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 = 128;
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_SEND_COMMAND_SG_MAX 2
43 #define VIRTNET_DRIVER_VERSION "1.0.0"
45 struct virtnet_stats {
46 struct u64_stats_sync tx_syncp;
47 struct u64_stats_sync rx_syncp;
55 /* Internal representation of a send virtqueue */
57 /* Virtqueue associated with this send _queue */
60 /* TX: fragments + linear part + virtio header */
61 struct scatterlist sg[MAX_SKB_FRAGS + 2];
63 /* Name of the send queue: output.$index */
67 /* Internal representation of a receive virtqueue */
68 struct receive_queue {
69 /* Virtqueue associated with this receive_queue */
72 struct napi_struct napi;
74 /* Number of input buffers, and max we've ever had. */
75 unsigned int num, max;
77 /* Chain pages by the private ptr. */
80 /* RX: fragments + linear part + virtio header */
81 struct scatterlist sg[MAX_SKB_FRAGS + 2];
83 /* Name of this receive queue: input.$index */
88 struct virtio_device *vdev;
89 struct virtqueue *cvq;
90 struct net_device *dev;
91 struct send_queue *sq;
92 struct receive_queue *rq;
95 /* Max # of queue pairs supported by the device */
98 /* # of queue pairs currently used by the driver */
101 /* I like... big packets and I cannot lie! */
104 /* Host will merge rx buffers for big packets (shake it! shake it!) */
105 bool mergeable_rx_bufs;
107 /* Has control virtqueue */
110 /* enable config space updates */
113 /* Active statistics */
114 struct virtnet_stats __percpu *stats;
116 /* Work struct for refilling if we run low on memory. */
117 struct delayed_work refill;
119 /* Work struct for config space updates */
120 struct work_struct config_work;
122 /* Lock for config space updates */
123 struct mutex config_lock;
125 /* Does the affinity hint is set for virtqueues? */
126 bool affinity_hint_set;
128 /* Per-cpu variable to show the mapping from CPU to virtqueue */
129 int __percpu *vq_index;
131 /* CPU hot plug notifier */
132 struct notifier_block nb;
135 struct skb_vnet_hdr {
137 struct virtio_net_hdr hdr;
138 struct virtio_net_hdr_mrg_rxbuf mhdr;
142 struct padded_vnet_hdr {
143 struct virtio_net_hdr hdr;
145 * virtio_net_hdr should be in a separated sg buffer because of a
146 * QEMU bug, and data sg buffer shares same page with this header sg.
147 * This padding makes next sg 16 byte aligned after virtio_net_hdr.
152 /* Converting between virtqueue no. and kernel tx/rx queue no.
153 * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq
155 static int vq2txq(struct virtqueue *vq)
157 return (virtqueue_get_queue_index(vq) - 1) / 2;
160 static int txq2vq(int txq)
165 static int vq2rxq(struct virtqueue *vq)
167 return virtqueue_get_queue_index(vq) / 2;
170 static int rxq2vq(int rxq)
175 static inline struct skb_vnet_hdr *skb_vnet_hdr(struct sk_buff *skb)
177 return (struct skb_vnet_hdr *)skb->cb;
181 * private is used to chain pages for big packets, put the whole
182 * most recent used list in the beginning for reuse
184 static void give_pages(struct receive_queue *rq, struct page *page)
188 /* Find end of list, sew whole thing into vi->rq.pages. */
189 for (end = page; end->private; end = (struct page *)end->private);
190 end->private = (unsigned long)rq->pages;
194 static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask)
196 struct page *p = rq->pages;
199 rq->pages = (struct page *)p->private;
200 /* clear private here, it is used to chain pages */
203 p = alloc_page(gfp_mask);
207 static void skb_xmit_done(struct virtqueue *vq)
209 struct virtnet_info *vi = vq->vdev->priv;
211 /* Suppress further interrupts. */
212 virtqueue_disable_cb(vq);
214 /* We were probably waiting for more output buffers. */
215 netif_wake_subqueue(vi->dev, vq2txq(vq));
218 static void set_skb_frag(struct sk_buff *skb, struct page *page,
219 unsigned int offset, unsigned int *len)
221 int size = min((unsigned)PAGE_SIZE - offset, *len);
222 int i = skb_shinfo(skb)->nr_frags;
224 __skb_fill_page_desc(skb, i, page, offset, size);
226 skb->data_len += size;
228 skb->truesize += PAGE_SIZE;
229 skb_shinfo(skb)->nr_frags++;
230 skb_shinfo(skb)->gso_type |= SKB_GSO_SHARED_FRAG;
234 /* Called from bottom half context */
235 static struct sk_buff *page_to_skb(struct receive_queue *rq,
236 struct page *page, unsigned int len)
238 struct virtnet_info *vi = rq->vq->vdev->priv;
240 struct skb_vnet_hdr *hdr;
241 unsigned int copy, hdr_len, offset;
244 p = page_address(page);
246 /* copy small packet so we can reuse these pages for small data */
247 skb = netdev_alloc_skb_ip_align(vi->dev, GOOD_COPY_LEN);
251 hdr = skb_vnet_hdr(skb);
253 if (vi->mergeable_rx_bufs) {
254 hdr_len = sizeof hdr->mhdr;
257 hdr_len = sizeof hdr->hdr;
258 offset = sizeof(struct padded_vnet_hdr);
261 memcpy(hdr, p, hdr_len);
267 if (copy > skb_tailroom(skb))
268 copy = skb_tailroom(skb);
269 memcpy(skb_put(skb, copy), p, copy);
275 * Verify that we can indeed put this data into a skb.
