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 static const char *vhost_message_str[VHOST_USER_MAX] = {
55 [VHOST_USER_NONE] = "VHOST_USER_NONE",
56 [VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
57 [VHOST_USER_SET_FEATURES] = "VHOST_USER_SET_FEATURES",
58 [VHOST_USER_SET_OWNER] = "VHOST_USER_SET_OWNER",
59 [VHOST_USER_RESET_OWNER] = "VHOST_USER_RESET_OWNER",
60 [VHOST_USER_SET_MEM_TABLE] = "VHOST_USER_SET_MEM_TABLE",
61 [VHOST_USER_SET_LOG_BASE] = "VHOST_USER_SET_LOG_BASE",
62 [VHOST_USER_SET_LOG_FD] = "VHOST_USER_SET_LOG_FD",
63 [VHOST_USER_SET_VRING_NUM] = "VHOST_USER_SET_VRING_NUM",
64 [VHOST_USER_SET_VRING_ADDR] = "VHOST_USER_SET_VRING_ADDR",
65 [VHOST_USER_SET_VRING_BASE] = "VHOST_USER_SET_VRING_BASE",
66 [VHOST_USER_GET_VRING_BASE] = "VHOST_USER_GET_VRING_BASE",
67 [VHOST_USER_SET_VRING_KICK] = "VHOST_USER_SET_VRING_KICK",
68 [VHOST_USER_SET_VRING_CALL] = "VHOST_USER_SET_VRING_CALL",
69 [VHOST_USER_SET_VRING_ERR] = "VHOST_USER_SET_VRING_ERR",
70 [VHOST_USER_GET_PROTOCOL_FEATURES] = "VHOST_USER_GET_PROTOCOL_FEATURES",
71 [VHOST_USER_SET_PROTOCOL_FEATURES] = "VHOST_USER_SET_PROTOCOL_FEATURES",
72 [VHOST_USER_GET_QUEUE_NUM] = "VHOST_USER_GET_QUEUE_NUM",
73 [VHOST_USER_SET_VRING_ENABLE] = "VHOST_USER_SET_VRING_ENABLE",
74 [VHOST_USER_SEND_RARP] = "VHOST_USER_SEND_RARP",
83 ret = fstat(fd, &stat);
84 return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
88 free_mem_region(struct virtio_net *dev)
91 struct virtio_memory_region *reg;
93 if (!dev || !dev->mem)
96 for (i = 0; i < dev->mem->nregions; i++) {
97 reg = &dev->mem->regions[i];
98 if (reg->host_user_addr) {
99 munmap(reg->mmap_addr, reg->mmap_size);
106 vhost_backend_cleanup(struct virtio_net *dev)
109 free_mem_region(dev);
114 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
120 * This function just returns success at the moment unless
121 * the device hasn't been initialised.
124 vhost_user_set_owner(void)
130 vhost_user_reset_owner(struct virtio_net *dev)
132 if (dev->flags & VIRTIO_DEV_RUNNING) {
133 dev->flags &= ~VIRTIO_DEV_RUNNING;
134 notify_ops->destroy_device(dev->vid);
137 cleanup_device(dev, 0);
143 * The features that we support are requested.
146 vhost_user_get_features(void)
148 return VHOST_FEATURES;
152 * We receive the negotiated features supported by us and the virtio device.
155 vhost_user_set_features(struct virtio_net *dev, uint64_t features)
157 if (features & ~VHOST_FEATURES)
160 dev->features = features;
162 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
163 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
165 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
167 LOG_DEBUG(VHOST_CONFIG,
168 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
170 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
171 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
177 * The virtio device sends us the size of the descriptor ring.
180 vhost_user_set_vring_num(struct virtio_net *dev,
181 struct vhost_vring_state *state)
183 struct vhost_virtqueue *vq = dev->virtqueue[state->index];
185 vq->size = state->num;
187 if (dev->dequeue_zero_copy) {
189 vq->last_zmbuf_idx = 0;
190 vq->zmbuf_size = vq->size;
191 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
192 sizeof(struct zcopy_mbuf), 0);
193 if (vq->zmbufs == NULL) {
194 RTE_LOG(WARNING, VHOST_CONFIG,
195 "failed to allocate mem for zero copy; "
196 "zero copy is force disabled\n");
197 dev->dequeue_zero_copy = 0;
201 vq->shadow_used_ring = rte_malloc(NULL,
202 vq->size * sizeof(struct vring_used_elem),
203 RTE_CACHE_LINE_SIZE);
204 if (!vq->shadow_used_ring) {
205 RTE_LOG(ERR, VHOST_CONFIG,
206 "failed to allocate memory for shadow used ring.\n");
214 * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
215 * same numa node as the memory of vring descriptor.
