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;
318 vhost_user_iotlb_init(dev, index);
323 static struct virtio_net*
324 numa_realloc(struct virtio_net *dev, int index __rte_unused)
330 /* Converts QEMU virtual address to Vhost virtual address. */
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 +
353 * Converts ring address to Vhost virtual address.
354 * If IOMMU is enabled, the ring address is a guest IO virtual address,
355 * else it is a QEMU virtual address.
358 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
359 uint64_t ra, uint64_t size)
361 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
364 vva = vhost_user_iotlb_cache_find(vq, ra,
365 &size, VHOST_ACCESS_RW);
367 vhost_user_iotlb_miss(dev, ra, VHOST_ACCESS_RW);
372 return qva_to_vva(dev, ra);
376 * The virtio device sends us the desc, used and avail ring addresses.
377 * This function then converts these to our address space.
380 vhost_user_set_vring_addr(struct virtio_net *dev, VhostUserMsg *msg)
382 struct vhost_virtqueue *vq;
383 struct vhost_vring_addr *addr = &msg->payload.addr;
385 if (dev->mem == NULL)
388 /* addr->index refers to the queue index. The txq 1, rxq is 0. */
389 vq = dev->virtqueue[msg->payload.addr.index];
392 * Rings addresses should not be interpreted as long as the ring is not
393 * started and enabled
395 memcpy(&vq->ring_addrs, addr, sizeof(*addr));
397 vring_invalidate(dev, vq);
402 static struct virtio_net *
403 translate_ring_addresses(struct virtio_net *dev, int vq_index)
405 struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
406 struct vhost_vring_addr *addr = &vq->ring_addrs;
408 /* The addresses are converted from QEMU virtual to Vhost virtual. */
409 if (vq->desc && vq->avail && vq->used)
412 vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
413 vq, addr->desc_user_addr, sizeof(struct vring_desc));
415 RTE_LOG(DEBUG, VHOST_CONFIG,
416 "(%d) failed to find desc ring address.\n",
421 dev = numa_realloc(dev, vq_index);
422 vq = dev->virtqueue[vq_index];
423 addr = &vq->ring_addrs;
425 vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
426 vq, addr->avail_user_addr, sizeof(struct vring_avail));
427 if (vq->avail == 0) {
428 RTE_LOG(DEBUG, VHOST_CONFIG,
429 "(%d) failed to find avail ring address.\n",
434 vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
435 vq, addr->used_user_addr, sizeof(struct vring_used));
437 RTE_LOG(DEBUG, VHOST_CONFIG,
438 "(%d) failed to find used ring address.\n",
443 if (vq->last_used_idx != vq->used->idx) {
444 RTE_LOG(WARNING, VHOST_CONFIG,
445 "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
446 "some packets maybe resent for Tx and dropped for Rx\n",
447 vq->last_used_idx, vq->used->idx);
448 vq->last_used_idx = vq->used->idx;
449 vq->last_avail_idx = vq->used->idx;
452 vq->log_guest_addr = addr->log_guest_addr;
454 LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
456 LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
457 dev->vid, vq->avail);
458 LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
460 LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
461 dev->vid, vq->log_guest_addr);
467 * The virtio device sends us the available ring last used index.