276 * This is here to handle cases when the device erroneously
277 * tries to receive more than is possible. This is usually
278 * the case of a broken device.
280 if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) {
281 net_dbg_ratelimited("%s: too much data\n", skb->dev->name);
287 set_skb_frag(skb, page, offset, &len);
288 page = (struct page *)page->private;
293 give_pages(rq, page);
298 static int receive_mergeable(struct receive_queue *rq, struct sk_buff *skb)
300 struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
304 num_buf = hdr->mhdr.num_buffers;
306 i = skb_shinfo(skb)->nr_frags;
307 if (i >= MAX_SKB_FRAGS) {
308 pr_debug("%s: packet too long\n", skb->dev->name);
309 skb->dev->stats.rx_length_errors++;
312 page = virtqueue_get_buf(rq->vq, &len);
314 pr_debug("%s: rx error: %d buffers missing\n",
315 skb->dev->name, hdr->mhdr.num_buffers);
316 skb->dev->stats.rx_length_errors++;
323 set_skb_frag(skb, page, 0, &len);
330 static void receive_buf(struct receive_queue *rq, void *buf, unsigned int len)
332 struct virtnet_info *vi = rq->vq->vdev->priv;
333 struct net_device *dev = vi->dev;
334 struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
337 struct skb_vnet_hdr *hdr;
339 if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
340 pr_debug("%s: short packet %i\n", dev->name, len);
341 dev->stats.rx_length_errors++;
342 if (vi->mergeable_rx_bufs || vi->big_packets)
349 if (!vi->mergeable_rx_bufs && !vi->big_packets) {
351 len -= sizeof(struct virtio_net_hdr);
355 skb = page_to_skb(rq, page, len);
356 if (unlikely(!skb)) {
357 dev->stats.rx_dropped++;
358 give_pages(rq, page);
361 if (vi->mergeable_rx_bufs)
362 if (receive_mergeable(rq, skb)) {
368 hdr = skb_vnet_hdr(skb);
370 u64_stats_update_begin(&stats->rx_syncp);
371 stats->rx_bytes += skb->len;
373 u64_stats_update_end(&stats->rx_syncp);
375 if (hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
376 pr_debug("Needs csum!\n");
377 if (!skb_partial_csum_set(skb,
379 hdr->hdr.csum_offset))
381 } else if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) {
382 skb->ip_summed = CHECKSUM_UNNECESSARY;
385 skb->protocol = eth_type_trans(skb, dev);
386 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
387 ntohs(skb->protocol), skb->len, skb->pkt_type);
389 if (hdr->hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
390 unsigned short gso_type = 0;
393 switch (hdr->hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
394 case VIRTIO_NET_HDR_GSO_TCPV4:
395 gso_type = SKB_GSO_TCPV4;
397 case VIRTIO_NET_HDR_GSO_UDP:
398 gso_type = SKB_GSO_UDP;
400 case VIRTIO_NET_HDR_GSO_TCPV6:
401 gso_type = SKB_GSO_TCPV6;
404 net_warn_ratelimited("%s: bad gso type %u.\n",
405 dev->name, hdr->hdr.gso_type);
409 if (hdr->hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
410 gso_type |= SKB_GSO_TCP_ECN;
412 skb_shinfo(skb)->gso_size = hdr->hdr.gso_size;
413 if (skb_shinfo(skb)->gso_size == 0) {
414 net_warn_ratelimited("%s: zero gso size.\n", dev->name);
418 skb_shinfo(skb)->gso_type |= gso_type;
419 /* Header must be checked, and gso_segs computed. */
420 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
421 skb_shinfo(skb)->gso_segs = 0;
424 netif_receive_skb(skb);
428 dev->stats.rx_frame_errors++;
432 static int add_recvbuf_small(struct receive_queue *rq, gfp_t gfp)
434 struct virtnet_info *vi = rq->vq->vdev->priv;
436 struct skb_vnet_hdr *hdr;
439 skb = __netdev_alloc_skb_ip_align(vi->dev, MAX_PACKET_LEN, gfp);
443 skb_put(skb, MAX_PACKET_LEN);
445 hdr = skb_vnet_hdr(skb);
446 sg_set_buf(rq->sg, &hdr->hdr, sizeof hdr->hdr);
448 skb_to_sgvec(skb, rq->sg + 1, 0, skb->len);
450 err = virtqueue_add_buf(rq->vq, rq->sg, 0, 2, skb, gfp);
457 static int add_recvbuf_big(struct receive_queue *rq, gfp_t gfp)
459 struct page *first, *list = NULL;
463 /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */
464 for (i = MAX_SKB_FRAGS + 1; i > 1; --i) {
465 first = get_a_page(rq, gfp);
468 give_pages(rq, list);
471 sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE);
473 /* chain new page in list head to match sg */
474 first->private = (unsigned long)list;
478 first = get_a_page(rq, gfp);
480 give_pages(rq, list);
483 p = page_address(first);
485 /* rq->sg[0], rq->sg[1] share the same page */
486 /* a separated rq->sg[0] for virtio_net_hdr only due to QEMU bug */
487 sg_set_buf(&rq->sg[0], p, sizeof(struct virtio_net_hdr));
489 /* rq->sg[1] for data packet, from offset */
490 offset = sizeof(struct padded_vnet_hdr);
491 sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset);
493 /* chain first in list head */
494 first->private = (unsigned long)list;
495 err = virtqueue_add_buf(rq->vq, rq->sg, 0, MAX_SKB_FRAGS + 2,
498 give_pages(rq, first);
503 static int add_recvbuf_mergeable(struct receive_queue *rq, gfp_t gfp)
508 page = get_a_page(rq, gfp);
512 sg_init_one(rq->sg, page_address(page), PAGE_SIZE);
514 err = virtqueue_add_buf(rq->vq, rq->sg, 0, 1, page, gfp);
516 give_pages(rq, page);
522 * Returns false if we couldn't fill entirely (OOM).
