]>
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 | |
adf8d3ff | 16 | * along with this program; if not, see <http://www.gnu.org/licenses/>. |
296f96fc RR |
17 | */ |
18 | //#define DEBUG | |
19 | #include <linux/netdevice.h> | |
20 | #include <linux/etherdevice.h> | |
a9ea3fc6 | 21 | #include <linux/ethtool.h> |
296f96fc RR |
22 | #include <linux/module.h> |
23 | #include <linux/virtio.h> | |
24 | #include <linux/virtio_net.h> | |
25 | #include <linux/scatterlist.h> | |
e918085a | 26 | #include <linux/if_vlan.h> |
5a0e3ad6 | 27 | #include <linux/slab.h> |
8de4b2f3 | 28 | #include <linux/cpu.h> |
ab7db917 | 29 | #include <linux/average.h> |
296f96fc | 30 | |
d34710e3 | 31 | static int napi_weight = NAPI_POLL_WEIGHT; |
6c0cd7c0 DL |
32 | module_param(napi_weight, int, 0444); |
33 | ||
eb939922 | 34 | static bool csum = true, gso = true; |
34a48579 RR |
35 | module_param(csum, bool, 0444); |
36 | module_param(gso, bool, 0444); | |
37 | ||
296f96fc | 38 | /* FIXME: MTU in config. */ |
5061de36 | 39 | #define GOOD_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN) |
3f2c31d9 | 40 | #define GOOD_COPY_LEN 128 |
296f96fc | 41 | |
ab7db917 MD |
42 | /* Weight used for the RX packet size EWMA. The average packet size is used to |
43 | * determine the packet buffer size when refilling RX rings. As the entire RX | |
44 | * ring may be refilled at once, the weight is chosen so that the EWMA will be | |
45 | * insensitive to short-term, transient changes in packet size. | |
46 | */ | |
47 | #define RECEIVE_AVG_WEIGHT 64 | |
48 | ||
49 | /* Minimum alignment for mergeable packet buffers. */ | |
50 | #define MERGEABLE_BUFFER_ALIGN max(L1_CACHE_BYTES, 256) | |
51 | ||
66846048 | 52 | #define VIRTNET_DRIVER_VERSION "1.0.0" |
2a41f71d | 53 | |
3fa2a1df | 54 | struct virtnet_stats { |
83a27052 ED |
55 | struct u64_stats_sync tx_syncp; |
56 | struct u64_stats_sync rx_syncp; | |
3fa2a1df | 57 | u64 tx_bytes; |
58 | u64 tx_packets; | |
59 | ||
60 | u64 rx_bytes; | |
61 | u64 rx_packets; | |
62 | }; | |
63 | ||
e9d7417b JW |
64 | /* Internal representation of a send virtqueue */ |
65 | struct send_queue { | |
66 | /* Virtqueue associated with this send _queue */ | |
67 | struct virtqueue *vq; | |
68 | ||
69 | /* TX: fragments + linear part + virtio header */ | |
70 | struct scatterlist sg[MAX_SKB_FRAGS + 2]; | |
986a4f4d JW |
71 | |
72 | /* Name of the send queue: output.$index */ | |
73 | char name[40]; | |
e9d7417b JW |
74 | }; |
75 | ||
76 | /* Internal representation of a receive virtqueue */ | |
77 | struct receive_queue { | |
78 | /* Virtqueue associated with this receive_queue */ | |
79 | struct virtqueue *vq; | |
80 | ||
296f96fc RR |
81 | struct napi_struct napi; |
82 | ||
e9d7417b JW |
83 | /* Chain pages by the private ptr. */ |
84 | struct page *pages; | |
85 | ||
ab7db917 MD |
86 | /* Average packet length for mergeable receive buffers. */ |
87 | struct ewma mrg_avg_pkt_len; | |
88 | ||
fb51879d MD |
89 | /* Page frag for packet buffer allocation. */ |
90 | struct page_frag alloc_frag; | |
91 | ||
e9d7417b JW |
92 | /* RX: fragments + linear part + virtio header */ |
93 | struct scatterlist sg[MAX_SKB_FRAGS + 2]; | |
986a4f4d JW |
94 | |
95 | /* Name of this receive queue: input.$index */ | |
96 | char name[40]; | |
e9d7417b JW |
97 | }; |
98 | ||
99 | struct virtnet_info { | |
100 | struct virtio_device *vdev; | |
101 | struct virtqueue *cvq; | |
102 | struct net_device *dev; | |
986a4f4d JW |
103 | struct send_queue *sq; |
104 | struct receive_queue *rq; | |
e9d7417b JW |
105 | unsigned int status; |
106 | ||
986a4f4d JW |
107 | /* Max # of queue pairs supported by the device */ |
108 | u16 max_queue_pairs; | |
109 | ||
110 | /* # of queue pairs currently used by the driver */ | |
111 | u16 curr_queue_pairs; | |
112 | ||
97402b96 HX |
113 | /* I like... big packets and I cannot lie! */ |
114 | bool big_packets; | |
115 | ||
3f2c31d9 MM |
116 | /* Host will merge rx buffers for big packets (shake it! shake it!) */ |
117 | bool mergeable_rx_bufs; | |
118 | ||
986a4f4d JW |
119 | /* Has control virtqueue */ |
120 | bool has_cvq; | |
121 | ||
e7428e95 MT |
122 | /* Host can handle any s/g split between our header and packet data */ |
123 | bool any_header_sg; | |
124 | ||
586d17c5 JW |
125 | /* enable config space updates */ |
126 | bool config_enable; | |
127 | ||
3fa2a1df | 128 | /* Active statistics */ |
129 | struct virtnet_stats __percpu *stats; | |
130 | ||
3161e453 RR |
131 | /* Work struct for refilling if we run low on memory. */ |
132 | struct delayed_work refill; | |
133 | ||
586d17c5 JW |
134 | /* Work struct for config space updates */ |
135 | struct work_struct config_work; | |
136 | ||
137 | /* Lock for config space updates */ | |
138 | struct mutex config_lock; | |
986a4f4d JW |
139 | |
140 | /* Does the affinity hint is set for virtqueues? */ | |
141 | bool affinity_hint_set; | |
47be2479 | 142 | |
8de4b2f3 WG |
143 | /* CPU hot plug notifier */ |
144 | struct notifier_block nb; | |
296f96fc RR |
145 | }; |
146 | ||
b3f24698 RR |
147 | struct skb_vnet_hdr { |
148 | union { | |
149 | struct virtio_net_hdr hdr; | |
150 | struct virtio_net_hdr_mrg_rxbuf mhdr; | |
151 | }; | |
152 | }; | |
153 | ||
9ab86bbc SM |
154 | struct padded_vnet_hdr { |
155 | struct virtio_net_hdr hdr; | |
156 | /* | |
157 | * virtio_net_hdr should be in a separated sg buffer because of a | |
158 | * QEMU bug, and data sg buffer shares same page with this header sg. | |
159 | * This padding makes next sg 16 byte aligned after virtio_net_hdr. | |
160 | */ | |
161 | char padding[6]; | |
162 | }; | |
163 | ||
986a4f4d JW |
164 | /* Converting between virtqueue no. and kernel tx/rx queue no. |
165 | * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq | |
166 | */ | |
167 | static int vq2txq(struct virtqueue *vq) | |
168 | { | |
9d0ca6ed | 169 | return (vq->index - 1) / 2; |
986a4f4d JW |
170 | } |
171 | ||
172 | static int txq2vq(int txq) | |
173 | { | |
174 | return txq * 2 + 1; | |
175 | } | |
176 | ||
177 | static int vq2rxq(struct virtqueue *vq) | |
178 | { | |
9d0ca6ed | 179 | return vq->index / 2; |
986a4f4d JW |
180 | } |
181 | ||
182 | static int rxq2vq(int rxq) | |
183 | { | |
184 | return rxq * 2; | |
185 | } | |
186 | ||
b3f24698 | 187 | static inline struct skb_vnet_hdr *skb_vnet_hdr(struct sk_buff *skb) |
296f96fc | 188 | { |
b3f24698 | 189 | return (struct skb_vnet_hdr *)skb->cb; |
296f96fc RR |
190 | } |
191 | ||
9ab86bbc SM |
192 | /* |
193 | * private is used to chain pages for big packets, put the whole | |
194 | * most recent used list in the beginning for reuse | |
195 | */ | |
e9d7417b | 196 | static void give_pages(struct receive_queue *rq, struct page *page) |
0a888fd1 | 197 | { |
9ab86bbc | 198 | struct page *end; |
0a888fd1 | 199 | |
e9d7417b | 200 | /* Find end of list, sew whole thing into vi->rq.pages. */ |
9ab86bbc | 201 | for (end = page; end->private; end = (struct page *)end->private); |
e9d7417b JW |
202 | end->private = (unsigned long)rq->pages; |
203 | rq->pages = page; | |
0a888fd1 MM |
204 | } |
205 | ||
e9d7417b | 206 | static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask) |
fb6813f4 | 207 | { |
e9d7417b | 208 | struct page *p = rq->pages; |
fb6813f4 | 209 | |
9ab86bbc | 210 | if (p) { |
e9d7417b | 211 | rq->pages = (struct page *)p->private; |
9ab86bbc SM |
212 | /* clear private here, it is used to chain pages */ |
213 | p->private = 0; | |
214 | } else | |
fb6813f4 RR |
215 | p = alloc_page(gfp_mask); |
216 | return p; | |
217 | } | |
218 | ||
e9d7417b | 219 | static void skb_xmit_done(struct virtqueue *vq) |
296f96fc | 220 | { |
e9d7417b | 221 | struct virtnet_info *vi = vq->vdev->priv; |
296f96fc | 222 | |
2cb9c6ba | 223 | /* Suppress further interrupts. */ |
e9d7417b | 224 | virtqueue_disable_cb(vq); |
11a3a154 | 225 | |
363f1514 | 226 | /* We were probably waiting for more output buffers. */ |
986a4f4d | 227 | netif_wake_subqueue(vi->dev, vq2txq(vq)); |
296f96fc RR |
228 | } |
229 | ||
ab7db917 MD |
230 | static unsigned int mergeable_ctx_to_buf_truesize(unsigned long mrg_ctx) |
231 | { | |
232 | unsigned int truesize = mrg_ctx & (MERGEABLE_BUFFER_ALIGN - 1); | |
233 | return (truesize + 1) * MERGEABLE_BUFFER_ALIGN; | |
234 | } | |
235 | ||
236 | static void *mergeable_ctx_to_buf_address(unsigned long mrg_ctx) | |
237 | { | |
238 | return (void *)(mrg_ctx & -MERGEABLE_BUFFER_ALIGN); | |
239 | ||
240 | } | |
241 | ||
242 | static unsigned long mergeable_buf_to_ctx(void *buf, unsigned int truesize) | |
243 | { | |
244 | unsigned int size = truesize / MERGEABLE_BUFFER_ALIGN; | |
245 | return (unsigned long)buf | (size - 1); | |
246 | } | |
247 | ||
3464645a | 248 | /* Called from bottom half context */ |
e9d7417b | 249 | static struct sk_buff *page_to_skb(struct receive_queue *rq, |
2613af0e MD |
250 | struct page *page, unsigned int offset, |
251 | unsigned int len, unsigned int truesize) | |
9ab86bbc | 252 | { |
e9d7417b | 253 | struct virtnet_info *vi = rq->vq->vdev->priv; |
9ab86bbc SM |
254 | struct sk_buff *skb; |
255 | struct skb_vnet_hdr *hdr; | |
2613af0e | 256 | unsigned int copy, hdr_len, hdr_padded_len; |
9ab86bbc | 257 | char *p; |
fb6813f4 | 258 | |
2613af0e | 259 | p = page_address(page) + offset; |
3f2c31d9 | 260 | |
9ab86bbc SM |
261 | /* copy small packet so we can reuse these pages for small data */ |
262 | skb = netdev_alloc_skb_ip_align(vi->dev, GOOD_COPY_LEN); | |
263 | if (unlikely(!skb)) | |
264 | return NULL; | |
3f2c31d9 | 265 | |
