1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2016 Intel Corporation
7 #include <linux/virtio_net.h>
10 #include <rte_memcpy.h>
11 #include <rte_ether.h>
13 #include <rte_vhost.h>
18 #include <rte_spinlock.h>
19 #include <rte_malloc.h>
20 #include <rte_vhost_async.h>
25 #define MAX_BATCH_LEN 256
27 #define VHOST_ASYNC_BATCH_THRESHOLD 32
29 static __rte_always_inline bool
30 rxvq_is_mergeable(struct virtio_net *dev)
32 return dev->features & (1ULL << VIRTIO_NET_F_MRG_RXBUF);
35 static __rte_always_inline bool
36 virtio_net_is_inorder(struct virtio_net *dev)
38 return dev->features & (1ULL << VIRTIO_F_IN_ORDER);
42 is_valid_virt_queue_idx(uint32_t idx, int is_tx, uint32_t nr_vring)
44 return (is_tx ^ (idx & 1)) == 0 && idx < nr_vring;
48 do_data_copy_enqueue(struct virtio_net *dev, struct vhost_virtqueue *vq)
50 struct batch_copy_elem *elem = vq->batch_copy_elems;
51 uint16_t count = vq->batch_copy_nb_elems;
54 for (i = 0; i < count; i++) {
55 rte_memcpy(elem[i].dst, elem[i].src, elem[i].len);
56 vhost_log_cache_write_iova(dev, vq, elem[i].log_addr,
58 PRINT_PACKET(dev, (uintptr_t)elem[i].dst, elem[i].len, 0);
61 vq->batch_copy_nb_elems = 0;
65 do_data_copy_dequeue(struct vhost_virtqueue *vq)
67 struct batch_copy_elem *elem = vq->batch_copy_elems;
68 uint16_t count = vq->batch_copy_nb_elems;
71 for (i = 0; i < count; i++)
72 rte_memcpy(elem[i].dst, elem[i].src, elem[i].len);
74 vq->batch_copy_nb_elems = 0;
77 static __rte_always_inline void
78 do_flush_shadow_used_ring_split(struct virtio_net *dev,
79 struct vhost_virtqueue *vq,
80 uint16_t to, uint16_t from, uint16_t size)
82 rte_memcpy(&vq->used->ring[to],
83 &vq->shadow_used_split[from],
84 size * sizeof(struct vring_used_elem));
85 vhost_log_cache_used_vring(dev, vq,
86 offsetof(struct vring_used, ring[to]),
87 size * sizeof(struct vring_used_elem));
90 static __rte_always_inline void
91 flush_shadow_used_ring_split(struct virtio_net *dev, struct vhost_virtqueue *vq)
93 uint16_t used_idx = vq->last_used_idx & (vq->size - 1);
95 if (used_idx + vq->shadow_used_idx <= vq->size) {
96 do_flush_shadow_used_ring_split(dev, vq, used_idx, 0,
101 /* update used ring interval [used_idx, vq->size] */
102 size = vq->size - used_idx;
103 do_flush_shadow_used_ring_split(dev, vq, used_idx, 0, size);
105 /* update the left half used ring interval [0, left_size] */
106 do_flush_shadow_used_ring_split(dev, vq, 0, size,
107 vq->shadow_used_idx - size);
109 vq->last_used_idx += vq->shadow_used_idx;
111 vhost_log_cache_sync(dev, vq);
113 __atomic_add_fetch(&vq->used->idx, vq->shadow_used_idx,
115 vq->shadow_used_idx = 0;
116 vhost_log_used_vring(dev, vq, offsetof(struct vring_used, idx),
117 sizeof(vq->used->idx));
120 static __rte_always_inline void
121 update_shadow_used_ring_split(struct vhost_virtqueue *vq,
122 uint16_t desc_idx, uint32_t len)
124 uint16_t i = vq->shadow_used_idx++;
126 vq->shadow_used_split[i].id = desc_idx;
127 vq->shadow_used_split[i].len = len;
130 static __rte_always_inline void
131 vhost_flush_enqueue_shadow_packed(struct virtio_net *dev,
132 struct vhost_virtqueue *vq)
135 uint16_t used_idx = vq->last_used_idx;
136 uint16_t head_idx = vq->last_used_idx;
137 uint16_t head_flags = 0;
139 /* Split loop in two to save memory barriers */
140 for (i = 0; i < vq->shadow_used_idx; i++) {
141 vq->desc_packed[used_idx].id = vq->shadow_used_packed[i].id;
142 vq->desc_packed[used_idx].len = vq->shadow_used_packed[i].len;
144 used_idx += vq->shadow_used_packed[i].count;
145 if (used_idx >= vq->size)
146 used_idx -= vq->size;
149 /* The ordering for storing desc flags needs to be enforced. */
150 rte_atomic_thread_fence(__ATOMIC_RELEASE);
152 for (i = 0; i < vq->shadow_used_idx; i++) {
155 if (vq->shadow_used_packed[i].len)
156 flags = VRING_DESC_F_WRITE;
160 if (vq->used_wrap_counter) {
161 flags |= VRING_DESC_F_USED;
162 flags |= VRING_DESC_F_AVAIL;
164 flags &= ~VRING_DESC_F_USED;
165 flags &= ~VRING_DESC_F_AVAIL;
169 vq->desc_packed[vq->last_used_idx].flags = flags;
171 vhost_log_cache_used_vring(dev, vq,
173 sizeof(struct vring_packed_desc),
174 sizeof(struct vring_packed_desc));
176 head_idx = vq->last_used_idx;
180 vq_inc_last_used_packed(vq, vq->shadow_used_packed[i].count);
183 vq->desc_packed[head_idx].flags = head_flags;
185 vhost_log_cache_used_vring(dev, vq,
187 sizeof(struct vring_packed_desc),
188 sizeof(struct vring_packed_desc));
190 vq->shadow_used_idx = 0;
191 vhost_log_cache_sync(dev, vq);
194 static __rte_always_inline void
195 vhost_flush_dequeue_shadow_packed(struct virtio_net *dev,
196 struct vhost_virtqueue *vq)
198 struct vring_used_elem_packed *used_elem = &vq->shadow_used_packed[0];
200 vq->desc_packed[vq->shadow_last_used_idx].id = used_elem->id;
201 /* desc flags is the synchronization point for virtio packed vring */
202 __atomic_store_n(&vq->desc_packed[vq->shadow_last_used_idx].flags,
203 used_elem->flags, __ATOMIC_RELEASE);
205 vhost_log_cache_used_vring(dev, vq, vq->shadow_last_used_idx *
206 sizeof(struct vring_packed_desc),
207 sizeof(struct vring_packed_desc));
208 vq->shadow_used_idx = 0;
209 vhost_log_cache_sync(dev, vq);
212 static __rte_always_inline void
213 vhost_flush_enqueue_batch_packed(struct virtio_net *dev,
214 struct vhost_virtqueue *vq,
221 if (vq->shadow_used_idx) {
222 do_data_copy_enqueue(dev, vq);
223 vhost_flush_enqueue_shadow_packed(dev, vq);
226 flags = PACKED_DESC_ENQUEUE_USED_FLAG(vq->used_wrap_counter);
228 vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE) {
229 vq->desc_packed[vq->last_used_idx + i].id = ids[i];
230 vq->desc_packed[vq->last_used_idx + i].len = lens[i];
233 rte_atomic_thread_fence(__ATOMIC_RELEASE);
235 vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE)
236 vq->desc_packed[vq->last_used_idx + i].flags = flags;
238 vhost_log_cache_used_vring(dev, vq, vq->last_used_idx *
239 sizeof(struct vring_packed_desc),
240 sizeof(struct vring_packed_desc) *
242 vhost_log_cache_sync(dev, vq);
244 vq_inc_last_used_packed(vq, PACKED_BATCH_SIZE);
247 static __rte_always_inline void
248 vhost_shadow_dequeue_batch_packed_inorder(struct vhost_virtqueue *vq,
251 vq->shadow_used_packed[0].id = id;
253 if (!vq->shadow_used_idx) {
254 vq->shadow_last_used_idx = vq->last_used_idx;
255 vq->shadow_used_packed[0].flags =
256 PACKED_DESC_DEQUEUE_USED_FLAG(vq->used_wrap_counter);
257 vq->shadow_used_packed[0].len = 0;
258 vq->shadow_used_packed[0].count = 1;
259 vq->shadow_used_idx++;
262 vq_inc_last_used_packed(vq, PACKED_BATCH_SIZE);
265 static __rte_always_inline void
266 vhost_shadow_dequeue_batch_packed(struct virtio_net *dev,
267 struct vhost_virtqueue *vq,
274 flags = PACKED_DESC_DEQUEUE_USED_FLAG(vq->used_wrap_counter);
276 if (!vq->shadow_used_idx) {
277 vq->shadow_last_used_idx = vq->last_used_idx;
278 vq->shadow_used_packed[0].id = ids[0];
279 vq->shadow_used_packed[0].len = 0;
280 vq->shadow_used_packed[0].count = 1;
281 vq->shadow_used_packed[0].flags = flags;
282 vq->shadow_used_idx++;
287 vhost_for_each_try_unroll(i, begin, PACKED_BATCH_SIZE) {
288 vq->desc_packed[vq->last_used_idx + i].id = ids[i];
289 vq->desc_packed[vq->last_used_idx + i].len = 0;
292 rte_atomic_thread_fence(__ATOMIC_RELEASE);
293 vhost_for_each_try_unroll(i, begin, PACKED_BATCH_SIZE)
294 vq->desc_packed[vq->last_used_idx + i].flags = flags;
296 vhost_log_cache_used_vring(dev, vq, vq->last_used_idx *
297 sizeof(struct vring_packed_desc),
298 sizeof(struct vring_packed_desc) *
300 vhost_log_cache_sync(dev, vq);
