Index: head/share/man/man4/em.4 =================================================================== --- head/share/man/man4/em.4 (revision 326983) +++ head/share/man/man4/em.4 (revision 326984) @@ -1,324 +1,324 @@ .\" Copyright (c) 2001-2003, 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. .\" .\" * Other names and brands may be claimed as the property of others. .\" .\" $FreeBSD$ .\" .Dd March 22, 2017 .Dt EM 4 .Os .Sh NAME .Nm em .Nd "Intel(R) PRO/1000 Gigabit Ethernet adapter driver" .Sh SYNOPSIS To compile this driver into the kernel, place the following line in your kernel configuration file: .Bd -ragged -offset indent .Cd "device em" .Ed .Pp Alternatively, to load the driver as a module at boot time, place the following line in .Xr loader.conf 5 : .Bd -literal -offset indent if_em_load="YES" .Ed .Sh DESCRIPTION The .Nm driver provides support for PCI Gigabit Ethernet adapters based on the Intel 82540, 82541ER, 82541PI, 82542, 82543, 82544, 82545, 82546, 82546EB, 82546GB, 82547, 82571, 81572, 82573, and 82574 Ethernet controller chips. .Pp The .Nm driver also provides support for PCI Express Gigabit Ethernet adapters based on the Intel 82575, 82576, 82580, i21x and i35x. .Pp The driver supports Transmit/Receive checksum offload and Jumbo Frames on all but 82542-based adapters. .Pp Furthermore it supports TCP segmentation offload (TSO) on all adapters but those based on the 82543, 82544 and 82547 controller chips. The identification LEDs of the adapters supported by the .Nm driver can be controlled via the .Xr led 4 API for localization purposes. For further hardware information, see the .Pa README included with the driver. .Pp For questions related to hardware requirements, refer to the documentation supplied with your Intel PRO/1000 adapter. All hardware requirements listed apply to use with .Fx . .Pp Support for Jumbo Frames is provided via the interface MTU setting. Selecting an MTU larger than 1500 bytes with the .Xr ifconfig 8 utility configures the adapter to receive and transmit Jumbo Frames. The maximum MTU size for Jumbo Frames is 16114. .Pp This driver supports hardware assisted VLANs. The .Nm driver supports the following media types: .Bl -tag -width ".Cm 10baseT/UTP" .It Cm autoselect Enables auto-negotiation for speed and duplex. .It Cm 10baseT/UTP Sets 10Mbps operation. Use the .Cm mediaopt option to select .Cm full-duplex mode. .It Cm 100baseTX Sets 100Mbps operation. Use the .Cm mediaopt option to select .Cm full-duplex mode. .It Cm 1000baseSX Sets 1000Mbps operation. Only .Cm full-duplex mode is supported at this speed. .It Cm 1000baseTX Sets 1000Mbps operation. Only .Cm full-duplex mode is supported at this speed. .El .Pp The .Nm driver supports the following media options: .Bl -tag -width ".Cm full-duplex" .It Cm full-duplex Forces full-duplex operation .It Cm half-duplex Forces half-duplex operation. .El .Pp Only use .Cm mediaopt to set the driver to .Cm full-duplex . If .Cm mediaopt is not specified, the driver defaults to .Cm half-duplex . .Pp For more information on configuring this device, see .Xr ifconfig 8 . .Sh HARDWARE The .Nm driver supports Gigabit Ethernet adapters based on the Intel 82540, 82541ER, 82541PI, 82542, 82543, 82544, 82545, 82546, 82546EB, 82546GB, 82547, 82571, 82572, 82573, 82574, 82575, 82576, and 82580 controller chips: .Pp .Bl -bullet -compact .It Intel Gigabit ET Dual Port Server Adapter (82576) .It Intel Gigabit VT Quad Port Server Adapter (82575) .It Intel Single, Dual and Quad Gigabit Ethernet Controller (82580) .It Intel i210 and i211 Gigabit Ethernet Controller .It Intel i350 and i354 Gigabit Ethernet Controller .It Intel PRO/1000 CT Network Connection (82547) .It Intel PRO/1000 F Server Adapter (82543) .It Intel PRO/1000 Gigabit Server Adapter (82542) .It Intel PRO/1000 GT Desktop Adapter (82541PI) .It Intel PRO/1000 MF Dual Port Server Adapter (82546) .It Intel PRO/1000 MF Server Adapter (82545) .It Intel PRO/1000 MF Server Adapter (LX) (82545) .It Intel PRO/1000 MT Desktop Adapter (82540) .It Intel PRO/1000 MT Desktop Adapter (82541) .It Intel PRO/1000 MT Dual Port Server Adapter (82546) .It Intel PRO/1000 MT Quad Port Server Adapter (82546EB) .It Intel PRO/1000 MT Server Adapter (82545) .It Intel PRO/1000 PF Dual Port Server Adapter (82571) .It Intel PRO/1000 PF Quad Port Server Adapter (82571) .It Intel PRO/1000 PF Server Adapter (82572) .It Intel PRO/1000 PT Desktop Adapter (82572) .It Intel PRO/1000 PT Dual Port Server Adapter (82571) .It Intel PRO/1000 PT Quad Port Server Adapter (82571) .It Intel PRO/1000 PT Server Adapter (82572) .It Intel PRO/1000 T Desktop Adapter (82544) .It Intel PRO/1000 T Server Adapter (82543) .It Intel PRO/1000 XF Server Adapter (82544) .It Intel PRO/1000 XT Server Adapter (82544) .El .Sh LOADER TUNABLES Tunables can be set at the .Xr loader 8 prompt before booting the kernel or stored in .Xr loader.conf 5 . .Bl -tag -width indent .It Va hw.em.disable_crc_stripping Disable or enable hardware stripping of CRC field. This is mostly useful on BMC/IPMI shared interfaces where stripping the CRC causes remote access over IPMI to fail. Default 0 (enabled). .It Va hw.em.eee_setting Disable or enable Energy Efficient Ethernet. Default 1 (disabled). .It Va hw.em.msix Enable or disable MSI-X style interrupts. Default 1 (enabled). .It Va hw.em.smart_pwr_down Enable or disable smart power down features on newer adapters. Default 0 (disabled). .It Va hw.em.sbp Show bad packets when in promiscuous mode. Default 0 (off). .It Va hw.em.rxd Number of receive descriptors allocated by the driver. The default value is 1024 for adapters newer than 82547, and 256 for older ones. The 82542 and 82543-based adapters can handle up to 256 descriptors, while others can have up to 4096. .It Va hw.em.txd Number of transmit descriptors allocated by the driver. The default value is 1024 for adapters newer than 82547, and 256 for older ones. The 82542 and 82543-based adapters can handle up to 256 descriptors, while others can have up to 4096. .It Va hw.em.rx_int_delay This value delays the generation of receive interrupts in units of 1.024 microseconds. The default value is 0, since adapters may hang with this feature being enabled. .It Va hw.em.rx_abs_int_delay If .Va hw.em.rx_int_delay is non-zero, this tunable limits the maximum delay in which a receive interrupt is generated. .It Va hw.em.tx_int_delay This value delays the generation of transmit interrupts in units of 1.024 microseconds. The default value is 64. .It Va hw.em.tx_abs_int_delay If .Va hw.em.tx_int_delay is non-zero, this tunable limits the maximum delay in which a transmit interrupt is generated. .Sh FILES .Bl -tag -width /dev/led/em* .It Pa /dev/led/em* identification LED device nodes .El .Sh EXAMPLES Make the identification LED of em0 blink: .Pp .Dl "echo f2 > /dev/led/em0" .Pp Turn the identification LED of em0 off again: .Pp .Dl "echo 0 > /dev/led/em0" .Sh DIAGNOSTICS .Bl -diag .It "em%d: Unable to allocate bus resource: memory" A fatal initialization error has occurred. .It "em%d: Unable to allocate bus resource: interrupt" A fatal initialization error has occurred. .It "em%d: watchdog timeout -- resetting" The device has stopped responding to the network, or there is a problem with the network connection (cable). .El .Sh SUPPORT For general information and support, go to the Intel support website at: .Pa http://support.intel.com . .Pp If an issue is identified with the released source code on the supported kernel with a supported adapter, email the specific information related to the issue to .Aq Mt freebsd@intel.com . .Sh SEE ALSO .Xr altq 4 , .Xr arp 4 , .Xr led 4 , .Xr netintro 4 , .Xr ng_ether 4 , .Xr polling 4 , .Xr vlan 4 , .Xr ifconfig 8 .Sh HISTORY The .Nm device driver first appeared in .Fx 4.4 . .Nm was merged with the igb device driver and converted to the iflib framework in .Fx 12.0 . .Sh AUTHORS .An -nosplit The .Nm driver was originally written by .An Intel Corporation Aq Mt freebsd@intel.com . It was merged with igb driver and converted to the iflib framework by -.An Matthew Macy Aq Mt mmacy@nextbsd.org +.An Matthew Macy Aq Mt mmacy@mattmacy.io and .An Sean Bruno Aq Mt sbruno@FreeBSD.org . Index: head/sys/compat/linuxkpi/common/src/linux_page.c =================================================================== --- head/sys/compat/linuxkpi/common/src/linux_page.c (revision 326983) +++ head/sys/compat/linuxkpi/common/src/linux_page.c (revision 326984) @@ -1,395 +1,395 @@ /*- * Copyright (c) 2010 Isilon Systems, Inc. - * Copyright (c) 2016 Matt Macy (mmacy@nextbsd.org) + * Copyright (c) 2016 Matthew Macy (mmacy@mattmacy.io) * Copyright (c) 2017 Mellanox Technologies, Ltd. * 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 unmodified, 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 ``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 #include #include #include #include #include #include #include #include #include #if defined(__amd64__) || defined(__aarch64__) || defined(__riscv) #define LINUXKPI_HAVE_DMAP #else #undef LINUXKPI_HAVE_DMAP #endif void * linux_page_address(struct page *page) { if (page->object != kmem_object && page->object != kernel_object) { #ifdef LINUXKPI_HAVE_DMAP return ((void *)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(page))); #else return (NULL); #endif } return ((void *)(uintptr_t)(VM_MIN_KERNEL_ADDRESS + IDX_TO_OFF(page->pindex))); } vm_page_t linux_alloc_pages(gfp_t flags, unsigned int order) { #ifdef LINUXKPI_HAVE_DMAP unsigned long npages = 1UL << order; int req = (flags & M_ZERO) ? (VM_ALLOC_ZERO | VM_ALLOC_NOOBJ | VM_ALLOC_NORMAL) : (VM_ALLOC_NOOBJ | VM_ALLOC_NORMAL); vm_page_t page; if (order == 0 && (flags & GFP_DMA32) == 0) { page = vm_page_alloc(NULL, 0, req); if (page == NULL) return (NULL); } else { vm_paddr_t pmax = (flags & GFP_DMA32) ? BUS_SPACE_MAXADDR_32BIT : BUS_SPACE_MAXADDR; retry: page = vm_page_alloc_contig(NULL, 0, req, npages, 0, pmax, PAGE_SIZE, 0, VM_MEMATTR_DEFAULT); if (page == NULL) { if (flags & M_WAITOK) { if (!vm_page_reclaim_contig(req, npages, 0, pmax, PAGE_SIZE, 0)) { VM_WAIT; } flags &= ~M_WAITOK; goto retry; } return (NULL); } } if (flags & M_ZERO) { unsigned long x; for (x = 0; x != npages; x++) { vm_page_t pgo = page + x; if ((pgo->flags & PG_ZERO) == 0) pmap_zero_page(pgo); } } #else vm_offset_t vaddr; vm_page_t page; vaddr = linux_alloc_kmem(flags, order); if (vaddr == 0) return (NULL); page = PHYS_TO_VM_PAGE(vtophys((void *)vaddr)); KASSERT(vaddr == (vm_offset_t)page_address(page), ("Page address mismatch")); #endif return (page); } void linux_free_pages(vm_page_t page, unsigned int order) { #ifdef LINUXKPI_HAVE_DMAP unsigned long npages = 1UL << order; unsigned long x; for (x = 0; x != npages; x++) { vm_page_t pgo = page + x; vm_page_lock(pgo); vm_page_free(pgo); vm_page_unlock(pgo); } #else vm_offset_t vaddr; vaddr = (vm_offset_t)page_address(page); linux_free_kmem(vaddr, order); #endif } vm_offset_t linux_alloc_kmem(gfp_t flags, unsigned int order) { size_t size = ((size_t)PAGE_SIZE) << order; vm_offset_t addr; if ((flags & GFP_DMA32) == 0) { addr = kmem_malloc(kmem_arena, size, flags & GFP_NATIVE_MASK); } else { addr = kmem_alloc_contig(kmem_arena, size, flags & GFP_NATIVE_MASK, 0, BUS_SPACE_MAXADDR_32BIT, PAGE_SIZE, 0, VM_MEMATTR_DEFAULT); } return (addr); } void linux_free_kmem(vm_offset_t addr, unsigned int order) { size_t size = ((size_t)PAGE_SIZE) << order; kmem_free(kmem_arena, addr, size); } static int linux_get_user_pages_internal(vm_map_t map, unsigned long start, int nr_pages, int write, struct page **pages) { vm_prot_t prot; size_t len; int count; int i; prot = write ? (VM_PROT_READ | VM_PROT_WRITE) : VM_PROT_READ; len = ((size_t)nr_pages) << PAGE_SHIFT; count = vm_fault_quick_hold_pages(map, start, len, prot, pages, nr_pages); if (count == -1) return (-EFAULT); for (i = 0; i != nr_pages; i++) { struct page *pg = pages[i]; vm_page_lock(pg); vm_page_wire(pg); vm_page_unhold(pg); vm_page_unlock(pg); } return (nr_pages); } int __get_user_pages_fast(unsigned long start, int nr_pages, int write, struct page **pages) { vm_map_t map; vm_page_t *mp; vm_offset_t va; vm_offset_t end; vm_prot_t prot; int count; if (nr_pages == 0 || in_interrupt()) return (0); MPASS(pages != NULL); va = start; map = &curthread->td_proc->p_vmspace->vm_map; end = start + (((size_t)nr_pages) << PAGE_SHIFT); if (start < vm_map_min(map) || end > vm_map_max(map)) return (-EINVAL); prot = write ? (VM_PROT_READ | VM_PROT_WRITE) : VM_PROT_READ; for (count = 0, mp = pages, va = start; va < end; mp++, va += PAGE_SIZE, count++) { *mp = pmap_extract_and_hold(map->pmap, va, prot); if (*mp == NULL) break; vm_page_lock(*mp); vm_page_wire(*mp); vm_page_unhold(*mp); vm_page_unlock(*mp); if ((prot & VM_PROT_WRITE) != 0 && (*mp)->dirty != VM_PAGE_BITS_ALL) { /* * Explicitly dirty the physical page. Otherwise, the * caller's changes may go unnoticed because they are * performed through an unmanaged mapping or by a DMA * operation. * * The object lock is not held here. * See vm_page_clear_dirty_mask(). */ vm_page_dirty(*mp); } } return (count); } long get_user_pages_remote(struct task_struct *task, struct mm_struct *mm, unsigned long start, unsigned long nr_pages, int gup_flags, struct page **pages, struct vm_area_struct **vmas) { vm_map_t map; map = &task->task_thread->td_proc->p_vmspace->vm_map; return (linux_get_user_pages_internal(map, start, nr_pages, !!(gup_flags & FOLL_WRITE), pages)); } long get_user_pages(unsigned long start, unsigned long nr_pages, int gup_flags, struct page **pages, struct vm_area_struct **vmas) { vm_map_t map; map = &curthread->td_proc->p_vmspace->vm_map; return (linux_get_user_pages_internal(map, start, nr_pages, !!(gup_flags & FOLL_WRITE), pages)); } int is_vmalloc_addr(const void *addr) { return (vtoslab((vm_offset_t)addr & ~UMA_SLAB_MASK) != NULL); } struct page * linux_shmem_read_mapping_page_gfp(vm_object_t obj, int pindex, gfp_t gfp) { vm_page_t page; int rv; if ((gfp & GFP_NOWAIT) != 0) panic("GFP_NOWAIT is unimplemented"); VM_OBJECT_WLOCK(obj); page = vm_page_grab(obj, pindex, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED); if (page->valid != VM_PAGE_BITS_ALL) { vm_page_xbusy(page); if (vm_pager_has_page(obj, pindex, NULL, NULL)) { rv = vm_pager_get_pages(obj, &page, 1, NULL, NULL); if (rv != VM_PAGER_OK) { vm_page_lock(page); vm_page_unwire(page, PQ_NONE); vm_page_free(page); vm_page_unlock(page); VM_OBJECT_WUNLOCK(obj); return (ERR_PTR(-EINVAL)); } MPASS(page->valid == VM_PAGE_BITS_ALL); } else { pmap_zero_page(page); page->valid = VM_PAGE_BITS_ALL; page->dirty = 0; } vm_page_xunbusy(page); } VM_OBJECT_WUNLOCK(obj); return (page); } struct linux_file * linux_shmem_file_setup(const char *name, loff_t size, unsigned long flags) { struct fileobj { struct linux_file file __aligned(sizeof(void *)); struct vnode vnode __aligned(sizeof(void *)); }; struct fileobj *fileobj; struct linux_file *filp; struct vnode *vp; int error; fileobj = kzalloc(sizeof(*fileobj), GFP_KERNEL); if (fileobj == NULL) { error = -ENOMEM; goto err_0; } filp = &fileobj->file; vp = &fileobj->vnode; filp->f_count = 1; filp->f_vnode = vp; filp->f_shmem = vm_pager_allocate(OBJT_DEFAULT, NULL, size, VM_PROT_READ | VM_PROT_WRITE, 0, curthread->td_ucred); if (filp->f_shmem == NULL) { error = -ENOMEM; goto err_1; } return (filp); err_1: kfree(filp); err_0: return (ERR_PTR(error)); } static vm_ooffset_t linux_invalidate_mapping_pages_sub(vm_object_t obj, vm_pindex_t start, vm_pindex_t end, int flags) { int start_count, end_count; VM_OBJECT_WLOCK(obj); start_count = obj->resident_page_count; vm_object_page_remove(obj, start, end, flags); end_count = obj->resident_page_count; VM_OBJECT_WUNLOCK(obj); return (start_count - end_count); } unsigned long linux_invalidate_mapping_pages(vm_object_t obj, pgoff_t start, pgoff_t end) { return (linux_invalidate_mapping_pages_sub(obj, start, end, OBJPR_CLEANONLY)); } void linux_shmem_truncate_range(vm_object_t obj, loff_t lstart, loff_t lend) { vm_pindex_t start = OFF_TO_IDX(lstart + PAGE_SIZE - 1); vm_pindex_t end = OFF_TO_IDX(lend + 1); (void) linux_invalidate_mapping_pages_sub(obj, start, end, 0); } Index: head/sys/compat/linuxkpi/common/src/linux_rcu.c =================================================================== --- head/sys/compat/linuxkpi/common/src/linux_rcu.c (revision 326983) +++ head/sys/compat/linuxkpi/common/src/linux_rcu.c (revision 326984) @@ -1,398 +1,398 @@ /*- - * Copyright (c) 2016 Matt Macy (mmacy@nextbsd.org) + * Copyright (c) 2016 Matthew Macy (mmacy@mattmacy.io) * 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 unmodified, 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 ``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 /* * By defining CONFIG_NO_RCU_SKIP LinuxKPI RCU locks and asserts will * not be skipped during panic(). */ #ifdef CONFIG_NO_RCU_SKIP #define RCU_SKIP(void) 0 #else #define RCU_SKIP(void) unlikely(SCHEDULER_STOPPED() || kdb_active) #endif struct callback_head { STAILQ_ENTRY(callback_head) entry; rcu_callback_t func; }; struct linux_epoch_head { STAILQ_HEAD(, callback_head) cb_head; struct mtx lock; struct task task; } __aligned(CACHE_LINE_SIZE); struct linux_epoch_record { ck_epoch_record_t epoch_record; TAILQ_HEAD(, task_struct) ts_head; int cpuid; } __aligned(CACHE_LINE_SIZE); /* * Verify that "struct rcu_head" is big enough to hold "struct * callback_head". This has been done to avoid having to add special * compile flags for including ck_epoch.h to all clients of the * LinuxKPI. */ CTASSERT(sizeof(struct rcu_head) == sizeof(struct callback_head)); /* * Verify that "epoch_record" is at beginning of "struct * linux_epoch_record": */ CTASSERT(offsetof(struct linux_epoch_record, epoch_record) == 0); static ck_epoch_t linux_epoch; static struct linux_epoch_head linux_epoch_head; static DPCPU_DEFINE(struct linux_epoch_record, linux_epoch_record); static void linux_rcu_cleaner_func(void *, int); static void linux_rcu_runtime_init(void *arg __unused) { struct linux_epoch_head *head; int i; ck_epoch_init(&linux_epoch); head = &linux_epoch_head; mtx_init(&head->lock, "LRCU-HEAD", NULL, MTX_DEF); TASK_INIT(&head->task, 0, linux_rcu_cleaner_func, NULL); STAILQ_INIT(&head->cb_head); CPU_FOREACH(i) { struct linux_epoch_record *record; record = &DPCPU_ID_GET(i, linux_epoch_record); record->cpuid = i; ck_epoch_register(&linux_epoch, &record->epoch_record, NULL); TAILQ_INIT(&record->ts_head); } } SYSINIT(linux_rcu_runtime, SI_SUB_CPU, SI_ORDER_ANY, linux_rcu_runtime_init, NULL); static void linux_rcu_runtime_uninit(void *arg __unused) { struct linux_epoch_head *head; head = &linux_epoch_head; /* destroy head lock */ mtx_destroy(&head->lock); } SYSUNINIT(linux_rcu_runtime, SI_SUB_LOCK, SI_ORDER_SECOND, linux_rcu_runtime_uninit, NULL); static void linux_rcu_cleaner_func(void *context __unused, int pending __unused) { struct linux_epoch_head *head; struct callback_head *rcu; STAILQ_HEAD(, callback_head) tmp_head; linux_set_current(curthread); head = &linux_epoch_head; /* move current callbacks into own queue */ mtx_lock(&head->lock); STAILQ_INIT(&tmp_head); STAILQ_CONCAT(&tmp_head, &head->cb_head); mtx_unlock(&head->lock); /* synchronize */ linux_synchronize_rcu(); /* dispatch all callbacks, if any */ while ((rcu = STAILQ_FIRST(&tmp_head)) != NULL) { uintptr_t offset; STAILQ_REMOVE_HEAD(&tmp_head, entry); offset = (uintptr_t)rcu->func; if (offset < LINUX_KFREE_RCU_OFFSET_MAX) kfree((char *)rcu - offset); else rcu->func((struct rcu_head *)rcu); } } void linux_rcu_read_lock(void) { struct linux_epoch_record *record; struct task_struct *ts; if (RCU_SKIP()) return; /* * Pin thread to current CPU so that the unlock code gets the * same per-CPU epoch record: */ sched_pin(); record = &DPCPU_GET(linux_epoch_record); ts = current; /* * Use a critical section to prevent recursion inside * ck_epoch_begin(). Else this function supports recursion. */ critical_enter(); ck_epoch_begin(&record->epoch_record, NULL); ts->rcu_recurse++; if (ts->rcu_recurse == 1) TAILQ_INSERT_TAIL(&record->ts_head, ts, rcu_entry); critical_exit(); } void linux_rcu_read_unlock(void) { struct linux_epoch_record *record; struct task_struct *ts; if (RCU_SKIP()) return; record = &DPCPU_GET(linux_epoch_record); ts = current; /* * Use a critical section to prevent recursion inside * ck_epoch_end(). Else this function supports recursion. */ critical_enter(); ck_epoch_end(&record->epoch_record, NULL); ts->rcu_recurse--; if (ts->rcu_recurse == 0) TAILQ_REMOVE(&record->ts_head, ts, rcu_entry); critical_exit(); sched_unpin(); } static void linux_synchronize_rcu_cb(ck_epoch_t *epoch __unused, ck_epoch_record_t *epoch_record, void *arg __unused) { struct linux_epoch_record *record = container_of(epoch_record, struct linux_epoch_record, epoch_record); struct thread *td = curthread; struct task_struct *ts; /* check if blocked on the current CPU */ if (record->cpuid == PCPU_GET(cpuid)) { bool is_sleeping = 0; u_char prio = 0; /* * Find the lowest priority or sleeping thread which * is blocking synchronization on this CPU core. All * the threads in the queue are CPU-pinned and cannot * go anywhere while the current thread is locked. */ TAILQ_FOREACH(ts, &record->ts_head, rcu_entry) { if (ts->task_thread->td_priority > prio) prio = ts->task_thread->td_priority; is_sleeping |= (ts->task_thread->td_inhibitors != 0); } if (is_sleeping) { thread_unlock(td); pause("W", 1); thread_lock(td); } else { /* set new thread priority */ sched_prio(td, prio); /* task switch */ mi_switch(SW_VOL | SWT_RELINQUISH, NULL); /* * Release the thread lock while yielding to * allow other threads to acquire the lock * pointed to by TDQ_LOCKPTR(td). Else a * deadlock like situation might happen. */ thread_unlock(td); thread_lock(td); } } else { /* * To avoid spinning move execution to the other CPU * which is blocking synchronization. Set highest * thread priority so that code gets run. The thread * priority will be restored later. */ sched_prio(td, 0); sched_bind(td, record->cpuid); } } void linux_synchronize_rcu(void) { struct thread *td; int was_bound; int old_cpu; int old_pinned; u_char old_prio; if (RCU_SKIP()) return; WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "linux_synchronize_rcu() can sleep"); td = curthread; DROP_GIANT(); /* * Synchronizing RCU might change the CPU core this function * is running on. Save current values: */ thread_lock(td); old_cpu = PCPU_GET(cpuid); old_pinned = td->td_pinned; old_prio = td->td_priority; was_bound = sched_is_bound(td); sched_unbind(td); td->td_pinned = 0; sched_bind(td, old_cpu); ck_epoch_synchronize_wait(&linux_epoch, &linux_synchronize_rcu_cb, NULL); /* restore CPU binding, if any */ if (was_bound != 0) { sched_bind(td, old_cpu); } else { /* get thread back to initial CPU, if any */ if (old_pinned != 0) sched_bind(td, old_cpu); sched_unbind(td); } /* restore pinned after bind */ td->td_pinned = old_pinned; /* restore thread priority */ sched_prio(td, old_prio); thread_unlock(td); PICKUP_GIANT(); } void linux_rcu_barrier(void) { struct linux_epoch_head *head; linux_synchronize_rcu(); head = &linux_epoch_head; /* wait for callbacks to complete */ taskqueue_drain(taskqueue_fast, &head->task); } void linux_call_rcu(struct rcu_head *context, rcu_callback_t func) { struct callback_head *rcu = (struct callback_head *)context; struct linux_epoch_head *head = &linux_epoch_head; mtx_lock(&head->lock); rcu->func = func; STAILQ_INSERT_TAIL(&head->cb_head, rcu, entry); taskqueue_enqueue(taskqueue_fast, &head->task); mtx_unlock(&head->lock); } int init_srcu_struct(struct srcu_struct *srcu) { return (0); } void cleanup_srcu_struct(struct srcu_struct *srcu) { } int srcu_read_lock(struct srcu_struct *srcu) { linux_rcu_read_lock(); return (0); } void srcu_read_unlock(struct srcu_struct *srcu, int key __unused) { linux_rcu_read_unlock(); } void synchronize_srcu(struct srcu_struct *srcu) { linux_synchronize_rcu(); } void srcu_barrier(struct srcu_struct *srcu) { linux_rcu_barrier(); } Index: head/sys/dev/e1000/em_txrx.c =================================================================== --- head/sys/dev/e1000/em_txrx.c (revision 326983) +++ head/sys/dev/e1000/em_txrx.c (revision 326984) @@ -1,815 +1,815 @@ /*- - * Copyright (c) 2016-2017 Matt Macy + * Copyright (c) 2016-2017 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. 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. */ /* $FreeBSD$ */ #include "if_em.h" #ifdef RSS #include #include #endif #ifdef VERBOSE_DEBUG #define DPRINTF device_printf #else #define DPRINTF(...) #endif /********************************************************************* * Local Function prototypes *********************************************************************/ static int em_tso_setup(struct adapter *adapter, if_pkt_info_t pi, u32 *txd_upper, u32 *txd_lower); static int em_transmit_checksum_setup(struct adapter *adapter, if_pkt_info_t pi, u32 *txd_upper, u32 *txd_lower); static int em_isc_txd_encap(void *arg, if_pkt_info_t pi); static void em_isc_txd_flush(void *arg, uint16_t txqid, qidx_t pidx); static int em_isc_txd_credits_update(void *arg, uint16_t txqid, bool clear); static void em_isc_rxd_refill(void *arg, if_rxd_update_t iru); static void em_isc_rxd_flush(void *arg, uint16_t rxqid, uint8_t flid __unused, qidx_t pidx); static int em_isc_rxd_available(void *arg, uint16_t rxqid, qidx_t idx, qidx_t budget); static int em_isc_rxd_pkt_get(void *arg, if_rxd_info_t ri); static void lem_isc_rxd_refill(void *arg, if_rxd_update_t iru); static int lem_isc_rxd_available(void *arg, uint16_t rxqid, qidx_t idx, qidx_t budget); static int lem_isc_rxd_pkt_get(void *arg, if_rxd_info_t ri); static void lem_receive_checksum(int status, int errors, if_rxd_info_t ri); static void em_receive_checksum(uint32_t status, if_rxd_info_t ri); static int em_determine_rsstype(u32 pkt_info); extern int em_intr(void *arg); struct if_txrx em_txrx = { em_isc_txd_encap, em_isc_txd_flush, em_isc_txd_credits_update, em_isc_rxd_available, em_isc_rxd_pkt_get, em_isc_rxd_refill, em_isc_rxd_flush, em_intr }; struct if_txrx lem_txrx = { em_isc_txd_encap, em_isc_txd_flush, em_isc_txd_credits_update, lem_isc_rxd_available, lem_isc_rxd_pkt_get, lem_isc_rxd_refill, em_isc_rxd_flush, em_intr }; extern if_shared_ctx_t em_sctx; void em_dump_rs(struct adapter *adapter) { if_softc_ctx_t scctx = adapter->shared; struct em_tx_queue *que; struct tx_ring *txr; qidx_t i, ntxd, qid, cur; int16_t rs_cidx; uint8_t status; printf("\n"); ntxd = scctx->isc_ntxd[0]; for (qid = 0; qid < adapter->tx_num_queues; qid++) { que = &adapter->tx_queues[qid]; txr = &que->txr; rs_cidx = txr->tx_rs_cidx; if (rs_cidx != txr->tx_rs_pidx) { cur = txr->tx_rsq[rs_cidx]; status = txr->tx_base[cur].upper.fields.status; if (!