Merge branch 'tcp_tsoffset'
[linux-block.git] / drivers / net / virtio_net.c
CommitLineData
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
31static int napi_weight = 128;
32module_param(napi_weight, int, 0444);
33
eb939922 34static bool csum = true, gso = true;
34a48579
RR
35module_param(csum, bool, 0444);
36module_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
f565a7c2 42#define VIRTNET_SEND_COMMAND_SG_MAX 2
66846048 43#define VIRTNET_DRIVER_VERSION "1.0.0"
2a41f71d 44
3fa2a1df 45struct virtnet_stats {
83a27052
ED
46 struct u64_stats_sync tx_syncp;
47 struct u64_stats_sync rx_syncp;
3fa2a1df 48 u64 tx_bytes;
49 u64 tx_packets;
50
51 u64 rx_bytes;
52 u64 rx_packets;
53};
54
e9d7417b
JW
55/* Internal representation of a send virtqueue */
56struct send_queue {
57 /* Virtqueue associated with this send _queue */
58 struct virtqueue *vq;
59
60 /* TX: fragments + linear part + virtio header */
61 struct scatterlist sg[MAX_SKB_FRAGS + 2];
986a4f4d
JW
62
63 /* Name of the send queue: output.$index */
64 char name[40];
e9d7417b
JW
65};
66
67/* Internal representation of a receive virtqueue */
68struct receive_queue {
69 /* Virtqueue associated with this receive_queue */
70 struct virtqueue *vq;
71
296f96fc
RR
72 struct napi_struct napi;
73
74 /* Number of input buffers, and max we've ever had. */
75 unsigned int num, max;
76
e9d7417b
JW
77 /* Chain pages by the private ptr. */
78 struct page *pages;
79
80 /* RX: fragments + linear part + virtio header */
81 struct scatterlist sg[MAX_SKB_FRAGS + 2];
986a4f4d
JW
82
83 /* Name of this receive queue: input.$index */
84 char name[40];
e9d7417b
JW
85};
86
87struct virtnet_info {
88 struct virtio_device *vdev;
89 struct virtqueue *cvq;
90 struct net_device *dev;
986a4f4d
JW
91 struct send_queue *sq;
92 struct receive_queue *rq;
e9d7417b
JW
93 unsigned int status;
94
986a4f4d
JW
95 /* Max # of queue pairs supported by the device */
96 u16 max_queue_pairs;
97
98 /* # of queue pairs currently used by the driver */
99 u16 curr_queue_pairs;
100
97402b96
HX
101 /* I like... big packets and I cannot lie! */
102 bool big_packets;
103
3f2c31d9
MM
104 /* Host will merge rx buffers for big packets (shake it! shake it!) */
105 bool mergeable_rx_bufs;
106
986a4f4d
JW
107 /* Has control virtqueue */
108 bool has_cvq;
109
586d17c5
JW
110 /* enable config space updates */
111 bool config_enable;
112
3fa2a1df 113 /* Active statistics */
114 struct virtnet_stats __percpu *stats;
115
3161e453
RR
116 /* Work struct for refilling if we run low on memory. */
117 struct delayed_work refill;
118
586d17c5
JW
119 /* Work struct for config space updates */
120 struct work_struct config_work;
121
122 /* Lock for config space updates */
123 struct mutex config_lock;
986a4f4d
JW
124
125 /* Does the affinity hint is set for virtqueues? */
126 bool affinity_hint_set;
47be2479
WG
127
128 /* Per-cpu variable to show the mapping from CPU to virtqueue */
129 int __percpu *vq_index;
8de4b2f3
WG
130
131 /* CPU hot plug notifier */
132 struct notifier_block nb;
296f96fc
RR
133};
134
b3f24698
RR
135struct skb_vnet_hdr {
136 union {
137 struct virtio_net_hdr hdr;
138 struct virtio_net_hdr_mrg_rxbuf mhdr;
139 };
140};
141
9ab86bbc
SM
142struct padded_vnet_hdr {
143 struct virtio_net_hdr hdr;
144 /*
145 * virtio_net_hdr should be in a separated sg buffer because of a
146 * QEMU bug, and data sg buffer shares same page with this header sg.
147 * This padding makes next sg 16 byte aligned after virtio_net_hdr.
148 */
149 char padding[6];
150};
151
986a4f4d
JW
152/* Converting between virtqueue no. and kernel tx/rx queue no.
153 * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq
154 */
155static int vq2txq(struct virtqueue *vq)
156{
157 return (virtqueue_get_queue_index(vq) - 1) / 2;
158}
159
160static int txq2vq(int txq)
161{
162 return txq * 2 + 1;
163}
164
165static int vq2rxq(struct virtqueue *vq)
166{
167 return virtqueue_get_queue_index(vq) / 2;
168}
169
170static int rxq2vq(int rxq)
171{
172 return rxq * 2;
173}
174
b3f24698 175static inline struct skb_vnet_hdr *skb_vnet_hdr(struct sk_buff *skb)
296f96fc 176{
b3f24698 177 return (struct skb_vnet_hdr *)skb->cb;
296f96fc
RR
178}
179
9ab86bbc
SM
180/*
181 * private is used to chain pages for big packets, put the whole
182 * most recent used list in the beginning for reuse
183 */
e9d7417b 184static void give_pages(struct receive_queue *rq, struct page *page)
0a888fd1 185{
9ab86bbc 186 struct page *end;
0a888fd1 187
e9d7417b 188 /* Find end of list, sew whole thing into vi->rq.pages. */
9ab86bbc 189 for (end = page; end->private; end = (struct page *)end->private);
e9d7417b
JW
190 end->private = (unsigned long)rq->pages;
191 rq->pages = page;
0a888fd1
MM
192}
193
e9d7417b 194static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask)
fb6813f4 195{
e9d7417b 196 struct page *p = rq->pages;
fb6813f4 197
9ab86bbc 198 if (p) {
e9d7417b 199 rq->pages = (struct page *)p->private;
9ab86bbc
SM
200 /* clear private here, it is used to chain pages */
201 p->private = 0;
202 } else
fb6813f4
RR
203 p = alloc_page(gfp_mask);
204 return p;
205}
206
e9d7417b 207static void skb_xmit_done(struct virtqueue *vq)
296f96fc 208{
e9d7417b 209 struct virtnet_info *vi = vq->vdev->priv;
296f96fc 210
2cb9c6ba 211 /* Suppress further interrupts. */
e9d7417b 212 virtqueue_disable_cb(vq);
11a3a154 213
363f1514 214 /* We were probably waiting for more output buffers. */
986a4f4d 215 netif_wake_subqueue(vi->dev, vq2txq(vq));
296f96fc
RR
216}
217
9ab86bbc
SM
218static void set_skb_frag(struct sk_buff *skb, struct page *page,
219 unsigned int offset, unsigned int *len)
296f96fc 220{
8a59a7b9 221 int size = min((unsigned)PAGE_SIZE - offset, *len);
9ab86bbc 222 int i = skb_shinfo(skb)->nr_frags;
9ab86bbc 223
8a59a7b9 224 __skb_fill_page_desc(skb, i, page, offset, size);
9ab86bbc 225
8a59a7b9
KK
226 skb->data_len += size;
227 skb->len += size;
4b727361 228 skb->truesize += PAGE_SIZE;
9ab86bbc 229 skb_shinfo(skb)->nr_frags++;
cef401de 230 skb_shinfo(skb)->gso_type |= SKB_GSO_SHARED_FRAG;
8a59a7b9 231 *len -= size;
9ab86bbc 232}
23cde76d 233
3464645a 234/* Called from bottom half context */
e9d7417b 235static struct sk_buff *page_to_skb(struct receive_queue *rq,
9ab86bbc
SM
236 struct page *page, unsigned int len)
237{
e9d7417b 238 struct virtnet_info *vi = rq->vq->vdev->priv;
9ab86bbc
SM
239 struct sk_buff *skb;
240 struct skb_vnet_hdr *hdr;
241 unsigned int copy, hdr_len, offset;
242 char *p;
fb6813f4 243
9ab86bbc 244 p = page_address(page);
3f2c31d9 245
9ab86bbc
SM
246 /* copy small packet so we can reuse these pages for small data */
247 skb = netdev_alloc_skb_ip_align(vi->dev, GOOD_COPY_LEN);
248 if (unlikely(!skb))
249 return NULL;
3f2c31d9 250
9ab86bbc 251 hdr = skb_vnet_hdr(skb);
3f2c31d9 252
9ab86bbc
SM
253 if (vi->mergeable_rx_bufs) {
254 hdr_len = sizeof hdr->mhdr;
255 offset = hdr_len;
256 } else {
257 hdr_len = sizeof hdr->hdr;
258 offset = sizeof(struct padded_vnet_hdr);
259 }
3f2c31d9 260
9ab86bbc 261 memcpy(hdr, p, hdr_len);
3f2c31d9 262
9ab86bbc
SM
263 len -= hdr_len;
264 p += offset;
3f2c31d9 265
9ab86bbc
SM
266 copy = len;
267 if (copy > skb_tailroom(skb))
268 copy = skb_tailroom(skb);
269 memcpy(skb_put(skb, copy), p, copy);
3f2c31d9 270
9ab86bbc
SM
271 len -= copy;
272 offset += copy;
3f2c31d9 273
e878d78b
SL
274 /*
275 * Verify that we can indeed put this data into a skb.
