blob: 6fb783ce20b987fe6b74377a3fa96a0e0a8be18f [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 Russell34a48572008-02-04 23:50:02 -050033static int csum = 1, gso = 1;
34module_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
Alex Williamson2a41f712009-02-04 09:02:34 +000042
Rusty Russell296f96f2007-10-22 11:03:37 +100043struct virtnet_info
44{
45 struct virtio_device *vdev;
Alex Williamson2a41f712009-02-04 09:02:34 +000046 struct virtqueue *rvq, *svq, *cvq;
Rusty Russell296f96f2007-10-22 11:03:37 +100047 struct net_device *dev;
48 struct napi_struct napi;
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -080049 unsigned int status;
Rusty Russell296f96f2007-10-22 11:03:37 +100050
51 /* Number of input buffers, and max we've ever had. */
52 unsigned int num, max;
53
Herbert Xu97402b92008-04-18 11:24:27 +080054 /* I like... big packets and I cannot lie! */
55 bool big_packets;
56
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -080057 /* Host will merge rx buffers for big packets (shake it! shake it!) */
58 bool mergeable_rx_bufs;
59
Rusty Russell3161e452009-08-26 12:22:32 -070060 /* Work struct for refilling if we run low on memory. */
61 struct delayed_work refill;
62
Rusty Russellfb6813f2008-07-25 12:06:01 -050063 /* Chain pages by the private ptr. */
64 struct page *pages;
Rusty Russell296f96f2007-10-22 11:03:37 +100065};
66
Rusty Russellb3f24692009-09-24 09:59:19 -060067struct skb_vnet_hdr {
68 union {
69 struct virtio_net_hdr hdr;
70 struct virtio_net_hdr_mrg_rxbuf mhdr;
71 };
Rusty Russell48925e32009-09-24 09:59:20 -060072 unsigned int num_sg;
Rusty Russellb3f24692009-09-24 09:59:19 -060073};
74
Shirley Ma9ab86bb2010-01-29 03:20:04 +000075struct padded_vnet_hdr {
76 struct virtio_net_hdr hdr;
77 /*
78 * virtio_net_hdr should be in a separated sg buffer because of a
79 * QEMU bug, and data sg buffer shares same page with this header sg.
80 * This padding makes next sg 16 byte aligned after virtio_net_hdr.
81 */
82 char padding[6];
83};
84
Rusty Russellb3f24692009-09-24 09:59:19 -060085static inline struct skb_vnet_hdr *skb_vnet_hdr(struct sk_buff *skb)
Rusty Russell296f96f2007-10-22 11:03:37 +100086{
Rusty Russellb3f24692009-09-24 09:59:19 -060087 return (struct skb_vnet_hdr *)skb->cb;
Rusty Russell296f96f2007-10-22 11:03:37 +100088}
89
Shirley Ma9ab86bb2010-01-29 03:20:04 +000090/*
91 * private is used to chain pages for big packets, put the whole
92 * most recent used list in the beginning for reuse
93 */
94static void give_pages(struct virtnet_info *vi, struct page *page)
Rusty Russellfb6813f2008-07-25 12:06:01 -050095{
Shirley Ma9ab86bb2010-01-29 03:20:04 +000096 struct page *end;
97
98 /* Find end of list, sew whole thing into vi->pages. */
99 for (end = page; end->private; end = (struct page *)end->private);
100 end->private = (unsigned long)vi->pages;
Rusty Russellfb6813f2008-07-25 12:06:01 -0500101 vi->pages = page;
102}
103
104static struct page *get_a_page(struct virtnet_info *vi, gfp_t gfp_mask)
105{
106 struct page *p = vi->pages;
107
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000108 if (p) {
Rusty Russellfb6813f2008-07-25 12:06:01 -0500109 vi->pages = (struct page *)p->private;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000110 /* clear private here, it is used to chain pages */
111 p->private = 0;
112 } else
Rusty Russellfb6813f2008-07-25 12:06:01 -0500113 p = alloc_page(gfp_mask);
114 return p;
115}
116
Rusty Russell2cb9c6b2008-02-04 23:50:07 -0500117static void skb_xmit_done(struct virtqueue *svq)
Rusty Russell296f96f2007-10-22 11:03:37 +1000118{
Rusty Russell2cb9c6b2008-02-04 23:50:07 -0500119 struct virtnet_info *vi = svq->vdev->priv;
Rusty Russell296f96f2007-10-22 11:03:37 +1000120
Rusty Russell2cb9c6b2008-02-04 23:50:07 -0500121 /* Suppress further interrupts. */
122 svq->vq_ops->disable_cb(svq);
Rusty Russell11a3a152008-05-26 17:48:13 +1000123
Rusty Russell363f1512008-06-08 20:51:55 +1000124 /* We were probably waiting for more output buffers. */
Rusty Russell296f96f2007-10-22 11:03:37 +1000125 netif_wake_queue(vi->dev);
Rusty Russell296f96f2007-10-22 11:03:37 +1000126}
127
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000128static void set_skb_frag(struct sk_buff *skb, struct page *page,
129 unsigned int offset, unsigned int *len)
130{
131 int i = skb_shinfo(skb)->nr_frags;
132 skb_frag_t *f;
133
134 f = &skb_shinfo(skb)->frags[i];
135 f->size = min((unsigned)PAGE_SIZE - offset, *len);
136 f->page_offset = offset;
137 f->page = page;
138
139 skb->data_len += f->size;
140 skb->len += f->size;
141 skb_shinfo(skb)->nr_frags++;
142 *len -= f->size;
143}
144
145static struct sk_buff *page_to_skb(struct virtnet_info *vi,
146 struct page *page, unsigned int len)
147{
148 struct sk_buff *skb;
149 struct skb_vnet_hdr *hdr;
150 unsigned int copy, hdr_len, offset;
151 char *p;
152
153 p = page_address(page);
154
155 /* copy small packet so we can reuse these pages for small data */
156 skb = netdev_alloc_skb_ip_align(vi->dev, GOOD_COPY_LEN);
157 if (unlikely(!skb))
158 return NULL;
159
160 hdr = skb_vnet_hdr(skb);
161
162 if (vi->mergeable_rx_bufs) {
163 hdr_len = sizeof hdr->mhdr;
164 offset = hdr_len;
165 } else {
166 hdr_len = sizeof hdr->hdr;
167 offset = sizeof(struct padded_vnet_hdr);
168 }
169
170 memcpy(hdr, p, hdr_len);
171
172 len -= hdr_len;
173 p += offset;
174
175 copy = len;
176 if (copy > skb_tailroom(skb))
177 copy = skb_tailroom(skb);
178 memcpy(skb_put(skb, copy), p, copy);
179
180 len -= copy;
181 offset += copy;
182
183 while (len) {
184 set_skb_frag(skb, page, offset, &len);
185 page = (struct page *)page->private;
186 offset = 0;
187 }
188
189 if (page)
190 give_pages(vi, page);
191
192 return skb;
193}
194
195static int receive_mergeable(struct virtnet_info *vi, struct sk_buff *skb)
