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/scatterlist.h>
26 #include <linux/if_vlan.h>
27 #include <linux/slab.h>
28 #include <linux/cpu.h>
29 #include <linux/average.h>
30 #include <net/busy_poll.h>
32 static int napi_weight
= NAPI_POLL_WEIGHT
;
33 module_param(napi_weight
, int, 0444);
35 static bool csum
= true, gso
= true;
36 module_param(csum
, bool, 0444);
37 module_param(gso
, bool, 0444);
39 /* FIXME: MTU in config. */
40 #define GOOD_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
41 #define GOOD_COPY_LEN 128
43 /* RX packet size EWMA. The average packet size is used to determine the packet
44 * buffer size when refilling RX rings. As the entire RX ring may be refilled
45 * at once, the weight is chosen so that the EWMA will be insensitive to short-
46 * term, transient changes in packet size.
48 DECLARE_EWMA(pkt_len
, 1, 64)
50 /* Minimum alignment for mergeable packet buffers. */
51 #define MERGEABLE_BUFFER_ALIGN max(L1_CACHE_BYTES, 256)
53 #define VIRTNET_DRIVER_VERSION "1.0.0"
55 struct virtnet_stats
{
56 struct u64_stats_sync tx_syncp
;
57 struct u64_stats_sync rx_syncp
;
65 /* Internal representation of a send virtqueue */
67 /* Virtqueue associated with this send _queue */
70 /* TX: fragments + linear part + virtio header */
71 struct scatterlist sg
[MAX_SKB_FRAGS
+ 2];
73 /* Name of the send queue: output.$index */
77 /* Internal representation of a receive virtqueue */
78 struct receive_queue
{
79 /* Virtqueue associated with this receive_queue */
82 struct napi_struct napi
;
84 /* Chain pages by the private ptr. */
87 /* Average packet length for mergeable receive buffers. */
88 struct ewma_pkt_len mrg_avg_pkt_len
;
90 /* Page frag for packet buffer allocation. */
91 struct page_frag alloc_frag
;
93 /* RX: fragments + linear part + virtio header */
94 struct scatterlist sg
[MAX_SKB_FRAGS
+ 2];
96 /* Name of this receive queue: input.$index */
100 struct virtnet_info
{
101 struct virtio_device
*vdev
;
102 struct virtqueue
*cvq
;
103 struct net_device
*dev
;
104 struct send_queue
*sq
;
105 struct receive_queue
*rq
;
108 /* Max # of queue pairs supported by the device */
111 /* # of queue pairs currently used by the driver */
112 u16 curr_queue_pairs
;
114 /* I like... big packets and I cannot lie! */
117 /* Host will merge rx buffers for big packets (shake it! shake it!) */
118 bool mergeable_rx_bufs
;
120 /* Has control virtqueue */
123 /* Host can handle any s/g split between our header and packet data */
126 /* Packet virtio header size */
129 /* Active statistics */
130 struct virtnet_stats __percpu
*stats
;
132 /* Work struct for refilling if we run low on memory. */
133 struct delayed_work refill
;
135 /* Work struct for config space updates */
136 struct work_struct config_work
;
138 /* Does the affinity hint is set for virtqueues? */
139 bool affinity_hint_set
;
141 /* CPU hot plug notifier */
142 struct notifier_block nb
;
144 /* Control VQ buffers: protected by the rtnl lock */
145 struct virtio_net_ctrl_hdr ctrl_hdr
;
146 virtio_net_ctrl_ack ctrl_status
;
147 struct virtio_net_ctrl_mq ctrl_mq
;
152 /* Ethtool settings */
157 struct padded_vnet_hdr
{
158 struct virtio_net_hdr_mrg_rxbuf hdr
;
160 * hdr is in a separate sg buffer, and data sg buffer shares same page
161 * with this header sg. This padding makes next sg 16 byte aligned
167 /* Converting between virtqueue no. and kernel tx/rx queue no.
168 * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq
170 static int vq2txq(struct virtqueue
*vq
)
172 return (vq
->index
- 1) / 2;
175 static int txq2vq(int txq
)
180 static int vq2rxq(struct virtqueue
*vq
)
182 return vq
->index
/ 2;
185 static int rxq2vq(int rxq
)
190 static inline struct virtio_net_hdr_mrg_rxbuf
*skb_vnet_hdr(struct sk_buff
*skb
)
192 return (struct virtio_net_hdr_mrg_rxbuf
*)skb
->cb
;
196 * private is used to chain pages for big packets, put the whole
197 * most recent used list in the beginning for reuse
199 static void give_pages(struct receive_queue
*rq
, struct page
*page
)
203 /* Find end of list, sew whole thing into vi->rq.pages. */
204 for (end
= page
; end
->private; end
= (struct page
*)end
->private);
205 end
->private = (unsigned long)rq
->pages
;
209 static struct page
*get_a_page(struct receive_queue
*rq
, gfp_t gfp_mask
)
211 struct page
*p
= rq
->pages
;
214 rq
->pages
= (struct page
*)p
->private;
215 /* clear private here, it is used to chain pages */
218 p
= alloc_page(gfp_mask
);
222 static void skb_xmit_done(struct virtqueue
*vq
)
224 struct virtnet_info
*vi
= vq
->vdev
->priv
;
226 /* Suppress further interrupts. */
227 virtqueue_disable_cb(vq
);
229 /* We were probably waiting for more output buffers. */
230 netif_wake_subqueue(vi
->dev
, vq2txq(vq
));
233 static unsigned int mergeable_ctx_to_buf_truesize(unsigned long mrg_ctx
)
235 unsigned int truesize
= mrg_ctx
& (MERGEABLE_BUFFER_ALIGN
- 1);
236 return (truesize
+ 1) * MERGEABLE_BUFFER_ALIGN
;
239 static void *mergeable_ctx_to_buf_address(unsigned long mrg_ctx
)
241 return (void *)(mrg_ctx
& -MERGEABLE_BUFFER_ALIGN
);
245 static unsigned long mergeable_buf_to_ctx(void *buf
, unsigned int truesize
)
247 unsigned int size
= truesize
/ MERGEABLE_BUFFER_ALIGN
;
248 return (unsigned long)buf
| (size
- 1);
251 /* Called from bottom half context */
252 static struct sk_buff
*page_to_skb(struct virtnet_info
*vi
,
253 struct receive_queue
*rq
,
254 struct page
*page
, unsigned int offset
,
255 unsigned int len
, unsigned int truesize
)
258 struct virtio_net_hdr_mrg_rxbuf
*hdr
;
259 unsigned int copy
, hdr_len
, hdr_padded_len
;
262 p
= page_address(page
) + offset
;
264 /* copy small packet so we can reuse these pages for small data */
265 skb
= napi_alloc_skb(&rq
->napi
, GOOD_COPY_LEN
);
269 hdr
= skb_vnet_hdr(skb
);
271 hdr_len
= vi
->hdr_len
;
272 if (vi
->mergeable_rx_bufs
)
273 hdr_padded_len
= sizeof *hdr
;
275 hdr_padded_len
= sizeof(struct padded_vnet_hdr
);
277 memcpy(hdr
, p
, hdr_len
);
280 offset
+= hdr_padded_len
;
284 if (copy
> skb_tailroom(skb
))
285 copy
= skb_tailroom(skb
);
286 memcpy(skb_put(skb
, copy
), p
, copy
);
291 if (vi
->mergeable_rx_bufs
) {
293 skb_add_rx_frag(skb
, 0, page
, offset
, len
, truesize
);
300 * Verify that we can indeed put this data into a skb.
301 * This is here to handle cases when the device erroneously
302 * tries to receive more than is possible. This is usually
303 * the case of a broken device.
