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 /* Active statistics */
127 struct virtnet_stats __percpu
*stats
;
129 /* Work struct for refilling if we run low on memory. */
130 struct delayed_work refill
;
132 /* Work struct for config space updates */
133 struct work_struct config_work
;
135 /* Lock for config space updates */
136 struct mutex config_lock
;
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 skb_vnet_hdr
{
147 struct virtio_net_hdr hdr
;
148 struct virtio_net_hdr_mrg_rxbuf mhdr
;
152 struct padded_vnet_hdr
{
153 struct virtio_net_hdr hdr
;
155 * virtio_net_hdr should be in a separated sg buffer because of a
156 * QEMU bug, and data sg buffer shares same page with this header sg.
157 * This padding makes next sg 16 byte aligned after virtio_net_hdr.
162 /* Converting between virtqueue no. and kernel tx/rx queue no.
163 * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq
165 static int vq2txq(struct virtqueue
*vq
)
167 return (vq
->index
- 1) / 2;
170 static int txq2vq(int txq
)
175 static int vq2rxq(struct virtqueue
*vq
)
177 return vq
->index
/ 2;
180 static int rxq2vq(int rxq
)
185 static inline struct skb_vnet_hdr
*skb_vnet_hdr(struct sk_buff
*skb
)
187 return (struct skb_vnet_hdr
*)skb
->cb
;
191 * private is used to chain pages for big packets, put the whole
192 * most recent used list in the beginning for reuse
194 static void give_pages(struct receive_queue
*rq
, struct page
*page
)
198 /* Find end of list, sew whole thing into vi->rq.pages. */
199 for (end
= page
; end
->private; end
= (struct page
*)end
->private);
200 end
->private = (unsigned long)rq
->pages
;
204 static struct page
*get_a_page(struct receive_queue
*rq
, gfp_t gfp_mask
)
206 struct page
*p
= rq
->pages
;
209 rq
->pages
= (struct page
*)p
->private;
210 /* clear private here, it is used to chain pages */
213 p
= alloc_page(gfp_mask
);
217 static void skb_xmit_done(struct virtqueue
*vq
)
219 struct virtnet_info
*vi
= vq
->vdev
->priv
;
221 /* Suppress further interrupts. */
222 virtqueue_disable_cb(vq
);
224 /* We were probably waiting for more output buffers. */
225 netif_wake_subqueue(vi
->dev
, vq2txq(vq
));
228 static unsigned int mergeable_ctx_to_buf_truesize(unsigned long mrg_ctx
)
230 unsigned int truesize
= mrg_ctx
& (MERGEABLE_BUFFER_ALIGN
- 1);
231 return (truesize
+ 1) * MERGEABLE_BUFFER_ALIGN
;
234 static void *mergeable_ctx_to_buf_address(unsigned long mrg_ctx
)
236 return (void *)(mrg_ctx
& -MERGEABLE_BUFFER_ALIGN
);
240 static unsigned long mergeable_buf_to_ctx(void *buf
, unsigned int truesize
)
242 unsigned int size
= truesize
/ MERGEABLE_BUFFER_ALIGN
;
243 return (unsigned long)buf
| (size
- 1);
246 /* Called from bottom half context */
247 static struct sk_buff
*page_to_skb(struct receive_queue
*rq
,
248 struct page
*page
, unsigned int offset
,
249 unsigned int len
, unsigned int truesize
)
251 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
253 struct skb_vnet_hdr
*hdr
;
254 unsigned int copy
, hdr_len
, hdr_padded_len
;
257 p
= page_address(page
) + offset
;
259 /* copy small packet so we can reuse these pages for small data */
260 skb
= netdev_alloc_skb_ip_align(vi
->dev
, GOOD_COPY_LEN
);
264 hdr
= skb_vnet_hdr(skb
);
266 if (vi
->mergeable_rx_bufs
) {
267 hdr_len
= sizeof hdr
->mhdr
;
268 hdr_padded_len
= sizeof hdr
->mhdr
;
270 hdr_len
= sizeof hdr
->hdr
;
271 hdr_padded_len
= sizeof(struct padded_vnet_hdr
);
274 memcpy(hdr
, p
, hdr_len
);
277 offset
+= hdr_padded_len
;
281 if (copy
> skb_tailroom(skb
))
282 copy
= skb_tailroom(skb
);
283 memcpy(skb_put(skb
, copy
), p
, copy
);
288 if (vi
->mergeable_rx_bufs
) {
290 skb_add_rx_frag(skb
, 0, page
, offset
, len
, truesize
);
297 * Verify that we can indeed put this data into a skb.
298 * This is here to handle cases when the device erroneously
299 * tries to receive more than is possible. This is usually
300 * the case of a broken device.
302 if (unlikely(len
> MAX_SKB_FRAGS
* PAGE_SIZE
)) {
303 net_dbg_ratelimited("%s: too much data\n", skb
->dev
->name
);
307 BUG_ON(offset
>= PAGE_SIZE
);
309 unsigned int frag_size
= min((unsigned)PAGE_SIZE
- offset
, len
);
310 skb_add_rx_frag(skb
, skb_shinfo(skb
)->nr_frags
, page
, offset
,
311 frag_size
, truesize
);
313 page
= (struct page
*)page
->private;
318 give_pages(rq
, page
);
323 static struct sk_buff
*receive_small(void *buf
, unsigned int len
)
325 struct sk_buff
* skb
= buf
;
327 len
-= sizeof(struct virtio_net_hdr
);
333 static struct sk_buff
*receive_big(struct net_device
*dev
,
334 struct receive_queue
*rq
,
338 struct page
*page
= buf
;
339 struct sk_buff
*skb
= page_to_skb(rq
, page
, 0, len
, PAGE_SIZE
);
347 dev
->stats
.rx_dropped
++;
348 give_pages(rq
, page
);
352 static struct sk_buff
*receive_mergeable(struct net_device
*dev
,
353 struct receive_queue
*rq
,
357 void *buf
= mergeable_ctx_to_buf_address(ctx
);
358 struct skb_vnet_hdr
*hdr
= buf
;
359 int num_buf
= hdr
->mhdr
.num_buffers
;
360 struct page
*page
= virt_to_head_page(buf
);
361 int offset
= buf
- page_address(page
);
362 unsigned int truesize
= max(len
, mergeable_ctx_to_buf_truesize(ctx
));
364 struct sk_buff
*head_skb
= page_to_skb(rq
, page
, offset
, len
, truesize
);
365 struct sk_buff
*curr_skb
= head_skb
;
367 if (unlikely(!curr_skb
))
372 ctx
= (unsigned long)virtqueue_get_buf(rq
->vq
, &len
);
373 if (unlikely(!ctx
)) {
