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) && dev->features != features) {
158 if (dev->notify_ops->features_changed)
159 dev->notify_ops->features_changed(dev->vid, features);
162 dev->features = features;
164 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
165 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
167 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
169 LOG_DEBUG(VHOST_CONFIG,
170 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
172 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
173 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
179 * The virtio device sends us the size of the descriptor ring.
182 vhost_user_set_vring_num(struct virtio_net *dev,
185 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
187 vq->size = msg->payload.state.num;
189 if (dev->dequeue_zero_copy) {
191 vq->last_zmbuf_idx = 0;
192 vq->zmbuf_size = vq->size;
193 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
194 sizeof(struct zcopy_mbuf), 0);
195 if (vq->zmbufs == NULL) {
196 RTE_LOG(WARNING, VHOST_CONFIG,
197 "failed to allocate mem for zero copy; "
198 "zero copy is force disabled\n");
199 dev->dequeue_zero_copy = 0;
203 vq->shadow_used_ring = rte_malloc(NULL,
204 vq->size * sizeof(struct vring_used_elem),
205 RTE_CACHE_LINE_SIZE);
206 if (!vq->shadow_used_ring) {
207 RTE_LOG(ERR, VHOST_CONFIG,
208 "failed to allocate memory for shadow used ring.\n");
212 vq->batch_copy_elems = rte_malloc(NULL,
213 vq->size * sizeof(struct batch_copy_elem),
214 RTE_CACHE_LINE_SIZE);
215 if (!vq->batch_copy_elems) {
216 RTE_LOG(ERR, VHOST_CONFIG,
217 "failed to allocate memory for batching copy.\n");
225 * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
226 * same numa node as the memory of vring descriptor.
228 #ifdef RTE_LIBRTE_VHOST_NUMA
229 static struct virtio_net*
230 numa_realloc(struct virtio_net *dev, int index)
232 int oldnode, newnode;
233 struct virtio_net *old_dev;
234 struct vhost_virtqueue *old_vq, *vq;
238 vq = old_vq = dev->virtqueue[index];
240 ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
241 MPOL_F_NODE | MPOL_F_ADDR);
243 /* check if we need to reallocate vq */
244 ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
245 MPOL_F_NODE | MPOL_F_ADDR);
247 RTE_LOG(ERR, VHOST_CONFIG,
248 "Unable to get vq numa information.\n");
251 if (oldnode != newnode) {
252 RTE_LOG(INFO, VHOST_CONFIG,
253 "reallocate vq from %d to %d node\n", oldnode, newnode);
254 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
258 memcpy(vq, old_vq, sizeof(*vq));
262 /* check if we need to reallocate dev */
263 ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
264 MPOL_F_NODE | MPOL_F_ADDR);
266 RTE_LOG(ERR, VHOST_CONFIG,
267 "Unable to get dev numa information.\n");
270 if (oldnode != newnode) {
271 RTE_LOG(INFO, VHOST_CONFIG,
272 "reallocate dev from %d to %d node\n",
274 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
280 memcpy(dev, old_dev, sizeof(*dev));
285 dev->virtqueue[index] = vq;
286 vhost_devices[dev->vid] = dev;
289 vhost_user_iotlb_init(dev, index);
294 static struct virtio_net*
295 numa_realloc(struct virtio_net *dev, int index __rte_unused)
301 /* Converts QEMU virtual address to Vhost virtual address. */
303 qva_to_vva(struct virtio_net *dev, uint64_t qva)
305 struct rte_vhost_mem_region *reg;
308 /* Find the region where the address lives. */
309 for (i = 0; i < dev->mem->nregions; i++) {
310 reg = &dev->mem->regions[i];
312 if (qva >= reg->guest_user_addr &&
313 qva < reg->guest_user_addr + reg->size) {
314 return qva - reg->guest_user_addr +
324 * Converts ring address to Vhost virtual address.
325 * If IOMMU is enabled, the ring address is a guest IO virtual address,
326 * else it is a QEMU virtual address.
