1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
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#include <linux/cpu.h>
30
31static int napi_weight = NAPI_POLL_WEIGHT;
32module_param(napi_weight, int, 0444);
33
34static bool csum = true, gso = true;
35module_param(csum, bool, 0444);
36module_param(gso, bool, 0444);
37
38
39#define MAX_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
40#define GOOD_COPY_LEN 128
41
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
54
55struct send_queue {
56
57 struct virtqueue *vq;
58
59
60 struct scatterlist sg[MAX_SKB_FRAGS + 2];
61
62
63 char name[40];
64};
65
66
67struct receive_queue {
68
69 struct virtqueue *vq;
70
71 struct napi_struct napi;
72
73
74 unsigned int num, max;
75
76
77 struct page *pages;
78
79
80 struct scatterlist sg[MAX_SKB_FRAGS + 2];
81
82
83 char name[40];
84};
85
86struct virtnet_info {
87 struct virtio_device *vdev;
88 struct virtqueue *cvq;
89 struct net_device *dev;
90 struct send_queue *sq;
91 struct receive_queue *rq;
92 unsigned int status;
93
94
95 u16 max_queue_pairs;
96
97
98 u16 curr_queue_pairs;
99
100
101 bool big_packets;
102
103
104 bool mergeable_rx_bufs;
105
106
107 bool has_cvq;
108
109
110 bool any_header_sg;
111
112
113 u8 hdr_len;
114
115
116 struct virtnet_stats __percpu *stats;
117
118
119 struct delayed_work refill;
120
121
122 struct work_struct config_work;
123
124
125 bool affinity_hint_set;
126
127
128 int __percpu *vq_index;
129
130
131 struct notifier_block nb;
132
133
134 u16 max_mtu;
135};
136
137struct padded_vnet_hdr {
138 struct virtio_net_hdr_mrg_rxbuf hdr;
139
140
141
142
143
144 char padding[4];
145};
146
147
148
149
150static int vq2txq(struct virtqueue *vq)
151{
152 return (vq->index - 1) / 2;
153}
154
155static int txq2vq(int txq)
156{
157 return txq * 2 + 1;
158}
159
160static int vq2rxq(struct virtqueue *vq)
161{
162 return vq->index / 2;
163}
164
165static int rxq2vq(int rxq)
166{
167 return rxq * 2;
168}
169
170static inline struct virtio_net_hdr_mrg_rxbuf *skb_vnet_hdr(struct sk_buff *skb)
171{
172 return (struct virtio_net_hdr_mrg_rxbuf *)skb->cb;
173}
174
175
176
177
178
179static void give_pages(struct receive_queue *rq, struct page *page)
180{
181 struct page *end;
182
183
184 for (end = page; end->private; end = (struct page *)end->private);
185 end->private = (unsigned long)rq->pages;
186 rq->pages = page;
187}
188
189static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask)
190{
191 struct page *p = rq->pages;
192
193 if (p) {
194 rq->pages = (struct page *)p->private;
195
196 p->private = 0;
197 } else
198 p = alloc_page(gfp_mask);
199 return p;
200}
201
202static void skb_xmit_done(struct virtqueue *vq)
203{
204 struct virtnet_info *vi = vq->vdev->priv;
205
206
207 virtqueue_disable_cb(vq);
208
209
210 netif_wake_subqueue(vi->dev, vq2txq(vq));
211}
212
213static void set_skb_frag(struct sk_buff *skb, struct page *page,
214 unsigned int offset, unsigned int *len)
215{
216 int size = min((unsigned)PAGE_SIZE - offset, *len);
217 int i = skb_shinfo(skb)->nr_frags;
218
219 __skb_fill_page_desc(skb, i, page, offset, size);
220
221 skb->data_len += size;
222 skb->len += size;
223 skb->truesize += PAGE_SIZE;
224 skb_shinfo(skb)->nr_frags++;
225 skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
226 *len -= size;
227}
228
229
230static struct sk_buff *page_to_skb(struct virtnet_info *vi,
231 struct receive_queue *rq,
232 struct page *page, unsigned int len)
233{
234 struct sk_buff *skb;
235 struct virtio_net_hdr_mrg_rxbuf *hdr;
236 unsigned int copy, hdr_len, offset;
237 char *p;
238
239 p = page_address(page);
240
241
242 skb = netdev_alloc_skb_ip_align(vi->dev, GOOD_COPY_LEN);
243 if (unlikely(!skb))
244 return NULL;
245
246 hdr = skb_vnet_hdr(skb);
247
248 hdr_len = vi->hdr_len;
249 if (vi->mergeable_rx_bufs)
250 offset = sizeof *hdr;
251 else
252 offset = sizeof(struct padded_vnet_hdr);
253
254 memcpy(hdr, p, hdr_len);
255
256 len -= hdr_len;
257 p += offset;
258
259 copy = len;
260 if (copy > skb_tailroom(skb))
261 copy = skb_tailroom(skb);
262 memcpy(skb_put(skb, copy), p, copy);
263
264 len -= copy;
265 offset += copy;
266
267
268
269
270
271
272
273 if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) {
274 net_dbg_ratelimited("%s: too much data\n", skb->dev->name);
275 dev_kfree_skb(skb);
276 return NULL;
277 }
278
279 while (len) {
280 set_skb_frag(skb, page, offset, &len);
281 page = (struct page *)page->private;
282 offset = 0;
283 }
284
285 if (page)
286 give_pages(rq, page);
287
288 return skb;
289}
290
291static struct sk_buff *receive_small(struct virtnet_info *vi, void *buf, unsigned int len)
292{
293 struct sk_buff * skb = buf;
294
295 len -= vi->hdr_len;
296 skb_trim(skb, len);
297
298 return skb;
299}
300
301static struct sk_buff *receive_big(struct net_device *dev,
302 struct virtnet_info *vi,
303 struct receive_queue *rq,
304 void *buf, unsigned int len)
305{
306 struct page *page = buf;
307 struct sk_buff *skb = page_to_skb(vi, rq, page, len);
308
309 if (unlikely(!skb))
310 goto err;
311
312 return skb;
313
314err:
315 dev->stats.rx_dropped++;
316 give_pages(rq, page);
317 return NULL;
318}
319
320static struct sk_buff *receive_mergeable(struct net_device *dev,
321 struct virtnet_info *vi,
322 struct receive_queue *rq,
323 void *buf,
324 unsigned int len)
325{
326 struct virtio_net_hdr_mrg_rxbuf *hdr = page_address(buf);
327 u16 num_buf = virtio16_to_cpu(rq->vq->vdev, hdr->num_buffers);
328 struct page *page = buf;
329 struct sk_buff *skb = page_to_skb(vi, rq, page, len);
330 int i;
331
332 if (unlikely(!skb))
333 goto err_skb;
334
335 while (--num_buf) {
336 i = skb_shinfo(skb)->nr_frags;
337 if (i >= MAX_SKB_FRAGS) {
