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/scatterlist.h>
27 #include <linux/if_vlan.h>
28 #include <linux/slab.h>
29 #include <linux/cpu.h>
30 #include <linux/average.h>
31 #include <net/busy_poll.h>
33 static int napi_weight
= NAPI_POLL_WEIGHT
;
34 module_param(napi_weight
, int, 0444);
36 static bool csum
= true, gso
= true;
37 module_param(csum
, bool, 0444);
38 module_param(gso
, bool, 0444);
40 /* FIXME: MTU in config. */
41 #define GOOD_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
42 #define GOOD_COPY_LEN 128
44 /* RX packet size EWMA. The average packet size is used to determine the packet
45 * buffer size when refilling RX rings. As the entire RX ring may be refilled
46 * at once, the weight is chosen so that the EWMA will be insensitive to short-
47 * term, transient changes in packet size.
49 DECLARE_EWMA(pkt_len
, 1, 64)
51 /* Minimum alignment for mergeable packet buffers. */
52 #define MERGEABLE_BUFFER_ALIGN max(L1_CACHE_BYTES, 256)
54 #define VIRTNET_DRIVER_VERSION "1.0.0"
56 struct virtnet_stats
{
57 struct u64_stats_sync tx_syncp
;
58 struct u64_stats_sync rx_syncp
;
66 /* Internal representation of a send virtqueue */
68 /* Virtqueue associated with this send _queue */
71 /* TX: fragments + linear part + virtio header */
72 struct scatterlist sg
[MAX_SKB_FRAGS
+ 2];
74 /* Name of the send queue: output.$index */
78 /* Internal representation of a receive virtqueue */
79 struct receive_queue
{
80 /* Virtqueue associated with this receive_queue */
83 struct napi_struct napi
;
85 struct bpf_prog __rcu
*xdp_prog
;
87 /* Chain pages by the private ptr. */
90 /* Average packet length for mergeable receive buffers. */
91 struct ewma_pkt_len mrg_avg_pkt_len
;
93 /* Page frag for packet buffer allocation. */
94 struct page_frag alloc_frag
;
96 /* RX: fragments + linear part + virtio header */
97 struct scatterlist sg
[MAX_SKB_FRAGS
+ 2];
99 /* Name of this receive queue: input.$index */
103 struct virtnet_info
{
104 struct virtio_device
*vdev
;
105 struct virtqueue
*cvq
;
106 struct net_device
*dev
;
107 struct send_queue
*sq
;
108 struct receive_queue
*rq
;
111 /* Max # of queue pairs supported by the device */
114 /* # of queue pairs currently used by the driver */
115 u16 curr_queue_pairs
;
117 /* # of XDP queue pairs currently used by the driver */
120 /* I like... big packets and I cannot lie! */
123 /* Host will merge rx buffers for big packets (shake it! shake it!) */
124 bool mergeable_rx_bufs
;
126 /* Has control virtqueue */
129 /* Host can handle any s/g split between our header and packet data */
132 /* Packet virtio header size */
135 /* Active statistics */
136 struct virtnet_stats __percpu
*stats
;
138 /* Work struct for refilling if we run low on memory. */
139 struct delayed_work refill
;
141 /* Work struct for config space updates */
142 struct work_struct config_work
;
144 /* Does the affinity hint is set for virtqueues? */
145 bool affinity_hint_set
;
147 /* CPU hotplug instances for online & dead */
148 struct hlist_node node
;
149 struct hlist_node node_dead
;
151 /* Control VQ buffers: protected by the rtnl lock */
152 struct virtio_net_ctrl_hdr ctrl_hdr
;
153 virtio_net_ctrl_ack ctrl_status
;
154 struct virtio_net_ctrl_mq ctrl_mq
;
159 /* Ethtool settings */
164 struct padded_vnet_hdr
{
165 struct virtio_net_hdr_mrg_rxbuf hdr
;
167 * hdr is in a separate sg buffer, and data sg buffer shares same page
168 * with this header sg. This padding makes next sg 16 byte aligned
174 /* Converting between virtqueue no. and kernel tx/rx queue no.
175 * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq
177 static int vq2txq(struct virtqueue
*vq
)
179 return (vq
->index
- 1) / 2;
182 static int txq2vq(int txq
)
187 static int vq2rxq(struct virtqueue
*vq
)
189 return vq
->index
/ 2;
192 static int rxq2vq(int rxq
)
197 static inline struct virtio_net_hdr_mrg_rxbuf
*skb_vnet_hdr(struct sk_buff
*skb
)
199 return (struct virtio_net_hdr_mrg_rxbuf
*)skb
->cb
;
203 * private is used to chain pages for big packets, put the whole
204 * most recent used list in the beginning for reuse
206 static void give_pages(struct receive_queue
*rq
, struct page
*page
)
210 /* Find end of list, sew whole thing into vi->rq.pages. */
211 for (end
= page
; end
->private; end
= (struct page
*)end
->private);
212 end
->private = (unsigned long)rq
->pages
;
216 static struct page
*get_a_page(struct receive_queue
*rq
, gfp_t gfp_mask
)
218 struct page
*p
= rq
->pages
;
221 rq
->pages
= (struct page
*)p
->private;
222 /* clear private here, it is used to chain pages */
225 p
= alloc_page(gfp_mask
);
229 static void skb_xmit_done(struct virtqueue
*vq
)
231 struct virtnet_info
*vi
= vq
->vdev
->priv
;
233 /* Suppress further interrupts. */
234 virtqueue_disable_cb(vq
);
236 /* We were probably waiting for more output buffers. */
237 netif_wake_subqueue(vi
->dev
, vq2txq(vq
));
240 static unsigned int mergeable_ctx_to_buf_truesize(unsigned long mrg_ctx
)
242 unsigned int truesize
= mrg_ctx
& (MERGEABLE_BUFFER_ALIGN
- 1);
243 return (truesize
+ 1) * MERGEABLE_BUFFER_ALIGN
;
246 static void *mergeable_ctx_to_buf_address(unsigned long mrg_ctx
)
248 return (void *)(mrg_ctx
& -MERGEABLE_BUFFER_ALIGN
);
252 static unsigned long mergeable_buf_to_ctx(void *buf
, unsigned int truesize
)
254 unsigned int size
= truesize
/ MERGEABLE_BUFFER_ALIGN
;
255 return (unsigned long)buf
| (size
- 1);
258 /* Called from bottom half context */
259 static struct sk_buff
*page_to_skb(struct virtnet_info
*vi
,
260 struct receive_queue
*rq
,
261 struct page
*page
, unsigned int offset
,
262 unsigned int len
, unsigned int truesize
)
265 struct virtio_net_hdr_mrg_rxbuf
*hdr
;
266 unsigned int copy
, hdr_len
, hdr_padded_len
;
269 p
= page_address(page
) + offset
;
271 /* copy small packet so we can reuse these pages for small data */
272 skb
= napi_alloc_skb(&rq
->napi
, GOOD_COPY_LEN
);
276 hdr
= skb_vnet_hdr(skb
);
278 hdr_len
= vi
->hdr_len
;
279 if (vi
->mergeable_rx_bufs
)
280 hdr_padded_len
= sizeof *hdr
;
282 hdr_padded_len
= sizeof(struct padded_vnet_hdr
);
284 memcpy(hdr
, p
, hdr_len
);
287 offset
+= hdr_padded_len
;
291 if (copy
> skb_tailroom(skb
))
292 copy
= skb_tailroom(skb
);
293 memcpy(skb_put(skb
, copy
), p
, copy
);
298 if (vi
->mergeable_rx_bufs
) {
300 skb_add_rx_frag(skb
, 0, page
, offset
, len
, truesize
);
307 * Verify that we can indeed put this data into a skb.
308 * This is here to handle cases when the device erroneously
309 * tries to receive more than is possible. This is usually
310 * the case of a broken device.