(status & E1000_TXD_STAT_DD)) printf("qid[%d]->tx_rsq[%d]: %d clear ", qid, rs_cidx, cur); } else { rs_cidx = (rs_cidx-1)&(ntxd-1); cur = txr->tx_rsq[rs_cidx]; printf("qid[%d]->tx_rsq[rs_cidx-1=%d]: %d ", qid, rs_cidx, cur); } printf("cidx_prev=%d rs_pidx=%d ",txr->tx_cidx_processed, txr->tx_rs_pidx); for (i = 0; i < ntxd; i++) { if (txr->tx_base[i].upper.fields.status & E1000_TXD_STAT_DD) printf("%d set ", i); } printf("\n"); } } /********************************************************************** * * Setup work for hardware segmentation offload (TSO) on * adapters using advanced tx descriptors * **********************************************************************/ static int em_tso_setup(struct adapter *adapter, if_pkt_info_t pi, u32 *txd_upper, u32 *txd_lower) { if_softc_ctx_t scctx = adapter->shared; struct em_tx_queue *que = &adapter->tx_queues[pi->ipi_qsidx]; struct tx_ring *txr = &que->txr; struct e1000_context_desc *TXD; int cur, hdr_len; hdr_len = pi->ipi_ehdrlen + pi->ipi_ip_hlen + pi->ipi_tcp_hlen; *txd_lower = (E1000_TXD_CMD_DEXT | /* Extended descr type */ E1000_TXD_DTYP_D | /* Data descr type */ E1000_TXD_CMD_TSE); /* Do TSE on this packet */ /* IP and/or TCP header checksum calculation and insertion. */ *txd_upper = (E1000_TXD_POPTS_IXSM | E1000_TXD_POPTS_TXSM) << 8; cur = pi->ipi_pidx; TXD = (struct e1000_context_desc *)&txr->tx_base[cur]; /* * Start offset for header checksum calculation. * End offset for header checksum calculation. * Offset of place put the checksum. */ TXD->lower_setup.ip_fields.ipcss = pi->ipi_ehdrlen; TXD->lower_setup.ip_fields.ipcse = htole16(pi->ipi_ehdrlen + pi->ipi_ip_hlen - 1); TXD->lower_setup.ip_fields.ipcso = pi->ipi_ehdrlen + offsetof(struct ip, ip_sum); /* * Start offset for payload checksum calculation. * End offset for payload checksum calculation. * Offset of place to put the checksum. */ TXD->upper_setup.tcp_fields.tucss = pi->ipi_ehdrlen + pi->ipi_ip_hlen; TXD->upper_setup.tcp_fields.tucse = 0; TXD->upper_setup.tcp_fields.tucso = pi->ipi_ehdrlen + pi->ipi_ip_hlen + offsetof(struct tcphdr, th_sum); /* * Payload size per packet w/o any headers. * Length of all headers up to payload. */ TXD->tcp_seg_setup.fields.mss = htole16(pi->ipi_tso_segsz); TXD->tcp_seg_setup.fields.hdr_len = hdr_len; TXD->cmd_and_length = htole32(adapter->txd_cmd | E1000_TXD_CMD_DEXT | /* Extended descr */ E1000_TXD_CMD_TSE | /* TSE context */ E1000_TXD_CMD_IP | /* Do IP csum */ E1000_TXD_CMD_TCP | /* Do TCP checksum */ (pi->ipi_len - hdr_len)); /* Total len */ txr->tx_tso = TRUE; if (++cur == scctx->isc_ntxd[0]) { cur = 0; } DPRINTF(iflib_get_dev(adapter->ctx), "%s: pidx: %d cur: %d\n", __FUNCTION__, pi->ipi_pidx, cur); return (cur); } #define TSO_WORKAROUND 4 #define DONT_FORCE_CTX 1 /********************************************************************* * The offload context is protocol specific (TCP/UDP) and thus * only needs to be set when the protocol changes. The occasion * of a context change can be a performance detriment, and * might be better just disabled. The reason arises in the way * in which the controller supports pipelined requests from the * Tx data DMA. Up to four requests can be pipelined, and they may * belong to the same packet or to multiple packets. However all * requests for one packet are issued before a request is issued * for a subsequent packet and if a request for the next packet * requires a context change, that request will be stalled * until the previous request completes. This means setting up * a new context effectively disables pipelined Tx data DMA which * in turn greatly slow down performance to send small sized * frames. **********************************************************************/ static int em_transmit_checksum_setup(struct adapter *adapter, if_pkt_info_t pi, u32 *txd_upper, u32 *txd_lower) { struct e1000_context_desc *TXD = NULL; if_softc_ctx_t scctx = adapter->shared; struct em_tx_queue *que = &adapter->tx_queues[pi->ipi_qsidx]; struct tx_ring *txr = &que->txr; int csum_flags = pi->ipi_csum_flags; int cur, hdr_len; u32 cmd; cur = pi->ipi_pidx; hdr_len = pi->ipi_ehdrlen + pi->ipi_ip_hlen; cmd = adapter->txd_cmd; /* * The 82574L can only remember the *last* context used * regardless of queue that it was use for. We cannot reuse * contexts on this hardware platform and must generate a new * context every time. 82574L hardware spec, section 7.2.6, * second note. */ if (DONT_FORCE_CTX && adapter->tx_num_queues == 1 && txr->csum_lhlen == pi->ipi_ehdrlen && txr->csum_iphlen == pi->ipi_ip_hlen && txr->csum_flags == csum_flags) { /* * Same csum offload context as the previous packets; * just return. */ *txd_upper = txr->csum_txd_upper; *txd_lower = txr->csum_txd_lower; return (cur); } TXD = (struct e1000_context_desc *)&txr->tx_base[cur]; if (csum_flags & CSUM_IP) { *txd_upper |= E1000_TXD_POPTS_IXSM << 8; /* * Start offset for header checksum calculation. * End offset for header checksum calculation. * Offset of place to put the checksum. */ TXD->lower_setup.ip_fields.ipcss = pi->ipi_ehdrlen; TXD->lower_setup.ip_fields.ipcse = htole16(hdr_len); TXD->lower_setup.ip_fields.ipcso = pi->ipi_ehdrlen + offsetof(struct ip, ip_sum); cmd |= E1000_TXD_CMD_IP; } if (csum_flags & (CSUM_TCP|CSUM_UDP)) { uint8_t tucso; *txd_upper |= E1000_TXD_POPTS_TXSM << 8; *txd_lower = E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D; if (csum_flags & CSUM_TCP) { tucso = hdr_len + offsetof(struct tcphdr, th_sum); cmd |= E1000_TXD_CMD_TCP; } else tucso = hdr_len + offsetof(struct udphdr, uh_sum); TXD->upper_setup.tcp_fields.tucss = hdr_len; TXD->upper_setup.tcp_fields.tucse = htole16(0); TXD->upper_setup.tcp_fields.tucso = tucso; } txr->csum_lhlen = pi->ipi_ehdrlen; txr->csum_iphlen = pi->ipi_ip_hlen; txr->csum_flags = csum_flags; txr->csum_txd_upper = *txd_upper; txr->csum_txd_lower = *txd_lower; TXD->tcp_seg_setup.data = htole32(0); TXD->cmd_and_length = htole32(E1000_TXD_CMD_IFCS | E1000_TXD_CMD_DEXT | cmd); if (++cur == scctx->isc_ntxd[0]) { cur = 0; } DPRINTF(iflib_get_dev(adapter->ctx), "checksum_setup csum_flags=%x txd_upper=%x txd_lower=%x hdr_len=%d cmd=%x\n", csum_flags, *txd_upper, *txd_lower, hdr_len, cmd); return (cur); } static int em_isc_txd_encap(void *arg, if_pkt_info_t pi) { struct adapter *sc = arg; if_softc_ctx_t scctx = sc->shared; struct em_tx_queue *que = &sc->tx_queues[pi->ipi_qsidx]; struct tx_ring *txr = &que->txr; bus_dma_segment_t *segs = pi->ipi_segs; int nsegs = pi->ipi_nsegs; int csum_flags = pi->ipi_csum_flags; int i, j, first, pidx_last; u32 txd_flags, txd_upper = 0, txd_lower = 0; struct e1000_tx_desc *ctxd = NULL; bool do_tso, tso_desc; qidx_t ntxd; txd_flags = pi->ipi_flags & IPI_TX_INTR ? E1000_TXD_CMD_RS : 0; i = first = pi->ipi_pidx; do_tso = (csum_flags & CSUM_TSO); tso_desc = FALSE; ntxd = scctx->isc_ntxd[0]; /* * TSO Hardware workaround, if this packet is not * TSO, and is only a single descriptor long, and * it follows a TSO burst, then we need to add a * sentinel descriptor to prevent premature writeback. */ if ((!do_tso) && (txr->tx_tso == TRUE)) { if (nsegs == 1) tso_desc = TRUE; txr->tx_tso = FALSE; } /* Do hardware assists */ if (do_tso) { i = em_tso_setup(sc, pi, &txd_upper, &txd_lower); tso_desc = TRUE; } else if (csum_flags & EM_CSUM_OFFLOAD) { i = em_transmit_checksum_setup(sc, pi, &txd_upper, &txd_lower); } if (pi->ipi_mflags & M_VLANTAG) { /* Set the vlan id. */ txd_upper |= htole16(pi->ipi_vtag) << 16; /* Tell hardware to add tag */ txd_lower |= htole32(E1000_TXD_CMD_VLE); } DPRINTF(iflib_get_dev(sc->ctx), "encap: set up tx: nsegs=%d first=%d i=%d\n", nsegs, first, i); /* XXX adapter->pcix_82544 -- lem_fill_descriptors */ /* Set up our transmit descriptors */ for (j = 0; j < nsegs; j++) { bus_size_t seg_len; bus_addr_t seg_addr; uint32_t cmd; ctxd = &txr->tx_base[i]; seg_addr = segs[j].ds_addr; seg_len = segs[j].ds_len; cmd = E1000_TXD_CMD_IFCS | sc->txd_cmd; /* * TSO Workaround: * If this is the last descriptor, we want to * split it so we have a small final sentinel */ if (tso_desc && (j == (nsegs - 1)) && (seg_len > 8)) { seg_len -= TSO_WORKAROUND; ctxd->buffer_addr = htole64(seg_addr); ctxd->lower.data = htole32(cmd | txd_lower | seg_len); ctxd->upper.data = htole32(txd_upper); if (++i == scctx->isc_ntxd[0]) i = 0; /* Now make the sentinel */ ctxd = &txr->tx_base[i]; ctxd->buffer_addr = htole64(seg_addr + seg_len); ctxd->lower.data = htole32(cmd | txd_lower | TSO_WORKAROUND); ctxd->upper.data = htole32(txd_upper); pidx_last = i; if (++i == scctx->isc_ntxd[0]) i = 0; DPRINTF(iflib_get_dev(sc->ctx), "TSO path pidx_last=%d i=%d ntxd[0]=%d\n", pidx_last, i, scctx->isc_ntxd[0]); } else { ctxd->buffer_addr = htole64(seg_addr); ctxd->lower.data = htole32(cmd | txd_lower | seg_len); ctxd->upper.data = htole32(txd_upper); pidx_last = i; if (++i == scctx->isc_ntxd[0]) i = 0; DPRINTF(iflib_get_dev(sc->ctx), "pidx_last=%d i=%d ntxd[0]=%d\n", pidx_last, i, scctx->isc_ntxd[0]); } } /* * Last Descriptor of Packet * needs End Of Packet (EOP) * and Report Status (RS) */ if (txd_flags) { txr->tx_rsq[txr->tx_rs_pidx] = pidx_last; DPRINTF(iflib_get_dev(sc->ctx), "setting to RS on %d rs_pidx %d first: %d\n", pidx_last, txr->tx_rs_pidx, first); txr->tx_rs_pidx = (txr->tx_rs_pidx+1) & (ntxd-1); MPASS(txr->tx_rs_pidx != txr->tx_rs_cidx); } ctxd->lower.data |= htole32(E1000_TXD_CMD_EOP | txd_flags); DPRINTF(iflib_get_dev(sc->ctx), "tx_buffers[%d]->eop = %d ipi_new_pidx=%d\n", first, pidx_last, i); pi->ipi_new_pidx = i; return (0); } static void em_isc_txd_flush(void *arg, uint16_t txqid, qidx_t pidx) { struct adapter *adapter = arg; struct em_tx_queue *que = &adapter->tx_queues[txqid]; struct tx_ring *txr = &que->txr; E1000_WRITE_REG(&adapter->hw, E1000_TDT(txr->me), pidx); } static int em_isc_txd_credits_update(void *arg, uint16_t txqid, bool clear) { struct adapter *adapter = arg; if_softc_ctx_t scctx = adapter->shared; struct em_tx_queue *que = &adapter->tx_queues[txqid]; struct tx_ring *txr = &que->txr; qidx_t processed = 0; int updated; qidx_t cur, prev, ntxd, rs_cidx; int32_t delta; uint8_t status; rs_cidx = txr->tx_rs_cidx; if (rs_cidx == txr->tx_rs_pidx) return (0); cur = txr->tx_rsq[rs_cidx]; MPASS(cur != QIDX_INVALID); status = txr->tx_base[cur].upper.fields.status; updated = !!(status & E1000_TXD_STAT_DD); if (clear == false || updated == 0) return (updated); prev = txr->tx_cidx_processed; ntxd = scctx->isc_ntxd[0]; do { delta = (int32_t)cur - (int32_t)prev; MPASS(prev == 0 || delta != 0); if (delta < 0) delta += ntxd; DPRINTF(iflib_get_dev(adapter->ctx), "%s: cidx_processed=%u cur=%u clear=%d delta=%d\n", __FUNCTION__, prev, cur, clear, delta); processed += delta; prev = cur; rs_cidx = (rs_cidx + 1) & (ntxd-1); if (rs_cidx == txr->tx_rs_pidx) break; cur = txr->tx_rsq[rs_cidx]; MPASS(cur != QIDX_INVALID); status = txr->tx_base[cur].upper.fields.status; } while ((status & E1000_TXD_STAT_DD)); txr->tx_rs_cidx = rs_cidx; txr->tx_cidx_processed = prev; return(processed); } static void lem_isc_rxd_refill(void *arg, if_rxd_update_t iru) { struct adapter *sc = arg; if_softc_ctx_t scctx = sc->shared; struct em_rx_queue *que = &sc->rx_queues[iru->iru_qsidx]; struct rx_ring *rxr = &que->rxr; struct e1000_rx_desc *rxd; uint64_t *paddrs; uint32_t next_pidx, pidx; uint16_t count; int i; paddrs = iru->iru_paddrs; pidx = iru->iru_pidx; count = iru->iru_count; for (i = 0, next_pidx = pidx; i < count; i++) { rxd = (struct e1000_rx_desc *)&rxr->rx_base[next_pidx]; rxd->buffer_addr = htole64(paddrs[i]); /* status bits must be cleared */ rxd->status = 0; if (++next_pidx == scctx->isc_nrxd[0]) next_pidx = 0; } } static void em_isc_rxd_refill(void *arg, if_rxd_update_t iru) { struct adapter *sc = arg; if_softc_ctx_t scctx = sc->shared; uint16_t rxqid = iru->iru_qsidx; struct em_rx_queue *que = &sc->rx_queues[rxqid]; struct rx_ring *rxr = &que->rxr; union e1000_rx_desc_extended *rxd; uint64_t *paddrs; uint32_t next_pidx, pidx; uint16_t count; int i; paddrs = iru->iru_paddrs; pidx = iru->iru_pidx; count = iru->iru_count; for (i = 0, next_pidx = pidx; i < count; i++) { rxd = &rxr->rx_base[next_pidx]; rxd->read.buffer_addr = htole64(paddrs[i]); /* DD bits must be cleared */ rxd->wb.upper.status_error = 0; if (++next_pidx == scctx->isc_nrxd[0]) next_pidx = 0; } } static void em_isc_rxd_flush(void *arg, uint16_t rxqid, uint8_t flid __unused, qidx_t pidx) { struct adapter *sc = arg; struct em_rx_queue *que = &sc->rx_queues[rxqid]; struct rx_ring *rxr = &que->rxr; E1000_WRITE_REG(&sc->hw, E1000_RDT(rxr->me), pidx); } static int lem_isc_rxd_available(void *arg, uint16_t rxqid, qidx_t idx, qidx_t budget) { struct adapter *sc = arg; if_softc_ctx_t scctx = sc->shared; struct em_rx_queue *que = &sc->rx_queues[rxqid]; struct rx_ring *rxr = &que->rxr; struct e1000_rx_desc *rxd; u32 staterr = 0; int cnt, i; if (budget == 1) { rxd = (struct e1000_rx_desc *)&rxr->rx_base[idx]; staterr = rxd->status; return (staterr & E1000_RXD_STAT_DD); } for (cnt = 0, i = idx; cnt < scctx->isc_nrxd[0] && cnt <= budget;) { rxd = (struct e1000_rx_desc *)&rxr->rx_base[i]; staterr = rxd->status; if ((staterr & E1000_RXD_STAT_DD) == 0) break; if (++i == scctx->isc_nrxd[0]) i = 0; if (staterr & E1000_RXD_STAT_EOP) cnt++; } return (cnt); } static int em_isc_rxd_available(void *arg, uint16_t rxqid, qidx_t idx, qidx_t budget) { struct adapter *sc = arg; if_softc_ctx_t scctx = sc->shared; struct em_rx_queue *que = &sc->rx_queues[rxqid]; struct rx_ring *rxr = &que->rxr; union e1000_rx_desc_extended *rxd; u32 staterr = 0; int cnt, i; if (budget == 1) { rxd = &rxr->rx_base[idx]; staterr = le32toh(rxd->wb.upper.status_error); return (staterr & E1000_RXD_STAT_DD); } for (cnt = 0, i = idx; cnt < scctx->isc_nrxd[0] && cnt <= budget;) { rxd = &rxr->rx_base[i]; staterr = le32toh(rxd->wb.upper.status_error); if ((staterr & E1000_RXD_STAT_DD) == 0) break; if (++i == scctx->isc_nrxd[0]) { i = 0; } if (staterr & E1000_RXD_STAT_EOP) cnt++; } return (cnt); } static int lem_isc_rxd_pkt_get(void *arg, if_rxd_info_t ri) { struct adapter *adapter = arg; if_softc_ctx_t scctx = adapter->shared; struct em_rx_queue *que = &adapter->rx_queues[ri->iri_qsidx]; struct rx_ring *rxr = &que->rxr; struct e1000_rx_desc *rxd; u16 len; u32 status, errors; bool eop; int i, cidx; status = errors = i = 0; cidx = ri->iri_cidx; do { rxd = (struct e1000_rx_desc *)&rxr->rx_base[cidx]; status = rxd->status; errors = rxd->errors; /* Error Checking then decrement count */ MPASS ((status & E1000_RXD_STAT_DD) != 0); len = le16toh(rxd->length); ri->iri_len += len; eop = (status & E1000_RXD_STAT_EOP) != 0; /* Make sure bad packets are discarded */ if (errors & E1000_RXD_ERR_FRAME_ERR_MASK) { adapter->dropped_pkts++; /* XXX fixup if common */ return (EBADMSG); } ri->iri_frags[i].irf_flid = 0; ri->iri_frags[i].irf_idx = cidx; ri->iri_frags[i].irf_len = len; /* Zero out the receive descriptors status. */ rxd->status = 0; if (++cidx == scctx->isc_nrxd[0]) cidx = 0; i++; } while (!eop); /* XXX add a faster way to look this up */ if (adapter->hw.mac.type >= e1000_82543 && !(status & E1000_RXD_STAT_IXSM)) lem_receive_checksum(status, errors, ri); if (status & E1000_RXD_STAT_VP) { ri->iri_vtag = le16toh(rxd->special); ri->iri_flags |= M_VLANTAG; } ri->iri_nfrags = i; return (0); } static int em_isc_rxd_pkt_get(void *arg, if_rxd_info_t ri) { struct adapter *adapter = arg; if_softc_ctx_t scctx = adapter->shared; struct em_rx_queue *que = &adapter->rx_queues[ri->iri_qsidx]; struct rx_ring *rxr = &que->rxr; union e1000_rx_desc_extended *rxd; u16 len; u32 pkt_info; u32 staterr = 0; bool eop; int i, cidx, vtag; i = vtag = 0; cidx = ri->iri_cidx; do { rxd = &rxr->rx_base[cidx]; staterr = le32toh(rxd->wb.upper.status_error); pkt_info = le32toh(rxd->wb.lower.mrq); /* Error Checking then decrement count */ MPASS ((staterr & E1000_RXD_STAT_DD) != 0); len = le16toh(rxd->wb.upper.length); ri->iri_len += len; eop = (staterr & E1000_RXD_STAT_EOP) != 0; /* Make sure bad packets are discarded */ if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) { adapter->dropped_pkts++; return EBADMSG; } ri->iri_frags[i].irf_flid = 0; ri->iri_frags[i].irf_idx = cidx; ri->iri_frags[i].irf_len = len; /* Zero out the receive descriptors status. */ rxd->wb.upper.status_error &= htole32(~0xFF); if (++cidx == scctx->isc_nrxd[0]) cidx = 0; i++; } while (!eop); /* XXX add a faster way to look this up */ if (adapter->hw.mac.type >= e1000_82543) em_receive_checksum(staterr, ri); if (staterr & E1000_RXD_STAT_VP) { vtag = le16toh(rxd->wb.upper.vlan); } ri->iri_vtag = vtag; if (vtag) ri->iri_flags |= M_VLANTAG; ri->iri_flowid = le32toh(rxd->wb.lower.hi_dword.rss); ri->iri_rsstype = em_determine_rsstype(pkt_info); ri->iri_nfrags = i; return (0); } /********************************************************************* * * Verify that the hardware indicated that the checksum is valid. * Inform the stack about the status of checksum so that stack * doesn't spend time verifying the checksum. * *********************************************************************/ static void lem_receive_checksum(int status, int errors, if_rxd_info_t ri) { /* Did it pass? */ if (status & E1000_RXD_STAT_IPCS && !(errors & E1000_RXD_ERR_IPE)) ri->iri_csum_flags = (CSUM_IP_CHECKED|CSUM_IP_VALID); if (status & E1000_RXD_STAT_TCPCS) { /* Did it pass? */ if (!(errors & E1000_RXD_ERR_TCPE)) { ri->iri_csum_flags |= (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); ri->iri_csum_data = htons(0xffff); } } } /******************************************************************** * * Parse the packet type to determine the appropriate hash * ******************************************************************/ static int em_determine_rsstype(u32 pkt_info) { switch (pkt_info & E1000_RXDADV_RSSTYPE_MASK) { case E1000_RXDADV_RSSTYPE_IPV4_TCP: return M_HASHTYPE_RSS_TCP_IPV4; case E1000_RXDADV_RSSTYPE_IPV4: return M_HASHTYPE_RSS_IPV4; case E1000_RXDADV_RSSTYPE_IPV6_TCP: return M_HASHTYPE_RSS_TCP_IPV6; case E1000_RXDADV_RSSTYPE_IPV6_EX: return M_HASHTYPE_RSS_IPV6_EX; case E1000_RXDADV_RSSTYPE_IPV6: return M_HASHTYPE_RSS_IPV6; case E1000_RXDADV_RSSTYPE_IPV6_TCP_EX: return M_HASHTYPE_RSS_TCP_IPV6_EX; default: return M_HASHTYPE_OPAQUE; } } static void em_receive_checksum(uint32_t status, if_rxd_info_t ri) { ri->iri_csum_flags = 0; /* Ignore Checksum bit is set */ if (status & E1000_RXD_STAT_IXSM) return; /* If the IP checksum exists and there is no IP Checksum error */ if ((status & (E1000_RXD_STAT_IPCS | E1000_RXDEXT_STATERR_IPE)) == E1000_RXD_STAT_IPCS) { ri->iri_csum_flags = (CSUM_IP_CHECKED | CSUM_IP_VALID); } /* TCP or UDP checksum */ if ((status & (E1000_RXD_STAT_TCPCS | E1000_RXDEXT_STATERR_TCPE)) == E1000_RXD_STAT_TCPCS) { ri->iri_csum_flags |= (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); ri->iri_csum_data = htons(0xffff); } if (status & E1000_RXD_STAT_UDPCS) { ri->iri_csum_flags |= (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); ri->iri_csum_data = htons(0xffff); } } Index: head/sys/dev/e1000/if_em.c =================================================================== --- head/sys/dev/e1000/if_em.c (revision 326983) +++ head/sys/dev/e1000/if_em.c (revision 326984) @@ -1,4545 +1,4545 @@ /*- * SPDX-License-Identifier: BSD-2-Clause * - * Copyright (c) 2016 Matt Macy + * Copyright (c) 2016 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. 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. */ /* $FreeBSD$ */ #include "if_em.h" #include #include #define em_mac_min e1000_82547 #define igb_mac_min e1000_82575 /********************************************************************* * Driver version: *********************************************************************/ char em_driver_version[] = "7.6.1-k"; /********************************************************************* * PCI Device ID Table * * Used by probe to select devices to load on * Last field stores an index into e1000_strings * Last entry must be all 0s * * { Vendor ID, Device ID, SubVendor ID, SubDevice ID, String Index } *********************************************************************/ static pci_vendor_info_t em_vendor_info_array[] = { /* Intel(R) PRO/1000 Network Connection - Legacy em*/ PVID(0x8086, E1000_DEV_ID_82540EM, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82540EM_LOM, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82540EP, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82540EP_LOM, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82540EP_LP, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82541EI, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82541ER, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82541ER_LOM, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82541EI_MOBILE, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82541GI, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82541GI_LF, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82541GI_MOBILE, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82542, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82543GC_FIBER, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82543GC_COPPER, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82544EI_COPPER, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82544EI_FIBER, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82544GC_COPPER, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82544GC_LOM, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82545EM_COPPER, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82545EM_FIBER, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82545GM_COPPER, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82545GM_FIBER, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82545GM_SERDES, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82546EB_COPPER, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82546EB_FIBER, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82546EB_QUAD_COPPER, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82546GB_COPPER, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82546GB_FIBER, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82546GB_SERDES, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82546GB_PCIE, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82546GB_QUAD_COPPER, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82546GB_QUAD_COPPER_KSP3, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82547EI, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82547EI_MOBILE, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82547GI, "Intel(R) PRO/1000 Network Connection"), /* Intel(R) PRO/1000 Network Connection - em */ PVID(0x8086, E1000_DEV_ID_82571EB_COPPER, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82571EB_FIBER, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82571EB_SERDES, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82571EB_SERDES_DUAL, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82571EB_SERDES_QUAD, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82571EB_QUAD_COPPER, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82571EB_QUAD_COPPER_LP, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82571EB_QUAD_FIBER, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82571PT_QUAD_COPPER, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82572EI, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82572EI_COPPER, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82572EI_FIBER, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82572EI_SERDES, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82573E, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82573E_IAMT, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82573L, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82583V, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_80003ES2LAN_COPPER_SPT, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_80003ES2LAN_SERDES_SPT, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_80003ES2LAN_COPPER_DPT, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_80003ES2LAN_SERDES_DPT, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_ICH8_IGP_M_AMT, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_ICH8_IGP_AMT, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_ICH8_IGP_C, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_ICH8_IFE, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_ICH8_IFE_GT, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_ICH8_IFE_G, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_ICH8_IGP_M, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_ICH8_82567V_3, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_ICH9_IGP_M_AMT, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_ICH9_IGP_AMT, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_ICH9_IGP_C, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_ICH9_IGP_M, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_ICH9_IGP_M_V, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_ICH9_IFE, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_ICH9_IFE_GT, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_ICH9_IFE_G, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_ICH9_BM, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82574L, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_82574LA, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_ICH10_R_BM_LM, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_ICH10_R_BM_LF, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_ICH10_R_BM_V, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_ICH10_D_BM_LM, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_ICH10_D_BM_LF, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_ICH10_D_BM_V, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_PCH_M_HV_LM, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_PCH_M_HV_LC, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_PCH_D_HV_DM, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_PCH_D_HV_DC, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_PCH2_LV_LM, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_PCH2_LV_V, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_PCH_LPT_I217_LM, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_PCH_LPT_I217_V, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_PCH_LPTLP_I218_LM, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_PCH_LPTLP_I218_V, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_PCH_I218_LM2, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_PCH_I218_V2, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_PCH_I218_LM3, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_PCH_I218_V3, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_PCH_SPT_I219_LM, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_PCH_SPT_I219_V, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_PCH_SPT_I219_LM2, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_PCH_SPT_I219_V2, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_PCH_LBG_I219_LM3, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_PCH_SPT_I219_LM4, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_PCH_SPT_I219_V4, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_PCH_SPT_I219_LM5, "Intel(R) PRO/1000 Network Connection"), PVID(0x8086, E1000_DEV_ID_PCH_SPT_I219_V5, "Intel(R) PRO/1000 Network Connection"), /* required last entry */ PVID_END }; static pci_vendor_info_t igb_vendor_info_array[] = { /* Intel(R) PRO/1000 Network Connection - igb */ PVID(0x8086, E1000_DEV_ID_82575EB_COPPER, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_82575EB_FIBER_SERDES, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_82575GB_QUAD_COPPER, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_82576, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_82576_NS, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_82576_NS_SERDES, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_82576_FIBER, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_82576_SERDES, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_82576_SERDES_QUAD, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_82576_QUAD_COPPER, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_82576_QUAD_COPPER_ET2, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_82576_VF, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_82580_COPPER, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_82580_FIBER, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_82580_SERDES, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_82580_SGMII, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_82580_COPPER_DUAL, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_82580_QUAD_FIBER, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_DH89XXCC_SERDES, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_DH89XXCC_SGMII, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_DH89XXCC_SFP, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_DH89XXCC_BACKPLANE, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_I350_COPPER, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_I350_FIBER, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_I350_SERDES, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_I350_SGMII, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_I350_VF, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_I210_COPPER, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_I210_COPPER_IT, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_I210_COPPER_OEM1, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_I210_COPPER_FLASHLESS, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_I210_SERDES_FLASHLESS, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_I210_FIBER, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_I210_SERDES, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_I210_SGMII, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_I211_COPPER, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_I354_BACKPLANE_1GBPS, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_I354_BACKPLANE_2_5GBPS, "Intel(R) PRO/1000 PCI-Express Network Driver"), PVID(0x8086, E1000_DEV_ID_I354_SGMII, "Intel(R) PRO/1000 PCI-Express Network Driver"), /* required last entry */ PVID_END }; /********************************************************************* * Function prototypes *********************************************************************/ static void *em_register(device_t dev); static void *igb_register(device_t dev); static int em_if_attach_pre(if_ctx_t ctx); static int em_if_attach_post(if_ctx_t ctx); static int em_if_detach(if_ctx_t ctx); static int em_if_shutdown(if_ctx_t ctx); static int em_if_suspend(if_ctx_t ctx); static int em_if_resume(if_ctx_t ctx); static int em_if_tx_queues_alloc(if_ctx_t ctx, caddr_t *vaddrs, uint64_t *paddrs, int ntxqs, int ntxqsets); static int em_if_rx_queues_alloc(if_ctx_t ctx, caddr_t *vaddrs, uint64_t *paddrs, int nrxqs, int nrxqsets); static void em_if_queues_free(if_ctx_t ctx); static uint64_t em_if_get_counter(if_ctx_t, ift_counter); static void em_if_init(if_ctx_t