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)
328 * Converts QEMU virtual address to Vhost virtual address. This function is
329 * used to convert the ring addresses to our address space.
332 qva_to_vva(struct virtio_net *dev, uint64_t qva)
334 struct rte_vhost_mem_region *reg;
337 /* Find the region where the address lives. */
338 for (i = 0; i < dev->mem->nregions; i++) {
339 reg = &dev->mem->regions[i];
341 if (qva >= reg->guest_user_addr &&
342 qva < reg->guest_user_addr + reg->size) {
343 return qva - reg->guest_user_addr +
352 * The virtio device sends us the desc, used and avail ring addresses.
353 * This function then converts these to our address space.
356 vhost_user_set_vring_addr(struct virtio_net *dev, VhostUserMsg *msg)
358 struct vhost_virtqueue *vq;
360 if (dev->mem == NULL)
363 /* addr->index refers to the queue index. The txq 1, rxq is 0. */
364 vq = dev->virtqueue[msg->payload.addr.index];
366 /* The addresses are converted from QEMU virtual to Vhost virtual. */
367 vq->desc = (struct vring_desc *)(uintptr_t)qva_to_vva(dev,
368 msg->payload.addr.desc_user_addr);
370 RTE_LOG(ERR, VHOST_CONFIG,
371 "(%d) failed to find desc ring address.\n",
376 dev = numa_realloc(dev, msg->payload.addr.index);
377 vq = dev->virtqueue[msg->payload.addr.index];
379 vq->avail = (struct vring_avail *)(uintptr_t)qva_to_vva(dev,
380 msg->payload.addr.avail_user_addr);
381 if (vq->avail == 0) {
382 RTE_LOG(ERR, VHOST_CONFIG,
383 "(%d) failed to find avail ring address.\n",
388 vq->used = (struct vring_used *)(uintptr_t)qva_to_vva(dev,
389 msg->payload.addr.used_user_addr);
391 RTE_LOG(ERR, VHOST_CONFIG,
392 "(%d) failed to find used ring address.\n",
397 if (vq->last_used_idx != vq->used->idx) {
398 RTE_LOG(WARNING, VHOST_CONFIG,
399 "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
400 "some packets maybe resent for Tx and dropped for Rx\n",
401 vq->last_used_idx, vq->used->idx);
402 vq->last_used_idx = vq->used->idx;
403 vq->last_avail_idx = vq->used->idx;
406 vq->log_guest_addr = msg->payload.addr.log_guest_addr;
408 LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
410 LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
411 dev->vid, vq->avail);
412 LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
414 LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
415 dev->vid, vq->log_guest_addr);
421 * The virtio device sends us the available ring last used index.
424 vhost_user_set_vring_base(struct virtio_net *dev,
427 dev->virtqueue[msg->payload.state.index]->last_used_idx =
428 msg->payload.state.num;
429 dev->virtqueue[msg->payload.state.index]->last_avail_idx =
430 msg->payload.state.num;
436 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
437 uint64_t host_phys_addr, uint64_t size)
439 struct guest_page *page, *last_page;
441 if (dev->nr_guest_pages == dev->max_guest_pages) {
442 dev->max_guest_pages *= 2;
443 dev->guest_pages = realloc(dev->guest_pages,
444 dev->max_guest_pages * sizeof(*page));
447 if (dev->nr_guest_pages > 0) {
448 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
449 /* merge if the two pages are continuous */
450 if (host_phys_addr == last_page->host_phys_addr +
452 last_page->size += size;
457 page = &dev->guest_pages[dev->nr_guest_pages++];
458 page->guest_phys_addr = guest_phys_addr;
459 page->host_phys_addr = host_phys_addr;
464 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
467 uint64_t reg_size = reg->size;
468 uint64_t host_user_addr = reg->host_user_addr;
469 uint64_t guest_phys_addr = reg->guest_phys_addr;
470 uint64_t host_phys_addr;
473 host_phys_addr = rte_mem_virt2phy((void *)(uintptr_t)host_user_addr);
474 size = page_size - (guest_phys_addr & (page_size - 1));
475 size = RTE_MIN(size, reg_size);
477 add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
478 host_user_addr += size;
479 guest_phys_addr += size;
482 while (reg_size > 0) {
483 size = RTE_MIN(reg_size, page_size);
484 host_phys_addr = rte_mem_virt2phy((void *)(uintptr_t)
486 add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
488 host_user_addr += size;
489 guest_phys_addr += size;
494 #ifdef RTE_LIBRTE_VHOST_DEBUG
495 /* TODO: enable it only in debug mode? */
497 dump_guest_pages(struct virtio_net *dev)
500 struct guest_page *page;
502 for (i = 0; i < dev->nr_guest_pages; i++) {
503 page = &dev->guest_pages[i];
