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>
29 static int napi_weight
= 128;
30 module_param(napi_weight
, int, 0444);
32 static int csum
= 1, gso
= 1;
33 module_param(csum
, bool, 0444);
34 module_param(gso
, bool, 0444);
36 /* FIXME: MTU in config. */
37 #define MAX_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
38 #define GOOD_COPY_LEN 128
40 #define VIRTNET_SEND_COMMAND_SG_MAX 2
44 struct virtio_device
*vdev
;
45 struct virtqueue
*rvq
, *svq
, *cvq
;
46 struct net_device
*dev
;
47 struct napi_struct napi
;
50 /* Number of input buffers, and max we've ever had. */
51 unsigned int num
, max
;
53 /* I like... big packets and I cannot lie! */
56 /* Host will merge rx buffers for big packets (shake it! shake it!) */
57 bool mergeable_rx_bufs
;
59 /* Work struct for refilling if we run low on memory. */
60 struct delayed_work refill
;
62 /* Chain pages by the private ptr. */
68 struct virtio_net_hdr hdr
;
69 struct virtio_net_hdr_mrg_rxbuf mhdr
;
74 struct padded_vnet_hdr
{
75 struct virtio_net_hdr hdr
;
77 * virtio_net_hdr should be in a separated sg buffer because of a
78 * QEMU bug, and data sg buffer shares same page with this header sg.
79 * This padding makes next sg 16 byte aligned after virtio_net_hdr.
84 static inline struct skb_vnet_hdr
*skb_vnet_hdr(struct sk_buff
*skb
)
86 return (struct skb_vnet_hdr
*)skb
->cb
;
90 * private is used to chain pages for big packets, put the whole
91 * most recent used list in the beginning for reuse
93 static void give_pages(struct virtnet_info
*vi
, struct page
*page
)
97 /* Find end of list, sew whole thing into vi->pages. */
98 for (end
= page
; end
->private; end
= (struct page
*)end
->private);
99 end
->private = (unsigned long)vi
->pages
;
103 static struct page
*get_a_page(struct virtnet_info
*vi
, gfp_t gfp_mask
)
105 struct page
*p
= vi
->pages
;
108 vi
->pages
= (struct page
*)p
->private;
109 /* clear private here, it is used to chain pages */
112 p
= alloc_page(gfp_mask
);
116 static void skb_xmit_done(struct virtqueue
*svq
)
118 struct virtnet_info
*vi
= svq
->vdev
->priv
;
120 /* Suppress further interrupts. */
121 svq
->vq_ops
->disable_cb(svq
);
123 /* We were probably waiting for more output buffers. */
124 netif_wake_queue(vi
->dev
);
127 static void set_skb_frag(struct sk_buff
*skb
, struct page
*page
,
128 unsigned int offset
, unsigned int *len
)
130 int i
= skb_shinfo(skb
)->nr_frags
;
133 f
= &skb_shinfo(skb
)->frags
[i
];
134 f
->size
= min((unsigned)PAGE_SIZE
- offset
, *len
);
135 f
->page_offset
= offset
;
138 skb
->data_len
+= f
->size
;
140 skb_shinfo(skb
)->nr_frags
++;
144 static struct sk_buff
*page_to_skb(struct virtnet_info
*vi
,
145 struct page
*page
, unsigned int len
)
148 struct skb_vnet_hdr
*hdr
;
149 unsigned int copy
, hdr_len
, offset
;
152 p
= page_address(page
);
154 /* copy small packet so we can reuse these pages for small data */
155 skb
= netdev_alloc_skb_ip_align(vi
->dev
, GOOD_COPY_LEN
);
159 hdr
= skb_vnet_hdr(skb
);
161 if (vi
->mergeable_rx_bufs
) {
162 hdr_len
= sizeof hdr
->mhdr
;
165 hdr_len
= sizeof hdr
