Index: head/sys/dev/virtio/balloon/virtio_balloon.c =================================================================== --- head/sys/dev/virtio/balloon/virtio_balloon.c (revision 348598) +++ head/sys/dev/virtio/balloon/virtio_balloon.c (revision 348599) @@ -1,566 +1,564 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2011, Bryan Venteicher * 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. */ /* Driver for VirtIO memory balloon devices. */ #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 "virtio_if.h" struct vtballoon_softc { device_t vtballoon_dev; struct mtx vtballoon_mtx; uint64_t vtballoon_features; uint32_t vtballoon_flags; #define VTBALLOON_FLAG_DETACH 0x01 struct virtqueue *vtballoon_inflate_vq; struct virtqueue *vtballoon_deflate_vq; uint32_t vtballoon_desired_npages; uint32_t vtballoon_current_npages; TAILQ_HEAD(,vm_page) vtballoon_pages; struct thread *vtballoon_td; uint32_t *vtballoon_page_frames; int vtballoon_timeout; }; static struct virtio_feature_desc vtballoon_feature_desc[] = { { VIRTIO_BALLOON_F_MUST_TELL_HOST, "MustTellHost" }, { VIRTIO_BALLOON_F_STATS_VQ, "StatsVq" }, { 0, NULL } }; static int vtballoon_probe(device_t); static int vtballoon_attach(device_t); static int vtballoon_detach(device_t); static int vtballoon_config_change(device_t); static void vtballoon_negotiate_features(struct vtballoon_softc *); static int vtballoon_alloc_virtqueues(struct vtballoon_softc *); static void vtballoon_vq_intr(void *); static void vtballoon_inflate(struct vtballoon_softc *, int); static void vtballoon_deflate(struct vtballoon_softc *, int); static void vtballoon_send_page_frames(struct vtballoon_softc *, struct virtqueue *, int); static void vtballoon_pop(struct vtballoon_softc *); static void vtballoon_stop(struct vtballoon_softc *); static vm_page_t vtballoon_alloc_page(struct vtballoon_softc *); static void vtballoon_free_page(struct vtballoon_softc *, vm_page_t); static int vtballoon_sleep(struct vtballoon_softc *); static void vtballoon_thread(void *); static void vtballoon_add_sysctl(struct vtballoon_softc *); /* Features desired/implemented by this driver. */ #define VTBALLOON_FEATURES 0 /* Timeout between retries when the balloon needs inflating. */ #define VTBALLOON_LOWMEM_TIMEOUT hz /* * Maximum number of pages we'll request to inflate or deflate * the balloon in one virtqueue request. Both Linux and NetBSD * have settled on 256, doing up to 1MB at a time. */ #define VTBALLOON_PAGES_PER_REQUEST 256 /* Must be able to fix all pages frames in one page (segment). */ CTASSERT(VTBALLOON_PAGES_PER_REQUEST * sizeof(uint32_t) <= PAGE_SIZE); #define VTBALLOON_MTX(_sc) &(_sc)->vtballoon_mtx #define VTBALLOON_LOCK_INIT(_sc, _name) mtx_init(VTBALLOON_MTX((_sc)), _name, \ "VirtIO Balloon Lock", MTX_DEF) #define VTBALLOON_LOCK(_sc) mtx_lock(VTBALLOON_MTX((_sc))) #define VTBALLOON_UNLOCK(_sc) mtx_unlock(VTBALLOON_MTX((_sc))) #define VTBALLOON_LOCK_DESTROY(_sc) mtx_destroy(VTBALLOON_MTX((_sc))) static device_method_t vtballoon_methods[] = { /* Device methods. */ DEVMETHOD(device_probe, vtballoon_probe), DEVMETHOD(device_attach, vtballoon_attach), DEVMETHOD(device_detach, vtballoon_detach), /* VirtIO methods. */ DEVMETHOD(virtio_config_change, vtballoon_config_change), DEVMETHOD_END }; static driver_t vtballoon_driver = { "vtballoon", vtballoon_methods, sizeof(struct vtballoon_softc) }; static devclass_t vtballoon_devclass; DRIVER_MODULE(virtio_balloon, virtio_pci, vtballoon_driver, vtballoon_devclass, 0, 0); MODULE_VERSION(virtio_balloon, 1); MODULE_DEPEND(virtio_balloon, virtio, 1, 1, 1); +VIRTIO_SIMPLE_PNPTABLE(virtio_balloon, VIRTIO_ID_BALLOON, + "VirtIO Balloon Adapter"); +VIRTIO_SIMPLE_PNPINFO(virtio_pci, virtio_balloon); + static int vtballoon_probe(device_t dev) { - - if (virtio_get_device_type(dev) != VIRTIO_ID_BALLOON) - return (ENXIO); - - device_set_desc(dev, "VirtIO Balloon Adapter"); - - return (BUS_PROBE_DEFAULT); + return (VIRTIO_SIMPLE_PROBE(dev, virtio_balloon)); } static int vtballoon_attach(device_t dev) { struct vtballoon_softc *sc; int error; sc = device_get_softc(dev); sc->vtballoon_dev = dev; VTBALLOON_LOCK_INIT(sc, device_get_nameunit(dev)); TAILQ_INIT(&sc->vtballoon_pages); vtballoon_add_sysctl(sc); virtio_set_feature_desc(dev, vtballoon_feature_desc); vtballoon_negotiate_features(sc); sc->vtballoon_page_frames = malloc(VTBALLOON_PAGES_PER_REQUEST * sizeof(uint32_t), M_DEVBUF, M_NOWAIT | M_ZERO); if (sc->vtballoon_page_frames == NULL) { error = ENOMEM; device_printf(dev, "cannot allocate page frame request array\n"); goto fail; } error = vtballoon_alloc_virtqueues(sc); if (error) { device_printf(dev, "cannot allocate virtqueues\n"); goto fail; } error = virtio_setup_intr(dev, INTR_TYPE_MISC); if (error) { device_printf(dev, "cannot setup virtqueue interrupts\n"); goto fail; } error = kthread_add(vtballoon_thread, sc, NULL, &sc->vtballoon_td, 0, 0, "virtio_balloon"); if (error) { device_printf(dev, "cannot create balloon kthread\n"); goto fail; } virtqueue_enable_intr(sc->vtballoon_inflate_vq); virtqueue_enable_intr(sc->vtballoon_deflate_vq); fail: if (error) vtballoon_detach(dev); return (error); } static int vtballoon_detach(device_t dev) { struct vtballoon_softc *sc; sc = device_get_softc(dev); if (sc->vtballoon_td != NULL) { VTBALLOON_LOCK(sc); sc->vtballoon_flags |= VTBALLOON_FLAG_DETACH; wakeup_one(sc); msleep(sc->vtballoon_td, VTBALLOON_MTX(sc), 0, "vtbdth", 0); VTBALLOON_UNLOCK(sc); sc->vtballoon_td = NULL; } if (device_is_attached(dev)) { vtballoon_pop(sc); vtballoon_stop(sc); } if (sc->vtballoon_page_frames != NULL) { free(sc->vtballoon_page_frames, M_DEVBUF); sc->vtballoon_page_frames = NULL; } VTBALLOON_LOCK_DESTROY(sc); return (0); } static int vtballoon_config_change(device_t dev) { struct vtballoon_softc *sc; sc = device_get_softc(dev); VTBALLOON_LOCK(sc); wakeup_one(sc); VTBALLOON_UNLOCK(sc); return (1); } static void vtballoon_negotiate_features(struct vtballoon_softc *sc) { device_t dev; uint64_t features; dev = sc->vtballoon_dev; features = virtio_negotiate_features(dev, VTBALLOON_FEATURES); sc->vtballoon_features = features; } static int vtballoon_alloc_virtqueues(struct vtballoon_softc *sc) { device_t dev; struct vq_alloc_info vq_info[2]; int nvqs; dev = sc->vtballoon_dev; nvqs = 2; VQ_ALLOC_INFO_INIT(&vq_info[0], 0, vtballoon_vq_intr, sc, &sc->vtballoon_inflate_vq, "%s inflate", device_get_nameunit(dev)); VQ_ALLOC_INFO_INIT(&vq_info[1], 0, vtballoon_vq_intr, sc, &sc->vtballoon_deflate_vq, "%s deflate", device_get_nameunit(dev)); return (virtio_alloc_virtqueues(dev, 0, nvqs, vq_info)); } static void vtballoon_vq_intr(void *xsc) { struct vtballoon_softc *sc; sc = xsc; VTBALLOON_LOCK(sc); wakeup_one(sc); VTBALLOON_UNLOCK(sc); } static void vtballoon_inflate(struct vtballoon_softc *sc, int npages) { struct virtqueue *vq; vm_page_t m; int i; vq = sc->vtballoon_inflate_vq; if (npages > VTBALLOON_PAGES_PER_REQUEST) npages = VTBALLOON_PAGES_PER_REQUEST; for (i = 0; i < npages; i++) { if ((m = vtballoon_alloc_page(sc)) == NULL) { sc->vtballoon_timeout = VTBALLOON_LOWMEM_TIMEOUT; break; } sc->vtballoon_page_frames[i] = VM_PAGE_TO_PHYS(m) >> VIRTIO_BALLOON_PFN_SHIFT; KASSERT(m->queue == PQ_NONE, ("%s: allocated page %p on queue", __func__, m)); TAILQ_INSERT_TAIL(&sc->vtballoon_pages, m, plinks.q); } if (i > 0) vtballoon_send_page_frames(sc, vq, i); } static void vtballoon_deflate(struct vtballoon_softc *sc, int npages) { TAILQ_HEAD(, vm_page) free_pages; struct virtqueue *vq; vm_page_t m; int i; vq = sc->vtballoon_deflate_vq; TAILQ_INIT(&free_pages); if (npages > VTBALLOON_PAGES_PER_REQUEST) npages = VTBALLOON_PAGES_PER_REQUEST; for (i = 0; i < npages; i++) { m = TAILQ_FIRST(&sc->vtballoon_pages); KASSERT(m != NULL, ("%s: no more pages to deflate", __func__)); sc->vtballoon_page_frames[i] = VM_PAGE_TO_PHYS(m) >> VIRTIO_BALLOON_PFN_SHIFT; TAILQ_REMOVE(&sc->vtballoon_pages, m, plinks.q); TAILQ_INSERT_TAIL(&free_pages, m, plinks.q); } if (i > 0) { /* Always tell host first before freeing the pages. */ vtballoon_send_page_frames(sc, vq, i); while ((m = TAILQ_FIRST(&free_pages)) != NULL) { TAILQ_REMOVE(&free_pages, m, plinks.q); vtballoon_free_page(sc, m); } } KASSERT((TAILQ_EMPTY(&sc->vtballoon_pages) && sc->vtballoon_current_npages == 0) || (!TAILQ_EMPTY(&sc->vtballoon_pages) && sc->vtballoon_current_npages != 0), ("%s: bogus page count %d", __func__, sc->vtballoon_current_npages)); } static void vtballoon_send_page_frames(struct vtballoon_softc *sc, struct virtqueue *vq, int npages) { struct sglist sg; struct sglist_seg segs[1]; void *c; int error; sglist_init(&sg, 1, segs); error = sglist_append(&sg, sc->vtballoon_page_frames, npages * sizeof(uint32_t)); KASSERT(error == 0, ("error adding page frames to sglist")); error = virtqueue_enqueue(vq, vq, &sg, 1, 0); KASSERT(error == 0, ("error enqueuing page frames to virtqueue")); virtqueue_notify(vq); /* * Inflate and deflate operations are done synchronously. The * interrupt handler will wake us up. */ VTBALLOON_LOCK(sc); while ((c = virtqueue_dequeue(vq, NULL)) == NULL) msleep(sc, VTBALLOON_MTX(sc), 0, "vtbspf", 0); VTBALLOON_UNLOCK(sc); KASSERT(c == vq, ("unexpected balloon operation response")); } static void vtballoon_pop(struct vtballoon_softc *sc) { while (!TAILQ_EMPTY(&sc->vtballoon_pages)) vtballoon_deflate(sc, sc->vtballoon_current_npages); } static void vtballoon_stop(struct vtballoon_softc *sc) { virtqueue_disable_intr(sc->vtballoon_inflate_vq); virtqueue_disable_intr(sc->vtballoon_deflate_vq); virtio_stop(sc->vtballoon_dev); } static vm_page_t vtballoon_alloc_page(struct vtballoon_softc *sc) { vm_page_t m; m = vm_page_alloc(NULL, 0, VM_ALLOC_NORMAL | VM_ALLOC_NOOBJ); if (m != NULL) sc->vtballoon_current_npages++; return (m); } static void vtballoon_free_page(struct vtballoon_softc *sc, vm_page_t m) { vm_page_free(m); sc->vtballoon_current_npages--; } static uint32_t vtballoon_desired_size(struct vtballoon_softc *sc) { uint32_t desired; desired = virtio_read_dev_config_4(sc->vtballoon_dev, offsetof(struct virtio_balloon_config, num_pages)); return (le32toh(desired)); } static void vtballoon_update_size(struct vtballoon_softc *sc) { virtio_write_dev_config_4(sc->vtballoon_dev, offsetof(struct virtio_balloon_config, actual), htole32(sc->vtballoon_current_npages)); } static int vtballoon_sleep(struct vtballoon_softc *sc) { int rc, timeout; uint32_t current, desired; rc = 0; current = sc->vtballoon_current_npages; VTBALLOON_LOCK(sc); for (;;) { if (sc->vtballoon_flags & VTBALLOON_FLAG_DETACH) { rc = 1; break; } desired = vtballoon_desired_size(sc); sc->vtballoon_desired_npages = desired; /* * If given, use non-zero timeout on the first time through * the loop. On subsequent times, timeout will be zero so * we will reevaluate the desired size of the balloon and * break out to retry if needed. */ timeout = sc->vtballoon_timeout; sc->vtballoon_timeout = 0; if (current > desired) break; if (current < desired && timeout == 0) break; msleep(sc, VTBALLOON_MTX(sc), 0, "vtbslp", timeout); } VTBALLOON_UNLOCK(sc); return (rc); } static void vtballoon_thread(void *xsc) { struct vtballoon_softc *sc; uint32_t current, desired; sc = xsc; for (;;) { if (vtballoon_sleep(sc) != 0) break; current = sc->vtballoon_current_npages; desired = sc->vtballoon_desired_npages; if (desired != current) { if (desired > current) vtballoon_inflate(sc, desired - current); else vtballoon_deflate(sc, current - desired); vtballoon_update_size(sc); } } kthread_exit(); } static void vtballoon_add_sysctl(struct vtballoon_softc *sc) { device_t dev; struct sysctl_ctx_list *ctx; struct sysctl_oid *tree; struct sysctl_oid_list *child; dev = sc->vtballoon_dev; ctx = device_get_sysctl_ctx(dev); tree = device_get_sysctl_tree(dev); child = SYSCTL_CHILDREN(tree); SYSCTL_ADD_UINT(ctx, child, OID_AUTO, "desired", CTLFLAG_RD, &sc->vtballoon_desired_npages, sizeof(uint32_t), "Desired balloon size in pages"); SYSCTL_ADD_UINT(ctx, child, OID_AUTO, "current", CTLFLAG_RD, &sc->vtballoon_current_npages, sizeof(uint32_t), "Current balloon size in pages"); } Index: head/sys/dev/virtio/block/virtio_blk.c =================================================================== --- head/sys/dev/virtio/block/virtio_blk.c (revision 348598) +++ head/sys/dev/virtio/block/virtio_blk.c (revision 348599) @@ -1,1405 +1,1403 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2011, Bryan Venteicher * 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. */ /* Driver for VirtIO block devices. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "virtio_if.h" struct vtblk_request { struct virtio_blk_outhdr vbr_hdr; struct bio *vbr_bp; uint8_t vbr_ack; TAILQ_ENTRY(vtblk_request) vbr_link; }; enum vtblk_cache_mode { VTBLK_CACHE_WRITETHROUGH, VTBLK_CACHE_WRITEBACK, VTBLK_CACHE_MAX }; struct vtblk_softc { device_t vtblk_dev; struct mtx vtblk_mtx; uint64_t vtblk_features; uint32_t vtblk_flags; #define VTBLK_FLAG_INDIRECT 0x0001 #define VTBLK_FLAG_READONLY 0x0002 #define VTBLK_FLAG_DETACH 0x0004 #define VTBLK_FLAG_SUSPEND 0x0008 #define VTBLK_FLAG_BARRIER 0x0010 #define VTBLK_FLAG_WC_CONFIG 0x0020 struct virtqueue *vtblk_vq; struct sglist *vtblk_sglist; struct disk *vtblk_disk; struct bio_queue_head vtblk_bioq; TAILQ_HEAD(, vtblk_request) vtblk_req_free; TAILQ_HEAD(, vtblk_request) vtblk_req_ready; struct vtblk_request *vtblk_req_ordered; int vtblk_max_nsegs; int vtblk_request_count; enum vtblk_cache_mode vtblk_write_cache; struct bio_queue vtblk_dump_queue; struct vtblk_request vtblk_dump_request; }; static struct virtio_feature_desc vtblk_feature_desc[] = { { VIRTIO_BLK_F_BARRIER, "HostBarrier" }, { VIRTIO_BLK_F_SIZE_MAX, "MaxSegSize" }, { VIRTIO_BLK_F_SEG_MAX, "MaxNumSegs" }, { VIRTIO_BLK_F_GEOMETRY, "DiskGeometry" }, { VIRTIO_BLK_F_RO, "ReadOnly" }, { VIRTIO_BLK_F_BLK_SIZE, "BlockSize" }, { VIRTIO_BLK_F_SCSI, "SCSICmds" }, { VIRTIO_BLK_F_WCE, "WriteCache" }, { VIRTIO_BLK_F_TOPOLOGY, "Topology" }, { VIRTIO_BLK_F_CONFIG_WCE, "ConfigWCE" }, { 0, NULL } }; static int vtblk_modevent(module_t, int, void *); static int vtblk_probe(device_t); static int vtblk_attach(device_t); static int vtblk_detach(device_t); static int vtblk_suspend(device_t); static int vtblk_resume(device_t); static int vtblk_shutdown(device_t); static int vtblk_config_change(device_t); static int vtblk_open(struct disk *); static int vtblk_close(struct disk *); static int vtblk_ioctl(struct disk *, u_long, void *, int, struct thread *); static int vtblk_dump(void *, void *, vm_offset_t, off_t, size_t); static void vtblk_strategy(struct bio *); static void vtblk_negotiate_features(struct vtblk_softc *); static void vtblk_setup_features(struct vtblk_softc *); static int vtblk_maximum_segments(struct vtblk_softc *, struct virtio_blk_config *); static int vtblk_alloc_virtqueue(struct vtblk_softc *); static void vtblk_resize_disk(struct vtblk_softc *, uint64_t); static void vtblk_alloc_disk(struct vtblk_softc *, struct virtio_blk_config *); static void vtblk_create_disk(struct vtblk_softc *); static int vtblk_request_prealloc(struct vtblk_softc *); static void vtblk_request_free(struct vtblk_softc *); static struct vtblk_request * vtblk_request_dequeue(struct vtblk_softc *); static void vtblk_request_enqueue(struct vtblk_softc *, struct vtblk_request *); static struct vtblk_request * vtblk_request_next_ready(struct vtblk_softc *); static void vtblk_request_requeue_ready(struct vtblk_softc *, struct vtblk_request *); static struct vtblk_request * vtblk_request_next(struct vtblk_softc *); static struct vtblk_request * vtblk_request_bio(struct vtblk_softc *); static int vtblk_request_execute(struct vtblk_softc *, struct vtblk_request *); static int vtblk_request_error(struct vtblk_request *); static void vtblk_queue_completed(struct vtblk_softc *, struct bio_queue *); static void vtblk_done_completed(struct vtblk_softc *, struct bio_queue *); static void vtblk_drain_vq(struct vtblk_softc *); static void vtblk_drain(struct vtblk_softc *); static void vtblk_startio(struct vtblk_softc *); static void vtblk_bio_done(struct vtblk_softc *, struct bio *, int); static void vtblk_read_config(struct vtblk_softc *, struct virtio_blk_config *); static void vtblk_ident(struct vtblk_softc *); static int vtblk_poll_request(struct vtblk_softc *, struct vtblk_request *); static int vtblk_quiesce(struct vtblk_softc *); static void vtblk_vq_intr(void *); static void vtblk_stop(struct vtblk_softc *); static void vtblk_dump_quiesce(struct vtblk_softc *); static int vtblk_dump_write(struct vtblk_softc *, void *, off_t, size_t); static int vtblk_dump_flush(struct vtblk_softc *); static void vtblk_dump_complete(struct vtblk_softc *); static void vtblk_set_write_cache(struct vtblk_softc *, int); static int vtblk_write_cache_enabled(struct vtblk_softc *sc, struct virtio_blk_config *); static int vtblk_write_cache_sysctl(SYSCTL_HANDLER_ARGS); static void vtblk_setup_sysctl(struct vtblk_softc *); static int vtblk_tunable_int(struct vtblk_softc *, const char *, int); /* Tunables. */ static int vtblk_no_ident = 0; TUNABLE_INT("hw.vtblk.no_ident", &vtblk_no_ident); static int vtblk_writecache_mode = -1; TUNABLE_INT("hw.vtblk.writecache_mode", &vtblk_writecache_mode); /* Features desired/implemented by this driver. */ #define VTBLK_FEATURES \ (VIRTIO_BLK_F_BARRIER | \ VIRTIO_BLK_F_SIZE_MAX | \ VIRTIO_BLK_F_SEG_MAX | \ VIRTIO_BLK_F_GEOMETRY | \ VIRTIO_BLK_F_RO | \ VIRTIO_BLK_F_BLK_SIZE | \ VIRTIO_BLK_F_WCE | \ VIRTIO_BLK_F_TOPOLOGY | \ VIRTIO_BLK_F_CONFIG_WCE | \ VIRTIO_RING_F_INDIRECT_DESC) #define VTBLK_MTX(_sc) &(_sc)->vtblk_mtx #define VTBLK_LOCK_INIT(_sc, _name) \ mtx_init(VTBLK_MTX((_sc)), (_name), \ "VirtIO Block Lock", MTX_DEF) #define VTBLK_LOCK(_sc) mtx_lock(VTBLK_MTX((_sc))) #define VTBLK_UNLOCK(_sc) mtx_unlock(VTBLK_MTX((_sc))) #define VTBLK_LOCK_DESTROY(_sc) mtx_destroy(VTBLK_MTX((_sc))) #define VTBLK_LOCK_ASSERT(_sc) mtx_assert(VTBLK_MTX((_sc)), MA_OWNED) #define VTBLK_LOCK_ASSERT_NOTOWNED(_sc) \ mtx_assert(VTBLK_MTX((_sc)), MA_NOTOWNED) #define VTBLK_DISK_NAME "vtbd" #define VTBLK_QUIESCE_TIMEOUT (30 * hz) /* * Each block request uses at least two segments - one for the header * and one for the status. */ #define VTBLK_MIN_SEGMENTS 2 static device_method_t vtblk_methods[] = { /* Device methods. */ DEVMETHOD(device_probe, vtblk_probe), DEVMETHOD(device_attach, vtblk_attach), DEVMETHOD(device_detach, vtblk_detach), DEVMETHOD(device_suspend, vtblk_suspend), DEVMETHOD(device_resume, vtblk_resume), DEVMETHOD(device_shutdown, vtblk_shutdown), /* VirtIO methods. */ DEVMETHOD(virtio_config_change, vtblk_config_change), DEVMETHOD_END }; static driver_t vtblk_driver = { "vtblk", vtblk_methods, sizeof(struct vtblk_softc) }; static devclass_t vtblk_devclass; DRIVER_MODULE(virtio_blk, virtio_mmio, vtblk_driver, vtblk_devclass, vtblk_modevent, 0); DRIVER_MODULE(virtio_blk, virtio_pci, vtblk_driver, vtblk_devclass, vtblk_modevent, 0); MODULE_VERSION(virtio_blk, 1); MODULE_DEPEND(virtio_blk, virtio, 1, 1, 1); +VIRTIO_SIMPLE_PNPTABLE(virtio_blk, VIRTIO_ID_BLOCK, "VirtIO Block Adapter"); +VIRTIO_SIMPLE_PNPINFO(virtio_mmio, virtio_blk); +VIRTIO_SIMPLE_PNPINFO(virtio_pci, virtio_blk); + static int vtblk_modevent(module_t mod, int type, void *unused) { int error; error = 0; switch (type) { case MOD_LOAD: case MOD_QUIESCE: case MOD_UNLOAD: case MOD_SHUTDOWN: break; default: error = EOPNOTSUPP; break; } return (error); } static int vtblk_probe(device_t dev) { - - if (virtio_get_device_type(dev) != VIRTIO_ID_BLOCK) - return (ENXIO); - - device_set_desc(dev, "VirtIO Block Adapter"); - - return (BUS_PROBE_DEFAULT); + return (VIRTIO_SIMPLE_PROBE(dev, virtio_blk)); } static int vtblk_attach(device_t dev) { struct vtblk_softc *sc; struct virtio_blk_config blkcfg; int error; virtio_set_feature_desc(dev, vtblk_feature_desc); sc = device_get_softc(dev); sc->vtblk_dev = dev; VTBLK_LOCK_INIT(sc, device_get_nameunit(dev)); bioq_init(&sc->vtblk_bioq); TAILQ_INIT(&sc->vtblk_dump_queue); TAILQ_INIT(&sc->vtblk_req_free); TAILQ_INIT(&sc->vtblk_req_ready); vtblk_setup_sysctl(sc); vtblk_setup_features(sc); vtblk_read_config(sc, &blkcfg); /* * With the current sglist(9) implementation, it is not easy * for us to support a maximum segment size as adjacent * segments are coalesced. For now, just make sure it's larger * than the maximum supported transfer size. */ if (virtio_with_feature(dev, VIRTIO_BLK_F_SIZE_MAX)) { if (blkcfg.size_max < MAXPHYS) { error = ENOTSUP; device_printf(dev, "host requires unsupported " "maximum segment size feature\n"); goto fail; } } sc->vtblk_max_nsegs = vtblk_maximum_segments(sc, &blkcfg); if (sc->vtblk_max_nsegs <= VTBLK_MIN_SEGMENTS) { error = EINVAL; device_printf(dev, "fewer than minimum number of segments " "allowed: %d\n", sc->vtblk_max_nsegs); goto fail; } sc->vtblk_sglist = sglist_alloc(sc->vtblk_max_nsegs, M_NOWAIT); if (sc->vtblk_sglist == NULL) { error = ENOMEM; device_printf(dev, "cannot allocate sglist\n"); goto fail; } error = vtblk_alloc_virtqueue(sc); if (error) { device_printf(dev, "cannot allocate virtqueue\n"); goto fail; } error = vtblk_request_prealloc(sc); if (error) { device_printf(dev, "cannot preallocate requests\n"); goto fail; } vtblk_alloc_disk(sc, &blkcfg); error = virtio_setup_intr(dev, INTR_TYPE_BIO | INTR_ENTROPY); if (error) { device_printf(dev, "cannot setup virtqueue interrupt\n"); goto fail; } vtblk_create_disk(sc); virtqueue_enable_intr(sc->vtblk_vq); fail: if (error) vtblk_detach(dev); return (error); } static int vtblk_detach(device_t dev) { struct vtblk_softc *sc; sc = device_get_softc(dev); VTBLK_LOCK(sc); sc->vtblk_flags |= VTBLK_FLAG_DETACH; if (device_is_attached(dev)) vtblk_stop(sc); VTBLK_UNLOCK(sc); vtblk_drain(sc); if (sc->vtblk_disk != NULL) { disk_destroy(sc->vtblk_disk); sc->vtblk_disk = NULL; } if (sc->vtblk_sglist != NULL) { sglist_free(sc->vtblk_sglist); sc->vtblk_sglist = NULL; } VTBLK_LOCK_DESTROY(sc); return (0); } static int vtblk_suspend(device_t dev) { struct vtblk_softc *sc; int error; sc = device_get_softc(dev); VTBLK_LOCK(sc); sc->vtblk_flags |= VTBLK_FLAG_SUSPEND; /* XXX BMV: virtio_stop(), etc needed here? */ error = vtblk_quiesce(sc); if (error) sc->vtblk_flags &= ~VTBLK_FLAG_SUSPEND; VTBLK_UNLOCK(sc); return (error); } static int vtblk_resume(device_t dev) { struct vtblk_softc *sc; sc = device_get_softc(dev); VTBLK_LOCK(sc); /* XXX BMV: virtio_reinit(), etc needed here? */ sc->vtblk_flags &= ~VTBLK_FLAG_SUSPEND; vtblk_startio(sc); VTBLK_UNLOCK(sc); return (0); } static int vtblk_shutdown(device_t dev) { return (0); } static int vtblk_config_change(device_t dev) { struct vtblk_softc *sc; struct virtio_blk_config blkcfg; uint64_t capacity; sc = device_get_softc(dev); vtblk_read_config(sc, &blkcfg); /* Capacity is always in 512-byte units. */ capacity = blkcfg.capacity * 512; if (sc->vtblk_disk->d_mediasize != capacity) vtblk_resize_disk(sc, capacity); return (0); } static int vtblk_open(struct disk *dp) { struct vtblk_softc *sc; if ((sc = dp->d_drv1) == NULL) return (ENXIO); return (sc->vtblk_flags & VTBLK_FLAG_DETACH ? ENXIO : 0); } static int vtblk_close(struct disk *dp) { struct vtblk_softc *sc; if ((sc = dp->d_drv1) == NULL) return (ENXIO); return (0); } static int vtblk_ioctl(struct disk *dp, u_long cmd, void *addr, int flag, struct thread *td) { struct vtblk_softc *sc; if ((sc = dp->d_drv1) == NULL) return (ENXIO); return (ENOTTY); } static int vtblk_dump(void *arg, void *virtual, vm_offset_t physical, off_t offset, size_t length) { struct disk *dp; struct vtblk_softc *sc; int error; dp = arg; error = 0; if ((sc = dp->d_drv1) == NULL) return (ENXIO); VTBLK_LOCK(sc); vtblk_dump_quiesce(sc); if (length > 0) error = vtblk_dump_write(sc, virtual, offset, length); if (error || (virtual == NULL && offset == 0)) vtblk_dump_complete(sc); VTBLK_UNLOCK(sc); return (error); } static void vtblk_strategy(struct bio *bp) { struct vtblk_softc *sc; if ((sc = bp->bio_disk->d_drv1) == NULL) { vtblk_bio_done(NULL, bp, EINVAL); return; } /* * Fail any write if RO. Unfortunately, there does not seem to * be a better way to report our readonly'ness to GEOM above. */ if (sc->vtblk_flags & VTBLK_FLAG_READONLY && (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_FLUSH)) { vtblk_bio_done(sc, bp, EROFS); return; } VTBLK_LOCK(sc); if (sc->vtblk_flags & VTBLK_FLAG_DETACH) { VTBLK_UNLOCK(sc); vtblk_bio_done(sc, bp, ENXIO); return; } bioq_insert_tail(&sc->vtblk_bioq, bp); vtblk_startio(sc); VTBLK_UNLOCK(sc); } static void vtblk_negotiate_features(struct vtblk_softc *sc) { device_t dev; uint64_t features; dev = sc->vtblk_dev; features = VTBLK_FEATURES; sc->vtblk_features = virtio_negotiate_features(dev, features); } static void vtblk_setup_features(struct vtblk_softc *sc) { device_t dev; dev = sc->vtblk_dev; vtblk_negotiate_features(sc); if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) sc->vtblk_flags |= VTBLK_FLAG_INDIRECT; if (virtio_with_feature(dev, VIRTIO_BLK_F_RO)) sc->vtblk_flags |= VTBLK_FLAG_READONLY; if (virtio_with_feature(dev, VIRTIO_BLK_F_BARRIER)) sc->vtblk_flags |= VTBLK_FLAG_BARRIER; if (virtio_with_feature(dev, VIRTIO_BLK_F_CONFIG_WCE)) sc->vtblk_flags |= VTBLK_FLAG_WC_CONFIG; } static int vtblk_maximum_segments(struct vtblk_softc *sc, struct virtio_blk_config *blkcfg) { device_t dev; int nsegs; dev = sc->vtblk_dev; nsegs = VTBLK_MIN_SEGMENTS; if (virtio_with_feature(dev, VIRTIO_BLK_F_SEG_MAX)) { nsegs += MIN(blkcfg->seg_max, MAXPHYS / PAGE_SIZE + 1); if (sc->vtblk_flags & VTBLK_FLAG_INDIRECT) nsegs = MIN(nsegs, VIRTIO_MAX_INDIRECT); } else nsegs += 1; return (nsegs); } static int vtblk_alloc_virtqueue(struct vtblk_softc *sc) { device_t dev; struct vq_alloc_info vq_info; dev = sc->vtblk_dev; VQ_ALLOC_INFO_INIT(&vq_info, sc->vtblk_max_nsegs, vtblk_vq_intr, sc, &sc->vtblk_vq, "%s request", device_get_nameunit(dev)); return (virtio_alloc_virtqueues(dev, 0, 1, &vq_info)); } static void vtblk_resize_disk(struct vtblk_softc *sc, uint64_t new_capacity) { device_t dev; struct disk *dp; int error; dev = sc->vtblk_dev; dp = sc->vtblk_disk; dp->d_mediasize = new_capacity; if (bootverbose) { device_printf(dev, "resized to %juMB (%ju %u byte sectors)\n", (uintmax_t) dp->d_mediasize >> 20, (uintmax_t) dp->d_mediasize / dp->d_sectorsize, dp->d_sectorsize); } error = disk_resize(dp, M_NOWAIT); if (error) { device_printf(dev, "disk_resize(9) failed, error: %d\n", error); } } static void vtblk_alloc_disk(struct vtblk_softc *sc, struct virtio_blk_config *blkcfg) { device_t dev; struct disk *dp; dev = sc->vtblk_dev; sc->vtblk_disk = dp = disk_alloc(); dp->d_open = vtblk_open; dp->d_close = vtblk_close; dp->d_ioctl = vtblk_ioctl; dp->d_strategy = vtblk_strategy; dp->d_name = VTBLK_DISK_NAME; dp->d_unit = device_get_unit(dev); dp->d_drv1 = sc; dp->d_flags = DISKFLAG_CANFLUSHCACHE | DISKFLAG_UNMAPPED_BIO | DISKFLAG_DIRECT_COMPLETION; dp->d_hba_vendor = virtio_get_vendor(dev); dp->d_hba_device = virtio_get_device(dev); dp->d_hba_subvendor = virtio_get_subvendor(dev); dp->d_hba_subdevice = virtio_get_subdevice(dev); if ((sc->vtblk_flags & VTBLK_FLAG_READONLY) == 0) dp->d_dump = vtblk_dump; /* Capacity is always in 512-byte units. */ dp->d_mediasize = blkcfg->capacity * 512; if (virtio_with_feature(dev, VIRTIO_BLK_F_BLK_SIZE)) dp->d_sectorsize = blkcfg->blk_size; else dp->d_sectorsize = 512; /* * The VirtIO maximum I/O size is given in terms of segments. * However, FreeBSD limits I/O size by logical buffer size, not * by physically contiguous pages. Therefore, we have to assume * no pages are contiguous. This may impose an artificially low * maximum I/O size. But in practice, since QEMU advertises 128 * segments, this gives us a maximum IO size of 125 * PAGE_SIZE, * which is typically greater than MAXPHYS. Eventually we should * just advertise MAXPHYS and split buffers that are too big. * * Note we must subtract one additional segment in case of non * page aligned buffers. */ dp->d_maxsize = (sc->vtblk_max_nsegs - VTBLK_MIN_SEGMENTS - 1) * PAGE_SIZE; if (dp->d_maxsize < PAGE_SIZE) dp->d_maxsize = PAGE_SIZE; /* XXX */ if (virtio_with_feature(dev, VIRTIO_BLK_F_GEOMETRY)) { dp->d_fwsectors = blkcfg->geometry.sectors; dp->d_fwheads = blkcfg->geometry.heads; } if (virtio_with_feature(dev, VIRTIO_BLK_F_TOPOLOGY) && blkcfg->topology.physical_block_exp > 0) { dp->d_stripesize = dp->d_sectorsize * (1 << blkcfg->topology.physical_block_exp); dp->d_stripeoffset = (dp->d_stripesize - blkcfg->topology.alignment_offset * dp->d_sectorsize) % dp->d_stripesize; } if (vtblk_write_cache_enabled(sc, blkcfg) != 0) sc->vtblk_write_cache = VTBLK_CACHE_WRITEBACK; else sc->vtblk_write_cache = VTBLK_CACHE_WRITETHROUGH; } static