302 vq_inc_last_used_packed(vq, PACKED_BATCH_SIZE);
305 static __rte_always_inline void
306 vhost_shadow_dequeue_single_packed(struct vhost_virtqueue *vq,
312 flags = vq->desc_packed[vq->last_used_idx].flags;
313 if (vq->used_wrap_counter) {
314 flags |= VRING_DESC_F_USED;
315 flags |= VRING_DESC_F_AVAIL;
317 flags &= ~VRING_DESC_F_USED;
318 flags &= ~VRING_DESC_F_AVAIL;
321 if (!vq->shadow_used_idx) {
322 vq->shadow_last_used_idx = vq->last_used_idx;
324 vq->shadow_used_packed[0].id = buf_id;
325 vq->shadow_used_packed[0].len = 0;
326 vq->shadow_used_packed[0].flags = flags;
327 vq->shadow_used_idx++;
329 vq->desc_packed[vq->last_used_idx].id = buf_id;
330 vq->desc_packed[vq->last_used_idx].len = 0;
331 vq->desc_packed[vq->last_used_idx].flags = flags;
334 vq_inc_last_used_packed(vq, count);
337 static __rte_always_inline void
338 vhost_shadow_dequeue_single_packed_inorder(struct vhost_virtqueue *vq,
344 vq->shadow_used_packed[0].id = buf_id;
346 flags = vq->desc_packed[vq->last_used_idx].flags;
347 if (vq->used_wrap_counter) {
348 flags |= VRING_DESC_F_USED;
349 flags |= VRING_DESC_F_AVAIL;
351 flags &= ~VRING_DESC_F_USED;
352 flags &= ~VRING_DESC_F_AVAIL;
355 if (!vq->shadow_used_idx) {
356 vq->shadow_last_used_idx = vq->last_used_idx;
357 vq->shadow_used_packed[0].len = 0;
358 vq->shadow_used_packed[0].flags = flags;
359 vq->shadow_used_idx++;
362 vq_inc_last_used_packed(vq, count);
365 static __rte_always_inline void
366 vhost_shadow_enqueue_single_packed(struct virtio_net *dev,
367 struct vhost_virtqueue *vq,
371 uint16_t num_buffers)
374 for (i = 0; i < num_buffers; i++) {
375 /* enqueue shadow flush action aligned with batch num */
376 if (!vq->shadow_used_idx)
377 vq->shadow_aligned_idx = vq->last_used_idx &
379 vq->shadow_used_packed[vq->shadow_used_idx].id = id[i];
380 vq->shadow_used_packed[vq->shadow_used_idx].len = len[i];
381 vq->shadow_used_packed[vq->shadow_used_idx].count = count[i];
382 vq->shadow_aligned_idx += count[i];
383 vq->shadow_used_idx++;
386 if (vq->shadow_aligned_idx >= PACKED_BATCH_SIZE) {
387 do_data_copy_enqueue(dev, vq);
388 vhost_flush_enqueue_shadow_packed(dev, vq);
392 /* avoid write operation when necessary, to lessen cache issues */
393 #define ASSIGN_UNLESS_EQUAL(var, val) do { \
394 if ((var) != (val)) \
398 static __rte_always_inline void
399 virtio_enqueue_offload(struct rte_mbuf *m_buf, struct virtio_net_hdr *net_hdr)
401 uint64_t csum_l4 = m_buf->ol_flags & PKT_TX_L4_MASK;
403 if (m_buf->ol_flags & PKT_TX_TCP_SEG)
404 csum_l4 |= PKT_TX_TCP_CKSUM;
407 net_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
408 net_hdr->csum_start = m_buf->l2_len + m_buf->l3_len;
411 case PKT_TX_TCP_CKSUM:
412 net_hdr->csum_offset = (offsetof(struct rte_tcp_hdr,
415 case PKT_TX_UDP_CKSUM:
416 net_hdr->csum_offset = (offsetof(struct rte_udp_hdr,
419 case PKT_TX_SCTP_CKSUM:
420 net_hdr->csum_offset = (offsetof(struct rte_sctp_hdr,
425 ASSIGN_UNLESS_EQUAL(net_hdr->csum_start, 0);
426 ASSIGN_UNLESS_EQUAL(net_hdr->csum_offset, 0);
427 ASSIGN_UNLESS_EQUAL(net_hdr->flags, 0);
430 /* IP cksum verification cannot be bypassed, then calculate here */
431 if (m_buf->ol_flags & PKT_TX_IP_CKSUM) {
432 struct rte_ipv4_hdr *ipv4_hdr;
434 ipv4_hdr = rte_pktmbuf_mtod_offset(m_buf, struct rte_ipv4_hdr *,
436 ipv4_hdr->hdr_checksum = 0;
437 ipv4_hdr->hdr_checksum = rte_ipv4_cksum(ipv4_hdr);
440 if (m_buf->ol_flags & PKT_TX_TCP_SEG) {
441 if (m_buf->ol_flags & PKT_TX_IPV4)
442 net_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
444 net_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
445 net_hdr->gso_size = m_buf->tso_segsz;
446 net_hdr->hdr_len = m_buf->l2_len + m_buf->l3_len
448 } else if (m_buf->ol_flags & PKT_TX_UDP_SEG) {
449 net_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
450 net_hdr->gso_size = m_buf->tso_segsz;
451 net_hdr->hdr_len = m_buf->l2_len + m_buf->l3_len +
454 ASSIGN_UNLESS_EQUAL(net_hdr->gso_type, 0);
455 ASSIGN_UNLESS_EQUAL(net_hdr->gso_size, 0);
456 ASSIGN_UNLESS_EQUAL(net_hdr->hdr_len, 0);
460 static __rte_always_inline int
461 map_one_desc(struct virtio_net *dev, struct vhost_virtqueue *vq,
462 struct buf_vector *buf_vec, uint16_t *vec_idx,
463 uint64_t desc_iova, uint64_t desc_len, uint8_t perm)
465 uint16_t vec_id = *vec_idx;
469 uint64_t desc_chunck_len = desc_len;
471 if (unlikely(vec_id >= BUF_VECTOR_MAX))
474 desc_addr = vhost_iova_to_vva(dev, vq,
478 if (unlikely(!desc_addr))
481 rte_prefetch0((void *)(uintptr_t)desc_addr);
483 buf_vec[vec_id].buf_iova = desc_iova;
484 buf_vec[vec_id].buf_addr = desc_addr;
485 buf_vec[vec_id].buf_len = desc_chunck_len;
487 desc_len -= desc_chunck_len;
488 desc_iova += desc_chunck_len;
496 static __rte_always_inline int
497 fill_vec_buf_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
498 uint32_t avail_idx, uint16_t *vec_idx,
499 struct buf_vector *buf_vec, uint16_t *desc_chain_head,
500 uint32_t *desc_chain_len, uint8_t perm)
502 uint16_t idx = vq->avail->ring[avail_idx & (vq->size - 1)];
503 uint16_t vec_id = *vec_idx;
506 uint32_t nr_descs = vq->size;
508 struct vring_desc *descs = vq->desc;
509 struct vring_desc *idesc = NULL;
511 if (unlikely(idx >= vq->size))
514 *desc_chain_head = idx;
516 if (vq->desc[idx].flags & VRING_DESC_F_INDIRECT) {
517 dlen = vq->desc[idx].len;
518 nr_descs = dlen / sizeof(struct vring_desc);
519 if (unlikely(nr_descs > vq->size))
522 descs = (struct vring_desc *)(uintptr_t)
523 vhost_iova_to_vva(dev, vq, vq->desc[idx].addr,
526 if (unlikely(!descs))
529 if (unlikely(dlen < vq->desc[idx].len)) {
531 * The indirect desc table is not contiguous
532 * in process VA space, we have to copy it.
534 idesc = vhost_alloc_copy_ind_table(dev, vq,
535 vq->desc[idx].addr, vq->desc[idx].len);
536 if (unlikely(!idesc))
546 if (unlikely(idx >= nr_descs || cnt++ >= nr_descs)) {
547 free_ind_table(idesc);
551 dlen = descs[idx].len;
554 if (unlikely(map_one_desc(dev, vq, buf_vec, &vec_id,
555 descs[idx].addr, dlen,
557 free_ind_table(idesc);
561 if ((descs[idx].flags & VRING_DESC_F_NEXT) == 0)
564 idx = descs[idx].next;
567 *desc_chain_len = len;
570 if (unlikely(!!idesc))
571 free_ind_table(idesc);
577 * Returns -1 on fail, 0 on success
580 reserve_avail_buf_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
581 uint32_t size, struct buf_vector *buf_vec,
582 uint16_t *num_buffers, uint16_t avail_head,
586 uint16_t vec_idx = 0;
587 uint16_t max_tries, tries = 0;
589 uint16_t head_idx = 0;
593 cur_idx = vq->last_avail_idx;
595 if (rxvq_is_mergeable(dev))
596 max_tries = vq->size - 1;
601 if (unlikely(cur_idx == avail_head))
604 * if we tried all available ring items, and still
605 * can't get enough buf, it means something abnormal
608 if (unlikely(++tries > max_tries))
611 if (unlikely(fill_vec_buf_split(dev, vq, cur_idx,
614 VHOST_ACCESS_RW) < 0))
616 len = RTE_MIN(len, size);
617 update_shadow_used_ring_split(vq, head_idx, len);
629 static __rte_always_inline int
630 fill_vec_buf_packed_indirect(struct virtio_net *dev,
631 struct vhost_virtqueue *vq,
632 struct vring_packed_desc *desc, uint16_t *vec_idx,
633 struct buf_vector *buf_vec, uint32_t *len, uint8_t perm)
637 uint16_t vec_id = *vec_idx;
639 struct vring_packed_desc *descs, *idescs = NULL;
642 descs = (struct vring_packed_desc *)(uintptr_t)
643 vhost_iova_to_vva(dev, vq, desc->addr, &dlen, VHOST_ACCESS_RO);
644 if (unlikely(!descs))
647 if (unlikely(dlen < desc->len)) {
649 * The indirect desc table is not contiguous
650 * in process VA space, we have to copy it.
