Loading...
1/* A network driver using virtio.
2 *
3 * Copyright 2007 Rusty Russell <rusty@rustcorp.com.au> IBM Corporation
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18 */
19//#define DEBUG
20#include <linux/netdevice.h>
21#include <linux/etherdevice.h>
22#include <linux/ethtool.h>
23#include <linux/module.h>
24#include <linux/virtio.h>
25#include <linux/virtio_net.h>
26#include <linux/scatterlist.h>
27#include <linux/if_vlan.h>
28#include <linux/slab.h>
29
30static int napi_weight = 128;
31module_param(napi_weight, int, 0444);
32
33static bool csum = true, gso = true;
34module_param(csum, bool, 0444);
35module_param(gso, bool, 0444);
36
37/* FIXME: MTU in config. */
38#define MAX_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
39#define GOOD_COPY_LEN 128
40
41#define VIRTNET_SEND_COMMAND_SG_MAX 2
42#define VIRTNET_DRIVER_VERSION "1.0.0"
43
44struct virtnet_stats {
45 struct u64_stats_sync tx_syncp;
46 struct u64_stats_sync rx_syncp;
47 u64 tx_bytes;
48 u64 tx_packets;
49
50 u64 rx_bytes;
51 u64 rx_packets;
52};
53
54struct virtnet_info {
55 struct virtio_device *vdev;
56 struct virtqueue *rvq, *svq, *cvq;
57 struct net_device *dev;
58 struct napi_struct napi;
59 unsigned int status;
60
61 /* Number of input buffers, and max we've ever had. */
62 unsigned int num, max;
63
64 /* I like... big packets and I cannot lie! */
65 bool big_packets;
66
67 /* Host will merge rx buffers for big packets (shake it! shake it!) */
68 bool mergeable_rx_bufs;
69
70 /* enable config space updates */
71 bool config_enable;
72
73 /* Active statistics */
74 struct virtnet_stats __percpu *stats;
75
76 /* Work struct for refilling if we run low on memory. */
77 struct delayed_work refill;
78
79 /* Work struct for config space updates */
80 struct work_struct config_work;
81
82 /* Lock for config space updates */
83 struct mutex config_lock;
84
85 /* Chain pages by the private ptr. */
86 struct page *pages;
87
88 /* fragments + linear part + virtio header */
89 struct scatterlist rx_sg[MAX_SKB_FRAGS + 2];
90 struct scatterlist tx_sg[MAX_SKB_FRAGS + 2];
91};
92
93struct skb_vnet_hdr {
94 union {
95 struct virtio_net_hdr hdr;
96 struct virtio_net_hdr_mrg_rxbuf mhdr;
97 };
98 unsigned int num_sg;
99};
100
101struct padded_vnet_hdr {
102 struct virtio_net_hdr hdr;
103 /*
104 * virtio_net_hdr should be in a separated sg buffer because of a
105 * QEMU bug, and data sg buffer shares same page with this header sg.
106 * This padding makes next sg 16 byte aligned after virtio_net_hdr.
107 */
108 char padding[6];
109};
110
111static inline struct skb_vnet_hdr *skb_vnet_hdr(struct sk_buff *skb)
112{
113 return (struct skb_vnet_hdr *)skb->cb;
114}
115
116/*
117 * private is used to chain pages for big packets, put the whole
118 * most recent used list in the beginning for reuse
119 */
120static void give_pages(struct virtnet_info *vi, struct page *page)
121{
122 struct page *end;
123
124 /* Find end of list, sew whole thing into vi->pages. */
125 for (end = page; end->private; end = (struct page *)end->private);
126 end->private = (unsigned long)vi->pages;
127 vi->pages = page;
128}
129
130static struct page *get_a_page(struct virtnet_info *vi, gfp_t gfp_mask)
131{
132 struct page *p = vi->pages;
133
134 if (p) {
135 vi->pages = (struct page *)p->private;
136 /* clear private here, it is used to chain pages */
137 p->private = 0;
138 } else
139 p = alloc_page(gfp_mask);
140 return p;
141}
142
143static void skb_xmit_done(struct virtqueue *svq)
144{
145 struct virtnet_info *vi = svq->vdev->priv;
146
147 /* Suppress further interrupts. */
148 virtqueue_disable_cb(svq);
149
150 /* We were probably waiting for more output buffers. */
151 netif_wake_queue(vi->dev);
152}
153
154static void set_skb_frag(struct sk_buff *skb, struct page *page,
155 unsigned int offset, unsigned int *len)
156{
157 int size = min((unsigned)PAGE_SIZE - offset, *len);
158 int i = skb_shinfo(skb)->nr_frags;
159
160 __skb_fill_page_desc(skb, i, page, offset, size);
161
162 skb->data_len += size;
163 skb->len += size;
164 skb->truesize += PAGE_SIZE;
165 skb_shinfo(skb)->nr_frags++;
166 *len -= size;
167}
168
169/* Called from bottom half context */
170static struct sk_buff *page_to_skb(struct virtnet_info *vi,
171 struct page *page, unsigned int len)
172{
173 struct sk_buff *skb;
174 struct skb_vnet_hdr *hdr;
175 unsigned int copy, hdr_len, offset;
176 char *p;
177
178 p = page_address(page);
179
180 /* copy small packet so we can reuse these pages for small data */
181 skb = netdev_alloc_skb_ip_align(vi->dev, GOOD_COPY_LEN);
182 if (unlikely(!skb))
183 return NULL;
184
185 hdr = skb_vnet_hdr(skb);
186
187 if (vi->mergeable_rx_bufs) {
188 hdr_len = sizeof hdr->mhdr;
189 offset = hdr_len;
190 } else {
191 hdr_len = sizeof hdr->hdr;
192 offset = sizeof(struct padded_vnet_hdr);
193 }
194
195 memcpy(hdr, p, hdr_len);
196
197 len -= hdr_len;
198 p += offset;
199
200 copy = len;
201 if (copy > skb_tailroom(skb))
202 copy = skb_tailroom(skb);
203 memcpy(skb_put(skb, copy), p, copy);
204
205 len -= copy;
206 offset += copy;
207
208 /*
209 * Verify that we can indeed put this data into a skb.
