4 * Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * * Neither the name of Intel Corporation nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
36 #include <linux/virtio_net.h>
39 #include <rte_memcpy.h>
40 #include <rte_ether.h>
42 #include <rte_virtio_net.h>
48 #include "vhost-net.h"
50 #define MAX_PKT_BURST 32
51 #define VHOST_LOG_PAGE 4096
53 static inline void __attribute__((always_inline))
54 vhost_log_page(uint8_t *log_base, uint64_t page)
56 log_base[page / 8] |= 1 << (page % 8);
59 static inline void __attribute__((always_inline))
60 vhost_log_write(struct virtio_net *dev, uint64_t addr, uint64_t len)
64 if (likely(((dev->features & (1ULL << VHOST_F_LOG_ALL)) == 0) ||
65 !dev->log_base || !len))
68 if (unlikely(dev->log_size <= ((addr + len - 1) / VHOST_LOG_PAGE / 8)))
71 /* To make sure guest memory updates are committed before logging */
74 page = addr / VHOST_LOG_PAGE;
75 while (page * VHOST_LOG_PAGE < addr + len) {
76 vhost_log_page((uint8_t *)(uintptr_t)dev->log_base, page);
81 static inline void __attribute__((always_inline))
82 vhost_log_used_vring(struct virtio_net *dev, struct vhost_virtqueue *vq,
83 uint64_t offset, uint64_t len)
85 vhost_log_write(dev, vq->log_guest_addr + offset, len);
89 is_valid_virt_queue_idx(uint32_t idx, int is_tx, uint32_t qp_nb)
91 return (is_tx ^ (idx & 1)) == 0 && idx < qp_nb * VIRTIO_QNUM;
95 virtio_enqueue_offload(struct rte_mbuf *m_buf, struct virtio_net_hdr *net_hdr)
97 if (m_buf->ol_flags & PKT_TX_L4_MASK) {
98 net_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
99 net_hdr->csum_start = m_buf->l2_len + m_buf->l3_len;
101 switch (m_buf->ol_flags & PKT_TX_L4_MASK) {
102 case PKT_TX_TCP_CKSUM:
103 net_hdr->csum_offset = (offsetof(struct tcp_hdr,
106 case PKT_TX_UDP_CKSUM:
107 net_hdr->csum_offset = (offsetof(struct udp_hdr,
110 case PKT_TX_SCTP_CKSUM:
111 net_hdr->csum_offset = (offsetof(struct sctp_hdr,
117 if (m_buf->ol_flags & PKT_TX_TCP_SEG) {
118 if (m_buf->ol_flags & PKT_TX_IPV4)
119 net_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
121 net_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
122 net_hdr->gso_size = m_buf->tso_segsz;
123 net_hdr->hdr_len = m_buf->l2_len + m_buf->l3_len
129 copy_virtio_net_hdr(struct vhost_virtqueue *vq, uint64_t desc_addr,
130 struct virtio_net_hdr_mrg_rxbuf hdr)
132 if (vq->vhost_hlen == sizeof(struct virtio_net_hdr_mrg_rxbuf))
133 *(struct virtio_net_hdr_mrg_rxbuf *)(uintptr_t)desc_addr = hdr;
135 *(struct virtio_net_hdr *)(uintptr_t)desc_addr = hdr.hdr;
138 static inline int __attribute__((always_inline))
139 copy_mbuf_to_desc(struct virtio_net *dev, struct vhost_virtqueue *vq,
140 struct rte_mbuf *m, uint16_t desc_idx, uint32_t *copied)
142 uint32_t desc_avail, desc_offset;
143 uint32_t mbuf_avail, mbuf_offset;