329 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
330 uint64_t ra, uint64_t size)
332 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
335 vva = vhost_user_iotlb_cache_find(vq, ra,
336 &size, VHOST_ACCESS_RW);
338 vhost_user_iotlb_miss(dev, ra, VHOST_ACCESS_RW);
343 return qva_to_vva(dev, ra);
346 static struct virtio_net *
347 translate_ring_addresses(struct virtio_net *dev, int vq_index)
349 struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
350 struct vhost_vring_addr *addr = &vq->ring_addrs;
352 /* The addresses are converted from QEMU virtual to Vhost virtual. */
353 if (vq->desc && vq->avail && vq->used)
356 vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
357 vq, addr->desc_user_addr, sizeof(struct vring_desc));
359 RTE_LOG(DEBUG, VHOST_CONFIG,
360 "(%d) failed to find desc ring address.\n",
365 dev = numa_realloc(dev, vq_index);
366 vq = dev->virtqueue[vq_index];
367 addr = &vq->ring_addrs;
369 vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
370 vq, addr->avail_user_addr, sizeof(struct vring_avail));
371 if (vq->avail == 0) {
372 RTE_LOG(DEBUG, VHOST_CONFIG,
373 "(%d) failed to find avail ring address.\n",
378 vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
379 vq, addr->used_user_addr, sizeof(struct vring_used));
381 RTE_LOG(DEBUG, VHOST_CONFIG,
382 "(%d) failed to find used ring address.\n",
387 if (vq->last_used_idx != vq->used->idx) {
388 RTE_LOG(WARNING, VHOST_CONFIG,
389 "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
390 "some packets maybe resent for Tx and dropped for Rx\n",
391 vq->last_used_idx, vq->used->idx);
392 vq->last_used_idx = vq->used->idx;
393 vq->last_avail_idx = vq->used->idx;
396 vq->log_guest_addr = addr->log_guest_addr;
398 LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
400 LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
401 dev->vid, vq->avail);
402 LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
404 LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
405 dev->vid, vq->log_guest_addr);
411 * The virtio device sends us the desc, used and avail ring addresses.
412 * This function then converts these to our address space.
415 vhost_user_set_vring_addr(struct virtio_net **pdev, VhostUserMsg *msg)
417 struct vhost_virtqueue *vq;
418 struct vhost_vring_addr *addr = &msg->payload.addr;
419 struct virtio_net *dev = *pdev;
421 if (dev->mem == NULL)
424 /* addr->index refers to the queue index. The txq 1, rxq is 0. */
425 vq = dev->virtqueue[msg->payload.addr.index];
428 * Rings addresses should not be interpreted as long as the ring is not
429 * started and enabled
431 memcpy(&vq->ring_addrs, addr, sizeof(*addr));
433 vring_invalidate(dev, vq);
435 if (vq->enabled && (dev->features &
436 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
437 dev = translate_ring_addresses(dev, msg->payload.state.index);
448 * The virtio device sends us the available ring last used index.
451 vhost_user_set_vring_base(struct virtio_net *dev,
454 dev->virtqueue[msg->payload.state.index]->last_used_idx =
455 msg->payload.state.num;
456 dev->virtqueue[msg->payload.state.index]->last_avail_idx =
457 msg->payload.state.num;
463 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
464 uint64_t host_phys_addr, uint64_t size)
466 struct guest_page *page, *last_page;
468 if (dev->nr_guest_pages == dev->max_guest_pages) {
469 dev->max_guest_pages *= 2;
470 dev->guest_pages = realloc(dev->guest_pages,
471 dev->max_guest_pages * sizeof(*page));
474 if (dev->nr_guest_pages > 0) {
475 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
476 /* merge if the two pages are continuous */
477 if (host_phys_addr == last_page->host_phys_addr +
479 last_page->size += size;
484 page = &dev->guest_pages[dev->nr_guest_pages++];
485 page->guest_phys_addr = guest_phys_addr;
486 page->host_phys_addr = host_phys_addr;
491 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
494 uint64_t reg_size = reg->size;
495 uint64_t host_user_addr = reg->host_user_addr;
496 uint64_t guest_phys_addr = reg->guest_phys_addr;
497 uint64_t host_phys_addr;
500 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
