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, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 #include <linux/netdevice.h>
21 #include <linux/etherdevice.h>
22 #include <linux/ethtool.h>
23 #include <linux/module.h>
24 #include <linux/virtio.h>
25 #include <linux/virtio_net.h>
26 #include <linux/scatterlist.h>
27 #include <linux/if_vlan.h>
28 #include <linux/slab.h>
29 #include <linux/cpu.h>
31 static int napi_weight = NAPI_POLL_WEIGHT;
32 module_param(napi_weight, int, 0444);
34 static bool csum = true, gso = true;
35 module_param(csum, bool, 0444);
36 module_param(gso, bool, 0444);
38 /* FIXME: MTU in config. */
39 #define GOOD_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
40 #define MERGE_BUFFER_LEN (ALIGN(GOOD_PACKET_LEN + \
41 sizeof(struct virtio_net_hdr_mrg_rxbuf), \
43 #define GOOD_COPY_LEN 128
45 #define VIRTNET_DRIVER_VERSION "1.0.0"
47 struct virtnet_stats {
48 struct u64_stats_sync tx_syncp;
49 struct u64_stats_sync rx_syncp;
57 /* Internal representation of a send virtqueue */
59 /* Virtqueue associated with this send _queue */
62 /* TX: fragments + linear part + virtio header */
63 struct scatterlist sg[MAX_SKB_FRAGS + 2];
65 /* Name of the send queue: output.$index */
69 /* Internal representation of a receive virtqueue */
70 struct receive_queue {
71 /* Virtqueue associated with this receive_queue */
74 struct napi_struct napi;
76 /* Number of input buffers, and max we've ever had. */
77 unsigned int num, max;
79 /* Chain pages by the private ptr. */
82 /* RX: fragments + linear part + virtio header */
83 struct scatterlist sg[MAX_SKB_FRAGS + 2];
85 /* Name of this receive queue: input.$index */
90 struct virtio_device *vdev;
91 struct virtqueue *cvq;
92 struct net_device *dev;
93 struct send_queue *sq;
94 struct receive_queue *rq;
97 /* Max # of queue pairs supported by the device */
100 /* # of queue pairs currently used by the driver */
101 u16 curr_queue_pairs;
103 /* I like... big packets and I cannot lie! */
106 /* Host will merge rx buffers for big packets (shake it! shake it!) */
107 bool mergeable_rx_bufs;
109 /* Has control virtqueue */
112 /* Host can handle any s/g split between our header and packet data */
115 /* enable config space updates */
118 /* Active statistics */
119 struct virtnet_stats __percpu *stats;
121 /* Work struct for refilling if we run low on memory. */
122 struct delayed_work refill;
124 /* Work struct for config space updates */
125 struct work_struct config_work;
127 /* Lock for config space updates */
128 struct mutex config_lock;
130 /* Page_frag for GFP_KERNEL packet buffer allocation when we run
133 struct page_frag alloc_frag;
135 /* Does the affinity hint is set for virtqueues? */
136 bool affinity_hint_set;
138 /* CPU hot plug notifier */
139 struct notifier_block nb;
142 struct skb_vnet_hdr {
144 struct virtio_net_hdr hdr;
145 struct virtio_net_hdr_mrg_rxbuf mhdr;
149 struct padded_vnet_hdr {
150 struct virtio_net_hdr hdr;
152 * virtio_net_hdr should be in a separated sg buffer because of a
153 * QEMU bug, and data sg buffer shares same page with this header sg.
154 * This padding makes next sg 16 byte aligned after virtio_net_hdr.
159 /* Converting between virtqueue no. and kernel tx/rx queue no.
160 * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq
162 static int vq2txq(struct virtqueue *vq)
164 return (vq->index - 1) / 2;
167 static int txq2vq(int txq)
172 static int vq2rxq(struct virtqueue *vq)
174 return vq->index / 2;
177 static int rxq2vq(int rxq)
182 static inline struct skb_vnet_hdr *skb_vnet_hdr(struct sk_buff *skb)
184 return (struct skb_vnet_hdr *)skb->cb;
188 * private is used to chain pages for big packets, put the whole
189 * most recent used list in the beginning for reuse
191 static void give_pages(struct receive_queue *rq, struct page *page)
195 /* Find end of list, sew whole thing into vi->rq.pages. */
196 for (end = page; end->private; end = (struct page *)end->private);
197 end->private = (unsigned long)rq->pages;
201 static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask)
203 struct page *p = rq->pages;
206 rq->pages = (struct page *)p->private;
207 /* clear private here, it is used to chain pages */
210 p = alloc_page(gfp_mask);
214 static void skb_xmit_done(struct virtqueue *vq)
216 struct virtnet_info *vi = vq->vdev->priv;
218 /* Suppress further interrupts. */
219 virtqueue_disable_cb(vq);
221 /* We were probably waiting for more output buffers. */
222 netif_wake_subqueue(vi->dev, vq2txq(vq));
225 /* Called from bottom half context */
226 static struct sk_buff *page_to_skb(struct receive_queue *rq,
227 struct page *page, unsigned int offset,
228 unsigned int len, unsigned int truesize)
230 struct virtnet_info *vi = rq->vq->vdev->priv;
232 struct skb_vnet_hdr *hdr;
233 unsigned int copy, hdr_len, hdr_padded_len;
236 p = page_address(page) + offset;
238 /* copy small packet so we can reuse these pages for small data */
239 skb = netdev_alloc_skb_ip_align(vi->dev, GOOD_COPY_LEN);
243 hdr = skb_vnet_hdr(skb);
245 if (vi->mergeable_rx_bufs) {
246 hdr_len = sizeof hdr->mhdr;
247 hdr_padded_len = sizeof hdr->mhdr;
249 hdr_len = sizeof hdr->hdr;
250 hdr_padded_len = sizeof(struct padded_vnet_hdr);
253 memcpy(hdr, p, hdr_len);
256 offset += hdr_padded_len;
260 if (copy > skb_tailroom(skb))
261 copy = skb_tailroom(skb);
262 memcpy(skb_put(skb, copy), p, copy);
267 if (vi->mergeable_rx_bufs) {
269 skb_add_rx_frag(skb, 0, page, offset, len, truesize);
276 * Verify that we can indeed put this data into a skb.
