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 | ||
eb939922 | 33 | static bool csum = true, gso = true; |
34a48579 RR |
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 | ||
586d17c5 JW |
69 | /* enable config space updates */ |
70 | bool config_enable; | |
71 | ||
3fa2a1df | 72 | /* Active statistics */ |
73 | struct virtnet_stats __percpu *stats; | |
74 | ||
3161e453 RR |
75 | /* Work struct for refilling if we run low on memory. */ |
76 | struct delayed_work refill; | |
77 | ||
586d17c5 JW |
78 | /* Work struct for config space updates */ |
79 | struct work_struct config_work; | |
80 | ||
81 | /* Lock for config space updates */ | |
82 | struct mutex config_lock; | |
83 | ||
fb6813f4 RR |
84 | /* Chain pages by the private ptr. */ |
85 | struct page *pages; | |
5e01d2f9 MT |
86 | |
87 | /* fragments + linear part + virtio header */ | |
88 | struct scatterlist rx_sg[MAX_SKB_FRAGS + 2]; | |
89 | struct scatterlist tx_sg[MAX_SKB_FRAGS + 2]; | |
296f96fc RR |
90 | }; |
91 | ||
b3f24698 RR |
92 | struct skb_vnet_hdr { |
93 | union { | |
94 | struct virtio_net_hdr hdr; | |
95 | struct virtio_net_hdr_mrg_rxbuf mhdr; | |
96 | }; | |
48925e37 | 97 | unsigned int num_sg; |
b3f24698 RR |
98 | }; |
99 | ||
9ab86bbc SM |
100 | struct padded_vnet_hdr { |
101 | struct virtio_net_hdr hdr; | |
102 | /* | |
103 | * virtio_net_hdr should be in a separated sg buffer because of a | |
104 | * QEMU bug, and data sg buffer shares same page with this header sg. | |
105 | * This padding makes next sg 16 byte aligned after virtio_net_hdr. | |
106 | */ | |
107 | char padding[6]; | |
108 | }; | |
109 | ||
b3f24698 | 110 | static inline struct skb_vnet_hdr *skb_vnet_hdr(struct sk_buff *skb) |
296f96fc | 111 | { |
b3f24698 | 112 | return (struct skb_vnet_hdr *)skb->cb; |
296f96fc RR |
113 | } |
114 | ||
9ab86bbc SM |
115 | /* |
116 | * private is used to chain pages for big packets, put the whole | |
117 | * most recent used list in the beginning for reuse | |
118 | */ | |
119 | static void give_pages(struct virtnet_info *vi, struct page *page) | |
0a888fd1 | 120 | { |
9ab86bbc | 121 | struct page *end; |
0a888fd1 | 122 | |
9ab86bbc SM |
123 | /* Find end of list, sew whole thing into vi->pages. */ |
124 | for (end = page; end->private; end = (struct page *)end->private); | |
125 | end->private = (unsigned long)vi->pages; | |
126 | vi->pages = page; | |
0a888fd1 MM |
127 | } |
128 | ||
fb6813f4 RR |
129 | static struct page *get_a_page(struct virtnet_info *vi, gfp_t gfp_mask) |
130 | { | |
131 | struct page *p = vi->pages; | |
132 | ||
9ab86bbc | 133 | if (p) { |
fb6813f4 | 134 | vi->pages = (struct page *)p->private; |
9ab86bbc SM |
135 | /* clear private here, it is used to chain pages */ |
136 | p->private = 0; | |
137 | } else | |
fb6813f4 RR |
138 | p = alloc_page(gfp_mask); |
139 | return p; | |
140 | } | |
141 | ||
2cb9c6ba | 142 | static void skb_xmit_done(struct virtqueue *svq) |
296f96fc | 143 | { |
2cb9c6ba | 144 | struct virtnet_info *vi = svq->vdev->priv; |
296f96fc | 145 | |
2cb9c6ba | 146 | /* Suppress further interrupts. */ |
1915a712 | 147 | virtqueue_disable_cb(svq); |
11a3a154 | 148 | |
363f1514 | 149 | /* We were probably waiting for more output buffers. */ |
296f96fc | 150 | netif_wake_queue(vi->dev); |
296f96fc RR |
151 | } |
152 | ||
9ab86bbc SM |
153 | static void set_skb_frag(struct sk_buff *skb, struct page *page, |
154 | unsigned int offset, unsigned int *len) | |
296f96fc | 155 | { |
8a59a7b9 | 156 | int size = min((unsigned)PAGE_SIZE - offset, *len); |
9ab86bbc | 157 | int i = skb_shinfo(skb)->nr_frags; |
9ab86bbc | 158 | |
8a59a7b9 | 159 | __skb_fill_page_desc(skb, i, page, offset, size); |
9ab86bbc | 160 | |
8a59a7b9 KK |
161 | skb->data_len += size; |
162 | skb->len += size; | |
4b727361 | 163 | skb->truesize += PAGE_SIZE; |
9ab86bbc | 164 | skb_shinfo(skb)->nr_frags++; |
8a59a7b9 | 165 | *len -= size; |
9ab86bbc | 166 | } |
23cde76d | 167 | |
3464645a | 168 | /* Called from bottom half context */ |
9ab86bbc SM |
169 | static struct sk_buff *page_to_skb(struct virtnet_info *vi, |
170 | struct page *page, unsigned int len) | |
171 | { | |
172 | struct sk_buff *skb; | |
173 | struct skb_vnet_hdr *hdr; | |
174 | unsigned int copy, hdr_len, offset; | |
175 | char *p; | |
fb6813f4 | 176 | |
9ab86bbc | 177 | p = page_address(page); |
3f2c31d9 | 178 | |
9ab86bbc SM |
179 | /* copy small packet so we can reuse these pages for small data */ |
180 | skb = netdev_alloc_skb_ip_align(vi->dev, GOOD_COPY_LEN); | |
181 | if (unlikely(!skb)) | |
182 | return NULL; | |
3f2c31d9 | 183 | |
9ab86bbc | 184 | hdr = skb_vnet_hdr(skb); |
3f2c31d9 | 185 | |
9ab86bbc SM |
186 | if (vi->mergeable_rx_bufs) { |
187 | hdr_len = sizeof hdr->mhdr; | |
188 | offset = hdr_len; | |
189 | } else { | |
190 | hdr_len = sizeof hdr->hdr; | |
191 | offset = sizeof(struct padded_vnet_hdr); | |
192 | } | |
3f2c31d9 | 193 | |
9ab86bbc | 194 | memcpy(hdr, p, hdr_len); |
3f2c31d9 | 195 | |
9ab86bbc SM |
196 | len -= hdr_len; |
197 | p += offset; | |
3f2c31d9 | 198 | |
9ab86bbc SM |
199 | copy = len; |
200 | if (copy > skb_tailroom(skb)) | |
201 | copy = skb_tailroom(skb); | |
202 | memcpy(skb_put(skb, copy), p, copy); | |
3f2c31d9 | 203 | |
9ab86bbc SM |
204 | len -= copy; |
205 | offset += copy; | |
3f2c31d9 | 206 | |
e878d78b SL |
207 | /* |
208 | * Verify that we can indeed put this data into a skb. | |
209 | * This is here to handle cases when the device erroneously | |
210 | * tries to receive more than is possible. This is usually | |
211 | * the case of a broken device. | |
212 | */ | |
213 | if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) { | |
214 | if (net_ratelimit()) | |
215 | pr_debug("%s: too much data\n", skb->dev->name); | |
216 | dev_kfree_skb(skb); | |
217 | return NULL; | |
218 | } | |
219 | ||
9ab86bbc SM |
220 | while (len) { |
221 | set_skb_frag(skb, page, offset, &len); | |
222 | page = (struct page *)page->private; | |
223 | offset = 0; | |
224 | } | |
3f2c31d9 | 225 | |
9ab86bbc SM |
226 | if (page) |
227 | give_pages(vi, page); | |
3f2c31d9 | 228 | |
