1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2018 Intel Corporation
7 * The vhost-user protocol connection is an external interface, so it must be
8 * robust against invalid inputs.
10 * This is important because the vhost-user master is only one step removed
11 * from the guest. Malicious guests that have escaped will then launch further
12 * attacks from the vhost-user master.
14 * Even in deployments where guests are trusted, a bug in the vhost-user master
15 * can still cause invalid messages to be sent. Such messages must not
16 * compromise the stability of the DPDK application by causing crashes, memory
17 * corruption, or other problematic behavior.
19 * Do not assume received VhostUserMsg fields contain sensible values!
28 #include <sys/ioctl.h>
30 #include <sys/types.h>
32 #include <sys/syscall.h>
34 #ifdef RTE_LIBRTE_VHOST_NUMA
37 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
38 #include <linux/userfaultfd.h>
40 #ifdef F_ADD_SEALS /* if file sealing is supported, so is memfd */
41 #include <linux/memfd.h>
42 #define MEMFD_SUPPORTED
45 #include <rte_common.h>
46 #include <rte_malloc.h>
51 #include "vhost_user.h"
53 #define VIRTIO_MIN_MTU 68
54 #define VIRTIO_MAX_MTU 65535
56 #define INFLIGHT_ALIGNMENT 64
57 #define INFLIGHT_VERSION 0x1
59 static const char *vhost_message_str[VHOST_USER_MAX] = {
60 [VHOST_USER_NONE] = "VHOST_USER_NONE",
61 [VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
62 [VHOST_USER_SET_FEATURES] = "VHOST_USER_SET_FEATURES",
63 [VHOST_USER_SET_OWNER] = "VHOST_USER_SET_OWNER",
64 [VHOST_USER_RESET_OWNER] = "VHOST_USER_RESET_OWNER",
65 [VHOST_USER_SET_MEM_TABLE] = "VHOST_USER_SET_MEM_TABLE",
66 [VHOST_USER_SET_LOG_BASE] = "VHOST_USER_SET_LOG_BASE",
67 [VHOST_USER_SET_LOG_FD] = "VHOST_USER_SET_LOG_FD",
68 [VHOST_USER_SET_VRING_NUM] = "VHOST_USER_SET_VRING_NUM",
69 [VHOST_USER_SET_VRING_ADDR] = "VHOST_USER_SET_VRING_ADDR",
70 [VHOST_USER_SET_VRING_BASE] = "VHOST_USER_SET_VRING_BASE",
71 [VHOST_USER_GET_VRING_BASE] = "VHOST_USER_GET_VRING_BASE",
72 [VHOST_USER_SET_VRING_KICK] = "VHOST_USER_SET_VRING_KICK",
73 [VHOST_USER_SET_VRING_CALL] = "VHOST_USER_SET_VRING_CALL",
74 [VHOST_USER_SET_VRING_ERR] = "VHOST_USER_SET_VRING_ERR",
75 [VHOST_USER_GET_PROTOCOL_FEATURES] = "VHOST_USER_GET_PROTOCOL_FEATURES",
76 [VHOST_USER_SET_PROTOCOL_FEATURES] = "VHOST_USER_SET_PROTOCOL_FEATURES",
77 [VHOST_USER_GET_QUEUE_NUM] = "VHOST_USER_GET_QUEUE_NUM",
78 [VHOST_USER_SET_VRING_ENABLE] = "VHOST_USER_SET_VRING_ENABLE",
79 [VHOST_USER_SEND_RARP] = "VHOST_USER_SEND_RARP",
80 [VHOST_USER_NET_SET_MTU] = "VHOST_USER_NET_SET_MTU",
81 [VHOST_USER_SET_SLAVE_REQ_FD] = "VHOST_USER_SET_SLAVE_REQ_FD",
82 [VHOST_USER_IOTLB_MSG] = "VHOST_USER_IOTLB_MSG",
83 [VHOST_USER_CRYPTO_CREATE_SESS] = "VHOST_USER_CRYPTO_CREATE_SESS",
84 [VHOST_USER_CRYPTO_CLOSE_SESS] = "VHOST_USER_CRYPTO_CLOSE_SESS",
85 [VHOST_USER_POSTCOPY_ADVISE] = "VHOST_USER_POSTCOPY_ADVISE",
86 [VHOST_USER_POSTCOPY_LISTEN] = "VHOST_USER_POSTCOPY_LISTEN",
87 [VHOST_USER_POSTCOPY_END] = "VHOST_USER_POSTCOPY_END",
88 [VHOST_USER_GET_INFLIGHT_FD] = "VHOST_USER_GET_INFLIGHT_FD",
89 [VHOST_USER_SET_INFLIGHT_FD] = "VHOST_USER_SET_INFLIGHT_FD",
90 [VHOST_USER_SET_STATUS] = "VHOST_USER_SET_STATUS",
91 [VHOST_USER_GET_STATUS] = "VHOST_USER_GET_STATUS",
94 static int send_vhost_reply(int sockfd, struct VhostUserMsg *msg);
95 static int read_vhost_message(int sockfd, struct VhostUserMsg *msg);
98 close_msg_fds(struct VhostUserMsg *msg)
102 for (i = 0; i < msg->fd_num; i++)
107 * Ensure the expected number of FDs is received,
108 * close all FDs and return an error if this is not the case.
111 validate_msg_fds(struct VhostUserMsg *msg, int expected_fds)
113 if (msg->fd_num == expected_fds)
116 VHOST_LOG_CONFIG(ERR,
117 " Expect %d FDs for request %s, received %d\n",
119 vhost_message_str[msg->request.master],
133 ret = fstat(fd, &stat);
134 return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
138 free_mem_region(struct virtio_net *dev)
141 struct rte_vhost_mem_region *reg;
143 if (!dev || !dev->mem)
146 for (i = 0; i < dev->mem->nregions; i++) {
147 reg = &dev->mem->regions[i];
148 if (reg->host_user_addr) {
149 munmap(reg->mmap_addr, reg->mmap_size);
156 vhost_backend_cleanup(struct virtio_net *dev)
159 free_mem_region(dev);
164 rte_free(dev->guest_pages);
165 dev->guest_pages = NULL;
168 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
172 if (dev->inflight_info) {
173 if (dev->inflight_info->addr) {
174 munmap(dev->inflight_info->addr,
175 dev->inflight_info->size);
176 dev->inflight_info->addr = NULL;
179 if (dev->inflight_info->fd >= 0) {
180 close(dev->inflight_info->fd);
181 dev->inflight_info->fd = -1;
184 free(dev->inflight_info);
185 dev->inflight_info = NULL;
188 if (dev->slave_req_fd >= 0) {
189 close(dev->slave_req_fd);
190 dev->slave_req_fd = -1;
193 if (dev->postcopy_ufd >= 0) {
194 close(dev->postcopy_ufd);
195 dev->postcopy_ufd = -1;
198 dev->postcopy_listening = 0;
202 vhost_user_notify_queue_state(struct virtio_net *dev, uint16_t index,
205 struct rte_vdpa_device *vdpa_dev = dev->vdpa_dev;
206 struct vhost_virtqueue *vq = dev->virtqueue[index];
208 /* Configure guest notifications on enable */
209 if (enable && vq->notif_enable != VIRTIO_UNINITIALIZED_NOTIF)
210 vhost_enable_guest_notification(dev, vq, vq->notif_enable);
212 if (vdpa_dev && vdpa_dev->ops->set_vring_state)
213 vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
215 if (dev->notify_ops->vring_state_changed)
216 dev->notify_ops->vring_state_changed(dev->vid,
221 * This function just returns success at the moment unless
222 * the device hasn't been initialised.
225 vhost_user_set_owner(struct virtio_net **pdev __rte_unused,
226 struct VhostUserMsg *msg,
227 int main_fd __rte_unused)
229 if (validate_msg_fds(msg, 0) != 0)
230 return RTE_VHOST_MSG_RESULT_ERR;
232 return RTE_VHOST_MSG_RESULT_OK;
236 vhost_user_reset_owner(struct virtio_net **pdev,
237 struct VhostUserMsg *msg,
238 int main_fd __rte_unused)
240 struct virtio_net *dev = *pdev;
242 if (validate_msg_fds(msg, 0) != 0)
243 return RTE_VHOST_MSG_RESULT_ERR;
245 vhost_destroy_device_notify(dev);
247 cleanup_device(dev, 0);
249 return RTE_VHOST_MSG_RESULT_OK;
253 * The features that we support are requested.
256 vhost_user_get_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
257 int main_fd __rte_unused)
259 struct virtio_net *dev = *pdev;
260 uint64_t features = 0;
262 if (validate_msg_fds(msg, 0) != 0)
263 return RTE_VHOST_MSG_RESULT_ERR;
265 rte_vhost_driver_get_features(dev->ifname, &features);
267 msg->payload.u64 = features;
268 msg->size = sizeof(msg->payload.u64);
271 return RTE_VHOST_MSG_RESULT_REPLY;
275 * The queue number that we support are requested.
278 vhost_user_get_queue_num(struct virtio_net **pdev, struct VhostUserMsg *msg,
279 int main_fd __rte_unused)
281 struct virtio_net *dev = *pdev;
282 uint32_t queue_num = 0;
284 if (validate_msg_fds(msg, 0) != 0)
285 return RTE_VHOST_MSG_RESULT_ERR;
287 rte_vhost_driver_get_queue_num(dev->ifname, &queue_num);
289 msg->payload.u64 = (uint64_t)queue_num;
290 msg->size = sizeof(msg->payload.u64);
293 return RTE_VHOST_MSG_RESULT_REPLY;
297 * We receive the negotiated features supported by us and the virtio device.
300 vhost_user_set_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
301 int main_fd __rte_unused)
303 struct virtio_net *dev = *pdev;
304 uint64_t features = msg->payload.u64;
305 uint64_t vhost_features = 0;
306 struct rte_vdpa_device *vdpa_dev;
308 if (validate_msg_fds(msg, 0) != 0)
309 return RTE_VHOST_MSG_RESULT_ERR;
311 rte_vhost_driver_get_features(dev->ifname, &vhost_features);
312 if (features & ~vhost_features) {
313 VHOST_LOG_CONFIG(ERR,
314 "(%d) received invalid negotiated features.\n",
316 dev->flags |= VIRTIO_DEV_FEATURES_FAILED;
317 dev->status &= ~VIRTIO_DEVICE_STATUS_FEATURES_OK;
319 return RTE_VHOST_MSG_RESULT_ERR;
322 if (dev->flags & VIRTIO_DEV_RUNNING) {
323 if (dev->features == features)
324 return RTE_VHOST_MSG_RESULT_OK;
327 * Error out if master tries to change features while device is
328 * in running state. The exception being VHOST_F_LOG_ALL, which
329 * is enabled when the live-migration starts.
