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>
52 #include "vhost_user.h"
54 #define VIRTIO_MIN_MTU 68
55 #define VIRTIO_MAX_MTU 65535
57 static const char *vhost_message_str[VHOST_USER_MAX] = {
58 [VHOST_USER_NONE] = "VHOST_USER_NONE",
59 [VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
60 [VHOST_USER_SET_FEATURES] = "VHOST_USER_SET_FEATURES",
61 [VHOST_USER_SET_OWNER] = "VHOST_USER_SET_OWNER",
62 [VHOST_USER_RESET_OWNER] = "VHOST_USER_RESET_OWNER",
63 [VHOST_USER_SET_MEM_TABLE] = "VHOST_USER_SET_MEM_TABLE",
64 [VHOST_USER_SET_LOG_BASE] = "VHOST_USER_SET_LOG_BASE",
65 [VHOST_USER_SET_LOG_FD] = "VHOST_USER_SET_LOG_FD",
66 [VHOST_USER_SET_VRING_NUM] = "VHOST_USER_SET_VRING_NUM",
67 [VHOST_USER_SET_VRING_ADDR] = "VHOST_USER_SET_VRING_ADDR",
68 [VHOST_USER_SET_VRING_BASE] = "VHOST_USER_SET_VRING_BASE",
69 [VHOST_USER_GET_VRING_BASE] = "VHOST_USER_GET_VRING_BASE",
70 [VHOST_USER_SET_VRING_KICK] = "VHOST_USER_SET_VRING_KICK",
71 [VHOST_USER_SET_VRING_CALL] = "VHOST_USER_SET_VRING_CALL",
72 [VHOST_USER_SET_VRING_ERR] = "VHOST_USER_SET_VRING_ERR",
73 [VHOST_USER_GET_PROTOCOL_FEATURES] = "VHOST_USER_GET_PROTOCOL_FEATURES",
74 [VHOST_USER_SET_PROTOCOL_FEATURES] = "VHOST_USER_SET_PROTOCOL_FEATURES",
75 [VHOST_USER_GET_QUEUE_NUM] = "VHOST_USER_GET_QUEUE_NUM",
76 [VHOST_USER_SET_VRING_ENABLE] = "VHOST_USER_SET_VRING_ENABLE",
77 [VHOST_USER_SEND_RARP] = "VHOST_USER_SEND_RARP",
78 [VHOST_USER_NET_SET_MTU] = "VHOST_USER_NET_SET_MTU",
87 ret = fstat(fd, &stat);
88 return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
92 free_mem_region(struct virtio_net *dev)
95 struct rte_vhost_mem_region *reg;
97 if (!dev || !dev->mem)
100 for (i = 0; i < dev->mem->nregions; i++) {
101 reg = &dev->mem->regions[i];
102 if (reg->host_user_addr) {
103 munmap(reg->mmap_addr, reg->mmap_size);
110 vhost_backend_cleanup(struct virtio_net *dev)
113 free_mem_region(dev);
118 free(dev->guest_pages);
119 dev->guest_pages = NULL;
122 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
128 * This function just returns success at the moment unless
129 * the device hasn't been initialised.
132 vhost_user_set_owner(void)
138 vhost_user_reset_owner(struct virtio_net *dev)
140 if (dev->flags & VIRTIO_DEV_RUNNING) {
141 dev->flags &= ~VIRTIO_DEV_RUNNING;
142 dev->notify_ops->destroy_device(dev->vid);
145 cleanup_device(dev, 0);
151 * The features that we support are requested.
154 vhost_user_get_features(struct virtio_net *dev)
156 uint64_t features = 0;
158 rte_vhost_driver_get_features(dev->ifname, &features);
163 * We receive the negotiated features supported by us and the virtio device.
