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, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 #include <linux/netdevice.h>
21 #include <linux/etherdevice.h>
22 #include <linux/ethtool.h>
23 #include <linux/module.h>
24 #include <linux/virtio.h>
25 #include <linux/virtio_net.h>
26 #include <linux/scatterlist.h>
27 #include <linux/if_vlan.h>
28 #include <linux/slab.h>
30 static int napi_weight
= 128;
31 module_param(napi_weight
, int, 0444);
33 static bool csum
= true, gso
= true;
34 module_param(csum
, bool, 0444);
35 module_param(gso
, bool, 0444);
37 /* FIXME: MTU in config. */
38 #define MAX_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
39 #define GOOD_COPY_LEN 128
41 #define VIRTNET_SEND_COMMAND_SG_MAX 2
42 #define VIRTNET_DRIVER_VERSION "1.0.0"
44 struct virtnet_stats
{
45 struct u64_stats_sync tx_syncp
;
46 struct u64_stats_sync rx_syncp
;
54 /* Internal representation of a send virtqueue */
56 /* Virtqueue associated with this send _queue */
59 /* TX: fragments + linear part + virtio header */
60 struct scatterlist sg
[MAX_SKB_FRAGS
+ 2];
62 /* Name of the send queue: output.$index */
66 /* Internal representation of a receive virtqueue */
67 struct receive_queue
{
68 /* Virtqueue associated with this receive_queue */
71 struct napi_struct napi
;
73 /* Number of input buffers, and max we've ever had. */
74 unsigned int num
, max
;
76 /* Chain pages by the private ptr. */
79 /* RX: fragments + linear part + virtio header */
80 struct scatterlist sg
[MAX_SKB_FRAGS
+ 2];
82 /* Name of this receive queue: input.$index */
87 struct virtio_device
*vdev
;
88 struct virtqueue
*cvq
;
89 struct net_device
*dev
;
90 struct send_queue
*sq
;
91 struct receive_queue
*rq
;
94 /* Max # of queue pairs supported by the device */
97 /* # of queue pairs currently used by the driver */
100 /* I like... big packets and I cannot lie! */
103 /* Host will merge rx buffers for big packets (shake it! shake it!) */
104 bool mergeable_rx_bufs
;
106 /* Has control virtqueue */
109 /* enable config space updates */
112 /* Active statistics */
113 struct virtnet_stats __percpu
*stats
;
115 /* Work struct for refilling if we run low on memory. */
116 struct delayed_work refill
;
118 /* Work struct for config space updates */
119 struct work_struct config_work
;
121 /* Lock for config space updates */
122 struct mutex config_lock
;
124 /* Does the affinity hint is set for virtqueues? */
125 bool affinity_hint_set
;
128 struct skb_vnet_hdr
{
130 struct virtio_net_hdr hdr
;
131 struct virtio_net_hdr_mrg_rxbuf mhdr
;
135 struct padded_vnet_hdr
{
136 struct virtio_net_hdr hdr
;
138 * virtio_net_hdr should be in a separated sg buffer because of a
139 * QEMU bug, and data sg buffer shares same page with this header sg.
140 * This padding makes next sg 16 byte aligned after virtio_net_hdr.
145 /* Converting between virtqueue no. and kernel tx/rx queue no.
146 * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq
148 static int vq2txq(struct virtqueue
*vq
)
150 return (virtqueue_get_queue_index(vq
) - 1) / 2;
153 static int txq2vq(int txq
)
158 static int vq2rxq(struct virtqueue
*vq
)
160 return virtqueue_get_queue_index(vq
) / 2;
163 static int rxq2vq(int rxq
)
168 static inline struct skb_vnet_hdr
*skb_vnet_hdr(struct sk_buff
*skb
)
170 return (struct skb_vnet_hdr
*)skb
->cb
;
174 * private is used to chain pages for big packets, put the whole
175 * most recent used list in the beginning for reuse
177 static void give_pages(struct receive_queue
*rq
, struct page
*page
)
181 /* Find end of list, sew whole thing into vi->rq.pages. */
182 for (end
= page
; end
->private; end
= (struct page
*)end
->private);
183 end
->private = (unsigned long)rq
->pages
;
187 static struct page
*get_a_page(struct receive_queue
*rq
, gfp_t gfp_mask
)
189 struct page
*p
= rq
->pages
;
192 rq
->pages
= (struct page
*)p
->private;
193 /* clear private here, it is used to chain pages */
196 p
= alloc_page(gfp_mask
);
200 static void skb_xmit_done(struct virtqueue
*vq
)
202 struct virtnet_info
*vi
= vq
->vdev
->priv
;
204 /* Suppress further interrupts. */
205 virtqueue_disable_cb(vq
);
207 /* We were probably waiting for more output buffers. */
208 netif_wake_subqueue(vi
->dev
, vq2txq(vq
));
211 static void set_skb_frag(struct sk_buff
*skb
, struct page
*page
,
212 unsigned int offset
, unsigned int *len
)
214 int size
= min((unsigned)PAGE_SIZE
- offset
, *len
);
215 int i
= skb_shinfo(skb
)->nr_frags
;
217 __skb_fill_page_desc(skb
, i
, page
, offset
, size
);
219 skb
->data_len
+= size
;
221 skb
->truesize
+= PAGE_SIZE
;
222 skb_shinfo(skb
)->nr_frags
++;
226 /* Called from bottom half context */
227 static struct sk_buff
*page_to_skb(struct receive_queue
*rq
,
228 struct page
*page
, unsigned int len
)
230 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
232 struct skb_vnet_hdr
*hdr
;
233 unsigned int copy
, hdr_len
, offset
;
236 p
= page_address(page
);
238 /* copy small packet so we can reuse these pages for small data */
239 skb
= netdev_alloc_skb_ip_align(vi
->dev
, GOOD_COPY_LEN
);
243 hdr
= skb_vnet_hdr(skb
);
245 if (vi
->mergeable_rx_bufs
) {
246 hdr_len
= sizeof hdr
->mhdr
;
249 hdr_len
= sizeof hdr
->hdr
;
250 offset
= sizeof(struct padded_vnet_hdr
);
253 memcpy(hdr
, p
, hdr_len
);
259 if (copy
> skb_tailroom(skb
))
260 copy
= skb_tailroom(skb
);
261 memcpy(skb_put(skb
, copy
), p
, copy
);
267 * Verify that we can indeed put this data into a skb.
