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 /* 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
;
145 struct padded_vnet_hdr
{
146 struct virtio_net_hdr_mrg_rxbuf hdr
;
148 * hdr is in a separate sg buffer, and data sg buffer shares same page
149 * with this header sg. This padding makes next sg 16 byte aligned
155 /* Converting between virtqueue no. and kernel tx/rx queue no.
156 * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq
158 static int vq2txq(struct virtqueue
*vq
)
160 return (vq
->index
- 1) / 2;
163 static int txq2vq(int txq
)
168 static int vq2rxq(struct virtqueue
*vq
)
170 return vq
->index
/ 2;
173 static int rxq2vq(int rxq
)
178 static inline struct virtio_net_hdr_mrg_rxbuf
*skb_vnet_hdr(struct sk_buff
*skb
)
180 return (struct virtio_net_hdr_mrg_rxbuf
*)skb
->cb
;
184 * private is used to chain pages for big packets, put the whole
185 * most recent used list in the beginning for reuse
187 static void give_pages(struct receive_queue
*rq
, struct page
*page
)
191 /* Find end of list, sew whole thing into vi->rq.pages. */
192 for (end
= page
; end
->private; end
= (struct page
*)end
->private);
193 end
->private = (unsigned long)rq
->pages
;
197 static struct page
*get_a_page(struct receive_queue
*rq
, gfp_t gfp_mask
)
199 struct page
*p
= rq
->pages
;
202 rq
->pages
= (struct page
*)p
->private;
203 /* clear private here, it is used to chain pages */
206 p
= alloc_page(gfp_mask
);
210 static void skb_xmit_done(struct virtqueue
*vq
)
212 struct virtnet_info
*vi
= vq
->vdev
->priv
;
214 /* Suppress further interrupts. */
215 virtqueue_disable_cb(vq
);
217 /* We were probably waiting for more output buffers. */
218 netif_wake_subqueue(vi
->dev
, vq2txq(vq
));
221 static unsigned int mergeable_ctx_to_buf_truesize(unsigned long mrg_ctx
)
223 unsigned int truesize
= mrg_ctx
& (MERGEABLE_BUFFER_ALIGN
- 1);
224 return (truesize
+ 1) * MERGEABLE_BUFFER_ALIGN
;
227 static void *mergeable_ctx_to_buf_address(unsigned long mrg_ctx
)
229 return (void *)(mrg_ctx
& -MERGEABLE_BUFFER_ALIGN
);
233 static unsigned long mergeable_buf_to_ctx(void *buf
, unsigned int truesize
)
235 unsigned int size
= truesize
/ MERGEABLE_BUFFER_ALIGN
;
236 return (unsigned long)buf
| (size
- 1);
239 /* Called from bottom half context */
240 static struct sk_buff
*page_to_skb(struct virtnet_info
*vi
,
241 struct receive_queue
*rq
,
242 struct page
*page
, unsigned int offset
,
243 unsigned int len
, unsigned int truesize
)
246 struct virtio_net_hdr_mrg_rxbuf
*hdr
;
247 unsigned int copy
, hdr_len
, hdr_padded_len
;
250 p
= page_address(page
) + offset
;
252 /* copy small packet so we can reuse these pages for small data */
253 skb
= netdev_alloc_skb_ip_align(vi
->dev
, GOOD_COPY_LEN
);
257 hdr
= skb_vnet_hdr(skb
);
259 hdr_len
= vi
->hdr_len
;
260 if (vi
->mergeable_rx_bufs
)
261 hdr_padded_len
= sizeof *hdr
;
263 hdr_padded_len
= sizeof(struct padded_vnet_hdr
);
265 memcpy(hdr
, p
, hdr_len
);
268 offset
+= hdr_padded_len
;
272 if (copy
> skb_tailroom(skb
))
273 copy
= skb_tailroom(skb
);
274 memcpy(skb_put(skb
, copy
), p
, copy
);
279 if (vi
->mergeable_rx_bufs
) {
281 skb_add_rx_frag(skb
, 0, page
, offset
, len
, truesize
);
288 * Verify that we can indeed put this data into a skb.
289 * This is here to handle cases when the device erroneously
290 * tries to receive more than is possible. This is usually
291 * the case of a broken device.
293 if (unlikely(len
> MAX_SKB_FRAGS
* PAGE_SIZE
)) {
294 net_dbg_ratelimited("%s: too much data\n", skb
->dev
->name
);
298 BUG_ON(offset
>= PAGE_SIZE
);
300 unsigned int frag_size
= min((unsigned)PAGE_SIZE
- offset
, len
);
301 skb_add_rx_frag(skb
, skb_shinfo(skb
)->nr_frags
, page
, offset
,
302 frag_size
, truesize
);
304 page
= (struct page
*)page
->private;
309 give_pages(rq
, page
);
314 static struct sk_buff
*receive_small(struct virtnet_info
*vi
, void *buf
, unsigned int len
)
316 struct sk_buff
* skb
= buf
;
324 static struct sk_buff
*receive_big(struct net_device
*dev
,
325 struct virtnet_info
*vi
,
326 struct receive_queue
*rq
,
330 struct page
*page
= buf
;
331 struct sk_buff
*skb
= page_to_skb(vi
, rq
, page
, 0, len
, PAGE_SIZE
);
339 dev
->stats
.rx_dropped
++;
340 give_pages(rq
, page
);
344 static struct sk_buff
*receive_mergeable(struct net_device
*dev
,
345 struct virtnet_info
*vi
,
346 struct receive_queue
*rq
,
350 void *buf
= mergeable_ctx_to_buf_address(ctx
);
351 struct virtio_net_hdr_mrg_rxbuf
*hdr
= buf
;
352 u16 num_buf
= virtio16_to_cpu(vi
->vdev
, hdr
->num_buffers
);
353 struct page
*page
= virt_to_head_page(buf
);
354 int offset
= buf
- page_address(page
);
355 unsigned int truesize
= max(len
, mergeable_ctx_to_buf_truesize(ctx
));
357 struct sk_buff
*head_skb
= page_to_skb(vi
, rq
, page
, offset
, len
,
359 struct sk_buff
*curr_skb
= head_skb
;
361 if (unlikely(!curr_skb
))
366 ctx
= (unsigned long)virtqueue_get_buf(rq
->vq
, &len
);
367 if (unlikely(!ctx
)) {
368 pr_debug("%s: rx error: %d buffers out of %d missing\n",
370 virtio16_to_cpu(vi
->vdev
,
372 dev
->stats
.rx_length_errors
++;
376 buf
= mergeable_ctx_to_buf_address(ctx
);
377 page
= virt_to_head_page(buf
);
379 num_skb_frags
= skb_shinfo(curr_skb
)->nr_frags
;
380 if (unlikely(num_skb_frags
== MAX_SKB_FRAGS
)) {
381 struct sk_buff
*nskb
= alloc_skb(0, GFP_ATOMIC
);
385 if (curr_skb
== head_skb
)
386 skb_shinfo(curr_skb
)->frag_list
= nskb
;
388 curr_skb
->next
= nskb
;
390 head_skb
->truesize
+= nskb
->truesize
;
393 truesize
= max(len
, mergeable_ctx_to_buf_truesize(ctx
));
394 if (curr_skb
!= head_skb
) {
395 head_skb
->data_len
+= len
;
396 head_skb
->len
+= len
;
397 head_skb
->truesize
+= truesize
;
399 offset
= buf
- page_address(page
);
400 if (skb_can_coalesce(curr_skb
, num_skb_frags
, page
, offset
