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 /* Weight used for the RX packet size EWMA. The average packet size is used to
44 * determine the packet buffer size when refilling RX rings. As the entire RX
45 * ring may be refilled at once, the weight is chosen so that the EWMA will be
46 * insensitive to short-term, transient changes in packet size.
48 #define RECEIVE_AVG_WEIGHT 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.$index */
77 /* Internal representation of a receive virtqueue */
78 struct receive_queue
{
79 /* Virtqueue associated with this receive_queue */
82 struct napi_struct napi
;
84 /* Chain pages by the private ptr. */
87 /* Average packet length for mergeable receive buffers. */
88 struct ewma mrg_avg_pkt_len
;
90 /* Page frag for packet buffer allocation. */
91 struct page_frag alloc_frag
;
93 /* RX: fragments + linear part + virtio header */
94 struct scatterlist sg
[MAX_SKB_FRAGS
+ 2];
96 /* Name of this receive queue: input.$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 /* enable config space updates */
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 /* Lock for config space updates */
139 struct mutex config_lock
;
141 /* Does the affinity hint is set for virtqueues? */
142 bool affinity_hint_set
;
144 /* CPU hot plug notifier */
145 struct notifier_block nb
;
148 struct skb_vnet_hdr
{
150 struct virtio_net_hdr hdr
;
151 struct virtio_net_hdr_mrg_rxbuf mhdr
;
155 struct padded_vnet_hdr
{
156 struct virtio_net_hdr hdr
;
158 * virtio_net_hdr should be in a separated sg buffer because of a
159 * QEMU bug, and data sg buffer shares same page with this header sg.
160 * This padding makes next sg 16 byte aligned after virtio_net_hdr.
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 skb_vnet_hdr
*skb_vnet_hdr(struct sk_buff
*skb
)
190 return (struct skb_vnet_hdr
*)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 receive_queue
*rq
,
251 struct page
*page
, unsigned int offset
,
252 unsigned int len
, unsigned int truesize
)
254 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
256 struct skb_vnet_hdr
*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
= netdev_alloc_skb_ip_align(vi
->dev
, GOOD_COPY_LEN
);
267 hdr
= skb_vnet_hdr(skb
);
269 if (vi
->mergeable_rx_bufs
) {
270 hdr_len
= sizeof hdr
->mhdr
;
271 hdr_padded_len
= sizeof hdr
->mhdr
;
273 hdr_len
= sizeof hdr
->hdr
;
274 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(void *buf
, unsigned int len
)
328 struct sk_buff
* skb
= buf
;
330 len
-= sizeof(struct virtio_net_hdr
);
336 static struct sk_buff
*receive_big(struct net_device
*dev
,
337 struct receive_queue
*rq
,
341 struct page
*page
= buf
;
342 struct sk_buff
*skb
= page_to_skb(rq
, page
, 0, len
, PAGE_SIZE
);
350 dev
->stats
.rx_dropped
++;
351 give_pages(rq
, page
);
355 static struct sk_buff
*receive_mergeable(struct net_device
*dev
,
356 struct receive_queue
*rq
,
360 void *buf
= mergeable_ctx_to_buf_address(ctx
);
361 struct skb_vnet_hdr
*hdr
= buf
;
362 int num_buf
= hdr
->mhdr
.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(rq
, page
, offset
, len
, truesize
);
368 struct sk_buff
*curr_skb
= head_skb
;
370 if (unlikely(!curr_skb
))
375 ctx
= (unsigned long)virtqueue_get_buf(rq
->vq
, &len
);
376 if (unlikely(!ctx
)) {
377 pr_debug("%s: rx error: %d buffers out of %d missing\n",
378 dev
->name
, num_buf
, hdr
->mhdr
.num_buffers
);
379 dev
->stats
.rx_length_errors
++;
383 buf
= mergeable_ctx_to_buf_address(ctx
);
384 page
= virt_to_head_page(buf
);
386 num_skb_frags
= skb_shinfo(curr_skb
)->nr_frags
;
387 if (unlikely(num_skb_frags
== MAX_SKB_FRAGS
)) {
388 struct sk_buff
*nskb
= alloc_skb(0, GFP_ATOMIC
);
392 if (curr_skb
== head_skb
)
393 skb_shinfo(curr_skb
)->frag_list
= nskb
;
395 curr_skb
->next
= nskb
;
397 head_skb
->truesize
+= nskb
->truesize
;
400 truesize
= max(len
, mergeable_ctx_to_buf_truesize(ctx
));
401 if (curr_skb
!= head_skb
) {
402 head_skb
->data_len
+= len
;
403 head_skb
->len
+= len
;
404 head_skb
->truesize
+= truesize
;
406 offset
= buf
- page_address(page
);
407 if (skb_can_coalesce(curr_skb
, num_skb_frags
, page
, offset
)) {
409 skb_coalesce_rx_frag(curr_skb
, num_skb_frags
- 1,
412 skb_add_rx_frag(curr_skb
, num_skb_frags
, page
,
413 offset
, len
, truesize
);
417 ewma_add(&rq
->mrg_avg_pkt_len
, head_skb
->len
);
423 ctx
= (unsigned long)virtqueue_get_buf(rq
->vq
, &len
);
424 if (unlikely(!ctx
)) {
425 pr_debug("%s: rx error: %d buffers missing\n",
427 dev
->stats
.rx_length_errors
++;
430 page
= virt_to_head_page(mergeable_ctx_to_buf_address(ctx
));
434 dev
->stats
.rx_dropped
++;
435 dev_kfree_skb(head_skb
);
439 static void receive_buf(struct receive_queue
*rq
, void *buf
, unsigned int len
)
441 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
442 struct net_device
*dev
= vi
->dev
;
443 struct virtnet_stats
*stats
= this_cpu_ptr(vi
->stats
);
445 struct skb_vnet_hdr
*hdr
;
447 if (unlikely(len
< sizeof(struct virtio_net_hdr
) + ETH_HLEN
)) {
448 pr_debug("%s: short packet %i\n", dev
->name
, len
);
449 dev
->stats
.rx_length_errors
++;
450 if (vi
->mergeable_rx_bufs
) {
451 unsigned long ctx
= (unsigned long)buf
;
452 void *base
= mergeable_ctx_to_buf_address(ctx
);
453 put_page(virt_to_head_page(base
));
454 } else if (vi
->big_packets
) {
462 if (vi
->mergeable_rx_bufs
)
463 skb
= receive_mergeable(dev
, rq
, (unsigned long)buf
, len
);
464 else if (vi
->big_packets
)
465 skb
= receive_big(dev
, rq
, buf
, len
);
467 skb
= receive_small(buf
, len
);
472 hdr
= skb_vnet_hdr(skb
);
474 u64_stats_update_begin(&stats
->rx_syncp
);
475 stats
->rx_bytes
+= skb
->len
;
477 u64_stats_update_end(&stats
->rx_syncp
);
479 if (hdr
->hdr
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) {
480 pr_debug("Needs csum!\n");
481 if (!skb_partial_csum_set(skb
,
483 hdr
->hdr
.csum_offset
))
485 } else if (hdr
->hdr
.flags
& VIRTIO_NET_HDR_F_DATA_VALID
) {
486 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
