]>
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 | ||
3fa2a1df | 126 | /* Active statistics */ |
127 | struct virtnet_stats __percpu *stats; | |
128 | ||
3161e453 RR |
129 | /* Work struct for refilling if we run low on memory. */ |
130 | struct delayed_work refill; | |
131 | ||
586d17c5 JW |
132 | /* Work struct for config space updates */ |
133 | struct work_struct config_work; | |
134 | ||
986a4f4d JW |
135 | /* Does the affinity hint is set for virtqueues? */ |
136 | bool affinity_hint_set; | |
47be2479 | 137 | |
8de4b2f3 WG |
138 | /* CPU hot plug notifier */ |
139 | struct notifier_block nb; | |
296f96fc RR |
140 | }; |
141 | ||
b3f24698 RR |
142 | struct skb_vnet_hdr { |
143 | union { | |
144 | struct virtio_net_hdr hdr; | |
145 | struct virtio_net_hdr_mrg_rxbuf mhdr; | |
146 | }; | |
147 | }; | |
148 | ||
9ab86bbc SM |
149 | struct padded_vnet_hdr { |
150 | struct virtio_net_hdr hdr; | |
151 | /* | |
152 | * virtio_net_hdr should be in a separated sg buffer because of a | |
153 | * QEMU bug, and data sg buffer shares same page with this header sg. | |
154 | * This padding makes next sg 16 byte aligned after virtio_net_hdr. | |
155 | */ | |
156 | char padding[6]; | |
157 | }; | |
158 | ||
986a4f4d JW |
159 | /* Converting between virtqueue no. and kernel tx/rx queue no. |
160 | * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq | |
161 | */ | |
162 | static int vq2txq(struct virtqueue *vq) | |
163 | { | |
9d0ca6ed | 164 | return (vq->index - 1) / 2; |
986a4f4d JW |
165 | } |
166 | ||
167 | static int txq2vq(int txq) | |
168 | { | |
169 | return txq * 2 + 1; | |
170 | } | |
171 | ||
172 | static int vq2rxq(struct virtqueue *vq) | |
173 | { | |
9d0ca6ed | 174 | return vq->index / 2; |
986a4f4d JW |
175 | } |
176 | ||
177 | static int rxq2vq(int rxq) | |
178 | { | |
179 | return rxq * 2; | |
180 | } | |
181 | ||
b3f24698 | 182 | static inline struct skb_vnet_hdr *skb_vnet_hdr(struct sk_buff *skb) |
296f96fc | 183 | { |
b3f24698 | 184 | return (struct skb_vnet_hdr *)skb->cb; |
296f96fc RR |
185 | } |
186 | ||
9ab86bbc SM |
187 | /* |
188 | * private is used to chain pages for big packets, put the whole | |
189 | * most recent used list in the beginning for reuse | |
190 | */ | |
e9d7417b | 191 | static void give_pages(struct receive_queue *rq, struct page *page) |
0a888fd1 | 192 | { |
9ab86bbc | 193 | struct page *end; |
0a888fd1 | 194 | |
e9d7417b | 195 | /* Find end of list, sew whole thing into vi->rq.pages. */ |
9ab86bbc | 196 | for (end = page; end->private; end = (struct page *)end->private); |
e9d7417b JW |
197 | end->private = (unsigned long)rq->pages; |
198 | rq->pages = page; | |
0a888fd1 MM |
199 | } |
200 | ||
e9d7417b | 201 | static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask) |
fb6813f4 | 202 | { |
e9d7417b | 203 | struct page *p = rq->pages; |
fb6813f4 | 204 | |
9ab86bbc | 205 | if (p) { |
e9d7417b | 206 | rq->pages = (struct page *)p->private; |
9ab86bbc SM |
207 | /* clear private here, it is used to chain pages */ |
208 | p->private = 0; | |
209 | } else | |
fb6813f4 RR |
210 | p = alloc_page(gfp_mask); |
211 | return p; | |
212 | } | |
213 | ||
e9d7417b | 214 | static void skb_xmit_done(struct virtqueue *vq) |
296f96fc | 215 | { |
e9d7417b | 216 | struct virtnet_info *vi = vq->vdev->priv; |
296f96fc | 217 | |
2cb9c6ba | 218 | /* Suppress further interrupts. */ |
e9d7417b | 219 | virtqueue_disable_cb(vq); |
11a3a154 | 220 | |
363f1514 | 221 | /* We were probably waiting for more output buffers. */ |
986a4f4d | 222 | netif_wake_subqueue(vi->dev, vq2txq(vq)); |
296f96fc RR |
223 | } |
224 | ||
ab7db917 MD |
225 | static unsigned int mergeable_ctx_to_buf_truesize(unsigned long mrg_ctx) |
226 | { | |
227 | unsigned int truesize = mrg_ctx & (MERGEABLE_BUFFER_ALIGN - 1); | |
228 | return (truesize + 1) * MERGEABLE_BUFFER_ALIGN; | |
229 | } | |
230 | ||
231 | static void *mergeable_ctx_to_buf_address(unsigned long mrg_ctx) | |
232 | { | |
233 | return (void *)(mrg_ctx & -MERGEABLE_BUFFER_ALIGN); | |
234 | ||
235 | } | |
236 | ||
237 | static unsigned long mergeable_buf_to_ctx(void *buf, unsigned int truesize) | |
238 | { | |
239 | unsigned int size = truesize / MERGEABLE_BUFFER_ALIGN; | |
240 | return (unsigned long)buf | (size - 1); | |
241 | } | |
242 | ||
3464645a | 243 | /* Called from bottom half context */ |
e9d7417b | 244 | static struct sk_buff *page_to_skb(struct receive_queue *rq, |
2613af0e MD |
245 | struct page *page, unsigned int offset, |
246 | unsigned int len, unsigned int truesize) | |
9ab86bbc | 247 | { |
e9d7417b | 248 | struct virtnet_info *vi = rq->vq->vdev->priv; |
9ab86bbc SM |
249 | struct sk_buff *skb; |
250 | struct skb_vnet_hdr *hdr; | |
2613af0e | 251 | unsigned int copy, hdr_len, hdr_padded_len; |
9ab86bbc | 252 | char *p; |
fb6813f4 | 253 | |
2613af0e | 254 | p = page_address(page) + offset; |
3f2c31d9 | 255 | |
9ab86bbc SM |
256 | /* copy small packet so we can reuse these pages for small data */ |
257 | skb = netdev_alloc_skb_ip_align(vi->dev, GOOD_COPY_LEN); | |
258 | if (unlikely(!skb)) | |
259 | return NULL; | |
3f2c31d9 | 260 | |
9ab86bbc | 261 | hdr = skb_vnet_hdr(skb); |
3f2c31d9 | 262 | |
9ab86bbc SM |
263 | if (vi->mergeable_rx_bufs) { |
264 | hdr_len = sizeof hdr->mhdr; | |
2613af0e | 265 | hdr_padded_len = sizeof hdr->mhdr; |
9ab86bbc SM |
266 | } else { |
267 | hdr_len = sizeof hdr->hdr; | |
2613af0e | 268 | hdr_padded_len = sizeof(struct padded_vnet_hdr); |
9ab86bbc | 269 | } |
3f2c31d9 | 270 | |
9ab86bbc | 271 | memcpy(hdr, p, hdr_len); |
3f2c31d9 | 272 | |
9ab86bbc | 273 | len -= hdr_len; |
2613af0e MD |
274 | offset += hdr_padded_len; |
275 | p += hdr_padded_len; | |
3f2c31d9 | 276 | |
9ab86bbc SM |
277 | copy = len; |
278 | if (copy > skb_tailroom(skb)) | |
279 | copy = skb_tailroom(skb); | |
280 | memcpy(skb_put(skb, copy), p, copy); | |
3f2c31d9 | 281 | |
9ab86bbc SM |
282 | len -= copy; |
283 | offset += copy; | |
3f2c31d9 | 284 | |
2613af0e MD |
285 | if (vi->mergeable_rx_bufs) { |
286 | if (len) | |
287 | skb_add_rx_frag(skb, 0, page, offset, len, truesize); | |
288 | else | |
289 | put_page(page); | |
290 | return skb; | |
291 | } | |
292 | ||
e878d78b SL |
293 | /* |
294 | * Verify that we can indeed put this data into a skb. | |
295 | * This is here to handle cases when the device erroneously | |
296 | * tries to receive more than is possible. This is usually | |
297 | * the case of a broken device. | |
298 | */ | |
299 | if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) { | |
be443899 | 300 | net_dbg_ratelimited("%s: too much data\n", skb->dev->name); |
e878d78b SL |
301 | dev_kfree_skb(skb); |
302 | return NULL; | |
303 | } | |
2613af0e | 304 | BUG_ON(offset >= PAGE_SIZE); |
9ab86bbc | 305 | while (len) { |
2613af0e MD |
306 | unsigned int frag_size = min((unsigned)PAGE_SIZE - offset, len); |
307 | skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page, offset, | |
308 | frag_size, truesize); | |
309 | len -= frag_size; | |
9ab86bbc SM |
310 | page = (struct page *)page->private; |
311 | offset = 0; | |
312 | } | |
3f2c31d9 | 313 | |
9ab86bbc | 314 | if (page) |
e9d7417b | 315 | give_pages(rq, page); |
3f2c31d9 | 316 | |
9ab86bbc SM |
317 | return skb; |
318 | } | |
3f2c31d9 | 319 | |
f121159d MT |
320 | static struct sk_buff *receive_small(void *buf, unsigned int len) |
321 | { | |
322 | struct sk_buff * skb = buf; | |
323 | ||
324 | len -= sizeof(struct virtio_net_hdr); | |
325 | skb_trim(skb, len); | |
326 | ||
327 | return skb; | |
328 | } | |
329 | ||
330 | static struct sk_buff *receive_big(struct net_device *dev, | |
331 | struct receive_queue *rq, | |
332 | void *buf, | |
333 | unsigned int len) | |
334 | { | |
335 | struct page *page = buf; | |
336 | struct sk_buff *skb = page_to_skb(rq, page, 0, len, PAGE_SIZE); | |
337 | ||
338 | if (unlikely(!skb)) | |
339 | goto err; | |
340 | ||
341 | return skb; | |
342 | ||
343 | err: | |
344 | dev->stats.rx_dropped++; | |
345 | give_pages(rq, page); | |
346 | return NULL; | |
347 | } | |
348 | ||
8fc3b9e9 | 349 | static struct sk_buff *receive_mergeable(struct net_device *dev, |
fdd819b2 | 350 | struct virtnet_info *vi, |
8fc3b9e9 | 351 | struct receive_queue *rq, |
ab7db917 | 352 | unsigned long ctx, |
8fc3b9e9 | 353 | unsigned int len) |
9ab86bbc | 354 | { |
ab7db917 | 355 | void *buf = mergeable_ctx_to_buf_address(ctx); |
8fc3b9e9 | 356 | struct skb_vnet_hdr *hdr = buf; |
fdd819b2 | 357 | u16 num_buf = virtio16_to_cpu(rq->vq->vdev, hdr->mhdr.num_buffers); |
8fc3b9e9 MT |
358 | struct page *page = virt_to_head_page(buf); |
359 | int offset = buf - page_address(page); | |
ab7db917 MD |
360 | unsigned int truesize = max(len, mergeable_ctx_to_buf_truesize(ctx)); |
361 | ||
fb51879d | 362 | struct sk_buff *head_skb = page_to_skb(rq, page, offset, len, truesize); |
2613af0e | 363 | struct sk_buff *curr_skb = head_skb; |
9ab86bbc | 364 | |
8fc3b9e9 MT |
365 | if (unlikely(!curr_skb)) |
366 | goto err_skb; | |
9ab86bbc | 367 | while (--num_buf) { |
8fc3b9e9 MT |
368 | int num_skb_frags; |
369 | ||
ab7db917 MD |
370 | ctx = (unsigned long)virtqueue_get_buf(rq->vq, &len); |
371 | if (unlikely(!ctx)) { | |
8fc3b9e9 | 372 | pr_debug("%s: rx error: %d buffers out of %d missing\n", |
fdd819b2 MT |
373 | dev->name, num_buf, |
374 | virtio16_to_cpu(rq->vq->vdev, | |
375 | hdr->mhdr.num_buffers)); | |
8fc3b9e9 MT |
376 | dev->stats.rx_length_errors++; |
377 | goto err_buf; | |
