Commit | Line | Data |
---|---|---|
48925e37 | 1 | /* A network driver using virtio. |
296f96fc RR |
2 | * |
3 | * Copyright 2007 Rusty Russell <rusty@rustcorp.com.au> IBM Corporation | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or modify | |
6 | * it under the terms of the GNU General Public License as published by | |
7 | * the Free Software Foundation; either version 2 of the License, or | |
8 | * (at your option) any later version. | |
9 | * | |
10 | * This program is distributed in the hope that it will be useful, | |
11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
13 | * GNU General Public License for more details. | |
14 | * | |
15 | * You should have received a copy of the GNU General Public License | |
16 | * along with this program; if not, write to the Free Software | |
17 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
18 | */ | |
19 | //#define DEBUG | |
20 | #include <linux/netdevice.h> | |
21 | #include <linux/etherdevice.h> | |
a9ea3fc6 | 22 | #include <linux/ethtool.h> |
296f96fc RR |
23 | #include <linux/module.h> |
24 | #include <linux/virtio.h> | |
25 | #include <linux/virtio_net.h> | |
26 | #include <linux/scatterlist.h> | |
e918085a | 27 | #include <linux/if_vlan.h> |
5a0e3ad6 | 28 | #include <linux/slab.h> |
8de4b2f3 | 29 | #include <linux/cpu.h> |
296f96fc | 30 | |
d34710e3 | 31 | static int napi_weight = NAPI_POLL_WEIGHT; |
6c0cd7c0 DL |
32 | module_param(napi_weight, int, 0444); |
33 | ||
eb939922 | 34 | static bool csum = true, gso = true; |
34a48579 RR |
35 | module_param(csum, bool, 0444); |
36 | module_param(gso, bool, 0444); | |
37 | ||
296f96fc | 38 | /* FIXME: MTU in config. */ |
e918085a | 39 | #define MAX_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN) |
3f2c31d9 | 40 | #define GOOD_COPY_LEN 128 |
296f96fc | 41 | |
66846048 | 42 | #define VIRTNET_DRIVER_VERSION "1.0.0" |
2a41f71d | 43 | |
3fa2a1df | 44 | struct virtnet_stats { |
83a27052 ED |
45 | struct u64_stats_sync tx_syncp; |
46 | struct u64_stats_sync rx_syncp; | |
3fa2a1df | 47 | u64 tx_bytes; |
48 | u64 tx_packets; | |
49 | ||
50 | u64 rx_bytes; | |
51 | u64 rx_packets; | |
52 | }; | |
53 | ||
e9d7417b JW |
54 | /* Internal representation of a send virtqueue */ |
55 | struct send_queue { | |
56 | /* Virtqueue associated with this send _queue */ | |
57 | struct virtqueue *vq; | |
58 | ||
59 | /* TX: fragments + linear part + virtio header */ | |
60 | struct scatterlist sg[MAX_SKB_FRAGS + 2]; | |
986a4f4d JW |
61 | |
62 | /* Name of the send queue: output.$index */ | |
63 | char name[40]; | |
e9d7417b JW |
64 | }; |
65 | ||
66 | /* Internal representation of a receive virtqueue */ | |
67 | struct receive_queue { | |
68 | /* Virtqueue associated with this receive_queue */ | |
69 | struct virtqueue *vq; | |
70 | ||
296f96fc RR |
71 | struct napi_struct napi; |
72 | ||
73 | /* Number of input buffers, and max we've ever had. */ | |
74 | unsigned int num, max; | |
75 | ||
e9d7417b JW |
76 | /* Chain pages by the private ptr. */ |
77 | struct page *pages; | |
78 | ||
79 | /* RX: fragments + linear part + virtio header */ | |
80 | struct scatterlist sg[MAX_SKB_FRAGS + 2]; | |
986a4f4d JW |
81 | |
82 | /* Name of this receive queue: input.$index */ | |
83 | char name[40]; | |
e9d7417b JW |
84 | }; |
85 | ||
86 | struct virtnet_info { | |
87 | struct virtio_device *vdev; | |
88 | struct virtqueue *cvq; | |
89 | struct net_device *dev; | |
986a4f4d JW |
90 | struct send_queue *sq; |
91 | struct receive_queue *rq; | |
e9d7417b JW |
92 | unsigned int status; |
93 | ||
986a4f4d JW |
94 | /* Max # of queue pairs supported by the device */ |
95 | u16 max_queue_pairs; | |
96 | ||
97 | /* # of queue pairs currently used by the driver */ | |
98 | u16 curr_queue_pairs; | |
99 | ||
97402b96 HX |
100 | /* I like... big packets and I cannot lie! */ |
101 | bool big_packets; | |
102 | ||
3f2c31d9 MM |
103 | /* Host will merge rx buffers for big packets (shake it! shake it!) */ |
104 | bool mergeable_rx_bufs; | |
105 | ||
986a4f4d JW |
106 | /* Has control virtqueue */ |
107 | bool has_cvq; | |
108 | ||
e7428e95 MT |
109 | /* Host can handle any s/g split between our header and packet data */ |
110 | bool any_header_sg; | |
111 | ||
586d17c5 JW |
112 | /* enable config space updates */ |
113 | bool config_enable; | |
114 | ||
3fa2a1df | 115 | /* Active statistics */ |
116 | struct virtnet_stats __percpu *stats; | |
117 | ||
3161e453 RR |
118 | /* Work struct for refilling if we run low on memory. */ |
119 | struct delayed_work refill; | |
120 | ||
586d17c5 JW |
121 | /* Work struct for config space updates */ |
122 | struct work_struct config_work; | |
123 | ||
124 | /* Lock for config space updates */ | |
125 | struct mutex config_lock; | |
986a4f4d JW |
126 | |
127 | /* Does the affinity hint is set for virtqueues? */ | |
128 | bool affinity_hint_set; | |
47be2479 WG |
129 | |
130 | /* Per-cpu variable to show the mapping from CPU to virtqueue */ | |
131 | int __percpu *vq_index; | |
8de4b2f3 WG |
132 | |
133 | /* CPU hot plug notifier */ | |
134 | struct notifier_block nb; | |
296f96fc RR |
135 | }; |
136 | ||
b3f24698 RR |
137 | struct skb_vnet_hdr { |
138 | union { | |
139 | struct virtio_net_hdr hdr; | |
140 | struct virtio_net_hdr_mrg_rxbuf mhdr; | |
141 | }; | |
142 | }; | |
143 | ||
9ab86bbc SM |
144 | struct padded_vnet_hdr { |
145 | struct virtio_net_hdr hdr; | |
146 | /* | |
147 | * virtio_net_hdr should be in a separated sg buffer because of a | |
148 | * QEMU bug, and data sg buffer shares same page with this header sg. | |
149 | * This padding makes next sg 16 byte aligned after virtio_net_hdr. | |
150 | */ | |
151 | char padding[6]; | |
152 | }; | |
153 | ||
986a4f4d JW |
154 | /* Converting between virtqueue no. and kernel tx/rx queue no. |
155 | * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq | |
156 | */ | |
157 | static int vq2txq(struct virtqueue *vq) | |
158 | { | |
9d0ca6ed | 159 | return (vq->index - 1) / 2; |
986a4f4d JW |
160 | } |
161 | ||
162 | static int txq2vq(int txq) | |
163 | { | |
164 | return txq * 2 + 1; | |
165 | } | |
166 | ||
167 | static int vq2rxq(struct virtqueue *vq) | |
168 | { | |
9d0ca6ed | 169 | return vq->index / 2; |
986a4f4d JW |
170 | } |
171 | ||
172 | static int rxq2vq(int rxq) | |
173 | { | |
174 | return rxq * 2; | |
175 | } | |
176 | ||
b3f24698 | 177 | static inline struct skb_vnet_hdr *skb_vnet_hdr(struct sk_buff *skb) |
296f96fc | 178 | { |
b3f24698 | 179 | return (struct skb_vnet_hdr *)skb->cb; |
296f96fc RR |
180 | } |
181 | ||
9ab86bbc SM |
182 | /* |
183 | * private is used to chain pages for big packets, put the whole | |
184 | * most recent used list in the beginning for reuse | |
185 | */ | |
e9d7417b | 186 | static void give_pages(struct receive_queue *rq, struct page *page) |
0a888fd1 | 187 | { |
9ab86bbc | 188 | struct page *end; |
0a888fd1 | 189 | |
e9d7417b | 190 | /* Find end of list, sew whole thing into vi->rq.pages. */ |
9ab86bbc | 191 | for (end = page; end->private; end = (struct page *)end->private); |
e9d7417b JW |
192 | end->private = (unsigned long)rq->pages; |
193 | rq->pages = page; | |
0a888fd1 MM |
194 | } |
195 | ||
e9d7417b | 196 | static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask) |
fb6813f4 | 197 | { |
e9d7417b | 198 | struct page *p = rq->pages; |
fb6813f4 | 199 | |
9ab86bbc | 200 | if (p) { |
e9d7417b | 201 | rq->pages = (struct page *)p->private; |
9ab86bbc SM |
202 | /* clear private here, it is used to chain pages */ |
203 | p->private = 0; | |
204 | } else | |
fb6813f4 RR |
205 | p = alloc_page(gfp_mask); |
206 | return p; | |
207 | } | |
208 | ||
e9d7417b | 209 | static void skb_xmit_done(struct virtqueue *vq) |
296f96fc | 210 | { |
e9d7417b | 211 | struct virtnet_info *vi = vq->vdev->priv; |
296f96fc | 212 | |
2cb9c6ba | 213 | /* Suppress further interrupts. */ |
e9d7417b | 214 | virtqueue_disable_cb(vq); |
11a3a154 | 215 | |
363f1514 | 216 | /* We were probably waiting for more output buffers. */ |
986a4f4d | 217 | netif_wake_subqueue(vi->dev, vq2txq(vq)); |
296f96fc RR |
218 | } |
219 | ||
9ab86bbc SM |
220 | static void set_skb_frag(struct sk_buff *skb, struct page *page, |
221 | unsigned int offset, unsigned int *len) | |
296f96fc | 222 | { |
8a59a7b9 | 223 | int size = min((unsigned)PAGE_SIZE - offset, *len); |
9ab86bbc | 224 | int i = skb_shinfo(skb)->nr_frags; |
9ab86bbc | 225 | |
8a59a7b9 | 226 | __skb_fill_page_desc(skb, i, page, offset, size); |
9ab86bbc | 227 | |
8a59a7b9 KK |
228 | skb->data_len += size; |
229 | skb->len += size; | |
4b727361 | 230 | skb->truesize += PAGE_SIZE; |
9ab86bbc | 231 | skb_shinfo(skb)->nr_frags++; |
