1 /* A network driver using virtio.
3 * Copyright 2007 Rusty Russell <rusty@rustcorp.com.au> IBM Corporation
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, see <http://www.gnu.org/licenses/>.
19 #include <linux/netdevice.h>
20 #include <linux/etherdevice.h>
21 #include <linux/ethtool.h>
22 #include <linux/module.h>
23 #include <linux/virtio.h>
24 #include <linux/virtio_net.h>
25 #include <linux/bpf.h>
26 #include <linux/bpf_trace.h>
27 #include <linux/scatterlist.h>
28 #include <linux/if_vlan.h>
29 #include <linux/slab.h>
30 #include <linux/cpu.h>
31 #include <linux/average.h>
32 #include <net/route.h>
34 static int napi_weight
= NAPI_POLL_WEIGHT
;
35 module_param(napi_weight
, int, 0444);
37 static bool csum
= true, gso
= true;
38 module_param(csum
, bool, 0444);
39 module_param(gso
, bool, 0444);
41 /* FIXME: MTU in config. */
42 #define GOOD_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
43 #define GOOD_COPY_LEN 128
45 #define VIRTNET_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
47 /* Amount of XDP headroom to prepend to packets for use by xdp_adjust_head */
48 #define VIRTIO_XDP_HEADROOM 256
50 /* RX packet size EWMA. The average packet size is used to determine the packet
51 * buffer size when refilling RX rings. As the entire RX ring may be refilled
52 * at once, the weight is chosen so that the EWMA will be insensitive to short-
53 * term, transient changes in packet size.
55 DECLARE_EWMA(pkt_len
, 0, 64)
57 #define VIRTNET_DRIVER_VERSION "1.0.0"
59 struct virtnet_stats
{
60 struct u64_stats_sync tx_syncp
;
61 struct u64_stats_sync rx_syncp
;
69 /* Internal representation of a send virtqueue */
71 /* Virtqueue associated with this send _queue */
74 /* TX: fragments + linear part + virtio header */
75 struct scatterlist sg
[MAX_SKB_FRAGS
+ 2];
77 /* Name of the send queue: output.$index */
81 /* Internal representation of a receive virtqueue */
82 struct receive_queue
{
83 /* Virtqueue associated with this receive_queue */
86 struct napi_struct napi
;
88 struct bpf_prog __rcu
*xdp_prog
;
90 /* Chain pages by the private ptr. */
93 /* Average packet length for mergeable receive buffers. */
94 struct ewma_pkt_len mrg_avg_pkt_len
;
96 /* Page frag for packet buffer allocation. */
97 struct page_frag alloc_frag
;
99 /* RX: fragments + linear part + virtio header */
100 struct scatterlist sg
[MAX_SKB_FRAGS
+ 2];
102 /* Min single buffer size for mergeable buffers case. */
103 unsigned int min_buf_len
;
105 /* Name of this receive queue: input.$index */
109 struct virtnet_info
{
110 struct virtio_device
*vdev
;
111 struct virtqueue
*cvq
;
112 struct net_device
*dev
;
113 struct send_queue
*sq
;
114 struct receive_queue
*rq
;
117 /* Max # of queue pairs supported by the device */
120 /* # of queue pairs currently used by the driver */
121 u16 curr_queue_pairs
;
123 /* # of XDP queue pairs currently used by the driver */
126 /* I like... big packets and I cannot lie! */
129 /* Host will merge rx buffers for big packets (shake it! shake it!) */
130 bool mergeable_rx_bufs
;
132 /* Has control virtqueue */
135 /* Host can handle any s/g split between our header and packet data */
138 /* Packet virtio header size */
141 /* Active statistics */
142 struct virtnet_stats __percpu
*stats
;
144 /* Work struct for refilling if we run low on memory. */
145 struct delayed_work refill
;
147 /* Work struct for config space updates */
148 struct work_struct config_work
;
150 /* Does the affinity hint is set for virtqueues? */
151 bool affinity_hint_set
;
153 /* CPU hotplug instances for online & dead */
154 struct hlist_node node
;
155 struct hlist_node node_dead
;
157 /* Control VQ buffers: protected by the rtnl lock */
158 struct virtio_net_ctrl_hdr ctrl_hdr
;
159 virtio_net_ctrl_ack ctrl_status
;
160 struct virtio_net_ctrl_mq ctrl_mq
;
165 /* Ethtool settings */
170 struct padded_vnet_hdr
{
171 struct virtio_net_hdr_mrg_rxbuf hdr
;
173 * hdr is in a separate sg buffer, and data sg buffer shares same page
174 * with this header sg. This padding makes next sg 16 byte aligned
180 /* Converting between virtqueue no. and kernel tx/rx queue no.
181 * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq
183 static int vq2txq(struct virtqueue
*vq
)
185 return (vq
->index
- 1) / 2;
188 static int txq2vq(int txq
)
193 static int vq2rxq(struct virtqueue
*vq
)
195 return vq
->index
/ 2;
198 static int rxq2vq(int rxq
)
203 static inline struct virtio_net_hdr_mrg_rxbuf
*skb_vnet_hdr(struct sk_buff
*skb
)
205 return (struct virtio_net_hdr_mrg_rxbuf
*)skb
->cb
;
209 * private is used to chain pages for big packets, put the whole
210 * most recent used list in the beginning for reuse
212 static void give_pages(struct receive_queue
*rq
, struct page
*page
)
216 /* Find end of list, sew whole thing into vi->rq.pages. */
217 for (end
= page
; end
->private; end
= (struct page
*)end
->private);
218 end
->private = (unsigned long)rq
->pages
;
222 static struct page
*get_a_page(struct receive_queue
*rq
, gfp_t gfp_mask
)
224 struct page
*p
= rq
->pages
;
227 rq
->pages
= (struct page
*)p
->private;
228 /* clear private here, it is used to chain pages */
231 p
= alloc_page(gfp_mask
);
235 static void skb_xmit_done(struct virtqueue
*vq
)
237 struct virtnet_info
*vi
= vq
->vdev
->priv
;
239 /* Suppress further interrupts. */
240 virtqueue_disable_cb(vq
);
242 /* We were probably waiting for more output buffers. */
243 netif_wake_subqueue(vi
->dev
, vq2txq(vq
));
246 /* Called from bottom half context */
247 static struct sk_buff
*page_to_skb(struct virtnet_info
*vi
,
248 struct receive_queue
*rq
,
249 struct page
*page
, unsigned int offset
,
250 unsigned int len
, unsigned int truesize
)
253 struct virtio_net_hdr_mrg_rxbuf
*hdr
;
254 unsigned int copy
, hdr_len
, hdr_padded_len
;
257 p
= page_address(page
) + offset
;
259 /* copy small packet so we can reuse these pages for small data */
260 skb
= napi_alloc_skb(&rq
->napi
, GOOD_COPY_LEN
);
264 hdr
= skb_vnet_hdr(skb
);
266 hdr_len
= vi
->hdr_len
;
267 if (vi
->mergeable_rx_bufs
)
268 hdr_padded_len
= sizeof *hdr
;
270 hdr_padded_len
= sizeof(struct padded_vnet_hdr
);
272 memcpy(hdr
, p
, hdr_len
);
275 offset
+= hdr_padded_len
;
279 if (copy
> skb_tailroom(skb
))
280 copy
= skb_tailroom(skb
);
281 memcpy(skb_put(skb
, copy
), p
, copy
);
286 if (vi
->mergeable_rx_bufs
) {
288 skb_add_rx_frag(skb
, 0, page
, offset
, len
, truesize
);
295 * Verify that we can indeed put this data into a skb.
296 * This is here to handle cases when the device erroneously
297 * tries to receive more than is possible. This is usually
298 * the case of a broken device.
300 if (unlikely(len
> MAX_SKB_FRAGS
* PAGE_SIZE
)) {
301 net_dbg_ratelimited("%s: too much data\n", skb
->dev
->name
);
305 BUG_ON(offset
>= PAGE_SIZE
);
307 unsigned int frag_size
= min((unsigned)PAGE_SIZE
- offset
, len
);
308 skb_add_rx_frag(skb
, skb_shinfo(skb
)->nr_frags
, page
, offset
,
309 frag_size
, truesize
);
311 page
= (struct page
*)page
->private;
316 give_pages(rq
, page
);
321 static bool virtnet_xdp_xmit(struct virtnet_info
*vi
,
322 struct receive_queue
*rq
,
323 struct xdp_buff
*xdp
)
325 struct virtio_net_hdr_mrg_rxbuf
*hdr
;
327 struct send_queue
*sq
;
332 qp
= vi
->curr_queue_pairs
- vi
->xdp_queue_pairs
+ smp_processor_id();
335 /* Free up any pending old buffers before queueing new ones. */
336 while ((xdp_sent
= virtqueue_get_buf(sq
->vq
, &len
)) != NULL
) {
337 struct page
*sent_page
= virt_to_head_page(xdp_sent
);
342 xdp
->data
-= vi
->hdr_len
;
343 /* Zero header and leave csum up to XDP layers */
345 memset(hdr
, 0, vi
->hdr_len
);
347 sg_init_one(sq
->sg
, xdp
->data
, xdp
->data_end
- xdp
->data
);
349 err
= virtqueue_add_outbuf(sq
->vq
, sq
->sg
, 1, xdp
->data
, GFP_ATOMIC
);
351 struct page
*page
= virt_to_head_page(xdp
->data
);
357 virtqueue_kick(sq
->vq
);
361 static unsigned int virtnet_get_headroom(struct virtnet_info
*vi
)
363 return vi
->xdp_queue_pairs
? VIRTIO_XDP_HEADROOM
: 0;
366 static struct sk_buff
*receive_small(struct net_device
*dev
,
367 struct virtnet_info
*vi
,
368 struct receive_queue
*rq
,
369 void *buf
, unsigned int len
)
372 struct bpf_prog
*xdp_prog
;
373 unsigned int xdp_headroom
= virtnet_get_headroom(vi
);
374 unsigned int header_offset
= VIRTNET_RX_PAD
+ xdp_headroom
;
375 unsigned int headroom
= vi
->hdr_len
+ header_offset
;
376 unsigned int buflen
= SKB_DATA_ALIGN(GOOD_PACKET_LEN
+ headroom
) +
377 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
378 unsigned int delta
= 0;
382 xdp_prog
= rcu_dereference(rq
->xdp_prog
);
384 struct virtio_net_hdr_mrg_rxbuf
*hdr
= buf
+ header_offset
;
389 if (unlikely(hdr
->hdr
.gso_type
|| hdr
->hdr
.flags
))
392 xdp
.data_hard_start
= buf
+ VIRTNET_RX_PAD
+ vi
->hdr_len
;
393 xdp
.data
= xdp
.data_hard_start
+ xdp_headroom
;
394 xdp
.data_end
= xdp
.data
+ len
;
395 orig_data
= xdp
.data
;
396 act
= bpf_prog_run_xdp(xdp_prog
, &xdp
);
400 /* Recalculate length in case bpf program changed it */
401 delta
= orig_data
- xdp
.data
;
404 if (unlikely(!virtnet_xdp_xmit(vi
, rq
, &xdp
)))
405 trace_xdp_exception(vi
->dev
, xdp_prog
, act
);
409 bpf_warn_invalid_xdp_action(act
);
411 trace_xdp_exception(vi
->dev
, xdp_prog
, act
);
418 skb
= build_skb(buf
, buflen
);
420 put_page(virt_to_head_page(buf
));
423 skb_reserve(skb
, headroom
- delta
);
424 skb_put(skb
, len
+ delta
);
426 buf
+= header_offset
;
427 memcpy(skb_vnet_hdr(skb
), buf
, vi
->hdr_len
);
428 } /* keep zeroed vnet hdr since packet was changed by bpf */
435 dev
->stats
.rx_dropped
++;
436 put_page(virt_to_head_page(buf
));
441 static struct sk_buff
*receive_big(struct net_device
*dev
,
442 struct virtnet_info
*vi
,
443 struct receive_queue
*rq
,
447 struct page
*page
= buf
;
448 struct sk_buff
*skb
= page_to_skb(vi
, rq
, page
, 0, len
, PAGE_SIZE
);
456 dev
->stats
.rx_dropped
++;
457 give_pages(rq
, page
);
461 /* The conditions to enable XDP should preclude the underlying device from
462 * sending packets across multiple buffers (num_buf > 1). However per spec
463 * it does not appear to be illegal to do so but rather just against convention.