3f2c31d9 | 378 | } |
8fc3b9e9 | 379 | |
ab7db917 | 380 | buf = mergeable_ctx_to_buf_address(ctx); |
8fc3b9e9 | 381 | page = virt_to_head_page(buf); |
8fc3b9e9 MT |
382 | |
383 | num_skb_frags = skb_shinfo(curr_skb)->nr_frags; | |
2613af0e MD |
384 | if (unlikely(num_skb_frags == MAX_SKB_FRAGS)) { |
385 | struct sk_buff *nskb = alloc_skb(0, GFP_ATOMIC); | |
8fc3b9e9 MT |
386 | |
387 | if (unlikely(!nskb)) | |
388 | goto err_skb; | |
2613af0e MD |
389 | if (curr_skb == head_skb) |
390 | skb_shinfo(curr_skb)->frag_list = nskb; | |
391 | else | |
392 | curr_skb->next = nskb; | |
393 | curr_skb = nskb; | |
394 | head_skb->truesize += nskb->truesize; | |
395 | num_skb_frags = 0; | |
396 | } | |
ab7db917 | 397 | truesize = max(len, mergeable_ctx_to_buf_truesize(ctx)); |
2613af0e MD |
398 | if (curr_skb != head_skb) { |
399 | head_skb->data_len += len; | |
400 | head_skb->len += len; | |
fb51879d | 401 | head_skb->truesize += truesize; |
2613af0e | 402 | } |
8fc3b9e9 | 403 | offset = buf - page_address(page); |
ba275241 JW |
404 | if (skb_can_coalesce(curr_skb, num_skb_frags, page, offset)) { |
405 | put_page(page); | |
406 | skb_coalesce_rx_frag(curr_skb, num_skb_frags - 1, | |
fb51879d | 407 | len, truesize); |
ba275241 JW |
408 | } else { |
409 | skb_add_rx_frag(curr_skb, num_skb_frags, page, | |
fb51879d | 410 | offset, len, truesize); |
ba275241 | 411 | } |
8fc3b9e9 MT |
412 | } |
413 | ||
ab7db917 | 414 | ewma_add(&rq->mrg_avg_pkt_len, head_skb->len); |
8fc3b9e9 MT |
415 | return head_skb; |
416 | ||
417 | err_skb: | |
418 | put_page(page); | |
419 | while (--num_buf) { | |
ab7db917 MD |
420 | ctx = (unsigned long)virtqueue_get_buf(rq->vq, &len); |
421 | if (unlikely(!ctx)) { | |
8fc3b9e9 MT |
422 | pr_debug("%s: rx error: %d buffers missing\n", |
423 | dev->name, num_buf); | |
424 | dev->stats.rx_length_errors++; | |
425 | break; | |
426 | } | |
ab7db917 | 427 | page = virt_to_head_page(mergeable_ctx_to_buf_address(ctx)); |
8fc3b9e9 | 428 | put_page(page); |
9ab86bbc | 429 | } |
8fc3b9e9 MT |
430 | err_buf: |
431 | dev->stats.rx_dropped++; | |
432 | dev_kfree_skb(head_skb); | |
433 | return NULL; | |
9ab86bbc SM |
434 | } |
435 | ||
e9d7417b | 436 | static void receive_buf(struct receive_queue *rq, void *buf, unsigned int len) |
9ab86bbc | 437 | { |
e9d7417b JW |
438 | struct virtnet_info *vi = rq->vq->vdev->priv; |
439 | struct net_device *dev = vi->dev; | |
58472a76 | 440 | struct virtnet_stats *stats = this_cpu_ptr(vi->stats); |
9ab86bbc | 441 | struct sk_buff *skb; |
9ab86bbc | 442 | struct skb_vnet_hdr *hdr; |
3f2c31d9 | 443 | |
9ab86bbc SM |
444 | if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) { |
445 | pr_debug("%s: short packet %i\n", dev->name, len); | |
446 | dev->stats.rx_length_errors++; | |
ab7db917 MD |
447 | if (vi->mergeable_rx_bufs) { |
448 | unsigned long ctx = (unsigned long)buf; | |
449 | void *base = mergeable_ctx_to_buf_address(ctx); | |
450 | put_page(virt_to_head_page(base)); | |
451 | } else if (vi->big_packets) { | |
98bfd23c | 452 | give_pages(rq, buf); |
ab7db917 | 453 | } else { |
9ab86bbc | 454 | dev_kfree_skb(buf); |
ab7db917 | 455 | } |
9ab86bbc SM |
456 | return; |
457 | } | |
3f2c31d9 | 458 | |
f121159d | 459 | if (vi->mergeable_rx_bufs) |
fdd819b2 | 460 | skb = receive_mergeable(dev, vi, rq, (unsigned long)buf, len); |
f121159d MT |
461 | else if (vi->big_packets) |
462 | skb = receive_big(dev, rq, buf, len); | |
463 | else | |
464 | skb = receive_small(buf, len); | |
465 | ||
466 | if (unlikely(!skb)) | |
467 | return; | |
3f2c31d9 | 468 | |
9ab86bbc | 469 | hdr = skb_vnet_hdr(skb); |
3fa2a1df | 470 | |
83a27052 | 471 | u64_stats_update_begin(&stats->rx_syncp); |
3fa2a1df | 472 | stats->rx_bytes += skb->len; |
473 | stats->rx_packets++; | |
83a27052 | 474 | u64_stats_update_end(&stats->rx_syncp); |
296f96fc | 475 | |
b3f24698 | 476 | if (hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { |
296f96fc | 477 | pr_debug("Needs csum!\n"); |
b3f24698 | 478 | if (!skb_partial_csum_set(skb, |
fdd819b2 MT |
479 | virtio16_to_cpu(vi->vdev, hdr->hdr.csum_start), |
480 | virtio16_to_cpu(vi->vdev, hdr->hdr.csum_offset))) | |
296f96fc | 481 | goto frame_err; |
10a8d94a JW |
482 | } else if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) { |
483 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
296f96fc RR |
484 | } |
485 | ||
23cde76d MM |
486 | skb->protocol = eth_type_trans(skb, dev); |
487 | pr_debug("Receiving skb proto 0x%04x len %i type %i\n", | |
488 | ntohs(skb->protocol), skb->len, skb->pkt_type); | |
489 | ||
b3f24698 | 490 | if (hdr->hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) { |
296f96fc | 491 | pr_debug("GSO!\n"); |
b3f24698 | 492 | switch (hdr->hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { |
296f96fc | 493 | case VIRTIO_NET_HDR_GSO_TCPV4: |
c9af6db4 | 494 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; |
296f96fc | 495 | break; |
296f96fc | 496 | case VIRTIO_NET_HDR_GSO_UDP: |
3d0ad094 BH |
497 | { |
498 | static bool warned; | |
499 | ||
500 | if (!warned) { | |
501 | warned = true; | |
502 | netdev_warn(dev, | |
503 | "host using disabled UFO feature; please fix it\n"); | |
504 | } | |
c9af6db4 | 505 | skb_shinfo(skb)->gso_type = SKB_GSO_UDP; |
296f96fc | 506 | break; |
3d0ad094 | 507 | } |
296f96fc | 508 | case VIRTIO_NET_HDR_GSO_TCPV6: |
c9af6db4 | 509 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6; |
296f96fc RR |
510 | break; |
511 | default: | |
be443899 AW |
512 | net_warn_ratelimited("%s: bad gso type %u.\n", |
513 | dev->name, hdr->hdr.gso_type); | |
296f96fc RR |
514 | goto frame_err; |
515 | } | |
516 | ||
b3f24698 | 517 | if (hdr->hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN) |
c9af6db4 | 518 | skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN; |
34a48579 | 519 | |
fdd819b2 MT |
520 | skb_shinfo(skb)->gso_size = virtio16_to_cpu(vi->vdev, |
521 | hdr->hdr.gso_size); | |
296f96fc | 522 | if (skb_shinfo(skb)->gso_size == 0) { |
be443899 | 523 | net_warn_ratelimited("%s: zero gso size.\n", dev->name); |
296f96fc RR |
524 | goto frame_err; |
525 | } | |
526 | ||
527 | /* Header must be checked, and gso_segs computed. */ | |
528 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
529 | skb_shinfo(skb)->gso_segs = 0; | |
530 | } | |
531 | ||
91815639 JW |
532 | skb_mark_napi_id(skb, &rq->napi); |
533 | ||
296f96fc RR |
534 | netif_receive_skb(skb); |
535 | return; | |
536 | ||
537 | frame_err: | |
538 | dev->stats.rx_frame_errors++; | |
296f96fc RR |
539 | dev_kfree_skb(skb); |
540 | } | |
541 | ||
e9d7417b | 542 | static int add_recvbuf_small(struct receive_queue *rq, gfp_t gfp) |
296f96fc | 543 | { |
e9d7417b | 544 | struct virtnet_info *vi = rq->vq->vdev->priv; |
296f96fc | 545 | struct sk_buff *skb; |
9ab86bbc | 546 | struct skb_vnet_hdr *hdr; |
9ab86bbc | 547 | int err; |
3f2c31d9 | 548 | |
5061de36 | 549 | skb = __netdev_alloc_skb_ip_align(vi->dev, GOOD_PACKET_LEN, gfp); |
9ab86bbc SM |
550 | if (unlikely(!skb)) |
551 | return -ENOMEM; | |
296f96fc | 552 | |
5061de36 | 553 | skb_put(skb, GOOD_PACKET_LEN); |
3f2c31d9 | 554 | |
9ab86bbc | 555 | hdr = skb_vnet_hdr(skb); |
a5835440 | 556 | sg_init_table(rq->sg, MAX_SKB_FRAGS + 2); |
e9d7417b | 557 | sg_set_buf(rq->sg, &hdr->hdr, sizeof hdr->hdr); |
e9d7417b | 558 | skb_to_sgvec(skb, rq->sg + 1, 0, skb->len); |
97402b96 | 559 | |
9dc7b9e4 | 560 | err = virtqueue_add_inbuf(rq->vq, rq->sg, 2, skb, gfp); |
9ab86bbc SM |
561 | if (err < 0) |
562 | dev_kfree_skb(skb); | |
97402b96 | 563 | |
9ab86bbc SM |
564 | return err; |
565 | } | |
97402b96 | 566 | |
e9d7417b | 567 | static int add_recvbuf_big(struct receive_queue *rq, gfp_t gfp) |
9ab86bbc | 568 | { |
9ab86bbc SM |
569 | struct page *first, *list = NULL; |
570 | char *p; | |
571 | int i, err, offset; | |
572 | ||
a5835440 RR |
573 | sg_init_table(rq->sg, MAX_SKB_FRAGS + 2); |
574 | ||
e9d7417b | 575 | /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */ |
9ab86bbc | 576 | for (i = MAX_SKB_FRAGS + 1; i > 1; --i) { |
e9d7417b | 577 | first = get_a_page(rq, gfp); |
9ab86bbc SM |
578 | if (!first) { |
579 | if (list) | |
e9d7417b | 580 | give_pages(rq, list); |
9ab86bbc | 581 | return -ENOMEM; |
97402b96 | 582 | } |
e9d7417b | 583 | sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE); |
97402b96 | 584 | |
9ab86bbc SM |
585 | /* chain new page in list head to match sg */ |
586 | first->private = (unsigned long)list; | |
587 | list = first; | |
588 | } | |
296f96fc | 589 | |
e9d7417b | 590 | first = get_a_page(rq, gfp); |
9ab86bbc | 591 | if (!first) { |
e9d7417b | 592 | give_pages(rq, list); |
9ab86bbc SM |
593 | return -ENOMEM; |
594 | } | |
595 | p = page_address(first); | |
596 | ||
e9d7417b JW |
597 | /* rq->sg[0], rq->sg[1] share the same page */ |
598 | /* a separated rq->sg[0] for virtio_net_hdr only due to QEMU bug */ | |
599 | sg_set_buf(&rq->sg[0], p, sizeof(struct virtio_net_hdr)); | |
9ab86bbc | 600 | |
e9d7417b | 601 | /* rq->sg[1] for data packet, from offset */ |
9ab86bbc | 602 | offset = sizeof(struct padded_vnet_hdr); |
e9d7417b | 603 | sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset); |
9ab86bbc SM |
604 | |
605 | /* chain first in list head */ | |
606 | first->private = (unsigned long)list; | |
9dc7b9e4 RR |
607 | err = virtqueue_add_inbuf(rq->vq, rq->sg, MAX_SKB_FRAGS + 2, |
608 | first, gfp); | |
9ab86bbc | 609 | if (err < 0) |
