blob: a6fcf15adc4ff3d36d928f44cc0ac1f49d249820 [file] [log] [blame]
Rusty Russell48925e32009-09-24 09:59:20 -06001/* A network driver using virtio.
Rusty Russell296f96f2007-10-22 11:03:37 +10002 *
3 * Copyright 2007 Rusty Russell <rusty@rustcorp.com.au> IBM Corporation
4 *
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.
9 *
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.
14 *
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
18 */
19//#define DEBUG
20#include <linux/netdevice.h>
21#include <linux/etherdevice.h>
Herbert Xua9ea3fc2008-04-18 11:21:42 +080022#include <linux/ethtool.h>
Rusty Russell296f96f2007-10-22 11:03:37 +100023#include <linux/module.h>
24#include <linux/virtio.h>
25#include <linux/virtio_net.h>
26#include <linux/scatterlist.h>
Alex Williamsone918085a2009-01-25 18:06:26 -080027#include <linux/if_vlan.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090028#include <linux/slab.h>
Rusty Russell296f96f2007-10-22 11:03:37 +100029
Dor Laor6c0cd7c2007-12-16 15:19:43 +020030static int napi_weight = 128;
31module_param(napi_weight, int, 0444);
32
Rusty Russelleb939922011-12-19 14:08:01 +000033static bool csum = true, gso = true;
Rusty Russell34a48572008-02-04 23:50:02 -050034module_param(csum, bool, 0444);
35module_param(gso, bool, 0444);
36
Rusty Russell296f96f2007-10-22 11:03:37 +100037/* FIXME: MTU in config. */
Alex Williamsone918085a2009-01-25 18:06:26 -080038#define MAX_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -080039#define GOOD_COPY_LEN 128
Rusty Russell296f96f2007-10-22 11:03:37 +100040
Alex Williamsonf565a7c2009-02-04 09:02:45 +000041#define VIRTNET_SEND_COMMAND_SG_MAX 2
Rick Jones66846042011-11-14 14:17:08 +000042#define VIRTNET_DRIVER_VERSION "1.0.0"
Alex Williamson2a41f712009-02-04 09:02:34 +000043
stephen hemminger3fa2a1d2011-06-15 06:36:29 +000044struct virtnet_stats {
Eric Dumazet83a27052012-06-05 22:35:24 +000045 struct u64_stats_sync tx_syncp;
46 struct u64_stats_sync rx_syncp;
stephen hemminger3fa2a1d2011-06-15 06:36:29 +000047 u64 tx_bytes;
48 u64 tx_packets;
49
50 u64 rx_bytes;
51 u64 rx_packets;
52};
53
Jason Wange9d74172012-12-07 07:04:55 +000054/* Internal representation of a send virtqueue */
55struct send_queue {
56 /* Virtqueue associated with this send _queue */
57 struct virtqueue *vq;
58
59 /* TX: fragments + linear part + virtio header */
60 struct scatterlist sg[MAX_SKB_FRAGS + 2];
Jason Wang986a4f42012-12-07 07:04:56 +000061
62 /* Name of the send queue: output.$index */
63 char name[40];
Jason Wange9d74172012-12-07 07:04:55 +000064};
65
66/* Internal representation of a receive virtqueue */
67struct receive_queue {
68 /* Virtqueue associated with this receive_queue */
69 struct virtqueue *vq;
70
Rusty Russell296f96f2007-10-22 11:03:37 +100071 struct napi_struct napi;
72
73 /* Number of input buffers, and max we've ever had. */
74 unsigned int num, max;
75
Jason Wange9d74172012-12-07 07:04:55 +000076 /* Chain pages by the private ptr. */
77 struct page *pages;
78
79 /* RX: fragments + linear part + virtio header */
80 struct scatterlist sg[MAX_SKB_FRAGS + 2];
Jason Wang986a4f42012-12-07 07:04:56 +000081
82 /* Name of this receive queue: input.$index */
83 char name[40];
Jason Wange9d74172012-12-07 07:04:55 +000084};
85
86struct virtnet_info {
87 struct virtio_device *vdev;
88 struct virtqueue *cvq;
89 struct net_device *dev;
Jason Wang986a4f42012-12-07 07:04:56 +000090 struct send_queue *sq;
91 struct receive_queue *rq;
Jason Wange9d74172012-12-07 07:04:55 +000092 unsigned int status;
93
Jason Wang986a4f42012-12-07 07:04:56 +000094 /* Max # of queue pairs supported by the device */
95 u16 max_queue_pairs;
96
97 /* # of queue pairs currently used by the driver */
98 u16 curr_queue_pairs;
99
Herbert Xu97402b92008-04-18 11:24:27 +0800100 /* I like... big packets and I cannot lie! */
101 bool big_packets;
102
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800103 /* Host will merge rx buffers for big packets (shake it! shake it!) */
104 bool mergeable_rx_bufs;
105
Jason Wang986a4f42012-12-07 07:04:56 +0000106 /* Has control virtqueue */
107 bool has_cvq;
108
Jason Wang586d17c2012-04-11 20:43:52 +0000109 /* enable config space updates */
110 bool config_enable;
111
stephen hemminger3fa2a1d2011-06-15 06:36:29 +0000112 /* Active statistics */
113 struct virtnet_stats __percpu *stats;
114
Rusty Russell3161e452009-08-26 12:22:32 -0700115 /* Work struct for refilling if we run low on memory. */
116 struct delayed_work refill;
117
Jason Wang586d17c2012-04-11 20:43:52 +0000118 /* Work struct for config space updates */
119 struct work_struct config_work;
120
121 /* Lock for config space updates */
122 struct mutex config_lock;
Jason Wang986a4f42012-12-07 07:04:56 +0000123
124 /* Does the affinity hint is set for virtqueues? */
125 bool affinity_hint_set;
Rusty Russell296f96f2007-10-22 11:03:37 +1000126};
127
Rusty Russellb3f24692009-09-24 09:59:19 -0600128struct skb_vnet_hdr {
129 union {
130 struct virtio_net_hdr hdr;
131 struct virtio_net_hdr_mrg_rxbuf mhdr;
132 };
133};
134
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000135struct padded_vnet_hdr {
136 struct virtio_net_hdr hdr;
137 /*
138 * virtio_net_hdr should be in a separated sg buffer because of a
139 * QEMU bug, and data sg buffer shares same page with this header sg.
140 * This padding makes next sg 16 byte aligned after virtio_net_hdr.
141 */
142 char padding[6];
143};
144
Jason Wang986a4f42012-12-07 07:04:56 +0000145/* Converting between virtqueue no. and kernel tx/rx queue no.
146 * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq
147 */
148static int vq2txq(struct virtqueue *vq)
149{
150 return (virtqueue_get_queue_index(vq) - 1) / 2;
151}
152
153static int txq2vq(int txq)
154{
155 return txq * 2 + 1;
156}
157
158static int vq2rxq(struct virtqueue *vq)
159{
160 return virtqueue_get_queue_index(vq) / 2;
161}
162
163static int rxq2vq(int rxq)
164{
165 return rxq * 2;
166}
167
Rusty Russellb3f24692009-09-24 09:59:19 -0600168static inline struct skb_vnet_hdr *skb_vnet_hdr(struct sk_buff *skb)
Rusty Russell296f96f2007-10-22 11:03:37 +1000169{
Rusty Russellb3f24692009-09-24 09:59:19 -0600170 return (struct skb_vnet_hdr *)skb->cb;
Rusty Russell296f96f2007-10-22 11:03:37 +1000171}
172
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000173/*
174 * private is used to chain pages for big packets, put the whole
175 * most recent used list in the beginning for reuse
176 */
Jason Wange9d74172012-12-07 07:04:55 +0000177static void give_pages(struct receive_queue *rq, struct page *page)
Rusty Russellfb6813f2008-07-25 12:06:01 -0500178{
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000179 struct page *end;
180
Jason Wange9d74172012-12-07 07:04:55 +0000181 /* Find end of list, sew whole thing into vi->rq.pages. */
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000182 for (end = page; end->private; end = (struct page *)end->private);
Jason Wange9d74172012-12-07 07:04:55 +0000183 end->private = (unsigned long)rq->pages;
184 rq->pages = page;
Rusty Russellfb6813f2008-07-25 12:06:01 -0500185}
186
Jason Wange9d74172012-12-07 07:04:55 +0000187static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask)
Rusty Russellfb6813f2008-07-25 12:06:01 -0500188{
Jason Wange9d74172012-12-07 07:04:55 +0000189 struct page *p = rq->pages;
Rusty Russellfb6813f2008-07-25 12:06:01 -0500190
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000191 if (p) {
Jason Wange9d74172012-12-07 07:04:55 +0000192 rq->pages = (struct page *)p->private;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000193 /* clear private here, it is used to chain pages */
194 p->private = 0;
195 } else
Rusty Russellfb6813f2008-07-25 12:06:01 -0500196 p = alloc_page(gfp_mask);
197 return p;
198}
199
Jason Wange9d74172012-12-07 07:04:55 +0000200static void skb_xmit_done(struct virtqueue *vq)
Rusty Russell296f96f2007-10-22 11:03:37 +1000201{
Jason Wange9d74172012-12-07 07:04:55 +0000202 struct virtnet_info *vi = vq->vdev->priv;
Rusty Russell296f96f2007-10-22 11:03:37 +1000203
Rusty Russell2cb9c6b2008-02-04 23:50:07 -0500204 /* Suppress further interrupts. */
Jason Wange9d74172012-12-07 07:04:55 +0000205 virtqueue_disable_cb(vq);
Rusty Russell11a3a152008-05-26 17:48:13 +1000206
Rusty Russell363f1512008-06-08 20:51:55 +1000207 /* We were probably waiting for more output buffers. */
Jason Wang986a4f42012-12-07 07:04:56 +0000208 netif_wake_subqueue(vi->dev, vq2txq(vq));
Rusty Russell296f96f2007-10-22 11:03:37 +1000209}
210
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000211static void set_skb_frag(struct sk_buff *skb, struct page *page,
212 unsigned int offset, unsigned int *len)
213{
Krishna Kumar8a59a7b2011-10-19 22:17:27 +0000214 int size = min((unsigned)PAGE_SIZE - offset, *len);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000215 int i = skb_shinfo(skb)->nr_frags;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000216
Krishna Kumar8a59a7b2011-10-19 22:17:27 +0000217 __skb_fill_page_desc(skb, i, page, offset, size);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000218
Krishna Kumar8a59a7b2011-10-19 22:17:27 +0000219 skb->data_len += size;
220 skb->len += size;
Eric Dumazet4b727362011-10-19 23:14:46 +0000221 skb->truesize += PAGE_SIZE;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000222 skb_shinfo(skb)->nr_frags++;
Krishna Kumar8a59a7b2011-10-19 22:17:27 +0000223 *len -= size;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000224}
225
Mike Waychison34646452012-01-04 12:52:32 +0000226/* Called from bottom half context */
Jason Wange9d74172012-12-07 07:04:55 +0000227static struct sk_buff *page_to_skb(struct receive_queue *rq,
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000228 struct page *page, unsigned int len)
229{
Jason Wange9d74172012-12-07 07:04:55 +0000230 struct virtnet_info *vi = rq->vq->vdev->priv;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000231 struct sk_buff *skb;
232 struct skb_vnet_hdr *hdr;
233 unsigned int copy, hdr_len, offset;
234 char *p;
235
236 p = page_address(page);
237
238 /* copy small packet so we can reuse these pages for small data */
239 skb = netdev_alloc_skb_ip_align(vi->dev, GOOD_COPY_LEN);
240 if (unlikely(!skb))
241 return NULL;
242
243 hdr = skb_vnet_hdr(skb);
244
245 if (vi->mergeable_rx_bufs) {
246 hdr_len = sizeof hdr->mhdr;
247 offset = hdr_len;
248 } else {
249 hdr_len = sizeof hdr->hdr;
250 offset = sizeof(struct padded_vnet_hdr);
251 }
252
253 memcpy(hdr, p, hdr_len);
254
255 len -= hdr_len;
256 p += offset;
257
258 copy = len;
259 if (copy > skb_tailroom(skb))
260 copy = skb_tailroom(skb);
261 memcpy(skb_put(skb, copy), p, copy);
262
263 len -= copy;
264 offset += copy;
265
Sasha Levine878d782011-09-28 04:40:54 +0000266 /*
267 * Verify that we can indeed put this data into a skb.