)) {
402 skb_coalesce_rx_frag(curr_skb
, num_skb_frags
- 1,
405 skb_add_rx_frag(curr_skb
, num_skb_frags
, page
,
406 offset
, len
, truesize
);
410 ewma_add(&rq
->mrg_avg_pkt_len
, head_skb
->len
);
416 ctx
= (unsigned long)virtqueue_get_buf(rq
->vq
, &len
);
417 if (unlikely(!ctx
)) {
418 pr_debug("%s: rx error: %d buffers missing\n",
420 dev
->stats
.rx_length_errors
++;
423 page
= virt_to_head_page(mergeable_ctx_to_buf_address(ctx
));
427 dev
->stats
.rx_dropped
++;
428 dev_kfree_skb(head_skb
);
432 static void receive_buf(struct virtnet_info
*vi
, struct receive_queue
*rq
,
433 void *buf
, unsigned int len
)
435 struct net_device
*dev
= vi
->dev
;
436 struct virtnet_stats
*stats
= this_cpu_ptr(vi
->stats
);
438 struct virtio_net_hdr_mrg_rxbuf
*hdr
;
440 if (unlikely(len
< vi
->hdr_len
+ ETH_HLEN
)) {
441 pr_debug("%s: short packet %i\n", dev
->name
, len
);
442 dev
->stats
.rx_length_errors
++;
443 if (vi
->mergeable_rx_bufs
) {
444 unsigned long ctx
= (unsigned long)buf
;
445 void *base
= mergeable_ctx_to_buf_address(ctx
);
446 put_page(virt_to_head_page(base
));
447 } else if (vi
->big_packets
) {
455 if (vi
->mergeable_rx_bufs
)
456 skb
= receive_mergeable(dev
, vi
, rq
, (unsigned long)buf
, len
);
457 else if (vi
->big_packets
)
458 skb
= receive_big(dev
, vi
, rq
, buf
, len
);
460 skb
= receive_small(vi
, buf
, len
);
465 hdr
= skb_vnet_hdr(skb
);
467 u64_stats_update_begin(&stats
->rx_syncp
);
468 stats
->rx_bytes
+= skb
->len
;
470 u64_stats_update_end(&stats
->rx_syncp
);
472 if (hdr
->hdr
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) {
473 pr_debug("Needs csum!\n");
474 if (!skb_partial_csum_set(skb
,
475 virtio16_to_cpu(vi
->vdev
, hdr
->hdr
.csum_start
),
476 virtio16_to_cpu(vi
->vdev
, hdr
->hdr
.csum_offset
)))
478 } else if (hdr
->hdr
.flags
& VIRTIO_NET_HDR_F_DATA_VALID
) {
479 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
482 skb
->protocol
= eth_type_trans(skb
, dev
);
483 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
484 ntohs(skb
->protocol
), skb
->len
, skb
->pkt_type
);
486 if (hdr
->hdr
.gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
488 switch (hdr
->hdr
.gso_type
& ~VIRTIO_NET_HDR_GSO_ECN
) {
489 case VIRTIO_NET_HDR_GSO_TCPV4
:
490 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV4
;
492 case VIRTIO_NET_HDR_GSO_UDP
:
493 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
495 case VIRTIO_NET_HDR_GSO_TCPV6
:
496 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV6
;
499 net_warn_ratelimited("%s: bad gso type %u.\n",
500 dev
->name
, hdr
->hdr
.gso_type
);
504 if (hdr
->hdr
.gso_type
& VIRTIO_NET_HDR_GSO_ECN
)
505 skb_shinfo(skb
)->gso_type
|= SKB_GSO_TCP_ECN
;
507 skb_shinfo(skb
)->gso_size
= virtio16_to_cpu(vi
->vdev
,
509 if (skb_shinfo(skb
)->gso_size
== 0) {
510 net_warn_ratelimited("%s: zero gso size.\n", dev
->name
);
514 /* Header must be checked, and gso_segs computed. */
515 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
516 skb_shinfo(skb
)->gso_segs
= 0;
519 skb_mark_napi_id(skb
, &rq
->napi
);
521 netif_receive_skb(skb
);
525 dev
->stats
.rx_frame_errors
++;
529 static int add_recvbuf_small(struct virtnet_info
*vi
, struct receive_queue
*rq
,
533 struct virtio_net_hdr_mrg_rxbuf
*hdr
;
536 skb
= __netdev_alloc_skb_ip_align(vi
->dev
, GOOD_PACKET_LEN
, gfp
);
540 skb_put(skb
, GOOD_PACKET_LEN
);
542 hdr
= skb_vnet_hdr(skb
);
543 sg_init_table(rq
->sg
, MAX_SKB_FRAGS
+ 2);
544 sg_set_buf(rq
->sg
, hdr
, vi
->hdr_len
);
545 skb_to_sgvec(skb
, rq
->sg
+ 1, 0, skb
->len
);
547 err
= virtqueue_add_inbuf(rq
->vq
, rq
->sg
, 2, skb
, gfp
);
554 static int add_recvbuf_big(struct virtnet_info
*vi
, struct receive_queue
*rq
,
557 struct page
*first
, *list
= NULL
;
561 sg_init_table(rq
->sg
, MAX_SKB_FRAGS
+ 2);
563 /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */
564 for (i
= MAX_SKB_FRAGS
+ 1; i
> 1; --i
) {
565 first
= get_a_page(rq
, gfp
);
568 give_pages(rq
, list
);
571 sg_set_buf(&rq
->sg
[i
], page_address(first
), PAGE_SIZE
);
573 /* chain new page in list head to match sg */
574 first
->private = (unsigned long)list
;
578 first
= get_a_page(rq
, gfp
);
580 give_pages(rq
, list
);
583 p
= page_address(first
);
585 /* rq->sg[0], rq->sg[1] share the same page */
586 /* a separated rq->sg[0] for header - required in case !any_header_sg */
587 sg_set_buf(&rq
->sg
[0], p
, vi
->hdr_len
);
589 /* rq->sg[1] for data packet, from offset */
590 offset
= sizeof(struct padded_vnet_hdr
);
591 sg_set_buf(&rq
->sg
[1], p
+ offset
, PAGE_SIZE
- offset
);
593 /* chain first in list head */
594 first
->private = (unsigned long)list
;
595 err
= virtqueue_add_inbuf(rq
->vq
, rq
->sg
, MAX_SKB_FRAGS
+ 2,
598 give_pages(rq
, first
);
603 static unsigned int get_mergeable_buf_len(struct ewma
*avg_pkt_len
)
605 const size_t hdr_len
= sizeof(struct virtio_net_hdr_mrg_rxbuf
);
608 len
= hdr_len
+ clamp_t(unsigned int, ewma_read(avg_pkt_len
),
609 GOOD_PACKET_LEN
, PAGE_SIZE
- hdr_len
);
610 return ALIGN(len
, MERGEABLE_BUFFER_ALIGN
);
613 static int add_recvbuf_mergeable(struct receive_queue
*rq
, gfp_t gfp
)
615 struct page_frag
*alloc_frag
= &rq
->alloc_frag
;
619 unsigned int len
, hole
;
621 len
= get_mergeable_buf_len(&rq
->mrg_avg_pkt_len
);
622 if (unlikely(!skb_page_frag_refill(len
, alloc_frag
, gfp
)))
625 buf
= (char *)page_address(alloc_frag
->page
) + alloc_frag
->offset
;
626 ctx
= mergeable_buf_to_ctx(buf
, len
);
627 get_page(alloc_frag
->page
);
628 alloc_frag
->offset
+= len
;
629 hole
= alloc_frag
->size
- alloc_frag
->offset
;
631 /* To avoid internal fragmentation, if there is very likely not
632 * enough space for another buffer, add the remaining space to
633 * the current buffer. This extra space is not included in
634 * the truesize stored in ctx.