->hdr
;
166 offset
= sizeof(struct padded_vnet_hdr
);
169 memcpy(hdr
, p
, hdr_len
);
175 if (copy
> skb_tailroom(skb
))
176 copy
= skb_tailroom(skb
);
177 memcpy(skb_put(skb
, copy
), p
, copy
);
183 set_skb_frag(skb
, page
, offset
, &len
);
184 page
= (struct page
*)page
->private;
189 give_pages(vi
, page
);
194 static int receive_mergeable(struct virtnet_info
*vi
, struct sk_buff
*skb
)
196 struct skb_vnet_hdr
*hdr
= skb_vnet_hdr(skb
);
200 num_buf
= hdr
->mhdr
.num_buffers
;
202 i
= skb_shinfo(skb
)->nr_frags
;
203 if (i
>= MAX_SKB_FRAGS
) {
204 pr_debug("%s: packet too long\n", skb
->dev
->name
);
205 skb
->dev
->stats
.rx_length_errors
++;
209 page
= vi
->rvq
->vq_ops
->get_buf(vi
->rvq
, &len
);
211 pr_debug("%s: rx error: %d buffers missing\n",
212 skb
->dev
->name
, hdr
->mhdr
.num_buffers
);
213 skb
->dev
->stats
.rx_length_errors
++;
219 set_skb_frag(skb
, page
, 0, &len
);
226 static void receive_buf(struct net_device
*dev
, void *buf
, unsigned int len
)
228 struct virtnet_info
*vi
= netdev_priv(dev
);
231 struct skb_vnet_hdr
*hdr
;
233 if (unlikely(len
< sizeof(struct virtio_net_hdr
) + ETH_HLEN
)) {
234 pr_debug("%s: short packet %i\n", dev
->name
, len
);
235 dev
->stats
.rx_length_errors
++;
236 if (vi
->mergeable_rx_bufs
|| vi
->big_packets
)
243 if (!vi
->mergeable_rx_bufs
&& !vi
->big_packets
) {
245 len
-= sizeof(struct virtio_net_hdr
);
249 skb
= page_to_skb(vi
, page
, len
);
250 if (unlikely(!skb
)) {
251 dev
->stats
.rx_dropped
++;
252 give_pages(vi
, page
);
255 if (vi
->mergeable_rx_bufs
)
256 if (receive_mergeable(vi
, skb
)) {
262 hdr
= skb_vnet_hdr(skb
);
263 skb
->truesize
+= skb
->data_len
;
264 dev
->stats
.rx_bytes
+= skb
->len
;
265 dev
->stats
.rx_packets
++;
267 if (hdr
->hdr
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) {
268 pr_debug("Needs csum!\n");
269 if (!skb_partial_csum_set(skb
,
271 hdr
->hdr
.csum_offset
))
275 skb
->protocol
= eth_type_trans(skb
, dev
);
276 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
277 ntohs(skb
->protocol
), skb
->len
, skb
->pkt_type
);
279 if (hdr
->hdr
.gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
281 switch (hdr
->hdr
.gso_type
& ~VIRTIO_NET_HDR_GSO_ECN
) {
282 case VIRTIO_NET_HDR_GSO_TCPV4
:
283 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV4
;
285 case VIRTIO_NET_HDR_GSO_UDP
:
286 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
288 case VIRTIO_NET_HDR_GSO_TCPV6
:
289 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV6
;
293 printk(KERN_WARNING
"%s: bad gso type %u.\n",
294 dev
->name
, hdr
->hdr
.gso_type
);
298 if (hdr
->hdr
.gso_type
& VIRTIO_NET_HDR_GSO_ECN
)
299 skb_shinfo(skb
)->gso_type
|= SKB_GSO_TCP_ECN
;
301 skb_shinfo(skb
)->gso_size
= hdr
->hdr
.gso_size
;
302 if (skb_shinfo(skb
)->gso_size
== 0) {
304 printk(KERN_WARNING
"%s: zero gso size.\n",
309 /* Header must be checked, and gso_segs computed. */
310 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
311 skb_shinfo(skb
)->gso_segs
= 0;
314 netif_receive_skb(skb
);
318 dev
->stats
.rx_frame_errors
++;