331 if ((dev->features ^ features) & ~(1ULL << VHOST_F_LOG_ALL)) {
332 VHOST_LOG_CONFIG(ERR,
333 "(%d) features changed while device is running.\n",
335 return RTE_VHOST_MSG_RESULT_ERR;
338 if (dev->notify_ops->features_changed)
339 dev->notify_ops->features_changed(dev->vid, features);
342 dev->features = features;
344 ((1ULL << VIRTIO_NET_F_MRG_RXBUF) |
345 (1ULL << VIRTIO_F_VERSION_1) |
346 (1ULL << VIRTIO_F_RING_PACKED))) {
347 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
349 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
351 VHOST_LOG_CONFIG(INFO,
352 "negotiated Virtio features: 0x%" PRIx64 "\n", dev->features);
353 VHOST_LOG_CONFIG(DEBUG,
354 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
356 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
357 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
359 if ((dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) &&
360 !(dev->features & (1ULL << VIRTIO_NET_F_MQ))) {
362 * Remove all but first queue pair if MQ hasn't been
363 * negotiated. This is safe because the device is not
364 * running at this stage.
366 while (dev->nr_vring > 2) {
367 struct vhost_virtqueue *vq;
369 vq = dev->virtqueue[--dev->nr_vring];
373 dev->virtqueue[dev->nr_vring] = NULL;
375 cleanup_vq_inflight(dev, vq);
380 vdpa_dev = dev->vdpa_dev;
382 vdpa_dev->ops->set_features(dev->vid);
384 dev->flags &= ~VIRTIO_DEV_FEATURES_FAILED;
385 return RTE_VHOST_MSG_RESULT_OK;
389 * The virtio device sends us the size of the descriptor ring.
392 vhost_user_set_vring_num(struct virtio_net **pdev,
393 struct VhostUserMsg *msg,
394 int main_fd __rte_unused)
396 struct virtio_net *dev = *pdev;
397 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
399 if (validate_msg_fds(msg, 0) != 0)
400 return RTE_VHOST_MSG_RESULT_ERR;
402 vq->size = msg->payload.state.num;
404 /* VIRTIO 1.0, 2.4 Virtqueues says:
406 * Queue Size value is always a power of 2. The maximum Queue Size
409 * VIRTIO 1.1 2.7 Virtqueues says:
411 * Packed virtqueues support up to 2^15 entries each.
413 if (!vq_is_packed(dev)) {
414 if (vq->size & (vq->size - 1)) {
415 VHOST_LOG_CONFIG(ERR,
416 "invalid virtqueue size %u\n", vq->size);
417 return RTE_VHOST_MSG_RESULT_ERR;
421 if (vq->size > 32768) {
422 VHOST_LOG_CONFIG(ERR,
423 "invalid virtqueue size %u\n", vq->size);
424 return RTE_VHOST_MSG_RESULT_ERR;
427 if (vq_is_packed(dev)) {
428 if (vq->shadow_used_packed)
429 rte_free(vq->shadow_used_packed);
430 vq->shadow_used_packed = rte_malloc(NULL,
432 sizeof(struct vring_used_elem_packed),
433 RTE_CACHE_LINE_SIZE);
434 if (!vq->shadow_used_packed) {
435 VHOST_LOG_CONFIG(ERR,
436 "failed to allocate memory for shadow used ring.\n");
437 return RTE_VHOST_MSG_RESULT_ERR;
441 if (vq->shadow_used_split)
442 rte_free(vq->shadow_used_split);
444 vq->shadow_used_split = rte_malloc(NULL,
445 vq->size * sizeof(struct vring_used_elem),
446 RTE_CACHE_LINE_SIZE);
448 if (!vq->shadow_used_split) {
449 VHOST_LOG_CONFIG(ERR,
450 "failed to allocate memory for vq internal data.\n");
451 return RTE_VHOST_MSG_RESULT_ERR;
455 if (vq->batch_copy_elems)
456 rte_free(vq->batch_copy_elems);
457 vq->batch_copy_elems = rte_malloc(NULL,
458 vq->size * sizeof(struct batch_copy_elem),
459 RTE_CACHE_LINE_SIZE);
460 if (!vq->batch_copy_elems) {
461 VHOST_LOG_CONFIG(ERR,
462 "failed to allocate memory for batching copy.\n");
463 return RTE_VHOST_MSG_RESULT_ERR;
466 return RTE_VHOST_MSG_RESULT_OK;
470 * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
471 * same numa node as the memory of vring descriptor.
473 #ifdef RTE_LIBRTE_VHOST_NUMA
474 static struct virtio_net*
475 numa_realloc(struct virtio_net *dev, int index)
477 int oldnode, newnode;
478 struct virtio_net *old_dev;
479 struct vhost_virtqueue *old_vq, *vq;
480 struct vring_used_elem *new_shadow_used_split;
481 struct vring_used_elem_packed *new_shadow_used_packed;
482 struct batch_copy_elem *new_batch_copy_elems;
485 if (dev->flags & VIRTIO_DEV_RUNNING)
489 vq = old_vq = dev->virtqueue[index];
491 ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
492 MPOL_F_NODE | MPOL_F_ADDR);
494 /* check if we need to reallocate vq */
495 ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
496 MPOL_F_NODE | MPOL_F_ADDR);
498 VHOST_LOG_CONFIG(ERR,
499 "Unable to get vq numa information.\n");
502 if (oldnode != newnode) {
503 VHOST_LOG_CONFIG(INFO,
504 "reallocate vq from %d to %d node\n", oldnode, newnode);
505 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
509 memcpy(vq, old_vq, sizeof(*vq));
511 if (vq_is_packed(dev)) {
512 new_shadow_used_packed = rte_malloc_socket(NULL,
514 sizeof(struct vring_used_elem_packed),
517 if (new_shadow_used_packed) {
518 rte_free(vq->shadow_used_packed);
519 vq->shadow_used_packed = new_shadow_used_packed;
522 new_shadow_used_split = rte_malloc_socket(NULL,
524 sizeof(struct vring_used_elem),
527 if (new_shadow_used_split) {
528 rte_free(vq->shadow_used_split);
529 vq->shadow_used_split = new_shadow_used_split;
533 new_batch_copy_elems = rte_malloc_socket(NULL,
534 vq->size * sizeof(struct batch_copy_elem),
537 if (new_batch_copy_elems) {
538 rte_free(vq->batch_copy_elems);
539 vq->batch_copy_elems = new_batch_copy_elems;
545 /* check if we need to reallocate dev */
546 ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
547 MPOL_F_NODE | MPOL_F_ADDR);
549 VHOST_LOG_CONFIG(ERR,
550 "Unable to get dev numa information.\n");
553 if (oldnode != newnode) {
554 VHOST_LOG_CONFIG(INFO,
555 "reallocate dev from %d to %d node\n",
557 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
563 memcpy(dev, old_dev, sizeof(*dev));
568 dev->virtqueue[index] = vq;
569 vhost_devices[dev->vid] = dev;
572 vhost_user_iotlb_init(dev, index);
577 static struct virtio_net*
578 numa_realloc(struct virtio_net *dev, int index __rte_unused)
584 /* Converts QEMU virtual address to Vhost virtual address. */
586 qva_to_vva(struct virtio_net *dev, uint64_t qva, uint64_t *len)
588 struct rte_vhost_mem_region *r;
591 if (unlikely(!dev || !dev->mem))
594 /* Find the region where the address lives. */
595 for (i = 0; i < dev->mem->nregions; i++) {
596 r = &dev->mem->regions[i];
598 if (qva >= r->guest_user_addr &&
599 qva < r->guest_user_addr + r->size) {
601 if (unlikely(*len > r->guest_user_addr + r->size - qva))
602 *len = r->guest_user_addr + r->size - qva;
604 return qva - r->guest_user_addr +
616 * Converts ring address to Vhost virtual address.
617 * If IOMMU is enabled, the ring address is a guest IO virtual address,
618 * else it is a QEMU virtual address.
621 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
622 uint64_t ra, uint64_t *size)
624 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
627 vhost_user_iotlb_rd_lock(vq);
628 vva = vhost_iova_to_vva(dev, vq, ra,
629 size, VHOST_ACCESS_RW);
630 vhost_user_iotlb_rd_unlock(vq);
635 return qva_to_vva(dev, ra, size);
639 log_addr_to_gpa(struct virtio_net *dev, struct vhost_virtqueue *vq)
643 vhost_user_iotlb_rd_lock(vq);
644 log_gpa = translate_log_addr(dev, vq, vq->ring_addrs.log_guest_addr);
645 vhost_user_iotlb_rd_unlock(vq);
650 static struct virtio_net *
651 translate_ring_addresses(struct virtio_net *dev, int vq_index)
653 struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
654 struct vhost_vring_addr *addr = &vq->ring_addrs;
655 uint64_t len, expected_len;
657 if (addr->flags & (1 << VHOST_VRING_F_LOG)) {
659 log_addr_to_gpa(dev, vq);
660 if (vq->log_guest_addr == 0) {
661 VHOST_LOG_CONFIG(DEBUG,
662 "(%d) failed to map log_guest_addr.\n",
668 if (vq_is_packed(dev)) {
669 len = sizeof(struct vring_packed_desc) * vq->size;
670 vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
671 ring_addr_to_vva(dev, vq, addr->desc_user_addr, &len);
672 if (vq->desc_packed == NULL ||
673 len != sizeof(struct vring_packed_desc) *
675 VHOST_LOG_CONFIG(DEBUG,
676 "(%d) failed to map desc_packed ring.\n",
681 dev = numa_realloc(dev, vq_index);
682 vq = dev->virtqueue[vq_index];
683 addr = &vq->ring_addrs;
685 len = sizeof(struct vring_packed_desc_event);
686 vq->driver_event = (struct vring_packed_desc_event *)
687 (uintptr_t)ring_addr_to_vva(dev,
688 vq, addr->avail_user_addr, &len);
689 if (vq->driver_event == NULL ||
690 len != sizeof(struct vring_packed_desc_event)) {
691 VHOST_LOG_CONFIG(DEBUG,
692 "(%d) failed to find driver area address.\n",
697 len = sizeof(struct vring_packed_desc_event);
698 vq->device_event = (struct vring_packed_desc_event *)
699 (uintptr_t)ring_addr_to_vva(dev,
700 vq, addr->used_user_addr, &len);
701 if (vq->device_event == NULL ||
702 len != sizeof(struct vring_packed_desc_event)) {
703 VHOST_LOG_CONFIG(DEBUG,
704 "(%d) failed to find device area address.\n",
713 /* The addresses are converted from QEMU virtual to Vhost virtual. */
714 if (vq->desc && vq->avail && vq->used)
717 len = sizeof(struct vring_desc) * vq->size;
718 vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
719 vq, addr->desc_user_addr, &len);
720 if (vq->desc == 0 || len != sizeof(struct vring_desc) * vq->size) {
721 VHOST_LOG_CONFIG(DEBUG,
722 "(%d) failed to map desc ring.\n",
727 dev = numa_realloc(dev, vq_index);
728 vq = dev->virtqueue[vq_index];
729 addr = &vq->ring_addrs;
731 len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
732 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
733 len += sizeof(uint16_t);
735 vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
736 vq, addr->avail_user_addr, &len);
737 if (vq->avail == 0 || len != expected_len) {
738 VHOST_LOG_CONFIG(DEBUG,
739 "(%d) failed to map avail ring.\n",
744 len = sizeof(struct vring_used) +
745 sizeof(struct vring_used_elem) * vq->size;
746 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
747 len += sizeof(uint16_t);
749 vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
750 vq, addr->used_user_addr, &len);
751 if (vq->used == 0 || len != expected_len) {
752 VHOST_LOG_CONFIG(DEBUG,
753 "(%d) failed to map used ring.\n",
758 if (vq->last_used_idx != vq->used->idx) {
759 VHOST_LOG_CONFIG(WARNING,
760 "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
761 "some packets maybe resent for Tx and dropped for Rx\n",
762 vq->last_used_idx, vq->used->idx);
763 vq->last_used_idx = vq->used->idx;
764 vq->last_avail_idx = vq->used->idx;
769 VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address desc: %p\n",
771 VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address avail: %p\n",
772 dev->vid, vq->avail);
773 VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address used: %p\n",
775 VHOST_LOG_CONFIG(DEBUG, "(%d) log_guest_addr: %" PRIx64 "\n",
776 dev->vid, vq->log_guest_addr);
782 * The virtio device sends us the desc, used and avail ring addresses.