166 vhost_user_set_features(struct virtio_net *dev, uint64_t features)
168 uint64_t vhost_features = 0;
170 rte_vhost_driver_get_features(dev->ifname, &vhost_features);
171 if (features & ~vhost_features) {
172 RTE_LOG(ERR, VHOST_CONFIG,
173 "(%d) received invalid negotiated features.\n",
178 if ((dev->flags & VIRTIO_DEV_RUNNING) && dev->features != features) {
179 if (dev->notify_ops->features_changed)
180 dev->notify_ops->features_changed(dev->vid, features);
183 dev->features = features;
185 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
186 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
188 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
190 LOG_DEBUG(VHOST_CONFIG,
191 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
193 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
194 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
200 * The virtio device sends us the size of the descriptor ring.
203 vhost_user_set_vring_num(struct virtio_net *dev,
206 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
208 vq->size = msg->payload.state.num;
210 if (dev->dequeue_zero_copy) {
212 vq->last_zmbuf_idx = 0;
213 vq->zmbuf_size = vq->size;
214 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
215 sizeof(struct zcopy_mbuf), 0);
216 if (vq->zmbufs == NULL) {
217 RTE_LOG(WARNING, VHOST_CONFIG,
218 "failed to allocate mem for zero copy; "
219 "zero copy is force disabled\n");
220 dev->dequeue_zero_copy = 0;
224 vq->shadow_used_ring = rte_malloc(NULL,
225 vq->size * sizeof(struct vring_used_elem),
226 RTE_CACHE_LINE_SIZE);
227 if (!vq->shadow_used_ring) {
228 RTE_LOG(ERR, VHOST_CONFIG,
229 "failed to allocate memory for shadow used ring.\n");
233 vq->batch_copy_elems = rte_malloc(NULL,
234 vq->size * sizeof(struct batch_copy_elem),
235 RTE_CACHE_LINE_SIZE);
236 if (!vq->batch_copy_elems) {
237 RTE_LOG(ERR, VHOST_CONFIG,
238 "failed to allocate memory for batching copy.\n");
246 * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
247 * same numa node as the memory of vring descriptor.
249 #ifdef RTE_LIBRTE_VHOST_NUMA
250 static struct virtio_net*
251 numa_realloc(struct virtio_net *dev, int index)
253 int oldnode, newnode;
254 struct virtio_net *old_dev;
255 struct vhost_virtqueue *old_vq, *vq;
259 vq = old_vq = dev->virtqueue[index];
261 ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
262 MPOL_F_NODE | MPOL_F_ADDR);
264 /* check if we need to reallocate vq */
265 ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
266 MPOL_F_NODE | MPOL_F_ADDR);
268 RTE_LOG(ERR, VHOST_CONFIG,
269 "Unable to get vq numa information.\n");
272 if (oldnode != newnode) {
273 RTE_LOG(INFO, VHOST_CONFIG,
274 "reallocate vq from %d to %d node\n", oldnode, newnode);
275 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
279 memcpy(vq, old_vq, sizeof(*vq));
283 /* check if we need to reallocate dev */
284 ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
285 MPOL_F_NODE | MPOL_F_ADDR);
287 RTE_LOG(ERR, VHOST_CONFIG,
288 "Unable to get dev numa information.\n");
291 if (oldnode != newnode) {
292 RTE_LOG(INFO, VHOST_CONFIG,
293 "reallocate dev from %d to %d node\n",
295 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
301 memcpy(dev, old_dev, sizeof(*dev));
306 dev->virtqueue[index] = vq;
307 vhost_devices[dev->vid] = dev;
312 static struct virtio_net*
313 numa_realloc(struct virtio_net *dev, int index __rte_unused)
320 * Converts QEMU virtual address to Vhost virtual address. This function is
321 * used to convert the ring addresses to our address space.
324 qva_to_vva(struct virtio_net *dev, uint64_t qva)
326 struct rte_vhost_mem_region *reg;
329 /* Find the region where the address lives. */
330 for (i = 0; i < dev->mem->nregions; i++) {
331 reg = &dev->mem->regions[i];
333 if (qva >= reg->guest_user_addr &&
334 qva < reg->guest_user_addr + reg->size) {
335 return qva - reg->guest_user_addr +
344 * The virtio device sends us the desc, used and avail ring addresses.
345 * This function then converts these to our address space.
