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/bpf.h>
26 #include <linux/bpf_trace.h>
27 #include <linux/scatterlist.h>
28 #include <linux/if_vlan.h>
29 #include <linux/slab.h>
30 #include <linux/cpu.h>
31 #include <linux/average.h>
32 #include <linux/filter.h>
33 #include <linux/kernel.h>
34 #include <linux/pci.h>
35 #include <net/route.h>
37 #include <net/net_failover.h>
39 static int napi_weight = NAPI_POLL_WEIGHT;
40 module_param(napi_weight, int, 0444);
42 static bool csum = true, gso = true, napi_tx;
43 module_param(csum, bool, 0444);
44 module_param(gso, bool, 0444);
45 module_param(napi_tx, bool, 0644);
47 /* FIXME: MTU in config. */
48 #define GOOD_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
49 #define GOOD_COPY_LEN 128
51 #define VIRTNET_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
53 /* Amount of XDP headroom to prepend to packets for use by xdp_adjust_head */
54 #define VIRTIO_XDP_HEADROOM 256
56 /* Separating two types of XDP xmit */
57 #define VIRTIO_XDP_TX BIT(0)
58 #define VIRTIO_XDP_REDIR BIT(1)
60 #define VIRTIO_XDP_FLAG BIT(0)
62 /* RX packet size EWMA. The average packet size is used to determine the packet
63 * buffer size when refilling RX rings. As the entire RX ring may be refilled
64 * at once, the weight is chosen so that the EWMA will be insensitive to short-
65 * term, transient changes in packet size.
67 DECLARE_EWMA(pkt_len, 0, 64)
69 #define VIRTNET_DRIVER_VERSION "1.0.0"
71 static const unsigned long guest_offloads[] = {
72 VIRTIO_NET_F_GUEST_TSO4,
73 VIRTIO_NET_F_GUEST_TSO6,
74 VIRTIO_NET_F_GUEST_ECN,
75 VIRTIO_NET_F_GUEST_UFO,
76 VIRTIO_NET_F_GUEST_CSUM
79 struct virtnet_stat_desc {
80 char desc[ETH_GSTRING_LEN];
84 struct virtnet_sq_stats {
85 struct u64_stats_sync syncp;
93 struct virtnet_rq_stats {
94 struct u64_stats_sync syncp;
105 #define VIRTNET_SQ_STAT(m) offsetof(struct virtnet_sq_stats, m)
106 #define VIRTNET_RQ_STAT(m) offsetof(struct virtnet_rq_stats, m)
108 static const struct virtnet_stat_desc virtnet_sq_stats_desc[] = {
109 { "packets", VIRTNET_SQ_STAT(packets) },
110 { "bytes", VIRTNET_SQ_STAT(bytes) },
111 { "xdp_tx", VIRTNET_SQ_STAT(xdp_tx) },
112 { "xdp_tx_drops", VIRTNET_SQ_STAT(xdp_tx_drops) },
113 { "kicks", VIRTNET_SQ_STAT(kicks) },
116 static const struct virtnet_stat_desc virtnet_rq_stats_desc[] = {
117 { "packets", VIRTNET_RQ_STAT(packets) },
118 { "bytes", VIRTNET_RQ_STAT(bytes) },
119 { "drops", VIRTNET_RQ_STAT(drops) },
120 { "xdp_packets", VIRTNET_RQ_STAT(xdp_packets) },
121 { "xdp_tx", VIRTNET_RQ_STAT(xdp_tx) },
122 { "xdp_redirects", VIRTNET_RQ_STAT(xdp_redirects) },
123 { "xdp_drops", VIRTNET_RQ_STAT(xdp_drops) },
124 { "kicks", VIRTNET_RQ_STAT(kicks) },
127 #define VIRTNET_SQ_STATS_LEN ARRAY_SIZE(virtnet_sq_stats_desc)
128 #define VIRTNET_RQ_STATS_LEN ARRAY_SIZE(virtnet_rq_stats_desc)
130 /* Internal representation of a send virtqueue */
132 /* Virtqueue associated with this send _queue */
133 struct virtqueue *vq;
135 /* TX: fragments + linear part + virtio header */
136 struct scatterlist sg[MAX_SKB_FRAGS + 2];
138 /* Name of the send queue: output.$index */
141 struct virtnet_sq_stats stats;
143 struct napi_struct napi;
146 /* Internal representation of a receive virtqueue */
147 struct receive_queue {
148 /* Virtqueue associated with this receive_queue */
149 struct virtqueue *vq;
151 struct napi_struct napi;
153 struct bpf_prog __rcu *xdp_prog;
155 struct virtnet_rq_stats stats;
157 /* Chain pages by the private ptr. */
160 /* Average packet length for mergeable receive buffers. */
161 struct ewma_pkt_len mrg_avg_pkt_len;
163 /* Page frag for packet buffer allocation. */
164 struct page_frag alloc_frag;
166 /* RX: fragments + linear part + virtio header */
167 struct scatterlist sg[MAX_SKB_FRAGS + 2];
169 /* Min single buffer size for mergeable buffers case. */
170 unsigned int min_buf_len;
172 /* Name of this receive queue: input.$index */
175 struct xdp_rxq_info xdp_rxq;
178 /* Control VQ buffers: protected by the rtnl lock */
180 struct virtio_net_ctrl_hdr hdr;
181 virtio_net_ctrl_ack status;
182 struct virtio_net_ctrl_mq mq;
189 struct virtnet_info {
190 struct virtio_device *vdev;
191 struct virtqueue *cvq;
192 struct net_device *dev;
193 struct send_queue *sq;
194 struct receive_queue *rq;
197 /* Max # of queue pairs supported by the device */
200 /* # of queue pairs currently used by the driver */
201 u16 curr_queue_pairs;
203 /* # of XDP queue pairs currently used by the driver */
206 /* I like... big packets and I cannot lie! */
209 /* Host will merge rx buffers for big packets (shake it! shake it!) */
210 bool mergeable_rx_bufs;
212 /* Has control virtqueue */
215 /* Host can handle any s/g split between our header and packet data */
218 /* Packet virtio header size */
221 /* Work struct for refilling if we run low on memory. */
222 struct delayed_work refill;
224 /* Work struct for config space updates */
225 struct work_struct config_work;
227 /* Does the affinity hint is set for virtqueues? */
228 bool affinity_hint_set;
230 /* CPU hotplug instances for online & dead */
231 struct hlist_node node;
232 struct hlist_node node_dead;
234 struct control_buf *ctrl;
236 /* Ethtool settings */
240 unsigned long guest_offloads;
242 /* failover when STANDBY feature enabled */
243 struct failover *failover;
246 struct padded_vnet_hdr {
247 struct virtio_net_hdr_mrg_rxbuf hdr;
249 * hdr is in a separate sg buffer, and data sg buffer shares same page
250 * with this header sg. This padding makes next sg 16 byte aligned
256 static bool is_xdp_frame(void *ptr)
258 return (unsigned long)ptr & VIRTIO_XDP_FLAG;
261 static void *xdp_to_ptr(struct xdp_frame *ptr)
263 return (void *)((unsigned long)ptr | VIRTIO_XDP_FLAG);
266 static struct xdp_frame *ptr_to_xdp(void *ptr)
268 return (struct xdp_frame *)((unsigned long)ptr & ~VIRTIO_XDP_FLAG);
271 /* Converting between virtqueue no. and kernel tx/rx queue no.
272 * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq
274 static int vq2txq(struct virtqueue *vq)
276 return (vq->index - 1) / 2;
279 static int txq2vq(int txq)
284 static int vq2rxq(struct virtqueue *vq)
286 return vq->index / 2;
289 static int rxq2vq(int rxq)
294 static inline struct virtio_net_hdr_mrg_rxbuf *skb_vnet_hdr(struct sk_buff *skb)
296 return (struct virtio_net_hdr_mrg_rxbuf *)skb->cb;
300 * private is used to chain pages for big packets, put the whole
301 * most recent used list in the beginning for reuse
303 static void give_pages(struct receive_queue *rq, struct page *page)
307 /* Find end of list, sew whole thing into vi->rq.pages. */
308 for (end = page; end->private; end = (struct page *)end->private);
309 end->private = (unsigned long)rq->pages;
313 static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask)
315 struct page *p = rq->pages;
318 rq->pages = (struct page *)p->private;
319 /* clear private here, it is used to chain pages */
322 p = alloc_page(gfp_mask);
326 static void virtqueue_napi_schedule(struct napi_struct *napi,
327 struct virtqueue *vq)
329 if (napi_schedule_prep(napi)) {
330 virtqueue_disable_cb(vq);
331 __napi_schedule(napi);
335 static void virtqueue_napi_complete(struct napi_struct *napi,
336 struct virtqueue *vq, int processed)
340 opaque = virtqueue_enable_cb_prepare(vq);
341 if (napi_complete_done(napi, processed)) {
342 if (unlikely(virtqueue_poll(vq, opaque)))
343 virtqueue_napi_schedule(napi, vq);
345 virtqueue_disable_cb(vq);
349 static void skb_xmit_done(struct virtqueue *vq)
351 struct virtnet_info *vi = vq->vdev->priv;
352 struct napi_struct *napi = &vi->sq[vq2txq(vq)].napi;
354 /* Suppress further interrupts. */
355 virtqueue_disable_cb(vq);
358 virtqueue_napi_schedule(napi, vq);
360 /* We were probably waiting for more output buffers. */
361 netif_wake_subqueue(vi->dev, vq2txq(vq));
364 #define MRG_CTX_HEADER_SHIFT 22
365 static void *mergeable_len_to_ctx(unsigned int truesize,
366 unsigned int headroom)
368 return (void *)(unsigned long)((headroom << MRG_CTX_HEADER_SHIFT) | truesize);
371 static unsigned int mergeable_ctx_to_headroom(void *mrg_ctx)
373 return (unsigned long)mrg_ctx >> MRG_CTX_HEADER_SHIFT;
376 static unsigned int mergeable_ctx_to_truesize(void *mrg_ctx)
378 return (unsigned long)mrg_ctx & ((1 << MRG_CTX_HEADER_SHIFT) - 1);
381 /* Called from bottom half context */
382 static struct sk_buff *page_to_skb(struct virtnet_info *vi,
383 struct receive_queue *rq,
384 struct page *page, unsigned int offset,
385 unsigned int len, unsigned int truesize,
386 bool hdr_valid, unsigned int metasize)
389 struct virtio_net_hdr_mrg_rxbuf *hdr;
390 unsigned int copy, hdr_len, hdr_padded_len;
393 p = page_address(page) + offset;
395 /* copy small packet so we can reuse these pages for small data */
396 skb = napi_alloc_skb(&rq->napi, GOOD_COPY_LEN);
400 hdr = skb_vnet_hdr(skb);
402 hdr_len = vi->hdr_len;
403 if (vi->mergeable_rx_bufs)
404 hdr_padded_len = sizeof(*hdr);
406 hdr_padded_len = sizeof(struct padded_vnet_hdr);
408 /* hdr_valid means no XDP, so we can copy the vnet header */
410 memcpy(hdr, p, hdr_len);
413 offset += hdr_padded_len;
416 /* Copy all frame if it fits skb->head, otherwise
417 * we let virtio_net_hdr_to_skb() and GRO pull headers as needed.
419 if (len <= skb_tailroom(skb))
422 copy = ETH_HLEN + metasize;
423 skb_put_data(skb, p, copy);
426 __skb_pull(skb, metasize);
427 skb_metadata_set(skb, metasize);
433 if (vi->mergeable_rx_bufs) {
435 skb_add_rx_frag(skb, 0, page, offset, len, truesize);
442 * Verify that we can indeed put this data into a skb.
443 * This is here to handle cases when the device erroneously
444 * tries to receive more than is possible. This is usually
445 * the case of a broken device.
