1 /* A network driver using virtio.
3 * Copyright 2007 Rusty Russell <rusty@rustcorp.com.au> IBM Corporation
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 #include <linux/netdevice.h>
21 #include <linux/etherdevice.h>
22 #include <linux/ethtool.h>
23 #include <linux/module.h>
24 #include <linux/virtio.h>
25 #include <linux/virtio_net.h>
26 #include <linux/scatterlist.h>
27 #include <linux/if_vlan.h>
28 #include <linux/slab.h>
30 static int napi_weight
= 128;
31 module_param(napi_weight
, int, 0444);
33 static bool csum
= true, gso
= true;
34 module_param(csum
, bool, 0444);
35 module_param(gso
, bool, 0444);
37 /* FIXME: MTU in config. */
38 #define MAX_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
39 #define GOOD_COPY_LEN 128
41 #define VIRTNET_SEND_COMMAND_SG_MAX 2
42 #define VIRTNET_DRIVER_VERSION "1.0.0"
44 struct virtnet_stats
{
45 struct u64_stats_sync tx_syncp
;
46 struct u64_stats_sync rx_syncp
;
55 struct virtio_device
*vdev
;
56 struct virtqueue
*rvq
, *svq
, *cvq
;
57 struct net_device
*dev
;
58 struct napi_struct napi
;
61 /* Number of input buffers, and max we've ever had. */
62 unsigned int num
, max
;
64 /* I like... big packets and I cannot lie! */
67 /* Host will merge rx buffers for big packets (shake it! shake it!) */
68 bool mergeable_rx_bufs
;
70 /* enable config space updates */
73 /* Active statistics */
74 struct virtnet_stats __percpu
*stats
;
76 /* Work struct for refilling if we run low on memory. */
77 struct delayed_work refill
;
79 /* Work struct for config space updates */
80 struct work_struct config_work
;
82 /* Lock for config space updates */
83 struct mutex config_lock
;
85 /* Chain pages by the private ptr. */
88 /* fragments + linear part + virtio header */
89 struct scatterlist rx_sg
[MAX_SKB_FRAGS
+ 2];
90 struct scatterlist tx_sg
[MAX_SKB_FRAGS
+ 2];
95 struct virtio_net_hdr hdr
;
96 struct virtio_net_hdr_mrg_rxbuf mhdr
;
101 struct padded_vnet_hdr
{
102 struct virtio_net_hdr hdr
;
104 * virtio_net_hdr should be in a separated sg buffer because of a
105 * QEMU bug, and data sg buffer shares same page with this header sg.
106 * This padding makes next sg 16 byte aligned after virtio_net_hdr.
111 static inline struct skb_vnet_hdr
*skb_vnet_hdr(struct sk_buff
*skb
)
113 return (struct skb_vnet_hdr
*)skb
->cb
;
117 * private is used to chain pages for big packets, put the whole
118 * most recent used list in the beginning for reuse
120 static void give_pages(struct virtnet_info
*vi
, struct page
*page
)
124 /* Find end of list, sew whole thing into vi->pages. */
125 for (end
= page
; end
->private; end
= (struct page
*)end
->private);
126 end
->private = (unsigned long)vi
->pages
;
130 static struct page
*get_a_page(struct virtnet_info
*vi
, gfp_t gfp_mask
)
132 struct page
*p
= vi
->pages
;
135 vi
->pages
= (struct page
*)p
->private;
136 /* clear private here, it is used to chain pages */
139 p
= alloc_page(gfp_mask
);
143 static void skb_xmit_done(struct virtqueue
*svq
)
145 struct virtnet_info
*vi
= svq
->vdev
->priv
;
147 /* Suppress further interrupts. */
148 virtqueue_disable_cb(svq
);
150 /* We were probably waiting for more output buffers. */
151 netif_wake_queue(vi
->dev
);
154 static void set_skb_frag(struct sk_buff
*skb
, struct page
*page
,
155 unsigned int offset
, unsigned int *len
)
157 int size
= min((unsigned)PAGE_SIZE
- offset
, *len
);
158 int i
= skb_shinfo(skb
)->nr_frags
;
160 __skb_fill_page_desc(skb
, i
, page
, offset
, size
);
162 skb
->data_len
+= size
;
164 skb
->truesize
+= PAGE_SIZE
;
165 skb_shinfo(skb
)->nr_frags
++;
169 /* Called from bottom half context */
170 static struct sk_buff
*page_to_skb(struct virtnet_info
*vi
,
171 struct page
*page
, unsigned int len
)
174 struct skb_vnet_hdr
*hdr
;
175 unsigned int copy
, hdr_len
, offset
;
178 p
= page_address(page
);
180 /* copy small packet so we can reuse these pages for small data */
181 skb
= netdev_alloc_skb_ip_align(vi
->dev
, GOOD_COPY_LEN
);
185 hdr
= skb_vnet_hdr(skb
);
187 if (vi
->mergeable_rx_bufs
) {
188 hdr_len
= sizeof hdr
->mhdr
;
191 hdr_len
= sizeof hdr
->hdr
;
192 offset
= sizeof(struct padded_vnet_hdr
);
195 memcpy(hdr
, p
, hdr_len
);
201 if (copy
> skb_tailroom(skb
))
202 copy
= skb_tailroom(skb
);
203 memcpy(skb_put(skb
, copy
), p
, copy
);
209 * Verify that we can indeed put this data into a skb.
