1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2016 Intel Corporation
11 #include <sys/types.h>
14 #ifdef RTE_LIBRTE_VHOST_NUMA
18 #include <rte_common.h>
19 #include <rte_malloc.h>
24 #include "vhost_user.h"
26 #define VIRTIO_MIN_MTU 68
27 #define VIRTIO_MAX_MTU 65535
29 static const char *vhost_message_str[VHOST_USER_MAX] = {
30 [VHOST_USER_NONE] = "VHOST_USER_NONE",
31 [VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
32 [VHOST_USER_SET_FEATURES] = "VHOST_USER_SET_FEATURES",
33 [VHOST_USER_SET_OWNER] = "VHOST_USER_SET_OWNER",
34 [VHOST_USER_RESET_OWNER] = "VHOST_USER_RESET_OWNER",
35 [VHOST_USER_SET_MEM_TABLE] = "VHOST_USER_SET_MEM_TABLE",
36 [VHOST_USER_SET_LOG_BASE] = "VHOST_USER_SET_LOG_BASE",
37 [VHOST_USER_SET_LOG_FD] = "VHOST_USER_SET_LOG_FD",
38 [VHOST_USER_SET_VRING_NUM] = "VHOST_USER_SET_VRING_NUM",
39 [VHOST_USER_SET_VRING_ADDR] = "VHOST_USER_SET_VRING_ADDR",
40 [VHOST_USER_SET_VRING_BASE] = "VHOST_USER_SET_VRING_BASE",
41 [VHOST_USER_GET_VRING_BASE] = "VHOST_USER_GET_VRING_BASE",
42 [VHOST_USER_SET_VRING_KICK] = "VHOST_USER_SET_VRING_KICK",
43 [VHOST_USER_SET_VRING_CALL] = "VHOST_USER_SET_VRING_CALL",
44 [VHOST_USER_SET_VRING_ERR] = "VHOST_USER_SET_VRING_ERR",
45 [VHOST_USER_GET_PROTOCOL_FEATURES] = "VHOST_USER_GET_PROTOCOL_FEATURES",
46 [VHOST_USER_SET_PROTOCOL_FEATURES] = "VHOST_USER_SET_PROTOCOL_FEATURES",
47 [VHOST_USER_GET_QUEUE_NUM] = "VHOST_USER_GET_QUEUE_NUM",
48 [VHOST_USER_SET_VRING_ENABLE] = "VHOST_USER_SET_VRING_ENABLE",
49 [VHOST_USER_SEND_RARP] = "VHOST_USER_SEND_RARP",
50 [VHOST_USER_NET_SET_MTU] = "VHOST_USER_NET_SET_MTU",
51 [VHOST_USER_SET_SLAVE_REQ_FD] = "VHOST_USER_SET_SLAVE_REQ_FD",
52 [VHOST_USER_IOTLB_MSG] = "VHOST_USER_IOTLB_MSG",
61 ret = fstat(fd, &stat);
62 return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
66 free_mem_region(struct virtio_net *dev)
69 struct rte_vhost_mem_region *reg;
71 if (!dev || !dev->mem)
74 for (i = 0; i < dev->mem->nregions; i++) {
75 reg = &dev->mem->regions[i];
76 if (reg->host_user_addr) {
77 munmap(reg->mmap_addr, reg->mmap_size);
84 vhost_backend_cleanup(struct virtio_net *dev)
92 free(dev->guest_pages);
93 dev->guest_pages = NULL;
96 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
100 if (dev->slave_req_fd >= 0) {
101 close(dev->slave_req_fd);
102 dev->slave_req_fd = -1;
107 * This function just returns success at the moment unless
108 * the device hasn't been initialised.
111 vhost_user_set_owner(void)
117 vhost_user_reset_owner(struct virtio_net *dev)
119 if (dev->flags & VIRTIO_DEV_RUNNING) {
120 dev->flags &= ~VIRTIO_DEV_RUNNING;
121 dev->notify_ops->destroy_device(dev->vid);
124 cleanup_device(dev, 0);
130 * The features that we support are requested.
133 vhost_user_get_features(struct virtio_net *dev)
135 uint64_t features = 0;
137 rte_vhost_driver_get_features(dev->ifname, &features);
142 * We receive the negotiated features supported by us and the virtio device.
