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 <net/route.h>
36 #include <net/net_failover.h>
38 static int napi_weight = NAPI_POLL_WEIGHT;
39 module_param(napi_weight, int, 0444);
41 static bool csum = true, gso = true, napi_tx;
42 module_param(csum, bool, 0444);
43 module_param(gso, bool, 0444);
44 module_param(napi_tx, bool, 0644);
46 /* FIXME: MTU in config. */
47 #define GOOD_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
48 #define GOOD_COPY_LEN 128
50 #define VIRTNET_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
52 /* Amount of XDP headroom to prepend to packets for use by xdp_adjust_head */
53 #define VIRTIO_XDP_HEADROOM 256
55 /* Separating two types of XDP xmit */
56 #define VIRTIO_XDP_TX BIT(0)
57 #define VIRTIO_XDP_REDIR BIT(1)
59 #define VIRTIO_XDP_FLAG BIT(0)
61 /* RX packet size EWMA. The average packet size is used to determine the packet
62 * buffer size when refilling RX rings. As the entire RX ring may be refilled
63 * at once, the weight is chosen so that the EWMA will be insensitive to short-
64 * term, transient changes in packet size.
66 DECLARE_EWMA(pkt_len, 0, 64)
68 #define VIRTNET_DRIVER_VERSION "1.0.0"
70 static const unsigned long guest_offloads[] = {
71 VIRTIO_NET_F_GUEST_TSO4,
72 VIRTIO_NET_F_GUEST_TSO6,
73 VIRTIO_NET_F_GUEST_ECN,
74 VIRTIO_NET_F_GUEST_UFO,
75 VIRTIO_NET_F_GUEST_CSUM
78 struct virtnet_stat_desc {
79 char desc[ETH_GSTRING_LEN];
83 struct virtnet_sq_stats {
84 struct u64_stats_sync syncp;
92 struct virtnet_rq_stats {
93 struct u64_stats_sync syncp;
104 #define VIRTNET_SQ_STAT(m) offsetof(struct virtnet_sq_stats, m)
105 #define VIRTNET_RQ_STAT(m) offsetof(struct virtnet_rq_stats, m)
107 static const struct virtnet_stat_desc virtnet_sq_stats_desc[] = {
108 { "packets", VIRTNET_SQ_STAT(packets) },
109 { "bytes", VIRTNET_SQ_STAT(bytes) },
110 { "xdp_tx", VIRTNET_SQ_STAT(xdp_tx) },
111 { "xdp_tx_drops", VIRTNET_SQ_STAT(xdp_tx_drops) },
112 { "kicks", VIRTNET_SQ_STAT(kicks) },
115 static const struct virtnet_stat_desc virtnet_rq_stats_desc[] = {
116 { "packets", VIRTNET_RQ_STAT(packets) },
117 { "bytes", VIRTNET_RQ_STAT(bytes) },
118 { "drops", VIRTNET_RQ_STAT(drops) },
119 { "xdp_packets", VIRTNET_RQ_STAT(xdp_packets) },
120 { "xdp_tx", VIRTNET_RQ_STAT(xdp_tx) },
121 { "xdp_redirects", VIRTNET_RQ_STAT(xdp_redirects) },
122 { "xdp_drops", VIRTNET_RQ_STAT(xdp_drops) },
123 { "kicks", VIRTNET_RQ_STAT(kicks) },
126 #define VIRTNET_SQ_STATS_LEN ARRAY_SIZE(virtnet_sq_stats_desc)
127 #define VIRTNET_RQ_STATS_LEN ARRAY_SIZE(virtnet_rq_stats_desc)
129 /* Internal representation of a send virtqueue */
131 /* Virtqueue associated with this send _queue */
132 struct virtqueue *vq;
134 /* TX: fragments + linear part + virtio header */
135 struct scatterlist sg[MAX_SKB_FRAGS + 2];
137 /* Name of the send queue: output.$index */
140 struct virtnet_sq_stats stats;
142 struct napi_struct napi;
145 /* Internal representation of a receive virtqueue */
146 struct receive_queue {
147 /* Virtqueue associated with this receive_queue */
148 struct virtqueue *vq;
150 struct napi_struct napi;
152 struct bpf_prog __rcu *xdp_prog;
154 struct virtnet_rq_stats stats;
156 /* Chain pages by the private ptr. */
159 /* Average packet length for mergeable receive buffers. */
160 struct ewma_pkt_len mrg_avg_pkt_len;
162 /* Page frag for packet buffer allocation. */
163 struct page_frag alloc_frag;
165 /* RX: fragments + linear part + virtio header */
166 struct scatterlist sg[MAX_SKB_FRAGS + 2];
168 /* Min single buffer size for mergeable buffers case. */
169 unsigned int min_buf_len;
171 /* Name of this receive queue: input.$index */
174 struct xdp_rxq_info xdp_rxq;
177 /* Control VQ buffers: protected by the rtnl lock */
179 struct virtio_net_ctrl_hdr hdr;
180 virtio_net_ctrl_ack status;
181 struct virtio_net_ctrl_mq mq;
188 struct virtnet_info {
189 struct virtio_device *vdev;
190 struct virtqueue *cvq;
191 struct net_device *dev;
192 struct send_queue *sq;
193 struct receive_queue *rq;
196 /* Max # of queue pairs supported by the device */
199 /* # of queue pairs currently used by the driver */
200 u16 curr_queue_pairs;
202 /* # of XDP queue pairs currently used by the driver */
205 /* I like... big packets and I cannot lie! */
208 /* Host will merge rx buffers for big packets (shake it! shake it!) */
209 bool mergeable_rx_bufs;
211 /* Has control virtqueue */
214 /* Host can handle any s/g split between our header and packet data */
217 /* Packet virtio header size */
220 /* Work struct for refilling if we run low on memory. */
221 struct delayed_work refill;
223 /* Work struct for config space updates */
224 struct work_struct config_work;
226 /* Does the affinity hint is set for virtqueues? */
227 bool affinity_hint_set;
229 /* CPU hotplug instances for online & dead */
230 struct hlist_node node;
231 struct hlist_node node_dead;
233 struct control_buf *ctrl;
235 /* Ethtool settings */
239 unsigned long guest_offloads;
240 unsigned long guest_offloads_capable;
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,
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);
409 memcpy(hdr, p, hdr_len);
412 offset += hdr_padded_len;
416 if (copy > skb_tailroom(skb))
417 copy = skb_tailroom(skb);
418 skb_put_data(skb, p, copy);
423 if (vi->mergeable_rx_bufs) {
425 skb_add_rx_frag(skb, 0, page, offset, len, truesize);
432 * Verify that we can indeed put this data into a skb.
433 * This is here to handle cases when the device erroneously
434 * tries to receive more than is possible. This is usually
435 * the case of a broken device.
437 if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) {
438 net_dbg_ratelimited("%s: too much data\n", skb->dev->name);
442 BUG_ON(offset >= PAGE_SIZE);
444 unsigned int frag_size = min((unsigned)PAGE_SIZE - offset, len);
445 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page, offset,
446 frag_size, truesize);
448 page = (struct page *)page->private;
453 give_pages(rq, page);
458 static int __virtnet_xdp_xmit_one(struct virtnet_info *vi,
459 struct send_queue *sq,
460 struct xdp_frame *xdpf)
462 struct virtio_net_hdr_mrg_rxbuf *hdr;
465 /* virtqueue want to use data area in-front of packet */
466 if (unlikely(xdpf->metasize > 0))
469 if (unlikely(xdpf->headroom < vi->hdr_len))
472 /* Make room for virtqueue hdr (also change xdpf->headroom?) */
473 xdpf->data -= vi->hdr_len;
474 /* Zero header and leave csum up to XDP layers */
476 memset(hdr, 0, vi->hdr_len);
477 xdpf->len += vi->hdr_len;
479 sg_init_one(sq->sg, xdpf->data, xdpf->len);
481 err = virtqueue_add_outbuf(sq->vq, sq->sg, 1, xdp_to_ptr(xdpf),
484 return -ENOSPC; /* Caller handle free/refcnt */
489 static struct send_queue *virtnet_xdp_sq(struct virtnet_info *vi)
493 qp = vi->curr_queue_pairs - vi->xdp_queue_pairs + smp_processor_id();
497 static int virtnet_xdp_xmit(struct net_device *dev,
498 int n, struct xdp_frame **frames, u32 flags)
500 struct virtnet_info *vi = netdev_priv(dev);
501 struct receive_queue *rq = vi->rq;
502 struct bpf_prog *xdp_prog;
503 struct send_queue *sq;
513 /* Only allow ndo_xdp_xmit if XDP is loaded on dev, as this
514 * indicate XDP resources have been successfully allocated.
516 xdp_prog = rcu_dereference(rq->xdp_prog);
520 sq = virtnet_xdp_sq(vi);
522 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK)) {
528 /* Free up any pending old buffers before queueing new ones. */
529 while ((ptr = virtqueue_get_buf(sq->vq, &len)) != NULL) {
530 if (likely(is_xdp_frame(ptr))) {
531 struct xdp_frame *frame = ptr_to_xdp(ptr);
534 xdp_return_frame(frame);
536 struct sk_buff *skb = ptr;
539 napi_consume_skb(skb, false);
544 for (i = 0; i < n; i++) {
545 struct xdp_frame *xdpf = frames[i];
547 err = __virtnet_xdp_xmit_one(vi, sq, xdpf);
549 xdp_return_frame_rx_napi(xdpf);
555 if (flags & XDP_XMIT_FLUSH) {
556 if (virtqueue_kick_prepare(sq->vq) && virtqueue_notify(sq->vq))
560 u64_stats_update_begin(&sq->stats.syncp);
561 sq->stats.bytes += bytes;
562 sq->stats.packets += packets;
563 sq->stats.xdp_tx += n;
564 sq->stats.xdp_tx_drops += drops;
565 sq->stats.kicks += kicks;
566 u64_stats_update_end(&sq->stats.syncp);
571 static unsigned int virtnet_get_headroom(struct virtnet_info *vi)
573 return vi->xdp_queue_pairs ? VIRTIO_XDP_HEADROOM : 0;
576 /* We copy the packet for XDP in the following cases:
578 * 1) Packet is scattered across multiple rx buffers.
579 * 2) Headroom space is insufficient.
581 * This is inefficient but it's a temporary condition that
582 * we hit right after XDP is enabled and until queue is refilled
583 * with large buffers with sufficient headroom - so it should affect
584 * at most queue size packets.
585 * Afterwards, the conditions to enable
586 * XDP should preclude the underlying device from sending packets
587 * across multiple buffers (num_buf > 1), and we make sure buffers
588 * have enough headroom.
590 static struct page *xdp_linearize_page(struct receive_queue *rq,
597 struct page *page = alloc_page(GFP_ATOMIC);
602 memcpy(page_address(page) + page_off, page_address(p) + offset, *len);
606 int tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
611 buf = virtqueue_get_buf(rq->vq, &buflen);
615 p = virt_to_head_page(buf);
616 off = buf - page_address(p);
618 /* guard against a misconfigured or uncooperative backend that
619 * is sending packet larger than the MTU.
