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:
505 "host using disabled UFO feature; please fix it\n");
507 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
510 case VIRTIO_NET_HDR_GSO_TCPV6:
511 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
514 net_warn_ratelimited("%s: bad gso type %u.\n",
515 dev->name, hdr->hdr.gso_type);
519 if (hdr->hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
520 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
522 skb_shinfo(skb)->gso_size = hdr->hdr.gso_size;
523 if (skb_shinfo(skb)->gso_size == 0) {
524 net_warn_ratelimited("%s: zero gso size.\n", dev->name);
528 /* Header must be checked, and gso_segs computed. */
529 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
530 skb_shinfo(skb)->gso_segs = 0;
533 netif_receive_skb(skb);
537 dev->stats.rx_frame_errors++;
541 static int add_recvbuf_small(struct receive_queue *rq, gfp_t gfp)
543 struct virtnet_info *vi = rq->vq->vdev->priv;
545 struct skb_vnet_hdr *hdr;
548 skb = __netdev_alloc_skb_ip_align(vi->dev, GOOD_PACKET_LEN, gfp);
552 skb_put(skb, GOOD_PACKET_LEN);
554 hdr = skb_vnet_hdr(skb);
555 sg_set_buf(rq->sg, &hdr->hdr, sizeof hdr->hdr);
557 skb_to_sgvec(skb, rq->sg + 1, 0, skb->len);
559 err = virtqueue_add_inbuf(rq->vq, rq->sg, 2, skb, gfp);
566 static int add_recvbuf_big(struct receive_queue *rq, gfp_t gfp)
568 struct page *first, *list = NULL;
572 /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */
573 for (i = MAX_SKB_FRAGS + 1; i > 1; --i) {
574 first = get_a_page(rq, gfp);
577 give_pages(rq, list);
580 sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE);
582 /* chain new page in list head to match sg */
583 first->private = (unsigned long)list;
587 first = get_a_page(rq, gfp);
589 give_pages(rq, list);
592 p = page_address(first);
594 /* rq->sg[0], rq->sg[1] share the same page */
595 /* a separated rq->sg[0] for virtio_net_hdr only due to QEMU bug */
596 sg_set_buf(&rq->sg[0], p, sizeof(struct virtio_net_hdr));
598 /* rq->sg[1] for data packet, from offset */
599 offset = sizeof(struct padded_vnet_hdr);
600 sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset);
602 /* chain first in list head */
603 first->private = (unsigned long)list;
604 err = virtqueue_add_inbuf(rq->vq, rq->sg, MAX_SKB_FRAGS + 2,
607 give_pages(rq, first);
612 static unsigned int get_mergeable_buf_len(struct ewma *avg_pkt_len)
614 const size_t hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
617 len = hdr_len + clamp_t(unsigned int, ewma_read(avg_pkt_len),
618 GOOD_PACKET_LEN, PAGE_SIZE - hdr_len);
619 return ALIGN(len, MERGEABLE_BUFFER_ALIGN);
622 static int add_recvbuf_mergeable(struct receive_queue *rq, gfp_t gfp)
624 struct page_frag *alloc_frag = &rq->alloc_frag;
628 unsigned int len, hole;
630 len = get_mergeable_buf_len(&rq->mrg_avg_pkt_len);
631 if (unlikely(!skb_page_frag_refill(len, alloc_frag, gfp)))
634 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
635 ctx = mergeable_buf_to_ctx(buf, len);
636 get_page(alloc_frag->page);
637 alloc_frag->offset += len;
638 hole = alloc_frag->size - alloc_frag->offset;
640 /* To avoid internal fragmentation, if there is very likely not
641 * enough space for another buffer, add the remaining space to
642 * the current buffer. This extra space is not included in
643 * the truesize stored in ctx.
646 alloc_frag->offset += hole;
649 sg_init_one(rq->sg, buf, len);
650 err = virtqueue_add_inbuf(rq->vq, rq->sg, 1, (void *)ctx, gfp);
652 put_page(virt_to_head_page(buf));
658 * Returns false if we couldn't fill entirely (OOM).
660 * Normally run in the receive path, but can also be run from ndo_open
661 * before we're receiving packets, or from refill_work which is
662 * careful to disable receiving (using napi_disable).
664 static bool try_fill_recv(struct receive_queue *rq, gfp_t gfp)
666 struct virtnet_info *vi = rq->vq->vdev->priv;
672 if (vi->mergeable_rx_bufs)
673 err = add_recvbuf_mergeable(rq, gfp);
674 else if (vi->big_packets)
675 err = add_recvbuf_big(rq, gfp);
677 err = add_recvbuf_small(rq, gfp);
679 oom = err == -ENOMEM;
682 } while (rq->vq->num_free);
683 virtqueue_kick(rq->vq);
687 static void skb_recv_done(struct virtqueue *rvq)
689 struct virtnet_info *vi = rvq->vdev->priv;
690 struct receive_queue *rq = &vi->rq[vq2rxq(rvq)];
692 /* Schedule NAPI, Suppress further interrupts if successful. */
693 if (napi_schedule_prep(&rq->napi)) {
694 virtqueue_disable_cb(rvq);
695 __napi_schedule(&rq->napi);
699 static void virtnet_napi_enable(struct receive_queue *rq)
701 napi_enable(&rq->napi);
703 /* If all buffers were filled by other side before we napi_enabled, we
704 * won't get another interrupt, so process any outstanding packets
705 * now. virtnet_poll wants re-enable the queue, so we disable here.
