4 * Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * * Neither the name of Intel Corporation nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
40 #include <sys/types.h>
43 #ifdef RTE_LIBRTE_VHOST_NUMA
47 #include <rte_common.h>
48 #include <rte_malloc.h>
53 #include "vhost_user.h"
55 #define VIRTIO_MIN_MTU 68
56 #define VIRTIO_MAX_MTU 65535
58 static const char *vhost_message_str[VHOST_USER_MAX] = {
59 [VHOST_USER_NONE] = "VHOST_USER_NONE",
60 [VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
61 [VHOST_USER_SET_FEATURES] = "VHOST_USER_SET_FEATURES",
62 [VHOST_USER_SET_OWNER] = "VHOST_USER_SET_OWNER",
63 [VHOST_USER_RESET_OWNER] = "VHOST_USER_RESET_OWNER",
64 [VHOST_USER_SET_MEM_TABLE] = "VHOST_USER_SET_MEM_TABLE",
65 [VHOST_USER_SET_LOG_BASE] = "VHOST_USER_SET_LOG_BASE",
66 [VHOST_USER_SET_LOG_FD] = "VHOST_USER_SET_LOG_FD",
67 [VHOST_USER_SET_VRING_NUM] = "VHOST_USER_SET_VRING_NUM",
68 [VHOST_USER_SET_VRING_ADDR] = "VHOST_USER_SET_VRING_ADDR",
69 [VHOST_USER_SET_VRING_BASE] = "VHOST_USER_SET_VRING_BASE",
70 [VHOST_USER_GET_VRING_BASE] = "VHOST_USER_GET_VRING_BASE",
71 [VHOST_USER_SET_VRING_KICK] = "VHOST_USER_SET_VRING_KICK",
72 [VHOST_USER_SET_VRING_CALL] = "VHOST_USER_SET_VRING_CALL",
73 [VHOST_USER_SET_VRING_ERR] = "VHOST_USER_SET_VRING_ERR",
74 [VHOST_USER_GET_PROTOCOL_FEATURES] = "VHOST_USER_GET_PROTOCOL_FEATURES",
75 [VHOST_USER_SET_PROTOCOL_FEATURES] = "VHOST_USER_SET_PROTOCOL_FEATURES",
76 [VHOST_USER_GET_QUEUE_NUM] = "VHOST_USER_GET_QUEUE_NUM",
77 [VHOST_USER_SET_VRING_ENABLE] = "VHOST_USER_SET_VRING_ENABLE",
78 [VHOST_USER_SEND_RARP] = "VHOST_USER_SEND_RARP",
79 [VHOST_USER_NET_SET_MTU] = "VHOST_USER_NET_SET_MTU",
80 [VHOST_USER_SET_SLAVE_REQ_FD] = "VHOST_USER_SET_SLAVE_REQ_FD",
81 [VHOST_USER_IOTLB_MSG] = "VHOST_USER_IOTLB_MSG",
90 ret = fstat(fd, &stat);
91 return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
95 free_mem_region(struct virtio_net *dev)
98 struct rte_vhost_mem_region *reg;
100 if (!dev || !dev->mem)
103 for (i = 0; i < dev->mem->nregions; i++) {
104 reg = &dev->mem->regions[i];
105 if (reg->host_user_addr) {
106 munmap(reg->mmap_addr, reg->mmap_size);
113 vhost_backend_cleanup(struct virtio_net *dev)
116 free_mem_region(dev);
121 free(dev->guest_pages);
122 dev->guest_pages = NULL;
125 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
129 if (dev->slave_req_fd >= 0) {
130 close(dev->slave_req_fd);
131 dev->slave_req_fd = -1;
136 * This function just returns success at the moment unless
137 * the device hasn't been initialised.
140 vhost_user_set_owner(void)
146 vhost_user_reset_owner(struct virtio_net *dev)
148 if (dev->flags & VIRTIO_DEV_RUNNING) {
149 dev->flags &= ~VIRTIO_DEV_RUNNING;
150 dev->notify_ops->destroy_device(dev->vid);
153 cleanup_device(dev, 0);
159 * The features that we support are requested.
162 vhost_user_get_features(struct virtio_net *dev)
164 uint64_t features = 0;
166 rte_vhost_driver_get_features(dev->ifname, &features);
171 * We receive the negotiated features supported by us and the virtio device.