524 * Normally run in the receive path, but can also be run from ndo_open
525 * before we're receiving packets, or from refill_work which is
526 * careful to disable receiving (using napi_disable).
528 static bool try_fill_recv(struct receive_queue *rq, gfp_t gfp)
530 struct virtnet_info *vi = rq->vq->vdev->priv;
535 if (vi->mergeable_rx_bufs)
536 err = add_recvbuf_mergeable(rq, gfp);
537 else if (vi->big_packets)
538 err = add_recvbuf_big(rq, gfp);
540 err = add_recvbuf_small(rq, gfp);
542 oom = err == -ENOMEM;
546 } while (rq->vq->num_free);
547 if (unlikely(rq->num > rq->max))
549 virtqueue_kick(rq->vq);
553 static void skb_recv_done(struct virtqueue *rvq)
555 struct virtnet_info *vi = rvq->vdev->priv;
556 struct receive_queue *rq = &vi->rq[vq2rxq(rvq)];
558 /* Schedule NAPI, Suppress further interrupts if successful. */
559 if (napi_schedule_prep(&rq->napi)) {
560 virtqueue_disable_cb(rvq);
561 __napi_schedule(&rq->napi);
565 static void virtnet_napi_enable(struct receive_queue *rq)
567 napi_enable(&rq->napi);
569 /* If all buffers were filled by other side before we napi_enabled, we
570 * won't get another interrupt, so process any outstanding packets
571 * now. virtnet_poll wants re-enable the queue, so we disable here.
572 * We synchronize against interrupts via NAPI_STATE_SCHED */
573 if (napi_schedule_prep(&rq->napi)) {
574 virtqueue_disable_cb(rq->vq);
576 __napi_schedule(&rq->napi);
581 static void refill_work(struct work_struct *work)
583 struct virtnet_info *vi =
584 container_of(work, struct virtnet_info, refill.work);
588 for (i = 0; i < vi->max_queue_pairs; i++) {
589 struct receive_queue *rq = &vi->rq[i];
591 napi_disable(&rq->napi);
592 still_empty = !try_fill_recv(rq, GFP_KERNEL);
593 virtnet_napi_enable(rq);
595 /* In theory, this can happen: if we don't get any buffers in
596 * we will *never* try to fill again.
599 schedule_delayed_work(&vi->refill, HZ/2);
603 static int virtnet_poll(struct napi_struct *napi, int budget)
605 struct receive_queue *rq =
606 container_of(napi, struct receive_queue, napi);
607 struct virtnet_info *vi = rq->vq->vdev->priv;
609 unsigned int len, received = 0;
612 while (received < budget &&
613 (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) {
614 receive_buf(rq, buf, len);
619 if (rq->num < rq->max / 2) {
620 if (!try_fill_recv(rq, GFP_ATOMIC))
621 schedule_delayed_work(&vi->refill, 0);
624 /* Out of packets? */
625 if (received < budget) {
627 if (unlikely(!virtqueue_enable_cb(rq->vq)) &&
628 napi_schedule_prep(napi)) {
629 virtqueue_disable_cb(rq->vq);
630 __napi_schedule(napi);
638 static int virtnet_open(struct net_device *dev)
640 struct virtnet_info *vi = netdev_priv(dev);
643 for (i = 0; i < vi->max_queue_pairs; i++) {
644 /* Make sure we have some buffers: if oom use wq. */
645 if (!try_fill_recv(&vi->rq[i], GFP_KERNEL))
646 schedule_delayed_work(&vi->refill, 0);
647 virtnet_napi_enable(&vi->rq[i]);
653 static void free_old_xmit_skbs(struct send_queue *sq)
657 struct virtnet_info *vi = sq->vq->vdev->priv;
658 struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
660 while ((skb = virtqueue_get_buf(sq->vq, &len)) != NULL) {
661 pr_debug("Sent skb %p\n", skb);
663 u64_stats_update_begin(&stats->tx_syncp);
664 stats->tx_bytes += skb->len;
666 u64_stats_update_end(&stats->tx_syncp);
668 dev_kfree_skb_any(skb);
672 static int xmit_skb(struct send_queue *sq, struct sk_buff *skb)
674 struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
675 const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
676 struct virtnet_info *vi = sq->vq->vdev->priv;
679 pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
681 if (skb->ip_summed == CHECKSUM_PARTIAL) {
682 hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
683 hdr->hdr.csum_start = skb_checksum_start_offset(skb);
684 hdr->hdr.csum_offset = skb->csum_offset;
687 hdr->hdr.csum_offset = hdr->hdr.csum_start = 0;
690 if (skb_is_gso(skb)) {
691 hdr->hdr.hdr_len = skb_headlen(skb);
692 hdr->hdr.gso_size = skb_shinfo(skb)->gso_size;
693 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
694 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
695 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
696 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
697 else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
698 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
701 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
702 hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
704 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
705 hdr->hdr.gso_size = hdr->hdr.hdr_len = 0;
708 hdr->mhdr.num_buffers = 0;
710 /* Encode metadata header at front. */
711 if (vi->mergeable_rx_bufs)
712 sg_set_buf(sq->sg, &hdr->mhdr, sizeof hdr->mhdr);
714 sg_set_buf(sq->sg, &hdr->hdr, sizeof hdr->hdr);
716 num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len) + 1;
717 return virtqueue_add_buf(sq->vq, sq->sg, num_sg,
721 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
723 struct virtnet_info *vi = netdev_priv(dev);
724 int qnum = skb_get_queue_mapping(skb);
725 struct send_queue *sq = &vi->sq[qnum];
728 /* Free up any pending old buffers before queueing new ones. */