706 * We synchronize against interrupts via NAPI_STATE_SCHED */
707 if (napi_schedule_prep(&rq->napi)) {
708 virtqueue_disable_cb(rq->vq);
710 __napi_schedule(&rq->napi);
715 static void refill_work(struct work_struct *work)
717 struct virtnet_info *vi =
718 container_of(work, struct virtnet_info, refill.work);
722 for (i = 0; i < vi->curr_queue_pairs; i++) {
723 struct receive_queue *rq = &vi->rq[i];
725 napi_disable(&rq->napi);
726 still_empty = !try_fill_recv(rq, GFP_KERNEL);
727 virtnet_napi_enable(rq);
729 /* In theory, this can happen: if we don't get any buffers in
730 * we will *never* try to fill again.
733 schedule_delayed_work(&vi->refill, HZ/2);
737 static int virtnet_poll(struct napi_struct *napi, int budget)
739 struct receive_queue *rq =
740 container_of(napi, struct receive_queue, napi);
741 struct virtnet_info *vi = rq->vq->vdev->priv;
743 unsigned int r, len, received = 0;
746 while (received < budget &&
747 (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) {
748 receive_buf(rq, buf, len);
752 if (rq->vq->num_free > virtqueue_get_vring_size(rq->vq) / 2) {
753 if (!try_fill_recv(rq, GFP_ATOMIC))
754 schedule_delayed_work(&vi->refill, 0);
757 /* Out of packets? */
758 if (received < budget) {
759 r = virtqueue_enable_cb_prepare(rq->vq);
761 if (unlikely(virtqueue_poll(rq->vq, r)) &&
762 napi_schedule_prep(napi)) {
763 virtqueue_disable_cb(rq->vq);
764 __napi_schedule(napi);
772 static int virtnet_open(struct net_device *dev)
774 struct virtnet_info *vi = netdev_priv(dev);
777 for (i = 0; i < vi->max_queue_pairs; i++) {
778 if (i < vi->curr_queue_pairs)
779 /* Make sure we have some buffers: if oom use wq. */
780 if (!try_fill_recv(&vi->rq[i], GFP_KERNEL))
781 schedule_delayed_work(&vi->refill, 0);
782 virtnet_napi_enable(&vi->rq[i]);
788 static void free_old_xmit_skbs(struct send_queue *sq)
792 struct virtnet_info *vi = sq->vq->vdev->priv;
793 struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
795 while ((skb = virtqueue_get_buf(sq->vq, &len)) != NULL) {
796 pr_debug("Sent skb %p\n", skb);
798 u64_stats_update_begin(&stats->tx_syncp);
799 stats->tx_bytes += skb->len;
801 u64_stats_update_end(&stats->tx_syncp);
803 dev_kfree_skb_any(skb);
807 static int xmit_skb(struct send_queue *sq, struct sk_buff *skb)
809 struct skb_vnet_hdr *hdr;
810 const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
811 struct virtnet_info *vi = sq->vq->vdev->priv;
816 pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
817 if (vi->mergeable_rx_bufs)
818 hdr_len = sizeof hdr->mhdr;
820 hdr_len = sizeof hdr->hdr;
822 can_push = vi->any_header_sg &&
823 !((unsigned long)skb->data & (__alignof__(*hdr) - 1)) &&
824 !skb_header_cloned(skb) && skb_headroom(skb) >= hdr_len;
825 /* Even if we can, don't push here yet as this would skew
826 * csum_start offset below. */
828 hdr = (struct skb_vnet_hdr *)(skb->data - hdr_len);
830 hdr = skb_vnet_hdr(skb);
832 if (skb->ip_summed == CHECKSUM_PARTIAL) {
833 hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
834 hdr->hdr.csum_start = skb_checksum_start_offset(skb);
835 hdr->hdr.csum_offset = skb->csum_offset;
838 hdr->hdr.csum_offset = hdr->hdr.csum_start = 0;
841 if (skb_is_gso(skb)) {
842 hdr->hdr.hdr_len = skb_headlen(skb);
843 hdr->hdr.gso_size = skb_shinfo(skb)->gso_size;
844 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
845 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
846 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
847 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
850 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
851 hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
853 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
854 hdr->hdr.gso_size = hdr->hdr.hdr_len = 0;
857 if (vi->mergeable_rx_bufs)
858 hdr->mhdr.num_buffers = 0;
861 __skb_push(skb, hdr_len);
862 num_sg = skb_to_sgvec(skb, sq->sg, 0, skb->len);
863 /* Pull header back to avoid skew in tx bytes calculations. */
864 __skb_pull(skb, hdr_len);
866 sg_set_buf(sq->sg, hdr, hdr_len);
867 num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len) + 1;
869 return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC);
872 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
874 struct virtnet_info *vi = netdev_priv(dev);
875 int qnum = skb_get_queue_mapping(skb);