338 pr_debug("%s: packet too long\n", skb->dev->name);
339 skb->dev->stats.rx_length_errors++;
340 return NULL;
341 }
342 page = virtqueue_get_buf(rq->vq, &len);
343 if (!page) {
344 pr_debug("%s: rx error: %d buffers %d missing\n",
345 dev->name,
346 virtio16_to_cpu(rq->vq->vdev,
347 hdr->num_buffers),
348 num_buf);
349 dev->stats.rx_length_errors++;
350 goto err_buf;
351 }
352
353 if (len > PAGE_SIZE)
354 len = PAGE_SIZE;
355
356 set_skb_frag(skb, page, 0, &len);
357
358 --rq->num;
359 }
360 return skb;
361err_skb:
362 give_pages(rq, page);
363 while (--num_buf) {
364 buf = virtqueue_get_buf(rq->vq, &len);
365 if (unlikely(!buf)) {
366 pr_debug("%s: rx error: %d buffers missing\n",
367 dev->name, num_buf);
368 dev->stats.rx_length_errors++;
369 break;
370 }
371 page = buf;
372 give_pages(rq, page);
373 --rq->num;
374 }
375err_buf:
376 dev->stats.rx_dropped++;
377 dev_kfree_skb(skb);
378 return NULL;
379}
380
381static void receive_buf(struct virtnet_info *vi, struct receive_queue *rq,
382 void *buf, unsigned int len)
383{
384 struct net_device *dev = vi->dev;
385 struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
386 struct sk_buff *skb;
387 struct virtio_net_hdr_mrg_rxbuf *hdr;
388
389 if (unlikely(len < vi->hdr_len + ETH_HLEN)) {
390 pr_debug("%s: short packet %i\n", dev->name, len);
391 dev->stats.rx_length_errors++;
392 if (vi->mergeable_rx_bufs || vi->big_packets)
393 give_pages(rq, buf);
394 else
395 dev_kfree_skb(buf);
396 return;
397 }
398 if (vi->mergeable_rx_bufs)
399 skb = receive_mergeable(dev, vi, rq, buf, len);
400 else if (vi->big_packets)
401 skb = receive_big(dev, vi, rq, buf, len);
402 else
403 skb = receive_small(vi, buf, len);
404
405 if (unlikely(!skb))
406 return;
407
408 hdr = skb_vnet_hdr(skb);
409
410 u64_stats_update_begin(&stats->rx_syncp);
411 stats->rx_bytes += skb->len;
412 stats->rx_packets++;
413 u64_stats_update_end(&stats->rx_syncp);
414
415 if (hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
416 pr_debug("Needs csum!\n");
417 if (!skb_partial_csum_set(skb,
418 virtio16_to_cpu(vi->vdev, hdr->hdr.csum_start),
419 virtio16_to_cpu(vi->vdev, hdr->hdr.csum_offset)))
420 goto frame_err;
421 } else if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) {
422 skb->ip_summed = CHECKSUM_UNNECESSARY;
423 }
424
425 skb->protocol = eth_type_trans(skb, dev);
426 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
427 ntohs(skb->protocol), skb->len, skb->pkt_type);
428
429 if (hdr->hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
430 pr_debug("GSO!\n");
431 switch (hdr->hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
432 case VIRTIO_NET_HDR_GSO_TCPV4:
433 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
434 break;
435 case VIRTIO_NET_HDR_GSO_UDP:
436 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
437 break;
438 case VIRTIO_NET_HDR_GSO_TCPV6:
439 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
440 break;
441 default:
442 net_warn_ratelimited("%s: bad gso type %u.\n",
443 dev->name, hdr->hdr.gso_type);
444 goto frame_err;
445 }
446
447 if (hdr->hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
448 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
449
450 skb_shinfo(skb)->gso_size = virtio16_to_cpu(vi->vdev,
451 hdr->hdr.gso_size);
452 if (skb_shinfo(skb)->gso_size == 0) {
453 net_warn_ratelimited("%s: zero gso size.\n", dev->name);
454 goto frame_err;
455 }
456
457
458 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
459 skb_shinfo(skb)->gso_segs = 0;
460 }
461
462 napi_gro_receive(&rq->napi, skb);
463 return;
464
465frame_err:
466 dev->stats.rx_frame_errors++;
467 dev_kfree_skb(skb);
468}
469
470static int add_recvbuf_small(struct virtnet_info *vi, struct receive_queue *rq,
471 gfp_t gfp)
472{
473 struct sk_buff *skb;
474 struct virtio_net_hdr_mrg_rxbuf *hdr;
475 int err;
476
477 skb = __netdev_alloc_skb_ip_align(vi->dev, MAX_PACKET_LEN, gfp);
478 if (unlikely(!skb))
479 return -ENOMEM;
480
481 skb_put(skb, MAX_PACKET_LEN);
482
483 hdr = skb_vnet_hdr(skb);
484 sg_init_table(rq->sg, MAX_SKB_FRAGS + 2);
485 sg_set_buf(rq->sg, hdr, vi->hdr_len);
486 skb_to_sgvec(skb, rq->sg + 1, 0, skb->len);
487
488 err = virtqueue_add_inbuf(rq->vq, rq->sg, 2, skb, gfp);
489 if (err < 0)
490 dev_kfree_skb(skb);
491
492 return err;
493}
494
495static int add_recvbuf_big(struct virtnet_info *vi, struct receive_queue *rq,
496 gfp_t gfp)
497{
498 struct page *first, *list = NULL;
499 char *p;
500 int i, err, offset;
501
502 sg_init_table(rq->sg, MAX_SKB_FRAGS + 2);
503
504
505 for (i = MAX_SKB_FRAGS + 1; i > 1; --i) {
506 first = get_a_page(rq, gfp);
507 if (!first) {
508 if (list)
509 give_pages(rq, list);
510 return -ENOMEM;
511 }
512 sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE);
513
514
515 first->private = (unsigned long)list;
516 list = first;
517 }
518
519 first = get_a_page(rq, gfp);
520 if (!first) {
521 give_pages(rq, list);
522 return -ENOMEM;
523 }
524 p = page_address(first);
525
526
527
528 sg_set_buf(&rq->sg[0], p, vi->hdr_len);
529
530
531 offset = sizeof(struct padded_vnet_hdr);
532 sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset);
533
534
535 first->private = (unsigned long)list;
536 err = virtqueue_add_inbuf(rq->vq, rq->sg, MAX_SKB_FRAGS + 2,
537 first, gfp);
538 if (err < 0)
539 give_pages(rq, first);
540
541 return err;
542}
543
544static int add_recvbuf_mergeable(struct receive_queue *rq, gfp_t gfp)
545{
546 struct page *page;
547 int err;
548
549 page = get_a_page(rq, gfp);
550 if (!page)
551 return -ENOMEM;
552
553 sg_init_one(rq->sg, page_address(page), PAGE_SIZE);
554
555 err = virtqueue_add_inbuf(rq->vq, rq->sg, 1, page, gfp);
556 if (err < 0)
557 give_pages(rq, page);