447 if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) {
448 net_dbg_ratelimited("%s: too much data\n", skb->dev->name);
452 BUG_ON(offset >= PAGE_SIZE);
454 unsigned int frag_size = min((unsigned)PAGE_SIZE - offset, len);
455 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page, offset,
456 frag_size, truesize);
458 page = (struct page *)page->private;
463 give_pages(rq, page);
468 static int __virtnet_xdp_xmit_one(struct virtnet_info *vi,
469 struct send_queue *sq,
470 struct xdp_frame *xdpf)
472 struct virtio_net_hdr_mrg_rxbuf *hdr;
475 if (unlikely(xdpf->headroom < vi->hdr_len))
478 /* Make room for virtqueue hdr (also change xdpf->headroom?) */
479 xdpf->data -= vi->hdr_len;
480 /* Zero header and leave csum up to XDP layers */
482 memset(hdr, 0, vi->hdr_len);
483 xdpf->len += vi->hdr_len;
485 sg_init_one(sq->sg, xdpf->data, xdpf->len);
487 err = virtqueue_add_outbuf(sq->vq, sq->sg, 1, xdp_to_ptr(xdpf),
490 return -ENOSPC; /* Caller handle free/refcnt */
495 static struct send_queue *virtnet_xdp_sq(struct virtnet_info *vi)
499 qp = vi->curr_queue_pairs - vi->xdp_queue_pairs + smp_processor_id();
503 static int virtnet_xdp_xmit(struct net_device *dev,
504 int n, struct xdp_frame **frames, u32 flags)
506 struct virtnet_info *vi = netdev_priv(dev);
507 struct receive_queue *rq = vi->rq;
508 struct bpf_prog *xdp_prog;
509 struct send_queue *sq;
519 /* Only allow ndo_xdp_xmit if XDP is loaded on dev, as this
520 * indicate XDP resources have been successfully allocated.
522 xdp_prog = rcu_dereference(rq->xdp_prog);
526 sq = virtnet_xdp_sq(vi);
528 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK)) {
534 /* Free up any pending old buffers before queueing new ones. */
535 while ((ptr = virtqueue_get_buf(sq->vq, &len)) != NULL) {
536 if (likely(is_xdp_frame(ptr))) {
537 struct xdp_frame *frame = ptr_to_xdp(ptr);
540 xdp_return_frame(frame);
542 struct sk_buff *skb = ptr;
545 napi_consume_skb(skb, false);
550 for (i = 0; i < n; i++) {
551 struct xdp_frame *xdpf = frames[i];
553 err = __virtnet_xdp_xmit_one(vi, sq, xdpf);
555 xdp_return_frame_rx_napi(xdpf);
561 if (flags & XDP_XMIT_FLUSH) {
562 if (virtqueue_kick_prepare(sq->vq) && virtqueue_notify(sq->vq))
566 u64_stats_update_begin(&sq->stats.syncp);
567 sq->stats.bytes += bytes;
568 sq->stats.packets += packets;
569 sq->stats.xdp_tx += n;
570 sq->stats.xdp_tx_drops += drops;
571 sq->stats.kicks += kicks;
572 u64_stats_update_end(&sq->stats.syncp);
577 static unsigned int virtnet_get_headroom(struct virtnet_info *vi)
579 return vi->xdp_queue_pairs ? VIRTIO_XDP_HEADROOM : 0;
582 /* We copy the packet for XDP in the following cases:
584 * 1) Packet is scattered across multiple rx buffers.
585 * 2) Headroom space is insufficient.
587 * This is inefficient but it's a temporary condition that
588 * we hit right after XDP is enabled and until queue is refilled
589 * with large buffers with sufficient headroom - so it should affect
590 * at most queue size packets.
591 * Afterwards, the conditions to enable
592 * XDP should preclude the underlying device from sending packets
593 * across multiple buffers (num_buf > 1), and we make sure buffers
594 * have enough headroom.
596 static struct page *xdp_linearize_page(struct receive_queue *rq,
603 struct page *page = alloc_page(GFP_ATOMIC);
608 memcpy(page_address(page) + page_off, page_address(p) + offset, *len);
612 int tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
617 buf = virtqueue_get_buf(rq->vq, &buflen);
621 p = virt_to_head_page(buf);
622 off = buf - page_address(p);
624 /* guard against a misconfigured or uncooperative backend that
625 * is sending packet larger than the MTU.
627 if ((page_off + buflen + tailroom) > PAGE_SIZE) {
632 memcpy(page_address(page) + page_off,
633 page_address(p) + off, buflen);
638 /* Headroom does not contribute to packet length */
639 *len = page_off - VIRTIO_XDP_HEADROOM;
642 __free_pages(page, 0);
646 static struct sk_buff *receive_small(struct net_device *dev,
647 struct virtnet_info *vi,
648 struct receive_queue *rq,
649 void *buf, void *ctx,
651 unsigned int *xdp_xmit,
652 struct virtnet_rq_stats *stats)
655 struct bpf_prog *xdp_prog;
656 unsigned int xdp_headroom = (unsigned long)ctx;
657 unsigned int header_offset = VIRTNET_RX_PAD + xdp_headroom;
658 unsigned int headroom = vi->hdr_len + header_offset;
659 unsigned int buflen = SKB_DATA_ALIGN(GOOD_PACKET_LEN + headroom) +
660 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
661 struct page *page = virt_to_head_page(buf);
662 unsigned int delta = 0;
663 struct page *xdp_page;
665 unsigned int metasize = 0;
671 xdp_prog = rcu_dereference(rq->xdp_prog);
673 struct virtio_net_hdr_mrg_rxbuf *hdr = buf + header_offset;
674 struct xdp_frame *xdpf;
679 if (unlikely(hdr->hdr.gso_type))
682 if (unlikely(xdp_headroom < virtnet_get_headroom(vi))) {
683 int offset = buf - page_address(page) + header_offset;
684 unsigned int tlen = len + vi->hdr_len;
687 xdp_headroom = virtnet_get_headroom(vi);
688 header_offset = VIRTNET_RX_PAD + xdp_headroom;
689 headroom = vi->hdr_len + header_offset;
690 buflen = SKB_DATA_ALIGN(GOOD_PACKET_LEN + headroom) +
691 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
692 xdp_page = xdp_linearize_page(rq, &num_buf, page,
693 offset, header_offset,
698 buf = page_address(xdp_page);
703 xdp.data_hard_start = buf + VIRTNET_RX_PAD + vi->hdr_len;
704 xdp.data = xdp.data_hard_start + xdp_headroom;
705 xdp.data_end = xdp.data + len;
706 xdp.data_meta = xdp.data;
707 xdp.rxq = &rq->xdp_rxq;
708 orig_data = xdp.data;
709 act = bpf_prog_run_xdp(xdp_prog, &xdp);
710 stats->xdp_packets++;
714 /* Recalculate length in case bpf program changed it */
715 delta = orig_data - xdp.data;
716 len = xdp.data_end - xdp.data;
717 metasize = xdp.data - xdp.data_meta;
721 xdpf = convert_to_xdp_frame(&xdp);
724 err = virtnet_xdp_xmit(dev, 1, &xdpf, 0);
725 if (unlikely(err < 0)) {
726 trace_xdp_exception(vi->dev, xdp_prog, act);
729 *xdp_xmit |= VIRTIO_XDP_TX;
733 stats->xdp_redirects++;
734 err = xdp_do_redirect(dev, &xdp, xdp_prog);
737 *xdp_xmit |= VIRTIO_XDP_REDIR;
741 bpf_warn_invalid_xdp_action(act);
744 trace_xdp_exception(vi->dev, xdp_prog, act);
751 skb = build_skb(buf, buflen);
756 skb_reserve(skb, headroom - delta);
759 buf += header_offset;
760 memcpy(skb_vnet_hdr(skb), buf, vi->hdr_len);
761 } /* keep zeroed vnet hdr since packet was changed by bpf */
764 skb_metadata_set(skb, metasize);
778 static struct sk_buff *receive_big(struct net_device *dev,
779 struct virtnet_info *vi,
780 struct receive_queue *rq,
783 struct virtnet_rq_stats *stats)
785 struct page *page = buf;
786 struct sk_buff *skb =
787 page_to_skb(vi, rq, page, 0, len, PAGE_SIZE, true, 0);
789 stats->bytes += len - vi->hdr_len;
797 give_pages(rq, page);
801 static struct sk_buff *receive_mergeable(struct net_device *dev,
802 struct virtnet_info *vi,
803 struct receive_queue *rq,
807 unsigned int *xdp_xmit,
808 struct virtnet_rq_stats *stats)
810 struct virtio_net_hdr_mrg_rxbuf *hdr = buf;
811 u16 num_buf = virtio16_to_cpu(vi->vdev, hdr->num_buffers);
812 struct page *page = virt_to_head_page(buf);
813 int offset = buf - page_address(page);
814 struct sk_buff *head_skb, *curr_skb;
815 struct bpf_prog *xdp_prog;
816 unsigned int truesize;
817 unsigned int headroom = mergeable_ctx_to_headroom(ctx);
819 unsigned int metasize = 0;
822 stats->bytes += len - vi->hdr_len;
825 xdp_prog = rcu_dereference(rq->xdp_prog);
827 struct xdp_frame *xdpf;
828 struct page *xdp_page;
833 /* Transient failure which in theory could occur if
834 * in-flight packets from before XDP was enabled reach
835 * the receive path after XDP is loaded.
837 if (unlikely(hdr->hdr.gso_type))
840 /* This happens when rx buffer size is underestimated
841 * or headroom is not enough because of the buffer
842 * was refilled before XDP is set. This should only
843 * happen for the first several packets, so we don't
844 * care much about its performance.
846 if (unlikely(num_buf > 1 ||
847 headroom < virtnet_get_headroom(vi))) {
848 /* linearize data for XDP */
849 xdp_page = xdp_linearize_page(rq, &num_buf,
855 offset = VIRTIO_XDP_HEADROOM;
860 /* Allow consuming headroom but reserve enough space to push
861 * the descriptor on if we get an XDP_TX return code.