276 * This is here to handle cases when the device erroneously
277 * tries to receive more than is possible. This is usually
278 * the case of a broken device.
279 */
280 if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) {
be443899 281 net_dbg_ratelimited("%s: too much data\n", skb->dev->name);
e878d78b
SL
282 dev_kfree_skb(skb);
283 return NULL;
284 }
285
9ab86bbc
SM
286 while (len) {
287 set_skb_frag(skb, page, offset, &len);
288 page = (struct page *)page->private;
289 offset = 0;
290 }
3f2c31d9 291
9ab86bbc 292 if (page)
e9d7417b 293 give_pages(rq, page);
3f2c31d9 294
9ab86bbc
SM
295 return skb;
296}
3f2c31d9 297
e9d7417b 298static int receive_mergeable(struct receive_queue *rq, struct sk_buff *skb)
9ab86bbc
SM
299{
300 struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
301 struct page *page;
302 int num_buf, i, len;
303
304 num_buf = hdr->mhdr.num_buffers;
305 while (--num_buf) {
306 i = skb_shinfo(skb)->nr_frags;
307 if (i >= MAX_SKB_FRAGS) {
308 pr_debug("%s: packet too long\n", skb->dev->name);
309 skb->dev->stats.rx_length_errors++;
310 return -EINVAL;
311 }
e9d7417b 312 page = virtqueue_get_buf(rq->vq, &len);
9ab86bbc
SM
313 if (!page) {
314 pr_debug("%s: rx error: %d buffers missing\n",
315 skb->dev->name, hdr->mhdr.num_buffers);
316 skb->dev->stats.rx_length_errors++;
317 return -EINVAL;
3f2c31d9 318 }
3fa2a1df 319
9ab86bbc
SM
320 if (len > PAGE_SIZE)
321 len = PAGE_SIZE;
322
323 set_skb_frag(skb, page, 0, &len);
324
e9d7417b 325 --rq->num;
9ab86bbc
SM
326 }
327 return 0;
328}
329
e9d7417b 330static void receive_buf(struct receive_queue *rq, void *buf, unsigned int len)
9ab86bbc 331{
e9d7417b
JW
332 struct virtnet_info *vi = rq->vq->vdev->priv;
333 struct net_device *dev = vi->dev;
58472a76 334 struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
9ab86bbc
SM
335 struct sk_buff *skb;
336 struct page *page;
337 struct skb_vnet_hdr *hdr;
3f2c31d9 338
9ab86bbc
SM
339 if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
340 pr_debug("%s: short packet %i\n", dev->name, len);
341 dev->stats.rx_length_errors++;
342 if (vi->mergeable_rx_bufs || vi->big_packets)
e9d7417b 343 give_pages(rq, buf);
9ab86bbc
SM
344 else
345 dev_kfree_skb(buf);
346 return;
347 }
3f2c31d9 348
9ab86bbc
SM
349 if (!vi->mergeable_rx_bufs && !vi->big_packets) {
350 skb = buf;
351 len -= sizeof(struct virtio_net_hdr);
352 skb_trim(skb, len);
353 } else {
354 page = buf;
e9d7417b 355 skb = page_to_skb(rq, page, len);
9ab86bbc 356 if (unlikely(!skb)) {
3f2c31d9 357 dev->stats.rx_dropped++;
e9d7417b 358 give_pages(rq, page);
9ab86bbc 359 return;
3f2c31d9 360 }
9ab86bbc 361 if (vi->mergeable_rx_bufs)
e9d7417b 362 if (receive_mergeable(rq, skb)) {
9ab86bbc
SM
363 dev_kfree_skb(skb);
364 return;
365 }
97402b96 366 }
3f2c31d9 367
9ab86bbc 368 hdr = skb_vnet_hdr(skb);
3fa2a1df 369
83a27052 370 u64_stats_update_begin(&stats->rx_syncp);
3fa2a1df 371 stats->rx_bytes += skb->len;
372 stats->rx_packets++;
83a27052 373 u64_stats_update_end(&stats->rx_syncp);
296f96fc 374
b3f24698 375 if (hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
296f96fc 376 pr_debug("Needs csum!\n");
b3f24698
RR
377 if (!skb_partial_csum_set(skb,
378 hdr->hdr.csum_start,
379 hdr->hdr.csum_offset))
296f96fc 380 goto frame_err;
10a8d94a
JW
381 } else if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) {
382 skb->ip_summed = CHECKSUM_UNNECESSARY;
296f96fc
RR
383 }
384
23cde76d
MM
385 skb->protocol = eth_type_trans(skb, dev);
386 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
387 ntohs(skb->protocol), skb->len, skb->pkt_type);
388
b3f24698 389 if (hdr->hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
cef401de
ED
390 unsigned short gso_type = 0;
391
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:
cef401de 395 gso_type = SKB_GSO_TCPV4;
296f96fc 396 break;
296f96fc 397 case VIRTIO_NET_HDR_GSO_UDP:
cef401de 398 gso_type = SKB_GSO_UDP;
296f96fc
RR
399 break;
400 case VIRTIO_NET_HDR_GSO_TCPV6:
cef401de 401 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)
cef401de 410 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
cef401de 418 skb_shinfo(skb)->gso_type |= gso_type;
296f96fc
RR
419 /* Header must be checked, and gso_segs computed. */
420 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
421 skb_shinfo(skb)->gso_segs = 0;
422 }
423
424 netif_receive_skb(skb);
425 return;
426
427frame_err:
428 dev->stats.rx_frame_errors++;
296f96fc
RR
429 dev_kfree_skb(skb);
430}
431
e9d7417b 432static int add_recvbuf_small(struct receive_queue *rq, gfp_t gfp)
296f96fc 433{
e9d7417b 434 struct virtnet_info *vi = rq->vq->vdev->priv;
296f96fc 435 struct sk_buff *skb;
9ab86bbc 436 struct skb_vnet_hdr *hdr;
9ab86bbc 437 int err;
3f2c31d9 438
3464645a 439 skb = __netdev_alloc_skb_ip_align(vi->dev, MAX_PACKET_LEN, gfp);
9ab86bbc
SM
440 if (unlikely(!skb))
441 return -ENOMEM;
296f96fc 442
9ab86bbc 443 skb_put(skb, MAX_PACKET_LEN);
3f2c31d9 444
9ab86bbc 445 hdr = skb_vnet_hdr(skb);
e9d7417b 446 sg_set_buf(rq->sg, &hdr->hdr, sizeof hdr->hdr);
97402b96 447
e9d7417b 448 skb_to_sgvec(skb, rq->sg + 1, 0, skb->len);
97402b96 449
e9d7417b 450 err = virtqueue_add_buf(rq->vq, rq->sg, 0, 2, skb, gfp);
9ab86bbc
SM
451 if (err < 0)
452 dev_kfree_skb(skb);
97402b96 453
9ab86bbc
SM
454 return err;
455}
97402b96 456
e9d7417b 457static int add_recvbuf_big(struct receive_queue *rq, gfp_t gfp)
9ab86bbc 458{
9ab86bbc
SM
459 struct page *first, *list = NULL;
460 char *p;
461 int i, err, offset;
462
e9d7417b 463 /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */
9ab86bbc 464 for (i = MAX_SKB_FRAGS + 1; i > 1; --i) {
e9d7417b 465 first = get_a_page(rq, gfp);
9ab86bbc
SM
466 if (!first) {
467 if (list)
e9d7417b 468 give_pages(rq, list);
9ab86bbc 469 return -ENOMEM;
97402b96 470 }
e9d7417b 471 sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE);
97402b96 472
9ab86bbc
SM
473 /* chain new page in list head to match sg */
474 first->private = (unsigned long)list;
475 list = first;
476 }
296f96fc 477
e9d7417b 478 first = get_a_page(rq, gfp);
9ab86bbc 479 if (!first) {
e9d7417b 480 give_pages(rq, list);
9ab86bbc
SM
481 return -ENOMEM;
482 }
483 p = page_address(first);
484
e9d7417b
JW
485 /* rq->sg[0], rq->sg[1] share the same page */
486 /* a separated rq->sg[0] for virtio_net_hdr only due to QEMU bug */
487 sg_set_buf(&rq->sg[0], p, sizeof(struct virtio_net_hdr));
9ab86bbc 488
e9d7417b 489 /* rq->sg[1] for data packet, from offset */
9ab86bbc 490 offset = sizeof(struct padded_vnet_hdr);
e9d7417b 491 sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset);
9ab86bbc
SM
492
493 /* chain first in list head */
494 first->private = (unsigned long)list;
e9d7417b 495 err = virtqueue_add_buf(rq->vq, rq->sg, 0, MAX_SKB_FRAGS + 2,
f96fde41 496 first, gfp);
9ab86bbc 497 if (err < 0)
e9d7417b 498 give_pages(rq, first);
9ab86bbc
SM
499
500 return err;
296f96fc
RR
501}
502
e9d7417b 503static int add_recvbuf_mergeable(struct receive_queue *rq, gfp_t gfp)
3f2c31d9 504{
9ab86bbc 505 struct page *page;
3f2c31d9 506 int err;
3f2c31d9 507
e9d7417b 508 page = get_a_page(rq, gfp);
9ab86bbc
SM
509 if (!page)
510 return -ENOMEM;
3f2c31d9 511
e9d7417b 512 sg_init_one(rq->sg, page_address(page), PAGE_SIZE);
3f2c31d9 513
e9d7417b 514 err = virtqueue_add_buf(rq->vq, rq->sg, 0, 1, page, gfp);
9ab86bbc 515 if (err < 0)
e9d7417b 516 give_pages(rq, page);
3f2c31d9 517
9ab86bbc
SM
518 return err;
519}
3f2c31d9 520
b2baed69
RR
521/*
522 * Returns false if we couldn't fill entirely (OOM).
