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