void vtblk_create_disk(struct vtblk_softc *sc) { struct disk *dp; dp = sc->vtblk_disk; vtblk_ident(sc); device_printf(sc->vtblk_dev, "%juMB (%ju %u byte sectors)\n", (uintmax_t) dp->d_mediasize >> 20, (uintmax_t) dp->d_mediasize / dp->d_sectorsize, dp->d_sectorsize); disk_create(dp, DISK_VERSION); } static int vtblk_request_prealloc(struct vtblk_softc *sc) { struct vtblk_request *req; int i, nreqs; nreqs = virtqueue_size(sc->vtblk_vq); /* * Preallocate sufficient requests to keep the virtqueue full. Each * request consumes VTBLK_MIN_SEGMENTS or more descriptors so reduce * the number allocated when indirect descriptors are not available. */ if ((sc->vtblk_flags & VTBLK_FLAG_INDIRECT) == 0) nreqs /= VTBLK_MIN_SEGMENTS; for (i = 0; i < nreqs; i++) { req = malloc(sizeof(struct vtblk_request), M_DEVBUF, M_NOWAIT); if (req == NULL) return (ENOMEM); MPASS(sglist_count(&req->vbr_hdr, sizeof(req->vbr_hdr)) == 1); MPASS(sglist_count(&req->vbr_ack, sizeof(req->vbr_ack)) == 1); sc->vtblk_request_count++; vtblk_request_enqueue(sc, req); } return (0); } static void vtblk_request_free(struct vtblk_softc *sc) { struct vtblk_request *req; MPASS(TAILQ_EMPTY(&sc->vtblk_req_ready)); while ((req = vtblk_request_dequeue(sc)) != NULL) { sc->vtblk_request_count--; free(req, M_DEVBUF); } KASSERT(sc->vtblk_request_count == 0, ("%s: leaked %d requests", __func__, sc->vtblk_request_count)); } static struct vtblk_request * vtblk_request_dequeue(struct vtblk_softc *sc) { struct vtblk_request *req; req = TAILQ_FIRST(&sc->vtblk_req_free); if (req != NULL) { TAILQ_REMOVE(&sc->vtblk_req_free, req, vbr_link); bzero(req, sizeof(struct vtblk_request)); } return (req); } static void vtblk_request_enqueue(struct vtblk_softc *sc, struct vtblk_request *req) { TAILQ_INSERT_HEAD(&sc->vtblk_req_free, req, vbr_link); } static struct vtblk_request * vtblk_request_next_ready(struct vtblk_softc *sc) { struct vtblk_request *req; req = TAILQ_FIRST(&sc->vtblk_req_ready); if (req != NULL) TAILQ_REMOVE(&sc->vtblk_req_ready, req, vbr_link); return (req); } static void vtblk_request_requeue_ready(struct vtblk_softc *sc, struct vtblk_request *req) { /* NOTE: Currently, there will be at most one request in the queue. */ TAILQ_INSERT_HEAD(&sc->vtblk_req_ready, req, vbr_link); } static struct vtblk_request * vtblk_request_next(struct vtblk_softc *sc) { struct vtblk_request *req; req = vtblk_request_next_ready(sc); if (req != NULL) return (req); return (vtblk_request_bio(sc)); } static struct vtblk_request * vtblk_request_bio(struct vtblk_softc *sc) { struct bio_queue_head *bioq; struct vtblk_request *req; struct bio *bp; bioq = &sc->vtblk_bioq; if (bioq_first(bioq) == NULL) return (NULL); req = vtblk_request_dequeue(sc); if (req == NULL) return (NULL); bp = bioq_takefirst(bioq); req->vbr_bp = bp; req->vbr_ack = -1; req->vbr_hdr.ioprio = 1; switch (bp->bio_cmd) { case BIO_FLUSH: req->vbr_hdr.type = VIRTIO_BLK_T_FLUSH; break; case BIO_READ: req->vbr_hdr.type = VIRTIO_BLK_T_IN; req->vbr_hdr.sector = bp->bio_offset / 512; break; case BIO_WRITE: req->vbr_hdr.type = VIRTIO_BLK_T_OUT; req->vbr_hdr.sector = bp->bio_offset / 512; break; default: panic("%s: bio with unhandled cmd: %d", __func__, bp->bio_cmd); } if (bp->bio_flags & BIO_ORDERED) req->vbr_hdr.type |= VIRTIO_BLK_T_BARRIER; return (req); } static int vtblk_request_execute(struct vtblk_softc *sc, struct vtblk_request *req) { struct virtqueue *vq; struct sglist *sg; struct bio *bp; int ordered, readable, writable, error; vq = sc->vtblk_vq; sg = sc->vtblk_sglist; bp = req->vbr_bp; ordered = 0; writable = 0; /* * Some hosts (such as bhyve) do not implement the barrier feature, * so we emulate it in the driver by allowing the barrier request * to be the only one in flight. */ if ((sc->vtblk_flags & VTBLK_FLAG_BARRIER) == 0) { if (sc->vtblk_req_ordered != NULL) return (EBUSY); if (bp->bio_flags & BIO_ORDERED) { if (!virtqueue_empty(vq)) return (EBUSY); ordered = 1; req->vbr_hdr.type &= ~VIRTIO_BLK_T_BARRIER; } } sglist_reset(sg); sglist_append(sg, &req->vbr_hdr, sizeof(struct virtio_blk_outhdr)); if (bp->bio_cmd == BIO_READ || bp->bio_cmd == BIO_WRITE) { error = sglist_append_bio(sg, bp); if (error || sg->sg_nseg == sg->sg_maxseg) { panic("%s: bio %p data buffer too big %d", __func__, bp, error); } /* BIO_READ means the host writes into our buffer. */ if (bp->bio_cmd == BIO_READ) writable = sg->sg_nseg - 1; } writable++; sglist_append(sg, &req->vbr_ack, sizeof(uint8_t)); readable = sg->sg_nseg - writable; error = virtqueue_enqueue(vq, req, sg, readable, writable); if (error == 0 && ordered) sc->vtblk_req_ordered = req; return (error); } static int vtblk_request_error(struct vtblk_request *req) { int error; switch (req->vbr_ack) { case VIRTIO_BLK_S_OK: error = 0; break; case VIRTIO_BLK_S_UNSUPP: error = ENOTSUP; break; default: error = EIO; break; } return (error); } static void vtblk_queue_completed(struct vtblk_softc *sc, struct bio_queue *queue) { struct vtblk_request *req; struct bio *bp; while ((req = virtqueue_dequeue(sc->vtblk_vq, NULL)) != NULL) { if (sc->vtblk_req_ordered != NULL) { MPASS(sc->vtblk_req_ordered == req); sc->vtblk_req_ordered = NULL; } bp = req->vbr_bp; bp->bio_error = vtblk_request_error(req); TAILQ_INSERT_TAIL(queue, bp, bio_queue); vtblk_request_enqueue(sc, req); } } static void vtblk_done_completed(struct vtblk_softc *sc, struct bio_queue *queue) { struct bio *bp, *tmp; TAILQ_FOREACH_SAFE(bp, queue, bio_queue, tmp) { if (bp->bio_error != 0) disk_err(bp, "hard error", -1, 1); vtblk_bio_done(sc, bp, bp->bio_error); } } static void vtblk_drain_vq(struct vtblk_softc *sc) { struct virtqueue *vq; struct vtblk_request *req; int last; vq = sc->vtblk_vq; last = 0; while ((req = virtqueue_drain(vq, &last)) != NULL) { vtblk_bio_done(sc, req->vbr_bp, ENXIO); vtblk_request_enqueue(sc, req); } sc->vtblk_req_ordered = NULL; KASSERT(virtqueue_empty(vq), ("virtqueue not empty")); } static void vtblk_drain(struct vtblk_softc *sc) { struct bio_queue queue; struct bio_queue_head *bioq; struct vtblk_request *req; struct bio *bp; bioq = &sc->vtblk_bioq; TAILQ_INIT(&queue); if (sc->vtblk_vq != NULL) { vtblk_queue_completed(sc, &queue); vtblk_done_completed(sc, &queue); vtblk_drain_vq(sc); } while ((req = vtblk_request_next_ready(sc)) != NULL) { vtblk_bio_done(sc, req->vbr_bp, ENXIO); vtblk_request_enqueue(sc, req); } while (bioq_first(bioq) != NULL) { bp = bioq_takefirst(bioq); vtblk_bio_done(sc, bp, ENXIO); } vtblk_request_free(sc); } static void vtblk_startio(struct vtblk_softc *sc) { struct virtqueue *vq; struct vtblk_request *req; int enq; VTBLK_LOCK_ASSERT(sc); vq = sc->vtblk_vq; enq = 0; if (sc->vtblk_flags & VTBLK_FLAG_SUSPEND) return; while (!virtqueue_full(vq)) { req = vtblk_request_next(sc); if (req == NULL) break; if (vtblk_request_execute(sc, req) != 0) { vtblk_request_requeue_ready(sc, req); break; } enq++; } if (enq > 0) virtqueue_notify(vq); } static void vtblk_bio_done(struct vtblk_softc *sc, struct bio *bp, int error) { /* Because of GEOM direct dispatch, we cannot hold any locks. */ if (sc != NULL) VTBLK_LOCK_ASSERT_NOTOWNED(sc); if (error) { bp->bio_resid = bp->bio_bcount; bp->bio_error = error; bp->bio_flags |= BIO_ERROR; } biodone(bp); } #define VTBLK_GET_CONFIG(_dev, _feature, _field, _cfg) \ if (virtio_with_feature(_dev, _feature)) { \ virtio_read_device_config(_dev, \ offsetof(struct virtio_blk_config, _field), \ &(_cfg)->_field, sizeof((_cfg)->_field)); \ } static void vtblk_read_config(struct vtblk_softc *sc, struct virtio_blk_config *blkcfg) { device_t dev; dev = sc->vtblk_dev; bzero(blkcfg, sizeof(struct virtio_blk_config)); /* The capacity is always available. */ virtio_read_device_config(dev, offsetof(struct virtio_blk_config, capacity), &blkcfg->capacity, sizeof(blkcfg->capacity)); /* Read the configuration if the feature was negotiated. */ VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_SIZE_MAX, size_max, blkcfg); VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_SEG_MAX, seg_max, blkcfg); VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_GEOMETRY, geometry, blkcfg); VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_BLK_SIZE, blk_size, blkcfg); VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_TOPOLOGY, topology, blkcfg); VTBLK_GET_CONFIG(dev, VIRTIO_BLK_F_CONFIG_WCE, writeback, blkcfg); } #undef VTBLK_GET_CONFIG static void vtblk_ident(struct vtblk_softc *sc) { struct bio buf; struct disk *dp; struct vtblk_request *req; int len, error; dp = sc->vtblk_disk; len = MIN(VIRTIO_BLK_ID_BYTES, DISK_IDENT_SIZE); if (vtblk_tunable_int(sc, "no_ident", vtblk_no_ident) != 0) return; req = vtblk_request_dequeue(sc); if (req == NULL) return; req->vbr_ack = -1; req->vbr_hdr.type = VIRTIO_BLK_T_GET_ID; req->vbr_hdr.ioprio = 1; req->vbr_hdr.sector = 0; req->vbr_bp = &buf; g_reset_bio(&buf); buf.bio_cmd = BIO_READ; buf.bio_data = dp->d_ident; buf.bio_bcount = len; VTBLK_LOCK(sc); error = vtblk_poll_request(sc, req); VTBLK_UNLOCK(sc); vtblk_request_enqueue(sc, req); if (error) { device_printf(sc->vtblk_dev, "error getting device identifier: %d\n", error); } } static int vtblk_poll_request(struct vtblk_softc *sc, struct vtblk_request *req) { struct virtqueue *vq; int error; vq = sc->vtblk_vq; if (!virtqueue_empty(vq)) return (EBUSY); error = vtblk_request_execute(sc, req); if (error) return (error); virtqueue_notify(vq); virtqueue_poll(vq, NULL); error = vtblk_request_error(req); if (error && bootverbose) { device_printf(sc->vtblk_dev, "%s: IO error: %d\n", __func__, error); } return (error); } static int vtblk_quiesce(struct vtblk_softc *sc) { int error; VTBLK_LOCK_ASSERT(sc); error = 0; while (!virtqueue_empty(sc->vtblk_vq)) { if (mtx_sleep(&sc->vtblk_vq, VTBLK_MTX(sc), PRIBIO, "vtblkq", VTBLK_QUIESCE_TIMEOUT) == EWOULDBLOCK) { error = EBUSY; break; } } return (error); } static void vtblk_vq_intr(void *xsc) { struct vtblk_softc *sc; struct virtqueue *vq; struct bio_queue queue; sc = xsc; vq = sc->vtblk_vq; TAILQ_INIT(&queue); VTBLK_LOCK(sc); again: if (sc->vtblk_flags & VTBLK_FLAG_DETACH) goto out; vtblk_queue_completed(sc, &queue); vtblk_startio(sc); if (virtqueue_enable_intr(vq) != 0) { virtqueue_disable_intr(vq); goto again; } if (sc->vtblk_flags & VTBLK_FLAG_SUSPEND) wakeup(&sc->vtblk_vq); out: VTBLK_UNLOCK(sc); vtblk_done_completed(sc, &queue); } static void vtblk_stop(struct vtblk_softc *sc) { virtqueue_disable_intr(sc->vtblk_vq); virtio_stop(sc->vtblk_dev); } static void vtblk_dump_quiesce(struct vtblk_softc *sc) { /* * Spin here until all the requests in-flight at the time of the * dump are completed and queued. The queued requests will be * biodone'd once the dump is finished. */ while (!virtqueue_empty(sc->vtblk_vq)) vtblk_queue_completed(sc, &sc->vtblk_dump_queue); } static int vtblk_dump_write(struct vtblk_softc *sc, void *virtual, off_t offset, size_t length) { struct bio buf; struct vtblk_request *req; req = &sc->vtblk_dump_request; req->vbr_ack = -1; req->vbr_hdr.type = VIRTIO_BLK_T_OUT; req->vbr_hdr.ioprio = 1; req->vbr_hdr.sector = offset / 512; req->vbr_bp = &buf; g_reset_bio(&buf); buf.bio_cmd = BIO_WRITE; buf.bio_data = virtual; buf.bio_bcount = length; return (vtblk_poll_request(sc, req)); } static int vtblk_dump_flush(struct vtblk_softc *sc) { struct bio buf; struct vtblk_request *req; req = &sc->vtblk_dump_request; req->vbr_ack = -1; req->vbr_hdr.type = VIRTIO_BLK_T_FLUSH; req->vbr_hdr.ioprio = 1; req->vbr_hdr.sector = 0; req->vbr_bp = &buf; g_reset_bio(&buf); buf.bio_cmd = BIO_FLUSH; return (vtblk_poll_request(sc, req)); } static void vtblk_dump_complete(struct vtblk_softc *sc) { vtblk_dump_flush(sc); VTBLK_UNLOCK(sc); vtblk_done_completed(sc, &sc->vtblk_dump_queue); VTBLK_LOCK(sc); } static void vtblk_set_write_cache(struct vtblk_softc *sc, int wc) { /* Set either writeback (1) or writethrough (0) mode. */ virtio_write_dev_config_1(sc->vtblk_dev, offsetof(struct virtio_blk_config, writeback), wc); } static int vtblk_write_cache_enabled(struct vtblk_softc *sc, struct virtio_blk_config *blkcfg) { int wc; if (sc->vtblk_flags & VTBLK_FLAG_WC_CONFIG) { wc = vtblk_tunable_int(sc, "writecache_mode", vtblk_writecache_mode); if (wc >= 0 && wc < VTBLK_CACHE_MAX) vtblk_set_write_cache(sc, wc); else wc = blkcfg->writeback; } else wc = virtio_with_feature(sc->vtblk_dev, VIRTIO_BLK_F_WCE); return (wc); } static int vtblk_write_cache_sysctl(SYSCTL_HANDLER_ARGS) { struct vtblk_softc *sc; int wc, error; sc = oidp->oid_arg1; wc = sc->vtblk_write_cache; error = sysctl_handle_int(oidp, &wc, 0, req); if (error || req->newptr == NULL) return (error); if ((sc->vtblk_flags & VTBLK_FLAG_WC_CONFIG) == 0) return (EPERM); if (wc < 0 || wc >= VTBLK_CACHE_MAX) return (EINVAL); VTBLK_LOCK(sc); sc->vtblk_write_cache = wc; vtblk_set_write_cache(sc, sc->vtblk_write_cache); VTBLK_UNLOCK(sc); return (0); } static void vtblk_setup_sysctl(struct vtblk_softc *sc) { device_t dev; struct sysctl_ctx_list *ctx; struct sysctl_oid *tree; struct sysctl_oid_list *child; dev = sc->vtblk_dev; ctx = device_get_sysctl_ctx(dev); tree = device_get_sysctl_tree(dev); child = SYSCTL_CHILDREN(tree); SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "writecache_mode", CTLTYPE_INT | CTLFLAG_RW, sc, 0, vtblk_write_cache_sysctl, "I", "Write cache mode (writethrough (0) or writeback (1))"); } static int vtblk_tunable_int(struct vtblk_softc *sc, const char *knob, int def) { char path[64]; snprintf(path, sizeof(path), "hw.vtblk.%d.%s", device_get_unit(sc->vtblk_dev), knob); TUNABLE_INT_FETCH(path, &def); return (def); } Index: head/sys/dev/virtio/console/virtio_console.c =================================================================== --- head/sys/dev/virtio/console/virtio_console.c (revision 348598) +++ head/sys/dev/virtio/console/virtio_console.c (revision 348599) @@ -1,1502 +1,1500 @@ /*- * Copyright (c) 2014, Bryan Venteicher * 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. */ /* Driver for VirtIO console devices. */ #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 "virtio_if.h" #define VTCON_MAX_PORTS 32 #define VTCON_TTY_PREFIX "V" #define VTCON_TTY_ALIAS_PREFIX "vtcon" #define VTCON_BULK_BUFSZ 128 #define VTCON_CTRL_BUFSZ 128 /* * The buffers cannot cross more than one page boundary due to the * size of the sglist segment array used. */ CTASSERT(VTCON_BULK_BUFSZ <= PAGE_SIZE); CTASSERT(VTCON_CTRL_BUFSZ <= PAGE_SIZE); CTASSERT(sizeof(struct virtio_console_config) <= VTCON_CTRL_BUFSZ); struct vtcon_softc; struct vtcon_softc_port; struct vtcon_port { struct mtx vtcport_mtx; struct vtcon_softc *vtcport_sc; struct vtcon_softc_port *vtcport_scport; struct tty *vtcport_tty; struct virtqueue *vtcport_invq; struct virtqueue *vtcport_outvq; int vtcport_id; int vtcport_flags; #define VTCON_PORT_FLAG_GONE 0x01 #define VTCON_PORT_FLAG_CONSOLE 0x02 #define VTCON_PORT_FLAG_ALIAS 0x04 #if defined(KDB) int vtcport_alt_break_state; #endif }; #define VTCON_PORT_LOCK(_port) mtx_lock(&(_port)->vtcport_mtx) #define VTCON_PORT_UNLOCK(_port) mtx_unlock(&(_port)->vtcport_mtx) struct vtcon_softc_port { struct vtcon_softc *vcsp_sc; struct vtcon_port *vcsp_port; struct virtqueue *vcsp_invq; struct virtqueue *vcsp_outvq; }; struct vtcon_softc { device_t vtcon_dev; struct mtx vtcon_mtx; uint64_t vtcon_features; uint32_t vtcon_max_ports; uint32_t vtcon_flags; #define VTCON_FLAG_DETACHED 0x01 #define VTCON_FLAG_SIZE 0x02 #define VTCON_FLAG_MULTIPORT 0x04 /* * Ports can be added and removed during runtime, but we have * to allocate all the virtqueues during attach. This array is * indexed by the port ID. */ struct vtcon_softc_port *vtcon_ports; struct task vtcon_ctrl_task; struct virtqueue *vtcon_ctrl_rxvq; struct virtqueue *vtcon_ctrl_txvq; struct mtx vtcon_ctrl_tx_mtx; }; #define VTCON_LOCK(_sc) mtx_lock(&(_sc)->vtcon_mtx) #define VTCON_UNLOCK(_sc) mtx_unlock(&(_sc)->vtcon_mtx) #define VTCON_LOCK_ASSERT(_sc) \ mtx_assert(&(_sc)->vtcon_mtx, MA_OWNED) #define VTCON_LOCK_ASSERT_NOTOWNED(_sc) \ mtx_assert(&(_sc)->vtcon_mtx, MA_NOTOWNED) #define VTCON_CTRL_TX_LOCK(_sc) mtx_lock(&(_sc)->vtcon_ctrl_tx_mtx) #define VTCON_CTRL_TX_UNLOCK(_sc) mtx_unlock(&(_sc)->vtcon_ctrl_tx_mtx) #define VTCON_ASSERT_VALID_PORTID(_sc, _id) \ KASSERT((_id) >= 0 && (_id) < (_sc)->vtcon_max_ports, \ ("%s: port ID %d out of range", __func__, _id)) #define VTCON_FEATURES VIRTIO_CONSOLE_F_MULTIPORT static struct virtio_feature_desc vtcon_feature_desc[] = { { VIRTIO_CONSOLE_F_SIZE, "ConsoleSize" }, { VIRTIO_CONSOLE_F_MULTIPORT, "MultiplePorts" }, { VIRTIO_CONSOLE_F_EMERG_WRITE, "EmergencyWrite" }, { 0, NULL } }; static int vtcon_modevent(module_t, int, void *); static void vtcon_drain_all(void); static int vtcon_probe(device_t); static int vtcon_attach(device_t); static int vtcon_detach(device_t); static int vtcon_config_change(device_t); static void vtcon_setup_features(struct vtcon_softc *); static void vtcon_negotiate_features(struct vtcon_softc *); static int vtcon_alloc_scports(struct vtcon_softc *); static int vtcon_alloc_virtqueues(struct vtcon_softc *); static void vtcon_read_config(struct vtcon_softc *, struct virtio_console_config *); static void vtcon_determine_max_ports(struct vtcon_softc *, struct virtio_console_config *); static void vtcon_destroy_ports(struct vtcon_softc *); static void vtcon_stop(struct vtcon_softc *); static int vtcon_ctrl_event_enqueue(struct vtcon_softc *, struct virtio_console_control *); static int vtcon_ctrl_event_create(struct vtcon_softc *); static void vtcon_ctrl_event_requeue(struct vtcon_softc *, struct virtio_console_control *); static int vtcon_ctrl_event_populate(struct vtcon_softc *); static void vtcon_ctrl_event_drain(struct vtcon_softc *); static int vtcon_ctrl_init(struct vtcon_softc *); static void vtcon_ctrl_deinit(struct vtcon_softc *); static void vtcon_ctrl_port_add_event(struct vtcon_softc *, int); static void vtcon_ctrl_port_remove_event(struct vtcon_softc *, int); static void vtcon_ctrl_port_console_event(struct vtcon_softc *, int); static void vtcon_ctrl_port_open_event(struct vtcon_softc *, int); static void vtcon_ctrl_port_name_event(struct vtcon_softc *, int, const char *, size_t); static void vtcon_ctrl_process_event(struct vtcon_softc *, struct virtio_console_control *, void *, size_t); static void vtcon_ctrl_task_cb(void *, int); static void vtcon_ctrl_event_intr(void *); static void vtcon_ctrl_poll(struct vtcon_softc *, struct virtio_console_control *control); static void vtcon_ctrl_send_control(struct vtcon_softc *, uint32_t, uint16_t, uint16_t); static int vtcon_port_enqueue_buf(struct vtcon_port *, void *, size_t); static int vtcon_port_create_buf(struct vtcon_port *); static void vtcon_port_requeue_buf(struct vtcon_port *, void *); static int vtcon_port_populate(struct vtcon_port *); static void vtcon_port_destroy(struct vtcon_port *); static int vtcon_port_create(struct vtcon_softc *, int); static void vtcon_port_dev_alias(struct vtcon_port *, const char *, size_t); static void vtcon_port_drain_bufs(struct virtqueue *); static void vtcon_port_drain(struct vtcon_port *); static void vtcon_port_teardown(struct vtcon_port *); static void vtcon_port_change_size(struct vtcon_port *, uint16_t, uint16_t); static void vtcon_port_update_console_size(struct vtcon_softc *); static void vtcon_port_enable_intr(struct vtcon_port *); static void vtcon_port_disable_intr(struct vtcon_port *); static void vtcon_port_in(struct vtcon_port *); static void vtcon_port_intr(void *); static void vtcon_port_out(struct vtcon_port *, void *, int); static void vtcon_port_submit_event(struct vtcon_port *, uint16_t, uint16_t); static int vtcon_tty_open(struct tty *); static void vtcon_tty_close(struct tty *); static void vtcon_tty_outwakeup(struct tty *); static void vtcon_tty_free(void *); static void vtcon_get_console_size(struct vtcon_softc *, uint16_t *, uint16_t *); static void vtcon_enable_interrupts(struct vtcon_softc *); static void vtcon_disable_interrupts(struct vtcon_softc *); static int vtcon_pending_free; static struct ttydevsw vtcon_tty_class = { .tsw_flags = 0, .tsw_open = vtcon_tty_open, .tsw_close = vtcon_tty_close, .tsw_outwakeup = vtcon_tty_outwakeup, .tsw_free = vtcon_tty_free, }; static device_method_t vtcon_methods[] = { /* Device methods. */ DEVMETHOD(device_probe, vtcon_probe), DEVMETHOD(device_attach, vtcon_attach), DEVMETHOD(device_detach, vtcon_detach), /* VirtIO methods. */ DEVMETHOD(virtio_config_change, vtcon_config_change), DEVMETHOD_END }; static driver_t vtcon_driver = { "vtcon", vtcon_methods, sizeof(struct vtcon_softc) }; static devclass_t vtcon_devclass; DRIVER_MODULE(virtio_console, virtio_pci, vtcon_driver, vtcon_devclass, vtcon_modevent, 0); MODULE_VERSION(virtio_console, 1); MODULE_DEPEND(virtio_console, virtio, 1, 1, 1); +VIRTIO_SIMPLE_PNPTABLE(virtio_console, VIRTIO_ID_CONSOLE, + "VirtIO Console Adapter"); +VIRTIO_SIMPLE_PNPINFO(virtio_pci, virtio_console); + static int vtcon_modevent(module_t mod, int type, void *unused) { int error; switch (type) { case MOD_LOAD: error = 0; break; case MOD_QUIESCE: error = 0; break; case MOD_UNLOAD: vtcon_drain_all(); error = 0; break; case MOD_SHUTDOWN: error = 0; break; default: error = EOPNOTSUPP; break; } return (error); } static void vtcon_drain_all(void) { int first; for (first = 1; vtcon_pending_free != 0; first = 0) { if (first != 0) { printf("virtio_console: Waiting for all detached TTY " "devices to have open fds closed.\n"); } pause("vtcondra", hz); } } static int vtcon_probe(device_t dev) { - - if (virtio_get_device_type(dev) != VIRTIO_ID_CONSOLE) - return (ENXIO); - - device_set_desc(dev, "VirtIO Console Adapter"); - - return (BUS_PROBE_DEFAULT); + return (VIRTIO_SIMPLE_PROBE(dev, virtio_console)); } static int vtcon_attach(device_t dev) { struct vtcon_softc *sc; struct virtio_console_config concfg; int error; sc = device_get_softc(dev); sc->vtcon_dev = dev; mtx_init(&sc->vtcon_mtx, "vtconmtx", NULL, MTX_DEF); mtx_init(&sc->vtcon_ctrl_tx_mtx, "vtconctrlmtx", NULL, MTX_DEF); virtio_set_feature_desc(dev, vtcon_feature_desc); vtcon_setup_features(sc); vtcon_read_config(sc, &concfg); vtcon_determine_max_ports(sc, &concfg); error = vtcon_alloc_scports(sc); if (error) { device_printf(dev, "cannot allocate softc port structures\n"); goto fail; } error = vtcon_alloc_virtqueues(sc); if (error) { device_printf(dev, "cannot allocate virtqueues\n"); goto fail; } if (sc->vtcon_flags & VTCON_FLAG_MULTIPORT) { TASK_INIT(&sc->vtcon_ctrl_task, 0, vtcon_ctrl_task_cb, sc); error = vtcon_ctrl_init(sc); if (error) goto fail; } else { error = vtcon_port_create(sc, 0); if (error) goto fail; if (sc->vtcon_flags & VTCON_FLAG_SIZE) vtcon_port_update_console_size(sc); } error = virtio_setup_intr(dev, INTR_TYPE_TTY); if (error) { device_printf(dev, "cannot setup virtqueue interrupts\n"); goto fail; } vtcon_enable_interrupts(sc); vtcon_ctrl_send_control(sc, VIRTIO_CONSOLE_BAD_ID, VIRTIO_CONSOLE_DEVICE_READY, 1); fail: if (error) vtcon_detach(dev); return (error); } static int vtcon_detach(device_t dev) { struct vtcon_softc *sc; sc = device_get_softc(dev); VTCON_LOCK(sc); sc->vtcon_flags |= VTCON_FLAG_DETACHED; if (device_is_attached(dev)) vtcon_stop(sc); VTCON_UNLOCK(sc); if (sc->vtcon_flags & VTCON_FLAG_MULTIPORT) { taskqueue_drain(taskqueue_thread, &sc->vtcon_ctrl_task); vtcon_ctrl_deinit(sc); } vtcon_destroy_ports(sc); mtx_destroy(&sc->vtcon_mtx); mtx_destroy(&sc->vtcon_ctrl_tx_mtx); return (0); } static int vtcon_config_change(device_t dev) { struct vtcon_softc *sc; sc = device_get_softc(dev); /* * When the multiport feature is negotiated, all configuration * changes are done through control virtqueue events. */ if ((sc->vtcon_flags & VTCON_FLAG_MULTIPORT) == 0) { if (sc->vtcon_flags & VTCON_FLAG_SIZE) vtcon_port_update_console_size(sc); } return (0); } static void vtcon_negotiate_features(struct vtcon_softc *sc) { device_t dev; uint64_t features; dev = sc->vtcon_dev; features = VTCON_FEATURES; sc->vtcon_features = virtio_negotiate_features(dev, features); } static void vtcon_setup_features(struct vtcon_softc *sc) { device_t dev; dev = sc->vtcon_dev; vtcon_negotiate_features(sc); if (virtio_with_feature(dev, VIRTIO_CONSOLE_F_SIZE)) sc->vtcon_flags |= VTCON_FLAG_SIZE; if (virtio_with_feature(dev, VIRTIO_CONSOLE_F_MULTIPORT)) sc->vtcon_flags |= VTCON_FLAG_MULTIPORT; } #define VTCON_GET_CONFIG(_dev, _feature, _field, _cfg) \ if (virtio_with_feature(_dev, _feature)) { \ virtio_read_device_config(_dev, \ offsetof(struct virtio_console_config, _field), \ &(_cfg)->_field, sizeof((_cfg)->_field)); \ } static void vtcon_read_config(struct vtcon_softc *sc, struct virtio_console_config *concfg) { device_t dev; dev = sc->vtcon_dev; bzero(concfg, sizeof(struct virtio_console_config)); VTCON_GET_CONFIG(dev, VIRTIO_CONSOLE_F_SIZE, cols, concfg); VTCON_GET_CONFIG(dev, VIRTIO_CONSOLE_F_SIZE, rows, concfg); VTCON_GET_CONFIG(dev, VIRTIO_CONSOLE_F_MULTIPORT, max_nr_ports, concfg); } #undef VTCON_GET_CONFIG static int vtcon_alloc_scports(struct vtcon_softc *sc) { struct vtcon_softc_port *scport; int max, i; max = sc->vtcon_max_ports; sc->vtcon_ports = malloc(sizeof(struct vtcon_softc_port) * max, M_DEVBUF, M_NOWAIT | M_ZERO); if (sc->vtcon_ports == NULL) return (ENOMEM); for (i = 0; i < max; i++) { scport = &sc->vtcon_ports[i]; scport->vcsp_sc = sc; } return (0); } static int vtcon_alloc_virtqueues(struct vtcon_softc *sc) { device_t dev; struct vq_alloc_info *info; struct vtcon_softc_port *scport; int i, idx, portidx, nvqs, error; dev = sc->vtcon_dev; nvqs = sc->vtcon_max_ports * 2; if (sc->vtcon_flags & VTCON_FLAG_MULTIPORT) nvqs += 2; info = malloc(sizeof(struct vq_alloc_info) * nvqs, M_TEMP, M_NOWAIT); if (info == NULL) return (ENOMEM); for (i = 0, idx = 0, portidx = 0; i < nvqs / 2; i++, idx += 2) { if (i == 1) { /* The control virtqueues are after the first port. */ VQ_ALLOC_INFO_INIT(&info[idx], 0, vtcon_ctrl_event_intr, sc, &sc->vtcon_ctrl_rxvq, "%s-control rx", device_get_nameunit(dev)); VQ_ALLOC_INFO_INIT(&info[idx+1], 0, NULL, sc, &sc->vtcon_ctrl_txvq, "%s-control tx", device_get_nameunit(dev)); continue; } scport = &sc->vtcon_ports[portidx]; VQ_ALLOC_INFO_INIT(&info[idx], 0, vtcon_port_intr, scport, &scport->vcsp_invq, "%s-port%d in", device_get_nameunit(dev), i); VQ_ALLOC_INFO_INIT(&info[idx+1], 0, NULL, NULL, &scport->vcsp_outvq, "%s-port%d out", device_get_nameunit(dev), i); portidx++; } error = virtio_alloc_virtqueues(dev, 0, nvqs, info); free(info, M_TEMP); return (error); } static void vtcon_determine_max_ports(struct vtcon_softc *sc, struct virtio_console_config *concfg) { if (sc->vtcon_flags & VTCON_FLAG_MULTIPORT) { sc->vtcon_max_ports = min(concfg->max_nr_ports, VTCON_MAX_PORTS); if (sc->vtcon_max_ports == 0) sc->vtcon_max_ports = 1; } else sc->vtcon_max_ports = 1; } static void vtcon_destroy_ports(struct vtcon_softc *sc) { struct vtcon_softc_port *scport; struct vtcon_port *port; struct virtqueue *vq; int i; if (sc->vtcon_ports == NULL) return; VTCON_LOCK(sc); for (i = 0; i < sc->vtcon_max_ports; i++) { scport = &sc->vtcon_ports[i]; port = scport->vcsp_port; if (port != NULL) { scport->vcsp_port = NULL; VTCON_PORT_LOCK(port); VTCON_UNLOCK(sc); vtcon_port_teardown(port); VTCON_LOCK(sc); } vq = scport->vcsp_invq; if (vq != NULL) vtcon_port_drain_bufs(vq); } VTCON_UNLOCK(sc); free(sc->vtcon_ports, M_DEVBUF); sc->vtcon_ports = NULL; } static void vtcon_stop(struct vtcon_softc *sc) { vtcon_disable_interrupts(sc); virtio_stop(sc->vtcon_dev); } static int vtcon_ctrl_event_enqueue(struct vtcon_softc *sc, struct virtio_console_control *control) { struct sglist_seg segs[2]; struct sglist sg; struct virtqueue *vq; int error; vq = sc->vtcon_ctrl_rxvq; sglist_init(&sg, 2, segs); error = sglist_append(&sg, control, VTCON_CTRL_BUFSZ); KASSERT(error == 0, ("%s: error %d adding control to sglist", __func__, error)); return (virtqueue_enqueue(vq, control, &sg, 0, sg.sg_nseg)); } static int vtcon_ctrl_event_create(struct vtcon_softc *sc) { struct virtio_console_control *control; int error; control = malloc(VTCON_CTRL_BUFSZ, M_DEVBUF, M_ZERO | M_NOWAIT); if (control == NULL) return (ENOMEM); error = vtcon_ctrl_event_enqueue(sc, control); if (error) free(control, M_DEVBUF); return (error); } static void vtcon_ctrl_event_requeue(struct vtcon_softc *sc, struct virtio_console_control *control) { int error; bzero(control, VTCON_CTRL_BUFSZ); error = vtcon_ctrl_event_enqueue(sc, control); KASSERT(error == 0, ("%s: cannot requeue control buffer %d", __func__, error)); } static int vtcon_ctrl_event_populate(struct vtcon_softc *sc) { struct virtqueue *vq; int nbufs, error; vq = sc->vtcon_ctrl_rxvq; error = ENOSPC; for (nbufs = 0; !virtqueue_full(vq); nbufs++) { error = vtcon_ctrl_event_create(sc); if (error) break; } if (nbufs > 0) { virtqueue_notify(vq); error = 0; } return (error); } static void vtcon_ctrl_event_drain(struct vtcon_softc *sc) { struct virtio_console_control *control; struct virtqueue *vq; int last; vq = sc->vtcon_ctrl_rxvq; last = 0; if (vq == NULL) return; VTCON_LOCK(sc); while ((control = virtqueue_drain(vq, &last)) != NULL) free(control, M_DEVBUF); VTCON_UNLOCK(sc); } static int vtcon_ctrl_init(struct vtcon_softc *sc) { int error; error = vtcon_ctrl_event_populate(sc); return (error); } static void vtcon_ctrl_deinit(struct vtcon_softc *sc) { vtcon_ctrl_event_drain(sc); } static void vtcon_ctrl_port_add_event(struct vtcon_softc *sc, int id) { device_t dev; int error; dev = sc->vtcon_dev; /* This single thread only way for ports to be created. */ if (sc->vtcon_ports[id].vcsp_port != NULL) { device_printf(dev, "%s: adding port %d, but already exists\n", __func__, id); return; } error = vtcon_port_create(sc, id); if (error) { device_printf(dev, "%s: cannot create port %d: %d\n", __func__, id, error); vtcon_ctrl_send_control(sc, id, VIRTIO_CONSOLE_PORT_READY, 0); return; } } static void vtcon_ctrl_port_remove_event(struct vtcon_softc *sc, int id) { device_t dev; struct vtcon_softc_port *scport; struct vtcon_port *port; dev = sc->vtcon_dev; scport = &sc->vtcon_ports[id]; VTCON_LOCK(sc); port = scport->vcsp_port; if (port == NULL) { VTCON_UNLOCK(sc); device_printf(dev, "%s: remove port %d, but does not exist\n", __func__, id); return; } scport->vcsp_port = NULL; VTCON_PORT_LOCK(port); VTCON_UNLOCK(sc); vtcon_port_teardown(port); } static void vtcon_ctrl_port_console_event(struct vtcon_softc *sc, int id) { device_t dev; struct vtcon_softc_port *scport; struct vtcon_port *port; dev = sc->vtcon_dev; scport = &sc->vtcon_ports[id]; VTCON_LOCK(sc); port = scport->vcsp_port; if (port == NULL) { VTCON_UNLOCK(sc); device_printf(dev, "%s: console port %d, but does not exist\n", __func__, id); return; } VTCON_PORT_LOCK(port); VTCON_UNLOCK(sc); port->vtcport_flags |= VTCON_PORT_FLAG_CONSOLE; vtcon_port_submit_event(port, VIRTIO_CONSOLE_PORT_OPEN, 1); VTCON_PORT_UNLOCK(port); } static void vtcon_ctrl_port_open_event(struct vtcon_softc *sc, int id) { device_t dev; struct vtcon_softc_port *scport; struct vtcon_port *port; dev = sc->vtcon_dev; scport = &sc->vtcon_ports[id]; VTCON_LOCK(sc); port = scport->vcsp_port; if (port == NULL) { VTCON_UNLOCK(sc); device_printf(dev, "%s: open port %d, but does not exist\n", __func__, id); return; } VTCON_PORT_LOCK(port); VTCON_UNLOCK(sc); vtcon_port_enable_intr(port); VTCON_PORT_UNLOCK(port); } static void vtcon_ctrl_port_name_event(struct vtcon_softc *sc, int id, const char *name, size_t len) { device_t dev; struct vtcon_softc_port *scport; struct vtcon_port *port; dev = sc->vtcon_dev; scport = &sc->vtcon_ports[id]; /* * The VirtIO specification says the NUL terminator is not included in * the length, but QEMU includes it. Adjust the length if needed. */ if (name == NULL || len == 0) return; if (name[len - 1] == '\0') { len--; if (len == 0) return; } VTCON_LOCK(sc); port = scport->vcsp_port; if (port == NULL) { VTCON_UNLOCK(sc); device_printf(dev, "%s: name port %d, but does not exist\n", __func__, id); return; } VTCON_PORT_LOCK(port); VTCON_UNLOCK(sc); vtcon_port_dev_alias(port, name, len); VTCON_PORT_UNLOCK(port); } static void vtcon_ctrl_process_event(struct vtcon_softc *sc, struct virtio_console_control *control, void *data, size_t data_len) { device_t dev; int id; dev = sc->vtcon_dev; id = control->id; if (id < 0 || id >= sc->vtcon_max_ports) { device_printf(dev, "%s: invalid port ID %d\n", __func__, id); return; } switch (control->event) { case VIRTIO_CONSOLE_PORT_ADD: vtcon_ctrl_port_add_event(sc, id); break; case VIRTIO_CONSOLE_PORT_REMOVE: vtcon_ctrl_port_remove_event(sc, id); break; case VIRTIO_CONSOLE_CONSOLE_PORT: vtcon_ctrl_port_console_event(sc, id); break; case VIRTIO_CONSOLE_RESIZE: break; case VIRTIO_CONSOLE_PORT_OPEN: vtcon_ctrl_port_open_event(sc, id); break; case VIRTIO_CONSOLE_PORT_NAME: vtcon_ctrl_port_name_event(sc, id, (const char *)data, data_len); break; } } static void vtcon_ctrl_task_cb(void *xsc, int pending) { struct vtcon_softc *sc; struct virtqueue *vq; struct virtio_console_control *control; void *data; size_t data_len; int detached; uint32_t len; sc = xsc; vq = sc->vtcon_ctrl_rxvq; VTCON_LOCK(sc); while ((detached = (sc->vtcon_flags & VTCON_FLAG_DETACHED)) == 0) { control = virtqueue_dequeue(vq, &len); if (control == NULL) break; if (len > sizeof(struct virtio_console_control)) { data = (void *) &control[1]; data_len = len - sizeof(struct virtio_console_control); } else { data = NULL; data_len = 0; } VTCON_UNLOCK(sc); vtcon_ctrl_process_event(sc, control, data, data_len); VTCON_LOCK(sc); vtcon_ctrl_event_requeue(sc, control); } if (!detached) { virtqueue_notify(vq); if (virtqueue_enable_intr(vq) != 0) taskqueue_enqueue(taskqueue_thread, &sc->vtcon_ctrl_task); } VTCON_UNLOCK(sc); } static void vtcon_ctrl_event_intr(void *xsc) { struct vtcon_softc *sc; sc = xsc; /* * Only some events require us to potentially block, but it * easier to just defer all event handling to the taskqueue. */ taskqueue_enqueue(taskqueue_thread, &sc->vtcon_ctrl_task); } static void vtcon_ctrl_poll(struct vtcon_softc *sc, struct virtio_console_control *control) { struct sglist_seg segs[2]; struct sglist sg; struct virtqueue *vq; int error; vq = sc->vtcon_ctrl_txvq; sglist_init(&sg, 2, segs); error = sglist_append(&sg, control, sizeof(struct virtio_console_control)); KASSERT(error == 0, ("%s: error %d adding control to sglist", __func__, error)); /* * We cannot use the softc lock to serialize access to this * virtqueue since this is called from the tty layer with the * port lock held. Acquiring the softc would violate our lock * ordering. */ VTCON_CTRL_TX_LOCK(sc); KASSERT(virtqueue_empty(vq), ("%s: virtqueue is not emtpy", __func__)); error = virtqueue_enqueue(vq, control, &sg, sg.sg_nseg, 0); if (error == 0) { virtqueue_notify(vq); virtqueue_poll(vq, NULL); } VTCON_CTRL_TX_UNLOCK(sc); } static void vtcon_ctrl_send_control(struct vtcon_softc *sc, uint32_t portid, uint16_t event, uint16_t value) { struct virtio_console_control control; if ((sc->vtcon_flags & VTCON_FLAG_MULTIPORT) == 0) return; control.id = portid; control.event = event; control.value = value; vtcon_ctrl_poll(sc, &control); } static int vtcon_port_enqueue_buf(struct vtcon_port *port, void *buf, size_t len) { struct sglist_seg segs[2]; struct sglist sg; struct virtqueue *vq; int error; vq = port->vtcport_invq; sglist_init(&sg, 2, segs); error = sglist_append(&sg, buf, len); KASSERT(error == 0, ("%s: error %d adding buffer to sglist", __func__, error)); error = virtqueue_enqueue(vq, buf, &sg, 0, sg.sg_nseg); return (error); } static int vtcon_port_create_buf(struct vtcon_port *port) { void *buf; int error; buf = malloc(VTCON_BULK_BUFSZ, M_DEVBUF, M_ZERO | M_NOWAIT); if (buf == NULL) return (ENOMEM); error = vtcon_port_enqueue_buf(port, buf, VTCON_BULK_BUFSZ); if (error) free(buf, M_DEVBUF); return (error); } static void vtcon_port_requeue_buf(struct vtcon_port *port, void *buf) { int error; error = vtcon_port_enqueue_buf(port, buf, VTCON_BULK_BUFSZ); KASSERT(error == 0, ("%s: cannot requeue input buffer %d", __func__, error)); } static int vtcon_port_populate(struct vtcon_port *port) { struct virtqueue *vq; int nbufs, error; vq = port->vtcport_invq; error = ENOSPC; for (nbufs = 0; !virtqueue_full(vq); nbufs++) { error = vtcon_port_create_buf(port); if (error) break; } if (nbufs > 0) { virtqueue_notify(vq); error = 0; } return (error); } static void vtcon_port_destroy(struct vtcon_port *port) { port->vtcport_sc = NULL; port->vtcport_scport = NULL; port->vtcport_invq = NULL; port->vtcport_outvq = NULL; port->vtcport_id = -1; mtx_destroy(&port->vtcport_mtx); free(port, M_DEVBUF); } static int vtcon_port_init_vqs(struct vtcon_port *port) { struct vtcon_softc_port *scport; int error; scport = port->vtcport_scport; port->vtcport_invq = scport->vcsp_invq; port->vtcport_outvq = scport->vcsp_outvq; /* * Free any data left over from when this virtqueue was in use by a * prior port. We have not yet notified the host that the port is * ready, so assume nothing in the virtqueue can be for us. */ vtcon_port_drain(port); KASSERT(virtqueue_empty(port->vtcport_invq), ("%s: in virtqueue is not empty", __func__)); KASSERT(virtqueue_empty(port->vtcport_outvq), ("%s: out virtqueue is not empty", __func__)); error = vtcon_port_populate(port); if (error) return (error); return (0); } static int vtcon_port_create(struct vtcon_softc *sc, int id) { device_t dev; struct vtcon_softc_port *scport; struct vtcon_port *port; int error; dev = sc->vtcon_dev; scport = &sc->vtcon_ports[id]; VTCON_ASSERT_VALID_PORTID(sc, id); MPASS(scport->vcsp_port == NULL); port = malloc(sizeof(struct vtcon_port), M_DEVBUF, M_NOWAIT | M_ZERO); if (port == NULL) return (ENOMEM); port->vtcport_sc = sc; port->vtcport_scport = scport; port->vtcport_id = id; mtx_init(&port->vtcport_mtx, "vtcpmtx", NULL, MTX_DEF); port->vtcport_tty = tty_alloc_mutex(&vtcon_tty_class, port, &port->vtcport_mtx); error = vtcon_port_init_vqs(port); if (error) { VTCON_PORT_LOCK(port); vtcon_port_teardown(port); return (error); } VTCON_LOCK(sc); VTCON_PORT_LOCK(port); scport->vcsp_port = port; vtcon_port_enable_intr(port); vtcon_port_submit_event(port, VIRTIO_CONSOLE_PORT_READY, 1); VTCON_PORT_UNLOCK(port); VTCON_UNLOCK(sc); tty_makedev(port->vtcport_tty, NULL, "%s%r.%r", VTCON_TTY_PREFIX, device_get_unit(dev), id); return (0); } static void vtcon_port_dev_alias(struct vtcon_port *port, const char *name, size_t len) { struct vtcon_softc *sc; struct cdev *pdev; struct tty *tp; int i, error; sc = port->vtcport_sc; tp = port->vtcport_tty; if (port->vtcport_flags & VTCON_PORT_FLAG_ALIAS) return; /* Port name is UTF-8, but we can only handle ASCII. */ for (i = 0; i < len; i++) { if (!isascii(name[i])) return; } /* * Port name may not conform to the devfs requirements so we cannot use * tty_makealias() because the MAKEDEV_CHECKNAME flag must be specified. */ error = make_dev_alias_p(MAKEDEV_NOWAIT | MAKEDEV_CHECKNAME, &pdev, tp->t_dev, "%s/%*s", VTCON_TTY_ALIAS_PREFIX, (int)len, name); if (error) { device_printf(sc->vtcon_dev, "%s: cannot make dev alias (%s/%*s) error %d\n", __func__, VTCON_TTY_ALIAS_PREFIX, (int)len, name, error); } else port->vtcport_flags |= VTCON_PORT_FLAG_ALIAS; } static void vtcon_port_drain_bufs(struct virtqueue *vq) { void *buf; int last; last = 0; while ((buf = virtqueue_drain(vq, &last)) != NULL) free(buf, M_DEVBUF); } static void vtcon_port_drain(struct vtcon_port *port) { vtcon_port_drain_bufs(port->vtcport_invq); } static void vtcon_port_teardown(struct vtcon_port *port) { struct tty *tp; tp = port->vtcport_tty; port->vtcport_flags |= VTCON_PORT_FLAG_GONE; if (tp != NULL) { atomic_add_int(&vtcon_pending_free, 1); tty_rel_gone(tp); } else vtcon_port_destroy(port); } static void vtcon_port_change_size(struct vtcon_port *port, uint16_t cols, uint16_t rows) { struct tty *tp; struct winsize sz; tp = port->vtcport_tty; if (tp == NULL) return; bzero(&sz, sizeof(struct winsize)); sz.ws_col = cols; sz.ws_row = rows; tty_set_winsize(tp, &sz); } static void vtcon_port_update_console_size(struct vtcon_softc *sc) { struct vtcon_port *port; struct vtcon_softc_port *scport; uint16_t cols, rows; vtcon_get_console_size(sc, &cols, &rows); /* * For now, assume the first (only) port is the console. Note * QEMU does not implement this feature yet. */ scport = &sc->vtcon_ports[0]; VTCON_LOCK(sc); port = scport->vcsp_port; if (port != NULL) { VTCON_PORT_LOCK(port); VTCON_UNLOCK(sc); vtcon_port_change_size(port, cols, rows); VTCON_PORT_UNLOCK(port); } else VTCON_UNLOCK(sc); } static void vtcon_port_enable_intr(struct vtcon_port *port) { /* * NOTE: The out virtqueue is always polled, so its interrupt * kept disabled. */ virtqueue_enable_intr(port->vtcport_invq); } static void vtcon_port_disable_intr(struct vtcon_port *port) { if (port->vtcport_invq != NULL) virtqueue_disable_intr(port->vtcport_invq); if (port->vtcport_outvq != NULL) virtqueue_disable_intr(port->vtcport_outvq); } static void vtcon_port_in(struct vtcon_port *port) { struct virtqueue *vq; struct tty *tp; char *buf; uint32_t len; int i, deq; tp = port->vtcport_tty; vq = port->vtcport_invq; again: deq = 0; while ((buf = virtqueue_dequeue(vq, &len)) != NULL) { for (i = 0; i < len; i++) { #if defined(KDB) if (port->vtcport_flags & VTCON_PORT_FLAG_CONSOLE) kdb_alt_break(buf[i], &port->vtcport_alt_break_state); #endif ttydisc_rint(tp, buf[i], 0); } vtcon_port_requeue_buf(port, buf); deq++; } ttydisc_rint_done(tp); if (deq > 0) virtqueue_notify(vq); if (virtqueue_enable_intr(vq) != 0) goto again; } static void vtcon_port_intr(void *scportx) { struct vtcon_softc_port *scport; struct vtcon_softc *sc; struct vtcon_port *port; scport = scportx; sc = scport->vcsp_sc; VTCON_LOCK(sc); port = scport->vcsp_port; if (port == NULL) { VTCON_UNLOCK(sc); return; } VTCON_PORT_LOCK(port); VTCON_UNLOCK(sc); if ((port->vtcport_flags & VTCON_PORT_FLAG_GONE) == 0) vtcon_port_in(port); VTCON_PORT_UNLOCK(port); } static void vtcon_port_out(struct vtcon_port *port, void *buf, int bufsize) { struct sglist_seg segs[2]; struct sglist sg; struct virtqueue *vq; int error; vq = port->vtcport_outvq; KASSERT(virtqueue_empty(vq), ("%s: port %p out virtqueue not emtpy", __func__, port)); sglist_init(&sg, 2, segs); error = sglist_append(&sg, buf, bufsize); KASSERT(error == 0, ("%s: error %d adding buffer to sglist", __func__, error)); error = virtqueue_enqueue(vq, buf, &sg, sg.sg_nseg, 0); if (error == 0) { virtqueue_notify(vq); virtqueue_poll(vq, NULL); } } static void vtcon_port_submit_event(struct vtcon_port *port, uint16_t event, uint16_t value) { struct vtcon_softc *sc; sc = port->vtcport_sc; vtcon_ctrl_send_control(sc, port->vtcport_id, event, value); } static int vtcon_tty_open(struct tty *tp) { struct vtcon_port *port; port = tty_softc(tp); if (port->vtcport_flags & VTCON_PORT_FLAG_GONE) return (ENXIO); vtcon_port_submit_event(port, VIRTIO_CONSOLE_PORT_OPEN, 1); return (0); } static void vtcon_tty_close(struct tty *tp) { struct vtcon_port *port; port = tty_softc(tp); if (port->vtcport_flags & VTCON_PORT_FLAG_GONE) return; vtcon_port_submit_event(port, VIRTIO_CONSOLE_PORT_OPEN, 0); } static void vtcon_tty_outwakeup(struct tty *tp) { struct vtcon_port *port; char buf[VTCON_BULK_BUFSZ]; int len; port = tty_softc(tp); if (port->vtcport_flags & VTCON_PORT_FLAG_GONE) return; while ((len = ttydisc_getc(tp, buf, sizeof(buf))) != 0) vtcon_port_out(port, buf, len); } static void vtcon_tty_free(void *xport) { struct vtcon_port *port; port = xport; vtcon_port_destroy(port); atomic_subtract_int(&vtcon_pending_free, 1); } static void vtcon_get_console_size(struct vtcon_softc *sc, uint16_t *cols, uint16_t *rows) { struct virtio_console_config concfg; KASSERT(sc->vtcon_flags & VTCON_FLAG_SIZE, ("%s: size feature not negotiated", __func__)); vtcon_read_config(sc, &concfg); *cols = concfg.cols; *rows = concfg.rows; } static void vtcon_enable_interrupts(struct vtcon_softc *sc) { struct vtcon_softc_port *scport; struct vtcon_port *port; int i; VTCON_LOCK(sc); if (sc->vtcon_flags & VTCON_FLAG_MULTIPORT) virtqueue_enable_intr(sc->vtcon_ctrl_rxvq); for (i = 0; i < sc->vtcon_max_ports; i++) { scport = &sc->vtcon_ports[i]; port = scport->vcsp_port; if (port == NULL) continue; VTCON_PORT_LOCK(port); vtcon_port_enable_intr(port); VTCON_PORT_UNLOCK(port); } VTCON_UNLOCK(sc); } static void vtcon_disable_interrupts(struct vtcon_softc *sc) { struct vtcon_softc_port *scport; struct vtcon_port *port; int i; VTCON_LOCK_ASSERT(sc); if (sc->vtcon_flags & VTCON_FLAG_MULTIPORT) virtqueue_disable_intr(sc->vtcon_ctrl_rxvq); for (i = 0; i < sc->vtcon_max_ports; i++) { scport = &sc->vtcon_ports[i]; port = scport->vcsp_port; if (port == NULL) continue; VTCON_PORT_LOCK(port); vtcon_port_disable_intr(port); VTCON_PORT_UNLOCK(port); } } Index: head/sys/dev/virtio/network/if_vtnet.c =================================================================== --- head/sys/dev/virtio/network/if_vtnet.c (revision 348598) +++ head/sys/dev/virtio/network/if_vtnet.c (revision 348599) @@ -1,4041 +1,4039 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2011, Bryan Venteicher * 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. */ /* Driver for VirtIO network devices. */ #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 #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "virtio_if.h" #include "opt_inet.h" #include "opt_inet6.h" static int vtnet_modevent(module_t, int, void *); static int vtnet_probe(device_t); static int vtnet_attach(device_t); static int vtnet_detach(device_t); static int vtnet_suspend(device_t); static int vtnet_resume(device_t); static int vtnet_shutdown(device_t); static int vtnet_attach_completed(device_t); static int vtnet_config_change(device_t); static void vtnet_negotiate_features(struct vtnet_softc *); static void vtnet_setup_features(struct vtnet_softc *); static int vtnet_init_rxq(struct vtnet_softc *, int); static int vtnet_init_txq(struct vtnet_softc *, int); static int vtnet_alloc_rxtx_queues(struct vtnet_softc *); static void vtnet_free_rxtx_queues(struct vtnet_softc *); static int vtnet_alloc_rx_filters(struct vtnet_softc *); static void vtnet_free_rx_filters(struct vtnet_softc *); static int vtnet_alloc_virtqueues(struct vtnet_softc *); static int vtnet_setup_interface(struct vtnet_softc *); static int vtnet_change_mtu(struct vtnet_softc *, int); static int vtnet_ioctl(struct ifnet *, u_long, caddr_t); static uint64_t vtnet_get_counter(struct ifnet *, ift_counter); static int vtnet_rxq_populate(struct vtnet_rxq *); static void vtnet_rxq_free_mbufs(struct vtnet_rxq *); static struct mbuf * vtnet_rx_alloc_buf(struct vtnet_softc *, int , struct mbuf **); static int vtnet_rxq_replace_lro_nomgr_buf(struct vtnet_rxq *, struct mbuf *, int); static int vtnet_rxq_replace_buf(struct vtnet_rxq *, struct mbuf *, int); static int vtnet_rxq_enqueue_buf(struct vtnet_rxq *, struct mbuf *); static int vtnet_rxq_new_buf(struct vtnet_rxq *); static int vtnet_rxq_csum(struct vtnet_rxq *, struct mbuf *, struct virtio_net_hdr *); static void vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *, int); static void vtnet_rxq_discard_buf(struct vtnet_rxq *, struct mbuf *); static int vtnet_rxq_merged_eof(struct vtnet_rxq *, struct mbuf *, int); static void vtnet_rxq_input(struct vtnet_rxq *, struct mbuf *, struct virtio_net_hdr *); static int vtnet_rxq_eof(struct vtnet_rxq *); static void vtnet_rx_vq_intr(void *); static void vtnet_rxq_tq_intr(void *, int); static int vtnet_txq_below_threshold(struct vtnet_txq *); static int vtnet_txq_notify(struct vtnet_txq *); static void vtnet_txq_free_mbufs(struct vtnet_txq *); static int vtnet_txq_offload_ctx(struct vtnet_txq *, struct mbuf *, int *, int *, int *); static int vtnet_txq_offload_tso(struct vtnet_txq *, struct mbuf *, int, int, struct virtio_net_hdr *); static struct mbuf * vtnet_txq_offload(struct vtnet_txq *, struct mbuf *, struct virtio_net_hdr *); static int vtnet_txq_enqueue_buf(struct vtnet_txq *, struct mbuf **, struct vtnet_tx_header *); static int vtnet_txq_encap(struct vtnet_txq *, struct mbuf **, int); #ifdef VTNET_LEGACY_TX static void vtnet_start_locked(struct vtnet_txq *, struct ifnet *); static void vtnet_start(struct ifnet *); #else static int vtnet_txq_mq_start_locked(struct vtnet_txq *, struct mbuf *); static int vtnet_txq_mq_start(struct ifnet *, struct mbuf *); static void vtnet_txq_tq_deferred(void *, int); #endif static void vtnet_txq_start(struct vtnet_txq *); static void vtnet_txq_tq_intr(void *, int); static int vtnet_txq_eof(struct vtnet_txq *); static void vtnet_tx_vq_intr(void *); static void vtnet_tx_start_all(struct vtnet_softc *); #ifndef VTNET_LEGACY_TX static void vtnet_qflush(struct ifnet *); #endif static int vtnet_watchdog(struct vtnet_txq *); static void vtnet_accum_stats(struct vtnet_softc *, struct vtnet_rxq_stats *, struct vtnet_txq_stats *); static void vtnet_tick(void *); static void vtnet_start_taskqueues(struct vtnet_softc *); static void vtnet_free_taskqueues(struct vtnet_softc *); static void vtnet_drain_taskqueues(struct vtnet_softc *); static void vtnet_drain_rxtx_queues(struct vtnet_softc *); static void vtnet_stop_rendezvous(struct vtnet_softc *); static void vtnet_stop(struct vtnet_softc *); static int vtnet_virtio_reinit(struct vtnet_softc *); static void vtnet_init_rx_filters(struct vtnet_softc *); static int vtnet_init_rx_queues(struct vtnet_softc *); static int vtnet_init_tx_queues(struct vtnet_softc *); static int vtnet_init_rxtx_queues(struct vtnet_softc *); static void vtnet_set_active_vq_pairs(struct vtnet_softc *); static int vtnet_reinit(struct vtnet_softc *); static void vtnet_init_locked(struct vtnet_softc *); static void vtnet_init(void *); static void vtnet_free_ctrl_vq(struct vtnet_softc *); static void vtnet_exec_ctrl_cmd(struct vtnet_softc *, void *, struct sglist *, int, int); static int vtnet_ctrl_mac_cmd(struct vtnet_softc *, uint8_t *); static int vtnet_ctrl_mq_cmd(struct vtnet_softc *, uint16_t); static int vtnet_ctrl_rx_cmd(struct vtnet_softc *, int, int); static int vtnet_set_promisc(struct vtnet_softc *, int); static int vtnet_set_allmulti(struct vtnet_softc *, int); static void vtnet_attach_disable_promisc(struct vtnet_softc *); static void vtnet_rx_filter(struct vtnet_softc *); static void vtnet_rx_filter_mac(struct vtnet_softc *); static int vtnet_exec_vlan_filter(struct vtnet_softc *, int, uint16_t); static void vtnet_rx_filter_vlan(struct vtnet_softc *); static void vtnet_update_vlan_filter(struct vtnet_softc *, int, uint16_t); static void vtnet_register_vlan(void *, struct ifnet *, uint16_t); static void vtnet_unregister_vlan(void *, struct ifnet *, uint16_t); static int vtnet_is_link_up(struct vtnet_softc *); static void vtnet_update_link_status(struct vtnet_softc *); static int vtnet_ifmedia_upd(struct ifnet *); static void vtnet_ifmedia_sts(struct ifnet *, struct ifmediareq *); static void vtnet_get_hwaddr(struct vtnet_softc *); static void vtnet_set_hwaddr(struct vtnet_softc *); static void vtnet_vlan_tag_remove(struct mbuf *); static void vtnet_set_rx_process_limit(struct vtnet_softc *); static void vtnet_set_tx_intr_threshold(struct vtnet_softc *); static void vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *, struct sysctl_oid_list *, struct vtnet_rxq *); static void vtnet_setup_txq_sysctl(struct sysctl_ctx_list *, struct sysctl_oid_list *, struct vtnet_txq *); static void vtnet_setup_queue_sysctl(struct vtnet_softc *); static void vtnet_setup_sysctl(struct vtnet_softc *); static int vtnet_rxq_enable_intr(struct vtnet_rxq *); static void vtnet_rxq_disable_intr(struct vtnet_rxq *); static int vtnet_txq_enable_intr(struct vtnet_txq *); static void vtnet_txq_disable_intr(struct vtnet_txq *); static void vtnet_enable_rx_interrupts(struct vtnet_softc *); static void vtnet_enable_tx_interrupts(struct vtnet_softc *); static void vtnet_enable_interrupts(struct vtnet_softc *); static void vtnet_disable_rx_interrupts(struct vtnet_softc *); static void vtnet_disable_tx_interrupts(struct vtnet_softc *); static void vtnet_disable_interrupts(struct vtnet_softc *); static int vtnet_tunable_int(struct vtnet_softc *, const char *, int); NETDUMP_DEFINE(vtnet); /* Tunables. */ static SYSCTL_NODE(_hw, OID_AUTO, vtnet, CTLFLAG_RD, 0, "VNET driver parameters"); static int vtnet_csum_disable = 0; TUNABLE_INT("hw.vtnet.csum_disable", &vtnet_csum_disable); SYSCTL_INT(_hw_vtnet, OID_AUTO, csum_disable, CTLFLAG_RDTUN, &vtnet_csum_disable, 0, "Disables receive and send checksum offload"); static int vtnet_tso_disable = 0; TUNABLE_INT("hw.vtnet.tso_disable", &vtnet_tso_disable); SYSCTL_INT(_hw_vtnet, OID_AUTO, tso_disable, CTLFLAG_RDTUN, &vtnet_tso_disable, 0, "Disables TCP Segmentation Offload"); static int vtnet_lro_disable = 0; TUNABLE_INT("hw.vtnet.lro_disable", &vtnet_lro_disable); SYSCTL_INT(_hw_vtnet, OID_AUTO, lro_disable, CTLFLAG_RDTUN, &vtnet_lro_disable, 0, "Disables TCP Large Receive Offload"); static int vtnet_mq_disable = 0; TUNABLE_INT("hw.vtnet.mq_disable", &vtnet_mq_disable); SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_disable, CTLFLAG_RDTUN, &vtnet_mq_disable, 0, "Disables Multi Queue support"); static int vtnet_mq_max_pairs = VTNET_MAX_QUEUE_PAIRS; TUNABLE_INT("hw.vtnet.mq_max_pairs", &vtnet_mq_max_pairs); SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_max_pairs, CTLFLAG_RDTUN, &vtnet_mq_max_pairs, 0, "Sets the maximum number of Multi Queue pairs"); static int vtnet_rx_process_limit = 512; TUNABLE_INT("hw.vtnet.rx_process_limit", &vtnet_rx_process_limit); SYSCTL_INT(_hw_vtnet, OID_AUTO, rx_process_limit, CTLFLAG_RDTUN, &vtnet_rx_process_limit, 0, "Limits the number RX segments processed in a single pass"); static uma_zone_t vtnet_tx_header_zone; static struct virtio_feature_desc vtnet_feature_desc[] = { { VIRTIO_NET_F_CSUM, "TxChecksum" }, { VIRTIO_NET_F_GUEST_CSUM, "RxChecksum" }, { VIRTIO_NET_F_MAC, "MacAddress" }, { VIRTIO_NET_F_GSO, "TxAllGSO" }, { VIRTIO_NET_F_GUEST_TSO4, "RxTSOv4" }, { VIRTIO_NET_F_GUEST_TSO6, "RxTSOv6" }, { VIRTIO_NET_F_GUEST_ECN, "RxECN" }, { VIRTIO_NET_F_GUEST_UFO, "RxUFO" }, { VIRTIO_NET_F_HOST_TSO4, "TxTSOv4" }, { VIRTIO_NET_F_HOST_TSO6, "TxTSOv6" }, { VIRTIO_NET_F_HOST_ECN, "TxTSOECN" }, { VIRTIO_NET_F_HOST_UFO, "TxUFO" }, { VIRTIO_NET_F_MRG_RXBUF, "MrgRxBuf" }, { VIRTIO_NET_F_STATUS, "Status" }, { VIRTIO_NET_F_CTRL_VQ, "ControlVq" }, { VIRTIO_NET_F_CTRL_RX, "RxMode" }, { VIRTIO_NET_F_CTRL_VLAN, "VLanFilter" }, { VIRTIO_NET_F_CTRL_RX_EXTRA, "RxModeExtra" }, { VIRTIO_NET_F_GUEST_ANNOUNCE, "GuestAnnounce" }, { VIRTIO_NET_F_MQ, "Multiqueue" }, { VIRTIO_NET_F_CTRL_MAC_ADDR, "SetMacAddress" }, { 0, NULL } }; static device_method_t vtnet_methods[] = { /* Device methods. */ DEVMETHOD(device_probe, vtnet_probe), DEVMETHOD(device_attach, vtnet_attach), DEVMETHOD(device_detach, vtnet_detach), DEVMETHOD(device_suspend, vtnet_suspend), DEVMETHOD(device_resume, vtnet_resume), DEVMETHOD(device_shutdown, vtnet_shutdown), /* VirtIO methods. */ DEVMETHOD(virtio_attach_completed, vtnet_attach_completed), DEVMETHOD(virtio_config_change, vtnet_config_change), DEVMETHOD_END }; #ifdef DEV_NETMAP #include #endif /* DEV_NETMAP */ static driver_t vtnet_driver = { "vtnet", vtnet_methods, sizeof(struct vtnet_softc) }; static devclass_t vtnet_devclass; DRIVER_MODULE(vtnet, virtio_mmio, vtnet_driver, vtnet_devclass, vtnet_modevent, 0); DRIVER_MODULE(vtnet, virtio_pci, vtnet_driver, vtnet_devclass, vtnet_modevent, 0); MODULE_VERSION(vtnet, 1); MODULE_DEPEND(vtnet, virtio, 1, 1, 1); #ifdef DEV_NETMAP MODULE_DEPEND(vtnet, netmap, 1, 1, 1); #endif /* DEV_NETMAP */ +VIRTIO_SIMPLE_PNPTABLE(vtnet, VIRTIO_ID_NETWORK, "VirtIO Networking Adapter"); +VIRTIO_SIMPLE_PNPINFO(virtio_mmio, vtnet); +VIRTIO_SIMPLE_PNPINFO(virtio_pci, vtnet); + static int vtnet_modevent(module_t mod, int type, void *unused) { int error = 0; static int loaded = 0; switch (type) { case MOD_LOAD: if (loaded++ == 0) vtnet_tx_header_zone = uma_zcreate("vtnet_tx_hdr", sizeof(struct vtnet_tx_header), NULL, NULL, NULL, NULL, 0, 0); break; case MOD_QUIESCE: if (uma_zone_get_cur(vtnet_tx_header_zone) > 0) error = EBUSY; break; case MOD_UNLOAD: if (--loaded == 0) { uma_zdestroy(vtnet_tx_header_zone); vtnet_tx_header_zone = NULL; } break; case MOD_SHUTDOWN: break; default: error = EOPNOTSUPP; break; } return (error); } static int vtnet_probe(device_t dev) { - - if (virtio_get_device_type(dev) != VIRTIO_ID_NETWORK) - return (ENXIO); - - device_set_desc(dev, "VirtIO Networking Adapter"); - - return (BUS_PROBE_DEFAULT); + return (VIRTIO_SIMPLE_PROBE(dev, vtnet)); } static int vtnet_attach(device_t dev) { struct vtnet_softc *sc; int error; sc = device_get_softc(dev); sc->vtnet_dev = dev; /* Register our feature descriptions. */ virtio_set_feature_desc(dev, vtnet_feature_desc); VTNET_CORE_LOCK_INIT(sc); callout_init_mtx(&sc->vtnet_tick_ch, VTNET_CORE_MTX(sc), 0); vtnet_setup_sysctl(sc); vtnet_setup_features(sc); error = vtnet_alloc_rx_filters(sc); if (error) { device_printf(dev, "cannot allocate Rx filters\n"); goto fail; } error = vtnet_alloc_rxtx_queues(sc); if (error) { device_printf(dev, "cannot allocate queues\n"); goto fail; } error = vtnet_alloc_virtqueues(sc); if (error) { device_printf(dev, "cannot allocate virtqueues\n"); goto fail; } error = vtnet_setup_interface(sc); if (error) { device_printf(dev, "cannot