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 MAX_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
40 #define GOOD_COPY_LEN 128
42 #define VIRTNET_DRIVER_VERSION "1.0.0"
44 struct virtnet_stats {
45 struct u64_stats_sync tx_syncp;
46 struct u64_stats_sync rx_syncp;
54 /* Internal representation of a send virtqueue */
56 /* Virtqueue associated with this send _queue */
59 /* TX: fragments + linear part + virtio header */
60 struct scatterlist sg[MAX_SKB_FRAGS + 2];
62 /* Name of the send queue: output.$index */
66 /* Internal representation of a receive virtqueue */
67 struct receive_queue {
68 /* Virtqueue associated with this receive_queue */
71 struct napi_struct napi;
73 /* Number of input buffers, and max we've ever had. */
74 unsigned int num, max;
76 /* Chain pages by the private ptr. */
79 /* RX: fragments + linear part + virtio header */
80 struct scatterlist sg[MAX_SKB_FRAGS + 2];
82 /* Name of this receive queue: input.$index */
87 struct virtio_device *vdev;
88 struct virtqueue *cvq;
89 struct net_device *dev;
90 struct send_queue *sq;
91 struct receive_queue *rq;
94 /* Max # of queue pairs supported by the device */
97 /* # of queue pairs currently used by the driver */
100 /* I like... big packets and I cannot lie! */
103 /* Host will merge rx buffers for big packets (shake it! shake it!) */
104 bool mergeable_rx_bufs;
106 /* Has control virtqueue */
109 /* Host can handle any s/g split between our header and packet data */
112 /* enable config space updates */
115 /* Active statistics */
116 struct virtnet_stats __percpu *stats;
118 /* Work struct for refilling if we run low on memory. */
119 struct delayed_work refill;
121 /* Work struct for config space updates */
122 struct work_struct config_work;
124 /* Lock for config space updates */
125 struct mutex config_lock;
127 /* Page_frag for GFP_KERNEL packet buffer allocation when we run
130 struct page_frag alloc_frag;
132 /* Does the affinity hint is set for virtqueues? */
133 bool affinity_hint_set;
135 /* CPU hot plug notifier */
136 struct notifier_block nb;
139 struct skb_vnet_hdr {
141 struct virtio_net_hdr hdr;
142 struct virtio_net_hdr_mrg_rxbuf mhdr;
146 struct padded_vnet_hdr {
147 struct virtio_net_hdr hdr;
149 * virtio_net_hdr should be in a separated sg buffer because of a
150 * QEMU bug, and data sg buffer shares same page with this header sg.
151 * This padding makes next sg 16 byte aligned after virtio_net_hdr.
156 /* Converting between virtqueue no. and kernel tx/rx queue no.
157 * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq
159 static int vq2txq(struct virtqueue *vq)
161 return (vq->index - 1) / 2;
164 static int txq2vq(int txq)
169 static int vq2rxq(struct virtqueue *vq)
171 return vq->index / 2;
174 static int rxq2vq(int rxq)
179 static inline struct skb_vnet_hdr *skb_vnet_hdr(struct sk_buff *skb)
181 return (struct skb_vnet_hdr *)skb->cb;
185 * private is used to chain pages for big packets, put the whole
186 * most recent used list in the beginning for reuse
188 static void give_pages(struct receive_queue *rq, struct page *page)
192 /* Find end of list, sew whole thing into vi->rq.pages. */
193 for (end = page; end->private; end = (struct page *)end->private);
194 end->private = (unsigned long)rq->pages;
198 static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask)
200 struct page *p = rq->pages;
203 rq->pages = (struct page *)p->private;
204 /* clear private here, it is used to chain pages */
207 p = alloc_page(gfp_mask);
211 static void skb_xmit_done(struct virtqueue *vq)
213 struct virtnet_info *vi = vq->vdev->priv;
215 /* Suppress further interrupts. */
216 virtqueue_disable_cb(vq);
218 /* We were probably waiting for more output buffers. */
219 netif_wake_subqueue(vi->dev, vq2txq(vq));
222 /* Called from bottom half context */
223 static struct sk_buff *page_to_skb(struct receive_queue *rq,
224 struct page *page, unsigned int offset,
225 unsigned int len, unsigned int truesize)
227 struct virtnet_info *vi = rq->vq->vdev->priv;
229 struct skb_vnet_hdr *hdr;
230 unsigned int copy, hdr_len, hdr_padded_len;
233 p = page_address(page) + offset;
235 /* copy small packet so we can reuse these pages for small data */
236 skb = netdev_alloc_skb_ip_align(vi->dev, GOOD_COPY_LEN);
240 hdr = skb_vnet_hdr(skb);
242 if (vi->mergeable_rx_bufs) {
243 hdr_len = sizeof hdr->mhdr;
244 hdr_padded_len = sizeof hdr->mhdr;
246 hdr_len = sizeof hdr->hdr;
247 hdr_padded_len = sizeof(struct padded_vnet_hdr);
250 memcpy(hdr, p, hdr_len);
253 offset += hdr_padded_len;
257 if (copy > skb_tailroom(skb))
258 copy = skb_tailroom(skb);
259 memcpy(skb_put(skb, copy), p, copy);
264 if (vi->mergeable_rx_bufs) {
266 skb_add_rx_frag(skb, 0, page, offset, len, truesize);
273 * Verify that we can indeed put this data into a skb.
274 * This is here to handle cases when the device erroneously
275 * tries to receive more than is possible. This is usually
276 * the case of a broken device.