558
559 return err;
560}
561
562
563
564
565
566
567
568
569static bool try_fill_recv(struct virtnet_info *vi, struct receive_queue *rq,
570 gfp_t gfp)
571{
572 int err;
573 bool oom;
574
575 do {
576 if (vi->mergeable_rx_bufs)
577 err = add_recvbuf_mergeable(rq, gfp);
578 else if (vi->big_packets)
579 err = add_recvbuf_big(vi, rq, gfp);
580 else
581 err = add_recvbuf_small(vi, rq, gfp);
582
583 oom = err == -ENOMEM;
584 if (err)
585 break;
586 ++rq->num;
587 } while (rq->vq->num_free);
588 if (unlikely(rq->num > rq->max))
589 rq->max = rq->num;
590 virtqueue_kick(rq->vq);
591 return !oom;
592}
593
594static void skb_recv_done(struct virtqueue *rvq)
595{
596 struct virtnet_info *vi = rvq->vdev->priv;
597 struct receive_queue *rq = &vi->rq[vq2rxq(rvq)];
598
599
600 if (napi_schedule_prep(&rq->napi)) {
601 virtqueue_disable_cb(rvq);
602 __napi_schedule(&rq->napi);
603 }
604}
605
606static void virtnet_napi_enable(struct receive_queue *rq)
607{
608 napi_enable(&rq->napi);
609
610
611
612
613
614 if (napi_schedule_prep(&rq->napi)) {
615 virtqueue_disable_cb(rq->vq);
616 local_bh_disable();
617 __napi_schedule(&rq->napi);
618 local_bh_enable();
619 }
620}
621
622static void refill_work(struct work_struct *work)
623{
624 struct virtnet_info *vi =
625 container_of(work, struct virtnet_info, refill.work);
626 bool still_empty;
627 int i;
628
629 for (i = 0; i < vi->curr_queue_pairs; i++) {
630 struct receive_queue *rq = &vi->rq[i];
631
632 napi_disable(&rq->napi);
633 still_empty = !try_fill_recv(vi, rq, GFP_KERNEL);
634 virtnet_napi_enable(rq);
635
636
637
638
639 if (still_empty)
640 schedule_delayed_work(&vi->refill, HZ/2);
641 }
642}
643
644static int virtnet_poll(struct napi_struct *napi, int budget)
645{
646 struct receive_queue *rq =
647 container_of(napi, struct receive_queue, napi);
648 struct virtnet_info *vi = rq->vq->vdev->priv;
649 void *buf;
650 unsigned int r, len, received = 0;
651
652 while (received < budget &&
653 (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) {
654 receive_buf(vi, rq, buf, len);
655 --rq->num;
656 received++;
657 }
658
659 if (rq->num < rq->max / 2) {
660 if (!try_fill_recv(vi, rq, GFP_ATOMIC))
661 schedule_delayed_work(&vi->refill, 0);
662 }
663
664
665 if (received < budget) {
666 r = virtqueue_enable_cb_prepare(rq->vq);
667 napi_complete_done(napi, received);
668 if (unlikely(virtqueue_poll(rq->vq, r)) &&
669 napi_schedule_prep(napi)) {
670 virtqueue_disable_cb(rq->vq);
671 __napi_schedule(napi);
672 }
673 }
674
675 return received;
676}
677
678static int virtnet_open(struct net_device *dev)
679{
680 struct virtnet_info *vi = netdev_priv(dev);
681 int i;
682
683 for (i = 0; i < vi->max_queue_pairs; i++) {
684 if (i < vi->curr_queue_pairs)
685
686 if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL))
687 schedule_delayed_work(&vi->refill, 0);
688 virtnet_napi_enable(&vi->rq[i]);
689 }
690
691 return 0;
692}
693
694static void free_old_xmit_skbs(struct send_queue *sq)
695{
696 struct sk_buff *skb;
697 unsigned int len;
698 struct virtnet_info *vi = sq->vq->vdev->priv;
699 struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
700
701 while ((skb = virtqueue_get_buf(sq->vq, &len)) != NULL) {
702 pr_debug("Sent skb %p\n", skb);
703
704 u64_stats_update_begin(&stats->tx_syncp);
705 stats->tx_bytes += skb->len;
706 stats->tx_packets++;
707 u64_stats_update_end(&stats->tx_syncp);
708
709 dev_kfree_skb_any(skb);
710 }
711}
712
713static int xmit_skb(struct send_queue *sq, struct sk_buff *skb)
714{
715 struct virtio_net_hdr_mrg_rxbuf *hdr;
716 const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
717 struct virtnet_info *vi = sq->vq->vdev->priv;
718 unsigned num_sg;
719 unsigned hdr_len = vi->hdr_len;
720 bool can_push;
721
722 pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
723
724 can_push = vi->any_header_sg &&
725 !((unsigned long)skb->data & (__alignof__(*hdr) - 1)) &&
726 !skb_header_cloned(skb) && skb_headroom(skb) >= hdr_len;
727
728
729 if (can_push)
730 hdr = (struct virtio_net_hdr_mrg_rxbuf *)(skb->data - hdr_len);
731 else
732 hdr = skb_vnet_hdr(skb);
733
734 if (skb->ip_summed == CHECKSUM_PARTIAL) {
735 hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
736 hdr->hdr.csum_start = cpu_to_virtio16(vi->vdev,
737 skb_checksum_start_offset(skb));
738 hdr->hdr.csum_offset = cpu_to_virtio16(vi->vdev,
739 skb->csum_offset);
740 } else {
741 hdr->hdr.flags = 0;
742 hdr->hdr.csum_offset = hdr->hdr.csum_start = 0;
743 }
744
745 if (skb_is_gso(skb)) {
746 hdr->hdr.hdr_len = cpu_to_virtio16(vi->vdev, skb_headlen(skb));
747 hdr->hdr.gso_size = cpu_to_virtio16(vi->vdev,
748 skb_shinfo(skb)->gso_size);
749 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
750 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
751 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
752 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
753 else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
754 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
755 else
756 BUG();
757 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
758 hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
759 } else {
760 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
761 hdr->hdr.gso_size = hdr->hdr.hdr_len = 0;
762 }
763
764 if (vi->mergeable_rx_bufs)
765 hdr->num_buffers = 0;
766
767 sg_init_table(sq->sg, MAX_SKB_FRAGS + 2);
768 if (can_push) {
769 __skb_push(skb, hdr_len);
770 num_sg = skb_to_sgvec(skb, sq->sg, 0, skb->len);
771
772 __skb_pull(skb, hdr_len);
773 } else {
774 sg_set_buf(sq->sg, hdr, hdr_len);
775 num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len) + 1;
776 }
777 return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC);