305 if (unlikely(len
> MAX_SKB_FRAGS
* PAGE_SIZE
)) {
306 net_dbg_ratelimited("%s: too much data\n", skb
->dev
->name
);
310 BUG_ON(offset
>= PAGE_SIZE
);
312 unsigned int frag_size
= min((unsigned)PAGE_SIZE
- offset
, len
);
313 skb_add_rx_frag(skb
, skb_shinfo(skb
)->nr_frags
, page
, offset
,
314 frag_size
, truesize
);
316 page
= (struct page
*)page
->private;
321 give_pages(rq
, page
);
326 static struct sk_buff
*receive_small(struct virtnet_info
*vi
, void *buf
, unsigned int len
)
328 struct sk_buff
* skb
= buf
;
336 static struct sk_buff
*receive_big(struct net_device
*dev
,
337 struct virtnet_info
*vi
,
338 struct receive_queue
*rq
,
342 struct page
*page
= buf
;
343 struct sk_buff
*skb
= page_to_skb(vi
, rq
, page
, 0, len
, PAGE_SIZE
);
351 dev
->stats
.rx_dropped
++;
352 give_pages(rq
, page
);
356 static struct sk_buff
*receive_mergeable(struct net_device
*dev
,
357 struct virtnet_info
*vi
,
358 struct receive_queue
*rq
,
362 void *buf
= mergeable_ctx_to_buf_address(ctx
);
363 struct virtio_net_hdr_mrg_rxbuf
*hdr
= buf
;
364 u16 num_buf
= virtio16_to_cpu(vi
->vdev
, hdr
->num_buffers
);
365 struct page
*page
= virt_to_head_page(buf
);
366 int offset
= buf
- page_address(page
);
367 unsigned int truesize
= max(len
, mergeable_ctx_to_buf_truesize(ctx
));
369 struct sk_buff
*head_skb
= page_to_skb(vi
, rq
, page
, offset
, len
,
371 struct sk_buff
*curr_skb
= head_skb
;
373 if (unlikely(!curr_skb
))
378 ctx
= (unsigned long)virtqueue_get_buf(rq
->vq
, &len
);
379 if (unlikely(!ctx
)) {
380 pr_debug("%s: rx error: %d buffers out of %d missing\n",
382 virtio16_to_cpu(vi
->vdev
,
384 dev
->stats
.rx_length_errors
++;
388 buf
= mergeable_ctx_to_buf_address(ctx
);
389 page
= virt_to_head_page(buf
);
391 num_skb_frags
= skb_shinfo(curr_skb
)->nr_frags
;
392 if (unlikely(num_skb_frags
== MAX_SKB_FRAGS
)) {
393 struct sk_buff
*nskb
= alloc_skb(0, GFP_ATOMIC
);
397 if (curr_skb
== head_skb
)
398 skb_shinfo(curr_skb
)->frag_list
= nskb
;
400 curr_skb
->next
= nskb
;
402 head_skb
->truesize
+= nskb
->truesize
;
405 truesize
= max(len
, mergeable_ctx_to_buf_truesize(ctx
));
406 if (curr_skb
!= head_skb
) {
407 head_skb
->data_len
+= len
;
408 head_skb
->len
+= len
;
409 head_skb
->truesize
+= truesize
;
411 offset
= buf
- page_address(page
);
412 if (skb_can_coalesce(curr_skb
, num_skb_frags
, page
, offset
)) {
414 skb_coalesce_rx_frag(curr_skb
, num_skb_frags
- 1,
417 skb_add_rx_frag(curr_skb
, num_skb_frags
, page
,
418 offset
, len
, truesize
);
422 ewma_pkt_len_add(&rq
->mrg_avg_pkt_len
, head_skb
->len
);
428 ctx
= (unsigned long)virtqueue_get_buf(rq
->vq
, &len
);
429 if (unlikely(!ctx
)) {
430 pr_debug("%s: rx error: %d buffers missing\n",
432 dev
->stats
.rx_length_errors
++;
435 page
= virt_to_head_page(mergeable_ctx_to_buf_address(ctx
));
439 dev
->stats
.rx_dropped
++;
440 dev_kfree_skb(head_skb
);
444 static void receive_buf(struct virtnet_info
*vi
, struct receive_queue
*rq
,
445 void *buf
, unsigned int len
)
447 struct net_device
*dev
= vi
->dev
;
448 struct virtnet_stats
*stats
= this_cpu_ptr(vi
->stats
);
450 struct virtio_net_hdr_mrg_rxbuf
*hdr
;
452 if (unlikely(len
< vi
->hdr_len
+ ETH_HLEN
)) {
453 pr_debug("%s: short packet %i\n", dev
->name
, len
);
454 dev
->stats
.rx_length_errors
++;
455 if (vi
->mergeable_rx_bufs
) {
456 unsigned long ctx
= (unsigned long)buf
;
457 void *base
= mergeable_ctx_to_buf_address(ctx
);
458 put_page(virt_to_head_page(base
));
459 } else if (vi
->big_packets
) {
467 if (vi
->mergeable_rx_bufs
)
468 skb
= receive_mergeable(dev
, vi
, rq
, (unsigned long)buf
, len
);
469 else if (vi
->big_packets
)
470 skb
= receive_big(dev
, vi
, rq
, buf
, len
);
472 skb
= receive_small(vi
, buf
, len
);
477 hdr
= skb_vnet_hdr(skb
);
479 u64_stats_update_begin(&stats
->rx_syncp
);
480 stats
->rx_bytes
+= skb
->len
;
482 u64_stats_update_end(&stats
->rx_syncp
);
484 if (hdr
->hdr
.flags
& VIRTIO_NET_HDR_F_DATA_VALID
)
485 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
487 if (virtio_net_hdr_to_skb(skb
, &hdr
->hdr
,
488 virtio_is_little_endian(vi
->vdev
))) {
489 net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n",
490 dev
->name
, hdr
->hdr
.gso_type
,
495 skb
->protocol
= eth_type_trans(skb
, dev
);
496 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
497 ntohs(skb
->protocol
), skb
->len
, skb
->pkt_type
);
499 napi_gro_receive(&rq
->napi
, skb
);
503 dev
->stats
.rx_frame_errors
++;
507 static int add_recvbuf_small(struct virtnet_info
*vi
, struct receive_queue
*rq
,
511 struct virtio_net_hdr_mrg_rxbuf
*hdr
;
514 skb
= __netdev_alloc_skb_ip_align(vi
->dev
, GOOD_PACKET_LEN
, gfp
);
518 skb_put(skb
, GOOD_PACKET_LEN
);
520 hdr
= skb_vnet_hdr(skb
);
521 sg_init_table(rq
->sg
, 2);
522 sg_set_buf(rq
->sg
, hdr
, vi
->hdr_len
);
523 skb_to_sgvec(skb
, rq
->sg
+ 1, 0, skb
->len
);
525 err
= virtqueue_add_inbuf(rq
->vq
, rq
->sg
, 2, skb
, gfp
);
532 static int add_recvbuf_big(struct virtnet_info
*vi
, struct receive_queue
*rq
,
535 struct page
*first
, *list
= NULL
;
539 sg_init_table(rq
->sg
, MAX_SKB_FRAGS
+ 2);
541 /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */
542 for (i
= MAX_SKB_FRAGS
+ 1; i
> 1; --i
) {
543 first
= get_a_page(rq
, gfp
);
546 give_pages(rq
, list
);
549 sg_set_buf(&rq
->sg
[i
], page_address(first
), PAGE_SIZE
);
551 /* chain new page in list head to match sg */
552 first
->private = (unsigned long)list
;
556 first
= get_a_page(rq
, gfp
);
558 give_pages(rq
, list
);
561 p
= page_address(first
);
563 /* rq->sg[0], rq->sg[1] share the same page */
564 /* a separated rq->sg[0] for header - required in case !any_header_sg */
565 sg_set_buf(&rq
->sg
[0], p
, vi
->hdr_len
);
567 /* rq->sg[1] for data packet, from offset */
568 offset
= sizeof(struct padded_vnet_hdr
);
569 sg_set_buf(&rq
->sg
[1], p
+ offset
, PAGE_SIZE
- offset
);
571 /* chain first in list head */
572 first
->private = (unsigned long)list
;
573 err
= virtqueue_add_inbuf(rq
->vq
, rq
->sg
, MAX_SKB_FRAGS
+ 2,
576 give_pages(rq
, first
);
581 static unsigned int get_mergeable_buf_len(struct ewma_pkt_len
*avg_pkt_len
)
583 const size_t hdr_len
= sizeof(struct virtio_net_hdr_mrg_rxbuf
);
586 len
= hdr_len
+ clamp_t(unsigned int, ewma_pkt_len_read(avg_pkt_len
),
587 GOOD_PACKET_LEN
, PAGE_SIZE
- hdr_len
);
588 return ALIGN(len
, MERGEABLE_BUFFER_ALIGN
);
591 static int add_recvbuf_mergeable(struct receive_queue
*rq
, gfp_t gfp
)
593 struct page_frag
*alloc_frag
= &rq
->alloc_frag
;
597 unsigned int len
, hole
;
599 len
= get_mergeable_buf_len(&rq
->mrg_avg_pkt_len
);
600 if (unlikely(!skb_page_frag_refill(len
, alloc_frag
, gfp
)))
603 buf
= (char *)page_address(alloc_frag
->page
) + alloc_frag
->offset
;
604 ctx
= mergeable_buf_to_ctx(buf
, len
);
605 get_page(alloc_frag
->page
);
606 alloc_frag
->offset
+= len
;
607 hole
= alloc_frag
->size
- alloc_frag
->offset
;
609 /* To avoid internal fragmentation, if there is very likely not
610 * enough space for another buffer, add the remaining space to
611 * the current buffer. This extra space is not included in
612 * the truesize stored in ctx.
