]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - drivers/net/virtio_net.c
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:


fe4562d395e3f09a33d1bdfc7beb175680453732
[mirror_ubuntu-bionic-kernel.git] / drivers / net / virtio_net.c
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, see <http://www.gnu.org/licenses/>.
17 */
18 //#define DEBUG
19 #include <linux/netdevice.h>
20 #include <linux/etherdevice.h>
21 #include <linux/ethtool.h>
22 #include <linux/module.h>
23 #include <linux/virtio.h>
24 #include <linux/virtio_net.h>
25 #include <linux/bpf.h>
26 #include <linux/scatterlist.h>
27 #include <linux/if_vlan.h>
28 #include <linux/slab.h>
29 #include <linux/cpu.h>
30 #include <linux/average.h>
31 #include <net/busy_poll.h>
32
33 static int napi_weight = NAPI_POLL_WEIGHT;
34 module_param(napi_weight, int, 0444);
35
36 static bool csum = true, gso = true;
37 module_param(csum, bool, 0444);
38 module_param(gso, bool, 0444);
39
40 /* FIXME: MTU in config. */
41 #define GOOD_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
42 #define GOOD_COPY_LEN 128
43
44 /* RX packet size EWMA. The average packet size is used to determine the packet
45 * buffer size when refilling RX rings. As the entire RX ring may be refilled
46 * at once, the weight is chosen so that the EWMA will be insensitive to short-
47 * term, transient changes in packet size.
48 */
49 DECLARE_EWMA(pkt_len, 1, 64)
50
51 /* Minimum alignment for mergeable packet buffers. */
52 #define MERGEABLE_BUFFER_ALIGN max(L1_CACHE_BYTES, 256)
53
54 #define VIRTNET_DRIVER_VERSION "1.0.0"
55
56 struct virtnet_stats {
57 struct u64_stats_sync tx_syncp;
58 struct u64_stats_sync rx_syncp;
59 u64 tx_bytes;
60 u64 tx_packets;
61
62 u64 rx_bytes;
63 u64 rx_packets;
64 };
65
66 /* Internal representation of a send virtqueue */
67 struct send_queue {
68 /* Virtqueue associated with this send _queue */
69 struct virtqueue *vq;
70
71 /* TX: fragments + linear part + virtio header */
72 struct scatterlist sg[MAX_SKB_FRAGS + 2];
73
74 /* Name of the send queue: output.$index */
75 char name[40];
76 };
77
78 /* Internal representation of a receive virtqueue */
79 struct receive_queue {
80 /* Virtqueue associated with this receive_queue */
81 struct virtqueue *vq;
82
83 struct napi_struct napi;
84
85 struct bpf_prog __rcu *xdp_prog;
86
87 /* Chain pages by the private ptr. */
88 struct page *pages;
89
90 /* Average packet length for mergeable receive buffers. */
91 struct ewma_pkt_len mrg_avg_pkt_len;
92
93 /* Page frag for packet buffer allocation. */
94 struct page_frag alloc_frag;
95
96 /* RX: fragments + linear part + virtio header */
97 struct scatterlist sg[MAX_SKB_FRAGS + 2];
98
99 /* Name of this receive queue: input.$index */
100 char name[40];
101 };
102
103 struct virtnet_info {
104 struct virtio_device *vdev;
105 struct virtqueue *cvq;
106 struct net_device *dev;
107 struct send_queue *sq;
108 struct receive_queue *rq;
109 unsigned int status;
110
111 /* Max # of queue pairs supported by the device */
112 u16 max_queue_pairs;
113
114 /* # of queue pairs currently used by the driver */
115 u16 curr_queue_pairs;
116
117 /* # of XDP queue pairs currently used by the driver */
118 u16 xdp_queue_pairs;
119
120 /* I like... big packets and I cannot lie! */
121 bool big_packets;
122
123 /* Host will merge rx buffers for big packets (shake it! shake it!) */
124 bool mergeable_rx_bufs;
125
126 /* Has control virtqueue */
127 bool has_cvq;
128
129 /* Host can handle any s/g split between our header and packet data */
130 bool any_header_sg;
131
132 /* Packet virtio header size */
133 u8 hdr_len;
134
135 /* Active statistics */
136 struct virtnet_stats __percpu *stats;
137
138 /* Work struct for refilling if we run low on memory. */
139 struct delayed_work refill;
140
141 /* Work struct for config space updates */
142 struct work_struct config_work;
143
144 /* Does the affinity hint is set for virtqueues? */
145 bool affinity_hint_set;
146
147 /* CPU hotplug instances for online & dead */
148 struct hlist_node node;
149 struct hlist_node node_dead;
150
151 /* Control VQ buffers: protected by the rtnl lock */
152 struct virtio_net_ctrl_hdr ctrl_hdr;
153 virtio_net_ctrl_ack ctrl_status;
154 struct virtio_net_ctrl_mq ctrl_mq;
155 u8 ctrl_promisc;
156 u8 ctrl_allmulti;
157 u16 ctrl_vid;
158
159 /* Ethtool settings */
160 u8 duplex;
161 u32 speed;
162 };
163
164 struct padded_vnet_hdr {
165 struct virtio_net_hdr_mrg_rxbuf hdr;
166 /*
167 * hdr is in a separate sg buffer, and data sg buffer shares same page
168 * with this header sg. This padding makes next sg 16 byte aligned
169 * after the header.
170 */
171 char padding[4];
172 };
173
174 /* Converting between virtqueue no. and kernel tx/rx queue no.
175 * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq
176 */
177 static int vq2txq(struct virtqueue *vq)
178 {
179 return (vq->index - 1) / 2;
180 }
181
182 static int txq2vq(int txq)
183 {
184 return txq * 2 + 1;
185 }
186
187 static int vq2rxq(struct virtqueue *vq)
188 {
189 return vq->index / 2;
190 }
191
192 static int rxq2vq(int rxq)
193 {
194 return rxq * 2;
195 }
196
197 static inline struct virtio_net_hdr_mrg_rxbuf *skb_vnet_hdr(struct sk_buff *skb)
198 {
199 return (struct virtio_net_hdr_mrg_rxbuf *)skb->cb;
200 }
201
202 /*
203 * private is used to chain pages for big packets, put the whole
204 * most recent used list in the beginning for reuse
205 */
206 static void give_pages(struct receive_queue *rq, struct page *page)
207 {
208 struct page *end;
209
210 /* Find end of list, sew whole thing into vi->rq.pages. */
211 for (end = page; end->private; end = (struct page *)end->private);
212 end->private = (unsigned long)rq->pages;
213 rq->pages = page;
214 }
215
216 static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask)
217 {
218 struct page *p = rq->pages;
219
220 if (p) {
221 rq->pages = (struct page *)p->private;
222 /* clear private here, it is used to chain pages */
223 p->private = 0;
224 } else
225 p = alloc_page(gfp_mask);
226 return p;
227 }
228
229 static void skb_xmit_done(struct virtqueue *vq)
230 {
231 struct virtnet_info *vi = vq->vdev->priv;
232
233 /* Suppress further interrupts. */
234 virtqueue_disable_cb(vq);
235
236 /* We were probably waiting for more output buffers. */
237 netif_wake_subqueue(vi->dev, vq2txq(vq));
238 }
239
240 static unsigned int mergeable_ctx_to_buf_truesize(unsigned long mrg_ctx)
241 {
242 unsigned int truesize = mrg_ctx & (MERGEABLE_BUFFER_ALIGN - 1);
243 return (truesize + 1) * MERGEABLE_BUFFER_ALIGN;
244 }
245
246 static void *mergeable_ctx_to_buf_address(unsigned long mrg_ctx)
247 {
248 return (void *)(mrg_ctx & -MERGEABLE_BUFFER_ALIGN);
249
250 }
251
252 static unsigned long mergeable_buf_to_ctx(void *buf, unsigned int truesize)
253 {
254 unsigned int size = truesize / MERGEABLE_BUFFER_ALIGN;
255 return (unsigned long)buf | (size - 1);
256 }
257
258 /* Called from bottom half context */
259 static struct sk_buff *page_to_skb(struct virtnet_info *vi,
260 struct receive_queue *rq,
261 struct page *page, unsigned int offset,
262 unsigned int len, unsigned int truesize)
263 {
264 struct sk_buff *skb;
265 struct virtio_net_hdr_mrg_rxbuf *hdr;
266 unsigned int copy, hdr_len, hdr_padded_len;
267 char *p;
268
269 p = page_address(page) + offset;
270
271 /* copy small packet so we can reuse these pages for small data */
272 skb = napi_alloc_skb(&rq->napi, GOOD_COPY_LEN);
273 if (unlikely(!skb))
274 return NULL;
275
276 hdr = skb_vnet_hdr(skb);
277
278 hdr_len = vi->hdr_len;
279 if (vi->mergeable_rx_bufs)
280 hdr_padded_len = sizeof *hdr;
281 else
282 hdr_padded_len = sizeof(struct padded_vnet_hdr);
283
284 memcpy(hdr, p, hdr_len);
285
286 len -= hdr_len;
287 offset += hdr_padded_len;
288 p += hdr_padded_len;
289
290 copy = len;
291 if (copy > skb_tailroom(skb))
292 copy = skb_tailroom(skb);
293 memcpy(skb_put(skb, copy), p, copy);
294
295 len -= copy;
296 offset += copy;
297
298 if (vi->mergeable_rx_bufs) {
299 if (len)
300 skb_add_rx_frag(skb, 0, page, offset, len, truesize);
301 else
302 put_page(page);
303 return skb;
304 }
305
306 /*
307 * Verify that we can indeed put this data into a skb.
308 * This is here to handle cases when the device erroneously
309 * tries to receive more than is possible. This is usually
310 * the case of a broken device.
