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
;
150 /* Ethtool settings */
155 struct padded_vnet_hdr
{
156 struct virtio_net_hdr_mrg_rxbuf hdr
;
158 * hdr is in a separate sg buffer, and data sg buffer shares same page
159 * with this header sg. This padding makes next sg 16 byte aligned
165 /* Converting between virtqueue no. and kernel tx/rx queue no.
166 * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq
168 static int vq2txq(struct virtqueue
*vq
)
170 return (vq
->index
- 1) / 2;
173 static int txq2vq(int txq
)
178 static int vq2rxq(struct virtqueue
*vq
)
180 return vq
->index
/ 2;
183 static int rxq2vq(int rxq
)
188 static inline struct virtio_net_hdr_mrg_rxbuf
*skb_vnet_hdr(struct sk_buff
*skb
)
190 return (struct virtio_net_hdr_mrg_rxbuf
*)skb
->cb
;
194 * private is used to chain pages for big packets, put the whole
195 * most recent used list in the beginning for reuse
197 static void give_pages(struct receive_queue
*rq
, struct page
*page
)
201 /* Find end of list, sew whole thing into vi->rq.pages. */
202 for (end
= page
; end
->private; end
= (struct page
*)end
->private);
203 end
->private = (unsigned long)rq
->pages
;
207 static struct page
*get_a_page(struct receive_queue
*rq
, gfp_t gfp_mask
)
209 struct page
*p
= rq
->pages
;
212 rq
->pages
= (struct page
*)p
->private;
213 /* clear private here, it is used to chain pages */
216 p
= alloc_page(gfp_mask
);
220 static void skb_xmit_done(struct virtqueue
*vq
)
222 struct virtnet_info
*vi
= vq
->vdev
->priv
;
224 /* Suppress further interrupts. */
225 virtqueue_disable_cb(vq
);
227 /* We were probably waiting for more output buffers. */
228 netif_wake_subqueue(vi
->dev
, vq2txq(vq
));
231 static unsigned int mergeable_ctx_to_buf_truesize(unsigned long mrg_ctx
)
233 unsigned int truesize
= mrg_ctx
& (MERGEABLE_BUFFER_ALIGN
- 1);
234 return (truesize
+ 1) * MERGEABLE_BUFFER_ALIGN
;
237 static void *mergeable_ctx_to_buf_address(unsigned long mrg_ctx
)
239 return (void *)(mrg_ctx
& -MERGEABLE_BUFFER_ALIGN
);
243 static unsigned long mergeable_buf_to_ctx(void *buf
, unsigned int truesize
)
245 unsigned int size
= truesize
/ MERGEABLE_BUFFER_ALIGN
;
246 return (unsigned long)buf
| (size
- 1);
249 /* Called from bottom half context */
250 static struct sk_buff
*page_to_skb(struct virtnet_info
*vi
,
251 struct receive_queue
*rq
,
252 struct page
*page
, unsigned int offset
,
253 unsigned int len
, unsigned int truesize
)
256 struct virtio_net_hdr_mrg_rxbuf
*hdr
;
257 unsigned int copy
, hdr_len
, hdr_padded_len
;
260 p
= page_address(page
) + offset
;
262 /* copy small packet so we can reuse these pages for small data */
263 skb
= napi_alloc_skb(&rq
->napi
, GOOD_COPY_LEN
);
267 hdr
= skb_vnet_hdr(skb
);
269 hdr_len
= vi
->hdr_len
;
270 if (vi
->mergeable_rx_bufs
)
271 hdr_padded_len
= sizeof *hdr
;
273 hdr_padded_len
= sizeof(struct padded_vnet_hdr
);
275 memcpy(hdr
, p
, hdr_len
);
278 offset
+= hdr_padded_len
;
282 if (copy
> skb_tailroom(skb
))
283 copy
= skb_tailroom(skb
);
284 memcpy(skb_put(skb
, copy
), p
, copy
);
289 if (vi
->mergeable_rx_bufs
) {
291 skb_add_rx_frag(skb
, 0, page
, offset
, len
, truesize
);
298 * Verify that we can indeed put this data into a skb.
299 * This is here to handle cases when the device erroneously
300 * tries to receive more than is possible. This is usually
301 * the case of a broken device.
303 if (unlikely(len
> MAX_SKB_FRAGS
* PAGE_SIZE
)) {
304 net_dbg_ratelimited("%s: too much data\n", skb
->dev
->name
);
308 BUG_ON(offset
>= PAGE_SIZE
);
310 unsigned int frag_size
= min((unsigned)PAGE_SIZE
- offset
, len
);
311 skb_add_rx_frag(skb
, skb_shinfo(skb
)->nr_frags
, page
, offset
,
312 frag_size
, truesize
);
314 page
= (struct page
*)page
->private;
319 give_pages(rq
, page
);
324 static struct sk_buff
*receive_small(struct virtnet_info
*vi
, void *buf
, unsigned int len
)
326 struct sk_buff
* skb
= buf
;
334 static struct sk_buff
*receive_big(struct net_device
*dev
,
335 struct virtnet_info
*vi
,
336 struct receive_queue
*rq
,
340 struct page
*page
= buf
;
341 struct sk_buff
*skb
= page_to_skb(vi
, rq
, page
, 0, len
, PAGE_SIZE
);
349 dev
->stats
.rx_dropped
++;
350 give_pages(rq
, page
);
354 static struct sk_buff
*receive_mergeable(struct net_device
*dev
,
355 struct virtnet_info
*vi
,
356 struct receive_queue
*rq
,
360 void *buf
= mergeable_ctx_to_buf_address(ctx
);
361 struct virtio_net_hdr_mrg_rxbuf
*hdr
= buf
;
362 u16 num_buf
= virtio16_to_cpu(vi
->vdev
, hdr
->num_buffers
);
363 struct page
*page
= virt_to_head_page(buf
);
364 int offset
= buf
- page_address(page
);
365 unsigned int truesize
= max(len
, mergeable_ctx_to_buf_truesize(ctx
));
367 struct sk_buff
*head_skb
= page_to_skb(vi
, rq
, page
, offset
, len
,
369 struct sk_buff
*curr_skb
= head_skb
;
371 if (unlikely(!curr_skb
))
376 ctx
= (unsigned long)virtqueue_get_buf(rq
->vq
, &len
);
377 if (unlikely(!ctx
)) {
378 pr_debug("%s: rx error: %d buffers out of %d missing\n",
380 virtio16_to_cpu(vi
->vdev
,
382 dev
->stats
.rx_length_errors
++;
386 buf
= mergeable_ctx_to_buf_address(ctx
);
387 page
= virt_to_head_page(buf
);
389 num_skb_frags
= skb_shinfo(curr_skb
)->nr_frags
;
390 if (unlikely(num_skb_frags
== MAX_SKB_FRAGS
)) {
391 struct sk_buff
*nskb
= alloc_skb(0, GFP_ATOMIC
);
395 if (curr_skb
== head_skb
)
396 skb_shinfo(curr_skb
)->frag_list
= nskb
;
398 curr_skb
->next
= nskb
;
400 head_skb
->truesize
+= nskb
->truesize
;
403 truesize
= max(len
, mergeable_ctx_to_buf_truesize(ctx
));
404 if (curr_skb
!= head_skb
) {
405 head_skb
->data_len
+= len
;
406 head_skb
->len
+= len
;
407 head_skb
->truesize
+= truesize
;
409 offset
= buf
- page_address(page
);
410 if (skb_can_coalesce(curr_skb
, num_skb_frags
, page
, offset
)) {
412 skb_coalesce_rx_frag(curr_skb
, num_skb_frags
- 1,
415 skb_add_rx_frag(curr_skb
, num_skb_frags
, page
,
416 offset
, len
, truesize
);
420 ewma_pkt_len_add(&rq
->mrg_avg_pkt_len
, head_skb
->len
);
426 ctx
