blob: 7b687469199b58357a74490cb266e4d47534c097 [file] [log] [blame]
Rusty Russell48925e32009-09-24 09:59:20 -06001/* A network driver using virtio.
Rusty Russell296f96f2007-10-22 11:03:37 +10002 *
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
Jeff Kirsheradf8d3f2013-12-06 06:28:47 -080016 * along with this program; if not, see <http://www.gnu.org/licenses/>.
Rusty Russell296f96f2007-10-22 11:03:37 +100017 */
18//#define DEBUG
19#include <linux/netdevice.h>
20#include <linux/etherdevice.h>
Herbert Xua9ea3fc2008-04-18 11:21:42 +080021#include <linux/ethtool.h>
Rusty Russell296f96f2007-10-22 11:03:37 +100022#include <linux/module.h>
23#include <linux/virtio.h>
24#include <linux/virtio_net.h>
25#include <linux/scatterlist.h>
Alex Williamsone918085a2009-01-25 18:06:26 -080026#include <linux/if_vlan.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090027#include <linux/slab.h>
Wanlong Gao8de4b2f2013-01-24 23:51:31 +000028#include <linux/cpu.h>
Michael Daltonab7db912014-01-16 22:23:27 -080029#include <linux/average.h>
Rusty Russell296f96f2007-10-22 11:03:37 +100030
Amerigo Wangd34710e2013-05-09 19:50:51 +000031static int napi_weight = NAPI_POLL_WEIGHT;
Dor Laor6c0cd7c2007-12-16 15:19:43 +020032module_param(napi_weight, int, 0444);
33
Rusty Russelleb939922011-12-19 14:08:01 +000034static bool csum = true, gso = true;
Rusty Russell34a48572008-02-04 23:50:02 -050035module_param(csum, bool, 0444);
36module_param(gso, bool, 0444);
37
Rusty Russell296f96f2007-10-22 11:03:37 +100038/* FIXME: MTU in config. */
Michael Dalton5061de32013-11-14 10:41:04 -080039#define GOOD_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -080040#define GOOD_COPY_LEN 128
Rusty Russell296f96f2007-10-22 11:03:37 +100041
Michael Daltonab7db912014-01-16 22:23:27 -080042/* Weight used for the RX packet size EWMA. The average packet size is used to
43 * determine the packet buffer size when refilling RX rings. As the entire RX
44 * ring may be refilled at once, the weight is chosen so that the EWMA will be
45 * insensitive to short-term, transient changes in packet size.
46 */
47#define RECEIVE_AVG_WEIGHT 64
48
49/* Minimum alignment for mergeable packet buffers. */
50#define MERGEABLE_BUFFER_ALIGN max(L1_CACHE_BYTES, 256)
51
Rick Jones66846042011-11-14 14:17:08 +000052#define VIRTNET_DRIVER_VERSION "1.0.0"
Alex Williamson2a41f712009-02-04 09:02:34 +000053
stephen hemminger3fa2a1d2011-06-15 06:36:29 +000054struct virtnet_stats {
Eric Dumazet83a27052012-06-05 22:35:24 +000055 struct u64_stats_sync tx_syncp;
56 struct u64_stats_sync rx_syncp;
stephen hemminger3fa2a1d2011-06-15 06:36:29 +000057 u64 tx_bytes;
58 u64 tx_packets;
59
60 u64 rx_bytes;
61 u64 rx_packets;
62};
63
Jason Wange9d74172012-12-07 07:04:55 +000064/* Internal representation of a send virtqueue */
65struct send_queue {
66 /* Virtqueue associated with this send _queue */
67 struct virtqueue *vq;
68
69 /* TX: fragments + linear part + virtio header */
70 struct scatterlist sg[MAX_SKB_FRAGS + 2];
Jason Wang986a4f42012-12-07 07:04:56 +000071
72 /* Name of the send queue: output.$index */
73 char name[40];
Jason Wange9d74172012-12-07 07:04:55 +000074};
75
76/* Internal representation of a receive virtqueue */
77struct receive_queue {
78 /* Virtqueue associated with this receive_queue */
79 struct virtqueue *vq;
80
Rusty Russell296f96f2007-10-22 11:03:37 +100081 struct napi_struct napi;
82
Jason Wange9d74172012-12-07 07:04:55 +000083 /* Chain pages by the private ptr. */
84 struct page *pages;
85
Michael Daltonab7db912014-01-16 22:23:27 -080086 /* Average packet length for mergeable receive buffers. */
87 struct ewma mrg_avg_pkt_len;
88
Michael Daltonfb518792014-01-16 22:23:26 -080089 /* Page frag for packet buffer allocation. */
90 struct page_frag alloc_frag;
91
Jason Wange9d74172012-12-07 07:04:55 +000092 /* RX: fragments + linear part + virtio header */
93 struct scatterlist sg[MAX_SKB_FRAGS + 2];
Jason Wang986a4f42012-12-07 07:04:56 +000094
95 /* Name of this receive queue: input.$index */
96 char name[40];
Jason Wange9d74172012-12-07 07:04:55 +000097};
98
99struct virtnet_info {
100 struct virtio_device *vdev;
101 struct virtqueue *cvq;
102 struct net_device *dev;
Jason Wang986a4f42012-12-07 07:04:56 +0000103 struct send_queue *sq;
104 struct receive_queue *rq;
Jason Wange9d74172012-12-07 07:04:55 +0000105 unsigned int status;
106
Jason Wang986a4f42012-12-07 07:04:56 +0000107 /* Max # of queue pairs supported by the device */
108 u16 max_queue_pairs;
109
110 /* # of queue pairs currently used by the driver */
111 u16 curr_queue_pairs;
112
Herbert Xu97402b92008-04-18 11:24:27 +0800113 /* I like... big packets and I cannot lie! */
114 bool big_packets;
115
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800116 /* Host will merge rx buffers for big packets (shake it! shake it!) */
117 bool mergeable_rx_bufs;
118
Jason Wang986a4f42012-12-07 07:04:56 +0000119 /* Has control virtqueue */
120 bool has_cvq;
121
Michael S. Tsirkine7428e92013-07-25 10:20:23 +0930122 /* Host can handle any s/g split between our header and packet data */
123 bool any_header_sg;
124
Jason Wang586d17c2012-04-11 20:43:52 +0000125 /* enable config space updates */
126 bool config_enable;
127
stephen hemminger3fa2a1d2011-06-15 06:36:29 +0000128 /* Active statistics */
129 struct virtnet_stats __percpu *stats;
130
Rusty Russell3161e452009-08-26 12:22:32 -0700131 /* Work struct for refilling if we run low on memory. */
132 struct delayed_work refill;
133
Jason Wang586d17c2012-04-11 20:43:52 +0000134 /* Work struct for config space updates */
135 struct work_struct config_work;
136
137 /* Lock for config space updates */
138 struct mutex config_lock;
Jason Wang986a4f42012-12-07 07:04:56 +0000139
140 /* Does the affinity hint is set for virtqueues? */
141 bool affinity_hint_set;
Wanlong Gao47be2472013-01-24 23:51:29 +0000142
Wanlong Gao8de4b2f2013-01-24 23:51:31 +0000143 /* CPU hot plug notifier */
144 struct notifier_block nb;
Rusty Russell296f96f2007-10-22 11:03:37 +1000145};
146
Rusty Russellb3f24692009-09-24 09:59:19 -0600147struct skb_vnet_hdr {
148 union {
149 struct virtio_net_hdr hdr;
150 struct virtio_net_hdr_mrg_rxbuf mhdr;
151 };
152};
153
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000154struct padded_vnet_hdr {
155 struct virtio_net_hdr hdr;
156 /*
157 * virtio_net_hdr should be in a separated sg buffer because of a
158 * QEMU bug, and data sg buffer shares same page with this header sg.
159 * This padding makes next sg 16 byte aligned after virtio_net_hdr.
160 */
161 char padding[6];
162};
163
Jason Wang986a4f42012-12-07 07:04:56 +0000164/* Converting between virtqueue no. and kernel tx/rx queue no.
165 * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq
166 */
167static int vq2txq(struct virtqueue *vq)
168{
Rusty Russell9d0ca6e2013-03-21 14:17:34 +0000169 return (vq->index - 1) / 2;
Jason Wang986a4f42012-12-07 07:04:56 +0000170}
171
172static int txq2vq(int txq)
173{
174 return txq * 2 + 1;
175}
176
177static int vq2rxq(struct virtqueue *vq)
178{
Rusty Russell9d0ca6e2013-03-21 14:17:34 +0000179 return vq->index / 2;
Jason Wang986a4f42012-12-07 07:04:56 +0000180}
181
182static int rxq2vq(int rxq)
183{
184 return rxq * 2;
185}
186
Rusty Russellb3f24692009-09-24 09:59:19 -0600187static inline struct skb_vnet_hdr *skb_vnet_hdr(struct sk_buff *skb)
Rusty Russell296f96f2007-10-22 11:03:37 +1000188{
Rusty Russellb3f24692009-09-24 09:59:19 -0600189 return (struct skb_vnet_hdr *)skb->cb;
Rusty Russell296f96f2007-10-22 11:03:37 +1000190}
191
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000192/*
193 * private is used to chain pages for big packets, put the whole
194 * most recent used list in the beginning for reuse
195 */
Jason Wange9d74172012-12-07 07:04:55 +0000196static void give_pages(struct receive_queue *rq, struct page *page)
Rusty Russellfb6813f2008-07-25 12:06:01 -0500197{
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000198 struct page *end;
199
Jason Wange9d74172012-12-07 07:04:55 +0000200 /* Find end of list, sew whole thing into vi->rq.pages. */
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000201 for (end = page; end->private; end = (struct page *)end->private);
Jason Wange9d74172012-12-07 07:04:55 +0000202 end->private = (unsigned long)rq->pages;
203 rq->pages = page;
Rusty Russellfb6813f2008-07-25 12:06:01 -0500204}
205
Jason Wange9d74172012-12-07 07:04:55 +0000206static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask)
Rusty Russellfb6813f2008-07-25 12:06:01 -0500207{
Jason Wange9d74172012-12-07 07:04:55 +0000208 struct page *p = rq->pages;
Rusty Russellfb6813f2008-07-25 12:06:01 -0500209
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000210 if (p) {
Jason Wange9d74172012-12-07 07:04:55 +0000211 rq->pages = (struct page *)p->private;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000212 /* clear private here, it is used to chain pages */
213 p->private = 0;
214 } else
Rusty Russellfb6813f2008-07-25 12:06:01 -0500215 p = alloc_page(gfp_mask);
216 return p;
217}
218
Jason Wange9d74172012-12-07 07:04:55 +0000219static void skb_xmit_done(struct virtqueue *vq)
Rusty Russell296f96f2007-10-22 11:03:37 +1000220{
Jason Wange9d74172012-12-07 07:04:55 +0000221 struct virtnet_info *vi = vq->vdev->priv;
Rusty Russell296f96f2007-10-22 11:03:37 +1000222
Rusty Russell2cb9c6b2008-02-04 23:50:07 -0500223 /* Suppress further interrupts. */
Jason Wange9d74172012-12-07 07:04:55 +0000224 virtqueue_disable_cb(vq);
Rusty Russell11a3a152008-05-26 17:48:13 +1000225
Rusty Russell363f1512008-06-08 20:51:55 +1000226 /* We were probably waiting for more output buffers. */
Jason Wang986a4f42012-12-07 07:04:56 +0000227 netif_wake_subqueue(vi->dev, vq2txq(vq));
Rusty Russell296f96f2007-10-22 11:03:37 +1000228}
229
Michael Daltonab7db912014-01-16 22:23:27 -0800230static unsigned int mergeable_ctx_to_buf_truesize(unsigned long mrg_ctx)
231{
232 unsigned int truesize = mrg_ctx & (MERGEABLE_BUFFER_ALIGN - 1);
233 return (truesize + 1) * MERGEABLE_BUFFER_ALIGN;
234}
235
236static void *mergeable_ctx_to_buf_address(unsigned long mrg_ctx)
237{
238 return (void *)(mrg_ctx & -MERGEABLE_BUFFER_ALIGN);
239
240}
241
242static unsigned long mergeable_buf_to_ctx(void *buf, unsigned int truesize)
243{
244 unsigned int size = truesize / MERGEABLE_BUFFER_ALIGN;
245 return (unsigned long)buf | (size - 1);
246}
247
Mike Waychison34646452012-01-04 12:52:32 +0000248/* Called from bottom half context */
Jason Wange9d74172012-12-07 07:04:55 +0000249static struct sk_buff *page_to_skb(struct receive_queue *rq,
Michael Dalton2613af02013-10-28 15:44:18 -0700250 struct page *page, unsigned int offset,
251 unsigned int len, unsigned int truesize)
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000252{
Jason Wange9d74172012-12-07 07:04:55 +0000253 struct virtnet_info *vi = rq->vq->vdev->priv;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000254 struct sk_buff *skb;
255 struct skb_vnet_hdr *hdr;
Michael Dalton2613af02013-10-28 15:44:18 -0700256 unsigned int copy, hdr_len, hdr_padded_len;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000257 char *p;
258
Michael Dalton2613af02013-10-28 15:44:18 -0700259 p = page_address(page) + offset;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000260
261 /* copy small packet so we can reuse these pages for small data */
262 skb = netdev_alloc_skb_ip_align(vi->dev, GOOD_COPY_LEN);
263 if (unlikely(!skb))
264 return NULL;
265
266 hdr = skb_vnet_hdr(skb);
267
268 if (vi->mergeable_rx_bufs) {
269 hdr_len = sizeof hdr->mhdr;
Michael Dalton2613af02013-10-28 15:44:18 -0700270 hdr_padded_len = sizeof hdr->mhdr;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000271 } else {
272 hdr_len = sizeof hdr->hdr;
Michael Dalton2613af02013-10-28 15:44:18 -0700273 hdr_padded_len = sizeof(struct padded_vnet_hdr);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000274 }
275
276 memcpy(hdr, p, hdr_len);
277
278 len -= hdr_len;
Michael Dalton2613af02013-10-28 15:44:18 -0700279 offset += hdr_padded_len;
280 p += hdr_padded_len;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000281
282 copy = len;
283 if (copy > skb_tailroom(skb))
284 copy = skb_tailroom(skb);
285 memcpy(skb_put(skb, copy), p, copy);
286
287 len -= copy;
288 offset += copy;
289
Michael Dalton2613af02013-10-28 15:44:18 -0700290 if (vi->mergeable_rx_bufs) {
291 if (len)
292 skb_add_rx_frag(skb, 0, page, offset, len, truesize);
293 else
294 put_page(page);
295 return skb;
296 }
297
Sasha Levine878d782011-09-28 04:40:54 +0000298 /*
299 * Verify that we can indeed put this data into a skb.