278 if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) {
279 net_dbg_ratelimited("%s: too much data\n", skb->dev->name);
283 BUG_ON(offset >= PAGE_SIZE);
285 unsigned int frag_size = min((unsigned)PAGE_SIZE - offset, len);
286 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page, offset,
287 frag_size, truesize);
289 page = (struct page *)page->private;
294 give_pages(rq, page);
299 static int receive_mergeable(struct receive_queue *rq, struct sk_buff *head_skb)
301 struct skb_vnet_hdr *hdr = skb_vnet_hdr(head_skb);
302 struct sk_buff *curr_skb = head_skb;
305 int num_buf, len, offset;
307 num_buf = hdr->mhdr.num_buffers;
309 int num_skb_frags = skb_shinfo(curr_skb)->nr_frags;
310 buf = virtqueue_get_buf(rq->vq, &len);
311 if (unlikely(!buf)) {
312 pr_debug("%s: rx error: %d buffers missing\n",
313 head_skb->dev->name, hdr->mhdr.num_buffers);
314 head_skb->dev->stats.rx_length_errors++;
317 if (unlikely(len > MAX_PACKET_LEN)) {
318 pr_debug("%s: rx error: merge buffer too long\n",
319 head_skb->dev->name);
320 len = MAX_PACKET_LEN;
322 if (unlikely(num_skb_frags == MAX_SKB_FRAGS)) {
323 struct sk_buff *nskb = alloc_skb(0, GFP_ATOMIC);
324 if (unlikely(!nskb)) {
325 head_skb->dev->stats.rx_dropped++;
328 if (curr_skb == head_skb)
329 skb_shinfo(curr_skb)->frag_list = nskb;
331 curr_skb->next = nskb;
333 head_skb->truesize += nskb->truesize;
336 if (curr_skb != head_skb) {
337 head_skb->data_len += len;
338 head_skb->len += len;
339 head_skb->truesize += MAX_PACKET_LEN;
341 page = virt_to_head_page(buf);
342 offset = buf - (char *)page_address(page);
343 if (skb_can_coalesce(curr_skb, num_skb_frags, page, offset)) {
345 skb_coalesce_rx_frag(curr_skb, num_skb_frags - 1,
346 len, MAX_PACKET_LEN);
348 skb_add_rx_frag(curr_skb, num_skb_frags, page,
357 static void receive_buf(struct receive_queue *rq, void *buf, unsigned int len)
359 struct virtnet_info *vi = rq->vq->vdev->priv;
360 struct net_device *dev = vi->dev;
361 struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
364 struct skb_vnet_hdr *hdr;
366 if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
367 pr_debug("%s: short packet %i\n", dev->name, len);
368 dev->stats.rx_length_errors++;
371 else if (vi->mergeable_rx_bufs)
372 put_page(virt_to_head_page(buf));
378 if (!vi->mergeable_rx_bufs && !vi->big_packets) {
380 len -= sizeof(struct virtio_net_hdr);
382 } else if (vi->mergeable_rx_bufs) {
383 struct page *page = virt_to_head_page(buf);
384 skb = page_to_skb(rq, page,
385 (char *)buf - (char *)page_address(page),
386 len, MAX_PACKET_LEN);
387 if (unlikely(!skb)) {
388 dev->stats.rx_dropped++;
392 if (receive_mergeable(rq, skb)) {
398 skb = page_to_skb(rq, page, 0, len, PAGE_SIZE);
399 if (unlikely(!skb)) {
400 dev->stats.rx_dropped++;
401 give_pages(rq, page);
406 hdr = skb_vnet_hdr(skb);
408 u64_stats_update_begin(&stats->rx_syncp);
409 stats->rx_bytes += skb->len;
411 u64_stats_update_end(&stats->rx_syncp);
413 if (hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
414 pr_debug("Needs csum!\n");
415 if (!skb_partial_csum_set(skb,
417 hdr->hdr.csum_offset))
419 } else if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) {
420 skb->ip_summed = CHECKSUM_UNNECESSARY;
423 skb->protocol = eth_type_trans(skb, dev);
424 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
425 ntohs(skb->protocol), skb->len, skb->pkt_type);
427 if (hdr->hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
429 switch (hdr->hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
430 case VIRTIO_NET_HDR_GSO_TCPV4:
431 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
433 case VIRTIO_NET_HDR_GSO_UDP:
434 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
436 case VIRTIO_NET_HDR_GSO_TCPV6:
437 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
440 net_warn_ratelimited("%s: bad gso type %u.\n",
441 dev->name, hdr->hdr.gso_type);
445 if (hdr->hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
446 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
448 skb_shinfo(skb)->gso_size = hdr->hdr.gso_size;
449 if (skb_shinfo(skb)->gso_size == 0) {
450 net_warn_ratelimited("%s: zero gso size.\n", dev->name);
454 /* Header must be checked, and gso_segs computed. */
455 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
456 skb_shinfo(skb)->gso_segs = 0;
459 netif_receive_skb(skb);
463 dev->stats.rx_frame_errors++;
467 static int add_recvbuf_small(struct receive_queue *rq, gfp_t gfp)
469 struct virtnet_info *vi = rq->vq->vdev->priv;
471 struct skb_vnet_hdr *hdr;
474 skb = __netdev_alloc_skb_ip_align(vi->dev, MAX_PACKET_LEN, gfp);
478 skb_put(skb, MAX_PACKET_LEN);
480 hdr = skb_vnet_hdr(skb);
481 sg_set_buf(rq->sg, &hdr->hdr, sizeof hdr->hdr);
483 skb_to_sgvec(skb, rq->sg + 1, 0, skb->len);
485 err = virtqueue_add_inbuf(rq->vq, rq->sg, 2, skb, gfp);
492 static int add_recvbuf_big(struct receive_queue *rq, gfp_t gfp)
494 struct page *first, *list = NULL;
498 /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */
499 for (i = MAX_SKB_FRAGS + 1; i > 1; --i) {
500 first = get_a_page(rq, gfp);
503 give_pages(rq, list);
506 sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE);
508 /* chain new page in list head to match sg */
509 first->private = (unsigned long)list;
513 first = get_a_page(rq, gfp);
515 give_pages(rq, list);
518 p = page_address(first);
520 /* rq->sg[0], rq->sg[1] share the same page */
521 /* a separated rq->sg[0] for virtio_net_hdr only due to QEMU bug */
522 sg_set_buf(&rq->sg[0], p, sizeof(struct virtio_net_hdr));
524 /* rq->sg[1] for data packet, from offset */
525 offset = sizeof(struct padded_vnet_hdr);
526 sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset);
528 /* chain first in list head */
529 first->private = (unsigned long)list;
530 err = virtqueue_add_inbuf(rq->vq, rq->sg, MAX_SKB_FRAGS + 2,
533 give_pages(rq, first);
538 static int add_recvbuf_mergeable(struct receive_queue *rq, gfp_t gfp)
540 struct virtnet_info *vi = rq->vq->vdev->priv;
544 if (gfp & __GFP_WAIT) {
545 if (skb_page_frag_refill(MAX_PACKET_LEN, &vi->alloc_frag,
547 buf = (char *)page_address(vi->alloc_frag.page) +
548 vi->alloc_frag.offset;
549 get_page(vi->alloc_frag.page);
550 vi->alloc_frag.offset += MAX_PACKET_LEN;
553 buf = netdev_alloc_frag(MAX_PACKET_LEN);
558 sg_init_one(rq->sg, buf, MAX_PACKET_LEN);
559 err = virtqueue_add_inbuf(rq->vq, rq->sg, 1, buf, gfp);
561 put_page(virt_to_head_page(buf));
567 * Returns false if we couldn't fill entirely (OOM).
