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> |
91815639 | 30 | #include <net/busy_poll.h> |
296f96fc | 31 | |
d34710e3 | 32 | static int napi_weight = NAPI_POLL_WEIGHT; |
6c0cd7c0 DL |
33 | module_param(napi_weight, int, 0444); |
34 | ||
eb939922 | 35 | static bool csum = true, gso = true; |
34a48579 RR |
36 | module_param(csum, bool, 0444); |
37 | module_param(gso, bool, 0444); | |
38 | ||
296f96fc | 39 | /* FIXME: MTU in config. */ |
5061de36 | 40 | #define GOOD_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN) |
3f2c31d9 | 41 | #define GOOD_COPY_LEN 128 |
296f96fc | 42 | |
ab7db917 MD |
43 | /* Weight used for the RX packet size EWMA. The average packet size is used to |
44 | * determine the packet buffer size when refilling RX rings. As the entire RX | |
45 | * ring may be refilled at once, the weight is chosen so that the EWMA will be | |
46 | * insensitive to short-term, transient changes in packet size. | |
47 | */ | |
48 | #define RECEIVE_AVG_WEIGHT 64 | |
49 | ||
50 | /* Minimum alignment for mergeable packet buffers. */ | |
51 | #define MERGEABLE_BUFFER_ALIGN max(L1_CACHE_BYTES, 256) | |
52 | ||
66846048 | 53 | #define VIRTNET_DRIVER_VERSION "1.0.0" |
2a41f71d | 54 | |
3fa2a1df | 55 | struct virtnet_stats { |
83a27052 ED |
56 | struct u64_stats_sync tx_syncp; |
57 | struct u64_stats_sync rx_syncp; | |
3fa2a1df | 58 | u64 tx_bytes; |
59 | u64 tx_packets; | |
60 | ||
61 | u64 rx_bytes; | |
62 | u64 rx_packets; | |
63 | }; | |
64 | ||
e9d7417b JW |
65 | /* Internal representation of a send virtqueue */ |
66 | struct send_queue { | |
67 | /* Virtqueue associated with this send _queue */ | |
68 | struct virtqueue *vq; | |
69 | ||
70 | /* TX: fragments + linear part + virtio header */ | |
71 | struct scatterlist sg[MAX_SKB_FRAGS + 2]; | |
986a4f4d JW |
72 | |
73 | /* Name of the send queue: output.$index */ | |
74 | char name[40]; | |
e9d7417b JW |
75 | }; |
76 | ||
77 | /* Internal representation of a receive virtqueue */ | |
78 | struct receive_queue { | |
79 | /* Virtqueue associated with this receive_queue */ | |
80 | struct virtqueue *vq; | |
81 | ||
296f96fc RR |
82 | struct napi_struct napi; |
83 | ||
e9d7417b JW |
84 | /* Chain pages by the private ptr. */ |
85 | struct page *pages; | |
86 | ||
ab7db917 MD |
87 | /* Average packet length for mergeable receive buffers. */ |
88 | struct ewma mrg_avg_pkt_len; | |
89 | ||
fb51879d MD |
90 | /* Page frag for packet buffer allocation. */ |
91 | struct page_frag alloc_frag; | |
92 | ||
e9d7417b JW |
93 | /* RX: fragments + linear part + virtio header */ |
94 | struct scatterlist sg[MAX_SKB_FRAGS + 2]; | |
986a4f4d JW |
95 | |
96 | /* Name of this receive queue: input.$index */ | |
97 | char name[40]; | |
e9d7417b JW |
98 | }; |
99 | ||
100 | struct virtnet_info { | |
101 | struct virtio_device *vdev; | |
102 | struct virtqueue *cvq; | |
103 | struct net_device *dev; | |
986a4f4d JW |
104 | struct send_queue *sq; |
105 | struct receive_queue *rq; | |
e9d7417b JW |
106 | unsigned int status; |
107 | ||
986a4f4d JW |
108 | /* Max # of queue pairs supported by the device */ |
109 | u16 max_queue_pairs; | |
110 | ||
111 | /* # of queue pairs currently used by the driver */ | |
112 | u16 curr_queue_pairs; | |
113 | ||
97402b96 HX |
114 | /* I like... big packets and I cannot lie! */ |
115 | bool big_packets; | |
116 | ||
3f2c31d9 MM |
117 | /* Host will merge rx buffers for big packets (shake it! shake it!) */ |
118 | bool mergeable_rx_bufs; | |
119 | ||
986a4f4d JW |
120 | /* Has control virtqueue */ |
121 | bool has_cvq; | |
122 | ||
e7428e95 MT |
123 | /* Host can handle any s/g split between our header and packet data */ |
124 | bool any_header_sg; | |
125 | ||
586d17c5 JW |
126 | /* enable config space updates */ |
127 | bool config_enable; | |
128 | ||
3fa2a1df | 129 | /* Active statistics */ |
130 | struct virtnet_stats __percpu *stats; | |
131 | ||
3161e453 RR |
132 | /* Work struct for refilling if we run low on memory. */ |
133 | struct delayed_work refill; | |
134 | ||
586d17c5 JW |
135 | /* Work struct for config space updates */ |
136 | struct work_struct config_work; | |
137 | ||
138 | /* Lock for config space updates */ | |
139 | struct mutex config_lock; | |
986a4f4d JW |
140 | |
141 | /* Does the affinity hint is set for virtqueues? */ | |
142 | bool affinity_hint_set; | |
47be2479 | 143 | |
8de4b2f3 WG |
144 | /* CPU hot plug notifier */ |
145 | struct notifier_block nb; | |
296f96fc RR |
146 | }; |
147 | ||
b3f24698 RR |
148 | struct skb_vnet_hdr { |
149 | union { | |
150 | struct virtio_net_hdr hdr; | |
151 | struct virtio_net_hdr_mrg_rxbuf mhdr; | |
152 | }; | |
153 | }; | |
154 | ||
9ab86bbc SM |
155 | struct padded_vnet_hdr { |
156 | struct virtio_net_hdr hdr; | |
157 | /* | |
158 | * virtio_net_hdr should be in a separated sg buffer because of a | |
159 | * QEMU bug, and data sg buffer shares same page with this header sg. | |
160 | * This padding makes next sg 16 byte aligned after virtio_net_hdr. | |
161 | */ | |
162 | char padding[6]; | |
163 | }; | |
164 | ||
986a4f4d JW |
165 | /* Converting between virtqueue no. and kernel tx/rx queue no. |
166 | * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq | |
167 | */ | |
168 | static int vq2txq(struct virtqueue *vq) | |
169 | { | |
9d0ca6ed | 170 | return (vq->index - 1) / 2; |
986a4f4d JW |
171 | } |
172 | ||
173 | static int txq2vq(int txq) | |
174 | { | |
175 | return txq * 2 + 1; | |
176 | } | |
177 | ||
178 | static int vq2rxq(struct virtqueue *vq) | |
179 | { | |
9d0ca6ed | 180 | return vq->index / 2; |
986a4f4d JW |
181 | } |
182 | ||
183 | static int rxq2vq(int rxq) | |
184 | { | |
185 | return rxq * 2; | |
186 | } | |
187 | ||
b3f24698 | 188 | static inline struct skb_vnet_hdr *skb_vnet_hdr(struct sk_buff *skb) |
296f96fc | 189 | { |
b3f24698 | 190 | return (struct skb_vnet_hdr *)skb->cb; |
296f96fc RR |
191 | } |
192 | ||
9ab86bbc SM |
193 | /* |
194 | * private is used to chain pages for big packets, put the whole | |
195 | * most recent used list in the beginning for reuse | |
196 | */ | |
e9d7417b | 197 | static void give_pages(struct receive_queue *rq, struct page *page) |
0a888fd1 | 198 | { |
9ab86bbc | 199 | struct page *end; |
0a888fd1 | 200 | |
e9d7417b | 201 | /* Find end of list, sew whole thing into vi->rq.pages. */ |
9ab86bbc | 202 | for (end = page; end->private; end = (struct page *)end->private); |
e9d7417b JW |
203 | end->private = (unsigned long)rq->pages; |
204 | rq->pages = page; | |
0a888fd1 MM |
205 | } |
206 | ||
e9d7417b | 207 | static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask) |
fb6813f4 | 208 | { |
e9d7417b | 209 | struct page *p = rq->pages; |
fb6813f4 | 210 | |
9ab86bbc | 211 | if (p) { |
e9d7417b | 212 | rq->pages = (struct page *)p->private; |
9ab86bbc SM |
213 | /* clear private here, it is used to chain pages */ |
214 | p->private = 0; | |
215 | } else | |
fb6813f4 RR |
216 | p = alloc_page(gfp_mask); |
217 | return p; | |
218 | } | |
219 | ||
e9d7417b | 220 | static void skb_xmit_done(struct virtqueue *vq) |
296f96fc | 221 | { |
e9d7417b | 222 | struct virtnet_info *vi = vq->vdev->priv; |
296f96fc | 223 | |
2cb9c6ba | 224 | /* Suppress further interrupts. */ |
e9d7417b | 225 | virtqueue_disable_cb(vq); |
11a3a154 | 226 | |
363f1514 | 227 | /* We were probably waiting for more output buffers. */ |
986a4f4d | 228 | netif_wake_subqueue(vi->dev, vq2txq(vq)); |
296f96fc RR |
229 | } |
230 | ||
ab7db917 MD |
231 | static unsigned int mergeable_ctx_to_buf_truesize(unsigned long mrg_ctx) |
232 | { | |
233 | unsigned int truesize = mrg_ctx & (MERGEABLE_BUFFER_ALIGN - 1); | |
234 | return (truesize + 1) * MERGEABLE_BUFFER_ALIGN; | |
235 | } | |
236 | ||
237 | static void *mergeable_ctx_to_buf_address(unsigned long mrg_ctx) | |
238 | { | |
239 | return (void *)(mrg_ctx & -MERGEABLE_BUFFER_ALIGN); | |
240 | ||
241 | } | |
242 | ||
243 | static unsigned long mergeable_buf_to_ctx(void *buf, unsigned int truesize) | |
244 | { | |
245 | unsigned int size = truesize / MERGEABLE_BUFFER_ALIGN; | |
246 | return (unsigned long)buf | (size - 1); | |
247 | } | |
248 | ||
3464645a | 249 | /* Called from bottom half context */ |
e9d7417b | 250 | static struct sk_buff *page_to_skb(struct receive_queue *rq, |
2613af0e MD |
251 | struct page *page, unsigned int offset, |
252 | unsigned int len, unsigned int truesize) | |
9ab86bbc | 253 | { |
e9d7417b | 254 | struct virtnet_info *vi = rq->vq->vdev->priv; |
9ab86bbc SM |
255 | struct sk_buff *skb; |
256 | struct skb_vnet_hdr *hdr; | |
2613af0e | 257 | unsigned int copy, hdr_len, hdr_padded_len; |
9ab86bbc | 258 | char *p; |
fb6813f4 | 259 | |
2613af0e | 260 | p = page_address(page) + offset; |
3f2c31d9 | 261 | |
9ab86bbc SM |
262 | /* copy small packet so we can reuse these pages for small data */ |
263 | skb = netdev_alloc_skb_ip_align(vi->dev, GOOD_COPY_LEN); | |
264 | if (unlikely(!skb)) | |
265 | return NULL; | |
3f2c31d9 | 266 | |
9ab86bbc | 267 | hdr = skb_vnet_hdr(skb); |
3f2c31d9 | 268 | |
9ab86bbc SM |
269 | if (vi->mergeable_rx_bufs) { |
270 | hdr_len = sizeof hdr->mhdr; | |
2613af0e | 271 | hdr_padded_len = sizeof hdr->mhdr; |
9ab86bbc SM |
272 | } else { |
273 | hdr_len = sizeof hdr->hdr; | |
2613af0e | 274 | hdr_padded_len = sizeof(struct padded_vnet_hdr); |
9ab86bbc | 275 | } |
3f2c31d9 | 276 | |
9ab86bbc | 277 | memcpy(hdr, p, hdr_len); |
3f2c31d9 | 278 | |
9ab86bbc | 279 | len -= hdr_len; |
2613af0e MD |
280 | offset += hdr_padded_len; |
281 | p += hdr_padded_len; | |
3f2c31d9 | 282 | |
9ab86bbc SM |
283 | copy = len; |
284 | if (copy > skb_tailroom(skb)) | |
285 | copy = skb_tailroom(skb); | |
286 | memcpy(skb_put(skb, copy), p, copy); | |
3f2c31d9 | 287 | |
9ab86bbc SM |
288 | len -= copy; |