863 data = page_address(xdp_page) + offset;
864 xdp.data_hard_start = data - VIRTIO_XDP_HEADROOM + vi->hdr_len;
865 xdp.data = data + vi->hdr_len;
866 xdp.data_end = xdp.data + (len - vi->hdr_len);
867 xdp.data_meta = xdp.data;
868 xdp.rxq = &rq->xdp_rxq;
870 act = bpf_prog_run_xdp(xdp_prog, &xdp);
871 stats->xdp_packets++;
875 metasize = xdp.data - xdp.data_meta;
877 /* recalculate offset to account for any header
878 * adjustments and minus the metasize to copy the
879 * metadata in page_to_skb(). Note other cases do not
880 * build an skb and avoid using offset
882 offset = xdp.data - page_address(xdp_page) -
883 vi->hdr_len - metasize;
885 /* recalculate len if xdp.data, xdp.data_end or
886 * xdp.data_meta were adjusted
888 len = xdp.data_end - xdp.data + vi->hdr_len + metasize;
889 /* We can only create skb based on xdp_page. */
890 if (unlikely(xdp_page != page)) {
893 head_skb = page_to_skb(vi, rq, xdp_page, offset,
894 len, PAGE_SIZE, false,
901 xdpf = convert_to_xdp_frame(&xdp);
904 err = virtnet_xdp_xmit(dev, 1, &xdpf, 0);
905 if (unlikely(err < 0)) {
906 trace_xdp_exception(vi->dev, xdp_prog, act);
907 if (unlikely(xdp_page != page))
911 *xdp_xmit |= VIRTIO_XDP_TX;
912 if (unlikely(xdp_page != page))
917 stats->xdp_redirects++;
918 err = xdp_do_redirect(dev, &xdp, xdp_prog);
920 if (unlikely(xdp_page != page))
924 *xdp_xmit |= VIRTIO_XDP_REDIR;
925 if (unlikely(xdp_page != page))
930 bpf_warn_invalid_xdp_action(act);
933 trace_xdp_exception(vi->dev, xdp_prog, act);
936 if (unlikely(xdp_page != page))
937 __free_pages(xdp_page, 0);
943 truesize = mergeable_ctx_to_truesize(ctx);
944 if (unlikely(len > truesize)) {
945 pr_debug("%s: rx error: len %u exceeds truesize %lu\n",
946 dev->name, len, (unsigned long)ctx);
947 dev->stats.rx_length_errors++;
951 head_skb = page_to_skb(vi, rq, page, offset, len, truesize, !xdp_prog,
955 if (unlikely(!curr_skb))
960 buf = virtqueue_get_buf_ctx(rq->vq, &len, &ctx);
961 if (unlikely(!buf)) {
962 pr_debug("%s: rx error: %d buffers out of %d missing\n",
964 virtio16_to_cpu(vi->vdev,
966 dev->stats.rx_length_errors++;
971 page = virt_to_head_page(buf);
973 truesize = mergeable_ctx_to_truesize(ctx);
974 if (unlikely(len > truesize)) {
975 pr_debug("%s: rx error: len %u exceeds truesize %lu\n",
976 dev->name, len, (unsigned long)ctx);
977 dev->stats.rx_length_errors++;
981 num_skb_frags = skb_shinfo(curr_skb)->nr_frags;
982 if (unlikely(num_skb_frags == MAX_SKB_FRAGS)) {
983 struct sk_buff *nskb = alloc_skb(0, GFP_ATOMIC);
987 if (curr_skb == head_skb)
988 skb_shinfo(curr_skb)->frag_list = nskb;
990 curr_skb->next = nskb;
992 head_skb->truesize += nskb->truesize;
995 if (curr_skb != head_skb) {
996 head_skb->data_len += len;
997 head_skb->len += len;
998 head_skb->truesize += truesize;
1000 offset = buf - page_address(page);
1001 if (skb_can_coalesce(curr_skb, num_skb_frags, page, offset)) {
1003 skb_coalesce_rx_frag(curr_skb, num_skb_frags - 1,
1006 skb_add_rx_frag(curr_skb, num_skb_frags, page,
1007 offset, len, truesize);
1011 ewma_pkt_len_add(&rq->mrg_avg_pkt_len, head_skb->len);
1019 while (num_buf-- > 1) {
1020 buf = virtqueue_get_buf(rq->vq, &len);
1021 if (unlikely(!buf)) {
1022 pr_debug("%s: rx error: %d buffers missing\n",
1023 dev->name, num_buf);
1024 dev->stats.rx_length_errors++;
1027 stats->bytes += len;
1028 page = virt_to_head_page(buf);
1033 dev_kfree_skb(head_skb);
1038 static void receive_buf(struct virtnet_info *vi, struct receive_queue *rq,
1039 void *buf, unsigned int len, void **ctx,
1040 unsigned int *xdp_xmit,
1041 struct virtnet_rq_stats *stats)
1043 struct net_device *dev = vi->dev;
1044 struct sk_buff *skb;
1045 struct virtio_net_hdr_mrg_rxbuf *hdr;
1047 if (unlikely(len < vi->hdr_len + ETH_HLEN)) {
1048 pr_debug("%s: short packet %i\n", dev->name, len);
1049 dev->stats.rx_length_errors++;
1050 if (vi->mergeable_rx_bufs) {
1051 put_page(virt_to_head_page(buf));
1052 } else if (vi->big_packets) {
1053 give_pages(rq, buf);
1055 put_page(virt_to_head_page(buf));
1060 if (vi->mergeable_rx_bufs)
1061 skb = receive_mergeable(dev, vi, rq, buf, ctx, len, xdp_xmit,
1063 else if (vi->big_packets)
1064 skb = receive_big(dev, vi, rq, buf, len, stats);
1066 skb = receive_small(dev, vi, rq, buf, ctx, len, xdp_xmit, stats);
1071 hdr = skb_vnet_hdr(skb);
1073 if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID)
1074 skb->ip_summed = CHECKSUM_UNNECESSARY;
1076 if (virtio_net_hdr_to_skb(skb, &hdr->hdr,
1077 virtio_is_little_endian(vi->vdev))) {
1078 net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n",
1079 dev->name, hdr->hdr.gso_type,
1084 skb->protocol = eth_type_trans(skb, dev);
1085 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
1086 ntohs(skb->protocol), skb->len, skb->pkt_type);
1088 napi_gro_receive(&rq->napi, skb);
1092 dev->stats.rx_frame_errors++;
1096 /* Unlike mergeable buffers, all buffers are allocated to the
1097 * same size, except for the headroom. For this reason we do
1098 * not need to use mergeable_len_to_ctx here - it is enough
1099 * to store the headroom as the context ignoring the truesize.
1101 static int add_recvbuf_small(struct virtnet_info *vi, struct receive_queue *rq,
1104 struct page_frag *alloc_frag = &rq->alloc_frag;
1106 unsigned int xdp_headroom = virtnet_get_headroom(vi);
1107 void *ctx = (void *)(unsigned long)xdp_headroom;
1108 int len = vi->hdr_len + VIRTNET_RX_PAD + GOOD_PACKET_LEN + xdp_headroom;
1111 len = SKB_DATA_ALIGN(len) +
1112 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1113 if (unlikely(!skb_page_frag_refill(len, alloc_frag, gfp)))
1116 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1117 get_page(alloc_frag->page);
1118 alloc_frag->offset += len;
1119 sg_init_one(rq->sg, buf + VIRTNET_RX_PAD + xdp_headroom,
1120 vi->hdr_len + GOOD_PACKET_LEN);
1121 err = virtqueue_add_inbuf_ctx(rq->vq, rq->sg, 1, buf, ctx, gfp);
1123 put_page(virt_to_head_page(buf));
1127 static int add_recvbuf_big(struct virtnet_info *vi, struct receive_queue *rq,
1130 struct page *first, *list = NULL;
1134 sg_init_table(rq->sg, MAX_SKB_FRAGS + 2);
1136 /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */
1137 for (i = MAX_SKB_FRAGS + 1; i > 1; --i) {
1138 first = get_a_page(rq, gfp);
1141 give_pages(rq, list);
1144 sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE);
1146 /* chain new page in list head to match sg */
1147 first->private = (unsigned long)list;
1151 first = get_a_page(rq, gfp);
1153 give_pages(rq, list);
1156 p = page_address(first);
1158 /* rq->sg[0], rq->sg[1] share the same page */
1159 /* a separated rq->sg[0] for header - required in case !any_header_sg */
1160 sg_set_buf(&rq->sg[0], p, vi->hdr_len);
1162 /* rq->sg[1] for data packet, from offset */
1163 offset = sizeof(struct padded_vnet_hdr);
1164 sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset);
1166 /* chain first in list head */
1167 first->private = (unsigned long)list;
1168 err = virtqueue_add_inbuf(rq->vq, rq->sg, MAX_SKB_FRAGS + 2,
1171 give_pages(rq, first);
1176 static unsigned int get_mergeable_buf_len(struct receive_queue *rq,
1177 struct ewma_pkt_len *avg_pkt_len,
1180 const size_t hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1184 return PAGE_SIZE - room;
1186 len = hdr_len + clamp_t(unsigned int, ewma_pkt_len_read(avg_pkt_len),
1187 rq->min_buf_len, PAGE_SIZE - hdr_len);
1189 return ALIGN(len, L1_CACHE_BYTES);
1192 static int add_recvbuf_mergeable(struct virtnet_info *vi,
1193 struct receive_queue *rq, gfp_t gfp)
1195 struct page_frag *alloc_frag = &rq->alloc_frag;
1196 unsigned int headroom = virtnet_get_headroom(vi);
1197 unsigned int tailroom = headroom ? sizeof(struct skb_shared_info) : 0;
1198 unsigned int room = SKB_DATA_ALIGN(headroom + tailroom);
1202 unsigned int len, hole;
1204 /* Extra tailroom is needed to satisfy XDP's assumption. This
1205 * means rx frags coalescing won't work, but consider we've
1206 * disabled GSO for XDP, it won't be a big issue.
1208 len = get_mergeable_buf_len(rq, &rq->mrg_avg_pkt_len, room);
1209 if (unlikely(!skb_page_frag_refill(len + room, alloc_frag, gfp)))
1212 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1213 buf += headroom; /* advance address leaving hole at front of pkt */
1214 get_page(alloc_frag->page);
1215 alloc_frag->offset += len + room;
1216 hole = alloc_frag->size - alloc_frag->offset;
1217 if (hole < len + room) {
1218 /* To avoid internal fragmentation, if there is very likely not
1219 * enough space for another buffer, add the remaining space to
1220 * the current buffer.
1223 alloc_frag->offset += hole;
1226 sg_init_one(rq->sg, buf, len);
1227 ctx = mergeable_len_to_ctx(len, headroom);
1228 err = virtqueue_add_inbuf_ctx(rq->vq, rq->sg, 1, buf, ctx, gfp);
1230 put_page(virt_to_head_page(buf));
1236 * Returns false if we couldn't fill entirely (OOM).
1238 * Normally run in the receive path, but can also be run from ndo_open
1239 * before we're receiving packets, or from refill_work which is
1240 * careful to disable receiving (using napi_disable).
1242 static bool try_fill_recv(struct virtnet_info *vi, struct receive_queue *rq,
1249 if (vi->mergeable_rx_bufs)
1250 err = add_recvbuf_mergeable(vi, rq, gfp);
1251 else if (vi->big_packets)
1252 err = add_recvbuf_big(vi, rq, gfp);
1254 err = add_recvbuf_small(vi, rq, gfp);
1256 oom = err == -ENOMEM;
1259 } while (rq->vq->num_free);
1260 if (virtqueue_kick_prepare(rq->vq) && virtqueue_notify(rq->vq)) {
1261 unsigned long flags;
1263 flags = u64_stats_update_begin_irqsave(&rq->stats.syncp);
1265 u64_stats_update_end_irqrestore(&rq->stats.syncp, flags);
1271 static void skb_recv_done(struct virtqueue *rvq)
1273 struct virtnet_info *vi = rvq->vdev->priv;
1274 struct receive_queue *rq = &vi->rq[vq2rxq(rvq)];
1276 virtqueue_napi_schedule(&rq->napi, rvq);
1279 static void virtnet_napi_enable(struct virtqueue *vq, struct napi_struct *napi)
1283 /* If all buffers were filled by other side before we napi_enabled, we
1284 * won't get another interrupt, so process any outstanding packets now.
1285 * Call local_bh_enable after to trigger softIRQ processing.
1288 virtqueue_napi_schedule(napi, vq);
1292 static void virtnet_napi_tx_enable(struct virtnet_info *vi,
1293 struct virtqueue *vq,
1294 struct napi_struct *napi)
1299 /* Tx napi touches cachelines on the cpu handling tx interrupts. Only
1300 * enable the feature if this is likely affine with the transmit path.
1302 if (!vi->affinity_hint_set) {
1307 return virtnet_napi_enable(vq, napi);
1310 static void virtnet_napi_tx_disable(struct napi_struct *napi)
1316 static void refill_work(struct work_struct *work)
1318 struct virtnet_info *vi =
1319 container_of(work, struct virtnet_info, refill.work);
1323 for (i = 0; i < vi->curr_queue_pairs; i++) {
1324 struct receive_queue *rq = &vi->rq[i];
1326 napi_disable(&rq->napi);
1327 still_empty = !try_fill_recv(vi, rq, GFP_KERNEL);
1328 virtnet_napi_enable(rq->vq, &rq->napi);
1330 /* In theory, this can happen: if we don't get any buffers in
1331 * we will *never* try to fill again.