489 skb
->protocol
= eth_type_trans(skb
, dev
);
490 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
491 ntohs(skb
->protocol
), skb
->len
, skb
->pkt_type
);
493 if (hdr
->hdr
.gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
495 switch (hdr
->hdr
.gso_type
& ~VIRTIO_NET_HDR_GSO_ECN
) {
496 case VIRTIO_NET_HDR_GSO_TCPV4
:
497 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV4
;
499 case VIRTIO_NET_HDR_GSO_UDP
:
500 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
502 case VIRTIO_NET_HDR_GSO_TCPV6
:
503 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV6
;
506 net_warn_ratelimited("%s: bad gso type %u.\n",
507 dev
->name
, hdr
->hdr
.gso_type
);
511 if (hdr
->hdr
.gso_type
& VIRTIO_NET_HDR_GSO_ECN
)
512 skb_shinfo(skb
)->gso_type
|= SKB_GSO_TCP_ECN
;
514 skb_shinfo(skb
)->gso_size
= hdr
->hdr
.gso_size
;
515 if (skb_shinfo(skb
)->gso_size
== 0) {
516 net_warn_ratelimited("%s: zero gso size.\n", dev
->name
);
520 /* Header must be checked, and gso_segs computed. */
521 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
522 skb_shinfo(skb
)->gso_segs
= 0;
525 skb_mark_napi_id(skb
, &rq
->napi
);
527 netif_receive_skb(skb
);
531 dev
->stats
.rx_frame_errors
++;
535 static int add_recvbuf_small(struct receive_queue
*rq
, gfp_t gfp
)
537 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
539 struct skb_vnet_hdr
*hdr
;
542 skb
= __netdev_alloc_skb_ip_align(vi
->dev
, GOOD_PACKET_LEN
, gfp
);
546 skb_put(skb
, GOOD_PACKET_LEN
);
548 hdr
= skb_vnet_hdr(skb
);
549 sg_set_buf(rq
->sg
, &hdr
->hdr
, sizeof hdr
->hdr
);
551 skb_to_sgvec(skb
, rq
->sg
+ 1, 0, skb
->len
);
553 err
= virtqueue_add_inbuf(rq
->vq
, rq
->sg
, 2, skb
, gfp
);
560 static int add_recvbuf_big(struct receive_queue
*rq
, gfp_t gfp
)
562 struct page
*first
, *list
= NULL
;
566 /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */
567 for (i
= MAX_SKB_FRAGS
+ 1; i
> 1; --i
) {
568 first
= get_a_page(rq
, gfp
);
571 give_pages(rq
, list
);
574 sg_set_buf(&rq
->sg
[i
], page_address(first
), PAGE_SIZE
);
576 /* chain new page in list head to match sg */
577 first
->private = (unsigned long)list
;
581 first
= get_a_page(rq
, gfp
);
583 give_pages(rq
, list
);
586 p
= page_address(first
);
588 /* rq->sg[0], rq->sg[1] share the same page */
589 /* a separated rq->sg[0] for virtio_net_hdr only due to QEMU bug */
590 sg_set_buf(&rq
->sg
[0], p
, sizeof(struct virtio_net_hdr
));
592 /* rq->sg[1] for data packet, from offset */
593 offset
= sizeof(struct padded_vnet_hdr
);
594 sg_set_buf(&rq
->sg
[1], p
+ offset
, PAGE_SIZE
- offset
);
596 /* chain first in list head */
597 first
->private = (unsigned long)list
;
598 err
= virtqueue_add_inbuf(rq
->vq
, rq
->sg
, MAX_SKB_FRAGS
+ 2,
601 give_pages(rq
, first
);
606 static unsigned int get_mergeable_buf_len(struct ewma
*avg_pkt_len
)
608 const size_t hdr_len
= sizeof(struct virtio_net_hdr_mrg_rxbuf
);
611 len
= hdr_len
+ clamp_t(unsigned int, ewma_read(avg_pkt_len
),
612 GOOD_PACKET_LEN
, PAGE_SIZE
- hdr_len
);
613 return ALIGN(len
, MERGEABLE_BUFFER_ALIGN
);
616 static int add_recvbuf_mergeable(struct receive_queue
*rq
, gfp_t gfp
)
618 struct page_frag
*alloc_frag
= &rq
->alloc_frag
;
622 unsigned int len
, hole
;
624 len
= get_mergeable_buf_len(&rq
->mrg_avg_pkt_len
);
625 if (unlikely(!skb_page_frag_refill(len
, alloc_frag
, gfp
)))
628 buf
= (char *)page_address(alloc_frag
->page
) + alloc_frag
->offset
;
629 ctx
= mergeable_buf_to_ctx(buf
, len
);
630 get_page(alloc_frag
->page
);
631 alloc_frag
->offset
+= len
;
632 hole
= alloc_frag
->size
- alloc_frag
->offset
;
634 /* To avoid internal fragmentation, if there is very likely not
635 * enough space for another buffer, add the remaining space to
636 * the current buffer. This extra space is not included in
637 * the truesize stored in ctx.
640 alloc_frag
->offset
+= hole
;
643 sg_init_one(rq
->sg
, buf
, len
);
644 err
= virtqueue_add_inbuf(rq
->vq
, rq
->sg
, 1, (void *)ctx
, gfp
);
646 put_page(virt_to_head_page(buf
));
652 * Returns false if we couldn't fill entirely (OOM).
654 * Normally run in the receive path, but can also be run from ndo_open
655 * before we're receiving packets, or from refill_work which is
656 * careful to disable receiving (using napi_disable).
658 static bool try_fill_recv(struct receive_queue
*rq
, gfp_t gfp
)
660 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
666 if (vi
->mergeable_rx_bufs
)
667 err
= add_recvbuf_mergeable(rq
, gfp
);
668 else if (vi
->big_packets
)
669 err
= add_recvbuf_big(rq
, gfp
);
671 err
= add_recvbuf_small(rq
, gfp
);
673 oom
= err
== -ENOMEM
;
676 } while (rq
->vq
->num_free
);
677 virtqueue_kick(rq
->vq
);
681 static void skb_recv_done(struct virtqueue
*rvq
)
683 struct virtnet_info
*vi
= rvq
->vdev
->priv
;
684 struct receive_queue
*rq
= &vi
->rq
[vq2rxq(rvq
)];
686 /* Schedule NAPI, Suppress further interrupts if successful. */
687 if (napi_schedule_prep(&rq
->napi
)) {
688 virtqueue_disable_cb(rvq
);
689 __napi_schedule(&rq
->napi
);
693 static void virtnet_napi_enable(struct receive_queue
*rq
)
695 napi_enable(&rq
->napi
);
697 /* If all buffers were filled by other side before we napi_enabled, we
698 * won't get another interrupt, so process any outstanding packets
699 * now. virtnet_poll wants re-enable the queue, so we disable here.
700 * We synchronize against interrupts via NAPI_STATE_SCHED */
701 if (napi_schedule_prep(&rq
->napi
)) {
702 virtqueue_disable_cb(rq
->vq
);
704 __napi_schedule(&rq
->napi
);
709 static void refill_work(struct work_struct
*work
)
711 struct virtnet_info
*vi
=
712 container_of(work
, struct virtnet_info
, refill
.work
);
716 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++) {
717 struct receive_queue
*rq
= &vi
->rq
[i
];
719 napi_disable(&rq
->napi
);
720 still_empty
= !try_fill_recv(rq
, GFP_KERNEL
);
721 virtnet_napi_enable(rq
);
723 /* In theory, this can happen: if we don't get any buffers in
724 * we will *never* try to fill again.