268 * This is here to handle cases when the device erroneously
269 * tries to receive more than is possible. This is usually
270 * the case of a broken device.
271 */
272 if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) {
Amerigo Wangbe443892012-11-08 17:47:28 +0000273 net_dbg_ratelimited("%s: too much data\n", skb->dev->name);
Sasha Levine878d782011-09-28 04:40:54 +0000274 dev_kfree_skb(skb);
275 return NULL;
276 }
277
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000278 while (len) {
279 set_skb_frag(skb, page, offset, &len);
280 page = (struct page *)page->private;
281 offset = 0;
282 }
283
284 if (page)
Jason Wange9d74172012-12-07 07:04:55 +0000285 give_pages(rq, page);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000286
287 return skb;
288}
289
Jason Wange9d74172012-12-07 07:04:55 +0000290static int receive_mergeable(struct receive_queue *rq, struct sk_buff *skb)
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000291{
292 struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
293 struct page *page;
294 int num_buf, i, len;
295
296 num_buf = hdr->mhdr.num_buffers;
297 while (--num_buf) {
298 i = skb_shinfo(skb)->nr_frags;
299 if (i >= MAX_SKB_FRAGS) {
300 pr_debug("%s: packet too long\n", skb->dev->name);
301 skb->dev->stats.rx_length_errors++;
302 return -EINVAL;
303 }
Jason Wange9d74172012-12-07 07:04:55 +0000304 page = virtqueue_get_buf(rq->vq, &len);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000305 if (!page) {
306 pr_debug("%s: rx error: %d buffers missing\n",
307 skb->dev->name, hdr->mhdr.num_buffers);
308 skb->dev->stats.rx_length_errors++;
309 return -EINVAL;
310 }
stephen hemminger3fa2a1d2011-06-15 06:36:29 +0000311
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000312 if (len > PAGE_SIZE)
313 len = PAGE_SIZE;
314
315 set_skb_frag(skb, page, 0, &len);
316
Jason Wange9d74172012-12-07 07:04:55 +0000317 --rq->num;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000318 }
319 return 0;
320}
321
Jason Wange9d74172012-12-07 07:04:55 +0000322static void receive_buf(struct receive_queue *rq, void *buf, unsigned int len)
Rusty Russell296f96f2007-10-22 11:03:37 +1000323{
Jason Wange9d74172012-12-07 07:04:55 +0000324 struct virtnet_info *vi = rq->vq->vdev->priv;
325 struct net_device *dev = vi->dev;
Eric Dumazet58472a72012-02-13 06:53:41 +0000326 struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000327 struct sk_buff *skb;
328 struct page *page;
329 struct skb_vnet_hdr *hdr;
Rusty Russell296f96f2007-10-22 11:03:37 +1000330
331 if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
332 pr_debug("%s: short packet %i\n", dev->name, len);
333 dev->stats.rx_length_errors++;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000334 if (vi->mergeable_rx_bufs || vi->big_packets)
Jason Wange9d74172012-12-07 07:04:55 +0000335 give_pages(rq, buf);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000336 else
337 dev_kfree_skb(buf);
338 return;
Rusty Russell296f96f2007-10-22 11:03:37 +1000339 }
Rusty Russell296f96f2007-10-22 11:03:37 +1000340
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000341 if (!vi->mergeable_rx_bufs && !vi->big_packets) {
342 skb = buf;
343 len -= sizeof(struct virtio_net_hdr);
344 skb_trim(skb, len);
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800345 } else {
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000346 page = buf;
Jason Wange9d74172012-12-07 07:04:55 +0000347 skb = page_to_skb(rq, page, len);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000348 if (unlikely(!skb)) {
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800349 dev->stats.rx_dropped++;
Jason Wange9d74172012-12-07 07:04:55 +0000350 give_pages(rq, page);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000351 return;
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800352 }
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000353 if (vi->mergeable_rx_bufs)
Jason Wange9d74172012-12-07 07:04:55 +0000354 if (receive_mergeable(rq, skb)) {
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000355 dev_kfree_skb(skb);
356 return;
357 }
Herbert Xu97402b92008-04-18 11:24:27 +0800358 }
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800359
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000360 hdr = skb_vnet_hdr(skb);
stephen hemminger3fa2a1d2011-06-15 06:36:29 +0000361
Eric Dumazet83a27052012-06-05 22:35:24 +0000362 u64_stats_update_begin(&stats->rx_syncp);
stephen hemminger3fa2a1d2011-06-15 06:36:29 +0000363 stats->rx_bytes += skb->len;
364 stats->rx_packets++;
Eric Dumazet83a27052012-06-05 22:35:24 +0000365 u64_stats_update_end(&stats->rx_syncp);
Rusty Russell296f96f2007-10-22 11:03:37 +1000366
Rusty Russellb3f24692009-09-24 09:59:19 -0600367 if (hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
Rusty Russell296f96f2007-10-22 11:03:37 +1000368 pr_debug("Needs csum!\n");
Rusty Russellb3f24692009-09-24 09:59:19 -0600369 if (!skb_partial_csum_set(skb,
370 hdr->hdr.csum_start,
371 hdr->hdr.csum_offset))
Rusty Russell296f96f2007-10-22 11:03:37 +1000372 goto frame_err;
Jason Wang10a8d942011-06-10 00:56:17 +0000373 } else if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) {
374 skb->ip_summed = CHECKSUM_UNNECESSARY;
Rusty Russell296f96f2007-10-22 11:03:37 +1000375 }
376
Mark McLoughlin23cde762008-06-08 20:49:00 +1000377 skb->protocol = eth_type_trans(skb, dev);
378 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
379 ntohs(skb->protocol), skb->len, skb->pkt_type);
380
Rusty Russellb3f24692009-09-24 09:59:19 -0600381 if (hdr->hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
Rusty Russell296f96f2007-10-22 11:03:37 +1000382 pr_debug("GSO!\n");
Rusty Russellb3f24692009-09-24 09:59:19 -0600383 switch (hdr->hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
Rusty Russell296f96f2007-10-22 11:03:37 +1000384 case VIRTIO_NET_HDR_GSO_TCPV4:
385 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
386 break;
Rusty Russell296f96f2007-10-22 11:03:37 +1000387 case VIRTIO_NET_HDR_GSO_UDP:
388 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
389 break;
390 case VIRTIO_NET_HDR_GSO_TCPV6:
391 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
392 break;
393 default:
Amerigo Wangbe443892012-11-08 17:47:28 +0000394 net_warn_ratelimited("%s: bad gso type %u.\n",
395 dev->name, hdr->hdr.gso_type);
Rusty Russell296f96f2007-10-22 11:03:37 +1000396 goto frame_err;
397 }
398
Rusty Russellb3f24692009-09-24 09:59:19 -0600399 if (hdr->hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
Rusty Russell34a48572008-02-04 23:50:02 -0500400 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
401
Rusty Russellb3f24692009-09-24 09:59:19 -0600402 skb_shinfo(skb)->gso_size = hdr->hdr.gso_size;
Rusty Russell296f96f2007-10-22 11:03:37 +1000403 if (skb_shinfo(skb)->gso_size == 0) {
Amerigo Wangbe443892012-11-08 17:47:28 +0000404 net_warn_ratelimited("%s: zero gso size.\n", dev->name);
Rusty Russell296f96f2007-10-22 11:03:37 +1000405 goto frame_err;
406 }
407
408 /* Header must be checked, and gso_segs computed. */
409 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
410 skb_shinfo(skb)->gso_segs = 0;
411 }
412
413 netif_receive_skb(skb);
414 return;
415
416frame_err:
417 dev->stats.rx_frame_errors++;
Rusty Russell296f96f2007-10-22 11:03:37 +1000418 dev_kfree_skb(skb);
419}
420
Jason Wange9d74172012-12-07 07:04:55 +0000421static int add_recvbuf_small(struct receive_queue *rq, gfp_t gfp)
Rusty Russell296f96f2007-10-22 11:03:37 +1000422{
Jason Wange9d74172012-12-07 07:04:55 +0000423 struct virtnet_info *vi = rq->vq->vdev->priv;
Rusty Russell296f96f2007-10-22 11:03:37 +1000424 struct sk_buff *skb;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000425 struct skb_vnet_hdr *hdr;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000426 int err;
Rusty Russell296f96f2007-10-22 11:03:37 +1000427
Mike Waychison34646452012-01-04 12:52:32 +0000428 skb = __netdev_alloc_skb_ip_align(vi->dev, MAX_PACKET_LEN, gfp);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000429 if (unlikely(!skb))
430 return -ENOMEM;
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800431
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000432 skb_put(skb, MAX_PACKET_LEN);
433
434 hdr = skb_vnet_hdr(skb);
Jason Wange9d74172012-12-07 07:04:55 +0000435 sg_set_buf(rq->sg, &hdr->hdr, sizeof hdr->hdr);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000436
Jason Wange9d74172012-12-07 07:04:55 +0000437 skb_to_sgvec(skb, rq->sg + 1, 0, skb->len);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000438
Jason Wange9d74172012-12-07 07:04:55 +0000439 err = virtqueue_add_buf(rq->vq, rq->sg, 0, 2, skb, gfp);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000440 if (err < 0)
441 dev_kfree_skb(skb);
442
443 return err;
444}
445
Jason Wange9d74172012-12-07 07:04:55 +0000446static int add_recvbuf_big(struct receive_queue *rq, gfp_t gfp)
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000447{
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000448 struct page *first, *list = NULL;
449 char *p;
450 int i, err, offset;
451
Jason Wange9d74172012-12-07 07:04:55 +0000452 /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000453 for (i = MAX_SKB_FRAGS + 1; i > 1; --i) {
Jason Wange9d74172012-12-07 07:04:55 +0000454 first = get_a_page(rq, gfp);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000455 if (!first) {
456 if (list)
Jason Wange9d74172012-12-07 07:04:55 +0000457 give_pages(rq, list);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000458 return -ENOMEM;
Rusty Russell3161e452009-08-26 12:22:32 -0700459 }
Jason Wange9d74172012-12-07 07:04:55 +0000460 sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE);
Rusty Russell296f96f2007-10-22 11:03:37 +1000461
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000462 /* chain new page in list head to match sg */
463 first->private = (unsigned long)list;
464 list = first;
465 }
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800466
Jason Wange9d74172012-12-07 07:04:55 +0000467 first = get_a_page(rq, gfp);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000468 if (!first) {
Jason Wange9d74172012-12-07 07:04:55 +0000469 give_pages(rq, list);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000470 return -ENOMEM;
471 }
472 p = page_address(first);
Herbert Xu97402b92008-04-18 11:24:27 +0800473
Jason Wange9d74172012-12-07 07:04:55 +0000474 /* rq->sg[0], rq->sg[1] share the same page */
475 /* a separated rq->sg[0] for virtio_net_hdr only due to QEMU bug */
476 sg_set_buf(&rq->sg[0], p, sizeof(struct virtio_net_hdr));
Herbert Xu97402b92008-04-18 11:24:27 +0800477
Jason Wange9d74172012-12-07 07:04:55 +0000478 /* rq->sg[1] for data packet, from offset */
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000479 offset = sizeof(struct padded_vnet_hdr);
Jason Wange9d74172012-12-07 07:04:55 +0000480 sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset);
Herbert Xu97402b92008-04-18 11:24:27 +0800481
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000482 /* chain first in list head */
483 first->private = (unsigned long)list;
Jason Wange9d74172012-12-07 07:04:55 +0000484 err = virtqueue_add_buf(rq->vq, rq->sg, 0, MAX_SKB_FRAGS + 2,
Rusty Russellf96fde42012-01-12 15:44:42 +1030485 first, gfp);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000486 if (err < 0)
Jason Wange9d74172012-12-07 07:04:55 +0000487 give_pages(rq, first);
Herbert Xu97402b92008-04-18 11:24:27 +0800488
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000489 return err;
490}
Herbert Xu97402b92008-04-18 11:24:27 +0800491
Jason Wange9d74172012-12-07 07:04:55 +0000492static int add_recvbuf_mergeable(struct receive_queue *rq, gfp_t gfp)
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000493{
494 struct page *page;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000495 int err;
Rusty Russell296f96f2007-10-22 11:03:37 +1000496
Jason Wange9d74172012-12-07 07:04:55 +0000497 page = get_a_page(rq, gfp);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000498 if (!page)
499 return -ENOMEM;
500
Jason Wange9d74172012-12-07 07:04:55 +0000501 sg_init_one(rq->sg, page_address(page), PAGE_SIZE);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000502
Jason Wange9d74172012-12-07 07:04:55 +0000503 err = virtqueue_add_buf(rq->vq, rq->sg, 0, 1, page, gfp);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000504 if (err < 0)
Jason Wange9d74172012-12-07 07:04:55 +0000505 give_pages(rq, page);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000506
507 return err;
Rusty Russell296f96f2007-10-22 11:03:37 +1000508}
509
Rusty Russellb2baed62011-12-29 00:42:38 +0000510/*
511 * Returns false if we couldn't fill entirely (OOM).