268 * This is here to handle cases when the device erroneously
269 * tries to receive more than is possible. This is usually
270 * the case of a broken device.
272 if (unlikely(len
> MAX_SKB_FRAGS
* PAGE_SIZE
)) {
273 net_dbg_ratelimited("%s: too much data\n", skb
->dev
->name
);
279 set_skb_frag(skb
, page
, offset
, &len
);
280 page
= (struct page
*)page
->private;
285 give_pages(rq
, page
);
290 static int receive_mergeable(struct receive_queue
*rq
, struct sk_buff
*skb
)
292 struct skb_vnet_hdr
*hdr
= skb_vnet_hdr(skb
);
296 num_buf
= hdr
->mhdr
.num_buffers
;
298 i
= skb_shinfo(skb
)->nr_frags
;
299 if (i
>= MAX_SKB_FRAGS
) {
300 pr_debug("%s: packet too long\n", skb
->dev
->name
);
301 skb
->dev
->stats
.rx_length_errors
++;
304 page
= virtqueue_get_buf(rq
->vq
, &len
);
306 pr_debug("%s: rx error: %d buffers missing\n",
307 skb
->dev
->name
, hdr
->mhdr
.num_buffers
);
308 skb
->dev
->stats
.rx_length_errors
++;
315 set_skb_frag(skb
, page
, 0, &len
);
322 static void receive_buf(struct receive_queue
*rq
, void *buf
, unsigned int len
)
324 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
325 struct net_device
*dev
= vi
->dev
;
326 struct virtnet_stats
*stats
= this_cpu_ptr(vi
->stats
);
329 struct skb_vnet_hdr
*hdr
;
331 if (unlikely(len
< sizeof(struct virtio_net_hdr
) + ETH_HLEN
)) {
332 pr_debug("%s: short packet %i\n", dev
->name
, len
);
333 dev
->stats
.rx_length_errors
++;
334 if (vi
->mergeable_rx_bufs
|| vi
->big_packets
)
341 if (!vi
->mergeable_rx_bufs
&& !vi
->big_packets
) {
343 len
-= sizeof(struct virtio_net_hdr
);
347 skb
= page_to_skb(rq
, page
, len
);
348 if (unlikely(!skb
)) {
349 dev
->stats
.rx_dropped
++;
350 give_pages(rq
, page
);
353 if (vi
->mergeable_rx_bufs
)
354 if (receive_mergeable(rq
, skb
)) {
360 hdr
= skb_vnet_hdr(skb
);
362 u64_stats_update_begin(&stats
->rx_syncp
);
363 stats
->rx_bytes
+= skb
->len
;
365 u64_stats_update_end(&stats
->rx_syncp
);
367 if (hdr
->hdr
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) {
368 pr_debug("Needs csum!\n");
369 if (!skb_partial_csum_set(skb
,
371 hdr
->hdr
.csum_offset
))
373 } else if (hdr
->hdr
.flags
& VIRTIO_NET_HDR_F_DATA_VALID
) {
374 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
377 skb
->protocol
= eth_type_trans(skb
, dev
);
378 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
379 ntohs(skb
->protocol
), skb
->len
, skb
->pkt_type
);
381 if (hdr
->hdr
.gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
383 switch (hdr
->hdr
.gso_type
& ~VIRTIO_NET_HDR_GSO_ECN
) {
384 case VIRTIO_NET_HDR_GSO_TCPV4
:
385 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV4
;
387 case VIRTIO_NET_HDR_GSO_UDP
:
388 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
390 case VIRTIO_NET_HDR_GSO_TCPV6
:
391 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV6
;
394 net_warn_ratelimited("%s: bad gso type %u.\n",
395 dev
->name
, hdr
->hdr
.gso_type
);
399 if (hdr
->hdr
.gso_type
& VIRTIO_NET_HDR_GSO_ECN
)
400 skb_shinfo(skb
)->gso_type
|= SKB_GSO_TCP_ECN
;
402 skb_shinfo(skb
)->gso_size
= hdr
->hdr
.gso_size
;
403 if (skb_shinfo(skb
)->gso_size
== 0) {
404 net_warn_ratelimited("%s: zero gso size.\n", dev
->name
);
408 /* Header must be checked, and gso_segs computed. */
409 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
410 skb_shinfo(skb
)->gso_segs
= 0;
413 netif_receive_skb(skb
);
417 dev
->stats
.rx_frame_errors
++;
421 static int add_recvbuf_small(struct receive_queue
*rq
, gfp_t gfp
)
423 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
425 struct skb_vnet_hdr
*hdr
;
428 skb
= __netdev_alloc_skb_ip_align(vi
->dev
, MAX_PACKET_LEN
, gfp
);
432 skb_put(skb
, MAX_PACKET_LEN
);
434 hdr
= skb_vnet_hdr(skb
);
435 sg_set_buf(rq
->sg
, &hdr
->hdr
, sizeof hdr
->hdr
);
437 skb_to_sgvec(skb
, rq
->sg
+ 1, 0, skb
->len
);
439 err
= virtqueue_add_buf(rq
->vq
, rq
->sg
, 0, 2, skb
, gfp
);
446 static int add_recvbuf_big(struct receive_queue
*rq
, gfp_t gfp
)
448 struct page
*first
, *list
= NULL
;
452 /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */
453 for (i
= MAX_SKB_FRAGS
+ 1; i
> 1; --i
) {
454 first
= get_a_page(rq
, gfp
);
457 give_pages(rq
, list
);
460 sg_set_buf(&rq
->sg
[i
], page_address(first
), PAGE_SIZE
);
462 /* chain new page in list head to match sg */
463 first
->private = (unsigned long)list
;
467 first
= get_a_page(rq
, gfp
);
469 give_pages(rq
, list
);
472 p
= page_address(first
);
474 /* rq->sg[0], rq->sg[1] share the same page */
475 /* a separated rq->sg[0] for virtio_net_hdr only due to QEMU bug */
476 sg_set_buf(&rq
->sg
[0], p
, sizeof(struct virtio_net_hdr
));
478 /* rq->sg[1] for data packet, from offset */
479 offset
= sizeof(struct padded_vnet_hdr
);
480 sg_set_buf(&rq
->sg
[1], p
+ offset
, PAGE_SIZE
- offset
);
482 /* chain first in list head */
483 first
->private = (unsigned long)list
;
484 err
= virtqueue_add_buf(rq
->vq
, rq
->sg
, 0, MAX_SKB_FRAGS
+ 2,
487 give_pages(rq
, first
);
492 static int add_recvbuf_mergeable(struct receive_queue
*rq
, gfp_t gfp
)
497 page
= get_a_page(rq
, gfp
);
501 sg_init_one(rq
->sg
, page_address(page
), PAGE_SIZE
);
503 err
= virtqueue_add_buf(rq
->vq
, rq
->sg
, 0, 1, page
, gfp
);
505 give_pages(rq
, page
);
511 * Returns false if we couldn't fill entirely (OOM).
