1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2017 Intel Corporation
5 #include <linux/vhost.h>
6 #include <linux/virtio_net.h>
10 #ifdef RTE_LIBRTE_VHOST_NUMA
15 #include <rte_errno.h>
16 #include <rte_ethdev.h>
18 #include <rte_string_fns.h>
19 #include <rte_memory.h>
20 #include <rte_malloc.h>
21 #include <rte_vhost.h>
22 #include <rte_rwlock.h>
26 #include "vhost_user.h"
28 struct virtio_net *vhost_devices[MAX_VHOST_DEVICE];
30 /* Called with iotlb_lock read-locked */
32 __vhost_iova_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
33 uint64_t iova, uint64_t *size, uint8_t perm)
35 uint64_t vva, tmp_size;
42 vva = vhost_user_iotlb_cache_find(vq, iova, &tmp_size, perm);
43 if (tmp_size == *size)
48 if (!vhost_user_iotlb_pending_miss(vq, iova, perm)) {
50 * iotlb_lock is read-locked for a full burst,
51 * but it only protects the iotlb cache.
52 * In case of IOTLB miss, we might block on the socket,
53 * which could cause a deadlock with QEMU if an IOTLB update
54 * is being handled. We can safely unlock here to avoid it.
56 vhost_user_iotlb_rd_unlock(vq);
58 vhost_user_iotlb_pending_insert(vq, iova, perm);
59 if (vhost_user_iotlb_miss(dev, iova, perm)) {
60 RTE_LOG(ERR, VHOST_CONFIG,
61 "IOTLB miss req failed for IOVA 0x%" PRIx64 "\n",
63 vhost_user_iotlb_pending_remove(vq, iova, 1, perm);
66 vhost_user_iotlb_rd_lock(vq);
72 #define VHOST_LOG_PAGE 4096
75 * Atomically set a bit in memory.
77 static __rte_always_inline void
78 vhost_set_bit(unsigned int nr, volatile uint8_t *addr)
80 #if defined(RTE_TOOLCHAIN_GCC) && (GCC_VERSION < 70100)
82 * __sync_ built-ins are deprecated, but __atomic_ ones
83 * are sub-optimized in older GCC versions.
85 __sync_fetch_and_or_1(addr, (1U << nr));
87 __atomic_fetch_or(addr, (1U << nr), __ATOMIC_RELAXED);
91 static __rte_always_inline void
92 vhost_log_page(uint8_t *log_base, uint64_t page)
94 vhost_set_bit(page % 8, &log_base[page / 8]);
98 __vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len)
102 if (unlikely(!dev->log_base || !len))
105 if (unlikely(dev->log_size <= ((addr + len - 1) / VHOST_LOG_PAGE / 8)))
108 /* To make sure guest memory updates are committed before logging */
111 page = addr / VHOST_LOG_PAGE;
112 while (page * VHOST_LOG_PAGE < addr + len) {
113 vhost_log_page((uint8_t *)(uintptr_t)dev->log_base, page);
119 __vhost_log_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
120 uint64_t iova, uint64_t len)
122 uint64_t hva, gpa, map_len;
125 hva = __vhost_iova_to_vva(dev, vq, iova, &map_len, VHOST_ACCESS_RW);
126 if (map_len != len) {
127 RTE_LOG(ERR, VHOST_CONFIG,
128 "Failed to write log for IOVA 0x%" PRIx64 ". No IOTLB entry found\n",
133 gpa = hva_to_gpa(dev, hva, len);
135 __vhost_log_write(dev, gpa, len);
139 __vhost_log_cache_sync(struct virtio_net *dev, struct vhost_virtqueue *vq)
141 unsigned long *log_base;
144 if (unlikely(!dev->log_base))
149 log_base = (unsigned long *)(uintptr_t)dev->log_base;
151 for (i = 0; i < vq->log_cache_nb_elem; i++) {
152 struct log_cache_entry *elem = vq->log_cache + i;
154 #if defined(RTE_TOOLCHAIN_GCC) && (GCC_VERSION < 70100)
156 * '__sync' builtins are deprecated, but '__atomic' ones
157 * are sub-optimized in older GCC versions.