1334 schedule_delayed_work(&vi->refill, HZ/2);
1338 static int virtnet_receive(struct receive_queue *rq, int budget,
1339 unsigned int *xdp_xmit)
1341 struct virtnet_info *vi = rq->vq->vdev->priv;
1342 struct virtnet_rq_stats stats = {};
1347 if (!vi->big_packets || vi->mergeable_rx_bufs) {
1350 while (stats.packets < budget &&
1351 (buf = virtqueue_get_buf_ctx(rq->vq, &len, &ctx))) {
1352 receive_buf(vi, rq, buf, len, ctx, xdp_xmit, &stats);
1356 while (stats.packets < budget &&
1357 (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) {
1358 receive_buf(vi, rq, buf, len, NULL, xdp_xmit, &stats);
1363 if (rq->vq->num_free > virtqueue_get_vring_size(rq->vq) / 2) {
1364 if (!try_fill_recv(vi, rq, GFP_ATOMIC))
1365 schedule_delayed_work(&vi->refill, 0);
1368 u64_stats_update_begin(&rq->stats.syncp);
1369 for (i = 0; i < VIRTNET_RQ_STATS_LEN; i++) {
1370 size_t offset = virtnet_rq_stats_desc[i].offset;
1373 item = (u64 *)((u8 *)&rq->stats + offset);
1374 *item += *(u64 *)((u8 *)&stats + offset);
1376 u64_stats_update_end(&rq->stats.syncp);
1378 return stats.packets;
1381 static void free_old_xmit_skbs(struct send_queue *sq, bool in_napi)
1384 unsigned int packets = 0;
1385 unsigned int bytes = 0;
1388 while ((ptr = virtqueue_get_buf(sq->vq, &len)) != NULL) {
1389 if (likely(!is_xdp_frame(ptr))) {
1390 struct sk_buff *skb = ptr;
1392 pr_debug("Sent skb %p\n", skb);
1395 napi_consume_skb(skb, in_napi);
1397 struct xdp_frame *frame = ptr_to_xdp(ptr);
1399 bytes += frame->len;
1400 xdp_return_frame(frame);
1405 /* Avoid overhead when no packets have been processed
1406 * happens when called speculatively from start_xmit.
1411 u64_stats_update_begin(&sq->stats.syncp);
1412 sq->stats.bytes += bytes;
1413 sq->stats.packets += packets;
1414 u64_stats_update_end(&sq->stats.syncp);
1417 static bool is_xdp_raw_buffer_queue(struct virtnet_info *vi, int q)
1419 if (q < (vi->curr_queue_pairs - vi->xdp_queue_pairs))
1421 else if (q < vi->curr_queue_pairs)
1427 static void virtnet_poll_cleantx(struct receive_queue *rq)
1429 struct virtnet_info *vi = rq->vq->vdev->priv;
1430 unsigned int index = vq2rxq(rq->vq);
1431 struct send_queue *sq = &vi->sq[index];
1432 struct netdev_queue *txq = netdev_get_tx_queue(vi->dev, index);
1434 if (!sq->napi.weight || is_xdp_raw_buffer_queue(vi, index))
1437 if (__netif_tx_trylock(txq)) {
1438 free_old_xmit_skbs(sq, true);
1439 __netif_tx_unlock(txq);
1442 if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS)
1443 netif_tx_wake_queue(txq);
1446 static int virtnet_poll(struct napi_struct *napi, int budget)
1448 struct receive_queue *rq =
1449 container_of(napi, struct receive_queue, napi);
1450 struct virtnet_info *vi = rq->vq->vdev->priv;
1451 struct send_queue *sq;
1452 unsigned int received;
1453 unsigned int xdp_xmit = 0;
1455 virtnet_poll_cleantx(rq);
1457 received = virtnet_receive(rq, budget, &xdp_xmit);
1459 /* Out of packets? */
1460 if (received < budget)
1461 virtqueue_napi_complete(napi, rq->vq, received);
1463 if (xdp_xmit & VIRTIO_XDP_REDIR)
1466 if (xdp_xmit & VIRTIO_XDP_TX) {
1467 sq = virtnet_xdp_sq(vi);
1468 if (virtqueue_kick_prepare(sq->vq) && virtqueue_notify(sq->vq)) {
1469 u64_stats_update_begin(&sq->stats.syncp);
1471 u64_stats_update_end(&sq->stats.syncp);
1478 static int virtnet_open(struct net_device *dev)
1480 struct virtnet_info *vi = netdev_priv(dev);
1483 for (i = 0; i < vi->max_queue_pairs; i++) {
1484 if (i < vi->curr_queue_pairs)
1485 /* Make sure we have some buffers: if oom use wq. */
1486 if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL))
1487 schedule_delayed_work(&vi->refill, 0);
1489 err = xdp_rxq_info_reg(&vi->rq[i].xdp_rxq, dev, i);
1493 err = xdp_rxq_info_reg_mem_model(&vi->rq[i].xdp_rxq,
1494 MEM_TYPE_PAGE_SHARED, NULL);
1496 xdp_rxq_info_unreg(&vi->rq[i].xdp_rxq);
1500 virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi);
1501 virtnet_napi_tx_enable(vi, vi->sq[i].vq, &vi->sq[i].napi);
1507 static int virtnet_poll_tx(struct napi_struct *napi, int budget)
1509 struct send_queue *sq = container_of(napi, struct send_queue, napi);
1510 struct virtnet_info *vi = sq->vq->vdev->priv;
1511 unsigned int index = vq2txq(sq->vq);
1512 struct netdev_queue *txq;
1516 if (unlikely(is_xdp_raw_buffer_queue(vi, index))) {
1517 /* We don't need to enable cb for XDP */
1518 napi_complete_done(napi, 0);
1522 txq = netdev_get_tx_queue(vi->dev, index);
1523 __netif_tx_lock(txq, raw_smp_processor_id());
1524 virtqueue_disable_cb(sq->vq);
1525 free_old_xmit_skbs(sq, true);
1527 opaque = virtqueue_enable_cb_prepare(sq->vq);
1529 done = napi_complete_done(napi, 0);
1532 virtqueue_disable_cb(sq->vq);
1534 __netif_tx_unlock(txq);
1537 if (unlikely(virtqueue_poll(sq->vq, opaque))) {
1538 if (napi_schedule_prep(napi)) {
1539 __netif_tx_lock(txq, raw_smp_processor_id());
1540 virtqueue_disable_cb(sq->vq);
1541 __netif_tx_unlock(txq);
1542 __napi_schedule(napi);
1547 if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS)
1548 netif_tx_wake_queue(txq);
1553 static int xmit_skb(struct send_queue *sq, struct sk_buff *skb)
1555 struct virtio_net_hdr_mrg_rxbuf *hdr;
1556 const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
1557 struct virtnet_info *vi = sq->vq->vdev->priv;
1559 unsigned hdr_len = vi->hdr_len;
1562 pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
1564 can_push = vi->any_header_sg &&
1565 !((unsigned long)skb->data & (__alignof__(*hdr) - 1)) &&
1566 !skb_header_cloned(skb) && skb_headroom(skb) >= hdr_len;
1567 /* Even if we can, don't push here yet as this would skew
1568 * csum_start offset below. */
1570 hdr = (struct virtio_net_hdr_mrg_rxbuf *)(skb->data - hdr_len);
1572 hdr = skb_vnet_hdr(skb);
1574 if (virtio_net_hdr_from_skb(skb, &hdr->hdr,
1575 virtio_is_little_endian(vi->vdev), false,
1579 if (vi->mergeable_rx_bufs)
1580 hdr->num_buffers = 0;
1582 sg_init_table(sq->sg, skb_shinfo(skb)->nr_frags + (can_push ? 1 : 2));
1584 __skb_push(skb, hdr_len);
1585 num_sg = skb_to_sgvec(skb, sq->sg, 0, skb->len);
1586 if (unlikely(num_sg < 0))
1588 /* Pull header back to avoid skew in tx bytes calculations. */
1589 __skb_pull(skb, hdr_len);
1591 sg_set_buf(sq->sg, hdr, hdr_len);
1592 num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len);
1593 if (unlikely(num_sg < 0))
1597 return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC);
1600 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
1602 struct virtnet_info *vi = netdev_priv(dev);
1603 int qnum = skb_get_queue_mapping(skb);
1604 struct send_queue *sq = &vi->sq[qnum];
1606 struct netdev_queue *txq = netdev_get_tx_queue(dev, qnum);
1607 bool kick = !skb->xmit_more;
1608 bool use_napi = sq->napi.weight;
1610 /* Free up any pending old buffers before queueing new ones. */
1611 free_old_xmit_skbs(sq, false);
1613 if (use_napi && kick)
1614 virtqueue_enable_cb_delayed(sq->vq);
1616 /* timestamp packet in software */
1617 skb_tx_timestamp(skb);
1619 /* Try to transmit */
1620 err = xmit_skb(sq, skb);
1622 /* This should not happen! */
1623 if (unlikely(err)) {
1624 dev->stats.tx_fifo_errors++;
1625 if (net_ratelimit())
1627 "Unexpected TXQ (%d) queue failure: %d\n", qnum, err);
1628 dev->stats.tx_dropped++;
1629 dev_kfree_skb_any(skb);
1630 return NETDEV_TX_OK;
1633 /* Don't wait up for transmitted skbs to be freed. */
1639 /* If running out of space, stop queue to avoid getting packets that we
1640 * are then unable to transmit.
1641 * An alternative would be to force queuing layer to requeue the skb by
1642 * returning NETDEV_TX_BUSY. However, NETDEV_TX_BUSY should not be
1643 * returned in a normal path of operation: it means that driver is not
1644 * maintaining the TX queue stop/start state properly, and causes
1645 * the stack to do a non-trivial amount of useless work.
1646 * Since most packets only take 1 or 2 ring slots, stopping the queue
1647 * early means 16 slots are typically wasted.
1649 if (sq->vq->num_free < 2+MAX_SKB_FRAGS) {
1650 netif_stop_subqueue(dev, qnum);
1652 unlikely(!virtqueue_enable_cb_delayed(sq->vq))) {
1653 /* More just got used, free them then recheck. */
1654 free_old_xmit_skbs(sq, false);
1655 if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) {
1656 netif_start_subqueue(dev, qnum);
1657 virtqueue_disable_cb(sq->vq);
1662 if (kick || netif_xmit_stopped(txq)) {
1663 if (virtqueue_kick_prepare(sq->vq) && virtqueue_notify(sq->vq)) {
1664 u64_stats_update_begin(&sq->stats.syncp);
1666 u64_stats_update_end(&sq->stats.syncp);
1670 return NETDEV_TX_OK;
1674 * Send command via the control virtqueue and check status. Commands
1675 * supported by the hypervisor, as indicated by feature bits, should
1676 * never fail unless improperly formatted.
1678 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
1679 struct scatterlist *out)
1681 struct scatterlist *sgs[4], hdr, stat;
1682 unsigned out_num = 0, tmp;
1684 /* Caller should know better */
1685 BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ));
1687 vi->ctrl->status = ~0;
1688 vi->ctrl->hdr.class = class;
1689 vi->ctrl->hdr.cmd = cmd;
1691 sg_init_one(&hdr, &vi->ctrl->hdr, sizeof(vi->ctrl->hdr));
1692 sgs[out_num++] = &hdr;
1695 sgs[out_num++] = out;
1697 /* Add return status. */
1698 sg_init_one(&stat, &vi->ctrl->status, sizeof(vi->ctrl->status));
1699 sgs[out_num] = &stat;
1701 BUG_ON(out_num + 1 > ARRAY_SIZE(sgs));
1702 virtqueue_add_sgs(vi->cvq, sgs, out_num, 1, vi, GFP_ATOMIC);
1704 if (unlikely(!virtqueue_kick(vi->cvq)))
1705 return vi->ctrl->status == VIRTIO_NET_OK;
1707 /* Spin for a response, the kick causes an ioport write, trapping
1708 * into the hypervisor, so the request should be handled immediately.