569 * Normally run in the receive path, but can also be run from ndo_open
570 * before we're receiving packets, or from refill_work which is
571 * careful to disable receiving (using napi_disable).
573 static bool try_fill_recv(struct receive_queue *rq, gfp_t gfp)
575 struct virtnet_info *vi = rq->vq->vdev->priv;
580 if (vi->mergeable_rx_bufs)
581 err = add_recvbuf_mergeable(rq, gfp);
582 else if (vi->big_packets)
583 err = add_recvbuf_big(rq, gfp);
585 err = add_recvbuf_small(rq, gfp);
587 oom = err == -ENOMEM;
591 } while (rq->vq->num_free);
592 if (unlikely(rq->num > rq->max))
594 if (unlikely(!virtqueue_kick(rq->vq)))
599 static void skb_recv_done(struct virtqueue *rvq)
601 struct virtnet_info *vi = rvq->vdev->priv;
602 struct receive_queue *rq = &vi->rq[vq2rxq(rvq)];
604 /* Schedule NAPI, Suppress further interrupts if successful. */
605 if (napi_schedule_prep(&rq->napi)) {
606 virtqueue_disable_cb(rvq);
607 __napi_schedule(&rq->napi);
611 static void virtnet_napi_enable(struct receive_queue *rq)
613 napi_enable(&rq->napi);
615 /* If all buffers were filled by other side before we napi_enabled, we
616 * won't get another interrupt, so process any outstanding packets
617 * now. virtnet_poll wants re-enable the queue, so we disable here.
618 * We synchronize against interrupts via NAPI_STATE_SCHED */
619 if (napi_schedule_prep(&rq->napi)) {
620 virtqueue_disable_cb(rq->vq);
622 __napi_schedule(&rq->napi);
627 static void refill_work(struct work_struct *work)
629 struct virtnet_info *vi =
630 container_of(work, struct virtnet_info, refill.work);
634 for (i = 0; i < vi->curr_queue_pairs; i++) {
635 struct receive_queue *rq = &vi->rq[i];
637 napi_disable(&rq->napi);
638 still_empty = !try_fill_recv(rq, GFP_KERNEL);
639 virtnet_napi_enable(rq);
641 /* In theory, this can happen: if we don't get any buffers in
642 * we will *never* try to fill again.
645 schedule_delayed_work(&vi->refill, HZ/2);
649 static int virtnet_poll(struct napi_struct *napi, int budget)
651 struct receive_queue *rq =
652 container_of(napi, struct receive_queue, napi);
653 struct virtnet_info *vi = rq->vq->vdev->priv;
655 unsigned int r, len, received = 0;
658 while (received < budget &&
659 (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) {
660 receive_buf(rq, buf, len);
665 if (rq->num < rq->max / 2) {
666 if (!try_fill_recv(rq, GFP_ATOMIC))
667 schedule_delayed_work(&vi->refill, 0);
670 /* Out of packets? */
671 if (received < budget) {
672 r = virtqueue_enable_cb_prepare(rq->vq);
674 if (unlikely(virtqueue_poll(rq->vq, r)) &&
675 napi_schedule_prep(napi)) {
676 virtqueue_disable_cb(rq->vq);
677 __napi_schedule(napi);
685 static int virtnet_open(struct net_device *dev)
687 struct virtnet_info *vi = netdev_priv(dev);
690 for (i = 0; i < vi->max_queue_pairs; i++) {
691 if (i < vi->curr_queue_pairs)
692 /* Make sure we have some buffers: if oom use wq. */
693 if (!try_fill_recv(&vi->rq[i], GFP_KERNEL))
694 schedule_delayed_work(&vi->refill, 0);
695 virtnet_napi_enable(&vi->rq[i]);
701 static void free_old_xmit_skbs(struct send_queue *sq)
705 struct virtnet_info *vi = sq->vq->vdev->priv;
706 struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
708 while ((skb = virtqueue_get_buf(sq->vq, &len)) != NULL) {
709 pr_debug("Sent skb %p\n", skb);
711 u64_stats_update_begin(&stats->tx_syncp);
712 stats->tx_bytes += skb->len;
714 u64_stats_update_end(&stats->tx_syncp);
716 dev_kfree_skb_any(skb);
720 static int xmit_skb(struct send_queue *sq, struct sk_buff *skb)
722 struct skb_vnet_hdr *hdr;
723 const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
724 struct virtnet_info *vi = sq->vq->vdev->priv;
729 pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
730 if (vi->mergeable_rx_bufs)
731 hdr_len = sizeof hdr->mhdr;
733 hdr_len = sizeof hdr->hdr;
735 can_push = vi->any_header_sg &&
736 !((unsigned long)skb->data & (__alignof__(*hdr) - 1)) &&
