1 /* A network driver using virtio.
3 * Copyright 2007 Rusty Russell <rusty@rustcorp.com.au> IBM Corporation
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, see <http://www.gnu.org/licenses/>.
19 #include <linux/netdevice.h>
20 #include <linux/etherdevice.h>
21 #include <linux/ethtool.h>
22 #include <linux/module.h>
23 #include <linux/virtio.h>
24 #include <linux/virtio_net.h>
25 #include <linux/scatterlist.h>
26 #include <linux/if_vlan.h>
27 #include <linux/slab.h>
28 #include <linux/cpu.h>
29 #include <linux/average.h>
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 GOOD_COPY_LEN 128
42 /* Weight used for the RX packet size EWMA. The average packet size is used to
43 * determine the packet buffer size when refilling RX rings. As the entire RX
44 * ring may be refilled at once, the weight is chosen so that the EWMA will be
45 * insensitive to short-term, transient changes in packet size.
47 #define RECEIVE_AVG_WEIGHT 64
49 /* Minimum alignment for mergeable packet buffers. */
50 #define MERGEABLE_BUFFER_ALIGN max(L1_CACHE_BYTES, 256)
52 #define VIRTNET_DRIVER_VERSION "1.0.0"
54 struct virtnet_stats {
55 struct u64_stats_sync tx_syncp;
56 struct u64_stats_sync rx_syncp;
64 /* Internal representation of a send virtqueue */
66 /* Virtqueue associated with this send _queue */
69 /* TX: fragments + linear part + virtio header */
70 struct scatterlist sg[MAX_SKB_FRAGS + 2];
72 /* Name of the send queue: output.$index */
76 /* Internal representation of a receive virtqueue */
77 struct receive_queue {
78 /* Virtqueue associated with this receive_queue */
81 struct napi_struct napi;
83 /* Chain pages by the private ptr. */
86 /* Average packet length for mergeable receive buffers. */
87 struct ewma mrg_avg_pkt_len;
89 /* Page frag for packet buffer allocation. */
90 struct page_frag alloc_frag;
92 /* RX: fragments + linear part + virtio header */
93 struct scatterlist sg[MAX_SKB_FRAGS + 2];
95 /* Name of this receive queue: input.$index */
100 struct virtio_device *vdev;
101 struct virtqueue *cvq;
102 struct net_device *dev;
103 struct send_queue *sq;
104 struct receive_queue *rq;
107 /* Max # of queue pairs supported by the device */
110 /* # of queue pairs currently used by the driver */
111 u16 curr_queue_pairs;
113 /* I like... big packets and I cannot lie! */
116 /* Host will merge rx buffers for big packets (shake it! shake it!) */
117 bool mergeable_rx_bufs;
119 /* Has control virtqueue */
122 /* Host can handle any s/g split between our header and packet data */
125 /* enable config space updates */
128 /* Active statistics */
129 struct virtnet_stats __percpu *stats;
131 /* Work struct for refilling if we run low on memory. */
132 struct delayed_work refill;
134 /* Work struct for config space updates */
135 struct work_struct config_work;
137 /* Lock for config space updates */
138 struct mutex config_lock;
140 /* Does the affinity hint is set for virtqueues? */
141 bool affinity_hint_set;
143 /* CPU hot plug notifier */
144 struct notifier_block nb;
147 struct skb_vnet_hdr {
149 struct virtio_net_hdr hdr;
150 struct virtio_net_hdr_mrg_rxbuf mhdr;
154 struct padded_vnet_hdr {
155 struct virtio_net_hdr hdr;
157 * virtio_net_hdr should be in a separated sg buffer because of a
158 * QEMU bug, and data sg buffer shares same page with this header sg.
159 * This padding makes next sg 16 byte aligned after virtio_net_hdr.
164 /* Converting between virtqueue no. and kernel tx/rx queue no.
165 * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq
167 static int vq2txq(struct virtqueue *vq)
169 return (vq->index - 1) / 2;
172 static int txq2vq(int txq)
177 static int vq2rxq(struct virtqueue *vq)
179 return vq->index / 2;
182 static int rxq2vq(int rxq)
187 static inline struct skb_vnet_hdr *skb_vnet_hdr(struct sk_buff *skb)
189 return (struct skb_vnet_hdr *)skb->cb;
193 * private is used to chain pages for big packets, put the whole
194 * most recent used list in the beginning for reuse
196 static void give_pages(struct receive_queue *rq, struct page *page)
200 /* Find end of list, sew whole thing into vi->rq.pages. */
201 for (end = page; end->private; end = (struct page *)end->private);
202 end->private = (unsigned long)rq->pages;
206 static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask)
208 struct page *p = rq->pages;
211 rq->pages = (struct page *)p->private;
212 /* clear private here, it is used to chain pages */
215 p = alloc_page(gfp_mask);
219 static void skb_xmit_done(struct virtqueue *vq)
221 struct virtnet_info *vi = vq->vdev->priv;
223 /* Suppress further interrupts. */
224 virtqueue_disable_cb(vq);
226 /* We were probably waiting for more output buffers. */
227 netif_wake_subqueue(vi->dev, vq2txq(vq));
230 static unsigned int mergeable_ctx_to_buf_truesize(unsigned long mrg_ctx)
232 unsigned int truesize = mrg_ctx & (MERGEABLE_BUFFER_ALIGN - 1);
233 return (truesize + 1) * MERGEABLE_BUFFER_ALIGN;
236 static void *mergeable_ctx_to_buf_address(unsigned long mrg_ctx)
238 return (void *)(mrg_ctx & -MERGEABLE_BUFFER_ALIGN);
242 static unsigned long mergeable_buf_to_ctx(void *buf, unsigned int truesize)
244 unsigned int size = truesize / MERGEABLE_BUFFER_ALIGN;
245 return (unsigned long)buf | (size - 1);
248 /* Called from bottom half context */
249 static struct sk_buff *page_to_skb(struct receive_queue *rq,
250 struct page *page, unsigned int offset,
251 unsigned int len, unsigned int truesize)
253 struct virtnet_info *vi = rq->vq->vdev->priv;
255 struct skb_vnet_hdr *hdr;
256 unsigned int copy, hdr_len, hdr_padded_len;
259 p = page_address(page) + offset;
261 /* copy small packet so we can reuse these pages for small data */
262 skb = netdev_alloc_skb_ip_align(vi->dev, GOOD_COPY_LEN);
266 hdr = skb_vnet_hdr(skb);
268 if (vi->mergeable_rx_bufs) {
269 hdr_len = sizeof hdr->mhdr;
270 hdr_padded_len = sizeof hdr->mhdr;
272 hdr_len = sizeof hdr->hdr;
273 hdr_padded_len = sizeof(struct padded_vnet_hdr);
276 memcpy(hdr, p, hdr_len);
279 offset += hdr_padded_len;
283 if (copy > skb_tailroom(skb))
284 copy = skb_tailroom(skb);
285 memcpy(skb_put(skb, copy), p, copy);
290 if (vi->mergeable_rx_bufs) {
292 skb_add_rx_frag(skb, 0, page, offset, len, truesize);
299 * Verify that we can indeed put this data into a skb.