145 struct vring_desc *desc;
147 struct virtio_net_hdr_mrg_rxbuf virtio_hdr = {{0, 0, 0, 0, 0, 0}, 0};
149 desc = &vq->desc[desc_idx];
150 if (unlikely(desc->len < vq->vhost_hlen))
153 desc_addr = gpa_to_vva(dev, desc->addr);
154 rte_prefetch0((void *)(uintptr_t)desc_addr);
156 virtio_enqueue_offload(m, &virtio_hdr.hdr);
157 copy_virtio_net_hdr(vq, desc_addr, virtio_hdr);
158 vhost_log_write(dev, desc->addr, vq->vhost_hlen);
159 PRINT_PACKET(dev, (uintptr_t)desc_addr, vq->vhost_hlen, 0);
161 desc_offset = vq->vhost_hlen;
162 desc_avail = desc->len - vq->vhost_hlen;
164 *copied = rte_pktmbuf_pkt_len(m);
165 mbuf_avail = rte_pktmbuf_data_len(m);
167 while (mbuf_avail != 0 || m->next != NULL) {
168 /* done with current mbuf, fetch next */
169 if (mbuf_avail == 0) {
173 mbuf_avail = rte_pktmbuf_data_len(m);
176 /* done with current desc buf, fetch next */
177 if (desc_avail == 0) {
178 if ((desc->flags & VRING_DESC_F_NEXT) == 0) {
179 /* Room in vring buffer is not enough */
182 if (unlikely(desc->next >= vq->size))
185 desc = &vq->desc[desc->next];
186 desc_addr = gpa_to_vva(dev, desc->addr);
188 desc_avail = desc->len;
191 cpy_len = RTE_MIN(desc_avail, mbuf_avail);
192 rte_memcpy((void *)((uintptr_t)(desc_addr + desc_offset)),
193 rte_pktmbuf_mtod_offset(m, void *, mbuf_offset),
195 vhost_log_write(dev, desc->addr + desc_offset, cpy_len);
196 PRINT_PACKET(dev, (uintptr_t)(desc_addr + desc_offset),
199 mbuf_avail -= cpy_len;
200 mbuf_offset += cpy_len;
201 desc_avail -= cpy_len;
202 desc_offset += cpy_len;
209 * As many data cores may want to access available buffers
210 * they need to be reserved.
212 static inline uint32_t
213 reserve_avail_buf(struct vhost_virtqueue *vq, uint32_t count,
214 uint16_t *start, uint16_t *end)
216 uint16_t res_start_idx;
217 uint16_t res_end_idx;
219 uint16_t free_entries;
222 count = RTE_MIN(count, (uint32_t)MAX_PKT_BURST);
225 res_start_idx = vq->last_used_idx_res;
226 avail_idx = *((volatile uint16_t *)&vq->avail->idx);
228 free_entries = avail_idx - res_start_idx;
229 count = RTE_MIN(count, free_entries);
233 res_end_idx = res_start_idx + count;
236 * update vq->last_used_idx_res atomically; try again if failed.
238 * TODO: Allow to disable cmpset if no concurrency in application.
240 success = rte_atomic16_cmpset(&vq->last_used_idx_res,
241 res_start_idx, res_end_idx);
242 if (unlikely(!success))
245 *start = res_start_idx;
252 * This function adds buffers to the virtio devices RX virtqueue. Buffers can
253 * be received from the physical port or from another virtio device. A packet
254 * count is returned to indicate the number of packets that are succesfully
255 * added to the RX queue. This function works when the mbuf is scattered, but
256 * it doesn't support the mergeable feature.