778}
779
780static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
781{
782 struct virtnet_info *vi = netdev_priv(dev);
783 int qnum = skb_get_queue_mapping(skb);
784 struct send_queue *sq = &vi->sq[qnum];
785 int err;
786 struct netdev_queue *txq = netdev_get_tx_queue(dev, qnum);
787 bool kick = !skb->xmit_more;
788
789
790 free_old_xmit_skbs(sq);
791
792
793 err = xmit_skb(sq, skb);
794
795
796 if (unlikely(err)) {
797 dev->stats.tx_fifo_errors++;
798 if (net_ratelimit())
799 dev_warn(&dev->dev,
800 "Unexpected TXQ (%d) queue failure: %d\n", qnum, err);
801 dev->stats.tx_dropped++;
802 kfree_skb(skb);
803 return NETDEV_TX_OK;
804 }
805
806
807 skb_orphan(skb);
808 nf_reset(skb);
809
810
811
812 if (sq->vq->num_free < 2+MAX_SKB_FRAGS) {
813 netif_stop_subqueue(dev, qnum);
814 if (unlikely(!virtqueue_enable_cb_delayed(sq->vq))) {
815
816 free_old_xmit_skbs(sq);
817 if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) {
818 netif_start_subqueue(dev, qnum);
819 virtqueue_disable_cb(sq->vq);
820 }
821 }
822 }
823
824 if (kick || netif_xmit_stopped(txq))
825 virtqueue_kick(sq->vq);
826
827 return NETDEV_TX_OK;
828}
829
830
831
832
833
834
835static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
836 struct scatterlist *out,
837 struct scatterlist *in)
838{
839 struct scatterlist *sgs[4], hdr, stat;
840 struct virtio_net_ctrl_hdr ctrl;
841 virtio_net_ctrl_ack status = ~0;
842 unsigned out_num = 0, in_num = 0, tmp;
843
844
845 BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ));
846
847 ctrl.class = class;
848 ctrl.cmd = cmd;
849
850 sg_init_one(&hdr, &ctrl, sizeof(ctrl));
851 sgs[out_num++] = &hdr;
852
853 if (out)
854 sgs[out_num++] = out;
855 if (in)
856 sgs[out_num + in_num++] = in;
857
858
859 sg_init_one(&stat, &status, sizeof(status));
860 sgs[out_num + in_num++] = &stat;
861
862 BUG_ON(out_num + in_num > ARRAY_SIZE(sgs));
863 virtqueue_add_sgs(vi->cvq, sgs, out_num, in_num, vi, GFP_ATOMIC);
864
865 virtqueue_kick(vi->cvq);
866
867
868
869
870 while (!virtqueue_get_buf(vi->cvq, &tmp) &&
871 !virtqueue_is_broken(vi->cvq))
872 cpu_relax();
873
874 return status == VIRTIO_NET_OK;
875}
876
877static int virtnet_set_mac_address(struct net_device *dev, void *p)
878{
879 struct virtnet_info *vi = netdev_priv(dev);
880 struct virtio_device *vdev = vi->vdev;
881 int ret;
882 struct sockaddr *addr = p;
883 struct scatterlist sg;
884
885 ret = eth_prepare_mac_addr_change(dev, p);
886 if (ret)
887 return ret;
888
889 if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
890 sg_init_one(&sg, addr->sa_data, dev->addr_len);
891 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
892 VIRTIO_NET_CTRL_MAC_ADDR_SET,
893 &sg, NULL)) {
894 dev_warn(&vdev->dev,
895 "Failed to set mac address by vq command.\n");
896 return -EINVAL;
897 }
898 } else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC) &&
899 !virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) {
900 unsigned int i;
901
902
903 for (i = 0; i < dev->addr_len; i++)
904 virtio_cwrite8(vdev,
905 offsetof(struct virtio_net_config, mac) +
906 i, addr->sa_data[i]);
907 }
908
909 eth_commit_mac_addr_change(dev, p);
910
911 return 0;
912}
913
914static struct rtnl_link_stats64 *virtnet_stats(struct net_device *dev,
915 struct rtnl_link_stats64 *tot)
916{
917 struct virtnet_info *vi = netdev_priv(dev);
918 int cpu;
919 unsigned int start;
920
921 for_each_possible_cpu(cpu) {
922 struct virtnet_stats *stats = per_cpu_ptr(vi->stats, cpu);
923 u64 tpackets, tbytes, rpackets, rbytes;
924
925 do {
926 start = u64_stats_fetch_begin_irq(&stats->tx_syncp);
927 tpackets = stats->tx_packets;
928 tbytes = stats->tx_bytes;
929 } while (u64_stats_fetch_retry_irq(&stats->tx_syncp, start));
930
931 do {
932 start = u64_stats_fetch_begin_irq(&stats->rx_syncp);
933 rpackets = stats->rx_packets;
934 rbytes = stats->rx_bytes;
935 } while (u64_stats_fetch_retry_irq(&stats->rx_syncp, start));
936
937 tot->rx_packets += rpackets;
938 tot->tx_packets += tpackets;
939 tot->rx_bytes += rbytes;
940 tot->tx_bytes += tbytes;
941 }
942
943 tot->tx_dropped = dev->stats.tx_dropped;
944 tot->tx_fifo_errors = dev->stats.tx_fifo_errors;
945 tot->rx_dropped = dev->stats.rx_dropped;
946 tot->rx_length_errors = dev->stats.rx_length_errors;
947 tot->rx_frame_errors = dev->stats.rx_frame_errors;
948
949 return tot;
950}
951
952#ifdef CONFIG_NET_POLL_CONTROLLER
953static void virtnet_netpoll(struct net_device *dev)
954{
955 struct virtnet_info *vi = netdev_priv(dev);
956 int i;
957
958 for (i = 0; i < vi->curr_queue_pairs; i++)
959 napi_schedule(&vi->rq[i].napi);
960}
961#endif
962
963static void virtnet_ack_link_announce(struct virtnet_info *vi)
964{
965 rtnl_lock();
966 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE,
967 VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL, NULL))
968 dev_warn(&vi->dev->dev, "Failed to ack link announce.\n");
969 rtnl_unlock();
970}
971
972static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs)
973{
974 struct scatterlist sg;
975 struct virtio_net_ctrl_mq s;
976 struct net_device *dev = vi->dev;
977
978 if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ))
979 return 0;
980
981 s.virtqueue_pairs = cpu_to_virtio16(vi->vdev, queue_pairs);
982 sg_init_one(&sg, &s, sizeof(s));
983
984 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ,
985 VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg, NULL)) {
986 dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n",
987 queue_pairs);
988 return -EINVAL;
989 } else {
990 vi->curr_queue_pairs = queue_pairs;
991
992 if (dev->flags & IFF_UP)
993 schedule_delayed_work(&vi->refill, 0);
994 }
995
996 return 0;
997}
998
999static int virtnet_close(struct net_device *dev)