464 * So in order to avoid making a system unresponsive the packets are pushed
465 * into a page and the XDP program is run. This will be extremely slow and we
466 * push a warning to the user to fix this as soon as possible. Fixing this may
467 * require resolving the underlying hardware to determine why multiple buffers
468 * are being received or simply loading the XDP program in the ingress stack
469 * after the skb is built because there is no advantage to running it here
472 static struct page
*xdp_linearize_page(struct receive_queue
*rq
,
478 struct page
*page
= alloc_page(GFP_ATOMIC
);
479 unsigned int page_off
= VIRTIO_XDP_HEADROOM
;
484 memcpy(page_address(page
) + page_off
, page_address(p
) + offset
, *len
);
492 buf
= virtqueue_get_buf(rq
->vq
, &buflen
);
496 p
= virt_to_head_page(buf
);
497 off
= buf
- page_address(p
);
499 /* guard against a misconfigured or uncooperative backend that
500 * is sending packet larger than the MTU.
502 if ((page_off
+ buflen
) > PAGE_SIZE
) {
507 memcpy(page_address(page
) + page_off
,
508 page_address(p
) + off
, buflen
);
513 /* Headroom does not contribute to packet length */
514 *len
= page_off
- VIRTIO_XDP_HEADROOM
;
517 __free_pages(page
, 0);
521 static struct sk_buff
*receive_mergeable(struct net_device
*dev
,
522 struct virtnet_info
*vi
,
523 struct receive_queue
*rq
,
528 struct virtio_net_hdr_mrg_rxbuf
*hdr
= buf
;
529 u16 num_buf
= virtio16_to_cpu(vi
->vdev
, hdr
->num_buffers
);
530 struct page
*page
= virt_to_head_page(buf
);
531 int offset
= buf
- page_address(page
);
532 struct sk_buff
*head_skb
, *curr_skb
;
533 struct bpf_prog
*xdp_prog
;
534 unsigned int truesize
;
539 xdp_prog
= rcu_dereference(rq
->xdp_prog
);
541 struct page
*xdp_page
;
546 /* This happens when rx buffer size is underestimated */
547 if (unlikely(num_buf
> 1)) {
548 /* linearize data for XDP */
549 xdp_page
= xdp_linearize_page(rq
, &num_buf
,
553 offset
= VIRTIO_XDP_HEADROOM
;
558 /* Transient failure which in theory could occur if
559 * in-flight packets from before XDP was enabled reach
560 * the receive path after XDP is loaded. In practice I
561 * was not able to create this condition.
563 if (unlikely(hdr
->hdr
.gso_type
))
566 /* Allow consuming headroom but reserve enough space to push
567 * the descriptor on if we get an XDP_TX return code.
569 data
= page_address(xdp_page
) + offset
;
570 xdp
.data_hard_start
= data
- VIRTIO_XDP_HEADROOM
+ vi
->hdr_len
;
571 xdp
.data
= data
+ vi
->hdr_len
;
572 xdp
.data_end
= xdp
.data
+ (len
- vi
->hdr_len
);
573 act
= bpf_prog_run_xdp(xdp_prog
, &xdp
);
577 /* recalculate offset to account for any header
578 * adjustments. Note other cases do not build an
579 * skb and avoid using offset
582 page_address(xdp_page
) - vi
->hdr_len
;
584 /* We can only create skb based on xdp_page. */
585 if (unlikely(xdp_page
!= page
)) {
588 head_skb
= page_to_skb(vi
, rq
, xdp_page
,
589 offset
, len
, PAGE_SIZE
);
590 ewma_pkt_len_add(&rq
->mrg_avg_pkt_len
, len
);
595 if (unlikely(!virtnet_xdp_xmit(vi
, rq
, &xdp
)))
596 trace_xdp_exception(vi
->dev
, xdp_prog
, act
);
597 ewma_pkt_len_add(&rq
->mrg_avg_pkt_len
, len
);
598 if (unlikely(xdp_page
!= page
))
603 bpf_warn_invalid_xdp_action(act
);
605 trace_xdp_exception(vi
->dev
, xdp_prog
, act
);
607 if (unlikely(xdp_page
!= page
))
608 __free_pages(xdp_page
, 0);
609 ewma_pkt_len_add(&rq
->mrg_avg_pkt_len
, len
);
615 if (unlikely(len
> (unsigned long)ctx
)) {
616 pr_debug("%s: rx error: len %u exceeds truesize %lu\n",
617 dev
->name
, len
, (unsigned long)ctx
);
618 dev
->stats
.rx_length_errors
++;
621 truesize
= (unsigned long)ctx
;
622 head_skb
= page_to_skb(vi
, rq
, page
, offset
, len
, truesize
);
625 if (unlikely(!curr_skb
))
630 buf
= virtqueue_get_buf_ctx(rq
->vq
, &len
, &ctx
);
631 if (unlikely(!ctx
)) {
632 pr_debug("%s: rx error: %d buffers out of %d missing\n",
634 virtio16_to_cpu(vi
->vdev
,
636 dev
->stats
.rx_length_errors
++;
640 page
= virt_to_head_page(buf
);
641 if (unlikely(len
> (unsigned long)ctx
)) {
642 pr_debug("%s: rx error: len %u exceeds truesize %lu\n",
643 dev
->name
, len
, (unsigned long)ctx
);
644 dev
->stats
.rx_length_errors
++;
647 truesize
= (unsigned long)ctx
;
649 num_skb_frags
= skb_shinfo(curr_skb
)->nr_frags
;
650 if (unlikely(num_skb_frags
== MAX_SKB_FRAGS
)) {
651 struct sk_buff
*nskb
= alloc_skb(0, GFP_ATOMIC
);
655 if (curr_skb
== head_skb
)
656 skb_shinfo(curr_skb
)->frag_list
= nskb
;
658 curr_skb
->next
= nskb
;
660 head_skb
->truesize
+= nskb
->truesize
;
663 if (curr_skb
!= head_skb
) {
664 head_skb
->data_len
+= len
;
665 head_skb
->len
+= len
;
666 head_skb
->truesize
+= truesize
;
668 offset
= buf
- page_address(page
);
669 if (skb_can_coalesce(curr_skb
, num_skb_frags
, page
, offset
)) {
671 skb_coalesce_rx_frag(curr_skb
, num_skb_frags
- 1,
674 skb_add_rx_frag(curr_skb
, num_skb_frags
, page
,
675 offset
, len
, truesize
);
679 ewma_pkt_len_add(&rq
->mrg_avg_pkt_len
, head_skb
->len
);
687 buf
= virtqueue_get_buf(rq
->vq
, &len
);
688 if (unlikely(!buf
)) {
689 pr_debug("%s: rx error: %d buffers missing\n",
691 dev
->stats
.rx_length_errors
++;
694 page
= virt_to_head_page(buf
);
698 dev
->stats
.rx_dropped
++;
699 dev_kfree_skb(head_skb
);
704 static int receive_buf(struct virtnet_info
*vi
, struct receive_queue
*rq
,
705 void *buf
, unsigned int len
, void **ctx
)
707 struct net_device
*dev
= vi
->dev
;
709 struct virtio_net_hdr_mrg_rxbuf
*hdr
;
712 if (unlikely(len
< vi
->hdr_len
+ ETH_HLEN
)) {
713 pr_debug("%s: short packet %i\n", dev
->name
, len
);
714 dev
->stats
.rx_length_errors
++;
715 if (vi
->mergeable_rx_bufs
) {
716 put_page(virt_to_head_page(buf
));
717 } else if (vi
->big_packets
) {
720 put_page(virt_to_head_page(buf
));
725 if (vi
->mergeable_rx_bufs
)
726 skb
= receive_mergeable(dev
, vi
, rq
, buf
, ctx
, len
);
727 else if (vi
->big_packets
)
728 skb
= receive_big(dev
, vi
, rq
, buf
, len
);
730 skb
= receive_small(dev
, vi
, rq
, buf
, len
);
735 hdr
= skb_vnet_hdr(skb
);
739 if (hdr
->hdr
.flags
& VIRTIO_NET_HDR_F_DATA_VALID
)
740 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
742 if (virtio_net_hdr_to_skb(skb
, &hdr
->hdr
,
743 virtio_is_little_endian(vi
->vdev
))) {
744 net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n",
745 dev
->name
, hdr
->hdr
.gso_type
,
750 skb
->protocol
= eth_type_trans(skb
, dev
);
751 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
752 ntohs(skb
->protocol
), skb
->len
, skb
->pkt_type
);
754 napi_gro_receive(&rq
->napi
, skb
);
758 dev
->stats
.rx_frame_errors
++;
763 static int add_recvbuf_small(struct virtnet_info
*vi
, struct receive_queue
*rq
,
766 struct page_frag
*alloc_frag
= &rq
->alloc_frag
;
768 unsigned int xdp_headroom
= virtnet_get_headroom(vi
);
769 int len
= vi
->hdr_len
+ VIRTNET_RX_PAD
+ GOOD_PACKET_LEN
+ xdp_headroom
;
772 len
= SKB_DATA_ALIGN(len
) +
