1 /* A network driver using virtio.
3 * Copyright 2007 Rusty Russell <rusty@rustcorp.com.au> IBM Corporation
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 #include <linux/netdevice.h>
21 #include <linux/etherdevice.h>
22 #include <linux/ethtool.h>
23 #include <linux/module.h>
24 #include <linux/virtio.h>
25 #include <linux/virtio_net.h>
26 #include <linux/scatterlist.h>
27 #include <linux/if_vlan.h>
28 #include <linux/slab.h>
30 static int napi_weight
= 128;
31 module_param(napi_weight
, int, 0444);
33 static bool csum
= true, gso
= true;
34 module_param(csum
, bool, 0444);
35 module_param(gso
, bool, 0444);
37 /* FIXME: MTU in config. */
38 #define MAX_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
39 #define GOOD_COPY_LEN 128
41 #define VIRTNET_SEND_COMMAND_SG_MAX 2
42 #define VIRTNET_DRIVER_VERSION "1.0.0"
44 struct virtnet_stats
{
45 struct u64_stats_sync syncp
;
54 struct virtio_device
*vdev
;
55 struct virtqueue
*rvq
, *svq
, *cvq
;
56 struct net_device
*dev
;
57 struct napi_struct napi
;
60 /* Number of input buffers, and max we've ever had. */
61 unsigned int num
, max
;
63 /* I like... big packets and I cannot lie! */
66 /* Host will merge rx buffers for big packets (shake it! shake it!) */
67 bool mergeable_rx_bufs
;
69 /* Active statistics */
70 struct virtnet_stats __percpu
*stats
;
72 /* Work struct for refilling if we run low on memory. */
73 struct delayed_work refill
;
75 /* Chain pages by the private ptr. */
78 /* fragments + linear part + virtio header */
79 struct scatterlist rx_sg
[MAX_SKB_FRAGS
+ 2];
80 struct scatterlist tx_sg
[MAX_SKB_FRAGS
+ 2];
85 struct virtio_net_hdr hdr
;
86 struct virtio_net_hdr_mrg_rxbuf mhdr
;
91 struct padded_vnet_hdr
{
92 struct virtio_net_hdr hdr
;
94 * virtio_net_hdr should be in a separated sg buffer because of a
95 * QEMU bug, and data sg buffer shares same page with this header sg.
96 * This padding makes next sg 16 byte aligned after virtio_net_hdr.
101 static inline struct skb_vnet_hdr
*skb_vnet_hdr(struct sk_buff
*skb
)
103 return (struct skb_vnet_hdr
*)skb
->cb
;
107 * private is used to chain pages for big packets, put the whole
108 * most recent used list in the beginning for reuse
110 static void give_pages(struct virtnet_info
*vi
, struct page
*page
)
114 /* Find end of list, sew whole thing into vi->pages. */
115 for (end
= page
; end
->private; end
= (struct page
*)end
->private);
116 end
->private = (unsigned long)vi
->pages
;
120 static struct page
*get_a_page(struct virtnet_info
*vi
, gfp_t gfp_mask
)
122 struct page
*p
= vi
->pages
;
125 vi
->pages
= (struct page
*)p
->private;
126 /* clear private here, it is used to chain pages */
129 p
= alloc_page(gfp_mask
);
133 static void skb_xmit_done(struct virtqueue
*svq
)
135 struct virtnet_info
*vi
= svq
->vdev
->priv
;
137 /* Suppress further interrupts. */
138 virtqueue_disable_cb(svq
);
140 /* We were probably waiting for more output buffers. */
141 netif_wake_queue(vi
->dev
);
144 static void set_skb_frag(struct sk_buff
*skb
, struct page
*page
,
145 unsigned int offset
, unsigned int *len
)
147 int size
= min((unsigned)PAGE_SIZE
- offset
, *len
);
148 int i
= skb_shinfo(skb
)->nr_frags
;
150 __skb_fill_page_desc(skb
, i
, page
, offset
, size
);
152 skb
->data_len
+= size
;
154 skb
->truesize
+= PAGE_SIZE
;
155 skb_shinfo(skb
)->nr_frags
++;
159 static struct sk_buff
*page_to_skb(struct virtnet_info
*vi
,
160 struct page
*page
, unsigned int len
)
163 struct skb_vnet_hdr
*hdr
;
164 unsigned int copy
, hdr_len
, offset
;
167 p
= page_address(page
);
169 /* copy small packet so we can reuse these pages for small data */
170 skb
= netdev_alloc_skb_ip_align(vi
->dev
, GOOD_COPY_LEN
);
174 hdr
= skb_vnet_hdr(skb
);
176 if (vi
->mergeable_rx_bufs
) {
177 hdr_len
= sizeof hdr
->mhdr
;
180 hdr_len
= sizeof hdr
->hdr
;
181 offset
= sizeof(struct padded_vnet_hdr
);
184 memcpy(hdr
, p
, hdr_len
);
190 if (copy
> skb_tailroom(skb
))
191 copy
= skb_tailroom(skb
);
192 memcpy(skb_put(skb
, copy
), p
, copy
);
198 * Verify that we can indeed put this data into a skb.