512 *
513 * Normally run in the receive path, but can also be run from ndo_open
514 * before we're receiving packets, or from refill_work which is
515 * careful to disable receiving (using napi_disable).
516 */
Jason Wange9d74172012-12-07 07:04:55 +0000517static bool try_fill_recv(struct receive_queue *rq, gfp_t gfp)
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800518{
Jason Wange9d74172012-12-07 07:04:55 +0000519 struct virtnet_info *vi = rq->vq->vdev->priv;
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800520 int err;
Michael S. Tsirkin1788f4952010-07-02 16:32:55 +0000521 bool oom;
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800522
Amit Shah0aea51c2009-08-26 14:58:28 +0530523 do {
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000524 if (vi->mergeable_rx_bufs)
Jason Wange9d74172012-12-07 07:04:55 +0000525 err = add_recvbuf_mergeable(rq, gfp);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000526 else if (vi->big_packets)
Jason Wange9d74172012-12-07 07:04:55 +0000527 err = add_recvbuf_big(rq, gfp);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000528 else
Jason Wange9d74172012-12-07 07:04:55 +0000529 err = add_recvbuf_small(rq, gfp);
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800530
Michael S. Tsirkin1788f4952010-07-02 16:32:55 +0000531 oom = err == -ENOMEM;
Rusty Russell9ed4cb02012-10-16 23:56:14 +1030532 if (err)
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800533 break;
Jason Wange9d74172012-12-07 07:04:55 +0000534 ++rq->num;
Linus Torvaldsb7dfde92012-12-20 08:37:04 -0800535 } while (rq->vq->num_free);
Jason Wange9d74172012-12-07 07:04:55 +0000536 if (unlikely(rq->num > rq->max))
537 rq->max = rq->num;
538 virtqueue_kick(rq->vq);
Rusty Russell3161e452009-08-26 12:22:32 -0700539 return !oom;
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800540}
541
Rusty Russell18445c42008-02-04 23:49:57 -0500542static void skb_recv_done(struct virtqueue *rvq)
Rusty Russell296f96f2007-10-22 11:03:37 +1000543{
544 struct virtnet_info *vi = rvq->vdev->priv;
Jason Wang986a4f42012-12-07 07:04:56 +0000545 struct receive_queue *rq = &vi->rq[vq2rxq(rvq)];
Jason Wange9d74172012-12-07 07:04:55 +0000546
Rusty Russell18445c42008-02-04 23:49:57 -0500547 /* Schedule NAPI, Suppress further interrupts if successful. */
Jason Wange9d74172012-12-07 07:04:55 +0000548 if (napi_schedule_prep(&rq->napi)) {
Michael S. Tsirkin1915a7122010-04-12 16:19:04 +0300549 virtqueue_disable_cb(rvq);
Jason Wange9d74172012-12-07 07:04:55 +0000550 __napi_schedule(&rq->napi);
Rusty Russell18445c42008-02-04 23:49:57 -0500551 }
Rusty Russell296f96f2007-10-22 11:03:37 +1000552}
553
Jason Wange9d74172012-12-07 07:04:55 +0000554static void virtnet_napi_enable(struct receive_queue *rq)
Bruce Rogers3e9d08e2011-02-10 11:03:31 -0800555{
Jason Wange9d74172012-12-07 07:04:55 +0000556 napi_enable(&rq->napi);
Bruce Rogers3e9d08e2011-02-10 11:03:31 -0800557
558 /* If all buffers were filled by other side before we napi_enabled, we
559 * won't get another interrupt, so process any outstanding packets
560 * now. virtnet_poll wants re-enable the queue, so we disable here.
561 * We synchronize against interrupts via NAPI_STATE_SCHED */
Jason Wange9d74172012-12-07 07:04:55 +0000562 if (napi_schedule_prep(&rq->napi)) {
563 virtqueue_disable_cb(rq->vq);
Michael S. Tsirkinec13ee82012-05-16 10:57:12 +0300564 local_bh_disable();
Jason Wange9d74172012-12-07 07:04:55 +0000565 __napi_schedule(&rq->napi);
Michael S. Tsirkinec13ee82012-05-16 10:57:12 +0300566 local_bh_enable();
Bruce Rogers3e9d08e2011-02-10 11:03:31 -0800567 }
568}
569
Rusty Russell3161e452009-08-26 12:22:32 -0700570static void refill_work(struct work_struct *work)
571{
Jason Wange9d74172012-12-07 07:04:55 +0000572 struct virtnet_info *vi =
573 container_of(work, struct virtnet_info, refill.work);
Rusty Russell3161e452009-08-26 12:22:32 -0700574 bool still_empty;
Jason Wang986a4f42012-12-07 07:04:56 +0000575 int i;
Rusty Russell3161e452009-08-26 12:22:32 -0700576
Jason Wang986a4f42012-12-07 07:04:56 +0000577 for (i = 0; i < vi->max_queue_pairs; i++) {
578 struct receive_queue *rq = &vi->rq[i];
Rusty Russell3161e452009-08-26 12:22:32 -0700579
Jason Wang986a4f42012-12-07 07:04:56 +0000580 napi_disable(&rq->napi);
581 still_empty = !try_fill_recv(rq, GFP_KERNEL);
582 virtnet_napi_enable(rq);
583
584 /* In theory, this can happen: if we don't get any buffers in
585 * we will *never* try to fill again.