523 *
524 * Normally run in the receive path, but can also be run from ndo_open
525 * before we're receiving packets, or from refill_work which is
526 * careful to disable receiving (using napi_disable).
527 */
e9d7417b 528static bool try_fill_recv(struct receive_queue *rq, gfp_t gfp)
9ab86bbc 529{
e9d7417b 530 struct virtnet_info *vi = rq->vq->vdev->priv;
9ab86bbc 531 int err;
1788f495 532 bool oom;
3f2c31d9 533
9ab86bbc
SM
534 do {
535 if (vi->mergeable_rx_bufs)
e9d7417b 536 err = add_recvbuf_mergeable(rq, gfp);
9ab86bbc 537 else if (vi->big_packets)
e9d7417b 538 err = add_recvbuf_big(rq, gfp);
9ab86bbc 539 else
e9d7417b 540 err = add_recvbuf_small(rq, gfp);
3f2c31d9 541
1788f495 542 oom = err == -ENOMEM;
9ed4cb07 543 if (err)
3f2c31d9 544 break;
e9d7417b 545 ++rq->num;
b7dfde95 546 } while (rq->vq->num_free);
e9d7417b
JW
547 if (unlikely(rq->num > rq->max))
548 rq->max = rq->num;
549 virtqueue_kick(rq->vq);
3161e453 550 return !oom;
3f2c31d9
MM
551}
552
18445c4d 553static void skb_recv_done(struct virtqueue *rvq)
296f96fc
RR
554{
555 struct virtnet_info *vi = rvq->vdev->priv;
986a4f4d 556 struct receive_queue *rq = &vi->rq[vq2rxq(rvq)];
e9d7417b 557
18445c4d 558 /* Schedule NAPI, Suppress further interrupts if successful. */
e9d7417b 559 if (napi_schedule_prep(&rq->napi)) {
1915a712 560 virtqueue_disable_cb(rvq);
e9d7417b 561 __napi_schedule(&rq->napi);
18445c4d 562 }
296f96fc
RR
563}
564
e9d7417b 565static void virtnet_napi_enable(struct receive_queue *rq)
3e9d08ec 566{
e9d7417b 567 napi_enable(&rq->napi);
3e9d08ec
BR
568
569 /* If all buffers were filled by other side before we napi_enabled, we
570 * won't get another interrupt, so process any outstanding packets
571 * now. virtnet_poll wants re-enable the queue, so we disable here.
572 * We synchronize against interrupts via NAPI_STATE_SCHED */
e9d7417b
JW
573 if (napi_schedule_prep(&rq->napi)) {
574 virtqueue_disable_cb(rq->vq);
ec13ee80 575 local_bh_disable();
e9d7417b 576 __napi_schedule(&rq->napi);
ec13ee80 577 local_bh_enable();
3e9d08ec
BR
578 }
579}
580
3161e453
RR
581static void refill_work(struct work_struct *work)
582{
e9d7417b
JW
583 struct virtnet_info *vi =
584 container_of(work, struct virtnet_info, refill.work);
3161e453 585 bool still_empty;
986a4f4d
JW
586 int i;
587
588 for (i = 0; i < vi->max_queue_pairs; i++) {
589 struct receive_queue *rq = &vi->rq[i];
3161e453 590
986a4f4d
JW
591 napi_disable(&rq->napi);
592 still_empty = !try_fill_recv(rq, GFP_KERNEL);
593 virtnet_napi_enable(rq);
3161e453 594
986a4f4d
JW
595 /* In theory, this can happen: if we don't get any buffers in
596 * we will *never* try to fill again.
597 */
598 if (still_empty)
599 schedule_delayed_work(&vi->refill, HZ/2);
600 }
3161e453
RR
601}
602
296f96fc
RR
603static int virtnet_poll(struct napi_struct *napi, int budget)
604{
e9d7417b
JW
605 struct receive_queue *rq =
606 container_of(napi, struct receive_queue, napi);
607 struct virtnet_info *vi = rq->vq->vdev->priv;
9ab86bbc 608 void *buf;
296f96fc
RR
609 unsigned int len, received = 0;
610
611again:
612 while (received < budget &&
e9d7417b
JW
613 (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) {
614 receive_buf(rq, buf, len);
615 --rq->num;
296f96fc
RR
616 received++;
617 }
618
e9d7417b
JW
619 if (rq->num < rq->max / 2) {
620 if (!try_fill_recv(rq, GFP_ATOMIC))
3b07e9ca 621 schedule_delayed_work(&vi->refill, 0);
3161e453 622 }
296f96fc 623
8329d98e
RR
624 /* Out of packets? */
625 if (received < budget) {
288379f0 626 napi_complete(napi);
e9d7417b 627 if (unlikely(!virtqueue_enable_cb(rq->vq)) &&
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
638static int virtnet_open(struct net_device *dev)
639{
640 struct virtnet_info *vi = netdev_priv(dev);
641 int i;
642
643 for (i = 0; i < vi->max_queue_pairs; i++) {
644 /* Make sure we have some buffers: if oom use wq. */
645 if (!try_fill_recv(&vi->rq[i], GFP_KERNEL))
646 schedule_delayed_work(&vi->refill, 0);
647 virtnet_napi_enable(&vi->rq[i]);
648 }
649
650 return 0;
651}
652
b7dfde95 653static void free_old_xmit_skbs(struct send_queue *sq)
296f96fc
RR
654{
655 struct sk_buff *skb;
6ee57bcc 656 unsigned int len;
e9d7417b 657 struct virtnet_info *vi = sq->vq->vdev->priv;
58472a76 658 struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
296f96fc 659
e9d7417b 660 while ((skb = virtqueue_get_buf(sq->vq, &len)) != NULL) {
296f96fc 661 pr_debug("Sent skb %p\n", skb);
3fa2a1df 662
83a27052 663 u64_stats_update_begin(&stats->tx_syncp);
3fa2a1df 664 stats->tx_bytes += skb->len;
665 stats->tx_packets++;
83a27052 666 u64_stats_update_end(&stats->tx_syncp);
3fa2a1df 667
ed79bab8 668 dev_kfree_skb_any(skb);
296f96fc
RR
669 }
670}
671
e9d7417b 672static int xmit_skb(struct send_queue *sq, struct sk_buff *skb)
296f96fc 673{
b3f24698 674 struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
296f96fc 675 const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
e9d7417b 676 struct virtnet_info *vi = sq->vq->vdev->priv;
7bedc7dc 677 unsigned num_sg;
296f96fc 678
e174961c 679 pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
296f96fc 680
296f96fc 681 if (skb->ip_summed == CHECKSUM_PARTIAL) {
b3f24698 682 hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
55508d60 683 hdr->hdr.csum_start = skb_checksum_start_offset(skb);
b3f24698 684 hdr->hdr.csum_offset = skb->csum_offset;
296f96fc 685 } else {
b3f24698
RR
686 hdr->hdr.flags = 0;
687 hdr->hdr.csum_offset = hdr->hdr.csum_start = 0;
296f96fc
RR
688 }
689
690 if (skb_is_gso(skb)) {
b3f24698
RR
691 hdr->hdr.hdr_len = skb_headlen(skb);
692 hdr->hdr.gso_size = skb_shinfo(skb)->gso_size;
34a48579 693 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
b3f24698 694 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
296f96fc 695 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
b3f24698 696 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
296f96fc 697 else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
b3f24698 698 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
296f96fc
RR
699 else
700 BUG();
34a48579 701 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
b3f24698 702 hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
