1 /* A network driver using virtio.
3 * Copyright 2007 Rusty Russell <rusty@rustcorp.com.au> IBM Corporation
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, see <http://www.gnu.org/licenses/>.
19 #include <linux/netdevice.h>
20 #include <linux/etherdevice.h>
21 #include <linux/ethtool.h>
22 #include <linux/module.h>
23 #include <linux/virtio.h>
24 #include <linux/virtio_net.h>
25 #include <linux/scatterlist.h>
26 #include <linux/if_vlan.h>
27 #include <linux/slab.h>
28 #include <linux/cpu.h>
29 #include <linux/average.h>
30 #include <net/busy_poll.h>
32 static int napi_weight = NAPI_POLL_WEIGHT;
33 module_param(napi_weight, int, 0444);
35 static bool csum = true, gso = true;
36 module_param(csum, bool, 0444);
37 module_param(gso, bool, 0444);
39 /* FIXME: MTU in config. */
40 #define GOOD_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
41 #define GOOD_COPY_LEN 128
43 /* RX packet size EWMA. The average packet size is used to determine the packet
44 * buffer size when refilling RX rings. As the entire RX ring may be refilled
45 * at once, the weight is chosen so that the EWMA will be insensitive to short-
46 * term, transient changes in packet size.
48 DECLARE_EWMA(pkt_len, 1, 64)
50 /* Minimum alignment for mergeable packet buffers. */
51 #define MERGEABLE_BUFFER_ALIGN max(L1_CACHE_BYTES, 256)
53 #define VIRTNET_DRIVER_VERSION "1.0.0"
55 struct virtnet_stats {
56 struct u64_stats_sync tx_syncp;
57 struct u64_stats_sync rx_syncp;
65 /* Internal representation of a send virtqueue */
67 /* Virtqueue associated with this send _queue */
70 /* TX: fragments + linear part + virtio header */
71 struct scatterlist sg[MAX_SKB_FRAGS + 2];
73 /* Name of the send queue: output.$index */
77 /* Internal representation of a receive virtqueue */
78 struct receive_queue {
79 /* Virtqueue associated with this receive_queue */
82 struct napi_struct napi;
84 /* Chain pages by the private ptr. */
87 /* Average packet length for mergeable receive buffers. */
88 struct ewma_pkt_len mrg_avg_pkt_len;
90 /* Page frag for packet buffer allocation. */
91 struct page_frag alloc_frag;
93 /* RX: fragments + linear part + virtio header */
94 struct scatterlist sg[MAX_SKB_FRAGS + 2];
96 /* Name of this receive queue: input.$index */
100 struct virtnet_info {
101 struct virtio_device *vdev;
102 struct virtqueue *cvq;
103 struct net_device *dev;
104 struct send_queue *sq;
105 struct receive_queue *rq;
108 /* Max # of queue pairs supported by the device */
111 /* # of queue pairs currently used by the driver */
112 u16 curr_queue_pairs;
114 /* I like... big packets and I cannot lie! */
117 /* Host will merge rx buffers for big packets (shake it! shake it!) */
118 bool mergeable_rx_bufs;
120 /* Has control virtqueue */
123 /* Host can handle any s/g split between our header and packet data */
126 /* Packet virtio header size */
129 /* Active statistics */
130 struct virtnet_stats __percpu *stats;
132 /* Work struct for refilling if we run low on memory. */
133 struct delayed_work refill;
135 /* Work struct for config space updates */
136 struct work_struct config_work;
138 /* Does the affinity hint is set for virtqueues? */
139 bool affinity_hint_set;
141 /* CPU hotplug instances for online & dead */
142 struct hlist_node node;
143 struct hlist_node node_dead;
145 /* Control VQ buffers: protected by the rtnl lock */
146 struct virtio_net_ctrl_hdr ctrl_hdr;
147 virtio_net_ctrl_ack ctrl_status;
148 struct virtio_net_ctrl_mq ctrl_mq;
153 /* Ethtool settings */
158 struct padded_vnet_hdr {
159 struct virtio_net_hdr_mrg_rxbuf hdr;
161 * hdr is in a separate sg buffer, and data sg buffer shares same page
162 * with this header sg. This padding makes next sg 16 byte aligned
168 /* Converting between virtqueue no. and kernel tx/rx queue no.
169 * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq
171 static int vq2txq(struct virtqueue *vq)
173 return (vq->index - 1) / 2;
176 static int txq2vq(int txq)
181 static int vq2rxq(struct virtqueue *vq)
183 return vq->index / 2;
186 static int rxq2vq(int rxq)
191 static inline struct virtio_net_hdr_mrg_rxbuf *skb_vnet_hdr(struct sk_buff *skb)
193 return (struct virtio_net_hdr_mrg_rxbuf *)skb->cb;
197 * private is used to chain pages for big packets, put the whole
198 * most recent used list in the beginning for reuse
200 static void give_pages(struct receive_queue *rq, struct page *page)
204 /* Find end of list, sew whole thing into vi->rq.pages. */
205 for (end = page; end->private; end = (struct page *)end->private);
206 end->private = (unsigned long)rq->pages;
210 static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask)
212 struct page *p = rq->pages;
215 rq->pages = (struct page *)p->private;
216 /* clear private here, it is used to chain pages */
219 p = alloc_page(gfp_mask);
223 static void skb_xmit_done(struct virtqueue *vq)
225 struct virtnet_info *vi = vq->vdev->priv;
227 /* Suppress further interrupts. */
228 virtqueue_disable_cb(vq);
230 /* We were probably waiting for more output buffers. */
231 netif_wake_subqueue(vi->dev, vq2txq(vq));
234 static unsigned int mergeable_ctx_to_buf_truesize(unsigned long mrg_ctx)
236 unsigned int truesize = mrg_ctx & (MERGEABLE_BUFFER_ALIGN - 1);
237 return (truesize + 1) * MERGEABLE_BUFFER_ALIGN;
240 static void *mergeable_ctx_to_buf_address(unsigned long mrg_ctx)
242 return (void *)(mrg_ctx & -MERGEABLE_BUFFER_ALIGN);
246 static unsigned long mergeable_buf_to_ctx(void *buf, unsigned int truesize)
248 unsigned int size = truesize / MERGEABLE_BUFFER_ALIGN;
249 return (unsigned long)buf | (size - 1);
252 /* Called from bottom half context */
253 static struct sk_buff *page_to_skb(struct virtnet_info *vi,
254 struct receive_queue *rq,
255 struct page *page, unsigned int offset,
256 unsigned int len, unsigned int truesize)
259 struct virtio_net_hdr_mrg_rxbuf *hdr;
260 unsigned int copy, hdr_len, hdr_padded_len;
263 p = page_address(page) + offset;
265 /* copy small packet so we can reuse these pages for small data */
266 skb = napi_alloc_skb(&rq->napi, GOOD_COPY_LEN);
270 hdr = skb_vnet_hdr(skb);
272 hdr_len = vi->hdr_len;
273 if (vi->mergeable_rx_bufs)
274 hdr_padded_len = sizeof *hdr;
276 hdr_padded_len = sizeof(struct padded_vnet_hdr);
278 memcpy(hdr, p, hdr_len);
281 offset += hdr_padded_len;
285 if (copy > skb_tailroom(skb))
286 copy = skb_tailroom(skb);
287 memcpy(skb_put(skb, copy), p, copy);
292 if (vi->mergeable_rx_bufs) {
294 skb_add_rx_frag(skb, 0, page, offset, len, truesize);
301 * Verify that we can indeed put this data into a skb.