289 | offset += copy; | |
3f2c31d9 | 290 | |
2613af0e MD |
291 | if (vi->mergeable_rx_bufs) { |
292 | if (len) | |
293 | skb_add_rx_frag(skb, 0, page, offset, len, truesize); | |
294 | else | |
295 | put_page(page); | |
296 | return skb; | |
297 | } | |
298 | ||
e878d78b SL |
299 | /* |
300 | * Verify that we can indeed put this data into a skb. | |
301 | * This is here to handle cases when the device erroneously | |
302 | * tries to receive more than is possible. This is usually | |
303 | * the case of a broken device. | |
304 | */ | |
305 | if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) { | |
be443899 | 306 | net_dbg_ratelimited("%s: too much data\n", skb->dev->name); |
e878d78b SL |
307 | dev_kfree_skb(skb); |
308 | return NULL; | |
309 | } | |
2613af0e | 310 | BUG_ON(offset >= PAGE_SIZE); |
9ab86bbc | 311 | while (len) { |
2613af0e MD |
312 | unsigned int frag_size = min((unsigned)PAGE_SIZE - offset, len); |
313 | skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page, offset, | |
314 | frag_size, truesize); | |
315 | len -= frag_size; | |
9ab86bbc SM |
316 | page = (struct page *)page->private; |
317 | offset = 0; | |
318 | } | |
3f2c31d9 | 319 | |
9ab86bbc | 320 | if (page) |
e9d7417b | 321 | give_pages(rq, page); |
3f2c31d9 | 322 | |
9ab86bbc SM |
323 | return skb; |
324 | } | |
3f2c31d9 | 325 | |
f121159d MT |
326 | static struct sk_buff *receive_small(void *buf, unsigned int len) |
327 | { | |
328 | struct sk_buff * skb = buf; | |
329 | ||
330 | len -= sizeof(struct virtio_net_hdr); | |
331 | skb_trim(skb, len); | |
332 | ||
333 | return skb; | |
334 | } | |
335 | ||
336 | static struct sk_buff *receive_big(struct net_device *dev, | |
337 | struct receive_queue *rq, | |
338 | void *buf, | |
339 | unsigned int len) | |
340 | { | |
341 | struct page *page = buf; | |
342 | struct sk_buff *skb = page_to_skb(rq, page, 0, len, PAGE_SIZE); | |
343 | ||
344 | if (unlikely(!skb)) | |
345 | goto err; | |
346 | ||
347 | return skb; | |
348 | ||
349 | err: | |
350 | dev->stats.rx_dropped++; | |
351 | give_pages(rq, page); | |
352 | return NULL; | |
353 | } | |
354 | ||
8fc3b9e9 MT |
355 | static struct sk_buff *receive_mergeable(struct net_device *dev, |
356 | struct receive_queue *rq, | |
ab7db917 | 357 | unsigned long ctx, |
8fc3b9e9 | 358 | unsigned int len) |
9ab86bbc | 359 | { |
ab7db917 | 360 | void *buf = mergeable_ctx_to_buf_address(ctx); |
8fc3b9e9 MT |
361 | struct skb_vnet_hdr *hdr = buf; |
362 | int num_buf = hdr->mhdr.num_buffers; | |
363 | struct page *page = virt_to_head_page(buf); | |
364 | int offset = buf - page_address(page); | |
ab7db917 MD |
365 | unsigned int truesize = max(len, mergeable_ctx_to_buf_truesize(ctx)); |
366 | ||
fb51879d | 367 | struct sk_buff *head_skb = page_to_skb(rq, page, offset, len, truesize); |
2613af0e | 368 | struct sk_buff *curr_skb = head_skb; |
9ab86bbc | 369 | |
8fc3b9e9 MT |
370 | if (unlikely(!curr_skb)) |
371 | goto err_skb; | |
9ab86bbc | 372 | while (--num_buf) { |
8fc3b9e9 MT |
373 | int num_skb_frags; |
374 | ||
ab7db917 MD |
375 | ctx = (unsigned long)virtqueue_get_buf(rq->vq, &len); |
376 | if (unlikely(!ctx)) { | |
8fc3b9e9 MT |
377 | pr_debug("%s: rx error: %d buffers out of %d missing\n", |
378 | dev->name, num_buf, hdr->mhdr.num_buffers); | |
379 | dev->stats.rx_length_errors++; | |
380 | goto err_buf; | |
3f2c31d9 | 381 | } |
8fc3b9e9 | 382 | |
ab7db917 | 383 | buf = mergeable_ctx_to_buf_address(ctx); |
8fc3b9e9 | 384 | page = virt_to_head_page(buf); |
8fc3b9e9 MT |
385 | |
386 | num_skb_frags = skb_shinfo(curr_skb)->nr_frags; | |
2613af0e MD |
387 | if (unlikely(num_skb_frags == MAX_SKB_FRAGS)) { |
388 | struct sk_buff *nskb = alloc_skb(0, GFP_ATOMIC); | |
8fc3b9e9 MT |
389 | |
390 | if (unlikely(!nskb)) | |
391 | goto err_skb; | |
2613af0e MD |
392 | if (curr_skb == head_skb) |
393 | skb_shinfo(curr_skb)->frag_list = nskb; | |
394 | else | |
395 | curr_skb->next = nskb; | |
396 | curr_skb = nskb; | |
397 | head_skb->truesize += nskb->truesize; | |
398 | num_skb_frags = 0; | |
399 | } | |
ab7db917 | 400 | truesize = max(len, mergeable_ctx_to_buf_truesize(ctx)); |
2613af0e MD |
401 | if (curr_skb != head_skb) { |
402 | head_skb->data_len += len; | |
403 | head_skb->len += len; | |
fb51879d | 404 | head_skb->truesize += truesize; |
2613af0e | 405 | } |
8fc3b9e9 | 406 | offset = buf - page_address(page); |
ba275241 JW |
407 | if (skb_can_coalesce(curr_skb, num_skb_frags, page, offset)) { |
408 | put_page(page); | |
409 | skb_coalesce_rx_frag(curr_skb, num_skb_frags - 1, | |
fb51879d | 410 | len, truesize); |
ba275241 JW |
411 | } else { |
412 | skb_add_rx_frag(curr_skb, num_skb_frags, page, | |
fb51879d | 413 | offset, len, truesize); |
ba275241 | 414 | } |
8fc3b9e9 MT |
415 | } |
416 | ||
ab7db917 | 417 | ewma_add(&rq->mrg_avg_pkt_len, head_skb->len); |
8fc3b9e9 MT |
418 | return head_skb; |
419 | ||
420 | err_skb: | |
421 | put_page(page); | |
422 | while (--num_buf) { | |
ab7db917 MD |
423 | ctx = (unsigned long)virtqueue_get_buf(rq->vq, &len); |
424 | if (unlikely(!ctx)) { | |
8fc3b9e9 MT |
425 | pr_debug("%s: rx error: %d buffers missing\n", |
426 | dev->name, num_buf); | |
427 | dev->stats.rx_length_errors++; | |
428 | break; | |
429 | } | |
ab7db917 | 430 | page = virt_to_head_page(mergeable_ctx_to_buf_address(ctx)); |
8fc3b9e9 | 431 | put_page(page); |
9ab86bbc | 432 | } |
8fc3b9e9 MT |
433 | err_buf: |
434 | dev->stats.rx_dropped++; | |
435 | dev_kfree_skb(head_skb); | |
436 | return NULL; | |
9ab86bbc SM |
437 | } |
438 | ||
e9d7417b | 439 | static void receive_buf(struct receive_queue *rq, void *buf, unsigned int len) |
9ab86bbc | 440 | { |
e9d7417b JW |
441 | struct virtnet_info *vi = rq->vq->vdev->priv; |
442 | struct net_device *dev = vi->dev; | |
58472a76 | 443 | struct virtnet_stats *stats = this_cpu_ptr(vi->stats); |
9ab86bbc | 444 | struct sk_buff *skb; |
9ab86bbc | 445 | struct skb_vnet_hdr *hdr; |
3f2c31d9 | 446 | |
9ab86bbc SM |
447 | if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) { |
448 | pr_debug("%s: short packet %i\n", dev->name, len); | |
449 | dev->stats.rx_length_errors++; | |
ab7db917 MD |
450 | if (vi->mergeable_rx_bufs) { |
451 | unsigned long ctx = (unsigned long)buf; | |
452 | void *base = mergeable_ctx_to_buf_address(ctx); | |
453 | put_page(virt_to_head_page(base)); | |
454 | } else if (vi->big_packets) { | |
98bfd23c | 455 | give_pages(rq, buf); |
ab7db917 | 456 | } else { |
9ab86bbc | 457 | dev_kfree_skb(buf); |
ab7db917 | 458 | } |
9ab86bbc SM |
459 | return; |
460 | } | |
3f2c31d9 | 461 | |
f121159d | 462 | if (vi->mergeable_rx_bufs) |
ab7db917 | 463 | skb = receive_mergeable(dev, rq, (unsigned long)buf, len); |
f121159d MT |
464 | else if (vi->big_packets) |
465 | skb = receive_big(dev, rq, buf, len); | |
466 | else | |
467 | skb = receive_small(buf, len); | |
468 | ||
469 | if (unlikely(!skb)) | |
470 | return; | |
3f2c31d9 | 471 | |
9ab86bbc | 472 | hdr = skb_vnet_hdr(skb); |
3fa2a1df | 473 | |
83a27052 | 474 | u64_stats_update_begin(&stats->rx_syncp); |
3fa2a1df | 475 | stats->rx_bytes += skb->len; |
476 | stats->rx_packets++; | |
83a27052 | 477 | u64_stats_update_end(&stats->rx_syncp); |
296f96fc | 478 | |
b3f24698 | 479 | if (hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { |
296f96fc | 480 | pr_debug("Needs csum!\n"); |
b3f24698 RR |
481 | if (!skb_partial_csum_set(skb, |
482 | hdr->hdr.csum_start, | |
483 | hdr->hdr.csum_offset)) | |
296f96fc | 484 | goto frame_err; |
10a8d94a JW |
485 | } else if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) { |
486 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
296f96fc RR |
487 | } |
488 | ||
23cde76d MM |
489 | skb->protocol = eth_type_trans(skb, dev); |
490 | pr_debug("Receiving skb proto 0x%04x len %i type %i\n", | |
491 | ntohs(skb->protocol), skb->len, skb->pkt_type); | |
492 | ||
b3f24698 | 493 | if (hdr->hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) { |
296f96fc | 494 | pr_debug("GSO!\n"); |
b3f24698 | 495 | switch (hdr->hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { |
296f96fc | 496 | case VIRTIO_NET_HDR_GSO_TCPV4: |
c9af6db4 | 497 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; |
296f96fc | 498 | break; |
296f96fc | 499 | case VIRTIO_NET_HDR_GSO_UDP: |
c9af6db4 | 500 | skb_shinfo(skb)->gso_type = SKB_GSO_UDP; |
296f96fc RR |
501 | break; |
502 | case VIRTIO_NET_HDR_GSO_TCPV6: | |
c9af6db4 | 503 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6; |
296f96fc RR |
504 | break; |
505 | default: | |
be443899 AW |
506 | net_warn_ratelimited("%s: bad gso type %u.\n", |
507 | dev->name, hdr->hdr.gso_type); | |
296f96fc RR |
508 | goto frame_err; |
509 | } | |
510 | ||
b3f24698 | 511 | if (hdr->hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN) |
c9af6db4 | 512 | skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN; |
34a48579 | 513 | |
b3f24698 | 514 | skb_shinfo(skb)->gso_size = hdr->hdr.gso_size; |
296f96fc | 515 | if (skb_shinfo(skb)->gso_size == 0) { |
be443899 | 516 | net_warn_ratelimited("%s: zero gso size.\n", dev->name); |
296f96fc RR |
517 | goto frame_err; |
518 | } | |
519 | ||
520 | /* Header must be checked, and gso_segs computed. */ | |
521 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
522 | skb_shinfo(skb)->gso_segs = 0; | |
523 | } | |
524 | ||
91815639 JW |
525 | skb_mark_napi_id(skb, &rq->napi); |
526 | ||
296f96fc RR |
527 | netif_receive_skb(skb); |
528 | return; | |
529 | ||
530 | frame_err: | |
531 | dev->stats.rx_frame_errors++; | |
296f96fc RR |
532 | dev_kfree_skb(skb); |
533 | } | |
534 | ||
e9d7417b | 535 | static int add_recvbuf_small(struct receive_queue *rq, gfp_t gfp) |
296f96fc | 536 | { |
e9d7417b | 537 | struct virtnet_info *vi = rq->vq->vdev->priv; |