300 * This is here to handle cases when the device erroneously
301 * tries to receive more than is possible. This is usually
302 * the case of a broken device.
303 */
304 if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) {
Amerigo Wangbe443892012-11-08 17:47:28 +0000305 net_dbg_ratelimited("%s: too much data\n", skb->dev->name);
Sasha Levine878d782011-09-28 04:40:54 +0000306 dev_kfree_skb(skb);
307 return NULL;
308 }
Michael Dalton2613af02013-10-28 15:44:18 -0700309 BUG_ON(offset >= PAGE_SIZE);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000310 while (len) {
Michael Dalton2613af02013-10-28 15:44:18 -0700311 unsigned int frag_size = min((unsigned)PAGE_SIZE - offset, len);
312 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page, offset,
313 frag_size, truesize);
314 len -= frag_size;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000315 page = (struct page *)page->private;
316 offset = 0;
317 }
318
319 if (page)
Jason Wange9d74172012-12-07 07:04:55 +0000320 give_pages(rq, page);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000321
322 return skb;
323}
324
Michael S. Tsirkinf1211592013-11-28 13:30:59 +0200325static struct sk_buff *receive_small(void *buf, unsigned int len)
326{
327 struct sk_buff * skb = buf;
328
329 len -= sizeof(struct virtio_net_hdr);
330 skb_trim(skb, len);
331
332 return skb;
333}
334
335static struct sk_buff *receive_big(struct net_device *dev,
336 struct receive_queue *rq,
337 void *buf,
338 unsigned int len)
339{
340 struct page *page = buf;
341 struct sk_buff *skb = page_to_skb(rq, page, 0, len, PAGE_SIZE);
342
343 if (unlikely(!skb))
344 goto err;
345
346 return skb;
347
348err:
349 dev->stats.rx_dropped++;
350 give_pages(rq, page);
351 return NULL;
352}
353
Michael S. Tsirkin8fc3b9e2013-11-28 13:30:55 +0200354static struct sk_buff *receive_mergeable(struct net_device *dev,
355 struct receive_queue *rq,
Michael Daltonab7db912014-01-16 22:23:27 -0800356 unsigned long ctx,
Michael S. Tsirkin8fc3b9e2013-11-28 13:30:55 +0200357 unsigned int len)
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000358{
Michael Daltonab7db912014-01-16 22:23:27 -0800359 void *buf = mergeable_ctx_to_buf_address(ctx);
Michael S. Tsirkin8fc3b9e2013-11-28 13:30:55 +0200360 struct skb_vnet_hdr *hdr = buf;
361 int num_buf = hdr->mhdr.num_buffers;
362 struct page *page = virt_to_head_page(buf);
363 int offset = buf - page_address(page);
Michael Daltonab7db912014-01-16 22:23:27 -0800364 unsigned int truesize = max(len, mergeable_ctx_to_buf_truesize(ctx));
365
Michael Daltonfb518792014-01-16 22:23:26 -0800366 struct sk_buff *head_skb = page_to_skb(rq, page, offset, len, truesize);
Michael Dalton2613af02013-10-28 15:44:18 -0700367 struct sk_buff *curr_skb = head_skb;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000368
Michael S. Tsirkin8fc3b9e2013-11-28 13:30:55 +0200369 if (unlikely(!curr_skb))
370 goto err_skb;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000371 while (--num_buf) {
Michael S. Tsirkin8fc3b9e2013-11-28 13:30:55 +0200372 int num_skb_frags;
373
Michael Daltonab7db912014-01-16 22:23:27 -0800374 ctx = (unsigned long)virtqueue_get_buf(rq->vq, &len);
375 if (unlikely(!ctx)) {
Michael S. Tsirkin8fc3b9e2013-11-28 13:30:55 +0200376 pr_debug("%s: rx error: %d buffers out of %d missing\n",
377 dev->name, num_buf, hdr->mhdr.num_buffers);
378 dev->stats.rx_length_errors++;
379 goto err_buf;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000380 }
Michael S. Tsirkin8fc3b9e2013-11-28 13:30:55 +0200381
Michael Daltonab7db912014-01-16 22:23:27 -0800382 buf = mergeable_ctx_to_buf_address(ctx);
Michael S. Tsirkin8fc3b9e2013-11-28 13:30:55 +0200383 page = virt_to_head_page(buf);
Michael S. Tsirkin8fc3b9e2013-11-28 13:30:55 +0200384
385 num_skb_frags = skb_shinfo(curr_skb)->nr_frags;
Michael Dalton2613af02013-10-28 15:44:18 -0700386 if (unlikely(num_skb_frags == MAX_SKB_FRAGS)) {
387 struct sk_buff *nskb = alloc_skb(0, GFP_ATOMIC);
Michael S. Tsirkin8fc3b9e2013-11-28 13:30:55 +0200388
389 if (unlikely(!nskb))
390 goto err_skb;
Michael Dalton2613af02013-10-28 15:44:18 -0700391 if (curr_skb == head_skb)
392 skb_shinfo(curr_skb)->frag_list = nskb;
393 else
394 curr_skb->next = nskb;
395 curr_skb = nskb;
396 head_skb->truesize += nskb->truesize;
397 num_skb_frags = 0;
398 }
Michael Daltonab7db912014-01-16 22:23:27 -0800399 truesize = max(len, mergeable_ctx_to_buf_truesize(ctx));
Michael Dalton2613af02013-10-28 15:44:18 -0700400 if (curr_skb != head_skb) {
401 head_skb->data_len += len;
402 head_skb->len += len;
Michael Daltonfb518792014-01-16 22:23:26 -0800403 head_skb->truesize += truesize;
Michael Dalton2613af02013-10-28 15:44:18 -0700404 }
Michael S. Tsirkin8fc3b9e2013-11-28 13:30:55 +0200405 offset = buf - page_address(page);
Jason Wangba275242013-11-01 14:07:48 +0800406 if (skb_can_coalesce(curr_skb, num_skb_frags, page, offset)) {
407 put_page(page);
408 skb_coalesce_rx_frag(curr_skb, num_skb_frags - 1,
Michael Daltonfb518792014-01-16 22:23:26 -0800409 len, truesize);
Jason Wangba275242013-11-01 14:07:48 +0800410 } else {
411 skb_add_rx_frag(curr_skb, num_skb_frags, page,
Michael Daltonfb518792014-01-16 22:23:26 -0800412 offset, len, truesize);
Jason Wangba275242013-11-01 14:07:48 +0800413 }
Michael S. Tsirkin8fc3b9e2013-11-28 13:30:55 +0200414 }
415
Michael Daltonab7db912014-01-16 22:23:27 -0800416 ewma_add(&rq->mrg_avg_pkt_len, head_skb->len);
Michael S. Tsirkin8fc3b9e2013-11-28 13:30:55 +0200417 return head_skb;
418
419err_skb:
420 put_page(page);
421 while (--num_buf) {
Michael Daltonab7db912014-01-16 22:23:27 -0800422 ctx = (unsigned long)virtqueue_get_buf(rq->vq, &len);
423 if (unlikely(!ctx)) {
Michael S. Tsirkin8fc3b9e2013-11-28 13:30:55 +0200424 pr_debug("%s: rx error: %d buffers missing\n",
425 dev->name, num_buf);
426 dev->stats.rx_length_errors++;
427 break;
428 }
Michael Daltonab7db912014-01-16 22:23:27 -0800429 page = virt_to_head_page(mergeable_ctx_to_buf_address(ctx));
Michael S. Tsirkin8fc3b9e2013-11-28 13:30:55 +0200430 put_page(page);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000431 }
Michael S. Tsirkin8fc3b9e2013-11-28 13:30:55 +0200432err_buf:
433 dev->stats.rx_dropped++;
434 dev_kfree_skb(head_skb);
435 return NULL;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000436}
437
Jason Wange9d74172012-12-07 07:04:55 +0000438static void receive_buf(struct receive_queue *rq, void *buf, unsigned int len)
Rusty Russell296f96f2007-10-22 11:03:37 +1000439{
Jason Wange9d74172012-12-07 07:04:55 +0000440 struct virtnet_info *vi = rq->vq->vdev->priv;
441 struct net_device *dev = vi->dev;
Eric Dumazet58472a72012-02-13 06:53:41 +0000442 struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000443 struct sk_buff *skb;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000444 struct skb_vnet_hdr *hdr;
Rusty Russell296f96f2007-10-22 11:03:37 +1000445
446 if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
447 pr_debug("%s: short packet %i\n", dev->name, len);
448 dev->stats.rx_length_errors++;
Michael Daltonab7db912014-01-16 22:23:27 -0800449 if (vi->mergeable_rx_bufs) {
450 unsigned long ctx = (unsigned long)buf;
451 void *base = mergeable_ctx_to_buf_address(ctx);
452 put_page(virt_to_head_page(base));
453 } else if (vi->big_packets) {
Michael Dalton98bfd232013-12-05 13:14:05 -0800454 give_pages(rq, buf);
Michael Daltonab7db912014-01-16 22:23:27 -0800455 } else {
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000456 dev_kfree_skb(buf);
Michael Daltonab7db912014-01-16 22:23:27 -0800457 }
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000458 return;
Rusty Russell296f96f2007-10-22 11:03:37 +1000459 }
Rusty Russell296f96f2007-10-22 11:03:37 +1000460
Michael S. Tsirkinf1211592013-11-28 13:30:59 +0200461 if (vi->mergeable_rx_bufs)
Michael Daltonab7db912014-01-16 22:23:27 -0800462 skb = receive_mergeable(dev, rq, (unsigned long)buf, len);
Michael S. Tsirkinf1211592013-11-28 13:30:59 +0200463 else if (vi->big_packets)
464 skb = receive_big(dev, rq, buf, len);
465 else
466 skb = receive_small(buf, len);
467
468 if (unlikely(!skb))
469 return;
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800470
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000471 hdr = skb_vnet_hdr(skb);
stephen hemminger3fa2a1d2011-06-15 06:36:29 +0000472
Eric Dumazet83a27052012-06-05 22:35:24 +0000473 u64_stats_update_begin(&stats->rx_syncp);
stephen hemminger3fa2a1d2011-06-15 06:36:29 +0000474 stats->rx_bytes += skb->len;
475 stats->rx_packets++;
Eric Dumazet83a27052012-06-05 22:35:24 +0000476 u64_stats_update_end(&stats->rx_syncp);
Rusty Russell296f96f2007-10-22 11:03:37 +1000477
Rusty Russellb3f24692009-09-24 09:59:19 -0600478 if (hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
Rusty Russell296f96f2007-10-22 11:03:37 +1000479 pr_debug("Needs csum!\n");
Rusty Russellb3f24692009-09-24 09:59:19 -0600480 if (!skb_partial_csum_set(skb,
481 hdr->hdr.csum_start,
482 hdr->hdr.csum_offset))
Rusty Russell296f96f2007-10-22 11:03:37 +1000483 goto frame_err;
Jason Wang10a8d942011-06-10 00:56:17 +0000484 } else if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) {
485 skb->ip_summed = CHECKSUM_UNNECESSARY;
Rusty Russell296f96f2007-10-22 11:03:37 +1000486 }
487
Mark McLoughlin23cde762008-06-08 20:49:00 +1000488 skb->protocol = eth_type_trans(skb, dev);
489 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
490 ntohs(skb->protocol), skb->len, skb->pkt_type);
491
Rusty Russellb3f24692009-09-24 09:59:19 -0600492 if (hdr->hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
Rusty Russell296f96f2007-10-22 11:03:37 +1000493 pr_debug("GSO!\n");
Rusty Russellb3f24692009-09-24 09:59:19 -0600494 switch (hdr->hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
Rusty Russell296f96f2007-10-22 11:03:37 +1000495 case VIRTIO_NET_HDR_GSO_TCPV4:
Pravin B Shelarc9af6db2013-02-11 09:27:41 +0000496 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
Rusty Russell296f96f2007-10-22 11:03:37 +1000497 break;
Rusty Russell296f96f2007-10-22 11:03:37 +1000498 case VIRTIO_NET_HDR_GSO_UDP:
Pravin B Shelarc9af6db2013-02-11 09:27:41 +0000499 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
Rusty Russell296f96f2007-10-22 11:03:37 +1000500 break;
501 case VIRTIO_NET_HDR_GSO_TCPV6:
Pravin B Shelarc9af6db2013-02-11 09:27:41 +0000502 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
Rusty Russell296f96f2007-10-22 11:03:37 +1000503 break;
504 default:
Amerigo Wangbe443892012-11-08 17:47:28 +0000505 net_warn_ratelimited("%s: bad gso type %u.\n",
506 dev->name, hdr->hdr.gso_type);
Rusty Russell296f96f2007-10-22 11:03:37 +1000507 goto frame_err;
508 }
509
Rusty Russellb3f24692009-09-24 09:59:19 -0600510 if (hdr->hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
Pravin B Shelarc9af6db2013-02-11 09:27:41 +0000511 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
Rusty Russell34a48572008-02-04 23:50:02 -0500512
Rusty Russellb3f24692009-09-24 09:59:19 -0600513 skb_shinfo(skb)->gso_size = hdr->hdr.gso_size;
Rusty Russell296f96f2007-10-22 11:03:37 +1000514 if (skb_shinfo(skb)->gso_size == 0) {
Amerigo Wangbe443892012-11-08 17:47:28 +0000515 net_warn_ratelimited("%s: zero gso size.\n", dev->name);
Rusty Russell296f96f2007-10-22 11:03:37 +1000516 goto frame_err;
517 }
518
519 /* Header must be checked, and gso_segs computed. */
520 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
521 skb_shinfo(skb)->gso_segs = 0;
522 }
523
524 netif_receive_skb(skb);
525 return;
526
527frame_err:
528 dev->stats.rx_frame_errors++;