277 * This is here to handle cases when the device erroneously
278 * tries to receive more than is possible. This is usually
279 * the case of a broken device.
281 if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) {
282 net_dbg_ratelimited("%s: too much data\n", skb->dev->name);
286 BUG_ON(offset >= PAGE_SIZE);
288 unsigned int frag_size = min((unsigned)PAGE_SIZE - offset, len);
289 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page, offset,
290 frag_size, truesize);
292 page = (struct page *)page->private;
297 give_pages(rq, page);
302 static int receive_mergeable(struct receive_queue *rq, struct sk_buff *head_skb)
304 struct skb_vnet_hdr *hdr = skb_vnet_hdr(head_skb);
305 struct sk_buff *curr_skb = head_skb;
308 int num_buf, len, offset;
310 num_buf = hdr->mhdr.num_buffers;
312 int num_skb_frags = skb_shinfo(curr_skb)->nr_frags;
313 buf = virtqueue_get_buf(rq->vq, &len);
314 if (unlikely(!buf)) {
315 pr_debug("%s: rx error: %d buffers missing\n",
316 head_skb->dev->name, hdr->mhdr.num_buffers);
317 head_skb->dev->stats.rx_length_errors++;
320 if (unlikely(len > MERGE_BUFFER_LEN)) {
321 pr_debug("%s: rx error: merge buffer too long\n",
322 head_skb->dev->name);
323 len = MERGE_BUFFER_LEN;
325 if (unlikely(num_skb_frags == MAX_SKB_FRAGS)) {
326 struct sk_buff *nskb = alloc_skb(0, GFP_ATOMIC);
327 if (unlikely(!nskb)) {
328 head_skb->dev->stats.rx_dropped++;
331 if (curr_skb == head_skb)
332 skb_shinfo(curr_skb)->frag_list = nskb;
334 curr_skb->next = nskb;
336 head_skb->truesize += nskb->truesize;
339 if (curr_skb != head_skb) {
340 head_skb->data_len += len;
341 head_skb->len += len;
342 head_skb->truesize += MERGE_BUFFER_LEN;
344 page = virt_to_head_page(buf);
345 offset = buf - (char *)page_address(page);
346 if (skb_can_coalesce(curr_skb, num_skb_frags, page, offset)) {
348 skb_coalesce_rx_frag(curr_skb, num_skb_frags - 1,
349 len, MERGE_BUFFER_LEN);
351 skb_add_rx_frag(curr_skb, num_skb_frags, page,
352 offset, len, MERGE_BUFFER_LEN);
359 static void receive_buf(struct receive_queue *rq, void *buf, unsigned int len)
361 struct virtnet_info *vi = rq->vq->vdev->priv;
362 struct net_device *dev = vi->dev;
363 struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
366 struct skb_vnet_hdr *hdr;
368 if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
369 pr_debug("%s: short packet %i\n", dev->name, len);
370 dev->stats.rx_length_errors++;
373 else if (vi->mergeable_rx_bufs)
374 put_page(virt_to_head_page(buf));
380 if (!vi->mergeable_rx_bufs && !vi->big_packets) {
382 len -= sizeof(struct virtio_net_hdr);
384 } else if (vi->mergeable_rx_bufs) {
385 struct page *page = virt_to_head_page(buf);
386 skb = page_to_skb(rq, page,
387 (char *)buf - (char *)page_address(page),
388 len, MERGE_BUFFER_LEN);
389 if (unlikely(!skb)) {
390 dev->stats.rx_dropped++;
394 if (receive_mergeable(rq, skb)) {
400 skb = page_to_skb(rq, page, 0, len, PAGE_SIZE);
401 if (unlikely(!skb)) {
402 dev->stats.rx_dropped++;
403 give_pages(rq, page);
408 hdr = skb_vnet_hdr(skb);
410 u64_stats_update_begin(&stats->rx_syncp);
411 stats->rx_bytes += skb->len;
413 u64_stats_update_end(&stats->rx_syncp);
415 if (hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
416 pr_debug("Needs csum!\n");
417 if (!skb_partial_csum_set(skb,
419 hdr->hdr.csum_offset))
421 } else if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) {
422 skb->ip_summed = CHECKSUM_UNNECESSARY;
425 skb->protocol = eth_type_trans(skb, dev);
426 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
427 ntohs(skb->protocol), skb->len, skb->pkt_type);
429 if (hdr->hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
431 switch (hdr->hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
432 case VIRTIO_NET_HDR_GSO_TCPV4:
433 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
435 case VIRTIO_NET_HDR_GSO_UDP:
436 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
438 case VIRTIO_NET_HDR_GSO_TCPV6:
439 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
442 net_warn_ratelimited("%s: bad gso type %u.\n",
443 dev->name, hdr->hdr.gso_type);
447 if (hdr->hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
448 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
450 skb_shinfo(skb)->gso_size = hdr->hdr.gso_size;