e9d7417b | 610 | give_pages(rq, first); |
9ab86bbc SM |
611 | |
612 | return err; | |
296f96fc RR |
613 | } |
614 | ||
fbf28d78 | 615 | static unsigned int get_mergeable_buf_len(struct ewma *avg_pkt_len) |
3f2c31d9 | 616 | { |
ab7db917 | 617 | const size_t hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf); |
fbf28d78 MD |
618 | unsigned int len; |
619 | ||
620 | len = hdr_len + clamp_t(unsigned int, ewma_read(avg_pkt_len), | |
621 | GOOD_PACKET_LEN, PAGE_SIZE - hdr_len); | |
622 | return ALIGN(len, MERGEABLE_BUFFER_ALIGN); | |
623 | } | |
624 | ||
625 | static int add_recvbuf_mergeable(struct receive_queue *rq, gfp_t gfp) | |
626 | { | |
fb51879d MD |
627 | struct page_frag *alloc_frag = &rq->alloc_frag; |
628 | char *buf; | |
ab7db917 | 629 | unsigned long ctx; |
3f2c31d9 | 630 | int err; |
fb51879d | 631 | unsigned int len, hole; |
3f2c31d9 | 632 | |
fbf28d78 | 633 | len = get_mergeable_buf_len(&rq->mrg_avg_pkt_len); |
ab7db917 | 634 | if (unlikely(!skb_page_frag_refill(len, alloc_frag, gfp))) |
9ab86bbc | 635 | return -ENOMEM; |
ab7db917 | 636 | |
fb51879d | 637 | buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset; |
ab7db917 | 638 | ctx = mergeable_buf_to_ctx(buf, len); |
fb51879d | 639 | get_page(alloc_frag->page); |
fb51879d MD |
640 | alloc_frag->offset += len; |
641 | hole = alloc_frag->size - alloc_frag->offset; | |
ab7db917 MD |
642 | if (hole < len) { |
643 | /* To avoid internal fragmentation, if there is very likely not | |
644 | * enough space for another buffer, add the remaining space to | |
645 | * the current buffer. This extra space is not included in | |
646 | * the truesize stored in ctx. | |
647 | */ | |
fb51879d MD |
648 | len += hole; |
649 | alloc_frag->offset += hole; | |
650 | } | |
3f2c31d9 | 651 | |
fb51879d | 652 | sg_init_one(rq->sg, buf, len); |
ab7db917 | 653 | err = virtqueue_add_inbuf(rq->vq, rq->sg, 1, (void *)ctx, gfp); |
9ab86bbc | 654 | if (err < 0) |
2613af0e | 655 | put_page(virt_to_head_page(buf)); |
3f2c31d9 | 656 | |
9ab86bbc SM |
657 | return err; |
658 | } | |
3f2c31d9 | 659 | |
b2baed69 RR |
660 | /* |
661 | * Returns false if we couldn't fill entirely (OOM). | |
662 | * | |
663 | * Normally run in the receive path, but can also be run from ndo_open | |
664 | * before we're receiving packets, or from refill_work which is | |
665 | * careful to disable receiving (using napi_disable). | |
666 | */ | |
e9d7417b | 667 | static bool try_fill_recv(struct receive_queue *rq, gfp_t gfp) |
9ab86bbc | 668 | { |
e9d7417b | 669 | struct virtnet_info *vi = rq->vq->vdev->priv; |
9ab86bbc | 670 | int err; |
1788f495 | 671 | bool oom; |
3f2c31d9 | 672 | |
fb51879d | 673 | gfp |= __GFP_COLD; |
9ab86bbc SM |
674 | do { |
675 | if (vi->mergeable_rx_bufs) | |
e9d7417b | 676 | err = add_recvbuf_mergeable(rq, gfp); |
9ab86bbc | 677 | else if (vi->big_packets) |
e9d7417b | 678 | err = add_recvbuf_big(rq, gfp); |
9ab86bbc | 679 | else |
e9d7417b | 680 | err = add_recvbuf_small(rq, gfp); |
3f2c31d9 | 681 | |
1788f495 | 682 | oom = err == -ENOMEM; |
9ed4cb07 | 683 | if (err) |
3f2c31d9 | 684 | break; |
b7dfde95 | 685 | } while (rq->vq->num_free); |
681daee2 | 686 | virtqueue_kick(rq->vq); |
3161e453 | 687 | return !oom; |
3f2c31d9 MM |
688 | } |
689 | ||
18445c4d | 690 | static void skb_recv_done(struct virtqueue *rvq) |
296f96fc RR |
691 | { |
692 | struct virtnet_info *vi = rvq->vdev->priv; | |
986a4f4d | 693 | struct receive_queue *rq = &vi->rq[vq2rxq(rvq)]; |
e9d7417b | 694 | |
18445c4d | 695 | /* Schedule NAPI, Suppress further interrupts if successful. */ |
e9d7417b | 696 | if (napi_schedule_prep(&rq->napi)) { |
1915a712 | 697 | virtqueue_disable_cb(rvq); |
e9d7417b | 698 | __napi_schedule(&rq->napi); |
18445c4d | 699 | } |
296f96fc RR |
700 | } |
701 | ||
e9d7417b | 702 | static void virtnet_napi_enable(struct receive_queue *rq) |
3e9d08ec | 703 | { |
e9d7417b | 704 | napi_enable(&rq->napi); |
3e9d08ec BR |
705 | |
706 | /* If all buffers were filled by other side before we napi_enabled, we | |
707 | * won't get another interrupt, so process any outstanding packets | |
708 | * now. virtnet_poll wants re-enable the queue, so we disable here. | |
709 | * We synchronize against interrupts via NAPI_STATE_SCHED */ | |
e9d7417b JW |
710 | if (napi_schedule_prep(&rq->napi)) { |
711 | virtqueue_disable_cb(rq->vq); | |
ec13ee80 | 712 | local_bh_disable(); |
e9d7417b | 713 | __napi_schedule(&rq->napi); |
ec13ee80 | 714 | local_bh_enable(); |
3e9d08ec BR |
715 | } |
716 | } | |
717 | ||
3161e453 RR |
718 | static void refill_work(struct work_struct *work) |
719 | { | |
e9d7417b JW |
720 | struct virtnet_info *vi = |
721 | container_of(work, struct virtnet_info, refill.work); | |
3161e453 | 722 | bool still_empty; |
986a4f4d JW |
723 | int i; |
724 | ||
55257d72 | 725 | for (i = 0; i < vi->curr_queue_pairs; i++) { |
986a4f4d | 726 | struct receive_queue *rq = &vi->rq[i]; |
3161e453 | 727 | |
986a4f4d JW |
728 | napi_disable(&rq->napi); |
729 | still_empty = !try_fill_recv(rq, GFP_KERNEL); | |
730 | virtnet_napi_enable(rq); | |
3161e453 | 731 | |
986a4f4d JW |
732 | /* In theory, this can happen: if we don't get any buffers in |
733 | * we will *never* try to fill again. | |
734 | */ | |
735 | if (still_empty) | |
736 | schedule_delayed_work(&vi->refill, HZ/2); | |
737 | } | |
3161e453 RR |
738 | } |
739 | ||
2ffa7598 | 740 | static int virtnet_receive(struct receive_queue *rq, int budget) |
296f96fc | 741 | { |
e9d7417b | 742 | struct virtnet_info *vi = rq->vq->vdev->priv; |
2ffa7598 | 743 | unsigned int len, received = 0; |
9ab86bbc | 744 | void *buf; |
296f96fc | 745 | |
296f96fc | 746 | while (received < budget && |
e9d7417b JW |
747 | (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) { |
748 | receive_buf(rq, buf, len); | |
296f96fc RR |
749 | received++; |
750 | } | |
751 | ||
be121f46 | 752 | if (rq->vq->num_free > virtqueue_get_vring_size(rq->vq) / 2) { |
e9d7417b | 753 | if (!try_fill_recv(rq, GFP_ATOMIC)) |
3b07e9ca | 754 | schedule_delayed_work(&vi->refill, 0); |
3161e453 | 755 | } |
296f96fc | 756 | |
2ffa7598 JW |
757 | return received; |
758 | } | |
759 | ||
760 | static int virtnet_poll(struct napi_struct *napi, int budget) | |
761 | { | |
762 | struct receive_queue *rq = | |
763 | container_of(napi, struct receive_queue, napi); | |
764 | unsigned int r, received = 0; | |
765 | ||
766 | again: | |
767 | received += virtnet_receive(rq, budget - received); | |
768 | ||
8329d98e RR |
769 | /* Out of packets? */ |
770 | if (received < budget) { | |
cbdadbbf | 771 | r = virtqueue_enable_cb_prepare(rq->vq); |
288379f0 | 772 | napi_complete(napi); |
cbdadbbf | 773 | if (unlikely(virtqueue_poll(rq->vq, r)) && |
8e95a202 | 774 | napi_schedule_prep(napi)) { |
e9d7417b | 775 | virtqueue_disable_cb(rq->vq); |
288379f0 | 776 | __napi_schedule(napi); |
296f96fc | 777 | goto again; |
4265f161 | 778 | } |
296f96fc RR |
779 | } |
780 | ||
781 | return received; | |
782 | } | |
783 | ||
91815639 JW |
784 | #ifdef CONFIG_NET_RX_BUSY_POLL |
785 | /* must be called with local_bh_disable()d */ | |
786 | static int virtnet_busy_poll(struct napi_struct *napi) | |
787 | { | |
788 | struct receive_queue *rq = | |
789 | container_of(napi, struct receive_queue, napi); | |
790 | struct virtnet_info *vi = rq->vq->vdev->priv; | |
791 | int r, received = 0, budget = 4; | |
792 | ||
793 | if (!(vi->status & VIRTIO_NET_S_LINK_UP)) | |
794 | return LL_FLUSH_FAILED; | |
795 | ||
796 | if (!napi_schedule_prep(napi)) | |
797 | return LL_FLUSH_BUSY; | |
798 | ||
799 | virtqueue_disable_cb(rq->vq); | |
800 | ||
801 | again: | |
802 | received += virtnet_receive(rq, budget); | |
803 | ||
804 | r = virtqueue_enable_cb_prepare(rq->vq); | |
805 | clear_bit(NAPI_STATE_SCHED, &napi->state); | |
806 | if (unlikely(virtqueue_poll(rq->vq, r)) && | |
807 | napi_schedule_prep(napi)) { | |
808 | virtqueue_disable_cb(rq->vq); | |
809 | if (received < budget) { | |
810 | budget -= received; | |
811 | goto again; | |
812 | } else { | |
813 | __napi_schedule(napi); | |
814 | } | |
815 | } | |
816 | ||
817 | return received; | |
818 | } | |
819 | #endif /* CONFIG_NET_RX_BUSY_POLL */ | |
820 | ||
986a4f4d JW |
821 | static int virtnet_open(struct net_device *dev) |
822 | { | |
823 | struct virtnet_info *vi = netdev_priv(dev); | |
824 | int i; | |
825 | ||
e4166625 JW |
826 | for (i = 0; i < vi->max_queue_pairs; i++) { |
827 | if (i < vi->curr_queue_pairs) | |
828 | /* Make sure we have some buffers: if oom use wq. */ | |
829 | if (!try_fill_recv(&vi->rq[i], GFP_KERNEL)) | |
830 | schedule_delayed_work(&vi->refill, 0); | |
986a4f4d JW |
831 | virtnet_napi_enable(&vi->rq[i]); |
832 | } | |
833 | ||
834 | return 0; | |
835 | } | |
836 | ||
b7dfde95 | 837 | static void free_old_xmit_skbs(struct send_queue *sq) |
296f96fc RR |
838 | { |
839 | struct sk_buff *skb; | |
6ee57bcc | 840 | unsigned int len; |
e9d7417b | 841 | struct virtnet_info *vi = sq->vq->vdev->priv; |
58472a76 | 842 | struct virtnet_stats *stats = this_cpu_ptr(vi->stats); |
296f96fc | 843 | |
e9d7417b | 844 | while ((skb = virtqueue_get_buf(sq->vq, &len)) != NULL) { |
296f96fc | 845 | pr_debug("Sent skb %p\n", skb); |
3fa2a1df | 846 | |
83a27052 | 847 | u64_stats_update_begin(&stats->tx_syncp); |
3fa2a1df | 848 | stats->tx_bytes += skb->len; |
849 | stats->tx_packets++; | |
83a27052 | 850 | u64_stats_update_end(&stats->tx_syncp); |
3fa2a1df | 851 | |
ed79bab8 | 852 | dev_kfree_skb_any(skb); |