615 alloc_frag
->offset
+= hole
;
618 sg_init_one(rq
->sg
, buf
, len
);
619 err
= virtqueue_add_inbuf(rq
->vq
, rq
->sg
, 1, (void *)ctx
, gfp
);
621 put_page(virt_to_head_page(buf
));
627 * Returns false if we couldn't fill entirely (OOM).
629 * Normally run in the receive path, but can also be run from ndo_open
630 * before we're receiving packets, or from refill_work which is
631 * careful to disable receiving (using napi_disable).
633 static bool try_fill_recv(struct virtnet_info
*vi
, struct receive_queue
*rq
,
641 if (vi
->mergeable_rx_bufs
)
642 err
= add_recvbuf_mergeable(rq
, gfp
);
643 else if (vi
->big_packets
)
644 err
= add_recvbuf_big(vi
, rq
, gfp
);
646 err
= add_recvbuf_small(vi
, rq
, gfp
);
648 oom
= err
== -ENOMEM
;
651 } while (rq
->vq
->num_free
);
652 virtqueue_kick(rq
->vq
);
656 static void skb_recv_done(struct virtqueue
*rvq
)
658 struct virtnet_info
*vi
= rvq
->vdev
->priv
;
659 struct receive_queue
*rq
= &vi
->rq
[vq2rxq(rvq
)];
661 /* Schedule NAPI, Suppress further interrupts if successful. */
662 if (napi_schedule_prep(&rq
->napi
)) {
663 virtqueue_disable_cb(rvq
);
664 __napi_schedule(&rq
->napi
);
668 static void virtnet_napi_enable(struct receive_queue
*rq
)
670 napi_enable(&rq
->napi
);
672 /* If all buffers were filled by other side before we napi_enabled, we
673 * won't get another interrupt, so process any outstanding packets
674 * now. virtnet_poll wants re-enable the queue, so we disable here.
675 * We synchronize against interrupts via NAPI_STATE_SCHED */
676 if (napi_schedule_prep(&rq
->napi
)) {
677 virtqueue_disable_cb(rq
->vq
);
679 __napi_schedule(&rq
->napi
);
684 static void refill_work(struct work_struct
*work
)
686 struct virtnet_info
*vi
=
687 container_of(work
, struct virtnet_info
, refill
.work
);
691 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++) {
692 struct receive_queue
*rq
= &vi
->rq
[i
];
694 napi_disable(&rq
->napi
);
695 still_empty
= !try_fill_recv(vi
, rq
, GFP_KERNEL
);
696 virtnet_napi_enable(rq
);
698 /* In theory, this can happen: if we don't get any buffers in
699 * we will *never* try to fill again.
702 schedule_delayed_work(&vi
->refill
, HZ
/2);
706 static int virtnet_receive(struct receive_queue
*rq
, int budget
)
708 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
709 unsigned int len
, received
= 0;
712 while (received
< budget
&&
713 (buf
= virtqueue_get_buf(rq
->vq
, &len
)) != NULL
) {
714 receive_buf(vi
, rq
, buf
, len
);
718 if (rq
->vq
->num_free
> virtqueue_get_vring_size(rq
->vq
) / 2) {
719 if (!try_fill_recv(vi
, rq
, GFP_ATOMIC
))
720 schedule_delayed_work(&vi
->refill
, 0);
726 static int virtnet_poll(struct napi_struct
*napi
, int budget
)
728 struct receive_queue
*rq
=
729 container_of(napi
, struct receive_queue
, napi
);
730 unsigned int r
, received
;
732 received
= virtnet_receive(rq
, budget
);
734 /* Out of packets? */
735 if (received
< budget
) {
736 r
= virtqueue_enable_cb_prepare(rq
->vq
);
737 napi_complete_done(napi
, received
);
738 if (unlikely(virtqueue_poll(rq
->vq
, r
)) &&
739 napi_schedule_prep(napi
)) {
740 virtqueue_disable_cb(rq
->vq
);
741 __napi_schedule(napi
);
748 #ifdef CONFIG_NET_RX_BUSY_POLL
749 /* must be called with local_bh_disable()d */
750 static int virtnet_busy_poll(struct napi_struct
*napi
)
752 struct receive_queue
*rq
=
753 container_of(napi
, struct receive_queue
, napi
);
754 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
755 int r
, received
= 0, budget
= 4;
757 if (!(vi
->status
& VIRTIO_NET_S_LINK_UP
))
758 return LL_FLUSH_FAILED
;
760 if (!napi_schedule_prep(napi
))
761 return LL_FLUSH_BUSY
;
763 virtqueue_disable_cb(rq
->vq
);
766 received
+= virtnet_receive(rq
, budget
);
768 r
= virtqueue_enable_cb_prepare(rq
->vq
);
769 clear_bit(NAPI_STATE_SCHED
, &napi
->state
);
770 if (unlikely(virtqueue_poll(rq
->vq
, r
)) &&
771 napi_schedule_prep(napi
)) {
772 virtqueue_disable_cb(rq
->vq
);
773 if (received
< budget
) {
777 __napi_schedule(napi
);
783 #endif /* CONFIG_NET_RX_BUSY_POLL */
785 static int virtnet_open(struct net_device
*dev
)
787 struct virtnet_info
*vi
= netdev_priv(dev
);
790 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
791 if (i
< vi
->curr_queue_pairs
)
792 /* Make sure we have some buffers: if oom use wq. */
793 if (!try_fill_recv(vi
, &vi
->rq
[i
], GFP_KERNEL
))
794 schedule_delayed_work(&vi
->refill
, 0);
795 virtnet_napi_enable(&vi
->rq
[i
]);
801 static void free_old_xmit_skbs(struct send_queue
*sq
)
805 struct virtnet_info
*vi
= sq
->vq
->vdev
->priv
;
806 struct virtnet_stats
*stats
= this_cpu_ptr(vi
->stats
);
808 while ((skb
= virtqueue_get_buf(sq
->vq
, &len
)) != NULL
) {
809 pr_debug("Sent skb %p\n", skb
);
811 u64_stats_update_begin(&stats
->tx_syncp
);
812 stats
->tx_bytes
+= skb
->len
;
814 u64_stats_update_end(&stats
->tx_syncp
);
816 dev_kfree_skb_any(skb
);
820 static int xmit_skb(struct send_queue
*sq
, struct sk_buff
*skb
)
822 struct virtio_net_hdr_mrg_rxbuf
*hdr
;
823 const unsigned char *dest
= ((struct ethhdr
*)skb
->data
)->h_dest
;
824 struct virtnet_info
*vi
= sq
->vq
->vdev
->priv
;
826 unsigned hdr_len
= vi
->hdr_len
;
829 pr_debug("%s: xmit %p %pM\n", vi
->dev
->name
, skb
, dest
);
831 can_push
= vi
->any_header_sg
&&
832 !((unsigned long)skb
->data
& (__alignof__(*hdr
) - 1)) &&
833 !skb_header_cloned(skb
) && skb_headroom(skb
) >= hdr_len
;
834 /* Even if we can, don't push here yet as this would skew
835 * csum_start offset below. */
837 hdr
= (struct virtio_net_hdr_mrg_rxbuf
*)(skb
->data
- hdr_len
);
839 hdr
= skb_vnet_hdr(skb
);
841 if (virtio_net_hdr_from_skb(skb
, &hdr
->hdr
,
842 virtio_is_little_endian(vi
->vdev
)))
845 if (vi
->mergeable_rx_bufs
)
846 hdr
->num_buffers
= 0;
848 sg_init_table(sq
->sg
, skb_shinfo(skb
)->nr_frags
+ (can_push
? 1 : 2));
850 __skb_push(skb
, hdr_len
);
851 num_sg
= skb_to_sgvec(skb
, sq
->sg
, 0, skb
->len
);
852 /* Pull header back to avoid skew in tx bytes calculations. */
853 __skb_pull(skb
, hdr_len
);
855 sg_set_buf(sq
->sg
, hdr
, hdr_len
);
856 num_sg
= skb_to_sgvec(skb
, sq
->sg
+ 1, 0, skb
->len
) + 1;
858 return virtqueue_add_outbuf(sq
->vq
, sq
->sg
, num_sg
, skb
, GFP_ATOMIC
);
861 static netdev_tx_t
start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
863 struct virtnet_info
*vi
= netdev_priv(dev
);
864 int qnum
= skb_get_queue_mapping(skb
);
865 struct send_queue
*sq
= &vi
->sq
[qnum
];
867 struct netdev_queue
*txq
= netdev_get_tx_queue(dev
, qnum
);
868 bool kick
= !skb
->xmit_more
;
870 /* Free up any pending old buffers before queueing new ones. */
871 free_old_xmit_skbs(sq
);
873 /* timestamp packet in software */
874 skb_tx_timestamp(skb
);
876 /* Try to transmit */
877 err
= xmit_skb(sq
, skb
);
879 /* This should not happen! */
881 dev
->stats
.tx_fifo_errors
++;
884 "Unexpected TXQ (%d) queue failure: %d\n", qnum
, err
);
885 dev
->stats
.tx_dropped
++;
886 dev_kfree_skb_any(skb
);
890 /* Don't wait up for transmitted skbs to be freed. */
894 /* If running out of space, stop queue to avoid getting packets that we
895 * are then unable to transmit.