727 schedule_delayed_work(&vi
->refill
, HZ
/2);
731 static int virtnet_receive(struct receive_queue
*rq
, int budget
)
733 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
734 unsigned int len
, received
= 0;
737 while (received
< budget
&&
738 (buf
= virtqueue_get_buf(rq
->vq
, &len
)) != NULL
) {
739 receive_buf(rq
, buf
, len
);
743 if (rq
->vq
->num_free
> virtqueue_get_vring_size(rq
->vq
) / 2) {
744 if (!try_fill_recv(rq
, GFP_ATOMIC
))
745 schedule_delayed_work(&vi
->refill
, 0);
751 static int virtnet_poll(struct napi_struct
*napi
, int budget
)
753 struct receive_queue
*rq
=
754 container_of(napi
, struct receive_queue
, napi
);
755 unsigned int r
, received
= 0;
758 received
+= virtnet_receive(rq
, budget
- received
);
760 /* Out of packets? */
761 if (received
< budget
) {
762 r
= virtqueue_enable_cb_prepare(rq
->vq
);
764 if (unlikely(virtqueue_poll(rq
->vq
, r
)) &&
765 napi_schedule_prep(napi
)) {
766 virtqueue_disable_cb(rq
->vq
);
767 __napi_schedule(napi
);
775 #ifdef CONFIG_NET_RX_BUSY_POLL
776 /* must be called with local_bh_disable()d */
777 static int virtnet_busy_poll(struct napi_struct
*napi
)
779 struct receive_queue
*rq
=
780 container_of(napi
, struct receive_queue
, napi
);
781 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
782 int r
, received
= 0, budget
= 4;
784 if (!(vi
->status
& VIRTIO_NET_S_LINK_UP
))
785 return LL_FLUSH_FAILED
;
787 if (!napi_schedule_prep(napi
))
788 return LL_FLUSH_BUSY
;
790 virtqueue_disable_cb(rq
->vq
);
793 received
+= virtnet_receive(rq
, budget
);
795 r
= virtqueue_enable_cb_prepare(rq
->vq
);
796 clear_bit(NAPI_STATE_SCHED
, &napi
->state
);
797 if (unlikely(virtqueue_poll(rq
->vq
, r
)) &&
798 napi_schedule_prep(napi
)) {
799 virtqueue_disable_cb(rq
->vq
);
800 if (received
< budget
) {
804 __napi_schedule(napi
);
810 #endif /* CONFIG_NET_RX_BUSY_POLL */
812 static int virtnet_open(struct net_device
*dev
)
814 struct virtnet_info
*vi
= netdev_priv(dev
);
817 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
818 if (i
< vi
->curr_queue_pairs
)
819 /* Make sure we have some buffers: if oom use wq. */
820 if (!try_fill_recv(&vi
->rq
[i
], GFP_KERNEL
))
821 schedule_delayed_work(&vi
->refill
, 0);
822 virtnet_napi_enable(&vi
->rq
[i
]);
828 static void free_old_xmit_skbs(struct send_queue
*sq
)
832 struct virtnet_info
*vi
= sq
->vq
->vdev
->priv
;
833 struct virtnet_stats
*stats
= this_cpu_ptr(vi
->stats
);
835 while ((skb
= virtqueue_get_buf(sq
->vq
, &len
)) != NULL
) {
836 pr_debug("Sent skb %p\n", skb
);
838 u64_stats_update_begin(&stats
->tx_syncp
);
839 stats
->tx_bytes
+= skb
->len
;
841 u64_stats_update_end(&stats
->tx_syncp
);
843 dev_kfree_skb_any(skb
);
847 static int xmit_skb(struct send_queue
*sq
, struct sk_buff
*skb
)
849 struct skb_vnet_hdr
*hdr
;
850 const unsigned char *dest
= ((struct ethhdr
*)skb
->data
)->h_dest
;
851 struct virtnet_info
*vi
= sq
->vq
->vdev
->priv
;
856 pr_debug("%s: xmit %p %pM\n", vi
->dev
->name
, skb
, dest
);
857 if (vi
->mergeable_rx_bufs
)
858 hdr_len
= sizeof hdr
->mhdr
;
860 hdr_len
= sizeof hdr
->hdr
;
862 can_push
= vi
->any_header_sg
&&
863 !((unsigned long)skb
->data
& (__alignof__(*hdr
) - 1)) &&
864 !skb_header_cloned(skb
) && skb_headroom(skb
) >= hdr_len
;
865 /* Even if we can, don't push here yet as this would skew
866 * csum_start offset below. */
868 hdr
= (struct skb_vnet_hdr
*)(skb
->data
- hdr_len
);
870 hdr
= skb_vnet_hdr(skb
);
872 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
873 hdr
->hdr
.flags
= VIRTIO_NET_HDR_F_NEEDS_CSUM
;
874 hdr
->hdr
.csum_start
= skb_checksum_start_offset(skb
);
875 hdr
->hdr
.csum_offset
= skb
->csum_offset
;
878 hdr
->hdr
.csum_offset
= hdr
->hdr
.csum_start
= 0;
881 if (skb_is_gso(skb
)) {
882 hdr
->hdr
.hdr_len
= skb_headlen(skb
);
883 hdr
->hdr
.gso_size
= skb_shinfo(skb
)->gso_size
;
884 if (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCPV4
)
885 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV4
;
886 else if (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCPV6
)
887 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV6
;
888 else if (skb_shinfo(skb
)->gso_type
& SKB_GSO_UDP
)
889 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_UDP
;
892 if (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCP_ECN
)
893 hdr
->hdr
.gso_type
|= VIRTIO_NET_HDR_GSO_ECN
;
895 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
896 hdr
->hdr
.gso_size
= hdr
->hdr
.hdr_len
= 0;
899 if (vi
->mergeable_rx_bufs
)
900 hdr
->mhdr
.num_buffers
= 0;
903 __skb_push(skb
, hdr_len
);
904 num_sg
= skb_to_sgvec(skb
, sq
->sg
, 0, skb
->len
);
905 /* Pull header back to avoid skew in tx bytes calculations. */
906 __skb_pull(skb
, hdr_len
);
908 sg_set_buf(sq
->sg
, hdr
, hdr_len
);
909 num_sg
= skb_to_sgvec(skb
, sq
->sg
+ 1, 0, skb
->len
) + 1;
911 return virtqueue_add_outbuf(sq
->vq
, sq
->sg
, num_sg
, skb
, GFP_ATOMIC
);
914 static netdev_tx_t
start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
916 struct virtnet_info
*vi
= netdev_priv(dev
);
917 int qnum
= skb_get_queue_mapping(skb
);
918 struct send_queue
*sq
= &vi
->sq
[qnum
];
921 /* Free up any pending old buffers before queueing new ones. */
922 free_old_xmit_skbs(sq
);
924 /* Try to transmit */
925 err
= xmit_skb(sq
, skb
);
927 /* This should not happen! */
929 dev
->stats
.tx_fifo_errors
++;
932 "Unexpected TXQ (%d) queue failure: %d\n", qnum
, err
);
933 dev
->stats
.tx_dropped
++;
934 dev_kfree_skb_any(skb
);
937 virtqueue_kick(sq
->vq
);
939 /* Don't wait up for transmitted skbs to be freed. */
943 /* Apparently nice girls don't return TX_BUSY; stop the queue
944 * before it gets out of hand. Naturally, this wastes entries. */
945 if (sq
->vq
->num_free
< 2+MAX_SKB_FRAGS
) {
946 netif_stop_subqueue(dev
, qnum
);
947 if (unlikely(!virtqueue_enable_cb_delayed(sq
->vq
))) {
948 /* More just got used, free them then recheck. */
949 free_old_xmit_skbs(sq
);
950 if (sq
->vq
->num_free
>= 2+MAX_SKB_FRAGS
) {
951 netif_start_subqueue(dev
, qnum
);
952 virtqueue_disable_cb(sq
->vq
);
961 * Send command via the control virtqueue and check status. Commands
962 * supported by the hypervisor, as indicated by feature bits, should
963 * never fail unless improperly formatted.