505 RTE_LOG(INFO, VHOST_CONFIG,
506 "guest physical page region %u\n"
507 "\t guest_phys_addr: %" PRIx64 "\n"
508 "\t host_phys_addr : %" PRIx64 "\n"
509 "\t size : %" PRIx64 "\n",
511 page->guest_phys_addr,
512 page->host_phys_addr,
517 #define dump_guest_pages(dev)
521 vhost_user_set_mem_table(struct virtio_net *dev, struct VhostUserMsg *pmsg)
523 struct VhostUserMemory memory = pmsg->payload.memory;
524 struct rte_vhost_mem_region *reg;
527 uint64_t mmap_offset;
533 free_mem_region(dev);
538 dev->nr_guest_pages = 0;
539 if (!dev->guest_pages) {
540 dev->max_guest_pages = 8;
541 dev->guest_pages = malloc(dev->max_guest_pages *
542 sizeof(struct guest_page));
543 if (dev->guest_pages == NULL) {
544 RTE_LOG(ERR, VHOST_CONFIG,
545 "(%d) failed to allocate memory "
546 "for dev->guest_pages\n",
552 dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
553 sizeof(struct rte_vhost_mem_region) * memory.nregions, 0);
554 if (dev->mem == NULL) {
555 RTE_LOG(ERR, VHOST_CONFIG,
556 "(%d) failed to allocate memory for dev->mem\n",
560 dev->mem->nregions = memory.nregions;
562 for (i = 0; i < memory.nregions; i++) {
564 reg = &dev->mem->regions[i];
566 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
567 reg->guest_user_addr = memory.regions[i].userspace_addr;
568 reg->size = memory.regions[i].memory_size;
571 mmap_offset = memory.regions[i].mmap_offset;
572 mmap_size = reg->size + mmap_offset;
574 /* mmap() without flag of MAP_ANONYMOUS, should be called
575 * with length argument aligned with hugepagesz at older
576 * longterm version Linux, like 2.6.32 and 3.2.72, or
577 * mmap() will fail with EINVAL.
579 * to avoid failure, make sure in caller to keep length
582 alignment = get_blk_size(fd);
583 if (alignment == (uint64_t)-1) {
584 RTE_LOG(ERR, VHOST_CONFIG,
585 "couldn't get hugepage size through fstat\n");
588 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
590 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
591 MAP_SHARED | MAP_POPULATE, fd, 0);
593 if (mmap_addr == MAP_FAILED) {
594 RTE_LOG(ERR, VHOST_CONFIG,
595 "mmap region %u failed.\n", i);
599 reg->mmap_addr = mmap_addr;
600 reg->mmap_size = mmap_size;
601 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
604 if (dev->dequeue_zero_copy)
605 add_guest_pages(dev, reg, alignment);
607 RTE_LOG(INFO, VHOST_CONFIG,
608 "guest memory region %u, size: 0x%" PRIx64 "\n"
609 "\t guest physical addr: 0x%" PRIx64 "\n"
610 "\t guest virtual addr: 0x%" PRIx64 "\n"
611 "\t host virtual addr: 0x%" PRIx64 "\n"
612 "\t mmap addr : 0x%" PRIx64 "\n"
613 "\t mmap size : 0x%" PRIx64 "\n"
614 "\t mmap align: 0x%" PRIx64 "\n"
615 "\t mmap off : 0x%" PRIx64 "\n",
617 reg->guest_phys_addr,
618 reg->guest_user_addr,
620 (uint64_t)(uintptr_t)mmap_addr,
626 dump_guest_pages(dev);
631 free_mem_region(dev);
638 vq_is_ready(struct vhost_virtqueue *vq)
640 return vq && vq->desc &&
641 vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
642 vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
646 virtio_is_ready(struct virtio_net *dev)
648 struct vhost_virtqueue *vq;
651 if (dev->nr_vring == 0)
654 for (i = 0; i < dev->nr_vring; i++) {
655 vq = dev->virtqueue[i];
657 if (!vq_is_ready(vq))
661 RTE_LOG(INFO, VHOST_CONFIG,
662 "virtio is now ready for processing.\n");
667 vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *pmsg)
669 struct vhost_vring_file file;
670 struct vhost_virtqueue *vq;
672 file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
673 if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
674 file.fd = VIRTIO_INVALID_EVENTFD;
676 file.fd = pmsg->fds[0];
677 RTE_LOG(INFO, VHOST_CONFIG,
678 "vring call idx:%d file:%d\n", file.index, file.fd);
680 vq = dev->virtqueue[file.index];
684 vq->callfd = file.fd;
688 vhost_user_set_vring_kick(struct virtio_net *dev, struct VhostUserMsg *pmsg)
690 struct vhost_vring_file file;
691 struct vhost_virtqueue *vq;
693 file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
694 if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
695 file.fd = VIRTIO_INVALID_EVENTFD;
697 file.fd = pmsg->fds[0];
698 RTE_LOG(INFO, VHOST_CONFIG,
699 "vring kick idx:%d file:%d\n", file.index, file.fd);
701 vq = dev->virtqueue[file.index];
704 * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
705 * the ring starts already enabled. Otherwise, it is enabled via
706 * the SET_VRING_ENABLE message.