Rusty Russell296f96f2007-10-22 11:03:37 +1000529 dev_kfree_skb(skb);
530}
531
Jason Wange9d74172012-12-07 07:04:55 +0000532static int add_recvbuf_small(struct receive_queue *rq, gfp_t gfp)
Rusty Russell296f96f2007-10-22 11:03:37 +1000533{
Jason Wange9d74172012-12-07 07:04:55 +0000534 struct virtnet_info *vi = rq->vq->vdev->priv;
Rusty Russell296f96f2007-10-22 11:03:37 +1000535 struct sk_buff *skb;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000536 struct skb_vnet_hdr *hdr;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000537 int err;
Rusty Russell296f96f2007-10-22 11:03:37 +1000538
Michael Dalton5061de32013-11-14 10:41:04 -0800539 skb = __netdev_alloc_skb_ip_align(vi->dev, GOOD_PACKET_LEN, gfp);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000540 if (unlikely(!skb))
541 return -ENOMEM;
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800542
Michael Dalton5061de32013-11-14 10:41:04 -0800543 skb_put(skb, GOOD_PACKET_LEN);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000544
545 hdr = skb_vnet_hdr(skb);
Jason Wange9d74172012-12-07 07:04:55 +0000546 sg_set_buf(rq->sg, &hdr->hdr, sizeof hdr->hdr);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000547
Jason Wange9d74172012-12-07 07:04:55 +0000548 skb_to_sgvec(skb, rq->sg + 1, 0, skb->len);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000549
Rusty Russell9dc7b9e2013-03-20 15:44:28 +1030550 err = virtqueue_add_inbuf(rq->vq, rq->sg, 2, skb, gfp);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000551 if (err < 0)
552 dev_kfree_skb(skb);
553
554 return err;
555}
556
Jason Wange9d74172012-12-07 07:04:55 +0000557static int add_recvbuf_big(struct receive_queue *rq, gfp_t gfp)
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000558{
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000559 struct page *first, *list = NULL;
560 char *p;
561 int i, err, offset;
562
Jason Wange9d74172012-12-07 07:04:55 +0000563 /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000564 for (i = MAX_SKB_FRAGS + 1; i > 1; --i) {
Jason Wange9d74172012-12-07 07:04:55 +0000565 first = get_a_page(rq, gfp);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000566 if (!first) {
567 if (list)
Jason Wange9d74172012-12-07 07:04:55 +0000568 give_pages(rq, list);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000569 return -ENOMEM;
Rusty Russell3161e452009-08-26 12:22:32 -0700570 }
Jason Wange9d74172012-12-07 07:04:55 +0000571 sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE);
Rusty Russell296f96f2007-10-22 11:03:37 +1000572
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000573 /* chain new page in list head to match sg */
574 first->private = (unsigned long)list;
575 list = first;
576 }
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800577
Jason Wange9d74172012-12-07 07:04:55 +0000578 first = get_a_page(rq, gfp);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000579 if (!first) {
Jason Wange9d74172012-12-07 07:04:55 +0000580 give_pages(rq, list);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000581 return -ENOMEM;
582 }
583 p = page_address(first);
Herbert Xu97402b92008-04-18 11:24:27 +0800584
Jason Wange9d74172012-12-07 07:04:55 +0000585 /* rq->sg[0], rq->sg[1] share the same page */
586 /* a separated rq->sg[0] for virtio_net_hdr only due to QEMU bug */
587 sg_set_buf(&rq->sg[0], p, sizeof(struct virtio_net_hdr));
Herbert Xu97402b92008-04-18 11:24:27 +0800588
Jason Wange9d74172012-12-07 07:04:55 +0000589 /* rq->sg[1] for data packet, from offset */
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000590 offset = sizeof(struct padded_vnet_hdr);
Jason Wange9d74172012-12-07 07:04:55 +0000591 sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset);
Herbert Xu97402b92008-04-18 11:24:27 +0800592
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000593 /* chain first in list head */
594 first->private = (unsigned long)list;
Rusty Russell9dc7b9e2013-03-20 15:44:28 +1030595 err = virtqueue_add_inbuf(rq->vq, rq->sg, MAX_SKB_FRAGS + 2,
596 first, gfp);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000597 if (err < 0)
Jason Wange9d74172012-12-07 07:04:55 +0000598 give_pages(rq, first);
Herbert Xu97402b92008-04-18 11:24:27 +0800599
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000600 return err;
601}
Herbert Xu97402b92008-04-18 11:24:27 +0800602
Michael Daltonfbf28d72014-01-16 22:23:30 -0800603static unsigned int get_mergeable_buf_len(struct ewma *avg_pkt_len)
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000604{
Michael Daltonab7db912014-01-16 22:23:27 -0800605 const size_t hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
Michael Daltonfbf28d72014-01-16 22:23:30 -0800606 unsigned int len;
607
608 len = hdr_len + clamp_t(unsigned int, ewma_read(avg_pkt_len),
609 GOOD_PACKET_LEN, PAGE_SIZE - hdr_len);
610 return ALIGN(len, MERGEABLE_BUFFER_ALIGN);
611}
612
613static int add_recvbuf_mergeable(struct receive_queue *rq, gfp_t gfp)
614{
Michael Daltonfb518792014-01-16 22:23:26 -0800615 struct page_frag *alloc_frag = &rq->alloc_frag;
616 char *buf;
Michael Daltonab7db912014-01-16 22:23:27 -0800617 unsigned long ctx;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000618 int err;
Michael Daltonfb518792014-01-16 22:23:26 -0800619 unsigned int len, hole;
Rusty Russell296f96f2007-10-22 11:03:37 +1000620
Michael Daltonfbf28d72014-01-16 22:23:30 -0800621 len = get_mergeable_buf_len(&rq->mrg_avg_pkt_len);
Michael Daltonab7db912014-01-16 22:23:27 -0800622 if (unlikely(!skb_page_frag_refill(len, alloc_frag, gfp)))
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000623 return -ENOMEM;
Michael Daltonab7db912014-01-16 22:23:27 -0800624
Michael Daltonfb518792014-01-16 22:23:26 -0800625 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
Michael Daltonab7db912014-01-16 22:23:27 -0800626 ctx = mergeable_buf_to_ctx(buf, len);
Michael Daltonfb518792014-01-16 22:23:26 -0800627 get_page(alloc_frag->page);
Michael Daltonfb518792014-01-16 22:23:26 -0800628 alloc_frag->offset += len;
629 hole = alloc_frag->size - alloc_frag->offset;
Michael Daltonab7db912014-01-16 22:23:27 -0800630 if (hole < len) {
631 /* To avoid internal fragmentation, if there is very likely not
632 * enough space for another buffer, add the remaining space to
633 * the current buffer. This extra space is not included in
634 * the truesize stored in ctx.
635 */
Michael Daltonfb518792014-01-16 22:23:26 -0800636 len += hole;
637 alloc_frag->offset += hole;
638 }
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000639
Michael Daltonfb518792014-01-16 22:23:26 -0800640 sg_init_one(rq->sg, buf, len);
Michael Daltonab7db912014-01-16 22:23:27 -0800641 err = virtqueue_add_inbuf(rq->vq, rq->sg, 1, (void *)ctx, gfp);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000642 if (err < 0)
Michael Dalton2613af02013-10-28 15:44:18 -0700643 put_page(virt_to_head_page(buf));
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000644
645 return err;
Rusty Russell296f96f2007-10-22 11:03:37 +1000646}
647
Rusty Russellb2baed62011-12-29 00:42:38 +0000648/*
649 * Returns false if we couldn't fill entirely (OOM).
650 *
651 * Normally run in the receive path, but can also be run from ndo_open
652 * before we're receiving packets, or from refill_work which is
653 * careful to disable receiving (using napi_disable).
654 */
Jason Wange9d74172012-12-07 07:04:55 +0000655static bool try_fill_recv(struct receive_queue *rq, gfp_t gfp)
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800656{
Jason Wange9d74172012-12-07 07:04:55 +0000657 struct virtnet_info *vi = rq->vq->vdev->priv;
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800658 int err;
Michael S. Tsirkin1788f4952010-07-02 16:32:55 +0000659 bool oom;
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800660
Michael Daltonfb518792014-01-16 22:23:26 -0800661 gfp |= __GFP_COLD;
Amit Shah0aea51c2009-08-26 14:58:28 +0530662 do {
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000663 if (vi->mergeable_rx_bufs)
Jason Wange9d74172012-12-07 07:04:55 +0000664 err = add_recvbuf_mergeable(rq, gfp);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000665 else if (vi->big_packets)
Jason Wange9d74172012-12-07 07:04:55 +0000666 err = add_recvbuf_big(rq, gfp);
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000667 else
Jason Wange9d74172012-12-07 07:04:55 +0000668 err = add_recvbuf_small(rq, gfp);
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800669
Michael S. Tsirkin1788f4952010-07-02 16:32:55 +0000670 oom = err == -ENOMEM;
Rusty Russell9ed4cb02012-10-16 23:56:14 +1030671 if (err)
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800672 break;
Linus Torvaldsb7dfde92012-12-20 08:37:04 -0800673 } while (rq->vq->num_free);
Jason Wang681daee22014-03-26 13:03:00 +0800674 virtqueue_kick(rq->vq);
Rusty Russell3161e452009-08-26 12:22:32 -0700675 return !oom;
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800676}
677
Rusty Russell18445c42008-02-04 23:49:57 -0500678static void skb_recv_done(struct virtqueue *rvq)
Rusty Russell296f96f2007-10-22 11:03:37 +1000679{
680 struct virtnet_info *vi = rvq->vdev->priv;
Jason Wang986a4f42012-12-07 07:04:56 +0000681 struct receive_queue *rq = &vi->rq[vq2rxq(rvq)];
Jason Wange9d74172012-12-07 07:04:55 +0000682
Rusty Russell18445c42008-02-04 23:49:57 -0500683 /* Schedule NAPI, Suppress further interrupts if successful. */
Jason Wange9d74172012-12-07 07:04:55 +0000684 if (napi_schedule_prep(&rq->napi)) {
Michael S. Tsirkin1915a7122010-04-12 16:19:04 +0300685 virtqueue_disable_cb(rvq);
Jason Wange9d74172012-12-07 07:04:55 +0000686 __napi_schedule(&rq->napi);
Rusty Russell18445c42008-02-04 23:49:57 -0500687 }
Rusty Russell296f96f2007-10-22 11:03:37 +1000688}
689
Jason Wange9d74172012-12-07 07:04:55 +0000690static void virtnet_napi_enable(struct receive_queue *rq)
Bruce Rogers3e9d08e2011-02-10 11:03:31 -0800691{
Jason Wange9d74172012-12-07 07:04:55 +0000692 napi_enable(&rq->napi);
Bruce Rogers3e9d08e2011-02-10 11:03:31 -0800693
694 /* If all buffers were filled by other side before we napi_enabled, we
695 * won't get another interrupt, so process any outstanding packets
696 * now. virtnet_poll wants re-enable the queue, so we disable here.
697 * We synchronize against interrupts via NAPI_STATE_SCHED */
Jason Wange9d74172012-12-07 07:04:55 +0000698 if (napi_schedule_prep(&rq->napi)) {
699 virtqueue_disable_cb(rq->vq);
Michael S. Tsirkinec13ee82012-05-16 10:57:12 +0300700 local_bh_disable();
Jason Wange9d74172012-12-07 07:04:55 +0000701 __napi_schedule(&rq->napi);
Michael S. Tsirkinec13ee82012-05-16 10:57:12 +0300702 local_bh_enable();
Bruce Rogers3e9d08e2011-02-10 11:03:31 -0800703 }
704}
705
Rusty Russell3161e452009-08-26 12:22:32 -0700706static void refill_work(struct work_struct *work)
707{
Jason Wange9d74172012-12-07 07:04:55 +0000708 struct virtnet_info *vi =
709 container_of(work, struct virtnet_info, refill.work);
Rusty Russell3161e452009-08-26 12:22:32 -0700710 bool still_empty;
Jason Wang986a4f42012-12-07 07:04:56 +0000711 int i;
Rusty Russell3161e452009-08-26 12:22:32 -0700712
Sasha Levin55257d72013-04-29 12:00:08 +0930713 for (i = 0; i < vi->curr_queue_pairs; i++) {
Jason Wang986a4f42012-12-07 07:04:56 +0000714 struct receive_queue *rq = &vi->rq[i];
Rusty Russell3161e452009-08-26 12:22:32 -0700715
Jason Wang986a4f42012-12-07 07:04:56 +0000716 napi_disable(&rq->napi);
717 still_empty = !try_fill_recv(rq, GFP_KERNEL);
718 virtnet_napi_enable(rq);
719
720 /* In theory, this can happen: if we don't get any buffers in
721 * we will *never* try to fill again.