322 static int add_recvbuf_small(struct virtnet_info
*vi
, gfp_t gfp
)
325 struct skb_vnet_hdr
*hdr
;
326 struct scatterlist sg
[2];
329 sg_init_table(sg
, 2);
330 skb
= netdev_alloc_skb_ip_align(vi
->dev
, MAX_PACKET_LEN
);
334 skb_put(skb
, MAX_PACKET_LEN
);
336 hdr
= skb_vnet_hdr(skb
);
337 sg_set_buf(sg
, &hdr
->hdr
, sizeof hdr
->hdr
);
339 skb_to_sgvec(skb
, sg
+ 1, 0, skb
->len
);
341 err
= vi
->rvq
->vq_ops
->add_buf(vi
->rvq
, sg
, 0, 2, skb
);
348 static int add_recvbuf_big(struct virtnet_info
*vi
, gfp_t gfp
)
350 struct scatterlist sg
[MAX_SKB_FRAGS
+ 2];
351 struct page
*first
, *list
= NULL
;
355 sg_init_table(sg
, MAX_SKB_FRAGS
+ 2);
356 /* page in sg[MAX_SKB_FRAGS + 1] is list tail */
357 for (i
= MAX_SKB_FRAGS
+ 1; i
> 1; --i
) {
358 first
= get_a_page(vi
, gfp
);
361 give_pages(vi
, list
);
364 sg_set_buf(&sg
[i
], page_address(first
), PAGE_SIZE
);
366 /* chain new page in list head to match sg */
367 first
->private = (unsigned long)list
;
371 first
= get_a_page(vi
, gfp
);
373 give_pages(vi
, list
);
376 p
= page_address(first
);
378 /* sg[0], sg[1] share the same page */
379 /* a separated sg[0] for virtio_net_hdr only during to QEMU bug*/
380 sg_set_buf(&sg
[0], p
, sizeof(struct virtio_net_hdr
));
382 /* sg[1] for data packet, from offset */
383 offset
= sizeof(struct padded_vnet_hdr
);
384 sg_set_buf(&sg
[1], p
+ offset
, PAGE_SIZE
- offset
);
386 /* chain first in list head */
387 first
->private = (unsigned long)list
;
388 err
= vi
->rvq
->vq_ops
->add_buf(vi
->rvq
, sg
, 0, MAX_SKB_FRAGS
+ 2,
391 give_pages(vi
, first
);
396 static int add_recvbuf_mergeable(struct virtnet_info
*vi
, gfp_t gfp
)
399 struct scatterlist sg
;
402 page
= get_a_page(vi
, gfp
);
406 sg_init_one(&sg
, page_address(page
), PAGE_SIZE
);
408 err
= vi
->rvq
->vq_ops
->add_buf(vi
->rvq
, &sg
, 0, 1, page
);
410 give_pages(vi
, page
);
415 /* Returns false if we couldn't fill entirely (OOM). */
416 static bool try_fill_recv(struct virtnet_info
*vi
, gfp_t gfp
)
422 if (vi
->mergeable_rx_bufs
)
423 err
= add_recvbuf_mergeable(vi
, gfp
);
424 else if (vi
->big_packets
)
425 err
= add_recvbuf_big(vi
, gfp
);
427 err
= add_recvbuf_small(vi
, gfp
);
435 if (unlikely(vi
->num
> vi
->max
))
437 vi
->rvq
->vq_ops
->kick(vi
->rvq
);
441 static void skb_recv_done(struct virtqueue
*rvq
)
443 struct virtnet_info
*vi
= rvq
->vdev
->priv
;
444 /* Schedule NAPI, Suppress further interrupts if successful. */
445 if (napi_schedule_prep(&vi
->napi
)) {
446 rvq
->vq_ops
->disable_cb(rvq
);
447 __napi_schedule(&vi
->napi
);
451 static void refill_work(struct work_struct
*work
)
453 struct virtnet_info
*vi
;
456 vi
= container_of(work
, struct virtnet_info
, refill
.work
);
457 napi_disable(&vi
->napi
);
458 still_empty
= !try_fill_recv(vi
, GFP_KERNEL
);
459 napi_enable(&vi
->napi
);
461 /* In theory, this can happen: if we don't get any buffers in
462 * we will *never* try to fill again. */
464 schedule_delayed_work(&vi
->refill
, HZ
/2);
467 static int virtnet_poll(struct napi_struct
*napi
, int budget
)
469 struct virtnet_info
*vi
= container_of(napi