174 vhost_user_set_features(struct virtio_net *dev, uint64_t features)
176 uint64_t vhost_features = 0;
178 rte_vhost_driver_get_features(dev->ifname, &vhost_features);
179 if (features & ~vhost_features) {
180 RTE_LOG(ERR, VHOST_CONFIG,
181 "(%d) received invalid negotiated features.\n",
186 if ((dev->flags & VIRTIO_DEV_RUNNING) && dev->features != features) {
187 if (dev->notify_ops->features_changed)
188 dev->notify_ops->features_changed(dev->vid, features);
191 dev->features = features;
193 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
194 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
196 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
198 LOG_DEBUG(VHOST_CONFIG,
199 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
201 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
202 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
208 * The virtio device sends us the size of the descriptor ring.
211 vhost_user_set_vring_num(struct virtio_net *dev,
214 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
216 vq->size = msg->payload.state.num;
218 if (dev->dequeue_zero_copy) {
220 vq->last_zmbuf_idx = 0;
221 vq->zmbuf_size = vq->size;
222 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
223 sizeof(struct zcopy_mbuf), 0);
224 if (vq->zmbufs == NULL) {
225 RTE_LOG(WARNING, VHOST_CONFIG,
226 "failed to allocate mem for zero copy; "
227 "zero copy is force disabled\n");
228 dev->dequeue_zero_copy = 0;
232 vq->shadow_used_ring = rte_malloc(NULL,
233 vq->size * sizeof(struct vring_used_elem),
234 RTE_CACHE_LINE_SIZE);
235 if (!vq->shadow_used_ring) {
236 RTE_LOG(ERR, VHOST_CONFIG,
237 "failed to allocate memory for shadow used ring.\n");
241 vq->batch_copy_elems = rte_malloc(NULL,
242 vq->size * sizeof(struct batch_copy_elem),
243 RTE_CACHE_LINE_SIZE);
244 if (!vq->batch_copy_elems) {
245 RTE_LOG(ERR, VHOST_CONFIG,
246 "failed to allocate memory for batching copy.\n");
254 * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
255 * same numa node as the memory of vring descriptor.
257 #ifdef RTE_LIBRTE_VHOST_NUMA
258 static struct virtio_net*
259 numa_realloc(struct virtio_net *dev, int index)
261 int oldnode, newnode;
262 struct virtio_net *old_dev;
263 struct vhost_virtqueue *old_vq, *vq;
267 vq = old_vq = dev->virtqueue[index];
269 ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
270 MPOL_F_NODE | MPOL_F_ADDR);
272 /* check if we need to reallocate vq */
273 ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
274 MPOL_F_NODE | MPOL_F_ADDR);
276 RTE_LOG(ERR, VHOST_CONFIG,
277 "Unable to get vq numa information.\n");
280 if (oldnode != newnode) {
281 RTE_LOG(INFO, VHOST_CONFIG,
282 "reallocate vq from %d to %d node\n", oldnode, newnode);
283 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
287 memcpy(vq, old_vq, sizeof(*vq));
291 /* check if we need to reallocate dev */
292 ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
293 MPOL_F_NODE | MPOL_F_ADDR);
295 RTE_LOG(ERR, VHOST_CONFIG,
296 "Unable to get dev numa information.\n");
299 if (oldnode != newnode) {
300 RTE_LOG(INFO, VHOST_CONFIG,
301 "reallocate dev from %d to %d node\n",
303 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
309 memcpy(dev, old_dev, sizeof(*dev));
314 dev->virtqueue[index] = vq;
315 vhost_devices[dev->vid] = dev;
320 static struct virtio_net*
321 numa_realloc(struct virtio_net *dev, int index __rte_unused)
327 /* Converts QEMU virtual address to Vhost virtual address. */
329 qva_to_vva(struct virtio_net *dev, uint64_t qva)
331 struct rte_vhost_mem_region *reg;
334 /* Find the region where the address lives. */
335 for (i = 0; i < dev->mem->nregions; i++) {
336 reg = &dev->mem->regions[i];
338 if (qva >= reg->guest_user_addr &&
339 qva < reg->guest_user_addr + reg->size) {
340 return qva - reg->guest_user_addr +
350 * Converts ring address to Vhost virtual address.