9ab86bbc SM |
229 | return skb; |
230 | } | |
3f2c31d9 | 231 | |
9ab86bbc SM |
232 | static int receive_mergeable(struct virtnet_info *vi, struct sk_buff *skb) |
233 | { | |
234 | struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb); | |
235 | struct page *page; | |
236 | int num_buf, i, len; | |
237 | ||
238 | num_buf = hdr->mhdr.num_buffers; | |
239 | while (--num_buf) { | |
240 | i = skb_shinfo(skb)->nr_frags; | |
241 | if (i >= MAX_SKB_FRAGS) { | |
242 | pr_debug("%s: packet too long\n", skb->dev->name); | |
243 | skb->dev->stats.rx_length_errors++; | |
244 | return -EINVAL; | |
245 | } | |
1915a712 | 246 | page = virtqueue_get_buf(vi->rvq, &len); |
9ab86bbc SM |
247 | if (!page) { |
248 | pr_debug("%s: rx error: %d buffers missing\n", | |
249 | skb->dev->name, hdr->mhdr.num_buffers); | |
250 | skb->dev->stats.rx_length_errors++; | |
251 | return -EINVAL; | |
3f2c31d9 | 252 | } |
3fa2a1df | 253 | |
9ab86bbc SM |
254 | if (len > PAGE_SIZE) |
255 | len = PAGE_SIZE; | |
256 | ||
257 | set_skb_frag(skb, page, 0, &len); | |
258 | ||
259 | --vi->num; | |
260 | } | |
261 | return 0; | |
262 | } | |
263 | ||
264 | static void receive_buf(struct net_device *dev, void *buf, unsigned int len) | |
265 | { | |
266 | struct virtnet_info *vi = netdev_priv(dev); | |
58472a76 | 267 | struct virtnet_stats *stats = this_cpu_ptr(vi->stats); |
9ab86bbc SM |
268 | struct sk_buff *skb; |
269 | struct page *page; | |
270 | struct skb_vnet_hdr *hdr; | |
3f2c31d9 | 271 | |
9ab86bbc SM |
272 | if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) { |
273 | pr_debug("%s: short packet %i\n", dev->name, len); | |
274 | dev->stats.rx_length_errors++; | |
275 | if (vi->mergeable_rx_bufs || vi->big_packets) | |
276 | give_pages(vi, buf); | |
277 | else | |
278 | dev_kfree_skb(buf); | |
279 | return; | |
280 | } | |
3f2c31d9 | 281 | |
9ab86bbc SM |
282 | if (!vi->mergeable_rx_bufs && !vi->big_packets) { |
283 | skb = buf; | |
284 | len -= sizeof(struct virtio_net_hdr); | |
285 | skb_trim(skb, len); | |
286 | } else { | |
287 | page = buf; | |
288 | skb = page_to_skb(vi, page, len); | |
289 | if (unlikely(!skb)) { | |
3f2c31d9 | 290 | dev->stats.rx_dropped++; |
9ab86bbc SM |
291 | give_pages(vi, page); |
292 | return; | |
3f2c31d9 | 293 | } |
9ab86bbc SM |
294 | if (vi->mergeable_rx_bufs) |
295 | if (receive_mergeable(vi, skb)) { | |
296 | dev_kfree_skb(skb); | |
297 | return; | |
298 | } | |
97402b96 | 299 | } |
3f2c31d9 | 300 | |
9ab86bbc | 301 | hdr = skb_vnet_hdr(skb); |
3fa2a1df | 302 | |
303 | u64_stats_update_begin(&stats->syncp); | |
304 | stats->rx_bytes += skb->len; | |
305 | stats->rx_packets++; | |
306 | u64_stats_update_end(&stats->syncp); | |
296f96fc | 307 | |
b3f24698 | 308 | if (hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { |
296f96fc | 309 | pr_debug("Needs csum!\n"); |
b3f24698 RR |
310 | if (!skb_partial_csum_set(skb, |
311 | hdr->hdr.csum_start, | |
312 | hdr->hdr.csum_offset)) | |
296f96fc | 313 | goto frame_err; |
10a8d94a JW |
314 | } else if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) { |
315 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
296f96fc RR |
316 | } |
317 | ||
23cde76d MM |
318 | skb->protocol = eth_type_trans(skb, dev); |
319 | pr_debug("Receiving skb proto 0x%04x len %i type %i\n", | |
320 | ntohs(skb->protocol), skb->len, skb->pkt_type); | |
321 | ||
b3f24698 | 322 | if (hdr->hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) { |
296f96fc | 323 | pr_debug("GSO!\n"); |
b3f24698 | 324 | switch (hdr->hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { |
296f96fc RR |
325 | case VIRTIO_NET_HDR_GSO_TCPV4: |
326 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; | |
327 | break; | |
296f96fc RR |
328 | case VIRTIO_NET_HDR_GSO_UDP: |
329 | skb_shinfo(skb)->gso_type = SKB_GSO_UDP; | |
330 | break; | |
331 | case VIRTIO_NET_HDR_GSO_TCPV6: | |
332 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6; | |
333 | break; | |
334 | default: | |
335 | if (net_ratelimit()) | |
336 | printk(KERN_WARNING "%s: bad gso type %u.\n", | |
b3f24698 | 337 | dev->name, hdr->hdr.gso_type); |
296f96fc RR |
338 | goto frame_err; |
339 | } | |
340 | ||
b3f24698 | 341 | if (hdr->hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN) |
34a48579 RR |
342 | skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN; |
343 | ||
b3f24698 | 344 | skb_shinfo(skb)->gso_size = hdr->hdr.gso_size; |
296f96fc RR |
345 | if (skb_shinfo(skb)->gso_size == 0) { |
346 | if (net_ratelimit()) | |
347 | printk(KERN_WARNING "%s: zero gso size.\n", | |
348 | dev->name); | |
349 | goto frame_err; | |
350 | } | |
351 | ||
352 | /* Header must be checked, and gso_segs computed. */ | |
353 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
354 | skb_shinfo(skb)->gso_segs = 0; | |
355 | } | |
356 | ||
357 | netif_receive_skb(skb); | |
358 | return; | |
359 | ||
360 | frame_err: | |
361 | dev->stats.rx_frame_errors++; | |
296f96fc RR |
362 | dev_kfree_skb(skb); |
363 | } | |
364 | ||
9ab86bbc | 365 | static int add_recvbuf_small(struct virtnet_info *vi, gfp_t gfp) |
296f96fc RR |
366 | { |
367 | struct sk_buff *skb; | |
9ab86bbc | 368 | struct skb_vnet_hdr *hdr; |
9ab86bbc | 369 | int err; |
3f2c31d9 | 370 | |
3464645a | 371 | skb = __netdev_alloc_skb_ip_align(vi->dev, MAX_PACKET_LEN, gfp); |
9ab86bbc SM |
372 | if (unlikely(!skb)) |
373 | return -ENOMEM; | |
296f96fc | 374 | |
9ab86bbc | 375 | skb_put(skb, MAX_PACKET_LEN); |
3f2c31d9 | 376 | |
9ab86bbc | 377 | hdr = skb_vnet_hdr(skb); |
5e01d2f9 | 378 | sg_set_buf(vi->rx_sg, &hdr->hdr, sizeof hdr->hdr); |
97402b96 | 379 | |
5e01d2f9 | 380 | skb_to_sgvec(skb, vi->rx_sg + 1, 0, skb->len); |
97402b96 | 381 | |
f96fde41 | 382 | err = virtqueue_add_buf(vi->rvq, vi->rx_sg, 0, 2, skb, gfp); |
9ab86bbc SM |
383 | if (err < 0) |
384 | dev_kfree_skb(skb); | |
97402b96 | 385 | |
9ab86bbc SM |
386 | return err; |
387 | } | |
97402b96 | 388 | |
9ab86bbc SM |
389 | static int add_recvbuf_big(struct virtnet_info *vi, gfp_t gfp) |
390 | { | |
9ab86bbc SM |
391 | struct page *first, *list = NULL; |
392 | char *p; | |