210 * This is here to handle cases when the device erroneously
211 * tries to receive more than is possible. This is usually
212 * the case of a broken device.
213 */
214 if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) {
215 if (net_ratelimit())
216 pr_debug("%s: too much data\n", skb->dev->name);
217 dev_kfree_skb(skb);
218 return NULL;
219 }
220
221 while (len) {
222 set_skb_frag(skb, page, offset, &len);
223 page = (struct page *)page->private;
224 offset = 0;
225 }
226
227 if (page)
228 give_pages(vi, page);
229
230 return skb;
231}
232
233static int receive_mergeable(struct virtnet_info *vi, struct sk_buff *skb)
234{
235 struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
236 struct page *page;
237 int num_buf, i, len;
238
239 num_buf = hdr->mhdr.num_buffers;
240 while (--num_buf) {
241 i = skb_shinfo(skb)->nr_frags;
242 if (i >= MAX_SKB_FRAGS) {
243 pr_debug("%s: packet too long\n", skb->dev->name);
244 skb->dev->stats.rx_length_errors++;
245 return -EINVAL;
246 }
247 page = virtqueue_get_buf(vi->rvq, &len);
248 if (!page) {
249 pr_debug("%s: rx error: %d buffers missing\n",
250 skb->dev->name, hdr->mhdr.num_buffers);
251 skb->dev->stats.rx_length_errors++;
252 return -EINVAL;
253 }
254
255 if (len > PAGE_SIZE)
256 len = PAGE_SIZE;
257
258 set_skb_frag(skb, page, 0, &len);
259
260 --vi->num;
261 }
262 return 0;
263}
264
265static void receive_buf(struct net_device *dev, void *buf, unsigned int len)
266{
267 struct virtnet_info *vi = netdev_priv(dev);
268 struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
269 struct sk_buff *skb;
270 struct page *page;
271 struct skb_vnet_hdr *hdr;
272
273 if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
274 pr_debug("%s: short packet %i\n", dev->name, len);
275 dev->stats.rx_length_errors++;
276 if (vi->mergeable_rx_bufs || vi->big_packets)
277 give_pages(vi, buf);
278 else
279 dev_kfree_skb(buf);
280 return;
281 }
282
283 if (!vi->mergeable_rx_bufs && !vi->big_packets) {
284 skb = buf;
285 len -= sizeof(struct virtio_net_hdr);
286 skb_trim(skb, len);
287 } else {
288 page = buf;
289 skb = page_to_skb(vi, page, len);
290 if (unlikely(!skb)) {
291 dev->stats.rx_dropped++;
292 give_pages(vi, page);
293 return;
294 }
295 if (vi->mergeable_rx_bufs)
296 if (receive_mergeable(vi, skb)) {
297 dev_kfree_skb(skb);
298 return;
299 }
300 }
301
302 hdr = skb_vnet_hdr(skb);
303
304 u64_stats_update_begin(&stats->rx_syncp);
305 stats->rx_bytes += skb->len;
306 stats->rx_packets++;
307 u64_stats_update_end(&stats->rx_syncp);
308
309 if (hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
310 pr_debug("Needs csum!\n");
311 if (!skb_partial_csum_set(skb,
312 hdr->hdr.csum_start,
313 hdr->hdr.csum_offset))
314 goto frame_err;
315 } else if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) {
316 skb->ip_summed = CHECKSUM_UNNECESSARY;
317 }
318
319 skb->protocol = eth_type_trans(skb, dev);
320 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
321 ntohs(skb->protocol), skb->len, skb->pkt_type);
322
323 if (hdr->hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
324 pr_debug("GSO!\n");
325 switch (hdr->hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
326 case VIRTIO_NET_HDR_GSO_TCPV4:
327 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
328 break;
329 case VIRTIO_NET_HDR_GSO_UDP:
330 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
331 break;
332 case VIRTIO_NET_HDR_GSO_TCPV6:
333 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
334 break;
335 default:
336 if (net_ratelimit())
337 printk(KERN_WARNING "%s: bad gso type %u.\n",
338 dev->name, hdr->hdr.gso_type);
339 goto frame_err;
340 }
341
342 if (hdr->hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
343 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
344
345 skb_shinfo(skb)->gso_size = hdr->hdr.gso_size;
346 if (skb_shinfo(skb)->gso_size == 0) {
347 if (net_ratelimit())
348 printk(KERN_WARNING "%s: zero gso size.\n",
349 dev->name);