351 * If IOMMU is enabled, the ring address is a guest IO virtual address,
352 * else it is a QEMU virtual address.
355 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
356 uint64_t ra, uint64_t size)
358 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
361 vva = vhost_user_iotlb_cache_find(vq, ra,
362 &size, VHOST_ACCESS_RW);
364 vhost_user_iotlb_miss(dev, ra, VHOST_ACCESS_RW);
369 return qva_to_vva(dev, ra);
373 * The virtio device sends us the desc, used and avail ring addresses.
374 * This function then converts these to our address space.
377 vhost_user_set_vring_addr(struct virtio_net *dev, VhostUserMsg *msg)
379 struct vhost_virtqueue *vq;
380 struct vhost_vring_addr *addr = &msg->payload.addr;
382 if (dev->mem == NULL)
385 /* addr->index refers to the queue index. The txq 1, rxq is 0. */
386 vq = dev->virtqueue[msg->payload.addr.index];
389 * Rings addresses should not be interpreted as long as the ring is not
390 * started and enabled
392 memcpy(&vq->ring_addrs, addr, sizeof(*addr));
394 vring_invalidate(dev, vq);
399 static struct virtio_net *
400 translate_ring_addresses(struct virtio_net *dev, int vq_index)
402 struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
403 struct vhost_vring_addr *addr = &vq->ring_addrs;
405 /* The addresses are converted from QEMU virtual to Vhost virtual. */
406 if (vq->desc && vq->avail && vq->used)
409 vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
410 vq, addr->desc_user_addr, sizeof(struct vring_desc));
412 RTE_LOG(DEBUG, VHOST_CONFIG,
413 "(%d) failed to find desc ring address.\n",
418 dev = numa_realloc(dev, vq_index);
419 vq = dev->virtqueue[vq_index];
421 vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
422 vq, addr->avail_user_addr, sizeof(struct vring_avail));
423 if (vq->avail == 0) {
424 RTE_LOG(DEBUG, VHOST_CONFIG,
425 "(%d) failed to find avail ring address.\n",
430 vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
431 vq, addr->used_user_addr, sizeof(struct vring_used));
433 RTE_LOG(DEBUG, VHOST_CONFIG,
434 "(%d) failed to find used ring address.\n",
439 if (vq->last_used_idx != vq->used->idx) {
440 RTE_LOG(WARNING, VHOST_CONFIG,
441 "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
442 "some packets maybe resent for Tx and dropped for Rx\n",
443 vq->last_used_idx, vq->used->idx);
444 vq->last_used_idx = vq->used->idx;
445 vq->last_avail_idx = vq->used->idx;
448 vq->log_guest_addr = addr->log_guest_addr;
450 LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
452 LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
453 dev->vid, vq->avail);
454 LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
456 LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
457 dev->vid, vq->log_guest_addr);
463 * The virtio device sends us the available ring last used index.
466 vhost_user_set_vring_base(struct virtio_net *dev,
469 dev->virtqueue[msg->payload.state.index]->last_used_idx =
470 msg->payload.state.num;
471 dev->virtqueue[msg->payload.state.index]->last_avail_idx =
472 msg->payload.state.num;
478 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
479 uint64_t host_phys_addr, uint64_t size)
481 struct guest_page *page, *last_page;
483 if (dev->nr_guest_pages == dev->max_guest_pages) {
484 dev->max_guest_pages *= 2;
485 dev->guest_pages = realloc(dev->guest_pages,
486 dev->max_guest_pages * sizeof(*page));
489 if (dev->nr_guest_pages > 0) {
490 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
491 /* merge if the two pages are continuous */
492 if (host_phys_addr == last_page->host_phys_addr +
494 last_page->size += size;
499 page = &dev->guest_pages[dev->nr_guest_pages++];
500 page->guest_phys_addr = guest_phys_addr;
501 page->host_phys_addr = host_phys_addr;
506 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