348 vhost_user_set_vring_addr(struct virtio_net *dev, VhostUserMsg *msg)
350 struct vhost_virtqueue *vq;
352 if (dev->mem == NULL)
355 /* addr->index refers to the queue index. The txq 1, rxq is 0. */
356 vq = dev->virtqueue[msg->payload.addr.index];
358 /* The addresses are converted from QEMU virtual to Vhost virtual. */
359 vq->desc = (struct vring_desc *)(uintptr_t)qva_to_vva(dev,
360 msg->payload.addr.desc_user_addr);
362 RTE_LOG(ERR, VHOST_CONFIG,
363 "(%d) failed to find desc ring address.\n",
368 dev = numa_realloc(dev, msg->payload.addr.index);
369 vq = dev->virtqueue[msg->payload.addr.index];
371 vq->avail = (struct vring_avail *)(uintptr_t)qva_to_vva(dev,
372 msg->payload.addr.avail_user_addr);
373 if (vq->avail == 0) {
374 RTE_LOG(ERR, VHOST_CONFIG,
375 "(%d) failed to find avail ring address.\n",
380 vq->used = (struct vring_used *)(uintptr_t)qva_to_vva(dev,
381 msg->payload.addr.used_user_addr);
383 RTE_LOG(ERR, VHOST_CONFIG,
384 "(%d) failed to find used ring address.\n",
389 if (vq->last_used_idx != vq->used->idx) {
390 RTE_LOG(WARNING, VHOST_CONFIG,
391 "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
392 "some packets maybe resent for Tx and dropped for Rx\n",
393 vq->last_used_idx, vq->used->idx);
394 vq->last_used_idx = vq->used->idx;
395 vq->last_avail_idx = vq->used->idx;
398 vq->log_guest_addr = msg->payload.addr.log_guest_addr;
400 LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
402 LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
403 dev->vid, vq->avail);
404 LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
406 LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
407 dev->vid, vq->log_guest_addr);
413 * The virtio device sends us the available ring last used index.
416 vhost_user_set_vring_base(struct virtio_net *dev,
419 dev->virtqueue[msg->payload.state.index]->last_used_idx =
420 msg->payload.state.num;
421 dev->virtqueue[msg->payload.state.index]->last_avail_idx =
422 msg->payload.state.num;
428 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
429 uint64_t host_phys_addr, uint64_t size)
431 struct guest_page *page, *last_page;
433 if (dev->nr_guest_pages == dev->max_guest_pages) {
434 dev->max_guest_pages *= 2;
435 dev->guest_pages = realloc(dev->guest_pages,
436 dev->max_guest_pages * sizeof(*page));
439 if (dev->nr_guest_pages > 0) {
440 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
441 /* merge if the two pages are continuous */
442 if (host_phys_addr == last_page->host_phys_addr +
444 last_page->size += size;
449 page = &dev->guest_pages[dev->nr_guest_pages++];
450 page->guest_phys_addr = guest_phys_addr;
451 page->host_phys_addr = host_phys_addr;
456 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
459 uint64_t reg_size = reg->size;
460 uint64_t host_user_addr = reg->host_user_addr;
461 uint64_t guest_phys_addr = reg->guest_phys_addr;
462 uint64_t host_phys_addr;
465 host_phys_addr = rte_mem_virt2phy((void *)(uintptr_t)host_user_addr);
466 size = page_size - (guest_phys_addr & (page_size - 1));
467 size = RTE_MIN(size, reg_size);
469 add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
470 host_user_addr += size;
471 guest_phys_addr += size;
474 while (reg_size > 0) {
475 size = RTE_MIN(reg_size, page_size);
476 host_phys_addr = rte_mem_virt2phy((void *)(uintptr_t)
478 add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
480 host_user_addr += size;
481 guest_phys_addr += size;
486 #ifdef RTE_LIBRTE_VHOST_DEBUG
487 /* TODO: enable it only in debug mode? */