773 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
774 if (unlikely(!skb_page_frag_refill(len
, alloc_frag
, gfp
)))
777 buf
= (char *)page_address(alloc_frag
->page
) + alloc_frag
->offset
;
778 get_page(alloc_frag
->page
);
779 alloc_frag
->offset
+= len
;
780 sg_init_one(rq
->sg
, buf
+ VIRTNET_RX_PAD
+ xdp_headroom
,
781 vi
->hdr_len
+ GOOD_PACKET_LEN
);
782 err
= virtqueue_add_inbuf(rq
->vq
, rq
->sg
, 1, buf
, gfp
);
784 put_page(virt_to_head_page(buf
));
789 static int add_recvbuf_big(struct virtnet_info
*vi
, struct receive_queue
*rq
,
792 struct page
*first
, *list
= NULL
;
796 sg_init_table(rq
->sg
, MAX_SKB_FRAGS
+ 2);
798 /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */
799 for (i
= MAX_SKB_FRAGS
+ 1; i
> 1; --i
) {
800 first
= get_a_page(rq
, gfp
);
803 give_pages(rq
, list
);
806 sg_set_buf(&rq
->sg
[i
], page_address(first
), PAGE_SIZE
);
808 /* chain new page in list head to match sg */
809 first
->private = (unsigned long)list
;
813 first
= get_a_page(rq
, gfp
);
815 give_pages(rq
, list
);
818 p
= page_address(first
);
820 /* rq->sg[0], rq->sg[1] share the same page */
821 /* a separated rq->sg[0] for header - required in case !any_header_sg */
822 sg_set_buf(&rq
->sg
[0], p
, vi
->hdr_len
);
824 /* rq->sg[1] for data packet, from offset */
825 offset
= sizeof(struct padded_vnet_hdr
);
826 sg_set_buf(&rq
->sg
[1], p
+ offset
, PAGE_SIZE
- offset
);
828 /* chain first in list head */
829 first
->private = (unsigned long)list
;
830 err
= virtqueue_add_inbuf(rq
->vq
, rq
->sg
, MAX_SKB_FRAGS
+ 2,
833 give_pages(rq
, first
);
838 static unsigned int get_mergeable_buf_len(struct receive_queue
*rq
,
839 struct ewma_pkt_len
*avg_pkt_len
)
841 const size_t hdr_len
= sizeof(struct virtio_net_hdr_mrg_rxbuf
);
844 len
= hdr_len
+ clamp_t(unsigned int, ewma_pkt_len_read(avg_pkt_len
),
845 rq
->min_buf_len
- hdr_len
, PAGE_SIZE
- hdr_len
);
846 return ALIGN(len
, L1_CACHE_BYTES
);
849 static int add_recvbuf_mergeable(struct virtnet_info
*vi
,
850 struct receive_queue
*rq
, gfp_t gfp
)
852 struct page_frag
*alloc_frag
= &rq
->alloc_frag
;
853 unsigned int headroom
= virtnet_get_headroom(vi
);
857 unsigned int len
, hole
;
859 len
= get_mergeable_buf_len(rq
, &rq
->mrg_avg_pkt_len
);
860 if (unlikely(!skb_page_frag_refill(len
+ headroom
, alloc_frag
, gfp
)))
863 buf
= (char *)page_address(alloc_frag
->page
) + alloc_frag
->offset
;
864 buf
+= headroom
; /* advance address leaving hole at front of pkt */
865 ctx
= (void *)(unsigned long)len
;
866 get_page(alloc_frag
->page
);
867 alloc_frag
->offset
+= len
+ headroom
;
868 hole
= alloc_frag
->size
- alloc_frag
->offset
;
869 if (hole
< len
+ headroom
) {
870 /* To avoid internal fragmentation, if there is very likely not
871 * enough space for another buffer, add the remaining space to
872 * the current buffer. This extra space is not included in
873 * the truesize stored in ctx.
876 alloc_frag
->offset
+= hole
;
879 sg_init_one(rq
->sg
, buf
, len
);
880 err
= virtqueue_add_inbuf_ctx(rq
->vq
, rq
->sg
, 1, buf
, ctx
, gfp
);
882 put_page(virt_to_head_page(buf
));
888 * Returns false if we couldn't fill entirely (OOM).
890 * Normally run in the receive path, but can also be run from ndo_open
891 * before we're receiving packets, or from refill_work which is
892 * careful to disable receiving (using napi_disable).
894 static bool try_fill_recv(struct virtnet_info
*vi
, struct receive_queue
*rq
,
902 if (vi
->mergeable_rx_bufs
)
903 err
= add_recvbuf_mergeable(vi
, rq
, gfp
);
904 else if (vi
->big_packets
)
905 err
= add_recvbuf_big(vi
, rq
, gfp
);
907 err
= add_recvbuf_small(vi
, rq
, gfp
);
909 oom
= err
== -ENOMEM
;
912 } while (rq
->vq
->num_free
);
913 virtqueue_kick(rq
->vq
);
917 static void skb_recv_done(struct virtqueue
*rvq
)
919 struct virtnet_info
*vi
= rvq
->vdev
->priv
;
920 struct receive_queue
*rq
= &vi
->rq
[vq2rxq(rvq
)];
922 /* Schedule NAPI, Suppress further interrupts if successful. */
923 if (napi_schedule_prep(&rq
->napi
)) {
924 virtqueue_disable_cb(rvq
);
925 __napi_schedule(&rq
->napi
);
929 static void virtnet_napi_enable(struct receive_queue
*rq
)
931 napi_enable(&rq
->napi
);
933 /* If all buffers were filled by other side before we napi_enabled, we
934 * won't get another interrupt, so process any outstanding packets
935 * now. virtnet_poll wants re-enable the queue, so we disable here.
936 * We synchronize against interrupts via NAPI_STATE_SCHED */
937 if (napi_schedule_prep(&rq
->napi
)) {
938 virtqueue_disable_cb(rq
->vq
);
940 __napi_schedule(&rq
->napi
);
945 static void refill_work(struct work_struct
*work
)
947 struct virtnet_info
*vi
=
948 container_of(work
, struct virtnet_info
, refill
.work
);
952 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++) {
953 struct receive_queue
*rq
= &vi
->rq
[i
];
955 napi_disable(&rq
->napi
);
956 still_empty
= !try_fill_recv(vi
, rq
, GFP_KERNEL
);
957 virtnet_napi_enable(rq
);
959 /* In theory, this can happen: if we don't get any buffers in
960 * we will *never* try to fill again.
963 schedule_delayed_work(&vi
->refill
, HZ
/2);
967 static int virtnet_receive(struct receive_queue
*rq
, int budget
)
969 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
970 unsigned int len
, received
= 0, bytes
= 0;
972 struct virtnet_stats
*stats
= this_cpu_ptr(vi
->stats
);
974 if (vi
->mergeable_rx_bufs
) {
977 while (received
< budget
&&
978 (buf
= virtqueue_get_buf_ctx(rq
->vq
, &len
, &ctx
))) {
979 bytes
+= receive_buf(vi
, rq
, buf
, len
, ctx
);
983 while (received
< budget
&&
984 (buf
= virtqueue_get_buf(rq
->vq
, &len
)) != NULL
) {
985 bytes
+= receive_buf(vi
, rq
, buf
, len
, NULL
);
990 if (rq
->vq
->num_free
> virtqueue_get_vring_size(rq
->vq
) / 2) {
991 if (!try_fill_recv(vi
, rq
, GFP_ATOMIC
))
992 schedule_delayed_work(&vi
->refill
, 0);
995 u64_stats_update_begin(&stats
->rx_syncp
);
996 stats
->rx_bytes
+= bytes
;
997 stats
->rx_packets
+= received
;
998 u64_stats_update_end(&stats
->rx_syncp
);
1003 static int virtnet_poll(struct napi_struct
*napi
, int budget
)
1005 struct receive_queue
*rq
=
1006 container_of(napi
, struct receive_queue
, napi
);
1007 unsigned int r
, received
;
1009 received
= virtnet_receive(rq
, budget
);
1011 /* Out of packets? */
1012 if (received
< budget
) {
1013 r
= virtqueue_enable_cb_prepare(rq
->vq
);
1014 if (napi_complete_done(napi
, received
)) {
1015 if (unlikely(virtqueue_poll(rq
->vq
, r
)) &&
1016 napi_schedule_prep(napi
)) {
1017 virtqueue_disable_cb(rq
->vq
);
1018 __napi_schedule(napi
);
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
);
1048 unsigned int packets
= 0;
1049 unsigned int bytes
= 0;
1051 while ((skb
= virtqueue_get_buf(sq
->vq
, &len
)) != NULL
) {
1052 pr_debug("Sent skb %p\n", skb
);
1057 dev_kfree_skb_any(skb
);
1060 /* Avoid overhead when no packets have been processed
1061 * happens when called speculatively from start_xmit.