509 uint64_t reg_size = reg->size;
510 uint64_t host_user_addr = reg->host_user_addr;
511 uint64_t guest_phys_addr = reg->guest_phys_addr;
512 uint64_t host_phys_addr;
515 host_phys_addr = rte_mem_virt2phy((void *)(uintptr_t)host_user_addr);
516 size = page_size - (guest_phys_addr & (page_size - 1));
517 size = RTE_MIN(size, reg_size);
519 add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
520 host_user_addr += size;
521 guest_phys_addr += size;
524 while (reg_size > 0) {
525 size = RTE_MIN(reg_size, page_size);
526 host_phys_addr = rte_mem_virt2phy((void *)(uintptr_t)
528 add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
530 host_user_addr += size;
531 guest_phys_addr += size;
536 #ifdef RTE_LIBRTE_VHOST_DEBUG
537 /* TODO: enable it only in debug mode? */
539 dump_guest_pages(struct virtio_net *dev)
542 struct guest_page *page;
544 for (i = 0; i < dev->nr_guest_pages; i++) {
545 page = &dev->guest_pages[i];
547 RTE_LOG(INFO, VHOST_CONFIG,
548 "guest physical page region %u\n"
549 "\t guest_phys_addr: %" PRIx64 "\n"
550 "\t host_phys_addr : %" PRIx64 "\n"
551 "\t size : %" PRIx64 "\n",
553 page->guest_phys_addr,
554 page->host_phys_addr,
559 #define dump_guest_pages(dev)
563 vhost_user_set_mem_table(struct virtio_net *dev, struct VhostUserMsg *pmsg)
565 struct VhostUserMemory memory = pmsg->payload.memory;
566 struct rte_vhost_mem_region *reg;
569 uint64_t mmap_offset;
575 free_mem_region(dev);
580 dev->nr_guest_pages = 0;
581 if (!dev->guest_pages) {
582 dev->max_guest_pages = 8;
583 dev->guest_pages = malloc(dev->max_guest_pages *
584 sizeof(struct guest_page));
585 if (dev->guest_pages == NULL) {
586 RTE_LOG(ERR, VHOST_CONFIG,
587 "(%d) failed to allocate memory "
588 "for dev->guest_pages\n",
594 dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
595 sizeof(struct rte_vhost_mem_region) * memory.nregions, 0);
596 if (dev->mem == NULL) {
597 RTE_LOG(ERR, VHOST_CONFIG,
598 "(%d) failed to allocate memory for dev->mem\n",
602 dev->mem->nregions = memory.nregions;
604 for (i = 0; i < memory.nregions; i++) {
606 reg = &dev->mem->regions[i];
608 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
609 reg->guest_user_addr = memory.regions[i].userspace_addr;
610 reg->size = memory.regions[i].memory_size;
613 mmap_offset = memory.regions[i].mmap_offset;
614 mmap_size = reg->size + mmap_offset;
616 /* mmap() without flag of MAP_ANONYMOUS, should be called
617 * with length argument aligned with hugepagesz at older
618 * longterm version Linux, like 2.6.32 and 3.2.72, or
619 * mmap() will fail with EINVAL.
621 * to avoid failure, make sure in caller to keep length
624 alignment = get_blk_size(fd);
625 if (alignment == (uint64_t)-1) {
626 RTE_LOG(ERR, VHOST_CONFIG,
627 "couldn't get hugepage size through fstat\n");
630 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
632 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
633 MAP_SHARED | MAP_POPULATE, fd, 0);
635 if (mmap_addr == MAP_FAILED) {
636 RTE_LOG(ERR, VHOST_CONFIG,
637 "mmap region %u failed.\n", i);
641 reg->mmap_addr = mmap_addr;
642 reg->mmap_size = mmap_size;
643 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
646 if (dev->dequeue_zero_copy)
647 add_guest_pages(dev, reg, alignment);
649 RTE_LOG(INFO, VHOST_CONFIG,
650 "guest memory region %u, size: 0x%" PRIx64 "\n"
651 "\t guest physical addr: 0x%" PRIx64 "\n"
652 "\t guest virtual addr: 0x%" PRIx64 "\n"
653 "\t host virtual addr: 0x%" PRIx64 "\n"
654 "\t mmap addr : 0x%" PRIx64 "\n"
655 "\t mmap size : 0x%" PRIx64 "\n"
656 "\t mmap align: 0x%" PRIx64 "\n"
657 "\t mmap off : 0x%" PRIx64 "\n",
659 reg->guest_phys_addr,
660 reg->guest_user_addr,
662 (uint64_t)(uintptr_t)mmap_addr,
668 dump_guest_pages(dev);
673 free_mem_region(dev);