896 * An alternative would be to force queuing layer to requeue the skb by
897 * returning NETDEV_TX_BUSY. However, NETDEV_TX_BUSY should not be
898 * returned in a normal path of operation: it means that driver is not
899 * maintaining the TX queue stop/start state properly, and causes
900 * the stack to do a non-trivial amount of useless work.
901 * Since most packets only take 1 or 2 ring slots, stopping the queue
902 * early means 16 slots are typically wasted.
904 if (sq
->vq
->num_free
< 2+MAX_SKB_FRAGS
) {
905 netif_stop_subqueue(dev
, qnum
);
906 if (unlikely(!virtqueue_enable_cb_delayed(sq
->vq
))) {
907 /* More just got used, free them then recheck. */
908 free_old_xmit_skbs(sq
);
909 if (sq
->vq
->num_free
>= 2+MAX_SKB_FRAGS
) {
910 netif_start_subqueue(dev
, qnum
);
911 virtqueue_disable_cb(sq
->vq
);
916 if (kick
|| netif_xmit_stopped(txq
))
917 virtqueue_kick(sq
->vq
);
923 * Send command via the control virtqueue and check status. Commands
924 * supported by the hypervisor, as indicated by feature bits, should
925 * never fail unless improperly formatted.
927 static bool virtnet_send_command(struct virtnet_info
*vi
, u8
class, u8 cmd
,
928 struct scatterlist
*out
)
930 struct scatterlist
*sgs
[4], hdr
, stat
;
931 unsigned out_num
= 0, tmp
;
933 /* Caller should know better */
934 BUG_ON(!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
));
936 vi
->ctrl_status
= ~0;
937 vi
->ctrl_hdr
.class = class;
938 vi
->ctrl_hdr
.cmd
= cmd
;
940 sg_init_one(&hdr
, &vi
->ctrl_hdr
, sizeof(vi
->ctrl_hdr
));
941 sgs
[out_num
++] = &hdr
;
944 sgs
[out_num
++] = out
;
946 /* Add return status. */
947 sg_init_one(&stat
, &vi
->ctrl_status
, sizeof(vi
->ctrl_status
));
948 sgs
[out_num
] = &stat
;
950 BUG_ON(out_num
+ 1 > ARRAY_SIZE(sgs
));
951 virtqueue_add_sgs(vi
->cvq
, sgs
, out_num
, 1, vi
, GFP_ATOMIC
);
953 if (unlikely(!virtqueue_kick(vi
->cvq
)))
954 return vi
->ctrl_status
== VIRTIO_NET_OK
;
956 /* Spin for a response, the kick causes an ioport write, trapping
957 * into the hypervisor, so the request should be handled immediately.
959 while (!virtqueue_get_buf(vi
->cvq
, &tmp
) &&
960 !virtqueue_is_broken(vi
->cvq
))
963 return vi
->ctrl_status
== VIRTIO_NET_OK
;
966 static int virtnet_set_mac_address(struct net_device
*dev
, void *p
)
968 struct virtnet_info
*vi
= netdev_priv(dev
);
969 struct virtio_device
*vdev
= vi
->vdev
;
971 struct sockaddr
*addr
= p
;
972 struct scatterlist sg
;
974 ret
= eth_prepare_mac_addr_change(dev
, p
);
978 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_MAC_ADDR
)) {
979 sg_init_one(&sg
, addr
->sa_data
, dev
->addr_len
);
980 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MAC
,
981 VIRTIO_NET_CTRL_MAC_ADDR_SET
, &sg
)) {
983 "Failed to set mac address by vq command.\n");
986 } else if (virtio_has_feature(vdev
, VIRTIO_NET_F_MAC
) &&
987 !virtio_has_feature(vdev
, VIRTIO_F_VERSION_1
)) {
990 /* Naturally, this has an atomicity problem. */
991 for (i
= 0; i
< dev
->addr_len
; i
++)
993 offsetof(struct virtio_net_config
, mac
) +
994 i
, addr
->sa_data
[i
]);
997 eth_commit_mac_addr_change(dev
, p
);
1002 static struct rtnl_link_stats64
*virtnet_stats(struct net_device
*dev
,
1003 struct rtnl_link_stats64
*tot
)
1005 struct virtnet_info
*vi
= netdev_priv(dev
);
1009 for_each_possible_cpu(cpu
) {
1010 struct virtnet_stats
*stats
= per_cpu_ptr(vi
->stats
, cpu
);
1011 u64 tpackets
, tbytes
, rpackets
, rbytes
;
1014 start
= u64_stats_fetch_begin_irq(&stats
->tx_syncp
);
1015 tpackets
= stats
->tx_packets
;
1016 tbytes
= stats
->tx_bytes
;
1017 } while (u64_stats_fetch_retry_irq(&stats
->tx_syncp
, start
));
1020 start
= u64_stats_fetch_begin_irq(&stats
->rx_syncp
);
1021 rpackets
= stats
->rx_packets
;
1022 rbytes
= stats
->rx_bytes
;
1023 } while (u64_stats_fetch_retry_irq(&stats
->rx_syncp
, start
));
1025 tot
->rx_packets
+= rpackets
;
1026 tot
->tx_packets
+= tpackets
;
1027 tot
->rx_bytes
+= rbytes
;
1028 tot
->tx_bytes
+= tbytes
;
1031 tot
->tx_dropped
= dev
->stats
.tx_dropped
;
1032 tot
->tx_fifo_errors
= dev
->stats
.tx_fifo_errors
;
1033 tot
->rx_dropped
= dev
->stats
.rx_dropped
;
1034 tot
->rx_length_errors
= dev
->stats
.rx_length_errors
;
1035 tot
->rx_frame_errors
= dev
->stats
.rx_frame_errors
;
1040 #ifdef CONFIG_NET_POLL_CONTROLLER
1041 static void virtnet_netpoll(struct net_device
*dev
)
1043 struct virtnet_info
*vi
= netdev_priv(dev
);
1046 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++)
1047 napi_schedule(&vi
->rq
[i
].napi
);
1051 static void virtnet_ack_link_announce(struct virtnet_info
*vi
)
1054 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_ANNOUNCE
,
1055 VIRTIO_NET_CTRL_ANNOUNCE_ACK
, NULL
))
1056 dev_warn(&vi
->dev
->dev
, "Failed to ack link announce.\n");
1060 static int virtnet_set_queues(struct virtnet_info
*vi
, u16 queue_pairs
)
1062 struct scatterlist sg
;
1063 struct net_device
*dev
= vi
->dev
;
1065 if (!vi
->has_cvq
|| !virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_MQ
))
1068 vi
->ctrl_mq
.virtqueue_pairs
= cpu_to_virtio16(vi
->vdev
, queue_pairs
);
1069 sg_init_one(&sg
, &vi
->ctrl_mq
, sizeof(vi
->ctrl_mq
));
1071 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MQ
,
1072 VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET
, &sg
)) {
1073 dev_warn(&dev
->dev
, "Fail to set num of queue pairs to %d\n",
1077 vi
->curr_queue_pairs
= queue_pairs
;
1078 /* virtnet_open() will refill when device is going to up. */
1079 if (dev
->flags
& IFF_UP
)
1080 schedule_delayed_work(&vi
->refill
, 0);
1086 static int virtnet_close(struct net_device
*dev
)
1088 struct virtnet_info
*vi
= netdev_priv(dev
);