350 goto frame_err;
351 }
352
353 /* Header must be checked, and gso_segs computed. */
354 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
355 skb_shinfo(skb)->gso_segs = 0;
356 }
357
358 netif_receive_skb(skb);
359 return;
360
361frame_err:
362 dev->stats.rx_frame_errors++;
363 dev_kfree_skb(skb);
364}
365
366static int add_recvbuf_small(struct virtnet_info *vi, gfp_t gfp)
367{
368 struct sk_buff *skb;
369 struct skb_vnet_hdr *hdr;
370 int err;
371
372 skb = __netdev_alloc_skb_ip_align(vi->dev, MAX_PACKET_LEN, gfp);
373 if (unlikely(!skb))
374 return -ENOMEM;
375
376 skb_put(skb, MAX_PACKET_LEN);
377
378 hdr = skb_vnet_hdr(skb);
379 sg_set_buf(vi->rx_sg, &hdr->hdr, sizeof hdr->hdr);
380
381 skb_to_sgvec(skb, vi->rx_sg + 1, 0, skb->len);
382
383 err = virtqueue_add_buf(vi->rvq, vi->rx_sg, 0, 2, skb, gfp);
384 if (err < 0)
385 dev_kfree_skb(skb);
386
387 return err;
388}
389
390static int add_recvbuf_big(struct virtnet_info *vi, gfp_t gfp)
391{
392 struct page *first, *list = NULL;
393 char *p;
394 int i, err, offset;
395
396 /* page in vi->rx_sg[MAX_SKB_FRAGS + 1] is list tail */
397 for (i = MAX_SKB_FRAGS + 1; i > 1; --i) {
398 first = get_a_page(vi, gfp);
399 if (!first) {
400 if (list)
401 give_pages(vi, list);
402 return -ENOMEM;
403 }
404 sg_set_buf(&vi->rx_sg[i], page_address(first), PAGE_SIZE);
405
406 /* chain new page in list head to match sg */
407 first->private = (unsigned long)list;
408 list = first;
409 }
410
411 first = get_a_page(vi, gfp);
412 if (!first) {
413 give_pages(vi, list);
414 return -ENOMEM;
415 }
416 p = page_address(first);
417
418 /* vi->rx_sg[0], vi->rx_sg[1] share the same page */
419 /* a separated vi->rx_sg[0] for virtio_net_hdr only due to QEMU bug */
420 sg_set_buf(&vi->rx_sg[0], p, sizeof(struct virtio_net_hdr));
421
422 /* vi->rx_sg[1] for data packet, from offset */
423 offset = sizeof(struct padded_vnet_hdr);
424 sg_set_buf(&vi->rx_sg[1], p + offset, PAGE_SIZE - offset);
425
426 /* chain first in list head */
427 first->private = (unsigned long)list;
428 err = virtqueue_add_buf(vi->rvq, vi->rx_sg, 0, MAX_SKB_FRAGS + 2,
429 first, gfp);
430 if (err < 0)
431 give_pages(vi, first);
432
433 return err;
434}
435
436static int add_recvbuf_mergeable(struct virtnet_info *vi, gfp_t gfp)
437{
438 struct page *page;
439 int err;
440
441 page = get_a_page(vi, gfp);
442 if (!page)
443 return -ENOMEM;
444
445 sg_init_one(vi->rx_sg, page_address(page), PAGE_SIZE);
446
447 err = virtqueue_add_buf(vi->rvq, vi->rx_sg, 0, 1, page, gfp);
448 if (err < 0)
449 give_pages(vi, page);
450
451 return err;
452}
453
454/*
455 * Returns false if we couldn't fill entirely (OOM).
456 *
457 * Normally run in the receive path, but can also be run from ndo_open
458 * before we're receiving packets, or from refill_work which is
459 * careful to disable receiving (using napi_disable).
460 */
461static bool try_fill_recv(struct virtnet_info *vi, gfp_t gfp)
462{
463 int err;
464 bool oom;
465
466 do {
467 if (vi->mergeable_rx_bufs)
468 err = add_recvbuf_mergeable(vi, gfp);
469 else if (vi->big_packets)
470 err = add_recvbuf_big(vi, gfp);
471 else
472 err = add_recvbuf_small(vi, gfp);
473
474 oom = err == -ENOMEM;
475 if (err < 0)
476 break;
477 ++vi->num;
478 } while (err > 0);
479 if (unlikely(vi->num > vi->max))
480 vi->max = vi->num;
481 virtqueue_kick(vi->rvq);
482 return !oom;
483}
484
485static void skb_recv_done(struct virtqueue *rvq)
486{
487 struct virtnet_info *vi = rvq->vdev->priv;
488 /* Schedule NAPI, Suppress further interrupts if successful. */
489 if (napi_schedule_prep(&vi->napi)) {
490 virtqueue_disable_cb(rvq);
491 __napi_schedule(&vi->napi);
492 }
493}
494
495static void virtnet_napi_enable(struct virtnet_info *vi)
496{
497 napi_enable(&vi->napi);
498
499 /* If all buffers were filled by other side before we napi_enabled, we
500 * won't get another interrupt, so process any outstanding packets
501 * now. virtnet_poll wants re-enable the queue, so we disable here.
