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 /* With mergeable buffers we align buffer address and use the low bits to
51 * encode its true size. Buffer size is up to 1 page so we need to align to
52 * square root of page size to ensure we reserve enough bits to encode the true
55 #define MERGEABLE_BUFFER_MIN_ALIGN_SHIFT ((PAGE_SHIFT + 1) / 2)
57 /* Minimum alignment for mergeable packet buffers. */
58 #define MERGEABLE_BUFFER_ALIGN max(L1_CACHE_BYTES, \
59 1 << MERGEABLE_BUFFER_MIN_ALIGN_SHIFT)
61 #define VIRTNET_DRIVER_VERSION "1.0.0"
63 struct virtnet_stats {
64 struct u64_stats_sync tx_syncp;
65 struct u64_stats_sync rx_syncp;
73 /* Internal representation of a send virtqueue */
75 /* Virtqueue associated with this send _queue */
78 /* TX: fragments + linear part + virtio header */
79 struct scatterlist sg[MAX_SKB_FRAGS + 2];
81 /* Name of the send queue: output.$index */
85 /* Internal representation of a receive virtqueue */
86 struct receive_queue {
87 /* Virtqueue associated with this receive_queue */
90 struct napi_struct napi;
92 /* Chain pages by the private ptr. */
95 /* Average packet length for mergeable receive buffers. */
96 struct ewma_pkt_len mrg_avg_pkt_len;
98 /* Page frag for packet buffer allocation. */
99 struct page_frag alloc_frag;
101 /* RX: fragments + linear part + virtio header */
102 struct scatterlist sg[MAX_SKB_FRAGS + 2];
104 /* Name of this receive queue: input.$index */
108 struct virtnet_info {
109 struct virtio_device *vdev;
110 struct virtqueue *cvq;
111 struct net_device *dev;
112 struct send_queue *sq;
113 struct receive_queue *rq;
116 /* Max # of queue pairs supported by the device */
119 /* # of queue pairs currently used by the driver */
120 u16 curr_queue_pairs;
122 /* I like... big packets and I cannot lie! */
125 /* Host will merge rx buffers for big packets (shake it! shake it!) */
126 bool mergeable_rx_bufs;
128 /* Has control virtqueue */
131 /* Host can handle any s/g split between our header and packet data */
134 /* Packet virtio header size */
137 /* Active statistics */
138 struct virtnet_stats __percpu *stats;
140 /* Work struct for refilling if we run low on memory. */
141 struct delayed_work refill;
143 /* Work struct for config space updates */
144 struct work_struct config_work;
146 /* Does the affinity hint is set for virtqueues? */
147 bool affinity_hint_set;
149 /* CPU hotplug instances for online & dead */
150 struct hlist_node node;
151 struct hlist_node node_dead;
153 /* Control VQ buffers: protected by the rtnl lock */
154 struct virtio_net_ctrl_hdr ctrl_hdr;
155 virtio_net_ctrl_ack ctrl_status;
156 struct virtio_net_ctrl_mq ctrl_mq;
161 /* Ethtool settings */
166 struct padded_vnet_hdr {
167 struct virtio_net_hdr_mrg_rxbuf hdr;
169 * hdr is in a separate sg buffer, and data sg buffer shares same page
170 * with this header sg. This padding makes next sg 16 byte aligned
176 /* Converting between virtqueue no. and kernel tx/rx queue no.
177 * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq
179 static int vq2txq(struct virtqueue *vq)
181 return (vq->index - 1) / 2;
184 static int txq2vq(int txq)
189 static int vq2rxq(struct virtqueue *vq)
191 return vq->index / 2;
194 static int rxq2vq(int rxq)
199 static inline struct virtio_net_hdr_mrg_rxbuf *skb_vnet_hdr(struct sk_buff *skb)
201 return (struct virtio_net_hdr_mrg_rxbuf *)skb->cb;
205 * private is used to chain pages for big packets, put the whole
206 * most recent used list in the beginning for reuse
208 static void give_pages(struct receive_queue *rq, struct page *page)
212 /* Find end of list, sew whole thing into vi->rq.pages. */
213 for (end = page; end->private; end = (struct page *)end->private);
214 end->private = (unsigned long)rq->pages;
218 static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask)
220 struct page *p = rq->pages;
223 rq->pages = (struct page *)p->private;
224 /* clear private here, it is used to chain pages */
227 p = alloc_page(gfp_mask);
231 static void skb_xmit_done(struct virtqueue *vq)
233 struct virtnet_info *vi = vq->vdev->priv;
235 /* Suppress further interrupts. */
236 virtqueue_disable_cb(vq);
238 /* We were probably waiting for more output buffers. */
239 netif_wake_subqueue(vi->dev, vq2txq(vq));
242 static unsigned int mergeable_ctx_to_buf_truesize(unsigned long mrg_ctx)
244 unsigned int truesize = mrg_ctx & (MERGEABLE_BUFFER_ALIGN - 1);
245 return (truesize + 1) * MERGEABLE_BUFFER_ALIGN;
248 static void *mergeable_ctx_to_buf_address(unsigned long mrg_ctx)
250 return (void *)(mrg_ctx & -MERGEABLE_BUFFER_ALIGN);
254 static unsigned long mergeable_buf_to_ctx(void *buf, unsigned int truesize)
256 unsigned int size = truesize / MERGEABLE_BUFFER_ALIGN;
257 return (unsigned long)buf | (size - 1);
260 /* Called from bottom half context */
261 static struct sk_buff *page_to_skb(struct virtnet_info *vi,
262 struct receive_queue *rq,
263 struct page *page, unsigned int offset,
264 unsigned int len, unsigned int truesize)
267 struct virtio_net_hdr_mrg_rxbuf *hdr;
268 unsigned int copy, hdr_len, hdr_padded_len;
271 p = page_address(page) + offset;
273 /* copy small packet so we can reuse these pages for small data */
274 skb = napi_alloc_skb(&rq->napi, GOOD_COPY_LEN);
278 hdr = skb_vnet_hdr(skb);
280 hdr_len = vi->hdr_len;
281 if (vi->mergeable_rx_bufs)
282 hdr_padded_len = sizeof *hdr;
284 hdr_padded_len = sizeof(struct padded_vnet_hdr);
286 memcpy(hdr, p, hdr_len);
289 offset += hdr_padded_len;
293 if (copy > skb_tailroom(skb))
294 copy = skb_tailroom(skb);
295 memcpy(skb_put(skb, copy), p, copy);
300 if (vi->mergeable_rx_bufs) {
302 skb_add_rx_frag(skb, 0, page, offset, len, truesize);
309 * Verify that we can indeed put this data into a skb.
310 * This is here to handle cases when the device erroneously
311 * tries to receive more than is possible. This is usually
312 * the case of a broken device.