258 static inline uint32_t __attribute__((always_inline))
259 virtio_dev_rx(struct virtio_net *dev, uint16_t queue_id,
260 struct rte_mbuf **pkts, uint32_t count)
262 struct vhost_virtqueue *vq;
263 uint16_t res_start_idx, res_end_idx;
264 uint16_t desc_indexes[MAX_PKT_BURST];
267 LOG_DEBUG(VHOST_DATA, "(%"PRIu64") virtio_dev_rx()\n", dev->device_fh);
268 if (unlikely(!is_valid_virt_queue_idx(queue_id, 0, dev->virt_qp_nb))) {
269 RTE_LOG(ERR, VHOST_DATA,
270 "%s (%"PRIu64"): virtqueue idx:%d invalid.\n",
271 __func__, dev->device_fh, queue_id);
275 vq = dev->virtqueue[queue_id];
276 if (unlikely(vq->enabled == 0))
279 count = reserve_avail_buf(vq, count, &res_start_idx, &res_end_idx);
283 LOG_DEBUG(VHOST_DATA,
284 "(%"PRIu64") res_start_idx %d| res_end_idx Index %d\n",
285 dev->device_fh, res_start_idx, res_end_idx);
287 /* Retrieve all of the desc indexes first to avoid caching issues. */
288 rte_prefetch0(&vq->avail->ring[res_start_idx & (vq->size - 1)]);
289 for (i = 0; i < count; i++) {
290 desc_indexes[i] = vq->avail->ring[(res_start_idx + i) &
294 rte_prefetch0(&vq->desc[desc_indexes[0]]);
295 for (i = 0; i < count; i++) {
296 uint16_t desc_idx = desc_indexes[i];
297 uint16_t used_idx = (res_start_idx + i) & (vq->size - 1);
301 err = copy_mbuf_to_desc(dev, vq, pkts[i], desc_idx, &copied);
303 vq->used->ring[used_idx].id = desc_idx;
305 vq->used->ring[used_idx].len = vq->vhost_hlen;
307 vq->used->ring[used_idx].len = copied + vq->vhost_hlen;
308 vhost_log_used_vring(dev, vq,
309 offsetof(struct vring_used, ring[used_idx]),
310 sizeof(vq->used->ring[used_idx]));
313 rte_prefetch0(&vq->desc[desc_indexes[i+1]]);
318 /* Wait until it's our turn to add our buffer to the used ring. */
319 while (unlikely(vq->last_used_idx != res_start_idx))
322 *(volatile uint16_t *)&vq->used->idx += count;
323 vq->last_used_idx = res_end_idx;
324 vhost_log_used_vring(dev, vq,
325 offsetof(struct vring_used, idx),
326 sizeof(vq->used->idx));
328 /* flush used->idx update before we read avail->flags. */
331 /* Kick the guest if necessary. */
332 if (!(vq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
333 && (vq->callfd >= 0))
334 eventfd_write(vq->callfd, (eventfd_t)1);
339 fill_vec_buf(struct vhost_virtqueue *vq, uint32_t avail_idx,
340 uint32_t *allocated, uint32_t *vec_idx)
342 uint16_t idx = vq->avail->ring[avail_idx & (vq->size - 1)];
343 uint32_t vec_id = *vec_idx;
344 uint32_t len = *allocated;
347 if (unlikely(vec_id >= BUF_VECTOR_MAX || idx >= vq->size))
350 len += vq->desc[idx].len;
351 vq->buf_vec[vec_id].buf_addr = vq->desc[idx].addr;
352 vq->buf_vec[vec_id].buf_len = vq->desc[idx].len;
353 vq->buf_vec[vec_id].desc_idx = idx;
356 if ((vq->desc[idx].flags & VRING_DESC_F_NEXT) == 0)
359 idx = vq->desc[idx].next;
369 * As many data cores may want to access available buffers concurrently,
370 * they need to be reserved.
372 * Returns -1 on fail, 0 on success
375 reserve_avail_buf_mergeable(struct vhost_virtqueue *vq, uint32_t size,
376 uint16_t *start, uint16_t *end)
378 uint16_t res_start_idx;
379 uint16_t res_cur_idx;
386 res_start_idx = vq->last_used_idx_res;
387 res_cur_idx = res_start_idx;
393 avail_idx = *((volatile uint16_t *)&vq->avail->idx);
394 if (unlikely(res_cur_idx == avail_idx))
397 if (unlikely(fill_vec_buf(vq, res_cur_idx, &allocated,
404 if (allocated >= size)
408 * if we tried all available ring items, and still
409 * can't get enough buf, it means something abnormal
412 if (unlikely(tries >= vq->size))
417 * update vq->last_used_idx_res atomically.
