xref: /openbmc/linux/drivers/net/virtio_net.c (revision 6774def6)
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/scatterlist.h>
26 #include <linux/if_vlan.h>
27 #include <linux/slab.h>
28 #include <linux/cpu.h>
29 #include <linux/average.h>
30 #include <net/busy_poll.h>
31 
32 static int napi_weight = NAPI_POLL_WEIGHT;
33 module_param(napi_weight, int, 0444);
34 
35 static bool csum = true, gso = true;
36 module_param(csum, bool, 0444);
37 module_param(gso, bool, 0444);
38 
39 /* FIXME: MTU in config. */
40 #define GOOD_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
41 #define GOOD_COPY_LEN	128
42 
43 /* Weight used for the RX packet size EWMA. The average packet size is used to
44  * determine the packet buffer size when refilling RX rings. As the entire RX
45  * ring may be refilled at once, the weight is chosen so that the EWMA will be
46  * insensitive to short-term, transient changes in packet size.
47  */
48 #define RECEIVE_AVG_WEIGHT 64
49 
50 /* Minimum alignment for mergeable packet buffers. */
51 #define MERGEABLE_BUFFER_ALIGN max(L1_CACHE_BYTES, 256)
52 
53 #define VIRTNET_DRIVER_VERSION "1.0.0"
54 
55 struct virtnet_stats {
56 	struct u64_stats_sync tx_syncp;
57 	struct u64_stats_sync rx_syncp;
58 	u64 tx_bytes;
59 	u64 tx_packets;
60 
61 	u64 rx_bytes;
62 	u64 rx_packets;
63 };
64 
65 /* Internal representation of a send virtqueue */
66 struct send_queue {
67 	/* Virtqueue associated with this send _queue */
68 	struct virtqueue *vq;
69 
70 	/* TX: fragments + linear part + virtio header */
71 	struct scatterlist sg[MAX_SKB_FRAGS + 2];
72 
73 	/* Name of the send queue: output.$index */
74 	char name[40];
75 };
76 
77 /* Internal representation of a receive virtqueue */
78 struct receive_queue {
79 	/* Virtqueue associated with this receive_queue */
80 	struct virtqueue *vq;
81 
82 	struct napi_struct napi;
83 
84 	/* Chain pages by the private ptr. */
85 	struct page *pages;
86 
87 	/* Average packet length for mergeable receive buffers. */
88 	struct ewma mrg_avg_pkt_len;
89 
90 	/* Page frag for packet buffer allocation. */
91 	struct page_frag alloc_frag;
92 
93 	/* RX: fragments + linear part + virtio header */
94 	struct scatterlist sg[MAX_SKB_FRAGS + 2];
95 
96 	/* Name of this receive queue: input.$index */
97 	char name[40];
98 };
99 
100 struct virtnet_info {
101 	struct virtio_device *vdev;
102 	struct virtqueue *cvq;
103 	struct net_device *dev;
104 	struct send_queue *sq;
105 	struct receive_queue *rq;
106 	unsigned int status;
107 
108 	/* Max # of queue pairs supported by the device */
109 	u16 max_queue_pairs;
110 
111 	/* # of queue pairs currently used by the driver */
112 	u16 curr_queue_pairs;
113 
114 	/* I like... big packets and I cannot lie! */
115 	bool big_packets;
116 
117 	/* Host will merge rx buffers for big packets (shake it! shake it!) */
118 	bool mergeable_rx_bufs;
119 
120 	/* Has control virtqueue */
121 	bool has_cvq;
122 
123 	/* Host can handle any s/g split between our header and packet data */
124 	bool any_header_sg;
125 
126 	/* Active statistics */
127 	struct virtnet_stats __percpu *stats;
128 
129 	/* Work struct for refilling if we run low on memory. */
130 	struct delayed_work refill;
131 
132 	/* Work struct for config space updates */
133 	struct work_struct config_work;
134 
135 	/* Does the affinity hint is set for virtqueues? */
136 	bool affinity_hint_set;
137 
138 	/* CPU hot plug notifier */
139 	struct notifier_block nb;
140 };
141 
142 struct skb_vnet_hdr {
143 	union {
144 		struct virtio_net_hdr hdr;
145 		struct virtio_net_hdr_mrg_rxbuf mhdr;
146 	};
147 };
148 
149 struct padded_vnet_hdr {
150 	struct virtio_net_hdr hdr;
151 	/*
152 	 * virtio_net_hdr should be in a separated sg buffer because of a
153 	 * QEMU bug, and data sg buffer shares same page with this header sg.
154 	 * This padding makes next sg 16 byte aligned after virtio_net_hdr.
155 	 */
156 	char padding[6];
157 };
158 
159 /* Converting between virtqueue no. and kernel tx/rx queue no.
160  * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq
161  */
162 static int vq2txq(struct virtqueue *vq)
163 {
164 	return (vq->index - 1) / 2;
165 }
166 
167 static int txq2vq(int txq)
168 {
169 	return txq * 2 + 1;
170 }
171 
172 static int vq2rxq(struct virtqueue *vq)
173 {
174 	return vq->index / 2;
175 }
176 
177 static int rxq2vq(int rxq)
178 {
179 	return rxq * 2;
180 }
181 
182 static inline struct skb_vnet_hdr *skb_vnet_hdr(struct sk_buff *skb)
183 {
184 	return (struct skb_vnet_hdr *)skb->cb;
185 }
186 
187 /*
188  * private is used to chain pages for big packets, put the whole
189  * most recent used list in the beginning for reuse
190  */
191 static void give_pages(struct receive_queue *rq, struct page *page)
192 {
193 	struct page *end;
194 
195 	/* Find end of list, sew whole thing into vi->rq.pages. */
196 	for (end = page; end->private; end = (struct page *)end->private);
197 	end->private = (unsigned long)rq->pages;
198 	rq->pages = page;
199 }
200 
201 static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask)
202 {
203 	struct page *p = rq->pages;
204 
205 	if (p) {
206 		rq->pages = (struct page *)p->private;
207 		/* clear private here, it is used to chain pages */
208 		p->private = 0;
209 	} else
210 		p = alloc_page(gfp_mask);
211 	return p;
212 }
213 
214 static void skb_xmit_done(struct virtqueue *vq)
215 {
216 	struct virtnet_info *vi = vq->vdev->priv;
217 
218 	/* Suppress further interrupts. */
219 	virtqueue_disable_cb(vq);
220 
221 	/* We were probably waiting for more output buffers. */
222 	netif_wake_subqueue(vi->dev, vq2txq(vq));
223 }
224 
225 static unsigned int mergeable_ctx_to_buf_truesize(unsigned long mrg_ctx)
226 {
227 	unsigned int truesize = mrg_ctx & (MERGEABLE_BUFFER_ALIGN - 1);
228 	return (truesize + 1) * MERGEABLE_BUFFER_ALIGN;
229 }
230 
231 static void *mergeable_ctx_to_buf_address(unsigned long mrg_ctx)
232 {
233 	return (void *)(mrg_ctx & -MERGEABLE_BUFFER_ALIGN);
234 
235 }
236 
237 static unsigned long mergeable_buf_to_ctx(void *buf, unsigned int truesize)
238 {
239 	unsigned int size = truesize / MERGEABLE_BUFFER_ALIGN;
240 	return (unsigned long)buf | (size - 1);
241 }
242 
243 /* Called from bottom half context */
244 static struct sk_buff *page_to_skb(struct receive_queue *rq,
245 				   struct page *page, unsigned int offset,
246 				   unsigned int len, unsigned int truesize)
247 {
248 	struct virtnet_info *vi = rq->vq->vdev->priv;
249 	struct sk_buff *skb;
250 	struct skb_vnet_hdr *hdr;
251 	unsigned int copy, hdr_len, hdr_padded_len;
252 	char *p;
253 
254 	p = page_address(page) + offset;
255 
256 	/* copy small packet so we can reuse these pages for small data */
257 	skb = netdev_alloc_skb_ip_align(vi->dev, GOOD_COPY_LEN);
258 	if (unlikely(!skb))
259 		return NULL;
260 
261 	hdr = skb_vnet_hdr(skb);
262 
263 	if (vi->mergeable_rx_bufs) {
264 		hdr_len = sizeof hdr->mhdr;
265 		hdr_padded_len = sizeof hdr->mhdr;
266 	} else {
267 		hdr_len = sizeof hdr->hdr;
268 		hdr_padded_len = sizeof(struct padded_vnet_hdr);
269 	}
270 
271 	memcpy(hdr, p, hdr_len);
272 
273 	len -= hdr_len;
274 	offset += hdr_padded_len;
275 	p += hdr_padded_len;
276 
277 	copy = len;
278 	if (copy > skb_tailroom(skb))
279 		copy = skb_tailroom(skb);
280 	memcpy(skb_put(skb, copy), p, copy);
281 
282 	len -= copy;
283 	offset += copy;
284 
285 	if (vi->mergeable_rx_bufs) {
286 		if (len)
287 			skb_add_rx_frag(skb, 0, page, offset, len, truesize);
288 		else
289 			put_page(page);
290 		return skb;
291 	}
292 
293 	/*
294 	 * Verify that we can indeed put this data into a skb.