729 free_old_xmit_skbs(sq);
731 /* Try to transmit */
732 err = xmit_skb(sq, skb);
734 /* This should not happen! */
736 dev->stats.tx_fifo_errors++;
739 "Unexpected TXQ (%d) queue failure: %d\n", qnum, err);
740 dev->stats.tx_dropped++;
744 virtqueue_kick(sq->vq);
746 /* Don't wait up for transmitted skbs to be freed. */
750 /* Apparently nice girls don't return TX_BUSY; stop the queue
751 * before it gets out of hand. Naturally, this wastes entries. */
752 if (sq->vq->num_free < 2+MAX_SKB_FRAGS) {
753 netif_stop_subqueue(dev, qnum);
754 if (unlikely(!virtqueue_enable_cb_delayed(sq->vq))) {
755 /* More just got used, free them then recheck. */
756 free_old_xmit_skbs(sq);
757 if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) {
758 netif_start_subqueue(dev, qnum);
759 virtqueue_disable_cb(sq->vq);
768 * Send command via the control virtqueue and check status. Commands
769 * supported by the hypervisor, as indicated by feature bits, should
770 * never fail unless improperly formated.
772 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
773 struct scatterlist *data, int out, int in)
775 struct scatterlist *s, sg[VIRTNET_SEND_COMMAND_SG_MAX + 2];
776 struct virtio_net_ctrl_hdr ctrl;
777 virtio_net_ctrl_ack status = ~0;
781 /* Caller should know better */
782 BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ||
783 (out + in > VIRTNET_SEND_COMMAND_SG_MAX));
785 out++; /* Add header */
786 in++; /* Add return status */
791 sg_init_table(sg, out + in);
793 sg_set_buf(&sg[0], &ctrl, sizeof(ctrl));
794 for_each_sg(data, s, out + in - 2, i)
795 sg_set_buf(&sg[i + 1], sg_virt(s), s->length);
796 sg_set_buf(&sg[out + in - 1], &status, sizeof(status));
798 BUG_ON(virtqueue_add_buf(vi->cvq, sg, out, in, vi, GFP_ATOMIC) < 0);
800 virtqueue_kick(vi->cvq);
802 /* Spin for a response, the kick causes an ioport write, trapping
803 * into the hypervisor, so the request should be handled immediately.
805 while (!virtqueue_get_buf(vi->cvq, &tmp))
808 return status == VIRTIO_NET_OK;
811 static int virtnet_set_mac_address(struct net_device *dev, void *p)
813 struct virtnet_info *vi = netdev_priv(dev);
814 struct virtio_device *vdev = vi->vdev;
816 struct sockaddr *addr = p;
817 struct scatterlist sg;
819 ret = eth_prepare_mac_addr_change(dev, p);
823 if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
824 sg_init_one(&sg, addr->sa_data, dev->addr_len);
825 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
826 VIRTIO_NET_CTRL_MAC_ADDR_SET,
829 "Failed to set mac address by vq command.\n");
832 } else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
833 vdev->config->set(vdev, offsetof(struct virtio_net_config, mac),
834 addr->sa_data, dev->addr_len);
837 eth_commit_mac_addr_change(dev, p);
842 static struct rtnl_link_stats64 *virtnet_stats(struct net_device *dev,
843 struct rtnl_link_stats64 *tot)
845 struct virtnet_info *vi = netdev_priv(dev);
849 for_each_possible_cpu(cpu) {
850 struct virtnet_stats *stats = per_cpu_ptr(vi->stats, cpu);
851 u64 tpackets, tbytes, rpackets, rbytes;
854 start = u64_stats_fetch_begin_bh(&stats->tx_syncp);
855 tpackets = stats->tx_packets;
856 tbytes = stats->tx_bytes;
857 } while (u64_stats_fetch_retry_bh(&stats->tx_syncp, start));
860 start = u64_stats_fetch_begin_bh(&stats->rx_syncp);
861 rpackets = stats->rx_packets;
862 rbytes = stats->rx_bytes;
863 } while (u64_stats_fetch_retry_bh(&stats->rx_syncp, start));
865 tot->rx_packets += rpackets;
866 tot->tx_packets += tpackets;
867 tot->rx_bytes += rbytes;
868 tot->tx_bytes += tbytes;
871 tot->tx_dropped = dev->stats.tx_dropped;
872 tot->tx_fifo_errors = dev->stats.tx_fifo_errors;
873 tot->rx_dropped = dev->stats.rx_dropped;
874 tot->rx_length_errors = dev->stats.rx_length_errors;
875 tot->rx_frame_errors = dev->stats.rx_frame_errors;
880 #ifdef CONFIG_NET_POLL_CONTROLLER
881 static void virtnet_netpoll(struct net_device *dev)
883 struct virtnet_info *vi = netdev_priv(dev);
886 for (i = 0; i < vi->curr_queue_pairs; i++)
887 napi_schedule(&vi->rq[i].napi);
891 static void virtnet_ack_link_announce(struct virtnet_info *vi)
894 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE,
895 VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL,
897 dev_warn(&vi->dev->dev, "Failed to ack link announce.\n");
901 static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs)
903 struct scatterlist sg;
904 struct virtio_net_ctrl_mq s;
905 struct net_device *dev = vi->dev;
907 if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ))
910 s.virtqueue_pairs = queue_pairs;
911 sg_init_one(&sg, &s, sizeof(s));
913 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ,
914 VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg, 1, 0)){
915 dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n",
919 vi->curr_queue_pairs = queue_pairs;
924 static int virtnet_close(struct net_device *dev)
926 struct virtnet_info *vi = netdev_priv(dev);
929 /* Make sure refill_work doesn't re-enable napi! */
930 cancel_delayed_work_sync(&vi->refill);
932 for (i = 0; i < vi->max_queue_pairs; i++)
933 napi_disable(&vi->rq[i].napi);
938 static void virtnet_set_rx_mode(struct net_device *dev)