1066 u64_stats_update_begin(&stats
->tx_syncp
);
1067 stats
->tx_bytes
+= bytes
;
1068 stats
->tx_packets
+= packets
;
1069 u64_stats_update_end(&stats
->tx_syncp
);
1072 static int xmit_skb(struct send_queue
*sq
, struct sk_buff
*skb
)
1074 struct virtio_net_hdr_mrg_rxbuf
*hdr
;
1075 const unsigned char *dest
= ((struct ethhdr
*)skb
->data
)->h_dest
;
1076 struct virtnet_info
*vi
= sq
->vq
->vdev
->priv
;
1078 unsigned hdr_len
= vi
->hdr_len
;
1081 pr_debug("%s: xmit %p %pM\n", vi
->dev
->name
, skb
, dest
);
1083 can_push
= vi
->any_header_sg
&&
1084 !((unsigned long)skb
->data
& (__alignof__(*hdr
) - 1)) &&
1085 !skb_header_cloned(skb
) && skb_headroom(skb
) >= hdr_len
;
1086 /* Even if we can, don't push here yet as this would skew
1087 * csum_start offset below. */
1089 hdr
= (struct virtio_net_hdr_mrg_rxbuf
*)(skb
->data
- hdr_len
);
1091 hdr
= skb_vnet_hdr(skb
);
1093 if (virtio_net_hdr_from_skb(skb
, &hdr
->hdr
,
1094 virtio_is_little_endian(vi
->vdev
), false))
1097 if (vi
->mergeable_rx_bufs
)
1098 hdr
->num_buffers
= 0;
1100 sg_init_table(sq
->sg
, skb_shinfo(skb
)->nr_frags
+ (can_push
? 1 : 2));
1102 __skb_push(skb
, hdr_len
);
1103 num_sg
= skb_to_sgvec(skb
, sq
->sg
, 0, skb
->len
);
1104 /* Pull header back to avoid skew in tx bytes calculations. */
1105 __skb_pull(skb
, hdr_len
);
1107 sg_set_buf(sq
->sg
, hdr
, hdr_len
);
1108 num_sg
= skb_to_sgvec(skb
, sq
->sg
+ 1, 0, skb
->len
) + 1;
1110 return virtqueue_add_outbuf(sq
->vq
, sq
->sg
, num_sg
, skb
, GFP_ATOMIC
);
1113 static netdev_tx_t
start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
1115 struct virtnet_info
*vi
= netdev_priv(dev
);
1116 int qnum
= skb_get_queue_mapping(skb
);
1117 struct send_queue
*sq
= &vi
->sq
[qnum
];
1119 struct netdev_queue
*txq
= netdev_get_tx_queue(dev
, qnum
);
1120 bool kick
= !skb
->xmit_more
;
1122 /* Free up any pending old buffers before queueing new ones. */
1123 free_old_xmit_skbs(sq
);
1125 /* timestamp packet in software */
1126 skb_tx_timestamp(skb
);
1128 /* Try to transmit */
1129 err
= xmit_skb(sq
, skb
);
1131 /* This should not happen! */
1132 if (unlikely(err
)) {
1133 dev
->stats
.tx_fifo_errors
++;
1134 if (net_ratelimit())
1136 "Unexpected TXQ (%d) queue failure: %d\n", qnum
, err
);
1137 dev
->stats
.tx_dropped
++;
1138 dev_kfree_skb_any(skb
);
1139 return NETDEV_TX_OK
;
1142 /* Don't wait up for transmitted skbs to be freed. */
1146 /* If running out of space, stop queue to avoid getting packets that we
1147 * are then unable to transmit.
1148 * An alternative would be to force queuing layer to requeue the skb by
1149 * returning NETDEV_TX_BUSY. However, NETDEV_TX_BUSY should not be
1150 * returned in a normal path of operation: it means that driver is not
1151 * maintaining the TX queue stop/start state properly, and causes
1152 * the stack to do a non-trivial amount of useless work.
1153 * Since most packets only take 1 or 2 ring slots, stopping the queue
1154 * early means 16 slots are typically wasted.
1156 if (sq
->vq
->num_free
< 2+MAX_SKB_FRAGS
) {
1157 netif_stop_subqueue(dev
, qnum
);
1158 if (unlikely(!virtqueue_enable_cb_delayed(sq
->vq
))) {
1159 /* More just got used, free them then recheck. */
1160 free_old_xmit_skbs(sq
);
1161 if (sq
->vq
->num_free
>= 2+MAX_SKB_FRAGS
) {
1162 netif_start_subqueue(dev
, qnum
);
1163 virtqueue_disable_cb(sq
->vq
);
1168 if (kick
|| netif_xmit_stopped(txq
))
1169 virtqueue_kick(sq
->vq
);
1171 return NETDEV_TX_OK
;
1175 * Send command via the control virtqueue and check status. Commands
1176 * supported by the hypervisor, as indicated by feature bits, should
1177 * never fail unless improperly formatted.
1179 static bool virtnet_send_command(struct virtnet_info
*vi
, u8
class, u8 cmd
,
1180 struct scatterlist
*out
)
1182 struct scatterlist
*sgs
[4], hdr
, stat
;
1183 unsigned out_num
= 0, tmp
;
1185 /* Caller should know better */
1186 BUG_ON(!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
));
1188 vi
->ctrl_status
= ~0;
1189 vi
->ctrl_hdr
.class = class;
1190 vi
->ctrl_hdr
.cmd
= cmd
;
1192 sg_init_one(&hdr
, &vi
->ctrl_hdr
, sizeof(vi
->ctrl_hdr
));
1193 sgs
[out_num
++] = &hdr
;
1196 sgs
[out_num
++] = out
;
1198 /* Add return status. */
1199 sg_init_one(&stat
, &vi
->ctrl_status
, sizeof(vi
->ctrl_status
));
1200 sgs
[out_num
] = &stat
;
1202 BUG_ON(out_num
+ 1 > ARRAY_SIZE(sgs
));
1203 virtqueue_add_sgs(vi
->cvq
, sgs
, out_num
, 1, vi
, GFP_ATOMIC
);
1205 if (unlikely(!virtqueue_kick(vi
->cvq
)))
1206 return vi
->ctrl_status
== VIRTIO_NET_OK
;
1208 /* Spin for a response, the kick causes an ioport write, trapping
1209 * into the hypervisor, so the request should be handled immediately.
1211 while (!virtqueue_get_buf(vi
->cvq
, &tmp
) &&
1212 !virtqueue_is_broken(vi
->cvq
))
1215 return vi
->ctrl_status
== VIRTIO_NET_OK
;
1218 static int virtnet_set_mac_address(struct net_device
*dev
, void *p
)
1220 struct virtnet_info
*vi
= netdev_priv(dev
);
1221 struct virtio_device
*vdev
= vi
->vdev
;
1223 struct sockaddr
*addr
;
1224 struct scatterlist sg
;
1226 addr
= kmemdup(p
, sizeof(*addr
), GFP_KERNEL
);
1230 ret
= eth_prepare_mac_addr_change(dev
, addr
);
1234 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_MAC_ADDR
)) {
1235 sg_init_one(&sg
, addr
->sa_data
, dev
->addr_len
);
1236 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MAC
,
1237 VIRTIO_NET_CTRL_MAC_ADDR_SET
, &sg
)) {
1238 dev_warn(&vdev
->dev
,
1239 "Failed to set mac address by vq command.\n");
1243 } else if (virtio_has_feature(vdev
, VIRTIO_NET_F_MAC
) &&
1244 !virtio_has_feature(vdev
, VIRTIO_F_VERSION_1
)) {
1247 /* Naturally, this has an atomicity problem. */
1248 for (i
= 0; i
< dev
->addr_len
; i
++)
1249 virtio_cwrite8(vdev
,
1250 offsetof(struct virtio_net_config
, mac
) +
1251 i
, addr
->sa_data
[i
]);
1254 eth_commit_mac_addr_change(dev
, p
);
1262 static void virtnet_stats(struct net_device
*dev
,
1263 struct rtnl_link_stats64
*tot
)
1265 struct virtnet_info
*vi
= netdev_priv(dev
);
1269 for_each_possible_cpu(cpu
) {
1270 struct virtnet_stats
*stats
= per_cpu_ptr(vi
->stats
, cpu
);
1271 u64 tpackets
, tbytes
, rpackets
, rbytes
;
1274 start
= u64_stats_fetch_begin_irq(&stats
->tx_syncp
);
1275 tpackets
= stats
->tx_packets
;
1276 tbytes
= stats
->tx_bytes
;
1277 } while (u64_stats_fetch_retry_irq(&stats
->tx_syncp
, start
));
1280 start
= u64_stats_fetch_begin_irq(&stats
->rx_syncp
);
1281 rpackets
= stats
->rx_packets
;
1282 rbytes
= stats
->rx_bytes
;
1283 } while (u64_stats_fetch_retry_irq(&stats
->rx_syncp
, start
));
1285 tot
->rx_packets
+= rpackets
;
1286 tot
->tx_packets
+= tpackets
;
1287 tot
->rx_bytes
+= rbytes
;
1288 tot
->tx_bytes
+= tbytes
;
1291 tot
->tx_dropped
= dev
->stats
.tx_dropped
;
1292 tot
->tx_fifo_errors
= dev
->stats
.tx_fifo_errors
;
1293 tot
->rx_dropped
= dev
->stats
.rx_dropped
;
1294 tot
->rx_length_errors
= dev
->stats
.rx_length_errors
;
1295 tot
->rx_frame_errors
= dev
->stats
.rx_frame_errors
;
1298 #ifdef CONFIG_NET_POLL_CONTROLLER
1299 static void virtnet_netpoll(struct net_device
*dev
)
1301 struct virtnet_info
*vi
= netdev_priv(dev
);
1304 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++)
1305 napi_schedule(&vi
->rq
[i
].napi
);
1309 static void virtnet_ack_link_announce(struct virtnet_info
*vi
)
1312 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_ANNOUNCE
,
1313 VIRTIO_NET_CTRL_ANNOUNCE_ACK
, NULL
))
1314 dev_warn(&vi
->dev
->dev
, "Failed to ack link announce.\n");
1318 static int _virtnet_set_queues(struct virtnet_info
*vi
, u16 queue_pairs
)
1320 struct scatterlist sg
;
1321 struct net_device
*dev
= vi
->dev
;
1323 if (!vi
->has_cvq