302 * This is here to handle cases when the device erroneously
303 * tries to receive more than is possible. This is usually
304 * the case of a broken device.
306 if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) {
307 net_dbg_ratelimited("%s: too much data\n", skb->dev->name);
311 BUG_ON(offset >= PAGE_SIZE);
313 unsigned int frag_size = min((unsigned)PAGE_SIZE - offset, len);
314 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page, offset,
315 frag_size, truesize);
317 page = (struct page *)page->private;
322 give_pages(rq, page);
327 static struct sk_buff *receive_small(struct virtnet_info *vi, void *buf, unsigned int len)
329 struct sk_buff * skb = buf;
337 static struct sk_buff *receive_big(struct net_device *dev,
338 struct virtnet_info *vi,
339 struct receive_queue *rq,
343 struct page *page = buf;
344 struct sk_buff *skb = page_to_skb(vi, rq, page, 0, len, PAGE_SIZE);
352 dev->stats.rx_dropped++;
353 give_pages(rq, page);
357 static struct sk_buff *receive_mergeable(struct net_device *dev,
358 struct virtnet_info *vi,
359 struct receive_queue *rq,
363 void *buf = mergeable_ctx_to_buf_address(ctx);
364 struct virtio_net_hdr_mrg_rxbuf *hdr = buf;
365 u16 num_buf = virtio16_to_cpu(vi->vdev, hdr->num_buffers);
366 struct page *page = virt_to_head_page(buf);
367 int offset = buf - page_address(page);
368 unsigned int truesize = max(len, mergeable_ctx_to_buf_truesize(ctx));
370 struct sk_buff *head_skb = page_to_skb(vi, rq, page, offset, len,
372 struct sk_buff *curr_skb = head_skb;
374 if (unlikely(!curr_skb))
379 ctx = (unsigned long)virtqueue_get_buf(rq->vq, &len);
380 if (unlikely(!ctx)) {
381 pr_debug("%s: rx error: %d buffers out of %d missing\n",
383 virtio16_to_cpu(vi->vdev,
385 dev->stats.rx_length_errors++;
389 buf = mergeable_ctx_to_buf_address(ctx);
390 page = virt_to_head_page(buf);
392 num_skb_frags = skb_shinfo(curr_skb)->nr_frags;
393 if (unlikely(num_skb_frags == MAX_SKB_FRAGS)) {
394 struct sk_buff *nskb = alloc_skb(0, GFP_ATOMIC);
398 if (curr_skb == head_skb)
399 skb_shinfo(curr_skb)->frag_list = nskb;
401 curr_skb->next = nskb;
403 head_skb->truesize += nskb->truesize;
406 truesize = max(len, mergeable_ctx_to_buf_truesize(ctx));
407 if (curr_skb != head_skb) {
408 head_skb->data_len += len;
409 head_skb->len += len;
410 head_skb->truesize += truesize;
412 offset = buf - page_address(page);
413 if (skb_can_coalesce(curr_skb, num_skb_frags, page, offset)) {
415 skb_coalesce_rx_frag(curr_skb, num_skb_frags - 1,
418 skb_add_rx_frag(curr_skb, num_skb_frags, page,
419 offset, len, truesize);
423 ewma_pkt_len_add(&rq->mrg_avg_pkt_len, head_skb->len);
429 ctx = (unsigned long)virtqueue_get_buf(rq->vq, &len);
430 if (unlikely(!ctx)) {
431 pr_debug("%s: rx error: %d buffers missing\n",
433 dev->stats.rx_length_errors++;
436 page = virt_to_head_page(mergeable_ctx_to_buf_address(ctx));
440 dev->stats.rx_dropped++;
441 dev_kfree_skb(head_skb);
445 static void receive_buf(struct virtnet_info *vi, struct receive_queue *rq,
446 void *buf, unsigned int len)
448 struct net_device *dev = vi->dev;
449 struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
451 struct virtio_net_hdr_mrg_rxbuf *hdr;
453 if (unlikely(len < vi->hdr_len + ETH_HLEN)) {
454 pr_debug("%s: short packet %i\n", dev->name, len);
455 dev->stats.rx_length_errors++;
456 if (vi->mergeable_rx_bufs) {
457 unsigned long ctx = (unsigned long)buf;
458 void *base = mergeable_ctx_to_buf_address(ctx);
459 put_page(virt_to_head_page(base));
460 } else if (vi->big_packets) {
468 if (vi->mergeable_rx_bufs)
469 skb = receive_mergeable(dev, vi, rq, (unsigned long)buf, len);
470 else if (vi->big_packets)
471 skb = receive_big(dev, vi, rq, buf, len);
473 skb = receive_small(vi, buf, len);
478 hdr = skb_vnet_hdr(skb);
480 u64_stats_update_begin(&stats->rx_syncp);
481 stats->rx_bytes += skb->len;
483 u64_stats_update_end(&stats->rx_syncp);
485 if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID)
486 skb->ip_summed = CHECKSUM_UNNECESSARY;
488 if (virtio_net_hdr_to_skb(skb, &hdr->hdr,
489 virtio_is_little_endian(vi->vdev))) {
490 net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n",
491 dev->name, hdr->hdr.gso_type,
496 skb->protocol = eth_type_trans(skb, dev);
497 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
498 ntohs(skb->protocol), skb->len, skb->pkt_type);
500 napi_gro_receive(&rq->napi, skb);
504 dev->stats.rx_frame_errors++;
508 static int add_recvbuf_small(struct virtnet_info *vi, struct receive_queue *rq,
512 struct virtio_net_hdr_mrg_rxbuf *hdr;
515 skb = __netdev_alloc_skb_ip_align(vi->dev, GOOD_PACKET_LEN, gfp);
519 skb_put(skb, GOOD_PACKET_LEN);
521 hdr = skb_vnet_hdr(skb);
522 sg_init_table(rq->sg, 2);
523 sg_set_buf(rq->sg, hdr, vi->hdr_len);
524 skb_to_sgvec(skb, rq->sg + 1, 0, skb->len);
526 err = virtqueue_add_inbuf(rq->vq, rq->sg, 2, skb, gfp);
533 static int add_recvbuf_big(struct virtnet_info *vi, struct receive_queue *rq,
536 struct page *first, *list = NULL;
540 sg_init_table(rq->sg, MAX_SKB_FRAGS + 2);
542 /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */
543 for (i = MAX_SKB_FRAGS + 1; i > 1; --i) {
544 first = get_a_page(rq, gfp);
547 give_pages(rq, list);
550 sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE);
552 /* chain new page in list head to match sg */
553 first->private = (unsigned long)list;
557 first = get_a_page(rq, gfp);
559 give_pages(rq, list);
562 p = page_address(first);
564 /* rq->sg[0], rq->sg[1] share the same page */
565 /* a separated rq->sg[0] for header - required in case !any_header_sg */
566 sg_set_buf(&rq->sg[0], p, vi->hdr_len);
568 /* rq->sg[1] for data packet, from offset */
569 offset = sizeof(struct padded_vnet_hdr);
570 sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset);
572 /* chain first in list head */
573 first->private = (unsigned long)list;
574 err = virtqueue_add_inbuf(rq->vq, rq->sg, MAX_SKB_FRAGS + 2,
577 give_pages(rq, first);
582 static unsigned int get_mergeable_buf_len(struct ewma_pkt_len *avg_pkt_len)
584 const size_t hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
587 len = hdr_len + clamp_t(unsigned int, ewma_pkt_len_read(avg_pkt_len),
588 GOOD_PACKET_LEN, PAGE_SIZE - hdr_len);
589 return ALIGN(len, MERGEABLE_BUFFER_ALIGN);
592 static int add_recvbuf_mergeable(struct receive_queue *rq, gfp_t gfp)
594 struct page_frag *alloc_frag = &rq->alloc_frag;
598 unsigned int len, hole;
600 len = get_mergeable_buf_len(&rq->mrg_avg_pkt_len);
601 if (unlikely(!skb_page_frag_refill(len, alloc_frag, gfp)))
604 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
605 ctx = mergeable_buf_to_ctx(buf, len);
606 get_page(alloc_frag->page);
607 alloc_frag->offset += len;
608 hole = alloc_frag->size - alloc_frag->offset;
610 /* To avoid internal fragmentation, if there is very likely not
611 * enough space for another buffer, add the remaining space to
612 * the current buffer. This extra space is not included in
613 * the truesize stored in ctx.