783 * This function then converts these to our address space.
786 vhost_user_set_vring_addr(struct virtio_net **pdev, struct VhostUserMsg *msg,
787 int main_fd __rte_unused)
789 struct virtio_net *dev = *pdev;
790 struct vhost_virtqueue *vq;
791 struct vhost_vring_addr *addr = &msg->payload.addr;
794 if (validate_msg_fds(msg, 0) != 0)
795 return RTE_VHOST_MSG_RESULT_ERR;
797 if (dev->mem == NULL)
798 return RTE_VHOST_MSG_RESULT_ERR;
800 /* addr->index refers to the queue index. The txq 1, rxq is 0. */
801 vq = dev->virtqueue[msg->payload.addr.index];
803 access_ok = vq->access_ok;
806 * Rings addresses should not be interpreted as long as the ring is not
807 * started and enabled
809 memcpy(&vq->ring_addrs, addr, sizeof(*addr));
811 vring_invalidate(dev, vq);
813 if ((vq->enabled && (dev->features &
814 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) ||
816 dev = translate_ring_addresses(dev, msg->payload.addr.index);
818 return RTE_VHOST_MSG_RESULT_ERR;
823 return RTE_VHOST_MSG_RESULT_OK;
827 * The virtio device sends us the available ring last used index.
830 vhost_user_set_vring_base(struct virtio_net **pdev,
831 struct VhostUserMsg *msg,
832 int main_fd __rte_unused)
834 struct virtio_net *dev = *pdev;
835 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
836 uint64_t val = msg->payload.state.num;
838 if (validate_msg_fds(msg, 0) != 0)
839 return RTE_VHOST_MSG_RESULT_ERR;
841 if (vq_is_packed(dev)) {
843 * Bit[0:14]: avail index
844 * Bit[15]: avail wrap counter
846 vq->last_avail_idx = val & 0x7fff;
847 vq->avail_wrap_counter = !!(val & (0x1 << 15));
849 * Set used index to same value as available one, as
850 * their values should be the same since ring processing
851 * was stopped at get time.
853 vq->last_used_idx = vq->last_avail_idx;
854 vq->used_wrap_counter = vq->avail_wrap_counter;
856 vq->last_used_idx = msg->payload.state.num;
857 vq->last_avail_idx = msg->payload.state.num;
860 return RTE_VHOST_MSG_RESULT_OK;
864 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
865 uint64_t host_phys_addr, uint64_t size)
867 struct guest_page *page, *last_page;
868 struct guest_page *old_pages;
870 if (dev->nr_guest_pages == dev->max_guest_pages) {
871 dev->max_guest_pages *= 2;
872 old_pages = dev->guest_pages;
873 dev->guest_pages = rte_realloc(dev->guest_pages,
874 dev->max_guest_pages * sizeof(*page),
875 RTE_CACHE_LINE_SIZE);
876 if (dev->guest_pages == NULL) {
877 VHOST_LOG_CONFIG(ERR, "cannot realloc guest_pages\n");
883 if (dev->nr_guest_pages > 0) {
884 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
885 /* merge if the two pages are continuous */
886 if (host_phys_addr == last_page->host_phys_addr +
888 last_page->size += size;
893 page = &dev->guest_pages[dev->nr_guest_pages++];
894 page->guest_phys_addr = guest_phys_addr;
895 page->host_phys_addr = host_phys_addr;
902 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
905 uint64_t reg_size = reg->size;
906 uint64_t host_user_addr = reg->host_user_addr;
907 uint64_t guest_phys_addr = reg->guest_phys_addr;
908 uint64_t host_phys_addr;
911 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
912 size = page_size - (guest_phys_addr & (page_size - 1));
913 size = RTE_MIN(size, reg_size);
915 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
918 host_user_addr += size;
919 guest_phys_addr += size;
922 while (reg_size > 0) {
923 size = RTE_MIN(reg_size, page_size);
924 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
926 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
930 host_user_addr += size;
931 guest_phys_addr += size;
935 /* sort guest page array if over binary search threshold */
936 if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
937 qsort((void *)dev->guest_pages, dev->nr_guest_pages,
938 sizeof(struct guest_page), guest_page_addrcmp);
944 #ifdef RTE_LIBRTE_VHOST_DEBUG
945 /* TODO: enable it only in debug mode? */
947 dump_guest_pages(struct virtio_net *dev)
950 struct guest_page *page;
952 for (i = 0; i < dev->nr_guest_pages; i++) {
953 page = &dev->guest_pages[i];
955 VHOST_LOG_CONFIG(INFO,
956 "guest physical page region %u\n"
957 "\t guest_phys_addr: %" PRIx64 "\n"
958 "\t host_phys_addr : %" PRIx64 "\n"
959 "\t size : %" PRIx64 "\n",
961 page->guest_phys_addr,
962 page->host_phys_addr,
967 #define dump_guest_pages(dev)
971 vhost_memory_changed(struct VhostUserMemory *new,
972 struct rte_vhost_memory *old)
976 if (new->nregions != old->nregions)
979 for (i = 0; i < new->nregions; ++i) {
980 VhostUserMemoryRegion *new_r = &new->regions[i];
981 struct rte_vhost_mem_region *old_r = &old->regions[i];
983 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
985 if (new_r->memory_size != old_r->size)
987 if (new_r->userspace_addr != old_r->guest_user_addr)
995 vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg,
998 struct virtio_net *dev = *pdev;
999 struct VhostUserMemory *memory = &msg->payload.memory;
1000 struct rte_vhost_mem_region *reg;
1003 uint64_t mmap_offset;
1009 if (validate_msg_fds(msg, memory->nregions) != 0)
1010 return RTE_VHOST_MSG_RESULT_ERR;
1012 if (memory->nregions > VHOST_MEMORY_MAX_NREGIONS) {
1013 VHOST_LOG_CONFIG(ERR,
1014 "too many memory regions (%u)\n", memory->nregions);
1015 return RTE_VHOST_MSG_RESULT_ERR;
1018 if (dev->mem && !vhost_memory_changed(memory, dev->mem)) {
1019 VHOST_LOG_CONFIG(INFO,
1020 "(%d) memory regions not changed\n", dev->vid);
1024 return RTE_VHOST_MSG_RESULT_OK;
1028 if (dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) {
1029 struct rte_vdpa_device *vdpa_dev = dev->vdpa_dev;
1031 if (vdpa_dev && vdpa_dev->ops->dev_close)
1032 vdpa_dev->ops->dev_close(dev->vid);
1033 dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1035 free_mem_region(dev);
1040 /* Flush IOTLB cache as previous HVAs are now invalid */
1041 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
1042 for (i = 0; i < dev->nr_vring; i++)
1043 vhost_user_iotlb_flush_all(dev->virtqueue[i]);
1045 dev->nr_guest_pages = 0;
1046 if (dev->guest_pages == NULL) {
1047 dev->max_guest_pages = 8;
1048 dev->guest_pages = rte_zmalloc(NULL,
1049 dev->max_guest_pages *
1050 sizeof(struct guest_page),
1051 RTE_CACHE_LINE_SIZE);
1052 if (dev->guest_pages == NULL) {
1053 VHOST_LOG_CONFIG(ERR,
1054 "(%d) failed to allocate memory "
1055 "for dev->guest_pages\n",
1057 return RTE_VHOST_MSG_RESULT_ERR;
1061 dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
1062 sizeof(struct rte_vhost_mem_region) * memory->nregions, 0);
1063 if (dev->mem == NULL) {
1064 VHOST_LOG_CONFIG(ERR,
1065 "(%d) failed to allocate memory for dev->mem\n",
1067 return RTE_VHOST_MSG_RESULT_ERR;
1069 dev->mem->nregions = memory->nregions;
1071 for (i = 0; i < memory->nregions; i++) {
1073 reg = &dev->mem->regions[i];
1075 reg->guest_phys_addr = memory->regions[i].guest_phys_addr;
1076 reg->guest_user_addr = memory->regions[i].userspace_addr;
1077 reg->size = memory->regions[i].memory_size;
1080 mmap_offset = memory->regions[i].mmap_offset;
1082 /* Check for memory_size + mmap_offset overflow */
1083 if (mmap_offset >= -reg->size) {
1084 VHOST_LOG_CONFIG(ERR,
1085 "mmap_offset (%#"PRIx64") and memory_size "
1086 "(%#"PRIx64") overflow\n",
1087 mmap_offset, reg->size);
1091 mmap_size = reg->size + mmap_offset;
1093 /* mmap() without flag of MAP_ANONYMOUS, should be called
1094 * with length argument aligned with hugepagesz at older
1095 * longterm version Linux, like 2.6.32 and 3.2.72, or
1096 * mmap() will fail with EINVAL.
1098 * to avoid failure, make sure in caller to keep length
1101 alignment = get_blk_size(fd);
1102 if (alignment == (uint64_t)-1) {
1103 VHOST_LOG_CONFIG(ERR,
1104 "couldn't get hugepage size through fstat\n");
1107 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
1108 if (mmap_size == 0) {
1110 * It could happen if initial mmap_size + alignment
1111 * overflows the sizeof uint64, which could happen if
1112 * either mmap_size or alignment value is wrong.
1114 * mmap() kernel implementation would return an error,
1115 * but better catch it before and provide useful info
1118 VHOST_LOG_CONFIG(ERR, "mmap size (0x%" PRIx64 ") "
1119 "or alignment (0x%" PRIx64 ") is invalid\n",
1120 reg->size + mmap_offset, alignment);
1124 populate = dev->async_copy ? MAP_POPULATE : 0;
1125 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
1126 MAP_SHARED | populate, fd, 0);
1128 if (mmap_addr == MAP_FAILED) {
1129 VHOST_LOG_CONFIG(ERR,
1130 "mmap region %u failed.\n", i);
1134 reg->mmap_addr = mmap_addr;
1135 reg->mmap_size = mmap_size;
1136 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
1139 if (dev->async_copy)
1140 if (add_guest_pages(dev, reg, alignment) < 0) {
1141 VHOST_LOG_CONFIG(ERR,
1142 "adding guest pages to region %u failed.\n",
1147 VHOST_LOG_CONFIG(INFO,
1148 "guest memory region %u, size: 0x%" PRIx64 "\n"
1149 "\t guest physical addr: 0x%" PRIx64 "\n"
1150 "\t guest virtual addr: 0x%" PRIx64 "\n"
1151 "\t host virtual addr: 0x%" PRIx64 "\n"
1152 "\t mmap addr : 0x%" PRIx64 "\n"
1153 "\t mmap size : 0x%" PRIx64 "\n"
1154 "\t mmap align: 0x%" PRIx64 "\n"
1155 "\t mmap off : 0x%" PRIx64 "\n",
1157 reg->guest_phys_addr,
1158 reg->guest_user_addr,
1159 reg->host_user_addr,
1160 (uint64_t)(uintptr_t)mmap_addr,
1165 if (dev->postcopy_listening) {
1167 * We haven't a better way right now than sharing
1168 * DPDK's virtual address with Qemu, so that Qemu can
1169 * retrieve the region offset when handling userfaults.