451 if (skb_shinfo(skb)->gso_size == 0) {
452 net_warn_ratelimited("%s: zero gso size.\n", dev->name);
456 /* Header must be checked, and gso_segs computed. */
457 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
458 skb_shinfo(skb)->gso_segs = 0;
461 netif_receive_skb(skb);
465 dev->stats.rx_frame_errors++;
469 static int add_recvbuf_small(struct receive_queue *rq, gfp_t gfp)
471 struct virtnet_info *vi = rq->vq->vdev->priv;
473 struct skb_vnet_hdr *hdr;
476 skb = __netdev_alloc_skb_ip_align(vi->dev, GOOD_PACKET_LEN, gfp);
480 skb_put(skb, GOOD_PACKET_LEN);
482 hdr = skb_vnet_hdr(skb);
483 sg_set_buf(rq->sg, &hdr->hdr, sizeof hdr->hdr);
485 skb_to_sgvec(skb, rq->sg + 1, 0, skb->len);
487 err = virtqueue_add_inbuf(rq->vq, rq->sg, 2, skb, gfp);
494 static int add_recvbuf_big(struct receive_queue *rq, gfp_t gfp)
496 struct page *first, *list = NULL;
500 /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */
501 for (i = MAX_SKB_FRAGS + 1; i > 1; --i) {
502 first = get_a_page(rq, gfp);
505 give_pages(rq, list);
508 sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE);
510 /* chain new page in list head to match sg */
511 first->private = (unsigned long)list;
515 first = get_a_page(rq, gfp);
517 give_pages(rq, list);
520 p = page_address(first);
522 /* rq->sg[0], rq->sg[1] share the same page */
523 /* a separated rq->sg[0] for virtio_net_hdr only due to QEMU bug */
524 sg_set_buf(&rq->sg[0], p, sizeof(struct virtio_net_hdr));
526 /* rq->sg[1] for data packet, from offset */
527 offset = sizeof(struct padded_vnet_hdr);
528 sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset);
530 /* chain first in list head */
531 first->private = (unsigned long)list;
532 err = virtqueue_add_inbuf(rq->vq, rq->sg, MAX_SKB_FRAGS + 2,
535 give_pages(rq, first);
540 static int add_recvbuf_mergeable(struct receive_queue *rq, gfp_t gfp)
542 struct virtnet_info *vi = rq->vq->vdev->priv;
546 if (gfp & __GFP_WAIT) {
547 if (skb_page_frag_refill(MERGE_BUFFER_LEN, &vi->alloc_frag,
549 buf = (char *)page_address(vi->alloc_frag.page) +
550 vi->alloc_frag.offset;
551 get_page(vi->alloc_frag.page);
552 vi->alloc_frag.offset += MERGE_BUFFER_LEN;
555 buf = netdev_alloc_frag(MERGE_BUFFER_LEN);
560 sg_init_one(rq->sg, buf, MERGE_BUFFER_LEN);
561 err = virtqueue_add_inbuf(rq->vq, rq->sg, 1, buf, gfp);
563 put_page(virt_to_head_page(buf));
569 * Returns false if we couldn't fill entirely (OOM).
571 * Normally run in the receive path, but can also be run from ndo_open
572 * before we're receiving packets, or from refill_work which is
573 * careful to disable receiving (using napi_disable).
575 static bool try_fill_recv(struct receive_queue *rq, gfp_t gfp)
577 struct virtnet_info *vi = rq->vq->vdev->priv;
582 if (vi->mergeable_rx_bufs)
583 err = add_recvbuf_mergeable(rq, gfp);
584 else if (vi->big_packets)
585 err = add_recvbuf_big(rq, gfp);
587 err = add_recvbuf_small(rq, gfp);
589 oom = err == -ENOMEM;
593 } while (rq->vq->num_free);
594 if (unlikely(rq->num > rq->max))
596 virtqueue_kick(rq->vq);
600 static void skb_recv_done(struct virtqueue *rvq)
602 struct virtnet_info *vi = rvq->vdev->priv;
603 struct receive_queue *rq = &vi->rq[vq2rxq(rvq)];
605 /* Schedule NAPI, Suppress further interrupts if successful. */
606 if (napi_schedule_prep(&rq->napi)) {
607 virtqueue_disable_cb(rvq);
608 __napi_schedule(&rq->napi);
612 static void virtnet_napi_enable(struct receive_queue *rq)
614 napi_enable(&rq->napi);
616 /* If all buffers were filled by other side before we napi_enabled, we
617 * won't get another interrupt, so process any outstanding packets
618 * now. virtnet_poll wants re-enable the queue, so we disable here.
619 * We synchronize against interrupts via NAPI_STATE_SCHED */
620 if (napi_schedule_prep(&rq->napi)) {
621 virtqueue_disable_cb(rq->vq);
623 __napi_schedule(&rq->napi);
628 static void refill_work(struct work_struct *work)
630 struct virtnet_info *vi =
631 container_of(work, struct virtnet_info, refill.work);
635 for (i = 0; i < vi->curr_queue_pairs; i++) {
636 struct receive_queue *rq = &vi->rq[i];
638 napi_disable(&rq->napi);
639 still_empty = !try_fill_recv(rq, GFP_KERNEL);
640 virtnet_napi_enable(rq);
642 /* In theory, this can happen: if we don't get any buffers in
643 * we will *never* try to fill again.
