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)
174 if ((dev->flags & VIRTIO_DEV_RUNNING) && dev->features != features) {
175 if (dev->notify_ops->features_changed)
176 dev->notify_ops->features_changed(dev->vid, features);
179 dev->features = features;
181 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
182 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
184 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
186 LOG_DEBUG(VHOST_CONFIG,
187 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
189 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
190 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
196 * The virtio device sends us the size of the descriptor ring.
199 vhost_user_set_vring_num(struct virtio_net *dev,
202 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
204 vq->size = msg->payload.state.num;
206 if (dev->dequeue_zero_copy) {
208 vq->last_zmbuf_idx = 0;
209 vq->zmbuf_size = vq->size;
210 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
211 sizeof(struct zcopy_mbuf), 0);
212 if (vq->zmbufs == NULL) {
213 RTE_LOG(WARNING, VHOST_CONFIG,
214 "failed to allocate mem for zero copy; "
215 "zero copy is force disabled\n");
216 dev->dequeue_zero_copy = 0;
220 vq->shadow_used_ring = rte_malloc(NULL,
221 vq->size * sizeof(struct vring_used_elem),
222 RTE_CACHE_LINE_SIZE);
223 if (!vq->shadow_used_ring) {
224 RTE_LOG(ERR, VHOST_CONFIG,
225 "failed to allocate memory for shadow used ring.\n");
233 * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
234 * same numa node as the memory of vring descriptor.
236 #ifdef RTE_LIBRTE_VHOST_NUMA
237 static struct virtio_net*
238 numa_realloc(struct virtio_net *dev, int index)
240 int oldnode, newnode;
241 struct virtio_net *old_dev;
242 struct vhost_virtqueue *old_vq, *vq;
245 enum {VIRTIO_RXQ, VIRTIO_TXQ, VIRTIO_QNUM};
248 vq = old_vq = dev->virtqueue[index];
250 ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
251 MPOL_F_NODE | MPOL_F_ADDR);
253 /* check if we need to reallocate vq */
254 ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
255 MPOL_F_NODE | MPOL_F_ADDR);
257 RTE_LOG(ERR, VHOST_CONFIG,
258 "Unable to get vq numa information.\n");
261 if (oldnode != newnode) {
262 RTE_LOG(INFO, VHOST_CONFIG,
263 "reallocate vq from %d to %d node\n", oldnode, newnode);
264 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
268 memcpy(vq, old_vq, sizeof(*vq) * VIRTIO_QNUM);
272 /* check if we need to reallocate dev */
273 ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
274 MPOL_F_NODE | MPOL_F_ADDR);
276 RTE_LOG(ERR, VHOST_CONFIG,
277 "Unable to get dev numa information.\n");
280 if (oldnode != newnode) {
281 RTE_LOG(INFO, VHOST_CONFIG,
282 "reallocate dev from %d to %d node\n",
284 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
290 memcpy(dev, old_dev, sizeof(*dev));
295 dev->virtqueue[index] = vq;
296 vhost_devices[dev->vid] = dev;
301 static struct virtio_net*
302 numa_realloc(struct virtio_net *dev, int index __rte_unused)
309 * Converts QEMU virtual address to Vhost virtual address. This function is
310 * used to convert the ring addresses to our address space.
313 qva_to_vva(struct virtio_net *dev, uint64_t qva)
315 struct rte_vhost_mem_region *reg;
318 /* Find the region where the address lives. */
319 for (i = 0; i < dev->mem->nregions; i++) {
320 reg = &dev->mem->regions[i];
322 if (qva >= reg->guest_user_addr &&
323 qva < reg->guest_user_addr + reg->size) {
324 return qva - reg->guest_user_addr +
333 * The virtio device sends us the desc, used and avail ring addresses.
334 * This function then converts these to our address space.