502 * We synchronize against interrupts via NAPI_STATE_SCHED */
503 if (napi_schedule_prep(&vi->napi)) {
504 virtqueue_disable_cb(vi->rvq);
505 local_bh_disable();
506 __napi_schedule(&vi->napi);
507 local_bh_enable();
508 }
509}
510
511static void refill_work(struct work_struct *work)
512{
513 struct virtnet_info *vi;
514 bool still_empty;
515
516 vi = container_of(work, struct virtnet_info, refill.work);
517 napi_disable(&vi->napi);
518 still_empty = !try_fill_recv(vi, GFP_KERNEL);
519 virtnet_napi_enable(vi);
520
521 /* In theory, this can happen: if we don't get any buffers in
522 * we will *never* try to fill again. */
523 if (still_empty)
524 queue_delayed_work(system_nrt_wq, &vi->refill, HZ/2);
525}
526
527static int virtnet_poll(struct napi_struct *napi, int budget)
528{
529 struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi);
530 void *buf;
531 unsigned int len, received = 0;
532
533again:
534 while (received < budget &&
535 (buf = virtqueue_get_buf(vi->rvq, &len)) != NULL) {
536 receive_buf(vi->dev, buf, len);
537 --vi->num;
538 received++;
539 }
540
541 if (vi->num < vi->max / 2) {
542 if (!try_fill_recv(vi, GFP_ATOMIC))
543 queue_delayed_work(system_nrt_wq, &vi->refill, 0);
544 }
545
546 /* Out of packets? */
547 if (received < budget) {
548 napi_complete(napi);
549 if (unlikely(!virtqueue_enable_cb(vi->rvq)) &&
550 napi_schedule_prep(napi)) {
551 virtqueue_disable_cb(vi->rvq);
552 __napi_schedule(napi);
553 goto again;
554 }
555 }
556
557 return received;
558}
559
560static unsigned int free_old_xmit_skbs(struct virtnet_info *vi)
561{
562 struct sk_buff *skb;
563 unsigned int len, tot_sgs = 0;
564 struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
565
566 while ((skb = virtqueue_get_buf(vi->svq, &len)) != NULL) {
567 pr_debug("Sent skb %p\n", skb);
568
569 u64_stats_update_begin(&stats->tx_syncp);
570 stats->tx_bytes += skb->len;
571 stats->tx_packets++;
572 u64_stats_update_end(&stats->tx_syncp);
573
574 tot_sgs += skb_vnet_hdr(skb)->num_sg;
575 dev_kfree_skb_any(skb);
576 }
577 return tot_sgs;
578}
579
580static int xmit_skb(struct virtnet_info *vi, struct sk_buff *skb)
581{
582 struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
583 const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
584
585 pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
586
587 if (skb->ip_summed == CHECKSUM_PARTIAL) {
588 hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
589 hdr->hdr.csum_start = skb_checksum_start_offset(skb);
590 hdr->hdr.csum_offset = skb->csum_offset;
591 } else {
592 hdr->hdr.flags = 0;
593 hdr->hdr.csum_offset = hdr->hdr.csum_start = 0;
594 }
595
596 if (skb_is_gso(skb)) {
597 hdr->hdr.hdr_len = skb_headlen(skb);
598 hdr->hdr.gso_size = skb_shinfo(skb)->gso_size;
599 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
600 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
601 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
602 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
603 else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
604 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
605 else
606 BUG();
607 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
608 hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
609 } else {
610 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
611 hdr->hdr.gso_size = hdr->hdr.hdr_len = 0;
612 }
613
614 hdr->mhdr.num_buffers = 0;
615
616 /* Encode metadata header at front. */
617 if (vi->mergeable_rx_bufs)
618 sg_set_buf(vi->tx_sg, &hdr->mhdr, sizeof hdr->mhdr);
619 else
620 sg_set_buf(vi->tx_sg, &hdr->hdr, sizeof hdr->hdr);
621
622 hdr->num_sg = skb_to_sgvec(skb, vi->tx_sg + 1, 0, skb->len) + 1;
623 return virtqueue_add_buf(vi->svq, vi->tx_sg, hdr->num_sg,
624 0, skb, GFP_ATOMIC);
625}
626
627static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
628{
629 struct virtnet_info *vi = netdev_priv(dev);
630 int capacity;
631
632 /* Free up any pending old buffers before queueing new ones. */
633 free_old_xmit_skbs(vi);
634
635 /* Try to transmit */
636 capacity = xmit_skb(vi, skb);
637
638 /* This can happen with OOM and indirect buffers. */
639 if (unlikely(capacity < 0)) {
640 if (likely(capacity == -ENOMEM)) {
641 if (net_ratelimit())
642 dev_warn(&dev->dev,