680 vq_is_ready(struct vhost_virtqueue *vq)
682 return vq && vq->desc && vq->avail && vq->used &&
683 vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
684 vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
688 virtio_is_ready(struct virtio_net *dev)
690 struct vhost_virtqueue *vq;
693 if (dev->nr_vring == 0)
696 for (i = 0; i < dev->nr_vring; i++) {
697 vq = dev->virtqueue[i];
699 if (!vq_is_ready(vq))
703 RTE_LOG(INFO, VHOST_CONFIG,
704 "virtio is now ready for processing.\n");
709 vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *pmsg)
711 struct vhost_vring_file file;
712 struct vhost_virtqueue *vq;
714 file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
715 if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
716 file.fd = VIRTIO_INVALID_EVENTFD;
718 file.fd = pmsg->fds[0];
719 RTE_LOG(INFO, VHOST_CONFIG,
720 "vring call idx:%d file:%d\n", file.index, file.fd);
722 vq = dev->virtqueue[file.index];
726 vq->callfd = file.fd;
730 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *pmsg)
732 struct vhost_vring_file file;
733 struct vhost_virtqueue *vq;
734 struct virtio_net *dev = *pdev;
736 file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
737 if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
738 file.fd = VIRTIO_INVALID_EVENTFD;
740 file.fd = pmsg->fds[0];
741 RTE_LOG(INFO, VHOST_CONFIG,
742 "vring kick idx:%d file:%d\n", file.index, file.fd);
745 * Interpret ring addresses only when ring is started and enabled.
746 * This is now if protocol features aren't supported.
748 if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
749 *pdev = dev = translate_ring_addresses(dev, file.index);
754 vq = dev->virtqueue[file.index];
757 * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
758 * the ring starts already enabled. Otherwise, it is enabled via
759 * the SET_VRING_ENABLE message.
761 if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)))
766 vq->kickfd = file.fd;
770 free_zmbufs(struct vhost_virtqueue *vq)
772 struct zcopy_mbuf *zmbuf, *next;
774 for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
775 zmbuf != NULL; zmbuf = next) {
776 next = TAILQ_NEXT(zmbuf, next);
778 rte_pktmbuf_free(zmbuf->mbuf);
779 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
782 rte_free(vq->zmbufs);
786 * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
789 vhost_user_get_vring_base(struct virtio_net *dev,
792 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
794 /* We have to stop the queue (virtio) if it is running. */
795 if (dev->flags & VIRTIO_DEV_RUNNING) {
796 dev->flags &= ~VIRTIO_DEV_RUNNING;
797 dev->notify_ops->destroy_device(dev->vid);
800 dev->flags &= ~VIRTIO_DEV_READY;
802 /* Here we are safe to get the last used index */
803 msg->payload.state.num = vq->last_used_idx;
805 RTE_LOG(INFO, VHOST_CONFIG,
806 "vring base idx:%d file:%d\n", msg->payload.state.index,
807 msg->payload.state.num);
809 * Based on current qemu vhost-user implementation, this message is
810 * sent and only sent in vhost_vring_stop.
811 * TODO: cleanup the vring, it isn't usable since here.
816 vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
818 if (dev->dequeue_zero_copy)
820 rte_free(vq->shadow_used_ring);
821 vq->shadow_used_ring = NULL;
823 rte_free(vq->batch_copy_elems);
824 vq->batch_copy_elems = NULL;
830 * when virtio queues are ready to work, qemu will send us to
831 * enable the virtio queue pair.
834 vhost_user_set_vring_enable(struct virtio_net **pdev,
837 struct virtio_net *dev = *pdev;
838 int enable = (int)msg->payload.state.num;
840 RTE_LOG(INFO, VHOST_CONFIG,
841 "set queue enable: %d to qp idx: %d\n",
842 enable, msg->payload.state.index);
845 * Interpret ring addresses only when ring is started and enabled.
846 * This is now if protocol features are supported.