393 | int i, err, offset; | |
394 | ||
5e01d2f9 | 395 | /* page in vi->rx_sg[MAX_SKB_FRAGS + 1] is list tail */ |
9ab86bbc SM |
396 | for (i = MAX_SKB_FRAGS + 1; i > 1; --i) { |
397 | first = get_a_page(vi, gfp); | |
398 | if (!first) { | |
399 | if (list) | |
400 | give_pages(vi, list); | |
401 | return -ENOMEM; | |
97402b96 | 402 | } |
5e01d2f9 | 403 | sg_set_buf(&vi->rx_sg[i], page_address(first), PAGE_SIZE); |
97402b96 | 404 | |
9ab86bbc SM |
405 | /* chain new page in list head to match sg */ |
406 | first->private = (unsigned long)list; | |
407 | list = first; | |
408 | } | |
296f96fc | 409 | |
9ab86bbc SM |
410 | first = get_a_page(vi, gfp); |
411 | if (!first) { | |
412 | give_pages(vi, list); | |
413 | return -ENOMEM; | |
414 | } | |
415 | p = page_address(first); | |
416 | ||
5e01d2f9 MT |
417 | /* vi->rx_sg[0], vi->rx_sg[1] share the same page */ |
418 | /* a separated vi->rx_sg[0] for virtio_net_hdr only due to QEMU bug */ | |
419 | sg_set_buf(&vi->rx_sg[0], p, sizeof(struct virtio_net_hdr)); | |
9ab86bbc | 420 | |
5e01d2f9 | 421 | /* vi->rx_sg[1] for data packet, from offset */ |
9ab86bbc | 422 | offset = sizeof(struct padded_vnet_hdr); |
5e01d2f9 | 423 | sg_set_buf(&vi->rx_sg[1], p + offset, PAGE_SIZE - offset); |
9ab86bbc SM |
424 | |
425 | /* chain first in list head */ | |
426 | first->private = (unsigned long)list; | |
f96fde41 RR |
427 | err = virtqueue_add_buf(vi->rvq, vi->rx_sg, 0, MAX_SKB_FRAGS + 2, |
428 | first, gfp); | |
9ab86bbc SM |
429 | if (err < 0) |
430 | give_pages(vi, first); | |
431 | ||
432 | return err; | |
296f96fc RR |
433 | } |
434 | ||
9ab86bbc | 435 | static int add_recvbuf_mergeable(struct virtnet_info *vi, gfp_t gfp) |
3f2c31d9 | 436 | { |
9ab86bbc | 437 | struct page *page; |
3f2c31d9 | 438 | int err; |
3f2c31d9 | 439 | |
9ab86bbc SM |
440 | page = get_a_page(vi, gfp); |
441 | if (!page) | |
442 | return -ENOMEM; | |
3f2c31d9 | 443 | |
5e01d2f9 | 444 | sg_init_one(vi->rx_sg, page_address(page), PAGE_SIZE); |
3f2c31d9 | 445 | |
f96fde41 | 446 | err = virtqueue_add_buf(vi->rvq, vi->rx_sg, 0, 1, page, gfp); |
9ab86bbc SM |
447 | if (err < 0) |
448 | give_pages(vi, page); | |
3f2c31d9 | 449 | |
9ab86bbc SM |
450 | return err; |
451 | } | |
3f2c31d9 | 452 | |
b2baed69 RR |
453 | /* |
454 | * Returns false if we couldn't fill entirely (OOM). | |
455 | * | |
456 | * Normally run in the receive path, but can also be run from ndo_open | |
457 | * before we're receiving packets, or from refill_work which is | |
458 | * careful to disable receiving (using napi_disable). | |
459 | */ | |
9ab86bbc SM |
460 | static bool try_fill_recv(struct virtnet_info *vi, gfp_t gfp) |
461 | { | |
462 | int err; | |
1788f495 | 463 | bool oom; |
3f2c31d9 | 464 | |
9ab86bbc SM |
465 | do { |
466 | if (vi->mergeable_rx_bufs) | |
467 | err = add_recvbuf_mergeable(vi, gfp); | |
468 | else if (vi->big_packets) | |
469 | err = add_recvbuf_big(vi, gfp); | |
470 | else | |
471 | err = add_recvbuf_small(vi, gfp); | |
3f2c31d9 | 472 | |
1788f495 MT |
473 | oom = err == -ENOMEM; |
474 | if (err < 0) | |
3f2c31d9 | 475 | break; |
9ab86bbc | 476 | ++vi->num; |
0aea51c3 | 477 | } while (err > 0); |
3f2c31d9 MM |
478 | if (unlikely(vi->num > vi->max)) |
479 | vi->max = vi->num; | |
1915a712 | 480 | virtqueue_kick(vi->rvq); |
3161e453 | 481 | return !oom; |
3f2c31d9 MM |
482 | } |
483 | ||
18445c4d | 484 | static void skb_recv_done(struct virtqueue *rvq) |
296f96fc RR |
485 | { |
486 | struct virtnet_info *vi = rvq->vdev->priv; | |
18445c4d | 487 | /* Schedule NAPI, Suppress further interrupts if successful. */ |
288379f0 | 488 | if (napi_schedule_prep(&vi->napi)) { |
1915a712 | 489 | virtqueue_disable_cb(rvq); |
288379f0 | 490 | __napi_schedule(&vi->napi); |
18445c4d | 491 | } |
296f96fc RR |
492 | } |
493 | ||
3e9d08ec BR |
494 | static void virtnet_napi_enable(struct virtnet_info *vi) |
495 | { | |
496 | napi_enable(&vi->napi); | |
497 | ||
498 | /* If all buffers were filled by other side before we napi_enabled, we | |
499 | * won't get another interrupt, so process any outstanding packets | |
500 | * now. virtnet_poll wants re-enable the queue, so we disable here. | |
501 | * We synchronize against interrupts via NAPI_STATE_SCHED */ | |
502 | if (napi_schedule_prep(&vi->napi)) { | |
503 | virtqueue_disable_cb(vi->rvq); | |
504 | __napi_schedule(&vi->napi); | |
505 | } | |
506 | } | |
507 | ||
3161e453 RR |
508 | static void refill_work(struct work_struct *work) |
509 | { | |
510 | struct virtnet_info *vi; | |
511 | bool still_empty; | |
512 | ||
513 | vi = container_of(work, struct virtnet_info, refill.work); | |
514 | napi_disable(&vi->napi); | |
39d32157 | 515 | still_empty = !try_fill_recv(vi, GFP_KERNEL); |
3e9d08ec | 516 | virtnet_napi_enable(vi); |
3161e453 RR |
517 | |
518 | /* In theory, this can happen: if we don't get any buffers in | |
519 | * we will *never* try to fill again. */ | |
520 | if (still_empty) | |
f1776dad | 521 | queue_delayed_work(system_nrt_wq, &vi->refill, HZ/2); |
3161e453 RR |
522 | } |
523 | ||
296f96fc RR |
524 | static int virtnet_poll(struct napi_struct *napi, int budget) |
525 | { | |
526 | struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi); | |
9ab86bbc | 527 | void *buf; |
296f96fc RR |
528 | unsigned int len, received = 0; |
529 | ||
530 | again: | |
531 | while (received < budget && | |
1915a712 | 532 | (buf = virtqueue_get_buf(vi->rvq, &len)) != NULL) { |
9ab86bbc SM |
533 | receive_buf(vi->dev, buf, len); |
534 | --vi->num; | |
296f96fc RR |
535 | received++; |
536 | } | |
537 | ||
3161e453 RR |
538 | if (vi->num < vi->max / 2) { |
539 | if (!try_fill_recv(vi, GFP_ATOMIC)) | |
f1776dad | 540 | queue_delayed_work(system_nrt_wq, &vi->refill, 0); |
3161e453 | 541 | } |
296f96fc | 542 | |
8329d98e RR |
543 | /* Out of packets? */ |
544 | if (received < budget) { | |
288379f0 | 545 | napi_complete(napi); |
1915a712 | 546 | if (unlikely(!virtqueue_enable_cb(vi->rvq)) && |
8e95a202 | 547 | napi_schedule_prep(napi)) { |
1915a712 | 548 | virtqueue_disable_cb(vi->rvq); |
288379f0 | 549 | __napi_schedule(napi); |