637 alloc_frag
->offset
+= hole
;
640 sg_init_one(rq
->sg
, buf
, len
);
641 err
= virtqueue_add_inbuf(rq
->vq
, rq
->sg
, 1, (void *)ctx
, gfp
);
643 put_page(virt_to_head_page(buf
));
649 * Returns false if we couldn't fill entirely (OOM).
651 * Normally run in the receive path, but can also be run from ndo_open
652 * before we're receiving packets, or from refill_work which is
653 * careful to disable receiving (using napi_disable).
655 static bool try_fill_recv(struct virtnet_info
*vi
, struct receive_queue
*rq
,
663 if (vi
->mergeable_rx_bufs
)
664 err
= add_recvbuf_mergeable(rq
, gfp
);
665 else if (vi
->big_packets
)
666 err
= add_recvbuf_big(vi
, rq
, gfp
);
668 err
= add_recvbuf_small(vi
, rq
, gfp
);
670 oom
= err
== -ENOMEM
;
673 } while (rq
->vq
->num_free
);
674 virtqueue_kick(rq
->vq
);
678 static void skb_recv_done(struct virtqueue
*rvq
)
680 struct virtnet_info
*vi
= rvq
->vdev
->priv
;
681 struct receive_queue
*rq
= &vi
->rq
[vq2rxq(rvq
)];
683 /* Schedule NAPI, Suppress further interrupts if successful. */
684 if (napi_schedule_prep(&rq
->napi
)) {
685 virtqueue_disable_cb(rvq
);
686 __napi_schedule(&rq
->napi
);
690 static void virtnet_napi_enable(struct receive_queue
*rq
)
692 napi_enable(&rq
->napi
);
694 /* If all buffers were filled by other side before we napi_enabled, we
695 * won't get another interrupt, so process any outstanding packets
696 * now. virtnet_poll wants re-enable the queue, so we disable here.
697 * We synchronize against interrupts via NAPI_STATE_SCHED */
698 if (napi_schedule_prep(&rq
->napi
)) {
699 virtqueue_disable_cb(rq
->vq
);
701 __napi_schedule(&rq
->napi
);
706 static void refill_work(struct work_struct
*work
)
708 struct virtnet_info
*vi
=
709 container_of(work
, struct virtnet_info
, refill
.work
);
713 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++) {
714 struct receive_queue
*rq
= &vi
->rq
[i
];
716 napi_disable(&rq
->napi
);
717 still_empty
= !try_fill_recv(vi
, rq
, GFP_KERNEL
);
718 virtnet_napi_enable(rq
);
720 /* In theory, this can happen: if we don't get any buffers in
721 * we will *never* try to fill again.
724 schedule_delayed_work(&vi
->refill
, HZ
/2);
728 static int virtnet_receive(struct receive_queue
*rq
, int budget
)
730 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
731 unsigned int len
, received
= 0;
734 while (received
< budget
&&
735 (buf
= virtqueue_get_buf(rq
->vq
, &len
)) != NULL
) {
736 receive_buf(vi
, rq
, buf
, len
);
740 if (rq
->vq
->num_free
> virtqueue_get_vring_size(rq
->vq
) / 2) {
741 if (!try_fill_recv(vi
, rq
, GFP_ATOMIC
))
742 schedule_delayed_work(&vi
->refill
, 0);
748 static int virtnet_poll(struct napi_struct
*napi
, int budget
)
750 struct receive_queue
*rq
=
751 container_of(napi
, struct receive_queue
, napi
);
752 unsigned int r
, received
= 0;
754 received
+= virtnet_receive(rq
, budget
- received
);
756 /* Out of packets? */
757 if (received
< budget
) {
758 r
= virtqueue_enable_cb_prepare(rq
->vq
);
760 if (unlikely(virtqueue_poll(rq
->vq
, r
)) &&
761 napi_schedule_prep(napi
)) {
762 virtqueue_disable_cb(rq
->vq
);
763 __napi_schedule(napi
);
770 #ifdef CONFIG_NET_RX_BUSY_POLL
771 /* must be called with local_bh_disable()d */
772 static int virtnet_busy_poll(struct napi_struct
*napi
)
774 struct receive_queue
*rq
=
775 container_of(napi
, struct receive_queue
, napi
);
776 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
777 int r
, received
= 0, budget
= 4;
779 if (!(vi
->status
& VIRTIO_NET_S_LINK_UP
))
780 return LL_FLUSH_FAILED
;
782 if (!napi_schedule_prep(napi
))
783 return LL_FLUSH_BUSY
;
785 virtqueue_disable_cb(rq
->vq
);
788 received
+= virtnet_receive(rq
, budget
);
790 r
= virtqueue_enable_cb_prepare(rq
->vq
);
791 clear_bit(NAPI_STATE_SCHED
, &napi
->state
);
792 if (unlikely(virtqueue_poll(rq
->vq
, r
)) &&
793 napi_schedule_prep(napi
)) {
794 virtqueue_disable_cb(rq
->vq
);
795 if (received
< budget
) {
799 __napi_schedule(napi
);
805 #endif /* CONFIG_NET_RX_BUSY_POLL */
807 static int virtnet_open(struct net_device
*dev
)
809 struct virtnet_info
*vi
= netdev_priv(dev
);
812 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
813 if (i
< vi
->curr_queue_pairs
)
814 /* Make sure we have some buffers: if oom use wq. */
815 if (!try_fill_recv(vi
, &vi
->rq
[i
], GFP_KERNEL
))
816 schedule_delayed_work(&vi
->refill
, 0);
817 virtnet_napi_enable(&vi
->rq
[i
]);
823 static void free_old_xmit_skbs(struct send_queue
*sq
)
827 struct virtnet_info
*vi
= sq
->vq
->vdev
->priv
;
828 struct virtnet_stats
*stats
= this_cpu_ptr(vi
->stats
);
830 while ((skb
= virtqueue_get_buf(sq
->vq
, &len
)) != NULL
) {
831 pr_debug("Sent skb %p\n", skb
);
833 u64_stats_update_begin(&stats
->tx_syncp
);
834 stats
->tx_bytes
+= skb
->len
;
836 u64_stats_update_end(&stats
->tx_syncp
);
838 dev_kfree_skb_any(skb
);
842 static int xmit_skb(struct send_queue
*sq
, struct sk_buff
*skb
)
844 struct virtio_net_hdr_mrg_rxbuf
*hdr
;
845 const unsigned char *dest
= ((struct ethhdr
*)skb
->data
)->h_dest
;
846 struct virtnet_info
*vi
= sq
->vq
->vdev
->priv
;
848 unsigned hdr_len
= vi
->hdr_len
;
851 pr_debug("%s: xmit %p %pM\n", vi
->dev
->name
, skb
, dest
);
853 can_push
= vi
->any_header_sg
&&
854 !((unsigned long)skb
->data
& (__alignof__(*hdr
) - 1)) &&
855 !skb_header_cloned(skb
) && skb_headroom(skb
) >= hdr_len
;
856 /* Even if we can, don't push here yet as this would skew
857 * csum_start offset below. */
859 hdr
= (struct virtio_net_hdr_mrg_rxbuf
*)(skb
->data
- hdr_len
);
861 hdr
= skb_vnet_hdr(skb
);
863 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
864 hdr
->hdr
.flags
= VIRTIO_NET_HDR_F_NEEDS_CSUM
;
865 hdr
->hdr
.csum_start
= cpu_to_virtio16(vi
->vdev
,
866 skb_checksum_start_offset(skb
));
867 hdr
->hdr
.csum_offset
= cpu_to_virtio16(vi
->vdev
,
871 hdr
->hdr
.csum_offset
= hdr
->hdr
.csum_start
= 0;
874 if (skb_is_gso(skb
)) {
875 hdr
->hdr
.hdr_len
= cpu_to_virtio16(vi
->vdev
, skb_headlen(skb
));
876 hdr
->hdr
.gso_size
= cpu_to_virtio16(vi
->vdev
,
877 skb_shinfo(skb
)->gso_size
);
878 if (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCPV4
)
879 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV4
;
880 else if (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCPV6