876 struct send_queue *sq = &vi->sq[qnum];
879 /* Free up any pending old buffers before queueing new ones. */
880 free_old_xmit_skbs(sq);
882 /* Try to transmit */
883 err = xmit_skb(sq, skb);
885 /* This should not happen! */
887 dev->stats.tx_fifo_errors++;
890 "Unexpected TXQ (%d) queue failure: %d\n", qnum, err);
891 dev->stats.tx_dropped++;
895 virtqueue_kick(sq->vq);
897 /* Don't wait up for transmitted skbs to be freed. */
901 /* Apparently nice girls don't return TX_BUSY; stop the queue
902 * before it gets out of hand. Naturally, this wastes entries. */
903 if (sq->vq->num_free < 2+MAX_SKB_FRAGS) {
904 netif_stop_subqueue(dev, qnum);
905 if (unlikely(!virtqueue_enable_cb_delayed(sq->vq))) {
906 /* More just got used, free them then recheck. */
907 free_old_xmit_skbs(sq);
908 if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) {
909 netif_start_subqueue(dev, qnum);
910 virtqueue_disable_cb(sq->vq);
919 * Send command via the control virtqueue and check status. Commands
920 * supported by the hypervisor, as indicated by feature bits, should
921 * never fail unless improperly formatted.
923 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
924 struct scatterlist *out)
926 struct scatterlist *sgs[4], hdr, stat;
927 struct virtio_net_ctrl_hdr ctrl;
928 virtio_net_ctrl_ack status = ~0;
929 unsigned out_num = 0, tmp;
931 /* Caller should know better */
932 BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ));
937 sg_init_one(&hdr, &ctrl, sizeof(ctrl));
938 sgs[out_num++] = &hdr;
941 sgs[out_num++] = out;
943 /* Add return status. */
944 sg_init_one(&stat, &status, sizeof(status));
945 sgs[out_num] = &stat;
947 BUG_ON(out_num + 1 > ARRAY_SIZE(sgs));
948 BUG_ON(virtqueue_add_sgs(vi->cvq, sgs, out_num, 1, vi, GFP_ATOMIC) < 0);
950 if (unlikely(!virtqueue_kick(vi->cvq)))
951 return status == VIRTIO_NET_OK;
953 /* Spin for a response, the kick causes an ioport write, trapping
954 * into the hypervisor, so the request should be handled immediately.
956 while (!virtqueue_get_buf(vi->cvq, &tmp) &&
957 !virtqueue_is_broken(vi->cvq))
960 return status == VIRTIO_NET_OK;
963 static int virtnet_set_mac_address(struct net_device *dev, void *p)
965 struct virtnet_info *vi = netdev_priv(dev);
966 struct virtio_device *vdev = vi->vdev;
968 struct sockaddr *addr = p;
969 struct scatterlist sg;
971 ret = eth_prepare_mac_addr_change(dev, p);
975 if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
976 sg_init_one(&sg, addr->sa_data, dev->addr_len);
977 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
978 VIRTIO_NET_CTRL_MAC_ADDR_SET, &sg)) {
980 "Failed to set mac address by vq command.\n");
983 } else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
986 /* Naturally, this has an atomicity problem. */
987 for (i = 0; i < dev->addr_len; i++)
989 offsetof(struct virtio_net_config, mac) +
990 i, addr->sa_data[i]);
993 eth_commit_mac_addr_change(dev, p);
998 static struct rtnl_link_stats64 *virtnet_stats(struct net_device *dev,
999 struct rtnl_link_stats64 *tot)
1001 struct virtnet_info *vi = netdev_priv(dev);
1005 for_each_possible_cpu(cpu) {
1006 struct virtnet_stats *stats = per_cpu_ptr(vi->stats, cpu);
1007 u64 tpackets, tbytes, rpackets, rbytes;
1010 start = u64_stats_fetch_begin_bh(&stats->tx_syncp);
1011 tpackets = stats->tx_packets;
1012 tbytes = stats->tx_bytes;
1013 } while (u64_stats_fetch_retry_bh(&stats->tx_syncp, start));
1016 start = u64_stats_fetch_begin_bh(&stats->rx_syncp);
1017 rpackets = stats->rx_packets;
1018 rbytes = stats->rx_bytes;
1019 } while (u64_stats_fetch_retry_bh(&stats->rx_syncp, start));
1021 tot->rx_packets += rpackets;
1022 tot->tx_packets += tpackets;
1023 tot->rx_bytes += rbytes;
1024 tot->tx_bytes += tbytes;
1027 tot->tx_dropped = dev->stats.tx_dropped;
1028 tot->tx_fifo_errors = dev->stats.tx_fifo_errors;
1029 tot->rx_dropped = dev->stats.rx_dropped;
1030 tot->rx_length_errors = dev->stats.rx_length_errors;
1031 tot->rx_frame_errors = dev->stats.rx_frame_errors;
1036 #ifdef CONFIG_NET_POLL_CONTROLLER
1037 static void virtnet_netpoll(struct net_device *dev)
1039 struct virtnet_info *vi = netdev_priv(dev);
1042 for (i = 0; i < vi->curr_queue_pairs; i++)
1043 napi_schedule(&vi->rq[i].napi);
1047 static void virtnet_ack_link_announce(struct virtnet_info *vi)
