]>
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 |
66846048 | 42 | #define VIRTNET_DRIVER_VERSION "1.0.0" |
2a41f71d | 43 | |
3fa2a1df | 44 | struct virtnet_stats { |
45 | struct u64_stats_sync syncp; | |
46 | u64 tx_bytes; | |
47 | u64 tx_packets; | |
48 | ||
49 | u64 rx_bytes; | |
50 | u64 rx_packets; | |
51 | }; | |
52 | ||
5e01d2f9 | 53 | struct virtnet_info { |
296f96fc | 54 | struct virtio_device *vdev; |
2a41f71d | 55 | struct virtqueue *rvq, *svq, *cvq; |
296f96fc RR |
56 | struct net_device *dev; |
57 | struct napi_struct napi; | |
9f4d26d0 | 58 | unsigned int status; |
296f96fc RR |
59 | |
60 | /* Number of input buffers, and max we've ever had. */ | |
61 | unsigned int num, max; | |
62 | ||
97402b96 HX |
63 | /* I like... big packets and I cannot lie! */ |
64 | bool big_packets; | |
65 | ||
3f2c31d9 MM |
66 | /* Host will merge rx buffers for big packets (shake it! shake it!) */ |
67 | bool mergeable_rx_bufs; | |
68 | ||
3fa2a1df | 69 | /* Active statistics */ |
70 | struct virtnet_stats __percpu *stats; | |
71 | ||
3161e453 RR |
72 | /* Work struct for refilling if we run low on memory. */ |
73 | struct delayed_work refill; | |
74 | ||
fb6813f4 RR |
75 | /* Chain pages by the private ptr. */ |
76 | struct page *pages; | |
5e01d2f9 MT |
77 | |
78 | /* fragments + linear part + virtio header */ | |
79 | struct scatterlist rx_sg[MAX_SKB_FRAGS + 2]; | |
80 | struct scatterlist tx_sg[MAX_SKB_FRAGS + 2]; | |
296f96fc RR |
81 | }; |
82 | ||
b3f24698 RR |
83 | struct skb_vnet_hdr { |
84 | union { | |
85 | struct virtio_net_hdr hdr; | |
86 | struct virtio_net_hdr_mrg_rxbuf mhdr; | |
87 | }; | |
48925e37 | 88 | unsigned int num_sg; |
b3f24698 RR |
89 | }; |
90 | ||
9ab86bbc SM |
91 | struct padded_vnet_hdr { |
92 | struct virtio_net_hdr hdr; | |
93 | /* | |
94 | * virtio_net_hdr should be in a separated sg buffer because of a | |
95 | * QEMU bug, and data sg buffer shares same page with this header sg. | |
96 | * This padding makes next sg 16 byte aligned after virtio_net_hdr. | |
97 | */ | |
98 | char padding[6]; | |
99 | }; | |
100 | ||
b3f24698 | 101 | static inline struct skb_vnet_hdr *skb_vnet_hdr(struct sk_buff *skb) |
296f96fc | 102 | { |
b3f24698 | 103 | return (struct skb_vnet_hdr *)skb->cb; |
296f96fc RR |
104 | } |
105 | ||
9ab86bbc SM |
106 | /* |
107 | * private is used to chain pages for big packets, put the whole | |
108 | * most recent used list in the beginning for reuse | |
109 | */ | |
110 | static void give_pages(struct virtnet_info *vi, struct page *page) | |
0a888fd1 | 111 | { |
9ab86bbc | 112 | struct page *end; |
0a888fd1 | 113 | |
9ab86bbc SM |
114 | /* Find end of list, sew whole thing into vi->pages. */ |
115 | for (end = page; end->private; end = (struct page *)end->private); | |
116 | end->private = (unsigned long)vi->pages; | |
117 | vi->pages = page; | |
0a888fd1 MM |
118 | } |
119 | ||
fb6813f4 RR |
120 | static struct page *get_a_page(struct virtnet_info *vi, gfp_t gfp_mask) |
121 | { | |
122 | struct page *p = vi->pages; | |
123 | ||
9ab86bbc | 124 | if (p) { |
fb6813f4 | 125 | vi->pages = (struct page *)p->private; |
9ab86bbc SM |
126 | /* clear private here, it is used to chain pages */ |
127 | p->private = 0; | |
128 | } else | |
fb6813f4 RR |
129 | p = alloc_page(gfp_mask); |
130 | return p; | |
131 | } | |
132 | ||
2cb9c6ba | 133 | static void skb_xmit_done(struct virtqueue *svq) |
296f96fc | 134 | { |
2cb9c6ba | 135 | struct virtnet_info *vi = svq->vdev->priv; |
296f96fc | 136 | |
2cb9c6ba | 137 | /* Suppress further interrupts. */ |
1915a712 | 138 | virtqueue_disable_cb(svq); |
11a3a154 | 139 | |
363f1514 | 140 | /* We were probably waiting for more output buffers. */ |
296f96fc | 141 | netif_wake_queue(vi->dev); |
296f96fc RR |
142 | } |
143 | ||
9ab86bbc SM |
144 | static void set_skb_frag(struct sk_buff *skb, struct page *page, |
145 | unsigned int offset, unsigned int *len) | |
296f96fc | 146 | { |
8a59a7b9 | 147 | int size = min((unsigned)PAGE_SIZE - offset, *len); |
9ab86bbc | 148 | int i = skb_shinfo(skb)->nr_frags; |
9ab86bbc | 149 | |
8a59a7b9 | 150 | __skb_fill_page_desc(skb, i, page, offset, size); |
9ab86bbc | 151 | |
8a59a7b9 KK |
152 | skb->data_len += size; |
153 | skb->len += size; | |
4b727361 | 154 | skb->truesize += PAGE_SIZE; |
9ab86bbc | 155 | skb_shinfo(skb)->nr_frags++; |
8a59a7b9 | 156 | *len -= size; |
9ab86bbc | 157 | } |
23cde76d | 158 | |
9ab86bbc SM |
159 | static struct sk_buff *page_to_skb(struct virtnet_info *vi, |
160 | struct page *page, unsigned int len) | |
161 | { | |
162 | struct sk_buff *skb; | |
163 | struct skb_vnet_hdr *hdr; | |
164 | unsigned int copy, hdr_len, offset; | |
165 | char *p; | |
fb6813f4 | 166 | |
9ab86bbc | 167 | p = page_address(page); |
3f2c31d9 | 168 | |
9ab86bbc SM |
169 | /* copy small packet so we can reuse these pages for small data */ |
170 | skb = netdev_alloc_skb_ip_align(vi->dev, GOOD_COPY_LEN); | |
171 | if (unlikely(!skb)) | |
172 | return NULL; | |
3f2c31d9 | 173 | |
9ab86bbc | 174 | hdr = skb_vnet_hdr(skb); |
3f2c31d9 | 175 | |
9ab86bbc SM |
176 | if (vi->mergeable_rx_bufs) { |
177 | hdr_len = sizeof hdr->mhdr; | |
178 | offset = hdr_len; | |
179 | } else { | |
180 | hdr_len = sizeof hdr->hdr; | |
181 | offset = sizeof(struct padded_vnet_hdr); | |
182 | } | |
3f2c31d9 | 183 | |
9ab86bbc | 184 | memcpy(hdr, p, hdr_len); |
3f2c31d9 | 185 | |
9ab86bbc SM |
186 | len -= hdr_len; |
187 | p += offset; | |
3f2c31d9 | 188 | |
9ab86bbc SM |
189 | copy = len; |
190 | if (copy > skb_tailroom(skb)) | |
191 | copy = skb_tailroom(skb); | |
192 | memcpy(skb_put(skb, copy), p, copy); | |