210 * This is here to handle cases when the device erroneously
211 * tries to receive more than is possible. This is usually
212 * the case of a broken device.
214 if (unlikely(len
> MAX_SKB_FRAGS
* PAGE_SIZE
)) {
216 pr_debug("%s: too much data\n", skb
->dev
->name
);
222 set_skb_frag(skb
, page
, offset
, &len
);
223 page
= (struct page
*)page
->private;
228 give_pages(vi
, page
);
233 static int receive_mergeable(struct virtnet_info
*vi
, struct sk_buff
*skb
)
235 struct skb_vnet_hdr
*hdr
= skb_vnet_hdr(skb
);
239 num_buf
= hdr
->mhdr
.num_buffers
;
241 i
= skb_shinfo(skb
)->nr_frags
;
242 if (i
>= MAX_SKB_FRAGS
) {
243 pr_debug("%s: packet too long\n", skb
->dev
->name
);
244 skb
->dev
->stats
.rx_length_errors
++;
247 page
= virtqueue_get_buf(vi
->rvq
, &len
);
249 pr_debug("%s: rx error: %d buffers missing\n",
250 skb
->dev
->name
, hdr
->mhdr
.num_buffers
);
251 skb
->dev
->stats
.rx_length_errors
++;
258 set_skb_frag(skb
, page
, 0, &len
);
265 static void receive_buf(struct net_device
*dev
, void *buf
, unsigned int len
)
267 struct virtnet_info
*vi
= netdev_priv(dev
);
268 struct virtnet_stats
*stats
= this_cpu_ptr(vi
->stats
);
271 struct skb_vnet_hdr
*hdr
;
273 if (unlikely(len
< sizeof(struct virtio_net_hdr
) + ETH_HLEN
)) {
274 pr_debug("%s: short packet %i\n", dev
->name
, len
);
275 dev
->stats
.rx_length_errors
++;
276 if (vi
->mergeable_rx_bufs
|| vi
->big_packets
)
283 if (!vi
->mergeable_rx_bufs
&& !vi
->big_packets
) {
285 len
-= sizeof(struct virtio_net_hdr
);
289 skb
= page_to_skb(vi
, page
, len
);
290 if (unlikely(!skb
)) {
291 dev
->stats
.rx_dropped
++;
292 give_pages(vi
, page
);
295 if (vi
->mergeable_rx_bufs
)
296 if (receive_mergeable(vi
, skb
)) {
302 hdr
= skb_vnet_hdr(skb
);
304 u64_stats_update_begin(&stats
->rx_syncp
);
305 stats
->rx_bytes
+= skb
->len
;
307 u64_stats_update_end(&stats
->rx_syncp
);
309 if (hdr
->hdr
.flags
& VIRTIO_NET_HDR_F_NEEDS_CSUM
) {
310 pr_debug("Needs csum!\n");
311 if (!skb_partial_csum_set(skb
,
313 hdr
->hdr
.csum_offset
))
315 } else if (hdr
->hdr
.flags
& VIRTIO_NET_HDR_F_DATA_VALID
) {
316 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
319 skb
->protocol
= eth_type_trans(skb
, dev
);
320 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
321 ntohs(skb
->protocol
), skb
->len
, skb
->pkt_type
);
323 if (hdr
->hdr
.gso_type
!= VIRTIO_NET_HDR_GSO_NONE
) {
325 switch (hdr
->hdr
.gso_type
& ~VIRTIO_NET_HDR_GSO_ECN
) {
326 case VIRTIO_NET_HDR_GSO_TCPV4
:
327 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV4
;
329 case VIRTIO_NET_HDR_GSO_UDP
:
330 skb_shinfo(skb
)->gso_type
= SKB_GSO_UDP
;
332 case VIRTIO_NET_HDR_GSO_TCPV6
:
333 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV6
;
337 printk(KERN_WARNING
"%s: bad gso type %u.\n",
338 dev
->name
, hdr
->hdr
.gso_type
);
342 if (hdr
->hdr
.gso_type
& VIRTIO_NET_HDR_GSO_ECN
)
343 skb_shinfo(skb
)->gso_type
|= SKB_GSO_TCP_ECN
;
345 skb_shinfo(skb
)->gso_size
= hdr
->hdr
.gso_size
;
346 if (skb_shinfo(skb
)->gso_size
== 0) {
348 printk(KERN_WARNING
"%s: zero gso size.\n",
353 /* Header must be checked, and gso_segs computed. */
354 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
355 skb_shinfo(skb
)->gso_segs
= 0;
358 netif_receive_skb(skb
);
362 dev
->stats
.rx_frame_errors
++;
366 static int add_recvbuf_small(struct virtnet_info
*vi
, gfp_t gfp
)
369 struct skb_vnet_hdr
*hdr
;
372 skb
= __netdev_alloc_skb_ip_align(vi
->dev
, MAX_PACKET_LEN
, gfp
);
376 skb_put(skb
, MAX_PACKET_LEN
);
378 hdr
= skb_vnet_hdr(skb
);
379 sg_set_buf(vi
->rx_sg
, &hdr
->hdr
, sizeof hdr
->hdr
);
381 skb_to_sgvec(skb
, vi
->rx_sg
+ 1, 0, skb
->len
);
383 err
= virtqueue_add_buf(vi
->rvq
, vi
->rx_sg
, 0, 2, skb
, gfp
);
390 static int add_recvbuf_big(struct virtnet_info
*vi
, gfp_t gfp
)
392 struct page
*first
, *list
= NULL
;
396 /* page in vi->rx_sg[MAX_SKB_FRAGS + 1] is list tail */
397 for (i
= MAX_SKB_FRAGS
+ 1; i
> 1; --i
) {
398 first
= get_a_page(vi