1050 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE,
1051 VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL))
1052 dev_warn(&vi->dev->dev, "Failed to ack link announce.\n");
1056 static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs)
1058 struct scatterlist sg;
1059 struct virtio_net_ctrl_mq s;
1060 struct net_device *dev = vi->dev;
1062 if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ))
1065 s.virtqueue_pairs = queue_pairs;
1066 sg_init_one(&sg, &s, sizeof(s));
1068 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ,
1069 VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg)) {
1070 dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n",
1074 vi->curr_queue_pairs = queue_pairs;
1075 /* virtnet_open() will refill when device is going to up. */
1076 if (dev->flags & IFF_UP)
1077 schedule_delayed_work(&vi->refill, 0);
1083 static int virtnet_close(struct net_device *dev)
1085 struct virtnet_info *vi = netdev_priv(dev);
1088 /* Make sure refill_work doesn't re-enable napi! */
1089 cancel_delayed_work_sync(&vi->refill);
1091 for (i = 0; i < vi->max_queue_pairs; i++)
1092 napi_disable(&vi->rq[i].napi);
1097 static void virtnet_set_rx_mode(struct net_device *dev)
1099 struct virtnet_info *vi = netdev_priv(dev);
1100 struct scatterlist sg[2];
1101 u8 promisc, allmulti;
1102 struct virtio_net_ctrl_mac *mac_data;
1103 struct netdev_hw_addr *ha;
1109 /* We can't dynamically set ndo_set_rx_mode, so return gracefully */
1110 if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
1113 promisc = ((dev->flags & IFF_PROMISC) != 0);
1114 allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
1116 sg_init_one(sg, &promisc, sizeof(promisc));
1118 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
1119 VIRTIO_NET_CTRL_RX_PROMISC, sg))
1120 dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
1121 promisc ? "en" : "dis");
1123 sg_init_one(sg, &allmulti, sizeof(allmulti));
1125 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
1126 VIRTIO_NET_CTRL_RX_ALLMULTI, sg))
1127 dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
1128 allmulti ? "en" : "dis");
1130 uc_count = netdev_uc_count(dev);
1131 mc_count = netdev_mc_count(dev);
1132 /* MAC filter - use one buffer for both lists */
1133 buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
1134 (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
1139 sg_init_table(sg, 2);
1141 /* Store the unicast list and count in the front of the buffer */
1142 mac_data->entries = uc_count;
1144 netdev_for_each_uc_addr(ha, dev)
1145 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
1147 sg_set_buf(&sg[0], mac_data,
1148 sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
1150 /* multicast list and count fill the end */
1151 mac_data = (void *)&mac_data->macs[uc_count][0];
1153 mac_data->entries = mc_count;
1155 netdev_for_each_mc_addr(ha, dev)
1156 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
1158 sg_set_buf(&sg[1], mac_data,
1159 sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
1161 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
1162 VIRTIO_NET_CTRL_MAC_TABLE_SET, sg))
1163 dev_warn(&dev->dev, "Failed to set MAC filter table.\n");
1168 static int virtnet_vlan_rx_add_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_ADD, &sg))
1178 dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
1182 static int virtnet_vlan_rx_kill_vid(struct net_device *dev,
1183 __be16 proto, u16 vid)
1185 struct virtnet_info *vi = netdev_priv(dev);
1186 struct scatterlist sg;
1188 sg_init_one(&sg, &vid, sizeof(vid));
1190 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1191 VIRTIO_NET_CTRL_VLAN_DEL, &sg))
1192 dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
1196 static void virtnet_clean_affinity(struct virtnet_info *vi, long hcpu)
1200 if (vi->affinity_hint_set) {
1201 for (i = 0; i < vi->max_queue_pairs; i++) {
1202 virtqueue_set_affinity(vi->rq[i].vq, -1);
1203 virtqueue_set_affinity(vi->sq[i].vq, -1);
1206 vi->affinity_hint_set = false;
1210 static void virtnet_set_affinity(struct virtnet_info *vi)
1215 /* In multiqueue mode, when the number of cpu is equal to the number of
1216 * queue pairs, we let the queue pairs to be private to one cpu by
1217 * setting the affinity hint to eliminate the contention.