312 if (unlikely(len
> MAX_SKB_FRAGS
* PAGE_SIZE
)) {
313 net_dbg_ratelimited("%s: too much data\n", skb
->dev
->name
);
317 BUG_ON(offset
>= PAGE_SIZE
);
319 unsigned int frag_size
= min((unsigned)PAGE_SIZE
- offset
, len
);
320 skb_add_rx_frag(skb
, skb_shinfo(skb
)->nr_frags
, page
, offset
,
321 frag_size
, truesize
);
323 page
= (struct page
*)page
->private;
328 give_pages(rq
, page
);
333 static void virtnet_xdp_xmit(struct virtnet_info
*vi
,
334 struct receive_queue
*rq
,
335 struct send_queue
*sq
,
336 struct xdp_buff
*xdp
)
338 struct page
*page
= virt_to_head_page(xdp
->data
);
339 struct virtio_net_hdr_mrg_rxbuf
*hdr
;
340 unsigned int num_sg
, len
;
344 /* Free up any pending old buffers before queueing new ones. */
345 while ((xdp_sent
= virtqueue_get_buf(sq
->vq
, &len
)) != NULL
) {
346 struct page
*sent_page
= virt_to_head_page(xdp_sent
);
348 if (vi
->mergeable_rx_bufs
)
351 give_pages(rq
, sent_page
);
354 /* Zero header and leave csum up to XDP layers */
356 memset(hdr
, 0, vi
->hdr_len
);
359 sg_init_one(sq
->sg
, xdp
->data
, xdp
->data_end
- xdp
->data
);
360 err
= virtqueue_add_outbuf(sq
->vq
, sq
->sg
, num_sg
,
361 xdp
->data
, GFP_ATOMIC
);
363 if (vi
->mergeable_rx_bufs
)
366 give_pages(rq
, page
);
367 return; // On error abort to avoid unnecessary kick
368 } else if (!vi
->mergeable_rx_bufs
) {
369 /* If not mergeable bufs must be big packets so cleanup pages */
370 give_pages(rq
, (struct page
*)page
->private);
374 virtqueue_kick(sq
->vq
);
377 static u32
do_xdp_prog(struct virtnet_info
*vi
,
378 struct receive_queue
*rq
,
379 struct bpf_prog
*xdp_prog
,
380 struct page
*page
, int offset
, int len
)
388 buf
= page_address(page
) + offset
;
390 if (vi
->mergeable_rx_bufs
)
391 hdr_padded_len
= sizeof(struct virtio_net_hdr_mrg_rxbuf
);
393 hdr_padded_len
= sizeof(struct padded_vnet_hdr
);
395 xdp
.data
= buf
+ hdr_padded_len
;
396 xdp
.data_end
= xdp
.data
+ (len
- vi
->hdr_len
);
398 act
= bpf_prog_run_xdp(xdp_prog
, &xdp
);
403 qp
= vi
->curr_queue_pairs
-
404 vi
->xdp_queue_pairs
+
406 xdp
.data
= buf
+ (vi
->mergeable_rx_bufs
? 0 : 4);
407 virtnet_xdp_xmit(vi
, rq
, &vi
->sq
[qp
], &xdp
);
410 bpf_warn_invalid_xdp_action(act
);
417 static struct sk_buff
*receive_small(struct virtnet_info
*vi
, void *buf
, unsigned int len
)
419 struct sk_buff
* skb
= buf
;
427 static struct sk_buff
*receive_big(struct net_device
*dev
,
428 struct virtnet_info
*vi
,
429 struct receive_queue
*rq
,
433 struct bpf_prog
*xdp_prog
;
434 struct page
*page
= buf
;
438 xdp_prog
= rcu_dereference(rq
->xdp_prog
);
440 struct virtio_net_hdr_mrg_rxbuf
*hdr
= buf
;
443 if (unlikely(hdr
->hdr
.gso_type
|| hdr
->hdr
.flags
))
445 act
= do_xdp_prog(vi
, rq
, xdp_prog
, page
, 0, len
);
459 skb
= page_to_skb(vi
, rq
, page
, 0, len
, PAGE_SIZE
);
468 dev
->stats
.rx_dropped
++;
469 give_pages(rq
, page
);
474 /* The conditions to enable XDP should preclude the underlying device from
475 * sending packets across multiple buffers (num_buf > 1). However per spec
476 * it does not appear to be illegal to do so but rather just against convention.
477 * So in order to avoid making a system unresponsive the packets are pushed
478 * into a page and the XDP program is run. This will be extremely slow and we
479 * push a warning to the user to fix this as soon as possible. Fixing this may
480 * require resolving the underlying hardware to determine why multiple buffers
481 * are being received or simply loading the XDP program in the ingress stack
482 * after the skb is built because there is no advantage to running it here
485 static struct page
*xdp_linearize_page(struct receive_queue
*rq
,
491 struct page
*page
= alloc_page(GFP_ATOMIC
);
492 unsigned int page_off
= 0;
497 memcpy(page_address(page
) + page_off
, page_address(p
) + offset
, *len
);
506 ctx
= (unsigned long)virtqueue_get_buf(rq
->vq
, &buflen
);
510 buf
= mergeable_ctx_to_buf_address(ctx
);
511 p
= virt_to_head_page(buf
);
512 off
= buf
- page_address(p
);
514 /* guard against a misconfigured or uncooperative backend that
515 * is sending packet larger than the MTU.
517 if ((page_off
+ buflen
) > PAGE_SIZE
) {
522 memcpy(page_address(page
) + page_off
,
523 page_address(p
) + off
, buflen
);
531 __free_pages(page
, 0);
535 static struct sk_buff
*receive_mergeable(struct net_device
*dev
,
536 struct virtnet_info
*vi
,
537 struct receive_queue
*rq
,
541 void *buf
= mergeable_ctx_to_buf_address(ctx
);
542 struct virtio_net_hdr_mrg_rxbuf
*hdr
= buf
;
543 u16 num_buf
= virtio16_to_cpu(vi
->vdev
, hdr
->num_buffers
);
544 struct page
*page
= virt_to_head_page(buf
);
545 int offset
= buf
- page_address(page
);
546 struct sk_buff
*head_skb
, *curr_skb
;
547 struct bpf_prog
*xdp_prog
;
548 unsigned int truesize
;
553 xdp_prog
= rcu_dereference(rq
->xdp_prog
);
555 struct page
*xdp_page
;
558 /* This happens when rx buffer size is underestimated */
559 if (unlikely(num_buf
> 1)) {
560 /* linearize data for XDP */
561 xdp_page
= xdp_linearize_page(rq
, &num_buf
,
570 /* Transient failure which in theory could occur if
571 * in-flight packets from before XDP was enabled reach
572 * the receive path after XDP is loaded. In practice I
573 * was not able to create this condition.
575 if (unlikely(hdr
->hdr
.gso_type
|| hdr
->hdr
.flags
))
578 act
= do_xdp_prog(vi
, rq
, xdp_prog
, xdp_page
, offset
, len
);
581 /* We can only create skb based on xdp_page. */
582 if (unlikely(xdp_page
!= page
)) {
585 head_skb
= page_to_skb(vi
, rq
, xdp_page
,
591 if (unlikely(xdp_page
!= page
))
597 if (unlikely(xdp_page
!= page
))
598 __free_pages(xdp_page
, 0);
604 truesize
= max(len
, mergeable_ctx_to_buf_truesize(ctx
));
605 head_skb
= page_to_skb(vi
, rq
, page
, offset
, len
, truesize
);
608 if (unlikely(!curr_skb
))
613 ctx
= (unsigned long)virtqueue_get_buf(rq
->vq
, &len
);
614 if (unlikely(!ctx
)) {
615 pr_debug("%s: rx error: %d buffers out of %d missing\n",
617 virtio16_to_cpu(vi
->vdev
,
619 dev
->stats
.rx_length_errors
++;
623 buf
= mergeable_ctx_to_buf_address(ctx
);
624 page
= virt_to_head_page(buf
);
626 num_skb_frags
= skb_shinfo(curr_skb
)->nr_frags
;
627 if (unlikely(num_skb_frags
== MAX_SKB_FRAGS
)) {
628 struct sk_buff
*nskb
= alloc_skb(0, GFP_ATOMIC
);
632 if (curr_skb
== head_skb
)
633 skb_shinfo(curr_skb
)->frag_list
= nskb
;
635 curr_skb
->next
= nskb
;
637 head_skb
->truesize
+= nskb
->truesize
;
640 truesize
= max(len
, mergeable_ctx_to_buf_truesize(ctx
));
641 if (curr_skb
!= head_skb
) {
642 head_skb
->data_len
+= len
;
643 head_skb
->len
+= len
;
644 head_skb
->truesize
+= truesize
;
646 offset
= buf
- page_address(page
);
647 if (skb_can_coalesce(curr_skb
, num_skb_frags
, page
, offset
)) {
649 skb_coalesce_rx_frag(curr_skb
, num_skb_frags
- 1,
652 skb_add_rx_frag(curr_skb
, num_skb_frags
, page
,
653 offset
, len
, truesize
);
657 ewma_pkt_len_add(&rq
->mrg_avg_pkt_len
, head_skb
->len
);
665 ctx
= (unsigned long)virtqueue_get_buf(rq
->vq
, &len
);
666 if (unlikely(!ctx
)) {
667 pr_debug("%s: rx error: %d buffers missing\n",
669 dev
->stats
.rx_length_errors
++;
672 page
= virt_to_head_page(mergeable_ctx_to_buf_address(ctx
));
676 dev
->stats
.rx_dropped
++;
677 dev_kfree_skb(head_skb
);
682 static void receive_buf(struct virtnet_info
*vi
, struct receive_queue
*rq
,
683 void *buf
, unsigned int len
)
685 struct net_device
*dev
= vi
->dev
;
686 struct virtnet_stats
*stats
= this_cpu_ptr(vi
->stats
);
688 struct virtio_net_hdr_mrg_rxbuf
*hdr
;
690 if (unlikely(len
< vi
->hdr_len
+ ETH_HLEN
)) {
691 pr_debug("%s: short packet %i\n", dev
->name
, len
);
692 dev
->stats
.rx_length_errors
++;
693 if (vi
->mergeable_rx_bufs
) {
694 unsigned long ctx
= (unsigned long)buf
;
695 void *base
= mergeable_ctx_to_buf_address(ctx
);