295 	 * This is here to handle cases when the device erroneously
296 	 * tries to receive more than is possible. This is usually
297 	 * the case of a broken device.
298 	 */
299 	if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) {
300 		net_dbg_ratelimited("%s: too much data\n", skb->dev->name);
301 		dev_kfree_skb(skb);
302 		return NULL;
303 	}
304 	BUG_ON(offset >= PAGE_SIZE);
305 	while (len) {
306 		unsigned int frag_size = min((unsigned)PAGE_SIZE - offset, len);
307 		skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page, offset,
308 				frag_size, truesize);
309 		len -= frag_size;
310 		page = (struct page *)page->private;
311 		offset = 0;
312 	}
313 
314 	if (page)
315 		give_pages(rq, page);
316 
317 	return skb;
318 }
319 
320 static struct sk_buff *receive_small(void *buf, unsigned int len)
321 {
322 	struct sk_buff * skb = buf;
323 
324 	len -= sizeof(struct virtio_net_hdr);
325 	skb_trim(skb, len);
326 
327 	return skb;
328 }
329 
330 static struct sk_buff *receive_big(struct net_device *dev,
331 				   struct receive_queue *rq,
332 				   void *buf,
333 				   unsigned int len)
334 {
335 	struct page *page = buf;
336 	struct sk_buff *skb = page_to_skb(rq, page, 0, len, PAGE_SIZE);
337 
338 	if (unlikely(!skb))
339 		goto err;
340 
341 	return skb;
342 
343 err:
344 	dev->stats.rx_dropped++;
345 	give_pages(rq, page);
346 	return NULL;
347 }
348 
349 static struct sk_buff *receive_mergeable(struct net_device *dev,
350 					 struct receive_queue *rq,
351 					 unsigned long ctx,
352 					 unsigned int len)
353 {
354 	void *buf = mergeable_ctx_to_buf_address(ctx);
355 	struct skb_vnet_hdr *hdr = buf;
356 	int num_buf = hdr->mhdr.num_buffers;
357 	struct page *page = virt_to_head_page(buf);
358 	int offset = buf - page_address(page);
359 	unsigned int truesize = max(len, mergeable_ctx_to_buf_truesize(ctx));
360 
361 	struct sk_buff *head_skb = page_to_skb(rq, page, offset, len, truesize);
362 	struct sk_buff *curr_skb = head_skb;
363 
364 	if (unlikely(!curr_skb))
365 		goto err_skb;
366 	while (--num_buf) {
367 		int num_skb_frags;
368 
369 		ctx = (unsigned long)virtqueue_get_buf(rq->vq, &len);
370 		if (unlikely(!ctx)) {
371 			pr_debug("%s: rx error: %d buffers out of %d missing\n",
372 				 dev->name, num_buf, hdr->mhdr.num_buffers);
373 			dev->stats.rx_length_errors++;
374 			goto err_buf;
375 		}
376 
377 		buf = mergeable_ctx_to_buf_address(ctx);
378 		page = virt_to_head_page(buf);
379 
380 		num_skb_frags = skb_shinfo(curr_skb)->nr_frags;
381 		if (unlikely(num_skb_frags == MAX_SKB_FRAGS)) {
382 			struct sk_buff *nskb = alloc_skb(0, GFP_ATOMIC);
383 
384 			if (unlikely(!nskb))
385 				goto err_skb;
386 			if (curr_skb == head_skb)
387 				skb_shinfo(curr_skb)->frag_list = nskb;
388 			else
389 				curr_skb->next = nskb;
390 			curr_skb = nskb;
391 			head_skb->truesize += nskb->truesize;
392 			num_skb_frags = 0;
393 		}
394 		truesize = max(len, mergeable_ctx_to_buf_truesize(ctx));
395 		if (curr_skb != head_skb) {
396 			head_skb->data_len += len;
397 			head_skb->len += len;
398 			head_skb->truesize += truesize;
399 		}
400 		offset = buf - page_address(page);
401 		if (skb_can_coalesce(curr_skb, num_skb_frags, page, offset)) {
402 			put_page(page);
403 			skb_coalesce_rx_frag(curr_skb, num_skb_frags - 1,
404 					     len, truesize);
405 		} else {
406 			skb_add_rx_frag(curr_skb, num_skb_frags, page,
407 					offset, len, truesize);
408 		}
409 	}
410 
411 	ewma_add(&rq->mrg_avg_pkt_len, head_skb->len);
412 	return head_skb;
413 
414 err_skb:
415 	put_page(page);
416 	while (--num_buf) {
417 		ctx = (unsigned long)virtqueue_get_buf(rq->vq, &len);
418 		if (unlikely(!ctx)) {
419 			pr_debug("%s: rx error: %d buffers missing\n",
420 				 dev->name, num_buf);
421 			dev->stats.rx_length_errors++;
422 			break;
423 		}
424 		page = virt_to_head_page(mergeable_ctx_to_buf_address(ctx));
425 		put_page(page);
426 	}
427 err_buf:
428 	dev->stats.rx_dropped++;
429 	dev_kfree_skb(head_skb);
430 	return NULL;
431 }
432 
433 static void receive_buf(struct receive_queue *rq, void *buf, unsigned int len)
434 {
435 	struct virtnet_info *vi = rq->vq->vdev->priv;
436 	struct net_device *dev = vi->dev;
437 	struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
438 	struct sk_buff *skb;
439 	struct skb_vnet_hdr *hdr;
440 
441 	if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
442 		pr_debug("%s: short packet %i\n", dev->name, len);
443 		dev->stats.rx_length_errors++;
444 		if (vi->mergeable_rx_bufs) {
445 			unsigned long ctx = (unsigned long)buf;
446 			void *base = mergeable_ctx_to_buf_address(ctx);
447 			put_page(virt_to_head_page(base));
448 		} else if (vi->big_packets) {
449 			give_pages(rq, buf);
450 		} else {
451 			dev_kfree_skb(buf);
452 		}
453 		return;
454 	}
455 
456 	if (vi->mergeable_rx_bufs)
457 		skb = receive_mergeable(dev, rq, (unsigned long)buf, len);
458 	else if (vi->big_packets)
459 		skb = receive_big(dev, rq, buf, len);
460 	else
461 		skb = receive_small(buf, len);
462 
463 	if (unlikely(!skb))
464 		return;
465 
466 	hdr = skb_vnet_hdr(skb);
467 
468 	u64_stats_update_begin(&stats->rx_syncp);
469 	stats->rx_bytes += skb->len;
470 	stats->rx_packets++;
471 	u64_stats_update_end(&stats->rx_syncp);
472 
473 	if (hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
474 		pr_debug("Needs csum!\n");
475 		if (!skb_partial_csum_set(skb,
476 					  hdr->hdr.csum_start,
477 					  hdr->hdr.csum_offset))
478 			goto frame_err;
479 	} else if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) {
480 		skb->ip_summed = CHECKSUM_UNNECESSARY;
481 	}
482 
483 	skb->protocol = eth_type_trans(skb, dev);
484 	pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
485 		 ntohs(skb->protocol), skb->len, skb->pkt_type);
486 
487 	if (hdr->hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
488 		pr_debug("GSO!\n");
489 		switch (hdr->hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
490 		case VIRTIO_NET_HDR_GSO_TCPV4:
491 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
492 			break;
493 		case VIRTIO_NET_HDR_GSO_UDP:
494 		{
495 			static bool warned;
496 
497 			if (!warned) {
498 				warned = true;
499 				netdev_warn(dev,
500 					    "host using disabled UFO feature; please fix it\n");
501 			}
502 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
503 			break;
504 		}
505 		case VIRTIO_NET_HDR_GSO_TCPV6:
506 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
507 			break;
508 		default:
509 			net_warn_ratelimited("%s: bad gso type %u.\n",
510 					     dev->name, hdr->hdr.gso_type);