374 pr_debug("%s: rx error: %d buffers out of %d missing\n",
375 dev
->name
, num_buf
, hdr
->mhdr
.num_buffers
);
376 dev
->stats
.rx_length_errors
++;
380 buf
= mergeable_ctx_to_buf_address(ctx
);
381 page
= virt_to_head_page(buf
);
383 num_skb_frags
= skb_shinfo(curr_skb
)->nr_frags
;
384 if (unlikely(num_skb_frags
== MAX_SKB_FRAGS
)) {
385 struct sk_buff
*nskb
= alloc_skb(0, GFP_ATOMIC
);
389 if (curr_skb
== head_skb
)
390 skb_shinfo(curr_skb
)->frag_list
= nskb
;
392 curr_skb
->next
= nskb
;
394 head_skb
->truesize
+= nskb
->truesize
;
397 truesize
= max(len
, mergeable_ctx_to_buf_truesize(ctx
));
398 if (curr_skb
!= head_skb
) {
399 head_skb
->data_len
+= len
;
400 head_skb
->len
+= len
;
401 head_skb
->truesize
+= truesize
;
403 offset
= buf
- page_address(page
);
404 if (skb_can_coalesce(curr_skb
, num_skb_frags
, page
, offset
)) {
406 skb_coalesce_rx_frag(curr_skb
, num_skb_frags
- 1,
409 skb_add_rx_frag(curr_skb
, num_skb_frags
, page
,
410 offset
, len
, truesize
);
414 ewma_add(&rq
->mrg_avg_pkt_len
, head_skb
->len
);
420 ctx
= (unsigned long)virtqueue_get_buf(rq
->vq
, &len
);
421 if (unlikely(!ctx
)) {
422 pr_debug("%s: rx error: %d buffers missing\n",
424 dev
->stats
.rx_length_errors
++;
427 page
= virt_to_head_page(mergeable_ctx_to_buf_address(ctx
));
431 dev
->stats
.rx_dropped
++;
432 dev_kfree_skb(head_skb
);
436 static void receive_buf(struct receive_queue
*rq
, void *buf
, unsigned int len
)
438 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
439 struct net_device
*dev
= vi
->dev
;
440 struct virtnet_stats
*stats
= this_cpu_ptr(vi
->stats
);
442 struct skb_vnet_hdr
*hdr
;
444 if (unlikely(len
< sizeof(struct virtio_net_hdr
) + ETH_HLEN
)) {
445 pr_debug("%s: short packet %i\n", dev
->name
, len
);
446 dev
->stats
.rx_length_errors
++;
447 if (vi
->mergeable_rx_bufs
) {
448 unsigned long ctx
= (unsigned long)buf
;
449 void *base
= mergeable_ctx_to_buf_address(ctx
);
450 put_page(virt_to_head_page(base
));
451 } else if (vi
->big_packets
) {
459 if (vi
->mergeable_rx_bufs
)
460 skb
= receive_mergeable(dev
, rq
, (unsigned long)buf
, len
);
461 else if (vi
->big_packets
)
462 skb
= receive_big(dev
, rq
, buf
, len
);
464 skb
= receive_small(buf
, len
);
469 hdr
= skb_vnet_hdr(skb
);
471 u64_stats_update_begin(&stats
->rx_syncp
);
472 stats
->rx_bytes
+= skb
->len
;
474 u64_stats_update_end(&stats
->rx_syncp
);
476 if (hdr
->hdr
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) {
477 pr_debug("Needs csum!\n");
478 if (!skb_partial_csum_set(skb
,
480 hdr
->hdr
.csum_offset
))
482 } else if (hdr
->hdr
.flags
& VIRTIO_NET_HDR_F_DATA_VALID
) {
483 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
486 skb
->protocol
= eth_type_trans(skb
, dev
);
487 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
488 ntohs(skb
->protocol
), skb
->len
, skb
->pkt_type
);
490 if (hdr
->hdr
.gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
492 switch (hdr
->hdr
.gso_type
& ~VIRTIO_NET_HDR_GSO_ECN
) {
493 case VIRTIO_NET_HDR_GSO_TCPV4
:
494 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV4
;
496 case VIRTIO_NET_HDR_GSO_UDP
:
497 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
499 case VIRTIO_NET_HDR_GSO_TCPV6
:
500 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV6
;
503 net_warn_ratelimited("%s: bad gso type %u.\n",
504 dev
->name
, hdr
->hdr
.gso_type
);
508 if (hdr
->hdr
.gso_type
& VIRTIO_NET_HDR_GSO_ECN
)
509 skb_shinfo(skb
)->gso_type
|= SKB_GSO_TCP_ECN
;
511 skb_shinfo(skb
)->gso_size
= hdr
->hdr
.gso_size
;
512 if (skb_shinfo(skb
)->gso_size
== 0) {
513 net_warn_ratelimited("%s: zero gso size.\n", dev
->name
);
517 /* Header must be checked, and gso_segs computed. */
518 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
519 skb_shinfo(skb
)->gso_segs
= 0;
522 skb_mark_napi_id(skb
, &rq
->napi
);
524 netif_receive_skb(skb
);
528 dev
->stats
.rx_frame_errors
++;
532 static int add_recvbuf_small(struct receive_queue
*rq
, gfp_t gfp
)
534 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
536 struct skb_vnet_hdr
*hdr
;
539 skb
= __netdev_alloc_skb_ip_align(vi
->dev
, GOOD_PACKET_LEN
, gfp
);
543 skb_put(skb
, GOOD_PACKET_LEN
);
545 hdr
= skb_vnet_hdr(skb
);
546 sg_set_buf(rq
->sg
, &hdr
->hdr
, sizeof hdr
->hdr
);
548 skb_to_sgvec(skb
, rq
->sg
+ 1, 0, skb
->len
);
550 err
= virtqueue_add_inbuf(rq
->vq
, rq
->sg
, 2, skb
, gfp
);
557 static int add_recvbuf_big(struct receive_queue
*rq
, gfp_t gfp
)
559 struct page
*first
, *list
= NULL
;
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 virtio_net_hdr only due to QEMU bug */
587 sg_set_buf(&rq
->sg
[0], p
, sizeof(struct virtio_net_hdr
));
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 receive_queue
*rq
, gfp_t gfp
)
657 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
663 if (vi
->mergeable_rx_bufs
)
664 err
= add_recvbuf_mergeable(rq
, gfp
);
665 else if (vi
->big_packets
)
666 err
= add_recvbuf_big(rq
, gfp
);
668 err
= add_recvbuf_small(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(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(rq
, buf
, len
);
740 if (rq
->vq
->num_free
> virtqueue_get_vring_size(rq
->vq
) / 2) {
741 if (!try_fill_recv(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;
755 received
+= virtnet_receive(rq
, budget
- received
);
757 /* Out of packets? */
758 if (received
< budget
) {
759 r
= virtqueue_enable_cb_prepare(rq
->vq
);