= (unsigned long)virtqueue_get_buf(rq
->vq
, &len
);
427 if (unlikely(!ctx
)) {
428 pr_debug("%s: rx error: %d buffers missing\n",
430 dev
->stats
.rx_length_errors
++;
433 page
= virt_to_head_page(mergeable_ctx_to_buf_address(ctx
));
437 dev
->stats
.rx_dropped
++;
438 dev_kfree_skb(head_skb
);
442 static void receive_buf(struct virtnet_info
*vi
, struct receive_queue
*rq
,
443 void *buf
, unsigned int len
)
445 struct net_device
*dev
= vi
->dev
;
446 struct virtnet_stats
*stats
= this_cpu_ptr(vi
->stats
);
448 struct virtio_net_hdr_mrg_rxbuf
*hdr
;
450 if (unlikely(len
< vi
->hdr_len
+ ETH_HLEN
)) {
451 pr_debug("%s: short packet %i\n", dev
->name
, len
);
452 dev
->stats
.rx_length_errors
++;
453 if (vi
->mergeable_rx_bufs
) {
454 unsigned long ctx
= (unsigned long)buf
;
455 void *base
= mergeable_ctx_to_buf_address(ctx
);
456 put_page(virt_to_head_page(base
));
457 } else if (vi
->big_packets
) {
465 if (vi
->mergeable_rx_bufs
)
466 skb
= receive_mergeable(dev
, vi
, rq
, (unsigned long)buf
, len
);
467 else if (vi
->big_packets
)
468 skb
= receive_big(dev
, vi
, rq
, buf
, len
);
470 skb
= receive_small(vi
, buf
, len
);
475 hdr
= skb_vnet_hdr(skb
);
477 u64_stats_update_begin(&stats
->rx_syncp
);
478 stats
->rx_bytes
+= skb
->len
;
480 u64_stats_update_end(&stats
->rx_syncp
);
482 if (hdr
->hdr
.flags
& VIRTIO_NET_HDR_F_DATA_VALID
)
483 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
485 if (virtio_net_hdr_to_skb(skb
, &hdr
->hdr
,
486 virtio_is_little_endian(vi
->vdev
))) {
487 net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n",
488 dev
->name
, hdr
->hdr
.gso_type
,
493 skb
->protocol
= eth_type_trans(skb
, dev
);
494 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
495 ntohs(skb
->protocol
), skb
->len
, skb
->pkt_type
);
497 napi_gro_receive(&rq
->napi
, skb
);
501 dev
->stats
.rx_frame_errors
++;
505 static int add_recvbuf_small(struct virtnet_info
*vi
, struct receive_queue
*rq
,
509 struct virtio_net_hdr_mrg_rxbuf
*hdr
;
512 skb
= __netdev_alloc_skb_ip_align(vi
->dev
, GOOD_PACKET_LEN
, gfp
);
516 skb_put(skb
, GOOD_PACKET_LEN
);
518 hdr
= skb_vnet_hdr(skb
);
519 sg_init_table(rq
->sg
, 2);
520 sg_set_buf(rq
->sg
, hdr
, vi
->hdr_len
);
521 skb_to_sgvec(skb
, rq
->sg
+ 1, 0, skb
->len
);
523 err
= virtqueue_add_inbuf(rq
->vq
, rq
->sg
, 2, skb
, gfp
);
530 static int add_recvbuf_big(struct virtnet_info
*vi
, struct receive_queue
*rq
,
533 struct page
*first
, *list
= NULL
;
537 sg_init_table(rq
->sg
, MAX_SKB_FRAGS
+ 2);
539 /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */
540 for (i
= MAX_SKB_FRAGS
+ 1; i
> 1; --i
) {
541 first
= get_a_page(rq
, gfp
);
544 give_pages(rq
, list
);
547 sg_set_buf(&rq
->sg
[i
], page_address(first
), PAGE_SIZE
);
549 /* chain new page in list head to match sg */
550 first
->private = (unsigned long)list
;
554 first
= get_a_page(rq
, gfp
);
556 give_pages(rq
, list
);
559 p
= page_address(first
);
561 /* rq->sg[0], rq->sg[1] share the same page */
562 /* a separated rq->sg[0] for header - required in case !any_header_sg */
563 sg_set_buf(&rq
->sg
[0], p
, vi
->hdr_len
);
565 /* rq->sg[1] for data packet, from offset */
566 offset
= sizeof(struct padded_vnet_hdr
);
567 sg_set_buf(&rq
->sg
[1], p
+ offset
, PAGE_SIZE
- offset
);
569 /* chain first in list head */
570 first
->private = (unsigned long)list
;
571 err
= virtqueue_add_inbuf(rq
->vq
, rq
->sg
, MAX_SKB_FRAGS
+ 2,
574 give_pages(rq
, first
);
579 static unsigned int get_mergeable_buf_len(struct ewma_pkt_len
*avg_pkt_len
)
581 const size_t hdr_len
= sizeof(struct virtio_net_hdr_mrg_rxbuf
);
584 len
= hdr_len
+ clamp_t(unsigned int, ewma_pkt_len_read(avg_pkt_len
),
585 GOOD_PACKET_LEN
, PAGE_SIZE
- hdr_len
);
586 return ALIGN(len
, MERGEABLE_BUFFER_ALIGN
);
589 static int add_recvbuf_mergeable(struct receive_queue
*rq
, gfp_t gfp
)
591 struct page_frag
*alloc_frag
= &rq
->alloc_frag
;
595 unsigned int len
, hole
;
597 len
= get_mergeable_buf_len(&rq
->mrg_avg_pkt_len
);
598 if (unlikely(!skb_page_frag_refill(len
, alloc_frag
, gfp
)))
601 buf
= (char *)page_address(alloc_frag
->page
) + alloc_frag
->offset
;
602 ctx
= mergeable_buf_to_ctx(buf
, len
);
603 get_page(alloc_frag
->page
);
604 alloc_frag
->offset
+= len
;
605 hole
= alloc_frag
->size
- alloc_frag
->offset
;
607 /* To avoid internal fragmentation, if there is very likely not
608 * enough space for another buffer, add the remaining space to
609 * the current buffer. This extra space is not included in
610 * the truesize stored in ctx.
613 alloc_frag
->offset
+= hole
;
616 sg_init_one(rq
->sg
, buf
, len
);
617 err
= virtqueue_add_inbuf(rq
->vq
, rq
->sg
, 1, (void *)ctx
, gfp
);
619 put_page(virt_to_head_page(buf
));
625 * Returns false if we couldn't fill entirely (OOM).
627 * Normally run in the receive path, but can also be run from ndo_open
628 * before we're receiving packets, or from refill_work which is
629 * careful to disable receiving (using napi_disable).
631 static bool try_fill_recv(struct virtnet_info
*vi
, struct receive_queue
*rq
,
639 if (vi
->mergeable_rx_bufs
)
640 err
= add_recvbuf_mergeable(rq
, gfp
);
641 else if (vi
->big_packets
)
642 err
= add_recvbuf_big(vi
, rq
, gfp
);
644 err
= add_recvbuf_small(vi
, rq
, gfp
);
646 oom
= err
== -ENOMEM
;
649 } while (rq
->vq
->num_free
);
650 virtqueue_kick(rq
->vq
);
654 static void skb_recv_done(struct virtqueue
*rvq
)
656 struct virtnet_info
*vi
= rvq
->vdev
->priv
;
657 struct receive_queue
*rq
= &vi
->rq
[vq2rxq(rvq
)];
659 /* Schedule NAPI, Suppress further interrupts if successful. */
660 if (napi_schedule_prep(&rq
->napi
)) {
661 virtqueue_disable_cb(rvq
);
662 __napi_schedule(&rq
->napi
);
666 static void virtnet_napi_enable(struct receive_queue
*rq
)
668 napi_enable(&rq
->napi
);
670 /* If all buffers were filled by other side before we napi_enabled, we
671 * won't get another interrupt, so process any outstanding packets
672 * now. virtnet_poll wants re-enable the queue, so we disable here.