511 			goto frame_err;
512 		}
513 
514 		if (hdr->hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
515 			skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
516 
517 		skb_shinfo(skb)->gso_size = hdr->hdr.gso_size;
518 		if (skb_shinfo(skb)->gso_size == 0) {
519 			net_warn_ratelimited("%s: zero gso size.\n", dev->name);
520 			goto frame_err;
521 		}
522 
523 		/* Header must be checked, and gso_segs computed. */
524 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
525 		skb_shinfo(skb)->gso_segs = 0;
526 	}
527 
528 	skb_mark_napi_id(skb, &rq->napi);
529 
530 	netif_receive_skb(skb);
531 	return;
532 
533 frame_err:
534 	dev->stats.rx_frame_errors++;
535 	dev_kfree_skb(skb);
536 }
537 
538 static int add_recvbuf_small(struct receive_queue *rq, gfp_t gfp)
539 {
540 	struct virtnet_info *vi = rq->vq->vdev->priv;
541 	struct sk_buff *skb;
542 	struct skb_vnet_hdr *hdr;
543 	int err;
544 
545 	skb = __netdev_alloc_skb_ip_align(vi->dev, GOOD_PACKET_LEN, gfp);
546 	if (unlikely(!skb))
547 		return -ENOMEM;
548 
549 	skb_put(skb, GOOD_PACKET_LEN);
550 
551 	hdr = skb_vnet_hdr(skb);
552 	sg_init_table(rq->sg, MAX_SKB_FRAGS + 2);
553 	sg_set_buf(rq->sg, &hdr->hdr, sizeof hdr->hdr);
554 	skb_to_sgvec(skb, rq->sg + 1, 0, skb->len);
555 
556 	err = virtqueue_add_inbuf(rq->vq, rq->sg, 2, skb, gfp);
557 	if (err < 0)
558 		dev_kfree_skb(skb);
559 
560 	return err;
561 }
562 
563 static int add_recvbuf_big(struct receive_queue *rq, gfp_t gfp)
564 {
565 	struct page *first, *list = NULL;
566 	char *p;
567 	int i, err, offset;
568 
569 	sg_init_table(rq->sg, MAX_SKB_FRAGS + 2);
570 
571 	/* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */
572 	for (i = MAX_SKB_FRAGS + 1; i > 1; --i) {
573 		first = get_a_page(rq, gfp);
574 		if (!first) {
575 			if (list)
576 				give_pages(rq, list);
577 			return -ENOMEM;
578 		}
579 		sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE);
580 
581 		/* chain new page in list head to match sg */
582 		first->private = (unsigned long)list;
583 		list = first;
584 	}
585 
586 	first = get_a_page(rq, gfp);
587 	if (!first) {
588 		give_pages(rq, list);
589 		return -ENOMEM;
590 	}
591 	p = page_address(first);
592 
593 	/* rq->sg[0], rq->sg[1] share the same page */
594 	/* a separated rq->sg[0] for virtio_net_hdr only due to QEMU bug */
595 	sg_set_buf(&rq->sg[0], p, sizeof(struct virtio_net_hdr));
596 
597 	/* rq->sg[1] for data packet, from offset */
598 	offset = sizeof(struct padded_vnet_hdr);
599 	sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset);
600 
601 	/* chain first in list head */
602 	first->private = (unsigned long)list;
603 	err = virtqueue_add_inbuf(rq->vq, rq->sg, MAX_SKB_FRAGS + 2,
604 				  first, gfp);
605 	if (err < 0)
606 		give_pages(rq, first);
607 
608 	return err;
609 }
610 
611 static unsigned int get_mergeable_buf_len(struct ewma *avg_pkt_len)
612 {
613 	const size_t hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
614 	unsigned int len;
615 
616 	len = hdr_len + clamp_t(unsigned int, ewma_read(avg_pkt_len),
617 			GOOD_PACKET_LEN, PAGE_SIZE - hdr_len);
618 	return ALIGN(len, MERGEABLE_BUFFER_ALIGN);
619 }
620 
621 static int add_recvbuf_mergeable(struct receive_queue *rq, gfp_t gfp)
622 {
623 	struct page_frag *alloc_frag = &rq->alloc_frag;
624 	char *buf;
625 	unsigned long ctx;
626 	int err;
627 	unsigned int len, hole;
628 
629 	len = get_mergeable_buf_len(&rq->mrg_avg_pkt_len);
630 	if (unlikely(!skb_page_frag_refill(len, alloc_frag, gfp)))
631 		return -ENOMEM;
632 
633 	buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
634 	ctx = mergeable_buf_to_ctx(buf, len);
635 	get_page(alloc_frag->page);
636 	alloc_frag->offset += len;
637 	hole = alloc_frag->size - alloc_frag->offset;
638 	if (hole < len) {
639 		/* To avoid internal fragmentation, if there is very likely not
640 		 * enough space for another buffer, add the remaining space to
641 		 * the current buffer. This extra space is not included in
642 		 * the truesize stored in ctx.
643 		 */
644 		len += hole;
645 		alloc_frag->offset += hole;
646 	}
647 
648 	sg_init_one(rq->sg, buf, len);
649 	err = virtqueue_add_inbuf(rq->vq, rq->sg, 1, (void *)ctx, gfp);
650 	if (err < 0)
651 		put_page(virt_to_head_page(buf));
652 
653 	return err;
654 }
655 
656 /*
657  * Returns false if we couldn't fill entirely (OOM).
658  *
659  * Normally run in the receive path, but can also be run from ndo_open
660  * before we're receiving packets, or from refill_work which is
661  * careful to disable receiving (using napi_disable).
662  */
663 static bool try_fill_recv(struct receive_queue *rq, gfp_t gfp)
664 {
665 	struct virtnet_info *vi = rq->vq->vdev->priv;
666 	int err;
667 	bool oom;
668 
669 	gfp |= __GFP_COLD;
670 	do {
671 		if (vi->mergeable_rx_bufs)
672 			err = add_recvbuf_mergeable(rq, gfp);
673 		else if (vi->big_packets)
674 			err = add_recvbuf_big(rq, gfp);
675 		else
676 			err = add_recvbuf_small(rq, gfp);
677 
678 		oom = err == -ENOMEM;
679 		if (err)
680 			break;
681 	} while (rq->vq->num_free);
682 	virtqueue_kick(rq->vq);
683 	return !oom;
684 }
685 
686 static void skb_recv_done(struct virtqueue *rvq)
687 {
688 	struct virtnet_info *vi = rvq->vdev->priv;
689 	struct receive_queue *rq = &vi->rq[vq2rxq(rvq)];
690 
691 	/* Schedule NAPI, Suppress further interrupts if successful. */
692 	if (napi_schedule_prep(&rq->napi)) {
693 		virtqueue_disable_cb(rvq);
694 		__napi_schedule(&rq->napi);
695 	}
696 }
697 
698 static void virtnet_napi_enable(struct receive_queue *rq)
699 {
700 	napi_enable(&rq->napi);
701 
702 	/* If all buffers were filled by other side before we napi_enabled, we
703 	 * won't get another interrupt, so process any outstanding packets
704 	 * now.  virtnet_poll wants re-enable the queue, so we disable here.
705 	 * We synchronize against interrupts via NAPI_STATE_SCHED */
706 	if (napi_schedule_prep(&rq->napi)) {
707 		virtqueue_disable_cb(rq->vq);
708 		local_bh_disable();
709 		__napi_schedule(&rq->napi);
710 		local_bh_enable();
711 	}
712 }
713 
714 static void refill_work(struct work_struct *work)
715 {
716 	struct virtnet_info *vi =
717 		container_of(work, struct virtnet_info, refill.work);
718 	bool still_empty;
719 	int i;
720 
721 	for (i = 0; i < vi->curr_queue_pairs; i++) {
722 		struct receive_queue *rq = &vi->rq[i];
723 
724 		napi_disable(&rq->napi);
725 		still_empty = !try_fill_recv(rq, GFP_KERNEL);
726 		virtnet_napi_enable(rq);
727 
728 		/* In theory, this can happen: if we don't get any buffers in
729 		 * we will *never* try to fill again.