761 if (unlikely(virtqueue_poll(rq
->vq
, r
)) &&
762 napi_schedule_prep(napi
)) {
763 virtqueue_disable_cb(rq
->vq
);
764 __napi_schedule(napi
);
772 #ifdef CONFIG_NET_RX_BUSY_POLL
773 /* must be called with local_bh_disable()d */
774 static int virtnet_busy_poll(struct napi_struct
*napi
)
776 struct receive_queue
*rq
=
777 container_of(napi
, struct receive_queue
, napi
);
778 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
779 int r
, received
= 0, budget
= 4;
781 if (!(vi
->status
& VIRTIO_NET_S_LINK_UP
))
782 return LL_FLUSH_FAILED
;
784 if (!napi_schedule_prep(napi
))
785 return LL_FLUSH_BUSY
;
787 virtqueue_disable_cb(rq
->vq
);
790 received
+= virtnet_receive(rq
, budget
);
792 r
= virtqueue_enable_cb_prepare(rq
->vq
);
793 clear_bit(NAPI_STATE_SCHED
, &napi
->state
);
794 if (unlikely(virtqueue_poll(rq
->vq
, r
)) &&
795 napi_schedule_prep(napi
)) {
796 virtqueue_disable_cb(rq
->vq
);
797 if (received
< budget
) {
801 __napi_schedule(napi
);
807 #endif /* CONFIG_NET_RX_BUSY_POLL */
809 static int virtnet_open(struct net_device
*dev
)
811 struct virtnet_info
*vi
= netdev_priv(dev
);
814 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
815 if (i
< vi
->curr_queue_pairs
)
816 /* Make sure we have some buffers: if oom use wq. */
817 if (!try_fill_recv(&vi
->rq
[i
], GFP_KERNEL
))
818 schedule_delayed_work(&vi
->refill
, 0);
819 virtnet_napi_enable(&vi
->rq
[i
]);
825 static void free_old_xmit_skbs(struct send_queue
*sq
)
829 struct virtnet_info
*vi
= sq
->vq
->vdev
->priv
;
830 struct virtnet_stats
*stats
= this_cpu_ptr(vi
->stats
);
832 while ((skb
= virtqueue_get_buf(sq
->vq
, &len
)) != NULL
) {
833 pr_debug("Sent skb %p\n", skb
);
835 u64_stats_update_begin(&stats
->tx_syncp
);
836 stats
->tx_bytes
+= skb
->len
;
838 u64_stats_update_end(&stats
->tx_syncp
);
840 dev_kfree_skb_any(skb
);
844 static int xmit_skb(struct send_queue
*sq
, struct sk_buff
*skb
)
846 struct skb_vnet_hdr
*hdr
;
847 const unsigned char *dest
= ((struct ethhdr
*)skb
->data
)->h_dest
;
848 struct virtnet_info
*vi
= sq
->vq
->vdev
->priv
;
853 pr_debug("%s: xmit %p %pM\n", vi
->dev
->name
, skb
, dest
);
854 if (vi
->mergeable_rx_bufs
)
855 hdr_len
= sizeof hdr
->mhdr
;
857 hdr_len
= sizeof hdr
->hdr
;
859 can_push
= vi
->any_header_sg
&&
860 !((unsigned long)skb
->data
& (__alignof__(*hdr
) - 1)) &&
861 !skb_header_cloned(skb
) && skb_headroom(skb
) >= hdr_len
;
862 /* Even if we can, don't push here yet as this would skew
863 * csum_start offset below. */
865 hdr
= (struct skb_vnet_hdr
*)(skb
->data
- hdr_len
);
867 hdr
= skb_vnet_hdr(skb
);
869 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
870 hdr
->hdr
.flags
= VIRTIO_NET_HDR_F_NEEDS_CSUM
;
871 hdr
->hdr
.csum_start
= skb_checksum_start_offset(skb
);
872 hdr
->hdr
.csum_offset
= skb
->csum_offset
;
875 hdr
->hdr
.csum_offset
= hdr
->hdr
.csum_start
= 0;
878 if (skb_is_gso(skb
)) {
879 hdr
->hdr
.hdr_len
= skb_headlen(skb
);
880 hdr
->hdr
.gso_size
= skb_shinfo(skb
)->gso_size
;
881 if (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCPV4
)
882 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV4
;
883 else if (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCPV6
)
884 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV6
;
885 else if (skb_shinfo(skb
)->gso_type
& SKB_GSO_UDP
)
886 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_UDP
;
889 if (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCP_ECN
)
890 hdr
->hdr
.gso_type
|= VIRTIO_NET_HDR_GSO_ECN
;
892 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
893 hdr
->hdr
.gso_size
= hdr
->hdr
.hdr_len
= 0;
896 if (vi
->mergeable_rx_bufs
)
897 hdr
->mhdr
.num_buffers
= 0;
900 __skb_push(skb
, hdr_len
);
901 num_sg
= skb_to_sgvec(skb
, sq
->sg
, 0, skb
->len
);
902 /* Pull header back to avoid skew in tx bytes calculations. */
903 __skb_pull(skb
, hdr_len
);
905 sg_set_buf(sq
->sg
, hdr
, hdr_len
);
906 num_sg
= skb_to_sgvec(skb
, sq
->sg
+ 1, 0, skb
->len
) + 1;
908 return virtqueue_add_outbuf(sq
->vq
, sq
->sg
, num_sg
, skb
, GFP_ATOMIC
);
911 static netdev_tx_t
start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
913 struct virtnet_info
*vi
= netdev_priv(dev
);
914 int qnum
= skb_get_queue_mapping(skb
);
915 struct send_queue
*sq
= &vi
->sq
[qnum
];
918 /* Free up any pending old buffers before queueing new ones. */
919 free_old_xmit_skbs(sq
);
921 /* Try to transmit */
922 err
= xmit_skb(sq
, skb
);
924 /* This should not happen! */
926 dev
->stats
.tx_fifo_errors
++;
929 "Unexpected TXQ (%d) queue failure: %d\n", qnum
, err
);
930 dev
->stats
.tx_dropped
++;
931 dev_kfree_skb_any(skb
);
934 virtqueue_kick(sq
->vq
);
936 /* Don't wait up for transmitted skbs to be freed. */
940 /* Apparently nice girls don't return TX_BUSY; stop the queue
941 * before it gets out of hand. Naturally, this wastes entries. */
942 if (sq
->vq
->num_free
< 2+MAX_SKB_FRAGS
) {
943 netif_stop_subqueue(dev
, qnum
);
944 if (unlikely(!virtqueue_enable_cb_delayed(sq
->vq
))) {
945 /* More just got used, free them then recheck. */
946 free_old_xmit_skbs(sq
);
947 if (sq
->vq
->num_free
>= 2+MAX_SKB_FRAGS
) {
948 netif_start_subqueue(dev
, qnum
);
949 virtqueue_disable_cb(sq
->vq
);
958 * Send command via the control virtqueue and check status. Commands
959 * supported by the hypervisor, as indicated by feature bits, should
960 * never fail unless improperly formatted.