708 if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)))
713 vq->kickfd = file.fd;
717 free_zmbufs(struct vhost_virtqueue *vq)
719 struct zcopy_mbuf *zmbuf, *next;
721 for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
722 zmbuf != NULL; zmbuf = next) {
723 next = TAILQ_NEXT(zmbuf, next);
725 rte_pktmbuf_free(zmbuf->mbuf);
726 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
729 rte_free(vq->zmbufs);
733 * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
736 vhost_user_get_vring_base(struct virtio_net *dev,
739 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
741 /* We have to stop the queue (virtio) if it is running. */
742 if (dev->flags & VIRTIO_DEV_RUNNING) {
743 dev->flags &= ~VIRTIO_DEV_RUNNING;
744 dev->notify_ops->destroy_device(dev->vid);
747 dev->flags &= ~VIRTIO_DEV_READY;
749 /* Here we are safe to get the last used index */
750 msg->payload.state.num = vq->last_used_idx;
752 RTE_LOG(INFO, VHOST_CONFIG,
753 "vring base idx:%d file:%d\n", msg->payload.state.index,
754 msg->payload.state.num);
756 * Based on current qemu vhost-user implementation, this message is
757 * sent and only sent in vhost_vring_stop.
758 * TODO: cleanup the vring, it isn't usable since here.
763 vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
765 if (dev->dequeue_zero_copy)
767 rte_free(vq->shadow_used_ring);
768 vq->shadow_used_ring = NULL;
770 rte_free(vq->batch_copy_elems);
771 vq->batch_copy_elems = NULL;
777 * when virtio queues are ready to work, qemu will send us to
778 * enable the virtio queue pair.
781 vhost_user_set_vring_enable(struct virtio_net *dev,
784 int enable = (int)msg->payload.state.num;
786 RTE_LOG(INFO, VHOST_CONFIG,
787 "set queue enable: %d to qp idx: %d\n",
788 enable, msg->payload.state.index);
790 if (dev->notify_ops->vring_state_changed)
791 dev->notify_ops->vring_state_changed(dev->vid,
792 msg->payload.state.index, enable);
794 dev->virtqueue[msg->payload.state.index]->enabled = enable;
800 vhost_user_set_protocol_features(struct virtio_net *dev,
801 uint64_t protocol_features)
803 if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
806 dev->protocol_features = protocol_features;
810 vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
812 int fd = msg->fds[0];
817 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
821 if (msg->size != sizeof(VhostUserLog)) {
822 RTE_LOG(ERR, VHOST_CONFIG,
823 "invalid log base msg size: %"PRId32" != %d\n",
824 msg->size, (int)sizeof(VhostUserLog));
828 size = msg->payload.log.mmap_size;
829 off = msg->payload.log.mmap_offset;
830 RTE_LOG(INFO, VHOST_CONFIG,
831 "log mmap size: %"PRId64", offset: %"PRId64"\n",
835 * mmap from 0 to workaround a hugepage mmap bug: mmap will
836 * fail when offset is not page size aligned.
838 addr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
840 if (addr == MAP_FAILED) {
841 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
846 * Free previously mapped log memory on occasionally
847 * multiple VHOST_USER_SET_LOG_BASE.
850 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
852 dev->log_addr = (uint64_t)(uintptr_t)addr;
853 dev->log_base = dev->log_addr + off;
854 dev->log_size = size;
860 * An rarp packet is constructed and broadcasted to notify switches about
861 * the new location of the migrated VM, so that packets from outside will
862 * not be lost after migration.
864 * However, we don't actually "send" a rarp packet here, instead, we set
865 * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
868 vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
870 uint8_t *mac = (uint8_t *)&msg->payload.u64;
872 RTE_LOG(DEBUG, VHOST_CONFIG,
873 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
874 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
875 memcpy(dev->mac.addr_bytes, mac, 6);
878 * Set the flag to inject a RARP broadcast packet at
879 * rte_vhost_dequeue_burst().