337 vhost_user_set_vring_addr(struct virtio_net *dev, VhostUserMsg *msg)
339 struct vhost_virtqueue *vq;
341 if (dev->mem == NULL)
344 /* addr->index refers to the queue index. The txq 1, rxq is 0. */
345 vq = dev->virtqueue[msg->payload.addr.index];
347 /* The addresses are converted from QEMU virtual to Vhost virtual. */
348 vq->desc = (struct vring_desc *)(uintptr_t)qva_to_vva(dev,
349 msg->payload.addr.desc_user_addr);
351 RTE_LOG(ERR, VHOST_CONFIG,
352 "(%d) failed to find desc ring address.\n",
357 dev = numa_realloc(dev, msg->payload.addr.index);
358 vq = dev->virtqueue[msg->payload.addr.index];
360 vq->avail = (struct vring_avail *)(uintptr_t)qva_to_vva(dev,
361 msg->payload.addr.avail_user_addr);
362 if (vq->avail == 0) {
363 RTE_LOG(ERR, VHOST_CONFIG,
364 "(%d) failed to find avail ring address.\n",
369 vq->used = (struct vring_used *)(uintptr_t)qva_to_vva(dev,
370 msg->payload.addr.used_user_addr);
372 RTE_LOG(ERR, VHOST_CONFIG,
373 "(%d) failed to find used ring address.\n",
378 if (vq->last_used_idx != vq->used->idx) {
379 RTE_LOG(WARNING, VHOST_CONFIG,
380 "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
381 "some packets maybe resent for Tx and dropped for Rx\n",
382 vq->last_used_idx, vq->used->idx);
383 vq->last_used_idx = vq->used->idx;
384 vq->last_avail_idx = vq->used->idx;
387 vq->log_guest_addr = msg->payload.addr.log_guest_addr;
389 LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
391 LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
392 dev->vid, vq->avail);
393 LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
395 LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
396 dev->vid, vq->log_guest_addr);
402 * The virtio device sends us the available ring last used index.
405 vhost_user_set_vring_base(struct virtio_net *dev,
408 dev->virtqueue[msg->payload.state.index]->last_used_idx =
409 msg->payload.state.num;
410 dev->virtqueue[msg->payload.state.index]->last_avail_idx =
411 msg->payload.state.num;
417 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
418 uint64_t host_phys_addr, uint64_t size)
420 struct guest_page *page, *last_page;
422 if (dev->nr_guest_pages == dev->max_guest_pages) {
423 dev->max_guest_pages *= 2;
424 dev->guest_pages = realloc(dev->guest_pages,
425 dev->max_guest_pages * sizeof(*page));
428 if (dev->nr_guest_pages > 0) {
429 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
430 /* merge if the two pages are continuous */
431 if (host_phys_addr == last_page->host_phys_addr +
433 last_page->size += size;
438 page = &dev->guest_pages[dev->nr_guest_pages++];
439 page->guest_phys_addr = guest_phys_addr;
440 page->host_phys_addr = host_phys_addr;
445 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
448 uint64_t reg_size = reg->size;
449 uint64_t host_user_addr = reg->host_user_addr;
450 uint64_t guest_phys_addr = reg->guest_phys_addr;
451 uint64_t host_phys_addr;
454 host_phys_addr = rte_mem_virt2phy((void *)(uintptr_t)host_user_addr);
455 size = page_size - (guest_phys_addr & (page_size - 1));
456 size = RTE_MIN(size, reg_size);
458 add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
459 host_user_addr += size;
460 guest_phys_addr += size;
463 while (reg_size > 0) {
464 size = RTE_MIN(reg_size, page_size);
465 host_phys_addr = rte_mem_virt2phy((void *)(uintptr_t)
467 add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
469 host_user_addr += size;
470 guest_phys_addr += size;
475 #ifdef RTE_LIBRTE_VHOST_DEBUG
476 /* TODO: enable it only in debug mode? */