, struct virtnet_info
, napi
);
471 unsigned int len
, received
= 0;
474 while (received
< budget
&&
475 (buf
= vi
->rvq
->vq_ops
->get_buf(vi
->rvq
, &len
)) != NULL
) {
476 receive_buf(vi
->dev
, buf
, len
);
481 if (vi
->num
< vi
->max
/ 2) {
482 if (!try_fill_recv(vi
, GFP_ATOMIC
))
483 schedule_delayed_work(&vi
->refill
, 0);
486 /* Out of packets? */
487 if (received
< budget
) {
489 if (unlikely(!vi
->rvq
->vq_ops
->enable_cb(vi
->rvq
)) &&
490 napi_schedule_prep(napi
)) {
491 vi
->rvq
->vq_ops
->disable_cb(vi
->rvq
);
492 __napi_schedule(napi
);
500 static unsigned int free_old_xmit_skbs(struct virtnet_info
*vi
)
503 unsigned int len
, tot_sgs
= 0;
505 while ((skb
= vi
->svq
->vq_ops
->get_buf(vi
->svq
, &len
)) != NULL
) {
506 pr_debug("Sent skb %p\n", skb
);
507 vi
->dev
->stats
.tx_bytes
+= skb
->len
;
508 vi
->dev
->stats
.tx_packets
++;
509 tot_sgs
+= skb_vnet_hdr(skb
)->num_sg
;
510 dev_kfree_skb_any(skb
);
515 static int xmit_skb(struct virtnet_info
*vi
, struct sk_buff
*skb
)
517 struct scatterlist sg
[2+MAX_SKB_FRAGS
];
518 struct skb_vnet_hdr
*hdr
= skb_vnet_hdr(skb
);
519 const unsigned char *dest
= ((struct ethhdr
*)skb
->data
)->h_dest
;
521 sg_init_table(sg
, 2+MAX_SKB_FRAGS
);
523 pr_debug("%s: xmit %p %pM\n", vi
->dev
->name
, skb
, dest
);
525 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
526 hdr
->hdr
.flags
= VIRTIO_NET_HDR_F_NEEDS_CSUM
;
527 hdr
->hdr
.csum_start
= skb
->csum_start
- skb_headroom(skb
);
528 hdr
->hdr
.csum_offset
= skb
->csum_offset
;
531 hdr
->hdr
.csum_offset
= hdr
->hdr
.csum_start
= 0;
534 if (skb_is_gso(skb
)) {
535 hdr
->hdr
.hdr_len
= skb_headlen(skb
);
536 hdr
->hdr
.gso_size
= skb_shinfo(skb
)->gso_size
;
537 if (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCPV4
)
538 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV4
;
539 else if (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCPV6
)
540 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV6
;
541 else if (skb_shinfo(skb
)->gso_type
& SKB_GSO_UDP
)
542 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_UDP
;
545 if (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCP_ECN
)
546 hdr
->hdr
.gso_type
|= VIRTIO_NET_HDR_GSO_ECN
;
548 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
549 hdr
->hdr
.gso_size
= hdr
->hdr
.hdr_len
= 0;
552 hdr
->mhdr
.num_buffers
= 0;
554 /* Encode metadata header at front. */
555 if (vi
->mergeable_rx_bufs
)
556 sg_set_buf(sg
, &hdr
->mhdr
, sizeof hdr
->mhdr
);
558 sg_set_buf(sg
, &hdr
->hdr
, sizeof hdr
->hdr
);
560 hdr
->num_sg
= skb_to_sgvec(skb
, sg
+1, 0, skb
->len
) + 1;
561 return vi
->svq
->vq_ops
->add_buf(vi
->svq
, sg
, hdr
->num_sg
, 0, skb
);
564 static netdev_tx_t
start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
566 struct virtnet_info
*vi
= netdev_priv(dev
);
570 /* Free up any pending old buffers before queueing new ones. */
571 free_old_xmit_skbs(vi
);
573 /* Try to transmit */
574 capacity
= xmit_skb(vi
, skb
);
576 /* This can happen with OOM and indirect buffers. */
577 if (unlikely(capacity
< 0)) {
578 netif_stop_queue(dev
);