1000{
1001 struct virtnet_info *vi = netdev_priv(dev);
1002 int i;
1003
1004
1005 cancel_delayed_work_sync(&vi->refill);
1006
1007 for (i = 0; i < vi->max_queue_pairs; i++)
1008 napi_disable(&vi->rq[i].napi);
1009
1010 return 0;
1011}
1012
1013static void virtnet_set_rx_mode(struct net_device *dev)
1014{
1015 struct virtnet_info *vi = netdev_priv(dev);
1016 struct scatterlist sg[2];
1017 u8 promisc, allmulti;
1018 struct virtio_net_ctrl_mac *mac_data;
1019 struct netdev_hw_addr *ha;
1020 int uc_count;
1021 int mc_count;
1022 void *buf;
1023 int i;
1024
1025
1026 if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
1027 return;
1028
1029 promisc = ((dev->flags & IFF_PROMISC) != 0);
1030 allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
1031
1032 sg_init_one(sg, &promisc, sizeof(promisc));
1033
1034 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
1035 VIRTIO_NET_CTRL_RX_PROMISC,
1036 sg, NULL))
1037 dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
1038 promisc ? "en" : "dis");
1039
1040 sg_init_one(sg, &allmulti, sizeof(allmulti));
1041
1042 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
1043 VIRTIO_NET_CTRL_RX_ALLMULTI,
1044 sg, NULL))
1045 dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
1046 allmulti ? "en" : "dis");
1047
1048 uc_count = netdev_uc_count(dev);
1049 mc_count = netdev_mc_count(dev);
1050
1051 buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
1052 (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
1053 mac_data = buf;
1054 if (!buf)
1055 return;
1056
1057 sg_init_table(sg, 2);
1058
1059
1060 mac_data->entries = cpu_to_virtio32(vi->vdev, uc_count);
1061 i = 0;
1062 netdev_for_each_uc_addr(ha, dev)
1063 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
1064
1065 sg_set_buf(&sg[0], mac_data,
1066 sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
1067
1068
1069 mac_data = (void *)&mac_data->macs[uc_count][0];
1070
1071 mac_data->entries = cpu_to_virtio32(vi->vdev, mc_count);
1072 i = 0;
1073 netdev_for_each_mc_addr(ha, dev)
1074 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
1075
1076 sg_set_buf(&sg[1], mac_data,
1077 sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
1078
1079 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
1080 VIRTIO_NET_CTRL_MAC_TABLE_SET,
1081 sg, NULL))
1082 dev_warn(&dev->dev, "Failed to set MAC fitler table.\n");
1083
1084 kfree(buf);
1085}
1086
1087static int virtnet_vlan_rx_add_vid(struct net_device *dev,
1088 __be16 proto, u16 vid)
1089{
1090 struct virtnet_info *vi = netdev_priv(dev);
1091 struct scatterlist sg;
1092
1093 sg_init_one(&sg, &vid, sizeof(vid));
1094
1095 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1096 VIRTIO_NET_CTRL_VLAN_ADD, &sg, NULL))
1097 dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
1098 return 0;
1099}
1100
1101static int virtnet_vlan_rx_kill_vid(struct net_device *dev,
1102 __be16 proto, u16 vid)
1103{
1104 struct virtnet_info *vi = netdev_priv(dev);
1105 struct scatterlist sg;
1106
1107 sg_init_one(&sg, &vid, sizeof(vid));
1108
1109 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1110 VIRTIO_NET_CTRL_VLAN_DEL, &sg, NULL))
1111 dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
1112 return 0;
1113}
1114
1115static void virtnet_clean_affinity(struct virtnet_info *vi, long hcpu)
1116{
1117 int i;
1118 int cpu;
1119
1120 if (vi->affinity_hint_set) {
1121 for (i = 0; i < vi->max_queue_pairs; i++) {
1122 virtqueue_set_affinity(vi->rq[i].vq, -1);
1123 virtqueue_set_affinity(vi->sq[i].vq, -1);
1124 }
1125
1126 vi->affinity_hint_set = false;
1127 }
1128
1129 i = 0;
1130 for_each_online_cpu(cpu) {
1131 if (cpu == hcpu) {
1132 *per_cpu_ptr(vi->vq_index, cpu) = -1;
1133 } else {
1134 *per_cpu_ptr(vi->vq_index, cpu) =
1135 ++i % vi->curr_queue_pairs;
1136 }
1137 }
1138}
1139
1140static void virtnet_set_affinity(struct virtnet_info *vi)
1141{
1142 int i;
1143 int cpu;
1144
1145
1146
1147
1148
1149 if (vi->curr_queue_pairs == 1 ||
1150 vi->max_queue_pairs != num_online_cpus()) {
1151 virtnet_clean_affinity(vi, -1);
1152 return;
1153 }
1154
1155 i = 0;
1156 for_each_online_cpu(cpu) {
1157 virtqueue_set_affinity(vi->rq[i].vq, cpu);
1158 virtqueue_set_affinity(vi->sq[i].vq, cpu);
1159 *per_cpu_ptr(vi->vq_index, cpu) = i;
1160 i++;
1161 }
1162
1163 vi->affinity_hint_set = true;
1164}
1165
1166static int virtnet_cpu_callback(struct notifier_block *nfb,
1167 unsigned long action, void *hcpu)
1168{
1169 struct virtnet_info *vi = container_of(nfb, struct virtnet_info, nb);
1170
1171 switch(action & ~CPU_TASKS_FROZEN) {
1172 case CPU_ONLINE:
1173 case CPU_DOWN_FAILED:
1174 case CPU_DEAD:
1175 virtnet_set_affinity(vi);
1176 break;
1177 case CPU_DOWN_PREPARE:
1178 virtnet_clean_affinity(vi, (long)hcpu);
1179 break;
1180 default:
1181 break;
1182 }
1183
1184 return NOTIFY_OK;
1185}
1186
1187static void virtnet_get_ringparam(struct net_device *dev,
1188 struct ethtool_ringparam *ring)
1189{
1190 struct virtnet_info *vi = netdev_priv(dev);
1191
1192 ring->rx_max_pending = virtqueue_get_vring_size(vi->rq[0].vq);
1193 ring->tx_max_pending = virtqueue_get_vring_size(vi->sq[0].vq);
1194 ring->rx_pending = ring->rx_max_pending;
1195 ring->tx_pending = ring->tx_max_pending;
1196}
1197
1198
1199static void virtnet_get_drvinfo(struct net_device *dev,
1200 struct ethtool_drvinfo *info)
1201{
1202 struct virtnet_info *vi = netdev_priv(dev);
1203 struct virtio_device *vdev = vi->vdev;
1204
1205 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
1206 strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version));
1207 strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info));
1208
1209}
1210
1211
1212static int virtnet_set_channels(struct net_device *dev,