296f96fc | 550 | goto again; |
4265f161 | 551 | } |
296f96fc RR |
552 | } |
553 | ||
554 | return received; | |
555 | } | |
556 | ||
48925e37 | 557 | static unsigned int free_old_xmit_skbs(struct virtnet_info *vi) |
296f96fc RR |
558 | { |
559 | struct sk_buff *skb; | |
48925e37 | 560 | unsigned int len, tot_sgs = 0; |
58472a76 | 561 | struct virtnet_stats *stats = this_cpu_ptr(vi->stats); |
296f96fc | 562 | |
1915a712 | 563 | while ((skb = virtqueue_get_buf(vi->svq, &len)) != NULL) { |
296f96fc | 564 | pr_debug("Sent skb %p\n", skb); |
3fa2a1df | 565 | |
566 | u64_stats_update_begin(&stats->syncp); | |
567 | stats->tx_bytes += skb->len; | |
568 | stats->tx_packets++; | |
569 | u64_stats_update_end(&stats->syncp); | |
570 | ||
48925e37 | 571 | tot_sgs += skb_vnet_hdr(skb)->num_sg; |
ed79bab8 | 572 | dev_kfree_skb_any(skb); |
296f96fc | 573 | } |
48925e37 | 574 | return tot_sgs; |
296f96fc RR |
575 | } |
576 | ||
99ffc696 | 577 | static int xmit_skb(struct virtnet_info *vi, struct sk_buff *skb) |
296f96fc | 578 | { |
b3f24698 | 579 | struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb); |
296f96fc | 580 | const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest; |
296f96fc | 581 | |
e174961c | 582 | pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest); |
296f96fc | 583 | |
296f96fc | 584 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
b3f24698 | 585 | hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; |
55508d60 | 586 | hdr->hdr.csum_start = skb_checksum_start_offset(skb); |
b3f24698 | 587 | hdr->hdr.csum_offset = skb->csum_offset; |
296f96fc | 588 | } else { |
b3f24698 RR |
589 | hdr->hdr.flags = 0; |
590 | hdr->hdr.csum_offset = hdr->hdr.csum_start = 0; | |
296f96fc RR |
591 | } |
592 | ||
593 | if (skb_is_gso(skb)) { | |
b3f24698 RR |
594 | hdr->hdr.hdr_len = skb_headlen(skb); |
595 | hdr->hdr.gso_size = skb_shinfo(skb)->gso_size; | |
34a48579 | 596 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) |
b3f24698 | 597 | hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4; |
296f96fc | 598 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) |
b3f24698 | 599 | hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6; |
296f96fc | 600 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP) |
b3f24698 | 601 | hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP; |
296f96fc RR |
602 | else |
603 | BUG(); | |
34a48579 | 604 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN) |
b3f24698 | 605 | hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN; |
296f96fc | 606 | } else { |
b3f24698 RR |
607 | hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE; |
608 | hdr->hdr.gso_size = hdr->hdr.hdr_len = 0; | |
296f96fc RR |
609 | } |
610 | ||
b3f24698 | 611 | hdr->mhdr.num_buffers = 0; |
3f2c31d9 MM |
612 | |
613 | /* Encode metadata header at front. */ | |
614 | if (vi->mergeable_rx_bufs) | |
5e01d2f9 | 615 | sg_set_buf(vi->tx_sg, &hdr->mhdr, sizeof hdr->mhdr); |
3f2c31d9 | 616 | else |
5e01d2f9 | 617 | sg_set_buf(vi->tx_sg, &hdr->hdr, sizeof hdr->hdr); |
3f2c31d9 | 618 | |
5e01d2f9 | 619 | hdr->num_sg = skb_to_sgvec(skb, vi->tx_sg + 1, 0, skb->len) + 1; |
1756ac3d | 620 | return virtqueue_add_buf(vi->svq, vi->tx_sg, hdr->num_sg, |
f96fde41 | 621 | 0, skb, GFP_ATOMIC); |
11a3a154 RR |
622 | } |
623 | ||
424efe9c | 624 | static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev) |
99ffc696 RR |
625 | { |
626 | struct virtnet_info *vi = netdev_priv(dev); | |
48925e37 | 627 | int capacity; |
2cb9c6ba | 628 | |
2cb9c6ba RR |
629 | /* Free up any pending old buffers before queueing new ones. */ |
630 | free_old_xmit_skbs(vi); | |
99ffc696 | 631 | |
03f191ba | 632 | /* Try to transmit */ |
48925e37 RR |
633 | capacity = xmit_skb(vi, skb); |
634 | ||
635 | /* This can happen with OOM and indirect buffers. */ | |
636 | if (unlikely(capacity < 0)) { | |
2e57b79c | 637 | if (likely(capacity == -ENOMEM)) { |
31304165 | 638 | if (net_ratelimit()) |
58eba97d RR |
639 | dev_warn(&dev->dev, |
640 | "TX queue failure: out of memory\n"); | |
31304165 | 641 | } else { |
2e57b79c RJ |
642 | dev->stats.tx_fifo_errors++; |
643 | if (net_ratelimit()) | |
58eba97d RR |
644 | dev_warn(&dev->dev, |
645 | "Unexpected TX queue failure: %d\n", | |
646 | capacity); | |
48925e37 | 647 | } |
58eba97d RR |
648 | dev->stats.tx_dropped++; |
649 | kfree_skb(skb); | |
650 | return NETDEV_TX_OK; | |
296f96fc | 651 | } |
1915a712 | 652 | virtqueue_kick(vi->svq); |
03f191ba | 653 | |
48925e37 RR |
654 | /* Don't wait up for transmitted skbs to be freed. */ |
655 | skb_orphan(skb); | |
656 | nf_reset(skb); | |
657 | ||
658 | /* Apparently nice girls don't return TX_BUSY; stop the queue | |
659 | * before it gets out of hand. Naturally, this wastes entries. */ | |
660 | if (capacity < 2+MAX_SKB_FRAGS) { | |
661 | netif_stop_queue(dev); | |
7a66f784 | 662 | if (unlikely(!virtqueue_enable_cb_delayed(vi->svq))) { |
48925e37 RR |
663 | /* More just got used, free them then recheck. */ |
664 | capacity += free_old_xmit_skbs(vi); | |
665 | if (capacity >= 2+MAX_SKB_FRAGS) { | |
666 | netif_start_queue(dev); | |
1915a712 | 667 | virtqueue_disable_cb(vi->svq); |
48925e37 RR |
668 | } |
669 | } | |
99ffc696 | 670 | } |
48925e37 RR |
671 | |
672 | return NETDEV_TX_OK; | |
296f96fc RR |
673 | } |
674 | ||
9c46f6d4 AW |
675 | static int virtnet_set_mac_address(struct net_device *dev, void *p) |
676 | { | |
677 | struct virtnet_info *vi = netdev_priv(dev); | |
678 | struct virtio_device *vdev = vi->vdev; | |
679 | int ret; | |
680 | ||
681 | ret = eth_mac_addr(dev, p); | |
682 | if (ret) | |
683 | return ret; | |
684 | ||
62994b2d AW |
685 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) |
686 | vdev->config->set(vdev, offsetof(struct virtio_net_config, mac), | |
687 | dev->dev_addr, dev->addr_len); | |
9c46f6d4 AW |
688 | |
689 | return 0; | |
690 | } | |
691 | ||
3fa2a1df | 692 | static struct rtnl_link_stats64 *virtnet_stats(struct net_device *dev, |
693 | struct rtnl_link_stats64 *tot) | |
694 | { | |
695 | struct virtnet_info *vi = netdev_priv(dev); | |