3f2c31d9 | 193 | |
9ab86bbc SM |
194 | len -= copy; |
195 | offset += copy; | |
3f2c31d9 | 196 | |
e878d78b SL |
197 | /* |
198 | * Verify that we can indeed put this data into a skb. | |
199 | * This is here to handle cases when the device erroneously | |
200 | * tries to receive more than is possible. This is usually | |
201 | * the case of a broken device. | |
202 | */ | |
203 | if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) { | |
204 | if (net_ratelimit()) | |
205 | pr_debug("%s: too much data\n", skb->dev->name); | |
206 | dev_kfree_skb(skb); | |
207 | return NULL; | |
208 | } | |
209 | ||
9ab86bbc SM |
210 | while (len) { |
211 | set_skb_frag(skb, page, offset, &len); | |
212 | page = (struct page *)page->private; | |
213 | offset = 0; | |
214 | } | |
3f2c31d9 | 215 | |
9ab86bbc SM |
216 | if (page) |
217 | give_pages(vi, page); | |
3f2c31d9 | 218 | |
9ab86bbc SM |
219 | return skb; |
220 | } | |
3f2c31d9 | 221 | |
9ab86bbc SM |
222 | static int receive_mergeable(struct virtnet_info *vi, struct sk_buff *skb) |
223 | { | |
224 | struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb); | |
225 | struct page *page; | |
226 | int num_buf, i, len; | |
227 | ||
228 | num_buf = hdr->mhdr.num_buffers; | |
229 | while (--num_buf) { | |
230 | i = skb_shinfo(skb)->nr_frags; | |
231 | if (i >= MAX_SKB_FRAGS) { | |
232 | pr_debug("%s: packet too long\n", skb->dev->name); | |
233 | skb->dev->stats.rx_length_errors++; | |
234 | return -EINVAL; | |
235 | } | |
1915a712 | 236 | page = virtqueue_get_buf(vi->rvq, &len); |
9ab86bbc SM |
237 | if (!page) { |
238 | pr_debug("%s: rx error: %d buffers missing\n", | |
239 | skb->dev->name, hdr->mhdr.num_buffers); | |
240 | skb->dev->stats.rx_length_errors++; | |
241 | return -EINVAL; | |
3f2c31d9 | 242 | } |
3fa2a1df | 243 | |
9ab86bbc SM |
244 | if (len > PAGE_SIZE) |
245 | len = PAGE_SIZE; | |
246 | ||
247 | set_skb_frag(skb, page, 0, &len); | |
248 | ||
249 | --vi->num; | |
250 | } | |
251 | return 0; | |
252 | } | |
253 | ||
254 | static void receive_buf(struct net_device *dev, void *buf, unsigned int len) | |
255 | { | |
256 | struct virtnet_info *vi = netdev_priv(dev); | |
3fa2a1df | 257 | struct virtnet_stats __percpu *stats = this_cpu_ptr(vi->stats); |
9ab86bbc SM |
258 | struct sk_buff *skb; |
259 | struct page *page; | |
260 | struct skb_vnet_hdr *hdr; | |
3f2c31d9 | 261 | |
9ab86bbc SM |
262 | if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) { |
263 | pr_debug("%s: short packet %i\n", dev->name, len); | |
264 | dev->stats.rx_length_errors++; | |
265 | if (vi->mergeable_rx_bufs || vi->big_packets) | |
266 | give_pages(vi, buf); | |
267 | else | |
268 | dev_kfree_skb(buf); | |
269 | return; | |
270 | } | |
3f2c31d9 | 271 | |
9ab86bbc SM |
272 | if (!vi->mergeable_rx_bufs && !vi->big_packets) { |
273 | skb = buf; | |
274 | len -= sizeof(struct virtio_net_hdr); | |
275 | skb_trim(skb, len); | |
276 | } else { | |
277 | page = buf; | |
278 | skb = page_to_skb(vi, page, len); | |
279 | if (unlikely(!skb)) { | |
3f2c31d9 | 280 | dev->stats.rx_dropped++; |
9ab86bbc SM |
281 | give_pages(vi, page); |
282 | return; | |
3f2c31d9 | 283 | } |
9ab86bbc SM |
284 | if (vi->mergeable_rx_bufs) |
285 | if (receive_mergeable(vi, skb)) { | |
286 | dev_kfree_skb(skb); | |
287 | return; | |
288 | } | |
97402b96 | 289 | } |
3f2c31d9 | 290 | |
9ab86bbc | 291 | hdr = skb_vnet_hdr(skb); |
3fa2a1df | 292 | |
293 | u64_stats_update_begin(&stats->syncp); | |
294 | stats->rx_bytes += skb->len; | |
295 | stats->rx_packets++; | |
296 | u64_stats_update_end(&stats->syncp); | |
296f96fc | 297 | |
b3f24698 | 298 | if (hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { |
296f96fc | 299 | pr_debug("Needs csum!\n"); |
b3f24698 RR |
300 | if (!skb_partial_csum_set(skb, |
301 | hdr->hdr.csum_start, | |
302 | hdr->hdr.csum_offset)) | |
296f96fc | 303 | goto frame_err; |
10a8d94a JW |
304 | } else if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) { |
305 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
296f96fc RR |
306 | } |
307 | ||
23cde76d MM |
308 | skb->protocol = eth_type_trans(skb, dev); |
309 | pr_debug("Receiving skb proto 0x%04x len %i type %i\n", | |
310 | ntohs(skb->protocol), skb->len, skb->pkt_type); | |
311 | ||
b3f24698 | 312 | if (hdr->hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) { |
296f96fc | 313 | pr_debug("GSO!\n"); |
b3f24698 | 314 | switch (hdr->hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { |
296f96fc RR |
315 | case VIRTIO_NET_HDR_GSO_TCPV4: |
316 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; | |
317 | break; | |
296f96fc RR |
318 | case VIRTIO_NET_HDR_GSO_UDP: |
319 | skb_shinfo(skb)->gso_type = SKB_GSO_UDP; | |
320 | break; | |
321 | case VIRTIO_NET_HDR_GSO_TCPV6: | |
322 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6; | |
323 | break; | |
324 | default: | |
325 | if (net_ratelimit()) | |
326 | printk(KERN_WARNING "%s: bad gso type %u.\n", | |
b3f24698 | 327 | dev->name, hdr->hdr.gso_type); |
296f96fc RR |
328 | goto frame_err; |
329 | } | |
330 | ||
b3f24698 | 331 | if (hdr->hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN) |
34a48579 RR |
332 | skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN; |
333 | ||
b3f24698 | 334 | skb_shinfo(skb)->gso_size = hdr->hdr.gso_size; |
296f96fc RR |
335 | if (skb_shinfo(skb)->gso_size == 0) { |
336 | if (net_ratelimit()) | |
337 | printk(KERN_WARNING "%s: zero gso size.\n", | |
338 | dev->name); | |
339 | goto frame_err; | |
340 | } | |
341 | ||
342 | /* Header must be checked, and gso_segs computed. */ | |