300 * This is here to handle cases when the device erroneously
301 * tries to receive more than is possible. This is usually
302 * the case of a broken device.
304 if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) {
305 net_dbg_ratelimited("%s: too much data\n", skb->dev->name);
309 BUG_ON(offset >= PAGE_SIZE);
311 unsigned int frag_size = min((unsigned)PAGE_SIZE - offset, len);
312 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page, offset,
313 frag_size, truesize);
315 page = (struct page *)page->private;
320 give_pages(rq, page);
325 static struct sk_buff *receive_small(void *buf, unsigned int len)
327 struct sk_buff * skb = buf;
329 len -= sizeof(struct virtio_net_hdr);
335 static struct sk_buff *receive_big(struct net_device *dev,
336 struct receive_queue *rq,
340 struct page *page = buf;
341 struct sk_buff *skb = page_to_skb(rq, page, 0, len, PAGE_SIZE);
349 dev->stats.rx_dropped++;
350 give_pages(rq, page);
354 static struct sk_buff *receive_mergeable(struct net_device *dev,
355 struct receive_queue *rq,
359 void *buf = mergeable_ctx_to_buf_address(ctx);
360 struct skb_vnet_hdr *hdr = buf;
361 int num_buf = hdr->mhdr.num_buffers;
362 struct page *page = virt_to_head_page(buf);
363 int offset = buf - page_address(page);
364 unsigned int truesize = max(len, mergeable_ctx_to_buf_truesize(ctx));
366 struct sk_buff *head_skb = page_to_skb(rq, page, offset, len, truesize);
367 struct sk_buff *curr_skb = head_skb;
369 if (unlikely(!curr_skb))
374 ctx = (unsigned long)virtqueue_get_buf(rq->vq, &len);
375 if (unlikely(!ctx)) {
376 pr_debug("%s: rx error: %d buffers out of %d missing\n",
377 dev->name, num_buf, hdr->mhdr.num_buffers);
378 dev->stats.rx_length_errors++;
382 buf = mergeable_ctx_to_buf_address(ctx);
383 page = virt_to_head_page(buf);
385 num_skb_frags = skb_shinfo(curr_skb)->nr_frags;
386 if (unlikely(num_skb_frags == MAX_SKB_FRAGS)) {
387 struct sk_buff *nskb = alloc_skb(0, GFP_ATOMIC);
391 if (curr_skb == head_skb)
392 skb_shinfo(curr_skb)->frag_list = nskb;
394 curr_skb->next = nskb;
396 head_skb->truesize += nskb->truesize;
399 truesize = max(len, mergeable_ctx_to_buf_truesize(ctx));
400 if (curr_skb != head_skb) {
401 head_skb->data_len += len;
402 head_skb->len += len;
403 head_skb->truesize += truesize;
405 offset = buf - page_address(page);
406 if (skb_can_coalesce(curr_skb, num_skb_frags, page, offset)) {
408 skb_coalesce_rx_frag(curr_skb, num_skb_frags - 1,
411 skb_add_rx_frag(curr_skb, num_skb_frags, page,
412 offset, len, truesize);
416 ewma_add(&rq->mrg_avg_pkt_len, head_skb->len);
422 ctx = (unsigned long)virtqueue_get_buf(rq->vq, &len);
423 if (unlikely(!ctx)) {
424 pr_debug("%s: rx error: %d buffers missing\n",
426 dev->stats.rx_length_errors++;
429 page = virt_to_head_page(mergeable_ctx_to_buf_address(ctx));
433 dev->stats.rx_dropped++;
434 dev_kfree_skb(head_skb);
438 static void receive_buf(struct receive_queue *rq, void *buf, unsigned int len)
440 struct virtnet_info *vi = rq->vq->vdev->priv;
441 struct net_device *dev = vi->dev;
442 struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
444 struct skb_vnet_hdr *hdr;
446 if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
447 pr_debug("%s: short packet %i\n", dev->name, len);
448 dev->stats.rx_length_errors++;
449 if (vi->mergeable_rx_bufs) {
450 unsigned long ctx = (unsigned long)buf;
451 void *base = mergeable_ctx_to_buf_address(ctx);
452 put_page(virt_to_head_page(base));
453 } else if (vi->big_packets) {
461 if (vi->mergeable_rx_bufs)
462 skb = receive_mergeable(dev, rq, (unsigned long)buf, len);
463 else if (vi->big_packets)
464 skb = receive_big(dev, rq, buf, len);
466 skb = receive_small(buf, len);
471 hdr = skb_vnet_hdr(skb);
473 u64_stats_update_begin(&stats->rx_syncp);
474 stats->rx_bytes += skb->len;
476 u64_stats_update_end(&stats->rx_syncp);
478 if (hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
479 pr_debug("Needs csum!\n");
480 if (!skb_partial_csum_set(skb,
482 hdr->hdr.csum_offset))
484 } else if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) {
485 skb->ip_summed = CHECKSUM_UNNECESSARY;
488 skb->protocol = eth_type_trans(skb, dev);
489 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
490 ntohs(skb->protocol), skb->len, skb->pkt_type);
492 if (hdr->hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
494 switch (hdr->hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
495 case VIRTIO_NET_HDR_GSO_TCPV4:
496 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
498 case VIRTIO_NET_HDR_GSO_UDP:
499 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
501 case VIRTIO_NET_HDR_GSO_TCPV6:
502 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
505 net_warn_ratelimited("%s: bad gso type %u.\n",
506 dev->name, hdr->hdr.gso_type);
510 if (hdr->hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