296f96fc | 538 | struct sk_buff *skb; |
9ab86bbc | 539 | struct skb_vnet_hdr *hdr; |
9ab86bbc | 540 | int err; |
3f2c31d9 | 541 | |
5061de36 | 542 | skb = __netdev_alloc_skb_ip_align(vi->dev, GOOD_PACKET_LEN, gfp); |
9ab86bbc SM |
543 | if (unlikely(!skb)) |
544 | return -ENOMEM; | |
296f96fc | 545 | |
5061de36 | 546 | skb_put(skb, GOOD_PACKET_LEN); |
3f2c31d9 | 547 | |
9ab86bbc | 548 | hdr = skb_vnet_hdr(skb); |
a5835440 | 549 | sg_init_table(rq->sg, MAX_SKB_FRAGS + 2); |
e9d7417b | 550 | sg_set_buf(rq->sg, &hdr->hdr, sizeof hdr->hdr); |
e9d7417b | 551 | skb_to_sgvec(skb, rq->sg + 1, 0, skb->len); |
97402b96 | 552 | |
9dc7b9e4 | 553 | err = virtqueue_add_inbuf(rq->vq, rq->sg, 2, skb, gfp); |
9ab86bbc SM |
554 | if (err < 0) |
555 | dev_kfree_skb(skb); | |
97402b96 | 556 | |
9ab86bbc SM |
557 | return err; |
558 | } | |
97402b96 | 559 | |
e9d7417b | 560 | static int add_recvbuf_big(struct receive_queue *rq, gfp_t gfp) |
9ab86bbc | 561 | { |
9ab86bbc SM |
562 | struct page *first, *list = NULL; |
563 | char *p; | |
564 | int i, err, offset; | |
565 | ||
a5835440 RR |
566 | sg_init_table(rq->sg, MAX_SKB_FRAGS + 2); |
567 | ||
e9d7417b | 568 | /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */ |
9ab86bbc | 569 | for (i = MAX_SKB_FRAGS + 1; i > 1; --i) { |
e9d7417b | 570 | first = get_a_page(rq, gfp); |
9ab86bbc SM |
571 | if (!first) { |
572 | if (list) | |
e9d7417b | 573 | give_pages(rq, list); |
9ab86bbc | 574 | return -ENOMEM; |
97402b96 | 575 | } |
e9d7417b | 576 | sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE); |
97402b96 | 577 | |
9ab86bbc SM |
578 | /* chain new page in list head to match sg */ |
579 | first->private = (unsigned long)list; | |
580 | list = first; | |
581 | } | |
296f96fc | 582 | |
e9d7417b | 583 | first = get_a_page(rq, gfp); |
9ab86bbc | 584 | if (!first) { |
e9d7417b | 585 | give_pages(rq, list); |
9ab86bbc SM |
586 | return -ENOMEM; |
587 | } | |
588 | p = page_address(first); | |
589 | ||
e9d7417b JW |
590 | /* rq->sg[0], rq->sg[1] share the same page */ |
591 | /* a separated rq->sg[0] for virtio_net_hdr only due to QEMU bug */ | |
592 | sg_set_buf(&rq->sg[0], p, sizeof(struct virtio_net_hdr)); | |
9ab86bbc | 593 | |
e9d7417b | 594 | /* rq->sg[1] for data packet, from offset */ |
9ab86bbc | 595 | offset = sizeof(struct padded_vnet_hdr); |
e9d7417b | 596 | sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset); |
9ab86bbc SM |
597 | |
598 | /* chain first in list head */ | |
599 | first->private = (unsigned long)list; | |
9dc7b9e4 RR |
600 | err = virtqueue_add_inbuf(rq->vq, rq->sg, MAX_SKB_FRAGS + 2, |
601 | first, gfp); | |
9ab86bbc | 602 | if (err < 0) |
e9d7417b | 603 | give_pages(rq, first); |
9ab86bbc SM |
604 | |
605 | return err; | |
296f96fc RR |
606 | } |
607 | ||
fbf28d78 | 608 | static unsigned int get_mergeable_buf_len(struct ewma *avg_pkt_len) |
3f2c31d9 | 609 | { |
ab7db917 | 610 | const size_t hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf); |
fbf28d78 MD |
611 | unsigned int len; |
612 | ||
613 | len = hdr_len + clamp_t(unsigned int, ewma_read(avg_pkt_len), | |
614 | GOOD_PACKET_LEN, PAGE_SIZE - hdr_len); | |
615 | return ALIGN(len, MERGEABLE_BUFFER_ALIGN); | |
616 | } | |
617 | ||
618 | static int add_recvbuf_mergeable(struct receive_queue *rq, gfp_t gfp) | |
619 | { | |
fb51879d MD |
620 | struct page_frag *alloc_frag = &rq->alloc_frag; |
621 | char *buf; | |
ab7db917 | 622 | unsigned long ctx; |
3f2c31d9 | 623 | int err; |
fb51879d | 624 | unsigned int len, hole; |
3f2c31d9 | 625 | |
fbf28d78 | 626 | len = get_mergeable_buf_len(&rq->mrg_avg_pkt_len); |
ab7db917 | 627 | if (unlikely(!skb_page_frag_refill(len, alloc_frag, gfp))) |
9ab86bbc | 628 | return -ENOMEM; |
ab7db917 | 629 | |
fb51879d | 630 | buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset; |
ab7db917 | 631 | ctx = mergeable_buf_to_ctx(buf, len); |
fb51879d | 632 | get_page(alloc_frag->page); |
fb51879d MD |
633 | alloc_frag->offset += len; |
634 | hole = alloc_frag->size - alloc_frag->offset; | |
ab7db917 MD |
635 | if (hole < len) { |
636 | /* To avoid internal fragmentation, if there is very likely not | |
637 | * enough space for another buffer, add the remaining space to | |
638 | * the current buffer. This extra space is not included in | |
639 | * the truesize stored in ctx. | |
640 | */ | |
fb51879d MD |
641 | len += hole; |
642 | alloc_frag->offset += hole; | |
643 | } | |
3f2c31d9 | 644 | |
fb51879d | 645 | sg_init_one(rq->sg, buf, len); |
ab7db917 | 646 | err = virtqueue_add_inbuf(rq->vq, rq->sg, 1, (void *)ctx, gfp); |
9ab86bbc | 647 | if (err < 0) |
2613af0e | 648 | put_page(virt_to_head_page(buf)); |
3f2c31d9 | 649 | |
9ab86bbc SM |
650 | return err; |
651 | } | |
3f2c31d9 | 652 | |
b2baed69 RR |
653 | /* |
654 | * Returns false if we couldn't fill entirely (OOM). | |
655 | * | |
656 | * Normally run in the receive path, but can also be run from ndo_open | |
657 | * before we're receiving packets, or from refill_work which is | |
658 | * careful to disable receiving (using napi_disable). | |
659 | */ | |
e9d7417b | 660 | static bool try_fill_recv(struct receive_queue *rq, gfp_t gfp) |
9ab86bbc | 661 | { |
e9d7417b | 662 | struct virtnet_info *vi = rq->vq->vdev->priv; |
9ab86bbc | 663 | int err; |
1788f495 | 664 | bool oom; |
3f2c31d9 | 665 | |
fb51879d | 666 | gfp |= __GFP_COLD; |
9ab86bbc SM |
667 | do { |
668 | if (vi->mergeable_rx_bufs) | |
e9d7417b | 669 | err = add_recvbuf_mergeable(rq, gfp); |
9ab86bbc | 670 | else if (vi->big_packets) |
e9d7417b | 671 | err = add_recvbuf_big(rq, gfp); |
9ab86bbc | 672 | else |
e9d7417b | 673 | err = add_recvbuf_small(rq, gfp); |
3f2c31d9 | 674 | |
1788f495 | 675 | oom = err == -ENOMEM; |
9ed4cb07 | 676 | if (err) |
3f2c31d9 | 677 | break; |
b7dfde95 | 678 | } while (rq->vq->num_free); |
681daee2 | 679 | virtqueue_kick(rq->vq); |
3161e453 | 680 | return !oom; |
3f2c31d9 MM |
681 | } |
682 | ||
18445c4d | 683 | static void skb_recv_done(struct virtqueue *rvq) |
296f96fc RR |
684 | { |
685 | struct virtnet_info *vi = rvq->vdev->priv; | |
986a4f4d | 686 | struct receive_queue *rq = &vi->rq[vq2rxq(rvq)]; |
e9d7417b | 687 | |
18445c4d | 688 | /* Schedule NAPI, Suppress further interrupts if successful. */ |
e9d7417b | 689 | if (napi_schedule_prep(&rq->napi)) { |
1915a712 | 690 | virtqueue_disable_cb(rvq); |
e9d7417b | 691 | __napi_schedule(&rq->napi); |
18445c4d | 692 | } |
296f96fc RR |
693 | } |
694 | ||
e9d7417b | 695 | static void virtnet_napi_enable(struct receive_queue *rq) |
3e9d08ec | 696 | { |
e9d7417b | 697 | napi_enable(&rq->napi); |
3e9d08ec BR |
698 | |
699 | /* If all buffers were filled by other side before we napi_enabled, we | |
700 | * won't get another interrupt, so process any outstanding packets | |
701 | * now. virtnet_poll wants re-enable the queue, so we disable here. | |
702 | * We synchronize against interrupts via NAPI_STATE_SCHED */ | |
e9d7417b JW |
703 | if (napi_schedule_prep(&rq->napi)) { |
704 | virtqueue_disable_cb(rq->vq); | |
ec13ee80 | 705 | local_bh_disable(); |
e9d7417b | 706 | __napi_schedule(&rq->napi); |
ec13ee80 | 707 | local_bh_enable(); |
3e9d08ec BR |
708 | } |
709 | } | |
710 | ||
3161e453 RR |
711 | static void refill_work(struct work_struct *work) |
712 | { | |
e9d7417b JW |
713 | struct virtnet_info *vi = |
714 | container_of(work, struct virtnet_info, refill.work); | |
3161e453 | 715 | bool still_empty; |
986a4f4d JW |
716 | int i; |
717 | ||
55257d72 | 718 | for (i = 0; i < vi->curr_queue_pairs; i++) { |
986a4f4d | 719 | struct receive_queue *rq = &vi->rq[i]; |
3161e453 | 720 | |
986a4f4d JW |
721 | napi_disable(&rq->napi); |
722 | still_empty = !try_fill_recv(rq, GFP_KERNEL); | |
723 | virtnet_napi_enable(rq); | |
3161e453 | 724 | |
986a4f4d JW |
725 | /* In theory, this can happen: if we don't get any buffers in |
726 | * we will *never* try to fill again. | |
727 | */ | |
728 | if (still_empty) | |
729 | schedule_delayed_work(&vi->refill, HZ/2); | |
730 | } | |
3161e453 RR |
731 | } |
732 | ||
2ffa7598 | 733 | static int virtnet_receive(struct receive_queue *rq, int budget) |
296f96fc | 734 | { |
e9d7417b | 735 | struct virtnet_info *vi = rq->vq->vdev->priv; |
2ffa7598 | 736 | unsigned int len, received = 0; |
9ab86bbc | 737 | void *buf; |
296f96fc | 738 | |
296f96fc | 739 | while (received < budget && |
e9d7417b JW |
740 | (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) { |
741 | receive_buf(rq, buf, len); | |
296f96fc RR |
742 | received++; |
743 | } | |
744 | ||
be121f46 | 745 | if (rq->vq->num_free > virtqueue_get_vring_size(rq->vq) / 2) { |
e9d7417b | 746 | if (!try_fill_recv(rq, GFP_ATOMIC)) |
3b07e9ca | 747 | schedule_delayed_work(&vi->refill, 0); |
3161e453 | 748 | } |
296f96fc | 749 | |
2ffa7598 JW |
750 | return received; |
751 | } | |
752 | ||
753 | static int virtnet_poll(struct napi_struct *napi, int budget) | |
754 | { | |
755 | struct receive_queue *rq = | |
756 | container_of(napi, struct receive_queue, napi); | |
757 | unsigned int r, received = 0; | |
758 | ||
759 | again: | |
760 | received += virtnet_receive(rq, budget - received); | |
761 | ||
8329d98e RR |
762 | /* Out of packets? */ |
763 | if (received < budget) { | |
cbdadbbf | 764 | r = virtqueue_enable_cb_prepare(rq->vq); |
288379f0 | 765 | napi_complete(napi); |
cbdadbbf | 766 | if (unlikely(virtqueue_poll(rq->vq, r)) && |
8e95a202 | 767 | napi_schedule_prep(napi)) { |
e9d7417b | 768 | virtqueue_disable_cb(rq->vq); |
288379f0 | 769 | __napi_schedule(napi); |
296f96fc | 770 | goto again; |
4265f161 | 771 | } |
296f96fc RR |
772 | } |
773 | ||
774 | return received; | |
775 | } | |
776 | ||
91815639 JW |
777 | #ifdef CONFIG_NET_RX_BUSY_POLL |
778 | /* must be called with local_bh_disable()d */ | |