418 * retry again if failed.
420 if (rte_atomic16_cmpset(&vq->last_used_idx_res,
421 res_start_idx, res_cur_idx) == 0)
424 *start = res_start_idx;
429 static inline uint32_t __attribute__((always_inline))
430 copy_mbuf_to_desc_mergeable(struct virtio_net *dev, struct vhost_virtqueue *vq,
431 uint16_t res_start_idx, uint16_t res_end_idx,
434 struct virtio_net_hdr_mrg_rxbuf virtio_hdr = {{0, 0, 0, 0, 0, 0}, 0};
435 uint32_t vec_idx = 0;
436 uint16_t cur_idx = res_start_idx;
438 uint32_t mbuf_offset, mbuf_avail;
439 uint32_t desc_offset, desc_avail;
441 uint16_t desc_idx, used_idx;
443 if (unlikely(m == NULL))
446 LOG_DEBUG(VHOST_DATA,
447 "(%"PRIu64") Current Index %d| End Index %d\n",
448 dev->device_fh, cur_idx, res_end_idx);
450 if (vq->buf_vec[vec_idx].buf_len < vq->vhost_hlen)
453 desc_addr = gpa_to_vva(dev, vq->buf_vec[vec_idx].buf_addr);
454 rte_prefetch0((void *)(uintptr_t)desc_addr);
456 virtio_hdr.num_buffers = res_end_idx - res_start_idx;
457 LOG_DEBUG(VHOST_DATA, "(%"PRIu64") RX: Num merge buffers %d\n",
458 dev->device_fh, virtio_hdr.num_buffers);
460 virtio_enqueue_offload(m, &virtio_hdr.hdr);
461 copy_virtio_net_hdr(vq, desc_addr, virtio_hdr);
462 vhost_log_write(dev, vq->buf_vec[vec_idx].buf_addr, vq->vhost_hlen);
463 PRINT_PACKET(dev, (uintptr_t)desc_addr, vq->vhost_hlen, 0);
465 desc_avail = vq->buf_vec[vec_idx].buf_len - vq->vhost_hlen;
466 desc_offset = vq->vhost_hlen;
468 mbuf_avail = rte_pktmbuf_data_len(m);
470 while (mbuf_avail != 0 || m->next != NULL) {
471 /* done with current desc buf, get the next one */
472 if (desc_avail == 0) {
473 desc_idx = vq->buf_vec[vec_idx].desc_idx;
475 if (!(vq->desc[desc_idx].flags & VRING_DESC_F_NEXT)) {
476 /* Update used ring with desc information */
477 used_idx = cur_idx++ & (vq->size - 1);
478 vq->used->ring[used_idx].id = desc_idx;
479 vq->used->ring[used_idx].len = desc_offset;
480 vhost_log_used_vring(dev, vq,
481 offsetof(struct vring_used,
483 sizeof(vq->used->ring[used_idx]));
487 desc_addr = gpa_to_vva(dev, vq->buf_vec[vec_idx].buf_addr);
489 /* Prefetch buffer address. */
490 rte_prefetch0((void *)(uintptr_t)desc_addr);
492 desc_avail = vq->buf_vec[vec_idx].buf_len;
495 /* done with current mbuf, get the next one */
496 if (mbuf_avail == 0) {
500 mbuf_avail = rte_pktmbuf_data_len(m);
503 cpy_len = RTE_MIN(desc_avail, mbuf_avail);
504 rte_memcpy((void *)((uintptr_t)(desc_addr + desc_offset)),
505 rte_pktmbuf_mtod_offset(m, void *, mbuf_offset),
507 vhost_log_write(dev, vq->buf_vec[vec_idx].buf_addr + desc_offset,
509 PRINT_PACKET(dev, (uintptr_t)(desc_addr + desc_offset),
512 mbuf_avail -= cpy_len;
513 mbuf_offset += cpy_len;
514 desc_avail -= cpy_len;
515 desc_offset += cpy_len;