1213 struct ethtool_channels *channels)
1214{
1215 struct virtnet_info *vi = netdev_priv(dev);
1216 u16 queue_pairs = channels->combined_count;
1217 int err;
1218
1219
1220
1221
1222 if (channels->rx_count || channels->tx_count || channels->other_count)
1223 return -EINVAL;
1224
1225 if (queue_pairs > vi->max_queue_pairs)
1226 return -EINVAL;
1227
1228 get_online_cpus();
1229 err = virtnet_set_queues(vi, queue_pairs);
1230 if (!err) {
1231 netif_set_real_num_tx_queues(dev, queue_pairs);
1232 netif_set_real_num_rx_queues(dev, queue_pairs);
1233
1234 virtnet_set_affinity(vi);
1235 }
1236 put_online_cpus();
1237
1238 return err;
1239}
1240
1241static void virtnet_get_channels(struct net_device *dev,
1242 struct ethtool_channels *channels)
1243{
1244 struct virtnet_info *vi = netdev_priv(dev);
1245
1246 channels->combined_count = vi->curr_queue_pairs;
1247 channels->max_combined = vi->max_queue_pairs;
1248 channels->max_other = 0;
1249 channels->rx_count = 0;
1250 channels->tx_count = 0;
1251 channels->other_count = 0;
1252}
1253
1254static const struct ethtool_ops virtnet_ethtool_ops = {
1255 .get_drvinfo = virtnet_get_drvinfo,
1256 .get_link = ethtool_op_get_link,
1257 .get_ringparam = virtnet_get_ringparam,
1258 .set_channels = virtnet_set_channels,
1259 .get_channels = virtnet_get_channels,
1260};
1261
1262#define MIN_MTU 68
1263#define MAX_MTU 65535
1264
1265static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
1266{
1267 struct virtnet_info *vi = netdev_priv(dev);
1268
1269 if (new_mtu < MIN_MTU || new_mtu > vi->max_mtu)
1270 return -EINVAL;
1271 dev->mtu = new_mtu;
1272 return 0;
1273}
1274
1275
1276
1277
1278static u16 virtnet_select_queue(struct net_device *dev, struct sk_buff *skb,
1279 void *accel_priv, select_queue_fallback_t fallback)
1280{
1281 int txq;
1282 struct virtnet_info *vi = netdev_priv(dev);
1283
1284 if (skb_rx_queue_recorded(skb)) {
1285 txq = skb_get_rx_queue(skb);
1286 } else {
1287 txq = *__this_cpu_ptr(vi->vq_index);
1288 if (txq == -1)
1289 txq = 0;
1290 }
1291
1292 while (unlikely(txq >= dev->real_num_tx_queues))
1293 txq -= dev->real_num_tx_queues;
1294
1295 return txq;
1296}
1297
1298static const struct net_device_ops virtnet_netdev = {
1299 .ndo_open = virtnet_open,
1300 .ndo_stop = virtnet_close,
1301 .ndo_start_xmit = start_xmit,
1302 .ndo_validate_addr = eth_validate_addr,
1303 .ndo_set_mac_address = virtnet_set_mac_address,
1304 .ndo_set_rx_mode = virtnet_set_rx_mode,
1305 .ndo_change_mtu = virtnet_change_mtu,
1306 .ndo_get_stats64 = virtnet_stats,
1307 .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
1308 .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
1309 .ndo_select_queue = virtnet_select_queue,
1310#ifdef CONFIG_NET_POLL_CONTROLLER
1311 .ndo_poll_controller = virtnet_netpoll,
1312#endif
1313 .ndo_features_check = passthru_features_check,
1314};
1315
1316static void virtnet_config_changed_work(struct work_struct *work)
1317{
1318 struct virtnet_info *vi =
1319 container_of(work, struct virtnet_info, config_work);
1320 u16 v;
1321
1322 if (virtio_cread_feature(vi->vdev, VIRTIO_NET_F_STATUS,
1323 struct virtio_net_config, status, &v) < 0)
1324 return;
1325
1326 if (v & VIRTIO_NET_S_ANNOUNCE) {
1327 netdev_notify_peers(vi->dev);
1328 virtnet_ack_link_announce(vi);
1329 }
1330
1331
1332 v &= VIRTIO_NET_S_LINK_UP;
1333
1334 if (vi->status == v)
1335 return;
1336
1337 vi->status = v;
1338
1339 if (vi->status & VIRTIO_NET_S_LINK_UP) {
1340 netif_carrier_on(vi->dev);
1341 netif_tx_wake_all_queues(vi->dev);
1342 } else {
1343 netif_carrier_off(vi->dev);
1344 netif_tx_stop_all_queues(vi->dev);
1345 }
1346}
1347
1348static void virtnet_config_changed(struct virtio_device *vdev)
1349{
1350 struct virtnet_info *vi = vdev->priv;
1351
1352 schedule_work(&vi->config_work);
1353}
1354
1355static void virtnet_free_queues(struct virtnet_info *vi)
1356{
1357 int i;
1358
1359 for (i = 0; i < vi->max_queue_pairs; i++)
1360 netif_napi_del(&vi->rq[i].napi);
1361
1362 kfree(vi->rq);
1363 kfree(vi->sq);
1364}
1365
1366static void free_receive_bufs(struct virtnet_info *vi)
1367{
1368 int i;
1369
1370 for (i = 0; i < vi->max_queue_pairs; i++) {
1371 while (vi->rq[i].pages)
1372 __free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0);
1373 }
1374}
1375
1376static void free_unused_bufs(struct virtnet_info *vi)
1377{
1378 void *buf;
1379 int i;
1380
1381 for (i = 0; i < vi->max_queue_pairs; i++) {
1382 struct virtqueue *vq = vi->sq[i].vq;
1383 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL)
1384 dev_kfree_skb(buf);
1385 }
1386
1387 for (i = 0; i < vi->max_queue_pairs; i++) {
1388 struct virtqueue *vq = vi->rq[i].vq;
1389
1390 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) {
1391 if (vi->mergeable_rx_bufs || vi->big_packets)
1392 give_pages(&vi->rq[i], buf);
1393 else
1394 dev_kfree_skb(buf);
1395 --vi->rq[i].num;
1396 }
1397 BUG_ON(vi->rq[i].num != 0);
1398 }
1399}
1400
1401static void virtnet_del_vqs(struct virtnet_info *vi)
1402{
1403 struct virtio_device *vdev = vi->vdev;
1404
1405 virtnet_clean_affinity(vi, -1);
1406
1407 vdev->config->del_vqs(vdev);
1408
1409 virtnet_free_queues(vi);
1410}
1411
1412static int virtnet_find_vqs(struct virtnet_info *vi)
1413{
1414 vq_callback_t **callbacks;
1415 struct virtqueue **vqs;
1416 int ret = -ENOMEM;
1417 int i, total_vqs;
1418 const char **names;
1419
1420
1421
1422
1423
1424 total_vqs = vi->max_queue_pairs * 2 +
1425 virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ);
1426
1427
1428 vqs = kzalloc(total_vqs * sizeof(*vqs), GFP_KERNEL);
1429 if (!vqs)
1430 goto err_vq;
1431 callbacks = kmalloc(total_vqs * sizeof(*callbacks), GFP_KERNEL);
1432 if (!callbacks)
