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 int csum
= 1, gso
= 1;
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
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. */
65 /* fragments + linear part + virtio header */
66 struct scatterlist rx_sg
[MAX_SKB_FRAGS
+ 2];
67 struct scatterlist tx_sg
[MAX_SKB_FRAGS
+ 2];
72 struct virtio_net_hdr hdr
;
73 struct virtio_net_hdr_mrg_rxbuf mhdr
;
78 struct padded_vnet_hdr
{
79 struct virtio_net_hdr hdr
;
81 * virtio_net_hdr should be in a separated sg buffer because of a
82 * QEMU bug, and data sg buffer shares same page with this header sg.
83 * This padding makes next sg 16 byte aligned after virtio_net_hdr.
88 static inline struct skb_vnet_hdr
*skb_vnet_hdr(struct sk_buff
*skb
)
90 return (struct skb_vnet_hdr
*)skb
->cb
;
94 * private is used to chain pages for big packets, put the whole
95 * most recent used list in the beginning for reuse
97 static void give_pages(struct virtnet_info
*vi
, struct page
*page
)
101 /* Find end of list, sew whole thing into vi->pages. */
102 for (end
= page
; end
->private; end
= (struct page
*)end
->private);
103 end
->private = (unsigned long)vi
->pages
;
107 static struct page
*get_a_page(struct virtnet_info
*vi
, gfp_t gfp_mask
)
109 struct page
*p
= vi
->pages
;
112 vi
->pages
= (struct page
*)p
->private;
113 /* clear private here, it is used to chain pages */
116 p
= alloc_page(gfp_mask
);
120 static void skb_xmit_done(struct virtqueue
*svq
)
122 struct virtnet_info
*vi
= svq
->vdev
->priv
;
124 /* Suppress further interrupts. */
125 virtqueue_disable_cb(svq
);
127 /* We were probably waiting for more output buffers. */
128 netif_wake_queue(vi
->dev
);
131 static void set_skb_frag(struct sk_buff
*skb
, struct page
*page
,
132 unsigned int offset
, unsigned int *len
)
134 int i
= skb_shinfo(skb
)->nr_frags
;
137 f
= &skb_shinfo(skb
)->frags
[i
];
138 f
->size
= min((unsigned)PAGE_SIZE
- offset
, *len
);
139 f
->page_offset
= offset
;
142 skb
->data_len
+= f
->size
;
144 skb_shinfo(skb
)->nr_frags
++;
148 static struct sk_buff
*page_to_skb(struct virtnet_info
*vi
,
149 struct page
*page
, unsigned int len
)
152 struct skb_vnet_hdr
*hdr
;
153 unsigned int copy
, hdr_len
, offset
;
156 p
= page_address(page
);
158 /* copy small packet so we can reuse these pages for small data */
159 skb
= netdev_alloc_skb_ip_align(vi
->dev
, GOOD_COPY_LEN
);
163 hdr
= skb_vnet_hdr(skb
);
165 if (vi
->mergeable_rx_bufs
) {
166 hdr_len
= sizeof hdr
->mhdr
;
169 hdr_len
= sizeof hdr
->hdr
;
170 offset
= sizeof(struct padded_vnet_hdr
);
173 memcpy(hdr
, p
, hdr_len
);
179 if (copy
> skb_tailroom(skb
))
180 copy
= skb_tailroom(skb
);
181 memcpy(skb_put(skb
, copy
), p
, copy
);
187 set_skb_frag(skb
, page
, offset
, &len
);
188 page