962 static bool virtnet_send_command(struct virtnet_info
*vi
, u8
class, u8 cmd
,
963 struct scatterlist
*out
)
965 struct scatterlist
*sgs
[4], hdr
, stat
;
966 struct virtio_net_ctrl_hdr ctrl
;
967 virtio_net_ctrl_ack status
= ~0;
968 unsigned out_num
= 0, tmp
;
970 /* Caller should know better */
971 BUG_ON(!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
));
976 sg_init_one(&hdr
, &ctrl
, sizeof(ctrl
));
977 sgs
[out_num
++] = &hdr
;
980 sgs
[out_num
++] = out
;
982 /* Add return status. */
983 sg_init_one(&stat
, &status
, sizeof(status
));
984 sgs
[out_num
] = &stat
;
986 BUG_ON(out_num
+ 1 > ARRAY_SIZE(sgs
));
987 virtqueue_add_sgs(vi
->cvq
, sgs
, out_num
, 1, vi
, GFP_ATOMIC
);
989 if (unlikely(!virtqueue_kick(vi
->cvq
)))
990 return status
== VIRTIO_NET_OK
;
992 /* Spin for a response, the kick causes an ioport write, trapping
993 * into the hypervisor, so the request should be handled immediately.
995 while (!virtqueue_get_buf(vi
->cvq
, &tmp
) &&
996 !virtqueue_is_broken(vi
->cvq
))
999 return status
== VIRTIO_NET_OK
;
1002 static int virtnet_set_mac_address(struct net_device
*dev
, void *p
)
1004 struct virtnet_info
*vi
= netdev_priv(dev
);
1005 struct virtio_device
*vdev
= vi
->vdev
;
1007 struct sockaddr
*addr
= p
;
1008 struct scatterlist sg
;
1010 ret
= eth_prepare_mac_addr_change(dev
, p
);
1014 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_MAC_ADDR
)) {
1015 sg_init_one(&sg
, addr
->sa_data
, dev
->addr_len
);
1016 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MAC
,
1017 VIRTIO_NET_CTRL_MAC_ADDR_SET
, &sg
)) {
1018 dev_warn(&vdev
->dev
,
1019 "Failed to set mac address by vq command.\n");
1022 } else if (virtio_has_feature(vdev
, VIRTIO_NET_F_MAC
)) {
1025 /* Naturally, this has an atomicity problem. */
1026 for (i
= 0; i
< dev
->addr_len
; i
++)
1027 virtio_cwrite8(vdev
,
1028 offsetof(struct virtio_net_config
, mac
) +
1029 i
, addr
->sa_data
[i
]);
1032 eth_commit_mac_addr_change(dev
, p
);
1037 static struct rtnl_link_stats64
*virtnet_stats(struct net_device
*dev
,
1038 struct rtnl_link_stats64
*tot
)
1040 struct virtnet_info
*vi
= netdev_priv(dev
);
1044 for_each_possible_cpu(cpu
) {
1045 struct virtnet_stats
*stats
= per_cpu_ptr(vi
->stats
, cpu
);
1046 u64 tpackets
, tbytes
, rpackets
, rbytes
;
1049 start
= u64_stats_fetch_begin_irq(&stats
->tx_syncp
);
1050 tpackets
= stats
->tx_packets
;
1051 tbytes
= stats
->tx_bytes
;
1052 } while (u64_stats_fetch_retry_irq(&stats
->tx_syncp
, start
));
1055 start
= u64_stats_fetch_begin_irq(&stats
->rx_syncp
);
1056 rpackets
= stats
->rx_packets
;
1057 rbytes
= stats
->rx_bytes
;
1058 } while (u64_stats_fetch_retry_irq(&stats
->rx_syncp
, start
));
1060 tot
->rx_packets
+= rpackets
;
1061 tot
->tx_packets
+= tpackets
;
1062 tot
->rx_bytes
+= rbytes
;
1063 tot
->tx_bytes
+= tbytes
;
1066 tot
->tx_dropped
= dev
->stats
.tx_dropped
;
1067 tot
->tx_fifo_errors
= dev
->stats
.tx_fifo_errors
;
1068 tot
->rx_dropped
= dev
->stats
.rx_dropped
;
1069 tot
->rx_length_errors
= dev
->stats
.rx_length_errors
;
1070 tot
->rx_frame_errors
= dev
->stats
.rx_frame_errors
;
1075 #ifdef CONFIG_NET_POLL_CONTROLLER
1076 static void virtnet_netpoll(struct net_device
*dev
)
1078 struct virtnet_info
*vi
= netdev_priv(dev
);
1081 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++)
1082 napi_schedule(&vi
->rq
[i
].napi
);
1086 static void virtnet_ack_link_announce(struct virtnet_info
*vi
)
1089 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_ANNOUNCE
,
1090 VIRTIO_NET_CTRL_ANNOUNCE_ACK
, NULL
))
1091 dev_warn(&vi
->dev
->dev
, "Failed to ack link announce.\n");
1095 static int virtnet_set_queues(struct virtnet_info
*vi
, u16 queue_pairs
)
1097 struct scatterlist sg
;
1098 struct virtio_net_ctrl_mq s
;
1099 struct net_device
*dev
= vi
->dev
;
1101 if (!vi
->has_cvq
|| !virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_MQ
))
1104 s
.virtqueue_pairs
= queue_pairs
;
1105 sg_init_one(&sg
, &s
, sizeof(s
));
1107 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MQ
,
1108 VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET
, &sg
)) {
1109 dev_warn(&dev
->dev
, "Fail to set num of queue pairs to %d\n",
1113 vi
->curr_queue_pairs
= queue_pairs
;
1114 /* virtnet_open() will refill when device is going to up. */
1115 if (dev
->flags
& IFF_UP
)
1116 schedule_delayed_work(&vi
->refill
, 0);
1122 static int virtnet_close(struct net_device
*dev
)
1124 struct virtnet_info
*vi
= netdev_priv(dev
);
1127 /* Make sure refill_work doesn't re-enable napi! */
1128 cancel_delayed_work_sync(&vi
->refill
);
1130 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
1131 napi_disable(&vi
->rq
[i
].napi
);
1136 static void virtnet_set_rx_mode(struct net_device
*dev
)
1138 struct virtnet_info
*vi
= netdev_priv(dev
);
1139 struct scatterlist sg
[2];
1140 u8 promisc
, allmulti
;
1141 struct virtio_net_ctrl_mac
*mac_data
;
1142 struct netdev_hw_addr
*ha
;
1148 /* We can't dynamically set ndo_set_rx_mode, so return gracefully */
1149 if (!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_RX