)
881 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV6
;
882 else if (skb_shinfo(skb
)->gso_type
& SKB_GSO_UDP
)
883 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_UDP
;
886 if (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCP_ECN
)
887 hdr
->hdr
.gso_type
|= VIRTIO_NET_HDR_GSO_ECN
;
889 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
890 hdr
->hdr
.gso_size
= hdr
->hdr
.hdr_len
= 0;
893 if (vi
->mergeable_rx_bufs
)
894 hdr
->num_buffers
= 0;
896 sg_init_table(sq
->sg
, MAX_SKB_FRAGS
+ 2);
898 __skb_push(skb
, hdr_len
);
899 num_sg
= skb_to_sgvec(skb
, sq
->sg
, 0, skb
->len
);
900 /* Pull header back to avoid skew in tx bytes calculations. */
901 __skb_pull(skb
, hdr_len
);
903 sg_set_buf(sq
->sg
, hdr
, hdr_len
);
904 num_sg
= skb_to_sgvec(skb
, sq
->sg
+ 1, 0, skb
->len
) + 1;
906 return virtqueue_add_outbuf(sq
->vq
, sq
->sg
, num_sg
, skb
, GFP_ATOMIC
);
909 static netdev_tx_t
start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
911 struct virtnet_info
*vi
= netdev_priv(dev
);
912 int qnum
= skb_get_queue_mapping(skb
);
913 struct send_queue
*sq
= &vi
->sq
[qnum
];
915 struct netdev_queue
*txq
= netdev_get_tx_queue(dev
, qnum
);
916 bool kick
= !skb
->xmit_more
;
918 /* Free up any pending old buffers before queueing new ones. */
919 free_old_xmit_skbs(sq
);
921 /* timestamp packet in software */
922 skb_tx_timestamp(skb
);
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
);
938 /* Don't wait up for transmitted skbs to be freed. */
942 /* Apparently nice girls don't return TX_BUSY; stop the queue
943 * before it gets out of hand. Naturally, this wastes entries. */
944 if (sq
->vq
->num_free
< 2+MAX_SKB_FRAGS
) {
945 netif_stop_subqueue(dev
, qnum
);
946 if (unlikely(!virtqueue_enable_cb_delayed(sq
->vq
))) {
947 /* More just got used, free them then recheck. */
948 free_old_xmit_skbs(sq
);
949 if (sq
->vq
->num_free
>= 2+MAX_SKB_FRAGS
) {
950 netif_start_subqueue(dev
, qnum
);
951 virtqueue_disable_cb(sq
->vq
);
956 if (kick
|| netif_xmit_stopped(txq
))
957 virtqueue_kick(sq
->vq
);
963 * Send command via the control virtqueue and check status. Commands
964 * supported by the hypervisor, as indicated by feature bits, should
965 * never fail unless improperly formatted.
967 static bool virtnet_send_command(struct virtnet_info
*vi
, u8
class, u8 cmd
,
968 struct scatterlist
*out
)
970 struct scatterlist
*sgs
[4], hdr
, stat
;
971 struct virtio_net_ctrl_hdr ctrl
;
972 virtio_net_ctrl_ack status
= ~0;
973 unsigned out_num
= 0, tmp
;
975 /* Caller should know better */
976 BUG_ON(!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
));
981 sg_init_one(&hdr
, &ctrl
, sizeof(ctrl
));
982 sgs
[out_num
++] = &hdr
;
985 sgs
[out_num
++] = out
;
987 /* Add return status. */
988 sg_init_one(&stat
, &status
, sizeof(status
));
989 sgs
[out_num
] = &stat
;
991 BUG_ON(out_num
+ 1 > ARRAY_SIZE(sgs
));
992 virtqueue_add_sgs(vi
->cvq
, sgs
, out_num
, 1, vi
, GFP_ATOMIC
);
994 if (unlikely(!virtqueue_kick(vi
->cvq
)))
995 return status
== VIRTIO_NET_OK
;
997 /* Spin for a response, the kick causes an ioport write, trapping
998 * into the hypervisor, so the request should be handled immediately.
1000 while (!virtqueue_get_buf(vi
->cvq
, &tmp
) &&
1001 !virtqueue_is_broken(vi
->cvq
))
1004 return status
== VIRTIO_NET_OK
;
1007 static int virtnet_set_mac_address(struct net_device
*dev
, void *p
)
1009 struct virtnet_info
*vi
= netdev_priv(dev
);
1010 struct virtio_device
*vdev
= vi
->vdev
;
1012 struct sockaddr
*addr
= p
;
1013 struct scatterlist sg
;
1015 ret
= eth_prepare_mac_addr_change(dev
, p
);
1019 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_MAC_ADDR
)) {
1020 sg_init_one(&sg
, addr
->sa_data
, dev
->addr_len
);
1021 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MAC
,
1022 VIRTIO_NET_CTRL_MAC_ADDR_SET
, &sg
)) {
1023 dev_warn(&vdev
->dev
,
1024 "Failed to set mac address by vq command.\n");
1027 } else if (virtio_has_feature(vdev
, VIRTIO_NET_F_MAC
) &&
1028 !virtio_has_feature(vdev
, VIRTIO_F_VERSION_1
)) {
1031 /* Naturally, this has an atomicity problem. */
1032 for (i
= 0; i
< dev
->addr_len
; i
++)
1033 virtio_cwrite8(vdev
,
1034 offsetof(struct virtio_net_config
, mac
) +
1035 i
, addr
->sa_data
[i
]);
1038 eth_commit_mac_addr_change(dev
, p
);
1043 static struct rtnl_link_stats64
*virtnet_stats(struct net_device
*dev
,
1044 struct rtnl_link_stats64
*tot
)
1046 struct virtnet_info
*vi
= netdev_priv(dev
);
1050 for_each_possible_cpu(cpu
) {
1051 struct virtnet_stats
*stats
= per_cpu_ptr(vi
->stats
, cpu
);
1052 u64 tpackets
, tbytes
, rpackets
, rbytes
;
1055 start
= u64_stats_fetch_begin_irq(&stats
->tx_syncp
);
1056 tpackets
= stats
->tx_packets
;
1057 tbytes
= stats
->tx_bytes
;
1058 } while (u64_stats_fetch_retry_irq(&stats
->tx_syncp
, start
));
1061 start
= u64_stats_fetch_begin_irq(&stats
->rx_syncp
);
1062 rpackets
= stats
->rx_packets
;
1063 rbytes
= stats
->rx_bytes
;
1064 } while (u64_stats_fetch_retry_irq(&stats
->rx_syncp
, start
));
1066 tot
->rx_packets
+= rpackets
;
1067 tot
->tx_packets
+= tpackets
;
1068 tot
->rx_bytes
+= rbytes
;
1069 tot
->tx_bytes
+= tbytes
;
1072 tot
->tx_dropped
= dev
->stats
.tx_dropped
;
1073 tot
->tx_fifo_errors
= dev
->stats
.tx_fifo_errors
;
1074 tot
->rx_dropped
= dev
->stats
.rx_dropped
;
1075 tot
->rx_length_errors
= dev
->stats
.rx_length_errors
;
1076 tot
->rx_frame_errors
= dev
->stats
.rx_frame_errors
;
1081 #ifdef CONFIG_NET_POLL_CONTROLLER
1082 static void virtnet_netpoll(struct net_device
*dev
)
1084 struct virtnet_info
*vi
= netdev_priv(dev
);
1087 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++)
1088 napi_schedule(&vi
->rq
[i
].napi
);
1092 static void virtnet_ack_link_announce(struct virtnet_info
*vi
)
1095 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_ANNOUNCE
,
1096 VIRTIO_NET_CTRL_ANNOUNCE_ACK
, NULL
))
1097 dev_warn(&vi
->dev
->dev
, "Failed to ack link announce.\n");
1101 static int virtnet_set_queues(struct virtnet_info
*vi
, u16 queue_pairs
)
1103 struct scatterlist sg
;
1104 struct virtio_net_ctrl_mq s
;
1105 struct net_device
*dev
= vi
->dev
;
1107 if (!vi
->has_cvq
|| !virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_MQ