311 */
312 if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) {
313 net_dbg_ratelimited("%s: too much data\n", skb->dev->name);
314 dev_kfree_skb(skb);
315 return NULL;
316 }
317 BUG_ON(offset >= PAGE_SIZE);
318 while (len) {
319 unsigned int frag_size = min((unsigned)PAGE_SIZE - offset, len);
320 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page, offset,
321 frag_size, truesize);
322 len -= frag_size;
323 page = (struct page *)page->private;
324 offset = 0;
325 }
326
327 if (page)
328 give_pages(rq, page);
329
330 return skb;
331 }
332
333 static void virtnet_xdp_xmit(struct virtnet_info *vi,
334 struct receive_queue *rq,
335 struct send_queue *sq,
336 struct xdp_buff *xdp)
337 {
338 struct page *page = virt_to_head_page(xdp->data);
339 struct virtio_net_hdr_mrg_rxbuf *hdr;
340 unsigned int num_sg, len;
341 void *xdp_sent;
342 int err;
343
344 /* Free up any pending old buffers before queueing new ones. */
345 while ((xdp_sent = virtqueue_get_buf(sq->vq, &len)) != NULL) {
346 struct page *sent_page = virt_to_head_page(xdp_sent);
347
348 if (vi->mergeable_rx_bufs)
349 put_page(sent_page);
350 else
351 give_pages(rq, sent_page);
352 }
353
354 /* Zero header and leave csum up to XDP layers */
355 hdr = xdp->data;
356 memset(hdr, 0, vi->hdr_len);
357
358 num_sg = 1;
359 sg_init_one(sq->sg, xdp->data, xdp->data_end - xdp->data);
360 err = virtqueue_add_outbuf(sq->vq, sq->sg, num_sg,
361 xdp->data, GFP_ATOMIC);
362 if (unlikely(err)) {
363 if (vi->mergeable_rx_bufs)
364 put_page(page);
365 else
366 give_pages(rq, page);
367 return; // On error abort to avoid unnecessary kick
368 } else if (!vi->mergeable_rx_bufs) {
369 /* If not mergeable bufs must be big packets so cleanup pages */
370 give_pages(rq, (struct page *)page->private);
371 page->private = 0;
372 }
373
374 virtqueue_kick(sq->vq);
375 }
376
377 static u32 do_xdp_prog(struct virtnet_info *vi,
378 struct receive_queue *rq,
379 struct bpf_prog *xdp_prog,
380 struct page *page, int offset, int len)
381 {
382 int hdr_padded_len;
383 struct xdp_buff xdp;
384 unsigned int qp;
385 u32 act;
386 u8 *buf;
387
388 buf = page_address(page) + offset;
389
390 if (vi->mergeable_rx_bufs)
391 hdr_padded_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
392 else
393 hdr_padded_len = sizeof(struct padded_vnet_hdr);
394
395 xdp.data = buf + hdr_padded_len;
396 xdp.data_end = xdp.data + (len - vi->hdr_len);
397
398 act = bpf_prog_run_xdp(xdp_prog, &xdp);
399 switch (act) {
400 case XDP_PASS:
401 return XDP_PASS;
402 case XDP_TX:
403 qp = vi->curr_queue_pairs -
404 vi->xdp_queue_pairs +
405 smp_processor_id();
406 xdp.data = buf + (vi->mergeable_rx_bufs ? 0 : 4);
407 virtnet_xdp_xmit(vi, rq, &vi->sq[qp], &xdp);
408 return XDP_TX;
409 default:
410 bpf_warn_invalid_xdp_action(act);
411 case XDP_ABORTED:
412 case XDP_DROP:
413 return XDP_DROP;
414 }
415 }
416
417 static struct sk_buff *receive_small(struct virtnet_info *vi, void *buf, unsigned int len)
418 {
419 struct sk_buff * skb = buf;
420
421 len -= vi->hdr_len;
422 skb_trim(skb, len);
423
424 return skb;
425 }
426
427 static struct sk_buff *receive_big(struct net_device *dev,
428 struct virtnet_info *vi,
429 struct receive_queue *rq,
430 void *buf,
431 unsigned int len)
432 {
433 struct bpf_prog *xdp_prog;
434 struct page *page = buf;
435 struct sk_buff *skb;
436
437 rcu_read_lock();
438 xdp_prog = rcu_dereference(rq->xdp_prog);
439 if (xdp_prog) {
440 struct virtio_net_hdr_mrg_rxbuf *hdr = buf;
441 u32 act;
442
443 if (unlikely(hdr->hdr.gso_type || hdr->hdr.flags))
444 goto err_xdp;
445 act = do_xdp_prog(vi, rq, xdp_prog, page, 0, len);
446 switch (act) {
447 case XDP_PASS:
448 break;
449 case XDP_TX:
450 rcu_read_unlock();
451 goto xdp_xmit;
452 case XDP_DROP:
453 default:
454 goto err_xdp;
455 }
456 }
457 rcu_read_unlock();
458
459 skb = page_to_skb(vi, rq, page, 0, len, PAGE_SIZE);
460 if (unlikely(!skb))
461 goto err;
462
463 return skb;
464
465 err_xdp:
466 rcu_read_unlock();
467 err:
468 dev->stats.rx_dropped++;
469 give_pages(rq, page);
470 xdp_xmit:
471 return NULL;
472 }
473
474 /* The conditions to enable XDP should preclude the underlying device from
475 * sending packets across multiple buffers (num_buf > 1). However per spec
476 * it does not appear to be illegal to do so but rather just against convention.
477 * So in order to avoid making a system unresponsive the packets are pushed
478 * into a page and the XDP program is run. This will be extremely slow and we
479 * push a warning to the user to fix this as soon as possible. Fixing this may
480 * require resolving the underlying hardware to determine why multiple buffers
481 * are being received or simply loading the XDP program in the ingress stack
482 * after the skb is built because there is no advantage to running it here
483 * anymore.
484 */
485 static struct page *xdp_linearize_page(struct receive_queue *rq,
486 u16 num_buf,
487 struct page *p,
488 int offset,
489 unsigned int *len)
490 {
491 struct page *page = alloc_page(GFP_ATOMIC);
492 unsigned int page_off = 0;
493
494 if (!page)
495 return NULL;
496
497 memcpy(page_address(page) + page_off, page_address(p) + offset, *len);
498 page_off += *len;
499
500 while (--num_buf) {
501 unsigned int buflen;
502 unsigned long ctx;
503 void *buf;
504 int off;
505
506 ctx = (unsigned long)virtqueue_get_buf(rq->vq, &buflen);
507 if (unlikely(!ctx))
508 goto err_buf;
509
510 /* guard against a misconfigured or uncooperative backend that
511 * is sending packet larger than the MTU.
512 */
513 if ((page_off + buflen) > PAGE_SIZE)
514 goto err_buf;
515
516 buf = mergeable_ctx_to_buf_address(ctx);
517 p = virt_to_head_page(buf);
518 off = buf - page_address(p);
519
520 memcpy(page_address(page) + page_off,
521 page_address(p) + off, buflen);
522 page_off += buflen;
523 }
524
525 *len = page_off;
526 return page;
527 err_buf:
528 __free_pages(page, 0);
529 return NULL;
530 }
531
532 static struct sk_buff *receive_mergeable(struct net_device *dev,
533 struct virtnet_info *vi,
534 struct receive_queue *rq,
535 unsigned long ctx,
536 unsigned int len)
537 {
538 void *buf = mergeable_ctx_to_buf_address(ctx);
539 struct virtio_net_hdr_mrg_rxbuf *hdr = buf;
540 u16 num_buf = virtio16_to_cpu(vi->vdev, hdr->num_buffers);
541 struct page *page = virt_to_head_page(buf);
542 int offset = buf - page_address(page);
543 struct sk_buff *head_skb, *curr_skb;
544 struct bpf_prog *xdp_prog;
545 unsigned int truesize;
546
547 head_skb = NULL;
548
549 rcu_read_lock();
550 xdp_prog = rcu_dereference(rq->xdp_prog);
551 if (xdp_prog) {
552 struct page *xdp_page;
553 u32 act;
554
555 /* This happens when rx buffer size is underestimated */
556 if (unlikely(num_buf > 1)) {
557 /* linearize data for XDP */
558 xdp_page = xdp_linearize_page(rq, num_buf,
559 page, offset, &len);
560 if (!xdp_page)
561 goto err_xdp;
562 offset = 0;
563 } else {
564 xdp_page = page;
565 }
566
567 /* Transient failure which in theory could occur if
568 * in-flight packets from before XDP was enabled reach
569 * the receive path after XDP is loaded. In practice I
570 * was not able to create this condition.
571 */
572 if (unlikely(hdr->hdr.gso_type || hdr->hdr.flags))
573 goto err_xdp;
574
575 act = do_xdp_prog(vi, rq, xdp_prog, xdp_page, offset, len);
576 switch (act) {
577 case XDP_PASS:
578 if (unlikely(xdp_page != page))
579 __free_pages(xdp_page, 0);
580 break;
581 case XDP_TX:
582 if (unlikely(xdp_page != page))
583 goto err_xdp;
584 rcu_read_unlock();
585 goto xdp_xmit;
586 case XDP_DROP:
587 default:
588 if (unlikely(xdp_page != page))
589 __free_pages(xdp_page, 0);
590 goto err_xdp;
591 }
592 }
593 rcu_read_unlock();
594
595 truesize = max(len, mergeable_ctx_to_buf_truesize(ctx));
596 head_skb = page_to_skb(vi, rq, page, offset, len, truesize);
597 curr_skb = head_skb;
598
599 if (unlikely(!curr_skb))
600 goto err_skb;
601 while (--num_buf) {
602 int num_skb_frags;
603
604 ctx = (unsigned long)virtqueue_get_buf(rq->vq, &len);
605 if (unlikely(!ctx)) {
606 pr_debug("%s: rx error: %d buffers out of %d missing\n",
607 dev->name, num_buf,
608 virtio16_to_cpu(vi->vdev,
609 hdr->num_buffers));
610 dev->stats.rx_length_errors++;
611 goto err_buf;
612 }
613
614 buf = mergeable_ctx_to_buf_address(ctx);
615 page = virt_to_head_page(buf);
616
617 num_skb_frags = skb_shinfo(curr_skb)->nr_frags;
618 if (unlikely(num_skb_frags == MAX_SKB_FRAGS)) {
619 struct sk_buff *nskb = alloc_skb(0, GFP_ATOMIC);
620
621 if (unlikely(!nskb))
622 goto err_skb;
623 if (curr_skb == head_skb)
624 skb_shinfo(curr_skb)->frag_list = nskb;
625 else
626 curr_skb->next = nskb;