))
1152 promisc
= ((dev
->flags
& IFF_PROMISC
) != 0);
1153 allmulti
= ((dev
->flags
& IFF_ALLMULTI
) != 0);
1155 sg_init_one(sg
, &promisc
, sizeof(promisc
));
1157 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_RX
,
1158 VIRTIO_NET_CTRL_RX_PROMISC
, sg
))
1159 dev_warn(&dev
->dev
, "Failed to %sable promisc mode.\n",
1160 promisc
? "en" : "dis");
1162 sg_init_one(sg
, &allmulti
, sizeof(allmulti
));
1164 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_RX
,
1165 VIRTIO_NET_CTRL_RX_ALLMULTI
, sg
))
1166 dev_warn(&dev
->dev
, "Failed to %sable allmulti mode.\n",
1167 allmulti
? "en" : "dis");
1169 uc_count
= netdev_uc_count(dev
);
1170 mc_count
= netdev_mc_count(dev
);
1171 /* MAC filter - use one buffer for both lists */
1172 buf
= kzalloc(((uc_count
+ mc_count
) * ETH_ALEN
) +
1173 (2 * sizeof(mac_data
->entries
)), GFP_ATOMIC
);
1178 sg_init_table(sg
, 2);
1180 /* Store the unicast list and count in the front of the buffer */
1181 mac_data
->entries
= uc_count
;
1183 netdev_for_each_uc_addr(ha
, dev
)
1184 memcpy(&mac_data
->macs
[i
++][0], ha
->addr
, ETH_ALEN
);
1186 sg_set_buf(&sg
[0], mac_data
,
1187 sizeof(mac_data
->entries
) + (uc_count
* ETH_ALEN
));
1189 /* multicast list and count fill the end */
1190 mac_data
= (void *)&mac_data
->macs
[uc_count
][0];
1192 mac_data
->entries
= mc_count
;
1194 netdev_for_each_mc_addr(ha
, dev
)
1195 memcpy(&mac_data
->macs
[i
++][0], ha
->addr
, ETH_ALEN
);
1197 sg_set_buf(&sg
[1], mac_data
,
1198 sizeof(mac_data
->entries
) + (mc_count
* ETH_ALEN
));
1200 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MAC
,
1201 VIRTIO_NET_CTRL_MAC_TABLE_SET
, sg
))
1202 dev_warn(&dev
->dev
, "Failed to set MAC filter table.\n");
1207 static int virtnet_vlan_rx_add_vid(struct net_device
*dev
,
1208 __be16 proto
, u16 vid
)
1210 struct virtnet_info
*vi
= netdev_priv(dev
);
1211 struct scatterlist sg
;
1213 sg_init_one(&sg
, &vid
, sizeof(vid
));
1215 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_VLAN
,
1216 VIRTIO_NET_CTRL_VLAN_ADD
, &sg
))
1217 dev_warn(&dev
->dev
, "Failed to add VLAN ID %d.\n", vid
);
1221 static int virtnet_vlan_rx_kill_vid(struct net_device
*dev
,
1222 __be16 proto
, u16 vid
)
1224 struct virtnet_info
*vi
= netdev_priv(dev
);
1225 struct scatterlist sg
;
1227 sg_init_one(&sg
, &vid
, sizeof(vid
));
1229 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_VLAN
,
1230 VIRTIO_NET_CTRL_VLAN_DEL
, &sg
))
1231 dev_warn(&dev
->dev
, "Failed to kill VLAN ID %d.\n", vid
);
1235 static void virtnet_clean_affinity(struct virtnet_info
*vi
, long hcpu
)
1239 if (vi
->affinity_hint_set
) {
1240 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1241 virtqueue_set_affinity(vi
->rq
[i
].vq
, -1);
1242 virtqueue_set_affinity(vi
->sq
[i
].vq
, -1);
1245 vi
->affinity_hint_set
= false;
1249 static void virtnet_set_affinity(struct virtnet_info
*vi
)
1254 /* In multiqueue mode, when the number of cpu is equal to the number of
1255 * queue pairs, we let the queue pairs to be private to one cpu by
1256 * setting the affinity hint to eliminate the contention.
1258 if (vi
->curr_queue_pairs
== 1 ||
1259 vi
->max_queue_pairs
!= num_online_cpus()) {
1260 virtnet_clean_affinity(vi
, -1);
1265 for_each_online_cpu(cpu
) {
1266 virtqueue_set_affinity(vi
->rq
[i
].vq
, cpu
);
1267 virtqueue_set_affinity(vi
->sq
[i
].vq
, cpu
);
1268 netif_set_xps_queue(vi
->dev
, cpumask_of(cpu
), i
);
1272 vi
->affinity_hint_set
= true;
1275 static int virtnet_cpu_callback(struct notifier_block
*nfb
,
1276 unsigned long action
, void *hcpu
)
1278 struct virtnet_info
*vi
= container_of(nfb
, struct virtnet_info
, nb
);
1280 switch(action
& ~CPU_TASKS_FROZEN
) {
1282 case CPU_DOWN_FAILED
:
1284 virtnet_set_affinity(vi
);
1286 case CPU_DOWN_PREPARE
:
1287 virtnet_clean_affinity(vi
, (long)hcpu
);
1296 static void virtnet_get_ringparam(struct net_device
*dev
,
1297 struct ethtool_ringparam
*ring
)
1299 struct virtnet_info
*vi
= netdev_priv(dev
);
1301 ring
->rx_max_pending
= virtqueue_get_vring_size(vi
->rq
[0].vq
);
1302 ring
->tx_max_pending
= virtqueue_get_vring_size(vi
->sq
[0].vq
);
1303 ring
->rx_pending
= ring
->rx_max_pending
;
1304 ring
->tx_pending
= ring
->tx_max_pending
;
1308 static void virtnet_get_drvinfo(struct net_device
*dev
,
1309 struct ethtool_drvinfo
*info
)
1311 struct virtnet_info
*vi
= netdev_priv(dev
);
1312 struct virtio_device
*vdev
= vi
->vdev
;
1314 strlcpy(info
->driver
, KBUILD_MODNAME
, sizeof(info
->driver
));
1315 strlcpy(info
->version
, VIRTNET_DRIVER_VERSION
, sizeof(info
->version
));
1316 strlcpy(info
->bus_info
, virtio_bus_name(vdev
), sizeof(info
->bus_info
));
1320 /* TODO: Eliminate OOO packets during switching */
1321 static int virtnet_set_channels(struct net_device
*dev
,
1322 struct ethtool_channels
*channels
)
1324 struct virtnet_info
*vi
= netdev_priv(dev
);
1325 u16 queue_pairs
= channels
->combined_count
;
1328 /* We don't support separate rx/tx channels.
1329 * We don't allow setting 'other' channels.