1219 if (vi->curr_queue_pairs == 1 ||
1220 vi->max_queue_pairs != num_online_cpus()) {
1221 virtnet_clean_affinity(vi, -1);
1226 for_each_online_cpu(cpu) {
1227 virtqueue_set_affinity(vi->rq[i].vq, cpu);
1228 virtqueue_set_affinity(vi->sq[i].vq, cpu);
1229 netif_set_xps_queue(vi->dev, cpumask_of(cpu), i);
1233 vi->affinity_hint_set = true;
1236 static int virtnet_cpu_callback(struct notifier_block *nfb,
1237 unsigned long action, void *hcpu)
1239 struct virtnet_info *vi = container_of(nfb, struct virtnet_info, nb);
1241 switch(action & ~CPU_TASKS_FROZEN) {
1243 case CPU_DOWN_FAILED:
1245 virtnet_set_affinity(vi);
1247 case CPU_DOWN_PREPARE:
1248 virtnet_clean_affinity(vi, (long)hcpu);
1257 static void virtnet_get_ringparam(struct net_device *dev,
1258 struct ethtool_ringparam *ring)
1260 struct virtnet_info *vi = netdev_priv(dev);
1262 ring->rx_max_pending = virtqueue_get_vring_size(vi->rq[0].vq);
1263 ring->tx_max_pending = virtqueue_get_vring_size(vi->sq[0].vq);
1264 ring->rx_pending = ring->rx_max_pending;
1265 ring->tx_pending = ring->tx_max_pending;
1269 static void virtnet_get_drvinfo(struct net_device *dev,
1270 struct ethtool_drvinfo *info)
1272 struct virtnet_info *vi = netdev_priv(dev);
1273 struct virtio_device *vdev = vi->vdev;
1275 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
1276 strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version));
1277 strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info));
1281 /* TODO: Eliminate OOO packets during switching */
1282 static int virtnet_set_channels(struct net_device *dev,
1283 struct ethtool_channels *channels)
1285 struct virtnet_info *vi = netdev_priv(dev);
1286 u16 queue_pairs = channels->combined_count;
1289 /* We don't support separate rx/tx channels.
1290 * We don't allow setting 'other' channels.
1292 if (channels->rx_count || channels->tx_count || channels->other_count)
1295 if (queue_pairs > vi->max_queue_pairs)
1299 err = virtnet_set_queues(vi, queue_pairs);
1301 netif_set_real_num_tx_queues(dev, queue_pairs);
1302 netif_set_real_num_rx_queues(dev, queue_pairs);
1304 virtnet_set_affinity(vi);
1311 static void virtnet_get_channels(struct net_device *dev,
1312 struct ethtool_channels *channels)
1314 struct virtnet_info *vi = netdev_priv(dev);
1316 channels->combined_count = vi->curr_queue_pairs;
1317 channels->max_combined = vi->max_queue_pairs;
1318 channels->max_other = 0;
1319 channels->rx_count = 0;
1320 channels->tx_count = 0;
1321 channels->other_count = 0;
1324 static const struct ethtool_ops virtnet_ethtool_ops = {
1325 .get_drvinfo = virtnet_get_drvinfo,
1326 .get_link = ethtool_op_get_link,
1327 .get_ringparam = virtnet_get_ringparam,
1328 .set_channels = virtnet_set_channels,
1329 .get_channels = virtnet_get_channels,
1333 #define MAX_MTU 65535
1335 static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
1337 if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
1343 static const struct net_device_ops virtnet_netdev = {
1344 .ndo_open = virtnet_open,
1345 .ndo_stop = virtnet_close,
1346 .ndo_start_xmit = start_xmit,
1347 .ndo_validate_addr = eth_validate_addr,
1348 .ndo_set_mac_address = virtnet_set_mac_address,
1349 .ndo_set_rx_mode = virtnet_set_rx_mode,
1350 .ndo_change_mtu = virtnet_change_mtu,
1351 .ndo_get_stats64 = virtnet_stats,
1352 .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
1353 .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
1354 #ifdef CONFIG_NET_POLL_CONTROLLER
1355 .ndo_poll_controller = virtnet_netpoll,
1359 static void virtnet_config_changed_work(struct work_struct *work)
1361 struct virtnet_info *vi =
1362 container_of(work, struct virtnet_info, config_work);
1365 mutex_lock(&vi->config_lock);
1366 if (!vi->config_enable)
1369 if (virtio_cread_feature(vi->vdev, VIRTIO_NET_F_STATUS,
1370 struct virtio_net_config, status, &v) < 0)
1373 if (v & VIRTIO_NET_S_ANNOUNCE) {
1374 netdev_notify_peers(vi->dev);
1375 virtnet_ack_link_announce(vi);
1378 /* Ignore unknown (future) status bits */
1379 v &= VIRTIO_NET_S_LINK_UP;
1381 if (vi->status == v)
1386 if (vi->status & VIRTIO_NET_S_LINK_UP) {
1387 netif_carrier_on(vi->dev);
1388 netif_tx_wake_all_queues(vi->dev);
1390 netif_carrier_off(vi->dev);
1391 netif_tx_stop_all_queues(vi->dev);
1394 mutex_unlock(&vi->config_lock);