1091 /* Make sure refill_work doesn't re-enable napi! */
1092 cancel_delayed_work_sync(&vi
->refill
);
1094 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
1095 napi_disable(&vi
->rq
[i
].napi
);
1100 static void virtnet_set_rx_mode(struct net_device
*dev
)
1102 struct virtnet_info
*vi
= netdev_priv(dev
);
1103 struct scatterlist sg
[2];
1104 struct virtio_net_ctrl_mac
*mac_data
;
1105 struct netdev_hw_addr
*ha
;
1111 /* We can't dynamically set ndo_set_rx_mode, so return gracefully */
1112 if (!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_RX
))
1115 vi
->ctrl_promisc
= ((dev
->flags
& IFF_PROMISC
) != 0);
1116 vi
->ctrl_allmulti
= ((dev
->flags
& IFF_ALLMULTI
) != 0);
1118 sg_init_one(sg
, &vi
->ctrl_promisc
, sizeof(vi
->ctrl_promisc
));
1120 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_RX
,
1121 VIRTIO_NET_CTRL_RX_PROMISC
, sg
))
1122 dev_warn(&dev
->dev
, "Failed to %sable promisc mode.\n",
1123 vi
->ctrl_promisc
? "en" : "dis");
1125 sg_init_one(sg
, &vi
->ctrl_allmulti
, sizeof(vi
->ctrl_allmulti
));
1127 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_RX
,
1128 VIRTIO_NET_CTRL_RX_ALLMULTI
, sg
))
1129 dev_warn(&dev
->dev
, "Failed to %sable allmulti mode.\n",
1130 vi
->ctrl_allmulti
? "en" : "dis");
1132 uc_count
= netdev_uc_count(dev
);
1133 mc_count
= netdev_mc_count(dev
);
1134 /* MAC filter - use one buffer for both lists */
1135 buf
= kzalloc(((uc_count
+ mc_count
) * ETH_ALEN
) +
1136 (2 * sizeof(mac_data
->entries
)), GFP_ATOMIC
);
1141 sg_init_table(sg
, 2);
1143 /* Store the unicast list and count in the front of the buffer */
1144 mac_data
->entries
= cpu_to_virtio32(vi
->vdev
, uc_count
);
1146 netdev_for_each_uc_addr(ha
, dev
)
1147 memcpy(&mac_data
->macs
[i
++][0], ha
->addr
, ETH_ALEN
);
1149 sg_set_buf(&sg
[0], mac_data
,
1150 sizeof(mac_data
->entries
) + (uc_count
* ETH_ALEN
));
1152 /* multicast list and count fill the end */
1153 mac_data
= (void *)&mac_data
->macs
[uc_count
][0];
1155 mac_data
->entries
= cpu_to_virtio32(vi
->vdev
, mc_count
);
1157 netdev_for_each_mc_addr(ha
, dev
)
1158 memcpy(&mac_data
->macs
[i
++][0], ha
->addr
, ETH_ALEN
);
1160 sg_set_buf(&sg
[1], mac_data
,
1161 sizeof(mac_data
->entries
) + (mc_count
* ETH_ALEN
));
1163 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MAC
,
1164 VIRTIO_NET_CTRL_MAC_TABLE_SET
, sg
))
1165 dev_warn(&dev
->dev
, "Failed to set MAC filter table.\n");
1170 static int virtnet_vlan_rx_add_vid(struct net_device
*dev
,
1171 __be16 proto
, u16 vid
)
1173 struct virtnet_info
*vi
= netdev_priv(dev
);
1174 struct scatterlist sg
;
1177 sg_init_one(&sg
, &vi
->ctrl_vid
, sizeof(vi
->ctrl_vid
));
1179 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_VLAN
,
1180 VIRTIO_NET_CTRL_VLAN_ADD
, &sg
))
1181 dev_warn(&dev
->dev
, "Failed to add VLAN ID %d.\n", vid
);
1185 static int virtnet_vlan_rx_kill_vid(struct net_device
*dev
,
1186 __be16 proto
, u16 vid
)
1188 struct virtnet_info
*vi
= netdev_priv(dev
);
1189 struct scatterlist sg
;
1192 sg_init_one(&sg
, &vi
->ctrl_vid
, sizeof(vi
->ctrl_vid
));
1194 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_VLAN
,
1195 VIRTIO_NET_CTRL_VLAN_DEL
, &sg
))
1196 dev_warn(&dev
->dev
, "Failed to kill VLAN ID %d.\n", vid
);
1200 static void virtnet_clean_affinity(struct virtnet_info
*vi
, long hcpu
)
1204 if (vi
->affinity_hint_set
) {
1205 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1206 virtqueue_set_affinity(vi
->rq
[i
].vq
, -1);
1207 virtqueue_set_affinity(vi
->sq
[i
].vq
, -1);
1210 vi
->affinity_hint_set
= false;
1214 static void virtnet_set_affinity(struct virtnet_info
*vi
)
1219 /* In multiqueue mode, when the number of cpu is equal to the number of
1220 * queue pairs, we let the queue pairs to be private to one cpu by
1221 * setting the affinity hint to eliminate the contention.
1223 if (vi
->curr_queue_pairs
== 1 ||
1224 vi
->max_queue_pairs
!= num_online_cpus()) {
1225 virtnet_clean_affinity(vi
, -1);
1230 for_each_online_cpu(cpu
) {
1231 virtqueue_set_affinity(vi
->rq
[i
].vq
, cpu
);
1232 virtqueue_set_affinity(vi
->sq
[i
].vq
, cpu
);
1233 netif_set_xps_queue(vi
->dev
, cpumask_of(cpu
), i
);
1237 vi
->affinity_hint_set
= true;
1240 static int virtnet_cpu_callback(struct notifier_block
*nfb
,
1241 unsigned long action
, void *hcpu
)
1243 struct virtnet_info
*vi
= container_of(nfb
, struct virtnet_info
, nb
);
1245 switch(action
& ~CPU_TASKS_FROZEN
) {
1247 case CPU_DOWN_FAILED
:
1249 virtnet_set_affinity(vi
);
1251 case CPU_DOWN_PREPARE
:
1252 virtnet_clean_affinity(vi
, (long)hcpu
);
1261 static void virtnet_get_ringparam(struct net_device
*dev
,
1262 struct ethtool_ringparam
*ring
)
1264 struct virtnet_info
*vi
= netdev_priv(dev
);
1266 ring
->rx_max_pending
= virtqueue_get_vring_size(vi
->rq
[0].vq
);
1267 ring
->tx_max_pending
= virtqueue_get_vring_size(vi
->sq
[0].vq
);
1268 ring
->rx_pending
= ring
->rx_max_pending
;
1269 ring
->tx_pending
= ring
->tx_max_pending
;
1273 static void virtnet_get_drvinfo(struct net_device
*dev
,
1274 struct ethtool_drvinfo
*info
)
1276 struct virtnet_info
*vi
= netdev_priv(dev
);
1277 struct virtio_device
*vdev
= vi
->vdev
;
1279 strlcpy(info
->driver
, KBUILD_MODNAME
, sizeof(info
->driver
));
1280 strlcpy(info
->version
, VIRTNET_DRIVER_VERSION
, sizeof(info
->version
));
1281 strlcpy(info
->bus_info
, virtio_bus_name(vdev
), sizeof(info
->bus_info
));
1285 /* TODO: Eliminate OOO packets during switching */
1286 static int virtnet_set_channels(struct net_device
*dev
,
1287 struct ethtool_channels
*channels
)
1289 struct virtnet_info
*vi
= netdev_priv(dev
);
1290 u16 queue_pairs
= channels
->combined_count
;
1293 /* We don't support separate rx/tx channels.