737 !skb_header_cloned(skb) && skb_headroom(skb) >= hdr_len;
738 /* Even if we can, don't push here yet as this would skew
739 * csum_start offset below. */
741 hdr = (struct skb_vnet_hdr *)(skb->data - hdr_len);
743 hdr = skb_vnet_hdr(skb);
745 if (skb->ip_summed == CHECKSUM_PARTIAL) {
746 hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
747 hdr->hdr.csum_start = skb_checksum_start_offset(skb);
748 hdr->hdr.csum_offset = skb->csum_offset;
751 hdr->hdr.csum_offset = hdr->hdr.csum_start = 0;
754 if (skb_is_gso(skb)) {
755 hdr->hdr.hdr_len = skb_headlen(skb);
756 hdr->hdr.gso_size = skb_shinfo(skb)->gso_size;
757 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
758 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
759 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
760 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
761 else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
762 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
765 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
766 hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
768 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
769 hdr->hdr.gso_size = hdr->hdr.hdr_len = 0;
772 if (vi->mergeable_rx_bufs)
773 hdr->mhdr.num_buffers = 0;
776 __skb_push(skb, hdr_len);
777 num_sg = skb_to_sgvec(skb, sq->sg, 0, skb->len);
778 /* Pull header back to avoid skew in tx bytes calculations. */
779 __skb_pull(skb, hdr_len);
781 sg_set_buf(sq->sg, hdr, hdr_len);
782 num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len) + 1;
784 return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC);
787 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
789 struct virtnet_info *vi = netdev_priv(dev);
790 int qnum = skb_get_queue_mapping(skb);
791 struct send_queue *sq = &vi->sq[qnum];
794 /* Free up any pending old buffers before queueing new ones. */
795 free_old_xmit_skbs(sq);
797 /* Try to transmit */
798 err = xmit_skb(sq, skb);
800 /* This should not happen! */
801 if (unlikely(err) || unlikely(!virtqueue_kick(sq->vq))) {
802 dev->stats.tx_fifo_errors++;
805 "Unexpected TXQ (%d) queue failure: %d\n", qnum, err);
806 dev->stats.tx_dropped++;
811 /* Don't wait up for transmitted skbs to be freed. */
815 /* Apparently nice girls don't return TX_BUSY; stop the queue
816 * before it gets out of hand. Naturally, this wastes entries. */
817 if (sq->vq->num_free < 2+MAX_SKB_FRAGS) {
818 netif_stop_subqueue(dev, qnum);
819 if (unlikely(!virtqueue_enable_cb_delayed(sq->vq))) {
820 /* More just got used, free them then recheck. */
821 free_old_xmit_skbs(sq);
822 if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) {
823 netif_start_subqueue(dev, qnum);
824 virtqueue_disable_cb(sq->vq);
833 * Send command via the control virtqueue and check status. Commands
834 * supported by the hypervisor, as indicated by feature bits, should
835 * never fail unless improperly formated.
837 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
838 struct scatterlist *out,
839 struct scatterlist *in)
841 struct scatterlist *sgs[4], hdr, stat;
842 struct virtio_net_ctrl_hdr ctrl;
843 virtio_net_ctrl_ack status = ~0;
844 unsigned out_num = 0, in_num = 0, tmp;
846 /* Caller should know better */
847 BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ));
852 sg_init_one(&hdr, &ctrl, sizeof(ctrl));
853 sgs[out_num++] = &hdr;
856 sgs[out_num++] = out;
858 sgs[out_num + in_num++] = in;
860 /* Add return status. */
861 sg_init_one(&stat, &status, sizeof(status));
862 sgs[out_num + in_num++] = &stat;
864 BUG_ON(out_num + in_num > ARRAY_SIZE(sgs));
865 BUG_ON(virtqueue_add_sgs(vi->cvq, sgs, out_num, in_num, vi, GFP_ATOMIC)
868 if (unlikely(!virtqueue_kick(vi->cvq)))
869 return status == VIRTIO_NET_OK;
871 /* Spin for a response, the kick causes an ioport write, trapping
872 * into the hypervisor, so the request should be handled immediately.