470 vhost_user_set_vring_base(struct virtio_net *dev,
473 dev->virtqueue[msg->payload.state.index]->last_used_idx =
474 msg->payload.state.num;
475 dev->virtqueue[msg->payload.state.index]->last_avail_idx =
476 msg->payload.state.num;
482 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
483 uint64_t host_phys_addr, uint64_t size)
485 struct guest_page *page, *last_page;
487 if (dev->nr_guest_pages == dev->max_guest_pages) {
488 dev->max_guest_pages *= 2;
489 dev->guest_pages = realloc(dev->guest_pages,
490 dev->max_guest_pages * sizeof(*page));
493 if (dev->nr_guest_pages > 0) {
494 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
495 /* merge if the two pages are continuous */
496 if (host_phys_addr == last_page->host_phys_addr +
498 last_page->size += size;
503 page = &dev->guest_pages[dev->nr_guest_pages++];
504 page->guest_phys_addr = guest_phys_addr;
505 page->host_phys_addr = host_phys_addr;
510 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
513 uint64_t reg_size = reg->size;
514 uint64_t host_user_addr = reg->host_user_addr;
515 uint64_t guest_phys_addr = reg->guest_phys_addr;
516 uint64_t host_phys_addr;
519 host_phys_addr = rte_mem_virt2phy((void *)(uintptr_t)host_user_addr);
520 size = page_size - (guest_phys_addr & (page_size - 1));
521 size = RTE_MIN(size, reg_size);
523 add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
524 host_user_addr += size;
525 guest_phys_addr += size;
528 while (reg_size > 0) {
529 size = RTE_MIN(reg_size, page_size);
530 host_phys_addr = rte_mem_virt2phy((void *)(uintptr_t)
532 add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
534 host_user_addr += size;
535 guest_phys_addr += size;
540 #ifdef RTE_LIBRTE_VHOST_DEBUG
541 /* TODO: enable it only in debug mode? */
543 dump_guest_pages(struct virtio_net *dev)
546 struct guest_page *page;
548 for (i = 0; i < dev->nr_guest_pages; i++) {
549 page = &dev->guest_pages[i];
551 RTE_LOG(INFO, VHOST_CONFIG,
552 "guest physical page region %u\n"
553 "\t guest_phys_addr: %" PRIx64 "\n"
554 "\t host_phys_addr : %" PRIx64 "\n"
555 "\t size : %" PRIx64 "\n",
557 page->guest_phys_addr,
558 page->host_phys_addr,
563 #define dump_guest_pages(dev)
567 vhost_user_set_mem_table(struct virtio_net *dev, struct VhostUserMsg *pmsg)
569 struct VhostUserMemory memory = pmsg->payload.memory;
570 struct rte_vhost_mem_region *reg;
573 uint64_t mmap_offset;
579 free_mem_region(dev);
584 dev->nr_guest_pages = 0;
585 if (!dev->guest_pages) {
586 dev->max_guest_pages = 8;
587 dev->guest_pages = malloc(dev->max_guest_pages *
588 sizeof(struct guest_page));
589 if (dev->guest_pages == NULL) {
590 RTE_LOG(ERR, VHOST_CONFIG,
591 "(%d) failed to allocate memory "
592 "for dev->guest_pages\n",
598 dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
599 sizeof(struct rte_vhost_mem_region) * memory.nregions, 0);
600 if (dev->mem == NULL) {
601 RTE_LOG(ERR, VHOST_CONFIG,
602 "(%d) failed to allocate memory for dev->mem\n",
606 dev->mem->nregions = memory.nregions;
608 for (i = 0; i < memory.nregions; i++) {
610 reg = &dev->mem->regions[i];
612 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
613 reg->guest_user_addr = memory.regions[i].userspace_addr;
614 reg->size = memory.regions[i].memory_size;
617 mmap_offset = memory.regions[i].mmap_offset;
618 mmap_size = reg->size + mmap_offset;
620 /* mmap() without flag of MAP_ANONYMOUS, should be called
621 * with length argument aligned with hugepagesz at older
622 * longterm version Linux, like 2.6.32 and 3.2.72, or
623 * mmap() will fail with EINVAL.
