xref: /linux/drivers/net/virtio_net.c (revision b889fcf63cb62e7fdb7816565e28f44dbe4a76a5)
1 /* A network driver using virtio.
2  *
3  * Copyright 2007 Rusty Russell <rusty@rustcorp.com.au> IBM Corporation
4  *
5  * This program is free software; you can redistribute it and/or modify
6  * it under the terms of the GNU General Public License as published by
7  * the Free Software Foundation; either version 2 of the License, or
8  * (at your option) any later version.
9  *
10  * This program is distributed in the hope that it will be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License for more details.
14  *
15  * You should have received a copy of the GNU General Public License
16  * along with this program; if not, write to the Free Software
17  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
18  */
19 //#define DEBUG
20 #include <linux/netdevice.h>
21 #include <linux/etherdevice.h>
22 #include <linux/ethtool.h>
23 #include <linux/module.h>
24 #include <linux/virtio.h>
25 #include <linux/virtio_net.h>
26 #include <linux/scatterlist.h>
27 #include <linux/if_vlan.h>
28 #include <linux/slab.h>
29 
30 static int napi_weight = 128;
31 module_param(napi_weight, int, 0444);
32 
33 static bool csum = true, gso = true;
34 module_param(csum, bool, 0444);
35 module_param(gso, bool, 0444);
36 
37 /* FIXME: MTU in config. */
38 #define MAX_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
39 #define GOOD_COPY_LEN	128
40 
41 #define VIRTNET_SEND_COMMAND_SG_MAX    2
42 #define VIRTNET_DRIVER_VERSION "1.0.0"
43 
44 struct virtnet_stats {
45 	struct u64_stats_sync tx_syncp;
46 	struct u64_stats_sync rx_syncp;
47 	u64 tx_bytes;
48 	u64 tx_packets;
49 
50 	u64 rx_bytes;
51 	u64 rx_packets;
52 };
53 
54 /* Internal representation of a send virtqueue */
55 struct send_queue {
56 	/* Virtqueue associated with this send _queue */
57 	struct virtqueue *vq;
58 
59 	/* TX: fragments + linear part + virtio header */
60 	struct scatterlist sg[MAX_SKB_FRAGS + 2];
61 
62 	/* Name of the send queue: output.$index */
63 	char name[40];
64 };
65 
66 /* Internal representation of a receive virtqueue */
67 struct receive_queue {
68 	/* Virtqueue associated with this receive_queue */
69 	struct virtqueue *vq;
70 
71 	struct napi_struct napi;
72 
73 	/* Number of input buffers, and max we've ever had. */
74 	unsigned int num, max;
75 
76 	/* Chain pages by the private ptr. */
77 	struct page *pages;
78 
79 	/* RX: fragments + linear part + virtio header */
80 	struct scatterlist sg[MAX_SKB_FRAGS + 2];
81 
82 	/* Name of this receive queue: input.$index */
83 	char name[40];
84 };
85 
86 struct virtnet_info {
87 	struct virtio_device *vdev;
88 	struct virtqueue *cvq;
89 	struct net_device *dev;
90 	struct send_queue *sq;
91 	struct receive_queue *rq;
92 	unsigned int status;
93 
94 	/* Max # of queue pairs supported by the device */
95 	u16 max_queue_pairs;
96 
97 	/* # of queue pairs currently used by the driver */
98 	u16 curr_queue_pairs;
99 
100 	/* I like... big packets and I cannot lie! */
101 	bool big_packets;
102 
103 	/* Host will merge rx buffers for big packets (shake it! shake it!) */
104 	bool mergeable_rx_bufs;
105 
106 	/* Has control virtqueue */
107 	bool has_cvq;
108 
109 	/* enable config space updates */
110 	bool config_enable;
111 
112 	/* Active statistics */
113 	struct virtnet_stats __percpu *stats;
114 
115 	/* Work struct for refilling if we run low on memory. */
116 	struct delayed_work refill;
117 
118 	/* Work struct for config space updates */
119 	struct work_struct config_work;
120 
121 	/* Lock for config space updates */
122 	struct mutex config_lock;
123 
124 	/* Does the affinity hint is set for virtqueues? */
125 	bool affinity_hint_set;
126 };
127 
128 struct skb_vnet_hdr {
129 	union {
130 		struct virtio_net_hdr hdr;
131 		struct virtio_net_hdr_mrg_rxbuf mhdr;
132 	};
133 };
134 
135 struct padded_vnet_hdr {
136 	struct virtio_net_hdr hdr;
137 	/*
138 	 * virtio_net_hdr should be in a separated sg buffer because of a
139 	 * QEMU bug, and data sg buffer shares same page with this header sg.
140 	 * This padding makes next sg 16 byte aligned after virtio_net_hdr.
141 	 */
142 	char padding[6];
143 };
144 
145 /* Converting between virtqueue no. and kernel tx/rx queue no.
146  * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq
147  */
148 static int vq2txq(struct virtqueue *vq)
149 {
150 	return (virtqueue_get_queue_index(vq) - 1) / 2;
151 }
152 
153 static int txq2vq(int txq)
154 {
155 	return txq * 2 + 1;
156 }
157 
158 static int vq2rxq(struct virtqueue *vq)
159 {
160 	return virtqueue_get_queue_index(vq) / 2;
161 }
162 
163 static int rxq2vq(int rxq)
164 {
165 	return rxq * 2;
166 }
167 
168 static inline struct skb_vnet_hdr *skb_vnet_hdr(struct sk_buff *skb)
169 {
170 	return (struct skb_vnet_hdr *)skb->cb;
171 }
172 
173 /*
174  * private is used to chain pages for big packets, put the whole
175  * most recent used list in the beginning for reuse
176  */
177 static void give_pages(struct receive_queue *rq, struct page *page)
178 {
179 	struct page *end;
180 
181 	/* Find end of list, sew whole thing into vi->rq.pages. */
182 	for (end = page; end->private; end = (struct page *)end->private);
183 	end->private = (unsigned long)rq->pages;
184 	rq->pages = page;
185 }
186 
187 static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask)
188 {
189 	struct page *p = rq->pages;
190 
191 	if (p) {
192 		rq->pages = (struct page *)p->private;
193 		/* clear private here, it is used to chain pages */
194 		p->private = 0;
195 	} else
196 		p = alloc_page(gfp_mask);
197 	return p;
198 }
199 
200 static void skb_xmit_done(struct virtqueue *vq)
201 {
202 	struct virtnet_info *vi = vq->vdev->priv;
203 
204 	/* Suppress further interrupts. */
205 	virtqueue_disable_cb(vq);
206 
207 	/* We were probably waiting for more output buffers. */
208 	netif_wake_subqueue(vi->dev, vq2txq(vq));
209 }
210 
211 static void set_skb_frag(struct sk_buff *skb, struct page *page,
212 			 unsigned int offset, unsigned int *len)
213 {
214 	int size = min((unsigned)PAGE_SIZE - offset, *len);
215 	int i = skb_shinfo(skb)->nr_frags;
216 
217 	__skb_fill_page_desc(skb, i, page, offset, size);
218 
219 	skb->data_len += size;
220 	skb->len += size;
221 	skb->truesize += PAGE_SIZE;
222 	skb_shinfo(skb)->nr_frags++;
223 	*len -= size;
224 }
225 
226 /* Called from bottom half context */
227 static struct sk_buff *page_to_skb(struct receive_queue *rq,
228 				   struct page *page, unsigned int len)
229 {
230 	struct virtnet_info *vi = rq->vq->vdev->priv;
231 	struct sk_buff *skb;
232 	struct skb_vnet_hdr *hdr;
233 	unsigned int copy, hdr_len, offset;
234 	char *p;
235 
236 	p = page_address(page);
237 
238 	/* copy small packet so we can reuse these pages for small data */
239 	skb = netdev_alloc_skb_ip_align(vi->dev, GOOD_COPY_LEN);
240 	if (unlikely(!skb))
241 		return NULL;
242 
243 	hdr = skb_vnet_hdr(skb);
244 
245 	if (vi->mergeable_rx_bufs) {
246 		hdr_len = sizeof hdr->mhdr;
247 		offset = hdr_len;
248 	} else {
249 		hdr_len = sizeof hdr->hdr;
250 		offset = sizeof(struct padded_vnet_hdr);
251 	}
252 
253 	memcpy(hdr, p, hdr_len);
254 
255 	len -= hdr_len;
256 	p += offset;
257 
258 	copy = len;
259 	if (copy > skb_tailroom(skb))
260 		copy = skb_tailroom(skb);
261 	memcpy(skb_put(skb, copy), p, copy);
262 
263 	len -= copy;
264 	offset += copy;
265 
266 	/*
267 	 * Verify that we can indeed put this data into a skb.