627 curr_skb = nskb;
628 head_skb->truesize += nskb->truesize;
629 num_skb_frags = 0;
630 }
631 truesize = max(len, mergeable_ctx_to_buf_truesize(ctx));
632 if (curr_skb != head_skb) {
633 head_skb->data_len += len;
634 head_skb->len += len;
635 head_skb->truesize += truesize;
636 }
637 offset = buf - page_address(page);
638 if (skb_can_coalesce(curr_skb, num_skb_frags, page, offset)) {
639 put_page(page);
640 skb_coalesce_rx_frag(curr_skb, num_skb_frags - 1,
641 len, truesize);
642 } else {
643 skb_add_rx_frag(curr_skb, num_skb_frags, page,
644 offset, len, truesize);
645 }
646 }
647
648 ewma_pkt_len_add(&rq->mrg_avg_pkt_len, head_skb->len);
649 return head_skb;
650
651 err_xdp:
652 rcu_read_unlock();
653 err_skb:
654 put_page(page);
655 while (--num_buf) {
656 ctx = (unsigned long)virtqueue_get_buf(rq->vq, &len);
657 if (unlikely(!ctx)) {
658 pr_debug("%s: rx error: %d buffers missing\n",
659 dev->name, num_buf);
660 dev->stats.rx_length_errors++;
661 break;
662 }
663 page = virt_to_head_page(mergeable_ctx_to_buf_address(ctx));
664 put_page(page);
665 }
666 err_buf:
667 dev->stats.rx_dropped++;
668 dev_kfree_skb(head_skb);
669 xdp_xmit:
670 return NULL;
671 }
672
673 static void receive_buf(struct virtnet_info *vi, struct receive_queue *rq,
674 void *buf, unsigned int len)
675 {
676 struct net_device *dev = vi->dev;
677 struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
678 struct sk_buff *skb;
679 struct virtio_net_hdr_mrg_rxbuf *hdr;
680
681 if (unlikely(len < vi->hdr_len + ETH_HLEN)) {
682 pr_debug("%s: short packet %i\n", dev->name, len);
683 dev->stats.rx_length_errors++;
684 if (vi->mergeable_rx_bufs) {
685 unsigned long ctx = (unsigned long)buf;
686 void *base = mergeable_ctx_to_buf_address(ctx);
687 put_page(virt_to_head_page(base));
688 } else if (vi->big_packets) {
689 give_pages(rq, buf);
690 } else {
691 dev_kfree_skb(buf);
692 }
693 return;
694 }
695
696 if (vi->mergeable_rx_bufs)
697 skb = receive_mergeable(dev, vi, rq, (unsigned long)buf, len);
698 else if (vi->big_packets)
699 skb = receive_big(dev, vi, rq, buf, len);
700 else
701 skb = receive_small(vi, buf, len);
702
703 if (unlikely(!skb))
704 return;
705
706 hdr = skb_vnet_hdr(skb);
707
708 u64_stats_update_begin(&stats->rx_syncp);
709 stats->rx_bytes += skb->len;
710 stats->rx_packets++;
711 u64_stats_update_end(&stats->rx_syncp);
712
713 if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID)
714 skb->ip_summed = CHECKSUM_UNNECESSARY;
715
716 if (virtio_net_hdr_to_skb(skb, &hdr->hdr,
717 virtio_is_little_endian(vi->vdev))) {
718 net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n",
719 dev->name, hdr->hdr.gso_type,
720 hdr->hdr.gso_size);
721 goto frame_err;
722 }
723
724 skb->protocol = eth_type_trans(skb, dev);
725 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
726 ntohs(skb->protocol), skb->len, skb->pkt_type);
727
728 napi_gro_receive(&rq->napi, skb);
729 return;
730
731 frame_err:
732 dev->stats.rx_frame_errors++;
733 dev_kfree_skb(skb);
734 }
735
736 static int add_recvbuf_small(struct virtnet_info *vi, struct receive_queue *rq,
737 gfp_t gfp)
738 {
739 struct sk_buff *skb;
740 struct virtio_net_hdr_mrg_rxbuf *hdr;
741 int err;
742
743 skb = __netdev_alloc_skb_ip_align(vi->dev, GOOD_PACKET_LEN, gfp);
744 if (unlikely(!skb))
745 return -ENOMEM;
746
747 skb_put(skb, GOOD_PACKET_LEN);
748
749 hdr = skb_vnet_hdr(skb);
750 sg_init_table(rq->sg, 2);
751 sg_set_buf(rq->sg, hdr, vi->hdr_len);
752 skb_to_sgvec(skb, rq->sg + 1, 0, skb->len);
753
754 err = virtqueue_add_inbuf(rq->vq, rq->sg, 2, skb, gfp);
755 if (err < 0)
756 dev_kfree_skb(skb);
757
758 return err;
759 }
760
761 static int add_recvbuf_big(struct virtnet_info *vi, struct receive_queue *rq,
762 gfp_t gfp)
763 {
764 struct page *first, *list = NULL;
765 char *p;
766 int i, err, offset;
767
768 sg_init_table(rq->sg, MAX_SKB_FRAGS + 2);
769
770 /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */
771 for (i = MAX_SKB_FRAGS + 1; i > 1; --i) {
772 first = get_a_page(rq, gfp);
773 if (!first) {
774 if (list)
775 give_pages(rq, list);
776 return -ENOMEM;
777 }
778 sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE);
779
780 /* chain new page in list head to match sg */
781 first->private = (unsigned long)list;
782 list = first;
783 }
784
785 first = get_a_page(rq, gfp);
786 if (!first) {
787 give_pages(rq, list);
788 return -ENOMEM;
789 }
790 p = page_address(first);
791
792 /* rq->sg[0], rq->sg[1] share the same page */
793 /* a separated rq->sg[0] for header - required in case !any_header_sg */
794 sg_set_buf(&rq->sg[0], p, vi->hdr_len);
795
796 /* rq->sg[1] for data packet, from offset */
797 offset = sizeof(struct padded_vnet_hdr);
798 sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset);
799
800 /* chain first in list head */
801 first->private = (unsigned long)list;
802 err = virtqueue_add_inbuf(rq->vq, rq->sg, MAX_SKB_FRAGS + 2,
803 first, gfp);
804 if (err < 0)
805 give_pages(rq, first);
806
807 return err;
808 }
809
810 static unsigned int get_mergeable_buf_len(struct ewma_pkt_len *avg_pkt_len)
811 {
812 const size_t hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
813 unsigned int len;
814
815 len = hdr_len + clamp_t(unsigned int, ewma_pkt_len_read(avg_pkt_len),
816 GOOD_PACKET_LEN, PAGE_SIZE - hdr_len);
817 return ALIGN(len, MERGEABLE_BUFFER_ALIGN);
818 }
819
820 static int add_recvbuf_mergeable(struct receive_queue *rq, gfp_t gfp)
821 {
822 struct page_frag *alloc_frag = &rq->alloc_frag;
823 char *buf;
824 unsigned long ctx;
825 int err;
826 unsigned int len, hole;
827
828 len = get_mergeable_buf_len(&rq->mrg_avg_pkt_len);
829 if (unlikely(!skb_page_frag_refill(len, alloc_frag, gfp)))
830 return -ENOMEM;
831
832 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
833 ctx = mergeable_buf_to_ctx(buf, len);
834 get_page(alloc_frag->page);
835 alloc_frag->offset += len;
836 hole = alloc_frag->size - alloc_frag->offset;
837 if (hole < len) {
838 /* To avoid internal fragmentation, if there is very likely not
839 * enough space for another buffer, add the remaining space to
840 * the current buffer. This extra space is not included in
841 * the truesize stored in ctx.
842 */
843 len += hole;
844 alloc_frag->offset += hole;
845 }
846
847 sg_init_one(rq->sg, buf, len);
848 err = virtqueue_add_inbuf(rq->vq, rq->sg, 1, (void *)ctx, gfp);
849 if (err < 0)
850 put_page(virt_to_head_page(buf));
851
852 return err;
853 }
854
855 /*
856 * Returns false if we couldn't fill entirely (OOM).
857 *
858 * Normally run in the receive path, but can also be run from ndo_open
859 * before we're receiving packets, or from refill_work which is
860 * careful to disable receiving (using napi_disable).
861 */
862 static bool try_fill_recv(struct virtnet_info *vi, struct receive_queue *rq,
863 gfp_t gfp)
864 {
865 int err;
866 bool oom;
867
868 gfp |= __GFP_COLD;
869 do {
870 if (vi->mergeable_rx_bufs)
871 err = add_recvbuf_mergeable(rq, gfp);
872 else if (vi->big_packets)
873 err = add_recvbuf_big(vi, rq, gfp);
874 else
875 err = add_recvbuf_small(vi, rq, gfp);
876
877 oom = err == -ENOMEM;
878 if (err)
879 break;
880 } while (rq->vq->num_free);
881 virtqueue_kick(rq->vq);
882 return !oom;
883 }
884
885 static void skb_recv_done(struct virtqueue *rvq)
886 {
887 struct virtnet_info *vi = rvq->vdev->priv;
888 struct receive_queue *rq = &vi->rq[vq2rxq(rvq)];
889
890 /* Schedule NAPI, Suppress further interrupts if successful. */
891 if (napi_schedule_prep(&rq->napi)) {
892 virtqueue_disable_cb(rvq);
893 __napi_schedule(&rq->napi);
894 }
895 }
896
897 static void virtnet_napi_enable(struct receive_queue *rq)
898 {
899 napi_enable(&rq->napi);
900
901 /* If all buffers were filled by other side before we napi_enabled, we
902 * won't get another interrupt, so process any outstanding packets
903 * now. virtnet_poll wants re-enable the queue, so we disable here.
904 * We synchronize against interrupts via NAPI_STATE_SCHED */
905 if (napi_schedule_prep(&rq->napi)) {
906 virtqueue_disable_cb(rq->vq);
907 local_bh_disable();
908 __napi_schedule(&rq->napi);
909 local_bh_enable();
910 }
911 }
912
913 static void refill_work(struct work_struct *work)
914 {
915 struct virtnet_info *vi =
916 container_of(work, struct virtnet_info, refill.work);
917 bool still_empty;
918 int i;
919
920 for (i = 0; i < vi->curr_queue_pairs; i++) {
921 struct receive_queue *rq = &vi->rq[i];
922
923 napi_disable(&rq->napi);
924 still_empty = !try_fill_recv(vi, rq, GFP_KERNEL);
925 virtnet_napi_enable(rq);
926
927 /* In theory, this can happen: if we don't get any buffers in
928 * we will *never* try to fill again.