145 vhost_user_set_features(struct virtio_net *dev, uint64_t features)
147 uint64_t vhost_features = 0;
149 rte_vhost_driver_get_features(dev->ifname, &vhost_features);
150 if (features & ~vhost_features) {
151 RTE_LOG(ERR, VHOST_CONFIG,
152 "(%d) received invalid negotiated features.\n",
157 if (dev->flags & VIRTIO_DEV_RUNNING) {
158 if (dev->features == features)
162 * Error out if master tries to change features while device is
163 * in running state. The exception being VHOST_F_LOG_ALL, which
164 * is enabled when the live-migration starts.
166 if ((dev->features ^ features) & ~(1ULL << VHOST_F_LOG_ALL)) {
167 RTE_LOG(ERR, VHOST_CONFIG,
168 "(%d) features changed while device is running.\n",
173 if (dev->notify_ops->features_changed)
174 dev->notify_ops->features_changed(dev->vid, features);
177 dev->features = features;
179 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
180 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
182 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
184 LOG_DEBUG(VHOST_CONFIG,
185 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
187 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
188 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
190 if (!(dev->features & (1ULL << VIRTIO_NET_F_MQ))) {
192 * Remove all but first queue pair if MQ hasn't been
193 * negotiated. This is safe because the device is not
194 * running at this stage.
196 while (dev->nr_vring > 2) {
197 struct vhost_virtqueue *vq;
199 vq = dev->virtqueue[--dev->nr_vring];
203 dev->virtqueue[dev->nr_vring] = NULL;
213 * The virtio device sends us the size of the descriptor ring.
216 vhost_user_set_vring_num(struct virtio_net *dev,
219 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
221 vq->size = msg->payload.state.num;
223 if (dev->dequeue_zero_copy) {
225 vq->last_zmbuf_idx = 0;
226 vq->zmbuf_size = vq->size;
227 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
228 sizeof(struct zcopy_mbuf), 0);
229 if (vq->zmbufs == NULL) {
230 RTE_LOG(WARNING, VHOST_CONFIG,
231 "failed to allocate mem for zero copy; "
232 "zero copy is force disabled\n");
233 dev->dequeue_zero_copy = 0;
237 vq->shadow_used_ring = rte_malloc(NULL,
238 vq->size * sizeof(struct vring_used_elem),
239 RTE_CACHE_LINE_SIZE);
240 if (!vq->shadow_used_ring) {
241 RTE_LOG(ERR, VHOST_CONFIG,
242 "failed to allocate memory for shadow used ring.\n");
246 vq->batch_copy_elems = rte_malloc(NULL,
247 vq->size * sizeof(struct batch_copy_elem),
248 RTE_CACHE_LINE_SIZE);
249 if (!vq->batch_copy_elems) {
250 RTE_LOG(ERR, VHOST_CONFIG,
251 "failed to allocate memory for batching copy.\n");
259 * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
260 * same numa node as the memory of vring descriptor.
262 #ifdef RTE_LIBRTE_VHOST_NUMA
263 static struct virtio_net*
264 numa_realloc(struct virtio_net *dev, int index)
266 int oldnode, newnode;
267 struct virtio_net *old_dev;
268 struct vhost_virtqueue *old_vq, *vq;
272 vq = old_vq = dev->virtqueue[index];
274 ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
275 MPOL_F_NODE | MPOL_F_ADDR);
277 /* check if we need to reallocate vq */
278 ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
279 MPOL_F_NODE | MPOL_F_ADDR);
281 RTE_LOG(ERR, VHOST_CONFIG,
282 "Unable to get vq numa information.\n");
285 if (oldnode != newnode) {
286 RTE_LOG(INFO, VHOST_CONFIG,
287 "reallocate vq from %d to %d node\n", oldnode, newnode);
288 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
292 memcpy(vq, old_vq, sizeof(*vq));
296 /* check if we need to reallocate dev */
297 ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
298 MPOL_F_NODE | MPOL_F_ADDR);
300 RTE_LOG(ERR, VHOST_CONFIG,
301 "Unable to get dev numa information.\n");
304 if (oldnode != newnode) {
305 RTE_LOG(INFO, VHOST_CONFIG,
306 "reallocate dev from %d to %d node\n",
308 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
314 memcpy(dev, old_dev, sizeof(*dev));
319 dev->virtqueue[index] = vq;
320 vhost_devices[dev->vid] = dev;
323 vhost_user_iotlb_init(dev, index);
328 static struct virtio_net*
329 numa_realloc(struct virtio_net *dev, int index __rte_unused)
335 /* Converts QEMU virtual address to Vhost virtual address. */
337 qva_to_vva(struct virtio_net *dev, uint64_t qva)
339 struct rte_vhost_mem_region *reg;
342 /* Find the region where the address lives. */
343 for (i = 0; i < dev->mem->nregions; i++) {
344 reg = &dev->mem->regions[i];
346 if (qva >= reg->guest_user_addr &&
347 qva < reg->guest_user_addr + reg->size) {
348 return qva - reg->guest_user_addr +
358 * Converts ring address to Vhost virtual address.