268 	 * This is here to handle cases when the device erroneously
269 	 * tries to receive more than is possible. This is usually
270 	 * the case of a broken device.
271 	 */
272 	if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) {
273 		net_dbg_ratelimited("%s: too much data\n", skb->dev->name);
274 		dev_kfree_skb(skb);
275 		return NULL;
276 	}
277 
278 	while (len) {
279 		set_skb_frag(skb, page, offset, &len);
280 		page = (struct page *)page->private;
281 		offset = 0;
282 	}
283 
284 	if (page)
285 		give_pages(rq, page);
286 
287 	return skb;
288 }
289 
290 static int receive_mergeable(struct receive_queue *rq, struct sk_buff *skb)
291 {
292 	struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
293 	struct page *page;
294 	int num_buf, i, len;
295 
296 	num_buf = hdr->mhdr.num_buffers;
297 	while (--num_buf) {
298 		i = skb_shinfo(skb)->nr_frags;
299 		if (i >= MAX_SKB_FRAGS) {
300 			pr_debug("%s: packet too long\n", skb->dev->name);
301 			skb->dev->stats.rx_length_errors++;
302 			return -EINVAL;
303 		}
304 		page = virtqueue_get_buf(rq->vq, &len);
305 		if (!page) {
306 			pr_debug("%s: rx error: %d buffers missing\n",
307 				 skb->dev->name, hdr->mhdr.num_buffers);
308 			skb->dev->stats.rx_length_errors++;
309 			return -EINVAL;
310 		}
311 
312 		if (len > PAGE_SIZE)
313 			len = PAGE_SIZE;
314 
315 		set_skb_frag(skb, page, 0, &len);
316 
317 		--rq->num;
318 	}
319 	return 0;
320 }
321 
322 static void receive_buf(struct receive_queue *rq, void *buf, unsigned int len)
323 {
324 	struct virtnet_info *vi = rq->vq->vdev->priv;
325 	struct net_device *dev = vi->dev;
326 	struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
327 	struct sk_buff *skb;
328 	struct page *page;
329 	struct skb_vnet_hdr *hdr;
330 
331 	if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
332 		pr_debug("%s: short packet %i\n", dev->name, len);
333 		dev->stats.rx_length_errors++;
334 		if (vi->mergeable_rx_bufs || vi->big_packets)
335 			give_pages(rq, buf);
336 		else
337 			dev_kfree_skb(buf);
338 		return;
339 	}
340 
341 	if (!vi->mergeable_rx_bufs && !vi->big_packets) {
342 		skb = buf;
343 		len -= sizeof(struct virtio_net_hdr);
344 		skb_trim(skb, len);
345 	} else {
346 		page = buf;
347 		skb = page_to_skb(rq, page, len);
348 		if (unlikely(!skb)) {
349 			dev->stats.rx_dropped++;
350 			give_pages(rq, page);
351 			return;
352 		}
353 		if (vi->mergeable_rx_bufs)
354 			if (receive_mergeable(rq, skb)) {
355 				dev_kfree_skb(skb);
356 				return;
357 			}
358 	}
359 
360 	hdr = skb_vnet_hdr(skb);
361 
362 	u64_stats_update_begin(&stats->rx_syncp);
363 	stats->rx_bytes += skb->len;
364 	stats->rx_packets++;
365 	u64_stats_update_end(&stats->rx_syncp);
366 
367 	if (hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
368 		pr_debug("Needs csum!\n");
369 		if (!skb_partial_csum_set(skb,
370 					  hdr->hdr.csum_start,
371 					  hdr->hdr.csum_offset))
372 			goto frame_err;
373 	} else if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) {
374 		skb->ip_summed = CHECKSUM_UNNECESSARY;
375 	}
376 
377 	skb->protocol = eth_type_trans(skb, dev);
378 	pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
379 		 ntohs(skb->protocol), skb->len, skb->pkt_type);
380 
381 	if (hdr->hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
382 		pr_debug("GSO!\n");
383 		switch (hdr->hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
384 		case VIRTIO_NET_HDR_GSO_TCPV4:
385 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
386 			break;
387 		case VIRTIO_NET_HDR_GSO_UDP:
388 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
389 			break;
390 		case VIRTIO_NET_HDR_GSO_TCPV6:
391 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
392 			break;
393 		default:
394 			net_warn_ratelimited("%s: bad gso type %u.\n",
395 					     dev->name, hdr->hdr.gso_type);
396 			goto frame_err;
397 		}
398 
399 		if (hdr->hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
400 			skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
401 
402 		skb_shinfo(skb)->gso_size = hdr->hdr.gso_size;
403 		if (skb_shinfo(skb)->gso_size == 0) {
404 			net_warn_ratelimited("%s: zero gso size.\n", dev->name);
405 			goto frame_err;
406 		}
407 
408 		/* Header must be checked, and gso_segs computed. */
409 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
410 		skb_shinfo(skb)->gso_segs = 0;
411 	}
412 
413 	netif_receive_skb(skb);
414 	return;
415 
416 frame_err:
417 	dev->stats.rx_frame_errors++;
418 	dev_kfree_skb(skb);
419 }
420 
421 static int add_recvbuf_small(struct receive_queue *rq, gfp_t gfp)
422 {
423 	struct virtnet_info *vi = rq->vq->vdev->priv;