673 * We synchronize against interrupts via NAPI_STATE_SCHED */
674 if (napi_schedule_prep(&rq
->napi
)) {
675 virtqueue_disable_cb(rq
->vq
);
677 __napi_schedule(&rq
->napi
);
682 static void refill_work(struct work_struct
*work
)
684 struct virtnet_info
*vi
=
685 container_of(work
, struct virtnet_info
, refill
.work
);
689 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++) {
690 struct receive_queue
*rq
= &vi
->rq
[i
];
692 napi_disable(&rq
->napi
);
693 still_empty
= !try_fill_recv(vi
, rq
, GFP_KERNEL
);
694 virtnet_napi_enable(rq
);
696 /* In theory, this can happen: if we don't get any buffers in
697 * we will *never* try to fill again.
700 schedule_delayed_work(&vi
->refill
, HZ
/2);
704 static int virtnet_receive(struct receive_queue
*rq
, int budget
)
706 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
707 unsigned int len
, received
= 0;
710 while (received
< budget
&&
711 (buf
= virtqueue_get_buf(rq
->vq
, &len
)) != NULL
) {
712 receive_buf(vi
, rq
, buf
, len
);
716 if (rq
->vq
->num_free
> virtqueue_get_vring_size(rq
->vq
) / 2) {
717 if (!try_fill_recv(vi
, rq
, GFP_ATOMIC
))
718 schedule_delayed_work(&vi
->refill
, 0);
724 static int virtnet_poll(struct napi_struct
*napi
, int budget
)
726 struct receive_queue
*rq
=
727 container_of(napi
, struct receive_queue
, napi
);
728 unsigned int r
, received
;
730 received
= virtnet_receive(rq
, budget
);
732 /* Out of packets? */
733 if (received
< budget
) {
734 r
= virtqueue_enable_cb_prepare(rq
->vq
);
735 napi_complete_done(napi
, received
);
736 if (unlikely(virtqueue_poll(rq
->vq
, r
)) &&
737 napi_schedule_prep(napi
)) {
738 virtqueue_disable_cb(rq
->vq
);
739 __napi_schedule(napi
);
746 #ifdef CONFIG_NET_RX_BUSY_POLL
747 /* must be called with local_bh_disable()d */
748 static int virtnet_busy_poll(struct napi_struct
*napi
)
750 struct receive_queue
*rq
=
751 container_of(napi
, struct receive_queue
, napi
);
752 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
753 int r
, received
= 0, budget
= 4;
755 if (!(vi
->status
& VIRTIO_NET_S_LINK_UP
))
756 return LL_FLUSH_FAILED
;
758 if (!napi_schedule_prep(napi
))
759 return LL_FLUSH_BUSY
;
761 virtqueue_disable_cb(rq
->vq
);
764 received
+= virtnet_receive(rq
, budget
);
766 r
= virtqueue_enable_cb_prepare(rq
->vq
);
767 clear_bit(NAPI_STATE_SCHED
, &napi
->state
);
768 if (unlikely(virtqueue_poll(rq
->vq
, r
)) &&
769 napi_schedule_prep(napi
)) {
770 virtqueue_disable_cb(rq
->vq
);
771 if (received
< budget
) {
775 __napi_schedule(napi
);
781 #endif /* CONFIG_NET_RX_BUSY_POLL */
783 static int virtnet_open(struct net_device
*dev
)
785 struct virtnet_info
*vi
= netdev_priv(dev
);
788 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
789 if (i
< vi
->curr_queue_pairs
)
790 /* Make sure we have some buffers: if oom use wq. */
791 if (!try_fill_recv(vi
, &vi
->rq
[i
], GFP_KERNEL
))
792 schedule_delayed_work(&vi
->refill
, 0);
793 virtnet_napi_enable(&vi
->rq
[i
]);
799 static void free_old_xmit_skbs(struct send_queue
*sq
)
803 struct virtnet_info
*vi
= sq
->vq
->vdev
->priv
;
804 struct virtnet_stats
*stats
= this_cpu_ptr(vi
->stats
);
806 while ((skb
= virtqueue_get_buf(sq
->vq
, &len
)) != NULL
) {
807 pr_debug("Sent skb %p\n", skb
);
809 u64_stats_update_begin(&stats
->tx_syncp
);
810 stats
->tx_bytes
+= skb
->len
;
812 u64_stats_update_end(&stats
->tx_syncp
);
814 dev_kfree_skb_any(skb
);
818 static int xmit_skb(struct send_queue
*sq
, struct sk_buff
*skb
)
820 struct virtio_net_hdr_mrg_rxbuf
*hdr
;
821 const unsigned char *dest
= ((struct ethhdr
*)skb
->data
)->h_dest
;
822 struct virtnet_info
*vi
= sq
->vq
->vdev
->priv
;
824 unsigned hdr_len
= vi
->hdr_len
;
827 pr_debug("%s: xmit %p %pM\n", vi
->dev
->name
, skb
, dest
);
829 can_push
= vi
->any_header_sg
&&
830 !((unsigned long)skb
->data
& (__alignof__(*hdr
) - 1)) &&
831 !skb_header_cloned(skb
) && skb_headroom(skb
) >= hdr_len
;
832 /* Even if we can, don't push here yet as this would skew
833 * csum_start offset below. */
835 hdr
= (struct virtio_net_hdr_mrg_rxbuf
*)(skb
->data
- hdr_len
);
837 hdr
= skb_vnet_hdr(skb
);
839 if (virtio_net_hdr_from_skb(skb
, &hdr
->hdr
,
840 virtio_is_little_endian(vi
->vdev
)))
843 if (vi
->mergeable_rx_bufs
)
844 hdr
->num_buffers
= 0;
846 sg_init_table(sq
->sg
, skb_shinfo(skb
)->nr_frags
+ (can_push
? 1 : 2));
848 __skb_push(skb
, hdr_len
);
849 num_sg
= skb_to_sgvec(skb
, sq
->sg
, 0, skb
->len
);
850 /* Pull header back to avoid skew in tx bytes calculations. */
851 __skb_pull(skb
, hdr_len
);
853 sg_set_buf(sq
->sg
, hdr
, hdr_len
);
854 num_sg
= skb_to_sgvec(skb
, sq
->sg
+ 1, 0, skb
->len
) + 1;
856 return virtqueue_add_outbuf(sq
->vq
, sq
->sg
, num_sg
, skb
, GFP_ATOMIC
);
859 static netdev_tx_t
start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
861 struct virtnet_info
*vi
= netdev_priv(dev
);
862 int qnum
= skb_get_queue_mapping(skb
);
863 struct send_queue
*sq
= &vi
->sq
[qnum
];
865 struct netdev_queue
*txq
= netdev_get_tx_queue(dev
, qnum
);
866 bool kick
= !skb
->xmit_more
;
868 /* Free up any pending old buffers before queueing new ones. */
869 free_old_xmit_skbs(sq
);
871 /* timestamp packet in software */
872 skb_tx_timestamp(skb
);
874 /* Try to transmit */
875 err
= xmit_skb(sq
, skb
);
877 /* This should not happen! */
879 dev
->stats
.tx_fifo_errors
++;
882 "Unexpected TXQ (%d) queue failure: %d\n", qnum
, err
);
883 dev
->stats
.tx_dropped
++;
884 dev_kfree_skb_any(skb
);
888 /* Don't wait up for transmitted skbs to be freed. */
892 /* If running out of space, stop queue to avoid getting packets that we
893 * are then unable to transmit.
894 * An alternative would be to force queuing layer to requeue the skb by
895 * returning NETDEV_TX_BUSY. However, NETDEV_TX_BUSY should not be
896 * returned in a normal path of operation: it means that driver is not
897 * maintaining the TX queue stop/start state properly, and causes
898 * the stack to do a non-trivial amount of useless work.