359 * If IOMMU is enabled, the ring address is a guest IO virtual address,
360 * else it is a QEMU virtual address.
363 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
364 uint64_t ra, uint64_t size)
366 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
369 vva = vhost_user_iotlb_cache_find(vq, ra,
370 &size, VHOST_ACCESS_RW);
372 vhost_user_iotlb_miss(dev, ra, VHOST_ACCESS_RW);
377 return qva_to_vva(dev, ra);
380 static struct virtio_net *
381 translate_ring_addresses(struct virtio_net *dev, int vq_index)
383 struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
384 struct vhost_vring_addr *addr = &vq->ring_addrs;
386 /* The addresses are converted from QEMU virtual to Vhost virtual. */
387 if (vq->desc && vq->avail && vq->used)
390 vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
391 vq, addr->desc_user_addr, sizeof(struct vring_desc));
393 RTE_LOG(DEBUG, VHOST_CONFIG,
394 "(%d) failed to find desc ring address.\n",
399 dev = numa_realloc(dev, vq_index);
400 vq = dev->virtqueue[vq_index];
401 addr = &vq->ring_addrs;
403 vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
404 vq, addr->avail_user_addr, sizeof(struct vring_avail));
405 if (vq->avail == 0) {
406 RTE_LOG(DEBUG, VHOST_CONFIG,
407 "(%d) failed to find avail ring address.\n",
412 vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
413 vq, addr->used_user_addr, sizeof(struct vring_used));
415 RTE_LOG(DEBUG, VHOST_CONFIG,
416 "(%d) failed to find used ring address.\n",
421 if (vq->last_used_idx != vq->used->idx) {
422 RTE_LOG(WARNING, VHOST_CONFIG,
423 "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
424 "some packets maybe resent for Tx and dropped for Rx\n",
425 vq->last_used_idx, vq->used->idx);
426 vq->last_used_idx = vq->used->idx;
427 vq->last_avail_idx = vq->used->idx;
430 vq->log_guest_addr = addr->log_guest_addr;
432 LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
434 LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
435 dev->vid, vq->avail);
436 LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
438 LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
439 dev->vid, vq->log_guest_addr);
445 * The virtio device sends us the desc, used and avail ring addresses.
446 * This function then converts these to our address space.
449 vhost_user_set_vring_addr(struct virtio_net **pdev, VhostUserMsg *msg)
451 struct vhost_virtqueue *vq;
452 struct vhost_vring_addr *addr = &msg->payload.addr;
453 struct virtio_net *dev = *pdev;
455 if (dev->mem == NULL)
458 /* addr->index refers to the queue index. The txq 1, rxq is 0. */
459 vq = dev->virtqueue[msg->payload.addr.index];
462 * Rings addresses should not be interpreted as long as the ring is not
463 * started and enabled
465 memcpy(&vq->ring_addrs, addr, sizeof(*addr));
467 vring_invalidate(dev, vq);
469 if (vq->enabled && (dev->features &
470 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
471 dev = translate_ring_addresses(dev, msg->payload.state.index);
482 * The virtio device sends us the available ring last used index.