616 alloc_frag->offset += hole;
619 sg_init_one(rq->sg, buf, len);
620 err = virtqueue_add_inbuf(rq->vq, rq->sg, 1, (void *)ctx, gfp);
622 put_page(virt_to_head_page(buf));
628 * Returns false if we couldn't fill entirely (OOM).
630 * Normally run in the receive path, but can also be run from ndo_open
631 * before we're receiving packets, or from refill_work which is
632 * careful to disable receiving (using napi_disable).
634 static bool try_fill_recv(struct virtnet_info *vi, struct receive_queue *rq,
642 if (vi->mergeable_rx_bufs)
643 err = add_recvbuf_mergeable(rq, gfp);
644 else if (vi->big_packets)
645 err = add_recvbuf_big(vi, rq, gfp);
647 err = add_recvbuf_small(vi, rq, gfp);
649 oom = err == -ENOMEM;
652 } while (rq->vq->num_free);
653 virtqueue_kick(rq->vq);
657 static void skb_recv_done(struct virtqueue *rvq)
659 struct virtnet_info *vi = rvq->vdev->priv;
660 struct receive_queue *rq = &vi->rq[vq2rxq(rvq)];
662 /* Schedule NAPI, Suppress further interrupts if successful. */
663 if (napi_schedule_prep(&rq->napi)) {
664 virtqueue_disable_cb(rvq);
665 __napi_schedule(&rq->napi);
669 static void virtnet_napi_enable(struct receive_queue *rq)
671 napi_enable(&rq->napi);
673 /* If all buffers were filled by other side before we napi_enabled, we
674 * won't get another interrupt, so process any outstanding packets
675 * now. virtnet_poll wants re-enable the queue, so we disable here.
676 * We synchronize against interrupts via NAPI_STATE_SCHED */
677 if (napi_schedule_prep(&rq->napi)) {
678 virtqueue_disable_cb(rq->vq);
680 __napi_schedule(&rq->napi);
685 static void refill_work(struct work_struct *work)
687 struct virtnet_info *vi =
688 container_of(work, struct virtnet_info, refill.work);
692 for (i = 0; i < vi->curr_queue_pairs; i++) {
693 struct receive_queue *rq = &vi->rq[i];
695 napi_disable(&rq->napi);
696 still_empty = !try_fill_recv(vi, rq, GFP_KERNEL);
697 virtnet_napi_enable(rq);
699 /* In theory, this can happen: if we don't get any buffers in
700 * we will *never* try to fill again.
703 schedule_delayed_work(&vi->refill, HZ/2);
707 static int virtnet_receive(struct receive_queue *rq, int budget)
709 struct virtnet_info *vi = rq->vq->vdev->priv;
710 unsigned int len, received = 0;
713 while (received < budget &&
714 (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) {
715 receive_buf(vi, rq, buf, len);
719 if (rq->vq->num_free > virtqueue_get_vring_size(rq->vq) / 2) {
720 if (!try_fill_recv(vi, rq, GFP_ATOMIC))
721 schedule_delayed_work(&vi->refill, 0);
727 static int virtnet_poll(struct napi_struct *napi, int budget)
729 struct receive_queue *rq =
730 container_of(napi, struct receive_queue, napi);
731 unsigned int r, received;
733 received = virtnet_receive(rq, budget);
735 /* Out of packets? */
736 if (received < budget) {
737 r = virtqueue_enable_cb_prepare(rq->vq);
738 napi_complete_done(napi, received);
739 if (unlikely(virtqueue_poll(rq->vq, r)) &&
740 napi_schedule_prep(napi)) {
741 virtqueue_disable_cb(rq->vq);
742 __napi_schedule(napi);
749 #ifdef CONFIG_NET_RX_BUSY_POLL
750 /* must be called with local_bh_disable()d */
751 static int virtnet_busy_poll(struct napi_struct *napi)
753 struct receive_queue *rq =
754 container_of(napi, struct receive_queue, napi);
755 struct virtnet_info *vi = rq->vq->vdev->priv;
756 int r, received = 0, budget = 4;
758 if (!(vi->status & VIRTIO_NET_S_LINK_UP))
759 return LL_FLUSH_FAILED;
761 if (!napi_schedule_prep(napi))
762 return LL_FLUSH_BUSY;
764 virtqueue_disable_cb(rq->vq);
767 received += virtnet_receive(rq, budget);
769 r = virtqueue_enable_cb_prepare(rq->vq);
770 clear_bit(NAPI_STATE_SCHED, &napi->state);
771 if (unlikely(virtqueue_poll(rq->vq, r)) &&
772 napi_schedule_prep(napi)) {
773 virtqueue_disable_cb(rq->vq);
774 if (received < budget) {
778 __napi_schedule(napi);
784 #endif /* CONFIG_NET_RX_BUSY_POLL */
786 static int virtnet_open(struct net_device *dev)
788 struct virtnet_info *vi = netdev_priv(dev);
791 for (i = 0; i < vi->max_queue_pairs; i++) {
792 if (i < vi->curr_queue_pairs)
793 /* Make sure we have some buffers: if oom use wq. */
794 if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL))
795 schedule_delayed_work(&vi->refill, 0);
796 virtnet_napi_enable(&vi->rq[i]);
802 static void free_old_xmit_skbs(struct send_queue *sq)
806 struct virtnet_info *vi = sq->vq->vdev->priv;
807 struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
809 while ((skb = virtqueue_get_buf(sq->vq, &len)) != NULL) {
810 pr_debug("Sent skb %p\n", skb);
812 u64_stats_update_begin(&stats->tx_syncp);
813 stats->tx_bytes += skb->len;
815 u64_stats_update_end(&stats->tx_syncp);
817 dev_kfree_skb_any(skb);
821 static int xmit_skb(struct send_queue *sq, struct sk_buff *skb)
823 struct virtio_net_hdr_mrg_rxbuf *hdr;
824 const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
825 struct virtnet_info *vi = sq->vq->vdev->priv;
827 unsigned hdr_len = vi->hdr_len;
830 pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
832 can_push = vi->any_header_sg &&
833 !((unsigned long)skb->data & (__alignof__(*hdr) - 1)) &&
834 !skb_header_cloned(skb) && skb_headroom(skb) >= hdr_len;
835 /* Even if we can, don't push here yet as this would skew
836 * csum_start offset below. */
838 hdr = (struct virtio_net_hdr_mrg_rxbuf *)(skb->data - hdr_len);
840 hdr = skb_vnet_hdr(skb);
842 if (virtio_net_hdr_from_skb(skb, &hdr->hdr,
843 virtio_is_little_endian(vi->vdev)))
846 if (vi->mergeable_rx_bufs)
847 hdr->num_buffers = 0;
849 sg_init_table(sq->sg, skb_shinfo(skb)->nr_frags + (can_push ? 1 : 2));
851 __skb_push(skb, hdr_len);
852 num_sg = skb_to_sgvec(skb, sq->sg, 0, skb->len);
853 /* Pull header back to avoid skew in tx bytes calculations. */
854 __skb_pull(skb, hdr_len);
856 sg_set_buf(sq->sg, hdr, hdr_len);
857 num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len) + 1;
859 return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC);
862 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
864 struct virtnet_info *vi = netdev_priv(dev);
865 int qnum = skb_get_queue_mapping(skb);
866 struct send_queue *sq = &vi->sq[qnum];
868 struct netdev_queue *txq = netdev_get_tx_queue(dev, qnum);
869 bool kick = !skb->xmit_more;
871 /* Free up any pending old buffers before queueing new ones. */
872 free_old_xmit_skbs(sq);
874 /* timestamp packet in software */
875 skb_tx_timestamp(skb);
877 /* Try to transmit */
878 err = xmit_skb(sq, skb);
880 /* This should not happen! */
882 dev->stats.tx_fifo_errors++;
885 "Unexpected TXQ (%d) queue failure: %d\n", qnum, err);
886 dev->stats.tx_dropped++;
887 dev_kfree_skb_any(skb);
891 /* Don't wait up for transmitted skbs to be freed. */
895 /* If running out of space, stop queue to avoid getting packets that we
896 * are then unable to transmit.