848 if (enable && (dev->features &
849 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
850 dev = translate_ring_addresses(dev, msg->payload.state.index);
857 if (dev->notify_ops->vring_state_changed)
858 dev->notify_ops->vring_state_changed(dev->vid,
859 msg->payload.state.index, enable);
861 dev->virtqueue[msg->payload.state.index]->enabled = enable;
867 vhost_user_set_protocol_features(struct virtio_net *dev,
868 uint64_t protocol_features)
870 if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
873 dev->protocol_features = protocol_features;
877 vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
879 int fd = msg->fds[0];
884 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
888 if (msg->size != sizeof(VhostUserLog)) {
889 RTE_LOG(ERR, VHOST_CONFIG,
890 "invalid log base msg size: %"PRId32" != %d\n",
891 msg->size, (int)sizeof(VhostUserLog));
895 size = msg->payload.log.mmap_size;
896 off = msg->payload.log.mmap_offset;
897 RTE_LOG(INFO, VHOST_CONFIG,
898 "log mmap size: %"PRId64", offset: %"PRId64"\n",
902 * mmap from 0 to workaround a hugepage mmap bug: mmap will
903 * fail when offset is not page size aligned.
905 addr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
907 if (addr == MAP_FAILED) {
908 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
913 * Free previously mapped log memory on occasionally
914 * multiple VHOST_USER_SET_LOG_BASE.
917 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
919 dev->log_addr = (uint64_t)(uintptr_t)addr;
920 dev->log_base = dev->log_addr + off;
921 dev->log_size = size;
927 * An rarp packet is constructed and broadcasted to notify switches about
928 * the new location of the migrated VM, so that packets from outside will
929 * not be lost after migration.
931 * However, we don't actually "send" a rarp packet here, instead, we set
932 * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
935 vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
937 uint8_t *mac = (uint8_t *)&msg->payload.u64;
939 RTE_LOG(DEBUG, VHOST_CONFIG,
940 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
941 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
942 memcpy(dev->mac.addr_bytes, mac, 6);
945 * Set the flag to inject a RARP broadcast packet at
946 * rte_vhost_dequeue_burst().
948 * rte_smp_wmb() is for making sure the mac is copied
949 * before the flag is set.
952 rte_atomic16_set(&dev->broadcast_rarp, 1);
958 vhost_user_net_set_mtu(struct virtio_net *dev, struct VhostUserMsg *msg)
960 if (msg->payload.u64 < VIRTIO_MIN_MTU ||
961 msg->payload.u64 > VIRTIO_MAX_MTU) {
962 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
968 dev->mtu = msg->payload.u64;
974 vhost_user_set_req_fd(struct virtio_net *dev, struct VhostUserMsg *msg)
976 int fd = msg->fds[0];
979 RTE_LOG(ERR, VHOST_CONFIG,
980 "Invalid file descriptor for slave channel (%d)\n",
985 dev->slave_req_fd = fd;
991 is_vring_iotlb_update(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
993 struct vhost_vring_addr *ra;
997 end = start + imsg->size;
999 ra = &vq->ring_addrs;
1000 if (ra->desc_user_addr >= start && ra->desc_user_addr < end)
1002 if (ra->avail_user_addr >= start && ra->avail_user_addr < end)
1004 if (ra->used_user_addr >= start && ra->used_user_addr < end)
1011 is_vring_iotlb_invalidate(struct vhost_virtqueue *vq,
1012 struct vhost_iotlb_msg *imsg)
1014 uint64_t istart, iend, vstart, vend;
1016 istart = imsg->iova;
1017 iend = istart + imsg->size - 1;
1019 vstart = (uintptr_t)vq->desc;
1020 vend = vstart + sizeof(struct vring_desc) * vq->size - 1;
1021 if (vstart <= iend && istart <= vend)
