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
))
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
))
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
,
587 ewma_pkt_len_add(&rq
->mrg_avg_pkt_len
, len
);
592 ewma_pkt_len_add(&rq
->mrg_avg_pkt_len
, len
);
593 if (unlikely(xdp_page
!= page
))
599 if (unlikely(xdp_page
!= page
))
600 __free_pages(xdp_page
, 0);
601 ewma_pkt_len_add(&rq
->mrg_avg_pkt_len
, len
);
607 truesize
= max(len
, mergeable_ctx_to_buf_truesize(ctx
));
608 head_skb
= page_to_skb(vi
, rq
, page
, offset
, len
, truesize
);
611 if (unlikely(!curr_skb
))
616 ctx
= (unsigned long)virtqueue_get_buf(rq
->vq
, &len
);
617 if (unlikely(!ctx
)) {
618 pr_debug("%s: rx error: %d buffers out of %d missing\n",
620 virtio16_to_cpu(vi
->vdev
,
622 dev
->stats
.rx_length_errors
++;
626 buf
= mergeable_ctx_to_buf_address(ctx
);
627 page
= virt_to_head_page(buf
);
629 num_skb_frags
= skb_shinfo(curr_skb
)->nr_frags
;
630 if (unlikely(num_skb_frags
== MAX_SKB_FRAGS
)) {
631 struct sk_buff
*nskb
= alloc_skb(0, GFP_ATOMIC
);
635 if (curr_skb
== head_skb
)
636 skb_shinfo(curr_skb
)->frag_list
= nskb
;
638 curr_skb
->next
= nskb
;
640 head_skb
->truesize
+= nskb
->truesize
;
643 truesize
= max(len
, mergeable_ctx_to_buf_truesize(ctx
));
644 if (curr_skb
!= head_skb
) {
645 head_skb
->data_len
+= len
;
646 head_skb
->len
+= len
;
647 head_skb
->truesize
+= truesize
;
649 offset
= buf
- page_address(page
);
650 if (skb_can_coalesce(curr_skb
, num_skb_frags
, page
, offset
)) {
652 skb_coalesce_rx_frag(curr_skb
, num_skb_frags
- 1,
655 skb_add_rx_frag(curr_skb
, num_skb_frags
, page
,
656 offset
, len
, truesize
);
660 ewma_pkt_len_add(&rq
->mrg_avg_pkt_len
, head_skb
->len
);
668 ctx
= (unsigned long)virtqueue_get_buf(rq
->vq
, &len
);
669 if (unlikely(!ctx
)) {
670 pr_debug("%s: rx error: %d buffers missing\n",
672 dev
->stats
.rx_length_errors
++;
675 page
= virt_to_head_page(mergeable_ctx_to_buf_address(ctx
));
679 dev
->stats
.rx_dropped
++;
680 dev_kfree_skb(head_skb
);
685 static void receive_buf(struct virtnet_info
*vi
, struct receive_queue
*rq
,
686 void *buf
, unsigned int len
)
688 struct net_device
*dev
= vi
->dev
;
689 struct virtnet_stats
*stats
= this_cpu_ptr(vi
->stats
);
691 struct virtio_net_hdr_mrg_rxbuf
*hdr
;
693 if (unlikely(len
< vi
->hdr_len
+ ETH_HLEN
)) {
694 pr_debug("%s: short packet %i\n", dev
->name
, len
);
695 dev
->stats
.rx_length_errors
++;
696 if (vi
->mergeable_rx_bufs
) {
697 unsigned long ctx
= (unsigned long)buf
;
698 void *base
= mergeable_ctx_to_buf_address(ctx
);
699 put_page(virt_to_head_page(base
));
700 } else if (vi
->big_packets
) {
708 if (vi
->mergeable_rx_bufs
)
709 skb
= receive_mergeable(dev
, vi
, rq
, (unsigned long)buf
, len
);
710 else if (vi
->big_packets
)
711 skb
= receive_big(dev
, vi
, rq
, buf
, len
);
713 skb
= receive_small(vi
, buf
, len
);
718 hdr
= skb_vnet_hdr(skb
);
720 u64_stats_update_begin(&stats
->rx_syncp
);
721 stats
->rx_bytes
+= skb
->len
;
723 u64_stats_update_end(&stats
->rx_syncp
);
725 if (hdr
->hdr
.flags
& VIRTIO_NET_HDR_F_DATA_VALID
)
726 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
728 if (virtio_net_hdr_to_skb(skb
, &hdr
->hdr
,
729 virtio_is_little_endian(vi
->vdev
))) {
730 net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n",
731 dev
->name
, hdr
->hdr
.gso_type
,
736 skb
->protocol
= eth_type_trans(skb
, dev
);
737 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
738 ntohs(skb
->protocol
), skb
->len
, skb
->pkt_type
);
740 napi_gro_receive(&rq
->napi
, skb
);
744 dev
->stats
.rx_frame_errors
++;
748 static int add_recvbuf_small(struct virtnet_info
*vi
, struct receive_queue
*rq
,
752 struct virtio_net_hdr_mrg_rxbuf
*hdr
;
755 skb
= __netdev_alloc_skb_ip_align(vi
->dev
, GOOD_PACKET_LEN
, gfp
);
759 skb_put(skb
, GOOD_PACKET_LEN
);
761 hdr
= skb_vnet_hdr(skb
);
762 sg_init_table(rq
->sg
, 2);
763 sg_set_buf(rq
->sg
, hdr
, vi
->hdr_len
);
764 skb_to_sgvec(skb
, rq
->sg
+ 1, 0, skb
->len
);
766 err
= virtqueue_add_inbuf(rq
->vq
, rq
->sg
, 2, skb
, gfp
);
773 static int add_recvbuf_big(struct virtnet_info
*vi
, struct receive_queue
*rq
,
776 struct page
*first
, *list
= NULL
;
780 sg_init_table(rq
->sg
, MAX_SKB_FRAGS
+ 2);
782 /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */
783 for (i
= MAX_SKB_FRAGS
+ 1; i
> 1; --i
) {
784 first
= get_a_page(rq
, gfp
);
787 give_pages(rq
, list
);
790 sg_set_buf(&rq
->sg
[i
], page_address(first
), PAGE_SIZE
);
792 /* chain new page in list head to match sg */
793 first
->private = (unsigned long)list
;
797 first
= get_a_page(rq
, gfp
);
799 give_pages(rq
, list
);
802 p
= page_address(first
);
804 /* rq->sg[0], rq->sg[1] share the same page */
805 /* a separated rq->sg[0] for header - required in case !any_header_sg */
806 sg_set_buf(&rq
->sg
[0], p
, vi
->hdr_len
);
808 /* rq->sg[1] for data packet, from offset */
809 offset
= sizeof(struct padded_vnet_hdr
);
810 sg_set_buf(&rq
->sg
[1], p
+ offset
, PAGE_SIZE
- offset
);
812 /* chain first in list head */
813 first
->private = (unsigned long)list
;
814 err
= virtqueue_add_inbuf(rq
->vq
, rq
->sg
, MAX_SKB_FRAGS
+ 2,
817 give_pages(rq
, first
);
822 static unsigned int get_mergeable_buf_len(struct ewma_pkt_len
*avg_pkt_len
)
824 const size_t hdr_len
= sizeof(struct virtio_net_hdr_mrg_rxbuf
);
827 len
= hdr_len
+ clamp_t(unsigned int, ewma_pkt_len_read(avg_pkt_len
),
828 GOOD_PACKET_LEN
, PAGE_SIZE
- hdr_len
);
829 return ALIGN(len
, MERGEABLE_BUFFER_ALIGN
);
832 static int add_recvbuf_mergeable(struct receive_queue
*rq
, gfp_t gfp
)
834 struct page_frag
*alloc_frag
= &rq
->alloc_frag
;
838 unsigned int len
, hole
;
840 len
= get_mergeable_buf_len(&rq
->mrg_avg_pkt_len
);
841 if (unlikely(!skb_page_frag_refill(len
, alloc_frag
, gfp
)))
844 buf
= (char *)page_address(alloc_frag
->page
) + alloc_frag
->offset
;
845 ctx
= mergeable_buf_to_ctx(buf
, len
);
846 get_page(alloc_frag
->page
);
847 alloc_frag
->offset
+= len
;
848 hole
= alloc_frag
->size
- alloc_frag
->offset
;
850 /* To avoid internal fragmentation, if there is very likely not
851 * enough space for another buffer, add the remaining space to
852 * the current buffer. This extra space is not included in
853 * the truesize stored in ctx.