, gfp
);
401 give_pages(vi
, list
);
404 sg_set_buf(&vi
->rx_sg
[i
], page_address(first
), PAGE_SIZE
);
406 /* chain new page in list head to match sg */
407 first
->private = (unsigned long)list
;
411 first
= get_a_page(vi
, gfp
);
413 give_pages(vi
, list
);
416 p
= page_address(first
);
418 /* vi->rx_sg[0], vi->rx_sg[1] share the same page */
419 /* a separated vi->rx_sg[0] for virtio_net_hdr only due to QEMU bug */
420 sg_set_buf(&vi
->rx_sg
[0], p
, sizeof(struct virtio_net_hdr
));
422 /* vi->rx_sg[1] for data packet, from offset */
423 offset
= sizeof(struct padded_vnet_hdr
);
424 sg_set_buf(&vi
->rx_sg
[1], p
+ offset
, PAGE_SIZE
- offset
);
426 /* chain first in list head */
427 first
->private = (unsigned long)list
;
428 err
= virtqueue_add_buf(vi
->rvq
, vi
->rx_sg
, 0, MAX_SKB_FRAGS
+ 2,
431 give_pages(vi
, first
);
436 static int add_recvbuf_mergeable(struct virtnet_info
*vi
, gfp_t gfp
)
441 page
= get_a_page(vi
, gfp
);
445 sg_init_one(vi
->rx_sg
, page_address(page
), PAGE_SIZE
);
447 err
= virtqueue_add_buf(vi
->rvq
, vi
->rx_sg
, 0, 1, page
, gfp
);
449 give_pages(vi
, page
);
455 * Returns false if we couldn't fill entirely (OOM).
457 * Normally run in the receive path, but can also be run from ndo_open
458 * before we're receiving packets, or from refill_work which is
459 * careful to disable receiving (using napi_disable).
461 static bool try_fill_recv(struct virtnet_info
*vi
, gfp_t gfp
)
467 if (vi
->mergeable_rx_bufs
)
468 err
= add_recvbuf_mergeable(vi
, gfp
);
469 else if (vi
->big_packets
)
470 err
= add_recvbuf_big(vi
, gfp
);
472 err
= add_recvbuf_small(vi
, gfp
);
474 oom
= err
== -ENOMEM
;
479 if (unlikely(vi
->num
> vi
->max
))
481 virtqueue_kick(vi
->rvq
);
485 static void skb_recv_done(struct virtqueue
*rvq
)
487 struct virtnet_info
*vi
= rvq
->vdev
->priv
;
488 /* Schedule NAPI, Suppress further interrupts if successful. */
489 if (napi_schedule_prep(&vi
->napi
)) {
490 virtqueue_disable_cb(rvq
);
491 __napi_schedule(&vi
->napi
);
495 static void virtnet_napi_enable(struct virtnet_info
*vi
)
497 napi_enable(&vi
->napi
);
499 /* If all buffers were filled by other side before we napi_enabled, we
500 * won't get another interrupt, so process any outstanding packets
501 * now. virtnet_poll wants re-enable the queue, so we disable here.
502 * We synchronize against interrupts via NAPI_STATE_SCHED */
503 if (napi_schedule_prep(&vi
->napi
)) {
504 virtqueue_disable_cb(vi
->rvq
);
506 __napi_schedule(&vi
->napi
);
511 static void refill_work(struct work_struct
*work
)
513 struct virtnet_info
*vi
;
516 vi
= container_of(work
, struct virtnet_info
, refill
.work
);
517 napi_disable(&vi
->napi
);
518 still_empty
= !try_fill_recv(vi
, GFP_KERNEL
);
519 virtnet_napi_enable(vi
);
521 /* In theory, this can happen: if we don't get any buffers in
522 * we will *never* try to fill again. */
524 queue_delayed_work(system_nrt_wq
, &vi
->refill
, HZ
/2);
527 static int virtnet_poll(struct napi_struct
*napi
, int budget
)
529 struct virtnet_info
*vi
= container_of(napi
, struct virtnet_info
, napi
);
531 unsigned int len
, received
= 0;
534 while (received
< budget
&&
535 (buf
= virtqueue_get_buf(vi
->rvq
, &len
)) != NULL
) {
536 receive_buf(vi
->dev
, buf
, len
);
541 if (vi
->num
< vi
->max
/ 2) {
542 if (!try_fill_recv(vi
, GFP_ATOMIC
))
543 queue_delayed_work(system_nrt_wq
, &vi
->refill
, 0);
546 /* Out of packets? */
547 if (received
< budget
) {
549 if (unlikely(!virtqueue_enable_cb(vi
->rvq
)) &&
550 napi_schedule_prep(napi
)) {
551 virtqueue_disable_cb(vi
->rvq
);
552 __napi_schedule(napi
);
560 static unsigned int free_old_xmit_skbs(struct virtnet_info
*vi
)
563 unsigned int len
, tot_sgs
= 0;
564 struct virtnet_stats
*stats
= this_cpu_ptr(vi
->stats
);
566 while ((skb
= virtqueue_get_buf(vi
->svq
, &len
)) != NULL
) {
567 pr_debug("Sent skb %p\n", skb
);
569 u64_stats_update_begin(&stats
->tx_syncp
);
570 stats
->tx_bytes
+= skb
->len
;