625 * to avoid failure, make sure in caller to keep length
628 alignment = get_blk_size(fd);
629 if (alignment == (uint64_t)-1) {
630 RTE_LOG(ERR, VHOST_CONFIG,
631 "couldn't get hugepage size through fstat\n");
634 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
636 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
637 MAP_SHARED | MAP_POPULATE, fd, 0);
639 if (mmap_addr == MAP_FAILED) {
640 RTE_LOG(ERR, VHOST_CONFIG,
641 "mmap region %u failed.\n", i);
645 reg->mmap_addr = mmap_addr;
646 reg->mmap_size = mmap_size;
647 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
650 if (dev->dequeue_zero_copy)
651 add_guest_pages(dev, reg, alignment);
653 RTE_LOG(INFO, VHOST_CONFIG,
654 "guest memory region %u, size: 0x%" PRIx64 "\n"
655 "\t guest physical addr: 0x%" PRIx64 "\n"
656 "\t guest virtual addr: 0x%" PRIx64 "\n"
657 "\t host virtual addr: 0x%" PRIx64 "\n"
658 "\t mmap addr : 0x%" PRIx64 "\n"
659 "\t mmap size : 0x%" PRIx64 "\n"
660 "\t mmap align: 0x%" PRIx64 "\n"
661 "\t mmap off : 0x%" PRIx64 "\n",
663 reg->guest_phys_addr,
664 reg->guest_user_addr,
666 (uint64_t)(uintptr_t)mmap_addr,
672 dump_guest_pages(dev);
677 free_mem_region(dev);
684 vq_is_ready(struct vhost_virtqueue *vq)
686 return vq && vq->desc && vq->avail && vq->used &&
687 vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
688 vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
692 virtio_is_ready(struct virtio_net *dev)
694 struct vhost_virtqueue *vq;
697 if (dev->nr_vring == 0)
700 for (i = 0; i < dev->nr_vring; i++) {
701 vq = dev->virtqueue[i];
703 if (!vq_is_ready(vq))
707 RTE_LOG(INFO, VHOST_CONFIG,
708 "virtio is now ready for processing.\n");
713 vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *pmsg)
715 struct vhost_vring_file file;
716 struct vhost_virtqueue *vq;
718 file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
719 if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
720 file.fd = VIRTIO_INVALID_EVENTFD;
722 file.fd = pmsg->fds[0];
723 RTE_LOG(INFO, VHOST_CONFIG,
724 "vring call idx:%d file:%d\n", file.index, file.fd);
726 vq = dev->virtqueue[file.index];
730 vq->callfd = file.fd;
734 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *pmsg)
736 struct vhost_vring_file file;
737 struct vhost_virtqueue *vq;
738 struct virtio_net *dev = *pdev;
740 file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
741 if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
742 file.fd = VIRTIO_INVALID_EVENTFD;
744 file.fd = pmsg->fds[0];
745 RTE_LOG(INFO, VHOST_CONFIG,
746 "vring kick idx:%d file:%d\n", file.index, file.fd);
749 * Interpret ring addresses only when ring is started and enabled.
750 * This is now if protocol features aren't supported.
752 if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
753 *pdev = dev = translate_ring_addresses(dev, file.index);
758 vq = dev->virtqueue[file.index];
761 * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
762 * the ring starts already enabled. Otherwise, it is enabled via
763 * the SET_VRING_ENABLE message.
765 if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)))
770 vq->kickfd = file.fd;
774 free_zmbufs(struct vhost_virtqueue *vq)
776 struct zcopy_mbuf *zmbuf, *next;
778 for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
779 zmbuf != NULL; zmbuf = next) {
780 next = TAILQ_NEXT(zmbuf, next);
782 rte_pktmbuf_free(zmbuf->mbuf);
783 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
786 rte_free(vq->zmbufs);
790 * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
793 vhost_user_get_vring_base(struct virtio_net *dev,
796 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
798 /* We have to stop the queue (virtio) if it is running. */
799 if (dev->flags & VIRTIO_DEV_RUNNING) {
800 dev->flags &= ~VIRTIO_DEV_RUNNING;
801 dev->notify_ops->destroy_device(dev->vid);
804 dev->flags &= ~VIRTIO_DEV_READY;
806 /* Here we are safe to get the last used index */
807 msg->payload.state.num = vq->last_used_idx;
809 RTE_LOG(INFO, VHOST_CONFIG,
810 "vring base idx:%d file:%d\n", msg->payload.state.index,
811 msg->payload.state.num);
813 * Based on current qemu vhost-user implementation, this message is
814 * sent and only sent in vhost_vring_stop.
815 * TODO: cleanup the vring, it isn't usable since here.
820 vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
822 if (dev->dequeue_zero_copy)
824 rte_free(vq->shadow_used_ring);
825 vq->shadow_used_ring = NULL;
827 rte_free(vq->batch_copy_elems);
828 vq->batch_copy_elems = NULL;
834 * when virtio queues are ready to work, qemu will send us to
835 * enable the virtio queue pair.
838 vhost_user_set_vring_enable(struct virtio_net **pdev,
841 struct virtio_net *dev = *pdev;
842 int enable = (int)msg->payload.state.num;
844 RTE_LOG(INFO, VHOST_CONFIG,
845 "set queue enable: %d to qp idx: %d\n",
846 enable, msg->payload.state.index);
849 * Interpret ring addresses only when ring is started and enabled.
850 * This is now if protocol features are supported.