1171 memory->regions[i].userspace_addr =
1172 reg->host_user_addr;
1175 if (dev->postcopy_listening) {
1176 /* Send the addresses back to qemu */
1178 send_vhost_reply(main_fd, msg);
1180 /* Wait for qemu to acknolwedge it's got the addresses
1181 * we've got to wait before we're allowed to generate faults.
1183 VhostUserMsg ack_msg;
1184 if (read_vhost_message(main_fd, &ack_msg) <= 0) {
1185 VHOST_LOG_CONFIG(ERR,
1186 "Failed to read qemu ack on postcopy set-mem-table\n");
1190 if (validate_msg_fds(&ack_msg, 0) != 0)
1193 if (ack_msg.request.master != VHOST_USER_SET_MEM_TABLE) {
1194 VHOST_LOG_CONFIG(ERR,
1195 "Bad qemu ack on postcopy set-mem-table (%d)\n",
1196 ack_msg.request.master);
1200 /* Now userfault register and we can use the memory */
1201 for (i = 0; i < memory->nregions; i++) {
1202 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
1203 reg = &dev->mem->regions[i];
1204 struct uffdio_register reg_struct;
1207 * Let's register all the mmap'ed area to ensure
1208 * alignment on page boundary.
1210 reg_struct.range.start =
1211 (uint64_t)(uintptr_t)reg->mmap_addr;
1212 reg_struct.range.len = reg->mmap_size;
1213 reg_struct.mode = UFFDIO_REGISTER_MODE_MISSING;
1215 if (ioctl(dev->postcopy_ufd, UFFDIO_REGISTER,
1217 VHOST_LOG_CONFIG(ERR,
1218 "Failed to register ufd for region %d: (ufd = %d) %s\n",
1219 i, dev->postcopy_ufd,
1223 VHOST_LOG_CONFIG(INFO,
1224 "\t userfaultfd registered for range : "
1225 "%" PRIx64 " - %" PRIx64 "\n",
1226 (uint64_t)reg_struct.range.start,
1227 (uint64_t)reg_struct.range.start +
1228 (uint64_t)reg_struct.range.len - 1);
1235 for (i = 0; i < dev->nr_vring; i++) {
1236 struct vhost_virtqueue *vq = dev->virtqueue[i];
1238 if (vq->desc || vq->avail || vq->used) {
1240 * If the memory table got updated, the ring addresses
1241 * need to be translated again as virtual addresses have
1244 vring_invalidate(dev, vq);
1246 dev = translate_ring_addresses(dev, i);
1256 dump_guest_pages(dev);
1258 return RTE_VHOST_MSG_RESULT_OK;
1261 free_mem_region(dev);
1264 return RTE_VHOST_MSG_RESULT_ERR;
1268 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
1275 if (vq_is_packed(dev))
1276 rings_ok = vq->desc_packed && vq->driver_event &&
1279 rings_ok = vq->desc && vq->avail && vq->used;
1282 vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1283 vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1287 #define VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY 2u
1290 virtio_is_ready(struct virtio_net *dev)
1292 struct vhost_virtqueue *vq;
1293 uint32_t i, nr_vring = dev->nr_vring;
1295 if (dev->flags & VIRTIO_DEV_READY)
1301 if (dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) {
1302 nr_vring = VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY;
1304 if (dev->nr_vring < nr_vring)
1308 for (i = 0; i < nr_vring; i++) {
1309 vq = dev->virtqueue[i];
1311 if (!vq_is_ready(dev, vq))
1315 /* If supported, ensure the frontend is really done with config */
1316 if (dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_STATUS))
1317 if (!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK))
1320 dev->flags |= VIRTIO_DEV_READY;
1322 if (!(dev->flags & VIRTIO_DEV_RUNNING))
1323 VHOST_LOG_CONFIG(INFO,
1324 "virtio is now ready for processing.\n");
1329 inflight_mem_alloc(const char *name, size_t size, int *fd)
1333 char fname[20] = "/tmp/memfd-XXXXXX";
1336 #ifdef MEMFD_SUPPORTED
1337 mfd = memfd_create(name, MFD_CLOEXEC);
1342 mfd = mkstemp(fname);
1344 VHOST_LOG_CONFIG(ERR,
1345 "failed to get inflight buffer fd\n");
1352 if (ftruncate(mfd, size) == -1) {
1353 VHOST_LOG_CONFIG(ERR,
1354 "failed to alloc inflight buffer\n");
1359 ptr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, mfd, 0);
1360 if (ptr == MAP_FAILED) {
1361 VHOST_LOG_CONFIG(ERR,
1362 "failed to mmap inflight buffer\n");
1372 get_pervq_shm_size_split(uint16_t queue_size)
1374 return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_split) *
1375 queue_size + sizeof(uint64_t) +
1376 sizeof(uint16_t) * 4, INFLIGHT_ALIGNMENT);
1380 get_pervq_shm_size_packed(uint16_t queue_size)
1382 return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_packed)
1383 * queue_size + sizeof(uint64_t) +
1384 sizeof(uint16_t) * 6 + sizeof(uint8_t) * 9,
1385 INFLIGHT_ALIGNMENT);
1389 vhost_user_get_inflight_fd(struct virtio_net **pdev,
1391 int main_fd __rte_unused)
1393 struct rte_vhost_inflight_info_packed *inflight_packed;
1394 uint64_t pervq_inflight_size, mmap_size;
1395 uint16_t num_queues, queue_size;
1396 struct virtio_net *dev = *pdev;
1400 if (msg->size != sizeof(msg->payload.inflight)) {
1401 VHOST_LOG_CONFIG(ERR,
1402 "invalid get_inflight_fd message size is %d\n",
1404 return RTE_VHOST_MSG_RESULT_ERR;
1407 if (dev->inflight_info == NULL) {
1408 dev->inflight_info = calloc(1,
1409 sizeof(struct inflight_mem_info));
1410 if (!dev->inflight_info) {
1411 VHOST_LOG_CONFIG(ERR,
1412 "failed to alloc dev inflight area\n");
1413 return RTE_VHOST_MSG_RESULT_ERR;
1415 dev->inflight_info->fd = -1;
1418 num_queues = msg->payload.inflight.num_queues;
1419 queue_size = msg->payload.inflight.queue_size;
1421 VHOST_LOG_CONFIG(INFO, "get_inflight_fd num_queues: %u\n",
1422 msg->payload.inflight.num_queues);
1423 VHOST_LOG_CONFIG(INFO, "get_inflight_fd queue_size: %u\n",
1424 msg->payload.inflight.queue_size);
1426 if (vq_is_packed(dev))
1427 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1429 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1431 mmap_size = num_queues * pervq_inflight_size;
1432 addr = inflight_mem_alloc("vhost-inflight", mmap_size, &fd);
1434 VHOST_LOG_CONFIG(ERR,
1435 "failed to alloc vhost inflight area\n");
1436 msg->payload.inflight.mmap_size = 0;
1437 return RTE_VHOST_MSG_RESULT_ERR;
1439 memset(addr, 0, mmap_size);
1441 if (dev->inflight_info->addr) {
1442 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1443 dev->inflight_info->addr = NULL;
1446 if (dev->inflight_info->fd >= 0) {
1447 close(dev->inflight_info->fd);
1448 dev->inflight_info->fd = -1;
1451 dev->inflight_info->addr = addr;
1452 dev->inflight_info->size = msg->payload.inflight.mmap_size = mmap_size;
1453 dev->inflight_info->fd = msg->fds[0] = fd;
1454 msg->payload.inflight.mmap_offset = 0;
1457 if (vq_is_packed(dev)) {
1458 for (i = 0; i < num_queues; i++) {
1460 (struct rte_vhost_inflight_info_packed *)addr;
1461 inflight_packed->used_wrap_counter = 1;
1462 inflight_packed->old_used_wrap_counter = 1;
1463 for (j = 0; j < queue_size; j++)
1464 inflight_packed->desc[j].next = j + 1;
1465 addr = (void *)((char *)addr + pervq_inflight_size);
1469 VHOST_LOG_CONFIG(INFO,
1470 "send inflight mmap_size: %"PRIu64"\n",
1471 msg->payload.inflight.mmap_size);
1472 VHOST_LOG_CONFIG(INFO,
1473 "send inflight mmap_offset: %"PRIu64"\n",
1474 msg->payload.inflight.mmap_offset);
1475 VHOST_LOG_CONFIG(INFO,
1476 "send inflight fd: %d\n", msg->fds[0]);
1478 return RTE_VHOST_MSG_RESULT_REPLY;
1482 vhost_user_set_inflight_fd(struct virtio_net **pdev, VhostUserMsg *msg,
1483 int main_fd __rte_unused)
1485 uint64_t mmap_size, mmap_offset;
1486 uint16_t num_queues, queue_size;
1487 struct virtio_net *dev = *pdev;
1488 uint32_t pervq_inflight_size;
1489 struct vhost_virtqueue *vq;
1494 if (msg->size != sizeof(msg->payload.inflight) || fd < 0) {
1495 VHOST_LOG_CONFIG(ERR,
1496 "invalid set_inflight_fd message size is %d,fd is %d\n",
1498 return RTE_VHOST_MSG_RESULT_ERR;
1501 mmap_size = msg->payload.inflight.mmap_size;
1502 mmap_offset = msg->payload.inflight.mmap_offset;
1503 num_queues = msg->payload.inflight.num_queues;
1504 queue_size = msg->payload.inflight.queue_size;
1506 if (vq_is_packed(dev))
1507 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1509 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1511 VHOST_LOG_CONFIG(INFO,
1512 "set_inflight_fd mmap_size: %"PRIu64"\n", mmap_size);
1513 VHOST_LOG_CONFIG(INFO,
1514 "set_inflight_fd mmap_offset: %"PRIu64"\n", mmap_offset);
1515 VHOST_LOG_CONFIG(INFO,
1516 "set_inflight_fd num_queues: %u\n", num_queues);
1517 VHOST_LOG_CONFIG(INFO,
1518 "set_inflight_fd queue_size: %u\n", queue_size);
1519 VHOST_LOG_CONFIG(INFO,
1520 "set_inflight_fd fd: %d\n", fd);
1521 VHOST_LOG_CONFIG(INFO,
1522 "set_inflight_fd pervq_inflight_size: %d\n",
1523 pervq_inflight_size);
1525 if (!dev->inflight_info) {
1526 dev->inflight_info = calloc(1,
1527 sizeof(struct inflight_mem_info));
1528 if (dev->inflight_info == NULL) {
1529 VHOST_LOG_CONFIG(ERR,
1530 "failed to alloc dev inflight area\n");
1531 return RTE_VHOST_MSG_RESULT_ERR;
1533 dev->inflight_info->fd = -1;
1536 if (dev->inflight_info->addr) {
1537 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1538 dev->inflight_info->addr = NULL;
1541 addr = mmap(0, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
1543 if (addr == MAP_FAILED) {
1544 VHOST_LOG_CONFIG(ERR, "failed to mmap share memory.\n");
1545 return RTE_VHOST_MSG_RESULT_ERR;
1548 if (dev->inflight_info->fd >= 0) {
1549 close(dev->inflight_info->fd);
1550 dev->inflight_info->fd = -1;
1553 dev->inflight_info->fd = fd;
1554 dev->inflight_info->addr = addr;
1555 dev->inflight_info->size = mmap_size;
1557 for (i = 0; i < num_queues; i++) {
1558 vq = dev->virtqueue[i];
1559 if (vq_is_packed(dev)) {
1560 vq->inflight_packed = addr;
1561 vq->inflight_packed->desc_num = queue_size;
1563 vq->inflight_split = addr;
1564 vq->inflight_split->desc_num = queue_size;
1566 addr = (void *)((char *)addr + pervq_inflight_size);
1569 return RTE_VHOST_MSG_RESULT_OK;
1573 vhost_user_set_vring_call(struct virtio_net **pdev, struct VhostUserMsg *msg,
1574 int main_fd __rte_unused)
1576 struct virtio_net *dev = *pdev;
1577 struct vhost_vring_file file;
1578 struct vhost_virtqueue *vq;
1581 expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1582 if (validate_msg_fds(msg, expected_fds) != 0)
1583 return RTE_VHOST_MSG_RESULT_ERR;
1585 file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1586 if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1587 file.fd = VIRTIO_INVALID_EVENTFD;
1589 file.fd = msg->fds[0];
1590 VHOST_LOG_CONFIG(INFO,
1591 "vring call idx:%d file:%d\n", file.index, file.fd);
1593 vq = dev->virtqueue[file.index];
1597 vhost_user_notify_queue_state(dev, file.index, 0);
1600 if (vq->callfd >= 0)
1603 vq->callfd = file.fd;
1605 return RTE_VHOST_MSG_RESULT_OK;
1608 static int vhost_user_set_vring_err(struct virtio_net **pdev __rte_unused,
1609 struct VhostUserMsg *msg,
1610 int main_fd __rte_unused)
1614 expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1615 if (validate_msg_fds(msg, expected_fds) != 0)
1616 return RTE_VHOST_MSG_RESULT_ERR;
1618 if (!(msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1620 VHOST_LOG_CONFIG(INFO, "not implemented\n");
1622 return RTE_VHOST_MSG_RESULT_OK;
1626 resubmit_desc_compare(const void *a, const void *b)
1628 const struct rte_vhost_resubmit_desc *desc0 = a;
1629 const struct rte_vhost_resubmit_desc *desc1 = b;
1631 if (desc1->counter > desc0->counter)
1638 vhost_check_queue_inflights_split(struct virtio_net *dev,
1639 struct vhost_virtqueue *vq)
1642 uint16_t resubmit_num = 0, last_io, num;
1643 struct vring_used *used = vq->used;
1644 struct rte_vhost_resubmit_info *resubmit;
1645 struct rte_vhost_inflight_info_split *inflight_split;
1647 if (!(dev->protocol_features &
1648 (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1649 return RTE_VHOST_MSG_RESULT_OK;
1651 /* The frontend may still not support the inflight feature
1652 * although we negotiate the protocol feature.