929 */
930 if (still_empty)
931 schedule_delayed_work(&vi->refill, HZ/2);
932 }
933 }
934
935 static int virtnet_receive(struct receive_queue *rq, int budget)
936 {
937 struct virtnet_info *vi = rq->vq->vdev->priv;
938 unsigned int len, received = 0;
939 void *buf;
940
941 while (received < budget &&
942 (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) {
943 receive_buf(vi, rq, buf, len);
944 received++;
945 }
946
947 if (rq->vq->num_free > virtqueue_get_vring_size(rq->vq) / 2) {
948 if (!try_fill_recv(vi, rq, GFP_ATOMIC))
949 schedule_delayed_work(&vi->refill, 0);
950 }
951
952 return received;
953 }
954
955 static int virtnet_poll(struct napi_struct *napi, int budget)
956 {
957 struct receive_queue *rq =
958 container_of(napi, struct receive_queue, napi);
959 unsigned int r, received;
960
961 received = virtnet_receive(rq, budget);
962
963 /* Out of packets? */
964 if (received < budget) {
965 r = virtqueue_enable_cb_prepare(rq->vq);
966 napi_complete_done(napi, received);
967 if (unlikely(virtqueue_poll(rq->vq, r)) &&
968 napi_schedule_prep(napi)) {
969 virtqueue_disable_cb(rq->vq);
970 __napi_schedule(napi);
971 }
972 }
973
974 return received;
975 }
976
977 #ifdef CONFIG_NET_RX_BUSY_POLL
978 /* must be called with local_bh_disable()d */
979 static int virtnet_busy_poll(struct napi_struct *napi)
980 {
981 struct receive_queue *rq =
982 container_of(napi, struct receive_queue, napi);
983 struct virtnet_info *vi = rq->vq->vdev->priv;
984 int r, received = 0, budget = 4;
985
986 if (!(vi->status & VIRTIO_NET_S_LINK_UP))
987 return LL_FLUSH_FAILED;
988
989 if (!napi_schedule_prep(napi))
990 return LL_FLUSH_BUSY;
991
992 virtqueue_disable_cb(rq->vq);
993
994 again:
995 received += virtnet_receive(rq, budget);
996
997 r = virtqueue_enable_cb_prepare(rq->vq);
998 clear_bit(NAPI_STATE_SCHED, &napi->state);
999 if (unlikely(virtqueue_poll(rq->vq, r)) &&
1000 napi_schedule_prep(napi)) {
1001 virtqueue_disable_cb(rq->vq);
1002 if (received < budget) {
1003 budget -= received;
1004 goto again;
1005 } else {
1006 __napi_schedule(napi);
1007 }
1008 }
1009
1010 return received;
1011 }
1012 #endif /* CONFIG_NET_RX_BUSY_POLL */
1013
1014 static int virtnet_open(struct net_device *dev)
1015 {
1016 struct virtnet_info *vi = netdev_priv(dev);
1017 int i;
1018
1019 for (i = 0; i < vi->max_queue_pairs; i++) {
1020 if (i < vi->curr_queue_pairs)
1021 /* Make sure we have some buffers: if oom use wq. */
1022 if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL))
1023 schedule_delayed_work(&vi->refill, 0);
1024 virtnet_napi_enable(&vi->rq[i]);
1025 }
1026
1027 return 0;
1028 }
1029
1030 static void free_old_xmit_skbs(struct send_queue *sq)
1031 {
1032 struct sk_buff *skb;
1033 unsigned int len;
1034 struct virtnet_info *vi = sq->vq->vdev->priv;
1035 struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
1036
1037 while ((skb = virtqueue_get_buf(sq->vq, &len)) != NULL) {
1038 pr_debug("Sent skb %p\n", skb);
1039
1040 u64_stats_update_begin(&stats->tx_syncp);
1041 stats->tx_bytes += skb->len;
1042 stats->tx_packets++;
1043 u64_stats_update_end(&stats->tx_syncp);
1044
1045 dev_kfree_skb_any(skb);
1046 }
1047 }
1048
1049 static int xmit_skb(struct send_queue *sq, struct sk_buff *skb)
1050 {
1051 struct virtio_net_hdr_mrg_rxbuf *hdr;
1052 const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
1053 struct virtnet_info *vi = sq->vq->vdev->priv;
1054 unsigned num_sg;
1055 unsigned hdr_len = vi->hdr_len;
1056 bool can_push;
1057
1058 pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
1059
1060 can_push = vi->any_header_sg &&
1061 !((unsigned long)skb->data & (__alignof__(*hdr) - 1)) &&
1062 !skb_header_cloned(skb) && skb_headroom(skb) >= hdr_len;
1063 /* Even if we can, don't push here yet as this would skew
1064 * csum_start offset below. */
1065 if (can_push)
1066 hdr = (struct virtio_net_hdr_mrg_rxbuf *)(skb->data - hdr_len);
1067 else
1068 hdr = skb_vnet_hdr(skb);
1069
1070 if (virtio_net_hdr_from_skb(skb, &hdr->hdr,
1071 virtio_is_little_endian(vi->vdev)))
1072 BUG();
1073
1074 if (vi->mergeable_rx_bufs)
1075 hdr->num_buffers = 0;
1076
1077 sg_init_table(sq->sg, skb_shinfo(skb)->nr_frags + (can_push ? 1 : 2));
1078 if (can_push) {
1079 __skb_push(skb, hdr_len);
1080 num_sg = skb_to_sgvec(skb, sq->sg, 0, skb->len);
1081 /* Pull header back to avoid skew in tx bytes calculations. */
1082 __skb_pull(skb, hdr_len);
1083 } else {
1084 sg_set_buf(sq->sg, hdr, hdr_len);
1085 num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len) + 1;
1086 }
1087 return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC);
1088 }
1089
1090 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
1091 {
1092 struct virtnet_info *vi = netdev_priv(dev);
1093 int qnum = skb_get_queue_mapping(skb);
1094 struct send_queue *sq = &vi->sq[qnum];
1095 int err;
1096 struct netdev_queue *txq = netdev_get_tx_queue(dev, qnum);
1097 bool kick = !skb->xmit_more;
1098
1099 /* Free up any pending old buffers before queueing new ones. */
1100 free_old_xmit_skbs(sq);
1101
1102 /* timestamp packet in software */
1103 skb_tx_timestamp(skb);
1104
1105 /* Try to transmit */
1106 err = xmit_skb(sq, skb);
1107
1108 /* This should not happen! */
1109 if (unlikely(err)) {
1110 dev->stats.tx_fifo_errors++;
1111 if (net_ratelimit())
1112 dev_warn(&dev->dev,
1113 "Unexpected TXQ (%d) queue failure: %d\n", qnum, err);
1114 dev->stats.tx_dropped++;
1115 dev_kfree_skb_any(skb);
1116 return NETDEV_TX_OK;
1117 }
1118
1119 /* Don't wait up for transmitted skbs to be freed. */
1120 skb_orphan(skb);
1121 nf_reset(skb);
1122
1123 /* If running out of space, stop queue to avoid getting packets that we
1124 * are then unable to transmit.
1125 * An alternative would be to force queuing layer to requeue the skb by
1126 * returning NETDEV_TX_BUSY. However, NETDEV_TX_BUSY should not be
1127 * returned in a normal path of operation: it means that driver is not
1128 * maintaining the TX queue stop/start state properly, and causes
1129 * the stack to do a non-trivial amount of useless work.
1130 * Since most packets only take 1 or 2 ring slots, stopping the queue
1131 * early means 16 slots are typically wasted.
1132 */
1133 if (sq->vq->num_free < 2+MAX_SKB_FRAGS) {
1134 netif_stop_subqueue(dev, qnum);
1135 if (unlikely(!virtqueue_enable_cb_delayed(sq->vq))) {
1136 /* More just got used, free them then recheck. */
1137 free_old_xmit_skbs(sq);
1138 if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) {
1139 netif_start_subqueue(dev, qnum);
1140 virtqueue_disable_cb(sq->vq);
1141 }
1142 }
1143 }
1144
1145 if (kick || netif_xmit_stopped(txq))
1146 virtqueue_kick(sq->vq);
1147
1148 return NETDEV_TX_OK;
1149 }
1150
1151 /*
1152 * Send command via the control virtqueue and check status. Commands
1153 * supported by the hypervisor, as indicated by feature bits, should
1154 * never fail unless improperly formatted.
1155 */
1156 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
1157 struct scatterlist *out)
1158 {
1159 struct scatterlist *sgs[4], hdr, stat;
1160 unsigned out_num = 0, tmp;
1161
1162 /* Caller should know better */
1163 BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ));
1164
1165 vi->ctrl_status = ~0;
1166 vi->ctrl_hdr.class = class;
1167 vi->ctrl_hdr.cmd = cmd;
1168 /* Add header */
1169 sg_init_one(&hdr, &vi->ctrl_hdr, sizeof(vi->ctrl_hdr));
1170 sgs[out_num++] = &hdr;
1171
1172 if (out)
1173 sgs[out_num++] = out;
1174
1175 /* Add return status. */
1176 sg_init_one(&stat, &vi->ctrl_status, sizeof(vi->ctrl_status));
1177 sgs[out_num] = &stat;
1178
1179 BUG_ON(out_num + 1 > ARRAY_SIZE(sgs));
1180 virtqueue_add_sgs(vi->cvq, sgs, out_num, 1, vi, GFP_ATOMIC);
1181
1182 if (unlikely(!virtqueue_kick(vi->cvq)))
1183 return vi->ctrl_status == VIRTIO_NET_OK;
1184
1185 /* Spin for a response, the kick causes an ioport write, trapping
1186 * into the hypervisor, so the request should be handled immediately.