setup interface\n"); goto fail; } error = virtio_setup_intr(dev, INTR_TYPE_NET); if (error) { device_printf(dev, "cannot setup virtqueue interrupts\n"); /* BMV: This will crash if during boot! */ ether_ifdetach(sc->vtnet_ifp); goto fail; } #ifdef DEV_NETMAP vtnet_netmap_attach(sc); #endif /* DEV_NETMAP */ vtnet_start_taskqueues(sc); fail: if (error) vtnet_detach(dev); return (error); } static int vtnet_detach(device_t dev) { struct vtnet_softc *sc; struct ifnet *ifp; sc = device_get_softc(dev); ifp = sc->vtnet_ifp; if (device_is_attached(dev)) { VTNET_CORE_LOCK(sc); vtnet_stop(sc); VTNET_CORE_UNLOCK(sc); callout_drain(&sc->vtnet_tick_ch); vtnet_drain_taskqueues(sc); ether_ifdetach(ifp); } #ifdef DEV_NETMAP netmap_detach(ifp); #endif /* DEV_NETMAP */ vtnet_free_taskqueues(sc); if (sc->vtnet_vlan_attach != NULL) { EVENTHANDLER_DEREGISTER(vlan_config, sc->vtnet_vlan_attach); sc->vtnet_vlan_attach = NULL; } if (sc->vtnet_vlan_detach != NULL) { EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vtnet_vlan_detach); sc->vtnet_vlan_detach = NULL; } ifmedia_removeall(&sc->vtnet_media); if (ifp != NULL) { if_free(ifp); sc->vtnet_ifp = NULL; } vtnet_free_rxtx_queues(sc); vtnet_free_rx_filters(sc); if (sc->vtnet_ctrl_vq != NULL) vtnet_free_ctrl_vq(sc); VTNET_CORE_LOCK_DESTROY(sc); return (0); } static int vtnet_suspend(device_t dev) { struct vtnet_softc *sc; sc = device_get_softc(dev); VTNET_CORE_LOCK(sc); vtnet_stop(sc); sc->vtnet_flags |= VTNET_FLAG_SUSPENDED; VTNET_CORE_UNLOCK(sc); return (0); } static int vtnet_resume(device_t dev) { struct vtnet_softc *sc; struct ifnet *ifp; sc = device_get_softc(dev); ifp = sc->vtnet_ifp; VTNET_CORE_LOCK(sc); if (ifp->if_flags & IFF_UP) vtnet_init_locked(sc); sc->vtnet_flags &= ~VTNET_FLAG_SUSPENDED; VTNET_CORE_UNLOCK(sc); return (0); } static int vtnet_shutdown(device_t dev) { /* * Suspend already does all of what we need to * do here; we just never expect to be resumed. */ return (vtnet_suspend(dev)); } static int vtnet_attach_completed(device_t dev) { vtnet_attach_disable_promisc(device_get_softc(dev)); return (0); } static int vtnet_config_change(device_t dev) { struct vtnet_softc *sc; sc = device_get_softc(dev); VTNET_CORE_LOCK(sc); vtnet_update_link_status(sc); if (sc->vtnet_link_active != 0) vtnet_tx_start_all(sc); VTNET_CORE_UNLOCK(sc); return (0); } static void vtnet_negotiate_features(struct vtnet_softc *sc) { device_t dev; uint64_t mask, features; dev = sc->vtnet_dev; mask = 0; /* * TSO and LRO are only available when their corresponding checksum * offload feature is also negotiated. */ if (vtnet_tunable_int(sc, "csum_disable", vtnet_csum_disable)) { mask |= VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM; mask |= VTNET_TSO_FEATURES | VTNET_LRO_FEATURES; } if (vtnet_tunable_int(sc, "tso_disable", vtnet_tso_disable)) mask |= VTNET_TSO_FEATURES; if (vtnet_tunable_int(sc, "lro_disable", vtnet_lro_disable)) mask |= VTNET_LRO_FEATURES; #ifndef VTNET_LEGACY_TX if (vtnet_tunable_int(sc, "mq_disable", vtnet_mq_disable)) mask |= VIRTIO_NET_F_MQ; #else mask |= VIRTIO_NET_F_MQ; #endif features = VTNET_FEATURES & ~mask; sc->vtnet_features = virtio_negotiate_features(dev, features); if (virtio_with_feature(dev, VTNET_LRO_FEATURES) && virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF) == 0) { /* * LRO without mergeable buffers requires special care. This * is not ideal because every receive buffer must be large * enough to hold the maximum TCP packet, the Ethernet header, * and the header. This requires up to 34 descriptors with * MCLBYTES clusters. If we do not have indirect descriptors, * LRO is disabled since the virtqueue will not contain very * many receive buffers. */ if (!virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) { device_printf(dev, "LRO disabled due to both mergeable buffers and " "indirect descriptors not negotiated\n"); features &= ~VTNET_LRO_FEATURES; sc->vtnet_features = virtio_negotiate_features(dev, features); } else sc->vtnet_flags |= VTNET_FLAG_LRO_NOMRG; } } static void vtnet_setup_features(struct vtnet_softc *sc) { device_t dev; dev = sc->vtnet_dev; vtnet_negotiate_features(sc); if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) sc->vtnet_flags |= VTNET_FLAG_INDIRECT; if (virtio_with_feature(dev, VIRTIO_RING_F_EVENT_IDX)) sc->vtnet_flags |= VTNET_FLAG_EVENT_IDX; if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) { /* This feature should always be negotiated. */ sc->vtnet_flags |= VTNET_FLAG_MAC; } if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF)) { sc->vtnet_flags |= VTNET_FLAG_MRG_RXBUFS; sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf); } else sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr); if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) sc->vtnet_rx_nsegs = VTNET_MRG_RX_SEGS; else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG) sc->vtnet_rx_nsegs = VTNET_MAX_RX_SEGS; else sc->vtnet_rx_nsegs = VTNET_MIN_RX_SEGS; if (virtio_with_feature(dev, VIRTIO_NET_F_GSO) || virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4) || virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6)) sc->vtnet_tx_nsegs = VTNET_MAX_TX_SEGS; else sc->vtnet_tx_nsegs = VTNET_MIN_TX_SEGS; if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VQ)) { sc->vtnet_flags |= VTNET_FLAG_CTRL_VQ; if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX)) sc->vtnet_flags |= VTNET_FLAG_CTRL_RX; if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VLAN)) sc->vtnet_flags |= VTNET_FLAG_VLAN_FILTER; if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_MAC_ADDR)) sc->vtnet_flags |= VTNET_FLAG_CTRL_MAC; } if (virtio_with_feature(dev, VIRTIO_NET_F_MQ) && sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) { sc->vtnet_max_vq_pairs = virtio_read_dev_config_2(dev, offsetof(struct virtio_net_config, max_virtqueue_pairs)); } else sc->vtnet_max_vq_pairs = 1; if (sc->vtnet_max_vq_pairs > 1) { /* * Limit the maximum number of queue pairs to the lower of * the number of CPUs and the configured maximum. * The actual number of queues that get used may be less. */ int max; max = vtnet_tunable_int(sc, "mq_max_pairs", vtnet_mq_max_pairs); if (max > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN) { if (max > mp_ncpus) max = mp_ncpus; if (max > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX) max = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX; if (max > 1) { sc->vtnet_requested_vq_pairs = max; sc->vtnet_flags |= VTNET_FLAG_MULTIQ; } } } } static int vtnet_init_rxq(struct vtnet_softc *sc, int id) { struct vtnet_rxq *rxq; rxq = &sc->vtnet_rxqs[id]; snprintf(rxq->vtnrx_name, sizeof(rxq->vtnrx_name), "%s-rx%d", device_get_nameunit(sc->vtnet_dev), id); mtx_init(&rxq->vtnrx_mtx, rxq->vtnrx_name, NULL, MTX_DEF); rxq->vtnrx_sc = sc; rxq->vtnrx_id = id; rxq->vtnrx_sg = sglist_alloc(sc->vtnet_rx_nsegs, M_NOWAIT); if (rxq->vtnrx_sg == NULL) return (ENOMEM); TASK_INIT(&rxq->vtnrx_intrtask, 0, vtnet_rxq_tq_intr, rxq); rxq->vtnrx_tq = taskqueue_create(rxq->vtnrx_name, M_NOWAIT, taskqueue_thread_enqueue, &rxq->vtnrx_tq); return (rxq->vtnrx_tq == NULL ? ENOMEM : 0); } static int vtnet_init_txq(struct vtnet_softc *sc, int id) { struct vtnet_txq *txq; txq = &sc->vtnet_txqs[id]; snprintf(txq->vtntx_name, sizeof(txq->vtntx_name), "%s-tx%d", device_get_nameunit(sc->vtnet_dev), id); mtx_init(&txq->vtntx_mtx, txq->vtntx_name, NULL, MTX_DEF); txq->vtntx_sc = sc; txq->vtntx_id = id; txq->vtntx_sg = sglist_alloc(sc->vtnet_tx_nsegs, M_NOWAIT); if (txq->vtntx_sg == NULL) return (ENOMEM); #ifndef VTNET_LEGACY_TX txq->vtntx_br = buf_ring_alloc(VTNET_DEFAULT_BUFRING_SIZE, M_DEVBUF, M_NOWAIT, &txq->vtntx_mtx); if (txq->vtntx_br == NULL) return (ENOMEM); TASK_INIT(&txq->vtntx_defrtask, 0, vtnet_txq_tq_deferred, txq); #endif TASK_INIT(&txq->vtntx_intrtask, 0, vtnet_txq_tq_intr, txq); txq->vtntx_tq = taskqueue_create(txq->vtntx_name, M_NOWAIT, taskqueue_thread_enqueue, &txq->vtntx_tq); if (txq->vtntx_tq == NULL) return (ENOMEM); return (0); } static int vtnet_alloc_rxtx_queues(struct vtnet_softc *sc) { int i, npairs, error; npairs = sc->vtnet_max_vq_pairs; sc->vtnet_rxqs = malloc(sizeof(struct vtnet_rxq) * npairs, M_DEVBUF, M_NOWAIT | M_ZERO); sc->vtnet_txqs = malloc(sizeof(struct vtnet_txq) * npairs, M_DEVBUF, M_NOWAIT | M_ZERO); if (sc->vtnet_rxqs == NULL || sc->vtnet_txqs == NULL) return (ENOMEM); for (i = 0; i < npairs; i++) { error = vtnet_init_rxq(sc, i); if (error) return (error); error = vtnet_init_txq(sc, i); if (error) return (error); } vtnet_setup_queue_sysctl(sc); return (0); } static void vtnet_destroy_rxq(struct vtnet_rxq *rxq) { rxq->vtnrx_sc = NULL; rxq->vtnrx_id = -1; if (rxq->vtnrx_sg != NULL) { sglist_free(rxq->vtnrx_sg); rxq->vtnrx_sg = NULL; } if (mtx_initialized(&rxq->vtnrx_mtx) != 0) mtx_destroy(&rxq->vtnrx_mtx); } static void vtnet_destroy_txq(struct vtnet_txq *txq) { txq->vtntx_sc = NULL; txq->vtntx_id = -1; if (txq->vtntx_sg != NULL) { sglist_free(txq->vtntx_sg); txq->vtntx_sg = NULL; } #ifndef VTNET_LEGACY_TX if (txq->vtntx_br != NULL) { buf_ring_free(txq->vtntx_br, M_DEVBUF); txq->vtntx_br = NULL; } #endif if (mtx_initialized(&txq->vtntx_mtx) != 0) mtx_destroy(&txq->vtntx_mtx); } static void vtnet_free_rxtx_queues(struct vtnet_softc *sc) { int i; if (sc->vtnet_rxqs != NULL) { for (i = 0; i < sc->vtnet_max_vq_pairs; i++) vtnet_destroy_rxq(&sc->vtnet_rxqs[i]); free(sc->vtnet_rxqs, M_DEVBUF); sc->vtnet_rxqs = NULL; } if (sc->vtnet_txqs != NULL) { for (i = 0; i < sc->vtnet_max_vq_pairs; i++) vtnet_destroy_txq(&sc->vtnet_txqs[i]); free(sc->vtnet_txqs, M_DEVBUF); sc->vtnet_txqs = NULL; } } static int vtnet_alloc_rx_filters(struct vtnet_softc *sc) { if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) { sc->vtnet_mac_filter = malloc(sizeof(struct vtnet_mac_filter), M_DEVBUF, M_NOWAIT | M_ZERO); if (sc->vtnet_mac_filter == NULL) return (ENOMEM); } if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) { sc->vtnet_vlan_filter = malloc(sizeof(uint32_t) * VTNET_VLAN_FILTER_NWORDS, M_DEVBUF, M_NOWAIT | M_ZERO); if (sc->vtnet_vlan_filter == NULL) return (ENOMEM); } return (0); } static void vtnet_free_rx_filters(struct vtnet_softc *sc) { if (sc->vtnet_mac_filter != NULL) { free(sc->vtnet_mac_filter, M_DEVBUF); sc->vtnet_mac_filter = NULL; } if (sc->vtnet_vlan_filter != NULL) { free(sc->vtnet_vlan_filter, M_DEVBUF); sc->vtnet_vlan_filter = NULL; } } static int vtnet_alloc_virtqueues(struct vtnet_softc *sc) { device_t dev; struct vq_alloc_info *info; struct vtnet_rxq *rxq; struct vtnet_txq *txq; int i, idx, flags, nvqs, error; dev = sc->vtnet_dev; flags = 0; nvqs = sc->vtnet_max_vq_pairs * 2; if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) nvqs++; info = malloc(sizeof(struct vq_alloc_info) * nvqs, M_TEMP, M_NOWAIT); if (info == NULL) return (ENOMEM); for (i = 0, idx = 0; i < sc->vtnet_max_vq_pairs; i++, idx+=2) { rxq = &sc->vtnet_rxqs[i]; VQ_ALLOC_INFO_INIT(&info[idx], sc->vtnet_rx_nsegs, vtnet_rx_vq_intr, rxq, &rxq->vtnrx_vq, "%s-%d rx", device_get_nameunit(dev), rxq->vtnrx_id); txq = &sc->vtnet_txqs[i]; VQ_ALLOC_INFO_INIT(&info[idx+1], sc->vtnet_tx_nsegs, vtnet_tx_vq_intr, txq, &txq->vtntx_vq, "%s-%d tx", device_get_nameunit(dev), txq->vtntx_id); } if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) { VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, NULL, &sc->vtnet_ctrl_vq, "%s ctrl", device_get_nameunit(dev)); } /* * Enable interrupt binding if this is multiqueue. This only matters * when per-vq MSIX is available. */ if (sc->vtnet_flags & VTNET_FLAG_MULTIQ) flags |= 0; error = virtio_alloc_virtqueues(dev, flags, nvqs, info); free(info, M_TEMP); return (error); } static int vtnet_setup_interface(struct vtnet_softc *sc) { device_t dev; struct ifnet *ifp; dev = sc->vtnet_dev; ifp = sc->vtnet_ifp = if_alloc(IFT_ETHER); if (ifp == NULL) { device_printf(dev, "cannot allocate ifnet structure\n"); return (ENOSPC); } if_initname(ifp, device_get_name(dev), device_get_unit(dev)); ifp->if_baudrate = IF_Gbps(10); /* Approx. */ ifp->if_softc = sc; ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST; ifp->if_init = vtnet_init; ifp->if_ioctl = vtnet_ioctl; ifp->if_get_counter = vtnet_get_counter; #ifndef VTNET_LEGACY_TX ifp->if_transmit = vtnet_txq_mq_start; ifp->if_qflush = vtnet_qflush; #else struct virtqueue *vq = sc->vtnet_txqs[0].vtntx_vq; ifp->if_start = vtnet_start; IFQ_SET_MAXLEN(&ifp->if_snd, virtqueue_size(vq) - 1); ifp->if_snd.ifq_drv_maxlen = virtqueue_size(vq) - 1; IFQ_SET_READY(&ifp->if_snd); #endif ifmedia_init(&sc->vtnet_media, IFM_IMASK, vtnet_ifmedia_upd, vtnet_ifmedia_sts); ifmedia_add(&sc->vtnet_media, VTNET_MEDIATYPE, 0, NULL); ifmedia_set(&sc->vtnet_media, VTNET_MEDIATYPE); /* Read (or generate) the MAC address for the adapter. */ vtnet_get_hwaddr(sc); ether_ifattach(ifp, sc->vtnet_hwaddr); if (virtio_with_feature(dev, VIRTIO_NET_F_STATUS)) ifp->if_capabilities |= IFCAP_LINKSTATE; /* Tell the upper layer(s) we support long frames. */ ifp->if_hdrlen = sizeof(struct ether_vlan_header); ifp->if_capabilities |= IFCAP_JUMBO_MTU | IFCAP_VLAN_MTU; if (virtio_with_feature(dev, VIRTIO_NET_F_CSUM)) { ifp->if_capabilities |= IFCAP_TXCSUM | IFCAP_TXCSUM_IPV6; if (virtio_with_feature(dev, VIRTIO_NET_F_GSO)) { ifp->if_capabilities |= IFCAP_TSO4 | IFCAP_TSO6; sc->vtnet_flags |= VTNET_FLAG_TSO_ECN; } else { if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4)) ifp->if_capabilities |= IFCAP_TSO4; if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6)) ifp->if_capabilities |= IFCAP_TSO6; if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_ECN)) sc->vtnet_flags |= VTNET_FLAG_TSO_ECN; } if (ifp->if_capabilities & IFCAP_TSO) ifp->if_capabilities |= IFCAP_VLAN_HWTSO; } if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_CSUM)) { ifp->if_capabilities |= IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6; if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO4) || virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO6)) ifp->if_capabilities |= IFCAP_LRO; } if (ifp->if_capabilities & IFCAP_HWCSUM) { /* * VirtIO does not support VLAN tagging, but we can fake * it by inserting and removing the 802.1Q header during * transmit and receive. We are then able to do checksum * offloading of VLAN frames. */ ifp->if_capabilities |= IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM; } ifp->if_capenable = ifp->if_capabilities; /* * Capabilities after here are not enabled by default. */ if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) { ifp->if_capabilities |= IFCAP_VLAN_HWFILTER; sc->vtnet_vlan_attach = EVENTHANDLER_REGISTER(vlan_config, vtnet_register_vlan, sc, EVENTHANDLER_PRI_FIRST); sc->vtnet_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig, vtnet_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST); } vtnet_set_rx_process_limit(sc); vtnet_set_tx_intr_threshold(sc); NETDUMP_SET(ifp, vtnet); return (0); } static int vtnet_change_mtu(struct vtnet_softc *sc, int new_mtu) { struct ifnet *ifp; int frame_size, clsize; ifp = sc->vtnet_ifp; if (new_mtu < ETHERMIN || new_mtu > VTNET_MAX_MTU) return (EINVAL); frame_size = sc->vtnet_hdr_size + sizeof(struct ether_vlan_header) + new_mtu; /* * Based on the new MTU (and hence frame size) determine which * cluster size is most appropriate for the receive queues. */ if (frame_size <= MCLBYTES) { clsize = MCLBYTES; } else if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) { /* Avoid going past 9K jumbos. */ if (frame_size > MJUM9BYTES) return (EINVAL); clsize = MJUM9BYTES; } else clsize = MJUMPAGESIZE; ifp->if_mtu = new_mtu; sc->vtnet_rx_new_clsize = clsize; if (ifp->if_drv_flags & IFF_DRV_RUNNING) { ifp->if_drv_flags &= ~IFF_DRV_RUNNING; vtnet_init_locked(sc); } return (0); } static int vtnet_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) { struct vtnet_softc *sc; struct ifreq *ifr; int reinit, mask, error; sc = ifp->if_softc; ifr = (struct ifreq *) data; error = 0; switch (cmd) { case SIOCSIFMTU: if (ifp->if_mtu != ifr->ifr_mtu) { VTNET_CORE_LOCK(sc); error = vtnet_change_mtu(sc, ifr->ifr_mtu); VTNET_CORE_UNLOCK(sc); } break; case SIOCSIFFLAGS: VTNET_CORE_LOCK(sc); if ((ifp->if_flags & IFF_UP) == 0) { if (ifp->if_drv_flags & IFF_DRV_RUNNING) vtnet_stop(sc); } else if (ifp->if_drv_flags & IFF_DRV_RUNNING) { if ((ifp->if_flags ^ sc->vtnet_if_flags) & (IFF_PROMISC | IFF_ALLMULTI)) { if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) vtnet_rx_filter(sc); else { ifp->if_flags |= IFF_PROMISC; if ((ifp->if_flags ^ sc->vtnet_if_flags) & IFF_ALLMULTI) error = ENOTSUP; } } } else vtnet_init_locked(sc); if (error == 0) sc->vtnet_if_flags = ifp->if_flags; VTNET_CORE_UNLOCK(sc); break; case SIOCADDMULTI: case SIOCDELMULTI: if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0) break; VTNET_CORE_LOCK(sc); if (ifp->if_drv_flags & IFF_DRV_RUNNING) vtnet_rx_filter_mac(sc); VTNET_CORE_UNLOCK(sc); break; case SIOCSIFMEDIA: case SIOCGIFMEDIA: error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd); break; case SIOCSIFCAP: VTNET_CORE_LOCK(sc); mask = ifr->ifr_reqcap ^ ifp->if_capenable; if (mask & IFCAP_TXCSUM) ifp->if_capenable ^= IFCAP_TXCSUM; if (mask & IFCAP_TXCSUM_IPV6) ifp->if_capenable ^= IFCAP_TXCSUM_IPV6; if (mask & IFCAP_TSO4) ifp->if_capenable ^= IFCAP_TSO4; if (mask & IFCAP_TSO6) ifp->if_capenable ^= IFCAP_TSO6; if (mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO | IFCAP_VLAN_HWFILTER)) { /* These Rx features require us to renegotiate. */ reinit = 1; if (mask & IFCAP_RXCSUM) ifp->if_capenable ^= IFCAP_RXCSUM; if (mask & IFCAP_RXCSUM_IPV6) ifp->if_capenable ^= IFCAP_RXCSUM_IPV6; if (mask & IFCAP_LRO) ifp->if_capenable ^= IFCAP_LRO; if (mask & IFCAP_VLAN_HWFILTER) ifp->if_capenable ^= IFCAP_VLAN_HWFILTER; } else reinit = 0; if (mask & IFCAP_VLAN_HWTSO) ifp->if_capenable ^= IFCAP_VLAN_HWTSO; if (mask & IFCAP_VLAN_HWTAGGING) ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING; if (reinit && (ifp->if_drv_flags & IFF_DRV_RUNNING)) { ifp->if_drv_flags &= ~IFF_DRV_RUNNING; vtnet_init_locked(sc); } VTNET_CORE_UNLOCK(sc); VLAN_CAPABILITIES(ifp); break; default: error = ether_ioctl(ifp, cmd, data); break; } VTNET_CORE_LOCK_ASSERT_NOTOWNED(sc); return (error); } static int vtnet_rxq_populate(struct vtnet_rxq *rxq) { struct virtqueue *vq; int nbufs, error; #ifdef DEV_NETMAP error = vtnet_netmap_rxq_populate(rxq); if (error >= 0) return (error); #endif /* DEV_NETMAP */ vq = rxq->vtnrx_vq; error = ENOSPC; for (nbufs = 0; !virtqueue_full(vq); nbufs++) { error = vtnet_rxq_new_buf(rxq); if (error) break; } if (nbufs > 0) { virtqueue_notify(vq); /* * EMSGSIZE signifies the virtqueue did not have enough * entries available to hold the last mbuf. This is not * an error. */ if (error == EMSGSIZE) error = 0; } return (error); } static void vtnet_rxq_free_mbufs(struct vtnet_rxq *rxq) { struct virtqueue *vq; struct mbuf *m; int last; #ifdef DEV_NETMAP int netmap_bufs = vtnet_netmap_queue_on(rxq->vtnrx_sc, NR_RX, rxq->vtnrx_id); #else /* !DEV_NETMAP */ int netmap_bufs = 0; #endif /* !DEV_NETMAP */ vq = rxq->vtnrx_vq; last = 0; while ((m = virtqueue_drain(vq, &last)) != NULL) { if (!netmap_bufs) m_freem(m); } KASSERT(virtqueue_empty(vq), ("%s: mbufs remaining in rx queue %p", __func__, rxq)); } static struct mbuf * vtnet_rx_alloc_buf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp) { struct mbuf *m_head, *m_tail, *m; int i, clsize; clsize = sc->vtnet_rx_clsize; KASSERT(nbufs == 1 || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG, ("%s: chained mbuf %d request without LRO_NOMRG", __func__, nbufs)); m_head = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, clsize); if (m_head == NULL) goto fail; m_head->m_len = clsize; m_tail = m_head; /* Allocate the rest of the chain. */ for (i = 1; i < nbufs; i++) { m = m_getjcl(M_NOWAIT, MT_DATA, 0, clsize); if (m == NULL) goto fail; m->m_len = clsize; m_tail->m_next = m; m_tail = m; } if (m_tailp != NULL) *m_tailp = m_tail; return (m_head); fail: sc->vtnet_stats.mbuf_alloc_failed++; m_freem(m_head); return (NULL); } /* * Slow path for when LRO without mergeable buffers is negotiated. */ static int vtnet_rxq_replace_lro_nomgr_buf(struct vtnet_rxq *rxq, struct mbuf *m0, int len0) { struct vtnet_softc *sc; struct mbuf *m, *m_prev; struct mbuf *m_new, *m_tail; int len, clsize, nreplace, error; sc = rxq->vtnrx_sc; clsize = sc->vtnet_rx_clsize; m_prev = NULL; m_tail = NULL; nreplace = 0; m = m0; len = len0; /* * Since these mbuf chains are so large, we avoid allocating an * entire replacement chain if possible. When the received frame * did not consume the entire chain, the unused mbufs are moved * to the replacement chain. */ while (len > 0) { /* * Something is seriously wrong if we received a frame * larger than the chain. Drop it. */ if (m == NULL) { sc->vtnet_stats.rx_frame_too_large++; return (EMSGSIZE); } /* We always allocate the same cluster size. */ KASSERT(m->m_len == clsize, ("%s: mbuf size %d is not the cluster size %d", __func__, m->m_len, clsize)); m->m_len = MIN(m->m_len, len); len -= m->m_len; m_prev = m; m = m->m_next; nreplace++; } KASSERT(nreplace <= sc->vtnet_rx_nmbufs, ("%s: too many replacement mbufs %d max %d", __func__, nreplace, sc->vtnet_rx_nmbufs)); m_new = vtnet_rx_alloc_buf(sc, nreplace, &m_tail); if (m_new == NULL) { m_prev->m_len = clsize; return (ENOBUFS); } /* * Move any unused mbufs from the received chain onto the end * of the new chain. */ if (m_prev->m_next != NULL) { m_tail->m_next = m_prev->m_next; m_prev->m_next = NULL; } error = vtnet_rxq_enqueue_buf(rxq, m_new); if (error) { /* * BAD! We could not enqueue the replacement mbuf chain. We * must restore the m0 chain to the original state if it was * modified so we can subsequently discard it. * * NOTE: The replacement is suppose to be an identical copy * to the one just dequeued so this is an unexpected error. */ sc->vtnet_stats.rx_enq_replacement_failed++; if (m_tail->m_next != NULL) { m_prev->m_next = m_tail->m_next; m_tail->m_next = NULL; } m_prev->m_len = clsize; m_freem(m_new); } return (error); } static int vtnet_rxq_replace_buf(struct vtnet_rxq *rxq, struct mbuf *m, int len) { struct vtnet_softc *sc; struct mbuf *m_new; int error; sc = rxq->vtnrx_sc; KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG || m->m_next == NULL, ("%s: chained mbuf without LRO_NOMRG", __func__)); if (m->m_next == NULL) { /* Fast-path for the common case of just one mbuf. */ if (m->m_len < len) return (EINVAL); m_new = vtnet_rx_alloc_buf(sc, 1, NULL); if (m_new == NULL) return (ENOBUFS); error = vtnet_rxq_enqueue_buf(rxq, m_new); if (error) { /* * The new mbuf is suppose to be an identical * copy of the one just dequeued so this is an * unexpected error. */ m_freem(m_new); sc->vtnet_stats.rx_enq_replacement_failed++; } else m->m_len = len; } else error = vtnet_rxq_replace_lro_nomgr_buf(rxq, m, len); return (error); } static int vtnet_rxq_enqueue_buf(struct vtnet_rxq *rxq, struct mbuf *m) { struct vtnet_softc *sc; struct sglist *sg; struct vtnet_rx_header *rxhdr; uint8_t *mdata; int offset, error; sc = rxq->vtnrx_sc; sg = rxq->vtnrx_sg; mdata = mtod(m, uint8_t *); VTNET_RXQ_LOCK_ASSERT(rxq); KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG || m->m_next == NULL, ("%s: chained mbuf without LRO_NOMRG", __func__)); KASSERT(m->m_len == sc->vtnet_rx_clsize, ("%s: unexpected cluster size %d/%d", __func__, m->m_len, sc->vtnet_rx_clsize)); sglist_reset(sg); if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) { MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr)); rxhdr = (struct vtnet_rx_header *) mdata; sglist_append(sg, &rxhdr->vrh_hdr, sc->vtnet_hdr_size); offset = sizeof(struct vtnet_rx_header); } else offset = 0; sglist_append(sg, mdata + offset, m->m_len - offset); if (m->m_next != NULL) { error = sglist_append_mbuf(sg, m->m_next); MPASS(error == 0); } error = virtqueue_enqueue(rxq->vtnrx_vq, m, sg, 0, sg->sg_nseg); return (error); } static int vtnet_rxq_new_buf(struct vtnet_rxq *rxq) { struct vtnet_softc *sc; struct mbuf *m; int error; sc = rxq->vtnrx_sc; m = vtnet_rx_alloc_buf(sc, sc->vtnet_rx_nmbufs, NULL); if (m == NULL) return (ENOBUFS); error = vtnet_rxq_enqueue_buf(rxq, m); if (error) m_freem(m); return (error); } /* * Use the checksum offset in the VirtIO header to set the * correct CSUM_* flags. */ static int vtnet_rxq_csum_by_offset(struct vtnet_rxq *rxq, struct mbuf *m, uint16_t eth_type, int ip_start, struct virtio_net_hdr *hdr) { struct vtnet_softc *sc; #if defined(INET) || defined(INET6) int offset = hdr->csum_start + hdr->csum_offset; #endif sc = rxq->vtnrx_sc; /* Only do a basic sanity check on the offset. */ switch (eth_type) { #if defined(INET) case ETHERTYPE_IP: if (__predict_false(offset < ip_start + sizeof(struct ip))) return (1); break; #endif #if defined(INET6) case ETHERTYPE_IPV6: if (__predict_false(offset < ip_start + sizeof(struct ip6_hdr))) return (1); break; #endif default: sc->vtnet_stats.rx_csum_bad_ethtype++; return (1); } /* * Use the offset to determine the appropriate CSUM_* flags. This is * a bit dirty, but we can get by with it since the checksum offsets * happen to be different. We assume the host host does not do IPv4 * header checksum offloading. */ switch (hdr->csum_offset) { case offsetof(struct udphdr, uh_sum): case offsetof(struct tcphdr, th_sum): m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR; m->m_pkthdr.csum_data = 0xFFFF; break; default: sc->vtnet_stats.rx_csum_bad_offset++; return (1); } return (0); } static int vtnet_rxq_csum_by_parse(struct vtnet_rxq *rxq, struct mbuf *m, uint16_t eth_type, int ip_start, struct virtio_net_hdr *hdr) { struct vtnet_softc *sc; int offset, proto; sc = rxq->vtnrx_sc; switch (eth_type) { #if defined(INET) case ETHERTYPE_IP: { struct ip *ip; if (__predict_false(m->m_len < ip_start + sizeof(struct ip))) return (1); ip = (struct ip *)(m->m_data + ip_start); proto = ip->ip_p; offset = ip_start + (ip->ip_hl << 2); break; } #endif #if defined(INET6) case ETHERTYPE_IPV6: if (__predict_false(m->m_len < ip_start + sizeof(struct ip6_hdr))) return (1); offset = ip6_lasthdr(m, ip_start, IPPROTO_IPV6, &proto); if (__predict_false(offset < 0)) return (1); break; #endif default: sc->vtnet_stats.rx_csum_bad_ethtype++; return (1); } switch (proto) { case IPPROTO_TCP: if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) return (1); m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR; m->m_pkthdr.csum_data = 0xFFFF; break; case IPPROTO_UDP: if (__predict_false(m->m_len < offset + sizeof(struct udphdr))) return (1); m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR; m->m_pkthdr.csum_data = 0xFFFF; break; default: /* * For the remaining protocols, FreeBSD does not support * checksum offloading, so the checksum will be recomputed. */ #if 0 if_printf(sc->vtnet_ifp, "cksum offload of unsupported " "protocol eth_type=%#x proto=%d csum_start=%d " "csum_offset=%d\n", __func__, eth_type, proto, hdr->csum_start, hdr->csum_offset); #endif break; } return (0); } /* * Set the appropriate CSUM_* flags. Unfortunately, the information * provided is not directly useful to us. The VirtIO header gives the * offset of the checksum, which is all Linux needs, but this is not * how FreeBSD does things. We are forced to peek inside the packet * a bit. * * It would be nice if VirtIO gave us the L4 protocol or if FreeBSD * could accept the offsets and let the stack figure it out. */ static int vtnet_rxq_csum(struct vtnet_rxq *rxq, struct mbuf *m, struct virtio_net_hdr *hdr) { struct ether_header *eh; struct ether_vlan_header *evh; uint16_t eth_type; int offset, error; eh = mtod(m, struct ether_header *); eth_type = ntohs(eh->ether_type); if (eth_type == ETHERTYPE_VLAN) { /* BMV: We should handle nested VLAN tags too. */ evh = mtod(m, struct ether_vlan_header *); eth_type = ntohs(evh->evl_proto); offset = sizeof(struct ether_vlan_header); } else offset = sizeof(struct ether_header); if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) error = vtnet_rxq_csum_by_offset(rxq, m, eth_type, offset, hdr); else error = vtnet_rxq_csum_by_parse(rxq, m, eth_type, offset, hdr); return (error); } static void vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *rxq, int nbufs) { struct mbuf *m; while (--nbufs > 0) { m = virtqueue_dequeue(rxq->vtnrx_vq, NULL); if (m == NULL) break; vtnet_rxq_discard_buf(rxq, m); } } static void vtnet_rxq_discard_buf(struct vtnet_rxq *rxq, struct mbuf *m) { int error; /* * Requeue the discarded mbuf. This should always be successful * since it was just dequeued. */ error = vtnet_rxq_enqueue_buf(rxq, m); KASSERT(error == 0, ("%s: cannot requeue discarded mbuf %d", __func__, error)); } static int vtnet_rxq_merged_eof(struct vtnet_rxq *rxq, struct mbuf *m_head, int nbufs) { struct vtnet_softc *sc; struct virtqueue *vq; struct mbuf *m, *m_tail; int len; sc = rxq->vtnrx_sc; vq = rxq->vtnrx_vq; m_tail = m_head; while (--nbufs > 0) { m = virtqueue_dequeue(vq, &len); if (m == NULL) { rxq->vtnrx_stats.vrxs_ierrors++; goto fail; } if (vtnet_rxq_new_buf(rxq) != 0) { rxq->vtnrx_stats.vrxs_iqdrops++; vtnet_rxq_discard_buf(rxq, m); if (nbufs > 1) vtnet_rxq_discard_merged_bufs(rxq, nbufs); goto fail; } if (m->m_len < len) len = m->m_len; m->m_len = len; m->m_flags &= ~M_PKTHDR; m_head->m_pkthdr.len += len; m_tail->m_next = m; m_tail = m; } return (0); fail: sc->vtnet_stats.rx_mergeable_failed++; m_freem(m_head); return (1); } static void vtnet_rxq_input(struct vtnet_rxq *rxq, struct mbuf *m, struct virtio_net_hdr *hdr) { struct vtnet_softc *sc; struct ifnet *ifp; struct ether_header *eh; sc = rxq->vtnrx_sc; ifp = sc->vtnet_ifp; if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) { eh = mtod(m, struct ether_header *); if (eh->ether_type == htons(ETHERTYPE_VLAN)) { vtnet_vlan_tag_remove(m); /* * With the 802.1Q header removed, update the * checksum starting location accordingly. */ if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) hdr->csum_start -= ETHER_VLAN_ENCAP_LEN; } } m->m_pkthdr.flowid = rxq->vtnrx_id; M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE); /* * BMV: FreeBSD does not have the UNNECESSARY and PARTIAL checksum * distinction that Linux does. Need to reevaluate if performing * offloading for the NEEDS_CSUM case is really appropriate. */ if (hdr->flags & (VIRTIO_NET_HDR_F_NEEDS_CSUM | VIRTIO_NET_HDR_F_DATA_VALID)) { if (vtnet_rxq_csum(rxq, m, hdr) == 0) rxq->vtnrx_stats.vrxs_csum++; else rxq->vtnrx_stats.vrxs_csum_failed++; } rxq->vtnrx_stats.vrxs_ipackets++; rxq->vtnrx_stats.vrxs_ibytes += m->m_pkthdr.len; VTNET_RXQ_UNLOCK(rxq); (*ifp->if_input)(ifp, m); VTNET_RXQ_LOCK(rxq); } static int vtnet_rxq_eof(struct vtnet_rxq *rxq) { struct virtio_net_hdr lhdr, *hdr; struct vtnet_softc *sc; struct ifnet *ifp; struct virtqueue *vq; struct mbuf *m; struct virtio_net_hdr_mrg_rxbuf *mhdr; int len, deq, nbufs, adjsz, count; sc = rxq->vtnrx_sc; vq = rxq->vtnrx_vq; ifp = sc->vtnet_ifp; hdr = &lhdr; deq = 0; count = sc->vtnet_rx_process_limit; VTNET_RXQ_LOCK_ASSERT(rxq); while (count-- > 0) { m = virtqueue_dequeue(vq, &len); if (m == NULL) break; deq++; if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) { rxq->vtnrx_stats.vrxs_ierrors++; vtnet_rxq_discard_buf(rxq, m); continue; } if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) { nbufs = 1; adjsz = sizeof(struct vtnet_rx_header); /* * Account for our pad inserted between the header * and the actual start of the frame. */ len += VTNET_RX_HEADER_PAD; } else { mhdr = mtod(m, struct virtio_net_hdr_mrg_rxbuf *); nbufs = mhdr->num_buffers; adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf); } if (vtnet_rxq_replace_buf(rxq, m, len) != 0) { rxq->vtnrx_stats.vrxs_iqdrops++; vtnet_rxq_discard_buf(rxq, m); if (nbufs > 1) vtnet_rxq_discard_merged_bufs(rxq, nbufs); continue; } m->m_pkthdr.len = len; m->m_pkthdr.rcvif = ifp; m->m_pkthdr.csum_flags = 0; if (nbufs > 1) { /* Dequeue the rest of chain. */ if (vtnet_rxq_merged_eof(rxq, m, nbufs) != 0) continue; } /* * Save copy of header before we strip it. For both mergeable * and non-mergeable, the header is at the beginning of the * mbuf data. We no longer need num_buffers, so always use a * regular header. * * BMV: Is this memcpy() expensive? We know the mbuf data is * still valid even after the m_adj(). */ memcpy(hdr, mtod(m, void *), sizeof(struct virtio_net_hdr)); m_adj(m, adjsz); vtnet_rxq_input(rxq, m, hdr); /* Must recheck after dropping the Rx lock. */ if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) break; } if (deq > 0) virtqueue_notify(vq); return (count > 0 ? 