696 | int cpu; | |
697 | unsigned int start; | |
698 | ||
699 | for_each_possible_cpu(cpu) { | |
58472a76 | 700 | struct virtnet_stats *stats = per_cpu_ptr(vi->stats, cpu); |
3fa2a1df | 701 | u64 tpackets, tbytes, rpackets, rbytes; |
702 | ||
703 | do { | |
704 | start = u64_stats_fetch_begin(&stats->syncp); | |
705 | tpackets = stats->tx_packets; | |
706 | tbytes = stats->tx_bytes; | |
707 | rpackets = stats->rx_packets; | |
708 | rbytes = stats->rx_bytes; | |
709 | } while (u64_stats_fetch_retry(&stats->syncp, start)); | |
710 | ||
711 | tot->rx_packets += rpackets; | |
712 | tot->tx_packets += tpackets; | |
713 | tot->rx_bytes += rbytes; | |
714 | tot->tx_bytes += tbytes; | |
715 | } | |
716 | ||
717 | tot->tx_dropped = dev->stats.tx_dropped; | |
021ac8d3 | 718 | tot->tx_fifo_errors = dev->stats.tx_fifo_errors; |
3fa2a1df | 719 | tot->rx_dropped = dev->stats.rx_dropped; |
720 | tot->rx_length_errors = dev->stats.rx_length_errors; | |
721 | tot->rx_frame_errors = dev->stats.rx_frame_errors; | |
722 | ||
723 | return tot; | |
724 | } | |
725 | ||
da74e89d AS |
726 | #ifdef CONFIG_NET_POLL_CONTROLLER |
727 | static void virtnet_netpoll(struct net_device *dev) | |
728 | { | |
729 | struct virtnet_info *vi = netdev_priv(dev); | |
730 | ||
731 | napi_schedule(&vi->napi); | |
732 | } | |
733 | #endif | |
734 | ||
296f96fc RR |
735 | static int virtnet_open(struct net_device *dev) |
736 | { | |
737 | struct virtnet_info *vi = netdev_priv(dev); | |
738 | ||
b2baed69 RR |
739 | /* Make sure we have some buffers: if oom use wq. */ |
740 | if (!try_fill_recv(vi, GFP_KERNEL)) | |
f1776dad | 741 | queue_delayed_work(system_nrt_wq, &vi->refill, 0); |
b2baed69 | 742 | |
3e9d08ec | 743 | virtnet_napi_enable(vi); |
296f96fc RR |
744 | return 0; |
745 | } | |
746 | ||
2a41f71d AW |
747 | /* |
748 | * Send command via the control virtqueue and check status. Commands | |
749 | * supported by the hypervisor, as indicated by feature bits, should | |
750 | * never fail unless improperly formated. | |
751 | */ | |
752 | static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd, | |
753 | struct scatterlist *data, int out, int in) | |
754 | { | |
23e258e1 | 755 | struct scatterlist *s, sg[VIRTNET_SEND_COMMAND_SG_MAX + 2]; |
2a41f71d AW |
756 | struct virtio_net_ctrl_hdr ctrl; |
757 | virtio_net_ctrl_ack status = ~0; | |
758 | unsigned int tmp; | |
23e258e1 | 759 | int i; |
2a41f71d | 760 | |
0ee904c3 AB |
761 | /* Caller should know better */ |
762 | BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) || | |
763 | (out + in > VIRTNET_SEND_COMMAND_SG_MAX)); | |
2a41f71d AW |
764 | |
765 | out++; /* Add header */ | |
766 | in++; /* Add return status */ | |
767 | ||
768 | ctrl.class = class; | |
769 | ctrl.cmd = cmd; | |
770 | ||
771 | sg_init_table(sg, out + in); | |
772 | ||
773 | sg_set_buf(&sg[0], &ctrl, sizeof(ctrl)); | |
23e258e1 AW |
774 | for_each_sg(data, s, out + in - 2, i) |
775 | sg_set_buf(&sg[i + 1], sg_virt(s), s->length); | |
2a41f71d AW |
776 | sg_set_buf(&sg[out + in - 1], &status, sizeof(status)); |
777 | ||
f96fde41 | 778 | BUG_ON(virtqueue_add_buf(vi->cvq, sg, out, in, vi, GFP_ATOMIC) < 0); |
2a41f71d | 779 | |
1915a712 | 780 | virtqueue_kick(vi->cvq); |
2a41f71d AW |
781 | |
782 | /* | |
783 | * Spin for a response, the kick causes an ioport write, trapping | |
784 | * into the hypervisor, so the request should be handled immediately. | |
785 | */ | |
1915a712 | 786 | while (!virtqueue_get_buf(vi->cvq, &tmp)) |
2a41f71d AW |
787 | cpu_relax(); |
788 | ||
789 | return status == VIRTIO_NET_OK; | |
790 | } | |
791 | ||
586d17c5 JW |
792 | static void virtnet_ack_link_announce(struct virtnet_info *vi) |
793 | { | |
794 | rtnl_lock(); | |
795 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE, | |
796 | VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL, | |
797 | 0, 0)) | |
798 | dev_warn(&vi->dev->dev, "Failed to ack link announce.\n"); | |
799 | rtnl_unlock(); | |
800 | } | |
801 | ||
296f96fc RR |
802 | static int virtnet_close(struct net_device *dev) |
803 | { | |
804 | struct virtnet_info *vi = netdev_priv(dev); | |
296f96fc | 805 | |
b2baed69 RR |
806 | /* Make sure refill_work doesn't re-enable napi! */ |
807 | cancel_delayed_work_sync(&vi->refill); | |
296f96fc RR |
808 | napi_disable(&vi->napi); |
809 | ||
296f96fc RR |
810 | return 0; |
811 | } | |
812 | ||
2af7698e AW |
813 | static void virtnet_set_rx_mode(struct net_device *dev) |
814 | { | |
815 | struct virtnet_info *vi = netdev_priv(dev); | |
f565a7c2 | 816 | struct scatterlist sg[2]; |
2af7698e | 817 | u8 promisc, allmulti; |
f565a7c2 | 818 | struct virtio_net_ctrl_mac *mac_data; |
ccffad25 | 819 | struct netdev_hw_addr *ha; |
32e7bfc4 | 820 | int uc_count; |
4cd24eaf | 821 | int mc_count; |
f565a7c2 AW |
822 | void *buf; |
823 | int i; | |
2af7698e AW |
824 | |
825 | /* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */ | |
826 | if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX)) | |
827 | return; | |
828 | ||
f565a7c2 AW |
829 | promisc = ((dev->flags & IFF_PROMISC) != 0); |
830 | allmulti = ((dev->flags & IFF_ALLMULTI) != 0); | |
2af7698e | 831 | |
23e258e1 | 832 | sg_init_one(sg, &promisc, sizeof(promisc)); |
2af7698e AW |
833 | |
834 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, | |
835 | VIRTIO_NET_CTRL_RX_PROMISC, | |
f565a7c2 | 836 | sg, 1, 0)) |
2af7698e AW |
837 | dev_warn(&dev->dev, "Failed to %sable promisc mode.\n", |
838 | promisc ? "en" : "dis"); | |
839 | ||
23e258e1 | 840 | sg_init_one(sg, &allmulti, sizeof(allmulti)); |
2af7698e AW |
841 | |
842 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, | |
843 | VIRTIO_NET_CTRL_RX_ALLMULTI, | |
f565a7c2 | 844 | sg, 1, 0)) |
2af7698e AW |
845 | dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n", |
846 | allmulti ? "en" : "dis"); | |
f565a7c2 | 847 | |
32e7bfc4 | 848 | uc_count = netdev_uc_count(dev); |
4cd24eaf | 849 | mc_count = netdev_mc_count(dev); |
f565a7c2 | 850 | /* MAC filter - use one buffer for both lists */ |
4cd24eaf JP |
851 | buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) + |