779 | static int virtnet_busy_poll(struct napi_struct *napi) | |
780 | { | |
781 | struct receive_queue *rq = | |
782 | container_of(napi, struct receive_queue, napi); | |
783 | struct virtnet_info *vi = rq->vq->vdev->priv; | |
784 | int r, received = 0, budget = 4; | |
785 | ||
786 | if (!(vi->status & VIRTIO_NET_S_LINK_UP)) | |
787 | return LL_FLUSH_FAILED; | |
788 | ||
789 | if (!napi_schedule_prep(napi)) | |
790 | return LL_FLUSH_BUSY; | |
791 | ||
792 | virtqueue_disable_cb(rq->vq); | |
793 | ||
794 | again: | |
795 | received += virtnet_receive(rq, budget); | |
796 | ||
797 | r = virtqueue_enable_cb_prepare(rq->vq); | |
798 | clear_bit(NAPI_STATE_SCHED, &napi->state); | |
799 | if (unlikely(virtqueue_poll(rq->vq, r)) && | |
800 | napi_schedule_prep(napi)) { | |
801 | virtqueue_disable_cb(rq->vq); | |
802 | if (received < budget) { | |
803 | budget -= received; | |
804 | goto again; | |
805 | } else { | |
806 | __napi_schedule(napi); | |
807 | } | |
808 | } | |
809 | ||
810 | return received; | |
811 | } | |
812 | #endif /* CONFIG_NET_RX_BUSY_POLL */ | |
813 | ||
986a4f4d JW |
814 | static int virtnet_open(struct net_device *dev) |
815 | { | |
816 | struct virtnet_info *vi = netdev_priv(dev); | |
817 | int i; | |
818 | ||
e4166625 JW |
819 | for (i = 0; i < vi->max_queue_pairs; i++) { |
820 | if (i < vi->curr_queue_pairs) | |
821 | /* Make sure we have some buffers: if oom use wq. */ | |
822 | if (!try_fill_recv(&vi->rq[i], GFP_KERNEL)) | |
823 | schedule_delayed_work(&vi->refill, 0); | |
986a4f4d JW |
824 | virtnet_napi_enable(&vi->rq[i]); |
825 | } | |
826 | ||
827 | return 0; | |
828 | } | |
829 | ||
b7dfde95 | 830 | static void free_old_xmit_skbs(struct send_queue *sq) |
296f96fc RR |
831 | { |
832 | struct sk_buff *skb; | |
6ee57bcc | 833 | unsigned int len; |
e9d7417b | 834 | struct virtnet_info *vi = sq->vq->vdev->priv; |
58472a76 | 835 | struct virtnet_stats *stats = this_cpu_ptr(vi->stats); |
296f96fc | 836 | |
e9d7417b | 837 | while ((skb = virtqueue_get_buf(sq->vq, &len)) != NULL) { |
296f96fc | 838 | pr_debug("Sent skb %p\n", skb); |
3fa2a1df | 839 | |
83a27052 | 840 | u64_stats_update_begin(&stats->tx_syncp); |
3fa2a1df | 841 | stats->tx_bytes += skb->len; |
842 | stats->tx_packets++; | |
83a27052 | 843 | u64_stats_update_end(&stats->tx_syncp); |
3fa2a1df | 844 | |
ed79bab8 | 845 | dev_kfree_skb_any(skb); |
296f96fc RR |
846 | } |
847 | } | |
848 | ||
e9d7417b | 849 | static int xmit_skb(struct send_queue *sq, struct sk_buff *skb) |
296f96fc | 850 | { |
e7428e95 | 851 | struct skb_vnet_hdr *hdr; |
296f96fc | 852 | const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest; |
e9d7417b | 853 | struct virtnet_info *vi = sq->vq->vdev->priv; |
7bedc7dc | 854 | unsigned num_sg; |
e7428e95 MT |
855 | unsigned hdr_len; |
856 | bool can_push; | |
296f96fc | 857 | |
e174961c | 858 | pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest); |
e7428e95 MT |
859 | if (vi->mergeable_rx_bufs) |
860 | hdr_len = sizeof hdr->mhdr; | |
861 | else | |
862 | hdr_len = sizeof hdr->hdr; | |
863 | ||
864 | can_push = vi->any_header_sg && | |
865 | !((unsigned long)skb->data & (__alignof__(*hdr) - 1)) && | |
866 | !skb_header_cloned(skb) && skb_headroom(skb) >= hdr_len; | |
867 | /* Even if we can, don't push here yet as this would skew | |
868 | * csum_start offset below. */ | |
869 | if (can_push) | |
870 | hdr = (struct skb_vnet_hdr *)(skb->data - hdr_len); | |
871 | else | |
872 | hdr = skb_vnet_hdr(skb); | |
296f96fc | 873 | |
296f96fc | 874 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
b3f24698 | 875 | hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; |
55508d60 | 876 | hdr->hdr.csum_start = skb_checksum_start_offset(skb); |
b3f24698 | 877 | hdr->hdr.csum_offset = skb->csum_offset; |
296f96fc | 878 | } else { |
b3f24698 RR |
879 | hdr->hdr.flags = 0; |
880 | hdr->hdr.csum_offset = hdr->hdr.csum_start = 0; | |
296f96fc RR |
881 | } |
882 | ||
883 | if (skb_is_gso(skb)) { | |
b3f24698 RR |
884 | hdr->hdr.hdr_len = skb_headlen(skb); |
885 | hdr->hdr.gso_size = skb_shinfo(skb)->gso_size; | |
34a48579 | 886 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) |
b3f24698 | 887 | hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4; |
296f96fc | 888 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) |
b3f24698 | 889 | hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6; |
296f96fc | 890 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP) |
b3f24698 | 891 | hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP; |
296f96fc RR |
892 | else |
893 | BUG(); | |
34a48579 | 894 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN) |
b3f24698 | 895 | hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN; |
296f96fc | 896 | } else { |
b3f24698 RR |
897 | hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE; |
898 | hdr->hdr.gso_size = hdr->hdr.hdr_len = 0; | |
296f96fc RR |
899 | } |
900 | ||
3f2c31d9 | 901 | if (vi->mergeable_rx_bufs) |
e7428e95 | 902 | hdr->mhdr.num_buffers = 0; |
3f2c31d9 | 903 | |
a5835440 | 904 | sg_init_table(sq->sg, MAX_SKB_FRAGS + 2); |
e7428e95 MT |
905 | if (can_push) { |
906 | __skb_push(skb, hdr_len); | |
907 | num_sg = skb_to_sgvec(skb, sq->sg, 0, skb->len); | |
908 | /* Pull header back to avoid skew in tx bytes calculations. */ | |
909 | __skb_pull(skb, hdr_len); | |
910 | } else { | |
911 | sg_set_buf(sq->sg, hdr, hdr_len); | |
912 | num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len) + 1; | |
913 | } | |
9dc7b9e4 | 914 | return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC); |
11a3a154 RR |
915 | } |
916 | ||
424efe9c | 917 | static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev) |
99ffc696 RR |
918 | { |
919 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d JW |
920 | int qnum = skb_get_queue_mapping(skb); |
921 | struct send_queue *sq = &vi->sq[qnum]; | |
9ed4cb07 | 922 | int err; |
4b7fd2e6 MT |
923 | struct netdev_queue *txq = netdev_get_tx_queue(dev, qnum); |
924 | bool kick = !skb->xmit_more; | |
2cb9c6ba | 925 | |
2cb9c6ba | 926 | /* Free up any pending old buffers before queueing new ones. */ |
e9d7417b | 927 | free_old_xmit_skbs(sq); |
99ffc696 | 928 | |
03f191ba | 929 | /* Try to transmit */ |
b7dfde95 | 930 | err = xmit_skb(sq, skb); |
48925e37 | 931 | |
9ed4cb07 | 932 | /* This should not happen! */ |
681daee2 | 933 | if (unlikely(err)) { |
9ed4cb07 RR |
934 | dev->stats.tx_fifo_errors++; |
935 | if (net_ratelimit()) | |
936 | dev_warn(&dev->dev, | |
b7dfde95 | 937 | "Unexpected TXQ (%d) queue failure: %d\n", qnum, err); |
58eba97d | 938 | dev->stats.tx_dropped++; |
85e94525 | 939 | dev_kfree_skb_any(skb); |
58eba97d | 940 | return NETDEV_TX_OK; |
296f96fc | 941 | } |
03f191ba | 942 | |
48925e37 RR |
943 | /* Don't wait up for transmitted skbs to be freed. */ |
944 | skb_orphan(skb); | |
945 | nf_reset(skb); | |
946 | ||
947 | /* Apparently nice girls don't return TX_BUSY; stop the queue | |
948 | * before it gets out of hand. Naturally, this wastes entries. */ | |
b7dfde95 | 949 | if (sq->vq->num_free < 2+MAX_SKB_FRAGS) { |
986a4f4d | 950 | netif_stop_subqueue(dev, qnum); |
e9d7417b | 951 | if (unlikely(!virtqueue_enable_cb_delayed(sq->vq))) { |
48925e37 | 952 | /* More just got used, free them then recheck. */ |
b7dfde95 LT |
953 | free_old_xmit_skbs(sq); |
954 | if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) { | |
986a4f4d | 955 | netif_start_subqueue(dev, qnum); |
e9d7417b | 956 | virtqueue_disable_cb(sq->vq); |
48925e37 RR |
957 | } |
958 | } | |
99ffc696 | 959 | } |
48925e37 | 960 | |
4b7fd2e6 | 961 | if (kick || netif_xmit_stopped(txq)) |
0b725a2c | 962 | virtqueue_kick(sq->vq); |
296f96fc | 963 | |
0b725a2c | 964 | return NETDEV_TX_OK; |
c223a078 DM |
965 | } |
966 | ||
40cbfc37 AK |
967 | /* |
968 | * Send command via the control virtqueue and check status. Commands | |
969 | * supported by the hypervisor, as indicated by feature bits, should | |
788a8b6d | 970 | * never fail unless improperly formatted. |
40cbfc37 AK |
971 | */ |
972 | static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd, | |
d24bae32 | 973 | struct scatterlist *out) |
40cbfc37 | 974 | { |
f7bc9594 | 975 | struct scatterlist *sgs[4], hdr, stat; |
40cbfc37 AK |
976 | struct virtio_net_ctrl_hdr ctrl; |
977 | virtio_net_ctrl_ack status = ~0; | |
d24bae32 | 978 | unsigned out_num = 0, tmp; |
40cbfc37 AK |
979 | |
980 | /* Caller should know better */ | |
f7bc9594 | 981 | BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)); |
40cbfc37 AK |
982 | |
983 | ctrl.class = class; | |
984 | ctrl.cmd = cmd; | |
f7bc9594 RR |
985 | /* Add header */ |
986 | sg_init_one(&hdr, &ctrl, sizeof(ctrl)); | |
987 | sgs[out_num++] = &hdr; | |
40cbfc37 | 988 | |
f7bc9594 RR |
989 | if (out) |
990 | sgs[out_num++] = out; | |
40cbfc37 | 991 | |
f7bc9594 RR |
992 | /* Add return status. */ |
993 | sg_init_one(&stat, &status, sizeof(status)); | |
d24bae32 | 994 | sgs[out_num] = &stat; |
40cbfc37 | 995 | |
d24bae32 | 996 | BUG_ON(out_num + 1 > ARRAY_SIZE(sgs)); |
a7c58146 | 997 | virtqueue_add_sgs(vi->cvq, sgs, out_num, 1, vi, GFP_ATOMIC); |
40cbfc37 | 998 | |
67975901 HG |
999 | if (unlikely(!virtqueue_kick(vi->cvq))) |
1000 | return status == VIRTIO_NET_OK; | |
40cbfc37 AK |
1001 | |
1002 | /* Spin for a response, the kick causes an ioport write, trapping | |
1003 | * into the hypervisor, so the request should be handled immediately. | |
1004 | */ | |
047b9b94 HG |
1005 | while (!virtqueue_get_buf(vi->cvq, &tmp) && |
1006 | !virtqueue_is_broken(vi->cvq)) | |
40cbfc37 AK |
1007 | cpu_relax(); |
1008 | ||
1009 | return status == VIRTIO_NET_OK; | |
1010 | } | |
1011 | ||
9c46f6d4 AW |
1012 | static int virtnet_set_mac_address(struct net_device *dev, void *p) |