199 * This is here to handle cases when the device erroneously
200 * tries to receive more than is possible. This is usually
201 * the case of a broken device.
203 if (unlikely(len
> MAX_SKB_FRAGS
* PAGE_SIZE
)) {
205 pr_debug("%s: too much data\n", skb
->dev
->name
);
211 set_skb_frag(skb
, page
, offset
, &len
);
212 page
= (struct page
*)page
->private;
217 give_pages(vi
, page
);
222 static int receive_mergeable(struct virtnet_info
*vi
, struct sk_buff
*skb
)
224 struct skb_vnet_hdr
*hdr
= skb_vnet_hdr(skb
);
228 num_buf
= hdr
->mhdr
.num_buffers
;
230 i
= skb_shinfo(skb
)->nr_frags
;
231 if (i
>= MAX_SKB_FRAGS
) {
232 pr_debug("%s: packet too long\n", skb
->dev
->name
);
233 skb
->dev
->stats
.rx_length_errors
++;
236 page
= virtqueue_get_buf(vi
->rvq
, &len
);
238 pr_debug("%s: rx error: %d buffers missing\n",
239 skb
->dev
->name
, hdr
->mhdr
.num_buffers
);
240 skb
->dev
->stats
.rx_length_errors
++;
247 set_skb_frag(skb
, page
, 0, &len
);
254 static void receive_buf(struct net_device
*dev
, void *buf
, unsigned int len
)
256 struct virtnet_info
*vi
= netdev_priv(dev
);
257 struct virtnet_stats __percpu
*stats
= this_cpu_ptr(vi
->stats
);
260 struct skb_vnet_hdr
*hdr
;
262 if (unlikely(len
< sizeof(struct virtio_net_hdr
) + ETH_HLEN
)) {
263 pr_debug("%s: short packet %i\n", dev
->name
, len
);
264 dev
->stats
.rx_length_errors
++;
265 if (vi
->mergeable_rx_bufs
|| vi
->big_packets
)
272 if (!vi
->mergeable_rx_bufs
&& !vi
->big_packets
) {
274 len
-= sizeof(struct virtio_net_hdr
);
278 skb
= page_to_skb(vi
, page
, len
);
279 if (unlikely(!skb
)) {
280 dev
->stats
.rx_dropped
++;
281 give_pages(vi
, page
);
284 if (vi
->mergeable_rx_bufs
)
285 if (receive_mergeable(vi
, skb
)) {
291 hdr
= skb_vnet_hdr(skb
);
293 u64_stats_update_begin(&stats
->syncp
);
294 stats
->rx_bytes
+= skb
->len
;
296 u64_stats_update_end(&stats
->syncp
);
298 if (hdr
->hdr
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) {
299 pr_debug("Needs csum!\n");
300 if (!skb_partial_csum_set(skb
,
302 hdr
->hdr
.csum_offset
))
304 } else if (hdr
->hdr
.flags
& VIRTIO_NET_HDR_F_DATA_VALID
) {
305 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
308 skb
->protocol
= eth_type_trans(skb
, dev
);
309 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
310 ntohs(skb
->protocol
), skb
->len
, skb
->pkt_type
);
312 if (hdr
->hdr
.gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
314 switch (hdr
->hdr
.gso_type
& ~VIRTIO_NET_HDR_GSO_ECN
) {
315 case VIRTIO_NET_HDR_GSO_TCPV4
:
316 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV4
;
318 case VIRTIO_NET_HDR_GSO_UDP
:
319 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
321 case VIRTIO_NET_HDR_GSO_TCPV6
:
322 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV6
;
326 printk(KERN_WARNING
"%s: bad gso type %u.\n",
327 dev
->name
, hdr
->hdr
.gso_type
);
331 if (hdr
->hdr
.gso_type
& VIRTIO_NET_HDR_GSO_ECN
)
332 skb_shinfo(skb
)->gso_type
|= SKB_GSO_TCP_ECN
;
334 skb_shinfo(skb
)->gso_size
= hdr
->hdr
.gso_size
;
335 if (skb_shinfo(skb
)->gso_size
== 0) {
337 printk(KERN_WARNING
"%s: zero gso size.\n",
342 /* Header must be checked, and gso_segs computed. */