159 __sync_fetch_and_or(log_base + elem->offset, elem->val);
161 __atomic_fetch_or(log_base + elem->offset, elem->val,
168 vq->log_cache_nb_elem = 0;
171 static __rte_always_inline void
172 vhost_log_cache_page(struct virtio_net *dev, struct vhost_virtqueue *vq,
175 uint32_t bit_nr = page % (sizeof(unsigned long) << 3);
176 uint32_t offset = page / (sizeof(unsigned long) << 3);
179 for (i = 0; i < vq->log_cache_nb_elem; i++) {
180 struct log_cache_entry *elem = vq->log_cache + i;
182 if (elem->offset == offset) {
183 elem->val |= (1UL << bit_nr);
188 if (unlikely(i >= VHOST_LOG_CACHE_NR)) {
190 * No more room for a new log cache entry,
191 * so write the dirty log map directly.
194 vhost_log_page((uint8_t *)(uintptr_t)dev->log_base, page);
199 vq->log_cache[i].offset = offset;
200 vq->log_cache[i].val = (1UL << bit_nr);
201 vq->log_cache_nb_elem++;
205 __vhost_log_cache_write(struct virtio_net *dev, struct vhost_virtqueue *vq,
206 uint64_t addr, uint64_t len)
210 if (unlikely(!dev->log_base || !len))
213 if (unlikely(dev->log_size <= ((addr + len - 1) / VHOST_LOG_PAGE / 8)))
216 page = addr / VHOST_LOG_PAGE;
217 while (page * VHOST_LOG_PAGE < addr + len) {
218 vhost_log_cache_page(dev, vq, page);
224 __vhost_log_cache_write_iova(struct virtio_net *dev, struct vhost_virtqueue *vq,
225 uint64_t iova, uint64_t len)
227 uint64_t hva, gpa, map_len;
230 hva = __vhost_iova_to_vva(dev, vq, iova, &map_len, VHOST_ACCESS_RW);
231 if (map_len != len) {
232 RTE_LOG(ERR, VHOST_CONFIG,
233 "Failed to write log for IOVA 0x%" PRIx64 ". No IOTLB entry found\n",
238 gpa = hva_to_gpa(dev, hva, len);
240 __vhost_log_cache_write(dev, vq, gpa, len);
244 vhost_alloc_copy_ind_table(struct virtio_net *dev, struct vhost_virtqueue *vq,
245 uint64_t desc_addr, uint64_t desc_len)
249 uint64_t len, remain = desc_len;
251 idesc = rte_malloc(__func__, desc_len, 0);
252 if (unlikely(!idesc))
255 dst = (uint64_t)(uintptr_t)idesc;
259 src = vhost_iova_to_vva(dev, vq, desc_addr, &len,
261 if (unlikely(!src || !len)) {
266 rte_memcpy((void *)(uintptr_t)dst, (void *)(uintptr_t)src, len);
277 cleanup_vq(struct vhost_virtqueue *vq, int destroy)
279 if ((vq->callfd >= 0) && (destroy != 0))
286 cleanup_vq_inflight(struct virtio_net *dev, struct vhost_virtqueue *vq)
288 if (!(dev->protocol_features &
289 (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
292 if (vq_is_packed(dev)) {
293 if (vq->inflight_packed)
294 vq->inflight_packed = NULL;
296 if (vq->inflight_split)
297 vq->inflight_split = NULL;
300 if (vq->resubmit_inflight) {
301 if (vq->resubmit_inflight->resubmit_list) {
302 free(vq->resubmit_inflight->resubmit_list);
303 vq->resubmit_inflight->resubmit_list = NULL;
305 free(vq->resubmit_inflight);
306 vq->resubmit_inflight = NULL;
311 * Unmap any memory, close any file descriptors and
312 * free any memory owned by a device.