643 "TX queue failure: out of memory\n");
644 } else {
645 dev->stats.tx_fifo_errors++;
646 if (net_ratelimit())
647 dev_warn(&dev->dev,
648 "Unexpected TX queue failure: %d\n",
649 capacity);
650 }
651 dev->stats.tx_dropped++;
652 kfree_skb(skb);
653 return NETDEV_TX_OK;
654 }
655 virtqueue_kick(vi->svq);
656
657 /* Don't wait up for transmitted skbs to be freed. */
658 skb_orphan(skb);
659 nf_reset(skb);
660
661 /* Apparently nice girls don't return TX_BUSY; stop the queue
662 * before it gets out of hand. Naturally, this wastes entries. */
663 if (capacity < 2+MAX_SKB_FRAGS) {
664 netif_stop_queue(dev);
665 if (unlikely(!virtqueue_enable_cb_delayed(vi->svq))) {
666 /* More just got used, free them then recheck. */
667 capacity += free_old_xmit_skbs(vi);
668 if (capacity >= 2+MAX_SKB_FRAGS) {
669 netif_start_queue(dev);
670 virtqueue_disable_cb(vi->svq);
671 }
672 }
673 }
674
675 return NETDEV_TX_OK;
676}
677
678static int virtnet_set_mac_address(struct net_device *dev, void *p)
679{
680 struct virtnet_info *vi = netdev_priv(dev);
681 struct virtio_device *vdev = vi->vdev;
682 int ret;
683
684 ret = eth_mac_addr(dev, p);
685 if (ret)
686 return ret;
687
688 if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
689 vdev->config->set(vdev, offsetof(struct virtio_net_config, mac),
690 dev->dev_addr, dev->addr_len);
691
692 return 0;
693}
694
695static struct rtnl_link_stats64 *virtnet_stats(struct net_device *dev,
696 struct rtnl_link_stats64 *tot)
697{
698 struct virtnet_info *vi = netdev_priv(dev);
699 int cpu;
700 unsigned int start;
701
702 for_each_possible_cpu(cpu) {
703 struct virtnet_stats *stats = per_cpu_ptr(vi->stats, cpu);
704 u64 tpackets, tbytes, rpackets, rbytes;
705
706 do {
707 start = u64_stats_fetch_begin(&stats->tx_syncp);
708 tpackets = stats->tx_packets;
709 tbytes = stats->tx_bytes;
710 } while (u64_stats_fetch_retry(&stats->tx_syncp, start));
711
712 do {
713 start = u64_stats_fetch_begin(&stats->rx_syncp);
714 rpackets = stats->rx_packets;
715 rbytes = stats->rx_bytes;
716 } while (u64_stats_fetch_retry(&stats->rx_syncp, start));
717
718 tot->rx_packets += rpackets;
719 tot->tx_packets += tpackets;
720 tot->rx_bytes += rbytes;
721 tot->tx_bytes += tbytes;
722 }
723
724 tot->tx_dropped = dev->stats.tx_dropped;
725 tot->tx_fifo_errors = dev->stats.tx_fifo_errors;
726 tot->rx_dropped = dev->stats.rx_dropped;
727 tot->rx_length_errors = dev->stats.rx_length_errors;
728 tot->rx_frame_errors = dev->stats.rx_frame_errors;
729
730 return tot;
731}
732
733#ifdef CONFIG_NET_POLL_CONTROLLER
734static void virtnet_netpoll(struct net_device *dev)
735{
736 struct virtnet_info *vi = netdev_priv(dev);
737
738 napi_schedule(&vi->napi);
739}
740#endif
741
742static int virtnet_open(struct net_device *dev)
743{
744 struct virtnet_info *vi = netdev_priv(dev);
745
746 /* Make sure we have some buffers: if oom use wq. */
747 if (!try_fill_recv(vi, GFP_KERNEL))
748 queue_delayed_work(system_nrt_wq, &vi->refill, 0);
749
750 virtnet_napi_enable(vi);
751 return 0;
752}
753
754/*
755 * Send command via the control virtqueue and check status. Commands
756 * supported by the hypervisor, as indicated by feature bits, should
757 * never fail unless improperly formated.
758 */
759static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
760 struct scatterlist *data, int out, int in)
761{
762 struct scatterlist *s, sg[VIRTNET_SEND_COMMAND_SG_MAX + 2];
763 struct virtio_net_ctrl_hdr ctrl;
764 virtio_net_ctrl_ack status = ~0;
765 unsigned int tmp;
766 int i;
767
768 /* Caller should know better */
769 BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ||
770 (out + in > VIRTNET_SEND_COMMAND_SG_MAX));
771
772 out++; /* Add header */
773 in++; /* Add return status */
774
775 ctrl.class = class;
776 ctrl.cmd = cmd;
777
778 sg_init_table(sg, out + in);
779
780 sg_set_buf(&sg[0], &ctrl, sizeof(ctrl));
781 for_each_sg(data, s, out + in - 2, i)
782 sg_set_buf(&sg[i + 1], sg_virt(s), s->length);
783 sg_set_buf(&sg[out + in - 1], &status, sizeof(status));
784
785 BUG_ON(virtqueue_add_buf(vi->cvq, sg, out, in, vi, GFP_ATOMIC) < 0);
786
787 virtqueue_kick(vi->cvq);
788
789 /*
790 * Spin for a response, the kick causes an ioport write, trapping
791 * into the hypervisor, so the request should be handled immediately.