572 u64_stats_update_end(&stats
->tx_syncp
);
574 tot_sgs
+= skb_vnet_hdr(skb
)->num_sg
;
575 dev_kfree_skb_any(skb
);
580 static int xmit_skb(struct virtnet_info
*vi
, struct sk_buff
*skb
)
582 struct skb_vnet_hdr
*hdr
= skb_vnet_hdr(skb
);
583 const unsigned char *dest
= ((struct ethhdr
*)skb
->data
)->h_dest
;
585 pr_debug("%s: xmit %p %pM\n", vi
->dev
->name
, skb
, dest
);
587 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
588 hdr
->hdr
.flags
= VIRTIO_NET_HDR_F_NEEDS_CSUM
;
589 hdr
->hdr
.csum_start
= skb_checksum_start_offset(skb
);
590 hdr
->hdr
.csum_offset
= skb
->csum_offset
;
593 hdr
->hdr
.csum_offset
= hdr
->hdr
.csum_start
= 0;
596 if (skb_is_gso(skb
)) {
597 hdr
->hdr
.hdr_len
= skb_headlen(skb
);
598 hdr
->hdr
.gso_size
= skb_shinfo(skb
)->gso_size
;
599 if (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCPV4
)
600 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV4
;
601 else if (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCPV6
)
602 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_TCPV6
;
603 else if (skb_shinfo(skb
)->gso_type
& SKB_GSO_UDP
)
604 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_UDP
;
607 if (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCP_ECN
)
608 hdr
->hdr
.gso_type
|= VIRTIO_NET_HDR_GSO_ECN
;
610 hdr
->hdr
.gso_type
= VIRTIO_NET_HDR_GSO_NONE
;
611 hdr
->hdr
.gso_size
= hdr
->hdr
.hdr_len
= 0;
614 hdr
->mhdr
.num_buffers
= 0;
616 /* Encode metadata header at front. */
617 if (vi
->mergeable_rx_bufs
)
618 sg_set_buf(vi
->tx_sg
, &hdr
->mhdr
, sizeof hdr
->mhdr
);
620 sg_set_buf(vi
->tx_sg
, &hdr
->hdr
, sizeof hdr
->hdr
);
622 hdr
->num_sg
= skb_to_sgvec(skb
, vi
->tx_sg
+ 1, 0, skb
->len
) + 1;
623 return virtqueue_add_buf(vi
->svq
, vi
->tx_sg
, hdr
->num_sg
,
627 static netdev_tx_t
start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
629 struct virtnet_info
*vi
= netdev_priv(dev
);
632 /* Free up any pending old buffers before queueing new ones. */
633 free_old_xmit_skbs(vi
);
635 /* Try to transmit */
636 capacity
= xmit_skb(vi
, skb
);
638 /* This can happen with OOM and indirect buffers. */
639 if (unlikely(capacity
< 0)) {
640 if (likely(capacity
== -ENOMEM
)) {
643 "TX queue failure: out of memory\n");
645 dev
->stats
.tx_fifo_errors
++;
648 "Unexpected TX queue failure: %d\n",
651 dev
->stats
.tx_dropped
++;
655 virtqueue_kick(vi
->svq
);
657 /* Don't wait up for transmitted skbs to be freed. */
661 /* Apparently nice girls don't return TX_BUSY; stop the queue
662 * before it gets out of hand. Naturally, this wastes entries. */
663 if (capacity
< 2+MAX_SKB_FRAGS
) {
664 netif_stop_queue(dev
);
665 if (unlikely(!virtqueue_enable_cb_delayed(vi
->svq
))) {
666 /* More just got used, free them then recheck. */
667 capacity
+= free_old_xmit_skbs(vi
);
668 if (capacity
>= 2+MAX_SKB_FRAGS
) {
669 netif_start_queue(dev
);
670 virtqueue_disable_cb(vi
->svq
);
678 static int virtnet_set_mac_address(struct net_device
*dev
, void *p
)
680 struct virtnet_info
*vi
= netdev_priv(dev
);
681 struct virtio_device
*vdev
= vi
->vdev
;
682 struct sockaddr
*addr
= p
;
684 if (!is_valid_ether_addr(addr
->sa_data
))
685 return -EADDRNOTAVAIL
;
686 memcpy(dev
->dev_addr
, addr
->sa_data
, ETH_ALEN
);
687 dev
->addr_assign_type
&= ~NET_ADDR_RANDOM
;
689 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MAC
))
690 vdev
->config
->set(vdev
, offsetof(struct virtio_net_config
, mac
),
691 dev
->dev_addr
, dev
->addr_len
);
696 static struct rtnl_link_stats64
*virtnet_stats(struct net_device
*dev
,
697 struct rtnl_link_stats64
*tot
)
699 struct virtnet_info
*vi
= netdev_priv(dev
);
703 for_each_possible_cpu(cpu
) {
704 struct virtnet_stats
*stats
= per_cpu_ptr(vi
->stats
, cpu
);
705 u64 tpackets
, tbytes
, rpackets
, rbytes
;
708 start
= u64_stats_fetch_begin(&stats
->tx_syncp
);
709 tpackets
= stats
->tx_packets
;
710 tbytes
= stats
->tx_bytes
;
711 } while (u64_stats_fetch_retry(&stats
->tx_syncp
, start