))
1110 s
.virtqueue_pairs
= cpu_to_virtio16(vi
->vdev
, queue_pairs
);
1111 sg_init_one(&sg
, &s
, sizeof(s
));
1113 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MQ
,
1114 VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET
, &sg
)) {
1115 dev_warn(&dev
->dev
, "Fail to set num of queue pairs to %d\n",
1119 vi
->curr_queue_pairs
= queue_pairs
;
1120 /* virtnet_open() will refill when device is going to up. */
1121 if (dev
->flags
& IFF_UP
)
1122 schedule_delayed_work(&vi
->refill
, 0);
1128 static int virtnet_close(struct net_device
*dev
)
1130 struct virtnet_info
*vi
= netdev_priv(dev
);
1133 /* Make sure refill_work doesn't re-enable napi! */
1134 cancel_delayed_work_sync(&vi
->refill
);
1136 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
1137 napi_disable(&vi
->rq
[i
].napi
);
1142 static void virtnet_set_rx_mode(struct net_device
*dev
)
1144 struct virtnet_info
*vi
= netdev_priv(dev
);
1145 struct scatterlist sg
[2];
1146 u8 promisc
, allmulti
;
1147 struct virtio_net_ctrl_mac
*mac_data
;
1148 struct netdev_hw_addr
*ha
;
1154 /* We can't dynamically set ndo_set_rx_mode, so return gracefully */
1155 if (!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_RX
))
1158 promisc
= ((dev
->flags
& IFF_PROMISC
) != 0);
1159 allmulti
= ((dev
->flags
& IFF_ALLMULTI
) != 0);
1161 sg_init_one(sg
, &promisc
, sizeof(promisc
));
1163 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_RX
,
1164 VIRTIO_NET_CTRL_RX_PROMISC
, sg
))
1165 dev_warn(&dev
->dev
, "Failed to %sable promisc mode.\n",
1166 promisc
? "en" : "dis");
1168 sg_init_one(sg
, &allmulti
, sizeof(allmulti
));
1170 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_RX
,
1171 VIRTIO_NET_CTRL_RX_ALLMULTI
, sg
))
1172 dev_warn(&dev
->dev
, "Failed to %sable allmulti mode.\n",
1173 allmulti
? "en" : "dis");
1175 uc_count
= netdev_uc_count(dev
);
1176 mc_count
= netdev_mc_count(dev
);
1177 /* MAC filter - use one buffer for both lists */
1178 buf
= kzalloc(((uc_count
+ mc_count
) * ETH_ALEN
) +
1179 (2 * sizeof(mac_data
->entries
)), GFP_ATOMIC
);
1184 sg_init_table(sg
, 2);
1186 /* Store the unicast list and count in the front of the buffer */
1187 mac_data
->entries
= cpu_to_virtio32(vi
->vdev
, uc_count
);
1189 netdev_for_each_uc_addr(ha
, dev
)
1190 memcpy(&mac_data
->macs
[i
++][0], ha
->addr
, ETH_ALEN
);
1192 sg_set_buf(&sg
[0], mac_data
,
1193 sizeof(mac_data
->entries
) + (uc_count
* ETH_ALEN
));
1195 /* multicast list and count fill the end */
1196 mac_data
= (void *)&mac_data
->macs
[uc_count
][0];
1198 mac_data
->entries
= cpu_to_virtio32(vi
->vdev
, mc_count
);
1200 netdev_for_each_mc_addr(ha
, dev
)
1201 memcpy(&mac_data
->macs
[i
++][0], ha
->addr
, ETH_ALEN
);
1203 sg_set_buf(&sg
[1], mac_data
,
1204 sizeof(mac_data
->entries
) + (mc_count
* ETH_ALEN
));
1206 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MAC
,
1207 VIRTIO_NET_CTRL_MAC_TABLE_SET
, sg
))
1208 dev_warn(&dev
->dev
, "Failed to set MAC filter table.\n");
1213 static int virtnet_vlan_rx_add_vid(struct net_device
*dev
,
1214 __be16 proto
, u16 vid
)
1216 struct virtnet_info
*vi
= netdev_priv(dev
);
1217 struct scatterlist sg
;
1219 sg_init_one(&sg
, &vid
, sizeof(vid
));
1221 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_VLAN
,
1222 VIRTIO_NET_CTRL_VLAN_ADD
, &sg
))
1223 dev_warn(&dev
->dev
, "Failed to add VLAN ID %d.\n", vid
);
1227 static int virtnet_vlan_rx_kill_vid(struct net_device
*dev
,
1228 __be16 proto
, u16 vid
)
1230 struct virtnet_info
*vi
= netdev_priv(dev
);
1231 struct scatterlist sg
;
1233 sg_init_one(&sg
, &vid
, sizeof(vid
));
1235 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_VLAN
,
1236 VIRTIO_NET_CTRL_VLAN_DEL
, &sg
))
1237 dev_warn(&dev
->dev
, "Failed to kill VLAN ID %d.\n", vid
);
1241 static void virtnet_clean_affinity(struct virtnet_info
*vi
, long hcpu
)
1245 if (vi
->affinity_hint_set
) {
1246 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1247 virtqueue_set_affinity(vi
->rq
[i
].vq
, -1);
1248 virtqueue_set_affinity(vi
->sq
[i
].vq
, -1);
1251 vi
->affinity_hint_set
= false;
1255 static void virtnet_set_affinity(struct virtnet_info
*vi
)
1260 /* In multiqueue mode, when the number of cpu is equal to the number of
1261 * queue pairs, we let the queue pairs to be private to one cpu by
1262 * setting the affinity hint to eliminate the contention.
1264 if (vi
->curr_queue_pairs
== 1 ||
1265 vi
->max_queue_pairs
!= num_online_cpus()) {
1266 virtnet_clean_affinity(vi
, -1);
1271 for_each_online_cpu(cpu
) {
1272 virtqueue_set_affinity(vi
->rq
[i
].vq
, cpu
);
1273 virtqueue_set_affinity(vi
->sq
[i
].vq
, cpu
);
1274 netif_set_xps_queue(vi
->dev
, cpumask_of(cpu
), i
);
1278 vi
->affinity_hint_set
= true;
1281 static int virtnet_cpu_callback(struct notifier_block
*nfb
,
1282 unsigned long action
, void *hcpu
)
1284 struct virtnet_info
*vi
= container_of(nfb
, struct virtnet_info
, nb
);
1286 switch(action
& ~CPU_TASKS_FROZEN
) {
1288 case CPU_DOWN_FAILED
:
1290 virtnet_set_affinity(vi
);
1292 case CPU_DOWN_PREPARE
:
1293 virtnet_clean_affinity(vi
, (long)hcpu
);
1302 static void virtnet_get_ringparam(struct net_device
*dev
,
1303 struct ethtool_ringparam
*ring
)
1305 struct virtnet_info
*vi
= netdev_priv(dev
);
1307 ring
->rx_max_pending
= virtqueue_get_vring_size(vi
->rq
[0].vq
);
1308 ring
->tx_max_pending
= virtqueue_get_vring_size(vi
->sq
[0].vq
);
1309 ring
->rx_pending
= ring
->rx_max_pending
;
1310 ring
->tx_pending
= ring
->tx_max_pending
;
1314 static void virtnet_get_drvinfo(struct net_device
*dev
,
1315 struct ethtool_drvinfo
*info
)
1317 struct virtnet_info
*vi
= netdev_priv(dev
);
1318 struct virtio_device
*vdev
= vi
->vdev
;
1320 strlcpy(info
->driver
, KBUILD_MODNAME
, sizeof(info
->driver
));
1321 strlcpy(info
->version
, VIRTNET_DRIVER_VERSION
, sizeof(info
->version
));
1322 strlcpy(info
->bus_info
, virtio_bus_name(vdev
), sizeof(info
->bus_info
));
1326 /* TODO: Eliminate OOO packets during switching */
1327 static int virtnet_set_channels(struct net_device
*dev
,
1328 struct ethtool_channels
*channels
)
1330 struct virtnet_info
*vi
= netdev_priv(dev
);
1331 u16 queue_pairs
= channels
->combined_count
;
1334 /* We don't support separate rx/tx channels.