= (struct page
*)page
->private;
193 give_pages(vi
, page
);
198 static int receive_mergeable(struct virtnet_info
*vi
, struct sk_buff
*skb
)
200 struct skb_vnet_hdr
*hdr
= skb_vnet_hdr(skb
);
204 num_buf
= hdr
->mhdr
.num_buffers
;
206 i
= skb_shinfo(skb
)->nr_frags
;
207 if (i
>= MAX_SKB_FRAGS
) {
208 pr_debug("%s: packet too long\n", skb
->dev
->name
);
209 skb
->dev
->stats
.rx_length_errors
++;
213 page
= virtqueue_get_buf(vi
->rvq
, &len
);
215 pr_debug("%s: rx error: %d buffers missing\n",
216 skb
->dev
->name
, hdr
->mhdr
.num_buffers
);
217 skb
->dev
->stats
.rx_length_errors
++;
223 set_skb_frag(skb
, page
, 0, &len
);
230 static void receive_buf(struct net_device
*dev
, void *buf
, unsigned int len
)
232 struct virtnet_info
*vi
= netdev_priv(dev
);
235 struct skb_vnet_hdr
*hdr
;
237 if (unlikely(len
< sizeof(struct virtio_net_hdr
) + ETH_HLEN
)) {
238 pr_debug("%s: short packet %i\n", dev
->name
, len
);
239 dev
->stats
.rx_length_errors
++;
240 if (vi
->mergeable_rx_bufs
|| vi
->big_packets
)
247 if (!vi
->mergeable_rx_bufs
&& !vi
->big_packets
) {
249 len
-= sizeof(struct virtio_net_hdr
);
253 skb
= page_to_skb(vi
, page
, len
);
254 if (unlikely(!skb
)) {
255 dev
->stats
.rx_dropped
++;
256 give_pages(vi
, page
);
259 if (vi
->mergeable_rx_bufs
)
260 if (receive_mergeable(vi
, skb
)) {
266 hdr
= skb_vnet_hdr(skb
);
267 skb
->truesize
+= skb
->data_len
;
268 dev
->stats
.rx_bytes
+= skb
->len
;
269 dev
->stats
.rx_packets
++;
271 if (hdr
->hdr
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) {
272 pr_debug("Needs csum!\n");
273 if (!skb_partial_csum_set(skb
,
275 hdr
->hdr
.csum_offset
))
277 } else if (hdr
->hdr
.flags
& VIRTIO_NET_HDR_F_DATA_VALID
) {
278 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
281 skb
->protocol
= eth_type_trans(skb
, dev
);
282 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
283 ntohs(skb
->protocol
), skb
->len
, skb
->pkt_type
);
285 if (hdr
->hdr
.gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
287 switch (hdr
->hdr
.gso_type
& ~VIRTIO_NET_HDR_GSO_ECN
) {
288 case VIRTIO_NET_HDR_GSO_TCPV4
:
289 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV4
;
291 case VIRTIO_NET_HDR_GSO_UDP
:
292 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
294 case VIRTIO_NET_HDR_GSO_TCPV6
:
295 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV6
;
299 printk(KERN_WARNING
"%s: bad gso type %u.\n",
300 dev
->name
, hdr
->hdr
.gso_type
);
304 if (hdr
->hdr
.gso_type
& VIRTIO_NET_HDR_GSO_ECN
)
305 skb_shinfo(skb
)->gso_type
|= SKB_GSO_TCP_ECN
;
307 skb_shinfo(skb
)->gso_size
= hdr
->hdr
.gso_size
;
308 if (skb_shinfo(skb
)->gso_size
== 0) {
310 printk(KERN_WARNING
"%s: zero gso size.\n",
315 /* Header must be checked, and gso_segs computed. */