940 struct virtnet_info *vi = netdev_priv(dev);
941 struct scatterlist sg[2];
942 u8 promisc, allmulti;
943 struct virtio_net_ctrl_mac *mac_data;
944 struct netdev_hw_addr *ha;
950 /* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */
951 if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
954 promisc = ((dev->flags & IFF_PROMISC) != 0);
955 allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
957 sg_init_one(sg, &promisc, sizeof(promisc));
959 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
960 VIRTIO_NET_CTRL_RX_PROMISC,
962 dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
963 promisc ? "en" : "dis");
965 sg_init_one(sg, &allmulti, sizeof(allmulti));
967 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
968 VIRTIO_NET_CTRL_RX_ALLMULTI,
970 dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
971 allmulti ? "en" : "dis");
973 uc_count = netdev_uc_count(dev);
974 mc_count = netdev_mc_count(dev);
975 /* MAC filter - use one buffer for both lists */
976 buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
977 (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
980 dev_warn(&dev->dev, "No memory for MAC address buffer\n");
984 sg_init_table(sg, 2);
986 /* Store the unicast list and count in the front of the buffer */
987 mac_data->entries = uc_count;
989 netdev_for_each_uc_addr(ha, dev)
990 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
992 sg_set_buf(&sg[0], mac_data,
993 sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
995 /* multicast list and count fill the end */
996 mac_data = (void *)&mac_data->macs[uc_count][0];
998 mac_data->entries = mc_count;
1000 netdev_for_each_mc_addr(ha, dev)
1001 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
1003 sg_set_buf(&sg[1], mac_data,
1004 sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
1006 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
1007 VIRTIO_NET_CTRL_MAC_TABLE_SET,
1009 dev_warn(&dev->dev, "Failed to set MAC fitler table.\n");
1014 static int virtnet_vlan_rx_add_vid(struct net_device *dev, u16 vid)
1016 struct virtnet_info *vi = netdev_priv(dev);
1017 struct scatterlist sg;
1019 sg_init_one(&sg, &vid, sizeof(vid));
1021 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1022 VIRTIO_NET_CTRL_VLAN_ADD, &sg, 1, 0))
1023 dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
1027 static int virtnet_vlan_rx_kill_vid(struct net_device *dev, u16 vid)
1029 struct virtnet_info *vi = netdev_priv(dev);
1030 struct scatterlist sg;
1032 sg_init_one(&sg, &vid, sizeof(vid));
1034 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1035 VIRTIO_NET_CTRL_VLAN_DEL, &sg, 1, 0))
1036 dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
1040 static void virtnet_clean_affinity(struct virtnet_info *vi, long hcpu)
1045 if (vi->affinity_hint_set) {
1046 for (i = 0; i < vi->max_queue_pairs; i++) {
1047 virtqueue_set_affinity(vi->rq[i].vq, -1);
1048 virtqueue_set_affinity(vi->sq[i].vq, -1);
1051 vi->affinity_hint_set = false;
1055 for_each_online_cpu(cpu) {
1057 *per_cpu_ptr(vi->vq_index, cpu) = -1;
1059 *per_cpu_ptr(vi->vq_index, cpu) =
1060 ++i % vi->curr_queue_pairs;
1065 static void virtnet_set_affinity(struct virtnet_info *vi)
1070 /* In multiqueue mode, when the number of cpu is equal to the number of
1071 * queue pairs, we let the queue pairs to be private to one cpu by
1072 * setting the affinity hint to eliminate the contention.
1074 if (vi->curr_queue_pairs == 1 ||
1075 vi->max_queue_pairs != num_online_cpus()) {
1076 virtnet_clean_affinity(vi, -1);
1081 for_each_online_cpu(cpu) {
1082 virtqueue_set_affinity(vi->rq[i].vq, cpu);
1083 virtqueue_set_affinity(vi->sq[i].vq, cpu);
1084 *per_cpu_ptr(vi->vq_index, cpu) = i;
1088 vi->affinity_hint_set = true;
1091 static int virtnet_cpu_callback(struct notifier_block *nfb,
1092 unsigned long action, void *hcpu)
1094 struct virtnet_info *vi = container_of(nfb, struct virtnet_info, nb);
1096 switch(action & ~CPU_TASKS_FROZEN) {
1098 case CPU_DOWN_FAILED:
1100 virtnet_set_affinity(vi);
1102 case CPU_DOWN_PREPARE:
1103 virtnet_clean_affinity(vi, (long)hcpu);
1111 static void virtnet_get_ringparam(struct net_device *dev,
1112 struct ethtool_ringparam *ring)
1114 struct virtnet_info *vi = netdev_priv(dev);
1116 ring->rx_max_pending = virtqueue_get_vring_size(vi->rq[0].vq);
1117 ring->tx_max_pending = virtqueue_get_vring_size(vi->sq[0].vq);
1118 ring->rx_pending = ring->rx_max_pending;
1119 ring->tx_pending = ring->tx_max_pending;
1123 static void virtnet_get_drvinfo(struct net_device *dev,
1124 struct ethtool_drvinfo *info)
1126 struct virtnet_info *vi = netdev_priv(dev);
1127 struct virtio_device *vdev = vi->vdev;
1129 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
1130 strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version));
1131 strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info));
1135 /* TODO: Eliminate OOO packets during switching */
1136 static int virtnet_set_channels(struct net_device *dev,
1137 struct ethtool_channels *channels)
1139 struct virtnet_info *vi = netdev_priv(dev);
1140 u16 queue_pairs = channels->combined_count;
1143 /* We don't support separate rx/tx channels.