897 * An alternative would be to force queuing layer to requeue the skb by
898 * returning NETDEV_TX_BUSY. However, NETDEV_TX_BUSY should not be
899 * returned in a normal path of operation: it means that driver is not
900 * maintaining the TX queue stop/start state properly, and causes
901 * the stack to do a non-trivial amount of useless work.
902 * Since most packets only take 1 or 2 ring slots, stopping the queue
903 * early means 16 slots are typically wasted.
905 if (sq->vq->num_free < 2+MAX_SKB_FRAGS) {
906 netif_stop_subqueue(dev, qnum);
907 if (unlikely(!virtqueue_enable_cb_delayed(sq->vq))) {
908 /* More just got used, free them then recheck. */
909 free_old_xmit_skbs(sq);
910 if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) {
911 netif_start_subqueue(dev, qnum);
912 virtqueue_disable_cb(sq->vq);
917 if (kick || netif_xmit_stopped(txq))
918 virtqueue_kick(sq->vq);
924 * Send command via the control virtqueue and check status. Commands
925 * supported by the hypervisor, as indicated by feature bits, should
926 * never fail unless improperly formatted.
928 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
929 struct scatterlist *out)
931 struct scatterlist *sgs[4], hdr, stat;
932 unsigned out_num = 0, tmp;
934 /* Caller should know better */
935 BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ));
937 vi->ctrl_status = ~0;
938 vi->ctrl_hdr.class = class;
939 vi->ctrl_hdr.cmd = cmd;
941 sg_init_one(&hdr, &vi->ctrl_hdr, sizeof(vi->ctrl_hdr));
942 sgs[out_num++] = &hdr;
945 sgs[out_num++] = out;
947 /* Add return status. */
948 sg_init_one(&stat, &vi->ctrl_status, sizeof(vi->ctrl_status));
949 sgs[out_num] = &stat;
951 BUG_ON(out_num + 1 > ARRAY_SIZE(sgs));
952 virtqueue_add_sgs(vi->cvq, sgs, out_num, 1, vi, GFP_ATOMIC);
954 if (unlikely(!virtqueue_kick(vi->cvq)))
955 return vi->ctrl_status == VIRTIO_NET_OK;
957 /* Spin for a response, the kick causes an ioport write, trapping
958 * into the hypervisor, so the request should be handled immediately.
960 while (!virtqueue_get_buf(vi->cvq, &tmp) &&
961 !virtqueue_is_broken(vi->cvq))
964 return vi->ctrl_status == VIRTIO_NET_OK;
967 static int virtnet_set_mac_address(struct net_device *dev, void *p)
969 struct virtnet_info *vi = netdev_priv(dev);
970 struct virtio_device *vdev = vi->vdev;
972 struct sockaddr *addr;
973 struct scatterlist sg;
975 addr = kmalloc(sizeof(*addr), GFP_KERNEL);
978 memcpy(addr, p, sizeof(*addr));
980 ret = eth_prepare_mac_addr_change(dev, addr);
984 if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
985 sg_init_one(&sg, addr->sa_data, dev->addr_len);
986 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
987 VIRTIO_NET_CTRL_MAC_ADDR_SET, &sg)) {
989 "Failed to set mac address by vq command.\n");
993 } else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC) &&
994 !virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) {
997 /* Naturally, this has an atomicity problem. */
998 for (i = 0; i < dev->addr_len; i++)
1000 offsetof(struct virtio_net_config, mac) +
1001 i, addr->sa_data[i]);
1004 eth_commit_mac_addr_change(dev, p);
1012 static struct rtnl_link_stats64 *virtnet_stats(struct net_device *dev,
1013 struct rtnl_link_stats64 *tot)
1015 struct virtnet_info *vi = netdev_priv(dev);
1019 for_each_possible_cpu(cpu) {
1020 struct virtnet_stats *stats = per_cpu_ptr(vi->stats, cpu);
1021 u64 tpackets, tbytes, rpackets, rbytes;
1024 start = u64_stats_fetch_begin_irq(&stats->tx_syncp);
1025 tpackets = stats->tx_packets;
1026 tbytes = stats->tx_bytes;
1027 } while (u64_stats_fetch_retry_irq(&stats->tx_syncp, start));
1030 start = u64_stats_fetch_begin_irq(&stats->rx_syncp);
1031 rpackets = stats->rx_packets;
1032 rbytes = stats->rx_bytes;
1033 } while (u64_stats_fetch_retry_irq(&stats->rx_syncp, start));
1035 tot->rx_packets += rpackets;
1036 tot->tx_packets += tpackets;
1037 tot->rx_bytes += rbytes;
1038 tot->tx_bytes += tbytes;
1041 tot->tx_dropped = dev->stats.tx_dropped;
1042 tot->tx_fifo_errors = dev->stats.tx_fifo_errors;
1043 tot->rx_dropped = dev->stats.rx_dropped;
1044 tot->rx_length_errors = dev->stats.rx_length_errors;
1045 tot->rx_frame_errors = dev->stats.rx_frame_errors;
1050 #ifdef CONFIG_NET_POLL_CONTROLLER
1051 static void virtnet_netpoll(struct net_device *dev)
1053 struct virtnet_info *vi = netdev_priv(dev);
1056 for (i = 0; i < vi->curr_queue_pairs; i++)
1057 napi_schedule(&vi->rq[i].napi);
1061 static void virtnet_ack_link_announce(struct virtnet_info *vi)
1064 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE,
1065 VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL))
1066 dev_warn(&vi->dev->dev, "Failed to ack link announce.\n");
1070 static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs)
1072 struct scatterlist sg;
1073 struct net_device *dev = vi->dev;
1075 if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ))
1078 vi->ctrl_mq.virtqueue_pairs = cpu_to_virtio16(vi->vdev, queue_pairs);
1079 sg_init_one(&sg, &vi->ctrl_mq, sizeof(vi->ctrl_mq));
1081 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ,
1082 VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg)) {
1083 dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n",
1087 vi->curr_queue_pairs = queue_pairs;
1088 /* virtnet_open() will refill when device is going to up. */
1089 if (dev->flags & IFF_UP)
1090 schedule_delayed_work(&vi->refill, 0);
1096 static int virtnet_close(struct net_device *dev)
1098 struct virtnet_info *vi = netdev_priv(dev);
1101 /* Make sure refill_work doesn't re-enable napi! */
1102 cancel_delayed_work_sync(&vi->refill);
1104 for (i = 0; i < vi->max_queue_pairs; i++)
1105 napi_disable(&vi->rq[i].napi);
1110 static void virtnet_set_rx_mode(struct net_device *dev)
1112 struct virtnet_info *vi = netdev_priv(dev);
1113 struct scatterlist sg[2];