316 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
317 skb_shinfo(skb
)->gso_segs
= 0;
320 netif_receive_skb(skb
);
324 dev
->stats
.rx_frame_errors
++;
328 static int add_recvbuf_small(struct virtnet_info
*vi
, gfp_t gfp
)
331 struct skb_vnet_hdr
*hdr
;
334 skb
= netdev_alloc_skb_ip_align(vi
->dev
, MAX_PACKET_LEN
);
338 skb_put(skb
, MAX_PACKET_LEN
);
340 hdr
= skb_vnet_hdr(skb
);
341 sg_set_buf(vi
->rx_sg
, &hdr
->hdr
, sizeof hdr
->hdr
);
343 skb_to_sgvec(skb
, vi
->rx_sg
+ 1, 0, skb
->len
);
345 err
= virtqueue_add_buf_gfp(vi
->rvq
, vi
->rx_sg
, 0, 2, skb
, gfp
);
352 static int add_recvbuf_big(struct virtnet_info
*vi
, gfp_t gfp
)
354 struct page
*first
, *list
= NULL
;
358 /* page in vi->rx_sg[MAX_SKB_FRAGS + 1] is list tail */
359 for (i
= MAX_SKB_FRAGS
+ 1; i
> 1; --i
) {
360 first
= get_a_page(vi
, gfp
);
363 give_pages(vi
, list
);
366 sg_set_buf(&vi
->rx_sg
[i
], page_address(first
), PAGE_SIZE
);
368 /* chain new page in list head to match sg */
369 first
->private = (unsigned long)list
;
373 first
= get_a_page(vi
, gfp
);
375 give_pages(vi
, list
);
378 p
= page_address(first
);
380 /* vi->rx_sg[0], vi->rx_sg[1] share the same page */
381 /* a separated vi->rx_sg[0] for virtio_net_hdr only due to QEMU bug */
382 sg_set_buf(&vi
->rx_sg
[0], p
, sizeof(struct virtio_net_hdr
));
384 /* vi->rx_sg[1] for data packet, from offset */
385 offset
= sizeof(struct padded_vnet_hdr
);
386 sg_set_buf(&vi
->rx_sg
[1], p
+ offset
, PAGE_SIZE
- offset
);
388 /* chain first in list head */
389 first
->private = (unsigned long)list
;
390 err
= virtqueue_add_buf_gfp(vi
->rvq
, vi
->rx_sg
, 0, MAX_SKB_FRAGS
+ 2,
393 give_pages(vi
, first
);
398 static int add_recvbuf_mergeable(struct virtnet_info
*vi
, gfp_t gfp
)
403 page
= get_a_page(vi
, gfp
);
407 sg_init_one(vi
->rx_sg
, page_address(page
), PAGE_SIZE
);
409 err
= virtqueue_add_buf_gfp(vi
->rvq
, vi
->rx_sg
, 0, 1, page
, gfp
);
411 give_pages(vi
, page
);
416 /* Returns false if we couldn't fill entirely (OOM). */
417 static bool try_fill_recv(struct virtnet_info
*vi
, gfp_t gfp
)
423 if (vi
->mergeable_rx_bufs
)
424 err
= add_recvbuf_mergeable(vi
, gfp
);
425 else if (vi
->big_packets
)
426 err
= add_recvbuf_big(vi
, gfp
);
428 err
= add_recvbuf_small(vi
, gfp
);
430 oom
= err
== -ENOMEM
;
435 if (unlikely(vi
->num
> vi
->max
))
437 virtqueue_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 virtqueue_disable_cb(rvq
);
447 __napi_schedule(&vi
->napi
);
451 static void virtnet_napi_enable(struct virtnet_info
*vi
)
453 napi_enable(&vi
->napi
);
455 /* If all buffers were filled by other side before we napi_enabled, we
456 * won't get another interrupt, so process any outstanding packets
457 * now. virtnet_poll wants re-enable the queue, so we disable here.