501 size = page_size - (guest_phys_addr & (page_size - 1));
502 size = RTE_MIN(size, reg_size);
504 add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
505 host_user_addr += size;
506 guest_phys_addr += size;
509 while (reg_size > 0) {
510 size = RTE_MIN(reg_size, page_size);
511 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
513 add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
515 host_user_addr += size;
516 guest_phys_addr += size;
521 #ifdef RTE_LIBRTE_VHOST_DEBUG
522 /* TODO: enable it only in debug mode? */
524 dump_guest_pages(struct virtio_net *dev)
527 struct guest_page *page;
529 for (i = 0; i < dev->nr_guest_pages; i++) {
530 page = &dev->guest_pages[i];
532 RTE_LOG(INFO, VHOST_CONFIG,
533 "guest physical page region %u\n"
534 "\t guest_phys_addr: %" PRIx64 "\n"
535 "\t host_phys_addr : %" PRIx64 "\n"
536 "\t size : %" PRIx64 "\n",
538 page->guest_phys_addr,
539 page->host_phys_addr,
544 #define dump_guest_pages(dev)
548 vhost_memory_changed(struct VhostUserMemory *new,
549 struct rte_vhost_memory *old)
553 if (new->nregions != old->nregions)
556 for (i = 0; i < new->nregions; ++i) {
557 VhostUserMemoryRegion *new_r = &new->regions[i];
558 struct rte_vhost_mem_region *old_r = &old->regions[i];
560 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
562 if (new_r->memory_size != old_r->size)
564 if (new_r->userspace_addr != old_r->guest_user_addr)
572 vhost_user_set_mem_table(struct virtio_net *dev, struct VhostUserMsg *pmsg)
574 struct VhostUserMemory memory = pmsg->payload.memory;
575 struct rte_vhost_mem_region *reg;
578 uint64_t mmap_offset;
583 if (dev->mem && !vhost_memory_changed(&memory, dev->mem)) {
584 RTE_LOG(INFO, VHOST_CONFIG,
585 "(%d) memory regions not changed\n", dev->vid);
587 for (i = 0; i < memory.nregions; i++)
594 free_mem_region(dev);
599 dev->nr_guest_pages = 0;
600 if (!dev->guest_pages) {
601 dev->max_guest_pages = 8;
602 dev->guest_pages = malloc(dev->max_guest_pages *
603 sizeof(struct guest_page));
604 if (dev->guest_pages == NULL) {
605 RTE_LOG(ERR, VHOST_CONFIG,
606 "(%d) failed to allocate memory "
607 "for dev->guest_pages\n",
613 dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
614 sizeof(struct rte_vhost_mem_region) * memory.nregions, 0);
615 if (dev->mem == NULL) {
616 RTE_LOG(ERR, VHOST_CONFIG,
617 "(%d) failed to allocate memory for dev->mem\n",
621 dev->mem->nregions = memory.nregions;
623 for (i = 0; i < memory.nregions; i++) {
625 reg = &dev->mem->regions[i];
627 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
628 reg->guest_user_addr = memory.regions[i].userspace_addr;
629 reg->size = memory.regions[i].memory_size;
632 mmap_offset = memory.regions[i].mmap_offset;
633 mmap_size = reg->size + mmap_offset;
635 /* mmap() without flag of MAP_ANONYMOUS, should be called
636 * with length argument aligned with hugepagesz at older
637 * longterm version Linux, like 2.6.32 and 3.2.72, or
638 * mmap() will fail with EINVAL.
640 * to avoid failure, make sure in caller to keep length
643 alignment = get_blk_size(fd);
644 if (alignment == (uint64_t)-1) {
645 RTE_LOG(ERR, VHOST_CONFIG,
646 "couldn't get hugepage size through fstat\n");
649 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
651 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
652 MAP_SHARED | MAP_POPULATE, fd, 0);
654 if (mmap_addr == MAP_FAILED) {
655 RTE_LOG(ERR, VHOST_CONFIG,
656 "mmap region %u failed.\n", i);
660 reg->mmap_addr = mmap_addr;
661 reg->mmap_size = mmap_size;
662 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
665 if (dev->dequeue_zero_copy)
666 add_guest_pages(dev, reg, alignment);
668 RTE_LOG(INFO, VHOST_CONFIG,
669 "guest memory region %u, size: 0x%" PRIx64 "\n"
670 "\t guest physical addr: 0x%" PRIx64 "\n"
671 "\t guest virtual addr: 0x%" PRIx64 "\n"
672 "\t host virtual addr: 0x%" PRIx64 "\n"
673 "\t mmap addr : 0x%" PRIx64 "\n"
674 "\t mmap size : 0x%" PRIx64 "\n"
675 "\t mmap align: 0x%" PRIx64 "\n"
676 "\t mmap off : 0x%" PRIx64 "\n",
678 reg->guest_phys_addr,
679 reg->guest_user_addr,
681 (uint64_t)(uintptr_t)mmap_addr,
687 dump_guest_pages(dev);
692 free_mem_region(dev);