1114 struct virtio_net_ctrl_mac *mac_data;
1115 struct netdev_hw_addr *ha;
1121 /* We can't dynamically set ndo_set_rx_mode, so return gracefully */
1122 if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
1125 vi->ctrl_promisc = ((dev->flags & IFF_PROMISC) != 0);
1126 vi->ctrl_allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
1128 sg_init_one(sg, &vi->ctrl_promisc, sizeof(vi->ctrl_promisc));
1130 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
1131 VIRTIO_NET_CTRL_RX_PROMISC, sg))
1132 dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
1133 vi->ctrl_promisc ? "en" : "dis");
1135 sg_init_one(sg, &vi->ctrl_allmulti, sizeof(vi->ctrl_allmulti));
1137 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
1138 VIRTIO_NET_CTRL_RX_ALLMULTI, sg))
1139 dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
1140 vi->ctrl_allmulti ? "en" : "dis");
1142 uc_count = netdev_uc_count(dev);
1143 mc_count = netdev_mc_count(dev);
1144 /* MAC filter - use one buffer for both lists */
1145 buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
1146 (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
1151 sg_init_table(sg, 2);
1153 /* Store the unicast list and count in the front of the buffer */
1154 mac_data->entries = cpu_to_virtio32(vi->vdev, uc_count);
1156 netdev_for_each_uc_addr(ha, dev)
1157 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
1159 sg_set_buf(&sg[0], mac_data,
1160 sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
1162 /* multicast list and count fill the end */
1163 mac_data = (void *)&mac_data->macs[uc_count][0];
1165 mac_data->entries = cpu_to_virtio32(vi->vdev, mc_count);
1167 netdev_for_each_mc_addr(ha, dev)
1168 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
1170 sg_set_buf(&sg[1], mac_data,
1171 sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
1173 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
1174 VIRTIO_NET_CTRL_MAC_TABLE_SET, sg))
1175 dev_warn(&dev->dev, "Failed to set MAC filter table.\n");
1180 static int virtnet_vlan_rx_add_vid(struct net_device *dev,
1181 __be16 proto, u16 vid)
1183 struct virtnet_info *vi = netdev_priv(dev);
1184 struct scatterlist sg;
1187 sg_init_one(&sg, &vi->ctrl_vid, sizeof(vi->ctrl_vid));
1189 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1190 VIRTIO_NET_CTRL_VLAN_ADD, &sg))
1191 dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
1195 static int virtnet_vlan_rx_kill_vid(struct net_device *dev,
1196 __be16 proto, u16 vid)
1198 struct virtnet_info *vi = netdev_priv(dev);
1199 struct scatterlist sg;
1202 sg_init_one(&sg, &vi->ctrl_vid, sizeof(vi->ctrl_vid));
1204 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1205 VIRTIO_NET_CTRL_VLAN_DEL, &sg))
1206 dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
1210 static void virtnet_clean_affinity(struct virtnet_info *vi, long hcpu)
1214 if (vi->affinity_hint_set) {
1215 for (i = 0; i < vi->max_queue_pairs; i++) {
1216 virtqueue_set_affinity(vi->rq[i].vq, -1);
1217 virtqueue_set_affinity(vi->sq[i].vq, -1);
1220 vi->affinity_hint_set = false;
1224 static void virtnet_set_affinity(struct virtnet_info *vi)
1229 /* In multiqueue mode, when the number of cpu is equal to the number of
1230 * queue pairs, we let the queue pairs to be private to one cpu by
1231 * setting the affinity hint to eliminate the contention.
1233 if (vi->curr_queue_pairs == 1 ||
1234 vi->max_queue_pairs != num_online_cpus()) {
1235 virtnet_clean_affinity(vi, -1);
1240 for_each_online_cpu(cpu) {
1241 virtqueue_set_affinity(vi->rq[i].vq, cpu);
1242 virtqueue_set_affinity(vi->sq[i].vq, cpu);
1243 netif_set_xps_queue(vi->dev, cpumask_of(cpu), i);
1247 vi->affinity_hint_set = true;
1250 static int virtnet_cpu_online(unsigned int cpu, struct hlist_node *node)
1252 struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
1254 virtnet_set_affinity(vi);
1258 static int virtnet_cpu_dead(unsigned int cpu, struct hlist_node *node)
1260 struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
1262 virtnet_set_affinity(vi);
1266 static int virtnet_cpu_down_prep(unsigned int cpu, struct hlist_node *node)
1268 struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
1271 virtnet_clean_affinity(vi, cpu);
1275 static enum cpuhp_state virtionet_online;
1277 static int virtnet_cpu_notif_add(struct virtnet_info *vi)
1281 ret = cpuhp_state_add_instance_nocalls(virtionet_online, &vi->node);
1284 ret = cpuhp_state_add_instance_nocalls(CPUHP_VIRT_NET_DEAD,
1288 cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node);
1292 static void virtnet_cpu_notif_remove(struct virtnet_info *vi)
1294 cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node);
1295 cpuhp_state_remove_instance_nocalls(CPUHP_VIRT_NET_DEAD,
1299 static void virtnet_get_ringparam(struct net_device *dev,
1300 struct ethtool_ringparam *ring)
1302 struct virtnet_info *vi = netdev_priv(dev);
1304 ring->rx_max_pending = virtqueue_get_vring_size(vi->rq[0].vq);
1305 ring->tx_max_pending = virtqueue_get_vring_size(vi->sq[0].vq);
1306 ring->rx_pending = ring->rx_max_pending;
1307 ring->tx_pending = ring->tx_max_pending;
1311 static void virtnet_get_drvinfo(struct net_device *dev,
1312 struct ethtool_drvinfo *info)
1314 struct virtnet_info *vi = netdev_priv(dev);
1315 struct virtio_device *vdev = vi->vdev;
1317 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
1318 strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version));
1319 strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info));
1323 /* TODO: Eliminate OOO packets during switching */
1324 static int virtnet_set_channels(struct net_device *dev,
1325 struct ethtool_channels *channels)
1327 struct virtnet_info *vi = netdev_priv(dev);
1328 u16 queue_pairs = channels->combined_count;
1331 /* We don't support separate rx/tx channels.