652 idescs = vhost_alloc_copy_ind_table(dev,
653 vq, desc->addr, desc->len);
654 if (unlikely(!idescs))
660 nr_descs = desc->len / sizeof(struct vring_packed_desc);
661 if (unlikely(nr_descs >= vq->size)) {
662 free_ind_table(idescs);
666 for (i = 0; i < nr_descs; i++) {
667 if (unlikely(vec_id >= BUF_VECTOR_MAX)) {
668 free_ind_table(idescs);
674 if (unlikely(map_one_desc(dev, vq, buf_vec, &vec_id,
681 if (unlikely(!!idescs))
682 free_ind_table(idescs);
687 static __rte_always_inline int
688 fill_vec_buf_packed(struct virtio_net *dev, struct vhost_virtqueue *vq,
689 uint16_t avail_idx, uint16_t *desc_count,
690 struct buf_vector *buf_vec, uint16_t *vec_idx,
691 uint16_t *buf_id, uint32_t *len, uint8_t perm)
693 bool wrap_counter = vq->avail_wrap_counter;
694 struct vring_packed_desc *descs = vq->desc_packed;
695 uint16_t vec_id = *vec_idx;
698 if (avail_idx < vq->last_avail_idx)
702 * Perform a load-acquire barrier in desc_is_avail to
703 * enforce the ordering between desc flags and desc
706 if (unlikely(!desc_is_avail(&descs[avail_idx], wrap_counter)))
713 if (unlikely(vec_id >= BUF_VECTOR_MAX))
716 if (unlikely(*desc_count >= vq->size))
720 *buf_id = descs[avail_idx].id;
722 if (descs[avail_idx].flags & VRING_DESC_F_INDIRECT) {
723 if (unlikely(fill_vec_buf_packed_indirect(dev, vq,
729 dlen = descs[avail_idx].len;
732 if (unlikely(map_one_desc(dev, vq, buf_vec, &vec_id,
733 descs[avail_idx].addr,
739 if ((descs[avail_idx].flags & VRING_DESC_F_NEXT) == 0)
742 if (++avail_idx >= vq->size) {
743 avail_idx -= vq->size;
753 static __rte_noinline void
754 copy_vnet_hdr_to_desc(struct virtio_net *dev, struct vhost_virtqueue *vq,
755 struct buf_vector *buf_vec,
756 struct virtio_net_hdr_mrg_rxbuf *hdr)
759 uint64_t remain = dev->vhost_hlen;
760 uint64_t src = (uint64_t)(uintptr_t)hdr, dst;
761 uint64_t iova = buf_vec->buf_iova;
764 len = RTE_MIN(remain,
766 dst = buf_vec->buf_addr;
767 rte_memcpy((void *)(uintptr_t)dst,
768 (void *)(uintptr_t)src,
771 PRINT_PACKET(dev, (uintptr_t)dst,
773 vhost_log_cache_write_iova(dev, vq,
783 static __rte_always_inline int
784 copy_mbuf_to_desc(struct virtio_net *dev, struct vhost_virtqueue *vq,
785 struct rte_mbuf *m, struct buf_vector *buf_vec,
786 uint16_t nr_vec, uint16_t num_buffers)
788 uint32_t vec_idx = 0;
789 uint32_t mbuf_offset, mbuf_avail;
790 uint32_t buf_offset, buf_avail;
791 uint64_t buf_addr, buf_iova, buf_len;
794 struct rte_mbuf *hdr_mbuf;
795 struct batch_copy_elem *batch_copy = vq->batch_copy_elems;
796 struct virtio_net_hdr_mrg_rxbuf tmp_hdr, *hdr = NULL;
799 if (unlikely(m == NULL)) {
804 buf_addr = buf_vec[vec_idx].buf_addr;
805 buf_iova = buf_vec[vec_idx].buf_iova;
806 buf_len = buf_vec[vec_idx].buf_len;
808 if (unlikely(buf_len < dev->vhost_hlen && nr_vec <= 1)) {
815 if (unlikely(buf_len < dev->vhost_hlen)) {
816 memset(&tmp_hdr, 0, sizeof(struct virtio_net_hdr_mrg_rxbuf));
819 hdr = (struct virtio_net_hdr_mrg_rxbuf *)(uintptr_t)hdr_addr;
821 VHOST_LOG_DATA(DEBUG, "(%d) RX: num merge buffers %d\n",
822 dev->vid, num_buffers);
824 if (unlikely(buf_len < dev->vhost_hlen)) {
825 buf_offset = dev->vhost_hlen - buf_len;
827 buf_addr = buf_vec[vec_idx].buf_addr;
828 buf_iova = buf_vec[vec_idx].buf_iova;
829 buf_len = buf_vec[vec_idx].buf_len;
830 buf_avail = buf_len - buf_offset;
832 buf_offset = dev->vhost_hlen;
833 buf_avail = buf_len - dev->vhost_hlen;
836 mbuf_avail = rte_pktmbuf_data_len(m);
838 while (mbuf_avail != 0 || m->next != NULL) {
839 /* done with current buf, get the next one */
840 if (buf_avail == 0) {
842 if (unlikely(vec_idx >= nr_vec)) {
847 buf_addr = buf_vec[vec_idx].buf_addr;
848 buf_iova = buf_vec[vec_idx].buf_iova;
849 buf_len = buf_vec[vec_idx].buf_len;
855 /* done with current mbuf, get the next one */
856 if (mbuf_avail == 0) {
860 mbuf_avail = rte_pktmbuf_data_len(m);
864 virtio_enqueue_offload(hdr_mbuf, &hdr->hdr);
865 if (rxvq_is_mergeable(dev))
866 ASSIGN_UNLESS_EQUAL(hdr->num_buffers,
869 if (unlikely(hdr == &tmp_hdr)) {
870 copy_vnet_hdr_to_desc(dev, vq, buf_vec, hdr);
872 PRINT_PACKET(dev, (uintptr_t)hdr_addr,
874 vhost_log_cache_write_iova(dev, vq,
882 cpy_len = RTE_MIN(buf_avail, mbuf_avail);
884 if (likely(cpy_len > MAX_BATCH_LEN ||
885 vq->batch_copy_nb_elems >= vq->size)) {
886 rte_memcpy((void *)((uintptr_t)(buf_addr + buf_offset)),
887 rte_pktmbuf_mtod_offset(m, void *, mbuf_offset),
889 vhost_log_cache_write_iova(dev, vq,
890 buf_iova + buf_offset,
892 PRINT_PACKET(dev, (uintptr_t)(buf_addr + buf_offset),
895 batch_copy[vq->batch_copy_nb_elems].dst =
896 (void *)((uintptr_t)(buf_addr + buf_offset));
897 batch_copy[vq->batch_copy_nb_elems].src =
898 rte_pktmbuf_mtod_offset(m, void *, mbuf_offset);
899 batch_copy[vq->batch_copy_nb_elems].log_addr =
900 buf_iova + buf_offset;
901 batch_copy[vq->batch_copy_nb_elems].len = cpy_len;
902 vq->batch_copy_nb_elems++;
905 mbuf_avail -= cpy_len;
906 mbuf_offset += cpy_len;
907 buf_avail -= cpy_len;
908 buf_offset += cpy_len;
916 static __rte_always_inline void
917 async_fill_vec(struct iovec *v, void *base, size_t len)
923 static __rte_always_inline void
924 async_fill_iter(struct rte_vhost_iov_iter *it, size_t count,
925 struct iovec *vec, unsigned long nr_seg)
932 it->nr_segs = nr_seg;
939 static __rte_always_inline void
940 async_fill_desc(struct rte_vhost_async_desc *desc,
941 struct rte_vhost_iov_iter *src, struct rte_vhost_iov_iter *dst)
947 static __rte_always_inline int
948 async_mbuf_to_desc(struct virtio_net *dev, struct vhost_virtqueue *vq,
949 struct rte_mbuf *m, struct buf_vector *buf_vec,
950 uint16_t nr_vec, uint16_t num_buffers,
951 struct iovec *src_iovec, struct iovec *dst_iovec,
952 struct rte_vhost_iov_iter *src_it,
953 struct rte_vhost_iov_iter *dst_it)
955 uint32_t vec_idx = 0;
956 uint32_t mbuf_offset, mbuf_avail;
957 uint32_t buf_offset, buf_avail;
958 uint64_t buf_addr, buf_iova, buf_len;
959 uint32_t cpy_len, cpy_threshold;
961 struct rte_mbuf *hdr_mbuf;
962 struct batch_copy_elem *batch_copy = vq->batch_copy_elems;
963 struct virtio_net_hdr_mrg_rxbuf tmp_hdr, *hdr = NULL;
971 if (unlikely(m == NULL)) {
976 cpy_threshold = vq->async_threshold;
978 buf_addr = buf_vec[vec_idx].buf_addr;
979 buf_iova = buf_vec[vec_idx].buf_iova;
980 buf_len = buf_vec[vec_idx].buf_len;
982 if (unlikely(buf_len < dev->vhost_hlen && nr_vec <= 1)) {
989 if (unlikely(buf_len < dev->vhost_hlen)) {
990 memset(&tmp_hdr, 0, sizeof(struct virtio_net_hdr_mrg_rxbuf));
993 hdr = (struct virtio_net_hdr_mrg_rxbuf *)(uintptr_t)hdr_addr;
995 VHOST_LOG_DATA(DEBUG, "(%d) RX: num merge buffers %d\n",
996 dev->vid, num_buffers);
998 if (unlikely(buf_len < dev->vhost_hlen)) {
999 buf_offset = dev->vhost_hlen - buf_len;
1001 buf_addr = buf_vec[vec_idx].buf_addr;
1002 buf_iova = buf_vec[vec_idx].buf_iova;
1003 buf_len = buf_vec[vec_idx].buf_len;
1004 buf_avail = buf_len - buf_offset;
1006 buf_offset = dev->vhost_hlen;
1007 buf_avail = buf_len - dev->vhost_hlen;
1010 mbuf_avail = rte_pktmbuf_data_len(m);
1013 while (mbuf_avail != 0 || m->next != NULL) {
1014 /* done with current buf, get the next one */
1015 if (buf_avail == 0) {
1017 if (unlikely(vec_idx >= nr_vec)) {
1022 buf_addr = buf_vec[vec_idx].buf_addr;
1023 buf_iova = buf_vec[vec_idx].buf_iova;
1024 buf_len = buf_vec[vec_idx].buf_len;
1027 buf_avail = buf_len;
1030 /* done with current mbuf, get the next one */
1031 if (mbuf_avail == 0) {
1035 mbuf_avail = rte_pktmbuf_data_len(m);
1039 virtio_enqueue_offload(hdr_mbuf, &hdr->hdr);
1040 if (rxvq_is_mergeable(dev))
1041 ASSIGN_UNLESS_EQUAL(hdr->num_buffers,
1044 if (unlikely(hdr == &tmp_hdr)) {
1045 copy_vnet_hdr_to_desc(dev, vq, buf_vec, hdr);
1047 PRINT_PACKET(dev, (uintptr_t)hdr_addr,