513 * Normally run in the receive path, but can also be run from ndo_open
514 * before we're receiving packets, or from refill_work which is
515 * careful to disable receiving (using napi_disable).
517 static bool try_fill_recv(struct receive_queue
*rq
, gfp_t gfp
)
519 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
524 if (vi
->mergeable_rx_bufs
)
525 err
= add_recvbuf_mergeable(rq
, gfp
);
526 else if (vi
->big_packets
)
527 err
= add_recvbuf_big(rq
, gfp
);
529 err
= add_recvbuf_small(rq
, gfp
);
531 oom
= err
== -ENOMEM
;
535 } while (rq
->vq
->num_free
);
536 if (unlikely(rq
->num
> rq
->max
))
538 virtqueue_kick(rq
->vq
);
542 static void skb_recv_done(struct virtqueue
*rvq
)
544 struct virtnet_info
*vi
= rvq
->vdev
->priv
;
545 struct receive_queue
*rq
= &vi
->rq
[vq2rxq(rvq
)];
547 /* Schedule NAPI, Suppress further interrupts if successful. */
548 if (napi_schedule_prep(&rq
->napi
)) {
549 virtqueue_disable_cb(rvq
);
550 __napi_schedule(&rq
->napi
);
554 static void virtnet_napi_enable(struct receive_queue
*rq
)
556 napi_enable(&rq
->napi
);
558 /* If all buffers were filled by other side before we napi_enabled, we
559 * won't get another interrupt, so process any outstanding packets
560 * now. virtnet_poll wants re-enable the queue, so we disable here.
561 * We synchronize against interrupts via NAPI_STATE_SCHED */
562 if (napi_schedule_prep(&rq
->napi
)) {
563 virtqueue_disable_cb(rq
->vq
);
565 __napi_schedule(&rq
->napi
);
570 static void refill_work(struct work_struct
*work
)
572 struct virtnet_info
*vi
=
573 container_of(work
, struct virtnet_info
, refill
.work
);
577 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
578 struct receive_queue
*rq
= &vi
->rq
[i
];
580 napi_disable(&rq
->napi
);
581 still_empty
= !try_fill_recv(rq
, GFP_KERNEL
);
582 virtnet_napi_enable(rq
);
584 /* In theory, this can happen: if we don't get any buffers in
585 * we will *never* try to fill again.
588 schedule_delayed_work(&vi
->refill
, HZ
/2);
592 static int virtnet_poll(struct napi_struct
*napi
, int budget
)
594 struct receive_queue
*rq
=
595 container_of(napi
, struct receive_queue
, napi
);
596 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
598 unsigned int len
, received
= 0;
601 while (received
< budget
&&
602 (buf
= virtqueue_get_buf(rq
->vq
, &len
)) != NULL
) {
603 receive_buf(rq
, buf
, len
);
608 if (rq
->num
< rq
->max
/ 2) {
609 if (!try_fill_recv(rq
, GFP_ATOMIC
))
610 schedule_delayed_work(&vi
->refill
, 0);
613 /* Out of packets? */
614 if (received
< budget
) {
616 if (unlikely(!virtqueue_enable_cb(rq
->vq
)) &&
617 napi_schedule_prep(napi
)) {
618 virtqueue_disable_cb(rq
->vq
);
619 __napi_schedule(napi
);
627 static int virtnet_open(struct net_device
*dev
)
629 struct virtnet_info
*vi
= netdev_priv(dev
);
632 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
633 /* Make sure we have some buffers: if oom use wq. */
634 if (!try_fill_recv(&vi
->rq
[i
], GFP_KERNEL
))
635 schedule_delayed_work(&vi
->refill
, 0);
636 virtnet_napi_enable(&vi
->rq
[i
]);
642 static void free_old_xmit_skbs(struct send_queue
*sq
)
646 struct virtnet_info
*vi
= sq
->vq
->vdev
->priv
;
647 struct virtnet_stats
*stats
= this_cpu_ptr(vi
->stats
);
649 while ((skb
= virtqueue_get_buf(sq
->vq
, &len
)) != NULL
) {
650 pr_debug("Sent skb %p\n", skb
);
652 u64_stats_update_begin(&stats
->tx_syncp
);
653 stats
->tx_bytes
+= skb
->len
;
655 u64_stats_update_end(&stats
->tx_syncp
);
657 dev_kfree_skb_any(skb
);
661 static int xmit_skb(struct send_queue
*sq
, struct sk_buff
*skb
)
663 struct skb_vnet_hdr
*hdr
= skb_vnet_hdr(skb
);
664 const unsigned char *dest
= ((struct ethhdr
*)skb
->data
)->h_dest
;
665 struct virtnet_info
*vi
= sq
->vq
->vdev
->priv
;
668 pr_debug("%s: xmit %p %pM\n", vi
->dev
->name
, skb
, dest
);
670 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
671 hdr
->hdr
.flags
= VIRTIO_NET_HDR_F_NEEDS_CSUM
;
672 hdr
->hdr
.csum_start
= skb_checksum_start_offset(skb
);
673 hdr
->hdr
.csum_offset
= skb
->csum_offset
;
676 hdr
->hdr
.csum_offset
= hdr
->hdr
.csum_start
= 0;
679 if (skb_is_gso(skb
)) {
680 hdr
->hdr
.hdr_len
= skb_headlen(skb
);
681 hdr
->hdr
.gso_size
= skb_shinfo(skb
)->gso_size
;
682 if (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCPV4
)
683 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV4
;
684 else if (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCPV6
)
685 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV6
;
686 else if (skb_shinfo(skb
)->gso_type
& SKB_GSO_UDP
)
687 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_UDP
;