9ab86bbc | 266 | hdr = skb_vnet_hdr(skb); |
3f2c31d9 | 267 | |
9ab86bbc SM |
268 | if (vi->mergeable_rx_bufs) { |
269 | hdr_len = sizeof hdr->mhdr; | |
2613af0e | 270 | hdr_padded_len = sizeof hdr->mhdr; |
9ab86bbc SM |
271 | } else { |
272 | hdr_len = sizeof hdr->hdr; | |
2613af0e | 273 | hdr_padded_len = sizeof(struct padded_vnet_hdr); |
9ab86bbc | 274 | } |
3f2c31d9 | 275 | |
9ab86bbc | 276 | memcpy(hdr, p, hdr_len); |
3f2c31d9 | 277 | |
9ab86bbc | 278 | len -= hdr_len; |
2613af0e MD |
279 | offset += hdr_padded_len; |
280 | p += hdr_padded_len; | |
3f2c31d9 | 281 | |
9ab86bbc SM |
282 | copy = len; |
283 | if (copy > skb_tailroom(skb)) | |
284 | copy = skb_tailroom(skb); | |
285 | memcpy(skb_put(skb, copy), p, copy); | |
3f2c31d9 | 286 | |
9ab86bbc SM |
287 | len -= copy; |
288 | offset += copy; | |
3f2c31d9 | 289 | |
2613af0e MD |
290 | if (vi->mergeable_rx_bufs) { |
291 | if (len) | |
292 | skb_add_rx_frag(skb, 0, page, offset, len, truesize); | |
293 | else | |
294 | put_page(page); | |
295 | return skb; | |
296 | } | |
297 | ||
e878d78b SL |
298 | /* |
299 | * Verify that we can indeed put this data into a skb. | |
300 | * This is here to handle cases when the device erroneously | |
301 | * tries to receive more than is possible. This is usually | |
302 | * the case of a broken device. | |
303 | */ | |
304 | if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) { | |
be443899 | 305 | net_dbg_ratelimited("%s: too much data\n", skb->dev->name); |
e878d78b SL |
306 | dev_kfree_skb(skb); |
307 | return NULL; | |
308 | } | |
2613af0e | 309 | BUG_ON(offset >= PAGE_SIZE); |
9ab86bbc | 310 | while (len) { |
2613af0e MD |
311 | unsigned int frag_size = min((unsigned)PAGE_SIZE - offset, len); |
312 | skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page, offset, | |
313 | frag_size, truesize); | |
314 | len -= frag_size; | |
9ab86bbc SM |
315 | page = (struct page *)page->private; |
316 | offset = 0; | |
317 | } | |
3f2c31d9 | 318 | |
9ab86bbc | 319 | if (page) |
e9d7417b | 320 | give_pages(rq, page); |
3f2c31d9 | 321 | |
9ab86bbc SM |
322 | return skb; |
323 | } | |
3f2c31d9 | 324 | |
f121159d MT |
325 | static struct sk_buff *receive_small(void *buf, unsigned int len) |
326 | { | |
327 | struct sk_buff * skb = buf; | |
328 | ||
329 | len -= sizeof(struct virtio_net_hdr); | |
330 | skb_trim(skb, len); | |
331 | ||
332 | return skb; | |
333 | } | |
334 | ||
335 | static struct sk_buff *receive_big(struct net_device *dev, | |
336 | struct receive_queue *rq, | |
337 | void *buf, | |
338 | unsigned int len) | |
339 | { | |
340 | struct page *page = buf; | |
341 | struct sk_buff *skb = page_to_skb(rq, page, 0, len, PAGE_SIZE); | |
342 | ||
343 | if (unlikely(!skb)) | |
344 | goto err; | |
345 | ||
346 | return skb; | |
347 | ||
348 | err: | |
349 | dev->stats.rx_dropped++; | |
350 | give_pages(rq, page); | |
351 | return NULL; | |
352 | } | |
353 | ||
8fc3b9e9 MT |
354 | static struct sk_buff *receive_mergeable(struct net_device *dev, |
355 | struct receive_queue *rq, | |
ab7db917 | 356 | unsigned long ctx, |
8fc3b9e9 | 357 | unsigned int len) |
9ab86bbc | 358 | { |
ab7db917 | 359 | void *buf = mergeable_ctx_to_buf_address(ctx); |
8fc3b9e9 MT |
360 | struct skb_vnet_hdr *hdr = buf; |
361 | int num_buf = hdr->mhdr.num_buffers; | |
362 | struct page *page = virt_to_head_page(buf); | |
363 | int offset = buf - page_address(page); | |
ab7db917 MD |
364 | unsigned int truesize = max(len, mergeable_ctx_to_buf_truesize(ctx)); |
365 | ||
fb51879d | 366 | struct sk_buff *head_skb = page_to_skb(rq, page, offset, len, truesize); |
2613af0e | 367 | struct sk_buff *curr_skb = head_skb; |
9ab86bbc | 368 | |
8fc3b9e9 MT |
369 | if (unlikely(!curr_skb)) |
370 | goto err_skb; | |
9ab86bbc | 371 | while (--num_buf) { |
8fc3b9e9 MT |
372 | int num_skb_frags; |
373 | ||
ab7db917 MD |
374 | ctx = (unsigned long)virtqueue_get_buf(rq->vq, &len); |
375 | if (unlikely(!ctx)) { | |
8fc3b9e9 MT |
376 | pr_debug("%s: rx error: %d buffers out of %d missing\n", |
377 | dev->name, num_buf, hdr->mhdr.num_buffers); | |
378 | dev->stats.rx_length_errors++; | |
379 | goto err_buf; | |
3f2c31d9 | 380 | } |
8fc3b9e9 | 381 | |
ab7db917 | 382 | buf = mergeable_ctx_to_buf_address(ctx); |
8fc3b9e9 | 383 | page = virt_to_head_page(buf); |
8fc3b9e9 MT |
384 | |
385 | num_skb_frags = skb_shinfo(curr_skb)->nr_frags; | |
2613af0e MD |
386 | if (unlikely(num_skb_frags == MAX_SKB_FRAGS)) { |
387 | struct sk_buff *nskb = alloc_skb(0, GFP_ATOMIC); | |
8fc3b9e9 MT |
388 | |
389 | if (unlikely(!nskb)) | |
390 | goto err_skb; | |
2613af0e MD |
391 | if (curr_skb == head_skb) |
392 | skb_shinfo(curr_skb)->frag_list = nskb; | |
393 | else | |
394 | curr_skb->next = nskb; | |
395 | curr_skb = nskb; | |
396 | head_skb->truesize += nskb->truesize; | |
397 | num_skb_frags = 0; | |
398 | } | |
ab7db917 | 399 | truesize = max(len, mergeable_ctx_to_buf_truesize(ctx)); |
2613af0e MD |
400 | if (curr_skb != head_skb) { |
401 | head_skb->data_len += len; | |
402 | head_skb->len += len; | |
fb51879d | 403 | head_skb->truesize += truesize; |
2613af0e | 404 | } |
8fc3b9e9 | 405 | offset = buf - page_address(page); |
ba275241 JW |
406 | if (skb_can_coalesce(curr_skb, num_skb_frags, page, offset)) { |
407 | put_page(page); | |
408 | skb_coalesce_rx_frag(curr_skb, num_skb_frags - 1, | |
fb51879d | 409 | len, truesize); |
ba275241 JW |
410 | } else { |
411 | skb_add_rx_frag(curr_skb, num_skb_frags, page, | |
fb51879d | 412 | offset, len, truesize); |
ba275241 | 413 | } |
8fc3b9e9 MT |
414 | } |
415 | ||
ab7db917 | 416 | ewma_add(&rq->mrg_avg_pkt_len, head_skb->len); |
8fc3b9e9 MT |
417 | return head_skb; |
418 | ||
419 | err_skb: | |
420 | put_page(page); | |
421 | while (--num_buf) { | |
ab7db917 MD |
422 | ctx = (unsigned long)virtqueue_get_buf(rq->vq, &len); |
423 | if (unlikely(!ctx)) { | |
8fc3b9e9 MT |
424 | pr_debug("%s: rx error: %d buffers missing\n", |
425 | dev->name, num_buf); | |
426 | dev->stats.rx_length_errors++; | |
427 | break; | |
428 | } | |
ab7db917 | 429 | page = virt_to_head_page(mergeable_ctx_to_buf_address(ctx)); |
8fc3b9e9 | 430 | put_page(page); |
9ab86bbc | 431 | } |
8fc3b9e9 MT |
432 | err_buf: |
433 | dev->stats.rx_dropped++; | |
434 | dev_kfree_skb(head_skb); | |
435 | return NULL; | |
9ab86bbc SM |
436 | } |
437 | ||
e9d7417b | 438 | static void receive_buf(struct receive_queue *rq, void *buf, unsigned int len) |
9ab86bbc | 439 | { |
e9d7417b JW |
440 | struct virtnet_info *vi = rq->vq->vdev->priv; |
441 | struct net_device *dev = vi->dev; | |
58472a76 | 442 | struct virtnet_stats *stats = this_cpu_ptr(vi->stats); |
9ab86bbc | 443 | struct sk_buff *skb; |
9ab86bbc | 444 | struct skb_vnet_hdr *hdr; |
3f2c31d9 | 445 | |
9ab86bbc SM |
446 | if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) { |
447 | pr_debug("%s: short packet %i\n", dev->name, len); | |
448 | dev->stats.rx_length_errors++; | |
ab7db917 MD |
449 | if (vi->mergeable_rx_bufs) { |
450 | unsigned long ctx = (unsigned long)buf; | |
451 | void *base = mergeable_ctx_to_buf_address(ctx); | |
452 | put_page(virt_to_head_page(base)); | |
453 | } else if (vi->big_packets) { | |
98bfd23c | 454 | give_pages(rq, buf); |
ab7db917 | 455 | } else { |
9ab86bbc | 456 | dev_kfree_skb(buf); |
ab7db917 | 457 | } |
9ab86bbc SM |
458 | return; |
459 | } | |
3f2c31d9 | 460 | |
f121159d | 461 | if (vi->mergeable_rx_bufs) |
ab7db917 | 462 | skb = receive_mergeable(dev, rq, (unsigned long)buf, len); |
f121159d MT |
463 | else if (vi->big_packets) |
464 | skb = receive_big(dev, rq, buf, len); | |
465 | else | |
466 | skb = receive_small(buf, len); | |
467 | ||
468 | if (unlikely(!skb)) | |
469 | return; | |
3f2c31d9 | 470 | |
9ab86bbc | 471 | hdr = skb_vnet_hdr(skb); |
3fa2a1df | 472 | |
83a27052 | 473 | u64_stats_update_begin(&stats->rx_syncp); |
3fa2a1df | 474 | stats->rx_bytes += skb->len; |
475 | stats->rx_packets++; | |
83a27052 | 476 | u64_stats_update_end(&stats->rx_syncp); |
296f96fc | 477 | |
b3f24698 | 478 | if (hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { |
296f96fc | 479 | pr_debug("Needs csum!\n"); |
b3f24698 RR |
480 | if (!skb_partial_csum_set(skb, |
481 | hdr->hdr.csum_start, | |
482 | hdr->hdr.csum_offset)) | |
296f96fc | 483 | goto frame_err; |
10a8d94a JW |
484 | } else if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) { |
485 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
296f96fc RR |
486 | } |
487 | ||
23cde76d MM |
488 | skb->protocol = eth_type_trans(skb, dev); |
489 | pr_debug("Receiving skb proto 0x%04x len %i type %i\n", | |
490 | ntohs(skb->protocol), skb->len, skb->pkt_type); | |
491 | ||
b3f24698 | 492 | if (hdr->hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) { |
296f96fc | 493 | pr_debug("GSO!\n"); |
b3f24698 | 494 | switch (hdr->hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { |
296f96fc | 495 | case VIRTIO_NET_HDR_GSO_TCPV4: |
c9af6db4 | 496 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; |
296f96fc | 497 | break; |
296f96fc | 498 | case VIRTIO_NET_HDR_GSO_UDP: |
c9af6db4 | 499 | skb_shinfo(skb)->gso_type = SKB_GSO_UDP; |
296f96fc RR |
500 | break; |
501 | case VIRTIO_NET_HDR_GSO_TCPV6: | |
c9af6db4 | 502 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6; |
296f96fc RR |
503 | break; |
504 | default: | |
be443899 AW |
505 | net_warn_ratelimited("%s: bad gso type %u.\n", |
506 | dev->name, hdr->hdr.gso_type); | |
296f96fc RR |
507 | goto frame_err; |
508 | } | |
509 | ||
b3f24698 | 510 | if (hdr->hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN) |