1397 static void virtnet_config_changed(struct virtio_device *vdev)
1399 struct virtnet_info *vi = vdev->priv;
1401 schedule_work(&vi->config_work);
1404 static void virtnet_free_queues(struct virtnet_info *vi)
1408 for (i = 0; i < vi->max_queue_pairs; i++)
1409 netif_napi_del(&vi->rq[i].napi);
1415 static void free_receive_bufs(struct virtnet_info *vi)
1419 for (i = 0; i < vi->max_queue_pairs; i++) {
1420 while (vi->rq[i].pages)
1421 __free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0);
1425 static void free_receive_page_frags(struct virtnet_info *vi)
1428 for (i = 0; i < vi->max_queue_pairs; i++)
1429 if (vi->rq[i].alloc_frag.page)
1430 put_page(vi->rq[i].alloc_frag.page);
1433 static void free_unused_bufs(struct virtnet_info *vi)
1438 for (i = 0; i < vi->max_queue_pairs; i++) {
1439 struct virtqueue *vq = vi->sq[i].vq;
1440 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL)
1444 for (i = 0; i < vi->max_queue_pairs; i++) {
1445 struct virtqueue *vq = vi->rq[i].vq;
1447 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) {
1448 if (vi->mergeable_rx_bufs) {
1449 unsigned long ctx = (unsigned long)buf;
1450 void *base = mergeable_ctx_to_buf_address(ctx);
1451 put_page(virt_to_head_page(base));
1452 } else if (vi->big_packets) {
1453 give_pages(&vi->rq[i], buf);
1461 static void virtnet_del_vqs(struct virtnet_info *vi)
1463 struct virtio_device *vdev = vi->vdev;
1465 virtnet_clean_affinity(vi, -1);
1467 vdev->config->del_vqs(vdev);
1469 virtnet_free_queues(vi);
1472 static int virtnet_find_vqs(struct virtnet_info *vi)
1474 vq_callback_t **callbacks;
1475 struct virtqueue **vqs;
1480 /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
1481 * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
1482 * possible control vq.
1484 total_vqs = vi->max_queue_pairs * 2 +
1485 virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ);
1487 /* Allocate space for find_vqs parameters */
1488 vqs = kzalloc(total_vqs * sizeof(*vqs), GFP_KERNEL);
1491 callbacks = kmalloc(total_vqs * sizeof(*callbacks), GFP_KERNEL);
1494 names = kmalloc(total_vqs * sizeof(*names), GFP_KERNEL);
1498 /* Parameters for control virtqueue, if any */
1500 callbacks[total_vqs - 1] = NULL;
1501 names[total_vqs - 1] = "control";
1504 /* Allocate/initialize parameters for send/receive virtqueues */
1505 for (i = 0; i < vi->max_queue_pairs; i++) {
1506 callbacks[rxq2vq(i)] = skb_recv_done;
1507 callbacks[txq2vq(i)] = skb_xmit_done;
1508 sprintf(vi->rq[i].name, "input.%d", i);
1509 sprintf(vi->sq[i].name, "output.%d", i);
1510 names[rxq2vq(i)] = vi->rq[i].name;
1511 names[txq2vq(i)] = vi->sq[i].name;
1514 ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs, callbacks,
1520 vi->cvq = vqs[total_vqs - 1];
1521 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
1522 vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
1525 for (i = 0; i < vi->max_queue_pairs; i++) {
1526 vi->rq[i].vq = vqs[rxq2vq(i)];
1527 vi->sq[i].vq = vqs[txq2vq(i)];
1546 static int virtnet_alloc_queues(struct virtnet_info *vi)
1550 vi->sq = kzalloc(sizeof(*vi->sq) * vi->max_queue_pairs, GFP_KERNEL);
1553 vi->rq = kzalloc(sizeof(*vi->rq) * vi->max_queue_pairs, GFP_KERNEL);
1557 INIT_DELAYED_WORK(&vi->refill, refill_work);
1558 for (i = 0; i < vi->max_queue_pairs; i++) {
1559 vi->rq[i].pages = NULL;
1560 netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll,
1563 sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg));
1564 ewma_init(&vi->rq[i].mrg_avg_pkt_len, 1, RECEIVE_AVG_WEIGHT);
1565 sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg));
1576 static int init_vqs(struct virtnet_info *vi)
1580 /* Allocate send & receive queues */
1581 ret = virtnet_alloc_queues(vi);
1585 ret = virtnet_find_vqs(vi);
1590 virtnet_set_affinity(vi);
1596 virtnet_free_queues(vi);
1602 static ssize_t mergeable_rx_buffer_size_show(struct netdev_rx_queue *queue,
1603 struct rx_queue_attribute *attribute, char *buf)
1605 struct virtnet_info *vi = netdev_priv(queue->dev);
1606 unsigned int queue_index = get_netdev_rx_queue_index(queue);
1609 BUG_ON(queue_index >= vi->max_queue_pairs);
1610 avg = &vi->rq[queue_index].mrg_avg_pkt_len;
1611 return sprintf(buf, "%u\n", get_mergeable_buf_len(avg));
1614 static struct rx_queue_attribute mergeable_rx_buffer_size_attribute =