690 if (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCP_ECN
)
691 hdr
->hdr
.gso_type
|= VIRTIO_NET_HDR_GSO_ECN
;
693 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
694 hdr
->hdr
.gso_size
= hdr
->hdr
.hdr_len
= 0;
697 hdr
->mhdr
.num_buffers
= 0;
699 /* Encode metadata header at front. */
700 if (vi
->mergeable_rx_bufs
)
701 sg_set_buf(sq
->sg
, &hdr
->mhdr
, sizeof hdr
->mhdr
);
703 sg_set_buf(sq
->sg
, &hdr
->hdr
, sizeof hdr
->hdr
);
705 num_sg
= skb_to_sgvec(skb
, sq
->sg
+ 1, 0, skb
->len
) + 1;
706 return virtqueue_add_buf(sq
->vq
, sq
->sg
, num_sg
,
710 static netdev_tx_t
start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
712 struct virtnet_info
*vi
= netdev_priv(dev
);
713 int qnum
= skb_get_queue_mapping(skb
);
714 struct send_queue
*sq
= &vi
->sq
[qnum
];
717 /* Free up any pending old buffers before queueing new ones. */
718 free_old_xmit_skbs(sq
);
720 /* Try to transmit */
721 err
= xmit_skb(sq
, skb
);
723 /* This should not happen! */
725 dev
->stats
.tx_fifo_errors
++;
728 "Unexpected TXQ (%d) queue failure: %d\n", qnum
, err
);
729 dev
->stats
.tx_dropped
++;
733 virtqueue_kick(sq
->vq
);
735 /* Don't wait up for transmitted skbs to be freed. */
739 /* Apparently nice girls don't return TX_BUSY; stop the queue
740 * before it gets out of hand. Naturally, this wastes entries. */
741 if (sq
->vq
->num_free
< 2+MAX_SKB_FRAGS
) {
742 netif_stop_subqueue(dev
, qnum
);
743 if (unlikely(!virtqueue_enable_cb_delayed(sq
->vq
))) {
744 /* More just got used, free them then recheck. */
745 free_old_xmit_skbs(sq
);
746 if (sq
->vq
->num_free
>= 2+MAX_SKB_FRAGS
) {
747 netif_start_subqueue(dev
, qnum
);
748 virtqueue_disable_cb(sq
->vq
);
756 static int virtnet_set_mac_address(struct net_device
*dev
, void *p
)
758 struct virtnet_info
*vi
= netdev_priv(dev
);
759 struct virtio_device
*vdev
= vi
->vdev
;
762 ret
= eth_mac_addr(dev
, p
);
766 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MAC
))
767 vdev
->config
->set(vdev
, offsetof(struct virtio_net_config
, mac
),
768 dev
->dev_addr
, dev
->addr_len
);
773 static struct rtnl_link_stats64
*virtnet_stats(struct net_device
*dev
,
774 struct rtnl_link_stats64
*tot
)
776 struct virtnet_info
*vi
= netdev_priv(dev
);
780 for_each_possible_cpu(cpu
) {
781 struct virtnet_stats
*stats
= per_cpu_ptr(vi
->stats
, cpu
);
782 u64 tpackets
, tbytes
, rpackets
, rbytes
;
785 start
= u64_stats_fetch_begin_bh(&stats
->tx_syncp
);
786 tpackets
= stats
->tx_packets
;
787 tbytes
= stats
->tx_bytes
;
788 } while (u64_stats_fetch_retry_bh(&stats
->tx_syncp
, start
));
791 start
= u64_stats_fetch_begin_bh(&stats
->rx_syncp
);
792 rpackets
= stats
->rx_packets
;
793 rbytes
= stats
->rx_bytes
;
794 } while (u64_stats_fetch_retry_bh(&stats
->rx_syncp
, start
));
796 tot
->rx_packets
+= rpackets
;
797 tot
->tx_packets
+= tpackets
;
798 tot
->rx_bytes
+= rbytes
;
799 tot
->tx_bytes
+= tbytes
;
802 tot
->tx_dropped
= dev
->stats
.tx_dropped
;
803 tot
->tx_fifo_errors
= dev
->stats
.tx_fifo_errors
;
804 tot
->rx_dropped
= dev
->stats
.rx_dropped
;
805 tot
->rx_length_errors
= dev
->stats
.rx_length_errors
;
806 tot
->rx_frame_errors
= dev
->stats
.rx_frame_errors
;
811 #ifdef CONFIG_NET_POLL_CONTROLLER
812 static void virtnet_netpoll(struct net_device
*dev
)
814 struct virtnet_info
*vi
= netdev_priv(dev
);
817 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++)
818 napi_schedule(&vi
->rq
[i
].napi
);
823 * Send command via the control virtqueue and check status. Commands
824 * supported by the hypervisor, as indicated by feature bits, should
825 * never fail unless improperly formated.
827 static bool virtnet_send_command(struct virtnet_info
*vi
, u8
class, u8 cmd
,
828 struct scatterlist
*data
, int out
, int in
)
830 struct scatterlist
*s
, sg
[VIRTNET_SEND_COMMAND_SG_MAX
+ 2];
831 struct virtio_net_ctrl_hdr ctrl
;
832 virtio_net_ctrl_ack status
= ~0;
836 /* Caller should know better */
837 BUG_ON(!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
) ||
838 (out
+ in
> VIRTNET_SEND_COMMAND_SG_MAX
));
840 out
++; /* Add header */
841 in
++; /* Add return status */
846 sg_init_table(sg
, out
+ in
);
848 sg_set_buf(&sg
[0], &ctrl
, sizeof(ctrl
));
849 for_each_sg(data
, s
, out
+ in
- 2, i
)
850 sg_set_buf(&sg
[i
+ 1], sg_virt(s
), s
->length
);
851 sg_set_buf(&sg
[out
+ in
- 1], &status
, sizeof(status
));
853 BUG_ON(virtqueue_add_buf(vi
->cvq
, sg
, out
, in
, vi
, GFP_ATOMIC
) < 0);
855 virtqueue_kick(vi
->cvq
);
858 * Spin for a response, the kick causes an ioport write, trapping
859 * into the hypervisor, so the request should be handled immediately.