518 used_idx = cur_idx & (vq->size - 1);
519 vq->used->ring[used_idx].id = vq->buf_vec[vec_idx].desc_idx;
520 vq->used->ring[used_idx].len = desc_offset;
521 vhost_log_used_vring(dev, vq,
522 offsetof(struct vring_used, ring[used_idx]),
523 sizeof(vq->used->ring[used_idx]));
525 return res_end_idx - res_start_idx;
528 static inline uint32_t __attribute__((always_inline))
529 virtio_dev_merge_rx(struct virtio_net *dev, uint16_t queue_id,
530 struct rte_mbuf **pkts, uint32_t count)
532 struct vhost_virtqueue *vq;
533 uint32_t pkt_idx = 0, nr_used = 0;
536 LOG_DEBUG(VHOST_DATA, "(%"PRIu64") virtio_dev_merge_rx()\n",
538 if (unlikely(!is_valid_virt_queue_idx(queue_id, 0, dev->virt_qp_nb))) {
539 RTE_LOG(ERR, VHOST_DATA,
540 "%s (%"PRIu64"): virtqueue idx:%d invalid.\n",
541 __func__, dev->device_fh, queue_id);
545 vq = dev->virtqueue[queue_id];
546 if (unlikely(vq->enabled == 0))
549 count = RTE_MIN((uint32_t)MAX_PKT_BURST, count);
553 for (pkt_idx = 0; pkt_idx < count; pkt_idx++) {
554 uint32_t pkt_len = pkts[pkt_idx]->pkt_len + vq->vhost_hlen;
556 if (unlikely(reserve_avail_buf_mergeable(vq, pkt_len,
557 &start, &end) < 0)) {
558 LOG_DEBUG(VHOST_DATA,
559 "(%" PRIu64 ") Failed to get enough desc from vring\n",
564 nr_used = copy_mbuf_to_desc_mergeable(dev, vq, start, end,
569 * Wait until it's our turn to add our buffer
572 while (unlikely(vq->last_used_idx != start))
575 *(volatile uint16_t *)&vq->used->idx += nr_used;
576 vhost_log_used_vring(dev, vq, offsetof(struct vring_used, idx),
577 sizeof(vq->used->idx));
578 vq->last_used_idx = end;
581 if (likely(pkt_idx)) {
582 /* flush used->idx update before we read avail->flags. */
585 /* Kick the guest if necessary. */
586 if (!(vq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
587 && (vq->callfd >= 0))
588 eventfd_write(vq->callfd, (eventfd_t)1);
595 rte_vhost_enqueue_burst(struct virtio_net *dev, uint16_t queue_id,
596 struct rte_mbuf **pkts, uint16_t count)
598 if (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF))
599 return virtio_dev_merge_rx(dev, queue_id, pkts, count);
601 return virtio_dev_rx(dev, queue_id, pkts, count);
605 parse_ethernet(struct rte_mbuf *m, uint16_t *l4_proto, void **l4_hdr)
607 struct ipv4_hdr *ipv4_hdr;
608 struct ipv6_hdr *ipv6_hdr;
610 struct ether_hdr *eth_hdr;
613 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
615 m->l2_len = sizeof(struct ether_hdr);
616 ethertype = rte_be_to_cpu_16(eth_hdr->ether_type);
618 if (ethertype == ETHER_TYPE_VLAN) {
619 struct vlan_hdr *vlan_hdr = (struct vlan_hdr *)(eth_hdr + 1);
621 m->l2_len += sizeof(struct vlan_hdr);
622 ethertype = rte_be_to_cpu_16(vlan_hdr->eth_proto);
625 l3_hdr = (char *)eth_hdr + m->l2_len;
628 case ETHER_TYPE_IPv4:
629 ipv4_hdr = (struct ipv4_hdr *)l3_hdr;
630 *l4_proto = ipv4_hdr->next_proto_id;
631 m->l3_len = (ipv4_hdr->version_ihl & 0x0f) * 4;
632 *l4_hdr = (char *)l3_hdr + m->l3_len;
633 m->ol_flags |= PKT_TX_IPV4;
635 case ETHER_TYPE_IPv6:
636 ipv6_hdr = (struct ipv6_hdr *)l3_hdr;
637 *l4_proto = ipv6_hdr->proto;
638 m->l3_len = sizeof(struct ipv6_hdr);
639 *l4_hdr = (char *)l3_hdr + m->l3_len;
640 m->ol_flags |= PKT_TX_IPV6;
649 static inline void __attribute__((always_inline))
650 vhost_dequeue_offload(struct virtio_net_hdr *hdr, struct rte_mbuf *m)
652 uint16_t l4_proto = 0;
654 struct tcp_hdr *tcp_hdr = NULL;
656 parse_ethernet(m, &l4_proto, &l4_hdr);
657 if (hdr->flags == VIRTIO_NET_HDR_F_NEEDS_CSUM) {
658 if (hdr->csum_start == (m->l2_len + m->l3_len)) {
659 switch (hdr->csum_offset) {
660 case (offsetof(struct tcp_hdr, cksum)):
661 if (l4_proto == IPPROTO_TCP)
662 m->ol_flags |= PKT_TX_TCP_CKSUM;
664 case (offsetof(struct udp_hdr, dgram_cksum)):
665 if (l4_proto == IPPROTO_UDP)
666 m->ol_flags |= PKT_TX_UDP_CKSUM;
668 case (offsetof(struct sctp_hdr, cksum)):
669 if (l4_proto == IPPROTO_SCTP)
670 m->ol_flags |= PKT_TX_SCTP_CKSUM;
678 if (hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
679 switch (hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
680 case VIRTIO_NET_HDR_GSO_TCPV4:
681 case VIRTIO_NET_HDR_GSO_TCPV6:
682 tcp_hdr = (struct tcp_hdr *)l4_hdr;
683 m->ol_flags |= PKT_TX_TCP_SEG;
684 m->tso_segsz = hdr->gso_size;
685 m->l4_len = (tcp_hdr->data_off & 0xf0) >> 2;
688 RTE_LOG(WARNING, VHOST_DATA,
689 "unsupported gso type %u.\n", hdr->gso_type);
695 #define RARP_PKT_SIZE 64
698 make_rarp_packet(struct rte_mbuf *rarp_mbuf, const struct ether_addr *mac)
700 struct ether_hdr *eth_hdr;
701 struct arp_hdr *rarp;
703 if (rarp_mbuf->buf_len < 64) {
704 RTE_LOG(WARNING, VHOST_DATA,
705 "failed to make RARP; mbuf size too small %u (< %d)\n",
706 rarp_mbuf->buf_len, RARP_PKT_SIZE);
710 /* Ethernet header. */
711 eth_hdr = rte_pktmbuf_mtod_offset(rarp_mbuf, struct ether_hdr *, 0);
712 memset(eth_hdr->d_addr.addr_bytes, 0xff, ETHER_ADDR_LEN);
713 ether_addr_copy(mac, ð_hdr->s_addr);
714 eth_hdr->ether_type = htons(ETHER_TYPE_RARP);
717 rarp = (struct arp_hdr *)(eth_hdr + 1);
718 rarp->arp_hrd = htons(ARP_HRD_ETHER);
719 rarp->arp_pro = htons(ETHER_TYPE_IPv4);
720 rarp->arp_hln = ETHER_ADDR_LEN;
722 rarp->arp_op = htons(ARP_OP_REVREQUEST);
724 ether_addr_copy(mac, &rarp->arp_data.arp_sha);
725 ether_addr_copy(mac, &rarp->arp_data.arp_tha);
726 memset(&rarp->arp_data.arp_sip, 0x00, 4);
727 memset(&rarp->arp_data.arp_tip, 0x00, 4);
729 rarp_mbuf->pkt_len = rarp_mbuf->data_len = RARP_PKT_SIZE;
734 static inline int __attribute__((always_inline))