621 if ((page_off + buflen + tailroom) > PAGE_SIZE) {
626 memcpy(page_address(page) + page_off,
627 page_address(p) + off, buflen);
632 /* Headroom does not contribute to packet length */
633 *len = page_off - VIRTIO_XDP_HEADROOM;
636 __free_pages(page, 0);
640 static struct sk_buff *receive_small(struct net_device *dev,
641 struct virtnet_info *vi,
642 struct receive_queue *rq,
643 void *buf, void *ctx,
645 unsigned int *xdp_xmit,
646 struct virtnet_rq_stats *stats)
649 struct bpf_prog *xdp_prog;
650 unsigned int xdp_headroom = (unsigned long)ctx;
651 unsigned int header_offset = VIRTNET_RX_PAD + xdp_headroom;
652 unsigned int headroom = vi->hdr_len + header_offset;
653 unsigned int buflen = SKB_DATA_ALIGN(GOOD_PACKET_LEN + headroom) +
654 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
655 struct page *page = virt_to_head_page(buf);
656 unsigned int delta = 0;
657 struct page *xdp_page;
664 xdp_prog = rcu_dereference(rq->xdp_prog);
666 struct virtio_net_hdr_mrg_rxbuf *hdr = buf + header_offset;
667 struct xdp_frame *xdpf;
672 if (unlikely(hdr->hdr.gso_type))
675 if (unlikely(xdp_headroom < virtnet_get_headroom(vi))) {
676 int offset = buf - page_address(page) + header_offset;
677 unsigned int tlen = len + vi->hdr_len;
680 xdp_headroom = virtnet_get_headroom(vi);
681 header_offset = VIRTNET_RX_PAD + xdp_headroom;
682 headroom = vi->hdr_len + header_offset;
683 buflen = SKB_DATA_ALIGN(GOOD_PACKET_LEN + headroom) +
684 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
685 xdp_page = xdp_linearize_page(rq, &num_buf, page,
686 offset, header_offset,
691 buf = page_address(xdp_page);
696 xdp.data_hard_start = buf + VIRTNET_RX_PAD + vi->hdr_len;
697 xdp.data = xdp.data_hard_start + xdp_headroom;
698 xdp_set_data_meta_invalid(&xdp);
699 xdp.data_end = xdp.data + len;
700 xdp.rxq = &rq->xdp_rxq;
701 orig_data = xdp.data;
702 act = bpf_prog_run_xdp(xdp_prog, &xdp);
703 stats->xdp_packets++;
707 /* Recalculate length in case bpf program changed it */
708 delta = orig_data - xdp.data;
709 len = xdp.data_end - xdp.data;
713 xdpf = convert_to_xdp_frame(&xdp);
716 err = virtnet_xdp_xmit(dev, 1, &xdpf, 0);
717 if (unlikely(err < 0)) {
718 trace_xdp_exception(vi->dev, xdp_prog, act);
721 *xdp_xmit |= VIRTIO_XDP_TX;
725 stats->xdp_redirects++;
726 err = xdp_do_redirect(dev, &xdp, xdp_prog);
729 *xdp_xmit |= VIRTIO_XDP_REDIR;
733 bpf_warn_invalid_xdp_action(act);
736 trace_xdp_exception(vi->dev, xdp_prog, act);
743 skb = build_skb(buf, buflen);
748 skb_reserve(skb, headroom - delta);
751 buf += header_offset;
752 memcpy(skb_vnet_hdr(skb), buf, vi->hdr_len);
753 } /* keep zeroed vnet hdr since packet was changed by bpf */
767 static struct sk_buff *receive_big(struct net_device *dev,
768 struct virtnet_info *vi,
769 struct receive_queue *rq,
772 struct virtnet_rq_stats *stats)
774 struct page *page = buf;
775 struct sk_buff *skb = page_to_skb(vi, rq, page, 0, len,
778 stats->bytes += len - vi->hdr_len;
786 give_pages(rq, page);
790 static struct sk_buff *receive_mergeable(struct net_device *dev,
791 struct virtnet_info *vi,
792 struct receive_queue *rq,
796 unsigned int *xdp_xmit,
797 struct virtnet_rq_stats *stats)
799 struct virtio_net_hdr_mrg_rxbuf *hdr = buf;
800 u16 num_buf = virtio16_to_cpu(vi->vdev, hdr->num_buffers);
801 struct page *page = virt_to_head_page(buf);
802 int offset = buf - page_address(page);
803 struct sk_buff *head_skb, *curr_skb;
804 struct bpf_prog *xdp_prog;
805 unsigned int truesize;
806 unsigned int headroom = mergeable_ctx_to_headroom(ctx);
810 stats->bytes += len - vi->hdr_len;
813 xdp_prog = rcu_dereference(rq->xdp_prog);
815 struct xdp_frame *xdpf;
816 struct page *xdp_page;
821 /* Transient failure which in theory could occur if
822 * in-flight packets from before XDP was enabled reach
823 * the receive path after XDP is loaded.
825 if (unlikely(hdr->hdr.gso_type))
828 /* This happens when rx buffer size is underestimated
829 * or headroom is not enough because of the buffer
830 * was refilled before XDP is set. This should only
831 * happen for the first several packets, so we don't
832 * care much about its performance.
834 if (unlikely(num_buf > 1 ||
835 headroom < virtnet_get_headroom(vi))) {
836 /* linearize data for XDP */
837 xdp_page = xdp_linearize_page(rq, &num_buf,
843 offset = VIRTIO_XDP_HEADROOM;
848 /* Allow consuming headroom but reserve enough space to push
849 * the descriptor on if we get an XDP_TX return code.
851 data = page_address(xdp_page) + offset;
852 xdp.data_hard_start = data - VIRTIO_XDP_HEADROOM + vi->hdr_len;
853 xdp.data = data + vi->hdr_len;
854 xdp_set_data_meta_invalid(&xdp);
855 xdp.data_end = xdp.data + (len - vi->hdr_len);
856 xdp.rxq = &rq->xdp_rxq;
858 act = bpf_prog_run_xdp(xdp_prog, &xdp);
859 stats->xdp_packets++;
863 /* recalculate offset to account for any header
864 * adjustments. Note other cases do not build an
865 * skb and avoid using offset
868 page_address(xdp_page) - vi->hdr_len;
870 /* recalculate len if xdp.data or xdp.data_end were
873 len = xdp.data_end - xdp.data + vi->hdr_len;
874 /* We can only create skb based on xdp_page. */
875 if (unlikely(xdp_page != page)) {
878 head_skb = page_to_skb(vi, rq, xdp_page,
886 xdpf = convert_to_xdp_frame(&xdp);
889 err = virtnet_xdp_xmit(dev, 1, &xdpf, 0);
890 if (unlikely(err < 0)) {
891 trace_xdp_exception(vi->dev, xdp_prog, act);
892 if (unlikely(xdp_page != page))
896 *xdp_xmit |= VIRTIO_XDP_TX;
897 if (unlikely(xdp_page != page))
902 stats->xdp_redirects++;
903 err = xdp_do_redirect(dev, &xdp, xdp_prog);
905 if (unlikely(xdp_page != page))
909 *xdp_xmit |= VIRTIO_XDP_REDIR;
910 if (unlikely(xdp_page != page))
915 bpf_warn_invalid_xdp_action(act);
918 trace_xdp_exception(vi->dev, xdp_prog, act);
921 if (unlikely(xdp_page != page))
922 __free_pages(xdp_page, 0);
928 truesize = mergeable_ctx_to_truesize(ctx);
929 if (unlikely(len > truesize)) {
930 pr_debug("%s: rx error: len %u exceeds truesize %lu\n",
931 dev->name, len, (unsigned long)ctx);
932 dev->stats.rx_length_errors++;
936 head_skb = page_to_skb(vi, rq, page, offset, len, truesize, !xdp_prog);
939 if (unlikely(!curr_skb))
944 buf = virtqueue_get_buf_ctx(rq->vq, &len, &ctx);
945 if (unlikely(!buf)) {
946 pr_debug("%s: rx error: %d buffers out of %d missing\n",
948 virtio16_to_cpu(vi->vdev,
950 dev->stats.rx_length_errors++;
955 page = virt_to_head_page(buf);
957 truesize = mergeable_ctx_to_truesize(ctx);
958 if (unlikely(len > truesize)) {
959 pr_debug("%s: rx error: len %u exceeds truesize %lu\n",
960 dev->name, len, (unsigned long)ctx);
961 dev->stats.rx_length_errors++;
965 num_skb_frags = skb_shinfo(curr_skb)->nr_frags;
966 if (unlikely(num_skb_frags == MAX_SKB_FRAGS)) {
967 struct sk_buff *nskb = alloc_skb(0, GFP_ATOMIC);
971 if (curr_skb == head_skb)
972 skb_shinfo(curr_skb)->frag_list = nskb;
974 curr_skb->next = nskb;
976 head_skb->truesize += nskb->truesize;
979 if (curr_skb != head_skb) {
980 head_skb->data_len += len;
981 head_skb->len += len;
982 head_skb->truesize += truesize;
984 offset = buf - page_address(page);
985 if (skb_can_coalesce(curr_skb, num_skb_frags, page, offset)) {
987 skb_coalesce_rx_frag(curr_skb, num_skb_frags - 1,
990 skb_add_rx_frag(curr_skb, num_skb_frags, page,
991 offset, len, truesize);
995 ewma_pkt_len_add(&rq->mrg_avg_pkt_len, head_skb->len);
1003 while (num_buf-- > 1) {
1004 buf = virtqueue_get_buf(rq->vq, &len);
1005 if (unlikely(!buf)) {
1006 pr_debug("%s: rx error: %d buffers missing\n",
1007 dev->name, num_buf);
1008 dev->stats.rx_length_errors++;
1011 stats->bytes += len;
1012 page = virt_to_head_page(buf);
1017 dev_kfree_skb(head_skb);
1022 static void receive_buf(struct virtnet_info *vi, struct receive_queue *rq,
1023 void *buf, unsigned int len, void **ctx,
1024 unsigned int *xdp_xmit,
1025 struct virtnet_rq_stats *stats)
1027 struct net_device *dev = vi->dev;
1028 struct sk_buff *skb;
1029 struct virtio_net_hdr_mrg_rxbuf *hdr;
1031 if (unlikely(len < vi->hdr_len + ETH_HLEN)) {
1032 pr_debug("%s: short packet %i\n", dev->name, len);
1033 dev->stats.rx_length_errors++;
1034 if (vi->mergeable_rx_bufs) {
1035 put_page(virt_to_head_page(buf));
1036 } else if (vi->big_packets) {
1037 give_pages(rq, buf);
1039 put_page(virt_to_head_page(buf));
1044 if (vi->mergeable_rx_bufs)
1045 skb = receive_mergeable(dev, vi, rq, buf, ctx, len, xdp_xmit,
1047 else if (vi->big_packets)
1048 skb = receive_big(dev, vi, rq, buf, len, stats);
1050 skb = receive_small(dev, vi, rq, buf, ctx, len, xdp_xmit, stats);
1055 hdr = skb_vnet_hdr(skb);
1057 if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID)
1058 skb->ip_summed = CHECKSUM_UNNECESSARY;
1060 if (virtio_net_hdr_to_skb(skb, &hdr->hdr,
1061 virtio_is_little_endian(vi->vdev))) {
1062 net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n",
1063 dev->name, hdr->hdr.gso_type,
1068 skb_record_rx_queue(skb, vq2rxq(rq->vq));
1069 skb->protocol = eth_type_trans(skb, dev);
1070 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
1071 ntohs(skb->protocol), skb->len, skb->pkt_type);
1073 napi_gro_receive(&rq->napi, skb);
1077 dev->stats.rx_frame_errors++;
1081 /* Unlike mergeable buffers, all buffers are allocated to the
1082 * same size, except for the headroom. For this reason we do
1083 * not need to use mergeable_len_to_ctx here - it is enough
1084 * to store the headroom as the context ignoring the truesize.