315 cleanup_device(struct virtio_net *dev, int destroy)
319 vhost_backend_cleanup(dev);
321 for (i = 0; i < dev->nr_vring; i++) {
322 cleanup_vq(dev->virtqueue[i], destroy);
323 cleanup_vq_inflight(dev, dev->virtqueue[i]);
328 free_vq(struct virtio_net *dev, struct vhost_virtqueue *vq)
330 if (vq_is_packed(dev))
331 rte_free(vq->shadow_used_packed);
333 rte_free(vq->shadow_used_split);
334 rte_free(vq->batch_copy_elems);
335 rte_mempool_free(vq->iotlb_pool);
340 * Release virtqueues and device memory.
343 free_device(struct virtio_net *dev)
347 for (i = 0; i < dev->nr_vring; i++)
348 free_vq(dev, dev->virtqueue[i]);
354 vring_translate_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
356 uint64_t req_size, size;
358 req_size = sizeof(struct vring_desc) * vq->size;
360 vq->desc = (struct vring_desc *)(uintptr_t)vhost_iova_to_vva(dev, vq,
361 vq->ring_addrs.desc_user_addr,
362 &size, VHOST_ACCESS_RW);
363 if (!vq->desc || size != req_size)
366 req_size = sizeof(struct vring_avail);
367 req_size += sizeof(uint16_t) * vq->size;
368 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
369 req_size += sizeof(uint16_t);
371 vq->avail = (struct vring_avail *)(uintptr_t)vhost_iova_to_vva(dev, vq,
372 vq->ring_addrs.avail_user_addr,
373 &size, VHOST_ACCESS_RW);
374 if (!vq->avail || size != req_size)
377 req_size = sizeof(struct vring_used);
378 req_size += sizeof(struct vring_used_elem) * vq->size;
379 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
380 req_size += sizeof(uint16_t);
382 vq->used = (struct vring_used *)(uintptr_t)vhost_iova_to_vva(dev, vq,
383 vq->ring_addrs.used_user_addr,
384 &size, VHOST_ACCESS_RW);
385 if (!vq->used || size != req_size)
392 vring_translate_packed(struct virtio_net *dev, struct vhost_virtqueue *vq)
394 uint64_t req_size, size;
396 req_size = sizeof(struct vring_packed_desc) * vq->size;
398 vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
399 vhost_iova_to_vva(dev, vq, vq->ring_addrs.desc_user_addr,
400 &size, VHOST_ACCESS_RW);
401 if (!vq->desc_packed || size != req_size)
404 req_size = sizeof(struct vring_packed_desc_event);
406 vq->driver_event = (struct vring_packed_desc_event *)(uintptr_t)
407 vhost_iova_to_vva(dev, vq, vq->ring_addrs.avail_user_addr,
408 &size, VHOST_ACCESS_RW);
409 if (!vq->driver_event || size != req_size)
412 req_size = sizeof(struct vring_packed_desc_event);
414 vq->device_event = (struct vring_packed_desc_event *)(uintptr_t)
415 vhost_iova_to_vva(dev, vq, vq->ring_addrs.used_user_addr,
416 &size, VHOST_ACCESS_RW);
417 if (!vq->device_event || size != req_size)
424 vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq)
427 if (!(dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
430 if (vq_is_packed(dev)) {
431 if (vring_translate_packed(dev, vq) < 0)
434 if (vring_translate_split(dev, vq) < 0)
443 vring_invalidate(struct virtio_net *dev, struct vhost_virtqueue *vq)
445 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
446 vhost_user_iotlb_wr_lock(vq);
452 vq->log_guest_addr = 0;