852 if (enable && (dev->features &
853 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
854 dev = translate_ring_addresses(dev, msg->payload.state.index);
861 if (dev->notify_ops->vring_state_changed)
862 dev->notify_ops->vring_state_changed(dev->vid,
863 msg->payload.state.index, enable);
865 dev->virtqueue[msg->payload.state.index]->enabled = enable;
871 vhost_user_set_protocol_features(struct virtio_net *dev,
872 uint64_t protocol_features)
874 if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
877 dev->protocol_features = protocol_features;
881 vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
883 int fd = msg->fds[0];
888 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
892 if (msg->size != sizeof(VhostUserLog)) {
893 RTE_LOG(ERR, VHOST_CONFIG,
894 "invalid log base msg size: %"PRId32" != %d\n",
895 msg->size, (int)sizeof(VhostUserLog));
899 size = msg->payload.log.mmap_size;
900 off = msg->payload.log.mmap_offset;
901 RTE_LOG(INFO, VHOST_CONFIG,
902 "log mmap size: %"PRId64", offset: %"PRId64"\n",
906 * mmap from 0 to workaround a hugepage mmap bug: mmap will
907 * fail when offset is not page size aligned.
909 addr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
911 if (addr == MAP_FAILED) {
912 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
917 * Free previously mapped log memory on occasionally
918 * multiple VHOST_USER_SET_LOG_BASE.
921 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
923 dev->log_addr = (uint64_t)(uintptr_t)addr;
924 dev->log_base = dev->log_addr + off;
925 dev->log_size = size;
931 * An rarp packet is constructed and broadcasted to notify switches about
932 * the new location of the migrated VM, so that packets from outside will
933 * not be lost after migration.
935 * However, we don't actually "send" a rarp packet here, instead, we set
936 * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
939 vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
941 uint8_t *mac = (uint8_t *)&msg->payload.u64;
943 RTE_LOG(DEBUG, VHOST_CONFIG,
944 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
945 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
946 memcpy(dev->mac.addr_bytes, mac, 6);
949 * Set the flag to inject a RARP broadcast packet at
950 * rte_vhost_dequeue_burst().
952 * rte_smp_wmb() is for making sure the mac is copied
953 * before the flag is set.
956 rte_atomic16_set(&dev->broadcast_rarp, 1);
962 vhost_user_net_set_mtu(struct virtio_net *dev, struct VhostUserMsg *msg)
964 if (msg->payload.u64 < VIRTIO_MIN_MTU ||
965 msg->payload.u64 > VIRTIO_MAX_MTU) {
966 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
972 dev->mtu = msg->payload.u64;
978 vhost_user_set_req_fd(struct virtio_net *dev, struct VhostUserMsg *msg)
980 int fd = msg->fds[0];
983 RTE_LOG(ERR, VHOST_CONFIG,
984 "Invalid file descriptor for slave channel (%d)\n",
989 dev->slave_req_fd = fd;
995 is_vring_iotlb_update(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
997 struct vhost_vring_addr *ra;
1001 end = start + imsg->size;
1003 ra = &vq->ring_addrs;
1004 if (ra->desc_user_addr >= start && ra->desc_user_addr < end)
1006 if (ra->avail_user_addr >= start && ra->avail_user_addr < end)
1008 if (ra->used_user_addr >= start && ra->used_user_addr < end)
1015 is_vring_iotlb_invalidate(struct vhost_virtqueue *vq,
1016 struct vhost_iotlb_msg *imsg)
1018 uint64_t istart, iend, vstart, vend;
1020 istart = imsg->iova;
1021 iend = istart + imsg->size - 1;
1023 vstart = (uintptr_t)vq->desc;
1024 vend = vstart + sizeof(struct vring_desc) * vq->size - 1;
1025 if (vstart <= iend && istart <= vend)