1331 if (channels
->rx_count
|| channels
->tx_count
|| channels
->other_count
)
1334 if (queue_pairs
> vi
->max_queue_pairs
|| queue_pairs
== 0)
1338 err
= virtnet_set_queues(vi
, queue_pairs
);
1340 netif_set_real_num_tx_queues(dev
, queue_pairs
);
1341 netif_set_real_num_rx_queues(dev
, queue_pairs
);
1343 virtnet_set_affinity(vi
);
1350 static void virtnet_get_channels(struct net_device
*dev
,
1351 struct ethtool_channels
*channels
)
1353 struct virtnet_info
*vi
= netdev_priv(dev
);
1355 channels
->combined_count
= vi
->curr_queue_pairs
;
1356 channels
->max_combined
= vi
->max_queue_pairs
;
1357 channels
->max_other
= 0;
1358 channels
->rx_count
= 0;
1359 channels
->tx_count
= 0;
1360 channels
->other_count
= 0;
1363 static const struct ethtool_ops virtnet_ethtool_ops
= {
1364 .get_drvinfo
= virtnet_get_drvinfo
,
1365 .get_link
= ethtool_op_get_link
,
1366 .get_ringparam
= virtnet_get_ringparam
,
1367 .set_channels
= virtnet_set_channels
,
1368 .get_channels
= virtnet_get_channels
,
1372 #define MAX_MTU 65535
1374 static int virtnet_change_mtu(struct net_device
*dev
, int new_mtu
)
1376 if (new_mtu
< MIN_MTU
|| new_mtu
> MAX_MTU
)
1382 static const struct net_device_ops virtnet_netdev
= {
1383 .ndo_open
= virtnet_open
,
1384 .ndo_stop
= virtnet_close
,
1385 .ndo_start_xmit
= start_xmit
,
1386 .ndo_validate_addr
= eth_validate_addr
,
1387 .ndo_set_mac_address
= virtnet_set_mac_address
,
1388 .ndo_set_rx_mode
= virtnet_set_rx_mode
,
1389 .ndo_change_mtu
= virtnet_change_mtu
,
1390 .ndo_get_stats64
= virtnet_stats
,
1391 .ndo_vlan_rx_add_vid
= virtnet_vlan_rx_add_vid
,
1392 .ndo_vlan_rx_kill_vid
= virtnet_vlan_rx_kill_vid
,
1393 #ifdef CONFIG_NET_POLL_CONTROLLER
1394 .ndo_poll_controller
= virtnet_netpoll
,
1396 #ifdef CONFIG_NET_RX_BUSY_POLL
1397 .ndo_busy_poll
= virtnet_busy_poll
,
1401 static void virtnet_config_changed_work(struct work_struct
*work
)
1403 struct virtnet_info
*vi
=
1404 container_of(work
, struct virtnet_info
, config_work
);
1407 mutex_lock(&vi
->config_lock
);
1408 if (virtio_cread_feature(vi
->vdev
, VIRTIO_NET_F_STATUS
,
1409 struct virtio_net_config
, status
, &v
) < 0)
1412 if (v
& VIRTIO_NET_S_ANNOUNCE
) {
1413 netdev_notify_peers(vi
->dev
);
1414 virtnet_ack_link_announce(vi
);
1417 /* Ignore unknown (future) status bits */
1418 v
&= VIRTIO_NET_S_LINK_UP
;
1420 if (vi
->status
== v
)
1425 if (vi
->status
& VIRTIO_NET_S_LINK_UP
) {
1426 netif_carrier_on(vi
->dev
);
1427 netif_tx_wake_all_queues(vi
->dev
);
1429 netif_carrier_off(vi
->dev
);
1430 netif_tx_stop_all_queues(vi
->dev
);
1433 mutex_unlock(&vi
->config_lock
);
1436 static void virtnet_config_changed(struct virtio_device
*vdev
)
1438 struct virtnet_info
*vi
= vdev
->priv
;
1440 schedule_work(&vi
->config_work
);
1443 static void virtnet_free_queues(struct virtnet_info
*vi
)
1447 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
1448 netif_napi_del(&vi
->rq
[i
].napi
);
1454 static void free_receive_bufs(struct virtnet_info
*vi
)
1458 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1459 while (vi
->rq
[i
].pages
)
1460 __free_pages(get_a_page(&vi
->rq
[i
], GFP_KERNEL
), 0);
1464 static void free_receive_page_frags(struct virtnet_info
*vi
)
1467 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
1468 if (vi
->rq
[i
].alloc_frag
.page
)
1469 put_page(vi
->rq
[i
].alloc_frag
.page
);
1472 static void free_unused_bufs(struct virtnet_info
*vi
)
1477 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1478 struct virtqueue
*vq
= vi
->sq
[i
].vq
;
1479 while ((buf
= virtqueue_detach_unused_buf(vq
)) != NULL
)
1483 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1484 struct virtqueue
*vq
= vi
->rq
[i
].vq
;
1486 while ((buf
= virtqueue_detach_unused_buf(vq
)) != NULL
) {
1487 if (vi
->mergeable_rx_bufs
) {
1488 unsigned long ctx
= (unsigned long)buf
;
1489 void *base
= mergeable_ctx_to_buf_address(ctx
);
1490 put_page(virt_to_head_page(base
));
1491 } else if (vi
->big_packets
) {
1492 give_pages(&vi
->rq
[i
], buf
);
1500 static void virtnet_del_vqs(struct virtnet_info
*vi
)
1502 struct virtio_device
*vdev
= vi
->vdev
;
1504 virtnet_clean_affinity(vi
, -1);
1506 vdev
->config
->del_vqs(vdev
);
1508 virtnet_free_queues(vi
);
1511 static int virtnet_find_vqs(struct virtnet_info
*vi
)
1513 vq_callback_t
**callbacks
;
1514 struct virtqueue
**vqs
;
1519 /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
1520 * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
1521 * possible control vq.