424 	struct sk_buff *skb;
425 	struct skb_vnet_hdr *hdr;
426 	int err;
427 
428 	skb = __netdev_alloc_skb_ip_align(vi->dev, MAX_PACKET_LEN, gfp);
429 	if (unlikely(!skb))
430 		return -ENOMEM;
431 
432 	skb_put(skb, MAX_PACKET_LEN);
433 
434 	hdr = skb_vnet_hdr(skb);
435 	sg_set_buf(rq->sg, &hdr->hdr, sizeof hdr->hdr);
436 
437 	skb_to_sgvec(skb, rq->sg + 1, 0, skb->len);
438 
439 	err = virtqueue_add_buf(rq->vq, rq->sg, 0, 2, skb, gfp);
440 	if (err < 0)
441 		dev_kfree_skb(skb);
442 
443 	return err;
444 }
445 
446 static int add_recvbuf_big(struct receive_queue *rq, gfp_t gfp)
447 {
448 	struct page *first, *list = NULL;
449 	char *p;
450 	int i, err, offset;
451 
452 	/* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */
453 	for (i = MAX_SKB_FRAGS + 1; i > 1; --i) {
454 		first = get_a_page(rq, gfp);
455 		if (!first) {
456 			if (list)
457 				give_pages(rq, list);
458 			return -ENOMEM;
459 		}
460 		sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE);
461 
462 		/* chain new page in list head to match sg */
463 		first->private = (unsigned long)list;
464 		list = first;
465 	}
466 
467 	first = get_a_page(rq, gfp);
468 	if (!first) {
469 		give_pages(rq, list);
470 		return -ENOMEM;
471 	}
472 	p = page_address(first);
473 
474 	/* rq->sg[0], rq->sg[1] share the same page */
475 	/* a separated rq->sg[0] for virtio_net_hdr only due to QEMU bug */
476 	sg_set_buf(&rq->sg[0], p, sizeof(struct virtio_net_hdr));
477 
478 	/* rq->sg[1] for data packet, from offset */
479 	offset = sizeof(struct padded_vnet_hdr);
480 	sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset);
481 
482 	/* chain first in list head */
483 	first->private = (unsigned long)list;
484 	err = virtqueue_add_buf(rq->vq, rq->sg, 0, MAX_SKB_FRAGS + 2,
485 				first, gfp);
486 	if (err < 0)
487 		give_pages(rq, first);
488 
489 	return err;
490 }
491 
492 static int add_recvbuf_mergeable(struct receive_queue *rq, gfp_t gfp)
493 {
494 	struct page *page;
495 	int err;
496 
497 	page = get_a_page(rq, gfp);
498 	if (!page)
499 		return -ENOMEM;
500 
501 	sg_init_one(rq->sg, page_address(page), PAGE_SIZE);
502 
503 	err = virtqueue_add_buf(rq->vq, rq->sg, 0, 1, page, gfp);
504 	if (err < 0)
505 		give_pages(rq, page);
506 
507 	return err;
508 }
509 
510 /*
511  * Returns false if we couldn't fill entirely (OOM).
512  *
513  * Normally run in the receive path, but can also be run from ndo_open
514  * before we're receiving packets, or from refill_work which is
515  * careful to disable receiving (using napi_disable).
516  */
517 static bool try_fill_recv(struct receive_queue *rq, gfp_t gfp)
518 {
519 	struct virtnet_info *vi = rq->vq->vdev->priv;
520 	int err;
521 	bool oom;
522 
523 	do {
524 		if (vi->mergeable_rx_bufs)
525 			err = add_recvbuf_mergeable(rq, gfp);
526 		else if (vi->big_packets)
527 			err = add_recvbuf_big(rq, gfp);
528 		else
529 			err = add_recvbuf_small(rq, gfp);
530 
531 		oom = err == -ENOMEM;
532 		if (err)
533 			break;
534 		++rq->num;
535 	} while (rq->vq->num_free);
536 	if (unlikely(rq->num > rq->max))
537 		rq->max = rq->num;
538 	virtqueue_kick(rq->vq);
539 	return !oom;
540 }
541 
542 static void skb_recv_done(struct virtqueue *rvq)
543 {
544 	struct virtnet_info *vi = rvq->vdev->priv;
545 	struct receive_queue *rq = &vi->rq[vq2rxq(rvq)];
546 
547 	/* Schedule NAPI, Suppress further interrupts if successful. */
548 	if (napi_schedule_prep(&rq->napi)) {
549 		virtqueue_disable_cb(rvq);
550 		__napi_schedule(&rq->napi);
551 	}
552 }
553 
554 static void virtnet_napi_enable(struct receive_queue *rq)
555 {
556 	napi_enable(&rq->napi);
557 
558 	/* If all buffers were filled by other side before we napi_enabled, we
559 	 * won't get another interrupt, so process any outstanding packets
560 	 * now.  virtnet_poll wants re-enable the queue, so we disable here.
561 	 * We synchronize against interrupts via NAPI_STATE_SCHED */
562 	if (napi_schedule_prep(&rq->napi)) {
563 		virtqueue_disable_cb(rq->vq);
564 		local_bh_disable();
565 		__napi_schedule(&rq->napi);
566 		local_bh_enable();
567 	}
568 }
569 
570 static void refill_work(struct work_struct *work)
571 {
572 	struct virtnet_info *vi =
573 		container_of(work, struct virtnet_info, refill.work);
574 	bool still_empty;
575 	int i;
576 
577 	for (i = 0; i < vi->max_queue_pairs; i++) {
578 		struct receive_queue *rq = &vi->rq[i];
579 
580 		napi_disable(&rq->napi);
581 		still_empty = !try_fill_recv(rq, GFP_KERNEL);
582 		virtnet_napi_enable(rq);
583 
584 		/* In theory, this can happen: if we don't get any buffers in
585 		 * we will *never* try to fill again.