1187 */
1188 while (!virtqueue_get_buf(vi->cvq, &tmp) &&
1189 !virtqueue_is_broken(vi->cvq))
1190 cpu_relax();
1191
1192 return vi->ctrl_status == VIRTIO_NET_OK;
1193 }
1194
1195 static int virtnet_set_mac_address(struct net_device *dev, void *p)
1196 {
1197 struct virtnet_info *vi = netdev_priv(dev);
1198 struct virtio_device *vdev = vi->vdev;
1199 int ret;
1200 struct sockaddr *addr;
1201 struct scatterlist sg;
1202
1203 addr = kmalloc(sizeof(*addr), GFP_KERNEL);
1204 if (!addr)
1205 return -ENOMEM;
1206 memcpy(addr, p, sizeof(*addr));
1207
1208 ret = eth_prepare_mac_addr_change(dev, addr);
1209 if (ret)
1210 goto out;
1211
1212 if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1213 sg_init_one(&sg, addr->sa_data, dev->addr_len);
1214 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
1215 VIRTIO_NET_CTRL_MAC_ADDR_SET, &sg)) {
1216 dev_warn(&vdev->dev,
1217 "Failed to set mac address by vq command.\n");
1218 ret = -EINVAL;
1219 goto out;
1220 }
1221 } else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC) &&
1222 !virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) {
1223 unsigned int i;
1224
1225 /* Naturally, this has an atomicity problem. */
1226 for (i = 0; i < dev->addr_len; i++)
1227 virtio_cwrite8(vdev,
1228 offsetof(struct virtio_net_config, mac) +
1229 i, addr->sa_data[i]);
1230 }
1231
1232 eth_commit_mac_addr_change(dev, p);
1233 ret = 0;
1234
1235 out:
1236 kfree(addr);
1237 return ret;
1238 }
1239
1240 static struct rtnl_link_stats64 *virtnet_stats(struct net_device *dev,
1241 struct rtnl_link_stats64 *tot)
1242 {
1243 struct virtnet_info *vi = netdev_priv(dev);
1244 int cpu;
1245 unsigned int start;
1246
1247 for_each_possible_cpu(cpu) {
1248 struct virtnet_stats *stats = per_cpu_ptr(vi->stats, cpu);
1249 u64 tpackets, tbytes, rpackets, rbytes;
1250
1251 do {
1252 start = u64_stats_fetch_begin_irq(&stats->tx_syncp);
1253 tpackets = stats->tx_packets;
1254 tbytes = stats->tx_bytes;
1255 } while (u64_stats_fetch_retry_irq(&stats->tx_syncp, start));
1256
1257 do {
1258 start = u64_stats_fetch_begin_irq(&stats->rx_syncp);
1259 rpackets = stats->rx_packets;
1260 rbytes = stats->rx_bytes;
1261 } while (u64_stats_fetch_retry_irq(&stats->rx_syncp, start));
1262
1263 tot->rx_packets += rpackets;
1264 tot->tx_packets += tpackets;
1265 tot->rx_bytes += rbytes;
1266 tot->tx_bytes += tbytes;
1267 }
1268
1269 tot->tx_dropped = dev->stats.tx_dropped;
1270 tot->tx_fifo_errors = dev->stats.tx_fifo_errors;
1271 tot->rx_dropped = dev->stats.rx_dropped;
1272 tot->rx_length_errors = dev->stats.rx_length_errors;
1273 tot->rx_frame_errors = dev->stats.rx_frame_errors;
1274
1275 return tot;
1276 }
1277
1278 #ifdef CONFIG_NET_POLL_CONTROLLER
1279 static void virtnet_netpoll(struct net_device *dev)
1280 {
1281 struct virtnet_info *vi = netdev_priv(dev);
1282 int i;
1283
1284 for (i = 0; i < vi->curr_queue_pairs; i++)
1285 napi_schedule(&vi->rq[i].napi);
1286 }
1287 #endif
1288
1289 static void virtnet_ack_link_announce(struct virtnet_info *vi)
1290 {
1291 rtnl_lock();
1292 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE,
1293 VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL))
1294 dev_warn(&vi->dev->dev, "Failed to ack link announce.\n");
1295 rtnl_unlock();
1296 }
1297
1298 static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs)
1299 {
1300 struct scatterlist sg;
1301 struct net_device *dev = vi->dev;
1302
1303 if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ))
1304 return 0;
1305
1306 vi->ctrl_mq.virtqueue_pairs = cpu_to_virtio16(vi->vdev, queue_pairs);
1307 sg_init_one(&sg, &vi->ctrl_mq, sizeof(vi->ctrl_mq));
1308
1309 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ,
1310 VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg)) {
1311 dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n",
1312 queue_pairs);
1313 return -EINVAL;
1314 } else {
1315 vi->curr_queue_pairs = queue_pairs;
1316 /* virtnet_open() will refill when device is going to up. */
1317 if (dev->flags & IFF_UP)
1318 schedule_delayed_work(&vi->refill, 0);
1319 }
1320
1321 return 0;
1322 }
1323
1324 static int virtnet_close(struct net_device *dev)
1325 {
1326 struct virtnet_info *vi = netdev_priv(dev);
1327 int i;
1328
1329 /* Make sure refill_work doesn't re-enable napi! */
1330 cancel_delayed_work_sync(&vi->refill);
1331
1332 for (i = 0; i < vi->max_queue_pairs; i++)
1333 napi_disable(&vi->rq[i].napi);
1334
1335 return 0;
1336 }
1337
1338 static void virtnet_set_rx_mode(struct net_device *dev)
1339 {
1340 struct virtnet_info *vi = netdev_priv(dev);
1341 struct scatterlist sg[2];
1342 struct virtio_net_ctrl_mac *mac_data;
1343 struct netdev_hw_addr *ha;
1344 int uc_count;
1345 int mc_count;
1346 void *buf;
1347 int i;
1348
1349 /* We can't dynamically set ndo_set_rx_mode, so return gracefully */
1350 if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
1351 return;
1352
1353 vi->ctrl_promisc = ((dev->flags & IFF_PROMISC) != 0);
1354 vi->ctrl_allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
1355
1356 sg_init_one(sg, &vi->ctrl_promisc, sizeof(vi->ctrl_promisc));
1357
1358 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
1359 VIRTIO_NET_CTRL_RX_PROMISC, sg))
1360 dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
1361 vi->ctrl_promisc ? "en" : "dis");
1362
1363 sg_init_one(sg, &vi->ctrl_allmulti, sizeof(vi->ctrl_allmulti));
1364
1365 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
1366 VIRTIO_NET_CTRL_RX_ALLMULTI, sg))
1367 dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
1368 vi->ctrl_allmulti ? "en" : "dis");
1369
1370 uc_count = netdev_uc_count(dev);
1371 mc_count = netdev_mc_count(dev);
1372 /* MAC filter - use one buffer for both lists */
1373 buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
1374 (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
1375 mac_data = buf;
1376 if (!buf)
1377 return;
1378
1379 sg_init_table(sg, 2);
1380
1381 /* Store the unicast list and count in the front of the buffer */
1382 mac_data->entries = cpu_to_virtio32(vi->vdev, uc_count);
1383 i = 0;
1384 netdev_for_each_uc_addr(ha, dev)
1385 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
1386
1387 sg_set_buf(&sg[0], mac_data,
1388 sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
1389
1390 /* multicast list and count fill the end */
1391 mac_data = (void *)&mac_data->macs[uc_count][0];
1392
1393 mac_data->entries = cpu_to_virtio32(vi->vdev, mc_count);
1394 i = 0;
1395 netdev_for_each_mc_addr(ha, dev)
1396 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
1397
1398 sg_set_buf(&sg[1], mac_data,
1399 sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
1400
1401 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
1402 VIRTIO_NET_CTRL_MAC_TABLE_SET, sg))
1403 dev_warn(&dev->dev, "Failed to set MAC filter table.\n");
1404
1405 kfree(buf);
1406 }
1407
1408 static int virtnet_vlan_rx_add_vid(struct net_device *dev,
1409 __be16 proto, u16 vid)
1410 {
1411 struct virtnet_info *vi = netdev_priv(dev);
1412 struct scatterlist sg;
1413
1414 vi->ctrl_vid = vid;
1415 sg_init_one(&sg, &vi->ctrl_vid, sizeof(vi->ctrl_vid));
1416
1417 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1418 VIRTIO_NET_CTRL_VLAN_ADD, &sg))
1419 dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
1420 return 0;
1421 }
1422
1423 static int virtnet_vlan_rx_kill_vid(struct net_device *dev,
1424 __be16 proto, u16 vid)
1425 {
1426 struct virtnet_info *vi = netdev_priv(dev);
1427 struct scatterlist sg;
1428
1429 vi->ctrl_vid = vid;
1430 sg_init_one(&sg, &vi->ctrl_vid, sizeof(vi->ctrl_vid));
1431
1432 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1433 VIRTIO_NET_CTRL_VLAN_DEL, &sg))
1434 dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
1435 return 0;
1436 }
1437
1438 static void virtnet_clean_affinity(struct virtnet_info *vi, long hcpu)
1439 {
1440 int i;
1441
1442 if (vi->affinity_hint_set) {
1443 for (i = 0; i < vi->max_queue_pairs; i++) {
1444 virtqueue_set_affinity(vi->rq[i].vq, -1);
1445 virtqueue_set_affinity(vi->sq[i].vq, -1);
1446 }
1447
1448 vi->affinity_hint_set = false;
1449 }
1450 }
1451
1452 static void virtnet_set_affinity(struct virtnet_info *vi)
1453 {
1454 int i;
1455 int cpu;
1456
1457 /* In multiqueue mode, when the number of cpu is equal to the number of
1458 * queue pairs, we let the queue pairs to be private to one cpu by
1459 * setting the affinity hint to eliminate the contention.
1460 */
1461 if (vi->curr_queue_pairs == 1 ||
1462 vi->max_queue_pairs != num_online_cpus()) {
1463 virtnet_clean_affinity(vi, -1);
1464 return;
1465 }
1466
1467 i = 0;
1468 for_each_online_cpu(cpu) {
1469 virtqueue_set_affinity(vi->rq[i].vq, cpu);
1470 virtqueue_set_affinity(vi->sq[i].vq, cpu);
1471 netif_set_xps_queue(vi->dev, cpumask_of(cpu), i);
1472 i++;
1473 }
1474
1475 vi->affinity_hint_set = true;
1476 }
1477
1478 static int virtnet_cpu_online(unsigned int cpu, struct hlist_node *node)
1479 {
1480 struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
1481 node);
1482 virtnet_set_affinity(vi);
1483 return 0;
1484 }
1485
1486 static int virtnet_cpu_dead(unsigned int cpu, struct hlist_node *node)
1487 {
1488 struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
1489 node_dead);
1490 virtnet_set_affinity(vi);
1491 return 0;
1492 }
1493
1494 static int virtnet_cpu_down_prep(unsigned int cpu, struct hlist_node *node)
1495 {
1496 struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
1497 node);
1498
1499 virtnet_clean_affinity(vi, cpu);
1500 return 0;
1501 }
1502
1503 static enum cpuhp_state virtionet_online;
1504
1505 static int virtnet_cpu_notif_add(struct virtnet_info *vi)
1506 {
1507 int ret;
1508
1509 ret = cpuhp_state_add_instance_nocalls(virtionet_online, &vi->node);
1510 if (ret)
1511 return ret;
1512 ret = cpuhp_state_add_instance_nocalls(CPUHP_VIRT_NET_DEAD,
1513 &vi->node_dead);
1514 if (!ret)
1515 return ret;
1516 cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node);
1517 return ret;
1518 }
1519
1520 static void virtnet_cpu_notif_remove(struct virtnet_info *vi)
1521 {
1522 cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node);
1523 cpuhp_state_remove_instance_nocalls(CPUHP_VIRT_NET_DEAD,
1524 &vi->node_dead);
1525 }
1526
1527 static void virtnet_get_ringparam(struct net_device *dev,
1528 struct ethtool_ringparam *ring)
1529 {
1530 struct virtnet_info *vi = netdev_priv(dev);
1531
1532 ring->rx_max_pending = virtqueue_get_vring_size(vi->rq[0].vq);
1533 ring->tx_max_pending = virtqueue_get_vring_size(vi->sq[0].vq);
1534 ring->rx_pending = ring->rx_max_pending;
1535 ring->tx_pending = ring->tx_max_pending;
1536 }
1537
1538
1539 static void virtnet_get_drvinfo(struct net_device *dev,
1540 struct ethtool_drvinfo *info)
1541 {
1542 struct virtnet_info *vi = netdev_priv(dev);
1543 struct virtio_device *vdev = vi->vdev;
1544
1545 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
1546 strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version));
1547 strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info));
1548
1549 }
1550
1551 /* TODO: Eliminate OOO packets during switching */
1552 static int virtnet_set_channels(struct net_device *dev,
1553 struct ethtool_channels *channels)
1554 {
1555 struct virtnet_info *vi = netdev_priv(dev);
1556 u16 queue_pairs = channels->combined_count;
1557 int err;
1558
1559 /* We don't support separate rx/tx channels.
1560 * We don't allow setting 'other' channels.
