]>
Commit | Line | Data |
---|---|---|
48925e37 | 1 | /* A network driver using virtio. |
296f96fc RR |
2 | * |
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> | |
a9ea3fc6 | 22 | #include <linux/ethtool.h> |
296f96fc RR |
23 | #include <linux/module.h> |
24 | #include <linux/virtio.h> | |
25 | #include <linux/virtio_net.h> | |
26 | #include <linux/scatterlist.h> | |
e918085a | 27 | #include <linux/if_vlan.h> |
5a0e3ad6 | 28 | #include <linux/slab.h> |
296f96fc | 29 | |
6c0cd7c0 DL |
30 | static int napi_weight = 128; |
31 | module_param(napi_weight, int, 0444); | |
32 | ||
34a48579 RR |
33 | static int csum = 1, gso = 1; |
34 | module_param(csum, bool, 0444); | |
35 | module_param(gso, bool, 0444); | |
36 | ||
296f96fc | 37 | /* FIXME: MTU in config. */ |
e918085a | 38 | #define MAX_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN) |
3f2c31d9 | 39 | #define GOOD_COPY_LEN 128 |
296f96fc | 40 | |
f565a7c2 | 41 | #define VIRTNET_SEND_COMMAND_SG_MAX 2 |
2a41f71d | 42 | |
5e01d2f9 | 43 | struct virtnet_info { |
296f96fc | 44 | struct virtio_device *vdev; |
2a41f71d | 45 | struct virtqueue *rvq, *svq, *cvq; |
296f96fc RR |
46 | struct net_device *dev; |
47 | struct napi_struct napi; | |
9f4d26d0 | 48 | unsigned int status; |
296f96fc RR |
49 | |
50 | /* Number of input buffers, and max we've ever had. */ | |
51 | unsigned int num, max; | |
52 | ||
97402b96 HX |
53 | /* I like... big packets and I cannot lie! */ |
54 | bool big_packets; | |
55 | ||
3f2c31d9 MM |
56 | /* Host will merge rx buffers for big packets (shake it! shake it!) */ |
57 | bool mergeable_rx_bufs; | |
58 | ||
3161e453 RR |
59 | /* Work struct for refilling if we run low on memory. */ |
60 | struct delayed_work refill; | |
61 | ||
fb6813f4 RR |
62 | /* Chain pages by the private ptr. */ |
63 | struct page *pages; | |
5e01d2f9 MT |
64 | |
65 | /* fragments + linear part + virtio header */ | |
66 | struct scatterlist rx_sg[MAX_SKB_FRAGS + 2]; | |
67 | struct scatterlist tx_sg[MAX_SKB_FRAGS + 2]; | |
296f96fc RR |
68 | }; |
69 | ||
b3f24698 RR |
70 | struct skb_vnet_hdr { |
71 | union { | |
72 | struct virtio_net_hdr hdr; | |
73 | struct virtio_net_hdr_mrg_rxbuf mhdr; | |
74 | }; | |
48925e37 | 75 | unsigned int num_sg; |
b3f24698 RR |
76 | }; |
77 | ||
9ab86bbc SM |
78 | struct padded_vnet_hdr { |
79 | struct virtio_net_hdr hdr; | |
80 | /* | |
81 | * virtio_net_hdr should be in a separated sg buffer because of a | |
82 | * QEMU bug, and data sg buffer shares same page with this header sg. | |
83 | * This padding makes next sg 16 byte aligned after virtio_net_hdr. | |
84 | */ | |
85 | char padding[6]; | |
86 | }; | |
87 | ||
b3f24698 | 88 | static inline struct skb_vnet_hdr *skb_vnet_hdr(struct sk_buff *skb) |
296f96fc | 89 | { |
b3f24698 | 90 | return (struct skb_vnet_hdr *)skb->cb; |
296f96fc RR |
91 | } |
92 | ||
9ab86bbc SM |
93 | /* |
94 | * private is used to chain pages for big packets, put the whole | |
95 | * most recent used list in the beginning for reuse | |
96 | */ | |
97 | static void give_pages(struct virtnet_info *vi, struct page *page) | |
0a888fd1 | 98 | { |
9ab86bbc | 99 | struct page *end; |
0a888fd1 | 100 | |
9ab86bbc SM |
101 | /* Find end of list, sew whole thing into vi->pages. */ |
102 | for (end = page; end->private; end = (struct page *)end->private); | |
103 | end->private = (unsigned long)vi->pages; | |
104 | vi->pages = page; | |
0a888fd1 MM |
105 | } |
106 | ||
fb6813f4 RR |
107 | static struct page *get_a_page(struct virtnet_info *vi, gfp_t gfp_mask) |
108 | { | |
109 | struct page *p = vi->pages; | |
110 | ||
9ab86bbc | 111 | if (p) { |
fb6813f4 | 112 | vi->pages = (struct page *)p->private; |
9ab86bbc SM |
113 | /* clear private here, it is used to chain pages */ |
114 | p->private = 0; | |
115 | } else | |
fb6813f4 RR |
116 | p = alloc_page(gfp_mask); |
117 | return p; | |
118 | } | |
119 | ||
2cb9c6ba | 120 | static void skb_xmit_done(struct virtqueue *svq) |
296f96fc | 121 | { |
2cb9c6ba | 122 | struct virtnet_info *vi = svq->vdev->priv; |
296f96fc | 123 | |
2cb9c6ba | 124 | /* Suppress further interrupts. */ |
1915a712 | 125 | virtqueue_disable_cb(svq); |
11a3a154 | 126 | |
363f1514 | 127 | /* We were probably waiting for more output buffers. */ |
296f96fc | 128 | netif_wake_queue(vi->dev); |
296f96fc RR |
129 | } |
130 | ||
9ab86bbc SM |
131 | static void set_skb_frag(struct sk_buff *skb, struct page *page, |
132 | unsigned int offset, unsigned int *len) | |
296f96fc | 133 | { |
9ab86bbc SM |
134 | int i = skb_shinfo(skb)->nr_frags; |
135 | skb_frag_t *f; | |
136 | ||
137 | f = &skb_shinfo(skb)->frags[i]; | |
138 | f->size = min((unsigned)PAGE_SIZE - offset, *len); | |
139 | f->page_offset = offset; | |
140 | f->page = page; | |
141 | ||
142 | skb->data_len += f->size; | |
143 | skb->len += f->size; | |
144 | skb_shinfo(skb)->nr_frags++; | |
145 | *len -= f->size; | |
146 | } | |
23cde76d | 147 | |
9ab86bbc SM |
148 | static struct sk_buff *page_to_skb(struct virtnet_info *vi, |
149 | struct page *page, unsigned int len) | |
150 | { | |
151 | struct sk_buff *skb; | |
152 | struct skb_vnet_hdr *hdr; | |
153 | unsigned int copy, hdr_len, offset; | |
154 | char *p; | |
fb6813f4 | 155 | |