861 while (!virtqueue_get_buf(vi
->cvq
, &tmp
))
864 return status
== VIRTIO_NET_OK
;
867 static void virtnet_ack_link_announce(struct virtnet_info
*vi
)
870 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_ANNOUNCE
,
871 VIRTIO_NET_CTRL_ANNOUNCE_ACK
, NULL
,
873 dev_warn(&vi
->dev
->dev
, "Failed to ack link announce.\n");
877 static int virtnet_set_queues(struct virtnet_info
*vi
, u16 queue_pairs
)
879 struct scatterlist sg
;
880 struct virtio_net_ctrl_mq s
;
881 struct net_device
*dev
= vi
->dev
;
883 if (!vi
->has_cvq
|| !virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_MQ
))
886 s
.virtqueue_pairs
= queue_pairs
;
887 sg_init_one(&sg
, &s
, sizeof(s
));
889 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MQ
,
890 VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET
, &sg
, 1, 0)){
891 dev_warn(&dev
->dev
, "Fail to set num of queue pairs to %d\n",
895 vi
->curr_queue_pairs
= queue_pairs
;
900 static int virtnet_close(struct net_device
*dev
)
902 struct virtnet_info
*vi
= netdev_priv(dev
);
905 /* Make sure refill_work doesn't re-enable napi! */
906 cancel_delayed_work_sync(&vi
->refill
);
908 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
909 napi_disable(&vi
->rq
[i
].napi
);
914 static void virtnet_set_rx_mode(struct net_device
*dev
)
916 struct virtnet_info
*vi
= netdev_priv(dev
);
917 struct scatterlist sg
[2];
918 u8 promisc
, allmulti
;
919 struct virtio_net_ctrl_mac
*mac_data
;
920 struct netdev_hw_addr
*ha
;
926 /* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */
927 if (!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_RX
))
930 promisc
= ((dev
->flags
& IFF_PROMISC
) != 0);
931 allmulti
= ((dev
->flags
& IFF_ALLMULTI
) != 0);
933 sg_init_one(sg
, &promisc
, sizeof(promisc
));
935 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_RX
,
936 VIRTIO_NET_CTRL_RX_PROMISC
,
938 dev_warn(&dev
->dev
, "Failed to %sable promisc mode.\n",
939 promisc
? "en" : "dis");
941 sg_init_one(sg
, &allmulti
, sizeof(allmulti
));
943 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_RX
,
944 VIRTIO_NET_CTRL_RX_ALLMULTI
,
946 dev_warn(&dev
->dev
, "Failed to %sable allmulti mode.\n",
947 allmulti
? "en" : "dis");
949 uc_count
= netdev_uc_count(dev
);
950 mc_count
= netdev_mc_count(dev
);
951 /* MAC filter - use one buffer for both lists */
952 buf
= kzalloc(((uc_count
+ mc_count
) * ETH_ALEN
) +
953 (2 * sizeof(mac_data
->entries
)), GFP_ATOMIC
);
956 dev_warn(&dev
->dev
, "No memory for MAC address buffer\n");
960 sg_init_table(sg
, 2);
962 /* Store the unicast list and count in the front of the buffer */
963 mac_data
->entries
= uc_count
;
965 netdev_for_each_uc_addr(ha
, dev
)
966 memcpy(&mac_data
->macs
[i
++][0], ha
->addr
, ETH_ALEN
);
968 sg_set_buf(&sg
[0], mac_data
,
969 sizeof(mac_data
->entries
) + (uc_count
* ETH_ALEN
));
971 /* multicast list and count fill the end */
972 mac_data
= (void *)&mac_data
->macs
[uc_count
][0];
974 mac_data
->entries
= mc_count
;
976 netdev_for_each_mc_addr(ha
, dev
)
977 memcpy(&mac_data
->macs
[i
++][0], ha
->addr
, ETH_ALEN
);
979 sg_set_buf(&sg
[1], mac_data
,
980 sizeof(mac_data
->entries
) + (mc_count
* ETH_ALEN
));
982 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MAC
,
983 VIRTIO_NET_CTRL_MAC_TABLE_SET
,
985 dev_warn(&dev
->dev
, "Failed to set MAC fitler table.\n");
990 static int virtnet_vlan_rx_add_vid(struct net_device
*dev
, u16 vid
)
992 struct virtnet_info
*vi
= netdev_priv(dev
);
993 struct scatterlist sg
;
995 sg_init_one(&sg
, &vid
, sizeof(vid
));
997 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_VLAN
,
998 VIRTIO_NET_CTRL_VLAN_ADD
, &sg
, 1, 0))
999 dev_warn(&dev
->dev
, "Failed to add VLAN ID %d.\n", vid
);
1003 static int virtnet_vlan_rx_kill_vid(struct net_device
*dev
, u16 vid
)
1005 struct virtnet_info
*vi
= netdev_priv(dev
);
1006 struct scatterlist sg
;
1008 sg_init_one(&sg
, &vid
, sizeof(vid
));
1010 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_VLAN
,
1011 VIRTIO_NET_CTRL_VLAN_DEL
, &sg
, 1, 0))
1012 dev_warn(&dev
->dev
, "Failed to kill VLAN ID %d.\n", vid
);
1016 static void virtnet_set_affinity(struct virtnet_info
*vi
, bool set
)
1020 /* In multiqueue mode, when the number of cpu is equal to the number of
1021 * queue pairs, we let the queue pairs to be private to one cpu by
1022 * setting the affinity hint to eliminate the contention.
1024 if ((vi
->curr_queue_pairs
== 1 ||
1025 vi
->max_queue_pairs
!= num_online_cpus()) && set
) {
1026 if (vi
->affinity_hint_set
)
1032 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1033 int cpu
= set
? i
: -1;
1034 virtqueue_set_affinity(vi
->rq
[i
].vq
, cpu
);
1035 virtqueue_set_affinity(vi
->sq
[i
].vq
, cpu
);
1039 vi
->affinity_hint_set
= true;
1041 vi
->affinity_hint_set
= false;
1044 static void virtnet_get_ringparam(struct net_device
*dev
,
1045 struct ethtool_ringparam
*ring
)
1047 struct virtnet_info
*vi
= netdev_priv(dev
);
1049 ring
->rx_max_pending
= virtqueue_get_vring_size(vi
->rq
[0].vq
);
1050 ring
->tx_max_pending
= virtqueue_get_vring_size(vi
->sq
[0].vq
);
1051 ring
->rx_pending
= ring
->rx_max_pending
;
1052 ring
->tx_pending
= ring
->tx_max_pending
;
1056 static void virtnet_get_drvinfo(struct net_device
*dev
,
1057 struct ethtool_drvinfo
*info
)
1059 struct virtnet_info
*vi
= netdev_priv(dev
);
1060 struct virtio_device
*vdev
= vi
->vdev
;
1062 strlcpy(info
->driver
, KBUILD_MODNAME
, sizeof(info
->driver
));
1063 strlcpy(info
->version
, VIRTNET_DRIVER_VERSION
, sizeof(info
->version
));
1064 strlcpy(info
->bus_info
, virtio_bus_name(vdev
), sizeof(info
->bus_info
));
1068 /* TODO: Eliminate OOO packets during switching */
1069 static int virtnet_set_channels(struct net_device
*dev
,
1070 struct ethtool_channels
*channels
)
1072 struct virtnet_info
*vi
= netdev_priv(dev
);
1073 u16 queue_pairs
= channels
->combined_count
;
1076 /* We don't support separate rx/tx channels.