1654 if ((!vq->inflight_split))
1655 return RTE_VHOST_MSG_RESULT_OK;
1657 if (!vq->inflight_split->version) {
1658 vq->inflight_split->version = INFLIGHT_VERSION;
1659 return RTE_VHOST_MSG_RESULT_OK;
1662 if (vq->resubmit_inflight)
1663 return RTE_VHOST_MSG_RESULT_OK;
1665 inflight_split = vq->inflight_split;
1666 vq->global_counter = 0;
1667 last_io = inflight_split->last_inflight_io;
1669 if (inflight_split->used_idx != used->idx) {
1670 inflight_split->desc[last_io].inflight = 0;
1672 inflight_split->used_idx = used->idx;
1675 for (i = 0; i < inflight_split->desc_num; i++) {
1676 if (inflight_split->desc[i].inflight == 1)
1680 vq->last_avail_idx += resubmit_num;
1683 resubmit = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1685 VHOST_LOG_CONFIG(ERR,
1686 "failed to allocate memory for resubmit info.\n");
1687 return RTE_VHOST_MSG_RESULT_ERR;
1690 resubmit->resubmit_list = calloc(resubmit_num,
1691 sizeof(struct rte_vhost_resubmit_desc));
1692 if (!resubmit->resubmit_list) {
1693 VHOST_LOG_CONFIG(ERR,
1694 "failed to allocate memory for inflight desc.\n");
1696 return RTE_VHOST_MSG_RESULT_ERR;
1700 for (i = 0; i < vq->inflight_split->desc_num; i++) {
1701 if (vq->inflight_split->desc[i].inflight == 1) {
1702 resubmit->resubmit_list[num].index = i;
1703 resubmit->resubmit_list[num].counter =
1704 inflight_split->desc[i].counter;
1708 resubmit->resubmit_num = num;
1710 if (resubmit->resubmit_num > 1)
1711 qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1712 sizeof(struct rte_vhost_resubmit_desc),
1713 resubmit_desc_compare);
1715 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1716 vq->resubmit_inflight = resubmit;
1719 return RTE_VHOST_MSG_RESULT_OK;
1723 vhost_check_queue_inflights_packed(struct virtio_net *dev,
1724 struct vhost_virtqueue *vq)
1727 uint16_t resubmit_num = 0, old_used_idx, num;
1728 struct rte_vhost_resubmit_info *resubmit;
1729 struct rte_vhost_inflight_info_packed *inflight_packed;
1731 if (!(dev->protocol_features &
1732 (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1733 return RTE_VHOST_MSG_RESULT_OK;
1735 /* The frontend may still not support the inflight feature
1736 * although we negotiate the protocol feature.
1738 if ((!vq->inflight_packed))
1739 return RTE_VHOST_MSG_RESULT_OK;
1741 if (!vq->inflight_packed->version) {
1742 vq->inflight_packed->version = INFLIGHT_VERSION;
1743 return RTE_VHOST_MSG_RESULT_OK;
1746 if (vq->resubmit_inflight)
1747 return RTE_VHOST_MSG_RESULT_OK;
1749 inflight_packed = vq->inflight_packed;
1750 vq->global_counter = 0;
1751 old_used_idx = inflight_packed->old_used_idx;
1753 if (inflight_packed->used_idx != old_used_idx) {
1754 if (inflight_packed->desc[old_used_idx].inflight == 0) {
1755 inflight_packed->old_used_idx =
1756 inflight_packed->used_idx;
1757 inflight_packed->old_used_wrap_counter =
1758 inflight_packed->used_wrap_counter;
1759 inflight_packed->old_free_head =
1760 inflight_packed->free_head;
1762 inflight_packed->used_idx =
1763 inflight_packed->old_used_idx;
1764 inflight_packed->used_wrap_counter =
1765 inflight_packed->old_used_wrap_counter;
1766 inflight_packed->free_head =
1767 inflight_packed->old_free_head;
1771 for (i = 0; i < inflight_packed->desc_num; i++) {
1772 if (inflight_packed->desc[i].inflight == 1)
1777 resubmit = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1778 if (resubmit == NULL) {
1779 VHOST_LOG_CONFIG(ERR,
1780 "failed to allocate memory for resubmit info.\n");
1781 return RTE_VHOST_MSG_RESULT_ERR;
1784 resubmit->resubmit_list = calloc(resubmit_num,
1785 sizeof(struct rte_vhost_resubmit_desc));
1786 if (resubmit->resubmit_list == NULL) {
1787 VHOST_LOG_CONFIG(ERR,
1788 "failed to allocate memory for resubmit desc.\n");
1790 return RTE_VHOST_MSG_RESULT_ERR;
1794 for (i = 0; i < inflight_packed->desc_num; i++) {
1795 if (vq->inflight_packed->desc[i].inflight == 1) {
1796 resubmit->resubmit_list[num].index = i;
1797 resubmit->resubmit_list[num].counter =
1798 inflight_packed->desc[i].counter;
1802 resubmit->resubmit_num = num;
1804 if (resubmit->resubmit_num > 1)
1805 qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1806 sizeof(struct rte_vhost_resubmit_desc),
1807 resubmit_desc_compare);
1809 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1810 vq->resubmit_inflight = resubmit;
1813 return RTE_VHOST_MSG_RESULT_OK;
1817 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
1818 int main_fd __rte_unused)
1820 struct virtio_net *dev = *pdev;
1821 struct vhost_vring_file file;
1822 struct vhost_virtqueue *vq;
1825 expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1826 if (validate_msg_fds(msg, expected_fds) != 0)
1827 return RTE_VHOST_MSG_RESULT_ERR;
1829 file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1830 if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1831 file.fd = VIRTIO_INVALID_EVENTFD;
1833 file.fd = msg->fds[0];
1834 VHOST_LOG_CONFIG(INFO,
1835 "vring kick idx:%d file:%d\n", file.index, file.fd);
1837 /* Interpret ring addresses only when ring is started. */
1838 dev = translate_ring_addresses(dev, file.index);
1840 return RTE_VHOST_MSG_RESULT_ERR;
1844 vq = dev->virtqueue[file.index];
1847 * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
1848 * the ring starts already enabled. Otherwise, it is enabled via
1849 * the SET_VRING_ENABLE message.
1851 if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
1853 if (dev->notify_ops->vring_state_changed)
1854 dev->notify_ops->vring_state_changed(
1855 dev->vid, file.index, 1);
1860 vhost_user_notify_queue_state(dev, file.index, 0);
1863 if (vq->kickfd >= 0)
1865 vq->kickfd = file.fd;
1867 if (vq_is_packed(dev)) {
1868 if (vhost_check_queue_inflights_packed(dev, vq)) {
1869 VHOST_LOG_CONFIG(ERR,
1870 "failed to inflights for vq: %d\n", file.index);
1871 return RTE_VHOST_MSG_RESULT_ERR;
1874 if (vhost_check_queue_inflights_split(dev, vq)) {
1875 VHOST_LOG_CONFIG(ERR,
1876 "failed to inflights for vq: %d\n", file.index);
1877 return RTE_VHOST_MSG_RESULT_ERR;
1881 return RTE_VHOST_MSG_RESULT_OK;
1885 * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
1888 vhost_user_get_vring_base(struct virtio_net **pdev,
1889 struct VhostUserMsg *msg,
1890 int main_fd __rte_unused)
1892 struct virtio_net *dev = *pdev;
1893 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
1896 if (validate_msg_fds(msg, 0) != 0)
1897 return RTE_VHOST_MSG_RESULT_ERR;
1899 /* We have to stop the queue (virtio) if it is running. */
1900 vhost_destroy_device_notify(dev);
1902 dev->flags &= ~VIRTIO_DEV_READY;
1903 dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1905 /* Here we are safe to get the indexes */
1906 if (vq_is_packed(dev)) {
1908 * Bit[0:14]: avail index
1909 * Bit[15]: avail wrap counter
1911 val = vq->last_avail_idx & 0x7fff;
1912 val |= vq->avail_wrap_counter << 15;
1913 msg->payload.state.num = val;
1915 msg->payload.state.num = vq->last_avail_idx;
1918 VHOST_LOG_CONFIG(INFO,
1919 "vring base idx:%d file:%d\n", msg->payload.state.index,
1920 msg->payload.state.num);
1922 * Based on current qemu vhost-user implementation, this message is
1923 * sent and only sent in vhost_vring_stop.