1523 total_vqs
= vi
->max_queue_pairs
* 2 +
1524 virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
);
1526 /* Allocate space for find_vqs parameters */
1527 vqs
= kzalloc(total_vqs
* sizeof(*vqs
), GFP_KERNEL
);
1530 callbacks
= kmalloc(total_vqs
* sizeof(*callbacks
), GFP_KERNEL
);
1533 names
= kmalloc(total_vqs
* sizeof(*names
), GFP_KERNEL
);
1537 /* Parameters for control virtqueue, if any */
1539 callbacks
[total_vqs
- 1] = NULL
;
1540 names
[total_vqs
- 1] = "control";
1543 /* Allocate/initialize parameters for send/receive virtqueues */
1544 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1545 callbacks
[rxq2vq(i
)] = skb_recv_done
;
1546 callbacks
[txq2vq(i
)] = skb_xmit_done
;
1547 sprintf(vi
->rq
[i
].name
, "input.%d", i
);
1548 sprintf(vi
->sq
[i
].name
, "output.%d", i
);
1549 names
[rxq2vq(i
)] = vi
->rq
[i
].name
;
1550 names
[txq2vq(i
)] = vi
->sq
[i
].name
;
1553 ret
= vi
->vdev
->config
->find_vqs(vi
->vdev
, total_vqs
, vqs
, callbacks
,
1559 vi
->cvq
= vqs
[total_vqs
- 1];
1560 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VLAN
))
1561 vi
->dev
->features
|= NETIF_F_HW_VLAN_CTAG_FILTER
;
1564 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1565 vi
->rq
[i
].vq
= vqs
[rxq2vq(i
)];
1566 vi
->sq
[i
].vq
= vqs
[txq2vq(i
)];
1585 static int virtnet_alloc_queues(struct virtnet_info
*vi
)
1589 vi
->sq
= kzalloc(sizeof(*vi
->sq
) * vi
->max_queue_pairs
, GFP_KERNEL
);
1592 vi
->rq
= kzalloc(sizeof(*vi
->rq
) * vi
->max_queue_pairs
, GFP_KERNEL
);
1596 INIT_DELAYED_WORK(&vi
->refill
, refill_work
);
1597 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1598 vi
->rq
[i
].pages
= NULL
;
1599 netif_napi_add(vi
->dev
, &vi
->rq
[i
].napi
, virtnet_poll
,
1601 napi_hash_add(&vi
->rq
[i
].napi
);
1603 sg_init_table(vi
->rq
[i
].sg
, ARRAY_SIZE(vi
->rq
[i
].sg
));
1604 ewma_init(&vi
->rq
[i
].mrg_avg_pkt_len
, 1, RECEIVE_AVG_WEIGHT
);
1605 sg_init_table(vi
->sq
[i
].sg
, ARRAY_SIZE(vi
->sq
[i
].sg
));
1616 static int init_vqs(struct virtnet_info
*vi
)
1620 /* Allocate send & receive queues */
1621 ret
= virtnet_alloc_queues(vi
);
1625 ret
= virtnet_find_vqs(vi
);
1630 virtnet_set_affinity(vi
);
1636 virtnet_free_queues(vi
);
1642 static ssize_t
mergeable_rx_buffer_size_show(struct netdev_rx_queue
*queue
,
1643 struct rx_queue_attribute
*attribute
, char *buf
)
1645 struct virtnet_info
*vi
= netdev_priv(queue
->dev
);
1646 unsigned int queue_index
= get_netdev_rx_queue_index(queue
);
1649 BUG_ON(queue_index
>= vi
->max_queue_pairs
);
1650 avg
= &vi
->rq
[queue_index
].mrg_avg_pkt_len
;
1651 return sprintf(buf
, "%u\n", get_mergeable_buf_len(avg
));
1654 static struct rx_queue_attribute mergeable_rx_buffer_size_attribute
=
1655 __ATTR_RO(mergeable_rx_buffer_size
);
1657 static struct attribute
*virtio_net_mrg_rx_attrs
[] = {
1658 &mergeable_rx_buffer_size_attribute
.attr
,
1662 static const struct attribute_group virtio_net_mrg_rx_group
= {
1663 .name
= "virtio_net",
1664 .attrs
= virtio_net_mrg_rx_attrs
1668 static int virtnet_probe(struct virtio_device
*vdev
)
1671 struct net_device
*dev
;
1672 struct virtnet_info
*vi
;
1673 u16 max_queue_pairs
;
1675 /* Find if host supports multiqueue virtio_net device */
1676 err
= virtio_cread_feature(vdev
, VIRTIO_NET_F_MQ
,
1677 struct virtio_net_config
,
1678 max_virtqueue_pairs
, &max_queue_pairs
);
1680 /* We need at least 2 queue's */
1681 if (err
|| max_queue_pairs
< VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN
||
1682 max_queue_pairs
> VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX
||
1683 !virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
))
1684 max_queue_pairs
= 1;
1686 /* Allocate ourselves a network device with room for our info */
1687 dev
= alloc_etherdev_mq(sizeof(struct virtnet_info
), max_queue_pairs
);
1691 /* Set up network device as normal. */
1692 dev
->priv_flags
|= IFF_UNICAST_FLT
| IFF_LIVE_ADDR_CHANGE
;
1693 dev
->netdev_ops
= &virtnet_netdev
;
1694 dev
->features
= NETIF_F_HIGHDMA
;
1696 dev
->ethtool_ops
= &virtnet_ethtool_ops
;
1697 SET_NETDEV_DEV(dev
, &vdev
->dev
);
1699 /* Do we support "hardware" checksums? */
1700 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CSUM
)) {
1701 /* This opens up the world of extra features. */
1702 dev
->hw_features
|= NETIF_F_HW_CSUM
|NETIF_F_SG
|NETIF_F_FRAGLIST
;
1704 dev
->features
|= NETIF_F_HW_CSUM
|NETIF_F_SG
|NETIF_F_FRAGLIST
;
1706 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GSO
)) {
1707 dev
->hw_features
|= NETIF_F_TSO
| NETIF_F_UFO
1708 | NETIF_F_TSO_ECN
| NETIF_F_TSO6
;
1710 /* Individual feature bits: what can host handle? */
1711 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_TSO4
))
1712 dev
->hw_features
|= NETIF_F_TSO
;
1713 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_TSO6
))
1714 dev
->hw_features
|= NETIF_F_TSO6
;
1715 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_ECN
))
1716 dev
->hw_features
|= NETIF_F_TSO_ECN
;
1717 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_UFO
))
1718 dev
->hw_features
|= NETIF_F_UFO
;
1721 dev
->features
|= dev
->hw_features
& (NETIF_F_ALL_TSO
|NETIF_F_UFO
);
1722 /* (!csum && gso) case will be fixed by register_netdev() */
1724 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_CSUM
))
1725 dev
->features
|= NETIF_F_RXCSUM
;
1727 dev
->vlan_features
= dev
->features
;
1729 /* Configuration may specify what MAC to use. Otherwise random. */
1730 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MAC
))
1731 virtio_cread_bytes(vdev
,
1732 offsetof(struct virtio_net_config
, mac
),
1733 dev
->dev_addr
, dev
->addr_len
);
1735 eth_hw_addr_random(dev
);
1737 /* Set up our device-specific information */
1738 vi
= netdev_priv(dev
);
1742 vi
->stats
= alloc_percpu(struct virtnet_stats
);
1744 if (vi
->stats
== NULL
)
1747 for_each_possible_cpu(i
) {
1748 struct virtnet_stats
*virtnet_stats
;
1749 virtnet_stats