852 | (2 * sizeof(mac_data->entries)), GFP_ATOMIC); | |
853 | mac_data = buf; | |
f565a7c2 AW |
854 | if (!buf) { |
855 | dev_warn(&dev->dev, "No memory for MAC address buffer\n"); | |
856 | return; | |
857 | } | |
858 | ||
23e258e1 AW |
859 | sg_init_table(sg, 2); |
860 | ||
f565a7c2 | 861 | /* Store the unicast list and count in the front of the buffer */ |
32e7bfc4 | 862 | mac_data->entries = uc_count; |
ccffad25 | 863 | i = 0; |
32e7bfc4 | 864 | netdev_for_each_uc_addr(ha, dev) |
ccffad25 | 865 | memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); |
f565a7c2 AW |
866 | |
867 | sg_set_buf(&sg[0], mac_data, | |
32e7bfc4 | 868 | sizeof(mac_data->entries) + (uc_count * ETH_ALEN)); |
f565a7c2 AW |
869 | |
870 | /* multicast list and count fill the end */ | |
32e7bfc4 | 871 | mac_data = (void *)&mac_data->macs[uc_count][0]; |
f565a7c2 | 872 | |
4cd24eaf | 873 | mac_data->entries = mc_count; |
567ec874 | 874 | i = 0; |
22bedad3 JP |
875 | netdev_for_each_mc_addr(ha, dev) |
876 | memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); | |
f565a7c2 AW |
877 | |
878 | sg_set_buf(&sg[1], mac_data, | |
4cd24eaf | 879 | sizeof(mac_data->entries) + (mc_count * ETH_ALEN)); |
f565a7c2 AW |
880 | |
881 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC, | |
882 | VIRTIO_NET_CTRL_MAC_TABLE_SET, | |
883 | sg, 2, 0)) | |
884 | dev_warn(&dev->dev, "Failed to set MAC fitler table.\n"); | |
885 | ||
886 | kfree(buf); | |
2af7698e AW |
887 | } |
888 | ||
8e586137 | 889 | static int virtnet_vlan_rx_add_vid(struct net_device *dev, u16 vid) |
0bde9569 AW |
890 | { |
891 | struct virtnet_info *vi = netdev_priv(dev); | |
892 | struct scatterlist sg; | |
893 | ||
23e258e1 | 894 | sg_init_one(&sg, &vid, sizeof(vid)); |
0bde9569 AW |
895 | |
896 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, | |
897 | VIRTIO_NET_CTRL_VLAN_ADD, &sg, 1, 0)) | |
898 | dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid); | |
8e586137 | 899 | return 0; |
0bde9569 AW |
900 | } |
901 | ||
8e586137 | 902 | static int virtnet_vlan_rx_kill_vid(struct net_device *dev, u16 vid) |
0bde9569 AW |
903 | { |
904 | struct virtnet_info *vi = netdev_priv(dev); | |
905 | struct scatterlist sg; | |
906 | ||
23e258e1 | 907 | sg_init_one(&sg, &vid, sizeof(vid)); |
0bde9569 AW |
908 | |
909 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, | |
910 | VIRTIO_NET_CTRL_VLAN_DEL, &sg, 1, 0)) | |
911 | dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid); | |
8e586137 | 912 | return 0; |
0bde9569 AW |
913 | } |
914 | ||
8f9f4668 RJ |
915 | static void virtnet_get_ringparam(struct net_device *dev, |
916 | struct ethtool_ringparam *ring) | |
917 | { | |
918 | struct virtnet_info *vi = netdev_priv(dev); | |
919 | ||
920 | ring->rx_max_pending = virtqueue_get_vring_size(vi->rvq); | |
921 | ring->tx_max_pending = virtqueue_get_vring_size(vi->svq); | |
922 | ring->rx_pending = ring->rx_max_pending; | |
923 | ring->tx_pending = ring->tx_max_pending; | |
924 | ||
925 | } | |
926 | ||
66846048 RJ |
927 | |
928 | static void virtnet_get_drvinfo(struct net_device *dev, | |
929 | struct ethtool_drvinfo *info) | |
930 | { | |
931 | struct virtnet_info *vi = netdev_priv(dev); | |
932 | struct virtio_device *vdev = vi->vdev; | |
933 | ||
934 | strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver)); | |
935 | strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version)); | |
936 | strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info)); | |
937 | ||
938 | } | |
939 | ||
0fc0b732 | 940 | static const struct ethtool_ops virtnet_ethtool_ops = { |
66846048 | 941 | .get_drvinfo = virtnet_get_drvinfo, |
9f4d26d0 | 942 | .get_link = ethtool_op_get_link, |
8f9f4668 | 943 | .get_ringparam = virtnet_get_ringparam, |
a9ea3fc6 HX |
944 | }; |
945 | ||
39da5814 MM |
946 | #define MIN_MTU 68 |
947 | #define MAX_MTU 65535 | |
948 | ||
949 | static int virtnet_change_mtu(struct net_device *dev, int new_mtu) | |
950 | { | |
951 | if (new_mtu < MIN_MTU || new_mtu > MAX_MTU) | |
952 | return -EINVAL; | |
953 | dev->mtu = new_mtu; | |
954 | return 0; | |
955 | } | |
956 | ||
76288b4e SH |
957 | static const struct net_device_ops virtnet_netdev = { |
958 | .ndo_open = virtnet_open, | |
959 | .ndo_stop = virtnet_close, | |
960 | .ndo_start_xmit = start_xmit, | |
961 | .ndo_validate_addr = eth_validate_addr, | |
9c46f6d4 | 962 | .ndo_set_mac_address = virtnet_set_mac_address, |
2af7698e | 963 | .ndo_set_rx_mode = virtnet_set_rx_mode, |
76288b4e | 964 | .ndo_change_mtu = virtnet_change_mtu, |
3fa2a1df | 965 | .ndo_get_stats64 = virtnet_stats, |
1824a989 AW |
966 | .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid, |
967 | .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid, | |
76288b4e SH |
968 | #ifdef CONFIG_NET_POLL_CONTROLLER |
969 | .ndo_poll_controller = virtnet_netpoll, | |
970 | #endif | |
971 | }; | |
972 | ||
586d17c5 | 973 | static void virtnet_config_changed_work(struct work_struct *work) |
9f4d26d0 | 974 | { |
586d17c5 JW |
975 | struct virtnet_info *vi = |
976 | container_of(work, struct virtnet_info, config_work); | |
9f4d26d0 MM |
977 | u16 v; |
978 | ||
586d17c5 JW |
979 | mutex_lock(&vi->config_lock); |
980 | if (!vi->config_enable) | |
981 | goto done; | |
982 | ||
77dd7693 | 983 | if (virtio_config_val(vi->vdev, VIRTIO_NET_F_STATUS, |
9f4d26d0 | 984 | offsetof(struct virtio_net_config, status), |
77dd7693 | 985 | &v) < 0) |
586d17c5 JW |
986 | goto done; |
987 | ||
988 | if (v & VIRTIO_NET_S_ANNOUNCE) { | |
989 | netif_notify_peers(vi->dev); | |
990 | virtnet_ack_link_announce(vi); | |
991 | } | |
9f4d26d0 MM |
992 | |
993 | /* Ignore unknown (future) status bits */ | |
994 | v &= VIRTIO_NET_S_LINK_UP; | |
995 | ||
996 | if (vi->status == v) | |
586d17c5 | 997 | goto done; |
9f4d26d0 MM |
998 | |
999 | vi->status = v; | |
1000 | ||
1001 | if (vi->status & VIRTIO_NET_S_LINK_UP) { | |
1002 | netif_carrier_on(vi->dev); | |
1003 | netif_wake_queue(vi->dev); | |
1004 | } else { | |
1005 | netif_carrier_off(vi->dev); | |
1006 | netif_stop_queue(vi->dev); | |
1007 | } | |
586d17c5 JW |