1024 vstart = (uintptr_t)vq->avail;
1025 vend = vstart + sizeof(struct vring_avail);
1026 vend += sizeof(uint16_t) * vq->size - 1;
1027 if (vstart <= iend && istart <= vend)
1030 vstart = (uintptr_t)vq->used;
1031 vend = vstart + sizeof(struct vring_used);
1032 vend += sizeof(struct vring_used_elem) * vq->size - 1;
1033 if (vstart <= iend && istart <= vend)
1040 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg)
1042 struct virtio_net *dev = *pdev;
1043 struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
1047 switch (imsg->type) {
1048 case VHOST_IOTLB_UPDATE:
1049 vva = qva_to_vva(dev, imsg->uaddr);
1053 for (i = 0; i < dev->nr_vring; i++) {
1054 struct vhost_virtqueue *vq = dev->virtqueue[i];
1056 vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
1057 imsg->size, imsg->perm);
1059 if (is_vring_iotlb_update(vq, imsg))
1060 *pdev = dev = translate_ring_addresses(dev, i);
1063 case VHOST_IOTLB_INVALIDATE:
1064 for (i = 0; i < dev->nr_vring; i++) {
1065 struct vhost_virtqueue *vq = dev->virtqueue[i];
1067 vhost_user_iotlb_cache_remove(vq, imsg->iova,
1070 if (is_vring_iotlb_invalidate(vq, imsg))
1071 vring_invalidate(dev, vq);
1075 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
1083 /* return bytes# of read on success or negative val on failure. */
1085 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
1089 ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
1090 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
1094 if (msg && msg->size) {
1095 if (msg->size > sizeof(msg->payload)) {
1096 RTE_LOG(ERR, VHOST_CONFIG,
1097 "invalid msg size: %d\n", msg->size);
1100 ret = read(sockfd, &msg->payload, msg->size);
1103 if (ret != (int)msg->size) {
1104 RTE_LOG(ERR, VHOST_CONFIG,
1105 "read control message failed\n");
1114 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
1119 return send_fd_message(sockfd, (char *)msg,
1120 VHOST_USER_HDR_SIZE + msg->size, NULL, 0);
1124 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
1129 msg->flags &= ~VHOST_USER_VERSION_MASK;
1130 msg->flags &= ~VHOST_USER_NEED_REPLY;
1131 msg->flags |= VHOST_USER_VERSION;
1132 msg->flags |= VHOST_USER_REPLY_MASK;
1134 return send_vhost_message(sockfd, msg);
1138 * Allocate a queue pair if it hasn't been allocated yet
1141 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev, VhostUserMsg *msg)
1145 switch (msg->request) {
1146 case VHOST_USER_SET_VRING_KICK:
1147 case VHOST_USER_SET_VRING_CALL:
1148 case VHOST_USER_SET_VRING_ERR:
1149 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1151 case VHOST_USER_SET_VRING_NUM:
1152 case VHOST_USER_SET_VRING_BASE:
1153 case VHOST_USER_SET_VRING_ENABLE:
1154 vring_idx = msg->payload.state.index;
1156 case VHOST_USER_SET_VRING_ADDR:
1157 vring_idx = msg->payload.addr.index;
1163 if (vring_idx >= VHOST_MAX_VRING) {
1164 RTE_LOG(ERR, VHOST_CONFIG,
1165 "invalid vring index: %u\n", vring_idx);
1169 if (dev->virtqueue[vring_idx])
1172 return alloc_vring_queue(dev, vring_idx);
1176 vhost_user_msg_handler(int vid, int fd)
1178 struct virtio_net *dev;
1179 struct VhostUserMsg msg;
1182 dev = get_device(vid);
1186 if (!dev->notify_ops) {
1187 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
1188 if (!dev->notify_ops) {
1189 RTE_LOG(ERR, VHOST_CONFIG,
1190 "failed to get callback ops for driver %s\n",
1196 ret = read_vhost_message(fd, &msg);
1197 if (ret <= 0 || msg.request >= VHOST_USER_MAX) {
1199 RTE_LOG(ERR, VHOST_CONFIG,
1200 "vhost read message failed\n");
1202 RTE_LOG(INFO, VHOST_CONFIG,
1203 "vhost peer closed\n");
1205 RTE_LOG(ERR, VHOST_CONFIG,