296f96fc RR |
853 | } |
854 | } | |
855 | ||
e9d7417b | 856 | static int xmit_skb(struct send_queue *sq, struct sk_buff *skb) |
296f96fc | 857 | { |
e7428e95 | 858 | struct skb_vnet_hdr *hdr; |
296f96fc | 859 | const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest; |
e9d7417b | 860 | struct virtnet_info *vi = sq->vq->vdev->priv; |
7bedc7dc | 861 | unsigned num_sg; |
e7428e95 MT |
862 | unsigned hdr_len; |
863 | bool can_push; | |
296f96fc | 864 | |
e174961c | 865 | pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest); |
e7428e95 MT |
866 | if (vi->mergeable_rx_bufs) |
867 | hdr_len = sizeof hdr->mhdr; | |
868 | else | |
869 | hdr_len = sizeof hdr->hdr; | |
870 | ||
871 | can_push = vi->any_header_sg && | |
872 | !((unsigned long)skb->data & (__alignof__(*hdr) - 1)) && | |
873 | !skb_header_cloned(skb) && skb_headroom(skb) >= hdr_len; | |
874 | /* Even if we can, don't push here yet as this would skew | |
875 | * csum_start offset below. */ | |
876 | if (can_push) | |
877 | hdr = (struct skb_vnet_hdr *)(skb->data - hdr_len); | |
878 | else | |
879 | hdr = skb_vnet_hdr(skb); | |
296f96fc | 880 | |
296f96fc | 881 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
b3f24698 | 882 | hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; |
fdd819b2 MT |
883 | hdr->hdr.csum_start = cpu_to_virtio16(vi->vdev, |
884 | skb_checksum_start_offset(skb)); | |
885 | hdr->hdr.csum_offset = cpu_to_virtio16(vi->vdev, | |
886 | skb->csum_offset); | |
296f96fc | 887 | } else { |
b3f24698 RR |
888 | hdr->hdr.flags = 0; |
889 | hdr->hdr.csum_offset = hdr->hdr.csum_start = 0; | |
296f96fc RR |
890 | } |
891 | ||
892 | if (skb_is_gso(skb)) { | |
fdd819b2 MT |
893 | hdr->hdr.hdr_len = cpu_to_virtio16(vi->vdev, skb_headlen(skb)); |
894 | hdr->hdr.gso_size = cpu_to_virtio16(vi->vdev, | |
895 | skb_shinfo(skb)->gso_size); | |
34a48579 | 896 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) |
b3f24698 | 897 | hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4; |
296f96fc | 898 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) |
b3f24698 | 899 | hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6; |
296f96fc RR |
900 | else |
901 | BUG(); | |
34a48579 | 902 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN) |
b3f24698 | 903 | hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN; |
296f96fc | 904 | } else { |
b3f24698 RR |
905 | hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE; |
906 | hdr->hdr.gso_size = hdr->hdr.hdr_len = 0; | |
296f96fc RR |
907 | } |
908 | ||
3f2c31d9 | 909 | if (vi->mergeable_rx_bufs) |
e7428e95 | 910 | hdr->mhdr.num_buffers = 0; |
3f2c31d9 | 911 | |
a5835440 | 912 | sg_init_table(sq->sg, MAX_SKB_FRAGS + 2); |
e7428e95 MT |
913 | if (can_push) { |
914 | __skb_push(skb, hdr_len); | |
915 | num_sg = skb_to_sgvec(skb, sq->sg, 0, skb->len); | |
916 | /* Pull header back to avoid skew in tx bytes calculations. */ | |
917 | __skb_pull(skb, hdr_len); | |
918 | } else { | |
919 | sg_set_buf(sq->sg, hdr, hdr_len); | |
920 | num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len) + 1; | |
921 | } | |
9dc7b9e4 | 922 | return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC); |
11a3a154 RR |
923 | } |
924 | ||
424efe9c | 925 | static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev) |
99ffc696 RR |
926 | { |
927 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d JW |
928 | int qnum = skb_get_queue_mapping(skb); |
929 | struct send_queue *sq = &vi->sq[qnum]; | |
9ed4cb07 | 930 | int err; |
4b7fd2e6 MT |
931 | struct netdev_queue *txq = netdev_get_tx_queue(dev, qnum); |
932 | bool kick = !skb->xmit_more; | |
2cb9c6ba | 933 | |
2cb9c6ba | 934 | /* Free up any pending old buffers before queueing new ones. */ |
e9d7417b | 935 | free_old_xmit_skbs(sq); |
99ffc696 | 936 | |
03f191ba | 937 | /* Try to transmit */ |
b7dfde95 | 938 | err = xmit_skb(sq, skb); |
48925e37 | 939 | |
9ed4cb07 | 940 | /* This should not happen! */ |
681daee2 | 941 | if (unlikely(err)) { |
9ed4cb07 RR |
942 | dev->stats.tx_fifo_errors++; |
943 | if (net_ratelimit()) | |
944 | dev_warn(&dev->dev, | |
b7dfde95 | 945 | "Unexpected TXQ (%d) queue failure: %d\n", qnum, err); |
58eba97d | 946 | dev->stats.tx_dropped++; |
85e94525 | 947 | dev_kfree_skb_any(skb); |
58eba97d | 948 | return NETDEV_TX_OK; |
296f96fc | 949 | } |
03f191ba | 950 | |
48925e37 RR |
951 | /* Don't wait up for transmitted skbs to be freed. */ |
952 | skb_orphan(skb); | |
953 | nf_reset(skb); | |
954 | ||
955 | /* Apparently nice girls don't return TX_BUSY; stop the queue | |
956 | * before it gets out of hand. Naturally, this wastes entries. */ | |
b7dfde95 | 957 | if (sq->vq->num_free < 2+MAX_SKB_FRAGS) { |
986a4f4d | 958 | netif_stop_subqueue(dev, qnum); |
e9d7417b | 959 | if (unlikely(!virtqueue_enable_cb_delayed(sq->vq))) { |
48925e37 | 960 | /* More just got used, free them then recheck. */ |
b7dfde95 LT |
961 | free_old_xmit_skbs(sq); |
962 | if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) { | |
986a4f4d | 963 | netif_start_subqueue(dev, qnum); |
e9d7417b | 964 | virtqueue_disable_cb(sq->vq); |
48925e37 RR |
965 | } |
966 | } | |
99ffc696 | 967 | } |
48925e37 | 968 | |
4b7fd2e6 | 969 | if (kick || netif_xmit_stopped(txq)) |
0b725a2c | 970 | virtqueue_kick(sq->vq); |
296f96fc | 971 | |
0b725a2c | 972 | return NETDEV_TX_OK; |
c223a078 DM |
973 | } |
974 | ||
40cbfc37 AK |
975 | /* |
976 | * Send command via the control virtqueue and check status. Commands | |
977 | * supported by the hypervisor, as indicated by feature bits, should | |
788a8b6d | 978 | * never fail unless improperly formatted. |
40cbfc37 AK |
979 | */ |
980 | static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd, | |
d24bae32 | 981 | struct scatterlist *out) |
40cbfc37 | 982 | { |
f7bc9594 | 983 | struct scatterlist *sgs[4], hdr, stat; |
40cbfc37 AK |
984 | struct virtio_net_ctrl_hdr ctrl; |
985 | virtio_net_ctrl_ack status = ~0; | |
d24bae32 | 986 | unsigned out_num = 0, tmp; |
40cbfc37 AK |
987 | |
988 | /* Caller should know better */ | |
f7bc9594 | 989 | BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)); |
40cbfc37 AK |
990 | |
991 | ctrl.class = class; | |
992 | ctrl.cmd = cmd; | |
f7bc9594 RR |
993 | /* Add header */ |
994 | sg_init_one(&hdr, &ctrl, sizeof(ctrl)); | |
995 | sgs[out_num++] = &hdr; | |
40cbfc37 | 996 | |
f7bc9594 RR |
997 | if (out) |
998 | sgs[out_num++] = out; | |
40cbfc37 | 999 | |
f7bc9594 RR |
1000 | /* Add return status. */ |
1001 | sg_init_one(&stat, &status, sizeof(status)); | |
d24bae32 | 1002 | sgs[out_num] = &stat; |
40cbfc37 | 1003 | |
d24bae32 | 1004 | BUG_ON(out_num + 1 > ARRAY_SIZE(sgs)); |
a7c58146 | 1005 | virtqueue_add_sgs(vi->cvq, sgs, out_num, 1, vi, GFP_ATOMIC); |
40cbfc37 | 1006 | |
67975901 HG |
1007 | if (unlikely(!virtqueue_kick(vi->cvq))) |
1008 | return status == VIRTIO_NET_OK; | |
40cbfc37 AK |
1009 | |
1010 | /* Spin for a response, the kick causes an ioport write, trapping | |
1011 | * into the hypervisor, so the request should be handled immediately. | |
1012 | */ | |
047b9b94 HG |
1013 | while (!virtqueue_get_buf(vi->cvq, &tmp) && |
1014 | !virtqueue_is_broken(vi->cvq)) | |
40cbfc37 AK |
1015 | cpu_relax(); |
1016 | ||
1017 | return status == VIRTIO_NET_OK; | |
1018 | } | |
1019 | ||
9c46f6d4 AW |
1020 | static int virtnet_set_mac_address(struct net_device *dev, void *p) |
1021 | { | |
1022 | struct virtnet_info *vi = netdev_priv(dev); | |
1023 | struct virtio_device *vdev = vi->vdev; | |
f2f2c8b4 | 1024 | int ret; |
7e58d5ae AK |
1025 | struct sockaddr *addr = p; |
1026 | struct scatterlist sg; | |
9c46f6d4 | 1027 | |
7e58d5ae | 1028 | ret = eth_prepare_mac_addr_change(dev, p); |
f2f2c8b4 JP |
1029 | if (ret) |
1030 | return ret; | |
9c46f6d4 | 1031 | |
7e58d5ae AK |
1032 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) { |
1033 | sg_init_one(&sg, addr->sa_data, dev->addr_len); | |
1034 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC, | |
d24bae32 | 1035 | VIRTIO_NET_CTRL_MAC_ADDR_SET, &sg)) { |
7e58d5ae AK |
1036 | dev_warn(&vdev->dev, |
1037 | "Failed to set mac address by vq command.\n"); | |
1038 | return -EINVAL; | |
1039 | } | |
1040 | } else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) { | |
855e0c52 RR |
1041 | unsigned int i; |
1042 | ||
1043 | /* Naturally, this has an atomicity problem. */ | |
1044 | for (i = 0; i < dev->addr_len; i++) | |
1045 | virtio_cwrite8(vdev, | |
1046 | offsetof(struct virtio_net_config, mac) + | |
1047 | i, addr->sa_data[i]); | |
7e58d5ae AK |
1048 | } |
1049 | ||
1050 | eth_commit_mac_addr_change(dev, p); | |
9c46f6d4 AW |
1051 | |
1052 | return 0; | |
1053 | } | |
1054 | ||
3fa2a1df | 1055 | static struct rtnl_link_stats64 *virtnet_stats(struct net_device *dev, |
1056 | struct rtnl_link_stats64 *tot) | |
1057 | { | |
1058 | struct virtnet_info *vi = netdev_priv(dev); | |
1059 | int cpu; | |
1060 | unsigned int start; | |
1061 | ||
1062 | for_each_possible_cpu(cpu) { | |
58472a76 | 1063 | struct virtnet_stats *stats = per_cpu_ptr(vi->stats, cpu); |
3fa2a1df | 1064 | u64 tpackets, tbytes, rpackets, rbytes; |
1065 | ||
1066 | do { | |
57a7744e | 1067 | start = u64_stats_fetch_begin_irq(&stats->tx_syncp); |
3fa2a1df | 1068 | tpackets = stats->tx_packets; |
1069 | tbytes = stats->tx_bytes; | |
57a7744e | 1070 | } while (u64_stats_fetch_retry_irq(&stats->tx_syncp, start)); |
83a27052 ED |
1071 | |
1072 | do { | |