485 vhost_user_set_vring_base(struct virtio_net *dev,
488 dev->virtqueue[msg->payload.state.index]->last_used_idx =
489 msg->payload.state.num;
490 dev->virtqueue[msg->payload.state.index]->last_avail_idx =
491 msg->payload.state.num;
497 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
498 uint64_t host_phys_addr, uint64_t size)
500 struct guest_page *page, *last_page;
502 if (dev->nr_guest_pages == dev->max_guest_pages) {
503 dev->max_guest_pages *= 2;
504 dev->guest_pages = realloc(dev->guest_pages,
505 dev->max_guest_pages * sizeof(*page));
508 if (dev->nr_guest_pages > 0) {
509 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
510 /* merge if the two pages are continuous */
511 if (host_phys_addr == last_page->host_phys_addr +
513 last_page->size += size;
518 page = &dev->guest_pages[dev->nr_guest_pages++];
519 page->guest_phys_addr = guest_phys_addr;
520 page->host_phys_addr = host_phys_addr;
525 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
528 uint64_t reg_size = reg->size;
529 uint64_t host_user_addr = reg->host_user_addr;
530 uint64_t guest_phys_addr = reg->guest_phys_addr;
531 uint64_t host_phys_addr;
534 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
535 size = page_size - (guest_phys_addr & (page_size - 1));
536 size = RTE_MIN(size, reg_size);
538 add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
539 host_user_addr += size;
540 guest_phys_addr += size;
543 while (reg_size > 0) {
544 size = RTE_MIN(reg_size, page_size);
545 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
547 add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
549 host_user_addr += size;
550 guest_phys_addr += size;
555 #ifdef RTE_LIBRTE_VHOST_DEBUG
556 /* TODO: enable it only in debug mode? */
558 dump_guest_pages(struct virtio_net *dev)
561 struct guest_page *page;
563 for (i = 0; i < dev->nr_guest_pages; i++) {
564 page = &dev->guest_pages[i];
566 RTE_LOG(INFO, VHOST_CONFIG,
567 "guest physical page region %u\n"
568 "\t guest_phys_addr: %" PRIx64 "\n"
569 "\t host_phys_addr : %" PRIx64 "\n"
570 "\t size : %" PRIx64 "\n",
572 page->guest_phys_addr,
573 page->host_phys_addr,
578 #define dump_guest_pages(dev)
582 vhost_memory_changed(struct VhostUserMemory *new,
583 struct rte_vhost_memory *old)
587 if (new->nregions != old->nregions)
590 for (i = 0; i < new->nregions; ++i) {
591 VhostUserMemoryRegion *new_r = &new->regions[i];
592 struct rte_vhost_mem_region *old_r = &old->regions[i];
594 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
596 if (new_r->memory_size != old_r->size)
598 if (new_r->userspace_addr != old_r->guest_user_addr)
606 vhost_user_set_mem_table(struct virtio_net *dev, struct VhostUserMsg *pmsg)
608 struct VhostUserMemory memory = pmsg->payload.memory;
609 struct rte_vhost_mem_region *reg;
612 uint64_t mmap_offset;
617 if (dev->mem && !vhost_memory_changed(&memory, dev->mem)) {
618 RTE_LOG(INFO, VHOST_CONFIG,
619 "(%d) memory regions not changed\n", dev->vid);
621 for (i = 0; i < memory.nregions; i++)
628 free_mem_region(dev);
633 dev->nr_guest_pages = 0;
634 if (!dev->guest_pages) {
635 dev->max_guest_pages = 8;
636 dev->guest_pages = malloc(dev->max_guest_pages *
637 sizeof(struct guest_page));
638 if (dev->guest_pages == NULL) {
639 RTE_LOG(ERR, VHOST_CONFIG,
640 "(%d) failed to allocate memory "
641 "for dev->guest_pages\n",
647 dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
648 sizeof(struct rte_vhost_mem_region) * memory.nregions, 0);
649 if (dev->mem == NULL) {
650 RTE_LOG(ERR, VHOST_CONFIG,
651 "(%d) failed to allocate memory for dev->mem\n",
655 dev->mem->nregions = memory.nregions;
657 for (i = 0; i < memory.nregions; i++) {
659 reg = &dev->mem->regions[i];
661 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
662 reg->guest_user_addr = memory.regions[i].userspace_addr;
663 reg->size = memory.regions[i].memory_size;
666 mmap_offset = memory.regions[i].mmap_offset;
667 mmap_size = reg->size + mmap_offset;
669 /* mmap() without flag of MAP_ANONYMOUS, should be called
670 * with length argument aligned with hugepagesz at older
671 * longterm version Linux, like 2.6.32 and 3.2.72, or
672 * mmap() will fail with EINVAL.