1077 * We don't allow setting 'other' channels.
1079 if (channels
->rx_count
|| channels
->tx_count
|| channels
->other_count
)
1082 if (queue_pairs
> vi
->max_queue_pairs
)
1085 err
= virtnet_set_queues(vi
, queue_pairs
);
1087 netif_set_real_num_tx_queues(dev
, queue_pairs
);
1088 netif_set_real_num_rx_queues(dev
, queue_pairs
);
1090 virtnet_set_affinity(vi
, true);
1096 static void virtnet_get_channels(struct net_device
*dev
,
1097 struct ethtool_channels
*channels
)
1099 struct virtnet_info
*vi
= netdev_priv(dev
);
1101 channels
->combined_count
= vi
->curr_queue_pairs
;
1102 channels
->max_combined
= vi
->max_queue_pairs
;
1103 channels
->max_other
= 0;
1104 channels
->rx_count
= 0;
1105 channels
->tx_count
= 0;
1106 channels
->other_count
= 0;
1109 static const struct ethtool_ops virtnet_ethtool_ops
= {
1110 .get_drvinfo
= virtnet_get_drvinfo
,
1111 .get_link
= ethtool_op_get_link
,
1112 .get_ringparam
= virtnet_get_ringparam
,
1113 .set_channels
= virtnet_set_channels
,
1114 .get_channels
= virtnet_get_channels
,
1118 #define MAX_MTU 65535
1120 static int virtnet_change_mtu(struct net_device
*dev
, int new_mtu
)
1122 if (new_mtu
< MIN_MTU
|| new_mtu
> MAX_MTU
)
1128 /* To avoid contending a lock hold by a vcpu who would exit to host, select the
1129 * txq based on the processor id.
1130 * TODO: handle cpu hotplug.
1132 static u16
virtnet_select_queue(struct net_device
*dev
, struct sk_buff
*skb
)
1134 int txq
= skb_rx_queue_recorded(skb
) ? skb_get_rx_queue(skb
) :
1137 while (unlikely(txq
>= dev
->real_num_tx_queues
))
1138 txq
-= dev
->real_num_tx_queues
;
1143 static const struct net_device_ops virtnet_netdev
= {
1144 .ndo_open
= virtnet_open
,
1145 .ndo_stop
= virtnet_close
,
1146 .ndo_start_xmit
= start_xmit
,
1147 .ndo_validate_addr
= eth_validate_addr
,
1148 .ndo_set_mac_address
= virtnet_set_mac_address
,
1149 .ndo_set_rx_mode
= virtnet_set_rx_mode
,
1150 .ndo_change_mtu
= virtnet_change_mtu
,
1151 .ndo_get_stats64
= virtnet_stats
,
1152 .ndo_vlan_rx_add_vid
= virtnet_vlan_rx_add_vid
,
1153 .ndo_vlan_rx_kill_vid
= virtnet_vlan_rx_kill_vid
,
1154 .ndo_select_queue
= virtnet_select_queue
,
1155 #ifdef CONFIG_NET_POLL_CONTROLLER
1156 .ndo_poll_controller
= virtnet_netpoll
,
1160 static void virtnet_config_changed_work(struct work_struct
*work
)
1162 struct virtnet_info
*vi
=
1163 container_of(work
, struct virtnet_info
, config_work
);
1166 mutex_lock(&vi
->config_lock
);
1167 if (!vi
->config_enable
)
1170 if (virtio_config_val(vi
->vdev
, VIRTIO_NET_F_STATUS
,
1171 offsetof(struct virtio_net_config
, status
),
1175 if (v
& VIRTIO_NET_S_ANNOUNCE
) {
1176 netdev_notify_peers(vi
->dev
);
1177 virtnet_ack_link_announce(vi
);
1180 /* Ignore unknown (future) status bits */
1181 v
&= VIRTIO_NET_S_LINK_UP
;
1183 if (vi
->status
== v
)
1188 if (vi
->status
& VIRTIO_NET_S_LINK_UP
) {
1189 netif_carrier_on(vi
->dev
);
1190 netif_tx_wake_all_queues(vi
->dev
);
1192 netif_carrier_off(vi
->dev
);
1193 netif_tx_stop_all_queues(vi
->dev
);
1196 mutex_unlock(&vi
->config_lock
);
1199 static void virtnet_config_changed(struct virtio_device
*vdev
)
1201 struct virtnet_info
*vi
= vdev
->priv
;
1203 schedule_work(&vi
->config_work
);
1206 static void virtnet_free_queues(struct virtnet_info
*vi
)
1212 static void free_receive_bufs(struct virtnet_info
*vi
)
1216 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1217 while (vi
->rq
[i
].pages
)
1218 __free_pages(get_a_page(&vi
->rq
[i
], GFP_KERNEL
), 0);
1222 static void free_unused_bufs(struct virtnet_info
*vi
)
1227 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1228 struct virtqueue
*vq
= vi
->sq
[i
].vq
;
1229 while ((buf
= virtqueue_detach_unused_buf(vq
)) != NULL
)
1233 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1234 struct virtqueue
*vq
= vi
->rq
[i
].vq
;
1236 while ((buf
= virtqueue_detach_unused_buf(vq
)) != NULL
) {
1237 if (vi
->mergeable_rx_bufs
|| vi
->big_packets
)
1238 give_pages(&vi
->rq
[i
], buf
);
1243 BUG_ON(vi
->rq
[i
].num
!= 0);
1247 static void virtnet_del_vqs(struct virtnet_info
*vi
)
1249 struct virtio_device
*vdev
= vi
->vdev
;
1251 virtnet_set_affinity(vi
, false);
1253 vdev
->config
->del_vqs(vdev
);
1255 virtnet_free_queues(vi
);
1258 static int virtnet_find_vqs(struct virtnet_info
*vi
)
1260 vq_callback_t
**callbacks
;
1261 struct virtqueue
**vqs
;
1266 /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
1267 * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
1268 * possible control vq.