1710 while (!virtqueue_get_buf(vi->cvq, &tmp) &&
1711 !virtqueue_is_broken(vi->cvq))
1714 return vi->ctrl->status == VIRTIO_NET_OK;
1717 static int virtnet_set_mac_address(struct net_device *dev, void *p)
1719 struct virtnet_info *vi = netdev_priv(dev);
1720 struct virtio_device *vdev = vi->vdev;
1722 struct sockaddr *addr;
1723 struct scatterlist sg;
1725 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STANDBY))
1728 addr = kmemdup(p, sizeof(*addr), GFP_KERNEL);
1732 ret = eth_prepare_mac_addr_change(dev, addr);
1736 if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1737 sg_init_one(&sg, addr->sa_data, dev->addr_len);
1738 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
1739 VIRTIO_NET_CTRL_MAC_ADDR_SET, &sg)) {
1740 dev_warn(&vdev->dev,
1741 "Failed to set mac address by vq command.\n");
1745 } else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC) &&
1746 !virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) {
1749 /* Naturally, this has an atomicity problem. */
1750 for (i = 0; i < dev->addr_len; i++)
1751 virtio_cwrite8(vdev,
1752 offsetof(struct virtio_net_config, mac) +
1753 i, addr->sa_data[i]);
1756 eth_commit_mac_addr_change(dev, p);
1764 static void virtnet_stats(struct net_device *dev,
1765 struct rtnl_link_stats64 *tot)
1767 struct virtnet_info *vi = netdev_priv(dev);
1771 for (i = 0; i < vi->max_queue_pairs; i++) {
1772 u64 tpackets, tbytes, rpackets, rbytes, rdrops;
1773 struct receive_queue *rq = &vi->rq[i];
1774 struct send_queue *sq = &vi->sq[i];
1777 start = u64_stats_fetch_begin_irq(&sq->stats.syncp);
1778 tpackets = sq->stats.packets;
1779 tbytes = sq->stats.bytes;
1780 } while (u64_stats_fetch_retry_irq(&sq->stats.syncp, start));
1783 start = u64_stats_fetch_begin_irq(&rq->stats.syncp);
1784 rpackets = rq->stats.packets;
1785 rbytes = rq->stats.bytes;
1786 rdrops = rq->stats.drops;
1787 } while (u64_stats_fetch_retry_irq(&rq->stats.syncp, start));
1789 tot->rx_packets += rpackets;
1790 tot->tx_packets += tpackets;
1791 tot->rx_bytes += rbytes;
1792 tot->tx_bytes += tbytes;
1793 tot->rx_dropped += rdrops;
1796 tot->tx_dropped = dev->stats.tx_dropped;
1797 tot->tx_fifo_errors = dev->stats.tx_fifo_errors;
1798 tot->rx_length_errors = dev->stats.rx_length_errors;
1799 tot->rx_frame_errors = dev->stats.rx_frame_errors;
1802 static void virtnet_ack_link_announce(struct virtnet_info *vi)
1805 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE,
1806 VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL))
1807 dev_warn(&vi->dev->dev, "Failed to ack link announce.\n");
1811 static int _virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs)
1813 struct scatterlist sg;
1814 struct net_device *dev = vi->dev;
1816 if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ))
1819 vi->ctrl->mq.virtqueue_pairs = cpu_to_virtio16(vi->vdev, queue_pairs);
1820 sg_init_one(&sg, &vi->ctrl->mq, sizeof(vi->ctrl->mq));
1822 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ,
1823 VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg)) {
1824 dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n",
1828 vi->curr_queue_pairs = queue_pairs;
1829 /* virtnet_open() will refill when device is going to up. */
1830 if (dev->flags & IFF_UP)
1831 schedule_delayed_work(&vi->refill, 0);
1837 static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs)
1842 err = _virtnet_set_queues(vi, queue_pairs);
1847 static int virtnet_close(struct net_device *dev)
1849 struct virtnet_info *vi = netdev_priv(dev);
1852 /* Make sure refill_work doesn't re-enable napi! */
1853 cancel_delayed_work_sync(&vi->refill);
1855 for (i = 0; i < vi->max_queue_pairs; i++) {
1856 xdp_rxq_info_unreg(&vi->rq[i].xdp_rxq);
1857 napi_disable(&vi->rq[i].napi);
1858 virtnet_napi_tx_disable(&vi->sq[i].napi);
1864 static void virtnet_set_rx_mode(struct net_device *dev)
1866 struct virtnet_info *vi = netdev_priv(dev);
1867 struct scatterlist sg[2];
1868 struct virtio_net_ctrl_mac *mac_data;
1869 struct netdev_hw_addr *ha;
1875 /* We can't dynamically set ndo_set_rx_mode, so return gracefully */
1876 if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
1879 vi->ctrl->promisc = ((dev->flags & IFF_PROMISC) != 0);
1880 vi->ctrl->allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
1882 sg_init_one(sg, &vi->ctrl->promisc, sizeof(vi->ctrl->promisc));
1884 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
1885 VIRTIO_NET_CTRL_RX_PROMISC, sg))
1886 dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
1887 vi->ctrl->promisc ? "en" : "dis");
1889 sg_init_one(sg, &vi->ctrl->allmulti, sizeof(vi->ctrl->allmulti));
1891 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
1892 VIRTIO_NET_CTRL_RX_ALLMULTI, sg))
1893 dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
1894 vi->ctrl->allmulti ? "en" : "dis");
1896 uc_count = netdev_uc_count(dev);
1897 mc_count = netdev_mc_count(dev);
1898 /* MAC filter - use one buffer for both lists */
1899 buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
1900 (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
1905 sg_init_table(sg, 2);
1907 /* Store the unicast list and count in the front of the buffer */
1908 mac_data->entries = cpu_to_virtio32(vi->vdev, uc_count);
1910 netdev_for_each_uc_addr(ha, dev)
1911 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
1913 sg_set_buf(&sg[0], mac_data,
1914 sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
1916 /* multicast list and count fill the end */
1917 mac_data = (void *)&mac_data->macs[uc_count][0];
1919 mac_data->entries = cpu_to_virtio32(vi->vdev, mc_count);
1921 netdev_for_each_mc_addr(ha, dev)
1922 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
1924 sg_set_buf(&sg[1], mac_data,
1925 sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
1927 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
1928 VIRTIO_NET_CTRL_MAC_TABLE_SET, sg))
1929 dev_warn(&dev->dev, "Failed to set MAC filter table.\n");
1934 static int virtnet_vlan_rx_add_vid(struct net_device *dev,
1935 __be16 proto, u16 vid)
1937 struct virtnet_info *vi = netdev_priv(dev);
1938 struct scatterlist sg;
1940 vi->ctrl->vid = cpu_to_virtio16(vi->vdev, vid);
1941 sg_init_one(&sg, &vi->ctrl->vid, sizeof(vi->ctrl->vid));
1943 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1944 VIRTIO_NET_CTRL_VLAN_ADD, &sg))
1945 dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
1949 static int virtnet_vlan_rx_kill_vid(struct net_device *dev,
1950 __be16 proto, u16 vid)
1952 struct virtnet_info *vi = netdev_priv(dev);
1953 struct scatterlist sg;
1955 vi->ctrl->vid = cpu_to_virtio16(vi->vdev, vid);
1956 sg_init_one(&sg, &vi->ctrl->vid, sizeof(vi->ctrl->vid));
1958 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1959 VIRTIO_NET_CTRL_VLAN_DEL, &sg))
1960 dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
1964 static void virtnet_clean_affinity(struct virtnet_info *vi, long hcpu)
1968 if (vi->affinity_hint_set) {
1969 for (i = 0; i < vi->max_queue_pairs; i++) {
1970 virtqueue_set_affinity(vi->rq[i].vq, NULL);
1971 virtqueue_set_affinity(vi->sq[i].vq, NULL);
1974 vi->affinity_hint_set = false;
1978 static void virtnet_set_affinity(struct virtnet_info *vi)
1987 if (!zalloc_cpumask_var(&mask, GFP_KERNEL)) {
1988 virtnet_clean_affinity(vi, -1);
1992 num_cpu = num_online_cpus();
1993 stride = max_t(int, num_cpu / vi->curr_queue_pairs, 1);
1994 stragglers = num_cpu >= vi->curr_queue_pairs ?
1995 num_cpu % vi->curr_queue_pairs :
1997 cpu = cpumask_next(-1, cpu_online_mask);
1999 for (i = 0; i < vi->curr_queue_pairs; i++) {
2000 group_size = stride + (i < stragglers ? 1 : 0);
2002 for (j = 0; j < group_size; j++) {
2003 cpumask_set_cpu(cpu, mask);
2004 cpu = cpumask_next_wrap(cpu, cpu_online_mask,
2007 virtqueue_set_affinity(vi->rq[i].vq, mask);
2008 virtqueue_set_affinity(vi->sq[i].vq, mask);
2009 __netif_set_xps_queue(vi->dev, cpumask_bits(mask), i, false);
2010 cpumask_clear(mask);
2013 vi->affinity_hint_set = true;
2014 free_cpumask_var(mask);
2017 static int virtnet_cpu_online(unsigned int cpu, struct hlist_node *node)
2019 struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
2021 virtnet_set_affinity(vi);
2025 static int virtnet_cpu_dead(unsigned int cpu, struct hlist_node *node)
2027 struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
2029 virtnet_set_affinity(vi);
2033 static int virtnet_cpu_down_prep(unsigned int cpu, struct hlist_node *node)
2035 struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
2038 virtnet_clean_affinity(vi, cpu);
2042 static enum cpuhp_state virtionet_online;
2044 static int virtnet_cpu_notif_add(struct virtnet_info *vi)
2048 ret = cpuhp_state_add_instance_nocalls(virtionet_online, &vi->node);
2051 ret = cpuhp_state_add_instance_nocalls(CPUHP_VIRT_NET_DEAD,
2055 cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node);
2059 static void virtnet_cpu_notif_remove(struct virtnet_info *vi)
2061 cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node);
2062 cpuhp_state_remove_instance_nocalls(CPUHP_VIRT_NET_DEAD,
2066 static void virtnet_get_ringparam(struct net_device *dev,
2067 struct ethtool_ringparam *ring)
2069 struct virtnet_info *vi = netdev_priv(dev);
2071 ring->rx_max_pending = virtqueue_get_vring_size(vi->rq[0].vq);
2072 ring->tx_max_pending = virtqueue_get_vring_size(vi->sq[0].vq);
2073 ring->rx_pending = ring->rx_max_pending;
2074 ring->tx_pending = ring->tx_max_pending;
2078 static void virtnet_get_drvinfo(struct net_device *dev,
2079 struct ethtool_drvinfo *info)
2081 struct virtnet_info *vi = netdev_priv(dev);
2082 struct virtio_device *vdev = vi->vdev;
2084 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
2085 strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version));
2086 strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info));
2090 /* TODO: Eliminate OOO packets during switching */
2091 static int virtnet_set_channels(struct net_device *dev,
2092 struct ethtool_channels *channels)
2094 struct virtnet_info *vi = netdev_priv(dev);
2095 u16 queue_pairs = channels->combined_count;
2098 /* We don't support separate rx/tx channels.
2099 * We don't allow setting 'other' channels.
2101 if (channels->rx_count || channels->tx_count || channels->other_count)
2104 if (queue_pairs > vi->max_queue_pairs || queue_pairs == 0)
2107 /* For now we don't support modifying channels while XDP is loaded
2108 * also when XDP is loaded all RX queues have XDP programs so we only
2109 * need to check a single RX queue.