ctx); static void em_if_stop(if_ctx_t ctx); static void em_if_media_status(if_ctx_t, struct ifmediareq *); static int em_if_media_change(if_ctx_t ctx); static int em_if_mtu_set(if_ctx_t ctx, uint32_t mtu); static void em_if_timer(if_ctx_t ctx, uint16_t qid); static void em_if_vlan_register(if_ctx_t ctx, u16 vtag); static void em_if_vlan_unregister(if_ctx_t ctx, u16 vtag); static void em_identify_hardware(if_ctx_t ctx); static int em_allocate_pci_resources(if_ctx_t ctx); static void em_free_pci_resources(if_ctx_t ctx); static void em_reset(if_ctx_t ctx); static int em_setup_interface(if_ctx_t ctx); static int em_setup_msix(if_ctx_t ctx); static void em_initialize_transmit_unit(if_ctx_t ctx); static void em_initialize_receive_unit(if_ctx_t ctx); static void em_if_enable_intr(if_ctx_t ctx); static void em_if_disable_intr(if_ctx_t ctx); static int em_if_rx_queue_intr_enable(if_ctx_t ctx, uint16_t rxqid); static int em_if_tx_queue_intr_enable(if_ctx_t ctx, uint16_t txqid); static void em_if_multi_set(if_ctx_t ctx); static void em_if_update_admin_status(if_ctx_t ctx); static void em_if_debug(if_ctx_t ctx); static void em_update_stats_counters(struct adapter *); static void em_add_hw_stats(struct adapter *adapter); static int em_if_set_promisc(if_ctx_t ctx, int flags); static void em_setup_vlan_hw_support(struct adapter *); static int em_sysctl_nvm_info(SYSCTL_HANDLER_ARGS); static void em_print_nvm_info(struct adapter *); static int em_sysctl_debug_info(SYSCTL_HANDLER_ARGS); static int em_get_rs(SYSCTL_HANDLER_ARGS); static void em_print_debug_info(struct adapter *); static int em_is_valid_ether_addr(u8 *); static int em_sysctl_int_delay(SYSCTL_HANDLER_ARGS); static void em_add_int_delay_sysctl(struct adapter *, const char *, const char *, struct em_int_delay_info *, int, int); /* Management and WOL Support */ static void em_init_manageability(struct adapter *); static void em_release_manageability(struct adapter *); static void em_get_hw_control(struct adapter *); static void em_release_hw_control(struct adapter *); static void em_get_wakeup(if_ctx_t ctx); static void em_enable_wakeup(if_ctx_t ctx); static int em_enable_phy_wakeup(struct adapter *); static void em_disable_aspm(struct adapter *); int em_intr(void *arg); static void em_disable_promisc(if_ctx_t ctx); /* MSIX handlers */ static int em_if_msix_intr_assign(if_ctx_t, int); static int em_msix_link(void *); static void em_handle_link(void *context); static void em_enable_vectors_82574(if_ctx_t); static int em_set_flowcntl(SYSCTL_HANDLER_ARGS); static int em_sysctl_eee(SYSCTL_HANDLER_ARGS); static void em_if_led_func(if_ctx_t ctx, int onoff); static int em_get_regs(SYSCTL_HANDLER_ARGS); static void lem_smartspeed(struct adapter *adapter); static void igb_configure_queues(struct adapter *adapter); /********************************************************************* * FreeBSD Device Interface Entry Points *********************************************************************/ static device_method_t em_methods[] = { /* Device interface */ DEVMETHOD(device_register, em_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 device_method_t igb_methods[] = { /* Device interface */ DEVMETHOD(device_register, igb_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 em_driver = { "em", em_methods, sizeof(struct adapter), }; static devclass_t em_devclass; DRIVER_MODULE(em, pci, em_driver, em_devclass, 0, 0); MODULE_DEPEND(em, pci, 1, 1, 1); MODULE_DEPEND(em, ether, 1, 1, 1); MODULE_DEPEND(em, iflib, 1, 1, 1); IFLIB_PNP_INFO(pci, em, em_vendor_info_array); static driver_t igb_driver = { "igb", igb_methods, sizeof(struct adapter), }; static devclass_t igb_devclass; DRIVER_MODULE(igb, pci, igb_driver, igb_devclass, 0, 0); MODULE_DEPEND(igb, pci, 1, 1, 1); MODULE_DEPEND(igb, ether, 1, 1, 1); MODULE_DEPEND(igb, iflib, 1, 1, 1); IFLIB_PNP_INFO(pci, igb, igb_vendor_info_array); static device_method_t em_if_methods[] = { DEVMETHOD(ifdi_attach_pre, em_if_attach_pre), DEVMETHOD(ifdi_attach_post, em_if_attach_post), DEVMETHOD(ifdi_detach, em_if_detach), DEVMETHOD(ifdi_shutdown, em_if_shutdown), DEVMETHOD(ifdi_suspend, em_if_suspend), DEVMETHOD(ifdi_resume, em_if_resume), DEVMETHOD(ifdi_init, em_if_init), DEVMETHOD(ifdi_stop, em_if_stop), DEVMETHOD(ifdi_msix_intr_assign, em_if_msix_intr_assign), DEVMETHOD(ifdi_intr_enable, em_if_enable_intr), DEVMETHOD(ifdi_intr_disable, em_if_disable_intr), DEVMETHOD(ifdi_tx_queues_alloc, em_if_tx_queues_alloc), DEVMETHOD(ifdi_rx_queues_alloc, em_if_rx_queues_alloc), DEVMETHOD(ifdi_queues_free, em_if_queues_free), DEVMETHOD(ifdi_update_admin_status, em_if_update_admin_status), DEVMETHOD(ifdi_multi_set, em_if_multi_set), DEVMETHOD(ifdi_media_status, em_if_media_status), DEVMETHOD(ifdi_media_change, em_if_media_change), DEVMETHOD(ifdi_mtu_set, em_if_mtu_set), DEVMETHOD(ifdi_promisc_set, em_if_set_promisc), DEVMETHOD(ifdi_timer, em_if_timer), DEVMETHOD(ifdi_vlan_register, em_if_vlan_register), DEVMETHOD(ifdi_vlan_unregister, em_if_vlan_unregister), DEVMETHOD(ifdi_get_counter, em_if_get_counter), DEVMETHOD(ifdi_led_func, em_if_led_func), DEVMETHOD(ifdi_rx_queue_intr_enable, em_if_rx_queue_intr_enable), DEVMETHOD(ifdi_tx_queue_intr_enable, em_if_tx_queue_intr_enable), DEVMETHOD(ifdi_debug, em_if_debug), DEVMETHOD_END }; /* * note that if (adapter->msix_mem) is replaced by: * if (adapter->intr_type == IFLIB_INTR_MSIX) */ static driver_t em_if_driver = { "em_if", em_if_methods, sizeof(struct adapter) }; /********************************************************************* * Tunable default values. *********************************************************************/ #define EM_TICKS_TO_USECS(ticks) ((1024 * (ticks) + 500) / 1000) #define EM_USECS_TO_TICKS(usecs) ((1000 * (usecs) + 512) / 1024) #define M_TSO_LEN 66 #define MAX_INTS_PER_SEC 8000 #define DEFAULT_ITR (1000000000/(MAX_INTS_PER_SEC * 256)) /* Allow common code without TSO */ #ifndef CSUM_TSO #define CSUM_TSO 0 #endif #define TSO_WORKAROUND 4 static SYSCTL_NODE(_hw, OID_AUTO, em, CTLFLAG_RD, 0, "EM driver parameters"); static int em_disable_crc_stripping = 0; SYSCTL_INT(_hw_em, OID_AUTO, disable_crc_stripping, CTLFLAG_RDTUN, &em_disable_crc_stripping, 0, "Disable CRC Stripping"); static int em_tx_int_delay_dflt = EM_TICKS_TO_USECS(EM_TIDV); static int em_rx_int_delay_dflt = EM_TICKS_TO_USECS(EM_RDTR); SYSCTL_INT(_hw_em, OID_AUTO, tx_int_delay, CTLFLAG_RDTUN, &em_tx_int_delay_dflt, 0, "Default transmit interrupt delay in usecs"); SYSCTL_INT(_hw_em, OID_AUTO, rx_int_delay, CTLFLAG_RDTUN, &em_rx_int_delay_dflt, 0, "Default receive interrupt delay in usecs"); static int em_tx_abs_int_delay_dflt = EM_TICKS_TO_USECS(EM_TADV); static int em_rx_abs_int_delay_dflt = EM_TICKS_TO_USECS(EM_RADV); SYSCTL_INT(_hw_em, OID_AUTO, tx_abs_int_delay, CTLFLAG_RDTUN, &em_tx_abs_int_delay_dflt, 0, "Default transmit interrupt delay limit in usecs"); SYSCTL_INT(_hw_em, OID_AUTO, rx_abs_int_delay, CTLFLAG_RDTUN, &em_rx_abs_int_delay_dflt, 0, "Default receive interrupt delay limit in usecs"); static int em_smart_pwr_down = FALSE; SYSCTL_INT(_hw_em, OID_AUTO, smart_pwr_down, CTLFLAG_RDTUN, &em_smart_pwr_down, 0, "Set to true to leave smart power down enabled on newer adapters"); /* Controls whether promiscuous also shows bad packets */ static int em_debug_sbp = TRUE; SYSCTL_INT(_hw_em, OID_AUTO, sbp, CTLFLAG_RDTUN, &em_debug_sbp, 0, "Show bad packets in promiscuous mode"); /* How many packets rxeof tries to clean at a time */ static int em_rx_process_limit = 100; SYSCTL_INT(_hw_em, OID_AUTO, rx_process_limit, CTLFLAG_RDTUN, &em_rx_process_limit, 0, "Maximum number of received packets to process " "at a time, -1 means unlimited"); /* Energy efficient ethernet - default to OFF */ static int eee_setting = 1; SYSCTL_INT(_hw_em, OID_AUTO, eee_setting, CTLFLAG_RDTUN, &eee_setting, 0, "Enable Energy Efficient Ethernet"); /* ** Tuneable Interrupt rate */ static int em_max_interrupt_rate = 8000; SYSCTL_INT(_hw_em, OID_AUTO, max_interrupt_rate, CTLFLAG_RDTUN, &em_max_interrupt_rate, 0, "Maximum interrupts per second"); /* Global used in WOL setup with multiport cards */ static int global_quad_port_a = 0; extern struct if_txrx igb_txrx; extern struct if_txrx em_txrx; extern struct if_txrx lem_txrx; static struct if_shared_ctx em_sctx_init = { .isc_magic = IFLIB_MAGIC, .isc_q_align = PAGE_SIZE, .isc_tx_maxsize = EM_TSO_SIZE, .isc_tx_maxsegsize = PAGE_SIZE, .isc_rx_maxsize = MJUM9BYTES, .isc_rx_nsegments = 1, .isc_rx_maxsegsize = MJUM9BYTES, .isc_nfl = 1, .isc_nrxqs = 1, .isc_ntxqs = 1, .isc_admin_intrcnt = 1, .isc_vendor_info = em_vendor_info_array, .isc_driver_version = em_driver_version, .isc_driver = &em_if_driver, .isc_flags = IFLIB_NEED_SCRATCH | IFLIB_TSO_INIT_IP | IFLIB_NEED_ZERO_CSUM, .isc_nrxd_min = {EM_MIN_RXD}, .isc_ntxd_min = {EM_MIN_TXD}, .isc_nrxd_max = {EM_MAX_RXD}, .isc_ntxd_max = {EM_MAX_TXD}, .isc_nrxd_default = {EM_DEFAULT_RXD}, .isc_ntxd_default = {EM_DEFAULT_TXD}, }; if_shared_ctx_t em_sctx = &em_sctx_init; static struct if_shared_ctx igb_sctx_init = { .isc_magic = IFLIB_MAGIC, .isc_q_align = PAGE_SIZE, .isc_tx_maxsize = EM_TSO_SIZE, .isc_tx_maxsegsize = PAGE_SIZE, .isc_rx_maxsize = MJUM9BYTES, .isc_rx_nsegments = 1, .isc_rx_maxsegsize = MJUM9BYTES, .isc_nfl = 1, .isc_nrxqs = 1, .isc_ntxqs = 1, .isc_admin_intrcnt = 1, .isc_vendor_info = igb_vendor_info_array, .isc_driver_version = em_driver_version, .isc_driver = &em_if_driver, .isc_flags = IFLIB_NEED_SCRATCH | IFLIB_TSO_INIT_IP | IFLIB_NEED_ZERO_CSUM, .isc_nrxd_min = {EM_MIN_RXD}, .isc_ntxd_min = {EM_MIN_TXD}, .isc_nrxd_max = {IGB_MAX_RXD}, .isc_ntxd_max = {IGB_MAX_TXD}, .isc_nrxd_default = {EM_DEFAULT_RXD}, .isc_ntxd_default = {EM_DEFAULT_TXD}, }; if_shared_ctx_t igb_sctx = &igb_sctx_init; /***************************************************************** * * Dump Registers * ****************************************************************/ #define IGB_REGS_LEN 739 static int em_get_regs(SYSCTL_HANDLER_ARGS) { struct adapter *adapter = (struct adapter *)arg1; struct e1000_hw *hw = &adapter->hw; struct sbuf *sb; u32 *regs_buff; int rc; regs_buff = malloc(sizeof(u32) * IGB_REGS_LEN, M_DEVBUF, M_WAITOK); memset(regs_buff, 0, IGB_REGS_LEN * sizeof(u32)); rc = sysctl_wire_old_buffer(req, 0); MPASS(rc == 0); if (rc != 0) { free(regs_buff, M_DEVBUF); return (rc); } sb = sbuf_new_for_sysctl(NULL, NULL, 32*400, req); MPASS(sb != NULL); if (sb == NULL) { free(regs_buff, M_DEVBUF); return (ENOMEM); } /* General Registers */ regs_buff[0] = E1000_READ_REG(hw, E1000_CTRL); regs_buff[1] = E1000_READ_REG(hw, E1000_STATUS); regs_buff[2] = E1000_READ_REG(hw, E1000_CTRL_EXT); regs_buff[3] = E1000_READ_REG(hw, E1000_ICR); regs_buff[4] = E1000_READ_REG(hw, E1000_RCTL); regs_buff[5] = E1000_READ_REG(hw, E1000_RDLEN(0)); regs_buff[6] = E1000_READ_REG(hw, E1000_RDH(0)); regs_buff[7] = E1000_READ_REG(hw, E1000_RDT(0)); regs_buff[8] = E1000_READ_REG(hw, E1000_RXDCTL(0)); regs_buff[9] = E1000_READ_REG(hw, E1000_RDBAL(0)); regs_buff[10] = E1000_READ_REG(hw, E1000_RDBAH(0)); regs_buff[11] = E1000_READ_REG(hw, E1000_TCTL); regs_buff[12] = E1000_READ_REG(hw, E1000_TDBAL(0)); regs_buff[13] = E1000_READ_REG(hw, E1000_TDBAH(0)); regs_buff[14] = E1000_READ_REG(hw, E1000_TDLEN(0)); regs_buff[15] = E1000_READ_REG(hw, E1000_TDH(0)); regs_buff[16] = E1000_READ_REG(hw, E1000_TDT(0)); regs_buff[17] = E1000_READ_REG(hw, E1000_TXDCTL(0)); regs_buff[18] = E1000_READ_REG(hw, E1000_TDFH); regs_buff[19] = E1000_READ_REG(hw, E1000_TDFT); regs_buff[20] = E1000_READ_REG(hw, E1000_TDFHS); regs_buff[21] = E1000_READ_REG(hw, E1000_TDFPC); sbuf_printf(sb, "General Registers\n"); sbuf_printf(sb, "\tCTRL\t %08x\n", regs_buff[0]); sbuf_printf(sb, "\tSTATUS\t %08x\n", regs_buff[1]); sbuf_printf(sb, "\tCTRL_EXIT\t %08x\n\n", regs_buff[2]); sbuf_printf(sb, "Interrupt Registers\n"); sbuf_printf(sb, "\tICR\t %08x\n\n", regs_buff[3]); sbuf_printf(sb, "RX Registers\n"); sbuf_printf(sb, "\tRCTL\t %08x\n", regs_buff[4]); sbuf_printf(sb, "\tRDLEN\t %08x\n", regs_buff[5]); sbuf_printf(sb, "\tRDH\t %08x\n", regs_buff[6]); sbuf_printf(sb, "\tRDT\t %08x\n", regs_buff[7]); sbuf_printf(sb, "\tRXDCTL\t %08x\n", regs_buff[8]); sbuf_printf(sb, "\tRDBAL\t %08x\n", regs_buff[9]); sbuf_printf(sb, "\tRDBAH\t %08x\n\n", regs_buff[10]); sbuf_printf(sb, "TX Registers\n"); sbuf_printf(sb, "\tTCTL\t %08x\n", regs_buff[11]); sbuf_printf(sb, "\tTDBAL\t %08x\n", regs_buff[12]); sbuf_printf(sb, "\tTDBAH\t %08x\n", regs_buff[13]); sbuf_printf(sb, "\tTDLEN\t %08x\n", regs_buff[14]); sbuf_printf(sb, "\tTDH\t %08x\n", regs_buff[15]); sbuf_printf(sb, "\tTDT\t %08x\n", regs_buff[16]); sbuf_printf(sb, "\tTXDCTL\t %08x\n", regs_buff[17]); sbuf_printf(sb, "\tTDFH\t %08x\n", regs_buff[18]); sbuf_printf(sb, "\tTDFT\t %08x\n", regs_buff[19]); sbuf_printf(sb, "\tTDFHS\t %08x\n", regs_buff[20]); sbuf_printf(sb, "\tTDFPC\t %08x\n\n", regs_buff[21]); free(regs_buff, M_DEVBUF); #ifdef DUMP_DESCS { if_softc_ctx_t scctx = adapter->shared; struct rx_ring *rxr = &rx_que->rxr; struct tx_ring *txr = &tx_que->txr; int ntxd = scctx->isc_ntxd[0]; int nrxd = scctx->isc_nrxd[0]; int j; for (j = 0; j < nrxd; j++) { u32 staterr = le32toh(rxr->rx_base[j].wb.upper.status_error); u32 length = le32toh(rxr->rx_base[j].wb.upper.length); sbuf_printf(sb, "\tReceive Descriptor Address %d: %08" PRIx64 " Error:%d Length:%d\n", j, rxr->rx_base[j].read.buffer_addr, staterr, length); } for (j = 0; j < min(ntxd, 256); j++) { unsigned int *ptr = (unsigned int *)&txr->tx_base[j]; sbuf_printf(sb, "\tTXD[%03d] [0]: %08x [1]: %08x [2]: %08x [3]: %08x eop: %d DD=%d\n", j, ptr[0], ptr[1], ptr[2], ptr[3], buf->eop, buf->eop != -1 ? txr->tx_base[buf->eop].upper.fields.status & E1000_TXD_STAT_DD : 0); } } #endif rc = sbuf_finish(sb); sbuf_delete(sb); return(rc); } static void * em_register(device_t dev) { return (em_sctx); } static void * igb_register(device_t dev) { return (igb_sctx); } static int em_set_num_queues(if_ctx_t ctx) { struct adapter *adapter = iflib_get_softc(ctx); int maxqueues; /* Sanity check based on HW */ switch (adapter->hw.mac.type) { case e1000_82576: case e1000_82580: case e1000_i350: case e1000_i354: maxqueues = 8; break; case e1000_i210: case e1000_82575: maxqueues = 4; break; case e1000_i211: case e1000_82574: maxqueues = 2; break; default: maxqueues = 1; break; } return (maxqueues); } #define EM_CAPS \ IFCAP_TSO4 | IFCAP_TXCSUM | IFCAP_LRO | IFCAP_RXCSUM | IFCAP_VLAN_HWFILTER | IFCAP_WOL_MAGIC | \ IFCAP_WOL_MCAST | IFCAP_WOL | IFCAP_VLAN_HWTSO | IFCAP_HWCSUM | IFCAP_VLAN_HWTAGGING | \ IFCAP_VLAN_HWCSUM | IFCAP_VLAN_HWTSO | IFCAP_VLAN_MTU; #define IGB_CAPS \ IFCAP_TSO4 | IFCAP_TXCSUM | IFCAP_LRO | IFCAP_RXCSUM | IFCAP_VLAN_HWFILTER | IFCAP_WOL_MAGIC | \ IFCAP_WOL_MCAST | IFCAP_WOL | IFCAP_VLAN_HWTSO | IFCAP_HWCSUM | IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM | \ IFCAP_VLAN_HWTSO | IFCAP_VLAN_MTU | IFCAP_TXCSUM_IPV6 | IFCAP_HWCSUM_IPV6 | IFCAP_JUMBO_MTU; /********************************************************************* * Device initialization routine * * The attach entry point is called when the driver is being loaded. * This routine identifies the type of hardware, allocates all resources * and initializes the hardware. * * return 0 on success, positive on failure *********************************************************************/ static int em_if_attach_pre(if_ctx_t ctx) { struct adapter *adapter; if_softc_ctx_t scctx; device_t dev; struct e1000_hw *hw; int error = 0; INIT_DEBUGOUT("em_if_attach_pre begin"); dev = iflib_get_dev(ctx); adapter = iflib_get_softc(ctx); if (resource_disabled("em", device_get_unit(dev))) { device_printf(dev, "Disabled by device hint\n"); return (ENXIO); } adapter->ctx = ctx; adapter->dev = adapter->osdep.dev = dev; scctx = adapter->shared = iflib_get_softc_ctx(ctx); adapter->media = iflib_get_media(ctx); hw = &adapter->hw; adapter->tx_process_limit = scctx->isc_ntxd[0]; /* SYSCTL stuff */ SYSCTL_ADD_PROC(device_get_sysctl_ctx(dev), SYSCTL_CHILDREN(device_get_sysctl_tree(dev)), OID_AUTO, "nvm", CTLTYPE_INT|CTLFLAG_RW, adapter, 0, em_sysctl_nvm_info, "I", "NVM Information"); SYSCTL_ADD_PROC(device_get_sysctl_ctx(dev), SYSCTL_CHILDREN(device_get_sysctl_tree(dev)), OID_AUTO, "debug", CTLTYPE_INT|CTLFLAG_RW, adapter, 0, em_sysctl_debug_info, "I", "Debug Information"); SYSCTL_ADD_PROC(device_get_sysctl_ctx(dev), SYSCTL_CHILDREN(device_get_sysctl_tree(dev)), OID_AUTO, "fc", CTLTYPE_INT|CTLFLAG_RW, adapter, 0, em_set_flowcntl, "I", "Flow Control"); SYSCTL_ADD_PROC(device_get_sysctl_ctx(dev), SYSCTL_CHILDREN(device_get_sysctl_tree(dev)), OID_AUTO, "reg_dump", CTLTYPE_STRING | CTLFLAG_RD, adapter, 0, em_get_regs, "A", "Dump Registers"); SYSCTL_ADD_PROC(device_get_sysctl_ctx(dev), SYSCTL_CHILDREN(device_get_sysctl_tree(dev)), OID_AUTO, "rs_dump", CTLTYPE_INT | CTLFLAG_RW, adapter, 0, em_get_rs, "I", "Dump RS indexes"); /* Determine hardware and mac info */ em_identify_hardware(ctx); /* Set isc_msix_bar */ scctx->isc_msix_bar = PCIR_BAR(EM_MSIX_BAR); scctx->isc_tx_nsegments = EM_MAX_SCATTER; scctx->isc_tx_tso_segments_max = scctx->isc_tx_nsegments; scctx->isc_tx_tso_size_max = EM_TSO_SIZE; scctx->isc_tx_tso_segsize_max = EM_TSO_SEG_SIZE; scctx->isc_nrxqsets_max = scctx->isc_ntxqsets_max = em_set_num_queues(ctx); device_printf(dev, "attach_pre capping queues at %d\n", scctx->isc_ntxqsets_max); scctx->isc_tx_csum_flags = CSUM_TCP | CSUM_UDP | CSUM_IP_TSO; if (adapter->hw.mac.type >= igb_mac_min) { int try_second_bar; scctx->isc_txqsizes[0] = roundup2(scctx->isc_ntxd[0] * sizeof(union e1000_adv_tx_desc), EM_DBA_ALIGN); scctx->isc_rxqsizes[0] = roundup2(scctx->isc_nrxd[0] * sizeof(union e1000_adv_rx_desc), EM_DBA_ALIGN); scctx->isc_txd_size[0] = sizeof(union e1000_adv_tx_desc); scctx->isc_rxd_size[0] = sizeof(union e1000_adv_rx_desc); scctx->isc_txrx = &igb_txrx; scctx->isc_capenable = IGB_CAPS; scctx->isc_tx_csum_flags = CSUM_TCP | CSUM_UDP | CSUM_TSO | CSUM_IP6_TCP \ | CSUM_IP6_UDP | CSUM_IP6_TCP; if (adapter->hw.mac.type != e1000_82575) scctx->isc_tx_csum_flags |= CSUM_SCTP | CSUM_IP6_SCTP; /* ** Some new devices, as with ixgbe, now may ** use a different BAR, so we need to keep ** track of which is used. */ try_second_bar = pci_read_config(dev, scctx->isc_msix_bar, 4); if (try_second_bar == 0) scctx->isc_msix_bar += 4; } else if (adapter->hw.mac.type >= em_mac_min) { scctx->isc_txqsizes[0] = roundup2(scctx->isc_ntxd[0]* sizeof(struct e1000_tx_desc), EM_DBA_ALIGN); scctx->isc_rxqsizes[0] = roundup2(scctx->isc_nrxd[0] * sizeof(union e1000_rx_desc_extended), EM_DBA_ALIGN); scctx->isc_txd_size[0] = sizeof(struct e1000_tx_desc); scctx->isc_rxd_size[0] = sizeof(union e1000_rx_desc_extended); scctx->isc_txrx = &em_txrx; scctx->isc_capenable = EM_CAPS; scctx->isc_tx_csum_flags = CSUM_TCP | CSUM_UDP | CSUM_IP_TSO; } else { scctx->isc_txqsizes[0] = roundup2((scctx->isc_ntxd[0] + 1) * sizeof(struct e1000_tx_desc), EM_DBA_ALIGN); scctx->isc_rxqsizes[0] = roundup2((scctx->isc_nrxd[0] + 1) * sizeof(struct e1000_rx_desc), EM_DBA_ALIGN); scctx->isc_txd_size[0] = sizeof(struct e1000_tx_desc); scctx->isc_rxd_size[0] = sizeof(struct e1000_rx_desc); scctx->isc_tx_csum_flags = CSUM_TCP | CSUM_UDP | CSUM_IP_TSO; scctx->isc_txrx = &lem_txrx; scctx->isc_capenable = EM_CAPS; if (adapter->hw.mac.type < e1000_82543) scctx->isc_capenable &= ~(IFCAP_HWCSUM|IFCAP_VLAN_HWCSUM); scctx->isc_tx_csum_flags = CSUM_TCP | CSUM_UDP | CSUM_IP_TSO; scctx->isc_msix_bar = 0; } /* Setup PCI resources */ if (em_allocate_pci_resources(ctx)) { device_printf(dev, "Allocation of PCI resources failed\n"); error = ENXIO; goto err_pci; } /* ** For ICH8 and family we need to ** map the flash memory, and this ** must happen after the MAC is ** identified */ if ((hw->mac.type == e1000_ich8lan) || (hw->mac.type == e1000_ich9lan) || (hw->mac.type == e1000_ich10lan) || (hw->mac.type == e1000_pchlan) || (hw->mac.type == e1000_pch2lan) || (hw->mac.type == e1000_pch_lpt)) { int rid = EM_BAR_TYPE_FLASH; adapter->flash = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid, RF_ACTIVE); if (adapter->flash == NULL) { device_printf(dev, "Mapping of Flash failed\n"); error = ENXIO; goto err_pci; } /* This is used in the shared code */ hw->flash_address = (u8 *)adapter->flash; adapter->osdep.flash_bus_space_tag = rman_get_bustag(adapter->flash); adapter->osdep.flash_bus_space_handle = rman_get_bushandle(adapter->flash); } /* ** In the new SPT device flash is not a ** separate BAR, rather it is also in BAR0, ** so use the same tag and an offset handle for the ** FLASH read/write macros in the shared code. */ else if (hw->mac.type == e1000_pch_spt) { adapter->osdep.flash_bus_space_tag = adapter->osdep.mem_bus_space_tag; adapter->osdep.flash_bus_space_handle = adapter->osdep.mem_bus_space_handle + E1000_FLASH_BASE_ADDR; } /* Do Shared Code initialization */ error = e1000_setup_init_funcs(hw, TRUE); if (error) { device_printf(dev, "Setup of Shared code failed, error %d\n", error); error = ENXIO; goto err_pci; } em_setup_msix(ctx); e1000_get_bus_info(hw); /* Set up some sysctls for the tunable interrupt delays */ em_add_int_delay_sysctl(adapter, "rx_int_delay", "receive interrupt delay in usecs", &adapter->rx_int_delay, E1000_REGISTER(hw, E1000_RDTR), em_rx_int_delay_dflt); em_add_int_delay_sysctl(adapter, "tx_int_delay", "transmit interrupt delay in usecs", &adapter->tx_int_delay, E1000_REGISTER(hw, E1000_TIDV), em_tx_int_delay_dflt); em_add_int_delay_sysctl(adapter, "rx_abs_int_delay", "receive interrupt delay limit in usecs", &adapter->rx_abs_int_delay, E1000_REGISTER(hw, E1000_RADV), em_rx_abs_int_delay_dflt); em_add_int_delay_sysctl(adapter, "tx_abs_int_delay", "transmit interrupt delay limit in usecs", &adapter->tx_abs_int_delay, E1000_REGISTER(hw, E1000_TADV), em_tx_abs_int_delay_dflt); em_add_int_delay_sysctl(adapter, "itr", "interrupt delay limit in usecs/4", &adapter->tx_itr, E1000_REGISTER(hw, E1000_ITR), DEFAULT_ITR); hw->mac.autoneg = DO_AUTO_NEG; hw->phy.autoneg_wait_to_complete = FALSE; hw->phy.autoneg_advertised = AUTONEG_ADV_DEFAULT; if (adapter->hw.mac.type < em_mac_min) { e1000_init_script_state_82541(&adapter->hw, TRUE); e1000_set_tbi_compatibility_82543(&adapter->hw, TRUE); } /* Copper options */ if (hw->phy.media_type == e1000_media_type_copper) { hw->phy.mdix = AUTO_ALL_MODES; hw->phy.disable_polarity_correction = FALSE; hw->phy.ms_type = EM_MASTER_SLAVE; } /* * Set the frame limits assuming * standard ethernet sized frames. */ scctx->isc_max_frame_size = adapter->hw.mac.max_frame_size = ETHERMTU + ETHER_HDR_LEN + ETHERNET_FCS_SIZE; /* * This controls when hardware reports transmit completion * status. */ hw->mac.report_tx_early = 1; /* Allocate multicast array memory. */ adapter->mta = malloc(sizeof(u8) * ETH_ADDR_LEN * MAX_NUM_MULTICAST_ADDRESSES, M_DEVBUF, M_NOWAIT); if (adapter->mta == NULL) { device_printf(dev, "Can not allocate multicast setup array\n"); error = ENOMEM; goto err_late; } /* Check SOL/IDER usage */ if (e1000_check_reset_block(hw)) device_printf(dev, "PHY reset is blocked" " due to SOL/IDER session.\n"); /* Sysctl for setting Energy Efficient Ethernet */ hw->dev_spec.ich8lan.eee_disable = eee_setting; SYSCTL_ADD_PROC(device_get_sysctl_ctx(dev), SYSCTL_CHILDREN(device_get_sysctl_tree(dev)), OID_AUTO, "eee_control", CTLTYPE_INT|CTLFLAG_RW, adapter, 0, em_sysctl_eee, "I", "Disable Energy Efficient Ethernet"); /* ** Start from a known state, this is ** important in reading the nvm and ** mac from that. */ e1000_reset_hw(hw); /* Make sure we have a good EEPROM before we read from it */ if (e1000_validate_nvm_checksum(hw) < 0) { /* ** Some PCI-E parts fail the first check due to ** the link being in sleep state, call it again, ** if it fails a second time its a real issue. */ if (e1000_validate_nvm_checksum(hw) < 0) { device_printf(dev, "The EEPROM Checksum Is Not Valid\n"); error = EIO; goto err_late; } } /* Copy the permanent MAC address out of the EEPROM */ if (e1000_read_mac_addr(hw) < 0) { device_printf(dev, "EEPROM read error while reading MAC" " address\n"); error = EIO; goto err_late; } if (!em_is_valid_ether_addr(hw->mac.addr)) { device_printf(dev, "Invalid MAC address\n"); error = EIO; goto err_late; } /* Disable ULP support */ e1000_disable_ulp_lpt_lp(hw, TRUE); /* * Get Wake-on-Lan and Management info for later use */ em_get_wakeup(ctx); iflib_set_mac(ctx, hw->mac.addr); return (0); err_late: em_release_hw_control(adapter); err_pci: em_free_pci_resources(ctx); free(adapter->mta, M_DEVBUF); return (error); } static int em_if_attach_post(if_ctx_t ctx) { struct adapter *adapter = iflib_get_softc(ctx); struct e1000_hw *hw = &adapter->hw; int error = 0; /* Setup OS specific network interface */ error = em_setup_interface(ctx); if (error != 0) { goto err_late; } em_reset(ctx); /* Initialize statistics */ em_update_stats_counters(adapter); hw->mac.get_link_status = 1; em_if_update_admin_status(ctx); em_add_hw_stats(adapter); /* Non-AMT based hardware can now take control from firmware */ if (adapter->has_manage && !adapter->has_amt) em_get_hw_control(adapter); INIT_DEBUGOUT("em_if_attach_post: end"); return (error); err_late: em_release_hw_control(adapter); em_free_pci_resources(ctx); em_if_queues_free(ctx); free(adapter->mta, M_DEVBUF); return (error); } /********************************************************************* * Device removal routine * * The detach entry point is called when the driver is being removed. * This routine stops the adapter and deallocates all the resources * that were allocated for driver operation. * * return 0 on success, positive on failure *********************************************************************/ static int em_if_detach(if_ctx_t ctx) { struct adapter *adapter = iflib_get_softc(ctx); INIT_DEBUGOUT("em_detach: begin"); e1000_phy_hw_reset(&adapter->hw); em_release_manageability(adapter); em_release_hw_control(adapter); em_free_pci_resources(ctx); return (0); } /********************************************************************* * * Shutdown entry point * **********************************************************************/ static int em_if_shutdown(if_ctx_t ctx) { return em_if_suspend(ctx); } /* * Suspend/resume device methods. */ static int em_if_suspend(if_ctx_t ctx) { struct adapter *adapter = iflib_get_softc(ctx); em_release_manageability(adapter); em_release_hw_control(adapter); em_enable_wakeup(ctx); return (0); } static int em_if_resume(if_ctx_t ctx) { struct adapter *adapter = iflib_get_softc(ctx); if (adapter->hw.mac.type == e1000_pch2lan) e1000_resume_workarounds_pchlan(&adapter->hw); em_if_init(ctx); em_init_manageability(adapter); return(0); } static int em_if_mtu_set(if_ctx_t ctx, uint32_t mtu) { int max_frame_size; struct adapter *adapter = iflib_get_softc(ctx); if_softc_ctx_t scctx = iflib_get_softc_ctx(ctx); IOCTL_DEBUGOUT("ioctl rcv'd: SIOCSIFMTU (Set Interface MTU)"); switch (adapter->hw.mac.type) { case e1000_82571: case e1000_82572: case e1000_ich9lan: case e1000_ich10lan: case e1000_pch2lan: case e1000_pch_lpt: case e1000_pch_spt: case e1000_82574: case e1000_82583: case e1000_80003es2lan: /* 9K Jumbo Frame size */ max_frame_size = 9234; break; case e1000_pchlan: max_frame_size = 4096; break; case e1000_82542: case e1000_ich8lan: /* Adapters that do not support jumbo frames */ max_frame_size = ETHER_MAX_LEN; break; default: if (adapter->hw.mac.type >= igb_mac_min) max_frame_size = 9234; else /* lem */ max_frame_size = MAX_JUMBO_FRAME_SIZE; } if (mtu > max_frame_size - ETHER_HDR_LEN - ETHER_CRC_LEN) { return (EINVAL); } scctx->isc_max_frame_size = adapter->hw.mac.max_frame_size = mtu + ETHER_HDR_LEN + ETHER_CRC_LEN; return (0); } /********************************************************************* * Init entry point * * This routine is used in two ways. It is used by the stack as * 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 **********************************************************************/ static void em_if_init(if_ctx_t ctx) { struct adapter *adapter = iflib_get_softc(ctx); struct ifnet *ifp = iflib_get_ifp(ctx); struct em_tx_queue *tx_que; int i; INIT_DEBUGOUT("em_if_init: begin"); /* Get the latest mac address, User can use a LAA */ bcopy(if_getlladdr(ifp), adapter->hw.mac.addr, ETHER_ADDR_LEN); /* Put the address into the Receive Address Array */ e1000_rar_set(&adapter->hw, adapter->hw.mac.addr, 0); /* * With the 82571 adapter, RAR[0] may be overwritten * when the other port is reset, we make a duplicate * in RAR[14] for that eventuality, this assures * the interface continues to function. */ if (adapter->hw.mac.type == e1000_82571) { e1000_set_laa_state_82571(&adapter->hw, TRUE); e1000_rar_set(&adapter->hw, adapter->hw.mac.addr, E1000_RAR_ENTRIES - 1); } /* Initialize the hardware */ em_reset(ctx); em_if_update_admin_status(ctx); for (i = 0, tx_que = adapter->tx_queues; i < adapter->tx_num_queues; i++, tx_que++) { struct tx_ring *txr = &tx_que->txr; txr->tx_rs_cidx = txr->tx_rs_pidx = txr->tx_cidx_processed = 