343 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
344 | skb_shinfo(skb)->gso_segs = 0; | |
345 | } | |
346 | ||
347 | netif_receive_skb(skb); | |
348 | return; | |
349 | ||
350 | frame_err: | |
351 | dev->stats.rx_frame_errors++; | |
296f96fc RR |
352 | dev_kfree_skb(skb); |
353 | } | |
354 | ||
9ab86bbc | 355 | static int add_recvbuf_small(struct virtnet_info *vi, gfp_t gfp) |
296f96fc RR |
356 | { |
357 | struct sk_buff *skb; | |
9ab86bbc | 358 | struct skb_vnet_hdr *hdr; |
9ab86bbc | 359 | int err; |
3f2c31d9 | 360 | |
9ab86bbc SM |
361 | skb = netdev_alloc_skb_ip_align(vi->dev, MAX_PACKET_LEN); |
362 | if (unlikely(!skb)) | |
363 | return -ENOMEM; | |
296f96fc | 364 | |
9ab86bbc | 365 | skb_put(skb, MAX_PACKET_LEN); |
3f2c31d9 | 366 | |
9ab86bbc | 367 | hdr = skb_vnet_hdr(skb); |
5e01d2f9 | 368 | sg_set_buf(vi->rx_sg, &hdr->hdr, sizeof hdr->hdr); |
97402b96 | 369 | |
5e01d2f9 | 370 | skb_to_sgvec(skb, vi->rx_sg + 1, 0, skb->len); |
97402b96 | 371 | |
aa989f5e | 372 | err = virtqueue_add_buf_gfp(vi->rvq, vi->rx_sg, 0, 2, skb, gfp); |
9ab86bbc SM |
373 | if (err < 0) |
374 | dev_kfree_skb(skb); | |
97402b96 | 375 | |
9ab86bbc SM |
376 | return err; |
377 | } | |
97402b96 | 378 | |
9ab86bbc SM |
379 | static int add_recvbuf_big(struct virtnet_info *vi, gfp_t gfp) |
380 | { | |
9ab86bbc SM |
381 | struct page *first, *list = NULL; |
382 | char *p; | |
383 | int i, err, offset; | |
384 | ||
5e01d2f9 | 385 | /* page in vi->rx_sg[MAX_SKB_FRAGS + 1] is list tail */ |
9ab86bbc SM |
386 | for (i = MAX_SKB_FRAGS + 1; i > 1; --i) { |
387 | first = get_a_page(vi, gfp); | |
388 | if (!first) { | |
389 | if (list) | |
390 | give_pages(vi, list); | |
391 | return -ENOMEM; | |
97402b96 | 392 | } |
5e01d2f9 | 393 | sg_set_buf(&vi->rx_sg[i], page_address(first), PAGE_SIZE); |
97402b96 | 394 | |
9ab86bbc SM |
395 | /* chain new page in list head to match sg */ |
396 | first->private = (unsigned long)list; | |
397 | list = first; | |
398 | } | |
296f96fc | 399 | |
9ab86bbc SM |
400 | first = get_a_page(vi, gfp); |
401 | if (!first) { | |
402 | give_pages(vi, list); | |
403 | return -ENOMEM; | |
404 | } | |
405 | p = page_address(first); | |
406 | ||
5e01d2f9 MT |
407 | /* vi->rx_sg[0], vi->rx_sg[1] share the same page */ |
408 | /* a separated vi->rx_sg[0] for virtio_net_hdr only due to QEMU bug */ | |
409 | sg_set_buf(&vi->rx_sg[0], p, sizeof(struct virtio_net_hdr)); | |
9ab86bbc | 410 | |
5e01d2f9 | 411 | /* vi->rx_sg[1] for data packet, from offset */ |
9ab86bbc | 412 | offset = sizeof(struct padded_vnet_hdr); |
5e01d2f9 | 413 | sg_set_buf(&vi->rx_sg[1], p + offset, PAGE_SIZE - offset); |
9ab86bbc SM |
414 | |
415 | /* chain first in list head */ | |
416 | first->private = (unsigned long)list; | |
aa989f5e MT |
417 | err = virtqueue_add_buf_gfp(vi->rvq, vi->rx_sg, 0, MAX_SKB_FRAGS + 2, |
418 | first, gfp); | |
9ab86bbc SM |
419 | if (err < 0) |
420 | give_pages(vi, first); | |
421 | ||
422 | return err; | |
296f96fc RR |
423 | } |
424 | ||
9ab86bbc | 425 | static int add_recvbuf_mergeable(struct virtnet_info *vi, gfp_t gfp) |
3f2c31d9 | 426 | { |
9ab86bbc | 427 | struct page *page; |
3f2c31d9 | 428 | int err; |
3f2c31d9 | 429 | |
9ab86bbc SM |
430 | page = get_a_page(vi, gfp); |
431 | if (!page) | |
432 | return -ENOMEM; | |
3f2c31d9 | 433 | |
5e01d2f9 | 434 | sg_init_one(vi->rx_sg, page_address(page), PAGE_SIZE); |
3f2c31d9 | 435 | |
aa989f5e | 436 | err = virtqueue_add_buf_gfp(vi->rvq, vi->rx_sg, 0, 1, page, gfp); |
9ab86bbc SM |
437 | if (err < 0) |
438 | give_pages(vi, page); | |
3f2c31d9 | 439 | |
9ab86bbc SM |
440 | return err; |
441 | } | |
3f2c31d9 | 442 | |
9ab86bbc SM |
443 | /* Returns false if we couldn't fill entirely (OOM). */ |
444 | static bool try_fill_recv(struct virtnet_info *vi, gfp_t gfp) | |
445 | { | |
446 | int err; | |
1788f495 | 447 | bool oom; |
3f2c31d9 | 448 | |
9ab86bbc SM |
449 | do { |
450 | if (vi->mergeable_rx_bufs) | |
451 | err = add_recvbuf_mergeable(vi, gfp); | |
452 | else if (vi->big_packets) | |
453 | err = add_recvbuf_big(vi, gfp); | |
454 | else | |
455 | err = add_recvbuf_small(vi, gfp); | |
3f2c31d9 | 456 | |
1788f495 MT |
457 | oom = err == -ENOMEM; |
458 | if (err < 0) | |
3f2c31d9 | 459 | break; |
9ab86bbc | 460 | ++vi->num; |
0aea51c3 | 461 | } while (err > 0); |
3f2c31d9 MM |
462 | if (unlikely(vi->num > vi->max)) |
463 | vi->max = vi->num; | |
1915a712 | 464 | virtqueue_kick(vi->rvq); |
3161e453 | 465 | return !oom; |
3f2c31d9 MM |
466 | } |
467 | ||
18445c4d | 468 | static void skb_recv_done(struct virtqueue *rvq) |
296f96fc RR |
469 | { |
470 | struct virtnet_info *vi = rvq->vdev->priv; | |
18445c4d | 471 | /* Schedule NAPI, Suppress further interrupts if successful. */ |
288379f0 | 472 | if (napi_schedule_prep(&vi->napi)) { |
1915a712 | 473 | virtqueue_disable_cb(rvq); |
288379f0 | 474 | __napi_schedule(&vi->napi); |
18445c4d | 475 | } |
296f96fc RR |
476 | } |
477 | ||
3e9d08ec BR |
478 | static void virtnet_napi_enable(struct virtnet_info *vi) |
479 | { | |
480 | napi_enable(&vi->napi); | |
481 | ||
482 | /* If all buffers were filled by other side before we napi_enabled, we | |
483 | * won't get another interrupt, so process any outstanding packets | |
484 | * now. virtnet_poll wants re-enable the queue, so we disable here. | |
485 | * We synchronize against interrupts via NAPI_STATE_SCHED */ | |
486 | if (napi_schedule_prep(&vi->napi)) { | |
487 | virtqueue_disable_cb(vi->rvq); | |
488 | __napi_schedule(&vi->napi); | |
489 | } | |
490 | } | |
491 | ||
3161e453 RR |
492 | static void refill_work(struct work_struct *work) |