478 dump_guest_pages(struct virtio_net *dev)
481 struct guest_page *page;
483 for (i = 0; i < dev->nr_guest_pages; i++) {
484 page = &dev->guest_pages[i];
486 RTE_LOG(INFO, VHOST_CONFIG,
487 "guest physical page region %u\n"
488 "\t guest_phys_addr: %" PRIx64 "\n"
489 "\t host_phys_addr : %" PRIx64 "\n"
490 "\t size : %" PRIx64 "\n",
492 page->guest_phys_addr,
493 page->host_phys_addr,
498 #define dump_guest_pages(dev)
502 vhost_user_set_mem_table(struct virtio_net *dev, struct VhostUserMsg *pmsg)
504 struct VhostUserMemory memory = pmsg->payload.memory;
505 struct rte_vhost_mem_region *reg;
508 uint64_t mmap_offset;
514 free_mem_region(dev);
519 dev->nr_guest_pages = 0;
520 if (!dev->guest_pages) {
521 dev->max_guest_pages = 8;
522 dev->guest_pages = malloc(dev->max_guest_pages *
523 sizeof(struct guest_page));
524 if (dev->guest_pages == NULL) {
525 RTE_LOG(ERR, VHOST_CONFIG,
526 "(%d) failed to allocate memory "
527 "for dev->guest_pages\n",
533 dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
534 sizeof(struct rte_vhost_mem_region) * memory.nregions, 0);
535 if (dev->mem == NULL) {
536 RTE_LOG(ERR, VHOST_CONFIG,
537 "(%d) failed to allocate memory for dev->mem\n",
541 dev->mem->nregions = memory.nregions;
543 for (i = 0; i < memory.nregions; i++) {
545 reg = &dev->mem->regions[i];
547 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
548 reg->guest_user_addr = memory.regions[i].userspace_addr;
549 reg->size = memory.regions[i].memory_size;
552 mmap_offset = memory.regions[i].mmap_offset;
553 mmap_size = reg->size + mmap_offset;
555 /* mmap() without flag of MAP_ANONYMOUS, should be called
556 * with length argument aligned with hugepagesz at older
557 * longterm version Linux, like 2.6.32 and 3.2.72, or
558 * mmap() will fail with EINVAL.
560 * to avoid failure, make sure in caller to keep length
563 alignment = get_blk_size(fd);
564 if (alignment == (uint64_t)-1) {
565 RTE_LOG(ERR, VHOST_CONFIG,
566 "couldn't get hugepage size through fstat\n");
569 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
571 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
572 MAP_SHARED | MAP_POPULATE, fd, 0);
574 if (mmap_addr == MAP_FAILED) {
575 RTE_LOG(ERR, VHOST_CONFIG,
576 "mmap region %u failed.\n", i);
580 reg->mmap_addr = mmap_addr;
581 reg->mmap_size = mmap_size;
582 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
585 if (dev->dequeue_zero_copy)
586 add_guest_pages(dev, reg, alignment);
588 RTE_LOG(INFO, VHOST_CONFIG,
589 "guest memory region %u, size: 0x%" PRIx64 "\n"
590 "\t guest physical addr: 0x%" PRIx64 "\n"
591 "\t guest virtual addr: 0x%" PRIx64 "\n"
592 "\t host virtual addr: 0x%" PRIx64 "\n"
593 "\t mmap addr : 0x%" PRIx64 "\n"
594 "\t mmap size : 0x%" PRIx64 "\n"
595 "\t mmap align: 0x%" PRIx64 "\n"
596 "\t mmap off : 0x%" PRIx64 "\n",
598 reg->guest_phys_addr,
599 reg->guest_user_addr,
601 (uint64_t)(uintptr_t)mmap_addr,
607 dump_guest_pages(dev);
612 free_mem_region(dev);
619 vq_is_ready(struct vhost_virtqueue *vq)
621 return vq && vq->desc &&
622 vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
623 vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
627 virtio_is_ready(struct virtio_net *dev)
629 struct vhost_virtqueue *vq;
632 if (dev->nr_vring == 0)
635 for (i = 0; i < dev->nr_vring; i++) {
636 vq = dev->virtqueue[i];
638 if (!vq_is_ready(vq))
642 RTE_LOG(INFO, VHOST_CONFIG,
643 "virtio is now ready for processing.\n");
648 vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *pmsg)