217 #ifdef RTE_LIBRTE_VHOST_NUMA
218 static struct virtio_net*
219 numa_realloc(struct virtio_net *dev, int index)
221 int oldnode, newnode;
222 struct virtio_net *old_dev;
223 struct vhost_virtqueue *old_vq, *vq;
227 * vq is allocated on pairs, we should try to do realloc
228 * on first queue of one queue pair only.
230 if (index % VIRTIO_QNUM != 0)
234 vq = old_vq = dev->virtqueue[index];
236 ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
237 MPOL_F_NODE | MPOL_F_ADDR);
239 /* check if we need to reallocate vq */
240 ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
241 MPOL_F_NODE | MPOL_F_ADDR);
243 RTE_LOG(ERR, VHOST_CONFIG,
244 "Unable to get vq numa information.\n");
247 if (oldnode != newnode) {
248 RTE_LOG(INFO, VHOST_CONFIG,
249 "reallocate vq from %d to %d node\n", oldnode, newnode);
250 vq = rte_malloc_socket(NULL, sizeof(*vq) * VIRTIO_QNUM, 0,
255 memcpy(vq, old_vq, sizeof(*vq) * VIRTIO_QNUM);
259 /* check if we need to reallocate dev */
260 ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
261 MPOL_F_NODE | MPOL_F_ADDR);
263 RTE_LOG(ERR, VHOST_CONFIG,
264 "Unable to get dev numa information.\n");
267 if (oldnode != newnode) {
268 RTE_LOG(INFO, VHOST_CONFIG,
269 "reallocate dev from %d to %d node\n",
271 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
277 memcpy(dev, old_dev, sizeof(*dev));
282 dev->virtqueue[index] = vq;
283 dev->virtqueue[index + 1] = vq + 1;
284 vhost_devices[dev->vid] = dev;
289 static struct virtio_net*
290 numa_realloc(struct virtio_net *dev, int index __rte_unused)
297 * Converts QEMU virtual address to Vhost virtual address. This function is
298 * used to convert the ring addresses to our address space.
301 qva_to_vva(struct virtio_net *dev, uint64_t qva)
303 struct virtio_memory_region *reg;
306 /* Find the region where the address lives. */
307 for (i = 0; i < dev->mem->nregions; i++) {
308 reg = &dev->mem->regions[i];
310 if (qva >= reg->guest_user_addr &&
311 qva < reg->guest_user_addr + reg->size) {
312 return qva - reg->guest_user_addr +
321 * The virtio device sends us the desc, used and avail ring addresses.
322 * This function then converts these to our address space.
325 vhost_user_set_vring_addr(struct virtio_net *dev, struct vhost_vring_addr *addr)
327 struct vhost_virtqueue *vq;
329 if (dev->mem == NULL)
332 /* addr->index refers to the queue index. The txq 1, rxq is 0. */
333 vq = dev->virtqueue[addr->index];
335 /* The addresses are converted from QEMU virtual to Vhost virtual. */
336 vq->desc = (struct vring_desc *)(uintptr_t)qva_to_vva(dev,
337 addr->desc_user_addr);
339 RTE_LOG(ERR, VHOST_CONFIG,
340 "(%d) failed to find desc ring address.\n",
345 dev = numa_realloc(dev, addr->index);
346 vq = dev->virtqueue[addr->index];
348 vq->avail = (struct vring_avail *)(uintptr_t)qva_to_vva(dev,
349 addr->avail_user_addr);
350 if (vq->avail == 0) {
351 RTE_LOG(ERR, VHOST_CONFIG,
352 "(%d) failed to find avail ring address.\n",
357 vq->used = (struct vring_used *)(uintptr_t)qva_to_vva(dev,
358 addr->used_user_addr);
360 RTE_LOG(ERR, VHOST_CONFIG,
361 "(%d) failed to find used ring address.\n",
366 if (vq->last_used_idx != vq->used->idx) {
367 RTE_LOG(WARNING, VHOST_CONFIG,
368 "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
369 "some packets maybe resent for Tx and dropped for Rx\n",
370 vq->last_used_idx, vq->used->idx);
371 vq->last_used_idx = vq->used->idx;
372 vq->last_avail_idx = vq->used->idx;
375 vq->log_guest_addr = addr->log_guest_addr;
377 LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
379 LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
380 dev->vid, vq->avail);
381 LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
383 LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
384 dev->vid, vq->log_guest_addr);
390 * The virtio device sends us the available ring last used index.