579 dev_warn(&dev
->dev
, "Unexpected full queue\n");
580 if (unlikely(!vi
->svq
->vq_ops
->enable_cb(vi
->svq
))) {
581 vi
->svq
->vq_ops
->disable_cb(vi
->svq
);
582 netif_start_queue(dev
);
585 return NETDEV_TX_BUSY
;
587 vi
->svq
->vq_ops
->kick(vi
->svq
);
589 /* Don't wait up for transmitted skbs to be freed. */
593 /* Apparently nice girls don't return TX_BUSY; stop the queue
594 * before it gets out of hand. Naturally, this wastes entries. */
595 if (capacity
< 2+MAX_SKB_FRAGS
) {
596 netif_stop_queue(dev
);
597 if (unlikely(!vi
->svq
->vq_ops
->enable_cb(vi
->svq
))) {
598 /* More just got used, free them then recheck. */
599 capacity
+= free_old_xmit_skbs(vi
);
600 if (capacity
>= 2+MAX_SKB_FRAGS
) {
601 netif_start_queue(dev
);
602 vi
->svq
->vq_ops
->disable_cb(vi
->svq
);
610 static int virtnet_set_mac_address(struct net_device
*dev
, void *p
)
612 struct virtnet_info
*vi
= netdev_priv(dev
);
613 struct virtio_device
*vdev
= vi
->vdev
;
616 ret
= eth_mac_addr(dev
, p
);
620 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MAC
))
621 vdev
->config
->set(vdev
, offsetof(struct virtio_net_config
, mac
),
622 dev
->dev_addr
, dev
->addr_len
);
627 #ifdef CONFIG_NET_POLL_CONTROLLER
628 static void virtnet_netpoll(struct net_device
*dev
)
630 struct virtnet_info
*vi
= netdev_priv(dev
);
632 napi_schedule(&vi
->napi
);
636 static int virtnet_open(struct net_device
*dev
)
638 struct virtnet_info
*vi
= netdev_priv(dev
);
640 napi_enable(&vi
->napi
);
642 /* If all buffers were filled by other side before we napi_enabled, we
643 * won't get another interrupt, so process any outstanding packets
644 * now. virtnet_poll wants re-enable the queue, so we disable here.
645 * We synchronize against interrupts via NAPI_STATE_SCHED */
646 if (napi_schedule_prep(&vi
->napi
)) {
647 vi
->rvq
->vq_ops
->disable_cb(vi
->rvq
);
648 __napi_schedule(&vi
->napi
);
654 * Send command via the control virtqueue and check status. Commands
655 * supported by the hypervisor, as indicated by feature bits, should
656 * never fail unless improperly formated.
658 static bool virtnet_send_command(struct virtnet_info
*vi
, u8
class, u8 cmd
,
659 struct scatterlist
*data
, int out
, int in
)
661 struct scatterlist
*s
, sg
[VIRTNET_SEND_COMMAND_SG_MAX
+ 2];
662 struct virtio_net_ctrl_hdr ctrl
;
663 virtio_net_ctrl_ack status
= ~0;
667 /* Caller should know better */
668 BUG_ON(!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
) ||
669 (out
+ in
> VIRTNET_SEND_COMMAND_SG_MAX
));
671 out
++; /* Add header */
672 in
++; /* Add return status */
677 sg_init_table(sg
, out
+ in
);
679 sg_set_buf(&sg
[0], &ctrl
, sizeof(ctrl
));
680 for_each_sg(data
, s
, out
+ in
- 2, i
)
681 sg_set_buf(&sg
[i
+ 1], sg_virt(s
), s
->length
);
682 sg_set_buf(&sg
[out
+ in
- 1], &status
, sizeof(status
));
684 BUG_ON(vi
->cvq
->vq_ops
->add_buf(vi
->cvq
, sg
, out
, in
, vi
) < 0);
686 vi
->cvq
->vq_ops
->kick(vi
->cvq
);
689 * Spin for a response, the kick causes an ioport write, trapping
690 * into the hypervisor, so the request should be handled immediately.