458 * We synchronize against interrupts via NAPI_STATE_SCHED */
459 if (napi_schedule_prep(&vi
->napi
)) {
460 virtqueue_disable_cb(vi
->rvq
);
461 __napi_schedule(&vi
->napi
);
465 static void refill_work(struct work_struct
*work
)
467 struct virtnet_info
*vi
;
470 vi
= container_of(work
, struct virtnet_info
, refill
.work
);
471 napi_disable(&vi
->napi
);
472 still_empty
= !try_fill_recv(vi
, GFP_KERNEL
);
473 virtnet_napi_enable(vi
);
475 /* In theory, this can happen: if we don't get any buffers in
476 * we will *never* try to fill again. */
478 schedule_delayed_work(&vi
->refill
, HZ
/2);
481 static int virtnet_poll(struct napi_struct
*napi
, int budget
)
483 struct virtnet_info
*vi
= container_of(napi
, struct virtnet_info
, napi
);
485 unsigned int len
, received
= 0;
488 while (received
< budget
&&
489 (buf
= virtqueue_get_buf(vi
->rvq
, &len
)) != NULL
) {
490 receive_buf(vi
->dev
, buf
, len
);
495 if (vi
->num
< vi
->max
/ 2) {
496 if (!try_fill_recv(vi
, GFP_ATOMIC
))
497 schedule_delayed_work(&vi
->refill
, 0);
500 /* Out of packets? */
501 if (received
< budget
) {
503 if (unlikely(!virtqueue_enable_cb(vi
->rvq
)) &&
504 napi_schedule_prep(napi
)) {
505 virtqueue_disable_cb(vi
->rvq
);
506 __napi_schedule(napi
);
514 static unsigned int free_old_xmit_skbs(struct virtnet_info
*vi
)
517 unsigned int len
, tot_sgs
= 0;
519 while ((skb
= virtqueue_get_buf(vi
->svq
, &len
)) != NULL
) {
520 pr_debug("Sent skb %p\n", skb
);
521 vi
->dev
->stats
.tx_bytes
+= skb
->len
;
522 vi
->dev
->stats
.tx_packets
++;
523 tot_sgs
+= skb_vnet_hdr(skb
)->num_sg
;
524 dev_kfree_skb_any(skb
);
529 static int xmit_skb(struct virtnet_info
*vi
, struct sk_buff
*skb
)
531 struct skb_vnet_hdr
*hdr
= skb_vnet_hdr(skb
);
532 const unsigned char *dest
= ((struct ethhdr
*)skb
->data
)->h_dest
;
534 pr_debug("%s: xmit %p %pM\n", vi
->dev
->name
, skb
, dest
);
536 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
537 hdr
->hdr
.flags
= VIRTIO_NET_HDR_F_NEEDS_CSUM
;
538 hdr
->hdr
.csum_start
= skb_checksum_start_offset(skb
);
539 hdr
->hdr
.csum_offset
= skb
->csum_offset
;
542 hdr
->hdr
.csum_offset
= hdr
->hdr
.csum_start
= 0;
545 if (skb_is_gso(skb
)) {
546 hdr
->hdr
.hdr_len
= skb_headlen(skb
);
547 hdr
->hdr
.gso_size
= skb_shinfo(skb
)->gso_size
;
548 if (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCPV4
)
549 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV4
;
550 else if (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCPV6
)
551 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV6
;
552 else if (skb_shinfo(skb
)->gso_type
& SKB_GSO_UDP
)
553 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_UDP
;
556 if (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCP_ECN
)
557 hdr
->hdr
.gso_type
|= VIRTIO_NET_HDR_GSO_ECN
;
559 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
560 hdr
->hdr
.gso_size
= hdr
->hdr
.hdr_len
= 0;
563 hdr
->mhdr
.num_buffers
= 0;
565 /* Encode metadata header at front. */
566 if (vi
->mergeable_rx_bufs
)
567 sg_set_buf(vi
->tx_sg
, &hdr
->mhdr
, sizeof hdr
->mhdr
);
569 sg_set_buf(vi
->tx_sg
, &hdr
->hdr
, sizeof hdr
->hdr
);
571 hdr