646 schedule_delayed_work(&vi->refill, HZ/2);
650 static int virtnet_poll(struct napi_struct *napi, int budget)
652 struct receive_queue *rq =
653 container_of(napi, struct receive_queue, napi);
654 struct virtnet_info *vi = rq->vq->vdev->priv;
656 unsigned int r, len, received = 0;
659 while (received < budget &&
660 (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) {
661 receive_buf(rq, buf, len);
666 if (rq->num < rq->max / 2) {
667 if (!try_fill_recv(rq, GFP_ATOMIC))
668 schedule_delayed_work(&vi->refill, 0);
671 /* Out of packets? */
672 if (received < budget) {
673 r = virtqueue_enable_cb_prepare(rq->vq);
675 if (unlikely(virtqueue_poll(rq->vq, r)) &&
676 napi_schedule_prep(napi)) {
677 virtqueue_disable_cb(rq->vq);
678 __napi_schedule(napi);
686 static int virtnet_open(struct net_device *dev)
688 struct virtnet_info *vi = netdev_priv(dev);
691 for (i = 0; i < vi->max_queue_pairs; i++) {
692 if (i < vi->curr_queue_pairs)
693 /* Make sure we have some buffers: if oom use wq. */
694 if (!try_fill_recv(&vi->rq[i], GFP_KERNEL))
695 schedule_delayed_work(&vi->refill, 0);
696 virtnet_napi_enable(&vi->rq[i]);
702 static void free_old_xmit_skbs(struct send_queue *sq)
706 struct virtnet_info *vi = sq->vq->vdev->priv;
707 struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
709 while ((skb = virtqueue_get_buf(sq->vq, &len)) != NULL) {
710 pr_debug("Sent skb %p\n", skb);
712 u64_stats_update_begin(&stats->tx_syncp);
713 stats->tx_bytes += skb->len;
715 u64_stats_update_end(&stats->tx_syncp);
717 dev_kfree_skb_any(skb);
721 static int xmit_skb(struct send_queue *sq, struct sk_buff *skb)
723 struct skb_vnet_hdr *hdr;
724 const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
725 struct virtnet_info *vi = sq->vq->vdev->priv;
730 pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
731 if (vi->mergeable_rx_bufs)
732 hdr_len = sizeof hdr->mhdr;
734 hdr_len = sizeof hdr->hdr;
736 can_push = vi->any_header_sg &&
737 !((unsigned long)skb->data & (__alignof__(*hdr) - 1)) &&
738 !skb_header_cloned(skb) && skb_headroom(skb) >= hdr_len;
739 /* Even if we can, don't push here yet as this would skew
740 * csum_start offset below. */
742 hdr = (struct skb_vnet_hdr *)(skb->data - hdr_len);
744 hdr = skb_vnet_hdr(skb);
746 if (skb->ip_summed == CHECKSUM_PARTIAL) {
747 hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
748 hdr->hdr.csum_start = skb_checksum_start_offset(skb);
749 hdr->hdr.csum_offset = skb->csum_offset;
752 hdr->hdr.csum_offset = hdr->hdr.csum_start = 0;
755 if (skb_is_gso(skb)) {
756 hdr->hdr.hdr_len = skb_headlen(skb);
757 hdr->hdr.gso_size = skb_shinfo(skb)->gso_size;
758 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
759 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
760 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
761 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
762 else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
763 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
766 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
767 hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
769 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
770 hdr->hdr.gso_size = hdr->hdr.hdr_len = 0;
773 if (vi->mergeable_rx_bufs)
774 hdr->mhdr.num_buffers = 0;
777 __skb_push(skb, hdr_len);
778 num_sg = skb_to_sgvec(skb, sq->sg, 0, skb->len);
779 /* Pull header back to avoid skew in tx bytes calculations. */
780 __skb_pull(skb, hdr_len);
782 sg_set_buf(sq->sg, hdr, hdr_len);
783 num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len) + 1;
785 return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC);
788 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
790 struct virtnet_info *vi = netdev_priv(dev);
791 int qnum = skb_get_queue_mapping(skb);
792 struct send_queue *sq = &vi->sq[qnum];
795 /* Free up any pending old buffers before queueing new ones. */
796 free_old_xmit_skbs(sq);
798 /* Try to transmit */
799 err = xmit_skb(sq, skb);
801 /* This should not happen! */
803 dev->stats.tx_fifo_errors++;
806 "Unexpected TXQ (%d) queue failure: %d\n", qnum, err);
807 dev->stats.tx_dropped++;
811 virtqueue_kick(sq->vq);
813 /* Don't wait up for transmitted skbs to be freed. */
817 /* Apparently nice girls don't return TX_BUSY; stop the queue
818 * before it gets out of hand. Naturally, this wastes entries. */
819 if (sq->vq->num_free < 2+MAX_SKB_FRAGS) {
820 netif_stop_subqueue(dev, qnum);
821 if (unlikely(!virtqueue_enable_cb_delayed(sq->vq))) {
822 /* More just got used, free them then recheck. */
823 free_old_xmit_skbs(sq);
824 if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) {
825 netif_start_subqueue(dev, qnum);
826 virtqueue_disable_cb(sq->vq);
835 * Send command via the control virtqueue and check status. Commands
836 * supported by the hypervisor, as indicated by feature bits, should
837 * never fail unless improperly formated.
839 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
840 struct scatterlist *out,
841 struct scatterlist *in)
843 struct scatterlist *sgs[4], hdr, stat;
844 struct virtio_net_ctrl_hdr ctrl;
845 virtio_net_ctrl_ack status = ~0;
846 unsigned out_num = 0, in_num = 0, tmp;
848 /* Caller should know better */
849 BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ));
854 sg_init_one(&hdr, &ctrl, sizeof(ctrl));
855 sgs[out_num++] = &hdr;
858 sgs[out_num++] = out;
860 sgs[out_num + in_num++] = in;
862 /* Add return status. */
863 sg_init_one(&stat, &status, sizeof(status));
864 sgs[out_num + in_num++] = &stat;
866 BUG_ON(out_num + in_num > ARRAY_SIZE(sgs));
867 BUG_ON(virtqueue_add_sgs(vi->cvq, sgs, out_num, in_num, vi, GFP_ATOMIC)
870 virtqueue_kick(vi->cvq);
872 /* Spin for a response, the kick causes an ioport write, trapping
873 * into the hypervisor, so the request should be handled immediately.
