xref: /openbmc/linux/drivers/net/virtio_net.c (revision 3a9a231d)
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 int csum = 1, gso = 1;
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 
43 struct virtnet_stats {
44 	struct u64_stats_sync syncp;
45 	u64 tx_bytes;
46 	u64 tx_packets;
47 
48 	u64 rx_bytes;
49 	u64 rx_packets;
50 };
51 
52 struct virtnet_info {
53 	struct virtio_device *vdev;
54 	struct virtqueue *rvq, *svq, *cvq;
55 	struct net_device *dev;
56 	struct napi_struct napi;
57 	unsigned int status;
58 
59 	/* Number of input buffers, and max we've ever had. */
60 	unsigned int num, max;
61 
62 	/* I like... big packets and I cannot lie! */
63 	bool big_packets;
64 
65 	/* Host will merge rx buffers for big packets (shake it! shake it!) */
66 	bool mergeable_rx_bufs;
67 
68 	/* Active statistics */
69 	struct virtnet_stats __percpu *stats;
70 
71 	/* Work struct for refilling if we run low on memory. */
72 	struct delayed_work refill;
73 
74 	/* Chain pages by the private ptr. */
75 	struct page *pages;
76 
77 	/* fragments + linear part + virtio header */
78 	struct scatterlist rx_sg[MAX_SKB_FRAGS + 2];
79 	struct scatterlist tx_sg[MAX_SKB_FRAGS + 2];
80 };
81 
82 struct skb_vnet_hdr {
83 	union {
84 		struct virtio_net_hdr hdr;
85 		struct virtio_net_hdr_mrg_rxbuf mhdr;
86 	};
87 	unsigned int num_sg;
88 };
89 
90 struct padded_vnet_hdr {
91 	struct virtio_net_hdr hdr;
92 	/*
93 	 * virtio_net_hdr should be in a separated sg buffer because of a
94 	 * QEMU bug, and data sg buffer shares same page with this header sg.
95 	 * This padding makes next sg 16 byte aligned after virtio_net_hdr.
96 	 */
97 	char padding[6];
98 };
99 
100 static inline struct skb_vnet_hdr *skb_vnet_hdr(struct sk_buff *skb)
101 {
102 	return (struct skb_vnet_hdr *)skb->cb;
103 }
104 
105 /*
106  * private is used to chain pages for big packets, put the whole
107  * most recent used list in the beginning for reuse
108  */
109 static void give_pages(struct virtnet_info *vi, struct page *page)
110 {
111 	struct page *end;
112 
113 	/* Find end of list, sew whole thing into vi->pages. */
114 	for (end = page; end->private; end = (struct page *)end->private);
115 	end->private = (unsigned long)vi->pages;
116 	vi->pages = page;
117 }
118 
119 static struct page *get_a_page(struct virtnet_info *vi, gfp_t gfp_mask)
120 {
121 	struct page *p = vi->pages;
122 
123 	if (p) {
124 		vi->pages = (struct page *)p->private;
125 		/* clear private here, it is used to chain pages */
126 		p->private = 0;
127 	} else
128 		p = alloc_page(gfp_mask);
129 	return p;
130 }
131 
132 static void skb_xmit_done(struct virtqueue *svq)
133 {
134 	struct virtnet_info *vi = svq->vdev->priv;
135 
136 	/* Suppress further interrupts. */
137 	virtqueue_disable_cb(svq);
138 
139 	/* We were probably waiting for more output buffers. */
140 	netif_wake_queue(vi->dev);
141 }
142 
143 static void set_skb_frag(struct sk_buff *skb, struct page *page,
144 			 unsigned int offset, unsigned int *len)
145 {
146 	int size = min((unsigned)PAGE_SIZE - offset, *len);
147 	int i = skb_shinfo(skb)->nr_frags;
148 
149 	__skb_fill_page_desc(skb, i, page, offset, size);
150 
151 	skb->data_len += size;
152 	skb->len += size;
153 	skb->truesize += PAGE_SIZE;
154 	skb_shinfo(skb)->nr_frags++;
155 	*len -= size;
156 }
157 
158 static struct sk_buff *page_to_skb(struct virtnet_info *vi,
159 				   struct page *page, unsigned int len)
160 {
161 	struct sk_buff *skb;
162 	struct skb_vnet_hdr *hdr;
163 	unsigned int copy, hdr_len, offset;
164 	char *p;
165 
166 	p = page_address(page);
167 
168 	/* copy small packet so we can reuse these pages for small data */
169 	skb = netdev_alloc_skb_ip_align(vi->dev, GOOD_COPY_LEN);
170 	if (unlikely(!skb))
171 		return NULL;
172 
173 	hdr = skb_vnet_hdr(skb);
174 
175 	if (vi->mergeable_rx_bufs) {
176 		hdr_len = sizeof hdr->mhdr;
177 		offset = hdr_len;
178 	} else {
179 		hdr_len = sizeof hdr->hdr;
180 		offset = sizeof(struct padded_vnet_hdr);
181 	}
182 
183 	memcpy(hdr, p, hdr_len);
184 
185 	len -= hdr_len;
186 	p += offset;
187 
188 	copy = len;
189 	if (copy > skb_tailroom(skb))
190 		copy = skb_tailroom(skb);
191 	memcpy(skb_put(skb, copy), p, copy);
192 
193 	len -= copy;
194 	offset += copy;
195 
196 	/*
197 	 * Verify that we can indeed put this data into a skb.