881 * rte_smp_wmb() is for making sure the mac is copied
882 * before the flag is set.
885 rte_atomic16_set(&dev->broadcast_rarp, 1);
891 vhost_user_net_set_mtu(struct virtio_net *dev, struct VhostUserMsg *msg)
893 if (msg->payload.u64 < VIRTIO_MIN_MTU ||
894 msg->payload.u64 > VIRTIO_MAX_MTU) {
895 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
901 dev->mtu = msg->payload.u64;
907 vhost_user_set_req_fd(struct virtio_net *dev, struct VhostUserMsg *msg)
909 int fd = msg->fds[0];
912 RTE_LOG(ERR, VHOST_CONFIG,
913 "Invalid file descriptor for slave channel (%d)\n",
918 dev->slave_req_fd = fd;
924 vhost_user_iotlb_msg(struct virtio_net *dev, struct VhostUserMsg *msg)
926 struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
930 switch (imsg->type) {
931 case VHOST_IOTLB_UPDATE:
932 vva = qva_to_vva(dev, imsg->uaddr);
936 for (i = 0; i < dev->nr_vring; i++) {
937 struct vhost_virtqueue *vq = dev->virtqueue[i];
939 vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
940 imsg->size, imsg->perm);
943 case VHOST_IOTLB_INVALIDATE:
944 for (i = 0; i < dev->nr_vring; i++) {
945 struct vhost_virtqueue *vq = dev->virtqueue[i];
947 vhost_user_iotlb_cache_remove(vq, imsg->iova,
952 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
960 /* return bytes# of read on success or negative val on failure. */
962 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
966 ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
967 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
971 if (msg && msg->size) {
972 if (msg->size > sizeof(msg->payload)) {
973 RTE_LOG(ERR, VHOST_CONFIG,
974 "invalid msg size: %d\n", msg->size);
977 ret = read(sockfd, &msg->payload, msg->size);
980 if (ret != (int)msg->size) {
981 RTE_LOG(ERR, VHOST_CONFIG,
982 "read control message failed\n");
991 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
996 return send_fd_message(sockfd, (char *)msg,
997 VHOST_USER_HDR_SIZE + msg->size, NULL, 0);
1001 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
1006 msg->flags &= ~VHOST_USER_VERSION_MASK;
1007 msg->flags &= ~VHOST_USER_NEED_REPLY;
1008 msg->flags |= VHOST_USER_VERSION;
1009 msg->flags |= VHOST_USER_REPLY_MASK;
1011 return send_vhost_message(sockfd, msg);
1015 * Allocate a queue pair if it hasn't been allocated yet
1018 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev, VhostUserMsg *msg)
1022 switch (msg->request) {
1023 case VHOST_USER_SET_VRING_KICK:
1024 case VHOST_USER_SET_VRING_CALL:
1025 case VHOST_USER_SET_VRING_ERR:
1026 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1028 case VHOST_USER_SET_VRING_NUM:
1029 case VHOST_USER_SET_VRING_BASE:
1030 case VHOST_USER_SET_VRING_ENABLE:
1031 vring_idx = msg->payload.state.index;
1033 case VHOST_USER_SET_VRING_ADDR:
1034 vring_idx = msg->payload.addr.index;
1040 if (vring_idx >= VHOST_MAX_VRING) {
1041 RTE_LOG(ERR, VHOST_CONFIG,
1042 "invalid vring index: %u\n", vring_idx);
1046 if (dev->virtqueue[vring_idx])
1049 return alloc_vring_queue(dev, vring_idx);
1053 vhost_user_msg_handler(int vid, int fd)
1055 struct virtio_net *dev;
1056 struct VhostUserMsg msg;
1059 dev = get_device(vid);
1063 if (!dev->notify_ops) {
1064 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
1065 if (!dev->notify_ops) {
1066 RTE_LOG(ERR, VHOST_CONFIG,
1067 "failed to get callback ops for driver %s\n",
1073 ret = read_vhost_message(fd, &msg);
1074 if (ret <= 0 || msg.request >= VHOST_USER_MAX) {
1076 RTE_LOG(ERR, VHOST_CONFIG,
1077 "vhost read message failed\n");
1079 RTE_LOG(INFO, VHOST_CONFIG,
1080 "vhost peer closed\n");
1082 RTE_LOG(ERR, VHOST_CONFIG,