493 | { | |
494 | struct virtnet_info *vi; | |
495 | bool still_empty; | |
496 | ||
497 | vi = container_of(work, struct virtnet_info, refill.work); | |
498 | napi_disable(&vi->napi); | |
39d32157 | 499 | still_empty = !try_fill_recv(vi, GFP_KERNEL); |
3e9d08ec | 500 | virtnet_napi_enable(vi); |
3161e453 RR |
501 | |
502 | /* In theory, this can happen: if we don't get any buffers in | |
503 | * we will *never* try to fill again. */ | |
504 | if (still_empty) | |
505 | schedule_delayed_work(&vi->refill, HZ/2); | |
506 | } | |
507 | ||
296f96fc RR |
508 | static int virtnet_poll(struct napi_struct *napi, int budget) |
509 | { | |
510 | struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi); | |
9ab86bbc | 511 | void *buf; |
296f96fc RR |
512 | unsigned int len, received = 0; |
513 | ||
514 | again: | |
515 | while (received < budget && | |
1915a712 | 516 | (buf = virtqueue_get_buf(vi->rvq, &len)) != NULL) { |
9ab86bbc SM |
517 | receive_buf(vi->dev, buf, len); |
518 | --vi->num; | |
296f96fc RR |
519 | received++; |
520 | } | |
521 | ||
3161e453 RR |
522 | if (vi->num < vi->max / 2) { |
523 | if (!try_fill_recv(vi, GFP_ATOMIC)) | |
524 | schedule_delayed_work(&vi->refill, 0); | |
525 | } | |
296f96fc | 526 | |
8329d98e RR |
527 | /* Out of packets? */ |
528 | if (received < budget) { | |
288379f0 | 529 | napi_complete(napi); |
1915a712 | 530 | if (unlikely(!virtqueue_enable_cb(vi->rvq)) && |
8e95a202 | 531 | napi_schedule_prep(napi)) { |
1915a712 | 532 | virtqueue_disable_cb(vi->rvq); |
288379f0 | 533 | __napi_schedule(napi); |
296f96fc | 534 | goto again; |
4265f161 | 535 | } |
296f96fc RR |
536 | } |
537 | ||
538 | return received; | |
539 | } | |
540 | ||
48925e37 | 541 | static unsigned int free_old_xmit_skbs(struct virtnet_info *vi) |
296f96fc RR |
542 | { |
543 | struct sk_buff *skb; | |
48925e37 | 544 | unsigned int len, tot_sgs = 0; |
3fa2a1df | 545 | struct virtnet_stats __percpu *stats = this_cpu_ptr(vi->stats); |
296f96fc | 546 | |
1915a712 | 547 | while ((skb = virtqueue_get_buf(vi->svq, &len)) != NULL) { |
296f96fc | 548 | pr_debug("Sent skb %p\n", skb); |
3fa2a1df | 549 | |
550 | u64_stats_update_begin(&stats->syncp); | |
551 | stats->tx_bytes += skb->len; | |
552 | stats->tx_packets++; | |
553 | u64_stats_update_end(&stats->syncp); | |
554 | ||
48925e37 | 555 | tot_sgs += skb_vnet_hdr(skb)->num_sg; |
ed79bab8 | 556 | dev_kfree_skb_any(skb); |
296f96fc | 557 | } |
48925e37 | 558 | return tot_sgs; |
296f96fc RR |
559 | } |
560 | ||
99ffc696 | 561 | static int xmit_skb(struct virtnet_info *vi, struct sk_buff *skb) |
296f96fc | 562 | { |
b3f24698 | 563 | struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb); |
296f96fc | 564 | const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest; |
296f96fc | 565 | |
e174961c | 566 | pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest); |
296f96fc | 567 | |
296f96fc | 568 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
b3f24698 | 569 | hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; |
55508d60 | 570 | hdr->hdr.csum_start = skb_checksum_start_offset(skb); |
b3f24698 | 571 | hdr->hdr.csum_offset = skb->csum_offset; |
296f96fc | 572 | } else { |
b3f24698 RR |
573 | hdr->hdr.flags = 0; |
574 | hdr->hdr.csum_offset = hdr->hdr.csum_start = 0; | |
296f96fc RR |
575 | } |
576 | ||
577 | if (skb_is_gso(skb)) { | |
b3f24698 RR |
578 | hdr->hdr.hdr_len = skb_headlen(skb); |
579 | hdr->hdr.gso_size = skb_shinfo(skb)->gso_size; | |
34a48579 | 580 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) |
b3f24698 | 581 | hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4; |
296f96fc | 582 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) |
b3f24698 | 583 | hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6; |
296f96fc | 584 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP) |
b3f24698 | 585 | hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP; |
296f96fc RR |
586 | else |
587 | BUG(); | |
34a48579 | 588 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN) |
b3f24698 | 589 | hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN; |
296f96fc | 590 | } else { |
b3f24698 RR |
591 | hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE; |
592 | hdr->hdr.gso_size = hdr->hdr.hdr_len = 0; | |
296f96fc RR |
593 | } |
594 | ||
b3f24698 | 595 | hdr->mhdr.num_buffers = 0; |
3f2c31d9 MM |
596 | |
597 | /* Encode metadata header at front. */ | |
598 | if (vi->mergeable_rx_bufs) | |
5e01d2f9 | 599 | sg_set_buf(vi->tx_sg, &hdr->mhdr, sizeof hdr->mhdr); |
3f2c31d9 | 600 | else |
5e01d2f9 | 601 | sg_set_buf(vi->tx_sg, &hdr->hdr, sizeof hdr->hdr); |
3f2c31d9 | 602 | |
5e01d2f9 | 603 | hdr->num_sg = skb_to_sgvec(skb, vi->tx_sg + 1, 0, skb->len) + 1; |
1756ac3d | 604 | return virtqueue_add_buf(vi->svq, vi->tx_sg, hdr->num_sg, |
5e01d2f9 | 605 | 0, skb); |
11a3a154 RR |
606 | } |
607 | ||
424efe9c | 608 | static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev) |
99ffc696 RR |
609 | { |
610 | struct virtnet_info *vi = netdev_priv(dev); | |
48925e37 | 611 | int capacity; |
2cb9c6ba | 612 | |
2cb9c6ba RR |
613 | /* Free up any pending old buffers before queueing new ones. */ |
614 | free_old_xmit_skbs(vi); | |
99ffc696 | 615 | |
03f191ba | 616 | /* Try to transmit */ |
48925e37 RR |
617 | capacity = xmit_skb(vi, skb); |
618 | ||
619 | /* This can happen with OOM and indirect buffers. */ | |
620 | if (unlikely(capacity < 0)) { | |
58eba97d RR |
621 | if (net_ratelimit()) { |
622 | if (likely(capacity == -ENOMEM)) { | |
623 | dev_warn(&dev->dev, | |
624 | "TX queue failure: out of memory\n"); | |