1924 * TODO: cleanup the vring, it isn't usable since here.
1926 if (vq->kickfd >= 0)
1929 vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
1931 if (vq->callfd >= 0)
1934 vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
1936 vq->signalled_used_valid = false;
1938 if (vq_is_packed(dev)) {
1939 rte_free(vq->shadow_used_packed);
1940 vq->shadow_used_packed = NULL;
1942 rte_free(vq->shadow_used_split);
1943 vq->shadow_used_split = NULL;
1944 if (vq->async_pkts_pending)
1945 rte_free(vq->async_pkts_pending);
1946 if (vq->async_pkts_info)
1947 rte_free(vq->async_pkts_info);
1948 vq->async_pkts_pending = NULL;
1949 vq->async_pkts_info = NULL;
1952 rte_free(vq->batch_copy_elems);
1953 vq->batch_copy_elems = NULL;
1955 msg->size = sizeof(msg->payload.state);
1958 vring_invalidate(dev, vq);
1960 return RTE_VHOST_MSG_RESULT_REPLY;
1964 * when virtio queues are ready to work, qemu will send us to
1965 * enable the virtio queue pair.
1968 vhost_user_set_vring_enable(struct virtio_net **pdev,
1969 struct VhostUserMsg *msg,
1970 int main_fd __rte_unused)
1972 struct virtio_net *dev = *pdev;
1973 int enable = (int)msg->payload.state.num;
1974 int index = (int)msg->payload.state.index;
1976 if (validate_msg_fds(msg, 0) != 0)
1977 return RTE_VHOST_MSG_RESULT_ERR;
1979 VHOST_LOG_CONFIG(INFO,
1980 "set queue enable: %d to qp idx: %d\n",
1983 if (enable && dev->virtqueue[index]->async_registered) {
1984 if (dev->virtqueue[index]->async_pkts_inflight_n) {
1985 VHOST_LOG_CONFIG(ERR, "failed to enable vring. "
1986 "async inflight packets must be completed first\n");
1987 return RTE_VHOST_MSG_RESULT_ERR;
1991 dev->virtqueue[index]->enabled = enable;
1993 return RTE_VHOST_MSG_RESULT_OK;
1997 vhost_user_get_protocol_features(struct virtio_net **pdev,
1998 struct VhostUserMsg *msg,
1999 int main_fd __rte_unused)
2001 struct virtio_net *dev = *pdev;
2002 uint64_t features, protocol_features;
2004 if (validate_msg_fds(msg, 0) != 0)
2005 return RTE_VHOST_MSG_RESULT_ERR;
2007 rte_vhost_driver_get_features(dev->ifname, &features);
2008 rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
2010 msg->payload.u64 = protocol_features;
2011 msg->size = sizeof(msg->payload.u64);
2014 return RTE_VHOST_MSG_RESULT_REPLY;
2018 vhost_user_set_protocol_features(struct virtio_net **pdev,
2019 struct VhostUserMsg *msg,
2020 int main_fd __rte_unused)
2022 struct virtio_net *dev = *pdev;
2023 uint64_t protocol_features = msg->payload.u64;
2024 uint64_t slave_protocol_features = 0;
2026 if (validate_msg_fds(msg, 0) != 0)
2027 return RTE_VHOST_MSG_RESULT_ERR;
2029 rte_vhost_driver_get_protocol_features(dev->ifname,
2030 &slave_protocol_features);
2031 if (protocol_features & ~slave_protocol_features) {
2032 VHOST_LOG_CONFIG(ERR,
2033 "(%d) received invalid protocol features.\n",
2035 return RTE_VHOST_MSG_RESULT_ERR;
2038 dev->protocol_features = protocol_features;
2039 VHOST_LOG_CONFIG(INFO,
2040 "negotiated Vhost-user protocol features: 0x%" PRIx64 "\n",
2041 dev->protocol_features);
2043 return RTE_VHOST_MSG_RESULT_OK;
2047 vhost_user_set_log_base(struct virtio_net **pdev, struct VhostUserMsg *msg,
2048 int main_fd __rte_unused)
2050 struct virtio_net *dev = *pdev;
2051 int fd = msg->fds[0];
2055 if (validate_msg_fds(msg, 1) != 0)
2056 return RTE_VHOST_MSG_RESULT_ERR;
2059 VHOST_LOG_CONFIG(ERR, "invalid log fd: %d\n", fd);
2060 return RTE_VHOST_MSG_RESULT_ERR;
2063 if (msg->size != sizeof(VhostUserLog)) {
2064 VHOST_LOG_CONFIG(ERR,
2065 "invalid log base msg size: %"PRId32" != %d\n",
2066 msg->size, (int)sizeof(VhostUserLog));
2067 return RTE_VHOST_MSG_RESULT_ERR;
2070 size = msg->payload.log.mmap_size;
2071 off = msg->payload.log.mmap_offset;
2073 /* Check for mmap size and offset overflow. */
2075 VHOST_LOG_CONFIG(ERR,
2076 "log offset %#"PRIx64" and log size %#"PRIx64" overflow\n",
2078 return RTE_VHOST_MSG_RESULT_ERR;
2081 VHOST_LOG_CONFIG(INFO,
2082 "log mmap size: %"PRId64", offset: %"PRId64"\n",
2086 * mmap from 0 to workaround a hugepage mmap bug: mmap will
2087 * fail when offset is not page size aligned.
2089 addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
2091 if (addr == MAP_FAILED) {
2092 VHOST_LOG_CONFIG(ERR, "mmap log base failed!\n");
2093 return RTE_VHOST_MSG_RESULT_ERR;
2097 * Free previously mapped log memory on occasionally
2098 * multiple VHOST_USER_SET_LOG_BASE.
2100 if (dev->log_addr) {
2101 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
2103 dev->log_addr = (uint64_t)(uintptr_t)addr;
2104 dev->log_base = dev->log_addr + off;
2105 dev->log_size = size;
2108 * The spec is not clear about it (yet), but QEMU doesn't expect
2109 * any payload in the reply.
2114 return RTE_VHOST_MSG_RESULT_REPLY;
2117 static int vhost_user_set_log_fd(struct virtio_net **pdev __rte_unused,
2118 struct VhostUserMsg *msg,
2119 int main_fd __rte_unused)
2121 if (validate_msg_fds(msg, 1) != 0)
2122 return RTE_VHOST_MSG_RESULT_ERR;
2125 VHOST_LOG_CONFIG(INFO, "not implemented.\n");
2127 return RTE_VHOST_MSG_RESULT_OK;
2131 * An rarp packet is constructed and broadcasted to notify switches about
2132 * the new location of the migrated VM, so that packets from outside will
2133 * not be lost after migration.
2135 * However, we don't actually "send" a rarp packet here, instead, we set
2136 * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
2139 vhost_user_send_rarp(struct virtio_net **pdev, struct VhostUserMsg *msg,
2140 int main_fd __rte_unused)
2142 struct virtio_net *dev = *pdev;
2143 uint8_t *mac = (uint8_t *)&msg->payload.u64;
2144 struct rte_vdpa_device *vdpa_dev;
2146 if (validate_msg_fds(msg, 0) != 0)
2147 return RTE_VHOST_MSG_RESULT_ERR;
2149 VHOST_LOG_CONFIG(DEBUG,
2150 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
2151 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
2152 memcpy(dev->mac.addr_bytes, mac, 6);
2155 * Set the flag to inject a RARP broadcast packet at
2156 * rte_vhost_dequeue_burst().
2158 * __ATOMIC_RELEASE ordering is for making sure the mac is
2159 * copied before the flag is set.
2161 __atomic_store_n(&dev->broadcast_rarp, 1, __ATOMIC_RELEASE);
2162 vdpa_dev = dev->vdpa_dev;
2163 if (vdpa_dev && vdpa_dev->ops->migration_done)
2164 vdpa_dev->ops->migration_done(dev->vid);
2166 return RTE_VHOST_MSG_RESULT_OK;
2170 vhost_user_net_set_mtu(struct virtio_net **pdev, struct VhostUserMsg *msg,
2171 int main_fd __rte_unused)
2173 struct virtio_net *dev = *pdev;
2175 if (validate_msg_fds(msg, 0) != 0)
2176 return RTE_VHOST_MSG_RESULT_ERR;
2178 if (msg->payload.u64 < VIRTIO_MIN_MTU ||
2179 msg->payload.u64 > VIRTIO_MAX_MTU) {
2180 VHOST_LOG_CONFIG(ERR, "Invalid MTU size (%"PRIu64")\n",
2183 return RTE_VHOST_MSG_RESULT_ERR;
2186 dev->mtu = msg->payload.u64;
2188 return RTE_VHOST_MSG_RESULT_OK;
2192 vhost_user_set_req_fd(struct virtio_net **pdev, struct VhostUserMsg *msg,
2193 int main_fd __rte_unused)
2195 struct virtio_net *dev = *pdev;
2196 int fd = msg->fds[0];
2198 if (validate_msg_fds(msg, 1) != 0)
2199 return RTE_VHOST_MSG_RESULT_ERR;
2202 VHOST_LOG_CONFIG(ERR,
2203 "Invalid file descriptor for slave channel (%d)\n",
2205 return RTE_VHOST_MSG_RESULT_ERR;
2208 if (dev->slave_req_fd >= 0)
2209 close(dev->slave_req_fd);
2211 dev->slave_req_fd = fd;
2213 return RTE_VHOST_MSG_RESULT_OK;
2217 is_vring_iotlb_split(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2219 struct vhost_vring_addr *ra;
2220 uint64_t start, end, len;
2223 end = start + imsg->size;
2225 ra = &vq->ring_addrs;
2226 len = sizeof(struct vring_desc) * vq->size;
2227 if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2230 len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
2231 if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2234 len = sizeof(struct vring_used) +
2235 sizeof(struct vring_used_elem) * vq->size;
2236 if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2239 if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2240 len = sizeof(uint64_t);
2241 if (ra->log_guest_addr < end &&
2242 (ra->log_guest_addr + len) > start)
2250 is_vring_iotlb_packed(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2252 struct vhost_vring_addr *ra;
2253 uint64_t start, end, len;
2256 end = start + imsg->size;
2258 ra = &vq->ring_addrs;
2259 len = sizeof(struct vring_packed_desc) * vq->size;
2260 if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2263 len = sizeof(struct vring_packed_desc_event);
2264 if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2267 len = sizeof(struct vring_packed_desc_event);
2268 if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2271 if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2272 len = sizeof(uint64_t);
2273 if (ra->log_guest_addr < end &&
2274 (ra->log_guest_addr + len) > start)
2281 static int is_vring_iotlb(struct virtio_net *dev,
2282 struct vhost_virtqueue *vq,
2283 struct vhost_iotlb_msg *imsg)
2285 if (vq_is_packed(dev))
2286 return is_vring_iotlb_packed(vq, imsg);
2288 return is_vring_iotlb_split(vq, imsg);
2292 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg,
2293 int main_fd __rte_unused)
2295 struct virtio_net *dev = *pdev;
2296 struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
2300 if (validate_msg_fds(msg, 0) != 0)
2301 return RTE_VHOST_MSG_RESULT_ERR;
2303 switch (imsg->type) {
2304 case VHOST_IOTLB_UPDATE:
2306 vva = qva_to_vva(dev, imsg->uaddr, &len);
2308 return RTE_VHOST_MSG_RESULT_ERR;
2310 for (i = 0; i < dev->nr_vring; i++) {
2311 struct vhost_virtqueue *vq = dev->virtqueue[i];
2313 vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
2316 if (is_vring_iotlb(dev, vq, imsg))
2317 *pdev = dev = translate_ring_addresses(dev, i);
2320 case VHOST_IOTLB_INVALIDATE:
2321 for (i = 0; i < dev->nr_vring; i++) {
2322 struct vhost_virtqueue *vq = dev->virtqueue[i];
2324 vhost_user_iotlb_cache_remove(vq, imsg->iova,
2327 if (is_vring_iotlb(dev, vq, imsg))
2328 vring_invalidate(dev, vq);