1335 * We don't allow setting 'other' channels.
1337 if (channels
->rx_count
|| channels
->tx_count
|| channels
->other_count
)
1340 if (queue_pairs
> vi
->max_queue_pairs
|| queue_pairs
== 0)
1344 err
= virtnet_set_queues(vi
, queue_pairs
);
1346 netif_set_real_num_tx_queues(dev
, queue_pairs
);
1347 netif_set_real_num_rx_queues(dev
, queue_pairs
);
1349 virtnet_set_affinity(vi
);
1356 static void virtnet_get_channels(struct net_device
*dev
,
1357 struct ethtool_channels
*channels
)
1359 struct virtnet_info
*vi
= netdev_priv(dev
);
1361 channels
->combined_count
= vi
->curr_queue_pairs
;
1362 channels
->max_combined
= vi
->max_queue_pairs
;
1363 channels
->max_other
= 0;
1364 channels
->rx_count
= 0;
1365 channels
->tx_count
= 0;
1366 channels
->other_count
= 0;
1369 static const struct ethtool_ops virtnet_ethtool_ops
= {
1370 .get_drvinfo
= virtnet_get_drvinfo
,
1371 .get_link
= ethtool_op_get_link
,
1372 .get_ringparam
= virtnet_get_ringparam
,
1373 .set_channels
= virtnet_set_channels
,
1374 .get_channels
= virtnet_get_channels
,
1375 .get_ts_info
= ethtool_op_get_ts_info
,
1379 #define MAX_MTU 65535
1381 static int virtnet_change_mtu(struct net_device
*dev
, int new_mtu
)
1383 if (new_mtu
< MIN_MTU
|| new_mtu
> MAX_MTU
)
1389 static const struct net_device_ops virtnet_netdev
= {
1390 .ndo_open
= virtnet_open
,
1391 .ndo_stop
= virtnet_close
,
1392 .ndo_start_xmit
= start_xmit
,
1393 .ndo_validate_addr
= eth_validate_addr
,
1394 .ndo_set_mac_address
= virtnet_set_mac_address
,
1395 .ndo_set_rx_mode
= virtnet_set_rx_mode
,
1396 .ndo_change_mtu
= virtnet_change_mtu
,
1397 .ndo_get_stats64
= virtnet_stats
,
1398 .ndo_vlan_rx_add_vid
= virtnet_vlan_rx_add_vid
,
1399 .ndo_vlan_rx_kill_vid
= virtnet_vlan_rx_kill_vid
,
1400 #ifdef CONFIG_NET_POLL_CONTROLLER
1401 .ndo_poll_controller
= virtnet_netpoll
,
1403 #ifdef CONFIG_NET_RX_BUSY_POLL
1404 .ndo_busy_poll
= virtnet_busy_poll
,
1408 static void virtnet_config_changed_work(struct work_struct
*work
)
1410 struct virtnet_info
*vi
=
1411 container_of(work
, struct virtnet_info
, config_work
);
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
);
1440 static void virtnet_config_changed(struct virtio_device
*vdev
)
1442 struct virtnet_info
*vi
= vdev
->priv
;
1444 schedule_work(&vi
->config_work
);
1447 static void virtnet_free_queues(struct virtnet_info
*vi
)
1451 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
1452 netif_napi_del(&vi
->rq
[i
].napi
);
1458 static void free_receive_bufs(struct virtnet_info
*vi
)
1462 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1463 while (vi
->rq
[i
].pages
)
1464 __free_pages(get_a_page(&vi
->rq
[i
], GFP_KERNEL
), 0);
1468 static void free_receive_page_frags(struct virtnet_info
*vi
)
1471 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
1472 if (vi
->rq
[i
].alloc_frag
.page
)
1473 put_page(vi
->rq
[i
].alloc_frag
.page
);
1476 static void free_unused_bufs(struct virtnet_info
*vi
)
1481 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1482 struct virtqueue
*vq
= vi
->sq
[i
].vq
;
1483 while ((buf
= virtqueue_detach_unused_buf(vq
)) != NULL
)
1487 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1488 struct virtqueue
*vq
= vi
->rq
[i
].vq
;
1490 while ((buf
= virtqueue_detach_unused_buf(vq
)) != NULL
) {
1491 if (vi
->mergeable_rx_bufs
) {
1492 unsigned long ctx
= (unsigned long)buf
;
1493 void *base
= mergeable_ctx_to_buf_address(ctx
);
1494 put_page(virt_to_head_page(base
));
1495 } else if (vi
->big_packets
) {
1496 give_pages(&vi
->rq
[i
], buf
);
1504 static void virtnet_del_vqs(struct virtnet_info
*vi
)
1506 struct virtio_device
*vdev
= vi
->vdev
;
1508 virtnet_clean_affinity(vi
, -1);
1510 vdev
->config
->del_vqs(vdev
);
1512 virtnet_free_queues(vi
);
1515 static int virtnet_find_vqs(struct virtnet_info
*vi
)
1517 vq_callback_t
**callbacks
;
1518 struct virtqueue
**vqs
;
1523 /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
1524 * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
1525 * possible control vq.