625 | } else { | |
626 | dev->stats.tx_fifo_errors++; | |
627 | dev_warn(&dev->dev, | |
628 | "Unexpected TX queue failure: %d\n", | |
629 | capacity); | |
630 | } | |
48925e37 | 631 | } |
58eba97d RR |
632 | dev->stats.tx_dropped++; |
633 | kfree_skb(skb); | |
634 | return NETDEV_TX_OK; | |
296f96fc | 635 | } |
1915a712 | 636 | virtqueue_kick(vi->svq); |
03f191ba | 637 | |
48925e37 RR |
638 | /* Don't wait up for transmitted skbs to be freed. */ |
639 | skb_orphan(skb); | |
640 | nf_reset(skb); | |
641 | ||
642 | /* Apparently nice girls don't return TX_BUSY; stop the queue | |
643 | * before it gets out of hand. Naturally, this wastes entries. */ | |
644 | if (capacity < 2+MAX_SKB_FRAGS) { | |
645 | netif_stop_queue(dev); | |
7a66f784 | 646 | if (unlikely(!virtqueue_enable_cb_delayed(vi->svq))) { |
48925e37 RR |
647 | /* More just got used, free them then recheck. */ |
648 | capacity += free_old_xmit_skbs(vi); | |
649 | if (capacity >= 2+MAX_SKB_FRAGS) { | |
650 | netif_start_queue(dev); | |
1915a712 | 651 | virtqueue_disable_cb(vi->svq); |
48925e37 RR |
652 | } |
653 | } | |
99ffc696 | 654 | } |
48925e37 RR |
655 | |
656 | return NETDEV_TX_OK; | |
296f96fc RR |
657 | } |
658 | ||
9c46f6d4 AW |
659 | static int virtnet_set_mac_address(struct net_device *dev, void *p) |
660 | { | |
661 | struct virtnet_info *vi = netdev_priv(dev); | |
662 | struct virtio_device *vdev = vi->vdev; | |
663 | int ret; | |
664 | ||
665 | ret = eth_mac_addr(dev, p); | |
666 | if (ret) | |
667 | return ret; | |
668 | ||
62994b2d AW |
669 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) |
670 | vdev->config->set(vdev, offsetof(struct virtio_net_config, mac), | |
671 | dev->dev_addr, dev->addr_len); | |
9c46f6d4 AW |
672 | |
673 | return 0; | |
674 | } | |
675 | ||
3fa2a1df | 676 | static struct rtnl_link_stats64 *virtnet_stats(struct net_device *dev, |
677 | struct rtnl_link_stats64 *tot) | |
678 | { | |
679 | struct virtnet_info *vi = netdev_priv(dev); | |
680 | int cpu; | |
681 | unsigned int start; | |
682 | ||
683 | for_each_possible_cpu(cpu) { | |
684 | struct virtnet_stats __percpu *stats | |
685 | = per_cpu_ptr(vi->stats, cpu); | |
686 | u64 tpackets, tbytes, rpackets, rbytes; | |
687 | ||
688 | do { | |
689 | start = u64_stats_fetch_begin(&stats->syncp); | |
690 | tpackets = stats->tx_packets; | |
691 | tbytes = stats->tx_bytes; | |
692 | rpackets = stats->rx_packets; | |
693 | rbytes = stats->rx_bytes; | |
694 | } while (u64_stats_fetch_retry(&stats->syncp, start)); | |
695 | ||
696 | tot->rx_packets += rpackets; | |
697 | tot->tx_packets += tpackets; | |
698 | tot->rx_bytes += rbytes; | |
699 | tot->tx_bytes += tbytes; | |
700 | } | |
701 | ||
702 | tot->tx_dropped = dev->stats.tx_dropped; | |
703 | tot->rx_dropped = dev->stats.rx_dropped; | |
704 | tot->rx_length_errors = dev->stats.rx_length_errors; | |
705 | tot->rx_frame_errors = dev->stats.rx_frame_errors; | |
706 | ||
707 | return tot; | |
708 | } | |
709 | ||
da74e89d AS |
710 | #ifdef CONFIG_NET_POLL_CONTROLLER |
711 | static void virtnet_netpoll(struct net_device *dev) | |
712 | { | |
713 | struct virtnet_info *vi = netdev_priv(dev); | |
714 | ||
715 | napi_schedule(&vi->napi); | |
716 | } | |
717 | #endif | |
718 | ||
296f96fc RR |
719 | static int virtnet_open(struct net_device *dev) |
720 | { | |
721 | struct virtnet_info *vi = netdev_priv(dev); | |
722 | ||
3e9d08ec | 723 | virtnet_napi_enable(vi); |
296f96fc RR |
724 | return 0; |
725 | } | |
726 | ||
2a41f71d AW |
727 | /* |
728 | * Send command via the control virtqueue and check status. Commands | |
729 | * supported by the hypervisor, as indicated by feature bits, should | |
730 | * never fail unless improperly formated. | |
731 | */ | |
732 | static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd, | |
733 | struct scatterlist *data, int out, int in) | |
734 | { | |
23e258e1 | 735 | struct scatterlist *s, sg[VIRTNET_SEND_COMMAND_SG_MAX + 2]; |
2a41f71d AW |
736 | struct virtio_net_ctrl_hdr ctrl; |
737 | virtio_net_ctrl_ack status = ~0; | |
738 | unsigned int tmp; | |
23e258e1 | 739 | int i; |
2a41f71d | 740 | |
0ee904c3 AB |
741 | /* Caller should know better */ |
742 | BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) || | |
743 | (out + in > VIRTNET_SEND_COMMAND_SG_MAX)); | |
2a41f71d AW |
744 | |
745 | out++; /* Add header */ | |
746 | in++; /* Add return status */ | |
747 | ||
748 | ctrl.class = class; | |
749 | ctrl.cmd = cmd; | |
750 | ||
751 | sg_init_table(sg, out + in); | |
752 | ||
753 | sg_set_buf(&sg[0], &ctrl, sizeof(ctrl)); | |
23e258e1 AW |
754 | for_each_sg(data, s, out + in - 2, i) |
755 | sg_set_buf(&sg[i + 1], sg_virt(s), s->length); | |
2a41f71d AW |
756 | sg_set_buf(&sg[out + in - 1], &status, sizeof(status)); |
757 | ||
1915a712 | 758 | BUG_ON(virtqueue_add_buf(vi->cvq, sg, out, in, vi) < 0); |
2a41f71d | 759 | |
1915a712 | 760 | virtqueue_kick(vi->cvq); |
2a41f71d AW |
761 | |
762 | /* | |
763 | * Spin for a response, the kick causes an ioport write, trapping | |
764 | * into the hypervisor, so the request should be handled immediately. | |
765 | */ | |
1915a712 | 766 | while (!virtqueue_get_buf(vi->cvq, &tmp)) |
2a41f71d AW |
767 | cpu_relax(); |
768 | ||
769 | return status == VIRTIO_NET_OK; | |
770 | } | |
771 | ||
296f96fc RR |
772 | static int virtnet_close(struct net_device *dev) |
773 | { | |
774 | struct virtnet_info *vi = netdev_priv(dev); | |
296f96fc RR |
775 | |
776 | napi_disable(&vi->napi); | |
777 | ||
296f96fc RR |
778 | return 0; |
779 | } | |
780 | ||
2af7698e AW |
781 | static void virtnet_set_rx_mode(struct net_device *dev) |