9ab86bbc | 156 | p = page_address(page); |
3f2c31d9 | 157 | |
9ab86bbc SM |
158 | /* copy small packet so we can reuse these pages for small data */ |
159 | skb = netdev_alloc_skb_ip_align(vi->dev, GOOD_COPY_LEN); | |
160 | if (unlikely(!skb)) | |
161 | return NULL; | |
3f2c31d9 | 162 | |
9ab86bbc | 163 | hdr = skb_vnet_hdr(skb); |
3f2c31d9 | 164 | |
9ab86bbc SM |
165 | if (vi->mergeable_rx_bufs) { |
166 | hdr_len = sizeof hdr->mhdr; | |
167 | offset = hdr_len; | |
168 | } else { | |
169 | hdr_len = sizeof hdr->hdr; | |
170 | offset = sizeof(struct padded_vnet_hdr); | |
171 | } | |
3f2c31d9 | 172 | |
9ab86bbc | 173 | memcpy(hdr, p, hdr_len); |
3f2c31d9 | 174 | |
9ab86bbc SM |
175 | len -= hdr_len; |
176 | p += offset; | |
3f2c31d9 | 177 | |
9ab86bbc SM |
178 | copy = len; |
179 | if (copy > skb_tailroom(skb)) | |
180 | copy = skb_tailroom(skb); | |
181 | memcpy(skb_put(skb, copy), p, copy); | |
3f2c31d9 | 182 | |
9ab86bbc SM |
183 | len -= copy; |
184 | offset += copy; | |
3f2c31d9 | 185 | |
9ab86bbc SM |
186 | while (len) { |
187 | set_skb_frag(skb, page, offset, &len); | |
188 | page = (struct page *)page->private; | |
189 | offset = 0; | |
190 | } | |
3f2c31d9 | 191 | |
9ab86bbc SM |
192 | if (page) |
193 | give_pages(vi, page); | |
3f2c31d9 | 194 | |
9ab86bbc SM |
195 | return skb; |
196 | } | |
3f2c31d9 | 197 | |
9ab86bbc SM |
198 | static int receive_mergeable(struct virtnet_info *vi, struct sk_buff *skb) |
199 | { | |
200 | struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb); | |
201 | struct page *page; | |
202 | int num_buf, i, len; | |
203 | ||
204 | num_buf = hdr->mhdr.num_buffers; | |
205 | while (--num_buf) { | |
206 | i = skb_shinfo(skb)->nr_frags; | |
207 | if (i >= MAX_SKB_FRAGS) { | |
208 | pr_debug("%s: packet too long\n", skb->dev->name); | |
209 | skb->dev->stats.rx_length_errors++; | |
210 | return -EINVAL; | |
211 | } | |
3f2c31d9 | 212 | |
1915a712 | 213 | page = virtqueue_get_buf(vi->rvq, &len); |
9ab86bbc SM |
214 | if (!page) { |
215 | pr_debug("%s: rx error: %d buffers missing\n", | |
216 | skb->dev->name, hdr->mhdr.num_buffers); | |
217 | skb->dev->stats.rx_length_errors++; | |
218 | return -EINVAL; | |
3f2c31d9 | 219 | } |
9ab86bbc SM |
220 | if (len > PAGE_SIZE) |
221 | len = PAGE_SIZE; | |
222 | ||
223 | set_skb_frag(skb, page, 0, &len); | |
224 | ||
225 | --vi->num; | |
226 | } | |
227 | return 0; | |
228 | } | |
229 | ||
230 | static void receive_buf(struct net_device *dev, void *buf, unsigned int len) | |
231 | { | |
232 | struct virtnet_info *vi = netdev_priv(dev); | |
233 | struct sk_buff *skb; | |
234 | struct page *page; | |
235 | struct skb_vnet_hdr *hdr; | |
3f2c31d9 | 236 | |
9ab86bbc SM |
237 | if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) { |
238 | pr_debug("%s: short packet %i\n", dev->name, len); | |
239 | dev->stats.rx_length_errors++; | |
240 | if (vi->mergeable_rx_bufs || vi->big_packets) | |
241 | give_pages(vi, buf); | |
242 | else | |
243 | dev_kfree_skb(buf); | |
244 | return; | |
245 | } | |
3f2c31d9 | 246 | |
9ab86bbc SM |
247 | if (!vi->mergeable_rx_bufs && !vi->big_packets) { |
248 | skb = buf; | |
249 | len -= sizeof(struct virtio_net_hdr); | |
250 | skb_trim(skb, len); | |
251 | } else { | |
252 | page = buf; | |
253 | skb = page_to_skb(vi, page, len); | |
254 | if (unlikely(!skb)) { | |
3f2c31d9 | 255 | dev->stats.rx_dropped++; |
9ab86bbc SM |
256 | give_pages(vi, page); |
257 | return; | |
3f2c31d9 | 258 | } |
9ab86bbc SM |
259 | if (vi->mergeable_rx_bufs) |
260 | if (receive_mergeable(vi, skb)) { | |
261 | dev_kfree_skb(skb); | |
262 | return; | |
263 | } | |
97402b96 | 264 | } |
3f2c31d9 | 265 | |
9ab86bbc | 266 | hdr = skb_vnet_hdr(skb); |
97402b96 | 267 | skb->truesize += skb->data_len; |
296f96fc RR |
268 | dev->stats.rx_bytes += skb->len; |
269 | dev->stats.rx_packets++; | |
270 | ||
b3f24698 | 271 | if (hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { |
296f96fc | 272 | pr_debug("Needs csum!\n"); |
b3f24698 RR |
273 | if (!skb_partial_csum_set(skb, |
274 | hdr->hdr.csum_start, | |
275 | hdr->hdr.csum_offset)) | |
296f96fc | 276 | goto frame_err; |
296f96fc RR |
277 | } |
278 | ||
23cde76d MM |
279 | skb->protocol = eth_type_trans(skb, dev); |
280 | pr_debug("Receiving skb proto 0x%04x len %i type %i\n", | |
281 | ntohs(skb->protocol), skb->len, skb->pkt_type); | |
282 | ||
b3f24698 | 283 | if (hdr->hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) { |
296f96fc | 284 | pr_debug("GSO!\n"); |
b3f24698 | 285 | switch (hdr->hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { |
296f96fc RR |
286 | case VIRTIO_NET_HDR_GSO_TCPV4: |
287 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; | |
288 | break; | |
296f96fc RR |
289 | case VIRTIO_NET_HDR_GSO_UDP: |
290 | skb_shinfo(skb)->gso_type = SKB_GSO_UDP; | |
291 | break; | |
292 | case VIRTIO_NET_HDR_GSO_TCPV6: | |
293 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6; | |
294 | break; | |
295 | default: | |
296 | if (net_ratelimit()) | |
297 | printk(KERN_WARNING "%s: bad gso type %u.\n", | |
b3f24698 | 298 | dev->name, hdr->hdr.gso_type); |
296f96fc RR |
299 | goto frame_err; |
300 | } | |
301 | ||
b3f24698 | 302 | if (hdr->hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN) |