196{
197 struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
198 struct page *page;
199 int num_buf, i, len;
200
201 num_buf = hdr->mhdr.num_buffers;
202 while (--num_buf) {
203 i = skb_shinfo(skb)->nr_frags;
204 if (i >= MAX_SKB_FRAGS) {
205 pr_debug("%s: packet too long\n", skb->dev->name);
206 skb->dev->stats.rx_length_errors++;
207 return -EINVAL;
208 }
209
210 page = vi->rvq->vq_ops->get_buf(vi->rvq, &len);
211 if (!page) {
212 pr_debug("%s: rx error: %d buffers missing\n",
213 skb->dev->name, hdr->mhdr.num_buffers);
214 skb->dev->stats.rx_length_errors++;
215 return -EINVAL;
216 }
217 if (len > PAGE_SIZE)
218 len = PAGE_SIZE;
219
220 set_skb_frag(skb, page, 0, &len);
221
222 --vi->num;
223 }
224 return 0;
225}
226
227static void receive_buf(struct net_device *dev, void *buf, unsigned int len)
Rusty Russell296f96f2007-10-22 11:03:37 +1000228{
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800229 struct virtnet_info *vi = netdev_priv(dev);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000230 struct sk_buff *skb;
231 struct page *page;
232 struct skb_vnet_hdr *hdr;
Rusty Russell296f96f2007-10-22 11:03:37 +1000233
234 if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
235 pr_debug("%s: short packet %i\n", dev->name, len);
236 dev->stats.rx_length_errors++;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000237 if (vi->mergeable_rx_bufs || vi->big_packets)
238 give_pages(vi, buf);
239 else
240 dev_kfree_skb(buf);
241 return;
Rusty Russell296f96f2007-10-22 11:03:37 +1000242 }
Rusty Russell296f96f2007-10-22 11:03:37 +1000243
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000244 if (!vi->mergeable_rx_bufs && !vi->big_packets) {
245 skb = buf;
246 len -= sizeof(struct virtio_net_hdr);
247 skb_trim(skb, len);
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800248 } else {
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000249 page = buf;
250 skb = page_to_skb(vi, page, len);
251 if (unlikely(!skb)) {
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800252 dev->stats.rx_dropped++;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000253 give_pages(vi, page);
254 return;
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800255 }
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000256 if (vi->mergeable_rx_bufs)
257 if (receive_mergeable(vi, skb)) {
258 dev_kfree_skb(skb);
259 return;
260 }
Herbert Xu97402b92008-04-18 11:24:27 +0800261 }
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800262
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000263 hdr = skb_vnet_hdr(skb);
Herbert Xu97402b92008-04-18 11:24:27 +0800264 skb->truesize += skb->data_len;
Rusty Russell296f96f2007-10-22 11:03:37 +1000265 dev->stats.rx_bytes += skb->len;
266 dev->stats.rx_packets++;
267
Rusty Russellb3f24692009-09-24 09:59:19 -0600268 if (hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
Rusty Russell296f96f2007-10-22 11:03:37 +1000269 pr_debug("Needs csum!\n");
Rusty Russellb3f24692009-09-24 09:59:19 -0600270 if (!skb_partial_csum_set(skb,
271 hdr->hdr.csum_start,
272 hdr->hdr.csum_offset))
Rusty Russell296f96f2007-10-22 11:03:37 +1000273 goto frame_err;
Rusty Russell296f96f2007-10-22 11:03:37 +1000274 }
275
Mark McLoughlin23cde762008-06-08 20:49:00 +1000276 skb->protocol = eth_type_trans(skb, dev);
277 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
278 ntohs(skb->protocol), skb->len, skb->pkt_type);
279
Rusty Russellb3f24692009-09-24 09:59:19 -0600280 if (hdr->hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
Rusty Russell296f96f2007-10-22 11:03:37 +1000281 pr_debug("GSO!\n");
Rusty Russellb3f24692009-09-24 09:59:19 -0600282 switch (hdr->hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
Rusty Russell296f96f2007-10-22 11:03:37 +1000283 case VIRTIO_NET_HDR_GSO_TCPV4:
284 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
285 break;
Rusty Russell296f96f2007-10-22 11:03:37 +1000286 case VIRTIO_NET_HDR_GSO_UDP:
287 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
288 break;
289 case VIRTIO_NET_HDR_GSO_TCPV6:
290 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
291 break;
292 default:
293 if (net_ratelimit())
294 printk(KERN_WARNING "%s: bad gso type %u.\n",
Rusty Russellb3f24692009-09-24 09:59:19 -0600295 dev->name, hdr->hdr.gso_type);
Rusty Russell296f96f2007-10-22 11:03:37 +1000296 goto frame_err;
297 }
298
Rusty Russellb3f24692009-09-24 09:59:19 -0600299 if (hdr->hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
Rusty Russell34a48572008-02-04 23:50:02 -0500300 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
301
Rusty Russellb3f24692009-09-24 09:59:19 -0600302 skb_shinfo(skb)->gso_size = hdr->hdr.gso_size;
Rusty Russell296f96f2007-10-22 11:03:37 +1000303 if (skb_shinfo(skb)->gso_size == 0) {
304 if (net_ratelimit())
305 printk(KERN_WARNING "%s: zero gso size.\n",
306 dev->name);
307 goto frame_err;
308 }
309
310 /* Header must be checked, and gso_segs computed. */
311 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
312 skb_shinfo(skb)->gso_segs = 0;
313 }
314
315 netif_receive_skb(skb);
316 return;
317
318frame_err:
319 dev->stats.rx_frame_errors++;
Rusty Russell296f96f2007-10-22 11:03:37 +1000320 dev_kfree_skb(skb);
321}
322
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000323static int add_recvbuf_small(struct virtnet_info *vi, gfp_t gfp)
Rusty Russell296f96f2007-10-22 11:03:37 +1000324{
325 struct sk_buff *skb;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000326 struct skb_vnet_hdr *hdr;
327 struct scatterlist sg[2];
328 int err;
Rusty Russell296f96f2007-10-22 11:03:37 +1000329
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000330 skb = netdev_alloc_skb_ip_align(vi->dev, MAX_PACKET_LEN);
331 if (unlikely(!skb))