0; } /* Setup VLAN support, basic and offload if available */ E1000_WRITE_REG(&adapter->hw, E1000_VET, ETHERTYPE_VLAN); /* Clear bad data from Rx FIFOs */ if (adapter->hw.mac.type >= igb_mac_min) e1000_rx_fifo_flush_82575(&adapter->hw); /* Configure for OS presence */ em_init_manageability(adapter); /* Prepare transmit descriptors and buffers */ em_initialize_transmit_unit(ctx); /* Setup Multicast table */ em_if_multi_set(ctx); /* * Figure out the desired mbuf * pool for doing jumbos */ if (adapter->hw.mac.max_frame_size <= 2048) adapter->rx_mbuf_sz = MCLBYTES; #ifndef CONTIGMALLOC_WORKS else adapter->rx_mbuf_sz = MJUMPAGESIZE; #else else if (adapter->hw.mac.max_frame_size <= 4096) adapter->rx_mbuf_sz = MJUMPAGESIZE; else adapter->rx_mbuf_sz = MJUM9BYTES; #endif em_initialize_receive_unit(ctx); /* Use real VLAN Filter support? */ if (if_getcapenable(ifp) & IFCAP_VLAN_HWTAGGING) { if (if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER) /* Use real VLAN Filter support */ em_setup_vlan_hw_support(adapter); else { u32 ctrl; ctrl = E1000_READ_REG(&adapter->hw, E1000_CTRL); ctrl |= E1000_CTRL_VME; E1000_WRITE_REG(&adapter->hw, E1000_CTRL, ctrl); } } /* Don't lose promiscuous settings */ em_if_set_promisc(ctx, IFF_PROMISC); e1000_clear_hw_cntrs_base_generic(&adapter->hw); /* MSI/X configuration for 82574 */ if (adapter->hw.mac.type == e1000_82574) { int tmp = E1000_READ_REG(&adapter->hw, E1000_CTRL_EXT); tmp |= E1000_CTRL_EXT_PBA_CLR; E1000_WRITE_REG(&adapter->hw, E1000_CTRL_EXT, tmp); /* Set the IVAR - interrupt vector routing. */ E1000_WRITE_REG(&adapter->hw, E1000_IVAR, adapter->ivars); } else if (adapter->intr_type == IFLIB_INTR_MSIX) /* Set up queue routing */ igb_configure_queues(adapter); /* this clears any pending interrupts */ E1000_READ_REG(&adapter->hw, E1000_ICR); E1000_WRITE_REG(&adapter->hw, E1000_ICS, E1000_ICS_LSC); /* AMT based hardware can now take control from firmware */ if (adapter->has_manage && adapter->has_amt) em_get_hw_control(adapter); /* Set Energy Efficient Ethernet */ if (adapter->hw.mac.type >= igb_mac_min && adapter->hw.phy.media_type == e1000_media_type_copper) { if (adapter->hw.mac.type == e1000_i354) e1000_set_eee_i354(&adapter->hw, TRUE, TRUE); else e1000_set_eee_i350(&adapter->hw, TRUE, TRUE); } } /********************************************************************* * * Fast Legacy/MSI Combined Interrupt Service routine * *********************************************************************/ int em_intr(void *arg) { struct adapter *adapter = arg; if_ctx_t ctx = adapter->ctx; u32 reg_icr; reg_icr = E1000_READ_REG(&adapter->hw, E1000_ICR); if (adapter->intr_type != IFLIB_INTR_LEGACY) goto skip_stray; /* Hot eject? */ if (reg_icr == 0xffffffff) return FILTER_STRAY; /* Definitely not our interrupt. */ if (reg_icr == 0x0) return FILTER_STRAY; /* * Starting with the 82571 chip, bit 31 should be used to * determine whether the interrupt belongs to us. */ if (adapter->hw.mac.type >= e1000_82571 && (reg_icr & E1000_ICR_INT_ASSERTED) == 0) return FILTER_STRAY; skip_stray: /* Link status change */ if (reg_icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) { adapter->hw.mac.get_link_status = 1; iflib_admin_intr_deferred(ctx); } if (reg_icr & E1000_ICR_RXO) adapter->rx_overruns++; return (FILTER_SCHEDULE_THREAD); } static void igb_rx_enable_queue(struct adapter *adapter, struct em_rx_queue *rxq) { E1000_WRITE_REG(&adapter->hw, E1000_EIMS, rxq->eims); } static void em_rx_enable_queue(struct adapter *adapter, struct em_rx_queue *rxq) { E1000_WRITE_REG(&adapter->hw, E1000_IMS, rxq->eims); } static void igb_tx_enable_queue(struct adapter *adapter, struct em_tx_queue *txq) { E1000_WRITE_REG(&adapter->hw, E1000_EIMS, txq->eims); } static void em_tx_enable_queue(struct adapter *adapter, struct em_tx_queue *txq) { E1000_WRITE_REG(&adapter->hw, E1000_IMS, txq->eims); } static int em_if_rx_queue_intr_enable(if_ctx_t ctx, uint16_t rxqid) { struct adapter *adapter = iflib_get_softc(ctx); struct em_rx_queue *rxq = &adapter->rx_queues[rxqid]; if (adapter->hw.mac.type >= igb_mac_min) igb_rx_enable_queue(adapter, rxq); else em_rx_enable_queue(adapter, rxq); return (0); } static int em_if_tx_queue_intr_enable(if_ctx_t ctx, uint16_t txqid) { struct adapter *adapter = iflib_get_softc(ctx); struct em_tx_queue *txq = &adapter->tx_queues[txqid]; if (adapter->hw.mac.type >= igb_mac_min) igb_tx_enable_queue(adapter, txq); else em_tx_enable_queue(adapter, txq); return (0); } /********************************************************************* * * MSIX RX Interrupt Service routine * **********************************************************************/ static int em_msix_que(void *arg) { struct em_rx_queue *que = arg; ++que->irqs; return (FILTER_SCHEDULE_THREAD); } /********************************************************************* * * MSIX Link Fast Interrupt Service routine * **********************************************************************/ static int em_msix_link(void *arg) { struct adapter *adapter = arg; u32 reg_icr; ++adapter->link_irq; MPASS(adapter->hw.back != NULL); reg_icr = E1000_READ_REG(&adapter->hw, E1000_ICR); if (reg_icr & E1000_ICR_RXO) adapter->rx_overruns++; if (reg_icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) { em_handle_link(adapter->ctx); } else { E1000_WRITE_REG(&adapter->hw, E1000_IMS, EM_MSIX_LINK | E1000_IMS_LSC); if (adapter->hw.mac.type >= igb_mac_min) E1000_WRITE_REG(&adapter->hw, E1000_EIMS, adapter->link_mask); } /* * Because we must read the ICR for this interrupt * it may clear other causes using autoclear, for * this reason we simply create a soft interrupt * for all these vectors. */ if (reg_icr && adapter->hw.mac.type < igb_mac_min) { E1000_WRITE_REG(&adapter->hw, E1000_ICS, adapter->ims); } return (FILTER_HANDLED); } static void em_handle_link(void *context) { if_ctx_t ctx = context; struct adapter *adapter = iflib_get_softc(ctx); adapter->hw.mac.get_link_status = 1; iflib_admin_intr_deferred(ctx); } /********************************************************************* * * Media Ioctl callback * * This routine is called whenever the user queries the status of * the interface using ifconfig. * **********************************************************************/ static void em_if_media_status(if_ctx_t ctx, struct ifmediareq *ifmr) { struct adapter *adapter = iflib_get_softc(ctx); u_char fiber_type = IFM_1000_SX; INIT_DEBUGOUT("em_if_media_status: begin"); iflib_admin_intr_deferred(ctx); ifmr->ifm_status = IFM_AVALID; ifmr->ifm_active = IFM_ETHER; if (!adapter->link_active) { return; } ifmr->ifm_status |= IFM_ACTIVE; if ((adapter->hw.phy.media_type == e1000_media_type_fiber) || (adapter->hw.phy.media_type == e1000_media_type_internal_serdes)) { if (adapter->hw.mac.type == e1000_82545) fiber_type = IFM_1000_LX; ifmr->ifm_active |= fiber_type | IFM_FDX; } else { switch (adapter->link_speed) { case 10: ifmr->ifm_active |= IFM_10_T; break; case 100: ifmr->ifm_active |= IFM_100_TX; break; case 1000: ifmr->ifm_active |= IFM_1000_T; break; } if (adapter->link_duplex == FULL_DUPLEX) ifmr->ifm_active |= IFM_FDX; else ifmr->ifm_active |= IFM_HDX; } } /********************************************************************* * * Media Ioctl callback * * This routine is called when the user changes speed/duplex using * media/mediopt option with ifconfig. * **********************************************************************/ static int em_if_media_change(if_ctx_t ctx) { struct adapter *adapter = iflib_get_softc(ctx); struct ifmedia *ifm = iflib_get_media(ctx); INIT_DEBUGOUT("em_if_media_change: begin"); if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER) return (EINVAL); switch (IFM_SUBTYPE(ifm->ifm_media)) { case IFM_AUTO: adapter->hw.mac.autoneg = DO_AUTO_NEG; adapter->hw.phy.autoneg_advertised = AUTONEG_ADV_DEFAULT; break; case IFM_1000_LX: case IFM_1000_SX: case IFM_1000_T: adapter->hw.mac.autoneg = DO_AUTO_NEG; adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL; break; case IFM_100_TX: adapter->hw.mac.autoneg = FALSE; adapter->hw.phy.autoneg_advertised = 0; if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX) adapter->hw.mac.forced_speed_duplex = ADVERTISE_100_FULL; else adapter->hw.mac.forced_speed_duplex = ADVERTISE_100_HALF; break; case IFM_10_T: adapter->hw.mac.autoneg = FALSE; adapter->hw.phy.autoneg_advertised = 0; if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX) adapter->hw.mac.forced_speed_duplex = ADVERTISE_10_FULL; else adapter->hw.mac.forced_speed_duplex = ADVERTISE_10_HALF; break; default: device_printf(adapter->dev, "Unsupported media type\n"); } em_if_init(ctx); return (0); } static int em_if_set_promisc(if_ctx_t ctx, int flags) { struct adapter *adapter = iflib_get_softc(ctx); u32 reg_rctl; em_disable_promisc(ctx); reg_rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL); if (flags & IFF_PROMISC) { reg_rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE); /* Turn this on if you want to see bad packets */ if (em_debug_sbp) reg_rctl |= E1000_RCTL_SBP; E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl); } else if (flags & IFF_ALLMULTI) { reg_rctl |= E1000_RCTL_MPE; reg_rctl &= ~E1000_RCTL_UPE; E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl); } return (0); } static void em_disable_promisc(if_ctx_t ctx) { struct adapter *adapter = iflib_get_softc(ctx); struct ifnet *ifp = iflib_get_ifp(ctx); u32 reg_rctl; int mcnt = 0; reg_rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL); reg_rctl &= (~E1000_RCTL_UPE); if (if_getflags(ifp) & IFF_ALLMULTI) mcnt = MAX_NUM_MULTICAST_ADDRESSES; else mcnt = if_multiaddr_count(ifp, MAX_NUM_MULTICAST_ADDRESSES); /* Don't disable if in MAX groups */ if (mcnt < MAX_NUM_MULTICAST_ADDRESSES) reg_rctl &= (~E1000_RCTL_MPE); reg_rctl &= (~E1000_RCTL_SBP); E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl); } /********************************************************************* * Multicast Update * * This routine is called whenever multicast address list is updated. * **********************************************************************/ static void em_if_multi_set(if_ctx_t ctx) { struct adapter *adapter = iflib_get_softc(ctx); struct ifnet *ifp = iflib_get_ifp(ctx); u32 reg_rctl = 0; u8 *mta; /* Multicast array memory */ int mcnt = 0; IOCTL_DEBUGOUT("em_set_multi: begin"); mta = adapter->mta; bzero(mta, sizeof(u8) * ETH_ADDR_LEN * MAX_NUM_MULTICAST_ADDRESSES); if (adapter->hw.mac.type == e1000_82542 && adapter->hw.revision_id == E1000_REVISION_2) { reg_rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL); if (adapter->hw.bus.pci_cmd_word & CMD_MEM_WRT_INVALIDATE) e1000_pci_clear_mwi(&adapter->hw); reg_rctl |= E1000_RCTL_RST; E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl); msec_delay(5); } if_multiaddr_array(ifp, mta, &mcnt, MAX_NUM_MULTICAST_ADDRESSES); if (mcnt >= MAX_NUM_MULTICAST_ADDRESSES) { reg_rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL); reg_rctl |= E1000_RCTL_MPE; E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl); } else e1000_update_mc_addr_list(&adapter->hw, mta, mcnt); if (adapter->hw.mac.type == e1000_82542 && adapter->hw.revision_id == E1000_REVISION_2) { reg_rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL); reg_rctl &= ~E1000_RCTL_RST; E1000_WRITE_REG(&adapter->hw, E1000_RCTL, reg_rctl); msec_delay(5); if (adapter->hw.bus.pci_cmd_word & CMD_MEM_WRT_INVALIDATE) e1000_pci_set_mwi(&adapter->hw); } } /********************************************************************* * Timer routine * * This routine checks for link status and updates statistics. * **********************************************************************/ static void em_if_timer(if_ctx_t ctx, uint16_t qid) { struct adapter *adapter = iflib_get_softc(ctx); struct em_rx_queue *que; int i; int trigger = 0; if (qid != 0) return; iflib_admin_intr_deferred(ctx); /* Reset LAA into RAR[0] on 82571 */ if ((adapter->hw.mac.type == e1000_82571) && e1000_get_laa_state_82571(&adapter->hw)) e1000_rar_set(&adapter->hw, adapter->hw.mac.addr, 0); if (adapter->hw.mac.type < em_mac_min) lem_smartspeed(adapter); /* Mask to use in the irq trigger */ if (adapter->intr_type == IFLIB_INTR_MSIX) { for (i = 0, que = adapter->rx_queues; i < adapter->rx_num_queues; i++, que++) trigger |= que->eims; } else { trigger = E1000_ICS_RXDMT0; } } static void em_if_update_admin_status(if_ctx_t ctx) { struct adapter *adapter = iflib_get_softc(ctx); struct e1000_hw *hw = &adapter->hw; struct ifnet *ifp = iflib_get_ifp(ctx); device_t dev = iflib_get_dev(ctx); u32 link_check, thstat, ctrl; link_check = thstat = ctrl = 0; /* Get the cached link value or read phy for real */ switch (hw->phy.media_type) { case e1000_media_type_copper: if (hw->mac.get_link_status) { if (hw->mac.type == e1000_pch_spt) msec_delay(50); /* Do the work to read phy */ e1000_check_for_link(hw); link_check = !hw->mac.get_link_status; if (link_check) /* ESB2 fix */ e1000_cfg_on_link_up(hw); } else { link_check = TRUE; } break; case e1000_media_type_fiber: e1000_check_for_link(hw); link_check = (E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU); break; case e1000_media_type_internal_serdes: e1000_check_for_link(hw); link_check = adapter->hw.mac.serdes_has_link; break; /* VF device is type_unknown */ case e1000_media_type_unknown: e1000_check_for_link(hw); link_check = !hw->mac.get_link_status; /* FALLTHROUGH */ default: break; } /* Check for thermal downshift or shutdown */ if (hw->mac.type == e1000_i350) { thstat = E1000_READ_REG(hw, E1000_THSTAT); ctrl = E1000_READ_REG(hw, E1000_CTRL_EXT); } /* Now check for a transition */ if (link_check && (adapter->link_active == 0)) { e1000_get_speed_and_duplex(hw, &adapter->link_speed, &adapter->link_duplex); /* Check if we must disable SPEED_MODE bit on PCI-E */ if ((adapter->link_speed != SPEED_1000) && ((hw->mac.type == e1000_82571) || (hw->mac.type == e1000_82572))) { int tarc0; tarc0 = E1000_READ_REG(hw, E1000_TARC(0)); tarc0 &= ~TARC_SPEED_MODE_BIT; E1000_WRITE_REG(hw, E1000_TARC(0), tarc0); } if (bootverbose) device_printf(dev, "Link is up %d Mbps %s\n", adapter->link_speed, ((adapter->link_duplex == FULL_DUPLEX) ? "Full Duplex" : "Half Duplex")); adapter->link_active = 1; adapter->smartspeed = 0; if_setbaudrate(ifp, adapter->link_speed * 1000000); if ((ctrl & E1000_CTRL_EXT_LINK_MODE_GMII) && (thstat & E1000_THSTAT_LINK_THROTTLE)) device_printf(dev, "Link: thermal downshift\n"); /* Delay Link Up for Phy update */ if (((hw->mac.type == e1000_i210) || (hw->mac.type == e1000_i211)) && (hw->phy.id == I210_I_PHY_ID)) msec_delay(I210_LINK_DELAY); /* Reset if the media type changed. */ if ((hw->dev_spec._82575.media_changed) && (adapter->hw.mac.type >= igb_mac_min)) { hw->dev_spec._82575.media_changed = false; adapter->flags |= IGB_MEDIA_RESET; em_reset(ctx); } iflib_link_state_change(ctx, LINK_STATE_UP, ifp->if_baudrate); printf("Link state changed to up\n"); } else if (!link_check && (adapter->link_active == 1)) { if_setbaudrate(ifp, 0); adapter->link_speed = 0; adapter->link_duplex = 0; if (bootverbose) device_printf(dev, "Link is Down\n"); adapter->link_active = 0; iflib_link_state_change(ctx, LINK_STATE_DOWN, ifp->if_baudrate); printf("link state changed to down\n"); } em_update_stats_counters(adapter); E1000_WRITE_REG(&adapter->hw, E1000_IMS, EM_MSIX_LINK | E1000_IMS_LSC); } /********************************************************************* * * This routine disables all traffic on the adapter by issuing a * global reset on the MAC and deallocates TX/RX buffers. * * This routine should always be called with BOTH the CORE * and TX locks. **********************************************************************/ static void em_if_stop(if_ctx_t ctx) { struct adapter *adapter = iflib_get_softc(ctx); INIT_DEBUGOUT("em_stop: begin"); e1000_reset_hw(&adapter->hw); if (adapter->hw.mac.type >= e1000_82544) E1000_WRITE_REG(&adapter->hw, E1000_WUFC, 0); e1000_led_off(&adapter->hw); e1000_cleanup_led(&adapter->hw); } /********************************************************************* * * Determine hardware revision. * **********************************************************************/ static void em_identify_hardware(if_ctx_t ctx) { device_t dev = iflib_get_dev(ctx); struct adapter *adapter = iflib_get_softc(ctx); /* Make sure our PCI config space has the necessary stuff set */ adapter->hw.bus.pci_cmd_word = pci_read_config(dev, PCIR_COMMAND, 2); /* Save off the information about this board */ adapter->hw.vendor_id = pci_get_vendor(dev); adapter->hw.device_id = pci_get_device(dev); adapter->hw.revision_id = pci_read_config(dev, PCIR_REVID, 1); adapter->hw.subsystem_vendor_id = pci_read_config(dev, PCIR_SUBVEND_0, 2); adapter->hw.subsystem_device_id = pci_read_config(dev, PCIR_SUBDEV_0, 2); /* Do Shared Code Init and Setup */ if (e1000_set_mac_type(&adapter->hw)) { device_printf(dev, "Setup init failure\n"); return; } } static int em_allocate_pci_resources(if_ctx_t ctx) { struct adapter *adapter = iflib_get_softc(ctx); device_t dev = iflib_get_dev(ctx); int rid, val; rid = PCIR_BAR(0); adapter->memory = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid, RF_ACTIVE); if (adapter->memory == NULL) { device_printf(dev, "Unable to allocate bus resource: memory\n"); return (ENXIO); } adapter->osdep.mem_bus_space_tag = rman_get_bustag(adapter->memory); adapter->osdep.mem_bus_space_handle = rman_get_bushandle(adapter->memory); adapter->hw.hw_addr = (u8 *)&adapter->osdep.mem_bus_space_handle; /* Only older adapters use IO mapping */ if (adapter->hw.mac.type < em_mac_min && adapter->hw.mac.type > e1000_82543) { /* Figure our where our IO BAR is ? */ for (rid = PCIR_BAR(0); rid < PCIR_CIS;) { val = pci_read_config(dev, rid, 4); if (EM_BAR_TYPE(val) == EM_BAR_TYPE_IO) { adapter->io_rid = rid; break; } rid += 4; /* check for 64bit BAR */ if (EM_BAR_MEM_TYPE(val) == EM_BAR_MEM_TYPE_64BIT) rid += 4; } if (rid >= PCIR_CIS) { device_printf(dev, "Unable to locate IO BAR\n"); return (ENXIO); } adapter->ioport = bus_alloc_resource_any(dev, SYS_RES_IOPORT, &adapter->io_rid, RF_ACTIVE); if (adapter->ioport == NULL) { device_printf(dev, "Unable to allocate bus resource: " "ioport\n"); return (ENXIO); } adapter->hw.io_base = 0; adapter->osdep.io_bus_space_tag = rman_get_bustag(adapter->ioport); adapter->osdep.io_bus_space_handle = rman_get_bushandle(adapter->ioport); } adapter->hw.back = &adapter->osdep; return (0); } /********************************************************************* * * Setup the MSIX Interrupt handlers * **********************************************************************/ static int em_if_msix_intr_assign(if_ctx_t ctx, int msix) { struct adapter *adapter = iflib_get_softc(ctx); struct em_rx_queue *rx_que = adapter->rx_queues; struct em_tx_queue *tx_que = adapter->tx_queues; int error, rid, i, vector = 0, rx_vectors; char buf[16]; /* First set up ring resources */ for (i = 0; i < adapter->rx_num_queues; i++, rx_que++, vector++) { rid = vector + 1; snprintf(buf, sizeof(buf), "rxq%d", i); error = iflib_irq_alloc_generic(ctx, &rx_que->que_irq, rid, IFLIB_INTR_RXTX, em_msix_que, rx_que, rx_que->me, buf); if (error) { device_printf(iflib_get_dev(ctx), "Failed to allocate que int %d err: %d", i, error); adapter->rx_num_queues = i + 1; goto fail; } rx_que->msix = vector; /* * Set the bit to enable interrupt * in E1000_IMS -- bits 20 and 21 * are for RX0 and RX1, note this has * NOTHING to do with the MSIX vector */ if (adapter->hw.mac.type == e1000_82574) { rx_que->eims = 1 << (20 + i); adapter->ims |= rx_que->eims; adapter->ivars |= (8 | rx_que->msix) << (i * 4); } else if (adapter->hw.mac.type == e1000_82575) rx_que->eims = E1000_EICR_TX_QUEUE0 << vector; else rx_que->eims = 1 << vector; } rx_vectors = vector; vector = 0; for (i = 0; i < adapter->tx_num_queues; i++, tx_que++, vector++) { rid = vector + 1; snprintf(buf, sizeof(buf), "txq%d", i); tx_que = &adapter->tx_queues[i]; iflib_softirq_alloc_generic(ctx, &adapter->rx_queues[i % adapter->rx_num_queues].que_irq, IFLIB_INTR_TX, tx_que, tx_que->me, buf); tx_que->msix = (vector % adapter->tx_num_queues); /* * Set the bit to enable interrupt * in E1000_IMS -- bits 22 and 23 * are for TX0 and TX1, note this has * NOTHING to do with the MSIX vector */ if (adapter->hw.mac.type == e1000_82574) { tx_que->eims = 1 << (22 + i); adapter->ims |= tx_que->eims; adapter->ivars |= (8 | tx_que->msix) << (8 + (i * 4)); } else if (adapter->hw.mac.type == e1000_82575) { tx_que->eims = E1000_EICR_TX_QUEUE0 << (i % adapter->tx_num_queues); } else { tx_que->eims = 1 << (i % adapter->tx_num_queues); } } /* Link interrupt */ rid = rx_vectors + 1; error = iflib_irq_alloc_generic(ctx, &adapter->irq, rid, IFLIB_INTR_ADMIN, em_msix_link, adapter, 0, "aq"); if (error) { device_printf(iflib_get_dev(ctx), "Failed to register admin handler"); goto fail; } adapter->linkvec = rx_vectors; if (adapter->hw.mac.type < igb_mac_min) { adapter->ivars |= (8 | rx_vectors) << 16; adapter->ivars |= 0x80000000; } return (0); fail: iflib_irq_free(ctx, &adapter->irq); rx_que = adapter->rx_queues; for (int i = 0; i < adapter->rx_num_queues; i++, rx_que++) iflib_irq_free(ctx, &rx_que->que_irq); return (error); } static void igb_configure_queues(struct adapter *adapter) { struct e1000_hw *hw = &adapter->hw; struct em_rx_queue *rx_que; struct em_tx_queue *tx_que; u32 tmp, ivar = 0, newitr = 0; /* First turn on RSS capability */ if (adapter->hw.mac.type != e1000_82575) E1000_WRITE_REG(hw, E1000_GPIE, E1000_GPIE_MSIX_MODE | E1000_GPIE_EIAME | E1000_GPIE_PBA | E1000_GPIE_NSICR); /* Turn on MSIX */ switch (adapter->hw.mac.type) { case e1000_82580: case e1000_i350: case e1000_i354: case e1000_i210: case e1000_i211: case e1000_vfadapt: case e1000_vfadapt_i350: /* RX entries */ for (int i = 0; i < adapter->rx_num_queues; i++) { u32 index = i >> 1; ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index); rx_que = &adapter->rx_queues[i]; if (i & 1) { ivar &= 0xFF00FFFF; ivar |= (rx_que->msix | E1000_IVAR_VALID) << 16; } else { ivar &= 0xFFFFFF00; ivar |= rx_que->msix | E1000_IVAR_VALID; } E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar); } /* TX entries */ for (int i = 0; i < adapter->tx_num_queues; i++) { u32 index = i >> 1; ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index); tx_que = &adapter->tx_queues[i]; if (i & 1) { ivar &= 0x00FFFFFF; ivar |= (tx_que->msix | E1000_IVAR_VALID) << 24; } else { ivar &= 0xFFFF00FF; ivar |= (tx_que->msix | E1000_IVAR_VALID) << 8; } E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar); adapter->que_mask |= tx_que->eims; } /* And for the link interrupt */ ivar = (adapter->linkvec | E1000_IVAR_VALID) << 8; adapter->link_mask = 1 << adapter->linkvec; E1000_WRITE_REG(hw, E1000_IVAR_MISC, ivar); break; case e1000_82576: /* RX entries */ for (int i = 0; i < adapter->rx_num_queues; i++) { u32 index = i & 0x7; /* Each IVAR has two entries */ ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index); rx_que = &adapter->rx_queues[i]; if (i < 8) { ivar &= 0xFFFFFF00; ivar |= rx_que->msix | E1000_IVAR_VALID; } else { ivar &= 0xFF00FFFF; ivar |= (rx_que->msix | E1000_IVAR_VALID) << 16; } E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar); adapter->que_mask |= rx_que->eims; } /* TX entries */ for (int i = 0; i < adapter->tx_num_queues; i++) { u32 index = i & 0x7; /* Each IVAR has two entries */ ivar = E1000_READ_REG_ARRAY(hw, E1000_IVAR0, index); tx_que = &adapter->tx_queues[i]; if (i < 8) { ivar &= 0xFFFF00FF; ivar |= (tx_que->msix | E1000_IVAR_VALID) << 8; } else { ivar &= 0x00FFFFFF; ivar |= (tx_que->msix | E1000_IVAR_VALID) << 24; } E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, index, ivar); adapter->que_mask |= tx_que->eims; } /* And for the link interrupt */ ivar = (adapter->linkvec | E1000_IVAR_VALID) << 8; adapter->link_mask = 1 << adapter->linkvec; E1000_WRITE_REG(hw, E1000_IVAR_MISC, ivar); break; case e1000_82575: /* enable MSI-X support*/ tmp = E1000_READ_REG(hw, E1000_CTRL_EXT); tmp |= E1000_CTRL_EXT_PBA_CLR; /* Auto-Mask interrupts upon ICR read. */ tmp |= E1000_CTRL_EXT_EIAME; tmp |= E1000_CTRL_EXT_IRCA; E1000_WRITE_REG(hw, E1000_CTRL_EXT, tmp); /* Queues */ for (int i = 0; i < adapter->rx_num_queues; i++) { rx_que = &adapter->rx_queues[i]; tmp = E1000_EICR_RX_QUEUE0 << i; tmp |= E1000_EICR_TX_QUEUE0 << i; rx_que->eims = tmp; E1000_WRITE_REG_ARRAY(hw, E1000_MSIXBM(0), i, rx_que->eims); adapter->que_mask |= rx_que->eims; } /* Link */ E1000_WRITE_REG(hw, E1000_MSIXBM(adapter->linkvec), E1000_EIMS_OTHER); adapter->link_mask |= E1000_EIMS_OTHER; default: break; } /* Set the starting interrupt rate */ if (em_max_interrupt_rate > 0) newitr = (4000000 / em_max_interrupt_rate) & 0x7FFC; if (hw->mac.type == e1000_82575) newitr |= newitr << 16; else newitr |= E1000_EITR_CNT_IGNR; for (int i = 0; i < adapter->rx_num_queues; i++) { rx_que = &adapter->rx_queues[i]; E1000_WRITE_REG(hw, E1000_EITR(rx_que->msix), newitr); } return; } static void em_free_pci_resources(if_ctx_t ctx) { struct adapter *adapter = iflib_get_softc(ctx); struct em_rx_queue *que = adapter->rx_queues; device_t dev = iflib_get_dev(ctx); /* Release all msix queue resources */ if (adapter->intr_type == IFLIB_INTR_MSIX) iflib_irq_free(ctx, &adapter->irq); for (int i = 0; i < adapter->rx_num_queues; i++, que++) { iflib_irq_free(ctx, &que->que_irq); } /* First release all the interrupt resources */ if (adapter->memory != NULL) { bus_release_resource(dev, SYS_RES_MEMORY, PCIR_BAR(0), adapter->memory); adapter->memory = NULL; } if (adapter->flash != NULL) { bus_release_resource(dev, SYS_RES_MEMORY, EM_FLASH, adapter->flash); adapter->flash = NULL; } if (adapter->ioport != NULL) bus_release_resource(dev, SYS_RES_IOPORT, adapter->io_rid, adapter->ioport); } /* Setup MSI or MSI/X */ static int em_setup_msix(if_ctx_t ctx) { struct adapter *adapter = iflib_get_softc(ctx); if (adapter->hw.mac.type == e1000_82574) { em_enable_vectors_82574(ctx); } return (0); } /********************************************************************* * * Initialize the hardware to a configuration * as specified by the adapter structure. * **********************************************************************/ static void lem_smartspeed(struct adapter *adapter) { u16 phy_tmp; if (adapter->link_active || (adapter->hw.phy.type != e1000_phy_igp) || adapter->hw.mac.autoneg == 0 || (adapter->hw.phy.autoneg_advertised & ADVERTISE_1000_FULL) == 0) return; if (adapter->smartspeed == 0) { /* If Master/Slave config fault is asserted twice, * we assume back-to-back */ e1000_read_phy_reg(&adapter->hw, PHY_1000T_STATUS, &phy_tmp); if (!(phy_tmp & SR_1000T_MS_CONFIG_FAULT)) return; e1000_read_phy_reg(&adapter->hw, PHY_1000T_STATUS, &phy_tmp); if (phy_tmp & SR_1000T_MS_CONFIG_FAULT) { e1000_read_phy_reg(&adapter->hw, PHY_1000T_CTRL, &phy_tmp); if(phy_tmp & CR_1000T_MS_ENABLE) { phy_tmp &= ~CR_1000T_MS_ENABLE; e1000_write_phy_reg(&adapter->hw, PHY_1000T_CTRL, phy_tmp); adapter->smartspeed++; if(adapter->hw.mac.autoneg && !e1000_copper_link_autoneg(&adapter->hw) && !e1000_read_phy_reg(&adapter->hw, PHY_CONTROL, &phy_tmp)) { phy_tmp |= (MII_CR_AUTO_NEG_EN | MII_CR_RESTART_AUTO_NEG); e1000_write_phy_reg(&adapter->hw, PHY_CONTROL, phy_tmp); } } } return; } else if(adapter->smartspeed == EM_SMARTSPEED_DOWNSHIFT) { /* If still no link, perhaps using 2/3 pair cable */ e1000_read_phy_reg(&adapter->hw, PHY_1000T_CTRL, &phy_tmp); phy_tmp |= CR_1000T_MS_ENABLE; e1000_write_phy_reg(&adapter->hw, PHY_1000T_CTRL, phy_tmp); if(adapter->hw.mac.autoneg && !e1000_copper_link_autoneg(&adapter->hw) && !e1000_read_phy_reg(&adapter->hw, PHY_CONTROL, &phy_tmp)) { phy_tmp |= (MII_CR_AUTO_NEG_EN | MII_CR_RESTART_AUTO_NEG); e1000_write_phy_reg(&adapter->hw, PHY_CONTROL, phy_tmp); } } /* Restart process after EM_SMARTSPEED_MAX iterations */ if(adapter->smartspeed++ == EM_SMARTSPEED_MAX) adapter->smartspeed = 0; } /********************************************************************* * * Initialize the DMA Coalescing feature * **********************************************************************/ static void igb_init_dmac(struct adapter *adapter, u32 pba) { device_t dev = adapter->dev; struct e1000_hw *hw = &adapter->hw; u32 dmac, reg = ~E1000_DMACR_DMAC_EN; u16 hwm; u16 max_frame_size; if (hw->mac.type == e1000_i211) return; max_frame_size = adapter->shared->isc_max_frame_size; if (hw->mac.type > e1000_82580) { if (adapter->dmac == 0) { /* Disabling it */ E1000_WRITE_REG(hw, E1000_DMACR, reg); return; } else device_printf(dev, "DMA Coalescing enabled\n"); /* Set starting threshold */ E1000_WRITE_REG(hw, E1000_DMCTXTH, 0); hwm = 64 * pba - max_frame_size / 16; if (hwm < 64 * (pba - 6)) hwm = 64 * (pba - 6); reg = E1000_READ_REG(hw, E1000_FCRTC); reg &= ~E1000_FCRTC_RTH_COAL_MASK; reg |= ((hwm << E1000_FCRTC_RTH_COAL_SHIFT) & E1000_FCRTC_RTH_COAL_MASK); E1000_WRITE_REG(hw, E1000_FCRTC, reg); dmac = pba - max_frame_size / 512; if (dmac < pba - 10) dmac = pba - 10; reg = E1000_READ_REG(hw, E1000_DMACR); reg &= ~E1000_DMACR_DMACTHR_MASK; reg = ((dmac << E1000_DMACR_DMACTHR_SHIFT) & E1000_DMACR_DMACTHR_MASK); /* transition to L0x or L1 if available..