->num_sg
= skb_to_sgvec(skb
, vi
->tx_sg
+ 1, 0, skb
->len
) + 1;
572 return virtqueue_add_buf(vi
->svq
, vi
->tx_sg
, hdr
->num_sg
,
576 static netdev_tx_t
start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
578 struct virtnet_info
*vi
= netdev_priv(dev
);
581 /* Free up any pending old buffers before queueing new ones. */
582 free_old_xmit_skbs(vi
);
584 /* Try to transmit */
585 capacity
= xmit_skb(vi
, skb
);
587 /* This can happen with OOM and indirect buffers. */
588 if (unlikely(capacity
< 0)) {
589 if (net_ratelimit()) {
590 if (likely(capacity
== -ENOMEM
)) {
592 "TX queue failure: out of memory\n");
594 dev
->stats
.tx_fifo_errors
++;
596 "Unexpected TX queue failure: %d\n",
600 dev
->stats
.tx_dropped
++;
604 virtqueue_kick(vi
->svq
);
606 /* Don't wait up for transmitted skbs to be freed. */
610 /* Apparently nice girls don't return TX_BUSY; stop the queue
611 * before it gets out of hand. Naturally, this wastes entries. */
612 if (capacity
< 2+MAX_SKB_FRAGS
) {
613 netif_stop_queue(dev
);
614 if (unlikely(!virtqueue_enable_cb_delayed(vi
->svq
))) {
615 /* More just got used, free them then recheck. */
616 capacity
+= free_old_xmit_skbs(vi
);
617 if (capacity
>= 2+MAX_SKB_FRAGS
) {
618 netif_start_queue(dev
);
619 virtqueue_disable_cb(vi
->svq
);
627 static int virtnet_set_mac_address(struct net_device
*dev
, void *p
)
629 struct virtnet_info
*vi
= netdev_priv(dev
);
630 struct virtio_device
*vdev
= vi
->vdev
;
633 ret
= eth_mac_addr(dev
, p
);
637 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MAC
))
638 vdev
->config
->set(vdev
, offsetof(struct virtio_net_config
, mac
),
639 dev
->dev_addr
, dev
->addr_len
);
644 #ifdef CONFIG_NET_POLL_CONTROLLER
645 static void virtnet_netpoll(struct net_device
*dev
)
647 struct virtnet_info
*vi
= netdev_priv(dev
);
649 napi_schedule(&vi
->napi
);
653 static int virtnet_open(struct net_device
*dev
)
655 struct virtnet_info
*vi
= netdev_priv(dev
);
657 virtnet_napi_enable(vi
);
662 * Send command via the control virtqueue and check status. Commands
663 * supported by the hypervisor, as indicated by feature bits, should
664 * never fail unless improperly formated.
666 static bool virtnet_send_command(struct virtnet_info
*vi
, u8
class, u8 cmd
,
667 struct scatterlist
*data
, int out
, int in
)
669 struct scatterlist
*s
, sg
[VIRTNET_SEND_COMMAND_SG_MAX
+ 2];
670 struct virtio_net_ctrl_hdr ctrl
;
671 virtio_net_ctrl_ack status
= ~0;
675 /* Caller should know better */
676 BUG_ON(!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
) ||
677 (out
+ in
> VIRTNET_SEND_COMMAND_SG_MAX
));
679 out
++; /* Add header */
680 in
++; /* Add return status */
685 sg_init_table(sg
, out
+ in
);
687 sg_set_buf(&sg
[0], &ctrl
, sizeof(ctrl
));
688 for_each_sg(data
, s
, out
+ in
- 2, i
)
689 sg_set_buf(&sg
[i
+ 1], sg_virt(s
), s
->length
);
690 sg_set_buf(&sg
[out
+ in
- 1], &status
, sizeof(status
));
692 BUG_ON(virtqueue_add_buf(vi
->cvq
, sg
, out
, in
, vi
) < 0);
694 virtqueue_kick(vi
->cvq
);
697 * Spin for a response, the kick causes an ioport write, trapping
698 * into the hypervisor, so the request should be handled immediately.