296f96fc 703 } else {
b3f24698
RR
704 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
705 hdr->hdr.gso_size = hdr->hdr.hdr_len = 0;
296f96fc
RR
706 }
707
b3f24698 708 hdr->mhdr.num_buffers = 0;
3f2c31d9
MM
709
710 /* Encode metadata header at front. */
711 if (vi->mergeable_rx_bufs)
e9d7417b 712 sg_set_buf(sq->sg, &hdr->mhdr, sizeof hdr->mhdr);
3f2c31d9 713 else
e9d7417b 714 sg_set_buf(sq->sg, &hdr->hdr, sizeof hdr->hdr);
3f2c31d9 715
b7dfde95
LT
716 num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len) + 1;
717 return virtqueue_add_buf(sq->vq, sq->sg, num_sg,
f96fde41 718 0, skb, GFP_ATOMIC);
11a3a154
RR
719}
720
424efe9c 721static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
99ffc696
RR
722{
723 struct virtnet_info *vi = netdev_priv(dev);
986a4f4d
JW
724 int qnum = skb_get_queue_mapping(skb);
725 struct send_queue *sq = &vi->sq[qnum];
9ed4cb07 726 int err;
2cb9c6ba 727
2cb9c6ba 728 /* Free up any pending old buffers before queueing new ones. */
e9d7417b 729 free_old_xmit_skbs(sq);
99ffc696 730
03f191ba 731 /* Try to transmit */
b7dfde95 732 err = xmit_skb(sq, skb);
48925e37 733
9ed4cb07 734 /* This should not happen! */
0e3daa64 735 if (unlikely(err)) {
9ed4cb07
RR
736 dev->stats.tx_fifo_errors++;
737 if (net_ratelimit())
738 dev_warn(&dev->dev,
b7dfde95 739 "Unexpected TXQ (%d) queue failure: %d\n", qnum, err);
58eba97d
RR
740 dev->stats.tx_dropped++;
741 kfree_skb(skb);
742 return NETDEV_TX_OK;
296f96fc 743 }
e9d7417b 744 virtqueue_kick(sq->vq);
03f191ba 745
48925e37
RR
746 /* Don't wait up for transmitted skbs to be freed. */
747 skb_orphan(skb);
748 nf_reset(skb);
749
750 /* Apparently nice girls don't return TX_BUSY; stop the queue
751 * before it gets out of hand. Naturally, this wastes entries. */
b7dfde95 752 if (sq->vq->num_free < 2+MAX_SKB_FRAGS) {
986a4f4d 753 netif_stop_subqueue(dev, qnum);
e9d7417b 754 if (unlikely(!virtqueue_enable_cb_delayed(sq->vq))) {
48925e37 755 /* More just got used, free them then recheck. */
b7dfde95
LT
756 free_old_xmit_skbs(sq);
757 if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) {
986a4f4d 758 netif_start_subqueue(dev, qnum);
e9d7417b 759 virtqueue_disable_cb(sq->vq);
48925e37
RR
760 }
761 }
99ffc696 762 }
48925e37
RR
763
764 return NETDEV_TX_OK;
296f96fc
RR
765}
766
40cbfc37
AK
767/*
768 * Send command via the control virtqueue and check status. Commands
769 * supported by the hypervisor, as indicated by feature bits, should
770 * never fail unless improperly formated.
771 */
772static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
773 struct scatterlist *data, int out, int in)
774{
775 struct scatterlist *s, sg[VIRTNET_SEND_COMMAND_SG_MAX + 2];
776 struct virtio_net_ctrl_hdr ctrl;
777 virtio_net_ctrl_ack status = ~0;
778 unsigned int tmp;
779 int i;
780
781 /* Caller should know better */
782 BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ||
783 (out + in > VIRTNET_SEND_COMMAND_SG_MAX));
784
785 out++; /* Add header */
786 in++; /* Add return status */
787
788 ctrl.class = class;
789 ctrl.cmd = cmd;
790
791 sg_init_table(sg, out + in);
792
793 sg_set_buf(&sg[0], &ctrl, sizeof(ctrl));
794 for_each_sg(data, s, out + in - 2, i)
795 sg_set_buf(&sg[i + 1], sg_virt(s), s->length);
796 sg_set_buf(&sg[out + in - 1], &status, sizeof(status));
797
798 BUG_ON(virtqueue_add_buf(vi->cvq, sg, out, in, vi, GFP_ATOMIC) < 0);
799
800 virtqueue_kick(vi->cvq);
801
802 /* Spin for a response, the kick causes an ioport write, trapping
803 * into the hypervisor, so the request should be handled immediately.
804 */
805 while (!virtqueue_get_buf(vi->cvq, &tmp))
806 cpu_relax();
807
808 return status == VIRTIO_NET_OK;
809}
810
9c46f6d4
AW
811static int virtnet_set_mac_address(struct net_device *dev, void *p)
812{
813 struct virtnet_info *vi = netdev_priv(dev);
814 struct virtio_device *vdev = vi->vdev;
f2f2c8b4 815 int ret;
7e58d5ae
AK
816 struct sockaddr *addr = p;
817 struct scatterlist sg;
9c46f6d4 818
7e58d5ae 819 ret = eth_prepare_mac_addr_change(dev, p);
f2f2c8b4
JP
820 if (ret)
821 return ret;
9c46f6d4 822
7e58d5ae
AK
823 if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
824 sg_init_one(&sg, addr->sa_data, dev->addr_len);
825 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
826 VIRTIO_NET_CTRL_MAC_ADDR_SET,
827 &sg, 1, 0)) {
828 dev_warn(&vdev->dev,
829 "Failed to set mac address by vq command.\n");
830 return -EINVAL;
831 }
832 } else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
62994b2d 833 vdev->config->set(vdev, offsetof(struct virtio_net_config, mac),
7e58d5ae
AK
834 addr->sa_data, dev->addr_len);
835 }
836
837 eth_commit_mac_addr_change(dev, p);
9c46f6d4
AW
838
839 return 0;
840}
841
3fa2a1df 842static struct rtnl_link_stats64 *virtnet_stats(struct net_device *dev,
843 struct rtnl_link_stats64 *tot)
844{
845 struct virtnet_info *vi = netdev_priv(dev);
846 int cpu;
847 unsigned int start;
848
849 for_each_possible_cpu(cpu) {
58472a76 850 struct virtnet_stats *stats = per_cpu_ptr(vi->stats, cpu);
3fa2a1df 851 u64 tpackets, tbytes, rpackets, rbytes;
852
853 do {
e3906486 854 start = u64_stats_fetch_begin_bh(&stats->tx_syncp);
3fa2a1df 855 tpackets = stats->tx_packets;
856 tbytes = stats->tx_bytes;
e3906486 857 } while (u64_stats_fetch_retry_bh(&stats->tx_syncp, start));
83a27052
ED
858
859 do {
e3906486 860 start = u64_stats_fetch_begin_bh(&stats->rx_syncp);
3fa2a1df 861 rpackets = stats->rx_packets;
862 rbytes = stats->rx_bytes;
e3906486 863 } while (u64_stats_fetch_retry_bh(&stats->rx_syncp, start));
3fa2a1df 864
865 tot->rx_packets += rpackets;
866 tot->tx_packets += tpackets;
867 tot->rx_bytes += rbytes;
868 tot->tx_bytes += tbytes;
869 }
870
871 tot->tx_dropped = dev->stats.tx_dropped;
021ac8d3 872 tot->tx_fifo_errors = dev->stats.tx_fifo_errors;
3fa2a1df 873 tot->rx_dropped = dev->stats.rx_dropped;
874 tot->rx_length_errors = dev->stats.rx_length_errors;
875 tot->rx_frame_errors = dev->stats.rx_frame_errors;
876
877 return tot;
878}
879
da74e89d
AS
880#ifdef CONFIG_NET_POLL_CONTROLLER
881static void virtnet_netpoll(struct net_device *dev)
882{
883 struct virtnet_info *vi = netdev_priv(dev);
986a4f4d 884 int i;