332 return -ENOMEM;
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800333
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000334 skb_put(skb, MAX_PACKET_LEN);
335
336 hdr = skb_vnet_hdr(skb);
337 sg_set_buf(sg, &hdr->hdr, sizeof hdr->hdr);
338
339 skb_to_sgvec(skb, sg + 1, 0, skb->len);
340
341 err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, 2, skb);
342 if (err < 0)
343 dev_kfree_skb(skb);
344
345 return err;
346}
347
348static int add_recvbuf_big(struct virtnet_info *vi, gfp_t gfp)
349{
350 struct scatterlist sg[MAX_SKB_FRAGS + 2];
351 struct page *first, *list = NULL;
352 char *p;
353 int i, err, offset;
354
355 /* page in sg[MAX_SKB_FRAGS + 1] is list tail */
356 for (i = MAX_SKB_FRAGS + 1; i > 1; --i) {
357 first = get_a_page(vi, gfp);
358 if (!first) {
359 if (list)
360 give_pages(vi, list);
361 return -ENOMEM;
Rusty Russell3161e452009-08-26 12:22:32 -0700362 }
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000363 sg_set_buf(&sg[i], page_address(first), PAGE_SIZE);
Rusty Russell296f96f2007-10-22 11:03:37 +1000364
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000365 /* chain new page in list head to match sg */
366 first->private = (unsigned long)list;
367 list = first;
368 }
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800369
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000370 first = get_a_page(vi, gfp);
371 if (!first) {
372 give_pages(vi, list);
373 return -ENOMEM;
374 }
375 p = page_address(first);
Herbert Xu97402b92008-04-18 11:24:27 +0800376
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000377 /* sg[0], sg[1] share the same page */
378 /* a separated sg[0] for virtio_net_hdr only during to QEMU bug*/
379 sg_set_buf(&sg[0], p, sizeof(struct virtio_net_hdr));
Herbert Xu97402b92008-04-18 11:24:27 +0800380
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000381 /* sg[1] for data packet, from offset */
382 offset = sizeof(struct padded_vnet_hdr);
383 sg_set_buf(&sg[1], p + offset, PAGE_SIZE - offset);
Herbert Xu97402b92008-04-18 11:24:27 +0800384
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000385 /* chain first in list head */
386 first->private = (unsigned long)list;
387 err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, MAX_SKB_FRAGS + 2,
388 first);
389 if (err < 0)
390 give_pages(vi, first);
Herbert Xu97402b92008-04-18 11:24:27 +0800391
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000392 return err;
393}
Herbert Xu97402b92008-04-18 11:24:27 +0800394
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000395static int add_recvbuf_mergeable(struct virtnet_info *vi, gfp_t gfp)
396{
397 struct page *page;
398 struct scatterlist sg;
399 int err;
Rusty Russell296f96f2007-10-22 11:03:37 +1000400
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000401 page = get_a_page(vi, gfp);
402 if (!page)
403 return -ENOMEM;
404
405 sg_init_one(&sg, page_address(page), PAGE_SIZE);
406
407 err = vi->rvq->vq_ops->add_buf(vi->rvq, &sg, 0, 1, page);
408 if (err < 0)
409 give_pages(vi, page);
410
411 return err;
Rusty Russell296f96f2007-10-22 11:03:37 +1000412}
413
Rusty Russell3161e452009-08-26 12:22:32 -0700414/* Returns false if we couldn't fill entirely (OOM). */
415static bool try_fill_recv(struct virtnet_info *vi, gfp_t gfp)
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800416{
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800417 int err;
Rusty Russell3161e452009-08-26 12:22:32 -0700418 bool oom = false;
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800419
Amit Shah0aea51c2009-08-26 14:58:28 +0530420 do {
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000421 if (vi->mergeable_rx_bufs)
422 err = add_recvbuf_mergeable(vi, gfp);
423 else if (vi->big_packets)
424 err = add_recvbuf_big(vi, gfp);
425 else
426 err = add_recvbuf_small(vi, gfp);
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800427
Rusty Russell3c1b27d2009-09-23 22:26:31 -0600428 if (err < 0) {
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000429 oom = true;
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800430 break;
431 }
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000432 ++vi->num;
Amit Shah0aea51c2009-08-26 14:58:28 +0530433 } while (err > 0);
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800434 if (unlikely(vi->num > vi->max))
435 vi->max = vi->num;
436 vi->rvq->vq_ops->kick(vi->rvq);
Rusty Russell3161e452009-08-26 12:22:32 -0700437 return !oom;
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800438}
439
Rusty Russell18445c42008-02-04 23:49:57 -0500440static void skb_recv_done(struct virtqueue *rvq)
Rusty Russell296f96f2007-10-22 11:03:37 +1000441{
442 struct virtnet_info *vi = rvq->vdev->priv;
Rusty Russell18445c42008-02-04 23:49:57 -0500443 /* Schedule NAPI, Suppress further interrupts if successful. */
Ben Hutchings288379f2009-01-19 16:43:59 -0800444 if (napi_schedule_prep(&vi->napi)) {
Rusty Russell18445c42008-02-04 23:49:57 -0500445 rvq->vq_ops->disable_cb(rvq);
Ben Hutchings288379f2009-01-19 16:43:59 -0800446 __napi_schedule(&vi->napi);
Rusty Russell18445c42008-02-04 23:49:57 -0500447 }
Rusty Russell296f96f2007-10-22 11:03:37 +1000448}
449
Rusty Russell3161e452009-08-26 12:22:32 -0700450static void refill_work(struct work_struct *work)
451{
452 struct virtnet_info *vi;
453 bool still_empty;
454
455 vi = container_of(work, struct virtnet_info, refill.work);
456 napi_disable(&vi->napi);
Herbert Xu39d32152010-01-25 15:51:01 -0800457 still_empty = !try_fill_recv(vi, GFP_KERNEL);
Rusty Russell3161e452009-08-26 12:22:32 -0700458 napi_enable(&vi->napi);
459
460 /* In theory, this can happen: if we don't get any buffers in
461 * we will *never* try to fill again. */
462 if (still_empty)
463 schedule_delayed_work(&vi->refill, HZ/2);