1294 * We don't allow setting 'other' channels.
1296 if (channels
->rx_count
|| channels
->tx_count
|| channels
->other_count
)
1299 if (queue_pairs
> vi
->max_queue_pairs
|| queue_pairs
== 0)
1303 err
= virtnet_set_queues(vi
, queue_pairs
);
1305 netif_set_real_num_tx_queues(dev
, queue_pairs
);
1306 netif_set_real_num_rx_queues(dev
, queue_pairs
);
1308 virtnet_set_affinity(vi
);
1315 static void virtnet_get_channels(struct net_device
*dev
,
1316 struct ethtool_channels
*channels
)
1318 struct virtnet_info
*vi
= netdev_priv(dev
);
1320 channels
->combined_count
= vi
->curr_queue_pairs
;
1321 channels
->max_combined
= vi
->max_queue_pairs
;
1322 channels
->max_other
= 0;
1323 channels
->rx_count
= 0;
1324 channels
->tx_count
= 0;
1325 channels
->other_count
= 0;
1328 /* Check if the user is trying to change anything besides speed/duplex */
1329 static bool virtnet_validate_ethtool_cmd(const struct ethtool_cmd
*cmd
)
1331 struct ethtool_cmd diff1
= *cmd
;
1332 struct ethtool_cmd diff2
= {};
1334 /* cmd is always set so we need to clear it, validate the port type
1335 * and also without autonegotiation we can ignore advertising
1337 ethtool_cmd_speed_set(&diff1
, 0);
1338 diff2
.port
= PORT_OTHER
;
1339 diff1
.advertising
= 0;
1343 return !memcmp(&diff1
, &diff2
, sizeof(diff1
));
1346 static int virtnet_set_settings(struct net_device
*dev
, struct ethtool_cmd
*cmd
)
1348 struct virtnet_info
*vi
= netdev_priv(dev
);
1351 speed
= ethtool_cmd_speed(cmd
);
1352 /* don't allow custom speed and duplex */
1353 if (!ethtool_validate_speed(speed
) ||
1354 !ethtool_validate_duplex(cmd
->duplex
) ||
1355 !virtnet_validate_ethtool_cmd(cmd
))
1358 vi
->duplex
= cmd
->duplex
;
1363 static int virtnet_get_settings(struct net_device
*dev
, struct ethtool_cmd
*cmd
)
1365 struct virtnet_info
*vi
= netdev_priv(dev
);
1367 ethtool_cmd_speed_set(cmd
, vi
->speed
);
1368 cmd
->duplex
= vi
->duplex
;
1369 cmd
->port
= PORT_OTHER
;
1374 static void virtnet_init_settings(struct net_device
*dev
)
1376 struct virtnet_info
*vi
= netdev_priv(dev
);
1378 vi
->speed
= SPEED_UNKNOWN
;
1379 vi
->duplex
= DUPLEX_UNKNOWN
;
1382 static const struct ethtool_ops virtnet_ethtool_ops
= {
1383 .get_drvinfo
= virtnet_get_drvinfo
,
1384 .get_link
= ethtool_op_get_link
,
1385 .get_ringparam
= virtnet_get_ringparam
,
1386 .set_channels
= virtnet_set_channels
,
1387 .get_channels
= virtnet_get_channels
,
1388 .get_ts_info
= ethtool_op_get_ts_info
,
1389 .get_settings
= virtnet_get_settings
,
1390 .set_settings
= virtnet_set_settings
,
1394 #define MAX_MTU 65535
1396 static int virtnet_change_mtu(struct net_device
*dev
, int new_mtu
)
1398 if (new_mtu
< MIN_MTU
|| new_mtu
> MAX_MTU
)
1404 static const struct net_device_ops virtnet_netdev
= {
1405 .ndo_open
= virtnet_open
,
1406 .ndo_stop
= virtnet_close
,
1407 .ndo_start_xmit
= start_xmit
,
1408 .ndo_validate_addr
= eth_validate_addr
,
1409 .ndo_set_mac_address
= virtnet_set_mac_address
,
1410 .ndo_set_rx_mode
= virtnet_set_rx_mode
,
1411 .ndo_change_mtu
= virtnet_change_mtu
,
1412 .ndo_get_stats64
= virtnet_stats
,
1413 .ndo_vlan_rx_add_vid
= virtnet_vlan_rx_add_vid
,
1414 .ndo_vlan_rx_kill_vid
= virtnet_vlan_rx_kill_vid
,
1415 #ifdef CONFIG_NET_POLL_CONTROLLER
1416 .ndo_poll_controller
= virtnet_netpoll
,
1418 #ifdef CONFIG_NET_RX_BUSY_POLL
1419 .ndo_busy_poll
= virtnet_busy_poll
,
1423 static void virtnet_config_changed_work(struct work_struct
*work
)
1425 struct virtnet_info
*vi
=
1426 container_of(work
, struct virtnet_info
, config_work
);
1429 if (virtio_cread_feature(vi
->vdev
, VIRTIO_NET_F_STATUS
,
1430 struct virtio_net_config
, status
, &v
) < 0)
1433 if (v
& VIRTIO_NET_S_ANNOUNCE
) {
1434 netdev_notify_peers(vi
->dev
);
1435 virtnet_ack_link_announce(vi
);
1438 /* Ignore unknown (future) status bits */
1439 v
&= VIRTIO_NET_S_LINK_UP
;
1441 if (vi
->status
== v
)
1446 if (vi
->status
& VIRTIO_NET_S_LINK_UP
) {
1447 netif_carrier_on(vi
->dev
);
1448 netif_tx_wake_all_queues(vi
->dev
);
1450 netif_carrier_off(vi
->dev
);
1451 netif_tx_stop_all_queues(vi
->dev
);
1455 static void virtnet_config_changed(struct virtio_device
*vdev
)
1457 struct virtnet_info
*vi
= vdev
->priv
;
1459 schedule_work(&vi
->config_work
);
1462 static void virtnet_free_queues(struct virtnet_info
*vi
)
1466 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1467 napi_hash_del(&vi
->rq
[i
].napi
);
1468 netif_napi_del(&vi
->rq
[i
].napi
);
1475 static void free_receive_bufs(struct virtnet_info
*vi
)
1479 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1480 while (vi
->rq
[i
].pages
)
1481 __free_pages(get_a_page(&vi
->rq
[i
], GFP_KERNEL
), 0);
1485 static void free_receive_page_frags(struct virtnet_info
*vi
)
1488 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
1489 if (vi
->rq
[i
].alloc_frag
.page
)
1490 put_page(vi
->rq
[i
].alloc_frag
.page
);
1493 static void free_unused_bufs(struct virtnet_info
*vi
)
1498 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1499 struct virtqueue
*vq
= vi
->sq
[i
].vq
;
1500 while ((buf
= virtqueue_detach_unused_buf(vq
)) != NULL
)
1504 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1505 struct virtqueue
*vq
= vi
->rq
[i
].vq
;
1507 while ((buf
= virtqueue_detach_unused_buf(vq
)) != NULL
) {
1508 if (vi
->mergeable_rx_bufs
) {
1509 unsigned long ctx
= (unsigned long)buf
;
1510 void *base
= mergeable_ctx_to_buf_address(ctx
);
1511 put_page(virt_to_head_page(base
));
1512 } else if (vi
->big_packets
) {
1513 give_pages(&vi
->rq
[i
], buf
);
1521 static void virtnet_del_vqs(struct virtnet_info
*vi
)
1523 struct virtio_device
*vdev
= vi
->vdev
;
1525 virtnet_clean_affinity(vi
, -1);
1527 vdev
->config
->del_vqs(vdev
);
1529 virtnet_free_queues(vi
);
1532 static int virtnet_find_vqs(struct virtnet_info
*vi
)
1534 vq_callback_t
**callbacks
;
1535 struct virtqueue
**vqs
;
1540 /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
1541 * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
1542 * possible control vq.