965 static bool virtnet_send_command(struct virtnet_info
*vi
, u8
class, u8 cmd
,
966 struct scatterlist
*out
)
968 struct scatterlist
*sgs
[4], hdr
, stat
;
969 struct virtio_net_ctrl_hdr ctrl
;
970 virtio_net_ctrl_ack status
= ~0;
971 unsigned out_num
= 0, tmp
;
973 /* Caller should know better */
974 BUG_ON(!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
));
979 sg_init_one(&hdr
, &ctrl
, sizeof(ctrl
));
980 sgs
[out_num
++] = &hdr
;
983 sgs
[out_num
++] = out
;
985 /* Add return status. */
986 sg_init_one(&stat
, &status
, sizeof(status
));
987 sgs
[out_num
] = &stat
;
989 BUG_ON(out_num
+ 1 > ARRAY_SIZE(sgs
));
990 virtqueue_add_sgs(vi
->cvq
, sgs
, out_num
, 1, vi
, GFP_ATOMIC
);
992 if (unlikely(!virtqueue_kick(vi
->cvq
)))
993 return status
== VIRTIO_NET_OK
;
995 /* Spin for a response, the kick causes an ioport write, trapping
996 * into the hypervisor, so the request should be handled immediately.
998 while (!virtqueue_get_buf(vi
->cvq
, &tmp
) &&
999 !virtqueue_is_broken(vi
->cvq
))
1002 return status
== VIRTIO_NET_OK
;
1005 static int virtnet_set_mac_address(struct net_device
*dev
, void *p
)
1007 struct virtnet_info
*vi
= netdev_priv(dev
);
1008 struct virtio_device
*vdev
= vi
->vdev
;
1010 struct sockaddr
*addr
= p
;
1011 struct scatterlist sg
;
1013 ret
= eth_prepare_mac_addr_change(dev
, p
);
1017 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_MAC_ADDR
)) {
1018 sg_init_one(&sg
, addr
->sa_data
, dev
->addr_len
);
1019 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MAC
,
1020 VIRTIO_NET_CTRL_MAC_ADDR_SET
, &sg
)) {
1021 dev_warn(&vdev
->dev
,
1022 "Failed to set mac address by vq command.\n");
1025 } else if (virtio_has_feature(vdev
, VIRTIO_NET_F_MAC
)) {
1028 /* Naturally, this has an atomicity problem. */
1029 for (i
= 0; i
< dev
->addr_len
; i
++)
1030 virtio_cwrite8(vdev
,
1031 offsetof(struct virtio_net_config
, mac
) +
1032 i
, addr
->sa_data
[i
]);
1035 eth_commit_mac_addr_change(dev
, p
);
1040 static struct rtnl_link_stats64
*virtnet_stats(struct net_device
*dev
,
1041 struct rtnl_link_stats64
*tot
)
1043 struct virtnet_info
*vi
= netdev_priv(dev
);
1047 for_each_possible_cpu(cpu
) {
1048 struct virtnet_stats
*stats
= per_cpu_ptr(vi
->stats
, cpu
);
1049 u64 tpackets
, tbytes
, rpackets
, rbytes
;
1052 start
= u64_stats_fetch_begin_irq(&stats
->tx_syncp
);
1053 tpackets
= stats
->tx_packets
;
1054 tbytes
= stats
->tx_bytes
;
1055 } while (u64_stats_fetch_retry_irq(&stats
->tx_syncp
, start
));
1058 start
= u64_stats_fetch_begin_irq(&stats
->rx_syncp
);
1059 rpackets
= stats
->rx_packets
;
1060 rbytes
= stats
->rx_bytes
;
1061 } while (u64_stats_fetch_retry_irq(&stats
->rx_syncp
, start
));
1063 tot
->rx_packets
+= rpackets
;
1064 tot
->tx_packets
+= tpackets
;
1065 tot
->rx_bytes
+= rbytes
;
1066 tot
->tx_bytes
+= tbytes
;
1069 tot
->tx_dropped
= dev
->stats
.tx_dropped
;
1070 tot
->tx_fifo_errors
= dev
->stats
.tx_fifo_errors
;
1071 tot
->rx_dropped
= dev
->stats
.rx_dropped
;
1072 tot
->rx_length_errors
= dev
->stats
.rx_length_errors
;
1073 tot
->rx_frame_errors
= dev
->stats
.rx_frame_errors
;
1078 #ifdef CONFIG_NET_POLL_CONTROLLER
1079 static void virtnet_netpoll(struct net_device
*dev
)
1081 struct virtnet_info
*vi
= netdev_priv(dev
);
1084 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++)
1085 napi_schedule(&vi
->rq
[i
].napi
);
1089 static void virtnet_ack_link_announce(struct virtnet_info
*vi
)
1092 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_ANNOUNCE
,
1093 VIRTIO_NET_CTRL_ANNOUNCE_ACK
, NULL
))
1094 dev_warn(&vi
->dev
->dev
, "Failed to ack link announce.\n");
1098 static int virtnet_set_queues(struct virtnet_info
*vi
, u16 queue_pairs
)
1100 struct scatterlist sg
;
1101 struct virtio_net_ctrl_mq s
;
1102 struct net_device
*dev
= vi
->dev
;
1104 if (!vi
->has_cvq
|| !virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_MQ
))
1107 s
.virtqueue_pairs
= queue_pairs
;
1108 sg_init_one(&sg
, &s
, sizeof(s
));
1110 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MQ
,
1111 VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET
, &sg
)) {
1112 dev_warn(&dev
->dev
, "Fail to set num of queue pairs to %d\n",
1116 vi
->curr_queue_pairs
= queue_pairs
;
1117 /* virtnet_open() will refill when device is going to up. */
1118 if (dev
->flags
& IFF_UP
)
1119 schedule_delayed_work(&vi
->refill
, 0);
1125 static int virtnet_close(struct net_device
*dev
)
1127 struct virtnet_info
*vi
= netdev_priv(dev
);
1130 /* Make sure refill_work doesn't re-enable napi! */
1131 cancel_delayed_work_sync(&vi
->refill
);
1133 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
1134 napi_disable(&vi
->rq
[i
].napi
);
1139 static void virtnet_set_rx_mode(struct net_device
*dev
)
1141 struct virtnet_info
*vi
= netdev_priv(dev
);
1142 struct scatterlist sg
[2];
1143 u8 promisc
, allmulti
;
1144 struct virtio_net_ctrl_mac
*mac_data
;
1145 struct netdev_hw_addr
*ha
;
1151 /* We can't dynamically set ndo_set_rx_mode, so return gracefully */