c9af6db4 | 232 | skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG; |
8a59a7b9 | 233 | *len -= size; |
9ab86bbc | 234 | } |
23cde76d | 235 | |
3464645a | 236 | /* Called from bottom half context */ |
e9d7417b | 237 | static struct sk_buff *page_to_skb(struct receive_queue *rq, |
9ab86bbc SM |
238 | struct page *page, unsigned int len) |
239 | { | |
e9d7417b | 240 | struct virtnet_info *vi = rq->vq->vdev->priv; |
9ab86bbc SM |
241 | struct sk_buff *skb; |
242 | struct skb_vnet_hdr *hdr; | |
243 | unsigned int copy, hdr_len, offset; | |
244 | char *p; | |
fb6813f4 | 245 | |
9ab86bbc | 246 | p = page_address(page); |
3f2c31d9 | 247 | |
9ab86bbc SM |
248 | /* copy small packet so we can reuse these pages for small data */ |
249 | skb = netdev_alloc_skb_ip_align(vi->dev, GOOD_COPY_LEN); | |
250 | if (unlikely(!skb)) | |
251 | return NULL; | |
3f2c31d9 | 252 | |
9ab86bbc | 253 | hdr = skb_vnet_hdr(skb); |
3f2c31d9 | 254 | |
9ab86bbc SM |
255 | if (vi->mergeable_rx_bufs) { |
256 | hdr_len = sizeof hdr->mhdr; | |
257 | offset = hdr_len; | |
258 | } else { | |
259 | hdr_len = sizeof hdr->hdr; | |
260 | offset = sizeof(struct padded_vnet_hdr); | |
261 | } | |
3f2c31d9 | 262 | |
9ab86bbc | 263 | memcpy(hdr, p, hdr_len); |
3f2c31d9 | 264 | |
9ab86bbc SM |
265 | len -= hdr_len; |
266 | p += offset; | |
3f2c31d9 | 267 | |
9ab86bbc SM |
268 | copy = len; |
269 | if (copy > skb_tailroom(skb)) | |
270 | copy = skb_tailroom(skb); | |
271 | memcpy(skb_put(skb, copy), p, copy); | |
3f2c31d9 | 272 | |
9ab86bbc SM |
273 | len -= copy; |
274 | offset += copy; | |
3f2c31d9 | 275 | |
e878d78b SL |
276 | /* |
277 | * Verify that we can indeed put this data into a skb. | |
278 | * This is here to handle cases when the device erroneously | |
279 | * tries to receive more than is possible. This is usually | |
280 | * the case of a broken device. | |
281 | */ | |
282 | if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) { | |
be443899 | 283 | net_dbg_ratelimited("%s: too much data\n", skb->dev->name); |
e878d78b SL |
284 | dev_kfree_skb(skb); |
285 | return NULL; | |
286 | } | |
287 | ||
9ab86bbc SM |
288 | while (len) { |
289 | set_skb_frag(skb, page, offset, &len); | |
290 | page = (struct page *)page->private; | |
291 | offset = 0; | |
292 | } | |
3f2c31d9 | 293 | |
9ab86bbc | 294 | if (page) |
e9d7417b | 295 | give_pages(rq, page); |
3f2c31d9 | 296 | |
9ab86bbc SM |
297 | return skb; |
298 | } | |
3f2c31d9 | 299 | |
e9d7417b | 300 | static int receive_mergeable(struct receive_queue *rq, struct sk_buff *skb) |
9ab86bbc SM |
301 | { |
302 | struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb); | |
303 | struct page *page; | |
304 | int num_buf, i, len; | |
305 | ||
306 | num_buf = hdr->mhdr.num_buffers; | |
307 | while (--num_buf) { | |
308 | i = skb_shinfo(skb)->nr_frags; | |
309 | if (i >= MAX_SKB_FRAGS) { | |
310 | pr_debug("%s: packet too long\n", skb->dev->name); | |
311 | skb->dev->stats.rx_length_errors++; | |
312 | return -EINVAL; | |
313 | } | |
e9d7417b | 314 | page = virtqueue_get_buf(rq->vq, &len); |
9ab86bbc SM |
315 | if (!page) { |
316 | pr_debug("%s: rx error: %d buffers missing\n", | |
317 | skb->dev->name, hdr->mhdr.num_buffers); | |
318 | skb->dev->stats.rx_length_errors++; | |
319 | return -EINVAL; | |
3f2c31d9 | 320 | } |
3fa2a1df | 321 | |
9ab86bbc SM |
322 | if (len > PAGE_SIZE) |
323 | len = PAGE_SIZE; | |
324 | ||
325 | set_skb_frag(skb, page, 0, &len); | |
326 | ||
e9d7417b | 327 | --rq->num; |
9ab86bbc SM |
328 | } |
329 | return 0; | |
330 | } | |
331 | ||
e9d7417b | 332 | static void receive_buf(struct receive_queue *rq, void *buf, unsigned int len) |
9ab86bbc | 333 | { |
e9d7417b JW |
334 | struct virtnet_info *vi = rq->vq->vdev->priv; |
335 | struct net_device *dev = vi->dev; | |
58472a76 | 336 | struct virtnet_stats *stats = this_cpu_ptr(vi->stats); |
9ab86bbc SM |
337 | struct sk_buff *skb; |
338 | struct page *page; | |
339 | struct skb_vnet_hdr *hdr; | |
3f2c31d9 | 340 | |
9ab86bbc SM |
341 | if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) { |
342 | pr_debug("%s: short packet %i\n", dev->name, len); | |
343 | dev->stats.rx_length_errors++; | |
344 | if (vi->mergeable_rx_bufs || vi->big_packets) | |
e9d7417b | 345 | give_pages(rq, buf); |
9ab86bbc SM |
346 | else |
347 | dev_kfree_skb(buf); | |
348 | return; | |
349 | } | |
3f2c31d9 | 350 | |
9ab86bbc SM |
351 | if (!vi->mergeable_rx_bufs && !vi->big_packets) { |
352 | skb = buf; | |
353 | len -= sizeof(struct virtio_net_hdr); | |
354 | skb_trim(skb, len); | |
355 | } else { | |
356 | page = buf; | |
e9d7417b | 357 | skb = page_to_skb(rq, page, len); |
9ab86bbc | 358 | if (unlikely(!skb)) { |
3f2c31d9 | 359 | dev->stats.rx_dropped++; |
e9d7417b | 360 | give_pages(rq, page); |
9ab86bbc | 361 | return; |
3f2c31d9 | 362 | } |
9ab86bbc | 363 | if (vi->mergeable_rx_bufs) |
e9d7417b | 364 | if (receive_mergeable(rq, skb)) { |
9ab86bbc SM |
365 | dev_kfree_skb(skb); |
366 | return; | |
367 | } | |
97402b96 | 368 | } |
3f2c31d9 | 369 | |
9ab86bbc | 370 | hdr = skb_vnet_hdr(skb); |
3fa2a1df | 371 | |
83a27052 | 372 | u64_stats_update_begin(&stats->rx_syncp); |
3fa2a1df | 373 | stats->rx_bytes += skb->len; |
374 | stats->rx_packets++; | |
83a27052 | 375 | u64_stats_update_end(&stats->rx_syncp); |
296f96fc | 376 | |
b3f24698 | 377 | if (hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { |
296f96fc | 378 | pr_debug("Needs csum!\n"); |
b3f24698 RR |
379 | if (!skb_partial_csum_set(skb, |
380 | hdr->hdr.csum_start, | |
381 | hdr->hdr.csum_offset)) | |
296f96fc | 382 | goto frame_err; |
10a8d94a JW |
383 | } else if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) { |
384 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
296f96fc RR |
385 | } |
386 | ||
23cde76d MM |
387 | skb->protocol = eth_type_trans(skb, dev); |
388 | pr_debug("Receiving skb proto 0x%04x len %i type %i\n", | |
389 | ntohs(skb->protocol), skb->len, skb->pkt_type); | |
390 | ||
b3f24698 | 391 | if (hdr->hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) { |
296f96fc | 392 | pr_debug("GSO!\n"); |
b3f24698 | 393 | switch (hdr->hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { |
296f96fc | 394 | case VIRTIO_NET_HDR_GSO_TCPV4: |
c9af6db4 | 395 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; |
296f96fc | 396 | break; |
296f96fc | 397 | case VIRTIO_NET_HDR_GSO_UDP: |
c9af6db4 | 398 | skb_shinfo(skb)->gso_type = SKB_GSO_UDP; |
296f96fc RR |
399 | break; |
400 | case VIRTIO_NET_HDR_GSO_TCPV6: | |
c9af6db4 | 401 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6; |
296f96fc RR |
402 | break; |
403 | default: | |
be443899 AW |
404 | net_warn_ratelimited("%s: bad gso type %u.\n", |
405 | dev->name, hdr->hdr.gso_type); | |
296f96fc RR |
406 | goto frame_err; |
407 | } | |
408 | ||
b3f24698 | 409 | if (hdr->hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN) |
c9af6db4 | 410 | skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN; |
34a48579 | 411 | |
b3f24698 | 412 | skb_shinfo(skb)->gso_size = hdr->hdr.gso_size; |
296f96fc | 413 | if (skb_shinfo(skb)->gso_size == 0) { |
be443899 | 414 | net_warn_ratelimited("%s: zero gso size.\n", dev->name); |
296f96fc RR |
415 | goto frame_err; |
416 | } | |
417 | ||
418 | /* Header must be checked, and gso_segs computed. */ | |
419 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
420 | skb_shinfo(skb)->gso_segs = 0; | |
421 | } | |
422 | ||
423 | netif_receive_skb(skb); | |
424 | return; | |
425 | ||
426 | frame_err: | |
427 | dev->stats.rx_frame_errors++; | |
296f96fc RR |
428 | dev_kfree_skb(skb); |
429 | } | |
430 | ||
e9d7417b | 431 | static int add_recvbuf_small(struct receive_queue *rq, gfp_t gfp) |
296f96fc | 432 | { |
e9d7417b | 433 | struct virtnet_info *vi = rq->vq->vdev->priv; |
296f96fc | 434 | struct sk_buff *skb; |
9ab86bbc | 435 | struct skb_vnet_hdr *hdr; |
9ab86bbc | 436 | int err; |
3f2c31d9 | 437 | |
3464645a | 438 | skb = __netdev_alloc_skb_ip_align(vi->dev, MAX_PACKET_LEN, gfp); |
9ab86bbc SM |
439 | if (unlikely(!skb)) |
440 | return -ENOMEM; | |
296f96fc | 441 | |
9ab86bbc | 442 | skb_put(skb, MAX_PACKET_LEN); |
3f2c31d9 | 443 | |
9ab86bbc | 444 | hdr = skb_vnet_hdr(skb); |
e9d7417b | 445 | sg_set_buf(rq->sg, &hdr->hdr, sizeof hdr->hdr); |
97402b96 | 446 | |
e9d7417b | 447 | skb_to_sgvec(skb, rq->sg + 1, 0, skb->len); |
97402b96 | 448 | |
9dc7b9e4 | 449 | err = virtqueue_add_inbuf(rq->vq, rq->sg, 2, skb, gfp); |
9ab86bbc SM |
450 | if (err < 0) |
451 | dev_kfree_skb(skb); | |
97402b96 | 452 | |
9ab86bbc SM |
453 | return err; |
454 | } | |
97402b96 | 455 | |
e9d7417b | 456 | static int add_recvbuf_big(struct receive_queue *rq, gfp_t gfp) |
9ab86bbc | 457 | { |