1561 */
1562 if (channels->rx_count || channels->tx_count || channels->other_count)
1563 return -EINVAL;
1564
1565 if (queue_pairs > vi->max_queue_pairs || queue_pairs == 0)
1566 return -EINVAL;
1567
1568 /* For now we don't support modifying channels while XDP is loaded
1569 * also when XDP is loaded all RX queues have XDP programs so we only
1570 * need to check a single RX queue.
1571 */
1572 if (vi->rq[0].xdp_prog)
1573 return -EINVAL;
1574
1575 get_online_cpus();
1576 err = virtnet_set_queues(vi, queue_pairs);
1577 if (!err) {
1578 netif_set_real_num_tx_queues(dev, queue_pairs);
1579 netif_set_real_num_rx_queues(dev, queue_pairs);
1580
1581 virtnet_set_affinity(vi);
1582 }
1583 put_online_cpus();
1584
1585 return err;
1586 }
1587
1588 static void virtnet_get_channels(struct net_device *dev,
1589 struct ethtool_channels *channels)
1590 {
1591 struct virtnet_info *vi = netdev_priv(dev);
1592
1593 channels->combined_count = vi->curr_queue_pairs;
1594 channels->max_combined = vi->max_queue_pairs;
1595 channels->max_other = 0;
1596 channels->rx_count = 0;
1597 channels->tx_count = 0;
1598 channels->other_count = 0;
1599 }
1600
1601 /* Check if the user is trying to change anything besides speed/duplex */
1602 static bool virtnet_validate_ethtool_cmd(const struct ethtool_cmd *cmd)
1603 {
1604 struct ethtool_cmd diff1 = *cmd;
1605 struct ethtool_cmd diff2 = {};
1606
1607 /* cmd is always set so we need to clear it, validate the port type
1608 * and also without autonegotiation we can ignore advertising
1609 */
1610 ethtool_cmd_speed_set(&diff1, 0);
1611 diff2.port = PORT_OTHER;
1612 diff1.advertising = 0;
1613 diff1.duplex = 0;
1614 diff1.cmd = 0;
1615
1616 return !memcmp(&diff1, &diff2, sizeof(diff1));
1617 }
1618
1619 static int virtnet_set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1620 {
1621 struct virtnet_info *vi = netdev_priv(dev);
1622 u32 speed;
1623
1624 speed = ethtool_cmd_speed(cmd);
1625 /* don't allow custom speed and duplex */
1626 if (!ethtool_validate_speed(speed) ||
1627 !ethtool_validate_duplex(cmd->duplex) ||
1628 !virtnet_validate_ethtool_cmd(cmd))
1629 return -EINVAL;
1630 vi->speed = speed;
1631 vi->duplex = cmd->duplex;
1632
1633 return 0;
1634 }
1635
1636 static int virtnet_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1637 {
1638 struct virtnet_info *vi = netdev_priv(dev);
1639
1640 ethtool_cmd_speed_set(cmd, vi->speed);
1641 cmd->duplex = vi->duplex;
1642 cmd->port = PORT_OTHER;
1643
1644 return 0;
1645 }
1646
1647 static void virtnet_init_settings(struct net_device *dev)
1648 {
1649 struct virtnet_info *vi = netdev_priv(dev);
1650
1651 vi->speed = SPEED_UNKNOWN;
1652 vi->duplex = DUPLEX_UNKNOWN;
1653 }
1654
1655 static const struct ethtool_ops virtnet_ethtool_ops = {
1656 .get_drvinfo = virtnet_get_drvinfo,
1657 .get_link = ethtool_op_get_link,
1658 .get_ringparam = virtnet_get_ringparam,
1659 .set_channels = virtnet_set_channels,
1660 .get_channels = virtnet_get_channels,
1661 .get_ts_info = ethtool_op_get_ts_info,
1662 .get_settings = virtnet_get_settings,
1663 .set_settings = virtnet_set_settings,
1664 };
1665
1666 static int virtnet_xdp_set(struct net_device *dev, struct bpf_prog *prog)
1667 {
1668 unsigned long int max_sz = PAGE_SIZE - sizeof(struct padded_vnet_hdr);
1669 struct virtnet_info *vi = netdev_priv(dev);
1670 struct bpf_prog *old_prog;
1671 u16 xdp_qp = 0, curr_qp;
1672 int i, err;
1673
1674 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO4) ||
1675 virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO6)) {
1676 netdev_warn(dev, "can't set XDP while host is implementing LRO, disable LRO first\n");
1677 return -EOPNOTSUPP;
1678 }
1679
1680 if (vi->mergeable_rx_bufs && !vi->any_header_sg) {
1681 netdev_warn(dev, "XDP expects header/data in single page, any_header_sg required\n");
1682 return -EINVAL;
1683 }
1684
1685 if (dev->mtu > max_sz) {
1686 netdev_warn(dev, "XDP requires MTU less than %lu\n", max_sz);
1687 return -EINVAL;
1688 }
1689
1690 curr_qp = vi->curr_queue_pairs - vi->xdp_queue_pairs;
1691 if (prog)
1692 xdp_qp = nr_cpu_ids;
1693
1694 /* XDP requires extra queues for XDP_TX */
1695 if (curr_qp + xdp_qp > vi->max_queue_pairs) {
1696 netdev_warn(dev, "request %i queues but max is %i\n",
1697 curr_qp + xdp_qp, vi->max_queue_pairs);
1698 return -ENOMEM;
1699 }
1700
1701 err = virtnet_set_queues(vi, curr_qp + xdp_qp);
1702 if (err) {
1703 dev_warn(&dev->dev, "XDP Device queue allocation failure.\n");
1704 return err;
1705 }
1706
1707 if (prog) {
1708 prog = bpf_prog_add(prog, vi->max_queue_pairs - 1);
1709 if (IS_ERR(prog)) {
1710 virtnet_set_queues(vi, curr_qp);
1711 return PTR_ERR(prog);
1712 }
1713 }
1714
1715 vi->xdp_queue_pairs = xdp_qp;
1716 netif_set_real_num_rx_queues(dev, curr_qp + xdp_qp);
1717
1718 for (i = 0; i < vi->max_queue_pairs; i++) {
1719 old_prog = rtnl_dereference(vi->rq[i].xdp_prog);
1720 rcu_assign_pointer(vi->rq[i].xdp_prog, prog);
1721 if (old_prog)
1722 bpf_prog_put(old_prog);
1723 }
1724
1725 return 0;
1726 }
1727
1728 static bool virtnet_xdp_query(struct net_device *dev)
1729 {
1730 struct virtnet_info *vi = netdev_priv(dev);
1731 int i;
1732
1733 for (i = 0; i < vi->max_queue_pairs; i++) {
1734 if (vi->rq[i].xdp_prog)
1735 return true;
1736 }
1737 return false;
1738 }
1739
1740 static int virtnet_xdp(struct net_device *dev, struct netdev_xdp *xdp)
1741 {
1742 switch (xdp->command) {
1743 case XDP_SETUP_PROG:
1744 return virtnet_xdp_set(dev, xdp->prog);
1745 case XDP_QUERY_PROG:
1746 xdp->prog_attached = virtnet_xdp_query(dev);
1747 return 0;
1748 default:
1749 return -EINVAL;
1750 }
1751 }
1752
1753 static const struct net_device_ops virtnet_netdev = {
1754 .ndo_open = virtnet_open,
1755 .ndo_stop = virtnet_close,
1756 .ndo_start_xmit = start_xmit,
1757 .ndo_validate_addr = eth_validate_addr,
1758 .ndo_set_mac_address = virtnet_set_mac_address,
1759 .ndo_set_rx_mode = virtnet_set_rx_mode,
1760 .ndo_get_stats64 = virtnet_stats,
1761 .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
1762 .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
1763 #ifdef CONFIG_NET_POLL_CONTROLLER
1764 .ndo_poll_controller = virtnet_netpoll,
1765 #endif
1766 #ifdef CONFIG_NET_RX_BUSY_POLL
1767 .ndo_busy_poll = virtnet_busy_poll,
1768 #endif
1769 .ndo_xdp = virtnet_xdp,
1770 };
1771
1772 static void virtnet_config_changed_work(struct work_struct *work)
1773 {
1774 struct virtnet_info *vi =
1775 container_of(work, struct virtnet_info, config_work);
1776 u16 v;
1777
1778 if (virtio_cread_feature(vi->vdev, VIRTIO_NET_F_STATUS,
1779 struct virtio_net_config, status, &v) < 0)
1780 return;
1781
1782 if (v & VIRTIO_NET_S_ANNOUNCE) {
1783 netdev_notify_peers(vi->dev);
1784 virtnet_ack_link_announce(vi);
1785 }
1786
1787 /* Ignore unknown (future) status bits */
1788 v &= VIRTIO_NET_S_LINK_UP;
1789
1790 if (vi->status == v)
1791 return;
1792
1793 vi->status = v;
1794
1795 if (vi->status & VIRTIO_NET_S_LINK_UP) {
1796 netif_carrier_on(vi->dev);
1797 netif_tx_wake_all_queues(vi->dev);
1798 } else {
1799 netif_carrier_off(vi->dev);
1800 netif_tx_stop_all_queues(vi->dev);
1801 }
1802 }
1803
1804 static void virtnet_config_changed(struct virtio_device *vdev)
1805 {
1806 struct virtnet_info *vi = vdev->priv;
1807
1808 schedule_work(&vi->config_work);
1809 }
1810
1811 static void virtnet_free_queues(struct virtnet_info *vi)
1812 {
1813 int i;
1814
1815 for (i = 0; i < vi->max_queue_pairs; i++) {
1816 napi_hash_del(&vi->rq[i].napi);
1817 netif_napi_del(&vi->rq[i].napi);
1818 }
1819
1820 /* We called napi_hash_del() before netif_napi_del(),
1821 * we need to respect an RCU grace period before freeing vi->rq
1822 */
1823 synchronize_net();
1824
1825 kfree(vi->rq);
1826 kfree(vi->sq);
1827 }
1828
1829 static void free_receive_bufs(struct virtnet_info *vi)
1830 {
1831 struct bpf_prog *old_prog;
1832 int i;
1833
1834 rtnl_lock();
1835 for (i = 0; i < vi->max_queue_pairs; i++) {
1836 while (vi->rq[i].pages)
1837 __free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0);
1838
1839 old_prog = rtnl_dereference(vi->rq[i].xdp_prog);
1840 RCU_INIT_POINTER(vi->rq[i].xdp_prog, NULL);
1841 if (old_prog)
1842 bpf_prog_put(old_prog);
1843 }
1844 rtnl_unlock();
1845 }
1846
1847 static void free_receive_page_frags(struct virtnet_info *vi)
1848 {
1849 int i;
1850 for (i = 0; i < vi->max_queue_pairs; i++)
1851 if (vi->rq[i].alloc_frag.page)
1852 put_page(vi->rq[i].alloc_frag.page);
1853 }
1854
1855 static bool is_xdp_queue(struct virtnet_info *vi, int q)
1856 {
1857 if (q < (vi->curr_queue_pairs - vi->xdp_queue_pairs))
1858 return false;
1859 else if (q < vi->curr_queue_pairs)
1860 return true;
1861 else
1862 return false;
1863 }
1864
1865 static void free_unused_bufs(struct virtnet_info *vi)
1866 {
1867 void *buf;
1868 int i;
1869
1870 for (i = 0; i < vi->max_queue_pairs; i++) {
1871 struct virtqueue *vq = vi->sq[i].vq;
1872 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) {
1873 if (!is_xdp_queue(vi, i))
1874 dev_kfree_skb(buf);
1875 else
1876 put_page(virt_to_head_page(buf));
1877 }
1878 }
1879
1880 for (i = 0; i < vi->max_queue_pairs; i++) {
1881 struct virtqueue *vq = vi->rq[i].vq;
1882
1883 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) {
1884 if (vi->mergeable_rx_bufs) {
1885 unsigned long ctx = (unsigned long)buf;
1886 void *base = mergeable_ctx_to_buf_address(ctx);
1887 put_page(virt_to_head_page(base));
1888 } else if (vi->big_packets) {
1889 give_pages(&vi->rq[i], buf);
1890 } else {
1891 dev_kfree_skb(buf);
1892 }
1893 }
1894 }
1895 }
1896
1897 static void virtnet_del_vqs(struct virtnet_info *vi)
1898 {
1899 struct virtio_device *vdev = vi->vdev;
1900
1901 virtnet_clean_affinity(vi, -1);
1902
1903 vdev->config->del_vqs(vdev);
1904
1905 virtnet_free_queues(vi);
1906 }
1907
1908 static int virtnet_find_vqs(struct virtnet_info *vi)
1909 {
1910 vq_callback_t **callbacks;
1911 struct virtqueue **vqs;
1912 int ret = -ENOMEM;
1913 int i, total_vqs;
1914 const char **names;
1915
1916 /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
1917 * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
1918 * possible control vq.