*/ reg |= (E1000_DMACR_DMAC_EN | E1000_DMACR_DMAC_LX_MASK); /* Check if status is 2.5Gb backplane connection * before configuration of watchdog timer, which is * in msec values in 12.8usec intervals * watchdog timer= msec values in 32usec intervals * for non 2.5Gb connection */ if (hw->mac.type == e1000_i354) { int status = E1000_READ_REG(hw, E1000_STATUS); if ((status & E1000_STATUS_2P5_SKU) && (!(status & E1000_STATUS_2P5_SKU_OVER))) reg |= ((adapter->dmac * 5) >> 6); else reg |= (adapter->dmac >> 5); } else { reg |= (adapter->dmac >> 5); } E1000_WRITE_REG(hw, E1000_DMACR, reg); E1000_WRITE_REG(hw, E1000_DMCRTRH, 0); /* Set the interval before transition */ reg = E1000_READ_REG(hw, E1000_DMCTLX); if (hw->mac.type == e1000_i350) reg |= IGB_DMCTLX_DCFLUSH_DIS; /* ** in 2.5Gb connection, TTLX unit is 0.4 usec ** which is 0x4*2 = 0xA. But delay is still 4 usec */ if (hw->mac.type == e1000_i354) { int status = E1000_READ_REG(hw, E1000_STATUS); if ((status & E1000_STATUS_2P5_SKU) && (!(status & E1000_STATUS_2P5_SKU_OVER))) reg |= 0xA; else reg |= 0x4; } else { reg |= 0x4; } E1000_WRITE_REG(hw, E1000_DMCTLX, reg); /* free space in tx packet buffer to wake from DMA coal */ E1000_WRITE_REG(hw, E1000_DMCTXTH, (IGB_TXPBSIZE - (2 * max_frame_size)) >> 6); /* make low power state decision controlled by DMA coal */ reg = E1000_READ_REG(hw, E1000_PCIEMISC); reg &= ~E1000_PCIEMISC_LX_DECISION; E1000_WRITE_REG(hw, E1000_PCIEMISC, reg); } else if (hw->mac.type == e1000_82580) { u32 reg = E1000_READ_REG(hw, E1000_PCIEMISC); E1000_WRITE_REG(hw, E1000_PCIEMISC, reg & ~E1000_PCIEMISC_LX_DECISION); E1000_WRITE_REG(hw, E1000_DMACR, 0); } } static void em_reset(if_ctx_t ctx) { device_t dev = iflib_get_dev(ctx); struct adapter *adapter = iflib_get_softc(ctx); struct ifnet *ifp = iflib_get_ifp(ctx); struct e1000_hw *hw = &adapter->hw; u16 rx_buffer_size; u32 pba; INIT_DEBUGOUT("em_reset: begin"); /* Let the firmware know the OS is in control */ em_get_hw_control(adapter); /* Set up smart power down as default off on newer adapters. */ if (!em_smart_pwr_down && (hw->mac.type == e1000_82571 || hw->mac.type == e1000_82572)) { u16 phy_tmp = 0; /* Speed up time to link by disabling smart power down. */ e1000_read_phy_reg(hw, IGP02E1000_PHY_POWER_MGMT, &phy_tmp); phy_tmp &= ~IGP02E1000_PM_SPD; e1000_write_phy_reg(hw, IGP02E1000_PHY_POWER_MGMT, phy_tmp); } /* * Packet Buffer Allocation (PBA) * Writing PBA sets the receive portion of the buffer * the remainder is used for the transmit buffer. */ switch (hw->mac.type) { /* Total Packet Buffer on these is 48K */ case e1000_82571: case e1000_82572: case e1000_80003es2lan: pba = E1000_PBA_32K; /* 32K for Rx, 16K for Tx */ break; case e1000_82573: /* 82573: Total Packet Buffer is 32K */ pba = E1000_PBA_12K; /* 12K for Rx, 20K for Tx */ break; case e1000_82574: case e1000_82583: pba = E1000_PBA_20K; /* 20K for Rx, 20K for Tx */ break; case e1000_ich8lan: pba = E1000_PBA_8K; break; case e1000_ich9lan: case e1000_ich10lan: /* Boost Receive side for jumbo frames */ if (adapter->hw.mac.max_frame_size > 4096) pba = E1000_PBA_14K; else pba = E1000_PBA_10K; break; case e1000_pchlan: case e1000_pch2lan: case e1000_pch_lpt: case e1000_pch_spt: pba = E1000_PBA_26K; break; case e1000_82575: pba = E1000_PBA_32K; break; case e1000_82576: case e1000_vfadapt: pba = E1000_READ_REG(hw, E1000_RXPBS); pba &= E1000_RXPBS_SIZE_MASK_82576; break; case e1000_82580: case e1000_i350: case e1000_i354: case e1000_vfadapt_i350: pba = E1000_READ_REG(hw, E1000_RXPBS); pba = e1000_rxpbs_adjust_82580(pba); break; case e1000_i210: case e1000_i211: pba = E1000_PBA_34K; break; default: if (adapter->hw.mac.max_frame_size > 8192) pba = E1000_PBA_40K; /* 40K for Rx, 24K for Tx */ else pba = E1000_PBA_48K; /* 48K for Rx, 16K for Tx */ } /* Special needs in case of Jumbo frames */ if ((hw->mac.type == e1000_82575) && (ifp->if_mtu > ETHERMTU)) { u32 tx_space, min_tx, min_rx; pba = E1000_READ_REG(hw, E1000_PBA); tx_space = pba >> 16; pba &= 0xffff; min_tx = (adapter->hw.mac.max_frame_size + sizeof(struct e1000_tx_desc) - ETHERNET_FCS_SIZE) * 2; min_tx = roundup2(min_tx, 1024); min_tx >>= 10; min_rx = adapter->hw.mac.max_frame_size; min_rx = roundup2(min_rx, 1024); min_rx >>= 10; if (tx_space < min_tx && ((min_tx - tx_space) < pba)) { pba = pba - (min_tx - tx_space); /* * if short on rx space, rx wins * and must trump tx adjustment */ if (pba < min_rx) pba = min_rx; } E1000_WRITE_REG(hw, E1000_PBA, pba); } if (hw->mac.type < igb_mac_min) E1000_WRITE_REG(&adapter->hw, E1000_PBA, pba); INIT_DEBUGOUT1("em_reset: pba=%dK",pba); /* * These parameters control the automatic generation (Tx) and * response (Rx) to Ethernet PAUSE frames. * - High water mark should allow for at least two frames to be * received after sending an XOFF. * - Low water mark works best when it is very near the high water mark. * This allows the receiver to restart by sending XON when it has * drained a bit. Here we use an arbitrary value of 1500 which will * restart after one full frame is pulled from the buffer. There * could be several smaller frames in the buffer and if so they will * not trigger the XON until their total number reduces the buffer * by 1500. * - The pause time is fairly large at 1000 x 512ns = 512 usec. */ rx_buffer_size = (pba & 0xffff) << 10; hw->fc.high_water = rx_buffer_size - roundup2(adapter->hw.mac.max_frame_size, 1024); hw->fc.low_water = hw->fc.high_water - 1500; if (adapter->fc) /* locally set flow control value? */ hw->fc.requested_mode = adapter->fc; else hw->fc.requested_mode = e1000_fc_full; if (hw->mac.type == e1000_80003es2lan) hw->fc.pause_time = 0xFFFF; else hw->fc.pause_time = EM_FC_PAUSE_TIME; hw->fc.send_xon = TRUE; /* Device specific overrides/settings */ switch (hw->mac.type) { case e1000_pchlan: /* Workaround: no TX flow ctrl for PCH */ hw->fc.requested_mode = e1000_fc_rx_pause; hw->fc.pause_time = 0xFFFF; /* override */ if (if_getmtu(ifp) > ETHERMTU) { hw->fc.high_water = 0x3500; hw->fc.low_water = 0x1500; } else { hw->fc.high_water = 0x5000; hw->fc.low_water = 0x3000; } hw->fc.refresh_time = 0x1000; break; case e1000_pch2lan: case e1000_pch_lpt: case e1000_pch_spt: hw->fc.high_water = 0x5C20; hw->fc.low_water = 0x5048; hw->fc.pause_time = 0x0650; hw->fc.refresh_time = 0x0400; /* Jumbos need adjusted PBA */ if (if_getmtu(ifp) > ETHERMTU) E1000_WRITE_REG(hw, E1000_PBA, 12); else E1000_WRITE_REG(hw, E1000_PBA, 26); break; case e1000_82575: case e1000_82576: /* 8-byte granularity */ hw->fc.low_water = hw->fc.high_water - 8; break; case e1000_82580: case e1000_i350: case e1000_i354: case e1000_i210: case e1000_i211: case e1000_vfadapt: case e1000_vfadapt_i350: /* 16-byte granularity */ hw->fc.low_water = hw->fc.high_water - 16; break; case e1000_ich9lan: case e1000_ich10lan: if (if_getmtu(ifp) > ETHERMTU) { hw->fc.high_water = 0x2800; hw->fc.low_water = hw->fc.high_water - 8; break; } /* FALLTHROUGH */ default: if (hw->mac.type == e1000_80003es2lan) hw->fc.pause_time = 0xFFFF; break; } /* Issue a global reset */ e1000_reset_hw(hw); if (adapter->hw.mac.type >= igb_mac_min) { E1000_WRITE_REG(hw, E1000_WUC, 0); } else { E1000_WRITE_REG(hw, E1000_WUFC, 0); em_disable_aspm(adapter); } if (adapter->flags & IGB_MEDIA_RESET) { e1000_setup_init_funcs(hw, TRUE); e1000_get_bus_info(hw); adapter->flags &= ~IGB_MEDIA_RESET; } /* and a re-init */ if (e1000_init_hw(hw) < 0) { device_printf(dev, "Hardware Initialization Failed\n"); return; } if (adapter->hw.mac.type >= igb_mac_min) igb_init_dmac(adapter, pba); E1000_WRITE_REG(hw, E1000_VET, ETHERTYPE_VLAN); e1000_get_phy_info(hw); e1000_check_for_link(hw); } #define RSSKEYLEN 10 static void em_initialize_rss_mapping(struct adapter *adapter) { uint8_t rss_key[4 * RSSKEYLEN]; uint32_t reta = 0; struct e1000_hw *hw = &adapter->hw; int i; /* * Configure RSS key */ arc4rand(rss_key, sizeof(rss_key), 0); for (i = 0; i < RSSKEYLEN; ++i) { uint32_t rssrk = 0; rssrk = EM_RSSRK_VAL(rss_key, i); E1000_WRITE_REG(hw,E1000_RSSRK(i), rssrk); } /* * Configure RSS redirect table in following fashion: * (hash & ring_cnt_mask) == rdr_table[(hash & rdr_table_mask)] */ for (i = 0; i < sizeof(reta); ++i) { uint32_t q; q = (i % adapter->rx_num_queues) << 7; reta |= q << (8 * i); } for (i = 0; i < 32; ++i) E1000_WRITE_REG(hw, E1000_RETA(i), reta); E1000_WRITE_REG(hw, E1000_MRQC, E1000_MRQC_RSS_ENABLE_2Q | E1000_MRQC_RSS_FIELD_IPV4_TCP | E1000_MRQC_RSS_FIELD_IPV4 | E1000_MRQC_RSS_FIELD_IPV6_TCP_EX | E1000_MRQC_RSS_FIELD_IPV6_EX | E1000_MRQC_RSS_FIELD_IPV6); } static void igb_initialize_rss_mapping(struct adapter *adapter) { struct e1000_hw *hw = &adapter->hw; int i; int queue_id; u32 reta; u32 rss_key[10], mrqc, shift = 0; /* XXX? */ if (adapter->hw.mac.type == e1000_82575) shift = 6; /* * The redirection table controls which destination * queue each bucket redirects traffic to. * Each DWORD represents four queues, with the LSB * being the first queue in the DWORD. * * This just allocates buckets to queues using round-robin * allocation. * * NOTE: It Just Happens to line up with the default * RSS allocation method. */ /* Warning FM follows */ reta = 0; for (i = 0; i < 128; i++) { #ifdef RSS queue_id = rss_get_indirection_to_bucket(i); /* * If we have more queues than buckets, we'll * end up mapping buckets to a subset of the * queues. * * If we have more buckets than queues, we'll * end up instead assigning multiple buckets * to queues. * * Both are suboptimal, but we need to handle * the case so we don't go out of bounds * indexing arrays and such. */ queue_id = queue_id % adapter->rx_num_queues; #else queue_id = (i % adapter->rx_num_queues); #endif /* Adjust if required */ queue_id = queue_id << shift; /* * 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) { E1000_WRITE_REG(hw, E1000_RETA(i >> 2), reta); reta = 0; } } /* Now fill in hash table */ /* * MRQC: Multiple Receive Queues Command * Set queuing to RSS control, number depends on the device. */ mrqc = E1000_MRQC_ENABLE_RSS_8Q; #ifdef RSS /* XXX ew typecasting */ rss_getkey((uint8_t *) &rss_key); #else arc4rand(&rss_key, sizeof(rss_key), 0); #endif for (i = 0; i < 10; i++) E1000_WRITE_REG_ARRAY(hw, E1000_RSSRK(0), i, rss_key[i]); /* * Configure the RSS fields to hash upon. */ mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 | E1000_MRQC_RSS_FIELD_IPV4_TCP); mrqc |= (E1000_MRQC_RSS_FIELD_IPV6 | E1000_MRQC_RSS_FIELD_IPV6_TCP); mrqc |=( E1000_MRQC_RSS_FIELD_IPV4_UDP | E1000_MRQC_RSS_FIELD_IPV6_UDP); mrqc |=( E1000_MRQC_RSS_FIELD_IPV6_UDP_EX | E1000_MRQC_RSS_FIELD_IPV6_TCP_EX); E1000_WRITE_REG(hw, E1000_MRQC, mrqc); } /********************************************************************* * * Setup networking device structure and register an interface. * **********************************************************************/ static int em_setup_interface(if_ctx_t ctx) { struct ifnet *ifp = iflib_get_ifp(ctx); struct adapter *adapter = iflib_get_softc(ctx); if_softc_ctx_t scctx = adapter->shared; uint64_t cap = 0; INIT_DEBUGOUT("em_setup_interface: begin"); /* TSO parameters */ if_sethwtsomax(ifp, IP_MAXPACKET); /* Take m_pullup(9)'s in em_xmit() w/ TSO into acount. */ if_sethwtsomaxsegcount(ifp, EM_MAX_SCATTER - 5); if_sethwtsomaxsegsize(ifp, EM_TSO_SEG_SIZE); /* Single Queue */ if (adapter->tx_num_queues == 1) { if_setsendqlen(ifp, scctx->isc_ntxd[0] - 1); if_setsendqready(ifp); } cap = IFCAP_HWCSUM | IFCAP_VLAN_HWCSUM | IFCAP_TSO4; cap |= IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWTSO | IFCAP_VLAN_MTU; /* * Tell the upper layer(s) we * support full VLAN capability */ if_setifheaderlen(ifp, sizeof(struct ether_vlan_header)); if_setcapabilitiesbit(ifp, cap, 0); /* * 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 em driver you can * enable this and get full hardware tag filtering. */ if_setcapabilitiesbit(ifp, IFCAP_VLAN_HWFILTER,0); /* Enable only WOL MAGIC by default */ if (adapter->wol) { if_setcapenablebit(ifp, IFCAP_WOL_MAGIC, IFCAP_WOL_MCAST| IFCAP_WOL_UCAST); } else { if_setcapenablebit(ifp, 0, IFCAP_WOL_MAGIC | IFCAP_WOL_MCAST| IFCAP_WOL_UCAST); } /* * Specify the media types supported by this adapter and register * callbacks to update media and link information */ if ((adapter->hw.phy.media_type == e1000_media_type_fiber) || (adapter->hw.phy.media_type == e1000_media_type_internal_serdes)) { u_char fiber_type = IFM_1000_SX; /* default type */ if (adapter->hw.mac.type == e1000_82545) fiber_type = IFM_1000_LX; ifmedia_add(adapter->media, IFM_ETHER | fiber_type | IFM_FDX, 0, NULL); ifmedia_add(adapter->media, IFM_ETHER | fiber_type, 0, NULL); } else { ifmedia_add(adapter->media, IFM_ETHER | IFM_10_T, 0, NULL); ifmedia_add(adapter->media, IFM_ETHER | IFM_10_T | IFM_FDX, 0, NULL); ifmedia_add(adapter->media, IFM_ETHER | IFM_100_TX, 0, NULL); ifmedia_add(adapter->media, IFM_ETHER | IFM_100_TX | IFM_FDX, 0, NULL); if (adapter->hw.phy.type != e1000_phy_ife) { 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); ifmedia_set(adapter->media, IFM_ETHER | IFM_AUTO); return (0); } static int em_if_tx_queues_alloc(if_ctx_t ctx, caddr_t *vaddrs, uint64_t *paddrs, int ntxqs, int ntxqsets) { struct adapter *adapter = iflib_get_softc(ctx); if_softc_ctx_t scctx = adapter->shared; int error = E1000_SUCCESS; struct em_tx_queue *que; int i, j; MPASS(adapter->tx_num_queues > 0); MPASS(adapter->tx_num_queues == ntxqsets); /* First allocate the top level queue structs */ if (!(adapter->tx_queues = (struct em_tx_queue *) malloc(sizeof(struct em_tx_queue) * adapter->tx_num_queues, M_DEVBUF, M_NOWAIT | M_ZERO))) { device_printf(iflib_get_dev(ctx), "Unable to allocate queue memory\n"); return(ENOMEM); } for (i = 0, que = adapter->tx_queues; i < adapter->tx_num_queues; i++, que++) { /* Set up some basics */ struct tx_ring *txr = &que->txr; txr->adapter = que->adapter = adapter; que->me = txr->me = i; /* Allocate report status array */ if (!(txr->tx_rsq = (qidx_t *) malloc(sizeof(qidx_t) * scctx->isc_ntxd[0], M_DEVBUF, M_NOWAIT | M_ZERO))) { device_printf(iflib_get_dev(ctx), "failed to allocate rs_idxs memory\n"); error = ENOMEM; goto fail; } for (j = 0; j < scctx->isc_ntxd[0]; j++) txr->tx_rsq[j] = QIDX_INVALID; /* get the virtual and physical address of the hardware queues */ txr->tx_base = (struct e1000_tx_desc *)vaddrs[i*ntxqs]; txr->tx_paddr = paddrs[i*ntxqs]; } device_printf(iflib_get_dev(ctx), "allocated for %d tx_queues\n", adapter->tx_num_queues); return (0); fail: em_if_queues_free(ctx); return (error); } static int em_if_rx_queues_alloc(if_ctx_t ctx, caddr_t *vaddrs, uint64_t *paddrs, int nrxqs, int nrxqsets) { struct adapter *adapter = iflib_get_softc(ctx); int error = E1000_SUCCESS; struct em_rx_queue *que; int i; MPASS(adapter->rx_num_queues > 0); MPASS(adapter->rx_num_queues == nrxqsets); /* First allocate the top level queue structs */ if (!(adapter->rx_queues = (struct em_rx_queue *) malloc(sizeof(struct em_rx_queue) * adapter->rx_num_queues, M_DEVBUF, M_NOWAIT | M_ZERO))) { device_printf(iflib_get_dev(ctx), "Unable to allocate queue memory\n"); error = ENOMEM; goto fail; } for (i = 0, que = adapter->rx_queues; i < nrxqsets; i++, que++) { /* Set up some basics */ struct rx_ring *rxr = &que->rxr; rxr->adapter = que->adapter = adapter; rxr->que = que; que->me = rxr->me = i; /* get the virtual and physical address of the hardware queues */ rxr->rx_base = (union e1000_rx_desc_extended *)vaddrs[i*nrxqs]; rxr->rx_paddr = paddrs[i*nrxqs]; } device_printf(iflib_get_dev(ctx), "allocated for %d rx_queues\n", adapter->rx_num_queues); return (0); fail: em_if_queues_free(ctx); return (error); } static void em_if_queues_free(if_ctx_t ctx) { struct adapter *adapter = iflib_get_softc(ctx); struct em_tx_queue *tx_que = adapter->tx_queues; struct em_rx_queue *rx_que = adapter->rx_queues; if (tx_que != NULL) { for (int i = 0; i < adapter->tx_num_queues; i++, tx_que++) { struct tx_ring *txr = &tx_que->txr; if (txr->tx_rsq == NULL) break; free(txr->tx_rsq, M_DEVBUF); txr->tx_rsq = NULL; } free(adapter->tx_queues, M_DEVBUF); adapter->tx_queues = NULL; } if (rx_que != NULL) { free(adapter->rx_queues, M_DEVBUF); adapter->rx_queues = NULL; } em_release_hw_control(adapter); if (adapter->mta != NULL) { free(adapter->mta, M_DEVBUF); } } /********************************************************************* * * Enable transmit unit. * **********************************************************************/ static void em_initialize_transmit_unit(if_ctx_t ctx) { struct adapter *adapter = iflib_get_softc(ctx); if_softc_ctx_t scctx = adapter->shared; struct em_tx_queue *que; struct tx_ring *txr; struct e1000_hw *hw = &adapter->hw; u32 tctl, txdctl = 0, tarc, tipg = 0; INIT_DEBUGOUT("em_initialize_transmit_unit: begin"); for (int i = 0; i < adapter->tx_num_queues; i++, txr++) { u64 bus_addr; caddr_t offp, endp; que = &adapter->tx_queues[i]; txr = &que->txr; bus_addr = txr->tx_paddr; /* Clear checksum offload context. */ offp = (caddr_t)&txr->csum_flags; endp = (caddr_t)(txr + 1); bzero(offp, endp - offp); /* Base and Len of TX Ring */ E1000_WRITE_REG(hw, E1000_TDLEN(i), scctx->isc_ntxd[0] * sizeof(struct e1000_tx_desc)); E1000_WRITE_REG(hw, E1000_TDBAH(i), (u32)(bus_addr >> 32)); E1000_WRITE_REG(hw, E1000_TDBAL(i), (u32)bus_addr); /* Init the HEAD/TAIL indices */ E1000_WRITE_REG(hw, E1000_TDT(i), 0); E1000_WRITE_REG(hw, E1000_TDH(i), 0); HW_DEBUGOUT2("Base = %x, Length = %x\n", E1000_READ_REG(&adapter->hw, E1000_TDBAL(i)), E1000_READ_REG(&adapter->hw, E1000_TDLEN(i))); txdctl = 0; /* clear txdctl */ txdctl |= 0x1f; /* PTHRESH */ txdctl |= 1 << 8; /* HTHRESH */ txdctl |= 1 << 16;/* WTHRESH */ txdctl |= 1 << 22; /* Reserved bit 22 must always be 1 */ txdctl |= E1000_TXDCTL_GRAN; txdctl |= 1 << 25; /* LWTHRESH */ E1000_WRITE_REG(hw, E1000_TXDCTL(i), txdctl); } /* Set the default values for the Tx Inter Packet Gap timer */ switch (adapter->hw.mac.type) { case e1000_80003es2lan: tipg = DEFAULT_82543_TIPG_IPGR1; tipg |= DEFAULT_80003ES2LAN_TIPG_IPGR2 << E1000_TIPG_IPGR2_SHIFT; break; case e1000_82542: tipg = DEFAULT_82542_TIPG_IPGT; tipg |= DEFAULT_82542_TIPG_IPGR1 << E1000_TIPG_IPGR1_SHIFT; tipg |= DEFAULT_82542_TIPG_IPGR2 << E1000_TIPG_IPGR2_SHIFT; break; default: if ((adapter->hw.phy.media_type == e1000_media_type_fiber) || (adapter->hw.phy.media_type == e1000_media_type_internal_serdes)) tipg = DEFAULT_82543_TIPG_IPGT_FIBER; else tipg = DEFAULT_82543_TIPG_IPGT_COPPER; tipg |= DEFAULT_82543_TIPG_IPGR1 << E1000_TIPG_IPGR1_SHIFT; tipg |= DEFAULT_82543_TIPG_IPGR2 << E1000_TIPG_IPGR2_SHIFT; } E1000_WRITE_REG(&adapter->hw, E1000_TIPG, tipg); E1000_WRITE_REG(&adapter->hw, E1000_TIDV, adapter->tx_int_delay.value); if(adapter->hw.mac.type >= e1000_82540) E1000_WRITE_REG(&adapter->hw, E1000_TADV, adapter->tx_abs_int_delay.value); if ((adapter->hw.mac.type == e1000_82571) || (adapter->hw.mac.type == e1000_82572)) { tarc = E1000_READ_REG(&adapter->hw, E1000_TARC(0)); tarc |= TARC_SPEED_MODE_BIT; E1000_WRITE_REG(&adapter->hw, E1000_TARC(0), tarc); } else if (adapter->hw.mac.type == e1000_80003es2lan) { /* errata: program both queues to unweighted RR */ tarc = E1000_READ_REG(&adapter->hw, E1000_TARC(0)); tarc |= 1; E1000_WRITE_REG(&adapter->hw, E1000_TARC(0), tarc); tarc = E1000_READ_REG(&adapter->hw, E1000_TARC(1)); tarc |= 1; E1000_WRITE_REG(&adapter->hw, E1000_TARC(1), tarc); } else if (adapter->hw.mac.type == e1000_82574) { tarc = E1000_READ_REG(&adapter->hw, E1000_TARC(0)); tarc |= TARC_ERRATA_BIT; if ( adapter->tx_num_queues > 1) { tarc |= (TARC_COMPENSATION_MODE | TARC_MQ_FIX); E1000_WRITE_REG(&adapter->hw, E1000_TARC(0), tarc); E1000_WRITE_REG(&adapter->hw, E1000_TARC(1), tarc); } else E1000_WRITE_REG(&adapter->hw, E1000_TARC(0), tarc); } if (adapter->tx_int_delay.value > 0) adapter->txd_cmd |= E1000_TXD_CMD_IDE; /* Program the Transmit Control Register */ tctl = E1000_READ_REG(&adapter->hw, E1000_TCTL); tctl &= ~E1000_TCTL_CT; tctl |= (E1000_TCTL_PSP | E1000_TCTL_RTLC | E1000_TCTL_EN | (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT)); if (adapter->hw.mac.type >= e1000_82571) tctl |= E1000_TCTL_MULR; /* This write will effectively turn on the transmit unit. */ E1000_WRITE_REG(&adapter->hw, E1000_TCTL, tctl); if (hw->mac.type == e1000_pch_spt) { u32 reg; reg = E1000_READ_REG(hw, E1000_IOSFPC); reg |= E1000_RCTL_RDMTS_HEX; E1000_WRITE_REG(hw, E1000_IOSFPC, reg); reg = E1000_READ_REG(hw, E1000_TARC(0)); reg |= E1000_TARC0_CB_MULTIQ_3_REQ; E1000_WRITE_REG(hw, E1000_TARC(0), reg); } } /********************************************************************* * * Enable receive unit. * **********************************************************************/ static void em_initialize_receive_unit(if_ctx_t ctx) { struct adapter *adapter = iflib_get_softc(ctx); if_softc_ctx_t scctx = adapter->shared; struct ifnet *ifp = iflib_get_ifp(ctx); struct e1000_hw *hw = &adapter->hw; struct em_rx_queue *que; int i; u32 rctl, rxcsum, rfctl; INIT_DEBUGOUT("em_initialize_receive_units: begin"); /* * Make sure receives are disabled while setting * up the descriptor ring */ rctl = E1000_READ_REG(hw, E1000_RCTL); /* Do not disable if ever enabled on this hardware */ if ((hw->mac.type != e1000_82574) && (hw->mac.type != e1000_82583)) E1000_WRITE_REG(hw, E1000_RCTL, rctl & ~E1000_RCTL_EN); /* Setup the Receive Control Register */ rctl &= ~(3 << E1000_RCTL_MO_SHIFT); rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF | (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT); /* Do not store bad packets */ rctl &= ~E1000_RCTL_SBP; /* Enable Long Packet receive */ if (if_getmtu(ifp) > ETHERMTU) rctl |= E1000_RCTL_LPE; else rctl &= ~E1000_RCTL_LPE; /* Strip the CRC */ if (!em_disable_crc_stripping) rctl |= E1000_RCTL_SECRC; if (adapter->hw.mac.type >= e1000_82540) { E1000_WRITE_REG(&adapter->hw, E1000_RADV, adapter->rx_abs_int_delay.value); /* * Set the interrupt throttling rate. Value is calculated * as DEFAULT_ITR = 1/(MAX_INTS_PER_SEC * 256ns) */ E1000_WRITE_REG(hw, E1000_ITR, DEFAULT_ITR); } E1000_WRITE_REG(&adapter->hw, E1000_RDTR, adapter->rx_int_delay.value); /* Use extended rx descriptor formats */ rfctl = E1000_READ_REG(hw, E1000_RFCTL); rfctl |= E1000_RFCTL_EXTEN; /* * When using MSIX interrupts we need to throttle * using the EITR register (82574 only) */ if (hw->mac.type == e1000_82574) { for (int i = 0; i < 4; i++) E1000_WRITE_REG(hw, E1000_EITR_82574(i), DEFAULT_ITR); /* Disable accelerated acknowledge */ rfctl |= E1000_RFCTL_ACK_DIS; } E1000_WRITE_REG(hw, E1000_RFCTL, rfctl); rxcsum = E1000_READ_REG(hw, E1000_RXCSUM); if (if_getcapenable(ifp) & IFCAP_RXCSUM && adapter->hw.mac.type >= e1000_82543) { if (adapter->tx_num_queues > 1) { if (adapter->hw.mac.type >= igb_mac_min) { rxcsum |= E1000_RXCSUM_PCSD; if (hw->mac.type != e1000_82575) rxcsum |= E1000_RXCSUM_CRCOFL; } else rxcsum |= E1000_RXCSUM_TUOFL | E1000_RXCSUM_IPOFL | E1000_RXCSUM_PCSD; } else { if (adapter->hw.mac.type >= igb_mac_min) rxcsum |= E1000_RXCSUM_IPPCSE; else rxcsum |= E1000_RXCSUM_TUOFL | E1000_RXCSUM_IPOFL; if (adapter->hw.mac.type > e1000_82575) rxcsum |= E1000_RXCSUM_CRCOFL; } } else rxcsum &= ~E1000_RXCSUM_TUOFL; E1000_WRITE_REG(hw, E1000_RXCSUM, rxcsum); if (adapter->rx_num_queues > 1) { if (adapter->hw.mac.type >= igb_mac_min) igb_initialize_rss_mapping(adapter); else em_initialize_rss_mapping(adapter); } /* * XXX TEMPORARY WORKAROUND: on some systems with 82573 * long latencies are observed, like Lenovo X60. This * change eliminates the problem, but since having positive * values in RDTR is a known source of problems on other * platforms another solution is being sought. */ if (hw->mac.type == e1000_82573) E1000_WRITE_REG(hw, E1000_RDTR, 0x20); for (i = 0, que = adapter->rx_queues; i < adapter->rx_num_queues; i++, que++) { struct rx_ring *rxr = &que->rxr; /* Setup the Base and Length of the Rx Descriptor Ring */ u64 bus_addr = rxr->rx_paddr; #if 0 u32 rdt = adapter->rx_num_queues -1; /* default */ #endif E1000_WRITE_REG(hw, E1000_RDLEN(i), scctx->isc_nrxd[0] * sizeof(union e1000_rx_desc_extended)); E1000_WRITE_REG(hw, E1000_RDBAH(i), (u32)(bus_addr >> 32)); E1000_WRITE_REG(hw, E1000_RDBAL(i), (u32)bus_addr); /* Setup the Head and Tail Descriptor Pointers */ E1000_WRITE_REG(hw, E1000_RDH(i), 0); E1000_WRITE_REG(hw, E1000_RDT(i), 0); } /* * Set PTHRESH for improved jumbo performance * According to 10.2.5.11 of Intel 82574 Datasheet, * RXDCTL(1) is written whenever RXDCTL(0) is written. * Only write to RXDCTL(1) if there is a need for different * settings. */ if (((adapter->hw.mac.type == e1000_ich9lan) || (adapter->hw.mac.type == e1000_pch2lan) || (adapter->hw.mac.type == e1000_ich10lan)) && (if_getmtu(ifp) > ETHERMTU)) { u32 rxdctl = E1000_READ_REG(hw, E1000_RXDCTL(0)); E1000_WRITE_REG(hw, E1000_RXDCTL(0), rxdctl | 3); } else if (adapter->hw.mac.type == e1000_82574) { for (int i = 0; i < adapter->rx_num_queues; i++) { u32 rxdctl = E1000_READ_REG(hw, E1000_RXDCTL(i)); rxdctl |= 0x20; /* PTHRESH */ rxdctl |= 4 << 8; /* HTHRESH */ rxdctl |= 4 << 16;/* WTHRESH */ rxdctl |= 1 << 24; /* Switch to granularity */ E1000_WRITE_REG(hw, E1000_RXDCTL(i), rxdctl); } } else if (adapter->hw.mac.type >= igb_mac_min) { u32 psize, srrctl = 0; if (if_getmtu(ifp) > ETHERMTU) { /* Set maximum packet len */ if (adapter->rx_mbuf_sz <= 4096) { srrctl |= 4096 >> E1000_SRRCTL_BSIZEPKT_SHIFT; rctl |= E1000_RCTL_SZ_4096 | E1000_RCTL_BSEX; } else if (adapter->rx_mbuf_sz > 4096) { srrctl |= 8192 >> E1000_SRRCTL_BSIZEPKT_SHIFT; rctl |= E1000_RCTL_SZ_8192 | E1000_RCTL_BSEX; } psize = scctx->isc_max_frame_size; /* are we on a vlan? */ if (ifp->if_vlantrunk != NULL) psize += VLAN_TAG_SIZE; E1000_WRITE_REG(&adapter->hw, E1000_RLPML, psize); } else { srrctl |= 2048 >> E1000_SRRCTL_BSIZEPKT_SHIFT; rctl |= E1000_RCTL_SZ_2048; } /* * If TX flow control is disabled and there's >1 queue defined, * enable DROP. * * This drops frames rather than hanging the RX MAC for all queues. */ if ((adapter->rx_num_queues > 1) && (adapter->fc == e1000_fc_none || adapter->fc == e1000_fc_rx_pause)) { srrctl |= E1000_SRRCTL_DROP_EN; } /* Setup the Base and Length of the Rx Descriptor Rings */ for (i = 0, que = adapter->rx_queues; i < adapter->rx_num_queues; i++, que++) { struct rx_ring *rxr = &que->rxr; u64 bus_addr = rxr->rx_paddr; u32 rxdctl; #ifdef notyet /* Configure for header split? -- ignore for now */ rxr->hdr_split = igb_header_split; #else srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF; #endif E1000_WRITE_REG(hw, E1000_RDLEN(i), scctx->isc_nrxd[0] * sizeof(struct e1000_rx_desc)); E1000_WRITE_REG(hw, E1000_RDBAH(i), (uint32_t)(bus_addr >> 32)); E1000_WRITE_REG(hw, E1000_RDBAL(i), (uint32_t)bus_addr); E1000_WRITE_REG(hw, E1000_SRRCTL(i), srrctl); /* Enable this Queue */ rxdctl = E1000_READ_REG(hw, E1000_RXDCTL(i)); rxdctl |= E1000_RXDCTL_QUEUE_ENABLE; rxdctl &= 0xFFF00000; rxdctl |= IGB_RX_PTHRESH; rxdctl |= IGB_RX_HTHRESH << 8; rxdctl |= IGB_RX_WTHRESH << 16; E1000_WRITE_REG(hw, E1000_RXDCTL(i), rxdctl); } } else if (adapter->hw.mac.type >= e1000_pch2lan) { if (if_getmtu(ifp) > ETHERMTU) e1000_lv_jumbo_workaround_ich8lan(hw, TRUE); else e1000_lv_jumbo_workaround_ich8lan(hw, FALSE); } /* Make sure VLAN Filters are off */ rctl &= ~E1000_RCTL_VFE; if (adapter->hw.mac.type < igb_mac_min) { if (adapter->rx_mbuf_sz == MCLBYTES) rctl |= E1000_RCTL_SZ_2048; else if (adapter->rx_mbuf_sz == MJUMPAGESIZE) rctl |= E1000_RCTL_SZ_4096 | E1000_RCTL_BSEX; else if (adapter->rx_mbuf_sz > MJUMPAGESIZE) rctl |= E1000_RCTL_SZ_8192 | E1000_RCTL_BSEX; /* ensure we clear use DTYPE of 00 here */ rctl &= ~0x00000C00; } /* Write out the settings */ E1000_WRITE_REG(hw, E1000_RCTL, rctl); return; } static void em_if_vlan_register(if_ctx_t ctx, u16 vtag) { struct adapter *adapter = iflib_get_softc(ctx); u32 index, bit; index = (vtag >> 5) & 0x7F; bit = vtag & 0x1F; adapter->shadow_vfta[index] |= (1 << bit); ++adapter->num_vlans; } static void em_if_vlan_unregister(if_ctx_t ctx, u16 vtag) { struct adapter *adapter = iflib_get_softc(ctx); u32 index, bit; index = (vtag >> 5) & 0x7F; bit = vtag & 0x1F; adapter->shadow_vfta[index] &= ~(1 << bit); --adapter->num_vlans; } static void em_setup_vlan_hw_support(struct adapter *adapter) { struct e1000_hw *hw = &adapter->hw; u32 reg; /* * 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; /* * A soft reset zero's out the VFTA, so * we need to repopulate it now. */ for (int i = 0; i < EM_VFTA_SIZE; i++) if (adapter->shadow_vfta[i] != 0) E1000_WRITE_REG_ARRAY(hw, E1000_VFTA, i, adapter->shadow_vfta[i]); reg = E1000_READ_REG(hw, E1000_CTRL); reg |= E1000_CTRL_VME; E1000_WRITE_REG(hw, E1000_CTRL, reg); /* Enable the Filter Table */ reg = E1000_READ_REG(hw, E1000_RCTL); reg &= ~E1000_RCTL_CFIEN; reg |= E1000_RCTL_VFE; E1000_WRITE_REG(hw, E1000_RCTL, reg); } static void em_if_enable_intr(if_ctx_t ctx) { struct