511 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
513 skb_shinfo(skb)->gso_size = hdr->hdr.gso_size;
514 if (skb_shinfo(skb)->gso_size == 0) {
515 net_warn_ratelimited("%s: zero gso size.\n", dev->name);
519 /* Header must be checked, and gso_segs computed. */
520 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
521 skb_shinfo(skb)->gso_segs = 0;
524 netif_receive_skb(skb);
528 dev->stats.rx_frame_errors++;
532 static int add_recvbuf_small(struct receive_queue *rq, gfp_t gfp)
534 struct virtnet_info *vi = rq->vq->vdev->priv;
536 struct skb_vnet_hdr *hdr;
539 skb = __netdev_alloc_skb_ip_align(vi->dev, GOOD_PACKET_LEN, gfp);
543 skb_put(skb, GOOD_PACKET_LEN);
545 hdr = skb_vnet_hdr(skb);
546 sg_set_buf(rq->sg, &hdr->hdr, sizeof hdr->hdr);
548 skb_to_sgvec(skb, rq->sg + 1, 0, skb->len);
550 err = virtqueue_add_inbuf(rq->vq, rq->sg, 2, skb, gfp);
557 static int add_recvbuf_big(struct receive_queue *rq, gfp_t gfp)
559 struct page *first, *list = NULL;
563 /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */
564 for (i = MAX_SKB_FRAGS + 1; i > 1; --i) {
565 first = get_a_page(rq, gfp);
568 give_pages(rq, list);
571 sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE);
573 /* chain new page in list head to match sg */
574 first->private = (unsigned long)list;
578 first = get_a_page(rq, gfp);
580 give_pages(rq, list);
583 p = page_address(first);
585 /* rq->sg[0], rq->sg[1] share the same page */
586 /* a separated rq->sg[0] for virtio_net_hdr only due to QEMU bug */
587 sg_set_buf(&rq->sg[0], p, sizeof(struct virtio_net_hdr));
589 /* rq->sg[1] for data packet, from offset */
590 offset = sizeof(struct padded_vnet_hdr);
591 sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset);
593 /* chain first in list head */
594 first->private = (unsigned long)list;
595 err = virtqueue_add_inbuf(rq->vq, rq->sg, MAX_SKB_FRAGS + 2,
598 give_pages(rq, first);
603 static int add_recvbuf_mergeable(struct receive_queue *rq, gfp_t gfp)
605 const size_t hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
606 struct page_frag *alloc_frag = &rq->alloc_frag;
610 unsigned int len, hole;
612 len = hdr_len + clamp_t(unsigned int, ewma_read(&rq->mrg_avg_pkt_len),
613 GOOD_PACKET_LEN, PAGE_SIZE - hdr_len);
614 len = ALIGN(len, MERGEABLE_BUFFER_ALIGN);
615 if (unlikely(!skb_page_frag_refill(len, alloc_frag, gfp)))
618 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
619 ctx = mergeable_buf_to_ctx(buf, len);
620 get_page(alloc_frag->page);
621 alloc_frag->offset += len;
622 hole = alloc_frag->size - alloc_frag->offset;
624 /* To avoid internal fragmentation, if there is very likely not
625 * enough space for another buffer, add the remaining space to
626 * the current buffer. This extra space is not included in
627 * the truesize stored in ctx.
630 alloc_frag->offset += hole;
633 sg_init_one(rq->sg, buf, len);
634 err = virtqueue_add_inbuf(rq->vq, rq->sg, 1, (void *)ctx, gfp);
636 put_page(virt_to_head_page(buf));
642 * Returns false if we couldn't fill entirely (OOM).
644 * Normally run in the receive path, but can also be run from ndo_open
645 * before we're receiving packets, or from refill_work which is
646 * careful to disable receiving (using napi_disable).
648 static bool try_fill_recv(struct receive_queue *rq, gfp_t gfp)
650 struct virtnet_info *vi = rq->vq->vdev->priv;
656 if (vi->mergeable_rx_bufs)
657 err = add_recvbuf_mergeable(rq, gfp);
658 else if (vi->big_packets)
659 err = add_recvbuf_big(rq, gfp);
661 err = add_recvbuf_small(rq, gfp);
663 oom = err == -ENOMEM;
666 } while (rq->vq->num_free);
667 if (unlikely(!virtqueue_kick(rq->vq)))
672 static void skb_recv_done(struct virtqueue *rvq)
674 struct virtnet_info *vi = rvq->vdev->priv;
675 struct receive_queue *rq = &vi->rq[vq2rxq(rvq)];
677 /* Schedule NAPI, Suppress further interrupts if successful. */
678 if (napi_schedule_prep(&rq->napi)) {
679 virtqueue_disable_cb(rvq);
680 __napi_schedule(&rq->napi);
684 static void virtnet_napi_enable(struct receive_queue *rq)
686 napi_enable(&rq->napi);
688 /* If all buffers were filled by other side before we napi_enabled, we
689 * won't get another interrupt, so process any outstanding packets
690 * now. virtnet_poll wants re-enable the queue, so we disable here.
691 * We synchronize against interrupts via NAPI_STATE_SCHED */
692 if (napi_schedule_prep(&rq->napi)) {
693 virtqueue_disable_cb(rq->vq);
695 __napi_schedule(&rq->napi);
700 static void refill_work(struct work_struct *work)
702 struct virtnet_info *vi =
703 container_of(work, struct virtnet_info, refill.work);
707 for (i = 0; i < vi->curr_queue_pairs; i++) {
708 struct receive_queue *rq = &vi->rq[i];
710 napi_disable(&rq->napi);
711 still_empty = !try_fill_recv(rq, GFP_KERNEL);
712 virtnet_napi_enable(rq);
714 /* In theory, this can happen: if we don't get any buffers in
715 * we will *never* try to fill again.