464}
465
Rusty Russell296f96f2007-10-22 11:03:37 +1000466static int virtnet_poll(struct napi_struct *napi, int budget)
467{
468 struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000469 void *buf;
Rusty Russell296f96f2007-10-22 11:03:37 +1000470 unsigned int len, received = 0;
471
472again:
473 while (received < budget &&
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000474 (buf = vi->rvq->vq_ops->get_buf(vi->rvq, &len)) != NULL) {
475 receive_buf(vi->dev, buf, len);
476 --vi->num;
Rusty Russell296f96f2007-10-22 11:03:37 +1000477 received++;
478 }
479
Rusty Russell3161e452009-08-26 12:22:32 -0700480 if (vi->num < vi->max / 2) {
481 if (!try_fill_recv(vi, GFP_ATOMIC))
482 schedule_delayed_work(&vi->refill, 0);
483 }
Rusty Russell296f96f2007-10-22 11:03:37 +1000484
Rusty Russell8329d982007-11-19 11:20:43 -0500485 /* Out of packets? */
486 if (received < budget) {
Ben Hutchings288379f2009-01-19 16:43:59 -0800487 napi_complete(napi);
Joe Perches8e95a202009-12-03 07:58:21 +0000488 if (unlikely(!vi->rvq->vq_ops->enable_cb(vi->rvq)) &&
489 napi_schedule_prep(napi)) {
Christian Borntraeger4265f162008-03-14 14:17:05 +0100490 vi->rvq->vq_ops->disable_cb(vi->rvq);
Ben Hutchings288379f2009-01-19 16:43:59 -0800491 __napi_schedule(napi);
Rusty Russell296f96f2007-10-22 11:03:37 +1000492 goto again;
Christian Borntraeger4265f162008-03-14 14:17:05 +0100493 }
Rusty Russell296f96f2007-10-22 11:03:37 +1000494 }
495
496 return received;
497}
498
Rusty Russell48925e32009-09-24 09:59:20 -0600499static unsigned int free_old_xmit_skbs(struct virtnet_info *vi)
Rusty Russell296f96f2007-10-22 11:03:37 +1000500{
501 struct sk_buff *skb;
Rusty Russell48925e32009-09-24 09:59:20 -0600502 unsigned int len, tot_sgs = 0;
Rusty Russell296f96f2007-10-22 11:03:37 +1000503
504 while ((skb = vi->svq->vq_ops->get_buf(vi->svq, &len)) != NULL) {
505 pr_debug("Sent skb %p\n", skb);
Rusty Russell655aa312008-05-02 21:50:43 -0500506 vi->dev->stats.tx_bytes += skb->len;
Rusty Russell296f96f2007-10-22 11:03:37 +1000507 vi->dev->stats.tx_packets++;
Rusty Russell48925e32009-09-24 09:59:20 -0600508 tot_sgs += skb_vnet_hdr(skb)->num_sg;
Eric Dumazeted79bab2009-10-14 14:36:43 +0000509 dev_kfree_skb_any(skb);
Rusty Russell296f96f2007-10-22 11:03:37 +1000510 }
Rusty Russell48925e32009-09-24 09:59:20 -0600511 return tot_sgs;
Rusty Russell296f96f2007-10-22 11:03:37 +1000512}
513
Rusty Russell99ffc692008-05-02 21:50:46 -0500514static int xmit_skb(struct virtnet_info *vi, struct sk_buff *skb)
Rusty Russell296f96f2007-10-22 11:03:37 +1000515{
Rusty Russell05271682008-05-02 21:50:45 -0500516 struct scatterlist sg[2+MAX_SKB_FRAGS];
Rusty Russellb3f24692009-09-24 09:59:19 -0600517 struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
Rusty Russell296f96f2007-10-22 11:03:37 +1000518 const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
Rusty Russell296f96f2007-10-22 11:03:37 +1000519
Rusty Russell05271682008-05-02 21:50:45 -0500520 sg_init_table(sg, 2+MAX_SKB_FRAGS);
Rusty Russell4d125de2007-11-07 16:34:49 +1100521
Johannes Berge1749612008-10-27 15:59:26 -0700522 pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
Rusty Russell296f96f2007-10-22 11:03:37 +1000523
Rusty Russell296f96f2007-10-22 11:03:37 +1000524 if (skb->ip_summed == CHECKSUM_PARTIAL) {
Rusty Russellb3f24692009-09-24 09:59:19 -0600525 hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
526 hdr->hdr.csum_start = skb->csum_start - skb_headroom(skb);
527 hdr->hdr.csum_offset = skb->csum_offset;
Rusty Russell296f96f2007-10-22 11:03:37 +1000528 } else {
Rusty Russellb3f24692009-09-24 09:59:19 -0600529 hdr->hdr.flags = 0;
530 hdr->hdr.csum_offset = hdr->hdr.csum_start = 0;
Rusty Russell296f96f2007-10-22 11:03:37 +1000531 }
532
533 if (skb_is_gso(skb)) {
Rusty Russellb3f24692009-09-24 09:59:19 -0600534 hdr->hdr.hdr_len = skb_headlen(skb);
535 hdr->hdr.gso_size = skb_shinfo(skb)->gso_size;
Rusty Russell34a48572008-02-04 23:50:02 -0500536 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
Rusty Russellb3f24692009-09-24 09:59:19 -0600537 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
Rusty Russell296f96f2007-10-22 11:03:37 +1000538 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
Rusty Russellb3f24692009-09-24 09:59:19 -0600539 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
Rusty Russell296f96f2007-10-22 11:03:37 +1000540 else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
Rusty Russellb3f24692009-09-24 09:59:19 -0600541 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
Rusty Russell296f96f2007-10-22 11:03:37 +1000542 else
543 BUG();
Rusty Russell34a48572008-02-04 23:50:02 -0500544 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
Rusty Russellb3f24692009-09-24 09:59:19 -0600545 hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
Rusty Russell296f96f2007-10-22 11:03:37 +1000546 } else {
Rusty Russellb3f24692009-09-24 09:59:19 -0600547 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
548 hdr->hdr.gso_size = hdr->hdr.hdr_len = 0;
Rusty Russell296f96f2007-10-22 11:03:37 +1000549 }
550
Rusty Russellb3f24692009-09-24 09:59:19 -0600551 hdr->mhdr.num_buffers = 0;
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800552
553 /* Encode metadata header at front. */
554 if (vi->mergeable_rx_bufs)
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000555 sg_set_buf(sg, &hdr->mhdr, sizeof hdr->mhdr);
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800556 else
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000557 sg_set_buf(sg, &hdr->hdr, sizeof hdr->hdr);
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800558
Rusty Russell48925e32009-09-24 09:59:20 -0600559 hdr->num_sg = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