1013 | { | |
1014 | struct virtnet_info *vi = netdev_priv(dev); | |
1015 | struct virtio_device *vdev = vi->vdev; | |
f2f2c8b4 | 1016 | int ret; |
7e58d5ae AK |
1017 | struct sockaddr *addr = p; |
1018 | struct scatterlist sg; | |
9c46f6d4 | 1019 | |
7e58d5ae | 1020 | ret = eth_prepare_mac_addr_change(dev, p); |
f2f2c8b4 JP |
1021 | if (ret) |
1022 | return ret; | |
9c46f6d4 | 1023 | |
7e58d5ae AK |
1024 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) { |
1025 | sg_init_one(&sg, addr->sa_data, dev->addr_len); | |
1026 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC, | |
d24bae32 | 1027 | VIRTIO_NET_CTRL_MAC_ADDR_SET, &sg)) { |
7e58d5ae AK |
1028 | dev_warn(&vdev->dev, |
1029 | "Failed to set mac address by vq command.\n"); | |
1030 | return -EINVAL; | |
1031 | } | |
1032 | } else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) { | |
855e0c52 RR |
1033 | unsigned int i; |
1034 | ||
1035 | /* Naturally, this has an atomicity problem. */ | |
1036 | for (i = 0; i < dev->addr_len; i++) | |
1037 | virtio_cwrite8(vdev, | |
1038 | offsetof(struct virtio_net_config, mac) + | |
1039 | i, addr->sa_data[i]); | |
7e58d5ae AK |
1040 | } |
1041 | ||
1042 | eth_commit_mac_addr_change(dev, p); | |
9c46f6d4 AW |
1043 | |
1044 | return 0; | |
1045 | } | |
1046 | ||
3fa2a1df | 1047 | static struct rtnl_link_stats64 *virtnet_stats(struct net_device *dev, |
1048 | struct rtnl_link_stats64 *tot) | |
1049 | { | |
1050 | struct virtnet_info *vi = netdev_priv(dev); | |
1051 | int cpu; | |
1052 | unsigned int start; | |
1053 | ||
1054 | for_each_possible_cpu(cpu) { | |
58472a76 | 1055 | struct virtnet_stats *stats = per_cpu_ptr(vi->stats, cpu); |
3fa2a1df | 1056 | u64 tpackets, tbytes, rpackets, rbytes; |
1057 | ||
1058 | do { | |
57a7744e | 1059 | start = u64_stats_fetch_begin_irq(&stats->tx_syncp); |
3fa2a1df | 1060 | tpackets = stats->tx_packets; |
1061 | tbytes = stats->tx_bytes; | |
57a7744e | 1062 | } while (u64_stats_fetch_retry_irq(&stats->tx_syncp, start)); |
83a27052 ED |
1063 | |
1064 | do { | |
57a7744e | 1065 | start = u64_stats_fetch_begin_irq(&stats->rx_syncp); |
3fa2a1df | 1066 | rpackets = stats->rx_packets; |
1067 | rbytes = stats->rx_bytes; | |
57a7744e | 1068 | } while (u64_stats_fetch_retry_irq(&stats->rx_syncp, start)); |
3fa2a1df | 1069 | |
1070 | tot->rx_packets += rpackets; | |
1071 | tot->tx_packets += tpackets; | |
1072 | tot->rx_bytes += rbytes; | |
1073 | tot->tx_bytes += tbytes; | |
1074 | } | |
1075 | ||
1076 | tot->tx_dropped = dev->stats.tx_dropped; | |
021ac8d3 | 1077 | tot->tx_fifo_errors = dev->stats.tx_fifo_errors; |
3fa2a1df | 1078 | tot->rx_dropped = dev->stats.rx_dropped; |
1079 | tot->rx_length_errors = dev->stats.rx_length_errors; | |
1080 | tot->rx_frame_errors = dev->stats.rx_frame_errors; | |
1081 | ||
1082 | return tot; | |
1083 | } | |
1084 | ||
da74e89d AS |
1085 | #ifdef CONFIG_NET_POLL_CONTROLLER |
1086 | static void virtnet_netpoll(struct net_device *dev) | |
1087 | { | |
1088 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d | 1089 | int i; |
da74e89d | 1090 | |
986a4f4d JW |
1091 | for (i = 0; i < vi->curr_queue_pairs; i++) |
1092 | napi_schedule(&vi->rq[i].napi); | |
da74e89d AS |
1093 | } |
1094 | #endif | |
1095 | ||
586d17c5 JW |
1096 | static void virtnet_ack_link_announce(struct virtnet_info *vi) |
1097 | { | |
1098 | rtnl_lock(); | |
1099 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE, | |
d24bae32 | 1100 | VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL)) |
586d17c5 JW |
1101 | dev_warn(&vi->dev->dev, "Failed to ack link announce.\n"); |
1102 | rtnl_unlock(); | |
1103 | } | |
1104 | ||
986a4f4d JW |
1105 | static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs) |
1106 | { | |
1107 | struct scatterlist sg; | |
1108 | struct virtio_net_ctrl_mq s; | |
1109 | struct net_device *dev = vi->dev; | |
1110 | ||
1111 | if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ)) | |
1112 | return 0; | |
1113 | ||
1114 | s.virtqueue_pairs = queue_pairs; | |
1115 | sg_init_one(&sg, &s, sizeof(s)); | |
1116 | ||
1117 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ, | |
d24bae32 | 1118 | VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg)) { |
986a4f4d JW |
1119 | dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n", |
1120 | queue_pairs); | |
1121 | return -EINVAL; | |
55257d72 | 1122 | } else { |
986a4f4d | 1123 | vi->curr_queue_pairs = queue_pairs; |
35ed159b JW |
1124 | /* virtnet_open() will refill when device is going to up. */ |
1125 | if (dev->flags & IFF_UP) | |
1126 | schedule_delayed_work(&vi->refill, 0); | |
55257d72 | 1127 | } |
986a4f4d JW |
1128 | |
1129 | return 0; | |
1130 | } | |
1131 | ||
296f96fc RR |
1132 | static int virtnet_close(struct net_device *dev) |
1133 | { | |
1134 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d | 1135 | int i; |
296f96fc | 1136 | |
b2baed69 RR |
1137 | /* Make sure refill_work doesn't re-enable napi! */ |
1138 | cancel_delayed_work_sync(&vi->refill); | |
986a4f4d JW |
1139 | |
1140 | for (i = 0; i < vi->max_queue_pairs; i++) | |
1141 | napi_disable(&vi->rq[i].napi); | |
296f96fc | 1142 | |
296f96fc RR |
1143 | return 0; |
1144 | } | |
1145 | ||
2af7698e AW |
1146 | static void virtnet_set_rx_mode(struct net_device *dev) |
1147 | { | |
1148 | struct virtnet_info *vi = netdev_priv(dev); | |
f565a7c2 | 1149 | struct scatterlist sg[2]; |
2af7698e | 1150 | u8 promisc, allmulti; |
f565a7c2 | 1151 | struct virtio_net_ctrl_mac *mac_data; |
ccffad25 | 1152 | struct netdev_hw_addr *ha; |
32e7bfc4 | 1153 | int uc_count; |
4cd24eaf | 1154 | int mc_count; |
f565a7c2 AW |
1155 | void *buf; |
1156 | int i; | |
2af7698e | 1157 | |
788a8b6d | 1158 | /* We can't dynamically set ndo_set_rx_mode, so return gracefully */ |
2af7698e AW |
1159 | if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX)) |
1160 | return; | |
1161 | ||
f565a7c2 AW |
1162 | promisc = ((dev->flags & IFF_PROMISC) != 0); |
1163 | allmulti = ((dev->flags & IFF_ALLMULTI) != 0); | |
2af7698e | 1164 | |
23e258e1 | 1165 | sg_init_one(sg, &promisc, sizeof(promisc)); |
2af7698e AW |
1166 | |
1167 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, | |
d24bae32 | 1168 | VIRTIO_NET_CTRL_RX_PROMISC, sg)) |
2af7698e AW |
1169 | dev_warn(&dev->dev, "Failed to %sable promisc mode.\n", |
1170 | promisc ? "en" : "dis"); | |
1171 | ||
23e258e1 | 1172 | sg_init_one(sg, &allmulti, sizeof(allmulti)); |
2af7698e AW |
1173 | |
1174 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, | |
d24bae32 | 1175 | VIRTIO_NET_CTRL_RX_ALLMULTI, sg)) |
2af7698e AW |
1176 | dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n", |
1177 | allmulti ? "en" : "dis"); | |
f565a7c2 | 1178 | |
32e7bfc4 | 1179 | uc_count = netdev_uc_count(dev); |
4cd24eaf | 1180 | mc_count = netdev_mc_count(dev); |
f565a7c2 | 1181 | /* MAC filter - use one buffer for both lists */ |
4cd24eaf JP |
1182 | buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) + |
1183 | (2 * sizeof(mac_data->entries)), GFP_ATOMIC); | |
1184 | mac_data = buf; | |
e68ed8f0 | 1185 | if (!buf) |
f565a7c2 | 1186 | return; |
f565a7c2 | 1187 | |
23e258e1 AW |
1188 | sg_init_table(sg, 2); |
1189 | ||
f565a7c2 | 1190 | /* Store the unicast list and count in the front of the buffer */ |
32e7bfc4 | 1191 | mac_data->entries = uc_count; |
ccffad25 | 1192 | i = 0; |
32e7bfc4 | 1193 | netdev_for_each_uc_addr(ha, dev) |
ccffad25 | 1194 | memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); |
f565a7c2 AW |
1195 | |
1196 | sg_set_buf(&sg[0], mac_data, | |
32e7bfc4 | 1197 | sizeof(mac_data->entries) + (uc_count * ETH_ALEN)); |
f565a7c2 AW |
1198 | |
1199 | /* multicast list and count fill the end */ | |
32e7bfc4 | 1200 | mac_data = (void *)&mac_data->macs[uc_count][0]; |
f565a7c2 | 1201 | |
4cd24eaf | 1202 | mac_data->entries = mc_count; |
567ec874 | 1203 | i = 0; |
22bedad3 JP |
1204 | netdev_for_each_mc_addr(ha, dev) |
1205 | memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); | |
f565a7c2 AW |
1206 | |
1207 | sg_set_buf(&sg[1], mac_data, | |
4cd24eaf | 1208 | sizeof(mac_data->entries) + (mc_count * ETH_ALEN)); |
f565a7c2 AW |
1209 | |
1210 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC, | |
d24bae32 | 1211 | VIRTIO_NET_CTRL_MAC_TABLE_SET, sg)) |
99e872ae | 1212 | dev_warn(&dev->dev, "Failed to set MAC filter table.\n"); |
f565a7c2 AW |
1213 | |
1214 | kfree(buf); | |
2af7698e AW |
1215 | } |
1216 | ||
80d5c368 PM |
1217 | static int virtnet_vlan_rx_add_vid(struct net_device *dev, |
1218 | __be16 proto, u16 vid) | |
0bde9569 AW |
1219 | { |
1220 | struct virtnet_info *vi = netdev_priv(dev); | |
1221 | struct scatterlist sg; | |
1222 | ||
23e258e1 | 1223 | sg_init_one(&sg, &vid, sizeof(vid)); |
0bde9569 AW |
1224 | |
1225 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, | |
d24bae32 | 1226 | VIRTIO_NET_CTRL_VLAN_ADD, &sg)) |
0bde9569 | 1227 | dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid); |
8e586137 | 1228 | return 0; |
0bde9569 AW |
1229 | } |
1230 | ||
80d5c368 PM |
1231 | static int virtnet_vlan_rx_kill_vid(struct net_device *dev, |
1232 | __be16 proto, u16 vid) | |
0bde9569 AW |
1233 | { |
1234 | struct virtnet_info *vi = netdev_priv(dev); | |
1235 | struct scatterlist sg; | |
1236 | ||
23e258e1 | 1237 | sg_init_one(&sg, &vid, sizeof(vid)); |
0bde9569 AW |
1238 | |
1239 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, | |
d24bae32 | 1240 | VIRTIO_NET_CTRL_VLAN_DEL, &sg)) |
0bde9569 | 1241 | dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid); |
8e586137 | 1242 | return 0; |
0bde9569 AW |
1243 | } |
1244 | ||
8898c21c | 1245 | static void virtnet_clean_affinity(struct virtnet_info *vi, long hcpu) |
986a4f4d JW |
1246 | { |
1247 | int i; | |