1332 * We don't allow setting 'other' channels.
1334 if (channels->rx_count || channels->tx_count || channels->other_count)
1337 if (queue_pairs > vi->max_queue_pairs || queue_pairs == 0)
1341 err = virtnet_set_queues(vi, queue_pairs);
1343 netif_set_real_num_tx_queues(dev, queue_pairs);
1344 netif_set_real_num_rx_queues(dev, queue_pairs);
1346 virtnet_set_affinity(vi);
1353 static void virtnet_get_channels(struct net_device *dev,
1354 struct ethtool_channels *channels)
1356 struct virtnet_info *vi = netdev_priv(dev);
1358 channels->combined_count = vi->curr_queue_pairs;
1359 channels->max_combined = vi->max_queue_pairs;
1360 channels->max_other = 0;
1361 channels->rx_count = 0;
1362 channels->tx_count = 0;
1363 channels->other_count = 0;
1366 /* Check if the user is trying to change anything besides speed/duplex */
1367 static bool virtnet_validate_ethtool_cmd(const struct ethtool_cmd *cmd)
1369 struct ethtool_cmd diff1 = *cmd;
1370 struct ethtool_cmd diff2 = {};
1372 /* cmd is always set so we need to clear it, validate the port type
1373 * and also without autonegotiation we can ignore advertising
1375 ethtool_cmd_speed_set(&diff1, 0);
1376 diff2.port = PORT_OTHER;
1377 diff1.advertising = 0;
1381 return !memcmp(&diff1, &diff2, sizeof(diff1));
1384 static int virtnet_set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1386 struct virtnet_info *vi = netdev_priv(dev);
1389 speed = ethtool_cmd_speed(cmd);
1390 /* don't allow custom speed and duplex */
1391 if (!ethtool_validate_speed(speed) ||
1392 !ethtool_validate_duplex(cmd->duplex) ||
1393 !virtnet_validate_ethtool_cmd(cmd))
1396 vi->duplex = cmd->duplex;
1401 static int virtnet_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1403 struct virtnet_info *vi = netdev_priv(dev);
1405 ethtool_cmd_speed_set(cmd, vi->speed);
1406 cmd->duplex = vi->duplex;
1407 cmd->port = PORT_OTHER;
1412 static void virtnet_init_settings(struct net_device *dev)
1414 struct virtnet_info *vi = netdev_priv(dev);
1416 vi->speed = SPEED_UNKNOWN;
1417 vi->duplex = DUPLEX_UNKNOWN;
1420 static const struct ethtool_ops virtnet_ethtool_ops = {
1421 .get_drvinfo = virtnet_get_drvinfo,
1422 .get_link = ethtool_op_get_link,
1423 .get_ringparam = virtnet_get_ringparam,
1424 .set_channels = virtnet_set_channels,
1425 .get_channels = virtnet_get_channels,
1426 .get_ts_info = ethtool_op_get_ts_info,
1427 .get_settings = virtnet_get_settings,
1428 .set_settings = virtnet_set_settings,
1431 static const struct net_device_ops virtnet_netdev = {
1432 .ndo_open = virtnet_open,
1433 .ndo_stop = virtnet_close,
1434 .ndo_start_xmit = start_xmit,
1435 .ndo_validate_addr = eth_validate_addr,
1436 .ndo_set_mac_address = virtnet_set_mac_address,
1437 .ndo_set_rx_mode = virtnet_set_rx_mode,
1438 .ndo_get_stats64 = virtnet_stats,
1439 .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
1440 .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
1441 #ifdef CONFIG_NET_POLL_CONTROLLER
1442 .ndo_poll_controller = virtnet_netpoll,
1444 #ifdef CONFIG_NET_RX_BUSY_POLL
1445 .ndo_busy_poll = virtnet_busy_poll,
1449 static void virtnet_config_changed_work(struct work_struct *work)
1451 struct virtnet_info *vi =
1452 container_of(work, struct virtnet_info, config_work);
1455 if (virtio_cread_feature(vi->vdev, VIRTIO_NET_F_STATUS,
1456 struct virtio_net_config, status, &v) < 0)
1459 if (v & VIRTIO_NET_S_ANNOUNCE) {
1460 netdev_notify_peers(vi->dev);
1461 virtnet_ack_link_announce(vi);
1464 /* Ignore unknown (future) status bits */
1465 v &= VIRTIO_NET_S_LINK_UP;
1467 if (vi->status == v)
1472 if (vi->status & VIRTIO_NET_S_LINK_UP) {
1473 netif_carrier_on(vi->dev);
1474 netif_tx_wake_all_queues(vi->dev);
1476 netif_carrier_off(vi->dev);
1477 netif_tx_stop_all_queues(vi->dev);
1481 static void virtnet_config_changed(struct virtio_device *vdev)
1483 struct virtnet_info *vi = vdev->priv;
1485 schedule_work(&vi->config_work);
1488 static void virtnet_free_queues(struct virtnet_info *vi)
1492 for (i = 0; i < vi->max_queue_pairs; i++) {
1493 napi_hash_del(&vi->rq[i].napi);
1494 netif_napi_del(&vi->rq[i].napi);
1497 /* We called napi_hash_del() before netif_napi_del(),
1498 * we need to respect an RCU grace period before freeing vi->rq
1506 static void free_receive_bufs(struct virtnet_info *vi)
1510 for (i = 0; i < vi->max_queue_pairs; i++) {
1511 while (vi->rq[i].pages)
1512 __free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0);
1516 static void free_receive_page_frags(struct virtnet_info *vi)
1519 for (i = 0; i < vi->max_queue_pairs; i++)
1520 if (vi->rq[i].alloc_frag.page)
1521 put_page(vi->rq[i].alloc_frag.page);
1524 static void free_unused_bufs(struct virtnet_info *vi)
1529 for (i = 0; i < vi->max_queue_pairs; i++) {
1530 struct virtqueue *vq = vi->sq[i].vq;
1531 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL)
1535 for (i = 0; i < vi->max_queue_pairs; i++) {
1536 struct virtqueue *vq = vi->rq[i].vq;
1538 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) {
1539 if (vi->mergeable_rx_bufs) {
1540 unsigned long ctx = (unsigned long)buf;
1541 void *base = mergeable_ctx_to_buf_address(ctx);
1542 put_page(virt_to_head_page(base));
1543 } else if (vi->big_packets) {
1544 give_pages(&vi->rq[i], buf);
1552 static void virtnet_del_vqs(struct virtnet_info *vi)
1554 struct virtio_device *vdev = vi->vdev;
1556 virtnet_clean_affinity(vi, -1);
1558 vdev->config->del_vqs(vdev);
1560 virtnet_free_queues(vi);
1563 static int virtnet_find_vqs(struct virtnet_info *vi)
1565 vq_callback_t **callbacks;
1566 struct virtqueue **vqs;
1571 /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
1572 * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
1573 * possible control vq.