|| !virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_MQ
))
1326 vi
->ctrl_mq
.virtqueue_pairs
= cpu_to_virtio16(vi
->vdev
, queue_pairs
);
1327 sg_init_one(&sg
, &vi
->ctrl_mq
, sizeof(vi
->ctrl_mq
));
1329 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MQ
,
1330 VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET
, &sg
)) {
1331 dev_warn(&dev
->dev
, "Fail to set num of queue pairs to %d\n",
1335 vi
->curr_queue_pairs
= queue_pairs
;
1336 /* virtnet_open() will refill when device is going to up. */
1337 if (dev
->flags
& IFF_UP
)
1338 schedule_delayed_work(&vi
->refill
, 0);
1344 static int virtnet_set_queues(struct virtnet_info
*vi
, u16 queue_pairs
)
1349 err
= _virtnet_set_queues(vi
, queue_pairs
);
1354 static int virtnet_close(struct net_device
*dev
)
1356 struct virtnet_info
*vi
= netdev_priv(dev
);
1359 /* Make sure refill_work doesn't re-enable napi! */
1360 cancel_delayed_work_sync(&vi
->refill
);
1362 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
1363 napi_disable(&vi
->rq
[i
].napi
);
1368 static void virtnet_set_rx_mode(struct net_device
*dev
)
1370 struct virtnet_info
*vi
= netdev_priv(dev
);
1371 struct scatterlist sg
[2];
1372 struct virtio_net_ctrl_mac
*mac_data
;
1373 struct netdev_hw_addr
*ha
;
1379 /* We can't dynamically set ndo_set_rx_mode, so return gracefully */
1380 if (!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_RX
))
1383 vi
->ctrl_promisc
= ((dev
->flags
& IFF_PROMISC
) != 0);
1384 vi
->ctrl_allmulti
= ((dev
->flags
& IFF_ALLMULTI
) != 0);
1386 sg_init_one(sg
, &vi
->ctrl_promisc
, sizeof(vi
->ctrl_promisc
));
1388 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_RX
,
1389 VIRTIO_NET_CTRL_RX_PROMISC
, sg
))
1390 dev_warn(&dev
->dev
, "Failed to %sable promisc mode.\n",
1391 vi
->ctrl_promisc
? "en" : "dis");
1393 sg_init_one(sg
, &vi
->ctrl_allmulti
, sizeof(vi
->ctrl_allmulti
));
1395 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_RX
,
1396 VIRTIO_NET_CTRL_RX_ALLMULTI
, sg
))
1397 dev_warn(&dev
->dev
, "Failed to %sable allmulti mode.\n",
1398 vi
->ctrl_allmulti
? "en" : "dis");
1400 uc_count
= netdev_uc_count(dev
);
1401 mc_count
= netdev_mc_count(dev
);
1402 /* MAC filter - use one buffer for both lists */
1403 buf
= kzalloc(((uc_count
+ mc_count
) * ETH_ALEN
) +
1404 (2 * sizeof(mac_data
->entries
)), GFP_ATOMIC
);
1409 sg_init_table(sg
, 2);
1411 /* Store the unicast list and count in the front of the buffer */
1412 mac_data
->entries
= cpu_to_virtio32(vi
->vdev
, uc_count
);
1414 netdev_for_each_uc_addr(ha
, dev
)
1415 memcpy(&mac_data
->macs
[i
++][0], ha
->addr
, ETH_ALEN
);
1417 sg_set_buf(&sg
[0], mac_data
,
1418 sizeof(mac_data
->entries
) + (uc_count
* ETH_ALEN
));
1420 /* multicast list and count fill the end */
1421 mac_data
= (void *)&mac_data
->macs
[uc_count
][0];
1423 mac_data
->entries
= cpu_to_virtio32(vi
->vdev
, mc_count
);
1425 netdev_for_each_mc_addr(ha
, dev
)
1426 memcpy(&mac_data
->macs
[i
++][0], ha
->addr
, ETH_ALEN
);
1428 sg_set_buf(&sg
[1], mac_data
,
1429 sizeof(mac_data
->entries
) + (mc_count
* ETH_ALEN
));
1431 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MAC
,
1432 VIRTIO_NET_CTRL_MAC_TABLE_SET
, sg
))
1433 dev_warn(&dev
->dev
, "Failed to set MAC filter table.\n");
1438 static int virtnet_vlan_rx_add_vid(struct net_device
*dev
,
1439 __be16 proto
, u16 vid
)
1441 struct virtnet_info
*vi
= netdev_priv(dev
);
1442 struct scatterlist sg
;
1445 sg_init_one(&sg
, &vi
->ctrl_vid
, sizeof(vi
->ctrl_vid
));
1447 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_VLAN
,
1448 VIRTIO_NET_CTRL_VLAN_ADD
, &sg
))
1449 dev_warn(&dev
->dev
, "Failed to add VLAN ID %d.\n", vid
);
1453 static int virtnet_vlan_rx_kill_vid(struct net_device
*dev
,
1454 __be16 proto
, u16 vid
)
1456 struct virtnet_info
*vi
= netdev_priv(dev
);
1457 struct scatterlist sg
;
1460 sg_init_one(&sg
, &vi
->ctrl_vid
, sizeof(vi
->ctrl_vid
));
1462 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_VLAN
,
1463 VIRTIO_NET_CTRL_VLAN_DEL
, &sg
))
1464 dev_warn(&dev
->dev
, "Failed to kill VLAN ID %d.\n", vid
);
1468 static void virtnet_clean_affinity(struct virtnet_info
*vi
, long hcpu
)
1472 if (vi
->affinity_hint_set
) {
1473 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1474 virtqueue_set_affinity(vi
->rq
[i
].vq
, -1);
1475 virtqueue_set_affinity(vi
->sq
[i
].vq
, -1);
1478 vi
->affinity_hint_set
= false;
1482 static void virtnet_set_affinity(struct virtnet_info
*vi
)
1487 /* In multiqueue mode, when the number of cpu is equal to the number of
1488 * queue pairs, we let the queue pairs to be private to one cpu by
1489 * setting the affinity hint to eliminate the contention.
1491 if (vi
->curr_queue_pairs
== 1 ||
1492 vi
->max_queue_pairs
!= num_online_cpus()) {
1493 virtnet_clean_affinity(vi
, -1);
1498 for_each_online_cpu(cpu
) {
1499 virtqueue_set_affinity(vi
->rq
[i
].vq
, cpu
);
1500 virtqueue_set_affinity(vi
->sq
[i
].vq
, cpu
);
1501 netif_set_xps_queue(vi
->dev
, cpumask_of(cpu
), i
);
1505 vi
->affinity_hint_set
= true;
1508 static int virtnet_cpu_online(unsigned int cpu
, struct hlist_node
*node
)
1510 struct virtnet_info
*vi
= hlist_entry_safe(node
, struct virtnet_info
,
1512 virtnet_set_affinity(vi
);
1516 static int virtnet_cpu_dead(unsigned int cpu
, struct hlist_node
*node
)
1518 struct virtnet_info
*vi
= hlist_entry_safe(node
, struct virtnet_info
,
1520 virtnet_set_affinity(vi
);
1524 static int virtnet_cpu_down_prep(unsigned int cpu
, struct hlist_node
*node
)
1526 struct virtnet_info
*vi
= hlist_entry_safe(node
, struct virtnet_info
,
1529 virtnet_clean_affinity(vi
, cpu
);
1533 static enum cpuhp_state virtionet_online
;
1535 static int virtnet_cpu_notif_add(struct virtnet_info
*vi
)
1539 ret
= cpuhp_state_add_instance_nocalls(virtionet_online
, &vi
->node
);
1542 ret
= cpuhp_state_add_instance_nocalls(CPUHP_VIRT_NET_DEAD
,
1546 cpuhp_state_remove_instance_nocalls(virtionet_online
, &vi
->node
);
1550 static void virtnet_cpu_notif_remove(struct virtnet_info
*vi
)
1552 cpuhp_state_remove_instance_nocalls(virtionet_online
, &vi
->node
);
1553 cpuhp_state_remove_instance_nocalls(CPUHP_VIRT_NET_DEAD
,
1557 static void virtnet_get_ringparam(struct net_device
*dev
,
1558 struct ethtool_ringparam
*ring
)
1560 struct virtnet_info
*vi
= netdev_priv(dev
);
1562 ring
->rx_max_pending
= virtqueue_get_vring_size(vi
->rq
[0].vq
);
1563 ring
->tx_max_pending
= virtqueue_get_vring_size(vi
->sq
[0].vq
);
1564 ring
->rx_pending
= ring
->rx_max_pending
;
1565 ring
->tx_pending
= ring
->tx_max_pending
;
1569 static void virtnet_get_drvinfo(struct net_device
*dev
,
1570 struct ethtool_drvinfo
*info
)
1572 struct virtnet_info
*vi
= netdev_priv(dev
);
1573 struct virtio_device
*vdev
= vi
->vdev
;
1575 strlcpy(info
->driver
, KBUILD_MODNAME
, sizeof(info
->driver
));
1576 strlcpy(info
->version
, VIRTNET_DRIVER_VERSION
, sizeof(info
->version
));
1577 strlcpy(info
->bus_info
, virtio_bus_name(vdev
), sizeof(info
->bus_info
));
1581 /* TODO: Eliminate OOO packets during switching */
1582 static int virtnet_set_channels(struct net_device
*dev
,
1583 struct ethtool_channels
*channels
)
1585 struct virtnet_info
*vi
= netdev_priv(dev
);
1586 u16 queue_pairs
= channels
->combined_count
;
1589 /* We don't support separate rx/tx channels.
1590 * We don't allow setting 'other' channels.
1592 if (channels
->rx_count
|| channels
->tx_count
|| channels
->other_count
)
1595 if (queue_pairs
> vi
->max_queue_pairs
|| queue_pairs
== 0)
1598 /* For now we don't support modifying channels while XDP is loaded
1599 * also when XDP is loaded all RX queues have XDP programs so we only
1600 * need to check a single RX queue.