560 return vi->svq->vq_ops->add_buf(vi->svq, sg, hdr->num_sg, 0, skb);
Rusty Russell11a3a152008-05-26 17:48:13 +1000561}
562
Stephen Hemminger424efe92009-08-31 19:50:51 +0000563static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
Rusty Russell99ffc692008-05-02 21:50:46 -0500564{
565 struct virtnet_info *vi = netdev_priv(dev);
Rusty Russell48925e32009-09-24 09:59:20 -0600566 int capacity;
Rusty Russell2cb9c6b2008-02-04 23:50:07 -0500567
568again:
569 /* Free up any pending old buffers before queueing new ones. */
570 free_old_xmit_skbs(vi);
Rusty Russell2cb9c6b2008-02-04 23:50:07 -0500571
Michael S. Tsirkin03f191b2009-10-28 04:03:38 -0700572 /* Try to transmit */
Rusty Russell48925e32009-09-24 09:59:20 -0600573 capacity = xmit_skb(vi, skb);
574
575 /* This can happen with OOM and indirect buffers. */
576 if (unlikely(capacity < 0)) {
577 netif_stop_queue(dev);
578 dev_warn(&dev->dev, "Unexpected full queue\n");
579 if (unlikely(!vi->svq->vq_ops->enable_cb(vi->svq))) {
580 vi->svq->vq_ops->disable_cb(vi->svq);
581 netif_start_queue(dev);
582 goto again;
583 }
584 return NETDEV_TX_BUSY;
Rusty Russell99ffc692008-05-02 21:50:46 -0500585 }
Rusty Russell48925e32009-09-24 09:59:20 -0600586 vi->svq->vq_ops->kick(vi->svq);
Michael S. Tsirkin03f191b2009-10-28 04:03:38 -0700587
Rusty Russell48925e32009-09-24 09:59:20 -0600588 /* Don't wait up for transmitted skbs to be freed. */
589 skb_orphan(skb);
590 nf_reset(skb);
Rusty Russell99ffc692008-05-02 21:50:46 -0500591
Rusty Russell48925e32009-09-24 09:59:20 -0600592 /* Apparently nice girls don't return TX_BUSY; stop the queue
593 * before it gets out of hand. Naturally, this wastes entries. */
594 if (capacity < 2+MAX_SKB_FRAGS) {
595 netif_stop_queue(dev);
596 if (unlikely(!vi->svq->vq_ops->enable_cb(vi->svq))) {
597 /* More just got used, free them then recheck. */
598 capacity += free_old_xmit_skbs(vi);
599 if (capacity >= 2+MAX_SKB_FRAGS) {
600 netif_start_queue(dev);
601 vi->svq->vq_ops->disable_cb(vi->svq);
602 }
603 }
Rusty Russell99ffc692008-05-02 21:50:46 -0500604 }
Rusty Russell48925e32009-09-24 09:59:20 -0600605
606 return NETDEV_TX_OK;
Rusty Russell296f96f2007-10-22 11:03:37 +1000607}
608
Alex Williamson9c46f6d2009-02-04 16:36:34 -0800609static int virtnet_set_mac_address(struct net_device *dev, void *p)
610{
611 struct virtnet_info *vi = netdev_priv(dev);
612 struct virtio_device *vdev = vi->vdev;
613 int ret;
614
615 ret = eth_mac_addr(dev, p);
616 if (ret)
617 return ret;
618
Alex Williamson62994b22009-04-04 16:40:19 -0700619 if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
620 vdev->config->set(vdev, offsetof(struct virtio_net_config, mac),
621 dev->dev_addr, dev->addr_len);
Alex Williamson9c46f6d2009-02-04 16:36:34 -0800622
623 return 0;
624}
625
Amit Shahda74e892008-02-29 16:24:50 +0530626#ifdef CONFIG_NET_POLL_CONTROLLER
627static void virtnet_netpoll(struct net_device *dev)
628{
629 struct virtnet_info *vi = netdev_priv(dev);
630
631 napi_schedule(&vi->napi);
632}
633#endif
634
Rusty Russell296f96f2007-10-22 11:03:37 +1000635static int virtnet_open(struct net_device *dev)
636{
637 struct virtnet_info *vi = netdev_priv(dev);
638
Rusty Russell296f96f2007-10-22 11:03:37 +1000639 napi_enable(&vi->napi);
Rusty Russella48bd8f2008-02-04 23:50:07 -0500640
641 /* If all buffers were filled by other side before we napi_enabled, we
642 * won't get another interrupt, so process any outstanding packets
Christian Borntraeger370076d2008-02-06 08:50:11 +0100643 * now. virtnet_poll wants re-enable the queue, so we disable here.
644 * We synchronize against interrupts via NAPI_STATE_SCHED */
Ben Hutchings288379f2009-01-19 16:43:59 -0800645 if (napi_schedule_prep(&vi->napi)) {
Christian Borntraeger370076d2008-02-06 08:50:11 +0100646 vi->rvq->vq_ops->disable_cb(vi->rvq);
Ben Hutchings288379f2009-01-19 16:43:59 -0800647 __napi_schedule(&vi->napi);
Christian Borntraeger370076d2008-02-06 08:50:11 +0100648 }
Rusty Russell296f96f2007-10-22 11:03:37 +1000649 return 0;
650}
651
Alex Williamson2a41f712009-02-04 09:02:34 +0000652/*
653 * Send command via the control virtqueue and check status. Commands
654 * supported by the hypervisor, as indicated by feature bits, should
655 * never fail unless improperly formated.
656 */
657static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
658 struct scatterlist *data, int out, int in)
659{
Alex Williamson23e258e2009-05-01 17:27:56 +0000660 struct scatterlist *s, sg[VIRTNET_SEND_COMMAND_SG_MAX + 2];
Alex Williamson2a41f712009-02-04 09:02:34 +0000661 struct virtio_net_ctrl_hdr ctrl;
662 virtio_net_ctrl_ack status = ~0;
663 unsigned int tmp;
Alex Williamson23e258e2009-05-01 17:27:56 +0000664 int i;
Alex Williamson2a41f712009-02-04 09:02:34 +0000665
Alexander Beregalov0ee904c2009-04-11 14:50:23 +0000666 /* Caller should know better */
667 BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ||
668 (out + in > VIRTNET_SEND_COMMAND_SG_MAX));
Alex Williamson2a41f712009-02-04 09:02:34 +0000669
670 out++; /* Add header */
671 in++; /* Add return status */
672
673 ctrl.class = class;
674 ctrl.cmd = cmd;
675
676 sg_init_table(sg, out + in);
677
678 sg_set_buf(&sg[0], &ctrl, sizeof(ctrl));
Alex Williamson23e258e2009-05-01 17:27:56 +0000679 for_each_sg(data, s, out + in - 2, i)
680 sg_set_buf(&sg[i + 1], sg_virt(s), s->length);
Alex Williamson2a41f712009-02-04 09:02:34 +0000681 sg_set_buf(&sg[out + in - 1], &status, sizeof(status));
682
Rusty Russell3c1b27d2009-09-23 22:26:31 -0600683 BUG_ON(vi->cvq->vq_ops->add_buf(vi->cvq, sg, out, in, vi) < 0);
Alex Williamson2a41f712009-02-04 09:02:34 +0000684
685 vi->cvq->vq_ops->kick(vi->cvq);
686
687 /*
688 * Spin for a response, the kick causes an ioport write, trapping
689 * into the hypervisor, so the request should be handled immediately.