c9af6db4 | 511 | skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN; |
34a48579 | 512 | |
b3f24698 | 513 | skb_shinfo(skb)->gso_size = hdr->hdr.gso_size; |
296f96fc | 514 | if (skb_shinfo(skb)->gso_size == 0) { |
be443899 | 515 | net_warn_ratelimited("%s: zero gso size.\n", dev->name); |
296f96fc RR |
516 | goto frame_err; |
517 | } | |
518 | ||
519 | /* Header must be checked, and gso_segs computed. */ | |
520 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
521 | skb_shinfo(skb)->gso_segs = 0; | |
522 | } | |
523 | ||
524 | netif_receive_skb(skb); | |
525 | return; | |
526 | ||
527 | frame_err: | |
528 | dev->stats.rx_frame_errors++; | |
296f96fc RR |
529 | dev_kfree_skb(skb); |
530 | } | |
531 | ||
e9d7417b | 532 | static int add_recvbuf_small(struct receive_queue *rq, gfp_t gfp) |
296f96fc | 533 | { |
e9d7417b | 534 | struct virtnet_info *vi = rq->vq->vdev->priv; |
296f96fc | 535 | struct sk_buff *skb; |
9ab86bbc | 536 | struct skb_vnet_hdr *hdr; |
9ab86bbc | 537 | int err; |
3f2c31d9 | 538 | |
5061de36 | 539 | skb = __netdev_alloc_skb_ip_align(vi->dev, GOOD_PACKET_LEN, gfp); |
9ab86bbc SM |
540 | if (unlikely(!skb)) |
541 | return -ENOMEM; | |
296f96fc | 542 | |
5061de36 | 543 | skb_put(skb, GOOD_PACKET_LEN); |
3f2c31d9 | 544 | |
9ab86bbc | 545 | hdr = skb_vnet_hdr(skb); |
e9d7417b | 546 | sg_set_buf(rq->sg, &hdr->hdr, sizeof hdr->hdr); |
97402b96 | 547 | |
e9d7417b | 548 | skb_to_sgvec(skb, rq->sg + 1, 0, skb->len); |
97402b96 | 549 | |
9dc7b9e4 | 550 | err = virtqueue_add_inbuf(rq->vq, rq->sg, 2, skb, gfp); |
9ab86bbc SM |
551 | if (err < 0) |
552 | dev_kfree_skb(skb); | |
97402b96 | 553 | |
9ab86bbc SM |
554 | return err; |
555 | } | |
97402b96 | 556 | |
e9d7417b | 557 | static int add_recvbuf_big(struct receive_queue *rq, gfp_t gfp) |
9ab86bbc | 558 | { |
9ab86bbc SM |
559 | struct page *first, *list = NULL; |
560 | char *p; | |
561 | int i, err, offset; | |
562 | ||
e9d7417b | 563 | /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */ |
9ab86bbc | 564 | for (i = MAX_SKB_FRAGS + 1; i > 1; --i) { |
e9d7417b | 565 | first = get_a_page(rq, gfp); |
9ab86bbc SM |
566 | if (!first) { |
567 | if (list) | |
e9d7417b | 568 | give_pages(rq, list); |
9ab86bbc | 569 | return -ENOMEM; |
97402b96 | 570 | } |
e9d7417b | 571 | sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE); |
97402b96 | 572 | |
9ab86bbc SM |
573 | /* chain new page in list head to match sg */ |
574 | first->private = (unsigned long)list; | |
575 | list = first; | |
576 | } | |
296f96fc | 577 | |
e9d7417b | 578 | first = get_a_page(rq, gfp); |
9ab86bbc | 579 | if (!first) { |
e9d7417b | 580 | give_pages(rq, list); |
9ab86bbc SM |
581 | return -ENOMEM; |
582 | } | |
583 | p = page_address(first); | |
584 | ||
e9d7417b JW |
585 | /* rq->sg[0], rq->sg[1] share the same page */ |
586 | /* a separated rq->sg[0] for virtio_net_hdr only due to QEMU bug */ | |
587 | sg_set_buf(&rq->sg[0], p, sizeof(struct virtio_net_hdr)); | |
9ab86bbc | 588 | |
e9d7417b | 589 | /* rq->sg[1] for data packet, from offset */ |
9ab86bbc | 590 | offset = sizeof(struct padded_vnet_hdr); |
e9d7417b | 591 | sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset); |
9ab86bbc SM |
592 | |
593 | /* chain first in list head */ | |
594 | first->private = (unsigned long)list; | |
9dc7b9e4 RR |
595 | err = virtqueue_add_inbuf(rq->vq, rq->sg, MAX_SKB_FRAGS + 2, |
596 | first, gfp); | |
9ab86bbc | 597 | if (err < 0) |
e9d7417b | 598 | give_pages(rq, first); |
9ab86bbc SM |
599 | |
600 | return err; | |
296f96fc RR |
601 | } |
602 | ||
fbf28d78 | 603 | static unsigned int get_mergeable_buf_len(struct ewma *avg_pkt_len) |
3f2c31d9 | 604 | { |
ab7db917 | 605 | const size_t hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf); |
fbf28d78 MD |
606 | unsigned int len; |
607 | ||
608 | len = hdr_len + clamp_t(unsigned int, ewma_read(avg_pkt_len), | |
609 | GOOD_PACKET_LEN, PAGE_SIZE - hdr_len); | |
610 | return ALIGN(len, MERGEABLE_BUFFER_ALIGN); | |
611 | } | |
612 | ||
613 | static int add_recvbuf_mergeable(struct receive_queue *rq, gfp_t gfp) | |
614 | { | |
fb51879d MD |
615 | struct page_frag *alloc_frag = &rq->alloc_frag; |
616 | char *buf; | |
ab7db917 | 617 | unsigned long ctx; |
3f2c31d9 | 618 | int err; |
fb51879d | 619 | unsigned int len, hole; |
3f2c31d9 | 620 | |
fbf28d78 | 621 | len = get_mergeable_buf_len(&rq->mrg_avg_pkt_len); |
ab7db917 | 622 | if (unlikely(!skb_page_frag_refill(len, alloc_frag, gfp))) |
9ab86bbc | 623 | return -ENOMEM; |
ab7db917 | 624 | |
fb51879d | 625 | buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset; |
ab7db917 | 626 | ctx = mergeable_buf_to_ctx(buf, len); |
fb51879d | 627 | get_page(alloc_frag->page); |
fb51879d MD |
628 | alloc_frag->offset += len; |
629 | hole = alloc_frag->size - alloc_frag->offset; | |
ab7db917 MD |
630 | if (hole < len) { |
631 | /* To avoid internal fragmentation, if there is very likely not | |
632 | * enough space for another buffer, add the remaining space to | |
633 | * the current buffer. This extra space is not included in | |
634 | * the truesize stored in ctx. | |
635 | */ | |
fb51879d MD |
636 | len += hole; |
637 | alloc_frag->offset += hole; | |
638 | } | |
3f2c31d9 | 639 | |
fb51879d | 640 | sg_init_one(rq->sg, buf, len); |
ab7db917 | 641 | err = virtqueue_add_inbuf(rq->vq, rq->sg, 1, (void *)ctx, gfp); |
9ab86bbc | 642 | if (err < 0) |
2613af0e | 643 | put_page(virt_to_head_page(buf)); |
3f2c31d9 | 644 | |
9ab86bbc SM |
645 | return err; |
646 | } | |
3f2c31d9 | 647 | |
b2baed69 RR |
648 | /* |
649 | * Returns false if we couldn't fill entirely (OOM). | |
650 | * | |
651 | * Normally run in the receive path, but can also be run from ndo_open | |
652 | * before we're receiving packets, or from refill_work which is | |
653 | * careful to disable receiving (using napi_disable). | |
654 | */ | |
e9d7417b | 655 | static bool try_fill_recv(struct receive_queue *rq, gfp_t gfp) |
9ab86bbc | 656 | { |
e9d7417b | 657 | struct virtnet_info *vi = rq->vq->vdev->priv; |
9ab86bbc | 658 | int err; |
1788f495 | 659 | bool oom; |
3f2c31d9 | 660 | |
fb51879d | 661 | gfp |= __GFP_COLD; |
9ab86bbc SM |
662 | do { |
663 | if (vi->mergeable_rx_bufs) | |
e9d7417b | 664 | err = add_recvbuf_mergeable(rq, gfp); |
9ab86bbc | 665 | else if (vi->big_packets) |
e9d7417b | 666 | err = add_recvbuf_big(rq, gfp); |
9ab86bbc | 667 | else |
e9d7417b | 668 | err = add_recvbuf_small(rq, gfp); |
3f2c31d9 | 669 | |
1788f495 | 670 | oom = err == -ENOMEM; |
9ed4cb07 | 671 | if (err) |
3f2c31d9 | 672 | break; |
b7dfde95 | 673 | } while (rq->vq->num_free); |
681daee2 | 674 | virtqueue_kick(rq->vq); |
3161e453 | 675 | return !oom; |
3f2c31d9 MM |
676 | } |
677 | ||
18445c4d | 678 | static void skb_recv_done(struct virtqueue *rvq) |
296f96fc RR |
679 | { |
680 | struct virtnet_info *vi = rvq->vdev->priv; | |
986a4f4d | 681 | struct receive_queue *rq = &vi->rq[vq2rxq(rvq)]; |
e9d7417b | 682 | |
18445c4d | 683 | /* Schedule NAPI, Suppress further interrupts if successful. */ |
e9d7417b | 684 | if (napi_schedule_prep(&rq->napi)) { |
1915a712 | 685 | virtqueue_disable_cb(rvq); |
e9d7417b | 686 | __napi_schedule(&rq->napi); |
18445c4d | 687 | } |
296f96fc RR |
688 | } |
689 | ||
e9d7417b | 690 | static void virtnet_napi_enable(struct receive_queue *rq) |
3e9d08ec | 691 | { |
e9d7417b | 692 | napi_enable(&rq->napi); |
3e9d08ec BR |
693 | |
694 | /* If all buffers were filled by other side before we napi_enabled, we | |
695 | * won't get another interrupt, so process any outstanding packets | |
696 | * now. virtnet_poll wants re-enable the queue, so we disable here. | |
697 | * We synchronize against interrupts via NAPI_STATE_SCHED */ | |
e9d7417b JW |
698 | if (napi_schedule_prep(&rq->napi)) { |
699 | virtqueue_disable_cb(rq->vq); | |
ec13ee80 | 700 | local_bh_disable(); |
e9d7417b | 701 | __napi_schedule(&rq->napi); |
ec13ee80 | 702 | local_bh_enable(); |
3e9d08ec BR |
703 | } |
704 | } | |
705 | ||
3161e453 RR |
706 | static void refill_work(struct work_struct *work) |
707 | { | |
e9d7417b JW |
708 | struct virtnet_info *vi = |
709 | container_of(work, struct virtnet_info, refill.work); | |
3161e453 | 710 | bool still_empty; |
986a4f4d JW |
711 | int i; |
712 | ||
55257d72 | 713 | for (i = 0; i < vi->curr_queue_pairs; i++) { |
986a4f4d | 714 | struct receive_queue *rq = &vi->rq[i]; |
3161e453 | 715 | |
986a4f4d JW |
716 | napi_disable(&rq->napi); |
717 | still_empty = !try_fill_recv(rq, GFP_KERNEL); | |
718 | virtnet_napi_enable(rq); | |
3161e453 | 719 | |
986a4f4d JW |
720 | /* In theory, this can happen: if we don't get any buffers in |
721 | * we will *never* try to fill again. | |
722 | */ | |
723 | if (still_empty) | |
724 | schedule_delayed_work(&vi->refill, HZ/2); | |
725 | } | |
3161e453 RR |
726 | } |
727 | ||
296f96fc RR |
728 | static int virtnet_poll(struct napi_struct *napi, int budget) |
729 | { | |
e9d7417b JW |
730 | struct receive_queue *rq = |
731 | container_of(napi, struct receive_queue, napi); | |
732 | struct virtnet_info *vi = rq->vq->vdev->priv; | |
9ab86bbc | 733 | void *buf; |
cbdadbbf | 734 | unsigned int r, len, received = 0; |
296f96fc RR |
735 | |
736 | again: | |
737 | while (received < budget && | |
e9d7417b JW |
738 | (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) { |
739 | receive_buf(rq, buf, len); | |
296f96fc RR |
740 | received++; |
741 | } | |
742 | ||
be121f46 | 743 | if (rq->vq->num_free > virtqueue_get_vring_size(rq->vq) / 2) { |
e9d7417b | 744 | if (!try_fill_recv(rq, GFP_ATOMIC)) |