393 vhost_user_set_vring_base(struct virtio_net *dev,
394 struct vhost_vring_state *state)
396 dev->virtqueue[state->index]->last_used_idx = state->num;
397 dev->virtqueue[state->index]->last_avail_idx = state->num;
403 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
404 uint64_t host_phys_addr, uint64_t size)
406 struct guest_page *page, *last_page;
408 if (dev->nr_guest_pages == dev->max_guest_pages) {
409 dev->max_guest_pages *= 2;
410 dev->guest_pages = realloc(dev->guest_pages,
411 dev->max_guest_pages * sizeof(*page));
414 if (dev->nr_guest_pages > 0) {
415 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
416 /* merge if the two pages are continuous */
417 if (host_phys_addr == last_page->host_phys_addr +
419 last_page->size += size;
424 page = &dev->guest_pages[dev->nr_guest_pages++];
425 page->guest_phys_addr = guest_phys_addr;
426 page->host_phys_addr = host_phys_addr;
431 add_guest_pages(struct virtio_net *dev, struct virtio_memory_region *reg,
434 uint64_t reg_size = reg->size;
435 uint64_t host_user_addr = reg->host_user_addr;
436 uint64_t guest_phys_addr = reg->guest_phys_addr;
437 uint64_t host_phys_addr;
440 host_phys_addr = rte_mem_virt2phy((void *)(uintptr_t)host_user_addr);
441 size = page_size - (guest_phys_addr & (page_size - 1));
442 size = RTE_MIN(size, reg_size);
444 add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
445 host_user_addr += size;
446 guest_phys_addr += size;
449 while (reg_size > 0) {
450 size = RTE_MIN(reg_size, page_size);
451 host_phys_addr = rte_mem_virt2phy((void *)(uintptr_t)
453 add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
455 host_user_addr += size;
456 guest_phys_addr += size;
461 #ifdef RTE_LIBRTE_VHOST_DEBUG
462 /* TODO: enable it only in debug mode? */
464 dump_guest_pages(struct virtio_net *dev)
467 struct guest_page *page;
469 for (i = 0; i < dev->nr_guest_pages; i++) {
470 page = &dev->guest_pages[i];
472 RTE_LOG(INFO, VHOST_CONFIG,
473 "guest physical page region %u\n"
474 "\t guest_phys_addr: %" PRIx64 "\n"
475 "\t host_phys_addr : %" PRIx64 "\n"
476 "\t size : %" PRIx64 "\n",
478 page->guest_phys_addr,
479 page->host_phys_addr,
484 #define dump_guest_pages(dev)
488 vhost_user_set_mem_table(struct virtio_net *dev, struct VhostUserMsg *pmsg)
490 struct VhostUserMemory memory = pmsg->payload.memory;
491 struct virtio_memory_region *reg;
494 uint64_t mmap_offset;
499 /* Remove from the data plane. */
500 if (dev->flags & VIRTIO_DEV_RUNNING) {
501 dev->flags &= ~VIRTIO_DEV_RUNNING;
502 notify_ops->destroy_device(dev->vid);
506 free_mem_region(dev);
511 dev->nr_guest_pages = 0;
512 if (!dev->guest_pages) {
513 dev->max_guest_pages = 8;
514 dev->guest_pages = malloc(dev->max_guest_pages *
515 sizeof(struct guest_page));
518 dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct virtio_memory) +
519 sizeof(struct virtio_memory_region) * memory.nregions, 0);
520 if (dev->mem == NULL) {
521 RTE_LOG(ERR, VHOST_CONFIG,
522 "(%d) failed to allocate memory for dev->mem\n",
526 dev->mem->nregions = memory.nregions;
528 for (i = 0; i < memory.nregions; i++) {
530 reg = &dev->mem->regions[i];
532 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
533 reg->guest_user_addr = memory.regions[i].userspace_addr;
534 reg->size = memory.regions[i].memory_size;
537 mmap_offset = memory.regions[i].mmap_offset;
538 mmap_size = reg->size + mmap_offset;
540 /* mmap() without flag of MAP_ANONYMOUS, should be called
541 * with length argument aligned with hugepagesz at older
542 * longterm version Linux, like 2.6.32 and 3.2.72, or
543 * mmap() will fail with EINVAL.