adapter *adapter = iflib_get_softc(ctx); struct e1000_hw *hw = &adapter->hw; u32 ims_mask = IMS_ENABLE_MASK; if (hw->mac.type == e1000_82574) { E1000_WRITE_REG(hw, EM_EIAC, EM_MSIX_MASK); ims_mask |= adapter->ims; } else if (adapter->intr_type == IFLIB_INTR_MSIX && hw->mac.type >= igb_mac_min) { u32 mask = (adapter->que_mask | adapter->link_mask); E1000_WRITE_REG(&adapter->hw, E1000_EIAC, mask); E1000_WRITE_REG(&adapter->hw, E1000_EIAM, mask); E1000_WRITE_REG(&adapter->hw, E1000_EIMS, mask); ims_mask = E1000_IMS_LSC; } E1000_WRITE_REG(hw, E1000_IMS, ims_mask); } static void em_if_disable_intr(if_ctx_t ctx) { struct adapter *adapter = iflib_get_softc(ctx); struct e1000_hw *hw = &adapter->hw; if (adapter->intr_type == IFLIB_INTR_MSIX) { if (hw->mac.type >= igb_mac_min) E1000_WRITE_REG(&adapter->hw, E1000_EIMC, ~0); E1000_WRITE_REG(&adapter->hw, E1000_EIAC, 0); } E1000_WRITE_REG(&adapter->hw, E1000_IMC, 0xffffffff); } /* * Bit of a misnomer, what this really means is * to enable OS management of the system... aka * to disable special hardware management features */ static void em_init_manageability(struct adapter *adapter) { /* A shared code workaround */ #define E1000_82542_MANC2H E1000_MANC2H if (adapter->has_manage) { int manc2h = E1000_READ_REG(&adapter->hw, E1000_MANC2H); int manc = E1000_READ_REG(&adapter->hw, E1000_MANC); /* disable hardware interception of ARP */ manc &= ~(E1000_MANC_ARP_EN); /* enable receiving management packets to the host */ manc |= E1000_MANC_EN_MNG2HOST; #define E1000_MNG2HOST_PORT_623 (1 << 5) #define E1000_MNG2HOST_PORT_664 (1 << 6) manc2h |= E1000_MNG2HOST_PORT_623; manc2h |= E1000_MNG2HOST_PORT_664; E1000_WRITE_REG(&adapter->hw, E1000_MANC2H, manc2h); E1000_WRITE_REG(&adapter->hw, E1000_MANC, manc); } } /* * Give control back to hardware management * controller if there is one. */ static void em_release_manageability(struct adapter *adapter) { if (adapter->has_manage) { int manc = E1000_READ_REG(&adapter->hw, E1000_MANC); /* re-enable hardware interception of ARP */ manc |= E1000_MANC_ARP_EN; manc &= ~E1000_MANC_EN_MNG2HOST; E1000_WRITE_REG(&adapter->hw, E1000_MANC, manc); } } /* * em_get_hw_control sets the {CTRL_EXT|FWSM}:DRV_LOAD bit. * For ASF and Pass Through versions of f/w this means * that the driver is loaded. For AMT version type f/w * this means that the network i/f is open. */ static void em_get_hw_control(struct adapter *adapter) { u32 ctrl_ext, swsm; if (adapter->vf_ifp) return; if (adapter->hw.mac.type == e1000_82573) { swsm = E1000_READ_REG(&adapter->hw, E1000_SWSM); E1000_WRITE_REG(&adapter->hw, E1000_SWSM, swsm | E1000_SWSM_DRV_LOAD); return; } /* else */ ctrl_ext = E1000_READ_REG(&adapter->hw, E1000_CTRL_EXT); E1000_WRITE_REG(&adapter->hw, E1000_CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD); } /* * em_release_hw_control resets {CTRL_EXT|FWSM}:DRV_LOAD bit. * For ASF and Pass Through versions of f/w this means that * the driver is no longer loaded. For AMT versions of the * f/w this means that the network i/f is closed. */ static void em_release_hw_control(struct adapter *adapter) { u32 ctrl_ext, swsm; if (!adapter->has_manage) return; if (adapter->hw.mac.type == e1000_82573) { swsm = E1000_READ_REG(&adapter->hw, E1000_SWSM); E1000_WRITE_REG(&adapter->hw, E1000_SWSM, swsm & ~E1000_SWSM_DRV_LOAD); return; } /* else */ ctrl_ext = E1000_READ_REG(&adapter->hw, E1000_CTRL_EXT); E1000_WRITE_REG(&adapter->hw, E1000_CTRL_EXT, ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD); return; } static int em_is_valid_ether_addr(u8 *addr) { char zero_addr[6] = { 0, 0, 0, 0, 0, 0 }; if ((addr[0] & 1) || (!bcmp(addr, zero_addr, ETHER_ADDR_LEN))) { return (FALSE); } return (TRUE); } /* ** Parse the interface capabilities with regard ** to both system management and wake-on-lan for ** later use. */ static void em_get_wakeup(if_ctx_t ctx) { struct adapter *adapter = iflib_get_softc(ctx); device_t dev = iflib_get_dev(ctx); u16 eeprom_data = 0, device_id, apme_mask; adapter->has_manage = e1000_enable_mng_pass_thru(&adapter->hw); apme_mask = EM_EEPROM_APME; switch (adapter->hw.mac.type) { case e1000_82542: case e1000_82543: break; case e1000_82544: e1000_read_nvm(&adapter->hw, NVM_INIT_CONTROL2_REG, 1, &eeprom_data); apme_mask = EM_82544_APME; break; case e1000_82546: case e1000_82546_rev_3: if (adapter->hw.bus.func == 1) { e1000_read_nvm(&adapter->hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data); break; } else e1000_read_nvm(&adapter->hw, NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data); break; case e1000_82573: case e1000_82583: adapter->has_amt = TRUE; /* FALLTHROUGH */ case e1000_82571: case e1000_82572: case e1000_80003es2lan: if (adapter->hw.bus.func == 1) { e1000_read_nvm(&adapter->hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data); break; } else e1000_read_nvm(&adapter->hw, NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data); break; case e1000_ich8lan: case e1000_ich9lan: case e1000_ich10lan: case e1000_pchlan: case e1000_pch2lan: case e1000_pch_lpt: case e1000_pch_spt: case e1000_82575: /* listing all igb devices */ case e1000_82576: case e1000_82580: case e1000_i350: case e1000_i354: case e1000_i210: case e1000_i211: case e1000_vfadapt: case e1000_vfadapt_i350: apme_mask = E1000_WUC_APME; adapter->has_amt = TRUE; eeprom_data = E1000_READ_REG(&adapter->hw, E1000_WUC); break; default: e1000_read_nvm(&adapter->hw, NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data); break; } if (eeprom_data & apme_mask) adapter->wol = (E1000_WUFC_MAG | E1000_WUFC_MC); /* * We have the eeprom settings, now apply the special cases * where the eeprom may be wrong or the board won't support * wake on lan on a particular port */ device_id = pci_get_device(dev); switch (device_id) { case E1000_DEV_ID_82546GB_PCIE: adapter->wol = 0; break; case E1000_DEV_ID_82546EB_FIBER: case E1000_DEV_ID_82546GB_FIBER: /* Wake events only supported on port A for dual fiber * regardless of eeprom setting */ if (E1000_READ_REG(&adapter->hw, E1000_STATUS) & E1000_STATUS_FUNC_1) adapter->wol = 0; break; case E1000_DEV_ID_82546GB_QUAD_COPPER_KSP3: /* if quad port adapter, disable WoL on all but port A */ if (global_quad_port_a != 0) adapter->wol = 0; /* Reset for multiple quad port adapters */ if (++global_quad_port_a == 4) global_quad_port_a = 0; break; case E1000_DEV_ID_82571EB_FIBER: /* Wake events only supported on port A for dual fiber * regardless of eeprom setting */ if (E1000_READ_REG(&adapter->hw, E1000_STATUS) & E1000_STATUS_FUNC_1) adapter->wol = 0; break; case E1000_DEV_ID_82571EB_QUAD_COPPER: case E1000_DEV_ID_82571EB_QUAD_FIBER: case E1000_DEV_ID_82571EB_QUAD_COPPER_LP: /* if quad port adapter, disable WoL on all but port A */ if (global_quad_port_a != 0) adapter->wol = 0; /* Reset for multiple quad port adapters */ if (++global_quad_port_a == 4) global_quad_port_a = 0; break; } return; } /* * Enable PCI Wake On Lan capability */ static void em_enable_wakeup(if_ctx_t ctx) { struct adapter *adapter = iflib_get_softc(ctx); device_t dev = iflib_get_dev(ctx); if_t ifp = iflib_get_ifp(ctx); int error = 0; u32 pmc, ctrl, ctrl_ext, rctl; u16 status; if (pci_find_cap(dev, PCIY_PMG, &pmc) != 0) return; /* * Determine type of Wakeup: note that wol * is set with all bits on by default. */ if ((if_getcapenable(ifp) & IFCAP_WOL_MAGIC) == 0) adapter->wol &= ~E1000_WUFC_MAG; if ((if_getcapenable(ifp) & IFCAP_WOL_UCAST) == 0) adapter->wol &= ~E1000_WUFC_EX; if ((if_getcapenable(ifp) & IFCAP_WOL_MCAST) == 0) adapter->wol &= ~E1000_WUFC_MC; else { rctl = E1000_READ_REG(&adapter->hw, E1000_RCTL); rctl |= E1000_RCTL_MPE; E1000_WRITE_REG(&adapter->hw, E1000_RCTL, rctl); } if (!(adapter->wol & (E1000_WUFC_EX | E1000_WUFC_MAG | E1000_WUFC_MC))) goto pme; /* Advertise the wakeup capability */ ctrl = E1000_READ_REG(&adapter->hw, E1000_CTRL); ctrl |= (E1000_CTRL_SWDPIN2 | E1000_CTRL_SWDPIN3); E1000_WRITE_REG(&adapter->hw, E1000_CTRL, ctrl); /* Keep the laser running on Fiber adapters */ if (adapter->hw.phy.media_type == e1000_media_type_fiber || adapter->hw.phy.media_type == e1000_media_type_internal_serdes) { ctrl_ext = E1000_READ_REG(&adapter->hw, E1000_CTRL_EXT); ctrl_ext |= E1000_CTRL_EXT_SDP3_DATA; E1000_WRITE_REG(&adapter->hw, E1000_CTRL_EXT, ctrl_ext); } if ((adapter->hw.mac.type == e1000_ich8lan) || (adapter->hw.mac.type == e1000_pchlan) || (adapter->hw.mac.type == e1000_ich9lan) || (adapter->hw.mac.type == e1000_ich10lan)) e1000_suspend_workarounds_ich8lan(&adapter->hw); if ( adapter->hw.mac.type >= e1000_pchlan) { error = em_enable_phy_wakeup(adapter); if (error) goto pme; } else { /* Enable wakeup by the MAC */ E1000_WRITE_REG(&adapter->hw, E1000_WUC, E1000_WUC_PME_EN); E1000_WRITE_REG(&adapter->hw, E1000_WUFC, adapter->wol); } if (adapter->hw.phy.type == e1000_phy_igp_3) e1000_igp3_phy_powerdown_workaround_ich8lan(&adapter->hw); pme: status = pci_read_config(dev, pmc + PCIR_POWER_STATUS, 2); status &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE); if (!error && (if_getcapenable(ifp) & IFCAP_WOL)) status |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE; pci_write_config(dev, pmc + PCIR_POWER_STATUS, status, 2); return; } /* * WOL in the newer chipset interfaces (pchlan) * require thing to be copied into the phy */ static int em_enable_phy_wakeup(struct adapter *adapter) { struct e1000_hw *hw = &adapter->hw; u32 mreg, ret = 0; u16 preg; /* copy MAC RARs to PHY RARs */ e1000_copy_rx_addrs_to_phy_ich8lan(hw); /* copy MAC MTA to PHY MTA */ for (int i = 0; i < adapter->hw.mac.mta_reg_count; i++) { mreg = E1000_READ_REG_ARRAY(hw, E1000_MTA, i); e1000_write_phy_reg(hw, BM_MTA(i), (u16)(mreg & 0xFFFF)); e1000_write_phy_reg(hw, BM_MTA(i) + 1, (u16)((mreg >> 16) & 0xFFFF)); } /* configure PHY Rx Control register */ e1000_read_phy_reg(&adapter->hw, BM_RCTL, &preg); mreg = E1000_READ_REG(hw, E1000_RCTL); if (mreg & E1000_RCTL_UPE) preg |= BM_RCTL_UPE; if (mreg & E1000_RCTL_MPE) preg |= BM_RCTL_MPE; preg &= ~(BM_RCTL_MO_MASK); if (mreg & E1000_RCTL_MO_3) preg |= (((mreg & E1000_RCTL_MO_3) >> E1000_RCTL_MO_SHIFT) << BM_RCTL_MO_SHIFT); if (mreg & E1000_RCTL_BAM) preg |= BM_RCTL_BAM; if (mreg & E1000_RCTL_PMCF) preg |= BM_RCTL_PMCF; mreg = E1000_READ_REG(hw, E1000_CTRL); if (mreg & E1000_CTRL_RFCE) preg |= BM_RCTL_RFCE; e1000_write_phy_reg(&adapter->hw, BM_RCTL, preg); /* enable PHY wakeup in MAC register */ E1000_WRITE_REG(hw, E1000_WUC, E1000_WUC_PHY_WAKE | E1000_WUC_PME_EN | E1000_WUC_APME); E1000_WRITE_REG(hw, E1000_WUFC, adapter->wol); /* configure and enable PHY wakeup in PHY registers */ e1000_write_phy_reg(&adapter->hw, BM_WUFC, adapter->wol); e1000_write_phy_reg(&adapter->hw, BM_WUC, E1000_WUC_PME_EN); /* activate PHY wakeup */ ret = hw->phy.ops.acquire(hw); if (ret) { printf("Could not acquire PHY\n"); return ret; } e1000_write_phy_reg_mdic(hw, IGP01E1000_PHY_PAGE_SELECT, (BM_WUC_ENABLE_PAGE << IGP_PAGE_SHIFT)); ret = e1000_read_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, &preg); if (ret) { printf("Could not read PHY page 769\n"); goto out; } preg |= BM_WUC_ENABLE_BIT | BM_WUC_HOST_WU_BIT; ret = e1000_write_phy_reg_mdic(hw, BM_WUC_ENABLE_REG, preg); if (ret) printf("Could not set PHY Host Wakeup bit\n"); out: hw->phy.ops.release(hw); return ret; } static void em_if_led_func(if_ctx_t ctx, int onoff) { struct adapter *adapter = iflib_get_softc(ctx); if (onoff) { e1000_setup_led(&adapter->hw); e1000_led_on(&adapter->hw); } else { e1000_led_off(&adapter->hw); e1000_cleanup_led(&adapter->hw); } } /* * Disable the L0S and L1 LINK states */ static void em_disable_aspm(struct adapter *adapter) { int base, reg; u16 link_cap,link_ctrl; device_t dev = adapter->dev; switch (adapter->hw.mac.type) { case e1000_82573: case e1000_82574: case e1000_82583: break; default: return; } if (pci_find_cap(dev, PCIY_EXPRESS, &base) != 0) return; reg = base + PCIER_LINK_CAP; link_cap = pci_read_config(dev, reg, 2); if ((link_cap & PCIEM_LINK_CAP_ASPM) == 0) return; reg = base + PCIER_LINK_CTL; link_ctrl = pci_read_config(dev, reg, 2); link_ctrl &= ~PCIEM_LINK_CTL_ASPMC; pci_write_config(dev, reg, link_ctrl, 2); return; } /********************************************************************** * * Update the board statistics counters. * **********************************************************************/ static void em_update_stats_counters(struct adapter *adapter) { if(adapter->hw.phy.media_type == e1000_media_type_copper || (E1000_READ_REG(&adapter->hw, E1000_STATUS) & E1000_STATUS_LU)) { adapter->stats.symerrs += E1000_READ_REG(&adapter->hw, E1000_SYMERRS); adapter->stats.sec += E1000_READ_REG(&adapter->hw, E1000_SEC); } adapter->stats.crcerrs += E1000_READ_REG(&adapter->hw, E1000_CRCERRS); adapter->stats.mpc += E1000_READ_REG(&adapter->hw, E1000_MPC); adapter->stats.scc += E1000_READ_REG(&adapter->hw, E1000_SCC); adapter->stats.ecol += E1000_READ_REG(&adapter->hw, E1000_ECOL); adapter->stats.mcc += E1000_READ_REG(&adapter->hw, E1000_MCC); adapter->stats.latecol += E1000_READ_REG(&adapter->hw, E1000_LATECOL); adapter->stats.colc += E1000_READ_REG(&adapter->hw, E1000_COLC); adapter->stats.dc += E1000_READ_REG(&adapter->hw, E1000_DC); adapter->stats.rlec += E1000_READ_REG(&adapter->hw, E1000_RLEC); adapter->stats.xonrxc += E1000_READ_REG(&adapter->hw, E1000_XONRXC); adapter->stats.xontxc += E1000_READ_REG(&adapter->hw, E1000_XONTXC); adapter->stats.xoffrxc += E1000_READ_REG(&adapter->hw, E1000_XOFFRXC); /* ** For watchdog management we need to know if we have been ** paused during the last interval, so capture that here. */ adapter->shared->isc_pause_frames = adapter->stats.xoffrxc; adapter->stats.xofftxc += E1000_READ_REG(&adapter->hw, E1000_XOFFTXC); adapter->stats.fcruc += E1000_READ_REG(&adapter->hw, E1000_FCRUC); adapter->stats.prc64 += E1000_READ_REG(&adapter->hw, E1000_PRC64); adapter->stats.prc127 += E1000_READ_REG(&adapter->hw, E1000_PRC127); adapter->stats.prc255 += E1000_READ_REG(&adapter->hw, E1000_PRC255); adapter->stats.prc511 += E1000_READ_REG(&adapter->hw, E1000_PRC511); adapter->stats.prc1023 += E1000_READ_REG(&adapter->hw, E1000_PRC1023); adapter->stats.prc1522 += E1000_READ_REG(&adapter->hw, E1000_PRC1522); adapter->stats.gprc += E1000_READ_REG(&adapter->hw, E1000_GPRC); adapter->stats.bprc += E1000_READ_REG(&adapter->hw, E1000_BPRC); adapter->stats.mprc += E1000_READ_REG(&adapter->hw, E1000_MPRC); adapter->stats.gptc += E1000_READ_REG(&adapter->hw, E1000_GPTC); /* For the 64-bit byte counters the low dword must be read first. */ /* Both registers clear on the read of the high dword */ adapter->stats.gorc += E1000_READ_REG(&adapter->hw, E1000_GORCL) + ((u64)E1000_READ_REG(&adapter->hw, E1000_GORCH) << 32); adapter->stats.gotc += E1000_READ_REG(&adapter->hw, E1000_GOTCL) + ((u64)E1000_READ_REG(&adapter->hw, E1000_GOTCH) << 32); adapter->stats.rnbc += E1000_READ_REG(&adapter->hw, E1000_RNBC); adapter->stats.ruc += E1000_READ_REG(&adapter->hw, E1000_RUC); adapter->stats.rfc += E1000_READ_REG(&adapter->hw, E1000_RFC); adapter->stats.roc += E1000_READ_REG(&adapter->hw, E1000_ROC); adapter->stats.rjc += E1000_READ_REG(&adapter->hw, E1000_RJC); adapter->stats.tor += E1000_READ_REG(&adapter->hw, E1000_TORH); adapter->stats.tot += E1000_READ_REG(&adapter->hw, E1000_TOTH); adapter->stats.tpr += E1000_READ_REG(&adapter->hw, E1000_TPR); adapter->stats.tpt += E1000_READ_REG(&adapter->hw, E1000_TPT); adapter->stats.ptc64 += E1000_READ_REG(&adapter->hw, E1000_PTC64); adapter->stats.ptc127 += E1000_READ_REG(&adapter->hw, E1000_PTC127); adapter->stats.ptc255 += E1000_READ_REG(&adapter->hw, E1000_PTC255); adapter->stats.ptc511 += E1000_READ_REG(&adapter->hw, E1000_PTC511); adapter->stats.ptc1023 += E1000_READ_REG(&adapter->hw, E1000_PTC1023); adapter->stats.ptc1522 += E1000_READ_REG(&adapter->hw, E1000_PTC1522); adapter->stats.mptc += E1000_READ_REG(&adapter->hw, E1000_MPTC); adapter->stats.bptc += E1000_READ_REG(&adapter->hw, E1000_BPTC); /* Interrupt Counts */ adapter->stats.iac += E1000_READ_REG(&adapter->hw, E1000_IAC); adapter->stats.icrxptc += E1000_READ_REG(&adapter->hw, E1000_ICRXPTC); adapter->stats.icrxatc += E1000_READ_REG(&adapter->hw, E1000_ICRXATC); adapter->stats.ictxptc += E1000_READ_REG(&adapter->hw, E1000_ICTXPTC); adapter->stats.ictxatc += E1000_READ_REG(&adapter->hw, E1000_ICTXATC); adapter->stats.ictxqec += E1000_READ_REG(&adapter->hw, E1000_ICTXQEC); adapter->stats.ictxqmtc += E1000_READ_REG(&adapter->hw, E1000_ICTXQMTC); adapter->stats.icrxdmtc += E1000_READ_REG(&adapter->hw, E1000_ICRXDMTC); adapter->stats.icrxoc += E1000_READ_REG(&adapter->hw, E1000_ICRXOC); if (adapter->hw.mac.type >= e1000_82543) { adapter->stats.algnerrc += E1000_READ_REG(&adapter->hw, E1000_ALGNERRC); adapter->stats.rxerrc += E1000_READ_REG(&adapter->hw, E1000_RXERRC); adapter->stats.tncrs += E1000_READ_REG(&adapter->hw, E1000_TNCRS); adapter->stats.cexterr += E1000_READ_REG(&adapter->hw, E1000_CEXTERR); adapter->stats.tsctc += E1000_READ_REG(&adapter->hw, E1000_TSCTC); adapter->stats.tsctfc += E1000_READ_REG(&adapter->hw, E1000_TSCTFC); } } static uint64_t em_if_get_counter(if_ctx_t ctx, ift_counter cnt) { struct adapter *adapter = iflib_get_softc(ctx); struct ifnet *ifp = iflib_get_ifp(ctx); switch (cnt) { case IFCOUNTER_COLLISIONS: return (adapter->stats.colc); case IFCOUNTER_IERRORS: return (adapter->dropped_pkts + adapter->stats.rxerrc + adapter->stats.crcerrs + adapter->stats.algnerrc + adapter->stats.ruc + adapter->stats.roc + adapter->stats.mpc + adapter->stats.cexterr); case IFCOUNTER_OERRORS: return (adapter->stats.ecol + adapter->stats.latecol + adapter->watchdog_events); default: return (if_get_counter_default(ifp, cnt)); } } /* Export a single 32-bit register via a read-only sysctl. */ static int em_sysctl_reg_handler(SYSCTL_HANDLER_ARGS) { struct adapter *adapter; u_int val; adapter = oidp->oid_arg1; val = E1000_READ_REG(&adapter->hw, oidp->oid_arg2); return (sysctl_handle_int(oidp, &val, 0, req)); } /* * Add sysctl variables, one per statistic, to the system. */ static void em_add_hw_stats(struct adapter *adapter) { device_t dev = iflib_get_dev(adapter->ctx); struct em_tx_queue *tx_que = adapter->tx_queues; struct em_rx_queue *rx_que = adapter->rx_queues; 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 e1000_hw_stats *stats = &adapter->stats; struct sysctl_oid *stat_node, *queue_node, *int_node; struct sysctl_oid_list *stat_list, *queue_list, *int_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, "link_irq", CTLFLAG_RD, &adapter->link_irq, "Link MSIX IRQ Handled"); SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "mbuf_defrag_fail", CTLFLAG_RD, &adapter->mbuf_defrag_failed, "Defragmenting mbuf chain failed"); SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "tx_dma_fail", CTLFLAG_RD, &adapter->no_tx_dma_setup, "Driver tx dma failure in xmit"); SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_overruns", CTLFLAG_RD, &adapter->rx_overruns, "RX overruns"); SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "watchdog_timeouts", CTLFLAG_RD, &adapter->watchdog_events, "Watchdog timeouts"); SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "device_control", CTLTYPE_UINT | CTLFLAG_RD, adapter, E1000_CTRL, em_sysctl_reg_handler, "IU", "Device Control Register"); SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "rx_control", CTLTYPE_UINT | CTLFLAG_RD, adapter, E1000_RCTL, em_sysctl_reg_handler, "IU", "Receiver Control Register"); SYSCTL_ADD_UINT(ctx, child, OID_AUTO, "fc_high_water", CTLFLAG_RD, &adapter->hw.fc.high_water, 0, "Flow Control High Watermark"); SYSCTL_ADD_UINT(ctx, child, OID_AUTO, "fc_low_water", CTLFLAG_RD, &adapter->hw.fc.low_water, 0, "Flow Control Low Watermark"); for (int i = 0; i < adapter->tx_num_queues; i++, tx_que++) { struct tx_ring *txr = &tx_que->txr; snprintf(namebuf, QUEUE_NAME_LEN, "queue_tx_%d", i); queue_node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf, CTLFLAG_RD, NULL, "TX Queue Name"); queue_list = SYSCTL_CHILDREN(queue_node); SYSCTL_ADD_PROC(ctx, queue_list, OID_AUTO, "txd_head", CTLTYPE_UINT | CTLFLAG_RD, adapter, E1000_TDH(txr->me), em_sysctl_reg_handler, "IU", "Transmit Descriptor Head"); SYSCTL_ADD_PROC(ctx, queue_list, OID_AUTO, "txd_tail", CTLTYPE_UINT | CTLFLAG_RD, adapter, E1000_TDT(txr->me), em_sysctl_reg_handler, "IU", "Transmit Descriptor Tail"); SYSCTL_ADD_ULONG(ctx, queue_list, OID_AUTO, "tx_irq", CTLFLAG_RD, &txr->tx_irq, "Queue MSI-X Transmit Interrupts"); } for (int j = 0; j < adapter->rx_num_queues; j++, rx_que++) { struct rx_ring *rxr = &rx_que->rxr; snprintf(namebuf, QUEUE_NAME_LEN, "queue_rx_%d", j); queue_node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf, CTLFLAG_RD, NULL, "RX Queue Name"); queue_list = SYSCTL_CHILDREN(queue_node); SYSCTL_ADD_PROC(ctx, queue_list, OID_AUTO, "rxd_head", CTLTYPE_UINT | CTLFLAG_RD, adapter, E1000_RDH(rxr->me), em_sysctl_reg_handler, "IU", "Receive Descriptor Head"); SYSCTL_ADD_PROC(ctx, queue_list, OID_AUTO, "rxd_tail", CTLTYPE_UINT | CTLFLAG_RD, adapter, E1000_RDT(rxr->me), em_sysctl_reg_handler, "IU", "Receive Descriptor Tail"); SYSCTL_ADD_ULONG(ctx, queue_list, OID_AUTO, "rx_irq", CTLFLAG_RD, &rxr->rx_irq, "Queue MSI-X Receive Interrupts"); } /* MAC stats get their own sub node */ stat_node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "mac_stats", CTLFLAG_RD, NULL, "Statistics"); stat_list = SYSCTL_CHILDREN(stat_node); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "excess_coll", CTLFLAG_RD, &stats->ecol, "Excessive collisions"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "single_coll", CTLFLAG_RD, &stats->scc, "Single collisions"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "multiple_coll", CTLFLAG_RD, &stats->mcc, "Multiple collisions"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "late_coll", CTLFLAG_RD, &stats->latecol, "Late collisions"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "collision_count", CTLFLAG_RD, &stats->colc, "Collision Count"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "symbol_errors", CTLFLAG_RD, &adapter->stats.symerrs, "Symbol Errors"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "sequence_errors", CTLFLAG_RD, &adapter->stats.sec, "Sequence Errors"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "defer_count", CTLFLAG_RD, &adapter->stats.dc, "Defer Count"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "missed_packets", CTLFLAG_RD, &adapter->stats.mpc, "Missed Packets"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "recv_no_buff", CTLFLAG_RD, &adapter->stats.rnbc, "Receive No Buffers"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "recv_undersize", CTLFLAG_RD, &adapter->stats.ruc, "Receive Undersize"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "recv_fragmented", CTLFLAG_RD, &adapter->stats.rfc, "Fragmented Packets Received "); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "recv_oversize", CTLFLAG_RD, &adapter->stats.roc, "Oversized Packets Received"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "recv_jabber", CTLFLAG_RD, &adapter->stats.rjc, "Recevied Jabber"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "recv_errs", CTLFLAG_RD, &adapter->stats.rxerrc, "Receive Errors"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "crc_errs", CTLFLAG_RD, &adapter->stats.crcerrs, "CRC errors"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "alignment_errs", CTLFLAG_RD, &adapter->stats.algnerrc, "Alignment Errors"); /* On 82575 these are collision counts */ SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "coll_ext_errs", CTLFLAG_RD, &adapter->stats.cexterr, "Collision/Carrier extension errors"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "xon_recvd", CTLFLAG_RD, &adapter->stats.xonrxc, "XON Received"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "xon_txd", CTLFLAG_RD, &adapter->stats.xontxc, "XON Transmitted"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "xoff_recvd", CTLFLAG_RD, &adapter->stats.xoffrxc, "XOFF Received"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "xoff_txd", CTLFLAG_RD, &adapter->stats.xofftxc, "XOFF Transmitted"); /* Packet Reception Stats */ SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "total_pkts_recvd", CTLFLAG_RD, &adapter->stats.tpr, "Total Packets Received "); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "good_pkts_recvd", CTLFLAG_RD, &adapter->stats.gprc, "Good Packets Received"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "bcast_pkts_recvd", CTLFLAG_RD, &adapter->stats.bprc, "Broadcast Packets Received"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "mcast_pkts_recvd", CTLFLAG_RD, &adapter->stats.mprc, "Multicast Packets Received"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "rx_frames_64", CTLFLAG_RD, &adapter->stats.prc64, "64 byte frames received "); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "rx_frames_65_127", CTLFLAG_RD, &adapter->stats.prc127, "65-127 byte frames received"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "rx_frames_128_255", CTLFLAG_RD, &adapter->stats.prc255, "128-255 byte frames received"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "rx_frames_256_511", CTLFLAG_RD, &adapter->stats.prc511, "256-511 byte frames received"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "rx_frames_512_1023", CTLFLAG_RD, &adapter->stats.prc1023, "512-1023 byte frames received"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "rx_frames_1024_1522", CTLFLAG_RD, &adapter->stats.prc1522, "1023-1522 byte frames received"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "good_octets_recvd", CTLFLAG_RD, &adapter->stats.gorc, "Good Octets Received"); /* Packet Transmission Stats */ SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "good_octets_txd", CTLFLAG_RD, &adapter->stats.gotc, "Good Octets Transmitted"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "total_pkts_txd", CTLFLAG_RD, &adapter->stats.tpt, "Total Packets Transmitted"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "good_pkts_txd", CTLFLAG_RD, &adapter->stats.gptc, "Good Packets Transmitted"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "bcast_pkts_txd", CTLFLAG_RD, &adapter->stats.bptc, "Broadcast Packets Transmitted"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "mcast_pkts_txd", CTLFLAG_RD, &adapter->stats.mptc, "Multicast Packets Transmitted"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "tx_frames_64", CTLFLAG_RD, &adapter->stats.ptc64, "64 byte frames transmitted "); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "tx_frames_65_127", CTLFLAG_RD, &adapter->stats.ptc127, "65-127 byte frames transmitted"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "tx_frames_128_255", CTLFLAG_RD, &adapter->stats.ptc255, "128-255 byte frames transmitted"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "tx_frames_256_511", CTLFLAG_RD, &adapter->stats.ptc511, "256-511 byte frames transmitted"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "tx_frames_512_1023", CTLFLAG_RD, &adapter->stats.ptc1023, "512-1023 byte frames transmitted"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "tx_frames_1024_1522", CTLFLAG_RD, &adapter->stats.ptc1522, "1024-1522 byte frames transmitted"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "tso_txd", CTLFLAG_RD, &adapter->stats.tsctc, "TSO Contexts Transmitted"); SYSCTL_ADD_UQUAD(ctx, stat_list, OID_AUTO, "tso_ctx_fail", CTLFLAG_RD, &adapter->stats.tsctfc, "TSO Contexts Failed"); /* Interrupt Stats */ int_node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "interrupts", CTLFLAG_RD, NULL, "Interrupt Statistics"); int_list = SYSCTL_CHILDREN(int_node); SYSCTL_ADD_UQUAD(ctx, int_list, OID_AUTO, "asserts", CTLFLAG_RD, &adapter->stats.iac, "Interrupt Assertion Count"); SYSCTL_ADD_UQUAD(ctx, int_list, OID_AUTO, "rx_pkt_timer", CTLFLAG_RD, &adapter->stats.icrxptc, "Interrupt Cause Rx Pkt Timer Expire Count"); SYSCTL_ADD_UQUAD(ctx, int_list, OID_AUTO, "rx_abs_timer", CTLFLAG_RD, &adapter->stats.icrxatc, "Interrupt Cause Rx Abs Timer Expire Count"); SYSCTL_ADD_UQUAD(ctx, int_list, OID_AUTO, "tx_pkt_timer", CTLFLAG_RD, &adapter->stats.ictxptc, "Interrupt Cause Tx Pkt Timer Expire Count"); SYSCTL_ADD_UQUAD(ctx, int_list, OID_AUTO, "tx_abs_timer", CTLFLAG_RD, &adapter->stats.ictxatc, "Interrupt Cause Tx Abs Timer Expire Count"); SYSCTL_ADD_UQUAD(ctx, int_list, OID_AUTO, "tx_queue_empty", CTLFLAG_RD, &adapter->stats.ictxqec, "Interrupt Cause Tx Queue Empty Count"); SYSCTL_ADD_UQUAD(ctx, int_list, OID_AUTO, "tx_queue_min_thresh", CTLFLAG_RD, &adapter->stats.ictxqmtc, "Interrupt Cause Tx Queue Min Thresh Count"); SYSCTL_ADD_UQUAD(ctx, int_list, OID_AUTO, "rx_desc_min_thresh", CTLFLAG_RD, &adapter->stats.icrxdmtc, "Interrupt Cause Rx Desc Min Thresh Count"); SYSCTL_ADD_UQUAD(ctx, int_list, OID_AUTO, "rx_overrun", CTLFLAG_RD, &adapter->stats.icrxoc, "Interrupt Cause Receiver Overrun Count"); } /********************************************************************** * * This routine provides a way to dump