489 dump_guest_pages(struct virtio_net *dev)
492 struct guest_page *page;
494 for (i = 0; i < dev->nr_guest_pages; i++) {
495 page = &dev->guest_pages[i];
497 RTE_LOG(INFO, VHOST_CONFIG,
498 "guest physical page region %u\n"
499 "\t guest_phys_addr: %" PRIx64 "\n"
500 "\t host_phys_addr : %" PRIx64 "\n"
501 "\t size : %" PRIx64 "\n",
503 page->guest_phys_addr,
504 page->host_phys_addr,
509 #define dump_guest_pages(dev)
513 vhost_user_set_mem_table(struct virtio_net *dev, struct VhostUserMsg *pmsg)
515 struct VhostUserMemory memory = pmsg->payload.memory;
516 struct rte_vhost_mem_region *reg;
519 uint64_t mmap_offset;
525 free_mem_region(dev);
530 dev->nr_guest_pages = 0;
531 if (!dev->guest_pages) {
532 dev->max_guest_pages = 8;
533 dev->guest_pages = malloc(dev->max_guest_pages *
534 sizeof(struct guest_page));
535 if (dev->guest_pages == NULL) {
536 RTE_LOG(ERR, VHOST_CONFIG,
537 "(%d) failed to allocate memory "
538 "for dev->guest_pages\n",
544 dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
545 sizeof(struct rte_vhost_mem_region) * memory.nregions, 0);
546 if (dev->mem == NULL) {
547 RTE_LOG(ERR, VHOST_CONFIG,
548 "(%d) failed to allocate memory for dev->mem\n",
552 dev->mem->nregions = memory.nregions;
554 for (i = 0; i < memory.nregions; i++) {
556 reg = &dev->mem->regions[i];
558 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
559 reg->guest_user_addr = memory.regions[i].userspace_addr;
560 reg->size = memory.regions[i].memory_size;
563 mmap_offset = memory.regions[i].mmap_offset;
564 mmap_size = reg->size + mmap_offset;
566 /* mmap() without flag of MAP_ANONYMOUS, should be called
567 * with length argument aligned with hugepagesz at older
568 * longterm version Linux, like 2.6.32 and 3.2.72, or
569 * mmap() will fail with EINVAL.
571 * to avoid failure, make sure in caller to keep length
574 alignment = get_blk_size(fd);
575 if (alignment == (uint64_t)-1) {
576 RTE_LOG(ERR, VHOST_CONFIG,
577 "couldn't get hugepage size through fstat\n");
580 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
582 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
583 MAP_SHARED | MAP_POPULATE, fd, 0);
585 if (mmap_addr == MAP_FAILED) {
586 RTE_LOG(ERR, VHOST_CONFIG,
587 "mmap region %u failed.\n", i);
591 reg->mmap_addr = mmap_addr;
592 reg->mmap_size = mmap_size;
593 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
596 if (dev->dequeue_zero_copy)
597 add_guest_pages(dev, reg, alignment);
599 RTE_LOG(INFO, VHOST_CONFIG,
600 "guest memory region %u, size: 0x%" PRIx64 "\n"
601 "\t guest physical addr: 0x%" PRIx64 "\n"
602 "\t guest virtual addr: 0x%" PRIx64 "\n"
603 "\t host virtual addr: 0x%" PRIx64 "\n"
604 "\t mmap addr : 0x%" PRIx64 "\n"
605 "\t mmap size : 0x%" PRIx64 "\n"
606 "\t mmap align: 0x%" PRIx64 "\n"
607 "\t mmap off : 0x%" PRIx64 "\n",
609 reg->guest_phys_addr,
610 reg->guest_user_addr,
612 (uint64_t)(uintptr_t)mmap_addr,
618 dump_guest_pages(dev);
623 free_mem_region(dev);
630 vq_is_ready(struct vhost_virtqueue *vq)
632 return vq && vq->desc &&
633 vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
634 vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
638 virtio_is_ready(struct virtio_net *dev)
640 struct vhost_virtqueue *vq;
643 if (dev->nr_vring == 0)
646 for (i = 0; i < dev->nr_vring; i++) {
647 vq = dev->virtqueue[i];
649 if (!vq_is_ready(vq))
653 RTE_LOG(INFO, VHOST_CONFIG,
654 "virtio is now ready for processing.\n");
659 vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *pmsg)
661 struct vhost_vring_file file;