718 schedule_delayed_work(&vi->refill, HZ/2);
722 static int virtnet_poll(struct napi_struct *napi, int budget)
724 struct receive_queue *rq =
725 container_of(napi, struct receive_queue, napi);
726 struct virtnet_info *vi = rq->vq->vdev->priv;
728 unsigned int r, len, received = 0;
731 while (received < budget &&
732 (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) {
733 receive_buf(rq, buf, len);
737 if (rq->vq->num_free > virtqueue_get_vring_size(rq->vq) / 2) {
738 if (!try_fill_recv(rq, GFP_ATOMIC))
739 schedule_delayed_work(&vi->refill, 0);
742 /* Out of packets? */
743 if (received < budget) {
744 r = virtqueue_enable_cb_prepare(rq->vq);
746 if (unlikely(virtqueue_poll(rq->vq, r)) &&
747 napi_schedule_prep(napi)) {
748 virtqueue_disable_cb(rq->vq);
749 __napi_schedule(napi);
757 static int virtnet_open(struct net_device *dev)
759 struct virtnet_info *vi = netdev_priv(dev);
762 for (i = 0; i < vi->max_queue_pairs; i++) {
763 if (i < vi->curr_queue_pairs)
764 /* Make sure we have some buffers: if oom use wq. */
765 if (!try_fill_recv(&vi->rq[i], GFP_KERNEL))
766 schedule_delayed_work(&vi->refill, 0);
767 virtnet_napi_enable(&vi->rq[i]);
773 static void free_old_xmit_skbs(struct send_queue *sq)
777 struct virtnet_info *vi = sq->vq->vdev->priv;
778 struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
780 while ((skb = virtqueue_get_buf(sq->vq, &len)) != NULL) {
781 pr_debug("Sent skb %p\n", skb);
783 u64_stats_update_begin(&stats->tx_syncp);
784 stats->tx_bytes += skb->len;
786 u64_stats_update_end(&stats->tx_syncp);
788 dev_kfree_skb_any(skb);
792 static int xmit_skb(struct send_queue *sq, struct sk_buff *skb)
794 struct skb_vnet_hdr *hdr;
795 const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
796 struct virtnet_info *vi = sq->vq->vdev->priv;
801 pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
802 if (vi->mergeable_rx_bufs)
803 hdr_len = sizeof hdr->mhdr;
805 hdr_len = sizeof hdr->hdr;
807 can_push = vi->any_header_sg &&
808 !((unsigned long)skb->data & (__alignof__(*hdr) - 1)) &&
809 !skb_header_cloned(skb) && skb_headroom(skb) >= hdr_len;
810 /* Even if we can, don't push here yet as this would skew
811 * csum_start offset below. */
813 hdr = (struct skb_vnet_hdr *)(skb->data - hdr_len);
815 hdr = skb_vnet_hdr(skb);
817 if (skb->ip_summed == CHECKSUM_PARTIAL) {
818 hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
819 hdr->hdr.csum_start = skb_checksum_start_offset(skb);
820 hdr->hdr.csum_offset = skb->csum_offset;
823 hdr->hdr.csum_offset = hdr->hdr.csum_start = 0;
826 if (skb_is_gso(skb)) {
827 hdr->hdr.hdr_len = skb_headlen(skb);
828 hdr->hdr.gso_size = skb_shinfo(skb)->gso_size;
829 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
830 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
831 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
832 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
833 else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
834 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
837 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
838 hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
840 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
841 hdr->hdr.gso_size = hdr->hdr.hdr_len = 0;
844 if (vi->mergeable_rx_bufs)
845 hdr->mhdr.num_buffers = 0;
848 __skb_push(skb, hdr_len);
849 num_sg = skb_to_sgvec(skb, sq->sg, 0, skb->len);
850 /* Pull header back to avoid skew in tx bytes calculations. */
851 __skb_pull(skb, hdr_len);
853 sg_set_buf(sq->sg, hdr, hdr_len);
854 num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len) + 1;
856 return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC);
859 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
861 struct virtnet_info *vi = netdev_priv(dev);
862 int qnum = skb_get_queue_mapping(skb);
863 struct send_queue *sq = &vi->sq[qnum];
866 /* Free up any pending old buffers before queueing new ones. */
867 free_old_xmit_skbs(sq);
869 /* Try to transmit */
870 err = xmit_skb(sq, skb);
872 /* This should not happen! */
873 if (unlikely(err) || unlikely(!virtqueue_kick(sq->vq))) {
874 dev->stats.tx_fifo_errors++;
877 "Unexpected TXQ (%d) queue failure: %d\n", qnum, err);
878 dev->stats.tx_dropped++;
883 /* Don't wait up for transmitted skbs to be freed. */
887 /* Apparently nice girls don't return TX_BUSY; stop the queue
888 * before it gets out of hand. Naturally, this wastes entries. */
889 if (sq->vq->num_free < 2+MAX_SKB_FRAGS) {
890 netif_stop_subqueue(dev, qnum);
891 if (unlikely(!virtqueue_enable_cb_delayed(sq->vq))) {
892 /* More just got used, free them then recheck. */
893 free_old_xmit_skbs(sq);
894 if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) {
895 netif_start_subqueue(dev, qnum);
896 virtqueue_disable_cb(sq->vq);
905 * Send command via the control virtqueue and check status. Commands
906 * supported by the hypervisor, as indicated by feature bits, should
907 * never fail unless improperly formatted.
909 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
910 struct scatterlist *out)
912 struct scatterlist *sgs[4], hdr, stat;
913 struct virtio_net_ctrl_hdr ctrl;
914 virtio_net_ctrl_ack status = ~0;
915 unsigned out_num = 0, tmp;
917 /* Caller should know better */
918 BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ));
923 sg_init_one(&hdr, &ctrl, sizeof(ctrl));
924 sgs[out_num++] = &hdr;
927 sgs[out_num++] = out;
929 /* Add return status. */
930 sg_init_one(&stat, &status, sizeof(status));
931 sgs[out_num] = &stat;
933 BUG_ON(out_num + 1 > ARRAY_SIZE(sgs));
934 BUG_ON(virtqueue_add_sgs(vi->cvq, sgs, out_num, 1, vi, GFP_ATOMIC) < 0);
936 if (unlikely(!virtqueue_kick(vi->cvq)))
937 return status == VIRTIO_NET_OK;
939 /* Spin for a response, the kick causes an ioport write, trapping
940 * into the hypervisor, so the request should be handled immediately.