1575 total_vqs = vi->max_queue_pairs * 2 +
1576 virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ);
1578 /* Allocate space for find_vqs parameters */
1579 vqs = kzalloc(total_vqs * sizeof(*vqs), GFP_KERNEL);
1582 callbacks = kmalloc(total_vqs * sizeof(*callbacks), GFP_KERNEL);
1585 names = kmalloc(total_vqs * sizeof(*names), GFP_KERNEL);
1589 /* Parameters for control virtqueue, if any */
1591 callbacks[total_vqs - 1] = NULL;
1592 names[total_vqs - 1] = "control";
1595 /* Allocate/initialize parameters for send/receive virtqueues */
1596 for (i = 0; i < vi->max_queue_pairs; i++) {
1597 callbacks[rxq2vq(i)] = skb_recv_done;
1598 callbacks[txq2vq(i)] = skb_xmit_done;
1599 sprintf(vi->rq[i].name, "input.%d", i);
1600 sprintf(vi->sq[i].name, "output.%d", i);
1601 names[rxq2vq(i)] = vi->rq[i].name;
1602 names[txq2vq(i)] = vi->sq[i].name;
1605 ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs, callbacks,
1611 vi->cvq = vqs[total_vqs - 1];
1612 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
1613 vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
1616 for (i = 0; i < vi->max_queue_pairs; i++) {
1617 vi->rq[i].vq = vqs[rxq2vq(i)];
1618 vi->sq[i].vq = vqs[txq2vq(i)];
1637 static int virtnet_alloc_queues(struct virtnet_info *vi)
1641 vi->sq = kzalloc(sizeof(*vi->sq) * vi->max_queue_pairs, GFP_KERNEL);
1644 vi->rq = kzalloc(sizeof(*vi->rq) * vi->max_queue_pairs, GFP_KERNEL);
1648 INIT_DELAYED_WORK(&vi->refill, refill_work);
1649 for (i = 0; i < vi->max_queue_pairs; i++) {
1650 vi->rq[i].pages = NULL;
1651 netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll,
1654 sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg));
1655 ewma_pkt_len_init(&vi->rq[i].mrg_avg_pkt_len);
1656 sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg));
1667 static int init_vqs(struct virtnet_info *vi)
1671 /* Allocate send & receive queues */
1672 ret = virtnet_alloc_queues(vi);
1676 ret = virtnet_find_vqs(vi);
1681 virtnet_set_affinity(vi);
1687 virtnet_free_queues(vi);
1693 static ssize_t mergeable_rx_buffer_size_show(struct netdev_rx_queue *queue,
1694 struct rx_queue_attribute *attribute, char *buf)
1696 struct virtnet_info *vi = netdev_priv(queue->dev);
1697 unsigned int queue_index = get_netdev_rx_queue_index(queue);
1698 struct ewma_pkt_len *avg;
1700 BUG_ON(queue_index >= vi->max_queue_pairs);
1701 avg = &vi->rq[queue_index].mrg_avg_pkt_len;
1702 return sprintf(buf, "%u\n", get_mergeable_buf_len(avg));
1705 static struct rx_queue_attribute mergeable_rx_buffer_size_attribute =
1706 __ATTR_RO(mergeable_rx_buffer_size);
1708 static struct attribute *virtio_net_mrg_rx_attrs[] = {
1709 &mergeable_rx_buffer_size_attribute.attr,
1713 static const struct attribute_group virtio_net_mrg_rx_group = {
1714 .name = "virtio_net",
1715 .attrs = virtio_net_mrg_rx_attrs
1719 static bool virtnet_fail_on_feature(struct virtio_device *vdev,
1721 const char *fname, const char *dname)
1723 if (!virtio_has_feature(vdev, fbit))
1726 dev_err(&vdev->dev, "device advertises feature %s but not %s",
1732 #define VIRTNET_FAIL_ON(vdev, fbit, dbit) \
1733 virtnet_fail_on_feature(vdev, fbit, #fbit, dbit)
1735 static bool virtnet_validate_features(struct virtio_device *vdev)
1737 if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) &&
1738 (VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_RX,
1739 "VIRTIO_NET_F_CTRL_VQ") ||
1740 VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_VLAN,
1741 "VIRTIO_NET_F_CTRL_VQ") ||
1742 VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE,
1743 "VIRTIO_NET_F_CTRL_VQ") ||
1744 VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_MQ, "VIRTIO_NET_F_CTRL_VQ") ||
1745 VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR,
1746 "VIRTIO_NET_F_CTRL_VQ"))) {
1753 #define MIN_MTU ETH_MIN_MTU
1754 #define MAX_MTU ETH_MAX_MTU
1756 static int virtnet_probe(struct virtio_device *vdev)
1759 struct net_device *dev;
1760 struct virtnet_info *vi;
1761 u16 max_queue_pairs;
1764 if (!vdev->config->get) {
1765 dev_err(&vdev->dev, "%s failure: config access disabled\n",
1770 if (!virtnet_validate_features(vdev))
1773 /* Find if host supports multiqueue virtio_net device */
1774 err = virtio_cread_feature(vdev, VIRTIO_NET_F_MQ,
1775 struct virtio_net_config,
1776 max_virtqueue_pairs, &max_queue_pairs);
1778 /* We need at least 2 queue's */
1779 if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
1780 max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
1781 !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
1782 max_queue_pairs = 1;
1784 /* Allocate ourselves a network device with room for our info */
1785 dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs);
1789 /* Set up network device as normal. */
1790 dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE;
1791 dev->netdev_ops = &virtnet_netdev;
1792 dev->features = NETIF_F_HIGHDMA;
1794 dev->ethtool_ops = &virtnet_ethtool_ops;
1795 SET_NETDEV_DEV(dev, &vdev->dev);
1797 /* Do we support "hardware" checksums? */
1798 if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
1799 /* This opens up the world of extra features. */
1800 dev->hw_features |= NETIF_F_HW_CSUM | NETIF_F_SG;
1802 dev->features |= NETIF_F_HW_CSUM | NETIF_F_SG;
1804 if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
1805 dev->hw_features |= NETIF_F_TSO | NETIF_F_UFO
1806 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
1808 /* Individual feature bits: what can host handle? */
1809 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
1810 dev->hw_features |= NETIF_F_TSO;
1811 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
1812 dev->hw_features |= NETIF_F_TSO6;
1813 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
1814 dev->hw_features |= NETIF_F_TSO_ECN;
1815 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
1816 dev->hw_features |= NETIF_F_UFO;
1818 dev->features |= NETIF_F_GSO_ROBUST;
1821 dev->features |= dev->hw_features & (NETIF_F_ALL_TSO|NETIF_F_UFO);
1822 /* (!csum && gso) case will be fixed by register_netdev() */
1824 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM))
1825 dev->features |= NETIF_F_RXCSUM;
1827 dev->vlan_features = dev->features;
1829 /* MTU range: 68 - 65535 */
1830 dev->min_mtu = MIN_MTU;
1831 dev->max_mtu = MAX_MTU;
1833 /* Configuration may specify what MAC to use. Otherwise random. */
1834 if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
1835 virtio_cread_bytes(vdev,
1836 offsetof(struct virtio_net_config, mac),
1837 dev->dev_addr, dev->addr_len);
1839 eth_hw_addr_random(dev);
1841 /* Set up our device-specific information */
1842 vi = netdev_priv(dev);
1846 vi->stats = alloc_percpu(struct virtnet_stats);
1848 if (vi->stats == NULL)