343 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
344 skb_shinfo(skb
)->gso_segs
= 0;
347 netif_receive_skb(skb
);
351 dev
->stats
.rx_frame_errors
++;
355 static int add_recvbuf_small(struct virtnet_info
*vi
, gfp_t gfp
)
358 struct skb_vnet_hdr
*hdr
;
361 skb
= netdev_alloc_skb_ip_align(vi
->dev
, MAX_PACKET_LEN
);
365 skb_put(skb
, MAX_PACKET_LEN
);
367 hdr
= skb_vnet_hdr(skb
);
368 sg_set_buf(vi
->rx_sg
, &hdr
->hdr
, sizeof hdr
->hdr
);
370 skb_to_sgvec(skb
, vi
->rx_sg
+ 1, 0, skb
->len
);
372 err
= virtqueue_add_buf_gfp(vi
->rvq
, vi
->rx_sg
, 0, 2, skb
, gfp
);
379 static int add_recvbuf_big(struct virtnet_info
*vi
, gfp_t gfp
)
381 struct page
*first
, *list
= NULL
;
385 /* page in vi->rx_sg[MAX_SKB_FRAGS + 1] is list tail */
386 for (i
= MAX_SKB_FRAGS
+ 1; i
> 1; --i
) {
387 first
= get_a_page(vi
, gfp
);
390 give_pages(vi
, list
);
393 sg_set_buf(&vi
->rx_sg
[i
], page_address(first
), PAGE_SIZE
);
395 /* chain new page in list head to match sg */
396 first
->private = (unsigned long)list
;
400 first
= get_a_page(vi
, gfp
);
402 give_pages(vi
, list
);
405 p
= page_address(first
);
407 /* vi->rx_sg[0], vi->rx_sg[1] share the same page */
408 /* a separated vi->rx_sg[0] for virtio_net_hdr only due to QEMU bug */
409 sg_set_buf(&vi
->rx_sg
[0], p
, sizeof(struct virtio_net_hdr
));
411 /* vi->rx_sg[1] for data packet, from offset */
412 offset
= sizeof(struct padded_vnet_hdr
);
413 sg_set_buf(&vi
->rx_sg
[1], p
+ offset
, PAGE_SIZE
- offset
);
415 /* chain first in list head */
416 first
->private = (unsigned long)list
;
417 err
= virtqueue_add_buf_gfp(vi
->rvq
, vi
->rx_sg
, 0, MAX_SKB_FRAGS
+ 2,
420 give_pages(vi
, first
);
425 static int add_recvbuf_mergeable(struct virtnet_info
*vi
, gfp_t gfp
)
430 page
= get_a_page(vi
, gfp
);
434 sg_init_one(vi
->rx_sg
, page_address(page
), PAGE_SIZE
);
436 err
= virtqueue_add_buf_gfp(vi
->rvq
, vi
->rx_sg
, 0, 1, page
, gfp
);
438 give_pages(vi
, page
);
444 * Returns false if we couldn't fill entirely (OOM).
446 * Normally run in the receive path, but can also be run from ndo_open
447 * before we're receiving packets, or from refill_work which is
448 * careful to disable receiving (using napi_disable).
450 static bool try_fill_recv(struct virtnet_info
*vi
, gfp_t gfp
)
456 if (vi
->mergeable_rx_bufs
)
457 err
= add_recvbuf_mergeable(vi
, gfp
);
458 else if (vi
->big_packets
)
459 err
= add_recvbuf_big(vi
, gfp
);
461 err
= add_recvbuf_small(vi
, gfp
);
463 oom
= err
== -ENOMEM
;
468 if (unlikely(vi
->num
> vi
->max
))
470 virtqueue_kick(vi
->rvq
);
474 static void skb_recv_done(struct virtqueue
*rvq
)
476 struct virtnet_info
*vi
= rvq
->vdev
->priv
;
477 /* Schedule NAPI, Suppress further interrupts if successful. */
478 if (napi_schedule_prep(&vi
->napi
)) {
479 virtqueue_disable_cb(rvq
);
480 __napi_schedule(&vi
->napi
);
484 static void virtnet_napi_enable(struct virtnet_info
*vi
)
486 napi_enable(&vi
->napi
);
488 /* If all buffers were filled by other side before we napi_enabled, we
489 * won't get another interrupt, so process any outstanding packets
490 * now. virtnet_poll wants re-enable the queue, so we disable here.