782 | { | |
783 | struct virtnet_info *vi = netdev_priv(dev); | |
f565a7c2 | 784 | struct scatterlist sg[2]; |
2af7698e | 785 | u8 promisc, allmulti; |
f565a7c2 | 786 | struct virtio_net_ctrl_mac *mac_data; |
ccffad25 | 787 | struct netdev_hw_addr *ha; |
32e7bfc4 | 788 | int uc_count; |
4cd24eaf | 789 | int mc_count; |
f565a7c2 AW |
790 | void *buf; |
791 | int i; | |
2af7698e AW |
792 | |
793 | /* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */ | |
794 | if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX)) | |
795 | return; | |
796 | ||
f565a7c2 AW |
797 | promisc = ((dev->flags & IFF_PROMISC) != 0); |
798 | allmulti = ((dev->flags & IFF_ALLMULTI) != 0); | |
2af7698e | 799 | |
23e258e1 | 800 | sg_init_one(sg, &promisc, sizeof(promisc)); |
2af7698e AW |
801 | |
802 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, | |
803 | VIRTIO_NET_CTRL_RX_PROMISC, | |
f565a7c2 | 804 | sg, 1, 0)) |
2af7698e AW |
805 | dev_warn(&dev->dev, "Failed to %sable promisc mode.\n", |
806 | promisc ? "en" : "dis"); | |
807 | ||
23e258e1 | 808 | sg_init_one(sg, &allmulti, sizeof(allmulti)); |
2af7698e AW |
809 | |
810 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, | |
811 | VIRTIO_NET_CTRL_RX_ALLMULTI, | |
f565a7c2 | 812 | sg, 1, 0)) |
2af7698e AW |
813 | dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n", |
814 | allmulti ? "en" : "dis"); | |
f565a7c2 | 815 | |
32e7bfc4 | 816 | uc_count = netdev_uc_count(dev); |
4cd24eaf | 817 | mc_count = netdev_mc_count(dev); |
f565a7c2 | 818 | /* MAC filter - use one buffer for both lists */ |
4cd24eaf JP |
819 | buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) + |
820 | (2 * sizeof(mac_data->entries)), GFP_ATOMIC); | |
821 | mac_data = buf; | |
f565a7c2 AW |
822 | if (!buf) { |
823 | dev_warn(&dev->dev, "No memory for MAC address buffer\n"); | |
824 | return; | |
825 | } | |
826 | ||
23e258e1 AW |
827 | sg_init_table(sg, 2); |
828 | ||
f565a7c2 | 829 | /* Store the unicast list and count in the front of the buffer */ |
32e7bfc4 | 830 | mac_data->entries = uc_count; |
ccffad25 | 831 | i = 0; |
32e7bfc4 | 832 | netdev_for_each_uc_addr(ha, dev) |
ccffad25 | 833 | memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); |
f565a7c2 AW |
834 | |
835 | sg_set_buf(&sg[0], mac_data, | |
32e7bfc4 | 836 | sizeof(mac_data->entries) + (uc_count * ETH_ALEN)); |
f565a7c2 AW |
837 | |
838 | /* multicast list and count fill the end */ | |
32e7bfc4 | 839 | mac_data = (void *)&mac_data->macs[uc_count][0]; |
f565a7c2 | 840 | |
4cd24eaf | 841 | mac_data->entries = mc_count; |
567ec874 | 842 | i = 0; |
22bedad3 JP |
843 | netdev_for_each_mc_addr(ha, dev) |
844 | memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); | |
f565a7c2 AW |
845 | |
846 | sg_set_buf(&sg[1], mac_data, | |
4cd24eaf | 847 | sizeof(mac_data->entries) + (mc_count * ETH_ALEN)); |
f565a7c2 AW |
848 | |
849 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC, | |
850 | VIRTIO_NET_CTRL_MAC_TABLE_SET, | |
851 | sg, 2, 0)) | |
852 | dev_warn(&dev->dev, "Failed to set MAC fitler table.\n"); | |
853 | ||
854 | kfree(buf); | |
2af7698e AW |
855 | } |
856 | ||
1824a989 | 857 | static void virtnet_vlan_rx_add_vid(struct net_device *dev, u16 vid) |
0bde9569 AW |
858 | { |
859 | struct virtnet_info *vi = netdev_priv(dev); | |
860 | struct scatterlist sg; | |
861 | ||
23e258e1 | 862 | sg_init_one(&sg, &vid, sizeof(vid)); |
0bde9569 AW |
863 | |
864 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, | |
865 | VIRTIO_NET_CTRL_VLAN_ADD, &sg, 1, 0)) | |
866 | dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid); | |
867 | } | |
868 | ||
1824a989 | 869 | static void virtnet_vlan_rx_kill_vid(struct net_device *dev, u16 vid) |
0bde9569 AW |
870 | { |
871 | struct virtnet_info *vi = netdev_priv(dev); | |
872 | struct scatterlist sg; | |
873 | ||
23e258e1 | 874 | sg_init_one(&sg, &vid, sizeof(vid)); |
0bde9569 AW |
875 | |
876 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, | |
877 | VIRTIO_NET_CTRL_VLAN_DEL, &sg, 1, 0)) | |
878 | dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid); | |
879 | } | |
880 | ||
8f9f4668 RJ |
881 | static void virtnet_get_ringparam(struct net_device *dev, |
882 | struct ethtool_ringparam *ring) | |
883 | { | |
884 | struct virtnet_info *vi = netdev_priv(dev); | |
885 | ||
886 | ring->rx_max_pending = virtqueue_get_vring_size(vi->rvq); | |
887 | ring->tx_max_pending = virtqueue_get_vring_size(vi->svq); | |
888 | ring->rx_pending = ring->rx_max_pending; | |
889 | ring->tx_pending = ring->tx_max_pending; | |
890 | ||
891 | } | |
892 | ||
66846048 RJ |
893 | |
894 | static void virtnet_get_drvinfo(struct net_device *dev, | |
895 | struct ethtool_drvinfo *info) | |
896 | { | |
897 | struct virtnet_info *vi = netdev_priv(dev); | |
898 | struct virtio_device *vdev = vi->vdev; | |
899 | ||
900 | strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver)); | |
901 | strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version)); | |
902 | strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info)); | |
903 | ||
904 | } | |
905 | ||
0fc0b732 | 906 | static const struct ethtool_ops virtnet_ethtool_ops = { |
66846048 | 907 | .get_drvinfo = virtnet_get_drvinfo, |
9f4d26d0 | 908 | .get_link = ethtool_op_get_link, |
8f9f4668 | 909 | .get_ringparam = virtnet_get_ringparam, |
a9ea3fc6 HX |
910 | }; |
911 | ||
39da5814 MM |
912 | #define MIN_MTU 68 |
913 | #define MAX_MTU 65535 | |
914 | ||
915 | static int virtnet_change_mtu(struct net_device *dev, int new_mtu) | |
916 | { | |
917 | if (new_mtu < MIN_MTU || new_mtu > MAX_MTU) | |