586 		 */
587 		if (still_empty)
588 			schedule_delayed_work(&vi->refill, HZ/2);
589 	}
590 }
591 
592 static int virtnet_poll(struct napi_struct *napi, int budget)
593 {
594 	struct receive_queue *rq =
595 		container_of(napi, struct receive_queue, napi);
596 	struct virtnet_info *vi = rq->vq->vdev->priv;
597 	void *buf;
598 	unsigned int len, received = 0;
599 
600 again:
601 	while (received < budget &&
602 	       (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) {
603 		receive_buf(rq, buf, len);
604 		--rq->num;
605 		received++;
606 	}
607 
608 	if (rq->num < rq->max / 2) {
609 		if (!try_fill_recv(rq, GFP_ATOMIC))
610 			schedule_delayed_work(&vi->refill, 0);
611 	}
612 
613 	/* Out of packets? */
614 	if (received < budget) {
615 		napi_complete(napi);
616 		if (unlikely(!virtqueue_enable_cb(rq->vq)) &&
617 		    napi_schedule_prep(napi)) {
618 			virtqueue_disable_cb(rq->vq);
619 			__napi_schedule(napi);
620 			goto again;
621 		}
622 	}
623 
624 	return received;
625 }
626 
627 static int virtnet_open(struct net_device *dev)
628 {
629 	struct virtnet_info *vi = netdev_priv(dev);
630 	int i;
631 
632 	for (i = 0; i < vi->max_queue_pairs; i++) {
633 		/* Make sure we have some buffers: if oom use wq. */
634 		if (!try_fill_recv(&vi->rq[i], GFP_KERNEL))
635 			schedule_delayed_work(&vi->refill, 0);
636 		virtnet_napi_enable(&vi->rq[i]);
637 	}
638 
639 	return 0;
640 }
641 
642 static void free_old_xmit_skbs(struct send_queue *sq)
643 {
644 	struct sk_buff *skb;
645 	unsigned int len;
646 	struct virtnet_info *vi = sq->vq->vdev->priv;
647 	struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
648 
649 	while ((skb = virtqueue_get_buf(sq->vq, &len)) != NULL) {
650 		pr_debug("Sent skb %p\n", skb);
651 
652 		u64_stats_update_begin(&stats->tx_syncp);
653 		stats->tx_bytes += skb->len;
654 		stats->tx_packets++;
655 		u64_stats_update_end(&stats->tx_syncp);
656 
657 		dev_kfree_skb_any(skb);
658 	}
659 }
660 
661 static int xmit_skb(struct send_queue *sq, struct sk_buff *skb)
662 {
663 	struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
664 	const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
665 	struct virtnet_info *vi = sq->vq->vdev->priv;
666 	unsigned num_sg;
667 
668 	pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
669 
670 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
671 		hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
672 		hdr->hdr.csum_start = skb_checksum_start_offset(skb);
673 		hdr->hdr.csum_offset = skb->csum_offset;
674 	} else {
675 		hdr->hdr.flags = 0;
676 		hdr->hdr.csum_offset = hdr->hdr.csum_start = 0;
677 	}
678 
679 	if (skb_is_gso(skb)) {
680 		hdr->hdr.hdr_len = skb_headlen(skb);
681 		hdr->hdr.gso_size = skb_shinfo(skb)->gso_size;
682 		if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
683 			hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
684 		else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
685 			hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
686 		else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
687 			hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
688 		else
689 			BUG();
690 		if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
691 			hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
692 	} else {
693 		hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
694 		hdr->hdr.gso_size = hdr->hdr.hdr_len = 0;
695 	}
696 
697 	hdr->mhdr.num_buffers = 0;
698 
699 	/* Encode metadata header at front. */
700 	if (vi->mergeable_rx_bufs)
701 		sg_set_buf(sq->sg, &hdr->mhdr, sizeof hdr->mhdr);
702 	else
703 		sg_set_buf(sq->sg, &hdr->hdr, sizeof hdr->hdr);
704 
705 	num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len) + 1;
706 	return virtqueue_add_buf(sq->vq, sq->sg, num_sg,
707 				 0, skb, GFP_ATOMIC);
708 }
709 
710 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
711 {
712 	struct virtnet_info *vi = netdev_priv(dev);
713 	int qnum = skb_get_queue_mapping(skb);
714 	struct send_queue *sq = &vi->sq[qnum];
715 	int err;
716 
717 	/* Free up any pending old buffers before queueing new ones. */
718 	free_old_xmit_skbs(sq);
719 
720 	/* Try to transmit */
721 	err = xmit_skb(sq, skb);
722 
723 	/* This should not happen! */
724 	if (unlikely(err)) {
725 		dev->stats.tx_fifo_errors++;
726 		if (net_ratelimit())
727 			dev_warn(&dev->dev,
728 				 "Unexpected TXQ (%d) queue failure: %d\n", qnum, err);
729 		dev->stats.tx_dropped++;
730 		kfree_skb(skb);
731 		return NETDEV_TX_OK;
732 	}
733 	virtqueue_kick(sq->vq);
734 
735 	/* Don't wait up for transmitted skbs to be freed. */
736 	skb_orphan(skb);
737 	nf_reset(skb);
738 
739 	/* Apparently nice girls don't return TX_BUSY; stop the queue
740 	 * before it gets out of hand.  Naturally, this wastes entries. */
741 	if (sq->vq->num_free < 2+MAX_SKB_FRAGS) {
742 		netif_stop_subqueue(dev, qnum);
743 		if (unlikely(!virtqueue_enable_cb_delayed(sq->vq))) {
744 			/* More just got used, free them then recheck. */
745 			free_old_xmit_skbs(sq);
746 			if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) {
747 				netif_start_subqueue(dev, qnum);
748 				virtqueue_disable_cb(sq->vq);
749 			}
750 		}
751 	}
752 
753 	return NETDEV_TX_OK;
754 }
755 
756 static int virtnet_set_mac_address(struct net_device *dev, void *p)
757 {
758 	struct virtnet_info *vi = netdev_priv(dev);
759 	struct virtio_device *vdev = vi->vdev;
760 	int ret;
761 
762 	ret = eth_mac_addr(dev, p);
763 	if (ret)
764 		return ret;
765 
766 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
767 		vdev->config->set(vdev, offsetof(struct virtio_net_config, mac),
768 		                  dev->dev_addr, dev->addr_len);
769 
770 	return 0;
771 }
772 
773 static struct rtnl_link_stats64 *virtnet_stats(struct net_device *dev,
774 					       struct rtnl_link_stats64 *tot)
775 {
776 	struct virtnet_info *vi = netdev_priv(dev);
777 	int cpu;
778 	unsigned int start;
779 
780 	for_each_possible_cpu(cpu) {
781 		struct virtnet_stats *stats = per_cpu_ptr(vi->stats, cpu);
782 		u64 tpackets, tbytes, rpackets, rbytes;
783 
784 		do {
785 			start = u64_stats_fetch_begin_bh(&stats->tx_syncp);
786 			tpackets = stats->tx_packets;
787 			tbytes   = stats->tx_bytes;
788 		} while (u64_stats_fetch_retry_bh(&stats->tx_syncp, start));
789 
790 		do {
791 			start = u64_stats_fetch_begin_bh(&stats->rx_syncp);
792 			rpackets = stats->rx_packets;
793 			rbytes   = stats->rx_bytes;
794 		} while (u64_stats_fetch_retry_bh(&stats->rx_syncp, start));
795 
796 		tot->rx_packets += rpackets;
797 		tot->tx_packets += tpackets;
798 		tot->rx_bytes   += rbytes;
799 		tot->tx_bytes   += tbytes;
800 	}
801 
802 	tot->tx_dropped = dev->stats.tx_dropped;
803 	tot->tx_fifo_errors = dev->stats.tx_fifo_errors;
804 	tot->rx_dropped = dev->stats.rx_dropped;
805 	tot->rx_length_errors = dev->stats.rx_length_errors;
806 	tot->rx_frame_errors = dev->stats.rx_frame_errors;
807 
808 	return tot;
809 }
810 
811 #ifdef CONFIG_NET_POLL_CONTROLLER
812 static void virtnet_netpoll(struct net_device *dev)
813 {
814 	struct virtnet_info *vi = netdev_priv(dev);
815 	int i;
816 
817 	for (i = 0; i < vi->curr_queue_pairs; i++)
818 		napi_schedule(&vi->rq[i].napi);
819 }
820 #endif
821 
822 /*
823  * Send command via the control virtqueue and check status.  Commands
824  * supported by the hypervisor, as indicated by feature bits, should
825  * never fail unless improperly formated.