674 * to avoid failure, make sure in caller to keep length
677 alignment = get_blk_size(fd);
678 if (alignment == (uint64_t)-1) {
679 RTE_LOG(ERR, VHOST_CONFIG,
680 "couldn't get hugepage size through fstat\n");
683 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
685 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
686 MAP_SHARED | MAP_POPULATE, fd, 0);
688 if (mmap_addr == MAP_FAILED) {
689 RTE_LOG(ERR, VHOST_CONFIG,
690 "mmap region %u failed.\n", i);
694 reg->mmap_addr = mmap_addr;
695 reg->mmap_size = mmap_size;
696 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
699 if (dev->dequeue_zero_copy)
700 add_guest_pages(dev, reg, alignment);
702 RTE_LOG(INFO, VHOST_CONFIG,
703 "guest memory region %u, size: 0x%" PRIx64 "\n"
704 "\t guest physical addr: 0x%" PRIx64 "\n"
705 "\t guest virtual addr: 0x%" PRIx64 "\n"
706 "\t host virtual addr: 0x%" PRIx64 "\n"
707 "\t mmap addr : 0x%" PRIx64 "\n"
708 "\t mmap size : 0x%" PRIx64 "\n"
709 "\t mmap align: 0x%" PRIx64 "\n"
710 "\t mmap off : 0x%" PRIx64 "\n",
712 reg->guest_phys_addr,
713 reg->guest_user_addr,
715 (uint64_t)(uintptr_t)mmap_addr,
721 dump_guest_pages(dev);
726 free_mem_region(dev);
733 vq_is_ready(struct vhost_virtqueue *vq)
735 return vq && vq->desc && vq->avail && vq->used &&
736 vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
737 vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
741 virtio_is_ready(struct virtio_net *dev)
743 struct vhost_virtqueue *vq;
746 if (dev->nr_vring == 0)
749 for (i = 0; i < dev->nr_vring; i++) {
750 vq = dev->virtqueue[i];
752 if (!vq_is_ready(vq))
756 RTE_LOG(INFO, VHOST_CONFIG,
757 "virtio is now ready for processing.\n");
762 vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *pmsg)
764 struct vhost_vring_file file;
765 struct vhost_virtqueue *vq;
767 file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
768 if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
769 file.fd = VIRTIO_INVALID_EVENTFD;
771 file.fd = pmsg->fds[0];
772 RTE_LOG(INFO, VHOST_CONFIG,
773 "vring call idx:%d file:%d\n", file.index, file.fd);
775 vq = dev->virtqueue[file.index];
779 vq->callfd = file.fd;
783 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *pmsg)
785 struct vhost_vring_file file;
786 struct vhost_virtqueue *vq;
787 struct virtio_net *dev = *pdev;
789 file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
790 if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
791 file.fd = VIRTIO_INVALID_EVENTFD;
793 file.fd = pmsg->fds[0];
794 RTE_LOG(INFO, VHOST_CONFIG,
795 "vring kick idx:%d file:%d\n", file.index, file.fd);
797 /* Interpret ring addresses only when ring is started. */
798 dev = translate_ring_addresses(dev, file.index);
804 vq = dev->virtqueue[file.index];
807 * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
808 * the ring starts already enabled. Otherwise, it is enabled via
809 * the SET_VRING_ENABLE message.
811 if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)))
816 vq->kickfd = file.fd;
820 free_zmbufs(struct vhost_virtqueue *vq)
822 struct zcopy_mbuf *zmbuf, *next;
824 for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
825 zmbuf != NULL; zmbuf = next) {
826 next = TAILQ_NEXT(zmbuf, next);
828 rte_pktmbuf_free(zmbuf->mbuf);
829 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
832 rte_free(vq->zmbufs);
836 * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
839 vhost_user_get_vring_base(struct virtio_net *dev,
842 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
844 /* We have to stop the queue (virtio) if it is running. */
845 if (dev->flags & VIRTIO_DEV_RUNNING) {
846 dev->flags &= ~VIRTIO_DEV_RUNNING;
847 dev->notify_ops->destroy_device(dev->vid);
850 dev->flags &= ~VIRTIO_DEV_READY;
852 /* Here we are safe to get the last used index */
853 msg->payload.state.num = vq->last_used_idx;
855 RTE_LOG(INFO, VHOST_CONFIG,
856 "vring base idx:%d file:%d\n", msg->payload.state.index,
857 msg->payload.state.num);
859 * Based on current qemu vhost-user implementation, this message is
860 * sent and only sent in vhost_vring_stop.
861 * TODO: cleanup the vring, it isn't usable since here.
866 vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
868 if (dev->dequeue_zero_copy)
870 rte_free(vq->shadow_used_ring);
871 vq->shadow_used_ring = NULL;
873 rte_free(vq->batch_copy_elems);
874 vq->batch_copy_elems = NULL;
880 * when virtio queues are ready to work, qemu will send us to
881 * enable the virtio queue pair.