730 		 */
731 		if (still_empty)
732 			schedule_delayed_work(&vi->refill, HZ/2);
733 	}
734 }
735 
736 static int virtnet_receive(struct receive_queue *rq, int budget)
737 {
738 	struct virtnet_info *vi = rq->vq->vdev->priv;
739 	unsigned int len, received = 0;
740 	void *buf;
741 
742 	while (received < budget &&
743 	       (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) {
744 		receive_buf(rq, buf, len);
745 		received++;
746 	}
747 
748 	if (rq->vq->num_free > virtqueue_get_vring_size(rq->vq) / 2) {
749 		if (!try_fill_recv(rq, GFP_ATOMIC))
750 			schedule_delayed_work(&vi->refill, 0);
751 	}
752 
753 	return received;
754 }
755 
756 static int virtnet_poll(struct napi_struct *napi, int budget)
757 {
758 	struct receive_queue *rq =
759 		container_of(napi, struct receive_queue, napi);
760 	unsigned int r, received = 0;
761 
762 again:
763 	received += virtnet_receive(rq, budget - received);
764 
765 	/* Out of packets? */
766 	if (received < budget) {
767 		r = virtqueue_enable_cb_prepare(rq->vq);
768 		napi_complete(napi);
769 		if (unlikely(virtqueue_poll(rq->vq, r)) &&
770 		    napi_schedule_prep(napi)) {
771 			virtqueue_disable_cb(rq->vq);
772 			__napi_schedule(napi);
773 			goto again;
774 		}
775 	}
776 
777 	return received;
778 }
779 
780 #ifdef CONFIG_NET_RX_BUSY_POLL
781 /* must be called with local_bh_disable()d */
782 static int virtnet_busy_poll(struct napi_struct *napi)
783 {
784 	struct receive_queue *rq =
785 		container_of(napi, struct receive_queue, napi);
786 	struct virtnet_info *vi = rq->vq->vdev->priv;
787 	int r, received = 0, budget = 4;
788 
789 	if (!(vi->status & VIRTIO_NET_S_LINK_UP))
790 		return LL_FLUSH_FAILED;
791 
792 	if (!napi_schedule_prep(napi))
793 		return LL_FLUSH_BUSY;
794 
795 	virtqueue_disable_cb(rq->vq);
796 
797 again:
798 	received += virtnet_receive(rq, budget);
799 
800 	r = virtqueue_enable_cb_prepare(rq->vq);
801 	clear_bit(NAPI_STATE_SCHED, &napi->state);
802 	if (unlikely(virtqueue_poll(rq->vq, r)) &&
803 	    napi_schedule_prep(napi)) {
804 		virtqueue_disable_cb(rq->vq);
805 		if (received < budget) {
806 			budget -= received;
807 			goto again;
808 		} else {
809 			__napi_schedule(napi);
810 		}
811 	}
812 
813 	return received;
814 }
815 #endif	/* CONFIG_NET_RX_BUSY_POLL */
816 
817 static int virtnet_open(struct net_device *dev)
818 {
819 	struct virtnet_info *vi = netdev_priv(dev);
820 	int i;
821 
822 	for (i = 0; i < vi->max_queue_pairs; i++) {
823 		if (i < vi->curr_queue_pairs)
824 			/* Make sure we have some buffers: if oom use wq. */
825 			if (!try_fill_recv(&vi->rq[i], GFP_KERNEL))
826 				schedule_delayed_work(&vi->refill, 0);
827 		virtnet_napi_enable(&vi->rq[i]);
828 	}
829 
830 	return 0;
831 }
832 
833 static void free_old_xmit_skbs(struct send_queue *sq)
834 {
835 	struct sk_buff *skb;
836 	unsigned int len;
837 	struct virtnet_info *vi = sq->vq->vdev->priv;
838 	struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
839 
840 	while ((skb = virtqueue_get_buf(sq->vq, &len)) != NULL) {
841 		pr_debug("Sent skb %p\n", skb);
842 
843 		u64_stats_update_begin(&stats->tx_syncp);
844 		stats->tx_bytes += skb->len;
845 		stats->tx_packets++;
846 		u64_stats_update_end(&stats->tx_syncp);
847 
848 		dev_kfree_skb_any(skb);
849 	}
850 }
851 
852 static int xmit_skb(struct send_queue *sq, struct sk_buff *skb)
853 {
854 	struct skb_vnet_hdr *hdr;
855 	const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
856 	struct virtnet_info *vi = sq->vq->vdev->priv;
857 	unsigned num_sg;
858 	unsigned hdr_len;
859 	bool can_push;
860 
861 	pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
862 	if (vi->mergeable_rx_bufs)
863 		hdr_len = sizeof hdr->mhdr;
864 	else
865 		hdr_len = sizeof hdr->hdr;
866 
867 	can_push = vi->any_header_sg &&
868 		!((unsigned long)skb->data & (__alignof__(*hdr) - 1)) &&
869 		!skb_header_cloned(skb) && skb_headroom(skb) >= hdr_len;
870 	/* Even if we can, don't push here yet as this would skew
871 	 * csum_start offset below. */
872 	if (can_push)
873 		hdr = (struct skb_vnet_hdr *)(skb->data - hdr_len);
874 	else
875 		hdr = skb_vnet_hdr(skb);
876 
877 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
878 		hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
879 		hdr->hdr.csum_start = skb_checksum_start_offset(skb);
880 		hdr->hdr.csum_offset = skb->csum_offset;
881 	} else {
882 		hdr->hdr.flags = 0;
883 		hdr->hdr.csum_offset = hdr->hdr.csum_start = 0;
884 	}
885 
886 	if (skb_is_gso(skb)) {
887 		hdr->hdr.hdr_len = skb_headlen(skb);
888 		hdr->hdr.gso_size = skb_shinfo(skb)->gso_size;
889 		if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
890 			hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
891 		else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
892 			hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
893 		else
894 			BUG();
895 		if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
896 			hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
897 	} else {
898 		hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
899 		hdr->hdr.gso_size = hdr->hdr.hdr_len = 0;
900 	}
901 
902 	if (vi->mergeable_rx_bufs)
903 		hdr->mhdr.num_buffers = 0;
904 
905 	sg_init_table(sq->sg, MAX_SKB_FRAGS + 2);
906 	if (can_push) {
907 		__skb_push(skb, hdr_len);
908 		num_sg = skb_to_sgvec(skb, sq->sg, 0, skb->len);
909 		/* Pull header back to avoid skew in tx bytes calculations. */
910 		__skb_pull(skb, hdr_len);
911 	} else {
912 		sg_set_buf(sq->sg, hdr, hdr_len);
913 		num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len) + 1;
914 	}
915 	return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC);
916 }
917 
918 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
919 {
920 	struct virtnet_info *vi = netdev_priv(dev);
921 	int qnum = skb_get_queue_mapping(skb);
922 	struct send_queue *sq = &vi->sq[qnum];
923 	int err;
924 	struct netdev_queue *txq = netdev_get_tx_queue(dev, qnum);
925 	bool kick = !skb->xmit_more;
926 
927 	/* Free up any pending old buffers before queueing new ones. */
928 	free_old_xmit_skbs(sq);
929 
930 	/* Try to transmit */
931 	err = xmit_skb(sq, skb);
932 
933 	/* This should not happen! */
934 	if (unlikely(err)) {
935 		dev->stats.tx_fifo_errors++;
936 		if (net_ratelimit())
937 			dev_warn(&dev->dev,
938 				 "Unexpected TXQ (%d) queue failure: %d\n", qnum, err);
939 		dev->stats.tx_dropped++;
940 		dev_kfree_skb_any(skb);
941 		return NETDEV_TX_OK;
942 	}
943 
944 	/* Don't wait up for transmitted skbs to be freed. */
945 	skb_orphan(skb);
946 	nf_reset(skb);
947 
948 	/* Apparently nice girls don't return TX_BUSY; stop the queue
949 	 * before it gets out of hand.  Naturally, this wastes entries. */
950 	if (sq->vq->num_free < 2+MAX_SKB_FRAGS) {
951 		netif_stop_subqueue(dev, qnum);
952 		if (unlikely(!virtqueue_enable_cb_delayed(sq->vq))) {
953 			/* More just got used, free them then recheck. */
954 			free_old_xmit_skbs(sq);
955 			if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) {
956 				netif_start_subqueue(dev, qnum);
957 				virtqueue_disable_cb(sq->vq);
958 			}
959 		}
960 	}
961 
962 	if (kick || netif_xmit_stopped(txq))
963 		virtqueue_kick(sq->vq);
964 
965 	return NETDEV_TX_OK;
966 }
967 
968 /*
969  * Send command via the control virtqueue and check status.  Commands
970  * supported by the hypervisor, as indicated by feature bits, should
971  * never fail unless improperly formatted.
972  */
973 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
974 				 struct scatterlist *out)
975 {
976 	struct scatterlist *sgs[4], hdr, stat;
977 	struct virtio_net_ctrl_hdr ctrl;
978 	virtio_net_ctrl_ack status = ~0;
979 	unsigned out_num = 0, tmp;
980 
981 	/* Caller should know better */
982 	BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ));
983 
984 	ctrl.class = class;
985 	ctrl.cmd = cmd;
986 	/* Add header */
987 	sg_init_one(&hdr, &ctrl, sizeof(ctrl));
988 	sgs[out_num++] = &hdr;
989 
990 	if (out)
991 		sgs[out_num++] = out;
992 
993 	/* Add return status. */
994 	sg_init_one(&stat, &status, sizeof(status));
995 	sgs[out_num] = &stat;
996 
997 	BUG_ON(out_num + 1 > ARRAY_SIZE(sgs));
998 	virtqueue_add_sgs(vi->cvq, sgs, out_num, 1, vi, GFP_ATOMIC);
999 
1000 	if (unlikely(!virtqueue_kick(vi->cvq)))
1001 		return status == VIRTIO_NET_OK;
1002 
1003 	/* Spin for a response, the kick causes an ioport write, trapping
1004 	 * into the hypervisor, so the request should be handled immediately.