1270 total_vqs
= vi
->max_queue_pairs
* 2 +
1271 virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
);
1273 /* Allocate space for find_vqs parameters */
1274 vqs
= kzalloc(total_vqs
* sizeof(*vqs
), GFP_KERNEL
);
1277 callbacks
= kmalloc(total_vqs
* sizeof(*callbacks
), GFP_KERNEL
);
1280 names
= kmalloc(total_vqs
* sizeof(*names
), GFP_KERNEL
);
1284 /* Parameters for control virtqueue, if any */
1286 callbacks
[total_vqs
- 1] = NULL
;
1287 names
[total_vqs
- 1] = "control";
1290 /* Allocate/initialize parameters for send/receive virtqueues */
1291 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1292 callbacks
[rxq2vq(i
)] = skb_recv_done
;
1293 callbacks
[txq2vq(i
)] = skb_xmit_done
;
1294 sprintf(vi
->rq
[i
].name
, "input.%d", i
);
1295 sprintf(vi
->sq
[i
].name
, "output.%d", i
);
1296 names
[rxq2vq(i
)] = vi
->rq
[i
].name
;
1297 names
[txq2vq(i
)] = vi
->sq
[i
].name
;
1300 ret
= vi
->vdev
->config
->find_vqs(vi
->vdev
, total_vqs
, vqs
, callbacks
,
1306 vi
->cvq
= vqs
[total_vqs
- 1];
1307 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VLAN
))
1308 vi
->dev
->features
|= NETIF_F_HW_VLAN_FILTER
;
1311 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1312 vi
->rq
[i
].vq
= vqs
[rxq2vq(i
)];
1313 vi
->sq
[i
].vq
= vqs
[txq2vq(i
)];
1332 static int virtnet_alloc_queues(struct virtnet_info
*vi
)
1336 vi
->sq
= kzalloc(sizeof(*vi
->sq
) * vi
->max_queue_pairs
, GFP_KERNEL
);
1339 vi
->rq
= kzalloc(sizeof(*vi
->rq
) * vi
->max_queue_pairs
, GFP_KERNEL
);
1343 INIT_DELAYED_WORK(&vi
->refill
, refill_work
);
1344 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1345 vi
->rq
[i
].pages
= NULL
;
1346 netif_napi_add(vi
->dev
, &vi
->rq
[i
].napi
, virtnet_poll
,
1349 sg_init_table(vi
->rq
[i
].sg
, ARRAY_SIZE(vi
->rq
[i
].sg
));
1350 sg_init_table(vi
->sq
[i
].sg
, ARRAY_SIZE(vi
->sq
[i
].sg
));
1361 static int init_vqs(struct virtnet_info
*vi
)
1365 /* Allocate send & receive queues */
1366 ret
= virtnet_alloc_queues(vi
);
1370 ret
= virtnet_find_vqs(vi
);
1374 virtnet_set_affinity(vi
, true);
1378 virtnet_free_queues(vi
);
1383 static int virtnet_probe(struct virtio_device
*vdev
)
1386 struct net_device
*dev
;
1387 struct virtnet_info
*vi
;
1388 u16 max_queue_pairs
;
1390 /* Find if host supports multiqueue virtio_net device */
1391 err
= virtio_config_val(vdev
, VIRTIO_NET_F_MQ
,
1392 offsetof(struct virtio_net_config
,
1393 max_virtqueue_pairs
), &max_queue_pairs
);
1395 /* We need at least 2 queue's */
1396 if (err
|| max_queue_pairs
< VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN
||
1397 max_queue_pairs
> VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX
||
1398 !virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
))
1399 max_queue_pairs
= 1;
1401 /* Allocate ourselves a network device with room for our info */
1402 dev
= alloc_etherdev_mq(sizeof(struct virtnet_info
), max_queue_pairs
);
1406 /* Set up network device as normal. */
1407 dev
->priv_flags
|= IFF_UNICAST_FLT
| IFF_LIVE_ADDR_CHANGE
;
1408 dev
->netdev_ops
= &virtnet_netdev
;
1409 dev
->features
= NETIF_F_HIGHDMA
;
1411 SET_ETHTOOL_OPS(dev
, &virtnet_ethtool_ops
);
1412 SET_NETDEV_DEV(dev
, &vdev
->dev
);
1414 /* Do we support "hardware" checksums? */
1415 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CSUM
)) {
1416 /* This opens up the world of extra features. */
1417 dev
->hw_features
|= NETIF_F_HW_CSUM
|NETIF_F_SG
|NETIF_F_FRAGLIST
;
1419 dev
->features
|= NETIF_F_HW_CSUM
|NETIF_F_SG
|NETIF_F_FRAGLIST
;
1421 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GSO
)) {
1422 dev
->hw_features
|= NETIF_F_TSO
| NETIF_F_UFO
1423 | NETIF_F_TSO_ECN
| NETIF_F_TSO6
;
1425 /* Individual feature bits: what can host handle? */
1426 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_TSO4
))
1427 dev
->hw_features
|= NETIF_F_TSO
;
1428 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_TSO6
))
1429 dev
->hw_features
|= NETIF_F_TSO6
;
1430 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_ECN
))
1431 dev
->hw_features
|= NETIF_F_TSO_ECN
;
1432 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_UFO
))
1433 dev
->hw_features
|= NETIF_F_UFO
;
1436 dev
->features
|= dev
->hw_features
& (NETIF_F_ALL_TSO
|NETIF_F_UFO
);
1437 /* (!csum && gso) case will be fixed by register_netdev() */
1440 /* Configuration may specify what MAC to use. Otherwise random. */
1441 if (virtio_config_val_len(vdev
, VIRTIO_NET_F_MAC
,
1442 offsetof(struct virtio_net_config
, mac
),
1443 dev
->dev_addr
, dev
->addr_len
) < 0)
1444 eth_hw_addr_random(dev
);
1446 /* Set up our device-specific information */
1447 vi
= netdev_priv(dev
);
1451 vi
->stats
= alloc_percpu(struct virtnet_stats
);
1453 if (vi
->stats
== NULL
)
1456 mutex_init(&vi
->config_lock
);
1457 vi
->config_enable
= true;
1458 INIT_WORK(&vi
->config_work
, virtnet_config_changed_work
);
1460 /* If we can receive ANY GSO packets, we must allocate large ones. */