692 while (!vi
->cvq
->vq_ops
->get_buf(vi
->cvq
, &tmp
))
695 return status
== VIRTIO_NET_OK
;
698 static int virtnet_close(struct net_device
*dev
)
700 struct virtnet_info
*vi
= netdev_priv(dev
);
702 napi_disable(&vi
->napi
);
707 static int virtnet_set_tx_csum(struct net_device
*dev
, u32 data
)
709 struct virtnet_info
*vi
= netdev_priv(dev
);
710 struct virtio_device
*vdev
= vi
->vdev
;
712 if (data
&& !virtio_has_feature(vdev
, VIRTIO_NET_F_CSUM
))
715 return ethtool_op_set_tx_hw_csum(dev
, data
);
718 static void virtnet_set_rx_mode(struct net_device
*dev
)
720 struct virtnet_info
*vi
= netdev_priv(dev
);
721 struct scatterlist sg
[2];
722 u8 promisc
, allmulti
;
723 struct virtio_net_ctrl_mac
*mac_data
;
724 struct dev_addr_list
*addr
;
725 struct netdev_hw_addr
*ha
;
731 /* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */
732 if (!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_RX
))
735 promisc
= ((dev
->flags
& IFF_PROMISC
) != 0);
736 allmulti
= ((dev
->flags
& IFF_ALLMULTI
) != 0);
738 sg_init_one(sg
, &promisc
, sizeof(promisc
));
740 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_RX
,
741 VIRTIO_NET_CTRL_RX_PROMISC
,
743 dev_warn(&dev
->dev
, "Failed to %sable promisc mode.\n",
744 promisc
? "en" : "dis");
746 sg_init_one(sg
, &allmulti
, sizeof(allmulti
));
748 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_RX
,
749 VIRTIO_NET_CTRL_RX_ALLMULTI
,
751 dev_warn(&dev
->dev
, "Failed to %sable allmulti mode.\n",
752 allmulti
? "en" : "dis");
754 uc_count
= netdev_uc_count(dev
);
755 mc_count
= netdev_mc_count(dev
);
756 /* MAC filter - use one buffer for both lists */
757 buf
= kzalloc(((uc_count
+ mc_count
) * ETH_ALEN
) +
758 (2 * sizeof(mac_data
->entries
)), GFP_ATOMIC
);
761 dev_warn(&dev
->dev
, "No memory for MAC address buffer\n");
765 sg_init_table(sg
, 2);
767 /* Store the unicast list and count in the front of the buffer */
768 mac_data
->entries
= uc_count
;
770 netdev_for_each_uc_addr(ha
, dev
)
771 memcpy(&mac_data
->macs
[i
++][0], ha
->addr
, ETH_ALEN
);
773 sg_set_buf(&sg
[0], mac_data
,
774 sizeof(mac_data
->entries
) + (uc_count
* ETH_ALEN
));
776 /* multicast list and count fill the end */
777 mac_data
= (void *)&mac_data
->macs
[uc_count
][0];
779 mac_data
->entries
= mc_count
;
781 netdev_for_each_mc_addr(addr
, dev
)
782 memcpy(&mac_data
->macs
[i
++][0], addr
->da_addr
, ETH_ALEN
);
784 sg_set_buf(&sg
[1], mac_data
,
785 sizeof(mac_data
->entries
) + (mc_count
* ETH_ALEN
));
787 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MAC
,
788 VIRTIO_NET_CTRL_MAC_TABLE_SET
,
790 dev_warn(&dev
->dev
, "Failed to set MAC fitler table.\n");
795 static void virtnet_vlan_rx_add_vid(struct net_device
*dev
, u16 vid
)
797 struct virtnet_info
*vi
= netdev_priv(dev
);
798 struct scatterlist sg
;
800 sg_init_one(&sg
, &vid
, sizeof(vid
));
802 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_VLAN
,
803 VIRTIO_NET_CTRL_VLAN_ADD
, &sg
, 1, 0))
804 dev_warn(&dev
->dev
, "Failed to add VLAN ID %d.\n", vid
);
807 static void virtnet_vlan_rx_kill_vid(struct net_device
*dev
, u16 vid
)
809 struct virtnet_info
*vi
= netdev_priv(dev
);
810 struct scatterlist sg
;
812 sg_init_one(&sg
, &vid
, sizeof(vid
));
814 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_VLAN
,
815 VIRTIO_NET_CTRL_VLAN_DEL
, &sg
, 1, 0))
816 dev_warn(&dev
->dev
, "Failed to kill VLAN ID %d.\n", vid
);
819 static const struct ethtool_ops virtnet_ethtool_ops