454 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
455 vhost_user_iotlb_wr_unlock(vq);
459 init_vring_queue(struct virtio_net *dev, uint32_t vring_idx)
461 struct vhost_virtqueue *vq;
463 if (vring_idx >= VHOST_MAX_VRING) {
464 RTE_LOG(ERR, VHOST_CONFIG,
465 "Failed not init vring, out of bound (%d)\n",
470 vq = dev->virtqueue[vring_idx];
472 memset(vq, 0, sizeof(struct vhost_virtqueue));
474 vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
475 vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
477 vhost_user_iotlb_init(dev, vring_idx);
478 /* Backends are set to -1 indicating an inactive device. */
481 TAILQ_INIT(&vq->zmbuf_list);
485 reset_vring_queue(struct virtio_net *dev, uint32_t vring_idx)
487 struct vhost_virtqueue *vq;
490 if (vring_idx >= VHOST_MAX_VRING) {
491 RTE_LOG(ERR, VHOST_CONFIG,
492 "Failed not init vring, out of bound (%d)\n",
497 vq = dev->virtqueue[vring_idx];
499 init_vring_queue(dev, vring_idx);
504 alloc_vring_queue(struct virtio_net *dev, uint32_t vring_idx)
506 struct vhost_virtqueue *vq;
508 vq = rte_malloc(NULL, sizeof(struct vhost_virtqueue), 0);
510 RTE_LOG(ERR, VHOST_CONFIG,
511 "Failed to allocate memory for vring:%u.\n", vring_idx);
515 dev->virtqueue[vring_idx] = vq;
516 init_vring_queue(dev, vring_idx);
517 rte_spinlock_init(&vq->access_lock);
518 vq->avail_wrap_counter = 1;
519 vq->used_wrap_counter = 1;
520 vq->signalled_used_valid = false;
528 * Reset some variables in device structure, while keeping few
529 * others untouched, such as vid, ifname, nr_vring: they
530 * should be same unless the device is removed.
533 reset_device(struct virtio_net *dev)
538 dev->protocol_features = 0;
539 dev->flags &= VIRTIO_DEV_BUILTIN_VIRTIO_NET;
541 for (i = 0; i < dev->nr_vring; i++)
542 reset_vring_queue(dev, i);
546 * Invoked when there is a new vhost-user connection established (when
547 * there is a new virtio device being attached).
550 vhost_new_device(void)
552 struct virtio_net *dev;
555 for (i = 0; i < MAX_VHOST_DEVICE; i++) {
556 if (vhost_devices[i] == NULL)
560 if (i == MAX_VHOST_DEVICE) {
561 RTE_LOG(ERR, VHOST_CONFIG,
562 "Failed to find a free slot for new device.\n");
566 dev = rte_zmalloc(NULL, sizeof(struct virtio_net), 0);
568 RTE_LOG(ERR, VHOST_CONFIG,
569 "Failed to allocate memory for new dev.\n");
573 vhost_devices[i] = dev;
575 dev->flags = VIRTIO_DEV_BUILTIN_VIRTIO_NET;
576 dev->slave_req_fd = -1;
577 dev->vdpa_dev_id = -1;
578 dev->postcopy_ufd = -1;
579 rte_spinlock_init(&dev->slave_req_lock);
585 vhost_destroy_device_notify(struct virtio_net *dev)
587 struct rte_vdpa_device *vdpa_dev;
590 if (dev->flags & VIRTIO_DEV_RUNNING) {
591 did = dev->vdpa_dev_id;
592 vdpa_dev = rte_vdpa_get_device(did);
593 if (vdpa_dev && vdpa_dev->ops->dev_close)
594 vdpa_dev->ops->dev_close(dev->vid);
595 dev->flags &= ~VIRTIO_DEV_RUNNING;
596 dev->notify_ops->destroy_device(dev->vid);
601 * Invoked when there is the vhost-user connection is broken (when
602 * the virtio device is being detached).