57a7744e | 1073 | start = u64_stats_fetch_begin_irq(&stats->rx_syncp); |
3fa2a1df | 1074 | rpackets = stats->rx_packets; |
1075 | rbytes = stats->rx_bytes; | |
57a7744e | 1076 | } while (u64_stats_fetch_retry_irq(&stats->rx_syncp, start)); |
3fa2a1df | 1077 | |
1078 | tot->rx_packets += rpackets; | |
1079 | tot->tx_packets += tpackets; | |
1080 | tot->rx_bytes += rbytes; | |
1081 | tot->tx_bytes += tbytes; | |
1082 | } | |
1083 | ||
1084 | tot->tx_dropped = dev->stats.tx_dropped; | |
021ac8d3 | 1085 | tot->tx_fifo_errors = dev->stats.tx_fifo_errors; |
3fa2a1df | 1086 | tot->rx_dropped = dev->stats.rx_dropped; |
1087 | tot->rx_length_errors = dev->stats.rx_length_errors; | |
1088 | tot->rx_frame_errors = dev->stats.rx_frame_errors; | |
1089 | ||
1090 | return tot; | |
1091 | } | |
1092 | ||
da74e89d AS |
1093 | #ifdef CONFIG_NET_POLL_CONTROLLER |
1094 | static void virtnet_netpoll(struct net_device *dev) | |
1095 | { | |
1096 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d | 1097 | int i; |
da74e89d | 1098 | |
986a4f4d JW |
1099 | for (i = 0; i < vi->curr_queue_pairs; i++) |
1100 | napi_schedule(&vi->rq[i].napi); | |
da74e89d AS |
1101 | } |
1102 | #endif | |
1103 | ||
586d17c5 JW |
1104 | static void virtnet_ack_link_announce(struct virtnet_info *vi) |
1105 | { | |
1106 | rtnl_lock(); | |
1107 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE, | |
d24bae32 | 1108 | VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL)) |
586d17c5 JW |
1109 | dev_warn(&vi->dev->dev, "Failed to ack link announce.\n"); |
1110 | rtnl_unlock(); | |
1111 | } | |
1112 | ||
986a4f4d JW |
1113 | static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs) |
1114 | { | |
1115 | struct scatterlist sg; | |
1116 | struct virtio_net_ctrl_mq s; | |
1117 | struct net_device *dev = vi->dev; | |
1118 | ||
1119 | if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ)) | |
1120 | return 0; | |
1121 | ||
fdd819b2 | 1122 | s.virtqueue_pairs = cpu_to_virtio16(vi->vdev, queue_pairs); |
986a4f4d JW |
1123 | sg_init_one(&sg, &s, sizeof(s)); |
1124 | ||
1125 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ, | |
d24bae32 | 1126 | VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg)) { |
986a4f4d JW |
1127 | dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n", |
1128 | queue_pairs); | |
1129 | return -EINVAL; | |
55257d72 | 1130 | } else { |
986a4f4d | 1131 | vi->curr_queue_pairs = queue_pairs; |
35ed159b JW |
1132 | /* virtnet_open() will refill when device is going to up. */ |
1133 | if (dev->flags & IFF_UP) | |
1134 | schedule_delayed_work(&vi->refill, 0); | |
55257d72 | 1135 | } |
986a4f4d JW |
1136 | |
1137 | return 0; | |
1138 | } | |
1139 | ||
296f96fc RR |
1140 | static int virtnet_close(struct net_device *dev) |
1141 | { | |
1142 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d | 1143 | int i; |
296f96fc | 1144 | |
b2baed69 RR |
1145 | /* Make sure refill_work doesn't re-enable napi! */ |
1146 | cancel_delayed_work_sync(&vi->refill); | |
986a4f4d JW |
1147 | |
1148 | for (i = 0; i < vi->max_queue_pairs; i++) | |
1149 | napi_disable(&vi->rq[i].napi); | |
296f96fc | 1150 | |
296f96fc RR |
1151 | return 0; |
1152 | } | |
1153 | ||
2af7698e AW |
1154 | static void virtnet_set_rx_mode(struct net_device *dev) |
1155 | { | |
1156 | struct virtnet_info *vi = netdev_priv(dev); | |
f565a7c2 | 1157 | struct scatterlist sg[2]; |
2af7698e | 1158 | u8 promisc, allmulti; |
f565a7c2 | 1159 | struct virtio_net_ctrl_mac *mac_data; |
ccffad25 | 1160 | struct netdev_hw_addr *ha; |
32e7bfc4 | 1161 | int uc_count; |
4cd24eaf | 1162 | int mc_count; |
f565a7c2 AW |
1163 | void *buf; |
1164 | int i; | |
2af7698e | 1165 | |
788a8b6d | 1166 | /* We can't dynamically set ndo_set_rx_mode, so return gracefully */ |
2af7698e AW |
1167 | if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX)) |
1168 | return; | |
1169 | ||
f565a7c2 AW |
1170 | promisc = ((dev->flags & IFF_PROMISC) != 0); |
1171 | allmulti = ((dev->flags & IFF_ALLMULTI) != 0); | |
2af7698e | 1172 | |
23e258e1 | 1173 | sg_init_one(sg, &promisc, sizeof(promisc)); |
2af7698e AW |
1174 | |
1175 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, | |
d24bae32 | 1176 | VIRTIO_NET_CTRL_RX_PROMISC, sg)) |
2af7698e AW |
1177 | dev_warn(&dev->dev, "Failed to %sable promisc mode.\n", |
1178 | promisc ? "en" : "dis"); | |
1179 | ||
23e258e1 | 1180 | sg_init_one(sg, &allmulti, sizeof(allmulti)); |
2af7698e AW |
1181 | |
1182 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, | |
d24bae32 | 1183 | VIRTIO_NET_CTRL_RX_ALLMULTI, sg)) |
2af7698e AW |
1184 | dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n", |
1185 | allmulti ? "en" : "dis"); | |
f565a7c2 | 1186 | |
32e7bfc4 | 1187 | uc_count = netdev_uc_count(dev); |
4cd24eaf | 1188 | mc_count = netdev_mc_count(dev); |
f565a7c2 | 1189 | /* MAC filter - use one buffer for both lists */ |
4cd24eaf JP |
1190 | buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) + |
1191 | (2 * sizeof(mac_data->entries)), GFP_ATOMIC); | |
1192 | mac_data = buf; | |
e68ed8f0 | 1193 | if (!buf) |
f565a7c2 | 1194 | return; |
f565a7c2 | 1195 | |
23e258e1 AW |
1196 | sg_init_table(sg, 2); |
1197 | ||
f565a7c2 | 1198 | /* Store the unicast list and count in the front of the buffer */ |
fdd819b2 | 1199 | mac_data->entries = cpu_to_virtio32(vi->vdev, uc_count); |
ccffad25 | 1200 | i = 0; |
32e7bfc4 | 1201 | netdev_for_each_uc_addr(ha, dev) |
ccffad25 | 1202 | memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); |
f565a7c2 AW |
1203 | |
1204 | sg_set_buf(&sg[0], mac_data, | |
32e7bfc4 | 1205 | sizeof(mac_data->entries) + (uc_count * ETH_ALEN)); |
f565a7c2 AW |
1206 | |
1207 | /* multicast list and count fill the end */ | |
32e7bfc4 | 1208 | mac_data = (void *)&mac_data->macs[uc_count][0]; |
f565a7c2 | 1209 | |
fdd819b2 | 1210 | mac_data->entries = cpu_to_virtio32(vi->vdev, mc_count); |
567ec874 | 1211 | i = 0; |
22bedad3 JP |
1212 | netdev_for_each_mc_addr(ha, dev) |
1213 | memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); | |
f565a7c2 AW |
1214 | |
1215 | sg_set_buf(&sg[1], mac_data, | |
4cd24eaf | 1216 | sizeof(mac_data->entries) + (mc_count * ETH_ALEN)); |
f565a7c2 AW |
1217 | |
1218 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC, | |
d24bae32 | 1219 | VIRTIO_NET_CTRL_MAC_TABLE_SET, sg)) |
99e872ae | 1220 | dev_warn(&dev->dev, "Failed to set MAC filter table.\n"); |
f565a7c2 AW |
1221 | |
1222 | kfree(buf); | |
2af7698e AW |
1223 | } |
1224 | ||
80d5c368 PM |
1225 | static int virtnet_vlan_rx_add_vid(struct net_device *dev, |
1226 | __be16 proto, u16 vid) | |
0bde9569 AW |
1227 | { |
1228 | struct virtnet_info *vi = netdev_priv(dev); | |
1229 | struct scatterlist sg; | |
1230 | ||
23e258e1 | 1231 | sg_init_one(&sg, &vid, sizeof(vid)); |
0bde9569 AW |
1232 | |
1233 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, | |
d24bae32 | 1234 | VIRTIO_NET_CTRL_VLAN_ADD, &sg)) |
0bde9569 | 1235 | dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid); |
8e586137 | 1236 | return 0; |
0bde9569 AW |
1237 | } |
1238 | ||
80d5c368 PM |
1239 | static int virtnet_vlan_rx_kill_vid(struct net_device *dev, |
1240 | __be16 proto, u16 vid) | |
0bde9569 AW |
1241 | { |
1242 | struct virtnet_info *vi = netdev_priv(dev); | |
1243 | struct scatterlist sg; | |
1244 | ||
23e258e1 | 1245 | sg_init_one(&sg, &vid, sizeof(vid)); |
0bde9569 AW |
1246 | |
1247 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, | |
d24bae32 | 1248 | VIRTIO_NET_CTRL_VLAN_DEL, &sg)) |
0bde9569 | 1249 | dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid); |
8e586137 | 1250 | return 0; |
0bde9569 AW |
1251 | } |
1252 | ||
8898c21c | 1253 | static void virtnet_clean_affinity(struct virtnet_info *vi, long hcpu) |
986a4f4d JW |
1254 | { |
1255 | int i; | |
1256 | ||
8898c21c WG |
1257 | if (vi->affinity_hint_set) { |
1258 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
47be2479 WG |
1259 | virtqueue_set_affinity(vi->rq[i].vq, -1); |
1260 | virtqueue_set_affinity(vi->sq[i].vq, -1); | |
1261 | } | |
1262 | ||
8898c21c WG |
1263 | vi->affinity_hint_set = false; |
1264 | } | |
8898c21c | 1265 | } |
47be2479 | 1266 | |
8898c21c WG |
1267 | static void virtnet_set_affinity(struct virtnet_info *vi) |
1268 | { | |
1269 | int i; | |
1270 | int cpu; | |
986a4f4d JW |
1271 | |
1272 | /* In multiqueue mode, when the number of cpu is equal to the number of | |
1273 | * queue pairs, we let the queue pairs to be private to one cpu by | |
1274 | * setting the affinity hint to eliminate the contention. | |
1275 | */ | |
8898c21c WG |
1276 | if (vi->curr_queue_pairs == 1 || |
1277 | vi->max_queue_pairs != num_online_cpus()) { | |
1278 | virtnet_clean_affinity(vi, -1); | |
1279 | return; | |
986a4f4d JW |
1280 | } |
1281 | ||
8898c21c WG |
1282 | i = 0; |
1283 | for_each_online_cpu(cpu) { | |
986a4f4d JW |
1284 | virtqueue_set_affinity(vi->rq[i].vq, cpu); |
1285 | virtqueue_set_affinity(vi->sq[i].vq, cpu); | |
9bb8ca86 | 1286 | netif_set_xps_queue(vi->dev, cpumask_of(cpu), i); |
8898c21c | 1287 | i++; |
986a4f4d JW |
1288 | } |
1289 | ||
8898c21c | 1290 | vi->affinity_hint_set = true; |
986a4f4d JW |
1291 | } |
1292 | ||
8de4b2f3 WG |
1293 | static int virtnet_cpu_callback(struct notifier_block *nfb, |
1294 | unsigned long action, void *hcpu) | |
1295 | { | |
1296 | struct virtnet_info *vi = container_of(nfb, struct virtnet_info, nb); | |
1297 | ||
1298 | switch(action & ~CPU_TASKS_FROZEN) { | |