699 vq_is_ready(struct vhost_virtqueue *vq)
701 return vq && vq->desc && vq->avail && vq->used &&
702 vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
703 vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
707 virtio_is_ready(struct virtio_net *dev)
709 struct vhost_virtqueue *vq;
712 if (dev->nr_vring == 0)
715 for (i = 0; i < dev->nr_vring; i++) {
716 vq = dev->virtqueue[i];
718 if (!vq_is_ready(vq))
722 RTE_LOG(INFO, VHOST_CONFIG,
723 "virtio is now ready for processing.\n");
728 vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *pmsg)
730 struct vhost_vring_file file;
731 struct vhost_virtqueue *vq;
733 file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
734 if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
735 file.fd = VIRTIO_INVALID_EVENTFD;
737 file.fd = pmsg->fds[0];
738 RTE_LOG(INFO, VHOST_CONFIG,
739 "vring call idx:%d file:%d\n", file.index, file.fd);
741 vq = dev->virtqueue[file.index];
745 vq->callfd = file.fd;
749 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *pmsg)
751 struct vhost_vring_file file;
752 struct vhost_virtqueue *vq;
753 struct virtio_net *dev = *pdev;
755 file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
756 if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
757 file.fd = VIRTIO_INVALID_EVENTFD;
759 file.fd = pmsg->fds[0];
760 RTE_LOG(INFO, VHOST_CONFIG,
761 "vring kick idx:%d file:%d\n", file.index, file.fd);
763 /* Interpret ring addresses only when ring is started. */
764 dev = translate_ring_addresses(dev, file.index);
770 vq = dev->virtqueue[file.index];
773 * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
774 * the ring starts already enabled. Otherwise, it is enabled via
775 * the SET_VRING_ENABLE message.
777 if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)))
782 vq->kickfd = file.fd;
786 free_zmbufs(struct vhost_virtqueue *vq)
788 struct zcopy_mbuf *zmbuf, *next;
790 for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
791 zmbuf != NULL; zmbuf = next) {
792 next = TAILQ_NEXT(zmbuf, next);
794 rte_pktmbuf_free(zmbuf->mbuf);
795 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
798 rte_free(vq->zmbufs);
802 * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
805 vhost_user_get_vring_base(struct virtio_net *dev,
808 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
810 /* We have to stop the queue (virtio) if it is running. */
811 if (dev->flags & VIRTIO_DEV_RUNNING) {
812 dev->flags &= ~VIRTIO_DEV_RUNNING;
813 dev->notify_ops->destroy_device(dev->vid);
816 dev->flags &= ~VIRTIO_DEV_READY;
818 /* Here we are safe to get the last used index */
819 msg->payload.state.num = vq->last_used_idx;
821 RTE_LOG(INFO, VHOST_CONFIG,
822 "vring base idx:%d file:%d\n", msg->payload.state.index,
823 msg->payload.state.num);
825 * Based on current qemu vhost-user implementation, this message is
826 * sent and only sent in vhost_vring_stop.
827 * TODO: cleanup the vring, it isn't usable since here.
832 vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
834 if (dev->dequeue_zero_copy)
836 rte_free(vq->shadow_used_ring);
837 vq->shadow_used_ring = NULL;
839 rte_free(vq->batch_copy_elems);
840 vq->batch_copy_elems = NULL;
846 * when virtio queues are ready to work, qemu will send us to
847 * enable the virtio queue pair.
850 vhost_user_set_vring_enable(struct virtio_net *dev,
853 int enable = (int)msg->payload.state.num;
855 RTE_LOG(INFO, VHOST_CONFIG,
856 "set queue enable: %d to qp idx: %d\n",
857 enable, msg->payload.state.index);
859 if (dev->notify_ops->vring_state_changed)
860 dev->notify_ops->vring_state_changed(dev->vid,
861 msg->payload.state.index, enable);
863 dev->virtqueue[msg->payload.state.index]->enabled = enable;
869 vhost_user_get_protocol_features(struct virtio_net *dev,
870 struct VhostUserMsg *msg)
872 uint64_t features, protocol_features = VHOST_USER_PROTOCOL_FEATURES;
874 rte_vhost_driver_get_features(dev->ifname, &features);
877 * REPLY_ACK protocol feature is only mandatory for now
878 * for IOMMU feature. If IOMMU is explicitly disabled by the
879 * application, disable also REPLY_ACK feature for older buggy
880 * Qemu versions (from v2.7.0 to v2.9.0).