942 while (!virtqueue_get_buf(vi->cvq, &tmp) &&
943 !virtqueue_is_broken(vi->cvq))
946 return status == VIRTIO_NET_OK;
949 static int virtnet_set_mac_address(struct net_device *dev, void *p)
951 struct virtnet_info *vi = netdev_priv(dev);
952 struct virtio_device *vdev = vi->vdev;
954 struct sockaddr *addr = p;
955 struct scatterlist sg;
957 ret = eth_prepare_mac_addr_change(dev, p);
961 if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
962 sg_init_one(&sg, addr->sa_data, dev->addr_len);
963 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
964 VIRTIO_NET_CTRL_MAC_ADDR_SET, &sg)) {
966 "Failed to set mac address by vq command.\n");
969 } else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
972 /* Naturally, this has an atomicity problem. */
973 for (i = 0; i < dev->addr_len; i++)
975 offsetof(struct virtio_net_config, mac) +
976 i, addr->sa_data[i]);
979 eth_commit_mac_addr_change(dev, p);
984 static struct rtnl_link_stats64 *virtnet_stats(struct net_device *dev,
985 struct rtnl_link_stats64 *tot)
987 struct virtnet_info *vi = netdev_priv(dev);
991 for_each_possible_cpu(cpu) {
992 struct virtnet_stats *stats = per_cpu_ptr(vi->stats, cpu);
993 u64 tpackets, tbytes, rpackets, rbytes;
996 start = u64_stats_fetch_begin_bh(&stats->tx_syncp);
997 tpackets = stats->tx_packets;
998 tbytes = stats->tx_bytes;
999 } while (u64_stats_fetch_retry_bh(&stats->tx_syncp, start));
1002 start = u64_stats_fetch_begin_bh(&stats->rx_syncp);
1003 rpackets = stats->rx_packets;
1004 rbytes = stats->rx_bytes;
1005 } while (u64_stats_fetch_retry_bh(&stats->rx_syncp, start));
1007 tot->rx_packets += rpackets;
1008 tot->tx_packets += tpackets;
1009 tot->rx_bytes += rbytes;
1010 tot->tx_bytes += tbytes;
1013 tot->tx_dropped = dev->stats.tx_dropped;
1014 tot->tx_fifo_errors = dev->stats.tx_fifo_errors;
1015 tot->rx_dropped = dev->stats.rx_dropped;
1016 tot->rx_length_errors = dev->stats.rx_length_errors;
1017 tot->rx_frame_errors = dev->stats.rx_frame_errors;
1022 #ifdef CONFIG_NET_POLL_CONTROLLER
1023 static void virtnet_netpoll(struct net_device *dev)
1025 struct virtnet_info *vi = netdev_priv(dev);
1028 for (i = 0; i < vi->curr_queue_pairs; i++)
1029 napi_schedule(&vi->rq[i].napi);
1033 static void virtnet_ack_link_announce(struct virtnet_info *vi)
1036 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE,
1037 VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL))
1038 dev_warn(&vi->dev->dev, "Failed to ack link announce.\n");
1042 static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs)
1044 struct scatterlist sg;
1045 struct virtio_net_ctrl_mq s;
1046 struct net_device *dev = vi->dev;
1048 if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ))
1051 s.virtqueue_pairs = queue_pairs;
1052 sg_init_one(&sg, &s, sizeof(s));
1054 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ,
1055 VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg)) {
1056 dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n",
1060 vi->curr_queue_pairs = queue_pairs;
1061 /* virtnet_open() will refill when device is going to up. */
1062 if (dev->flags & IFF_UP)
1063 schedule_delayed_work(&vi->refill, 0);
1069 static int virtnet_close(struct net_device *dev)
1071 struct virtnet_info *vi = netdev_priv(dev);
1074 /* Make sure refill_work doesn't re-enable napi! */
1075 cancel_delayed_work_sync(&vi->refill);
1077 for (i = 0; i < vi->max_queue_pairs; i++)
1078 napi_disable(&vi->rq[i].napi);
1083 static void virtnet_set_rx_mode(struct net_device *dev)
1085 struct virtnet_info *vi = netdev_priv(dev);
1086 struct scatterlist sg[2];
1087 u8 promisc, allmulti;
1088 struct virtio_net_ctrl_mac *mac_data;
1089 struct netdev_hw_addr *ha;
1095 /* We can't dynamically set ndo_set_rx_mode, so return gracefully */
1096 if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
1099 promisc = ((dev->flags & IFF_PROMISC) != 0);
1100 allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
1102 sg_init_one(sg, &promisc, sizeof(promisc));
1104 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
1105 VIRTIO_NET_CTRL_RX_PROMISC, sg))
1106 dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
1107 promisc ? "en" : "dis");
1109 sg_init_one(sg, &allmulti, sizeof(allmulti));
1111 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
1112 VIRTIO_NET_CTRL_RX_ALLMULTI, sg))
1113 dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
1114 allmulti ? "en" : "dis");
1116 uc_count = netdev_uc_count(dev);
1117 mc_count = netdev_mc_count(dev);
1118 /* MAC filter - use one buffer for both lists */
1119 buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
1120 (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
1125 sg_init_table(sg, 2);
1127 /* Store the unicast list and count in the front of the buffer */
1128 mac_data->entries = uc_count;
1130 netdev_for_each_uc_addr(ha, dev)
1131 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
1133 sg_set_buf(&sg[0], mac_data,
1134 sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
1136 /* multicast list and count fill the end */
1137 mac_data = (void *)&mac_data->macs[uc_count][0];
1139 mac_data->entries = mc_count;
1141 netdev_for_each_mc_addr(ha, dev)
1142 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
1144 sg_set_buf(&sg[1], mac_data,
1145 sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
1147 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
1148 VIRTIO_NET_CTRL_MAC_TABLE_SET, sg))
1149 dev_warn(&dev->dev, "Failed to set MAC filter table.\n");
1154 static int virtnet_vlan_rx_add_vid(struct net_device *dev,
1155 __be16 proto, u16 vid)
1157 struct virtnet_info *vi = netdev_priv(dev);
1158 struct scatterlist sg;
1160 sg_init_one(&sg, &vid, sizeof(vid));
1162 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1163 VIRTIO_NET_CTRL_VLAN_ADD, &sg))
1164 dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
1168 static int virtnet_vlan_rx_kill_vid(struct net_device *dev,
1169 __be16 proto, u16 vid)
1171 struct virtnet_info *vi = netdev_priv(dev);
1172 struct scatterlist sg;
1174 sg_init_one(&sg, &vid, sizeof(vid));
1176 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1177 VIRTIO_NET_CTRL_VLAN_DEL, &sg))
1178 dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
1182 static void virtnet_clean_affinity(struct virtnet_info *vi, long hcpu)
1186 if (vi->affinity_hint_set) {
1187 for (i = 0; i < vi->max_queue_pairs; i++) {
1188 virtqueue_set_affinity(vi->rq[i].vq, -1);
1189 virtqueue_set_affinity(vi->sq[i].vq, -1);
1192 vi->affinity_hint_set = false;
1196 static void virtnet_set_affinity(struct virtnet_info *vi)
1201 /* In multiqueue mode, when the number of cpu is equal to the number of
1202 * queue pairs, we let the queue pairs to be private to one cpu by
1203 * setting the affinity hint to eliminate the contention.