34a48579 RR |
303 | skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN; |
304 | ||
b3f24698 | 305 | skb_shinfo(skb)->gso_size = hdr->hdr.gso_size; |
296f96fc RR |
306 | if (skb_shinfo(skb)->gso_size == 0) { |
307 | if (net_ratelimit()) | |
308 | printk(KERN_WARNING "%s: zero gso size.\n", | |
309 | dev->name); | |
310 | goto frame_err; | |
311 | } | |
312 | ||
313 | /* Header must be checked, and gso_segs computed. */ | |
314 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
315 | skb_shinfo(skb)->gso_segs = 0; | |
316 | } | |
317 | ||
318 | netif_receive_skb(skb); | |
319 | return; | |
320 | ||
321 | frame_err: | |
322 | dev->stats.rx_frame_errors++; | |
296f96fc RR |
323 | dev_kfree_skb(skb); |
324 | } | |
325 | ||
9ab86bbc | 326 | static int add_recvbuf_small(struct virtnet_info *vi, gfp_t gfp) |
296f96fc RR |
327 | { |
328 | struct sk_buff *skb; | |
9ab86bbc | 329 | struct skb_vnet_hdr *hdr; |
9ab86bbc | 330 | int err; |
3f2c31d9 | 331 | |
9ab86bbc SM |
332 | skb = netdev_alloc_skb_ip_align(vi->dev, MAX_PACKET_LEN); |
333 | if (unlikely(!skb)) | |
334 | return -ENOMEM; | |
296f96fc | 335 | |
9ab86bbc | 336 | skb_put(skb, MAX_PACKET_LEN); |
3f2c31d9 | 337 | |
9ab86bbc | 338 | hdr = skb_vnet_hdr(skb); |
5e01d2f9 | 339 | sg_set_buf(vi->rx_sg, &hdr->hdr, sizeof hdr->hdr); |
97402b96 | 340 | |
5e01d2f9 | 341 | skb_to_sgvec(skb, vi->rx_sg + 1, 0, skb->len); |
97402b96 | 342 | |
aa989f5e | 343 | err = virtqueue_add_buf_gfp(vi->rvq, vi->rx_sg, 0, 2, skb, gfp); |
9ab86bbc SM |
344 | if (err < 0) |
345 | dev_kfree_skb(skb); | |
97402b96 | 346 | |
9ab86bbc SM |
347 | return err; |
348 | } | |
97402b96 | 349 | |
9ab86bbc SM |
350 | static int add_recvbuf_big(struct virtnet_info *vi, gfp_t gfp) |
351 | { | |
9ab86bbc SM |
352 | struct page *first, *list = NULL; |
353 | char *p; | |
354 | int i, err, offset; | |
355 | ||
5e01d2f9 | 356 | /* page in vi->rx_sg[MAX_SKB_FRAGS + 1] is list tail */ |
9ab86bbc SM |
357 | for (i = MAX_SKB_FRAGS + 1; i > 1; --i) { |
358 | first = get_a_page(vi, gfp); | |
359 | if (!first) { | |
360 | if (list) | |
361 | give_pages(vi, list); | |
362 | return -ENOMEM; | |
97402b96 | 363 | } |
5e01d2f9 | 364 | sg_set_buf(&vi->rx_sg[i], page_address(first), PAGE_SIZE); |
97402b96 | 365 | |
9ab86bbc SM |
366 | /* chain new page in list head to match sg */ |
367 | first->private = (unsigned long)list; | |
368 | list = first; | |
369 | } | |
296f96fc | 370 | |
9ab86bbc SM |
371 | first = get_a_page(vi, gfp); |
372 | if (!first) { | |
373 | give_pages(vi, list); | |
374 | return -ENOMEM; | |
375 | } | |
376 | p = page_address(first); | |
377 | ||
5e01d2f9 MT |
378 | /* vi->rx_sg[0], vi->rx_sg[1] share the same page */ |
379 | /* a separated vi->rx_sg[0] for virtio_net_hdr only due to QEMU bug */ | |
380 | sg_set_buf(&vi->rx_sg[0], p, sizeof(struct virtio_net_hdr)); | |
9ab86bbc | 381 | |
5e01d2f9 | 382 | /* vi->rx_sg[1] for data packet, from offset */ |
9ab86bbc | 383 | offset = sizeof(struct padded_vnet_hdr); |
5e01d2f9 | 384 | sg_set_buf(&vi->rx_sg[1], p + offset, PAGE_SIZE - offset); |
9ab86bbc SM |
385 | |
386 | /* chain first in list head */ | |
387 | first->private = (unsigned long)list; | |
aa989f5e MT |
388 | err = virtqueue_add_buf_gfp(vi->rvq, vi->rx_sg, 0, MAX_SKB_FRAGS + 2, |
389 | first, gfp); | |
9ab86bbc SM |
390 | if (err < 0) |
391 | give_pages(vi, first); | |
392 | ||
393 | return err; | |
296f96fc RR |
394 | } |
395 | ||
9ab86bbc | 396 | static int add_recvbuf_mergeable(struct virtnet_info *vi, gfp_t gfp) |
3f2c31d9 | 397 | { |
9ab86bbc | 398 | struct page *page; |
3f2c31d9 | 399 | int err; |
3f2c31d9 | 400 | |
9ab86bbc SM |
401 | page = get_a_page(vi, gfp); |
402 | if (!page) | |
403 | return -ENOMEM; | |
3f2c31d9 | 404 | |
5e01d2f9 | 405 | sg_init_one(vi->rx_sg, page_address(page), PAGE_SIZE); |
3f2c31d9 | 406 | |
aa989f5e | 407 | err = virtqueue_add_buf_gfp(vi->rvq, vi->rx_sg, 0, 1, page, gfp); |
9ab86bbc SM |
408 | if (err < 0) |
409 | give_pages(vi, page); | |
3f2c31d9 | 410 | |
9ab86bbc SM |
411 | return err; |
412 | } | |
3f2c31d9 | 413 | |
9ab86bbc SM |
414 | /* Returns false if we couldn't fill entirely (OOM). */ |
415 | static bool try_fill_recv(struct virtnet_info *vi, gfp_t gfp) | |
416 | { | |
417 | int err; | |
1788f495 | 418 | bool oom; |
3f2c31d9 | 419 | |
9ab86bbc SM |
420 | do { |
421 | 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); | |
3f2c31d9 | 427 | |
1788f495 MT |
428 | oom = err == -ENOMEM; |
429 | if (err < 0) | |
3f2c31d9 | 430 | break; |
9ab86bbc | 431 | ++vi->num; |
0aea51c3 | 432 | } while (err > 0); |
3f2c31d9 MM |
433 | if (unlikely(vi->num > vi->max)) |
434 | vi->max = vi->num; | |
1915a712 | 435 | virtqueue_kick(vi->rvq); |
3161e453 | 436 | return !oom; |
3f2c31d9 MM |
437 | } |
438 | ||
18445c4d | 439 | static void skb_recv_done(struct virtqueue *rvq) |
296f96fc RR |
440 | { |
441 | struct virtnet_info *vi = rvq->vdev->priv; | |
18445c4d | 442 | /* Schedule NAPI, Suppress further interrupts if successful. */ |
288379f0 | 443 | if (napi_schedule_prep(&vi->napi)) { |
1915a712 | 444 | virtqueue_disable_cb(rvq); |
288379f0 | 445 | __napi_schedule(&vi->napi); |
18445c4d | 446 | } |
296f96fc RR |