874 while (!virtqueue_get_buf(vi->cvq, &tmp) &&
875 !virtqueue_is_broken(vi->cvq))
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)) {
905 /* Naturally, this has an atomicity problem. */
906 for (i = 0; i < dev->addr_len; i++)
908 offsetof(struct virtio_net_config, mac) +
909 i, addr->sa_data[i]);
912 eth_commit_mac_addr_change(dev, p);
917 static struct rtnl_link_stats64 *virtnet_stats(struct net_device *dev,
918 struct rtnl_link_stats64 *tot)
920 struct virtnet_info *vi = netdev_priv(dev);
924 for_each_possible_cpu(cpu) {
925 struct virtnet_stats *stats = per_cpu_ptr(vi->stats, cpu);
926 u64 tpackets, tbytes, rpackets, rbytes;
929 start = u64_stats_fetch_begin_bh(&stats->tx_syncp);
930 tpackets = stats->tx_packets;
931 tbytes = stats->tx_bytes;
932 } while (u64_stats_fetch_retry_bh(&stats->tx_syncp, start));
935 start = u64_stats_fetch_begin_bh(&stats->rx_syncp);
936 rpackets = stats->rx_packets;
937 rbytes = stats->rx_bytes;
938 } while (u64_stats_fetch_retry_bh(&stats->rx_syncp, start));
940 tot->rx_packets += rpackets;
941 tot->tx_packets += tpackets;
942 tot->rx_bytes += rbytes;
943 tot->tx_bytes += tbytes;
946 tot->tx_dropped = dev->stats.tx_dropped;
947 tot->tx_fifo_errors = dev->stats.tx_fifo_errors;
948 tot->rx_dropped = dev->stats.rx_dropped;
949 tot->rx_length_errors = dev->stats.rx_length_errors;
950 tot->rx_frame_errors = dev->stats.rx_frame_errors;
955 #ifdef CONFIG_NET_POLL_CONTROLLER
956 static void virtnet_netpoll(struct net_device *dev)
958 struct virtnet_info *vi = netdev_priv(dev);
961 for (i = 0; i < vi->curr_queue_pairs; i++)
962 napi_schedule(&vi->rq[i].napi);
966 static void virtnet_ack_link_announce(struct virtnet_info *vi)
969 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE,
970 VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL, NULL))
971 dev_warn(&vi->dev->dev, "Failed to ack link announce.\n");
975 static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs)
977 struct scatterlist sg;
978 struct virtio_net_ctrl_mq s;
979 struct net_device *dev = vi->dev;
981 if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ))
984 s.virtqueue_pairs = queue_pairs;
985 sg_init_one(&sg, &s, sizeof(s));
987 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ,
988 VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg, NULL)) {
989 dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n",
993 vi->curr_queue_pairs = queue_pairs;
994 /* virtnet_open() will refill when device is going to up. */
995 if (dev->flags & IFF_UP)
996 schedule_delayed_work(&vi->refill, 0);
1002 static int virtnet_close(struct net_device *dev)
1004 struct virtnet_info *vi = netdev_priv(dev);
1007 /* Make sure refill_work doesn't re-enable napi! */
1008 cancel_delayed_work_sync(&vi->refill);
1010 for (i = 0; i < vi->max_queue_pairs; i++)
1011 napi_disable(&vi->rq[i].napi);
1016 static void virtnet_set_rx_mode(struct net_device *dev)
1018 struct virtnet_info *vi = netdev_priv(dev);
1019 struct scatterlist sg[2];
1020 u8 promisc, allmulti;
1021 struct virtio_net_ctrl_mac *mac_data;
1022 struct netdev_hw_addr *ha;
1028 /* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */
1029 if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
1032 promisc = ((dev->flags & IFF_PROMISC) != 0);
1033 allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
1035 sg_init_one(sg, &promisc, sizeof(promisc));
1037 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
1038 VIRTIO_NET_CTRL_RX_PROMISC,
1040 dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
1041 promisc ? "en" : "dis");
1043 sg_init_one(sg, &allmulti, sizeof(allmulti));
1045 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
1046 VIRTIO_NET_CTRL_RX_ALLMULTI,
1048 dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
1049 allmulti ? "en" : "dis");
1051 uc_count = netdev_uc_count(dev);
1052 mc_count = netdev_mc_count(dev);
1053 /* MAC filter - use one buffer for both lists */
1054 buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
1055 (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
1060 sg_init_table(sg, 2);
1062 /* Store the unicast list and count in the front of the buffer */
1063 mac_data->entries = uc_count;
1065 netdev_for_each_uc_addr(ha, dev)
1066 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
1068 sg_set_buf(&sg[0], mac_data,
1069 sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
1071 /* multicast list and count fill the end */
1072 mac_data = (void *)&mac_data->macs[uc_count][0];
1074 mac_data->entries = mc_count;
1076 netdev_for_each_mc_addr(ha, dev)
1077 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
1079 sg_set_buf(&sg[1], mac_data,
1080 sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
1082 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
1083 VIRTIO_NET_CTRL_MAC_TABLE_SET,
1085 dev_warn(&dev->dev, "Failed to set MAC fitler table.\n");
1090 static int virtnet_vlan_rx_add_vid(struct net_device *dev,
1091 __be16 proto, u16 vid)
1093 struct virtnet_info *vi = netdev_priv(dev);
1094 struct scatterlist sg;
1096 sg_init_one(&sg, &vid, sizeof(vid));
1098 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1099 VIRTIO_NET_CTRL_VLAN_ADD, &sg, NULL))
1100 dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
1104 static int virtnet_vlan_rx_kill_vid(struct net_device *dev,
1105 __be16 proto, u16 vid)
1107 struct virtnet_info *vi = netdev_priv(dev);
1108 struct scatterlist sg;
1110 sg_init_one(&sg, &vid, sizeof(vid));
1112 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1113 VIRTIO_NET_CTRL_VLAN_DEL, &sg, NULL))
1114 dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
1118 static void virtnet_clean_affinity(struct virtnet_info *vi, long hcpu)
1122 if (vi->affinity_hint_set) {
1123 for (i = 0; i < vi->max_queue_pairs; i++) {
1124 virtqueue_set_affinity(vi->rq[i].vq, -1);
1125 virtqueue_set_affinity(vi->sq[i].vq, -1);
1128 vi->affinity_hint_set = false;
1132 static void virtnet_set_affinity(struct virtnet_info *vi)
1137 /* In multiqueue mode, when the number of cpu is equal to the number of
1138 * queue pairs, we let the queue pairs to be private to one cpu by
1139 * setting the affinity hint to eliminate the contention.