605 vhost_destroy_device(int vid)
607 struct virtio_net *dev = get_device(vid);
612 vhost_destroy_device_notify(dev);
614 cleanup_device(dev, 1);
617 vhost_devices[vid] = NULL;
621 vhost_attach_vdpa_device(int vid, int did)
623 struct virtio_net *dev = get_device(vid);
628 if (rte_vdpa_get_device(did) == NULL)
631 dev->vdpa_dev_id = did;
635 vhost_set_ifname(int vid, const char *if_name, unsigned int if_len)
637 struct virtio_net *dev;
640 dev = get_device(vid);
644 len = if_len > sizeof(dev->ifname) ?
645 sizeof(dev->ifname) : if_len;
647 strncpy(dev->ifname, if_name, len);
648 dev->ifname[sizeof(dev->ifname) - 1] = '\0';
652 vhost_enable_dequeue_zero_copy(int vid)
654 struct virtio_net *dev = get_device(vid);
659 dev->dequeue_zero_copy = 1;
663 vhost_set_builtin_virtio_net(int vid, bool enable)
665 struct virtio_net *dev = get_device(vid);
671 dev->flags |= VIRTIO_DEV_BUILTIN_VIRTIO_NET;
673 dev->flags &= ~VIRTIO_DEV_BUILTIN_VIRTIO_NET;
677 rte_vhost_get_mtu(int vid, uint16_t *mtu)
679 struct virtio_net *dev = get_device(vid);
681 if (dev == NULL || mtu == NULL)
684 if (!(dev->flags & VIRTIO_DEV_READY))
687 if (!(dev->features & (1ULL << VIRTIO_NET_F_MTU)))
696 rte_vhost_get_numa_node(int vid)
698 #ifdef RTE_LIBRTE_VHOST_NUMA
699 struct virtio_net *dev = get_device(vid);
703 if (dev == NULL || numa_available() != 0)
706 ret = get_mempolicy(&numa_node, NULL, 0, dev,
707 MPOL_F_NODE | MPOL_F_ADDR);
709 RTE_LOG(ERR, VHOST_CONFIG,
710 "(%d) failed to query numa node: %s\n",
711 vid, rte_strerror(errno));
723 rte_vhost_get_queue_num(int vid)
725 struct virtio_net *dev = get_device(vid);
730 return dev->nr_vring / 2;
734 rte_vhost_get_vring_num(int vid)
736 struct virtio_net *dev = get_device(vid);
741 return dev->nr_vring;
745 rte_vhost_get_ifname(int vid, char *buf, size_t len)
747 struct virtio_net *dev = get_device(vid);
749 if (dev == NULL || buf == NULL)
752 len = RTE_MIN(len, sizeof(dev->ifname));
754 strncpy(buf, dev->ifname, len);
761 rte_vhost_get_negotiated_features(int vid, uint64_t *features)
763 struct virtio_net *dev;
765 dev = get_device(vid);
766 if (dev == NULL || features == NULL)
769 *features = dev->features;
774 rte_vhost_get_mem_table(int vid, struct rte_vhost_memory **mem)
776 struct virtio_net *dev;
777 struct rte_vhost_memory *m;
780 dev = get_device(vid);
781 if (dev == NULL || mem == NULL)
784 size = dev->mem->nregions * sizeof(struct rte_vhost_mem_region);
785 m = malloc(sizeof(struct rte_vhost_memory) + size);
789 m->nregions = dev->mem->nregions;
790 memcpy(m->regions, dev->mem->regions, size);
797 rte_vhost_get_vhost_vring(int vid, uint16_t vring_idx,
798 struct rte_vhost_vring *vring)
800 struct virtio_net *dev;
801 struct vhost_virtqueue *vq;
803 dev = get_device(vid);
804 if (dev == NULL || vring == NULL)
807 if (vring_idx >= VHOST_MAX_VRING)
810 vq = dev->virtqueue[vring_idx];
814 vring->desc = vq->desc;
815 vring->avail = vq->avail;
816 vring->used = vq->used;
817 vring->log_guest_addr = vq->log_guest_addr;
819 vring->callfd = vq->callfd;
820 vring->kickfd = vq->kickfd;
821 vring->size = vq->size;