491 * We synchronize against interrupts via NAPI_STATE_SCHED */
492 if (napi_schedule_prep(&vi
->napi
)) {
493 virtqueue_disable_cb(vi
->rvq
);
494 __napi_schedule(&vi
->napi
);
498 static void refill_work(struct work_struct
*work
)
500 struct virtnet_info
*vi
;
503 vi
= container_of(work
, struct virtnet_info
, refill
.work
);
504 napi_disable(&vi
->napi
);
505 still_empty
= !try_fill_recv(vi
, GFP_KERNEL
);
506 virtnet_napi_enable(vi
);
508 /* In theory, this can happen: if we don't get any buffers in
509 * we will *never* try to fill again. */
511 queue_delayed_work(system_nrt_wq
, &vi
->refill
, HZ
/2);
514 static int virtnet_poll(struct napi_struct
*napi
, int budget
)
516 struct virtnet_info
*vi
= container_of(napi
, struct virtnet_info
, napi
);
518 unsigned int len
, received
= 0;
521 while (received
< budget
&&
522 (buf
= virtqueue_get_buf(vi
->rvq
, &len
)) != NULL
) {
523 receive_buf(vi
->dev
, buf
, len
);
528 if (vi
->num
< vi
->max
/ 2) {
529 if (!try_fill_recv(vi
, GFP_ATOMIC
))
530 queue_delayed_work(system_nrt_wq
, &vi
->refill
, 0);
533 /* Out of packets? */
534 if (received
< budget
) {
536 if (unlikely(!virtqueue_enable_cb(vi
->rvq
)) &&
537 napi_schedule_prep(napi
)) {
538 virtqueue_disable_cb(vi
->rvq
);
539 __napi_schedule(napi
);
547 static unsigned int free_old_xmit_skbs(struct virtnet_info
*vi
)
550 unsigned int len
, tot_sgs
= 0;
551 struct virtnet_stats __percpu
*stats
= this_cpu_ptr(vi
->stats
);
553 while ((skb
= virtqueue_get_buf(vi
->svq
, &len
)) != NULL
) {
554 pr_debug("Sent skb %p\n", skb
);
556 u64_stats_update_begin(&stats
->syncp
);
557 stats
->tx_bytes
+= skb
->len
;
559 u64_stats_update_end(&stats
->syncp
);
561 tot_sgs
+= skb_vnet_hdr(skb
)->num_sg
;
562 dev_kfree_skb_any(skb
);
567 static int xmit_skb(struct virtnet_info
*vi
, struct sk_buff
*skb
)
569 struct skb_vnet_hdr
*hdr
= skb_vnet_hdr(skb
);
570 const unsigned char *dest
= ((struct ethhdr
*)skb
->data
)->h_dest
;
572 pr_debug("%s: xmit %p %pM\n", vi
->dev
->name
, skb
, dest
);
574 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
575 hdr
->hdr
.flags
= VIRTIO_NET_HDR_F_NEEDS_CSUM
;
576 hdr
->hdr
.csum_start
= skb_checksum_start_offset(skb
);
577 hdr
->hdr
.csum_offset
= skb
->csum_offset
;
580 hdr
->hdr
.csum_offset
= hdr
->hdr
.csum_start
= 0;
583 if (skb_is_gso(skb
)) {
584 hdr
->hdr
.hdr_len
= skb_headlen(skb
);
585 hdr
->hdr
.gso_size
= skb_shinfo(skb
)->gso_size
;
586 if (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCPV4
)
587 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV4
;
588 else if (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCPV6
)
589 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV6
;
590 else if (skb_shinfo(skb
)->gso_type
& SKB_GSO_UDP
)
591 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_UDP
;
594 if (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCP_ECN
)
595 hdr
->hdr
.gso_type
|= VIRTIO_NET_HDR_GSO_ECN
;
597 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
598 hdr
->hdr
.gso_size
= hdr
->hdr
.hdr_len
= 0;
601 hdr
->mhdr
.num_buffers
= 0;
603 /* Encode metadata header at front. */
604 if (vi
->mergeable_rx_bufs
)
605 sg_set_buf(vi
->tx_sg
, &hdr
->mhdr
, sizeof hdr
->mhdr
);
607 sg_set_buf(vi
->tx_sg
, &hdr
->hdr
, sizeof hdr
->hdr
);
609 hdr
->num_sg
= skb_to_sgvec(skb
, vi
->tx_sg
+ 1, 0, skb
->len
) + 1;
610 return virtqueue_add_buf(vi
->svq
, vi
->tx_sg
, hdr
->num_sg
,
614 static netdev_tx_t
start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
616 struct virtnet_info
*vi
= netdev_priv(dev
);
619 /* Free up any pending old buffers before queueing new ones. */
620 free_old_xmit_skbs(vi
);
622 /* Try to transmit */
623 capacity
= xmit_skb(vi
, skb
);
625 /* This can happen with OOM and indirect buffers. */
626 if (unlikely(capacity