1141 if (vi->curr_queue_pairs == 1 ||
1142 vi->max_queue_pairs != num_online_cpus()) {
1143 virtnet_clean_affinity(vi, -1);
1148 for_each_online_cpu(cpu) {
1149 virtqueue_set_affinity(vi->rq[i].vq, cpu);
1150 virtqueue_set_affinity(vi->sq[i].vq, cpu);
1151 netif_set_xps_queue(vi->dev, cpumask_of(cpu), i);
1155 vi->affinity_hint_set = true;
1158 static int virtnet_cpu_callback(struct notifier_block *nfb,
1159 unsigned long action, void *hcpu)
1161 struct virtnet_info *vi = container_of(nfb, struct virtnet_info, nb);
1163 switch(action & ~CPU_TASKS_FROZEN) {
1165 case CPU_DOWN_FAILED:
1167 virtnet_set_affinity(vi);
1169 case CPU_DOWN_PREPARE:
1170 virtnet_clean_affinity(vi, (long)hcpu);
1179 static void virtnet_get_ringparam(struct net_device *dev,
1180 struct ethtool_ringparam *ring)
1182 struct virtnet_info *vi = netdev_priv(dev);
1184 ring->rx_max_pending = virtqueue_get_vring_size(vi->rq[0].vq);
1185 ring->tx_max_pending = virtqueue_get_vring_size(vi->sq[0].vq);
1186 ring->rx_pending = ring->rx_max_pending;
1187 ring->tx_pending = ring->tx_max_pending;
1191 static void virtnet_get_drvinfo(struct net_device *dev,
1192 struct ethtool_drvinfo *info)
1194 struct virtnet_info *vi = netdev_priv(dev);
1195 struct virtio_device *vdev = vi->vdev;
1197 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
1198 strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version));
1199 strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info));
1203 /* TODO: Eliminate OOO packets during switching */
1204 static int virtnet_set_channels(struct net_device *dev,
1205 struct ethtool_channels *channels)
1207 struct virtnet_info *vi = netdev_priv(dev);
1208 u16 queue_pairs = channels->combined_count;
1211 /* We don't support separate rx/tx channels.
1212 * We don't allow setting 'other' channels.
1214 if (channels->rx_count || channels->tx_count || channels->other_count)
1217 if (queue_pairs > vi->max_queue_pairs)
1221 err = virtnet_set_queues(vi, queue_pairs);
1223 netif_set_real_num_tx_queues(dev, queue_pairs);
1224 netif_set_real_num_rx_queues(dev, queue_pairs);
1226 virtnet_set_affinity(vi);
1233 static void virtnet_get_channels(struct net_device *dev,
1234 struct ethtool_channels *channels)
1236 struct virtnet_info *vi = netdev_priv(dev);
1238 channels->combined_count = vi->curr_queue_pairs;
1239 channels->max_combined = vi->max_queue_pairs;
1240 channels->max_other = 0;
1241 channels->rx_count = 0;
1242 channels->tx_count = 0;
1243 channels->other_count = 0;
1246 static const struct ethtool_ops virtnet_ethtool_ops = {
1247 .get_drvinfo = virtnet_get_drvinfo,
1248 .get_link = ethtool_op_get_link,
1249 .get_ringparam = virtnet_get_ringparam,
1250 .set_channels = virtnet_set_channels,
1251 .get_channels = virtnet_get_channels,
1255 #define MAX_MTU 65535
1257 static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
1259 if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
1265 static const struct net_device_ops virtnet_netdev = {
1266 .ndo_open = virtnet_open,
1267 .ndo_stop = virtnet_close,
1268 .ndo_start_xmit = start_xmit,
1269 .ndo_validate_addr = eth_validate_addr,
1270 .ndo_set_mac_address = virtnet_set_mac_address,
1271 .ndo_set_rx_mode = virtnet_set_rx_mode,
1272 .ndo_change_mtu = virtnet_change_mtu,
1273 .ndo_get_stats64 = virtnet_stats,
1274 .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
1275 .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
1276 #ifdef CONFIG_NET_POLL_CONTROLLER
1277 .ndo_poll_controller = virtnet_netpoll,
1281 static void virtnet_config_changed_work(struct work_struct *work)
1283 struct virtnet_info *vi =
1284 container_of(work, struct virtnet_info, config_work);
1287 mutex_lock(&vi->config_lock);
1288 if (!vi->config_enable)
1291 if (virtio_cread_feature(vi->vdev, VIRTIO_NET_F_STATUS,
1292 struct virtio_net_config, status, &v) < 0)
1295 if (v & VIRTIO_NET_S_ANNOUNCE) {
1296 netdev_notify_peers(vi->dev);
1297 virtnet_ack_link_announce(vi);
1300 /* Ignore unknown (future) status bits */
1301 v &= VIRTIO_NET_S_LINK_UP;
1303 if (vi->status == v)
1308 if (vi->status & VIRTIO_NET_S_LINK_UP) {
1309 netif_carrier_on(vi->dev);
1310 netif_tx_wake_all_queues(vi->dev);
1312 netif_carrier_off(vi->dev);
1313 netif_tx_stop_all_queues(vi->dev);
1316 mutex_unlock(&vi->config_lock);
1319 static void virtnet_config_changed(struct virtio_device *vdev)
1321 struct virtnet_info *vi = vdev->priv;
1323 schedule_work(&vi->config_work);
1326 static void virtnet_free_queues(struct virtnet_info *vi)
1332 static void free_receive_bufs(struct virtnet_info *vi)
1336 for (i = 0; i < vi->max_queue_pairs; i++) {
1337 while (vi->rq[i].pages)
1338 __free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0);
1342 static void free_unused_bufs(struct virtnet_info *vi)
1347 for (i = 0; i < vi->max_queue_pairs; i++) {
1348 struct virtqueue *vq = vi->sq[i].vq;
1349 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL)
1353 for (i = 0; i < vi->max_queue_pairs; i++) {
1354 struct virtqueue *vq = vi->rq[i].vq;
1356 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) {
1357 if (vi->big_packets)
1358 give_pages(&vi->rq[i], buf);
1359 else if (vi->mergeable_rx_bufs)
1360 put_page(virt_to_head_page(buf));
1365 BUG_ON(vi->rq[i].num != 0);
1369 static void virtnet_del_vqs(struct virtnet_info *vi)
1371 struct virtio_device *vdev = vi->vdev;
1373 virtnet_clean_affinity(vi, -1);
1375 vdev->config->del_vqs(vdev);
1377 virtnet_free_queues(vi);
1380 static int virtnet_find_vqs(struct virtnet_info *vi)
1382 vq_callback_t **callbacks;
1383 struct virtqueue **vqs;
1388 /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
1389 * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
1390 * possible control vq.