1086 static int add_recvbuf_small(struct virtnet_info *vi, struct receive_queue *rq,
1089 struct page_frag *alloc_frag = &rq->alloc_frag;
1091 unsigned int xdp_headroom = virtnet_get_headroom(vi);
1092 void *ctx = (void *)(unsigned long)xdp_headroom;
1093 int len = vi->hdr_len + VIRTNET_RX_PAD + GOOD_PACKET_LEN + xdp_headroom;
1096 len = SKB_DATA_ALIGN(len) +
1097 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1098 if (unlikely(!skb_page_frag_refill(len, alloc_frag, gfp)))
1101 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1102 get_page(alloc_frag->page);
1103 alloc_frag->offset += len;
1104 sg_init_one(rq->sg, buf + VIRTNET_RX_PAD + xdp_headroom,
1105 vi->hdr_len + GOOD_PACKET_LEN);
1106 err = virtqueue_add_inbuf_ctx(rq->vq, rq->sg, 1, buf, ctx, gfp);
1108 put_page(virt_to_head_page(buf));
1112 static int add_recvbuf_big(struct virtnet_info *vi, struct receive_queue *rq,
1115 struct page *first, *list = NULL;
1119 sg_init_table(rq->sg, MAX_SKB_FRAGS + 2);
1121 /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */
1122 for (i = MAX_SKB_FRAGS + 1; i > 1; --i) {
1123 first = get_a_page(rq, gfp);
1126 give_pages(rq, list);
1129 sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE);
1131 /* chain new page in list head to match sg */
1132 first->private = (unsigned long)list;
1136 first = get_a_page(rq, gfp);
1138 give_pages(rq, list);
1141 p = page_address(first);
1143 /* rq->sg[0], rq->sg[1] share the same page */
1144 /* a separated rq->sg[0] for header - required in case !any_header_sg */
1145 sg_set_buf(&rq->sg[0], p, vi->hdr_len);
1147 /* rq->sg[1] for data packet, from offset */
1148 offset = sizeof(struct padded_vnet_hdr);
1149 sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset);
1151 /* chain first in list head */
1152 first->private = (unsigned long)list;
1153 err = virtqueue_add_inbuf(rq->vq, rq->sg, MAX_SKB_FRAGS + 2,
1156 give_pages(rq, first);
1161 static unsigned int get_mergeable_buf_len(struct receive_queue *rq,
1162 struct ewma_pkt_len *avg_pkt_len,
1165 const size_t hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1169 return PAGE_SIZE - room;
1171 len = hdr_len + clamp_t(unsigned int, ewma_pkt_len_read(avg_pkt_len),
1172 rq->min_buf_len, PAGE_SIZE - hdr_len);
1174 return ALIGN(len, L1_CACHE_BYTES);
1177 static int add_recvbuf_mergeable(struct virtnet_info *vi,
1178 struct receive_queue *rq, gfp_t gfp)
1180 struct page_frag *alloc_frag = &rq->alloc_frag;
1181 unsigned int headroom = virtnet_get_headroom(vi);
1182 unsigned int tailroom = headroom ? sizeof(struct skb_shared_info) : 0;
1183 unsigned int room = SKB_DATA_ALIGN(headroom + tailroom);
1187 unsigned int len, hole;
1189 /* Extra tailroom is needed to satisfy XDP's assumption. This
1190 * means rx frags coalescing won't work, but consider we've
1191 * disabled GSO for XDP, it won't be a big issue.
1193 len = get_mergeable_buf_len(rq, &rq->mrg_avg_pkt_len, room);
1194 if (unlikely(!skb_page_frag_refill(len + room, alloc_frag, gfp)))
1197 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1198 buf += headroom; /* advance address leaving hole at front of pkt */
1199 get_page(alloc_frag->page);
1200 alloc_frag->offset += len + room;
1201 hole = alloc_frag->size - alloc_frag->offset;
1202 if (hole < len + room) {
1203 /* To avoid internal fragmentation, if there is very likely not
1204 * enough space for another buffer, add the remaining space to
1205 * the current buffer.
1208 alloc_frag->offset += hole;
1211 sg_init_one(rq->sg, buf, len);
1212 ctx = mergeable_len_to_ctx(len, headroom);
1213 err = virtqueue_add_inbuf_ctx(rq->vq, rq->sg, 1, buf, ctx, gfp);
1215 put_page(virt_to_head_page(buf));
1221 * Returns false if we couldn't fill entirely (OOM).
1223 * Normally run in the receive path, but can also be run from ndo_open
1224 * before we're receiving packets, or from refill_work which is
1225 * careful to disable receiving (using napi_disable).
1227 static bool try_fill_recv(struct virtnet_info *vi, struct receive_queue *rq,
1234 if (vi->mergeable_rx_bufs)
1235 err = add_recvbuf_mergeable(vi, rq, gfp);
1236 else if (vi->big_packets)
1237 err = add_recvbuf_big(vi, rq, gfp);
1239 err = add_recvbuf_small(vi, rq, gfp);
1241 oom = err == -ENOMEM;
1244 } while (rq->vq->num_free);
1245 if (virtqueue_kick_prepare(rq->vq) && virtqueue_notify(rq->vq)) {
1246 u64_stats_update_begin(&rq->stats.syncp);
1248 u64_stats_update_end(&rq->stats.syncp);
1254 static void skb_recv_done(struct virtqueue *rvq)
1256 struct virtnet_info *vi = rvq->vdev->priv;
1257 struct receive_queue *rq = &vi->rq[vq2rxq(rvq)];
1259 virtqueue_napi_schedule(&rq->napi, rvq);
1262 static void virtnet_napi_enable(struct virtqueue *vq, struct napi_struct *napi)
1266 /* If all buffers were filled by other side before we napi_enabled, we
1267 * won't get another interrupt, so process any outstanding packets now.
1268 * Call local_bh_enable after to trigger softIRQ processing.
1271 virtqueue_napi_schedule(napi, vq);
1275 static void virtnet_napi_tx_enable(struct virtnet_info *vi,
1276 struct virtqueue *vq,
1277 struct napi_struct *napi)
1282 /* Tx napi touches cachelines on the cpu handling tx interrupts. Only
1283 * enable the feature if this is likely affine with the transmit path.
1285 if (!vi->affinity_hint_set) {
1290 return virtnet_napi_enable(vq, napi);
1293 static void virtnet_napi_tx_disable(struct napi_struct *napi)
1299 static void refill_work(struct work_struct *work)
1301 struct virtnet_info *vi =
1302 container_of(work, struct virtnet_info, refill.work);
1306 for (i = 0; i < vi->curr_queue_pairs; i++) {
1307 struct receive_queue *rq = &vi->rq[i];
1309 napi_disable(&rq->napi);
1310 still_empty = !try_fill_recv(vi, rq, GFP_KERNEL);
1311 virtnet_napi_enable(rq->vq, &rq->napi);
1313 /* In theory, this can happen: if we don't get any buffers in
1314 * we will *never* try to fill again.
1317 schedule_delayed_work(&vi->refill, HZ/2);
1321 static int virtnet_receive(struct receive_queue *rq, int budget,
1322 unsigned int *xdp_xmit)
1324 struct virtnet_info *vi = rq->vq->vdev->priv;
1325 struct virtnet_rq_stats stats = {};
1330 if (!vi->big_packets || vi->mergeable_rx_bufs) {
1333 while (stats.packets < budget &&
1334 (buf = virtqueue_get_buf_ctx(rq->vq, &len, &ctx))) {
1335 receive_buf(vi, rq, buf, len, ctx, xdp_xmit, &stats);
1339 while (stats.packets < budget &&
1340 (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) {
1341 receive_buf(vi, rq, buf, len, NULL, xdp_xmit, &stats);
1346 if (rq->vq->num_free > virtqueue_get_vring_size(rq->vq) / 2) {
1347 if (!try_fill_recv(vi, rq, GFP_ATOMIC))
1348 schedule_delayed_work(&vi->refill, 0);
1351 u64_stats_update_begin(&rq->stats.syncp);
1352 for (i = 0; i < VIRTNET_RQ_STATS_LEN; i++) {
1353 size_t offset = virtnet_rq_stats_desc[i].offset;
1356 item = (u64 *)((u8 *)&rq->stats + offset);
1357 *item += *(u64 *)((u8 *)&stats + offset);
1359 u64_stats_update_end(&rq->stats.syncp);
1361 return stats.packets;
1364 static void free_old_xmit_skbs(struct send_queue *sq, bool in_napi)
1367 unsigned int packets = 0;
1368 unsigned int bytes = 0;
1371 while ((ptr = virtqueue_get_buf(sq->vq, &len)) != NULL) {
1372 if (likely(!is_xdp_frame(ptr))) {
1373 struct sk_buff *skb = ptr;
1375 pr_debug("Sent skb %p\n", skb);
1378 napi_consume_skb(skb, in_napi);
1380 struct xdp_frame *frame = ptr_to_xdp(ptr);
1382 bytes += frame->len;
1383 xdp_return_frame(frame);
1388 /* Avoid overhead when no packets have been processed
1389 * happens when called speculatively from start_xmit.