875 while (!virtqueue_get_buf(vi->cvq, &tmp))
878 return status == VIRTIO_NET_OK;
881 static int virtnet_set_mac_address(struct net_device *dev, void *p)
883 struct virtnet_info *vi = netdev_priv(dev);
884 struct virtio_device *vdev = vi->vdev;
886 struct sockaddr *addr = p;
887 struct scatterlist sg;
889 ret = eth_prepare_mac_addr_change(dev, p);
893 if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
894 sg_init_one(&sg, addr->sa_data, dev->addr_len);
895 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
896 VIRTIO_NET_CTRL_MAC_ADDR_SET,
899 "Failed to set mac address by vq command.\n");
902 } else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
903 vdev->config->set(vdev, offsetof(struct virtio_net_config, mac),
904 addr->sa_data, dev->addr_len);
907 eth_commit_mac_addr_change(dev, p);
912 static struct rtnl_link_stats64 *virtnet_stats(struct net_device *dev,
913 struct rtnl_link_stats64 *tot)
915 struct virtnet_info *vi = netdev_priv(dev);
919 for_each_possible_cpu(cpu) {
920 struct virtnet_stats *stats = per_cpu_ptr(vi->stats, cpu);
921 u64 tpackets, tbytes, rpackets, rbytes;
924 start = u64_stats_fetch_begin_bh(&stats->tx_syncp);
925 tpackets = stats->tx_packets;
926 tbytes = stats->tx_bytes;
927 } while (u64_stats_fetch_retry_bh(&stats->tx_syncp, start));
930 start = u64_stats_fetch_begin_bh(&stats->rx_syncp);
931 rpackets = stats->rx_packets;
932 rbytes = stats->rx_bytes;
933 } while (u64_stats_fetch_retry_bh(&stats->rx_syncp, start));
935 tot->rx_packets += rpackets;
936 tot->tx_packets += tpackets;
937 tot->rx_bytes += rbytes;
938 tot->tx_bytes += tbytes;
941 tot->tx_dropped = dev->stats.tx_dropped;
942 tot->tx_fifo_errors = dev->stats.tx_fifo_errors;
943 tot->rx_dropped = dev->stats.rx_dropped;
944 tot->rx_length_errors = dev->stats.rx_length_errors;
945 tot->rx_frame_errors = dev->stats.rx_frame_errors;
950 #ifdef CONFIG_NET_POLL_CONTROLLER
951 static void virtnet_netpoll(struct net_device *dev)
953 struct virtnet_info *vi = netdev_priv(dev);
956 for (i = 0; i < vi->curr_queue_pairs; i++)
957 napi_schedule(&vi->rq[i].napi);
961 static void virtnet_ack_link_announce(struct virtnet_info *vi)
964 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE,
965 VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL, NULL))
966 dev_warn(&vi->dev->dev, "Failed to ack link announce.\n");
970 static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs)
972 struct scatterlist sg;
973 struct virtio_net_ctrl_mq s;
974 struct net_device *dev = vi->dev;
976 if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ))
979 s.virtqueue_pairs = queue_pairs;
980 sg_init_one(&sg, &s, sizeof(s));
982 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ,
983 VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg, NULL)) {
984 dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n",
988 vi->curr_queue_pairs = queue_pairs;
989 /* virtnet_open() will refill when device is going to up. */
990 if (dev->flags & IFF_UP)
991 schedule_delayed_work(&vi->refill, 0);
997 static int virtnet_close(struct net_device *dev)
999 struct virtnet_info *vi = netdev_priv(dev);
1002 /* Make sure refill_work doesn't re-enable napi! */
1003 cancel_delayed_work_sync(&vi->refill);
1005 for (i = 0; i < vi->max_queue_pairs; i++)
1006 napi_disable(&vi->rq[i].napi);
1011 static void virtnet_set_rx_mode(struct net_device *dev)
1013 struct virtnet_info *vi = netdev_priv(dev);
1014 struct scatterlist sg[2];
1015 u8 promisc, allmulti;
1016 struct virtio_net_ctrl_mac *mac_data;
1017 struct netdev_hw_addr *ha;
1023 /* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */
1024 if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
1027 promisc = ((dev->flags & IFF_PROMISC) != 0);
1028 allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
1030 sg_init_one(sg, &promisc, sizeof(promisc));
1032 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
1033 VIRTIO_NET_CTRL_RX_PROMISC,
1035 dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
1036 promisc ? "en" : "dis");
1038 sg_init_one(sg, &allmulti, sizeof(allmulti));
1040 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
1041 VIRTIO_NET_CTRL_RX_ALLMULTI,
1043 dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
1044 allmulti ? "en" : "dis");
1046 uc_count = netdev_uc_count(dev);
1047 mc_count = netdev_mc_count(dev);
1048 /* MAC filter - use one buffer for both lists */
1049 buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
1050 (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
1055 sg_init_table(sg, 2);
1057 /* Store the unicast list and count in the front of the buffer */
1058 mac_data->entries = uc_count;
1060 netdev_for_each_uc_addr(ha, dev)
1061 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
1063 sg_set_buf(&sg[0], mac_data,
1064 sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
1066 /* multicast list and count fill the end */
1067 mac_data = (void *)&mac_data->macs[uc_count][0];
1069 mac_data->entries = mc_count;
1071 netdev_for_each_mc_addr(ha, dev)
1072 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
1074 sg_set_buf(&sg[1], mac_data,
1075 sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
1077 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
1078 VIRTIO_NET_CTRL_MAC_TABLE_SET,
1080 dev_warn(&dev->dev, "Failed to set MAC fitler table.\n");
1085 static int virtnet_vlan_rx_add_vid(struct net_device *dev,
1086 __be16 proto, u16 vid)
1088 struct virtnet_info *vi = netdev_priv(dev);
1089 struct scatterlist sg;
1091 sg_init_one(&sg, &vid, sizeof(vid));
1093 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1094 VIRTIO_NET_CTRL_VLAN_ADD, &sg, NULL))
1095 dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
1099 static int virtnet_vlan_rx_kill_vid(struct net_device *dev,
1100 __be16 proto, u16 vid)
1102 struct virtnet_info *vi = netdev_priv(dev);
1103 struct scatterlist sg;
1105 sg_init_one(&sg, &vid, sizeof(vid));
1107 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1108 VIRTIO_NET_CTRL_VLAN_DEL, &sg, NULL))
1109 dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
1113 static void virtnet_clean_affinity(struct virtnet_info *vi, long hcpu)
1117 if (vi->affinity_hint_set) {
1118 for (i = 0; i < vi->max_queue_pairs; i++) {
1119 virtqueue_set_affinity(vi->rq[i].vq, -1);
1120 virtqueue_set_affinity(vi->sq[i].vq, -1);
1123 vi->affinity_hint_set = false;
1127 static void virtnet_set_affinity(struct virtnet_info *vi)
1132 /* In multiqueue mode, when the number of cpu is equal to the number of
1133 * queue pairs, we let the queue pairs to be private to one cpu by
1134 * setting the affinity hint to eliminate the contention.