826  */
827 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
828 				 struct scatterlist *data, int out, int in)
829 {
830 	struct scatterlist *s, sg[VIRTNET_SEND_COMMAND_SG_MAX + 2];
831 	struct virtio_net_ctrl_hdr ctrl;
832 	virtio_net_ctrl_ack status = ~0;
833 	unsigned int tmp;
834 	int i;
835 
836 	/* Caller should know better */
837 	BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ||
838 		(out + in > VIRTNET_SEND_COMMAND_SG_MAX));
839 
840 	out++; /* Add header */
841 	in++; /* Add return status */
842 
843 	ctrl.class = class;
844 	ctrl.cmd = cmd;
845 
846 	sg_init_table(sg, out + in);
847 
848 	sg_set_buf(&sg[0], &ctrl, sizeof(ctrl));
849 	for_each_sg(data, s, out + in - 2, i)
850 		sg_set_buf(&sg[i + 1], sg_virt(s), s->length);
851 	sg_set_buf(&sg[out + in - 1], &status, sizeof(status));
852 
853 	BUG_ON(virtqueue_add_buf(vi->cvq, sg, out, in, vi, GFP_ATOMIC) < 0);
854 
855 	virtqueue_kick(vi->cvq);
856 
857 	/*
858 	 * Spin for a response, the kick causes an ioport write, trapping
859 	 * into the hypervisor, so the request should be handled immediately.
860 	 */
861 	while (!virtqueue_get_buf(vi->cvq, &tmp))
862 		cpu_relax();
863 
864 	return status == VIRTIO_NET_OK;
865 }
866 
867 static void virtnet_ack_link_announce(struct virtnet_info *vi)
868 {
869 	rtnl_lock();
870 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE,
871 				  VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL,
872 				  0, 0))
873 		dev_warn(&vi->dev->dev, "Failed to ack link announce.\n");
874 	rtnl_unlock();
875 }
876 
877 static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs)
878 {
879 	struct scatterlist sg;
880 	struct virtio_net_ctrl_mq s;
881 	struct net_device *dev = vi->dev;
882 
883 	if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ))
884 		return 0;
885 
886 	s.virtqueue_pairs = queue_pairs;
887 	sg_init_one(&sg, &s, sizeof(s));
888 
889 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ,
890 				  VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg, 1, 0)){
891 		dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n",
892 			 queue_pairs);
893 		return -EINVAL;
894 	} else
895 		vi->curr_queue_pairs = queue_pairs;
896 
897 	return 0;
898 }
899 
900 static int virtnet_close(struct net_device *dev)
901 {
902 	struct virtnet_info *vi = netdev_priv(dev);
903 	int i;
904 
905 	/* Make sure refill_work doesn't re-enable napi! */
906 	cancel_delayed_work_sync(&vi->refill);
907 
908 	for (i = 0; i < vi->max_queue_pairs; i++)
909 		napi_disable(&vi->rq[i].napi);
910 
911 	return 0;
912 }
913 
914 static void virtnet_set_rx_mode(struct net_device *dev)
915 {
916 	struct virtnet_info *vi = netdev_priv(dev);
917 	struct scatterlist sg[2];
918 	u8 promisc, allmulti;
919 	struct virtio_net_ctrl_mac *mac_data;
920 	struct netdev_hw_addr *ha;
921 	int uc_count;
922 	int mc_count;
923 	void *buf;
924 	int i;
925 
926 	/* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */
927 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
928 		return;
929 
930 	promisc = ((dev->flags & IFF_PROMISC) != 0);
931 	allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
932 
933 	sg_init_one(sg, &promisc, sizeof(promisc));
934 
935 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
936 				  VIRTIO_NET_CTRL_RX_PROMISC,
937 				  sg, 1, 0))
938 		dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
939 			 promisc ? "en" : "dis");
940 
941 	sg_init_one(sg, &allmulti, sizeof(allmulti));
942 
943 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
944 				  VIRTIO_NET_CTRL_RX_ALLMULTI,
945 				  sg, 1, 0))
946 		dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
947 			 allmulti ? "en" : "dis");
948 
949 	uc_count = netdev_uc_count(dev);
950 	mc_count = netdev_mc_count(dev);
951 	/* MAC filter - use one buffer for both lists */
952 	buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
953 		      (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
954 	mac_data = buf;
955 	if (!buf) {
956 		dev_warn(&dev->dev, "No memory for MAC address buffer\n");
957 		return;
958 	}
959 
960 	sg_init_table(sg, 2);
961 
962 	/* Store the unicast list and count in the front of the buffer */
963 	mac_data->entries = uc_count;
964 	i = 0;
965 	netdev_for_each_uc_addr(ha, dev)
966 		memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
967 
968 	sg_set_buf(&sg[0], mac_data,
969 		   sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
970 
971 	/* multicast list and count fill the end */
972 	mac_data = (void *)&mac_data->macs[uc_count][0];
973 
974 	mac_data->entries = mc_count;
975 	i = 0;
976 	netdev_for_each_mc_addr(ha, dev)
977 		memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
978 
979 	sg_set_buf(&sg[1], mac_data,
980 		   sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
981 
982 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
983 				  VIRTIO_NET_CTRL_MAC_TABLE_SET,
984 				  sg, 2, 0))
985 		dev_warn(&dev->dev, "Failed to set MAC fitler table.\n");
986 
987 	kfree(buf);
988 }
989 
990 static int virtnet_vlan_rx_add_vid(struct net_device *dev, u16 vid)
991 {
992 	struct virtnet_info *vi = netdev_priv(dev);
993 	struct scatterlist sg;
994 
995 	sg_init_one(&sg, &vid, sizeof(vid));
996 
997 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
998 				  VIRTIO_NET_CTRL_VLAN_ADD, &sg, 1, 0))
999 		dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
1000 	return 0;
1001 }
1002 
1003 static int virtnet_vlan_rx_kill_vid(struct net_device *dev, u16 vid)
1004 {
1005 	struct virtnet_info *vi = netdev_priv(dev);
1006 	struct scatterlist sg;
1007 
1008 	sg_init_one(&sg, &vid, sizeof(vid));
1009 
1010 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1011 				  VIRTIO_NET_CTRL_VLAN_DEL, &sg, 1, 0))
1012 		dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
1013 	return 0;
1014 }
1015 
1016 static void virtnet_set_affinity(struct virtnet_info *vi, bool set)
1017 {
1018 	int i;
1019 
1020 	/* In multiqueue mode, when the number of cpu is equal to the number of
1021 	 * queue pairs, we let the queue pairs to be private to one cpu by
1022 	 * setting the affinity hint to eliminate the contention.
1023 	 */
1024 	if ((vi->curr_queue_pairs == 1 ||
1025 	     vi->max_queue_pairs != num_online_cpus()) && set) {
1026 		if (vi->affinity_hint_set)
1027 			set = false;
1028 		else
1029 			return;
1030 	}
1031 
1032 	for (i = 0; i < vi->max_queue_pairs; i++) {
1033 		int cpu = set ? i : -1;
1034 		virtqueue_set_affinity(vi->rq[i].vq, cpu);
1035 		virtqueue_set_affinity(vi->sq[i].vq, cpu);
1036 	}
1037 
1038 	if (set)
1039 		vi->affinity_hint_set = true;
1040 	else
1041 		vi->affinity_hint_set = false;
1042 }
1043 
1044 static void virtnet_get_ringparam(struct net_device *dev,
1045 				struct ethtool_ringparam *ring)
1046 {
1047 	struct virtnet_info *vi = netdev_priv(dev);
1048 
1049 	ring->rx_max_pending = virtqueue_get_vring_size(vi->rq[0].vq);
1050 	ring->tx_max_pending = virtqueue_get_vring_size(vi->sq[0].vq);
1051 	ring->rx_pending = ring->rx_max_pending;
1052 	ring->tx_pending = ring->tx_max_pending;
1053 }
1054 
1055 
1056 static void virtnet_get_drvinfo(struct net_device *dev,
1057 				struct ethtool_drvinfo *info)
1058 {
1059 	struct virtnet_info *vi = netdev_priv(dev);
1060 	struct virtio_device *vdev = vi->vdev;
1061 
1062 	strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
1063 	strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version));
1064 	strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info));
1065 
1066 }
1067 
1068 /* TODO: Eliminate OOO packets during switching */
1069 static int virtnet_set_channels(struct net_device *dev,
1070 				struct ethtool_channels *channels)
1071 {
1072 	struct virtnet_info *vi = netdev_priv(dev);
1073 	u16 queue_pairs = channels->combined_count;
1074 	int err;
1075 
1076 	/* We don't support separate rx/tx channels.