1394 u64_stats_update_begin(&sq->stats.syncp);
1395 sq->stats.bytes += bytes;
1396 sq->stats.packets += packets;
1397 u64_stats_update_end(&sq->stats.syncp);
1400 static bool is_xdp_raw_buffer_queue(struct virtnet_info *vi, int q)
1402 if (q < (vi->curr_queue_pairs - vi->xdp_queue_pairs))
1404 else if (q < vi->curr_queue_pairs)
1410 static void virtnet_poll_cleantx(struct receive_queue *rq)
1412 struct virtnet_info *vi = rq->vq->vdev->priv;
1413 unsigned int index = vq2rxq(rq->vq);
1414 struct send_queue *sq = &vi->sq[index];
1415 struct netdev_queue *txq = netdev_get_tx_queue(vi->dev, index);
1417 if (!sq->napi.weight || is_xdp_raw_buffer_queue(vi, index))
1420 if (__netif_tx_trylock(txq)) {
1421 free_old_xmit_skbs(sq, true);
1422 __netif_tx_unlock(txq);
1425 if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS)
1426 netif_tx_wake_queue(txq);
1429 static int virtnet_poll(struct napi_struct *napi, int budget)
1431 struct receive_queue *rq =
1432 container_of(napi, struct receive_queue, napi);
1433 struct virtnet_info *vi = rq->vq->vdev->priv;
1434 struct send_queue *sq;
1435 unsigned int received;
1436 unsigned int xdp_xmit = 0;
1438 virtnet_poll_cleantx(rq);
1440 received = virtnet_receive(rq, budget, &xdp_xmit);
1442 /* Out of packets? */
1443 if (received < budget)
1444 virtqueue_napi_complete(napi, rq->vq, received);
1446 if (xdp_xmit & VIRTIO_XDP_REDIR)
1449 if (xdp_xmit & VIRTIO_XDP_TX) {
1450 sq = virtnet_xdp_sq(vi);
1451 if (virtqueue_kick_prepare(sq->vq) && virtqueue_notify(sq->vq)) {
1452 u64_stats_update_begin(&sq->stats.syncp);
1454 u64_stats_update_end(&sq->stats.syncp);
1461 static int virtnet_open(struct net_device *dev)
1463 struct virtnet_info *vi = netdev_priv(dev);
1466 for (i = 0; i < vi->max_queue_pairs; i++) {
1467 if (i < vi->curr_queue_pairs)
1468 /* Make sure we have some buffers: if oom use wq. */
1469 if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL))
1470 schedule_delayed_work(&vi->refill, 0);
1472 err = xdp_rxq_info_reg(&vi->rq[i].xdp_rxq, dev, i);
1476 err = xdp_rxq_info_reg_mem_model(&vi->rq[i].xdp_rxq,
1477 MEM_TYPE_PAGE_SHARED, NULL);
1479 xdp_rxq_info_unreg(&vi->rq[i].xdp_rxq);
1483 virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi);
1484 virtnet_napi_tx_enable(vi, vi->sq[i].vq, &vi->sq[i].napi);
1490 static int virtnet_poll_tx(struct napi_struct *napi, int budget)
1492 struct send_queue *sq = container_of(napi, struct send_queue, napi);
1493 struct virtnet_info *vi = sq->vq->vdev->priv;
1494 unsigned int index = vq2txq(sq->vq);
1495 struct netdev_queue *txq;
1497 if (unlikely(is_xdp_raw_buffer_queue(vi, index))) {
1498 /* We don't need to enable cb for XDP */
1499 napi_complete_done(napi, 0);
1503 txq = netdev_get_tx_queue(vi->dev, index);
1504 __netif_tx_lock(txq, raw_smp_processor_id());
1505 free_old_xmit_skbs(sq, true);
1506 __netif_tx_unlock(txq);
1508 virtqueue_napi_complete(napi, sq->vq, 0);
1510 if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS)
1511 netif_tx_wake_queue(txq);
1516 static int xmit_skb(struct send_queue *sq, struct sk_buff *skb)
1518 struct virtio_net_hdr_mrg_rxbuf *hdr;
1519 const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
1520 struct virtnet_info *vi = sq->vq->vdev->priv;
1522 unsigned hdr_len = vi->hdr_len;
1525 pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
1527 can_push = vi->any_header_sg &&
1528 !((unsigned long)skb->data & (__alignof__(*hdr) - 1)) &&
1529 !skb_header_cloned(skb) && skb_headroom(skb) >= hdr_len;
1530 /* Even if we can, don't push here yet as this would skew
1531 * csum_start offset below. */
1533 hdr = (struct virtio_net_hdr_mrg_rxbuf *)(skb->data - hdr_len);
1535 hdr = skb_vnet_hdr(skb);
1537 if (virtio_net_hdr_from_skb(skb, &hdr->hdr,
1538 virtio_is_little_endian(vi->vdev), false,
1542 if (vi->mergeable_rx_bufs)
1543 hdr->num_buffers = 0;
1545 sg_init_table(sq->sg, skb_shinfo(skb)->nr_frags + (can_push ? 1 : 2));
1547 __skb_push(skb, hdr_len);
1548 num_sg = skb_to_sgvec(skb, sq->sg, 0, skb->len);
1549 if (unlikely(num_sg < 0))
1551 /* Pull header back to avoid skew in tx bytes calculations. */
1552 __skb_pull(skb, hdr_len);
1554 sg_set_buf(sq->sg, hdr, hdr_len);
1555 num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len);
1556 if (unlikely(num_sg < 0))
1560 return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC);
1563 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
1565 struct virtnet_info *vi = netdev_priv(dev);
1566 int qnum = skb_get_queue_mapping(skb);
1567 struct send_queue *sq = &vi->sq[qnum];
1569 struct netdev_queue *txq = netdev_get_tx_queue(dev, qnum);
1570 bool kick = !netdev_xmit_more();
1571 bool use_napi = sq->napi.weight;
1573 /* Free up any pending old buffers before queueing new ones. */
1574 free_old_xmit_skbs(sq, false);
1576 if (use_napi && kick)
1577 virtqueue_enable_cb_delayed(sq->vq);
1579 /* timestamp packet in software */
1580 skb_tx_timestamp(skb);
1582 /* Try to transmit */
1583 err = xmit_skb(sq, skb);
1585 /* This should not happen! */
1586 if (unlikely(err)) {
1587 dev->stats.tx_fifo_errors++;
1588 if (net_ratelimit())
1590 "Unexpected TXQ (%d) queue failure: %d\n",
1592 dev->stats.tx_dropped++;
1593 dev_kfree_skb_any(skb);
1594 return NETDEV_TX_OK;
1597 /* Don't wait up for transmitted skbs to be freed. */
1603 /* If running out of space, stop queue to avoid getting packets that we
1604 * are then unable to transmit.
1605 * An alternative would be to force queuing layer to requeue the skb by
1606 * returning NETDEV_TX_BUSY. However, NETDEV_TX_BUSY should not be
1607 * returned in a normal path of operation: it means that driver is not
1608 * maintaining the TX queue stop/start state properly, and causes
1609 * the stack to do a non-trivial amount of useless work.
1610 * Since most packets only take 1 or 2 ring slots, stopping the queue
1611 * early means 16 slots are typically wasted.
1613 if (sq->vq->num_free < 2+MAX_SKB_FRAGS) {
1614 netif_stop_subqueue(dev, qnum);
1616 unlikely(!virtqueue_enable_cb_delayed(sq->vq))) {
1617 /* More just got used, free them then recheck. */
1618 free_old_xmit_skbs(sq, false);
1619 if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) {
1620 netif_start_subqueue(dev, qnum);
1621 virtqueue_disable_cb(sq->vq);
1626 if (kick || netif_xmit_stopped(txq)) {
1627 if (virtqueue_kick_prepare(sq->vq) && virtqueue_notify(sq->vq)) {
1628 u64_stats_update_begin(&sq->stats.syncp);
1630 u64_stats_update_end(&sq->stats.syncp);
1634 return NETDEV_TX_OK;
1638 * Send command via the control virtqueue and check status. Commands
1639 * supported by the hypervisor, as indicated by feature bits, should
1640 * never fail unless improperly formatted.
1642 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
1643 struct scatterlist *out)
1645 struct scatterlist *sgs[4], hdr, stat;
1646 unsigned out_num = 0, tmp;
1648 /* Caller should know better */
1649 BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ));
1651 vi->ctrl->status = ~0;
1652 vi->ctrl->hdr.class = class;
1653 vi->ctrl->hdr.cmd = cmd;
1655 sg_init_one(&hdr, &vi->ctrl->hdr, sizeof(vi->ctrl->hdr));
1656 sgs[out_num++] = &hdr;
1659 sgs[out_num++] = out;
1661 /* Add return status. */
1662 sg_init_one(&stat, &vi->ctrl->status, sizeof(vi->ctrl->status));
1663 sgs[out_num] = &stat;
1665 BUG_ON(out_num + 1 > ARRAY_SIZE(sgs));
1666 virtqueue_add_sgs(vi->cvq, sgs, out_num, 1, vi, GFP_ATOMIC);
1668 if (unlikely(!virtqueue_kick(vi->cvq)))
1669 return vi->ctrl->status == VIRTIO_NET_OK;
1671 /* Spin for a response, the kick causes an ioport write, trapping
1672 * into the hypervisor, so the request should be handled immediately.
1674 while (!virtqueue_get_buf(vi->cvq, &tmp) &&
1675 !virtqueue_is_broken(vi->cvq))
1678 return vi->ctrl->status == VIRTIO_NET_OK;
1681 static int virtnet_set_mac_address(struct net_device *dev, void *p)
1683 struct virtnet_info *vi = netdev_priv(dev);
1684 struct virtio_device *vdev = vi->vdev;
1686 struct sockaddr *addr;
1687 struct scatterlist sg;
1689 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STANDBY))
1692 addr = kmemdup(p, sizeof(*addr), GFP_KERNEL);
1696 ret = eth_prepare_mac_addr_change(dev, addr);
1700 if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1701 sg_init_one(&sg, addr->sa_data, dev->addr_len);
1702 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
1703 VIRTIO_NET_CTRL_MAC_ADDR_SET, &sg)) {
1704 dev_warn(&vdev->dev,
1705 "Failed to set mac address by vq command.\n");
1709 } else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC) &&
1710 !virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) {
1713 /* Naturally, this has an atomicity problem. */
1714 for (i = 0; i < dev->addr_len; i++)
1715 virtio_cwrite8(vdev,
1716 offsetof(struct virtio_net_config, mac) +
1717 i, addr->sa_data[i]);
1720 eth_commit_mac_addr_change(dev, p);
1728 static void virtnet_stats(struct net_device *dev,
1729 struct rtnl_link_stats64 *tot)
1731 struct virtnet_info *vi = netdev_priv(dev);
1735 for (i = 0; i < vi->max_queue_pairs; i++) {
1736 u64 tpackets, tbytes, rpackets, rbytes, rdrops;
1737 struct receive_queue *rq = &vi->rq[i];
1738 struct send_queue *sq = &vi->sq[i];
1741 start = u64_stats_fetch_begin_irq(&sq->stats.syncp);
1742 tpackets = sq->stats.packets;
1743 tbytes = sq->stats.bytes;
1744 } while (u64_stats_fetch_retry_irq(&sq->stats.syncp, start));
1747 start = u64_stats_fetch_begin_irq(&rq->stats.syncp);
1748 rpackets = rq->stats.packets;
1749 rbytes = rq->stats.bytes;
1750 rdrops = rq->stats.drops;
1751 } while (u64_stats_fetch_retry_irq(&rq->stats.syncp, start));
1753 tot->rx_packets += rpackets;
1754 tot->tx_packets += tpackets;
1755 tot->rx_bytes += rbytes;
1756 tot->tx_bytes += tbytes;
1757 tot->rx_dropped += rdrops;
1760 tot->tx_dropped = dev->stats.tx_dropped;
1761 tot->tx_fifo_errors = dev->stats.tx_fifo_errors;
1762 tot->rx_length_errors = dev->stats.rx_length_errors;