1144 * We don't allow setting 'other' channels.
1146 if (channels->rx_count || channels->tx_count || channels->other_count)
1149 if (queue_pairs > vi->max_queue_pairs)
1153 err = virtnet_set_queues(vi, queue_pairs);
1155 netif_set_real_num_tx_queues(dev, queue_pairs);
1156 netif_set_real_num_rx_queues(dev, queue_pairs);
1158 virtnet_set_affinity(vi);
1165 static void virtnet_get_channels(struct net_device *dev,
1166 struct ethtool_channels *channels)
1168 struct virtnet_info *vi = netdev_priv(dev);
1170 channels->combined_count = vi->curr_queue_pairs;
1171 channels->max_combined = vi->max_queue_pairs;
1172 channels->max_other = 0;
1173 channels->rx_count = 0;
1174 channels->tx_count = 0;
1175 channels->other_count = 0;
1178 static const struct ethtool_ops virtnet_ethtool_ops = {
1179 .get_drvinfo = virtnet_get_drvinfo,
1180 .get_link = ethtool_op_get_link,
1181 .get_ringparam = virtnet_get_ringparam,
1182 .set_channels = virtnet_set_channels,
1183 .get_channels = virtnet_get_channels,
1187 #define MAX_MTU 65535
1189 static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
1191 if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
1197 /* To avoid contending a lock hold by a vcpu who would exit to host, select the
1198 * txq based on the processor id.
1200 static u16 virtnet_select_queue(struct net_device *dev, struct sk_buff *skb)
1203 struct virtnet_info *vi = netdev_priv(dev);
1205 if (skb_rx_queue_recorded(skb)) {
1206 txq = skb_get_rx_queue(skb);
1208 txq = *__this_cpu_ptr(vi->vq_index);
1213 while (unlikely(txq >= dev->real_num_tx_queues))
1214 txq -= dev->real_num_tx_queues;
1219 static const struct net_device_ops virtnet_netdev = {
1220 .ndo_open = virtnet_open,
1221 .ndo_stop = virtnet_close,
1222 .ndo_start_xmit = start_xmit,
1223 .ndo_validate_addr = eth_validate_addr,
1224 .ndo_set_mac_address = virtnet_set_mac_address,
1225 .ndo_set_rx_mode = virtnet_set_rx_mode,
1226 .ndo_change_mtu = virtnet_change_mtu,
1227 .ndo_get_stats64 = virtnet_stats,
1228 .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
1229 .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
1230 .ndo_select_queue = virtnet_select_queue,
1231 #ifdef CONFIG_NET_POLL_CONTROLLER
1232 .ndo_poll_controller = virtnet_netpoll,
1236 static void virtnet_config_changed_work(struct work_struct *work)
1238 struct virtnet_info *vi =
1239 container_of(work, struct virtnet_info, config_work);
1242 mutex_lock(&vi->config_lock);
1243 if (!vi->config_enable)
1246 if (virtio_config_val(vi->vdev, VIRTIO_NET_F_STATUS,
1247 offsetof(struct virtio_net_config, status),
1251 if (v & VIRTIO_NET_S_ANNOUNCE) {
1252 netdev_notify_peers(vi->dev);
1253 virtnet_ack_link_announce(vi);
1256 /* Ignore unknown (future) status bits */
1257 v &= VIRTIO_NET_S_LINK_UP;
1259 if (vi->status == v)
1264 if (vi->status & VIRTIO_NET_S_LINK_UP) {
1265 netif_carrier_on(vi->dev);
1266 netif_tx_wake_all_queues(vi->dev);
1268 netif_carrier_off(vi->dev);
1269 netif_tx_stop_all_queues(vi->dev);
1272 mutex_unlock(&vi->config_lock);
1275 static void virtnet_config_changed(struct virtio_device *vdev)
1277 struct virtnet_info *vi = vdev->priv;
1279 schedule_work(&vi->config_work);
1282 static void virtnet_free_queues(struct virtnet_info *vi)
1288 static void free_receive_bufs(struct virtnet_info *vi)
1292 for (i = 0; i < vi->max_queue_pairs; i++) {
1293 while (vi->rq[i].pages)
1294 __free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0);
1298 static void free_unused_bufs(struct virtnet_info *vi)
1303 for (i = 0; i < vi->max_queue_pairs; i++) {
1304 struct virtqueue *vq = vi->sq[i].vq;
1305 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL)
1309 for (i = 0; i < vi->max_queue_pairs; i++) {
1310 struct virtqueue *vq = vi->rq[i].vq;
1312 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) {
1313 if (vi->mergeable_rx_bufs || vi->big_packets)
1314 give_pages(&vi->rq[i], buf);
1319 BUG_ON(vi->rq[i].num != 0);
1323 static void virtnet_del_vqs(struct virtnet_info *vi)
1325 struct virtio_device *vdev = vi->vdev;
1327 virtnet_clean_affinity(vi, -1);
1329 vdev->config->del_vqs(vdev);
1331 virtnet_free_queues(vi);
1334 static int virtnet_find_vqs(struct virtnet_info *vi)
1336 vq_callback_t **callbacks;
1337 struct virtqueue **vqs;
1342 /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
1343 * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
1344 * possible control vq.