1544 total_vqs
= vi
->max_queue_pairs
* 2 +
1545 virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
);
1547 /* Allocate space for find_vqs parameters */
1548 vqs
= kzalloc(total_vqs
* sizeof(*vqs
), GFP_KERNEL
);
1551 callbacks
= kmalloc(total_vqs
* sizeof(*callbacks
), GFP_KERNEL
);
1554 names
= kmalloc(total_vqs
* sizeof(*names
), GFP_KERNEL
);
1558 /* Parameters for control virtqueue, if any */
1560 callbacks
[total_vqs
- 1] = NULL
;
1561 names
[total_vqs
- 1] = "control";
1564 /* Allocate/initialize parameters for send/receive virtqueues */
1565 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1566 callbacks
[rxq2vq(i
)] = skb_recv_done
;
1567 callbacks
[txq2vq(i
)] = skb_xmit_done
;
1568 sprintf(vi
->rq
[i
].name
, "input.%d", i
);
1569 sprintf(vi
->sq
[i
].name
, "output.%d", i
);
1570 names
[rxq2vq(i
)] = vi
->rq
[i
].name
;
1571 names
[txq2vq(i
)] = vi
->sq
[i
].name
;
1574 ret
= vi
->vdev
->config
->find_vqs(vi
->vdev
, total_vqs
, vqs
, callbacks
,
1580 vi
->cvq
= vqs
[total_vqs
- 1];
1581 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VLAN
))
1582 vi
->dev
->features
|= NETIF_F_HW_VLAN_CTAG_FILTER
;
1585 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1586 vi
->rq
[i
].vq
= vqs
[rxq2vq(i
)];
1587 vi
->sq
[i
].vq
= vqs
[txq2vq(i
)];
1606 static int virtnet_alloc_queues(struct virtnet_info
*vi
)
1610 vi
->sq
= kzalloc(sizeof(*vi
->sq
) * vi
->max_queue_pairs
, GFP_KERNEL
);
1613 vi
->rq
= kzalloc(sizeof(*vi
->rq
) * vi
->max_queue_pairs
, GFP_KERNEL
);
1617 INIT_DELAYED_WORK(&vi
->refill
, refill_work
);
1618 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1619 vi
->rq
[i
].pages
= NULL
;
1620 netif_napi_add(vi
->dev
, &vi
->rq
[i
].napi
, virtnet_poll
,
1623 sg_init_table(vi
->rq
[i
].sg
, ARRAY_SIZE(vi
->rq
[i
].sg
));
1624 ewma_pkt_len_init(&vi
->rq
[i
].mrg_avg_pkt_len
);
1625 sg_init_table(vi
->sq
[i
].sg
, ARRAY_SIZE(vi
->sq
[i
].sg
));
1636 static int init_vqs(struct virtnet_info
*vi
)
1640 /* Allocate send & receive queues */
1641 ret
= virtnet_alloc_queues(vi
);
1645 ret
= virtnet_find_vqs(vi
);
1650 virtnet_set_affinity(vi
);
1656 virtnet_free_queues(vi
);
1662 static ssize_t
mergeable_rx_buffer_size_show(struct netdev_rx_queue
*queue
,
1663 struct rx_queue_attribute
*attribute
, char *buf
)
1665 struct virtnet_info
*vi
= netdev_priv(queue
->dev
);
1666 unsigned int queue_index
= get_netdev_rx_queue_index(queue
);
1667 struct ewma_pkt_len
*avg
;
1669 BUG_ON(queue_index
>= vi
->max_queue_pairs
);
1670 avg
= &vi
->rq
[queue_index
].mrg_avg_pkt_len
;
1671 return sprintf(buf
, "%u\n", get_mergeable_buf_len(avg
));
1674 static struct rx_queue_attribute mergeable_rx_buffer_size_attribute
=
1675 __ATTR_RO(mergeable_rx_buffer_size
);
1677 static struct attribute
*virtio_net_mrg_rx_attrs
[] = {
1678 &mergeable_rx_buffer_size_attribute
.attr
,
1682 static const struct attribute_group virtio_net_mrg_rx_group
= {
1683 .name
= "virtio_net",
1684 .attrs
= virtio_net_mrg_rx_attrs
1688 static bool virtnet_fail_on_feature(struct virtio_device
*vdev
,
1690 const char *fname
, const char *dname
)
1692 if (!virtio_has_feature(vdev
, fbit
))
1695 dev_err(&vdev
->dev
, "device advertises feature %s but not %s",
1701 #define VIRTNET_FAIL_ON(vdev, fbit, dbit) \
1702 virtnet_fail_on_feature(vdev, fbit, #fbit, dbit)
1704 static bool virtnet_validate_features(struct virtio_device
*vdev
)
1706 if (!virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
) &&
1707 (VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_CTRL_RX
,
1708 "VIRTIO_NET_F_CTRL_VQ") ||
1709 VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_CTRL_VLAN
,
1710 "VIRTIO_NET_F_CTRL_VQ") ||
1711 VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_GUEST_ANNOUNCE
,
1712 "VIRTIO_NET_F_CTRL_VQ") ||
1713 VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_MQ
, "VIRTIO_NET_F_CTRL_VQ") ||
1714 VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_CTRL_MAC_ADDR
,
1715 "VIRTIO_NET_F_CTRL_VQ"))) {
1722 static int virtnet_probe(struct virtio_device
*vdev
)
1725 struct net_device
*dev
;
1726 struct virtnet_info
*vi
;
1727 u16 max_queue_pairs
;
1730 if (!vdev
->config
->get
) {
1731 dev_err(&vdev
->dev
, "%s failure: config access disabled\n",
1736 if (!virtnet_validate_features(vdev
))
1739 /* Find if host supports multiqueue virtio_net device */
1740 err
= virtio_cread_feature(vdev
, VIRTIO_NET_F_MQ
,
1741 struct virtio_net_config
,
1742 max_virtqueue_pairs
, &max_queue_pairs
);
1744 /* We need at least 2 queue's */
1745 if (err
|| max_queue_pairs
< VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN
||
1746 max_queue_pairs
> VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX
||
1747 !virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
))
1748 max_queue_pairs
= 1;
1750 /* Allocate ourselves a network device with room for our info */
1751 dev
= alloc_etherdev_mq(sizeof(struct virtnet_info
), max_queue_pairs
);
1755 /* Set up network device as normal. */
1756 dev
->priv_flags
|= IFF_UNICAST_FLT
| IFF_LIVE_ADDR_CHANGE
;
1757 dev
->netdev_ops
= &virtnet_netdev
;
1758 dev
->features
= NETIF_F_HIGHDMA
;
1760 dev
->ethtool_ops
= &virtnet_ethtool_ops
;
1761 SET_NETDEV_DEV(dev
, &vdev
->dev
);
1763 /* Do we support "hardware" checksums? */
1764 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CSUM
)) {
1765 /* This opens up the world of extra features. */
1766 dev
->hw_features
|= NETIF_F_HW_CSUM
| NETIF_F_SG
;
1768 dev
->features
|= NETIF_F_HW_CSUM
| NETIF_F_SG
;
1770 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GSO
)) {
1771 dev
->hw_features
|= NETIF_F_TSO
| NETIF_F_UFO
1772 | NETIF_F_TSO_ECN
| NETIF_F_TSO6
;
1774 /* Individual feature bits: what can host handle? */
1775 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_TSO4
))
1776 dev
->hw_features
|= NETIF_F_TSO
;
1777 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_TSO6
))
1778 dev
->hw_features
|= NETIF_F_TSO6
;
1779 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_ECN
))
1780 dev
->hw_features
|= NETIF_F_TSO_ECN
;
1781 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_UFO
))
1782 dev
->hw_features
|= NETIF_F_UFO
;
1784 dev
->features
|= NETIF_F_GSO_ROBUST
;
1787 dev
->features
|= dev
->hw_features
& (NETIF_F_ALL_TSO
|NETIF_F_UFO
);
1788 /* (!csum && gso) case will be fixed by register_netdev() */
1790 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_CSUM
))
1791 dev
->features
|= NETIF_F_RXCSUM
;
1793 dev
->vlan_features
= dev
->features
;