1248 | ||
8898c21c WG |
1249 | if (vi->affinity_hint_set) { |
1250 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
47be2479 WG |
1251 | virtqueue_set_affinity(vi->rq[i].vq, -1); |
1252 | virtqueue_set_affinity(vi->sq[i].vq, -1); | |
1253 | } | |
1254 | ||
8898c21c WG |
1255 | vi->affinity_hint_set = false; |
1256 | } | |
8898c21c | 1257 | } |
47be2479 | 1258 | |
8898c21c WG |
1259 | static void virtnet_set_affinity(struct virtnet_info *vi) |
1260 | { | |
1261 | int i; | |
1262 | int cpu; | |
986a4f4d JW |
1263 | |
1264 | /* In multiqueue mode, when the number of cpu is equal to the number of | |
1265 | * queue pairs, we let the queue pairs to be private to one cpu by | |
1266 | * setting the affinity hint to eliminate the contention. | |
1267 | */ | |
8898c21c WG |
1268 | if (vi->curr_queue_pairs == 1 || |
1269 | vi->max_queue_pairs != num_online_cpus()) { | |
1270 | virtnet_clean_affinity(vi, -1); | |
1271 | return; | |
986a4f4d JW |
1272 | } |
1273 | ||
8898c21c WG |
1274 | i = 0; |
1275 | for_each_online_cpu(cpu) { | |
986a4f4d JW |
1276 | virtqueue_set_affinity(vi->rq[i].vq, cpu); |
1277 | virtqueue_set_affinity(vi->sq[i].vq, cpu); | |
9bb8ca86 | 1278 | netif_set_xps_queue(vi->dev, cpumask_of(cpu), i); |
8898c21c | 1279 | i++; |
986a4f4d JW |
1280 | } |
1281 | ||
8898c21c | 1282 | vi->affinity_hint_set = true; |
986a4f4d JW |
1283 | } |
1284 | ||
8de4b2f3 WG |
1285 | static int virtnet_cpu_callback(struct notifier_block *nfb, |
1286 | unsigned long action, void *hcpu) | |
1287 | { | |
1288 | struct virtnet_info *vi = container_of(nfb, struct virtnet_info, nb); | |
1289 | ||
1290 | switch(action & ~CPU_TASKS_FROZEN) { | |
1291 | case CPU_ONLINE: | |
1292 | case CPU_DOWN_FAILED: | |
1293 | case CPU_DEAD: | |
1294 | virtnet_set_affinity(vi); | |
1295 | break; | |
1296 | case CPU_DOWN_PREPARE: | |
1297 | virtnet_clean_affinity(vi, (long)hcpu); | |
1298 | break; | |
1299 | default: | |
1300 | break; | |
1301 | } | |
3ab098df | 1302 | |
8de4b2f3 | 1303 | return NOTIFY_OK; |
986a4f4d JW |
1304 | } |
1305 | ||
8f9f4668 RJ |
1306 | static void virtnet_get_ringparam(struct net_device *dev, |
1307 | struct ethtool_ringparam *ring) | |
1308 | { | |
1309 | struct virtnet_info *vi = netdev_priv(dev); | |
1310 | ||
986a4f4d JW |
1311 | ring->rx_max_pending = virtqueue_get_vring_size(vi->rq[0].vq); |
1312 | ring->tx_max_pending = virtqueue_get_vring_size(vi->sq[0].vq); | |
8f9f4668 RJ |
1313 | ring->rx_pending = ring->rx_max_pending; |
1314 | ring->tx_pending = ring->tx_max_pending; | |
8f9f4668 RJ |
1315 | } |
1316 | ||
66846048 RJ |
1317 | |
1318 | static void virtnet_get_drvinfo(struct net_device *dev, | |
1319 | struct ethtool_drvinfo *info) | |
1320 | { | |
1321 | struct virtnet_info *vi = netdev_priv(dev); | |
1322 | struct virtio_device *vdev = vi->vdev; | |
1323 | ||
1324 | strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver)); | |
1325 | strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version)); | |
1326 | strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info)); | |
1327 | ||
1328 | } | |
1329 | ||
d73bcd2c JW |
1330 | /* TODO: Eliminate OOO packets during switching */ |
1331 | static int virtnet_set_channels(struct net_device *dev, | |
1332 | struct ethtool_channels *channels) | |
1333 | { | |
1334 | struct virtnet_info *vi = netdev_priv(dev); | |
1335 | u16 queue_pairs = channels->combined_count; | |
1336 | int err; | |
1337 | ||
1338 | /* We don't support separate rx/tx channels. | |
1339 | * We don't allow setting 'other' channels. | |
1340 | */ | |
1341 | if (channels->rx_count || channels->tx_count || channels->other_count) | |
1342 | return -EINVAL; | |
1343 | ||
c18e9cd6 | 1344 | if (queue_pairs > vi->max_queue_pairs || queue_pairs == 0) |
d73bcd2c JW |
1345 | return -EINVAL; |
1346 | ||
47be2479 | 1347 | get_online_cpus(); |
d73bcd2c JW |
1348 | err = virtnet_set_queues(vi, queue_pairs); |
1349 | if (!err) { | |
1350 | netif_set_real_num_tx_queues(dev, queue_pairs); | |
1351 | netif_set_real_num_rx_queues(dev, queue_pairs); | |
1352 | ||
8898c21c | 1353 | virtnet_set_affinity(vi); |
d73bcd2c | 1354 | } |
47be2479 | 1355 | put_online_cpus(); |
d73bcd2c JW |
1356 | |
1357 | return err; | |
1358 | } | |
1359 | ||
1360 | static void virtnet_get_channels(struct net_device *dev, | |
1361 | struct ethtool_channels *channels) | |
1362 | { | |
1363 | struct virtnet_info *vi = netdev_priv(dev); | |
1364 | ||
1365 | channels->combined_count = vi->curr_queue_pairs; | |
1366 | channels->max_combined = vi->max_queue_pairs; | |
1367 | channels->max_other = 0; | |
1368 | channels->rx_count = 0; | |
1369 | channels->tx_count = 0; | |
1370 | channels->other_count = 0; | |
1371 | } | |
1372 | ||
0fc0b732 | 1373 | static const struct ethtool_ops virtnet_ethtool_ops = { |
66846048 | 1374 | .get_drvinfo = virtnet_get_drvinfo, |
9f4d26d0 | 1375 | .get_link = ethtool_op_get_link, |
8f9f4668 | 1376 | .get_ringparam = virtnet_get_ringparam, |
d73bcd2c JW |
1377 | .set_channels = virtnet_set_channels, |
1378 | .get_channels = virtnet_get_channels, | |
a9ea3fc6 HX |
1379 | }; |
1380 | ||
39da5814 MM |
1381 | #define MIN_MTU 68 |
1382 | #define MAX_MTU 65535 | |
1383 | ||
1384 | static int virtnet_change_mtu(struct net_device *dev, int new_mtu) | |
1385 | { | |
1386 | if (new_mtu < MIN_MTU || new_mtu > MAX_MTU) | |
1387 | return -EINVAL; | |
1388 | dev->mtu = new_mtu; | |
1389 | return 0; | |
1390 | } | |
1391 | ||
76288b4e SH |
1392 | static const struct net_device_ops virtnet_netdev = { |
1393 | .ndo_open = virtnet_open, | |
1394 | .ndo_stop = virtnet_close, | |
1395 | .ndo_start_xmit = start_xmit, | |
1396 | .ndo_validate_addr = eth_validate_addr, | |
9c46f6d4 | 1397 | .ndo_set_mac_address = virtnet_set_mac_address, |
2af7698e | 1398 | .ndo_set_rx_mode = virtnet_set_rx_mode, |
76288b4e | 1399 | .ndo_change_mtu = virtnet_change_mtu, |
3fa2a1df | 1400 | .ndo_get_stats64 = virtnet_stats, |
1824a989 AW |
1401 | .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid, |
1402 | .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid, | |
76288b4e SH |
1403 | #ifdef CONFIG_NET_POLL_CONTROLLER |
1404 | .ndo_poll_controller = virtnet_netpoll, | |
1405 | #endif | |
91815639 JW |
1406 | #ifdef CONFIG_NET_RX_BUSY_POLL |
1407 | .ndo_busy_poll = virtnet_busy_poll, | |
1408 | #endif | |
76288b4e SH |
1409 | }; |
1410 | ||
586d17c5 | 1411 | static void virtnet_config_changed_work(struct work_struct *work) |
9f4d26d0 | 1412 | { |
586d17c5 JW |
1413 | struct virtnet_info *vi = |
1414 | container_of(work, struct virtnet_info, config_work); | |
9f4d26d0 MM |
1415 | u16 v; |
1416 | ||
586d17c5 JW |
1417 | mutex_lock(&vi->config_lock); |
1418 | if (!vi->config_enable) | |
1419 | goto done; | |
1420 | ||
855e0c52 RR |
1421 | if (virtio_cread_feature(vi->vdev, VIRTIO_NET_F_STATUS, |
1422 | struct virtio_net_config, status, &v) < 0) | |
586d17c5 JW |
1423 | goto done; |
1424 | ||
1425 | if (v & VIRTIO_NET_S_ANNOUNCE) { | |
ee89bab1 | 1426 | netdev_notify_peers(vi->dev); |
586d17c5 JW |
1427 | virtnet_ack_link_announce(vi); |
1428 | } | |
9f4d26d0 MM |
1429 | |
1430 | /* Ignore unknown (future) status bits */ | |
1431 | v &= VIRTIO_NET_S_LINK_UP; | |
1432 | ||
1433 | if (vi->status == v) | |
586d17c5 | 1434 | goto done; |
9f4d26d0 MM |
1435 | |
1436 | vi->status = v; | |
1437 | ||
1438 | if (vi->status & VIRTIO_NET_S_LINK_UP) { | |
1439 | netif_carrier_on(vi->dev); | |
986a4f4d | 1440 | netif_tx_wake_all_queues(vi->dev); |
9f4d26d0 MM |
1441 | } else { |
1442 | netif_carrier_off(vi->dev); | |
986a4f4d | 1443 | netif_tx_stop_all_queues(vi->dev); |
9f4d26d0 | 1444 | } |
586d17c5 JW |
1445 | done: |
1446 | mutex_unlock(&vi->config_lock); | |
9f4d26d0 MM |
1447 | } |
1448 | ||
1449 | static void virtnet_config_changed(struct virtio_device *vdev) | |
1450 | { | |
1451 | struct virtnet_info *vi = vdev->priv; | |
1452 | ||
3b07e9ca | 1453 | schedule_work(&vi->config_work); |
9f4d26d0 MM |
1454 | } |
1455 | ||
986a4f4d JW |
1456 | static void virtnet_free_queues(struct virtnet_info *vi) |
1457 | { | |
d4fb84ee AV |
1458 | int i; |
1459 | ||
1460 | for (i = 0; i < vi->max_queue_pairs; i++) | |
1461 | netif_napi_del(&vi->rq[i].napi); | |
1462 | ||
986a4f4d JW |
1463 | kfree(vi->rq); |
1464 | kfree(vi->sq); | |
1465 | } | |
1466 | ||
1467 | static void free_receive_bufs(struct virtnet_info *vi) | |
1468 | { | |
1469 | int i; | |
1470 | ||
1471 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
1472 | while (vi->rq[i].pages) | |
1473 | __free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0); | |
1474 | } | |
1475 | } | |
1476 | ||
fb51879d MD |
1477 | static void free_receive_page_frags(struct virtnet_info *vi) |
1478 | { | |
1479 | int i; | |
1480 | for (i = 0; i < vi->max_queue_pairs; i++) | |
1481 | if (vi->rq[i].alloc_frag.page) | |
1482 | put_page(vi->rq[i].alloc_frag.page); | |
1483 | } | |
1484 | ||
986a4f4d JW |
1485 | static void free_unused_bufs(struct virtnet_info *vi) |
1486 | { | |
1487 | void *buf; | |
1488 | int i; | |
1489 | ||
1490 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
1491 | struct virtqueue *vq = vi->sq[i].vq; | |
1492 | while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) | |
1493 | dev_kfree_skb(buf); | |
1494 | } | |
1495 | ||
1496 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
1497 | struct virtqueue *vq = vi->rq[i].vq; | |
1498 | ||
1499 | while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) { | |
ab7db917 MD |
1500 | if (vi->mergeable_rx_bufs) { |
1501 | unsigned long ctx = (unsigned long)buf; | |
1502 | void *base = mergeable_ctx_to_buf_address(ctx); | |
1503 | put_page(virt_to_head_page(base)); | |
1504 | } else if (vi->big_packets) { | |
fa9fac17 | 1505 | give_pages(&vi->rq[i], buf); |