1346 total_vqs = vi->max_queue_pairs * 2 +
1347 virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ);
1349 /* Allocate space for find_vqs parameters */
1350 vqs = kzalloc(total_vqs * sizeof(*vqs), GFP_KERNEL);
1353 callbacks = kmalloc(total_vqs * sizeof(*callbacks), GFP_KERNEL);
1356 names = kmalloc(total_vqs * sizeof(*names), GFP_KERNEL);
1360 /* Parameters for control virtqueue, if any */
1362 callbacks[total_vqs - 1] = NULL;
1363 names[total_vqs - 1] = "control";
1366 /* Allocate/initialize parameters for send/receive virtqueues */
1367 for (i = 0; i < vi->max_queue_pairs; i++) {
1368 callbacks[rxq2vq(i)] = skb_recv_done;
1369 callbacks[txq2vq(i)] = skb_xmit_done;
1370 sprintf(vi->rq[i].name, "input.%d", i);
1371 sprintf(vi->sq[i].name, "output.%d", i);
1372 names[rxq2vq(i)] = vi->rq[i].name;
1373 names[txq2vq(i)] = vi->sq[i].name;
1376 ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs, callbacks,
1382 vi->cvq = vqs[total_vqs - 1];
1383 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
1384 vi->dev->features |= NETIF_F_HW_VLAN_FILTER;
1387 for (i = 0; i < vi->max_queue_pairs; i++) {
1388 vi->rq[i].vq = vqs[rxq2vq(i)];
1389 vi->sq[i].vq = vqs[txq2vq(i)];
1408 static int virtnet_alloc_queues(struct virtnet_info *vi)
1412 vi->sq = kzalloc(sizeof(*vi->sq) * vi->max_queue_pairs, GFP_KERNEL);
1415 vi->rq = kzalloc(sizeof(*vi->rq) * vi->max_queue_pairs, GFP_KERNEL);
1419 INIT_DELAYED_WORK(&vi->refill, refill_work);
1420 for (i = 0; i < vi->max_queue_pairs; i++) {
1421 vi->rq[i].pages = NULL;
1422 netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll,
1425 sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg));
1426 sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg));
1437 static int init_vqs(struct virtnet_info *vi)
1441 /* Allocate send & receive queues */
1442 ret = virtnet_alloc_queues(vi);
1446 ret = virtnet_find_vqs(vi);
1451 virtnet_set_affinity(vi);
1457 virtnet_free_queues(vi);
1462 static int virtnet_probe(struct virtio_device *vdev)
1465 struct net_device *dev;
1466 struct virtnet_info *vi;
1467 u16 max_queue_pairs;
1469 /* Find if host supports multiqueue virtio_net device */
1470 err = virtio_config_val(vdev, VIRTIO_NET_F_MQ,
1471 offsetof(struct virtio_net_config,
1472 max_virtqueue_pairs), &max_queue_pairs);
1474 /* We need at least 2 queue's */
1475 if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
1476 max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
1477 !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
1478 max_queue_pairs = 1;
1480 /* Allocate ourselves a network device with room for our info */
1481 dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs);
1485 /* Set up network device as normal. */
1486 dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE;
1487 dev->netdev_ops = &virtnet_netdev;
1488 dev->features = NETIF_F_HIGHDMA;
1490 SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
1491 SET_NETDEV_DEV(dev, &vdev->dev);
1493 /* Do we support "hardware" checksums? */
1494 if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
1495 /* This opens up the world of extra features. */
1496 dev->hw_features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1498 dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1500 if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
1501 dev->hw_features |= NETIF_F_TSO | NETIF_F_UFO
1502 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
1504 /* Individual feature bits: what can host handle? */
1505 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
1506 dev->hw_features |= NETIF_F_TSO;
1507 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
1508 dev->hw_features |= NETIF_F_TSO6;
1509 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
1510 dev->hw_features |= NETIF_F_TSO_ECN;
1511 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
1512 dev->hw_features |= NETIF_F_UFO;
1515 dev->features |= dev->hw_features & (NETIF_F_ALL_TSO|NETIF_F_UFO);
1516 /* (!csum && gso) case will be fixed by register_netdev() */
1519 /* Configuration may specify what MAC to use. Otherwise random. */
1520 if (virtio_config_val_len(vdev, VIRTIO_NET_F_MAC,
1521 offsetof(struct virtio_net_config, mac),
1522 dev->dev_addr, dev->addr_len) < 0)
1523 eth_hw_addr_random(dev);
1525 /* Set up our device-specific information */
1526 vi = netdev_priv(dev);
1530 vi->stats = alloc_percpu(struct virtnet_stats);
1532 if (vi->stats == NULL)
1535 vi->vq_index = alloc_percpu(int);
1536 if (vi->vq_index == NULL)
1539 mutex_init(&vi->config_lock);
1540 vi->config_enable = true;
1541 INIT_WORK(&vi->config_work, virtnet_config_changed_work);
1543 /* If we can receive ANY GSO packets, we must allocate large ones. */
1544 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
1545 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
1546 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