792 */
793 while (!virtqueue_get_buf(vi->cvq, &tmp))
794 cpu_relax();
795
796 return status == VIRTIO_NET_OK;
797}
798
799static void virtnet_ack_link_announce(struct virtnet_info *vi)
800{
801 rtnl_lock();
802 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE,
803 VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL,
804 0, 0))
805 dev_warn(&vi->dev->dev, "Failed to ack link announce.\n");
806 rtnl_unlock();
807}
808
809static int virtnet_close(struct net_device *dev)
810{
811 struct virtnet_info *vi = netdev_priv(dev);
812
813 /* Make sure refill_work doesn't re-enable napi! */
814 cancel_delayed_work_sync(&vi->refill);
815 napi_disable(&vi->napi);
816
817 return 0;
818}
819
820static void virtnet_set_rx_mode(struct net_device *dev)
821{
822 struct virtnet_info *vi = netdev_priv(dev);
823 struct scatterlist sg[2];
824 u8 promisc, allmulti;
825 struct virtio_net_ctrl_mac *mac_data;
826 struct netdev_hw_addr *ha;
827 int uc_count;
828 int mc_count;
829 void *buf;
830 int i;
831
832 /* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */
833 if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
834 return;
835
836 promisc = ((dev->flags & IFF_PROMISC) != 0);
837 allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
838
839 sg_init_one(sg, &promisc, sizeof(promisc));
840
841 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
842 VIRTIO_NET_CTRL_RX_PROMISC,
843 sg, 1, 0))
844 dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
845 promisc ? "en" : "dis");
846
847 sg_init_one(sg, &allmulti, sizeof(allmulti));
848
849 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
850 VIRTIO_NET_CTRL_RX_ALLMULTI,
851 sg, 1, 0))
852 dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
853 allmulti ? "en" : "dis");
854
855 uc_count = netdev_uc_count(dev);
856 mc_count = netdev_mc_count(dev);
857 /* MAC filter - use one buffer for both lists */
858 buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
859 (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
860 mac_data = buf;
861 if (!buf) {
862 dev_warn(&dev->dev, "No memory for MAC address buffer\n");
863 return;
864 }
865
866 sg_init_table(sg, 2);
867
868 /* Store the unicast list and count in the front of the buffer */
869 mac_data->entries = uc_count;
870 i = 0;
871 netdev_for_each_uc_addr(ha, dev)
872 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
873
874 sg_set_buf(&sg[0], mac_data,
875 sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
876
877 /* multicast list and count fill the end */
878 mac_data = (void *)&mac_data->macs[uc_count][0];
879
880 mac_data->entries = mc_count;
881 i = 0;
882 netdev_for_each_mc_addr(ha, dev)
883 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
884
885 sg_set_buf(&sg[1], mac_data,
886 sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
887
888 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
889 VIRTIO_NET_CTRL_MAC_TABLE_SET,
890 sg, 2, 0))
891 dev_warn(&dev->dev, "Failed to set MAC fitler table.\n");
892
893 kfree(buf);
894}
895
896static int virtnet_vlan_rx_add_vid(struct net_device *dev, u16 vid)
897{
898 struct virtnet_info *vi = netdev_priv(dev);
899 struct scatterlist sg;
900
901 sg_init_one(&sg, &vid, sizeof(vid));
902
903 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
904 VIRTIO_NET_CTRL_VLAN_ADD, &sg, 1, 0))
905 dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
906 return 0;
907}
908
909static int virtnet_vlan_rx_kill_vid(struct net_device *dev, u16 vid)
910{
911 struct virtnet_info *vi = netdev_priv(dev);
912 struct scatterlist sg;
913
914 sg_init_one(&sg, &vid, sizeof(vid));
915
916 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
917 VIRTIO_NET_CTRL_VLAN_DEL, &sg, 1, 0))
918 dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
919 return 0;
920}
921
922static void virtnet_get_ringparam(struct net_device *dev,
923 struct ethtool_ringparam *ring)
924{
925 struct virtnet_info *vi = netdev_priv(dev);
926
927 ring->rx_max_pending = virtqueue_get_vring_size(vi->rvq);
928 ring->tx_max_pending = virtqueue_get_vring_size(vi->svq);
929 ring->rx_pending = ring->rx_max_pending;
930 ring->tx_pending = ring->tx_max_pending;
931
932}
933
934
935static void virtnet_get_drvinfo(struct net_device *dev,
936 struct ethtool_drvinfo *info)
937{
938 struct virtnet_info *vi = netdev_priv(dev);
939 struct virtio_device *vdev = vi->vdev;
940
941 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
942 strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version));
943 strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info));
944
945}
946
947static const struct ethtool_ops virtnet_ethtool_ops = {
948 .get_drvinfo = virtnet_get_drvinfo,
949 .get_link = ethtool_op_get_link,
950 .get_ringparam = virtnet_get_ringparam,
951};
952
953#define MIN_MTU 68
954#define MAX_MTU 65535
955
956static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
957{
958 if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
959 return -EINVAL;
960 dev->mtu = new_mtu;
961 return 0;
962}
963
964static const struct net_device_ops virtnet_netdev = {
965 .ndo_open = virtnet_open,
966 .ndo_stop = virtnet_close,