447 | } |
448 | ||
3161e453 RR |
449 | static void refill_work(struct work_struct *work) |
450 | { | |
451 | struct virtnet_info *vi; | |
452 | bool still_empty; | |
453 | ||
454 | vi = container_of(work, struct virtnet_info, refill.work); | |
455 | napi_disable(&vi->napi); | |
39d32157 | 456 | still_empty = !try_fill_recv(vi, GFP_KERNEL); |
3161e453 RR |
457 | napi_enable(&vi->napi); |
458 | ||
459 | /* In theory, this can happen: if we don't get any buffers in | |
460 | * we will *never* try to fill again. */ | |
461 | if (still_empty) | |
462 | schedule_delayed_work(&vi->refill, HZ/2); | |
463 | } | |
464 | ||
296f96fc RR |
465 | static int virtnet_poll(struct napi_struct *napi, int budget) |
466 | { | |
467 | struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi); | |
9ab86bbc | 468 | void *buf; |
296f96fc RR |
469 | unsigned int len, received = 0; |
470 | ||
471 | again: | |
472 | while (received < budget && | |
1915a712 | 473 | (buf = virtqueue_get_buf(vi->rvq, &len)) != NULL) { |
9ab86bbc SM |
474 | receive_buf(vi->dev, buf, len); |
475 | --vi->num; | |
296f96fc RR |
476 | received++; |
477 | } | |
478 | ||
3161e453 RR |
479 | if (vi->num < vi->max / 2) { |
480 | if (!try_fill_recv(vi, GFP_ATOMIC)) | |
481 | schedule_delayed_work(&vi->refill, 0); | |
482 | } | |
296f96fc | 483 | |
8329d98e RR |
484 | /* Out of packets? */ |
485 | if (received < budget) { | |
288379f0 | 486 | napi_complete(napi); |
1915a712 | 487 | if (unlikely(!virtqueue_enable_cb(vi->rvq)) && |
8e95a202 | 488 | napi_schedule_prep(napi)) { |
1915a712 | 489 | virtqueue_disable_cb(vi->rvq); |
288379f0 | 490 | __napi_schedule(napi); |
296f96fc | 491 | goto again; |
4265f161 | 492 | } |
296f96fc RR |
493 | } |
494 | ||
495 | return received; | |
496 | } | |
497 | ||
48925e37 | 498 | static unsigned int free_old_xmit_skbs(struct virtnet_info *vi) |
296f96fc RR |
499 | { |
500 | struct sk_buff *skb; | |
48925e37 | 501 | unsigned int len, tot_sgs = 0; |
296f96fc | 502 | |
1915a712 | 503 | while ((skb = virtqueue_get_buf(vi->svq, &len)) != NULL) { |
296f96fc | 504 | pr_debug("Sent skb %p\n", skb); |
655aa31f | 505 | vi->dev->stats.tx_bytes += skb->len; |
296f96fc | 506 | vi->dev->stats.tx_packets++; |
48925e37 | 507 | tot_sgs += skb_vnet_hdr(skb)->num_sg; |
ed79bab8 | 508 | dev_kfree_skb_any(skb); |
296f96fc | 509 | } |
48925e37 | 510 | return tot_sgs; |
296f96fc RR |
511 | } |
512 | ||
99ffc696 | 513 | static int xmit_skb(struct virtnet_info *vi, struct sk_buff *skb) |
296f96fc | 514 | { |
b3f24698 | 515 | struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb); |
296f96fc | 516 | const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest; |
296f96fc | 517 | |
e174961c | 518 | pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest); |
296f96fc | 519 | |
296f96fc | 520 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
b3f24698 RR |
521 | hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; |
522 | hdr->hdr.csum_start = skb->csum_start - skb_headroom(skb); | |
523 | hdr->hdr.csum_offset = skb->csum_offset; | |
296f96fc | 524 | } else { |
b3f24698 RR |
525 | hdr->hdr.flags = 0; |
526 | hdr->hdr.csum_offset = hdr->hdr.csum_start = 0; | |
296f96fc RR |
527 | } |
528 | ||
529 | if (skb_is_gso(skb)) { | |
b3f24698 RR |
530 | hdr->hdr.hdr_len = skb_headlen(skb); |
531 | hdr->hdr.gso_size = skb_shinfo(skb)->gso_size; | |
34a48579 | 532 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) |
b3f24698 | 533 | hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4; |
296f96fc | 534 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) |
b3f24698 | 535 | hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6; |
296f96fc | 536 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP) |
b3f24698 | 537 | hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP; |
296f96fc RR |
538 | else |
539 | BUG(); | |
34a48579 | 540 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN) |
b3f24698 | 541 | hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN; |
296f96fc | 542 | } else { |
b3f24698 RR |
543 | hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE; |
544 | hdr->hdr.gso_size = hdr->hdr.hdr_len = 0; | |
296f96fc RR |
545 | } |
546 | ||
b3f24698 | 547 | hdr->mhdr.num_buffers = 0; |
3f2c31d9 MM |
548 | |
549 | /* Encode metadata header at front. */ | |
550 | if (vi->mergeable_rx_bufs) | |
5e01d2f9 | 551 | sg_set_buf(vi->tx_sg, &hdr->mhdr, sizeof hdr->mhdr); |
3f2c31d9 | 552 | else |
5e01d2f9 | 553 | sg_set_buf(vi->tx_sg, &hdr->hdr, sizeof hdr->hdr); |
3f2c31d9 | 554 | |
5e01d2f9 | 555 | hdr->num_sg = skb_to_sgvec(skb, vi->tx_sg + 1, 0, skb->len) + 1; |
1756ac3d | 556 | return virtqueue_add_buf(vi->svq, vi->tx_sg, hdr->num_sg, |
5e01d2f9 | 557 | 0, skb); |
11a3a154 RR |
558 | } |
559 | ||
424efe9c | 560 | static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev) |
99ffc696 RR |
561 | { |
562 | struct virtnet_info *vi = netdev_priv(dev); | |
48925e37 | 563 | int capacity; |
2cb9c6ba | 564 | |
2cb9c6ba RR |
565 | /* Free up any pending old buffers before queueing new ones. */ |
566 | free_old_xmit_skbs(vi); | |
99ffc696 | 567 | |
03f191ba | 568 | /* Try to transmit */ |
48925e37 RR |
569 | capacity = xmit_skb(vi, skb); |