1299 | case CPU_ONLINE: | |
1300 | case CPU_DOWN_FAILED: | |
1301 | case CPU_DEAD: | |
1302 | virtnet_set_affinity(vi); | |
1303 | break; | |
1304 | case CPU_DOWN_PREPARE: | |
1305 | virtnet_clean_affinity(vi, (long)hcpu); | |
1306 | break; | |
1307 | default: | |
1308 | break; | |
1309 | } | |
3ab098df | 1310 | |
8de4b2f3 | 1311 | return NOTIFY_OK; |
986a4f4d JW |
1312 | } |
1313 | ||
8f9f4668 RJ |
1314 | static void virtnet_get_ringparam(struct net_device *dev, |
1315 | struct ethtool_ringparam *ring) | |
1316 | { | |
1317 | struct virtnet_info *vi = netdev_priv(dev); | |
1318 | ||
986a4f4d JW |
1319 | ring->rx_max_pending = virtqueue_get_vring_size(vi->rq[0].vq); |
1320 | ring->tx_max_pending = virtqueue_get_vring_size(vi->sq[0].vq); | |
8f9f4668 RJ |
1321 | ring->rx_pending = ring->rx_max_pending; |
1322 | ring->tx_pending = ring->tx_max_pending; | |
8f9f4668 RJ |
1323 | } |
1324 | ||
66846048 RJ |
1325 | |
1326 | static void virtnet_get_drvinfo(struct net_device *dev, | |
1327 | struct ethtool_drvinfo *info) | |
1328 | { | |
1329 | struct virtnet_info *vi = netdev_priv(dev); | |
1330 | struct virtio_device *vdev = vi->vdev; | |
1331 | ||
1332 | strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver)); | |
1333 | strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version)); | |
1334 | strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info)); | |
1335 | ||
1336 | } | |
1337 | ||
d73bcd2c JW |
1338 | /* TODO: Eliminate OOO packets during switching */ |
1339 | static int virtnet_set_channels(struct net_device *dev, | |
1340 | struct ethtool_channels *channels) | |
1341 | { | |
1342 | struct virtnet_info *vi = netdev_priv(dev); | |
1343 | u16 queue_pairs = channels->combined_count; | |
1344 | int err; | |
1345 | ||
1346 | /* We don't support separate rx/tx channels. | |
1347 | * We don't allow setting 'other' channels. | |
1348 | */ | |
1349 | if (channels->rx_count || channels->tx_count || channels->other_count) | |
1350 | return -EINVAL; | |
1351 | ||
c18e9cd6 | 1352 | if (queue_pairs > vi->max_queue_pairs || queue_pairs == 0) |
d73bcd2c JW |
1353 | return -EINVAL; |
1354 | ||
47be2479 | 1355 | get_online_cpus(); |
d73bcd2c JW |
1356 | err = virtnet_set_queues(vi, queue_pairs); |
1357 | if (!err) { | |
1358 | netif_set_real_num_tx_queues(dev, queue_pairs); | |
1359 | netif_set_real_num_rx_queues(dev, queue_pairs); | |
1360 | ||
8898c21c | 1361 | virtnet_set_affinity(vi); |
d73bcd2c | 1362 | } |
47be2479 | 1363 | put_online_cpus(); |
d73bcd2c JW |
1364 | |
1365 | return err; | |
1366 | } | |
1367 | ||
1368 | static void virtnet_get_channels(struct net_device *dev, | |
1369 | struct ethtool_channels *channels) | |
1370 | { | |
1371 | struct virtnet_info *vi = netdev_priv(dev); | |
1372 | ||
1373 | channels->combined_count = vi->curr_queue_pairs; | |
1374 | channels->max_combined = vi->max_queue_pairs; | |
1375 | channels->max_other = 0; | |
1376 | channels->rx_count = 0; | |
1377 | channels->tx_count = 0; | |
1378 | channels->other_count = 0; | |
1379 | } | |
1380 | ||
0fc0b732 | 1381 | static const struct ethtool_ops virtnet_ethtool_ops = { |
66846048 | 1382 | .get_drvinfo = virtnet_get_drvinfo, |
9f4d26d0 | 1383 | .get_link = ethtool_op_get_link, |
8f9f4668 | 1384 | .get_ringparam = virtnet_get_ringparam, |
d73bcd2c JW |
1385 | .set_channels = virtnet_set_channels, |
1386 | .get_channels = virtnet_get_channels, | |
a9ea3fc6 HX |
1387 | }; |
1388 | ||
39da5814 MM |
1389 | #define MIN_MTU 68 |
1390 | #define MAX_MTU 65535 | |
1391 | ||
1392 | static int virtnet_change_mtu(struct net_device *dev, int new_mtu) | |
1393 | { | |
1394 | if (new_mtu < MIN_MTU || new_mtu > MAX_MTU) | |
1395 | return -EINVAL; | |
1396 | dev->mtu = new_mtu; | |
1397 | return 0; | |
1398 | } | |
1399 | ||
76288b4e SH |
1400 | static const struct net_device_ops virtnet_netdev = { |
1401 | .ndo_open = virtnet_open, | |
1402 | .ndo_stop = virtnet_close, | |
1403 | .ndo_start_xmit = start_xmit, | |
1404 | .ndo_validate_addr = eth_validate_addr, | |
9c46f6d4 | 1405 | .ndo_set_mac_address = virtnet_set_mac_address, |
2af7698e | 1406 | .ndo_set_rx_mode = virtnet_set_rx_mode, |
76288b4e | 1407 | .ndo_change_mtu = virtnet_change_mtu, |
3fa2a1df | 1408 | .ndo_get_stats64 = virtnet_stats, |
1824a989 AW |
1409 | .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid, |
1410 | .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid, | |
76288b4e SH |
1411 | #ifdef CONFIG_NET_POLL_CONTROLLER |
1412 | .ndo_poll_controller = virtnet_netpoll, | |
1413 | #endif | |
91815639 JW |
1414 | #ifdef CONFIG_NET_RX_BUSY_POLL |
1415 | .ndo_busy_poll = virtnet_busy_poll, | |
1416 | #endif | |
76288b4e SH |
1417 | }; |
1418 | ||
586d17c5 | 1419 | static void virtnet_config_changed_work(struct work_struct *work) |
9f4d26d0 | 1420 | { |
586d17c5 JW |
1421 | struct virtnet_info *vi = |
1422 | container_of(work, struct virtnet_info, config_work); | |
9f4d26d0 MM |
1423 | u16 v; |
1424 | ||
855e0c52 RR |
1425 | if (virtio_cread_feature(vi->vdev, VIRTIO_NET_F_STATUS, |
1426 | struct virtio_net_config, status, &v) < 0) | |
507613bf | 1427 | return; |
586d17c5 JW |
1428 | |
1429 | if (v & VIRTIO_NET_S_ANNOUNCE) { | |
ee89bab1 | 1430 | netdev_notify_peers(vi->dev); |
586d17c5 JW |
1431 | virtnet_ack_link_announce(vi); |
1432 | } | |
9f4d26d0 MM |
1433 | |
1434 | /* Ignore unknown (future) status bits */ | |
1435 | v &= VIRTIO_NET_S_LINK_UP; | |
1436 | ||
1437 | if (vi->status == v) | |
507613bf | 1438 | return; |
9f4d26d0 MM |
1439 | |
1440 | vi->status = v; | |
1441 | ||
1442 | if (vi->status & VIRTIO_NET_S_LINK_UP) { | |
1443 | netif_carrier_on(vi->dev); | |
986a4f4d | 1444 | netif_tx_wake_all_queues(vi->dev); |
9f4d26d0 MM |
1445 | } else { |
1446 | netif_carrier_off(vi->dev); | |
986a4f4d | 1447 | netif_tx_stop_all_queues(vi->dev); |
9f4d26d0 MM |
1448 | } |
1449 | } | |
1450 | ||
1451 | static void virtnet_config_changed(struct virtio_device *vdev) | |
1452 | { | |
1453 | struct virtnet_info *vi = vdev->priv; | |
1454 | ||
3b07e9ca | 1455 | schedule_work(&vi->config_work); |
9f4d26d0 MM |
1456 | } |
1457 | ||
986a4f4d JW |
1458 | static void virtnet_free_queues(struct virtnet_info *vi) |
1459 | { | |
d4fb84ee AV |
1460 | int i; |
1461 | ||
1462 | for (i = 0; i < vi->max_queue_pairs; i++) | |
1463 | netif_napi_del(&vi->rq[i].napi); | |
1464 | ||
986a4f4d JW |
1465 | kfree(vi->rq); |
1466 | kfree(vi->sq); | |
1467 | } | |
1468 | ||
1469 | static void free_receive_bufs(struct virtnet_info *vi) | |
1470 | { | |
1471 | int i; | |
1472 | ||
1473 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
1474 | while (vi->rq[i].pages) | |
1475 | __free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0); | |
1476 | } | |
1477 | } | |
1478 | ||
fb51879d MD |
1479 | static void free_receive_page_frags(struct virtnet_info *vi) |
1480 | { | |
1481 | int i; | |
1482 | for (i = 0; i < vi->max_queue_pairs; i++) | |
1483 | if (vi->rq[i].alloc_frag.page) | |
1484 | put_page(vi->rq[i].alloc_frag.page); | |
1485 | } | |
1486 | ||
986a4f4d JW |
1487 | static void free_unused_bufs(struct virtnet_info *vi) |
1488 | { | |
1489 | void *buf; | |
1490 | int i; | |
1491 | ||
1492 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
1493 | struct virtqueue *vq = vi->sq[i].vq; | |
1494 | while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) | |
1495 | dev_kfree_skb(buf); | |
1496 | } | |
1497 | ||
1498 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
1499 | struct virtqueue *vq = vi->rq[i].vq; | |
1500 | ||
1501 | while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) { | |
ab7db917 MD |
1502 | if (vi->mergeable_rx_bufs) { |
1503 | unsigned long ctx = (unsigned long)buf; | |
1504 | void *base = mergeable_ctx_to_buf_address(ctx); | |
1505 | put_page(virt_to_head_page(base)); | |
1506 | } else if (vi->big_packets) { | |
fa9fac17 | 1507 | give_pages(&vi->rq[i], buf); |
ab7db917 | 1508 | } else { |
986a4f4d | 1509 | dev_kfree_skb(buf); |
ab7db917 | 1510 | } |
986a4f4d | 1511 | } |
986a4f4d JW |
1512 | } |
1513 | } | |
1514 | ||
e9d7417b JW |
1515 | static void virtnet_del_vqs(struct virtnet_info *vi) |
1516 | { | |
1517 | struct virtio_device *vdev = vi->vdev; | |
1518 | ||
8898c21c | 1519 | virtnet_clean_affinity(vi, -1); |
986a4f4d | 1520 | |
e9d7417b | 1521 | vdev->config->del_vqs(vdev); |
986a4f4d JW |
1522 | |
1523 | virtnet_free_queues(vi); | |
e9d7417b JW |
1524 | } |
1525 | ||
986a4f4d | 1526 | static int virtnet_find_vqs(struct virtnet_info *vi) |
3f9c10b0 | 1527 | { |
986a4f4d JW |
1528 | vq_callback_t **callbacks; |
1529 | struct virtqueue **vqs; | |
1530 | int ret = -ENOMEM; | |
1531 | int i, total_vqs; | |
1532 | const char **names; | |
1533 | ||
1534 | /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by | |
1535 | * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by | |
1536 | * possible control vq. | |
1537 | */ | |
1538 | total_vqs = vi->max_queue_pairs * 2 + | |
1539 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ); | |
1540 | ||
1541 | /* Allocate space for find_vqs parameters */ | |
1542 | vqs = kzalloc(total_vqs * sizeof(*vqs), GFP_KERNEL); | |
1543 | if (!vqs) | |
1544 | goto err_vq; | |
1545 | callbacks = kmalloc(total_vqs * sizeof(*callbacks), GFP_KERNEL); | |
1546 | if (!callbacks) | |
1547 | goto err_callback; | |
1548 | names = kmalloc(total_vqs * sizeof(*names), GFP_KERNEL); | |
1549 | if (!names) | |
1550 | goto err_names; | |