9ab86bbc SM |
458 | struct page *first, *list = NULL; |
459 | char *p; | |
460 | int i, err, offset; | |
461 | ||
e9d7417b | 462 | /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */ |
9ab86bbc | 463 | for (i = MAX_SKB_FRAGS + 1; i > 1; --i) { |
e9d7417b | 464 | first = get_a_page(rq, gfp); |
9ab86bbc SM |
465 | if (!first) { |
466 | if (list) | |
e9d7417b | 467 | give_pages(rq, list); |
9ab86bbc | 468 | return -ENOMEM; |
97402b96 | 469 | } |
e9d7417b | 470 | sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE); |
97402b96 | 471 | |
9ab86bbc SM |
472 | /* chain new page in list head to match sg */ |
473 | first->private = (unsigned long)list; | |
474 | list = first; | |
475 | } | |
296f96fc | 476 | |
e9d7417b | 477 | first = get_a_page(rq, gfp); |
9ab86bbc | 478 | if (!first) { |
e9d7417b | 479 | give_pages(rq, list); |
9ab86bbc SM |
480 | return -ENOMEM; |
481 | } | |
482 | p = page_address(first); | |
483 | ||
e9d7417b JW |
484 | /* rq->sg[0], rq->sg[1] share the same page */ |
485 | /* a separated rq->sg[0] for virtio_net_hdr only due to QEMU bug */ | |
486 | sg_set_buf(&rq->sg[0], p, sizeof(struct virtio_net_hdr)); | |
9ab86bbc | 487 | |
e9d7417b | 488 | /* rq->sg[1] for data packet, from offset */ |
9ab86bbc | 489 | offset = sizeof(struct padded_vnet_hdr); |
e9d7417b | 490 | sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset); |
9ab86bbc SM |
491 | |
492 | /* chain first in list head */ | |
493 | first->private = (unsigned long)list; | |
9dc7b9e4 RR |
494 | err = virtqueue_add_inbuf(rq->vq, rq->sg, MAX_SKB_FRAGS + 2, |
495 | first, gfp); | |
9ab86bbc | 496 | if (err < 0) |
e9d7417b | 497 | give_pages(rq, first); |
9ab86bbc SM |
498 | |
499 | return err; | |
296f96fc RR |
500 | } |
501 | ||
e9d7417b | 502 | static int add_recvbuf_mergeable(struct receive_queue *rq, gfp_t gfp) |
3f2c31d9 | 503 | { |
9ab86bbc | 504 | struct page *page; |
3f2c31d9 | 505 | int err; |
3f2c31d9 | 506 | |
e9d7417b | 507 | page = get_a_page(rq, gfp); |
9ab86bbc SM |
508 | if (!page) |
509 | return -ENOMEM; | |
3f2c31d9 | 510 | |
e9d7417b | 511 | sg_init_one(rq->sg, page_address(page), PAGE_SIZE); |
3f2c31d9 | 512 | |
9dc7b9e4 | 513 | err = virtqueue_add_inbuf(rq->vq, rq->sg, 1, page, gfp); |
9ab86bbc | 514 | if (err < 0) |
e9d7417b | 515 | give_pages(rq, page); |
3f2c31d9 | 516 | |
9ab86bbc SM |
517 | return err; |
518 | } | |
3f2c31d9 | 519 | |
b2baed69 RR |
520 | /* |
521 | * Returns false if we couldn't fill entirely (OOM). | |
522 | * | |
523 | * Normally run in the receive path, but can also be run from ndo_open | |
524 | * before we're receiving packets, or from refill_work which is | |
525 | * careful to disable receiving (using napi_disable). | |
526 | */ | |
e9d7417b | 527 | static bool try_fill_recv(struct receive_queue *rq, gfp_t gfp) |
9ab86bbc | 528 | { |
e9d7417b | 529 | struct virtnet_info *vi = rq->vq->vdev->priv; |
9ab86bbc | 530 | int err; |
1788f495 | 531 | bool oom; |
3f2c31d9 | 532 | |
9ab86bbc SM |
533 | do { |
534 | if (vi->mergeable_rx_bufs) | |
e9d7417b | 535 | err = add_recvbuf_mergeable(rq, gfp); |
9ab86bbc | 536 | else if (vi->big_packets) |
e9d7417b | 537 | err = add_recvbuf_big(rq, gfp); |
9ab86bbc | 538 | else |
e9d7417b | 539 | err = add_recvbuf_small(rq, gfp); |
3f2c31d9 | 540 | |
1788f495 | 541 | oom = err == -ENOMEM; |
9ed4cb07 | 542 | if (err) |
3f2c31d9 | 543 | break; |
e9d7417b | 544 | ++rq->num; |
b7dfde95 | 545 | } while (rq->vq->num_free); |
e9d7417b JW |
546 | if (unlikely(rq->num > rq->max)) |
547 | rq->max = rq->num; | |
548 | virtqueue_kick(rq->vq); | |
3161e453 | 549 | return !oom; |
3f2c31d9 MM |
550 | } |
551 | ||
18445c4d | 552 | static void skb_recv_done(struct virtqueue *rvq) |
296f96fc RR |
553 | { |
554 | struct virtnet_info *vi = rvq->vdev->priv; | |
986a4f4d | 555 | struct receive_queue *rq = &vi->rq[vq2rxq(rvq)]; |
e9d7417b | 556 | |
18445c4d | 557 | /* Schedule NAPI, Suppress further interrupts if successful. */ |
e9d7417b | 558 | if (napi_schedule_prep(&rq->napi)) { |
1915a712 | 559 | virtqueue_disable_cb(rvq); |
e9d7417b | 560 | __napi_schedule(&rq->napi); |
18445c4d | 561 | } |
296f96fc RR |
562 | } |
563 | ||
e9d7417b | 564 | static void virtnet_napi_enable(struct receive_queue *rq) |
3e9d08ec | 565 | { |
e9d7417b | 566 | napi_enable(&rq->napi); |
3e9d08ec BR |
567 | |
568 | /* If all buffers were filled by other side before we napi_enabled, we | |
569 | * won't get another interrupt, so process any outstanding packets | |
570 | * now. virtnet_poll wants re-enable the queue, so we disable here. | |
571 | * We synchronize against interrupts via NAPI_STATE_SCHED */ | |
e9d7417b JW |
572 | if (napi_schedule_prep(&rq->napi)) { |
573 | virtqueue_disable_cb(rq->vq); | |
ec13ee80 | 574 | local_bh_disable(); |
e9d7417b | 575 | __napi_schedule(&rq->napi); |
ec13ee80 | 576 | local_bh_enable(); |
3e9d08ec BR |
577 | } |
578 | } | |
579 | ||
3161e453 RR |
580 | static void refill_work(struct work_struct *work) |
581 | { | |
e9d7417b JW |
582 | struct virtnet_info *vi = |
583 | container_of(work, struct virtnet_info, refill.work); | |
3161e453 | 584 | bool still_empty; |
986a4f4d JW |
585 | int i; |
586 | ||
55257d72 | 587 | for (i = 0; i < vi->curr_queue_pairs; i++) { |
986a4f4d | 588 | struct receive_queue *rq = &vi->rq[i]; |
3161e453 | 589 | |
986a4f4d JW |
590 | napi_disable(&rq->napi); |
591 | still_empty = !try_fill_recv(rq, GFP_KERNEL); | |
592 | virtnet_napi_enable(rq); | |
3161e453 | 593 | |
986a4f4d JW |
594 | /* In theory, this can happen: if we don't get any buffers in |
595 | * we will *never* try to fill again. | |
596 | */ | |
597 | if (still_empty) | |
598 | schedule_delayed_work(&vi->refill, HZ/2); | |
599 | } | |
3161e453 RR |
600 | } |
601 | ||
296f96fc RR |
602 | static int virtnet_poll(struct napi_struct *napi, int budget) |
603 | { | |
e9d7417b JW |
604 | struct receive_queue *rq = |
605 | container_of(napi, struct receive_queue, napi); | |
606 | struct virtnet_info *vi = rq->vq->vdev->priv; | |
9ab86bbc | 607 | void *buf; |
cbdadbbf | 608 | unsigned int r, len, received = 0; |
296f96fc RR |
609 | |
610 | again: | |
611 | while (received < budget && | |
e9d7417b JW |
612 | (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) { |
613 | receive_buf(rq, buf, len); | |
614 | --rq->num; | |
296f96fc RR |
615 | received++; |
616 | } | |
617 | ||
e9d7417b JW |
618 | if (rq->num < rq->max / 2) { |
619 | if (!try_fill_recv(rq, GFP_ATOMIC)) | |
3b07e9ca | 620 | schedule_delayed_work(&vi->refill, 0); |
3161e453 | 621 | } |
296f96fc | 622 | |
8329d98e RR |
623 | /* Out of packets? */ |
624 | if (received < budget) { | |
cbdadbbf | 625 | r = virtqueue_enable_cb_prepare(rq->vq); |
288379f0 | 626 | napi_complete(napi); |
cbdadbbf | 627 | if (unlikely(virtqueue_poll(rq->vq, r)) && |
8e95a202 | 628 | napi_schedule_prep(napi)) { |
e9d7417b | 629 | virtqueue_disable_cb(rq->vq); |
288379f0 | 630 | __napi_schedule(napi); |
296f96fc | 631 | goto again; |
4265f161 | 632 | } |
296f96fc RR |
633 | } |
634 | ||
635 | return received; | |
636 | } | |
637 | ||
986a4f4d JW |
638 | static int virtnet_open(struct net_device *dev) |
639 | { | |
640 | struct virtnet_info *vi = netdev_priv(dev); | |
641 | int i; | |
642 | ||
e4166625 JW |
643 | for (i = 0; i < vi->max_queue_pairs; i++) { |
644 | if (i < vi->curr_queue_pairs) | |
645 | /* Make sure we have some buffers: if oom use wq. */ | |
646 | if (!try_fill_recv(&vi->rq[i], GFP_KERNEL)) | |
647 | schedule_delayed_work(&vi->refill, 0); | |
986a4f4d JW |
648 | virtnet_napi_enable(&vi->rq[i]); |
649 | } | |
650 | ||
651 | return 0; | |
652 | } | |
653 | ||
b7dfde95 | 654 | static void free_old_xmit_skbs(struct send_queue *sq) |
296f96fc RR |
655 | { |
656 | struct sk_buff *skb; | |
6ee57bcc | 657 | unsigned int len; |
e9d7417b | 658 | struct virtnet_info *vi = sq->vq->vdev->priv; |
58472a76 | 659 | struct virtnet_stats *stats = this_cpu_ptr(vi->stats); |
296f96fc | 660 | |
e9d7417b | 661 | while ((skb = virtqueue_get_buf(sq->vq, &len)) != NULL) { |
296f96fc | 662 | pr_debug("Sent skb %p\n", skb); |
3fa2a1df | 663 | |
83a27052 | 664 | u64_stats_update_begin(&stats->tx_syncp); |
3fa2a1df | 665 | stats->tx_bytes += skb->len; |
666 | stats->tx_packets++; | |
83a27052 | 667 | u64_stats_update_end(&stats->tx_syncp); |
3fa2a1df | 668 | |
ed79bab8 | 669 | dev_kfree_skb_any(skb); |
296f96fc RR |
670 | } |
671 | } | |
672 | ||
e9d7417b | 673 | static int xmit_skb(struct send_queue *sq, struct sk_buff *skb) |
296f96fc | 674 | { |
e7428e95 | 675 | struct skb_vnet_hdr *hdr; |
296f96fc | 676 | const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest; |
e9d7417b | 677 | struct virtnet_info *vi = sq->vq->vdev->priv; |