662 struct vhost_virtqueue *vq;
664 file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
665 if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
666 file.fd = VIRTIO_INVALID_EVENTFD;
668 file.fd = pmsg->fds[0];
669 RTE_LOG(INFO, VHOST_CONFIG,
670 "vring call idx:%d file:%d\n", file.index, file.fd);
672 vq = dev->virtqueue[file.index];
676 vq->callfd = file.fd;
680 vhost_user_set_vring_kick(struct virtio_net *dev, struct VhostUserMsg *pmsg)
682 struct vhost_vring_file file;
683 struct vhost_virtqueue *vq;
685 file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
686 if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
687 file.fd = VIRTIO_INVALID_EVENTFD;
689 file.fd = pmsg->fds[0];
690 RTE_LOG(INFO, VHOST_CONFIG,
691 "vring kick idx:%d file:%d\n", file.index, file.fd);
693 vq = dev->virtqueue[file.index];
696 vq->kickfd = file.fd;
700 free_zmbufs(struct vhost_virtqueue *vq)
702 struct zcopy_mbuf *zmbuf, *next;
704 for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
705 zmbuf != NULL; zmbuf = next) {
706 next = TAILQ_NEXT(zmbuf, next);
708 rte_pktmbuf_free(zmbuf->mbuf);
709 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
712 rte_free(vq->zmbufs);
716 * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
719 vhost_user_get_vring_base(struct virtio_net *dev,
722 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
724 /* We have to stop the queue (virtio) if it is running. */
725 if (dev->flags & VIRTIO_DEV_RUNNING) {
726 dev->flags &= ~VIRTIO_DEV_RUNNING;
727 dev->notify_ops->destroy_device(dev->vid);
730 dev->flags &= ~VIRTIO_DEV_READY;
732 /* Here we are safe to get the last used index */
733 msg->payload.state.num = vq->last_used_idx;
735 RTE_LOG(INFO, VHOST_CONFIG,
736 "vring base idx:%d file:%d\n", msg->payload.state.index,
737 msg->payload.state.num);
739 * Based on current qemu vhost-user implementation, this message is
740 * sent and only sent in vhost_vring_stop.
741 * TODO: cleanup the vring, it isn't usable since here.
746 vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
748 if (dev->dequeue_zero_copy)
750 rte_free(vq->shadow_used_ring);
751 vq->shadow_used_ring = NULL;
753 rte_free(vq->batch_copy_elems);
754 vq->batch_copy_elems = NULL;
760 * when virtio queues are ready to work, qemu will send us to
761 * enable the virtio queue pair.
764 vhost_user_set_vring_enable(struct virtio_net *dev,
767 int enable = (int)msg->payload.state.num;
769 RTE_LOG(INFO, VHOST_CONFIG,
770 "set queue enable: %d to qp idx: %d\n",
771 enable, msg->payload.state.index);
773 if (dev->notify_ops->vring_state_changed)
774 dev->notify_ops->vring_state_changed(dev->vid,
775 msg->payload.state.index, enable);
777 dev->virtqueue[msg->payload.state.index]->enabled = enable;
783 vhost_user_set_protocol_features(struct virtio_net *dev,
784 uint64_t protocol_features)
786 if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
789 dev->protocol_features = protocol_features;
793 vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
795 int fd = msg->fds[0];
800 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
804 if (msg->size != sizeof(VhostUserLog)) {
805 RTE_LOG(ERR, VHOST_CONFIG,
806 "invalid log base msg size: %"PRId32" != %d\n",
807 msg->size, (int)sizeof(VhostUserLog));
811 size = msg->payload.log.mmap_size;
812 off = msg->payload.log.mmap_offset;
813 RTE_LOG(INFO, VHOST_CONFIG,
814 "log mmap size: %"PRId64", offset: %"PRId64"\n",
818 * mmap from 0 to workaround a hugepage mmap bug: mmap will
819 * fail when offset is not page size aligned.
