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>
29 #include <linux/cpu.h>
31 static int napi_weight
= NAPI_POLL_WEIGHT
;
32 module_param(napi_weight
, int, 0444);
34 static bool csum
= true, gso
= true;
35 module_param(csum
, bool, 0444);
36 module_param(gso
, bool, 0444);
38 /* FIXME: MTU in config. */
39 #define MAX_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
40 #define GOOD_COPY_LEN 128
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 /* Host can handle any s/g split between our header and packet data */
112 /* enable config space updates */
115 /* Active statistics */
116 struct virtnet_stats __percpu
*stats
;
118 /* Work struct for refilling if we run low on memory. */
119 struct delayed_work refill
;
121 /* Work struct for config space updates */
122 struct work_struct config_work
;
124 /* Lock for config space updates */
125 struct mutex config_lock
;
127 /* Does the affinity hint is set for virtqueues? */
128 bool affinity_hint_set
;
130 /* Per-cpu variable to show the mapping from CPU to virtqueue */
131 int __percpu
*vq_index
;
133 /* CPU hot plug notifier */
134 struct notifier_block nb
;
137 struct skb_vnet_hdr
{
139 struct virtio_net_hdr hdr
;
140 struct virtio_net_hdr_mrg_rxbuf mhdr
;
144 struct padded_vnet_hdr
{
145 struct virtio_net_hdr hdr
;
147 * virtio_net_hdr should be in a separated sg buffer because of a
148 * QEMU bug, and data sg buffer shares same page with this header sg.
149 * This padding makes next sg 16 byte aligned after virtio_net_hdr.
154 /* Converting between virtqueue no. and kernel tx/rx queue no.
155 * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq
157 static int vq2txq(struct virtqueue
*vq
)
159 return (vq
->index
- 1) / 2;
162 static int txq2vq(int txq
)
167 static int vq2rxq(struct virtqueue
*vq
)
169 return vq
->index
/ 2;
172 static int rxq2vq(int rxq
)
177 static inline struct skb_vnet_hdr
*skb_vnet_hdr(struct sk_buff
*skb
)
179 return (struct skb_vnet_hdr
*)skb
->cb
;
183 * private is used to chain pages for big packets, put the whole
184 * most recent used list in the beginning for reuse
186 static void give_pages(struct receive_queue
*rq
, struct page
*page
)
190 /* Find end of list, sew whole thing into vi->rq.pages. */
191 for (end
= page
; end
->private; end
= (struct page
*)end
->private);
192 end
->private = (unsigned long)rq
->pages
;
196 static struct page
*get_a_page(struct receive_queue
*rq
, gfp_t gfp_mask
)
198 struct page
*p
= rq
->pages
;
201 rq
->pages
= (struct page
*)p
->private;
202 /* clear private here, it is used to chain pages */
205 p
= alloc_page(gfp_mask
);
209 static void skb_xmit_done(struct virtqueue
*vq
)
211 struct virtnet_info
*vi
= vq
->vdev
->priv
;
213 /* Suppress further interrupts. */
214 virtqueue_disable_cb(vq
);
216 /* We were probably waiting for more output buffers. */
217 netif_wake_subqueue(vi
->dev
, vq2txq(vq
));
220 static void set_skb_frag(struct sk_buff
*skb
, struct page
*page
,
221 unsigned int offset
, unsigned int *len
)
223 int size
= min((unsigned)PAGE_SIZE
- offset
, *len
);
224 int i
= skb_shinfo(skb
)->nr_frags
;
226 __skb_fill_page_desc(skb
, i
, page
, offset
, size
);
228 skb
->data_len
+= size
;
230 skb
->truesize
+= PAGE_SIZE
;
231 skb_shinfo(skb
)->nr_frags
++;
232 skb_shinfo(skb
)->tx_flags
|= SKBTX_SHARED_FRAG
;
236 /* Called from bottom half context */
237 static struct sk_buff
*page_to_skb(struct receive_queue
*rq
,
238 struct page
*page
, unsigned int len
)
240 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
242 struct skb_vnet_hdr
*hdr
;
243 unsigned int copy
, hdr_len
, offset
;
246 p
= page_address(page
);
248 /* copy small packet so we can reuse these pages for small data */
249 skb
= netdev_alloc_skb_ip_align(vi
->dev
, GOOD_COPY_LEN
);
253 hdr
= skb_vnet_hdr(skb
);
255 if (vi
->mergeable_rx_bufs
) {
256 hdr_len
= sizeof hdr
->mhdr
;
259 hdr_len
= sizeof hdr
->hdr
;
260 offset
= sizeof(struct padded_vnet_hdr
);
263 memcpy(hdr
, p
, hdr_len
);
269 if (copy
> skb_tailroom(skb
))
270 copy
= skb_tailroom(skb
);
271 memcpy(skb_put(skb
, copy
), p
, copy
);
277 * Verify that we can indeed put this data into a skb.
278 * This is here to handle cases when the device erroneously
279 * tries to receive more than is possible. This is usually
280 * the case of a broken device.