1602 if (vi
->rq
[0].xdp_prog
)
1606 err
= _virtnet_set_queues(vi
, queue_pairs
);
1608 netif_set_real_num_tx_queues(dev
, queue_pairs
);
1609 netif_set_real_num_rx_queues(dev
, queue_pairs
);
1611 virtnet_set_affinity(vi
);
1618 static void virtnet_get_channels(struct net_device
*dev
,
1619 struct ethtool_channels
*channels
)
1621 struct virtnet_info
*vi
= netdev_priv(dev
);
1623 channels
->combined_count
= vi
->curr_queue_pairs
;
1624 channels
->max_combined
= vi
->max_queue_pairs
;
1625 channels
->max_other
= 0;
1626 channels
->rx_count
= 0;
1627 channels
->tx_count
= 0;
1628 channels
->other_count
= 0;
1631 /* Check if the user is trying to change anything besides speed/duplex */
1632 static bool virtnet_validate_ethtool_cmd(const struct ethtool_cmd
*cmd
)
1634 struct ethtool_cmd diff1
= *cmd
;
1635 struct ethtool_cmd diff2
= {};
1637 /* cmd is always set so we need to clear it, validate the port type
1638 * and also without autonegotiation we can ignore advertising
1640 ethtool_cmd_speed_set(&diff1
, 0);
1641 diff2
.port
= PORT_OTHER
;
1642 diff1
.advertising
= 0;
1646 return !memcmp(&diff1
, &diff2
, sizeof(diff1
));
1649 static int virtnet_set_settings(struct net_device
*dev
, struct ethtool_cmd
*cmd
)
1651 struct virtnet_info
*vi
= netdev_priv(dev
);
1654 speed
= ethtool_cmd_speed(cmd
);
1655 /* don't allow custom speed and duplex */
1656 if (!ethtool_validate_speed(speed
) ||
1657 !ethtool_validate_duplex(cmd
->duplex
) ||
1658 !virtnet_validate_ethtool_cmd(cmd
))
1661 vi
->duplex
= cmd
->duplex
;
1666 static int virtnet_get_settings(struct net_device
*dev
, struct ethtool_cmd
*cmd
)
1668 struct virtnet_info
*vi
= netdev_priv(dev
);
1670 ethtool_cmd_speed_set(cmd
, vi
->speed
);
1671 cmd
->duplex
= vi
->duplex
;
1672 cmd
->port
= PORT_OTHER
;
1677 static void virtnet_init_settings(struct net_device
*dev
)
1679 struct virtnet_info
*vi
= netdev_priv(dev
);
1681 vi
->speed
= SPEED_UNKNOWN
;
1682 vi
->duplex
= DUPLEX_UNKNOWN
;
1685 static const struct ethtool_ops virtnet_ethtool_ops
= {
1686 .get_drvinfo
= virtnet_get_drvinfo
,
1687 .get_link
= ethtool_op_get_link
,
1688 .get_ringparam
= virtnet_get_ringparam
,
1689 .set_channels
= virtnet_set_channels
,
1690 .get_channels
= virtnet_get_channels
,
1691 .get_ts_info
= ethtool_op_get_ts_info
,
1692 .get_settings
= virtnet_get_settings
,
1693 .set_settings
= virtnet_set_settings
,
1696 static void virtnet_freeze_down(struct virtio_device
*vdev
)
1698 struct virtnet_info
*vi
= vdev
->priv
;
1701 /* Make sure no work handler is accessing the device */
1702 flush_work(&vi
->config_work
);
1704 netif_device_detach(vi
->dev
);
1705 cancel_delayed_work_sync(&vi
->refill
);
1707 if (netif_running(vi
->dev
)) {
1708 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
1709 napi_disable(&vi
->rq
[i
].napi
);
1713 static int init_vqs(struct virtnet_info
*vi
);
1714 static void _remove_vq_common(struct virtnet_info
*vi
);
1716 static int virtnet_restore_up(struct virtio_device
*vdev
)
1718 struct virtnet_info
*vi
= vdev
->priv
;
1725 virtio_device_ready(vdev
);
1727 if (netif_running(vi
->dev
)) {
1728 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++)
1729 if (!try_fill_recv(vi
, &vi
->rq
[i
], GFP_KERNEL
))
1730 schedule_delayed_work(&vi
->refill
, 0);
1732 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
1733 virtnet_napi_enable(&vi
->rq
[i
]);
1736 netif_device_attach(vi
->dev
);
1740 static int virtnet_reset(struct virtnet_info
*vi
, int curr_qp
, int xdp_qp
)
1742 struct virtio_device
*dev
= vi
->vdev
;
1745 virtio_config_disable(dev
);
1746 dev
->failed
= dev
->config
->get_status(dev
) & VIRTIO_CONFIG_S_FAILED
;
1747 virtnet_freeze_down(dev
);
1748 _remove_vq_common(vi
);
1750 virtio_add_status(dev
, VIRTIO_CONFIG_S_ACKNOWLEDGE
);
1751 virtio_add_status(dev
, VIRTIO_CONFIG_S_DRIVER
);
1753 ret
= virtio_finalize_features(dev
);
1757 vi
->xdp_queue_pairs
= xdp_qp
;
1758 ret
= virtnet_restore_up(dev
);
1761 ret
= _virtnet_set_queues(vi
, curr_qp
);
1765 virtio_add_status(dev
, VIRTIO_CONFIG_S_DRIVER_OK
);
1766 virtio_config_enable(dev
);
1769 virtio_add_status(dev
, VIRTIO_CONFIG_S_FAILED
);
1773 static int virtnet_xdp_set(struct net_device
*dev
, struct bpf_prog
*prog
)
1775 unsigned long int max_sz
= PAGE_SIZE
- sizeof(struct padded_vnet_hdr
);
1776 struct virtnet_info
*vi
= netdev_priv(dev
);
1777 struct bpf_prog
*old_prog
;
1778 u16 xdp_qp
= 0, curr_qp
;
1781 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_GUEST_TSO4
) ||
1782 virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_GUEST_TSO6
) ||
1783 virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_GUEST_ECN
) ||
1784 virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_GUEST_UFO
)) {
1785 netdev_warn(dev
, "can't set XDP while host is implementing LRO, disable LRO first\n");
1789 if (vi
->mergeable_rx_bufs
&& !vi
->any_header_sg
) {
1790 netdev_warn(dev
, "XDP expects header/data in single page, any_header_sg required\n");
1794 if (dev
->mtu
> max_sz
) {
1795 netdev_warn(dev
, "XDP requires MTU less than %lu\n", max_sz
);
1799 curr_qp
= vi
->curr_queue_pairs
- vi
->xdp_queue_pairs
;
1801 xdp_qp
= nr_cpu_ids
;
1803 /* XDP requires extra queues for XDP_TX */
1804 if (curr_qp
+ xdp_qp
> vi
->max_queue_pairs
) {
1805 netdev_warn(dev
, "request %i queues but max is %i\n",
1806 curr_qp
+ xdp_qp
, vi
->max_queue_pairs
);
1811 prog
= bpf_prog_add(prog
, vi
->max_queue_pairs
- 1);
1813 return PTR_ERR(prog
);
1816 /* Changing the headroom in buffers is a disruptive operation because
1817 * existing buffers must be flushed and reallocated. This will happen
1818 * when a xdp program is initially added or xdp is disabled by removing
1819 * the xdp program resulting in number of XDP queues changing.
1821 if (vi
->xdp_queue_pairs
!= xdp_qp
) {
1822 err
= virtnet_reset(vi
, curr_qp
+ xdp_qp
, xdp_qp
);
1824 dev_warn(&dev
->dev
, "XDP reset failure.\n");
1825 goto virtio_reset_err
;
1829 netif_set_real_num_rx_queues(dev
, curr_qp
+ xdp_qp
);
1831 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1832 old_prog
= rtnl_dereference(vi
->rq
[i
].xdp_prog
);
1833 rcu_assign_pointer(vi
->rq
[i
].xdp_prog
, prog
);
1835 bpf_prog_put(old_prog
);
1841 /* On reset error do our best to unwind XDP changes inflight and return
1842 * error up to user space for resolution. The underlying reset hung on
1843 * us so not much we can do here.
1846 bpf_prog_sub(prog
, vi
->max_queue_pairs
- 1);
1850 static bool virtnet_xdp_query(struct net_device
*dev
)
1852 struct virtnet_info
*vi
= netdev_priv(dev
);
1855 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1856 if (vi
->rq
[i
].xdp_prog
)
1862 static int virtnet_xdp(struct net_device
*dev
, struct netdev_xdp
*xdp
)
1864 switch (xdp
->command
) {
1865 case XDP_SETUP_PROG
:
1866 return virtnet_xdp_set(dev
, xdp
->prog
);
1867 case XDP_QUERY_PROG
:
1868 xdp
->prog_attached
= virtnet_xdp_query(dev
);
1875 static const struct net_device_ops virtnet_netdev
= {
1876 .ndo_open
= virtnet_open
,
1877 .ndo_stop
= virtnet_close
,
1878 .ndo_start_xmit
= start_xmit
,
1879 .ndo_validate_addr
= eth_validate_addr
,
1880 .ndo_set_mac_address
= virtnet_set_mac_address
,
1881 .ndo_set_rx_mode
= virtnet_set_rx_mode
,
1882 .ndo_get_stats64
= virtnet_stats
,
1883 .ndo_vlan_rx_add_vid
= virtnet_vlan_rx_add_vid
,
1884 .ndo_vlan_rx_kill_vid
= virtnet_vlan_rx_kill_vid
,
1885 #ifdef CONFIG_NET_POLL_CONTROLLER
1886 .ndo_poll_controller
= virtnet_netpoll
,
1888 .ndo_xdp
= virtnet_xdp
,
1891 static void virtnet_config_changed_work(struct work_struct
*work
)
1893 struct virtnet_info
*vi
=
1894 container_of(work
, struct virtnet_info
, config_work
);
1897 if (virtio_cread_feature(vi
->vdev
, VIRTIO_NET_F_STATUS
,
1898 struct virtio_net_config
, status
, &v
) < 0)
1901 if (v
& VIRTIO_NET_S_ANNOUNCE
) {
1902 netdev_notify_peers(vi
->dev
);
1903 virtnet_ack_link_announce(vi
);
1906 /* Ignore unknown (future) status bits */
1907 v
&= VIRTIO_NET_S_LINK_UP
;
1909 if (vi
->status
== v
)
1914 if (vi
->status
& VIRTIO_NET_S_LINK_UP
) {
1915 netif_carrier_on(vi
->dev
);
1916 netif_tx_wake_all_queues(vi
->dev
);
1918 netif_carrier_off(vi
->dev
);
1919 netif_tx_stop_all_queues(vi
->dev
);
1923 static void virtnet_config_changed(struct virtio_device
*vdev
)
1925 struct virtnet_info
*vi
= vdev
->priv
;
1927 schedule_work(&vi
->config_work
);
1930 static void virtnet_free_queues(struct virtnet_info
*vi
)
1934 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1935 napi_hash_del(&vi
->rq
[i
].napi
);
1936 netif_napi_del(&vi
->rq
[i
].napi
);
1939 /* We called napi_hash_del() before netif_napi_del(),
1940 * we need to respect an RCU grace period before freeing vi->rq
1948 static void _free_receive_bufs(struct virtnet_info
*vi
)
1950 struct bpf_prog
*old_prog
;
1953 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1954 while (vi
->rq
[i
].pages
)
1955 __free_pages(get_a_page(&vi
->rq
[i
], GFP_KERNEL
), 0);
1957 old_prog
= rtnl_dereference(vi
->rq
[i
].xdp_prog
);
1958 RCU_INIT_POINTER(vi
->rq
[i
].xdp_prog
, NULL
);
1960 bpf_prog_put(old_prog
);
1964 static void free_receive_bufs(struct virtnet_info
*vi
)
1967 _free_receive_bufs(vi
);
1971 static void free_receive_page_frags(struct virtnet_info
*vi
)
1974 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
1975 if (vi
->rq
[i
].alloc_frag
.page
)
1976 put_page(vi
->rq
[i
].alloc_frag
.page
);
1979 static bool is_xdp_raw_buffer_queue(struct virtnet_info
*vi
, int q
)
1981 if (q
< (vi
->curr_queue_pairs
- vi
->xdp_queue_pairs
))
1983 else if (q
< vi
->curr_queue_pairs
)
1989 static void free_unused_bufs(struct virtnet_info
*vi
)
1994 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1995 struct virtqueue
*vq
= vi
->sq
[i
].vq
;
1996 while ((buf
= virtqueue_detach_unused_buf(vq
)) != NULL
) {
1997 if (!is_xdp_raw_buffer_queue(vi
, i
))
2000 put_page(virt_to_head_page(buf
));
2004 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
2005 struct virtqueue
*vq
= vi
->rq
[i
].vq
;
2007 while ((buf
= virtqueue_detach_unused_buf(vq
)) != NULL
) {
2008 if (vi
->mergeable_rx_bufs
) {
2009 put_page(virt_to_head_page(buf
));
2010 } else if (vi
->big_packets
) {
2011 give_pages(&vi
->rq
[i
], buf
);
2013 put_page(virt_to_head_page(buf
));
2019 static void virtnet_del_vqs(struct virtnet_info
*vi
)
2021 struct virtio_device
*vdev
= vi
->vdev
;
2023 virtnet_clean_affinity(vi
, -1);
2025 vdev
->config
->del_vqs(vdev
);
2027 virtnet_free_queues(vi
);
2030 /* How large should a single buffer be so a queue full of these can fit at
2031 * least one full packet?