884 vhost_user_set_vring_enable(struct virtio_net *dev,
887 int enable = (int)msg->payload.state.num;
889 RTE_LOG(INFO, VHOST_CONFIG,
890 "set queue enable: %d to qp idx: %d\n",
891 enable, msg->payload.state.index);
893 if (dev->notify_ops->vring_state_changed)
894 dev->notify_ops->vring_state_changed(dev->vid,
895 msg->payload.state.index, enable);
897 dev->virtqueue[msg->payload.state.index]->enabled = enable;
903 vhost_user_get_protocol_features(struct virtio_net *dev,
904 struct VhostUserMsg *msg)
906 uint64_t features, protocol_features = VHOST_USER_PROTOCOL_FEATURES;
908 rte_vhost_driver_get_features(dev->ifname, &features);
911 * REPLY_ACK protocol feature is only mandatory for now
912 * for IOMMU feature. If IOMMU is explicitly disabled by the
913 * application, disable also REPLY_ACK feature for older buggy
914 * Qemu versions (from v2.7.0 to v2.9.0).
916 if (!(features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
917 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK);
919 msg->payload.u64 = protocol_features;
920 msg->size = sizeof(msg->payload.u64);
924 vhost_user_set_protocol_features(struct virtio_net *dev,
925 uint64_t protocol_features)
927 if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
930 dev->protocol_features = protocol_features;
934 vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
936 int fd = msg->fds[0];
941 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
945 if (msg->size != sizeof(VhostUserLog)) {
946 RTE_LOG(ERR, VHOST_CONFIG,
947 "invalid log base msg size: %"PRId32" != %d\n",
948 msg->size, (int)sizeof(VhostUserLog));
952 size = msg->payload.log.mmap_size;
953 off = msg->payload.log.mmap_offset;
954 RTE_LOG(INFO, VHOST_CONFIG,
955 "log mmap size: %"PRId64", offset: %"PRId64"\n",
959 * mmap from 0 to workaround a hugepage mmap bug: mmap will
960 * fail when offset is not page size aligned.
962 addr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
964 if (addr == MAP_FAILED) {
965 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
970 * Free previously mapped log memory on occasionally
971 * multiple VHOST_USER_SET_LOG_BASE.
974 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
976 dev->log_addr = (uint64_t)(uintptr_t)addr;
977 dev->log_base = dev->log_addr + off;
978 dev->log_size = size;
984 * An rarp packet is constructed and broadcasted to notify switches about
985 * the new location of the migrated VM, so that packets from outside will
986 * not be lost after migration.
988 * However, we don't actually "send" a rarp packet here, instead, we set
989 * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
992 vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
994 uint8_t *mac = (uint8_t *)&msg->payload.u64;
996 RTE_LOG(DEBUG, VHOST_CONFIG,
997 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
998 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
999 memcpy(dev->mac.addr_bytes, mac, 6);
1002 * Set the flag to inject a RARP broadcast packet at
1003 * rte_vhost_dequeue_burst().
1005 * rte_smp_wmb() is for making sure the mac is copied
1006 * before the flag is set.
1009 rte_atomic16_set(&dev->broadcast_rarp, 1);
1015 vhost_user_net_set_mtu(struct virtio_net *dev, struct VhostUserMsg *msg)
1017 if (msg->payload.u64 < VIRTIO_MIN_MTU ||
1018 msg->payload.u64 > VIRTIO_MAX_MTU) {
1019 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
1025 dev->mtu = msg->payload.u64;
1031 vhost_user_set_req_fd(struct virtio_net *dev, struct VhostUserMsg *msg)
1033 int fd = msg->fds[0];
1036 RTE_LOG(ERR, VHOST_CONFIG,
1037 "Invalid file descriptor for slave channel (%d)\n",
1042 dev->slave_req_fd = fd;
1048 is_vring_iotlb_update(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
1050 struct vhost_vring_addr *ra;
1051 uint64_t start, end;
1054 end = start + imsg->size;
1056 ra = &vq->ring_addrs;