570 | ||
571 | /* This can happen with OOM and indirect buffers. */ | |
572 | if (unlikely(capacity < 0)) { | |
58eba97d RR |
573 | if (net_ratelimit()) { |
574 | if (likely(capacity == -ENOMEM)) { | |
575 | dev_warn(&dev->dev, | |
576 | "TX queue failure: out of memory\n"); | |
577 | } else { | |
578 | dev->stats.tx_fifo_errors++; | |
579 | dev_warn(&dev->dev, | |
580 | "Unexpected TX queue failure: %d\n", | |
581 | capacity); | |
582 | } | |
48925e37 | 583 | } |
58eba97d RR |
584 | dev->stats.tx_dropped++; |
585 | kfree_skb(skb); | |
586 | return NETDEV_TX_OK; | |
296f96fc | 587 | } |
1915a712 | 588 | virtqueue_kick(vi->svq); |
03f191ba | 589 | |
48925e37 RR |
590 | /* Don't wait up for transmitted skbs to be freed. */ |
591 | skb_orphan(skb); | |
592 | nf_reset(skb); | |
593 | ||
594 | /* Apparently nice girls don't return TX_BUSY; stop the queue | |
595 | * before it gets out of hand. Naturally, this wastes entries. */ | |
596 | if (capacity < 2+MAX_SKB_FRAGS) { | |
597 | netif_stop_queue(dev); | |
1915a712 | 598 | if (unlikely(!virtqueue_enable_cb(vi->svq))) { |
48925e37 RR |
599 | /* More just got used, free them then recheck. */ |
600 | capacity += free_old_xmit_skbs(vi); | |
601 | if (capacity >= 2+MAX_SKB_FRAGS) { | |
602 | netif_start_queue(dev); | |
1915a712 | 603 | virtqueue_disable_cb(vi->svq); |
48925e37 RR |
604 | } |
605 | } | |
99ffc696 | 606 | } |
48925e37 RR |
607 | |
608 | return NETDEV_TX_OK; | |
296f96fc RR |
609 | } |
610 | ||
9c46f6d4 AW |
611 | static int virtnet_set_mac_address(struct net_device *dev, void *p) |
612 | { | |
613 | struct virtnet_info *vi = netdev_priv(dev); | |
614 | struct virtio_device *vdev = vi->vdev; | |
615 | int ret; | |
616 | ||
617 | ret = eth_mac_addr(dev, p); | |
618 | if (ret) | |
619 | return ret; | |
620 | ||
62994b2d AW |
621 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) |
622 | vdev->config->set(vdev, offsetof(struct virtio_net_config, mac), | |
623 | dev->dev_addr, dev->addr_len); | |
9c46f6d4 AW |
624 | |
625 | return 0; | |
626 | } | |
627 | ||
da74e89d AS |
628 | #ifdef CONFIG_NET_POLL_CONTROLLER |
629 | static void virtnet_netpoll(struct net_device *dev) | |
630 | { | |
631 | struct virtnet_info *vi = netdev_priv(dev); | |
632 | ||
633 | napi_schedule(&vi->napi); | |
634 | } | |
635 | #endif | |
636 | ||
296f96fc RR |
637 | static int virtnet_open(struct net_device *dev) |
638 | { | |
639 | struct virtnet_info *vi = netdev_priv(dev); | |
640 | ||
296f96fc | 641 | napi_enable(&vi->napi); |
a48bd8f6 RR |
642 | |
643 | /* If all buffers were filled by other side before we napi_enabled, we | |
644 | * won't get another interrupt, so process any outstanding packets | |
370076d9 CB |
645 | * now. virtnet_poll wants re-enable the queue, so we disable here. |
646 | * We synchronize against interrupts via NAPI_STATE_SCHED */ | |
288379f0 | 647 | if (napi_schedule_prep(&vi->napi)) { |
1915a712 | 648 | virtqueue_disable_cb(vi->rvq); |
288379f0 | 649 | __napi_schedule(&vi->napi); |
370076d9 | 650 | } |
296f96fc RR |
651 | return 0; |
652 | } | |
653 | ||
2a41f71d AW |
654 | /* |
655 | * Send command via the control virtqueue and check status. Commands | |
656 | * supported by the hypervisor, as indicated by feature bits, should | |
657 | * never fail unless improperly formated. | |
658 | */ | |
659 | static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd, | |
660 | struct scatterlist *data, int out, int in) | |
661 | { | |
23e258e1 | 662 | struct scatterlist *s, sg[VIRTNET_SEND_COMMAND_SG_MAX + 2]; |
2a41f71d AW |
663 | struct virtio_net_ctrl_hdr ctrl; |
664 | virtio_net_ctrl_ack status = ~0; | |
665 | unsigned int tmp; | |
23e258e1 | 666 | int i; |
2a41f71d | 667 | |
0ee904c3 AB |
668 | /* Caller should know better */ |
669 | BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) || | |
670 | (out + in > VIRTNET_SEND_COMMAND_SG_MAX)); | |
2a41f71d AW |
671 | |
672 | out++; /* Add header */ | |
673 | in++; /* Add return status */ | |
674 | ||
675 | ctrl.class = class; | |
676 | ctrl.cmd = cmd; | |
677 | ||
678 | sg_init_table(sg, out + in); | |
679 | ||
680 | sg_set_buf(&sg[0], &ctrl, sizeof(ctrl)); | |
23e258e1 AW |
681 | for_each_sg(data, s, out + in - 2, i) |
682 | sg_set_buf(&sg[i + 1], sg_virt(s), s->length); | |
2a41f71d AW |
683 | sg_set_buf(&sg[out + in - 1], &status, sizeof(status)); |
684 | ||
1915a712 | 685 | BUG_ON(virtqueue_add_buf(vi->cvq, sg, out, in, vi) < 0); |
2a41f71d | 686 | |
1915a712 | 687 | virtqueue_kick(vi->cvq); |
2a41f71d AW |
688 | |
689 | /* | |
690 | * Spin for a response, the kick causes an ioport write, trapping | |
691 | * into the hypervisor, so the request should be handled immediately. | |
692 | */ | |
1915a712 | 693 | while (!virtqueue_get_buf(vi->cvq, &tmp)) |
2a41f71d AW |
694 | cpu_relax(); |
695 | ||
696 | return status == VIRTIO_NET_OK; | |
697 | } | |
698 | ||
296f96fc RR |
699 | static int virtnet_close(struct net_device *dev) |
700 | { | |
701 | struct virtnet_info *vi = netdev_priv(dev); | |
296f96fc RR |
702 | |
703 | napi_disable(&vi->napi); | |
704 | ||
296f96fc RR |
705 | return 0; |
706 | } | |
707 | ||
a9ea3fc6 HX |
708 | static int virtnet_set_tx_csum(struct net_device *dev, u32 data) |