586 */
587 if (still_empty)
588 schedule_delayed_work(&vi->refill, HZ/2);
589 }
Rusty Russell3161e452009-08-26 12:22:32 -0700590}
591
Rusty Russell296f96f2007-10-22 11:03:37 +1000592static int virtnet_poll(struct napi_struct *napi, int budget)
593{
Jason Wange9d74172012-12-07 07:04:55 +0000594 struct receive_queue *rq =
595 container_of(napi, struct receive_queue, napi);
596 struct virtnet_info *vi = rq->vq->vdev->priv;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000597 void *buf;
Rusty Russell296f96f2007-10-22 11:03:37 +1000598 unsigned int len, received = 0;
599
600again:
601 while (received < budget &&
Jason Wange9d74172012-12-07 07:04:55 +0000602 (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) {
603 receive_buf(rq, buf, len);
604 --rq->num;
Rusty Russell296f96f2007-10-22 11:03:37 +1000605 received++;
606 }
607
Jason Wange9d74172012-12-07 07:04:55 +0000608 if (rq->num < rq->max / 2) {
609 if (!try_fill_recv(rq, GFP_ATOMIC))
Tejun Heo3b07e9c2012-08-20 14:51:24 -0700610 schedule_delayed_work(&vi->refill, 0);
Rusty Russell3161e452009-08-26 12:22:32 -0700611 }
Rusty Russell296f96f2007-10-22 11:03:37 +1000612
Rusty Russell8329d982007-11-19 11:20:43 -0500613 /* Out of packets? */
614 if (received < budget) {
Ben Hutchings288379f2009-01-19 16:43:59 -0800615 napi_complete(napi);
Jason Wange9d74172012-12-07 07:04:55 +0000616 if (unlikely(!virtqueue_enable_cb(rq->vq)) &&
Joe Perches8e95a202009-12-03 07:58:21 +0000617 napi_schedule_prep(napi)) {
Jason Wange9d74172012-12-07 07:04:55 +0000618 virtqueue_disable_cb(rq->vq);
Ben Hutchings288379f2009-01-19 16:43:59 -0800619 __napi_schedule(napi);
Rusty Russell296f96f2007-10-22 11:03:37 +1000620 goto again;
Christian Borntraeger4265f162008-03-14 14:17:05 +0100621 }
Rusty Russell296f96f2007-10-22 11:03:37 +1000622 }
623
624 return received;
625}
626
Jason Wang986a4f42012-12-07 07:04:56 +0000627static int virtnet_open(struct net_device *dev)
628{
629 struct virtnet_info *vi = netdev_priv(dev);
630 int i;
631
632 for (i = 0; i < vi->max_queue_pairs; i++) {
633 /* Make sure we have some buffers: if oom use wq. */
634 if (!try_fill_recv(&vi->rq[i], GFP_KERNEL))
635 schedule_delayed_work(&vi->refill, 0);
636 virtnet_napi_enable(&vi->rq[i]);
637 }
638
639 return 0;
640}
641
Linus Torvaldsb7dfde92012-12-20 08:37:04 -0800642static void free_old_xmit_skbs(struct send_queue *sq)
Rusty Russell296f96f2007-10-22 11:03:37 +1000643{
644 struct sk_buff *skb;
Michael S. Tsirkin6ee57bc2012-10-16 23:56:14 +1030645 unsigned int len;
Jason Wange9d74172012-12-07 07:04:55 +0000646 struct virtnet_info *vi = sq->vq->vdev->priv;
Eric Dumazet58472a72012-02-13 06:53:41 +0000647 struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
Rusty Russell296f96f2007-10-22 11:03:37 +1000648
Jason Wange9d74172012-12-07 07:04:55 +0000649 while ((skb = virtqueue_get_buf(sq->vq, &len)) != NULL) {
Rusty Russell296f96f2007-10-22 11:03:37 +1000650 pr_debug("Sent skb %p\n", skb);
stephen hemminger3fa2a1d2011-06-15 06:36:29 +0000651
Eric Dumazet83a27052012-06-05 22:35:24 +0000652 u64_stats_update_begin(&stats->tx_syncp);
stephen hemminger3fa2a1d2011-06-15 06:36:29 +0000653 stats->tx_bytes += skb->len;
654 stats->tx_packets++;
Eric Dumazet83a27052012-06-05 22:35:24 +0000655 u64_stats_update_end(&stats->tx_syncp);
stephen hemminger3fa2a1d2011-06-15 06:36:29 +0000656
Eric Dumazeted79bab2009-10-14 14:36:43 +0000657 dev_kfree_skb_any(skb);
Rusty Russell296f96f2007-10-22 11:03:37 +1000658 }
659}
660
Jason Wange9d74172012-12-07 07:04:55 +0000661static int xmit_skb(struct send_queue *sq, struct sk_buff *skb)
Rusty Russell296f96f2007-10-22 11:03:37 +1000662{
Rusty Russellb3f24692009-09-24 09:59:19 -0600663 struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
Rusty Russell296f96f2007-10-22 11:03:37 +1000664 const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
Jason Wange9d74172012-12-07 07:04:55 +0000665 struct virtnet_info *vi = sq->vq->vdev->priv;
Michael S. Tsirkin7bedc7d2012-10-16 23:56:14 +1030666 unsigned num_sg;
Rusty Russell296f96f2007-10-22 11:03:37 +1000667
Johannes Berge1749612008-10-27 15:59:26 -0700668 pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
Rusty Russell296f96f2007-10-22 11:03:37 +1000669
Rusty Russell296f96f2007-10-22 11:03:37 +1000670 if (skb->ip_summed == CHECKSUM_PARTIAL) {
Rusty Russellb3f24692009-09-24 09:59:19 -0600671 hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
Michał Mirosław55508d62010-12-14 15:24:08 +0000672 hdr->hdr.csum_start = skb_checksum_start_offset(skb);
Rusty Russellb3f24692009-09-24 09:59:19 -0600673 hdr->hdr.csum_offset = skb->csum_offset;
Rusty Russell296f96f2007-10-22 11:03:37 +1000674 } else {
Rusty Russellb3f24692009-09-24 09:59:19 -0600675 hdr->hdr.flags = 0;
676 hdr->hdr.csum_offset = hdr->hdr.csum_start = 0;
Rusty Russell296f96f2007-10-22 11:03:37 +1000677 }
678
679 if (skb_is_gso(skb)) {
Rusty Russellb3f24692009-09-24 09:59:19 -0600680 hdr->hdr.hdr_len = skb_headlen(skb);
681 hdr->hdr.gso_size = skb_shinfo(skb)->gso_size;
Rusty Russell34a48572008-02-04 23:50:02 -0500682 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
Rusty Russellb3f24692009-09-24 09:59:19 -0600683 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
Rusty Russell296f96f2007-10-22 11:03:37 +1000684 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
Rusty Russellb3f24692009-09-24 09:59:19 -0600685 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
Rusty Russell296f96f2007-10-22 11:03:37 +1000686 else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
Rusty Russellb3f24692009-09-24 09:59:19 -0600687 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
Rusty Russell296f96f2007-10-22 11:03:37 +1000688 else
689 BUG();
Rusty Russell34a48572008-02-04 23:50:02 -0500690 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
Rusty Russellb3f24692009-09-24 09:59:19 -0600691 hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
Rusty Russell296f96f2007-10-22 11:03:37 +1000692 } else {
Rusty Russellb3f24692009-09-24 09:59:19 -0600693 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
694 hdr->hdr.gso_size = hdr->hdr.hdr_len = 0;
Rusty Russell296f96f2007-10-22 11:03:37 +1000695 }
696
Rusty Russellb3f24692009-09-24 09:59:19 -0600697 hdr->mhdr.num_buffers = 0;
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800698
699 /* Encode metadata header at front. */
700 if (vi->mergeable_rx_bufs)
Jason Wange9d74172012-12-07 07:04:55 +0000701 sg_set_buf(sq->sg, &hdr->mhdr, sizeof hdr->mhdr);
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800702 else
Jason Wange9d74172012-12-07 07:04:55 +0000703 sg_set_buf(sq->sg, &hdr->hdr, sizeof hdr->hdr);
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800704
Linus Torvaldsb7dfde92012-12-20 08:37:04 -0800705 num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len) + 1;
706 return virtqueue_add_buf(sq->vq, sq->sg, num_sg,
Rusty Russellf96fde42012-01-12 15:44:42 +1030707 0, skb, GFP_ATOMIC);
Rusty Russell11a3a152008-05-26 17:48:13 +1000708}
709
Stephen Hemminger424efe92009-08-31 19:50:51 +0000710static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
Rusty Russell99ffc692008-05-02 21:50:46 -0500711{
712 struct virtnet_info *vi = netdev_priv(dev);
Jason Wang986a4f42012-12-07 07:04:56 +0000713 int qnum = skb_get_queue_mapping(skb);
714 struct send_queue *sq = &vi->sq[qnum];
Rusty Russell9ed4cb02012-10-16 23:56:14 +1030715 int err;
Rusty Russell2cb9c6b2008-02-04 23:50:07 -0500716
Rusty Russell2cb9c6b2008-02-04 23:50:07 -0500717 /* Free up any pending old buffers before queueing new ones. */
Jason Wange9d74172012-12-07 07:04:55 +0000718 free_old_xmit_skbs(sq);
Rusty Russell2cb9c6b2008-02-04 23:50:07 -0500719
Michael S. Tsirkin03f191b2009-10-28 04:03:38 -0700720 /* Try to transmit */
Linus Torvaldsb7dfde92012-12-20 08:37:04 -0800721 err = xmit_skb(sq, skb);
Rusty Russell48925e32009-09-24 09:59:20 -0600722
Rusty Russell9ed4cb02012-10-16 23:56:14 +1030723 /* This should not happen! */
Rusty Russell0e3daa62012-10-16 23:56:15 +1030724 if (unlikely(err)) {
Rusty Russell9ed4cb02012-10-16 23:56:14 +1030725 dev->stats.tx_fifo_errors++;
726 if (net_ratelimit())
727 dev_warn(&dev->dev,
Linus Torvaldsb7dfde92012-12-20 08:37:04 -0800728 "Unexpected TXQ (%d) queue failure: %d\n", qnum, err);
Rusty Russell58eba972010-07-02 16:34:01 +0000729 dev->stats.tx_dropped++;
730 kfree_skb(skb);
731 return NETDEV_TX_OK;
Rusty Russell99ffc692008-05-02 21:50:46 -0500732 }
Jason Wange9d74172012-12-07 07:04:55 +0000733 virtqueue_kick(sq->vq);
Michael S. Tsirkin03f191b2009-10-28 04:03:38 -0700734
Rusty Russell48925e32009-09-24 09:59:20 -0600735 /* Don't wait up for transmitted skbs to be freed. */
736 skb_orphan(skb);
737 nf_reset(skb);
Rusty Russell99ffc692008-05-02 21:50:46 -0500738
Rusty Russell48925e32009-09-24 09:59:20 -0600739 /* Apparently nice girls don't return TX_BUSY; stop the queue
740 * before it gets out of hand. Naturally, this wastes entries. */
Linus Torvaldsb7dfde92012-12-20 08:37:04 -0800741 if (sq->vq->num_free < 2+MAX_SKB_FRAGS) {
Jason Wang986a4f42012-12-07 07:04:56 +0000742 netif_stop_subqueue(dev, qnum);
Jason Wange9d74172012-12-07 07:04:55 +0000743 if (unlikely(!virtqueue_enable_cb_delayed(sq->vq))) {
Rusty Russell48925e32009-09-24 09:59:20 -0600744 /* More just got used, free them then recheck. */
Linus Torvaldsb7dfde92012-12-20 08:37:04 -0800745 free_old_xmit_skbs(sq);
746 if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) {
Jason Wang986a4f42012-12-07 07:04:56 +0000747 netif_start_subqueue(dev, qnum);
Jason Wange9d74172012-12-07 07:04:55 +0000748 virtqueue_disable_cb(sq->vq);
Rusty Russell48925e32009-09-24 09:59:20 -0600749 }
750 }
Rusty Russell99ffc692008-05-02 21:50:46 -0500751 }
Rusty Russell48925e32009-09-24 09:59:20 -0600752
753 return NETDEV_TX_OK;
Rusty Russell296f96f2007-10-22 11:03:37 +1000754}
755
Alex Williamson9c46f6d2009-02-04 16:36:34 -0800756static int virtnet_set_mac_address(struct net_device *dev, void *p)
757{
758 struct virtnet_info *vi = netdev_priv(dev);
759 struct virtio_device *vdev = vi->vdev;
Jiri Pirkof2f2c8b2012-06-29 05:10:06 +0000760 int ret;
Alex Williamson9c46f6d2009-02-04 16:36:34 -0800761
Jiri Pirkof2f2c8b2012-06-29 05:10:06 +0000762 ret = eth_mac_addr(dev, p);
763 if (ret)
764 return ret;
Alex Williamson9c46f6d2009-02-04 16:36:34 -0800765
Alex Williamson62994b22009-04-04 16:40:19 -0700766 if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
767 vdev->config->set(vdev, offsetof(struct virtio_net_config, mac),
768 dev->dev_addr, dev->addr_len);
Alex Williamson9c46f6d2009-02-04 16:36:34 -0800769
770 return 0;
771}
772
stephen hemminger3fa2a1d2011-06-15 06:36:29 +0000773static struct rtnl_link_stats64 *virtnet_stats(struct net_device *dev,
774 struct rtnl_link_stats64 *tot)
775{
776 struct virtnet_info *vi = netdev_priv(dev);
777 int cpu;
778 unsigned int start;
779
780 for_each_possible_cpu(cpu) {
Eric Dumazet58472a72012-02-13 06:53:41 +0000781 struct virtnet_stats *stats = per_cpu_ptr(vi->stats, cpu);
stephen hemminger3fa2a1d2011-06-15 06:36:29 +0000782 u64 tpackets, tbytes, rpackets, rbytes;
783
784 do {
Kevin Groenevelde3906482012-07-21 06:30:50 +0000785 start = u64_stats_fetch_begin_bh(&stats->tx_syncp);
stephen hemminger3fa2a1d2011-06-15 06:36:29 +0000786 tpackets = stats->tx_packets;
787 tbytes = stats->tx_bytes;
Kevin Groenevelde3906482012-07-21 06:30:50 +0000788 } while (u64_stats_fetch_retry_bh(&stats->tx_syncp, start));
Eric Dumazet83a27052012-06-05 22:35:24 +0000789
790 do {
Kevin Groenevelde3906482012-07-21 06:30:50 +0000791 start = u64_stats_fetch_begin_bh(&stats->rx_syncp);
stephen hemminger3fa2a1d2011-06-15 06:36:29 +0000792 rpackets = stats->rx_packets;
793 rbytes = stats->rx_bytes;
Kevin Groenevelde3906482012-07-21 06:30:50 +0000794 } while (u64_stats_fetch_retry_bh(&stats->rx_syncp, start));
stephen hemminger3fa2a1d2011-06-15 06:36:29 +0000795
796 tot->rx_packets += rpackets;
797 tot->tx_packets += tpackets;
798 tot->rx_bytes += rbytes;
799 tot->tx_bytes += tbytes;
800 }
801
802 tot->tx_dropped = dev->stats.tx_dropped;
Rick Jones021ac8d2011-11-21 09:28:17 +0000803 tot->tx_fifo_errors = dev->stats.tx_fifo_errors;
stephen hemminger3fa2a1d2011-06-15 06:36:29 +0000804 tot->rx_dropped = dev->stats.rx_dropped;
805 tot->rx_length_errors = dev->stats.rx_length_errors;
806 tot->rx_frame_errors = dev->stats.rx_frame_errors;
807
808 return tot;
809}
810
Amit Shahda74e892008-02-29 16:24:50 +0530811#ifdef CONFIG_NET_POLL_CONTROLLER
812static void virtnet_netpoll(struct net_device *dev)
813{
814 struct virtnet_info *vi = netdev_priv(dev);
Jason Wang986a4f42012-12-07 07:04:56 +0000815 int i;
Amit Shahda74e892008-02-29 16:24:50 +0530816
Jason Wang986a4f42012-12-07 07:04:56 +0000817 for (i = 0; i < vi->curr_queue_pairs; i++)
818 napi_schedule(&vi->rq[i].napi);
Amit Shahda74e892008-02-29 16:24:50 +0530819}
820#endif
821
Alex Williamson2a41f712009-02-04 09:02:34 +0000822/*
823 * Send command via the control virtqueue and check status. Commands
824 * supported by the hypervisor, as indicated by feature bits, should
825 * never fail unless improperly formated.