3b07e9ca | 745 | schedule_delayed_work(&vi->refill, 0); |
3161e453 | 746 | } |
296f96fc | 747 | |
8329d98e RR |
748 | /* Out of packets? */ |
749 | if (received < budget) { | |
cbdadbbf | 750 | r = virtqueue_enable_cb_prepare(rq->vq); |
288379f0 | 751 | napi_complete(napi); |
cbdadbbf | 752 | if (unlikely(virtqueue_poll(rq->vq, r)) && |
8e95a202 | 753 | napi_schedule_prep(napi)) { |
e9d7417b | 754 | virtqueue_disable_cb(rq->vq); |
288379f0 | 755 | __napi_schedule(napi); |
296f96fc | 756 | goto again; |
4265f161 | 757 | } |
296f96fc RR |
758 | } |
759 | ||
760 | return received; | |
761 | } | |
762 | ||
986a4f4d JW |
763 | static int virtnet_open(struct net_device *dev) |
764 | { | |
765 | struct virtnet_info *vi = netdev_priv(dev); | |
766 | int i; | |
767 | ||
e4166625 JW |
768 | for (i = 0; i < vi->max_queue_pairs; i++) { |
769 | if (i < vi->curr_queue_pairs) | |
770 | /* Make sure we have some buffers: if oom use wq. */ | |
771 | if (!try_fill_recv(&vi->rq[i], GFP_KERNEL)) | |
772 | schedule_delayed_work(&vi->refill, 0); | |
986a4f4d JW |
773 | virtnet_napi_enable(&vi->rq[i]); |
774 | } | |
775 | ||
776 | return 0; | |
777 | } | |
778 | ||
b7dfde95 | 779 | static void free_old_xmit_skbs(struct send_queue *sq) |
296f96fc RR |
780 | { |
781 | struct sk_buff *skb; | |
6ee57bcc | 782 | unsigned int len; |
e9d7417b | 783 | struct virtnet_info *vi = sq->vq->vdev->priv; |
58472a76 | 784 | struct virtnet_stats *stats = this_cpu_ptr(vi->stats); |
296f96fc | 785 | |
e9d7417b | 786 | while ((skb = virtqueue_get_buf(sq->vq, &len)) != NULL) { |
296f96fc | 787 | pr_debug("Sent skb %p\n", skb); |
3fa2a1df | 788 | |
83a27052 | 789 | u64_stats_update_begin(&stats->tx_syncp); |
3fa2a1df | 790 | stats->tx_bytes += skb->len; |
791 | stats->tx_packets++; | |
83a27052 | 792 | u64_stats_update_end(&stats->tx_syncp); |
3fa2a1df | 793 | |
ed79bab8 | 794 | dev_kfree_skb_any(skb); |
296f96fc RR |
795 | } |
796 | } | |
797 | ||
e9d7417b | 798 | static int xmit_skb(struct send_queue *sq, struct sk_buff *skb) |
296f96fc | 799 | { |
e7428e95 | 800 | struct skb_vnet_hdr *hdr; |
296f96fc | 801 | const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest; |
e9d7417b | 802 | struct virtnet_info *vi = sq->vq->vdev->priv; |
7bedc7dc | 803 | unsigned num_sg; |
e7428e95 MT |
804 | unsigned hdr_len; |
805 | bool can_push; | |
296f96fc | 806 | |
e174961c | 807 | pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest); |
e7428e95 MT |
808 | if (vi->mergeable_rx_bufs) |
809 | hdr_len = sizeof hdr->mhdr; | |
810 | else | |
811 | hdr_len = sizeof hdr->hdr; | |
812 | ||
813 | can_push = vi->any_header_sg && | |
814 | !((unsigned long)skb->data & (__alignof__(*hdr) - 1)) && | |
815 | !skb_header_cloned(skb) && skb_headroom(skb) >= hdr_len; | |
816 | /* Even if we can, don't push here yet as this would skew | |
817 | * csum_start offset below. */ | |
818 | if (can_push) | |
819 | hdr = (struct skb_vnet_hdr *)(skb->data - hdr_len); | |
820 | else | |
821 | hdr = skb_vnet_hdr(skb); | |
296f96fc | 822 | |
296f96fc | 823 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
b3f24698 | 824 | hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; |
55508d60 | 825 | hdr->hdr.csum_start = skb_checksum_start_offset(skb); |
b3f24698 | 826 | hdr->hdr.csum_offset = skb->csum_offset; |
296f96fc | 827 | } else { |
b3f24698 RR |
828 | hdr->hdr.flags = 0; |
829 | hdr->hdr.csum_offset = hdr->hdr.csum_start = 0; | |
296f96fc RR |
830 | } |
831 | ||
832 | if (skb_is_gso(skb)) { | |
b3f24698 RR |
833 | hdr->hdr.hdr_len = skb_headlen(skb); |
834 | hdr->hdr.gso_size = skb_shinfo(skb)->gso_size; | |
34a48579 | 835 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) |
b3f24698 | 836 | hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4; |
296f96fc | 837 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) |
b3f24698 | 838 | hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6; |
296f96fc | 839 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP) |
b3f24698 | 840 | hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP; |
296f96fc RR |
841 | else |
842 | BUG(); | |
34a48579 | 843 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN) |
b3f24698 | 844 | hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN; |
296f96fc | 845 | } else { |
b3f24698 RR |
846 | hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE; |
847 | hdr->hdr.gso_size = hdr->hdr.hdr_len = 0; | |
296f96fc RR |
848 | } |
849 | ||
3f2c31d9 | 850 | if (vi->mergeable_rx_bufs) |
e7428e95 | 851 | hdr->mhdr.num_buffers = 0; |
3f2c31d9 | 852 | |
e7428e95 MT |
853 | if (can_push) { |
854 | __skb_push(skb, hdr_len); | |
855 | num_sg = skb_to_sgvec(skb, sq->sg, 0, skb->len); | |
856 | /* Pull header back to avoid skew in tx bytes calculations. */ | |
857 | __skb_pull(skb, hdr_len); | |
858 | } else { | |
859 | sg_set_buf(sq->sg, hdr, hdr_len); | |
860 | num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len) + 1; | |
861 | } | |
9dc7b9e4 | 862 | return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC); |
11a3a154 RR |
863 | } |
864 | ||
424efe9c | 865 | static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev) |
99ffc696 RR |
866 | { |
867 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d JW |
868 | int qnum = skb_get_queue_mapping(skb); |
869 | struct send_queue *sq = &vi->sq[qnum]; | |
9ed4cb07 | 870 | int err; |
2cb9c6ba | 871 | |
2cb9c6ba | 872 | /* Free up any pending old buffers before queueing new ones. */ |
e9d7417b | 873 | free_old_xmit_skbs(sq); |
99ffc696 | 874 | |
03f191ba | 875 | /* Try to transmit */ |
b7dfde95 | 876 | err = xmit_skb(sq, skb); |
48925e37 | 877 | |
9ed4cb07 | 878 | /* This should not happen! */ |
681daee2 | 879 | if (unlikely(err)) { |
9ed4cb07 RR |
880 | dev->stats.tx_fifo_errors++; |
881 | if (net_ratelimit()) | |
882 | dev_warn(&dev->dev, | |
b7dfde95 | 883 | "Unexpected TXQ (%d) queue failure: %d\n", qnum, err); |
58eba97d | 884 | dev->stats.tx_dropped++; |
85e94525 | 885 | dev_kfree_skb_any(skb); |
58eba97d | 886 | return NETDEV_TX_OK; |
296f96fc | 887 | } |
681daee2 | 888 | virtqueue_kick(sq->vq); |
03f191ba | 889 | |
48925e37 RR |
890 | /* Don't wait up for transmitted skbs to be freed. */ |
891 | skb_orphan(skb); | |
892 | nf_reset(skb); | |
893 | ||
894 | /* Apparently nice girls don't return TX_BUSY; stop the queue | |
895 | * before it gets out of hand. Naturally, this wastes entries. */ | |
b7dfde95 | 896 | if (sq->vq->num_free < 2+MAX_SKB_FRAGS) { |
986a4f4d | 897 | netif_stop_subqueue(dev, qnum); |
e9d7417b | 898 | if (unlikely(!virtqueue_enable_cb_delayed(sq->vq))) { |
48925e37 | 899 | /* More just got used, free them then recheck. */ |
b7dfde95 LT |
900 | free_old_xmit_skbs(sq); |
901 | if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) { | |
986a4f4d | 902 | netif_start_subqueue(dev, qnum); |
e9d7417b | 903 | virtqueue_disable_cb(sq->vq); |
48925e37 RR |
904 | } |
905 | } | |
99ffc696 | 906 | } |
48925e37 RR |
907 | |
908 | return NETDEV_TX_OK; | |
296f96fc RR |
909 | } |
910 | ||
40cbfc37 AK |
911 | /* |
912 | * Send command via the control virtqueue and check status. Commands | |
913 | * supported by the hypervisor, as indicated by feature bits, should | |
788a8b6d | 914 | * never fail unless improperly formatted. |
40cbfc37 AK |
915 | */ |
916 | static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd, | |
d24bae32 | 917 | struct scatterlist *out) |
40cbfc37 | 918 | { |
f7bc9594 | 919 | struct scatterlist *sgs[4], hdr, stat; |
40cbfc37 AK |
920 | struct virtio_net_ctrl_hdr ctrl; |
921 | virtio_net_ctrl_ack status = ~0; | |
d24bae32 | 922 | unsigned out_num = 0, tmp; |
40cbfc37 AK |
923 | |
924 | /* Caller should know better */ | |
f7bc9594 | 925 | BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)); |
40cbfc37 AK |
926 | |
927 | ctrl.class = class; | |
928 | ctrl.cmd = cmd; | |
f7bc9594 RR |
929 | /* Add header */ |
930 | sg_init_one(&hdr, &ctrl, sizeof(ctrl)); | |
931 | sgs[out_num++] = &hdr; | |
40cbfc37 | 932 | |
f7bc9594 RR |
933 | if (out) |
934 | sgs[out_num++] = out; | |
40cbfc37 | 935 | |
f7bc9594 RR |
936 | /* Add return status. */ |
937 | sg_init_one(&stat, &status, sizeof(status)); | |
d24bae32 | 938 | sgs[out_num] = &stat; |
40cbfc37 | 939 | |
d24bae32 | 940 | BUG_ON(out_num + 1 > ARRAY_SIZE(sgs)); |
a7c58146 | 941 | virtqueue_add_sgs(vi->cvq, sgs, out_num, 1, vi, GFP_ATOMIC); |
40cbfc37 | 942 | |
67975901 HG |
943 | if (unlikely(!virtqueue_kick(vi->cvq))) |
944 | return status == VIRTIO_NET_OK; | |
40cbfc37 AK |
945 | |
946 | /* Spin for a response, the kick causes an ioport write, trapping | |
947 | * into the hypervisor, so the request should be handled immediately. | |
948 | */ | |
047b9b94 HG |
949 | while (!virtqueue_get_buf(vi->cvq, &tmp) && |
950 | !virtqueue_is_broken(vi->cvq)) | |
40cbfc37 AK |
951 | cpu_relax(); |
952 | ||
953 | return status == VIRTIO_NET_OK; | |
954 | } | |
955 | ||
9c46f6d4 AW |
956 | static int virtnet_set_mac_address(struct net_device *dev, void *p) |
957 | { | |
958 | struct virtnet_info *vi = netdev_priv(dev); | |
959 | struct virtio_device *vdev = vi->vdev; | |
f2f2c8b4 | 960 | int ret; |
7e58d5ae AK |
961 | struct sockaddr *addr = p; |
962 | struct scatterlist sg; | |
9c46f6d4 | 963 | |
7e58d5ae | 964 | ret = eth_prepare_mac_addr_change(dev, p); |
f2f2c8b4 JP |
965 | if (ret) |
966 | return ret; | |
9c46f6d4 | 967 | |
7e58d5ae AK |
968 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) { |
969 | sg_init_one(&sg, addr->sa_data, dev->addr_len); | |