856 alloc_frag
->offset
+= hole
;
859 sg_init_one(rq
->sg
, buf
, len
);
860 err
= virtqueue_add_inbuf(rq
->vq
, rq
->sg
, 1, (void *)ctx
, gfp
);
862 put_page(virt_to_head_page(buf
));
868 * Returns false if we couldn't fill entirely (OOM).
870 * Normally run in the receive path, but can also be run from ndo_open
871 * before we're receiving packets, or from refill_work which is
872 * careful to disable receiving (using napi_disable).
874 static bool try_fill_recv(struct virtnet_info
*vi
, struct receive_queue
*rq
,
882 if (vi
->mergeable_rx_bufs
)
883 err
= add_recvbuf_mergeable(rq
, gfp
);
884 else if (vi
->big_packets
)
885 err
= add_recvbuf_big(vi
, rq
, gfp
);
887 err
= add_recvbuf_small(vi
, rq
, gfp
);
889 oom
= err
== -ENOMEM
;
892 } while (rq
->vq
->num_free
);
893 virtqueue_kick(rq
->vq
);
897 static void skb_recv_done(struct virtqueue
*rvq
)
899 struct virtnet_info
*vi
= rvq
->vdev
->priv
;
900 struct receive_queue
*rq
= &vi
->rq
[vq2rxq(rvq
)];
902 /* Schedule NAPI, Suppress further interrupts if successful. */
903 if (napi_schedule_prep(&rq
->napi
)) {
904 virtqueue_disable_cb(rvq
);
905 __napi_schedule(&rq
->napi
);
909 static void virtnet_napi_enable(struct receive_queue
*rq
)
911 napi_enable(&rq
->napi
);
913 /* If all buffers were filled by other side before we napi_enabled, we
914 * won't get another interrupt, so process any outstanding packets
915 * now. virtnet_poll wants re-enable the queue, so we disable here.
916 * We synchronize against interrupts via NAPI_STATE_SCHED */
917 if (napi_schedule_prep(&rq
->napi
)) {
918 virtqueue_disable_cb(rq
->vq
);
920 __napi_schedule(&rq
->napi
);
925 static void refill_work(struct work_struct
*work
)
927 struct virtnet_info
*vi
=
928 container_of(work
, struct virtnet_info
, refill
.work
);
932 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++) {
933 struct receive_queue
*rq
= &vi
->rq
[i
];
935 napi_disable(&rq
->napi
);
936 still_empty
= !try_fill_recv(vi
, rq
, GFP_KERNEL
);
937 virtnet_napi_enable(rq
);
939 /* In theory, this can happen: if we don't get any buffers in
940 * we will *never* try to fill again.
943 schedule_delayed_work(&vi
->refill
, HZ
/2);
947 static int virtnet_receive(struct receive_queue
*rq
, int budget
)
949 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
950 unsigned int len
, received
= 0;
953 while (received
< budget
&&
954 (buf
= virtqueue_get_buf(rq
->vq
, &len
)) != NULL
) {
955 receive_buf(vi
, rq
, buf
, len
);
959 if (rq
->vq
->num_free
> virtqueue_get_vring_size(rq
->vq
) / 2) {
960 if (!try_fill_recv(vi
, rq
, GFP_ATOMIC
))
961 schedule_delayed_work(&vi
->refill
, 0);
967 static int virtnet_poll(struct napi_struct
*napi
, int budget
)
969 struct receive_queue
*rq
=
970 container_of(napi
, struct receive_queue
, napi
);
971 unsigned int r
, received
;
973 received
= virtnet_receive(rq
, budget
);
975 /* Out of packets? */
976 if (received
< budget
) {
977 r
= virtqueue_enable_cb_prepare(rq
->vq
);
978 napi_complete_done(napi
, received
);
979 if (unlikely(virtqueue_poll(rq
->vq
, r
)) &&
980 napi_schedule_prep(napi
)) {
981 virtqueue_disable_cb(rq
->vq
);
982 __napi_schedule(napi
);
989 #ifdef CONFIG_NET_RX_BUSY_POLL
990 /* must be called with local_bh_disable()d */
991 static int virtnet_busy_poll(struct napi_struct
*napi
)
993 struct receive_queue
*rq
=
994 container_of(napi
, struct receive_queue
, napi
);
995 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
996 int r
, received
= 0, budget
= 4;
998 if (!(vi
->status
& VIRTIO_NET_S_LINK_UP
))
999 return LL_FLUSH_FAILED
;
1001 if (!napi_schedule_prep(napi
))
1002 return LL_FLUSH_BUSY
;
1004 virtqueue_disable_cb(rq
->vq
);
1007 received
+= virtnet_receive(rq
, budget
);
1009 r
= virtqueue_enable_cb_prepare(rq
->vq
);
1010 clear_bit(NAPI_STATE_SCHED
, &napi
->state
);
1011 if (unlikely(virtqueue_poll(rq
->vq
, r
)) &&
1012 napi_schedule_prep(napi
)) {
1013 virtqueue_disable_cb(rq
->vq
);
1014 if (received
< budget
) {
1018 __napi_schedule(napi
);
1024 #endif /* CONFIG_NET_RX_BUSY_POLL */
1026 static int virtnet_open(struct net_device
*dev
)
1028 struct virtnet_info
*vi
= netdev_priv(dev
);
1031 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1032 if (i
< vi
->curr_queue_pairs
)
1033 /* Make sure we have some buffers: if oom use wq. */
1034 if (!try_fill_recv(vi
, &vi
->rq
[i
], GFP_KERNEL
))
1035 schedule_delayed_work(&vi
->refill
, 0);
1036 virtnet_napi_enable(&vi
->rq
[i
]);
1042 static void free_old_xmit_skbs(struct send_queue
*sq
)
1044 struct sk_buff
*skb
;
1046 struct virtnet_info
*vi
= sq
->vq
->vdev
->priv
;
1047 struct virtnet_stats
*stats
= this_cpu_ptr(vi
->stats
);
1049 while ((skb
= virtqueue_get_buf(sq
->vq
, &len
)) != NULL
) {
1050 pr_debug("Sent skb %p\n", skb
);
1052 u64_stats_update_begin(&stats
->tx_syncp
);
1053 stats
->tx_bytes
+= skb
->len
;
1054 stats
->tx_packets
++;
1055 u64_stats_update_end(&stats
->tx_syncp
);
1057 dev_kfree_skb_any(skb
);
1061 static int xmit_skb(struct send_queue
*sq
, struct sk_buff
*skb
)
1063 struct virtio_net_hdr_mrg_rxbuf
*hdr
;
1064 const unsigned char *dest
= ((struct ethhdr
*)skb
->data
)->h_dest
;
1065 struct virtnet_info
*vi
= sq
->vq
->vdev
->priv
;
1067 unsigned hdr_len
= vi
->hdr_len
;
1070 pr_debug("%s: xmit %p %pM\n", vi
->dev
->name
, skb
, dest
);
1072 can_push
= vi
->any_header_sg
&&
1073 !((unsigned long)skb
->data
& (__alignof__(*hdr
) - 1)) &&
1074 !skb_header_cloned(skb
) && skb_headroom(skb
) >= hdr_len
;
1075 /* Even if we can, don't push here yet as this would skew
1076 * csum_start offset below. */
1078 hdr
= (struct virtio_net_hdr_mrg_rxbuf
*)(skb
->data
- hdr_len
);
1080 hdr
= skb_vnet_hdr(skb
);
1082 if (virtio_net_hdr_from_skb(skb
, &hdr
->hdr
,
1083 virtio_is_little_endian(vi
->vdev
)))
1086 if (vi
->mergeable_rx_bufs
)
1087 hdr
->num_buffers
= 0;
1089 sg_init_table(sq
->sg
, skb_shinfo(skb
)->nr_frags
+ (can_push
? 1 : 2));
1091 __skb_push(skb
, hdr_len
);
1092 num_sg
= skb_to_sgvec(skb
, sq
->sg
, 0, skb
->len
);
1093 /* Pull header back to avoid skew in tx bytes calculations. */
1094 __skb_pull(skb
, hdr_len
);
1096 sg_set_buf(sq
->sg
, hdr
, hdr_len
);
1097 num_sg
= skb_to_sgvec(skb
, sq
->sg
+ 1, 0, skb
->len
) + 1;
1099 return virtqueue_add_outbuf(sq
->vq
, sq
->sg
, num_sg
, skb
, GFP_ATOMIC
);
1102 static netdev_tx_t
start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
1104 struct virtnet_info
*vi
= netdev_priv(dev
);
1105 int qnum
= skb_get_queue_mapping(skb
);
1106 struct send_queue
*sq
= &vi
->sq
[qnum
];
1108 struct netdev_queue
*txq
= netdev_get_tx_queue(dev
, qnum
);
1109 bool kick
= !skb
->xmit_more
;
1111 /* Free up any pending old buffers before queueing new ones. */
1112 free_old_xmit_skbs(sq
);
1114 /* timestamp packet in software */
1115 skb_tx_timestamp(skb
);
1117 /* Try to transmit */
1118 err
= xmit_skb(sq
, skb
);
1120 /* This should not happen! */
1121 if (unlikely(err
)) {
1122 dev
->stats
.tx_fifo_errors
++;
1123 if (net_ratelimit())
1125 "Unexpected TXQ (%d) queue failure: %d\n", qnum
, err
);
1126 dev
->stats
.tx_dropped
++;
1127 dev_kfree_skb_any(skb
);
1128 return NETDEV_TX_OK
;
1131 /* Don't wait up for transmitted skbs to be freed. */
1135 /* If running out of space, stop queue to avoid getting packets that we
1136 * are then unable to transmit.