1919 */
1920 total_vqs = vi->max_queue_pairs * 2 +
1921 virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ);
1922
1923 /* Allocate space for find_vqs parameters */
1924 vqs = kzalloc(total_vqs * sizeof(*vqs), GFP_KERNEL);
1925 if (!vqs)
1926 goto err_vq;
1927 callbacks = kmalloc(total_vqs * sizeof(*callbacks), GFP_KERNEL);
1928 if (!callbacks)
1929 goto err_callback;
1930 names = kmalloc(total_vqs * sizeof(*names), GFP_KERNEL);
1931 if (!names)
1932 goto err_names;
1933
1934 /* Parameters for control virtqueue, if any */
1935 if (vi->has_cvq) {
1936 callbacks[total_vqs - 1] = NULL;
1937 names[total_vqs - 1] = "control";
1938 }
1939
1940 /* Allocate/initialize parameters for send/receive virtqueues */
1941 for (i = 0; i < vi->max_queue_pairs; i++) {
1942 callbacks[rxq2vq(i)] = skb_recv_done;
1943 callbacks[txq2vq(i)] = skb_xmit_done;
1944 sprintf(vi->rq[i].name, "input.%d", i);
1945 sprintf(vi->sq[i].name, "output.%d", i);
1946 names[rxq2vq(i)] = vi->rq[i].name;
1947 names[txq2vq(i)] = vi->sq[i].name;
1948 }
1949
1950 ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs, callbacks,
1951 names);
1952 if (ret)
1953 goto err_find;
1954
1955 if (vi->has_cvq) {
1956 vi->cvq = vqs[total_vqs - 1];
1957 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
1958 vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
1959 }
1960
1961 for (i = 0; i < vi->max_queue_pairs; i++) {
1962 vi->rq[i].vq = vqs[rxq2vq(i)];
1963 vi->sq[i].vq = vqs[txq2vq(i)];
1964 }
1965
1966 kfree(names);
1967 kfree(callbacks);
1968 kfree(vqs);
1969
1970 return 0;
1971
1972 err_find:
1973 kfree(names);
1974 err_names:
1975 kfree(callbacks);
1976 err_callback:
1977 kfree(vqs);
1978 err_vq:
1979 return ret;
1980 }
1981
1982 static int virtnet_alloc_queues(struct virtnet_info *vi)
1983 {
1984 int i;
1985
1986 vi->sq = kzalloc(sizeof(*vi->sq) * vi->max_queue_pairs, GFP_KERNEL);
1987 if (!vi->sq)
1988 goto err_sq;
1989 vi->rq = kzalloc(sizeof(*vi->rq) * vi->max_queue_pairs, GFP_KERNEL);
1990 if (!vi->rq)
1991 goto err_rq;
1992
1993 INIT_DELAYED_WORK(&vi->refill, refill_work);
1994 for (i = 0; i < vi->max_queue_pairs; i++) {
1995 vi->rq[i].pages = NULL;
1996 netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll,
1997 napi_weight);
1998
1999 sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg));
2000 ewma_pkt_len_init(&vi->rq[i].mrg_avg_pkt_len);
2001 sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg));
2002 }
2003
2004 return 0;
2005
2006 err_rq:
2007 kfree(vi->sq);
2008 err_sq:
2009 return -ENOMEM;
2010 }
2011
2012 static int init_vqs(struct virtnet_info *vi)
2013 {
2014 int ret;
2015
2016 /* Allocate send & receive queues */
2017 ret = virtnet_alloc_queues(vi);
2018 if (ret)
2019 goto err;
2020
2021 ret = virtnet_find_vqs(vi);
2022 if (ret)
2023 goto err_free;
2024
2025 get_online_cpus();
2026 virtnet_set_affinity(vi);
2027 put_online_cpus();
2028
2029 return 0;
2030
2031 err_free:
2032 virtnet_free_queues(vi);
2033 err:
2034 return ret;
2035 }
2036
2037 #ifdef CONFIG_SYSFS
2038 static ssize_t mergeable_rx_buffer_size_show(struct netdev_rx_queue *queue,
2039 struct rx_queue_attribute *attribute, char *buf)
2040 {
2041 struct virtnet_info *vi = netdev_priv(queue->dev);
2042 unsigned int queue_index = get_netdev_rx_queue_index(queue);
2043 struct ewma_pkt_len *avg;
2044
2045 BUG_ON(queue_index >= vi->max_queue_pairs);
2046 avg = &vi->rq[queue_index].mrg_avg_pkt_len;
2047 return sprintf(buf, "%u\n", get_mergeable_buf_len(avg));
2048 }
2049
2050 static struct rx_queue_attribute mergeable_rx_buffer_size_attribute =
2051 __ATTR_RO(mergeable_rx_buffer_size);
2052
2053 static struct attribute *virtio_net_mrg_rx_attrs[] = {
2054 &mergeable_rx_buffer_size_attribute.attr,
2055 NULL
2056 };
2057
2058 static const struct attribute_group virtio_net_mrg_rx_group = {
2059 .name = "virtio_net",
2060 .attrs = virtio_net_mrg_rx_attrs
2061 };
2062 #endif
2063
2064 static bool virtnet_fail_on_feature(struct virtio_device *vdev,
2065 unsigned int fbit,
2066 const char *fname, const char *dname)
2067 {
2068 if (!virtio_has_feature(vdev, fbit))
2069 return false;
2070
2071 dev_err(&vdev->dev, "device advertises feature %s but not %s",
2072 fname, dname);
2073
2074 return true;
2075 }
2076
2077 #define VIRTNET_FAIL_ON(vdev, fbit, dbit) \
2078 virtnet_fail_on_feature(vdev, fbit, #fbit, dbit)
2079
2080 static bool virtnet_validate_features(struct virtio_device *vdev)
2081 {
2082 if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) &&
2083 (VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_RX,
2084 "VIRTIO_NET_F_CTRL_VQ") ||
2085 VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_VLAN,
2086 "VIRTIO_NET_F_CTRL_VQ") ||
2087 VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE,
2088 "VIRTIO_NET_F_CTRL_VQ") ||
2089 VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_MQ, "VIRTIO_NET_F_CTRL_VQ") ||
2090 VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR,
2091 "VIRTIO_NET_F_CTRL_VQ"))) {
2092 return false;
2093 }
2094
2095 return true;
2096 }
2097
2098 #define MIN_MTU ETH_MIN_MTU
2099 #define MAX_MTU ETH_MAX_MTU
2100
2101 static int virtnet_probe(struct virtio_device *vdev)
2102 {
2103 int i, err;
2104 struct net_device *dev;
2105 struct virtnet_info *vi;
2106 u16 max_queue_pairs;
2107 int mtu;
2108
2109 if (!vdev->config->get) {
2110 dev_err(&vdev->dev, "%s failure: config access disabled\n",
2111 __func__);
2112 return -EINVAL;
2113 }
2114
2115 if (!virtnet_validate_features(vdev))
2116 return -EINVAL;
2117
2118 /* Find if host supports multiqueue virtio_net device */
2119 err = virtio_cread_feature(vdev, VIRTIO_NET_F_MQ,
2120 struct virtio_net_config,
2121 max_virtqueue_pairs, &max_queue_pairs);
2122
2123 /* We need at least 2 queue's */
2124 if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
2125 max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
2126 !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
2127 max_queue_pairs = 1;
2128
2129 /* Allocate ourselves a network device with room for our info */
2130 dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs);
2131 if (!dev)
2132 return -ENOMEM;
2133
2134 /* Set up network device as normal. */
2135 dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE;
2136 dev->netdev_ops = &virtnet_netdev;
2137 dev->features = NETIF_F_HIGHDMA;
2138
2139 dev->ethtool_ops = &virtnet_ethtool_ops;
2140 SET_NETDEV_DEV(dev, &vdev->dev);
2141
2142 /* Do we support "hardware" checksums? */
2143 if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
2144 /* This opens up the world of extra features. */
2145 dev->hw_features |= NETIF_F_HW_CSUM | NETIF_F_SG;
2146 if (csum)
2147 dev->features |= NETIF_F_HW_CSUM | NETIF_F_SG;
2148
2149 if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
2150 dev->hw_features |= NETIF_F_TSO | NETIF_F_UFO
2151 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
2152 }
2153 /* Individual feature bits: what can host handle? */
2154 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
2155 dev->hw_features |= NETIF_F_TSO;
2156 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
2157 dev->hw_features |= NETIF_F_TSO6;
2158 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
2159 dev->hw_features |= NETIF_F_TSO_ECN;
2160 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
2161 dev->hw_features |= NETIF_F_UFO;
2162
2163 dev->features |= NETIF_F_GSO_ROBUST;
2164
2165 if (gso)
2166 dev->features |= dev->hw_features & (NETIF_F_ALL_TSO|NETIF_F_UFO);
2167 /* (!csum && gso) case will be fixed by register_netdev() */
2168 }
2169 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM))
2170 dev->features |= NETIF_F_RXCSUM;
2171
2172 dev->vlan_features = dev->features;
2173
2174 /* MTU range: 68 - 65535 */
2175 dev->min_mtu = MIN_MTU;
2176 dev->max_mtu = MAX_MTU;
2177
2178 /* Configuration may specify what MAC to use. Otherwise random. */
2179 if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
2180 virtio_cread_bytes(vdev,
2181 offsetof(struct virtio_net_config, mac),
2182 dev->dev_addr, dev->addr_len);
2183 else
2184 eth_hw_addr_random(dev);
2185
2186 /* Set up our device-specific information */
2187 vi = netdev_priv(dev);
2188 vi->dev = dev;
2189 vi->vdev = vdev;
2190 vdev->priv = vi;
2191 vi->stats = alloc_percpu(struct virtnet_stats);