ab7db917 | 1506 | } else { |
986a4f4d | 1507 | dev_kfree_skb(buf); |
ab7db917 | 1508 | } |
986a4f4d | 1509 | } |
986a4f4d JW |
1510 | } |
1511 | } | |
1512 | ||
e9d7417b JW |
1513 | static void virtnet_del_vqs(struct virtnet_info *vi) |
1514 | { | |
1515 | struct virtio_device *vdev = vi->vdev; | |
1516 | ||
8898c21c | 1517 | virtnet_clean_affinity(vi, -1); |
986a4f4d | 1518 | |
e9d7417b | 1519 | vdev->config->del_vqs(vdev); |
986a4f4d JW |
1520 | |
1521 | virtnet_free_queues(vi); | |
e9d7417b JW |
1522 | } |
1523 | ||
986a4f4d | 1524 | static int virtnet_find_vqs(struct virtnet_info *vi) |
3f9c10b0 | 1525 | { |
986a4f4d JW |
1526 | vq_callback_t **callbacks; |
1527 | struct virtqueue **vqs; | |
1528 | int ret = -ENOMEM; | |
1529 | int i, total_vqs; | |
1530 | const char **names; | |
1531 | ||
1532 | /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by | |
1533 | * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by | |
1534 | * possible control vq. | |
1535 | */ | |
1536 | total_vqs = vi->max_queue_pairs * 2 + | |
1537 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ); | |
1538 | ||
1539 | /* Allocate space for find_vqs parameters */ | |
1540 | vqs = kzalloc(total_vqs * sizeof(*vqs), GFP_KERNEL); | |
1541 | if (!vqs) | |
1542 | goto err_vq; | |
1543 | callbacks = kmalloc(total_vqs * sizeof(*callbacks), GFP_KERNEL); | |
1544 | if (!callbacks) | |
1545 | goto err_callback; | |
1546 | names = kmalloc(total_vqs * sizeof(*names), GFP_KERNEL); | |
1547 | if (!names) | |
1548 | goto err_names; | |
1549 | ||
1550 | /* Parameters for control virtqueue, if any */ | |
1551 | if (vi->has_cvq) { | |
1552 | callbacks[total_vqs - 1] = NULL; | |
1553 | names[total_vqs - 1] = "control"; | |
1554 | } | |
3f9c10b0 | 1555 | |
986a4f4d JW |
1556 | /* Allocate/initialize parameters for send/receive virtqueues */ |
1557 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
1558 | callbacks[rxq2vq(i)] = skb_recv_done; | |
1559 | callbacks[txq2vq(i)] = skb_xmit_done; | |
1560 | sprintf(vi->rq[i].name, "input.%d", i); | |
1561 | sprintf(vi->sq[i].name, "output.%d", i); | |
1562 | names[rxq2vq(i)] = vi->rq[i].name; | |
1563 | names[txq2vq(i)] = vi->sq[i].name; | |
1564 | } | |
3f9c10b0 | 1565 | |
986a4f4d JW |
1566 | ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs, callbacks, |
1567 | names); | |
1568 | if (ret) | |
1569 | goto err_find; | |
3f9c10b0 | 1570 | |
986a4f4d JW |
1571 | if (vi->has_cvq) { |
1572 | vi->cvq = vqs[total_vqs - 1]; | |
3f9c10b0 | 1573 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN)) |
f646968f | 1574 | vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER; |
3f9c10b0 | 1575 | } |
986a4f4d JW |
1576 | |
1577 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
1578 | vi->rq[i].vq = vqs[rxq2vq(i)]; | |
1579 | vi->sq[i].vq = vqs[txq2vq(i)]; | |
1580 | } | |
1581 | ||
1582 | kfree(names); | |
1583 | kfree(callbacks); | |
1584 | kfree(vqs); | |
1585 | ||
3f9c10b0 | 1586 | return 0; |
986a4f4d JW |
1587 | |
1588 | err_find: | |
1589 | kfree(names); | |
1590 | err_names: | |
1591 | kfree(callbacks); | |
1592 | err_callback: | |
1593 | kfree(vqs); | |
1594 | err_vq: | |
1595 | return ret; | |
1596 | } | |
1597 | ||
1598 | static int virtnet_alloc_queues(struct virtnet_info *vi) | |
1599 | { | |
1600 | int i; | |
1601 | ||
1602 | vi->sq = kzalloc(sizeof(*vi->sq) * vi->max_queue_pairs, GFP_KERNEL); | |
1603 | if (!vi->sq) | |
1604 | goto err_sq; | |
1605 | vi->rq = kzalloc(sizeof(*vi->rq) * vi->max_queue_pairs, GFP_KERNEL); | |
008d4278 | 1606 | if (!vi->rq) |
986a4f4d JW |
1607 | goto err_rq; |
1608 | ||
1609 | INIT_DELAYED_WORK(&vi->refill, refill_work); | |
1610 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
1611 | vi->rq[i].pages = NULL; | |
1612 | netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll, | |
1613 | napi_weight); | |
91815639 | 1614 | napi_hash_add(&vi->rq[i].napi); |
986a4f4d JW |
1615 | |
1616 | sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg)); | |
ab7db917 | 1617 | ewma_init(&vi->rq[i].mrg_avg_pkt_len, 1, RECEIVE_AVG_WEIGHT); |
986a4f4d JW |
1618 | sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg)); |
1619 | } | |
1620 | ||
1621 | return 0; | |
1622 | ||
1623 | err_rq: | |
1624 | kfree(vi->sq); | |
1625 | err_sq: | |
1626 | return -ENOMEM; | |
1627 | } | |
1628 | ||
1629 | static int init_vqs(struct virtnet_info *vi) | |
1630 | { | |
1631 | int ret; | |
1632 | ||
1633 | /* Allocate send & receive queues */ | |
1634 | ret = virtnet_alloc_queues(vi); | |
1635 | if (ret) | |
1636 | goto err; | |
1637 | ||
1638 | ret = virtnet_find_vqs(vi); | |
1639 | if (ret) | |
1640 | goto err_free; | |
1641 | ||
47be2479 | 1642 | get_online_cpus(); |
8898c21c | 1643 | virtnet_set_affinity(vi); |
47be2479 WG |
1644 | put_online_cpus(); |
1645 | ||
986a4f4d JW |
1646 | return 0; |
1647 | ||
1648 | err_free: | |
1649 | virtnet_free_queues(vi); | |
1650 | err: | |
1651 | return ret; | |
3f9c10b0 AS |
1652 | } |
1653 | ||
fbf28d78 MD |
1654 | #ifdef CONFIG_SYSFS |
1655 | static ssize_t mergeable_rx_buffer_size_show(struct netdev_rx_queue *queue, | |
1656 | struct rx_queue_attribute *attribute, char *buf) | |
1657 | { | |
1658 | struct virtnet_info *vi = netdev_priv(queue->dev); | |
1659 | unsigned int queue_index = get_netdev_rx_queue_index(queue); | |
1660 | struct ewma *avg; | |
1661 | ||
1662 | BUG_ON(queue_index >= vi->max_queue_pairs); | |
1663 | avg = &vi->rq[queue_index].mrg_avg_pkt_len; | |
1664 | return sprintf(buf, "%u\n", get_mergeable_buf_len(avg)); | |
1665 | } | |
1666 | ||
1667 | static struct rx_queue_attribute mergeable_rx_buffer_size_attribute = | |
1668 | __ATTR_RO(mergeable_rx_buffer_size); | |
1669 | ||
1670 | static struct attribute *virtio_net_mrg_rx_attrs[] = { | |
1671 | &mergeable_rx_buffer_size_attribute.attr, | |
1672 | NULL | |
1673 | }; | |
1674 | ||
1675 | static const struct attribute_group virtio_net_mrg_rx_group = { | |
1676 | .name = "virtio_net", | |
1677 | .attrs = virtio_net_mrg_rx_attrs | |
1678 | }; | |
1679 | #endif | |
1680 | ||
296f96fc RR |
1681 | static int virtnet_probe(struct virtio_device *vdev) |
1682 | { | |
986a4f4d | 1683 | int i, err; |
296f96fc RR |
1684 | struct net_device *dev; |
1685 | struct virtnet_info *vi; | |
986a4f4d JW |
1686 | u16 max_queue_pairs; |
1687 | ||
1688 | /* Find if host supports multiqueue virtio_net device */ | |
855e0c52 RR |
1689 | err = virtio_cread_feature(vdev, VIRTIO_NET_F_MQ, |
1690 | struct virtio_net_config, | |
1691 | max_virtqueue_pairs, &max_queue_pairs); | |
986a4f4d JW |
1692 | |
1693 | /* We need at least 2 queue's */ | |
1694 | if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN || | |
1695 | max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX || | |
1696 | !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) | |
1697 | max_queue_pairs = 1; | |
296f96fc RR |
1698 | |
1699 | /* Allocate ourselves a network device with room for our info */ | |
986a4f4d | 1700 | dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs); |
296f96fc RR |
1701 | if (!dev) |
1702 | return -ENOMEM; | |
1703 | ||
1704 | /* Set up network device as normal. */ | |
f2f2c8b4 | 1705 | dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE; |
76288b4e | 1706 | dev->netdev_ops = &virtnet_netdev; |
296f96fc | 1707 | dev->features = NETIF_F_HIGHDMA; |
3fa2a1df | 1708 | |
7ad24ea4 | 1709 | dev->ethtool_ops = &virtnet_ethtool_ops; |
296f96fc RR |
1710 | SET_NETDEV_DEV(dev, &vdev->dev); |
1711 | ||
1712 | /* Do we support "hardware" checksums? */ | |
98e778c9 | 1713 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) { |
296f96fc | 1714 | /* This opens up the world of extra features. */ |
98e778c9 MM |
1715 | dev->hw_features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST; |
1716 | if (csum) | |
1717 | dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST; | |
1718 | ||
1719 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) { | |
1720 | dev->hw_features |= NETIF_F_TSO | NETIF_F_UFO | |
34a48579 RR |
1721 | | NETIF_F_TSO_ECN | NETIF_F_TSO6; |
1722 | } | |
5539ae96 | 1723 | /* Individual feature bits: what can host handle? */ |
98e778c9 MM |
1724 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4)) |
1725 | dev->hw_features |= NETIF_F_TSO; | |
1726 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6)) | |
1727 | dev->hw_features |= NETIF_F_TSO6; | |
1728 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN)) | |
1729 | dev->hw_features |= NETIF_F_TSO_ECN; | |
1730 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO)) | |
1731 | dev->hw_features |= NETIF_F_UFO; | |
1732 | ||
1733 | if (gso) | |
1734 | dev->features |= dev->hw_features & (NETIF_F_ALL_TSO|NETIF_F_UFO); | |
1735 | /* (!csum && gso) case will be fixed by register_netdev() */ | |
296f96fc | 1736 | } |
4f49129b TH |
1737 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM)) |
1738 | dev->features |= NETIF_F_RXCSUM; | |
296f96fc | 1739 | |
4fda8302 JW |
1740 | dev->vlan_features = dev->features; |
1741 | ||
296f96fc | 1742 | /* Configuration may specify what MAC to use. Otherwise random. */ |
855e0c52 RR |
1743 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) |
1744 | virtio_cread_bytes(vdev, | |
1745 | offsetof(struct virtio_net_config, mac), | |
1746 | dev->dev_addr, dev->addr_len); | |
1747 | else | |
f2cedb63 | 1748 | eth_hw_addr_random(dev); |
296f96fc RR |
1749 | |
1750 | /* Set up our device-specific information */ | |
1751 | vi = netdev_priv(dev); | |
296f96fc RR |
1752 | vi->dev = dev; |
1753 | vi->vdev = vdev; | |
d9d5dcc8 | 1754 | vdev->priv = vi; |
3fa2a1df | 1755 | vi->stats = alloc_percpu(struct virtnet_stats); |
1756 | err = -ENOMEM; | |
1757 | if (vi->stats == NULL) | |
1758 | goto free; | |
