskq_thread_sequential && taskq_thread_spawn(tq)) seq_tasks = 0; tqt->tqt_id = TASKQID_INVALID; tqt->tqt_flags = 0; wake_up_all(&tq->tq_wait_waitq); } else { if (taskq_thread_should_stop(tq, tqt)) break; } set_current_state(TASK_INTERRUPTIBLE); } __set_current_state(TASK_RUNNING); tq->tq_nthreads--; list_del_init(&tqt->tqt_thread_list); error: kmem_free(tqt, sizeof (taskq_thread_t)); spin_unlock_irqrestore(&tq->tq_lock, flags); tsd_set(taskq_tsd, NULL); return (0); } static taskq_thread_t * taskq_thread_create(taskq_t *tq) { static int last_used_cpu = 0; taskq_thread_t *tqt; tqt = kmem_alloc(sizeof (*tqt), KM_PUSHPAGE); INIT_LIST_HEAD(&tqt->tqt_thread_list); INIT_LIST_HEAD(&tqt->tqt_active_list); tqt->tqt_tq = tq; tqt->tqt_id = TASKQID_INVALID; tqt->tqt_thread = spl_kthread_create(taskq_thread, tqt, "%s", tq->tq_name); if (tqt->tqt_thread == NULL) { kmem_free(tqt, sizeof (taskq_thread_t)); return (NULL); } if (spl_taskq_thread_bind) { last_used_cpu = (last_used_cpu + 1) % num_online_cpus(); kthread_bind(tqt->tqt_thread, last_used_cpu); } if (spl_taskq_thread_priority) set_user_nice(tqt->tqt_thread, PRIO_TO_NICE(tq->tq_pri)); wake_up_process(tqt->tqt_thread); return (tqt); } taskq_t * taskq_create(const char *name, int nthreads, pri_t pri, int minalloc, int maxalloc, uint_t flags) { taskq_t *tq; taskq_thread_t *tqt; int count = 0, rc = 0, i; unsigned long irqflags; ASSERT(name != NULL); ASSERT(minalloc >= 0); ASSERT(maxalloc <= INT_MAX); ASSERT(!(flags & (TASKQ_CPR_SAFE))); /* Unsupported */ /* Scale the number of threads using nthreads as a percentage */ if (flags & TASKQ_THREADS_CPU_PCT) { ASSERT(nthreads <= 100); ASSERT(nthreads >= 0); nthreads = MIN(nthreads, 100); nthreads = MAX(nthreads, 0); nthreads = MAX((num_online_cpus() * nthreads) / 100, 1); } tq = kmem_alloc(sizeof (*tq), KM_PUSHPAGE); if (tq == NULL) return (NULL); spin_lock_init(&tq->tq_lock); INIT_LIST_HEAD(&tq->tq_thread_list); INIT_LIST_HEAD(&tq->tq_active_list); tq->tq_name = strdup(name); tq->tq_nactive = 0; tq->tq_nthreads = 0; tq->tq_nspawn = 0; tq->tq_maxthreads = nthreads; tq->tq_pri = pri; tq->tq_minalloc = minalloc; tq->tq_maxalloc = maxalloc; tq->tq_nalloc = 0; tq->tq_flags = (flags | TASKQ_ACTIVE); tq->tq_next_id = TASKQID_INITIAL; tq->tq_lowest_id = TASKQID_INITIAL; INIT_LIST_HEAD(&tq->tq_free_list); INIT_LIST_HEAD(&tq->tq_pend_list); INIT_LIST_HEAD(&tq->tq_prio_list); INIT_LIST_HEAD(&tq->tq_delay_list); init_waitqueue_head(&tq->tq_work_waitq); init_waitqueue_head(&tq->tq_wait_waitq); tq->tq_lock_class = TQ_LOCK_GENERAL; INIT_LIST_HEAD(&tq->tq_taskqs); if (flags & TASKQ_PREPOPULATE) { spin_lock_irqsave_nested(&tq->tq_lock, irqflags, tq->tq_lock_class); for (i = 0; i < minalloc; i++) task_done(tq, task_alloc(tq, TQ_PUSHPAGE | TQ_NEW, &irqflags)); spin_unlock_irqrestore(&tq->tq_lock, irqflags); } if ((flags & TASKQ_DYNAMIC) && spl_taskq_thread_dynamic) nthreads = 1; for (i = 0; i < nthreads; i++) { tqt = taskq_thread_create(tq); if (tqt == NULL) rc = 1; else count++; } /* Wait for all threads to be started before potential destroy */ wait_event(tq->tq_wait_waitq, tq->tq_nthreads == count); /* * taskq_thread might have touched nspawn, but we don't want them to * because they're not dynamically spawned. So we reset it to 0 */ tq->tq_nspawn = 0; if (rc) { taskq_destroy(tq); tq = NULL; } else { down_write(&tq_list_sem); tq->tq_instance = taskq_find_by_name(name) + 1; list_add_tail(&tq->tq_taskqs, &tq_list); up_write(&tq_list_sem); } return (tq); } EXPORT_SYMBOL(taskq_create); void taskq_destroy(taskq_t *tq) { struct task_struct *thread; taskq_thread_t *tqt; taskq_ent_t *t; unsigned long flags; ASSERT(tq); spin_lock_irqsave_nested(&tq->tq_lock, flags, tq->tq_lock_class); tq->tq_flags &= ~TASKQ_ACTIVE; spin_unlock_irqrestore(&tq->tq_lock, flags); /* * When TASKQ_ACTIVE is clear new tasks may not be added nor may * new worker threads be spawned for dynamic taskq. */ if (dynamic_taskq != NULL) taskq_wait_outstanding(dynamic_taskq, 0); taskq_wait(tq); /* remove taskq from global list used by the kstats */ down_write(&tq_list_sem); list_del(&tq->tq_taskqs); up_write(&tq_list_sem); spin_lock_irqsave_nested(&tq->tq_lock, flags, tq->tq_lock_class); /* wait