899 * Since most packets only take 1 or 2 ring slots, stopping the queue
900 * early means 16 slots are typically wasted.
902 if (sq
->vq
->num_free
< 2+MAX_SKB_FRAGS
) {
903 netif_stop_subqueue(dev
, qnum
);
904 if (unlikely(!virtqueue_enable_cb_delayed(sq
->vq
))) {
905 /* More just got used, free them then recheck. */
906 free_old_xmit_skbs(sq
);
907 if (sq
->vq
->num_free
>= 2+MAX_SKB_FRAGS
) {
908 netif_start_subqueue(dev
, qnum
);
909 virtqueue_disable_cb(sq
->vq
);
914 if (kick
|| netif_xmit_stopped(txq
))
915 virtqueue_kick(sq
->vq
);
921 * Send command via the control virtqueue and check status. Commands
922 * supported by the hypervisor, as indicated by feature bits, should
923 * never fail unless improperly formatted.
925 static bool virtnet_send_command(struct virtnet_info
*vi
, u8
class, u8 cmd
,
926 struct scatterlist
*out
)
928 struct scatterlist
*sgs
[4], hdr
, stat
;
929 unsigned out_num
= 0, tmp
;
931 /* Caller should know better */
932 BUG_ON(!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
));
934 vi
->ctrl_status
= ~0;
935 vi
->ctrl_hdr
.class = class;
936 vi
->ctrl_hdr
.cmd
= cmd
;
938 sg_init_one(&hdr
, &vi
->ctrl_hdr
, sizeof(vi
->ctrl_hdr
));
939 sgs
[out_num
++] = &hdr
;
942 sgs
[out_num
++] = out
;
944 /* Add return status. */
945 sg_init_one(&stat
, &vi
->ctrl_status
, sizeof(vi
->ctrl_status
));
946 sgs
[out_num
] = &stat
;
948 BUG_ON(out_num
+ 1 > ARRAY_SIZE(sgs
));
949 virtqueue_add_sgs(vi
->cvq
, sgs
, out_num
, 1, vi
, GFP_ATOMIC
);
951 if (unlikely(!virtqueue_kick(vi
->cvq
)))
952 return vi
->ctrl_status
== VIRTIO_NET_OK
;
954 /* Spin for a response, the kick causes an ioport write, trapping
955 * into the hypervisor, so the request should be handled immediately.
957 while (!virtqueue_get_buf(vi
->cvq
, &tmp
) &&
958 !virtqueue_is_broken(vi
->cvq
))
961 return vi
->ctrl_status
== VIRTIO_NET_OK
;
964 static int virtnet_set_mac_address(struct net_device
*dev
, void *p
)
966 struct virtnet_info
*vi
= netdev_priv(dev
);
967 struct virtio_device
*vdev
= vi
->vdev
;
969 struct sockaddr
*addr
= p
;
970 struct scatterlist sg
;
972 ret
= eth_prepare_mac_addr_change(dev
, p
);
976 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_MAC_ADDR
)) {
977 sg_init_one(&sg
, addr
->sa_data
, dev
->addr_len
);
978 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MAC
,
979 VIRTIO_NET_CTRL_MAC_ADDR_SET
, &sg
)) {
981 "Failed to set mac address by vq command.\n");
984 } else if (virtio_has_feature(vdev
, VIRTIO_NET_F_MAC
) &&
985 !virtio_has_feature(vdev
, VIRTIO_F_VERSION_1
)) {
988 /* Naturally, this has an atomicity problem. */
989 for (i
= 0; i
< dev
->addr_len
; i
++)
991 offsetof(struct virtio_net_config
, mac
) +
992 i
, addr
->sa_data
[i
]);
995 eth_commit_mac_addr_change(dev
, p
);
1000 static struct rtnl_link_stats64
*virtnet_stats(struct net_device
*dev
,
1001 struct rtnl_link_stats64
*tot
)
1003 struct virtnet_info
*vi
= netdev_priv(dev
);
1007 for_each_possible_cpu(cpu
) {
1008 struct virtnet_stats
*stats
= per_cpu_ptr(vi
->stats
, cpu
);
1009 u64 tpackets
, tbytes
, rpackets
, rbytes
;
1012 start
= u64_stats_fetch_begin_irq(&stats
->tx_syncp
);
1013 tpackets
= stats
->tx_packets
;
1014 tbytes
= stats
->tx_bytes
;
1015 } while (u64_stats_fetch_retry_irq(&stats
->tx_syncp
, start
));
1018 start
= u64_stats_fetch_begin_irq(&stats
->rx_syncp
);
1019 rpackets
= stats
->rx_packets
;
1020 rbytes
= stats
->rx_bytes
;
1021 } while (u64_stats_fetch_retry_irq(&stats
->rx_syncp
, start
));
1023 tot
->rx_packets
+= rpackets
;
1024 tot
->tx_packets
+= tpackets
;
1025 tot
->rx_bytes
+= rbytes
;
1026 tot
->tx_bytes
+= tbytes
;
1029 tot
->tx_dropped
= dev
->stats
.tx_dropped
;
1030 tot
->tx_fifo_errors
= dev
->stats
.tx_fifo_errors
;
1031 tot
->rx_dropped
= dev
->stats
.rx_dropped
;
1032 tot
->rx_length_errors
= dev
->stats
.rx_length_errors
;
1033 tot
->rx_frame_errors
= dev
->stats
.rx_frame_errors
;
1038 #ifdef CONFIG_NET_POLL_CONTROLLER
1039 static void virtnet_netpoll(struct net_device
*dev
)
1041 struct virtnet_info
*vi
= netdev_priv(dev
);
1044 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++)
1045 napi_schedule(&vi
->rq
[i
].napi
);
1049 static void virtnet_ack_link_announce(struct virtnet_info
*vi
)
1052 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_ANNOUNCE
,
1053 VIRTIO_NET_CTRL_ANNOUNCE_ACK
, NULL
))
1054 dev_warn(&vi
->dev
->dev
, "Failed to ack link announce.\n");
1058 static int virtnet_set_queues(struct virtnet_info
*vi
, u16 queue_pairs
)
1060 struct scatterlist sg
;
1061 struct virtio_net_ctrl_mq s
;
1062 struct net_device
*dev
= vi
->dev
;
1064 if (!vi
->has_cvq
|| !virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_MQ
))
1067 s
.virtqueue_pairs
= cpu_to_virtio16(vi
->vdev
, queue_pairs
);
1068 sg_init_one(&sg
, &s
, sizeof(s
));
1070 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MQ
,
1071 VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET
, &sg
)) {
1072 dev_warn(&dev
->dev
, "Fail to set num of queue pairs to %d\n",
1076 vi
->curr_queue_pairs
= queue_pairs
;
1077 /* virtnet_open() will refill when device is going to up. */
1078 if (dev
->flags
& IFF_UP
)
1079 schedule_delayed_work(&vi
->refill
, 0);
1085 static int virtnet_close(struct net_device
*dev
)
1087 struct virtnet_info
*vi
= netdev_priv(dev
);
1090 /* Make sure refill_work doesn't re-enable napi! */
1091 cancel_delayed_work_sync(&vi
->refill
);
1093 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
1094 napi_disable(&vi
->rq
[i
].napi
);