2192 err = -ENOMEM;
2193 if (vi->stats == NULL)
2194 goto free;
2195
2196 for_each_possible_cpu(i) {
2197 struct virtnet_stats *virtnet_stats;
2198 virtnet_stats = per_cpu_ptr(vi->stats, i);
2199 u64_stats_init(&virtnet_stats->tx_syncp);
2200 u64_stats_init(&virtnet_stats->rx_syncp);
2201 }
2202
2203 INIT_WORK(&vi->config_work, virtnet_config_changed_work);
2204
2205 /* If we can receive ANY GSO packets, we must allocate large ones. */
2206 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
2207 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
2208 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN) ||
2209 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_UFO))
2210 vi->big_packets = true;
2211
2212 if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
2213 vi->mergeable_rx_bufs = true;
2214
2215 if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF) ||
2216 virtio_has_feature(vdev, VIRTIO_F_VERSION_1))
2217 vi->hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
2218 else
2219 vi->hdr_len = sizeof(struct virtio_net_hdr);
2220
2221 if (virtio_has_feature(vdev, VIRTIO_F_ANY_LAYOUT) ||
2222 virtio_has_feature(vdev, VIRTIO_F_VERSION_1))
2223 vi->any_header_sg = true;
2224
2225 if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
2226 vi->has_cvq = true;
2227
2228 if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) {
2229 mtu = virtio_cread16(vdev,
2230 offsetof(struct virtio_net_config,
2231 mtu));
2232 if (mtu < dev->min_mtu) {
2233 __virtio_clear_bit(vdev, VIRTIO_NET_F_MTU);
2234 } else {
2235 dev->mtu = mtu;
2236 dev->max_mtu = mtu;
2237 }
2238 }
2239
2240 if (vi->any_header_sg)
2241 dev->needed_headroom = vi->hdr_len;
2242
2243 /* Enable multiqueue by default */
2244 if (num_online_cpus() >= max_queue_pairs)
2245 vi->curr_queue_pairs = max_queue_pairs;
2246 else
2247 vi->curr_queue_pairs = num_online_cpus();
2248 vi->max_queue_pairs = max_queue_pairs;
2249
2250 /* Allocate/initialize the rx/tx queues, and invoke find_vqs */
2251 err = init_vqs(vi);
2252 if (err)
2253 goto free_stats;
2254
2255 #ifdef CONFIG_SYSFS
2256 if (vi->mergeable_rx_bufs)
2257 dev->sysfs_rx_queue_group = &virtio_net_mrg_rx_group;
2258 #endif
2259 netif_set_real_num_tx_queues(dev, vi->curr_queue_pairs);
2260 netif_set_real_num_rx_queues(dev, vi->curr_queue_pairs);
2261
2262 virtnet_init_settings(dev);
2263
2264 err = register_netdev(dev);
2265 if (err) {
2266 pr_debug("virtio_net: registering device failed\n");
2267 goto free_vqs;
2268 }
2269
2270 virtio_device_ready(vdev);
2271
2272 err = virtnet_cpu_notif_add(vi);
2273 if (err) {
2274 pr_debug("virtio_net: registering cpu notifier failed\n");
2275 goto free_unregister_netdev;
2276 }
2277
2278 rtnl_lock();
2279 virtnet_set_queues(vi, vi->curr_queue_pairs);
2280 rtnl_unlock();
2281
2282 /* Assume link up if device can't report link status,
2283 otherwise get link status from config. */
2284 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
2285 netif_carrier_off(dev);
2286 schedule_work(&vi->config_work);
2287 } else {
2288 vi->status = VIRTIO_NET_S_LINK_UP;
2289 netif_carrier_on(dev);
2290 }
2291
2292 pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
2293 dev->name, max_queue_pairs);
2294
2295 return 0;
2296
2297 free_unregister_netdev:
2298 vi->vdev->config->reset(vdev);
2299
2300 unregister_netdev(dev);
2301 free_vqs:
2302 cancel_delayed_work_sync(&vi->refill);
2303 free_receive_page_frags(vi);
2304 virtnet_del_vqs(vi);
2305 free_stats:
2306 free_percpu(vi->stats);
2307 free:
2308 free_netdev(dev);
2309 return err;
2310 }
2311
2312 static void remove_vq_common(struct virtnet_info *vi)
2313 {
2314 vi->vdev->config->reset(vi->vdev);
2315
2316 /* Free unused buffers in both send and recv, if any. */
2317 free_unused_bufs(vi);
2318
2319 free_receive_bufs(vi);
2320
2321 free_receive_page_frags(vi);
2322
2323 virtnet_del_vqs(vi);
2324 }
2325
2326 static void virtnet_remove(struct virtio_device *vdev)
2327 {
2328 struct virtnet_info *vi = vdev->priv;
2329
2330 virtnet_cpu_notif_remove(vi);
2331
2332 /* Make sure no work handler is accessing the device. */
2333 flush_work(&vi->config_work);
2334
2335 unregister_netdev(vi->dev);
2336
2337 remove_vq_common(vi);
2338
2339 free_percpu(vi->stats);
2340 free_netdev(vi->dev);
2341 }
2342
2343 #ifdef CONFIG_PM_SLEEP
2344 static int virtnet_freeze(struct virtio_device *vdev)
2345 {
2346 struct virtnet_info *vi = vdev->priv;
2347 int i;
2348
2349 virtnet_cpu_notif_remove(vi);
2350
2351 /* Make sure no work handler is accessing the device */
2352 flush_work(&vi->config_work);
2353
2354 netif_device_detach(vi->dev);
2355 cancel_delayed_work_sync(&vi->refill);
2356
2357 if (netif_running(vi->dev)) {
2358 for (i = 0; i < vi->max_queue_pairs; i++)
2359 napi_disable(&vi->rq[i].napi);
2360 }
2361
2362 remove_vq_common(vi);
2363
2364 return 0;
2365 }
2366
2367 static int virtnet_restore(struct virtio_device *vdev)
2368 {
2369 struct virtnet_info *vi = vdev->priv;
2370 int err, i;
2371
2372 err = init_vqs(vi);
2373 if (err)
2374 return err;
2375
2376 virtio_device_ready(vdev);
2377
2378 if (netif_running(vi->dev)) {
2379 for (i = 0; i < vi->curr_queue_pairs; i++)
2380 if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL))
2381 schedule_delayed_work(&vi->refill, 0);
2382
2383 for (i = 0; i < vi->max_queue_pairs; i++)
2384 virtnet_napi_enable(&vi->rq[i]);
2385 }
2386
2387 netif_device_attach(vi->dev);
2388
2389 rtnl_lock();
2390 virtnet_set_queues(vi, vi->curr_queue_pairs);
2391 rtnl_unlock();
2392
2393 err = virtnet_cpu_notif_add(vi);
2394 if (err)
2395 return err;
2396
2397 return 0;
2398 }
2399 #endif
2400
2401 static struct virtio_device_id id_table[] = {
2402 { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
2403 { 0 },
2404 };
2405
2406 #define VIRTNET_FEATURES \
2407 VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM, \
2408 VIRTIO_NET_F_MAC, \
2409 VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6, \
2410 VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, \
2411 VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO, \
2412 VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ, \
2413 VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, \
2414 VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ, \
2415 VIRTIO_NET_F_CTRL_MAC_ADDR, \
2416 VIRTIO_NET_F_MTU
2417
2418 static unsigned int features[] = {
2419 VIRTNET_FEATURES,
2420 };
2421
2422 static unsigned int features_legacy[] = {
2423 VIRTNET_FEATURES,
2424 VIRTIO_NET_F_GSO,
2425 VIRTIO_F_ANY_LAYOUT,
2426 };
2427
2428 static struct virtio_driver virtio_net_driver = {
2429 .feature_table = features,
2430 .feature_table_size = ARRAY_SIZE(features),
2431 .feature_table_legacy = features_legacy,
2432 .feature_table_size_legacy = ARRAY_SIZE(features_legacy),
2433 .driver.name = KBUILD_MODNAME,
2434 .driver.owner = THIS_MODULE,
2435 .id_table = id_table,
2436 .probe = virtnet_probe,
2437 .remove = virtnet_remove,
2438 .config_changed = virtnet_config_changed,
2439 #ifdef CONFIG_PM_SLEEP
2440 .freeze = virtnet_freeze,
2441 .restore = virtnet_restore,
2442 #endif
2443 };
2444
2445 static __init int virtio_net_driver_init(void)
2446 {
2447 int ret;
2448
2449 ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "AP_VIRT_NET_ONLINE",
2450 virtnet_cpu_online,
2451 virtnet_cpu_down_prep);
2452 if (ret < 0)
2453 goto out;
2454 virtionet_online = ret;
2455 ret = cpuhp_setup_state_multi(CPUHP_VIRT_NET_DEAD, "VIRT_NET_DEAD",
2456 NULL, virtnet_cpu_dead);
2457 if (ret)
2458 goto err_dead;
2459
2460 ret = register_virtio_driver(&virtio_net_driver);
2461 if (ret)
2462 goto err_virtio;
2463 return 0;
2464 err_virtio:
2465 cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD);
2466 err_dead:
2467 cpuhp_remove_multi_state(virtionet_online);
2468 out:
2469 return ret;
2470 }
2471 module_init(virtio_net_driver_init);
2472
2473 static __exit void virtio_net_driver_exit(void)
2474 {
2475 cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD);
2476 cpuhp_remove_multi_state(virtionet_online);
2477 unregister_virtio_driver(&virtio_net_driver);
2478 }
2479 module_exit(virtio_net_driver_exit);
2480
2481 MODULE_DEVICE_TABLE(virtio, id_table);
2482 MODULE_DESCRIPTION("Virtio network driver");
2483 MODULE_LICENSE("GPL");
ubuntu-bionic-kernel mirror