da74e89d 885
986a4f4d
JW
886 for (i = 0; i < vi->curr_queue_pairs; i++)
887 napi_schedule(&vi->rq[i].napi);
da74e89d
AS
888}
889#endif
890
586d17c5
JW
891static void virtnet_ack_link_announce(struct virtnet_info *vi)
892{
893 rtnl_lock();
894 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE,
895 VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL,
896 0, 0))
897 dev_warn(&vi->dev->dev, "Failed to ack link announce.\n");
898 rtnl_unlock();
899}
900
986a4f4d
JW
901static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs)
902{
903 struct scatterlist sg;
904 struct virtio_net_ctrl_mq s;
905 struct net_device *dev = vi->dev;
906
907 if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ))
908 return 0;
909
910 s.virtqueue_pairs = queue_pairs;
911 sg_init_one(&sg, &s, sizeof(s));
912
913 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ,
914 VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg, 1, 0)){
915 dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n",
916 queue_pairs);
917 return -EINVAL;
918 } else
919 vi->curr_queue_pairs = queue_pairs;
920
921 return 0;
922}
923
296f96fc
RR
924static int virtnet_close(struct net_device *dev)
925{
926 struct virtnet_info *vi = netdev_priv(dev);
986a4f4d 927 int i;
296f96fc 928
b2baed69
RR
929 /* Make sure refill_work doesn't re-enable napi! */
930 cancel_delayed_work_sync(&vi->refill);
986a4f4d
JW
931
932 for (i = 0; i < vi->max_queue_pairs; i++)
933 napi_disable(&vi->rq[i].napi);
296f96fc 934
296f96fc
RR
935 return 0;
936}
937
2af7698e
AW
938static void virtnet_set_rx_mode(struct net_device *dev)
939{
940 struct virtnet_info *vi = netdev_priv(dev);
f565a7c2 941 struct scatterlist sg[2];
2af7698e 942 u8 promisc, allmulti;
f565a7c2 943 struct virtio_net_ctrl_mac *mac_data;
ccffad25 944 struct netdev_hw_addr *ha;
32e7bfc4 945 int uc_count;
4cd24eaf 946 int mc_count;
f565a7c2
AW
947 void *buf;
948 int i;
2af7698e
AW
949
950 /* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */
951 if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
952 return;
953
f565a7c2
AW
954 promisc = ((dev->flags & IFF_PROMISC) != 0);
955 allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
2af7698e 956
23e258e1 957 sg_init_one(sg, &promisc, sizeof(promisc));
2af7698e
AW
958
959 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
960 VIRTIO_NET_CTRL_RX_PROMISC,
f565a7c2 961 sg, 1, 0))
2af7698e
AW
962 dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
963 promisc ? "en" : "dis");
964
23e258e1 965 sg_init_one(sg, &allmulti, sizeof(allmulti));
2af7698e
AW
966
967 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
968 VIRTIO_NET_CTRL_RX_ALLMULTI,
f565a7c2 969 sg, 1, 0))
2af7698e
AW
970 dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
971 allmulti ? "en" : "dis");
f565a7c2 972
32e7bfc4 973 uc_count = netdev_uc_count(dev);
4cd24eaf 974 mc_count = netdev_mc_count(dev);
f565a7c2 975 /* MAC filter - use one buffer for both lists */
4cd24eaf
JP
976 buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
977 (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
978 mac_data = buf;
e68ed8f0 979 if (!buf)
f565a7c2 980 return;
f565a7c2 981
23e258e1
AW
982 sg_init_table(sg, 2);
983
f565a7c2 984 /* Store the unicast list and count in the front of the buffer */
32e7bfc4 985 mac_data->entries = uc_count;
ccffad25 986 i = 0;
32e7bfc4 987 netdev_for_each_uc_addr(ha, dev)
ccffad25 988 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
f565a7c2
AW
989
990 sg_set_buf(&sg[0], mac_data,
32e7bfc4 991 sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
f565a7c2
AW
992
993 /* multicast list and count fill the end */
32e7bfc4 994 mac_data = (void *)&mac_data->macs[uc_count][0];
f565a7c2 995
4cd24eaf 996 mac_data->entries = mc_count;
567ec874 997 i = 0;
22bedad3
JP
998 netdev_for_each_mc_addr(ha, dev)
999 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
f565a7c2
AW
1000
1001 sg_set_buf(&sg[1], mac_data,
4cd24eaf 1002 sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
f565a7c2
AW
1003
1004 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
1005 VIRTIO_NET_CTRL_MAC_TABLE_SET,
1006 sg, 2, 0))
1007 dev_warn(&dev->dev, "Failed to set MAC fitler table.\n");
1008
1009 kfree(buf);
2af7698e
AW
1010}
1011
8e586137 1012static int virtnet_vlan_rx_add_vid(struct net_device *dev, u16 vid)
0bde9569
AW
1013{
1014 struct virtnet_info *vi = netdev_priv(dev);
1015 struct scatterlist sg;
1016
23e258e1 1017 sg_init_one(&sg, &vid, sizeof(vid));
0bde9569
AW
1018
1019 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1020 VIRTIO_NET_CTRL_VLAN_ADD, &sg, 1, 0))
1021 dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
8e586137 1022 return 0;
0bde9569
AW
1023}
1024
8e586137 1025static int virtnet_vlan_rx_kill_vid(struct net_device *dev, u16 vid)
0bde9569
AW
1026{
1027 struct virtnet_info *vi = netdev_priv(dev);
1028 struct scatterlist sg;
1029
23e258e1 1030 sg_init_one(&sg, &vid, sizeof(vid));
0bde9569
AW
1031
1032 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1033 VIRTIO_NET_CTRL_VLAN_DEL, &sg, 1, 0))
1034 dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
8e586137 1035 return 0;
0bde9569
AW
1036}
1037
8898c21c 1038static void virtnet_clean_affinity(struct virtnet_info *vi, long hcpu)
986a4f4d
JW
1039{
1040 int i;
47be2479 1041 int cpu;
986a4f4d 1042
8898c21c
WG
1043 if (vi->affinity_hint_set) {
1044 for (i = 0; i < vi->max_queue_pairs; i++) {
47be2479
WG
1045 virtqueue_set_affinity(vi->rq[i].vq, -1);
1046 virtqueue_set_affinity(vi->sq[i].vq, -1);
1047 }
1048
8898c21c
WG
1049 vi->affinity_hint_set = false;
1050 }
1051
1052 i = 0;
1053 for_each_online_cpu(cpu) {
1054 if (cpu == hcpu) {
1055 *per_cpu_ptr(vi->vq_index, cpu) = -1;
1056 } else {
47be2479
WG
1057 *per_cpu_ptr(vi->vq_index, cpu) =
1058 ++i % vi->curr_queue_pairs;
8898c21c
WG
1059 }
1060 }
1061}
47be2479 1062
8898c21c
WG
1063static void virtnet_set_affinity(struct virtnet_info *vi)
1064{
1065 int i;
1066 int cpu;
986a4f4d
JW
1067
1068 /* In multiqueue mode, when the number of cpu is equal to the number of
1069 * queue pairs, we let the queue pairs to be private to one cpu by
1070 * setting the affinity hint to eliminate the contention.