696 put_page(virt_to_head_page(base
));
697 } else if (vi
->big_packets
) {
705 if (vi
->mergeable_rx_bufs
)
706 skb
= receive_mergeable(dev
, vi
, rq
, (unsigned long)buf
, len
);
707 else if (vi
->big_packets
)
708 skb
= receive_big(dev
, vi
, rq
, buf
, len
);
710 skb
= receive_small(vi
, buf
, len
);
715 hdr
= skb_vnet_hdr(skb
);
717 u64_stats_update_begin(&stats
->rx_syncp
);
718 stats
->rx_bytes
+= skb
->len
;
720 u64_stats_update_end(&stats
->rx_syncp
);
722 if (hdr
->hdr
.flags
& VIRTIO_NET_HDR_F_DATA_VALID
)
723 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
725 if (virtio_net_hdr_to_skb(skb
, &hdr
->hdr
,
726 virtio_is_little_endian(vi
->vdev
))) {
727 net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n",
728 dev
->name
, hdr
->hdr
.gso_type
,
733 skb
->protocol
= eth_type_trans(skb
, dev
);
734 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
735 ntohs(skb
->protocol
), skb
->len
, skb
->pkt_type
);
737 napi_gro_receive(&rq
->napi
, skb
);
741 dev
->stats
.rx_frame_errors
++;
745 static int add_recvbuf_small(struct virtnet_info
*vi
, struct receive_queue
*rq
,
749 struct virtio_net_hdr_mrg_rxbuf
*hdr
;
752 skb
= __netdev_alloc_skb_ip_align(vi
->dev
, GOOD_PACKET_LEN
, gfp
);
756 skb_put(skb
, GOOD_PACKET_LEN
);
758 hdr
= skb_vnet_hdr(skb
);
759 sg_init_table(rq
->sg
, 2);
760 sg_set_buf(rq
->sg
, hdr
, vi
->hdr_len
);
761 skb_to_sgvec(skb
, rq
->sg
+ 1, 0, skb
->len
);
763 err
= virtqueue_add_inbuf(rq
->vq
, rq
->sg
, 2, skb
, gfp
);
770 static int add_recvbuf_big(struct virtnet_info
*vi
, struct receive_queue
*rq
,
773 struct page
*first
, *list
= NULL
;
777 sg_init_table(rq
->sg
, MAX_SKB_FRAGS
+ 2);
779 /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */
780 for (i
= MAX_SKB_FRAGS
+ 1; i
> 1; --i
) {
781 first
= get_a_page(rq
, gfp
);
784 give_pages(rq
, list
);
787 sg_set_buf(&rq
->sg
[i
], page_address(first
), PAGE_SIZE
);
789 /* chain new page in list head to match sg */
790 first
->private = (unsigned long)list
;
794 first
= get_a_page(rq
, gfp
);
796 give_pages(rq
, list
);
799 p
= page_address(first
);
801 /* rq->sg[0], rq->sg[1] share the same page */
802 /* a separated rq->sg[0] for header - required in case !any_header_sg */
803 sg_set_buf(&rq
->sg
[0], p
, vi
->hdr_len
);
805 /* rq->sg[1] for data packet, from offset */
806 offset
= sizeof(struct padded_vnet_hdr
);
807 sg_set_buf(&rq
->sg
[1], p
+ offset
, PAGE_SIZE
- offset
);
809 /* chain first in list head */
810 first
->private = (unsigned long)list
;
811 err
= virtqueue_add_inbuf(rq
->vq
, rq
->sg
, MAX_SKB_FRAGS
+ 2,
814 give_pages(rq
, first
);
819 static unsigned int get_mergeable_buf_len(struct ewma_pkt_len
*avg_pkt_len
)
821 const size_t hdr_len
= sizeof(struct virtio_net_hdr_mrg_rxbuf
);
824 len
= hdr_len
+ clamp_t(unsigned int, ewma_pkt_len_read(avg_pkt_len
),
825 GOOD_PACKET_LEN
, PAGE_SIZE
- hdr_len
);
826 return ALIGN(len
, MERGEABLE_BUFFER_ALIGN
);
829 static int add_recvbuf_mergeable(struct receive_queue
*rq
, gfp_t gfp
)
831 struct page_frag
*alloc_frag
= &rq
->alloc_frag
;
835 unsigned int len
, hole
;
837 len
= get_mergeable_buf_len(&rq
->mrg_avg_pkt_len
);
838 if (unlikely(!skb_page_frag_refill(len
, alloc_frag
, gfp
)))
841 buf
= (char *)page_address(alloc_frag
->page
) + alloc_frag
->offset
;
842 ctx
= mergeable_buf_to_ctx(buf
, len
);
843 get_page(alloc_frag
->page
);
844 alloc_frag
->offset
+= len
;
845 hole
= alloc_frag
->size
- alloc_frag
->offset
;
847 /* To avoid internal fragmentation, if there is very likely not
848 * enough space for another buffer, add the remaining space to
849 * the current buffer. This extra space is not included in
850 * the truesize stored in ctx.
853 alloc_frag
->offset
+= hole
;
856 sg_init_one(rq
->sg
, buf
, len
);
857 err
= virtqueue_add_inbuf(rq
->vq
, rq
->sg
, 1, (void *)ctx
, gfp
);
859 put_page(virt_to_head_page(buf
));
865 * Returns false if we couldn't fill entirely (OOM).
867 * Normally run in the receive path, but can also be run from ndo_open
868 * before we're receiving packets, or from refill_work which is
869 * careful to disable receiving (using napi_disable).
871 static bool try_fill_recv(struct virtnet_info
*vi
, struct receive_queue
*rq
,
879 if (vi
->mergeable_rx_bufs
)
880 err
= add_recvbuf_mergeable(rq
, gfp
);
881 else if (vi
->big_packets
)
882 err
= add_recvbuf_big(vi
, rq
, gfp
);
884 err
= add_recvbuf_small(vi
, rq
, gfp
);
886 oom
= err
== -ENOMEM
;
889 } while (rq
->vq
->num_free
);
890 virtqueue_kick(rq
->vq
);
894 static void skb_recv_done(struct virtqueue
*rvq
)
896 struct virtnet_info
*vi
= rvq
->vdev
->priv
;
897 struct receive_queue
*rq
= &vi
->rq
[vq2rxq(rvq
)];
899 /* Schedule NAPI, Suppress further interrupts if successful. */
900 if (napi_schedule_prep(&rq
->napi
)) {
901 virtqueue_disable_cb(rvq
);
902 __napi_schedule(&rq
->napi
);
906 static void virtnet_napi_enable(struct receive_queue
*rq
)
908 napi_enable(&rq
->napi
);
910 /* If all buffers were filled by other side before we napi_enabled, we
911 * won't get another interrupt, so process any outstanding packets
912 * now. virtnet_poll wants re-enable the queue, so we disable here.
913 * We synchronize against interrupts via NAPI_STATE_SCHED */
914 if (napi_schedule_prep(&rq
->napi
)) {
915 virtqueue_disable_cb(rq
->vq
);
917 __napi_schedule(&rq
->napi
);
922 static void refill_work(struct work_struct
*work
)
924 struct virtnet_info
*vi
=
925 container_of(work
, struct virtnet_info
, refill
.work
);
929 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++) {
930 struct receive_queue
*rq
= &vi
->rq
[i
];
932 napi_disable(&rq
->napi
);
933 still_empty
= !try_fill_recv(vi
, rq
, GFP_KERNEL
);
934 virtnet_napi_enable(rq
);
936 /* In theory, this can happen: if we don't get any buffers in
937 * we will *never* try to fill again.
940 schedule_delayed_work(&vi
->refill
, HZ
/2);
944 static int virtnet_receive(struct receive_queue
*rq
, int budget
)
946 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
947 unsigned int len
, received
= 0;
950 while (received
< budget
&&
951 (buf
= virtqueue_get_buf(rq
->vq
, &len
)) != NULL
) {
952 receive_buf(vi
, rq
, buf
, len
);
956 if (rq
->vq
->num_free
> virtqueue_get_vring_size(rq
->vq
) / 2) {
957 if (!try_fill_recv(vi
, rq
, GFP_ATOMIC
))
958 schedule_delayed_work(&vi
->refill
, 0);
964 static int virtnet_poll(struct napi_struct
*napi
, int budget
)
966 struct receive_queue
*rq
=
967 container_of(napi
, struct receive_queue
, napi
);
968 unsigned int r
, received
;
970 received
= virtnet_receive(rq
, budget
);
972 /* Out of packets? */
973 if (received
< budget
) {
974 r
= virtqueue_enable_cb_prepare(rq
->vq
);
975 napi_complete_done(napi
, received
);
976 if (unlikely(virtqueue_poll(rq
->vq
, r
)) &&
977 napi_schedule_prep(napi
)) {
978 virtqueue_disable_cb(rq
->vq
);
979 __napi_schedule(napi
);
986 #ifdef CONFIG_NET_RX_BUSY_POLL
987 /* must be called with local_bh_disable()d */
988 static int virtnet_busy_poll(struct napi_struct
*napi
)
990 struct receive_queue
*rq
=
991 container_of(napi
, struct receive_queue
, napi
);
992 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
993 int r
, received
= 0, budget
= 4;
995 if (!(vi
->status
& VIRTIO_NET_S_LINK_UP
))
996 return LL_FLUSH_FAILED
;
998 if (!napi_schedule_prep(napi
))
999 return LL_FLUSH_BUSY
;
1001 virtqueue_disable_cb(rq
->vq
);
1004 received
+= virtnet_receive(rq
, budget
);
1006 r
= virtqueue_enable_cb_prepare(rq
->vq
);
1007 clear_bit(NAPI_STATE_SCHED
, &napi
->state
);
1008 if (unlikely(virtqueue_poll(rq
->vq
, r
)) &&
1009 napi_schedule_prep(napi
)) {
1010 virtqueue_disable_cb(rq
->vq
);
1011 if (received
< budget
) {
1015 __napi_schedule(napi
);
1021 #endif /* CONFIG_NET_RX_BUSY_POLL */
1023 static int virtnet_open(struct net_device