282 if (unlikely(len
> MAX_SKB_FRAGS
* PAGE_SIZE
)) {
283 net_dbg_ratelimited("%s: too much data\n", skb
->dev
->name
);
289 set_skb_frag(skb
, page
, offset
, &len
);
290 page
= (struct page
*)page
->private;
295 give_pages(rq
, page
);
300 static int receive_mergeable(struct receive_queue
*rq
, struct sk_buff
*skb
)
302 struct skb_vnet_hdr
*hdr
= skb_vnet_hdr(skb
);
306 num_buf
= hdr
->mhdr
.num_buffers
;
308 i
= skb_shinfo(skb
)->nr_frags
;
309 if (i
>= MAX_SKB_FRAGS
) {
310 pr_debug("%s: packet too long\n", skb
->dev
->name
);
311 skb
->dev
->stats
.rx_length_errors
++;
314 page
= virtqueue_get_buf(rq
->vq
, &len
);
316 pr_debug("%s: rx error: %d buffers missing\n",
317 skb
->dev
->name
, hdr
->mhdr
.num_buffers
);
318 skb
->dev
->stats
.rx_length_errors
++;
325 set_skb_frag(skb
, page
, 0, &len
);
332 static void receive_buf(struct receive_queue
*rq
, void *buf
, unsigned int len
)
334 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
335 struct net_device
*dev
= vi
->dev
;
336 struct virtnet_stats
*stats
= this_cpu_ptr(vi
->stats
);
339 struct skb_vnet_hdr
*hdr
;
341 if (unlikely(len
< sizeof(struct virtio_net_hdr
) + ETH_HLEN
)) {
342 pr_debug("%s: short packet %i\n", dev
->name
, len
);
343 dev
->stats
.rx_length_errors
++;
344 if (vi
->mergeable_rx_bufs
|| vi
->big_packets
)
351 if (!vi
->mergeable_rx_bufs
&& !vi
->big_packets
) {
353 len
-= sizeof(struct virtio_net_hdr
);
357 skb
= page_to_skb(rq
, page
, len
);
358 if (unlikely(!skb
)) {
359 dev
->stats
.rx_dropped
++;
360 give_pages(rq
, page
);
363 if (vi
->mergeable_rx_bufs
)
364 if (receive_mergeable(rq
, skb
)) {
370 hdr
= skb_vnet_hdr(skb
);
372 u64_stats_update_begin(&stats
->rx_syncp
);
373 stats
->rx_bytes
+= skb
->len
;
375 u64_stats_update_end(&stats
->rx_syncp
);
377 if (hdr
->hdr
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) {
378 pr_debug("Needs csum!\n");
379 if (!skb_partial_csum_set(skb
,
381 hdr
->hdr
.csum_offset
))
383 } else if (hdr
->hdr
.flags
& VIRTIO_NET_HDR_F_DATA_VALID
) {
384 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
387 skb
->protocol
= eth_type_trans(skb
, dev
);
388 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
389 ntohs(skb
->protocol
), skb
->len
, skb
->pkt_type
);
391 if (hdr
->hdr
.gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
393 switch (hdr
->hdr
.gso_type
& ~VIRTIO_NET_HDR_GSO_ECN
) {
394 case VIRTIO_NET_HDR_GSO_TCPV4
:
395 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV4
;
397 case VIRTIO_NET_HDR_GSO_UDP
:
398 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
400 case VIRTIO_NET_HDR_GSO_TCPV6
:
401 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV6
;
404 net_warn_ratelimited("%s: bad gso type %u.\n",
405 dev
->name
, hdr
->hdr
.gso_type
);
409 if (hdr
->hdr
.gso_type
& VIRTIO_NET_HDR_GSO_ECN
)
410 skb_shinfo(skb
)->gso_type
|= SKB_GSO_TCP_ECN
;
412 skb_shinfo(skb
)->gso_size
= hdr
->hdr
.gso_size
;
413 if (skb_shinfo(skb
)->gso_size
== 0) {
414 net_warn_ratelimited("%s: zero gso size.\n", dev
->name
);
418 /* Header must be checked, and gso_segs computed. */
419 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
420 skb_shinfo(skb
)->gso_segs
= 0;
423 netif_receive_skb(skb
);
427 dev
->stats
.rx_frame_errors
++;
431 static int add_recvbuf_small(struct receive_queue
*rq
, gfp_t gfp
)
433 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
435 struct skb_vnet_hdr
*hdr
;
438 skb
= __netdev_alloc_skb_ip_align(vi
->dev
, MAX_PACKET_LEN
, gfp
);
442 skb_put(skb
, MAX_PACKET_LEN
);
444 hdr
= skb_vnet_hdr(skb
);
445 sg_set_buf(rq
->sg
, &hdr
->hdr
, sizeof hdr
->hdr
);
447 skb_to_sgvec(skb
, rq
->sg
+ 1, 0, skb
->len
);
449 err
= virtqueue_add_inbuf(rq
->vq
, rq
->sg
, 2, skb
, gfp
);
456 static int add_recvbuf_big(struct receive_queue
*rq
, gfp_t gfp
)
458 struct page
*first
, *list
= NULL
;
462 /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */
463 for (i
= MAX_SKB_FRAGS
+ 1; i
> 1; --i
) {
464 first
= get_a_page(rq
, gfp
);
467 give_pages(rq
, list
);
470 sg_set_buf(&rq
->sg
[i
], page_address(first
), PAGE_SIZE
);
472 /* chain new page in list head to match sg */
473 first
->private = (unsigned long)list
;
477 first
= get_a_page(rq
, gfp
);
479 give_pages(rq
, list
);
482 p
= page_address(first
);
484 /* rq->sg[0], rq->sg[1] share the same page */
485 /* a separated rq->sg[0] for virtio_net_hdr only due to QEMU bug */
486 sg_set_buf(&rq
->sg
[0], p
, sizeof(struct virtio_net_hdr
));
488 /* rq->sg[1] for data packet, from offset */
489 offset
= sizeof(struct padded_vnet_hdr
);
490 sg_set_buf(&rq
->sg
[1], p
+ offset
, PAGE_SIZE
- offset
);
492 /* chain first in list head */
493 first
->private = (unsigned long)list
;
494 err
= virtqueue_add_inbuf(rq
->vq
, rq
->sg
, MAX_SKB_FRAGS
+ 2,
497 give_pages(rq
, first
);
502 static int add_recvbuf_mergeable(struct receive_queue
*rq
, gfp_t gfp
)
507 page
= get_a_page(rq
, gfp
);
511 sg_init_one(rq
->sg
, page_address(page
), PAGE_SIZE
);
513 err
= virtqueue_add_inbuf(rq
->vq
, rq
->sg
, 1, page
, gfp
);
515 give_pages(rq
, page
);
521 * Returns false if we couldn't fill entirely (OOM).
523 * Normally run in the receive path, but can also be run from ndo_open
524 * before we're receiving packets, or from refill_work which is
525 * careful to disable receiving (using napi_disable).
527 static bool try_fill_recv(struct receive_queue
*rq
, gfp_t gfp
)
529 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
534 if (vi
->mergeable_rx_bufs
)
535 err
= add_recvbuf_mergeable(rq
, gfp
);
536 else if (vi
->big_packets
)
537 err
= add_recvbuf_big(rq
, gfp
);
539 err
= add_recvbuf_small(rq
, gfp
);
541 oom
= err
== -ENOMEM
;
545 } while (rq
->vq
->num_free
);
546 if (unlikely(rq
->num
> rq
->max
))
548 virtqueue_kick(rq
->vq
);
552 static void skb_recv_done(struct virtqueue
*rvq
)
554 struct virtnet_info
*vi
= rvq
->vdev
->priv
;
555 struct receive_queue
*rq
= &vi
->rq
[vq2rxq(rvq
)];
557 /* Schedule NAPI, Suppress further interrupts if successful. */
558 if (napi_schedule_prep(&rq
->napi
)) {
559 virtqueue_disable_cb(rvq
);
560 __napi_schedule(&rq
->napi
);
564 static void virtnet_napi_enable(struct receive_queue
*rq
)
566 napi_enable(&rq
->napi
);
568 /* If all buffers were filled by other side before we napi_enabled, we
569 * won't get another interrupt, so process any outstanding packets
570 * now. virtnet_poll wants re-enable the queue, so we disable here.