1136 if (vi->curr_queue_pairs == 1 ||
1137 vi->max_queue_pairs != num_online_cpus()) {
1138 virtnet_clean_affinity(vi, -1);
1143 for_each_online_cpu(cpu) {
1144 virtqueue_set_affinity(vi->rq[i].vq, cpu);
1145 virtqueue_set_affinity(vi->sq[i].vq, cpu);
1146 netif_set_xps_queue(vi->dev, cpumask_of(cpu), i);
1150 vi->affinity_hint_set = true;
1153 static int virtnet_cpu_callback(struct notifier_block *nfb,
1154 unsigned long action, void *hcpu)
1156 struct virtnet_info *vi = container_of(nfb, struct virtnet_info, nb);
1158 switch(action & ~CPU_TASKS_FROZEN) {
1160 case CPU_DOWN_FAILED:
1162 virtnet_set_affinity(vi);
1164 case CPU_DOWN_PREPARE:
1165 virtnet_clean_affinity(vi, (long)hcpu);
1174 static void virtnet_get_ringparam(struct net_device *dev,
1175 struct ethtool_ringparam *ring)
1177 struct virtnet_info *vi = netdev_priv(dev);
1179 ring->rx_max_pending = virtqueue_get_vring_size(vi->rq[0].vq);
1180 ring->tx_max_pending = virtqueue_get_vring_size(vi->sq[0].vq);
1181 ring->rx_pending = ring->rx_max_pending;
1182 ring->tx_pending = ring->tx_max_pending;
1186 static void virtnet_get_drvinfo(struct net_device *dev,
1187 struct ethtool_drvinfo *info)
1189 struct virtnet_info *vi = netdev_priv(dev);
1190 struct virtio_device *vdev = vi->vdev;
1192 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
1193 strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version));
1194 strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info));
1198 /* TODO: Eliminate OOO packets during switching */
1199 static int virtnet_set_channels(struct net_device *dev,
1200 struct ethtool_channels *channels)
1202 struct virtnet_info *vi = netdev_priv(dev);
1203 u16 queue_pairs = channels->combined_count;
1206 /* We don't support separate rx/tx channels.
1207 * We don't allow setting 'other' channels.
1209 if (channels->rx_count || channels->tx_count || channels->other_count)
1212 if (queue_pairs > vi->max_queue_pairs)
1216 err = virtnet_set_queues(vi, queue_pairs);
1218 netif_set_real_num_tx_queues(dev, queue_pairs);
1219 netif_set_real_num_rx_queues(dev, queue_pairs);
1221 virtnet_set_affinity(vi);
1228 static void virtnet_get_channels(struct net_device *dev,
1229 struct ethtool_channels *channels)
1231 struct virtnet_info *vi = netdev_priv(dev);
1233 channels->combined_count = vi->curr_queue_pairs;
1234 channels->max_combined = vi->max_queue_pairs;
1235 channels->max_other = 0;
1236 channels->rx_count = 0;
1237 channels->tx_count = 0;
1238 channels->other_count = 0;
1241 static const struct ethtool_ops virtnet_ethtool_ops = {
1242 .get_drvinfo = virtnet_get_drvinfo,
1243 .get_link = ethtool_op_get_link,
1244 .get_ringparam = virtnet_get_ringparam,
1245 .set_channels = virtnet_set_channels,
1246 .get_channels = virtnet_get_channels,
1250 #define MAX_MTU 65535
1252 static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
1254 if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
1260 static const struct net_device_ops virtnet_netdev = {
1261 .ndo_open = virtnet_open,
1262 .ndo_stop = virtnet_close,
1263 .ndo_start_xmit = start_xmit,
1264 .ndo_validate_addr = eth_validate_addr,
1265 .ndo_set_mac_address = virtnet_set_mac_address,
1266 .ndo_set_rx_mode = virtnet_set_rx_mode,
1267 .ndo_change_mtu = virtnet_change_mtu,
1268 .ndo_get_stats64 = virtnet_stats,
1269 .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
1270 .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
1271 #ifdef CONFIG_NET_POLL_CONTROLLER
1272 .ndo_poll_controller = virtnet_netpoll,
1276 static void virtnet_config_changed_work(struct work_struct *work)
1278 struct virtnet_info *vi =
1279 container_of(work, struct virtnet_info, config_work);
1282 mutex_lock(&vi->config_lock);
1283 if (!vi->config_enable)
1286 if (virtio_config_val(vi->vdev, VIRTIO_NET_F_STATUS,
1287 offsetof(struct virtio_net_config, status),
1291 if (v & VIRTIO_NET_S_ANNOUNCE) {
1292 netdev_notify_peers(vi->dev);
1293 virtnet_ack_link_announce(vi);
1296 /* Ignore unknown (future) status bits */
1297 v &= VIRTIO_NET_S_LINK_UP;
1299 if (vi->status == v)
1304 if (vi->status & VIRTIO_NET_S_LINK_UP) {
1305 netif_carrier_on(vi->dev);
1306 netif_tx_wake_all_queues(vi->dev);
1308 netif_carrier_off(vi->dev);
1309 netif_tx_stop_all_queues(vi->dev);
1312 mutex_unlock(&vi->config_lock);
1315 static void virtnet_config_changed(struct virtio_device *vdev)
1317 struct virtnet_info *vi = vdev->priv;
1319 schedule_work(&vi->config_work);
1322 static void virtnet_free_queues(struct virtnet_info *vi)
1328 static void free_receive_bufs(struct virtnet_info *vi)
1332 for (i = 0; i < vi->max_queue_pairs; i++) {
1333 while (vi->rq[i].pages)
1334 __free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0);
1338 static void free_unused_bufs(struct virtnet_info *vi)
1343 for (i = 0; i < vi->max_queue_pairs; i++) {
1344 struct virtqueue *vq = vi->sq[i].vq;
1345 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL)
1349 for (i = 0; i < vi->max_queue_pairs; i++) {
1350 struct virtqueue *vq = vi->rq[i].vq;
1352 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) {
1353 if (vi->big_packets)
1354 give_pages(&vi->rq[i], buf);
1355 else if (vi->mergeable_rx_bufs)
1356 put_page(virt_to_head_page(buf));
1361 BUG_ON(vi->rq[i].num != 0);
1365 static void virtnet_del_vqs(struct virtnet_info *vi)
1367 struct virtio_device *vdev = vi->vdev;
1369 virtnet_clean_affinity(vi, -1);
1371 vdev->config->del_vqs(vdev);
1373 virtnet_free_queues(vi);
1376 static int virtnet_find_vqs(struct virtnet_info *vi)
1378 vq_callback_t **callbacks;
1379 struct virtqueue **vqs;
1384 /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
1385 * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
1386 * possible control vq.