827 rte_vhost_set_inflight_desc_split(int vid, uint16_t vring_idx,
830 struct vhost_virtqueue *vq;
831 struct virtio_net *dev;
833 dev = get_device(vid);
837 if (unlikely(!(dev->protocol_features &
838 (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
841 if (unlikely(vq_is_packed(dev)))
844 if (unlikely(vring_idx >= VHOST_MAX_VRING))
847 vq = dev->virtqueue[vring_idx];
851 if (unlikely(!vq->inflight_split))
854 if (unlikely(idx >= vq->size))
857 vq->inflight_split->desc[idx].counter = vq->global_counter++;
858 vq->inflight_split->desc[idx].inflight = 1;
863 rte_vhost_set_inflight_desc_packed(int vid, uint16_t vring_idx,
864 uint16_t head, uint16_t last,
865 uint16_t *inflight_entry)
867 struct rte_vhost_inflight_info_packed *inflight_info;
868 struct virtio_net *dev;
869 struct vhost_virtqueue *vq;
870 struct vring_packed_desc *desc;
871 uint16_t old_free_head, free_head;
873 dev = get_device(vid);
877 if (unlikely(!(dev->protocol_features &
878 (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
881 if (unlikely(!vq_is_packed(dev)))
884 if (unlikely(vring_idx >= VHOST_MAX_VRING))
887 vq = dev->virtqueue[vring_idx];
891 inflight_info = vq->inflight_packed;
892 if (unlikely(!inflight_info))
895 if (unlikely(head >= vq->size))
898 desc = vq->desc_packed;
899 old_free_head = inflight_info->old_free_head;
900 if (unlikely(old_free_head >= vq->size))
903 free_head = old_free_head;
905 /* init header descriptor */
906 inflight_info->desc[old_free_head].num = 0;
907 inflight_info->desc[old_free_head].counter = vq->global_counter++;
908 inflight_info->desc[old_free_head].inflight = 1;
910 /* save desc entry in flight entry */
911 while (head != ((last + 1) % vq->size)) {
912 inflight_info->desc[old_free_head].num++;
913 inflight_info->desc[free_head].addr = desc[head].addr;
914 inflight_info->desc[free_head].len = desc[head].len;
915 inflight_info->desc[free_head].flags = desc[head].flags;
916 inflight_info->desc[free_head].id = desc[head].id;
918 inflight_info->desc[old_free_head].last = free_head;
919 free_head = inflight_info->desc[free_head].next;
920 inflight_info->free_head = free_head;
921 head = (head + 1) % vq->size;
924 inflight_info->old_free_head = free_head;
925 *inflight_entry = old_free_head;
931 rte_vhost_clr_inflight_desc_split(int vid, uint16_t vring_idx,
932 uint16_t last_used_idx, uint16_t idx)
934 struct virtio_net *dev;
935 struct vhost_virtqueue *vq;
937 dev = get_device(vid);
941 if (unlikely(!(dev->protocol_features &
942 (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
945 if (unlikely(vq_is_packed(dev)))
948 if (unlikely(vring_idx >= VHOST_MAX_VRING))
951 vq = dev->virtqueue[vring_idx];
955 if (unlikely(!vq->inflight_split))
958 if (unlikely(idx >= vq->size))
963 vq->inflight_split->desc[idx].inflight = 0;
967 vq->inflight_split->used_idx = last_used_idx;
972 rte_vhost_clr_inflight_desc_packed(int vid, uint16_t vring_idx,
975 struct rte_vhost_inflight_info_packed *inflight_info;
976 struct virtio_net *dev;
977 struct vhost_virtqueue *vq;
979 dev = get_device(vid);
983 if (unlikely(!(dev->protocol_features &
984 (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