882 if (!(features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
883 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK);
885 msg->payload.u64 = protocol_features;
886 msg->size = sizeof(msg->payload.u64);
890 vhost_user_set_protocol_features(struct virtio_net *dev,
891 uint64_t protocol_features)
893 if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
896 dev->protocol_features = protocol_features;
900 vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
902 int fd = msg->fds[0];
907 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
911 if (msg->size != sizeof(VhostUserLog)) {
912 RTE_LOG(ERR, VHOST_CONFIG,
913 "invalid log base msg size: %"PRId32" != %d\n",
914 msg->size, (int)sizeof(VhostUserLog));
918 size = msg->payload.log.mmap_size;
919 off = msg->payload.log.mmap_offset;
920 RTE_LOG(INFO, VHOST_CONFIG,
921 "log mmap size: %"PRId64", offset: %"PRId64"\n",
925 * mmap from 0 to workaround a hugepage mmap bug: mmap will
926 * fail when offset is not page size aligned.
928 addr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
930 if (addr == MAP_FAILED) {
931 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
936 * Free previously mapped log memory on occasionally
937 * multiple VHOST_USER_SET_LOG_BASE.
940 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
942 dev->log_addr = (uint64_t)(uintptr_t)addr;
943 dev->log_base = dev->log_addr + off;
944 dev->log_size = size;
950 * An rarp packet is constructed and broadcasted to notify switches about
951 * the new location of the migrated VM, so that packets from outside will
952 * not be lost after migration.
954 * However, we don't actually "send" a rarp packet here, instead, we set
955 * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
958 vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
960 uint8_t *mac = (uint8_t *)&msg->payload.u64;
962 RTE_LOG(DEBUG, VHOST_CONFIG,
963 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
964 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
965 memcpy(dev->mac.addr_bytes, mac, 6);
968 * Set the flag to inject a RARP broadcast packet at
969 * rte_vhost_dequeue_burst().
971 * rte_smp_wmb() is for making sure the mac is copied
972 * before the flag is set.
975 rte_atomic16_set(&dev->broadcast_rarp, 1);
981 vhost_user_net_set_mtu(struct virtio_net *dev, struct VhostUserMsg *msg)
983 if (msg->payload.u64 < VIRTIO_MIN_MTU ||
984 msg->payload.u64 > VIRTIO_MAX_MTU) {
985 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
991 dev->mtu = msg->payload.u64;
997 vhost_user_set_req_fd(struct virtio_net *dev, struct VhostUserMsg *msg)
999 int fd = msg->fds[0];
1002 RTE_LOG(ERR, VHOST_CONFIG,
1003 "Invalid file descriptor for slave channel (%d)\n",
1008 dev->slave_req_fd = fd;
1014 is_vring_iotlb_update(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
1016 struct vhost_vring_addr *ra;
1017 uint64_t start, end;
1020 end = start + imsg->size;
1022 ra = &vq->ring_addrs;
1023 if (ra->desc_user_addr >= start && ra->desc_user_addr < end)
1025 if (ra->avail_user_addr >= start && ra->avail_user_addr < end)
1027 if (ra->used_user_addr >= start && ra->used_user_addr < end)
1034 is_vring_iotlb_invalidate(struct vhost_virtqueue *vq,
1035 struct vhost_iotlb_msg *imsg)
1037 uint64_t istart, iend, vstart, vend;
1039 istart = imsg->iova;
1040 iend = istart + imsg->size - 1;
1042 vstart = (uintptr_t)vq->desc;
1043 vend = vstart + sizeof(struct vring_desc) * vq->size - 1;
1044 if (vstart <= iend && istart <= vend)
1047 vstart = (uintptr_t)vq->avail;
1048 vend = vstart + sizeof(struct vring_avail);