700 while (!virtqueue_get_buf(vi
->cvq
, &tmp
))
703 return status
== VIRTIO_NET_OK
;
706 static int virtnet_close(struct net_device
*dev
)
708 struct virtnet_info
*vi
= netdev_priv(dev
);
710 napi_disable(&vi
->napi
);
715 static void virtnet_set_rx_mode(struct net_device
*dev
)
717 struct virtnet_info
*vi
= netdev_priv(dev
);
718 struct scatterlist sg
[2];
719 u8 promisc
, allmulti
;
720 struct virtio_net_ctrl_mac
*mac_data
;
721 struct netdev_hw_addr
*ha
;
727 /* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */
728 if (!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_RX
))
731 promisc
= ((dev
->flags
& IFF_PROMISC
) != 0);
732 allmulti
= ((dev
->flags
& IFF_ALLMULTI
) != 0);
734 sg_init_one(sg
, &promisc
, sizeof(promisc
));
736 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_RX
,
737 VIRTIO_NET_CTRL_RX_PROMISC
,
739 dev_warn(&dev
->dev
, "Failed to %sable promisc mode.\n",
740 promisc
? "en" : "dis");
742 sg_init_one(sg
, &allmulti
, sizeof(allmulti
));
744 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_RX
,
745 VIRTIO_NET_CTRL_RX_ALLMULTI
,
747 dev_warn(&dev
->dev
, "Failed to %sable allmulti mode.\n",
748 allmulti
? "en" : "dis");
750 uc_count
= netdev_uc_count(dev
);
751 mc_count
= netdev_mc_count(dev
);
752 /* MAC filter - use one buffer for both lists */
753 buf
= kzalloc(((uc_count
+ mc_count
) * ETH_ALEN
) +
754 (2 * sizeof(mac_data
->entries
)), GFP_ATOMIC
);
757 dev_warn(&dev
->dev
, "No memory for MAC address buffer\n");
761 sg_init_table(sg
, 2);
763 /* Store the unicast list and count in the front of the buffer */
764 mac_data
->entries
= uc_count
;
766 netdev_for_each_uc_addr(ha
, dev
)
767 memcpy(&mac_data
->macs
[i
++][0], ha
->addr
, ETH_ALEN
);
769 sg_set_buf(&sg
[0], mac_data
,
770 sizeof(mac_data
->entries
) + (uc_count
* ETH_ALEN
));
772 /* multicast list and count fill the end */
773 mac_data
= (void *)&mac_data
->macs
[uc_count
][0];
775 mac_data
->entries
= mc_count
;
777 netdev_for_each_mc_addr(ha
, dev
)
778 memcpy(&mac_data
->macs
[i
++][0], ha
->addr
, ETH_ALEN
);
780 sg_set_buf(&sg
[1], mac_data
,
781 sizeof(mac_data
->entries
) + (mc_count
* ETH_ALEN
));
783 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MAC
,
784 VIRTIO_NET_CTRL_MAC_TABLE_SET
,
786 dev_warn(&dev
->dev
, "Failed to set MAC fitler table.\n");
791 static void virtnet_vlan_rx_add_vid(struct net_device
*dev
, u16 vid
)
793 struct virtnet_info
*vi
= netdev_priv(dev
);
794 struct scatterlist sg
;
796 sg_init_one(&sg
, &vid
, sizeof(vid
));
798 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_VLAN
,
799 VIRTIO_NET_CTRL_VLAN_ADD
, &sg
, 1, 0))
800 dev_warn(&dev
->dev
, "Failed to add VLAN ID %d.\n", vid
);
803 static void virtnet_vlan_rx_kill_vid(struct net_device
*dev
, u16 vid
)
805 struct virtnet_info
*vi
= netdev_priv(dev
);
806 struct scatterlist sg
;
808 sg_init_one(&sg
, &vid
, sizeof(vid
));
810 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_VLAN
,
811 VIRTIO_NET_CTRL_VLAN_DEL
, &sg
, 1, 0))
812 dev_warn(&dev
->dev
, "Failed to kill VLAN ID %d.\n", vid
);
815 static const struct ethtool_ops virtnet_ethtool_ops
= {
816 .get_link
= ethtool_op_get_link
,
820 #define MAX_MTU 65535
822 static int virtnet_change_mtu(struct net_device
*dev
, int new_mtu
)
824 if (new_mtu
< MIN_MTU
|| new_mtu
> MAX_MTU
)
830 static const struct net_device_ops virtnet_netdev
= {
831 .ndo_open
= virtnet_open
,
832 .ndo_stop
= virtnet_close
,
833 .ndo_start_xmit