2332 VHOST_LOG_CONFIG(ERR, "Invalid IOTLB message type (%d)\n",
2334 return RTE_VHOST_MSG_RESULT_ERR;
2337 return RTE_VHOST_MSG_RESULT_OK;
2341 vhost_user_set_postcopy_advise(struct virtio_net **pdev,
2342 struct VhostUserMsg *msg,
2343 int main_fd __rte_unused)
2345 struct virtio_net *dev = *pdev;
2346 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
2347 struct uffdio_api api_struct;
2349 if (validate_msg_fds(msg, 0) != 0)
2350 return RTE_VHOST_MSG_RESULT_ERR;
2352 dev->postcopy_ufd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
2354 if (dev->postcopy_ufd == -1) {
2355 VHOST_LOG_CONFIG(ERR, "Userfaultfd not available: %s\n",
2357 return RTE_VHOST_MSG_RESULT_ERR;
2359 api_struct.api = UFFD_API;
2360 api_struct.features = 0;
2361 if (ioctl(dev->postcopy_ufd, UFFDIO_API, &api_struct)) {
2362 VHOST_LOG_CONFIG(ERR, "UFFDIO_API ioctl failure: %s\n",
2364 close(dev->postcopy_ufd);
2365 dev->postcopy_ufd = -1;
2366 return RTE_VHOST_MSG_RESULT_ERR;
2368 msg->fds[0] = dev->postcopy_ufd;
2371 return RTE_VHOST_MSG_RESULT_REPLY;
2373 dev->postcopy_ufd = -1;
2376 return RTE_VHOST_MSG_RESULT_ERR;
2381 vhost_user_set_postcopy_listen(struct virtio_net **pdev,
2382 struct VhostUserMsg *msg __rte_unused,
2383 int main_fd __rte_unused)
2385 struct virtio_net *dev = *pdev;
2387 if (validate_msg_fds(msg, 0) != 0)
2388 return RTE_VHOST_MSG_RESULT_ERR;
2390 if (dev->mem && dev->mem->nregions) {
2391 VHOST_LOG_CONFIG(ERR,
2392 "Regions already registered at postcopy-listen\n");
2393 return RTE_VHOST_MSG_RESULT_ERR;
2395 dev->postcopy_listening = 1;
2397 return RTE_VHOST_MSG_RESULT_OK;
2401 vhost_user_postcopy_end(struct virtio_net **pdev, struct VhostUserMsg *msg,
2402 int main_fd __rte_unused)
2404 struct virtio_net *dev = *pdev;
2406 if (validate_msg_fds(msg, 0) != 0)
2407 return RTE_VHOST_MSG_RESULT_ERR;
2409 dev->postcopy_listening = 0;
2410 if (dev->postcopy_ufd >= 0) {
2411 close(dev->postcopy_ufd);
2412 dev->postcopy_ufd = -1;
2415 msg->payload.u64 = 0;
2416 msg->size = sizeof(msg->payload.u64);
2419 return RTE_VHOST_MSG_RESULT_REPLY;
2423 vhost_user_get_status(struct virtio_net **pdev, struct VhostUserMsg *msg,
2424 int main_fd __rte_unused)
2426 struct virtio_net *dev = *pdev;
2428 if (validate_msg_fds(msg, 0) != 0)
2429 return RTE_VHOST_MSG_RESULT_ERR;
2431 msg->payload.u64 = dev->status;
2432 msg->size = sizeof(msg->payload.u64);
2435 return RTE_VHOST_MSG_RESULT_REPLY;
2439 vhost_user_set_status(struct virtio_net **pdev, struct VhostUserMsg *msg,
2440 int main_fd __rte_unused)
2442 struct virtio_net *dev = *pdev;
2444 if (validate_msg_fds(msg, 0) != 0)
2445 return RTE_VHOST_MSG_RESULT_ERR;
2447 /* As per Virtio specification, the device status is 8bits long */
2448 if (msg->payload.u64 > UINT8_MAX) {
2449 VHOST_LOG_CONFIG(ERR, "Invalid VHOST_USER_SET_STATUS payload 0x%" PRIx64 "\n",
2451 return RTE_VHOST_MSG_RESULT_ERR;
2454 dev->status = msg->payload.u64;
2456 if ((dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK) &&
2457 (dev->flags & VIRTIO_DEV_FEATURES_FAILED)) {
2458 VHOST_LOG_CONFIG(ERR, "FEATURES_OK bit is set but feature negotiation failed\n");
2460 * Clear the bit to let the driver know about the feature
2461 * negotiation failure
2463 dev->status &= ~VIRTIO_DEVICE_STATUS_FEATURES_OK;
2466 VHOST_LOG_CONFIG(INFO, "New device status(0x%08x):\n"
2468 "\t-ACKNOWLEDGE: %u\n"
2470 "\t-FEATURES_OK: %u\n"
2471 "\t-DRIVER_OK: %u\n"
2472 "\t-DEVICE_NEED_RESET: %u\n"
2475 (dev->status == VIRTIO_DEVICE_STATUS_RESET),
2476 !!(dev->status & VIRTIO_DEVICE_STATUS_ACK),
2477 !!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER),
2478 !!(dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK),
2479 !!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK),
2480 !!(dev->status & VIRTIO_DEVICE_STATUS_DEV_NEED_RESET),
2481 !!(dev->status & VIRTIO_DEVICE_STATUS_FAILED));
2483 return RTE_VHOST_MSG_RESULT_OK;
2486 typedef int (*vhost_message_handler_t)(struct virtio_net **pdev,
2487 struct VhostUserMsg *msg,
2489 static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
2490 [VHOST_USER_NONE] = NULL,
2491 [VHOST_USER_GET_FEATURES] = vhost_user_get_features,
2492 [VHOST_USER_SET_FEATURES] = vhost_user_set_features,
2493 [VHOST_USER_SET_OWNER] = vhost_user_set_owner,
2494 [VHOST_USER_RESET_OWNER] = vhost_user_reset_owner,
2495 [VHOST_USER_SET_MEM_TABLE] = vhost_user_set_mem_table,
2496 [VHOST_USER_SET_LOG_BASE] = vhost_user_set_log_base,
2497 [VHOST_USER_SET_LOG_FD] = vhost_user_set_log_fd,
2498 [VHOST_USER_SET_VRING_NUM] = vhost_user_set_vring_num,
2499 [VHOST_USER_SET_VRING_ADDR] = vhost_user_set_vring_addr,
2500 [VHOST_USER_SET_VRING_BASE] = vhost_user_set_vring_base,
2501 [VHOST_USER_GET_VRING_BASE] = vhost_user_get_vring_base,
2502 [VHOST_USER_SET_VRING_KICK] = vhost_user_set_vring_kick,
2503 [VHOST_USER_SET_VRING_CALL] = vhost_user_set_vring_call,
2504 [VHOST_USER_SET_VRING_ERR] = vhost_user_set_vring_err,
2505 [VHOST_USER_GET_PROTOCOL_FEATURES] = vhost_user_get_protocol_features,
2506 [VHOST_USER_SET_PROTOCOL_FEATURES] = vhost_user_set_protocol_features,
2507 [VHOST_USER_GET_QUEUE_NUM] = vhost_user_get_queue_num,
2508 [VHOST_USER_SET_VRING_ENABLE] = vhost_user_set_vring_enable,
2509 [VHOST_USER_SEND_RARP] = vhost_user_send_rarp,
2510 [VHOST_USER_NET_SET_MTU] = vhost_user_net_set_mtu,
2511 [VHOST_USER_SET_SLAVE_REQ_FD] = vhost_user_set_req_fd,
2512 [VHOST_USER_IOTLB_MSG] = vhost_user_iotlb_msg,
2513 [VHOST_USER_POSTCOPY_ADVISE] = vhost_user_set_postcopy_advise,
2514 [VHOST_USER_POSTCOPY_LISTEN] = vhost_user_set_postcopy_listen,
2515 [VHOST_USER_POSTCOPY_END] = vhost_user_postcopy_end,
2516 [VHOST_USER_GET_INFLIGHT_FD] = vhost_user_get_inflight_fd,
2517 [VHOST_USER_SET_INFLIGHT_FD] = vhost_user_set_inflight_fd,
2518 [VHOST_USER_SET_STATUS] = vhost_user_set_status,
2519 [VHOST_USER_GET_STATUS] = vhost_user_get_status,
2522 /* return bytes# of read on success or negative val on failure. */
2524 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
2528 ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
2529 msg->fds, VHOST_MEMORY_MAX_NREGIONS, &msg->fd_num);
2532 } else if (ret != VHOST_USER_HDR_SIZE) {
2533 VHOST_LOG_CONFIG(ERR, "Unexpected header size read\n");
2539 if (msg->size > sizeof(msg->payload)) {
2540 VHOST_LOG_CONFIG(ERR,
2541 "invalid msg size: %d\n", msg->size);
2544 ret = read(sockfd, &msg->payload, msg->size);
2547 if (ret != (int)msg->size) {
2548 VHOST_LOG_CONFIG(ERR,
2549 "read control message failed\n");
2558 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
2563 return send_fd_message(sockfd, (char *)msg,
2564 VHOST_USER_HDR_SIZE + msg->size, msg->fds, msg->fd_num);
2568 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
2573 msg->flags &= ~VHOST_USER_VERSION_MASK;
2574 msg->flags &= ~VHOST_USER_NEED_REPLY;
2575 msg->flags |= VHOST_USER_VERSION;
2576 msg->flags |= VHOST_USER_REPLY_MASK;
2578 return send_vhost_message(sockfd, msg);
2582 send_vhost_slave_message(struct virtio_net *dev, struct VhostUserMsg *msg)
2586 if (msg->flags & VHOST_USER_NEED_REPLY)
2587 rte_spinlock_lock(&dev->slave_req_lock);
2589 ret = send_vhost_message(dev->slave_req_fd, msg);
2590 if (ret < 0 && (msg->flags & VHOST_USER_NEED_REPLY))
2591 rte_spinlock_unlock(&dev->slave_req_lock);
2597 * Allocate a queue pair if it hasn't been allocated yet
2600 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev,
2601 struct VhostUserMsg *msg)
2605 switch (msg->request.master) {
2606 case VHOST_USER_SET_VRING_KICK:
2607 case VHOST_USER_SET_VRING_CALL:
2608 case VHOST_USER_SET_VRING_ERR:
2609 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
2611 case VHOST_USER_SET_VRING_NUM:
2612 case VHOST_USER_SET_VRING_BASE:
2613 case VHOST_USER_SET_VRING_ENABLE:
2614 vring_idx = msg->payload.state.index;
2616 case VHOST_USER_SET_VRING_ADDR:
2617 vring_idx = msg->payload.addr.index;
2623 if (vring_idx >= VHOST_MAX_VRING) {
2624 VHOST_LOG_CONFIG(ERR,
2625 "invalid vring index: %u\n", vring_idx);
2629 if (dev->virtqueue[vring_idx])
2632 return alloc_vring_queue(dev, vring_idx);
2636 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
2639 unsigned int vq_num = 0;
2641 while (vq_num < dev->nr_vring) {
2642 struct vhost_virtqueue *vq = dev->virtqueue[i];
2645 rte_spinlock_lock(&vq->access_lock);
2653 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
2656 unsigned int vq_num = 0;
2658 while (vq_num < dev->nr_vring) {
2659 struct vhost_virtqueue *vq = dev->virtqueue[i];
2662 rte_spinlock_unlock(&vq->access_lock);
2670 vhost_user_msg_handler(int vid, int fd)
2672 struct virtio_net *dev;
2673 struct VhostUserMsg msg;
2674 struct rte_vdpa_device *vdpa_dev;
2676 int unlock_required = 0;
2681 dev = get_device(vid);
2685 if (!dev->notify_ops) {
2686 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
2687 if (!dev->notify_ops) {
2688 VHOST_LOG_CONFIG(ERR,
2689 "failed to get callback ops for driver %s\n",
2695 ret = read_vhost_message(fd, &msg);
2698 VHOST_LOG_CONFIG(ERR,
2699 "vhost read message failed\n");
2701 VHOST_LOG_CONFIG(INFO,
2702 "vhost peer closed\n");
2708 request = msg.request.master;
2709 if (request > VHOST_USER_NONE && request < VHOST_USER_MAX &&
2710 vhost_message_str[request]) {
2711 if (request != VHOST_USER_IOTLB_MSG)
2712 VHOST_LOG_CONFIG(INFO, "read message %s\n",
2713 vhost_message_str[request]);
2715 VHOST_LOG_CONFIG(DEBUG, "read message %s\n",
2716 vhost_message_str[request]);
2718 VHOST_LOG_CONFIG(DEBUG, "External request %d\n", request);
2721 ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
2723 VHOST_LOG_CONFIG(ERR,
2724 "failed to alloc queue\n");
2729 * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
2730 * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
2731 * and device is destroyed. destroy_device waits for queues to be
2732 * inactive, so it is safe. Otherwise taking the access_lock
2733 * would cause a dead lock.