7bedc7dc | 678 | unsigned num_sg; |
e7428e95 MT |
679 | unsigned hdr_len; |
680 | bool can_push; | |
296f96fc | 681 | |
e174961c | 682 | pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest); |
e7428e95 MT |
683 | if (vi->mergeable_rx_bufs) |
684 | hdr_len = sizeof hdr->mhdr; | |
685 | else | |
686 | hdr_len = sizeof hdr->hdr; | |
687 | ||
688 | can_push = vi->any_header_sg && | |
689 | !((unsigned long)skb->data & (__alignof__(*hdr) - 1)) && | |
690 | !skb_header_cloned(skb) && skb_headroom(skb) >= hdr_len; | |
691 | /* Even if we can, don't push here yet as this would skew | |
692 | * csum_start offset below. */ | |
693 | if (can_push) | |
694 | hdr = (struct skb_vnet_hdr *)(skb->data - hdr_len); | |
695 | else | |
696 | hdr = skb_vnet_hdr(skb); | |
296f96fc | 697 | |
296f96fc | 698 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
b3f24698 | 699 | hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; |
55508d60 | 700 | hdr->hdr.csum_start = skb_checksum_start_offset(skb); |
b3f24698 | 701 | hdr->hdr.csum_offset = skb->csum_offset; |
296f96fc | 702 | } else { |
b3f24698 RR |
703 | hdr->hdr.flags = 0; |
704 | hdr->hdr.csum_offset = hdr->hdr.csum_start = 0; | |
296f96fc RR |
705 | } |
706 | ||
707 | if (skb_is_gso(skb)) { | |
b3f24698 RR |
708 | hdr->hdr.hdr_len = skb_headlen(skb); |
709 | hdr->hdr.gso_size = skb_shinfo(skb)->gso_size; | |
34a48579 | 710 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) |
b3f24698 | 711 | hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4; |
296f96fc | 712 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) |
b3f24698 | 713 | hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6; |
296f96fc | 714 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP) |
b3f24698 | 715 | hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP; |
296f96fc RR |
716 | else |
717 | BUG(); | |
34a48579 | 718 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN) |
b3f24698 | 719 | hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN; |
296f96fc | 720 | } else { |
b3f24698 RR |
721 | hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE; |
722 | hdr->hdr.gso_size = hdr->hdr.hdr_len = 0; | |
296f96fc RR |
723 | } |
724 | ||
3f2c31d9 | 725 | if (vi->mergeable_rx_bufs) |
e7428e95 | 726 | hdr->mhdr.num_buffers = 0; |
3f2c31d9 | 727 | |
e7428e95 MT |
728 | if (can_push) { |
729 | __skb_push(skb, hdr_len); | |
730 | num_sg = skb_to_sgvec(skb, sq->sg, 0, skb->len); | |
731 | /* Pull header back to avoid skew in tx bytes calculations. */ | |
732 | __skb_pull(skb, hdr_len); | |
733 | } else { | |
734 | sg_set_buf(sq->sg, hdr, hdr_len); | |
735 | num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len) + 1; | |
736 | } | |
9dc7b9e4 | 737 | return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC); |
11a3a154 RR |
738 | } |
739 | ||
424efe9c | 740 | static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev) |
99ffc696 RR |
741 | { |
742 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d JW |
743 | int qnum = skb_get_queue_mapping(skb); |
744 | struct send_queue *sq = &vi->sq[qnum]; | |
9ed4cb07 | 745 | int err; |
2cb9c6ba | 746 | |
2cb9c6ba | 747 | /* Free up any pending old buffers before queueing new ones. */ |
e9d7417b | 748 | free_old_xmit_skbs(sq); |
99ffc696 | 749 | |
03f191ba | 750 | /* Try to transmit */ |
b7dfde95 | 751 | err = xmit_skb(sq, skb); |
48925e37 | 752 | |
9ed4cb07 | 753 | /* This should not happen! */ |
0e3daa64 | 754 | if (unlikely(err)) { |
9ed4cb07 RR |
755 | dev->stats.tx_fifo_errors++; |
756 | if (net_ratelimit()) | |
757 | dev_warn(&dev->dev, | |
b7dfde95 | 758 | "Unexpected TXQ (%d) queue failure: %d\n", qnum, err); |
58eba97d RR |
759 | dev->stats.tx_dropped++; |
760 | kfree_skb(skb); | |
761 | return NETDEV_TX_OK; | |
296f96fc | 762 | } |
e9d7417b | 763 | virtqueue_kick(sq->vq); |
03f191ba | 764 | |
48925e37 RR |
765 | /* Don't wait up for transmitted skbs to be freed. */ |
766 | skb_orphan(skb); | |
767 | nf_reset(skb); | |
768 | ||
769 | /* Apparently nice girls don't return TX_BUSY; stop the queue | |
770 | * before it gets out of hand. Naturally, this wastes entries. */ | |
b7dfde95 | 771 | if (sq->vq->num_free < 2+MAX_SKB_FRAGS) { |
986a4f4d | 772 | netif_stop_subqueue(dev, qnum); |
e9d7417b | 773 | if (unlikely(!virtqueue_enable_cb_delayed(sq->vq))) { |
48925e37 | 774 | /* More just got used, free them then recheck. */ |
b7dfde95 LT |
775 | free_old_xmit_skbs(sq); |
776 | if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) { | |
986a4f4d | 777 | netif_start_subqueue(dev, qnum); |
e9d7417b | 778 | virtqueue_disable_cb(sq->vq); |
48925e37 RR |
779 | } |
780 | } | |
99ffc696 | 781 | } |
48925e37 RR |
782 | |
783 | return NETDEV_TX_OK; | |
296f96fc RR |
784 | } |
785 | ||
40cbfc37 AK |
786 | /* |
787 | * Send command via the control virtqueue and check status. Commands | |
788 | * supported by the hypervisor, as indicated by feature bits, should | |
789 | * never fail unless improperly formated. | |
790 | */ | |
791 | static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd, | |
f7bc9594 RR |
792 | struct scatterlist *out, |
793 | struct scatterlist *in) | |
40cbfc37 | 794 | { |
f7bc9594 | 795 | struct scatterlist *sgs[4], hdr, stat; |
40cbfc37 AK |
796 | struct virtio_net_ctrl_hdr ctrl; |
797 | virtio_net_ctrl_ack status = ~0; | |
f7bc9594 | 798 | unsigned out_num = 0, in_num = 0, tmp; |
40cbfc37 AK |
799 | |
800 | /* Caller should know better */ | |
f7bc9594 | 801 | BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)); |
40cbfc37 AK |
802 | |
803 | ctrl.class = class; | |
804 | ctrl.cmd = cmd; | |
f7bc9594 RR |
805 | /* Add header */ |
806 | sg_init_one(&hdr, &ctrl, sizeof(ctrl)); | |
807 | sgs[out_num++] = &hdr; | |
40cbfc37 | 808 | |
f7bc9594 RR |
809 | if (out) |
810 | sgs[out_num++] = out; | |
811 | if (in) | |
812 | sgs[out_num + in_num++] = in; | |
40cbfc37 | 813 | |
f7bc9594 RR |
814 | /* Add return status. */ |
815 | sg_init_one(&stat, &status, sizeof(status)); | |
816 | sgs[out_num + in_num++] = &stat; | |
40cbfc37 | 817 | |
f7bc9594 RR |
818 | BUG_ON(out_num + in_num > ARRAY_SIZE(sgs)); |
819 | BUG_ON(virtqueue_add_sgs(vi->cvq, sgs, out_num, in_num, vi, GFP_ATOMIC) | |
820 | < 0); | |
40cbfc37 AK |
821 | |
822 | virtqueue_kick(vi->cvq); | |
823 | ||
824 | /* Spin for a response, the kick causes an ioport write, trapping | |
825 | * into the hypervisor, so the request should be handled immediately. | |
826 | */ | |
827 | while (!virtqueue_get_buf(vi->cvq, &tmp)) | |
828 | cpu_relax(); | |
829 | ||
830 | return status == VIRTIO_NET_OK; | |
831 | } | |
832 | ||
9c46f6d4 AW |
833 | static int virtnet_set_mac_address(struct net_device *dev, void *p) |
834 | { | |
835 | struct virtnet_info *vi = netdev_priv(dev); | |
836 | struct virtio_device *vdev = vi->vdev; | |
f2f2c8b4 | 837 | int ret; |
7e58d5ae AK |
838 | struct sockaddr *addr = p; |
839 | struct scatterlist sg; | |
9c46f6d4 | 840 | |
7e58d5ae | 841 | ret = eth_prepare_mac_addr_change(dev, p); |
f2f2c8b4 JP |
842 | if (ret) |
843 | return ret; | |
9c46f6d4 | 844 | |
7e58d5ae AK |
845 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) { |
846 | sg_init_one(&sg, addr->sa_data, dev->addr_len); | |
847 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC, | |
848 | VIRTIO_NET_CTRL_MAC_ADDR_SET, | |
f7bc9594 | 849 | &sg, NULL)) { |
7e58d5ae AK |
850 | dev_warn(&vdev->dev, |
851 | "Failed to set mac address by vq command.\n"); | |
852 | return -EINVAL; | |
853 | } | |
854 | } else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) { | |
62994b2d | 855 | vdev->config->set(vdev, offsetof(struct virtio_net_config, mac), |
7e58d5ae AK |
856 | addr->sa_data, dev->addr_len); |
857 | } | |
858 | ||
859 | eth_commit_mac_addr_change(dev, p); | |
9c46f6d4 AW |
860 | |
861 | return 0; | |
862 | } | |
863 | ||
3fa2a1df | 864 | static struct rtnl_link_stats64 *virtnet_stats(struct net_device *dev, |
865 | struct rtnl_link_stats64 *tot) | |
866 | { | |
867 | struct virtnet_info *vi = netdev_priv(dev); | |
868 | int cpu; | |
869 | unsigned int start; | |
870 | ||
871 | for_each_possible_cpu(cpu) { | |
58472a76 | 872 | struct virtnet_stats *stats = per_cpu_ptr(vi->stats, cpu); |
3fa2a1df | 873 | u64 tpackets, tbytes, rpackets, rbytes; |
874 | ||
875 | do { | |
e3906486 | 876 | start = u64_stats_fetch_begin_bh(&stats->tx_syncp); |
3fa2a1df | 877 | tpackets = stats->tx_packets; |
878 | tbytes = stats->tx_bytes; | |
e3906486 | 879 | } while (u64_stats_fetch_retry_bh(&stats->tx_syncp, start)); |
83a27052 ED |
880 | |
881 | do { | |
e3906486 | 882 | start = u64_stats_fetch_begin_bh(&stats->rx_syncp); |
3fa2a1df | 883 | rpackets = stats->rx_packets; |