1071 */
8898c21c
WG
1072 if (vi->curr_queue_pairs == 1 ||
1073 vi->max_queue_pairs != num_online_cpus()) {
1074 virtnet_clean_affinity(vi, -1);
1075 return;
986a4f4d
JW
1076 }
1077
8898c21c
WG
1078 i = 0;
1079 for_each_online_cpu(cpu) {
986a4f4d
JW
1080 virtqueue_set_affinity(vi->rq[i].vq, cpu);
1081 virtqueue_set_affinity(vi->sq[i].vq, cpu);
8898c21c
WG
1082 *per_cpu_ptr(vi->vq_index, cpu) = i;
1083 i++;
986a4f4d
JW
1084 }
1085
8898c21c 1086 vi->affinity_hint_set = true;
986a4f4d
JW
1087}
1088
8de4b2f3
WG
1089static int virtnet_cpu_callback(struct notifier_block *nfb,
1090 unsigned long action, void *hcpu)
1091{
1092 struct virtnet_info *vi = container_of(nfb, struct virtnet_info, nb);
1093
1094 switch(action & ~CPU_TASKS_FROZEN) {
1095 case CPU_ONLINE:
1096 case CPU_DOWN_FAILED:
1097 case CPU_DEAD:
1098 virtnet_set_affinity(vi);
1099 break;
1100 case CPU_DOWN_PREPARE:
1101 virtnet_clean_affinity(vi, (long)hcpu);
1102 break;
1103 default:
1104 break;
1105 }
1106 return NOTIFY_OK;
986a4f4d
JW
1107}
1108
8f9f4668
RJ
1109static void virtnet_get_ringparam(struct net_device *dev,
1110 struct ethtool_ringparam *ring)
1111{
1112 struct virtnet_info *vi = netdev_priv(dev);
1113
986a4f4d
JW
1114 ring->rx_max_pending = virtqueue_get_vring_size(vi->rq[0].vq);
1115 ring->tx_max_pending = virtqueue_get_vring_size(vi->sq[0].vq);
8f9f4668
RJ
1116 ring->rx_pending = ring->rx_max_pending;
1117 ring->tx_pending = ring->tx_max_pending;
8f9f4668
RJ
1118}
1119
66846048
RJ
1120
1121static void virtnet_get_drvinfo(struct net_device *dev,
1122 struct ethtool_drvinfo *info)
1123{
1124 struct virtnet_info *vi = netdev_priv(dev);
1125 struct virtio_device *vdev = vi->vdev;
1126
1127 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
1128 strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version));
1129 strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info));
1130
1131}
1132
d73bcd2c
JW
1133/* TODO: Eliminate OOO packets during switching */
1134static int virtnet_set_channels(struct net_device *dev,
1135 struct ethtool_channels *channels)
1136{
1137 struct virtnet_info *vi = netdev_priv(dev);
1138 u16 queue_pairs = channels->combined_count;
1139 int err;
1140
1141 /* We don't support separate rx/tx channels.
1142 * We don't allow setting 'other' channels.
1143 */
1144 if (channels->rx_count || channels->tx_count || channels->other_count)
1145 return -EINVAL;
1146
1147 if (queue_pairs > vi->max_queue_pairs)
1148 return -EINVAL;
1149
47be2479 1150 get_online_cpus();
d73bcd2c
JW
1151 err = virtnet_set_queues(vi, queue_pairs);
1152 if (!err) {
1153 netif_set_real_num_tx_queues(dev, queue_pairs);
1154 netif_set_real_num_rx_queues(dev, queue_pairs);
1155
8898c21c 1156 virtnet_set_affinity(vi);
d73bcd2c 1157 }
47be2479 1158 put_online_cpus();
d73bcd2c
JW
1159
1160 return err;
1161}
1162
1163static void virtnet_get_channels(struct net_device *dev,
1164 struct ethtool_channels *channels)
1165{
1166 struct virtnet_info *vi = netdev_priv(dev);
1167
1168 channels->combined_count = vi->curr_queue_pairs;
1169 channels->max_combined = vi->max_queue_pairs;
1170 channels->max_other = 0;
1171 channels->rx_count = 0;
1172 channels->tx_count = 0;
1173 channels->other_count = 0;
1174}
1175
0fc0b732 1176static const struct ethtool_ops virtnet_ethtool_ops = {
66846048 1177 .get_drvinfo = virtnet_get_drvinfo,
9f4d26d0 1178 .get_link = ethtool_op_get_link,
8f9f4668 1179 .get_ringparam = virtnet_get_ringparam,
d73bcd2c
JW
1180 .set_channels = virtnet_set_channels,
1181 .get_channels = virtnet_get_channels,
a9ea3fc6
HX
1182};
1183
39da5814
MM
1184#define MIN_MTU 68
1185#define MAX_MTU 65535
1186
1187static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
1188{
1189 if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
1190 return -EINVAL;
1191 dev->mtu = new_mtu;
1192 return 0;
1193}
1194
986a4f4d
JW
1195/* To avoid contending a lock hold by a vcpu who would exit to host, select the
1196 * txq based on the processor id.
986a4f4d
JW
1197 */
1198static u16 virtnet_select_queue(struct net_device *dev, struct sk_buff *skb)
1199{
47be2479
WG
1200 int txq;
1201 struct virtnet_info *vi = netdev_priv(dev);
1202
1203 if (skb_rx_queue_recorded(skb)) {
1204 txq = skb_get_rx_queue(skb);
1205 } else {
1206 txq = *__this_cpu_ptr(vi->vq_index);
1207 if (txq == -1)
1208 txq = 0;
1209 }
986a4f4d
JW
1210
1211 while (unlikely(txq >= dev->real_num_tx_queues))
1212 txq -= dev->real_num_tx_queues;
1213
1214 return txq;
1215}
1216
76288b4e
SH
1217static const struct net_device_ops virtnet_netdev = {
1218 .ndo_open = virtnet_open,
1219 .ndo_stop = virtnet_close,
1220 .ndo_start_xmit = start_xmit,
1221 .ndo_validate_addr = eth_validate_addr,
9c46f6d4 1222 .ndo_set_mac_address = virtnet_set_mac_address,
2af7698e 1223 .ndo_set_rx_mode = virtnet_set_rx_mode,
76288b4e 1224 .ndo_change_mtu = virtnet_change_mtu,
3fa2a1df 1225 .ndo_get_stats64 = virtnet_stats,
1824a989
AW
1226 .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
1227 .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
986a4f4d 1228 .ndo_select_queue = virtnet_select_queue,
76288b4e
SH
1229#ifdef CONFIG_NET_POLL_CONTROLLER
1230 .ndo_poll_controller = virtnet_netpoll,
1231#endif
1232};
1233
586d17c5 1234static void virtnet_config_changed_work(struct work_struct *work)
9f4d26d0 1235{
586d17c5
JW
1236 struct virtnet_info *vi =
1237 container_of(work, struct virtnet_info, config_work);
9f4d26d0
MM
1238 u16 v;
1239
586d17c5
JW
1240 mutex_lock(&vi->config_lock);
1241 if (!vi->config_enable)
1242 goto done;
1243
77dd7693 1244 if (virtio_config_val(vi->vdev, VIRTIO_NET_F_STATUS,
9f4d26d0 1245 offsetof(struct virtio_net_config, status),
77dd7693 1246 &v) < 0)
586d17c5
JW
1247 goto done;
1248
1249 if (v & VIRTIO_NET_S_ANNOUNCE) {
ee89bab1 1250 netdev_notify_peers(vi->dev);
586d17c5
JW
1251 virtnet_ack_link_announce(vi);
1252 }
9f4d26d0
MM
1253
1254 /* Ignore unknown (future) status bits */
1255 v &= VIRTIO_NET_S_LINK_UP;
1256
1257 if (vi->status == v)
586d17c5 1258 goto done;
9f4d26d0
MM
1259
1260 vi->status = v;
1261
1262 if (vi->status & VIRTIO_NET_S_LINK_UP) {
1263 netif_carrier_on(vi->dev);
986a4f4d 1264 netif_tx_wake_all_queues(vi->dev);
9f4d26d0
MM
1265 } else {
1266 netif_carrier_off(vi->dev);
986a4f4d 1267 netif_tx_stop_all_queues(vi->dev);
9f4d26d0 1268 }
586d17c5
JW
1269done:
1270 mutex_unlock(&vi->config_lock);
9f4d26d0
MM
1271}
1272
1273static void virtnet_config_changed(struct virtio_device *vdev)
1274{
1275 struct virtnet_info *vi = vdev->priv;
1276
3b07e9ca 1277 schedule_work(&vi->config_work);
9f4d26d0
MM
1278}
1279
986a4f4d
JW
1280static void virtnet_free_queues(struct virtnet_info *vi)
1281{
1282 kfree(vi->rq);
1283 kfree(vi->sq);
1284}
1285
1286static void free_receive_bufs(struct virtnet_info *vi)
1287{
1288 int i;
1289
1290 for (i = 0; i < vi->max_queue_pairs; i++) {
1291 while (vi->rq[i].pages)
1292 __free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0);
1293 }
1294}
1295
1296static void free_unused_bufs(struct virtnet_info *vi)
1297{
1298 void *buf;
1299 int i;
1300
1301 for (i = 0; i < vi->max_queue_pairs; i++) {
1302 struct virtqueue *vq = vi->sq[i].vq;
1303 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL)
1304 dev_kfree_skb(buf);
1305 }
1306
1307 for (i = 0; i < vi->max_queue_pairs; i++) {
1308 struct virtqueue *vq = vi->rq[i].vq;
1309
1310 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) {
1311 if (vi->mergeable_rx_bufs || vi->big_packets)
1312 give_pages(&vi->rq[i], buf);
1313 else
1314 dev_kfree_skb(buf);
1315 --vi->rq[i].num;
1316 }
1317 BUG_ON(vi->rq[i].num != 0);
1318 }
1319}
1320
e9d7417b
JW
1321static void virtnet_del_vqs(struct virtnet_info *vi)
1322{
1323 struct virtio_device *vdev = vi->vdev;
1324
8898c21c 1325 virtnet_clean_affinity(vi, -1);
986a4f4d 1326
e9d7417b 1327 vdev->config->del_vqs(vdev);
986a4f4d
JW
1328
1329 virtnet_free_queues(vi);
e9d7417b
JW
1330}
1331
986a4f4d 1332static int virtnet_find_vqs(struct virtnet_info *vi)
3f9c10b0 1333{
986a4f4d
JW
1334 vq_callback_t **callbacks;
1335 struct virtqueue **vqs;
1336 int ret = -ENOMEM;
1337 int i, total_vqs;
1338 const char **names;
1339
1340 /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
1341 * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
1342 * possible control vq.