1005 	 */
1006 	while (!virtqueue_get_buf(vi->cvq, &tmp) &&
1007 	       !virtqueue_is_broken(vi->cvq))
1008 		cpu_relax();
1009 
1010 	return status == VIRTIO_NET_OK;
1011 }
1012 
1013 static int virtnet_set_mac_address(struct net_device *dev, void *p)
1014 {
1015 	struct virtnet_info *vi = netdev_priv(dev);
1016 	struct virtio_device *vdev = vi->vdev;
1017 	int ret;
1018 	struct sockaddr *addr = p;
1019 	struct scatterlist sg;
1020 
1021 	ret = eth_prepare_mac_addr_change(dev, p);
1022 	if (ret)
1023 		return ret;
1024 
1025 	if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1026 		sg_init_one(&sg, addr->sa_data, dev->addr_len);
1027 		if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
1028 					  VIRTIO_NET_CTRL_MAC_ADDR_SET, &sg)) {
1029 			dev_warn(&vdev->dev,
1030 				 "Failed to set mac address by vq command.\n");
1031 			return -EINVAL;
1032 		}
1033 	} else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
1034 		unsigned int i;
1035 
1036 		/* Naturally, this has an atomicity problem. */
1037 		for (i = 0; i < dev->addr_len; i++)
1038 			virtio_cwrite8(vdev,
1039 				       offsetof(struct virtio_net_config, mac) +
1040 				       i, addr->sa_data[i]);
1041 	}
1042 
1043 	eth_commit_mac_addr_change(dev, p);
1044 
1045 	return 0;
1046 }
1047 
1048 static struct rtnl_link_stats64 *virtnet_stats(struct net_device *dev,
1049 					       struct rtnl_link_stats64 *tot)
1050 {
1051 	struct virtnet_info *vi = netdev_priv(dev);
1052 	int cpu;
1053 	unsigned int start;
1054 
1055 	for_each_possible_cpu(cpu) {
1056 		struct virtnet_stats *stats = per_cpu_ptr(vi->stats, cpu);
1057 		u64 tpackets, tbytes, rpackets, rbytes;
1058 
1059 		do {
1060 			start = u64_stats_fetch_begin_irq(&stats->tx_syncp);
1061 			tpackets = stats->tx_packets;
1062 			tbytes   = stats->tx_bytes;
1063 		} while (u64_stats_fetch_retry_irq(&stats->tx_syncp, start));
1064 
1065 		do {
1066 			start = u64_stats_fetch_begin_irq(&stats->rx_syncp);
1067 			rpackets = stats->rx_packets;
1068 			rbytes   = stats->rx_bytes;
1069 		} while (u64_stats_fetch_retry_irq(&stats->rx_syncp, start));
1070 
1071 		tot->rx_packets += rpackets;
1072 		tot->tx_packets += tpackets;
1073 		tot->rx_bytes   += rbytes;
1074 		tot->tx_bytes   += tbytes;
1075 	}
1076 
1077 	tot->tx_dropped = dev->stats.tx_dropped;
1078 	tot->tx_fifo_errors = dev->stats.tx_fifo_errors;
1079 	tot->rx_dropped = dev->stats.rx_dropped;
1080 	tot->rx_length_errors = dev->stats.rx_length_errors;
1081 	tot->rx_frame_errors = dev->stats.rx_frame_errors;
1082 
1083 	return tot;
1084 }
1085 
1086 #ifdef CONFIG_NET_POLL_CONTROLLER
1087 static void virtnet_netpoll(struct net_device *dev)
1088 {
1089 	struct virtnet_info *vi = netdev_priv(dev);
1090 	int i;
1091 
1092 	for (i = 0; i < vi->curr_queue_pairs; i++)
1093 		napi_schedule(&vi->rq[i].napi);
1094 }
1095 #endif
1096 
1097 static void virtnet_ack_link_announce(struct virtnet_info *vi)
1098 {
1099 	rtnl_lock();
1100 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE,
1101 				  VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL))
1102 		dev_warn(&vi->dev->dev, "Failed to ack link announce.\n");
1103 	rtnl_unlock();
1104 }
1105 
1106 static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs)
1107 {
1108 	struct scatterlist sg;
1109 	struct virtio_net_ctrl_mq s;
1110 	struct net_device *dev = vi->dev;
1111 
1112 	if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ))
1113 		return 0;
1114 
1115 	s.virtqueue_pairs = queue_pairs;
1116 	sg_init_one(&sg, &s, sizeof(s));
1117 
1118 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ,
1119 				  VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg)) {
1120 		dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n",
1121 			 queue_pairs);
1122 		return -EINVAL;
1123 	} else {
1124 		vi->curr_queue_pairs = queue_pairs;
1125 		/* virtnet_open() will refill when device is going to up. */
1126 		if (dev->flags & IFF_UP)
1127 			schedule_delayed_work(&vi->refill, 0);
1128 	}
1129 
1130 	return 0;
1131 }
1132 
1133 static int virtnet_close(struct net_device *dev)
1134 {
1135 	struct virtnet_info *vi = netdev_priv(dev);
1136 	int i;
1137 
1138 	/* Make sure refill_work doesn't re-enable napi! */
1139 	cancel_delayed_work_sync(&vi->refill);
1140 
1141 	for (i = 0; i < vi->max_queue_pairs; i++)
1142 		napi_disable(&vi->rq[i].napi);
1143 
1144 	return 0;
1145 }
1146 
1147 static void virtnet_set_rx_mode(struct net_device *dev)
1148 {
1149 	struct virtnet_info *vi = netdev_priv(dev);
1150 	struct scatterlist sg[2];
1151 	u8 promisc, allmulti;
1152 	struct virtio_net_ctrl_mac *mac_data;
1153 	struct netdev_hw_addr *ha;
1154 	int uc_count;
1155 	int mc_count;
1156 	void *buf;
1157 	int i;
1158 
1159 	/* We can't dynamically set ndo_set_rx_mode, so return gracefully */
1160 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
1161 		return;
1162 
1163 	promisc = ((dev->flags & IFF_PROMISC) != 0);
1164 	allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
1165 
1166 	sg_init_one(sg, &promisc, sizeof(promisc));
1167 
1168 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
1169 				  VIRTIO_NET_CTRL_RX_PROMISC, sg))
1170 		dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
1171 			 promisc ? "en" : "dis");
1172 
1173 	sg_init_one(sg, &allmulti, sizeof(allmulti));
1174 
1175 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
1176 				  VIRTIO_NET_CTRL_RX_ALLMULTI, sg))
1177 		dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
1178 			 allmulti ? "en" : "dis");
1179 
1180 	uc_count = netdev_uc_count(dev);
1181 	mc_count = netdev_mc_count(dev);
1182 	/* MAC filter - use one buffer for both lists */
1183 	buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
1184 		      (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
1185 	mac_data = buf;
1186 	if (!buf)
1187 		return;
1188 
1189 	sg_init_table(sg, 2);
1190 
1191 	/* Store the unicast list and count in the front of the buffer */
1192 	mac_data->entries = uc_count;
1193 	i = 0;
1194 	netdev_for_each_uc_addr(ha, dev)
1195 		memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
1196 
1197 	sg_set_buf(&sg[0], mac_data,
1198 		   sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
1199 
1200 	/* multicast list and count fill the end */
1201 	mac_data = (void *)&mac_data->macs[uc_count][0];
1202 
1203 	mac_data->entries = mc_count;
1204 	i = 0;
1205 	netdev_for_each_mc_addr(ha, dev)
1206 		memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
1207 
1208 	sg_set_buf(&sg[1], mac_data,
1209 		   sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
1210 
1211 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
1212 				  VIRTIO_NET_CTRL_MAC_TABLE_SET, sg))
1213 		dev_warn(&dev->dev, "Failed to set MAC filter table.\n");
1214 
1215 	kfree(buf);
1216 }
1217 
1218 static int virtnet_vlan_rx_add_vid(struct net_device *dev,
1219 				   __be16 proto, u16 vid)
1220 {
1221 	struct virtnet_info *vi = netdev_priv(dev);
1222 	struct scatterlist sg;
1223 
1224 	sg_init_one(&sg, &vid, sizeof(vid));
1225 
1226 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1227 				  VIRTIO_NET_CTRL_VLAN_ADD, &sg))
1228 		dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
1229 	return 0;
1230 }
1231 
1232 static int virtnet_vlan_rx_kill_vid(struct net_device *dev,
1233 				    __be16 proto, u16 vid)
1234 {
1235 	struct virtnet_info *vi = netdev_priv(dev);
1236 	struct scatterlist sg;
1237 
1238 	sg_init_one(&sg, &vid, sizeof(vid));
1239 
1240 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1241 				  VIRTIO_NET_CTRL_VLAN_DEL, &sg))
1242 		dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
1243 	return 0;
1244 }
1245 
1246 static void virtnet_clean_affinity(struct virtnet_info *vi, long hcpu)
1247 {
1248 	int i;
1249 
1250 	if (vi->affinity_hint_set) {
1251 		for (i = 0; i < vi->max_queue_pairs; i++) {
1252 			virtqueue_set_affinity(vi->rq[i].vq, -1);
1253 			virtqueue_set_affinity(vi->sq[i].vq, -1);
1254 		}
1255 
1256 		vi->affinity_hint_set = false;
1257 	}
1258 }
1259 
1260 static void virtnet_set_affinity(struct virtnet_info *vi)
1261 {
1262 	int i;
1263 	int cpu;
1264 
1265 	/* In multiqueue mode, when the number of cpu is equal to the number of
1266 	 * queue pairs, we let the queue pairs to be private to one cpu by
1267 	 * setting the affinity hint to eliminate the contention.