545 * to avoid failure, make sure in caller to keep length
548 alignment = get_blk_size(fd);
549 if (alignment == (uint64_t)-1) {
550 RTE_LOG(ERR, VHOST_CONFIG,
551 "couldn't get hugepage size through fstat\n");
554 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
556 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
557 MAP_SHARED | MAP_POPULATE, fd, 0);
559 if (mmap_addr == MAP_FAILED) {
560 RTE_LOG(ERR, VHOST_CONFIG,
561 "mmap region %u failed.\n", i);
565 reg->mmap_addr = mmap_addr;
566 reg->mmap_size = mmap_size;
567 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
570 if (dev->dequeue_zero_copy)
571 add_guest_pages(dev, reg, alignment);
573 RTE_LOG(INFO, VHOST_CONFIG,
574 "guest memory region %u, size: 0x%" PRIx64 "\n"
575 "\t guest physical addr: 0x%" PRIx64 "\n"
576 "\t guest virtual addr: 0x%" PRIx64 "\n"
577 "\t host virtual addr: 0x%" PRIx64 "\n"
578 "\t mmap addr : 0x%" PRIx64 "\n"
579 "\t mmap size : 0x%" PRIx64 "\n"
580 "\t mmap align: 0x%" PRIx64 "\n"
581 "\t mmap off : 0x%" PRIx64 "\n",
583 reg->guest_phys_addr,
584 reg->guest_user_addr,
586 (uint64_t)(uintptr_t)mmap_addr,
592 dump_guest_pages(dev);
597 free_mem_region(dev);
604 vq_is_ready(struct vhost_virtqueue *vq)
606 return vq && vq->desc &&
607 vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
608 vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
612 virtio_is_ready(struct virtio_net *dev)
614 struct vhost_virtqueue *rvq, *tvq;
617 for (i = 0; i < dev->virt_qp_nb; i++) {
618 rvq = dev->virtqueue[i * VIRTIO_QNUM + VIRTIO_RXQ];
619 tvq = dev->virtqueue[i * VIRTIO_QNUM + VIRTIO_TXQ];
621 if (!vq_is_ready(rvq) || !vq_is_ready(tvq)) {
622 RTE_LOG(INFO, VHOST_CONFIG,
623 "virtio is not ready for processing.\n");
628 RTE_LOG(INFO, VHOST_CONFIG,
629 "virtio is now ready for processing.\n");
634 vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *pmsg)
636 struct vhost_vring_file file;
637 struct vhost_virtqueue *vq;
640 file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
641 if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
642 file.fd = VIRTIO_INVALID_EVENTFD;
644 file.fd = pmsg->fds[0];
645 RTE_LOG(INFO, VHOST_CONFIG,
646 "vring call idx:%d file:%d\n", file.index, file.fd);
649 * FIXME: VHOST_SET_VRING_CALL is the first per-vring message
650 * we get, so we do vring queue pair allocation here.
652 cur_qp_idx = file.index / VIRTIO_QNUM;
653 if (cur_qp_idx + 1 > dev->virt_qp_nb) {
654 if (alloc_vring_queue_pair(dev, cur_qp_idx) < 0)
658 vq = dev->virtqueue[file.index];
664 vq->callfd = file.fd;
668 * In vhost-user, when we receive kick message, will test whether virtio
669 * device is ready for packet processing.