1343 */
1344 total_vqs = vi->max_queue_pairs * 2 +
1345 virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ);
1346
1347 /* Allocate space for find_vqs parameters */
1348 vqs = kzalloc(total_vqs * sizeof(*vqs), GFP_KERNEL);
1349 if (!vqs)
1350 goto err_vq;
1351 callbacks = kmalloc(total_vqs * sizeof(*callbacks), GFP_KERNEL);
1352 if (!callbacks)
1353 goto err_callback;
1354 names = kmalloc(total_vqs * sizeof(*names), GFP_KERNEL);
1355 if (!names)
1356 goto err_names;
1357
1358 /* Parameters for control virtqueue, if any */
1359 if (vi->has_cvq) {
1360 callbacks[total_vqs - 1] = NULL;
1361 names[total_vqs - 1] = "control";
1362 }
3f9c10b0 1363
986a4f4d
JW
1364 /* Allocate/initialize parameters for send/receive virtqueues */
1365 for (i = 0; i < vi->max_queue_pairs; i++) {
1366 callbacks[rxq2vq(i)] = skb_recv_done;
1367 callbacks[txq2vq(i)] = skb_xmit_done;
1368 sprintf(vi->rq[i].name, "input.%d", i);
1369 sprintf(vi->sq[i].name, "output.%d", i);
1370 names[rxq2vq(i)] = vi->rq[i].name;
1371 names[txq2vq(i)] = vi->sq[i].name;
1372 }
3f9c10b0 1373
986a4f4d
JW
1374 ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs, callbacks,
1375 names);
1376 if (ret)
1377 goto err_find;
3f9c10b0 1378
986a4f4d
JW
1379 if (vi->has_cvq) {
1380 vi->cvq = vqs[total_vqs - 1];
3f9c10b0
AS
1381 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
1382 vi->dev->features |= NETIF_F_HW_VLAN_FILTER;
1383 }
986a4f4d
JW
1384
1385 for (i = 0; i < vi->max_queue_pairs; i++) {
1386 vi->rq[i].vq = vqs[rxq2vq(i)];
1387 vi->sq[i].vq = vqs[txq2vq(i)];
1388 }
1389
1390 kfree(names);
1391 kfree(callbacks);
1392 kfree(vqs);
1393
3f9c10b0 1394 return 0;
986a4f4d
JW
1395
1396err_find:
1397 kfree(names);
1398err_names:
1399 kfree(callbacks);
1400err_callback:
1401 kfree(vqs);
1402err_vq:
1403 return ret;
1404}
1405
1406static int virtnet_alloc_queues(struct virtnet_info *vi)
1407{
1408 int i;
1409
1410 vi->sq = kzalloc(sizeof(*vi->sq) * vi->max_queue_pairs, GFP_KERNEL);
1411 if (!vi->sq)
1412 goto err_sq;
1413 vi->rq = kzalloc(sizeof(*vi->rq) * vi->max_queue_pairs, GFP_KERNEL);
008d4278 1414 if (!vi->rq)
986a4f4d
JW
1415 goto err_rq;
1416
1417 INIT_DELAYED_WORK(&vi->refill, refill_work);
1418 for (i = 0; i < vi->max_queue_pairs; i++) {
1419 vi->rq[i].pages = NULL;
1420 netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll,
1421 napi_weight);
1422
1423 sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg));
1424 sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg));
1425 }
1426
1427 return 0;
1428
1429err_rq:
1430 kfree(vi->sq);
1431err_sq:
1432 return -ENOMEM;
1433}
1434
1435static int init_vqs(struct virtnet_info *vi)
1436{
1437 int ret;
1438
1439 /* Allocate send & receive queues */
1440 ret = virtnet_alloc_queues(vi);
1441 if (ret)
1442 goto err;
1443
1444 ret = virtnet_find_vqs(vi);
1445 if (ret)
1446 goto err_free;
1447
47be2479 1448 get_online_cpus();
8898c21c 1449 virtnet_set_affinity(vi);
47be2479
WG
1450 put_online_cpus();
1451
986a4f4d
JW
1452 return 0;
1453
1454err_free:
1455 virtnet_free_queues(vi);
1456err:
1457 return ret;
3f9c10b0
AS
1458}
1459
296f96fc
RR
1460static int virtnet_probe(struct virtio_device *vdev)
1461{
986a4f4d 1462 int i, err;
296f96fc
RR
1463 struct net_device *dev;
1464 struct virtnet_info *vi;
986a4f4d
JW
1465 u16 max_queue_pairs;
1466
1467 /* Find if host supports multiqueue virtio_net device */
1468 err = virtio_config_val(vdev, VIRTIO_NET_F_MQ,
1469 offsetof(struct virtio_net_config,
1470 max_virtqueue_pairs), &max_queue_pairs);
1471
1472 /* We need at least 2 queue's */
1473 if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
1474 max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
1475 !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
1476 max_queue_pairs = 1;
296f96fc
RR
1477
1478 /* Allocate ourselves a network device with room for our info */
986a4f4d 1479 dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs);
296f96fc
RR
1480 if (!dev)
1481 return -ENOMEM;
1482
1483 /* Set up network device as normal. */
f2f2c8b4 1484 dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE;
76288b4e 1485 dev->netdev_ops = &virtnet_netdev;
296f96fc 1486 dev->features = NETIF_F_HIGHDMA;
3fa2a1df 1487
a9ea3fc6 1488 SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
296f96fc
RR
1489 SET_NETDEV_DEV(dev, &vdev->dev);
1490
1491 /* Do we support "hardware" checksums? */
98e778c9 1492 if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
296f96fc 1493 /* This opens up the world of extra features. */
98e778c9
MM
1494 dev->hw_features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1495 if (csum)
1496 dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1497
1498 if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
1499 dev->hw_features |= NETIF_F_TSO | NETIF_F_UFO
34a48579
RR
1500 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
1501 }
5539ae96 1502 /* Individual feature bits: what can host handle? */
98e778c9
MM
1503 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
1504 dev->hw_features |= NETIF_F_TSO;
1505 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
1506 dev->hw_features |= NETIF_F_TSO6;
1507 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
1508 dev->hw_features |= NETIF_F_TSO_ECN;
1509 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
1510 dev->hw_features |= NETIF_F_UFO;
1511
1512 if (gso)
1513 dev->features |= dev->hw_features & (NETIF_F_ALL_TSO|NETIF_F_UFO);
1514 /* (!csum && gso) case will be fixed by register_netdev() */
296f96fc
RR
1515 }
1516
1517 /* Configuration may specify what MAC to use. Otherwise random. */
77dd7693 1518 if (virtio_config_val_len(vdev, VIRTIO_NET_F_MAC,
a586d4f6 1519 offsetof(struct virtio_net_config, mac),
77dd7693 1520 dev->dev_addr, dev->addr_len) < 0)
f2cedb63 1521 eth_hw_addr_random(dev);
296f96fc
RR
1522
1523 /* Set up our device-specific information */
1524 vi = netdev_priv(dev);
296f96fc
RR
1525 vi->dev = dev;
1526 vi->vdev = vdev;
d9d5dcc8 1527 vdev->priv = vi;
3fa2a1df 1528 vi->stats = alloc_percpu(struct virtnet_stats);
1529 err = -ENOMEM;
1530 if (vi->stats == NULL)
1531 goto free;
1532
47be2479
WG
1533 vi->vq_index = alloc_percpu(int);
1534 if (vi->vq_index == NULL)
1535 goto free_stats;
1536
586d17c5
JW
1537 mutex_init(&vi->config_lock);
1538 vi->config_enable = true;
1539 INIT_WORK(&vi->config_work, virtnet_config_changed_work);
296f96fc 1540
97402b96 1541 /* If we can receive ANY GSO packets, we must allocate large ones. */
8e95a202
JP
1542 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
1543 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
1544 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
97402b96
HX
1545 vi->big_packets = true;