650 struct vhost_vring_file file;
651 struct vhost_virtqueue *vq;
653 file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
654 if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
655 file.fd = VIRTIO_INVALID_EVENTFD;
657 file.fd = pmsg->fds[0];
658 RTE_LOG(INFO, VHOST_CONFIG,
659 "vring call idx:%d file:%d\n", file.index, file.fd);
661 vq = dev->virtqueue[file.index];
665 vq->callfd = file.fd;
669 vhost_user_set_vring_kick(struct virtio_net *dev, struct VhostUserMsg *pmsg)
671 struct vhost_vring_file file;
672 struct vhost_virtqueue *vq;
674 file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
675 if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
676 file.fd = VIRTIO_INVALID_EVENTFD;
678 file.fd = pmsg->fds[0];
679 RTE_LOG(INFO, VHOST_CONFIG,
680 "vring kick idx:%d file:%d\n", file.index, file.fd);
682 vq = dev->virtqueue[file.index];
685 vq->kickfd = file.fd;
689 free_zmbufs(struct vhost_virtqueue *vq)
691 struct zcopy_mbuf *zmbuf, *next;
693 for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
694 zmbuf != NULL; zmbuf = next) {
695 next = TAILQ_NEXT(zmbuf, next);
697 rte_pktmbuf_free(zmbuf->mbuf);
698 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
701 rte_free(vq->zmbufs);
705 * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
708 vhost_user_get_vring_base(struct virtio_net *dev,
711 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
713 /* We have to stop the queue (virtio) if it is running. */
714 if (dev->flags & VIRTIO_DEV_RUNNING) {
715 dev->flags &= ~VIRTIO_DEV_RUNNING;
716 dev->notify_ops->destroy_device(dev->vid);
719 dev->flags &= ~VIRTIO_DEV_READY;
721 /* Here we are safe to get the last used index */
722 msg->payload.state.num = vq->last_used_idx;
724 RTE_LOG(INFO, VHOST_CONFIG,
725 "vring base idx:%d file:%d\n", msg->payload.state.index,
726 msg->payload.state.num);
728 * Based on current qemu vhost-user implementation, this message is
729 * sent and only sent in vhost_vring_stop.
730 * TODO: cleanup the vring, it isn't usable since here.
735 vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
737 if (dev->dequeue_zero_copy)
739 rte_free(vq->shadow_used_ring);
740 vq->shadow_used_ring = NULL;
746 * when virtio queues are ready to work, qemu will send us to
747 * enable the virtio queue pair.
750 vhost_user_set_vring_enable(struct virtio_net *dev,
753 int enable = (int)msg->payload.state.num;
755 RTE_LOG(INFO, VHOST_CONFIG,
756 "set queue enable: %d to qp idx: %d\n",
757 enable, msg->payload.state.index);
759 if (dev->notify_ops->vring_state_changed)
760 dev->notify_ops->vring_state_changed(dev->vid,
761 msg->payload.state.index, enable);
763 dev->virtqueue[msg->payload.state.index]->enabled = enable;
769 vhost_user_set_protocol_features(struct virtio_net *dev,
770 uint64_t protocol_features)
772 if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
775 dev->protocol_features = protocol_features;
779 vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
781 int fd = msg->fds[0];
786 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
790 if (msg->size != sizeof(VhostUserLog)) {
791 RTE_LOG(ERR, VHOST_CONFIG,
792 "invalid log base msg size: %"PRId32" != %d\n",
793 msg->size, (int)sizeof(VhostUserLog));
797 size = msg->payload.log.mmap_size;
798 off = msg->payload.log.mmap_offset;
799 RTE_LOG(INFO, VHOST_CONFIG,
800 "log mmap size: %"PRId64", offset: %"PRId64"\n",
804 * mmap from 0 to workaround a hugepage mmap bug: mmap will
805 * fail when offset is not page size aligned.