918 | return -EINVAL; | |
919 | dev->mtu = new_mtu; | |
920 | return 0; | |
921 | } | |
922 | ||
76288b4e SH |
923 | static const struct net_device_ops virtnet_netdev = { |
924 | .ndo_open = virtnet_open, | |
925 | .ndo_stop = virtnet_close, | |
926 | .ndo_start_xmit = start_xmit, | |
927 | .ndo_validate_addr = eth_validate_addr, | |
9c46f6d4 | 928 | .ndo_set_mac_address = virtnet_set_mac_address, |
2af7698e | 929 | .ndo_set_rx_mode = virtnet_set_rx_mode, |
76288b4e | 930 | .ndo_change_mtu = virtnet_change_mtu, |
3fa2a1df | 931 | .ndo_get_stats64 = virtnet_stats, |
1824a989 AW |
932 | .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid, |
933 | .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid, | |
76288b4e SH |
934 | #ifdef CONFIG_NET_POLL_CONTROLLER |
935 | .ndo_poll_controller = virtnet_netpoll, | |
936 | #endif | |
937 | }; | |
938 | ||
9f4d26d0 MM |
939 | static void virtnet_update_status(struct virtnet_info *vi) |
940 | { | |
941 | u16 v; | |
942 | ||
77dd7693 | 943 | if (virtio_config_val(vi->vdev, VIRTIO_NET_F_STATUS, |
9f4d26d0 | 944 | offsetof(struct virtio_net_config, status), |
77dd7693 SL |
945 | &v) < 0) |
946 | return; | |
9f4d26d0 MM |
947 | |
948 | /* Ignore unknown (future) status bits */ | |
949 | v &= VIRTIO_NET_S_LINK_UP; | |
950 | ||
951 | if (vi->status == v) | |
952 | return; | |
953 | ||
954 | vi->status = v; | |
955 | ||
956 | if (vi->status & VIRTIO_NET_S_LINK_UP) { | |
957 | netif_carrier_on(vi->dev); | |
958 | netif_wake_queue(vi->dev); | |
959 | } else { | |
960 | netif_carrier_off(vi->dev); | |
961 | netif_stop_queue(vi->dev); | |
962 | } | |
963 | } | |
964 | ||
965 | static void virtnet_config_changed(struct virtio_device *vdev) | |
966 | { | |
967 | struct virtnet_info *vi = vdev->priv; | |
968 | ||
969 | virtnet_update_status(vi); | |
970 | } | |
971 | ||
296f96fc RR |
972 | static int virtnet_probe(struct virtio_device *vdev) |
973 | { | |
974 | int err; | |
296f96fc RR |
975 | struct net_device *dev; |
976 | struct virtnet_info *vi; | |
d2a7ddda MT |
977 | struct virtqueue *vqs[3]; |
978 | vq_callback_t *callbacks[] = { skb_recv_done, skb_xmit_done, NULL}; | |
979 | const char *names[] = { "input", "output", "control" }; | |
980 | int nvqs; | |
296f96fc RR |
981 | |
982 | /* Allocate ourselves a network device with room for our info */ | |
983 | dev = alloc_etherdev(sizeof(struct virtnet_info)); | |
984 | if (!dev) | |
985 | return -ENOMEM; | |
986 | ||
987 | /* Set up network device as normal. */ | |
01789349 | 988 | dev->priv_flags |= IFF_UNICAST_FLT; |
76288b4e | 989 | dev->netdev_ops = &virtnet_netdev; |
296f96fc | 990 | dev->features = NETIF_F_HIGHDMA; |
3fa2a1df | 991 | |
a9ea3fc6 | 992 | SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops); |
296f96fc RR |
993 | SET_NETDEV_DEV(dev, &vdev->dev); |
994 | ||
995 | /* Do we support "hardware" checksums? */ | |
98e778c9 | 996 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) { |
296f96fc | 997 | /* This opens up the world of extra features. */ |
98e778c9 MM |
998 | dev->hw_features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST; |
999 | if (csum) | |
1000 | dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST; | |
1001 | ||
1002 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) { | |
1003 | dev->hw_features |= NETIF_F_TSO | NETIF_F_UFO | |
34a48579 RR |
1004 | | NETIF_F_TSO_ECN | NETIF_F_TSO6; |
1005 | } | |
5539ae96 | 1006 | /* Individual feature bits: what can host handle? */ |
98e778c9 MM |
1007 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4)) |
1008 | dev->hw_features |= NETIF_F_TSO; | |
1009 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6)) | |
1010 | dev->hw_features |= NETIF_F_TSO6; | |
1011 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN)) | |
1012 | dev->hw_features |= NETIF_F_TSO_ECN; | |
1013 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO)) | |
1014 | dev->hw_features |= NETIF_F_UFO; | |
1015 | ||
1016 | if (gso) | |
1017 | dev->features |= dev->hw_features & (NETIF_F_ALL_TSO|NETIF_F_UFO); | |
1018 | /* (!csum && gso) case will be fixed by register_netdev() */ | |
296f96fc RR |
1019 | } |
1020 | ||
1021 | /* Configuration may specify what MAC to use. Otherwise random. */ | |
77dd7693 | 1022 | if (virtio_config_val_len(vdev, VIRTIO_NET_F_MAC, |
a586d4f6 | 1023 | offsetof(struct virtio_net_config, mac), |
77dd7693 | 1024 | dev->dev_addr, dev->addr_len) < 0) |
296f96fc RR |
1025 | random_ether_addr(dev->dev_addr); |
1026 | ||
1027 | /* Set up our device-specific information */ | |
1028 | vi = netdev_priv(dev); | |
6c0cd7c0 | 1029 | netif_napi_add(dev, &vi->napi, virtnet_poll, napi_weight); |
296f96fc RR |
1030 | vi->dev = dev; |
1031 | vi->vdev = vdev; | |
d9d5dcc8 | 1032 | vdev->priv = vi; |
fb6813f4 | 1033 | vi->pages = NULL; |
3fa2a1df | 1034 | vi->stats = alloc_percpu(struct virtnet_stats); |
1035 | err = -ENOMEM; | |
1036 | if (vi->stats == NULL) | |
1037 | goto free; | |
1038 | ||
3161e453 | 1039 | INIT_DELAYED_WORK(&vi->refill, refill_work); |
5e01d2f9 MT |
1040 | sg_init_table(vi->rx_sg, ARRAY_SIZE(vi->rx_sg)); |
1041 | sg_init_table(vi->tx_sg, ARRAY_SIZE(vi->tx_sg)); | |
296f96fc | 1042 | |
97402b96 | 1043 | /* If we can receive ANY GSO packets, we must allocate large ones. */ |
8e95a202 JP |
1044 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) || |
1045 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) || | |
1046 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN)) | |
97402b96 HX |
1047 | vi->big_packets = true; |
1048 | ||
3f2c31d9 MM |