826 */
827static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
828 struct scatterlist *data, int out, int in)
829{
Alex Williamson23e258e2009-05-01 17:27:56 +0000830 struct scatterlist *s, sg[VIRTNET_SEND_COMMAND_SG_MAX + 2];
Alex Williamson2a41f712009-02-04 09:02:34 +0000831 struct virtio_net_ctrl_hdr ctrl;
832 virtio_net_ctrl_ack status = ~0;
833 unsigned int tmp;
Alex Williamson23e258e2009-05-01 17:27:56 +0000834 int i;
Alex Williamson2a41f712009-02-04 09:02:34 +0000835
Alexander Beregalov0ee904c2009-04-11 14:50:23 +0000836 /* Caller should know better */
837 BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ||
838 (out + in > VIRTNET_SEND_COMMAND_SG_MAX));
Alex Williamson2a41f712009-02-04 09:02:34 +0000839
840 out++; /* Add header */
841 in++; /* Add return status */
842
843 ctrl.class = class;
844 ctrl.cmd = cmd;
845
846 sg_init_table(sg, out + in);
847
848 sg_set_buf(&sg[0], &ctrl, sizeof(ctrl));
Alex Williamson23e258e2009-05-01 17:27:56 +0000849 for_each_sg(data, s, out + in - 2, i)
850 sg_set_buf(&sg[i + 1], sg_virt(s), s->length);
Alex Williamson2a41f712009-02-04 09:02:34 +0000851 sg_set_buf(&sg[out + in - 1], &status, sizeof(status));
852
Rusty Russellf96fde42012-01-12 15:44:42 +1030853 BUG_ON(virtqueue_add_buf(vi->cvq, sg, out, in, vi, GFP_ATOMIC) < 0);
Alex Williamson2a41f712009-02-04 09:02:34 +0000854
Michael S. Tsirkin1915a7122010-04-12 16:19:04 +0300855 virtqueue_kick(vi->cvq);
Alex Williamson2a41f712009-02-04 09:02:34 +0000856
857 /*
858 * Spin for a response, the kick causes an ioport write, trapping
859 * into the hypervisor, so the request should be handled immediately.
860 */
Michael S. Tsirkin1915a7122010-04-12 16:19:04 +0300861 while (!virtqueue_get_buf(vi->cvq, &tmp))
Alex Williamson2a41f712009-02-04 09:02:34 +0000862 cpu_relax();
863
864 return status == VIRTIO_NET_OK;
865}
866
Jason Wang586d17c2012-04-11 20:43:52 +0000867static void virtnet_ack_link_announce(struct virtnet_info *vi)
868{
869 rtnl_lock();
870 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE,
871 VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL,
872 0, 0))
873 dev_warn(&vi->dev->dev, "Failed to ack link announce.\n");
874 rtnl_unlock();
875}
876
Jason Wang986a4f42012-12-07 07:04:56 +0000877static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs)
878{
879 struct scatterlist sg;
880 struct virtio_net_ctrl_mq s;
881 struct net_device *dev = vi->dev;
882
883 if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ))
884 return 0;
885
886 s.virtqueue_pairs = queue_pairs;
887 sg_init_one(&sg, &s, sizeof(s));
888
889 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ,
890 VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg, 1, 0)){
891 dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n",
892 queue_pairs);
893 return -EINVAL;
894 } else
895 vi->curr_queue_pairs = queue_pairs;
896
897 return 0;
898}
899
Rusty Russell296f96f2007-10-22 11:03:37 +1000900static int virtnet_close(struct net_device *dev)
901{
902 struct virtnet_info *vi = netdev_priv(dev);
Jason Wang986a4f42012-12-07 07:04:56 +0000903 int i;
Rusty Russell296f96f2007-10-22 11:03:37 +1000904
Rusty Russellb2baed62011-12-29 00:42:38 +0000905 /* Make sure refill_work doesn't re-enable napi! */
906 cancel_delayed_work_sync(&vi->refill);
Jason Wang986a4f42012-12-07 07:04:56 +0000907
908 for (i = 0; i < vi->max_queue_pairs; i++)
909 napi_disable(&vi->rq[i].napi);
Rusty Russell296f96f2007-10-22 11:03:37 +1000910
Rusty Russell296f96f2007-10-22 11:03:37 +1000911 return 0;
912}
913
Alex Williamson2af76982009-02-04 09:02:40 +0000914static void virtnet_set_rx_mode(struct net_device *dev)
915{
916 struct virtnet_info *vi = netdev_priv(dev);
Alex Williamsonf565a7c2009-02-04 09:02:45 +0000917 struct scatterlist sg[2];
Alex Williamson2af76982009-02-04 09:02:40 +0000918 u8 promisc, allmulti;
Alex Williamsonf565a7c2009-02-04 09:02:45 +0000919 struct virtio_net_ctrl_mac *mac_data;
Jiri Pirkoccffad252009-05-22 23:22:17 +0000920 struct netdev_hw_addr *ha;
Jiri Pirko32e7bfc2010-01-25 13:36:10 -0800921 int uc_count;
Jiri Pirko4cd24ea2010-02-08 04:30:35 +0000922 int mc_count;
Alex Williamsonf565a7c2009-02-04 09:02:45 +0000923 void *buf;
924 int i;
Alex Williamson2af76982009-02-04 09:02:40 +0000925
926 /* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */
927 if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
928 return;
929
Alex Williamsonf565a7c2009-02-04 09:02:45 +0000930 promisc = ((dev->flags & IFF_PROMISC) != 0);
931 allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
Alex Williamson2af76982009-02-04 09:02:40 +0000932
Alex Williamson23e258e2009-05-01 17:27:56 +0000933 sg_init_one(sg, &promisc, sizeof(promisc));
Alex Williamson2af76982009-02-04 09:02:40 +0000934
935 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
936 VIRTIO_NET_CTRL_RX_PROMISC,
Alex Williamsonf565a7c2009-02-04 09:02:45 +0000937 sg, 1, 0))
Alex Williamson2af76982009-02-04 09:02:40 +0000938 dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
939 promisc ? "en" : "dis");
940
Alex Williamson23e258e2009-05-01 17:27:56 +0000941 sg_init_one(sg, &allmulti, sizeof(allmulti));
Alex Williamson2af76982009-02-04 09:02:40 +0000942
943 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
944 VIRTIO_NET_CTRL_RX_ALLMULTI,
Alex Williamsonf565a7c2009-02-04 09:02:45 +0000945 sg, 1, 0))
Alex Williamson2af76982009-02-04 09:02:40 +0000946 dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
947 allmulti ? "en" : "dis");
Alex Williamsonf565a7c2009-02-04 09:02:45 +0000948
Jiri Pirko32e7bfc2010-01-25 13:36:10 -0800949 uc_count = netdev_uc_count(dev);
Jiri Pirko4cd24ea2010-02-08 04:30:35 +0000950 mc_count = netdev_mc_count(dev);
Alex Williamsonf565a7c2009-02-04 09:02:45 +0000951 /* MAC filter - use one buffer for both lists */
Jiri Pirko4cd24ea2010-02-08 04:30:35 +0000952 buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
953 (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
954 mac_data = buf;
Alex Williamsonf565a7c2009-02-04 09:02:45 +0000955 if (!buf) {
956 dev_warn(&dev->dev, "No memory for MAC address buffer\n");
957 return;
958 }
959
Alex Williamson23e258e2009-05-01 17:27:56 +0000960 sg_init_table(sg, 2);
961
Alex Williamsonf565a7c2009-02-04 09:02:45 +0000962 /* Store the unicast list and count in the front of the buffer */
Jiri Pirko32e7bfc2010-01-25 13:36:10 -0800963 mac_data->entries = uc_count;
Jiri Pirkoccffad252009-05-22 23:22:17 +0000964 i = 0;
Jiri Pirko32e7bfc2010-01-25 13:36:10 -0800965 netdev_for_each_uc_addr(ha, dev)
Jiri Pirkoccffad252009-05-22 23:22:17 +0000966 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
Alex Williamsonf565a7c2009-02-04 09:02:45 +0000967
968 sg_set_buf(&sg[0], mac_data,
Jiri Pirko32e7bfc2010-01-25 13:36:10 -0800969 sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
Alex Williamsonf565a7c2009-02-04 09:02:45 +0000970
971 /* multicast list and count fill the end */
Jiri Pirko32e7bfc2010-01-25 13:36:10 -0800972 mac_data = (void *)&mac_data->macs[uc_count][0];
Alex Williamsonf565a7c2009-02-04 09:02:45 +0000973
Jiri Pirko4cd24ea2010-02-08 04:30:35 +0000974 mac_data->entries = mc_count;
Jiri Pirko567ec872010-02-23 23:17:07 +0000975 i = 0;
Jiri Pirko22bedad32010-04-01 21:22:57 +0000976 netdev_for_each_mc_addr(ha, dev)
977 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
Alex Williamsonf565a7c2009-02-04 09:02:45 +0000978
979 sg_set_buf(&sg[1], mac_data,
Jiri Pirko4cd24ea2010-02-08 04:30:35 +0000980 sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
Alex Williamsonf565a7c2009-02-04 09:02:45 +0000981
982 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
983 VIRTIO_NET_CTRL_MAC_TABLE_SET,
984 sg, 2, 0))
985 dev_warn(&dev->dev, "Failed to set MAC fitler table.\n");
986
987 kfree(buf);
Alex Williamson2af76982009-02-04 09:02:40 +0000988}
989
Jiri Pirko8e586132011-12-08 19:52:37 -0500990static int virtnet_vlan_rx_add_vid(struct net_device *dev, u16 vid)
Alex Williamson0bde95692009-02-04 09:02:50 +0000991{
992 struct virtnet_info *vi = netdev_priv(dev);
993 struct scatterlist sg;
994
Alex Williamson23e258e2009-05-01 17:27:56 +0000995 sg_init_one(&sg, &vid, sizeof(vid));
Alex Williamson0bde95692009-02-04 09:02:50 +0000996
997 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
998 VIRTIO_NET_CTRL_VLAN_ADD, &sg, 1, 0))
999 dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
Jiri Pirko8e586132011-12-08 19:52:37 -05001000 return 0;
Alex Williamson0bde95692009-02-04 09:02:50 +00001001}
1002
Jiri Pirko8e586132011-12-08 19:52:37 -05001003static int virtnet_vlan_rx_kill_vid(struct net_device *dev, u16 vid)
Alex Williamson0bde95692009-02-04 09:02:50 +00001004{
1005 struct virtnet_info *vi = netdev_priv(dev);
1006 struct scatterlist sg;
1007
Alex Williamson23e258e2009-05-01 17:27:56 +00001008 sg_init_one(&sg, &vid, sizeof(vid));
Alex Williamson0bde95692009-02-04 09:02:50 +00001009
1010 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1011 VIRTIO_NET_CTRL_VLAN_DEL, &sg, 1, 0))
1012 dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
Jiri Pirko8e586132011-12-08 19:52:37 -05001013 return 0;
Alex Williamson0bde95692009-02-04 09:02:50 +00001014}
1015
Jason Wang986a4f42012-12-07 07:04:56 +00001016static void virtnet_set_affinity(struct virtnet_info *vi, bool set)
1017{
1018 int i;
1019
1020 /* In multiqueue mode, when the number of cpu is equal to the number of
1021 * queue pairs, we let the queue pairs to be private to one cpu by
1022 * setting the affinity hint to eliminate the contention.