*dev
)
1025 struct virtnet_info
*vi
= netdev_priv(dev
);
1028 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1029 if (i
< vi
->curr_queue_pairs
)
1030 /* Make sure we have some buffers: if oom use wq. */
1031 if (!try_fill_recv(vi
, &vi
->rq
[i
], GFP_KERNEL
))
1032 schedule_delayed_work(&vi
->refill
, 0);
1033 virtnet_napi_enable(&vi
->rq
[i
]);
1039 static void free_old_xmit_skbs(struct send_queue
*sq
)
1041 struct sk_buff
*skb
;
1043 struct virtnet_info
*vi
= sq
->vq
->vdev
->priv
;
1044 struct virtnet_stats
*stats
= this_cpu_ptr(vi
->stats
);
1046 while ((skb
= virtqueue_get_buf(sq
->vq
, &len
)) != NULL
) {
1047 pr_debug("Sent skb %p\n", skb
);
1049 u64_stats_update_begin(&stats
->tx_syncp
);
1050 stats
->tx_bytes
+= skb
->len
;
1051 stats
->tx_packets
++;
1052 u64_stats_update_end(&stats
->tx_syncp
);
1054 dev_kfree_skb_any(skb
);
1058 static int xmit_skb(struct send_queue
*sq
, struct sk_buff
*skb
)
1060 struct virtio_net_hdr_mrg_rxbuf
*hdr
;
1061 const unsigned char *dest
= ((struct ethhdr
*)skb
->data
)->h_dest
;
1062 struct virtnet_info
*vi
= sq
->vq
->vdev
->priv
;
1064 unsigned hdr_len
= vi
->hdr_len
;
1067 pr_debug("%s: xmit %p %pM\n", vi
->dev
->name
, skb
, dest
);
1069 can_push
= vi
->any_header_sg
&&
1070 !((unsigned long)skb
->data
& (__alignof__(*hdr
) - 1)) &&
1071 !skb_header_cloned(skb
) && skb_headroom(skb
) >= hdr_len
;
1072 /* Even if we can, don't push here yet as this would skew
1073 * csum_start offset below. */
1075 hdr
= (struct virtio_net_hdr_mrg_rxbuf
*)(skb
->data
- hdr_len
);
1077 hdr
= skb_vnet_hdr(skb
);
1079 if (virtio_net_hdr_from_skb(skb
, &hdr
->hdr
,
1080 virtio_is_little_endian(vi
->vdev
)))
1083 if (vi
->mergeable_rx_bufs
)
1084 hdr
->num_buffers
= 0;
1086 sg_init_table(sq
->sg
, skb_shinfo(skb
)->nr_frags
+ (can_push
? 1 : 2));
1088 __skb_push(skb
, hdr_len
);
1089 num_sg
= skb_to_sgvec(skb
, sq
->sg
, 0, skb
->len
);
1090 /* Pull header back to avoid skew in tx bytes calculations. */
1091 __skb_pull(skb
, hdr_len
);
1093 sg_set_buf(sq
->sg
, hdr
, hdr_len
);
1094 num_sg
= skb_to_sgvec(skb
, sq
->sg
+ 1, 0, skb
->len
) + 1;
1096 return virtqueue_add_outbuf(sq
->vq
, sq
->sg
, num_sg
, skb
, GFP_ATOMIC
);
1099 static netdev_tx_t
start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
1101 struct virtnet_info
*vi
= netdev_priv(dev
);
1102 int qnum
= skb_get_queue_mapping(skb
);
1103 struct send_queue
*sq
= &vi
->sq
[qnum
];
1105 struct netdev_queue
*txq
= netdev_get_tx_queue(dev
, qnum
);
1106 bool kick
= !skb
->xmit_more
;
1108 /* Free up any pending old buffers before queueing new ones. */
1109 free_old_xmit_skbs(sq
);
1111 /* timestamp packet in software */
1112 skb_tx_timestamp(skb
);
1114 /* Try to transmit */
1115 err
= xmit_skb(sq
, skb
);
1117 /* This should not happen! */
1118 if (unlikely(err
)) {
1119 dev
->stats
.tx_fifo_errors
++;
1120 if (net_ratelimit())
1122 "Unexpected TXQ (%d) queue failure: %d\n", qnum
, err
);
1123 dev
->stats
.tx_dropped
++;
1124 dev_kfree_skb_any(skb
);
1125 return NETDEV_TX_OK
;
1128 /* Don't wait up for transmitted skbs to be freed. */
1132 /* If running out of space, stop queue to avoid getting packets that we
1133 * are then unable to transmit.
1134 * An alternative would be to force queuing layer to requeue the skb by
1135 * returning NETDEV_TX_BUSY. However, NETDEV_TX_BUSY should not be
1136 * returned in a normal path of operation: it means that driver is not
1137 * maintaining the TX queue stop/start state properly, and causes
1138 * the stack to do a non-trivial amount of useless work.
1139 * Since most packets only take 1 or 2 ring slots, stopping the queue
1140 * early means 16 slots are typically wasted.
1142 if (sq
->vq
->num_free
< 2+MAX_SKB_FRAGS
) {
1143 netif_stop_subqueue(dev
, qnum
);
1144 if (unlikely(!virtqueue_enable_cb_delayed(sq
->vq
))) {
1145 /* More just got used, free them then recheck. */
1146 free_old_xmit_skbs(sq
);
1147 if (sq
->vq
->num_free
>= 2+MAX_SKB_FRAGS
) {
1148 netif_start_subqueue(dev
, qnum
);
1149 virtqueue_disable_cb(sq
->vq
);
1154 if (kick
|| netif_xmit_stopped(txq
))
1155 virtqueue_kick(sq
->vq
);
1157 return NETDEV_TX_OK
;
1161 * Send command via the control virtqueue and check status. Commands
1162 * supported by the hypervisor, as indicated by feature bits, should
1163 * never fail unless improperly formatted.
1165 static bool virtnet_send_command(struct virtnet_info
*vi
, u8
class, u8 cmd
,
1166 struct scatterlist
*out
)
1168 struct scatterlist
*sgs
[4], hdr
, stat
;
1169 unsigned out_num
= 0, tmp
;
1171 /* Caller should know better */
1172 BUG_ON(!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
));
1174 vi
->ctrl_status
= ~0;
1175 vi
->ctrl_hdr
.class = class;
1176 vi
->ctrl_hdr
.cmd
= cmd
;
1178 sg_init_one(&hdr
, &vi
->ctrl_hdr
, sizeof(vi
->ctrl_hdr
));
1179 sgs
[out_num
++] = &hdr
;
1182 sgs
[out_num
++] = out
;
1184 /* Add return status. */
1185 sg_init_one(&stat
, &vi
->ctrl_status
, sizeof(vi
->ctrl_status
));
1186 sgs
[out_num
] = &stat
;
1188 BUG_ON(out_num
+ 1 > ARRAY_SIZE(sgs
));
1189 virtqueue_add_sgs(vi
->cvq
, sgs
, out_num
, 1, vi
, GFP_ATOMIC
);
1191 if (unlikely(!virtqueue_kick(vi
->cvq
)))
1192 return vi
->ctrl_status
== VIRTIO_NET_OK
;
1194 /* Spin for a response, the kick causes an ioport write, trapping
1195 * into the hypervisor, so the request should be handled immediately.
1197 while (!virtqueue_get_buf(vi
->cvq
, &tmp
) &&
1198 !virtqueue_is_broken(vi
->cvq
))
1201 return vi
->ctrl_status
== VIRTIO_NET_OK
;
1204 static int virtnet_set_mac_address(struct net_device
*dev
, void *p
)
1206 struct virtnet_info
*vi
= netdev_priv(dev
);
1207 struct virtio_device
*vdev
= vi
->vdev
;
1209 struct sockaddr
*addr
;
1210 struct scatterlist sg
;
1212 addr
= kmalloc(sizeof(*addr
), GFP_KERNEL
);
1215 memcpy(addr
, p
, sizeof(*addr
));
1217 ret
= eth_prepare_mac_addr_change(dev
, addr
);
1221 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_MAC_ADDR
)) {
1222 sg_init_one(&sg
, addr
->sa_data
, dev
->addr_len
);
1223 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MAC
,
1224 VIRTIO_NET_CTRL_MAC_ADDR_SET
, &sg
)) {
1225 dev_warn(&vdev
->dev
,
1226 "Failed to set mac address by vq command.\n");
1230 } else if (virtio_has_feature(vdev
, VIRTIO_NET_F_MAC
) &&
1231 !virtio_has_feature(vdev
, VIRTIO_F_VERSION_1
)) {
1234 /* Naturally, this has an atomicity problem. */
1235 for (i
= 0; i
< dev
->addr_len
; i
++)
1236 virtio_cwrite8(vdev
,
1237 offsetof(struct virtio_net_config
, mac
) +
1238 i
, addr
->sa_data
[i
]);
1241 eth_commit_mac_addr_change(dev
, p
);
1249 static struct rtnl_link_stats64
*virtnet_stats(struct net_device
*dev
,
1250 struct rtnl_link_stats64
*tot
)
1252 struct virtnet_info
*vi
= netdev_priv(dev
);
1256 for_each_possible_cpu(cpu
) {
1257 struct virtnet_stats
*stats
= per_cpu_ptr(vi
->stats
, cpu
);
1258 u64 tpackets
, tbytes
, rpackets
, rbytes
;
1261 start
= u64_stats_fetch_begin_irq(&stats
->tx_syncp
);
1262 tpackets
= stats
->tx_packets
;
1263 tbytes
= stats
->tx_bytes
;
1264 } while (u64_stats_fetch_retry_irq(&stats
->tx_syncp
, start
));
1267 start
= u64_stats_fetch_begin_irq(&stats
->rx_syncp
);
1268 rpackets
= stats
->rx_packets
;
1269 rbytes
= stats
->rx_bytes
;
1270 } while (u64_stats_fetch_retry_irq(&stats
->rx_syncp
, start
));
1272 tot
->rx_packets
+= rpackets
;
1273 tot
->tx_packets
+= tpackets
;
1274 tot
->rx_bytes
+= rbytes
;
1275 tot
->tx_bytes
+= tbytes
;
1278 tot
->tx_dropped
= dev
->stats
.tx_dropped
;
1279 tot
->tx_fifo_errors
= dev
->stats
.tx_fifo_errors
;
1280 tot
->rx_dropped
= dev
->stats
.rx_dropped
;
1281 tot
->rx_length_errors
= dev
->stats
.rx_length_errors
;
1282 tot
->rx_frame_errors
= dev
->stats
.rx_frame_errors
;
1287 #ifdef CONFIG_NET_POLL_CONTROLLER