314 if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) {
315 net_dbg_ratelimited("%s: too much data\n", skb->dev->name);
319 BUG_ON(offset >= PAGE_SIZE);
321 unsigned int frag_size = min((unsigned)PAGE_SIZE - offset, len);
322 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page, offset,
323 frag_size, truesize);
325 page = (struct page *)page->private;
330 give_pages(rq, page);
335 static struct sk_buff *receive_small(struct virtnet_info *vi, void *buf, unsigned int len)
337 struct sk_buff * skb = buf;
345 static struct sk_buff *receive_big(struct net_device *dev,
346 struct virtnet_info *vi,
347 struct receive_queue *rq,
351 struct page *page = buf;
352 struct sk_buff *skb = page_to_skb(vi, rq, page, 0, len, PAGE_SIZE);
360 dev->stats.rx_dropped++;
361 give_pages(rq, page);
365 static struct sk_buff *receive_mergeable(struct net_device *dev,
366 struct virtnet_info *vi,
367 struct receive_queue *rq,
371 void *buf = mergeable_ctx_to_buf_address(ctx);
372 struct virtio_net_hdr_mrg_rxbuf *hdr = buf;
373 u16 num_buf = virtio16_to_cpu(vi->vdev, hdr->num_buffers);
374 struct page *page = virt_to_head_page(buf);
375 int offset = buf - page_address(page);
376 unsigned int truesize = max(len, mergeable_ctx_to_buf_truesize(ctx));
378 struct sk_buff *head_skb = page_to_skb(vi, rq, page, offset, len,
380 struct sk_buff *curr_skb = head_skb;
382 if (unlikely(!curr_skb))
387 ctx = (unsigned long)virtqueue_get_buf(rq->vq, &len);
388 if (unlikely(!ctx)) {
389 pr_debug("%s: rx error: %d buffers out of %d missing\n",
391 virtio16_to_cpu(vi->vdev,
393 dev->stats.rx_length_errors++;
397 buf = mergeable_ctx_to_buf_address(ctx);
398 page = virt_to_head_page(buf);
400 num_skb_frags = skb_shinfo(curr_skb)->nr_frags;
401 if (unlikely(num_skb_frags == MAX_SKB_FRAGS)) {
402 struct sk_buff *nskb = alloc_skb(0, GFP_ATOMIC);
406 if (curr_skb == head_skb)
407 skb_shinfo(curr_skb)->frag_list = nskb;
409 curr_skb->next = nskb;
411 head_skb->truesize += nskb->truesize;
414 truesize = max(len, mergeable_ctx_to_buf_truesize(ctx));
415 if (curr_skb != head_skb) {
416 head_skb->data_len += len;
417 head_skb->len += len;
418 head_skb->truesize += truesize;
420 offset = buf - page_address(page);
421 if (skb_can_coalesce(curr_skb, num_skb_frags, page, offset)) {
423 skb_coalesce_rx_frag(curr_skb, num_skb_frags - 1,
426 skb_add_rx_frag(curr_skb, num_skb_frags, page,
427 offset, len, truesize);
431 ewma_pkt_len_add(&rq->mrg_avg_pkt_len, head_skb->len);
437 ctx = (unsigned long)virtqueue_get_buf(rq->vq, &len);
438 if (unlikely(!ctx)) {
439 pr_debug("%s: rx error: %d buffers missing\n",
441 dev->stats.rx_length_errors++;
444 page = virt_to_head_page(mergeable_ctx_to_buf_address(ctx));
448 dev->stats.rx_dropped++;
449 dev_kfree_skb(head_skb);
453 static void receive_buf(struct virtnet_info *vi, struct receive_queue *rq,
454 void *buf, unsigned int len)
456 struct net_device *dev = vi->dev;
457 struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
459 struct virtio_net_hdr_mrg_rxbuf *hdr;
461 if (unlikely(len < vi->hdr_len + ETH_HLEN)) {
462 pr_debug("%s: short packet %i\n", dev->name, len);
463 dev->stats.rx_length_errors++;
464 if (vi->mergeable_rx_bufs) {
465 unsigned long ctx = (unsigned long)buf;
466 void *base = mergeable_ctx_to_buf_address(ctx);
467 put_page(virt_to_head_page(base));
468 } else if (vi->big_packets) {
476 if (vi->mergeable_rx_bufs)
477 skb = receive_mergeable(dev, vi, rq, (unsigned long)buf, len);
478 else if (vi->big_packets)
479 skb = receive_big(dev, vi, rq, buf, len);
481 skb = receive_small(vi, buf, len);
486 hdr = skb_vnet_hdr(skb);
488 u64_stats_update_begin(&stats->rx_syncp);
489 stats->rx_bytes += skb->len;
491 u64_stats_update_end(&stats->rx_syncp);
493 if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID)
494 skb->ip_summed = CHECKSUM_UNNECESSARY;
496 if (virtio_net_hdr_to_skb(skb, &hdr->hdr,
497 virtio_is_little_endian(vi->vdev))) {
498 net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n",
499 dev->name, hdr->hdr.gso_type,
504 skb->protocol = eth_type_trans(skb, dev);
505 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
506 ntohs(skb->protocol), skb->len, skb->pkt_type);
508 napi_gro_receive(&rq->napi, skb);
512 dev->stats.rx_frame_errors++;
516 static int add_recvbuf_small(struct virtnet_info *vi, struct receive_queue *rq,
520 struct virtio_net_hdr_mrg_rxbuf *hdr;
523 skb = __netdev_alloc_skb_ip_align(vi->dev, GOOD_PACKET_LEN, gfp);
527 skb_put(skb, GOOD_PACKET_LEN);
529 hdr = skb_vnet_hdr(skb);
530 sg_init_table(rq->sg, 2);
531 sg_set_buf(rq->sg, hdr, vi->hdr_len);
533 err = skb_to_sgvec(skb, rq->sg + 1, 0, skb->len);
534 if (unlikely(err < 0)) {
539 err = virtqueue_add_inbuf(rq->vq, rq->sg, 2, skb, gfp);
546 static int add_recvbuf_big(struct virtnet_info *vi, struct receive_queue *rq,
549 struct page *first, *list = NULL;
553 sg_init_table(rq->sg, MAX_SKB_FRAGS + 2);
555 /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */
556 for (i = MAX_SKB_FRAGS + 1; i > 1; --i) {
557 first = get_a_page(rq, gfp);
560 give_pages(rq, list);
563 sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE);
565 /* chain new page in list head to match sg */
566 first->private = (unsigned long)list;
570 first = get_a_page(rq, gfp);
572 give_pages(rq, list);
575 p = page_address(first);
577 /* rq->sg[0], rq->sg[1] share the same page */
578 /* a separated rq->sg[0] for header - required in case !any_header_sg */
579 sg_set_buf(&rq->sg[0], p, vi->hdr_len);
581 /* rq->sg[1] for data packet, from offset */
582 offset = sizeof(struct padded_vnet_hdr);
583 sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset);
585 /* chain first in list head */
586 first->private = (unsigned long)list;
587 err = virtqueue_add_inbuf(rq->vq, rq->sg, MAX_SKB_FRAGS + 2,
590 give_pages(rq, first);
595 static unsigned int get_mergeable_buf_len(struct ewma_pkt_len *avg_pkt_len)
597 const size_t hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
600 len = hdr_len + clamp_t(unsigned int, ewma_pkt_len_read(avg_pkt_len),
601 GOOD_PACKET_LEN, PAGE_SIZE - hdr_len);
602 return ALIGN(len, MERGEABLE_BUFFER_ALIGN);
605 static int add_recvbuf_mergeable(struct receive_queue *rq, gfp_t gfp)
607 struct page_frag *alloc_frag = &rq->alloc_frag;
611 unsigned int len, hole;
613 len = get_mergeable_buf_len(&rq->mrg_avg_pkt_len);
614 if (unlikely(!skb_page_frag_refill(len, alloc_frag, gfp)))
617 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
618 ctx = mergeable_buf_to_ctx(buf, len);
619 get_page(alloc_frag->page);
620 alloc_frag->offset += len;
621 hole = alloc_frag->size - alloc_frag->offset;
623 /* To avoid internal fragmentation, if there is very likely not
624 * enough space for another buffer, add the remaining space to
625 * the current buffer. This extra space is not included in
626 * the truesize stored in ctx.