1099 static void virtnet_set_rx_mode(struct net_device
*dev
)
1101 struct virtnet_info
*vi
= netdev_priv(dev
);
1102 struct scatterlist sg
[2];
1103 struct virtio_net_ctrl_mac
*mac_data
;
1104 struct netdev_hw_addr
*ha
;
1110 /* We can't dynamically set ndo_set_rx_mode, so return gracefully */
1111 if (!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_RX
))
1114 vi
->ctrl_promisc
= ((dev
->flags
& IFF_PROMISC
) != 0);
1115 vi
->ctrl_allmulti
= ((dev
->flags
& IFF_ALLMULTI
) != 0);
1117 sg_init_one(sg
, &vi
->ctrl_promisc
, sizeof(vi
->ctrl_promisc
));
1119 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_RX
,
1120 VIRTIO_NET_CTRL_RX_PROMISC
, sg
))
1121 dev_warn(&dev
->dev
, "Failed to %sable promisc mode.\n",
1122 vi
->ctrl_promisc
? "en" : "dis");
1124 sg_init_one(sg
, &vi
->ctrl_allmulti
, sizeof(vi
->ctrl_allmulti
));
1126 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_RX
,
1127 VIRTIO_NET_CTRL_RX_ALLMULTI
, sg
))
1128 dev_warn(&dev
->dev
, "Failed to %sable allmulti mode.\n",
1129 vi
->ctrl_allmulti
? "en" : "dis");
1131 uc_count
= netdev_uc_count(dev
);
1132 mc_count
= netdev_mc_count(dev
);
1133 /* MAC filter - use one buffer for both lists */
1134 buf
= kzalloc(((uc_count
+ mc_count
) * ETH_ALEN
) +
1135 (2 * sizeof(mac_data
->entries
)), GFP_ATOMIC
);
1140 sg_init_table(sg
, 2);
1142 /* Store the unicast list and count in the front of the buffer */
1143 mac_data
->entries
= cpu_to_virtio32(vi
->vdev
, uc_count
);
1145 netdev_for_each_uc_addr(ha
, dev
)
1146 memcpy(&mac_data
->macs
[i
++][0], ha
->addr
, ETH_ALEN
);
1148 sg_set_buf(&sg
[0], mac_data
,
1149 sizeof(mac_data
->entries
) + (uc_count
* ETH_ALEN
));
1151 /* multicast list and count fill the end */
1152 mac_data
= (void *)&mac_data
->macs
[uc_count
][0];
1154 mac_data
->entries
= cpu_to_virtio32(vi
->vdev
, mc_count
);
1156 netdev_for_each_mc_addr(ha
, dev
)
1157 memcpy(&mac_data
->macs
[i
++][0], ha
->addr
, ETH_ALEN
);
1159 sg_set_buf(&sg
[1], mac_data
,
1160 sizeof(mac_data
->entries
) + (mc_count
* ETH_ALEN
));
1162 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MAC
,
1163 VIRTIO_NET_CTRL_MAC_TABLE_SET
, sg
))
1164 dev_warn(&dev
->dev
, "Failed to set MAC filter table.\n");
1169 static int virtnet_vlan_rx_add_vid(struct net_device
*dev
,
1170 __be16 proto
, u16 vid
)
1172 struct virtnet_info
*vi
= netdev_priv(dev
);
1173 struct scatterlist sg
;
1175 sg_init_one(&sg
, &vid
, sizeof(vid
));
1177 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_VLAN
,
1178 VIRTIO_NET_CTRL_VLAN_ADD
, &sg
))
1179 dev_warn(&dev
->dev
, "Failed to add VLAN ID %d.\n", vid
);
1183 static int virtnet_vlan_rx_kill_vid(struct net_device
*dev
,
1184 __be16 proto
, u16 vid
)
1186 struct virtnet_info
*vi
= netdev_priv(dev
);
1187 struct scatterlist sg
;
1189 sg_init_one(&sg
, &vid
, sizeof(vid
));
1191 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_VLAN
,
1192 VIRTIO_NET_CTRL_VLAN_DEL
, &sg
))
1193 dev_warn(&dev
->dev
, "Failed to kill VLAN ID %d.\n", vid
);
1197 static void virtnet_clean_affinity(struct virtnet_info
*vi
, long hcpu
)
1201 if (vi
->affinity_hint_set
) {
1202 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1203 virtqueue_set_affinity(vi
->rq
[i
].vq
, -1);
1204 virtqueue_set_affinity(vi
->sq
[i
].vq
, -1);
1207 vi
->affinity_hint_set
= false;
1211 static void virtnet_set_affinity(struct virtnet_info
*vi
)
1216 /* In multiqueue mode, when the number of cpu is equal to the number of
1217 * queue pairs, we let the queue pairs to be private to one cpu by
1218 * setting the affinity hint to eliminate the contention.
1220 if (vi
->curr_queue_pairs
== 1 ||
1221 vi
->max_queue_pairs
!= num_online_cpus()) {
1222 virtnet_clean_affinity(vi
, -1);
1227 for_each_online_cpu(cpu
) {
1228 virtqueue_set_affinity(vi
->rq
[i
].vq
, cpu
);
1229 virtqueue_set_affinity(vi
->sq
[i
].vq
, cpu
);
1230 netif_set_xps_queue(vi
->dev
, cpumask_of(cpu
), i
);
1234 vi
->affinity_hint_set
= true;
1237 static int virtnet_cpu_callback(struct notifier_block
*nfb
,
1238 unsigned long action
, void *hcpu
)
1240 struct virtnet_info
*vi
= container_of(nfb
, struct virtnet_info
, nb
);
1242 switch(action
& ~CPU_TASKS_FROZEN
) {
1244 case CPU_DOWN_FAILED
:
1246 virtnet_set_affinity(vi
);
1248 case CPU_DOWN_PREPARE
:
1249 virtnet_clean_affinity(vi
, (long)hcpu
);
1258 static void virtnet_get_ringparam(struct net_device
*dev
,
1259 struct ethtool_ringparam
*ring
)
1261 struct virtnet_info
*vi
= netdev_priv(dev
);
1263 ring
->rx_max_pending
= virtqueue_get_vring_size(vi
->rq
[0].vq
);
1264 ring
->tx_max_pending
= virtqueue_get_vring_size(vi
->sq
[0].vq
);
1265 ring
->rx_pending
= ring
->rx_max_pending
;
1266 ring
->tx_pending
= ring
->tx_max_pending
;
1270 static void virtnet_get_drvinfo(struct net_device
*dev
,
1271 struct ethtool_drvinfo
*info
)
1273 struct virtnet_info
*vi
= netdev_priv(dev
);
1274 struct virtio_device
*vdev
= vi
->vdev
;
1276 strlcpy(info
->driver
, KBUILD_MODNAME
, sizeof(info
->driver
));
1277 strlcpy(info
->version
, VIRTNET_DRIVER_VERSION
, sizeof(info
->version
));
1278 strlcpy(info
->bus_info
, virtio_bus_name(vdev
), sizeof(info
->bus_info
));
1282 /* TODO: Eliminate OOO packets during switching */
1283 static int virtnet_set_channels(struct net_device
*dev
,
1284 struct ethtool_channels
*channels
)
1286 struct virtnet_info
*vi
= netdev_priv(dev
);
1287 u16 queue_pairs
= channels
->combined_count
;
1290 /* We don't support separate rx/tx channels.
1291 * We don't allow setting 'other' channels.