1288 static void virtnet_netpoll(struct net_device
*dev
)
1290 struct virtnet_info
*vi
= netdev_priv(dev
);
1293 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++)
1294 napi_schedule(&vi
->rq
[i
].napi
);
1298 static void virtnet_ack_link_announce(struct virtnet_info
*vi
)
1301 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_ANNOUNCE
,
1302 VIRTIO_NET_CTRL_ANNOUNCE_ACK
, NULL
))
1303 dev_warn(&vi
->dev
->dev
, "Failed to ack link announce.\n");
1307 static int virtnet_set_queues(struct virtnet_info
*vi
, u16 queue_pairs
)
1309 struct scatterlist sg
;
1310 struct net_device
*dev
= vi
->dev
;
1312 if (!vi
->has_cvq
|| !virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_MQ
))
1315 vi
->ctrl_mq
.virtqueue_pairs
= cpu_to_virtio16(vi
->vdev
, queue_pairs
);
1316 sg_init_one(&sg
, &vi
->ctrl_mq
, sizeof(vi
->ctrl_mq
));
1318 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MQ
,
1319 VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET
, &sg
)) {
1320 dev_warn(&dev
->dev
, "Fail to set num of queue pairs to %d\n",
1324 vi
->curr_queue_pairs
= queue_pairs
;
1325 /* virtnet_open() will refill when device is going to up. */
1326 if (dev
->flags
& IFF_UP
)
1327 schedule_delayed_work(&vi
->refill
, 0);
1333 static int virtnet_close(struct net_device
*dev
)
1335 struct virtnet_info
*vi
= netdev_priv(dev
);
1338 /* Make sure refill_work doesn't re-enable napi! */
1339 cancel_delayed_work_sync(&vi
->refill
);
1341 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
1342 napi_disable(&vi
->rq
[i
].napi
);
1347 static void virtnet_set_rx_mode(struct net_device
*dev
)
1349 struct virtnet_info
*vi
= netdev_priv(dev
);
1350 struct scatterlist sg
[2];
1351 struct virtio_net_ctrl_mac
*mac_data
;
1352 struct netdev_hw_addr
*ha
;
1358 /* We can't dynamically set ndo_set_rx_mode, so return gracefully */
1359 if (!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_RX
))
1362 vi
->ctrl_promisc
= ((dev
->flags
& IFF_PROMISC
) != 0);
1363 vi
->ctrl_allmulti
= ((dev
->flags
& IFF_ALLMULTI
) != 0);
1365 sg_init_one(sg
, &vi
->ctrl_promisc
, sizeof(vi
->ctrl_promisc
));
1367 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_RX
,
1368 VIRTIO_NET_CTRL_RX_PROMISC
, sg
))
1369 dev_warn(&dev
->dev
, "Failed to %sable promisc mode.\n",
1370 vi
->ctrl_promisc
? "en" : "dis");
1372 sg_init_one(sg
, &vi
->ctrl_allmulti
, sizeof(vi
->ctrl_allmulti
));
1374 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_RX
,
1375 VIRTIO_NET_CTRL_RX_ALLMULTI
, sg
))
1376 dev_warn(&dev
->dev
, "Failed to %sable allmulti mode.\n",
1377 vi
->ctrl_allmulti
? "en" : "dis");
1379 uc_count
= netdev_uc_count(dev
);
1380 mc_count
= netdev_mc_count(dev
);
1381 /* MAC filter - use one buffer for both lists */
1382 buf
= kzalloc(((uc_count
+ mc_count
) * ETH_ALEN
) +
1383 (2 * sizeof(mac_data
->entries
)), GFP_ATOMIC
);
1388 sg_init_table(sg
, 2);
1390 /* Store the unicast list and count in the front of the buffer */
1391 mac_data
->entries
= cpu_to_virtio32(vi
->vdev
, uc_count
);
1393 netdev_for_each_uc_addr(ha
, dev
)
1394 memcpy(&mac_data
->macs
[i
++][0], ha
->addr
, ETH_ALEN
);
1396 sg_set_buf(&sg
[0], mac_data
,
1397 sizeof(mac_data
->entries
) + (uc_count
* ETH_ALEN
));
1399 /* multicast list and count fill the end */
1400 mac_data
= (void *)&mac_data
->macs
[uc_count
][0];
1402 mac_data
->entries
= cpu_to_virtio32(vi
->vdev
, mc_count
);
1404 netdev_for_each_mc_addr(ha
, dev
)
1405 memcpy(&mac_data
->macs
[i
++][0], ha
->addr
, ETH_ALEN
);
1407 sg_set_buf(&sg
[1], mac_data
,
1408 sizeof(mac_data
->entries
) + (mc_count
* ETH_ALEN
));
1410 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MAC
,
1411 VIRTIO_NET_CTRL_MAC_TABLE_SET
, sg
))
1412 dev_warn(&dev
->dev
, "Failed to set MAC filter table.\n");
1417 static int virtnet_vlan_rx_add_vid(struct net_device
*dev
,
1418 __be16 proto
, u16 vid
)
1420 struct virtnet_info
*vi
= netdev_priv(dev
);
1421 struct scatterlist sg
;
1424 sg_init_one(&sg
, &vi
->ctrl_vid
, sizeof(vi
->ctrl_vid
));
1426 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_VLAN
,
1427 VIRTIO_NET_CTRL_VLAN_ADD
, &sg
))
1428 dev_warn(&dev
->dev
, "Failed to add VLAN ID %d.\n", vid
);
1432 static int virtnet_vlan_rx_kill_vid(struct net_device
*dev
,
1433 __be16 proto
, u16 vid
)
1435 struct virtnet_info
*vi
= netdev_priv(dev
);
1436 struct scatterlist sg
;
1439 sg_init_one(&sg
, &vi
->ctrl_vid
, sizeof(vi
->ctrl_vid
));
1441 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_VLAN
,
1442 VIRTIO_NET_CTRL_VLAN_DEL
, &sg
))
1443 dev_warn(&dev
->dev
, "Failed to kill VLAN ID %d.\n", vid
);
1447 static void virtnet_clean_affinity(struct virtnet_info
*vi
, long hcpu
)
1451 if (vi
->affinity_hint_set
) {
1452 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1453 virtqueue_set_affinity(vi
->rq
[i
].vq
, -1);
1454 virtqueue_set_affinity(vi
->sq
[i
].vq
, -1);
1457 vi
->affinity_hint_set
= false;
1461 static void virtnet_set_affinity(struct virtnet_info
*vi
)
1466 /* In multiqueue mode, when the number of cpu is equal to the number of
1467 * queue pairs, we let the queue pairs to be private to one cpu by
1468 * setting the affinity hint to eliminate the contention.
1470 if (vi
->curr_queue_pairs
== 1 ||
1471 vi
->max_queue_pairs
!= num_online_cpus()) {
1472 virtnet_clean_affinity(vi
, -1);
1477 for_each_online_cpu(cpu
) {
1478 virtqueue_set_affinity(vi
->rq
[i
].vq
, cpu
);
1479 virtqueue_set_affinity(vi
->sq
[i
].vq
, cpu
);
1480 netif_set_xps_queue(vi
->dev
, cpumask_of(cpu
), i
);
1484 vi
->affinity_hint_set
= true;
1487 static int virtnet_cpu_online(unsigned int cpu
, struct hlist_node
*node
)
1489 struct virtnet_info
*vi
= hlist_entry_safe(node
, struct virtnet_info
,
1491 virtnet_set_affinity(vi
);
1495 static int virtnet_cpu_dead(unsigned int cpu
, struct hlist_node
*node
)
1497 struct virtnet_info
*vi
= hlist_entry_safe(node
, struct virtnet_info
,
1499 virtnet_set_affinity(vi
);
1503 static int virtnet_cpu_down_prep(unsigned int cpu
, struct hlist_node
*node
)
1505 struct virtnet_info
*vi
= hlist_entry_safe(node
, struct virtnet_info
,
1508 virtnet_clean_affinity(vi
, cpu
);
1512 static enum cpuhp_state virtionet_online
;
1514 static int virtnet_cpu_notif_add(struct virtnet_info
*vi
)
1518 ret
= cpuhp_state_add_instance_nocalls(virtionet_online
, &vi
->node
);
1521 ret
= cpuhp_state_add_instance_nocalls(CPUHP_VIRT_NET_DEAD
,
1525 cpuhp_state_remove_instance_nocalls(virtionet_online
, &vi
->node
);
1529 static void virtnet_cpu_notif_remove(struct virtnet_info
*vi
)
1531 cpuhp_state_remove_instance_nocalls(virtionet_online
, &vi
->node
);
1532 cpuhp_state_remove_instance_nocalls(CPUHP_VIRT_NET_DEAD
,
1536 static void virtnet_get_ringparam(struct net_device
*dev
,
1537 struct ethtool_ringparam
*ring
)
1539 struct virtnet_info
*vi
= netdev_priv(dev
);
1541 ring
->rx_max_pending
= virtqueue_get_vring_size(vi
->rq
[0].vq
);
1542 ring
->tx_max_pending
= virtqueue_get_vring_size(vi
->sq
[0].vq
);
1543 ring
->rx_pending
= ring
->rx_max_pending
;
1544 ring
->tx_pending
= ring
->tx_max_pending
;
1548 static void virtnet_get_drvinfo(struct net_device
*dev
,
1549 struct ethtool_drvinfo
*info
)
1551 struct virtnet_info
*vi
= netdev_priv(dev
);
1552 struct virtio_device
*vdev
= vi
->vdev
;
1554 strlcpy(info
->driver
, KBUILD_MODNAME
, sizeof(info
->driver
));
1555 strlcpy(info
->version
, VIRTNET_DRIVER_VERSION
, sizeof(info
->version
));
1556 strlcpy(info
->bus_info
, virtio_bus_name(vdev
), sizeof(info
->bus_info
));
1560 /* TODO: Eliminate OOO packets during switching */
1561 static int virtnet_set_channels(struct net_device
*dev
,
1562 struct ethtool_channels
*channels
)
1564 struct virtnet_info
*vi
= netdev_priv(dev
);
1565 u16 queue_pairs
= channels
->combined_count
;
1568 /* We don't support separate rx/tx channels.