1049 vend += sizeof(uint16_t) * vq->size - 1;
1050 if (vstart <= iend && istart <= vend)
1053 vstart = (uintptr_t)vq->used;
1054 vend = vstart + sizeof(struct vring_used);
1055 vend += sizeof(struct vring_used_elem) * vq->size - 1;
1056 if (vstart <= iend && istart <= vend)
1063 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg)
1065 struct virtio_net *dev = *pdev;
1066 struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
1070 switch (imsg->type) {
1071 case VHOST_IOTLB_UPDATE:
1072 vva = qva_to_vva(dev, imsg->uaddr);
1076 for (i = 0; i < dev->nr_vring; i++) {
1077 struct vhost_virtqueue *vq = dev->virtqueue[i];
1079 vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
1080 imsg->size, imsg->perm);
1082 if (is_vring_iotlb_update(vq, imsg))
1083 *pdev = dev = translate_ring_addresses(dev, i);
1086 case VHOST_IOTLB_INVALIDATE:
1087 for (i = 0; i < dev->nr_vring; i++) {
1088 struct vhost_virtqueue *vq = dev->virtqueue[i];
1090 vhost_user_iotlb_cache_remove(vq, imsg->iova,
1093 if (is_vring_iotlb_invalidate(vq, imsg))
1094 vring_invalidate(dev, vq);
1098 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
1106 /* return bytes# of read on success or negative val on failure. */
1108 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
1112 ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
1113 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
1117 if (msg && msg->size) {
1118 if (msg->size > sizeof(msg->payload)) {
1119 RTE_LOG(ERR, VHOST_CONFIG,
1120 "invalid msg size: %d\n", msg->size);
1123 ret = read(sockfd, &msg->payload, msg->size);
1126 if (ret != (int)msg->size) {
1127 RTE_LOG(ERR, VHOST_CONFIG,
1128 "read control message failed\n");
1137 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
1142 return send_fd_message(sockfd, (char *)msg,
1143 VHOST_USER_HDR_SIZE + msg->size, NULL, 0);
1147 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
1152 msg->flags &= ~VHOST_USER_VERSION_MASK;
1153 msg->flags &= ~VHOST_USER_NEED_REPLY;
1154 msg->flags |= VHOST_USER_VERSION;
1155 msg->flags |= VHOST_USER_REPLY_MASK;
1157 return send_vhost_message(sockfd, msg);
1161 * Allocate a queue pair if it hasn't been allocated yet
1164 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev, VhostUserMsg *msg)
1168 switch (msg->request.master) {
1169 case VHOST_USER_SET_VRING_KICK:
1170 case VHOST_USER_SET_VRING_CALL:
1171 case VHOST_USER_SET_VRING_ERR:
1172 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1174 case VHOST_USER_SET_VRING_NUM:
1175 case VHOST_USER_SET_VRING_BASE:
1176 case VHOST_USER_SET_VRING_ENABLE:
1177 vring_idx = msg->payload.state.index;
1179 case VHOST_USER_SET_VRING_ADDR:
1180 vring_idx = msg->payload.addr.index;
1186 if (vring_idx >= VHOST_MAX_VRING) {
1187 RTE_LOG(ERR, VHOST_CONFIG,
1188 "invalid vring index: %u\n", vring_idx);
1192 if (dev->virtqueue[vring_idx])
1195 return alloc_vring_queue(dev, vring_idx);
1199 vhost_user_msg_handler(int vid, int fd)
1201 struct virtio_net *dev;
1202 struct VhostUserMsg msg;
1205 dev = get_device(vid);
1209 if (!dev->notify_ops) {
1210 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
1211 if (!dev->notify_ops) {
1212 RTE_LOG(ERR, VHOST_CONFIG,
1213 "failed to get callback ops for driver %s\n",
1219 ret = read_vhost_message(fd, &msg);
1220 if (ret <= 0 || msg.request.master >= VHOST_USER_MAX) {
1222 RTE_LOG(ERR, VHOST_CONFIG,
1223 "vhost read message failed\n");
1225 RTE_LOG(INFO, VHOST_CONFIG,
1226 "vhost peer closed\n");
1228 RTE_LOG(ERR, VHOST_CONFIG,
1229 "vhost read incorrect message\n");