970 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC, | |
d24bae32 | 971 | VIRTIO_NET_CTRL_MAC_ADDR_SET, &sg)) { |
7e58d5ae AK |
972 | dev_warn(&vdev->dev, |
973 | "Failed to set mac address by vq command.\n"); | |
974 | return -EINVAL; | |
975 | } | |
976 | } else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) { | |
855e0c52 RR |
977 | unsigned int i; |
978 | ||
979 | /* Naturally, this has an atomicity problem. */ | |
980 | for (i = 0; i < dev->addr_len; i++) | |
981 | virtio_cwrite8(vdev, | |
982 | offsetof(struct virtio_net_config, mac) + | |
983 | i, addr->sa_data[i]); | |
7e58d5ae AK |
984 | } |
985 | ||
986 | eth_commit_mac_addr_change(dev, p); | |
9c46f6d4 AW |
987 | |
988 | return 0; | |
989 | } | |
990 | ||
3fa2a1df | 991 | static struct rtnl_link_stats64 *virtnet_stats(struct net_device *dev, |
992 | struct rtnl_link_stats64 *tot) | |
993 | { | |
994 | struct virtnet_info *vi = netdev_priv(dev); | |
995 | int cpu; | |
996 | unsigned int start; | |
997 | ||
998 | for_each_possible_cpu(cpu) { | |
58472a76 | 999 | struct virtnet_stats *stats = per_cpu_ptr(vi->stats, cpu); |
3fa2a1df | 1000 | u64 tpackets, tbytes, rpackets, rbytes; |
1001 | ||
1002 | do { | |
57a7744e | 1003 | start = u64_stats_fetch_begin_irq(&stats->tx_syncp); |
3fa2a1df | 1004 | tpackets = stats->tx_packets; |
1005 | tbytes = stats->tx_bytes; | |
57a7744e | 1006 | } while (u64_stats_fetch_retry_irq(&stats->tx_syncp, start)); |
83a27052 ED |
1007 | |
1008 | do { | |
57a7744e | 1009 | start = u64_stats_fetch_begin_irq(&stats->rx_syncp); |
3fa2a1df | 1010 | rpackets = stats->rx_packets; |
1011 | rbytes = stats->rx_bytes; | |
57a7744e | 1012 | } while (u64_stats_fetch_retry_irq(&stats->rx_syncp, start)); |
3fa2a1df | 1013 | |
1014 | tot->rx_packets += rpackets; | |
1015 | tot->tx_packets += tpackets; | |
1016 | tot->rx_bytes += rbytes; | |
1017 | tot->tx_bytes += tbytes; | |
1018 | } | |
1019 | ||
1020 | tot->tx_dropped = dev->stats.tx_dropped; | |
021ac8d3 | 1021 | tot->tx_fifo_errors = dev->stats.tx_fifo_errors; |
3fa2a1df | 1022 | tot->rx_dropped = dev->stats.rx_dropped; |
1023 | tot->rx_length_errors = dev->stats.rx_length_errors; | |
1024 | tot->rx_frame_errors = dev->stats.rx_frame_errors; | |
1025 | ||
1026 | return tot; | |
1027 | } | |
1028 | ||
da74e89d AS |
1029 | #ifdef CONFIG_NET_POLL_CONTROLLER |
1030 | static void virtnet_netpoll(struct net_device *dev) | |
1031 | { | |
1032 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d | 1033 | int i; |
da74e89d | 1034 | |
986a4f4d JW |
1035 | for (i = 0; i < vi->curr_queue_pairs; i++) |
1036 | napi_schedule(&vi->rq[i].napi); | |
da74e89d AS |
1037 | } |
1038 | #endif | |
1039 | ||
586d17c5 JW |
1040 | static void virtnet_ack_link_announce(struct virtnet_info *vi) |
1041 | { | |
1042 | rtnl_lock(); | |
1043 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE, | |
d24bae32 | 1044 | VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL)) |
586d17c5 JW |
1045 | dev_warn(&vi->dev->dev, "Failed to ack link announce.\n"); |
1046 | rtnl_unlock(); | |
1047 | } | |
1048 | ||
986a4f4d JW |
1049 | static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs) |
1050 | { | |
1051 | struct scatterlist sg; | |
1052 | struct virtio_net_ctrl_mq s; | |
1053 | struct net_device *dev = vi->dev; | |
1054 | ||
1055 | if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ)) | |
1056 | return 0; | |
1057 | ||
1058 | s.virtqueue_pairs = queue_pairs; | |
1059 | sg_init_one(&sg, &s, sizeof(s)); | |
1060 | ||
1061 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ, | |
d24bae32 | 1062 | VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg)) { |
986a4f4d JW |
1063 | dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n", |
1064 | queue_pairs); | |
1065 | return -EINVAL; | |
55257d72 | 1066 | } else { |
986a4f4d | 1067 | vi->curr_queue_pairs = queue_pairs; |
35ed159b JW |
1068 | /* virtnet_open() will refill when device is going to up. */ |
1069 | if (dev->flags & IFF_UP) | |
1070 | schedule_delayed_work(&vi->refill, 0); | |
55257d72 | 1071 | } |
986a4f4d JW |
1072 | |
1073 | return 0; | |
1074 | } | |
1075 | ||
296f96fc RR |
1076 | static int virtnet_close(struct net_device *dev) |
1077 | { | |
1078 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d | 1079 | int i; |
296f96fc | 1080 | |
b2baed69 RR |
1081 | /* Make sure refill_work doesn't re-enable napi! */ |
1082 | cancel_delayed_work_sync(&vi->refill); | |
986a4f4d JW |
1083 | |
1084 | for (i = 0; i < vi->max_queue_pairs; i++) | |
1085 | napi_disable(&vi->rq[i].napi); | |
296f96fc | 1086 | |
296f96fc RR |
1087 | return 0; |
1088 | } | |
1089 | ||
2af7698e AW |
1090 | static void virtnet_set_rx_mode(struct net_device *dev) |
1091 | { | |
1092 | struct virtnet_info *vi = netdev_priv(dev); | |
f565a7c2 | 1093 | struct scatterlist sg[2]; |
2af7698e | 1094 | u8 promisc, allmulti; |
f565a7c2 | 1095 | struct virtio_net_ctrl_mac *mac_data; |
ccffad25 | 1096 | struct netdev_hw_addr *ha; |
32e7bfc4 | 1097 | int uc_count; |
4cd24eaf | 1098 | int mc_count; |
f565a7c2 AW |
1099 | void *buf; |
1100 | int i; | |
2af7698e | 1101 | |
788a8b6d | 1102 | /* We can't dynamically set ndo_set_rx_mode, so return gracefully */ |
2af7698e AW |
1103 | if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX)) |
1104 | return; | |
1105 | ||
f565a7c2 AW |
1106 | promisc = ((dev->flags & IFF_PROMISC) != 0); |
1107 | allmulti = ((dev->flags & IFF_ALLMULTI) != 0); | |
2af7698e | 1108 | |
23e258e1 | 1109 | sg_init_one(sg, &promisc, sizeof(promisc)); |
2af7698e AW |
1110 | |
1111 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, | |
d24bae32 | 1112 | VIRTIO_NET_CTRL_RX_PROMISC, sg)) |
2af7698e AW |
1113 | dev_warn(&dev->dev, "Failed to %sable promisc mode.\n", |
1114 | promisc ? "en" : "dis"); | |
1115 | ||
23e258e1 | 1116 | sg_init_one(sg, &allmulti, sizeof(allmulti)); |
2af7698e AW |
1117 | |
1118 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, | |
d24bae32 | 1119 | VIRTIO_NET_CTRL_RX_ALLMULTI, sg)) |
2af7698e AW |
1120 | dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n", |
1121 | allmulti ? "en" : "dis"); | |
f565a7c2 | 1122 | |
32e7bfc4 | 1123 | uc_count = netdev_uc_count(dev); |
4cd24eaf | 1124 | mc_count = netdev_mc_count(dev); |
f565a7c2 | 1125 | /* MAC filter - use one buffer for both lists */ |
4cd24eaf JP |
1126 | buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) + |
1127 | (2 * sizeof(mac_data->entries)), GFP_ATOMIC); | |
1128 | mac_data = buf; | |
e68ed8f0 | 1129 | if (!buf) |
f565a7c2 | 1130 | return; |
f565a7c2 | 1131 | |
23e258e1 AW |
1132 | sg_init_table(sg, 2); |
1133 | ||
f565a7c2 | 1134 | /* Store the unicast list and count in the front of the buffer */ |
32e7bfc4 | 1135 | mac_data->entries = uc_count; |
ccffad25 | 1136 | i = 0; |
32e7bfc4 | 1137 | netdev_for_each_uc_addr(ha, dev) |
ccffad25 | 1138 | memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); |
f565a7c2 AW |
1139 | |
1140 | sg_set_buf(&sg[0], mac_data, | |
32e7bfc4 | 1141 | sizeof(mac_data->entries) + (uc_count * ETH_ALEN)); |
f565a7c2 AW |
1142 | |
1143 | /* multicast list and count fill the end */ | |
32e7bfc4 | 1144 | mac_data = (void *)&mac_data->macs[uc_count][0]; |
f565a7c2 | 1145 | |
4cd24eaf | 1146 | mac_data->entries = mc_count; |
567ec874 | 1147 | i = 0; |
22bedad3 JP |
1148 | netdev_for_each_mc_addr(ha, dev) |
1149 | memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); | |
f565a7c2 AW |
1150 | |
1151 | sg_set_buf(&sg[1], mac_data, | |
4cd24eaf | 1152 | sizeof(mac_data->entries) + (mc_count * ETH_ALEN)); |
f565a7c2 AW |
1153 | |
1154 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC, | |
d24bae32 | 1155 | VIRTIO_NET_CTRL_MAC_TABLE_SET, sg)) |
99e872ae | 1156 | dev_warn(&dev->dev, "Failed to set MAC filter table.\n"); |
f565a7c2 AW |
1157 | |
1158 | kfree(buf); | |
2af7698e AW |
1159 | } |
1160 | ||
80d5c368 PM |
1161 | static int virtnet_vlan_rx_add_vid(struct net_device *dev, |
1162 | __be16 proto, u16 vid) | |
0bde9569 AW |
1163 | { |
1164 | struct virtnet_info *vi = netdev_priv(dev); | |
1165 | struct scatterlist sg; | |
1166 | ||
23e258e1 | 1167 | sg_init_one(&sg, &vid, sizeof(vid)); |
0bde9569 AW |
1168 | |
1169 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, | |
d24bae32 | 1170 | VIRTIO_NET_CTRL_VLAN_ADD, &sg)) |
0bde9569 | 1171 | dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid); |
8e586137 | 1172 | return 0; |
0bde9569 AW |
1173 | } |
1174 | ||
80d5c368 PM |
1175 | static int virtnet_vlan_rx_kill_vid(struct net_device *dev, |
1176 | __be16 proto, u16 vid) | |
0bde9569 AW |
1177 | { |
1178 | struct virtnet_info *vi = netdev_priv(dev); | |
1179 | struct scatterlist sg; | |
1180 | ||
23e258e1 | 1181 | sg_init_one(&sg, &vid, sizeof(vid)); |
0bde9569 AW |
1182 | |
1183 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, | |
d24bae32 | 1184 | VIRTIO_NET_CTRL_VLAN_DEL, &sg)) |
0bde9569 | 1185 | dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid); |
8e586137 | 1186 | return 0; |
0bde9569 AW |
1187 | } |
1188 | ||
8898c21c | 1189 | static void virtnet_clean_affinity(struct virtnet_info *vi, long hcpu) |
986a4f4d JW |
1190 | { |
1191 | int i; | |
1192 | ||
8898c21c WG |
1193 | if (vi->affinity_hint_set) { |
1194 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
47be2479 WG |
1195 | virtqueue_set_affinity(vi->rq[i].vq, -1); |
1196 | virtqueue_set_affinity(vi->sq[i].vq, -1); | |
1197 | } | |
1198 | ||
8898c21c WG |
1199 | vi->affinity_hint_set = false; |
1200 | } | |
8898c21c | 1201 | } |
47be2479 | 1202 | |