1083 "vhost read incorrect message\n");
1089 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1090 vhost_message_str[msg.request]);
1092 ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1094 RTE_LOG(ERR, VHOST_CONFIG,
1095 "failed to alloc queue\n");
1099 switch (msg.request) {
1100 case VHOST_USER_GET_FEATURES:
1101 msg.payload.u64 = vhost_user_get_features(dev);
1102 msg.size = sizeof(msg.payload.u64);
1103 send_vhost_reply(fd, &msg);
1105 case VHOST_USER_SET_FEATURES:
1106 vhost_user_set_features(dev, msg.payload.u64);
1109 case VHOST_USER_GET_PROTOCOL_FEATURES:
1110 msg.payload.u64 = VHOST_USER_PROTOCOL_FEATURES;
1111 msg.size = sizeof(msg.payload.u64);
1112 send_vhost_reply(fd, &msg);
1114 case VHOST_USER_SET_PROTOCOL_FEATURES:
1115 vhost_user_set_protocol_features(dev, msg.payload.u64);
1118 case VHOST_USER_SET_OWNER:
1119 vhost_user_set_owner();
1121 case VHOST_USER_RESET_OWNER:
1122 vhost_user_reset_owner(dev);
1125 case VHOST_USER_SET_MEM_TABLE:
1126 ret = vhost_user_set_mem_table(dev, &msg);
1129 case VHOST_USER_SET_LOG_BASE:
1130 vhost_user_set_log_base(dev, &msg);
1132 /* it needs a reply */
1133 msg.size = sizeof(msg.payload.u64);
1134 send_vhost_reply(fd, &msg);
1136 case VHOST_USER_SET_LOG_FD:
1138 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1141 case VHOST_USER_SET_VRING_NUM:
1142 vhost_user_set_vring_num(dev, &msg);
1144 case VHOST_USER_SET_VRING_ADDR:
1145 vhost_user_set_vring_addr(dev, &msg);
1147 case VHOST_USER_SET_VRING_BASE:
1148 vhost_user_set_vring_base(dev, &msg);
1151 case VHOST_USER_GET_VRING_BASE:
1152 vhost_user_get_vring_base(dev, &msg);
1153 msg.size = sizeof(msg.payload.state);
1154 send_vhost_reply(fd, &msg);
1157 case VHOST_USER_SET_VRING_KICK:
1158 vhost_user_set_vring_kick(dev, &msg);
1160 case VHOST_USER_SET_VRING_CALL:
1161 vhost_user_set_vring_call(dev, &msg);
1164 case VHOST_USER_SET_VRING_ERR:
1165 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1167 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1170 case VHOST_USER_GET_QUEUE_NUM:
1171 msg.payload.u64 = VHOST_MAX_QUEUE_PAIRS;
1172 msg.size = sizeof(msg.payload.u64);
1173 send_vhost_reply(fd, &msg);
1176 case VHOST_USER_SET_VRING_ENABLE:
1177 vhost_user_set_vring_enable(dev, &msg);
1179 case VHOST_USER_SEND_RARP:
1180 vhost_user_send_rarp(dev, &msg);
1183 case VHOST_USER_NET_SET_MTU:
1184 ret = vhost_user_net_set_mtu(dev, &msg);
1187 case VHOST_USER_SET_SLAVE_REQ_FD:
1188 ret = vhost_user_set_req_fd(dev, &msg);
1191 case VHOST_USER_IOTLB_MSG:
1192 ret = vhost_user_iotlb_msg(dev, &msg);
1202 * The virtio_net struct might have been reallocated on a different
1203 * NUMA node, so dev pointer might no more be valid.
1205 dev = get_device(vid);
1207 if (msg.flags & VHOST_USER_NEED_REPLY) {
1208 msg.payload.u64 = !!ret;
1209 msg.size = sizeof(msg.payload.u64);
1210 send_vhost_reply(fd, &msg);
1213 if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1214 dev->flags |= VIRTIO_DEV_READY;
1216 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1217 if (dev->dequeue_zero_copy) {
1218 RTE_LOG(INFO, VHOST_CONFIG,
1219 "dequeue zero copy is enabled\n");
1222 if (dev->notify_ops->new_device(dev->vid) == 0)
1223 dev->flags |= VIRTIO_DEV_RUNNING;
1231 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
1234 struct VhostUserMsg msg = {
1235 .request = (enum VhostUserRequest)VHOST_USER_SLAVE_IOTLB_MSG,
1236 .flags = VHOST_USER_VERSION,
1237 .size = sizeof(msg.payload.iotlb),
1241 .type = VHOST_IOTLB_MISS,
1245 ret = send_vhost_message(dev->slave_req_fd, &msg);
1247 RTE_LOG(ERR, VHOST_CONFIG,
1248 "Failed to send IOTLB miss message (%d)\n",