722 */
723 if (still_empty)
724 schedule_delayed_work(&vi->refill, HZ/2);
725 }
Rusty Russell3161e452009-08-26 12:22:32 -0700726}
727
Rusty Russell296f96f2007-10-22 11:03:37 +1000728static int virtnet_poll(struct napi_struct *napi, int budget)
729{
Jason Wange9d74172012-12-07 07:04:55 +0000730 struct receive_queue *rq =
731 container_of(napi, struct receive_queue, napi);
732 struct virtnet_info *vi = rq->vq->vdev->priv;
Shirley Ma9ab86bb2010-01-29 03:20:04 +0000733 void *buf;
Michael S. Tsirkincbdadbb2013-07-09 08:13:04 +0300734 unsigned int r, len, received = 0;
Rusty Russell296f96f2007-10-22 11:03:37 +1000735
736again:
737 while (received < budget &&
Jason Wange9d74172012-12-07 07:04:55 +0000738 (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) {
739 receive_buf(rq, buf, len);
Rusty Russell296f96f2007-10-22 11:03:37 +1000740 received++;
741 }
742
Jason Wangbe121f42014-01-16 14:45:24 +0800743 if (rq->vq->num_free > virtqueue_get_vring_size(rq->vq) / 2) {
Jason Wange9d74172012-12-07 07:04:55 +0000744 if (!try_fill_recv(rq, GFP_ATOMIC))
Tejun Heo3b07e9c2012-08-20 14:51:24 -0700745 schedule_delayed_work(&vi->refill, 0);
Rusty Russell3161e452009-08-26 12:22:32 -0700746 }
Rusty Russell296f96f2007-10-22 11:03:37 +1000747
Rusty Russell8329d982007-11-19 11:20:43 -0500748 /* Out of packets? */
749 if (received < budget) {
Michael S. Tsirkincbdadbb2013-07-09 08:13:04 +0300750 r = virtqueue_enable_cb_prepare(rq->vq);
Ben Hutchings288379f2009-01-19 16:43:59 -0800751 napi_complete(napi);
Michael S. Tsirkincbdadbb2013-07-09 08:13:04 +0300752 if (unlikely(virtqueue_poll(rq->vq, r)) &&
Joe Perches8e95a202009-12-03 07:58:21 +0000753 napi_schedule_prep(napi)) {
Jason Wange9d74172012-12-07 07:04:55 +0000754 virtqueue_disable_cb(rq->vq);
Ben Hutchings288379f2009-01-19 16:43:59 -0800755 __napi_schedule(napi);
Rusty Russell296f96f2007-10-22 11:03:37 +1000756 goto again;
Christian Borntraeger4265f162008-03-14 14:17:05 +0100757 }
Rusty Russell296f96f2007-10-22 11:03:37 +1000758 }
759
760 return received;
761}
762
Jason Wang986a4f42012-12-07 07:04:56 +0000763static int virtnet_open(struct net_device *dev)
764{
765 struct virtnet_info *vi = netdev_priv(dev);
766 int i;
767
Jason Wange4166622013-05-21 20:03:58 +0000768 for (i = 0; i < vi->max_queue_pairs; i++) {
769 if (i < vi->curr_queue_pairs)
770 /* Make sure we have some buffers: if oom use wq. */
771 if (!try_fill_recv(&vi->rq[i], GFP_KERNEL))
772 schedule_delayed_work(&vi->refill, 0);
Jason Wang986a4f42012-12-07 07:04:56 +0000773 virtnet_napi_enable(&vi->rq[i]);
774 }
775
776 return 0;
777}
778
Linus Torvaldsb7dfde92012-12-20 08:37:04 -0800779static void free_old_xmit_skbs(struct send_queue *sq)
Rusty Russell296f96f2007-10-22 11:03:37 +1000780{
781 struct sk_buff *skb;
Michael S. Tsirkin6ee57bc2012-10-16 23:56:14 +1030782 unsigned int len;
Jason Wange9d74172012-12-07 07:04:55 +0000783 struct virtnet_info *vi = sq->vq->vdev->priv;
Eric Dumazet58472a72012-02-13 06:53:41 +0000784 struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
Rusty Russell296f96f2007-10-22 11:03:37 +1000785
Jason Wange9d74172012-12-07 07:04:55 +0000786 while ((skb = virtqueue_get_buf(sq->vq, &len)) != NULL) {
Rusty Russell296f96f2007-10-22 11:03:37 +1000787 pr_debug("Sent skb %p\n", skb);
stephen hemminger3fa2a1d2011-06-15 06:36:29 +0000788
Eric Dumazet83a27052012-06-05 22:35:24 +0000789 u64_stats_update_begin(&stats->tx_syncp);
stephen hemminger3fa2a1d2011-06-15 06:36:29 +0000790 stats->tx_bytes += skb->len;
791 stats->tx_packets++;
Eric Dumazet83a27052012-06-05 22:35:24 +0000792 u64_stats_update_end(&stats->tx_syncp);
stephen hemminger3fa2a1d2011-06-15 06:36:29 +0000793
Eric Dumazeted79bab2009-10-14 14:36:43 +0000794 dev_kfree_skb_any(skb);
Rusty Russell296f96f2007-10-22 11:03:37 +1000795 }
796}
797
Jason Wange9d74172012-12-07 07:04:55 +0000798static int xmit_skb(struct send_queue *sq, struct sk_buff *skb)
Rusty Russell296f96f2007-10-22 11:03:37 +1000799{
Michael S. Tsirkine7428e92013-07-25 10:20:23 +0930800 struct skb_vnet_hdr *hdr;
Rusty Russell296f96f2007-10-22 11:03:37 +1000801 const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
Jason Wange9d74172012-12-07 07:04:55 +0000802 struct virtnet_info *vi = sq->vq->vdev->priv;
Michael S. Tsirkin7bedc7d2012-10-16 23:56:14 +1030803 unsigned num_sg;
Michael S. Tsirkine7428e92013-07-25 10:20:23 +0930804 unsigned hdr_len;
805 bool can_push;
Rusty Russell296f96f2007-10-22 11:03:37 +1000806
Johannes Berge1749612008-10-27 15:59:26 -0700807 pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
Michael S. Tsirkine7428e92013-07-25 10:20:23 +0930808 if (vi->mergeable_rx_bufs)
809 hdr_len = sizeof hdr->mhdr;
810 else
811 hdr_len = sizeof hdr->hdr;
812
813 can_push = vi->any_header_sg &&
814 !((unsigned long)skb->data & (__alignof__(*hdr) - 1)) &&
815 !skb_header_cloned(skb) && skb_headroom(skb) >= hdr_len;
816 /* Even if we can, don't push here yet as this would skew
817 * csum_start offset below. */
818 if (can_push)
819 hdr = (struct skb_vnet_hdr *)(skb->data - hdr_len);
820 else
821 hdr = skb_vnet_hdr(skb);
Rusty Russell296f96f2007-10-22 11:03:37 +1000822
Rusty Russell296f96f2007-10-22 11:03:37 +1000823 if (skb->ip_summed == CHECKSUM_PARTIAL) {
Rusty Russellb3f24692009-09-24 09:59:19 -0600824 hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
Michał Mirosław55508d62010-12-14 15:24:08 +0000825 hdr->hdr.csum_start = skb_checksum_start_offset(skb);
Rusty Russellb3f24692009-09-24 09:59:19 -0600826 hdr->hdr.csum_offset = skb->csum_offset;
Rusty Russell296f96f2007-10-22 11:03:37 +1000827 } else {
Rusty Russellb3f24692009-09-24 09:59:19 -0600828 hdr->hdr.flags = 0;
829 hdr->hdr.csum_offset = hdr->hdr.csum_start = 0;
Rusty Russell296f96f2007-10-22 11:03:37 +1000830 }
831
832 if (skb_is_gso(skb)) {
Rusty Russellb3f24692009-09-24 09:59:19 -0600833 hdr->hdr.hdr_len = skb_headlen(skb);
834 hdr->hdr.gso_size = skb_shinfo(skb)->gso_size;
Rusty Russell34a48572008-02-04 23:50:02 -0500835 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
Rusty Russellb3f24692009-09-24 09:59:19 -0600836 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
Rusty Russell296f96f2007-10-22 11:03:37 +1000837 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
Rusty Russellb3f24692009-09-24 09:59:19 -0600838 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
Rusty Russell296f96f2007-10-22 11:03:37 +1000839 else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
Rusty Russellb3f24692009-09-24 09:59:19 -0600840 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
Rusty Russell296f96f2007-10-22 11:03:37 +1000841 else
842 BUG();
Rusty Russell34a48572008-02-04 23:50:02 -0500843 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
Rusty Russellb3f24692009-09-24 09:59:19 -0600844 hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
Rusty Russell296f96f2007-10-22 11:03:37 +1000845 } else {
Rusty Russellb3f24692009-09-24 09:59:19 -0600846 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
847 hdr->hdr.gso_size = hdr->hdr.hdr_len = 0;
Rusty Russell296f96f2007-10-22 11:03:37 +1000848 }
849
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800850 if (vi->mergeable_rx_bufs)
Michael S. Tsirkine7428e92013-07-25 10:20:23 +0930851 hdr->mhdr.num_buffers = 0;
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -0800852
Michael S. Tsirkine7428e92013-07-25 10:20:23 +0930853 if (can_push) {
854 __skb_push(skb, hdr_len);
855 num_sg = skb_to_sgvec(skb, sq->sg, 0, skb->len);
856 /* Pull header back to avoid skew in tx bytes calculations. */
857 __skb_pull(skb, hdr_len);
858 } else {
859 sg_set_buf(sq->sg, hdr, hdr_len);
860 num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len) + 1;
861 }
Rusty Russell9dc7b9e2013-03-20 15:44:28 +1030862 return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC);
Rusty Russell11a3a152008-05-26 17:48:13 +1000863}
864
Stephen Hemminger424efe92009-08-31 19:50:51 +0000865static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
Rusty Russell99ffc692008-05-02 21:50:46 -0500866{
867 struct virtnet_info *vi = netdev_priv(dev);
Jason Wang986a4f42012-12-07 07:04:56 +0000868 int qnum = skb_get_queue_mapping(skb);
869 struct send_queue *sq = &vi->sq[qnum];
Rusty Russell9ed4cb02012-10-16 23:56:14 +1030870 int err;
Rusty Russell2cb9c6b2008-02-04 23:50:07 -0500871
Rusty Russell2cb9c6b2008-02-04 23:50:07 -0500872 /* Free up any pending old buffers before queueing new ones. */
Jason Wange9d74172012-12-07 07:04:55 +0000873 free_old_xmit_skbs(sq);
Rusty Russell2cb9c6b2008-02-04 23:50:07 -0500874
Michael S. Tsirkin03f191b2009-10-28 04:03:38 -0700875 /* Try to transmit */
Linus Torvaldsb7dfde92012-12-20 08:37:04 -0800876 err = xmit_skb(sq, skb);
Rusty Russell48925e32009-09-24 09:59:20 -0600877
Rusty Russell9ed4cb02012-10-16 23:56:14 +1030878 /* This should not happen! */
Jason Wang681daee22014-03-26 13:03:00 +0800879 if (unlikely(err)) {
Rusty Russell9ed4cb02012-10-16 23:56:14 +1030880 dev->stats.tx_fifo_errors++;
881 if (net_ratelimit())
882 dev_warn(&dev->dev,
Linus Torvaldsb7dfde92012-12-20 08:37:04 -0800883 "Unexpected TXQ (%d) queue failure: %d\n", qnum, err);
Rusty Russell58eba972010-07-02 16:34:01 +0000884 dev->stats.tx_dropped++;
Eric W. Biederman85e94522014-03-15 18:43:33 -0700885 dev_kfree_skb_any(skb);
Rusty Russell58eba972010-07-02 16:34:01 +0000886 return NETDEV_TX_OK;
Rusty Russell99ffc692008-05-02 21:50:46 -0500887 }
Jason Wang681daee22014-03-26 13:03:00 +0800888 virtqueue_kick(sq->vq);
Michael S. Tsirkin03f191b2009-10-28 04:03:38 -0700889
Rusty Russell48925e32009-09-24 09:59:20 -0600890 /* Don't wait up for transmitted skbs to be freed. */
891 skb_orphan(skb);
892 nf_reset(skb);
Rusty Russell99ffc692008-05-02 21:50:46 -0500893
Rusty Russell48925e32009-09-24 09:59:20 -0600894 /* Apparently nice girls don't return TX_BUSY; stop the queue
895 * before it gets out of hand. Naturally, this wastes entries. */
Linus Torvaldsb7dfde92012-12-20 08:37:04 -0800896 if (sq->vq->num_free < 2+MAX_SKB_FRAGS) {
Jason Wang986a4f42012-12-07 07:04:56 +0000897 netif_stop_subqueue(dev, qnum);
Jason Wange9d74172012-12-07 07:04:55 +0000898 if (unlikely(!virtqueue_enable_cb_delayed(sq->vq))) {
Rusty Russell48925e32009-09-24 09:59:20 -0600899 /* More just got used, free them then recheck. */
Linus Torvaldsb7dfde92012-12-20 08:37:04 -0800900 free_old_xmit_skbs(sq);
901 if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) {
Jason Wang986a4f42012-12-07 07:04:56 +0000902 netif_start_subqueue(dev, qnum);
Jason Wange9d74172012-12-07 07:04:55 +0000903 virtqueue_disable_cb(sq->vq);
Rusty Russell48925e32009-09-24 09:59:20 -0600904 }
905 }
Rusty Russell99ffc692008-05-02 21:50:46 -0500906 }
Rusty Russell48925e32009-09-24 09:59:20 -0600907
908 return NETDEV_TX_OK;
Rusty Russell296f96f2007-10-22 11:03:37 +1000909}
910
Amos Kong40cbfc32013-01-21 01:17:21 +0000911/*
912 * Send command via the control virtqueue and check status. Commands
913 * supported by the hypervisor, as indicated by feature bits, should
stephen hemminger788a8b62013-12-09 16:18:45 -0800914 * never fail unless improperly formatted.
Amos Kong40cbfc32013-01-21 01:17:21 +0000915 */
916static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
stephen hemmingerd24bae32013-12-09 16:17:40 -0800917 struct scatterlist *out)
Amos Kong40cbfc32013-01-21 01:17:21 +0000918{
Rusty Russellf7bc9592013-03-20 15:44:28 +1030919 struct scatterlist *sgs[4], hdr, stat;
Amos Kong40cbfc32013-01-21 01:17:21 +0000920 struct virtio_net_ctrl_hdr ctrl;
921 virtio_net_ctrl_ack status = ~0;
stephen hemmingerd24bae32013-12-09 16:17:40 -0800922 unsigned out_num = 0, tmp;
Amos Kong40cbfc32013-01-21 01:17:21 +0000923
924 /* Caller should know better */
Rusty Russellf7bc9592013-03-20 15:44:28 +1030925 BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ));
Amos Kong40cbfc32013-01-21 01:17:21 +0000926
927 ctrl.class = class;
928 ctrl.cmd = cmd;
Rusty Russellf7bc9592013-03-20 15:44:28 +1030929 /* Add header */
930 sg_init_one(&hdr, &ctrl, sizeof(ctrl));
931 sgs[out_num++] = &hdr;
Amos Kong40cbfc32013-01-21 01:17:21 +0000932
Rusty Russellf7bc9592013-03-20 15:44:28 +1030933 if (out)
934 sgs[out_num++] = out;
Amos Kong40cbfc32013-01-21 01:17:21 +0000935
Rusty Russellf7bc9592013-03-20 15:44:28 +1030936 /* Add return status. */
937 sg_init_one(&stat, &status, sizeof(status));
stephen hemmingerd24bae32013-12-09 16:17:40 -0800938 sgs[out_num] = &stat;
Amos Kong40cbfc32013-01-21 01:17:21 +0000939
stephen hemmingerd24bae32013-12-09 16:17:40 -0800940 BUG_ON(out_num + 1 > ARRAY_SIZE(sgs));
Rusty Russella7c58142014-03-13 11:23:39 +1030941 virtqueue_add_sgs(vi->cvq, sgs, out_num, 1, vi, GFP_ATOMIC);
Amos Kong40cbfc32013-01-21 01:17:21 +0000942
Heinz Graalfs67975902013-10-29 09:40:02 +1030943 if (unlikely(!virtqueue_kick(vi->cvq)))
944 return status == VIRTIO_NET_OK;
Amos Kong40cbfc32013-01-21 01:17:21 +0000945
946 /* Spin for a response, the kick causes an ioport write, trapping
947 * into the hypervisor, so the request should be handled immediately.