1023 */
1024 if ((vi->curr_queue_pairs == 1 ||
1025 vi->max_queue_pairs != num_online_cpus()) && set) {
1026 if (vi->affinity_hint_set)
1027 set = false;
1028 else
1029 return;
1030 }
1031
1032 for (i = 0; i < vi->max_queue_pairs; i++) {
1033 int cpu = set ? i : -1;
1034 virtqueue_set_affinity(vi->rq[i].vq, cpu);
1035 virtqueue_set_affinity(vi->sq[i].vq, cpu);
1036 }
1037
1038 if (set)
1039 vi->affinity_hint_set = true;
1040 else
1041 vi->affinity_hint_set = false;
1042}
1043
Rick Jones8f9f4662011-10-19 08:10:59 +00001044static void virtnet_get_ringparam(struct net_device *dev,
1045 struct ethtool_ringparam *ring)
1046{
1047 struct virtnet_info *vi = netdev_priv(dev);
1048
Jason Wang986a4f42012-12-07 07:04:56 +00001049 ring->rx_max_pending = virtqueue_get_vring_size(vi->rq[0].vq);
1050 ring->tx_max_pending = virtqueue_get_vring_size(vi->sq[0].vq);
Rick Jones8f9f4662011-10-19 08:10:59 +00001051 ring->rx_pending = ring->rx_max_pending;
1052 ring->tx_pending = ring->tx_max_pending;
Rick Jones8f9f4662011-10-19 08:10:59 +00001053}
1054
Rick Jones66846042011-11-14 14:17:08 +00001055
1056static void virtnet_get_drvinfo(struct net_device *dev,
1057 struct ethtool_drvinfo *info)
1058{
1059 struct virtnet_info *vi = netdev_priv(dev);
1060 struct virtio_device *vdev = vi->vdev;
1061
1062 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
1063 strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version));
1064 strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info));
1065
1066}
1067
Jason Wangd73bcd22012-12-07 07:04:57 +00001068/* TODO: Eliminate OOO packets during switching */
1069static int virtnet_set_channels(struct net_device *dev,
1070 struct ethtool_channels *channels)
1071{
1072 struct virtnet_info *vi = netdev_priv(dev);
1073 u16 queue_pairs = channels->combined_count;
1074 int err;
1075
1076 /* We don't support separate rx/tx channels.
1077 * We don't allow setting 'other' channels.
1078 */
1079 if (channels->rx_count || channels->tx_count || channels->other_count)
1080 return -EINVAL;
1081
1082 if (queue_pairs > vi->max_queue_pairs)
1083 return -EINVAL;
1084
1085 err = virtnet_set_queues(vi, queue_pairs);
1086 if (!err) {
1087 netif_set_real_num_tx_queues(dev, queue_pairs);
1088 netif_set_real_num_rx_queues(dev, queue_pairs);
1089
1090 virtnet_set_affinity(vi, true);
1091 }
1092
1093 return err;
1094}
1095
1096static void virtnet_get_channels(struct net_device *dev,
1097 struct ethtool_channels *channels)
1098{
1099 struct virtnet_info *vi = netdev_priv(dev);
1100
1101 channels->combined_count = vi->curr_queue_pairs;
1102 channels->max_combined = vi->max_queue_pairs;
1103 channels->max_other = 0;
1104 channels->rx_count = 0;
1105 channels->tx_count = 0;
1106 channels->other_count = 0;
1107}
1108
Stephen Hemminger0fc0b732009-09-02 01:03:33 -07001109static const struct ethtool_ops virtnet_ethtool_ops = {
Rick Jones66846042011-11-14 14:17:08 +00001110 .get_drvinfo = virtnet_get_drvinfo,
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -08001111 .get_link = ethtool_op_get_link,
Rick Jones8f9f4662011-10-19 08:10:59 +00001112 .get_ringparam = virtnet_get_ringparam,
Jason Wangd73bcd22012-12-07 07:04:57 +00001113 .set_channels = virtnet_set_channels,
1114 .get_channels = virtnet_get_channels,
Herbert Xua9ea3fc2008-04-18 11:21:42 +08001115};
1116
Mark McLoughlin39da5812008-11-26 13:58:11 +00001117#define MIN_MTU 68
1118#define MAX_MTU 65535
1119
1120static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
1121{
1122 if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
1123 return -EINVAL;
1124 dev->mtu = new_mtu;
1125 return 0;
1126}
1127
Jason Wang986a4f42012-12-07 07:04:56 +00001128/* To avoid contending a lock hold by a vcpu who would exit to host, select the
1129 * txq based on the processor id.
1130 * TODO: handle cpu hotplug.
1131 */
1132static u16 virtnet_select_queue(struct net_device *dev, struct sk_buff *skb)
1133{
1134 int txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
1135 smp_processor_id();
1136
1137 while (unlikely(txq >= dev->real_num_tx_queues))
1138 txq -= dev->real_num_tx_queues;
1139
1140 return txq;
1141}
1142
Stephen Hemminger76288b42009-01-06 10:44:22 -08001143static const struct net_device_ops virtnet_netdev = {
1144 .ndo_open = virtnet_open,
1145 .ndo_stop = virtnet_close,
1146 .ndo_start_xmit = start_xmit,
1147 .ndo_validate_addr = eth_validate_addr,
Alex Williamson9c46f6d2009-02-04 16:36:34 -08001148 .ndo_set_mac_address = virtnet_set_mac_address,
Alex Williamson2af76982009-02-04 09:02:40 +00001149 .ndo_set_rx_mode = virtnet_set_rx_mode,
Stephen Hemminger76288b42009-01-06 10:44:22 -08001150 .ndo_change_mtu = virtnet_change_mtu,
stephen hemminger3fa2a1d2011-06-15 06:36:29 +00001151 .ndo_get_stats64 = virtnet_stats,
Alex Williamson1824a982009-05-01 17:31:10 +00001152 .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
1153 .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
Jason Wang986a4f42012-12-07 07:04:56 +00001154 .ndo_select_queue = virtnet_select_queue,
Stephen Hemminger76288b42009-01-06 10:44:22 -08001155#ifdef CONFIG_NET_POLL_CONTROLLER
1156 .ndo_poll_controller = virtnet_netpoll,
1157#endif
1158};
1159
Jason Wang586d17c2012-04-11 20:43:52 +00001160static void virtnet_config_changed_work(struct work_struct *work)
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -08001161{
Jason Wang586d17c2012-04-11 20:43:52 +00001162 struct virtnet_info *vi =
1163 container_of(work, struct virtnet_info, config_work);
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -08001164 u16 v;
1165
Jason Wang586d17c2012-04-11 20:43:52 +00001166 mutex_lock(&vi->config_lock);
1167 if (!vi->config_enable)
1168 goto done;
1169
Sasha Levin77dd7692011-08-14 17:52:33 +03001170 if (virtio_config_val(vi->vdev, VIRTIO_NET_F_STATUS,
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -08001171 offsetof(struct virtio_net_config, status),
Sasha Levin77dd7692011-08-14 17:52:33 +03001172 &v) < 0)
Jason Wang586d17c2012-04-11 20:43:52 +00001173 goto done;
1174
1175 if (v & VIRTIO_NET_S_ANNOUNCE) {
Amerigo Wangee89bab2012-08-09 22:14:56 +00001176 netdev_notify_peers(vi->dev);
Jason Wang586d17c2012-04-11 20:43:52 +00001177 virtnet_ack_link_announce(vi);
1178 }
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -08001179
1180 /* Ignore unknown (future) status bits */
1181 v &= VIRTIO_NET_S_LINK_UP;
1182
1183 if (vi->status == v)
Jason Wang586d17c2012-04-11 20:43:52 +00001184 goto done;
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -08001185
1186 vi->status = v;
1187
1188 if (vi->status & VIRTIO_NET_S_LINK_UP) {
1189 netif_carrier_on(vi->dev);
Jason Wang986a4f42012-12-07 07:04:56 +00001190 netif_tx_wake_all_queues(vi->dev);
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -08001191 } else {
1192 netif_carrier_off(vi->dev);
Jason Wang986a4f42012-12-07 07:04:56 +00001193 netif_tx_stop_all_queues(vi->dev);
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -08001194 }
Jason Wang586d17c2012-04-11 20:43:52 +00001195done:
1196 mutex_unlock(&vi->config_lock);
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -08001197}
1198
1199static void virtnet_config_changed(struct virtio_device *vdev)
1200{
1201 struct virtnet_info *vi = vdev->priv;
1202
Tejun Heo3b07e9c2012-08-20 14:51:24 -07001203 schedule_work(&vi->config_work);
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -08001204}
1205
Jason Wang986a4f42012-12-07 07:04:56 +00001206static void virtnet_free_queues(struct virtnet_info *vi)
1207{
1208 kfree(vi->rq);
1209 kfree(vi->sq);
1210}
1211
1212static void free_receive_bufs(struct virtnet_info *vi)
1213{
1214 int i;
1215
1216 for (i = 0; i < vi->max_queue_pairs; i++) {
1217 while (vi->rq[i].pages)
1218 __free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0);
1219 }
1220}
1221
1222static void free_unused_bufs(struct virtnet_info *vi)
1223{
1224 void *buf;
1225 int i;
1226
1227 for (i = 0; i < vi->max_queue_pairs; i++) {
1228 struct virtqueue *vq = vi->sq[i].vq;
1229 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL)
1230 dev_kfree_skb(buf);
1231 }
1232
1233 for (i = 0; i < vi->max_queue_pairs; i++) {
1234 struct virtqueue *vq = vi->rq[i].vq;
1235
1236 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) {
1237 if (vi->mergeable_rx_bufs || vi->big_packets)
1238 give_pages(&vi->rq[i], buf);
1239 else
1240 dev_kfree_skb(buf);
1241 --vi->rq[i].num;
1242 }
1243 BUG_ON(vi->rq[i].num != 0);
1244 }
1245}
1246
Jason Wange9d74172012-12-07 07:04:55 +00001247static void virtnet_del_vqs(struct virtnet_info *vi)
1248{
1249 struct virtio_device *vdev = vi->vdev;
1250
Jason Wang986a4f42012-12-07 07:04:56 +00001251 virtnet_set_affinity(vi, false);
1252
Jason Wange9d74172012-12-07 07:04:55 +00001253 vdev->config->del_vqs(vdev);
Jason Wang986a4f42012-12-07 07:04:56 +00001254
1255 virtnet_free_queues(vi);
1256}
1257
1258static int virtnet_find_vqs(struct virtnet_info *vi)
1259{
1260 vq_callback_t **callbacks;
1261 struct virtqueue **vqs;
1262 int ret = -ENOMEM;
1263 int i, total_vqs;
1264 const char **names;
1265
1266 /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
1267 * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
1268 * possible control vq.