1137 * An alternative would be to force queuing layer to requeue the skb by
1138 * returning NETDEV_TX_BUSY. However, NETDEV_TX_BUSY should not be
1139 * returned in a normal path of operation: it means that driver is not
1140 * maintaining the TX queue stop/start state properly, and causes
1141 * the stack to do a non-trivial amount of useless work.
1142 * Since most packets only take 1 or 2 ring slots, stopping the queue
1143 * early means 16 slots are typically wasted.
1145 if (sq
->vq
->num_free
< 2+MAX_SKB_FRAGS
) {
1146 netif_stop_subqueue(dev
, qnum
);
1147 if (unlikely(!virtqueue_enable_cb_delayed(sq
->vq
))) {
1148 /* More just got used, free them then recheck. */
1149 free_old_xmit_skbs(sq
);
1150 if (sq
->vq
->num_free
>= 2+MAX_SKB_FRAGS
) {
1151 netif_start_subqueue(dev
, qnum
);
1152 virtqueue_disable_cb(sq
->vq
);
1157 if (kick
|| netif_xmit_stopped(txq
))
1158 virtqueue_kick(sq
->vq
);
1160 return NETDEV_TX_OK
;
1164 * Send command via the control virtqueue and check status. Commands
1165 * supported by the hypervisor, as indicated by feature bits, should
1166 * never fail unless improperly formatted.
1168 static bool virtnet_send_command(struct virtnet_info
*vi
, u8
class, u8 cmd
,
1169 struct scatterlist
*out
)
1171 struct scatterlist
*sgs
[4], hdr
, stat
;
1172 unsigned out_num
= 0, tmp
;
1174 /* Caller should know better */
1175 BUG_ON(!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
));
1177 vi
->ctrl_status
= ~0;
1178 vi
->ctrl_hdr
.class = class;
1179 vi
->ctrl_hdr
.cmd
= cmd
;
1181 sg_init_one(&hdr
, &vi
->ctrl_hdr
, sizeof(vi
->ctrl_hdr
));
1182 sgs
[out_num
++] = &hdr
;
1185 sgs
[out_num
++] = out
;
1187 /* Add return status. */
1188 sg_init_one(&stat
, &vi
->ctrl_status
, sizeof(vi
->ctrl_status
));
1189 sgs
[out_num
] = &stat
;
1191 BUG_ON(out_num
+ 1 > ARRAY_SIZE(sgs
));
1192 virtqueue_add_sgs(vi
->cvq
, sgs
, out_num
, 1, vi
, GFP_ATOMIC
);
1194 if (unlikely(!virtqueue_kick(vi
->cvq
)))
1195 return vi
->ctrl_status
== VIRTIO_NET_OK
;
1197 /* Spin for a response, the kick causes an ioport write, trapping
1198 * into the hypervisor, so the request should be handled immediately.
1200 while (!virtqueue_get_buf(vi
->cvq
, &tmp
) &&
1201 !virtqueue_is_broken(vi
->cvq
))
1204 return vi
->ctrl_status
== VIRTIO_NET_OK
;
1207 static int virtnet_set_mac_address(struct net_device
*dev
, void *p
)
1209 struct virtnet_info
*vi
= netdev_priv(dev
);
1210 struct virtio_device
*vdev
= vi
->vdev
;
1212 struct sockaddr
*addr
;
1213 struct scatterlist sg
;
1215 addr
= kmalloc(sizeof(*addr
), GFP_KERNEL
);
1218 memcpy(addr
, p
, sizeof(*addr
));
1220 ret
= eth_prepare_mac_addr_change(dev
, addr
);
1224 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_MAC_ADDR
)) {
1225 sg_init_one(&sg
, addr
->sa_data
, dev
->addr_len
);
1226 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MAC
,
1227 VIRTIO_NET_CTRL_MAC_ADDR_SET
, &sg
)) {
1228 dev_warn(&vdev
->dev
,
1229 "Failed to set mac address by vq command.\n");
1233 } else if (virtio_has_feature(vdev
, VIRTIO_NET_F_MAC
) &&
1234 !virtio_has_feature(vdev
, VIRTIO_F_VERSION_1
)) {
1237 /* Naturally, this has an atomicity problem. */
1238 for (i
= 0; i
< dev
->addr_len
; i
++)
1239 virtio_cwrite8(vdev
,
1240 offsetof(struct virtio_net_config
, mac
) +
1241 i
, addr
->sa_data
[i
]);
1244 eth_commit_mac_addr_change(dev
, p
);
1252 static struct rtnl_link_stats64
*virtnet_stats(struct net_device
*dev
,
1253 struct rtnl_link_stats64
*tot
)
1255 struct virtnet_info
*vi
= netdev_priv(dev
);
1259 for_each_possible_cpu(cpu
) {
1260 struct virtnet_stats
*stats
= per_cpu_ptr(vi
->stats
, cpu
);
1261 u64 tpackets
, tbytes
, rpackets
, rbytes
;
1264 start
= u64_stats_fetch_begin_irq(&stats
->tx_syncp
);
1265 tpackets
= stats
->tx_packets
;
1266 tbytes
= stats
->tx_bytes
;
1267 } while (u64_stats_fetch_retry_irq(&stats
->tx_syncp
, start
));
1270 start
= u64_stats_fetch_begin_irq(&stats
->rx_syncp
);
1271 rpackets
= stats
->rx_packets
;
1272 rbytes
= stats
->rx_bytes
;
1273 } while (u64_stats_fetch_retry_irq(&stats
->rx_syncp
, start
));
1275 tot
->rx_packets
+= rpackets
;
1276 tot
->tx_packets
+= tpackets
;
1277 tot
->rx_bytes
+= rbytes
;
1278 tot
->tx_bytes
+= tbytes
;
1281 tot
->tx_dropped
= dev
->stats
.tx_dropped
;
1282 tot
->tx_fifo_errors
= dev
->stats
.tx_fifo_errors
;
1283 tot
->rx_dropped
= dev
->stats
.rx_dropped
;
1284 tot
->rx_length_errors
= dev
->stats
.rx_length_errors
;
1285 tot
->rx_frame_errors
= dev
->stats
.rx_frame_errors
;
1290 #ifdef CONFIG_NET_POLL_CONTROLLER
1291 static void virtnet_netpoll(struct net_device
*dev
)
1293 struct virtnet_info
*vi
= netdev_priv(dev
);
1296 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++)
1297 napi_schedule(&vi
->rq
[i
].napi
);
1301 static void virtnet_ack_link_announce(struct virtnet_info
*vi
)
1304 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_ANNOUNCE