< 0)) {
627 if (net_ratelimit()) {
628 if (likely(capacity
== -ENOMEM
)) {
630 "TX queue failure: out of memory\n");
632 dev
->stats
.tx_fifo_errors
++;
634 "Unexpected TX queue failure: %d\n",
638 dev
->stats
.tx_dropped
++;
642 virtqueue_kick(vi
->svq
);
644 /* Don't wait up for transmitted skbs to be freed. */
648 /* Apparently nice girls don't return TX_BUSY; stop the queue
649 * before it gets out of hand. Naturally, this wastes entries. */
650 if (capacity
< 2+MAX_SKB_FRAGS
) {
651 netif_stop_queue(dev
);
652 if (unlikely(!virtqueue_enable_cb_delayed(vi
->svq
))) {
653 /* More just got used, free them then recheck. */
654 capacity
+= free_old_xmit_skbs(vi
);
655 if (capacity
>= 2+MAX_SKB_FRAGS
) {
656 netif_start_queue(dev
);
657 virtqueue_disable_cb(vi
->svq
);
665 static int virtnet_set_mac_address(struct net_device
*dev
, void *p
)
667 struct virtnet_info
*vi
= netdev_priv(dev
);
668 struct virtio_device
*vdev
= vi
->vdev
;
671 ret
= eth_mac_addr(dev
, p
);
675 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MAC
))
676 vdev
->config
->set(vdev
, offsetof(struct virtio_net_config
, mac
),
677 dev
->dev_addr
, dev
->addr_len
);
682 static struct rtnl_link_stats64
*virtnet_stats(struct net_device
*dev
,
683 struct rtnl_link_stats64
*tot
)
685 struct virtnet_info
*vi
= netdev_priv(dev
);
689 for_each_possible_cpu(cpu
) {
690 struct virtnet_stats __percpu
*stats
691 = per_cpu_ptr(vi
->stats
, cpu
);
692 u64 tpackets
, tbytes
, rpackets
, rbytes
;
695 start
= u64_stats_fetch_begin(&stats
->syncp
);
696 tpackets
= stats
->tx_packets
;
697 tbytes
= stats
->tx_bytes
;
698 rpackets
= stats
->rx_packets
;
699 rbytes
= stats
->rx_bytes
;
700 } while (u64_stats_fetch_retry(&stats
->syncp
, start
));
702 tot
->rx_packets
+= rpackets
;
703 tot
->tx_packets
+= tpackets
;
704 tot
->rx_bytes
+= rbytes
;
705 tot
->tx_bytes
+= tbytes
;
708 tot
->tx_dropped
= dev
->stats
.tx_dropped
;
709 tot
->tx_fifo_errors
= dev
->stats
.tx_fifo_errors
;
710 tot
->rx_dropped
= dev
->stats
.rx_dropped
;
711 tot
->rx_length_errors
= dev
->stats
.rx_length_errors
;
712 tot
->rx_frame_errors
= dev
->stats
.rx_frame_errors
;
717 #ifdef CONFIG_NET_POLL_CONTROLLER
718 static void virtnet_netpoll(struct net_device
*dev
)
720 struct virtnet_info
*vi
= netdev_priv(dev
);
722 napi_schedule(&vi
->napi
);
726 static int virtnet_open(struct net_device
*dev
)
728 struct virtnet_info
*vi
= netdev_priv(dev
);
730 /* Make sure we have some buffers: if oom use wq. */
731 if (!try_fill_recv(vi
, GFP_KERNEL
))
732 queue_delayed_work(system_nrt_wq
, &vi
->refill
, 0);
734 virtnet_napi_enable(vi
);
739 * Send command via the control virtqueue and check status. Commands
740 * supported by the hypervisor, as indicated by feature bits, should
741 * never fail unless improperly formated.
743 static bool virtnet_send_command(struct virtnet_info
*vi
, u8
class, u8 cmd
,
744 struct scatterlist
*data
, int out
, int in
)
746 struct scatterlist
*s
, sg
[VIRTNET_SEND_COMMAND_SG_MAX
+ 2];
747 struct virtio_net_ctrl_hdr ctrl
;
748 virtio_net_ctrl_ack status
= ~0;
752 /* Caller should know better */
753 BUG_ON(!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
) ||
754 (out
+ in
> VIRTNET_SEND_COMMAND_SG_MAX
));
756 out
++; /* Add header */
757 in
++; /* Add return status */
762 sg_init_table(sg
, out
+ in
);
764 sg_set_buf(&sg
[0], &ctrl
, sizeof(ctrl
));
765 for_each_sg(data
, s
, out
+ in
- 2, i
)
766 sg_set_buf(&sg
[i
+ 1], sg_virt(s
), s
->length
);
767 sg_set_buf(&sg
[out
+ in
- 1], &status
, sizeof(status
));
769 BUG_ON(virtqueue_add_buf(vi
->cvq
, sg
, out
, in
, vi
) < 0);
771 virtqueue_kick(vi
->cvq
);
774 * Spin for a response, the kick causes an ioport write, trapping
775 * into the hypervisor, so the request should be handled immediately.