1235 if (msg.request.master != VHOST_USER_IOTLB_MSG)
1236 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1237 vhost_message_str[msg.request.master]);
1239 RTE_LOG(DEBUG, VHOST_CONFIG, "read message %s\n",
1240 vhost_message_str[msg.request.master]);
1242 ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1244 RTE_LOG(ERR, VHOST_CONFIG,
1245 "failed to alloc queue\n");
1249 switch (msg.request.master) {
1250 case VHOST_USER_GET_FEATURES:
1251 msg.payload.u64 = vhost_user_get_features(dev);
1252 msg.size = sizeof(msg.payload.u64);
1253 send_vhost_reply(fd, &msg);
1255 case VHOST_USER_SET_FEATURES:
1256 vhost_user_set_features(dev, msg.payload.u64);
1259 case VHOST_USER_GET_PROTOCOL_FEATURES:
1260 vhost_user_get_protocol_features(dev, &msg);
1261 send_vhost_reply(fd, &msg);
1263 case VHOST_USER_SET_PROTOCOL_FEATURES:
1264 vhost_user_set_protocol_features(dev, msg.payload.u64);
1267 case VHOST_USER_SET_OWNER:
1268 vhost_user_set_owner();
1270 case VHOST_USER_RESET_OWNER:
1271 vhost_user_reset_owner(dev);
1274 case VHOST_USER_SET_MEM_TABLE:
1275 ret = vhost_user_set_mem_table(dev, &msg);
1278 case VHOST_USER_SET_LOG_BASE:
1279 vhost_user_set_log_base(dev, &msg);
1281 /* it needs a reply */
1282 msg.size = sizeof(msg.payload.u64);
1283 send_vhost_reply(fd, &msg);
1285 case VHOST_USER_SET_LOG_FD:
1287 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1290 case VHOST_USER_SET_VRING_NUM:
1291 vhost_user_set_vring_num(dev, &msg);
1293 case VHOST_USER_SET_VRING_ADDR:
1294 vhost_user_set_vring_addr(&dev, &msg);
1296 case VHOST_USER_SET_VRING_BASE:
1297 vhost_user_set_vring_base(dev, &msg);
1300 case VHOST_USER_GET_VRING_BASE:
1301 vhost_user_get_vring_base(dev, &msg);
1302 msg.size = sizeof(msg.payload.state);
1303 send_vhost_reply(fd, &msg);
1306 case VHOST_USER_SET_VRING_KICK:
1307 vhost_user_set_vring_kick(&dev, &msg);
1309 case VHOST_USER_SET_VRING_CALL:
1310 vhost_user_set_vring_call(dev, &msg);
1313 case VHOST_USER_SET_VRING_ERR:
1314 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1316 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1319 case VHOST_USER_GET_QUEUE_NUM:
1320 msg.payload.u64 = VHOST_MAX_QUEUE_PAIRS;
1321 msg.size = sizeof(msg.payload.u64);
1322 send_vhost_reply(fd, &msg);
1325 case VHOST_USER_SET_VRING_ENABLE:
1326 vhost_user_set_vring_enable(dev, &msg);
1328 case VHOST_USER_SEND_RARP:
1329 vhost_user_send_rarp(dev, &msg);
1332 case VHOST_USER_NET_SET_MTU:
1333 ret = vhost_user_net_set_mtu(dev, &msg);
1336 case VHOST_USER_SET_SLAVE_REQ_FD:
1337 ret = vhost_user_set_req_fd(dev, &msg);
1340 case VHOST_USER_IOTLB_MSG:
1341 ret = vhost_user_iotlb_msg(&dev, &msg);
1350 if (msg.flags & VHOST_USER_NEED_REPLY) {
1351 msg.payload.u64 = !!ret;
1352 msg.size = sizeof(msg.payload.u64);
1353 send_vhost_reply(fd, &msg);
1356 if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1357 dev->flags |= VIRTIO_DEV_READY;
1359 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1360 if (dev->dequeue_zero_copy) {
1361 RTE_LOG(INFO, VHOST_CONFIG,
1362 "dequeue zero copy is enabled\n");
1365 if (dev->notify_ops->new_device(dev->vid) == 0)
1366 dev->flags |= VIRTIO_DEV_RUNNING;
1374 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
1377 struct VhostUserMsg msg = {
1378 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
1379 .flags = VHOST_USER_VERSION,
1380 .size = sizeof(msg.payload.iotlb),
1384 .type = VHOST_IOTLB_MISS,
1388 ret = send_vhost_message(dev->slave_req_fd, &msg);
1390 RTE_LOG(ERR, VHOST_CONFIG,
1391 "Failed to send IOTLB miss message (%d)\n",