1615 __ATTR_RO(mergeable_rx_buffer_size);
1617 static struct attribute *virtio_net_mrg_rx_attrs[] = {
1618 &mergeable_rx_buffer_size_attribute.attr,
1622 static const struct attribute_group virtio_net_mrg_rx_group = {
1623 .name = "virtio_net",
1624 .attrs = virtio_net_mrg_rx_attrs
1628 static int virtnet_probe(struct virtio_device *vdev)
1631 struct net_device *dev;
1632 struct virtnet_info *vi;
1633 u16 max_queue_pairs;
1635 /* Find if host supports multiqueue virtio_net device */
1636 err = virtio_cread_feature(vdev, VIRTIO_NET_F_MQ,
1637 struct virtio_net_config,
1638 max_virtqueue_pairs, &max_queue_pairs);
1640 /* We need at least 2 queue's */
1641 if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
1642 max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
1643 !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
1644 max_queue_pairs = 1;
1646 /* Allocate ourselves a network device with room for our info */
1647 dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs);
1651 /* Set up network device as normal. */
1652 dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE;
1653 dev->netdev_ops = &virtnet_netdev;
1654 dev->features = NETIF_F_HIGHDMA;
1656 SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
1657 SET_NETDEV_DEV(dev, &vdev->dev);
1659 /* Do we support "hardware" checksums? */
1660 if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
1661 /* This opens up the world of extra features. */
1662 dev->hw_features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1664 dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1666 if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
1667 dev->hw_features |= NETIF_F_TSO
1668 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
1670 /* Individual feature bits: what can host handle? */
1671 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
1672 dev->hw_features |= NETIF_F_TSO;
1673 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
1674 dev->hw_features |= NETIF_F_TSO6;
1675 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
1676 dev->hw_features |= NETIF_F_TSO_ECN;
1679 dev->features |= dev->hw_features & NETIF_F_ALL_TSO;
1680 /* (!csum && gso) case will be fixed by register_netdev() */
1682 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM))
1683 dev->features |= NETIF_F_RXCSUM;
1685 dev->vlan_features = dev->features;
1687 /* Configuration may specify what MAC to use. Otherwise random. */
1688 if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
1689 virtio_cread_bytes(vdev,
1690 offsetof(struct virtio_net_config, mac),
1691 dev->dev_addr, dev->addr_len);
1693 eth_hw_addr_random(dev);
1695 /* Set up our device-specific information */
1696 vi = netdev_priv(dev);
1700 vi->stats = alloc_percpu(struct virtnet_stats);
1702 if (vi->stats == NULL)
1705 for_each_possible_cpu(i) {
1706 struct virtnet_stats *virtnet_stats;
1707 virtnet_stats = per_cpu_ptr(vi->stats, i);
1708 u64_stats_init(&virtnet_stats->tx_syncp);
1709 u64_stats_init(&virtnet_stats->rx_syncp);
1712 mutex_init(&vi->config_lock);
1713 vi->config_enable = true;
1714 INIT_WORK(&vi->config_work, virtnet_config_changed_work);
1716 /* If we can receive ANY GSO packets, we must allocate large ones. */
1717 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
1718 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
1719 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
1720 vi->big_packets = true;
1722 if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
1723 vi->mergeable_rx_bufs = true;
1725 if (virtio_has_feature(vdev, VIRTIO_F_ANY_LAYOUT))
1726 vi->any_header_sg = true;
1728 if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
1731 /* Use single tx/rx queue pair as default */
1732 vi->curr_queue_pairs = 1;
1733 vi->max_queue_pairs = max_queue_pairs;
1735 /* Allocate/initialize the rx/tx queues, and invoke find_vqs */
1741 if (vi->mergeable_rx_bufs)
1742 dev->sysfs_rx_queue_group = &virtio_net_mrg_rx_group;
1744 netif_set_real_num_tx_queues(dev, vi->curr_queue_pairs);
1745 netif_set_real_num_rx_queues(dev, vi->curr_queue_pairs);
1747 err = register_netdev(dev);
1749 pr_debug("virtio_net: registering device failed\n");
1753 /* Last of all, set up some receive buffers. */
1754 for (i = 0; i < vi->curr_queue_pairs; i++) {