1205 if (vi->curr_queue_pairs == 1 ||
1206 vi->max_queue_pairs != num_online_cpus()) {
1207 virtnet_clean_affinity(vi, -1);
1212 for_each_online_cpu(cpu) {
1213 virtqueue_set_affinity(vi->rq[i].vq, cpu);
1214 virtqueue_set_affinity(vi->sq[i].vq, cpu);
1215 netif_set_xps_queue(vi->dev, cpumask_of(cpu), i);
1219 vi->affinity_hint_set = true;
1222 static int virtnet_cpu_callback(struct notifier_block *nfb,
1223 unsigned long action, void *hcpu)
1225 struct virtnet_info *vi = container_of(nfb, struct virtnet_info, nb);
1227 switch(action & ~CPU_TASKS_FROZEN) {
1229 case CPU_DOWN_FAILED:
1231 virtnet_set_affinity(vi);
1233 case CPU_DOWN_PREPARE:
1234 virtnet_clean_affinity(vi, (long)hcpu);
1243 static void virtnet_get_ringparam(struct net_device *dev,
1244 struct ethtool_ringparam *ring)
1246 struct virtnet_info *vi = netdev_priv(dev);
1248 ring->rx_max_pending = virtqueue_get_vring_size(vi->rq[0].vq);
1249 ring->tx_max_pending = virtqueue_get_vring_size(vi->sq[0].vq);
1250 ring->rx_pending = ring->rx_max_pending;
1251 ring->tx_pending = ring->tx_max_pending;
1255 static void virtnet_get_drvinfo(struct net_device *dev,
1256 struct ethtool_drvinfo *info)
1258 struct virtnet_info *vi = netdev_priv(dev);
1259 struct virtio_device *vdev = vi->vdev;
1261 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
1262 strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version));
1263 strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info));
1267 /* TODO: Eliminate OOO packets during switching */
1268 static int virtnet_set_channels(struct net_device *dev,
1269 struct ethtool_channels *channels)
1271 struct virtnet_info *vi = netdev_priv(dev);
1272 u16 queue_pairs = channels->combined_count;
1275 /* We don't support separate rx/tx channels.
1276 * We don't allow setting 'other' channels.
1278 if (channels->rx_count || channels->tx_count || channels->other_count)
1281 if (queue_pairs > vi->max_queue_pairs)
1285 err = virtnet_set_queues(vi, queue_pairs);
1287 netif_set_real_num_tx_queues(dev, queue_pairs);
1288 netif_set_real_num_rx_queues(dev, queue_pairs);
1290 virtnet_set_affinity(vi);
1297 static void virtnet_get_channels(struct net_device *dev,
1298 struct ethtool_channels *channels)
1300 struct virtnet_info *vi = netdev_priv(dev);
1302 channels->combined_count = vi->curr_queue_pairs;
1303 channels->max_combined = vi->max_queue_pairs;
1304 channels->max_other = 0;
1305 channels->rx_count = 0;
1306 channels->tx_count = 0;
1307 channels->other_count = 0;
1310 static const struct ethtool_ops virtnet_ethtool_ops = {
1311 .get_drvinfo = virtnet_get_drvinfo,
1312 .get_link = ethtool_op_get_link,
1313 .get_ringparam = virtnet_get_ringparam,
1314 .set_channels = virtnet_set_channels,
1315 .get_channels = virtnet_get_channels,
1319 #define MAX_MTU 65535
1321 static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
1323 if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
1329 static const struct net_device_ops virtnet_netdev = {
1330 .ndo_open = virtnet_open,
1331 .ndo_stop = virtnet_close,
1332 .ndo_start_xmit = start_xmit,
1333 .ndo_validate_addr = eth_validate_addr,
1334 .ndo_set_mac_address = virtnet_set_mac_address,
1335 .ndo_set_rx_mode = virtnet_set_rx_mode,
1336 .ndo_change_mtu = virtnet_change_mtu,
1337 .ndo_get_stats64 = virtnet_stats,
1338 .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
1339 .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
1340 #ifdef CONFIG_NET_POLL_CONTROLLER
1341 .ndo_poll_controller = virtnet_netpoll,
1345 static void virtnet_config_changed_work(struct work_struct *work)
1347 struct virtnet_info *vi =
1348 container_of(work, struct virtnet_info, config_work);
1351 mutex_lock(&vi->config_lock);
1352 if (!vi->config_enable)
1355 if (virtio_cread_feature(vi->vdev, VIRTIO_NET_F_STATUS,
1356 struct virtio_net_config, status, &v) < 0)
1359 if (v & VIRTIO_NET_S_ANNOUNCE) {
1360 netdev_notify_peers(vi->dev);
1361 virtnet_ack_link_announce(vi);
1364 /* Ignore unknown (future) status bits */
1365 v &= VIRTIO_NET_S_LINK_UP;
1367 if (vi->status == v)
1372 if (vi->status & VIRTIO_NET_S_LINK_UP) {
1373 netif_carrier_on(vi->dev);
1374 netif_tx_wake_all_queues(vi->dev);
1376 netif_carrier_off(vi->dev);
1377 netif_tx_stop_all_queues(vi->dev);
1380 mutex_unlock(&vi->config_lock);
1383 static void virtnet_config_changed(struct virtio_device *vdev)
1385 struct virtnet_info *vi = vdev->priv;
1387 schedule_work(&vi->config_work);
1390 static void virtnet_free_queues(struct virtnet_info *vi)
1394 for (i = 0; i < vi->max_queue_pairs; i++)
1395 netif_napi_del(&vi->rq[i].napi);
1401 static void free_receive_bufs(struct virtnet_info *vi)
1405 for (i = 0; i < vi->max_queue_pairs; i++) {
1406 while (vi->rq[i].pages)
1407 __free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0);
1411 static void free_receive_page_frags(struct virtnet_info *vi)
1414 for (i = 0; i < vi->max_queue_pairs; i++)
1415 if (vi->rq[i].alloc_frag.page)
1416 put_page(vi->rq[i].alloc_frag.page);
1419 static void free_unused_bufs(struct virtnet_info *vi)
1424 for (i = 0; i < vi->max_queue_pairs; i++) {
1425 struct virtqueue *vq = vi->sq[i].vq;
1426 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL)
1430 for (i = 0; i < vi->max_queue_pairs; i++) {
1431 struct virtqueue *vq = vi->rq[i].vq;
1433 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) {
1434 if (vi->mergeable_rx_bufs) {
1435 unsigned long ctx = (unsigned long)buf;
1436 void *base = mergeable_ctx_to_buf_address(ctx);
1437 put_page(virt_to_head_page(base));
1438 } else if (vi->big_packets) {
1439 give_pages(&vi->rq[i], buf);
1447 static void virtnet_del_vqs(struct virtnet_info *vi)
1449 struct virtio_device *vdev = vi->vdev;
1451 virtnet_clean_affinity(vi, -1);
1453 vdev->config->del_vqs(vdev);
1455 virtnet_free_queues(vi);
1458 static int virtnet_find_vqs(struct virtnet_info *vi)
1460 vq_callback_t **callbacks;
1461 struct virtqueue **vqs;
1466 /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
1467 * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
1468 * possible control vq.