948 */
Heinz Graalfs047b9b92013-10-29 09:40:47 +1030949 while (!virtqueue_get_buf(vi->cvq, &tmp) &&
950 !virtqueue_is_broken(vi->cvq))
Amos Kong40cbfc32013-01-21 01:17:21 +0000951 cpu_relax();
952
953 return status == VIRTIO_NET_OK;
954}
955
Alex Williamson9c46f6d2009-02-04 16:36:34 -0800956static int virtnet_set_mac_address(struct net_device *dev, void *p)
957{
958 struct virtnet_info *vi = netdev_priv(dev);
959 struct virtio_device *vdev = vi->vdev;
Jiri Pirkof2f2c8b2012-06-29 05:10:06 +0000960 int ret;
Amos Kong7e58d5a2013-01-21 01:17:23 +0000961 struct sockaddr *addr = p;
962 struct scatterlist sg;
Alex Williamson9c46f6d2009-02-04 16:36:34 -0800963
Amos Kong7e58d5a2013-01-21 01:17:23 +0000964 ret = eth_prepare_mac_addr_change(dev, p);
Jiri Pirkof2f2c8b2012-06-29 05:10:06 +0000965 if (ret)
966 return ret;
Alex Williamson9c46f6d2009-02-04 16:36:34 -0800967
Amos Kong7e58d5a2013-01-21 01:17:23 +0000968 if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
969 sg_init_one(&sg, addr->sa_data, dev->addr_len);
970 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
stephen hemmingerd24bae32013-12-09 16:17:40 -0800971 VIRTIO_NET_CTRL_MAC_ADDR_SET, &sg)) {
Amos Kong7e58d5a2013-01-21 01:17:23 +0000972 dev_warn(&vdev->dev,
973 "Failed to set mac address by vq command.\n");
974 return -EINVAL;
975 }
976 } else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
Rusty Russell855e0c52013-10-14 18:11:51 +1030977 unsigned int i;
978
979 /* Naturally, this has an atomicity problem. */
980 for (i = 0; i < dev->addr_len; i++)
981 virtio_cwrite8(vdev,
982 offsetof(struct virtio_net_config, mac) +
983 i, addr->sa_data[i]);
Amos Kong7e58d5a2013-01-21 01:17:23 +0000984 }
985
986 eth_commit_mac_addr_change(dev, p);
Alex Williamson9c46f6d2009-02-04 16:36:34 -0800987
988 return 0;
989}
990
stephen hemminger3fa2a1d2011-06-15 06:36:29 +0000991static struct rtnl_link_stats64 *virtnet_stats(struct net_device *dev,
992 struct rtnl_link_stats64 *tot)
993{
994 struct virtnet_info *vi = netdev_priv(dev);
995 int cpu;
996 unsigned int start;
997
998 for_each_possible_cpu(cpu) {
Eric Dumazet58472a72012-02-13 06:53:41 +0000999 struct virtnet_stats *stats = per_cpu_ptr(vi->stats, cpu);
stephen hemminger3fa2a1d2011-06-15 06:36:29 +00001000 u64 tpackets, tbytes, rpackets, rbytes;
1001
1002 do {
Eric W. Biederman57a77442014-03-13 21:26:42 -07001003 start = u64_stats_fetch_begin_irq(&stats->tx_syncp);
stephen hemminger3fa2a1d2011-06-15 06:36:29 +00001004 tpackets = stats->tx_packets;
1005 tbytes = stats->tx_bytes;
Eric W. Biederman57a77442014-03-13 21:26:42 -07001006 } while (u64_stats_fetch_retry_irq(&stats->tx_syncp, start));
Eric Dumazet83a27052012-06-05 22:35:24 +00001007
1008 do {
Eric W. Biederman57a77442014-03-13 21:26:42 -07001009 start = u64_stats_fetch_begin_irq(&stats->rx_syncp);
stephen hemminger3fa2a1d2011-06-15 06:36:29 +00001010 rpackets = stats->rx_packets;
1011 rbytes = stats->rx_bytes;
Eric W. Biederman57a77442014-03-13 21:26:42 -07001012 } while (u64_stats_fetch_retry_irq(&stats->rx_syncp, start));
stephen hemminger3fa2a1d2011-06-15 06:36:29 +00001013
1014 tot->rx_packets += rpackets;
1015 tot->tx_packets += tpackets;
1016 tot->rx_bytes += rbytes;
1017 tot->tx_bytes += tbytes;
1018 }
1019
1020 tot->tx_dropped = dev->stats.tx_dropped;
Rick Jones021ac8d2011-11-21 09:28:17 +00001021 tot->tx_fifo_errors = dev->stats.tx_fifo_errors;
stephen hemminger3fa2a1d2011-06-15 06:36:29 +00001022 tot->rx_dropped = dev->stats.rx_dropped;
1023 tot->rx_length_errors = dev->stats.rx_length_errors;
1024 tot->rx_frame_errors = dev->stats.rx_frame_errors;
1025
1026 return tot;
1027}
1028
Amit Shahda74e892008-02-29 16:24:50 +05301029#ifdef CONFIG_NET_POLL_CONTROLLER
1030static void virtnet_netpoll(struct net_device *dev)
1031{
1032 struct virtnet_info *vi = netdev_priv(dev);
Jason Wang986a4f42012-12-07 07:04:56 +00001033 int i;
Amit Shahda74e892008-02-29 16:24:50 +05301034
Jason Wang986a4f42012-12-07 07:04:56 +00001035 for (i = 0; i < vi->curr_queue_pairs; i++)
1036 napi_schedule(&vi->rq[i].napi);
Amit Shahda74e892008-02-29 16:24:50 +05301037}
1038#endif
1039
Jason Wang586d17c2012-04-11 20:43:52 +00001040static void virtnet_ack_link_announce(struct virtnet_info *vi)
1041{
1042 rtnl_lock();
1043 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE,
stephen hemmingerd24bae32013-12-09 16:17:40 -08001044 VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL))
Jason Wang586d17c2012-04-11 20:43:52 +00001045 dev_warn(&vi->dev->dev, "Failed to ack link announce.\n");
1046 rtnl_unlock();
1047}
1048
Jason Wang986a4f42012-12-07 07:04:56 +00001049static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs)
1050{
1051 struct scatterlist sg;
1052 struct virtio_net_ctrl_mq s;
1053 struct net_device *dev = vi->dev;
1054
1055 if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ))
1056 return 0;
1057
1058 s.virtqueue_pairs = queue_pairs;
1059 sg_init_one(&sg, &s, sizeof(s));
1060
1061 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ,
stephen hemmingerd24bae32013-12-09 16:17:40 -08001062 VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg)) {
Jason Wang986a4f42012-12-07 07:04:56 +00001063 dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n",
1064 queue_pairs);
1065 return -EINVAL;
Sasha Levin55257d72013-04-29 12:00:08 +09301066 } else {
Jason Wang986a4f42012-12-07 07:04:56 +00001067 vi->curr_queue_pairs = queue_pairs;
Jason Wang35ed1592013-10-15 11:18:59 +08001068 /* virtnet_open() will refill when device is going to up. */
1069 if (dev->flags & IFF_UP)
1070 schedule_delayed_work(&vi->refill, 0);
Sasha Levin55257d72013-04-29 12:00:08 +09301071 }
Jason Wang986a4f42012-12-07 07:04:56 +00001072
1073 return 0;
1074}
1075
Rusty Russell296f96f2007-10-22 11:03:37 +10001076static int virtnet_close(struct net_device *dev)
1077{
1078 struct virtnet_info *vi = netdev_priv(dev);
Jason Wang986a4f42012-12-07 07:04:56 +00001079 int i;
Rusty Russell296f96f2007-10-22 11:03:37 +10001080
Rusty Russellb2baed62011-12-29 00:42:38 +00001081 /* Make sure refill_work doesn't re-enable napi! */
1082 cancel_delayed_work_sync(&vi->refill);
Jason Wang986a4f42012-12-07 07:04:56 +00001083
1084 for (i = 0; i < vi->max_queue_pairs; i++)
1085 napi_disable(&vi->rq[i].napi);
Rusty Russell296f96f2007-10-22 11:03:37 +10001086
Rusty Russell296f96f2007-10-22 11:03:37 +10001087 return 0;
1088}
1089
Alex Williamson2af76982009-02-04 09:02:40 +00001090static void virtnet_set_rx_mode(struct net_device *dev)
1091{
1092 struct virtnet_info *vi = netdev_priv(dev);
Alex Williamsonf565a7c2009-02-04 09:02:45 +00001093 struct scatterlist sg[2];
Alex Williamson2af76982009-02-04 09:02:40 +00001094 u8 promisc, allmulti;
Alex Williamsonf565a7c2009-02-04 09:02:45 +00001095 struct virtio_net_ctrl_mac *mac_data;
Jiri Pirkoccffad252009-05-22 23:22:17 +00001096 struct netdev_hw_addr *ha;
Jiri Pirko32e7bfc2010-01-25 13:36:10 -08001097 int uc_count;
Jiri Pirko4cd24ea2010-02-08 04:30:35 +00001098 int mc_count;
Alex Williamsonf565a7c2009-02-04 09:02:45 +00001099 void *buf;
1100 int i;
Alex Williamson2af76982009-02-04 09:02:40 +00001101
stephen hemminger788a8b62013-12-09 16:18:45 -08001102 /* We can't dynamically set ndo_set_rx_mode, so return gracefully */
Alex Williamson2af76982009-02-04 09:02:40 +00001103 if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
1104 return;
1105
Alex Williamsonf565a7c2009-02-04 09:02:45 +00001106 promisc = ((dev->flags & IFF_PROMISC) != 0);
1107 allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
Alex Williamson2af76982009-02-04 09:02:40 +00001108
Alex Williamson23e258e2009-05-01 17:27:56 +00001109 sg_init_one(sg, &promisc, sizeof(promisc));
Alex Williamson2af76982009-02-04 09:02:40 +00001110
1111 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
stephen hemmingerd24bae32013-12-09 16:17:40 -08001112 VIRTIO_NET_CTRL_RX_PROMISC, sg))
Alex Williamson2af76982009-02-04 09:02:40 +00001113 dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
1114 promisc ? "en" : "dis");
1115
Alex Williamson23e258e2009-05-01 17:27:56 +00001116 sg_init_one(sg, &allmulti, sizeof(allmulti));
Alex Williamson2af76982009-02-04 09:02:40 +00001117
1118 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
stephen hemmingerd24bae32013-12-09 16:17:40 -08001119 VIRTIO_NET_CTRL_RX_ALLMULTI, sg))
Alex Williamson2af76982009-02-04 09:02:40 +00001120 dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
1121 allmulti ? "en" : "dis");
Alex Williamsonf565a7c2009-02-04 09:02:45 +00001122
Jiri Pirko32e7bfc2010-01-25 13:36:10 -08001123 uc_count = netdev_uc_count(dev);
Jiri Pirko4cd24ea2010-02-08 04:30:35 +00001124 mc_count = netdev_mc_count(dev);
Alex Williamsonf565a7c2009-02-04 09:02:45 +00001125 /* MAC filter - use one buffer for both lists */
Jiri Pirko4cd24ea2010-02-08 04:30:35 +00001126 buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
1127 (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
1128 mac_data = buf;
Joe Perchese68ed8f2013-02-03 17:28:15 +00001129 if (!buf)
Alex Williamsonf565a7c2009-02-04 09:02:45 +00001130 return;
Alex Williamsonf565a7c2009-02-04 09:02:45 +00001131
Alex Williamson23e258e2009-05-01 17:27:56 +00001132 sg_init_table(sg, 2);
1133
Alex Williamsonf565a7c2009-02-04 09:02:45 +00001134 /* Store the unicast list and count in the front of the buffer */
Jiri Pirko32e7bfc2010-01-25 13:36:10 -08001135 mac_data->entries = uc_count;
Jiri Pirkoccffad252009-05-22 23:22:17 +00001136 i = 0;
Jiri Pirko32e7bfc2010-01-25 13:36:10 -08001137 netdev_for_each_uc_addr(ha, dev)
Jiri Pirkoccffad252009-05-22 23:22:17 +00001138 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
Alex Williamsonf565a7c2009-02-04 09:02:45 +00001139
1140 sg_set_buf(&sg[0], mac_data,
Jiri Pirko32e7bfc2010-01-25 13:36:10 -08001141 sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
Alex Williamsonf565a7c2009-02-04 09:02:45 +00001142
1143 /* multicast list and count fill the end */
Jiri Pirko32e7bfc2010-01-25 13:36:10 -08001144 mac_data = (void *)&mac_data->macs[uc_count][0];
Alex Williamsonf565a7c2009-02-04 09:02:45 +00001145
Jiri Pirko4cd24ea2010-02-08 04:30:35 +00001146 mac_data->entries = mc_count;
Jiri Pirko567ec872010-02-23 23:17:07 +00001147 i = 0;
Jiri Pirko22bedad32010-04-01 21:22:57 +00001148 netdev_for_each_mc_addr(ha, dev)
1149 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
Alex Williamsonf565a7c2009-02-04 09:02:45 +00001150
1151 sg_set_buf(&sg[1], mac_data,
Jiri Pirko4cd24ea2010-02-08 04:30:35 +00001152 sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
Alex Williamsonf565a7c2009-02-04 09:02:45 +00001153
1154 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
stephen hemmingerd24bae32013-12-09 16:17:40 -08001155 VIRTIO_NET_CTRL_MAC_TABLE_SET, sg))
Thomas Huth99e872a2013-11-29 10:02:19 +01001156 dev_warn(&dev->dev, "Failed to set MAC filter table.\n");
Alex Williamsonf565a7c2009-02-04 09:02:45 +00001157
1158 kfree(buf);
Alex Williamson2af76982009-02-04 09:02:40 +00001159}
1160
Patrick McHardy80d5c362013-04-19 02:04:28 +00001161static int virtnet_vlan_rx_add_vid(struct net_device *dev,
1162 __be16 proto, u16 vid)
Alex Williamson0bde95692009-02-04 09:02:50 +00001163{
1164 struct virtnet_info *vi = netdev_priv(dev);
1165 struct scatterlist sg;
1166
Alex Williamson23e258e2009-05-01 17:27:56 +00001167 sg_init_one(&sg, &vid, sizeof(vid));
Alex Williamson0bde95692009-02-04 09:02:50 +00001168
1169 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
stephen hemmingerd24bae32013-12-09 16:17:40 -08001170 VIRTIO_NET_CTRL_VLAN_ADD, &sg))
Alex Williamson0bde95692009-02-04 09:02:50 +00001171 dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
Jiri Pirko8e586132011-12-08 19:52:37 -05001172 return 0;
Alex Williamson0bde95692009-02-04 09:02:50 +00001173}
1174
Patrick McHardy80d5c362013-04-19 02:04:28 +00001175static int virtnet_vlan_rx_kill_vid(struct net_device *dev,
1176 __be16 proto, u16 vid)
Alex Williamson0bde95692009-02-04 09:02:50 +00001177{
1178 struct virtnet_info *vi = netdev_priv(dev);
1179 struct scatterlist sg;
1180
Alex Williamson23e258e2009-05-01 17:27:56 +00001181 sg_init_one(&sg, &vid, sizeof(vid));
Alex Williamson0bde95692009-02-04 09:02:50 +00001182
1183 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
stephen hemmingerd24bae32013-12-09 16:17:40 -08001184 VIRTIO_NET_CTRL_VLAN_DEL, &sg))
Alex Williamson0bde95692009-02-04 09:02:50 +00001185 dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
Jiri Pirko8e586132011-12-08 19:52:37 -05001186 return 0;
Alex Williamson0bde95692009-02-04 09:02:50 +00001187}
1188
Wanlong Gao8898c212013-01-24 23:51:30 +00001189static void virtnet_clean_affinity(struct virtnet_info *vi, long hcpu)
Jason Wang986a4f42012-12-07 07:04:56 +00001190{
1191 int i;
Wanlong Gao8898c212013-01-24 23:51:30 +00001192
1193 if (vi->affinity_hint_set) {
1194 for (i = 0; i < vi->max_queue_pairs; i++) {
1195 virtqueue_set_affinity(vi->rq[i].vq, -1);
1196 virtqueue_set_affinity(vi->sq[i].vq, -1);
1197 }
1198
1199 vi->affinity_hint_set = false;
1200 }
Wanlong Gao8898c212013-01-24 23:51:30 +00001201}
1202
1203static void virtnet_set_affinity(struct virtnet_info *vi)
Jason Wang986a4f42012-12-07 07:04:56 +00001204{
1205 int i;
Wanlong Gao47be2472013-01-24 23:51:29 +00001206 int cpu;
Jason Wang986a4f42012-12-07 07:04:56 +00001207
1208 /* In multiqueue mode, when the number of cpu is equal to the number of
1209 * queue pairs, we let the queue pairs to be private to one cpu by
1210 * setting the affinity hint to eliminate the contention.