,
1305 VIRTIO_NET_CTRL_ANNOUNCE_ACK
, NULL
))
1306 dev_warn(&vi
->dev
->dev
, "Failed to ack link announce.\n");
1310 static int virtnet_set_queues(struct virtnet_info
*vi
, u16 queue_pairs
)
1312 struct scatterlist sg
;
1313 struct net_device
*dev
= vi
->dev
;
1315 if (!vi
->has_cvq
|| !virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_MQ
))
1318 vi
->ctrl_mq
.virtqueue_pairs
= cpu_to_virtio16(vi
->vdev
, queue_pairs
);
1319 sg_init_one(&sg
, &vi
->ctrl_mq
, sizeof(vi
->ctrl_mq
));
1321 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MQ
,
1322 VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET
, &sg
)) {
1323 dev_warn(&dev
->dev
, "Fail to set num of queue pairs to %d\n",
1327 vi
->curr_queue_pairs
= queue_pairs
;
1328 /* virtnet_open() will refill when device is going to up. */
1329 if (dev
->flags
& IFF_UP
)
1330 schedule_delayed_work(&vi
->refill
, 0);
1336 static int virtnet_close(struct net_device
*dev
)
1338 struct virtnet_info
*vi
= netdev_priv(dev
);
1341 /* Make sure refill_work doesn't re-enable napi! */
1342 cancel_delayed_work_sync(&vi
->refill
);
1344 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
1345 napi_disable(&vi
->rq
[i
].napi
);
1350 static void virtnet_set_rx_mode(struct net_device
*dev
)
1352 struct virtnet_info
*vi
= netdev_priv(dev
);
1353 struct scatterlist sg
[2];
1354 struct virtio_net_ctrl_mac
*mac_data
;
1355 struct netdev_hw_addr
*ha
;
1361 /* We can't dynamically set ndo_set_rx_mode, so return gracefully */
1362 if (!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_RX
))
1365 vi
->ctrl_promisc
= ((dev
->flags
& IFF_PROMISC
) != 0);
1366 vi
->ctrl_allmulti
= ((dev
->flags
& IFF_ALLMULTI
) != 0);
1368 sg_init_one(sg
, &vi
->ctrl_promisc
, sizeof(vi
->ctrl_promisc
));
1370 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_RX
,
1371 VIRTIO_NET_CTRL_RX_PROMISC
, sg
))
1372 dev_warn(&dev
->dev
, "Failed to %sable promisc mode.\n",
1373 vi
->ctrl_promisc
? "en" : "dis");
1375 sg_init_one(sg
, &vi
->ctrl_allmulti
, sizeof(vi
->ctrl_allmulti
));
1377 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_RX
,
1378 VIRTIO_NET_CTRL_RX_ALLMULTI
, sg
))
1379 dev_warn(&dev
->dev
, "Failed to %sable allmulti mode.\n",
1380 vi
->ctrl_allmulti
? "en" : "dis");
1382 uc_count
= netdev_uc_count(dev
);
1383 mc_count
= netdev_mc_count(dev
);
1384 /* MAC filter - use one buffer for both lists */
1385 buf
= kzalloc(((uc_count
+ mc_count
) * ETH_ALEN
) +
1386 (2 * sizeof(mac_data
->entries
)), GFP_ATOMIC
);
1391 sg_init_table(sg
, 2);
1393 /* Store the unicast list and count in the front of the buffer */
1394 mac_data
->entries
= cpu_to_virtio32(vi
->vdev
, uc_count
);
1396 netdev_for_each_uc_addr(ha
, dev
)
1397 memcpy(&mac_data
->macs
[i
++][0], ha
->addr
, ETH_ALEN
);
1399 sg_set_buf(&sg
[0], mac_data
,
1400 sizeof(mac_data
->entries
) + (uc_count
* ETH_ALEN
));
1402 /* multicast list and count fill the end */
1403 mac_data
= (void *)&mac_data
->macs
[uc_count
][0];
1405 mac_data
->entries
= cpu_to_virtio32(vi
->vdev
, mc_count
);
1407 netdev_for_each_mc_addr(ha
, dev
)
1408 memcpy(&mac_data
->macs
[i
++][0], ha
->addr
, ETH_ALEN
);
1410 sg_set_buf(&sg
[1], mac_data
,
1411 sizeof(mac_data
->entries
) + (mc_count
* ETH_ALEN
));
1413 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MAC
,
1414 VIRTIO_NET_CTRL_MAC_TABLE_SET
, sg
))
1415 dev_warn(&dev
->dev
, "Failed to set MAC filter table.\n");
1420 static int virtnet_vlan_rx_add_vid(struct net_device
*dev
,
1421 __be16 proto
, u16 vid
)
1423 struct virtnet_info
*vi
= netdev_priv(dev
);
1424 struct scatterlist sg
;
1427 sg_init_one(&sg
, &vi
->ctrl_vid
, sizeof(vi
->ctrl_vid
));
1429 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_VLAN
,
1430 VIRTIO_NET_CTRL_VLAN_ADD
, &sg
))
1431 dev_warn(&dev
->dev
, "Failed to add VLAN ID %d.\n", vid
);
1435 static int virtnet_vlan_rx_kill_vid(struct net_device
*dev
,
1436 __be16 proto
, u16 vid
)
1438 struct virtnet_info
*vi
= netdev_priv(dev
);
1439 struct scatterlist sg
;
1442 sg_init_one(&sg
, &vi
->ctrl_vid
, sizeof(vi
->ctrl_vid
));
1444 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_VLAN
,
1445 VIRTIO_NET_CTRL_VLAN_DEL
, &sg
))
1446 dev_warn(&dev
->dev
, "Failed to kill VLAN ID %d.\n", vid
);
1450 static void virtnet_clean_affinity(struct virtnet_info
*vi
, long hcpu
)
1454 if (vi
->affinity_hint_set
) {
1455 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1456 virtqueue_set_affinity(vi
->rq
[i
].vq
, -1);
1457 virtqueue_set_affinity(vi
->sq
[i
].vq
, -1);
1460 vi
->affinity_hint_set
= false;
1464 static void virtnet_set_affinity(struct virtnet_info
*vi
)
1469 /* In multiqueue mode, when the number of cpu is equal to the number of
1470 * queue pairs, we let the queue pairs to be private to one cpu by
1471 * setting the affinity hint to eliminate the contention.