1569 * We don't allow setting 'other' channels.
1571 if (channels
->rx_count
|| channels
->tx_count
|| channels
->other_count
)
1574 if (queue_pairs
> vi
->max_queue_pairs
|| queue_pairs
== 0)
1577 /* For now we don't support modifying channels while XDP is loaded
1578 * also when XDP is loaded all RX queues have XDP programs so we only
1579 * need to check a single RX queue.
1581 if (vi
->rq
[0].xdp_prog
)
1585 err
= virtnet_set_queues(vi
, queue_pairs
);
1587 netif_set_real_num_tx_queues(dev
, queue_pairs
);
1588 netif_set_real_num_rx_queues(dev
, queue_pairs
);
1590 virtnet_set_affinity(vi
);
1597 static void virtnet_get_channels(struct net_device
*dev
,
1598 struct ethtool_channels
*channels
)
1600 struct virtnet_info
*vi
= netdev_priv(dev
);
1602 channels
->combined_count
= vi
->curr_queue_pairs
;
1603 channels
->max_combined
= vi
->max_queue_pairs
;
1604 channels
->max_other
= 0;
1605 channels
->rx_count
= 0;
1606 channels
->tx_count
= 0;
1607 channels
->other_count
= 0;
1610 /* Check if the user is trying to change anything besides speed/duplex */
1611 static bool virtnet_validate_ethtool_cmd(const struct ethtool_cmd
*cmd
)
1613 struct ethtool_cmd diff1
= *cmd
;
1614 struct ethtool_cmd diff2
= {};
1616 /* cmd is always set so we need to clear it, validate the port type
1617 * and also without autonegotiation we can ignore advertising
1619 ethtool_cmd_speed_set(&diff1
, 0);
1620 diff2
.port
= PORT_OTHER
;
1621 diff1
.advertising
= 0;
1625 return !memcmp(&diff1
, &diff2
, sizeof(diff1
));
1628 static int virtnet_set_settings(struct net_device
*dev
, struct ethtool_cmd
*cmd
)
1630 struct virtnet_info
*vi
= netdev_priv(dev
);
1633 speed
= ethtool_cmd_speed(cmd
);
1634 /* don't allow custom speed and duplex */
1635 if (!ethtool_validate_speed(speed
) ||
1636 !ethtool_validate_duplex(cmd
->duplex
) ||
1637 !virtnet_validate_ethtool_cmd(cmd
))
1640 vi
->duplex
= cmd
->duplex
;
1645 static int virtnet_get_settings(struct net_device
*dev
, struct ethtool_cmd
*cmd
)
1647 struct virtnet_info
*vi
= netdev_priv(dev
);
1649 ethtool_cmd_speed_set(cmd
, vi
->speed
);
1650 cmd
->duplex
= vi
->duplex
;
1651 cmd
->port
= PORT_OTHER
;
1656 static void virtnet_init_settings(struct net_device
*dev
)
1658 struct virtnet_info
*vi
= netdev_priv(dev
);
1660 vi
->speed
= SPEED_UNKNOWN
;
1661 vi
->duplex
= DUPLEX_UNKNOWN
;
1664 static const struct ethtool_ops virtnet_ethtool_ops
= {
1665 .get_drvinfo
= virtnet_get_drvinfo
,
1666 .get_link
= ethtool_op_get_link
,
1667 .get_ringparam
= virtnet_get_ringparam
,
1668 .set_channels
= virtnet_set_channels
,
1669 .get_channels
= virtnet_get_channels
,
1670 .get_ts_info
= ethtool_op_get_ts_info
,
1671 .get_settings
= virtnet_get_settings
,
1672 .set_settings
= virtnet_set_settings
,
1675 static int virtnet_xdp_set(struct net_device
*dev
, struct bpf_prog
*prog
)
1677 unsigned long int max_sz
= PAGE_SIZE
- sizeof(struct padded_vnet_hdr
);
1678 struct virtnet_info
*vi
= netdev_priv(dev
);
1679 struct bpf_prog
*old_prog
;
1680 u16 xdp_qp
= 0, curr_qp
;
1683 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_GUEST_TSO4
) ||
1684 virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_GUEST_TSO6
)) {
1685 netdev_warn(dev
, "can't set XDP while host is implementing LRO, disable LRO first\n");
1689 if (vi
->mergeable_rx_bufs
&& !vi
->any_header_sg
) {
1690 netdev_warn(dev
, "XDP expects header/data in single page, any_header_sg required\n");
1694 if (dev
->mtu
> max_sz
) {
1695 netdev_warn(dev
, "XDP requires MTU less than %lu\n", max_sz
);
1699 curr_qp
= vi
->curr_queue_pairs
- vi
->xdp_queue_pairs
;
1701 xdp_qp
= nr_cpu_ids
;
1703 /* XDP requires extra queues for XDP_TX */
1704 if (curr_qp
+ xdp_qp
> vi
->max_queue_pairs
) {
1705 netdev_warn(dev
, "request %i queues but max is %i\n",
1706 curr_qp
+ xdp_qp
, vi
->max_queue_pairs
);
1710 err
= virtnet_set_queues(vi
, curr_qp
+ xdp_qp
);
1712 dev_warn(&dev
->dev
, "XDP Device queue allocation failure.\n");
1717 prog
= bpf_prog_add(prog
, vi
->max_queue_pairs
- 1);
1719 virtnet_set_queues(vi
, curr_qp
);
1720 return PTR_ERR(prog
);
1724 vi
->xdp_queue_pairs
= xdp_qp
;
1725 netif_set_real_num_rx_queues(dev
, curr_qp
+ xdp_qp
);
1727 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1728 old_prog
= rtnl_dereference(vi
->rq
[i
].xdp_prog
);
1729 rcu_assign_pointer(vi
->rq
[i
].xdp_prog
, prog
);
1731 bpf_prog_put(old_prog
);
1737 static bool virtnet_xdp_query(struct net_device
*dev
)
1739 struct virtnet_info
*vi
= netdev_priv(dev
);
1742 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1743 if (vi
->rq
[i
].xdp_prog
)
1749 static int virtnet_xdp(struct net_device
*dev
, struct netdev_xdp
*xdp
)
1751 switch (xdp
->command
) {
1752 case XDP_SETUP_PROG
:
1753 return virtnet_xdp_set(dev
, xdp
->prog
);
1754 case XDP_QUERY_PROG
:
1755 xdp
->prog_attached
= virtnet_xdp_query(dev
);
1762 static const struct net_device_ops virtnet_netdev
= {
1763 .ndo_open
= virtnet_open
,
1764 .ndo_stop
= virtnet_close
,
1765 .ndo_start_xmit
= start_xmit
,
1766 .ndo_validate_addr
= eth_validate_addr
,
1767 .ndo_set_mac_address
= virtnet_set_mac_address
,
1768 .ndo_set_rx_mode
= virtnet_set_rx_mode
,
1769 .ndo_get_stats64
= virtnet_stats
,
1770 .ndo_vlan_rx_add_vid
= virtnet_vlan_rx_add_vid
,
1771 .ndo_vlan_rx_kill_vid
= virtnet_vlan_rx_kill_vid
,
1772 #ifdef CONFIG_NET_POLL_CONTROLLER
1773 .ndo_poll_controller
= virtnet_netpoll
,
1775 #ifdef CONFIG_NET_RX_BUSY_POLL
1776 .ndo_busy_poll
= virtnet_busy_poll
,
1778 .ndo_xdp
= virtnet_xdp
,
1781 static void virtnet_config_changed_work(struct work_struct
*work
)
1783 struct virtnet_info
*vi
=
1784 container_of(work
, struct virtnet_info
, config_work
);
1787 if (virtio_cread_feature(vi
->vdev
, VIRTIO_NET_F_STATUS
,
1788 struct virtio_net_config
, status
, &v
) < 0)
1791 if (v
& VIRTIO_NET_S_ANNOUNCE
) {
1792 netdev_notify_peers(vi
->dev
);
1793 virtnet_ack_link_announce(vi
);
1796 /* Ignore unknown (future) status bits */
1797 v
&= VIRTIO_NET_S_LINK_UP
;
1799 if (vi
->status
== v
)
1804 if (vi
->status
& VIRTIO_NET_S_LINK_UP
) {
1805 netif_carrier_on(vi
->dev
);
1806 netif_tx_wake_all_queues(vi
->dev
);
1808 netif_carrier_off(vi
->dev
);
1809 netif_tx_stop_all_queues(vi
->dev
);
1813 static void virtnet_config_changed(struct virtio_device
*vdev
)
1815 struct virtnet_info
*vi
= vdev
->priv
;
1817 schedule_work(&vi
->config_work
);
1820 static void virtnet_free_queues(struct virtnet_info
*vi
)
1824 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1825 napi_hash_del(&vi
->rq
[i
].napi
);
1826 netif_napi_del(&vi
->rq
[i
].napi
);
1829 /* We called napi_hash_del() before netif_napi_del(),
1830 * we need to respect an RCU grace period before freeing vi->rq
1838 static void free_receive_bufs(struct virtnet_info
*vi
)
1840 struct bpf_prog
*old_prog
;
1844 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1845 while (vi
->rq
[i
].pages
)
1846 __free_pages(get_a_page(&vi
->rq
[i
], GFP_KERNEL
), 0);
1848 old_prog
= rtnl_dereference(vi
->rq
[i
].xdp_prog
);
1849 RCU_INIT_POINTER(vi
->rq
[i
].xdp_prog
, NULL
);
1851 bpf_prog_put(old_prog
);
1856 static void free_receive_page_frags(struct virtnet_info
*vi
)
1859 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
1860 if (vi
->rq
[i
].alloc_frag
.page
)
1861 put_page(vi
->rq
[i
].alloc_frag
.page
);
1864 static bool is_xdp_queue(struct virtnet_info
*vi
, int q
)
1866 if (q
< (vi
->curr_queue_pairs
- vi
->xdp_queue_pairs
))
1868 else if (q
< vi
->curr_queue_pairs
)
1874 static void free_unused_bufs(struct virtnet_info
*vi
)
1879 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1880 struct virtqueue
*vq
= vi
->sq
[i
].vq
;
1881 while ((buf
= virtqueue_detach_unused_buf(vq
)) != NULL
) {
1882 if (!is_xdp_queue(vi
, i
))
1885 put_page(virt_to_head_page(buf
));
1889 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1890 struct virtqueue
*vq
= vi
->rq
[i
].vq
;
1892 while ((buf
= virtqueue_detach_unused_buf(vq
)) != NULL
) {
1893 if (vi
->mergeable_rx_bufs
) {
1894 unsigned long ctx
= (unsigned long)buf
;
1895 void *base
= mergeable_ctx_to_buf_address(ctx
);
1896 put_page(virt_to_head_page(base
));
1897 } else if (vi
->big_packets
) {
1898 give_pages(&vi
->rq
[i
], buf
);
1906 static void virtnet_del_vqs(struct virtnet_info
*vi
)
1908 struct virtio_device
*vdev
= vi
->vdev
;
1910 virtnet_clean_affinity(vi
, -1);
1912 vdev
->config
->del_vqs(vdev
);
1914 virtnet_free_queues(vi
);
1917 static int virtnet_find_vqs(struct virtnet_info
*vi
)
1919 vq_callback_t
**callbacks
;
1920 struct virtqueue
**vqs
;
1925 /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
1926 * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
1927 * possible control vq.