));
714 start
= u64_stats_fetch_begin(&stats
->rx_syncp
);
715 rpackets
= stats
->rx_packets
;
716 rbytes
= stats
->rx_bytes
;
717 } while (u64_stats_fetch_retry(&stats
->rx_syncp
, start
));
719 tot
->rx_packets
+= rpackets
;
720 tot
->tx_packets
+= tpackets
;
721 tot
->rx_bytes
+= rbytes
;
722 tot
->tx_bytes
+= tbytes
;
725 tot
->tx_dropped
= dev
->stats
.tx_dropped
;
726 tot
->tx_fifo_errors
= dev
->stats
.tx_fifo_errors
;
727 tot
->rx_dropped
= dev
->stats
.rx_dropped
;
728 tot
->rx_length_errors
= dev
->stats
.rx_length_errors
;
729 tot
->rx_frame_errors
= dev
->stats
.rx_frame_errors
;
734 #ifdef CONFIG_NET_POLL_CONTROLLER
735 static void virtnet_netpoll(struct net_device
*dev
)
737 struct virtnet_info
*vi
= netdev_priv(dev
);
739 napi_schedule(&vi
->napi
);
743 static int virtnet_open(struct net_device
*dev
)
745 struct virtnet_info
*vi
= netdev_priv(dev
);
747 /* Make sure we have some buffers: if oom use wq. */
748 if (!try_fill_recv(vi
, GFP_KERNEL
))
749 queue_delayed_work(system_nrt_wq
, &vi
->refill
, 0);
751 virtnet_napi_enable(vi
);
756 * Send command via the control virtqueue and check status. Commands
757 * supported by the hypervisor, as indicated by feature bits, should
758 * never fail unless improperly formated.
760 static bool virtnet_send_command(struct virtnet_info
*vi
, u8
class, u8 cmd
,
761 struct scatterlist
*data
, int out
, int in
)
763 struct scatterlist
*s
, sg
[VIRTNET_SEND_COMMAND_SG_MAX
+ 2];
764 struct virtio_net_ctrl_hdr ctrl
;
765 virtio_net_ctrl_ack status
= ~0;
769 /* Caller should know better */
770 BUG_ON(!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
) ||
771 (out
+ in
> VIRTNET_SEND_COMMAND_SG_MAX
));
773 out
++; /* Add header */
774 in
++; /* Add return status */
779 sg_init_table(sg
, out
+ in
);
781 sg_set_buf(&sg
[0], &ctrl
, sizeof(ctrl
));
782 for_each_sg(data
, s
, out
+ in
- 2, i
)
783 sg_set_buf(&sg
[i
+ 1], sg_virt(s
), s
->length
);
784 sg_set_buf(&sg
[out
+ in
- 1], &status
, sizeof(status
));
786 BUG_ON(virtqueue_add_buf(vi
->cvq
, sg
, out
, in
, vi
, GFP_ATOMIC
) < 0);
788 virtqueue_kick(vi
->cvq
);
791 * Spin for a response, the kick causes an ioport write, trapping
792 * into the hypervisor, so the request should be handled immediately.
794 while (!virtqueue_get_buf(vi
->cvq
, &tmp
))
797 return status
== VIRTIO_NET_OK
;
800 static void virtnet_ack_link_announce(struct virtnet_info
*vi
)
803 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_ANNOUNCE
,
804 VIRTIO_NET_CTRL_ANNOUNCE_ACK
, NULL
,
806 dev_warn(&vi
->dev
->dev
, "Failed to ack link announce.\n");
810 static int virtnet_close(struct net_device
*dev
)
812 struct virtnet_info
*vi
= netdev_priv(dev
);
814 /* Make sure refill_work doesn't re-enable napi! */
815 cancel_delayed_work_sync(&vi
->refill
);
816 napi_disable(&vi
->napi
);
821 static void virtnet_set_rx_mode(struct net_device
*dev
)
823 struct virtnet_info
*vi
= netdev_priv(dev
);
824 struct scatterlist sg
[2];
825 u8 promisc
, allmulti
;
826 struct virtio_net_ctrl_mac
*mac_data
;
827 struct netdev_hw_addr
*ha
;
833 /* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */
834 if (!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_RX
))
837 promisc
= ((dev
->flags
& IFF_PROMISC
) != 0);
838 allmulti
= ((dev
->flags
& IFF_ALLMULTI
) != 0);
840 sg_init_one(sg
, &promisc
, sizeof(promisc
));
842 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_RX
,
843 VIRTIO_NET_CTRL_RX_PROMISC
,
845 dev_warn(&dev
->dev
, "Failed to %sable promisc mode.\n",
846 promisc
? "en" : "dis");
848 sg_init_one(sg
, &allmulti
, sizeof(allmulti
));
850 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_RX
,
851 VIRTIO_NET_CTRL_RX_ALLMULTI
,
853 dev_warn(&dev
->dev
, "Failed to %sable allmulti mode.\n",
854 allmulti
? "en" : "dis");
856 uc_count
= netdev_uc_count(dev
);
857 mc_count
= netdev_mc_count(dev
);
858 /* MAC filter - use one buffer for both lists */
859 buf
= kzalloc(((uc_count
+ mc_count