1527 total_vqs
= vi
->max_queue_pairs
* 2 +
1528 virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
);
1530 /* Allocate space for find_vqs parameters */
1531 vqs
= kzalloc(total_vqs
* sizeof(*vqs
), GFP_KERNEL
);
1534 callbacks
= kmalloc(total_vqs
* sizeof(*callbacks
), GFP_KERNEL
);
1537 names
= kmalloc(total_vqs
* sizeof(*names
), GFP_KERNEL
);
1541 /* Parameters for control virtqueue, if any */
1543 callbacks
[total_vqs
- 1] = NULL
;
1544 names
[total_vqs
- 1] = "control";
1547 /* Allocate/initialize parameters for send/receive virtqueues */
1548 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1549 callbacks
[rxq2vq(i
)] = skb_recv_done
;
1550 callbacks
[txq2vq(i
)] = skb_xmit_done
;
1551 sprintf(vi
->rq
[i
].name
, "input.%d", i
);
1552 sprintf(vi
->sq
[i
].name
, "output.%d", i
);
1553 names
[rxq2vq(i
)] = vi
->rq
[i
].name
;
1554 names
[txq2vq(i
)] = vi
->sq
[i
].name
;
1557 ret
= vi
->vdev
->config
->find_vqs(vi
->vdev
, total_vqs
, vqs
, callbacks
,
1563 vi
->cvq
= vqs
[total_vqs
- 1];
1564 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VLAN
))
1565 vi
->dev
->features
|= NETIF_F_HW_VLAN_CTAG_FILTER
;
1568 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1569 vi
->rq
[i
].vq
= vqs
[rxq2vq(i
)];
1570 vi
->sq
[i
].vq
= vqs
[txq2vq(i
)];
1589 static int virtnet_alloc_queues(struct virtnet_info
*vi
)
1593 vi
->sq
= kzalloc(sizeof(*vi
->sq
) * vi
->max_queue_pairs
, GFP_KERNEL
);
1596 vi
->rq
= kzalloc(sizeof(*vi
->rq
) * vi
->max_queue_pairs
, GFP_KERNEL
);
1600 INIT_DELAYED_WORK(&vi
->refill
, refill_work
);
1601 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1602 vi
->rq
[i
].pages
= NULL
;
1603 netif_napi_add(vi
->dev
, &vi
->rq
[i
].napi
, virtnet_poll
,
1605 napi_hash_add(&vi
->rq
[i
].napi
);
1607 sg_init_table(vi
->rq
[i
].sg
, ARRAY_SIZE(vi
->rq
[i
].sg
));
1608 ewma_init(&vi
->rq
[i
].mrg_avg_pkt_len
, 1, RECEIVE_AVG_WEIGHT
);
1609 sg_init_table(vi
->sq
[i
].sg
, ARRAY_SIZE(vi
->sq
[i
].sg
));
1620 static int init_vqs(struct virtnet_info
*vi
)
1624 /* Allocate send & receive queues */
1625 ret
= virtnet_alloc_queues(vi
);
1629 ret
= virtnet_find_vqs(vi
);
1634 virtnet_set_affinity(vi
);
1640 virtnet_free_queues(vi
);
1646 static ssize_t
mergeable_rx_buffer_size_show(struct netdev_rx_queue
*queue
,
1647 struct rx_queue_attribute
*attribute
, char *buf
)
1649 struct virtnet_info
*vi
= netdev_priv(queue
->dev
);
1650 unsigned int queue_index
= get_netdev_rx_queue_index(queue
);
1653 BUG_ON(queue_index
>= vi
->max_queue_pairs
);
1654 avg
= &vi
->rq
[queue_index
].mrg_avg_pkt_len
;
1655 return sprintf(buf
, "%u\n", get_mergeable_buf_len(avg
));
1658 static struct rx_queue_attribute mergeable_rx_buffer_size_attribute
=
1659 __ATTR_RO(mergeable_rx_buffer_size
);
1661 static struct attribute
*virtio_net_mrg_rx_attrs
[] = {
1662 &mergeable_rx_buffer_size_attribute
.attr
,
1666 static const struct attribute_group virtio_net_mrg_rx_group
= {
1667 .name
= "virtio_net",
1668 .attrs
= virtio_net_mrg_rx_attrs
1672 static bool virtnet_fail_on_feature(struct virtio_device
*vdev
,
1674 const char *fname
, const char *dname
)
1676 if (!virtio_has_feature(vdev
, fbit
))
1679 dev_err(&vdev
->dev
, "device advertises feature %s but not %s",
1685 #define VIRTNET_FAIL_ON(vdev, fbit, dbit) \
1686 virtnet_fail_on_feature(vdev, fbit, #fbit, dbit)
1688 static bool virtnet_validate_features(struct virtio_device
*vdev
)
1690 if (!virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
) &&
1691 (VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_CTRL_RX
,
1692 "VIRTIO_NET_F_CTRL_VQ") ||
1693 VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_CTRL_VLAN
,
1694 "VIRTIO_NET_F_CTRL_VQ") ||
1695 VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_GUEST_ANNOUNCE
,
1696 "VIRTIO_NET_F_CTRL_VQ") ||
1697 VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_MQ
, "VIRTIO_NET_F_CTRL_VQ") ||
1698 VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_CTRL_MAC_ADDR
,
1699 "VIRTIO_NET_F_CTRL_VQ"))) {
1706 static int virtnet_probe(struct virtio_device
*vdev
)
1709 struct net_device
*dev
;
1710 struct virtnet_info
*vi
;
1711 u16 max_queue_pairs
;
1713 if (!vdev
->config
->get
) {
1714 dev_err(&vdev
->dev
, "%s failure: config access disabled\n",
1719 if (!virtnet_validate_features(vdev
))
1722 /* Find if host supports multiqueue virtio_net device */
1723 err
= virtio_cread_feature(vdev
, VIRTIO_NET_F_MQ
,
1724 struct virtio_net_config
,
1725 max_virtqueue_pairs
, &max_queue_pairs
);
1727 /* We need at least 2 queue's */
1728 if (err
|| max_queue_pairs
< VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN
||
1729 max_queue_pairs
> VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX
||
1730 !virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
))
1731 max_queue_pairs
= 1;
1733 /* Allocate ourselves a network device with room for our info */
1734 dev
= alloc_etherdev_mq(sizeof(struct virtnet_info
), max_queue_pairs
);
1738 /* Set up network device as normal. */
1739 dev
->priv_flags
|= IFF_UNICAST_FLT
| IFF_LIVE_ADDR_CHANGE
;
1740 dev
->netdev_ops
= &virtnet_netdev
;
1741 dev
->features
= NETIF_F_HIGHDMA
;
1743 dev
->ethtool_ops
= &virtnet_ethtool_ops
;
1744 SET_NETDEV_DEV(dev
, &vdev
->dev
);
1746 /* Do we support "hardware" checksums? */
1747 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CSUM
)) {
1748 /* This opens up the world of extra features. */
1749 dev
->hw_features
|= NETIF_F_HW_CSUM
|NETIF_F_SG
|NETIF_F_FRAGLIST
;
1751 dev
->features
|= NETIF_F_HW_CSUM
|NETIF_F_SG
|NETIF_F_FRAGLIST
;
1753 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GSO
)) {
1754 dev
->hw_features
|= NETIF_F_TSO
| NETIF_F_UFO
1755 | NETIF_F_TSO_ECN
| NETIF_F_TSO6
;
1757 /* Individual feature bits: what can host handle? */
1758 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_TSO4
))
1759 dev
->hw_features
|= NETIF_F_TSO
;
1760 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_TSO6
))
1761 dev
->hw_features
|= NETIF_F_TSO6
;
1762 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_ECN
))
1763 dev
->hw_features
|= NETIF_F_TSO_ECN
;
1764 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_UFO
))
1765 dev
->hw_features
|= NETIF_F_UFO
;
1767 dev
->features
|= NETIF_F_GSO_ROBUST
;
1770 dev
->features
|= dev
->hw_features
& (NETIF_F_ALL_TSO
|NETIF_F_UFO
);
1771 /* (!csum && gso) case will be fixed by register_netdev() */
1773 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_CSUM
))
1774 dev
->features
|= NETIF_F_RXCSUM
;
1776 dev
->vlan_features
= dev
->features
;
1778 /* Configuration may specify what MAC to use. Otherwise random. */