709 | { | |
710 | struct virtnet_info *vi = netdev_priv(dev); | |
711 | struct virtio_device *vdev = vi->vdev; | |
712 | ||
713 | if (data && !virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) | |
714 | return -ENOSYS; | |
715 | ||
716 | return ethtool_op_set_tx_hw_csum(dev, data); | |
717 | } | |
718 | ||
2af7698e AW |
719 | static void virtnet_set_rx_mode(struct net_device *dev) |
720 | { | |
721 | struct virtnet_info *vi = netdev_priv(dev); | |
f565a7c2 | 722 | struct scatterlist sg[2]; |
2af7698e | 723 | u8 promisc, allmulti; |
f565a7c2 | 724 | struct virtio_net_ctrl_mac *mac_data; |
ccffad25 | 725 | struct netdev_hw_addr *ha; |
32e7bfc4 | 726 | int uc_count; |
4cd24eaf | 727 | int mc_count; |
f565a7c2 AW |
728 | void *buf; |
729 | int i; | |
2af7698e AW |
730 | |
731 | /* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */ | |
732 | if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX)) | |
733 | return; | |
734 | ||
f565a7c2 AW |
735 | promisc = ((dev->flags & IFF_PROMISC) != 0); |
736 | allmulti = ((dev->flags & IFF_ALLMULTI) != 0); | |
2af7698e | 737 | |
23e258e1 | 738 | sg_init_one(sg, &promisc, sizeof(promisc)); |
2af7698e AW |
739 | |
740 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, | |
741 | VIRTIO_NET_CTRL_RX_PROMISC, | |
f565a7c2 | 742 | sg, 1, 0)) |
2af7698e AW |
743 | dev_warn(&dev->dev, "Failed to %sable promisc mode.\n", |
744 | promisc ? "en" : "dis"); | |
745 | ||
23e258e1 | 746 | sg_init_one(sg, &allmulti, sizeof(allmulti)); |
2af7698e AW |
747 | |
748 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, | |
749 | VIRTIO_NET_CTRL_RX_ALLMULTI, | |
f565a7c2 | 750 | sg, 1, 0)) |
2af7698e AW |
751 | dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n", |
752 | allmulti ? "en" : "dis"); | |
f565a7c2 | 753 | |
32e7bfc4 | 754 | uc_count = netdev_uc_count(dev); |
4cd24eaf | 755 | mc_count = netdev_mc_count(dev); |
f565a7c2 | 756 | /* MAC filter - use one buffer for both lists */ |
4cd24eaf JP |
757 | buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) + |
758 | (2 * sizeof(mac_data->entries)), GFP_ATOMIC); | |
759 | mac_data = buf; | |
f565a7c2 AW |
760 | if (!buf) { |
761 | dev_warn(&dev->dev, "No memory for MAC address buffer\n"); | |
762 | return; | |
763 | } | |
764 | ||
23e258e1 AW |
765 | sg_init_table(sg, 2); |
766 | ||
f565a7c2 | 767 | /* Store the unicast list and count in the front of the buffer */ |
32e7bfc4 | 768 | mac_data->entries = uc_count; |
ccffad25 | 769 | i = 0; |
32e7bfc4 | 770 | netdev_for_each_uc_addr(ha, dev) |
ccffad25 | 771 | memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); |
f565a7c2 AW |
772 | |
773 | sg_set_buf(&sg[0], mac_data, | |
32e7bfc4 | 774 | sizeof(mac_data->entries) + (uc_count * ETH_ALEN)); |
f565a7c2 AW |
775 | |
776 | /* multicast list and count fill the end */ | |
32e7bfc4 | 777 | mac_data = (void *)&mac_data->macs[uc_count][0]; |
f565a7c2 | 778 | |
4cd24eaf | 779 | mac_data->entries = mc_count; |
567ec874 | 780 | i = 0; |
22bedad3 JP |
781 | netdev_for_each_mc_addr(ha, dev) |
782 | memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); | |
f565a7c2 AW |
783 | |
784 | sg_set_buf(&sg[1], mac_data, | |
4cd24eaf | 785 | sizeof(mac_data->entries) + (mc_count * ETH_ALEN)); |
f565a7c2 AW |
786 | |
787 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC, | |
788 | VIRTIO_NET_CTRL_MAC_TABLE_SET, | |
789 | sg, 2, 0)) | |
790 | dev_warn(&dev->dev, "Failed to set MAC fitler table.\n"); | |
791 | ||
792 | kfree(buf); | |
2af7698e AW |
793 | } |
794 | ||
1824a989 | 795 | static void virtnet_vlan_rx_add_vid(struct net_device *dev, u16 vid) |
0bde9569 AW |
796 | { |
797 | struct virtnet_info *vi = netdev_priv(dev); | |
798 | struct scatterlist sg; | |
799 | ||
23e258e1 | 800 | sg_init_one(&sg, &vid, sizeof(vid)); |
0bde9569 AW |
801 | |
802 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, | |
803 | VIRTIO_NET_CTRL_VLAN_ADD, &sg, 1, 0)) | |
804 | dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid); | |
805 | } | |
806 | ||
1824a989 | 807 | static void virtnet_vlan_rx_kill_vid(struct net_device *dev, u16 vid) |
0bde9569 AW |
808 | { |
809 | struct virtnet_info *vi = netdev_priv(dev); | |
810 | struct scatterlist sg; | |
811 | ||
23e258e1 | 812 | sg_init_one(&sg, &vid, sizeof(vid)); |
0bde9569 AW |
813 | |
814 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, | |
815 | VIRTIO_NET_CTRL_VLAN_DEL, &sg, 1, 0)) | |
816 | dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid); | |
817 | } | |
818 | ||
0fc0b732 | 819 | static const struct ethtool_ops virtnet_ethtool_ops = { |
a9ea3fc6 HX |
820 | .set_tx_csum = virtnet_set_tx_csum, |
821 | .set_sg = ethtool_op_set_sg, | |
0276b497 | 822 | .set_tso = ethtool_op_set_tso, |
5c516751 | 823 | .set_ufo = ethtool_op_set_ufo, |
9f4d26d0 | 824 | .get_link = ethtool_op_get_link, |
a9ea3fc6 HX |
825 | }; |
826 | ||
39da5814 MM |
827 | #define MIN_MTU 68 |
828 | #define MAX_MTU 65535 | |
829 | ||
830 | static int virtnet_change_mtu(struct net_device *dev, int new_mtu) | |
831 | { | |
832 | if (new_mtu < MIN_MTU || new_mtu > MAX_MTU) | |
833 | return -EINVAL; | |
834 | dev->mtu = new_mtu; | |
835 | return 0; | |
836 | } | |
837 | ||
76288b4e SH |