1077 	 * We don't allow setting 'other' channels.
1078 	 */
1079 	if (channels->rx_count || channels->tx_count || channels->other_count)
1080 		return -EINVAL;
1081 
1082 	if (queue_pairs > vi->max_queue_pairs)
1083 		return -EINVAL;
1084 
1085 	err = virtnet_set_queues(vi, queue_pairs);
1086 	if (!err) {
1087 		netif_set_real_num_tx_queues(dev, queue_pairs);
1088 		netif_set_real_num_rx_queues(dev, queue_pairs);
1089 
1090 		virtnet_set_affinity(vi, true);
1091 	}
1092 
1093 	return err;
1094 }
1095 
1096 static void virtnet_get_channels(struct net_device *dev,
1097 				 struct ethtool_channels *channels)
1098 {
1099 	struct virtnet_info *vi = netdev_priv(dev);
1100 
1101 	channels->combined_count = vi->curr_queue_pairs;
1102 	channels->max_combined = vi->max_queue_pairs;
1103 	channels->max_other = 0;
1104 	channels->rx_count = 0;
1105 	channels->tx_count = 0;
1106 	channels->other_count = 0;
1107 }
1108 
1109 static const struct ethtool_ops virtnet_ethtool_ops = {
1110 	.get_drvinfo = virtnet_get_drvinfo,
1111 	.get_link = ethtool_op_get_link,
1112 	.get_ringparam = virtnet_get_ringparam,
1113 	.set_channels = virtnet_set_channels,
1114 	.get_channels = virtnet_get_channels,
1115 };
1116 
1117 #define MIN_MTU 68
1118 #define MAX_MTU 65535
1119 
1120 static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
1121 {
1122 	if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
1123 		return -EINVAL;
1124 	dev->mtu = new_mtu;
1125 	return 0;
1126 }
1127 
1128 /* To avoid contending a lock hold by a vcpu who would exit to host, select the
1129  * txq based on the processor id.
1130  * TODO: handle cpu hotplug.
1131  */
1132 static u16 virtnet_select_queue(struct net_device *dev, struct sk_buff *skb)
1133 {
1134 	int txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
1135 		  smp_processor_id();
1136 
1137 	while (unlikely(txq >= dev->real_num_tx_queues))
1138 		txq -= dev->real_num_tx_queues;
1139 
1140 	return txq;
1141 }
1142 
1143 static const struct net_device_ops virtnet_netdev = {
1144 	.ndo_open            = virtnet_open,
1145 	.ndo_stop   	     = virtnet_close,
1146 	.ndo_start_xmit      = start_xmit,
1147 	.ndo_validate_addr   = eth_validate_addr,
1148 	.ndo_set_mac_address = virtnet_set_mac_address,
1149 	.ndo_set_rx_mode     = virtnet_set_rx_mode,
1150 	.ndo_change_mtu	     = virtnet_change_mtu,
1151 	.ndo_get_stats64     = virtnet_stats,
1152 	.ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
1153 	.ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
1154 	.ndo_select_queue     = virtnet_select_queue,
1155 #ifdef CONFIG_NET_POLL_CONTROLLER
1156 	.ndo_poll_controller = virtnet_netpoll,
1157 #endif
1158 };
1159 
1160 static void virtnet_config_changed_work(struct work_struct *work)
1161 {
1162 	struct virtnet_info *vi =
1163 		container_of(work, struct virtnet_info, config_work);
1164 	u16 v;
1165 
1166 	mutex_lock(&vi->config_lock);
1167 	if (!vi->config_enable)
1168 		goto done;
1169 
1170 	if (virtio_config_val(vi->vdev, VIRTIO_NET_F_STATUS,
1171 			      offsetof(struct virtio_net_config, status),
1172 			      &v) < 0)
1173 		goto done;
1174 
1175 	if (v & VIRTIO_NET_S_ANNOUNCE) {
1176 		netdev_notify_peers(vi->dev);
1177 		virtnet_ack_link_announce(vi);
1178 	}
1179 
1180 	/* Ignore unknown (future) status bits */
1181 	v &= VIRTIO_NET_S_LINK_UP;
1182 
1183 	if (vi->status == v)
1184 		goto done;
1185 
1186 	vi->status = v;
1187 
1188 	if (vi->status & VIRTIO_NET_S_LINK_UP) {
1189 		netif_carrier_on(vi->dev);
1190 		netif_tx_wake_all_queues(vi->dev);
1191 	} else {
1192 		netif_carrier_off(vi->dev);
1193 		netif_tx_stop_all_queues(vi->dev);
1194 	}
1195 done:
1196 	mutex_unlock(&vi->config_lock);
1197 }
1198 
1199 static void virtnet_config_changed(struct virtio_device *vdev)
1200 {
1201 	struct virtnet_info *vi = vdev->priv;
1202 
1203 	schedule_work(&vi->config_work);
1204 }
1205 
1206 static void virtnet_free_queues(struct virtnet_info *vi)
1207 {
1208 	kfree(vi->rq);
1209 	kfree(vi->sq);
1210 }
1211 
1212 static void free_receive_bufs(struct virtnet_info *vi)
1213 {
1214 	int i;
1215 
1216 	for (i = 0; i < vi->max_queue_pairs; i++) {
1217 		while (vi->rq[i].pages)
1218 			__free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0);
1219 	}
1220 }
1221 
1222 static void free_unused_bufs(struct virtnet_info *vi)
1223 {
1224 	void *buf;
1225 	int i;
1226 
1227 	for (i = 0; i < vi->max_queue_pairs; i++) {
1228 		struct virtqueue *vq = vi->sq[i].vq;
1229 		while ((buf = virtqueue_detach_unused_buf(vq)) != NULL)
1230 			dev_kfree_skb(buf);
1231 	}
1232 
1233 	for (i = 0; i < vi->max_queue_pairs; i++) {
1234 		struct virtqueue *vq = vi->rq[i].vq;
1235 
1236 		while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) {
1237 			if (vi->mergeable_rx_bufs || vi->big_packets)
1238 				give_pages(&vi->rq[i], buf);
1239 			else
1240 				dev_kfree_skb(buf);
1241 			--vi->rq[i].num;
1242 		}
1243 		BUG_ON(vi->rq[i].num != 0);
1244 	}
1245 }
1246 
1247 static void virtnet_del_vqs(struct virtnet_info *vi)
1248 {
1249 	struct virtio_device *vdev = vi->vdev;
1250 
1251 	virtnet_set_affinity(vi, false);
1252 
1253 	vdev->config->del_vqs(vdev);
1254 
1255 	virtnet_free_queues(vi);
1256 }
1257 
1258 static int virtnet_find_vqs(struct virtnet_info *vi)
1259 {
1260 	vq_callback_t **callbacks;
1261 	struct virtqueue **vqs;
1262 	int ret = -ENOMEM;
1263 	int i, total_vqs;
1264 	const char **names;
1265 
1266 	/* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
1267 	 * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
1268 	 * possible control vq.