690 */
691 while (!vi->cvq->vq_ops->get_buf(vi->cvq, &tmp))
692 cpu_relax();
693
694 return status == VIRTIO_NET_OK;
695}
696
Rusty Russell296f96f2007-10-22 11:03:37 +1000697static int virtnet_close(struct net_device *dev)
698{
699 struct virtnet_info *vi = netdev_priv(dev);
Rusty Russell296f96f2007-10-22 11:03:37 +1000700
701 napi_disable(&vi->napi);
702
Rusty Russell296f96f2007-10-22 11:03:37 +1000703 return 0;
704}
705
Herbert Xua9ea3fc2008-04-18 11:21:42 +0800706static int virtnet_set_tx_csum(struct net_device *dev, u32 data)
707{
708 struct virtnet_info *vi = netdev_priv(dev);
709 struct virtio_device *vdev = vi->vdev;
710
711 if (data && !virtio_has_feature(vdev, VIRTIO_NET_F_CSUM))
712 return -ENOSYS;
713
714 return ethtool_op_set_tx_hw_csum(dev, data);
715}
716
Alex Williamson2af76982009-02-04 09:02:40 +0000717static void virtnet_set_rx_mode(struct net_device *dev)
718{
719 struct virtnet_info *vi = netdev_priv(dev);
Alex Williamsonf565a7c2009-02-04 09:02:45 +0000720 struct scatterlist sg[2];
Alex Williamson2af76982009-02-04 09:02:40 +0000721 u8 promisc, allmulti;
Alex Williamsonf565a7c2009-02-04 09:02:45 +0000722 struct virtio_net_ctrl_mac *mac_data;
723 struct dev_addr_list *addr;
Jiri Pirkoccffad252009-05-22 23:22:17 +0000724 struct netdev_hw_addr *ha;
Jiri Pirko32e7bfc2010-01-25 13:36:10 -0800725 int uc_count;
Jiri Pirko4cd24ea2010-02-08 04:30:35 +0000726 int mc_count;
Alex Williamsonf565a7c2009-02-04 09:02:45 +0000727 void *buf;
728 int i;
Alex Williamson2af76982009-02-04 09:02:40 +0000729
730 /* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */
731 if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
732 return;
733
Alex Williamsonf565a7c2009-02-04 09:02:45 +0000734 promisc = ((dev->flags & IFF_PROMISC) != 0);
735 allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
Alex Williamson2af76982009-02-04 09:02:40 +0000736
Alex Williamson23e258e2009-05-01 17:27:56 +0000737 sg_init_one(sg, &promisc, sizeof(promisc));
Alex Williamson2af76982009-02-04 09:02:40 +0000738
739 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
740 VIRTIO_NET_CTRL_RX_PROMISC,
Alex Williamsonf565a7c2009-02-04 09:02:45 +0000741 sg, 1, 0))
Alex Williamson2af76982009-02-04 09:02:40 +0000742 dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
743 promisc ? "en" : "dis");
744
Alex Williamson23e258e2009-05-01 17:27:56 +0000745 sg_init_one(sg, &allmulti, sizeof(allmulti));
Alex Williamson2af76982009-02-04 09:02:40 +0000746
747 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
748 VIRTIO_NET_CTRL_RX_ALLMULTI,
Alex Williamsonf565a7c2009-02-04 09:02:45 +0000749 sg, 1, 0))
Alex Williamson2af76982009-02-04 09:02:40 +0000750 dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
751 allmulti ? "en" : "dis");
Alex Williamsonf565a7c2009-02-04 09:02:45 +0000752
Jiri Pirko32e7bfc2010-01-25 13:36:10 -0800753 uc_count = netdev_uc_count(dev);
Jiri Pirko4cd24ea2010-02-08 04:30:35 +0000754 mc_count = netdev_mc_count(dev);
Alex Williamsonf565a7c2009-02-04 09:02:45 +0000755 /* MAC filter - use one buffer for both lists */
Jiri Pirko4cd24ea2010-02-08 04:30:35 +0000756 buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
757 (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
758 mac_data = buf;
Alex Williamsonf565a7c2009-02-04 09:02:45 +0000759 if (!buf) {
760 dev_warn(&dev->dev, "No memory for MAC address buffer\n");
761 return;
762 }
763
Alex Williamson23e258e2009-05-01 17:27:56 +0000764 sg_init_table(sg, 2);
765
Alex Williamsonf565a7c2009-02-04 09:02:45 +0000766 /* Store the unicast list and count in the front of the buffer */
Jiri Pirko32e7bfc2010-01-25 13:36:10 -0800767 mac_data->entries = uc_count;
Jiri Pirkoccffad252009-05-22 23:22:17 +0000768 i = 0;
Jiri Pirko32e7bfc2010-01-25 13:36:10 -0800769 netdev_for_each_uc_addr(ha, dev)
Jiri Pirkoccffad252009-05-22 23:22:17 +0000770 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
Alex Williamsonf565a7c2009-02-04 09:02:45 +0000771
772 sg_set_buf(&sg[0], mac_data,
Jiri Pirko32e7bfc2010-01-25 13:36:10 -0800773 sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
Alex Williamsonf565a7c2009-02-04 09:02:45 +0000774
775 /* multicast list and count fill the end */
Jiri Pirko32e7bfc2010-01-25 13:36:10 -0800776 mac_data = (void *)&mac_data->macs[uc_count][0];
Alex Williamsonf565a7c2009-02-04 09:02:45 +0000777
Jiri Pirko4cd24ea2010-02-08 04:30:35 +0000778 mac_data->entries = mc_count;
Jiri Pirko567ec872010-02-23 23:17:07 +0000779 i = 0;
780 netdev_for_each_mc_addr(addr, dev)
781 memcpy(&mac_data->macs[i++][0], addr->da_addr, ETH_ALEN);
Alex Williamsonf565a7c2009-02-04 09:02:45 +0000782
783 sg_set_buf(&sg[1], mac_data,
Jiri Pirko4cd24ea2010-02-08 04:30:35 +0000784 sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
Alex Williamsonf565a7c2009-02-04 09:02:45 +0000785
786 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
787 VIRTIO_NET_CTRL_MAC_TABLE_SET,
788 sg, 2, 0))
789 dev_warn(&dev->dev, "Failed to set MAC fitler table.\n");
790
791 kfree(buf);
Alex Williamson2af76982009-02-04 09:02:40 +0000792}
793
Alex Williamson1824a982009-05-01 17:31:10 +0000794static void virtnet_vlan_rx_add_vid(struct net_device *dev, u16 vid)
Alex Williamson0bde95692009-02-04 09:02:50 +0000795{
796 struct virtnet_info *vi = netdev_priv(dev);
797 struct scatterlist sg;
798
Alex Williamson23e258e2009-05-01 17:27:56 +0000799 sg_init_one(&sg, &vid, sizeof(vid));
Alex Williamson0bde95692009-02-04 09:02:50 +0000800
801 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
802 VIRTIO_NET_CTRL_VLAN_ADD, &sg, 1, 0))
803 dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
804}
805
Alex Williamson1824a982009-05-01 17:31:10 +0000806static void virtnet_vlan_rx_kill_vid(struct net_device *dev, u16 vid)
Alex Williamson0bde95692009-02-04 09:02:50 +0000807{
808 struct virtnet_info *vi = netdev_priv(dev);
809 struct scatterlist sg;
810
Alex Williamson23e258e2009-05-01 17:27:56 +0000811 sg_init_one(&sg, &vid, sizeof(vid));
Alex Williamson0bde95692009-02-04 09:02:50 +0000812
813 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
814 VIRTIO_NET_CTRL_VLAN_DEL, &sg, 1, 0))
815 dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
816}
817
Stephen Hemminger0fc0b732009-09-02 01:03:33 -0700818static const struct ethtool_ops virtnet_ethtool_ops = {
Herbert Xua9ea3fc2008-04-18 11:21:42 +0800819 .set_tx_csum = virtnet_set_tx_csum,
820 .set_sg = ethtool_op_set_sg,