1048 dev->vhost_hlen, 0);
1049 vhost_log_cache_write_iova(dev, vq,
1050 buf_vec[0].buf_iova,
1057 cpy_len = RTE_MIN(buf_avail, mbuf_avail);
1059 while (unlikely(cpy_len && cpy_len >= cpy_threshold)) {
1060 hpa = (void *)(uintptr_t)gpa_to_first_hpa(dev,
1061 buf_iova + buf_offset,
1062 cpy_len, &mapped_len);
1064 if (unlikely(!hpa || mapped_len < cpy_threshold))
1067 async_fill_vec(src_iovec + tvec_idx,
1068 (void *)(uintptr_t)rte_pktmbuf_iova_offset(m,
1069 mbuf_offset), (size_t)mapped_len);
1071 async_fill_vec(dst_iovec + tvec_idx,
1072 hpa, (size_t)mapped_len);
1074 tlen += (uint32_t)mapped_len;
1075 cpy_len -= (uint32_t)mapped_len;
1076 mbuf_avail -= (uint32_t)mapped_len;
1077 mbuf_offset += (uint32_t)mapped_len;
1078 buf_avail -= (uint32_t)mapped_len;
1079 buf_offset += (uint32_t)mapped_len;
1083 if (likely(cpy_len)) {
1084 if (unlikely(vq->batch_copy_nb_elems >= vq->size)) {
1086 (void *)((uintptr_t)(buf_addr + buf_offset)),
1087 rte_pktmbuf_mtod_offset(m, void *, mbuf_offset),
1091 (uintptr_t)(buf_addr + buf_offset),
1094 batch_copy[vq->batch_copy_nb_elems].dst =
1095 (void *)((uintptr_t)(buf_addr + buf_offset));
1096 batch_copy[vq->batch_copy_nb_elems].src =
1097 rte_pktmbuf_mtod_offset(m, void *, mbuf_offset);
1098 batch_copy[vq->batch_copy_nb_elems].log_addr =
1099 buf_iova + buf_offset;
1100 batch_copy[vq->batch_copy_nb_elems].len =
1102 vq->batch_copy_nb_elems++;
1105 mbuf_avail -= cpy_len;
1106 mbuf_offset += cpy_len;
1107 buf_avail -= cpy_len;
1108 buf_offset += cpy_len;
1115 async_fill_iter(src_it, tlen, src_iovec, tvec_idx);
1116 async_fill_iter(dst_it, tlen, dst_iovec, tvec_idx);
1124 static __rte_always_inline int
1125 vhost_enqueue_single_packed(struct virtio_net *dev,
1126 struct vhost_virtqueue *vq,
1127 struct rte_mbuf *pkt,
1128 struct buf_vector *buf_vec,
1131 uint16_t nr_vec = 0;
1132 uint16_t avail_idx = vq->last_avail_idx;
1133 uint16_t max_tries, tries = 0;
1134 uint16_t buf_id = 0;
1136 uint16_t desc_count;
1137 uint32_t size = pkt->pkt_len + sizeof(struct virtio_net_hdr_mrg_rxbuf);
1138 uint16_t num_buffers = 0;
1139 uint32_t buffer_len[vq->size];
1140 uint16_t buffer_buf_id[vq->size];
1141 uint16_t buffer_desc_count[vq->size];
1143 if (rxvq_is_mergeable(dev))
1144 max_tries = vq->size - 1;
1150 * if we tried all available ring items, and still
1151 * can't get enough buf, it means something abnormal
1154 if (unlikely(++tries > max_tries))
1157 if (unlikely(fill_vec_buf_packed(dev, vq,
1158 avail_idx, &desc_count,
1161 VHOST_ACCESS_RW) < 0))
1164 len = RTE_MIN(len, size);
1167 buffer_len[num_buffers] = len;
1168 buffer_buf_id[num_buffers] = buf_id;
1169 buffer_desc_count[num_buffers] = desc_count;
1172 *nr_descs += desc_count;
1173 avail_idx += desc_count;
1174 if (avail_idx >= vq->size)
1175 avail_idx -= vq->size;
1178 if (copy_mbuf_to_desc(dev, vq, pkt, buf_vec, nr_vec, num_buffers) < 0)
1181 vhost_shadow_enqueue_single_packed(dev, vq, buffer_len, buffer_buf_id,
1182 buffer_desc_count, num_buffers);
1187 static __rte_noinline uint32_t
1188 virtio_dev_rx_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
1189 struct rte_mbuf **pkts, uint32_t count)
1191 uint32_t pkt_idx = 0;
1192 uint16_t num_buffers;
1193 struct buf_vector buf_vec[BUF_VECTOR_MAX];
1194 uint16_t avail_head;
1197 * The ordering between avail index and
1198 * desc reads needs to be enforced.
1200 avail_head = __atomic_load_n(&vq->avail->idx, __ATOMIC_ACQUIRE);
1202 rte_prefetch0(&vq->avail->ring[vq->last_avail_idx & (vq->size - 1)]);
1204 for (pkt_idx = 0; pkt_idx < count; pkt_idx++) {
1205 uint32_t pkt_len = pkts[pkt_idx]->pkt_len + dev->vhost_hlen;
1206 uint16_t nr_vec = 0;
1208 if (unlikely(reserve_avail_buf_split(dev, vq,
1209 pkt_len, buf_vec, &num_buffers,
1210 avail_head, &nr_vec) < 0)) {
1211 VHOST_LOG_DATA(DEBUG,
1212 "(%d) failed to get enough desc from vring\n",
1214 vq->shadow_used_idx -= num_buffers;
1218 VHOST_LOG_DATA(DEBUG, "(%d) current index %d | end index %d\n",
1219 dev->vid, vq->last_avail_idx,
1220 vq->last_avail_idx + num_buffers);
1222 if (copy_mbuf_to_desc(dev, vq, pkts[pkt_idx],
1225 vq->shadow_used_idx -= num_buffers;
1229 vq->last_avail_idx += num_buffers;
1232 do_data_copy_enqueue(dev, vq);
1234 if (likely(vq->shadow_used_idx)) {
1235 flush_shadow_used_ring_split(dev, vq);
1236 vhost_vring_call_split(dev, vq);
1242 static __rte_always_inline int
1243 virtio_dev_rx_batch_packed(struct virtio_net *dev,
1244 struct vhost_virtqueue *vq,
1245 struct rte_mbuf **pkts)
1247 bool wrap_counter = vq->avail_wrap_counter;
1248 struct vring_packed_desc *descs = vq->desc_packed;
1249 uint16_t avail_idx = vq->last_avail_idx;
1250 uint64_t desc_addrs[PACKED_BATCH_SIZE];
1251 struct virtio_net_hdr_mrg_rxbuf *hdrs[PACKED_BATCH_SIZE];
1252 uint32_t buf_offset = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1253 uint64_t lens[PACKED_BATCH_SIZE];
1254 uint16_t ids[PACKED_BATCH_SIZE];
1257 if (unlikely(avail_idx & PACKED_BATCH_MASK))
1260 if (unlikely((avail_idx + PACKED_BATCH_SIZE) > vq->size))
1263 vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE) {
1264 if (unlikely(pkts[i]->next != NULL))
1266 if (unlikely(!desc_is_avail(&descs[avail_idx + i],
1271 vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE)
1272 lens[i] = descs[avail_idx + i].len;
1274 vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE) {
1275 if (unlikely(pkts[i]->pkt_len > (lens[i] - buf_offset)))
1279 vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE)
1280 desc_addrs[i] = vhost_iova_to_vva(dev, vq,
1281 descs[avail_idx + i].addr,
1285 vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE) {
1286 if (unlikely(!desc_addrs[i]))
1288 if (unlikely(lens[i] != descs[avail_idx + i].len))
1292 vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE) {
1293 rte_prefetch0((void *)(uintptr_t)desc_addrs[i]);
1294 hdrs[i] = (struct virtio_net_hdr_mrg_rxbuf *)
1295 (uintptr_t)desc_addrs[i];
1296 lens[i] = pkts[i]->pkt_len +
1297 sizeof(struct virtio_net_hdr_mrg_rxbuf);
1300 vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE)
1301 virtio_enqueue_offload(pkts[i], &hdrs[i]->hdr);
1303 vq_inc_last_avail_packed(vq, PACKED_BATCH_SIZE);
1305 vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE) {
1306 rte_memcpy((void *)(uintptr_t)(desc_addrs[i] + buf_offset),
1307 rte_pktmbuf_mtod_offset(pkts[i], void *, 0),
1311 vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE)
1312 vhost_log_cache_write_iova(dev, vq, descs[avail_idx + i].addr,
1315 vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE)
1316 ids[i] = descs[avail_idx + i].id;
1318 vhost_flush_enqueue_batch_packed(dev, vq, lens, ids);
1323 static __rte_always_inline int16_t
1324 virtio_dev_rx_single_packed(struct virtio_net *dev,
1325 struct vhost_virtqueue *vq,
1326 struct rte_mbuf *pkt)
1328 struct buf_vector buf_vec[BUF_VECTOR_MAX];
1329 uint16_t nr_descs = 0;
1331 if (unlikely(vhost_enqueue_single_packed(dev, vq, pkt, buf_vec,
1333 VHOST_LOG_DATA(DEBUG,
1334 "(%d) failed to get enough desc from vring\n",
1339 VHOST_LOG_DATA(DEBUG, "(%d) current index %d | end index %d\n",
1340 dev->vid, vq->last_avail_idx,
1341 vq->last_avail_idx + nr_descs);
1343 vq_inc_last_avail_packed(vq, nr_descs);
1348 static __rte_noinline uint32_t
1349 virtio_dev_rx_packed(struct virtio_net *dev,
1350 struct vhost_virtqueue *__rte_restrict vq,
1351 struct rte_mbuf **__rte_restrict pkts,
1354 uint32_t pkt_idx = 0;
1355 uint32_t remained = count;
1358 rte_prefetch0(&vq->desc_packed[vq->last_avail_idx]);
1360 if (remained >= PACKED_BATCH_SIZE) {
1361 if (!virtio_dev_rx_batch_packed(dev, vq,
1363 pkt_idx += PACKED_BATCH_SIZE;
1364 remained -= PACKED_BATCH_SIZE;