821 addr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
823 if (addr == MAP_FAILED) {
824 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
829 * Free previously mapped log memory on occasionally
830 * multiple VHOST_USER_SET_LOG_BASE.
833 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
835 dev->log_addr = (uint64_t)(uintptr_t)addr;
836 dev->log_base = dev->log_addr + off;
837 dev->log_size = size;
843 * An rarp packet is constructed and broadcasted to notify switches about
844 * the new location of the migrated VM, so that packets from outside will
845 * not be lost after migration.
847 * However, we don't actually "send" a rarp packet here, instead, we set
848 * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
851 vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
853 uint8_t *mac = (uint8_t *)&msg->payload.u64;
855 RTE_LOG(DEBUG, VHOST_CONFIG,
856 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
857 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
858 memcpy(dev->mac.addr_bytes, mac, 6);
861 * Set the flag to inject a RARP broadcast packet at
862 * rte_vhost_dequeue_burst().
864 * rte_smp_wmb() is for making sure the mac is copied
865 * before the flag is set.
868 rte_atomic16_set(&dev->broadcast_rarp, 1);
874 vhost_user_net_set_mtu(struct virtio_net *dev, struct VhostUserMsg *msg)
876 if (msg->payload.u64 < VIRTIO_MIN_MTU ||
877 msg->payload.u64 > VIRTIO_MAX_MTU) {
878 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
884 dev->mtu = msg->payload.u64;
889 /* return bytes# of read on success or negative val on failure. */
891 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
895 ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
896 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
900 if (msg && msg->size) {
901 if (msg->size > sizeof(msg->payload)) {
902 RTE_LOG(ERR, VHOST_CONFIG,
903 "invalid msg size: %d\n", msg->size);
906 ret = read(sockfd, &msg->payload, msg->size);
909 if (ret != (int)msg->size) {
910 RTE_LOG(ERR, VHOST_CONFIG,
911 "read control message failed\n");
920 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
927 msg->flags &= ~VHOST_USER_VERSION_MASK;
928 msg->flags &= ~VHOST_USER_NEED_REPLY;
929 msg->flags |= VHOST_USER_VERSION;
930 msg->flags |= VHOST_USER_REPLY_MASK;
932 ret = send_fd_message(sockfd, (char *)msg,
933 VHOST_USER_HDR_SIZE + msg->size, NULL, 0);
939 * Allocate a queue pair if it hasn't been allocated yet
942 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev, VhostUserMsg *msg)
946 switch (msg->request) {
947 case VHOST_USER_SET_VRING_KICK:
948 case VHOST_USER_SET_VRING_CALL:
949 case VHOST_USER_SET_VRING_ERR:
950 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
952 case VHOST_USER_SET_VRING_NUM:
953 case VHOST_USER_SET_VRING_BASE:
954 case VHOST_USER_SET_VRING_ENABLE:
955 vring_idx = msg->payload.state.index;
957 case VHOST_USER_SET_VRING_ADDR:
958 vring_idx = msg->payload.addr.index;
964 if (vring_idx >= VHOST_MAX_VRING) {
965 RTE_LOG(ERR, VHOST_CONFIG,
966 "invalid vring index: %u\n", vring_idx);
970 if (dev->virtqueue[vring_idx])
973 return alloc_vring_queue(dev, vring_idx);
977 vhost_user_msg_handler(int vid, int fd)
979 struct virtio_net *dev;
980 struct VhostUserMsg msg;
983 dev = get_device(vid);
987 if (!dev->notify_ops) {