Mark McLoughlin0276b492008-11-16 22:40:36 -0800821 .set_tso = ethtool_op_set_tso,
Sridhar Samudrala5c516752009-07-14 14:21:02 +0000822 .set_ufo = ethtool_op_set_ufo,
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -0800823 .get_link = ethtool_op_get_link,
Herbert Xua9ea3fc2008-04-18 11:21:42 +0800824};
825
Mark McLoughlin39da5812008-11-26 13:58:11 +0000826#define MIN_MTU 68
827#define MAX_MTU 65535
828
829static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
830{
831 if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
832 return -EINVAL;
833 dev->mtu = new_mtu;
834 return 0;
835}
836
Stephen Hemminger76288b42009-01-06 10:44:22 -0800837static const struct net_device_ops virtnet_netdev = {
838 .ndo_open = virtnet_open,
839 .ndo_stop = virtnet_close,
840 .ndo_start_xmit = start_xmit,
841 .ndo_validate_addr = eth_validate_addr,
Alex Williamson9c46f6d2009-02-04 16:36:34 -0800842 .ndo_set_mac_address = virtnet_set_mac_address,
Alex Williamson2af76982009-02-04 09:02:40 +0000843 .ndo_set_rx_mode = virtnet_set_rx_mode,
Stephen Hemminger76288b42009-01-06 10:44:22 -0800844 .ndo_change_mtu = virtnet_change_mtu,
Alex Williamson1824a982009-05-01 17:31:10 +0000845 .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
846 .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
Stephen Hemminger76288b42009-01-06 10:44:22 -0800847#ifdef CONFIG_NET_POLL_CONTROLLER
848 .ndo_poll_controller = virtnet_netpoll,
849#endif
850};
851
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -0800852static void virtnet_update_status(struct virtnet_info *vi)
853{
854 u16 v;
855
856 if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS))
857 return;
858
859 vi->vdev->config->get(vi->vdev,
860 offsetof(struct virtio_net_config, status),
861 &v, sizeof(v));
862
863 /* Ignore unknown (future) status bits */
864 v &= VIRTIO_NET_S_LINK_UP;
865
866 if (vi->status == v)
867 return;
868
869 vi->status = v;
870
871 if (vi->status & VIRTIO_NET_S_LINK_UP) {
872 netif_carrier_on(vi->dev);
873 netif_wake_queue(vi->dev);
874 } else {
875 netif_carrier_off(vi->dev);
876 netif_stop_queue(vi->dev);
877 }
878}
879
880static void virtnet_config_changed(struct virtio_device *vdev)
881{
882 struct virtnet_info *vi = vdev->priv;
883
884 virtnet_update_status(vi);
885}
886
Rusty Russell296f96f2007-10-22 11:03:37 +1000887static int virtnet_probe(struct virtio_device *vdev)
888{
889 int err;
Rusty Russell296f96f2007-10-22 11:03:37 +1000890 struct net_device *dev;
891 struct virtnet_info *vi;
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -0600892 struct virtqueue *vqs[3];
893 vq_callback_t *callbacks[] = { skb_recv_done, skb_xmit_done, NULL};
894 const char *names[] = { "input", "output", "control" };
895 int nvqs;
Rusty Russell296f96f2007-10-22 11:03:37 +1000896
897 /* Allocate ourselves a network device with room for our info */
898 dev = alloc_etherdev(sizeof(struct virtnet_info));
899 if (!dev)
900 return -ENOMEM;
901
902 /* Set up network device as normal. */
Stephen Hemminger76288b42009-01-06 10:44:22 -0800903 dev->netdev_ops = &virtnet_netdev;
Rusty Russell296f96f2007-10-22 11:03:37 +1000904 dev->features = NETIF_F_HIGHDMA;
Herbert Xua9ea3fc2008-04-18 11:21:42 +0800905 SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
Rusty Russell296f96f2007-10-22 11:03:37 +1000906 SET_NETDEV_DEV(dev, &vdev->dev);
907
908 /* Do we support "hardware" checksums? */
Rusty Russellc45a6812008-05-02 21:50:50 -0500909 if (csum && virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
Rusty Russell296f96f2007-10-22 11:03:37 +1000910 /* This opens up the world of extra features. */
911 dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
Rusty Russellc45a6812008-05-02 21:50:50 -0500912 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
Rusty Russell34a48572008-02-04 23:50:02 -0500913 dev->features |= NETIF_F_TSO | NETIF_F_UFO
914 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
915 }
Rusty Russell5539ae92008-05-02 21:50:46 -0500916 /* Individual feature bits: what can host handle? */
Rusty Russellc45a6812008-05-02 21:50:50 -0500917 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
Rusty Russell5539ae92008-05-02 21:50:46 -0500918 dev->features |= NETIF_F_TSO;
Rusty Russellc45a6812008-05-02 21:50:50 -0500919 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
Rusty Russell5539ae92008-05-02 21:50:46 -0500920 dev->features |= NETIF_F_TSO6;
Rusty Russellc45a6812008-05-02 21:50:50 -0500921 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
Rusty Russell5539ae92008-05-02 21:50:46 -0500922 dev->features |= NETIF_F_TSO_ECN;
Rusty Russellc45a6812008-05-02 21:50:50 -0500923 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
Rusty Russell5539ae92008-05-02 21:50:46 -0500924 dev->features |= NETIF_F_UFO;
Rusty Russell296f96f2007-10-22 11:03:37 +1000925 }
926
927 /* Configuration may specify what MAC to use. Otherwise random. */
Rusty Russellc45a6812008-05-02 21:50:50 -0500928 if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
Rusty Russella586d4f2008-02-04 23:49:56 -0500929 vdev->config->get(vdev,
930 offsetof(struct virtio_net_config, mac),
931 dev->dev_addr, dev->addr_len);
Alex Williamson62994b22009-04-04 16:40:19 -0700932 } else
Rusty Russell296f96f2007-10-22 11:03:37 +1000933 random_ether_addr(dev->dev_addr);
934
935 /* Set up our device-specific information */
936 vi = netdev_priv(dev);
Dor Laor6c0cd7c2007-12-16 15:19:43 +0200937 netif_napi_add(dev, &vi->napi, virtnet_poll, napi_weight);
Rusty Russell296f96f2007-10-22 11:03:37 +1000938 vi->dev = dev;
939 vi->vdev = vdev;
Christian Borntraegerd9d5dcc2008-02-18 10:02:51 +0100940 vdev->priv = vi;
Rusty Russellfb6813f2008-07-25 12:06:01 -0500941 vi->pages = NULL;
Rusty Russell3161e452009-08-26 12:22:32 -0700942 INIT_DELAYED_WORK(&vi->refill, refill_work);
Rusty Russell296f96f2007-10-22 11:03:37 +1000943
Herbert Xu97402b92008-04-18 11:24:27 +0800944 /* If we can receive ANY GSO packets, we must allocate large ones. */
Joe Perches8e95a202009-12-03 07:58:21 +0000945 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
946 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
947 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
Herbert Xu97402b92008-04-18 11:24:27 +0800948 vi->big_packets = true;
949
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800950 if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
951 vi->mergeable_rx_bufs = true;