838 | static const struct net_device_ops virtnet_netdev = { |
839 | .ndo_open = virtnet_open, | |
840 | .ndo_stop = virtnet_close, | |
841 | .ndo_start_xmit = start_xmit, | |
842 | .ndo_validate_addr = eth_validate_addr, | |
9c46f6d4 | 843 | .ndo_set_mac_address = virtnet_set_mac_address, |
2af7698e | 844 | .ndo_set_rx_mode = virtnet_set_rx_mode, |
76288b4e | 845 | .ndo_change_mtu = virtnet_change_mtu, |
1824a989 AW |
846 | .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid, |
847 | .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid, | |
76288b4e SH |
848 | #ifdef CONFIG_NET_POLL_CONTROLLER |
849 | .ndo_poll_controller = virtnet_netpoll, | |
850 | #endif | |
851 | }; | |
852 | ||
9f4d26d0 MM |
853 | static void virtnet_update_status(struct virtnet_info *vi) |
854 | { | |
855 | u16 v; | |
856 | ||
857 | if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) | |
858 | return; | |
859 | ||
860 | vi->vdev->config->get(vi->vdev, | |
861 | offsetof(struct virtio_net_config, status), | |
862 | &v, sizeof(v)); | |
863 | ||
864 | /* Ignore unknown (future) status bits */ | |
865 | v &= VIRTIO_NET_S_LINK_UP; | |
866 | ||
867 | if (vi->status == v) | |
868 | return; | |
869 | ||
870 | vi->status = v; | |
871 | ||
872 | if (vi->status & VIRTIO_NET_S_LINK_UP) { | |
873 | netif_carrier_on(vi->dev); | |
874 | netif_wake_queue(vi->dev); | |
875 | } else { | |
876 | netif_carrier_off(vi->dev); | |
877 | netif_stop_queue(vi->dev); | |
878 | } | |
879 | } | |
880 | ||
881 | static void virtnet_config_changed(struct virtio_device *vdev) | |
882 | { | |
883 | struct virtnet_info *vi = vdev->priv; | |
884 | ||
885 | virtnet_update_status(vi); | |
886 | } | |
887 | ||
296f96fc RR |
888 | static int virtnet_probe(struct virtio_device *vdev) |
889 | { | |
890 | int err; | |
296f96fc RR |
891 | struct net_device *dev; |
892 | struct virtnet_info *vi; | |
d2a7ddda MT |
893 | struct virtqueue *vqs[3]; |
894 | vq_callback_t *callbacks[] = { skb_recv_done, skb_xmit_done, NULL}; | |
895 | const char *names[] = { "input", "output", "control" }; | |
896 | int nvqs; | |
296f96fc RR |
897 | |
898 | /* Allocate ourselves a network device with room for our info */ | |
899 | dev = alloc_etherdev(sizeof(struct virtnet_info)); | |
900 | if (!dev) | |
901 | return -ENOMEM; | |
902 | ||
903 | /* Set up network device as normal. */ | |
76288b4e | 904 | dev->netdev_ops = &virtnet_netdev; |
296f96fc | 905 | dev->features = NETIF_F_HIGHDMA; |
a9ea3fc6 | 906 | SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops); |
296f96fc RR |
907 | SET_NETDEV_DEV(dev, &vdev->dev); |
908 | ||
909 | /* Do we support "hardware" checksums? */ | |
c45a6816 | 910 | if (csum && virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) { |
296f96fc RR |
911 | /* This opens up the world of extra features. */ |
912 | dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST; | |
c45a6816 | 913 | if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) { |
34a48579 RR |
914 | dev->features |= NETIF_F_TSO | NETIF_F_UFO |
915 | | NETIF_F_TSO_ECN | NETIF_F_TSO6; | |
916 | } | |
5539ae96 | 917 | /* Individual feature bits: what can host handle? */ |
c45a6816 | 918 | if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4)) |
5539ae96 | 919 | dev->features |= NETIF_F_TSO; |
c45a6816 | 920 | if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6)) |
5539ae96 | 921 | dev->features |= NETIF_F_TSO6; |
c45a6816 | 922 | if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN)) |
5539ae96 | 923 | dev->features |= NETIF_F_TSO_ECN; |
c45a6816 | 924 | if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO)) |
5539ae96 | 925 | dev->features |= NETIF_F_UFO; |
296f96fc RR |
926 | } |
927 | ||
928 | /* Configuration may specify what MAC to use. Otherwise random. */ | |
c45a6816 | 929 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) { |
a586d4f6 RR |
930 | vdev->config->get(vdev, |
931 | offsetof(struct virtio_net_config, mac), | |
932 | dev->dev_addr, dev->addr_len); | |
62994b2d | 933 | } else |
296f96fc RR |
934 | random_ether_addr(dev->dev_addr); |
935 | ||
936 | /* Set up our device-specific information */ | |
937 | vi = netdev_priv(dev); | |
6c0cd7c0 | 938 | netif_napi_add(dev, &vi->napi, virtnet_poll, napi_weight); |
296f96fc RR |
939 | vi->dev = dev; |
940 | vi->vdev = vdev; | |
d9d5dcc8 | 941 | vdev->priv = vi; |
fb6813f4 | 942 | vi->pages = NULL; |
3161e453 | 943 | INIT_DELAYED_WORK(&vi->refill, refill_work); |
5e01d2f9 MT |
944 | sg_init_table(vi->rx_sg, ARRAY_SIZE(vi->rx_sg)); |
945 | sg_init_table(vi->tx_sg, ARRAY_SIZE(vi->tx_sg)); | |
296f96fc | 946 | |
97402b96 | 947 | /* If we can receive ANY GSO packets, we must allocate large ones. */ |
8e95a202 JP |
948 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) || |
949 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) || | |
950 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN)) | |
97402b96 HX |
951 | vi->big_packets = true; |
952 | ||
3f2c31d9 MM |
953 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF)) |
954 | vi->mergeable_rx_bufs = true; | |
955 | ||
d2a7ddda MT |
956 | /* We expect two virtqueues, receive then send, |
957 | * and optionally control. */ | |