1470 total_vqs = vi->max_queue_pairs * 2 +
1471 virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ);
1473 /* Allocate space for find_vqs parameters */
1474 vqs = kzalloc(total_vqs * sizeof(*vqs), GFP_KERNEL);
1477 callbacks = kmalloc(total_vqs * sizeof(*callbacks), GFP_KERNEL);
1480 names = kmalloc(total_vqs * sizeof(*names), GFP_KERNEL);
1484 /* Parameters for control virtqueue, if any */
1486 callbacks[total_vqs - 1] = NULL;
1487 names[total_vqs - 1] = "control";
1490 /* Allocate/initialize parameters for send/receive virtqueues */
1491 for (i = 0; i < vi->max_queue_pairs; i++) {
1492 callbacks[rxq2vq(i)] = skb_recv_done;
1493 callbacks[txq2vq(i)] = skb_xmit_done;
1494 sprintf(vi->rq[i].name, "input.%d", i);
1495 sprintf(vi->sq[i].name, "output.%d", i);
1496 names[rxq2vq(i)] = vi->rq[i].name;
1497 names[txq2vq(i)] = vi->sq[i].name;
1500 ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs, callbacks,
1506 vi->cvq = vqs[total_vqs - 1];
1507 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
1508 vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
1511 for (i = 0; i < vi->max_queue_pairs; i++) {
1512 vi->rq[i].vq = vqs[rxq2vq(i)];
1513 vi->sq[i].vq = vqs[txq2vq(i)];
1532 static int virtnet_alloc_queues(struct virtnet_info *vi)
1536 vi->sq = kzalloc(sizeof(*vi->sq) * vi->max_queue_pairs, GFP_KERNEL);
1539 vi->rq = kzalloc(sizeof(*vi->rq) * vi->max_queue_pairs, GFP_KERNEL);
1543 INIT_DELAYED_WORK(&vi->refill, refill_work);
1544 for (i = 0; i < vi->max_queue_pairs; i++) {
1545 vi->rq[i].pages = NULL;
1546 netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll,
1549 sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg));
1550 ewma_init(&vi->rq[i].mrg_avg_pkt_len, 1, RECEIVE_AVG_WEIGHT);
1551 sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg));
1562 static int init_vqs(struct virtnet_info *vi)
1566 /* Allocate send & receive queues */
1567 ret = virtnet_alloc_queues(vi);
1571 ret = virtnet_find_vqs(vi);
1576 virtnet_set_affinity(vi);
1582 virtnet_free_queues(vi);
1587 static int virtnet_probe(struct virtio_device *vdev)
1590 struct net_device *dev;
1591 struct virtnet_info *vi;
1592 u16 max_queue_pairs;
1594 /* Find if host supports multiqueue virtio_net device */
1595 err = virtio_cread_feature(vdev, VIRTIO_NET_F_MQ,
1596 struct virtio_net_config,
1597 max_virtqueue_pairs, &max_queue_pairs);
1599 /* We need at least 2 queue's */
1600 if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
1601 max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
1602 !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
1603 max_queue_pairs = 1;
1605 /* Allocate ourselves a network device with room for our info */
1606 dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs);
1610 /* Set up network device as normal. */
1611 dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE;
1612 dev->netdev_ops = &virtnet_netdev;
1613 dev->features = NETIF_F_HIGHDMA;
1615 SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
1616 SET_NETDEV_DEV(dev, &vdev->dev);
1618 /* Do we support "hardware" checksums? */
1619 if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
1620 /* This opens up the world of extra features. */
1621 dev->hw_features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1623 dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1625 if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
1626 dev->hw_features |= NETIF_F_TSO | NETIF_F_UFO
1627 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
1629 /* Individual feature bits: what can host handle? */
1630 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
1631 dev->hw_features |= NETIF_F_TSO;
1632 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
1633 dev->hw_features |= NETIF_F_TSO6;
1634 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
1635 dev->hw_features |= NETIF_F_TSO_ECN;
1636 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
1637 dev->hw_features |= NETIF_F_UFO;
1640 dev->features |= dev->hw_features & (NETIF_F_ALL_TSO|NETIF_F_UFO);
1641 /* (!csum && gso) case will be fixed by register_netdev() */
1643 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM))
1644 dev->features |= NETIF_F_RXCSUM;
1646 dev->vlan_features = dev->features;
1648 /* Configuration may specify what MAC to use. Otherwise random. */
1649 if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
1650 virtio_cread_bytes(vdev,
1651 offsetof(struct virtio_net_config, mac),
1652 dev->dev_addr, dev->addr_len);
1654 eth_hw_addr_random(dev);
1656 /* Set up our device-specific information */
1657 vi = netdev_priv(dev);
1661 vi->stats = alloc_percpu(struct virtnet_stats);
1663 if (vi->stats == NULL)
1666 for_each_possible_cpu(i) {
1667 struct virtnet_stats *virtnet_stats;
1668 virtnet_stats = per_cpu_ptr(vi->stats, i);
1669 u64_stats_init(&virtnet_stats->tx_syncp);
1670 u64_stats_init(&virtnet_stats->rx_syncp);
1673 mutex_init(&vi->config_lock);
1674 vi->config_enable = true;
1675 INIT_WORK(&vi->config_work, virtnet_config_changed_work);
1677 /* If we can receive ANY GSO packets, we must allocate large ones. */
1678 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
1679 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