8898c21c WG |
1203 | static void virtnet_set_affinity(struct virtnet_info *vi) |
1204 | { | |
1205 | int i; | |
1206 | int cpu; | |
986a4f4d JW |
1207 | |
1208 | /* In multiqueue mode, when the number of cpu is equal to the number of | |
1209 | * queue pairs, we let the queue pairs to be private to one cpu by | |
1210 | * setting the affinity hint to eliminate the contention. | |
1211 | */ | |
8898c21c WG |
1212 | if (vi->curr_queue_pairs == 1 || |
1213 | vi->max_queue_pairs != num_online_cpus()) { | |
1214 | virtnet_clean_affinity(vi, -1); | |
1215 | return; | |
986a4f4d JW |
1216 | } |
1217 | ||
8898c21c WG |
1218 | i = 0; |
1219 | for_each_online_cpu(cpu) { | |
986a4f4d JW |
1220 | virtqueue_set_affinity(vi->rq[i].vq, cpu); |
1221 | virtqueue_set_affinity(vi->sq[i].vq, cpu); | |
9bb8ca86 | 1222 | netif_set_xps_queue(vi->dev, cpumask_of(cpu), i); |
8898c21c | 1223 | i++; |
986a4f4d JW |
1224 | } |
1225 | ||
8898c21c | 1226 | vi->affinity_hint_set = true; |
986a4f4d JW |
1227 | } |
1228 | ||
8de4b2f3 WG |
1229 | static int virtnet_cpu_callback(struct notifier_block *nfb, |
1230 | unsigned long action, void *hcpu) | |
1231 | { | |
1232 | struct virtnet_info *vi = container_of(nfb, struct virtnet_info, nb); | |
1233 | ||
1234 | switch(action & ~CPU_TASKS_FROZEN) { | |
1235 | case CPU_ONLINE: | |
1236 | case CPU_DOWN_FAILED: | |
1237 | case CPU_DEAD: | |
1238 | virtnet_set_affinity(vi); | |
1239 | break; | |
1240 | case CPU_DOWN_PREPARE: | |
1241 | virtnet_clean_affinity(vi, (long)hcpu); | |
1242 | break; | |
1243 | default: | |
1244 | break; | |
1245 | } | |
3ab098df | 1246 | |
8de4b2f3 | 1247 | return NOTIFY_OK; |
986a4f4d JW |
1248 | } |
1249 | ||
8f9f4668 RJ |
1250 | static void virtnet_get_ringparam(struct net_device *dev, |
1251 | struct ethtool_ringparam *ring) | |
1252 | { | |
1253 | struct virtnet_info *vi = netdev_priv(dev); | |
1254 | ||
986a4f4d JW |
1255 | ring->rx_max_pending = virtqueue_get_vring_size(vi->rq[0].vq); |
1256 | ring->tx_max_pending = virtqueue_get_vring_size(vi->sq[0].vq); | |
8f9f4668 RJ |
1257 | ring->rx_pending = ring->rx_max_pending; |
1258 | ring->tx_pending = ring->tx_max_pending; | |
8f9f4668 RJ |
1259 | } |
1260 | ||
66846048 RJ |
1261 | |
1262 | static void virtnet_get_drvinfo(struct net_device *dev, | |
1263 | struct ethtool_drvinfo *info) | |
1264 | { | |
1265 | struct virtnet_info *vi = netdev_priv(dev); | |
1266 | struct virtio_device *vdev = vi->vdev; | |
1267 | ||
1268 | strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver)); | |
1269 | strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version)); | |
1270 | strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info)); | |
1271 | ||
1272 | } | |
1273 | ||
d73bcd2c JW |
1274 | /* TODO: Eliminate OOO packets during switching */ |
1275 | static int virtnet_set_channels(struct net_device *dev, | |
1276 | struct ethtool_channels *channels) | |
1277 | { | |
1278 | struct virtnet_info *vi = netdev_priv(dev); | |
1279 | u16 queue_pairs = channels->combined_count; | |
1280 | int err; | |
1281 | ||
1282 | /* We don't support separate rx/tx channels. | |
1283 | * We don't allow setting 'other' channels. | |
1284 | */ | |
1285 | if (channels->rx_count || channels->tx_count || channels->other_count) | |
1286 | return -EINVAL; | |
1287 | ||
c18e9cd6 | 1288 | if (queue_pairs > vi->max_queue_pairs || queue_pairs == 0) |
d73bcd2c JW |
1289 | return -EINVAL; |
1290 | ||
47be2479 | 1291 | get_online_cpus(); |
d73bcd2c JW |
1292 | err = virtnet_set_queues(vi, queue_pairs); |
1293 | if (!err) { | |
1294 | netif_set_real_num_tx_queues(dev, queue_pairs); | |
1295 | netif_set_real_num_rx_queues(dev, queue_pairs); | |
1296 | ||
8898c21c | 1297 | virtnet_set_affinity(vi); |
d73bcd2c | 1298 | } |
47be2479 | 1299 | put_online_cpus(); |
d73bcd2c JW |
1300 | |
1301 | return err; | |
1302 | } | |
1303 | ||
1304 | static void virtnet_get_channels(struct net_device *dev, | |
1305 | struct ethtool_channels *channels) | |
1306 | { | |
1307 | struct virtnet_info *vi = netdev_priv(dev); | |
1308 | ||
1309 | channels->combined_count = vi->curr_queue_pairs; | |
1310 | channels->max_combined = vi->max_queue_pairs; | |
1311 | channels->max_other = 0; | |
1312 | channels->rx_count = 0; | |
1313 | channels->tx_count = 0; | |
1314 | channels->other_count = 0; | |
1315 | } | |
1316 | ||
0fc0b732 | 1317 | static const struct ethtool_ops virtnet_ethtool_ops = { |
66846048 | 1318 | .get_drvinfo = virtnet_get_drvinfo, |
9f4d26d0 | 1319 | .get_link = ethtool_op_get_link, |
8f9f4668 | 1320 | .get_ringparam = virtnet_get_ringparam, |
d73bcd2c JW |
1321 | .set_channels = virtnet_set_channels, |
1322 | .get_channels = virtnet_get_channels, | |
a9ea3fc6 HX |
1323 | }; |
1324 | ||
39da5814 MM |
1325 | #define MIN_MTU 68 |
1326 | #define MAX_MTU 65535 | |
1327 | ||
1328 | static int virtnet_change_mtu(struct net_device *dev, int new_mtu) | |
1329 | { | |
1330 | if (new_mtu < MIN_MTU || new_mtu > MAX_MTU) | |
1331 | return -EINVAL; | |
1332 | dev->mtu = new_mtu; | |
1333 | return 0; | |
1334 | } | |
1335 | ||
76288b4e SH |
1336 | static const struct net_device_ops virtnet_netdev = { |
1337 | .ndo_open = virtnet_open, | |
1338 | .ndo_stop = virtnet_close, | |
1339 | .ndo_start_xmit = start_xmit, | |
1340 | .ndo_validate_addr = eth_validate_addr, | |
9c46f6d4 | 1341 | .ndo_set_mac_address = virtnet_set_mac_address, |
2af7698e | 1342 | .ndo_set_rx_mode = virtnet_set_rx_mode, |
76288b4e | 1343 | .ndo_change_mtu = virtnet_change_mtu, |
3fa2a1df | 1344 | .ndo_get_stats64 = virtnet_stats, |
1824a989 AW |
1345 | .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid, |
1346 | .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid, | |
76288b4e SH |
1347 | #ifdef CONFIG_NET_POLL_CONTROLLER |
1348 | .ndo_poll_controller = virtnet_netpoll, | |
1349 | #endif | |
1350 | }; | |
1351 | ||
586d17c5 | 1352 | static void virtnet_config_changed_work(struct work_struct *work) |
9f4d26d0 | 1353 | { |
586d17c5 JW |
1354 | struct virtnet_info *vi = |
1355 | container_of(work, struct virtnet_info, config_work); | |
9f4d26d0 MM |
1356 | u16 v; |
1357 | ||
586d17c5 JW |
1358 | mutex_lock(&vi->config_lock); |
1359 | if (!vi->config_enable) | |
1360 | goto done; | |
1361 | ||
855e0c52 RR |
1362 | if (virtio_cread_feature(vi->vdev, VIRTIO_NET_F_STATUS, |
1363 | struct virtio_net_config, status, &v) < 0) | |
586d17c5 JW |
1364 | goto done; |
1365 | ||
1366 | if (v & VIRTIO_NET_S_ANNOUNCE) { | |
ee89bab1 | 1367 | netdev_notify_peers(vi->dev); |
586d17c5 JW |
1368 | virtnet_ack_link_announce(vi); |
1369 | } | |
9f4d26d0 MM |
1370 | |
1371 | /* Ignore unknown (future) status bits */ | |
1372 | v &= VIRTIO_NET_S_LINK_UP; | |
1373 | ||
1374 | if (vi->status == v) | |
586d17c5 | 1375 | goto done; |
9f4d26d0 MM |
1376 | |
1377 | vi->status = v; | |
1378 | ||
1379 | if (vi->status & VIRTIO_NET_S_LINK_UP) { | |
1380 | netif_carrier_on(vi->dev); | |
986a4f4d | 1381 | netif_tx_wake_all_queues(vi->dev); |
9f4d26d0 MM |
1382 | } else { |
1383 | netif_carrier_off(vi->dev); | |
986a4f4d | 1384 | netif_tx_stop_all_queues(vi->dev); |
9f4d26d0 | 1385 | } |
586d17c5 JW |
1386 | done: |
1387 | mutex_unlock(&vi->config_lock); | |
9f4d26d0 MM |
1388 | } |
1389 | ||
1390 | static void virtnet_config_changed(struct virtio_device *vdev) | |
1391 | { | |
1392 | struct virtnet_info *vi = vdev->priv; | |
1393 | ||
3b07e9ca | 1394 | schedule_work(&vi->config_work); |
9f4d26d0 MM |
1395 | } |
1396 | ||
986a4f4d JW |
1397 | static void virtnet_free_queues(struct virtnet_info *vi) |
1398 | { | |
d4fb84ee AV |
1399 | int i; |
1400 | ||
1401 | for (i = 0; i < vi->max_queue_pairs; i++) | |
1402 | netif_napi_del(&vi->rq[i].napi); | |
1403 | ||
986a4f4d JW |
1404 | kfree(vi->rq); |
1405 | kfree(vi->sq); | |
1406 | } | |
1407 | ||
1408 | static void free_receive_bufs(struct virtnet_info *vi) | |
1409 | { | |
1410 | int i; | |
1411 | ||
1412 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
1413 | while (vi->rq[i].pages) | |
1414 | __free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0); | |
1415 | } | |
1416 | } | |
1417 | ||
fb51879d MD |
1418 | static void free_receive_page_frags(struct virtnet_info *vi) |
1419 | { | |
1420 | int i; | |
1421 | for (i = 0; i < vi->max_queue_pairs; i++) | |
1422 | if (vi->rq[i].alloc_frag.page) | |
1423 | put_page(vi->rq[i].alloc_frag.page); | |
1424 | } | |
1425 | ||
986a4f4d JW |
1426 | static void free_unused_bufs(struct virtnet_info *vi) |
1427 | { | |
1428 | void *buf; | |
1429 | int i; | |
1430 | ||
1431 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
1432 | struct virtqueue *vq = vi->sq[i].vq; | |
1433 | while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) | |
1434 | dev_kfree_skb(buf); | |
1435 | } | |
1436 | ||
1437 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
1438 | struct virtqueue *vq = vi->rq[i].vq; | |
1439 | ||
1440 | while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) { | |
ab7db917 MD |
1441 | if (vi->mergeable_rx_bufs) { |
1442 | unsigned long ctx = (unsigned long)buf; | |
1443 | void *base = mergeable_ctx_to_buf_address(ctx); | |
1444 | put_page(virt_to_head_page(base)); | |
1445 | } else if (vi->big_packets) { | |
fa9fac17 | 1446 | give_pages(&vi->rq[i], buf); |
ab7db917 | 1447 | } else { |
986a4f4d | 1448 | dev_kfree_skb(buf); |
ab7db917 | 1449 | } |
986a4f4d | 1450 | } |
986a4f4d JW |
1451 | } |
1452 | } | |
1453 | ||
e9d7417b JW |
1454 | static void virtnet_del_vqs(struct virtnet_info *vi) |
1455 | { | |