629 alloc_frag->offset += hole;
632 sg_init_one(rq->sg, buf, len);
633 err = virtqueue_add_inbuf(rq->vq, rq->sg, 1, (void *)ctx, gfp);
635 put_page(virt_to_head_page(buf));
641 * Returns false if we couldn't fill entirely (OOM).
643 * Normally run in the receive path, but can also be run from ndo_open
644 * before we're receiving packets, or from refill_work which is
645 * careful to disable receiving (using napi_disable).
647 static bool try_fill_recv(struct virtnet_info *vi, struct receive_queue *rq,
655 if (vi->mergeable_rx_bufs)
656 err = add_recvbuf_mergeable(rq, gfp);
657 else if (vi->big_packets)
658 err = add_recvbuf_big(vi, rq, gfp);
660 err = add_recvbuf_small(vi, rq, gfp);
662 oom = err == -ENOMEM;
665 } while (rq->vq->num_free);
666 virtqueue_kick(rq->vq);
670 static void skb_recv_done(struct virtqueue *rvq)
672 struct virtnet_info *vi = rvq->vdev->priv;
673 struct receive_queue *rq = &vi->rq[vq2rxq(rvq)];
675 /* Schedule NAPI, Suppress further interrupts if successful. */
676 if (napi_schedule_prep(&rq->napi)) {
677 virtqueue_disable_cb(rvq);
678 __napi_schedule(&rq->napi);
682 static void virtnet_napi_enable(struct receive_queue *rq)
684 napi_enable(&rq->napi);
686 /* If all buffers were filled by other side before we napi_enabled, we
687 * won't get another interrupt, so process any outstanding packets
688 * now. virtnet_poll wants re-enable the queue, so we disable here.
689 * We synchronize against interrupts via NAPI_STATE_SCHED */
690 if (napi_schedule_prep(&rq->napi)) {
691 virtqueue_disable_cb(rq->vq);
693 __napi_schedule(&rq->napi);
698 static void refill_work(struct work_struct *work)
700 struct virtnet_info *vi =
701 container_of(work, struct virtnet_info, refill.work);
705 for (i = 0; i < vi->curr_queue_pairs; i++) {
706 struct receive_queue *rq = &vi->rq[i];
708 napi_disable(&rq->napi);
709 still_empty = !try_fill_recv(vi, rq, GFP_KERNEL);
710 virtnet_napi_enable(rq);
712 /* In theory, this can happen: if we don't get any buffers in
713 * we will *never* try to fill again.
716 schedule_delayed_work(&vi->refill, HZ/2);
720 static int virtnet_receive(struct receive_queue *rq, int budget)
722 struct virtnet_info *vi = rq->vq->vdev->priv;
723 unsigned int len, received = 0;
726 while (received < budget &&
727 (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) {
728 receive_buf(vi, rq, buf, len);
732 if (rq->vq->num_free > virtqueue_get_vring_size(rq->vq) / 2) {
733 if (!try_fill_recv(vi, rq, GFP_ATOMIC))
734 schedule_delayed_work(&vi->refill, 0);
740 static int virtnet_poll(struct napi_struct *napi, int budget)
742 struct receive_queue *rq =
743 container_of(napi, struct receive_queue, napi);
744 unsigned int r, received;
746 received = virtnet_receive(rq, budget);
748 /* Out of packets? */
749 if (received < budget) {
750 r = virtqueue_enable_cb_prepare(rq->vq);
751 napi_complete_done(napi, received);
752 if (unlikely(virtqueue_poll(rq->vq, r)) &&
753 napi_schedule_prep(napi)) {
754 virtqueue_disable_cb(rq->vq);
755 __napi_schedule(napi);
762 #ifdef CONFIG_NET_RX_BUSY_POLL
763 /* must be called with local_bh_disable()d */
764 static int virtnet_busy_poll(struct napi_struct *napi)
766 struct receive_queue *rq =
767 container_of(napi, struct receive_queue, napi);
768 struct virtnet_info *vi = rq->vq->vdev->priv;
769 int r, received = 0, budget = 4;
771 if (!(vi->status & VIRTIO_NET_S_LINK_UP))
772 return LL_FLUSH_FAILED;
774 if (!napi_schedule_prep(napi))
775 return LL_FLUSH_BUSY;
777 virtqueue_disable_cb(rq->vq);
780 received += virtnet_receive(rq, budget);
782 r = virtqueue_enable_cb_prepare(rq->vq);
783 clear_bit(NAPI_STATE_SCHED, &napi->state);
784 if (unlikely(virtqueue_poll(rq->vq, r)) &&
785 napi_schedule_prep(napi)) {
786 virtqueue_disable_cb(rq->vq);
787 if (received < budget) {
791 __napi_schedule(napi);
797 #endif /* CONFIG_NET_RX_BUSY_POLL */
799 static int virtnet_open(struct net_device *dev)
801 struct virtnet_info *vi = netdev_priv(dev);
804 for (i = 0; i < vi->max_queue_pairs; i++) {
805 if (i < vi->curr_queue_pairs)
806 /* Make sure we have some buffers: if oom use wq. */
807 if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL))
808 schedule_delayed_work(&vi->refill, 0);
809 virtnet_napi_enable(&vi->rq[i]);
815 static void free_old_xmit_skbs(struct send_queue *sq)
819 struct virtnet_info *vi = sq->vq->vdev->priv;
820 struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
822 while ((skb = virtqueue_get_buf(sq->vq, &len)) != NULL) {
823 pr_debug("Sent skb %p\n", skb);
825 u64_stats_update_begin(&stats->tx_syncp);
826 stats->tx_bytes += skb->len;
828 u64_stats_update_end(&stats->tx_syncp);
830 dev_kfree_skb_any(skb);
834 static int xmit_skb(struct send_queue *sq, struct sk_buff *skb)
836 struct virtio_net_hdr_mrg_rxbuf *hdr;
837 const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
838 struct virtnet_info *vi = sq->vq->vdev->priv;
840 unsigned hdr_len = vi->hdr_len;
843 pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
845 can_push = vi->any_header_sg &&
846 !((unsigned long)skb->data & (__alignof__(*hdr) - 1)) &&
847 !skb_header_cloned(skb) && skb_headroom(skb) >= hdr_len;
848 /* Even if we can, don't push here yet as this would skew
849 * csum_start offset below. */
851 hdr = (struct virtio_net_hdr_mrg_rxbuf *)(skb->data - hdr_len);
853 hdr = skb_vnet_hdr(skb);
855 if (virtio_net_hdr_from_skb(skb, &hdr->hdr,
856 virtio_is_little_endian(vi->vdev), false))
859 if (vi->mergeable_rx_bufs)
860 hdr->num_buffers = 0;
862 sg_init_table(sq->sg, skb_shinfo(skb)->nr_frags + (can_push ? 1 : 2));
864 __skb_push(skb, hdr_len);
865 num_sg = skb_to_sgvec(skb, sq->sg, 0, skb->len);
866 if (unlikely(num_sg < 0))
868 /* Pull header back to avoid skew in tx bytes calculations. */
869 __skb_pull(skb, hdr_len);
871 sg_set_buf(sq->sg, hdr, hdr_len);
872 num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len);
873 if (unlikely(num_sg < 0))
877 return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC);
880 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
882 struct virtnet_info *vi = netdev_priv(dev);
883 int qnum = skb_get_queue_mapping(skb);
884 struct send_queue *sq = &vi->sq[qnum];
886 struct netdev_queue *txq = netdev_get_tx_queue(dev, qnum);
887 bool kick = !skb->xmit_more;
889 /* Free up any pending old buffers before queueing new ones. */
890 free_old_xmit_skbs(sq);
892 /* timestamp packet in software */
893 skb_tx_timestamp(skb);
895 /* Try to transmit */
896 err = xmit_skb(sq, skb);
898 /* This should not happen! */
900 dev->stats.tx_fifo_errors++;
903 "Unexpected TXQ (%d) queue failure: %d\n", qnum, err);
904 dev->stats.tx_dropped++;
905 dev_kfree_skb_any(skb);
909 /* Don't wait up for transmitted skbs to be freed. */
913 /* If running out of space, stop queue to avoid getting packets that we
914 * are then unable to transmit.