1268 	 */
1269 	if (vi->curr_queue_pairs == 1 ||
1270 	    vi->max_queue_pairs != num_online_cpus()) {
1271 		virtnet_clean_affinity(vi, -1);
1272 		return;
1273 	}
1274 
1275 	i = 0;
1276 	for_each_online_cpu(cpu) {
1277 		virtqueue_set_affinity(vi->rq[i].vq, cpu);
1278 		virtqueue_set_affinity(vi->sq[i].vq, cpu);
1279 		netif_set_xps_queue(vi->dev, cpumask_of(cpu), i);
1280 		i++;
1281 	}
1282 
1283 	vi->affinity_hint_set = true;
1284 }
1285 
1286 static int virtnet_cpu_callback(struct notifier_block *nfb,
1287 			        unsigned long action, void *hcpu)
1288 {
1289 	struct virtnet_info *vi = container_of(nfb, struct virtnet_info, nb);
1290 
1291 	switch(action & ~CPU_TASKS_FROZEN) {
1292 	case CPU_ONLINE:
1293 	case CPU_DOWN_FAILED:
1294 	case CPU_DEAD:
1295 		virtnet_set_affinity(vi);
1296 		break;
1297 	case CPU_DOWN_PREPARE:
1298 		virtnet_clean_affinity(vi, (long)hcpu);
1299 		break;
1300 	default:
1301 		break;
1302 	}
1303 
1304 	return NOTIFY_OK;
1305 }
1306 
1307 static void virtnet_get_ringparam(struct net_device *dev,
1308 				struct ethtool_ringparam *ring)
1309 {
1310 	struct virtnet_info *vi = netdev_priv(dev);
1311 
1312 	ring->rx_max_pending = virtqueue_get_vring_size(vi->rq[0].vq);
1313 	ring->tx_max_pending = virtqueue_get_vring_size(vi->sq[0].vq);
1314 	ring->rx_pending = ring->rx_max_pending;
1315 	ring->tx_pending = ring->tx_max_pending;
1316 }
1317 
1318 
1319 static void virtnet_get_drvinfo(struct net_device *dev,
1320 				struct ethtool_drvinfo *info)
1321 {
1322 	struct virtnet_info *vi = netdev_priv(dev);
1323 	struct virtio_device *vdev = vi->vdev;
1324 
1325 	strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
1326 	strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version));
1327 	strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info));
1328 
1329 }
1330 
1331 /* TODO: Eliminate OOO packets during switching */
1332 static int virtnet_set_channels(struct net_device *dev,
1333 				struct ethtool_channels *channels)
1334 {
1335 	struct virtnet_info *vi = netdev_priv(dev);
1336 	u16 queue_pairs = channels->combined_count;
1337 	int err;
1338 
1339 	/* We don't support separate rx/tx channels.
1340 	 * We don't allow setting 'other' channels.
1341 	 */
1342 	if (channels->rx_count || channels->tx_count || channels->other_count)
1343 		return -EINVAL;
1344 
1345 	if (queue_pairs > vi->max_queue_pairs || queue_pairs == 0)
1346 		return -EINVAL;
1347 
1348 	get_online_cpus();
1349 	err = virtnet_set_queues(vi, queue_pairs);
1350 	if (!err) {
1351 		netif_set_real_num_tx_queues(dev, queue_pairs);
1352 		netif_set_real_num_rx_queues(dev, queue_pairs);
1353 
1354 		virtnet_set_affinity(vi);
1355 	}
1356 	put_online_cpus();
1357 
1358 	return err;
1359 }
1360 
1361 static void virtnet_get_channels(struct net_device *dev,
1362 				 struct ethtool_channels *channels)
1363 {
1364 	struct virtnet_info *vi = netdev_priv(dev);
1365 
1366 	channels->combined_count = vi->curr_queue_pairs;
1367 	channels->max_combined = vi->max_queue_pairs;
1368 	channels->max_other = 0;
1369 	channels->rx_count = 0;
1370 	channels->tx_count = 0;
1371 	channels->other_count = 0;
1372 }
1373 
1374 static const struct ethtool_ops virtnet_ethtool_ops = {
1375 	.get_drvinfo = virtnet_get_drvinfo,
1376 	.get_link = ethtool_op_get_link,
1377 	.get_ringparam = virtnet_get_ringparam,
1378 	.set_channels = virtnet_set_channels,
1379 	.get_channels = virtnet_get_channels,
1380 };
1381 
1382 #define MIN_MTU 68
1383 #define MAX_MTU 65535
1384 
1385 static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
1386 {
1387 	if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
1388 		return -EINVAL;
1389 	dev->mtu = new_mtu;
1390 	return 0;
1391 }
1392 
1393 static const struct net_device_ops virtnet_netdev = {
1394 	.ndo_open            = virtnet_open,
1395 	.ndo_stop   	     = virtnet_close,
1396 	.ndo_start_xmit      = start_xmit,
1397 	.ndo_validate_addr   = eth_validate_addr,
1398 	.ndo_set_mac_address = virtnet_set_mac_address,
1399 	.ndo_set_rx_mode     = virtnet_set_rx_mode,
1400 	.ndo_change_mtu	     = virtnet_change_mtu,
1401 	.ndo_get_stats64     = virtnet_stats,
1402 	.ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
1403 	.ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
1404 #ifdef CONFIG_NET_POLL_CONTROLLER
1405 	.ndo_poll_controller = virtnet_netpoll,
1406 #endif
1407 #ifdef CONFIG_NET_RX_BUSY_POLL
1408 	.ndo_busy_poll		= virtnet_busy_poll,
1409 #endif
1410 };
1411 
1412 static void virtnet_config_changed_work(struct work_struct *work)
1413 {
1414 	struct virtnet_info *vi =
1415 		container_of(work, struct virtnet_info, config_work);
1416 	u16 v;
1417 
1418 	if (virtio_cread_feature(vi->vdev, VIRTIO_NET_F_STATUS,
1419 				 struct virtio_net_config, status, &v) < 0)
1420 		return;
1421 
1422 	if (v & VIRTIO_NET_S_ANNOUNCE) {
1423 		netdev_notify_peers(vi->dev);
1424 		virtnet_ack_link_announce(vi);
1425 	}
1426 
1427 	/* Ignore unknown (future) status bits */
1428 	v &= VIRTIO_NET_S_LINK_UP;
1429 
1430 	if (vi->status == v)
1431 		return;
1432 
1433 	vi->status = v;
1434 
1435 	if (vi->status & VIRTIO_NET_S_LINK_UP) {
1436 		netif_carrier_on(vi->dev);
1437 		netif_tx_wake_all_queues(vi->dev);
1438 	} else {
1439 		netif_carrier_off(vi->dev);
1440 		netif_tx_stop_all_queues(vi->dev);
1441 	}
1442 }
1443 
1444 static void virtnet_config_changed(struct virtio_device *vdev)
1445 {
1446 	struct virtnet_info *vi = vdev->priv;
1447 
1448 	schedule_work(&vi->config_work);
1449 }
1450 
1451 static void virtnet_free_queues(struct virtnet_info *vi)
1452 {
1453 	int i;
1454 
1455 	for (i = 0; i < vi->max_queue_pairs; i++)
1456 		netif_napi_del(&vi->rq[i].napi);
1457 
1458 	kfree(vi->rq);
1459 	kfree(vi->sq);
1460 }
1461 
1462 static void free_receive_bufs(struct virtnet_info *vi)
1463 {
1464 	int i;
1465 
1466 	for (i = 0; i < vi->max_queue_pairs; i++) {
1467 		while (vi->rq[i].pages)
1468 			__free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0);
1469 	}
1470 }
1471 
1472 static void free_receive_page_frags(struct virtnet_info *vi)
1473 {
1474 	int i;
1475 	for (i = 0; i < vi->max_queue_pairs; i++)
1476 		if (vi->rq[i].alloc_frag.page)
1477 			put_page(vi->rq[i].alloc_frag.page);
1478 }
1479 
1480 static void free_unused_bufs(struct virtnet_info *vi)
1481 {
1482 	void *buf;
1483 	int i;
1484 
1485 	for (i = 0; i < vi->max_queue_pairs; i++) {
1486 		struct virtqueue *vq = vi->sq[i].vq;
1487 		while ((buf = virtqueue_detach_unused_buf(vq)) != NULL)
1488 			dev_kfree_skb(buf);
1489 	}
1490 
1491 	for (i = 0; i < vi->max_queue_pairs; i++) {
1492 		struct virtqueue *vq = vi->rq[i].vq;
1493 
1494 		while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) {
1495 			if (vi->mergeable_rx_bufs) {
1496 				unsigned long ctx = (unsigned long)buf;
1497 				void *base = mergeable_ctx_to_buf_address(ctx);
1498 				put_page(virt_to_head_page(base));
1499 			} else if (vi->big_packets) {
1500 				give_pages(&vi->rq[i], buf);
1501 			} else {
1502 				dev_kfree_skb(buf);
1503 			}
1504 		}
1505 	}
1506 }
1507 
1508 static void virtnet_del_vqs(struct virtnet_info *vi)
1509 {
1510 	struct virtio_device *vdev = vi->vdev;
1511 
1512 	virtnet_clean_affinity(vi, -1);
1513 
1514 	vdev->config->del_vqs(vdev);
1515 
1516 	virtnet_free_queues(vi);
1517 }
1518 
1519 static int virtnet_find_vqs(struct virtnet_info *vi)
1520 {
1521 	vq_callback_t **callbacks;
1522 	struct virtqueue **vqs;
1523 	int ret = -ENOMEM;
1524 	int i, total_vqs;
1525 	const char **names;
1526 
1527 	/* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
1528 	 * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
1529 	 * possible control vq.