1057 if (ra->desc_user_addr >= start && ra->desc_user_addr < end)
1059 if (ra->avail_user_addr >= start && ra->avail_user_addr < end)
1061 if (ra->used_user_addr >= start && ra->used_user_addr < end)
1068 is_vring_iotlb_invalidate(struct vhost_virtqueue *vq,
1069 struct vhost_iotlb_msg *imsg)
1071 uint64_t istart, iend, vstart, vend;
1073 istart = imsg->iova;
1074 iend = istart + imsg->size - 1;
1076 vstart = (uintptr_t)vq->desc;
1077 vend = vstart + sizeof(struct vring_desc) * vq->size - 1;
1078 if (vstart <= iend && istart <= vend)
1081 vstart = (uintptr_t)vq->avail;
1082 vend = vstart + sizeof(struct vring_avail);
1083 vend += sizeof(uint16_t) * vq->size - 1;
1084 if (vstart <= iend && istart <= vend)
1087 vstart = (uintptr_t)vq->used;
1088 vend = vstart + sizeof(struct vring_used);
1089 vend += sizeof(struct vring_used_elem) * vq->size - 1;
1090 if (vstart <= iend && istart <= vend)
1097 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg)
1099 struct virtio_net *dev = *pdev;
1100 struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
1104 switch (imsg->type) {
1105 case VHOST_IOTLB_UPDATE:
1106 vva = qva_to_vva(dev, imsg->uaddr);
1110 for (i = 0; i < dev->nr_vring; i++) {
1111 struct vhost_virtqueue *vq = dev->virtqueue[i];
1113 vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
1114 imsg->size, imsg->perm);
1116 if (is_vring_iotlb_update(vq, imsg))
1117 *pdev = dev = translate_ring_addresses(dev, i);
1120 case VHOST_IOTLB_INVALIDATE:
1121 for (i = 0; i < dev->nr_vring; i++) {
1122 struct vhost_virtqueue *vq = dev->virtqueue[i];
1124 vhost_user_iotlb_cache_remove(vq, imsg->iova,
1127 if (is_vring_iotlb_invalidate(vq, imsg))
1128 vring_invalidate(dev, vq);
1132 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
1140 /* return bytes# of read on success or negative val on failure. */
1142 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
1146 ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
1147 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
1151 if (msg && msg->size) {
1152 if (msg->size > sizeof(msg->payload)) {
1153 RTE_LOG(ERR, VHOST_CONFIG,
1154 "invalid msg size: %d\n", msg->size);
1157 ret = read(sockfd, &msg->payload, msg->size);
1160 if (ret != (int)msg->size) {
1161 RTE_LOG(ERR, VHOST_CONFIG,
1162 "read control message failed\n");
1171 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
1176 return send_fd_message(sockfd, (char *)msg,
1177 VHOST_USER_HDR_SIZE + msg->size, NULL, 0);
1181 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
1186 msg->flags &= ~VHOST_USER_VERSION_MASK;
1187 msg->flags &= ~VHOST_USER_NEED_REPLY;
1188 msg->flags |= VHOST_USER_VERSION;
1189 msg->flags |= VHOST_USER_REPLY_MASK;
1191 return send_vhost_message(sockfd, msg);
1195 * Allocate a queue pair if it hasn't been allocated yet
1198 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev, VhostUserMsg *msg)
1202 switch (msg->request.master) {
1203 case VHOST_USER_SET_VRING_KICK:
1204 case VHOST_USER_SET_VRING_CALL:
1205 case VHOST_USER_SET_VRING_ERR:
1206 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1208 case VHOST_USER_SET_VRING_NUM:
1209 case VHOST_USER_SET_VRING_BASE:
1210 case VHOST_USER_SET_VRING_ENABLE:
1211 vring_idx = msg->payload.state.index;
1213 case VHOST_USER_SET_VRING_ADDR:
1214 vring_idx = msg->payload.addr.index;
1220 if (vring_idx >= VHOST_MAX_VRING) {
1221 RTE_LOG(ERR, VHOST_CONFIG,
1222 "invalid vring index: %u\n", vring_idx);
1226 if (dev->virtqueue[vring_idx])
1229 return alloc_vring_queue(dev, vring_idx);
1233 vhost_user_msg_handler(int vid, int fd)
1235 struct virtio_net *dev;
1236 struct VhostUserMsg msg;
1239 dev = get_device(vid);
1243 if (!dev->notify_ops) {
1244 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
1245 if (!dev->notify_ops) {
1246 RTE_LOG(ERR, VHOST_CONFIG,
1247 "failed to get callback ops for driver %s\n",