915 * An alternative would be to force queuing layer to requeue the skb by
916 * returning NETDEV_TX_BUSY. However, NETDEV_TX_BUSY should not be
917 * returned in a normal path of operation: it means that driver is not
918 * maintaining the TX queue stop/start state properly, and causes
919 * the stack to do a non-trivial amount of useless work.
920 * Since most packets only take 1 or 2 ring slots, stopping the queue
921 * early means 16 slots are typically wasted.
923 if (sq->vq->num_free < 2+MAX_SKB_FRAGS) {
924 netif_stop_subqueue(dev, qnum);
925 if (unlikely(!virtqueue_enable_cb_delayed(sq->vq))) {
926 /* More just got used, free them then recheck. */
927 free_old_xmit_skbs(sq);
928 if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) {
929 netif_start_subqueue(dev, qnum);
930 virtqueue_disable_cb(sq->vq);
935 if (kick || netif_xmit_stopped(txq))
936 virtqueue_kick(sq->vq);
942 * Send command via the control virtqueue and check status. Commands
943 * supported by the hypervisor, as indicated by feature bits, should
944 * never fail unless improperly formatted.
946 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
947 struct scatterlist *out)
949 struct scatterlist *sgs[4], hdr, stat;
950 unsigned out_num = 0, tmp;
952 /* Caller should know better */
953 BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ));
955 vi->ctrl_status = ~0;
956 vi->ctrl_hdr.class = class;
957 vi->ctrl_hdr.cmd = cmd;
959 sg_init_one(&hdr, &vi->ctrl_hdr, sizeof(vi->ctrl_hdr));
960 sgs[out_num++] = &hdr;
963 sgs[out_num++] = out;
965 /* Add return status. */
966 sg_init_one(&stat, &vi->ctrl_status, sizeof(vi->ctrl_status));
967 sgs[out_num] = &stat;
969 BUG_ON(out_num + 1 > ARRAY_SIZE(sgs));
970 virtqueue_add_sgs(vi->cvq, sgs, out_num, 1, vi, GFP_ATOMIC);
972 if (unlikely(!virtqueue_kick(vi->cvq)))
973 return vi->ctrl_status == VIRTIO_NET_OK;
975 /* Spin for a response, the kick causes an ioport write, trapping
976 * into the hypervisor, so the request should be handled immediately.
978 while (!virtqueue_get_buf(vi->cvq, &tmp) &&
979 !virtqueue_is_broken(vi->cvq))
982 return vi->ctrl_status == VIRTIO_NET_OK;
985 static int virtnet_set_mac_address(struct net_device *dev, void *p)
987 struct virtnet_info *vi = netdev_priv(dev);
988 struct virtio_device *vdev = vi->vdev;
990 struct sockaddr *addr;
991 struct scatterlist sg;
993 addr = kmalloc(sizeof(*addr), GFP_KERNEL);
996 memcpy(addr, p, sizeof(*addr));
998 ret = eth_prepare_mac_addr_change(dev, addr);
1002 if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1003 sg_init_one(&sg, addr->sa_data, dev->addr_len);
1004 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
1005 VIRTIO_NET_CTRL_MAC_ADDR_SET, &sg)) {
1006 dev_warn(&vdev->dev,
1007 "Failed to set mac address by vq command.\n");
1011 } else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC) &&
1012 !virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) {
1015 /* Naturally, this has an atomicity problem. */
1016 for (i = 0; i < dev->addr_len; i++)
1017 virtio_cwrite8(vdev,
1018 offsetof(struct virtio_net_config, mac) +
1019 i, addr->sa_data[i]);
1022 eth_commit_mac_addr_change(dev, p);
1030 static struct rtnl_link_stats64 *virtnet_stats(struct net_device *dev,
1031 struct rtnl_link_stats64 *tot)
1033 struct virtnet_info *vi = netdev_priv(dev);
1037 for_each_possible_cpu(cpu) {
1038 struct virtnet_stats *stats = per_cpu_ptr(vi->stats, cpu);
1039 u64 tpackets, tbytes, rpackets, rbytes;
1042 start = u64_stats_fetch_begin_irq(&stats->tx_syncp);
1043 tpackets = stats->tx_packets;
1044 tbytes = stats->tx_bytes;
1045 } while (u64_stats_fetch_retry_irq(&stats->tx_syncp, start));
1048 start = u64_stats_fetch_begin_irq(&stats->rx_syncp);
1049 rpackets = stats->rx_packets;
1050 rbytes = stats->rx_bytes;
1051 } while (u64_stats_fetch_retry_irq(&stats->rx_syncp, start));
1053 tot->rx_packets += rpackets;
1054 tot->tx_packets += tpackets;
1055 tot->rx_bytes += rbytes;
1056 tot->tx_bytes += tbytes;
1059 tot->tx_dropped = dev->stats.tx_dropped;
1060 tot->tx_fifo_errors = dev->stats.tx_fifo_errors;
1061 tot->rx_dropped = dev->stats.rx_dropped;
1062 tot->rx_length_errors = dev->stats.rx_length_errors;
1063 tot->rx_frame_errors = dev->stats.rx_frame_errors;
1068 #ifdef CONFIG_NET_POLL_CONTROLLER
1069 static void virtnet_netpoll(struct net_device *dev)
1071 struct virtnet_info *vi = netdev_priv(dev);
1074 for (i = 0; i < vi->curr_queue_pairs; i++)
1075 napi_schedule(&vi->rq[i].napi);
1079 static void virtnet_ack_link_announce(struct virtnet_info *vi)
1082 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE,
1083 VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL))
1084 dev_warn(&vi->dev->dev, "Failed to ack link announce.\n");
1088 static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs)
1090 struct scatterlist sg;
1091 struct net_device *dev = vi->dev;
1093 if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ))
1096 vi->ctrl_mq.virtqueue_pairs = cpu_to_virtio16(vi->vdev, queue_pairs);
1097 sg_init_one(&sg, &vi->ctrl_mq, sizeof(vi->ctrl_mq));
1099 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ,
1100 VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg)) {
1101 dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n",
1105 vi->curr_queue_pairs = queue_pairs;
1106 /* virtnet_open() will refill when device is going to up. */
1107 if (dev->flags & IFF_UP)
1108 schedule_delayed_work(&vi->refill, 0);
1114 static int virtnet_close(struct net_device *dev)
1116 struct virtnet_info *vi = netdev_priv(dev);
1119 /* Make sure refill_work doesn't re-enable napi! */
1120 cancel_delayed_work_sync(&vi->refill);
1122 for (i = 0; i < vi->max_queue_pairs; i++)
1123 napi_disable(&vi->rq[i].napi);
1128 static void virtnet_set_rx_mode(struct net_device *dev)
1130 struct virtnet_info *vi = netdev_priv(dev);