1530 	 */
1531 	total_vqs = vi->max_queue_pairs * 2 +
1532 		    virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ);
1533 
1534 	/* Allocate space for find_vqs parameters */
1535 	vqs = kzalloc(total_vqs * sizeof(*vqs), GFP_KERNEL);
1536 	if (!vqs)
1537 		goto err_vq;
1538 	callbacks = kmalloc(total_vqs * sizeof(*callbacks), GFP_KERNEL);
1539 	if (!callbacks)
1540 		goto err_callback;
1541 	names = kmalloc(total_vqs * sizeof(*names), GFP_KERNEL);
1542 	if (!names)
1543 		goto err_names;
1544 
1545 	/* Parameters for control virtqueue, if any */
1546 	if (vi->has_cvq) {
1547 		callbacks[total_vqs - 1] = NULL;
1548 		names[total_vqs - 1] = "control";
1549 	}
1550 
1551 	/* Allocate/initialize parameters for send/receive virtqueues */
1552 	for (i = 0; i < vi->max_queue_pairs; i++) {
1553 		callbacks[rxq2vq(i)] = skb_recv_done;
1554 		callbacks[txq2vq(i)] = skb_xmit_done;
1555 		sprintf(vi->rq[i].name, "input.%d", i);
1556 		sprintf(vi->sq[i].name, "output.%d", i);
1557 		names[rxq2vq(i)] = vi->rq[i].name;
1558 		names[txq2vq(i)] = vi->sq[i].name;
1559 	}
1560 
1561 	ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs, callbacks,
1562 					 names);
1563 	if (ret)
1564 		goto err_find;
1565 
1566 	if (vi->has_cvq) {
1567 		vi->cvq = vqs[total_vqs - 1];
1568 		if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
1569 			vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
1570 	}
1571 
1572 	for (i = 0; i < vi->max_queue_pairs; i++) {
1573 		vi->rq[i].vq = vqs[rxq2vq(i)];
1574 		vi->sq[i].vq = vqs[txq2vq(i)];
1575 	}
1576 
1577 	kfree(names);
1578 	kfree(callbacks);
1579 	kfree(vqs);
1580 
1581 	return 0;
1582 
1583 err_find:
1584 	kfree(names);
1585 err_names:
1586 	kfree(callbacks);
1587 err_callback:
1588 	kfree(vqs);
1589 err_vq:
1590 	return ret;
1591 }
1592 
1593 static int virtnet_alloc_queues(struct virtnet_info *vi)
1594 {
1595 	int i;
1596 
1597 	vi->sq = kzalloc(sizeof(*vi->sq) * vi->max_queue_pairs, GFP_KERNEL);
1598 	if (!vi->sq)
1599 		goto err_sq;
1600 	vi->rq = kzalloc(sizeof(*vi->rq) * vi->max_queue_pairs, GFP_KERNEL);
1601 	if (!vi->rq)
1602 		goto err_rq;
1603 
1604 	INIT_DELAYED_WORK(&vi->refill, refill_work);
1605 	for (i = 0; i < vi->max_queue_pairs; i++) {
1606 		vi->rq[i].pages = NULL;
1607 		netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll,
1608 			       napi_weight);
1609 		napi_hash_add(&vi->rq[i].napi);
1610 
1611 		sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg));
1612 		ewma_init(&vi->rq[i].mrg_avg_pkt_len, 1, RECEIVE_AVG_WEIGHT);
1613 		sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg));
1614 	}
1615 
1616 	return 0;
1617 
1618 err_rq:
1619 	kfree(vi->sq);
1620 err_sq:
1621 	return -ENOMEM;
1622 }
1623 
1624 static int init_vqs(struct virtnet_info *vi)
1625 {
1626 	int ret;
1627 
1628 	/* Allocate send & receive queues */
1629 	ret = virtnet_alloc_queues(vi);
1630 	if (ret)
1631 		goto err;
1632 
1633 	ret = virtnet_find_vqs(vi);
1634 	if (ret)
1635 		goto err_free;
1636 
1637 	get_online_cpus();
1638 	virtnet_set_affinity(vi);
1639 	put_online_cpus();
1640 
1641 	return 0;
1642 
1643 err_free:
1644 	virtnet_free_queues(vi);
1645 err:
1646 	return ret;
1647 }
1648 
1649 #ifdef CONFIG_SYSFS
1650 static ssize_t mergeable_rx_buffer_size_show(struct netdev_rx_queue *queue,
1651 		struct rx_queue_attribute *attribute, char *buf)
1652 {
1653 	struct virtnet_info *vi = netdev_priv(queue->dev);
1654 	unsigned int queue_index = get_netdev_rx_queue_index(queue);
1655 	struct ewma *avg;
1656 
1657 	BUG_ON(queue_index >= vi->max_queue_pairs);
1658 	avg = &vi->rq[queue_index].mrg_avg_pkt_len;
1659 	return sprintf(buf, "%u\n", get_mergeable_buf_len(avg));
1660 }
1661 
1662 static struct rx_queue_attribute mergeable_rx_buffer_size_attribute =
1663 	__ATTR_RO(mergeable_rx_buffer_size);
1664 
1665 static struct attribute *virtio_net_mrg_rx_attrs[] = {
1666 	&mergeable_rx_buffer_size_attribute.attr,
1667 	NULL
1668 };
1669 
1670 static const struct attribute_group virtio_net_mrg_rx_group = {
1671 	.name = "virtio_net",
1672 	.attrs = virtio_net_mrg_rx_attrs
1673 };
1674 #endif
1675 
1676 static int virtnet_probe(struct virtio_device *vdev)
1677 {
1678 	int i, err;
1679 	struct net_device *dev;
1680 	struct virtnet_info *vi;
1681 	u16 max_queue_pairs;
1682 
1683 	/* Find if host supports multiqueue virtio_net device */
1684 	err = virtio_cread_feature(vdev, VIRTIO_NET_F_MQ,
1685 				   struct virtio_net_config,
1686 				   max_virtqueue_pairs, &max_queue_pairs);
1687 
1688 	/* We need at least 2 queue's */
1689 	if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
1690 	    max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
1691 	    !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
1692 		max_queue_pairs = 1;
1693 
1694 	/* Allocate ourselves a network device with room for our info */
1695 	dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs);
1696 	if (!dev)
1697 		return -ENOMEM;
1698 
1699 	/* Set up network device as normal. */
1700 	dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE;
1701 	dev->netdev_ops = &virtnet_netdev;
1702 	dev->features = NETIF_F_HIGHDMA;
1703 
1704 	dev->ethtool_ops = &virtnet_ethtool_ops;
1705 	SET_NETDEV_DEV(dev, &vdev->dev);
1706 
1707 	/* Do we support "hardware" checksums? */
1708 	if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
1709 		/* This opens up the world of extra features. */
1710 		dev->hw_features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1711 		if (csum)
1712 			dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1713 
1714 		if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
1715 			dev->hw_features |= NETIF_F_TSO
1716 				| NETIF_F_TSO_ECN | NETIF_F_TSO6;
1717 		}
1718 		/* Individual feature bits: what can host handle? */
1719 		if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
1720 			dev->hw_features |= NETIF_F_TSO;
1721 		if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
1722 			dev->hw_features |= NETIF_F_TSO6;
1723 		if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
1724 			dev->hw_features |= NETIF_F_TSO_ECN;
1725 
1726 		if (gso)
1727 			dev->features |= dev->hw_features & NETIF_F_ALL_TSO;
1728 		/* (!csum && gso) case will be fixed by register_netdev() */
1729 	}
1730 	if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM))
1731 		dev->features |= NETIF_F_RXCSUM;
1732 
1733 	dev->vlan_features = dev->features;
1734 
1735 	/* Configuration may specify what MAC to use.  Otherwise random. */
1736 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
1737 		virtio_cread_bytes(vdev,
1738 				   offsetof(struct virtio_net_config, mac),
1739 				   dev->dev_addr, dev->addr_len);
1740 	else
1741 		eth_hw_addr_random(dev);
1742 
1743 	/* Set up our device-specific information */
1744 	vi = netdev_priv(dev);
1745 	vi->dev = dev;
1746 	vi->vdev = vdev;
1747 	vdev->priv = vi;
1748 	vi->stats = alloc_percpu(struct virtnet_stats);
1749 	err = -ENOMEM;
1750 	if (vi->stats == NULL)