672 vhost_user_set_vring_kick(struct virtio_net *dev, struct VhostUserMsg *pmsg)
674 struct vhost_vring_file file;
675 struct vhost_virtqueue *vq;
677 file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
678 if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
679 file.fd = VIRTIO_INVALID_EVENTFD;
681 file.fd = pmsg->fds[0];
682 RTE_LOG(INFO, VHOST_CONFIG,
683 "vring kick idx:%d file:%d\n", file.index, file.fd);
685 vq = dev->virtqueue[file.index];
688 vq->kickfd = file.fd;
690 if (virtio_is_ready(dev) && !(dev->flags & VIRTIO_DEV_RUNNING)) {
691 if (dev->dequeue_zero_copy) {
692 RTE_LOG(INFO, VHOST_CONFIG,
693 "dequeue zero copy is enabled\n");
696 if (notify_ops->new_device(dev->vid) == 0)
697 dev->flags |= VIRTIO_DEV_RUNNING;
702 free_zmbufs(struct vhost_virtqueue *vq)
704 struct zcopy_mbuf *zmbuf, *next;
706 for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
707 zmbuf != NULL; zmbuf = next) {
708 next = TAILQ_NEXT(zmbuf, next);
710 rte_pktmbuf_free(zmbuf->mbuf);
711 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
714 rte_free(vq->zmbufs);
718 * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
721 vhost_user_get_vring_base(struct virtio_net *dev,
722 struct vhost_vring_state *state)
724 struct vhost_virtqueue *vq = dev->virtqueue[state->index];
726 /* We have to stop the queue (virtio) if it is running. */
727 if (dev->flags & VIRTIO_DEV_RUNNING) {
728 dev->flags &= ~VIRTIO_DEV_RUNNING;
729 notify_ops->destroy_device(dev->vid);
732 /* Here we are safe to get the last used index */
733 state->num = vq->last_used_idx;
735 RTE_LOG(INFO, VHOST_CONFIG,
736 "vring base idx:%d file:%d\n", state->index, state->num);
738 * Based on current qemu vhost-user implementation, this message is
739 * sent and only sent in vhost_vring_stop.
740 * TODO: cleanup the vring, it isn't usable since here.
745 vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
747 if (dev->dequeue_zero_copy)
749 rte_free(vq->shadow_used_ring);
750 vq->shadow_used_ring = NULL;
756 * when virtio queues are ready to work, qemu will send us to
757 * enable the virtio queue pair.
760 vhost_user_set_vring_enable(struct virtio_net *dev,
761 struct vhost_vring_state *state)
763 int enable = (int)state->num;
765 RTE_LOG(INFO, VHOST_CONFIG,
766 "set queue enable: %d to qp idx: %d\n",
767 enable, state->index);
769 if (notify_ops->vring_state_changed)
770 notify_ops->vring_state_changed(dev->vid, state->index, enable);
772 dev->virtqueue[state->index]->enabled = enable;
778 vhost_user_set_protocol_features(struct virtio_net *dev,
779 uint64_t protocol_features)
781 if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
784 dev->protocol_features = protocol_features;
788 vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
790 int fd = msg->fds[0];
795 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
799 if (msg->size != sizeof(VhostUserLog)) {
800 RTE_LOG(ERR, VHOST_CONFIG,
801 "invalid log base msg size: %"PRId32" != %d\n",
802 msg->size, (int)sizeof(VhostUserLog));
806 size = msg->payload.log.mmap_size;
807 off = msg->payload.log.mmap_offset;
808 RTE_LOG(INFO, VHOST_CONFIG,
809 "log mmap size: %"PRId64", offset: %"PRId64"\n",
813 * mmap from 0 to workaround a hugepage mmap bug: mmap will
814 * fail when offset is not page size aligned.
816 addr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
818 if (addr == MAP_FAILED) {
819 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
824 * Free previously mapped log memory on occasionally
825 * multiple VHOST_USER_SET_LOG_BASE.
828 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
830 dev->log_addr = (uint64_t)(uintptr_t)addr;
831 dev->log_base = dev->log_addr + off;
832 dev->log_size = size;
838 * An rarp packet is constructed and broadcasted to notify switches about
839 * the new location of the migrated VM, so that packets from outside will
840 * not be lost after migration.
842 * However, we don't actually "send" a rarp packet here, instead, we set
843 * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
846 vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
848 uint8_t *mac = (uint8_t *)&msg->payload.u64;
850 RTE_LOG(DEBUG, VHOST_CONFIG,
851 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
852 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
853 memcpy(dev->mac.addr_bytes, mac, 6);
856 * Set the flag to inject a RARP broadcast packet at
857 * rte_vhost_dequeue_burst().
859 * rte_smp_wmb() is for making sure the mac is copied
860 * before the flag is set.