1269 	 */
1270 	total_vqs = vi->max_queue_pairs * 2 +
1271 		    virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ);
1272 
1273 	/* Allocate space for find_vqs parameters */
1274 	vqs = kzalloc(total_vqs * sizeof(*vqs), GFP_KERNEL);
1275 	if (!vqs)
1276 		goto err_vq;
1277 	callbacks = kmalloc(total_vqs * sizeof(*callbacks), GFP_KERNEL);
1278 	if (!callbacks)
1279 		goto err_callback;
1280 	names = kmalloc(total_vqs * sizeof(*names), GFP_KERNEL);
1281 	if (!names)
1282 		goto err_names;
1283 
1284 	/* Parameters for control virtqueue, if any */
1285 	if (vi->has_cvq) {
1286 		callbacks[total_vqs - 1] = NULL;
1287 		names[total_vqs - 1] = "control";
1288 	}
1289 
1290 	/* Allocate/initialize parameters for send/receive virtqueues */
1291 	for (i = 0; i < vi->max_queue_pairs; i++) {
1292 		callbacks[rxq2vq(i)] = skb_recv_done;
1293 		callbacks[txq2vq(i)] = skb_xmit_done;
1294 		sprintf(vi->rq[i].name, "input.%d", i);
1295 		sprintf(vi->sq[i].name, "output.%d", i);
1296 		names[rxq2vq(i)] = vi->rq[i].name;
1297 		names[txq2vq(i)] = vi->sq[i].name;
1298 	}
1299 
1300 	ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs, callbacks,
1301 					 names);
1302 	if (ret)
1303 		goto err_find;
1304 
1305 	if (vi->has_cvq) {
1306 		vi->cvq = vqs[total_vqs - 1];
1307 		if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
1308 			vi->dev->features |= NETIF_F_HW_VLAN_FILTER;
1309 	}
1310 
1311 	for (i = 0; i < vi->max_queue_pairs; i++) {
1312 		vi->rq[i].vq = vqs[rxq2vq(i)];
1313 		vi->sq[i].vq = vqs[txq2vq(i)];
1314 	}
1315 
1316 	kfree(names);
1317 	kfree(callbacks);
1318 	kfree(vqs);
1319 
1320 	return 0;
1321 
1322 err_find:
1323 	kfree(names);
1324 err_names:
1325 	kfree(callbacks);
1326 err_callback:
1327 	kfree(vqs);
1328 err_vq:
1329 	return ret;
1330 }
1331 
1332 static int virtnet_alloc_queues(struct virtnet_info *vi)
1333 {
1334 	int i;
1335 
1336 	vi->sq = kzalloc(sizeof(*vi->sq) * vi->max_queue_pairs, GFP_KERNEL);
1337 	if (!vi->sq)
1338 		goto err_sq;
1339 	vi->rq = kzalloc(sizeof(*vi->rq) * vi->max_queue_pairs, GFP_KERNEL);
1340 	if (!vi->rq)
1341 		goto err_rq;
1342 
1343 	INIT_DELAYED_WORK(&vi->refill, refill_work);
1344 	for (i = 0; i < vi->max_queue_pairs; i++) {
1345 		vi->rq[i].pages = NULL;
1346 		netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll,
1347 			       napi_weight);
1348 
1349 		sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg));
1350 		sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg));
1351 	}
1352 
1353 	return 0;
1354 
1355 err_rq:
1356 	kfree(vi->sq);
1357 err_sq:
1358 	return -ENOMEM;
1359 }
1360 
1361 static int init_vqs(struct virtnet_info *vi)
1362 {
1363 	int ret;
1364 
1365 	/* Allocate send & receive queues */
1366 	ret = virtnet_alloc_queues(vi);
1367 	if (ret)
1368 		goto err;
1369 
1370 	ret = virtnet_find_vqs(vi);
1371 	if (ret)
1372 		goto err_free;
1373 
1374 	virtnet_set_affinity(vi, true);
1375 	return 0;
1376 
1377 err_free:
1378 	virtnet_free_queues(vi);
1379 err:
1380 	return ret;
1381 }
1382 
1383 static int virtnet_probe(struct virtio_device *vdev)
1384 {
1385 	int i, err;
1386 	struct net_device *dev;
1387 	struct virtnet_info *vi;
1388 	u16 max_queue_pairs;
1389 
1390 	/* Find if host supports multiqueue virtio_net device */
1391 	err = virtio_config_val(vdev, VIRTIO_NET_F_MQ,
1392 				offsetof(struct virtio_net_config,
1393 				max_virtqueue_pairs), &max_queue_pairs);
1394 
1395 	/* We need at least 2 queue's */
1396 	if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
1397 	    max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
1398 	    !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
1399 		max_queue_pairs = 1;
1400 
1401 	/* Allocate ourselves a network device with room for our info */
1402 	dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs);
1403 	if (!dev)
1404 		return -ENOMEM;
1405 
1406 	/* Set up network device as normal. */
1407 	dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE;
1408 	dev->netdev_ops = &virtnet_netdev;
1409 	dev->features = NETIF_F_HIGHDMA;
1410 
1411 	SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
1412 	SET_NETDEV_DEV(dev, &vdev->dev);
1413 
1414 	/* Do we support "hardware" checksums? */
1415 	if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
1416 		/* This opens up the world of extra features. */
1417 		dev->hw_features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1418 		if (csum)
1419 			dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1420 
1421 		if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
1422 			dev->hw_features |= NETIF_F_TSO | NETIF_F_UFO
1423 				| NETIF_F_TSO_ECN | NETIF_F_TSO6;
1424 		}
1425 		/* Individual feature bits: what can host handle? */
1426 		if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
1427 			dev->hw_features |= NETIF_F_TSO;
1428 		if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
1429 			dev->hw_features |= NETIF_F_TSO6;
1430 		if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
1431 			dev->hw_features |= NETIF_F_TSO_ECN;
1432 		if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
1433 			dev->hw_features |= NETIF_F_UFO;
1434 
1435 		if (gso)
1436 			dev->features |= dev->hw_features & (NETIF_F_ALL_TSO|NETIF_F_UFO);
1437 		/* (!csum && gso) case will be fixed by register_netdev() */
1438 	}
1439 
1440 	/* Configuration may specify what MAC to use.  Otherwise random. */
1441 	if (virtio_config_val_len(vdev, VIRTIO_NET_F_MAC,
1442 				  offsetof(struct virtio_net_config, mac),
1443 				  dev->dev_addr, dev->addr_len) < 0)
1444 		eth_hw_addr_random(dev);
1445 
1446 	/* Set up our device-specific information */
1447 	vi = netdev_priv(dev);
1448 	vi->dev = dev;
1449 	vi->vdev = vdev;
1450 	vdev->priv = vi;
1451 	vi->stats = alloc_percpu(struct virtnet_stats);
1452 	err = -ENOMEM;
1453 	if (vi->stats == NULL)
1454 		goto free;
1455 
1456 	mutex_init(&vi->config_lock);
1457 	vi->config_enable = true;