1473 if (vi
->curr_queue_pairs
== 1 ||
1474 vi
->max_queue_pairs
!= num_online_cpus()) {
1475 virtnet_clean_affinity(vi
, -1);
1480 for_each_online_cpu(cpu
) {
1481 virtqueue_set_affinity(vi
->rq
[i
].vq
, cpu
);
1482 virtqueue_set_affinity(vi
->sq
[i
].vq
, cpu
);
1483 netif_set_xps_queue(vi
->dev
, cpumask_of(cpu
), i
);
1487 vi
->affinity_hint_set
= true;
1490 static int virtnet_cpu_online(unsigned int cpu
, struct hlist_node
*node
)
1492 struct virtnet_info
*vi
= hlist_entry_safe(node
, struct virtnet_info
,
1494 virtnet_set_affinity(vi
);
1498 static int virtnet_cpu_dead(unsigned int cpu
, struct hlist_node
*node
)
1500 struct virtnet_info
*vi
= hlist_entry_safe(node
, struct virtnet_info
,
1502 virtnet_set_affinity(vi
);
1506 static int virtnet_cpu_down_prep(unsigned int cpu
, struct hlist_node
*node
)
1508 struct virtnet_info
*vi
= hlist_entry_safe(node
, struct virtnet_info
,
1511 virtnet_clean_affinity(vi
, cpu
);
1515 static enum cpuhp_state virtionet_online
;
1517 static int virtnet_cpu_notif_add(struct virtnet_info
*vi
)
1521 ret
= cpuhp_state_add_instance_nocalls(virtionet_online
, &vi
->node
);
1524 ret
= cpuhp_state_add_instance_nocalls(CPUHP_VIRT_NET_DEAD
,
1528 cpuhp_state_remove_instance_nocalls(virtionet_online
, &vi
->node
);
1532 static void virtnet_cpu_notif_remove(struct virtnet_info
*vi
)
1534 cpuhp_state_remove_instance_nocalls(virtionet_online
, &vi
->node
);
1535 cpuhp_state_remove_instance_nocalls(CPUHP_VIRT_NET_DEAD
,
1539 static void virtnet_get_ringparam(struct net_device
*dev
,
1540 struct ethtool_ringparam
*ring
)
1542 struct virtnet_info
*vi
= netdev_priv(dev
);
1544 ring
->rx_max_pending
= virtqueue_get_vring_size(vi
->rq
[0].vq
);
1545 ring
->tx_max_pending
= virtqueue_get_vring_size(vi
->sq
[0].vq
);
1546 ring
->rx_pending
= ring
->rx_max_pending
;
1547 ring
->tx_pending
= ring
->tx_max_pending
;
1551 static void virtnet_get_drvinfo(struct net_device
*dev
,
1552 struct ethtool_drvinfo
*info
)
1554 struct virtnet_info
*vi
= netdev_priv(dev
);
1555 struct virtio_device
*vdev
= vi
->vdev
;
1557 strlcpy(info
->driver
, KBUILD_MODNAME
, sizeof(info
->driver
));
1558 strlcpy(info
->version
, VIRTNET_DRIVER_VERSION
, sizeof(info
->version
));
1559 strlcpy(info
->bus_info
, virtio_bus_name(vdev
), sizeof(info
->bus_info
));
1563 /* TODO: Eliminate OOO packets during switching */
1564 static int virtnet_set_channels(struct net_device
*dev
,
1565 struct ethtool_channels
*channels
)
1567 struct virtnet_info
*vi
= netdev_priv(dev
);
1568 u16 queue_pairs
= channels
->combined_count
;
1571 /* We don't support separate rx/tx channels.
1572 * We don't allow setting 'other' channels.
1574 if (channels
->rx_count
|| channels
->tx_count
|| channels
->other_count
)
1577 if (queue_pairs
> vi
->max_queue_pairs
|| queue_pairs
== 0)
1580 /* For now we don't support modifying channels while XDP is loaded
1581 * also when XDP is loaded all RX queues have XDP programs so we only
1582 * need to check a single RX queue.
1584 if (vi
->rq
[0].xdp_prog
)
1588 err
= virtnet_set_queues(vi
, queue_pairs
);
1590 netif_set_real_num_tx_queues(dev
, queue_pairs
);
1591 netif_set_real_num_rx_queues(dev
, queue_pairs
);
1593 virtnet_set_affinity(vi
);
1600 static void virtnet_get_channels(struct net_device
*dev
,
1601 struct ethtool_channels
*channels
)
1603 struct virtnet_info
*vi
= netdev_priv(dev
);
1605 channels
->combined_count
= vi
->curr_queue_pairs
;
1606 channels
->max_combined
= vi
->max_queue_pairs
;
1607 channels
->max_other
= 0;
1608 channels
->rx_count
= 0;
1609 channels
->tx_count
= 0;
1610 channels
->other_count
= 0;
1613 /* Check if the user is trying to change anything besides speed/duplex */
1614 static bool virtnet_validate_ethtool_cmd(const struct ethtool_cmd
*cmd
)
1616 struct ethtool_cmd diff1
= *cmd
;
1617 struct ethtool_cmd diff2
= {};
1619 /* cmd is always set so we need to clear it, validate the port type
1620 * and also without autonegotiation we can ignore advertising
1622 ethtool_cmd_speed_set(&diff1
, 0);
1623 diff2
.port
= PORT_OTHER
;
1624 diff1
.advertising
= 0;
1628 return !memcmp(&diff1
, &diff2
, sizeof(diff1
));
1631 static int virtnet_set_settings(struct net_device
*dev
, struct ethtool_cmd
*cmd
)
1633 struct virtnet_info
*vi
= netdev_priv(dev
);
1636 speed
= ethtool_cmd_speed(cmd
);
1637 /* don't allow custom speed and duplex */
1638 if (!ethtool_validate_speed(speed
) ||
1639 !ethtool_validate_duplex(cmd
->duplex
) ||
1640 !virtnet_validate_ethtool_cmd(cmd
))
1643 vi
->duplex
= cmd
->duplex
;
1648 static int virtnet_get_settings(struct net_device
*dev
, struct ethtool_cmd
*cmd
)
1650 struct virtnet_info
*vi
= netdev_priv(dev
);
1652 ethtool_cmd_speed_set(cmd
, vi
->speed
);
1653 cmd
->duplex
= vi
->duplex
;
1654 cmd
->port
= PORT_OTHER
;
1659 static void virtnet_init_settings(struct net_device
*dev
)
1661 struct virtnet_info
*vi
= netdev_priv(dev
);
1663 vi
->speed
= SPEED_UNKNOWN
;
1664 vi
->duplex
= DUPLEX_UNKNOWN
;
1667 static const struct ethtool_ops virtnet_ethtool_ops
= {
1668 .get_drvinfo
= virtnet_get_drvinfo
,
1669 .get_link
= ethtool_op_get_link
,
1670 .get_ringparam
= virtnet_get_ringparam
,
1671 .set_channels
= virtnet_set_channels
,
1672 .get_channels
= virtnet_get_channels
,
1673 .get_ts_info
= ethtool_op_get_ts_info
,
1674 .get_settings
= virtnet_get_settings
,
1675 .set_settings
= virtnet_set_settings
,
1678 static int virtnet_xdp_set(struct net_device
*dev
, struct bpf_prog
*prog
)
1680 unsigned long int max_sz
= PAGE_SIZE
- sizeof(struct padded_vnet_hdr
);
1681 struct virtnet_info
*vi
= netdev_priv(dev
);
1682 struct bpf_prog
*old_prog
;
1683 u16 xdp_qp
= 0, curr_qp
;
1686 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_GUEST_TSO4
) ||