0 : EAGAIN); } static void vtnet_rx_vq_intr(void *xrxq) { struct vtnet_softc *sc; struct vtnet_rxq *rxq; struct ifnet *ifp; int tries, more; rxq = xrxq; sc = rxq->vtnrx_sc; ifp = sc->vtnet_ifp; tries = 0; if (__predict_false(rxq->vtnrx_id >= sc->vtnet_act_vq_pairs)) { /* * Ignore this interrupt. Either this is a spurious interrupt * or multiqueue without per-VQ MSIX so every queue needs to * be polled (a brain dead configuration we could try harder * to avoid). */ vtnet_rxq_disable_intr(rxq); return; } #ifdef DEV_NETMAP if (netmap_rx_irq(ifp, rxq->vtnrx_id, &more) != NM_IRQ_PASS) return; #endif /* DEV_NETMAP */ VTNET_RXQ_LOCK(rxq); again: if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { VTNET_RXQ_UNLOCK(rxq); return; } more = vtnet_rxq_eof(rxq); if (more || vtnet_rxq_enable_intr(rxq) != 0) { if (!more) vtnet_rxq_disable_intr(rxq); /* * This is an occasional condition or race (when !more), * so retry a few times before scheduling the taskqueue. */ if (tries++ < VTNET_INTR_DISABLE_RETRIES) goto again; VTNET_RXQ_UNLOCK(rxq); rxq->vtnrx_stats.vrxs_rescheduled++; taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask); } else VTNET_RXQ_UNLOCK(rxq); } static void vtnet_rxq_tq_intr(void *xrxq, int pending) { struct vtnet_softc *sc; struct vtnet_rxq *rxq; struct ifnet *ifp; int more; rxq = xrxq; sc = rxq->vtnrx_sc; ifp = sc->vtnet_ifp; VTNET_RXQ_LOCK(rxq); if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { VTNET_RXQ_UNLOCK(rxq); return; } more = vtnet_rxq_eof(rxq); if (more || vtnet_rxq_enable_intr(rxq) != 0) { if (!more) vtnet_rxq_disable_intr(rxq); rxq->vtnrx_stats.vrxs_rescheduled++; taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask); } VTNET_RXQ_UNLOCK(rxq); } static int vtnet_txq_below_threshold(struct vtnet_txq *txq) { struct vtnet_softc *sc; struct virtqueue *vq; sc = txq->vtntx_sc; vq = txq->vtntx_vq; return (virtqueue_nfree(vq) <= sc->vtnet_tx_intr_thresh); } static int vtnet_txq_notify(struct vtnet_txq *txq) { struct virtqueue *vq; vq = txq->vtntx_vq; txq->vtntx_watchdog = VTNET_TX_TIMEOUT; virtqueue_notify(vq); if (vtnet_txq_enable_intr(txq) == 0) return (0); /* * Drain frames that were completed since last checked. If this * causes the queue to go above the threshold, the caller should * continue transmitting. */ if (vtnet_txq_eof(txq) != 0 && vtnet_txq_below_threshold(txq) == 0) { virtqueue_disable_intr(vq); return (1); } return (0); } static void vtnet_txq_free_mbufs(struct vtnet_txq *txq) { struct virtqueue *vq; struct vtnet_tx_header *txhdr; int last; #ifdef DEV_NETMAP int netmap_bufs = vtnet_netmap_queue_on(txq->vtntx_sc, NR_TX, txq->vtntx_id); #else /* !DEV_NETMAP */ int netmap_bufs = 0; #endif /* !DEV_NETMAP */ vq = txq->vtntx_vq; last = 0; while ((txhdr = virtqueue_drain(vq, &last)) != NULL) { if (!netmap_bufs) { m_freem(txhdr->vth_mbuf); uma_zfree(vtnet_tx_header_zone, txhdr); } } KASSERT(virtqueue_empty(vq), ("%s: mbufs remaining in tx queue %p", __func__, txq)); } /* * BMV: Much of this can go away once we finally have offsets in * the mbuf packet header. Bug andre@. */ static int vtnet_txq_offload_ctx(struct vtnet_txq *txq, struct mbuf *m, int *etype, int *proto, int *start) { struct vtnet_softc *sc; struct ether_vlan_header *evh; int offset; sc = txq->vtntx_sc; evh = mtod(m, struct ether_vlan_header *); if (evh->evl_encap_proto == htons(ETHERTYPE_VLAN)) { /* BMV: We should handle nested VLAN tags too. */ *etype = ntohs(evh->evl_proto); offset = sizeof(struct ether_vlan_header); } else { *etype = ntohs(evh->evl_encap_proto); offset = sizeof(struct ether_header); } switch (*etype) { #if defined(INET) case ETHERTYPE_IP: { struct ip *ip, iphdr; if (__predict_false(m->m_len < offset + sizeof(struct ip))) { m_copydata(m, offset, sizeof(struct ip), (caddr_t) &iphdr); ip = &iphdr; } else ip = (struct ip *)(m->m_data + offset); *proto = ip->ip_p; *start = offset + (ip->ip_hl << 2); break; } #endif #if defined(INET6) case ETHERTYPE_IPV6: *proto = -1; *start = ip6_lasthdr(m, offset, IPPROTO_IPV6, proto); /* Assert the network stack sent us a valid packet. */ KASSERT(*start > offset, ("%s: mbuf %p start %d offset %d proto %d", __func__, m, *start, offset, *proto)); break; #endif default: sc->vtnet_stats.tx_csum_bad_ethtype++; return (EINVAL); } return (0); } static int vtnet_txq_offload_tso(struct vtnet_txq *txq, struct mbuf *m, int eth_type, int offset, struct virtio_net_hdr *hdr) { static struct timeval lastecn; static int curecn; struct vtnet_softc *sc; struct tcphdr *tcp, tcphdr; sc = txq->vtntx_sc; if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) { m_copydata(m, offset, sizeof(struct tcphdr), (caddr_t) &tcphdr); tcp = &tcphdr; } else tcp = (struct tcphdr *)(m->m_data + offset); hdr->hdr_len = offset + (tcp->th_off << 2); hdr->gso_size = m->m_pkthdr.tso_segsz; hdr->gso_type = eth_type == ETHERTYPE_IP ? VIRTIO_NET_HDR_GSO_TCPV4 : VIRTIO_NET_HDR_GSO_TCPV6; if (tcp->th_flags & TH_CWR) { /* * Drop if VIRTIO_NET_F_HOST_ECN was not negotiated. In FreeBSD, * ECN support is not on a per-interface basis, but globally via * the net.inet.tcp.ecn.enable sysctl knob. The default is off. */ if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) { if (ppsratecheck(&lastecn, &curecn, 1)) if_printf(sc->vtnet_ifp, "TSO with ECN not negotiated with host\n"); return (ENOTSUP); } hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN; } txq->vtntx_stats.vtxs_tso++; return (0); } static struct mbuf * vtnet_txq_offload(struct vtnet_txq *txq, struct mbuf *m, struct virtio_net_hdr *hdr) { struct vtnet_softc *sc; int flags, etype, csum_start, proto, error; sc = txq->vtntx_sc; flags = m->m_pkthdr.csum_flags; error = vtnet_txq_offload_ctx(txq, m, &etype, &proto, &csum_start); if (error) goto drop; if ((etype == ETHERTYPE_IP && flags & VTNET_CSUM_OFFLOAD) || (etype == ETHERTYPE_IPV6 && flags & VTNET_CSUM_OFFLOAD_IPV6)) { /* * We could compare the IP protocol vs the CSUM_ flag too, * but that really should not be necessary. */ hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM; hdr->csum_start = csum_start; hdr->csum_offset = m->m_pkthdr.csum_data; txq->vtntx_stats.vtxs_csum++; } if (flags & CSUM_TSO) { if (__predict_false(proto != IPPROTO_TCP)) { /* Likely failed to correctly parse the mbuf. */ sc->vtnet_stats.tx_tso_not_tcp++; goto drop; } KASSERT(hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM, ("%s: mbuf %p TSO without checksum offload %#x", __func__, m, flags)); error = vtnet_txq_offload_tso(txq, m, etype, csum_start, hdr); if (error) goto drop; } return (m); drop: m_freem(m); return (NULL); } static int vtnet_txq_enqueue_buf(struct vtnet_txq *txq, struct mbuf **m_head, struct vtnet_tx_header *txhdr) { struct vtnet_softc *sc; struct virtqueue *vq; struct sglist *sg; struct mbuf *m; int error; sc = txq->vtntx_sc; vq = txq->vtntx_vq; sg = txq->vtntx_sg; m = *m_head; sglist_reset(sg); error = sglist_append(sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size); KASSERT(error == 0 && sg->sg_nseg == 1, ("%s: error %d adding header to sglist", __func__, error)); error = sglist_append_mbuf(sg, m); if (error) { m = m_defrag(m, M_NOWAIT); if (m == NULL) goto fail; *m_head = m; sc->vtnet_stats.tx_defragged++; error = sglist_append_mbuf(sg, m); if (error) goto fail; } txhdr->vth_mbuf = m; error = virtqueue_enqueue(vq, txhdr, sg, sg->sg_nseg, 0); return (error); fail: sc->vtnet_stats.tx_defrag_failed++; m_freem(*m_head); *m_head = NULL; return (ENOBUFS); } static int vtnet_txq_encap(struct vtnet_txq *txq, struct mbuf **m_head, int flags) { struct vtnet_tx_header *txhdr; struct virtio_net_hdr *hdr; struct mbuf *m; int error; m = *m_head; M_ASSERTPKTHDR(m); txhdr = uma_zalloc(vtnet_tx_header_zone, flags | M_ZERO); if (txhdr == NULL) { m_freem(m); *m_head = NULL; return (ENOMEM); } /* * Always use the non-mergeable header, regardless if the feature * was negotiated. For transmit, num_buffers is always zero. The * vtnet_hdr_size is used to enqueue the correct header size. */ hdr = &txhdr->vth_uhdr.hdr; if (m->m_flags & M_VLANTAG) { m = ether_vlanencap(m, m->m_pkthdr.ether_vtag); if ((*m_head = m) == NULL) { error = ENOBUFS; goto fail; } m->m_flags &= ~M_VLANTAG; } if (m->m_pkthdr.csum_flags & VTNET_CSUM_ALL_OFFLOAD) { m = vtnet_txq_offload(txq, m, hdr); if ((*m_head = m) == NULL) { error = ENOBUFS; goto fail; } } error = vtnet_txq_enqueue_buf(txq, m_head, txhdr); if (error == 0) return (0); fail: uma_zfree(vtnet_tx_header_zone, txhdr); return (error); } #ifdef VTNET_LEGACY_TX static void vtnet_start_locked(struct vtnet_txq *txq, struct ifnet *ifp) { struct vtnet_softc *sc; struct virtqueue *vq; struct mbuf *m0; int tries, enq; sc = txq->vtntx_sc; vq = txq->vtntx_vq; tries = 0; VTNET_TXQ_LOCK_ASSERT(txq); if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || sc->vtnet_link_active == 0) return; vtnet_txq_eof(txq); again: enq = 0; while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) { if (virtqueue_full(vq)) break; IFQ_DRV_DEQUEUE(&ifp->if_snd, m0); if (m0 == NULL) break; if (vtnet_txq_encap(txq, &m0, M_NOWAIT) != 0) { if (m0 != NULL) IFQ_DRV_PREPEND(&ifp->if_snd, m0); break; } enq++; ETHER_BPF_MTAP(ifp, m0); } if (enq > 0 && vtnet_txq_notify(txq) != 0) { if (tries++ < VTNET_NOTIFY_RETRIES) goto again; txq->vtntx_stats.vtxs_rescheduled++; taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask); } } static void vtnet_start(struct ifnet *ifp) { struct vtnet_softc *sc; struct vtnet_txq *txq; sc = ifp->if_softc; txq = &sc->vtnet_txqs[0]; VTNET_TXQ_LOCK(txq); vtnet_start_locked(txq, ifp); VTNET_TXQ_UNLOCK(txq); } #else /* !VTNET_LEGACY_TX */ static int vtnet_txq_mq_start_locked(struct vtnet_txq *txq, struct mbuf *m) { struct vtnet_softc *sc; struct virtqueue *vq; struct buf_ring *br; struct ifnet *ifp; int enq, tries, error; sc = txq->vtntx_sc; vq = txq->vtntx_vq; br = txq->vtntx_br; ifp = sc->vtnet_ifp; tries = 0; error = 0; VTNET_TXQ_LOCK_ASSERT(txq); if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || sc->vtnet_link_active == 0) { if (m != NULL) error = drbr_enqueue(ifp, br, m); return (error); } if (m != NULL) { error = drbr_enqueue(ifp, br, m); if (error) return (error); } vtnet_txq_eof(txq); again: enq = 0; while ((m = drbr_peek(ifp, br)) != NULL) { if (virtqueue_full(vq)) { drbr_putback(ifp, br, m); break; } if (vtnet_txq_encap(txq, &m, M_NOWAIT) != 0) { if (m != NULL) drbr_putback(ifp, br, m); else drbr_advance(ifp, br); break; } drbr_advance(ifp, br); enq++; ETHER_BPF_MTAP(ifp, m); } if (enq > 0 && vtnet_txq_notify(txq) != 0) { if (tries++ < VTNET_NOTIFY_RETRIES) goto again; txq->vtntx_stats.vtxs_rescheduled++; taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask); } return (0); } static int vtnet_txq_mq_start(struct ifnet *ifp, struct mbuf *m) { struct vtnet_softc *sc; struct vtnet_txq *txq; int i, npairs, error; sc = ifp->if_softc; npairs = sc->vtnet_act_vq_pairs; /* check if flowid is set */ if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) i = m->m_pkthdr.flowid % npairs; else i = curcpu % npairs; txq = &sc->vtnet_txqs[i]; if (VTNET_TXQ_TRYLOCK(txq) != 0) { error = vtnet_txq_mq_start_locked(txq, m); VTNET_TXQ_UNLOCK(txq); } else { error = drbr_enqueue(ifp, txq->vtntx_br, m); taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_defrtask); } return (error); } static void vtnet_txq_tq_deferred(void *xtxq, int pending) { struct vtnet_softc *sc; struct vtnet_txq *txq; txq = xtxq; sc = txq->vtntx_sc; VTNET_TXQ_LOCK(txq); if (!drbr_empty(sc->vtnet_ifp, txq->vtntx_br)) vtnet_txq_mq_start_locked(txq, NULL); VTNET_TXQ_UNLOCK(txq); } #endif /* VTNET_LEGACY_TX */ static void vtnet_txq_start(struct vtnet_txq *txq) { struct vtnet_softc *sc; struct ifnet *ifp; sc = txq->vtntx_sc; ifp = sc->vtnet_ifp; #ifdef VTNET_LEGACY_TX if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) vtnet_start_locked(txq, ifp); #else if (!drbr_empty(ifp, txq->vtntx_br)) vtnet_txq_mq_start_locked(txq, NULL); #endif } static void vtnet_txq_tq_intr(void *xtxq, int pending) { struct vtnet_softc *sc; struct vtnet_txq *txq; struct ifnet *ifp; txq = xtxq; sc = txq->vtntx_sc; ifp = sc->vtnet_ifp; VTNET_TXQ_LOCK(txq); if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { VTNET_TXQ_UNLOCK(txq); return; } vtnet_txq_eof(txq); vtnet_txq_start(txq); VTNET_TXQ_UNLOCK(txq); } static int vtnet_txq_eof(struct vtnet_txq *txq) { struct virtqueue *vq; struct vtnet_tx_header *txhdr; struct mbuf *m; int deq; vq = txq->vtntx_vq; deq = 0; VTNET_TXQ_LOCK_ASSERT(txq); while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) { m = txhdr->vth_mbuf; deq++; txq->vtntx_stats.vtxs_opackets++; txq->vtntx_stats.vtxs_obytes += m->m_pkthdr.len; if (m->m_flags & M_MCAST) txq->vtntx_stats.vtxs_omcasts++; m_freem(m); uma_zfree(vtnet_tx_header_zone, txhdr); } if (virtqueue_empty(vq)) txq->vtntx_watchdog = 0; return (deq); } static void vtnet_tx_vq_intr(void *xtxq) { struct vtnet_softc *sc; struct vtnet_txq *txq; struct ifnet *ifp; txq = xtxq; sc = txq->vtntx_sc; ifp = sc->vtnet_ifp; if (__predict_false(txq->vtntx_id >= sc->vtnet_act_vq_pairs)) { /* * Ignore this interrupt. Either this is a spurious interrupt * or multiqueue without per-VQ MSIX so every queue needs to * be polled (a brain dead configuration we could try harder * to avoid). */ vtnet_txq_disable_intr(txq); return; } #ifdef DEV_NETMAP if (netmap_tx_irq(ifp, txq->vtntx_id) != NM_IRQ_PASS) return; #endif /* DEV_NETMAP */ VTNET_TXQ_LOCK(txq); if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { VTNET_TXQ_UNLOCK(txq); return; } vtnet_txq_eof(txq); vtnet_txq_start(txq); VTNET_TXQ_UNLOCK(txq); } static void vtnet_tx_start_all(struct vtnet_softc *sc) { struct vtnet_txq *txq; int i; VTNET_CORE_LOCK_ASSERT(sc); for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { txq = &sc->vtnet_txqs[i]; VTNET_TXQ_LOCK(txq); vtnet_txq_start(txq); VTNET_TXQ_UNLOCK(txq); } } #ifndef VTNET_LEGACY_TX static void vtnet_qflush(struct ifnet *ifp) { struct vtnet_softc *sc; struct vtnet_txq *txq; struct mbuf *m; int i; sc = ifp->if_softc; for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { txq = &sc->vtnet_txqs[i]; VTNET_TXQ_LOCK(txq); while ((m = buf_ring_dequeue_sc(txq->vtntx_br)) != NULL) m_freem(m); VTNET_TXQ_UNLOCK(txq); } if_qflush(ifp); } #endif static int vtnet_watchdog(struct vtnet_txq *txq) { struct ifnet *ifp; ifp = txq->vtntx_sc->vtnet_ifp; VTNET_TXQ_LOCK(txq); if (txq->vtntx_watchdog == 1) { /* * Only drain completed frames if the watchdog is about to * expire. If any frames were drained, there may be enough * free descriptors now available to transmit queued frames. * In that case, the timer will immediately be decremented * below, but the timeout is generous enough that should not * be a problem. */ if (vtnet_txq_eof(txq) != 0) vtnet_txq_start(txq); } if (txq->vtntx_watchdog == 0 || --txq->vtntx_watchdog) { VTNET_TXQ_UNLOCK(txq); return (0); } VTNET_TXQ_UNLOCK(txq); if_printf(ifp, "watchdog timeout on queue %d\n", txq->vtntx_id); return (1); } static void vtnet_accum_stats(struct vtnet_softc *sc, struct vtnet_rxq_stats *rxacc, struct vtnet_txq_stats *txacc) { bzero(rxacc, sizeof(struct vtnet_rxq_stats)); bzero(txacc, sizeof(struct vtnet_txq_stats)); for (int i = 0; i < sc->vtnet_max_vq_pairs; i++) { struct vtnet_rxq_stats *rxst; struct vtnet_txq_stats *txst; rxst = &sc->vtnet_rxqs[i].vtnrx_stats; rxacc->vrxs_ipackets += rxst->vrxs_ipackets; rxacc->vrxs_ibytes += rxst->vrxs_ibytes; rxacc->vrxs_iqdrops += rxst->vrxs_iqdrops; rxacc->vrxs_csum += rxst->vrxs_csum; rxacc->vrxs_csum_failed += rxst->vrxs_csum_failed; rxacc->vrxs_rescheduled += rxst->vrxs_rescheduled; txst = &sc->vtnet_txqs[i].vtntx_stats; txacc->vtxs_opackets += txst->vtxs_opackets; txacc->vtxs_obytes += txst->vtxs_obytes; txacc->vtxs_csum += txst->vtxs_csum; txacc->vtxs_tso += txst->vtxs_tso; txacc->vtxs_rescheduled += txst->vtxs_rescheduled; } } static uint64_t vtnet_get_counter(if_t ifp, ift_counter cnt) { struct vtnet_softc *sc; struct vtnet_rxq_stats rxaccum; struct vtnet_txq_stats txaccum; sc = if_getsoftc(ifp); vtnet_accum_stats(sc, &rxaccum, &txaccum); switch (cnt) { case IFCOUNTER_IPACKETS: return (rxaccum.vrxs_ipackets); case IFCOUNTER_IQDROPS: return (rxaccum.vrxs_iqdrops); case IFCOUNTER_IERRORS: return (rxaccum.vrxs_ierrors); case IFCOUNTER_OPACKETS: return (txaccum.vtxs_opackets); #ifndef VTNET_LEGACY_TX case IFCOUNTER_OBYTES: return (txaccum.vtxs_obytes); case IFCOUNTER_OMCASTS: return (txaccum.vtxs_omcasts); #endif default: return (if_get_counter_default(ifp, cnt)); } } static void vtnet_tick(void *xsc) { struct vtnet_softc *sc; struct ifnet *ifp; int i, timedout; sc = xsc; ifp = sc->vtnet_ifp; timedout = 0; VTNET_CORE_LOCK_ASSERT(sc); for (i = 0; i < sc->vtnet_act_vq_pairs; i++) timedout |= vtnet_watchdog(&sc->vtnet_txqs[i]); if (timedout != 0) { ifp->if_drv_flags &= ~IFF_DRV_RUNNING; vtnet_init_locked(sc); } else callout_schedule(&sc->vtnet_tick_ch, hz); } static void vtnet_start_taskqueues(struct vtnet_softc *sc) { device_t dev; struct vtnet_rxq *rxq; struct vtnet_txq *txq; int i, error; dev = sc->vtnet_dev; /* * Errors here are very difficult to recover from - we cannot * easily fail because, if this is during boot, we will hang * when freeing any successfully started taskqueues because * the scheduler isn't up yet. * * Most drivers just ignore the return value - it only fails * with ENOMEM so an error is not likely. */ for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { rxq = &sc->vtnet_rxqs[i]; error = taskqueue_start_threads(&rxq->vtnrx_tq, 1, PI_NET, "%s rxq %d", device_get_nameunit(dev), rxq->vtnrx_id); if (error) { device_printf(dev, "failed to start rx taskq %d\n", rxq->vtnrx_id); } txq = &sc->vtnet_txqs[i]; error = taskqueue_start_threads(&txq->vtntx_tq, 1, PI_NET, "%s txq %d", device_get_nameunit(dev), txq->vtntx_id); if (error) { device_printf(dev, "failed to start tx taskq %d\n", txq->vtntx_id); } } } static void vtnet_free_taskqueues(struct vtnet_softc *sc) { struct vtnet_rxq *rxq; struct vtnet_txq *txq; int i; for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { rxq = &sc->vtnet_rxqs[i]; if (rxq->vtnrx_tq != NULL) { taskqueue_free(rxq->vtnrx_tq); rxq->vtnrx_tq = NULL; } txq = &sc->vtnet_txqs[i]; if (txq->vtntx_tq != NULL) { taskqueue_free(txq->vtntx_tq); txq->vtntx_tq = NULL; } } } static void vtnet_drain_taskqueues(struct vtnet_softc *sc) { struct vtnet_rxq *rxq; struct vtnet_txq *txq; int i; for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { rxq = &sc->vtnet_rxqs[i]; if (rxq->vtnrx_tq != NULL) taskqueue_drain(rxq->vtnrx_tq, &rxq->vtnrx_intrtask); txq = &sc->vtnet_txqs[i]; if (txq->vtntx_tq != NULL) { taskqueue_drain(txq->vtntx_tq, &txq->vtntx_intrtask); #ifndef VTNET_LEGACY_TX taskqueue_drain(txq->vtntx_tq, &txq->vtntx_defrtask); #endif } } } static void vtnet_drain_rxtx_queues(struct vtnet_softc *sc) { struct vtnet_rxq *rxq; struct vtnet_txq *txq; int i; for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { rxq = &sc->vtnet_rxqs[i]; vtnet_rxq_free_mbufs(rxq); txq = &sc->vtnet_txqs[i]; vtnet_txq_free_mbufs(txq); } } static void vtnet_stop_rendezvous(struct vtnet_softc *sc) { struct vtnet_rxq *rxq; struct vtnet_txq *txq; int i; /* * Lock and unlock the per-queue mutex so we known the stop * state is visible. Doing only the active queues should be * sufficient, but it does not cost much extra to do all the * queues. Note we hold the core mutex here too. */ for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { rxq = &sc->vtnet_rxqs[i]; VTNET_RXQ_LOCK(rxq); VTNET_RXQ_UNLOCK(rxq); txq = &sc->vtnet_txqs[i]; VTNET_TXQ_LOCK(txq); VTNET_TXQ_UNLOCK(txq); } } static void vtnet_stop(struct vtnet_softc *sc) { device_t dev; struct ifnet *ifp; dev = sc->vtnet_dev; ifp = sc->vtnet_ifp; VTNET_CORE_LOCK_ASSERT(sc); ifp->if_drv_flags &= ~IFF_DRV_RUNNING; sc->vtnet_link_active = 0; callout_stop(&sc->vtnet_tick_ch); /* Only advisory. */ vtnet_disable_interrupts(sc); /* * Stop the host adapter. This resets it to the pre-initialized * state. It will not generate any interrupts until after it is * reinitialized. */ virtio_stop(dev); vtnet_stop_rendezvous(sc); /* Free any mbufs left in the virtqueues. */ vtnet_drain_rxtx_queues(sc); } static int vtnet_virtio_reinit(struct vtnet_softc *sc) { device_t dev; struct ifnet *ifp; uint64_t features; int mask, error; dev = sc->vtnet_dev; ifp = sc->vtnet_ifp; features = sc->vtnet_features; mask = 0; #if defined(INET) mask |= IFCAP_RXCSUM; #endif #if defined (INET6) mask |= IFCAP_RXCSUM_IPV6; #endif /* * Re-negotiate with the host, removing any disabled receive * features. Transmit features are disabled only on our side * via if_capenable and if_hwassist. */ if (ifp->if_capabilities & mask) { /* * We require both IPv4 and IPv6 offloading to be enabled * in order to negotiated it: VirtIO does not distinguish * between the two. */ if ((ifp->if_capenable & mask) != mask) features &= ~VIRTIO_NET_F_GUEST_CSUM; } if (ifp->if_capabilities & IFCAP_LRO) { if ((ifp->if_capenable & IFCAP_LRO) == 0) features &= ~VTNET_LRO_FEATURES; } if (ifp->if_capabilities & IFCAP_VLAN_HWFILTER) { if ((ifp->if_capenable & IFCAP_VLAN_HWFILTER) == 0) features &= ~VIRTIO_NET_F_CTRL_VLAN; } error = virtio_reinit(dev, features); if (error) device_printf(dev, "virtio reinit error %d\n", error); return (error); } static void vtnet_init_rx_filters(struct vtnet_softc *sc) { struct ifnet *ifp; ifp = sc->vtnet_ifp; if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) { /* Restore promiscuous and all-multicast modes. */ vtnet_rx_filter(sc); /* Restore filtered MAC addresses. */ vtnet_rx_filter_mac(sc); } if (ifp->if_capenable & IFCAP_VLAN_HWFILTER) vtnet_rx_filter_vlan(sc); } static int vtnet_init_rx_queues(struct vtnet_softc *sc) { device_t dev; struct vtnet_rxq *rxq; int i, clsize, error; dev = sc->vtnet_dev; /* * Use the new cluster size if one has been set (via a MTU * change). Otherwise, use the standard 2K clusters. * * BMV: It might make sense to use page sized clusters as * the default (depending on the features negotiated). */ if (sc->vtnet_rx_new_clsize != 0) { clsize = sc->vtnet_rx_new_clsize; sc->vtnet_rx_new_clsize = 0; } else clsize = MCLBYTES; sc->vtnet_rx_clsize = clsize; sc->vtnet_rx_nmbufs = VTNET_NEEDED_RX_MBUFS(sc, clsize); KASSERT(sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS || sc->vtnet_rx_nmbufs < sc->vtnet_rx_nsegs, ("%s: too many rx mbufs %d for %d segments", __func__, sc->vtnet_rx_nmbufs, sc->vtnet_rx_nsegs)); for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { rxq = &sc->vtnet_rxqs[i]; /* Hold the lock to satisfy asserts. */ VTNET_RXQ_LOCK(rxq); error = vtnet_rxq_populate(rxq); VTNET_RXQ_UNLOCK(rxq); if (error) { device_printf(dev, "cannot allocate mbufs for Rx queue %d\n", i); return (error); } } return (0); } static int vtnet_init_tx_queues(struct vtnet_softc *sc) { struct vtnet_txq *txq; int i; for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { txq = &sc->vtnet_txqs[i]; txq->vtntx_watchdog = 0; } return (0); } static int vtnet_init_rxtx_queues(struct vtnet_softc *sc) { int error; error = vtnet_init_rx_queues(sc); if (error) return (error); error = vtnet_init_tx_queues(sc); if (error) return (error); return (0); } static void vtnet_set_active_vq_pairs(struct vtnet_softc *sc) { device_t dev; int npairs; dev = sc->vtnet_dev; if ((sc->vtnet_flags & VTNET_FLAG_MULTIQ) == 0) { sc->vtnet_act_vq_pairs = 1; return; } npairs = sc->vtnet_requested_vq_pairs; if (vtnet_ctrl_mq_cmd(sc, npairs) != 0) { device_printf(dev, "cannot set active queue pairs to %d\n", npairs); npairs = 1; } sc->vtnet_act_vq_pairs = npairs; } static int vtnet_reinit(struct vtnet_softc *sc) { struct ifnet *ifp; int error; ifp = sc->vtnet_ifp; /* Use the current MAC address. */ bcopy(IF_LLADDR(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN); vtnet_set_hwaddr(sc); vtnet_set_active_vq_pairs(sc); ifp->if_hwassist = 0; if (ifp->if_capenable & IFCAP_TXCSUM) ifp->if_hwassist |= VTNET_CSUM_OFFLOAD; if (ifp->if_capenable & IFCAP_TXCSUM_IPV6) ifp->if_hwassist |= VTNET_CSUM_OFFLOAD_IPV6; if (ifp->if_capenable & IFCAP_TSO4) ifp->if_hwassist |= CSUM_IP_TSO; if (ifp->if_capenable & IFCAP_TSO6) ifp->if_hwassist |= CSUM_IP6_TSO; if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) vtnet_init_rx_filters(sc); error = vtnet_init_rxtx_queues(sc); if (error) return (error); vtnet_enable_interrupts(sc); ifp->if_drv_flags |= IFF_DRV_RUNNING; return (0); } static void vtnet_init_locked(struct vtnet_softc *sc) { device_t dev; struct ifnet *ifp; dev = sc->vtnet_dev; ifp = sc->vtnet_ifp; VTNET_CORE_LOCK_ASSERT(sc); if (ifp->if_drv_flags & IFF_DRV_RUNNING) return; vtnet_stop(sc); /* Reinitialize with the host. */ if (vtnet_virtio_reinit(sc) != 0) goto fail; if (vtnet_reinit(sc) != 0) goto fail; virtio_reinit_complete(dev); vtnet_update_link_status(sc); callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc); return; fail: vtnet_stop(sc); } static void vtnet_init(void *xsc) { struct vtnet_softc *sc; sc = xsc; VTNET_CORE_LOCK(sc); vtnet_init_locked(sc); VTNET_CORE_UNLOCK(sc); } static void vtnet_free_ctrl_vq(struct vtnet_softc *sc) { struct virtqueue *vq; vq = sc->vtnet_ctrl_vq; /* * The control virtqueue is only polled and therefore it should * already be empty. */ KASSERT(virtqueue_empty(vq), ("%s: ctrl vq %p not empty", __func__, vq)); } static void vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie, struct sglist *sg, int readable, int writable) { struct virtqueue *vq; vq = sc->vtnet_ctrl_vq; VTNET_CORE_LOCK_ASSERT(sc); KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ, ("%s: CTRL_VQ feature not negotiated", __func__)); if (!virtqueue_empty(vq)) return; if (virtqueue_enqueue(vq, cookie, sg, readable, writable) != 0) return; /* * Poll for the response, but the command is likely already * done when we return from the notify. */ virtqueue_notify(vq); virtqueue_poll(vq, NULL); } static int vtnet_ctrl_mac_cmd(struct vtnet_softc *sc, uint8_t *hwaddr) { struct virtio_net_ctrl_hdr hdr __aligned(2); struct sglist_seg segs[3]; struct sglist sg; uint8_t ack; int error; hdr.class = VIRTIO_NET_CTRL_MAC; hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET; ack = VIRTIO_NET_ERR; sglist_init(&sg, 3, segs); error = 0; error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr)); error |= sglist_append(&sg, hwaddr, ETHER_ADDR_LEN); error |= sglist_append(&sg, &ack, sizeof(uint8_t)); KASSERT(error == 0 && sg.sg_nseg == 3, ("%s: error %d adding set MAC msg to sglist", __func__, error)); vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1); return (ack == VIRTIO_NET_OK ? 