571 * We synchronize against interrupts via NAPI_STATE_SCHED */
572 if (napi_schedule_prep(&rq
->napi
)) {
573 virtqueue_disable_cb(rq
->vq
);
575 __napi_schedule(&rq
->napi
);
580 static void refill_work(struct work_struct
*work
)
582 struct virtnet_info
*vi
=
583 container_of(work
, struct virtnet_info
, refill
.work
);
587 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++) {
588 struct receive_queue
*rq
= &vi
->rq
[i
];
590 napi_disable(&rq
->napi
);
591 still_empty
= !try_fill_recv(rq
, GFP_KERNEL
);
592 virtnet_napi_enable(rq
);
594 /* In theory, this can happen: if we don't get any buffers in
595 * we will *never* try to fill again.
598 schedule_delayed_work(&vi
->refill
, HZ
/2);
602 static int virtnet_poll(struct napi_struct
*napi
, int budget
)
604 struct receive_queue
*rq
=
605 container_of(napi
, struct receive_queue
, napi
);
606 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
608 unsigned int r
, len
, received
= 0;
611 while (received
< budget
&&
612 (buf
= virtqueue_get_buf(rq
->vq
, &len
)) != NULL
) {
613 receive_buf(rq
, buf
, len
);
618 if (rq
->num
< rq
->max
/ 2) {
619 if (!try_fill_recv(rq
, GFP_ATOMIC
))
620 schedule_delayed_work(&vi
->refill
, 0);
623 /* Out of packets? */
624 if (received
< budget
) {
625 r
= virtqueue_enable_cb_prepare(rq
->vq
);
627 if (unlikely(virtqueue_poll(rq
->vq
, r
)) &&
628 napi_schedule_prep(napi
)) {
629 virtqueue_disable_cb(rq
->vq
);
630 __napi_schedule(napi
);
638 static int virtnet_open(struct net_device
*dev
)
640 struct virtnet_info
*vi
= netdev_priv(dev
);
643 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
644 if (i
< vi
->curr_queue_pairs
)
645 /* Make sure we have some buffers: if oom use wq. */
646 if (!try_fill_recv(&vi
->rq
[i
], GFP_KERNEL
))
647 schedule_delayed_work(&vi
->refill
, 0);
648 virtnet_napi_enable(&vi
->rq
[i
]);
654 static void free_old_xmit_skbs(struct send_queue
*sq
)
658 struct virtnet_info
*vi
= sq
->vq
->vdev
->priv
;
659 struct virtnet_stats
*stats
= this_cpu_ptr(vi
->stats
);
661 while ((skb
= virtqueue_get_buf(sq
->vq
, &len
)) != NULL
) {
662 pr_debug("Sent skb %p\n", skb
);
664 u64_stats_update_begin(&stats
->tx_syncp
);
665 stats
->tx_bytes
+= skb
->len
;
667 u64_stats_update_end(&stats
->tx_syncp
);
669 dev_kfree_skb_any(skb
);
673 static int xmit_skb(struct send_queue
*sq
, struct sk_buff
*skb
)
675 struct skb_vnet_hdr
*hdr
;
676 const unsigned char *dest
= ((struct ethhdr
*)skb
->data
)->h_dest
;
677 struct virtnet_info
*vi
= sq
->vq
->vdev
->priv
;
682 pr_debug("%s: xmit %p %pM\n", vi
->dev
->name
, skb
, dest
);
683 if (vi
->mergeable_rx_bufs
)
684 hdr_len
= sizeof hdr
->mhdr
;
686 hdr_len
= sizeof hdr
->hdr
;
688 can_push
= vi
->any_header_sg
&&
689 !((unsigned long)skb
->data
& (__alignof__(*hdr
) - 1)) &&
690 !skb_header_cloned(skb
) && skb_headroom(skb
) >= hdr_len
;
691 /* Even if we can, don't push here yet as this would skew
692 * csum_start offset below. */
694 hdr
= (struct skb_vnet_hdr
*)(skb
->data
- hdr_len
);
696 hdr
= skb_vnet_hdr(skb
);
698 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
699 hdr
->hdr
.flags
= VIRTIO_NET_HDR_F_NEEDS_CSUM
;
700 hdr
->hdr
.csum_start
= skb_checksum_start_offset(skb
);
701 hdr
->hdr
.csum_offset
= skb
->csum_offset
;
704 hdr
->hdr
.csum_offset
= hdr
->hdr
.csum_start
= 0;
707 if (skb_is_gso(skb
)) {
708 hdr
->hdr
.hdr_len
= skb_headlen(skb
);
709 hdr
->hdr
.gso_size
= skb_shinfo(skb
)->gso_size
;
710 if (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCPV4
)
711 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV4
;
712 else if (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCPV6
)
713 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV6
;
714 else if (skb_shinfo(skb
)->gso_type
& SKB_GSO_UDP
)
715 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_UDP
;
718 if (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCP_ECN
)
719 hdr
->hdr
.gso_type
|= VIRTIO_NET_HDR_GSO_ECN
;
721 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
722 hdr
->hdr
.gso_size
= hdr
->hdr
.hdr_len
= 0;
725 if (vi
->mergeable_rx_bufs
)
726 hdr
->mhdr
.num_buffers
= 0;
729 __skb_push(skb
, hdr_len
);
730 num_sg
= skb_to_sgvec(skb
, sq
->sg
, 0, skb
->len
);
731 /* Pull header back to avoid skew in tx bytes calculations. */
732 __skb_pull(skb
, hdr_len
);
734 sg_set_buf(sq
->sg
, hdr
, hdr_len
);
735 num_sg
= skb_to_sgvec(skb
, sq
->sg
+ 1, 0, skb
->len
) + 1;
737 return virtqueue_add_outbuf(sq
->vq
, sq
->sg
, num_sg
, skb
, GFP_ATOMIC
);
740 static netdev_tx_t
start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
742 struct virtnet_info
*vi
= netdev_priv(dev
);
743 int qnum
= skb_get_queue_mapping(skb
);
744 struct send_queue
*sq
= &vi
->sq
[qnum
];
747 /* Free up any pending old buffers before queueing new ones. */
748 free_old_xmit_skbs(sq
);
750 /* Try to transmit */
751 err
= xmit_skb(sq
, skb
);
753 /* This should not happen! */
755 dev
->stats
.tx_fifo_errors
++;
758 "Unexpected TXQ (%d) queue failure: %d\n", qnum
, err
);
759 dev
->stats
.tx_dropped
++;
763 virtqueue_kick(sq
->vq
);
765 /* Don't wait up for transmitted skbs to be freed. */
769 /* Apparently nice girls don't return TX_BUSY; stop the queue
770 * before it gets out of hand. Naturally, this wastes entries. */
771 if (sq
->vq
->num_free
< 2+MAX_SKB_FRAGS
) {
772 netif_stop_subqueue(dev
, qnum
);
773 if (unlikely(!virtqueue_enable_cb_delayed(sq
->vq
))) {
774 /* More just got used, free them then recheck. */
775 free_old_xmit_skbs(sq
);
776 if (sq
->vq
->num_free
>= 2+MAX_SKB_FRAGS
) {
777 netif_start_subqueue(dev
, qnum
);
778 virtqueue_disable_cb(sq
->vq
);
787 * Send command via the control virtqueue and check status. Commands
788 * supported by the hypervisor, as indicated by feature bits, should
789 * never fail unless improperly formated.