out the adapter eeprom, * often a useful debug/service tool. This only dumps the first * 32 words, stuff that matters is in that extent. * **********************************************************************/ static int em_sysctl_nvm_info(SYSCTL_HANDLER_ARGS) { struct adapter *adapter = (struct adapter *)arg1; int error; int result; result = -1; error = sysctl_handle_int(oidp, &result, 0, req); if (error || !req->newptr) return (error); /* * This value will cause a hex dump of the * first 32 16-bit words of the EEPROM to * the screen. */ if (result == 1) em_print_nvm_info(adapter); return (error); } static void em_print_nvm_info(struct adapter *adapter) { u16 eeprom_data; int i, j, row = 0; /* Its a bit crude, but it gets the job done */ printf("\nInterface EEPROM Dump:\n"); printf("Offset\n0x0000 "); for (i = 0, j = 0; i < 32; i++, j++) { if (j == 8) { /* Make the offset block */ j = 0; ++row; printf("\n0x00%x0 ",row); } e1000_read_nvm(&adapter->hw, i, 1, &eeprom_data); printf("%04x ", eeprom_data); } printf("\n"); } static int em_sysctl_int_delay(SYSCTL_HANDLER_ARGS) { struct em_int_delay_info *info; struct adapter *adapter; u32 regval; int error, usecs, ticks; info = (struct em_int_delay_info *) arg1; usecs = info->value; error = sysctl_handle_int(oidp, &usecs, 0, req); if (error != 0 || req->newptr == NULL) return (error); if (usecs < 0 || usecs > EM_TICKS_TO_USECS(65535)) return (EINVAL); info->value = usecs; ticks = EM_USECS_TO_TICKS(usecs); if (info->offset == E1000_ITR) /* units are 256ns here */ ticks *= 4; adapter = info->adapter; regval = E1000_READ_OFFSET(&adapter->hw, info->offset); regval = (regval & ~0xffff) | (ticks & 0xffff); /* Handle a few special cases. */ switch (info->offset) { case E1000_RDTR: break; case E1000_TIDV: if (ticks == 0) { adapter->txd_cmd &= ~E1000_TXD_CMD_IDE; /* Don't write 0 into the TIDV register. */ regval++; } else adapter->txd_cmd |= E1000_TXD_CMD_IDE; break; } E1000_WRITE_OFFSET(&adapter->hw, info->offset, regval); return (0); } static void em_add_int_delay_sysctl(struct adapter *adapter, const char *name, const char *description, struct em_int_delay_info *info, int offset, int value) { info->adapter = adapter; info->offset = offset; info->value = value; SYSCTL_ADD_PROC(device_get_sysctl_ctx(adapter->dev), SYSCTL_CHILDREN(device_get_sysctl_tree(adapter->dev)), OID_AUTO, name, CTLTYPE_INT|CTLFLAG_RW, info, 0, em_sysctl_int_delay, "I", description); } /* * Set flow control using sysctl: * Flow control values: * 0 - off * 1 - rx pause * 2 - tx pause * 3 - full */ static int em_set_flowcntl(SYSCTL_HANDLER_ARGS) { int error; static int input = 3; /* default is full */ struct adapter *adapter = (struct adapter *) arg1; error = sysctl_handle_int(oidp, &input, 0, req); if ((error) || (req->newptr == NULL)) return (error); if (input == adapter->fc) /* no change? */ return (error); switch (input) { case e1000_fc_rx_pause: case e1000_fc_tx_pause: case e1000_fc_full: case e1000_fc_none: adapter->hw.fc.requested_mode = input; adapter->fc = input; break; default: /* Do nothing */ return (error); } adapter->hw.fc.current_mode = adapter->hw.fc.requested_mode; e1000_force_mac_fc(&adapter->hw); return (error); } /* * Manage Energy Efficient Ethernet: * Control values: * 0/1 - enabled/disabled */ static int em_sysctl_eee(SYSCTL_HANDLER_ARGS) { struct adapter *adapter = (struct adapter *) arg1; int error, value; value = adapter->hw.dev_spec.ich8lan.eee_disable; error = sysctl_handle_int(oidp, &value, 0, req); if (error || req->newptr == NULL) return (error); adapter->hw.dev_spec.ich8lan.eee_disable = (value != 0); em_if_init(adapter->ctx); return (0); } static int em_sysctl_debug_info(SYSCTL_HANDLER_ARGS) { struct adapter *adapter; int error; int result; result = -1; error = sysctl_handle_int(oidp, &result, 0, req); if (error || !req->newptr) return (error); if (result == 1) { adapter = (struct adapter *) arg1; em_print_debug_info(adapter); } return (error); } static int em_get_rs(SYSCTL_HANDLER_ARGS) { struct adapter *adapter = (struct adapter *) arg1; int error; int result; result = 0; error = sysctl_handle_int(oidp, &result, 0, req); if (error || !req->newptr || result != 1) return (error); em_dump_rs(adapter); return (error); } static void em_if_debug(if_ctx_t ctx) { em_dump_rs(iflib_get_softc(ctx)); } /* * This routine is meant to be fluid, add whatever is * needed for debugging a problem. -jfv */ static void em_print_debug_info(struct adapter *adapter) { device_t dev = iflib_get_dev(adapter->ctx); struct ifnet *ifp = iflib_get_ifp(adapter->ctx); struct tx_ring *txr = &adapter->tx_queues->txr; struct rx_ring *rxr = &adapter->rx_queues->rxr; if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) printf("Interface is RUNNING "); else printf("Interface is NOT RUNNING\n"); if (if_getdrvflags(ifp) & IFF_DRV_OACTIVE) printf("and INACTIVE\n"); else printf("and ACTIVE\n"); for (int i = 0; i < adapter->tx_num_queues; i++, txr++) { device_printf(dev, "TX Queue %d ------\n", i); device_printf(dev, "hw tdh = %d, hw tdt = %d\n", E1000_READ_REG(&adapter->hw, E1000_TDH(i)), E1000_READ_REG(&adapter->hw, E1000_TDT(i))); } for (int j=0; j < adapter->rx_num_queues; j++, rxr++) { device_printf(dev, "RX Queue %d ------\n", j); device_printf(dev, "hw rdh = %d, hw rdt = %d\n", E1000_READ_REG(&adapter->hw, E1000_RDH(j)), E1000_READ_REG(&adapter->hw, E1000_RDT(j))); } } /* * 82574 only: * Write a new value to the EEPROM increasing the number of MSIX * vectors from 3 to 5, for proper multiqueue support. */ static void em_enable_vectors_82574(if_ctx_t ctx) { struct adapter *adapter = iflib_get_softc(ctx); struct e1000_hw *hw = &adapter->hw; device_t dev = iflib_get_dev(ctx); u16 edata; e1000_read_nvm(hw, EM_NVM_PCIE_CTRL, 1, &edata); printf("Current cap: %#06x\n", edata); if (((edata & EM_NVM_MSIX_N_MASK) >> EM_NVM_MSIX_N_SHIFT) != 4) { device_printf(dev, "Writing to eeprom: increasing " "reported MSIX vectors from 3 to 5...\n"); edata &= ~(EM_NVM_MSIX_N_MASK); edata |= 4 << EM_NVM_MSIX_N_SHIFT; e1000_write_nvm(hw, EM_NVM_PCIE_CTRL, 1, &edata); e1000_update_nvm_checksum(hw); device_printf(dev, "Writing to eeprom: done\n"); } } Index: head/sys/dev/e1000/if_em.h =================================================================== --- head/sys/dev/e1000/if_em.h (revision 326983) +++ head/sys/dev/e1000/if_em.h (revision 326984) @@ -1,567 +1,567 @@ /*- * SPDX-License-Identifier: BSD-2-Clause * - * Copyright (c) 2016 Matt Macy + * Copyright (c) 2016 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. 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. */ /*$FreeBSD$*/ #include "opt_ddb.h" #include "opt_inet.h" #include "opt_inet6.h" #ifdef HAVE_KERNEL_OPTION_HEADERS #include "opt_device_polling.h" #endif #include #include #ifdef DDB #include #include #endif #if __FreeBSD_version >= 800000 #include #endif #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 "e1000_api.h" #include "e1000_82571.h" #include "ifdi_if.h" #ifndef _EM_H_DEFINED_ #define _EM_H_DEFINED_ /* Tunables */ /* * EM_MAX_TXD: Maximum number of Transmit Descriptors * Valid Range: 80-256 for 82542 and 82543-based adapters * 80-4096 for others * Default Value: 1024 * This value is the number of transmit descriptors allocated by the driver. * Increasing this value allows the driver to queue more transmits. Each * descriptor is 16 bytes. * Since TDLEN should be multiple of 128bytes, the number of transmit * desscriptors should meet the following condition. * (num_tx_desc * sizeof(struct e1000_tx_desc)) % 128 == 0 */ #define EM_MIN_TXD 128 #define EM_MAX_TXD 4096 #define EM_DEFAULT_TXD 1024 #define EM_DEFAULT_MULTI_TXD 4096 #define IGB_MAX_TXD 4096 /* * EM_MAX_RXD - Maximum number of receive Descriptors * Valid Range: 80-256 for 82542 and 82543-based adapters * 80-4096 for others * Default Value: 1024 * This value is the number of receive descriptors allocated by the driver. * Increasing this value allows the driver to buffer more incoming packets. * Each descriptor is 16 bytes. A receive buffer is also allocated for each * descriptor. The maximum MTU size is 16110. * Since TDLEN should be multiple of 128bytes, the number of transmit * desscriptors should meet the following condition. * (num_tx_desc * sizeof(struct e1000_tx_desc)) % 128 == 0 */ #define EM_MIN_RXD 128 #define EM_MAX_RXD 4096 #define EM_DEFAULT_RXD 1024 #define EM_DEFAULT_MULTI_RXD 4096 #define IGB_MAX_RXD 4096 /* * EM_TIDV - Transmit Interrupt Delay Value * Valid Range: 0-65535 (0=off) * Default Value: 64 * This value delays the generation of transmit interrupts in units of * 1.024 microseconds. Transmit interrupt reduction can improve CPU * efficiency if properly tuned for specific network traffic. If the * system is reporting dropped transmits, this value may be set too high * causing the driver to run out of available transmit descriptors. */ #define EM_TIDV 64 /* * EM_TADV - Transmit Absolute Interrupt Delay Value * (Not valid for 82542/82543/82544) * Valid Range: 0-65535 (0=off) * Default Value: 64 * This value, in units of 1.024 microseconds, limits the delay in which a * transmit interrupt is generated. Useful only if EM_TIDV is non-zero, * this value ensures that an interrupt is generated after the initial * packet is sent on the wire within the set amount of time. Proper tuning, * along with EM_TIDV, may improve traffic throughput in specific * network conditions. */ #define EM_TADV 64 /* * EM_RDTR - Receive Interrupt Delay Timer (Packet Timer) * Valid Range: 0-65535 (0=off) * Default Value: 0 * This value delays the generation of receive interrupts in units of 1.024 * microseconds. Receive interrupt reduction can improve CPU efficiency if * properly tuned for specific network traffic. Increasing this value adds * extra latency to frame reception and can end up decreasing the throughput * of TCP traffic. If the system is reporting dropped receives, this value * may be set too high, causing the driver to run out of available receive * descriptors. * * CAUTION: When setting EM_RDTR to a value other than 0, adapters * may hang (stop transmitting) under certain network conditions. * If this occurs a WATCHDOG message is logged in the system * event log. In addition, the controller is automatically reset, * restoring the network connection. To eliminate the potential * for the hang ensure that EM_RDTR is set to 0. */ #define EM_RDTR 0 /* * Receive Interrupt Absolute Delay Timer (Not valid for 82542/82543/82544) * Valid Range: 0-65535 (0=off) * Default Value: 64 * This value, in units of 1.024 microseconds, limits the delay in which a * receive interrupt is generated. Useful only if EM_RDTR is non-zero, * this value ensures that an interrupt is generated after the initial * packet is received within the set amount of time. Proper tuning, * along with EM_RDTR, may improve traffic throughput in specific network * conditions. */ #define EM_RADV 64 /* * This parameter controls whether or not autonegotation is enabled. * 0 - Disable autonegotiation * 1 - Enable autonegotiation */ #define DO_AUTO_NEG 1 /* * This parameter control whether or not the driver will wait for * autonegotiation to complete. * 1 - Wait for autonegotiation to complete * 0 - Don't wait for autonegotiation to complete */ #define WAIT_FOR_AUTO_NEG_DEFAULT 0 /* Tunables -- End */ #define AUTONEG_ADV_DEFAULT (ADVERTISE_10_HALF | ADVERTISE_10_FULL | \ ADVERTISE_100_HALF | ADVERTISE_100_FULL | \ ADVERTISE_1000_FULL) #define AUTO_ALL_MODES 0 /* PHY master/slave setting */ #define EM_MASTER_SLAVE e1000_ms_hw_default /* * Micellaneous constants */ #define EM_VENDOR_ID 0x8086 #define EM_FLASH 0x0014 #define EM_JUMBO_PBA 0x00000028 #define EM_DEFAULT_PBA 0x00000030 #define EM_SMARTSPEED_DOWNSHIFT 3 #define EM_SMARTSPEED_MAX 15 #define EM_MAX_LOOP 10 #define MAX_NUM_MULTICAST_ADDRESSES 128 #define PCI_ANY_ID (~0U) #define ETHER_ALIGN 2 #define EM_FC_PAUSE_TIME 0x0680 #define EM_EEPROM_APME 0x400; #define EM_82544_APME 0x0004; /* Support AutoMediaDetect for Marvell M88 PHY in i354 */ #define IGB_MEDIA_RESET (1 << 0) /* Define the starting Interrupt rate per Queue */ #define IGB_INTS_PER_SEC 8000 #define IGB_DEFAULT_ITR ((1000000/IGB_INTS_PER_SEC) << 2) #define IGB_LINK_ITR 2000 #define I210_LINK_DELAY 1000 #define IGB_MAX_SCATTER 40 #define IGB_VFTA_SIZE 128 #define IGB_BR_SIZE 4096 /* ring buf size */ #define IGB_TSO_SIZE (65535 + sizeof(struct ether_vlan_header)) #define IGB_TSO_SEG_SIZE 4096 /* Max dma segment size */ #define IGB_TXPBSIZE 20408 #define IGB_HDR_BUF 128 #define IGB_PKTTYPE_MASK 0x0000FFF0 #define IGB_DMCTLX_DCFLUSH_DIS 0x80000000 /* Disable DMA Coalesce Flush */ /* * Driver state logic for the detection of a hung state * in hardware. Set TX_HUNG whenever a TX packet is used * (data is sent) and clear it when txeof() is invoked if * any descriptors from the ring are cleaned/reclaimed. * Increment internal counter if no descriptors are cleaned * and compare to TX_MAXTRIES. When counter > TX_MAXTRIES, * reset adapter. */ #define EM_TX_IDLE 0x00000000 #define EM_TX_BUSY 0x00000001 #define EM_TX_HUNG 0x80000000 #define EM_TX_MAXTRIES 10 #define PCICFG_DESC_RING_STATUS 0xe4 #define FLUSH_DESC_REQUIRED 0x100 #define IGB_RX_PTHRESH ((hw->mac.type == e1000_i354) ? 12 : \ ((hw->mac.type <= e1000_82576) ? 16 : 8)) #define IGB_RX_HTHRESH 8 #define IGB_RX_WTHRESH ((hw->mac.type == e1000_82576 && \ (adapter->intr_type == IFLIB_INTR_MSIX)) ? 1 : 4) #define IGB_TX_PTHRESH ((hw->mac.type == e1000_i354) ? 20 : 8) #define IGB_TX_HTHRESH 1 #define IGB_TX_WTHRESH ((hw->mac.type != e1000_82575 && \ (adapter->intr_type == IFLIB_INTR_MSIX) ? 1 : 16) /* * TDBA/RDBA should be aligned on 16 byte boundary. But TDLEN/RDLEN should be * multiple of 128 bytes. So we align TDBA/RDBA on 128 byte boundary. This will * also optimize cache line size effect. H/W supports up to cache line size 128. */ #define EM_DBA_ALIGN 128 /* * See Intel 82574 Driver Programming Interface Manual, Section 10.2.6.9 */ #define TARC_COMPENSATION_MODE (1 << 7) /* Compensation Mode */ #define TARC_SPEED_MODE_BIT (1 << 21) /* On PCI-E MACs only */ #define TARC_MQ_FIX (1 << 23) | \ (1 << 24) | \ (1 << 25) /* Handle errata in MQ mode */ #define TARC_ERRATA_BIT (1 << 26) /* Note from errata on 82574 */ /* PCI Config defines */ #define EM_BAR_TYPE(v) ((v) & EM_BAR_TYPE_MASK) #define EM_BAR_TYPE_MASK 0x00000001 #define EM_BAR_TYPE_MMEM 0x00000000 #define EM_BAR_TYPE_IO 0x00000001 #define EM_BAR_TYPE_FLASH 0x0014 #define EM_BAR_MEM_TYPE(v) ((v) & EM_BAR_MEM_TYPE_MASK) #define EM_BAR_MEM_TYPE_MASK 0x00000006 #define EM_BAR_MEM_TYPE_32BIT 0x00000000 #define EM_BAR_MEM_TYPE_64BIT 0x00000004 #define EM_MSIX_BAR 3 /* On 82575 */ /* More backward compatibility */ #if __FreeBSD_version < 900000 #define SYSCTL_ADD_UQUAD SYSCTL_ADD_QUAD #endif /* Defines for printing debug information */ #define DEBUG_INIT 0 #define DEBUG_IOCTL 0 #define DEBUG_HW 0 #define INIT_DEBUGOUT(S) if (DEBUG_INIT) printf(S "\n") #define INIT_DEBUGOUT1(S, A) if (DEBUG_INIT) printf(S "\n", A) #define INIT_DEBUGOUT2(S, A, B) if (DEBUG_INIT) printf(S "\n", A, B) #define IOCTL_DEBUGOUT(S) if (DEBUG_IOCTL) printf(S "\n") #define IOCTL_DEBUGOUT1(S, A) if (DEBUG_IOCTL) printf(S "\n", A) #define IOCTL_DEBUGOUT2(S, A, B) if (DEBUG_IOCTL) printf(S "\n", A, B) #define HW_DEBUGOUT(S) if (DEBUG_HW) printf(S "\n") #define HW_DEBUGOUT1(S, A) if (DEBUG_HW) printf(S "\n", A) #define HW_DEBUGOUT2(S, A, B) if (DEBUG_HW) printf(S "\n", A, B) #define EM_MAX_SCATTER 40 #define EM_VFTA_SIZE 128 #define EM_TSO_SIZE (65535 + sizeof(struct ether_vlan_header)) #define EM_TSO_SEG_SIZE 4096 /* Max dma segment size */ #define EM_MSIX_MASK 0x01F00000 /* For 82574 use */ #define EM_MSIX_LINK 0x01000000 /* For 82574 use */ #define ETH_ZLEN 60 #define ETH_ADDR_LEN 6 #define EM_CSUM_OFFLOAD 7 /* Offload bits in mbuf flag */ #define IGB_CSUM_OFFLOAD 0x0E0F /* Offload bits in mbuf flag */ #define IGB_PKTTYPE_MASK 0x0000FFF0 #define IGB_DMCTLX_DCFLUSH_DIS 0x80000000 /* Disable DMA Coalesce Flush */ /* * 82574 has a nonstandard address for EIAC * and since its only used in MSIX, and in * the em driver only 82574 uses MSIX we can * solve it just using this define. */ #define EM_EIAC 0x000DC /* * 82574 only reports 3 MSI-X vectors by default; * defines assisting with making it report 5 are * located here. */ #define EM_NVM_PCIE_CTRL 0x1B #define EM_NVM_MSIX_N_MASK (0x7 << EM_NVM_MSIX_N_SHIFT) #define EM_NVM_MSIX_N_SHIFT 7 struct adapter; struct em_int_delay_info { struct adapter *adapter; /* Back-pointer to the adapter struct */ int offset; /* Register offset to read/write */ int value; /* Current value in usecs */ }; /* * The transmit ring, one per tx queue */ struct tx_ring { struct adapter *adapter; struct e1000_tx_desc *tx_base; uint64_t tx_paddr; qidx_t *tx_rsq; bool tx_tso; /* last tx was tso */ uint8_t me; qidx_t tx_rs_cidx; qidx_t tx_rs_pidx; qidx_t tx_cidx_processed; /* Interrupt resources */ void *tag; struct resource *res; unsigned long tx_irq; /* Saved csum offloading context information */ int csum_flags; int csum_lhlen; int csum_iphlen; int csum_thlen; int csum_mss; int csum_pktlen; uint32_t csum_txd_upper; uint32_t csum_txd_lower; /* last field */ }; /* * The Receive ring, one per rx queue */ struct rx_ring { struct adapter *adapter; struct em_rx_queue *que; u32 me; u32 payload; union e1000_rx_desc_extended *rx_base; uint64_t rx_paddr; /* Interrupt resources */ void *tag; struct resource *res; bool discard; /* Soft stats */ unsigned long rx_irq; unsigned long rx_discarded; unsigned long rx_packets; unsigned long rx_bytes; }; struct em_tx_queue { struct adapter *adapter; u32 msix; u32 eims; /* This queue's EIMS bit */ u32 me; struct tx_ring txr; }; struct em_rx_queue { struct adapter *adapter; u32 me; u32 msix; u32 eims; struct rx_ring rxr; u64 irqs; struct if_irq que_irq; }; /* Our adapter structure */ struct adapter { struct ifnet *ifp; struct e1000_hw hw; if_softc_ctx_t shared; if_ctx_t ctx; #define tx_num_queues shared->isc_ntxqsets #define rx_num_queues shared->isc_nrxqsets #define intr_type shared->isc_intr /* FreeBSD operating-system-specific structures. */ struct e1000_osdep osdep; device_t dev; struct cdev *led_dev; struct em_tx_queue *tx_queues; struct em_rx_queue *rx_queues; struct if_irq irq; struct resource *memory; struct resource *flash; struct resource *ioport; int io_rid; struct resource *res; void *tag; u32 linkvec; u32 ivars; struct ifmedia *media; int msix; int if_flags; int em_insert_vlan_header; u32 ims; bool in_detach; u32 flags; /* Task for FAST handling */ struct grouptask link_task; u16 num_vlans; u32 txd_cmd; u32 tx_process_limit; u32 rx_process_limit; u32 rx_mbuf_sz; /* Management and WOL features */ u32 wol; bool has_manage; bool has_amt; /* Multicast array memory */ u8 *mta; /* ** Shadow VFTA table, this is needed because ** the real vlan filter table gets cleared during ** a soft reset and the driver needs to be able ** to repopulate it. */ u32 shadow_vfta[EM_VFTA_SIZE]; /* Info about the interface */ u16 link_active; u16 fc; u16 link_speed; u16 link_duplex; u32 smartspeed; u32 dmac; int link_mask; u64 que_mask; struct em_int_delay_info tx_int_delay; struct em_int_delay_info tx_abs_int_delay; struct em_int_delay_info rx_int_delay; struct em_int_delay_info rx_abs_int_delay; struct em_int_delay_info tx_itr; /* Misc stats maintained by the driver */ unsigned long dropped_pkts; unsigned long link_irq; unsigned long mbuf_defrag_failed; unsigned long no_tx_dma_setup; unsigned long no_tx_map_avail; unsigned long rx_overruns; unsigned long watchdog_events; struct e1000_hw_stats stats; u16 vf_ifp; }; /******************************************************************************** * vendor_info_array * * This array contains the list of Subvendor/Subdevice IDs on which the driver * should load. * ********************************************************************************/ typedef struct _em_vendor_info_t { unsigned int vendor_id; unsigned int device_id; unsigned int subvendor_id; unsigned int subdevice_id; unsigned int index; } em_vendor_info_t; void em_dump_rs(struct adapter *); #define EM_RSSRK_SIZE 4 #define EM_RSSRK_VAL(key, i) (key[(i) * EM_RSSRK_SIZE] | \ key[(i) * EM_RSSRK_SIZE + 1] << 8 | \ key[(i) * EM_RSSRK_SIZE + 2] << 16 | \ key[(i) * EM_RSSRK_SIZE + 3] << 24) #endif /* _EM_H_DEFINED_ */ Index: head/sys/dev/e1000/igb_txrx.c =================================================================== --- head/sys/dev/e1000/igb_txrx.c (revision 326983) +++ head/sys/dev/e1000/igb_txrx.c (revision 326984) @@ -1,584 +1,584 @@ /*- - * Copyright (c) 2016 Matt Macy + * Copyright (c) 2016 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. 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. */ /* $FreeBSD$ */ #include "if_em.h" #ifdef RSS #include #include #endif #ifdef VERBOSE_DEBUG #define DPRINTF device_printf #else #define DPRINTF(...) #endif /********************************************************************* * Local Function prototypes *********************************************************************/ static int igb_isc_txd_encap(void *arg, if_pkt_info_t pi); static void igb_isc_txd_flush(void *arg, uint16_t txqid, qidx_t pidx); static int igb_isc_txd_credits_update(void *arg, uint16_t txqid, bool clear); static void igb_isc_rxd_refill(void *arg, if_rxd_update_t iru); static void igb_isc_rxd_flush(void *arg, uint16_t rxqid, uint8_t flid __unused, qidx_t pidx); static int igb_isc_rxd_available(void *arg, uint16_t rxqid, qidx_t idx, qidx_t budget); static int igb_isc_rxd_pkt_get(void *arg, if_rxd_info_t ri); static int igb_tx_ctx_setup(struct tx_ring *txr, if_pkt_info_t pi, u32 *cmd_type_len, u32 *olinfo_status); static int igb_tso_setup(struct tx_ring *txr, if_pkt_info_t pi, u32 *cmd_type_len, u32 *olinfo_status); static void igb_rx_checksum(u32 staterr, if_rxd_info_t ri, u32 ptype); static int igb_determine_rsstype(u16 pkt_info); extern void igb_if_enable_intr(if_ctx_t ctx); extern int em_intr(void *arg); struct if_txrx igb_txrx = { igb_isc_txd_encap, igb_isc_txd_flush, igb_isc_txd_credits_update, igb_isc_rxd_available, igb_isc_rxd_pkt_get, igb_isc_rxd_refill, igb_isc_rxd_flush, em_intr }; extern if_shared_ctx_t em_sctx; /********************************************************************** * * Setup work for hardware segmentation offload (TSO) on * adapters using advanced tx descriptors * **********************************************************************/ static int igb_tso_setup(struct tx_ring *txr, if_pkt_info_t pi, u32 *cmd_type_len, u32 *olinfo_status) { struct e1000_adv_tx_context_desc *TXD; struct adapter *adapter = txr->adapter; u32 type_tucmd_mlhl = 0, vlan_macip_lens = 0; u32 mss_l4len_idx = 0; u32 paylen; switch(pi->ipi_etype) { case ETHERTYPE_IPV6: type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_IPV6; break; case ETHERTYPE_IP: type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_IPV4; /* Tell transmit desc to also do IPv4 checksum. */ *olinfo_status |= E1000_TXD_POPTS_IXSM << 8; break; default: panic("%s: CSUM_TSO but no supported IP version (0x%04x)", __func__, ntohs(pi->ipi_etype)); break; } TXD = (struct e1000_adv_tx_context_desc *) &txr->tx_base[pi->ipi_pidx]; /* This is used in the transmit desc in encap */ paylen = pi->ipi_len - pi->ipi_ehdrlen - pi->ipi_ip_hlen - pi->ipi_tcp_hlen; /* VLAN MACLEN IPLEN */ if (pi->ipi_mflags & M_VLANTAG) { vlan_macip_lens |= (pi->ipi_vtag << E1000_ADVTXD_VLAN_SHIFT); } vlan_macip_lens |= pi->ipi_ehdrlen << E1000_ADVTXD_MACLEN_SHIFT; vlan_macip_lens |= pi->ipi_ip_hlen; TXD->vlan_macip_lens = htole32(vlan_macip_lens); /* ADV DTYPE TUCMD */ type_tucmd_mlhl |= E1000_ADVTXD_DCMD_DEXT | E1000_ADVTXD_DTYP_CTXT; type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_L4T_TCP; TXD->type_tucmd_mlhl = htole32(type_tucmd_mlhl); /* MSS L4LEN IDX */ mss_l4len_idx |= (pi->ipi_tso_segsz << E1000_ADVTXD_MSS_SHIFT); mss_l4len_idx |= (pi->ipi_tcp_hlen << E1000_ADVTXD_L4LEN_SHIFT); /* 82575 needs the queue index added */ if (adapter->hw.mac.type == e1000_82575) mss_l4len_idx |= txr->me << 4; TXD->mss_l4len_idx = htole32(mss_l4len_idx); TXD->seqnum_seed = htole32(0); *cmd_type_len |= E1000_ADVTXD_DCMD_TSE; *olinfo_status |= E1000_TXD_POPTS_TXSM << 8; *olinfo_status |= paylen << E1000_ADVTXD_PAYLEN_SHIFT; return (1); } /********************************************************************* * * Advanced Context Descriptor setup for VLAN, CSUM or TSO * **********************************************************************/ static int igb_tx_ctx_setup(struct tx_ring *txr, if_pkt_info_t pi, u32 *cmd_type_len, u32 *olinfo_status) { struct e1000_adv_tx_context_desc *TXD; struct adapter *adapter = txr->adapter; u32 vlan_macip_lens, type_tucmd_mlhl; u32 mss_l4len_idx; mss_l4len_idx = vlan_macip_lens = type_tucmd_mlhl = 0; int offload = TRUE; /* First check if TSO is to be used */ if (pi->ipi_csum_flags & CSUM_TSO) return (igb_tso_setup(txr, pi, cmd_type_len, olinfo_status)); /* Indicate the whole packet as payload when not doing TSO */ *olinfo_status |= pi->ipi_len << E1000_ADVTXD_PAYLEN_SHIFT; /* Now ready a context descriptor */ TXD = (struct e1000_adv_tx_context_desc *) &txr->tx_base[pi->ipi_pidx]; /* ** In advanced descriptors the vlan tag must ** be placed into the context descriptor. Hence ** we need to make one even if not doing offloads. */ if (pi->ipi_mflags & M_VLANTAG) { vlan_macip_lens |= (pi->ipi_vtag << E1000_ADVTXD_VLAN_SHIFT); } else if ((pi->ipi_csum_flags & IGB_CSUM_OFFLOAD) == 0) { return (0); } /* Set the ether header length */ vlan_macip_lens |= pi->ipi_ehdrlen << E1000_ADVTXD_MACLEN_SHIFT; switch(pi->ipi_etype) { case ETHERTYPE_IP: type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_IPV4; break; case ETHERTYPE_IPV6: type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_IPV6; break; default: offload = FALSE; break; } vlan_macip_lens |= pi->ipi_ip_hlen; type_tucmd_mlhl |= E1000_ADVTXD_DCMD_DEXT | E1000_ADVTXD_DTYP_CTXT; switch (pi->ipi_ipproto) { case IPPROTO_TCP: if (pi->ipi_csum_flags & (CSUM_IP_TCP | CSUM_IP6_TCP)) type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_L4T_TCP; break; case IPPROTO_UDP: if (pi->ipi_csum_flags & (CSUM_IP_UDP | CSUM_IP6_UDP)) type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_L4T_UDP; break; case IPPROTO_SCTP: if (pi->ipi_csum_flags & (CSUM_IP_SCTP | CSUM_IP6_SCTP)) type_tucmd_mlhl |= E1000_ADVTXD_TUCMD_L4T_SCTP; break; default: offload = FALSE; break; } if (offload) /* For the TX descriptor setup */ *olinfo_status |= E1000_TXD_POPTS_TXSM << 8; /* 82575 needs the queue index added */ if (adapter->hw.mac.type == e1000_82575) mss_l4len_idx = txr->me << 4; /* Now copy bits into descriptor */ TXD->vlan_macip_lens = htole32(vlan_macip_lens); TXD->type_tucmd_mlhl = htole32(type_tucmd_mlhl); TXD->seqnum_seed = htole32(0); TXD->mss_l4len_idx = htole32(mss_l4len_idx); return (1); } static int igb_isc_txd_encap(void *arg, if_pkt_info_t pi) { struct adapter *sc = arg; if_softc_ctx_t scctx = sc->shared; struct em_tx_queue *que = &sc->tx_queues[pi->ipi_qsidx]; struct tx_ring *txr = &que->txr; int nsegs = pi->ipi_nsegs; bus_dma_segment_t *segs = pi->ipi_segs; union e1000_adv_tx_desc *txd = NULL; int i, j, first, pidx_last; u32 olinfo_status, cmd_type_len, txd_flags; qidx_t ntxd; pidx_last = olinfo_status = 0; /* Basic descriptor defines */ cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS | E1000_ADVTXD_DCMD_DEXT); if (pi->ipi_mflags & M_VLANTAG) cmd_type_len |= E1000_ADVTXD_DCMD_VLE; first = i = pi->ipi_pidx; ntxd = scctx->isc_ntxd[0]; txd_flags = pi->ipi_flags & IPI_TX_INTR ? E1000_ADVTXD_DCMD_RS : 0; /* Consume the first descriptor */ i += igb_tx_ctx_setup(txr, pi, &cmd_type_len, &olinfo_status); if (i == scctx->isc_ntxd[0]) i = 0; /* 82575 needs the queue index added */ if (sc->hw.mac.type == e1000_82575) olinfo_status |= txr->me << 4; for (j = 0; j < nsegs; j++) { bus_size_t seglen; bus_addr_t segaddr; txd = (union e1000_adv_tx_desc *)&txr->tx_base[i]; seglen = segs[j].ds_len; segaddr = htole64(segs[j].ds_addr); txd->read.buffer_addr = segaddr; txd->read.cmd_type_len = htole32(E1000_TXD_CMD_IFCS | cmd_type_len | seglen); txd->read.olinfo_status = htole32(olinfo_status); pidx_last = i; if (++i == scctx->isc_ntxd[0]) { i = 0; } } if (txd_flags) { txr->tx_rsq[txr->tx_rs_pidx] = pidx_last; txr->tx_rs_pidx = (txr->tx_rs_pidx+1) & (ntxd-1); MPASS(txr->tx_rs_pidx != txr->tx_rs_cidx); } txd->read.cmd_type_len |= htole32(E1000_TXD_CMD_EOP | txd_flags); pi->ipi_new_pidx = i; return (0); } static void igb_isc_txd_flush(void *arg, uint16_t txqid, qidx_t pidx) { struct adapter *adapter = arg; struct em_tx_queue *que = &adapter->tx_queues[txqid]; struct tx_ring *txr = &que->txr; E1000_WRITE_REG(&adapter->hw, E1000_TDT(txr->me), pidx); } static int igb_isc_txd_credits_update(void *arg, uint16_t txqid, bool clear) { struct adapter *adapter = arg; if_softc_ctx_t scctx = adapter->shared; struct em_tx_queue *que = &adapter->tx_queues[txqid]; struct tx_ring *txr = &que->txr; qidx_t processed = 0; int updated; qidx_t cur, prev, ntxd, rs_cidx; int32_t delta; uint8_t status; rs_cidx = txr->tx_rs_cidx; if (rs_cidx == txr->tx_rs_pidx) return (0); cur = txr->tx_rsq[rs_cidx]; status = ((union e1000_adv_tx_desc *)&txr->tx_base[cur])->wb.status; updated = !!