1269 */
1270 total_vqs = vi->max_queue_pairs * 2 +
1271 virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ);
1272
1273 /* Allocate space for find_vqs parameters */
1274 vqs = kzalloc(total_vqs * sizeof(*vqs), GFP_KERNEL);
1275 if (!vqs)
1276 goto err_vq;
1277 callbacks = kmalloc(total_vqs * sizeof(*callbacks), GFP_KERNEL);
1278 if (!callbacks)
1279 goto err_callback;
1280 names = kmalloc(total_vqs * sizeof(*names), GFP_KERNEL);
1281 if (!names)
1282 goto err_names;
1283
1284 /* Parameters for control virtqueue, if any */
1285 if (vi->has_cvq) {
1286 callbacks[total_vqs - 1] = NULL;
1287 names[total_vqs - 1] = "control";
1288 }
1289
1290 /* Allocate/initialize parameters for send/receive virtqueues */
1291 for (i = 0; i < vi->max_queue_pairs; i++) {
1292 callbacks[rxq2vq(i)] = skb_recv_done;
1293 callbacks[txq2vq(i)] = skb_xmit_done;
1294 sprintf(vi->rq[i].name, "input.%d", i);
1295 sprintf(vi->sq[i].name, "output.%d", i);
1296 names[rxq2vq(i)] = vi->rq[i].name;
1297 names[txq2vq(i)] = vi->sq[i].name;
1298 }
1299
1300 ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs, callbacks,
1301 names);
1302 if (ret)
1303 goto err_find;
1304
1305 if (vi->has_cvq) {
1306 vi->cvq = vqs[total_vqs - 1];
1307 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
1308 vi->dev->features |= NETIF_F_HW_VLAN_FILTER;
1309 }
1310
1311 for (i = 0; i < vi->max_queue_pairs; i++) {
1312 vi->rq[i].vq = vqs[rxq2vq(i)];
1313 vi->sq[i].vq = vqs[txq2vq(i)];
1314 }
1315
1316 kfree(names);
1317 kfree(callbacks);
1318 kfree(vqs);
1319
1320 return 0;
1321
1322err_find:
1323 kfree(names);
1324err_names:
1325 kfree(callbacks);
1326err_callback:
1327 kfree(vqs);
1328err_vq:
1329 return ret;
1330}
1331
1332static int virtnet_alloc_queues(struct virtnet_info *vi)
1333{
1334 int i;
1335
1336 vi->sq = kzalloc(sizeof(*vi->sq) * vi->max_queue_pairs, GFP_KERNEL);
1337 if (!vi->sq)
1338 goto err_sq;
1339 vi->rq = kzalloc(sizeof(*vi->rq) * vi->max_queue_pairs, GFP_KERNEL);
Amerigo Wang008d4272012-12-10 02:24:08 +00001340 if (!vi->rq)
Jason Wang986a4f42012-12-07 07:04:56 +00001341 goto err_rq;
1342
1343 INIT_DELAYED_WORK(&vi->refill, refill_work);
1344 for (i = 0; i < vi->max_queue_pairs; i++) {
1345 vi->rq[i].pages = NULL;
1346 netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll,
1347 napi_weight);
1348
1349 sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg));
1350 sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg));
1351 }
1352
1353 return 0;
1354
1355err_rq:
1356 kfree(vi->sq);
1357err_sq:
1358 return -ENOMEM;
Jason Wange9d74172012-12-07 07:04:55 +00001359}
1360
Amit Shah3f9c10b2011-12-22 16:58:31 +05301361static int init_vqs(struct virtnet_info *vi)
1362{
Jason Wang986a4f42012-12-07 07:04:56 +00001363 int ret;
Amit Shah3f9c10b2011-12-22 16:58:31 +05301364
Jason Wang986a4f42012-12-07 07:04:56 +00001365 /* Allocate send & receive queues */
1366 ret = virtnet_alloc_queues(vi);
1367 if (ret)
1368 goto err;
Amit Shah3f9c10b2011-12-22 16:58:31 +05301369
Jason Wang986a4f42012-12-07 07:04:56 +00001370 ret = virtnet_find_vqs(vi);
1371 if (ret)
1372 goto err_free;
Amit Shah3f9c10b2011-12-22 16:58:31 +05301373
Jason Wang986a4f42012-12-07 07:04:56 +00001374 virtnet_set_affinity(vi, true);
Amit Shah3f9c10b2011-12-22 16:58:31 +05301375 return 0;
Jason Wang986a4f42012-12-07 07:04:56 +00001376
1377err_free:
1378 virtnet_free_queues(vi);
1379err:
1380 return ret;
Amit Shah3f9c10b2011-12-22 16:58:31 +05301381}
1382
Rusty Russell296f96f2007-10-22 11:03:37 +10001383static int virtnet_probe(struct virtio_device *vdev)
1384{
Jason Wang986a4f42012-12-07 07:04:56 +00001385 int i, err;
Rusty Russell296f96f2007-10-22 11:03:37 +10001386 struct net_device *dev;
1387 struct virtnet_info *vi;
Jason Wang986a4f42012-12-07 07:04:56 +00001388 u16 max_queue_pairs;
1389
1390 /* Find if host supports multiqueue virtio_net device */
1391 err = virtio_config_val(vdev, VIRTIO_NET_F_MQ,
1392 offsetof(struct virtio_net_config,
1393 max_virtqueue_pairs), &max_queue_pairs);
1394
1395 /* We need at least 2 queue's */
1396 if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
1397 max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
1398 !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
1399 max_queue_pairs = 1;
Rusty Russell296f96f2007-10-22 11:03:37 +10001400
1401 /* Allocate ourselves a network device with room for our info */
Jason Wang986a4f42012-12-07 07:04:56 +00001402 dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs);
Rusty Russell296f96f2007-10-22 11:03:37 +10001403 if (!dev)
1404 return -ENOMEM;
1405
1406 /* Set up network device as normal. */
Jiri Pirkof2f2c8b2012-06-29 05:10:06 +00001407 dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE;
Stephen Hemminger76288b42009-01-06 10:44:22 -08001408 dev->netdev_ops = &virtnet_netdev;
Rusty Russell296f96f2007-10-22 11:03:37 +10001409 dev->features = NETIF_F_HIGHDMA;
stephen hemminger3fa2a1d2011-06-15 06:36:29 +00001410
Herbert Xua9ea3fc2008-04-18 11:21:42 +08001411 SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
Rusty Russell296f96f2007-10-22 11:03:37 +10001412 SET_NETDEV_DEV(dev, &vdev->dev);
1413
1414 /* Do we support "hardware" checksums? */
Michał Mirosław98e778c2011-03-31 01:01:35 +00001415 if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
Rusty Russell296f96f2007-10-22 11:03:37 +10001416 /* This opens up the world of extra features. */
Michał Mirosław98e778c2011-03-31 01:01:35 +00001417 dev->hw_features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1418 if (csum)
1419 dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1420
1421 if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
1422 dev->hw_features |= NETIF_F_TSO | NETIF_F_UFO
Rusty Russell34a48572008-02-04 23:50:02 -05001423 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
1424 }
Rusty Russell5539ae92008-05-02 21:50:46 -05001425 /* Individual feature bits: what can host handle? */
Michał Mirosław98e778c2011-03-31 01:01:35 +00001426 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
1427 dev->hw_features |= NETIF_F_TSO;
1428 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
1429 dev->hw_features |= NETIF_F_TSO6;
1430 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
1431 dev->hw_features |= NETIF_F_TSO_ECN;
1432 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
1433 dev->hw_features |= NETIF_F_UFO;
1434
1435 if (gso)
1436 dev->features |= dev->hw_features & (NETIF_F_ALL_TSO|NETIF_F_UFO);
1437 /* (!csum && gso) case will be fixed by register_netdev() */
Rusty Russell296f96f2007-10-22 11:03:37 +10001438 }
1439
1440 /* Configuration may specify what MAC to use. Otherwise random. */
Sasha Levin77dd7692011-08-14 17:52:33 +03001441 if (virtio_config_val_len(vdev, VIRTIO_NET_F_MAC,
Rusty Russella586d4f2008-02-04 23:49:56 -05001442 offsetof(struct virtio_net_config, mac),
Sasha Levin77dd7692011-08-14 17:52:33 +03001443 dev->dev_addr, dev->addr_len) < 0)
Danny Kukawkaf2cedb62012-02-15 06:45:39 +00001444 eth_hw_addr_random(dev);
Rusty Russell296f96f2007-10-22 11:03:37 +10001445
1446 /* Set up our device-specific information */
1447 vi = netdev_priv(dev);
Rusty Russell296f96f2007-10-22 11:03:37 +10001448 vi->dev = dev;
1449 vi->vdev = vdev;
Christian Borntraegerd9d5dcc2008-02-18 10:02:51 +01001450 vdev->priv = vi;
stephen hemminger3fa2a1d2011-06-15 06:36:29 +00001451 vi->stats = alloc_percpu(struct virtnet_stats);
1452 err = -ENOMEM;
1453 if (vi->stats == NULL)
1454 goto free;
1455
Jason Wang586d17c2012-04-11 20:43:52 +00001456 mutex_init(&vi->config_lock);
1457 vi->config_enable = true;
1458 INIT_WORK(&vi->config_work, virtnet_config_changed_work);
Rusty Russell296f96f2007-10-22 11:03:37 +10001459
Herbert Xu97402b92008-04-18 11:24:27 +08001460 /* If we can receive ANY GSO packets, we must allocate large ones. */
Joe Perches8e95a202009-12-03 07:58:21 +00001461 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
1462 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
1463 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
Herbert Xu97402b92008-04-18 11:24:27 +08001464 vi->big_packets = true;
1465
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -08001466 if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
1467 vi->mergeable_rx_bufs = true;
1468
Jason Wang986a4f42012-12-07 07:04:56 +00001469 if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
1470 vi->has_cvq = true;
1471
1472 /* Use single tx/rx queue pair as default */
1473 vi->curr_queue_pairs = 1;
1474 vi->max_queue_pairs = max_queue_pairs;
1475
1476 /* Allocate/initialize the rx/tx queues, and invoke find_vqs */
Amit Shah3f9c10b2011-12-22 16:58:31 +05301477 err = init_vqs(vi);