791 static bool virtnet_send_command(struct virtnet_info
*vi
, u8
class, u8 cmd
,
792 struct scatterlist
*out
,
793 struct scatterlist
*in
)
795 struct scatterlist
*sgs
[4], hdr
, stat
;
796 struct virtio_net_ctrl_hdr ctrl
;
797 virtio_net_ctrl_ack status
= ~0;
798 unsigned out_num
= 0, in_num
= 0, tmp
;
800 /* Caller should know better */
801 BUG_ON(!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
));
806 sg_init_one(&hdr
, &ctrl
, sizeof(ctrl
));
807 sgs
[out_num
++] = &hdr
;
810 sgs
[out_num
++] = out
;
812 sgs
[out_num
+ in_num
++] = in
;
814 /* Add return status. */
815 sg_init_one(&stat
, &status
, sizeof(status
));
816 sgs
[out_num
+ in_num
++] = &stat
;
818 BUG_ON(out_num
+ in_num
> ARRAY_SIZE(sgs
));
819 BUG_ON(virtqueue_add_sgs(vi
->cvq
, sgs
, out_num
, in_num
, vi
, GFP_ATOMIC
)
822 virtqueue_kick(vi
->cvq
);
824 /* Spin for a response, the kick causes an ioport write, trapping
825 * into the hypervisor, so the request should be handled immediately.
827 while (!virtqueue_get_buf(vi
->cvq
, &tmp
))
830 return status
== VIRTIO_NET_OK
;
833 static int virtnet_set_mac_address(struct net_device
*dev
, void *p
)
835 struct virtnet_info
*vi
= netdev_priv(dev
);
836 struct virtio_device
*vdev
= vi
->vdev
;
838 struct sockaddr
*addr
= p
;
839 struct scatterlist sg
;
841 ret
= eth_prepare_mac_addr_change(dev
, p
);
845 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_MAC_ADDR
)) {
846 sg_init_one(&sg
, addr
->sa_data
, dev
->addr_len
);
847 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MAC
,
848 VIRTIO_NET_CTRL_MAC_ADDR_SET
,
851 "Failed to set mac address by vq command.\n");
854 } else if (virtio_has_feature(vdev
, VIRTIO_NET_F_MAC
)) {
855 vdev
->config
->set(vdev
, offsetof(struct virtio_net_config
, mac
),
856 addr
->sa_data
, dev
->addr_len
);
859 eth_commit_mac_addr_change(dev
, p
);
864 static struct rtnl_link_stats64
*virtnet_stats(struct net_device
*dev
,
865 struct rtnl_link_stats64
*tot
)
867 struct virtnet_info
*vi
= netdev_priv(dev
);
871 for_each_possible_cpu(cpu
) {
872 struct virtnet_stats
*stats
= per_cpu_ptr(vi
->stats
, cpu
);
873 u64 tpackets
, tbytes
, rpackets
, rbytes
;
876 start
= u64_stats_fetch_begin_bh(&stats
->tx_syncp
);
877 tpackets
= stats
->tx_packets
;
878 tbytes
= stats
->tx_bytes
;
879 } while (u64_stats_fetch_retry_bh(&stats
->tx_syncp
, start
));
882 start
= u64_stats_fetch_begin_bh(&stats
->rx_syncp
);
883 rpackets
= stats
->rx_packets
;
884 rbytes
= stats
->rx_bytes
;
885 } while (u64_stats_fetch_retry_bh(&stats
->rx_syncp
, start
));
887 tot
->rx_packets
+= rpackets
;
888 tot
->tx_packets
+= tpackets
;
889 tot
->rx_bytes
+= rbytes
;
890 tot
->tx_bytes
+= tbytes
;
893 tot
->tx_dropped
= dev
->stats
.tx_dropped
;
894 tot
->tx_fifo_errors
= dev
->stats
.tx_fifo_errors
;
895 tot
->rx_dropped
= dev
->stats
.rx_dropped
;
896 tot
->rx_length_errors
= dev
->stats
.rx_length_errors
;
897 tot
->rx_frame_errors
= dev
->stats
.rx_frame_errors
;
902 #ifdef CONFIG_NET_POLL_CONTROLLER
903 static void virtnet_netpoll(struct net_device
*dev
)
905 struct virtnet_info
*vi
= netdev_priv(dev
);
908 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++)
909 napi_schedule(&vi
->rq
[i
].napi
);
913 static void virtnet_ack_link_announce(struct virtnet_info
*vi
)
916 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_ANNOUNCE
,
917 VIRTIO_NET_CTRL_ANNOUNCE_ACK
, NULL
, NULL
))
918 dev_warn(&vi
->dev
->dev
, "Failed to ack link announce.\n");
922 static int virtnet_set_queues(struct virtnet_info
*vi
, u16 queue_pairs
)
924 struct scatterlist sg
;
925 struct virtio_net_ctrl_mq s
;
926 struct net_device
*dev
= vi
->dev
;
928 if (!vi
->has_cvq
|| !virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_MQ
))
931 s
.virtqueue_pairs
= queue_pairs
;
932 sg_init_one(&sg
, &s
, sizeof(s
));
934 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MQ
,
935 VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET
, &sg
, NULL
)) {
936 dev_warn(&dev
->dev
, "Fail to set num of queue pairs to %d\n",
940 vi
->curr_queue_pairs
= queue_pairs
;
941 /* virtnet_open() will refill when device is going to up. */
942 if (dev
->flags
& IFF_UP
)
943 schedule_delayed_work(&vi
->refill
, 0);
949 static int virtnet_close(struct net_device
*dev
)
951 struct virtnet_info
*vi
= netdev_priv(dev
);
954 /* Make sure refill_work doesn't re-enable napi! */
955 cancel_delayed_work_sync(&vi
->refill
);
957 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
958 napi_disable(&vi
->rq
[i
].napi
);
963 static void virtnet_set_rx_mode(struct net_device
*dev
)
965 struct virtnet_info
*vi
= netdev_priv(dev
);
966 struct scatterlist sg
[2];
967 u8 promisc
, allmulti
;
968 struct virtio_net_ctrl_mac
*mac_data
;
969 struct netdev_hw_addr
*ha
;
975 /* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */
976 if (!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_RX
))
979 promisc
= ((dev
->flags
& IFF_PROMISC
) != 0);
980 allmulti
= ((dev
->flags
& IFF_ALLMULTI
) != 0);
982 sg_init_one(sg
, &promisc
, sizeof(promisc