1547 vi->big_packets = true;
1549 if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
1550 vi->mergeable_rx_bufs = true;
1552 if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
1555 /* Use single tx/rx queue pair as default */
1556 vi->curr_queue_pairs = 1;
1557 vi->max_queue_pairs = max_queue_pairs;
1559 /* Allocate/initialize the rx/tx queues, and invoke find_vqs */
1564 netif_set_real_num_tx_queues(dev, 1);
1565 netif_set_real_num_rx_queues(dev, 1);
1567 err = register_netdev(dev);
1569 pr_debug("virtio_net: registering device failed\n");
1573 /* Last of all, set up some receive buffers. */
1574 for (i = 0; i < vi->max_queue_pairs; i++) {
1575 try_fill_recv(&vi->rq[i], GFP_KERNEL);
1577 /* If we didn't even get one input buffer, we're useless. */
1578 if (vi->rq[i].num == 0) {
1579 free_unused_bufs(vi);
1581 goto free_recv_bufs;
1585 vi->nb.notifier_call = &virtnet_cpu_callback;
1586 err = register_hotcpu_notifier(&vi->nb);
1588 pr_debug("virtio_net: registering cpu notifier failed\n");
1589 goto free_recv_bufs;
1592 /* Assume link up if device can't report link status,
1593 otherwise get link status from config. */
1594 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
1595 netif_carrier_off(dev);
1596 schedule_work(&vi->config_work);
1598 vi->status = VIRTIO_NET_S_LINK_UP;
1599 netif_carrier_on(dev);
1602 pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
1603 dev->name, max_queue_pairs);
1608 free_receive_bufs(vi);
1609 unregister_netdev(dev);
1611 cancel_delayed_work_sync(&vi->refill);
1612 virtnet_del_vqs(vi);
1614 free_percpu(vi->vq_index);
1616 free_percpu(vi->stats);
1622 static void remove_vq_common(struct virtnet_info *vi)
1624 vi->vdev->config->reset(vi->vdev);
1626 /* Free unused buffers in both send and recv, if any. */
1627 free_unused_bufs(vi);
1629 free_receive_bufs(vi);
1631 virtnet_del_vqs(vi);
1634 static void virtnet_remove(struct virtio_device *vdev)
1636 struct virtnet_info *vi = vdev->priv;
1638 unregister_hotcpu_notifier(&vi->nb);
1640 /* Prevent config work handler from accessing the device. */
1641 mutex_lock(&vi->config_lock);
1642 vi->config_enable = false;
1643 mutex_unlock(&vi->config_lock);
1645 unregister_netdev(vi->dev);
1647 remove_vq_common(vi);
1649 flush_work(&vi->config_work);
1651 free_percpu(vi->vq_index);
1652 free_percpu(vi->stats);
1653 free_netdev(vi->dev);
1657 static int virtnet_freeze(struct virtio_device *vdev)
1659 struct virtnet_info *vi = vdev->priv;
1662 /* Prevent config work handler from accessing the device */
1663 mutex_lock(&vi->config_lock);
1664 vi->config_enable = false;
1665 mutex_unlock(&vi->config_lock);
1667 netif_device_detach(vi->dev);
1668 cancel_delayed_work_sync(&vi->refill);
1670 if (netif_running(vi->dev))
1671 for (i = 0; i < vi->max_queue_pairs; i++) {
1672 napi_disable(&vi->rq[i].napi);
1673 netif_napi_del(&vi->rq[i].napi);
1676 remove_vq_common(vi);
1678 flush_work(&vi->config_work);
1683 static int virtnet_restore(struct virtio_device *vdev)
1685 struct virtnet_info *vi = vdev->priv;
1692 if (netif_running(vi->dev))
1693 for (i = 0; i < vi->max_queue_pairs; i++)
1694 virtnet_napi_enable(&vi->rq[i]);
1696 netif_device_attach(vi->dev);
1698 for (i = 0; i < vi->max_queue_pairs; i++)
1699 if (!try_fill_recv(&vi->rq[i], GFP_KERNEL))
1700 schedule_delayed_work(&vi->refill, 0);
1702 mutex_lock(&vi->config_lock);
1703 vi->config_enable = true;
1704 mutex_unlock(&vi->config_lock);
1706 virtnet_set_queues(vi, vi->curr_queue_pairs);
1712 static struct virtio_device_id id_table[] = {
1713 { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
1717 static unsigned int features[] = {
1718 VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
1719 VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
1720 VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
1721 VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
1722 VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO,
1723 VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ,
1724 VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN,
1725 VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ,
1726 VIRTIO_NET_F_CTRL_MAC_ADDR,
1729 static struct virtio_driver virtio_net_driver = {
1730 .feature_table = features,
1731 .feature_table_size = ARRAY_SIZE(features),
1732 .driver.name = KBUILD_MODNAME,
1733 .driver.owner = THIS_MODULE,
1734 .id_table = id_table,
1735 .probe = virtnet_probe,
1736 .remove = virtnet_remove,
1737 .config_changed = virtnet_config_changed,
1739 .freeze = virtnet_freeze,
1740 .restore = virtnet_restore,
1744 static int __init init(void)
1746 return register_virtio_driver(&virtio_net_driver);
1749 static void __exit fini(void)
1751 unregister_virtio_driver(&virtio_net_driver);
1756 MODULE_DEVICE_TABLE(virtio, id_table);
1757 MODULE_DESCRIPTION("Virtio network driver");
1758 MODULE_LICENSE("GPL");