0 : EIO); } static int vtnet_ctrl_mq_cmd(struct vtnet_softc *sc, uint16_t npairs) { struct sglist_seg segs[3]; struct sglist sg; struct { struct virtio_net_ctrl_hdr hdr; uint8_t pad1; struct virtio_net_ctrl_mq mq; uint8_t pad2; uint8_t ack; } s __aligned(2); int error; s.hdr.class = VIRTIO_NET_CTRL_MQ; s.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET; s.mq.virtqueue_pairs = npairs; s.ack = VIRTIO_NET_ERR; sglist_init(&sg, 3, segs); error = 0; error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr)); error |= sglist_append(&sg, &s.mq, sizeof(struct virtio_net_ctrl_mq)); error |= sglist_append(&sg, &s.ack, sizeof(uint8_t)); KASSERT(error == 0 && sg.sg_nseg == 3, ("%s: error %d adding MQ message to sglist", __func__, error)); vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1); return (s.ack == VIRTIO_NET_OK ? 0 : EIO); } static int vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, int cmd, int on) { struct sglist_seg segs[3]; struct sglist sg; struct { struct virtio_net_ctrl_hdr hdr; uint8_t pad1; uint8_t onoff; uint8_t pad2; uint8_t ack; } s __aligned(2); int error; KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX, ("%s: CTRL_RX feature not negotiated", __func__)); s.hdr.class = VIRTIO_NET_CTRL_RX; s.hdr.cmd = cmd; s.onoff = !!on; s.ack = VIRTIO_NET_ERR; sglist_init(&sg, 3, segs); error = 0; error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr)); error |= sglist_append(&sg, &s.onoff, sizeof(uint8_t)); error |= sglist_append(&sg, &s.ack, sizeof(uint8_t)); KASSERT(error == 0 && sg.sg_nseg == 3, ("%s: error %d adding Rx message to sglist", __func__, error)); vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1); return (s.ack == VIRTIO_NET_OK ? 0 : EIO); } static int vtnet_set_promisc(struct vtnet_softc *sc, int on) { return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on)); } static int vtnet_set_allmulti(struct vtnet_softc *sc, int on) { return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on)); } /* * The device defaults to promiscuous mode for backwards compatibility. * Turn it off at attach time if possible. */ static void vtnet_attach_disable_promisc(struct vtnet_softc *sc) { struct ifnet *ifp; ifp = sc->vtnet_ifp; VTNET_CORE_LOCK(sc); if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0) { ifp->if_flags |= IFF_PROMISC; } else if (vtnet_set_promisc(sc, 0) != 0) { ifp->if_flags |= IFF_PROMISC; device_printf(sc->vtnet_dev, "cannot disable default promiscuous mode\n"); } VTNET_CORE_UNLOCK(sc); } static void vtnet_rx_filter(struct vtnet_softc *sc) { device_t dev; struct ifnet *ifp; dev = sc->vtnet_dev; ifp = sc->vtnet_ifp; VTNET_CORE_LOCK_ASSERT(sc); if (vtnet_set_promisc(sc, ifp->if_flags & IFF_PROMISC) != 0) device_printf(dev, "cannot %s promiscuous mode\n", ifp->if_flags & IFF_PROMISC ? "enable" : "disable"); if (vtnet_set_allmulti(sc, ifp->if_flags & IFF_ALLMULTI) != 0) device_printf(dev, "cannot %s all-multicast mode\n", ifp->if_flags & IFF_ALLMULTI ? "enable" : "disable"); } static void vtnet_rx_filter_mac(struct vtnet_softc *sc) { struct virtio_net_ctrl_hdr hdr __aligned(2); struct vtnet_mac_filter *filter; struct sglist_seg segs[4]; struct sglist sg; struct ifnet *ifp; struct ifaddr *ifa; struct ifmultiaddr *ifma; int ucnt, mcnt, promisc, allmulti, error; uint8_t ack; ifp = sc->vtnet_ifp; filter = sc->vtnet_mac_filter; ucnt = 0; mcnt = 0; promisc = 0; allmulti = 0; VTNET_CORE_LOCK_ASSERT(sc); KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX, ("%s: CTRL_RX feature not negotiated", __func__)); /* Unicast MAC addresses: */ if_addr_rlock(ifp); CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { if (ifa->ifa_addr->sa_family != AF_LINK) continue; else if (memcmp(LLADDR((struct sockaddr_dl *)ifa->ifa_addr), sc->vtnet_hwaddr, ETHER_ADDR_LEN) == 0) continue; else if (ucnt == VTNET_MAX_MAC_ENTRIES) { promisc = 1; break; } bcopy(LLADDR((struct sockaddr_dl *)ifa->ifa_addr), &filter->vmf_unicast.macs[ucnt], ETHER_ADDR_LEN); ucnt++; } if_addr_runlock(ifp); if (promisc != 0) { filter->vmf_unicast.nentries = 0; if_printf(ifp, "more than %d MAC addresses assigned, " "falling back to promiscuous mode\n", VTNET_MAX_MAC_ENTRIES); } else filter->vmf_unicast.nentries = ucnt; /* Multicast MAC addresses: */ if_maddr_rlock(ifp); CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { if (ifma->ifma_addr->sa_family != AF_LINK) continue; else if (mcnt == VTNET_MAX_MAC_ENTRIES) { allmulti = 1; break; } bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr), &filter->vmf_multicast.macs[mcnt], ETHER_ADDR_LEN); mcnt++; } if_maddr_runlock(ifp); if (allmulti != 0) { filter->vmf_multicast.nentries = 0; if_printf(ifp, "more than %d multicast MAC addresses " "assigned, falling back to all-multicast mode\n", VTNET_MAX_MAC_ENTRIES); } else filter->vmf_multicast.nentries = mcnt; if (promisc != 0 && allmulti != 0) goto out; hdr.class = VIRTIO_NET_CTRL_MAC; hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET; ack = VIRTIO_NET_ERR; sglist_init(&sg, 4, segs); error = 0; error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr)); error |= sglist_append(&sg, &filter->vmf_unicast, sizeof(uint32_t) + filter->vmf_unicast.nentries * ETHER_ADDR_LEN); error |= sglist_append(&sg, &filter->vmf_multicast, sizeof(uint32_t) + filter->vmf_multicast.nentries * ETHER_ADDR_LEN); error |= sglist_append(&sg, &ack, sizeof(uint8_t)); KASSERT(error == 0 && sg.sg_nseg == 4, ("%s: error %d adding MAC filter msg to sglist", __func__, error)); vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1); if (ack != VIRTIO_NET_OK) if_printf(ifp, "error setting host MAC filter table\n"); out: if (promisc != 0 && vtnet_set_promisc(sc, 1) != 0) if_printf(ifp, "cannot enable promiscuous mode\n"); if (allmulti != 0 && vtnet_set_allmulti(sc, 1) != 0) if_printf(ifp, "cannot enable all-multicast mode\n"); } static int vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag) { struct sglist_seg segs[3]; struct sglist sg; struct { struct virtio_net_ctrl_hdr hdr; uint8_t pad1; uint16_t tag; uint8_t pad2; uint8_t ack; } s __aligned(2); int error; s.hdr.class = VIRTIO_NET_CTRL_VLAN; s.hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL; s.tag = tag; s.ack = VIRTIO_NET_ERR; sglist_init(&sg, 3, segs); error = 0; error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr)); error |= sglist_append(&sg, &s.tag, sizeof(uint16_t)); error |= sglist_append(&sg, &s.ack, sizeof(uint8_t)); KASSERT(error == 0 && sg.sg_nseg == 3, ("%s: error %d adding VLAN message to sglist", __func__, error)); vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1); return (s.ack == VIRTIO_NET_OK ? 0 : EIO); } static void vtnet_rx_filter_vlan(struct vtnet_softc *sc) { uint32_t w; uint16_t tag; int i, bit; VTNET_CORE_LOCK_ASSERT(sc); KASSERT(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER, ("%s: VLAN_FILTER feature not negotiated", __func__)); /* Enable the filter for each configured VLAN. */ for (i = 0; i < VTNET_VLAN_FILTER_NWORDS; i++) { w = sc->vtnet_vlan_filter[i]; while ((bit = ffs(w) - 1) != -1) { w &= ~(1 << bit); tag = sizeof(w) * CHAR_BIT * i + bit; if (vtnet_exec_vlan_filter(sc, 1, tag) != 0) { device_printf(sc->vtnet_dev, "cannot enable VLAN %d filter\n", tag); } } } } static void vtnet_update_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag) { struct ifnet *ifp; int idx, bit; ifp = sc->vtnet_ifp; idx = (tag >> 5) & 0x7F; bit = tag & 0x1F; if (tag == 0 || tag > 4095) return; VTNET_CORE_LOCK(sc); if (add) sc->vtnet_vlan_filter[idx] |= (1 << bit); else sc->vtnet_vlan_filter[idx] &= ~(1 << bit); if (ifp->if_capenable & IFCAP_VLAN_HWFILTER && ifp->if_drv_flags & IFF_DRV_RUNNING && vtnet_exec_vlan_filter(sc, add, tag) != 0) { device_printf(sc->vtnet_dev, "cannot %s VLAN %d %s the host filter table\n", add ? "add" : "remove", tag, add ? "to" : "from"); } VTNET_CORE_UNLOCK(sc); } static void vtnet_register_vlan(void *arg, struct ifnet *ifp, uint16_t tag) { if (ifp->if_softc != arg) return; vtnet_update_vlan_filter(arg, 1, tag); } static void vtnet_unregister_vlan(void *arg, struct ifnet *ifp, uint16_t tag) { if (ifp->if_softc != arg) return; vtnet_update_vlan_filter(arg, 0, tag); } static int vtnet_is_link_up(struct vtnet_softc *sc) { device_t dev; struct ifnet *ifp; uint16_t status; dev = sc->vtnet_dev; ifp = sc->vtnet_ifp; if ((ifp->if_capabilities & IFCAP_LINKSTATE) == 0) status = VIRTIO_NET_S_LINK_UP; else status = virtio_read_dev_config_2(dev, offsetof(struct virtio_net_config, status)); return ((status & VIRTIO_NET_S_LINK_UP) != 0); } static void vtnet_update_link_status(struct vtnet_softc *sc) { struct ifnet *ifp; int link; ifp = sc->vtnet_ifp; VTNET_CORE_LOCK_ASSERT(sc); link = vtnet_is_link_up(sc); /* Notify if the link status has changed. */ if (link != 0 && sc->vtnet_link_active == 0) { sc->vtnet_link_active = 1; if_link_state_change(ifp, LINK_STATE_UP); } else if (link == 0 && sc->vtnet_link_active != 0) { sc->vtnet_link_active = 0; if_link_state_change(ifp, LINK_STATE_DOWN); } } static int vtnet_ifmedia_upd(struct ifnet *ifp) { struct vtnet_softc *sc; struct ifmedia *ifm; sc = ifp->if_softc; ifm = &sc->vtnet_media; if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER) return (EINVAL); return (0); } static void vtnet_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr) { struct vtnet_softc *sc; sc = ifp->if_softc; ifmr->ifm_status = IFM_AVALID; ifmr->ifm_active = IFM_ETHER; VTNET_CORE_LOCK(sc); if (vtnet_is_link_up(sc) != 0) { ifmr->ifm_status |= IFM_ACTIVE; ifmr->ifm_active |= VTNET_MEDIATYPE; } else ifmr->ifm_active |= IFM_NONE; VTNET_CORE_UNLOCK(sc); } static void vtnet_set_hwaddr(struct vtnet_softc *sc) { device_t dev; int i; dev = sc->vtnet_dev; if (sc->vtnet_flags & VTNET_FLAG_CTRL_MAC) { if (vtnet_ctrl_mac_cmd(sc, sc->vtnet_hwaddr) != 0) device_printf(dev, "unable to set MAC address\n"); } else if (sc->vtnet_flags & VTNET_FLAG_MAC) { for (i = 0; i < ETHER_ADDR_LEN; i++) { virtio_write_dev_config_1(dev, offsetof(struct virtio_net_config, mac) + i, sc->vtnet_hwaddr[i]); } } } static void vtnet_get_hwaddr(struct vtnet_softc *sc) { device_t dev; int i; dev = sc->vtnet_dev; if ((sc->vtnet_flags & VTNET_FLAG_MAC) == 0) { /* * Generate a random locally administered unicast address. * * It would be nice to generate the same MAC address across * reboots, but it seems all the hosts currently available * support the MAC feature, so this isn't too important. */ sc->vtnet_hwaddr[0] = 0xB2; arc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1, 0); vtnet_set_hwaddr(sc); return; } for (i = 0; i < ETHER_ADDR_LEN; i++) { sc->vtnet_hwaddr[i] = virtio_read_dev_config_1(dev, offsetof(struct virtio_net_config, mac) + i); } } static void vtnet_vlan_tag_remove(struct mbuf *m) { struct ether_vlan_header *evh; evh = mtod(m, struct ether_vlan_header *); m->m_pkthdr.ether_vtag = ntohs(evh->evl_tag); m->m_flags |= M_VLANTAG; /* Strip the 802.1Q header. */ bcopy((char *) evh, (char *) evh + ETHER_VLAN_ENCAP_LEN, ETHER_HDR_LEN - ETHER_TYPE_LEN); m_adj(m, ETHER_VLAN_ENCAP_LEN); } static void vtnet_set_rx_process_limit(struct vtnet_softc *sc) { int limit; limit = vtnet_tunable_int(sc, "rx_process_limit", vtnet_rx_process_limit); if (limit < 0) limit = INT_MAX; sc->vtnet_rx_process_limit = limit; } static void vtnet_set_tx_intr_threshold(struct vtnet_softc *sc) { int size, thresh; size = virtqueue_size(sc->vtnet_txqs[0].vtntx_vq); /* * The Tx interrupt is disabled until the queue free count falls * below our threshold. Completed frames are drained from the Tx * virtqueue before transmitting new frames and in the watchdog * callout, so the frequency of Tx interrupts is greatly reduced, * at the cost of not freeing mbufs as quickly as they otherwise * would be. * * N.B. We assume all the Tx queues are the same size. */ thresh = size / 4; /* * Without indirect descriptors, leave enough room for the most * segments we handle. */ if ((sc->vtnet_flags & VTNET_FLAG_INDIRECT) == 0 && thresh < sc->vtnet_tx_nsegs) thresh = sc->vtnet_tx_nsegs; sc->vtnet_tx_intr_thresh = thresh; } static void vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *ctx, struct sysctl_oid_list *child, struct vtnet_rxq *rxq) { struct sysctl_oid *node; struct sysctl_oid_list *list; struct vtnet_rxq_stats *stats; char namebuf[16]; snprintf(namebuf, sizeof(namebuf), "rxq%d", rxq->vtnrx_id); node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf, CTLFLAG_RD, NULL, "Receive Queue"); list = SYSCTL_CHILDREN(node); stats = &rxq->vtnrx_stats; SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ipackets", CTLFLAG_RD, &stats->vrxs_ipackets, "Receive packets"); SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ibytes", CTLFLAG_RD, &stats->vrxs_ibytes, "Receive bytes"); SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "iqdrops", CTLFLAG_RD, &stats->vrxs_iqdrops, "Receive drops"); SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ierrors", CTLFLAG_RD, &stats->vrxs_ierrors, "Receive errors"); SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD, &stats->vrxs_csum, "Receive checksum offloaded"); SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum_failed", CTLFLAG_RD, &stats->vrxs_csum_failed, "Receive checksum offload failed"); SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD, &stats->vrxs_rescheduled, "Receive interrupt handler rescheduled"); } static void vtnet_setup_txq_sysctl(struct sysctl_ctx_list *ctx, struct sysctl_oid_list *child, struct vtnet_txq *txq) { struct sysctl_oid *node; struct sysctl_oid_list *list; struct vtnet_txq_stats *stats; char namebuf[16]; snprintf(namebuf, sizeof(namebuf), "txq%d", txq->vtntx_id); node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf, CTLFLAG_RD, NULL, "Transmit Queue"); list = SYSCTL_CHILDREN(node); stats = &txq->vtntx_stats; SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "opackets", CTLFLAG_RD, &stats->vtxs_opackets, "Transmit packets"); SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "obytes", CTLFLAG_RD, &stats->vtxs_obytes, "Transmit bytes"); SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "omcasts", CTLFLAG_RD, &stats->vtxs_omcasts, "Transmit multicasts"); SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD, &stats->vtxs_csum, "Transmit checksum offloaded"); SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "tso", CTLFLAG_RD, &stats->vtxs_tso, "Transmit segmentation offloaded"); SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD, &stats->vtxs_rescheduled, "Transmit interrupt handler rescheduled"); } static void vtnet_setup_queue_sysctl(struct vtnet_softc *sc) { device_t dev; struct sysctl_ctx_list *ctx; struct sysctl_oid *tree; struct sysctl_oid_list *child; int i; dev = sc->vtnet_dev; ctx = device_get_sysctl_ctx(dev); tree = device_get_sysctl_tree(dev); child = SYSCTL_CHILDREN(tree); for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { vtnet_setup_rxq_sysctl(ctx, child, &sc->vtnet_rxqs[i]); vtnet_setup_txq_sysctl(ctx, child, &sc->vtnet_txqs[i]); } } static void vtnet_setup_stat_sysctl(struct sysctl_ctx_list *ctx, struct sysctl_oid_list *child, struct vtnet_softc *sc) { struct vtnet_statistics *stats; struct vtnet_rxq_stats rxaccum; struct vtnet_txq_stats txaccum; vtnet_accum_stats(sc, &rxaccum, &txaccum); stats = &sc->vtnet_stats; stats->rx_csum_offloaded = rxaccum.vrxs_csum; stats->rx_csum_failed = rxaccum.vrxs_csum_failed; stats->rx_task_rescheduled = rxaccum.vrxs_rescheduled; stats->tx_csum_offloaded = txaccum.vtxs_csum; stats->tx_tso_offloaded = txaccum.vtxs_tso; stats->tx_task_rescheduled = txaccum.vtxs_rescheduled; SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "mbuf_alloc_failed", CTLFLAG_RD, &stats->mbuf_alloc_failed, "Mbuf cluster allocation failures"); SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_frame_too_large", CTLFLAG_RD, &stats->rx_frame_too_large, "Received frame larger than the mbuf chain"); SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_enq_replacement_failed", CTLFLAG_RD, &stats->rx_enq_replacement_failed, "Enqueuing the replacement receive mbuf failed"); SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_mergeable_failed", CTLFLAG_RD, &stats->rx_mergeable_failed, "Mergeable buffers receive failures"); SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ethtype", CTLFLAG_RD, &stats->rx_csum_bad_ethtype, "Received checksum offloaded buffer with unsupported " "Ethernet type"); SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ipproto", CTLFLAG_RD, &stats->rx_csum_bad_ipproto, "Received checksum offloaded buffer with incorrect IP protocol"); SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_offset", CTLFLAG_RD, &stats->rx_csum_bad_offset, "Received checksum offloaded buffer with incorrect offset"); SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_proto", CTLFLAG_RD, &stats->rx_csum_bad_proto, "Received checksum offloaded buffer with incorrect protocol"); SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_failed", CTLFLAG_RD, &stats->rx_csum_failed, "Received buffer checksum offload failed"); SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_offloaded", CTLFLAG_RD, &stats->rx_csum_offloaded, "Received buffer checksum offload succeeded"); SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_task_rescheduled", CTLFLAG_RD, &stats->rx_task_rescheduled, "Times the receive interrupt task rescheduled itself"); SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_bad_ethtype", CTLFLAG_RD, &stats->tx_csum_bad_ethtype, "Aborted transmit of checksum offloaded buffer with unknown " "Ethernet type"); SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_bad_ethtype", CTLFLAG_RD, &stats->tx_tso_bad_ethtype, "Aborted transmit of TSO buffer with unknown Ethernet type"); SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_not_tcp", CTLFLAG_RD, &stats->tx_tso_not_tcp, "Aborted transmit of TSO buffer with non TCP protocol"); SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defragged", CTLFLAG_RD, &stats->tx_defragged, "Transmit mbufs defragged"); SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defrag_failed", CTLFLAG_RD, &stats->tx_defrag_failed, "Aborted transmit of buffer because defrag failed"); SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_offloaded", CTLFLAG_RD, &stats->tx_csum_offloaded, "Offloaded checksum of transmitted buffer"); SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_offloaded", CTLFLAG_RD, &stats->tx_tso_offloaded, "Segmentation offload of transmitted buffer"); SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_task_rescheduled", CTLFLAG_RD, &stats->tx_task_rescheduled, "Times the transmit interrupt task rescheduled itself"); } static void vtnet_setup_sysctl(struct vtnet_softc *sc) { device_t dev; struct sysctl_ctx_list *ctx; struct sysctl_oid *tree; struct sysctl_oid_list *child; dev = sc->vtnet_dev; ctx = device_get_sysctl_ctx(dev); tree = device_get_sysctl_tree(dev); child = SYSCTL_CHILDREN(tree); SYSCTL_ADD_INT(ctx, child, OID_AUTO, "max_vq_pairs", CTLFLAG_RD, &sc->vtnet_max_vq_pairs, 0, "Maximum number of supported virtqueue pairs"); SYSCTL_ADD_INT(ctx, child, OID_AUTO, "requested_vq_pairs", CTLFLAG_RD, &sc->vtnet_requested_vq_pairs, 0, "Requested number of virtqueue pairs"); SYSCTL_ADD_INT(ctx, child, OID_AUTO, "act_vq_pairs", CTLFLAG_RD, &sc->vtnet_act_vq_pairs, 0, "Number of active virtqueue pairs"); vtnet_setup_stat_sysctl(ctx, child, sc); } static int vtnet_rxq_enable_intr(struct vtnet_rxq *rxq) { return (virtqueue_enable_intr(rxq->vtnrx_vq)); } static void vtnet_rxq_disable_intr(struct vtnet_rxq *rxq) { virtqueue_disable_intr(rxq->vtnrx_vq); } static int vtnet_txq_enable_intr(struct vtnet_txq *txq) { struct virtqueue *vq; vq = txq->vtntx_vq; if (vtnet_txq_below_threshold(txq) != 0) return (virtqueue_postpone_intr(vq, VQ_POSTPONE_LONG)); /* * The free count is above our threshold. Keep the Tx interrupt * disabled until the queue is fuller. */ return (0); } static void vtnet_txq_disable_intr(struct vtnet_txq *txq) { virtqueue_disable_intr(txq->vtntx_vq); } static void vtnet_enable_rx_interrupts(struct vtnet_softc *sc) { int i; for (i = 0; i < sc->vtnet_act_vq_pairs; i++) vtnet_rxq_enable_intr(&sc->vtnet_rxqs[i]); } static void vtnet_enable_tx_interrupts(struct vtnet_softc *sc) { int i; for (i = 0; i < sc->vtnet_act_vq_pairs; i++) vtnet_txq_enable_intr(&sc->vtnet_txqs[i]); } static void vtnet_enable_interrupts(struct vtnet_softc *sc) { vtnet_enable_rx_interrupts(sc); vtnet_enable_tx_interrupts(sc); } static void vtnet_disable_rx_interrupts(struct vtnet_softc *sc) { int i; for (i = 0; i < sc->vtnet_act_vq_pairs; i++) vtnet_rxq_disable_intr(&sc->vtnet_rxqs[i]); } static void vtnet_disable_tx_interrupts(struct vtnet_softc *sc) { int i; for (i = 0; i < sc->vtnet_act_vq_pairs; i++) vtnet_txq_disable_intr(&sc->vtnet_txqs[i]); } static void vtnet_disable_interrupts(struct vtnet_softc *sc) { vtnet_disable_rx_interrupts(sc); vtnet_disable_tx_interrupts(sc); } static int vtnet_tunable_int(struct vtnet_softc *sc, const char *knob, int def) { char path[64]; snprintf(path, sizeof(path), "hw.vtnet.%d.%s", device_get_unit(sc->vtnet_dev), knob); TUNABLE_INT_FETCH(path, &def); return (def); } #ifdef NETDUMP static void vtnet_netdump_init(struct ifnet *ifp, int *nrxr, int *ncl, int *clsize) { struct vtnet_softc *sc; sc = if_getsoftc(ifp); VTNET_CORE_LOCK(sc); *nrxr = sc->vtnet_max_vq_pairs; *ncl = NETDUMP_MAX_IN_FLIGHT; *clsize = sc->vtnet_rx_clsize; VTNET_CORE_UNLOCK(sc); /* * We need to allocate from this zone in the transmit path, so ensure * that we have at least one item per header available. * XXX add a separate zone like we do for mbufs? otherwise we may alloc * buckets */ uma_zone_reserve(vtnet_tx_header_zone, NETDUMP_MAX_IN_FLIGHT * 2); uma_prealloc(vtnet_tx_header_zone, NETDUMP_MAX_IN_FLIGHT * 2); } static void vtnet_netdump_event(struct ifnet *ifp __unused, enum netdump_ev event __unused) { } static int vtnet_netdump_transmit(struct ifnet *ifp, struct mbuf *m) { struct vtnet_softc *sc; struct vtnet_txq *txq; int error; sc = if_getsoftc(ifp); if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) != IFF_DRV_RUNNING) return (EBUSY); txq = &sc->vtnet_txqs[0]; error = vtnet_txq_encap(txq, &m, M_NOWAIT | M_USE_RESERVE); if (error == 0) (void)vtnet_txq_notify(txq); return (error); } static int vtnet_netdump_poll(struct ifnet *ifp, int count) { struct vtnet_softc *sc; int i; sc = if_getsoftc(ifp); if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) != IFF_DRV_RUNNING) return (EBUSY); (void)vtnet_txq_eof(&sc->vtnet_txqs[0]); for (i = 0; i < sc->vtnet_max_vq_pairs; i++) (void)vtnet_rxq_eof(&sc->vtnet_rxqs[i]); return (0); } #endif /* NETDUMP */ Index: head/sys/dev/virtio/random/virtio_random.c =================================================================== --- head/sys/dev/virtio/random/virtio_random.c (revision 348598) +++ head/sys/dev/virtio/random/virtio_random.c (revision 348599) @@ -1,275 +1,273 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2013, Bryan Venteicher * 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. */ /* Driver for VirtIO entropy device. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include struct vtrnd_softc { uint64_t vtrnd_features; struct virtqueue *vtrnd_vq; }; static int vtrnd_modevent(module_t, int, void *); static int vtrnd_probe(device_t); static int vtrnd_attach(device_t); static int vtrnd_detach(device_t); static void vtrnd_negotiate_features(device_t); static int vtrnd_alloc_virtqueue(device_t); static int vtrnd_harvest(struct vtrnd_softc *, void *, size_t *); static unsigned vtrnd_read(void *, unsigned); #define VTRND_FEATURES 0 static struct virtio_feature_desc vtrnd_feature_desc[] = { { 0, NULL } }; static struct random_source random_vtrnd = { .rs_ident = "VirtIO Entropy Adapter", .rs_source = RANDOM_PURE_VIRTIO, .rs_read = vtrnd_read, }; /* Kludge for API limitations of random(4). */ static _Atomic(struct vtrnd_softc *) g_vtrnd_softc; static device_method_t vtrnd_methods[] = { /* Device methods. */ DEVMETHOD(device_probe, vtrnd_probe), DEVMETHOD(device_attach, vtrnd_attach), DEVMETHOD(device_detach, vtrnd_detach), DEVMETHOD_END }; static driver_t vtrnd_driver = { "vtrnd", vtrnd_methods, sizeof(struct vtrnd_softc) }; static devclass_t vtrnd_devclass; DRIVER_MODULE(virtio_random, virtio_pci, vtrnd_driver, vtrnd_devclass, vtrnd_modevent, 0); MODULE_VERSION(virtio_random, 1); MODULE_DEPEND(virtio_random, virtio, 1, 1, 1); MODULE_DEPEND(virtio_random, random_device, 1, 1, 1); +VIRTIO_SIMPLE_PNPTABLE(virtio_random, VIRTIO_ID_ENTROPY, + "VirtIO Entropy Adapter"); +VIRTIO_SIMPLE_PNPINFO(virtio_pci, virtio_random); + static int vtrnd_modevent(module_t mod, int type, void *unused) { int error; switch (type) { case MOD_LOAD: case MOD_QUIESCE: case MOD_UNLOAD: case MOD_SHUTDOWN: error = 0; break; default: error = EOPNOTSUPP; break; } return (error); } static int vtrnd_probe(device_t dev) { - - if (virtio_get_device_type(dev) != VIRTIO_ID_ENTROPY) - return (ENXIO); - - device_set_desc(dev, "VirtIO Entropy Adapter"); - - return (BUS_PROBE_DEFAULT); + return (VIRTIO_SIMPLE_PROBE(dev, virtio_random)); } static int vtrnd_attach(device_t dev) { struct vtrnd_softc *sc, *exp; int error; sc = device_get_softc(dev); virtio_set_feature_desc(dev, vtrnd_feature_desc); vtrnd_negotiate_features(dev); error = vtrnd_alloc_virtqueue(dev); if (error) { device_printf(dev, "cannot allocate virtqueue\n"); goto fail; } exp = NULL; if (!atomic_compare_exchange_strong_explicit(&g_vtrnd_softc, &exp, sc, memory_order_release, memory_order_acquire)) { error = EEXIST; goto fail; } random_source_register(&random_vtrnd); fail: if (error) vtrnd_detach(dev); return (error); } static int vtrnd_detach(device_t dev) { struct vtrnd_softc *sc; sc = device_get_softc(dev); KASSERT( atomic_load_explicit(&g_vtrnd_softc, memory_order_acquire) == sc, ("only one global instance at a time")); random_source_deregister(&random_vtrnd); atomic_store_explicit(&g_vtrnd_softc, NULL, memory_order_release); /* * Unfortunately, deregister does not guarantee our source callback * will not be invoked after it returns. Use a kludge to prevent some, * but not all, possible races. */ tsleep_sbt(&g_vtrnd_softc, 0, "vtrnddet", mstosbt(50), 0, C_HARDCLOCK); return (0); } static void vtrnd_negotiate_features(device_t dev) { struct vtrnd_softc *sc; sc = device_get_softc(dev); sc->vtrnd_features = virtio_negotiate_features(dev, VTRND_FEATURES); } static int vtrnd_alloc_virtqueue(device_t dev) { struct vtrnd_softc *sc; struct vq_alloc_info vq_info; sc = device_get_softc(dev); VQ_ALLOC_INFO_INIT(&vq_info, 0, NULL, sc, &sc->vtrnd_vq, "%s request", device_get_nameunit(dev)); return (virtio_alloc_virtqueues(dev, 0, 1, &vq_info)); } static int vtrnd_harvest(struct vtrnd_softc *sc, void *buf, size_t *sz) { struct sglist_seg segs[1]; struct sglist sg; struct virtqueue *vq; uint32_t value[HARVESTSIZE] __aligned(sizeof(uint32_t) * HARVESTSIZE); uint32_t rdlen; int error; _Static_assert(sizeof(value) < PAGE_SIZE, "sglist assumption"); sglist_init(&sg, 1, segs); error = sglist_append(&sg, value, *sz); if (error != 0) panic("%s: sglist_append error=%d", __func__, error); vq = sc->vtrnd_vq; KASSERT(virtqueue_empty(vq), ("%s: non-empty queue", __func__)); error = virtqueue_enqueue(vq, buf, &sg, 0, 1); if (error != 0) return (error); /* * Poll for the response, but the command is likely already * done when we return from the notify. */ virtqueue_notify(vq); virtqueue_poll(vq, &rdlen); if (rdlen > *sz) panic("%s: random device wrote %zu bytes beyond end of provided" " buffer %p:%zu", __func__, (size_t)rdlen - *sz, (void *)value, *sz); else if (rdlen == 0) return (EAGAIN); *sz = MIN(rdlen, *sz); memcpy(buf, value, *sz); explicit_bzero(value, *sz); return (0); } static unsigned vtrnd_read(void *buf, unsigned usz) { struct vtrnd_softc *sc; size_t sz; int error; sc = g_vtrnd_softc; if (sc == NULL) return (0); sz = usz; error = vtrnd_harvest(sc, buf, &sz); if (error != 0) return (0); return (sz); } Index: head/sys/dev/virtio/scsi/virtio_scsi.c =================================================================== --- head/sys/dev/virtio/scsi/virtio_scsi.c (revision 348598) +++ head/sys/dev/virtio/scsi/virtio_scsi.c (revision 348599) @@ -1,2318 +1,2315 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2012, Bryan Venteicher * 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. */ /* Driver for VirtIO SCSI devices. */ #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 #include #include "virtio_if.h" static int vtscsi_modevent(module_t, int, void *); static int vtscsi_probe(device_t); static int vtscsi_attach(device_t); static int vtscsi_detach(device_t); static int vtscsi_suspend(device_t); static int vtscsi_resume(device_t); static void vtscsi_negotiate_features(struct vtscsi_softc *); static void vtscsi_read_config(struct vtscsi_softc *, struct virtio_scsi_config *); static int vtscsi_maximum_segments(struct vtscsi_softc *, int); static int vtscsi_alloc_virtqueues(struct vtscsi_softc *); static void vtscsi_write_device_config(struct vtscsi_softc *); static int vtscsi_reinit(struct vtscsi_softc *); static int vtscsi_alloc_cam(struct vtscsi_softc *); static int vtscsi_register_cam(struct vtscsi_softc *); static void vtscsi_free_cam(struct vtscsi_softc *); static void vtscsi_cam_async(void *, uint32_t, struct cam_path *, void *); static int vtscsi_register_async(struct vtscsi_softc *); static void vtscsi_deregister_async(struct vtscsi_softc *); static void vtscsi_cam_action(struct cam_sim *, union ccb *); static void vtscsi_cam_poll(struct cam_sim *); static void vtscsi_cam_scsi_io(struct vtscsi_softc *, struct cam_sim *, union ccb *); static void vtscsi_cam_get_tran_settings(struct vtscsi_softc *, union ccb *); static void vtscsi_cam_reset_bus(struct vtscsi_softc *, union ccb *); static void vtscsi_cam_reset_dev(struct vtscsi_softc *, union ccb *); static void vtscsi_cam_abort(struct vtscsi_softc *, union ccb *); static void vtscsi_cam_path_inquiry(struct vtscsi_softc *, struct cam_sim *, union ccb *); static int vtscsi_sg_append_scsi_buf(struct vtscsi_softc *, struct sglist *, struct ccb_scsiio *); static int vtscsi_fill_scsi_cmd_sglist(struct vtscsi_softc *, struct vtscsi_request *, int *, int *); static int vtscsi_execute_scsi_cmd(struct vtscsi_softc *, struct vtscsi_request *); static int vtscsi_start_scsi_cmd(struct vtscsi_softc *, union ccb *); static void vtscsi_complete_abort_timedout_scsi_cmd(struct vtscsi_softc *, struct vtscsi_request *); static int vtscsi_abort_timedout_scsi_cmd(struct vtscsi_softc *, struct vtscsi_request *); static void vtscsi_timedout_scsi_cmd(void *); static cam_status vtscsi_scsi_cmd_cam_status(struct virtio_scsi_cmd_resp *); static cam_status vtscsi_complete_scsi_cmd_response(struct vtscsi_softc *, struct ccb_scsiio *, struct virtio_scsi_cmd_resp *); static void vtscsi_complete_scsi_cmd(struct vtscsi_softc *, struct vtscsi_request *); static void vtscsi_poll_ctrl_req(struct vtscsi_softc *, struct vtscsi_request *); static int vtscsi_execute_ctrl_req(struct vtscsi_softc *, struct vtscsi_request *, struct sglist *, int, int, int); static void vtscsi_complete_abort_task_cmd(struct vtscsi_softc *c, struct vtscsi_request *); static int vtscsi_execute_abort_task_cmd(struct vtscsi_softc *, struct vtscsi_request *); static int vtscsi_execute_reset_dev_cmd(struct vtscsi_softc *, struct vtscsi_request *); static void vtscsi_get_request_lun(uint8_t [], target_id_t *, lun_id_t *); static void vtscsi_set_request_lun(struct ccb_hdr *, uint8_t []); static void vtscsi_init_scsi_cmd_req(struct ccb_scsiio *, struct virtio_scsi_cmd_req *); static void vtscsi_init_ctrl_tmf_req(struct ccb_hdr *, uint32_t, uintptr_t, struct virtio_scsi_ctrl_tmf_req *); static void vtscsi_freeze_simq(struct vtscsi_softc *, int); static int vtscsi_thaw_simq(struct vtscsi_softc *, int); static void vtscsi_announce(struct vtscsi_softc *, uint32_t, target_id_t, lun_id_t); static void vtscsi_execute_rescan(struct vtscsi_softc *, target_id_t, lun_id_t); static void vtscsi_execute_rescan_bus(struct vtscsi_softc *); static void vtscsi_handle_event(struct vtscsi_softc *, struct virtio_scsi_event *); static int vtscsi_enqueue_event_buf(struct vtscsi_softc *, struct virtio_scsi_event *); static int vtscsi_init_event_vq(struct vtscsi_softc *); static void vtscsi_reinit_event_vq(struct vtscsi_softc *); static void vtscsi_drain_event_vq(struct vtscsi_softc *); static void vtscsi_complete_vqs_locked(struct vtscsi_softc *); static void vtscsi_complete_vqs(struct vtscsi_softc *); static void vtscsi_drain_vqs(struct vtscsi_softc *); static void vtscsi_cancel_request(struct vtscsi_softc *, struct vtscsi_request *); static void vtscsi_drain_vq(struct vtscsi_softc *, struct virtqueue *); static void vtscsi_stop(struct vtscsi_softc *); static int vtscsi_reset_bus(struct vtscsi_softc *); static void vtscsi_init_request(struct vtscsi_softc *, struct vtscsi_request *); static int vtscsi_alloc_requests(struct vtscsi_softc *); static void vtscsi_free_requests(struct vtscsi_softc *); static void vtscsi_enqueue_request(struct vtscsi_softc *, struct vtscsi_request *); static struct vtscsi_request * vtscsi_dequeue_request(struct vtscsi_softc *); static void vtscsi_complete_request(struct vtscsi_request *); static void vtscsi_complete_vq(struct vtscsi_softc *, struct virtqueue *); static void vtscsi_control_vq_intr(void *); static void vtscsi_event_vq_intr(void *); static void vtscsi_request_vq_intr(void *); static void vtscsi_disable_vqs_intr(struct vtscsi_softc *); static void vtscsi_enable_vqs_intr(struct vtscsi_softc *); static void vtscsi_get_tunables(struct vtscsi_softc *); static void vtscsi_add_sysctl(struct vtscsi_softc *); static void vtscsi_printf_req(struct vtscsi_request *, const char *, const char *, ...); /* Global tunables. */ /* * The current QEMU VirtIO SCSI implementation does not cancel in-flight * IO during virtio_stop(). So in-flight requests still complete after the * device reset. We would have to wait for all the in-flight IO to complete, * which defeats the typical purpose of a bus reset. We could simulate the * bus reset with either I_T_NEXUS_RESET of all the targets, or with * LOGICAL_UNIT_RESET of all the LUNs (assuming there is space in the * control virtqueue). But this isn't very useful if things really go off * the rails, so default to disabled for now. */ static int vtscsi_bus_reset_disable = 1; TUNABLE_INT("hw.vtscsi.bus_reset_disable", &vtscsi_bus_reset_disable); static struct virtio_feature_desc vtscsi_feature_desc[] = { { VIRTIO_SCSI_F_INOUT, "InOut" }, { VIRTIO_SCSI_F_HOTPLUG, "Hotplug" }, { 0, NULL } }; static device_method_t vtscsi_methods[] = { /* Device methods. */ DEVMETHOD(device_probe, vtscsi_probe), DEVMETHOD(device_attach, vtscsi_attach), DEVMETHOD(device_detach, vtscsi_detach), DEVMETHOD(device_suspend, vtscsi_suspend), DEVMETHOD(device_resume, vtscsi_resume), DEVMETHOD_END }; static driver_t vtscsi_driver = { "vtscsi", vtscsi_methods, sizeof(struct vtscsi_softc) }; static devclass_t vtscsi_devclass; DRIVER_MODULE(virtio_scsi, virtio_pci, vtscsi_driver, vtscsi_devclass, vtscsi_modevent, 0); MODULE_VERSION(virtio_scsi, 1); MODULE_DEPEND(virtio_scsi, virtio, 1, 1, 1); MODULE_DEPEND(virtio_scsi, cam, 1, 1, 1); +VIRTIO_SIMPLE_PNPTABLE(virtio_scsi, VIRTIO_ID_SCSI, "VirtIO SCSI Adapter"); +VIRTIO_SIMPLE_PNPINFO(virtio_pci, virtio_scsi); + static int vtscsi_modevent(module_t mod, int type, void *unused) { int error; switch (type) { case MOD_LOAD: case MOD_QUIESCE: case MOD_UNLOAD: case MOD_SHUTDOWN: error = 0; break; default: error = EOPNOTSUPP; break; } return (error); } static int vtscsi_probe(device_t dev) { - - if (virtio_get_device_type(dev) != VIRTIO_ID_SCSI) - return (ENXIO); - - device_set_desc(dev, "VirtIO SCSI Adapter"); - - return (BUS_PROBE_DEFAULT); + return (VIRTIO_SIMPLE_PROBE(dev, virtio_scsi)); } static int vtscsi_attach(device_t dev) { struct vtscsi_softc *sc; struct virtio_scsi_config scsicfg; int error; sc = device_get_softc(dev); sc->vtscsi_dev = dev; VTSCSI_LOCK_INIT(sc, device_get_nameunit(dev)); TAILQ_INIT(&sc->vtscsi_req_free); vtscsi_get_tunables(sc); vtscsi_add_sysctl(sc); virtio_set_feature_desc(dev, vtscsi_feature_desc); vtscsi_negotiate_features(sc); if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) sc->vtscsi_flags |= VTSCSI_FLAG_INDIRECT; if (virtio_with_feature(dev, VIRTIO_SCSI_F_INOUT)) sc->vtscsi_flags |= VTSCSI_FLAG_BIDIRECTIONAL; if (virtio_with_feature(dev, VIRTIO_SCSI_F_HOTPLUG)) sc->vtscsi_flags |= VTSCSI_FLAG_HOTPLUG; vtscsi_read_config(sc, &scsicfg); sc->vtscsi_max_channel = scsicfg.max_channel; sc->vtscsi_max_target = scsicfg.max_target; sc->vtscsi_max_lun = scsicfg.max_lun; sc->vtscsi_event_buf_size = scsicfg.event_info_size; vtscsi_write_device_config(sc); sc->vtscsi_max_nsegs = vtscsi_maximum_segments(sc, scsicfg.seg_max); sc->vtscsi_sglist = sglist_alloc(sc->vtscsi_max_nsegs, M_NOWAIT); if (sc->vtscsi_sglist == NULL) { error = ENOMEM; device_printf(dev, "cannot allocate sglist\n"); goto fail; } error = vtscsi_alloc_virtqueues(sc); if (error) { device_printf(dev, "cannot allocate virtqueues\n"); goto fail; } error = vtscsi_init_event_vq(sc); if (error) { device_printf(dev, "cannot populate the eventvq\n"); goto fail; } error = vtscsi_alloc_requests(sc); if (error) { device_printf(dev, "cannot allocate requests\n"); goto fail; } error = vtscsi_alloc_cam(sc); if (error) { device_printf(dev, "cannot allocate CAM structures\n"); goto fail; } error = virtio_setup_intr(dev, INTR_TYPE_CAM); if (error) { device_printf(dev, "cannot setup virtqueue interrupts\n"); goto fail; } vtscsi_enable_vqs_intr(sc); /* * Register with CAM after interrupts are enabled so we will get * notified of the probe responses. */ error = vtscsi_register_cam(sc); if (error) { device_printf(dev, "cannot register with CAM\n"); goto fail; } fail: if (error) vtscsi_detach(dev); return (error); } static int vtscsi_detach(device_t dev) { struct vtscsi_softc *sc; sc = device_get_softc(dev); VTSCSI_LOCK(sc); sc->vtscsi_flags |= VTSCSI_FLAG_DETACH; if (device_is_attached(dev)) vtscsi_stop(sc); VTSCSI_UNLOCK(sc); vtscsi_complete_vqs(sc); vtscsi_drain_vqs(sc); vtscsi_free_cam(sc); vtscsi_free_requests(sc); if (sc->vtscsi_sglist != NULL) { sglist_free(sc->vtscsi_sglist); sc->vtscsi_sglist = NULL; } VTSCSI_LOCK_DESTROY(sc); return (0); } static int vtscsi_suspend(device_t dev) { return (0); } static int vtscsi_resume(device_t dev) { return (0); } static void vtscsi_negotiate_features(struct vtscsi_softc *sc) { device_t dev; uint64_t features; dev = sc->vtscsi_dev; features = virtio_negotiate_features(dev, VTSCSI_FEATURES); sc->vtscsi_features = features; } #define VTSCSI_GET_CONFIG(_dev, _field, _cfg) \ virtio_read_device_config(_dev, \ offsetof(struct virtio_scsi_config, _field), \ &(_cfg)->_field, sizeof((_cfg)->_field)) \ static void vtscsi_read_config(struct vtscsi_softc *sc, struct virtio_scsi_config *scsicfg) { device_t dev; dev = sc->vtscsi_dev; bzero(scsicfg, sizeof(struct virtio_scsi_config)); VTSCSI_GET_CONFIG(dev, num_queues, scsicfg); VTSCSI_GET_CONFIG(dev, seg_max, scsicfg); VTSCSI_GET_CONFIG(dev, max_sectors, scsicfg); VTSCSI_GET_CONFIG(dev, cmd_per_lun, scsicfg); VTSCSI_GET_CONFIG(dev, event_info_size, scsicfg); VTSCSI_GET_CONFIG(dev, sense_size, scsicfg); VTSCSI_GET_CONFIG(dev, cdb_size, scsicfg); VTSCSI_GET_CONFIG(dev, max_channel, scsicfg); VTSCSI_GET_CONFIG(dev, max_target, scsicfg); VTSCSI_GET_CONFIG(dev, max_lun, scsicfg); } #undef VTSCSI_GET_CONFIG static int vtscsi_maximum_segments(struct vtscsi_softc *sc, int seg_max) { int nsegs; nsegs = VTSCSI_MIN_SEGMENTS; if (seg_max > 0) { nsegs += MIN(seg_max, MAXPHYS / PAGE_SIZE + 1); if (sc->vtscsi_flags & VTSCSI_FLAG_INDIRECT) nsegs = MIN(nsegs, VIRTIO_MAX_INDIRECT); } else nsegs += 1; return (nsegs); } static int vtscsi_alloc_virtqueues(struct vtscsi_softc *sc) { device_t dev; struct vq_alloc_info vq_info[3]; int nvqs; dev = sc->vtscsi_dev; nvqs = 3; VQ_ALLOC_INFO_INIT(&vq_info[0], 0, vtscsi_control_vq_intr, sc, &sc->vtscsi_control_vq, "%s control", device_get_nameunit(dev)); VQ_ALLOC_INFO_INIT(&vq_info[1], 0, vtscsi_event_vq_intr, sc, &sc->vtscsi_event_vq, "%s event", device_get_nameunit(dev)); VQ_ALLOC_INFO_INIT(&vq_info[2], sc->vtscsi_max_nsegs, vtscsi_request_vq_intr, sc, &sc->vtscsi_request_vq, "%s request", device_get_nameunit(dev)); return (virtio_alloc_virtqueues(dev, 0, nvqs, vq_info)); } static void vtscsi_write_device_config(struct vtscsi_softc *sc) { virtio_write_dev_config_4(sc->vtscsi_dev, offsetof(struct virtio_scsi_config, sense_size), VIRTIO_SCSI_SENSE_SIZE); /* * This is the size in the virtio_scsi_cmd_req structure. Note * this value (32) is larger than the maximum CAM CDB size (16). */ virtio_write_dev_config_4(sc->vtscsi_dev, offsetof(struct virtio_scsi_config, cdb_size), VIRTIO_SCSI_CDB_SIZE); } static int vtscsi_reinit(struct vtscsi_softc *sc) { device_t dev; int error; dev = sc->vtscsi_dev; error = virtio_reinit(dev, sc->vtscsi_features); if (error == 0) { vtscsi_write_device_config(sc); vtscsi_reinit_event_vq(sc); virtio_reinit_complete(dev); vtscsi_enable_vqs_intr(sc); } vtscsi_dprintf(sc, VTSCSI_TRACE, "error=%d\n", error); return (error); } static int vtscsi_alloc_cam(struct vtscsi_softc *sc) { device_t dev; struct cam_devq *devq; int openings; dev = sc->vtscsi_dev; openings = sc->vtscsi_nrequests - VTSCSI_RESERVED_REQUESTS; devq = cam_simq_alloc(openings); if (devq == NULL) { device_printf(dev, "cannot allocate SIM queue\n"); return (ENOMEM); } sc->vtscsi_sim = cam_sim_alloc(vtscsi_cam_action, vtscsi_cam_poll, "vtscsi", sc, device_get_unit(dev), VTSCSI_MTX(sc), 1, openings, devq); if (sc->vtscsi_sim == NULL) { cam_simq_free(devq); device_printf(dev, "cannot allocate SIM\n"); return (ENOMEM); } return (0); } static int vtscsi_register_cam(struct vtscsi_softc *sc) { device_t dev; int registered, error; dev = sc->vtscsi_dev; registered = 0; VTSCSI_LOCK(sc); if (xpt_bus_register(sc->vtscsi_sim, dev, 0) != CAM_SUCCESS) { error = ENOMEM; device_printf(dev, "cannot register XPT bus\n"); goto fail; } registered = 1; if (xpt_create_path(&sc->vtscsi_path, NULL, cam_sim_path(sc->vtscsi_sim), CAM_TARGET_WILDCARD, CAM_LUN_WILDCARD) != CAM_REQ_CMP) { error = ENOMEM; device_printf(dev, "cannot create bus path\n"); goto fail; } if (vtscsi_register_async(sc) != CAM_REQ_CMP) { error = EIO; device_printf(dev, "cannot register async callback\n"); goto fail; } VTSCSI_UNLOCK(sc); return (0); fail: if (sc->vtscsi_path != NULL) { xpt_free_path(sc->vtscsi_path); sc->vtscsi_path = NULL; } if (registered != 0) xpt_bus_deregister(cam_sim_path(sc->vtscsi_sim)); VTSCSI_UNLOCK(sc); return (error); } static void vtscsi_free_cam(struct vtscsi_softc *sc) { VTSCSI_LOCK(sc); if (sc->vtscsi_path != NULL) { vtscsi_deregister_async(sc); xpt_free_path(sc->vtscsi_path); sc->vtscsi_path = NULL; xpt_bus_deregister(cam_sim_path(sc->vtscsi_sim)); } if (sc->vtscsi_sim != NULL) { cam_sim_free(sc->vtscsi_sim, 1); sc->vtscsi_sim = NULL; } VTSCSI_UNLOCK(sc); } static void vtscsi_cam_async(void *cb_arg, uint32_t code, struct cam_path *path, void *arg) { struct cam_sim *sim; struct vtscsi_softc *sc; sim = cb_arg; sc = cam_sim_softc(sim); vtscsi_dprintf(sc, VTSCSI_TRACE, "code=%u\n", code); /* * TODO Once QEMU supports event reporting, we should * (un)subscribe to events here. */ switch (code) { case AC_FOUND_DEVICE: break; case AC_LOST_DEVICE: break; } } static int vtscsi_register_async(struct vtscsi_softc *sc) { struct ccb_setasync csa; xpt_setup_ccb(&csa.ccb_h, sc->vtscsi_path, 5); csa.ccb_h.func_code = XPT_SASYNC_CB; csa.event_enable = AC_LOST_DEVICE | AC_FOUND_DEVICE; csa.callback = vtscsi_cam_async; csa.callback_arg = sc->vtscsi_sim; xpt_action((union ccb *) &csa); return (csa.ccb_h.status); } static void vtscsi_deregister_async(struct vtscsi_softc *sc) { struct ccb_setasync csa; xpt_setup_ccb(&csa.ccb_h, sc->vtscsi_path, 5); csa.ccb_h.func_code = XPT_SASYNC_CB; csa.event_enable = 0; csa.callback = vtscsi_cam_async; csa.callback_arg = sc->vtscsi_sim; xpt_action((union ccb *) &csa); } static void vtscsi_cam_action(struct cam_sim *sim, union ccb *ccb) { struct vtscsi_softc *sc; struct ccb_hdr *ccbh; sc = cam_sim_softc(sim); ccbh = &ccb->ccb_h; VTSCSI_LOCK_OWNED(sc); if (sc->vtscsi_flags & VTSCSI_FLAG_DETACH) { /* * The VTSCSI_MTX is briefly dropped between setting * VTSCSI_FLAG_DETACH and deregistering with CAM, so * drop any CCBs that come in during that window. */ ccbh->status = CAM_NO_HBA; xpt_done(ccb); return; } switch (ccbh->func_code) { case XPT_SCSI_IO: vtscsi_cam_scsi_io(sc, sim, ccb); break; case XPT_SET_TRAN_SETTINGS: ccbh->status = CAM_FUNC_NOTAVAIL; xpt_done(ccb); break; case XPT_GET_TRAN_SETTINGS: vtscsi_cam_get_tran_settings(sc, ccb); break; case XPT_RESET_BUS: vtscsi_cam_reset_bus(sc, ccb); break; case XPT_RESET_DEV: vtscsi_cam_reset_dev(sc, ccb); break; case XPT_ABORT: vtscsi_cam_abort(sc, ccb); break; case XPT_CALC_GEOMETRY: cam_calc_geometry(&ccb->ccg, 1); xpt_done(ccb); break; case XPT_PATH_INQ: vtscsi_cam_path_inquiry(sc, sim, ccb); break; default: vtscsi_dprintf(sc, VTSCSI_ERROR, "invalid ccb=%p func=%#x\n", ccb, ccbh->func_code); ccbh->status = CAM_REQ_INVALID; xpt_done(ccb); break; } } static void vtscsi_cam_poll(struct cam_sim *sim) { struct vtscsi_softc *sc; sc = cam_sim_softc(sim); vtscsi_complete_vqs_locked(sc); } static void vtscsi_cam_scsi_io(struct vtscsi_softc *sc, struct cam_sim *sim, union ccb *ccb) { struct ccb_hdr *ccbh; struct ccb_scsiio *csio; int error; ccbh = &ccb->ccb_h; csio = &ccb->csio; if (csio->cdb_len > VIRTIO_SCSI_CDB_SIZE) { error = EINVAL; ccbh->status = CAM_REQ_INVALID; goto done; } if ((ccbh->flags & CAM_DIR_MASK) == CAM_DIR_BOTH && (sc->vtscsi_flags & VTSCSI_FLAG_BIDIRECTIONAL) == 0) { error = EINVAL; ccbh->status = CAM_REQ_INVALID; goto done; } error = vtscsi_start_scsi_cmd(sc, ccb); done: if (error) { vtscsi_dprintf(sc, VTSCSI_ERROR, "error=%d ccb=%p status=%#x\n", error, ccb, ccbh->status); xpt_done(ccb); } } static void vtscsi_cam_get_tran_settings(struct vtscsi_softc *sc, union ccb *ccb) { struct ccb_trans_settings *cts; struct ccb_trans_settings_scsi *scsi; cts = &ccb->cts; scsi = &cts->proto_specific.scsi; cts->protocol = PROTO_SCSI; cts->protocol_version = SCSI_REV_SPC3; cts->transport = XPORT_SAS; cts->transport_version = 0; scsi->valid = CTS_SCSI_VALID_TQ; scsi->flags = CTS_SCSI_FLAGS_TAG_ENB; ccb->ccb_h.status = CAM_REQ_CMP; xpt_done(ccb); } static void vtscsi_cam_reset_bus(struct vtscsi_softc *sc, union ccb *ccb) { int error; error = vtscsi_reset_bus(sc); if (error == 0) ccb->ccb_h.status = CAM_REQ_CMP; else ccb->ccb_h.status = CAM_REQ_CMP_ERR; vtscsi_dprintf(sc, VTSCSI_TRACE, "error=%d ccb=%p status=%#x\n", error, ccb, ccb->ccb_h.status); xpt_done(ccb); } static void vtscsi_cam_reset_dev(struct vtscsi_softc *sc, union ccb *ccb) { struct ccb_hdr *ccbh; struct vtscsi_request *req; int error; ccbh = &ccb->ccb_h; req = vtscsi_dequeue_request(sc); if (req == NULL) { error = EAGAIN; vtscsi_freeze_simq(sc, VTSCSI_REQUEST); goto fail; } req->vsr_ccb = ccb; error = vtscsi_execute_reset_dev_cmd(sc, req); if (error == 0) return; vtscsi_enqueue_request(sc, req); fail: vtscsi_dprintf(sc, VTSCSI_ERROR, "error=%d req=%p ccb=%p\n", error, req, ccb); if (error == EAGAIN) ccbh->status = CAM_RESRC_UNAVAIL; else ccbh->status = CAM_REQ_CMP_ERR; xpt_done(ccb); } static void vtscsi_cam_abort(struct vtscsi_softc *sc, union ccb *ccb) { struct vtscsi_request *req; struct ccb_hdr *ccbh; int error; ccbh = &ccb->ccb_h; req = vtscsi_dequeue_request(sc); if (req == NULL) { error = EAGAIN; vtscsi_freeze_simq(sc, VTSCSI_REQUEST); goto fail; } req->vsr_ccb = ccb; error = vtscsi_execute_abort_task_cmd(sc, req); if (error == 0) return; vtscsi_enqueue_request(sc, req); fail: vtscsi_dprintf(sc, VTSCSI_ERROR, "error=%d req=%p ccb=%p\n", error, req, ccb); if (error == EAGAIN) ccbh->status = CAM_RESRC_UNAVAIL; else ccbh->status = CAM_REQ_CMP_ERR; xpt_done(ccb); } static void vtscsi_cam_path_inquiry(struct vtscsi_softc *sc, struct cam_sim *sim, union ccb *ccb) { device_t dev; struct ccb_pathinq *cpi; dev = sc->vtscsi_dev; cpi = &ccb->cpi; vtscsi_dprintf(sc, VTSCSI_TRACE, "sim=%p ccb=%p\n", sim, ccb); cpi->version_num = 1; cpi->hba_inquiry = PI_TAG_ABLE; cpi->target_sprt = 0; cpi->hba_misc = PIM_SEQSCAN | PIM_UNMAPPED; if (vtscsi_bus_reset_disable != 0) cpi->hba_misc |= PIM_NOBUSRESET; cpi->hba_eng_cnt = 0; cpi->max_target = sc->vtscsi_max_target; cpi->max_lun = sc->vtscsi_max_lun; cpi->initiator_id = VTSCSI_INITIATOR_ID; strlcpy(cpi->sim_vid, "FreeBSD", SIM_IDLEN); strlcpy(cpi->hba_vid, "VirtIO", HBA_IDLEN); strlcpy(cpi->dev_name, cam_sim_name(sim), DEV_IDLEN); cpi->unit_number = cam_sim_unit(sim); cpi->bus_id = cam_sim_bus(sim); cpi->base_transfer_speed = 300000; cpi->protocol = PROTO_SCSI; cpi->protocol_version = SCSI_REV_SPC3; cpi->transport = XPORT_SAS; cpi->transport_version = 0; cpi->maxio = (sc->vtscsi_max_nsegs - VTSCSI_MIN_SEGMENTS - 1) * PAGE_SIZE; cpi->hba_vendor = virtio_get_vendor(dev); cpi->hba_device = virtio_get_device(dev); cpi->hba_subvendor = virtio_get_subvendor(dev); cpi->hba_subdevice = virtio_get_subdevice(dev); ccb->ccb_h.status = CAM_REQ_CMP; xpt_done(ccb); } static int vtscsi_sg_append_scsi_buf(struct vtscsi_softc *sc, struct sglist *sg, struct ccb_scsiio *csio) { struct ccb_hdr *ccbh; struct bus_dma_segment *dseg; int i, error; ccbh = &csio->ccb_h; error = 0; switch ((ccbh->flags & CAM_DATA_MASK)) { case CAM_DATA_VADDR: error = sglist_append(sg, csio->data_ptr, csio->dxfer_len); break; case CAM_DATA_PADDR: error = sglist_append_phys(sg, (vm_paddr_t)(vm_offset_t) csio->data_ptr, csio->dxfer_len); break; case CAM_DATA_SG: for (i = 0; i < csio->sglist_cnt && error == 0; i++) { dseg = &((struct bus_dma_segment *)csio->data_ptr)[i]; error = sglist_append(sg, (void *)(vm_offset_t) dseg->ds_addr, dseg->ds_len); } break; case CAM_DATA_SG_PADDR: for (i = 0; i < csio->sglist_cnt && error == 0; i++) { dseg = &((struct bus_dma_segment *)csio->data_ptr)[i]; error = sglist_append_phys(sg, (vm_paddr_t) dseg->ds_addr, dseg->ds_len); } break; case CAM_DATA_BIO: error = sglist_append_bio(sg, (struct bio *) csio->data_ptr); break; default: error = EINVAL; break; } return (error); } static int vtscsi_fill_scsi_cmd_sglist(struct vtscsi_softc *sc, struct vtscsi_request *req, int *readable, int *writable) { struct sglist *sg; struct ccb_hdr *ccbh; struct ccb_scsiio *csio; struct virtio_scsi_cmd_req *cmd_req; struct virtio_scsi_cmd_resp *cmd_resp; int error; sg = sc->vtscsi_sglist; csio = &req->vsr_ccb->csio; ccbh = &csio->ccb_h; cmd_req = &req->vsr_cmd_req; cmd_resp = &req->vsr_cmd_resp; sglist_reset(sg); sglist_append(sg, cmd_req, sizeof(struct virtio_scsi_cmd_req)); if ((ccbh->flags & CAM_DIR_MASK) == CAM_DIR_OUT) { error = vtscsi_sg_append_scsi_buf(sc, sg, csio); /* At least one segment must be left for the response. */ if (error || sg->sg_nseg == sg->sg_maxseg) goto fail; } *readable = sg->sg_nseg; sglist_append(sg, cmd_resp, sizeof(struct virtio_scsi_cmd_resp)); if ((ccbh->flags & CAM_DIR_MASK) == CAM_DIR_IN) { error = vtscsi_sg_append_scsi_buf(sc, sg, csio); if (error) goto fail; } *writable = sg->sg_nseg - *readable; vtscsi_dprintf(sc, VTSCSI_TRACE, "req=%p ccb=%p readable=%d " "writable=%d\n", req, ccbh, *readable, *writable); return (0); fail: /* * This should never happen unless maxio was incorrectly set. */ vtscsi_set_ccb_status(ccbh, CAM_REQ_TOO_BIG, 0); vtscsi_dprintf(sc, VTSCSI_ERROR, "error=%d req=%p ccb=%p " "nseg=%d maxseg=%d\n", error, req, ccbh, sg->sg_nseg, sg->sg_maxseg); return (EFBIG); } static int vtscsi_execute_scsi_cmd(struct vtscsi_softc *sc, struct vtscsi_request *req) { struct sglist *sg; struct virtqueue *vq; struct ccb_scsiio *csio; struct ccb_hdr *ccbh; struct virtio_scsi_cmd_req *cmd_req; struct virtio_scsi_cmd_resp *cmd_resp; int readable, writable, error; sg = sc->vtscsi_sglist; vq = sc->vtscsi_request_vq; csio = &req->vsr_ccb->csio; ccbh = &csio->ccb_h; cmd_req = &req->vsr_cmd_req; cmd_resp = &req->vsr_cmd_resp; vtscsi_init_scsi_cmd_req(csio, cmd_req); error = vtscsi_fill_scsi_cmd_sglist(sc, req, &readable, &writable); if (error) return (error); req->vsr_complete = vtscsi_complete_scsi_cmd; cmd_resp->response = -1; error = virtqueue_enqueue(vq, req, sg, readable, writable); if (error) { vtscsi_dprintf(sc, VTSCSI_ERROR, "enqueue error=%d req=%p ccb=%p\n", error, req, ccbh); ccbh->status = CAM_REQUEUE_REQ; vtscsi_freeze_simq(sc, VTSCSI_REQUEST_VQ); return (error); } ccbh->status |= CAM_SIM_QUEUED; ccbh->ccbh_vtscsi_req = req; virtqueue_notify(vq); if (ccbh->timeout != CAM_TIME_INFINITY) { req->vsr_flags |= VTSCSI_REQ_FLAG_TIMEOUT_SET; callout_reset_sbt(&req->vsr_callout, SBT_1MS * ccbh->timeout, 0, vtscsi_timedout_scsi_cmd, req, 0); } vtscsi_dprintf_req(req, VTSCSI_TRACE, "enqueued req=%p ccb=%p\n", req, ccbh); return (0); } static int vtscsi_start_scsi_cmd(struct vtscsi_softc *sc, union ccb *ccb) { struct vtscsi_request *req; int error; req = vtscsi_dequeue_request(sc); if (req == NULL) { ccb->ccb_h.status = CAM_REQUEUE_REQ; vtscsi_freeze_simq(sc, VTSCSI_REQUEST); return (ENOBUFS); } req->vsr_ccb = ccb; error = vtscsi_execute_scsi_cmd(sc, req); if (error) vtscsi_enqueue_request(sc, req); return (error); } static void vtscsi_complete_abort_timedout_scsi_cmd(struct vtscsi_softc *sc, struct vtscsi_request *req) { struct virtio_scsi_ctrl_tmf_resp *tmf_resp; struct vtscsi_request *to_req; uint8_t response; tmf_resp = &req->vsr_tmf_resp; response = tmf_resp->response; to_req = req->vsr_timedout_req; vtscsi_dprintf(sc, VTSCSI_TRACE, "req=%p to_req=%p response=%d\n", req, to_req, response); vtscsi_enqueue_request(sc, req); /* * The timedout request could have completed between when the * abort task was sent and when the host processed it. */ if (to_req->vsr_state != VTSCSI_REQ_STATE_TIMEDOUT) return; /* The timedout request was successfully aborted. */ if (response == VIRTIO_SCSI_S_FUNCTION_COMPLETE) return; /* Don't bother if the device is going away. */ if (sc->vtscsi_flags & VTSCSI_FLAG_DETACH) return; /* The timedout request will be aborted by the reset. */ if (sc->vtscsi_flags & VTSCSI_FLAG_RESET) return; vtscsi_reset_bus(sc); } static int vtscsi_abort_timedout_scsi_cmd(struct vtscsi_softc *sc, struct vtscsi_request *to_req) { struct sglist *sg; struct ccb_hdr *to_ccbh; struct vtscsi_request *req; struct virtio_scsi_ctrl_tmf_req *tmf_req; struct virtio_scsi_ctrl_tmf_resp *tmf_resp; int error; sg = sc->vtscsi_sglist; to_ccbh = &to_req->vsr_ccb->ccb_h; req = vtscsi_dequeue_request(sc); if (req == NULL) { error = ENOBUFS; goto fail; } tmf_req = &req->vsr_tmf_req; tmf_resp = &req->vsr_tmf_resp; vtscsi_init_ctrl_tmf_req(to_ccbh, VIRTIO_SCSI_T_TMF_ABORT_TASK, (uintptr_t) to_ccbh, tmf_req); sglist_reset(sg); sglist_append(sg, tmf_req, sizeof(struct virtio_scsi_ctrl_tmf_req)); sglist_append(sg, tmf_resp, sizeof(struct virtio_scsi_ctrl_tmf_resp)); req->vsr_timedout_req = to_req; req->vsr_complete = vtscsi_complete_abort_timedout_scsi_cmd; tmf_resp->response = -1; error = vtscsi_execute_ctrl_req(sc, req, sg, 1, 1, VTSCSI_EXECUTE_ASYNC); if (error == 0) return (0); vtscsi_enqueue_request(sc, req); fail: vtscsi_dprintf(sc, VTSCSI_ERROR, "error=%d req=%p " "timedout req=%p ccb=%p\n", error, req, to_req, to_ccbh); return (error); } static void vtscsi_timedout_scsi_cmd(void *xreq) { struct vtscsi_softc *sc; struct vtscsi_request *to_req; to_req = xreq; sc = to_req->vsr_softc; vtscsi_dprintf(sc, VTSCSI_INFO, "timedout