(status & E1000_TXD_STAT_DD); if (!clear || !updated) return (updated); prev = txr->tx_cidx_processed; ntxd = scctx->isc_ntxd[0]; do { delta = (int32_t)cur - (int32_t)prev; MPASS(prev == 0 || delta != 0); if (delta < 0) delta += ntxd; processed += delta; prev = cur; rs_cidx = (rs_cidx + 1) & (ntxd-1); if (rs_cidx == txr->tx_rs_pidx) break; cur = txr->tx_rsq[rs_cidx]; status = ((union e1000_adv_tx_desc *)&txr->tx_base[cur])->wb.status; } while ((status & E1000_TXD_STAT_DD)); txr->tx_rs_cidx = rs_cidx; txr->tx_cidx_processed = prev; return (processed); } static void igb_isc_rxd_refill(void *arg, if_rxd_update_t iru) { struct adapter *sc = arg; if_softc_ctx_t scctx = sc->shared; uint16_t rxqid = iru->iru_qsidx; struct em_rx_queue *que = &sc->rx_queues[rxqid]; union e1000_adv_rx_desc *rxd; struct rx_ring *rxr = &que->rxr; uint64_t *paddrs; uint32_t next_pidx, pidx; uint16_t count; int i; paddrs = iru->iru_paddrs; pidx = iru->iru_pidx; count = iru->iru_count; for (i = 0, next_pidx = pidx; i < count; i++) { rxd = (union e1000_adv_rx_desc *)&rxr->rx_base[next_pidx]; rxd->read.pkt_addr = htole64(paddrs[i]); if (++next_pidx == scctx->isc_nrxd[0]) next_pidx = 0; } } static void igb_isc_rxd_flush(void *arg, uint16_t rxqid, uint8_t flid __unused, qidx_t pidx) { struct adapter *sc = arg; struct em_rx_queue *que = &sc->rx_queues[rxqid]; struct rx_ring *rxr = &que->rxr; E1000_WRITE_REG(&sc->hw, E1000_RDT(rxr->me), pidx); } static int igb_isc_rxd_available(void *arg, uint16_t rxqid, qidx_t idx, qidx_t budget) { struct adapter *sc = arg; if_softc_ctx_t scctx = sc->shared; struct em_rx_queue *que = &sc->rx_queues[rxqid]; struct rx_ring *rxr = &que->rxr; union e1000_adv_rx_desc *rxd; u32 staterr = 0; int cnt, i, iter; if (budget == 1) { rxd = (union e1000_adv_rx_desc *)&rxr->rx_base[idx]; staterr = le32toh(rxd->wb.upper.status_error); return (staterr & E1000_RXD_STAT_DD); } for (iter = cnt = 0, i = idx; iter < scctx->isc_nrxd[0] && iter <= budget;) { rxd = (union e1000_adv_rx_desc *)&rxr->rx_base[i]; staterr = le32toh(rxd->wb.upper.status_error); if ((staterr & E1000_RXD_STAT_DD) == 0) break; if (++i == scctx->isc_nrxd[0]) { i = 0; } if (staterr & E1000_RXD_STAT_EOP) cnt++; iter++; } return (cnt); } /**************************************************************** * Routine sends data which has been dma'ed into host memory * to upper layer. Initialize ri structure. * * Returns 0 upon success, errno on failure ***************************************************************/ static int igb_isc_rxd_pkt_get(void *arg, if_rxd_info_t ri) { struct adapter *adapter = arg; if_softc_ctx_t scctx = adapter->shared; struct em_rx_queue *que = &adapter->rx_queues[ri->iri_qsidx]; struct rx_ring *rxr = &que->rxr; struct ifnet *ifp = iflib_get_ifp(adapter->ctx); union e1000_adv_rx_desc *rxd; u16 pkt_info, len; u16 vtag = 0; u32 ptype; u32 staterr = 0; bool eop; int i = 0; int cidx = ri->iri_cidx; do { rxd = (union e1000_adv_rx_desc *)&rxr->rx_base[cidx]; staterr = le32toh(rxd->wb.upper.status_error); pkt_info = le16toh(rxd->wb.lower.lo_dword.hs_rss.pkt_info); MPASS ((staterr & E1000_RXD_STAT_DD) != 0); len = le16toh(rxd->wb.upper.length); ptype = le32toh(rxd->wb.lower.lo_dword.data) & IGB_PKTTYPE_MASK; ri->iri_len += len; rxr->rx_bytes += ri->iri_len; rxd->wb.upper.status_error = 0; eop = ((staterr & E1000_RXD_STAT_EOP) == E1000_RXD_STAT_EOP); if (((adapter->hw.mac.type == e1000_i350) || (adapter->hw.mac.type == e1000_i354)) && (staterr & E1000_RXDEXT_STATERR_LB)) vtag = be16toh(rxd->wb.upper.vlan); else vtag = le16toh(rxd->wb.upper.vlan); /* Make sure bad packets are discarded */ if (eop && ((staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) != 0)) { adapter->dropped_pkts++; ++rxr->rx_discarded; return (EBADMSG); } ri->iri_frags[i].irf_flid = 0; ri->iri_frags[i].irf_idx = cidx; ri->iri_frags[i].irf_len = len; if (++cidx == scctx->isc_nrxd[0]) cidx = 0; #ifdef notyet if (rxr->hdr_split == TRUE) { ri->iri_frags[i].irf_flid = 1; ri->iri_frags[i].irf_idx = cidx; if (++cidx == scctx->isc_nrxd[0]) cidx = 0; } #endif i++; } while (!eop); rxr->rx_packets++; if ((ifp->if_capenable & IFCAP_RXCSUM) != 0) igb_rx_checksum(staterr, ri, ptype); if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0 && (staterr & E1000_RXD_STAT_VP) != 0) { ri->iri_vtag = vtag; ri->iri_flags |= M_VLANTAG; } ri->iri_flowid = le32toh(rxd->wb.lower.hi_dword.rss); ri->iri_rsstype = igb_determine_rsstype(pkt_info); ri->iri_nfrags = i; return (0); } /********************************************************************* * * Verify that the hardware indicated that the checksum is valid. * Inform the stack about the status of checksum so that stack * doesn't spend time verifying the checksum. * *********************************************************************/ static void igb_rx_checksum(u32 staterr, if_rxd_info_t ri, u32 ptype) { u16 status = (u16)staterr; u8 errors = (u8) (staterr >> 24); bool sctp = FALSE; /* Ignore Checksum bit is set */ if (status & E1000_RXD_STAT_IXSM) { ri->iri_csum_flags = 0; return; } if ((ptype & E1000_RXDADV_PKTTYPE_ETQF) == 0 && (ptype & E1000_RXDADV_PKTTYPE_SCTP) != 0) sctp = 1; else sctp = 0; if (status & E1000_RXD_STAT_IPCS) { /* Did it pass? */ if (!(errors & E1000_RXD_ERR_IPE)) { /* IP Checksum Good */ ri->iri_csum_flags = CSUM_IP_CHECKED; ri->iri_csum_flags |= CSUM_IP_VALID; } else ri->iri_csum_flags = 0; } if (status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)) { u64 type = (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); if (sctp) /* reassign */ type = CSUM_SCTP_VALID; /* Did it pass? */ if (!(errors & E1000_RXD_ERR_TCPE)) { ri->iri_csum_flags |= type; if (sctp == 0) ri->iri_csum_data = htons(0xffff); } } return; } /******************************************************************** * * Parse the packet type to determine the appropriate hash * ******************************************************************/ static int igb_determine_rsstype(u16 pkt_info) { switch (pkt_info & E1000_RXDADV_RSSTYPE_MASK) { case E1000_RXDADV_RSSTYPE_IPV4_TCP: return M_HASHTYPE_RSS_TCP_IPV4; case E1000_RXDADV_RSSTYPE_IPV4: return M_HASHTYPE_RSS_IPV4; case E1000_RXDADV_RSSTYPE_IPV6_TCP: return M_HASHTYPE_RSS_TCP_IPV6; case E1000_RXDADV_RSSTYPE_IPV6_EX: return M_HASHTYPE_RSS_IPV6_EX; case E1000_RXDADV_RSSTYPE_IPV6: return M_HASHTYPE_RSS_IPV6; case E1000_RXDADV_RSSTYPE_IPV6_TCP_EX: return M_HASHTYPE_RSS_TCP_IPV6_EX; default: return M_HASHTYPE_OPAQUE; } } Index: head/sys/net/ifdi_if.m =================================================================== --- head/sys/net/ifdi_if.m (revision 326983) +++ head/sys/net/ifdi_if.m (revision 326984) @@ -1,343 +1,343 @@ # -# Copyright (c) 2014, Matthew Macy (kmacy@freebsd.org) +# Copyright (c) 2014, Matthew Macy (mmacy@mattmacy.io) # 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. # # $FreeBSD$ # #include #include #include #include #include #include #include #include #include #include INTERFACE ifdi; CODE { static void null_void_op(if_ctx_t _ctx __unused) { } static void null_timer_op(if_ctx_t _ctx __unused, uint16_t _qsidx __unused) { } static int null_int_op(if_ctx_t _ctx __unused) { return (0); } static int null_queue_intr_enable(if_ctx_t _ctx __unused, uint16_t _qid __unused) { return (ENOTSUP); } static void null_led_func(if_ctx_t _ctx __unused, int _onoff __unused) { } static void null_vlan_register_op(if_ctx_t _ctx __unused, uint16_t vtag __unused) { } static int null_q_setup(if_ctx_t _ctx __unused, uint32_t _qid __unused) { return (0); } static int null_i2c_req(if_ctx_t _sctx __unused, struct ifi2creq *_i2c __unused) { return (ENOTSUP); } static int null_sysctl_int_delay(if_ctx_t _sctx __unused, if_int_delay_info_t _iidi __unused) { return (0); } static int null_iov_init(if_ctx_t _ctx __unused, uint16_t num_vfs __unused, const nvlist_t *params __unused) { return (ENOTSUP); } static int null_vf_add(if_ctx_t _ctx __unused, uint16_t num_vfs __unused, const nvlist_t *params __unused) { return (ENOTSUP); } static int null_priv_ioctl(if_ctx_t _ctx __unused, u_long command, caddr_t *data __unused) { return (ENOTSUP); } }; # # bus interfaces # METHOD int attach_pre { if_ctx_t _ctx; }; METHOD int attach_post { if_ctx_t _ctx; }; METHOD int detach { if_ctx_t _ctx; }; METHOD int suspend { if_ctx_t _ctx; } DEFAULT null_int_op; METHOD int shutdown { if_ctx_t _ctx; } DEFAULT null_int_op; METHOD int resume { if_ctx_t _ctx; } DEFAULT null_int_op; # # downcall to driver to allocate its # own queue state and tie it to the parent # METHOD int tx_queues_alloc { if_ctx_t _ctx; caddr_t *_vaddrs; uint64_t *_paddrs; int ntxqs; int ntxqsets; }; METHOD int rx_queues_alloc { if_ctx_t _ctx; caddr_t *_vaddrs; uint64_t *_paddrs; int nrxqs; int nrxqsets; }; METHOD void queues_free { if_ctx_t _ctx; }; # # interface reset / stop # METHOD void init { if_ctx_t _ctx; }; METHOD void stop { if_ctx_t _ctx; }; # # interrupt setup and manipulation # METHOD int msix_intr_assign { if_ctx_t _sctx; int msix; }; METHOD void intr_enable { if_ctx_t _ctx; }; METHOD void intr_disable { if_ctx_t _ctx; }; METHOD int rx_queue_intr_enable { if_ctx_t _ctx; uint16_t _qid; } DEFAULT null_queue_intr_enable; METHOD int tx_queue_intr_enable { if_ctx_t _ctx; uint16_t _qid; } DEFAULT null_queue_intr_enable; METHOD void link_intr_enable { if_ctx_t _ctx; } DEFAULT null_void_op; # # interface configuration # METHOD void multi_set { if_ctx_t _ctx; }; METHOD int mtu_set { if_ctx_t _ctx; uint32_t _mtu; }; METHOD void media_set{ if_ctx_t _ctx; } DEFAULT null_void_op; METHOD int promisc_set { if_ctx_t _ctx; int _flags; }; METHOD void crcstrip_set { if_ctx_t _ctx; int _onoff; int _strip; }; # # IOV handling # METHOD void vflr_handle { if_ctx_t _ctx; } DEFAULT null_void_op; METHOD int iov_init { if_ctx_t _ctx; uint16_t num_vfs; const nvlist_t * params; } DEFAULT null_iov_init; METHOD void iov_uninit { if_ctx_t _ctx; } DEFAULT null_void_op; METHOD int iov_vf_add { if_ctx_t _ctx; uint16_t num_vfs; const nvlist_t * params; } DEFAULT null_vf_add; # # Device status # METHOD void update_admin_status { if_ctx_t _ctx; }; METHOD void media_status { if_ctx_t _ctx; struct ifmediareq *_ifm; }; METHOD int media_change { if_ctx_t _ctx; }; METHOD uint64_t get_counter { if_ctx_t _ctx; ift_counter cnt; }; METHOD int priv_ioctl { if_ctx_t _ctx; u_long _cmd; caddr_t _data; } DEFAULT null_priv_ioctl; # # optional methods # METHOD int i2c_req { if_ctx_t _ctx; struct ifi2creq *_req; } DEFAULT null_i2c_req; METHOD int txq_setup { if_ctx_t _ctx; uint32_t _txqid; } DEFAULT null_q_setup; METHOD int rxq_setup { if_ctx_t _ctx; uint32_t _txqid; } DEFAULT null_q_setup; METHOD void timer { if_ctx_t _ctx; uint16_t _txqid; } DEFAULT null_timer_op; METHOD void watchdog_reset { if_ctx_t _ctx; } DEFAULT null_void_op; METHOD void led_func { if_ctx_t _ctx; int _onoff; } DEFAULT null_led_func; METHOD void vlan_register { if_ctx_t _ctx; uint16_t _vtag; } DEFAULT null_vlan_register_op; METHOD void vlan_unregister { if_ctx_t _ctx; uint16_t _vtag; } DEFAULT null_vlan_register_op; METHOD int sysctl_int_delay { if_ctx_t _sctx; if_int_delay_info_t _iidi; } DEFAULT null_sysctl_int_delay; METHOD void debug { if_ctx_t _ctx; } DEFAULT null_void_op; Index: head/sys/net/iflib.c =================================================================== --- head/sys/net/iflib.c (revision 326983) +++ head/sys/net/iflib.c (revision 326984) @@ -1,5842 +1,5842 @@ /*- - * Copyright (c) 2014-2017, Matthew Macy + * 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_pad_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_pad_mbuf_fail, CTLFLAG_RD, &iflib_encap_pad_mbuf_fail, 0, "# runt frame pad 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_pad_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 - 1); /* ** 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) static void iflib_get_ip_forwarding(struct lro_ctrl *lc, bool *v4, bool *v6) { CURVNET_SET(lc->ifp->if_vnet); #if defined(INET6) *v6 = VNET(ip6_forwarding); #endif #if defined(INET) *v4 = VNET(ipforwarding); #endif CURVNET_RESTORE(); } /* * 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 mbuf *m, bool v4_forwarding, bool v6_forwarding) { 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: return !v6_forwarding; #endif #if defined (INET) case ETHERTYPE_IP: return !v4_forwarding; #endif } return false; } #else static void iflib_get_ip_forwarding(struct lro_ctrl *lc __unused, bool *v4 __unused, bool *v6 __unused) { } #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; bool v4_forwarding, v6_forwarding; /* * 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); if (lro_enabled) iflib_get_ip_forwarding(&rxq->ifr_lc, &v4_forwarding, &v6_forwarding); mt = mf = NULL; while (mh != NULL) { m = mh; 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) { if (!lro_possible) { lro_possible = iflib_check_lro_possible(m, v4_forwarding, v6_forwarding); 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); } /* * Pad an mbuf to ensure a minimum ethernet frame size. * min_frame_size is the frame size (less CRC) to pad the mbuf to */ static __noinline int iflib_ether_pad(device_t dev, struct mbuf **m_head, uint16_t min_frame_size) { /* * 18 is enough bytes to pad an ARP packet to 46 bytes, and * and ARP message is the smallest common payload I can think of */ static char pad[18]; /* just zeros */ int n; struct mbuf *new_head; if (!M_WRITABLE(*m_head)) { new_head = m_dup(*m_head, M_NOWAIT); if (new_head == NULL) { m_freem(*m_head); device_printf(dev, "cannot pad short frame, m_dup() failed"); DBG_COUNTER_INC(encap_pad_mbuf_fail); return ENOMEM; } m_freem(*m_head); *m_head = new_head; } for (n = min_frame_size - (*m_head)->m_pkthdr.len; n > 0; n -= sizeof(pad)) if (!m_append(*m_head, min(n, sizeof(pad)), pad)) break; if (n > 0) { m_freem(*m_head); device_printf(dev, "cannot pad short frame\n"); DBG_COUNTER_INC(encap_pad_mbuf_fail); return (ENOBUFS); } return 0; } 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; } if ((sctx->isc_flags & IFLIB_NEED_ETHER_PAD) && __predict_false(m_head->m_pkthdr.len < scctx->isc_min_frame_size)) { err = iflib_ether_pad(ctx->ifc_dev, m_headp, scctx->isc_min_frame_size); if (err) return err; } 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: case SIOCGIFXMEDIA: 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"); /* Set up cpu set. If it fails, use the set of all CPUs. */ if (bus_get_cpus(dev, INTR_CPUS, sizeof(ctx->ifc_cpus), &ctx->ifc_cpus) != 0) { device_printf(dev, "Unable to fetch CPU list\n"); CPU_COPY(&all_cpus, &ctx->ifc_cpus); } MPASS(CPU_COUNT(&ctx->ifc_cpus) > 0); /* ** 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(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 = ctx->ifc_sysctl_ntxqs; iflib_num_rx_queues = ctx->ifc_sysctl_nrxqs; device_printf(dev, "msix_init qsets capped at %d\n", imax(scctx->isc_ntxqsets, scctx->isc_nrxqsets)); 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 #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); #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; if (rx_queues > scctx->isc_nrxqsets) rx_queues = scctx->isc_nrxqsets; /* * 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 (tx_queues > scctx->isc_ntxqsets) tx_queues = scctx->isc_ntxqsets; 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: head/sys/net/iflib.h =================================================================== --- head/sys/net/iflib.h (revision 326983) +++ head/sys/net/iflib.h (revision 326984) @@ -1,407 +1,407 @@ /*- - * Copyright (c) 2014-2017, Matthew Macy (mmacy@nextbsd.org) + * Copyright (c) 2014-2017, Matthew Macy (mmacy@mattmacy.io) * 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. * * $FreeBSD$ */ #ifndef __IFLIB_H_ #define __IFLIB_H_ #include #include #include #include #include #include /* * The value type for indexing, limits max descriptors * to 65535 can be conditionally redefined to uint32_t * in the future if the need arises. */ typedef uint16_t qidx_t; #define QIDX_INVALID 0xFFFF /* * Most cards can handle much larger TSO requests * but the FreeBSD TCP stack will break on larger * values */ #define FREEBSD_TSO_SIZE_MAX 65518 struct iflib_ctx; typedef struct iflib_ctx *if_ctx_t; struct if_shared_ctx; typedef struct if_shared_ctx *if_shared_ctx_t; struct if_int_delay_info; typedef struct if_int_delay_info *if_int_delay_info_t; /* * File organization: * - public structures * - iflib accessors * - iflib utility functions * - iflib core functions */ typedef struct if_rxd_frag { uint8_t irf_flid; qidx_t irf_idx; uint16_t irf_len; } *if_rxd_frag_t; typedef struct if_rxd_info { /* set by iflib */ uint16_t iri_qsidx; /* qset index */ uint16_t iri_vtag; /* vlan tag - if flag set */ /* XXX redundant with the new irf_len field */ uint16_t iri_len; /* packet length */ qidx_t iri_cidx; /* consumer index of cq */ struct ifnet *iri_ifp; /* some drivers >1 interface per softc */ /* updated by driver */ if_rxd_frag_t iri_frags; uint32_t iri_flowid; /* RSS hash for packet */ uint32_t iri_csum_flags; /* m_pkthdr csum flags */ uint32_t iri_csum_data; /* m_pkthdr csum data */ uint8_t iri_flags; /* mbuf flags for packet */ uint8_t iri_nfrags; /* number of fragments in packet */ uint8_t iri_rsstype; /* RSS hash type */ uint8_t iri_pad; /* any padding in the received data */ } *if_rxd_info_t; typedef struct if_rxd_update { uint64_t *iru_paddrs; caddr_t *iru_vaddrs; qidx_t *iru_idxs; qidx_t iru_pidx; uint16_t iru_qsidx; uint16_t iru_count; uint16_t iru_buf_size; uint8_t iru_flidx; } *if_rxd_update_t; #define IPI_TX_INTR 0x1 /* send an interrupt when this packet is sent */ #define IPI_TX_IPV4 0x2 /* ethertype IPv4 */ #define IPI_TX_IPV6 0x4 /* ethertype IPv6 */ typedef struct if_pkt_info { bus_dma_segment_t *ipi_segs; /* physical addresses */ uint32_t ipi_len; /* packet length */ uint16_t ipi_qsidx; /* queue set index */ qidx_t ipi_nsegs; /* number of segments */ qidx_t ipi_ndescs; /* number of descriptors used by encap */ uint16_t ipi_flags; /* iflib per-packet flags */ qidx_t ipi_pidx; /* start pidx for encap */ qidx_t ipi_new_pidx; /* next available pidx post-encap */ /* offload handling */ uint8_t ipi_ehdrlen; /* ether header length */ uint8_t ipi_ip_hlen; /* ip header length */ uint8_t ipi_tcp_hlen; /* tcp header length */ uint8_t ipi_ipproto; /* ip protocol */ uint32_t ipi_csum_flags; /* packet checksum flags */ uint16_t ipi_tso_segsz; /* tso segment size */ uint16_t ipi_vtag; /* VLAN tag */ uint16_t ipi_etype; /* ether header type */ uint8_t ipi_tcp_hflags; /* tcp header flags */ uint8_t ipi_mflags; /* packet mbuf flags */ uint32_t ipi_tcp_seq; /* tcp seqno */ uint32_t ipi_tcp_sum; /* tcp csum */ } *if_pkt_info_t; typedef struct if_irq { struct resource *ii_res; int ii_rid; void *ii_tag; } *if_irq_t; struct if_int_delay_info { if_ctx_t iidi_ctx; /* Back-pointer to the iflib ctx (softc) */ int iidi_offset; /* Register offset to read/write */ int iidi_value; /* Current value in usecs */ struct sysctl_oid *iidi_oidp; struct sysctl_req *iidi_req; }; typedef enum { IFLIB_INTR_LEGACY, IFLIB_INTR_MSI, IFLIB_INTR_MSIX } iflib_intr_mode_t; /* * This really belongs in pciio.h or some place more general * but this is the only consumer for now. */ typedef struct pci_vendor_info { uint32_t pvi_vendor_id; uint32_t pvi_device_id; uint32_t pvi_subvendor_id; uint32_t pvi_subdevice_id; uint32_t pvi_rev_id; uint32_t pvi_class_mask; caddr_t pvi_name; } pci_vendor_info_t; #define PVID(vendor, devid, name) {vendor, devid, 0, 0, 0, 0, name} #define PVID_OEM(vendor, devid, svid, sdevid, revid, name) {vendor, devid, svid, sdevid, revid, 0, name} #define PVID_END {0, 0, 0, 0, 0, 0, NULL} #define IFLIB_PNP_DESCR "U32:vendor;U32:device;U32:subvendor;U32:subdevice;" \ "U32:revision;U32:class;D:human" #define IFLIB_PNP_INFO(b, u, t) \ MODULE_PNP_INFO(IFLIB_PNP_DESCR, b, u, t, sizeof(t[0]), nitems(t)) typedef struct if_txrx { int (*ift_txd_encap) (void *, if_pkt_info_t); void (*ift_txd_flush) (void *, uint16_t, qidx_t pidx); int (*ift_txd_credits_update) (void *, uint16_t qsidx, bool clear); int (*ift_rxd_available) (void *, uint16_t qsidx, qidx_t pidx, qidx_t budget); int (*ift_rxd_pkt_get) (void *, if_rxd_info_t ri); void (*ift_rxd_refill) (void * , if_rxd_update_t iru); void (*ift_rxd_flush) (void *, uint16_t qsidx, uint8_t flidx, qidx_t pidx); int (*ift_legacy_intr) (void *); } *if_txrx_t; typedef struct if_softc_ctx { int isc_vectors; int isc_nrxqsets; int isc_ntxqsets; int isc_msix_bar; /* can be model specific - initialize in attach_pre */ int isc_tx_nsegments; /* can be model specific - initialize in attach_pre */ int isc_ntxd[8]; int isc_nrxd[8]; uint32_t isc_txqsizes[8]; uint32_t isc_rxqsizes[8]; /* is there such thing as a descriptor that is more than 248 bytes ? */ uint8_t isc_txd_size[8]; uint8_t isc_rxd_size[8]; int isc_tx_tso_segments_max; int isc_tx_tso_size_max; int isc_tx_tso_segsize_max; int isc_tx_csum_flags; int isc_capenable; int isc_rss_table_size; int isc_rss_table_mask; int isc_nrxqsets_max; int isc_ntxqsets_max; iflib_intr_mode_t isc_intr; uint16_t isc_max_frame_size; /* set at init time by driver */ uint16_t isc_min_frame_size; /* set at init time by driver, only used if IFLIB_NEED_ETHER_PAD is set. */ uint32_t isc_pause_frames; /* set by driver for iflib_timer to detect */ pci_vendor_info_t isc_vendor_info; /* set by iflib prior to attach_pre */ int isc_disable_msix; if_txrx_t isc_txrx; } *if_softc_ctx_t; /* * Initialization values for device */ struct if_shared_ctx { int isc_magic; driver_t *isc_driver; bus_size_t isc_q_align; bus_size_t isc_tx_maxsize; bus_size_t isc_tx_maxsegsize; bus_size_t isc_rx_maxsize; bus_size_t isc_rx_maxsegsize; int isc_rx_nsegments; int isc_admin_intrcnt; /* # of admin/link interrupts */ /* fields necessary for probe */ pci_vendor_info_t *isc_vendor_info; char *isc_driver_version; /* optional function to transform the read values to match the table*/ void (*isc_parse_devinfo) (uint16_t *device_id, uint16_t *subvendor_id, uint16_t *subdevice_id, uint16_t *rev_id); int isc_nrxd_min[8]; int isc_nrxd_default[8]; int isc_nrxd_max[8]; int isc_ntxd_min[8]; int isc_ntxd_default[8]; int isc_ntxd_max[8]; /* actively used during operation */ int isc_nfl __aligned(CACHE_LINE_SIZE); int isc_ntxqs; /* # of tx queues per tx qset - usually 1 */ int isc_nrxqs; /* # of rx queues per rx qset - intel 1, chelsio 2, broadcom 3 */ int isc_rx_process_limit; int isc_tx_reclaim_thresh; int isc_flags; }; typedef struct iflib_dma_info { bus_addr_t idi_paddr; caddr_t idi_vaddr; bus_dma_tag_t idi_tag; bus_dmamap_t idi_map; uint32_t idi_size; } *iflib_dma_info_t; #define IFLIB_MAGIC 0xCAFEF00D typedef enum { IFLIB_INTR_RX, IFLIB_INTR_TX, IFLIB_INTR_RXTX, IFLIB_INTR_ADMIN, IFLIB_INTR_IOV, } iflib_intr_type_t; #ifndef ETH_ADDR_LEN #define ETH_ADDR_LEN 6 #endif /* * Interface has a separate command queue for RX */ #define IFLIB_HAS_RXCQ 0x01 /* * Driver has already allocated vectors */ #define IFLIB_SKIP_MSIX 0x02 /* * Interface is a virtual function */ #define IFLIB_IS_VF 0x04 /* * Interface has a separate command queue for TX */ #define IFLIB_HAS_TXCQ 0x08 /* * Interface does checksum in place */ #define IFLIB_NEED_SCRATCH 0x10 /* * Interface doesn't expect in_pseudo for th_sum */ #define IFLIB_TSO_INIT_IP 0x20 /* * Interface doesn't align IP header */ #define IFLIB_DO_RX_FIXUP 0x40 /* * Driver needs csum zeroed for offloading */ #define IFLIB_NEED_ZERO_CSUM 0x80 /* * Driver needs frames padded to some minimum length */ #define IFLIB_NEED_ETHER_PAD 0x100 /* * field accessors */ void *iflib_get_softc(if_ctx_t ctx); device_t iflib_get_dev(if_ctx_t ctx); if_t iflib_get_ifp(if_ctx_t ctx); struct ifmedia *iflib_get_media(if_ctx_t ctx); if_softc_ctx_t iflib_get_softc_ctx(if_ctx_t ctx); if_shared_ctx_t iflib_get_sctx(if_ctx_t ctx); void iflib_set_mac(if_ctx_t ctx, uint8_t mac[ETHER_ADDR_LEN]); /* * If the driver can plug cleanly in to newbus use these */ int iflib_device_probe(device_t); int iflib_device_attach(device_t); int iflib_device_detach(device_t); int iflib_device_suspend(device_t); int iflib_device_resume(device_t); int iflib_device_shutdown(device_t); int iflib_device_iov_init(device_t, uint16_t, const nvlist_t *); void iflib_device_iov_uninit(device_t); int iflib_device_iov_add_vf(device_t, uint16_t, const nvlist_t *); /* * If the driver can't plug cleanly in to newbus * use these */ int iflib_device_register(device_t dev, void *softc, if_shared_ctx_t sctx, if_ctx_t *ctxp); int iflib_device_deregister(if_ctx_t); int iflib_irq_alloc(if_ctx_t, if_irq_t, int, driver_filter_t, void *filter_arg, driver_intr_t, void *arg, char *name); 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); void iflib_softirq_alloc_generic(if_ctx_t ctx, if_irq_t irq, iflib_intr_type_t type, void *arg, int qid, char *name); void iflib_irq_free(if_ctx_t ctx, if_irq_t irq); void iflib_io_tqg_attach(struct grouptask *gt, void *uniq, int cpu, char *name); void iflib_config_gtask_init(if_ctx_t ctx, struct grouptask *gtask, gtask_fn_t *fn, char *name); void iflib_config_gtask_deinit(struct grouptask *gtask); void iflib_tx_intr_deferred(if_ctx_t ctx, int txqid); void iflib_rx_intr_deferred(if_ctx_t ctx, int rxqid); void iflib_admin_intr_deferred(if_ctx_t ctx); void iflib_iov_intr_deferred(if_ctx_t ctx); void iflib_link_state_change(if_ctx_t ctx, int linkstate, uint64_t baudrate); int iflib_dma_alloc(if_ctx_t ctx, int size, iflib_dma_info_t dma, int mapflags); void iflib_dma_free(iflib_dma_info_t dma); int iflib_dma_alloc_multi(if_ctx_t ctx, int *sizes, iflib_dma_info_t *dmalist, int mapflags, int count); void iflib_dma_free_multi(iflib_dma_info_t *dmalist, int count); struct mtx *iflib_ctx_lock_get(if_ctx_t); struct mtx *iflib_qset_lock_get(if_ctx_t, uint16_t); void iflib_led_create(if_ctx_t ctx); void iflib_add_int_delay_sysctl(if_ctx_t, const char *, const char *, if_int_delay_info_t, int, int); #endif /* __IFLIB_H_ */ Index: head/sys/sys/gtaskqueue.h =================================================================== --- head/sys/sys/gtaskqueue.h (revision 326983) +++ head/sys/sys/gtaskqueue.h (revision 326984) @@ -1,110 +1,110 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2014 Jeffrey Roberson - * Copyright (c) 2016 Matthew Macy + * Copyright (c) 2016 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. 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. * * $FreeBSD$ */ #ifndef _SYS_GTASKQUEUE_H_ #define _SYS_GTASKQUEUE_H_ #include #ifndef _KERNEL #error "no user-serviceable parts inside" #endif struct gtaskqueue; typedef void (*gtaskqueue_enqueue_fn)(void *context); /* * Taskqueue groups. Manages dynamic thread groups and irq binding for * device and other tasks. */ void gtaskqueue_block(struct gtaskqueue *queue); void gtaskqueue_unblock(struct gtaskqueue *queue); int gtaskqueue_cancel(struct gtaskqueue *queue, struct gtask *gtask); void gtaskqueue_drain(struct gtaskqueue *queue, struct gtask *task); void gtaskqueue_drain_all(struct gtaskqueue *queue); int grouptaskqueue_enqueue(struct gtaskqueue *queue, struct gtask *task); void taskqgroup_attach(struct taskqgroup *qgroup, struct grouptask *grptask, void *uniq, int irq, char *name); int taskqgroup_attach_cpu(struct taskqgroup *qgroup, struct grouptask *grptask, void *uniq, int cpu, int irq, char *name); void taskqgroup_detach(struct taskqgroup *qgroup, struct grouptask *gtask); struct taskqgroup *taskqgroup_create(char *name); void taskqgroup_destroy(struct taskqgroup *qgroup); int taskqgroup_adjust(struct taskqgroup *qgroup, int cnt, int stride); #define TASK_ENQUEUED 0x1 #define TASK_SKIP_WAKEUP 0x2 #define GTASK_INIT(task, flags, priority, func, context) do { \ (task)->ta_flags = flags; \ (task)->ta_priority = (priority); \ (task)->ta_func = (func); \ (task)->ta_context = (context); \ } while (0) #define GROUPTASK_INIT(gtask, priority, func, context) \ GTASK_INIT(&(gtask)->gt_task, TASK_SKIP_WAKEUP, priority, func, context) #define GROUPTASK_ENQUEUE(gtask) \ grouptaskqueue_enqueue((gtask)->gt_taskqueue, &(gtask)->gt_task) #define TASKQGROUP_DECLARE(name) \ extern struct taskqgroup *qgroup_##name #define TASKQGROUP_DEFINE(name, cnt, stride) \ \ struct taskqgroup *qgroup_##name; \ \ static void \ taskqgroup_define_##name(void *arg) \ { \ qgroup_##name = taskqgroup_create(#name); \ } \ \ SYSINIT(taskqgroup_##name, SI_SUB_TASKQ, SI_ORDER_FIRST, \ taskqgroup_define_##name, NULL); \ \ static void \ taskqgroup_adjust_##name(void *arg) \ { \ taskqgroup_adjust(qgroup_##name, (cnt), (stride)); \ } \ \ SYSINIT(taskqgroup_adj_##name, SI_SUB_SMP, SI_ORDER_ANY, \ taskqgroup_adjust_##name, NULL) TASKQGROUP_DECLARE(net); TASKQGROUP_DECLARE(softirq); #endif /* !_SYS_GTASKQUEUE_H_ */