1779 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MAC
))
1780 virtio_cread_bytes(vdev
,
1781 offsetof(struct virtio_net_config
, mac
),
1782 dev
->dev_addr
, dev
->addr_len
);
1784 eth_hw_addr_random(dev
);
1786 /* Set up our device-specific information */
1787 vi
= netdev_priv(dev
);
1791 vi
->stats
= alloc_percpu(struct virtnet_stats
);
1793 if (vi
->stats
== NULL
)
1796 for_each_possible_cpu(i
) {
1797 struct virtnet_stats
*virtnet_stats
;
1798 virtnet_stats
= per_cpu_ptr(vi
->stats
, i
);
1799 u64_stats_init(&virtnet_stats
->tx_syncp
);
1800 u64_stats_init(&virtnet_stats
->rx_syncp
);
1803 INIT_WORK(&vi
->config_work
, virtnet_config_changed_work
);
1805 /* If we can receive ANY GSO packets, we must allocate large ones. */
1806 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_TSO4
) ||
1807 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_TSO6
) ||
1808 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_ECN
) ||
1809 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_UFO
))
1810 vi
->big_packets
= true;
1812 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MRG_RXBUF
))
1813 vi
->mergeable_rx_bufs
= true;
1815 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MRG_RXBUF
) ||
1816 virtio_has_feature(vdev
, VIRTIO_F_VERSION_1
))
1817 vi
->hdr_len
= sizeof(struct virtio_net_hdr_mrg_rxbuf
);
1819 vi
->hdr_len
= sizeof(struct virtio_net_hdr
);
1821 if (virtio_has_feature(vdev
, VIRTIO_F_ANY_LAYOUT
))
1822 vi
->any_header_sg
= true;
1824 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
))
1827 if (vi
->any_header_sg
)
1828 dev
->needed_headroom
= vi
->hdr_len
;
1830 /* Use single tx/rx queue pair as default */
1831 vi
->curr_queue_pairs
= 1;
1832 vi
->max_queue_pairs
= max_queue_pairs
;
1834 /* Allocate/initialize the rx/tx queues, and invoke find_vqs */
1840 if (vi
->mergeable_rx_bufs
)
1841 dev
->sysfs_rx_queue_group
= &virtio_net_mrg_rx_group
;
1843 netif_set_real_num_tx_queues(dev
, vi
->curr_queue_pairs
);
1844 netif_set_real_num_rx_queues(dev
, vi
->curr_queue_pairs
);
1846 err
= register_netdev(dev
);
1848 pr_debug("virtio_net: registering device failed\n");
1852 virtio_device_ready(vdev
);
1854 /* Last of all, set up some receive buffers. */
1855 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++) {
1856 try_fill_recv(vi
, &vi
->rq
[i
], GFP_KERNEL
);
1858 /* If we didn't even get one input buffer, we're useless. */
1859 if (vi
->rq
[i
].vq
->num_free
==
1860 virtqueue_get_vring_size(vi
->rq
[i
].vq
)) {
1861 free_unused_bufs(vi
);
1863 goto free_recv_bufs
;
1867 vi
->nb
.notifier_call
= &virtnet_cpu_callback
;
1868 err
= register_hotcpu_notifier(&vi
->nb
);
1870 pr_debug("virtio_net: registering cpu notifier failed\n");
1871 goto free_recv_bufs
;
1874 /* Assume link up if device can't report link status,
1875 otherwise get link status from config. */
1876 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_STATUS
)) {
1877 netif_carrier_off(dev
);
1878 schedule_work(&vi
->config_work
);
1880 vi
->status
= VIRTIO_NET_S_LINK_UP
;
1881 netif_carrier_on(dev
);
1884 pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
1885 dev
->name
, max_queue_pairs
);
1890 vi
->vdev
->config
->reset(vdev
);
1892 free_receive_bufs(vi
);
1893 unregister_netdev(dev
);
1895 cancel_delayed_work_sync(&vi
->refill
);
1896 free_receive_page_frags(vi
);
1897 virtnet_del_vqs(vi
);
1899 free_percpu(vi
->stats
);
1905 static void remove_vq_common(struct virtnet_info
*vi
)
1907 vi
->vdev
->config
->reset(vi
->vdev
);
1909 /* Free unused buffers in both send and recv, if any. */
1910 free_unused_bufs(vi
);
1912 free_receive_bufs(vi
);
1914 free_receive_page_frags(vi
);
1916 virtnet_del_vqs(vi
);
1919 static void virtnet_remove(struct virtio_device
*vdev
)
1921 struct virtnet_info
*vi
= vdev
->priv
;
1923 unregister_hotcpu_notifier(&vi
->nb
);
1925 /* Make sure no work handler is accessing the device. */
1926 flush_work(&vi
->config_work
);
1928 unregister_netdev(vi
->dev
);
1930 remove_vq_common(vi
);
1932 free_percpu(vi
->stats
);
1933 free_netdev(vi
->dev
);
1936 #ifdef CONFIG_PM_SLEEP
1937 static int virtnet_freeze(struct virtio_device
*vdev
)
1939 struct virtnet_info
*vi
= vdev
->priv
;
1942 unregister_hotcpu_notifier(&vi
->nb
);
1944 /* Make sure no work handler is accessing the device */
1945 flush_work(&vi
->config_work
);
1947 netif_device_detach(vi
->dev
);
1948 cancel_delayed_work_sync(&vi
->refill
);
1950 if (netif_running(vi
->dev
)) {
1951 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1952 napi_disable(&vi
->rq
[i
].napi
);
1953 napi_hash_del(&vi
->rq
[i
].napi
);
1954 netif_napi_del(&vi
->rq
[i
].napi
);
1958 remove_vq_common(vi
);
1963 static int virtnet_restore(struct virtio_device
*vdev
)
1965 struct virtnet_info
*vi
= vdev
->priv
;
1972 virtio_device_ready(vdev
);
1974 if (netif_running(vi
->dev
)) {
1975 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++)
1976 if (!try_fill_recv(vi
, &vi
->rq
[i
], GFP_KERNEL
))
1977 schedule_delayed_work(&vi
->refill
, 0);
1979 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
1980 virtnet_napi_enable(&vi
->rq
[i
]);
1983 netif_device_attach(vi
->dev
);
1986 virtnet_set_queues(vi
, vi
->curr_queue_pairs
);
1989 err
= register_hotcpu_notifier(&vi
->nb
);
1997 static struct virtio_device_id id_table
[] = {
1998 { VIRTIO_ID_NET
, VIRTIO_DEV_ANY_ID
},
2002 static unsigned int features
[] = {
2003 VIRTIO_NET_F_CSUM
, VIRTIO_NET_F_GUEST_CSUM
,
2004 VIRTIO_NET_F_GSO
, VIRTIO_NET_F_MAC
,
2005 VIRTIO_NET_F_HOST_TSO4
, VIRTIO_NET_F_HOST_UFO
, VIRTIO_NET_F_HOST_TSO6
,
2006 VIRTIO_NET_F_HOST_ECN
, VIRTIO_NET_F_GUEST_TSO4
, VIRTIO_NET_F_GUEST_TSO6
,
2007 VIRTIO_NET_F_GUEST_ECN
, VIRTIO_NET_F_GUEST_UFO
,
2008 VIRTIO_NET_F_MRG_RXBUF
, VIRTIO_NET_F_STATUS
, VIRTIO_NET_F_CTRL_VQ
,
2009 VIRTIO_NET_F_CTRL_RX
, VIRTIO_NET_F_CTRL_VLAN
,
2010 VIRTIO_NET_F_GUEST_ANNOUNCE
, VIRTIO_NET_F_MQ
,
2011 VIRTIO_NET_F_CTRL_MAC_ADDR
,
2012 VIRTIO_F_ANY_LAYOUT
,
2015 static struct virtio_driver virtio_net_driver
= {
2016 .feature_table
= features
,
2017 .feature_table_size
= ARRAY_SIZE(features
),
2018 .driver
.name
= KBUILD_MODNAME
,
2019 .driver
.owner
= THIS_MODULE
,
2020 .id_table
= id_table
,
2021 .probe
= virtnet_probe
,
2022 .remove
= virtnet_remove
,
2023 .config_changed
= virtnet_config_changed
,
2024 #ifdef CONFIG_PM_SLEEP
2025 .freeze
= virtnet_freeze
,
2026 .restore
= virtnet_restore
,
2030 module_virtio_driver(virtio_net_driver
);
2032 MODULE_DEVICE_TABLE(virtio
, id_table
);
2033 MODULE_DESCRIPTION("Virtio network driver");
2034 MODULE_LICENSE("GPL");