1152 if (!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_RX
))
1155 promisc
= ((dev
->flags
& IFF_PROMISC
) != 0);
1156 allmulti
= ((dev
->flags
& IFF_ALLMULTI
) != 0);
1158 sg_init_one(sg
, &promisc
, sizeof(promisc
));
1160 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_RX
,
1161 VIRTIO_NET_CTRL_RX_PROMISC
, sg
))
1162 dev_warn(&dev
->dev
, "Failed to %sable promisc mode.\n",
1163 promisc
? "en" : "dis");
1165 sg_init_one(sg
, &allmulti
, sizeof(allmulti
));
1167 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_RX
,
1168 VIRTIO_NET_CTRL_RX_ALLMULTI
, sg
))
1169 dev_warn(&dev
->dev
, "Failed to %sable allmulti mode.\n",
1170 allmulti
? "en" : "dis");
1172 uc_count
= netdev_uc_count(dev
);
1173 mc_count
= netdev_mc_count(dev
);
1174 /* MAC filter - use one buffer for both lists */
1175 buf
= kzalloc(((uc_count
+ mc_count
) * ETH_ALEN
) +
1176 (2 * sizeof(mac_data
->entries
)), GFP_ATOMIC
);
1181 sg_init_table(sg
, 2);
1183 /* Store the unicast list and count in the front of the buffer */
1184 mac_data
->entries
= uc_count
;
1186 netdev_for_each_uc_addr(ha
, dev
)
1187 memcpy(&mac_data
->macs
[i
++][0], ha
->addr
, ETH_ALEN
);
1189 sg_set_buf(&sg
[0], mac_data
,
1190 sizeof(mac_data
->entries
) + (uc_count
* ETH_ALEN
));
1192 /* multicast list and count fill the end */
1193 mac_data
= (void *)&mac_data
->macs
[uc_count
][0];
1195 mac_data
->entries
= mc_count
;
1197 netdev_for_each_mc_addr(ha
, dev
)
1198 memcpy(&mac_data
->macs
[i
++][0], ha
->addr
, ETH_ALEN
);
1200 sg_set_buf(&sg
[1], mac_data
,
1201 sizeof(mac_data
->entries
) + (mc_count
* ETH_ALEN
));
1203 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MAC
,
1204 VIRTIO_NET_CTRL_MAC_TABLE_SET
, sg
))
1205 dev_warn(&dev
->dev
, "Failed to set MAC filter table.\n");
1210 static int virtnet_vlan_rx_add_vid(struct net_device
*dev
,
1211 __be16 proto
, u16 vid
)
1213 struct virtnet_info
*vi
= netdev_priv(dev
);
1214 struct scatterlist sg
;
1216 sg_init_one(&sg
, &vid
, sizeof(vid
));
1218 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_VLAN
,
1219 VIRTIO_NET_CTRL_VLAN_ADD
, &sg
))
1220 dev_warn(&dev
->dev
, "Failed to add VLAN ID %d.\n", vid
);
1224 static int virtnet_vlan_rx_kill_vid(struct net_device
*dev
,
1225 __be16 proto
, u16 vid
)
1227 struct virtnet_info
*vi
= netdev_priv(dev
);
1228 struct scatterlist sg
;
1230 sg_init_one(&sg
, &vid
, sizeof(vid
));
1232 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_VLAN
,
1233 VIRTIO_NET_CTRL_VLAN_DEL
, &sg
))
1234 dev_warn(&dev
->dev
, "Failed to kill VLAN ID %d.\n", vid
);
1238 static void virtnet_clean_affinity(struct virtnet_info
*vi
, long hcpu
)
1242 if (vi
->affinity_hint_set
) {
1243 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1244 virtqueue_set_affinity(vi
->rq
[i
].vq
, -1);
1245 virtqueue_set_affinity(vi
->sq
[i
].vq
, -1);
1248 vi
->affinity_hint_set
= false;
1252 static void virtnet_set_affinity(struct virtnet_info
*vi
)
1257 /* In multiqueue mode, when the number of cpu is equal to the number of
1258 * queue pairs, we let the queue pairs to be private to one cpu by
1259 * setting the affinity hint to eliminate the contention.
1261 if (vi
->curr_queue_pairs
== 1 ||
1262 vi
->max_queue_pairs
!= num_online_cpus()) {
1263 virtnet_clean_affinity(vi
, -1);
1268 for_each_online_cpu(cpu
) {
1269 virtqueue_set_affinity(vi
->rq
[i
].vq
, cpu
);
1270 virtqueue_set_affinity(vi
->sq
[i
].vq
, cpu
);
1271 netif_set_xps_queue(vi
->dev
, cpumask_of(cpu
), i
);
1275 vi
->affinity_hint_set
= true;
1278 static int virtnet_cpu_callback(struct notifier_block
*nfb
,
1279 unsigned long action
, void *hcpu
)
1281 struct virtnet_info
*vi
= container_of(nfb
, struct virtnet_info
, nb
);
1283 switch(action
& ~CPU_TASKS_FROZEN
) {
1285 case CPU_DOWN_FAILED
:
1287 virtnet_set_affinity(vi
);
1289 case CPU_DOWN_PREPARE
:
1290 virtnet_clean_affinity(vi
, (long)hcpu
);
1299 static void virtnet_get_ringparam(struct net_device
*dev
,
1300 struct ethtool_ringparam
*ring
)
1302 struct virtnet_info
*vi
= netdev_priv(dev
);
1304 ring
->rx_max_pending
= virtqueue_get_vring_size(vi
->rq
[0].vq
);
1305 ring
->tx_max_pending
= virtqueue_get_vring_size(vi
->sq
[0].vq
);
1306 ring
->rx_pending
= ring
->rx_max_pending
;
1307 ring
->tx_pending
= ring
->tx_max_pending
;
1311 static void virtnet_get_drvinfo(struct net_device
*dev
,
1312 struct ethtool_drvinfo
*info
)
1314 struct virtnet_info
*vi
= netdev_priv(dev
);
1315 struct virtio_device
*vdev
= vi
->vdev
;
1317 strlcpy(info
->driver
, KBUILD_MODNAME
, sizeof(info
->driver
));
1318 strlcpy(info
->version
, VIRTNET_DRIVER_VERSION
, sizeof(info
->version
));
1319 strlcpy(info
->bus_info
, virtio_bus_name(vdev
), sizeof(info
->bus_info
));
1323 /* TODO: Eliminate OOO packets during switching */
1324 static int virtnet_set_channels(struct net_device
*dev
,
1325 struct ethtool_channels
*channels
)
1327 struct virtnet_info
*vi
= netdev_priv(dev
);
1328 u16 queue_pairs
= channels
->combined_count
;
1331 /* We don't support separate rx/tx channels.
1332 * We don't allow setting 'other' channels.