807 addr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
809 if (addr == MAP_FAILED) {
810 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
815 * Free previously mapped log memory on occasionally
816 * multiple VHOST_USER_SET_LOG_BASE.
819 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
821 dev->log_addr = (uint64_t)(uintptr_t)addr;
822 dev->log_base = dev->log_addr + off;
823 dev->log_size = size;
829 * An rarp packet is constructed and broadcasted to notify switches about
830 * the new location of the migrated VM, so that packets from outside will
831 * not be lost after migration.
833 * However, we don't actually "send" a rarp packet here, instead, we set
834 * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
837 vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
839 uint8_t *mac = (uint8_t *)&msg->payload.u64;
841 RTE_LOG(DEBUG, VHOST_CONFIG,
842 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
843 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
844 memcpy(dev->mac.addr_bytes, mac, 6);
847 * Set the flag to inject a RARP broadcast packet at
848 * rte_vhost_dequeue_burst().
850 * rte_smp_wmb() is for making sure the mac is copied
851 * before the flag is set.
854 rte_atomic16_set(&dev->broadcast_rarp, 1);
860 vhost_user_net_set_mtu(struct virtio_net *dev, struct VhostUserMsg *msg)
862 if (msg->payload.u64 < VIRTIO_MIN_MTU ||
863 msg->payload.u64 > VIRTIO_MAX_MTU) {
864 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
870 dev->mtu = msg->payload.u64;
875 /* return bytes# of read on success or negative val on failure. */
877 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
881 ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
882 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
886 if (msg && msg->size) {
887 if (msg->size > sizeof(msg->payload)) {
888 RTE_LOG(ERR, VHOST_CONFIG,
889 "invalid msg size: %d\n", msg->size);
892 ret = read(sockfd, &msg->payload, msg->size);
895 if (ret != (int)msg->size) {
896 RTE_LOG(ERR, VHOST_CONFIG,
897 "read control message failed\n");
906 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
913 msg->flags &= ~VHOST_USER_VERSION_MASK;
914 msg->flags &= ~VHOST_USER_NEED_REPLY;
915 msg->flags |= VHOST_USER_VERSION;
916 msg->flags |= VHOST_USER_REPLY_MASK;
918 ret = send_fd_message(sockfd, (char *)msg,
919 VHOST_USER_HDR_SIZE + msg->size, NULL, 0);
925 * Allocate a queue pair if it hasn't been allocated yet
928 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev, VhostUserMsg *msg)
932 switch (msg->request) {
933 case VHOST_USER_SET_VRING_KICK:
934 case VHOST_USER_SET_VRING_CALL:
935 case VHOST_USER_SET_VRING_ERR:
936 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
938 case VHOST_USER_SET_VRING_NUM:
939 case VHOST_USER_SET_VRING_BASE:
940 case VHOST_USER_SET_VRING_ENABLE:
941 vring_idx = msg->payload.state.index;
943 case VHOST_USER_SET_VRING_ADDR:
944 vring_idx = msg->payload.addr.index;
950 if (vring_idx >= VHOST_MAX_VRING) {
951 RTE_LOG(ERR, VHOST_CONFIG,
952 "invalid vring index: %u\n", vring_idx);
956 if (dev->virtqueue[vring_idx])
959 return alloc_vring_queue(dev, vring_idx);
963 vhost_user_msg_handler(int vid, int fd)
965 struct virtio_net *dev;
966 struct VhostUserMsg msg;
969 dev = get_device(vid);
973 if (!dev->notify_ops) {