1392 total_vqs = vi->max_queue_pairs * 2 +
1393 virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ);
1395 /* Allocate space for find_vqs parameters */
1396 vqs = kzalloc(total_vqs * sizeof(*vqs), GFP_KERNEL);
1399 callbacks = kmalloc(total_vqs * sizeof(*callbacks), GFP_KERNEL);
1402 names = kmalloc(total_vqs * sizeof(*names), GFP_KERNEL);
1406 /* Parameters for control virtqueue, if any */
1408 callbacks[total_vqs - 1] = NULL;
1409 names[total_vqs - 1] = "control";
1412 /* Allocate/initialize parameters for send/receive virtqueues */
1413 for (i = 0; i < vi->max_queue_pairs; i++) {
1414 callbacks[rxq2vq(i)] = skb_recv_done;
1415 callbacks[txq2vq(i)] = skb_xmit_done;
1416 sprintf(vi->rq[i].name, "input.%d", i);
1417 sprintf(vi->sq[i].name, "output.%d", i);
1418 names[rxq2vq(i)] = vi->rq[i].name;
1419 names[txq2vq(i)] = vi->sq[i].name;
1422 ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs, callbacks,
1428 vi->cvq = vqs[total_vqs - 1];
1429 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
1430 vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
1433 for (i = 0; i < vi->max_queue_pairs; i++) {
1434 vi->rq[i].vq = vqs[rxq2vq(i)];
1435 vi->sq[i].vq = vqs[txq2vq(i)];
1454 static int virtnet_alloc_queues(struct virtnet_info *vi)
1458 vi->sq = kzalloc(sizeof(*vi->sq) * vi->max_queue_pairs, GFP_KERNEL);
1461 vi->rq = kzalloc(sizeof(*vi->rq) * vi->max_queue_pairs, GFP_KERNEL);
1465 INIT_DELAYED_WORK(&vi->refill, refill_work);
1466 for (i = 0; i < vi->max_queue_pairs; i++) {
1467 vi->rq[i].pages = NULL;
1468 netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll,
1471 sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg));
1472 sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg));
1483 static int init_vqs(struct virtnet_info *vi)
1487 /* Allocate send & receive queues */
1488 ret = virtnet_alloc_queues(vi);
1492 ret = virtnet_find_vqs(vi);
1497 virtnet_set_affinity(vi);
1503 virtnet_free_queues(vi);
1508 static int virtnet_probe(struct virtio_device *vdev)
1511 struct net_device *dev;
1512 struct virtnet_info *vi;
1513 u16 max_queue_pairs;
1515 /* Find if host supports multiqueue virtio_net device */
1516 err = virtio_cread_feature(vdev, VIRTIO_NET_F_MQ,
1517 struct virtio_net_config,
1518 max_virtqueue_pairs, &max_queue_pairs);
1520 /* We need at least 2 queue's */
1521 if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
1522 max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
1523 !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
1524 max_queue_pairs = 1;
1526 /* Allocate ourselves a network device with room for our info */
1527 dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs);
1531 /* Set up network device as normal. */
1532 dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE;
1533 dev->netdev_ops = &virtnet_netdev;
1534 dev->features = NETIF_F_HIGHDMA;
1536 SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
1537 SET_NETDEV_DEV(dev, &vdev->dev);
1539 /* Do we support "hardware" checksums? */
1540 if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
1541 /* This opens up the world of extra features. */
1542 dev->hw_features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1544 dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1546 if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
1547 dev->hw_features |= NETIF_F_TSO | NETIF_F_UFO
1548 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
1550 /* Individual feature bits: what can host handle? */
1551 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
1552 dev->hw_features |= NETIF_F_TSO;
1553 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
1554 dev->hw_features |= NETIF_F_TSO6;
1555 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
1556 dev->hw_features |= NETIF_F_TSO_ECN;
1557 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
1558 dev->hw_features |= NETIF_F_UFO;
1561 dev->features |= dev->hw_features & (NETIF_F_ALL_TSO|NETIF_F_UFO);
1562 /* (!csum && gso) case will be fixed by register_netdev() */
1564 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM))
1565 dev->features |= NETIF_F_RXCSUM;
1567 dev->vlan_features = dev->features;
1569 /* Configuration may specify what MAC to use. Otherwise random. */
1570 if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
1571 virtio_cread_bytes(vdev,
1572 offsetof(struct virtio_net_config, mac),
1573 dev->dev_addr, dev->addr_len);
1575 eth_hw_addr_random(dev);
1577 /* Set up our device-specific information */
1578 vi = netdev_priv(dev);
1582 vi->stats = alloc_percpu(struct virtnet_stats);
1584 if (vi->stats == NULL)
1587 for_each_possible_cpu(i) {
1588 struct virtnet_stats *virtnet_stats;
1589 virtnet_stats = per_cpu_ptr(vi->stats, i);
1590 u64_stats_init(&virtnet_stats->tx_syncp);
1591 u64_stats_init(&virtnet_stats->rx_syncp);
1594 mutex_init(&vi->config_lock);
1595 vi->config_enable = true;
1596 INIT_WORK(&vi->config_work, virtnet_config_changed_work);
1598 /* If we can receive ANY GSO packets, we must allocate large ones. */
1599 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
1600 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