1763 tot->rx_frame_errors = dev->stats.rx_frame_errors;
1766 static void virtnet_ack_link_announce(struct virtnet_info *vi)
1769 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE,
1770 VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL))
1771 dev_warn(&vi->dev->dev, "Failed to ack link announce.\n");
1775 static int _virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs)
1777 struct scatterlist sg;
1778 struct net_device *dev = vi->dev;
1780 if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ))
1783 vi->ctrl->mq.virtqueue_pairs = cpu_to_virtio16(vi->vdev, queue_pairs);
1784 sg_init_one(&sg, &vi->ctrl->mq, sizeof(vi->ctrl->mq));
1786 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ,
1787 VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg)) {
1788 dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n",
1792 vi->curr_queue_pairs = queue_pairs;
1793 /* virtnet_open() will refill when device is going to up. */
1794 if (dev->flags & IFF_UP)
1795 schedule_delayed_work(&vi->refill, 0);
1801 static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs)
1806 err = _virtnet_set_queues(vi, queue_pairs);
1811 static int virtnet_close(struct net_device *dev)
1813 struct virtnet_info *vi = netdev_priv(dev);
1816 /* Make sure refill_work doesn't re-enable napi! */
1817 cancel_delayed_work_sync(&vi->refill);
1819 for (i = 0; i < vi->max_queue_pairs; i++) {
1820 xdp_rxq_info_unreg(&vi->rq[i].xdp_rxq);
1821 napi_disable(&vi->rq[i].napi);
1822 virtnet_napi_tx_disable(&vi->sq[i].napi);
1828 static void virtnet_set_rx_mode(struct net_device *dev)
1830 struct virtnet_info *vi = netdev_priv(dev);
1831 struct scatterlist sg[2];
1832 struct virtio_net_ctrl_mac *mac_data;
1833 struct netdev_hw_addr *ha;
1839 /* We can't dynamically set ndo_set_rx_mode, so return gracefully */
1840 if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
1843 vi->ctrl->promisc = ((dev->flags & IFF_PROMISC) != 0);
1844 vi->ctrl->allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
1846 sg_init_one(sg, &vi->ctrl->promisc, sizeof(vi->ctrl->promisc));
1848 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
1849 VIRTIO_NET_CTRL_RX_PROMISC, sg))
1850 dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
1851 vi->ctrl->promisc ? "en" : "dis");
1853 sg_init_one(sg, &vi->ctrl->allmulti, sizeof(vi->ctrl->allmulti));
1855 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
1856 VIRTIO_NET_CTRL_RX_ALLMULTI, sg))
1857 dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
1858 vi->ctrl->allmulti ? "en" : "dis");
1860 uc_count = netdev_uc_count(dev);
1861 mc_count = netdev_mc_count(dev);
1862 /* MAC filter - use one buffer for both lists */
1863 buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
1864 (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
1869 sg_init_table(sg, 2);
1871 /* Store the unicast list and count in the front of the buffer */
1872 mac_data->entries = cpu_to_virtio32(vi->vdev, uc_count);
1874 netdev_for_each_uc_addr(ha, dev)
1875 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
1877 sg_set_buf(&sg[0], mac_data,
1878 sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
1880 /* multicast list and count fill the end */
1881 mac_data = (void *)&mac_data->macs[uc_count][0];
1883 mac_data->entries = cpu_to_virtio32(vi->vdev, mc_count);
1885 netdev_for_each_mc_addr(ha, dev)
1886 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
1888 sg_set_buf(&sg[1], mac_data,
1889 sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
1891 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
1892 VIRTIO_NET_CTRL_MAC_TABLE_SET, sg))
1893 dev_warn(&dev->dev, "Failed to set MAC filter table.\n");
1898 static int virtnet_vlan_rx_add_vid(struct net_device *dev,
1899 __be16 proto, u16 vid)
1901 struct virtnet_info *vi = netdev_priv(dev);
1902 struct scatterlist sg;
1904 vi->ctrl->vid = cpu_to_virtio16(vi->vdev, vid);
1905 sg_init_one(&sg, &vi->ctrl->vid, sizeof(vi->ctrl->vid));
1907 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1908 VIRTIO_NET_CTRL_VLAN_ADD, &sg))
1909 dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
1913 static int virtnet_vlan_rx_kill_vid(struct net_device *dev,
1914 __be16 proto, u16 vid)
1916 struct virtnet_info *vi = netdev_priv(dev);
1917 struct scatterlist sg;
1919 vi->ctrl->vid = cpu_to_virtio16(vi->vdev, vid);
1920 sg_init_one(&sg, &vi->ctrl->vid, sizeof(vi->ctrl->vid));
1922 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1923 VIRTIO_NET_CTRL_VLAN_DEL, &sg))
1924 dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
1928 static void virtnet_clean_affinity(struct virtnet_info *vi)
1932 if (vi->affinity_hint_set) {
1933 for (i = 0; i < vi->max_queue_pairs; i++) {
1934 virtqueue_set_affinity(vi->rq[i].vq, NULL);
1935 virtqueue_set_affinity(vi->sq[i].vq, NULL);
1938 vi->affinity_hint_set = false;
1942 static void virtnet_set_affinity(struct virtnet_info *vi)
1951 if (!zalloc_cpumask_var(&mask, GFP_KERNEL)) {
1952 virtnet_clean_affinity(vi);
1956 num_cpu = num_online_cpus();
1957 stride = max_t(int, num_cpu / vi->curr_queue_pairs, 1);
1958 stragglers = num_cpu >= vi->curr_queue_pairs ?
1959 num_cpu % vi->curr_queue_pairs :
1961 cpu = cpumask_next(-1, cpu_online_mask);
1963 for (i = 0; i < vi->curr_queue_pairs; i++) {
1964 group_size = stride + (i < stragglers ? 1 : 0);
1966 for (j = 0; j < group_size; j++) {
1967 cpumask_set_cpu(cpu, mask);
1968 cpu = cpumask_next_wrap(cpu, cpu_online_mask,
1971 virtqueue_set_affinity(vi->rq[i].vq, mask);
1972 virtqueue_set_affinity(vi->sq[i].vq, mask);
1973 __netif_set_xps_queue(vi->dev, cpumask_bits(mask), i, false);
1974 cpumask_clear(mask);
1977 vi->affinity_hint_set = true;
1978 free_cpumask_var(mask);
1981 static int virtnet_cpu_online(unsigned int cpu, struct hlist_node *node)
1983 struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
1985 virtnet_set_affinity(vi);
1989 static int virtnet_cpu_dead(unsigned int cpu, struct hlist_node *node)
1991 struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
1993 virtnet_set_affinity(vi);
1997 static int virtnet_cpu_down_prep(unsigned int cpu, struct hlist_node *node)
1999 struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
2002 virtnet_clean_affinity(vi);
2006 static enum cpuhp_state virtionet_online;
2008 static int virtnet_cpu_notif_add(struct virtnet_info *vi)
2012 ret = cpuhp_state_add_instance_nocalls(virtionet_online, &vi->node);
2015 ret = cpuhp_state_add_instance_nocalls(CPUHP_VIRT_NET_DEAD,
2019 cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node);
2023 static void virtnet_cpu_notif_remove(struct virtnet_info *vi)
2025 cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node);
2026 cpuhp_state_remove_instance_nocalls(CPUHP_VIRT_NET_DEAD,
2030 static void virtnet_get_ringparam(struct net_device *dev,
2031 struct ethtool_ringparam *ring)
2033 struct virtnet_info *vi = netdev_priv(dev);
2035 ring->rx_max_pending = virtqueue_get_vring_size(vi->rq[0].vq);
2036 ring->tx_max_pending = virtqueue_get_vring_size(vi->sq[0].vq);
2037 ring->rx_pending = ring->rx_max_pending;
2038 ring->tx_pending = ring->tx_max_pending;
2042 static void virtnet_get_drvinfo(struct net_device *dev,
2043 struct ethtool_drvinfo *info)
2045 struct virtnet_info *vi = netdev_priv(dev);
2046 struct virtio_device *vdev = vi->vdev;
2048 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
2049 strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version));
2050 strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info));
2054 /* TODO: Eliminate OOO packets during switching */
2055 static int virtnet_set_channels(struct net_device *dev,
2056 struct ethtool_channels *channels)
2058 struct virtnet_info *vi = netdev_priv(dev);
2059 u16 queue_pairs = channels->combined_count;
2062 /* We don't support separate rx/tx channels.
2063 * We don't allow setting 'other' channels.
2065 if (channels->rx_count || channels->tx_count || channels->other_count)
2068 if (queue_pairs > vi->max_queue_pairs || queue_pairs == 0)
2071 /* For now we don't support modifying channels while XDP is loaded
2072 * also when XDP is loaded all RX queues have XDP programs so we only
2073 * need to check a single RX queue.
2075 if (vi->rq[0].xdp_prog)
2079 err = _virtnet_set_queues(vi, queue_pairs);
2081 netif_set_real_num_tx_queues(dev, queue_pairs);
2082 netif_set_real_num_rx_queues(dev, queue_pairs);
2084 virtnet_set_affinity(vi);
2091 static void virtnet_get_strings(struct net_device *dev, u32 stringset, u8 *data)
2093 struct virtnet_info *vi = netdev_priv(dev);
2094 char *p = (char *)data;
2097 switch (stringset) {
2099 for (i = 0; i < vi->curr_queue_pairs; i++) {
2100 for (j = 0; j < VIRTNET_RQ_STATS_LEN; j++) {
2101 snprintf(p, ETH_GSTRING_LEN, "rx_queue_%u_%s",
2102 i, virtnet_rq_stats_desc[j].desc);
2103 p += ETH_GSTRING_LEN;
2107 for (i = 0; i < vi->curr_queue_pairs; i++) {
2108 for (j = 0; j < VIRTNET_SQ_STATS_LEN; j++) {
2109 snprintf(p, ETH_GSTRING_LEN, "tx_queue_%u_%s",
2110 i, virtnet_sq_stats_desc[j].desc);
2111 p += ETH_GSTRING_LEN;
2118 static int virtnet_get_sset_count(struct net_device *dev, int sset)
2120 struct virtnet_info *vi = netdev_priv(dev);
2124 return vi->curr_queue_pairs * (VIRTNET_RQ_STATS_LEN +
2125 VIRTNET_SQ_STATS_LEN);
2131 static void virtnet_get_ethtool_stats(struct net_device *dev,
2132 struct ethtool_stats *stats, u64 *data)
2134 struct virtnet_info *vi = netdev_priv(dev);
2135 unsigned int idx = 0, start, i, j;
2136 const u8 *stats_base;
2139 for (i = 0; i < vi->curr_queue_pairs; i++) {
2140 struct receive_queue *rq = &vi->rq[i];
2142 stats_base = (u8 *)&rq->stats;
2144 start = u64_stats_fetch_begin_irq(&rq->stats.syncp);
2145 for (j = 0; j < VIRTNET_RQ_STATS_LEN; j++) {
2146 offset = virtnet_rq_stats_desc[j].offset;
2147 data[idx + j] = *(u64 *)(stats_base + offset);
2149 } while (u64_stats_fetch_retry_irq(&rq->stats.syncp, start));
2150 idx += VIRTNET_RQ_STATS_LEN;