1211 */
Wanlong Gao8898c212013-01-24 23:51:30 +00001212 if (vi->curr_queue_pairs == 1 ||
1213 vi->max_queue_pairs != num_online_cpus()) {
1214 virtnet_clean_affinity(vi, -1);
1215 return;
Jason Wang986a4f42012-12-07 07:04:56 +00001216 }
1217
Wanlong Gao8898c212013-01-24 23:51:30 +00001218 i = 0;
1219 for_each_online_cpu(cpu) {
Jason Wang986a4f42012-12-07 07:04:56 +00001220 virtqueue_set_affinity(vi->rq[i].vq, cpu);
1221 virtqueue_set_affinity(vi->sq[i].vq, cpu);
Jason Wang9bb8ca82013-11-05 18:19:45 +08001222 netif_set_xps_queue(vi->dev, cpumask_of(cpu), i);
Wanlong Gao8898c212013-01-24 23:51:30 +00001223 i++;
Jason Wang986a4f42012-12-07 07:04:56 +00001224 }
1225
Wanlong Gao8898c212013-01-24 23:51:30 +00001226 vi->affinity_hint_set = true;
Jason Wang986a4f42012-12-07 07:04:56 +00001227}
1228
Wanlong Gao8de4b2f2013-01-24 23:51:31 +00001229static int virtnet_cpu_callback(struct notifier_block *nfb,
1230 unsigned long action, void *hcpu)
1231{
1232 struct virtnet_info *vi = container_of(nfb, struct virtnet_info, nb);
1233
1234 switch(action & ~CPU_TASKS_FROZEN) {
1235 case CPU_ONLINE:
1236 case CPU_DOWN_FAILED:
1237 case CPU_DEAD:
1238 virtnet_set_affinity(vi);
1239 break;
1240 case CPU_DOWN_PREPARE:
1241 virtnet_clean_affinity(vi, (long)hcpu);
1242 break;
1243 default:
1244 break;
1245 }
Jason Wang3ab098d2013-10-15 11:18:58 +08001246
Wanlong Gao8de4b2f2013-01-24 23:51:31 +00001247 return NOTIFY_OK;
Herbert Xua9ea3fc2008-04-18 11:21:42 +08001248}
1249
Rick Jones8f9f4662011-10-19 08:10:59 +00001250static void virtnet_get_ringparam(struct net_device *dev,
1251 struct ethtool_ringparam *ring)
1252{
1253 struct virtnet_info *vi = netdev_priv(dev);
1254
Jason Wang986a4f42012-12-07 07:04:56 +00001255 ring->rx_max_pending = virtqueue_get_vring_size(vi->rq[0].vq);
1256 ring->tx_max_pending = virtqueue_get_vring_size(vi->sq[0].vq);
Rick Jones8f9f4662011-10-19 08:10:59 +00001257 ring->rx_pending = ring->rx_max_pending;
1258 ring->tx_pending = ring->tx_max_pending;
Rick Jones8f9f4662011-10-19 08:10:59 +00001259}
1260
Rick Jones66846042011-11-14 14:17:08 +00001261
1262static void virtnet_get_drvinfo(struct net_device *dev,
1263 struct ethtool_drvinfo *info)
1264{
1265 struct virtnet_info *vi = netdev_priv(dev);
1266 struct virtio_device *vdev = vi->vdev;
1267
1268 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
1269 strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version));
1270 strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info));
1271
1272}
1273
Jason Wangd73bcd22012-12-07 07:04:57 +00001274/* TODO: Eliminate OOO packets during switching */
1275static int virtnet_set_channels(struct net_device *dev,
1276 struct ethtool_channels *channels)
1277{
1278 struct virtnet_info *vi = netdev_priv(dev);
1279 u16 queue_pairs = channels->combined_count;
1280 int err;
1281
1282 /* We don't support separate rx/tx channels.
1283 * We don't allow setting 'other' channels.
1284 */
1285 if (channels->rx_count || channels->tx_count || channels->other_count)
1286 return -EINVAL;
1287
1288 if (queue_pairs > vi->max_queue_pairs)
1289 return -EINVAL;
1290
Wanlong Gao47be2472013-01-24 23:51:29 +00001291 get_online_cpus();
Jason Wangd73bcd22012-12-07 07:04:57 +00001292 err = virtnet_set_queues(vi, queue_pairs);
1293 if (!err) {
1294 netif_set_real_num_tx_queues(dev, queue_pairs);
1295 netif_set_real_num_rx_queues(dev, queue_pairs);
1296
Wanlong Gao8898c212013-01-24 23:51:30 +00001297 virtnet_set_affinity(vi);
Jason Wangd73bcd22012-12-07 07:04:57 +00001298 }
Wanlong Gao47be2472013-01-24 23:51:29 +00001299 put_online_cpus();
Jason Wangd73bcd22012-12-07 07:04:57 +00001300
1301 return err;
1302}
1303
1304static void virtnet_get_channels(struct net_device *dev,
1305 struct ethtool_channels *channels)
1306{
1307 struct virtnet_info *vi = netdev_priv(dev);
1308
1309 channels->combined_count = vi->curr_queue_pairs;
1310 channels->max_combined = vi->max_queue_pairs;
1311 channels->max_other = 0;
1312 channels->rx_count = 0;
1313 channels->tx_count = 0;
1314 channels->other_count = 0;
1315}
1316
Stephen Hemminger0fc0b732009-09-02 01:03:33 -07001317static const struct ethtool_ops virtnet_ethtool_ops = {
Rick Jones66846042011-11-14 14:17:08 +00001318 .get_drvinfo = virtnet_get_drvinfo,
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -08001319 .get_link = ethtool_op_get_link,
Rick Jones8f9f4662011-10-19 08:10:59 +00001320 .get_ringparam = virtnet_get_ringparam,
Jason Wangd73bcd22012-12-07 07:04:57 +00001321 .set_channels = virtnet_set_channels,
1322 .get_channels = virtnet_get_channels,
Herbert Xua9ea3fc2008-04-18 11:21:42 +08001323};
1324
Mark McLoughlin39da5812008-11-26 13:58:11 +00001325#define MIN_MTU 68
1326#define MAX_MTU 65535
1327
1328static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
1329{
1330 if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
1331 return -EINVAL;
1332 dev->mtu = new_mtu;
1333 return 0;
1334}
1335
Stephen Hemminger76288b42009-01-06 10:44:22 -08001336static const struct net_device_ops virtnet_netdev = {
1337 .ndo_open = virtnet_open,
1338 .ndo_stop = virtnet_close,
1339 .ndo_start_xmit = start_xmit,
1340 .ndo_validate_addr = eth_validate_addr,
Alex Williamson9c46f6d2009-02-04 16:36:34 -08001341 .ndo_set_mac_address = virtnet_set_mac_address,
Alex Williamson2af76982009-02-04 09:02:40 +00001342 .ndo_set_rx_mode = virtnet_set_rx_mode,
Stephen Hemminger76288b42009-01-06 10:44:22 -08001343 .ndo_change_mtu = virtnet_change_mtu,
stephen hemminger3fa2a1d2011-06-15 06:36:29 +00001344 .ndo_get_stats64 = virtnet_stats,
Alex Williamson1824a982009-05-01 17:31:10 +00001345 .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
1346 .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
Stephen Hemminger76288b42009-01-06 10:44:22 -08001347#ifdef CONFIG_NET_POLL_CONTROLLER
1348 .ndo_poll_controller = virtnet_netpoll,
1349#endif
1350};
1351
Jason Wang586d17c2012-04-11 20:43:52 +00001352static void virtnet_config_changed_work(struct work_struct *work)
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -08001353{
Jason Wang586d17c2012-04-11 20:43:52 +00001354 struct virtnet_info *vi =
1355 container_of(work, struct virtnet_info, config_work);
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -08001356 u16 v;
1357
Jason Wang586d17c2012-04-11 20:43:52 +00001358 mutex_lock(&vi->config_lock);
1359 if (!vi->config_enable)
1360 goto done;
1361
Rusty Russell855e0c52013-10-14 18:11:51 +10301362 if (virtio_cread_feature(vi->vdev, VIRTIO_NET_F_STATUS,
1363 struct virtio_net_config, status, &v) < 0)
Jason Wang586d17c2012-04-11 20:43:52 +00001364 goto done;
1365
1366 if (v & VIRTIO_NET_S_ANNOUNCE) {
Amerigo Wangee89bab2012-08-09 22:14:56 +00001367 netdev_notify_peers(vi->dev);
Jason Wang586d17c2012-04-11 20:43:52 +00001368 virtnet_ack_link_announce(vi);
1369 }
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -08001370
1371 /* Ignore unknown (future) status bits */
1372 v &= VIRTIO_NET_S_LINK_UP;
1373
1374 if (vi->status == v)
Jason Wang586d17c2012-04-11 20:43:52 +00001375 goto done;
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -08001376
1377 vi->status = v;
1378
1379 if (vi->status & VIRTIO_NET_S_LINK_UP) {
1380 netif_carrier_on(vi->dev);
Jason Wang986a4f42012-12-07 07:04:56 +00001381 netif_tx_wake_all_queues(vi->dev);
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -08001382 } else {
1383 netif_carrier_off(vi->dev);
Jason Wang986a4f42012-12-07 07:04:56 +00001384 netif_tx_stop_all_queues(vi->dev);
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -08001385 }
Jason Wang586d17c2012-04-11 20:43:52 +00001386done:
1387 mutex_unlock(&vi->config_lock);
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -08001388}
1389
1390static void virtnet_config_changed(struct virtio_device *vdev)
1391{
1392 struct virtnet_info *vi = vdev->priv;
1393
Tejun Heo3b07e9c2012-08-20 14:51:24 -07001394 schedule_work(&vi->config_work);
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -08001395}
1396
Jason Wang986a4f42012-12-07 07:04:56 +00001397static void virtnet_free_queues(struct virtnet_info *vi)
1398{
Andrey Vagind4fb84e2013-12-05 18:36:21 +04001399 int i;
1400
1401 for (i = 0; i < vi->max_queue_pairs; i++)
1402 netif_napi_del(&vi->rq[i].napi);
1403
Jason Wang986a4f42012-12-07 07:04:56 +00001404 kfree(vi->rq);
1405 kfree(vi->sq);
1406}
1407
1408static void free_receive_bufs(struct virtnet_info *vi)
1409{
1410 int i;
1411
1412 for (i = 0; i < vi->max_queue_pairs; i++) {
1413 while (vi->rq[i].pages)
1414 __free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0);
1415 }
1416}
1417
Michael Daltonfb518792014-01-16 22:23:26 -08001418static void free_receive_page_frags(struct virtnet_info *vi)
1419{
1420 int i;
1421 for (i = 0; i < vi->max_queue_pairs; i++)
1422 if (vi->rq[i].alloc_frag.page)
1423 put_page(vi->rq[i].alloc_frag.page);
1424}
1425
Jason Wang986a4f42012-12-07 07:04:56 +00001426static void free_unused_bufs(struct virtnet_info *vi)
1427{
1428 void *buf;
1429 int i;
1430
1431 for (i = 0; i < vi->max_queue_pairs; i++) {
1432 struct virtqueue *vq = vi->sq[i].vq;
1433 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL)
1434 dev_kfree_skb(buf);
1435 }
1436
1437 for (i = 0; i < vi->max_queue_pairs; i++) {
1438 struct virtqueue *vq = vi->rq[i].vq;
1439
1440 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) {
Michael Daltonab7db912014-01-16 22:23:27 -08001441 if (vi->mergeable_rx_bufs) {
1442 unsigned long ctx = (unsigned long)buf;
1443 void *base = mergeable_ctx_to_buf_address(ctx);
1444 put_page(virt_to_head_page(base));
1445 } else if (vi->big_packets) {
Andrey Vaginfa9fac12013-12-05 18:36:20 +04001446 give_pages(&vi->rq[i], buf);
Michael Daltonab7db912014-01-16 22:23:27 -08001447 } else {
Jason Wang986a4f42012-12-07 07:04:56 +00001448 dev_kfree_skb(buf);
Michael Daltonab7db912014-01-16 22:23:27 -08001449 }
Jason Wang986a4f42012-12-07 07:04:56 +00001450 }
Jason Wang986a4f42012-12-07 07:04:56 +00001451 }
1452}
1453
Jason Wange9d74172012-12-07 07:04:55 +00001454static void virtnet_del_vqs(struct virtnet_info *vi)
1455{
1456 struct virtio_device *vdev = vi->vdev;
1457
Wanlong Gao8898c212013-01-24 23:51:30 +00001458 virtnet_clean_affinity(vi, -1);
Jason Wang986a4f42012-12-07 07:04:56 +00001459
Jason Wange9d74172012-12-07 07:04:55 +00001460 vdev->config->del_vqs(vdev);
Jason Wang986a4f42012-12-07 07:04:56 +00001461
1462 virtnet_free_queues(vi);
1463}
1464
1465static int virtnet_find_vqs(struct virtnet_info *vi)
1466{
1467 vq_callback_t **callbacks;
1468 struct virtqueue **vqs;
1469 int ret = -ENOMEM;
1470 int i, total_vqs;
1471 const char **names;
1472
1473 /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
1474 * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
1475 * possible control vq.