1929 total_vqs
= vi
->max_queue_pairs
* 2 +
1930 virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
);
1932 /* Allocate space for find_vqs parameters */
1933 vqs
= kzalloc(total_vqs
* sizeof(*vqs
), GFP_KERNEL
);
1936 callbacks
= kmalloc(total_vqs
* sizeof(*callbacks
), GFP_KERNEL
);
1939 names
= kmalloc(total_vqs
* sizeof(*names
), GFP_KERNEL
);
1943 /* Parameters for control virtqueue, if any */
1945 callbacks
[total_vqs
- 1] = NULL
;
1946 names
[total_vqs
- 1] = "control";
1949 /* Allocate/initialize parameters for send/receive virtqueues */
1950 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1951 callbacks
[rxq2vq(i
)] = skb_recv_done
;
1952 callbacks
[txq2vq(i
)] = skb_xmit_done
;
1953 sprintf(vi
->rq
[i
].name
, "input.%d", i
);
1954 sprintf(vi
->sq
[i
].name
, "output.%d", i
);
1955 names
[rxq2vq(i
)] = vi
->rq
[i
].name
;
1956 names
[txq2vq(i
)] = vi
->sq
[i
].name
;
1959 ret
= vi
->vdev
->config
->find_vqs(vi
->vdev
, total_vqs
, vqs
, callbacks
,
1965 vi
->cvq
= vqs
[total_vqs
- 1];
1966 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VLAN
))
1967 vi
->dev
->features
|= NETIF_F_HW_VLAN_CTAG_FILTER
;
1970 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1971 vi
->rq
[i
].vq
= vqs
[rxq2vq(i
)];
1972 vi
->sq
[i
].vq
= vqs
[txq2vq(i
)];
1991 static int virtnet_alloc_queues(struct virtnet_info
*vi
)
1995 vi
->sq
= kzalloc(sizeof(*vi
->sq
) * vi
->max_queue_pairs
, GFP_KERNEL
);
1998 vi
->rq
= kzalloc(sizeof(*vi
->rq
) * vi
->max_queue_pairs
, GFP_KERNEL
);
2002 INIT_DELAYED_WORK(&vi
->refill
, refill_work
);
2003 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
2004 vi
->rq
[i
].pages
= NULL
;
2005 netif_napi_add(vi
->dev
, &vi
->rq
[i
].napi
, virtnet_poll
,
2008 sg_init_table(vi
->rq
[i
].sg
, ARRAY_SIZE(vi
->rq
[i
].sg
));
2009 ewma_pkt_len_init(&vi
->rq
[i
].mrg_avg_pkt_len
);
2010 sg_init_table(vi
->sq
[i
].sg
, ARRAY_SIZE(vi
->sq
[i
].sg
));
2021 static int init_vqs(struct virtnet_info
*vi
)
2025 /* Allocate send & receive queues */
2026 ret
= virtnet_alloc_queues(vi
);
2030 ret
= virtnet_find_vqs(vi
);
2035 virtnet_set_affinity(vi
);
2041 virtnet_free_queues(vi
);
2047 static ssize_t
mergeable_rx_buffer_size_show(struct netdev_rx_queue
*queue
,
2048 struct rx_queue_attribute
*attribute
, char *buf
)
2050 struct virtnet_info
*vi
= netdev_priv(queue
->dev
);
2051 unsigned int queue_index
= get_netdev_rx_queue_index(queue
);
2052 struct ewma_pkt_len
*avg
;
2054 BUG_ON(queue_index
>= vi
->max_queue_pairs
);
2055 avg
= &vi
->rq
[queue_index
].mrg_avg_pkt_len
;
2056 return sprintf(buf
, "%u\n", get_mergeable_buf_len(avg
));
2059 static struct rx_queue_attribute mergeable_rx_buffer_size_attribute
=
2060 __ATTR_RO(mergeable_rx_buffer_size
);
2062 static struct attribute
*virtio_net_mrg_rx_attrs
[] = {
2063 &mergeable_rx_buffer_size_attribute
.attr
,
2067 static const struct attribute_group virtio_net_mrg_rx_group
= {
2068 .name
= "virtio_net",
2069 .attrs
= virtio_net_mrg_rx_attrs
2073 static bool virtnet_fail_on_feature(struct virtio_device
*vdev
,
2075 const char *fname
, const char *dname
)
2077 if (!virtio_has_feature(vdev
, fbit
))
2080 dev_err(&vdev
->dev
, "device advertises feature %s but not %s",
2086 #define VIRTNET_FAIL_ON(vdev, fbit, dbit) \
2087 virtnet_fail_on_feature(vdev, fbit, #fbit, dbit)
2089 static bool virtnet_validate_features(struct virtio_device
*vdev
)
2091 if (!virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
) &&
2092 (VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_CTRL_RX
,
2093 "VIRTIO_NET_F_CTRL_VQ") ||
2094 VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_CTRL_VLAN
,
2095 "VIRTIO_NET_F_CTRL_VQ") ||
2096 VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_GUEST_ANNOUNCE
,
2097 "VIRTIO_NET_F_CTRL_VQ") ||
2098 VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_MQ
, "VIRTIO_NET_F_CTRL_VQ") ||
2099 VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_CTRL_MAC_ADDR
,
2100 "VIRTIO_NET_F_CTRL_VQ"))) {
2107 #define MIN_MTU ETH_MIN_MTU
2108 #define MAX_MTU ETH_MAX_MTU
2110 static int virtnet_probe(struct virtio_device
*vdev
)
2113 struct net_device
*dev
;
2114 struct virtnet_info
*vi
;
2115 u16 max_queue_pairs
;
2118 if (!vdev
->config
->get
) {
2119 dev_err(&vdev
->dev
, "%s failure: config access disabled\n",
2124 if (!virtnet_validate_features(vdev
))
2127 /* Find if host supports multiqueue virtio_net device */
2128 err
= virtio_cread_feature(vdev
, VIRTIO_NET_F_MQ
,
2129 struct virtio_net_config
,
2130 max_virtqueue_pairs
, &max_queue_pairs
);
2132 /* We need at least 2 queue's */
2133 if (err
|| max_queue_pairs
< VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN
||
2134 max_queue_pairs
> VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX
||
2135 !virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
))
2136 max_queue_pairs
= 1;
2138 /* Allocate ourselves a network device with room for our info */
2139 dev
= alloc_etherdev_mq(sizeof(struct virtnet_info
), max_queue_pairs
);
2143 /* Set up network device as normal. */
2144 dev
->priv_flags
|= IFF_UNICAST_FLT
| IFF_LIVE_ADDR_CHANGE
;
2145 dev
->netdev_ops
= &virtnet_netdev
;
2146 dev
->features
= NETIF_F_HIGHDMA
;
2148 dev
->ethtool_ops
= &virtnet_ethtool_ops
;
2149 SET_NETDEV_DEV(dev
, &vdev
->dev
);
2151 /* Do we support "hardware" checksums? */
2152 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CSUM
)) {
2153 /* This opens up the world of extra features. */
2154 dev
->hw_features
|= NETIF_F_HW_CSUM
| NETIF_F_SG
;
2156 dev
->features
|= NETIF_F_HW_CSUM
| NETIF_F_SG
;
2158 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GSO
)) {
2159 dev
->hw_features
|= NETIF_F_TSO
| NETIF_F_UFO
2160 | NETIF_F_TSO_ECN
| NETIF_F_TSO6
;
2162 /* Individual feature bits: what can host handle? */
2163 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_TSO4
))
2164 dev
->hw_features
|= NETIF_F_TSO
;
2165 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_TSO6
))
2166 dev
->hw_features
|= NETIF_F_TSO6
;
2167 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_ECN
))
2168 dev
->hw_features
|= NETIF_F_TSO_ECN
;
2169 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_UFO
))
2170 dev
->hw_features
|= NETIF_F_UFO
;
2172 dev
->features
|= NETIF_F_GSO_ROBUST
;
2175 dev
->features
|= dev
->hw_features
& (NETIF_F_ALL_TSO
|NETIF_F_UFO
);
2176 /* (!csum && gso) case will be fixed by register_netdev() */
2178 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_CSUM
))
2179 dev
->features
|= NETIF_F_RXCSUM
;
2181 dev
->vlan_features
= dev
->features
;
2183 /* MTU range: 68 - 65535 */
2184 dev
->min_mtu
= MIN_MTU
;
2185 dev
->max_mtu
= MAX_MTU
;
2187 /* Configuration may specify what MAC to use. Otherwise random. */
2188 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MAC
))
2189 virtio_cread_bytes(vdev
,
2190 offsetof(struct virtio_net_config
, mac
),
2191 dev
->dev_addr
, dev
->addr_len
);
2193 eth_hw_addr_random(dev
);
2195 /* Set up our device-specific information */
2196 vi
= netdev_priv(dev
);
2200 vi
->stats
= alloc_percpu(struct virtnet_stats