2032 * Logic below assumes the mergeable buffer header is used.
2034 static unsigned int mergeable_min_buf_len(struct virtnet_info
*vi
, struct virtqueue
*vq
)
2036 const unsigned int hdr_len
= sizeof(struct virtio_net_hdr_mrg_rxbuf
);
2037 unsigned int rq_size
= virtqueue_get_vring_size(vq
);
2038 unsigned int packet_len
= vi
->big_packets
? IP_MAX_MTU
: vi
->dev
->max_mtu
;
2039 unsigned int buf_len
= hdr_len
+ ETH_HLEN
+ VLAN_HLEN
+ packet_len
;
2040 unsigned int min_buf_len
= DIV_ROUND_UP(buf_len
, rq_size
);
2042 return max(min_buf_len
, hdr_len
);
2045 static int virtnet_find_vqs(struct virtnet_info
*vi
)
2047 vq_callback_t
**callbacks
;
2048 struct virtqueue
**vqs
;
2054 /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
2055 * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
2056 * possible control vq.
2058 total_vqs
= vi
->max_queue_pairs
* 2 +
2059 virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
);
2061 /* Allocate space for find_vqs parameters */
2062 vqs
= kzalloc(total_vqs
* sizeof(*vqs
), GFP_KERNEL
);
2065 callbacks
= kmalloc(total_vqs
* sizeof(*callbacks
), GFP_KERNEL
);
2068 names
= kmalloc(total_vqs
* sizeof(*names
), GFP_KERNEL
);
2071 if (vi
->mergeable_rx_bufs
) {
2072 ctx
= kzalloc(total_vqs
* sizeof(*ctx
), GFP_KERNEL
);
2079 /* Parameters for control virtqueue, if any */
2081 callbacks
[total_vqs
- 1] = NULL
;
2082 names
[total_vqs
- 1] = "control";
2085 /* Allocate/initialize parameters for send/receive virtqueues */
2086 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
2087 callbacks
[rxq2vq(i
)] = skb_recv_done
;
2088 callbacks
[txq2vq(i
)] = skb_xmit_done
;
2089 sprintf(vi
->rq
[i
].name
, "input.%d", i
);
2090 sprintf(vi
->sq
[i
].name
, "output.%d", i
);
2091 names
[rxq2vq(i
)] = vi
->rq
[i
].name
;
2092 names
[txq2vq(i
)] = vi
->sq
[i
].name
;
2094 ctx
[rxq2vq(i
)] = true;
2097 ret
= vi
->vdev
->config
->find_vqs(vi
->vdev
, total_vqs
, vqs
, callbacks
,
2103 vi
->cvq
= vqs
[total_vqs
- 1];
2104 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VLAN
))
2105 vi
->dev
->features
|= NETIF_F_HW_VLAN_CTAG_FILTER
;
2108 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
2109 vi
->rq
[i
].vq
= vqs
[rxq2vq(i
)];
2110 vi
->rq
[i
].min_buf_len
= mergeable_min_buf_len(vi
, vi
->rq
[i
].vq
);
2111 vi
->sq
[i
].vq
= vqs
[txq2vq(i
)];
2132 static int virtnet_alloc_queues(struct virtnet_info
*vi
)
2136 vi
->sq
= kzalloc(sizeof(*vi
->sq
) * vi
->max_queue_pairs
, GFP_KERNEL
);
2139 vi
->rq
= kzalloc(sizeof(*vi
->rq
) * vi
->max_queue_pairs
, GFP_KERNEL
);
2143 INIT_DELAYED_WORK(&vi
->refill
, refill_work
);
2144 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
2145 vi
->rq
[i
].pages
= NULL
;
2146 netif_napi_add(vi
->dev
, &vi
->rq
[i
].napi
, virtnet_poll
,
2149 sg_init_table(vi
->rq
[i
].sg
, ARRAY_SIZE(vi
->rq
[i
].sg
));
2150 ewma_pkt_len_init(&vi
->rq
[i
].mrg_avg_pkt_len
);
2151 sg_init_table(vi
->sq
[i
].sg
, ARRAY_SIZE(vi
->sq
[i
].sg
));
2162 static int init_vqs(struct virtnet_info
*vi
)
2166 /* Allocate send & receive queues */
2167 ret
= virtnet_alloc_queues(vi
);
2171 ret
= virtnet_find_vqs(vi
);
2176 virtnet_set_affinity(vi
);
2182 virtnet_free_queues(vi
);
2188 static ssize_t
mergeable_rx_buffer_size_show(struct netdev_rx_queue
*queue
,
2189 struct rx_queue_attribute
*attribute
, char *buf
)
2191 struct virtnet_info
*vi
= netdev_priv(queue
->dev
);
2192 unsigned int queue_index
= get_netdev_rx_queue_index(queue
);
2193 struct ewma_pkt_len
*avg
;
2195 BUG_ON(queue_index
>= vi
->max_queue_pairs
);
2196 avg
= &vi
->rq
[queue_index
].mrg_avg_pkt_len
;
2197 return sprintf(buf
, "%u\n",
2198 get_mergeable_buf_len(&vi
->rq
[queue_index
], avg
));
2201 static struct rx_queue_attribute mergeable_rx_buffer_size_attribute
=
2202 __ATTR_RO(mergeable_rx_buffer_size
);
2204 static struct attribute
*virtio_net_mrg_rx_attrs
[] = {
2205 &mergeable_rx_buffer_size_attribute
.attr
,
2209 static const struct attribute_group virtio_net_mrg_rx_group
= {
2210 .name
= "virtio_net",
2211 .attrs
= virtio_net_mrg_rx_attrs
2215 static bool virtnet_fail_on_feature(struct virtio_device
*vdev
,
2217 const char *fname
, const char *dname
)
2219 if (!virtio_has_feature(vdev
, fbit
))
2222 dev_err(&vdev
->dev
, "device advertises feature %s but not %s",
2228 #define VIRTNET_FAIL_ON(vdev, fbit, dbit) \
2229 virtnet_fail_on_feature(vdev, fbit, #fbit, dbit)
2231 static bool virtnet_validate_features(struct virtio_device
*vdev
)
2233 if (!virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
) &&
2234 (VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_CTRL_RX
,
2235 "VIRTIO_NET_F_CTRL_VQ") ||
2236 VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_CTRL_VLAN
,
2237 "VIRTIO_NET_F_CTRL_VQ") ||
2238 VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_GUEST_ANNOUNCE
,
2239 "VIRTIO_NET_F_CTRL_VQ") ||
2240 VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_MQ
, "VIRTIO_NET_F_CTRL_VQ") ||
2241 VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_CTRL_MAC_ADDR
,
2242 "VIRTIO_NET_F_CTRL_VQ"))) {
2249 #define MIN_MTU ETH_MIN_MTU
2250 #define MAX_MTU ETH_MAX_MTU
2252 static int virtnet_validate(struct virtio_device
*vdev
)
2254 if (!vdev
->config
->get
) {
2255 dev_err(&vdev
->dev
, "%s failure: config access disabled\n",
2260 if (!virtnet_validate_features(vdev
))
2263 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MTU
)) {
2264 int mtu
= virtio_cread16(vdev
,
2265 offsetof(struct virtio_net_config
,
2268 __virtio_clear_bit(vdev
, VIRTIO_NET_F_MTU
);
2274 static int virtnet_probe(struct virtio_device
*vdev
)
2277 struct net_device
*dev
;
2278 struct virtnet_info
*vi
;
2279 u16 max_queue_pairs
;
2282 /* Find if host supports multiqueue virtio_net device */
2283 err
= virtio_cread_feature(vdev
, VIRTIO_NET_F_MQ
,
2284 struct virtio_net_config
,
2285 max_virtqueue_pairs
, &max_queue_pairs
);
2287 /* We need at least 2 queue's */
2288 if (err
|| max_queue_pairs
< VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN
||
2289 max_queue_pairs
> VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX
||
2290 !virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
))
2291 max_queue_pairs
= 1;
2293 /* Allocate ourselves a network device with room for our info */
2294 dev
= alloc_etherdev_mq(sizeof(struct virtnet_info
), max_queue_pairs
);
2298 /* Set up network device as normal. */
2299 dev
->priv_flags
|= IFF_UNICAST_FLT
| IFF_LIVE_ADDR_CHANGE
;
2300 dev
->netdev_ops
= &virtnet_netdev
;
2301 dev
->features
= NETIF_F_HIGHDMA
;
2303 dev
->ethtool_ops
= &virtnet_ethtool_ops
;
2304 SET_NETDEV_DEV(dev
, &vdev
->dev
);
2306 /* Do we support "hardware" checksums? */
2307 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CSUM
)) {
2308 /* This opens up the world of extra features. */
2309 dev
->hw_features
|= NETIF_F_HW_CSUM
| NETIF_F_SG
;
2311 dev
->features
|= NETIF_F_HW_CSUM
| NETIF_F_SG
;
2313 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GSO
)) {
2314 dev
->hw_features
|= NETIF_F_TSO
| NETIF_F_UFO
2315 | NETIF_F_TSO_ECN
| NETIF_F_TSO6
;
2317 /* Individual feature bits: what can host handle? */
2318 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_TSO4
))
2319 dev
->hw_features
|= NETIF_F_TSO
;
2320 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_TSO6
))
2321 dev
->hw_features
|= NETIF_F_TSO6
;
2322 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_ECN
))
2323 dev
->hw_features
|= NETIF_F_TSO_ECN
;
2324 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_UFO
))
2325 dev
->hw_features
|= NETIF_F_UFO
;
2327 dev
->features
|= NETIF_F_GSO_ROBUST
;
2330 dev
->features
|= dev
->hw_features
& (NETIF_F_ALL_TSO
|NETIF_F_UFO
);