1546
3f2c31d9
MM
1547 if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
1548 vi->mergeable_rx_bufs = true;
1549
986a4f4d
JW
1550 if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
1551 vi->has_cvq = true;
1552
1553 /* Use single tx/rx queue pair as default */
1554 vi->curr_queue_pairs = 1;
1555 vi->max_queue_pairs = max_queue_pairs;
1556
1557 /* Allocate/initialize the rx/tx queues, and invoke find_vqs */
3f9c10b0 1558 err = init_vqs(vi);
d2a7ddda 1559 if (err)
47be2479 1560 goto free_index;
296f96fc 1561
986a4f4d
JW
1562 netif_set_real_num_tx_queues(dev, 1);
1563 netif_set_real_num_rx_queues(dev, 1);
1564
296f96fc
RR
1565 err = register_netdev(dev);
1566 if (err) {
1567 pr_debug("virtio_net: registering device failed\n");
d2a7ddda 1568 goto free_vqs;
296f96fc 1569 }
b3369c1f
RR
1570
1571 /* Last of all, set up some receive buffers. */
986a4f4d
JW
1572 for (i = 0; i < vi->max_queue_pairs; i++) {
1573 try_fill_recv(&vi->rq[i], GFP_KERNEL);
1574
1575 /* If we didn't even get one input buffer, we're useless. */
1576 if (vi->rq[i].num == 0) {
1577 free_unused_bufs(vi);
1578 err = -ENOMEM;
1579 goto free_recv_bufs;
1580 }
b3369c1f
RR
1581 }
1582
8de4b2f3
WG
1583 vi->nb.notifier_call = &virtnet_cpu_callback;
1584 err = register_hotcpu_notifier(&vi->nb);
1585 if (err) {
1586 pr_debug("virtio_net: registering cpu notifier failed\n");
1587 goto free_recv_bufs;
1588 }
1589
167c25e4
JW
1590 /* Assume link up if device can't report link status,
1591 otherwise get link status from config. */
1592 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
1593 netif_carrier_off(dev);
3b07e9ca 1594 schedule_work(&vi->config_work);
167c25e4
JW
1595 } else {
1596 vi->status = VIRTIO_NET_S_LINK_UP;
1597 netif_carrier_on(dev);
1598 }
9f4d26d0 1599
986a4f4d
JW
1600 pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
1601 dev->name, max_queue_pairs);
1602
296f96fc
RR
1603 return 0;
1604
986a4f4d
JW
1605free_recv_bufs:
1606 free_receive_bufs(vi);
b3369c1f 1607 unregister_netdev(dev);
d2a7ddda 1608free_vqs:
986a4f4d 1609 cancel_delayed_work_sync(&vi->refill);
e9d7417b 1610 virtnet_del_vqs(vi);
47be2479
WG
1611free_index:
1612 free_percpu(vi->vq_index);
3fa2a1df 1613free_stats:
1614 free_percpu(vi->stats);
296f96fc
RR
1615free:
1616 free_netdev(dev);
1617 return err;
1618}
1619
04486ed0 1620static void remove_vq_common(struct virtnet_info *vi)
296f96fc 1621{
04486ed0 1622 vi->vdev->config->reset(vi->vdev);
830a8a97
SM
1623
1624 /* Free unused buffers in both send and recv, if any. */
9ab86bbc 1625 free_unused_bufs(vi);
fb6813f4 1626
986a4f4d 1627 free_receive_bufs(vi);
d2a7ddda 1628
986a4f4d 1629 virtnet_del_vqs(vi);
04486ed0
AS
1630}
1631
8cc085d6 1632static void virtnet_remove(struct virtio_device *vdev)
04486ed0
AS
1633{
1634 struct virtnet_info *vi = vdev->priv;
1635
8de4b2f3
WG
1636 unregister_hotcpu_notifier(&vi->nb);
1637
586d17c5
JW
1638 /* Prevent config work handler from accessing the device. */
1639 mutex_lock(&vi->config_lock);
1640 vi->config_enable = false;
1641 mutex_unlock(&vi->config_lock);
1642
04486ed0
AS
1643 unregister_netdev(vi->dev);
1644
1645 remove_vq_common(vi);
fb6813f4 1646
586d17c5
JW
1647 flush_work(&vi->config_work);
1648
47be2479 1649 free_percpu(vi->vq_index);
2e66f55b 1650 free_percpu(vi->stats);
74b2553f 1651 free_netdev(vi->dev);
296f96fc
RR
1652}
1653
0741bcb5
AS
1654#ifdef CONFIG_PM
1655static int virtnet_freeze(struct virtio_device *vdev)
1656{
1657 struct virtnet_info *vi = vdev->priv;
986a4f4d 1658 int i;
0741bcb5 1659
586d17c5
JW
1660 /* Prevent config work handler from accessing the device */
1661 mutex_lock(&vi->config_lock);
1662 vi->config_enable = false;
1663 mutex_unlock(&vi->config_lock);
1664
0741bcb5
AS
1665 netif_device_detach(vi->dev);
1666 cancel_delayed_work_sync(&vi->refill);
1667
1668 if (netif_running(vi->dev))
986a4f4d
JW
1669 for (i = 0; i < vi->max_queue_pairs; i++) {
1670 napi_disable(&vi->rq[i].napi);
1671 netif_napi_del(&vi->rq[i].napi);
1672 }
0741bcb5
AS
1673
1674 remove_vq_common(vi);
1675
586d17c5
JW
1676 flush_work(&vi->config_work);
1677
0741bcb5
AS
1678 return 0;
1679}
1680
1681static int virtnet_restore(struct virtio_device *vdev)
1682{
1683 struct virtnet_info *vi = vdev->priv;
986a4f4d 1684 int err, i;
0741bcb5
AS
1685
1686 err = init_vqs(vi);
1687 if (err)
1688 return err;
1689
1690 if (netif_running(vi->dev))
986a4f4d
JW
1691 for (i = 0; i < vi->max_queue_pairs; i++)
1692 virtnet_napi_enable(&vi->rq[i]);
0741bcb5
AS
1693
1694 netif_device_attach(vi->dev);
1695
986a4f4d
JW
1696 for (i = 0; i < vi->max_queue_pairs; i++)
1697 if (!try_fill_recv(&vi->rq[i], GFP_KERNEL))
1698 schedule_delayed_work(&vi->refill, 0);
0741bcb5 1699
586d17c5
JW
1700 mutex_lock(&vi->config_lock);
1701 vi->config_enable = true;
1702 mutex_unlock(&vi->config_lock);
1703
986a4f4d
JW
1704 virtnet_set_queues(vi, vi->curr_queue_pairs);
1705
0741bcb5
AS
1706 return 0;
1707}
1708#endif
1709
296f96fc
RR
1710static struct virtio_device_id id_table[] = {
1711 { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
1712 { 0 },
1713};
1714
c45a6816 1715static unsigned int features[] = {
5e4fe5c4
MM
1716 VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
1717 VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
c45a6816 1718 VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
97402b96 1719 VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
5c516751 1720 VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO,
2a41f71d 1721 VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ,
0bde9569 1722 VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN,
986a4f4d 1723 VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ,
7e58d5ae 1724 VIRTIO_NET_F_CTRL_MAC_ADDR,
c45a6816
RR
1725};
1726
22402529 1727static struct virtio_driver virtio_net_driver = {
c45a6816
RR
1728 .feature_table = features,
1729 .feature_table_size = ARRAY_SIZE(features),
296f96fc
RR
1730 .driver.name = KBUILD_MODNAME,
1731 .driver.owner = THIS_MODULE,
1732 .id_table = id_table,
1733 .probe = virtnet_probe,
8cc085d6 1734 .remove = virtnet_remove,
9f4d26d0 1735 .config_changed = virtnet_config_changed,
0741bcb5
AS
1736#ifdef CONFIG_PM
1737 .freeze = virtnet_freeze,
1738 .restore = virtnet_restore,
1739#endif
296f96fc
RR
1740};
1741
1742static int __init init(void)
1743{
22402529 1744 return register_virtio_driver(&virtio_net_driver);
296f96fc
RR
1745}
1746
1747static void __exit fini(void)
1748{
22402529 1749 unregister_virtio_driver(&virtio_net_driver);
296f96fc
RR
1750}
1751module_init(init);
1752module_exit(fini);
1753
1754MODULE_DEVICE_TABLE(virtio, id_table);
1755MODULE_DESCRIPTION("Virtio network driver");
1756MODULE_LICENSE("GPL");