777 while (!virtqueue_get_buf(vi
->cvq
, &tmp
))
780 return status
== VIRTIO_NET_OK
;
783 static int virtnet_close(struct net_device
*dev
)
785 struct virtnet_info
*vi
= netdev_priv(dev
);
787 /* Make sure refill_work doesn't re-enable napi! */
788 cancel_delayed_work_sync(&vi
->refill
);
789 napi_disable(&vi
->napi
);
794 static void virtnet_set_rx_mode(struct net_device
*dev
)
796 struct virtnet_info
*vi
= netdev_priv(dev
);
797 struct scatterlist sg
[2];
798 u8 promisc
, allmulti
;
799 struct virtio_net_ctrl_mac
*mac_data
;
800 struct netdev_hw_addr
*ha
;
806 /* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */
807 if (!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_RX
))
810 promisc
= ((dev
->flags
& IFF_PROMISC
) != 0);
811 allmulti
= ((dev
->flags
& IFF_ALLMULTI
) != 0);
813 sg_init_one(sg
, &promisc
, sizeof(promisc
));
815 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_RX
,
816 VIRTIO_NET_CTRL_RX_PROMISC
,
818 dev_warn(&dev
->dev
, "Failed to %sable promisc mode.\n",
819 promisc
? "en" : "dis");
821 sg_init_one(sg
, &allmulti
, sizeof(allmulti
));
823 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_RX
,
824 VIRTIO_NET_CTRL_RX_ALLMULTI
,
826 dev_warn(&dev
->dev
, "Failed to %sable allmulti mode.\n",
827 allmulti
? "en" : "dis");
829 uc_count
= netdev_uc_count(dev
);
830 mc_count
= netdev_mc_count(dev
);
831 /* MAC filter - use one buffer for both lists */
832 buf
= kzalloc(((uc_count
+ mc_count
) * ETH_ALEN
) +
833 (2 * sizeof(mac_data
->entries
)), GFP_ATOMIC
);
836 dev_warn(&dev
->dev
, "No memory for MAC address buffer\n");
840 sg_init_table(sg
, 2);
842 /* Store the unicast list and count in the front of the buffer */
843 mac_data
->entries
= uc_count
;
845 netdev_for_each_uc_addr(ha
, dev
)
846 memcpy(&mac_data
->macs
[i
++][0], ha
->addr
, ETH_ALEN
);
848 sg_set_buf(&sg
[0], mac_data
,
849 sizeof(mac_data
->entries
) + (uc_count
* ETH_ALEN
));
851 /* multicast list and count fill the end */
852 mac_data
= (void *)&mac_data
->macs
[uc_count
][0];
854 mac_data
->entries
= mc_count
;
856 netdev_for_each_mc_addr(ha
, dev
)
857 memcpy(&mac_data
->macs
[i
++][0], ha
->addr
, ETH_ALEN
);
859 sg_set_buf(&sg
[1], mac_data
,
860 sizeof(mac_data
->entries
) + (mc_count
* ETH_ALEN
));
862 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MAC
,
863 VIRTIO_NET_CTRL_MAC_TABLE_SET
,
865 dev_warn(&dev
->dev
, "Failed to set MAC fitler table.\n");
870 static int virtnet_vlan_rx_add_vid(struct net_device
*dev
, u16 vid
)
872 struct virtnet_info
*vi
= netdev_priv(dev
);
873 struct scatterlist sg
;
875 sg_init_one(&sg
, &vid
, sizeof(vid
));
877 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_VLAN
,
878 VIRTIO_NET_CTRL_VLAN_ADD
, &sg
, 1, 0))
879 dev_warn(&dev
->dev
, "Failed to add VLAN ID %d.\n", vid
);
883 static int virtnet_vlan_rx_kill_vid(struct net_device
*dev
, u16 vid
)
885 struct virtnet_info
*vi
= netdev_priv(dev
);
886 struct scatterlist sg
;
888 sg_init_one(&sg
, &vid
, sizeof(vid
));
890 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_VLAN
,
891 VIRTIO_NET_CTRL_VLAN_DEL
, &sg
, 1, 0))
892 dev_warn(&dev
->dev
, "Failed to kill VLAN ID %d.\n", vid
);
896 static void virtnet_get_ringparam(struct net_device
*dev
,
897 struct ethtool_ringparam
*ring
)
899 struct virtnet_info
*vi
= netdev_priv(dev
);
901 ring
->rx_max_pending
= virtqueue_get_vring_size(vi
->rvq
);
902 ring
->tx_max_pending
= virtqueue_get_vring_size(vi
->svq
);
903 ring
->rx_pending
= ring
->rx_max_pending
;
904 ring
->tx_pending
= ring
->tx_max_pending
;
909 static void virtnet_get_drvinfo(struct net_device
*dev
,
910 struct ethtool_drvinfo
*info
)
912 struct virtnet_info
*vi
= netdev_priv(dev
);
913 struct virtio_device
*vdev
= vi
->vdev
;
915 strlcpy(info
->driver
, KBUILD_MODNAME
, sizeof(info
->driver
));
916 strlcpy(info