1049 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF)) |
1050 | vi->mergeable_rx_bufs = true; | |
1051 | ||
d2a7ddda MT |
1052 | /* We expect two virtqueues, receive then send, |
1053 | * and optionally control. */ | |
1054 | nvqs = virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ? 3 : 2; | |
1055 | ||
1056 | err = vdev->config->find_vqs(vdev, nvqs, vqs, callbacks, names); | |
1057 | if (err) | |
3fa2a1df | 1058 | goto free_stats; |
296f96fc | 1059 | |
d2a7ddda MT |
1060 | vi->rvq = vqs[0]; |
1061 | vi->svq = vqs[1]; | |
296f96fc | 1062 | |
2a41f71d | 1063 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)) { |
d2a7ddda | 1064 | vi->cvq = vqs[2]; |
0bde9569 AW |
1065 | |
1066 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN)) | |
1067 | dev->features |= NETIF_F_HW_VLAN_FILTER; | |
2a41f71d AW |
1068 | } |
1069 | ||
296f96fc RR |
1070 | err = register_netdev(dev); |
1071 | if (err) { | |
1072 | pr_debug("virtio_net: registering device failed\n"); | |
d2a7ddda | 1073 | goto free_vqs; |
296f96fc | 1074 | } |
b3369c1f RR |
1075 | |
1076 | /* Last of all, set up some receive buffers. */ | |
3161e453 | 1077 | try_fill_recv(vi, GFP_KERNEL); |
b3369c1f RR |
1078 | |
1079 | /* If we didn't even get one input buffer, we're useless. */ | |
1080 | if (vi->num == 0) { | |
1081 | err = -ENOMEM; | |
1082 | goto unregister; | |
1083 | } | |
1084 | ||
167c25e4 JW |
1085 | /* Assume link up if device can't report link status, |
1086 | otherwise get link status from config. */ | |
1087 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) { | |
1088 | netif_carrier_off(dev); | |
1089 | virtnet_update_status(vi); | |
1090 | } else { | |
1091 | vi->status = VIRTIO_NET_S_LINK_UP; | |
1092 | netif_carrier_on(dev); | |
1093 | } | |
9f4d26d0 | 1094 | |
296f96fc | 1095 | pr_debug("virtnet: registered device %s\n", dev->name); |
296f96fc RR |
1096 | return 0; |
1097 | ||
b3369c1f RR |
1098 | unregister: |
1099 | unregister_netdev(dev); | |
3161e453 | 1100 | cancel_delayed_work_sync(&vi->refill); |
d2a7ddda MT |
1101 | free_vqs: |
1102 | vdev->config->del_vqs(vdev); | |
3fa2a1df | 1103 | free_stats: |
1104 | free_percpu(vi->stats); | |
296f96fc RR |
1105 | free: |
1106 | free_netdev(dev); | |
1107 | return err; | |
1108 | } | |
1109 | ||
9ab86bbc SM |
1110 | static void free_unused_bufs(struct virtnet_info *vi) |
1111 | { | |
1112 | void *buf; | |
830a8a97 | 1113 | while (1) { |
1915a712 | 1114 | buf = virtqueue_detach_unused_buf(vi->svq); |
830a8a97 SM |
1115 | if (!buf) |
1116 | break; | |
1117 | dev_kfree_skb(buf); | |
1118 | } | |
9ab86bbc | 1119 | while (1) { |
1915a712 | 1120 | buf = virtqueue_detach_unused_buf(vi->rvq); |
9ab86bbc SM |
1121 | if (!buf) |
1122 | break; | |
1123 | if (vi->mergeable_rx_bufs || vi->big_packets) | |
1124 | give_pages(vi, buf); | |
1125 | else | |
1126 | dev_kfree_skb(buf); | |
1127 | --vi->num; | |
1128 | } | |
1129 | BUG_ON(vi->num != 0); | |
1130 | } | |
1131 | ||
3d1285be | 1132 | static void __devexit virtnet_remove(struct virtio_device *vdev) |
296f96fc | 1133 | { |
74b2553f | 1134 | struct virtnet_info *vi = vdev->priv; |
b3369c1f | 1135 | |
6e5aa7ef RR |
1136 | /* Stop all the virtqueues. */ |
1137 | vdev->config->reset(vdev); | |
1138 | ||
b3369c1f | 1139 | |
74b2553f | 1140 | unregister_netdev(vi->dev); |
3161e453 | 1141 | cancel_delayed_work_sync(&vi->refill); |
830a8a97 SM |
1142 | |
1143 | /* Free unused buffers in both send and recv, if any. */ | |
9ab86bbc | 1144 | free_unused_bufs(vi); |
fb6813f4 | 1145 | |
d2a7ddda MT |
1146 | vdev->config->del_vqs(vi->vdev); |
1147 | ||
fb6813f4 RR |
1148 | while (vi->pages) |
1149 | __free_pages(get_a_page(vi, GFP_KERNEL), 0); | |
1150 | ||
2e66f55b | 1151 | free_percpu(vi->stats); |
74b2553f | 1152 | free_netdev(vi->dev); |
296f96fc RR |
1153 | } |
1154 | ||
1155 | static struct virtio_device_id id_table[] = { | |
1156 | { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID }, | |
1157 | { 0 }, | |
1158 | }; | |
1159 | ||
c45a6816 | 1160 | static unsigned int features[] = { |
5e4fe5c4 MM |
1161 | VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM, |
1162 | VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC, | |
c45a6816 | 1163 | VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6, |
97402b96 | 1164 | VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, |
5c516751 | 1165 | VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO, |
2a41f71d | 1166 | VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ, |
0bde9569 | 1167 | VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, |
c45a6816 RR |
1168 | }; |
1169 | ||
22402529 | 1170 | static struct virtio_driver virtio_net_driver = { |
c45a6816 RR |
1171 | .feature_table = features, |
1172 | .feature_table_size = ARRAY_SIZE(features), | |
296f96fc RR |
1173 | .driver.name = KBUILD_MODNAME, |
1174 | .driver.owner = THIS_MODULE, | |
1175 | .id_table = id_table, | |
1176 | .probe = virtnet_probe, | |
1177 | .remove = __devexit_p(virtnet_remove), | |
9f4d26d0 | 1178 | .config_changed = virtnet_config_changed, |
296f96fc RR |
1179 | }; |
1180 | ||
1181 | static int __init init(void) | |
1182 | { | |
22402529 | 1183 | return register_virtio_driver(&virtio_net_driver); |
296f96fc RR |
1184 | } |
1185 | ||
1186 | static void __exit fini(void) | |
1187 | { | |
22402529 | 1188 | unregister_virtio_driver(&virtio_net_driver); |
296f96fc RR |
1189 | } |
1190 | module_init(init); | |
1191 | module_exit(fini); | |
1192 | ||
1193 | MODULE_DEVICE_TABLE(virtio, id_table); | |
1194 | MODULE_DESCRIPTION("Virtio network driver"); | |
1195 | MODULE_LICENSE("GPL"); |