1206 "vhost read incorrect message\n");
1212 if (msg.request != VHOST_USER_IOTLB_MSG)
1213 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1214 vhost_message_str[msg.request]);
1216 RTE_LOG(DEBUG, VHOST_CONFIG, "read message %s\n",
1217 vhost_message_str[msg.request]);
1219 ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1221 RTE_LOG(ERR, VHOST_CONFIG,
1222 "failed to alloc queue\n");
1226 switch (msg.request) {
1227 case VHOST_USER_GET_FEATURES:
1228 msg.payload.u64 = vhost_user_get_features(dev);
1229 msg.size = sizeof(msg.payload.u64);
1230 send_vhost_reply(fd, &msg);
1232 case VHOST_USER_SET_FEATURES:
1233 vhost_user_set_features(dev, msg.payload.u64);
1236 case VHOST_USER_GET_PROTOCOL_FEATURES:
1237 msg.payload.u64 = VHOST_USER_PROTOCOL_FEATURES;
1238 msg.size = sizeof(msg.payload.u64);
1239 send_vhost_reply(fd, &msg);
1241 case VHOST_USER_SET_PROTOCOL_FEATURES:
1242 vhost_user_set_protocol_features(dev, msg.payload.u64);
1245 case VHOST_USER_SET_OWNER:
1246 vhost_user_set_owner();
1248 case VHOST_USER_RESET_OWNER:
1249 vhost_user_reset_owner(dev);
1252 case VHOST_USER_SET_MEM_TABLE:
1253 ret = vhost_user_set_mem_table(dev, &msg);
1256 case VHOST_USER_SET_LOG_BASE:
1257 vhost_user_set_log_base(dev, &msg);
1259 /* it needs a reply */
1260 msg.size = sizeof(msg.payload.u64);
1261 send_vhost_reply(fd, &msg);
1263 case VHOST_USER_SET_LOG_FD:
1265 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1268 case VHOST_USER_SET_VRING_NUM:
1269 vhost_user_set_vring_num(dev, &msg);
1271 case VHOST_USER_SET_VRING_ADDR:
1272 vhost_user_set_vring_addr(dev, &msg);
1274 case VHOST_USER_SET_VRING_BASE:
1275 vhost_user_set_vring_base(dev, &msg);
1278 case VHOST_USER_GET_VRING_BASE:
1279 vhost_user_get_vring_base(dev, &msg);
1280 msg.size = sizeof(msg.payload.state);
1281 send_vhost_reply(fd, &msg);
1284 case VHOST_USER_SET_VRING_KICK:
1285 vhost_user_set_vring_kick(&dev, &msg);
1287 case VHOST_USER_SET_VRING_CALL:
1288 vhost_user_set_vring_call(dev, &msg);
1291 case VHOST_USER_SET_VRING_ERR:
1292 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1294 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1297 case VHOST_USER_GET_QUEUE_NUM:
1298 msg.payload.u64 = VHOST_MAX_QUEUE_PAIRS;
1299 msg.size = sizeof(msg.payload.u64);
1300 send_vhost_reply(fd, &msg);
1303 case VHOST_USER_SET_VRING_ENABLE:
1304 vhost_user_set_vring_enable(&dev, &msg);
1306 case VHOST_USER_SEND_RARP:
1307 vhost_user_send_rarp(dev, &msg);
1310 case VHOST_USER_NET_SET_MTU:
1311 ret = vhost_user_net_set_mtu(dev, &msg);
1314 case VHOST_USER_SET_SLAVE_REQ_FD:
1315 ret = vhost_user_set_req_fd(dev, &msg);
1318 case VHOST_USER_IOTLB_MSG:
1319 ret = vhost_user_iotlb_msg(&dev, &msg);
1328 if (msg.flags & VHOST_USER_NEED_REPLY) {
1329 msg.payload.u64 = !!ret;
1330 msg.size = sizeof(msg.payload.u64);
1331 send_vhost_reply(fd, &msg);
1334 if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1335 dev->flags |= VIRTIO_DEV_READY;
1337 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1338 if (dev->dequeue_zero_copy) {
1339 RTE_LOG(INFO, VHOST_CONFIG,
1340 "dequeue zero copy is enabled\n");
1343 if (dev->notify_ops->new_device(dev->vid) == 0)
1344 dev->flags |= VIRTIO_DEV_RUNNING;
1352 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
1355 struct VhostUserMsg msg = {
1356 .request = (enum VhostUserRequest)VHOST_USER_SLAVE_IOTLB_MSG,
1357 .flags = VHOST_USER_VERSION,
1358 .size = sizeof(msg.payload.iotlb),
1362 .type = VHOST_IOTLB_MISS,
1366 ret = send_vhost_message(dev->slave_req_fd, &msg);
1368 RTE_LOG(ERR, VHOST_CONFIG,
1369 "Failed to send IOTLB miss message (%d)\n",