974 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
975 if (!dev->notify_ops) {
976 RTE_LOG(ERR, VHOST_CONFIG,
977 "failed to get callback ops for driver %s\n",
983 ret = read_vhost_message(fd, &msg);
984 if (ret <= 0 || msg.request >= VHOST_USER_MAX) {
986 RTE_LOG(ERR, VHOST_CONFIG,
987 "vhost read message failed\n");
989 RTE_LOG(INFO, VHOST_CONFIG,
990 "vhost peer closed\n");
992 RTE_LOG(ERR, VHOST_CONFIG,
993 "vhost read incorrect message\n");
999 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1000 vhost_message_str[msg.request]);
1002 ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1004 RTE_LOG(ERR, VHOST_CONFIG,
1005 "failed to alloc queue\n");
1009 switch (msg.request) {
1010 case VHOST_USER_GET_FEATURES:
1011 msg.payload.u64 = vhost_user_get_features(dev);
1012 msg.size = sizeof(msg.payload.u64);
1013 send_vhost_message(fd, &msg);
1015 case VHOST_USER_SET_FEATURES:
1016 vhost_user_set_features(dev, msg.payload.u64);
1019 case VHOST_USER_GET_PROTOCOL_FEATURES:
1020 msg.payload.u64 = VHOST_USER_PROTOCOL_FEATURES;
1021 msg.size = sizeof(msg.payload.u64);
1022 send_vhost_message(fd, &msg);
1024 case VHOST_USER_SET_PROTOCOL_FEATURES:
1025 vhost_user_set_protocol_features(dev, msg.payload.u64);
1028 case VHOST_USER_SET_OWNER:
1029 vhost_user_set_owner();
1031 case VHOST_USER_RESET_OWNER:
1032 vhost_user_reset_owner(dev);
1035 case VHOST_USER_SET_MEM_TABLE:
1036 ret = vhost_user_set_mem_table(dev, &msg);
1039 case VHOST_USER_SET_LOG_BASE:
1040 vhost_user_set_log_base(dev, &msg);
1042 /* it needs a reply */
1043 msg.size = sizeof(msg.payload.u64);
1044 send_vhost_message(fd, &msg);
1046 case VHOST_USER_SET_LOG_FD:
1048 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1051 case VHOST_USER_SET_VRING_NUM:
1052 vhost_user_set_vring_num(dev, &msg);
1054 case VHOST_USER_SET_VRING_ADDR:
1055 vhost_user_set_vring_addr(dev, &msg);
1057 case VHOST_USER_SET_VRING_BASE:
1058 vhost_user_set_vring_base(dev, &msg);
1061 case VHOST_USER_GET_VRING_BASE:
1062 vhost_user_get_vring_base(dev, &msg);
1063 msg.size = sizeof(msg.payload.state);
1064 send_vhost_message(fd, &msg);
1067 case VHOST_USER_SET_VRING_KICK:
1068 vhost_user_set_vring_kick(dev, &msg);
1070 case VHOST_USER_SET_VRING_CALL:
1071 vhost_user_set_vring_call(dev, &msg);
1074 case VHOST_USER_SET_VRING_ERR:
1075 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1077 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1080 case VHOST_USER_GET_QUEUE_NUM:
1081 msg.payload.u64 = VHOST_MAX_QUEUE_PAIRS;
1082 msg.size = sizeof(msg.payload.u64);
1083 send_vhost_message(fd, &msg);
1086 case VHOST_USER_SET_VRING_ENABLE:
1087 vhost_user_set_vring_enable(dev, &msg);
1089 case VHOST_USER_SEND_RARP:
1090 vhost_user_send_rarp(dev, &msg);
1093 case VHOST_USER_NET_SET_MTU:
1094 ret = vhost_user_net_set_mtu(dev, &msg);
1103 if (msg.flags & VHOST_USER_NEED_REPLY) {
1104 msg.payload.u64 = !!ret;
1105 msg.size = sizeof(msg.payload.u64);
1106 send_vhost_message(fd, &msg);
1109 if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1110 dev->flags |= VIRTIO_DEV_READY;
1112 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1113 if (dev->dequeue_zero_copy) {
1114 RTE_LOG(INFO, VHOST_CONFIG,
1115 "dequeue zero copy is enabled\n");
1118 if (dev->notify_ops->new_device(dev->vid) == 0)
1119 dev->flags |= VIRTIO_DEV_RUNNING;