1293 if (channels
->rx_count
|| channels
->tx_count
|| channels
->other_count
)
1296 if (queue_pairs
> vi
->max_queue_pairs
|| queue_pairs
== 0)
1300 err
= virtnet_set_queues(vi
, queue_pairs
);
1302 netif_set_real_num_tx_queues(dev
, queue_pairs
);
1303 netif_set_real_num_rx_queues(dev
, queue_pairs
);
1305 virtnet_set_affinity(vi
);
1312 static void virtnet_get_channels(struct net_device
*dev
,
1313 struct ethtool_channels
*channels
)
1315 struct virtnet_info
*vi
= netdev_priv(dev
);
1317 channels
->combined_count
= vi
->curr_queue_pairs
;
1318 channels
->max_combined
= vi
->max_queue_pairs
;
1319 channels
->max_other
= 0;
1320 channels
->rx_count
= 0;
1321 channels
->tx_count
= 0;
1322 channels
->other_count
= 0;
1325 /* Check if the user is trying to change anything besides speed/duplex */
1326 static bool virtnet_validate_ethtool_cmd(const struct ethtool_cmd
*cmd
)
1328 struct ethtool_cmd diff1
= *cmd
;
1329 struct ethtool_cmd diff2
= {};
1331 /* cmd is always set so we need to clear it, validate the port type
1332 * and also without autonegotiation we can ignore advertising
1334 ethtool_cmd_speed_set(&diff1
, 0);
1335 diff2
.port
= PORT_OTHER
;
1336 diff1
.advertising
= 0;
1340 return !memcmp(&diff1
, &diff2
, sizeof(diff1
));
1343 static int virtnet_set_settings(struct net_device
*dev
, struct ethtool_cmd
*cmd
)
1345 struct virtnet_info
*vi
= netdev_priv(dev
);
1348 speed
= ethtool_cmd_speed(cmd
);
1349 /* don't allow custom speed and duplex */
1350 if (!ethtool_validate_speed(speed
) ||
1351 !ethtool_validate_duplex(cmd
->duplex
) ||
1352 !virtnet_validate_ethtool_cmd(cmd
))
1355 vi
->duplex
= cmd
->duplex
;
1360 static int virtnet_get_settings(struct net_device
*dev
, struct ethtool_cmd
*cmd
)
1362 struct virtnet_info
*vi
= netdev_priv(dev
);
1364 ethtool_cmd_speed_set(cmd
, vi
->speed
);
1365 cmd
->duplex
= vi
->duplex
;
1366 cmd
->port
= PORT_OTHER
;
1371 static void virtnet_init_settings(struct net_device
*dev
)
1373 struct virtnet_info
*vi
= netdev_priv(dev
);
1375 vi
->speed
= SPEED_UNKNOWN
;
1376 vi
->duplex
= DUPLEX_UNKNOWN
;
1379 static const struct ethtool_ops virtnet_ethtool_ops
= {
1380 .get_drvinfo
= virtnet_get_drvinfo
,
1381 .get_link
= ethtool_op_get_link
,
1382 .get_ringparam
= virtnet_get_ringparam
,
1383 .set_channels
= virtnet_set_channels
,
1384 .get_channels
= virtnet_get_channels
,
1385 .get_ts_info
= ethtool_op_get_ts_info
,
1386 .get_settings
= virtnet_get_settings
,
1387 .set_settings
= virtnet_set_settings
,
1391 #define MAX_MTU 65535
1393 static int virtnet_change_mtu(struct net_device
*dev
, int new_mtu
)
1395 if (new_mtu
< MIN_MTU
|| new_mtu
> MAX_MTU
)
1401 static const struct net_device_ops virtnet_netdev
= {
1402 .ndo_open
= virtnet_open
,
1403 .ndo_stop
= virtnet_close
,
1404 .ndo_start_xmit
= start_xmit
,
1405 .ndo_validate_addr
= eth_validate_addr
,
1406 .ndo_set_mac_address
= virtnet_set_mac_address
,
1407 .ndo_set_rx_mode
= virtnet_set_rx_mode
,
1408 .ndo_change_mtu
= virtnet_change_mtu
,
1409 .ndo_get_stats64
= virtnet_stats
,
1410 .ndo_vlan_rx_add_vid
= virtnet_vlan_rx_add_vid
,
1411 .ndo_vlan_rx_kill_vid
= virtnet_vlan_rx_kill_vid
,
1412 #ifdef CONFIG_NET_POLL_CONTROLLER
1413 .ndo_poll_controller
= virtnet_netpoll
,
1415 #ifdef CONFIG_NET_RX_BUSY_POLL
1416 .ndo_busy_poll
= virtnet_busy_poll
,
1420 static void virtnet_config_changed_work(struct work_struct
*work
)
1422 struct virtnet_info
*vi
=
1423 container_of(work
, struct virtnet_info
, config_work
);
1426 if (virtio_cread_feature(vi
->vdev
, VIRTIO_NET_F_STATUS
,
1427 struct virtio_net_config
, status
, &v
) < 0)
1430 if (v
& VIRTIO_NET_S_ANNOUNCE
) {
1431 netdev_notify_peers(vi
->dev
);
1432 virtnet_ack_link_announce(vi
);
1435 /* Ignore unknown (future) status bits */
1436 v
&= VIRTIO_NET_S_LINK_UP
;
1438 if (vi
->status
== v
)
1443 if (vi
->status
& VIRTIO_NET_S_LINK_UP
) {
1444 netif_carrier_on(vi
->dev
);
1445 netif_tx_wake_all_queues(vi
->dev
);
1447 netif_carrier_off(vi
->dev
);
1448 netif_tx_stop_all_queues(vi
->dev
);
1452 static void virtnet_config_changed(struct virtio_device
*vdev
)
1454 struct virtnet_info
*vi
= vdev
->priv
;
1456 schedule_work(&vi
->config_work
);
1459 static void virtnet_free_queues(struct virtnet_info
*vi
)
1463 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1464 napi_hash_del(&vi
->rq
[i
].napi
);
1465 netif_napi_del(&vi
->rq
[i
].napi
);
1472 static void free_receive_bufs(struct virtnet_info
*vi
)
1476 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1477 while (vi
->rq
[i
].pages
)
1478 __free_pages(get_a_page(&vi
->rq
[i
], GFP_KERNEL
), 0);
1482 static void free_receive_page_frags(struct virtnet_info
*vi
)
1485 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
1486 if (vi
->rq
[i
].alloc_frag
.page
)
1487 put_page(vi
->rq
[i
].alloc_frag
.page
);
1490 static void free_unused_bufs(struct virtnet_info
*vi
)
1495 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1496 struct virtqueue
*vq
= vi
->sq
[i
].vq
;
1497 while ((buf
= virtqueue_detach_unused_buf(vq
)) != NULL
)
1501 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1502 struct virtqueue
*vq
= vi
->rq
[i
].vq
;
1504 while ((buf
= virtqueue_detach_unused_buf(vq
)) != NULL
) {
1505 if (vi
->mergeable_rx_bufs
) {
1506 unsigned long ctx
= (unsigned long)buf
;
1507 void *base
= mergeable_ctx_to_buf_address(ctx
);
1508 put_page(virt_to_head_page(base
));
1509 } else if (vi
->big_packets
) {
1510 give_pages(&vi
->rq
[i
], buf
);
1518 static void virtnet_del_vqs(struct virtnet_info
*vi
)
1520 struct virtio_device
*vdev
= vi
->vdev
;
1522 virtnet_clean_affinity(vi
, -1);
1524 vdev
->config
->del_vqs(vdev
);
1526 virtnet_free_queues(vi
);
1529 static int virtnet_find_vqs(struct virtnet_info
*vi
)
1531 vq_callback_t
**callbacks
;
1532 struct virtqueue
**vqs
;
1537 /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
1538 * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
1539 * possible control vq.