1028 vstart = (uintptr_t)vq->avail;
1029 vend = vstart + sizeof(struct vring_avail);
1030 vend += sizeof(uint16_t) * vq->size - 1;
1031 if (vstart <= iend && istart <= vend)
1034 vstart = (uintptr_t)vq->used;
1035 vend = vstart + sizeof(struct vring_used);
1036 vend += sizeof(struct vring_used_elem) * vq->size - 1;
1037 if (vstart <= iend && istart <= vend)
1044 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg)
1046 struct virtio_net *dev = *pdev;
1047 struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
1051 switch (imsg->type) {
1052 case VHOST_IOTLB_UPDATE:
1053 vva = qva_to_vva(dev, imsg->uaddr);
1057 for (i = 0; i < dev->nr_vring; i++) {
1058 struct vhost_virtqueue *vq = dev->virtqueue[i];
1060 vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
1061 imsg->size, imsg->perm);
1063 if (is_vring_iotlb_update(vq, imsg))
1064 *pdev = dev = translate_ring_addresses(dev, i);
1067 case VHOST_IOTLB_INVALIDATE:
1068 for (i = 0; i < dev->nr_vring; i++) {
1069 struct vhost_virtqueue *vq = dev->virtqueue[i];
1071 vhost_user_iotlb_cache_remove(vq, imsg->iova,
1074 if (is_vring_iotlb_invalidate(vq, imsg))
1075 vring_invalidate(dev, vq);
1079 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
1087 /* return bytes# of read on success or negative val on failure. */
1089 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
1093 ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
1094 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
1098 if (msg && msg->size) {
1099 if (msg->size > sizeof(msg->payload)) {
1100 RTE_LOG(ERR, VHOST_CONFIG,
1101 "invalid msg size: %d\n", msg->size);
1104 ret = read(sockfd, &msg->payload, msg->size);
1107 if (ret != (int)msg->size) {
1108 RTE_LOG(ERR, VHOST_CONFIG,
1109 "read control message failed\n");
1118 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
1123 return send_fd_message(sockfd, (char *)msg,
1124 VHOST_USER_HDR_SIZE + msg->size, NULL, 0);
1128 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
1133 msg->flags &= ~VHOST_USER_VERSION_MASK;
1134 msg->flags &= ~VHOST_USER_NEED_REPLY;
1135 msg->flags |= VHOST_USER_VERSION;
1136 msg->flags |= VHOST_USER_REPLY_MASK;
1138 return send_vhost_message(sockfd, msg);
1142 * Allocate a queue pair if it hasn't been allocated yet
1145 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev, VhostUserMsg *msg)
1149 switch (msg->request.master) {
1150 case VHOST_USER_SET_VRING_KICK:
1151 case VHOST_USER_SET_VRING_CALL:
1152 case VHOST_USER_SET_VRING_ERR:
1153 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1155 case VHOST_USER_SET_VRING_NUM:
1156 case VHOST_USER_SET_VRING_BASE:
1157 case VHOST_USER_SET_VRING_ENABLE:
1158 vring_idx = msg->payload.state.index;
1160 case VHOST_USER_SET_VRING_ADDR:
1161 vring_idx = msg->payload.addr.index;
1167 if (vring_idx >= VHOST_MAX_VRING) {
1168 RTE_LOG(ERR, VHOST_CONFIG,
1169 "invalid vring index: %u\n", vring_idx);
1173 if (dev->virtqueue[vring_idx])
1176 return alloc_vring_queue(dev, vring_idx);
1180 vhost_user_msg_handler(int vid, int fd)
1182 struct virtio_net *dev;
1183 struct VhostUserMsg msg;
1186 dev = get_device(vid);
1190 if (!dev->notify_ops) {
1191 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
1192 if (!dev->notify_ops) {
1193 RTE_LOG(ERR, VHOST_CONFIG,
1194 "failed to get callback ops for driver %s\n",
1200 ret = read_vhost_message(fd, &msg);
1201 if (ret <= 0 || msg.request.master >= VHOST_USER_MAX) {
1203 RTE_LOG(ERR, VHOST_CONFIG,
1204 "vhost read message failed\n");
1206 RTE_LOG(INFO, VHOST_CONFIG,
1207 "vhost peer closed\n");
1209 RTE_LOG(ERR, VHOST_CONFIG,