967 .ndo_start_xmit = start_xmit,
968 .ndo_validate_addr = eth_validate_addr,
969 .ndo_set_mac_address = virtnet_set_mac_address,
970 .ndo_set_rx_mode = virtnet_set_rx_mode,
971 .ndo_change_mtu = virtnet_change_mtu,
972 .ndo_get_stats64 = virtnet_stats,
973 .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
974 .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
975#ifdef CONFIG_NET_POLL_CONTROLLER
976 .ndo_poll_controller = virtnet_netpoll,
977#endif
978};
979
980static void virtnet_config_changed_work(struct work_struct *work)
981{
982 struct virtnet_info *vi =
983 container_of(work, struct virtnet_info, config_work);
984 u16 v;
985
986 mutex_lock(&vi->config_lock);
987 if (!vi->config_enable)
988 goto done;
989
990 if (virtio_config_val(vi->vdev, VIRTIO_NET_F_STATUS,
991 offsetof(struct virtio_net_config, status),
992 &v) < 0)
993 goto done;
994
995 if (v & VIRTIO_NET_S_ANNOUNCE) {
996 netif_notify_peers(vi->dev);
997 virtnet_ack_link_announce(vi);
998 }
999
1000 /* Ignore unknown (future) status bits */
1001 v &= VIRTIO_NET_S_LINK_UP;
1002
1003 if (vi->status == v)
1004 goto done;
1005
1006 vi->status = v;
1007
1008 if (vi->status & VIRTIO_NET_S_LINK_UP) {
1009 netif_carrier_on(vi->dev);
1010 netif_wake_queue(vi->dev);
1011 } else {
1012 netif_carrier_off(vi->dev);
1013 netif_stop_queue(vi->dev);
1014 }
1015done:
1016 mutex_unlock(&vi->config_lock);
1017}
1018
1019static void virtnet_config_changed(struct virtio_device *vdev)
1020{
1021 struct virtnet_info *vi = vdev->priv;
1022
1023 queue_work(system_nrt_wq, &vi->config_work);
1024}
1025
1026static int init_vqs(struct virtnet_info *vi)
1027{
1028 struct virtqueue *vqs[3];
1029 vq_callback_t *callbacks[] = { skb_recv_done, skb_xmit_done, NULL};
1030 const char *names[] = { "input", "output", "control" };
1031 int nvqs, err;
1032
1033 /* We expect two virtqueues, receive then send,
1034 * and optionally control. */
1035 nvqs = virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ? 3 : 2;
1036
1037 err = vi->vdev->config->find_vqs(vi->vdev, nvqs, vqs, callbacks, names);
1038 if (err)
1039 return err;
1040
1041 vi->rvq = vqs[0];
1042 vi->svq = vqs[1];
1043
1044 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)) {
1045 vi->cvq = vqs[2];
1046
1047 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
1048 vi->dev->features |= NETIF_F_HW_VLAN_FILTER;
1049 }
1050 return 0;
1051}
1052
1053static int virtnet_probe(struct virtio_device *vdev)
1054{
1055 int err;
1056 struct net_device *dev;
1057 struct virtnet_info *vi;
1058
1059 /* Allocate ourselves a network device with room for our info */
1060 dev = alloc_etherdev(sizeof(struct virtnet_info));
1061 if (!dev)
1062 return -ENOMEM;
1063
1064 /* Set up network device as normal. */
1065 dev->priv_flags |= IFF_UNICAST_FLT;
1066 dev->netdev_ops = &virtnet_netdev;
1067 dev->features = NETIF_F_HIGHDMA;
1068
1069 SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
1070 SET_NETDEV_DEV(dev, &vdev->dev);
1071
1072 /* Do we support "hardware" checksums? */
1073 if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
1074 /* This opens up the world of extra features. */
1075 dev->hw_features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1076 if (csum)
1077 dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1078
1079 if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
1080 dev->hw_features |= NETIF_F_TSO | NETIF_F_UFO
1081 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
1082 }
1083 /* Individual feature bits: what can host handle? */
1084 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
1085 dev->hw_features |= NETIF_F_TSO;
1086 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
1087 dev->hw_features |= NETIF_F_TSO6;
1088 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
1089 dev->hw_features |= NETIF_F_TSO_ECN;
1090 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
1091 dev->hw_features |= NETIF_F_UFO;
1092
1093 if (gso)
1094 dev->features |= dev->hw_features & (NETIF_F_ALL_TSO|NETIF_F_UFO);
1095 /* (!csum && gso) case will be fixed by register_netdev() */
1096 }
1097
1098 /* Configuration may specify what MAC to use. Otherwise random. */
1099 if (virtio_config_val_len(vdev, VIRTIO_NET_F_MAC,
1100 offsetof(struct virtio_net_config, mac),
1101 dev->dev_addr, dev->addr_len) < 0)
1102 eth_hw_addr_random(dev);
1103
1104 /* Set up our device-specific information */
1105 vi = netdev_priv(dev);
1106 netif_napi_add(dev, &vi->napi, virtnet_poll, napi_weight);
1107 vi->dev = dev;
1108 vi->vdev = vdev;
1109 vdev->priv = vi;
1110 vi->pages = NULL;
1111 vi->stats = alloc_percpu(struct virtnet_stats);
1112 err = -ENOMEM;
1113 if (vi->stats == NULL)
1114 goto free;
1115
1116 INIT_DELAYED_WORK(&vi->refill, refill_work);
1117 mutex_init(&vi->config_lock);
1118 vi->config_enable = true;
1119 INIT_WORK(&vi->config_work, virtnet_config_changed_work);
1120 sg_init_table(vi->rx_sg, ARRAY_SIZE(vi->rx_sg));
1121 sg_init_table(vi->tx_sg, ARRAY_SIZE(vi->tx_sg));
1122
1123 /* If we can receive ANY GSO packets, we must allocate large ones. */
1124 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
1125 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