2111 if (vi->rq[0].xdp_prog)
2115 err = _virtnet_set_queues(vi, queue_pairs);
2120 virtnet_set_affinity(vi);
2123 netif_set_real_num_tx_queues(dev, queue_pairs);
2124 netif_set_real_num_rx_queues(dev, queue_pairs);
2129 static void virtnet_get_strings(struct net_device *dev, u32 stringset, u8 *data)
2131 struct virtnet_info *vi = netdev_priv(dev);
2132 char *p = (char *)data;
2135 switch (stringset) {
2137 for (i = 0; i < vi->curr_queue_pairs; i++) {
2138 for (j = 0; j < VIRTNET_RQ_STATS_LEN; j++) {
2139 snprintf(p, ETH_GSTRING_LEN, "rx_queue_%u_%s",
2140 i, virtnet_rq_stats_desc[j].desc);
2141 p += ETH_GSTRING_LEN;
2145 for (i = 0; i < vi->curr_queue_pairs; i++) {
2146 for (j = 0; j < VIRTNET_SQ_STATS_LEN; j++) {
2147 snprintf(p, ETH_GSTRING_LEN, "tx_queue_%u_%s",
2148 i, virtnet_sq_stats_desc[j].desc);
2149 p += ETH_GSTRING_LEN;
2156 static int virtnet_get_sset_count(struct net_device *dev, int sset)
2158 struct virtnet_info *vi = netdev_priv(dev);
2162 return vi->curr_queue_pairs * (VIRTNET_RQ_STATS_LEN +
2163 VIRTNET_SQ_STATS_LEN);
2169 static void virtnet_get_ethtool_stats(struct net_device *dev,
2170 struct ethtool_stats *stats, u64 *data)
2172 struct virtnet_info *vi = netdev_priv(dev);
2173 unsigned int idx = 0, start, i, j;
2174 const u8 *stats_base;
2177 for (i = 0; i < vi->curr_queue_pairs; i++) {
2178 struct receive_queue *rq = &vi->rq[i];
2180 stats_base = (u8 *)&rq->stats;
2182 start = u64_stats_fetch_begin_irq(&rq->stats.syncp);
2183 for (j = 0; j < VIRTNET_RQ_STATS_LEN; j++) {
2184 offset = virtnet_rq_stats_desc[j].offset;
2185 data[idx + j] = *(u64 *)(stats_base + offset);
2187 } while (u64_stats_fetch_retry_irq(&rq->stats.syncp, start));
2188 idx += VIRTNET_RQ_STATS_LEN;
2191 for (i = 0; i < vi->curr_queue_pairs; i++) {
2192 struct send_queue *sq = &vi->sq[i];
2194 stats_base = (u8 *)&sq->stats;
2196 start = u64_stats_fetch_begin_irq(&sq->stats.syncp);
2197 for (j = 0; j < VIRTNET_SQ_STATS_LEN; j++) {
2198 offset = virtnet_sq_stats_desc[j].offset;
2199 data[idx + j] = *(u64 *)(stats_base + offset);
2201 } while (u64_stats_fetch_retry_irq(&sq->stats.syncp, start));
2202 idx += VIRTNET_SQ_STATS_LEN;
2206 static void virtnet_get_channels(struct net_device *dev,
2207 struct ethtool_channels *channels)
2209 struct virtnet_info *vi = netdev_priv(dev);
2211 channels->combined_count = vi->curr_queue_pairs;
2212 channels->max_combined = vi->max_queue_pairs;
2213 channels->max_other = 0;
2214 channels->rx_count = 0;
2215 channels->tx_count = 0;
2216 channels->other_count = 0;
2219 /* Check if the user is trying to change anything besides speed/duplex */
2221 virtnet_validate_ethtool_cmd(const struct ethtool_link_ksettings *cmd)
2223 struct ethtool_link_ksettings diff1 = *cmd;
2224 struct ethtool_link_ksettings diff2 = {};
2226 /* cmd is always set so we need to clear it, validate the port type
2227 * and also without autonegotiation we can ignore advertising
2229 diff1.base.speed = 0;
2230 diff2.base.port = PORT_OTHER;
2231 ethtool_link_ksettings_zero_link_mode(&diff1, advertising);
2232 diff1.base.duplex = 0;
2234 diff1.base.link_mode_masks_nwords = 0;
2236 return !memcmp(&diff1.base, &diff2.base, sizeof(diff1.base)) &&
2237 bitmap_empty(diff1.link_modes.supported,
2238 __ETHTOOL_LINK_MODE_MASK_NBITS) &&
2239 bitmap_empty(diff1.link_modes.advertising,
2240 __ETHTOOL_LINK_MODE_MASK_NBITS) &&
2241 bitmap_empty(diff1.link_modes.lp_advertising,
2242 __ETHTOOL_LINK_MODE_MASK_NBITS);
2245 static int virtnet_set_link_ksettings(struct net_device *dev,
2246 const struct ethtool_link_ksettings *cmd)
2248 struct virtnet_info *vi = netdev_priv(dev);
2251 speed = cmd->base.speed;
2252 /* don't allow custom speed and duplex */
2253 if (!ethtool_validate_speed(speed) ||
2254 !ethtool_validate_duplex(cmd->base.duplex) ||
2255 !virtnet_validate_ethtool_cmd(cmd))
2258 vi->duplex = cmd->base.duplex;
2263 static int virtnet_get_link_ksettings(struct net_device *dev,
2264 struct ethtool_link_ksettings *cmd)
2266 struct virtnet_info *vi = netdev_priv(dev);
2268 cmd->base.speed = vi->speed;
2269 cmd->base.duplex = vi->duplex;
2270 cmd->base.port = PORT_OTHER;
2275 static void virtnet_init_settings(struct net_device *dev)
2277 struct virtnet_info *vi = netdev_priv(dev);
2279 vi->speed = SPEED_UNKNOWN;
2280 vi->duplex = DUPLEX_UNKNOWN;
2283 static void virtnet_update_settings(struct virtnet_info *vi)
2288 if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_SPEED_DUPLEX))
2291 speed = virtio_cread32(vi->vdev, offsetof(struct virtio_net_config,
2293 if (ethtool_validate_speed(speed))
2295 duplex = virtio_cread8(vi->vdev, offsetof(struct virtio_net_config,
2297 if (ethtool_validate_duplex(duplex))
2298 vi->duplex = duplex;
2301 static const struct ethtool_ops virtnet_ethtool_ops = {
2302 .get_drvinfo = virtnet_get_drvinfo,
2303 .get_link = ethtool_op_get_link,
2304 .get_ringparam = virtnet_get_ringparam,
2305 .get_strings = virtnet_get_strings,
2306 .get_sset_count = virtnet_get_sset_count,
2307 .get_ethtool_stats = virtnet_get_ethtool_stats,
2308 .set_channels = virtnet_set_channels,
2309 .get_channels = virtnet_get_channels,
2310 .get_ts_info = ethtool_op_get_ts_info,
2311 .get_link_ksettings = virtnet_get_link_ksettings,
2312 .set_link_ksettings = virtnet_set_link_ksettings,
2315 static void virtnet_freeze_down(struct virtio_device *vdev)
2317 struct virtnet_info *vi = vdev->priv;
2320 /* Make sure no work handler is accessing the device */
2321 flush_work(&vi->config_work);
2323 netif_tx_lock_bh(vi->dev);
2324 netif_device_detach(vi->dev);
2325 netif_tx_unlock_bh(vi->dev);
2326 cancel_delayed_work_sync(&vi->refill);
2328 if (netif_running(vi->dev)) {
2329 for (i = 0; i < vi->max_queue_pairs; i++) {
2330 napi_disable(&vi->rq[i].napi);
2331 virtnet_napi_tx_disable(&vi->sq[i].napi);
2336 static int init_vqs(struct virtnet_info *vi);
2338 static int virtnet_restore_up(struct virtio_device *vdev)
2340 struct virtnet_info *vi = vdev->priv;
2347 virtio_device_ready(vdev);
2349 if (netif_running(vi->dev)) {
2350 for (i = 0; i < vi->curr_queue_pairs; i++)
2351 if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL))
2352 schedule_delayed_work(&vi->refill, 0);
2354 for (i = 0; i < vi->max_queue_pairs; i++) {
2355 virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi);
2356 virtnet_napi_tx_enable(vi, vi->sq[i].vq,
2361 netif_tx_lock_bh(vi->dev);
2362 netif_device_attach(vi->dev);
2363 netif_tx_unlock_bh(vi->dev);
2367 static int virtnet_set_guest_offloads(struct virtnet_info *vi, u64 offloads)
2369 struct scatterlist sg;
2370 vi->ctrl->offloads = cpu_to_virtio64(vi->vdev, offloads);
2372 sg_init_one(&sg, &vi->ctrl->offloads, sizeof(vi->ctrl->offloads));
2374 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_GUEST_OFFLOADS,
2375 VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET, &sg)) {
2376 dev_warn(&vi->dev->dev, "Fail to set guest offload. \n");
2383 static int virtnet_clear_guest_offloads(struct virtnet_info *vi)
2387 if (!vi->guest_offloads)
2390 return virtnet_set_guest_offloads(vi, offloads);
2393 static int virtnet_restore_guest_offloads(struct virtnet_info *vi)
2395 u64 offloads = vi->guest_offloads;
2397 if (!vi->guest_offloads)
2400 return virtnet_set_guest_offloads(vi, offloads);
2403 static int virtnet_xdp_set(struct net_device *dev, struct bpf_prog *prog,
2404 struct netlink_ext_ack *extack)
2406 unsigned long int max_sz = PAGE_SIZE - sizeof(struct padded_vnet_hdr);
2407 struct virtnet_info *vi = netdev_priv(dev);
2408 struct bpf_prog *old_prog;
2409 u16 xdp_qp = 0, curr_qp;
2412 if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)
2413 && (virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO4) ||
2414 virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO6) ||
2415 virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_ECN) ||
2416 virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_UFO) ||
2417 virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_CSUM))) {
2418 NL_SET_ERR_MSG_MOD(extack, "Can't set XDP while host is implementing LRO/CSUM, disable LRO/CSUM first");
2422 if (vi->mergeable_rx_bufs && !vi->any_header_sg) {
2423 NL_SET_ERR_MSG_MOD(extack, "XDP expects header/data in single page, any_header_sg required");
2427 if (dev->mtu > max_sz) {
2428 NL_SET_ERR_MSG_MOD(extack, "MTU too large to enable XDP");
2429 netdev_warn(dev, "XDP requires MTU less than %lu\n", max_sz);
2433 curr_qp = vi->curr_queue_pairs - vi->xdp_queue_pairs;
2435 xdp_qp = nr_cpu_ids;
2437 /* XDP requires extra queues for XDP_TX */
2438 if (curr_qp + xdp_qp > vi->max_queue_pairs) {
2439 NL_SET_ERR_MSG_MOD(extack, "Too few free TX rings available");
2440 netdev_warn(dev, "request %i queues but max is %i\n",
2441 curr_qp + xdp_qp, vi->max_queue_pairs);
2445 old_prog = rtnl_dereference(vi->rq[0].xdp_prog);
2446 if (!prog && !old_prog)
2450 prog = bpf_prog_add(prog, vi->max_queue_pairs - 1);
2452 return PTR_ERR(prog);
2455 /* Make sure NAPI is not using any XDP TX queues for RX. */
2456 if (netif_running(dev)) {
2457 for (i = 0; i < vi->max_queue_pairs; i++) {
2458 napi_disable(&vi->rq[i].napi);
2459 virtnet_napi_tx_disable(&vi->sq[i].napi);
2464 for (i = 0; i < vi->max_queue_pairs; i++) {
2465 rcu_assign_pointer(vi->rq[i].xdp_prog, prog);
2467 virtnet_restore_guest_offloads(vi);
2472 err = _virtnet_set_queues(vi, curr_qp + xdp_qp);
2475 netif_set_real_num_rx_queues(dev, curr_qp + xdp_qp);
2476 vi->xdp_queue_pairs = xdp_qp;
2479 for (i = 0; i < vi->max_queue_pairs; i++) {
2480 rcu_assign_pointer(vi->rq[i].xdp_prog, prog);
2481 if (i == 0 && !old_prog)
2482 virtnet_clear_guest_offloads(vi);
2486 for (i = 0; i < vi->max_queue_pairs; i++) {
2488 bpf_prog_put(old_prog);
2489 if (netif_running(dev)) {
2490 virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi);
2491 virtnet_napi_tx_enable(vi, vi->sq[i].vq,
2500 virtnet_clear_guest_offloads(vi);