= {
820 .set_tx_csum
= virtnet_set_tx_csum
,
821 .set_sg
= ethtool_op_set_sg
,
822 .set_tso
= ethtool_op_set_tso
,
823 .set_ufo
= ethtool_op_set_ufo
,
824 .get_link
= ethtool_op_get_link
,
828 #define MAX_MTU 65535
830 static int virtnet_change_mtu(struct net_device
*dev
, int new_mtu
)
832 if (new_mtu
< MIN_MTU
|| new_mtu
> MAX_MTU
)
838 static const struct net_device_ops virtnet_netdev
= {
839 .ndo_open
= virtnet_open
,
840 .ndo_stop
= virtnet_close
,
841 .ndo_start_xmit
= start_xmit
,
842 .ndo_validate_addr
= eth_validate_addr
,
843 .ndo_set_mac_address
= virtnet_set_mac_address
,
844 .ndo_set_rx_mode
= virtnet_set_rx_mode
,
845 .ndo_change_mtu
= virtnet_change_mtu
,
846 .ndo_vlan_rx_add_vid
= virtnet_vlan_rx_add_vid
,
847 .ndo_vlan_rx_kill_vid
= virtnet_vlan_rx_kill_vid
,
848 #ifdef CONFIG_NET_POLL_CONTROLLER
849 .ndo_poll_controller
= virtnet_netpoll
,
853 static void virtnet_update_status(struct virtnet_info
*vi
)
857 if (!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_STATUS
))
860 vi
->vdev
->config
->get(vi
->vdev
,
861 offsetof(struct virtio_net_config
, status
),
864 /* Ignore unknown (future) status bits */
865 v
&= VIRTIO_NET_S_LINK_UP
;
872 if (vi
->status
& VIRTIO_NET_S_LINK_UP
) {
873 netif_carrier_on(vi
->dev
);
874 netif_wake_queue(vi
->dev
);
876 netif_carrier_off(vi
->dev
);
877 netif_stop_queue(vi
->dev
);
881 static void virtnet_config_changed(struct virtio_device
*vdev
)
883 struct virtnet_info
*vi
= vdev
->priv
;
885 virtnet_update_status(vi
);
888 static int virtnet_probe(struct virtio_device
*vdev
)
891 struct net_device
*dev
;
892 struct virtnet_info
*vi
;
893 struct virtqueue
*vqs
[3];
894 vq_callback_t
*callbacks
[] = { skb_recv_done
, skb_xmit_done
, NULL
};
895 const char *names
[] = { "input", "output", "control" };
898 /* Allocate ourselves a network device with room for our info */
899 dev
= alloc_etherdev(sizeof(struct virtnet_info
));
903 /* Set up network device as normal. */
904 dev
->netdev_ops
= &virtnet_netdev
;
905 dev
->features
= NETIF_F_HIGHDMA
;
906 SET_ETHTOOL_OPS(dev
, &virtnet_ethtool_ops
);
907 SET_NETDEV_DEV(dev
, &vdev
->dev
);
909 /* Do we support "hardware" checksums? */
910 if (csum
&& virtio_has_feature(vdev
, VIRTIO_NET_F_CSUM
)) {
911 /* This opens up the world of extra features. */
912 dev
->features
|= NETIF_F_HW_CSUM
|NETIF_F_SG
|NETIF_F_FRAGLIST
;
913 if (gso
&& virtio_has_feature(vdev
, VIRTIO_NET_F_GSO
)) {
914 dev
->features
|= NETIF_F_TSO
| NETIF_F_UFO
915 | NETIF_F_TSO_ECN
| NETIF_F_TSO6
;
917 /* Individual feature bits: what can host handle? */
918 if (gso
&& virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_TSO4
))
919 dev
->features
|= NETIF_F_TSO
;
920 if (gso
&& virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_TSO6
))
921 dev
->features
|= NETIF_F_TSO6
;
922 if (gso
&& virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_ECN
))
923 dev
->features
|= NETIF_F_TSO_ECN
;
924 if (gso
&& virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_UFO
))
925 dev
->features
|= NETIF_F_UFO
;
928 /* Configuration may specify what MAC to use. Otherwise random. */
929 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MAC
)) {
930 vdev
->config
->get(vdev
,
931 offsetof(struct virtio_net_config
, mac
),
932 dev
->dev_addr
, dev
->addr_len
);
934 random_ether_addr(dev
->dev_addr
);
936 /* Set up our device-specific information */
937 vi
= netdev_priv(dev
);
938 netif_napi_add(dev
, &vi
->napi