1126 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
1127 vi->big_packets = true;
1128
1129 if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
1130 vi->mergeable_rx_bufs = true;
1131
1132 err = init_vqs(vi);
1133 if (err)
1134 goto free_stats;
1135
1136 err = register_netdev(dev);
1137 if (err) {
1138 pr_debug("virtio_net: registering device failed\n");
1139 goto free_vqs;
1140 }
1141
1142 /* Last of all, set up some receive buffers. */
1143 try_fill_recv(vi, GFP_KERNEL);
1144
1145 /* If we didn't even get one input buffer, we're useless. */
1146 if (vi->num == 0) {
1147 err = -ENOMEM;
1148 goto unregister;
1149 }
1150
1151 /* Assume link up if device can't report link status,
1152 otherwise get link status from config. */
1153 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
1154 netif_carrier_off(dev);
1155 queue_work(system_nrt_wq, &vi->config_work);
1156 } else {
1157 vi->status = VIRTIO_NET_S_LINK_UP;
1158 netif_carrier_on(dev);
1159 }
1160
1161 pr_debug("virtnet: registered device %s\n", dev->name);
1162 return 0;
1163
1164unregister:
1165 unregister_netdev(dev);
1166free_vqs:
1167 vdev->config->del_vqs(vdev);
1168free_stats:
1169 free_percpu(vi->stats);
1170free:
1171 free_netdev(dev);
1172 return err;
1173}
1174
1175static void free_unused_bufs(struct virtnet_info *vi)
1176{
1177 void *buf;
1178 while (1) {
1179 buf = virtqueue_detach_unused_buf(vi->svq);
1180 if (!buf)
1181 break;
1182 dev_kfree_skb(buf);
1183 }
1184 while (1) {
1185 buf = virtqueue_detach_unused_buf(vi->rvq);
1186 if (!buf)
1187 break;
1188 if (vi->mergeable_rx_bufs || vi->big_packets)
1189 give_pages(vi, buf);
1190 else
1191 dev_kfree_skb(buf);
1192 --vi->num;
1193 }
1194 BUG_ON(vi->num != 0);
1195}
1196
1197static void remove_vq_common(struct virtnet_info *vi)
1198{
1199 vi->vdev->config->reset(vi->vdev);
1200
1201 /* Free unused buffers in both send and recv, if any. */
1202 free_unused_bufs(vi);
1203
1204 vi->vdev->config->del_vqs(vi->vdev);
1205
1206 while (vi->pages)
1207 __free_pages(get_a_page(vi, GFP_KERNEL), 0);
1208}
1209
1210static void __devexit virtnet_remove(struct virtio_device *vdev)
1211{
1212 struct virtnet_info *vi = vdev->priv;
1213
1214 /* Prevent config work handler from accessing the device. */
1215 mutex_lock(&vi->config_lock);
1216 vi->config_enable = false;
1217 mutex_unlock(&vi->config_lock);
1218
1219 unregister_netdev(vi->dev);
1220
1221 remove_vq_common(vi);
1222
1223 flush_work(&vi->config_work);
1224
1225 free_percpu(vi->stats);
1226 free_netdev(vi->dev);
1227}
1228
1229#ifdef CONFIG_PM
1230static int virtnet_freeze(struct virtio_device *vdev)
1231{
1232 struct virtnet_info *vi = vdev->priv;
1233
1234 /* Prevent config work handler from accessing the device */
1235 mutex_lock(&vi->config_lock);
1236 vi->config_enable = false;
1237 mutex_unlock(&vi->config_lock);
1238
1239 netif_device_detach(vi->dev);
1240 cancel_delayed_work_sync(&vi->refill);
1241
1242 if (netif_running(vi->dev))
1243 napi_disable(&vi->napi);
1244
1245 remove_vq_common(vi);
1246
1247 flush_work(&vi->config_work);
1248
1249 return 0;
1250}
1251
1252static int virtnet_restore(struct virtio_device *vdev)
1253{
1254 struct virtnet_info *vi = vdev->priv;
1255 int err;
1256
1257 err = init_vqs(vi);
1258 if (err)
1259 return err;
1260
1261 if (netif_running(vi->dev))
1262 virtnet_napi_enable(vi);
1263
1264 netif_device_attach(vi->dev);
1265
1266 if (!try_fill_recv(vi, GFP_KERNEL))
1267 queue_delayed_work(system_nrt_wq, &vi->refill, 0);
1268
1269 mutex_lock(&vi->config_lock);
1270 vi->config_enable = true;
1271 mutex_unlock(&vi->config_lock);
1272
1273 return 0;
1274}
1275#endif
1276
1277static struct virtio_device_id id_table[] = {
1278 { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
1279 { 0 },
1280};
1281
1282static unsigned int features[] = {
1283 VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
1284 VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
1285 VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
1286 VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
1287 VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO,
1288 VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ,
1289 VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN,
1290 VIRTIO_NET_F_GUEST_ANNOUNCE,
1291};
1292
1293static struct virtio_driver virtio_net_driver = {
1294 .feature_table = features,
1295 .feature_table_size = ARRAY_SIZE(features),
1296 .driver.name = KBUILD_MODNAME,
1297 .driver.owner = THIS_MODULE,
1298 .id_table = id_table,
1299 .probe = virtnet_probe,
1300 .remove = __devexit_p(virtnet_remove),
1301 .config_changed = virtnet_config_changed,
1302#ifdef CONFIG_PM
1303 .freeze = virtnet_freeze,
1304 .restore = virtnet_restore,
1305#endif
1306};
1307
1308static int __init init(void)
1309{
1310 return register_virtio_driver(&virtio_net_driver);
1311}
1312
1313static void __exit fini(void)
1314{
1315 unregister_virtio_driver(&virtio_net_driver);
1316}
1317module_init(init);
1318module_exit(fini);
1319
1320MODULE_DEVICE_TABLE(virtio, id_table);
1321MODULE_DESCRIPTION("Virtio network driver");
1322MODULE_LICENSE("GPL");