->version
, VIRTNET_DRIVER_VERSION
, sizeof(info
->version
));
917 strlcpy(info
->bus_info
, virtio_bus_name(vdev
), sizeof(info
->bus_info
));
921 static const struct ethtool_ops virtnet_ethtool_ops
= {
922 .get_drvinfo
= virtnet_get_drvinfo
,
923 .get_link
= ethtool_op_get_link
,
924 .get_ringparam
= virtnet_get_ringparam
,
928 #define MAX_MTU 65535
930 static int virtnet_change_mtu(struct net_device
*dev
, int new_mtu
)
932 if (new_mtu
< MIN_MTU
|| new_mtu
> MAX_MTU
)
938 static const struct net_device_ops virtnet_netdev
= {
939 .ndo_open
= virtnet_open
,
940 .ndo_stop
= virtnet_close
,
941 .ndo_start_xmit
= start_xmit
,
942 .ndo_validate_addr
= eth_validate_addr
,
943 .ndo_set_mac_address
= virtnet_set_mac_address
,
944 .ndo_set_rx_mode
= virtnet_set_rx_mode
,
945 .ndo_change_mtu
= virtnet_change_mtu
,
946 .ndo_get_stats64
= virtnet_stats
,
947 .ndo_vlan_rx_add_vid
= virtnet_vlan_rx_add_vid
,
948 .ndo_vlan_rx_kill_vid
= virtnet_vlan_rx_kill_vid
,
949 #ifdef CONFIG_NET_POLL_CONTROLLER
950 .ndo_poll_controller
= virtnet_netpoll
,
954 static void virtnet_update_status(struct virtnet_info
*vi
)
958 if (virtio_config_val(vi
->vdev
, VIRTIO_NET_F_STATUS
,
959 offsetof(struct virtio_net_config
, status
),
963 /* Ignore unknown (future) status bits */
964 v
&= VIRTIO_NET_S_LINK_UP
;
971 if (vi
->status
& VIRTIO_NET_S_LINK_UP
) {
972 netif_carrier_on(vi
->dev
);
973 netif_wake_queue(vi
->dev
);
975 netif_carrier_off(vi
->dev
);
976 netif_stop_queue(vi
->dev
);
980 static void virtnet_config_changed(struct virtio_device
*vdev
)
982 struct virtnet_info
*vi
= vdev
->priv
;
984 virtnet_update_status(vi
);
987 static int virtnet_probe(struct virtio_device
*vdev
)
990 struct net_device
*dev
;
991 struct virtnet_info
*vi
;
992 struct virtqueue
*vqs
[3];
993 vq_callback_t
*callbacks
[] = { skb_recv_done
, skb_xmit_done
, NULL
};
994 const char *names
[] = { "input", "output", "control" };
997 /* Allocate ourselves a network device with room for our info */
998 dev
= alloc_etherdev(sizeof(struct virtnet_info
));
1002 /* Set up network device as normal. */
1003 dev
->priv_flags
|= IFF_UNICAST_FLT
;
1004 dev
->netdev_ops
= &virtnet_netdev
;
1005 dev
->features
= NETIF_F_HIGHDMA
;
1007 SET_ETHTOOL_OPS(dev
, &virtnet_ethtool_ops
);
1008 SET_NETDEV_DEV(dev
, &vdev
->dev
);
1010 /* Do we support "hardware" checksums? */
1011 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CSUM
)) {
1012 /* This opens up the world of extra features. */
1013 dev
->hw_features
|= NETIF_F_HW_CSUM
|NETIF_F_SG
|NETIF_F_FRAGLIST
;
1015 dev
->features
|= NETIF_F_HW_CSUM
|NETIF_F_SG
|NETIF_F_FRAGLIST
;
1017 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GSO
)) {
1018 dev
->hw_features
|= NETIF_F_TSO
| NETIF_F_UFO
1019 | NETIF_F_TSO_ECN
| NETIF_F_TSO6
;
1021 /* Individual feature bits: what can host handle? */
1022 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_TSO4
))
1023 dev
->hw_features
|= NETIF_F_TSO
;
1024 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_TSO6
))
1025 dev
->hw_features
|= NETIF_F_TSO6
;
1026 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_ECN
))
1027 dev
->hw_features
|= NETIF_F_TSO_ECN
;
1028 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_UFO
))
1029 dev
->hw_features
|= NETIF_F_UFO
;
1032 dev
->features
|= dev
->hw_features
& (NETIF_F_ALL_TSO
|NETIF_F_UFO
);
1033 /* (!csum && gso) case will be fixed by register_netdev() */
1036 /* Configuration may specify what MAC to use. Otherwise random. */
1037 if (virtio_config_val_len(vdev
, VIRTIO_NET_F_MAC
,
1038 offsetof(struct virtio_net_config
, mac
),
1039 dev
->dev_addr
, dev
->addr_len
) < 0)
1040 random_ether_addr(dev
->dev_addr
);
1042 /* Set up our device-specific information */
1043 vi
= netdev_priv(dev
);
1044 netif_napi_add(dev
, &vi
->napi
, virtnet_poll