1795 /* Configuration may specify what MAC to use. Otherwise random. */
1796 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MAC
))
1797 virtio_cread_bytes(vdev
,
1798 offsetof(struct virtio_net_config
, mac
),
1799 dev
->dev_addr
, dev
->addr_len
);
1801 eth_hw_addr_random(dev
);
1803 /* Set up our device-specific information */
1804 vi
= netdev_priv(dev
);
1808 vi
->stats
= alloc_percpu(struct virtnet_stats
);
1810 if (vi
->stats
== NULL
)
1813 for_each_possible_cpu(i
) {
1814 struct virtnet_stats
*virtnet_stats
;
1815 virtnet_stats
= per_cpu_ptr(vi
->stats
, i
);
1816 u64_stats_init(&virtnet_stats
->tx_syncp
);
1817 u64_stats_init(&virtnet_stats
->rx_syncp
);
1820 INIT_WORK(&vi
->config_work
, virtnet_config_changed_work
);
1822 /* If we can receive ANY GSO packets, we must allocate large ones. */
1823 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_TSO4
) ||
1824 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_TSO6
) ||
1825 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_ECN
) ||
1826 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_UFO
))
1827 vi
->big_packets
= true;
1829 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MRG_RXBUF
))
1830 vi
->mergeable_rx_bufs
= true;
1832 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MRG_RXBUF
) ||
1833 virtio_has_feature(vdev
, VIRTIO_F_VERSION_1
))
1834 vi
->hdr_len
= sizeof(struct virtio_net_hdr_mrg_rxbuf
);
1836 vi
->hdr_len
= sizeof(struct virtio_net_hdr
);
1838 if (virtio_has_feature(vdev
, VIRTIO_F_ANY_LAYOUT
) ||
1839 virtio_has_feature(vdev
, VIRTIO_F_VERSION_1
))
1840 vi
->any_header_sg
= true;
1842 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
))
1845 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MTU
)) {
1846 mtu
= virtio_cread16(vdev
,
1847 offsetof(struct virtio_net_config
,
1849 if (virtnet_change_mtu(dev
, mtu
))
1850 __virtio_clear_bit(vdev
, VIRTIO_NET_F_MTU
);
1853 if (vi
->any_header_sg
)
1854 dev
->needed_headroom
= vi
->hdr_len
;
1856 /* Use single tx/rx queue pair as default */
1857 vi
->curr_queue_pairs
= 1;
1858 vi
->max_queue_pairs
= max_queue_pairs
;
1860 /* Allocate/initialize the rx/tx queues, and invoke find_vqs */
1866 if (vi
->mergeable_rx_bufs
)
1867 dev
->sysfs_rx_queue_group
= &virtio_net_mrg_rx_group
;
1869 netif_set_real_num_tx_queues(dev
, vi
->curr_queue_pairs
);
1870 netif_set_real_num_rx_queues(dev
, vi
->curr_queue_pairs
);
1872 virtnet_init_settings(dev
);
1874 err
= register_netdev(dev
);
1876 pr_debug("virtio_net: registering device failed\n");
1880 virtio_device_ready(vdev
);
1882 vi
->nb
.notifier_call
= &virtnet_cpu_callback
;
1883 err
= register_hotcpu_notifier(&vi
->nb
);
1885 pr_debug("virtio_net: registering cpu notifier failed\n");
1886 goto free_unregister_netdev
;
1889 /* Assume link up if device can't report link status,
1890 otherwise get link status from config. */
1891 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_STATUS
)) {
1892 netif_carrier_off(dev
);
1893 schedule_work(&vi
->config_work
);
1895 vi
->status
= VIRTIO_NET_S_LINK_UP
;
1896 netif_carrier_on(dev
);
1899 pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
1900 dev
->name
, max_queue_pairs
);
1904 free_unregister_netdev
:
1905 vi
->vdev
->config
->reset(vdev
);
1907 unregister_netdev(dev
);
1909 cancel_delayed_work_sync(&vi
->refill
);
1910 free_receive_page_frags(vi
);
1911 virtnet_del_vqs(vi
);
1913 free_percpu(vi
->stats
);
1919 static void remove_vq_common(struct virtnet_info
*vi
)
1921 vi
->vdev
->config
->reset(vi
->vdev
);
1923 /* Free unused buffers in both send and recv, if any. */
1924 free_unused_bufs(vi
);
1926 free_receive_bufs(vi
);
1928 free_receive_page_frags(vi
);
1930 virtnet_del_vqs(vi
);
1933 static void virtnet_remove(struct virtio_device
*vdev
)
1935 struct virtnet_info
*vi
= vdev
->priv
;
1937 unregister_hotcpu_notifier(&vi
->nb
);
1939 /* Make sure no work handler is accessing the device. */
1940 flush_work(&vi
->config_work
);
1942 unregister_netdev(vi
->dev
);
1944 remove_vq_common(vi
);
1946 free_percpu(vi
->stats
);
1947 free_netdev(vi
->dev
);
1950 #ifdef CONFIG_PM_SLEEP
1951 static int virtnet_freeze(struct virtio_device
*vdev
)
1953 struct virtnet_info
*vi
= vdev
->priv
;
1956 unregister_hotcpu_notifier(&vi
->nb
);
1958 /* Make sure no work handler is accessing the device */
1959 flush_work(&vi
->config_work
);
1961 netif_device_detach(vi
->dev
);
1962 cancel_delayed_work_sync(&vi
->refill
);
1964 if (netif_running(vi
->dev
)) {
1965 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
1966 napi_disable(&vi
->rq
[i
].napi
);
1969 remove_vq_common(vi
);
1974 static int virtnet_restore(struct virtio_device
*vdev
)
1976 struct virtnet_info
*vi
= vdev
->priv
;
1983 virtio_device_ready(vdev
);
1985 if (netif_running(vi
->dev
)) {
1986 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++)
1987 if (!try_fill_recv(vi
, &vi
->rq
[i
], GFP_KERNEL
))
1988 schedule_delayed_work(&vi
->refill
, 0);
1990 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
1991 virtnet_napi_enable(&vi
->rq
[i
]);
1994 netif_device_attach(vi
->dev
);
1997 virtnet_set_queues(vi
, vi
->curr_queue_pairs
);
2000 err
= register_hotcpu_notifier(&vi
->nb
);
2008 static struct virtio_device_id id_table
[] = {
2009 { VIRTIO_ID_NET
, VIRTIO_DEV_ANY_ID
},
2013 static unsigned int features
[] = {
2014 VIRTIO_NET_F_CSUM
, VIRTIO_NET_F_GUEST_CSUM
,
2015 VIRTIO_NET_F_GSO
, VIRTIO_NET_F_MAC
,
2016 VIRTIO_NET_F_HOST_TSO4
, VIRTIO_NET_F_HOST_UFO
, VIRTIO_NET_F_HOST_TSO6
,
2017 VIRTIO_NET_F_HOST_ECN
, VIRTIO_NET_F_GUEST_TSO4
, VIRTIO_NET_F_GUEST_TSO6
,
2018 VIRTIO_NET_F_GUEST_ECN
, VIRTIO_NET_F_GUEST_UFO
,
2019 VIRTIO_NET_F_MRG_RXBUF
, VIRTIO_NET_F_STATUS
, VIRTIO_NET_F_CTRL_VQ
,
2020 VIRTIO_NET_F_CTRL_RX
, VIRTIO_NET_F_CTRL_VLAN
,
2021 VIRTIO_NET_F_GUEST_ANNOUNCE
, VIRTIO_NET_F_MQ
,
2022 VIRTIO_NET_F_CTRL_MAC_ADDR
,
2023 VIRTIO_F_ANY_LAYOUT
,
2027 static struct virtio_driver virtio_net_driver
= {
2028 .feature_table
= features
,
2029 .feature_table_size
= ARRAY_SIZE(features
),
2030 .driver
.name
= KBUILD_MODNAME
,
2031 .driver
.owner
= THIS_MODULE
,
2032 .id_table
= id_table
,
2033 .probe
= virtnet_probe
,
2034 .remove
= virtnet_remove
,
2035 .config_changed
= virtnet_config_changed
,
2036 #ifdef CONFIG_PM_SLEEP
2037 .freeze
= virtnet_freeze
,
2038 .restore
= virtnet_restore
,
2042 module_virtio_driver(virtio_net_driver
);
2044 MODULE_DEVICE_TABLE(virtio
, id_table
);
2045 MODULE_DESCRIPTION("Virtio network driver");
2046 MODULE_LICENSE("GPL");