735 copy_desc_to_mbuf(struct virtio_net *dev, struct vhost_virtqueue *vq,
736 struct rte_mbuf *m, uint16_t desc_idx,
737 struct rte_mempool *mbuf_pool)
739 struct vring_desc *desc;
741 uint32_t desc_avail, desc_offset;
742 uint32_t mbuf_avail, mbuf_offset;
744 struct rte_mbuf *cur = m, *prev = m;
745 struct virtio_net_hdr *hdr;
746 /* A counter to avoid desc dead loop chain */
747 uint32_t nr_desc = 1;
749 desc = &vq->desc[desc_idx];
750 if (unlikely(desc->len < vq->vhost_hlen))
753 desc_addr = gpa_to_vva(dev, desc->addr);
754 rte_prefetch0((void *)(uintptr_t)desc_addr);
756 /* Retrieve virtio net header */
757 hdr = (struct virtio_net_hdr *)((uintptr_t)desc_addr);
758 desc_avail = desc->len - vq->vhost_hlen;
759 desc_offset = vq->vhost_hlen;
762 mbuf_avail = m->buf_len - RTE_PKTMBUF_HEADROOM;
763 while (desc_avail != 0 || (desc->flags & VRING_DESC_F_NEXT) != 0) {
764 /* This desc reaches to its end, get the next one */
765 if (desc_avail == 0) {
766 if (unlikely(desc->next >= vq->size ||
767 ++nr_desc >= vq->size))
769 desc = &vq->desc[desc->next];
771 desc_addr = gpa_to_vva(dev, desc->addr);
772 rte_prefetch0((void *)(uintptr_t)desc_addr);
775 desc_avail = desc->len;
777 PRINT_PACKET(dev, (uintptr_t)desc_addr, desc->len, 0);
781 * This mbuf reaches to its end, get a new one
784 if (mbuf_avail == 0) {
785 cur = rte_pktmbuf_alloc(mbuf_pool);
786 if (unlikely(cur == NULL)) {
787 RTE_LOG(ERR, VHOST_DATA, "Failed to "
788 "allocate memory for mbuf.\n");
793 prev->data_len = mbuf_offset;
795 m->pkt_len += mbuf_offset;
799 mbuf_avail = cur->buf_len - RTE_PKTMBUF_HEADROOM;
802 cpy_len = RTE_MIN(desc_avail, mbuf_avail);
803 rte_memcpy(rte_pktmbuf_mtod_offset(cur, void *, mbuf_offset),
804 (void *)((uintptr_t)(desc_addr + desc_offset)),
807 mbuf_avail -= cpy_len;
808 mbuf_offset += cpy_len;
809 desc_avail -= cpy_len;
810 desc_offset += cpy_len;
813 prev->data_len = mbuf_offset;
814 m->pkt_len += mbuf_offset;
816 if (hdr->flags != 0 || hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE)
817 vhost_dequeue_offload(hdr, m);
823 rte_vhost_dequeue_burst(struct virtio_net *dev, uint16_t queue_id,
824 struct rte_mempool *mbuf_pool, struct rte_mbuf **pkts, uint16_t count)
826 struct rte_mbuf *rarp_mbuf = NULL;
827 struct vhost_virtqueue *vq;
828 uint32_t desc_indexes[MAX_PKT_BURST];
831 uint16_t free_entries;
834 if (unlikely(!is_valid_virt_queue_idx(queue_id, 1, dev->virt_qp_nb))) {
835 RTE_LOG(ERR, VHOST_DATA,
836 "%s (%"PRIu64"): virtqueue idx:%d invalid.\n",
837 __func__, dev->device_fh, queue_id);
841 vq = dev->virtqueue[queue_id];
842 if (unlikely(vq->enabled == 0))
846 * Construct a RARP broadcast packet, and inject it to the "pkts"
847 * array, to looks like that guest actually send such packet.