1369 if (virtio_dev_rx_single_packed(dev, vq, pkts[pkt_idx]))
1374 } while (pkt_idx < count);
1376 if (vq->shadow_used_idx) {
1377 do_data_copy_enqueue(dev, vq);
1378 vhost_flush_enqueue_shadow_packed(dev, vq);
1382 vhost_vring_call_packed(dev, vq);
1387 static __rte_always_inline uint32_t
1388 virtio_dev_rx(struct virtio_net *dev, uint16_t queue_id,
1389 struct rte_mbuf **pkts, uint32_t count)
1391 struct vhost_virtqueue *vq;
1394 VHOST_LOG_DATA(DEBUG, "(%d) %s\n", dev->vid, __func__);
1395 if (unlikely(!is_valid_virt_queue_idx(queue_id, 0, dev->nr_vring))) {
1396 VHOST_LOG_DATA(ERR, "(%d) %s: invalid virtqueue idx %d.\n",
1397 dev->vid, __func__, queue_id);
1401 vq = dev->virtqueue[queue_id];
1403 rte_spinlock_lock(&vq->access_lock);
1405 if (unlikely(!vq->enabled))
1406 goto out_access_unlock;
1408 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
1409 vhost_user_iotlb_rd_lock(vq);
1411 if (unlikely(!vq->access_ok))
1412 if (unlikely(vring_translate(dev, vq) < 0))
1415 count = RTE_MIN((uint32_t)MAX_PKT_BURST, count);
1419 if (vq_is_packed(dev))
1420 nb_tx = virtio_dev_rx_packed(dev, vq, pkts, count);
1422 nb_tx = virtio_dev_rx_split(dev, vq, pkts, count);
1425 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
1426 vhost_user_iotlb_rd_unlock(vq);
1429 rte_spinlock_unlock(&vq->access_lock);
1435 rte_vhost_enqueue_burst(int vid, uint16_t queue_id,
1436 struct rte_mbuf **__rte_restrict pkts, uint16_t count)
1438 struct virtio_net *dev = get_device(vid);
1443 if (unlikely(!(dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET))) {
1445 "(%d) %s: built-in vhost net backend is disabled.\n",
1446 dev->vid, __func__);
1450 return virtio_dev_rx(dev, queue_id, pkts, count);
1453 static __rte_always_inline uint16_t
1454 virtio_dev_rx_async_get_info_idx(uint16_t pkts_idx,
1455 uint16_t vq_size, uint16_t n_inflight)
1457 return pkts_idx > n_inflight ? (pkts_idx - n_inflight) :
1458 (vq_size - n_inflight + pkts_idx) & (vq_size - 1);
1461 static __rte_noinline uint32_t
1462 virtio_dev_rx_async_submit_split(struct virtio_net *dev,
1463 struct vhost_virtqueue *vq, uint16_t queue_id,
1464 struct rte_mbuf **pkts, uint32_t count,
1465 struct rte_mbuf **comp_pkts, uint32_t *comp_count)
1467 uint32_t pkt_idx = 0, pkt_burst_idx = 0;
1468 uint16_t num_buffers;
1469 struct buf_vector buf_vec[BUF_VECTOR_MAX];
1470 uint16_t avail_head;
1472 struct rte_vhost_iov_iter *it_pool = vq->it_pool;
1473 struct iovec *vec_pool = vq->vec_pool;
1474 struct rte_vhost_async_desc tdes[MAX_PKT_BURST];
1475 struct iovec *src_iovec = vec_pool;
1476 struct iovec *dst_iovec = vec_pool + (VHOST_MAX_ASYNC_VEC >> 1);
1477 struct rte_vhost_iov_iter *src_it = it_pool;
1478 struct rte_vhost_iov_iter *dst_it = it_pool + 1;
1479 uint16_t slot_idx = 0;
1480 uint16_t segs_await = 0;
1481 struct async_inflight_info *pkts_info = vq->async_pkts_info;
1482 uint32_t n_pkts = 0, pkt_err = 0;
1483 uint32_t num_async_pkts = 0, num_done_pkts = 0;
1486 uint16_t last_avail_idx;
1487 } async_pkts_log[MAX_PKT_BURST];
1490 * The ordering between avail index and desc reads need to be enforced.
1492 avail_head = __atomic_load_n(&vq->avail->idx, __ATOMIC_ACQUIRE);
1494 rte_prefetch0(&vq->avail->ring[vq->last_avail_idx & (vq->size - 1)]);
1496 for (pkt_idx = 0; pkt_idx < count; pkt_idx++) {
1497 uint32_t pkt_len = pkts[pkt_idx]->pkt_len + dev->vhost_hlen;
1498 uint16_t nr_vec = 0;
1500 if (unlikely(reserve_avail_buf_split(dev, vq,
1501 pkt_len, buf_vec, &num_buffers,
1502 avail_head, &nr_vec) < 0)) {
1503 VHOST_LOG_DATA(DEBUG,
1504 "(%d) failed to get enough desc from vring\n",
1506 vq->shadow_used_idx -= num_buffers;
1510 VHOST_LOG_DATA(DEBUG, "(%d) current index %d | end index %d\n",
1511 dev->vid, vq->last_avail_idx,
1512 vq->last_avail_idx + num_buffers);
1514 if (async_mbuf_to_desc(dev, vq, pkts[pkt_idx],
1515 buf_vec, nr_vec, num_buffers,
1516 src_iovec, dst_iovec, src_it, dst_it) < 0) {
1517 vq->shadow_used_idx -= num_buffers;
1521 slot_idx = (vq->async_pkts_idx + num_async_pkts) &
1523 if (src_it->count) {
1526 async_fill_desc(&tdes[pkt_burst_idx++], src_it, dst_it);
1527 pkts_info[slot_idx].descs = num_buffers;
1528 pkts_info[slot_idx].mbuf = pkts[pkt_idx];
1529 async_pkts_log[num_async_pkts].pkt_idx = pkt_idx;
1530 async_pkts_log[num_async_pkts++].last_avail_idx =
1532 src_iovec += src_it->nr_segs;
1533 dst_iovec += dst_it->nr_segs;
1536 segs_await += src_it->nr_segs;
1539 * recover shadow used ring and keep DMA-occupied
1542 from = vq->shadow_used_idx - num_buffers;
1543 to = vq->async_desc_idx & (vq->size - 1);
1544 if (num_buffers + to <= vq->size) {
1545 rte_memcpy(&vq->async_descs_split[to],
1546 &vq->shadow_used_split[from],
1548 sizeof(struct vring_used_elem));
1550 int size = vq->size - to;
1552 rte_memcpy(&vq->async_descs_split[to],
1553 &vq->shadow_used_split[from],
1555 sizeof(struct vring_used_elem));
1556 rte_memcpy(vq->async_descs_split,
1557 &vq->shadow_used_split[from +
1558 size], (num_buffers - size) *
1559 sizeof(struct vring_used_elem));
1561 vq->async_desc_idx += num_buffers;
1562 vq->shadow_used_idx -= num_buffers;
1564 comp_pkts[num_done_pkts++] = pkts[pkt_idx];
1566 vq->last_avail_idx += num_buffers;
1569 * conditions to trigger async device transfer:
1570 * - buffered packet number reaches transfer threshold
1571 * - unused async iov number is less than max vhost vector
1573 if (unlikely(pkt_burst_idx >= VHOST_ASYNC_BATCH_THRESHOLD ||
1574 ((VHOST_MAX_ASYNC_VEC >> 1) - segs_await <
1576 n_pkts = vq->async_ops.transfer_data(dev->vid,
1577 queue_id, tdes, 0, pkt_burst_idx);
1578 src_iovec = vec_pool;
1579 dst_iovec = vec_pool + (VHOST_MAX_ASYNC_VEC >> 1);
1581 dst_it = it_pool + 1;
1583 vq->async_pkts_inflight_n += n_pkts;
1585 if (unlikely(n_pkts < pkt_burst_idx)) {
1587 * log error packets number here and do actual
1588 * error processing when applications poll
1591 pkt_err = pkt_burst_idx - n_pkts;
1600 if (pkt_burst_idx) {
1601 n_pkts = vq->async_ops.transfer_data(dev->vid,
1602 queue_id, tdes, 0, pkt_burst_idx);
1603 vq->async_pkts_inflight_n += n_pkts;
1605 if (unlikely(n_pkts < pkt_burst_idx))
1606 pkt_err = pkt_burst_idx - n_pkts;
1609 do_data_copy_enqueue(dev, vq);
1611 if (unlikely(pkt_err)) {
1612 uint16_t num_descs = 0;
1614 num_async_pkts -= pkt_err;
1615 /* calculate the sum of descriptors of DMA-error packets. */
1616 while (pkt_err-- > 0) {
1617 num_descs += pkts_info[slot_idx & (vq->size - 1)].descs;
1620 vq->async_desc_idx -= num_descs;
1621 /* recover shadow used ring and available ring */
1622 vq->shadow_used_idx -= (vq->last_avail_idx -
1623 async_pkts_log[num_async_pkts].last_avail_idx -
1625 vq->last_avail_idx =
1626 async_pkts_log[num_async_pkts].last_avail_idx;
1627 pkt_idx = async_pkts_log[num_async_pkts].pkt_idx;
1628 num_done_pkts = pkt_idx - num_async_pkts;
1631 vq->async_pkts_idx += num_async_pkts;
1632 *comp_count = num_done_pkts;
1634 if (likely(vq->shadow_used_idx)) {
1635 flush_shadow_used_ring_split(dev, vq);
1636 vhost_vring_call_split(dev, vq);
1642 uint16_t rte_vhost_poll_enqueue_completed(int vid, uint16_t queue_id,
1643 struct rte_mbuf **pkts, uint16_t count)
1645 struct virtio_net *dev = get_device(vid);
1646 struct vhost_virtqueue *vq;
1647 uint16_t n_pkts_cpl = 0, n_pkts_put = 0, n_descs = 0;
1648 uint16_t start_idx, pkts_idx, vq_size;
1649 struct async_inflight_info *pkts_info;
1655 VHOST_LOG_DATA(DEBUG, "(%d) %s\n", dev->vid, __func__);
1656 if (unlikely(!is_valid_virt_queue_idx(queue_id, 0, dev->nr_vring))) {
1657 VHOST_LOG_DATA(ERR, "(%d) %s: invalid virtqueue idx %d.\n",
1658 dev->vid, __func__, queue_id);
1662 vq = dev->virtqueue[queue_id];
1664 if (unlikely(!vq->async_registered)) {