1388 total_vqs = vi->max_queue_pairs * 2 +
1389 virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ);
1391 /* Allocate space for find_vqs parameters */
1392 vqs = kzalloc(total_vqs * sizeof(*vqs), GFP_KERNEL);
1395 callbacks = kmalloc(total_vqs * sizeof(*callbacks), GFP_KERNEL);
1398 names = kmalloc(total_vqs * sizeof(*names), GFP_KERNEL);
1402 /* Parameters for control virtqueue, if any */
1404 callbacks[total_vqs - 1] = NULL;
1405 names[total_vqs - 1] = "control";
1408 /* Allocate/initialize parameters for send/receive virtqueues */
1409 for (i = 0; i < vi->max_queue_pairs; i++) {
1410 callbacks[rxq2vq(i)] = skb_recv_done;
1411 callbacks[txq2vq(i)] = skb_xmit_done;
1412 sprintf(vi->rq[i].name, "input.%d", i);
1413 sprintf(vi->sq[i].name, "output.%d", i);
1414 names[rxq2vq(i)] = vi->rq[i].name;
1415 names[txq2vq(i)] = vi->sq[i].name;
1418 ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs, callbacks,
1424 vi->cvq = vqs[total_vqs - 1];
1425 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
1426 vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
1429 for (i = 0; i < vi->max_queue_pairs; i++) {
1430 vi->rq[i].vq = vqs[rxq2vq(i)];
1431 vi->sq[i].vq = vqs[txq2vq(i)];
1450 static int virtnet_alloc_queues(struct virtnet_info *vi)
1454 vi->sq = kzalloc(sizeof(*vi->sq) * vi->max_queue_pairs, GFP_KERNEL);
1457 vi->rq = kzalloc(sizeof(*vi->rq) * vi->max_queue_pairs, GFP_KERNEL);
1461 INIT_DELAYED_WORK(&vi->refill, refill_work);
1462 for (i = 0; i < vi->max_queue_pairs; i++) {
1463 vi->rq[i].pages = NULL;
1464 netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll,
1467 sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg));
1468 sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg));
1479 static int init_vqs(struct virtnet_info *vi)
1483 /* Allocate send & receive queues */
1484 ret = virtnet_alloc_queues(vi);
1488 ret = virtnet_find_vqs(vi);
1493 virtnet_set_affinity(vi);
1499 virtnet_free_queues(vi);
1504 static int virtnet_probe(struct virtio_device *vdev)
1507 struct net_device *dev;
1508 struct virtnet_info *vi;
1509 u16 max_queue_pairs;
1511 /* Find if host supports multiqueue virtio_net device */
1512 err = virtio_config_val(vdev, VIRTIO_NET_F_MQ,
1513 offsetof(struct virtio_net_config,
1514 max_virtqueue_pairs), &max_queue_pairs);
1516 /* We need at least 2 queue's */
1517 if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
1518 max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
1519 !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
1520 max_queue_pairs = 1;
1522 /* Allocate ourselves a network device with room for our info */
1523 dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs);
1527 /* Set up network device as normal. */
1528 dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE;
1529 dev->netdev_ops = &virtnet_netdev;
1530 dev->features = NETIF_F_HIGHDMA;
1532 SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
1533 SET_NETDEV_DEV(dev, &vdev->dev);
1535 /* Do we support "hardware" checksums? */
1536 if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
1537 /* This opens up the world of extra features. */
1538 dev->hw_features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1540 dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1542 if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
1543 dev->hw_features |= NETIF_F_TSO | NETIF_F_UFO
1544 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
1546 /* Individual feature bits: what can host handle? */
1547 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
1548 dev->hw_features |= NETIF_F_TSO;
1549 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
1550 dev->hw_features |= NETIF_F_TSO6;
1551 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
1552 dev->hw_features |= NETIF_F_TSO_ECN;
1553 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
1554 dev->hw_features |= NETIF_F_UFO;
1557 dev->features |= dev->hw_features & (NETIF_F_ALL_TSO|NETIF_F_UFO);
1558 /* (!csum && gso) case will be fixed by register_netdev() */
1560 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM))
1561 dev->features |= NETIF_F_RXCSUM;
1563 dev->vlan_features = dev->features;
1565 /* Configuration may specify what MAC to use. Otherwise random. */
1566 if (virtio_config_val_len(vdev, VIRTIO_NET_F_MAC,
1567 offsetof(struct virtio_net_config, mac),
1568 dev->dev_addr, dev->addr_len) < 0)
1569 eth_hw_addr_random(dev);
1571 /* Set up our device-specific information */
1572 vi = netdev_priv(dev);
1576 vi->stats = alloc_percpu(struct virtnet_stats);
1578 if (vi->stats == NULL)
1581 mutex_init(&vi->config_lock);
1582 vi->config_enable = true;
1583 INIT_WORK(&vi->config_work, virtnet_config_changed_work);
1585 /* If we can receive ANY GSO packets, we must allocate large ones. */
1586 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
1587 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
1588 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
1589 vi->big_packets = true;