863 rte_atomic16_set(&dev->broadcast_rarp, 1);
868 /* return bytes# of read on success or negative val on failure. */
870 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
874 ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
875 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
879 if (msg && msg->size) {
880 if (msg->size > sizeof(msg->payload)) {
881 RTE_LOG(ERR, VHOST_CONFIG,
882 "invalid msg size: %d\n", msg->size);
885 ret = read(sockfd, &msg->payload, msg->size);
888 if (ret != (int)msg->size) {
889 RTE_LOG(ERR, VHOST_CONFIG,
890 "read control message failed\n");
899 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
906 msg->flags &= ~VHOST_USER_VERSION_MASK;
907 msg->flags &= ~VHOST_USER_NEED_REPLY;
908 msg->flags |= VHOST_USER_VERSION;
909 msg->flags |= VHOST_USER_REPLY_MASK;
911 ret = send_fd_message(sockfd, (char *)msg,
912 VHOST_USER_HDR_SIZE + msg->size, NULL, 0);
918 vhost_user_msg_handler(int vid, int fd)
920 struct virtio_net *dev;
921 struct VhostUserMsg msg;
924 dev = get_device(vid);
928 ret = read_vhost_message(fd, &msg);
929 if (ret <= 0 || msg.request >= VHOST_USER_MAX) {
931 RTE_LOG(ERR, VHOST_CONFIG,
932 "vhost read message failed\n");
934 RTE_LOG(INFO, VHOST_CONFIG,
935 "vhost peer closed\n");
937 RTE_LOG(ERR, VHOST_CONFIG,
938 "vhost read incorrect message\n");
944 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
945 vhost_message_str[msg.request]);
946 switch (msg.request) {
947 case VHOST_USER_GET_FEATURES:
948 msg.payload.u64 = vhost_user_get_features();
949 msg.size = sizeof(msg.payload.u64);
950 send_vhost_message(fd, &msg);
952 case VHOST_USER_SET_FEATURES:
953 vhost_user_set_features(dev, msg.payload.u64);
956 case VHOST_USER_GET_PROTOCOL_FEATURES:
957 msg.payload.u64 = VHOST_USER_PROTOCOL_FEATURES;
958 msg.size = sizeof(msg.payload.u64);
959 send_vhost_message(fd, &msg);
961 case VHOST_USER_SET_PROTOCOL_FEATURES:
962 vhost_user_set_protocol_features(dev, msg.payload.u64);
965 case VHOST_USER_SET_OWNER:
966 vhost_user_set_owner();
968 case VHOST_USER_RESET_OWNER:
969 vhost_user_reset_owner(dev);
972 case VHOST_USER_SET_MEM_TABLE:
973 ret = vhost_user_set_mem_table(dev, &msg);
976 case VHOST_USER_SET_LOG_BASE:
977 vhost_user_set_log_base(dev, &msg);
979 /* it needs a reply */
980 msg.size = sizeof(msg.payload.u64);
981 send_vhost_message(fd, &msg);
983 case VHOST_USER_SET_LOG_FD:
985 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
988 case VHOST_USER_SET_VRING_NUM:
989 vhost_user_set_vring_num(dev, &msg.payload.state);
991 case VHOST_USER_SET_VRING_ADDR:
992 vhost_user_set_vring_addr(dev, &msg.payload.addr);
994 case VHOST_USER_SET_VRING_BASE:
995 vhost_user_set_vring_base(dev, &msg.payload.state);
998 case VHOST_USER_GET_VRING_BASE:
999 vhost_user_get_vring_base(dev, &msg.payload.state);
1000 msg.size = sizeof(msg.payload.state);
1001 send_vhost_message(fd, &msg);
1004 case VHOST_USER_SET_VRING_KICK:
1005 vhost_user_set_vring_kick(dev, &msg);
1007 case VHOST_USER_SET_VRING_CALL:
1008 vhost_user_set_vring_call(dev, &msg);
1011 case VHOST_USER_SET_VRING_ERR:
1012 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1014 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1017 case VHOST_USER_GET_QUEUE_NUM:
1018 msg.payload.u64 = VHOST_MAX_QUEUE_PAIRS;
1019 msg.size = sizeof(msg.payload.u64);
1020 send_vhost_message(fd, &msg);
1023 case VHOST_USER_SET_VRING_ENABLE:
1024 vhost_user_set_vring_enable(dev, &msg.payload.state);
1026 case VHOST_USER_SEND_RARP:
1027 vhost_user_send_rarp(dev, &msg);
1036 if (msg.flags & VHOST_USER_NEED_REPLY) {
1037 msg.payload.u64 = !!ret;
1038 msg.size = sizeof(msg.payload.u64);
1039 send_vhost_message(fd, &msg);