1433 goto err_callback;
1434 names = kmalloc(total_vqs * sizeof(*names), GFP_KERNEL);
1435 if (!names)
1436 goto err_names;
1437
1438
1439 if (vi->has_cvq) {
1440 callbacks[total_vqs - 1] = NULL;
1441 names[total_vqs - 1] = "control";
1442 }
1443
1444
1445 for (i = 0; i < vi->max_queue_pairs; i++) {
1446 callbacks[rxq2vq(i)] = skb_recv_done;
1447 callbacks[txq2vq(i)] = skb_xmit_done;
1448 sprintf(vi->rq[i].name, "input.%d", i);
1449 sprintf(vi->sq[i].name, "output.%d", i);
1450 names[rxq2vq(i)] = vi->rq[i].name;
1451 names[txq2vq(i)] = vi->sq[i].name;
1452 }
1453
1454 ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs, callbacks,
1455 names);
1456 if (ret)
1457 goto err_find;
1458
1459 if (vi->has_cvq) {
1460 vi->cvq = vqs[total_vqs - 1];
1461 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
1462 vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
1463 }
1464
1465 for (i = 0; i < vi->max_queue_pairs; i++) {
1466 vi->rq[i].vq = vqs[rxq2vq(i)];
1467 vi->sq[i].vq = vqs[txq2vq(i)];
1468 }
1469
1470 kfree(names);
1471 kfree(callbacks);
1472 kfree(vqs);
1473
1474 return 0;
1475
1476err_find:
1477 kfree(names);
1478err_names:
1479 kfree(callbacks);
1480err_callback:
1481 kfree(vqs);
1482err_vq:
1483 return ret;
1484}
1485
1486static int virtnet_alloc_queues(struct virtnet_info *vi)
1487{
1488 int i;
1489
1490 vi->sq = kzalloc(sizeof(*vi->sq) * vi->max_queue_pairs, GFP_KERNEL);
1491 if (!vi->sq)
1492 goto err_sq;
1493 vi->rq = kzalloc(sizeof(*vi->rq) * vi->max_queue_pairs, GFP_KERNEL);
1494 if (!vi->rq)
1495 goto err_rq;
1496
1497 INIT_DELAYED_WORK(&vi->refill, refill_work);
1498 for (i = 0; i < vi->max_queue_pairs; i++) {
1499 vi->rq[i].pages = NULL;
1500 netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll,
1501 napi_weight);
1502
1503 sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg));
1504 sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg));
1505 }
1506
1507 return 0;
1508
1509err_rq:
1510 kfree(vi->sq);
1511err_sq:
1512 return -ENOMEM;
1513}
1514
1515static int init_vqs(struct virtnet_info *vi)
1516{
1517 int ret;
1518
1519
1520 ret = virtnet_alloc_queues(vi);
1521 if (ret)
1522 goto err;
1523
1524 ret = virtnet_find_vqs(vi);
1525 if (ret)
1526 goto err_free;
1527
1528 get_online_cpus();
1529 virtnet_set_affinity(vi);
1530 put_online_cpus();
1531
1532 return 0;
1533
1534err_free:
1535 virtnet_free_queues(vi);
1536err:
1537 return ret;
1538}
1539
1540static int virtnet_probe(struct virtio_device *vdev)
1541{
1542 int i, err;
1543 struct net_device *dev;
1544 struct virtnet_info *vi;
1545 u16 max_queue_pairs;
1546 int mtu;
1547
1548 if (!vdev->config->get) {
1549 dev_err(&vdev->dev, "%s failure: config access disabled\n",
1550 __func__);
1551 return -EINVAL;
1552 }
1553
1554
1555 err = virtio_cread_feature(vdev, VIRTIO_NET_F_MQ,
1556 struct virtio_net_config,
1557 max_virtqueue_pairs, &max_queue_pairs);
1558
1559
1560 if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
1561 max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
1562 !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
1563 max_queue_pairs = 1;
1564
1565
1566 dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs);
1567 if (!dev)
1568 return -ENOMEM;
1569
1570
1571 dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE;
1572 dev->netdev_ops = &virtnet_netdev;
1573 dev->features = NETIF_F_HIGHDMA;
1574
1575 SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
1576 SET_NETDEV_DEV(dev, &vdev->dev);
1577
1578
1579 if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
1580
1581 dev->hw_features |= NETIF_F_HW_CSUM | NETIF_F_SG;
1582 if (csum)
1583 dev->features |= NETIF_F_HW_CSUM | NETIF_F_SG;
1584
1585 if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
1586 dev->hw_features |= NETIF_F_TSO | NETIF_F_UFO
1587 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
1588 }
1589
1590 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
1591 dev->hw_features |= NETIF_F_TSO;
1592 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
1593 dev->hw_features |= NETIF_F_TSO6;
1594 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
1595 dev->hw_features |= NETIF_F_TSO_ECN;
1596 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
1597 dev->hw_features |= NETIF_F_UFO;
1598
1599 if (gso)
1600 dev->features |= dev->hw_features & (NETIF_F_ALL_TSO|NETIF_F_UFO);
1601
1602 }
1603 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM))
1604 dev->features |= NETIF_F_RXCSUM;
1605
1606 dev->vlan_features = dev->features;
1607
1608
1609 if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
1610 virtio_cread_bytes(vdev,
1611 offsetof(struct virtio_net_config, mac),
1612 dev->dev_addr, dev->addr_len);
1613 else
1614 eth_hw_addr_random(dev);
1615
1616
1617 vi = netdev_priv(dev);
1618 vi->dev = dev;
1619 vi->vdev = vdev;
1620 vdev->priv = vi;
1621 vi->stats = alloc_percpu(struct virtnet_stats);
1622 err = -ENOMEM;
1623 if (vi->stats == NULL)
1624 goto free;
1625
1626 vi->vq_index = alloc_percpu(int);
1627 if (vi->vq_index == NULL)
1628 goto free_stats;
1629
1630 INIT_WORK(&vi->config_work, virtnet_config_changed_work);
1631
1632
1633 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
1634 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
1635 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
1636 vi->big_packets = true;
1637
1638 if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
1639 vi->mergeable_rx_bufs = true;
1640
1641 if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF) ||
1642 virtio_has_feature(vdev, VIRTIO_F_VERSION_1))
1643 vi->hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1644 else
1645 vi->hdr_len = sizeof(struct virtio_net_hdr);
1646