2153 for (i = 0; i < vi->curr_queue_pairs; i++) {
2154 struct send_queue *sq = &vi->sq[i];
2156 stats_base = (u8 *)&sq->stats;
2158 start = u64_stats_fetch_begin_irq(&sq->stats.syncp);
2159 for (j = 0; j < VIRTNET_SQ_STATS_LEN; j++) {
2160 offset = virtnet_sq_stats_desc[j].offset;
2161 data[idx + j] = *(u64 *)(stats_base + offset);
2163 } while (u64_stats_fetch_retry_irq(&sq->stats.syncp, start));
2164 idx += VIRTNET_SQ_STATS_LEN;
2168 static void virtnet_get_channels(struct net_device *dev,
2169 struct ethtool_channels *channels)
2171 struct virtnet_info *vi = netdev_priv(dev);
2173 channels->combined_count = vi->curr_queue_pairs;
2174 channels->max_combined = vi->max_queue_pairs;
2175 channels->max_other = 0;
2176 channels->rx_count = 0;
2177 channels->tx_count = 0;
2178 channels->other_count = 0;
2181 /* Check if the user is trying to change anything besides speed/duplex */
2183 virtnet_validate_ethtool_cmd(const struct ethtool_link_ksettings *cmd)
2185 struct ethtool_link_ksettings diff1 = *cmd;
2186 struct ethtool_link_ksettings diff2 = {};
2188 /* cmd is always set so we need to clear it, validate the port type
2189 * and also without autonegotiation we can ignore advertising
2191 diff1.base.speed = 0;
2192 diff2.base.port = PORT_OTHER;
2193 ethtool_link_ksettings_zero_link_mode(&diff1, advertising);
2194 diff1.base.duplex = 0;
2196 diff1.base.link_mode_masks_nwords = 0;
2198 return !memcmp(&diff1.base, &diff2.base, sizeof(diff1.base)) &&
2199 bitmap_empty(diff1.link_modes.supported,
2200 __ETHTOOL_LINK_MODE_MASK_NBITS) &&
2201 bitmap_empty(diff1.link_modes.advertising,
2202 __ETHTOOL_LINK_MODE_MASK_NBITS) &&
2203 bitmap_empty(diff1.link_modes.lp_advertising,
2204 __ETHTOOL_LINK_MODE_MASK_NBITS);
2207 static int virtnet_set_link_ksettings(struct net_device *dev,
2208 const struct ethtool_link_ksettings *cmd)
2210 struct virtnet_info *vi = netdev_priv(dev);
2213 speed = cmd->base.speed;
2214 /* don't allow custom speed and duplex */
2215 if (!ethtool_validate_speed(speed) ||
2216 !ethtool_validate_duplex(cmd->base.duplex) ||
2217 !virtnet_validate_ethtool_cmd(cmd))
2220 vi->duplex = cmd->base.duplex;
2225 static int virtnet_get_link_ksettings(struct net_device *dev,
2226 struct ethtool_link_ksettings *cmd)
2228 struct virtnet_info *vi = netdev_priv(dev);
2230 cmd->base.speed = vi->speed;
2231 cmd->base.duplex = vi->duplex;
2232 cmd->base.port = PORT_OTHER;
2237 static int virtnet_set_coalesce(struct net_device *dev,
2238 struct ethtool_coalesce *ec)
2240 struct ethtool_coalesce ec_default = {
2241 .cmd = ETHTOOL_SCOALESCE,
2242 .rx_max_coalesced_frames = 1,
2244 struct virtnet_info *vi = netdev_priv(dev);
2247 if (ec->tx_max_coalesced_frames > 1)
2250 ec_default.tx_max_coalesced_frames = ec->tx_max_coalesced_frames;
2251 napi_weight = ec->tx_max_coalesced_frames ? NAPI_POLL_WEIGHT : 0;
2253 /* disallow changes to fields not explicitly tested above */
2254 if (memcmp(ec, &ec_default, sizeof(ec_default)))
2257 if (napi_weight ^ vi->sq[0].napi.weight) {
2258 if (dev->flags & IFF_UP)
2260 for (i = 0; i < vi->max_queue_pairs; i++)
2261 vi->sq[i].napi.weight = napi_weight;
2267 static int virtnet_get_coalesce(struct net_device *dev,
2268 struct ethtool_coalesce *ec)
2270 struct ethtool_coalesce ec_default = {
2271 .cmd = ETHTOOL_GCOALESCE,
2272 .rx_max_coalesced_frames = 1,
2274 struct virtnet_info *vi = netdev_priv(dev);
2276 memcpy(ec, &ec_default, sizeof(ec_default));
2278 if (vi->sq[0].napi.weight)
2279 ec->tx_max_coalesced_frames = 1;
2284 static void virtnet_init_settings(struct net_device *dev)
2286 struct virtnet_info *vi = netdev_priv(dev);
2288 vi->speed = SPEED_UNKNOWN;
2289 vi->duplex = DUPLEX_UNKNOWN;
2292 static void virtnet_update_settings(struct virtnet_info *vi)
2297 if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_SPEED_DUPLEX))
2300 speed = virtio_cread32(vi->vdev, offsetof(struct virtio_net_config,
2302 if (ethtool_validate_speed(speed))
2304 duplex = virtio_cread8(vi->vdev, offsetof(struct virtio_net_config,
2306 if (ethtool_validate_duplex(duplex))
2307 vi->duplex = duplex;
2310 static const struct ethtool_ops virtnet_ethtool_ops = {
2311 .get_drvinfo = virtnet_get_drvinfo,
2312 .get_link = ethtool_op_get_link,
2313 .get_ringparam = virtnet_get_ringparam,
2314 .get_strings = virtnet_get_strings,
2315 .get_sset_count = virtnet_get_sset_count,
2316 .get_ethtool_stats = virtnet_get_ethtool_stats,
2317 .set_channels = virtnet_set_channels,
2318 .get_channels = virtnet_get_channels,
2319 .get_ts_info = ethtool_op_get_ts_info,
2320 .get_link_ksettings = virtnet_get_link_ksettings,
2321 .set_link_ksettings = virtnet_set_link_ksettings,
2322 .set_coalesce = virtnet_set_coalesce,
2323 .get_coalesce = virtnet_get_coalesce,
2326 static void virtnet_freeze_down(struct virtio_device *vdev)
2328 struct virtnet_info *vi = vdev->priv;
2331 /* Make sure no work handler is accessing the device */
2332 flush_work(&vi->config_work);
2334 netif_tx_lock_bh(vi->dev);
2335 netif_device_detach(vi->dev);
2336 netif_tx_unlock_bh(vi->dev);
2337 cancel_delayed_work_sync(&vi->refill);
2339 if (netif_running(vi->dev)) {
2340 for (i = 0; i < vi->max_queue_pairs; i++) {
2341 napi_disable(&vi->rq[i].napi);
2342 virtnet_napi_tx_disable(&vi->sq[i].napi);
2347 static int init_vqs(struct virtnet_info *vi);
2349 static int virtnet_restore_up(struct virtio_device *vdev)
2351 struct virtnet_info *vi = vdev->priv;
2358 virtio_device_ready(vdev);
2360 if (netif_running(vi->dev)) {
2361 for (i = 0; i < vi->curr_queue_pairs; i++)
2362 if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL))
2363 schedule_delayed_work(&vi->refill, 0);
2365 for (i = 0; i < vi->max_queue_pairs; i++) {
2366 virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi);
2367 virtnet_napi_tx_enable(vi, vi->sq[i].vq,
2372 netif_tx_lock_bh(vi->dev);
2373 netif_device_attach(vi->dev);
2374 netif_tx_unlock_bh(vi->dev);
2378 static int virtnet_set_guest_offloads(struct virtnet_info *vi, u64 offloads)
2380 struct scatterlist sg;
2381 vi->ctrl->offloads = cpu_to_virtio64(vi->vdev, offloads);
2383 sg_init_one(&sg, &vi->ctrl->offloads, sizeof(vi->ctrl->offloads));
2385 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_GUEST_OFFLOADS,
2386 VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET, &sg)) {
2387 dev_warn(&vi->dev->dev, "Fail to set guest offload.\n");
2394 static int virtnet_clear_guest_offloads(struct virtnet_info *vi)
2398 if (!vi->guest_offloads)
2401 return virtnet_set_guest_offloads(vi, offloads);
2404 static int virtnet_restore_guest_offloads(struct virtnet_info *vi)
2406 u64 offloads = vi->guest_offloads;
2408 if (!vi->guest_offloads)
2411 return virtnet_set_guest_offloads(vi, offloads);
2414 static int virtnet_xdp_set(struct net_device *dev, struct bpf_prog *prog,
2415 struct netlink_ext_ack *extack)
2417 unsigned long int max_sz = PAGE_SIZE - sizeof(struct padded_vnet_hdr);
2418 struct virtnet_info *vi = netdev_priv(dev);
2419 struct bpf_prog *old_prog;
2420 u16 xdp_qp = 0, curr_qp;
2423 if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)
2424 && (virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO4) ||
2425 virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO6) ||
2426 virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_ECN) ||
2427 virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_UFO) ||
2428 virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_CSUM))) {
2429 NL_SET_ERR_MSG_MOD(extack, "Can't set XDP while host is implementing LRO/CSUM, disable LRO/CSUM first");
2433 if (vi->mergeable_rx_bufs && !vi->any_header_sg) {
2434 NL_SET_ERR_MSG_MOD(extack, "XDP expects header/data in single page, any_header_sg required");
2438 if (dev->mtu > max_sz) {
2439 NL_SET_ERR_MSG_MOD(extack, "MTU too large to enable XDP");
2440 netdev_warn(dev, "XDP requires MTU less than %lu\n", max_sz);
2444 curr_qp = vi->curr_queue_pairs - vi->xdp_queue_pairs;
2446 xdp_qp = nr_cpu_ids;
2448 /* XDP requires extra queues for XDP_TX */
2449 if (curr_qp + xdp_qp > vi->max_queue_pairs) {
2450 NL_SET_ERR_MSG_MOD(extack, "Too few free TX rings available");
2451 netdev_warn(dev, "request %i queues but max is %i\n",
2452 curr_qp + xdp_qp, vi->max_queue_pairs);
2456 old_prog = rtnl_dereference(vi->rq[0].xdp_prog);
2457 if (!prog && !old_prog)
2461 prog = bpf_prog_add(prog, vi->max_queue_pairs - 1);
2463 return PTR_ERR(prog);
2466 /* Make sure NAPI is not using any XDP TX queues for RX. */
2467 if (netif_running(dev)) {
2468 for (i = 0; i < vi->max_queue_pairs; i++) {
2469 napi_disable(&vi->rq[i].napi);
2470 virtnet_napi_tx_disable(&vi->sq[i].napi);
2475 for (i = 0; i < vi->max_queue_pairs; i++) {
2476 rcu_assign_pointer(vi->rq[i].xdp_prog, prog);
2478 virtnet_restore_guest_offloads(vi);
2483 err = _virtnet_set_queues(vi, curr_qp + xdp_qp);
2486 netif_set_real_num_rx_queues(dev, curr_qp + xdp_qp);
2487 vi->xdp_queue_pairs = xdp_qp;
2490 for (i = 0; i < vi->max_queue_pairs; i++) {
2491 rcu_assign_pointer(vi->rq[i].xdp_prog, prog);
2492 if (i == 0 && !old_prog)
2493 virtnet_clear_guest_offloads(vi);
2497 for (i = 0; i < vi->max_queue_pairs; i++) {
2499 bpf_prog_put(old_prog);
2500 if (netif_running(dev)) {
2501 virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi);
2502 virtnet_napi_tx_enable(vi, vi->sq[i].vq,
2511 virtnet_clear_guest_offloads(vi);
2512 for (i = 0; i < vi->max_queue_pairs; i++)
2513 rcu_assign_pointer(vi->rq[i].xdp_prog, old_prog);
2516 if (netif_running(dev)) {
2517 for (i = 0; i < vi->max_queue_pairs; i++) {
2518 virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi);
2519 virtnet_napi_tx_enable(vi, vi->sq[i].vq,
2524 bpf_prog_sub(prog, vi->max_queue_pairs - 1);
2528 static u32 virtnet_xdp_query(struct net_device *dev)
2530 struct virtnet_info *vi = netdev_priv(dev);
2531 const struct bpf_prog *xdp_prog;
2534 for (i = 0; i < vi->max_queue_pairs; i++) {