for spawning threads to insert themselves to the list */ while (tq->tq_nspawn) { spin_unlock_irqrestore(&tq->tq_lock, flags); schedule_timeout_interruptible(1); spin_lock_irqsave_nested(&tq->tq_lock, flags, tq->tq_lock_class); } /* * Signal each thread to exit and block until it does. Each thread * is responsible for removing itself from the list and freeing its * taskq_thread_t. This allows for idle threads to opt to remove * themselves from the taskq. They can be recreated as needed. */ while (!list_empty(&tq->tq_thread_list)) { tqt = list_entry(tq->tq_thread_list.next, taskq_thread_t, tqt_thread_list); thread = tqt->tqt_thread; spin_unlock_irqrestore(&tq->tq_lock, flags); kthread_stop(thread); spin_lock_irqsave_nested(&tq->tq_lock, flags, tq->tq_lock_class); } while (!list_empty(&tq->tq_free_list)) { t = list_entry(tq->tq_free_list.next, taskq_ent_t, tqent_list); ASSERT(!(t->tqent_flags & TQENT_FLAG_PREALLOC)); list_del_init(&t->tqent_list); task_free(tq, t); } ASSERT0(tq->tq_nthreads); ASSERT0(tq->tq_nalloc); ASSERT0(tq->tq_nspawn); ASSERT(list_empty(&tq->tq_thread_list)); ASSERT(list_empty(&tq->tq_active_list)); ASSERT(list_empty(&tq->tq_free_list)); ASSERT(list_empty(&tq->tq_pend_list)); ASSERT(list_empty(&tq->tq_prio_list)); ASSERT(list_empty(&tq->tq_delay_list)); spin_unlock_irqrestore(&tq->tq_lock, flags); strfree(tq->tq_name); kmem_free(tq, sizeof (taskq_t)); } EXPORT_SYMBOL(taskq_destroy); static unsigned int spl_taskq_kick = 0; /* * 2.6.36 API Change * module_param_cb is introduced to take kernel_param_ops and * module_param_call is marked as obsolete. Also set and get operations * were changed to take a 'const struct kernel_param *'. */ static int #ifdef module_param_cb param_set_taskq_kick(const char *val, const struct kernel_param *kp) #else param_set_taskq_kick(const char *val, struct kernel_param *kp) #endif { int ret; taskq_t *tq; taskq_ent_t *t; unsigned long flags; ret = param_set_uint(val, kp); if (ret < 0 || !spl_taskq_kick) return (ret); /* reset value */ spl_taskq_kick = 0; down_read(&tq_list_sem); list_for_each_entry(tq, &tq_list, tq_taskqs) { spin_lock_irqsave_nested(&tq->tq_lock, flags, tq->tq_lock_class); /* Check if the first pending is older than 5 seconds */ t = taskq_next_ent(tq); if (t && time_after(jiffies, t->tqent_birth + 5*HZ)) { (void) taskq_thread_spawn(tq); printk(KERN_INFO "spl: Kicked taskq %s/%d\n", tq->tq_name, tq->tq_instance); } spin_unlock_irqrestore(&tq->tq_lock, flags); } up_read(&tq_list_sem); return (ret); } #ifdef module_param_cb static const struct kernel_param_ops param_ops_taskq_kick = { .set = param_set_taskq_kick, .get = param_get_uint, }; module_param_cb(spl_taskq_kick, ¶m_ops_taskq_kick, &spl_taskq_kick, 0644); #else module_param_call(spl_taskq_kick, param_set_taskq_kick, param_get_uint, &spl_taskq_kick, 0644); #endif MODULE_PARM_DESC(spl_taskq_kick, "Write nonzero to kick stuck taskqs to spawn more threads"); int spl_taskq_init(void) { init_rwsem(&tq_list_sem); tsd_create(&taskq_tsd, NULL); system_taskq = taskq_create("spl_system_taskq", MAX(boot_ncpus, 64), maxclsyspri, boot_ncpus, INT_MAX, TASKQ_PREPOPULATE|TASKQ_DYNAMIC); if (system_taskq == NULL) return (1); system_delay_taskq = taskq_create("spl_delay_taskq", MAX(boot_ncpus, 4), maxclsyspri, boot_ncpus, INT_MAX, TASKQ_PREPOPULATE|TASKQ_DYNAMIC); if (system_delay_taskq == NULL) { taskq_destroy(system_taskq); return (1); } dynamic_taskq = taskq_create("spl_dynamic_taskq", 1, maxclsyspri, boot_ncpus, INT_MAX, TASKQ_PREPOPULATE); if (dynamic_taskq == NULL) { taskq_destroy(system_taskq); taskq_destroy(system_delay_taskq); return (1); } /* * This is used to annotate tq_lock, so * taskq_dispatch -> taskq_thread_spawn -> taskq_dispatch * does not trigger a lockdep warning re: possible recursive locking */ dynamic_taskq->tq_lock_class = TQ_LOCK_DYNAMIC; return (0); } void spl_taskq_fini(void) { taskq_destroy(dynamic_taskq); dynamic_taskq = NULL; taskq_destroy(system_delay_taskq); system_delay_taskq = NULL; taskq_destroy(system_taskq); system_taskq = NULL; tsd_destroy(&taskq_tsd); }