1647 if (virtio_has_feature(vdev, VIRTIO_F_ANY_LAYOUT))
1648 vi->any_header_sg = true;
1649
1650 if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
1651 vi->has_cvq = true;
1652
1653 vi->max_mtu = MAX_MTU;
1654 if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) {
1655 mtu = virtio_cread16(vdev,
1656 offsetof(struct virtio_net_config,
1657 mtu));
1658 if (virtnet_change_mtu(dev, mtu))
1659 __virtio_clear_bit(vdev, VIRTIO_NET_F_MTU);
1660 else
1661 vi->max_mtu = mtu;
1662 }
1663
1664 if (vi->any_header_sg)
1665 dev->needed_headroom = vi->hdr_len;
1666
1667
1668 if (num_online_cpus() >= max_queue_pairs)
1669 vi->curr_queue_pairs = max_queue_pairs;
1670 else
1671 vi->curr_queue_pairs = num_online_cpus();
1672 vi->max_queue_pairs = max_queue_pairs;
1673
1674
1675 err = init_vqs(vi);
1676 if (err)
1677 goto free_index;
1678
1679 netif_set_real_num_tx_queues(dev, vi->curr_queue_pairs);
1680 netif_set_real_num_rx_queues(dev, vi->curr_queue_pairs);
1681
1682 err = register_netdev(dev);
1683 if (err) {
1684 pr_debug("virtio_net: registering device failed\n");
1685 goto free_vqs;
1686 }
1687
1688 virtio_device_ready(vdev);
1689
1690
1691 for (i = 0; i < vi->curr_queue_pairs; i++) {
1692 try_fill_recv(vi, &vi->rq[i], GFP_KERNEL);
1693
1694
1695 if (vi->rq[i].num == 0) {
1696 free_unused_bufs(vi);
1697 err = -ENOMEM;
1698 goto free_recv_bufs;
1699 }
1700 }
1701
1702 vi->nb.notifier_call = &virtnet_cpu_callback;
1703 err = register_hotcpu_notifier(&vi->nb);
1704 if (err) {
1705 pr_debug("virtio_net: registering cpu notifier failed\n");
1706 goto free_recv_bufs;
1707 }
1708
1709 rtnl_lock();
1710 virtnet_set_queues(vi, vi->curr_queue_pairs);
1711 rtnl_unlock();
1712
1713
1714
1715 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
1716 netif_carrier_off(dev);
1717 schedule_work(&vi->config_work);
1718 } else {
1719 vi->status = VIRTIO_NET_S_LINK_UP;
1720 netif_carrier_on(dev);
1721 }
1722
1723 pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
1724 dev->name, max_queue_pairs);
1725
1726 return 0;
1727
1728free_recv_bufs:
1729 vi->vdev->config->reset(vdev);
1730
1731 free_receive_bufs(vi);
1732 unregister_netdev(dev);
1733free_vqs:
1734 cancel_delayed_work_sync(&vi->refill);
1735 virtnet_del_vqs(vi);
1736free_index:
1737 free_percpu(vi->vq_index);
1738free_stats:
1739 free_percpu(vi->stats);
1740free:
1741 free_netdev(dev);
1742 return err;
1743}
1744
1745static void remove_vq_common(struct virtnet_info *vi)
1746{
1747 vi->vdev->config->reset(vi->vdev);
1748
1749
1750 free_unused_bufs(vi);
1751
1752 free_receive_bufs(vi);
1753
1754 virtnet_del_vqs(vi);
1755}
1756
1757static void virtnet_remove(struct virtio_device *vdev)
1758{
1759 struct virtnet_info *vi = vdev->priv;
1760
1761 unregister_hotcpu_notifier(&vi->nb);
1762
1763
1764 flush_work(&vi->config_work);
1765
1766 unregister_netdev(vi->dev);
1767
1768 remove_vq_common(vi);
1769
1770
1771 free_percpu(vi->vq_index);
1772 free_percpu(vi->stats);
1773 free_netdev(vi->dev);
1774}
1775
1776#ifdef CONFIG_PM_SLEEP
1777static int virtnet_freeze(struct virtio_device *vdev)
1778{
1779 struct virtnet_info *vi = vdev->priv;
1780 int i;
1781
1782 unregister_hotcpu_notifier(&vi->nb);
1783
1784
1785 flush_work(&vi->config_work);
1786
1787 netif_device_detach(vi->dev);
1788 cancel_delayed_work_sync(&vi->refill);
1789
1790 if (netif_running(vi->dev))
1791 for (i = 0; i < vi->max_queue_pairs; i++) {
1792 napi_disable(&vi->rq[i].napi);
1793 netif_napi_del(&vi->rq[i].napi);
1794 }
1795
1796 remove_vq_common(vi);
1797
1798 return 0;
1799}
1800
1801static int virtnet_restore(struct virtio_device *vdev)
1802{
1803 struct virtnet_info *vi = vdev->priv;
1804 int err, i;
1805
1806 err = init_vqs(vi);
1807 if (err)
1808 return err;
1809
1810 virtio_device_ready(vdev);
1811
1812 if (netif_running(vi->dev)) {
1813 for (i = 0; i < vi->curr_queue_pairs; i++)
1814 if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL))
1815 schedule_delayed_work(&vi->refill, 0);
1816
1817 for (i = 0; i < vi->max_queue_pairs; i++)
1818 virtnet_napi_enable(&vi->rq[i]);
1819 }
1820
1821 netif_device_attach(vi->dev);
1822
1823 rtnl_lock();
1824 virtnet_set_queues(vi, vi->curr_queue_pairs);
1825 rtnl_unlock();
1826
1827 err = register_hotcpu_notifier(&vi->nb);
1828 if (err)
1829 return err;
1830
1831 return 0;
1832}
1833#endif
1834
1835static struct virtio_device_id id_table[] = {
1836 { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
1837 { 0 },
1838};
1839
1840static unsigned int features[] = {
1841 VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
1842 VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
1843 VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
1844 VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
1845 VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO,
1846 VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ,
1847 VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN,
1848 VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ,
1849 VIRTIO_NET_F_CTRL_MAC_ADDR,
1850 VIRTIO_F_ANY_LAYOUT,
1851 VIRTIO_NET_F_MTU,
1852};
1853
1854static struct virtio_driver virtio_net_driver = {
1855 .feature_table = features,
1856 .feature_table_size = ARRAY_SIZE(features),
1857 .driver.name = KBUILD_MODNAME,
1858 .driver.owner = THIS_MODULE,
1859 .id_table = id_table,
1860 .probe = virtnet_probe,
1861 .remove = virtnet_remove,
1862 .config_changed = virtnet_config_changed,
1863#ifdef CONFIG_PM_SLEEP
1864 .freeze = virtnet_freeze,
1865 .restore = virtnet_restore,
1866#endif
1867};
1868
1869module_virtio_driver(virtio_net_driver);
1870
1871MODULE_DEVICE_TABLE(virtio, id_table);
1872MODULE_DESCRIPTION("Virtio network driver");
1873MODULE_LICENSE("GPL");
1874