2535 xdp_prog = rtnl_dereference(vi->rq[i].xdp_prog);
2537 return xdp_prog->aux->id;
2542 static int virtnet_xdp(struct net_device *dev, struct netdev_bpf *xdp)
2544 switch (xdp->command) {
2545 case XDP_SETUP_PROG:
2546 return virtnet_xdp_set(dev, xdp->prog, xdp->extack);
2547 case XDP_QUERY_PROG:
2548 xdp->prog_id = virtnet_xdp_query(dev);
2555 static int virtnet_get_phys_port_name(struct net_device *dev, char *buf,
2558 struct virtnet_info *vi = netdev_priv(dev);
2561 if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_STANDBY))
2564 ret = snprintf(buf, len, "sby");
2571 static int virtnet_set_features(struct net_device *dev,
2572 netdev_features_t features)
2574 struct virtnet_info *vi = netdev_priv(dev);
2578 if ((dev->features ^ features) & NETIF_F_LRO) {
2579 if (vi->xdp_queue_pairs)
2582 if (features & NETIF_F_LRO)
2583 offloads = vi->guest_offloads_capable;
2587 err = virtnet_set_guest_offloads(vi, offloads);
2590 vi->guest_offloads = offloads;
2596 static const struct net_device_ops virtnet_netdev = {
2597 .ndo_open = virtnet_open,
2598 .ndo_stop = virtnet_close,
2599 .ndo_start_xmit = start_xmit,
2600 .ndo_validate_addr = eth_validate_addr,
2601 .ndo_set_mac_address = virtnet_set_mac_address,
2602 .ndo_set_rx_mode = virtnet_set_rx_mode,
2603 .ndo_get_stats64 = virtnet_stats,
2604 .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
2605 .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
2606 .ndo_bpf = virtnet_xdp,
2607 .ndo_xdp_xmit = virtnet_xdp_xmit,
2608 .ndo_features_check = passthru_features_check,
2609 .ndo_get_phys_port_name = virtnet_get_phys_port_name,
2610 .ndo_set_features = virtnet_set_features,
2613 static void virtnet_config_changed_work(struct work_struct *work)
2615 struct virtnet_info *vi =
2616 container_of(work, struct virtnet_info, config_work);
2619 if (virtio_cread_feature(vi->vdev, VIRTIO_NET_F_STATUS,
2620 struct virtio_net_config, status, &v) < 0)
2623 if (v & VIRTIO_NET_S_ANNOUNCE) {
2624 netdev_notify_peers(vi->dev);
2625 virtnet_ack_link_announce(vi);
2628 /* Ignore unknown (future) status bits */
2629 v &= VIRTIO_NET_S_LINK_UP;
2631 if (vi->status == v)
2636 if (vi->status & VIRTIO_NET_S_LINK_UP) {
2637 virtnet_update_settings(vi);
2638 netif_carrier_on(vi->dev);
2639 netif_tx_wake_all_queues(vi->dev);
2641 netif_carrier_off(vi->dev);
2642 netif_tx_stop_all_queues(vi->dev);
2646 static void virtnet_config_changed(struct virtio_device *vdev)
2648 struct virtnet_info *vi = vdev->priv;
2650 schedule_work(&vi->config_work);
2653 static void virtnet_free_queues(struct virtnet_info *vi)
2657 for (i = 0; i < vi->max_queue_pairs; i++) {
2658 napi_hash_del(&vi->rq[i].napi);
2659 netif_napi_del(&vi->rq[i].napi);
2660 netif_napi_del(&vi->sq[i].napi);
2663 /* We called napi_hash_del() before netif_napi_del(),
2664 * we need to respect an RCU grace period before freeing vi->rq
2673 static void _free_receive_bufs(struct virtnet_info *vi)
2675 struct bpf_prog *old_prog;
2678 for (i = 0; i < vi->max_queue_pairs; i++) {
2679 while (vi->rq[i].pages)
2680 __free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0);
2682 old_prog = rtnl_dereference(vi->rq[i].xdp_prog);
2683 RCU_INIT_POINTER(vi->rq[i].xdp_prog, NULL);
2685 bpf_prog_put(old_prog);
2689 static void free_receive_bufs(struct virtnet_info *vi)
2692 _free_receive_bufs(vi);
2696 static void free_receive_page_frags(struct virtnet_info *vi)
2699 for (i = 0; i < vi->max_queue_pairs; i++)
2700 if (vi->rq[i].alloc_frag.page)
2701 put_page(vi->rq[i].alloc_frag.page);
2704 static void free_unused_bufs(struct virtnet_info *vi)
2709 for (i = 0; i < vi->max_queue_pairs; i++) {
2710 struct virtqueue *vq = vi->sq[i].vq;
2711 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) {
2712 if (!is_xdp_frame(buf))
2715 xdp_return_frame(ptr_to_xdp(buf));
2719 for (i = 0; i < vi->max_queue_pairs; i++) {
2720 struct virtqueue *vq = vi->rq[i].vq;
2722 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) {
2723 if (vi->mergeable_rx_bufs) {
2724 put_page(virt_to_head_page(buf));
2725 } else if (vi->big_packets) {
2726 give_pages(&vi->rq[i], buf);
2728 put_page(virt_to_head_page(buf));
2734 static void virtnet_del_vqs(struct virtnet_info *vi)
2736 struct virtio_device *vdev = vi->vdev;
2738 virtnet_clean_affinity(vi);
2740 vdev->config->del_vqs(vdev);
2742 virtnet_free_queues(vi);
2745 /* How large should a single buffer be so a queue full of these can fit at
2746 * least one full packet?
2747 * Logic below assumes the mergeable buffer header is used.
2749 static unsigned int mergeable_min_buf_len(struct virtnet_info *vi, struct virtqueue *vq)
2751 const unsigned int hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
2752 unsigned int rq_size = virtqueue_get_vring_size(vq);
2753 unsigned int packet_len = vi->big_packets ? IP_MAX_MTU : vi->dev->max_mtu;
2754 unsigned int buf_len = hdr_len + ETH_HLEN + VLAN_HLEN + packet_len;
2755 unsigned int min_buf_len = DIV_ROUND_UP(buf_len, rq_size);
2757 return max(max(min_buf_len, hdr_len) - hdr_len,
2758 (unsigned int)GOOD_PACKET_LEN);
2761 static int virtnet_find_vqs(struct virtnet_info *vi)
2763 vq_callback_t **callbacks;
2764 struct virtqueue **vqs;
2770 /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
2771 * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
2772 * possible control vq.
2774 total_vqs = vi->max_queue_pairs * 2 +
2775 virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ);
2777 /* Allocate space for find_vqs parameters */
2778 vqs = kcalloc(total_vqs, sizeof(*vqs), GFP_KERNEL);
2781 callbacks = kmalloc_array(total_vqs, sizeof(*callbacks), GFP_KERNEL);
2784 names = kmalloc_array(total_vqs, sizeof(*names), GFP_KERNEL);
2787 if (!vi->big_packets || vi->mergeable_rx_bufs) {
2788 ctx = kcalloc(total_vqs, sizeof(*ctx), GFP_KERNEL);
2795 /* Parameters for control virtqueue, if any */
2797 callbacks[total_vqs - 1] = NULL;
2798 names[total_vqs - 1] = "control";
2801 /* Allocate/initialize parameters for send/receive virtqueues */
2802 for (i = 0; i < vi->max_queue_pairs; i++) {
2803 callbacks[rxq2vq(i)] = skb_recv_done;
2804 callbacks[txq2vq(i)] = skb_xmit_done;
2805 sprintf(vi->rq[i].name, "input.%d", i);
2806 sprintf(vi->sq[i].name, "output.%d", i);
2807 names[rxq2vq(i)] = vi->rq[i].name;
2808 names[txq2vq(i)] = vi->sq[i].name;
2810 ctx[rxq2vq(i)] = true;
2813 ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs, callbacks,
2819 vi->cvq = vqs[total_vqs - 1];
2820 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
2821 vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
2824 for (i = 0; i < vi->max_queue_pairs; i++) {
2825 vi->rq[i].vq = vqs[rxq2vq(i)];
2826 vi->rq[i].min_buf_len = mergeable_min_buf_len(vi, vi->rq[i].vq);
2827 vi->sq[i].vq = vqs[txq2vq(i)];
2830 /* run here: ret == 0. */
2845 static int virtnet_alloc_queues(struct virtnet_info *vi)
2849 vi->ctrl = kzalloc(sizeof(*vi->ctrl), GFP_KERNEL);
2852 vi->sq = kcalloc(vi->max_queue_pairs, sizeof(*vi->sq), GFP_KERNEL);
2855 vi->rq = kcalloc(vi->max_queue_pairs, sizeof(*vi->rq), GFP_KERNEL);
2859 INIT_DELAYED_WORK(&vi->refill, refill_work);
2860 for (i = 0; i < vi->max_queue_pairs; i++) {
2861 vi->rq[i].pages = NULL;
2862 netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll,
2864 netif_tx_napi_add(vi->dev, &vi->sq[i].napi, virtnet_poll_tx,
2865 napi_tx ? napi_weight : 0);
2867 sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg));
2868 ewma_pkt_len_init(&vi->rq[i].mrg_avg_pkt_len);
2869 sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg));
2871 u64_stats_init(&vi->rq[i].stats.syncp);
2872 u64_stats_init(&vi->sq[i].stats.syncp);
2885 static int init_vqs(struct virtnet_info *vi)
2889 /* Allocate send & receive queues */
2890 ret = virtnet_alloc_queues(vi);
2894 ret = virtnet_find_vqs(vi);
2899 virtnet_set_affinity(vi);
2905 virtnet_free_queues(vi);
2911 static ssize_t mergeable_rx_buffer_size_show(struct netdev_rx_queue *queue,
2914 struct virtnet_info *vi = netdev_priv(queue->dev);
2915 unsigned int queue_index = get_netdev_rx_queue_index(queue);
2916 unsigned int headroom = virtnet_get_headroom(vi);
2917 unsigned int tailroom = headroom ? sizeof(struct skb_shared_info) : 0;
2918 struct ewma_pkt_len *avg;
2920 BUG_ON(queue_index >= vi->max_queue_pairs);
2921 avg = &vi->rq[queue_index].mrg_avg_pkt_len;
2922 return sprintf(buf, "%u\n",
2923 get_mergeable_buf_len(&vi->rq[queue_index], avg,
2924 SKB_DATA_ALIGN(headroom + tailroom)));
2927 static struct rx_queue_attribute mergeable_rx_buffer_size_attribute =
2928 __ATTR_RO(mergeable_rx_buffer_size);
2930 static struct attribute *virtio_net_mrg_rx_attrs[] = {
2931 &mergeable_rx_buffer_size_attribute.attr,
2935 static const struct attribute_group virtio_net_mrg_rx_group = {
2936 .name = "virtio_net",
2937 .attrs = virtio_net_mrg_rx_attrs
2941 static bool virtnet_fail_on_feature(struct virtio_device *vdev,
2943 const char *fname, const char *dname)
2945 if (!virtio_has_feature(vdev, fbit))
2948 dev_err(&vdev->dev, "device advertises feature %s but not %s",
2954 #define VIRTNET_FAIL_ON(vdev, fbit, dbit) \
2955 virtnet_fail_on_feature(vdev, fbit, #fbit, dbit)
2957 static bool virtnet_validate_features(struct virtio_device *vdev)
2959 if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) &&
2960 (VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_RX,
2961 "VIRTIO_NET_F_CTRL_VQ") ||
2962 VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_VLAN,
2963 "VIRTIO_NET_F_CTRL_VQ") ||
2964 VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE,
2965 "VIRTIO_NET_F_CTRL_VQ") ||
2966 VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_MQ, "VIRTIO_NET_F_CTRL_VQ") ||
2967 VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR,
2968 "VIRTIO_NET_F_CTRL_VQ"))) {
2975 #define MIN_MTU ETH_MIN_MTU
2976 #define MAX_MTU ETH_MAX_MTU
2978 static int virtnet_validate(struct virtio_device *vdev)
2980 if (!vdev->config->get) {