1601 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
1602 vi->big_packets = true;
1604 if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
1605 vi->mergeable_rx_bufs = true;
1607 if (virtio_has_feature(vdev, VIRTIO_F_ANY_LAYOUT))
1608 vi->any_header_sg = true;
1610 if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
1613 /* Use single tx/rx queue pair as default */
1614 vi->curr_queue_pairs = 1;
1615 vi->max_queue_pairs = max_queue_pairs;
1617 /* Allocate/initialize the rx/tx queues, and invoke find_vqs */
1622 netif_set_real_num_tx_queues(dev, 1);
1623 netif_set_real_num_rx_queues(dev, 1);
1625 err = register_netdev(dev);
1627 pr_debug("virtio_net: registering device failed\n");
1631 /* Last of all, set up some receive buffers. */
1632 for (i = 0; i < vi->curr_queue_pairs; i++) {
1633 try_fill_recv(&vi->rq[i], GFP_KERNEL);
1635 /* If we didn't even get one input buffer, we're useless. */
1636 if (vi->rq[i].num == 0) {
1637 free_unused_bufs(vi);
1639 goto free_recv_bufs;
1643 vi->nb.notifier_call = &virtnet_cpu_callback;
1644 err = register_hotcpu_notifier(&vi->nb);
1646 pr_debug("virtio_net: registering cpu notifier failed\n");
1647 goto free_recv_bufs;
1650 /* Assume link up if device can't report link status,
1651 otherwise get link status from config. */
1652 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
1653 netif_carrier_off(dev);
1654 schedule_work(&vi->config_work);
1656 vi->status = VIRTIO_NET_S_LINK_UP;
1657 netif_carrier_on(dev);
1660 pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
1661 dev->name, max_queue_pairs);
1666 free_receive_bufs(vi);
1667 unregister_netdev(dev);
1669 cancel_delayed_work_sync(&vi->refill);
1670 virtnet_del_vqs(vi);
1671 if (vi->alloc_frag.page)
1672 put_page(vi->alloc_frag.page);
1674 free_percpu(vi->stats);
1680 static void remove_vq_common(struct virtnet_info *vi)
1682 vi->vdev->config->reset(vi->vdev);
1684 /* Free unused buffers in both send and recv, if any. */
1685 free_unused_bufs(vi);
1687 free_receive_bufs(vi);
1689 virtnet_del_vqs(vi);
1692 static void virtnet_remove(struct virtio_device *vdev)
1694 struct virtnet_info *vi = vdev->priv;
1696 unregister_hotcpu_notifier(&vi->nb);
1698 /* Prevent config work handler from accessing the device. */
1699 mutex_lock(&vi->config_lock);
1700 vi->config_enable = false;
1701 mutex_unlock(&vi->config_lock);
1703 unregister_netdev(vi->dev);
1705 remove_vq_common(vi);
1706 if (vi->alloc_frag.page)
1707 put_page(vi->alloc_frag.page);
1709 flush_work(&vi->config_work);
1711 free_percpu(vi->stats);
1712 free_netdev(vi->dev);
1715 #ifdef CONFIG_PM_SLEEP
1716 static int virtnet_freeze(struct virtio_device *vdev)
1718 struct virtnet_info *vi = vdev->priv;
1721 unregister_hotcpu_notifier(&vi->nb);
1723 /* Prevent config work handler from accessing the device */
1724 mutex_lock(&vi->config_lock);
1725 vi->config_enable = false;
1726 mutex_unlock(&vi->config_lock);
1728 netif_device_detach(vi->dev);
1729 cancel_delayed_work_sync(&vi->refill);
1731 if (netif_running(vi->dev))
1732 for (i = 0; i < vi->max_queue_pairs; i++) {
1733 napi_disable(&vi->rq[i].napi);
1734 netif_napi_del(&vi->rq[i].napi);
1737 remove_vq_common(vi);
1739 flush_work(&vi->config_work);
1744 static int virtnet_restore(struct virtio_device *vdev)
1746 struct virtnet_info *vi = vdev->priv;
1753 if (netif_running(vi->dev))
1754 for (i = 0; i < vi->max_queue_pairs; i++)
1755 virtnet_napi_enable(&vi->rq[i]);
1757 netif_device_attach(vi->dev);
1759 for (i = 0; i < vi->curr_queue_pairs; i++)
1760 if (!try_fill_recv(&vi->rq[i], GFP_KERNEL))
1761 schedule_delayed_work(&vi->refill, 0);
1763 mutex_lock(&vi->config_lock);
1764 vi->config_enable = true;
1765 mutex_unlock(&vi->config_lock);
1768 virtnet_set_queues(vi, vi->curr_queue_pairs);
1771 err = register_hotcpu_notifier(&vi->nb);
1779 static struct virtio_device_id id_table[] = {
1780 { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
1784 static unsigned int features[] = {
1785 VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
1786 VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
1787 VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
1788 VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
1789 VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO,
1790 VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ,
1791 VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN,
1792 VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ,
1793 VIRTIO_NET_F_CTRL_MAC_ADDR,
1794 VIRTIO_F_ANY_LAYOUT,
1797 static struct virtio_driver virtio_net_driver = {
1798 .feature_table = features,
1799 .feature_table_size = ARRAY_SIZE(features),
1800 .driver.name = KBUILD_MODNAME,
1801 .driver.owner = THIS_MODULE,
1802 .id_table = id_table,
1803 .probe = virtnet_probe,
1804 .remove = virtnet_remove,
1805 .config_changed = virtnet_config_changed,
1806 #ifdef CONFIG_PM_SLEEP
1807 .freeze = virtnet_freeze,
1808 .restore = virtnet_restore,
1812 module_virtio_driver(virtio_net_driver);
1814 MODULE_DEVICE_TABLE(virtio, id_table);
1815 MODULE_DESCRIPTION("Virtio network driver");
1816 MODULE_LICENSE("GPL");