1476 */
1477 total_vqs = vi->max_queue_pairs * 2 +
1478 virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ);
1479
1480 /* Allocate space for find_vqs parameters */
1481 vqs = kzalloc(total_vqs * sizeof(*vqs), GFP_KERNEL);
1482 if (!vqs)
1483 goto err_vq;
1484 callbacks = kmalloc(total_vqs * sizeof(*callbacks), GFP_KERNEL);
1485 if (!callbacks)
1486 goto err_callback;
1487 names = kmalloc(total_vqs * sizeof(*names), GFP_KERNEL);
1488 if (!names)
1489 goto err_names;
1490
1491 /* Parameters for control virtqueue, if any */
1492 if (vi->has_cvq) {
1493 callbacks[total_vqs - 1] = NULL;
1494 names[total_vqs - 1] = "control";
1495 }
1496
1497 /* Allocate/initialize parameters for send/receive virtqueues */
1498 for (i = 0; i < vi->max_queue_pairs; i++) {
1499 callbacks[rxq2vq(i)] = skb_recv_done;
1500 callbacks[txq2vq(i)] = skb_xmit_done;
1501 sprintf(vi->rq[i].name, "input.%d", i);
1502 sprintf(vi->sq[i].name, "output.%d", i);
1503 names[rxq2vq(i)] = vi->rq[i].name;
1504 names[txq2vq(i)] = vi->sq[i].name;
1505 }
1506
1507 ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs, callbacks,
1508 names);
1509 if (ret)
1510 goto err_find;
1511
1512 if (vi->has_cvq) {
1513 vi->cvq = vqs[total_vqs - 1];
1514 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
Patrick McHardyf6469682013-04-19 02:04:27 +00001515 vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
Jason Wang986a4f42012-12-07 07:04:56 +00001516 }
1517
1518 for (i = 0; i < vi->max_queue_pairs; i++) {
1519 vi->rq[i].vq = vqs[rxq2vq(i)];
1520 vi->sq[i].vq = vqs[txq2vq(i)];
1521 }
1522
1523 kfree(names);
1524 kfree(callbacks);
1525 kfree(vqs);
1526
1527 return 0;
1528
1529err_find:
1530 kfree(names);
1531err_names:
1532 kfree(callbacks);
1533err_callback:
1534 kfree(vqs);
1535err_vq:
1536 return ret;
1537}
1538
1539static int virtnet_alloc_queues(struct virtnet_info *vi)
1540{
1541 int i;
1542
1543 vi->sq = kzalloc(sizeof(*vi->sq) * vi->max_queue_pairs, GFP_KERNEL);
1544 if (!vi->sq)
1545 goto err_sq;
1546 vi->rq = kzalloc(sizeof(*vi->rq) * vi->max_queue_pairs, GFP_KERNEL);
Amerigo Wang008d4272012-12-10 02:24:08 +00001547 if (!vi->rq)
Jason Wang986a4f42012-12-07 07:04:56 +00001548 goto err_rq;
1549
1550 INIT_DELAYED_WORK(&vi->refill, refill_work);
1551 for (i = 0; i < vi->max_queue_pairs; i++) {
1552 vi->rq[i].pages = NULL;
1553 netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll,
1554 napi_weight);
1555
1556 sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg));
Michael Daltonab7db912014-01-16 22:23:27 -08001557 ewma_init(&vi->rq[i].mrg_avg_pkt_len, 1, RECEIVE_AVG_WEIGHT);
Jason Wang986a4f42012-12-07 07:04:56 +00001558 sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg));
1559 }
1560
1561 return 0;
1562
1563err_rq:
1564 kfree(vi->sq);
1565err_sq:
1566 return -ENOMEM;
Jason Wange9d74172012-12-07 07:04:55 +00001567}
1568
Amit Shah3f9c10b2011-12-22 16:58:31 +05301569static int init_vqs(struct virtnet_info *vi)
1570{
Jason Wang986a4f42012-12-07 07:04:56 +00001571 int ret;
Amit Shah3f9c10b2011-12-22 16:58:31 +05301572
Jason Wang986a4f42012-12-07 07:04:56 +00001573 /* Allocate send & receive queues */
1574 ret = virtnet_alloc_queues(vi);
1575 if (ret)
1576 goto err;
Amit Shah3f9c10b2011-12-22 16:58:31 +05301577
Jason Wang986a4f42012-12-07 07:04:56 +00001578 ret = virtnet_find_vqs(vi);
1579 if (ret)
1580 goto err_free;
Amit Shah3f9c10b2011-12-22 16:58:31 +05301581
Wanlong Gao47be2472013-01-24 23:51:29 +00001582 get_online_cpus();
Wanlong Gao8898c212013-01-24 23:51:30 +00001583 virtnet_set_affinity(vi);
Wanlong Gao47be2472013-01-24 23:51:29 +00001584 put_online_cpus();
1585
Amit Shah3f9c10b2011-12-22 16:58:31 +05301586 return 0;
Jason Wang986a4f42012-12-07 07:04:56 +00001587
1588err_free:
1589 virtnet_free_queues(vi);
1590err:
1591 return ret;
Amit Shah3f9c10b2011-12-22 16:58:31 +05301592}
1593
Michael Daltonfbf28d72014-01-16 22:23:30 -08001594#ifdef CONFIG_SYSFS
1595static ssize_t mergeable_rx_buffer_size_show(struct netdev_rx_queue *queue,
1596 struct rx_queue_attribute *attribute, char *buf)
1597{
1598 struct virtnet_info *vi = netdev_priv(queue->dev);
1599 unsigned int queue_index = get_netdev_rx_queue_index(queue);
1600 struct ewma *avg;
1601
1602 BUG_ON(queue_index >= vi->max_queue_pairs);
1603 avg = &vi->rq[queue_index].mrg_avg_pkt_len;
1604 return sprintf(buf, "%u\n", get_mergeable_buf_len(avg));
1605}
1606
1607static struct rx_queue_attribute mergeable_rx_buffer_size_attribute =
1608 __ATTR_RO(mergeable_rx_buffer_size);
1609
1610static struct attribute *virtio_net_mrg_rx_attrs[] = {
1611 &mergeable_rx_buffer_size_attribute.attr,
1612 NULL
1613};
1614
1615static const struct attribute_group virtio_net_mrg_rx_group = {
1616 .name = "virtio_net",
1617 .attrs = virtio_net_mrg_rx_attrs
1618};
1619#endif
1620
Rusty Russell296f96f2007-10-22 11:03:37 +10001621static int virtnet_probe(struct virtio_device *vdev)
1622{
Jason Wang986a4f42012-12-07 07:04:56 +00001623 int i, err;
Rusty Russell296f96f2007-10-22 11:03:37 +10001624 struct net_device *dev;
1625 struct virtnet_info *vi;
Jason Wang986a4f42012-12-07 07:04:56 +00001626 u16 max_queue_pairs;
1627
1628 /* Find if host supports multiqueue virtio_net device */
Rusty Russell855e0c52013-10-14 18:11:51 +10301629 err = virtio_cread_feature(vdev, VIRTIO_NET_F_MQ,
1630 struct virtio_net_config,
1631 max_virtqueue_pairs, &max_queue_pairs);
Jason Wang986a4f42012-12-07 07:04:56 +00001632
1633 /* We need at least 2 queue's */
1634 if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
1635 max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
1636 !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
1637 max_queue_pairs = 1;
Rusty Russell296f96f2007-10-22 11:03:37 +10001638
1639 /* Allocate ourselves a network device with room for our info */
Jason Wang986a4f42012-12-07 07:04:56 +00001640 dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs);
Rusty Russell296f96f2007-10-22 11:03:37 +10001641 if (!dev)
1642 return -ENOMEM;
1643
1644 /* Set up network device as normal. */
Jiri Pirkof2f2c8b2012-06-29 05:10:06 +00001645 dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE;
Stephen Hemminger76288b42009-01-06 10:44:22 -08001646 dev->netdev_ops = &virtnet_netdev;
Rusty Russell296f96f2007-10-22 11:03:37 +10001647 dev->features = NETIF_F_HIGHDMA;
stephen hemminger3fa2a1d2011-06-15 06:36:29 +00001648
Herbert Xua9ea3fc2008-04-18 11:21:42 +08001649 SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
Rusty Russell296f96f2007-10-22 11:03:37 +10001650 SET_NETDEV_DEV(dev, &vdev->dev);
1651
1652 /* Do we support "hardware" checksums? */
Michał Mirosław98e778c2011-03-31 01:01:35 +00001653 if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
Rusty Russell296f96f2007-10-22 11:03:37 +10001654 /* This opens up the world of extra features. */
Michał Mirosław98e778c2011-03-31 01:01:35 +00001655 dev->hw_features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1656 if (csum)
1657 dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1658
1659 if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
1660 dev->hw_features |= NETIF_F_TSO | NETIF_F_UFO
Rusty Russell34a48572008-02-04 23:50:02 -05001661 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
1662 }
Rusty Russell5539ae92008-05-02 21:50:46 -05001663 /* Individual feature bits: what can host handle? */
Michał Mirosław98e778c2011-03-31 01:01:35 +00001664 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
1665 dev->hw_features |= NETIF_F_TSO;
1666 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
1667 dev->hw_features |= NETIF_F_TSO6;
1668 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
1669 dev->hw_features |= NETIF_F_TSO_ECN;
1670 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
1671 dev->hw_features |= NETIF_F_UFO;
1672
1673 if (gso)
1674 dev->features |= dev->hw_features & (NETIF_F_ALL_TSO|NETIF_F_UFO);
1675 /* (!csum && gso) case will be fixed by register_netdev() */
Rusty Russell296f96f2007-10-22 11:03:37 +10001676 }
Thomas Huth4f491292013-08-27 17:09:02 +02001677 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM))
1678 dev->features |= NETIF_F_RXCSUM;
Rusty Russell296f96f2007-10-22 11:03:37 +10001679
Jason Wang4fda8302013-04-10 23:32:21 +00001680 dev->vlan_features = dev->features;
1681
Rusty Russell296f96f2007-10-22 11:03:37 +10001682 /* Configuration may specify what MAC to use. Otherwise random. */
Rusty Russell855e0c52013-10-14 18:11:51 +10301683 if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
1684 virtio_cread_bytes(vdev,
1685 offsetof(struct virtio_net_config, mac),
1686 dev->dev_addr, dev->addr_len);
1687 else
Danny Kukawkaf2cedb62012-02-15 06:45:39 +00001688 eth_hw_addr_random(dev);
Rusty Russell296f96f2007-10-22 11:03:37 +10001689
1690 /* Set up our device-specific information */
1691 vi = netdev_priv(dev);
Rusty Russell296f96f2007-10-22 11:03:37 +10001692 vi->dev = dev;
1693 vi->vdev = vdev;
Christian Borntraegerd9d5dcc2008-02-18 10:02:51 +01001694 vdev->priv = vi;
stephen hemminger3fa2a1d2011-06-15 06:36:29 +00001695 vi->stats = alloc_percpu(struct virtnet_stats);
1696 err = -ENOMEM;
1697 if (vi->stats == NULL)
1698 goto free;
1699
John Stultz827da442013-10-07 15:51:58 -07001700 for_each_possible_cpu(i) {
1701 struct virtnet_stats *virtnet_stats;
1702 virtnet_stats = per_cpu_ptr(vi->stats, i);
1703 u64_stats_init(&virtnet_stats->tx_syncp);
1704 u64_stats_init(&virtnet_stats->rx_syncp);
1705 }
1706
Jason Wang586d17c2012-04-11 20:43:52 +00001707 mutex_init(&vi->config_lock);
1708 vi->config_enable = true;
1709 INIT_WORK(&vi->config_work, virtnet_config_changed_work);
Rusty Russell296f96f2007-10-22 11:03:37 +10001710
Herbert Xu97402b92008-04-18 11:24:27 +08001711 /* If we can receive ANY GSO packets, we must allocate large ones. */
Joe Perches8e95a202009-12-03 07:58:21 +00001712 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
1713 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
Jason Wang0e7ede82014-02-21 13:08:04 +08001714 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN) ||
1715 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_UFO))
Herbert Xu97402b92008-04-18 11:24:27 +08001716 vi->big_packets = true;
1717
Mark McLoughlin3f2c31d2008-11-16 22:41:34 -08001718 if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
1719 vi->mergeable_rx_bufs = true;
1720
Michael S. Tsirkine7428e92013-07-25 10:20:23 +09301721 if (virtio_has_feature(vdev, VIRTIO_F_ANY_LAYOUT))