884 | rbytes = stats->rx_bytes; | |
e3906486 | 885 | } while (u64_stats_fetch_retry_bh(&stats->rx_syncp, start)); |
3fa2a1df | 886 | |
887 | tot->rx_packets += rpackets; | |
888 | tot->tx_packets += tpackets; | |
889 | tot->rx_bytes += rbytes; | |
890 | tot->tx_bytes += tbytes; | |
891 | } | |
892 | ||
893 | tot->tx_dropped = dev->stats.tx_dropped; | |
021ac8d3 | 894 | tot->tx_fifo_errors = dev->stats.tx_fifo_errors; |
3fa2a1df | 895 | tot->rx_dropped = dev->stats.rx_dropped; |
896 | tot->rx_length_errors = dev->stats.rx_length_errors; | |
897 | tot->rx_frame_errors = dev->stats.rx_frame_errors; | |
898 | ||
899 | return tot; | |
900 | } | |
901 | ||
da74e89d AS |
902 | #ifdef CONFIG_NET_POLL_CONTROLLER |
903 | static void virtnet_netpoll(struct net_device *dev) | |
904 | { | |
905 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d | 906 | int i; |
da74e89d | 907 | |
986a4f4d JW |
908 | for (i = 0; i < vi->curr_queue_pairs; i++) |
909 | napi_schedule(&vi->rq[i].napi); | |
da74e89d AS |
910 | } |
911 | #endif | |
912 | ||
586d17c5 JW |
913 | static void virtnet_ack_link_announce(struct virtnet_info *vi) |
914 | { | |
915 | rtnl_lock(); | |
916 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE, | |
f7bc9594 | 917 | VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL, NULL)) |
586d17c5 JW |
918 | dev_warn(&vi->dev->dev, "Failed to ack link announce.\n"); |
919 | rtnl_unlock(); | |
920 | } | |
921 | ||
986a4f4d JW |
922 | static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs) |
923 | { | |
924 | struct scatterlist sg; | |
925 | struct virtio_net_ctrl_mq s; | |
926 | struct net_device *dev = vi->dev; | |
927 | ||
928 | if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ)) | |
929 | return 0; | |
930 | ||
931 | s.virtqueue_pairs = queue_pairs; | |
932 | sg_init_one(&sg, &s, sizeof(s)); | |
933 | ||
934 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ, | |
f7bc9594 | 935 | VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg, NULL)) { |
986a4f4d JW |
936 | dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n", |
937 | queue_pairs); | |
938 | return -EINVAL; | |
55257d72 | 939 | } else { |
986a4f4d | 940 | vi->curr_queue_pairs = queue_pairs; |
9b9cd802 | 941 | schedule_delayed_work(&vi->refill, 0); |
55257d72 | 942 | } |
986a4f4d JW |
943 | |
944 | return 0; | |
945 | } | |
946 | ||
296f96fc RR |
947 | static int virtnet_close(struct net_device *dev) |
948 | { | |
949 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d | 950 | int i; |
296f96fc | 951 | |
b2baed69 RR |
952 | /* Make sure refill_work doesn't re-enable napi! */ |
953 | cancel_delayed_work_sync(&vi->refill); | |
986a4f4d JW |
954 | |
955 | for (i = 0; i < vi->max_queue_pairs; i++) | |
956 | napi_disable(&vi->rq[i].napi); | |
296f96fc | 957 | |
296f96fc RR |
958 | return 0; |
959 | } | |
960 | ||
2af7698e AW |
961 | static void virtnet_set_rx_mode(struct net_device *dev) |
962 | { | |
963 | struct virtnet_info *vi = netdev_priv(dev); | |
f565a7c2 | 964 | struct scatterlist sg[2]; |
2af7698e | 965 | u8 promisc, allmulti; |
f565a7c2 | 966 | struct virtio_net_ctrl_mac *mac_data; |
ccffad25 | 967 | struct netdev_hw_addr *ha; |
32e7bfc4 | 968 | int uc_count; |
4cd24eaf | 969 | int mc_count; |
f565a7c2 AW |
970 | void *buf; |
971 | int i; | |
2af7698e AW |
972 | |
973 | /* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */ | |
974 | if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX)) | |
975 | return; | |
976 | ||
f565a7c2 AW |
977 | promisc = ((dev->flags & IFF_PROMISC) != 0); |
978 | allmulti = ((dev->flags & IFF_ALLMULTI) != 0); | |
2af7698e | 979 | |
23e258e1 | 980 | sg_init_one(sg, &promisc, sizeof(promisc)); |
2af7698e AW |
981 | |
982 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, | |
983 | VIRTIO_NET_CTRL_RX_PROMISC, | |
f7bc9594 | 984 | sg, NULL)) |
2af7698e AW |
985 | dev_warn(&dev->dev, "Failed to %sable promisc mode.\n", |
986 | promisc ? "en" : "dis"); | |
987 | ||
23e258e1 | 988 | sg_init_one(sg, &allmulti, sizeof(allmulti)); |
2af7698e AW |
989 | |
990 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, | |
991 | VIRTIO_NET_CTRL_RX_ALLMULTI, | |
f7bc9594 | 992 | sg, NULL)) |
2af7698e AW |
993 | dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n", |
994 | allmulti ? "en" : "dis"); | |
f565a7c2 | 995 | |
32e7bfc4 | 996 | uc_count = netdev_uc_count(dev); |
4cd24eaf | 997 | mc_count = netdev_mc_count(dev); |
f565a7c2 | 998 | /* MAC filter - use one buffer for both lists */ |
4cd24eaf JP |
999 | buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) + |
1000 | (2 * sizeof(mac_data->entries)), GFP_ATOMIC); | |
1001 | mac_data = buf; | |
e68ed8f0 | 1002 | if (!buf) |
f565a7c2 | 1003 | return; |
f565a7c2 | 1004 | |
23e258e1 AW |
1005 | sg_init_table(sg, 2); |
1006 | ||
f565a7c2 | 1007 | /* Store the unicast list and count in the front of the buffer */ |
32e7bfc4 | 1008 | mac_data->entries = uc_count; |
ccffad25 | 1009 | i = 0; |
32e7bfc4 | 1010 | netdev_for_each_uc_addr(ha, dev) |
ccffad25 | 1011 | memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); |
f565a7c2 AW |
1012 | |
1013 | sg_set_buf(&sg[0], mac_data, | |
32e7bfc4 | 1014 | sizeof(mac_data->entries) + (uc_count * ETH_ALEN)); |
f565a7c2 AW |
1015 | |
1016 | /* multicast list and count fill the end */ | |
32e7bfc4 | 1017 | mac_data = (void *)&mac_data->macs[uc_count][0]; |
f565a7c2 | 1018 | |
4cd24eaf | 1019 | mac_data->entries = mc_count; |
567ec874 | 1020 | i = 0; |
22bedad3 JP |
1021 | netdev_for_each_mc_addr(ha, dev) |
1022 | memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); | |
f565a7c2 AW |
1023 | |
1024 | sg_set_buf(&sg[1], mac_data, | |
4cd24eaf | 1025 | sizeof(mac_data->entries) + (mc_count * ETH_ALEN)); |
f565a7c2 AW |
1026 | |
1027 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC, | |
1028 | VIRTIO_NET_CTRL_MAC_TABLE_SET, | |
f7bc9594 | 1029 | sg, NULL)) |
f565a7c2 AW |
1030 | dev_warn(&dev->dev, "Failed to set MAC fitler table.\n"); |
1031 | ||
1032 | kfree(buf); | |
2af7698e AW |
1033 | } |
1034 | ||
80d5c368 PM |
1035 | static int virtnet_vlan_rx_add_vid(struct net_device *dev, |
1036 | __be16 proto, u16 vid) | |
0bde9569 AW |
1037 | { |
1038 | struct virtnet_info *vi = netdev_priv(dev); | |
1039 | struct scatterlist sg; | |
1040 | ||
23e258e1 | 1041 | sg_init_one(&sg, &vid, sizeof(vid)); |
0bde9569 AW |
1042 | |
1043 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, | |
f7bc9594 | 1044 | VIRTIO_NET_CTRL_VLAN_ADD, &sg, NULL)) |
0bde9569 | 1045 | dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid); |
8e586137 | 1046 | return 0; |
0bde9569 AW |
1047 | } |
1048 | ||
80d5c368 PM |
1049 | static int virtnet_vlan_rx_kill_vid(struct net_device *dev, |
1050 | __be16 proto, u16 vid) | |
0bde9569 AW |
1051 | { |
1052 | struct virtnet_info *vi = netdev_priv(dev); | |
1053 | struct scatterlist sg; | |
1054 | ||
23e258e1 | 1055 | sg_init_one(&sg, &vid, sizeof(vid)); |
0bde9569 AW |
1056 | |
1057 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, | |
f7bc9594 | 1058 | VIRTIO_NET_CTRL_VLAN_DEL, &sg, NULL)) |
0bde9569 | 1059 | dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid); |
8e586137 | 1060 | return 0; |
0bde9569 AW |
1061 | } |
1062 | ||
8898c21c | 1063 | static void virtnet_clean_affinity(struct virtnet_info *vi, long hcpu) |
986a4f4d JW |
1064 | { |
1065 | int i; | |
47be2479 | 1066 | int cpu; |
986a4f4d | 1067 | |
8898c21c WG |
1068 | if (vi->affinity_hint_set) { |
1069 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
47be2479 WG |
1070 | virtqueue_set_affinity(vi->rq[i].vq, -1); |
1071 | virtqueue_set_affinity(vi->sq[i].vq, -1); | |
1072 | } | |
1073 | ||
8898c21c WG |
1074 | vi->affinity_hint_set = false; |
1075 | } | |
1076 | ||
1077 | i = 0; | |
1078 | for_each_online_cpu(cpu) { | |
1079 | if (cpu == hcpu) { | |
1080 | *per_cpu_ptr(vi->vq_index, cpu) = -1; | |
1081 | } else { | |
47be2479 WG |
1082 | *per_cpu_ptr(vi->vq_index, cpu) = |
1083 | ++i % vi->curr_queue_pairs; | |
8898c21c WG |
1084 | } |
1085 | } | |
1086 | } | |
47be2479 | 1087 | |
8898c21c WG |
1088 | static void virtnet_set_affinity(struct virtnet_info *vi) |
1089 | { | |
1090 | int i; | |
1091 | int cpu; | |
986a4f4d JW |
1092 | |
1093 | /* In multiqueue mode, when the number of cpu is equal to the number of | |
1094 | * queue pairs, we let the queue pairs to be private to one cpu by | |
1095 | * setting the affinity hint to eliminate the contention. | |
1096 | */ | |
8898c21c WG |
1097 | if (vi->curr_queue_pairs == 1 || |
1098 | vi->max_queue_pairs != num_online_cpus()) { | |
1099 | virtnet_clean_affinity(vi, -1); | |
1100 | return; | |
986a4f4d JW |
1101 | } |