1008 | done: |
1009 | mutex_unlock(&vi->config_lock); | |
9f4d26d0 MM |
1010 | } |
1011 | ||
1012 | static void virtnet_config_changed(struct virtio_device *vdev) | |
1013 | { | |
1014 | struct virtnet_info *vi = vdev->priv; | |
1015 | ||
586d17c5 | 1016 | queue_work(system_nrt_wq, &vi->config_work); |
9f4d26d0 MM |
1017 | } |
1018 | ||
3f9c10b0 AS |
1019 | static int init_vqs(struct virtnet_info *vi) |
1020 | { | |
1021 | struct virtqueue *vqs[3]; | |
1022 | vq_callback_t *callbacks[] = { skb_recv_done, skb_xmit_done, NULL}; | |
1023 | const char *names[] = { "input", "output", "control" }; | |
1024 | int nvqs, err; | |
1025 | ||
1026 | /* We expect two virtqueues, receive then send, | |
1027 | * and optionally control. */ | |
1028 | nvqs = virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ? 3 : 2; | |
1029 | ||
1030 | err = vi->vdev->config->find_vqs(vi->vdev, nvqs, vqs, callbacks, names); | |
1031 | if (err) | |
1032 | return err; | |
1033 | ||
1034 | vi->rvq = vqs[0]; | |
1035 | vi->svq = vqs[1]; | |
1036 | ||
1037 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)) { | |
1038 | vi->cvq = vqs[2]; | |
1039 | ||
1040 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN)) | |
1041 | vi->dev->features |= NETIF_F_HW_VLAN_FILTER; | |
1042 | } | |
1043 | return 0; | |
1044 | } | |
1045 | ||
296f96fc RR |
1046 | static int virtnet_probe(struct virtio_device *vdev) |
1047 | { | |
1048 | int err; | |
296f96fc RR |
1049 | struct net_device *dev; |
1050 | struct virtnet_info *vi; | |
296f96fc RR |
1051 | |
1052 | /* Allocate ourselves a network device with room for our info */ | |
1053 | dev = alloc_etherdev(sizeof(struct virtnet_info)); | |
1054 | if (!dev) | |
1055 | return -ENOMEM; | |
1056 | ||
1057 | /* Set up network device as normal. */ | |
01789349 | 1058 | dev->priv_flags |= IFF_UNICAST_FLT; |
76288b4e | 1059 | dev->netdev_ops = &virtnet_netdev; |
296f96fc | 1060 | dev->features = NETIF_F_HIGHDMA; |
3fa2a1df | 1061 | |
a9ea3fc6 | 1062 | SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops); |
296f96fc RR |
1063 | SET_NETDEV_DEV(dev, &vdev->dev); |
1064 | ||
1065 | /* Do we support "hardware" checksums? */ | |
98e778c9 | 1066 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) { |
296f96fc | 1067 | /* This opens up the world of extra features. */ |
98e778c9 MM |
1068 | dev->hw_features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST; |
1069 | if (csum) | |
1070 | dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST; | |
1071 | ||
1072 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) { | |
1073 | dev->hw_features |= NETIF_F_TSO | NETIF_F_UFO | |
34a48579 RR |
1074 | | NETIF_F_TSO_ECN | NETIF_F_TSO6; |
1075 | } | |
5539ae96 | 1076 | /* Individual feature bits: what can host handle? */ |
98e778c9 MM |
1077 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4)) |
1078 | dev->hw_features |= NETIF_F_TSO; | |
1079 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6)) | |
1080 | dev->hw_features |= NETIF_F_TSO6; | |
1081 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN)) | |
1082 | dev->hw_features |= NETIF_F_TSO_ECN; | |
1083 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO)) | |
1084 | dev->hw_features |= NETIF_F_UFO; | |
1085 | ||
1086 | if (gso) | |
1087 | dev->features |= dev->hw_features & (NETIF_F_ALL_TSO|NETIF_F_UFO); | |
1088 | /* (!csum && gso) case will be fixed by register_netdev() */ | |
296f96fc RR |
1089 | } |
1090 | ||
1091 | /* Configuration may specify what MAC to use. Otherwise random. */ | |
77dd7693 | 1092 | if (virtio_config_val_len(vdev, VIRTIO_NET_F_MAC, |
a586d4f6 | 1093 | offsetof(struct virtio_net_config, mac), |
77dd7693 | 1094 | dev->dev_addr, dev->addr_len) < 0) |
f2cedb63 | 1095 | eth_hw_addr_random(dev); |
296f96fc RR |
1096 | |
1097 | /* Set up our device-specific information */ | |
1098 | vi = netdev_priv(dev); | |
6c0cd7c0 | 1099 | netif_napi_add(dev, &vi->napi, virtnet_poll, napi_weight); |
296f96fc RR |
1100 | vi->dev = dev; |
1101 | vi->vdev = vdev; | |
d9d5dcc8 | 1102 | vdev->priv = vi; |
fb6813f4 | 1103 | vi->pages = NULL; |
3fa2a1df | 1104 | vi->stats = alloc_percpu(struct virtnet_stats); |
1105 | err = -ENOMEM; | |
1106 | if (vi->stats == NULL) | |
1107 | goto free; | |
1108 | ||
3161e453 | 1109 | INIT_DELAYED_WORK(&vi->refill, refill_work); |
586d17c5 JW |
1110 | mutex_init(&vi->config_lock); |
1111 | vi->config_enable = true; | |
1112 | INIT_WORK(&vi->config_work, virtnet_config_changed_work); | |
5e01d2f9 MT |
1113 | sg_init_table(vi->rx_sg, ARRAY_SIZE(vi->rx_sg)); |
1114 | sg_init_table(vi->tx_sg, ARRAY_SIZE(vi->tx_sg)); | |
296f96fc | 1115 | |
97402b96 | 1116 | /* If we can receive ANY GSO packets, we must allocate large ones. */ |
8e95a202 JP |
1117 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) || |
1118 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) || | |
1119 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN)) | |
97402b96 HX |
1120 | vi->big_packets = true; |
1121 | ||
3f2c31d9 MM |
1122 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF)) |
1123 | vi->mergeable_rx_bufs = true; | |
1124 | ||
3f9c10b0 | 1125 | err = init_vqs(vi); |
d2a7ddda | 1126 | if (err) |
3fa2a1df | 1127 | goto free_stats; |
296f96fc | 1128 | |
296f96fc RR |
1129 | err = register_netdev(dev); |
1130 | if (err) { | |
1131 | pr_debug("virtio_net: registering device failed\n"); | |
d2a7ddda | 1132 | goto free_vqs; |
296f96fc | 1133 | } |
b3369c1f RR |
1134 | |
1135 | /* Last of all, set up some receive buffers. */ | |
3161e453 | 1136 | try_fill_recv(vi, GFP_KERNEL); |
b3369c1f RR |
1137 | |
1138 | /* If we didn't even get one input buffer, we're useless. */ | |
1139 | if (vi->num == 0) { | |
1140 | err = -ENOMEM; | |
1141 | goto unregister; | |
1142 | } | |
1143 | ||
167c25e4 JW |
1144 | /* Assume link up if device can't report link status, |
1145 | otherwise get link status from config. */ | |
1146 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) { | |
1147 | netif_carrier_off(dev); | |
586d17c5 | 1148 | queue_work(system_nrt_wq, &vi->config_work); |
167c25e4 JW |