1755 try_fill_recv(&vi->rq[i], GFP_KERNEL);
1757 /* If we didn't even get one input buffer, we're useless. */
1758 if (vi->rq[i].vq->num_free ==
1759 virtqueue_get_vring_size(vi->rq[i].vq)) {
1760 free_unused_bufs(vi);
1762 goto free_recv_bufs;
1766 vi->nb.notifier_call = &virtnet_cpu_callback;
1767 err = register_hotcpu_notifier(&vi->nb);
1769 pr_debug("virtio_net: registering cpu notifier failed\n");
1770 goto free_recv_bufs;
1773 /* Assume link up if device can't report link status,
1774 otherwise get link status from config. */
1775 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
1776 netif_carrier_off(dev);
1777 schedule_work(&vi->config_work);
1779 vi->status = VIRTIO_NET_S_LINK_UP;
1780 netif_carrier_on(dev);
1783 pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
1784 dev->name, max_queue_pairs);
1789 free_receive_bufs(vi);
1790 unregister_netdev(dev);
1792 cancel_delayed_work_sync(&vi->refill);
1793 free_receive_page_frags(vi);
1794 virtnet_del_vqs(vi);
1796 free_percpu(vi->stats);
1802 static void remove_vq_common(struct virtnet_info *vi)
1804 vi->vdev->config->reset(vi->vdev);
1806 /* Free unused buffers in both send and recv, if any. */
1807 free_unused_bufs(vi);
1809 free_receive_bufs(vi);
1811 free_receive_page_frags(vi);
1813 virtnet_del_vqs(vi);
1816 static void virtnet_remove(struct virtio_device *vdev)
1818 struct virtnet_info *vi = vdev->priv;
1820 unregister_hotcpu_notifier(&vi->nb);
1822 /* Prevent config work handler from accessing the device. */
1823 mutex_lock(&vi->config_lock);
1824 vi->config_enable = false;
1825 mutex_unlock(&vi->config_lock);
1827 unregister_netdev(vi->dev);
1829 remove_vq_common(vi);
1831 flush_work(&vi->config_work);
1833 free_percpu(vi->stats);
1834 free_netdev(vi->dev);
1837 #ifdef CONFIG_PM_SLEEP
1838 static int virtnet_freeze(struct virtio_device *vdev)
1840 struct virtnet_info *vi = vdev->priv;
1843 unregister_hotcpu_notifier(&vi->nb);
1845 /* Prevent config work handler from accessing the device */
1846 mutex_lock(&vi->config_lock);
1847 vi->config_enable = false;
1848 mutex_unlock(&vi->config_lock);
1850 netif_device_detach(vi->dev);
1851 cancel_delayed_work_sync(&vi->refill);
1853 if (netif_running(vi->dev))
1854 for (i = 0; i < vi->max_queue_pairs; i++) {
1855 napi_disable(&vi->rq[i].napi);
1856 netif_napi_del(&vi->rq[i].napi);
1859 remove_vq_common(vi);
1861 flush_work(&vi->config_work);
1866 static int virtnet_restore(struct virtio_device *vdev)
1868 struct virtnet_info *vi = vdev->priv;
1875 if (netif_running(vi->dev)) {
1876 for (i = 0; i < vi->curr_queue_pairs; i++)
1877 if (!try_fill_recv(&vi->rq[i], GFP_KERNEL))
1878 schedule_delayed_work(&vi->refill, 0);
1880 for (i = 0; i < vi->max_queue_pairs; i++)
1881 virtnet_napi_enable(&vi->rq[i]);
1884 netif_device_attach(vi->dev);
1886 mutex_lock(&vi->config_lock);
1887 vi->config_enable = true;
1888 mutex_unlock(&vi->config_lock);
1891 virtnet_set_queues(vi, vi->curr_queue_pairs);
1894 err = register_hotcpu_notifier(&vi->nb);
1902 static struct virtio_device_id id_table[] = {
1903 { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
1907 static unsigned int features[] = {
1908 VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
1909 VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
1910 VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_TSO6,
1911 VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
1912 VIRTIO_NET_F_GUEST_ECN,
1913 VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ,
1914 VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN,
1915 VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ,
1916 VIRTIO_NET_F_CTRL_MAC_ADDR,
1917 VIRTIO_F_ANY_LAYOUT,
1920 static struct virtio_driver virtio_net_driver = {
1921 .feature_table = features,
1922 .feature_table_size = ARRAY_SIZE(features),
1923 .driver.name = KBUILD_MODNAME,
1924 .driver.owner = THIS_MODULE,
1925 .id_table = id_table,
1926 .probe = virtnet_probe,
1927 .remove = virtnet_remove,
1928 .config_changed = virtnet_config_changed,
1929 #ifdef CONFIG_PM_SLEEP
1930 .freeze = virtnet_freeze,
1931 .restore = virtnet_restore,
1935 module_virtio_driver(virtio_net_driver);
1937 MODULE_DEVICE_TABLE(virtio, id_table);
1938 MODULE_DESCRIPTION("Virtio network driver");
1939 MODULE_LICENSE("GPL");