1551 | ||
1552 | /* Parameters for control virtqueue, if any */ | |
1553 | if (vi->has_cvq) { | |
1554 | callbacks[total_vqs - 1] = NULL; | |
1555 | names[total_vqs - 1] = "control"; | |
1556 | } | |
3f9c10b0 | 1557 | |
986a4f4d JW |
1558 | /* Allocate/initialize parameters for send/receive virtqueues */ |
1559 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
1560 | callbacks[rxq2vq(i)] = skb_recv_done; | |
1561 | callbacks[txq2vq(i)] = skb_xmit_done; | |
1562 | sprintf(vi->rq[i].name, "input.%d", i); | |
1563 | sprintf(vi->sq[i].name, "output.%d", i); | |
1564 | names[rxq2vq(i)] = vi->rq[i].name; | |
1565 | names[txq2vq(i)] = vi->sq[i].name; | |
1566 | } | |
3f9c10b0 | 1567 | |
986a4f4d JW |
1568 | ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs, callbacks, |
1569 | names); | |
1570 | if (ret) | |
1571 | goto err_find; | |
3f9c10b0 | 1572 | |
986a4f4d JW |
1573 | if (vi->has_cvq) { |
1574 | vi->cvq = vqs[total_vqs - 1]; | |
3f9c10b0 | 1575 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN)) |
f646968f | 1576 | vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER; |
3f9c10b0 | 1577 | } |
986a4f4d JW |
1578 | |
1579 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
1580 | vi->rq[i].vq = vqs[rxq2vq(i)]; | |
1581 | vi->sq[i].vq = vqs[txq2vq(i)]; | |
1582 | } | |
1583 | ||
1584 | kfree(names); | |
1585 | kfree(callbacks); | |
1586 | kfree(vqs); | |
1587 | ||
3f9c10b0 | 1588 | return 0; |
986a4f4d JW |
1589 | |
1590 | err_find: | |
1591 | kfree(names); | |
1592 | err_names: | |
1593 | kfree(callbacks); | |
1594 | err_callback: | |
1595 | kfree(vqs); | |
1596 | err_vq: | |
1597 | return ret; | |
1598 | } | |
1599 | ||
1600 | static int virtnet_alloc_queues(struct virtnet_info *vi) | |
1601 | { | |
1602 | int i; | |
1603 | ||
1604 | vi->sq = kzalloc(sizeof(*vi->sq) * vi->max_queue_pairs, GFP_KERNEL); | |
1605 | if (!vi->sq) | |
1606 | goto err_sq; | |
1607 | vi->rq = kzalloc(sizeof(*vi->rq) * vi->max_queue_pairs, GFP_KERNEL); | |
008d4278 | 1608 | if (!vi->rq) |
986a4f4d JW |
1609 | goto err_rq; |
1610 | ||
1611 | INIT_DELAYED_WORK(&vi->refill, refill_work); | |
1612 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
1613 | vi->rq[i].pages = NULL; | |
1614 | netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll, | |
1615 | napi_weight); | |
91815639 | 1616 | napi_hash_add(&vi->rq[i].napi); |
986a4f4d JW |
1617 | |
1618 | sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg)); | |
ab7db917 | 1619 | ewma_init(&vi->rq[i].mrg_avg_pkt_len, 1, RECEIVE_AVG_WEIGHT); |
986a4f4d JW |
1620 | sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg)); |
1621 | } | |
1622 | ||
1623 | return 0; | |
1624 | ||
1625 | err_rq: | |
1626 | kfree(vi->sq); | |
1627 | err_sq: | |
1628 | return -ENOMEM; | |
1629 | } | |
1630 | ||
1631 | static int init_vqs(struct virtnet_info *vi) | |
1632 | { | |
1633 | int ret; | |
1634 | ||
1635 | /* Allocate send & receive queues */ | |
1636 | ret = virtnet_alloc_queues(vi); | |
1637 | if (ret) | |
1638 | goto err; | |
1639 | ||
1640 | ret = virtnet_find_vqs(vi); | |
1641 | if (ret) | |
1642 | goto err_free; | |
1643 | ||
47be2479 | 1644 | get_online_cpus(); |
8898c21c | 1645 | virtnet_set_affinity(vi); |
47be2479 WG |
1646 | put_online_cpus(); |
1647 | ||
986a4f4d JW |
1648 | return 0; |
1649 | ||
1650 | err_free: | |
1651 | virtnet_free_queues(vi); | |
1652 | err: | |
1653 | return ret; | |
3f9c10b0 AS |
1654 | } |
1655 | ||
fbf28d78 MD |
1656 | #ifdef CONFIG_SYSFS |
1657 | static ssize_t mergeable_rx_buffer_size_show(struct netdev_rx_queue *queue, | |
1658 | struct rx_queue_attribute *attribute, char *buf) | |
1659 | { | |
1660 | struct virtnet_info *vi = netdev_priv(queue->dev); | |
1661 | unsigned int queue_index = get_netdev_rx_queue_index(queue); | |
1662 | struct ewma *avg; | |
1663 | ||
1664 | BUG_ON(queue_index >= vi->max_queue_pairs); | |
1665 | avg = &vi->rq[queue_index].mrg_avg_pkt_len; | |
1666 | return sprintf(buf, "%u\n", get_mergeable_buf_len(avg)); | |
1667 | } | |
1668 | ||
1669 | static struct rx_queue_attribute mergeable_rx_buffer_size_attribute = | |
1670 | __ATTR_RO(mergeable_rx_buffer_size); | |
1671 | ||
1672 | static struct attribute *virtio_net_mrg_rx_attrs[] = { | |
1673 | &mergeable_rx_buffer_size_attribute.attr, | |
1674 | NULL | |
1675 | }; | |
1676 | ||
1677 | static const struct attribute_group virtio_net_mrg_rx_group = { | |
1678 | .name = "virtio_net", | |
1679 | .attrs = virtio_net_mrg_rx_attrs | |
1680 | }; | |
1681 | #endif | |
1682 | ||
892d6eb1 JW |
1683 | static bool virtnet_fail_on_feature(struct virtio_device *vdev, |
1684 | unsigned int fbit, | |
1685 | const char *fname, const char *dname) | |
1686 | { | |
1687 | if (!virtio_has_feature(vdev, fbit)) | |
1688 | return false; | |
1689 | ||
1690 | dev_err(&vdev->dev, "device advertises feature %s but not %s", | |
1691 | fname, dname); | |
1692 | ||
1693 | return true; | |
1694 | } | |
1695 | ||
1696 | #define VIRTNET_FAIL_ON(vdev, fbit, dbit) \ | |
1697 | virtnet_fail_on_feature(vdev, fbit, #fbit, dbit) | |
1698 | ||
1699 | static bool virtnet_validate_features(struct virtio_device *vdev) | |
1700 | { | |
1701 | if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) && | |
1702 | (VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_RX, | |
1703 | "VIRTIO_NET_F_CTRL_VQ") || | |
1704 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_VLAN, | |
1705 | "VIRTIO_NET_F_CTRL_VQ") || | |
1706 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE, | |
1707 | "VIRTIO_NET_F_CTRL_VQ") || | |
1708 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_MQ, "VIRTIO_NET_F_CTRL_VQ") || | |
1709 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR, | |
1710 | "VIRTIO_NET_F_CTRL_VQ"))) { | |
1711 | return false; | |
1712 | } | |
1713 | ||
1714 | return true; | |
1715 | } | |
1716 | ||
296f96fc RR |
1717 | static int virtnet_probe(struct virtio_device *vdev) |
1718 | { | |
986a4f4d | 1719 | int i, err; |
296f96fc RR |
1720 | struct net_device *dev; |
1721 | struct virtnet_info *vi; | |
986a4f4d JW |
1722 | u16 max_queue_pairs; |
1723 | ||
892d6eb1 JW |
1724 | if (!virtnet_validate_features(vdev)) |
1725 | return -EINVAL; | |
1726 | ||
986a4f4d | 1727 | /* Find if host supports multiqueue virtio_net device */ |
855e0c52 RR |
1728 | err = virtio_cread_feature(vdev, VIRTIO_NET_F_MQ, |
1729 | struct virtio_net_config, | |
1730 | max_virtqueue_pairs, &max_queue_pairs); | |
986a4f4d JW |
1731 | |
1732 | /* We need at least 2 queue's */ | |
1733 | if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN || | |
1734 | max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX || | |
1735 | !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) | |
1736 | max_queue_pairs = 1; | |
296f96fc RR |
1737 | |
1738 | /* Allocate ourselves a network device with room for our info */ | |
986a4f4d | 1739 | dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs); |
296f96fc RR |
1740 | if (!dev) |
1741 | return -ENOMEM; | |
1742 | ||
1743 | /* Set up network device as normal. */ | |
f2f2c8b4 | 1744 | dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE; |
76288b4e | 1745 | dev->netdev_ops = &virtnet_netdev; |
296f96fc | 1746 | dev->features = NETIF_F_HIGHDMA; |
3fa2a1df | 1747 | |
7ad24ea4 | 1748 | dev->ethtool_ops = &virtnet_ethtool_ops; |
296f96fc RR |
1749 | SET_NETDEV_DEV(dev, &vdev->dev); |
1750 | ||
1751 | /* Do we support "hardware" checksums? */ | |
98e778c9 | 1752 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) { |
296f96fc | 1753 | /* This opens up the world of extra features. */ |
98e778c9 MM |
1754 | dev->hw_features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST; |
1755 | if (csum) | |
1756 | dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST; | |
1757 | ||
1758 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) { | |
3d0ad094 | 1759 | dev->hw_features |= NETIF_F_TSO |
34a48579 RR |
1760 | | NETIF_F_TSO_ECN | NETIF_F_TSO6; |
1761 | } | |
5539ae96 | 1762 | /* Individual feature bits: what can host handle? */ |
98e778c9 MM |
1763 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4)) |
1764 | dev->hw_features |= NETIF_F_TSO; | |
1765 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6)) | |
1766 | dev->hw_features |= NETIF_F_TSO6; | |
1767 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN)) | |
1768 | dev->hw_features |= NETIF_F_TSO_ECN; | |
98e778c9 MM |
1769 | |
1770 | if (gso) | |
3d0ad094 | 1771 | dev->features |= dev->hw_features & NETIF_F_ALL_TSO; |
98e778c9 | 1772 | /* (!csum && gso) case will be fixed by register_netdev() */ |
296f96fc | 1773 | } |
4f49129b TH |
1774 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM)) |
1775 | dev->features |= NETIF_F_RXCSUM; | |
296f96fc | 1776 | |
4fda8302 JW |
1777 | dev->vlan_features = dev->features; |
1778 | ||
296f96fc | 1779 | /* Configuration may specify what MAC to use. Otherwise random. */ |
855e0c52 RR |
1780 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) |
1781 | virtio_cread_bytes(vdev, | |
1782 | offsetof(struct virtio_net_config, mac), | |
1783 | dev->dev_addr, dev->addr_len); | |
1784 | else | |
f2cedb63 | 1785 | eth_hw_addr_random(dev); |
296f96fc RR |
1786 | |
1787 | /* Set up our device-specific information */ | |
1788 | vi = netdev_priv(dev); | |
296f96fc RR |
1789 | vi->dev = dev; |
1790 | vi->vdev = vdev; | |
d9d5dcc8 | 1791 | vdev->priv = vi; |