1722 vi->any_header_sg = true;
1723
Jason Wang986a4f42012-12-07 07:04:56 +00001724 if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
1725 vi->has_cvq = true;
1726
1727 /* Use single tx/rx queue pair as default */
1728 vi->curr_queue_pairs = 1;
1729 vi->max_queue_pairs = max_queue_pairs;
1730
1731 /* Allocate/initialize the rx/tx queues, and invoke find_vqs */
Amit Shah3f9c10b2011-12-22 16:58:31 +05301732 err = init_vqs(vi);
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -06001733 if (err)
Jason Wang9bb8ca82013-11-05 18:19:45 +08001734 goto free_stats;
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -06001735
Michael Daltonfbf28d72014-01-16 22:23:30 -08001736#ifdef CONFIG_SYSFS
1737 if (vi->mergeable_rx_bufs)
1738 dev->sysfs_rx_queue_group = &virtio_net_mrg_rx_group;
1739#endif
Zhi Yong Wu0f13b662013-11-18 21:19:27 +08001740 netif_set_real_num_tx_queues(dev, vi->curr_queue_pairs);
1741 netif_set_real_num_rx_queues(dev, vi->curr_queue_pairs);
Jason Wang986a4f42012-12-07 07:04:56 +00001742
Rusty Russell296f96f2007-10-22 11:03:37 +10001743 err = register_netdev(dev);
1744 if (err) {
1745 pr_debug("virtio_net: registering device failed\n");
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -06001746 goto free_vqs;
Rusty Russell296f96f2007-10-22 11:03:37 +10001747 }
Rusty Russellb3369c12008-02-04 23:50:02 -05001748
1749 /* Last of all, set up some receive buffers. */
Sasha Levin55257d72013-04-29 12:00:08 +09301750 for (i = 0; i < vi->curr_queue_pairs; i++) {
Jason Wang986a4f42012-12-07 07:04:56 +00001751 try_fill_recv(&vi->rq[i], GFP_KERNEL);
Rusty Russellb3369c12008-02-04 23:50:02 -05001752
Jason Wang986a4f42012-12-07 07:04:56 +00001753 /* If we didn't even get one input buffer, we're useless. */
Jason Wangbe121f42014-01-16 14:45:24 +08001754 if (vi->rq[i].vq->num_free ==
1755 virtqueue_get_vring_size(vi->rq[i].vq)) {
Jason Wang986a4f42012-12-07 07:04:56 +00001756 free_unused_bufs(vi);
1757 err = -ENOMEM;
1758 goto free_recv_bufs;
1759 }
Rusty Russellb3369c12008-02-04 23:50:02 -05001760 }
1761
Wanlong Gao8de4b2f2013-01-24 23:51:31 +00001762 vi->nb.notifier_call = &virtnet_cpu_callback;
1763 err = register_hotcpu_notifier(&vi->nb);
1764 if (err) {
1765 pr_debug("virtio_net: registering cpu notifier failed\n");
1766 goto free_recv_bufs;
1767 }
1768
Jason Wang167c25e2010-11-10 14:45:41 +00001769 /* Assume link up if device can't report link status,
1770 otherwise get link status from config. */
1771 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
1772 netif_carrier_off(dev);
Tejun Heo3b07e9c2012-08-20 14:51:24 -07001773 schedule_work(&vi->config_work);
Jason Wang167c25e2010-11-10 14:45:41 +00001774 } else {
1775 vi->status = VIRTIO_NET_S_LINK_UP;
1776 netif_carrier_on(dev);
1777 }
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -08001778
Jason Wang986a4f42012-12-07 07:04:56 +00001779 pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
1780 dev->name, max_queue_pairs);
1781
Rusty Russell296f96f2007-10-22 11:03:37 +10001782 return 0;
1783
Jason Wang986a4f42012-12-07 07:04:56 +00001784free_recv_bufs:
1785 free_receive_bufs(vi);
Rusty Russellb3369c12008-02-04 23:50:02 -05001786 unregister_netdev(dev);
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -06001787free_vqs:
Jason Wang986a4f42012-12-07 07:04:56 +00001788 cancel_delayed_work_sync(&vi->refill);
Michael Daltonfb518792014-01-16 22:23:26 -08001789 free_receive_page_frags(vi);
Jason Wange9d74172012-12-07 07:04:55 +00001790 virtnet_del_vqs(vi);
stephen hemminger3fa2a1d2011-06-15 06:36:29 +00001791free_stats:
1792 free_percpu(vi->stats);
Rusty Russell296f96f2007-10-22 11:03:37 +10001793free:
1794 free_netdev(dev);
1795 return err;
1796}
1797
Amit Shah04486ed2011-12-22 16:58:32 +05301798static void remove_vq_common(struct virtnet_info *vi)
Rusty Russell296f96f2007-10-22 11:03:37 +10001799{
Amit Shah04486ed2011-12-22 16:58:32 +05301800 vi->vdev->config->reset(vi->vdev);
Shirley Ma830a8a92010-02-08 14:14:42 +00001801
1802 /* Free unused buffers in both send and recv, if any. */
Shirley Ma9ab86bb2010-01-29 03:20:04 +00001803 free_unused_bufs(vi);
Rusty Russellfb6813f2008-07-25 12:06:01 -05001804
Jason Wang986a4f42012-12-07 07:04:56 +00001805 free_receive_bufs(vi);
Michael S. Tsirkind2a7ddd2009-06-12 22:16:36 -06001806
Michael Daltonfb518792014-01-16 22:23:26 -08001807 free_receive_page_frags(vi);
1808
Jason Wang986a4f42012-12-07 07:04:56 +00001809 virtnet_del_vqs(vi);
Amit Shah04486ed2011-12-22 16:58:32 +05301810}
1811
Bill Pemberton8cc085d2012-12-03 09:24:15 -05001812static void virtnet_remove(struct virtio_device *vdev)
Amit Shah04486ed2011-12-22 16:58:32 +05301813{
1814 struct virtnet_info *vi = vdev->priv;
1815
Wanlong Gao8de4b2f2013-01-24 23:51:31 +00001816 unregister_hotcpu_notifier(&vi->nb);
1817
Jason Wang586d17c2012-04-11 20:43:52 +00001818 /* Prevent config work handler from accessing the device. */
1819 mutex_lock(&vi->config_lock);
1820 vi->config_enable = false;
1821 mutex_unlock(&vi->config_lock);
1822
Amit Shah04486ed2011-12-22 16:58:32 +05301823 unregister_netdev(vi->dev);
1824
1825 remove_vq_common(vi);
Rusty Russellfb6813f2008-07-25 12:06:01 -05001826
Jason Wang586d17c2012-04-11 20:43:52 +00001827 flush_work(&vi->config_work);
1828
Krishna Kumar2e66f552011-07-20 03:56:02 +00001829 free_percpu(vi->stats);
Rusty Russell74b25532007-11-19 11:20:42 -05001830 free_netdev(vi->dev);
Rusty Russell296f96f2007-10-22 11:03:37 +10001831}
1832
Aaron Lu89107002013-09-17 09:25:23 +09301833#ifdef CONFIG_PM_SLEEP
Amit Shah0741bcb2011-12-22 16:58:33 +05301834static int virtnet_freeze(struct virtio_device *vdev)
1835{
1836 struct virtnet_info *vi = vdev->priv;
Jason Wang986a4f42012-12-07 07:04:56 +00001837 int i;
Amit Shah0741bcb2011-12-22 16:58:33 +05301838
Jason Wangec9debb2013-10-29 15:11:07 +08001839 unregister_hotcpu_notifier(&vi->nb);
1840
Jason Wang586d17c2012-04-11 20:43:52 +00001841 /* Prevent config work handler from accessing the device */
1842 mutex_lock(&vi->config_lock);
1843 vi->config_enable = false;
1844 mutex_unlock(&vi->config_lock);
1845
Amit Shah0741bcb2011-12-22 16:58:33 +05301846 netif_device_detach(vi->dev);
1847 cancel_delayed_work_sync(&vi->refill);
1848
1849 if (netif_running(vi->dev))
Jason Wang986a4f42012-12-07 07:04:56 +00001850 for (i = 0; i < vi->max_queue_pairs; i++) {
1851 napi_disable(&vi->rq[i].napi);
1852 netif_napi_del(&vi->rq[i].napi);
1853 }
Amit Shah0741bcb2011-12-22 16:58:33 +05301854
1855 remove_vq_common(vi);
1856
Jason Wang586d17c2012-04-11 20:43:52 +00001857 flush_work(&vi->config_work);
1858
Amit Shah0741bcb2011-12-22 16:58:33 +05301859 return 0;
1860}
1861
1862static int virtnet_restore(struct virtio_device *vdev)
1863{
1864 struct virtnet_info *vi = vdev->priv;
Jason Wang986a4f42012-12-07 07:04:56 +00001865 int err, i;
Amit Shah0741bcb2011-12-22 16:58:33 +05301866
1867 err = init_vqs(vi);
1868 if (err)
1869 return err;
1870
Jason Wang6cd4ce02013-12-30 11:34:40 +08001871 if (netif_running(vi->dev)) {
1872 for (i = 0; i < vi->curr_queue_pairs; i++)
1873 if (!try_fill_recv(&vi->rq[i], GFP_KERNEL))
1874 schedule_delayed_work(&vi->refill, 0);
1875
Jason Wang986a4f42012-12-07 07:04:56 +00001876 for (i = 0; i < vi->max_queue_pairs; i++)
1877 virtnet_napi_enable(&vi->rq[i]);
Jason Wang6cd4ce02013-12-30 11:34:40 +08001878 }
Amit Shah0741bcb2011-12-22 16:58:33 +05301879
1880 netif_device_attach(vi->dev);
1881
Jason Wang586d17c2012-04-11 20:43:52 +00001882 mutex_lock(&vi->config_lock);
1883 vi->config_enable = true;
1884 mutex_unlock(&vi->config_lock);
1885
Jason Wang35ed1592013-10-15 11:18:59 +08001886 rtnl_lock();
Jason Wang986a4f42012-12-07 07:04:56 +00001887 virtnet_set_queues(vi, vi->curr_queue_pairs);
Jason Wang35ed1592013-10-15 11:18:59 +08001888 rtnl_unlock();
Jason Wang986a4f42012-12-07 07:04:56 +00001889
Jason Wangec9debb2013-10-29 15:11:07 +08001890 err = register_hotcpu_notifier(&vi->nb);
1891 if (err)
1892 return err;
1893
Amit Shah0741bcb2011-12-22 16:58:33 +05301894 return 0;
1895}
1896#endif
1897
Rusty Russell296f96f2007-10-22 11:03:37 +10001898static struct virtio_device_id id_table[] = {
1899 { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
1900 { 0 },
1901};
1902
Rusty Russellc45a6812008-05-02 21:50:50 -05001903static unsigned int features[] = {
Mark McLoughlin5e4fe5c2008-07-08 17:10:42 +10001904 VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
1905 VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
Rusty Russellc45a6812008-05-02 21:50:50 -05001906 VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
Herbert Xu97402b92008-04-18 11:24:27 +08001907 VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
Sridhar Samudrala5c516752009-07-14 14:21:02 +00001908 VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO,
Alex Williamson2a41f712009-02-04 09:02:34 +00001909 VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ,
Alex Williamson0bde95692009-02-04 09:02:50 +00001910 VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN,
Jason Wang986a4f42012-12-07 07:04:56 +00001911 VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ,
Amos Kong7e58d5a2013-01-21 01:17:23 +00001912 VIRTIO_NET_F_CTRL_MAC_ADDR,
Michael S. Tsirkine7428e92013-07-25 10:20:23 +09301913 VIRTIO_F_ANY_LAYOUT,
Rusty Russellc45a6812008-05-02 21:50:50 -05001914};
1915
Uwe Kleine-König22402522009-11-05 01:32:44 -08001916static struct virtio_driver virtio_net_driver = {
Rusty Russellc45a6812008-05-02 21:50:50 -05001917 .feature_table = features,
1918 .feature_table_size = ARRAY_SIZE(features),
Rusty Russell296f96f2007-10-22 11:03:37 +10001919 .driver.name = KBUILD_MODNAME,
1920 .driver.owner = THIS_MODULE,
1921 .id_table = id_table,
1922 .probe = virtnet_probe,
Bill Pemberton8cc085d2012-12-03 09:24:15 -05001923 .remove = virtnet_remove,
Mark McLoughlin9f4d26d2009-01-19 17:09:49 -08001924 .config_changed = virtnet_config_changed,
Aaron Lu89107002013-09-17 09:25:23 +09301925#ifdef CONFIG_PM_SLEEP
Amit Shah0741bcb2011-12-22 16:58:33 +05301926 .freeze = virtnet_freeze,
1927 .restore = virtnet_restore,
1928#endif
Rusty Russell296f96f2007-10-22 11:03:37 +10001929};
1930
Rusty Russellb2a17022013-02-13 16:59:28 +10301931module_virtio_driver(virtio_net_driver);
Rusty Russell296f96f2007-10-22 11:03:37 +10001932
1933MODULE_DEVICE_TABLE(virtio, id_table);
1934MODULE_DESCRIPTION("Virtio network driver");
1935MODULE_LICENSE("GPL");