1131 struct scatterlist sg[2];
1132 struct virtio_net_ctrl_mac *mac_data;
1133 struct netdev_hw_addr *ha;
1139 /* We can't dynamically set ndo_set_rx_mode, so return gracefully */
1140 if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
1143 vi->ctrl_promisc = ((dev->flags & IFF_PROMISC) != 0);
1144 vi->ctrl_allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
1146 sg_init_one(sg, &vi->ctrl_promisc, sizeof(vi->ctrl_promisc));
1148 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
1149 VIRTIO_NET_CTRL_RX_PROMISC, sg))
1150 dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
1151 vi->ctrl_promisc ? "en" : "dis");
1153 sg_init_one(sg, &vi->ctrl_allmulti, sizeof(vi->ctrl_allmulti));
1155 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
1156 VIRTIO_NET_CTRL_RX_ALLMULTI, sg))
1157 dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
1158 vi->ctrl_allmulti ? "en" : "dis");
1160 uc_count = netdev_uc_count(dev);
1161 mc_count = netdev_mc_count(dev);
1162 /* MAC filter - use one buffer for both lists */
1163 buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
1164 (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
1169 sg_init_table(sg, 2);
1171 /* Store the unicast list and count in the front of the buffer */
1172 mac_data->entries = cpu_to_virtio32(vi->vdev, uc_count);
1174 netdev_for_each_uc_addr(ha, dev)
1175 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
1177 sg_set_buf(&sg[0], mac_data,
1178 sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
1180 /* multicast list and count fill the end */
1181 mac_data = (void *)&mac_data->macs[uc_count][0];
1183 mac_data->entries = cpu_to_virtio32(vi->vdev, mc_count);
1185 netdev_for_each_mc_addr(ha, dev)
1186 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
1188 sg_set_buf(&sg[1], mac_data,
1189 sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
1191 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
1192 VIRTIO_NET_CTRL_MAC_TABLE_SET, sg))
1193 dev_warn(&dev->dev, "Failed to set MAC filter table.\n");
1198 static int virtnet_vlan_rx_add_vid(struct net_device *dev,
1199 __be16 proto, u16 vid)
1201 struct virtnet_info *vi = netdev_priv(dev);
1202 struct scatterlist sg;
1205 sg_init_one(&sg, &vi->ctrl_vid, sizeof(vi->ctrl_vid));
1207 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1208 VIRTIO_NET_CTRL_VLAN_ADD, &sg))
1209 dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
1213 static int virtnet_vlan_rx_kill_vid(struct net_device *dev,
1214 __be16 proto, u16 vid)
1216 struct virtnet_info *vi = netdev_priv(dev);
1217 struct scatterlist sg;
1220 sg_init_one(&sg, &vi->ctrl_vid, sizeof(vi->ctrl_vid));
1222 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1223 VIRTIO_NET_CTRL_VLAN_DEL, &sg))
1224 dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
1228 static void virtnet_clean_affinity(struct virtnet_info *vi, long hcpu)
1232 if (vi->affinity_hint_set) {
1233 for (i = 0; i < vi->max_queue_pairs; i++) {
1234 virtqueue_set_affinity(vi->rq[i].vq, -1);
1235 virtqueue_set_affinity(vi->sq[i].vq, -1);
1238 vi->affinity_hint_set = false;
1242 static void virtnet_set_affinity(struct virtnet_info *vi)
1247 /* In multiqueue mode, when the number of cpu is equal to the number of
1248 * queue pairs, we let the queue pairs to be private to one cpu by
1249 * setting the affinity hint to eliminate the contention.
1251 if (vi->curr_queue_pairs == 1 ||
1252 vi->max_queue_pairs != num_online_cpus()) {
1253 virtnet_clean_affinity(vi, -1);
1258 for_each_online_cpu(cpu) {
1259 virtqueue_set_affinity(vi->rq[i].vq, cpu);
1260 virtqueue_set_affinity(vi->sq[i].vq, cpu);
1261 netif_set_xps_queue(vi->dev, cpumask_of(cpu), i);
1265 vi->affinity_hint_set = true;
1268 static int virtnet_cpu_online(unsigned int cpu, struct hlist_node *node)
1270 struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
1272 virtnet_set_affinity(vi);
1276 static int virtnet_cpu_dead(unsigned int cpu, struct hlist_node *node)
1278 struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
1280 virtnet_set_affinity(vi);
1284 static int virtnet_cpu_down_prep(unsigned int cpu, struct hlist_node *node)
1286 struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
1289 virtnet_clean_affinity(vi, cpu);
1293 static enum cpuhp_state virtionet_online;
1295 static int virtnet_cpu_notif_add(struct virtnet_info *vi)
1299 ret = cpuhp_state_add_instance_nocalls(virtionet_online, &vi->node);
1302 ret = cpuhp_state_add_instance_nocalls(CPUHP_VIRT_NET_DEAD,
1306 cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node);
1310 static void virtnet_cpu_notif_remove(struct virtnet_info *vi)
1312 cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node);
1313 cpuhp_state_remove_instance_nocalls(CPUHP_VIRT_NET_DEAD,
1317 static void virtnet_get_ringparam(struct net_device *dev,
1318 struct ethtool_ringparam *ring)
1320 struct virtnet_info *vi = netdev_priv(dev);
1322 ring->rx_max_pending = virtqueue_get_vring_size(vi->rq[0].vq);
1323 ring->tx_max_pending = virtqueue_get_vring_size(vi->sq[0].vq);
1324 ring->rx_pending = ring->rx_max_pending;
1325 ring->tx_pending = ring->tx_max_pending;
1329 static void virtnet_get_drvinfo(struct net_device *dev,
1330 struct ethtool_drvinfo *info)
1332 struct virtnet_info *vi = netdev_priv(dev);
1333 struct virtio_device *vdev = vi->vdev;
1335 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
1336 strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version));
1337 strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info));
1341 /* TODO: Eliminate OOO packets during switching */
1342 static int virtnet_set_channels(struct net_device *dev,
1343 struct ethtool_channels *channels)
1345 struct virtnet_info *vi = netdev_priv(dev);
1346 u16 queue_pairs = channels->combined_count;
1349 /* We don't support separate rx/tx channels.