1759 | ||
827da44c JS |
1760 | for_each_possible_cpu(i) { |
1761 | struct virtnet_stats *virtnet_stats; | |
1762 | virtnet_stats = per_cpu_ptr(vi->stats, i); | |
1763 | u64_stats_init(&virtnet_stats->tx_syncp); | |
1764 | u64_stats_init(&virtnet_stats->rx_syncp); | |
1765 | } | |
1766 | ||
586d17c5 JW |
1767 | mutex_init(&vi->config_lock); |
1768 | vi->config_enable = true; | |
1769 | INIT_WORK(&vi->config_work, virtnet_config_changed_work); | |
296f96fc | 1770 | |
97402b96 | 1771 | /* If we can receive ANY GSO packets, we must allocate large ones. */ |
8e95a202 JP |
1772 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) || |
1773 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) || | |
0e7ede80 JW |
1774 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN) || |
1775 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_UFO)) | |
97402b96 HX |
1776 | vi->big_packets = true; |
1777 | ||
3f2c31d9 MM |
1778 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF)) |
1779 | vi->mergeable_rx_bufs = true; | |
1780 | ||
e7428e95 MT |
1781 | if (virtio_has_feature(vdev, VIRTIO_F_ANY_LAYOUT)) |
1782 | vi->any_header_sg = true; | |
1783 | ||
986a4f4d JW |
1784 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) |
1785 | vi->has_cvq = true; | |
1786 | ||
6ebbc1a6 ZH |
1787 | if (vi->any_header_sg) { |
1788 | if (vi->mergeable_rx_bufs) | |
1789 | dev->needed_headroom = sizeof(struct virtio_net_hdr_mrg_rxbuf); | |
1790 | else | |
1791 | dev->needed_headroom = sizeof(struct virtio_net_hdr); | |
1792 | } | |
1793 | ||
986a4f4d JW |
1794 | /* Use single tx/rx queue pair as default */ |
1795 | vi->curr_queue_pairs = 1; | |
1796 | vi->max_queue_pairs = max_queue_pairs; | |
1797 | ||
1798 | /* Allocate/initialize the rx/tx queues, and invoke find_vqs */ | |
3f9c10b0 | 1799 | err = init_vqs(vi); |
d2a7ddda | 1800 | if (err) |
9bb8ca86 | 1801 | goto free_stats; |
296f96fc | 1802 | |
fbf28d78 MD |
1803 | #ifdef CONFIG_SYSFS |
1804 | if (vi->mergeable_rx_bufs) | |
1805 | dev->sysfs_rx_queue_group = &virtio_net_mrg_rx_group; | |
1806 | #endif | |
0f13b66b ZYW |
1807 | netif_set_real_num_tx_queues(dev, vi->curr_queue_pairs); |
1808 | netif_set_real_num_rx_queues(dev, vi->curr_queue_pairs); | |
986a4f4d | 1809 | |
296f96fc RR |
1810 | err = register_netdev(dev); |
1811 | if (err) { | |
1812 | pr_debug("virtio_net: registering device failed\n"); | |
d2a7ddda | 1813 | goto free_vqs; |
296f96fc | 1814 | } |
b3369c1f RR |
1815 | |
1816 | /* Last of all, set up some receive buffers. */ | |
55257d72 | 1817 | for (i = 0; i < vi->curr_queue_pairs; i++) { |
986a4f4d JW |
1818 | try_fill_recv(&vi->rq[i], GFP_KERNEL); |
1819 | ||
1820 | /* If we didn't even get one input buffer, we're useless. */ | |
be121f46 JW |
1821 | if (vi->rq[i].vq->num_free == |
1822 | virtqueue_get_vring_size(vi->rq[i].vq)) { | |
986a4f4d JW |
1823 | free_unused_bufs(vi); |
1824 | err = -ENOMEM; | |
1825 | goto free_recv_bufs; | |
1826 | } | |
b3369c1f RR |
1827 | } |
1828 | ||
8de4b2f3 WG |
1829 | vi->nb.notifier_call = &virtnet_cpu_callback; |
1830 | err = register_hotcpu_notifier(&vi->nb); | |
1831 | if (err) { | |
1832 | pr_debug("virtio_net: registering cpu notifier failed\n"); | |
1833 | goto free_recv_bufs; | |
1834 | } | |
1835 | ||
167c25e4 JW |
1836 | /* Assume link up if device can't report link status, |
1837 | otherwise get link status from config. */ | |
1838 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) { | |
1839 | netif_carrier_off(dev); | |
3b07e9ca | 1840 | schedule_work(&vi->config_work); |
167c25e4 JW |
1841 | } else { |
1842 | vi->status = VIRTIO_NET_S_LINK_UP; | |
1843 | netif_carrier_on(dev); | |
1844 | } | |
9f4d26d0 | 1845 | |
986a4f4d JW |
1846 | pr_debug("virtnet: registered device %s with %d RX and TX vq's\n", |
1847 | dev->name, max_queue_pairs); | |
1848 | ||
296f96fc RR |
1849 | return 0; |
1850 | ||
986a4f4d JW |
1851 | free_recv_bufs: |
1852 | free_receive_bufs(vi); | |
b3369c1f | 1853 | unregister_netdev(dev); |
d2a7ddda | 1854 | free_vqs: |
986a4f4d | 1855 | cancel_delayed_work_sync(&vi->refill); |
fb51879d | 1856 | free_receive_page_frags(vi); |
e9d7417b | 1857 | virtnet_del_vqs(vi); |
3fa2a1df | 1858 | free_stats: |
1859 | free_percpu(vi->stats); | |
296f96fc RR |
1860 | free: |
1861 | free_netdev(dev); | |
1862 | return err; | |
1863 | } | |
1864 | ||
04486ed0 | 1865 | static void remove_vq_common(struct virtnet_info *vi) |
296f96fc | 1866 | { |
04486ed0 | 1867 | vi->vdev->config->reset(vi->vdev); |
830a8a97 SM |
1868 | |
1869 | /* Free unused buffers in both send and recv, if any. */ | |
9ab86bbc | 1870 | free_unused_bufs(vi); |
fb6813f4 | 1871 | |
986a4f4d | 1872 | free_receive_bufs(vi); |
d2a7ddda | 1873 | |
fb51879d MD |
1874 | free_receive_page_frags(vi); |
1875 | ||
986a4f4d | 1876 | virtnet_del_vqs(vi); |
04486ed0 AS |
1877 | } |
1878 | ||
8cc085d6 | 1879 | static void virtnet_remove(struct virtio_device *vdev) |
04486ed0 AS |
1880 | { |
1881 | struct virtnet_info *vi = vdev->priv; | |
1882 | ||
8de4b2f3 WG |
1883 | unregister_hotcpu_notifier(&vi->nb); |
1884 | ||
586d17c5 JW |
1885 | /* Prevent config work handler from accessing the device. */ |
1886 | mutex_lock(&vi->config_lock); | |
1887 | vi->config_enable = false; | |
1888 | mutex_unlock(&vi->config_lock); | |
1889 | ||
04486ed0 AS |
1890 | unregister_netdev(vi->dev); |
1891 | ||
1892 | remove_vq_common(vi); | |
fb6813f4 | 1893 | |
586d17c5 JW |
1894 | flush_work(&vi->config_work); |
1895 | ||
2e66f55b | 1896 | free_percpu(vi->stats); |
74b2553f | 1897 | free_netdev(vi->dev); |
296f96fc RR |
1898 | } |
1899 | ||
89107000 | 1900 | #ifdef CONFIG_PM_SLEEP |
0741bcb5 AS |
1901 | static int virtnet_freeze(struct virtio_device *vdev) |
1902 | { | |
1903 | struct virtnet_info *vi = vdev->priv; | |
986a4f4d | 1904 | int i; |
0741bcb5 | 1905 | |
ec9debbd JW |
1906 | unregister_hotcpu_notifier(&vi->nb); |
1907 | ||
586d17c5 JW |
1908 | /* Prevent config work handler from accessing the device */ |
1909 | mutex_lock(&vi->config_lock); | |
1910 | vi->config_enable = false; | |
1911 | mutex_unlock(&vi->config_lock); | |
1912 | ||
0741bcb5 AS |
1913 | netif_device_detach(vi->dev); |
1914 | cancel_delayed_work_sync(&vi->refill); | |
1915 | ||
91815639 | 1916 | if (netif_running(vi->dev)) { |
986a4f4d JW |
1917 | for (i = 0; i < vi->max_queue_pairs; i++) { |
1918 | napi_disable(&vi->rq[i].napi); | |
91815639 | 1919 | napi_hash_del(&vi->rq[i].napi); |
986a4f4d JW |
1920 | netif_napi_del(&vi->rq[i].napi); |
1921 | } | |
91815639 | 1922 | } |
0741bcb5 AS |
1923 | |
1924 | remove_vq_common(vi); | |
1925 | ||
586d17c5 JW |
1926 | flush_work(&vi->config_work); |
1927 | ||
0741bcb5 AS |
1928 | return 0; |
1929 | } | |
1930 | ||
1931 | static int virtnet_restore(struct virtio_device *vdev) | |
1932 | { | |
1933 | struct virtnet_info *vi = vdev->priv; | |
986a4f4d | 1934 | int err, i; |
0741bcb5 AS |
1935 | |
1936 | err = init_vqs(vi); | |
1937 | if (err) | |
1938 | return err; | |
1939 | ||
6cd4ce00 JW |
1940 | if (netif_running(vi->dev)) { |
1941 | for (i = 0; i < vi->curr_queue_pairs; i++) | |
1942 | if (!try_fill_recv(&vi->rq[i], GFP_KERNEL)) | |
1943 | schedule_delayed_work(&vi->refill, 0); | |
1944 | ||
986a4f4d JW |
1945 | for (i = 0; i < vi->max_queue_pairs; i++) |
1946 | virtnet_napi_enable(&vi->rq[i]); | |
6cd4ce00 | 1947 | } |
0741bcb5 AS |
1948 | |
1949 | netif_device_attach(vi->dev); | |
1950 | ||
586d17c5 JW |
1951 | mutex_lock(&vi->config_lock); |
1952 | vi->config_enable = true; | |
1953 | mutex_unlock(&vi->config_lock); | |
1954 | ||
35ed159b | 1955 | rtnl_lock(); |
986a4f4d | 1956 | virtnet_set_queues(vi, vi->curr_queue_pairs); |
35ed159b | 1957 | rtnl_unlock(); |
986a4f4d | 1958 | |
ec9debbd JW |
1959 | err = register_hotcpu_notifier(&vi->nb); |
1960 | if (err) | |
1961 | return err; | |
1962 | ||
0741bcb5 AS |
1963 | return 0; |
1964 | } | |
1965 | #endif | |
1966 | ||
296f96fc RR |
1967 | static struct virtio_device_id id_table[] = { |
1968 | { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID }, | |
1969 | { 0 }, | |
1970 | }; | |
1971 | ||
c45a6816 | 1972 | static unsigned int features[] = { |
5e4fe5c4 MM |
1973 | VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM, |
1974 | VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC, | |
c45a6816 | 1975 | VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6, |
97402b96 | 1976 | VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, |
5c516751 | 1977 | VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO, |
2a41f71d | 1978 | VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ, |
0bde9569 | 1979 | VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, |
986a4f4d | 1980 | VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ, |
7e58d5ae | 1981 | VIRTIO_NET_F_CTRL_MAC_ADDR, |
e7428e95 | 1982 | VIRTIO_F_ANY_LAYOUT, |
c45a6816 RR |
1983 | }; |
1984 | ||
22402529 | 1985 | static struct virtio_driver virtio_net_driver = { |
c45a6816 RR |
1986 | .feature_table = features, |
1987 | .feature_table_size = ARRAY_SIZE(features), | |
296f96fc RR |
1988 | .driver.name = KBUILD_MODNAME, |
1989 | .driver.owner = THIS_MODULE, | |
1990 | .id_table = id_table, | |
1991 | .probe = virtnet_probe, | |
8cc085d6 | 1992 | .remove = virtnet_remove, |
9f4d26d0 | 1993 | .config_changed = virtnet_config_changed, |
89107000 | 1994 | #ifdef CONFIG_PM_SLEEP |
0741bcb5 AS |
1995 | .freeze = virtnet_freeze, |
1996 | .restore = virtnet_restore, | |
1997 | #endif | |
296f96fc RR |
1998 | }; |
1999 | ||
b2a17029 | 2000 | module_virtio_driver(virtio_net_driver); |
296f96fc RR |
2001 | |
2002 | MODULE_DEVICE_TABLE(virtio, id_table); | |
2003 | MODULE_DESCRIPTION("Virtio network driver"); | |
2004 | MODULE_LICENSE("GPL"); |