1334 if (channels
->rx_count
|| channels
->tx_count
|| channels
->other_count
)
1337 if (queue_pairs
> vi
->max_queue_pairs
|| queue_pairs
== 0)
1341 err
= virtnet_set_queues(vi
, queue_pairs
);
1343 netif_set_real_num_tx_queues(dev
, queue_pairs
);
1344 netif_set_real_num_rx_queues(dev
, queue_pairs
);
1346 virtnet_set_affinity(vi
);
1353 static void virtnet_get_channels(struct net_device
*dev
,
1354 struct ethtool_channels
*channels
)
1356 struct virtnet_info
*vi
= netdev_priv(dev
);
1358 channels
->combined_count
= vi
->curr_queue_pairs
;
1359 channels
->max_combined
= vi
->max_queue_pairs
;
1360 channels
->max_other
= 0;
1361 channels
->rx_count
= 0;
1362 channels
->tx_count
= 0;
1363 channels
->other_count
= 0;
1366 static const struct ethtool_ops virtnet_ethtool_ops
= {
1367 .get_drvinfo
= virtnet_get_drvinfo
,
1368 .get_link
= ethtool_op_get_link
,
1369 .get_ringparam
= virtnet_get_ringparam
,
1370 .set_channels
= virtnet_set_channels
,
1371 .get_channels
= virtnet_get_channels
,
1375 #define MAX_MTU 65535
1377 static int virtnet_change_mtu(struct net_device
*dev
, int new_mtu
)
1379 if (new_mtu
< MIN_MTU
|| new_mtu
> MAX_MTU
)
1385 static const struct net_device_ops virtnet_netdev
= {
1386 .ndo_open
= virtnet_open
,
1387 .ndo_stop
= virtnet_close
,
1388 .ndo_start_xmit
= start_xmit
,
1389 .ndo_validate_addr
= eth_validate_addr
,
1390 .ndo_set_mac_address
= virtnet_set_mac_address
,
1391 .ndo_set_rx_mode
= virtnet_set_rx_mode
,
1392 .ndo_change_mtu
= virtnet_change_mtu
,
1393 .ndo_get_stats64
= virtnet_stats
,
1394 .ndo_vlan_rx_add_vid
= virtnet_vlan_rx_add_vid
,
1395 .ndo_vlan_rx_kill_vid
= virtnet_vlan_rx_kill_vid
,
1396 #ifdef CONFIG_NET_POLL_CONTROLLER
1397 .ndo_poll_controller
= virtnet_netpoll
,
1399 #ifdef CONFIG_NET_RX_BUSY_POLL
1400 .ndo_busy_poll
= virtnet_busy_poll
,
1404 static void virtnet_config_changed_work(struct work_struct
*work
)
1406 struct virtnet_info
*vi
=
1407 container_of(work
, struct virtnet_info
, config_work
);
1410 mutex_lock(&vi
->config_lock
);
1411 if (!vi
->config_enable
)
1414 if (virtio_cread_feature(vi
->vdev
, VIRTIO_NET_F_STATUS
,
1415 struct virtio_net_config
, status
, &v
) < 0)
1418 if (v
& VIRTIO_NET_S_ANNOUNCE
) {
1419 netdev_notify_peers(vi
->dev
);
1420 virtnet_ack_link_announce(vi
);
1423 /* Ignore unknown (future) status bits */
1424 v
&= VIRTIO_NET_S_LINK_UP
;
1426 if (vi
->status
== v
)
1431 if (vi
->status
& VIRTIO_NET_S_LINK_UP
) {
1432 netif_carrier_on(vi
->dev
);
1433 netif_tx_wake_all_queues(vi
->dev
);
1435 netif_carrier_off(vi
->dev
);
1436 netif_tx_stop_all_queues(vi
->dev
);
1439 mutex_unlock(&vi
->config_lock
);
1442 static void virtnet_config_changed(struct virtio_device
*vdev
)
1444 struct virtnet_info
*vi
= vdev
->priv
;
1446 schedule_work(&vi
->config_work
);
1449 static void virtnet_free_queues(struct virtnet_info
*vi
)
1453 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
1454 netif_napi_del(&vi
->rq
[i
].napi
);
1460 static void free_receive_bufs(struct virtnet_info
*vi
)
1464 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1465 while (vi
->rq
[i
].pages
)
1466 __free_pages(get_a_page(&vi
->rq
[i
], GFP_KERNEL
), 0);
1470 static void free_receive_page_frags(struct virtnet_info
*vi
)
1473 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
1474 if (vi
->rq
[i
].alloc_frag
.page
)
1475 put_page(vi
->rq
[i
].alloc_frag
.page
);
1478 static void free_unused_bufs(struct virtnet_info
*vi
)
1483 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1484 struct virtqueue
*vq
= vi
->sq
[i
].vq
;
1485 while ((buf
= virtqueue_detach_unused_buf(vq
)) != NULL
)
1489 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1490 struct virtqueue
*vq
= vi
->rq
[i
].vq
;
1492 while ((buf
= virtqueue_detach_unused_buf(vq
)) != NULL
) {
1493 if (vi
->mergeable_rx_bufs
) {
1494 unsigned long ctx
= (unsigned long)buf
;
1495 void *base
= mergeable_ctx_to_buf_address(ctx
);
1496 put_page(virt_to_head_page(base
));
1497 } else if (vi
->big_packets
) {
1498 give_pages(&vi
->rq
[i
], buf
);
1506 static void virtnet_del_vqs(struct virtnet_info
*vi
)
1508 struct virtio_device
*vdev
= vi
->vdev
;
1510 virtnet_clean_affinity(vi
, -1);
1512 vdev
->config
->del_vqs(vdev
);
1514 virtnet_free_queues(vi
);
1517 static int virtnet_find_vqs(struct virtnet_info
*vi
)
1519 vq_callback_t
**callbacks
;
1520 struct virtqueue
**vqs
;
1525 /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
1526 * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
1527 * possible control vq.