1253 ret = read_vhost_message(fd, &msg);
1254 if (ret <= 0 || msg.request.master >= VHOST_USER_MAX) {
1256 RTE_LOG(ERR, VHOST_CONFIG,
1257 "vhost read message failed\n");
1259 RTE_LOG(INFO, VHOST_CONFIG,
1260 "vhost peer closed\n");
1262 RTE_LOG(ERR, VHOST_CONFIG,
1263 "vhost read incorrect message\n");
1269 if (msg.request.master != VHOST_USER_IOTLB_MSG)
1270 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1271 vhost_message_str[msg.request.master]);
1273 RTE_LOG(DEBUG, VHOST_CONFIG, "read message %s\n",
1274 vhost_message_str[msg.request.master]);
1276 ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1278 RTE_LOG(ERR, VHOST_CONFIG,
1279 "failed to alloc queue\n");
1283 switch (msg.request.master) {
1284 case VHOST_USER_GET_FEATURES:
1285 msg.payload.u64 = vhost_user_get_features(dev);
1286 msg.size = sizeof(msg.payload.u64);
1287 send_vhost_reply(fd, &msg);
1289 case VHOST_USER_SET_FEATURES:
1290 ret = vhost_user_set_features(dev, msg.payload.u64);
1295 case VHOST_USER_GET_PROTOCOL_FEATURES:
1296 vhost_user_get_protocol_features(dev, &msg);
1297 send_vhost_reply(fd, &msg);
1299 case VHOST_USER_SET_PROTOCOL_FEATURES:
1300 vhost_user_set_protocol_features(dev, msg.payload.u64);
1303 case VHOST_USER_SET_OWNER:
1304 vhost_user_set_owner();
1306 case VHOST_USER_RESET_OWNER:
1307 vhost_user_reset_owner(dev);
1310 case VHOST_USER_SET_MEM_TABLE:
1311 ret = vhost_user_set_mem_table(dev, &msg);
1314 case VHOST_USER_SET_LOG_BASE:
1315 vhost_user_set_log_base(dev, &msg);
1317 /* it needs a reply */
1318 msg.size = sizeof(msg.payload.u64);
1319 send_vhost_reply(fd, &msg);
1321 case VHOST_USER_SET_LOG_FD:
1323 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1326 case VHOST_USER_SET_VRING_NUM:
1327 vhost_user_set_vring_num(dev, &msg);
1329 case VHOST_USER_SET_VRING_ADDR:
1330 vhost_user_set_vring_addr(&dev, &msg);
1332 case VHOST_USER_SET_VRING_BASE:
1333 vhost_user_set_vring_base(dev, &msg);
1336 case VHOST_USER_GET_VRING_BASE:
1337 vhost_user_get_vring_base(dev, &msg);
1338 msg.size = sizeof(msg.payload.state);
1339 send_vhost_reply(fd, &msg);
1342 case VHOST_USER_SET_VRING_KICK:
1343 vhost_user_set_vring_kick(&dev, &msg);
1345 case VHOST_USER_SET_VRING_CALL:
1346 vhost_user_set_vring_call(dev, &msg);
1349 case VHOST_USER_SET_VRING_ERR:
1350 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1352 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1355 case VHOST_USER_GET_QUEUE_NUM:
1356 msg.payload.u64 = VHOST_MAX_QUEUE_PAIRS;
1357 msg.size = sizeof(msg.payload.u64);
1358 send_vhost_reply(fd, &msg);
1361 case VHOST_USER_SET_VRING_ENABLE:
1362 vhost_user_set_vring_enable(dev, &msg);
1364 case VHOST_USER_SEND_RARP:
1365 vhost_user_send_rarp(dev, &msg);
1368 case VHOST_USER_NET_SET_MTU:
1369 ret = vhost_user_net_set_mtu(dev, &msg);
1372 case VHOST_USER_SET_SLAVE_REQ_FD:
1373 ret = vhost_user_set_req_fd(dev, &msg);
1376 case VHOST_USER_IOTLB_MSG:
1377 ret = vhost_user_iotlb_msg(&dev, &msg);
1386 if (msg.flags & VHOST_USER_NEED_REPLY) {
1387 msg.payload.u64 = !!ret;
1388 msg.size = sizeof(msg.payload.u64);
1389 send_vhost_reply(fd, &msg);
1392 if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1393 dev->flags |= VIRTIO_DEV_READY;
1395 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1396 if (dev->dequeue_zero_copy) {
1397 RTE_LOG(INFO, VHOST_CONFIG,
1398 "dequeue zero copy is enabled\n");
1401 if (dev->notify_ops->new_device(dev->vid) == 0)
1402 dev->flags |= VIRTIO_DEV_RUNNING;
1410 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
1413 struct VhostUserMsg msg = {
1414 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
1415 .flags = VHOST_USER_VERSION,
1416 .size = sizeof(msg.payload.iotlb),
1420 .type = VHOST_IOTLB_MISS,
1424 ret = send_vhost_message(dev->slave_req_fd, &msg);
1426 RTE_LOG(ERR, VHOST_CONFIG,
1427 "Failed to send IOTLB miss message (%d)\n",