1456 | struct virtio_device *vdev = vi->vdev; | |
1457 | ||
8898c21c | 1458 | virtnet_clean_affinity(vi, -1); |
986a4f4d | 1459 | |
e9d7417b | 1460 | vdev->config->del_vqs(vdev); |
986a4f4d JW |
1461 | |
1462 | virtnet_free_queues(vi); | |
e9d7417b JW |
1463 | } |
1464 | ||
986a4f4d | 1465 | static int virtnet_find_vqs(struct virtnet_info *vi) |
3f9c10b0 | 1466 | { |
986a4f4d JW |
1467 | vq_callback_t **callbacks; |
1468 | struct virtqueue **vqs; | |
1469 | int ret = -ENOMEM; | |
1470 | int i, total_vqs; | |
1471 | const char **names; | |
1472 | ||
1473 | /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by | |
1474 | * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by | |
1475 | * possible control vq. | |
1476 | */ | |
1477 | total_vqs = vi->max_queue_pairs * 2 + | |
1478 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ); | |
1479 | ||
1480 | /* Allocate space for find_vqs parameters */ | |
1481 | vqs = kzalloc(total_vqs * sizeof(*vqs), GFP_KERNEL); | |
1482 | if (!vqs) | |
1483 | goto err_vq; | |
1484 | callbacks = kmalloc(total_vqs * sizeof(*callbacks), GFP_KERNEL); | |
1485 | if (!callbacks) | |
1486 | goto err_callback; | |
1487 | names = kmalloc(total_vqs * sizeof(*names), GFP_KERNEL); | |
1488 | if (!names) | |
1489 | goto err_names; | |
1490 | ||
1491 | /* Parameters for control virtqueue, if any */ | |
1492 | if (vi->has_cvq) { | |
1493 | callbacks[total_vqs - 1] = NULL; | |
1494 | names[total_vqs - 1] = "control"; | |
1495 | } | |
3f9c10b0 | 1496 | |
986a4f4d JW |
1497 | /* Allocate/initialize parameters for send/receive virtqueues */ |
1498 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
1499 | callbacks[rxq2vq(i)] = skb_recv_done; | |
1500 | callbacks[txq2vq(i)] = skb_xmit_done; | |
1501 | sprintf(vi->rq[i].name, "input.%d", i); | |
1502 | sprintf(vi->sq[i].name, "output.%d", i); | |
1503 | names[rxq2vq(i)] = vi->rq[i].name; | |
1504 | names[txq2vq(i)] = vi->sq[i].name; | |
1505 | } | |
3f9c10b0 | 1506 | |
986a4f4d JW |
1507 | ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs, callbacks, |
1508 | names); | |
1509 | if (ret) | |
1510 | goto err_find; | |
3f9c10b0 | 1511 | |
986a4f4d JW |
1512 | if (vi->has_cvq) { |
1513 | vi->cvq = vqs[total_vqs - 1]; | |
3f9c10b0 | 1514 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN)) |
f646968f | 1515 | vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER; |
3f9c10b0 | 1516 | } |
986a4f4d JW |
1517 | |
1518 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
1519 | vi->rq[i].vq = vqs[rxq2vq(i)]; | |
1520 | vi->sq[i].vq = vqs[txq2vq(i)]; | |
1521 | } | |
1522 | ||
1523 | kfree(names); | |
1524 | kfree(callbacks); | |
1525 | kfree(vqs); | |
1526 | ||
3f9c10b0 | 1527 | return 0; |
986a4f4d JW |
1528 | |
1529 | err_find: | |
1530 | kfree(names); | |
1531 | err_names: | |
1532 | kfree(callbacks); | |
1533 | err_callback: | |
1534 | kfree(vqs); | |
1535 | err_vq: | |
1536 | return ret; | |
1537 | } | |
1538 | ||
1539 | static int virtnet_alloc_queues(struct virtnet_info *vi) | |
1540 | { | |
1541 | int i; | |
1542 | ||
1543 | vi->sq = kzalloc(sizeof(*vi->sq) * vi->max_queue_pairs, GFP_KERNEL); | |
1544 | if (!vi->sq) | |
1545 | goto err_sq; | |
1546 | vi->rq = kzalloc(sizeof(*vi->rq) * vi->max_queue_pairs, GFP_KERNEL); | |
008d4278 | 1547 | if (!vi->rq) |
986a4f4d JW |
1548 | goto err_rq; |
1549 | ||
1550 | INIT_DELAYED_WORK(&vi->refill, refill_work); | |
1551 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
1552 | vi->rq[i].pages = NULL; | |
1553 | netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll, | |
1554 | napi_weight); | |
1555 | ||
1556 | sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg)); | |
ab7db917 | 1557 | ewma_init(&vi->rq[i].mrg_avg_pkt_len, 1, RECEIVE_AVG_WEIGHT); |
986a4f4d JW |
1558 | sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg)); |
1559 | } | |
1560 | ||
1561 | return 0; | |
1562 | ||
1563 | err_rq: | |
1564 | kfree(vi->sq); | |
1565 | err_sq: | |
1566 | return -ENOMEM; | |
1567 | } | |
1568 | ||
1569 | static int init_vqs(struct virtnet_info *vi) | |
1570 | { | |
1571 | int ret; | |
1572 | ||
1573 | /* Allocate send & receive queues */ | |
1574 | ret = virtnet_alloc_queues(vi); | |
1575 | if (ret) | |
1576 | goto err; | |
1577 | ||
1578 | ret = virtnet_find_vqs(vi); | |
1579 | if (ret) | |
1580 | goto err_free; | |
1581 | ||
47be2479 | 1582 | get_online_cpus(); |
8898c21c | 1583 | virtnet_set_affinity(vi); |
47be2479 WG |
1584 | put_online_cpus(); |
1585 | ||
986a4f4d JW |
1586 | return 0; |
1587 | ||
1588 | err_free: | |
1589 | virtnet_free_queues(vi); | |
1590 | err: | |
1591 | return ret; | |
3f9c10b0 AS |
1592 | } |
1593 | ||
fbf28d78 MD |
1594 | #ifdef CONFIG_SYSFS |
1595 | static ssize_t mergeable_rx_buffer_size_show(struct netdev_rx_queue *queue, | |
1596 | struct rx_queue_attribute *attribute, char *buf) | |
1597 | { | |
1598 | struct virtnet_info *vi = netdev_priv(queue->dev); | |
1599 | unsigned int queue_index = get_netdev_rx_queue_index(queue); | |
1600 | struct ewma *avg; | |
1601 | ||
1602 | BUG_ON(queue_index >= vi->max_queue_pairs); | |
1603 | avg = &vi->rq[queue_index].mrg_avg_pkt_len; | |
1604 | return sprintf(buf, "%u\n", get_mergeable_buf_len(avg)); | |
1605 | } | |
1606 | ||
1607 | static struct rx_queue_attribute mergeable_rx_buffer_size_attribute = | |
1608 | __ATTR_RO(mergeable_rx_buffer_size); | |
1609 | ||
1610 | static struct attribute *virtio_net_mrg_rx_attrs[] = { | |
1611 | &mergeable_rx_buffer_size_attribute.attr, | |
1612 | NULL | |
1613 | }; | |
1614 | ||
1615 | static const struct attribute_group virtio_net_mrg_rx_group = { | |
1616 | .name = "virtio_net", | |
1617 | .attrs = virtio_net_mrg_rx_attrs | |
1618 | }; | |
1619 | #endif | |
1620 | ||
296f96fc RR |
1621 | static int virtnet_probe(struct virtio_device *vdev) |
1622 | { | |
986a4f4d | 1623 | int i, err; |
296f96fc RR |
1624 | struct net_device *dev; |
1625 | struct virtnet_info *vi; | |
986a4f4d JW |
1626 | u16 max_queue_pairs; |
1627 | ||
1628 | /* Find if host supports multiqueue virtio_net device */ | |
855e0c52 RR |
1629 | err = virtio_cread_feature(vdev, VIRTIO_NET_F_MQ, |
1630 | struct virtio_net_config, | |
1631 | max_virtqueue_pairs, &max_queue_pairs); | |
986a4f4d JW |
1632 | |
1633 | /* We need at least 2 queue's */ | |
1634 | if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN || | |
1635 | max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX || | |
1636 | !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) | |
1637 | max_queue_pairs = 1; | |
296f96fc RR |
1638 | |
1639 | /* Allocate ourselves a network device with room for our info */ | |
986a4f4d | 1640 | dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs); |
296f96fc RR |
1641 | if (!dev) |
1642 | return -ENOMEM; | |
1643 | ||
1644 | /* Set up network device as normal. */ | |
f2f2c8b4 | 1645 | dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE; |
76288b4e | 1646 | dev->netdev_ops = &virtnet_netdev; |
296f96fc | 1647 | dev->features = NETIF_F_HIGHDMA; |
3fa2a1df | 1648 | |
a9ea3fc6 | 1649 | SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops); |
296f96fc RR |
1650 | SET_NETDEV_DEV(dev, &vdev->dev); |
1651 | ||
1652 | /* Do we support "hardware" checksums? */ | |
98e778c9 | 1653 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) { |
296f96fc | 1654 | /* This opens up the world of extra features. */ |
98e778c9 MM |
1655 | dev->hw_features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST; |
1656 | if (csum) | |
1657 | dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST; | |
1658 | ||
1659 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) { | |
1660 | dev->hw_features |= NETIF_F_TSO | NETIF_F_UFO | |
34a48579 RR |
1661 | | NETIF_F_TSO_ECN | NETIF_F_TSO6; |
1662 | } | |
5539ae96 | 1663 | /* Individual feature bits: what can host handle? */ |
98e778c9 MM |
1664 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4)) |
1665 | dev->hw_features |= NETIF_F_TSO; | |
1666 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6)) | |
1667 | dev->hw_features |= NETIF_F_TSO6; | |
1668 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN)) | |
1669 | dev->hw_features |= NETIF_F_TSO_ECN; | |
1670 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO)) | |
1671 | dev->hw_features |= NETIF_F_UFO; | |
1672 | ||
1673 | if (gso) | |
1674 | dev->features |= dev->hw_features & (NETIF_F_ALL_TSO|NETIF_F_UFO); | |
1675 | /* (!csum && gso) case will be fixed by register_netdev() */ | |
296f96fc | 1676 | } |
4f49129b TH |
1677 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM)) |
1678 | dev->features |= NETIF_F_RXCSUM; | |
296f96fc | 1679 | |
4fda8302 JW |
1680 | dev->vlan_features = dev->features; |
1681 | ||
296f96fc | 1682 | /* Configuration may specify what MAC to use. Otherwise random. */ |
855e0c52 RR |
1683 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) |
1684 | virtio_cread_bytes(vdev, | |
1685 | offsetof(struct virtio_net_config, mac), | |
1686 | dev->dev_addr, dev->addr_len); | |
1687 | else | |
f2cedb63 | 1688 | eth_hw_addr_random(dev); |
296f96fc RR |
1689 | |
1690 | /* Set up our device-specific information */ | |
1691 | vi = netdev_priv(dev); | |
296f96fc RR |
1692 | vi->dev = dev; |
1693 | vi->vdev = vdev; | |
d9d5dcc8 | 1694 | vdev->priv = vi; |
3fa2a1df | 1695 | vi->stats = alloc_percpu(struct virtnet_stats); |
1696 | err = -ENOMEM; | |
1697 | if (vi->stats == NULL) | |
1698 | goto free; | |
1699 | ||
827da44c JS |
1700 | for_each_possible_cpu(i) { |
1701 | struct virtnet_stats *virtnet_stats; | |