987 if (unlikely(!vq_is_packed(dev)))
990 if (unlikely(vring_idx >= VHOST_MAX_VRING))
993 vq = dev->virtqueue[vring_idx];
997 inflight_info = vq->inflight_packed;
998 if (unlikely(!inflight_info))
1001 if (unlikely(head >= vq->size))
1006 inflight_info->desc[head].inflight = 0;
1010 inflight_info->old_free_head = inflight_info->free_head;
1011 inflight_info->old_used_idx = inflight_info->used_idx;
1012 inflight_info->old_used_wrap_counter = inflight_info->used_wrap_counter;
1018 rte_vhost_set_last_inflight_io_split(int vid, uint16_t vring_idx,
1021 struct virtio_net *dev;
1022 struct vhost_virtqueue *vq;
1024 dev = get_device(vid);
1028 if (unlikely(!(dev->protocol_features &
1029 (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
1032 if (unlikely(vq_is_packed(dev)))
1035 if (unlikely(vring_idx >= VHOST_MAX_VRING))
1038 vq = dev->virtqueue[vring_idx];
1042 if (unlikely(!vq->inflight_split))
1045 vq->inflight_split->last_inflight_io = idx;
1050 rte_vhost_set_last_inflight_io_packed(int vid, uint16_t vring_idx,
1053 struct rte_vhost_inflight_info_packed *inflight_info;
1054 struct virtio_net *dev;
1055 struct vhost_virtqueue *vq;
1058 dev = get_device(vid);
1062 if (unlikely(!(dev->protocol_features &
1063 (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
1066 if (unlikely(!vq_is_packed(dev)))
1069 if (unlikely(vring_idx >= VHOST_MAX_VRING))
1072 vq = dev->virtqueue[vring_idx];
1076 inflight_info = vq->inflight_packed;
1077 if (unlikely(!inflight_info))
1080 if (unlikely(head >= vq->size))
1083 last = inflight_info->desc[head].last;
1084 if (unlikely(last >= vq->size))
1087 inflight_info->desc[last].next = inflight_info->free_head;
1088 inflight_info->free_head = head;
1089 inflight_info->used_idx += inflight_info->desc[head].num;
1090 if (inflight_info->used_idx >= inflight_info->desc_num) {
1091 inflight_info->used_idx -= inflight_info->desc_num;
1092 inflight_info->used_wrap_counter =
1093 !inflight_info->used_wrap_counter;
1100 rte_vhost_vring_call(int vid, uint16_t vring_idx)
1102 struct virtio_net *dev;
1103 struct vhost_virtqueue *vq;
1105 dev = get_device(vid);
1109 if (vring_idx >= VHOST_MAX_VRING)
1112 vq = dev->virtqueue[vring_idx];
1116 if (vq_is_packed(dev))
1117 vhost_vring_call_packed(dev, vq);
1119 vhost_vring_call_split(dev, vq);
1125 rte_vhost_avail_entries(int vid, uint16_t queue_id)
1127 struct virtio_net *dev;
1128 struct vhost_virtqueue *vq;
1131 dev = get_device(vid);
1135 vq = dev->virtqueue[queue_id];
1137 rte_spinlock_lock(&vq->access_lock);
1139 if (unlikely(!vq->enabled || vq->avail == NULL))
1142 ret = *(volatile uint16_t *)&vq->avail->idx - vq->last_used_idx;
1145 rte_spinlock_unlock(&vq->access_lock);
1150 vhost_enable_notify_split(struct virtio_net *dev,
1151 struct vhost_virtqueue *vq, int enable)
1153 if (vq->used == NULL)
1156 if (!(dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))) {
1158 vq->used->flags &= ~VRING_USED_F_NO_NOTIFY;
1160 vq->used->flags |= VRING_USED_F_NO_NOTIFY;
1163 vhost_avail_event(vq) = vq->last_avail_idx;
1169 vhost_enable_notify_packed(struct virtio_net *dev,