1102 | ||
8898c21c WG |
1103 | i = 0; |
1104 | for_each_online_cpu(cpu) { | |
986a4f4d JW |
1105 | virtqueue_set_affinity(vi->rq[i].vq, cpu); |
1106 | virtqueue_set_affinity(vi->sq[i].vq, cpu); | |
8898c21c WG |
1107 | *per_cpu_ptr(vi->vq_index, cpu) = i; |
1108 | i++; | |
986a4f4d JW |
1109 | } |
1110 | ||
8898c21c | 1111 | vi->affinity_hint_set = true; |
986a4f4d JW |
1112 | } |
1113 | ||
8de4b2f3 WG |
1114 | static int virtnet_cpu_callback(struct notifier_block *nfb, |
1115 | unsigned long action, void *hcpu) | |
1116 | { | |
1117 | struct virtnet_info *vi = container_of(nfb, struct virtnet_info, nb); | |
1118 | ||
3ab098df JW |
1119 | mutex_lock(&vi->config_lock); |
1120 | ||
1121 | if (!vi->config_enable) | |
1122 | goto done; | |
1123 | ||
8de4b2f3 WG |
1124 | switch(action & ~CPU_TASKS_FROZEN) { |
1125 | case CPU_ONLINE: | |
1126 | case CPU_DOWN_FAILED: | |
1127 | case CPU_DEAD: | |
1128 | virtnet_set_affinity(vi); | |
1129 | break; | |
1130 | case CPU_DOWN_PREPARE: | |
1131 | virtnet_clean_affinity(vi, (long)hcpu); | |
1132 | break; | |
1133 | default: | |
1134 | break; | |
1135 | } | |
3ab098df JW |
1136 | |
1137 | done: | |
1138 | mutex_unlock(&vi->config_lock); | |
8de4b2f3 | 1139 | return NOTIFY_OK; |
986a4f4d JW |
1140 | } |
1141 | ||
8f9f4668 RJ |
1142 | static void virtnet_get_ringparam(struct net_device *dev, |
1143 | struct ethtool_ringparam *ring) | |
1144 | { | |
1145 | struct virtnet_info *vi = netdev_priv(dev); | |
1146 | ||
986a4f4d JW |
1147 | ring->rx_max_pending = virtqueue_get_vring_size(vi->rq[0].vq); |
1148 | ring->tx_max_pending = virtqueue_get_vring_size(vi->sq[0].vq); | |
8f9f4668 RJ |
1149 | ring->rx_pending = ring->rx_max_pending; |
1150 | ring->tx_pending = ring->tx_max_pending; | |
8f9f4668 RJ |
1151 | } |
1152 | ||
66846048 RJ |
1153 | |
1154 | static void virtnet_get_drvinfo(struct net_device *dev, | |
1155 | struct ethtool_drvinfo *info) | |
1156 | { | |
1157 | struct virtnet_info *vi = netdev_priv(dev); | |
1158 | struct virtio_device *vdev = vi->vdev; | |
1159 | ||
1160 | strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver)); | |
1161 | strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version)); | |
1162 | strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info)); | |
1163 | ||
1164 | } | |
1165 | ||
d73bcd2c JW |
1166 | /* TODO: Eliminate OOO packets during switching */ |
1167 | static int virtnet_set_channels(struct net_device *dev, | |
1168 | struct ethtool_channels *channels) | |
1169 | { | |
1170 | struct virtnet_info *vi = netdev_priv(dev); | |
1171 | u16 queue_pairs = channels->combined_count; | |
1172 | int err; | |
1173 | ||
1174 | /* We don't support separate rx/tx channels. | |
1175 | * We don't allow setting 'other' channels. | |
1176 | */ | |
1177 | if (channels->rx_count || channels->tx_count || channels->other_count) | |
1178 | return -EINVAL; | |
1179 | ||
1180 | if (queue_pairs > vi->max_queue_pairs) | |
1181 | return -EINVAL; | |
1182 | ||
47be2479 | 1183 | get_online_cpus(); |
d73bcd2c JW |
1184 | err = virtnet_set_queues(vi, queue_pairs); |
1185 | if (!err) { | |
1186 | netif_set_real_num_tx_queues(dev, queue_pairs); | |
1187 | netif_set_real_num_rx_queues(dev, queue_pairs); | |
1188 | ||
8898c21c | 1189 | virtnet_set_affinity(vi); |
d73bcd2c | 1190 | } |
47be2479 | 1191 | put_online_cpus(); |
d73bcd2c JW |
1192 | |
1193 | return err; | |
1194 | } | |
1195 | ||
1196 | static void virtnet_get_channels(struct net_device *dev, | |
1197 | struct ethtool_channels *channels) | |
1198 | { | |
1199 | struct virtnet_info *vi = netdev_priv(dev); | |
1200 | ||
1201 | channels->combined_count = vi->curr_queue_pairs; | |
1202 | channels->max_combined = vi->max_queue_pairs; | |
1203 | channels->max_other = 0; | |
1204 | channels->rx_count = 0; | |
1205 | channels->tx_count = 0; | |
1206 | channels->other_count = 0; | |
1207 | } | |
1208 | ||
0fc0b732 | 1209 | static const struct ethtool_ops virtnet_ethtool_ops = { |
66846048 | 1210 | .get_drvinfo = virtnet_get_drvinfo, |
9f4d26d0 | 1211 | .get_link = ethtool_op_get_link, |
8f9f4668 | 1212 | .get_ringparam = virtnet_get_ringparam, |
d73bcd2c JW |
1213 | .set_channels = virtnet_set_channels, |
1214 | .get_channels = virtnet_get_channels, | |
a9ea3fc6 HX |
1215 | }; |
1216 | ||
39da5814 MM |
1217 | #define MIN_MTU 68 |
1218 | #define MAX_MTU 65535 | |
1219 | ||
1220 | static int virtnet_change_mtu(struct net_device *dev, int new_mtu) | |
1221 | { | |
1222 | if (new_mtu < MIN_MTU || new_mtu > MAX_MTU) | |
1223 | return -EINVAL; | |
1224 | dev->mtu = new_mtu; | |
1225 | return 0; | |
1226 | } | |
1227 | ||
986a4f4d JW |
1228 | /* To avoid contending a lock hold by a vcpu who would exit to host, select the |
1229 | * txq based on the processor id. | |
986a4f4d JW |
1230 | */ |
1231 | static u16 virtnet_select_queue(struct net_device *dev, struct sk_buff *skb) | |
1232 | { | |
47be2479 WG |
1233 | int txq; |
1234 | struct virtnet_info *vi = netdev_priv(dev); | |
1235 | ||
1236 | if (skb_rx_queue_recorded(skb)) { | |
1237 | txq = skb_get_rx_queue(skb); | |
1238 | } else { | |
1239 | txq = *__this_cpu_ptr(vi->vq_index); | |
1240 | if (txq == -1) | |
1241 | txq = 0; | |
1242 | } | |
986a4f4d JW |
1243 | |
1244 | while (unlikely(txq >= dev->real_num_tx_queues)) | |
1245 | txq -= dev->real_num_tx_queues; | |
1246 | ||
1247 | return txq; | |
1248 | } | |
1249 | ||
76288b4e SH |
1250 | static const struct net_device_ops virtnet_netdev = { |
1251 | .ndo_open = virtnet_open, | |
1252 | .ndo_stop = virtnet_close, | |
1253 | .ndo_start_xmit = start_xmit, | |
1254 | .ndo_validate_addr = eth_validate_addr, | |
9c46f6d4 | 1255 | .ndo_set_mac_address = virtnet_set_mac_address, |
2af7698e | 1256 | .ndo_set_rx_mode = virtnet_set_rx_mode, |
76288b4e | 1257 | .ndo_change_mtu = virtnet_change_mtu, |
3fa2a1df | 1258 | .ndo_get_stats64 = virtnet_stats, |
1824a989 AW |
1259 | .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid, |
1260 | .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid, | |
986a4f4d | 1261 | .ndo_select_queue = virtnet_select_queue, |
76288b4e SH |
1262 | #ifdef CONFIG_NET_POLL_CONTROLLER |
1263 | .ndo_poll_controller = virtnet_netpoll, | |
1264 | #endif | |
1265 | }; | |
1266 | ||
586d17c5 | 1267 | static void virtnet_config_changed_work(struct work_struct *work) |
9f4d26d0 | 1268 | { |
586d17c5 JW |
1269 | struct virtnet_info *vi = |
1270 | container_of(work, struct virtnet_info, config_work); | |
9f4d26d0 MM |
1271 | u16 v; |
1272 | ||
586d17c5 JW |
1273 | mutex_lock(&vi->config_lock); |
1274 | if (!vi->config_enable) | |
1275 | goto done; | |
1276 | ||
77dd7693 | 1277 | if (virtio_config_val(vi->vdev, VIRTIO_NET_F_STATUS, |
9f4d26d0 | 1278 | offsetof(struct virtio_net_config, status), |
77dd7693 | 1279 | &v) < 0) |
586d17c5 JW |
1280 | goto done; |
1281 | ||
1282 | if (v & VIRTIO_NET_S_ANNOUNCE) { | |
ee89bab1 | 1283 | netdev_notify_peers(vi->dev); |
586d17c5 JW |
1284 | virtnet_ack_link_announce(vi); |
1285 | } | |
9f4d26d0 MM |
1286 | |
1287 | /* Ignore unknown (future) status bits */ | |
1288 | v &= VIRTIO_NET_S_LINK_UP; | |
1289 | ||
1290 | if (vi->status == v) | |
586d17c5 | 1291 | goto done; |
9f4d26d0 MM |
1292 | |
1293 | vi->status = v; | |
1294 | ||
1295 | if (vi->status & VIRTIO_NET_S_LINK_UP) { | |
1296 | netif_carrier_on(vi->dev); | |
986a4f4d | 1297 | netif_tx_wake_all_queues(vi->dev); |
9f4d26d0 MM |
1298 | } else { |
1299 | netif_carrier_off(vi->dev); | |
986a4f4d | 1300 | netif_tx_stop_all_queues(vi->dev); |
9f4d26d0 | 1301 | } |
586d17c5 JW |
1302 | done: |
1303 | mutex_unlock(&vi->config_lock); | |
9f4d26d0 MM |
1304 | } |
1305 | ||
1306 | static void virtnet_config_changed(struct virtio_device *vdev) | |
1307 | { | |
1308 | struct virtnet_info *vi = vdev->priv; | |
1309 | ||
3b07e9ca | 1310 | schedule_work(&vi->config_work); |
9f4d26d0 MM |
1311 | } |
1312 | ||
986a4f4d JW |
1313 | static void virtnet_free_queues(struct virtnet_info *vi) |
1314 | { | |
1315 | kfree(vi->rq); | |
1316 | kfree(vi->sq); | |
1317 | } | |
1318 | ||
1319 | static void free_receive_bufs(struct virtnet_info *vi) | |
1320 | { | |
1321 | int i; | |
1322 | ||
1323 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
1324 | while (vi->rq[i].pages) | |
1325 | __free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0); | |
1326 | } | |
1327 | } | |
1328 | ||
1329 | static void free_unused_bufs(struct virtnet_info *vi) | |
1330 | { | |
1331 | void *buf; | |
1332 | int i; | |
1333 | ||
1334 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
1335 | struct virtqueue *vq = vi->sq[i].vq; | |
1336 | while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) | |