= start_xmit
,
834 .ndo_validate_addr
= eth_validate_addr
,
835 .ndo_set_mac_address
= virtnet_set_mac_address
,
836 .ndo_set_rx_mode
= virtnet_set_rx_mode
,
837 .ndo_change_mtu
= virtnet_change_mtu
,
838 .ndo_vlan_rx_add_vid
= virtnet_vlan_rx_add_vid
,
839 .ndo_vlan_rx_kill_vid
= virtnet_vlan_rx_kill_vid
,
840 #ifdef CONFIG_NET_POLL_CONTROLLER
841 .ndo_poll_controller
= virtnet_netpoll
,
845 static void virtnet_update_status(struct virtnet_info
*vi
)
849 if (!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_STATUS
))
852 vi
->vdev
->config
->get(vi
->vdev
,
853 offsetof(struct virtio_net_config
, status
),
856 /* Ignore unknown (future) status bits */
857 v
&= VIRTIO_NET_S_LINK_UP
;
864 if (vi
->status
& VIRTIO_NET_S_LINK_UP
) {
865 netif_carrier_on(vi
->dev
);
866 netif_wake_queue(vi
->dev
);
868 netif_carrier_off(vi
->dev
);
869 netif_stop_queue(vi
->dev
);
873 static void virtnet_config_changed(struct virtio_device
*vdev
)
875 struct virtnet_info
*vi
= vdev
->priv
;
877 virtnet_update_status(vi
);
880 static int virtnet_probe(struct virtio_device
*vdev
)
883 struct net_device
*dev
;
884 struct virtnet_info
*vi
;
885 struct virtqueue
*vqs
[3];
886 vq_callback_t
*callbacks
[] = { skb_recv_done
, skb_xmit_done
, NULL
};
887 const char *names
[] = { "input", "output", "control" };
890 /* Allocate ourselves a network device with room for our info */
891 dev
= alloc_etherdev(sizeof(struct virtnet_info
));
895 /* Set up network device as normal. */
896 dev
->netdev_ops
= &virtnet_netdev
;
897 dev
->features
= NETIF_F_HIGHDMA
;
898 SET_ETHTOOL_OPS(dev
, &virtnet_ethtool_ops
);
899 SET_NETDEV_DEV(dev
, &vdev
->dev
);
901 /* Do we support "hardware" checksums? */
902 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CSUM
)) {
903 /* This opens up the world of extra features. */
904 dev
->hw_features
|= NETIF_F_HW_CSUM
|NETIF_F_SG
|NETIF_F_FRAGLIST
;
906 dev
->features
|= NETIF_F_HW_CSUM
|NETIF_F_SG
|NETIF_F_FRAGLIST
;
908 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GSO
)) {
909 dev
->hw_features
|= NETIF_F_TSO
| NETIF_F_UFO
910 | NETIF_F_TSO_ECN
| NETIF_F_TSO6
;
912 /* Individual feature bits: what can host handle? */
913 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_TSO4
))
914 dev
->hw_features
|= NETIF_F_TSO
;
915 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_TSO6
))
916 dev
->hw_features
|= NETIF_F_TSO6
;
917 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_ECN
))
918 dev
->hw_features
|= NETIF_F_TSO_ECN
;
919 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_UFO
))
920 dev
->hw_features
|= NETIF_F_UFO
;
923 dev
->features
|= dev
->hw_features
& (NETIF_F_ALL_TSO
|NETIF_F_UFO
);
924 /* (!csum && gso) case will be fixed by register_netdev() */
927 /* Configuration may specify what MAC to use. Otherwise random. */
928 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MAC
)) {
929 vdev
->config
->get(vdev
,
930 offsetof(struct virtio_net_config
, mac
),
931 dev
->dev_addr
, dev
->addr_len
);
933 random_ether_addr(dev
->dev_addr
);
935 /* Set up our device-specific information */
936 vi
= netdev_priv(dev
);
937 netif_napi_add(dev
, &vi
->napi
, virtnet_poll
, napi_weight
);
942 INIT_DELAYED_WORK(&vi
->refill
, refill_work
);
943 sg_init_table(vi
->rx_sg
, ARRAY_SIZE(vi
->rx_sg
));
944 sg_init_table(vi
->tx_sg
, ARRAY_SIZE(vi
->tx_sg
));
946 /* If we can receive ANY GSO packets, we must allocate large ones. */