1541 total_vqs
= vi
->max_queue_pairs
* 2 +
1542 virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
);
1544 /* Allocate space for find_vqs parameters */
1545 vqs
= kzalloc(total_vqs
* sizeof(*vqs
), GFP_KERNEL
);
1548 callbacks
= kmalloc(total_vqs
* sizeof(*callbacks
), GFP_KERNEL
);
1551 names
= kmalloc(total_vqs
* sizeof(*names
), GFP_KERNEL
);
1555 /* Parameters for control virtqueue, if any */
1557 callbacks
[total_vqs
- 1] = NULL
;
1558 names
[total_vqs
- 1] = "control";
1561 /* Allocate/initialize parameters for send/receive virtqueues */
1562 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1563 callbacks
[rxq2vq(i
)] = skb_recv_done
;
1564 callbacks
[txq2vq(i
)] = skb_xmit_done
;
1565 sprintf(vi
->rq
[i
].name
, "input.%d", i
);
1566 sprintf(vi
->sq
[i
].name
, "output.%d", i
);
1567 names
[rxq2vq(i
)] = vi
->rq
[i
].name
;
1568 names
[txq2vq(i
)] = vi
->sq
[i
].name
;
1571 ret
= vi
->vdev
->config
->find_vqs(vi
->vdev
, total_vqs
, vqs
, callbacks
,
1577 vi
->cvq
= vqs
[total_vqs
- 1];
1578 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VLAN
))
1579 vi
->dev
->features
|= NETIF_F_HW_VLAN_CTAG_FILTER
;
1582 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1583 vi
->rq
[i
].vq
= vqs
[rxq2vq(i
)];
1584 vi
->sq
[i
].vq
= vqs
[txq2vq(i
)];
1603 static int virtnet_alloc_queues(struct virtnet_info
*vi
)
1607 vi
->sq
= kzalloc(sizeof(*vi
->sq
) * vi
->max_queue_pairs
, GFP_KERNEL
);
1610 vi
->rq
= kzalloc(sizeof(*vi
->rq
) * vi
->max_queue_pairs
, GFP_KERNEL
);
1614 INIT_DELAYED_WORK(&vi
->refill
, refill_work
);
1615 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1616 vi
->rq
[i
].pages
= NULL
;
1617 netif_napi_add(vi
->dev
, &vi
->rq
[i
].napi
, virtnet_poll
,
1620 sg_init_table(vi
->rq
[i
].sg
, ARRAY_SIZE(vi
->rq
[i
].sg
));
1621 ewma_pkt_len_init(&vi
->rq
[i
].mrg_avg_pkt_len
);
1622 sg_init_table(vi
->sq
[i
].sg
, ARRAY_SIZE(vi
->sq
[i
].sg
));
1633 static int init_vqs(struct virtnet_info
*vi
)
1637 /* Allocate send & receive queues */
1638 ret
= virtnet_alloc_queues(vi
);
1642 ret
= virtnet_find_vqs(vi
);
1647 virtnet_set_affinity(vi
);
1653 virtnet_free_queues(vi
);
1659 static ssize_t
mergeable_rx_buffer_size_show(struct netdev_rx_queue
*queue
,
1660 struct rx_queue_attribute
*attribute
, char *buf
)
1662 struct virtnet_info
*vi
= netdev_priv(queue
->dev
);
1663 unsigned int queue_index
= get_netdev_rx_queue_index(queue
);
1664 struct ewma_pkt_len
*avg
;
1666 BUG_ON(queue_index
>= vi
->max_queue_pairs
);
1667 avg
= &vi
->rq
[queue_index
].mrg_avg_pkt_len
;
1668 return sprintf(buf
, "%u\n", get_mergeable_buf_len(avg
));
1671 static struct rx_queue_attribute mergeable_rx_buffer_size_attribute
=
1672 __ATTR_RO(mergeable_rx_buffer_size
);
1674 static struct attribute
*virtio_net_mrg_rx_attrs
[] = {
1675 &mergeable_rx_buffer_size_attribute
.attr
,
1679 static const struct attribute_group virtio_net_mrg_rx_group
= {
1680 .name
= "virtio_net",
1681 .attrs
= virtio_net_mrg_rx_attrs
1685 static bool virtnet_fail_on_feature(struct virtio_device
*vdev
,
1687 const char *fname
, const char *dname
)
1689 if (!virtio_has_feature(vdev
, fbit
))
1692 dev_err(&vdev
->dev
, "device advertises feature %s but not %s",
1698 #define VIRTNET_FAIL_ON(vdev, fbit, dbit) \
1699 virtnet_fail_on_feature(vdev, fbit, #fbit, dbit)
1701 static bool virtnet_validate_features(struct virtio_device
*vdev
)
1703 if (!virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
) &&
1704 (VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_CTRL_RX
,
1705 "VIRTIO_NET_F_CTRL_VQ") ||
1706 VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_CTRL_VLAN
,
1707 "VIRTIO_NET_F_CTRL_VQ") ||
1708 VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_GUEST_ANNOUNCE
,
1709 "VIRTIO_NET_F_CTRL_VQ") ||
1710 VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_MQ
, "VIRTIO_NET_F_CTRL_VQ") ||
1711 VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_CTRL_MAC_ADDR
,
1712 "VIRTIO_NET_F_CTRL_VQ"))) {
1719 static int virtnet_probe(struct virtio_device
*vdev
)
1722 struct net_device
*dev
;
1723 struct virtnet_info
*vi
;
1724 u16 max_queue_pairs
;
1727 if (!vdev
->config
->get
) {
1728 dev_err(&vdev
->dev
, "%s failure: config access disabled\n",
1733 if (!virtnet_validate_features(vdev
))
1736 /* Find if host supports multiqueue virtio_net device */
1737 err
= virtio_cread_feature(vdev
, VIRTIO_NET_F_MQ
,
1738 struct virtio_net_config
,
1739 max_virtqueue_pairs
, &max_queue_pairs
);
1741 /* We need at least 2 queue's */
1742 if (err
|| max_queue_pairs
< VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN
||
1743 max_queue_pairs
> VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX
||
1744 !virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
))
1745 max_queue_pairs
= 1;
1747 /* Allocate ourselves a network device with room for our info */
1748 dev
= alloc_etherdev_mq(sizeof(struct virtnet_info
), max_queue_pairs
);
1752 /* Set up network device as normal. */
1753 dev
->priv_flags
|= IFF_UNICAST_FLT
| IFF_LIVE_ADDR_CHANGE
;
1754 dev
->netdev_ops
= &virtnet_netdev
;
1755 dev
->features
= NETIF_F_HIGHDMA
;
1757 dev
->ethtool_ops
= &virtnet_ethtool_ops
;
1758 SET_NETDEV_DEV(dev
, &vdev
->dev
);
1760 /* Do we support "hardware" checksums? */
1761 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CSUM
)) {
1762 /* This opens up the world of extra features. */
1763 dev
->hw_features
|= NETIF_F_HW_CSUM
| NETIF_F_SG
;
1765 dev
->features
|= NETIF_F_HW_CSUM
| NETIF_F_SG
;
1767 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GSO
)) {
1768 dev
->hw_features
|= NETIF_F_TSO
| NETIF_F_UFO
1769 | NETIF_F_TSO_ECN
| NETIF_F_TSO6
;
1771 /* Individual feature bits: what can host handle? */
1772 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_TSO4
))
1773 dev
->hw_features
|= NETIF_F_TSO
;
1774 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_TSO6
))
1775 dev
->hw_features
|= NETIF_F_TSO6
;
1776 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_ECN
))
1777 dev
->hw_features
|= NETIF_F_TSO_ECN
;
1778 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_UFO
))
1779 dev
->hw_features
|= NETIF_F_UFO
;
1781 dev
->features
|= NETIF_F_GSO_ROBUST
;
1784 dev
->features
|= dev
->hw_features
& (NETIF_F_ALL_TSO
|NETIF_F_UFO
);
1785 /* (!csum && gso) case will be fixed by register_netdev() */
1787 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_CSUM
))
1788 dev
->features
|= NETIF_F_RXCSUM
;
1790 dev
->vlan_features
= dev
->features
;