));
984 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_RX
,
985 VIRTIO_NET_CTRL_RX_PROMISC
,
987 dev_warn(&dev
->dev
, "Failed to %sable promisc mode.\n",
988 promisc
? "en" : "dis");
990 sg_init_one(sg
, &allmulti
, sizeof(allmulti
));
992 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_RX
,
993 VIRTIO_NET_CTRL_RX_ALLMULTI
,
995 dev_warn(&dev
->dev
, "Failed to %sable allmulti mode.\n",
996 allmulti
? "en" : "dis");
998 uc_count
= netdev_uc_count(dev
);
999 mc_count
= netdev_mc_count(dev
);
1000 /* MAC filter - use one buffer for both lists */
1001 buf
= kzalloc(((uc_count
+ mc_count
) * ETH_ALEN
) +
1002 (2 * sizeof(mac_data
->entries
)), GFP_ATOMIC
);
1007 sg_init_table(sg
, 2);
1009 /* Store the unicast list and count in the front of the buffer */
1010 mac_data
->entries
= uc_count
;
1012 netdev_for_each_uc_addr(ha
, dev
)
1013 memcpy(&mac_data
->macs
[i
++][0], ha
->addr
, ETH_ALEN
);
1015 sg_set_buf(&sg
[0], mac_data
,
1016 sizeof(mac_data
->entries
) + (uc_count
* ETH_ALEN
));
1018 /* multicast list and count fill the end */
1019 mac_data
= (void *)&mac_data
->macs
[uc_count
][0];
1021 mac_data
->entries
= mc_count
;
1023 netdev_for_each_mc_addr(ha
, dev
)
1024 memcpy(&mac_data
->macs
[i
++][0], ha
->addr
, ETH_ALEN
);
1026 sg_set_buf(&sg
[1], mac_data
,
1027 sizeof(mac_data
->entries
) + (mc_count
* ETH_ALEN
));
1029 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MAC
,
1030 VIRTIO_NET_CTRL_MAC_TABLE_SET
,
1032 dev_warn(&dev
->dev
, "Failed to set MAC fitler table.\n");
1037 static int virtnet_vlan_rx_add_vid(struct net_device
*dev
,
1038 __be16 proto
, u16 vid
)
1040 struct virtnet_info
*vi
= netdev_priv(dev
);
1041 struct scatterlist sg
;
1043 sg_init_one(&sg
, &vid
, sizeof(vid
));
1045 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_VLAN
,
1046 VIRTIO_NET_CTRL_VLAN_ADD
, &sg
, NULL
))
1047 dev_warn(&dev
->dev
, "Failed to add VLAN ID %d.\n", vid
);
1051 static int virtnet_vlan_rx_kill_vid(struct net_device
*dev
,
1052 __be16 proto
, u16 vid
)
1054 struct virtnet_info
*vi
= netdev_priv(dev
);
1055 struct scatterlist sg
;
1057 sg_init_one(&sg
, &vid
, sizeof(vid
));
1059 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_VLAN
,
1060 VIRTIO_NET_CTRL_VLAN_DEL
, &sg
, NULL
))
1061 dev_warn(&dev
->dev
, "Failed to kill VLAN ID %d.\n", vid
);
1065 static void virtnet_clean_affinity(struct virtnet_info
*vi
, long hcpu
)
1070 if (vi
->affinity_hint_set
) {
1071 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1072 virtqueue_set_affinity(vi
->rq
[i
].vq
, -1);
1073 virtqueue_set_affinity(vi
->sq
[i
].vq
, -1);
1076 vi
->affinity_hint_set
= false;
1080 for_each_online_cpu(cpu
) {
1082 *per_cpu_ptr(vi
->vq_index
, cpu
) = -1;
1084 *per_cpu_ptr(vi
->vq_index
, cpu
) =
1085 ++i
% vi
->curr_queue_pairs
;
1090 static void virtnet_set_affinity(struct virtnet_info
*vi
)
1095 /* In multiqueue mode, when the number of cpu is equal to the number of
1096 * queue pairs, we let the queue pairs to be private to one cpu by
1097 * setting the affinity hint to eliminate the contention.
1099 if (vi
->curr_queue_pairs
== 1 ||
1100 vi
->max_queue_pairs
!= num_online_cpus()) {
1101 virtnet_clean_affinity(vi
, -1);
1106 for_each_online_cpu(cpu
) {
1107 virtqueue_set_affinity(vi
->rq
[i
].vq
, cpu
);
1108 virtqueue_set_affinity(vi
->sq
[i
].vq
, cpu
);
1109 *per_cpu_ptr(vi
->vq_index
, cpu
) = i
;
1113 vi
->affinity_hint_set
= true;
1116 static int virtnet_cpu_callback(struct notifier_block
*nfb
,
1117 unsigned long action
, void *hcpu
)
1119 struct virtnet_info
*vi
= container_of(nfb
, struct virtnet_info
, nb
);
1121 mutex_lock(&vi
->config_lock
);
1123 if (!vi
->config_enable
)
1126 switch(action
& ~CPU_TASKS_FROZEN
) {
1128 case CPU_DOWN_FAILED
:
1130 virtnet_set_affinity(vi
);
1132 case CPU_DOWN_PREPARE
:
1133 virtnet_clean_affinity(vi
, (long)hcpu
);
1140 mutex_unlock(&vi
->config_lock
);
1144 static void virtnet_get_ringparam(struct net_device
*dev
,
1145 struct ethtool_ringparam
*ring
)
1147 struct virtnet_info
*vi
= netdev_priv(dev
);
1149 ring
->rx_max_pending
= virtqueue_get_vring_size(vi
->rq
[0].vq
);
1150 ring
->tx_max_pending
= virtqueue_get_vring_size(vi
->sq
[0].vq
);
1151 ring
->rx_pending
= ring
->rx_max_pending
;
1152 ring
->tx_pending
= ring
->tx_max_pending
;
1156 static void virtnet_get_drvinfo(struct net_device
*dev
,
1157 struct ethtool_drvinfo
*info
)
1159 struct virtnet_info
*vi
= netdev_priv(dev
);
1160 struct virtio_device
*vdev
= vi
->vdev
;
1162 strlcpy(info
->driver
, KBUILD_MODNAME
, sizeof(info
->driver
));
1163 strlcpy(info
->version
, VIRTNET_DRIVER_VERSION
, sizeof(info
->version
));
1164 strlcpy(info
->bus_info
, virtio_bus_name(vdev
), sizeof(info
->bus_info
));
1168 /* TODO: Eliminate OOO packets during switching */
1169 static int virtnet_set_channels(struct net_device
*dev
,
1170 struct ethtool_channels
*channels
)
1172 struct virtnet_info
*vi
= netdev_priv(dev
);
1173 u16 queue_pairs
= channels
->combined_count
;
1176 /* We don't support separate rx/tx channels.