1350 * We don't allow setting 'other' channels.
1352 if (channels->rx_count || channels->tx_count || channels->other_count)
1355 if (queue_pairs > vi->max_queue_pairs || queue_pairs == 0)
1359 err = virtnet_set_queues(vi, queue_pairs);
1364 virtnet_set_affinity(vi);
1367 netif_set_real_num_tx_queues(dev, queue_pairs);
1368 netif_set_real_num_rx_queues(dev, queue_pairs);
1373 static void virtnet_get_channels(struct net_device *dev,
1374 struct ethtool_channels *channels)
1376 struct virtnet_info *vi = netdev_priv(dev);
1378 channels->combined_count = vi->curr_queue_pairs;
1379 channels->max_combined = vi->max_queue_pairs;
1380 channels->max_other = 0;
1381 channels->rx_count = 0;
1382 channels->tx_count = 0;
1383 channels->other_count = 0;
1386 /* Check if the user is trying to change anything besides speed/duplex */
1387 static bool virtnet_validate_ethtool_cmd(const struct ethtool_cmd *cmd)
1389 struct ethtool_cmd diff1 = *cmd;
1390 struct ethtool_cmd diff2 = {};
1392 /* cmd is always set so we need to clear it, validate the port type
1393 * and also without autonegotiation we can ignore advertising
1395 ethtool_cmd_speed_set(&diff1, 0);
1396 diff2.port = PORT_OTHER;
1397 diff1.advertising = 0;
1401 return !memcmp(&diff1, &diff2, sizeof(diff1));
1404 static int virtnet_set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1406 struct virtnet_info *vi = netdev_priv(dev);
1409 speed = ethtool_cmd_speed(cmd);
1410 /* don't allow custom speed and duplex */
1411 if (!ethtool_validate_speed(speed) ||
1412 !ethtool_validate_duplex(cmd->duplex) ||
1413 !virtnet_validate_ethtool_cmd(cmd))
1416 vi->duplex = cmd->duplex;
1421 static int virtnet_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1423 struct virtnet_info *vi = netdev_priv(dev);
1425 ethtool_cmd_speed_set(cmd, vi->speed);
1426 cmd->duplex = vi->duplex;
1427 cmd->port = PORT_OTHER;
1432 static void virtnet_init_settings(struct net_device *dev)
1434 struct virtnet_info *vi = netdev_priv(dev);
1436 vi->speed = SPEED_UNKNOWN;
1437 vi->duplex = DUPLEX_UNKNOWN;
1440 static const struct ethtool_ops virtnet_ethtool_ops = {
1441 .get_drvinfo = virtnet_get_drvinfo,
1442 .get_link = ethtool_op_get_link,
1443 .get_ringparam = virtnet_get_ringparam,
1444 .set_channels = virtnet_set_channels,
1445 .get_channels = virtnet_get_channels,
1446 .get_ts_info = ethtool_op_get_ts_info,
1447 .get_settings = virtnet_get_settings,
1448 .set_settings = virtnet_set_settings,
1452 #define MAX_MTU 65535
1454 static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
1456 if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
1462 static const struct net_device_ops virtnet_netdev = {
1463 .ndo_open = virtnet_open,
1464 .ndo_stop = virtnet_close,
1465 .ndo_start_xmit = start_xmit,
1466 .ndo_validate_addr = eth_validate_addr,
1467 .ndo_set_mac_address = virtnet_set_mac_address,
1468 .ndo_set_rx_mode = virtnet_set_rx_mode,
1469 .ndo_change_mtu = virtnet_change_mtu,
1470 .ndo_get_stats64 = virtnet_stats,
1471 .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
1472 .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
1473 #ifdef CONFIG_NET_POLL_CONTROLLER
1474 .ndo_poll_controller = virtnet_netpoll,
1476 #ifdef CONFIG_NET_RX_BUSY_POLL
1477 .ndo_busy_poll = virtnet_busy_poll,
1479 .ndo_features_check = passthru_features_check,
1482 static void virtnet_config_changed_work(struct work_struct *work)
1484 struct virtnet_info *vi =
1485 container_of(work, struct virtnet_info, config_work);
1488 if (virtio_cread_feature(vi->vdev, VIRTIO_NET_F_STATUS,
1489 struct virtio_net_config, status, &v) < 0)
1492 if (v & VIRTIO_NET_S_ANNOUNCE) {
1493 netdev_notify_peers(vi->dev);
1494 virtnet_ack_link_announce(vi);
1497 /* Ignore unknown (future) status bits */
1498 v &= VIRTIO_NET_S_LINK_UP;
1500 if (vi->status == v)
1505 if (vi->status & VIRTIO_NET_S_LINK_UP) {
1506 netif_carrier_on(vi->dev);
1507 netif_tx_wake_all_queues(vi->dev);
1509 netif_carrier_off(vi->dev);
1510 netif_tx_stop_all_queues(vi->dev);
1514 static void virtnet_config_changed(struct virtio_device *vdev)
1516 struct virtnet_info *vi = vdev->priv;
1518 schedule_work(&vi->config_work);
1521 static void virtnet_free_queues(struct virtnet_info *vi)
1525 for (i = 0; i < vi->max_queue_pairs; i++) {
1526 napi_hash_del(&vi->rq[i].napi);
1527 netif_napi_del(&vi->rq[i].napi);
1530 /* We called napi_hash_del() before netif_napi_del(),
1531 * we need to respect an RCU grace period before freeing vi->rq
1539 static void free_receive_bufs(struct virtnet_info *vi)
1543 for (i = 0; i < vi->max_queue_pairs; i++) {
1544 while (vi->rq[i].pages)
1545 __free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0);
1549 static void free_receive_page_frags(struct virtnet_info *vi)
1552 for (i = 0; i < vi->max_queue_pairs; i++)
1553 if (vi->rq[i].alloc_frag.page)
1554 put_page(vi->rq[i].alloc_frag.page);
1557 static void free_unused_bufs(struct virtnet_info *vi)
1562 for (i = 0; i < vi->max_queue_pairs; i++) {
1563 struct virtqueue *vq = vi->sq[i].vq;
1564 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL)
1568 for (i = 0; i < vi->max_queue_pairs; i++) {
1569 struct virtqueue *vq = vi->rq[i].vq;
1571 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) {
1572 if (vi->mergeable_rx_bufs) {
1573 unsigned long ctx = (unsigned long)buf;
1574 void *base = mergeable_ctx_to_buf_address(ctx);
1575 put_page(virt_to_head_page(base));
1576 } else if (vi->big_packets) {
1577 give_pages(&vi->rq[i], buf);
1585 static void virtnet_del_vqs(struct virtnet_info *vi)
1587 struct virtio_device *vdev = vi->vdev;
1589 virtnet_clean_affinity(vi, -1);
1591 vdev->config->del_vqs(vdev);
1593 virtnet_free_queues(vi);
1596 static int virtnet_find_vqs(struct virtnet_info *vi)
1598 vq_callback_t **callbacks;
1599 struct virtqueue **vqs;
1604 /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
1605 * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
1606 * possible control vq.