1665 VHOST_LOG_DATA(ERR, "(%d) %s: async not registered for queue id %d.\n",
1666 dev->vid, __func__, queue_id);
1670 rte_spinlock_lock(&vq->access_lock);
1672 pkts_idx = vq->async_pkts_idx & (vq->size - 1);
1673 pkts_info = vq->async_pkts_info;
1675 start_idx = virtio_dev_rx_async_get_info_idx(pkts_idx,
1676 vq_size, vq->async_pkts_inflight_n);
1678 if (count > vq->async_last_pkts_n)
1679 n_pkts_cpl = vq->async_ops.check_completed_copies(vid,
1680 queue_id, 0, count - vq->async_last_pkts_n);
1681 n_pkts_cpl += vq->async_last_pkts_n;
1683 n_pkts_put = RTE_MIN(count, n_pkts_cpl);
1684 if (unlikely(n_pkts_put == 0)) {
1685 vq->async_last_pkts_n = n_pkts_cpl;
1689 for (i = 0; i < n_pkts_put; i++) {
1690 from = (start_idx + i) & (vq_size - 1);
1691 n_descs += pkts_info[from].descs;
1692 pkts[i] = pkts_info[from].mbuf;
1694 vq->async_last_pkts_n = n_pkts_cpl - n_pkts_put;
1695 vq->async_pkts_inflight_n -= n_pkts_put;
1697 if (likely(vq->enabled && vq->access_ok)) {
1698 uint16_t nr_left = n_descs;
1702 /* write back completed descriptors to used ring */
1704 from = vq->last_async_desc_idx & (vq->size - 1);
1705 nr_copy = nr_left + from <= vq->size ? nr_left :
1707 to = vq->last_used_idx & (vq->size - 1);
1709 if (to + nr_copy <= vq->size) {
1710 rte_memcpy(&vq->used->ring[to],
1711 &vq->async_descs_split[from],
1713 sizeof(struct vring_used_elem));
1715 uint16_t size = vq->size - to;
1717 rte_memcpy(&vq->used->ring[to],
1718 &vq->async_descs_split[from],
1720 sizeof(struct vring_used_elem));
1721 rte_memcpy(vq->used->ring,
1722 &vq->async_descs_split[from +
1723 size], (nr_copy - size) *
1724 sizeof(struct vring_used_elem));
1727 vq->last_async_desc_idx += nr_copy;
1728 vq->last_used_idx += nr_copy;
1730 } while (nr_left > 0);
1732 __atomic_add_fetch(&vq->used->idx, n_descs, __ATOMIC_RELEASE);
1733 vhost_vring_call_split(dev, vq);
1735 vq->last_async_desc_idx += n_descs;
1738 rte_spinlock_unlock(&vq->access_lock);
1743 static __rte_always_inline uint32_t
1744 virtio_dev_rx_async_submit(struct virtio_net *dev, uint16_t queue_id,
1745 struct rte_mbuf **pkts, uint32_t count,
1746 struct rte_mbuf **comp_pkts, uint32_t *comp_count)
1748 struct vhost_virtqueue *vq;
1751 VHOST_LOG_DATA(DEBUG, "(%d) %s\n", dev->vid, __func__);
1752 if (unlikely(!is_valid_virt_queue_idx(queue_id, 0, dev->nr_vring))) {
1753 VHOST_LOG_DATA(ERR, "(%d) %s: invalid virtqueue idx %d.\n",
1754 dev->vid, __func__, queue_id);
1758 vq = dev->virtqueue[queue_id];
1760 rte_spinlock_lock(&vq->access_lock);
1762 if (unlikely(!vq->enabled || !vq->async_registered))
1763 goto out_access_unlock;
1765 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
1766 vhost_user_iotlb_rd_lock(vq);
1768 if (unlikely(!vq->access_ok))
1769 if (unlikely(vring_translate(dev, vq) < 0))
1772 count = RTE_MIN((uint32_t)MAX_PKT_BURST, count);
1776 /* TODO: packed queue not implemented */
1777 if (vq_is_packed(dev))
1780 nb_tx = virtio_dev_rx_async_submit_split(dev,
1781 vq, queue_id, pkts, count, comp_pkts,
1785 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
1786 vhost_user_iotlb_rd_unlock(vq);
1789 rte_spinlock_unlock(&vq->access_lock);
1795 rte_vhost_submit_enqueue_burst(int vid, uint16_t queue_id,
1796 struct rte_mbuf **pkts, uint16_t count,
1797 struct rte_mbuf **comp_pkts, uint32_t *comp_count)
1799 struct virtio_net *dev = get_device(vid);
1805 if (unlikely(!(dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET))) {
1807 "(%d) %s: built-in vhost net backend is disabled.\n",
1808 dev->vid, __func__);
1812 return virtio_dev_rx_async_submit(dev, queue_id, pkts, count, comp_pkts,
1817 virtio_net_with_host_offload(struct virtio_net *dev)
1820 ((1ULL << VIRTIO_NET_F_CSUM) |
1821 (1ULL << VIRTIO_NET_F_HOST_ECN) |
1822 (1ULL << VIRTIO_NET_F_HOST_TSO4) |
1823 (1ULL << VIRTIO_NET_F_HOST_TSO6) |
1824 (1ULL << VIRTIO_NET_F_HOST_UFO)))
1831 parse_ethernet(struct rte_mbuf *m, uint16_t *l4_proto, void **l4_hdr)
1833 struct rte_ipv4_hdr *ipv4_hdr;
1834 struct rte_ipv6_hdr *ipv6_hdr;
1835 void *l3_hdr = NULL;
1836 struct rte_ether_hdr *eth_hdr;
1839 eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
1841 m->l2_len = sizeof(struct rte_ether_hdr);
1842 ethertype = rte_be_to_cpu_16(eth_hdr->ether_type);
1844 if (ethertype == RTE_ETHER_TYPE_VLAN) {
1845 struct rte_vlan_hdr *vlan_hdr =
1846 (struct rte_vlan_hdr *)(eth_hdr + 1);
1848 m->l2_len += sizeof(struct rte_vlan_hdr);
1849 ethertype = rte_be_to_cpu_16(vlan_hdr->eth_proto);
1852 l3_hdr = (char *)eth_hdr + m->l2_len;
1854 switch (ethertype) {
1855 case RTE_ETHER_TYPE_IPV4:
1857 *l4_proto = ipv4_hdr->next_proto_id;
1858 m->l3_len = rte_ipv4_hdr_len(ipv4_hdr);
1859 *l4_hdr = (char *)l3_hdr + m->l3_len;
1860 m->ol_flags |= PKT_TX_IPV4;
1862 case RTE_ETHER_TYPE_IPV6:
1864 *l4_proto = ipv6_hdr->proto;
1865 m->l3_len = sizeof(struct rte_ipv6_hdr);
1866 *l4_hdr = (char *)l3_hdr + m->l3_len;
1867 m->ol_flags |= PKT_TX_IPV6;
1877 static __rte_always_inline void
1878 vhost_dequeue_offload(struct virtio_net_hdr *hdr, struct rte_mbuf *m)
1880 uint16_t l4_proto = 0;
1881 void *l4_hdr = NULL;
1882 struct rte_tcp_hdr *tcp_hdr = NULL;
1884 if (hdr->flags == 0 && hdr->gso_type == VIRTIO_NET_HDR_GSO_NONE)
1887 parse_ethernet(m, &l4_proto, &l4_hdr);
1888 if (hdr->flags == VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1889 if (hdr->csum_start == (m->l2_len + m->l3_len)) {
1890 switch (hdr->csum_offset) {
1891 case (offsetof(struct rte_tcp_hdr, cksum)):
1892 if (l4_proto == IPPROTO_TCP)
1893 m->ol_flags |= PKT_TX_TCP_CKSUM;
1895 case (offsetof(struct rte_udp_hdr, dgram_cksum)):
1896 if (l4_proto == IPPROTO_UDP)
1897 m->ol_flags |= PKT_TX_UDP_CKSUM;
1899 case (offsetof(struct rte_sctp_hdr, cksum)):
1900 if (l4_proto == IPPROTO_SCTP)
1901 m->ol_flags |= PKT_TX_SCTP_CKSUM;
1909 if (l4_hdr && hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1910 switch (hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1911 case VIRTIO_NET_HDR_GSO_TCPV4:
1912 case VIRTIO_NET_HDR_GSO_TCPV6:
1914 m->ol_flags |= PKT_TX_TCP_SEG;
1915 m->tso_segsz = hdr->gso_size;
1916 m->l4_len = (tcp_hdr->data_off & 0xf0) >> 2;
1918 case VIRTIO_NET_HDR_GSO_UDP:
1919 m->ol_flags |= PKT_TX_UDP_SEG;
1920 m->tso_segsz = hdr->gso_size;
1921 m->l4_len = sizeof(struct rte_udp_hdr);
1924 VHOST_LOG_DATA(WARNING,
1925 "unsupported gso type %u.\n", hdr->gso_type);
1931 static __rte_noinline void
1932 copy_vnet_hdr_from_desc(struct virtio_net_hdr *hdr,
1933 struct buf_vector *buf_vec)
1936 uint64_t remain = sizeof(struct virtio_net_hdr);
1938 uint64_t dst = (uint64_t)(uintptr_t)hdr;
1941 len = RTE_MIN(remain, buf_vec->buf_len);
1942 src = buf_vec->buf_addr;
1943 rte_memcpy((void *)(uintptr_t)dst,
1944 (void *)(uintptr_t)src, len);
1952 static __rte_always_inline int
1953 copy_desc_to_mbuf(struct virtio_net *dev, struct vhost_virtqueue *vq,
1954 struct buf_vector *buf_vec, uint16_t nr_vec,
1955 struct rte_mbuf *m, struct rte_mempool *mbuf_pool)
1957 uint32_t buf_avail, buf_offset;
1958 uint64_t buf_addr, buf_len;
1959 uint32_t mbuf_avail, mbuf_offset;
1961 struct rte_mbuf *cur = m, *prev = m;
1962 struct virtio_net_hdr tmp_hdr;
1963 struct virtio_net_hdr *hdr = NULL;
1964 /* A counter to avoid desc dead loop chain */
1965 uint16_t vec_idx = 0;
1966 struct batch_copy_elem *batch_copy = vq->batch_copy_elems;
1969 buf_addr = buf_vec[vec_idx].buf_addr;
1970 buf_len = buf_vec[vec_idx].buf_len;
1972 if (unlikely(buf_len < dev->vhost_hlen && nr_vec <= 1)) {
1977 if (virtio_net_with_host_offload(dev)) {
1978 if (unlikely(buf_len < sizeof(struct virtio_net_hdr))) {
1980 * No luck, the virtio-net header doesn't fit
1981 * in a contiguous virtual area.