1177 * We don't allow setting 'other' channels.
1179 if (channels
->rx_count
|| channels
->tx_count
|| channels
->other_count
)
1182 if (queue_pairs
> vi
->max_queue_pairs
)
1186 err
= virtnet_set_queues(vi
, queue_pairs
);
1188 netif_set_real_num_tx_queues(dev
, queue_pairs
);
1189 netif_set_real_num_rx_queues(dev
, queue_pairs
);
1191 virtnet_set_affinity(vi
);
1198 static void virtnet_get_channels(struct net_device
*dev
,
1199 struct ethtool_channels
*channels
)
1201 struct virtnet_info
*vi
= netdev_priv(dev
);
1203 channels
->combined_count
= vi
->curr_queue_pairs
;
1204 channels
->max_combined
= vi
->max_queue_pairs
;
1205 channels
->max_other
= 0;
1206 channels
->rx_count
= 0;
1207 channels
->tx_count
= 0;
1208 channels
->other_count
= 0;
1211 static const struct ethtool_ops virtnet_ethtool_ops
= {
1212 .get_drvinfo
= virtnet_get_drvinfo
,
1213 .get_link
= ethtool_op_get_link
,
1214 .get_ringparam
= virtnet_get_ringparam
,
1215 .set_channels
= virtnet_set_channels
,
1216 .get_channels
= virtnet_get_channels
,
1220 #define MAX_MTU 65535
1222 static int virtnet_change_mtu(struct net_device
*dev
, int new_mtu
)
1224 if (new_mtu
< MIN_MTU
|| new_mtu
> MAX_MTU
)
1230 /* To avoid contending a lock hold by a vcpu who would exit to host, select the
1231 * txq based on the processor id.
1233 static u16
virtnet_select_queue(struct net_device
*dev
, struct sk_buff
*skb
)
1236 struct virtnet_info
*vi
= netdev_priv(dev
);
1238 if (skb_rx_queue_recorded(skb
)) {
1239 txq
= skb_get_rx_queue(skb
);
1241 txq
= *__this_cpu_ptr(vi
->vq_index
);
1246 while (unlikely(txq
>= dev
->real_num_tx_queues
))
1247 txq
-= dev
->real_num_tx_queues
;
1252 static const struct net_device_ops virtnet_netdev
= {
1253 .ndo_open
= virtnet_open
,
1254 .ndo_stop
= virtnet_close
,
1255 .ndo_start_xmit
= start_xmit
,
1256 .ndo_validate_addr
= eth_validate_addr
,
1257 .ndo_set_mac_address
= virtnet_set_mac_address
,
1258 .ndo_set_rx_mode
= virtnet_set_rx_mode
,
1259 .ndo_change_mtu
= virtnet_change_mtu
,
1260 .ndo_get_stats64
= virtnet_stats
,
1261 .ndo_vlan_rx_add_vid
= virtnet_vlan_rx_add_vid
,
1262 .ndo_vlan_rx_kill_vid
= virtnet_vlan_rx_kill_vid
,
1263 .ndo_select_queue
= virtnet_select_queue
,
1264 #ifdef CONFIG_NET_POLL_CONTROLLER
1265 .ndo_poll_controller
= virtnet_netpoll
,
1269 static void virtnet_config_changed_work(struct work_struct
*work
)
1271 struct virtnet_info
*vi
=
1272 container_of(work
, struct virtnet_info
, config_work
);
1275 mutex_lock(&vi
->config_lock
);
1276 if (!vi
->config_enable
)
1279 if (virtio_config_val(vi
->vdev
, VIRTIO_NET_F_STATUS
,
1280 offsetof(struct virtio_net_config
, status
),
1284 if (v
& VIRTIO_NET_S_ANNOUNCE
) {
1285 netdev_notify_peers(vi
->dev
);
1286 virtnet_ack_link_announce(vi
);
1289 /* Ignore unknown (future) status bits */
1290 v
&= VIRTIO_NET_S_LINK_UP
;
1292 if (vi
->status
== v
)
1297 if (vi
->status
& VIRTIO_NET_S_LINK_UP
) {
1298 netif_carrier_on(vi
->dev
);
1299 netif_tx_wake_all_queues(vi
->dev
);
1301 netif_carrier_off(vi
->dev
);
1302 netif_tx_stop_all_queues(vi
->dev
);
1305 mutex_unlock(&vi
->config_lock
);
1308 static void virtnet_config_changed(struct virtio_device
*vdev
)
1310 struct virtnet_info
*vi
= vdev
->priv
;
1312 schedule_work(&vi
->config_work
);
1315 static void virtnet_free_queues(struct virtnet_info
*vi
)
1321 static void free_receive_bufs(struct virtnet_info
*vi
)
1325 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1326 while (vi
->rq
[i
].pages
)
1327 __free_pages(get_a_page(&vi
->rq
[i
], GFP_KERNEL
), 0);
1331 static void free_unused_bufs(struct virtnet_info
*vi
)
1336 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1337 struct virtqueue
*vq
= vi
->sq
[i
].vq
;
1338 while ((buf
= virtqueue_detach_unused_buf(vq
)) != NULL
)
1342 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1343 struct virtqueue
*vq
= vi
->rq
[i
].vq
;
1345 while ((buf
= virtqueue_detach_unused_buf(vq
)) != NULL
) {
1346 if (vi
->mergeable_rx_bufs
|| vi
->big_packets
)
1347 give_pages(&vi
->rq
[i
], buf
);
1352 BUG_ON(vi
->rq
[i
].num
!= 0);
1356 static void virtnet_del_vqs(struct virtnet_info
*vi
)
1358 struct virtio_device
*vdev
= vi
->vdev
;
1360 virtnet_clean_affinity(vi
, -1);
1362 vdev
->config
->del_vqs(vdev
);
1364 virtnet_free_queues(vi
);
1367 static int virtnet_find_vqs(struct virtnet_info
*vi
)
1369 vq_callback_t
**callbacks
;
1370 struct virtqueue
**vqs
;
1375 /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
1376 * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
1377 * possible control vq.