1751 		goto free;
1752 
1753 	for_each_possible_cpu(i) {
1754 		struct virtnet_stats *virtnet_stats;
1755 		virtnet_stats = per_cpu_ptr(vi->stats, i);
1756 		u64_stats_init(&virtnet_stats->tx_syncp);
1757 		u64_stats_init(&virtnet_stats->rx_syncp);
1758 	}
1759 
1760 	INIT_WORK(&vi->config_work, virtnet_config_changed_work);
1761 
1762 	/* If we can receive ANY GSO packets, we must allocate large ones. */
1763 	if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
1764 	    virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
1765 	    virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
1766 		vi->big_packets = true;
1767 
1768 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
1769 		vi->mergeable_rx_bufs = true;
1770 
1771 	if (virtio_has_feature(vdev, VIRTIO_F_ANY_LAYOUT))
1772 		vi->any_header_sg = true;
1773 
1774 	if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
1775 		vi->has_cvq = true;
1776 
1777 	if (vi->any_header_sg) {
1778 		if (vi->mergeable_rx_bufs)
1779 			dev->needed_headroom = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1780 		else
1781 			dev->needed_headroom = sizeof(struct virtio_net_hdr);
1782 	}
1783 
1784 	/* Use single tx/rx queue pair as default */
1785 	vi->curr_queue_pairs = 1;
1786 	vi->max_queue_pairs = max_queue_pairs;
1787 
1788 	/* Allocate/initialize the rx/tx queues, and invoke find_vqs */
1789 	err = init_vqs(vi);
1790 	if (err)
1791 		goto free_stats;
1792 
1793 #ifdef CONFIG_SYSFS
1794 	if (vi->mergeable_rx_bufs)
1795 		dev->sysfs_rx_queue_group = &virtio_net_mrg_rx_group;
1796 #endif
1797 	netif_set_real_num_tx_queues(dev, vi->curr_queue_pairs);
1798 	netif_set_real_num_rx_queues(dev, vi->curr_queue_pairs);
1799 
1800 	err = register_netdev(dev);
1801 	if (err) {
1802 		pr_debug("virtio_net: registering device failed\n");
1803 		goto free_vqs;
1804 	}
1805 
1806 	virtio_device_ready(vdev);
1807 
1808 	/* Last of all, set up some receive buffers. */
1809 	for (i = 0; i < vi->curr_queue_pairs; i++) {
1810 		try_fill_recv(&vi->rq[i], GFP_KERNEL);
1811 
1812 		/* If we didn't even get one input buffer, we're useless. */
1813 		if (vi->rq[i].vq->num_free ==
1814 		    virtqueue_get_vring_size(vi->rq[i].vq)) {
1815 			free_unused_bufs(vi);
1816 			err = -ENOMEM;
1817 			goto free_recv_bufs;
1818 		}
1819 	}
1820 
1821 	vi->nb.notifier_call = &virtnet_cpu_callback;
1822 	err = register_hotcpu_notifier(&vi->nb);
1823 	if (err) {
1824 		pr_debug("virtio_net: registering cpu notifier failed\n");
1825 		goto free_recv_bufs;
1826 	}
1827 
1828 	/* Assume link up if device can't report link status,
1829 	   otherwise get link status from config. */
1830 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
1831 		netif_carrier_off(dev);
1832 		schedule_work(&vi->config_work);
1833 	} else {
1834 		vi->status = VIRTIO_NET_S_LINK_UP;
1835 		netif_carrier_on(dev);
1836 	}
1837 
1838 	pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
1839 		 dev->name, max_queue_pairs);
1840 
1841 	return 0;
1842 
1843 free_recv_bufs:
1844 	vi->vdev->config->reset(vdev);
1845 
1846 	free_receive_bufs(vi);
1847 	unregister_netdev(dev);
1848 free_vqs:
1849 	cancel_delayed_work_sync(&vi->refill);
1850 	free_receive_page_frags(vi);
1851 	virtnet_del_vqs(vi);
1852 free_stats:
1853 	free_percpu(vi->stats);
1854 free:
1855 	free_netdev(dev);
1856 	return err;
1857 }
1858 
1859 static void remove_vq_common(struct virtnet_info *vi)
1860 {
1861 	vi->vdev->config->reset(vi->vdev);
1862 
1863 	/* Free unused buffers in both send and recv, if any. */
1864 	free_unused_bufs(vi);
1865 
1866 	free_receive_bufs(vi);
1867 
1868 	free_receive_page_frags(vi);
1869 
1870 	virtnet_del_vqs(vi);
1871 }
1872 
1873 static void virtnet_remove(struct virtio_device *vdev)
1874 {
1875 	struct virtnet_info *vi = vdev->priv;
1876 
1877 	unregister_hotcpu_notifier(&vi->nb);
1878 
1879 	/* Make sure no work handler is accessing the device. */
1880 	flush_work(&vi->config_work);
1881 
1882 	unregister_netdev(vi->dev);
1883 
1884 	remove_vq_common(vi);
1885 
1886 	free_percpu(vi->stats);
1887 	free_netdev(vi->dev);
1888 }
1889 
1890 #ifdef CONFIG_PM_SLEEP
1891 static int virtnet_freeze(struct virtio_device *vdev)
1892 {
1893 	struct virtnet_info *vi = vdev->priv;
1894 	int i;
1895 
1896 	unregister_hotcpu_notifier(&vi->nb);
1897 
1898 	/* Make sure no work handler is accessing the device */
1899 	flush_work(&vi->config_work);
1900 
1901 	netif_device_detach(vi->dev);
1902 	cancel_delayed_work_sync(&vi->refill);
1903 
1904 	if (netif_running(vi->dev)) {
1905 		for (i = 0; i < vi->max_queue_pairs; i++) {
1906 			napi_disable(&vi->rq[i].napi);
1907 			napi_hash_del(&vi->rq[i].napi);
1908 			netif_napi_del(&vi->rq[i].napi);
1909 		}
1910 	}
1911 
1912 	remove_vq_common(vi);
1913 
1914 	return 0;
1915 }
1916 
1917 static int virtnet_restore(struct virtio_device *vdev)
1918 {
1919 	struct virtnet_info *vi = vdev->priv;
1920 	int err, i;
1921 
1922 	err = init_vqs(vi);
1923 	if (err)
1924 		return err;
1925 
1926 	virtio_device_ready(vdev);
1927 
1928 	if (netif_running(vi->dev)) {
1929 		for (i = 0; i < vi->curr_queue_pairs; i++)
1930 			if (!try_fill_recv(&vi->rq[i], GFP_KERNEL))
1931 				schedule_delayed_work(&vi->refill, 0);
1932 
1933 		for (i = 0; i < vi->max_queue_pairs; i++)
1934 			virtnet_napi_enable(&vi->rq[i]);
1935 	}
1936 
1937 	netif_device_attach(vi->dev);
1938 
1939 	rtnl_lock();
1940 	virtnet_set_queues(vi, vi->curr_queue_pairs);
1941 	rtnl_unlock();
1942 
1943 	err = register_hotcpu_notifier(&vi->nb);
1944 	if (err)
1945 		return err;
1946 
1947 	return 0;
1948 }
1949 #endif
1950 
1951 static struct virtio_device_id id_table[] = {
1952 	{ VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
1953 	{ 0 },
1954 };
1955 
1956 static unsigned int features[] = {
1957 	VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
1958 	VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
1959 	VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_TSO6,
1960 	VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
1961 	VIRTIO_NET_F_GUEST_ECN,
1962 	VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ,
1963 	VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN,
1964 	VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ,
1965 	VIRTIO_NET_F_CTRL_MAC_ADDR,
1966 	VIRTIO_F_ANY_LAYOUT,
1967 };
1968 
1969 static struct virtio_driver virtio_net_driver = {
1970 	.feature_table = features,
1971 	.feature_table_size = ARRAY_SIZE(features),
1972 	.driver.name =	KBUILD_MODNAME,
1973 	.driver.owner =	THIS_MODULE,
1974 	.id_table =	id_table,
1975 	.probe =	virtnet_probe,
1976 	.remove =	virtnet_remove,
1977 	.config_changed = virtnet_config_changed,
1978 #ifdef CONFIG_PM_SLEEP
1979 	.freeze =	virtnet_freeze,
1980 	.restore =	virtnet_restore,
1981 #endif
1982 };
1983 
1984 module_virtio_driver(virtio_net_driver);
1985 
1986 MODULE_DEVICE_TABLE(virtio, id_table);
1987 MODULE_DESCRIPTION("Virtio network driver");
1988 MODULE_LICENSE("GPL");
1989