958 | nvqs = virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ? 3 : 2; | |
959 | ||
960 | err = vdev->config->find_vqs(vdev, nvqs, vqs, callbacks, names); | |
961 | if (err) | |
296f96fc | 962 | goto free; |
296f96fc | 963 | |
d2a7ddda MT |
964 | vi->rvq = vqs[0]; |
965 | vi->svq = vqs[1]; | |
296f96fc | 966 | |
2a41f71d | 967 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)) { |
d2a7ddda | 968 | vi->cvq = vqs[2]; |
0bde9569 AW |
969 | |
970 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN)) | |
971 | dev->features |= NETIF_F_HW_VLAN_FILTER; | |
2a41f71d AW |
972 | } |
973 | ||
296f96fc RR |
974 | err = register_netdev(dev); |
975 | if (err) { | |
976 | pr_debug("virtio_net: registering device failed\n"); | |
d2a7ddda | 977 | goto free_vqs; |
296f96fc | 978 | } |
b3369c1f RR |
979 | |
980 | /* Last of all, set up some receive buffers. */ | |
3161e453 | 981 | try_fill_recv(vi, GFP_KERNEL); |
b3369c1f RR |
982 | |
983 | /* If we didn't even get one input buffer, we're useless. */ | |
984 | if (vi->num == 0) { | |
985 | err = -ENOMEM; | |
986 | goto unregister; | |
987 | } | |
988 | ||
167c25e4 JW |
989 | /* Assume link up if device can't report link status, |
990 | otherwise get link status from config. */ | |
991 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) { | |
992 | netif_carrier_off(dev); | |
993 | virtnet_update_status(vi); | |
994 | } else { | |
995 | vi->status = VIRTIO_NET_S_LINK_UP; | |
996 | netif_carrier_on(dev); | |
997 | } | |
9f4d26d0 | 998 | |
296f96fc | 999 | pr_debug("virtnet: registered device %s\n", dev->name); |
296f96fc RR |
1000 | return 0; |
1001 | ||
b3369c1f RR |
1002 | unregister: |
1003 | unregister_netdev(dev); | |
3161e453 | 1004 | cancel_delayed_work_sync(&vi->refill); |
d2a7ddda MT |
1005 | free_vqs: |
1006 | vdev->config->del_vqs(vdev); | |
296f96fc RR |
1007 | free: |
1008 | free_netdev(dev); | |
1009 | return err; | |
1010 | } | |
1011 | ||
9ab86bbc SM |
1012 | static void free_unused_bufs(struct virtnet_info *vi) |
1013 | { | |
1014 | void *buf; | |
830a8a97 | 1015 | while (1) { |
1915a712 | 1016 | buf = virtqueue_detach_unused_buf(vi->svq); |
830a8a97 SM |
1017 | if (!buf) |
1018 | break; | |
1019 | dev_kfree_skb(buf); | |
1020 | } | |
9ab86bbc | 1021 | while (1) { |
1915a712 | 1022 | buf = virtqueue_detach_unused_buf(vi->rvq); |
9ab86bbc SM |
1023 | if (!buf) |
1024 | break; | |
1025 | if (vi->mergeable_rx_bufs || vi->big_packets) | |
1026 | give_pages(vi, buf); | |
1027 | else | |
1028 | dev_kfree_skb(buf); | |
1029 | --vi->num; | |
1030 | } | |
1031 | BUG_ON(vi->num != 0); | |
1032 | } | |
1033 | ||
3d1285be | 1034 | static void __devexit virtnet_remove(struct virtio_device *vdev) |
296f96fc | 1035 | { |
74b2553f | 1036 | struct virtnet_info *vi = vdev->priv; |
b3369c1f | 1037 | |
6e5aa7ef RR |
1038 | /* Stop all the virtqueues. */ |
1039 | vdev->config->reset(vdev); | |
1040 | ||
b3369c1f | 1041 | |
74b2553f | 1042 | unregister_netdev(vi->dev); |
3161e453 | 1043 | cancel_delayed_work_sync(&vi->refill); |
830a8a97 SM |
1044 | |
1045 | /* Free unused buffers in both send and recv, if any. */ | |
9ab86bbc | 1046 | free_unused_bufs(vi); |
fb6813f4 | 1047 | |
d2a7ddda MT |
1048 | vdev->config->del_vqs(vi->vdev); |
1049 | ||
fb6813f4 RR |
1050 | while (vi->pages) |
1051 | __free_pages(get_a_page(vi, GFP_KERNEL), 0); | |
1052 | ||
74b2553f | 1053 | free_netdev(vi->dev); |
296f96fc RR |
1054 | } |
1055 | ||
1056 | static struct virtio_device_id id_table[] = { | |
1057 | { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID }, | |
1058 | { 0 }, | |
1059 | }; | |
1060 | ||
c45a6816 | 1061 | static unsigned int features[] = { |
5e4fe5c4 MM |
1062 | VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM, |
1063 | VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC, | |
c45a6816 | 1064 | VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6, |
97402b96 | 1065 | VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, |
5c516751 | 1066 | VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO, |
2a41f71d | 1067 | VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ, |
0bde9569 | 1068 | VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, |
c45a6816 RR |
1069 | }; |
1070 | ||
22402529 | 1071 | static struct virtio_driver virtio_net_driver = { |
c45a6816 RR |
1072 | .feature_table = features, |
1073 | .feature_table_size = ARRAY_SIZE(features), | |
296f96fc RR |
1074 | .driver.name = KBUILD_MODNAME, |
1075 | .driver.owner = THIS_MODULE, | |
1076 | .id_table = id_table, | |
1077 | .probe = virtnet_probe, | |
1078 | .remove = __devexit_p(virtnet_remove), | |
9f4d26d0 | 1079 | .config_changed = virtnet_config_changed, |
296f96fc RR |
1080 | }; |
1081 | ||
1082 | static int __init init(void) | |
1083 | { | |
22402529 | 1084 | return register_virtio_driver(&virtio_net_driver); |
296f96fc RR |
1085 | } |
1086 | ||
1087 | static void __exit fini(void) | |
1088 | { | |
22402529 | 1089 | unregister_virtio_driver(&virtio_net_driver); |
296f96fc RR |
1090 | } |
1091 | module_init(init); | |
1092 | module_exit(fini); | |
1093 | ||
1094 | MODULE_DEVICE_TABLE(virtio, id_table); | |
1095 | MODULE_DESCRIPTION("Virtio network driver"); | |
1096 | MODULE_LICENSE("GPL"); |