2736 case VHOST_USER_SET_FEATURES:
2737 case VHOST_USER_SET_PROTOCOL_FEATURES:
2738 case VHOST_USER_SET_OWNER:
2739 case VHOST_USER_SET_MEM_TABLE:
2740 case VHOST_USER_SET_LOG_BASE:
2741 case VHOST_USER_SET_LOG_FD:
2742 case VHOST_USER_SET_VRING_NUM:
2743 case VHOST_USER_SET_VRING_ADDR:
2744 case VHOST_USER_SET_VRING_BASE:
2745 case VHOST_USER_SET_VRING_KICK:
2746 case VHOST_USER_SET_VRING_CALL:
2747 case VHOST_USER_SET_VRING_ERR:
2748 case VHOST_USER_SET_VRING_ENABLE:
2749 case VHOST_USER_SEND_RARP:
2750 case VHOST_USER_NET_SET_MTU:
2751 case VHOST_USER_SET_SLAVE_REQ_FD:
2752 if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
2753 vhost_user_lock_all_queue_pairs(dev);
2754 unlock_required = 1;
2763 if (dev->extern_ops.pre_msg_handle) {
2764 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
2767 case RTE_VHOST_MSG_RESULT_REPLY:
2768 send_vhost_reply(fd, &msg);
2770 case RTE_VHOST_MSG_RESULT_ERR:
2771 case RTE_VHOST_MSG_RESULT_OK:
2773 goto skip_to_post_handle;
2774 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2780 if (request > VHOST_USER_NONE && request < VHOST_USER_MAX) {
2781 if (!vhost_message_handlers[request])
2782 goto skip_to_post_handle;
2783 ret = vhost_message_handlers[request](&dev, &msg, fd);
2786 case RTE_VHOST_MSG_RESULT_ERR:
2787 VHOST_LOG_CONFIG(ERR,
2788 "Processing %s failed.\n",
2789 vhost_message_str[request]);
2792 case RTE_VHOST_MSG_RESULT_OK:
2793 VHOST_LOG_CONFIG(DEBUG,
2794 "Processing %s succeeded.\n",
2795 vhost_message_str[request]);
2798 case RTE_VHOST_MSG_RESULT_REPLY:
2799 VHOST_LOG_CONFIG(DEBUG,
2800 "Processing %s succeeded and needs reply.\n",
2801 vhost_message_str[request]);
2802 send_vhost_reply(fd, &msg);
2810 skip_to_post_handle:
2811 if (ret != RTE_VHOST_MSG_RESULT_ERR &&
2812 dev->extern_ops.post_msg_handle) {
2813 ret = (*dev->extern_ops.post_msg_handle)(dev->vid,
2816 case RTE_VHOST_MSG_RESULT_REPLY:
2817 send_vhost_reply(fd, &msg);
2819 case RTE_VHOST_MSG_RESULT_ERR:
2820 case RTE_VHOST_MSG_RESULT_OK:
2822 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2828 if (unlock_required)
2829 vhost_user_unlock_all_queue_pairs(dev);
2831 /* If message was not handled at this stage, treat it as an error */
2833 VHOST_LOG_CONFIG(ERR,
2834 "vhost message (req: %d) was not handled.\n", request);
2835 close_msg_fds(&msg);
2836 ret = RTE_VHOST_MSG_RESULT_ERR;
2840 * If the request required a reply that was already sent,
2841 * this optional reply-ack won't be sent as the
2842 * VHOST_USER_NEED_REPLY was cleared in send_vhost_reply().
2844 if (msg.flags & VHOST_USER_NEED_REPLY) {
2845 msg.payload.u64 = ret == RTE_VHOST_MSG_RESULT_ERR;
2846 msg.size = sizeof(msg.payload.u64);
2848 send_vhost_reply(fd, &msg);
2849 } else if (ret == RTE_VHOST_MSG_RESULT_ERR) {
2850 VHOST_LOG_CONFIG(ERR,
2851 "vhost message handling failed.\n");
2855 for (i = 0; i < dev->nr_vring; i++) {
2856 struct vhost_virtqueue *vq = dev->virtqueue[i];
2857 bool cur_ready = vq_is_ready(dev, vq);
2859 if (cur_ready != (vq && vq->ready)) {
2860 vq->ready = cur_ready;
2861 vhost_user_notify_queue_state(dev, i, cur_ready);
2866 if (!virtio_is_ready(dev))
2870 * Virtio is now ready. If not done already, it is time
2871 * to notify the application it can process the rings and
2872 * configure the vDPA device if present.
2875 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
2876 if (dev->notify_ops->new_device(dev->vid) == 0)
2877 dev->flags |= VIRTIO_DEV_RUNNING;
2880 vdpa_dev = dev->vdpa_dev;
2884 if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
2885 if (vdpa_dev->ops->dev_conf(dev->vid))
2886 VHOST_LOG_CONFIG(ERR,
2887 "Failed to configure vDPA device\n");
2889 dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
2896 static int process_slave_message_reply(struct virtio_net *dev,
2897 const struct VhostUserMsg *msg)
2899 struct VhostUserMsg msg_reply;
2902 if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
2905 ret = read_vhost_message(dev->slave_req_fd, &msg_reply);
2908 VHOST_LOG_CONFIG(ERR,
2909 "vhost read slave message reply failed\n");
2911 VHOST_LOG_CONFIG(INFO,
2912 "vhost peer closed\n");
2918 if (msg_reply.request.slave != msg->request.slave) {
2919 VHOST_LOG_CONFIG(ERR,
2920 "Received unexpected msg type (%u), expected %u\n",
2921 msg_reply.request.slave, msg->request.slave);
2926 ret = msg_reply.payload.u64 ? -1 : 0;
2929 rte_spinlock_unlock(&dev->slave_req_lock);
2934 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
2937 struct VhostUserMsg msg = {
2938 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
2939 .flags = VHOST_USER_VERSION,
2940 .size = sizeof(msg.payload.iotlb),
2944 .type = VHOST_IOTLB_MISS,
2948 ret = send_vhost_message(dev->slave_req_fd, &msg);
2950 VHOST_LOG_CONFIG(ERR,
2951 "Failed to send IOTLB miss message (%d)\n",
2960 vhost_user_slave_config_change(struct virtio_net *dev, bool need_reply)
2963 struct VhostUserMsg msg = {
2964 .request.slave = VHOST_USER_SLAVE_CONFIG_CHANGE_MSG,
2965 .flags = VHOST_USER_VERSION,
2970 msg.flags |= VHOST_USER_NEED_REPLY;
2972 ret = send_vhost_slave_message(dev, &msg);
2974 VHOST_LOG_CONFIG(ERR,
2975 "Failed to send config change (%d)\n",
2980 return process_slave_message_reply(dev, &msg);
2984 rte_vhost_slave_config_change(int vid, bool need_reply)
2986 struct virtio_net *dev;
2988 dev = get_device(vid);
2992 return vhost_user_slave_config_change(dev, need_reply);
2995 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
3001 struct VhostUserMsg msg = {
3002 .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
3003 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
3004 .size = sizeof(msg.payload.area),
3006 .u64 = index & VHOST_USER_VRING_IDX_MASK,
3013 msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
3019 ret = send_vhost_slave_message(dev, &msg);
3021 VHOST_LOG_CONFIG(ERR,
3022 "Failed to set host notifier (%d)\n", ret);
3026 return process_slave_message_reply(dev, &msg);
3029 int rte_vhost_host_notifier_ctrl(int vid, uint16_t qid, bool enable)
3031 struct virtio_net *dev;
3032 struct rte_vdpa_device *vdpa_dev;
3033 int vfio_device_fd, ret = 0;
3034 uint64_t offset, size;
3035 unsigned int i, q_start, q_last;
3037 dev = get_device(vid);
3041 vdpa_dev = dev->vdpa_dev;
3042 if (vdpa_dev == NULL)
3045 if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
3046 !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
3047 !(dev->protocol_features &
3048 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
3049 !(dev->protocol_features &
3050 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
3051 !(dev->protocol_features &
3052 (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
3055 if (qid == RTE_VHOST_QUEUE_ALL) {
3057 q_last = dev->nr_vring - 1;
3059 if (qid >= dev->nr_vring)
3065 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
3066 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
3068 vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
3069 if (vfio_device_fd < 0)
3073 for (i = q_start; i <= q_last; i++) {
3074 if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
3080 if (vhost_user_slave_set_vring_host_notifier(dev, i,
3081 vfio_device_fd, offset, size) < 0) {
3088 for (i = q_start; i <= q_last; i++) {
3089 vhost_user_slave_set_vring_host_notifier(dev, i, -1,