1983 copy_vnet_hdr_from_desc(&tmp_hdr, buf_vec);
1986 hdr = (struct virtio_net_hdr *)((uintptr_t)buf_addr);
1991 * A virtio driver normally uses at least 2 desc buffers
1992 * for Tx: the first for storing the header, and others
1993 * for storing the data.
1995 if (unlikely(buf_len < dev->vhost_hlen)) {
1996 buf_offset = dev->vhost_hlen - buf_len;
1998 buf_addr = buf_vec[vec_idx].buf_addr;
1999 buf_len = buf_vec[vec_idx].buf_len;
2000 buf_avail = buf_len - buf_offset;
2001 } else if (buf_len == dev->vhost_hlen) {
2002 if (unlikely(++vec_idx >= nr_vec))
2004 buf_addr = buf_vec[vec_idx].buf_addr;
2005 buf_len = buf_vec[vec_idx].buf_len;
2008 buf_avail = buf_len;
2010 buf_offset = dev->vhost_hlen;
2011 buf_avail = buf_vec[vec_idx].buf_len - dev->vhost_hlen;
2015 (uintptr_t)(buf_addr + buf_offset),
2016 (uint32_t)buf_avail, 0);
2019 mbuf_avail = m->buf_len - RTE_PKTMBUF_HEADROOM;
2021 cpy_len = RTE_MIN(buf_avail, mbuf_avail);
2023 if (likely(cpy_len > MAX_BATCH_LEN ||
2024 vq->batch_copy_nb_elems >= vq->size ||
2025 (hdr && cur == m))) {
2026 rte_memcpy(rte_pktmbuf_mtod_offset(cur, void *,
2028 (void *)((uintptr_t)(buf_addr +
2029 buf_offset)), cpy_len);
2031 batch_copy[vq->batch_copy_nb_elems].dst =
2032 rte_pktmbuf_mtod_offset(cur, void *,
2034 batch_copy[vq->batch_copy_nb_elems].src =
2035 (void *)((uintptr_t)(buf_addr + buf_offset));
2036 batch_copy[vq->batch_copy_nb_elems].len = cpy_len;
2037 vq->batch_copy_nb_elems++;
2040 mbuf_avail -= cpy_len;
2041 mbuf_offset += cpy_len;
2042 buf_avail -= cpy_len;
2043 buf_offset += cpy_len;
2045 /* This buf reaches to its end, get the next one */
2046 if (buf_avail == 0) {
2047 if (++vec_idx >= nr_vec)
2050 buf_addr = buf_vec[vec_idx].buf_addr;
2051 buf_len = buf_vec[vec_idx].buf_len;
2054 buf_avail = buf_len;
2056 PRINT_PACKET(dev, (uintptr_t)buf_addr,
2057 (uint32_t)buf_avail, 0);
2061 * This mbuf reaches to its end, get a new one
2062 * to hold more data.
2064 if (mbuf_avail == 0) {
2065 cur = rte_pktmbuf_alloc(mbuf_pool);
2066 if (unlikely(cur == NULL)) {
2067 VHOST_LOG_DATA(ERR, "Failed to "
2068 "allocate memory for mbuf.\n");
2074 prev->data_len = mbuf_offset;
2076 m->pkt_len += mbuf_offset;
2080 mbuf_avail = cur->buf_len - RTE_PKTMBUF_HEADROOM;
2084 prev->data_len = mbuf_offset;
2085 m->pkt_len += mbuf_offset;
2088 vhost_dequeue_offload(hdr, m);
2096 virtio_dev_extbuf_free(void *addr __rte_unused, void *opaque)
2102 virtio_dev_extbuf_alloc(struct rte_mbuf *pkt, uint32_t size)
2104 struct rte_mbuf_ext_shared_info *shinfo = NULL;
2105 uint32_t total_len = RTE_PKTMBUF_HEADROOM + size;
2110 total_len += sizeof(*shinfo) + sizeof(uintptr_t);
2111 total_len = RTE_ALIGN_CEIL(total_len, sizeof(uintptr_t));
2113 if (unlikely(total_len > UINT16_MAX))
2116 buf_len = total_len;
2117 buf = rte_malloc(NULL, buf_len, RTE_CACHE_LINE_SIZE);
2118 if (unlikely(buf == NULL))
2121 /* Initialize shinfo */
2122 shinfo = rte_pktmbuf_ext_shinfo_init_helper(buf, &buf_len,
2123 virtio_dev_extbuf_free, buf);
2124 if (unlikely(shinfo == NULL)) {
2126 VHOST_LOG_DATA(ERR, "Failed to init shinfo\n");
2130 iova = rte_malloc_virt2iova(buf);
2131 rte_pktmbuf_attach_extbuf(pkt, buf, iova, buf_len, shinfo);
2132 rte_pktmbuf_reset_headroom(pkt);
2138 * Allocate a host supported pktmbuf.
2140 static __rte_always_inline struct rte_mbuf *
2141 virtio_dev_pktmbuf_alloc(struct virtio_net *dev, struct rte_mempool *mp,
2144 struct rte_mbuf *pkt = rte_pktmbuf_alloc(mp);
2146 if (unlikely(pkt == NULL)) {
2148 "Failed to allocate memory for mbuf.\n");
2152 if (rte_pktmbuf_tailroom(pkt) >= data_len)
2155 /* attach an external buffer if supported */
2156 if (dev->extbuf && !virtio_dev_extbuf_alloc(pkt, data_len))
2159 /* check if chained buffers are allowed */
2160 if (!dev->linearbuf)
2163 /* Data doesn't fit into the buffer and the host supports
2164 * only linear buffers
2166 rte_pktmbuf_free(pkt);
2171 static __rte_noinline uint16_t
2172 virtio_dev_tx_split(struct virtio_net *dev, struct vhost_virtqueue *vq,
2173 struct rte_mempool *mbuf_pool, struct rte_mbuf **pkts, uint16_t count)
2176 uint16_t free_entries;
2177 uint16_t dropped = 0;
2178 static bool allocerr_warned;
2181 * The ordering between avail index and
2182 * desc reads needs to be enforced.
2184 free_entries = __atomic_load_n(&vq->avail->idx, __ATOMIC_ACQUIRE) -
2186 if (free_entries == 0)
2189 rte_prefetch0(&vq->avail->ring[vq->last_avail_idx & (vq->size - 1)]);
2191 VHOST_LOG_DATA(DEBUG, "(%d) %s\n", dev->vid, __func__);
2193 count = RTE_MIN(count, MAX_PKT_BURST);
2194 count = RTE_MIN(count, free_entries);
2195 VHOST_LOG_DATA(DEBUG, "(%d) about to dequeue %u buffers\n",
2198 for (i = 0; i < count; i++) {
2199 struct buf_vector buf_vec[BUF_VECTOR_MAX];
2202 uint16_t nr_vec = 0;
2205 if (unlikely(fill_vec_buf_split(dev, vq,
2206 vq->last_avail_idx + i,
2208 &head_idx, &buf_len,
2209 VHOST_ACCESS_RO) < 0))
2212 update_shadow_used_ring_split(vq, head_idx, 0);
2214 pkts[i] = virtio_dev_pktmbuf_alloc(dev, mbuf_pool, buf_len);
2215 if (unlikely(pkts[i] == NULL)) {
2217 * mbuf allocation fails for jumbo packets when external
2218 * buffer allocation is not allowed and linear buffer
2219 * is required. Drop this packet.