1687 virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_GUEST_TSO6
) ||
1688 virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_GUEST_ECN
) ||
1689 virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_GUEST_UFO
)) {
1690 netdev_warn(dev
, "can't set XDP while host is implementing LRO, disable LRO first\n");
1694 if (vi
->mergeable_rx_bufs
&& !vi
->any_header_sg
) {
1695 netdev_warn(dev
, "XDP expects header/data in single page, any_header_sg required\n");
1699 if (dev
->mtu
> max_sz
) {
1700 netdev_warn(dev
, "XDP requires MTU less than %lu\n", max_sz
);
1704 curr_qp
= vi
->curr_queue_pairs
- vi
->xdp_queue_pairs
;
1706 xdp_qp
= nr_cpu_ids
;
1708 /* XDP requires extra queues for XDP_TX */
1709 if (curr_qp
+ xdp_qp
> vi
->max_queue_pairs
) {
1710 netdev_warn(dev
, "request %i queues but max is %i\n",
1711 curr_qp
+ xdp_qp
, vi
->max_queue_pairs
);
1715 err
= virtnet_set_queues(vi
, curr_qp
+ xdp_qp
);
1717 dev_warn(&dev
->dev
, "XDP Device queue allocation failure.\n");
1722 prog
= bpf_prog_add(prog
, vi
->max_queue_pairs
- 1);
1724 virtnet_set_queues(vi
, curr_qp
);
1725 return PTR_ERR(prog
);
1729 vi
->xdp_queue_pairs
= xdp_qp
;
1730 netif_set_real_num_rx_queues(dev
, curr_qp
+ xdp_qp
);
1732 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1733 old_prog
= rtnl_dereference(vi
->rq
[i
].xdp_prog
);
1734 rcu_assign_pointer(vi
->rq
[i
].xdp_prog
, prog
);
1736 bpf_prog_put(old_prog
);
1742 static bool virtnet_xdp_query(struct net_device
*dev
)
1744 struct virtnet_info
*vi
= netdev_priv(dev
);
1747 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1748 if (vi
->rq
[i
].xdp_prog
)
1754 static int virtnet_xdp(struct net_device
*dev
, struct netdev_xdp
*xdp
)
1756 switch (xdp
->command
) {
1757 case XDP_SETUP_PROG
:
1758 return virtnet_xdp_set(dev
, xdp
->prog
);
1759 case XDP_QUERY_PROG
:
1760 xdp
->prog_attached
= virtnet_xdp_query(dev
);
1767 static const struct net_device_ops virtnet_netdev
= {
1768 .ndo_open
= virtnet_open
,
1769 .ndo_stop
= virtnet_close
,
1770 .ndo_start_xmit
= start_xmit
,
1771 .ndo_validate_addr
= eth_validate_addr
,
1772 .ndo_set_mac_address
= virtnet_set_mac_address
,
1773 .ndo_set_rx_mode
= virtnet_set_rx_mode
,
1774 .ndo_get_stats64
= virtnet_stats
,
1775 .ndo_vlan_rx_add_vid
= virtnet_vlan_rx_add_vid
,
1776 .ndo_vlan_rx_kill_vid
= virtnet_vlan_rx_kill_vid
,
1777 #ifdef CONFIG_NET_POLL_CONTROLLER
1778 .ndo_poll_controller
= virtnet_netpoll
,
1780 #ifdef CONFIG_NET_RX_BUSY_POLL
1781 .ndo_busy_poll
= virtnet_busy_poll
,
1783 .ndo_xdp
= virtnet_xdp
,
1786 static void virtnet_config_changed_work(struct work_struct
*work
)
1788 struct virtnet_info
*vi
=
1789 container_of(work
, struct virtnet_info
, config_work
);
1792 if (virtio_cread_feature(vi
->vdev
, VIRTIO_NET_F_STATUS
,
1793 struct virtio_net_config
, status
, &v
) < 0)
1796 if (v
& VIRTIO_NET_S_ANNOUNCE
) {
1797 netdev_notify_peers(vi
->dev
);
1798 virtnet_ack_link_announce(vi
);
1801 /* Ignore unknown (future) status bits */
1802 v
&= VIRTIO_NET_S_LINK_UP
;
1804 if (vi
->status
== v
)
1809 if (vi
->status
& VIRTIO_NET_S_LINK_UP
) {
1810 netif_carrier_on(vi
->dev
);
1811 netif_tx_wake_all_queues(vi
->dev
);
1813 netif_carrier_off(vi
->dev
);
1814 netif_tx_stop_all_queues(vi
->dev
);
1818 static void virtnet_config_changed(struct virtio_device
*vdev
)
1820 struct virtnet_info
*vi
= vdev
->priv
;
1822 schedule_work(&vi
->config_work
);
1825 static void virtnet_free_queues(struct virtnet_info
*vi
)
1829 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1830 napi_hash_del(&vi
->rq
[i
].napi
);
1831 netif_napi_del(&vi
->rq
[i
].napi
);
1834 /* We called napi_hash_del() before netif_napi_del(),
1835 * we need to respect an RCU grace period before freeing vi->rq
1843 static void free_receive_bufs(struct virtnet_info
*vi
)
1845 struct bpf_prog
*old_prog
;
1849 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1850 while (vi
->rq
[i
].pages
)
1851 __free_pages(get_a_page(&vi
->rq
[i
], GFP_KERNEL
), 0);
1853 old_prog
= rtnl_dereference(vi
->rq
[i
].xdp_prog
);
1854 RCU_INIT_POINTER(vi
->rq
[i
].xdp_prog
, NULL
);
1856 bpf_prog_put(old_prog
);
1861 static void free_receive_page_frags(struct virtnet_info
*vi
)
1864 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
1865 if (vi
->rq
[i
].alloc_frag
.page
)
1866 put_page(vi
->rq
[i
].alloc_frag
.page
);
1869 static bool is_xdp_queue(struct virtnet_info
*vi
, int q
)
1871 if (q
< (vi
->curr_queue_pairs
- vi
->xdp_queue_pairs
))
1873 else if (q
< vi
->curr_queue_pairs
)
1879 static void free_unused_bufs(struct virtnet_info
*vi
)
1884 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1885 struct virtqueue
*vq
= vi
->sq
[i
].vq
;
1886 while ((buf
= virtqueue_detach_unused_buf(vq
)) != NULL
) {
1887 if (!is_xdp_queue(vi
, i
))
1890 put_page(virt_to_head_page(buf
));
1894 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1895 struct virtqueue
*vq
= vi
->rq
[i
].vq
;
1897 while ((buf
= virtqueue_detach_unused_buf(vq
)) != NULL
) {
1898 if (vi
->mergeable_rx_bufs
) {
1899 unsigned long ctx
= (unsigned long)buf
;
1900 void *base
= mergeable_ctx_to_buf_address(ctx
);
1901 put_page(virt_to_head_page(base
));
1902 } else if (vi
->big_packets
) {
1903 give_pages(&vi
->rq
[i
], buf
);
1911 static void virtnet_del_vqs(struct virtnet_info
*vi
)
1913 struct virtio_device
*vdev
= vi
->vdev
;
1915 virtnet_clean_affinity(vi
, -1);
1917 vdev
->config
->del_vqs(vdev
);
1919 virtnet_free_queues(vi
);
1922 static int virtnet_find_vqs(struct virtnet_info
*vi
)
1924 vq_callback_t
**callbacks
;
1925 struct virtqueue
**vqs
;
1930 /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
1931 * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
1932 * possible control vq.