= per_cpu_ptr(vi
->stats
, i
);
1750 u64_stats_init(&virtnet_stats
->tx_syncp
);
1751 u64_stats_init(&virtnet_stats
->rx_syncp
);
1754 mutex_init(&vi
->config_lock
);
1755 INIT_WORK(&vi
->config_work
, virtnet_config_changed_work
);
1757 /* If we can receive ANY GSO packets, we must allocate large ones. */
1758 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_TSO4
) ||
1759 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_TSO6
) ||
1760 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_ECN
) ||
1761 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_UFO
))
1762 vi
->big_packets
= true;
1764 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MRG_RXBUF
))
1765 vi
->mergeable_rx_bufs
= true;
1767 if (virtio_has_feature(vdev
, VIRTIO_F_ANY_LAYOUT
))
1768 vi
->any_header_sg
= true;
1770 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
))
1773 if (vi
->any_header_sg
) {
1774 if (vi
->mergeable_rx_bufs
)
1775 dev
->needed_headroom
= sizeof(struct virtio_net_hdr_mrg_rxbuf
);
1777 dev
->needed_headroom
= sizeof(struct virtio_net_hdr
);
1780 /* Use single tx/rx queue pair as default */
1781 vi
->curr_queue_pairs
= 1;
1782 vi
->max_queue_pairs
= max_queue_pairs
;
1784 /* Allocate/initialize the rx/tx queues, and invoke find_vqs */
1790 if (vi
->mergeable_rx_bufs
)
1791 dev
->sysfs_rx_queue_group
= &virtio_net_mrg_rx_group
;
1793 netif_set_real_num_tx_queues(dev
, vi
->curr_queue_pairs
);
1794 netif_set_real_num_rx_queues(dev
, vi
->curr_queue_pairs
);
1796 err
= register_netdev(dev
);
1798 pr_debug("virtio_net: registering device failed\n");
1802 /* Last of all, set up some receive buffers. */
1803 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++) {
1804 try_fill_recv(&vi
->rq
[i
], GFP_KERNEL
);
1806 /* If we didn't even get one input buffer, we're useless. */
1807 if (vi
->rq
[i
].vq
->num_free
==
1808 virtqueue_get_vring_size(vi
->rq
[i
].vq
)) {
1809 free_unused_bufs(vi
);
1811 goto free_recv_bufs
;
1815 vi
->nb
.notifier_call
= &virtnet_cpu_callback
;
1816 err
= register_hotcpu_notifier(&vi
->nb
);
1818 pr_debug("virtio_net: registering cpu notifier failed\n");
1819 goto free_recv_bufs
;
1822 /* Assume link up if device can't report link status,
1823 otherwise get link status from config. */
1824 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_STATUS
)) {
1825 netif_carrier_off(dev
);
1826 schedule_work(&vi
->config_work
);
1828 vi
->status
= VIRTIO_NET_S_LINK_UP
;
1829 netif_carrier_on(dev
);
1832 pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
1833 dev
->name
, max_queue_pairs
);
1838 free_receive_bufs(vi
);
1839 unregister_netdev(dev
);
1841 cancel_delayed_work_sync(&vi
->refill
);
1842 free_receive_page_frags(vi
);
1843 virtnet_del_vqs(vi
);
1845 free_percpu(vi
->stats
);
1851 static void remove_vq_common(struct virtnet_info
*vi
)
1853 vi
->vdev
->config
->reset(vi
->vdev
);
1855 /* Free unused buffers in both send and recv, if any. */
1856 free_unused_bufs(vi
);
1858 free_receive_bufs(vi
);
1860 free_receive_page_frags(vi
);
1862 virtnet_del_vqs(vi
);
1865 static void virtnet_remove(struct virtio_device
*vdev
)
1867 struct virtnet_info
*vi
= vdev
->priv
;
1869 unregister_hotcpu_notifier(&vi
->nb
);
1871 /* Make sure no work handler is accessing the device. */
1872 flush_work(&vi
->config_work
);
1874 unregister_netdev(vi
->dev
);
1876 remove_vq_common(vi
);
1878 free_percpu(vi
->stats
);
1879 free_netdev(vi
->dev
);
1882 #ifdef CONFIG_PM_SLEEP
1883 static int virtnet_freeze(struct virtio_device
*vdev
)
1885 struct virtnet_info
*vi
= vdev
->priv
;
1888 unregister_hotcpu_notifier(&vi
->nb
);
1890 /* Make sure no work handler is accessing the device */
1891 flush_work(&vi
->config_work
);
1893 netif_device_detach(vi
->dev
);
1894 cancel_delayed_work_sync(&vi
->refill
);
1896 if (netif_running(vi
->dev
)) {
1897 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1898 napi_disable(&vi
->rq
[i
].napi
);
1899 napi_hash_del(&vi
->rq
[i
].napi
);
1900 netif_napi_del(&vi
->rq
[i
].napi
);
1904 remove_vq_common(vi
);
1909 static int virtnet_restore(struct virtio_device
*vdev
)
1911 struct virtnet_info
*vi
= vdev
->priv
;
1918 if (netif_running(vi
->dev
)) {
1919 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++)
1920 if (!try_fill_recv(&vi
->rq
[i
], GFP_KERNEL
))
1921 schedule_delayed_work(&vi
->refill
, 0);
1923 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
1924 virtnet_napi_enable(&vi
->rq
[i
]);
1927 netif_device_attach(vi
->dev
);
1930 virtnet_set_queues(vi
, vi
->curr_queue_pairs
);
1933 err
= register_hotcpu_notifier(&vi
->nb
);
1941 static struct virtio_device_id id_table
[] = {
1942 { VIRTIO_ID_NET
, VIRTIO_DEV_ANY_ID
},
1946 static unsigned int features
[] = {
1947 VIRTIO_NET_F_CSUM
, VIRTIO_NET_F_GUEST_CSUM
,
1948 VIRTIO_NET_F_GSO
, VIRTIO_NET_F_MAC
,
1949 VIRTIO_NET_F_HOST_TSO4
, VIRTIO_NET_F_HOST_UFO
, VIRTIO_NET_F_HOST_TSO6
,
1950 VIRTIO_NET_F_HOST_ECN
, VIRTIO_NET_F_GUEST_TSO4
, VIRTIO_NET_F_GUEST_TSO6
,
1951 VIRTIO_NET_F_GUEST_ECN
, VIRTIO_NET_F_GUEST_UFO
,
1952 VIRTIO_NET_F_MRG_RXBUF
, VIRTIO_NET_F_STATUS
, VIRTIO_NET_F_CTRL_VQ
,
1953 VIRTIO_NET_F_CTRL_RX
, VIRTIO_NET_F_CTRL_VLAN
,
1954 VIRTIO_NET_F_GUEST_ANNOUNCE
, VIRTIO_NET_F_MQ
,
1955 VIRTIO_NET_F_CTRL_MAC_ADDR
,
1956 VIRTIO_F_ANY_LAYOUT
,
1959 static struct virtio_driver virtio_net_driver
= {
1960 .feature_table
= features
,
1961 .feature_table_size
= ARRAY_SIZE(features
),
1962 .driver
.name
= KBUILD_MODNAME
,
1963 .driver
.owner
= THIS_MODULE
,
1964 .id_table
= id_table
,
1965 .probe
= virtnet_probe
,
1966 .remove
= virtnet_remove
,
1967 .config_changed
= virtnet_config_changed
,
1968 #ifdef CONFIG_PM_SLEEP
1969 .freeze
= virtnet_freeze
,
1970 .restore
= virtnet_restore
,
1974 module_virtio_driver(virtio_net_driver
);
1976 MODULE_DEVICE_TABLE(virtio
, id_table
);
1977 MODULE_DESCRIPTION("Virtio network driver");
1978 MODULE_LICENSE("GPL");