2501 for (i = 0; i < vi->max_queue_pairs; i++)
2502 rcu_assign_pointer(vi->rq[i].xdp_prog, old_prog);
2505 if (netif_running(dev)) {
2506 for (i = 0; i < vi->max_queue_pairs; i++) {
2507 virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi);
2508 virtnet_napi_tx_enable(vi, vi->sq[i].vq,
2513 bpf_prog_sub(prog, vi->max_queue_pairs - 1);
2517 static u32 virtnet_xdp_query(struct net_device *dev)
2519 struct virtnet_info *vi = netdev_priv(dev);
2520 const struct bpf_prog *xdp_prog;
2523 for (i = 0; i < vi->max_queue_pairs; i++) {
2524 xdp_prog = rtnl_dereference(vi->rq[i].xdp_prog);
2526 return xdp_prog->aux->id;
2531 static int virtnet_xdp(struct net_device *dev, struct netdev_bpf *xdp)
2533 switch (xdp->command) {
2534 case XDP_SETUP_PROG:
2535 return virtnet_xdp_set(dev, xdp->prog, xdp->extack);
2536 case XDP_QUERY_PROG:
2537 xdp->prog_id = virtnet_xdp_query(dev);
2544 static int virtnet_get_phys_port_name(struct net_device *dev, char *buf,
2547 struct virtnet_info *vi = netdev_priv(dev);
2550 if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_STANDBY))
2553 ret = snprintf(buf, len, "sby");
2560 static const struct net_device_ops virtnet_netdev = {
2561 .ndo_open = virtnet_open,
2562 .ndo_stop = virtnet_close,
2563 .ndo_start_xmit = start_xmit,
2564 .ndo_validate_addr = eth_validate_addr,
2565 .ndo_set_mac_address = virtnet_set_mac_address,
2566 .ndo_set_rx_mode = virtnet_set_rx_mode,
2567 .ndo_get_stats64 = virtnet_stats,
2568 .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
2569 .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
2570 .ndo_bpf = virtnet_xdp,
2571 .ndo_xdp_xmit = virtnet_xdp_xmit,
2572 .ndo_features_check = passthru_features_check,
2573 .ndo_get_phys_port_name = virtnet_get_phys_port_name,
2576 static void virtnet_config_changed_work(struct work_struct *work)
2578 struct virtnet_info *vi =
2579 container_of(work, struct virtnet_info, config_work);
2582 if (virtio_cread_feature(vi->vdev, VIRTIO_NET_F_STATUS,
2583 struct virtio_net_config, status, &v) < 0)
2586 if (v & VIRTIO_NET_S_ANNOUNCE) {
2587 netdev_notify_peers(vi->dev);
2588 virtnet_ack_link_announce(vi);
2591 /* Ignore unknown (future) status bits */
2592 v &= VIRTIO_NET_S_LINK_UP;
2594 if (vi->status == v)
2599 if (vi->status & VIRTIO_NET_S_LINK_UP) {
2600 virtnet_update_settings(vi);
2601 netif_carrier_on(vi->dev);
2602 netif_tx_wake_all_queues(vi->dev);
2604 netif_carrier_off(vi->dev);
2605 netif_tx_stop_all_queues(vi->dev);
2609 static void virtnet_config_changed(struct virtio_device *vdev)
2611 struct virtnet_info *vi = vdev->priv;
2613 schedule_work(&vi->config_work);
2616 static void virtnet_free_queues(struct virtnet_info *vi)
2620 for (i = 0; i < vi->max_queue_pairs; i++) {
2621 napi_hash_del(&vi->rq[i].napi);
2622 netif_napi_del(&vi->rq[i].napi);
2623 netif_napi_del(&vi->sq[i].napi);
2626 /* We called napi_hash_del() before netif_napi_del(),
2627 * we need to respect an RCU grace period before freeing vi->rq
2636 static void _free_receive_bufs(struct virtnet_info *vi)
2638 struct bpf_prog *old_prog;
2641 for (i = 0; i < vi->max_queue_pairs; i++) {
2642 while (vi->rq[i].pages)
2643 __free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0);
2645 old_prog = rtnl_dereference(vi->rq[i].xdp_prog);
2646 RCU_INIT_POINTER(vi->rq[i].xdp_prog, NULL);
2648 bpf_prog_put(old_prog);
2652 static void free_receive_bufs(struct virtnet_info *vi)
2655 _free_receive_bufs(vi);
2659 static void free_receive_page_frags(struct virtnet_info *vi)
2662 for (i = 0; i < vi->max_queue_pairs; i++)
2663 if (vi->rq[i].alloc_frag.page)
2664 put_page(vi->rq[i].alloc_frag.page);
2667 static void free_unused_bufs(struct virtnet_info *vi)
2672 for (i = 0; i < vi->max_queue_pairs; i++) {
2673 struct virtqueue *vq = vi->sq[i].vq;
2674 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) {
2675 if (!is_xdp_frame(buf))
2678 xdp_return_frame(ptr_to_xdp(buf));
2682 for (i = 0; i < vi->max_queue_pairs; i++) {
2683 struct virtqueue *vq = vi->rq[i].vq;
2685 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) {
2686 if (vi->mergeable_rx_bufs) {
2687 put_page(virt_to_head_page(buf));
2688 } else if (vi->big_packets) {
2689 give_pages(&vi->rq[i], buf);
2691 put_page(virt_to_head_page(buf));
2697 static void virtnet_del_vqs(struct virtnet_info *vi)
2699 struct virtio_device *vdev = vi->vdev;
2701 virtnet_clean_affinity(vi, -1);
2703 vdev->config->del_vqs(vdev);
2705 virtnet_free_queues(vi);
2708 /* How large should a single buffer be so a queue full of these can fit at
2709 * least one full packet?
2710 * Logic below assumes the mergeable buffer header is used.
2712 static unsigned int mergeable_min_buf_len(struct virtnet_info *vi, struct virtqueue *vq)
2714 const unsigned int hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
2715 unsigned int rq_size = virtqueue_get_vring_size(vq);
2716 unsigned int packet_len = vi->big_packets ? IP_MAX_MTU : vi->dev->max_mtu;
2717 unsigned int buf_len = hdr_len + ETH_HLEN + VLAN_HLEN + packet_len;
2718 unsigned int min_buf_len = DIV_ROUND_UP(buf_len, rq_size);
2720 return max(max(min_buf_len, hdr_len) - hdr_len,
2721 (unsigned int)GOOD_PACKET_LEN);
2724 static int virtnet_find_vqs(struct virtnet_info *vi)
2726 vq_callback_t **callbacks;
2727 struct virtqueue **vqs;
2733 /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
2734 * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
2735 * possible control vq.
2737 total_vqs = vi->max_queue_pairs * 2 +
2738 virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ);
2740 /* Allocate space for find_vqs parameters */
2741 vqs = kcalloc(total_vqs, sizeof(*vqs), GFP_KERNEL);
2744 callbacks = kmalloc_array(total_vqs, sizeof(*callbacks), GFP_KERNEL);
2747 names = kmalloc_array(total_vqs, sizeof(*names), GFP_KERNEL);
2750 if (!vi->big_packets || vi->mergeable_rx_bufs) {
2751 ctx = kcalloc(total_vqs, sizeof(*ctx), GFP_KERNEL);
2758 /* Parameters for control virtqueue, if any */
2760 callbacks[total_vqs - 1] = NULL;
2761 names[total_vqs - 1] = "control";
2764 /* Allocate/initialize parameters for send/receive virtqueues */
2765 for (i = 0; i < vi->max_queue_pairs; i++) {
2766 callbacks[rxq2vq(i)] = skb_recv_done;
2767 callbacks[txq2vq(i)] = skb_xmit_done;
2768 sprintf(vi->rq[i].name, "input.%d", i);
2769 sprintf(vi->sq[i].name, "output.%d", i);
2770 names[rxq2vq(i)] = vi->rq[i].name;
2771 names[txq2vq(i)] = vi->sq[i].name;
2773 ctx[rxq2vq(i)] = true;
2776 ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs, callbacks,
2782 vi->cvq = vqs[total_vqs - 1];
2783 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
2784 vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
2787 for (i = 0; i < vi->max_queue_pairs; i++) {
2788 vi->rq[i].vq = vqs[rxq2vq(i)];
2789 vi->rq[i].min_buf_len = mergeable_min_buf_len(vi, vi->rq[i].vq);
2790 vi->sq[i].vq = vqs[txq2vq(i)];
2793 /* run here: ret == 0. */
2808 static int virtnet_alloc_queues(struct virtnet_info *vi)
2812 vi->ctrl = kzalloc(sizeof(*vi->ctrl), GFP_KERNEL);
2815 vi->sq = kcalloc(vi->max_queue_pairs, sizeof(*vi->sq), GFP_KERNEL);
2818 vi->rq = kcalloc(vi->max_queue_pairs, sizeof(*vi->rq), GFP_KERNEL);
2822 INIT_DELAYED_WORK(&vi->refill, refill_work);
2823 for (i = 0; i < vi->max_queue_pairs; i++) {
2824 vi->rq[i].pages = NULL;
2825 netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll,
2827 netif_tx_napi_add(vi->dev, &vi->sq[i].napi, virtnet_poll_tx,
2828 napi_tx ? napi_weight : 0);
2830 sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg));
2831 ewma_pkt_len_init(&vi->rq[i].mrg_avg_pkt_len);
2832 sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg));
2834 u64_stats_init(&vi->rq[i].stats.syncp);
2835 u64_stats_init(&vi->sq[i].stats.syncp);
2848 static int init_vqs(struct virtnet_info *vi)
2852 /* Allocate send & receive queues */
2853 ret = virtnet_alloc_queues(vi);
2857 ret = virtnet_find_vqs(vi);
2862 virtnet_set_affinity(vi);
2868 virtnet_free_queues(vi);
2874 static ssize_t mergeable_rx_buffer_size_show(struct netdev_rx_queue *queue,
2877 struct virtnet_info *vi = netdev_priv(queue->dev);
2878 unsigned int queue_index = get_netdev_rx_queue_index(queue);
2879 unsigned int headroom = virtnet_get_headroom(vi);
2880 unsigned int tailroom = headroom ? sizeof(struct skb_shared_info) : 0;
2881 struct ewma_pkt_len *avg;
2883 BUG_ON(queue_index >= vi->max_queue_pairs);
2884 avg = &vi->rq[queue_index].mrg_avg_pkt_len;
2885 return sprintf(buf, "%u\n",
2886 get_mergeable_buf_len(&vi->rq[queue_index], avg,
2887 SKB_DATA_ALIGN(headroom + tailroom)));
2890 static struct rx_queue_attribute mergeable_rx_buffer_size_attribute =
2891 __ATTR_RO(mergeable_rx_buffer_size);
2893 static struct attribute *virtio_net_mrg_rx_attrs[] = {
2894 &mergeable_rx_buffer_size_attribute.attr,
2898 static const struct attribute_group virtio_net_mrg_rx_group = {
2899 .name = "virtio_net",
2900 .attrs = virtio_net_mrg_rx_attrs
2904 static bool virtnet_fail_on_feature(struct virtio_device *vdev,
2906 const char *fname, const char *dname)
2908 if (!virtio_has_feature(vdev, fbit))
2911 dev_err(&vdev->dev, "device advertises feature %s but not %s",
2917 #define VIRTNET_FAIL_ON(vdev, fbit, dbit) \
2918 virtnet_fail_on_feature(vdev, fbit, #fbit, dbit)
2920 static bool virtnet_validate_features(struct virtio_device *vdev)
2922 if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) &&
2923 (VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_RX,
2924 "VIRTIO_NET_F_CTRL_VQ") ||
2925 VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_VLAN,
2926 "VIRTIO_NET_F_CTRL_VQ") ||
2927 VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE,
2928 "VIRTIO_NET_F_CTRL_VQ") ||
2929 VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_MQ, "VIRTIO_NET_F_CTRL_VQ") ||
2930 VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR,
2931 "VIRTIO_NET_F_CTRL_VQ"))) {