) * ETH_ALEN
) +
860 (2 * sizeof(mac_data
->entries
)), GFP_ATOMIC
);
863 dev_warn(&dev
->dev
, "No memory for MAC address buffer\n");
867 sg_init_table(sg
, 2);
869 /* Store the unicast list and count in the front of the buffer */
870 mac_data
->entries
= uc_count
;
872 netdev_for_each_uc_addr(ha
, dev
)
873 memcpy(&mac_data
->macs
[i
++][0], ha
->addr
, ETH_ALEN
);
875 sg_set_buf(&sg
[0], mac_data
,
876 sizeof(mac_data
->entries
) + (uc_count
* ETH_ALEN
));
878 /* multicast list and count fill the end */
879 mac_data
= (void *)&mac_data
->macs
[uc_count
][0];
881 mac_data
->entries
= mc_count
;
883 netdev_for_each_mc_addr(ha
, dev
)
884 memcpy(&mac_data
->macs
[i
++][0], ha
->addr
, ETH_ALEN
);
886 sg_set_buf(&sg
[1], mac_data
,
887 sizeof(mac_data
->entries
) + (mc_count
* ETH_ALEN
));
889 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MAC
,
890 VIRTIO_NET_CTRL_MAC_TABLE_SET
,
892 dev_warn(&dev
->dev
, "Failed to set MAC fitler table.\n");
897 static int virtnet_vlan_rx_add_vid(struct net_device
*dev
, u16 vid
)
899 struct virtnet_info
*vi
= netdev_priv(dev
);
900 struct scatterlist sg
;
902 sg_init_one(&sg
, &vid
, sizeof(vid
));
904 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_VLAN
,
905 VIRTIO_NET_CTRL_VLAN_ADD
, &sg
, 1, 0))
906 dev_warn(&dev
->dev
, "Failed to add VLAN ID %d.\n", vid
);
910 static int virtnet_vlan_rx_kill_vid(struct net_device
*dev
, u16 vid
)
912 struct virtnet_info
*vi
= netdev_priv(dev
);
913 struct scatterlist sg
;
915 sg_init_one(&sg
, &vid
, sizeof(vid
));
917 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_VLAN
,
918 VIRTIO_NET_CTRL_VLAN_DEL
, &sg
, 1, 0))
919 dev_warn(&dev
->dev
, "Failed to kill VLAN ID %d.\n", vid
);
923 static void virtnet_get_ringparam(struct net_device
*dev
,
924 struct ethtool_ringparam
*ring
)
926 struct virtnet_info
*vi
= netdev_priv(dev
);
928 ring
->rx_max_pending
= virtqueue_get_vring_size(vi
->rvq
);
929 ring
->tx_max_pending
= virtqueue_get_vring_size(vi
->svq
);
930 ring
->rx_pending
= ring
->rx_max_pending
;
931 ring
->tx_pending
= ring
->tx_max_pending
;
936 static void virtnet_get_drvinfo(struct net_device
*dev
,
937 struct ethtool_drvinfo
*info
)
939 struct virtnet_info
*vi
= netdev_priv(dev
);
940 struct virtio_device
*vdev
= vi
->vdev
;
942 strlcpy(info
->driver
, KBUILD_MODNAME
, sizeof(info
->driver
));
943 strlcpy(info
->version
, VIRTNET_DRIVER_VERSION
, sizeof(info
->version
));
944 strlcpy(info
->bus_info
, virtio_bus_name(vdev
), sizeof(info
->bus_info
));
948 static const struct ethtool_ops virtnet_ethtool_ops
= {
949 .get_drvinfo
= virtnet_get_drvinfo
,
950 .get_link
= ethtool_op_get_link
,
951 .get_ringparam
= virtnet_get_ringparam
,
955 #define MAX_MTU 65535
957 static int virtnet_change_mtu(struct net_device
*dev
, int new_mtu
)
959 if (new_mtu
< MIN_MTU
|| new_mtu
> MAX_MTU
)
965 static const struct net_device_ops virtnet_netdev
= {
966 .ndo_open
= virtnet_open
,
967 .ndo_stop
= virtnet_close
,
968 .ndo_start_xmit
= start_xmit
,
969 .ndo_validate_addr
= eth_validate_addr
,
970 .ndo_set_mac_address
= virtnet_set_mac_address
,
971 .ndo_set_rx_mode
= virtnet_set_rx_mode
,
972 .ndo_change_mtu
= virtnet_change_mtu
,
973 .ndo_get_stats64
= virtnet_stats
,
974 .ndo_vlan_rx_add_vid
= virtnet_vlan_rx_add_vid
,
975 .ndo_vlan_rx_kill_vid
= virtnet_vlan_rx_kill_vid
,
976 #ifdef CONFIG_NET_POLL_CONTROLLER
977 .ndo_poll_controller
= virtnet_netpoll
,
981 static void virtnet_config_changed_work(struct work_struct
*work
)
983 struct virtnet_info
*vi
=
984 container_of(work
, struct virtnet_info
, config_work
);
987 mutex_lock(&vi
->config_lock
);
988 if (!vi
->config_enable
)
991 if (virtio_config_val(vi
->vdev
, VIRTIO_NET_F_STATUS
,
992 offsetof(struct virtio_net_config
, status
),
996 if (v
& VIRTIO_NET_S_ANNOUNCE
) {
997 netif_notify_peers(vi
->dev
);
998 virtnet_ack_link_announce(vi
);
1001 /* Ignore unknown (future) status bits */
1002 v
&= VIRTIO_NET_S_LINK_UP
;
1004 if (vi
->status