1591 if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
1592 vi->mergeable_rx_bufs = true;
1594 if (virtio_has_feature(vdev, VIRTIO_F_ANY_LAYOUT))
1595 vi->any_header_sg = true;
1597 if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
1600 /* Use single tx/rx queue pair as default */
1601 vi->curr_queue_pairs = 1;
1602 vi->max_queue_pairs = max_queue_pairs;
1604 /* Allocate/initialize the rx/tx queues, and invoke find_vqs */
1609 netif_set_real_num_tx_queues(dev, 1);
1610 netif_set_real_num_rx_queues(dev, 1);
1612 err = register_netdev(dev);
1614 pr_debug("virtio_net: registering device failed\n");
1618 /* Last of all, set up some receive buffers. */
1619 for (i = 0; i < vi->curr_queue_pairs; i++) {
1620 try_fill_recv(&vi->rq[i], GFP_KERNEL);
1622 /* If we didn't even get one input buffer, we're useless. */
1623 if (vi->rq[i].num == 0) {
1624 free_unused_bufs(vi);
1626 goto free_recv_bufs;
1630 vi->nb.notifier_call = &virtnet_cpu_callback;
1631 err = register_hotcpu_notifier(&vi->nb);
1633 pr_debug("virtio_net: registering cpu notifier failed\n");
1634 goto free_recv_bufs;
1637 /* Assume link up if device can't report link status,
1638 otherwise get link status from config. */
1639 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
1640 netif_carrier_off(dev);
1641 schedule_work(&vi->config_work);
1643 vi->status = VIRTIO_NET_S_LINK_UP;
1644 netif_carrier_on(dev);
1647 pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
1648 dev->name, max_queue_pairs);
1653 free_receive_bufs(vi);
1654 unregister_netdev(dev);
1656 cancel_delayed_work_sync(&vi->refill);
1657 virtnet_del_vqs(vi);
1658 if (vi->alloc_frag.page)
1659 put_page(vi->alloc_frag.page);
1661 free_percpu(vi->stats);
1667 static void remove_vq_common(struct virtnet_info *vi)
1669 vi->vdev->config->reset(vi->vdev);
1671 /* Free unused buffers in both send and recv, if any. */
1672 free_unused_bufs(vi);
1674 free_receive_bufs(vi);
1676 virtnet_del_vqs(vi);
1679 static void virtnet_remove(struct virtio_device *vdev)
1681 struct virtnet_info *vi = vdev->priv;
1683 unregister_hotcpu_notifier(&vi->nb);
1685 /* Prevent config work handler from accessing the device. */
1686 mutex_lock(&vi->config_lock);
1687 vi->config_enable = false;
1688 mutex_unlock(&vi->config_lock);
1690 unregister_netdev(vi->dev);
1692 remove_vq_common(vi);
1693 if (vi->alloc_frag.page)
1694 put_page(vi->alloc_frag.page);
1696 flush_work(&vi->config_work);
1698 free_percpu(vi->stats);
1699 free_netdev(vi->dev);
1703 static int virtnet_freeze(struct virtio_device *vdev)
1705 struct virtnet_info *vi = vdev->priv;
1708 unregister_hotcpu_notifier(&vi->nb);
1710 /* Prevent config work handler from accessing the device */
1711 mutex_lock(&vi->config_lock);
1712 vi->config_enable = false;
1713 mutex_unlock(&vi->config_lock);
1715 netif_device_detach(vi->dev);
1716 cancel_delayed_work_sync(&vi->refill);
1718 if (netif_running(vi->dev))
1719 for (i = 0; i < vi->max_queue_pairs; i++) {
1720 napi_disable(&vi->rq[i].napi);
1721 netif_napi_del(&vi->rq[i].napi);
1724 remove_vq_common(vi);
1726 flush_work(&vi->config_work);
1731 static int virtnet_restore(struct virtio_device *vdev)
1733 struct virtnet_info *vi = vdev->priv;
1740 if (netif_running(vi->dev))
1741 for (i = 0; i < vi->max_queue_pairs; i++)
1742 virtnet_napi_enable(&vi->rq[i]);
1744 netif_device_attach(vi->dev);
1746 for (i = 0; i < vi->curr_queue_pairs; i++)
1747 if (!try_fill_recv(&vi->rq[i], GFP_KERNEL))
1748 schedule_delayed_work(&vi->refill, 0);
1750 mutex_lock(&vi->config_lock);
1751 vi->config_enable = true;
1752 mutex_unlock(&vi->config_lock);
1755 virtnet_set_queues(vi, vi->curr_queue_pairs);
1758 err = register_hotcpu_notifier(&vi->nb);
1766 static struct virtio_device_id id_table[] = {
1767 { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
1771 static unsigned int features[] = {
1772 VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
1773 VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
1774 VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
1775 VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
1776 VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO,
1777 VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ,
1778 VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN,
1779 VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ,
1780 VIRTIO_NET_F_CTRL_MAC_ADDR,
1781 VIRTIO_F_ANY_LAYOUT,
1784 static struct virtio_driver virtio_net_driver = {
1785 .feature_table = features,
1786 .feature_table_size = ARRAY_SIZE(features),
1787 .driver.name = KBUILD_MODNAME,
1788 .driver.owner = THIS_MODULE,
1789 .id_table = id_table,
1790 .probe = virtnet_probe,
1791 .remove = virtnet_remove,
1792 .config_changed = virtnet_config_changed,
1794 .freeze = virtnet_freeze,
1795 .restore = virtnet_restore,
1799 module_virtio_driver(virtio_net_driver);
1801 MODULE_DEVICE_TABLE(virtio, id_table);
1802 MODULE_DESCRIPTION("Virtio network driver");
1803 MODULE_LICENSE("GPL");