1702 | virtnet_stats = per_cpu_ptr(vi->stats, i); | |
1703 | u64_stats_init(&virtnet_stats->tx_syncp); | |
1704 | u64_stats_init(&virtnet_stats->rx_syncp); | |
1705 | } | |
1706 | ||
586d17c5 JW |
1707 | mutex_init(&vi->config_lock); |
1708 | vi->config_enable = true; | |
1709 | INIT_WORK(&vi->config_work, virtnet_config_changed_work); | |
296f96fc | 1710 | |
97402b96 | 1711 | /* If we can receive ANY GSO packets, we must allocate large ones. */ |
8e95a202 JP |
1712 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) || |
1713 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) || | |
0e7ede80 JW |
1714 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN) || |
1715 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_UFO)) | |
97402b96 HX |
1716 | vi->big_packets = true; |
1717 | ||
3f2c31d9 MM |
1718 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF)) |
1719 | vi->mergeable_rx_bufs = true; | |
1720 | ||
e7428e95 MT |
1721 | if (virtio_has_feature(vdev, VIRTIO_F_ANY_LAYOUT)) |
1722 | vi->any_header_sg = true; | |
1723 | ||
986a4f4d JW |
1724 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) |
1725 | vi->has_cvq = true; | |
1726 | ||
6ebbc1a6 ZH |
1727 | if (vi->any_header_sg) { |
1728 | if (vi->mergeable_rx_bufs) | |
1729 | dev->needed_headroom = sizeof(struct virtio_net_hdr_mrg_rxbuf); | |
1730 | else | |
1731 | dev->needed_headroom = sizeof(struct virtio_net_hdr); | |
1732 | } | |
1733 | ||
986a4f4d JW |
1734 | /* Use single tx/rx queue pair as default */ |
1735 | vi->curr_queue_pairs = 1; | |
1736 | vi->max_queue_pairs = max_queue_pairs; | |
1737 | ||
1738 | /* Allocate/initialize the rx/tx queues, and invoke find_vqs */ | |
3f9c10b0 | 1739 | err = init_vqs(vi); |
d2a7ddda | 1740 | if (err) |
9bb8ca86 | 1741 | goto free_stats; |
296f96fc | 1742 | |
fbf28d78 MD |
1743 | #ifdef CONFIG_SYSFS |
1744 | if (vi->mergeable_rx_bufs) | |
1745 | dev->sysfs_rx_queue_group = &virtio_net_mrg_rx_group; | |
1746 | #endif | |
0f13b66b ZYW |
1747 | netif_set_real_num_tx_queues(dev, vi->curr_queue_pairs); |
1748 | netif_set_real_num_rx_queues(dev, vi->curr_queue_pairs); | |
986a4f4d | 1749 | |
296f96fc RR |
1750 | err = register_netdev(dev); |
1751 | if (err) { | |
1752 | pr_debug("virtio_net: registering device failed\n"); | |
d2a7ddda | 1753 | goto free_vqs; |
296f96fc | 1754 | } |
b3369c1f RR |
1755 | |
1756 | /* Last of all, set up some receive buffers. */ | |
55257d72 | 1757 | for (i = 0; i < vi->curr_queue_pairs; i++) { |
986a4f4d JW |
1758 | try_fill_recv(&vi->rq[i], GFP_KERNEL); |
1759 | ||
1760 | /* If we didn't even get one input buffer, we're useless. */ | |
be121f46 JW |
1761 | if (vi->rq[i].vq->num_free == |
1762 | virtqueue_get_vring_size(vi->rq[i].vq)) { | |
986a4f4d JW |
1763 | free_unused_bufs(vi); |
1764 | err = -ENOMEM; | |
1765 | goto free_recv_bufs; | |
1766 | } | |
b3369c1f RR |
1767 | } |
1768 | ||
8de4b2f3 WG |
1769 | vi->nb.notifier_call = &virtnet_cpu_callback; |
1770 | err = register_hotcpu_notifier(&vi->nb); | |
1771 | if (err) { | |
1772 | pr_debug("virtio_net: registering cpu notifier failed\n"); | |
1773 | goto free_recv_bufs; | |
1774 | } | |
1775 | ||
167c25e4 JW |
1776 | /* Assume link up if device can't report link status, |
1777 | otherwise get link status from config. */ | |
1778 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) { | |
1779 | netif_carrier_off(dev); | |
3b07e9ca | 1780 | schedule_work(&vi->config_work); |
167c25e4 JW |
1781 | } else { |
1782 | vi->status = VIRTIO_NET_S_LINK_UP; | |
1783 | netif_carrier_on(dev); | |
1784 | } | |
9f4d26d0 | 1785 | |
986a4f4d JW |
1786 | pr_debug("virtnet: registered device %s with %d RX and TX vq's\n", |
1787 | dev->name, max_queue_pairs); | |
1788 | ||
296f96fc RR |
1789 | return 0; |
1790 | ||
986a4f4d JW |
1791 | free_recv_bufs: |
1792 | free_receive_bufs(vi); | |
b3369c1f | 1793 | unregister_netdev(dev); |
d2a7ddda | 1794 | free_vqs: |
986a4f4d | 1795 | cancel_delayed_work_sync(&vi->refill); |
fb51879d | 1796 | free_receive_page_frags(vi); |
e9d7417b | 1797 | virtnet_del_vqs(vi); |
3fa2a1df | 1798 | free_stats: |
1799 | free_percpu(vi->stats); | |
296f96fc RR |
1800 | free: |
1801 | free_netdev(dev); | |
1802 | return err; | |
1803 | } | |
1804 | ||
04486ed0 | 1805 | static void remove_vq_common(struct virtnet_info *vi) |
296f96fc | 1806 | { |
04486ed0 | 1807 | vi->vdev->config->reset(vi->vdev); |
830a8a97 SM |
1808 | |
1809 | /* Free unused buffers in both send and recv, if any. */ | |
9ab86bbc | 1810 | free_unused_bufs(vi); |
fb6813f4 | 1811 | |
986a4f4d | 1812 | free_receive_bufs(vi); |
d2a7ddda | 1813 | |
fb51879d MD |
1814 | free_receive_page_frags(vi); |
1815 | ||
986a4f4d | 1816 | virtnet_del_vqs(vi); |
04486ed0 AS |
1817 | } |
1818 | ||
8cc085d6 | 1819 | static void virtnet_remove(struct virtio_device *vdev) |
04486ed0 AS |
1820 | { |
1821 | struct virtnet_info *vi = vdev->priv; | |
1822 | ||
8de4b2f3 WG |
1823 | unregister_hotcpu_notifier(&vi->nb); |
1824 | ||
586d17c5 JW |
1825 | /* Prevent config work handler from accessing the device. */ |
1826 | mutex_lock(&vi->config_lock); | |
1827 | vi->config_enable = false; | |
1828 | mutex_unlock(&vi->config_lock); | |
1829 | ||
04486ed0 AS |
1830 | unregister_netdev(vi->dev); |
1831 | ||
1832 | remove_vq_common(vi); | |
fb6813f4 | 1833 | |
586d17c5 JW |
1834 | flush_work(&vi->config_work); |
1835 | ||
2e66f55b | 1836 | free_percpu(vi->stats); |
74b2553f | 1837 | free_netdev(vi->dev); |
296f96fc RR |
1838 | } |
1839 | ||
89107000 | 1840 | #ifdef CONFIG_PM_SLEEP |
0741bcb5 AS |
1841 | static int virtnet_freeze(struct virtio_device *vdev) |
1842 | { | |
1843 | struct virtnet_info *vi = vdev->priv; | |
986a4f4d | 1844 | int i; |
0741bcb5 | 1845 | |
ec9debbd JW |
1846 | unregister_hotcpu_notifier(&vi->nb); |
1847 | ||
586d17c5 JW |
1848 | /* Prevent config work handler from accessing the device */ |
1849 | mutex_lock(&vi->config_lock); | |
1850 | vi->config_enable = false; | |
1851 | mutex_unlock(&vi->config_lock); | |
1852 | ||
0741bcb5 AS |
1853 | netif_device_detach(vi->dev); |
1854 | cancel_delayed_work_sync(&vi->refill); | |
1855 | ||
1856 | if (netif_running(vi->dev)) | |
986a4f4d JW |
1857 | for (i = 0; i < vi->max_queue_pairs; i++) { |
1858 | napi_disable(&vi->rq[i].napi); | |
1859 | netif_napi_del(&vi->rq[i].napi); | |
1860 | } | |
0741bcb5 AS |
1861 | |
1862 | remove_vq_common(vi); | |
1863 | ||
586d17c5 JW |
1864 | flush_work(&vi->config_work); |
1865 | ||
0741bcb5 AS |
1866 | return 0; |
1867 | } | |
1868 | ||
1869 | static int virtnet_restore(struct virtio_device *vdev) | |
1870 | { | |
1871 | struct virtnet_info *vi = vdev->priv; | |
986a4f4d | 1872 | int err, i; |
0741bcb5 AS |
1873 | |
1874 | err = init_vqs(vi); | |
1875 | if (err) | |
1876 | return err; | |
1877 | ||
6cd4ce00 JW |
1878 | if (netif_running(vi->dev)) { |
1879 | for (i = 0; i < vi->curr_queue_pairs; i++) | |
1880 | if (!try_fill_recv(&vi->rq[i], GFP_KERNEL)) | |
1881 | schedule_delayed_work(&vi->refill, 0); | |
1882 | ||
986a4f4d JW |
1883 | for (i = 0; i < vi->max_queue_pairs; i++) |
1884 | virtnet_napi_enable(&vi->rq[i]); | |
6cd4ce00 | 1885 | } |
0741bcb5 AS |
1886 | |
1887 | netif_device_attach(vi->dev); | |
1888 | ||
586d17c5 JW |
1889 | mutex_lock(&vi->config_lock); |
1890 | vi->config_enable = true; | |
1891 | mutex_unlock(&vi->config_lock); | |
1892 | ||
35ed159b | 1893 | rtnl_lock(); |
986a4f4d | 1894 | virtnet_set_queues(vi, vi->curr_queue_pairs); |
35ed159b | 1895 | rtnl_unlock(); |
986a4f4d | 1896 | |
ec9debbd JW |
1897 | err = register_hotcpu_notifier(&vi->nb); |
1898 | if (err) | |
1899 | return err; | |
1900 | ||
0741bcb5 AS |
1901 | return 0; |
1902 | } | |
1903 | #endif | |
1904 | ||
296f96fc RR |
1905 | static struct virtio_device_id id_table[] = { |
1906 | { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID }, | |
1907 | { 0 }, | |
1908 | }; | |
1909 | ||
c45a6816 | 1910 | static unsigned int features[] = { |
5e4fe5c4 MM |
1911 | VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM, |
1912 | VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC, | |
c45a6816 | 1913 | VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6, |
97402b96 | 1914 | VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, |
5c516751 | 1915 | VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO, |
2a41f71d | 1916 | VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ, |
0bde9569 | 1917 | VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, |
986a4f4d | 1918 | VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ, |
7e58d5ae | 1919 | VIRTIO_NET_F_CTRL_MAC_ADDR, |
e7428e95 | 1920 | VIRTIO_F_ANY_LAYOUT, |
c45a6816 RR |
1921 | }; |
1922 | ||
22402529 | 1923 | static struct virtio_driver virtio_net_driver = { |
c45a6816 RR |
1924 | .feature_table = features, |
1925 | .feature_table_size = ARRAY_SIZE(features), | |
296f96fc RR |
1926 | .driver.name = KBUILD_MODNAME, |
1927 | .driver.owner = THIS_MODULE, | |
1928 | .id_table = id_table, | |
1929 | .probe = virtnet_probe, | |
8cc085d6 | 1930 | .remove = virtnet_remove, |
9f4d26d0 | 1931 | .config_changed = virtnet_config_changed, |
89107000 | 1932 | #ifdef CONFIG_PM_SLEEP |
0741bcb5 AS |
1933 | .freeze = virtnet_freeze, |
1934 | .restore = virtnet_restore, | |
1935 | #endif | |
296f96fc RR |
1936 | }; |
1937 | ||
b2a17029 | 1938 | module_virtio_driver(virtio_net_driver); |
296f96fc RR |
1939 | |
1940 | MODULE_DEVICE_TABLE(virtio, id_table); | |
1941 | MODULE_DESCRIPTION("Virtio network driver"); | |
1942 | MODULE_LICENSE("GPL"); |