, virtnet_poll
, napi_weight
);
943 INIT_DELAYED_WORK(&vi
->refill
, refill_work
);
945 /* If we can receive ANY GSO packets, we must allocate large ones. */
946 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_TSO4
) ||
947 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_TSO6
) ||
948 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_ECN
))
949 vi
->big_packets
= true;
951 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MRG_RXBUF
))
952 vi
->mergeable_rx_bufs
= true;
954 /* We expect two virtqueues, receive then send,
955 * and optionally control. */
956 nvqs
= virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
) ? 3 : 2;
958 err
= vdev
->config
->find_vqs(vdev
, nvqs
, vqs
, callbacks
, names
);
965 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
)) {
968 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VLAN
))
969 dev
->features
|= NETIF_F_HW_VLAN_FILTER
;
972 err
= register_netdev(dev
);
974 pr_debug("virtio_net: registering device failed\n");
978 /* Last of all, set up some receive buffers. */
979 try_fill_recv(vi
, GFP_KERNEL
);
981 /* If we didn't even get one input buffer, we're useless. */
987 vi
->status
= VIRTIO_NET_S_LINK_UP
;
988 virtnet_update_status(vi
);
989 netif_carrier_on(dev
);
991 pr_debug("virtnet: registered device %s\n", dev
->name
);
995 unregister_netdev(dev
);
996 cancel_delayed_work_sync(&vi
->refill
);
998 vdev
->config
->del_vqs(vdev
);
1004 static void free_unused_bufs(struct virtnet_info
*vi
)
1008 buf
= vi
->svq
->vq_ops
->detach_unused_buf(vi
->svq
);
1014 buf
= vi
->rvq
->vq_ops
->detach_unused_buf(vi
->rvq
);
1017 if (vi
->mergeable_rx_bufs
|| vi
->big_packets
)
1018 give_pages(vi
, buf
);
1023 BUG_ON(vi
->num
!= 0);
1026 static void __devexit
virtnet_remove(struct virtio_device
*vdev
)
1028 struct virtnet_info
*vi
= vdev
->priv
;
1030 /* Stop all the virtqueues. */
1031 vdev
->config
->reset(vdev
);
1034 unregister_netdev(vi
->dev
);
1035 cancel_delayed_work_sync(&vi
->refill
);
1037 /* Free unused buffers in both send and recv, if any. */
1038 free_unused_bufs(vi
);
1040 vdev
->config
->del_vqs(vi
->vdev
);
1043 __free_pages(get_a_page(vi
, GFP_KERNEL
), 0);
1045 free_netdev(vi
->dev
);
1048 static struct virtio_device_id id_table
[] = {
1049 { VIRTIO_ID_NET
, VIRTIO_DEV_ANY_ID
},
1053 static unsigned int features
[] = {
1054 VIRTIO_NET_F_CSUM
, VIRTIO_NET_F_GUEST_CSUM
,
1055 VIRTIO_NET_F_GSO
, VIRTIO_NET_F_MAC
,
1056 VIRTIO_NET_F_HOST_TSO4
, VIRTIO_NET_F_HOST_UFO
, VIRTIO_NET_F_HOST_TSO6
,
1057 VIRTIO_NET_F_HOST_ECN
, VIRTIO_NET_F_GUEST_TSO4
, VIRTIO_NET_F_GUEST_TSO6
,
1058 VIRTIO_NET_F_GUEST_ECN
, VIRTIO_NET_F_GUEST_UFO
,
1059 VIRTIO_NET_F_MRG_RXBUF
, VIRTIO_NET_F_STATUS
, VIRTIO_NET_F_CTRL_VQ
,
1060 VIRTIO_NET_F_CTRL_RX
, VIRTIO_NET_F_CTRL_VLAN
,
1063 static struct virtio_driver virtio_net_driver
= {
1064 .feature_table
= features
,
1065 .feature_table_size
= ARRAY_SIZE(features
),
1066 .driver
.name
= KBUILD_MODNAME
,
1067 .driver
.owner
= THIS_MODULE
,
1068 .id_table
= id_table
,
1069 .probe
= virtnet_probe
,
1070 .remove
= __devexit_p(virtnet_remove
),
1071 .config_changed
= virtnet_config_changed
,
1074 static int __init
init(void)
1076 return register_virtio_driver(&virtio_net_driver
);
1079 static void __exit
fini(void)
1081 unregister_virtio_driver(&virtio_net_driver
);
1086 MODULE_DEVICE_TABLE(virtio
, id_table
);
1087 MODULE_DESCRIPTION("Virtio network driver");
1088 MODULE_LICENSE("GPL");