947 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_TSO4
) ||
948 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_TSO6
) ||
949 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_ECN
))
950 vi
->big_packets
= true;
952 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MRG_RXBUF
))
953 vi
->mergeable_rx_bufs
= true;
955 /* We expect two virtqueues, receive then send,
956 * and optionally control. */
957 nvqs
= virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
) ? 3 : 2;
959 err
= vdev
->config
->find_vqs(vdev
, nvqs
, vqs
, callbacks
, names
);
966 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
)) {
969 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VLAN
))
970 dev
->features
|= NETIF_F_HW_VLAN_FILTER
;
973 err
= register_netdev(dev
);
975 pr_debug("virtio_net: registering device failed\n");
979 /* Last of all, set up some receive buffers. */
980 try_fill_recv(vi
, GFP_KERNEL
);
982 /* If we didn't even get one input buffer, we're useless. */
988 /* Assume link up if device can't report link status,
989 otherwise get link status from config. */
990 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_STATUS
)) {
991 netif_carrier_off(dev
);
992 virtnet_update_status(vi
);
994 vi
->status
= VIRTIO_NET_S_LINK_UP
;
995 netif_carrier_on(dev
);
998 pr_debug("virtnet: registered device %s\n", dev
->name
);
1002 unregister_netdev(dev
);
1003 cancel_delayed_work_sync(&vi
->refill
);
1005 vdev
->config
->del_vqs(vdev
);
1011 static void free_unused_bufs(struct virtnet_info
*vi
)
1015 buf
= virtqueue_detach_unused_buf(vi
->svq
);
1021 buf
= virtqueue_detach_unused_buf(vi
->rvq
);
1024 if (vi
->mergeable_rx_bufs
|| vi
->big_packets
)
1025 give_pages(vi
, buf
);
1030 BUG_ON(vi
->num
!= 0);
1033 static void __devexit
virtnet_remove(struct virtio_device
*vdev
)
1035 struct virtnet_info
*vi
= vdev
->priv
;
1037 /* Stop all the virtqueues. */
1038 vdev
->config
->reset(vdev
);
1041 unregister_netdev(vi
->dev
);
1042 cancel_delayed_work_sync(&vi
->refill
);
1044 /* Free unused buffers in both send and recv, if any. */
1045 free_unused_bufs(vi
);
1047 vdev
->config
->del_vqs(vi
->vdev
);
1050 __free_pages(get_a_page(vi
, GFP_KERNEL
), 0);
1052 free_netdev(vi
->dev
);
1055 static struct virtio_device_id id_table
[] = {
1056 { VIRTIO_ID_NET
, VIRTIO_DEV_ANY_ID
},
1060 static unsigned int features
[] = {
1061 VIRTIO_NET_F_CSUM
, VIRTIO_NET_F_GUEST_CSUM
,
1062 VIRTIO_NET_F_GSO
, VIRTIO_NET_F_MAC
,
1063 VIRTIO_NET_F_HOST_TSO4
, VIRTIO_NET_F_HOST_UFO
, VIRTIO_NET_F_HOST_TSO6
,
1064 VIRTIO_NET_F_HOST_ECN
, VIRTIO_NET_F_GUEST_TSO4
, VIRTIO_NET_F_GUEST_TSO6
,
1065 VIRTIO_NET_F_GUEST_ECN
, VIRTIO_NET_F_GUEST_UFO
,
1066 VIRTIO_NET_F_MRG_RXBUF
, VIRTIO_NET_F_STATUS
, VIRTIO_NET_F_CTRL_VQ
,
1067 VIRTIO_NET_F_CTRL_RX
, VIRTIO_NET_F_CTRL_VLAN
,
1070 static struct virtio_driver virtio_net_driver
= {
1071 .feature_table
= features
,
1072 .feature_table_size
= ARRAY_SIZE(features
),
1073 .driver
.name
= KBUILD_MODNAME
,
1074 .driver
.owner
= THIS_MODULE
,
1075 .id_table
= id_table
,
1076 .probe
= virtnet_probe
,
1077 .remove
= __devexit_p(virtnet_remove
),
1078 .config_changed
= virtnet_config_changed
,
1081 static int __init
init(void)
1083 return register_virtio_driver(&virtio_net_driver
);
1086 static void __exit
fini(void)
1088 unregister_virtio_driver(&virtio_net_driver
);
1093 MODULE_DEVICE_TABLE(virtio
, id_table
);
1094 MODULE_DESCRIPTION("Virtio network driver");
1095 MODULE_LICENSE("GPL");