== v
)
1009 if (vi
->status
& VIRTIO_NET_S_LINK_UP
) {
1010 netif_carrier_on(vi
->dev
);
1011 netif_wake_queue(vi
->dev
);
1013 netif_carrier_off(vi
->dev
);
1014 netif_stop_queue(vi
->dev
);
1017 mutex_unlock(&vi
->config_lock
);
1020 static void virtnet_config_changed(struct virtio_device
*vdev
)
1022 struct virtnet_info
*vi
= vdev
->priv
;
1024 queue_work(system_nrt_wq
, &vi
->config_work
);
1027 static int init_vqs(struct virtnet_info
*vi
)
1029 struct virtqueue
*vqs
[3];
1030 vq_callback_t
*callbacks
[] = { skb_recv_done
, skb_xmit_done
, NULL
};
1031 const char *names
[] = { "input", "output", "control" };
1034 /* We expect two virtqueues, receive then send,
1035 * and optionally control. */
1036 nvqs
= virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
) ? 3 : 2;
1038 err
= vi
->vdev
->config
->find_vqs(vi
->vdev
, nvqs
, vqs
, callbacks
, names
);
1045 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
)) {
1048 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VLAN
))
1049 vi
->dev
->features
|= NETIF_F_HW_VLAN_FILTER
;
1054 static int virtnet_probe(struct virtio_device
*vdev
)
1057 struct net_device
*dev
;
1058 struct virtnet_info
*vi
;
1060 /* Allocate ourselves a network device with room for our info */
1061 dev
= alloc_etherdev(sizeof(struct virtnet_info
));
1065 /* Set up network device as normal. */
1066 dev
->priv_flags
|= IFF_UNICAST_FLT
;
1067 dev
->netdev_ops
= &virtnet_netdev
;
1068 dev
->features
= NETIF_F_HIGHDMA
;
1070 SET_ETHTOOL_OPS(dev
, &virtnet_ethtool_ops
);
1071 SET_NETDEV_DEV(dev
, &vdev
->dev
);
1073 /* Do we support "hardware" checksums? */
1074 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CSUM
)) {
1075 /* This opens up the world of extra features. */
1076 dev
->hw_features
|= NETIF_F_HW_CSUM
|NETIF_F_SG
|NETIF_F_FRAGLIST
;
1078 dev
->features
|= NETIF_F_HW_CSUM
|NETIF_F_SG
|NETIF_F_FRAGLIST
;
1080 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GSO
)) {
1081 dev
->hw_features
|= NETIF_F_TSO
| NETIF_F_UFO
1082 | NETIF_F_TSO_ECN
| NETIF_F_TSO6
;
1084 /* Individual feature bits: what can host handle? */
1085 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_TSO4
))
1086 dev
->hw_features
|= NETIF_F_TSO
;
1087 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_TSO6
))
1088 dev
->hw_features
|= NETIF_F_TSO6
;
1089 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_ECN
))
1090 dev
->hw_features
|= NETIF_F_TSO_ECN
;
1091 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_UFO
))
1092 dev
->hw_features
|= NETIF_F_UFO
;
1095 dev
->features
|= dev
->hw_features
& (NETIF_F_ALL_TSO
|NETIF_F_UFO
);
1096 /* (!csum && gso) case will be fixed by register_netdev() */
1099 /* Configuration may specify what MAC to use. Otherwise random. */
1100 if (virtio_config_val_len(vdev
, VIRTIO_NET_F_MAC
,
1101 offsetof(struct virtio_net_config
, mac
),
1102 dev
->dev_addr
, dev
->addr_len
) < 0)
1103 eth_hw_addr_random(dev
);
1105 /* Set up our device-specific information */
1106 vi
= netdev_priv(dev
);
1107 netif_napi_add(dev
, &vi
->napi
, virtnet_poll
, napi_weight
);
1112 vi
->stats
= alloc_percpu(struct virtnet_stats
);
1114 if (vi
->stats
== NULL
)
1117 INIT_DELAYED_WORK(&vi
->refill
, refill_work
);
1118 mutex_init(&vi
->config_lock
);
1119 vi
->config_enable
= true;
1120 INIT_WORK(&vi
->config_work
, virtnet_config_changed_work
);
1121 sg_init_table(vi
->rx_sg
, ARRAY_SIZE(vi
->rx_sg
));
1122 sg_init_table(vi
->tx_sg
, ARRAY_SIZE(vi
->tx_sg
));
1124 /* If we can receive ANY GSO packets, we must allocate large ones. */
1125 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_TSO4
) ||
1126 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_TSO6
) ||
1127 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_ECN
))
1128 vi
->big_packets
= true;
1130 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MRG_RXBUF
))
1131 vi
->mergeable_rx_bufs
= true;
1137 err
= register_netdev(dev
);
1139 pr_debug("virtio_net: registering device failed\n");