1934 total_vqs
= vi
->max_queue_pairs
* 2 +
1935 virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
);
1937 /* Allocate space for find_vqs parameters */
1938 vqs
= kzalloc(total_vqs
* sizeof(*vqs
), GFP_KERNEL
);
1941 callbacks
= kmalloc(total_vqs
* sizeof(*callbacks
), GFP_KERNEL
);
1944 names
= kmalloc(total_vqs
* sizeof(*names
), GFP_KERNEL
);
1948 /* Parameters for control virtqueue, if any */
1950 callbacks
[total_vqs
- 1] = NULL
;
1951 names
[total_vqs
- 1] = "control";
1954 /* Allocate/initialize parameters for send/receive virtqueues */
1955 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1956 callbacks
[rxq2vq(i
)] = skb_recv_done
;
1957 callbacks
[txq2vq(i
)] = skb_xmit_done
;
1958 sprintf(vi
->rq
[i
].name
, "input.%d", i
);
1959 sprintf(vi
->sq
[i
].name
, "output.%d", i
);
1960 names
[rxq2vq(i
)] = vi
->rq
[i
].name
;
1961 names
[txq2vq(i
)] = vi
->sq
[i
].name
;
1964 ret
= vi
->vdev
->config
->find_vqs(vi
->vdev
, total_vqs
, vqs
, callbacks
,
1970 vi
->cvq
= vqs
[total_vqs
- 1];
1971 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VLAN
))
1972 vi
->dev
->features
|= NETIF_F_HW_VLAN_CTAG_FILTER
;
1975 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1976 vi
->rq
[i
].vq
= vqs
[rxq2vq(i
)];
1977 vi
->sq
[i
].vq
= vqs
[txq2vq(i
)];
1996 static int virtnet_alloc_queues(struct virtnet_info
*vi
)
2000 vi
->sq
= kzalloc(sizeof(*vi
->sq
) * vi
->max_queue_pairs
, GFP_KERNEL
);
2003 vi
->rq
= kzalloc(sizeof(*vi
->rq
) * vi
->max_queue_pairs
, GFP_KERNEL
);
2007 INIT_DELAYED_WORK(&vi
->refill
, refill_work
);
2008 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
2009 vi
->rq
[i
].pages
= NULL
;
2010 netif_napi_add(vi
->dev
, &vi
->rq
[i
].napi
, virtnet_poll
,
2013 sg_init_table(vi
->rq
[i
].sg
, ARRAY_SIZE(vi
->rq
[i
].sg
));
2014 ewma_pkt_len_init(&vi
->rq
[i
].mrg_avg_pkt_len
);
2015 sg_init_table(vi
->sq
[i
].sg
, ARRAY_SIZE(vi
->sq
[i
].sg
));
2026 static int init_vqs(struct virtnet_info
*vi
)
2030 /* Allocate send & receive queues */
2031 ret
= virtnet_alloc_queues(vi
);
2035 ret
= virtnet_find_vqs(vi
);
2040 virtnet_set_affinity(vi
);
2046 virtnet_free_queues(vi
);
2052 static ssize_t
mergeable_rx_buffer_size_show(struct netdev_rx_queue
*queue
,
2053 struct rx_queue_attribute
*attribute
, char *buf
)
2055 struct virtnet_info
*vi
= netdev_priv(queue
->dev
);
2056 unsigned int queue_index
= get_netdev_rx_queue_index(queue
);
2057 struct ewma_pkt_len
*avg
;
2059 BUG_ON(queue_index
>= vi
->max_queue_pairs
);
2060 avg
= &vi
->rq
[queue_index
].mrg_avg_pkt_len
;
2061 return sprintf(buf
, "%u\n", get_mergeable_buf_len(avg
));
2064 static struct rx_queue_attribute mergeable_rx_buffer_size_attribute
=
2065 __ATTR_RO(mergeable_rx_buffer_size
);
2067 static struct attribute
*virtio_net_mrg_rx_attrs
[] = {
2068 &mergeable_rx_buffer_size_attribute
.attr
,
2072 static const struct attribute_group virtio_net_mrg_rx_group
= {
2073 .name
= "virtio_net",
2074 .attrs
= virtio_net_mrg_rx_attrs
2078 static bool virtnet_fail_on_feature(struct virtio_device
*vdev
,
2080 const char *fname
, const char *dname
)
2082 if (!virtio_has_feature(vdev
, fbit
))
2085 dev_err(&vdev
->dev
, "device advertises feature %s but not %s",
2091 #define VIRTNET_FAIL_ON(vdev, fbit, dbit) \
2092 virtnet_fail_on_feature(vdev, fbit, #fbit, dbit)
2094 static bool virtnet_validate_features(struct virtio_device
*vdev
)
2096 if (!virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
) &&
2097 (VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_CTRL_RX
,
2098 "VIRTIO_NET_F_CTRL_VQ") ||
2099 VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_CTRL_VLAN
,
2100 "VIRTIO_NET_F_CTRL_VQ") ||
2101 VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_GUEST_ANNOUNCE
,
2102 "VIRTIO_NET_F_CTRL_VQ") ||
2103 VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_MQ
, "VIRTIO_NET_F_CTRL_VQ") ||
2104 VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_CTRL_MAC_ADDR
,
2105 "VIRTIO_NET_F_CTRL_VQ"))) {
2112 #define MIN_MTU ETH_MIN_MTU
2113 #define MAX_MTU ETH_MAX_MTU
2115 static int virtnet_probe(struct virtio_device
*vdev
)
2118 struct net_device
*dev
;
2119 struct virtnet_info
*vi
;
2120 u16 max_queue_pairs
;
2123 if (!vdev
->config
->get
) {
2124 dev_err(&vdev
->dev
, "%s failure: config access disabled\n",
2129 if (!virtnet_validate_features(vdev
))
2132 /* Find if host supports multiqueue virtio_net device */
2133 err
= virtio_cread_feature(vdev
, VIRTIO_NET_F_MQ
,
2134 struct virtio_net_config
,
2135 max_virtqueue_pairs
, &max_queue_pairs
);
2137 /* We need at least 2 queue's */
2138 if (err
|| max_queue_pairs
< VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN
||
2139 max_queue_pairs
> VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX
||
2140 !virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
))
2141 max_queue_pairs
= 1;
2143 /* Allocate ourselves a network device with room for our info */
2144 dev
= alloc_etherdev_mq(sizeof(struct virtnet_info
), max_queue_pairs
);
2148 /* Set up network device as normal. */
2149 dev
->priv_flags
|= IFF_UNICAST_FLT
| IFF_LIVE_ADDR_CHANGE
;
2150 dev
->netdev_ops
= &virtnet_netdev
;
2151 dev
->features
= NETIF_F_HIGHDMA
;
2153 dev
->ethtool_ops
= &virtnet_ethtool_ops
;
2154 SET_NETDEV_DEV(dev
, &vdev
->dev
);
2156 /* Do we support "hardware" checksums? */
2157 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CSUM
)) {
2158 /* This opens up the world of extra features. */
2159 dev
->hw_features
|= NETIF_F_HW_CSUM
| NETIF_F_SG
;
2161 dev
->features
|= NETIF_F_HW_CSUM
| NETIF_F_SG
;
2163 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GSO
)) {
2164 dev
->hw_features
|= NETIF_F_TSO
| NETIF_F_UFO
2165 | NETIF_F_TSO_ECN
| NETIF_F_TSO6
;
2167 /* Individual feature bits: what can host handle? */
2168 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_TSO4
))
2169 dev
->hw_features
|= NETIF_F_TSO
;
2170 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_TSO6
))
2171 dev
->hw_features
|= NETIF_F_TSO6
;
2172 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_ECN
))
2173 dev
->hw_features
|= NETIF_F_TSO_ECN
;
2174 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_UFO
))
2175 dev
->hw_features
|= NETIF_F_UFO
;
2177 dev
->features
|= NETIF_F_GSO_ROBUST
;
2180 dev
->features
|= dev
->hw_features
& (NETIF_F_ALL_TSO
|NETIF_F_UFO
);
2181 /* (!csum && gso) case will be fixed by register_netdev() */
2183 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_CSUM
))
2184 dev
->features
|= NETIF_F_RXCSUM
;
2186 dev
->vlan_features
= dev
->features
;
2188 /* MTU range: 68 - 65535 */
2189 dev
->min_mtu
= MIN_MTU
;
2190 dev
->max_mtu
= MAX_MTU
;
2192 /* Configuration may specify what MAC to use. Otherwise random. */
2193 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MAC
))
2194 virtio_cread_bytes(vdev
,
2195 offsetof(struct virtio_net_config
, mac
),
2196 dev
->dev_addr
, dev
->addr_len
);
2198 eth_hw_addr_random(dev
);
2200 /* Set up our device-specific information */
2201 vi
= netdev_priv(dev
);
2205 vi
->stats
= alloc_percpu(struct virtnet_stats