1379 total_vqs
= vi
->max_queue_pairs
* 2 +
1380 virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
);
1382 /* Allocate space for find_vqs parameters */
1383 vqs
= kzalloc(total_vqs
* sizeof(*vqs
), GFP_KERNEL
);
1386 callbacks
= kmalloc(total_vqs
* sizeof(*callbacks
), GFP_KERNEL
);
1389 names
= kmalloc(total_vqs
* sizeof(*names
), GFP_KERNEL
);
1393 /* Parameters for control virtqueue, if any */
1395 callbacks
[total_vqs
- 1] = NULL
;
1396 names
[total_vqs
- 1] = "control";
1399 /* Allocate/initialize parameters for send/receive virtqueues */
1400 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1401 callbacks
[rxq2vq(i
)] = skb_recv_done
;
1402 callbacks
[txq2vq(i
)] = skb_xmit_done
;
1403 sprintf(vi
->rq
[i
].name
, "input.%d", i
);
1404 sprintf(vi
->sq
[i
].name
, "output.%d", i
);
1405 names
[rxq2vq(i
)] = vi
->rq
[i
].name
;
1406 names
[txq2vq(i
)] = vi
->sq
[i
].name
;
1409 ret
= vi
->vdev
->config
->find_vqs(vi
->vdev
, total_vqs
, vqs
, callbacks
,
1415 vi
->cvq
= vqs
[total_vqs
- 1];
1416 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VLAN
))
1417 vi
->dev
->features
|= NETIF_F_HW_VLAN_CTAG_FILTER
;
1420 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1421 vi
->rq
[i
].vq
= vqs
[rxq2vq(i
)];
1422 vi
->sq
[i
].vq
= vqs
[txq2vq(i
)];
1441 static int virtnet_alloc_queues(struct virtnet_info
*vi
)
1445 vi
->sq
= kzalloc(sizeof(*vi
->sq
) * vi
->max_queue_pairs
, GFP_KERNEL
);
1448 vi
->rq
= kzalloc(sizeof(*vi
->rq
) * vi
->max_queue_pairs
, GFP_KERNEL
);
1452 INIT_DELAYED_WORK(&vi
->refill
, refill_work
);
1453 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1454 vi
->rq
[i
].pages
= NULL
;
1455 netif_napi_add(vi
->dev
, &vi
->rq
[i
].napi
, virtnet_poll
,
1458 sg_init_table(vi
->rq
[i
].sg
, ARRAY_SIZE(vi
->rq
[i
].sg
));
1459 sg_init_table(vi
->sq
[i
].sg
, ARRAY_SIZE(vi
->sq
[i
].sg
));
1470 static int init_vqs(struct virtnet_info
*vi
)
1474 /* Allocate send & receive queues */
1475 ret
= virtnet_alloc_queues(vi
);
1479 ret
= virtnet_find_vqs(vi
);
1484 virtnet_set_affinity(vi
);
1490 virtnet_free_queues(vi
);
1495 static int virtnet_probe(struct virtio_device
*vdev
)
1498 struct net_device
*dev
;
1499 struct virtnet_info
*vi
;
1500 u16 max_queue_pairs
;
1502 /* Find if host supports multiqueue virtio_net device */
1503 err
= virtio_config_val(vdev
, VIRTIO_NET_F_MQ
,
1504 offsetof(struct virtio_net_config
,
1505 max_virtqueue_pairs
), &max_queue_pairs
);
1507 /* We need at least 2 queue's */
1508 if (err
|| max_queue_pairs
< VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN
||
1509 max_queue_pairs
> VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX
||
1510 !virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
))
1511 max_queue_pairs
= 1;
1513 /* Allocate ourselves a network device with room for our info */
1514 dev
= alloc_etherdev_mq(sizeof(struct virtnet_info
), max_queue_pairs
);
1518 /* Set up network device as normal. */
1519 dev
->priv_flags
|= IFF_UNICAST_FLT
| IFF_LIVE_ADDR_CHANGE
;
1520 dev
->netdev_ops
= &virtnet_netdev
;
1521 dev
->features
= NETIF_F_HIGHDMA
;
1523 SET_ETHTOOL_OPS(dev
, &virtnet_ethtool_ops
);
1524 SET_NETDEV_DEV(dev
, &vdev
->dev
);
1526 /* Do we support "hardware" checksums? */
1527 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CSUM
)) {
1528 /* This opens up the world of extra features. */
1529 dev
->hw_features
|= NETIF_F_HW_CSUM
|NETIF_F_SG
|NETIF_F_FRAGLIST
;
1531 dev
->features
|= NETIF_F_HW_CSUM
|NETIF_F_SG
|NETIF_F_FRAGLIST
;
1533 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GSO
)) {
1534 dev
->hw_features
|= NETIF_F_TSO
| NETIF_F_UFO
1535 | NETIF_F_TSO_ECN
| NETIF_F_TSO6
;
1537 /* Individual feature bits: what can host handle? */
1538 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_TSO4
))
1539 dev
->hw_features
|= NETIF_F_TSO
;
1540 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_TSO6
))
1541 dev
->hw_features
|= NETIF_F_TSO6
;
1542 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_ECN
))
1543 dev
->hw_features
|= NETIF_F_TSO_ECN
;
1544 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_UFO
))
1545 dev
->hw_features
|= NETIF_F_UFO
;
1548 dev
->features
|= dev
->hw_features
& (NETIF_F_ALL_TSO
|NETIF_F_UFO
);
1549 /* (!csum && gso) case will be fixed by register_netdev() */
1551 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_CSUM
))
1552 dev
->features
|= NETIF_F_RXCSUM
;
1554 dev
->vlan_features
= dev
->features
;
1556 /* Configuration may specify what MAC to use. Otherwise random. */
1557 if (virtio_config_val_len(vdev
, VIRTIO_NET_F_MAC
,
1558 offsetof(struct virtio_net_config
, mac
),
1559 dev
->dev_addr
, dev
->addr_len
) < 0)
1560 eth_hw_addr_random(dev
);
1562 /* Set up our device-specific information */
1563 vi
= netdev_priv(dev
);
1567 vi
->stats
= alloc_percpu(struct virtnet_stats
);
1569 if (vi
->stats
== NULL
)
1572 vi
->vq_index
= alloc_percpu(int);
1573 if (vi
->vq_index
== NULL
)
1576 mutex_init(&vi
->config_lock
);
1577 vi
->config_enable
= true;
1578 INIT_WORK(&vi
->config_work
, virtnet_config_changed_work
);
1580 /* If we can receive ANY GSO packets, we must allocate large ones. */
1581 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_TSO4