1210 "vhost read incorrect message\n");
1216 if (msg.request.master != VHOST_USER_IOTLB_MSG)
1217 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1218 vhost_message_str[msg.request.master]);
1220 RTE_LOG(DEBUG, VHOST_CONFIG, "read message %s\n",
1221 vhost_message_str[msg.request.master]);
1223 ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1225 RTE_LOG(ERR, VHOST_CONFIG,
1226 "failed to alloc queue\n");
1230 switch (msg.request.master) {
1231 case VHOST_USER_GET_FEATURES:
1232 msg.payload.u64 = vhost_user_get_features(dev);
1233 msg.size = sizeof(msg.payload.u64);
1234 send_vhost_reply(fd, &msg);
1236 case VHOST_USER_SET_FEATURES:
1237 vhost_user_set_features(dev, msg.payload.u64);
1240 case VHOST_USER_GET_PROTOCOL_FEATURES:
1241 msg.payload.u64 = VHOST_USER_PROTOCOL_FEATURES;
1242 msg.size = sizeof(msg.payload.u64);
1243 send_vhost_reply(fd, &msg);
1245 case VHOST_USER_SET_PROTOCOL_FEATURES:
1246 vhost_user_set_protocol_features(dev, msg.payload.u64);
1249 case VHOST_USER_SET_OWNER:
1250 vhost_user_set_owner();
1252 case VHOST_USER_RESET_OWNER:
1253 vhost_user_reset_owner(dev);
1256 case VHOST_USER_SET_MEM_TABLE:
1257 ret = vhost_user_set_mem_table(dev, &msg);
1260 case VHOST_USER_SET_LOG_BASE:
1261 vhost_user_set_log_base(dev, &msg);
1263 /* it needs a reply */
1264 msg.size = sizeof(msg.payload.u64);
1265 send_vhost_reply(fd, &msg);
1267 case VHOST_USER_SET_LOG_FD:
1269 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1272 case VHOST_USER_SET_VRING_NUM:
1273 vhost_user_set_vring_num(dev, &msg);
1275 case VHOST_USER_SET_VRING_ADDR:
1276 vhost_user_set_vring_addr(dev, &msg);
1278 case VHOST_USER_SET_VRING_BASE:
1279 vhost_user_set_vring_base(dev, &msg);
1282 case VHOST_USER_GET_VRING_BASE:
1283 vhost_user_get_vring_base(dev, &msg);
1284 msg.size = sizeof(msg.payload.state);
1285 send_vhost_reply(fd, &msg);
1288 case VHOST_USER_SET_VRING_KICK:
1289 vhost_user_set_vring_kick(&dev, &msg);
1291 case VHOST_USER_SET_VRING_CALL:
1292 vhost_user_set_vring_call(dev, &msg);
1295 case VHOST_USER_SET_VRING_ERR:
1296 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1298 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1301 case VHOST_USER_GET_QUEUE_NUM:
1302 msg.payload.u64 = VHOST_MAX_QUEUE_PAIRS;
1303 msg.size = sizeof(msg.payload.u64);
1304 send_vhost_reply(fd, &msg);
1307 case VHOST_USER_SET_VRING_ENABLE:
1308 vhost_user_set_vring_enable(&dev, &msg);
1310 case VHOST_USER_SEND_RARP:
1311 vhost_user_send_rarp(dev, &msg);
1314 case VHOST_USER_NET_SET_MTU:
1315 ret = vhost_user_net_set_mtu(dev, &msg);
1318 case VHOST_USER_SET_SLAVE_REQ_FD:
1319 ret = vhost_user_set_req_fd(dev, &msg);
1322 case VHOST_USER_IOTLB_MSG:
1323 ret = vhost_user_iotlb_msg(&dev, &msg);
1332 if (msg.flags & VHOST_USER_NEED_REPLY) {
1333 msg.payload.u64 = !!ret;
1334 msg.size = sizeof(msg.payload.u64);
1335 send_vhost_reply(fd, &msg);
1338 if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1339 dev->flags |= VIRTIO_DEV_READY;
1341 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1342 if (dev->dequeue_zero_copy) {
1343 RTE_LOG(INFO, VHOST_CONFIG,
1344 "dequeue zero copy is enabled\n");
1347 if (dev->notify_ops->new_device(dev->vid) == 0)
1348 dev->flags |= VIRTIO_DEV_RUNNING;
1356 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
1359 struct VhostUserMsg msg = {
1360 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
1361 .flags = VHOST_USER_VERSION,
1362 .size = sizeof(msg.payload.iotlb),
1366 .type = VHOST_IOTLB_MISS,
1370 ret = send_vhost_message(dev->slave_req_fd, &msg);
1372 RTE_LOG(ERR, VHOST_CONFIG,
1373 "Failed to send IOTLB miss message (%d)\n",