2938 #define MIN_MTU ETH_MIN_MTU
2939 #define MAX_MTU ETH_MAX_MTU
2941 static int virtnet_validate(struct virtio_device *vdev)
2943 if (!vdev->config->get) {
2944 dev_err(&vdev->dev, "%s failure: config access disabled\n",
2949 if (!virtnet_validate_features(vdev))
2952 if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) {
2953 int mtu = virtio_cread16(vdev,
2954 offsetof(struct virtio_net_config,
2957 __virtio_clear_bit(vdev, VIRTIO_NET_F_MTU);
2963 static int virtnet_probe(struct virtio_device *vdev)
2965 int i, err = -ENOMEM;
2966 struct net_device *dev;
2967 struct virtnet_info *vi;
2968 u16 max_queue_pairs;
2971 /* Find if host supports multiqueue virtio_net device */
2972 err = virtio_cread_feature(vdev, VIRTIO_NET_F_MQ,
2973 struct virtio_net_config,
2974 max_virtqueue_pairs, &max_queue_pairs);
2976 /* We need at least 2 queue's */
2977 if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
2978 max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
2979 !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
2980 max_queue_pairs = 1;
2982 /* Allocate ourselves a network device with room for our info */
2983 dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs);
2987 /* Set up network device as normal. */
2988 dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE;
2989 dev->netdev_ops = &virtnet_netdev;
2990 dev->features = NETIF_F_HIGHDMA;
2992 dev->ethtool_ops = &virtnet_ethtool_ops;
2993 SET_NETDEV_DEV(dev, &vdev->dev);
2995 /* Do we support "hardware" checksums? */
2996 if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
2997 /* This opens up the world of extra features. */
2998 dev->hw_features |= NETIF_F_HW_CSUM | NETIF_F_SG;
3000 dev->features |= NETIF_F_HW_CSUM | NETIF_F_SG;
3002 if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
3003 dev->hw_features |= NETIF_F_TSO
3004 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
3006 /* Individual feature bits: what can host handle? */
3007 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
3008 dev->hw_features |= NETIF_F_TSO;
3009 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
3010 dev->hw_features |= NETIF_F_TSO6;
3011 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
3012 dev->hw_features |= NETIF_F_TSO_ECN;
3014 dev->features |= NETIF_F_GSO_ROBUST;
3017 dev->features |= dev->hw_features & NETIF_F_ALL_TSO;
3018 /* (!csum && gso) case will be fixed by register_netdev() */
3020 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM))
3021 dev->features |= NETIF_F_RXCSUM;
3023 dev->vlan_features = dev->features;
3025 /* MTU range: 68 - 65535 */
3026 dev->min_mtu = MIN_MTU;
3027 dev->max_mtu = MAX_MTU;
3029 /* Configuration may specify what MAC to use. Otherwise random. */
3030 if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
3031 virtio_cread_bytes(vdev,
3032 offsetof(struct virtio_net_config, mac),
3033 dev->dev_addr, dev->addr_len);
3035 eth_hw_addr_random(dev);
3037 /* Set up our device-specific information */
3038 vi = netdev_priv(dev);
3043 INIT_WORK(&vi->config_work, virtnet_config_changed_work);
3045 /* If we can receive ANY GSO packets, we must allocate large ones. */
3046 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
3047 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
3048 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN) ||
3049 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_UFO))
3050 vi->big_packets = true;
3052 if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
3053 vi->mergeable_rx_bufs = true;
3055 if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF) ||
3056 virtio_has_feature(vdev, VIRTIO_F_VERSION_1))
3057 vi->hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
3059 vi->hdr_len = sizeof(struct virtio_net_hdr);
3061 if (virtio_has_feature(vdev, VIRTIO_F_ANY_LAYOUT) ||
3062 virtio_has_feature(vdev, VIRTIO_F_VERSION_1))
3063 vi->any_header_sg = true;
3065 if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
3068 if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) {
3069 mtu = virtio_cread16(vdev,
3070 offsetof(struct virtio_net_config,
3072 if (mtu < dev->min_mtu) {
3073 /* Should never trigger: MTU was previously validated
3074 * in virtnet_validate.
3076 dev_err(&vdev->dev, "device MTU appears to have changed "
3077 "it is now %d < %d", mtu, dev->min_mtu);
3084 /* TODO: size buffers correctly in this case. */
3085 if (dev->mtu > ETH_DATA_LEN)
3086 vi->big_packets = true;
3089 if (vi->any_header_sg)
3090 dev->needed_headroom = vi->hdr_len;
3092 /* Enable multiqueue by default */
3093 if (num_online_cpus() >= max_queue_pairs)
3094 vi->curr_queue_pairs = max_queue_pairs;
3096 vi->curr_queue_pairs = num_online_cpus();
3097 vi->max_queue_pairs = max_queue_pairs;
3099 /* Allocate/initialize the rx/tx queues, and invoke find_vqs */
3105 if (vi->mergeable_rx_bufs)
3106 dev->sysfs_rx_queue_group = &virtio_net_mrg_rx_group;
3108 netif_set_real_num_tx_queues(dev, vi->curr_queue_pairs);
3109 netif_set_real_num_rx_queues(dev, vi->curr_queue_pairs);
3111 virtnet_init_settings(dev);
3113 if (virtio_has_feature(vdev, VIRTIO_NET_F_STANDBY)) {
3114 vi->failover = net_failover_create(vi->dev);
3115 if (IS_ERR(vi->failover)) {
3116 err = PTR_ERR(vi->failover);
3121 err = register_netdev(dev);
3123 pr_debug("virtio_net: registering device failed\n");
3127 virtio_device_ready(vdev);
3129 err = virtnet_cpu_notif_add(vi);
3131 pr_debug("virtio_net: registering cpu notifier failed\n");
3132 goto free_unregister_netdev;
3135 virtnet_set_queues(vi, vi->curr_queue_pairs);
3137 /* Assume link up if device can't report link status,
3138 otherwise get link status from config. */
3139 netif_carrier_off(dev);
3140 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
3141 schedule_work(&vi->config_work);
3143 vi->status = VIRTIO_NET_S_LINK_UP;
3144 virtnet_update_settings(vi);
3145 netif_carrier_on(dev);
3148 for (i = 0; i < ARRAY_SIZE(guest_offloads); i++)
3149 if (virtio_has_feature(vi->vdev, guest_offloads[i]))
3150 set_bit(guest_offloads[i], &vi->guest_offloads);
3152 pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
3153 dev->name, max_queue_pairs);
3157 free_unregister_netdev:
3158 vi->vdev->config->reset(vdev);
3160 unregister_netdev(dev);
3162 net_failover_destroy(vi->failover);
3164 cancel_delayed_work_sync(&vi->refill);
3165 free_receive_page_frags(vi);
3166 virtnet_del_vqs(vi);
3172 static void remove_vq_common(struct virtnet_info *vi)
3174 vi->vdev->config->reset(vi->vdev);
3176 /* Free unused buffers in both send and recv, if any. */
3177 free_unused_bufs(vi);
3179 free_receive_bufs(vi);
3181 free_receive_page_frags(vi);
3183 virtnet_del_vqs(vi);
3186 static void virtnet_remove(struct virtio_device *vdev)
3188 struct virtnet_info *vi = vdev->priv;
3190 virtnet_cpu_notif_remove(vi);
3192 /* Make sure no work handler is accessing the device. */
3193 flush_work(&vi->config_work);
3195 unregister_netdev(vi->dev);
3197 net_failover_destroy(vi->failover);
3199 remove_vq_common(vi);
3201 free_netdev(vi->dev);
3204 static __maybe_unused int virtnet_freeze(struct virtio_device *vdev)
3206 struct virtnet_info *vi = vdev->priv;
3208 virtnet_cpu_notif_remove(vi);
3209 virtnet_freeze_down(vdev);
3210 remove_vq_common(vi);
3215 static __maybe_unused int virtnet_restore(struct virtio_device *vdev)
3217 struct virtnet_info *vi = vdev->priv;
3220 err = virtnet_restore_up(vdev);
3223 virtnet_set_queues(vi, vi->curr_queue_pairs);
3225 err = virtnet_cpu_notif_add(vi);
3227 virtnet_freeze_down(vdev);
3228 remove_vq_common(vi);
3235 static struct virtio_device_id id_table[] = {
3236 { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
3240 #define VIRTNET_FEATURES \
3241 VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM, \
3243 VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6, \
3244 VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, \
3245 VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO, \
3246 VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ, \
3247 VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, \
3248 VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ, \
3249 VIRTIO_NET_F_CTRL_MAC_ADDR, \
3250 VIRTIO_NET_F_MTU, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, \
3251 VIRTIO_NET_F_SPEED_DUPLEX, VIRTIO_NET_F_STANDBY
3253 static unsigned int features[] = {
3257 static unsigned int features_legacy[] = {
3260 VIRTIO_F_ANY_LAYOUT,
3263 static struct virtio_driver virtio_net_driver = {
3264 .feature_table = features,
3265 .feature_table_size = ARRAY_SIZE(features),
3266 .feature_table_legacy = features_legacy,
3267 .feature_table_size_legacy = ARRAY_SIZE(features_legacy),
3268 .driver.name = KBUILD_MODNAME,
3269 .driver.owner = THIS_MODULE,
3270 .id_table = id_table,
3271 .validate = virtnet_validate,
3272 .probe = virtnet_probe,
3273 .remove = virtnet_remove,
3274 .config_changed = virtnet_config_changed,
3275 #ifdef CONFIG_PM_SLEEP
3276 .freeze = virtnet_freeze,
3277 .restore = virtnet_restore,
3281 static __init int virtio_net_driver_init(void)
3285 ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "virtio/net:online",
3287 virtnet_cpu_down_prep);
3290 virtionet_online = ret;
3291 ret = cpuhp_setup_state_multi(CPUHP_VIRT_NET_DEAD, "virtio/net:dead",
3292 NULL, virtnet_cpu_dead);
3296 ret = register_virtio_driver(&virtio_net_driver);
3301 cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD);
3303 cpuhp_remove_multi_state(virtionet_online);
3307 module_init(virtio_net_driver_init);
3309 static __exit void virtio_net_driver_exit(void)
3311 unregister_virtio_driver(&virtio_net_driver);
3312 cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD);
3313 cpuhp_remove_multi_state(virtionet_online);
3315 module_exit(virtio_net_driver_exit);
3317 MODULE_DEVICE_TABLE(virtio, id_table);
3318 MODULE_DESCRIPTION("Virtio network driver");
3319 MODULE_LICENSE("GPL");