1529 total_vqs
= vi
->max_queue_pairs
* 2 +
1530 virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
);
1532 /* Allocate space for find_vqs parameters */
1533 vqs
= kzalloc(total_vqs
* sizeof(*vqs
), GFP_KERNEL
);
1536 callbacks
= kmalloc(total_vqs
* sizeof(*callbacks
), GFP_KERNEL
);
1539 names
= kmalloc(total_vqs
* sizeof(*names
), GFP_KERNEL
);
1543 /* Parameters for control virtqueue, if any */
1545 callbacks
[total_vqs
- 1] = NULL
;
1546 names
[total_vqs
- 1] = "control";
1549 /* Allocate/initialize parameters for send/receive virtqueues */
1550 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1551 callbacks
[rxq2vq(i
)] = skb_recv_done
;
1552 callbacks
[txq2vq(i
)] = skb_xmit_done
;
1553 sprintf(vi
->rq
[i
].name
, "input.%d", i
);
1554 sprintf(vi
->sq
[i
].name
, "output.%d", i
);
1555 names
[rxq2vq(i
)] = vi
->rq
[i
].name
;
1556 names
[txq2vq(i
)] = vi
->sq
[i
].name
;
1559 ret
= vi
->vdev
->config
->find_vqs(vi
->vdev
, total_vqs
, vqs
, callbacks
,
1565 vi
->cvq
= vqs
[total_vqs
- 1];
1566 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VLAN
))
1567 vi
->dev
->features
|= NETIF_F_HW_VLAN_CTAG_FILTER
;
1570 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1571 vi
->rq
[i
].vq
= vqs
[rxq2vq(i
)];
1572 vi
->sq
[i
].vq
= vqs
[txq2vq(i
)];
1591 static int virtnet_alloc_queues(struct virtnet_info
*vi
)
1595 vi
->sq
= kzalloc(sizeof(*vi
->sq
) * vi
->max_queue_pairs
, GFP_KERNEL
);
1598 vi
->rq
= kzalloc(sizeof(*vi
->rq
) * vi
->max_queue_pairs
, GFP_KERNEL
);
1602 INIT_DELAYED_WORK(&vi
->refill
, refill_work
);
1603 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1604 vi
->rq
[i
].pages
= NULL
;
1605 netif_napi_add(vi
->dev
, &vi
->rq
[i
].napi
, virtnet_poll
,
1607 napi_hash_add(&vi
->rq
[i
].napi
);
1609 sg_init_table(vi
->rq
[i
].sg
, ARRAY_SIZE(vi
->rq
[i
].sg
));
1610 ewma_init(&vi
->rq
[i
].mrg_avg_pkt_len
, 1, RECEIVE_AVG_WEIGHT
);
1611 sg_init_table(vi
->sq
[i
].sg
, ARRAY_SIZE(vi
->sq
[i
].sg
));
1622 static int init_vqs(struct virtnet_info
*vi
)
1626 /* Allocate send & receive queues */
1627 ret
= virtnet_alloc_queues(vi
);
1631 ret
= virtnet_find_vqs(vi
);
1636 virtnet_set_affinity(vi
);
1642 virtnet_free_queues(vi
);
1648 static ssize_t
mergeable_rx_buffer_size_show(struct netdev_rx_queue
*queue
,
1649 struct rx_queue_attribute
*attribute
, char *buf
)
1651 struct virtnet_info
*vi
= netdev_priv(queue
->dev
);
1652 unsigned int queue_index
= get_netdev_rx_queue_index(queue
);
1655 BUG_ON(queue_index
>= vi
->max_queue_pairs
);
1656 avg
= &vi
->rq
[queue_index
].mrg_avg_pkt_len
;
1657 return sprintf(buf
, "%u\n", get_mergeable_buf_len(avg
));
1660 static struct rx_queue_attribute mergeable_rx_buffer_size_attribute
=
1661 __ATTR_RO(mergeable_rx_buffer_size
);
1663 static struct attribute
*virtio_net_mrg_rx_attrs
[] = {
1664 &mergeable_rx_buffer_size_attribute
.attr
,
1668 static const struct attribute_group virtio_net_mrg_rx_group
= {
1669 .name
= "virtio_net",
1670 .attrs
= virtio_net_mrg_rx_attrs
1674 static int virtnet_probe(struct virtio_device
*vdev
)
1677 struct net_device
*dev
;
1678 struct virtnet_info
*vi
;
1679 u16 max_queue_pairs
;
1681 /* Find if host supports multiqueue virtio_net device */
1682 err
= virtio_cread_feature(vdev
, VIRTIO_NET_F_MQ
,
1683 struct virtio_net_config
,
1684 max_virtqueue_pairs
, &max_queue_pairs
);
1686 /* We need at least 2 queue's */
1687 if (err
|| max_queue_pairs
< VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN
||
1688 max_queue_pairs
> VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX
||
1689 !virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
))
1690 max_queue_pairs
= 1;
1692 /* Allocate ourselves a network device with room for our info */
1693 dev
= alloc_etherdev_mq(sizeof(struct virtnet_info
), max_queue_pairs
);
1697 /* Set up network device as normal. */
1698 dev
->priv_flags
|= IFF_UNICAST_FLT
| IFF_LIVE_ADDR_CHANGE
;
1699 dev
->netdev_ops
= &virtnet_netdev
;
1700 dev
->features
= NETIF_F_HIGHDMA
;
1702 dev
->ethtool_ops
= &virtnet_ethtool_ops
;
1703 SET_NETDEV_DEV(dev
, &vdev
->dev
);
1705 /* Do we support "hardware" checksums? */
1706 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CSUM
)) {
1707 /* This opens up the world of extra features. */
1708 dev
->hw_features
|= NETIF_F_HW_CSUM
|NETIF_F_SG
|NETIF_F_FRAGLIST
;
1710 dev
->features
|= NETIF_F_HW_CSUM
|NETIF_F_SG
|NETIF_F_FRAGLIST
;
1712 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GSO
)) {
1713 dev
->hw_features
|= NETIF_F_TSO
| NETIF_F_UFO
1714 | NETIF_F_TSO_ECN
| NETIF_F_TSO6
;
1716 /* Individual feature bits: what can host handle? */
1717 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_TSO4
))
1718 dev
->hw_features
|= NETIF_F_TSO
;
1719 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_TSO6
))
1720 dev
->hw_features
|= NETIF_F_TSO6
;
1721 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_ECN
))
1722 dev
->hw_features
|= NETIF_F_TSO_ECN
;
1723 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_UFO
))
1724 dev
->hw_features
|= NETIF_F_UFO
;
1727 dev
->features
|= dev
->hw_features
& (NETIF_F_ALL_TSO
|NETIF_F_UFO
);
1728 /* (!csum && gso) case will be fixed by register_netdev() */
1730 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_CSUM
))
1731 dev
->features
|= NETIF_F_RXCSUM
;
1733 dev
->vlan_features
= dev
->features
;
1735 /* Configuration may specify what MAC to use. Otherwise random. */
1736 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MAC
))
1737 virtio_cread_bytes(vdev
,
1738 offsetof(struct virtio_net_config
, mac
),
1739 dev
->dev_addr
, dev
->addr_len
);
1741 eth_hw_addr_random(dev
);
1743 /* Set up our device-specific information */
1744 vi
= netdev_priv(dev
);
1748 vi
->stats
= alloc_percpu(struct virtnet_stats
);
1750 if (vi
->stats
== NULL
)
1753 for_each_possible_cpu(i
) {
1754 struct virtnet_stats
*virtnet_stats
;
1755 virtnet_stats
= per_cpu_ptr(vi
->stats
, i
);
1756 u64_stats_init(&virtnet_stats
->tx_syncp
);
1757 u64_stats_init(&virtnet_stats
->rx_syncp
);
1760 mutex_init(&vi
->config_lock
);
1761 vi
->config_enable