1680 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
1681 vi->big_packets = true;
1683 if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
1684 vi->mergeable_rx_bufs = true;
1686 if (virtio_has_feature(vdev, VIRTIO_F_ANY_LAYOUT))
1687 vi->any_header_sg = true;
1689 if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
1692 /* Use single tx/rx queue pair as default */
1693 vi->curr_queue_pairs = 1;
1694 vi->max_queue_pairs = max_queue_pairs;
1696 /* Allocate/initialize the rx/tx queues, and invoke find_vqs */
1701 netif_set_real_num_tx_queues(dev, vi->curr_queue_pairs);
1702 netif_set_real_num_rx_queues(dev, vi->curr_queue_pairs);
1704 err = register_netdev(dev);
1706 pr_debug("virtio_net: registering device failed\n");
1710 /* Last of all, set up some receive buffers. */
1711 for (i = 0; i < vi->curr_queue_pairs; i++) {
1712 try_fill_recv(&vi->rq[i], GFP_KERNEL);
1714 /* If we didn't even get one input buffer, we're useless. */
1715 if (vi->rq[i].vq->num_free ==
1716 virtqueue_get_vring_size(vi->rq[i].vq)) {
1717 free_unused_bufs(vi);
1719 goto free_recv_bufs;
1723 vi->nb.notifier_call = &virtnet_cpu_callback;
1724 err = register_hotcpu_notifier(&vi->nb);
1726 pr_debug("virtio_net: registering cpu notifier failed\n");
1727 goto free_recv_bufs;
1730 /* Assume link up if device can't report link status,
1731 otherwise get link status from config. */
1732 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
1733 netif_carrier_off(dev);
1734 schedule_work(&vi->config_work);
1736 vi->status = VIRTIO_NET_S_LINK_UP;
1737 netif_carrier_on(dev);
1740 pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
1741 dev->name, max_queue_pairs);
1746 free_receive_bufs(vi);
1747 unregister_netdev(dev);
1749 cancel_delayed_work_sync(&vi->refill);
1750 free_receive_page_frags(vi);
1751 virtnet_del_vqs(vi);
1753 free_percpu(vi->stats);
1759 static void remove_vq_common(struct virtnet_info *vi)
1761 vi->vdev->config->reset(vi->vdev);
1763 /* Free unused buffers in both send and recv, if any. */
1764 free_unused_bufs(vi);
1766 free_receive_bufs(vi);
1768 free_receive_page_frags(vi);
1770 virtnet_del_vqs(vi);
1773 static void virtnet_remove(struct virtio_device *vdev)
1775 struct virtnet_info *vi = vdev->priv;
1777 unregister_hotcpu_notifier(&vi->nb);
1779 /* Prevent config work handler from accessing the device. */
1780 mutex_lock(&vi->config_lock);
1781 vi->config_enable = false;
1782 mutex_unlock(&vi->config_lock);
1784 unregister_netdev(vi->dev);
1786 remove_vq_common(vi);
1788 flush_work(&vi->config_work);
1790 free_percpu(vi->stats);
1791 free_netdev(vi->dev);
1794 #ifdef CONFIG_PM_SLEEP
1795 static int virtnet_freeze(struct virtio_device *vdev)
1797 struct virtnet_info *vi = vdev->priv;
1800 unregister_hotcpu_notifier(&vi->nb);
1802 /* Prevent config work handler from accessing the device */
1803 mutex_lock(&vi->config_lock);
1804 vi->config_enable = false;
1805 mutex_unlock(&vi->config_lock);
1807 netif_device_detach(vi->dev);
1808 cancel_delayed_work_sync(&vi->refill);
1810 if (netif_running(vi->dev))
1811 for (i = 0; i < vi->max_queue_pairs; i++) {
1812 napi_disable(&vi->rq[i].napi);
1813 netif_napi_del(&vi->rq[i].napi);
1816 remove_vq_common(vi);
1818 flush_work(&vi->config_work);
1823 static int virtnet_restore(struct virtio_device *vdev)
1825 struct virtnet_info *vi = vdev->priv;
1832 if (netif_running(vi->dev)) {
1833 for (i = 0; i < vi->curr_queue_pairs; i++)
1834 if (!try_fill_recv(&vi->rq[i], GFP_KERNEL))
1835 schedule_delayed_work(&vi->refill, 0);
1837 for (i = 0; i < vi->max_queue_pairs; i++)
1838 virtnet_napi_enable(&vi->rq[i]);
1841 netif_device_attach(vi->dev);
1843 mutex_lock(&vi->config_lock);
1844 vi->config_enable = true;
1845 mutex_unlock(&vi->config_lock);
1848 virtnet_set_queues(vi, vi->curr_queue_pairs);
1851 err = register_hotcpu_notifier(&vi->nb);
1859 static struct virtio_device_id id_table[] = {
1860 { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
1864 static unsigned int features[] = {
1865 VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
1866 VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
1867 VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
1868 VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
1869 VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO,
1870 VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ,
1871 VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN,
1872 VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ,
1873 VIRTIO_NET_F_CTRL_MAC_ADDR,
1874 VIRTIO_F_ANY_LAYOUT,
1877 static struct virtio_driver virtio_net_driver = {
1878 .feature_table = features,
1879 .feature_table_size = ARRAY_SIZE(features),
1880 .driver.name = KBUILD_MODNAME,
1881 .driver.owner = THIS_MODULE,
1882 .id_table = id_table,
1883 .probe = virtnet_probe,
1884 .remove = virtnet_remove,
1885 .config_changed = virtnet_config_changed,
1886 #ifdef CONFIG_PM_SLEEP
1887 .freeze = virtnet_freeze,
1888 .restore = virtnet_restore,
1892 module_virtio_driver(virtio_net_driver);
1894 MODULE_DEVICE_TABLE(virtio, id_table);
1895 MODULE_DESCRIPTION("Virtio network driver");
1896 MODULE_LICENSE("GPL");