1143 /* Last of all, set up some receive buffers. */
1144 try_fill_recv(vi
, GFP_KERNEL
);
1146 /* If we didn't even get one input buffer, we're useless. */
1152 /* Assume link up if device can't report link status,
1153 otherwise get link status from config. */
1154 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_STATUS
)) {
1155 netif_carrier_off(dev
);
1156 queue_work(system_nrt_wq
, &vi
->config_work
);
1158 vi
->status
= VIRTIO_NET_S_LINK_UP
;
1159 netif_carrier_on(dev
);
1162 pr_debug("virtnet: registered device %s\n", dev
->name
);
1166 unregister_netdev(dev
);
1168 vdev
->config
->del_vqs(vdev
);
1170 free_percpu(vi
->stats
);
1176 static void free_unused_bufs(struct virtnet_info
*vi
)
1180 buf
= virtqueue_detach_unused_buf(vi
->svq
);
1186 buf
= virtqueue_detach_unused_buf(vi
->rvq
);
1189 if (vi
->mergeable_rx_bufs
|| vi
->big_packets
)
1190 give_pages(vi
, buf
);
1195 BUG_ON(vi
->num
!= 0);
1198 static void remove_vq_common(struct virtnet_info
*vi
)
1200 vi
->vdev
->config
->reset(vi
->vdev
);
1202 /* Free unused buffers in both send and recv, if any. */
1203 free_unused_bufs(vi
);
1205 vi
->vdev
->config
->del_vqs(vi
->vdev
);
1208 __free_pages(get_a_page(vi
, GFP_KERNEL
), 0);
1211 static void __devexit
virtnet_remove(struct virtio_device
*vdev
)
1213 struct virtnet_info
*vi
= vdev
->priv
;
1215 /* Prevent config work handler from accessing the device. */
1216 mutex_lock(&vi
->config_lock
);
1217 vi
->config_enable
= false;
1218 mutex_unlock(&vi
->config_lock
);
1220 unregister_netdev(vi
->dev
);
1222 remove_vq_common(vi
);
1224 flush_work(&vi
->config_work
);
1226 free_percpu(vi
->stats
);
1227 free_netdev(vi
->dev
);
1231 static int virtnet_freeze(struct virtio_device
*vdev
)
1233 struct virtnet_info
*vi
= vdev
->priv
;
1235 /* Prevent config work handler from accessing the device */
1236 mutex_lock(&vi
->config_lock
);
1237 vi
->config_enable
= false;
1238 mutex_unlock(&vi
->config_lock
);
1240 netif_device_detach(vi
->dev
);
1241 cancel_delayed_work_sync(&vi
->refill
);
1243 if (netif_running(vi
->dev
))
1244 napi_disable(&vi
->napi
);
1246 remove_vq_common(vi
);
1248 flush_work(&vi
->config_work
);
1253 static int virtnet_restore(struct virtio_device
*vdev
)
1255 struct virtnet_info
*vi
= vdev
->priv
;
1262 if (netif_running(vi
->dev
))
1263 virtnet_napi_enable(vi
);
1265 netif_device_attach(vi
->dev
);
1267 if (!try_fill_recv(vi
, GFP_KERNEL
))
1268 queue_delayed_work(system_nrt_wq
, &vi
->refill
, 0);
1270 mutex_lock(&vi
->config_lock
);
1271 vi
->config_enable
= true;
1272 mutex_unlock(&vi
->config_lock
);
1278 static struct virtio_device_id id_table
[] = {
1279 { VIRTIO_ID_NET
, VIRTIO_DEV_ANY_ID
},
1283 static unsigned int features
[] = {
1284 VIRTIO_NET_F_CSUM
, VIRTIO_NET_F_GUEST_CSUM
,
1285 VIRTIO_NET_F_GSO
, VIRTIO_NET_F_MAC
,
1286 VIRTIO_NET_F_HOST_TSO4
, VIRTIO_NET_F_HOST_UFO
, VIRTIO_NET_F_HOST_TSO6
,
1287 VIRTIO_NET_F_HOST_ECN
, VIRTIO_NET_F_GUEST_TSO4
, VIRTIO_NET_F_GUEST_TSO6
,
1288 VIRTIO_NET_F_GUEST_ECN
, VIRTIO_NET_F_GUEST_UFO
,
1289 VIRTIO_NET_F_MRG_RXBUF
, VIRTIO_NET_F_STATUS
, VIRTIO_NET_F_CTRL_VQ
,
1290 VIRTIO_NET_F_CTRL_RX
, VIRTIO_NET_F_CTRL_VLAN
,
1291 VIRTIO_NET_F_GUEST_ANNOUNCE
,
1294 static struct virtio_driver virtio_net_driver
= {
1295 .feature_table
= features
,
1296 .feature_table_size
= ARRAY_SIZE(features
),
1297 .driver
.name
= KBUILD_MODNAME
,
1298 .driver
.owner
= THIS_MODULE
,
1299 .id_table
= id_table
,
1300 .probe
= virtnet_probe
,
1301 .remove
= __devexit_p(virtnet_remove
),
1302 .config_changed
= virtnet_config_changed
,
1304 .freeze
= virtnet_freeze
,
1305 .restore
= virtnet_restore
,
1309 static int __init
init(void)
1311 return register_virtio_driver(&virtio_net_driver
);
1314 static void __exit
fini(void)
1316 unregister_virtio_driver(&virtio_net_driver
);
1321 MODULE_DEVICE_TABLE(virtio
, id_table
);
1322 MODULE_DESCRIPTION("Virtio network driver");
1323 MODULE_LICENSE("GPL");