= true;
1762 INIT_WORK(&vi
->config_work
, virtnet_config_changed_work
);
1764 /* If we can receive ANY GSO packets, we must allocate large ones. */
1765 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_TSO4
) ||
1766 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_TSO6
) ||
1767 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_ECN
) ||
1768 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_UFO
))
1769 vi
->big_packets
= true;
1771 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MRG_RXBUF
))
1772 vi
->mergeable_rx_bufs
= true;
1774 if (virtio_has_feature(vdev
, VIRTIO_F_ANY_LAYOUT
))
1775 vi
->any_header_sg
= true;
1777 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
))
1780 if (vi
->any_header_sg
) {
1781 if (vi
->mergeable_rx_bufs
)
1782 dev
->needed_headroom
= sizeof(struct virtio_net_hdr_mrg_rxbuf
);
1784 dev
->needed_headroom
= sizeof(struct virtio_net_hdr
);
1787 /* Use single tx/rx queue pair as default */
1788 vi
->curr_queue_pairs
= 1;
1789 vi
->max_queue_pairs
= max_queue_pairs
;
1791 /* Allocate/initialize the rx/tx queues, and invoke find_vqs */
1797 if (vi
->mergeable_rx_bufs
)
1798 dev
->sysfs_rx_queue_group
= &virtio_net_mrg_rx_group
;
1800 netif_set_real_num_tx_queues(dev
, vi
->curr_queue_pairs
);
1801 netif_set_real_num_rx_queues(dev
, vi
->curr_queue_pairs
);
1803 err
= register_netdev(dev
);
1805 pr_debug("virtio_net: registering device failed\n");
1809 /* Last of all, set up some receive buffers. */
1810 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++) {
1811 try_fill_recv(&vi
->rq
[i
], GFP_KERNEL
);
1813 /* If we didn't even get one input buffer, we're useless. */
1814 if (vi
->rq
[i
].vq
->num_free
==
1815 virtqueue_get_vring_size(vi
->rq
[i
].vq
)) {
1816 free_unused_bufs(vi
);
1818 goto free_recv_bufs
;
1822 vi
->nb
.notifier_call
= &virtnet_cpu_callback
;
1823 err
= register_hotcpu_notifier(&vi
->nb
);
1825 pr_debug("virtio_net: registering cpu notifier failed\n");
1826 goto free_recv_bufs
;
1829 /* Assume link up if device can't report link status,
1830 otherwise get link status from config. */
1831 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_STATUS
)) {
1832 netif_carrier_off(dev
);
1833 schedule_work(&vi
->config_work
);
1835 vi
->status
= VIRTIO_NET_S_LINK_UP
;
1836 netif_carrier_on(dev
);
1839 pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
1840 dev
->name
, max_queue_pairs
);
1845 free_receive_bufs(vi
);
1846 unregister_netdev(dev
);
1848 cancel_delayed_work_sync(&vi
->refill
);
1849 free_receive_page_frags(vi
);
1850 virtnet_del_vqs(vi
);
1852 free_percpu(vi
->stats
);
1858 static void remove_vq_common(struct virtnet_info
*vi
)
1860 vi
->vdev
->config
->reset(vi
->vdev
);
1862 /* Free unused buffers in both send and recv, if any. */
1863 free_unused_bufs(vi
);
1865 free_receive_bufs(vi
);
1867 free_receive_page_frags(vi
);
1869 virtnet_del_vqs(vi
);
1872 static void virtnet_remove(struct virtio_device
*vdev
)
1874 struct virtnet_info
*vi
= vdev
->priv
;
1876 unregister_hotcpu_notifier(&vi
->nb
);
1878 /* Prevent config work handler from accessing the device. */
1879 mutex_lock(&vi
->config_lock
);
1880 vi
->config_enable
= false;
1881 mutex_unlock(&vi
->config_lock
);
1883 unregister_netdev(vi
->dev
);
1885 remove_vq_common(vi
);
1887 flush_work(&vi
->config_work
);
1889 free_percpu(vi
->stats
);
1890 free_netdev(vi
->dev
);
1893 #ifdef CONFIG_PM_SLEEP
1894 static int virtnet_freeze(struct virtio_device
*vdev
)
1896 struct virtnet_info
*vi
= vdev
->priv
;
1899 unregister_hotcpu_notifier(&vi
->nb
);
1901 /* Prevent config work handler from accessing the device */
1902 mutex_lock(&vi
->config_lock
);
1903 vi
->config_enable
= false;
1904 mutex_unlock(&vi
->config_lock
);
1906 netif_device_detach(vi
->dev
);
1907 cancel_delayed_work_sync(&vi
->refill
);
1909 if (netif_running(vi
->dev
)) {
1910 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1911 napi_disable(&vi
->rq
[i
].napi
);
1912 napi_hash_del(&vi
->rq
[i
].napi
);
1913 netif_napi_del(&vi
->rq
[i
].napi
);
1917 remove_vq_common(vi
);
1919 flush_work(&vi
->config_work
);
1924 static int virtnet_restore(struct virtio_device
*vdev
)
1926 struct virtnet_info
*vi
= vdev
->priv
;
1933 if (netif_running(vi
->dev
)) {
1934 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++)
1935 if (!try_fill_recv(&vi
->rq
[i
], GFP_KERNEL
))
1936 schedule_delayed_work(&vi
->refill
, 0);
1938 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
1939 virtnet_napi_enable(&vi
->rq
[i
]);
1942 netif_device_attach(vi
->dev
);
1944 mutex_lock(&vi
->config_lock
);
1945 vi
->config_enable
= true;
1946 mutex_unlock(&vi
->config_lock
);
1949 virtnet_set_queues(vi
, vi
->curr_queue_pairs
);
1952 err
= register_hotcpu_notifier(&vi
->nb
);
1960 static struct virtio_device_id id_table
[] = {
1961 { VIRTIO_ID_NET
, VIRTIO_DEV_ANY_ID
},
1965 static unsigned int features
[] = {
1966 VIRTIO_NET_F_CSUM
, VIRTIO_NET_F_GUEST_CSUM
,
1967 VIRTIO_NET_F_GSO
, VIRTIO_NET_F_MAC
,
1968 VIRTIO_NET_F_HOST_TSO4
, VIRTIO_NET_F_HOST_UFO
, VIRTIO_NET_F_HOST_TSO6
,
1969 VIRTIO_NET_F_HOST_ECN
, VIRTIO_NET_F_GUEST_TSO4
, VIRTIO_NET_F_GUEST_TSO6
,
1970 VIRTIO_NET_F_GUEST_ECN
, VIRTIO_NET_F_GUEST_UFO
,
1971 VIRTIO_NET_F_MRG_RXBUF
, VIRTIO_NET_F_STATUS
, VIRTIO_NET_F_CTRL_VQ
,
1972 VIRTIO_NET_F_CTRL_RX
, VIRTIO_NET_F_CTRL_VLAN
,
1973 VIRTIO_NET_F_GUEST_ANNOUNCE
, VIRTIO_NET_F_MQ
,
1974 VIRTIO_NET_F_CTRL_MAC_ADDR
,
1975 VIRTIO_F_ANY_LAYOUT
,
1978 static struct virtio_driver virtio_net_driver
= {
1979 .feature_table
= features
,
1980 .feature_table_size
= ARRAY_SIZE(features
),
1981 .driver
.name
= KBUILD_MODNAME
,
1982 .driver
.owner
= THIS_MODULE
,
1983 .id_table
= id_table
,
1984 .probe
= virtnet_probe
,
1985 .remove
= virtnet_remove
,
1986 .config_changed
= virtnet_config_changed
,
1987 #ifdef CONFIG_PM_SLEEP
1988 .freeze
= virtnet_freeze
,
1989 .restore
= virtnet_restore
,
1993 module_virtio_driver(virtio_net_driver
);
1995 MODULE_DEVICE_TABLE(virtio
, id_table
);
1996 MODULE_DESCRIPTION("Virtio network driver");
1997 MODULE_LICENSE("GPL");