849 * Check user_send_rarp() for more information.
851 if (unlikely(rte_atomic16_cmpset((volatile uint16_t *)
852 &dev->broadcast_rarp.cnt, 1, 0))) {
853 rarp_mbuf = rte_pktmbuf_alloc(mbuf_pool);
854 if (rarp_mbuf == NULL) {
855 RTE_LOG(ERR, VHOST_DATA,
856 "Failed to allocate memory for mbuf.\n");
860 if (make_rarp_packet(rarp_mbuf, &dev->mac)) {
861 rte_pktmbuf_free(rarp_mbuf);
868 avail_idx = *((volatile uint16_t *)&vq->avail->idx);
869 free_entries = avail_idx - vq->last_used_idx;
870 if (free_entries == 0)
873 LOG_DEBUG(VHOST_DATA, "%s (%"PRIu64")\n", __func__, dev->device_fh);
875 /* Prefetch available ring to retrieve head indexes. */
876 used_idx = vq->last_used_idx & (vq->size - 1);
877 rte_prefetch0(&vq->avail->ring[used_idx]);
879 count = RTE_MIN(count, MAX_PKT_BURST);
880 count = RTE_MIN(count, free_entries);
881 LOG_DEBUG(VHOST_DATA, "(%"PRIu64") about to dequeue %u buffers\n",
882 dev->device_fh, count);
884 /* Retrieve all of the head indexes first to avoid caching issues. */
885 for (i = 0; i < count; i++) {
886 desc_indexes[i] = vq->avail->ring[(vq->last_used_idx + i) &
890 /* Prefetch descriptor index. */
891 rte_prefetch0(&vq->desc[desc_indexes[0]]);
892 rte_prefetch0(&vq->used->ring[vq->last_used_idx & (vq->size - 1)]);
894 for (i = 0; i < count; i++) {
897 if (likely(i + 1 < count)) {
898 rte_prefetch0(&vq->desc[desc_indexes[i + 1]]);
899 rte_prefetch0(&vq->used->ring[(used_idx + 1) &
903 pkts[i] = rte_pktmbuf_alloc(mbuf_pool);
904 if (unlikely(pkts[i] == NULL)) {
905 RTE_LOG(ERR, VHOST_DATA,
906 "Failed to allocate memory for mbuf.\n");
909 err = copy_desc_to_mbuf(dev, vq, pkts[i], desc_indexes[i],
912 rte_pktmbuf_free(pkts[i]);
916 used_idx = vq->last_used_idx++ & (vq->size - 1);
917 vq->used->ring[used_idx].id = desc_indexes[i];
918 vq->used->ring[used_idx].len = 0;
919 vhost_log_used_vring(dev, vq,
920 offsetof(struct vring_used, ring[used_idx]),
921 sizeof(vq->used->ring[used_idx]));
927 vhost_log_used_vring(dev, vq, offsetof(struct vring_used, idx),
928 sizeof(vq->used->idx));
930 /* Kick guest if required. */
931 if (!(vq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT)
932 && (vq->callfd >= 0))
933 eventfd_write(vq->callfd, (eventfd_t)1);
936 if (unlikely(rarp_mbuf != NULL)) {
938 * Inject it to the head of "pkts" array, so that switch's mac
939 * learning table will get updated first.
941 memmove(&pkts[1], pkts, i * sizeof(struct rte_mbuf *));