1337 | dev_kfree_skb(buf); | |
1338 | } | |
1339 | ||
1340 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
1341 | struct virtqueue *vq = vi->rq[i].vq; | |
1342 | ||
1343 | while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) { | |
1344 | if (vi->mergeable_rx_bufs || vi->big_packets) | |
1345 | give_pages(&vi->rq[i], buf); | |
1346 | else | |
1347 | dev_kfree_skb(buf); | |
1348 | --vi->rq[i].num; | |
1349 | } | |
1350 | BUG_ON(vi->rq[i].num != 0); | |
1351 | } | |
1352 | } | |
1353 | ||
e9d7417b JW |
1354 | static void virtnet_del_vqs(struct virtnet_info *vi) |
1355 | { | |
1356 | struct virtio_device *vdev = vi->vdev; | |
1357 | ||
8898c21c | 1358 | virtnet_clean_affinity(vi, -1); |
986a4f4d | 1359 | |
e9d7417b | 1360 | vdev->config->del_vqs(vdev); |
986a4f4d JW |
1361 | |
1362 | virtnet_free_queues(vi); | |
e9d7417b JW |
1363 | } |
1364 | ||
986a4f4d | 1365 | static int virtnet_find_vqs(struct virtnet_info *vi) |
3f9c10b0 | 1366 | { |
986a4f4d JW |
1367 | vq_callback_t **callbacks; |
1368 | struct virtqueue **vqs; | |
1369 | int ret = -ENOMEM; | |
1370 | int i, total_vqs; | |
1371 | const char **names; | |
1372 | ||
1373 | /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by | |
1374 | * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by | |
1375 | * possible control vq. | |
1376 | */ | |
1377 | total_vqs = vi->max_queue_pairs * 2 + | |
1378 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ); | |
1379 | ||
1380 | /* Allocate space for find_vqs parameters */ | |
1381 | vqs = kzalloc(total_vqs * sizeof(*vqs), GFP_KERNEL); | |
1382 | if (!vqs) | |
1383 | goto err_vq; | |
1384 | callbacks = kmalloc(total_vqs * sizeof(*callbacks), GFP_KERNEL); | |
1385 | if (!callbacks) | |
1386 | goto err_callback; | |
1387 | names = kmalloc(total_vqs * sizeof(*names), GFP_KERNEL); | |
1388 | if (!names) | |
1389 | goto err_names; | |
1390 | ||
1391 | /* Parameters for control virtqueue, if any */ | |
1392 | if (vi->has_cvq) { | |
1393 | callbacks[total_vqs - 1] = NULL; | |
1394 | names[total_vqs - 1] = "control"; | |
1395 | } | |
3f9c10b0 | 1396 | |
986a4f4d JW |
1397 | /* Allocate/initialize parameters for send/receive virtqueues */ |
1398 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
1399 | callbacks[rxq2vq(i)] = skb_recv_done; | |
1400 | callbacks[txq2vq(i)] = skb_xmit_done; | |
1401 | sprintf(vi->rq[i].name, "input.%d", i); | |
1402 | sprintf(vi->sq[i].name, "output.%d", i); | |
1403 | names[rxq2vq(i)] = vi->rq[i].name; | |
1404 | names[txq2vq(i)] = vi->sq[i].name; | |
1405 | } | |
3f9c10b0 | 1406 | |
986a4f4d JW |
1407 | ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs, callbacks, |
1408 | names); | |
1409 | if (ret) | |
1410 | goto err_find; | |
3f9c10b0 | 1411 | |
986a4f4d JW |
1412 | if (vi->has_cvq) { |
1413 | vi->cvq = vqs[total_vqs - 1]; | |
3f9c10b0 | 1414 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN)) |
f646968f | 1415 | vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER; |
3f9c10b0 | 1416 | } |
986a4f4d JW |
1417 | |
1418 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
1419 | vi->rq[i].vq = vqs[rxq2vq(i)]; | |
1420 | vi->sq[i].vq = vqs[txq2vq(i)]; | |
1421 | } | |
1422 | ||
1423 | kfree(names); | |
1424 | kfree(callbacks); | |
1425 | kfree(vqs); | |
1426 | ||
3f9c10b0 | 1427 | return 0; |
986a4f4d JW |
1428 | |
1429 | err_find: | |
1430 | kfree(names); | |
1431 | err_names: | |
1432 | kfree(callbacks); | |
1433 | err_callback: | |
1434 | kfree(vqs); | |
1435 | err_vq: | |
1436 | return ret; | |
1437 | } | |
1438 | ||
1439 | static int virtnet_alloc_queues(struct virtnet_info *vi) | |
1440 | { | |
1441 | int i; | |
1442 | ||
1443 | vi->sq = kzalloc(sizeof(*vi->sq) * vi->max_queue_pairs, GFP_KERNEL); | |
1444 | if (!vi->sq) | |
1445 | goto err_sq; | |
1446 | vi->rq = kzalloc(sizeof(*vi->rq) * vi->max_queue_pairs, GFP_KERNEL); | |
008d4278 | 1447 | if (!vi->rq) |
986a4f4d JW |
1448 | goto err_rq; |
1449 | ||
1450 | INIT_DELAYED_WORK(&vi->refill, refill_work); | |
1451 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
1452 | vi->rq[i].pages = NULL; | |
1453 | netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll, | |
1454 | napi_weight); | |
1455 | ||
1456 | sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg)); | |
1457 | sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg)); | |
1458 | } | |
1459 | ||
1460 | return 0; | |
1461 | ||
1462 | err_rq: | |
1463 | kfree(vi->sq); | |
1464 | err_sq: | |
1465 | return -ENOMEM; | |
1466 | } | |
1467 | ||
1468 | static int init_vqs(struct virtnet_info *vi) | |
1469 | { | |
1470 | int ret; | |
1471 | ||
1472 | /* Allocate send & receive queues */ | |
1473 | ret = virtnet_alloc_queues(vi); | |
1474 | if (ret) | |
1475 | goto err; | |
1476 | ||
1477 | ret = virtnet_find_vqs(vi); | |
1478 | if (ret) | |
1479 | goto err_free; | |
1480 | ||
47be2479 | 1481 | get_online_cpus(); |
8898c21c | 1482 | virtnet_set_affinity(vi); |
47be2479 WG |
1483 | put_online_cpus(); |
1484 | ||
986a4f4d JW |
1485 | return 0; |
1486 | ||
1487 | err_free: | |
1488 | virtnet_free_queues(vi); | |
1489 | err: | |
1490 | return ret; | |
3f9c10b0 AS |
1491 | } |
1492 | ||
296f96fc RR |
1493 | static int virtnet_probe(struct virtio_device *vdev) |
1494 | { | |
986a4f4d | 1495 | int i, err; |
296f96fc RR |
1496 | struct net_device *dev; |
1497 | struct virtnet_info *vi; | |
986a4f4d JW |
1498 | u16 max_queue_pairs; |
1499 | ||
1500 | /* Find if host supports multiqueue virtio_net device */ | |
1501 | err = virtio_config_val(vdev, VIRTIO_NET_F_MQ, | |
1502 | offsetof(struct virtio_net_config, | |
1503 | max_virtqueue_pairs), &max_queue_pairs); | |
1504 | ||
1505 | /* We need at least 2 queue's */ | |
1506 | if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN || | |
1507 | max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX || | |
1508 | !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) | |
1509 | max_queue_pairs = 1; | |
296f96fc RR |
1510 | |
1511 | /* Allocate ourselves a network device with room for our info */ | |
986a4f4d | 1512 | dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs); |
296f96fc RR |
1513 | if (!dev) |
1514 | return -ENOMEM; | |
1515 | ||
1516 | /* Set up network device as normal. */ | |
f2f2c8b4 | 1517 | dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE; |
76288b4e | 1518 | dev->netdev_ops = &virtnet_netdev; |
296f96fc | 1519 | dev->features = NETIF_F_HIGHDMA; |
3fa2a1df | 1520 | |
a9ea3fc6 | 1521 | SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops); |
296f96fc RR |
1522 | SET_NETDEV_DEV(dev, &vdev->dev); |
1523 | ||
1524 | /* Do we support "hardware" checksums? */ | |
98e778c9 | 1525 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) { |
296f96fc | 1526 | /* This opens up the world of extra features. */ |
98e778c9 MM |
1527 | dev->hw_features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST; |
1528 | if (csum) | |
1529 | dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST; | |
1530 | ||
1531 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) { | |
1532 | dev->hw_features |= NETIF_F_TSO | NETIF_F_UFO | |
34a48579 RR |
1533 | | NETIF_F_TSO_ECN | NETIF_F_TSO6; |
1534 | } | |
5539ae96 | 1535 | /* Individual feature bits: what can host handle? */ |
98e778c9 MM |
1536 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4)) |
1537 | dev->hw_features |= NETIF_F_TSO; | |
1538 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6)) | |
1539 | dev->hw_features |= NETIF_F_TSO6; | |
1540 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN)) | |
1541 | dev->hw_features |= NETIF_F_TSO_ECN; | |
1542 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO)) | |
1543 | dev->hw_features |= NETIF_F_UFO; | |
1544 | ||
1545 | if (gso) | |
1546 | dev->features |= dev->hw_features & (NETIF_F_ALL_TSO|NETIF_F_UFO); | |
1547 | /* (!csum && gso) case will be fixed by register_netdev() */ | |
296f96fc | 1548 | } |
4f49129b TH |
1549 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM)) |
1550 | dev->features |= NETIF_F_RXCSUM; | |
296f96fc | 1551 | |
4fda8302 JW |
1552 | dev->vlan_features = dev->features; |
1553 | ||
296f96fc | 1554 | /* Configuration may specify what MAC to use. Otherwise random. */ |
77dd7693 | 1555 | if (virtio_config_val_len(vdev, VIRTIO_NET_F_MAC, |
a586d4f6 | 1556 | offsetof(struct virtio_net_config, mac), |
77dd7693 | 1557 | dev->dev_addr, dev->addr_len) < 0) |
f2cedb63 | 1558 | eth_hw_addr_random(dev); |
296f96fc RR |
1559 | |
1560 | /* Set up our device-specific information */ | |
1561 | vi = netdev_priv(dev); | |
296f96fc RR |
1562 | vi->dev = dev; |
1563 | vi->vdev = vdev; | |
d9d5dcc8 | 1564 | vdev->priv = vi; |