);
2202 if (vi
->stats
== NULL
)
2205 for_each_possible_cpu(i
) {
2206 struct virtnet_stats
*virtnet_stats
;
2207 virtnet_stats
= per_cpu_ptr(vi
->stats
, i
);
2208 u64_stats_init(&virtnet_stats
->tx_syncp
);
2209 u64_stats_init(&virtnet_stats
->rx_syncp
);
2212 INIT_WORK(&vi
->config_work
, virtnet_config_changed_work
);
2214 /* If we can receive ANY GSO packets, we must allocate large ones. */
2215 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_TSO4
) ||
2216 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_TSO6
) ||
2217 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_ECN
) ||
2218 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_UFO
))
2219 vi
->big_packets
= true;
2221 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MRG_RXBUF
))
2222 vi
->mergeable_rx_bufs
= true;
2224 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MRG_RXBUF
) ||
2225 virtio_has_feature(vdev
, VIRTIO_F_VERSION_1
))
2226 vi
->hdr_len
= sizeof(struct virtio_net_hdr_mrg_rxbuf
);
2228 vi
->hdr_len
= sizeof(struct virtio_net_hdr
);
2230 if (virtio_has_feature(vdev
, VIRTIO_F_ANY_LAYOUT
) ||
2231 virtio_has_feature(vdev
, VIRTIO_F_VERSION_1
))
2232 vi
->any_header_sg
= true;
2234 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
))
2237 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MTU
)) {
2238 mtu
= virtio_cread16(vdev
,
2239 offsetof(struct virtio_net_config
,
2241 if (mtu
< dev
->min_mtu
) {
2242 __virtio_clear_bit(vdev
, VIRTIO_NET_F_MTU
);
2249 if (vi
->any_header_sg
)
2250 dev
->needed_headroom
= vi
->hdr_len
;
2252 /* Enable multiqueue by default */
2253 if (num_online_cpus() >= max_queue_pairs
)
2254 vi
->curr_queue_pairs
= max_queue_pairs
;
2256 vi
->curr_queue_pairs
= num_online_cpus();
2257 vi
->max_queue_pairs
= max_queue_pairs
;
2259 /* Allocate/initialize the rx/tx queues, and invoke find_vqs */
2265 if (vi
->mergeable_rx_bufs
)
2266 dev
->sysfs_rx_queue_group
= &virtio_net_mrg_rx_group
;
2268 netif_set_real_num_tx_queues(dev
, vi
->curr_queue_pairs
);
2269 netif_set_real_num_rx_queues(dev
, vi
->curr_queue_pairs
);
2271 virtnet_init_settings(dev
);
2273 err
= register_netdev(dev
);
2275 pr_debug("virtio_net: registering device failed\n");
2279 virtio_device_ready(vdev
);
2281 err
= virtnet_cpu_notif_add(vi
);
2283 pr_debug("virtio_net: registering cpu notifier failed\n");
2284 goto free_unregister_netdev
;
2288 virtnet_set_queues(vi
, vi
->curr_queue_pairs
);
2291 /* Assume link up if device can't report link status,
2292 otherwise get link status from config. */
2293 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_STATUS
)) {
2294 netif_carrier_off(dev
);
2295 schedule_work(&vi
->config_work
);
2297 vi
->status
= VIRTIO_NET_S_LINK_UP
;
2298 netif_carrier_on(dev
);
2301 pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
2302 dev
->name
, max_queue_pairs
);
2306 free_unregister_netdev
:
2307 vi
->vdev
->config
->reset(vdev
);
2309 unregister_netdev(dev
);
2311 cancel_delayed_work_sync(&vi
->refill
);
2312 free_receive_page_frags(vi
);
2313 virtnet_del_vqs(vi
);
2315 free_percpu(vi
->stats
);
2321 static void remove_vq_common(struct virtnet_info
*vi
)
2323 vi
->vdev
->config
->reset(vi
->vdev
);
2325 /* Free unused buffers in both send and recv, if any. */
2326 free_unused_bufs(vi
);
2328 free_receive_bufs(vi
);
2330 free_receive_page_frags(vi
);
2332 virtnet_del_vqs(vi
);
2335 static void virtnet_remove(struct virtio_device
*vdev
)
2337 struct virtnet_info
*vi
= vdev
->priv
;
2339 virtnet_cpu_notif_remove(vi
);
2341 /* Make sure no work handler is accessing the device. */
2342 flush_work(&vi
->config_work
);
2344 unregister_netdev(vi
->dev
);
2346 remove_vq_common(vi
);
2348 free_percpu(vi
->stats
);
2349 free_netdev(vi
->dev
);
2352 #ifdef CONFIG_PM_SLEEP
2353 static int virtnet_freeze(struct virtio_device
*vdev
)
2355 struct virtnet_info
*vi
= vdev
->priv
;
2358 virtnet_cpu_notif_remove(vi
);
2360 /* Make sure no work handler is accessing the device */
2361 flush_work(&vi
->config_work
);
2363 netif_device_detach(vi
->dev
);
2364 cancel_delayed_work_sync(&vi
->refill
);
2366 if (netif_running(vi
->dev
)) {
2367 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
2368 napi_disable(&vi
->rq
[i
].napi
);
2371 remove_vq_common(vi
);
2376 static int virtnet_restore(struct virtio_device
*vdev
)
2378 struct virtnet_info
*vi
= vdev
->priv
;
2385 virtio_device_ready(vdev
);
2387 if (netif_running(vi
->dev
)) {
2388 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++)
2389 if (!try_fill_recv(vi
, &vi
->rq
[i
], GFP_KERNEL
))
2390 schedule_delayed_work(&vi
->refill
, 0);
2392 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
2393 virtnet_napi_enable(&vi
->rq
[i
]);
2396 netif_device_attach(vi
->dev
);
2399 virtnet_set_queues(vi
, vi
->curr_queue_pairs
);
2402 err
= virtnet_cpu_notif_add(vi
);
2410 static struct virtio_device_id id_table
[] = {
2411 { VIRTIO_ID_NET
, VIRTIO_DEV_ANY_ID
},
2415 #define VIRTNET_FEATURES \
2416 VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM, \
2418 VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6, \
2419 VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, \
2420 VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO, \
2421 VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ, \
2422 VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, \
2423 VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ, \
2424 VIRTIO_NET_F_CTRL_MAC_ADDR, \
2427 static unsigned int features
[] = {
2431 static unsigned int features_legacy
[] = {
2434 VIRTIO_F_ANY_LAYOUT
,
2437 static struct virtio_driver virtio_net_driver
= {
2438 .feature_table
= features
,
2439 .feature_table_size
= ARRAY_SIZE(features
),
2440 .feature_table_legacy
= features_legacy
,
2441 .feature_table_size_legacy
= ARRAY_SIZE(features_legacy
),
2442 .driver
.name
= KBUILD_MODNAME
,
2443 .driver
.owner
= THIS_MODULE
,
2444 .id_table
= id_table
,
2445 .probe
= virtnet_probe
,
2446 .remove
= virtnet_remove
,
2447 .config_changed
= virtnet_config_changed
,
2448 #ifdef CONFIG_PM_SLEEP
2449 .freeze
= virtnet_freeze
,
2450 .restore
= virtnet_restore
,
2454 static __init
int virtio_net_driver_init(void)
2458 ret
= cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN
, "AP_VIRT_NET_ONLINE",
2460 virtnet_cpu_down_prep
);
2463 virtionet_online
= ret
;
2464 ret
= cpuhp_setup_state_multi(CPUHP_VIRT_NET_DEAD
, "VIRT_NET_DEAD",
2465 NULL
, virtnet_cpu_dead
);
2469 ret
= register_virtio_driver(&virtio_net_driver
);
2474 cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD
);
2476 cpuhp_remove_multi_state(virtionet_online
);
2480 module_init(virtio_net_driver_init
);
2482 static __exit
void virtio_net_driver_exit(void)
2484 cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD
);
2485 cpuhp_remove_multi_state(virtionet_online
);
2486 unregister_virtio_driver(&virtio_net_driver
);
2488 module_exit(virtio_net_driver_exit
);
2490 MODULE_DEVICE_TABLE(virtio
, id_table
);
2491 MODULE_DESCRIPTION("Virtio network driver");
2492 MODULE_LICENSE("GPL");