1792 /* Configuration may specify what MAC to use. Otherwise random. */
1793 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MAC
))
1794 virtio_cread_bytes(vdev
,
1795 offsetof(struct virtio_net_config
, mac
),
1796 dev
->dev_addr
, dev
->addr_len
);
1798 eth_hw_addr_random(dev
);
1800 /* Set up our device-specific information */
1801 vi
= netdev_priv(dev
);
1805 vi
->stats
= alloc_percpu(struct virtnet_stats
);
1807 if (vi
->stats
== NULL
)
1810 for_each_possible_cpu(i
) {
1811 struct virtnet_stats
*virtnet_stats
;
1812 virtnet_stats
= per_cpu_ptr(vi
->stats
, i
);
1813 u64_stats_init(&virtnet_stats
->tx_syncp
);
1814 u64_stats_init(&virtnet_stats
->rx_syncp
);
1817 INIT_WORK(&vi
->config_work
, virtnet_config_changed_work
);
1819 /* If we can receive ANY GSO packets, we must allocate large ones. */
1820 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_TSO4
) ||
1821 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_TSO6
) ||
1822 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_ECN
) ||
1823 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_UFO
))
1824 vi
->big_packets
= true;
1826 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MRG_RXBUF
))
1827 vi
->mergeable_rx_bufs
= true;
1829 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MRG_RXBUF
) ||
1830 virtio_has_feature(vdev
, VIRTIO_F_VERSION_1
))
1831 vi
->hdr_len
= sizeof(struct virtio_net_hdr_mrg_rxbuf
);
1833 vi
->hdr_len
= sizeof(struct virtio_net_hdr
);
1835 if (virtio_has_feature(vdev
, VIRTIO_F_ANY_LAYOUT
) ||
1836 virtio_has_feature(vdev
, VIRTIO_F_VERSION_1
))
1837 vi
->any_header_sg
= true;
1839 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
))
1842 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MTU
)) {
1843 mtu
= virtio_cread16(vdev
,
1844 offsetof(struct virtio_net_config
,
1846 if (virtnet_change_mtu(dev
, mtu
))
1847 __virtio_clear_bit(vdev
, VIRTIO_NET_F_MTU
);
1850 if (vi
->any_header_sg
)
1851 dev
->needed_headroom
= vi
->hdr_len
;
1853 /* Use single tx/rx queue pair as default */
1854 vi
->curr_queue_pairs
= 1;
1855 vi
->max_queue_pairs
= max_queue_pairs
;
1857 /* Allocate/initialize the rx/tx queues, and invoke find_vqs */
1863 if (vi
->mergeable_rx_bufs
)
1864 dev
->sysfs_rx_queue_group
= &virtio_net_mrg_rx_group
;
1866 netif_set_real_num_tx_queues(dev
, vi
->curr_queue_pairs
);
1867 netif_set_real_num_rx_queues(dev
, vi
->curr_queue_pairs
);
1869 virtnet_init_settings(dev
);
1871 err
= register_netdev(dev
);
1873 pr_debug("virtio_net: registering device failed\n");
1877 virtio_device_ready(vdev
);
1879 vi
->nb
.notifier_call
= &virtnet_cpu_callback
;
1880 err
= register_hotcpu_notifier(&vi
->nb
);
1882 pr_debug("virtio_net: registering cpu notifier failed\n");
1883 goto free_unregister_netdev
;
1886 /* Assume link up if device can't report link status,
1887 otherwise get link status from config. */
1888 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_STATUS
)) {
1889 netif_carrier_off(dev
);
1890 schedule_work(&vi
->config_work
);
1892 vi
->status
= VIRTIO_NET_S_LINK_UP
;
1893 netif_carrier_on(dev
);
1896 pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
1897 dev
->name
, max_queue_pairs
);
1901 free_unregister_netdev
:
1902 vi
->vdev
->config
->reset(vdev
);
1904 unregister_netdev(dev
);
1906 cancel_delayed_work_sync(&vi
->refill
);
1907 free_receive_page_frags(vi
);
1908 virtnet_del_vqs(vi
);
1910 free_percpu(vi
->stats
);
1916 static void remove_vq_common(struct virtnet_info
*vi
)
1918 vi
->vdev
->config
->reset(vi
->vdev
);
1920 /* Free unused buffers in both send and recv, if any. */
1921 free_unused_bufs(vi
);
1923 free_receive_bufs(vi
);
1925 free_receive_page_frags(vi
);
1927 virtnet_del_vqs(vi
);
1930 static void virtnet_remove(struct virtio_device
*vdev
)
1932 struct virtnet_info
*vi
= vdev
->priv
;
1934 unregister_hotcpu_notifier(&vi
->nb
);
1936 /* Make sure no work handler is accessing the device. */
1937 flush_work(&vi
->config_work
);
1939 unregister_netdev(vi
->dev
);
1941 remove_vq_common(vi
);
1943 free_percpu(vi
->stats
);
1944 free_netdev(vi
->dev
);
1947 #ifdef CONFIG_PM_SLEEP
1948 static int virtnet_freeze(struct virtio_device
*vdev
)
1950 struct virtnet_info
*vi
= vdev
->priv
;
1953 unregister_hotcpu_notifier(&vi
->nb
);
1955 /* Make sure no work handler is accessing the device */
1956 flush_work(&vi
->config_work
);
1958 netif_device_detach(vi
->dev
);
1959 cancel_delayed_work_sync(&vi
->refill
);
1961 if (netif_running(vi
->dev
)) {
1962 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
1963 napi_disable(&vi
->rq
[i
].napi
);
1966 remove_vq_common(vi
);
1971 static int virtnet_restore(struct virtio_device
*vdev
)
1973 struct virtnet_info
*vi
= vdev
->priv
;
1980 virtio_device_ready(vdev
);
1982 if (netif_running(vi
->dev
)) {
1983 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++)
1984 if (!try_fill_recv(vi
, &vi
->rq
[i
], GFP_KERNEL
))
1985 schedule_delayed_work(&vi
->refill
, 0);
1987 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
1988 virtnet_napi_enable(&vi
->rq
[i
]);
1991 netif_device_attach(vi
->dev
);
1994 virtnet_set_queues(vi
, vi
->curr_queue_pairs
);
1997 err
= register_hotcpu_notifier(&vi
->nb
);
2005 static struct virtio_device_id id_table
[] = {
2006 { VIRTIO_ID_NET
, VIRTIO_DEV_ANY_ID
},
2010 static unsigned int features
[] = {
2011 VIRTIO_NET_F_CSUM
, VIRTIO_NET_F_GUEST_CSUM
,
2012 VIRTIO_NET_F_GSO
, VIRTIO_NET_F_MAC
,
2013 VIRTIO_NET_F_HOST_TSO4
, VIRTIO_NET_F_HOST_UFO
, VIRTIO_NET_F_HOST_TSO6
,
2014 VIRTIO_NET_F_HOST_ECN
, VIRTIO_NET_F_GUEST_TSO4
, VIRTIO_NET_F_GUEST_TSO6
,
2015 VIRTIO_NET_F_GUEST_ECN
, VIRTIO_NET_F_GUEST_UFO
,
2016 VIRTIO_NET_F_MRG_RXBUF
, VIRTIO_NET_F_STATUS
, VIRTIO_NET_F_CTRL_VQ
,
2017 VIRTIO_NET_F_CTRL_RX
, VIRTIO_NET_F_CTRL_VLAN
,
2018 VIRTIO_NET_F_GUEST_ANNOUNCE
, VIRTIO_NET_F_MQ
,
2019 VIRTIO_NET_F_CTRL_MAC_ADDR
,
2020 VIRTIO_F_ANY_LAYOUT
,
2024 static struct virtio_driver virtio_net_driver
= {
2025 .feature_table
= features
,
2026 .feature_table_size
= ARRAY_SIZE(features
),
2027 .driver
.name
= KBUILD_MODNAME
,
2028 .driver
.owner
= THIS_MODULE
,
2029 .id_table
= id_table
,
2030 .probe
= virtnet_probe
,
2031 .remove
= virtnet_remove
,
2032 .config_changed
= virtnet_config_changed
,
2033 #ifdef CONFIG_PM_SLEEP
2034 .freeze
= virtnet_freeze
,
2035 .restore
= virtnet_restore
,
2039 module_virtio_driver(virtio_net_driver
);
2041 MODULE_DEVICE_TABLE(virtio
, id_table
);
2042 MODULE_DESCRIPTION("Virtio network driver");
2043 MODULE_LICENSE("GPL");