2331 /* (!csum && gso) case will be fixed by register_netdev() */
2333 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_CSUM
))
2334 dev
->features
|= NETIF_F_RXCSUM
;
2336 dev
->vlan_features
= dev
->features
;
2338 /* MTU range: 68 - 65535 */
2339 dev
->min_mtu
= MIN_MTU
;
2340 dev
->max_mtu
= MAX_MTU
;
2342 /* Configuration may specify what MAC to use. Otherwise random. */
2343 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MAC
))
2344 virtio_cread_bytes(vdev
,
2345 offsetof(struct virtio_net_config
, mac
),
2346 dev
->dev_addr
, dev
->addr_len
);
2348 eth_hw_addr_random(dev
);
2350 /* Set up our device-specific information */
2351 vi
= netdev_priv(dev
);
2355 vi
->stats
= alloc_percpu(struct virtnet_stats
);
2357 if (vi
->stats
== NULL
)
2360 for_each_possible_cpu(i
) {
2361 struct virtnet_stats
*virtnet_stats
;
2362 virtnet_stats
= per_cpu_ptr(vi
->stats
, i
);
2363 u64_stats_init(&virtnet_stats
->tx_syncp
);
2364 u64_stats_init(&virtnet_stats
->rx_syncp
);
2367 INIT_WORK(&vi
->config_work
, virtnet_config_changed_work
);
2369 /* If we can receive ANY GSO packets, we must allocate large ones. */
2370 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_TSO4
) ||
2371 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_TSO6
) ||
2372 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_ECN
) ||
2373 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_UFO
))
2374 vi
->big_packets
= true;
2376 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MRG_RXBUF
))
2377 vi
->mergeable_rx_bufs
= true;
2379 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MRG_RXBUF
) ||
2380 virtio_has_feature(vdev
, VIRTIO_F_VERSION_1
))
2381 vi
->hdr_len
= sizeof(struct virtio_net_hdr_mrg_rxbuf
);
2383 vi
->hdr_len
= sizeof(struct virtio_net_hdr
);
2385 if (virtio_has_feature(vdev
, VIRTIO_F_ANY_LAYOUT
) ||
2386 virtio_has_feature(vdev
, VIRTIO_F_VERSION_1
))
2387 vi
->any_header_sg
= true;
2389 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
))
2392 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MTU
)) {
2393 mtu
= virtio_cread16(vdev
,
2394 offsetof(struct virtio_net_config
,
2396 if (mtu
< dev
->min_mtu
) {
2397 /* Should never trigger: MTU was previously validated
2398 * in virtnet_validate.
2400 dev_err(&vdev
->dev
, "device MTU appears to have changed "
2401 "it is now %d < %d", mtu
, dev
->min_mtu
);
2408 /* TODO: size buffers correctly in this case. */
2409 if (dev
->mtu
> ETH_DATA_LEN
)
2410 vi
->big_packets
= true;
2413 if (vi
->any_header_sg
)
2414 dev
->needed_headroom
= vi
->hdr_len
;
2416 /* Enable multiqueue by default */
2417 if (num_online_cpus() >= max_queue_pairs
)
2418 vi
->curr_queue_pairs
= max_queue_pairs
;
2420 vi
->curr_queue_pairs
= num_online_cpus();
2421 vi
->max_queue_pairs
= max_queue_pairs
;
2423 /* Allocate/initialize the rx/tx queues, and invoke find_vqs */
2429 if (vi
->mergeable_rx_bufs
)
2430 dev
->sysfs_rx_queue_group
= &virtio_net_mrg_rx_group
;
2432 netif_set_real_num_tx_queues(dev
, vi
->curr_queue_pairs
);
2433 netif_set_real_num_rx_queues(dev
, vi
->curr_queue_pairs
);
2435 virtnet_init_settings(dev
);
2437 err
= register_netdev(dev
);
2439 pr_debug("virtio_net: registering device failed\n");
2443 virtio_device_ready(vdev
);
2445 err
= virtnet_cpu_notif_add(vi
);
2447 pr_debug("virtio_net: registering cpu notifier failed\n");
2448 goto free_unregister_netdev
;
2451 virtnet_set_queues(vi
, vi
->curr_queue_pairs
);
2453 /* Assume link up if device can't report link status,
2454 otherwise get link status from config. */
2455 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_STATUS
)) {
2456 netif_carrier_off(dev
);
2457 schedule_work(&vi
->config_work
);
2459 vi
->status
= VIRTIO_NET_S_LINK_UP
;
2460 netif_carrier_on(dev
);
2463 pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
2464 dev
->name
, max_queue_pairs
);
2468 free_unregister_netdev
:
2469 vi
->vdev
->config
->reset(vdev
);
2471 unregister_netdev(dev
);
2473 cancel_delayed_work_sync(&vi
->refill
);
2474 free_receive_page_frags(vi
);
2475 virtnet_del_vqs(vi
);
2477 free_percpu(vi
->stats
);
2483 static void _remove_vq_common(struct virtnet_info
*vi
)
2485 vi
->vdev
->config
->reset(vi
->vdev
);
2486 free_unused_bufs(vi
);
2487 _free_receive_bufs(vi
);
2488 free_receive_page_frags(vi
);
2489 virtnet_del_vqs(vi
);
2492 static void remove_vq_common(struct virtnet_info
*vi
)
2494 vi
->vdev
->config
->reset(vi
->vdev
);
2496 /* Free unused buffers in both send and recv, if any. */
2497 free_unused_bufs(vi
);
2499 free_receive_bufs(vi
);
2501 free_receive_page_frags(vi
);
2503 virtnet_del_vqs(vi
);
2506 static void virtnet_remove(struct virtio_device
*vdev
)
2508 struct virtnet_info
*vi
= vdev
->priv
;
2510 virtnet_cpu_notif_remove(vi
);
2512 /* Make sure no work handler is accessing the device. */
2513 flush_work(&vi
->config_work
);
2515 unregister_netdev(vi
->dev
);
2517 remove_vq_common(vi
);
2519 free_percpu(vi
->stats
);
2520 free_netdev(vi
->dev
);
2523 #ifdef CONFIG_PM_SLEEP
2524 static int virtnet_freeze(struct virtio_device
*vdev
)
2526 struct virtnet_info
*vi
= vdev
->priv
;
2528 virtnet_cpu_notif_remove(vi
);
2529 virtnet_freeze_down(vdev
);
2530 remove_vq_common(vi
);
2535 static int virtnet_restore(struct virtio_device
*vdev
)
2537 struct virtnet_info
*vi
= vdev
->priv
;
2540 err
= virtnet_restore_up(vdev
);
2543 virtnet_set_queues(vi
, vi
->curr_queue_pairs
);
2545 err
= virtnet_cpu_notif_add(vi
);
2553 static struct virtio_device_id id_table
[] = {
2554 { VIRTIO_ID_NET
, VIRTIO_DEV_ANY_ID
},
2558 #define VIRTNET_FEATURES \
2559 VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM, \
2561 VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6, \
2562 VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, \
2563 VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO, \
2564 VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ, \
2565 VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, \
2566 VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ, \
2567 VIRTIO_NET_F_CTRL_MAC_ADDR, \
2570 static unsigned int features
[] = {
2574 static unsigned int features_legacy
[] = {
2577 VIRTIO_F_ANY_LAYOUT
,
2580 static struct virtio_driver virtio_net_driver
= {
2581 .feature_table
= features
,
2582 .feature_table_size
= ARRAY_SIZE(features
),
2583 .feature_table_legacy
= features_legacy
,
2584 .feature_table_size_legacy
= ARRAY_SIZE(features_legacy
),
2585 .driver
.name
= KBUILD_MODNAME
,
2586 .driver
.owner
= THIS_MODULE
,
2587 .id_table
= id_table
,
2588 .validate
= virtnet_validate
,
2589 .probe
= virtnet_probe
,
2590 .remove
= virtnet_remove
,
2591 .config_changed
= virtnet_config_changed
,
2592 #ifdef CONFIG_PM_SLEEP
2593 .freeze
= virtnet_freeze
,
2594 .restore
= virtnet_restore
,
2598 static __init
int virtio_net_driver_init(void)
2602 ret
= cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN
, "virtio/net:online",
2604 virtnet_cpu_down_prep
);
2607 virtionet_online
= ret
;
2608 ret
= cpuhp_setup_state_multi(CPUHP_VIRT_NET_DEAD
, "virtio/net:dead",
2609 NULL
, virtnet_cpu_dead
);
2613 ret
= register_virtio_driver(&virtio_net_driver
);
2618 cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD
);
2620 cpuhp_remove_multi_state(virtionet_online
);
2624 module_init(virtio_net_driver_init
);
2626 static __exit
void virtio_net_driver_exit(void)
2628 cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD
);
2629 cpuhp_remove_multi_state(virtionet_online
);
2630 unregister_virtio_driver(&virtio_net_driver
);
2632 module_exit(virtio_net_driver_exit
);
2634 MODULE_DEVICE_TABLE(virtio
, id_table
);
2635 MODULE_DESCRIPTION("Virtio network driver");
2636 MODULE_LICENSE("GPL");