1608 total_vqs = vi->max_queue_pairs * 2 +
1609 virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ);
1611 /* Allocate space for find_vqs parameters */
1612 vqs = kzalloc(total_vqs * sizeof(*vqs), GFP_KERNEL);
1615 callbacks = kmalloc(total_vqs * sizeof(*callbacks), GFP_KERNEL);
1618 names = kmalloc(total_vqs * sizeof(*names), GFP_KERNEL);
1622 /* Parameters for control virtqueue, if any */
1624 callbacks[total_vqs - 1] = NULL;
1625 names[total_vqs - 1] = "control";
1628 /* Allocate/initialize parameters for send/receive virtqueues */
1629 for (i = 0; i < vi->max_queue_pairs; i++) {
1630 callbacks[rxq2vq(i)] = skb_recv_done;
1631 callbacks[txq2vq(i)] = skb_xmit_done;
1632 sprintf(vi->rq[i].name, "input.%d", i);
1633 sprintf(vi->sq[i].name, "output.%d", i);
1634 names[rxq2vq(i)] = vi->rq[i].name;
1635 names[txq2vq(i)] = vi->sq[i].name;
1638 ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs, callbacks,
1644 vi->cvq = vqs[total_vqs - 1];
1645 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
1646 vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
1649 for (i = 0; i < vi->max_queue_pairs; i++) {
1650 vi->rq[i].vq = vqs[rxq2vq(i)];
1651 vi->sq[i].vq = vqs[txq2vq(i)];
1670 static int virtnet_alloc_queues(struct virtnet_info *vi)
1674 vi->sq = kzalloc(sizeof(*vi->sq) * vi->max_queue_pairs, GFP_KERNEL);
1677 vi->rq = kzalloc(sizeof(*vi->rq) * vi->max_queue_pairs, GFP_KERNEL);
1681 INIT_DELAYED_WORK(&vi->refill, refill_work);
1682 for (i = 0; i < vi->max_queue_pairs; i++) {
1683 vi->rq[i].pages = NULL;
1684 netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll,
1687 sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg));
1688 ewma_pkt_len_init(&vi->rq[i].mrg_avg_pkt_len);
1689 sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg));
1700 static int init_vqs(struct virtnet_info *vi)
1704 /* Allocate send & receive queues */
1705 ret = virtnet_alloc_queues(vi);
1709 ret = virtnet_find_vqs(vi);
1714 virtnet_set_affinity(vi);
1720 virtnet_free_queues(vi);
1726 static ssize_t mergeable_rx_buffer_size_show(struct netdev_rx_queue *queue,
1727 struct rx_queue_attribute *attribute, char *buf)
1729 struct virtnet_info *vi = netdev_priv(queue->dev);
1730 unsigned int queue_index = get_netdev_rx_queue_index(queue);
1731 struct ewma_pkt_len *avg;
1733 BUG_ON(queue_index >= vi->max_queue_pairs);
1734 avg = &vi->rq[queue_index].mrg_avg_pkt_len;
1735 return sprintf(buf, "%u\n", get_mergeable_buf_len(avg));
1738 static struct rx_queue_attribute mergeable_rx_buffer_size_attribute =
1739 __ATTR_RO(mergeable_rx_buffer_size);
1741 static struct attribute *virtio_net_mrg_rx_attrs[] = {
1742 &mergeable_rx_buffer_size_attribute.attr,
1746 static const struct attribute_group virtio_net_mrg_rx_group = {
1747 .name = "virtio_net",
1748 .attrs = virtio_net_mrg_rx_attrs
1752 static bool virtnet_fail_on_feature(struct virtio_device *vdev,
1754 const char *fname, const char *dname)
1756 if (!virtio_has_feature(vdev, fbit))
1759 dev_err(&vdev->dev, "device advertises feature %s but not %s",
1765 #define VIRTNET_FAIL_ON(vdev, fbit, dbit) \
1766 virtnet_fail_on_feature(vdev, fbit, #fbit, dbit)
1768 static bool virtnet_validate_features(struct virtio_device *vdev)
1770 if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) &&
1771 (VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_RX,
1772 "VIRTIO_NET_F_CTRL_VQ") ||
1773 VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_VLAN,
1774 "VIRTIO_NET_F_CTRL_VQ") ||
1775 VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE,
1776 "VIRTIO_NET_F_CTRL_VQ") ||
1777 VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_MQ, "VIRTIO_NET_F_CTRL_VQ") ||
1778 VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR,
1779 "VIRTIO_NET_F_CTRL_VQ"))) {
1786 static int virtnet_probe(struct virtio_device *vdev)
1789 struct net_device *dev;
1790 struct virtnet_info *vi;
1791 u16 max_queue_pairs;
1794 if (!vdev->config->get) {
1795 dev_err(&vdev->dev, "%s failure: config access disabled\n",
1800 if (!virtnet_validate_features(vdev))
1803 /* Find if host supports multiqueue virtio_net device */
1804 err = virtio_cread_feature(vdev, VIRTIO_NET_F_MQ,
1805 struct virtio_net_config,
1806 max_virtqueue_pairs, &max_queue_pairs);
1808 /* We need at least 2 queue's */
1809 if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
1810 max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
1811 !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
1812 max_queue_pairs = 1;
1814 /* Allocate ourselves a network device with room for our info */
1815 dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs);
1819 /* Set up network device as normal. */
1820 dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE;
1821 dev->netdev_ops = &virtnet_netdev;
1822 dev->features = NETIF_F_HIGHDMA;
1824 dev->ethtool_ops = &virtnet_ethtool_ops;
1825 SET_NETDEV_DEV(dev, &vdev->dev);
1827 /* Do we support "hardware" checksums? */
1828 if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
1829 /* This opens up the world of extra features. */
1830 dev->hw_features |= NETIF_F_HW_CSUM | NETIF_F_SG;
1832 dev->features |= NETIF_F_HW_CSUM | NETIF_F_SG;
1834 if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
1835 dev->hw_features |= NETIF_F_TSO | NETIF_F_UFO
1836 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
1838 /* Individual feature bits: what can host handle? */
1839 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
1840 dev->hw_features |= NETIF_F_TSO;
1841 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
1842 dev->hw_features |= NETIF_F_TSO6;
1843 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
1844 dev->hw_features |= NETIF_F_TSO_ECN;
1845 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
1846 dev->hw_features |= NETIF_F_UFO;
1848 dev->features |= NETIF_F_GSO_ROBUST;
1851 dev->features |= dev->hw_features & (NETIF_F_ALL_TSO|NETIF_F_UFO);
1852 /* (!csum && gso) case will be fixed by register_netdev() */
1854 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM))
1855 dev->features |= NETIF_F_RXCSUM;
1857 dev->vlan_features = dev->features;
1859 /* Configuration may specify what MAC to use. Otherwise random. */
1860 if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
1861 virtio_cread_bytes(vdev,
1862 offsetof(struct virtio_net_config, mac),
1863 dev->dev_addr, dev->addr_len);
1865 eth_hw_addr_random(dev);
1867 /* Set up our device-specific information */
1868 vi = netdev_priv(dev);
1872 vi->stats = alloc_percpu(struct virtnet_stats);
1874 if (vi->stats == NULL)
1877 for_each_possible_cpu(i) {