3fa2a1df | 1565 | vi->stats = alloc_percpu(struct virtnet_stats); |
1566 | err = -ENOMEM; | |
1567 | if (vi->stats == NULL) | |
1568 | goto free; | |
1569 | ||
47be2479 WG |
1570 | vi->vq_index = alloc_percpu(int); |
1571 | if (vi->vq_index == NULL) | |
1572 | goto free_stats; | |
1573 | ||
586d17c5 JW |
1574 | mutex_init(&vi->config_lock); |
1575 | vi->config_enable = true; | |
1576 | INIT_WORK(&vi->config_work, virtnet_config_changed_work); | |
296f96fc | 1577 | |
97402b96 | 1578 | /* If we can receive ANY GSO packets, we must allocate large ones. */ |
8e95a202 JP |
1579 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) || |
1580 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) || | |
1581 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN)) | |
97402b96 HX |
1582 | vi->big_packets = true; |
1583 | ||
3f2c31d9 MM |
1584 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF)) |
1585 | vi->mergeable_rx_bufs = true; | |
1586 | ||
e7428e95 MT |
1587 | if (virtio_has_feature(vdev, VIRTIO_F_ANY_LAYOUT)) |
1588 | vi->any_header_sg = true; | |
1589 | ||
986a4f4d JW |
1590 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) |
1591 | vi->has_cvq = true; | |
1592 | ||
1593 | /* Use single tx/rx queue pair as default */ | |
1594 | vi->curr_queue_pairs = 1; | |
1595 | vi->max_queue_pairs = max_queue_pairs; | |
1596 | ||
1597 | /* Allocate/initialize the rx/tx queues, and invoke find_vqs */ | |
3f9c10b0 | 1598 | err = init_vqs(vi); |
d2a7ddda | 1599 | if (err) |
47be2479 | 1600 | goto free_index; |
296f96fc | 1601 | |
986a4f4d JW |
1602 | netif_set_real_num_tx_queues(dev, 1); |
1603 | netif_set_real_num_rx_queues(dev, 1); | |
1604 | ||
296f96fc RR |
1605 | err = register_netdev(dev); |
1606 | if (err) { | |
1607 | pr_debug("virtio_net: registering device failed\n"); | |
d2a7ddda | 1608 | goto free_vqs; |
296f96fc | 1609 | } |
b3369c1f RR |
1610 | |
1611 | /* Last of all, set up some receive buffers. */ | |
55257d72 | 1612 | for (i = 0; i < vi->curr_queue_pairs; i++) { |
986a4f4d JW |
1613 | try_fill_recv(&vi->rq[i], GFP_KERNEL); |
1614 | ||
1615 | /* If we didn't even get one input buffer, we're useless. */ | |
1616 | if (vi->rq[i].num == 0) { | |
1617 | free_unused_bufs(vi); | |
1618 | err = -ENOMEM; | |
1619 | goto free_recv_bufs; | |
1620 | } | |
b3369c1f RR |
1621 | } |
1622 | ||
8de4b2f3 WG |
1623 | vi->nb.notifier_call = &virtnet_cpu_callback; |
1624 | err = register_hotcpu_notifier(&vi->nb); | |
1625 | if (err) { | |
1626 | pr_debug("virtio_net: registering cpu notifier failed\n"); | |
1627 | goto free_recv_bufs; | |
1628 | } | |
1629 | ||
167c25e4 JW |
1630 | /* Assume link up if device can't report link status, |
1631 | otherwise get link status from config. */ | |
1632 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) { | |
1633 | netif_carrier_off(dev); | |
3b07e9ca | 1634 | schedule_work(&vi->config_work); |
167c25e4 JW |
1635 | } else { |
1636 | vi->status = VIRTIO_NET_S_LINK_UP; | |
1637 | netif_carrier_on(dev); | |
1638 | } | |
9f4d26d0 | 1639 | |
986a4f4d JW |
1640 | pr_debug("virtnet: registered device %s with %d RX and TX vq's\n", |
1641 | dev->name, max_queue_pairs); | |
1642 | ||
296f96fc RR |
1643 | return 0; |
1644 | ||
986a4f4d JW |
1645 | free_recv_bufs: |
1646 | free_receive_bufs(vi); | |
b3369c1f | 1647 | unregister_netdev(dev); |
d2a7ddda | 1648 | free_vqs: |
986a4f4d | 1649 | cancel_delayed_work_sync(&vi->refill); |
e9d7417b | 1650 | virtnet_del_vqs(vi); |
47be2479 WG |
1651 | free_index: |
1652 | free_percpu(vi->vq_index); | |
3fa2a1df | 1653 | free_stats: |
1654 | free_percpu(vi->stats); | |
296f96fc RR |
1655 | free: |
1656 | free_netdev(dev); | |
1657 | return err; | |
1658 | } | |
1659 | ||
04486ed0 | 1660 | static void remove_vq_common(struct virtnet_info *vi) |
296f96fc | 1661 | { |
04486ed0 | 1662 | vi->vdev->config->reset(vi->vdev); |
830a8a97 SM |
1663 | |
1664 | /* Free unused buffers in both send and recv, if any. */ | |
9ab86bbc | 1665 | free_unused_bufs(vi); |
fb6813f4 | 1666 | |
986a4f4d | 1667 | free_receive_bufs(vi); |
d2a7ddda | 1668 | |
986a4f4d | 1669 | virtnet_del_vqs(vi); |
04486ed0 AS |
1670 | } |
1671 | ||
8cc085d6 | 1672 | static void virtnet_remove(struct virtio_device *vdev) |
04486ed0 AS |
1673 | { |
1674 | struct virtnet_info *vi = vdev->priv; | |
1675 | ||
8de4b2f3 WG |
1676 | unregister_hotcpu_notifier(&vi->nb); |
1677 | ||
586d17c5 JW |
1678 | /* Prevent config work handler from accessing the device. */ |
1679 | mutex_lock(&vi->config_lock); | |
1680 | vi->config_enable = false; | |
1681 | mutex_unlock(&vi->config_lock); | |
1682 | ||
04486ed0 AS |
1683 | unregister_netdev(vi->dev); |
1684 | ||
1685 | remove_vq_common(vi); | |
fb6813f4 | 1686 | |
586d17c5 JW |
1687 | flush_work(&vi->config_work); |
1688 | ||
47be2479 | 1689 | free_percpu(vi->vq_index); |
2e66f55b | 1690 | free_percpu(vi->stats); |
74b2553f | 1691 | free_netdev(vi->dev); |
296f96fc RR |
1692 | } |
1693 | ||
0741bcb5 AS |
1694 | #ifdef CONFIG_PM |
1695 | static int virtnet_freeze(struct virtio_device *vdev) | |
1696 | { | |
1697 | struct virtnet_info *vi = vdev->priv; | |
986a4f4d | 1698 | int i; |
0741bcb5 | 1699 | |
586d17c5 JW |
1700 | /* Prevent config work handler from accessing the device */ |
1701 | mutex_lock(&vi->config_lock); | |
1702 | vi->config_enable = false; | |
1703 | mutex_unlock(&vi->config_lock); | |
1704 | ||
0741bcb5 AS |
1705 | netif_device_detach(vi->dev); |
1706 | cancel_delayed_work_sync(&vi->refill); | |
1707 | ||
1708 | if (netif_running(vi->dev)) | |
986a4f4d JW |
1709 | for (i = 0; i < vi->max_queue_pairs; i++) { |
1710 | napi_disable(&vi->rq[i].napi); | |
1711 | netif_napi_del(&vi->rq[i].napi); | |
1712 | } | |
0741bcb5 AS |
1713 | |
1714 | remove_vq_common(vi); | |
1715 | ||
586d17c5 JW |
1716 | flush_work(&vi->config_work); |
1717 | ||
0741bcb5 AS |
1718 | return 0; |
1719 | } | |
1720 | ||
1721 | static int virtnet_restore(struct virtio_device *vdev) | |
1722 | { | |
1723 | struct virtnet_info *vi = vdev->priv; | |
986a4f4d | 1724 | int err, i; |
0741bcb5 AS |
1725 | |
1726 | err = init_vqs(vi); | |
1727 | if (err) | |
1728 | return err; | |
1729 | ||
1730 | if (netif_running(vi->dev)) | |
986a4f4d JW |
1731 | for (i = 0; i < vi->max_queue_pairs; i++) |
1732 | virtnet_napi_enable(&vi->rq[i]); | |
0741bcb5 AS |
1733 | |
1734 | netif_device_attach(vi->dev); | |
1735 | ||
55257d72 | 1736 | for (i = 0; i < vi->curr_queue_pairs; i++) |
986a4f4d JW |
1737 | if (!try_fill_recv(&vi->rq[i], GFP_KERNEL)) |
1738 | schedule_delayed_work(&vi->refill, 0); | |
0741bcb5 | 1739 | |
586d17c5 JW |
1740 | mutex_lock(&vi->config_lock); |
1741 | vi->config_enable = true; | |
1742 | mutex_unlock(&vi->config_lock); | |
1743 | ||
986a4f4d JW |
1744 | virtnet_set_queues(vi, vi->curr_queue_pairs); |
1745 | ||
0741bcb5 AS |
1746 | return 0; |
1747 | } | |
1748 | #endif | |
1749 | ||
296f96fc RR |
1750 | static struct virtio_device_id id_table[] = { |
1751 | { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID }, | |
1752 | { 0 }, | |
1753 | }; | |
1754 | ||
c45a6816 | 1755 | static unsigned int features[] = { |
5e4fe5c4 MM |
1756 | VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM, |
1757 | VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC, | |
c45a6816 | 1758 | VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6, |
97402b96 | 1759 | VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, |
5c516751 | 1760 | VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO, |
2a41f71d | 1761 | VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ, |
0bde9569 | 1762 | VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, |
986a4f4d | 1763 | VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ, |
7e58d5ae | 1764 | VIRTIO_NET_F_CTRL_MAC_ADDR, |
e7428e95 | 1765 | VIRTIO_F_ANY_LAYOUT, |
c45a6816 RR |
1766 | }; |
1767 | ||
22402529 | 1768 | static struct virtio_driver virtio_net_driver = { |
c45a6816 RR |
1769 | .feature_table = features, |
1770 | .feature_table_size = ARRAY_SIZE(features), | |
296f96fc RR |
1771 | .driver.name = KBUILD_MODNAME, |
1772 | .driver.owner = THIS_MODULE, | |
1773 | .id_table = id_table, | |
1774 | .probe = virtnet_probe, | |
8cc085d6 | 1775 | .remove = virtnet_remove, |
9f4d26d0 | 1776 | .config_changed = virtnet_config_changed, |
0741bcb5 AS |
1777 | #ifdef CONFIG_PM |
1778 | .freeze = virtnet_freeze, | |
1779 | .restore = virtnet_restore, | |
1780 | #endif | |
296f96fc RR |
1781 | }; |
1782 | ||
b2a17029 | 1783 | module_virtio_driver(virtio_net_driver); |
296f96fc RR |
1784 | |
1785 | MODULE_DEVICE_TABLE(virtio, id_table); | |
1786 | MODULE_DESCRIPTION("Virtio network driver"); | |
1787 | MODULE_LICENSE("GPL"); |