) ||
1582 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_TSO6
) ||
1583 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_ECN
))
1584 vi
->big_packets
= true;
1586 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MRG_RXBUF
))
1587 vi
->mergeable_rx_bufs
= true;
1589 if (virtio_has_feature(vdev
, VIRTIO_F_ANY_LAYOUT
))
1590 vi
->any_header_sg
= true;
1592 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
))
1595 /* Use single tx/rx queue pair as default */
1596 vi
->curr_queue_pairs
= 1;
1597 vi
->max_queue_pairs
= max_queue_pairs
;
1599 /* Allocate/initialize the rx/tx queues, and invoke find_vqs */
1604 netif_set_real_num_tx_queues(dev
, 1);
1605 netif_set_real_num_rx_queues(dev
, 1);
1607 err
= register_netdev(dev
);
1609 pr_debug("virtio_net: registering device failed\n");
1613 /* Last of all, set up some receive buffers. */
1614 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++) {
1615 try_fill_recv(&vi
->rq
[i
], GFP_KERNEL
);
1617 /* If we didn't even get one input buffer, we're useless. */
1618 if (vi
->rq
[i
].num
== 0) {
1619 free_unused_bufs(vi
);
1621 goto free_recv_bufs
;
1625 vi
->nb
.notifier_call
= &virtnet_cpu_callback
;
1626 err
= register_hotcpu_notifier(&vi
->nb
);
1628 pr_debug("virtio_net: registering cpu notifier failed\n");
1629 goto free_recv_bufs
;
1632 /* Assume link up if device can't report link status,
1633 otherwise get link status from config. */
1634 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_STATUS
)) {
1635 netif_carrier_off(dev
);
1636 schedule_work(&vi
->config_work
);
1638 vi
->status
= VIRTIO_NET_S_LINK_UP
;
1639 netif_carrier_on(dev
);
1642 pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
1643 dev
->name
, max_queue_pairs
);
1648 free_receive_bufs(vi
);
1649 unregister_netdev(dev
);
1651 cancel_delayed_work_sync(&vi
->refill
);
1652 virtnet_del_vqs(vi
);
1654 free_percpu(vi
->vq_index
);
1656 free_percpu(vi
->stats
);
1662 static void remove_vq_common(struct virtnet_info
*vi
)
1664 vi
->vdev
->config
->reset(vi
->vdev
);
1666 /* Free unused buffers in both send and recv, if any. */
1667 free_unused_bufs(vi
);
1669 free_receive_bufs(vi
);
1671 virtnet_del_vqs(vi
);
1674 static void virtnet_remove(struct virtio_device
*vdev
)
1676 struct virtnet_info
*vi
= vdev
->priv
;
1678 unregister_hotcpu_notifier(&vi
->nb
);
1680 /* Prevent config work handler from accessing the device. */
1681 mutex_lock(&vi
->config_lock
);
1682 vi
->config_enable
= false;
1683 mutex_unlock(&vi
->config_lock
);
1685 unregister_netdev(vi
->dev
);
1687 remove_vq_common(vi
);
1689 flush_work(&vi
->config_work
);
1691 free_percpu(vi
->vq_index
);
1692 free_percpu(vi
->stats
);
1693 free_netdev(vi
->dev
);
1697 static int virtnet_freeze(struct virtio_device
*vdev
)
1699 struct virtnet_info
*vi
= vdev
->priv
;
1702 /* Prevent config work handler from accessing the device */
1703 mutex_lock(&vi
->config_lock
);
1704 vi
->config_enable
= false;
1705 mutex_unlock(&vi
->config_lock
);
1707 netif_device_detach(vi
->dev
);
1708 cancel_delayed_work_sync(&vi
->refill
);
1710 if (netif_running(vi
->dev
))
1711 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1712 napi_disable(&vi
->rq
[i
].napi
);
1713 netif_napi_del(&vi
->rq
[i
].napi
);
1716 remove_vq_common(vi
);
1718 flush_work(&vi
->config_work
);
1723 static int virtnet_restore(struct virtio_device
*vdev
)
1725 struct virtnet_info
*vi
= vdev
->priv
;
1732 if (netif_running(vi
->dev
))
1733 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
1734 virtnet_napi_enable(&vi
->rq
[i
]);
1736 netif_device_attach(vi
->dev
);
1738 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++)
1739 if (!try_fill_recv(&vi
->rq
[i
], GFP_KERNEL
))
1740 schedule_delayed_work(&vi
->refill
, 0);
1742 mutex_lock(&vi
->config_lock
);
1743 vi
->config_enable
= true;
1744 mutex_unlock(&vi
->config_lock
);
1747 virtnet_set_queues(vi
, vi
->curr_queue_pairs
);
1754 static struct virtio_device_id id_table
[] = {
1755 { VIRTIO_ID_NET
, VIRTIO_DEV_ANY_ID
},
1759 static unsigned int features
[] = {
1760 VIRTIO_NET_F_CSUM
, VIRTIO_NET_F_GUEST_CSUM
,
1761 VIRTIO_NET_F_GSO
, VIRTIO_NET_F_MAC
,
1762 VIRTIO_NET_F_HOST_TSO4
, VIRTIO_NET_F_HOST_UFO
, VIRTIO_NET_F_HOST_TSO6
,
1763 VIRTIO_NET_F_HOST_ECN
, VIRTIO_NET_F_GUEST_TSO4
, VIRTIO_NET_F_GUEST_TSO6
,
1764 VIRTIO_NET_F_GUEST_ECN
, VIRTIO_NET_F_GUEST_UFO
,
1765 VIRTIO_NET_F_MRG_RXBUF
, VIRTIO_NET_F_STATUS
, VIRTIO_NET_F_CTRL_VQ
,
1766 VIRTIO_NET_F_CTRL_RX
, VIRTIO_NET_F_CTRL_VLAN
,
1767 VIRTIO_NET_F_GUEST_ANNOUNCE
, VIRTIO_NET_F_MQ
,
1768 VIRTIO_NET_F_CTRL_MAC_ADDR
,
1769 VIRTIO_F_ANY_LAYOUT
,
1772 static struct virtio_driver virtio_net_driver
= {
1773 .feature_table
= features
,
1774 .feature_table_size
= ARRAY_SIZE(features
),
1775 .driver
.name
= KBUILD_MODNAME
,
1776 .driver
.owner
= THIS_MODULE
,
1777 .id_table
= id_table
,
1778 .probe
= virtnet_probe
,
1779 .remove
= virtnet_remove
,
1780 .config_changed
= virtnet_config_changed
,
1782 .freeze
= virtnet_freeze
,
1783 .restore
= virtnet_restore
,
1787 module_virtio_driver(virtio_net_driver
);
1789 MODULE_DEVICE_TABLE(virtio
, id_table
);
1790 MODULE_DESCRIPTION("Virtio network driver");
1791 MODULE_LICENSE("GPL");