1461 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_TSO4
) ||
1462 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_TSO6
) ||
1463 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_ECN
))
1464 vi
->big_packets
= true;
1466 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MRG_RXBUF
))
1467 vi
->mergeable_rx_bufs
= true;
1469 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
))
1472 /* Use single tx/rx queue pair as default */
1473 vi
->curr_queue_pairs
= 1;
1474 vi
->max_queue_pairs
= max_queue_pairs
;
1476 /* Allocate/initialize the rx/tx queues, and invoke find_vqs */
1481 netif_set_real_num_tx_queues(dev
, 1);
1482 netif_set_real_num_rx_queues(dev
, 1);
1484 err
= register_netdev(dev
);
1486 pr_debug("virtio_net: registering device failed\n");
1490 /* Last of all, set up some receive buffers. */
1491 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1492 try_fill_recv(&vi
->rq
[i
], GFP_KERNEL
);
1494 /* If we didn't even get one input buffer, we're useless. */
1495 if (vi
->rq
[i
].num
== 0) {
1496 free_unused_bufs(vi
);
1498 goto free_recv_bufs
;
1502 /* Assume link up if device can't report link status,
1503 otherwise get link status from config. */
1504 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_STATUS
)) {
1505 netif_carrier_off(dev
);
1506 schedule_work(&vi
->config_work
);
1508 vi
->status
= VIRTIO_NET_S_LINK_UP
;
1509 netif_carrier_on(dev
);
1512 pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
1513 dev
->name
, max_queue_pairs
);
1518 free_receive_bufs(vi
);
1519 unregister_netdev(dev
);
1521 cancel_delayed_work_sync(&vi
->refill
);
1522 virtnet_del_vqs(vi
);
1524 free_percpu(vi
->stats
);
1530 static void remove_vq_common(struct virtnet_info
*vi
)
1532 vi
->vdev
->config
->reset(vi
->vdev
);
1534 /* Free unused buffers in both send and recv, if any. */
1535 free_unused_bufs(vi
);
1537 free_receive_bufs(vi
);
1539 virtnet_del_vqs(vi
);
1542 static void virtnet_remove(struct virtio_device
*vdev
)
1544 struct virtnet_info
*vi
= vdev
->priv
;
1546 /* Prevent config work handler from accessing the device. */
1547 mutex_lock(&vi
->config_lock
);
1548 vi
->config_enable
= false;
1549 mutex_unlock(&vi
->config_lock
);
1551 unregister_netdev(vi
->dev
);
1553 remove_vq_common(vi
);
1555 flush_work(&vi
->config_work
);
1557 free_percpu(vi
->stats
);
1558 free_netdev(vi
->dev
);
1562 static int virtnet_freeze(struct virtio_device
*vdev
)
1564 struct virtnet_info
*vi
= vdev
->priv
;
1567 /* Prevent config work handler from accessing the device */
1568 mutex_lock(&vi
->config_lock
);
1569 vi
->config_enable
= false;
1570 mutex_unlock(&vi
->config_lock
);
1572 netif_device_detach(vi
->dev
);
1573 cancel_delayed_work_sync(&vi
->refill
);
1575 if (netif_running(vi
->dev
))
1576 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1577 napi_disable(&vi
->rq
[i
].napi
);
1578 netif_napi_del(&vi
->rq
[i
].napi
);
1581 remove_vq_common(vi
);
1583 flush_work(&vi
->config_work
);
1588 static int virtnet_restore(struct virtio_device
*vdev
)
1590 struct virtnet_info
*vi
= vdev
->priv
;
1597 if (netif_running(vi
->dev
))
1598 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
1599 virtnet_napi_enable(&vi
->rq
[i
]);
1601 netif_device_attach(vi
->dev
);
1603 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
1604 if (!try_fill_recv(&vi
->rq
[i
], GFP_KERNEL
))
1605 schedule_delayed_work(&vi
->refill
, 0);
1607 mutex_lock(&vi
->config_lock
);
1608 vi
->config_enable
= true;
1609 mutex_unlock(&vi
->config_lock
);
1611 virtnet_set_queues(vi
, vi
->curr_queue_pairs
);
1617 static struct virtio_device_id id_table
[] = {
1618 { VIRTIO_ID_NET
, VIRTIO_DEV_ANY_ID
},
1622 static unsigned int features
[] = {
1623 VIRTIO_NET_F_CSUM
, VIRTIO_NET_F_GUEST_CSUM
,
1624 VIRTIO_NET_F_GSO
, VIRTIO_NET_F_MAC
,
1625 VIRTIO_NET_F_HOST_TSO4
, VIRTIO_NET_F_HOST_UFO
, VIRTIO_NET_F_HOST_TSO6
,
1626 VIRTIO_NET_F_HOST_ECN
, VIRTIO_NET_F_GUEST_TSO4
, VIRTIO_NET_F_GUEST_TSO6
,
1627 VIRTIO_NET_F_GUEST_ECN
, VIRTIO_NET_F_GUEST_UFO
,
1628 VIRTIO_NET_F_MRG_RXBUF
, VIRTIO_NET_F_STATUS
, VIRTIO_NET_F_CTRL_VQ
,
1629 VIRTIO_NET_F_CTRL_RX
, VIRTIO_NET_F_CTRL_VLAN
,
1630 VIRTIO_NET_F_GUEST_ANNOUNCE
, VIRTIO_NET_F_MQ
,
1633 static struct virtio_driver virtio_net_driver
= {
1634 .feature_table
= features
,
1635 .feature_table_size
= ARRAY_SIZE(features
),
1636 .driver
.name
= KBUILD_MODNAME
,
1637 .driver
.owner
= THIS_MODULE
,
1638 .id_table
= id_table
,
1639 .probe
= virtnet_probe
,
1640 .remove
= virtnet_remove
,
1641 .config_changed
= virtnet_config_changed
,
1643 .freeze
= virtnet_freeze
,
1644 .restore
= virtnet_restore
,
1648 static int __init
init(void)
1650 return register_virtio_driver(&virtio_net_driver
);
1653 static void __exit
fini(void)
1655 unregister_virtio_driver(&virtio_net_driver
);
1660 MODULE_DEVICE_TABLE(virtio
, id_table
);
1661 MODULE_DESCRIPTION("Virtio network driver");
1662 MODULE_LICENSE("GPL");