1458 	INIT_WORK(&vi->config_work, virtnet_config_changed_work);
1459 
1460 	/* If we can receive ANY GSO packets, we must allocate large ones. */
1461 	if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
1462 	    virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
1463 	    virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
1464 		vi->big_packets = true;
1465 
1466 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
1467 		vi->mergeable_rx_bufs = true;
1468 
1469 	if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
1470 		vi->has_cvq = true;
1471 
1472 	/* Use single tx/rx queue pair as default */
1473 	vi->curr_queue_pairs = 1;
1474 	vi->max_queue_pairs = max_queue_pairs;
1475 
1476 	/* Allocate/initialize the rx/tx queues, and invoke find_vqs */
1477 	err = init_vqs(vi);
1478 	if (err)
1479 		goto free_stats;
1480 
1481 	netif_set_real_num_tx_queues(dev, 1);
1482 	netif_set_real_num_rx_queues(dev, 1);
1483 
1484 	err = register_netdev(dev);
1485 	if (err) {
1486 		pr_debug("virtio_net: registering device failed\n");
1487 		goto free_vqs;
1488 	}
1489 
1490 	/* Last of all, set up some receive buffers. */
1491 	for (i = 0; i < vi->max_queue_pairs; i++) {
1492 		try_fill_recv(&vi->rq[i], GFP_KERNEL);
1493 
1494 		/* If we didn't even get one input buffer, we're useless. */
1495 		if (vi->rq[i].num == 0) {
1496 			free_unused_bufs(vi);
1497 			err = -ENOMEM;
1498 			goto free_recv_bufs;
1499 		}
1500 	}
1501 
1502 	/* Assume link up if device can't report link status,
1503 	   otherwise get link status from config. */
1504 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
1505 		netif_carrier_off(dev);
1506 		schedule_work(&vi->config_work);
1507 	} else {
1508 		vi->status = VIRTIO_NET_S_LINK_UP;
1509 		netif_carrier_on(dev);
1510 	}
1511 
1512 	pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
1513 		 dev->name, max_queue_pairs);
1514 
1515 	return 0;
1516 
1517 free_recv_bufs:
1518 	free_receive_bufs(vi);
1519 	unregister_netdev(dev);
1520 free_vqs:
1521 	cancel_delayed_work_sync(&vi->refill);
1522 	virtnet_del_vqs(vi);
1523 free_stats:
1524 	free_percpu(vi->stats);
1525 free:
1526 	free_netdev(dev);
1527 	return err;
1528 }
1529 
1530 static void remove_vq_common(struct virtnet_info *vi)
1531 {
1532 	vi->vdev->config->reset(vi->vdev);
1533 
1534 	/* Free unused buffers in both send and recv, if any. */
1535 	free_unused_bufs(vi);
1536 
1537 	free_receive_bufs(vi);
1538 
1539 	virtnet_del_vqs(vi);
1540 }
1541 
1542 static void virtnet_remove(struct virtio_device *vdev)
1543 {
1544 	struct virtnet_info *vi = vdev->priv;
1545 
1546 	/* Prevent config work handler from accessing the device. */
1547 	mutex_lock(&vi->config_lock);
1548 	vi->config_enable = false;
1549 	mutex_unlock(&vi->config_lock);
1550 
1551 	unregister_netdev(vi->dev);
1552 
1553 	remove_vq_common(vi);
1554 
1555 	flush_work(&vi->config_work);
1556 
1557 	free_percpu(vi->stats);
1558 	free_netdev(vi->dev);
1559 }
1560 
1561 #ifdef CONFIG_PM
1562 static int virtnet_freeze(struct virtio_device *vdev)
1563 {
1564 	struct virtnet_info *vi = vdev->priv;
1565 	int i;
1566 
1567 	/* Prevent config work handler from accessing the device */
1568 	mutex_lock(&vi->config_lock);
1569 	vi->config_enable = false;
1570 	mutex_unlock(&vi->config_lock);
1571 
1572 	netif_device_detach(vi->dev);
1573 	cancel_delayed_work_sync(&vi->refill);
1574 
1575 	if (netif_running(vi->dev))
1576 		for (i = 0; i < vi->max_queue_pairs; i++) {
1577 			napi_disable(&vi->rq[i].napi);
1578 			netif_napi_del(&vi->rq[i].napi);
1579 		}
1580 
1581 	remove_vq_common(vi);
1582 
1583 	flush_work(&vi->config_work);
1584 
1585 	return 0;
1586 }
1587 
1588 static int virtnet_restore(struct virtio_device *vdev)
1589 {
1590 	struct virtnet_info *vi = vdev->priv;
1591 	int err, i;
1592 
1593 	err = init_vqs(vi);
1594 	if (err)
1595 		return err;
1596 
1597 	if (netif_running(vi->dev))
1598 		for (i = 0; i < vi->max_queue_pairs; i++)
1599 			virtnet_napi_enable(&vi->rq[i]);
1600 
1601 	netif_device_attach(vi->dev);
1602 
1603 	for (i = 0; i < vi->max_queue_pairs; i++)
1604 		if (!try_fill_recv(&vi->rq[i], GFP_KERNEL))
1605 			schedule_delayed_work(&vi->refill, 0);
1606 
1607 	mutex_lock(&vi->config_lock);
1608 	vi->config_enable = true;
1609 	mutex_unlock(&vi->config_lock);
1610 
1611 	virtnet_set_queues(vi, vi->curr_queue_pairs);
1612 
1613 	return 0;
1614 }
1615 #endif
1616 
1617 static struct virtio_device_id id_table[] = {
1618 	{ VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
1619 	{ 0 },
1620 };
1621 
1622 static unsigned int features[] = {
1623 	VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
1624 	VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
1625 	VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
1626 	VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
1627 	VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO,
1628 	VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ,
1629 	VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN,
1630 	VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ,
1631 };
1632 
1633 static struct virtio_driver virtio_net_driver = {
1634 	.feature_table = features,
1635 	.feature_table_size = ARRAY_SIZE(features),
1636 	.driver.name =	KBUILD_MODNAME,
1637 	.driver.owner =	THIS_MODULE,
1638 	.id_table =	id_table,
1639 	.probe =	virtnet_probe,
1640 	.remove =	virtnet_remove,
1641 	.config_changed = virtnet_config_changed,
1642 #ifdef CONFIG_PM
1643 	.freeze =	virtnet_freeze,
1644 	.restore =	virtnet_restore,
1645 #endif
1646 };
1647 
1648 static int __init init(void)
1649 {
1650 	return register_virtio_driver(&virtio_net_driver);
1651 }
1652 
1653 static void __exit fini(void)
1654 {
1655 	unregister_virtio_driver(&virtio_net_driver);
1656 }
1657 module_init(init);
1658 module_exit(fini);
1659 
1660 MODULE_DEVICE_TABLE(virtio, id_table);
1661 MODULE_DESCRIPTION("Virtio network driver");
1662 MODULE_LICENSE("GPL");
1663