, napi_weight
);
1049 vi
->stats
= alloc_percpu(struct virtnet_stats
);
1051 if (vi
->stats
== NULL
)
1054 INIT_DELAYED_WORK(&vi
->refill
, refill_work
);
1055 sg_init_table(vi
->rx_sg
, ARRAY_SIZE(vi
->rx_sg
));
1056 sg_init_table(vi
->tx_sg
, ARRAY_SIZE(vi
->tx_sg
));
1058 /* If we can receive ANY GSO packets, we must allocate large ones. */
1059 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_TSO4
) ||
1060 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_TSO6
) ||
1061 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_ECN
))
1062 vi
->big_packets
= true;
1064 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MRG_RXBUF
))
1065 vi
->mergeable_rx_bufs
= true;
1067 /* We expect two virtqueues, receive then send,
1068 * and optionally control. */
1069 nvqs
= virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
) ? 3 : 2;
1071 err
= vdev
->config
->find_vqs(vdev
, nvqs
, vqs
, callbacks
, names
);
1078 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
)) {
1081 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VLAN
))
1082 dev
->features
|= NETIF_F_HW_VLAN_FILTER
;
1085 err
= register_netdev(dev
);
1087 pr_debug("virtio_net: registering device failed\n");
1091 /* Last of all, set up some receive buffers. */
1092 try_fill_recv(vi
, GFP_KERNEL
);
1094 /* If we didn't even get one input buffer, we're useless. */
1100 /* Assume link up if device can't report link status,
1101 otherwise get link status from config. */
1102 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_STATUS
)) {
1103 netif_carrier_off(dev
);
1104 virtnet_update_status(vi
);
1106 vi
->status
= VIRTIO_NET_S_LINK_UP
;
1107 netif_carrier_on(dev
);
1110 pr_debug("virtnet: registered device %s\n", dev
->name
);
1114 unregister_netdev(dev
);
1116 vdev
->config
->del_vqs(vdev
);
1118 free_percpu(vi
->stats
);
1124 static void free_unused_bufs(struct virtnet_info
*vi
)
1128 buf
= virtqueue_detach_unused_buf(vi
->svq
);
1134 buf
= virtqueue_detach_unused_buf(vi
->rvq
);
1137 if (vi
->mergeable_rx_bufs
|| vi
->big_packets
)
1138 give_pages(vi
, buf
);
1143 BUG_ON(vi
->num
!= 0);
1146 static void __devexit
virtnet_remove(struct virtio_device
*vdev
)
1148 struct virtnet_info
*vi
= vdev
->priv
;
1150 /* Stop all the virtqueues. */
1151 vdev
->config
->reset(vdev
);
1153 unregister_netdev(vi
->dev
);
1155 /* Free unused buffers in both send and recv, if any. */
1156 free_unused_bufs(vi
);
1158 vdev
->config
->del_vqs(vi
->vdev
);
1161 __free_pages(get_a_page(vi
, GFP_KERNEL
), 0);
1163 free_percpu(vi
->stats
);
1164 free_netdev(vi
->dev
);
1167 static struct virtio_device_id id_table
[] = {
1168 { VIRTIO_ID_NET
, VIRTIO_DEV_ANY_ID
},
1172 static unsigned int features
[] = {
1173 VIRTIO_NET_F_CSUM
, VIRTIO_NET_F_GUEST_CSUM
,
1174 VIRTIO_NET_F_GSO
, VIRTIO_NET_F_MAC
,
1175 VIRTIO_NET_F_HOST_TSO4
, VIRTIO_NET_F_HOST_UFO
, VIRTIO_NET_F_HOST_TSO6
,
1176 VIRTIO_NET_F_HOST_ECN
, VIRTIO_NET_F_GUEST_TSO4
, VIRTIO_NET_F_GUEST_TSO6
,
1177 VIRTIO_NET_F_GUEST_ECN
, VIRTIO_NET_F_GUEST_UFO
,
1178 VIRTIO_NET_F_MRG_RXBUF
, VIRTIO_NET_F_STATUS
, VIRTIO_NET_F_CTRL_VQ
,
1179 VIRTIO_NET_F_CTRL_RX
, VIRTIO_NET_F_CTRL_VLAN
,
1182 static struct virtio_driver virtio_net_driver
= {
1183 .feature_table
= features
,
1184 .feature_table_size
= ARRAY_SIZE(features
),
1185 .driver
.name
= KBUILD_MODNAME
,
1186 .driver
.owner
= THIS_MODULE
,
1187 .id_table
= id_table
,
1188 .probe
= virtnet_probe
,
1189 .remove
= __devexit_p(virtnet_remove
),
1190 .config_changed
= virtnet_config_changed
,
1193 static int __init
init(void)
1195 return register_virtio_driver(&virtio_net_driver
);
1198 static void __exit
fini(void)
1200 unregister_virtio_driver(&virtio_net_driver
);
1205 MODULE_DEVICE_TABLE(virtio
, id_table
);
1206 MODULE_DESCRIPTION("Virtio network driver");
1207 MODULE_LICENSE("GPL");