1878 struct virtnet_stats *virtnet_stats;
1879 virtnet_stats = per_cpu_ptr(vi->stats, i);
1880 u64_stats_init(&virtnet_stats->tx_syncp);
1881 u64_stats_init(&virtnet_stats->rx_syncp);
1884 INIT_WORK(&vi->config_work, virtnet_config_changed_work);
1886 /* If we can receive ANY GSO packets, we must allocate large ones. */
1887 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
1888 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
1889 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN) ||
1890 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_UFO))
1891 vi->big_packets = true;
1893 if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
1894 vi->mergeable_rx_bufs = true;
1896 if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF) ||
1897 virtio_has_feature(vdev, VIRTIO_F_VERSION_1))
1898 vi->hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1900 vi->hdr_len = sizeof(struct virtio_net_hdr);
1902 if (virtio_has_feature(vdev, VIRTIO_F_ANY_LAYOUT) ||
1903 virtio_has_feature(vdev, VIRTIO_F_VERSION_1))
1904 vi->any_header_sg = true;
1906 if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
1909 if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) {
1910 mtu = virtio_cread16(vdev,
1911 offsetof(struct virtio_net_config,
1913 if (virtnet_change_mtu(dev, mtu))
1914 __virtio_clear_bit(vdev, VIRTIO_NET_F_MTU);
1917 if (vi->any_header_sg)
1918 dev->needed_headroom = vi->hdr_len;
1920 /* Use single tx/rx queue pair as default */
1921 vi->curr_queue_pairs = 1;
1922 vi->max_queue_pairs = max_queue_pairs;
1924 /* Allocate/initialize the rx/tx queues, and invoke find_vqs */
1930 if (vi->mergeable_rx_bufs)
1931 dev->sysfs_rx_queue_group = &virtio_net_mrg_rx_group;
1933 netif_set_real_num_tx_queues(dev, vi->curr_queue_pairs);
1934 netif_set_real_num_rx_queues(dev, vi->curr_queue_pairs);
1936 virtnet_init_settings(dev);
1938 err = register_netdev(dev);
1940 pr_debug("virtio_net: registering device failed\n");
1944 virtio_device_ready(vdev);
1946 err = virtnet_cpu_notif_add(vi);
1948 pr_debug("virtio_net: registering cpu notifier failed\n");
1949 goto free_unregister_netdev;
1952 /* Assume link up if device can't report link status,
1953 otherwise get link status from config. */
1954 netif_carrier_off(dev);
1955 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
1956 schedule_work(&vi->config_work);
1958 vi->status = VIRTIO_NET_S_LINK_UP;
1959 netif_carrier_on(dev);
1962 pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
1963 dev->name, max_queue_pairs);
1967 free_unregister_netdev:
1968 vi->vdev->config->reset(vdev);
1970 unregister_netdev(dev);
1972 cancel_delayed_work_sync(&vi->refill);
1973 free_receive_page_frags(vi);
1974 virtnet_del_vqs(vi);
1976 free_percpu(vi->stats);
1982 static void remove_vq_common(struct virtnet_info *vi)
1984 vi->vdev->config->reset(vi->vdev);
1986 /* Free unused buffers in both send and recv, if any. */
1987 free_unused_bufs(vi);
1989 free_receive_bufs(vi);
1991 free_receive_page_frags(vi);
1993 virtnet_del_vqs(vi);
1996 static void virtnet_remove(struct virtio_device *vdev)
1998 struct virtnet_info *vi = vdev->priv;
2000 virtnet_cpu_notif_remove(vi);
2002 /* Make sure no work handler is accessing the device. */
2003 flush_work(&vi->config_work);
2005 unregister_netdev(vi->dev);
2007 remove_vq_common(vi);
2009 free_percpu(vi->stats);
2010 free_netdev(vi->dev);
2013 #ifdef CONFIG_PM_SLEEP
2014 static int virtnet_freeze(struct virtio_device *vdev)
2016 struct virtnet_info *vi = vdev->priv;
2019 virtnet_cpu_notif_remove(vi);
2021 /* Make sure no work handler is accessing the device */
2022 flush_work(&vi->config_work);
2024 netif_device_detach(vi->dev);
2025 cancel_delayed_work_sync(&vi->refill);
2027 if (netif_running(vi->dev)) {
2028 for (i = 0; i < vi->max_queue_pairs; i++)
2029 napi_disable(&vi->rq[i].napi);
2032 remove_vq_common(vi);
2037 static int virtnet_restore(struct virtio_device *vdev)
2039 struct virtnet_info *vi = vdev->priv;
2046 virtio_device_ready(vdev);
2048 if (netif_running(vi->dev)) {
2049 for (i = 0; i < vi->curr_queue_pairs; i++)
2050 if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL))
2051 schedule_delayed_work(&vi->refill, 0);
2053 for (i = 0; i < vi->max_queue_pairs; i++)
2054 virtnet_napi_enable(&vi->rq[i]);
2057 netif_device_attach(vi->dev);
2060 virtnet_set_queues(vi, vi->curr_queue_pairs);
2063 err = virtnet_cpu_notif_add(vi);
2071 static struct virtio_device_id id_table[] = {
2072 { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
2076 #define VIRTNET_FEATURES \
2077 VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM, \
2079 VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6, \
2080 VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, \
2081 VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO, \
2082 VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ, \
2083 VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, \
2084 VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ, \
2085 VIRTIO_NET_F_CTRL_MAC_ADDR, \
2088 static unsigned int features[] = {
2092 static unsigned int features_legacy[] = {
2095 VIRTIO_F_ANY_LAYOUT,
2098 static struct virtio_driver virtio_net_driver = {
2099 .feature_table = features,
2100 .feature_table_size = ARRAY_SIZE(features),
2101 .feature_table_legacy = features_legacy,
2102 .feature_table_size_legacy = ARRAY_SIZE(features_legacy),
2103 .driver.name = KBUILD_MODNAME,
2104 .driver.owner = THIS_MODULE,
2105 .id_table = id_table,
2106 .probe = virtnet_probe,
2107 .remove = virtnet_remove,
2108 .config_changed = virtnet_config_changed,
2109 #ifdef CONFIG_PM_SLEEP
2110 .freeze = virtnet_freeze,
2111 .restore = virtnet_restore,
2115 static __init int virtio_net_driver_init(void)
2119 ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "AP_VIRT_NET_ONLINE",
2121 virtnet_cpu_down_prep);
2124 virtionet_online = ret;
2125 ret = cpuhp_setup_state_multi(CPUHP_VIRT_NET_DEAD, "VIRT_NET_DEAD",
2126 NULL, virtnet_cpu_dead);
2130 ret = register_virtio_driver(&virtio_net_driver);
2135 cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD);
2137 cpuhp_remove_multi_state(virtionet_online);
2141 module_init(virtio_net_driver_init);
2143 static __exit void virtio_net_driver_exit(void)
2145 cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD);
2146 cpuhp_remove_multi_state(virtionet_online);
2147 unregister_virtio_driver(&virtio_net_driver);
2149 module_exit(virtio_net_driver_exit);
2151 MODULE_DEVICE_TABLE(virtio, id_table);
2152 MODULE_DESCRIPTION("Virtio network driver");
2153 MODULE_LICENSE("GPL");