2981 dev_err(&vdev->dev, "%s failure: config access disabled\n",
2986 if (!virtnet_validate_features(vdev))
2989 if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) {
2990 int mtu = virtio_cread16(vdev,
2991 offsetof(struct virtio_net_config,
2994 __virtio_clear_bit(vdev, VIRTIO_NET_F_MTU);
3000 static int virtnet_probe(struct virtio_device *vdev)
3002 int i, err = -ENOMEM;
3003 struct net_device *dev;
3004 struct virtnet_info *vi;
3005 u16 max_queue_pairs;
3008 /* Find if host supports multiqueue virtio_net device */
3009 err = virtio_cread_feature(vdev, VIRTIO_NET_F_MQ,
3010 struct virtio_net_config,
3011 max_virtqueue_pairs, &max_queue_pairs);
3013 /* We need at least 2 queue's */
3014 if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
3015 max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
3016 !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
3017 max_queue_pairs = 1;
3019 /* Allocate ourselves a network device with room for our info */
3020 dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs);
3024 /* Set up network device as normal. */
3025 dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE;
3026 dev->netdev_ops = &virtnet_netdev;
3027 dev->features = NETIF_F_HIGHDMA;
3029 dev->ethtool_ops = &virtnet_ethtool_ops;
3030 SET_NETDEV_DEV(dev, &vdev->dev);
3032 /* Do we support "hardware" checksums? */
3033 if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
3034 /* This opens up the world of extra features. */
3035 dev->hw_features |= NETIF_F_HW_CSUM | NETIF_F_SG;
3037 dev->features |= NETIF_F_HW_CSUM | NETIF_F_SG;
3039 if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
3040 dev->hw_features |= NETIF_F_TSO
3041 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
3043 /* Individual feature bits: what can host handle? */
3044 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
3045 dev->hw_features |= NETIF_F_TSO;
3046 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
3047 dev->hw_features |= NETIF_F_TSO6;
3048 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
3049 dev->hw_features |= NETIF_F_TSO_ECN;
3051 dev->features |= NETIF_F_GSO_ROBUST;
3054 dev->features |= dev->hw_features & NETIF_F_ALL_TSO;
3055 /* (!csum && gso) case will be fixed by register_netdev() */
3057 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM))
3058 dev->features |= NETIF_F_RXCSUM;
3059 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
3060 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6))
3061 dev->features |= NETIF_F_LRO;
3062 if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS))
3063 dev->hw_features |= NETIF_F_LRO;
3065 dev->vlan_features = dev->features;
3067 /* MTU range: 68 - 65535 */
3068 dev->min_mtu = MIN_MTU;
3069 dev->max_mtu = MAX_MTU;
3071 /* Configuration may specify what MAC to use. Otherwise random. */
3072 if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
3073 virtio_cread_bytes(vdev,
3074 offsetof(struct virtio_net_config, mac),
3075 dev->dev_addr, dev->addr_len);
3077 eth_hw_addr_random(dev);
3079 /* Set up our device-specific information */
3080 vi = netdev_priv(dev);
3085 INIT_WORK(&vi->config_work, virtnet_config_changed_work);
3087 /* If we can receive ANY GSO packets, we must allocate large ones. */
3088 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
3089 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
3090 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN) ||
3091 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_UFO))
3092 vi->big_packets = true;
3094 if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
3095 vi->mergeable_rx_bufs = true;
3097 if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF) ||
3098 virtio_has_feature(vdev, VIRTIO_F_VERSION_1))
3099 vi->hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
3101 vi->hdr_len = sizeof(struct virtio_net_hdr);
3103 if (virtio_has_feature(vdev, VIRTIO_F_ANY_LAYOUT) ||
3104 virtio_has_feature(vdev, VIRTIO_F_VERSION_1))
3105 vi->any_header_sg = true;
3107 if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
3110 if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) {
3111 mtu = virtio_cread16(vdev,
3112 offsetof(struct virtio_net_config,
3114 if (mtu < dev->min_mtu) {
3115 /* Should never trigger: MTU was previously validated
3116 * in virtnet_validate.
3119 "device MTU appears to have changed it is now %d < %d",
3127 /* TODO: size buffers correctly in this case. */
3128 if (dev->mtu > ETH_DATA_LEN)
3129 vi->big_packets = true;
3132 if (vi->any_header_sg)
3133 dev->needed_headroom = vi->hdr_len;
3135 /* Enable multiqueue by default */
3136 if (num_online_cpus() >= max_queue_pairs)
3137 vi->curr_queue_pairs = max_queue_pairs;
3139 vi->curr_queue_pairs = num_online_cpus();
3140 vi->max_queue_pairs = max_queue_pairs;
3142 /* Allocate/initialize the rx/tx queues, and invoke find_vqs */
3148 if (vi->mergeable_rx_bufs)
3149 dev->sysfs_rx_queue_group = &virtio_net_mrg_rx_group;
3151 netif_set_real_num_tx_queues(dev, vi->curr_queue_pairs);
3152 netif_set_real_num_rx_queues(dev, vi->curr_queue_pairs);
3154 virtnet_init_settings(dev);
3156 if (virtio_has_feature(vdev, VIRTIO_NET_F_STANDBY)) {
3157 vi->failover = net_failover_create(vi->dev);
3158 if (IS_ERR(vi->failover)) {
3159 err = PTR_ERR(vi->failover);
3164 err = register_netdev(dev);
3166 pr_debug("virtio_net: registering device failed\n");
3170 virtio_device_ready(vdev);
3172 err = virtnet_cpu_notif_add(vi);
3174 pr_debug("virtio_net: registering cpu notifier failed\n");
3175 goto free_unregister_netdev;
3178 virtnet_set_queues(vi, vi->curr_queue_pairs);
3180 /* Assume link up if device can't report link status,
3181 otherwise get link status from config. */
3182 netif_carrier_off(dev);
3183 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
3184 schedule_work(&vi->config_work);
3186 vi->status = VIRTIO_NET_S_LINK_UP;
3187 virtnet_update_settings(vi);
3188 netif_carrier_on(dev);
3191 for (i = 0; i < ARRAY_SIZE(guest_offloads); i++)
3192 if (virtio_has_feature(vi->vdev, guest_offloads[i]))
3193 set_bit(guest_offloads[i], &vi->guest_offloads);
3194 vi->guest_offloads_capable = vi->guest_offloads;
3196 pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
3197 dev->name, max_queue_pairs);
3201 free_unregister_netdev:
3202 vi->vdev->config->reset(vdev);
3204 unregister_netdev(dev);
3206 net_failover_destroy(vi->failover);
3208 cancel_delayed_work_sync(&vi->refill);
3209 free_receive_page_frags(vi);
3210 virtnet_del_vqs(vi);
3216 static void remove_vq_common(struct virtnet_info *vi)
3218 vi->vdev->config->reset(vi->vdev);
3220 /* Free unused buffers in both send and recv, if any. */
3221 free_unused_bufs(vi);
3223 free_receive_bufs(vi);
3225 free_receive_page_frags(vi);
3227 virtnet_del_vqs(vi);
3230 static void virtnet_remove(struct virtio_device *vdev)
3232 struct virtnet_info *vi = vdev->priv;
3234 virtnet_cpu_notif_remove(vi);
3236 /* Make sure no work handler is accessing the device. */
3237 flush_work(&vi->config_work);
3239 unregister_netdev(vi->dev);
3241 net_failover_destroy(vi->failover);
3243 remove_vq_common(vi);
3245 free_netdev(vi->dev);
3248 static __maybe_unused int virtnet_freeze(struct virtio_device *vdev)
3250 struct virtnet_info *vi = vdev->priv;
3252 virtnet_cpu_notif_remove(vi);
3253 virtnet_freeze_down(vdev);
3254 remove_vq_common(vi);
3259 static __maybe_unused int virtnet_restore(struct virtio_device *vdev)
3261 struct virtnet_info *vi = vdev->priv;
3264 err = virtnet_restore_up(vdev);
3267 virtnet_set_queues(vi, vi->curr_queue_pairs);
3269 err = virtnet_cpu_notif_add(vi);
3276 static struct virtio_device_id id_table[] = {
3277 { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
3281 #define VIRTNET_FEATURES \
3282 VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM, \
3284 VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6, \
3285 VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, \
3286 VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO, \
3287 VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ, \
3288 VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, \
3289 VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ, \
3290 VIRTIO_NET_F_CTRL_MAC_ADDR, \
3291 VIRTIO_NET_F_MTU, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, \
3292 VIRTIO_NET_F_SPEED_DUPLEX, VIRTIO_NET_F_STANDBY
3294 static unsigned int features[] = {
3298 static unsigned int features_legacy[] = {
3301 VIRTIO_F_ANY_LAYOUT,
3304 static struct virtio_driver virtio_net_driver = {
3305 .feature_table = features,
3306 .feature_table_size = ARRAY_SIZE(features),
3307 .feature_table_legacy = features_legacy,
3308 .feature_table_size_legacy = ARRAY_SIZE(features_legacy),
3309 .driver.name = KBUILD_MODNAME,
3310 .driver.owner = THIS_MODULE,
3311 .id_table = id_table,
3312 .validate = virtnet_validate,
3313 .probe = virtnet_probe,
3314 .remove = virtnet_remove,
3315 .config_changed = virtnet_config_changed,
3316 #ifdef CONFIG_PM_SLEEP
3317 .freeze = virtnet_freeze,
3318 .restore = virtnet_restore,
3322 static __init int virtio_net_driver_init(void)
3326 ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "virtio/net:online",
3328 virtnet_cpu_down_prep);
3331 virtionet_online = ret;
3332 ret = cpuhp_setup_state_multi(CPUHP_VIRT_NET_DEAD, "virtio/net:dead",
3333 NULL, virtnet_cpu_dead);
3337 ret = register_virtio_driver(&virtio_net_driver);
3342 cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD);
3344 cpuhp_remove_multi_state(virtionet_online);
3348 module_init(virtio_net_driver_init);
3350 static __exit void virtio_net_driver_exit(void)
3352 unregister_virtio_driver(&virtio_net_driver);
3353 cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD);
3354 cpuhp_remove_multi_state(virtionet_online);
3356 module_exit(virtio_net_driver_exit);
3358 MODULE_DEVICE_TABLE(virtio, id_table);
3359 MODULE_DESCRIPTION("Virtio network driver");
3360 MODULE_LICENSE("GPL");