952
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -0600953 /* We expect two virtqueues, receive then send,
954 * and optionally control. */
955 nvqs = virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ? 3 : 2;
Rusty Russell296f96f2007-10-22 11:03:37 +1000956
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -0600957 err = vdev->config->find_vqs(vdev, nvqs, vqs, callbacks, names);
958 if (err)
959 goto free;
960
961 vi->rvq = vqs[0];
962 vi->svq = vqs[1];
Rusty Russell296f96f2007-10-22 11:03:37 +1000963
Alex Williamson2a41f712009-02-04 09:02:34 +0000964 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)) {
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -0600965 vi->cvq = vqs[2];
Alex Williamson0bde95692009-02-04 09:02:50 +0000966
967 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
968 dev->features |= NETIF_F_HW_VLAN_FILTER;
Alex Williamson2a41f712009-02-04 09:02:34 +0000969 }
970
Rusty Russell296f96f2007-10-22 11:03:37 +1000971 err = register_netdev(dev);
972 if (err) {
973 pr_debug("virtio_net: registering device failed\n");
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -0600974 goto free_vqs;
Rusty Russell296f96f2007-10-22 11:03:37 +1000975 }
Rusty Russellb3369c12008-02-04 23:50:02 -0500976
977 /* Last of all, set up some receive buffers. */
Rusty Russell3161e452009-08-26 12:22:32 -0700978 try_fill_recv(vi, GFP_KERNEL);
Rusty Russellb3369c12008-02-04 23:50:02 -0500979
980 /* If we didn't even get one input buffer, we're useless. */
981 if (vi->num == 0) {
982 err = -ENOMEM;
983 goto unregister;
984 }
985
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -0800986 vi->status = VIRTIO_NET_S_LINK_UP;
987 virtnet_update_status(vi);
Pantelis Koukousoulas47832562009-03-18 18:40:02 -0700988 netif_carrier_on(dev);
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -0800989
Rusty Russell296f96f2007-10-22 11:03:37 +1000990 pr_debug("virtnet: registered device %s\n", dev->name);
Rusty Russell296f96f2007-10-22 11:03:37 +1000991 return 0;
992
Rusty Russellb3369c12008-02-04 23:50:02 -0500993unregister:
994 unregister_netdev(dev);
Rusty Russell3161e452009-08-26 12:22:32 -0700995 cancel_delayed_work_sync(&vi->refill);
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -0600996free_vqs:
997 vdev->config->del_vqs(vdev);
Rusty Russell296f96f2007-10-22 11:03:37 +1000998free:
999 free_netdev(dev);
1000 return err;
1001}
1002
Shirley Ma9ab86bb2010-01-29 03:20:04 +00001003static void free_unused_bufs(struct virtnet_info *vi)
1004{
1005 void *buf;
1006 while (1) {
Shirley Ma830a8a92010-02-08 14:14:42 +00001007 buf = vi->svq->vq_ops->detach_unused_buf(vi->svq);
1008 if (!buf)
1009 break;
1010 dev_kfree_skb(buf);
1011 }
1012 while (1) {
Shirley Ma9ab86bb2010-01-29 03:20:04 +00001013 buf = vi->rvq->vq_ops->detach_unused_buf(vi->rvq);
1014 if (!buf)
1015 break;
1016 if (vi->mergeable_rx_bufs || vi->big_packets)
1017 give_pages(vi, buf);
1018 else
1019 dev_kfree_skb(buf);
1020 --vi->num;
1021 }
1022 BUG_ON(vi->num != 0);
1023}
1024
Uwe Kleine-König3d1285b2009-09-30 22:28:34 +00001025static void __devexit virtnet_remove(struct virtio_device *vdev)
Rusty Russell296f96f2007-10-22 11:03:37 +10001026{
Rusty Russell74b25532007-11-19 11:20:42 -05001027 struct virtnet_info *vi = vdev->priv;
Rusty Russellb3369c12008-02-04 23:50:02 -05001028
Rusty Russell6e5aa7e2008-02-04 23:50:03 -05001029 /* Stop all the virtqueues. */
1030 vdev->config->reset(vdev);
1031
Rusty Russellb3369c12008-02-04 23:50:02 -05001032
Rusty Russell74b25532007-11-19 11:20:42 -05001033 unregister_netdev(vi->dev);
Rusty Russell3161e452009-08-26 12:22:32 -07001034 cancel_delayed_work_sync(&vi->refill);
Shirley Ma830a8a92010-02-08 14:14:42 +00001035
1036 /* Free unused buffers in both send and recv, if any. */
Shirley Ma9ab86bb2010-01-29 03:20:04 +00001037 free_unused_bufs(vi);
Rusty Russellfb6813f2008-07-25 12:06:01 -05001038
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -06001039 vdev->config->del_vqs(vi->vdev);
1040
Rusty Russellfb6813f2008-07-25 12:06:01 -05001041 while (vi->pages)
1042 __free_pages(get_a_page(vi, GFP_KERNEL), 0);
1043
Rusty Russell74b25532007-11-19 11:20:42 -05001044 free_netdev(vi->dev);
Rusty Russell296f96f2007-10-22 11:03:37 +10001045}
1046
1047static struct virtio_device_id id_table[] = {
1048 { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
1049 { 0 },
1050};
1051
Rusty Russellc45a6812008-05-02 21:50:50 -05001052static unsigned int features[] = {
Mark McLoughlin5e4fe5c2008-07-08 17:10:42 +10001053 VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
1054 VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
Rusty Russellc45a6812008-05-02 21:50:50 -05001055 VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
Herbert Xu97402b92008-04-18 11:24:27 +08001056 VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
Sridhar Samudrala5c516752009-07-14 14:21:02 +00001057 VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO,
Alex Williamson2a41f712009-02-04 09:02:34 +00001058 VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ,
Alex Williamson0bde95692009-02-04 09:02:50 +00001059 VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN,
Rusty Russellc45a6812008-05-02 21:50:50 -05001060};
1061
Uwe Kleine-König22402522009-11-05 01:32:44 -08001062static struct virtio_driver virtio_net_driver = {
Rusty Russellc45a6812008-05-02 21:50:50 -05001063 .feature_table = features,
1064 .feature_table_size = ARRAY_SIZE(features),
Rusty Russell296f96f2007-10-22 11:03:37 +10001065 .driver.name = KBUILD_MODNAME,
1066 .driver.owner = THIS_MODULE,
1067 .id_table = id_table,
1068 .probe = virtnet_probe,
1069 .remove = __devexit_p(virtnet_remove),
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -08001070 .config_changed = virtnet_config_changed,
Rusty Russell296f96f2007-10-22 11:03:37 +10001071};
1072
1073static int __init init(void)
1074{
Uwe Kleine-König22402522009-11-05 01:32:44 -08001075 return register_virtio_driver(&virtio_net_driver);
Rusty Russell296f96f2007-10-22 11:03:37 +10001076}
1077
1078static void __exit fini(void)
1079{
Uwe Kleine-König22402522009-11-05 01:32:44 -08001080 unregister_virtio_driver(&virtio_net_driver);
Rusty Russell296f96f2007-10-22 11:03:37 +10001081}
1082module_init(init);
1083module_exit(fini);
1084
1085MODULE_DEVICE_TABLE(virtio, id_table);
1086MODULE_DESCRIPTION("Virtio network driver");
1087MODULE_LICENSE("GPL");