3fa2a1df | 1792 | vi->stats = alloc_percpu(struct virtnet_stats); |
1793 | err = -ENOMEM; | |
1794 | if (vi->stats == NULL) | |
1795 | goto free; | |
1796 | ||
827da44c JS |
1797 | for_each_possible_cpu(i) { |
1798 | struct virtnet_stats *virtnet_stats; | |
1799 | virtnet_stats = per_cpu_ptr(vi->stats, i); | |
1800 | u64_stats_init(&virtnet_stats->tx_syncp); | |
1801 | u64_stats_init(&virtnet_stats->rx_syncp); | |
1802 | } | |
1803 | ||
586d17c5 | 1804 | INIT_WORK(&vi->config_work, virtnet_config_changed_work); |
296f96fc | 1805 | |
97402b96 | 1806 | /* If we can receive ANY GSO packets, we must allocate large ones. */ |
8e95a202 JP |
1807 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) || |
1808 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) || | |
3d0ad094 | 1809 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN)) |
97402b96 HX |
1810 | vi->big_packets = true; |
1811 | ||
3f2c31d9 MM |
1812 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF)) |
1813 | vi->mergeable_rx_bufs = true; | |
1814 | ||
e7428e95 MT |
1815 | if (virtio_has_feature(vdev, VIRTIO_F_ANY_LAYOUT)) |
1816 | vi->any_header_sg = true; | |
1817 | ||
986a4f4d JW |
1818 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) |
1819 | vi->has_cvq = true; | |
1820 | ||
6ebbc1a6 ZH |
1821 | if (vi->any_header_sg) { |
1822 | if (vi->mergeable_rx_bufs) | |
1823 | dev->needed_headroom = sizeof(struct virtio_net_hdr_mrg_rxbuf); | |
1824 | else | |
1825 | dev->needed_headroom = sizeof(struct virtio_net_hdr); | |
1826 | } | |
1827 | ||
986a4f4d JW |
1828 | /* Use single tx/rx queue pair as default */ |
1829 | vi->curr_queue_pairs = 1; | |
1830 | vi->max_queue_pairs = max_queue_pairs; | |
1831 | ||
1832 | /* Allocate/initialize the rx/tx queues, and invoke find_vqs */ | |
3f9c10b0 | 1833 | err = init_vqs(vi); |
d2a7ddda | 1834 | if (err) |
9bb8ca86 | 1835 | goto free_stats; |
296f96fc | 1836 | |
fbf28d78 MD |
1837 | #ifdef CONFIG_SYSFS |
1838 | if (vi->mergeable_rx_bufs) | |
1839 | dev->sysfs_rx_queue_group = &virtio_net_mrg_rx_group; | |
1840 | #endif | |
0f13b66b ZYW |
1841 | netif_set_real_num_tx_queues(dev, vi->curr_queue_pairs); |
1842 | netif_set_real_num_rx_queues(dev, vi->curr_queue_pairs); | |
986a4f4d | 1843 | |
296f96fc RR |
1844 | err = register_netdev(dev); |
1845 | if (err) { | |
1846 | pr_debug("virtio_net: registering device failed\n"); | |
d2a7ddda | 1847 | goto free_vqs; |
296f96fc | 1848 | } |
b3369c1f | 1849 | |
4baf1e33 MT |
1850 | virtio_device_ready(vdev); |
1851 | ||
b3369c1f | 1852 | /* Last of all, set up some receive buffers. */ |
55257d72 | 1853 | for (i = 0; i < vi->curr_queue_pairs; i++) { |
986a4f4d JW |
1854 | try_fill_recv(&vi->rq[i], GFP_KERNEL); |
1855 | ||
1856 | /* If we didn't even get one input buffer, we're useless. */ | |
be121f46 JW |
1857 | if (vi->rq[i].vq->num_free == |
1858 | virtqueue_get_vring_size(vi->rq[i].vq)) { | |
986a4f4d JW |
1859 | free_unused_bufs(vi); |
1860 | err = -ENOMEM; | |
1861 | goto free_recv_bufs; | |
1862 | } | |
b3369c1f RR |
1863 | } |
1864 | ||
8de4b2f3 WG |
1865 | vi->nb.notifier_call = &virtnet_cpu_callback; |
1866 | err = register_hotcpu_notifier(&vi->nb); | |
1867 | if (err) { | |
1868 | pr_debug("virtio_net: registering cpu notifier failed\n"); | |
1869 | goto free_recv_bufs; | |
1870 | } | |
1871 | ||
167c25e4 JW |
1872 | /* Assume link up if device can't report link status, |
1873 | otherwise get link status from config. */ | |
1874 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) { | |
1875 | netif_carrier_off(dev); | |
3b07e9ca | 1876 | schedule_work(&vi->config_work); |
167c25e4 JW |
1877 | } else { |
1878 | vi->status = VIRTIO_NET_S_LINK_UP; | |
1879 | netif_carrier_on(dev); | |
1880 | } | |
9f4d26d0 | 1881 | |
986a4f4d JW |
1882 | pr_debug("virtnet: registered device %s with %d RX and TX vq's\n", |
1883 | dev->name, max_queue_pairs); | |
1884 | ||
296f96fc RR |
1885 | return 0; |
1886 | ||
986a4f4d | 1887 | free_recv_bufs: |
02465555 MT |
1888 | vi->vdev->config->reset(vdev); |
1889 | ||
986a4f4d | 1890 | free_receive_bufs(vi); |
b3369c1f | 1891 | unregister_netdev(dev); |
d2a7ddda | 1892 | free_vqs: |
986a4f4d | 1893 | cancel_delayed_work_sync(&vi->refill); |
fb51879d | 1894 | free_receive_page_frags(vi); |
e9d7417b | 1895 | virtnet_del_vqs(vi); |
3fa2a1df | 1896 | free_stats: |
1897 | free_percpu(vi->stats); | |
296f96fc RR |
1898 | free: |
1899 | free_netdev(dev); | |
1900 | return err; | |
1901 | } | |
1902 | ||
04486ed0 | 1903 | static void remove_vq_common(struct virtnet_info *vi) |
296f96fc | 1904 | { |
04486ed0 | 1905 | vi->vdev->config->reset(vi->vdev); |
830a8a97 SM |
1906 | |
1907 | /* Free unused buffers in both send and recv, if any. */ | |
9ab86bbc | 1908 | free_unused_bufs(vi); |
fb6813f4 | 1909 | |
986a4f4d | 1910 | free_receive_bufs(vi); |
d2a7ddda | 1911 | |
fb51879d MD |
1912 | free_receive_page_frags(vi); |
1913 | ||
986a4f4d | 1914 | virtnet_del_vqs(vi); |
04486ed0 AS |
1915 | } |
1916 | ||
8cc085d6 | 1917 | static void virtnet_remove(struct virtio_device *vdev) |
04486ed0 AS |
1918 | { |
1919 | struct virtnet_info *vi = vdev->priv; | |
1920 | ||
8de4b2f3 WG |
1921 | unregister_hotcpu_notifier(&vi->nb); |
1922 | ||
102a2786 MT |
1923 | /* Make sure no work handler is accessing the device. */ |
1924 | flush_work(&vi->config_work); | |
586d17c5 | 1925 | |
04486ed0 AS |
1926 | unregister_netdev(vi->dev); |
1927 | ||
1928 | remove_vq_common(vi); | |
fb6813f4 | 1929 | |
2e66f55b | 1930 | free_percpu(vi->stats); |
74b2553f | 1931 | free_netdev(vi->dev); |
296f96fc RR |
1932 | } |
1933 | ||
89107000 | 1934 | #ifdef CONFIG_PM_SLEEP |
0741bcb5 AS |
1935 | static int virtnet_freeze(struct virtio_device *vdev) |
1936 | { | |
1937 | struct virtnet_info *vi = vdev->priv; | |
986a4f4d | 1938 | int i; |
0741bcb5 | 1939 | |
ec9debbd JW |
1940 | unregister_hotcpu_notifier(&vi->nb); |
1941 | ||
102a2786 MT |
1942 | /* Make sure no work handler is accessing the device */ |
1943 | flush_work(&vi->config_work); | |
586d17c5 | 1944 | |
0741bcb5 AS |
1945 | netif_device_detach(vi->dev); |
1946 | cancel_delayed_work_sync(&vi->refill); | |
1947 | ||
91815639 | 1948 | if (netif_running(vi->dev)) { |
986a4f4d JW |
1949 | for (i = 0; i < vi->max_queue_pairs; i++) { |
1950 | napi_disable(&vi->rq[i].napi); | |
91815639 | 1951 | napi_hash_del(&vi->rq[i].napi); |
986a4f4d JW |
1952 | netif_napi_del(&vi->rq[i].napi); |
1953 | } | |
91815639 | 1954 | } |
0741bcb5 AS |
1955 | |
1956 | remove_vq_common(vi); | |
1957 | ||
1958 | return 0; | |
1959 | } | |
1960 | ||
1961 | static int virtnet_restore(struct virtio_device *vdev) | |
1962 | { | |
1963 | struct virtnet_info *vi = vdev->priv; | |
986a4f4d | 1964 | int err, i; |
0741bcb5 AS |
1965 | |
1966 | err = init_vqs(vi); | |
1967 | if (err) | |
1968 | return err; | |
1969 | ||
e53fbd11 MT |
1970 | virtio_device_ready(vdev); |
1971 | ||
6cd4ce00 JW |
1972 | if (netif_running(vi->dev)) { |
1973 | for (i = 0; i < vi->curr_queue_pairs; i++) | |
1974 | if (!try_fill_recv(&vi->rq[i], GFP_KERNEL)) | |
1975 | schedule_delayed_work(&vi->refill, 0); | |
1976 | ||
986a4f4d JW |
1977 | for (i = 0; i < vi->max_queue_pairs; i++) |
1978 | virtnet_napi_enable(&vi->rq[i]); | |
6cd4ce00 | 1979 | } |
0741bcb5 AS |
1980 | |
1981 | netif_device_attach(vi->dev); | |
1982 | ||
35ed159b | 1983 | rtnl_lock(); |
986a4f4d | 1984 | virtnet_set_queues(vi, vi->curr_queue_pairs); |
35ed159b | 1985 | rtnl_unlock(); |
986a4f4d | 1986 | |
ec9debbd JW |
1987 | err = register_hotcpu_notifier(&vi->nb); |
1988 | if (err) | |
1989 | return err; | |
1990 | ||
0741bcb5 AS |
1991 | return 0; |
1992 | } | |
1993 | #endif | |
1994 | ||
296f96fc RR |
1995 | static struct virtio_device_id id_table[] = { |
1996 | { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID }, | |
1997 | { 0 }, | |
1998 | }; | |
1999 | ||
c45a6816 | 2000 | static unsigned int features[] = { |
5e4fe5c4 MM |
2001 | VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM, |
2002 | VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC, | |
3d0ad094 | 2003 | VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_TSO6, |
97402b96 | 2004 | VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, |
3d0ad094 | 2005 | VIRTIO_NET_F_GUEST_ECN, |
2a41f71d | 2006 | VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ, |
0bde9569 | 2007 | VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, |
986a4f4d | 2008 | VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ, |
7e58d5ae | 2009 | VIRTIO_NET_F_CTRL_MAC_ADDR, |
e7428e95 | 2010 | VIRTIO_F_ANY_LAYOUT, |
c45a6816 RR |
2011 | }; |
2012 | ||
22402529 | 2013 | static struct virtio_driver virtio_net_driver = { |
c45a6816 RR |
2014 | .feature_table = features, |
2015 | .feature_table_size = ARRAY_SIZE(features), | |
296f96fc RR |
2016 | .driver.name = KBUILD_MODNAME, |
2017 | .driver.owner = THIS_MODULE, | |
2018 | .id_table = id_table, | |
2019 | .probe = virtnet_probe, | |
8cc085d6 | 2020 | .remove = virtnet_remove, |
9f4d26d0 | 2021 | .config_changed = virtnet_config_changed, |
89107000 | 2022 | #ifdef CONFIG_PM_SLEEP |
0741bcb5 AS |
2023 | .freeze = virtnet_freeze, |
2024 | .restore = virtnet_restore, | |
2025 | #endif | |
296f96fc RR |
2026 | }; |
2027 | ||
b2a17029 | 2028 | module_virtio_driver(virtio_net_driver); |
296f96fc RR |
2029 | |
2030 | MODULE_DEVICE_TABLE(virtio, id_table); | |
2031 | MODULE_DESCRIPTION("Virtio network driver"); | |
2032 | MODULE_LICENSE("GPL"); |