);
2207 if (vi
->stats
== NULL
)
2210 for_each_possible_cpu(i
) {
2211 struct virtnet_stats
*virtnet_stats
;
2212 virtnet_stats
= per_cpu_ptr(vi
->stats
, i
);
2213 u64_stats_init(&virtnet_stats
->tx_syncp
);
2214 u64_stats_init(&virtnet_stats
->rx_syncp
);
2217 INIT_WORK(&vi
->config_work
, virtnet_config_changed_work
);
2219 /* If we can receive ANY GSO packets, we must allocate large ones. */
2220 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_TSO4
) ||
2221 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_TSO6
) ||
2222 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_ECN
) ||
2223 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_UFO
))
2224 vi
->big_packets
= true;
2226 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MRG_RXBUF
))
2227 vi
->mergeable_rx_bufs
= true;
2229 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MRG_RXBUF
) ||
2230 virtio_has_feature(vdev
, VIRTIO_F_VERSION_1
))
2231 vi
->hdr_len
= sizeof(struct virtio_net_hdr_mrg_rxbuf
);
2233 vi
->hdr_len
= sizeof(struct virtio_net_hdr
);
2235 if (virtio_has_feature(vdev
, VIRTIO_F_ANY_LAYOUT
) ||
2236 virtio_has_feature(vdev
, VIRTIO_F_VERSION_1
))
2237 vi
->any_header_sg
= true;
2239 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
))
2242 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MTU
)) {
2243 mtu
= virtio_cread16(vdev
,
2244 offsetof(struct virtio_net_config
,
2246 if (mtu
< dev
->min_mtu
) {
2247 __virtio_clear_bit(vdev
, VIRTIO_NET_F_MTU
);
2254 if (vi
->any_header_sg
)
2255 dev
->needed_headroom
= vi
->hdr_len
;
2257 /* Enable multiqueue by default */
2258 if (num_online_cpus() >= max_queue_pairs
)
2259 vi
->curr_queue_pairs
= max_queue_pairs
;
2261 vi
->curr_queue_pairs
= num_online_cpus();
2262 vi
->max_queue_pairs
= max_queue_pairs
;
2264 /* Allocate/initialize the rx/tx queues, and invoke find_vqs */
2270 if (vi
->mergeable_rx_bufs
)
2271 dev
->sysfs_rx_queue_group
= &virtio_net_mrg_rx_group
;
2273 netif_set_real_num_tx_queues(dev
, vi
->curr_queue_pairs
);
2274 netif_set_real_num_rx_queues(dev
, vi
->curr_queue_pairs
);
2276 virtnet_init_settings(dev
);
2278 err
= register_netdev(dev
);
2280 pr_debug("virtio_net: registering device failed\n");
2284 virtio_device_ready(vdev
);
2286 err
= virtnet_cpu_notif_add(vi
);
2288 pr_debug("virtio_net: registering cpu notifier failed\n");
2289 goto free_unregister_netdev
;
2293 virtnet_set_queues(vi
, vi
->curr_queue_pairs
);
2296 /* Assume link up if device can't report link status,
2297 otherwise get link status from config. */
2298 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_STATUS
)) {
2299 netif_carrier_off(dev
);
2300 schedule_work(&vi
->config_work
);
2302 vi
->status
= VIRTIO_NET_S_LINK_UP
;
2303 netif_carrier_on(dev
);
2306 pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
2307 dev
->name
, max_queue_pairs
);
2311 free_unregister_netdev
:
2312 vi
->vdev
->config
->reset(vdev
);
2314 unregister_netdev(dev
);
2316 cancel_delayed_work_sync(&vi
->refill
);
2317 free_receive_page_frags(vi
);
2318 virtnet_del_vqs(vi
);
2320 free_percpu(vi
->stats
);
2326 static void remove_vq_common(struct virtnet_info
*vi
)
2328 vi
->vdev
->config
->reset(vi
->vdev
);
2330 /* Free unused buffers in both send and recv, if any. */
2331 free_unused_bufs(vi
);
2333 free_receive_bufs(vi
);
2335 free_receive_page_frags(vi
);
2337 virtnet_del_vqs(vi
);
2340 static void virtnet_remove(struct virtio_device
*vdev
)
2342 struct virtnet_info
*vi
= vdev
->priv
;
2344 virtnet_cpu_notif_remove(vi
);
2346 /* Make sure no work handler is accessing the device. */
2347 flush_work(&vi
->config_work
);
2349 unregister_netdev(vi
->dev
);
2351 remove_vq_common(vi
);
2353 free_percpu(vi
->stats
);
2354 free_netdev(vi
->dev
);
2357 #ifdef CONFIG_PM_SLEEP
2358 static int virtnet_freeze(struct virtio_device
*vdev
)
2360 struct virtnet_info
*vi
= vdev
->priv
;
2363 virtnet_cpu_notif_remove(vi
);
2365 /* Make sure no work handler is accessing the device */
2366 flush_work(&vi
->config_work
);
2368 netif_device_detach(vi
->dev
);
2369 cancel_delayed_work_sync(&vi
->refill
);
2371 if (netif_running(vi
->dev
)) {
2372 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
2373 napi_disable(&vi
->rq
[i
].napi
);
2376 remove_vq_common(vi
);
2381 static int virtnet_restore(struct virtio_device
*vdev
)
2383 struct virtnet_info
*vi
= vdev
->priv
;
2390 virtio_device_ready(vdev
);
2392 if (netif_running(vi
->dev
)) {
2393 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++)
2394 if (!try_fill_recv(vi
, &vi
->rq
[i
], GFP_KERNEL
))
2395 schedule_delayed_work(&vi
->refill
, 0);
2397 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
2398 virtnet_napi_enable(&vi
->rq
[i
]);
2401 netif_device_attach(vi
->dev
);
2404 virtnet_set_queues(vi
, vi
->curr_queue_pairs
);
2407 err
= virtnet_cpu_notif_add(vi
);
2415 static struct virtio_device_id id_table
[] = {
2416 { VIRTIO_ID_NET
, VIRTIO_DEV_ANY_ID
},
2420 #define VIRTNET_FEATURES \
2421 VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM, \
2423 VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6, \
2424 VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, \
2425 VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO, \
2426 VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ, \
2427 VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, \
2428 VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ, \
2429 VIRTIO_NET_F_CTRL_MAC_ADDR, \
2432 static unsigned int features
[] = {
2436 static unsigned int features_legacy
[] = {
2439 VIRTIO_F_ANY_LAYOUT
,
2442 static struct virtio_driver virtio_net_driver
= {
2443 .feature_table
= features
,
2444 .feature_table_size
= ARRAY_SIZE(features
),
2445 .feature_table_legacy
= features_legacy
,
2446 .feature_table_size_legacy
= ARRAY_SIZE(features_legacy
),
2447 .driver
.name
= KBUILD_MODNAME
,
2448 .driver
.owner
= THIS_MODULE
,
2449 .id_table
= id_table
,
2450 .probe
= virtnet_probe
,
2451 .remove
= virtnet_remove
,
2452 .config_changed
= virtnet_config_changed
,
2453 #ifdef CONFIG_PM_SLEEP
2454 .freeze
= virtnet_freeze
,
2455 .restore
= virtnet_restore
,
2459 static __init
int virtio_net_driver_init(void)
2463 ret
= cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN
, "AP_VIRT_NET_ONLINE",
2465 virtnet_cpu_down_prep
);
2468 virtionet_online
= ret
;
2469 ret
= cpuhp_setup_state_multi(CPUHP_VIRT_NET_DEAD
, "VIRT_NET_DEAD",
2470 NULL
, virtnet_cpu_dead
);
2474 ret
= register_virtio_driver(&virtio_net_driver
);
2479 cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD
);
2481 cpuhp_remove_multi_state(virtionet_online
);
2485 module_init(virtio_net_driver_init
);
2487 static __exit
void virtio_net_driver_exit(void)
2489 cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD
);
2490 cpuhp_remove_multi_state(virtionet_online
);
2491 unregister_virtio_driver(&virtio_net_driver
);
2493 module_exit(virtio_net_driver_exit
);
2495 MODULE_DEVICE_TABLE(virtio
, id_table
);
2496 MODULE_DESCRIPTION("Virtio network driver");
2497 MODULE_LICENSE("GPL");