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -06001478 if (err)
stephen hemminger3fa2a1d2011-06-15 06:36:29 +00001479 goto free_stats;
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -06001480
Jason Wang986a4f42012-12-07 07:04:56 +00001481 netif_set_real_num_tx_queues(dev, 1);
1482 netif_set_real_num_rx_queues(dev, 1);
1483
Rusty Russell296f96f2007-10-22 11:03:37 +10001484 err = register_netdev(dev);
1485 if (err) {
1486 pr_debug("virtio_net: registering device failed\n");
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -06001487 goto free_vqs;
Rusty Russell296f96f2007-10-22 11:03:37 +10001488 }
Rusty Russellb3369c12008-02-04 23:50:02 -05001489
1490 /* Last of all, set up some receive buffers. */
Jason Wang986a4f42012-12-07 07:04:56 +00001491 for (i = 0; i < vi->max_queue_pairs; i++) {
1492 try_fill_recv(&vi->rq[i], GFP_KERNEL);
Rusty Russellb3369c12008-02-04 23:50:02 -05001493
Jason Wang986a4f42012-12-07 07:04:56 +00001494 /* If we didn't even get one input buffer, we're useless. */
1495 if (vi->rq[i].num == 0) {
1496 free_unused_bufs(vi);
1497 err = -ENOMEM;
1498 goto free_recv_bufs;
1499 }
Rusty Russellb3369c12008-02-04 23:50:02 -05001500 }
1501
Jason Wang167c25e2010-11-10 14:45:41 +00001502 /* Assume link up if device can't report link status,
1503 otherwise get link status from config. */
1504 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
1505 netif_carrier_off(dev);
Tejun Heo3b07e9c2012-08-20 14:51:24 -07001506 schedule_work(&vi->config_work);
Jason Wang167c25e2010-11-10 14:45:41 +00001507 } else {
1508 vi->status = VIRTIO_NET_S_LINK_UP;
1509 netif_carrier_on(dev);
1510 }
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -08001511
Jason Wang986a4f42012-12-07 07:04:56 +00001512 pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
1513 dev->name, max_queue_pairs);
1514
Rusty Russell296f96f2007-10-22 11:03:37 +10001515 return 0;
1516
Jason Wang986a4f42012-12-07 07:04:56 +00001517free_recv_bufs:
1518 free_receive_bufs(vi);
Rusty Russellb3369c12008-02-04 23:50:02 -05001519 unregister_netdev(dev);
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -06001520free_vqs:
Jason Wang986a4f42012-12-07 07:04:56 +00001521 cancel_delayed_work_sync(&vi->refill);
Jason Wange9d74172012-12-07 07:04:55 +00001522 virtnet_del_vqs(vi);
stephen hemminger3fa2a1d2011-06-15 06:36:29 +00001523free_stats:
1524 free_percpu(vi->stats);
Rusty Russell296f96f2007-10-22 11:03:37 +10001525free:
1526 free_netdev(dev);
1527 return err;
1528}
1529
Amit Shah04486ed2011-12-22 16:58:32 +05301530static void remove_vq_common(struct virtnet_info *vi)
Rusty Russell296f96f2007-10-22 11:03:37 +10001531{
Amit Shah04486ed2011-12-22 16:58:32 +05301532 vi->vdev->config->reset(vi->vdev);
Shirley Ma830a8a92010-02-08 14:14:42 +00001533
1534 /* Free unused buffers in both send and recv, if any. */
Shirley Ma9ab86bb2010-01-29 03:20:04 +00001535 free_unused_bufs(vi);
Rusty Russellfb6813f2008-07-25 12:06:01 -05001536
Jason Wang986a4f42012-12-07 07:04:56 +00001537 free_receive_bufs(vi);
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -06001538
Jason Wang986a4f42012-12-07 07:04:56 +00001539 virtnet_del_vqs(vi);
Amit Shah04486ed2011-12-22 16:58:32 +05301540}
1541
Bill Pemberton8cc085d2012-12-03 09:24:15 -05001542static void virtnet_remove(struct virtio_device *vdev)
Amit Shah04486ed2011-12-22 16:58:32 +05301543{
1544 struct virtnet_info *vi = vdev->priv;
1545
Jason Wang586d17c2012-04-11 20:43:52 +00001546 /* Prevent config work handler from accessing the device. */
1547 mutex_lock(&vi->config_lock);
1548 vi->config_enable = false;
1549 mutex_unlock(&vi->config_lock);
1550
Amit Shah04486ed2011-12-22 16:58:32 +05301551 unregister_netdev(vi->dev);
1552
1553 remove_vq_common(vi);
Rusty Russellfb6813f2008-07-25 12:06:01 -05001554
Jason Wang586d17c2012-04-11 20:43:52 +00001555 flush_work(&vi->config_work);
1556
Krishna Kumar2e66f552011-07-20 03:56:02 +00001557 free_percpu(vi->stats);
Rusty Russell74b25532007-11-19 11:20:42 -05001558 free_netdev(vi->dev);
Rusty Russell296f96f2007-10-22 11:03:37 +10001559}
1560
Amit Shah0741bcb2011-12-22 16:58:33 +05301561#ifdef CONFIG_PM
1562static int virtnet_freeze(struct virtio_device *vdev)
1563{
1564 struct virtnet_info *vi = vdev->priv;
Jason Wang986a4f42012-12-07 07:04:56 +00001565 int i;
Amit Shah0741bcb2011-12-22 16:58:33 +05301566
Jason Wang586d17c2012-04-11 20:43:52 +00001567 /* Prevent config work handler from accessing the device */
1568 mutex_lock(&vi->config_lock);
1569 vi->config_enable = false;
1570 mutex_unlock(&vi->config_lock);
1571
Amit Shah0741bcb2011-12-22 16:58:33 +05301572 netif_device_detach(vi->dev);
1573 cancel_delayed_work_sync(&vi->refill);
1574
1575 if (netif_running(vi->dev))
Jason Wang986a4f42012-12-07 07:04:56 +00001576 for (i = 0; i < vi->max_queue_pairs; i++) {
1577 napi_disable(&vi->rq[i].napi);
1578 netif_napi_del(&vi->rq[i].napi);
1579 }
Amit Shah0741bcb2011-12-22 16:58:33 +05301580
1581 remove_vq_common(vi);
1582
Jason Wang586d17c2012-04-11 20:43:52 +00001583 flush_work(&vi->config_work);
1584
Amit Shah0741bcb2011-12-22 16:58:33 +05301585 return 0;
1586}
1587
1588static int virtnet_restore(struct virtio_device *vdev)
1589{
1590 struct virtnet_info *vi = vdev->priv;
Jason Wang986a4f42012-12-07 07:04:56 +00001591 int err, i;
Amit Shah0741bcb2011-12-22 16:58:33 +05301592
1593 err = init_vqs(vi);
1594 if (err)
1595 return err;
1596
1597 if (netif_running(vi->dev))
Jason Wang986a4f42012-12-07 07:04:56 +00001598 for (i = 0; i < vi->max_queue_pairs; i++)
1599 virtnet_napi_enable(&vi->rq[i]);
Amit Shah0741bcb2011-12-22 16:58:33 +05301600
1601 netif_device_attach(vi->dev);
1602
Jason Wang986a4f42012-12-07 07:04:56 +00001603 for (i = 0; i < vi->max_queue_pairs; i++)
1604 if (!try_fill_recv(&vi->rq[i], GFP_KERNEL))
1605 schedule_delayed_work(&vi->refill, 0);
Amit Shah0741bcb2011-12-22 16:58:33 +05301606
Jason Wang586d17c2012-04-11 20:43:52 +00001607 mutex_lock(&vi->config_lock);
1608 vi->config_enable = true;
1609 mutex_unlock(&vi->config_lock);
1610
Jason Wang986a4f42012-12-07 07:04:56 +00001611 virtnet_set_queues(vi, vi->curr_queue_pairs);
1612
Amit Shah0741bcb2011-12-22 16:58:33 +05301613 return 0;
1614}
1615#endif
1616
Rusty Russell296f96f2007-10-22 11:03:37 +10001617static struct virtio_device_id id_table[] = {
1618 { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
1619 { 0 },
1620};
1621
Rusty Russellc45a6812008-05-02 21:50:50 -05001622static unsigned int features[] = {
Mark McLoughlin5e4fe5c2008-07-08 17:10:42 +10001623 VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
1624 VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
Rusty Russellc45a6812008-05-02 21:50:50 -05001625 VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
Herbert Xu97402b92008-04-18 11:24:27 +08001626 VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
Sridhar Samudrala5c516752009-07-14 14:21:02 +00001627 VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO,
Alex Williamson2a41f712009-02-04 09:02:34 +00001628 VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ,
Alex Williamson0bde95692009-02-04 09:02:50 +00001629 VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN,
Jason Wang986a4f42012-12-07 07:04:56 +00001630 VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ,
Rusty Russellc45a6812008-05-02 21:50:50 -05001631};
1632
Uwe Kleine-König22402522009-11-05 01:32:44 -08001633static struct virtio_driver virtio_net_driver = {
Rusty Russellc45a6812008-05-02 21:50:50 -05001634 .feature_table = features,
1635 .feature_table_size = ARRAY_SIZE(features),
Rusty Russell296f96f2007-10-22 11:03:37 +10001636 .driver.name = KBUILD_MODNAME,
1637 .driver.owner = THIS_MODULE,
1638 .id_table = id_table,
1639 .probe = virtnet_probe,
Bill Pemberton8cc085d2012-12-03 09:24:15 -05001640 .remove = virtnet_remove,
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -08001641 .config_changed = virtnet_config_changed,
Amit Shah0741bcb2011-12-22 16:58:33 +05301642#ifdef CONFIG_PM
1643 .freeze = virtnet_freeze,
1644 .restore = virtnet_restore,
1645#endif
Rusty Russell296f96f2007-10-22 11:03:37 +10001646};
1647
1648static int __init init(void)
1649{
Uwe Kleine-König22402522009-11-05 01:32:44 -08001650 return register_virtio_driver(&virtio_net_driver);
Rusty Russell296f96f2007-10-22 11:03:37 +10001651}
1652
1653static void __exit fini(void)
1654{
Uwe Kleine-König22402522009-11-05 01:32:44 -08001655 unregister_virtio_driver(&virtio_net_driver);
Rusty Russell296f96f2007-10-22 11:03:37 +10001656}
1657module_init(init);
1658module_exit(fini);
1659
1660MODULE_DEVICE_TABLE(virtio, id_table);
1661MODULE_DESCRIPTION("Virtio network driver");
1662MODULE_LICENSE("GPL");