1851 for_each_possible_cpu(i) {
1852 struct virtnet_stats *virtnet_stats;
1853 virtnet_stats = per_cpu_ptr(vi->stats, i);
1854 u64_stats_init(&virtnet_stats->tx_syncp);
1855 u64_stats_init(&virtnet_stats->rx_syncp);
1858 INIT_WORK(&vi->config_work, virtnet_config_changed_work);
1860 /* If we can receive ANY GSO packets, we must allocate large ones. */
1861 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
1862 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
1863 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN) ||
1864 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_UFO))
1865 vi->big_packets = true;
1867 if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
1868 vi->mergeable_rx_bufs = true;
1870 if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF) ||
1871 virtio_has_feature(vdev, VIRTIO_F_VERSION_1))
1872 vi->hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1874 vi->hdr_len = sizeof(struct virtio_net_hdr);
1876 if (virtio_has_feature(vdev, VIRTIO_F_ANY_LAYOUT) ||
1877 virtio_has_feature(vdev, VIRTIO_F_VERSION_1))
1878 vi->any_header_sg = true;
1880 if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
1883 if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) {
1884 mtu = virtio_cread16(vdev,
1885 offsetof(struct virtio_net_config,
1887 if (mtu < dev->min_mtu) {
1888 __virtio_clear_bit(vdev, VIRTIO_NET_F_MTU);
1895 if (vi->any_header_sg)
1896 dev->needed_headroom = vi->hdr_len;
1898 /* Enable multiqueue by default */
1899 if (num_online_cpus() >= max_queue_pairs)
1900 vi->curr_queue_pairs = max_queue_pairs;
1902 vi->curr_queue_pairs = num_online_cpus();
1903 vi->max_queue_pairs = max_queue_pairs;
1905 /* Allocate/initialize the rx/tx queues, and invoke find_vqs */
1911 if (vi->mergeable_rx_bufs)
1912 dev->sysfs_rx_queue_group = &virtio_net_mrg_rx_group;
1914 netif_set_real_num_tx_queues(dev, vi->curr_queue_pairs);
1915 netif_set_real_num_rx_queues(dev, vi->curr_queue_pairs);
1917 virtnet_init_settings(dev);
1919 err = register_netdev(dev);
1921 pr_debug("virtio_net: registering device failed\n");
1925 virtio_device_ready(vdev);
1927 err = virtnet_cpu_notif_add(vi);
1929 pr_debug("virtio_net: registering cpu notifier failed\n");
1930 goto free_unregister_netdev;
1933 virtnet_set_affinity(vi);
1935 /* Assume link up if device can't report link status,
1936 otherwise get link status from config. */
1937 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
1938 netif_carrier_off(dev);
1939 schedule_work(&vi->config_work);
1941 vi->status = VIRTIO_NET_S_LINK_UP;
1942 netif_carrier_on(dev);
1945 pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
1946 dev->name, max_queue_pairs);
1950 free_unregister_netdev:
1951 vi->vdev->config->reset(vdev);
1953 unregister_netdev(dev);
1955 cancel_delayed_work_sync(&vi->refill);
1956 free_receive_page_frags(vi);
1957 virtnet_del_vqs(vi);
1959 free_percpu(vi->stats);
1965 static void remove_vq_common(struct virtnet_info *vi)
1967 vi->vdev->config->reset(vi->vdev);
1969 /* Free unused buffers in both send and recv, if any. */
1970 free_unused_bufs(vi);
1972 free_receive_bufs(vi);
1974 free_receive_page_frags(vi);
1976 virtnet_del_vqs(vi);
1979 static void virtnet_remove(struct virtio_device *vdev)
1981 struct virtnet_info *vi = vdev->priv;
1983 virtnet_cpu_notif_remove(vi);
1985 /* Make sure no work handler is accessing the device. */
1986 flush_work(&vi->config_work);
1988 unregister_netdev(vi->dev);
1990 remove_vq_common(vi);
1992 free_percpu(vi->stats);
1993 free_netdev(vi->dev);
1996 #ifdef CONFIG_PM_SLEEP
1997 static int virtnet_freeze(struct virtio_device *vdev)
1999 struct virtnet_info *vi = vdev->priv;
2002 virtnet_cpu_notif_remove(vi);
2004 /* Make sure no work handler is accessing the device */
2005 flush_work(&vi->config_work);
2007 netif_device_detach(vi->dev);
2008 cancel_delayed_work_sync(&vi->refill);
2010 if (netif_running(vi->dev)) {
2011 for (i = 0; i < vi->max_queue_pairs; i++)
2012 napi_disable(&vi->rq[i].napi);
2015 remove_vq_common(vi);
2020 static int virtnet_restore(struct virtio_device *vdev)
2022 struct virtnet_info *vi = vdev->priv;
2029 virtio_device_ready(vdev);
2031 if (netif_running(vi->dev)) {
2032 for (i = 0; i < vi->curr_queue_pairs; i++)
2033 if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL))
2034 schedule_delayed_work(&vi->refill, 0);
2036 for (i = 0; i < vi->max_queue_pairs; i++)
2037 virtnet_napi_enable(&vi->rq[i]);
2040 netif_device_attach(vi->dev);
2043 virtnet_set_queues(vi, vi->curr_queue_pairs);
2046 err = virtnet_cpu_notif_add(vi);
2054 static struct virtio_device_id id_table[] = {
2055 { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
2059 #define VIRTNET_FEATURES \
2060 VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM, \
2062 VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6, \
2063 VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, \
2064 VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO, \
2065 VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ, \
2066 VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, \
2067 VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ, \
2068 VIRTIO_NET_F_CTRL_MAC_ADDR, \
2071 static unsigned int features[] = {
2075 static unsigned int features_legacy[] = {
2078 VIRTIO_F_ANY_LAYOUT,
2081 static struct virtio_driver virtio_net_driver = {
2082 .feature_table = features,
2083 .feature_table_size = ARRAY_SIZE(features),
2084 .feature_table_legacy = features_legacy,
2085 .feature_table_size_legacy = ARRAY_SIZE(features_legacy),
2086 .driver.name = KBUILD_MODNAME,
2087 .driver.owner = THIS_MODULE,
2088 .id_table = id_table,
2089 .probe = virtnet_probe,
2090 .remove = virtnet_remove,
2091 .config_changed = virtnet_config_changed,
2092 #ifdef CONFIG_PM_SLEEP
2093 .freeze = virtnet_freeze,
2094 .restore = virtnet_restore,
2098 static __init int virtio_net_driver_init(void)
2102 ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "AP_VIRT_NET_ONLINE",
2104 virtnet_cpu_down_prep);
2107 virtionet_online = ret;
2108 ret = cpuhp_setup_state_multi(CPUHP_VIRT_NET_DEAD, "VIRT_NET_DEAD",
2109 NULL, virtnet_cpu_dead);
2113 ret = register_virtio_driver(&virtio_net_driver);
2118 cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD);
2120 cpuhp_remove_multi_state(virtionet_online);
2124 module_init(virtio_net_driver_init);
2126 static __exit void virtio_net_driver_exit(void)
2128 cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD);
2129 cpuhp_remove_multi_state(virtionet_online);
2130 unregister_virtio_driver(&virtio_net_driver);
2132 module_exit(virtio_net_driver_exit);
2134 MODULE_DEVICE_TABLE(virtio, id_table);
2135 MODULE_DESCRIPTION("Virtio network driver");
2136 MODULE_LICENSE("GPL");