req=%p ccb=%p state=%#x\n", to_req, to_req->vsr_ccb, to_req->vsr_state); /* Don't bother if the device is going away. */ if (sc->vtscsi_flags & VTSCSI_FLAG_DETACH) return; /* * Bail if the request is not in use. We likely raced when * stopping the callout handler or it has already been aborted. */ if (to_req->vsr_state != VTSCSI_REQ_STATE_INUSE || (to_req->vsr_flags & VTSCSI_REQ_FLAG_TIMEOUT_SET) == 0) return; /* * Complete the request queue in case the timedout request is * actually just pending. */ vtscsi_complete_vq(sc, sc->vtscsi_request_vq); if (to_req->vsr_state == VTSCSI_REQ_STATE_FREE) return; sc->vtscsi_stats.scsi_cmd_timeouts++; to_req->vsr_state = VTSCSI_REQ_STATE_TIMEDOUT; if (vtscsi_abort_timedout_scsi_cmd(sc, to_req) == 0) return; vtscsi_dprintf(sc, VTSCSI_ERROR, "resetting bus\n"); vtscsi_reset_bus(sc); } static cam_status vtscsi_scsi_cmd_cam_status(struct virtio_scsi_cmd_resp *cmd_resp) { cam_status status; switch (cmd_resp->response) { case VIRTIO_SCSI_S_OK: status = CAM_REQ_CMP; break; case VIRTIO_SCSI_S_OVERRUN: status = CAM_DATA_RUN_ERR; break; case VIRTIO_SCSI_S_ABORTED: status = CAM_REQ_ABORTED; break; case VIRTIO_SCSI_S_BAD_TARGET: status = CAM_SEL_TIMEOUT; break; case VIRTIO_SCSI_S_RESET: status = CAM_SCSI_BUS_RESET; break; case VIRTIO_SCSI_S_BUSY: status = CAM_SCSI_BUSY; break; case VIRTIO_SCSI_S_TRANSPORT_FAILURE: case VIRTIO_SCSI_S_TARGET_FAILURE: case VIRTIO_SCSI_S_NEXUS_FAILURE: status = CAM_SCSI_IT_NEXUS_LOST; break; default: /* VIRTIO_SCSI_S_FAILURE */ status = CAM_REQ_CMP_ERR; break; } return (status); } static cam_status vtscsi_complete_scsi_cmd_response(struct vtscsi_softc *sc, struct ccb_scsiio *csio, struct virtio_scsi_cmd_resp *cmd_resp) { cam_status status; csio->scsi_status = cmd_resp->status; csio->resid = cmd_resp->resid; if (csio->scsi_status == SCSI_STATUS_OK) status = CAM_REQ_CMP; else status = CAM_SCSI_STATUS_ERROR; if (cmd_resp->sense_len > 0) { status |= CAM_AUTOSNS_VALID; if (cmd_resp->sense_len < csio->sense_len) csio->sense_resid = csio->sense_len - cmd_resp->sense_len; else csio->sense_resid = 0; memcpy(&csio->sense_data, cmd_resp->sense, csio->sense_len - csio->sense_resid); } vtscsi_dprintf(sc, status == CAM_REQ_CMP ? VTSCSI_TRACE : VTSCSI_ERROR, "ccb=%p scsi_status=%#x resid=%u sense_resid=%u\n", csio, csio->scsi_status, csio->resid, csio->sense_resid); return (status); } static void vtscsi_complete_scsi_cmd(struct vtscsi_softc *sc, struct vtscsi_request *req) { struct ccb_hdr *ccbh; struct ccb_scsiio *csio; struct virtio_scsi_cmd_resp *cmd_resp; cam_status status; csio = &req->vsr_ccb->csio; ccbh = &csio->ccb_h; cmd_resp = &req->vsr_cmd_resp; KASSERT(ccbh->ccbh_vtscsi_req == req, ("ccb %p req mismatch %p/%p", ccbh, ccbh->ccbh_vtscsi_req, req)); if (req->vsr_flags & VTSCSI_REQ_FLAG_TIMEOUT_SET) callout_stop(&req->vsr_callout); status = vtscsi_scsi_cmd_cam_status(cmd_resp); if (status == CAM_REQ_ABORTED) { if (req->vsr_state == VTSCSI_REQ_STATE_TIMEDOUT) status = CAM_CMD_TIMEOUT; } else if (status == CAM_REQ_CMP) status = vtscsi_complete_scsi_cmd_response(sc, csio, cmd_resp); if ((status & CAM_STATUS_MASK) != CAM_REQ_CMP) { status |= CAM_DEV_QFRZN; xpt_freeze_devq(ccbh->path, 1); } if (vtscsi_thaw_simq(sc, VTSCSI_REQUEST | VTSCSI_REQUEST_VQ) != 0) status |= CAM_RELEASE_SIMQ; vtscsi_dprintf(sc, VTSCSI_TRACE, "req=%p ccb=%p status=%#x\n", req, ccbh, status); ccbh->status = status; xpt_done(req->vsr_ccb); vtscsi_enqueue_request(sc, req); } static void vtscsi_poll_ctrl_req(struct vtscsi_softc *sc, struct vtscsi_request *req) { /* XXX We probably shouldn't poll forever. */ req->vsr_flags |= VTSCSI_REQ_FLAG_POLLED; do vtscsi_complete_vq(sc, sc->vtscsi_control_vq); while ((req->vsr_flags & VTSCSI_REQ_FLAG_COMPLETE) == 0); req->vsr_flags &= ~VTSCSI_REQ_FLAG_POLLED; } static int vtscsi_execute_ctrl_req(struct vtscsi_softc *sc, struct vtscsi_request *req, struct sglist *sg, int readable, int writable, int flag) { struct virtqueue *vq; int error; vq = sc->vtscsi_control_vq; MPASS(flag == VTSCSI_EXECUTE_POLL || req->vsr_complete != NULL); error = virtqueue_enqueue(vq, req, sg, readable, writable); if (error) { /* * Return EAGAIN when the virtqueue does not have enough * descriptors available. */ if (error == ENOSPC || error == EMSGSIZE) error = EAGAIN; return (error); } virtqueue_notify(vq); if (flag == VTSCSI_EXECUTE_POLL) vtscsi_poll_ctrl_req(sc, req); return (0); } static void vtscsi_complete_abort_task_cmd(struct vtscsi_softc *sc, struct vtscsi_request *req) { union ccb *ccb; struct ccb_hdr *ccbh; struct virtio_scsi_ctrl_tmf_resp *tmf_resp; ccb = req->vsr_ccb; ccbh = &ccb->ccb_h; tmf_resp = &req->vsr_tmf_resp; switch (tmf_resp->response) { case VIRTIO_SCSI_S_FUNCTION_COMPLETE: ccbh->status = CAM_REQ_CMP; break; case VIRTIO_SCSI_S_FUNCTION_REJECTED: ccbh->status = CAM_UA_ABORT; break; default: ccbh->status = CAM_REQ_CMP_ERR; break; } xpt_done(ccb); vtscsi_enqueue_request(sc, req); } static int vtscsi_execute_abort_task_cmd(struct vtscsi_softc *sc, struct vtscsi_request *req) { struct sglist *sg; struct ccb_abort *cab; struct ccb_hdr *ccbh; struct ccb_hdr *abort_ccbh; struct vtscsi_request *abort_req; struct virtio_scsi_ctrl_tmf_req *tmf_req; struct virtio_scsi_ctrl_tmf_resp *tmf_resp; int error; sg = sc->vtscsi_sglist; cab = &req->vsr_ccb->cab; ccbh = &cab->ccb_h; tmf_req = &req->vsr_tmf_req; tmf_resp = &req->vsr_tmf_resp; /* CCB header and request that's to be aborted. */ abort_ccbh = &cab->abort_ccb->ccb_h; abort_req = abort_ccbh->ccbh_vtscsi_req; if (abort_ccbh->func_code != XPT_SCSI_IO || abort_req == NULL) { error = EINVAL; goto fail; } /* Only attempt to abort requests that could be in-flight. */ if (abort_req->vsr_state != VTSCSI_REQ_STATE_INUSE) { error = EALREADY; goto fail; } abort_req->vsr_state = VTSCSI_REQ_STATE_ABORTED; if (abort_req->vsr_flags & VTSCSI_REQ_FLAG_TIMEOUT_SET) callout_stop(&abort_req->vsr_callout); vtscsi_init_ctrl_tmf_req(ccbh, VIRTIO_SCSI_T_TMF_ABORT_TASK, (uintptr_t) abort_ccbh, tmf_req); sglist_reset(sg); sglist_append(sg, tmf_req, sizeof(struct virtio_scsi_ctrl_tmf_req)); sglist_append(sg, tmf_resp, sizeof(struct virtio_scsi_ctrl_tmf_resp)); req->vsr_complete = vtscsi_complete_abort_task_cmd; tmf_resp->response = -1; error = vtscsi_execute_ctrl_req(sc, req, sg, 1, 1, VTSCSI_EXECUTE_ASYNC); fail: vtscsi_dprintf(sc, VTSCSI_TRACE, "error=%d req=%p abort_ccb=%p " "abort_req=%p\n", error, req, abort_ccbh, abort_req); return (error); } static void vtscsi_complete_reset_dev_cmd(struct vtscsi_softc *sc, struct vtscsi_request *req) { union ccb *ccb; struct ccb_hdr *ccbh; struct virtio_scsi_ctrl_tmf_resp *tmf_resp; ccb = req->vsr_ccb; ccbh = &ccb->ccb_h; tmf_resp = &req->vsr_tmf_resp; vtscsi_dprintf(sc, VTSCSI_TRACE, "req=%p ccb=%p response=%d\n", req, ccb, tmf_resp->response); if (tmf_resp->response == VIRTIO_SCSI_S_FUNCTION_COMPLETE) { ccbh->status = CAM_REQ_CMP; vtscsi_announce(sc, AC_SENT_BDR, ccbh->target_id, ccbh->target_lun); } else ccbh->status = CAM_REQ_CMP_ERR; xpt_done(ccb); vtscsi_enqueue_request(sc, req); } static int vtscsi_execute_reset_dev_cmd(struct vtscsi_softc *sc, struct vtscsi_request *req) { struct sglist *sg; struct ccb_resetdev *crd; struct ccb_hdr *ccbh; struct virtio_scsi_ctrl_tmf_req *tmf_req; struct virtio_scsi_ctrl_tmf_resp *tmf_resp; uint32_t subtype; int error; sg = sc->vtscsi_sglist; crd = &req->vsr_ccb->crd; ccbh = &crd->ccb_h; tmf_req = &req->vsr_tmf_req; tmf_resp = &req->vsr_tmf_resp; if (ccbh->target_lun == CAM_LUN_WILDCARD) subtype = VIRTIO_SCSI_T_TMF_I_T_NEXUS_RESET; else subtype = VIRTIO_SCSI_T_TMF_LOGICAL_UNIT_RESET; vtscsi_init_ctrl_tmf_req(ccbh, subtype, 0, tmf_req); sglist_reset(sg); sglist_append(sg, tmf_req, sizeof(struct virtio_scsi_ctrl_tmf_req)); sglist_append(sg, tmf_resp, sizeof(struct virtio_scsi_ctrl_tmf_resp)); req->vsr_complete = vtscsi_complete_reset_dev_cmd; tmf_resp->response = -1; error = vtscsi_execute_ctrl_req(sc, req, sg, 1, 1, VTSCSI_EXECUTE_ASYNC); vtscsi_dprintf(sc, VTSCSI_TRACE, "error=%d req=%p ccb=%p\n", error, req, ccbh); return (error); } static void vtscsi_get_request_lun(uint8_t lun[], target_id_t *target_id, lun_id_t *lun_id) { *target_id = lun[1]; *lun_id = (lun[2] << 8) | lun[3]; } static void vtscsi_set_request_lun(struct ccb_hdr *ccbh, uint8_t lun[]) { lun[0] = 1; lun[1] = ccbh->target_id; lun[2] = 0x40 | ((ccbh->target_lun >> 8) & 0x3F); lun[3] = ccbh->target_lun & 0xFF; } static void vtscsi_init_scsi_cmd_req(struct ccb_scsiio *csio, struct virtio_scsi_cmd_req *cmd_req) { uint8_t attr; switch (csio->tag_action) { case MSG_HEAD_OF_Q_TAG: attr = VIRTIO_SCSI_S_HEAD; break; case MSG_ORDERED_Q_TAG: attr = VIRTIO_SCSI_S_ORDERED; break; case MSG_ACA_TASK: attr = VIRTIO_SCSI_S_ACA; break; default: /* MSG_SIMPLE_Q_TAG */ attr = VIRTIO_SCSI_S_SIMPLE; break; } vtscsi_set_request_lun(&csio->ccb_h, cmd_req->lun); cmd_req->tag = (uintptr_t) csio; cmd_req->task_attr = attr; memcpy(cmd_req->cdb, csio->ccb_h.flags & CAM_CDB_POINTER ? csio->cdb_io.cdb_ptr : csio->cdb_io.cdb_bytes, csio->cdb_len); } static void vtscsi_init_ctrl_tmf_req(struct ccb_hdr *ccbh, uint32_t subtype, uintptr_t tag, struct virtio_scsi_ctrl_tmf_req *tmf_req) { vtscsi_set_request_lun(ccbh, tmf_req->lun); tmf_req->type = VIRTIO_SCSI_T_TMF; tmf_req->subtype = subtype; tmf_req->tag = tag; } static void vtscsi_freeze_simq(struct vtscsi_softc *sc, int reason) { int frozen; frozen = sc->vtscsi_frozen; if (reason & VTSCSI_REQUEST && (sc->vtscsi_frozen & VTSCSI_FROZEN_NO_REQUESTS) == 0) sc->vtscsi_frozen |= VTSCSI_FROZEN_NO_REQUESTS; if (reason & VTSCSI_REQUEST_VQ && (sc->vtscsi_frozen & VTSCSI_FROZEN_REQUEST_VQ_FULL) == 0) sc->vtscsi_frozen |= VTSCSI_FROZEN_REQUEST_VQ_FULL; /* Freeze the SIMQ if transitioned to frozen. */ if (frozen == 0 && sc->vtscsi_frozen != 0) { vtscsi_dprintf(sc, VTSCSI_INFO, "SIMQ frozen\n"); xpt_freeze_simq(sc->vtscsi_sim, 1); } } static int vtscsi_thaw_simq(struct vtscsi_softc *sc, int reason) { int thawed; if (sc->vtscsi_frozen == 0 || reason == 0) return (0); if (reason & VTSCSI_REQUEST && sc->vtscsi_frozen & VTSCSI_FROZEN_NO_REQUESTS) sc->vtscsi_frozen &= ~VTSCSI_FROZEN_NO_REQUESTS; if (reason & VTSCSI_REQUEST_VQ && sc->vtscsi_frozen & VTSCSI_FROZEN_REQUEST_VQ_FULL) sc->vtscsi_frozen &= ~VTSCSI_FROZEN_REQUEST_VQ_FULL; thawed = sc->vtscsi_frozen == 0; if (thawed != 0) vtscsi_dprintf(sc, VTSCSI_INFO, "SIMQ thawed\n"); return (thawed); } static void vtscsi_announce(struct vtscsi_softc *sc, uint32_t ac_code, target_id_t target_id, lun_id_t lun_id) { struct cam_path *path; /* Use the wildcard path from our softc for bus announcements. */ if (target_id == CAM_TARGET_WILDCARD && lun_id == CAM_LUN_WILDCARD) { xpt_async(ac_code, sc->vtscsi_path, NULL); return; } if (xpt_create_path(&path, NULL, cam_sim_path(sc->vtscsi_sim), target_id, lun_id) != CAM_REQ_CMP) { vtscsi_dprintf(sc, VTSCSI_ERROR, "cannot create path\n"); return; } xpt_async(ac_code, path, NULL); xpt_free_path(path); } static void vtscsi_execute_rescan(struct vtscsi_softc *sc, target_id_t target_id, lun_id_t lun_id) { union ccb *ccb; cam_status status; ccb = xpt_alloc_ccb_nowait(); if (ccb == NULL) { vtscsi_dprintf(sc, VTSCSI_ERROR, "cannot allocate CCB\n"); return; } status = xpt_create_path(&ccb->ccb_h.path, NULL, cam_sim_path(sc->vtscsi_sim), target_id, lun_id); if (status != CAM_REQ_CMP) { xpt_free_ccb(ccb); return; } xpt_rescan(ccb); } static void vtscsi_execute_rescan_bus(struct vtscsi_softc *sc) { vtscsi_execute_rescan(sc, CAM_TARGET_WILDCARD, CAM_LUN_WILDCARD); } static void vtscsi_transport_reset_event(struct vtscsi_softc *sc, struct virtio_scsi_event *event) { target_id_t target_id; lun_id_t lun_id; vtscsi_get_request_lun(event->lun, &target_id, &lun_id); switch (event->reason) { case VIRTIO_SCSI_EVT_RESET_RESCAN: case VIRTIO_SCSI_EVT_RESET_REMOVED: vtscsi_execute_rescan(sc, target_id, lun_id); break; default: device_printf(sc->vtscsi_dev, "unhandled transport event reason: %d\n", event->reason); break; } } static void vtscsi_handle_event(struct vtscsi_softc *sc, struct virtio_scsi_event *event) { int error; if ((event->event & VIRTIO_SCSI_T_EVENTS_MISSED) == 0) { switch (event->event) { case VIRTIO_SCSI_T_TRANSPORT_RESET: vtscsi_transport_reset_event(sc, event); break; default: device_printf(sc->vtscsi_dev, "unhandled event: %d\n", event->event); break; } } else vtscsi_execute_rescan_bus(sc); /* * This should always be successful since the buffer * was just dequeued. */ error = vtscsi_enqueue_event_buf(sc, event); KASSERT(error == 0, ("cannot requeue event buffer: %d", error)); } static int vtscsi_enqueue_event_buf(struct vtscsi_softc *sc, struct virtio_scsi_event *event) { struct sglist *sg; struct virtqueue *vq; int size, error; sg = sc->vtscsi_sglist; vq = sc->vtscsi_event_vq; size = sc->vtscsi_event_buf_size; bzero(event, size); sglist_reset(sg); error = sglist_append(sg, event, size); if (error) return (error); error = virtqueue_enqueue(vq, event, sg, 0, sg->sg_nseg); if (error) return (error); virtqueue_notify(vq); return (0); } static int vtscsi_init_event_vq(struct vtscsi_softc *sc) { struct virtio_scsi_event *event; int i, size, error; /* * The first release of QEMU with VirtIO SCSI support would crash * when attempting to notify the event virtqueue. This was fixed * when hotplug support was added. */ if (sc->vtscsi_flags & VTSCSI_FLAG_HOTPLUG) size = sc->vtscsi_event_buf_size; else size = 0; if (size < sizeof(struct virtio_scsi_event)) return (0); for (i = 0; i < VTSCSI_NUM_EVENT_BUFS; i++) { event = &sc->vtscsi_event_bufs[i]; error = vtscsi_enqueue_event_buf(sc, event); if (error) break; } /* * Even just one buffer is enough. Missed events are * denoted with the VIRTIO_SCSI_T_EVENTS_MISSED flag. */ if (i > 0) error = 0; return (error); } static void vtscsi_reinit_event_vq(struct vtscsi_softc *sc) { struct virtio_scsi_event *event; int i, error; if ((sc->vtscsi_flags & VTSCSI_FLAG_HOTPLUG) == 0 || sc->vtscsi_event_buf_size < sizeof(struct virtio_scsi_event)) return; for (i = 0; i < VTSCSI_NUM_EVENT_BUFS; i++) { event = &sc->vtscsi_event_bufs[i]; error = vtscsi_enqueue_event_buf(sc, event); if (error) break; } KASSERT(i > 0, ("cannot reinit event vq: %d", error)); } static void vtscsi_drain_event_vq(struct vtscsi_softc *sc) { struct virtqueue *vq; int last; vq = sc->vtscsi_event_vq; last = 0; while (virtqueue_drain(vq, &last) != NULL) ; KASSERT(virtqueue_empty(vq), ("eventvq not empty")); } static void vtscsi_complete_vqs_locked(struct vtscsi_softc *sc) { VTSCSI_LOCK_OWNED(sc); if (sc->vtscsi_request_vq != NULL) vtscsi_complete_vq(sc, sc->vtscsi_request_vq); if (sc->vtscsi_control_vq != NULL) vtscsi_complete_vq(sc, sc->vtscsi_control_vq); } static void vtscsi_complete_vqs(struct vtscsi_softc *sc) { VTSCSI_LOCK(sc); vtscsi_complete_vqs_locked(sc); VTSCSI_UNLOCK(sc); } static void vtscsi_cancel_request(struct vtscsi_softc *sc, struct vtscsi_request *req) { union ccb *ccb; int detach; ccb = req->vsr_ccb; vtscsi_dprintf(sc, VTSCSI_TRACE, "req=%p ccb=%p\n", req, ccb); /* * The callout must be drained when detaching since the request is * about to be freed. The VTSCSI_MTX must not be held for this in * case the callout is pending because there is a deadlock potential. * Otherwise, the virtqueue is being drained because of a bus reset * so we only need to attempt to stop the callouts. */ detach = (sc->vtscsi_flags & VTSCSI_FLAG_DETACH) != 0; if (detach != 0) VTSCSI_LOCK_NOTOWNED(sc); else VTSCSI_LOCK_OWNED(sc); if (req->vsr_flags & VTSCSI_REQ_FLAG_TIMEOUT_SET) { if (detach != 0) callout_drain(&req->vsr_callout); else callout_stop(&req->vsr_callout); } if (ccb != NULL) { if (detach != 0) { VTSCSI_LOCK(sc); ccb->ccb_h.status = CAM_NO_HBA; } else ccb->ccb_h.status = CAM_REQUEUE_REQ; xpt_done(ccb); if (detach != 0) VTSCSI_UNLOCK(sc); } vtscsi_enqueue_request(sc, req); } static void vtscsi_drain_vq(struct vtscsi_softc *sc, struct virtqueue *vq) { struct vtscsi_request *req; int last; last = 0; vtscsi_dprintf(sc, VTSCSI_TRACE, "vq=%p\n", vq); while ((req = virtqueue_drain(vq, &last)) != NULL) vtscsi_cancel_request(sc, req); KASSERT(virtqueue_empty(vq), ("virtqueue not empty")); } static void vtscsi_drain_vqs(struct vtscsi_softc *sc) { if (sc->vtscsi_control_vq != NULL) vtscsi_drain_vq(sc, sc->vtscsi_control_vq); if (sc->vtscsi_request_vq != NULL) vtscsi_drain_vq(sc, sc->vtscsi_request_vq); if (sc->vtscsi_event_vq != NULL) vtscsi_drain_event_vq(sc); } static void vtscsi_stop(struct vtscsi_softc *sc) { vtscsi_disable_vqs_intr(sc); virtio_stop(sc->vtscsi_dev); } static int vtscsi_reset_bus(struct vtscsi_softc *sc) { int error; VTSCSI_LOCK_OWNED(sc); if (vtscsi_bus_reset_disable != 0) { device_printf(sc->vtscsi_dev, "bus reset disabled\n"); return (0); } sc->vtscsi_flags |= VTSCSI_FLAG_RESET; /* * vtscsi_stop() will cause the in-flight requests to be canceled. * Those requests are then completed here so CAM will retry them * after the reset is complete. */ vtscsi_stop(sc); vtscsi_complete_vqs_locked(sc); /* Rid the virtqueues of any remaining requests. */ vtscsi_drain_vqs(sc); /* * Any resource shortage that froze the SIMQ cannot persist across * a bus reset so ensure it gets thawed here. */ if (vtscsi_thaw_simq(sc, VTSCSI_REQUEST | VTSCSI_REQUEST_VQ) != 0) xpt_release_simq(sc->vtscsi_sim, 0); error = vtscsi_reinit(sc); if (error) { device_printf(sc->vtscsi_dev, "reinitialization failed, stopping device...\n"); vtscsi_stop(sc); } else vtscsi_announce(sc, AC_BUS_RESET, CAM_TARGET_WILDCARD, CAM_LUN_WILDCARD); sc->vtscsi_flags &= ~VTSCSI_FLAG_RESET; return (error); } static void vtscsi_init_request(struct vtscsi_softc *sc, struct vtscsi_request *req) { #ifdef INVARIANTS int req_nsegs, resp_nsegs; req_nsegs = sglist_count(&req->vsr_ureq, sizeof(req->vsr_ureq)); resp_nsegs = sglist_count(&req->vsr_uresp, sizeof(req->vsr_uresp)); KASSERT(req_nsegs == 1, ("request crossed page boundary")); KASSERT(resp_nsegs == 1, ("response crossed page boundary")); #endif req->vsr_softc = sc; callout_init_mtx(&req->vsr_callout, VTSCSI_MTX(sc), 0); } static int vtscsi_alloc_requests(struct vtscsi_softc *sc) { struct vtscsi_request *req; int i, nreqs; /* * Commands destined for either the request or control queues come * from the same SIM queue. Use the size of the request virtqueue * as it (should) be much more frequently used. Some additional * requests are allocated for internal (TMF) use. */ nreqs = virtqueue_size(sc->vtscsi_request_vq); if ((sc->vtscsi_flags & VTSCSI_FLAG_INDIRECT) == 0) nreqs /= VTSCSI_MIN_SEGMENTS; nreqs += VTSCSI_RESERVED_REQUESTS; for (i = 0; i < nreqs; i++) { req = malloc(sizeof(struct vtscsi_request), M_DEVBUF, M_NOWAIT); if (req == NULL) return (ENOMEM); vtscsi_init_request(sc, req); sc->vtscsi_nrequests++; vtscsi_enqueue_request(sc, req); } return (0); } static void vtscsi_free_requests(struct vtscsi_softc *sc) { struct vtscsi_request *req; while ((req = vtscsi_dequeue_request(sc)) != NULL) { KASSERT(callout_active(&req->vsr_callout) == 0, ("request callout still active")); sc->vtscsi_nrequests--; free(req, M_DEVBUF); } KASSERT(sc->vtscsi_nrequests == 0, ("leaked requests: %d", sc->vtscsi_nrequests)); } static void vtscsi_enqueue_request(struct vtscsi_softc *sc, struct vtscsi_request *req) { KASSERT(req->vsr_softc == sc, ("non-matching request vsr_softc %p/%p", req->vsr_softc, sc)); vtscsi_dprintf(sc, VTSCSI_TRACE, "req=%p\n", req); /* A request is available so the SIMQ could be released. */ if (vtscsi_thaw_simq(sc, VTSCSI_REQUEST) != 0) xpt_release_simq(sc->vtscsi_sim, 1); req->vsr_ccb = NULL; req->vsr_complete = NULL; req->vsr_ptr0 = NULL; req->vsr_state = VTSCSI_REQ_STATE_FREE; req->vsr_flags = 0; bzero(&req->vsr_ureq, sizeof(req->vsr_ureq)); bzero(&req->vsr_uresp, sizeof(req->vsr_uresp)); /* * We insert at the tail of the queue in order to make it * very unlikely a request will be reused if we race with * stopping its callout handler. */ TAILQ_INSERT_TAIL(&sc->vtscsi_req_free, req, vsr_link); } static struct vtscsi_request * vtscsi_dequeue_request(struct vtscsi_softc *sc) { struct vtscsi_request *req; req = TAILQ_FIRST(&sc->vtscsi_req_free); if (req != NULL) { req->vsr_state = VTSCSI_REQ_STATE_INUSE; TAILQ_REMOVE(&sc->vtscsi_req_free, req, vsr_link); } else sc->vtscsi_stats.dequeue_no_requests++; vtscsi_dprintf(sc, VTSCSI_TRACE, "req=%p\n", req); return (req); } static void vtscsi_complete_request(struct vtscsi_request *req) { if (req->vsr_flags & VTSCSI_REQ_FLAG_POLLED) req->vsr_flags |= VTSCSI_REQ_FLAG_COMPLETE; if (req->vsr_complete != NULL) req->vsr_complete(req->vsr_softc, req); } static void vtscsi_complete_vq(struct vtscsi_softc *sc, struct virtqueue *vq) { struct vtscsi_request *req; VTSCSI_LOCK_OWNED(sc); while ((req = virtqueue_dequeue(vq, NULL)) != NULL) vtscsi_complete_request(req); } static void vtscsi_control_vq_intr(void *xsc) { struct vtscsi_softc *sc; struct virtqueue *vq; sc = xsc; vq = sc->vtscsi_control_vq; again: VTSCSI_LOCK(sc); vtscsi_complete_vq(sc, sc->vtscsi_control_vq); if (virtqueue_enable_intr(vq) != 0) { virtqueue_disable_intr(vq); VTSCSI_UNLOCK(sc); goto again; } VTSCSI_UNLOCK(sc); } static void vtscsi_event_vq_intr(void *xsc) { struct vtscsi_softc *sc; struct virtqueue *vq; struct virtio_scsi_event *event; sc = xsc; vq = sc->vtscsi_event_vq; again: VTSCSI_LOCK(sc); while ((event = virtqueue_dequeue(vq, NULL)) != NULL) vtscsi_handle_event(sc, event); if (virtqueue_enable_intr(vq) != 0) { virtqueue_disable_intr(vq); VTSCSI_UNLOCK(sc); goto again; } VTSCSI_UNLOCK(sc); } static void vtscsi_request_vq_intr(void *xsc) { struct vtscsi_softc *sc; struct virtqueue *vq; sc = xsc; vq = sc->vtscsi_request_vq; again: VTSCSI_LOCK(sc); vtscsi_complete_vq(sc, sc->vtscsi_request_vq); if (virtqueue_enable_intr(vq) != 0) { virtqueue_disable_intr(vq); VTSCSI_UNLOCK(sc); goto again; } VTSCSI_UNLOCK(sc); } static void vtscsi_disable_vqs_intr(struct vtscsi_softc *sc) { virtqueue_disable_intr(sc->vtscsi_control_vq); virtqueue_disable_intr(sc->vtscsi_event_vq); virtqueue_disable_intr(sc->vtscsi_request_vq); } static void vtscsi_enable_vqs_intr(struct vtscsi_softc *sc) { virtqueue_enable_intr(sc->vtscsi_control_vq); virtqueue_enable_intr(sc->vtscsi_event_vq); virtqueue_enable_intr(sc->vtscsi_request_vq); } static void vtscsi_get_tunables(struct vtscsi_softc *sc) { char tmpstr[64]; TUNABLE_INT_FETCH("hw.vtscsi.debug_level", &sc->vtscsi_debug); snprintf(tmpstr, sizeof(tmpstr), "dev.vtscsi.%d.debug_level", device_get_unit(sc->vtscsi_dev)); TUNABLE_INT_FETCH(tmpstr, &sc->vtscsi_debug); } static void vtscsi_add_sysctl(struct vtscsi_softc *sc) { device_t dev; struct vtscsi_statistics *stats; struct sysctl_ctx_list *ctx; struct sysctl_oid *tree; struct sysctl_oid_list *child; dev = sc->vtscsi_dev; stats = &sc->vtscsi_stats; ctx = device_get_sysctl_ctx(dev); tree = device_get_sysctl_tree(dev); child = SYSCTL_CHILDREN(tree); SYSCTL_ADD_INT(ctx, child, OID_AUTO, "debug_level", CTLFLAG_RW, &sc->vtscsi_debug, 0, "Debug level"); SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "scsi_cmd_timeouts", CTLFLAG_RD, &stats->scsi_cmd_timeouts, "SCSI command timeouts"); SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "dequeue_no_requests", CTLFLAG_RD, &stats->dequeue_no_requests, "No available requests to dequeue"); } static void vtscsi_printf_req(struct vtscsi_request *req, const char *func, const char *fmt, ...) { struct vtscsi_softc *sc; union ccb *ccb; struct sbuf sb; va_list ap; char str[192]; char path_str[64]; if (req == NULL) return; sc = req->vsr_softc; ccb = req->vsr_ccb; va_start(ap, fmt); sbuf_new(&sb, str, sizeof(str), 0); if (ccb == NULL) { sbuf_printf(&sb, "(noperiph:%s%d:%u): ", cam_sim_name(sc->vtscsi_sim), cam_sim_unit(sc->vtscsi_sim), cam_sim_bus(sc->vtscsi_sim)); } else { xpt_path_string(ccb->ccb_h.path, path_str, sizeof(path_str)); sbuf_cat(&sb, path_str); if (ccb->ccb_h.func_code == XPT_SCSI_IO) { scsi_command_string(&ccb->csio, &sb); sbuf_printf(&sb, "length %d ", ccb->csio.dxfer_len); } } sbuf_vprintf(&sb, fmt, ap); va_end(ap); sbuf_finish(&sb); printf("%s: %s: %s", device_get_nameunit(sc->vtscsi_dev), func, sbuf_data(&sb)); } Index: head/sys/dev/virtio/virtio.h =================================================================== --- head/sys/dev/virtio/virtio.h (revision 348598) +++ head/sys/dev/virtio/virtio.h (revision 348599) @@ -1,149 +1,174 @@ /*- * SPDX-License-Identifier: BSD-2-Clause * * Copyright (c) 2014, Bryan Venteicher * 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. * * $FreeBSD$ */ #ifndef _VIRTIO_H_ #define _VIRTIO_H_ #include #include struct vq_alloc_info; /* * Each virtqueue indirect descriptor list must be physically contiguous. * To allow us to malloc(9) each list individually, limit the number * supported to what will fit in one page. With 4KB pages, this is a limit * of 256 descriptors. If there is ever a need for more, we can switch to * contigmalloc(9) for the larger allocations, similar to what * bus_dmamem_alloc(9) does. * * Note the sizeof(struct vring_desc) is 16 bytes. */ #define VIRTIO_MAX_INDIRECT ((int) (PAGE_SIZE / 16)) /* * VirtIO instance variables indices. */ #define VIRTIO_IVAR_DEVTYPE 1 #define VIRTIO_IVAR_FEATURE_DESC 2 #define VIRTIO_IVAR_VENDOR 3 #define VIRTIO_IVAR_DEVICE 4 #define VIRTIO_IVAR_SUBVENDOR 5 #define VIRTIO_IVAR_SUBDEVICE 6 struct virtio_feature_desc { uint64_t vfd_val; const char *vfd_str; }; +struct virtio_pnp_match { + uint32_t device_type; + const char *description; +}; +#define VIRTIO_SIMPLE_PNPTABLE(driver, devtype, desc) \ + static const struct virtio_pnp_match driver ## _match = { \ + .device_type = devtype, \ + .description = desc, \ + } +#define VIRTIO_SIMPLE_PNPINFO(bus, driver) \ + MODULE_PNP_INFO("U32:device_type;D:#", bus, driver, \ + &driver ## _match, 1) +#define VIRTIO_SIMPLE_PROBE(dev, driver) \ + (virtio_simple_probe(dev, &driver ## _match)) + const char *virtio_device_name(uint16_t devid); void virtio_describe(device_t dev, const char *msg, uint64_t features, struct virtio_feature_desc *feature_desc); /* * VirtIO Bus Methods. */ void virtio_read_ivar(device_t dev, int ivar, uintptr_t *val); void virtio_write_ivar(device_t dev, int ivar, uintptr_t val); uint64_t virtio_negotiate_features(device_t dev, uint64_t child_features); int virtio_alloc_virtqueues(device_t dev, int flags, int nvqs, struct vq_alloc_info *info); int virtio_setup_intr(device_t dev, enum intr_type type); int virtio_with_feature(device_t dev, uint64_t feature); void virtio_stop(device_t dev); int virtio_config_generation(device_t dev); int virtio_reinit(device_t dev, uint64_t features); void virtio_reinit_complete(device_t dev); int virtio_child_pnpinfo_str(device_t busdev, device_t child, char *buf, size_t buflen); /* * Read/write a variable amount from the device specific (ie, network) * configuration region. This region is encoded in the same endian as * the guest. */ void virtio_read_device_config(device_t dev, bus_size_t offset, void *dst, int length); void virtio_write_device_config(device_t dev, bus_size_t offset, void *src, int length); /* Inlined device specific read/write functions for common lengths. */ #define VIRTIO_RDWR_DEVICE_CONFIG(size, type) \ static inline type \ __CONCAT(virtio_read_dev_config_,size)(device_t dev, \ bus_size_t offset) \ { \ type val; \ virtio_read_device_config(dev, offset, &val, sizeof(type)); \ return (val); \ } \ \ static inline void \ __CONCAT(virtio_write_dev_config_,size)(device_t dev, \ bus_size_t offset, type val) \ { \ virtio_write_device_config(dev, offset, &val, sizeof(type)); \ } VIRTIO_RDWR_DEVICE_CONFIG(1, uint8_t); VIRTIO_RDWR_DEVICE_CONFIG(2, uint16_t); VIRTIO_RDWR_DEVICE_CONFIG(4, uint32_t); #undef VIRTIO_RDWR_DEVICE_CONFIG #define VIRTIO_READ_IVAR(name, ivar) \ static inline int \ __CONCAT(virtio_get_,name)(device_t dev) \ { \ uintptr_t val; \ virtio_read_ivar(dev, ivar, &val); \ return ((int) val); \ } VIRTIO_READ_IVAR(device_type, VIRTIO_IVAR_DEVTYPE); VIRTIO_READ_IVAR(vendor, VIRTIO_IVAR_VENDOR); VIRTIO_READ_IVAR(device, VIRTIO_IVAR_DEVICE); VIRTIO_READ_IVAR(subvendor, VIRTIO_IVAR_SUBVENDOR); VIRTIO_READ_IVAR(subdevice, VIRTIO_IVAR_SUBDEVICE); #undef VIRTIO_READ_IVAR #define VIRTIO_WRITE_IVAR(name, ivar) \ static inline void \ __CONCAT(virtio_set_,name)(device_t dev, void *val) \ { \ virtio_write_ivar(dev, ivar, (uintptr_t) val); \ } VIRTIO_WRITE_IVAR(feature_desc, VIRTIO_IVAR_FEATURE_DESC); #undef VIRTIO_WRITE_IVAR + +static inline int +virtio_simple_probe(device_t dev, const struct virtio_pnp_match *match) +{ + + if (virtio_get_device_type(dev) != match->device_type) + return (ENXIO); + device_set_desc(dev, match->description); + return (BUS_PROBE_DEFAULT); +} #endif /* _VIRTIO_H_ */