1170 struct vhost_virtqueue *vq, int enable)
1174 if (vq->device_event == NULL)
1178 vq->device_event->flags = VRING_EVENT_F_DISABLE;
1182 flags = VRING_EVENT_F_ENABLE;
1183 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX)) {
1184 flags = VRING_EVENT_F_DESC;
1185 vq->device_event->off_wrap = vq->last_avail_idx |
1186 vq->avail_wrap_counter << 15;
1191 vq->device_event->flags = flags;
1196 rte_vhost_enable_guest_notification(int vid, uint16_t queue_id, int enable)
1198 struct virtio_net *dev = get_device(vid);
1199 struct vhost_virtqueue *vq;
1205 vq = dev->virtqueue[queue_id];
1207 rte_spinlock_lock(&vq->access_lock);
1209 if (vq_is_packed(dev))
1210 ret = vhost_enable_notify_packed(dev, vq, enable);
1212 ret = vhost_enable_notify_split(dev, vq, enable);
1214 rte_spinlock_unlock(&vq->access_lock);
1220 rte_vhost_log_write(int vid, uint64_t addr, uint64_t len)
1222 struct virtio_net *dev = get_device(vid);
1227 vhost_log_write(dev, addr, len);
1231 rte_vhost_log_used_vring(int vid, uint16_t vring_idx,
1232 uint64_t offset, uint64_t len)
1234 struct virtio_net *dev;
1235 struct vhost_virtqueue *vq;
1237 dev = get_device(vid);
1241 if (vring_idx >= VHOST_MAX_VRING)
1243 vq = dev->virtqueue[vring_idx];
1247 vhost_log_used_vring(dev, vq, offset, len);
1251 rte_vhost_rx_queue_count(int vid, uint16_t qid)
1253 struct virtio_net *dev;
1254 struct vhost_virtqueue *vq;
1257 dev = get_device(vid);
1261 if (unlikely(qid >= dev->nr_vring || (qid & 1) == 0)) {
1262 RTE_LOG(ERR, VHOST_DATA, "(%d) %s: invalid virtqueue idx %d.\n",
1263 dev->vid, __func__, qid);
1267 vq = dev->virtqueue[qid];
1271 rte_spinlock_lock(&vq->access_lock);
1273 if (unlikely(vq->enabled == 0 || vq->avail == NULL))
1276 ret = *((volatile uint16_t *)&vq->avail->idx) - vq->last_avail_idx;
1279 rte_spinlock_unlock(&vq->access_lock);
1283 int rte_vhost_get_vdpa_device_id(int vid)
1285 struct virtio_net *dev = get_device(vid);
1290 return dev->vdpa_dev_id;
1293 int rte_vhost_get_log_base(int vid, uint64_t *log_base,
1296 struct virtio_net *dev = get_device(vid);
1298 if (dev == NULL || log_base == NULL || log_size == NULL)
1301 *log_base = dev->log_base;
1302 *log_size = dev->log_size;
1307 int rte_vhost_get_vring_base(int vid, uint16_t queue_id,
1308 uint16_t *last_avail_idx, uint16_t *last_used_idx)
1310 struct virtio_net *dev = get_device(vid);
1312 if (dev == NULL || last_avail_idx == NULL || last_used_idx == NULL)
1315 *last_avail_idx = dev->virtqueue[queue_id]->last_avail_idx;
1316 *last_used_idx = dev->virtqueue[queue_id]->last_used_idx;
1321 int rte_vhost_set_vring_base(int vid, uint16_t queue_id,
1322 uint16_t last_avail_idx, uint16_t last_used_idx)
1324 struct virtio_net *dev = get_device(vid);
1329 dev->virtqueue[queue_id]->last_avail_idx = last_avail_idx;
1330 dev->virtqueue[queue_id]->last_used_idx = last_used_idx;
1335 int rte_vhost_extern_callback_register(int vid,
1336 struct rte_vhost_user_extern_ops const * const ops, void *ctx)
1338 struct virtio_net *dev = get_device(vid);
1340 if (dev == NULL || ops == NULL)
1343 dev->extern_ops = *ops;
1344 dev->extern_data = ctx;