988 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
989 if (!dev->notify_ops) {
990 RTE_LOG(ERR, VHOST_CONFIG,
991 "failed to get callback ops for driver %s\n",
997 ret = read_vhost_message(fd, &msg);
998 if (ret <= 0 || msg.request >= VHOST_USER_MAX) {
1000 RTE_LOG(ERR, VHOST_CONFIG,
1001 "vhost read message failed\n");
1003 RTE_LOG(INFO, VHOST_CONFIG,
1004 "vhost peer closed\n");
1006 RTE_LOG(ERR, VHOST_CONFIG,
1007 "vhost read incorrect message\n");
1013 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1014 vhost_message_str[msg.request]);
1016 ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1018 RTE_LOG(ERR, VHOST_CONFIG,
1019 "failed to alloc queue\n");
1023 switch (msg.request) {
1024 case VHOST_USER_GET_FEATURES:
1025 msg.payload.u64 = vhost_user_get_features(dev);
1026 msg.size = sizeof(msg.payload.u64);
1027 send_vhost_message(fd, &msg);
1029 case VHOST_USER_SET_FEATURES:
1030 vhost_user_set_features(dev, msg.payload.u64);
1033 case VHOST_USER_GET_PROTOCOL_FEATURES:
1034 msg.payload.u64 = VHOST_USER_PROTOCOL_FEATURES;
1035 msg.size = sizeof(msg.payload.u64);
1036 send_vhost_message(fd, &msg);
1038 case VHOST_USER_SET_PROTOCOL_FEATURES:
1039 vhost_user_set_protocol_features(dev, msg.payload.u64);
1042 case VHOST_USER_SET_OWNER:
1043 vhost_user_set_owner();
1045 case VHOST_USER_RESET_OWNER:
1046 vhost_user_reset_owner(dev);
1049 case VHOST_USER_SET_MEM_TABLE:
1050 ret = vhost_user_set_mem_table(dev, &msg);
1053 case VHOST_USER_SET_LOG_BASE:
1054 vhost_user_set_log_base(dev, &msg);
1056 /* it needs a reply */
1057 msg.size = sizeof(msg.payload.u64);
1058 send_vhost_message(fd, &msg);
1060 case VHOST_USER_SET_LOG_FD:
1062 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1065 case VHOST_USER_SET_VRING_NUM:
1066 vhost_user_set_vring_num(dev, &msg);
1068 case VHOST_USER_SET_VRING_ADDR:
1069 vhost_user_set_vring_addr(dev, &msg);
1071 case VHOST_USER_SET_VRING_BASE:
1072 vhost_user_set_vring_base(dev, &msg);
1075 case VHOST_USER_GET_VRING_BASE:
1076 vhost_user_get_vring_base(dev, &msg);
1077 msg.size = sizeof(msg.payload.state);
1078 send_vhost_message(fd, &msg);
1081 case VHOST_USER_SET_VRING_KICK:
1082 vhost_user_set_vring_kick(dev, &msg);
1084 case VHOST_USER_SET_VRING_CALL:
1085 vhost_user_set_vring_call(dev, &msg);
1088 case VHOST_USER_SET_VRING_ERR:
1089 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1091 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1094 case VHOST_USER_GET_QUEUE_NUM:
1095 msg.payload.u64 = VHOST_MAX_QUEUE_PAIRS;
1096 msg.size = sizeof(msg.payload.u64);
1097 send_vhost_message(fd, &msg);
1100 case VHOST_USER_SET_VRING_ENABLE:
1101 vhost_user_set_vring_enable(dev, &msg);
1103 case VHOST_USER_SEND_RARP:
1104 vhost_user_send_rarp(dev, &msg);
1107 case VHOST_USER_NET_SET_MTU:
1108 ret = vhost_user_net_set_mtu(dev, &msg);
1117 if (msg.flags & VHOST_USER_NEED_REPLY) {
1118 msg.payload.u64 = !!ret;
1119 msg.size = sizeof(msg.payload.u64);
1120 send_vhost_message(fd, &msg);
1123 if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1124 dev->flags |= VIRTIO_DEV_READY;
1126 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1127 if (dev->dequeue_zero_copy) {
1128 RTE_LOG(INFO, VHOST_CONFIG,
1129 "dequeue zero copy is enabled\n");
1132 if (dev->notify_ops->new_device(dev->vid) == 0)
1133 dev->flags |= VIRTIO_DEV_RUNNING;