2221 if (!allocerr_warned) {
2223 "Failed mbuf alloc of size %d from %s on %s.\n",
2224 buf_len, mbuf_pool->name, dev->ifname);
2225 allocerr_warned = true;
2232 err = copy_desc_to_mbuf(dev, vq, buf_vec, nr_vec, pkts[i],
2234 if (unlikely(err)) {
2235 rte_pktmbuf_free(pkts[i]);
2236 if (!allocerr_warned) {
2238 "Failed to copy desc to mbuf on %s.\n",
2240 allocerr_warned = true;
2248 vq->last_avail_idx += i;
2250 do_data_copy_dequeue(vq);
2251 if (unlikely(i < count))
2252 vq->shadow_used_idx = i;
2253 if (likely(vq->shadow_used_idx)) {
2254 flush_shadow_used_ring_split(dev, vq);
2255 vhost_vring_call_split(dev, vq);
2258 return (i - dropped);
2261 static __rte_always_inline int
2262 vhost_reserve_avail_batch_packed(struct virtio_net *dev,
2263 struct vhost_virtqueue *vq,
2264 struct rte_mempool *mbuf_pool,
2265 struct rte_mbuf **pkts,
2267 uintptr_t *desc_addrs,
2270 bool wrap = vq->avail_wrap_counter;
2271 struct vring_packed_desc *descs = vq->desc_packed;
2272 uint64_t lens[PACKED_BATCH_SIZE];
2273 uint64_t buf_lens[PACKED_BATCH_SIZE];
2274 uint32_t buf_offset = sizeof(struct virtio_net_hdr_mrg_rxbuf);
2277 if (unlikely(avail_idx & PACKED_BATCH_MASK))
2279 if (unlikely((avail_idx + PACKED_BATCH_SIZE) > vq->size))
2282 vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE) {
2283 flags = descs[avail_idx + i].flags;
2284 if (unlikely((wrap != !!(flags & VRING_DESC_F_AVAIL)) ||
2285 (wrap == !!(flags & VRING_DESC_F_USED)) ||
2286 (flags & PACKED_DESC_SINGLE_DEQUEUE_FLAG)))
2290 rte_atomic_thread_fence(__ATOMIC_ACQUIRE);
2292 vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE)
2293 lens[i] = descs[avail_idx + i].len;
2295 vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE) {
2296 desc_addrs[i] = vhost_iova_to_vva(dev, vq,
2297 descs[avail_idx + i].addr,
2298 &lens[i], VHOST_ACCESS_RW);
2301 vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE) {
2302 if (unlikely(!desc_addrs[i]))
2304 if (unlikely((lens[i] != descs[avail_idx + i].len)))
2308 vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE) {
2309 pkts[i] = virtio_dev_pktmbuf_alloc(dev, mbuf_pool, lens[i]);
2314 vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE)
2315 buf_lens[i] = pkts[i]->buf_len - pkts[i]->data_off;
2317 vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE) {
2318 if (unlikely(buf_lens[i] < (lens[i] - buf_offset)))
2322 vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE) {
2323 pkts[i]->pkt_len = lens[i] - buf_offset;
2324 pkts[i]->data_len = pkts[i]->pkt_len;
2325 ids[i] = descs[avail_idx + i].id;
2331 for (i = 0; i < PACKED_BATCH_SIZE; i++)
2332 rte_pktmbuf_free(pkts[i]);
2337 static __rte_always_inline int
2338 virtio_dev_tx_batch_packed(struct virtio_net *dev,
2339 struct vhost_virtqueue *vq,
2340 struct rte_mempool *mbuf_pool,
2341 struct rte_mbuf **pkts)
2343 uint16_t avail_idx = vq->last_avail_idx;
2344 uint32_t buf_offset = sizeof(struct virtio_net_hdr_mrg_rxbuf);
2345 struct virtio_net_hdr *hdr;
2346 uintptr_t desc_addrs[PACKED_BATCH_SIZE];
2347 uint16_t ids[PACKED_BATCH_SIZE];
2350 if (vhost_reserve_avail_batch_packed(dev, vq, mbuf_pool, pkts,
2351 avail_idx, desc_addrs, ids))
2354 vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE)
2355 rte_prefetch0((void *)(uintptr_t)desc_addrs[i]);
2357 vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE)
2358 rte_memcpy(rte_pktmbuf_mtod_offset(pkts[i], void *, 0),
2359 (void *)(uintptr_t)(desc_addrs[i] + buf_offset),
2362 if (virtio_net_with_host_offload(dev)) {
2363 vhost_for_each_try_unroll(i, 0, PACKED_BATCH_SIZE) {
2364 hdr = (struct virtio_net_hdr *)(desc_addrs[i]);
2365 vhost_dequeue_offload(hdr, pkts[i]);
2369 if (virtio_net_is_inorder(dev))
2370 vhost_shadow_dequeue_batch_packed_inorder(vq,
2371 ids[PACKED_BATCH_SIZE - 1]);
2373 vhost_shadow_dequeue_batch_packed(dev, vq, ids);
2375 vq_inc_last_avail_packed(vq, PACKED_BATCH_SIZE);
2380 static __rte_always_inline int
2381 vhost_dequeue_single_packed(struct virtio_net *dev,
2382 struct vhost_virtqueue *vq,
2383 struct rte_mempool *mbuf_pool,
2384 struct rte_mbuf **pkts,
2386 uint16_t *desc_count)
2388 struct buf_vector buf_vec[BUF_VECTOR_MAX];
2390 uint16_t nr_vec = 0;
2392 static bool allocerr_warned;
2394 if (unlikely(fill_vec_buf_packed(dev, vq,
2395 vq->last_avail_idx, desc_count,
2398 VHOST_ACCESS_RO) < 0))
2401 *pkts = virtio_dev_pktmbuf_alloc(dev, mbuf_pool, buf_len);
2402 if (unlikely(*pkts == NULL)) {
2403 if (!allocerr_warned) {
2405 "Failed mbuf alloc of size %d from %s on %s.\n",
2406 buf_len, mbuf_pool->name, dev->ifname);
2407 allocerr_warned = true;
2412 err = copy_desc_to_mbuf(dev, vq, buf_vec, nr_vec, *pkts,
2414 if (unlikely(err)) {
2415 if (!allocerr_warned) {
2417 "Failed to copy desc to mbuf on %s.\n",
2419 allocerr_warned = true;
2421 rte_pktmbuf_free(*pkts);
2428 static __rte_always_inline int
2429 virtio_dev_tx_single_packed(struct virtio_net *dev,
2430 struct vhost_virtqueue *vq,
2431 struct rte_mempool *mbuf_pool,
2432 struct rte_mbuf **pkts)
2435 uint16_t buf_id, desc_count = 0;
2438 ret = vhost_dequeue_single_packed(dev, vq, mbuf_pool, pkts, &buf_id,
2441 if (likely(desc_count > 0)) {
2442 if (virtio_net_is_inorder(dev))
2443 vhost_shadow_dequeue_single_packed_inorder(vq, buf_id,
2446 vhost_shadow_dequeue_single_packed(vq, buf_id,
2449 vq_inc_last_avail_packed(vq, desc_count);
2455 static __rte_noinline uint16_t
2456 virtio_dev_tx_packed(struct virtio_net *dev,
2457 struct vhost_virtqueue *__rte_restrict vq,
2458 struct rte_mempool *mbuf_pool,
2459 struct rte_mbuf **__rte_restrict pkts,
2462 uint32_t pkt_idx = 0;
2463 uint32_t remained = count;
2466 rte_prefetch0(&vq->desc_packed[vq->last_avail_idx]);
2468 if (remained >= PACKED_BATCH_SIZE) {
2469 if (!virtio_dev_tx_batch_packed(dev, vq, mbuf_pool,
2471 pkt_idx += PACKED_BATCH_SIZE;
2472 remained -= PACKED_BATCH_SIZE;
2477 if (virtio_dev_tx_single_packed(dev, vq, mbuf_pool,
2485 if (vq->shadow_used_idx) {
2486 do_data_copy_dequeue(vq);
2488 vhost_flush_dequeue_shadow_packed(dev, vq);
2489 vhost_vring_call_packed(dev, vq);
2496 rte_vhost_dequeue_burst(int vid, uint16_t queue_id,
2497 struct rte_mempool *mbuf_pool, struct rte_mbuf **pkts, uint16_t count)
2499 struct virtio_net *dev;
2500 struct rte_mbuf *rarp_mbuf = NULL;
2501 struct vhost_virtqueue *vq;
2502 int16_t success = 1;
2504 dev = get_device(vid);
2508 if (unlikely(!(dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET))) {
2510 "(%d) %s: built-in vhost net backend is disabled.\n",
2511 dev->vid, __func__);
2515 if (unlikely(!is_valid_virt_queue_idx(queue_id, 1, dev->nr_vring))) {
2517 "(%d) %s: invalid virtqueue idx %d.\n",
2518 dev->vid, __func__, queue_id);
2522 vq = dev->virtqueue[queue_id];
2524 if (unlikely(rte_spinlock_trylock(&vq->access_lock) == 0))
2527 if (unlikely(!vq->enabled)) {
2529 goto out_access_unlock;
2532 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
2533 vhost_user_iotlb_rd_lock(vq);
2535 if (unlikely(!vq->access_ok))
2536 if (unlikely(vring_translate(dev, vq) < 0)) {
2542 * Construct a RARP broadcast packet, and inject it to the "pkts"
2543 * array, to looks like that guest actually send such packet.
2545 * Check user_send_rarp() for more information.
2547 * broadcast_rarp shares a cacheline in the virtio_net structure
2548 * with some fields that are accessed during enqueue and
2549 * __atomic_compare_exchange_n causes a write if performed compare
2550 * and exchange. This could result in false sharing between enqueue
2553 * Prevent unnecessary false sharing by reading broadcast_rarp first
2554 * and only performing compare and exchange if the read indicates it
2555 * is likely to be set.
2557 if (unlikely(__atomic_load_n(&dev->broadcast_rarp, __ATOMIC_ACQUIRE) &&
2558 __atomic_compare_exchange_n(&dev->broadcast_rarp,
2559 &success, 0, 0, __ATOMIC_RELEASE, __ATOMIC_RELAXED))) {
2561 rarp_mbuf = rte_net_make_rarp_packet(mbuf_pool, &dev->mac);
2562 if (rarp_mbuf == NULL) {
2563 VHOST_LOG_DATA(ERR, "Failed to make RARP packet.\n");
2570 if (vq_is_packed(dev))
2571 count = virtio_dev_tx_packed(dev, vq, mbuf_pool, pkts, count);
2573 count = virtio_dev_tx_split(dev, vq, mbuf_pool, pkts, count);
2576 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
2577 vhost_user_iotlb_rd_unlock(vq);
2580 rte_spinlock_unlock(&vq->access_lock);
2582 if (unlikely(rarp_mbuf != NULL)) {
2584 * Inject it to the head of "pkts" array, so that switch's mac
2585 * learning table will get updated first.
2587 memmove(&pkts[1], pkts, count * sizeof(struct rte_mbuf *));
2588 pkts[0] = rarp_mbuf;