1149 | } else { |
1150 | vi->status = VIRTIO_NET_S_LINK_UP; | |
1151 | netif_carrier_on(dev); | |
1152 | } | |
9f4d26d0 | 1153 | |
296f96fc | 1154 | pr_debug("virtnet: registered device %s\n", dev->name); |
296f96fc RR |
1155 | return 0; |
1156 | ||
b3369c1f RR |
1157 | unregister: |
1158 | unregister_netdev(dev); | |
d2a7ddda MT |
1159 | free_vqs: |
1160 | vdev->config->del_vqs(vdev); | |
3fa2a1df | 1161 | free_stats: |
1162 | free_percpu(vi->stats); | |
296f96fc RR |
1163 | free: |
1164 | free_netdev(dev); | |
1165 | return err; | |
1166 | } | |
1167 | ||
9ab86bbc SM |
1168 | static void free_unused_bufs(struct virtnet_info *vi) |
1169 | { | |
1170 | void *buf; | |
830a8a97 | 1171 | while (1) { |
1915a712 | 1172 | buf = virtqueue_detach_unused_buf(vi->svq); |
830a8a97 SM |
1173 | if (!buf) |
1174 | break; | |
1175 | dev_kfree_skb(buf); | |
1176 | } | |
9ab86bbc | 1177 | while (1) { |
1915a712 | 1178 | buf = virtqueue_detach_unused_buf(vi->rvq); |
9ab86bbc SM |
1179 | if (!buf) |
1180 | break; | |
1181 | if (vi->mergeable_rx_bufs || vi->big_packets) | |
1182 | give_pages(vi, buf); | |
1183 | else | |
1184 | dev_kfree_skb(buf); | |
1185 | --vi->num; | |
1186 | } | |
1187 | BUG_ON(vi->num != 0); | |
1188 | } | |
1189 | ||
04486ed0 | 1190 | static void remove_vq_common(struct virtnet_info *vi) |
296f96fc | 1191 | { |
04486ed0 | 1192 | vi->vdev->config->reset(vi->vdev); |
830a8a97 SM |
1193 | |
1194 | /* Free unused buffers in both send and recv, if any. */ | |
9ab86bbc | 1195 | free_unused_bufs(vi); |
fb6813f4 | 1196 | |
04486ed0 | 1197 | vi->vdev->config->del_vqs(vi->vdev); |
d2a7ddda | 1198 | |
fb6813f4 RR |
1199 | while (vi->pages) |
1200 | __free_pages(get_a_page(vi, GFP_KERNEL), 0); | |
04486ed0 AS |
1201 | } |
1202 | ||
1203 | static void __devexit virtnet_remove(struct virtio_device *vdev) | |
1204 | { | |
1205 | struct virtnet_info *vi = vdev->priv; | |
1206 | ||
586d17c5 JW |
1207 | /* Prevent config work handler from accessing the device. */ |
1208 | mutex_lock(&vi->config_lock); | |
1209 | vi->config_enable = false; | |
1210 | mutex_unlock(&vi->config_lock); | |
1211 | ||
04486ed0 AS |
1212 | unregister_netdev(vi->dev); |
1213 | ||
1214 | remove_vq_common(vi); | |
fb6813f4 | 1215 | |
586d17c5 JW |
1216 | flush_work(&vi->config_work); |
1217 | ||
2e66f55b | 1218 | free_percpu(vi->stats); |
74b2553f | 1219 | free_netdev(vi->dev); |
296f96fc RR |
1220 | } |
1221 | ||
0741bcb5 AS |
1222 | #ifdef CONFIG_PM |
1223 | static int virtnet_freeze(struct virtio_device *vdev) | |
1224 | { | |
1225 | struct virtnet_info *vi = vdev->priv; | |
1226 | ||
586d17c5 JW |
1227 | /* Prevent config work handler from accessing the device */ |
1228 | mutex_lock(&vi->config_lock); | |
1229 | vi->config_enable = false; | |
1230 | mutex_unlock(&vi->config_lock); | |
1231 | ||
0741bcb5 AS |
1232 | virtqueue_disable_cb(vi->rvq); |
1233 | virtqueue_disable_cb(vi->svq); | |
1234 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)) | |
1235 | virtqueue_disable_cb(vi->cvq); | |
1236 | ||
1237 | netif_device_detach(vi->dev); | |
1238 | cancel_delayed_work_sync(&vi->refill); | |
1239 | ||
1240 | if (netif_running(vi->dev)) | |
1241 | napi_disable(&vi->napi); | |
1242 | ||
1243 | remove_vq_common(vi); | |
1244 | ||
586d17c5 JW |
1245 | flush_work(&vi->config_work); |
1246 | ||
0741bcb5 AS |
1247 | return 0; |
1248 | } | |
1249 | ||
1250 | static int virtnet_restore(struct virtio_device *vdev) | |
1251 | { | |
1252 | struct virtnet_info *vi = vdev->priv; | |
1253 | int err; | |
1254 | ||
1255 | err = init_vqs(vi); | |
1256 | if (err) | |
1257 | return err; | |
1258 | ||
1259 | if (netif_running(vi->dev)) | |
1260 | virtnet_napi_enable(vi); | |
1261 | ||
1262 | netif_device_attach(vi->dev); | |
1263 | ||
1264 | if (!try_fill_recv(vi, GFP_KERNEL)) | |
1265 | queue_delayed_work(system_nrt_wq, &vi->refill, 0); | |
1266 | ||
586d17c5 JW |
1267 | mutex_lock(&vi->config_lock); |
1268 | vi->config_enable = true; | |
1269 | mutex_unlock(&vi->config_lock); | |
1270 | ||
0741bcb5 AS |
1271 | return 0; |
1272 | } | |
1273 | #endif | |
1274 | ||
296f96fc RR |
1275 | static struct virtio_device_id id_table[] = { |
1276 | { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID }, | |
1277 | { 0 }, | |
1278 | }; | |
1279 | ||
c45a6816 | 1280 | static unsigned int features[] = { |
5e4fe5c4 MM |
1281 | VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM, |
1282 | VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC, | |
c45a6816 | 1283 | VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6, |
97402b96 | 1284 | VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, |
5c516751 | 1285 | VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO, |
2a41f71d | 1286 | VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ, |
0bde9569 | 1287 | VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, |
586d17c5 | 1288 | VIRTIO_NET_F_GUEST_ANNOUNCE, |
c45a6816 RR |
1289 | }; |
1290 | ||
22402529 | 1291 | static struct virtio_driver virtio_net_driver = { |
c45a6816 RR |
1292 | .feature_table = features, |
1293 | .feature_table_size = ARRAY_SIZE(features), | |
296f96fc RR |
1294 | .driver.name = KBUILD_MODNAME, |
1295 | .driver.owner = THIS_MODULE, | |
1296 | .id_table = id_table, | |
1297 | .probe = virtnet_probe, | |
1298 | .remove = __devexit_p(virtnet_remove), | |
9f4d26d0 | 1299 | .config_changed = virtnet_config_changed, |
0741bcb5 AS |
1300 | #ifdef CONFIG_PM |
1301 | .freeze = virtnet_freeze, | |
1302 | .restore = virtnet_restore, | |
1303 | #endif | |
296f96fc RR |
1304 | }; |
1305 | ||
1306 | static int __init init(void) | |
1307 | { | |
22402529 | 1308 | return register_virtio_driver(&virtio_net_driver); |
296f96fc RR |
1309 | } |
1310 | ||
1311 | static void __exit fini(void) | |
1312 | { | |
22402529 | 1313 | unregister_virtio_driver(&virtio_net_driver); |
296f96fc RR |
1314 | } |
1315 | module_init(init); | |
1316 | module_exit(fini); | |
1317 | ||
1318 | MODULE_DEVICE_TABLE(virtio, id_table); | |
1319 | MODULE_DESCRIPTION("Virtio network driver"); | |
1320 | MODULE_LICENSE("GPL"); |