198 	 * This is here to handle cases when the device erroneously
199 	 * tries to receive more than is possible. This is usually
200 	 * the case of a broken device.
201 	 */
202 	if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) {
203 		if (net_ratelimit())
204 			pr_debug("%s: too much data\n", skb->dev->name);
205 		dev_kfree_skb(skb);
206 		return NULL;
207 	}
208 
209 	while (len) {
210 		set_skb_frag(skb, page, offset, &len);
211 		page = (struct page *)page->private;
212 		offset = 0;
213 	}
214 
215 	if (page)
216 		give_pages(vi, page);
217 
218 	return skb;
219 }
220 
221 static int receive_mergeable(struct virtnet_info *vi, struct sk_buff *skb)
222 {
223 	struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
224 	struct page *page;
225 	int num_buf, i, len;
226 
227 	num_buf = hdr->mhdr.num_buffers;
228 	while (--num_buf) {
229 		i = skb_shinfo(skb)->nr_frags;
230 		if (i >= MAX_SKB_FRAGS) {
231 			pr_debug("%s: packet too long\n", skb->dev->name);
232 			skb->dev->stats.rx_length_errors++;
233 			return -EINVAL;
234 		}
235 		page = virtqueue_get_buf(vi->rvq, &len);
236 		if (!page) {
237 			pr_debug("%s: rx error: %d buffers missing\n",
238 				 skb->dev->name, hdr->mhdr.num_buffers);
239 			skb->dev->stats.rx_length_errors++;
240 			return -EINVAL;
241 		}
242 
243 		if (len > PAGE_SIZE)
244 			len = PAGE_SIZE;
245 
246 		set_skb_frag(skb, page, 0, &len);
247 
248 		--vi->num;
249 	}
250 	return 0;
251 }
252 
253 static void receive_buf(struct net_device *dev, void *buf, unsigned int len)
254 {
255 	struct virtnet_info *vi = netdev_priv(dev);
256 	struct virtnet_stats __percpu *stats = this_cpu_ptr(vi->stats);
257 	struct sk_buff *skb;
258 	struct page *page;
259 	struct skb_vnet_hdr *hdr;
260 
261 	if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
262 		pr_debug("%s: short packet %i\n", dev->name, len);
263 		dev->stats.rx_length_errors++;
264 		if (vi->mergeable_rx_bufs || vi->big_packets)
265 			give_pages(vi, buf);
266 		else
267 			dev_kfree_skb(buf);
268 		return;
269 	}
270 
271 	if (!vi->mergeable_rx_bufs && !vi->big_packets) {
272 		skb = buf;
273 		len -= sizeof(struct virtio_net_hdr);
274 		skb_trim(skb, len);
275 	} else {
276 		page = buf;
277 		skb = page_to_skb(vi, page, len);
278 		if (unlikely(!skb)) {
279 			dev->stats.rx_dropped++;
280 			give_pages(vi, page);
281 			return;
282 		}
283 		if (vi->mergeable_rx_bufs)
284 			if (receive_mergeable(vi, skb)) {
285 				dev_kfree_skb(skb);
286 				return;
287 			}
288 	}
289 
290 	hdr = skb_vnet_hdr(skb);
291 
292 	u64_stats_update_begin(&stats->syncp);
293 	stats->rx_bytes += skb->len;
294 	stats->rx_packets++;
295 	u64_stats_update_end(&stats->syncp);
296 
297 	if (hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
298 		pr_debug("Needs csum!\n");
299 		if (!skb_partial_csum_set(skb,
300 					  hdr->hdr.csum_start,
301 					  hdr->hdr.csum_offset))
302 			goto frame_err;
303 	} else if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) {
304 		skb->ip_summed = CHECKSUM_UNNECESSARY;
305 	}
306 
307 	skb->protocol = eth_type_trans(skb, dev);
308 	pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
309 		 ntohs(skb->protocol), skb->len, skb->pkt_type);
310 
311 	if (hdr->hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
312 		pr_debug("GSO!\n");
313 		switch (hdr->hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
314 		case VIRTIO_NET_HDR_GSO_TCPV4:
315 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
316 			break;
317 		case VIRTIO_NET_HDR_GSO_UDP:
318 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
319 			break;
320 		case VIRTIO_NET_HDR_GSO_TCPV6:
321 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
322 			break;
323 		default:
324 			if (net_ratelimit())
325 				printk(KERN_WARNING "%s: bad gso type %u.\n",
326 				       dev->name, hdr->hdr.gso_type);
327 			goto frame_err;
328 		}
329 
330 		if (hdr->hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
331 			skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
332 
333 		skb_shinfo(skb)->gso_size = hdr->hdr.gso_size;
334 		if (skb_shinfo(skb)->gso_size == 0) {
335 			if (net_ratelimit())
336 				printk(KERN_WARNING "%s: zero gso size.\n",
337 				       dev->name);
338 			goto frame_err;
339 		}
340 
341 		/* Header must be checked, and gso_segs computed. */
342 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
343 		skb_shinfo(skb)->gso_segs = 0;
344 	}
345 
346 	netif_receive_skb(skb);
347 	return;
348 
349 frame_err:
350 	dev->stats.rx_frame_errors++;
351 	dev_kfree_skb(skb);
352 }
353 
354 static int add_recvbuf_small(struct virtnet_info *vi, gfp_t gfp)
355 {
356 	struct sk_buff *skb;
357 	struct skb_vnet_hdr *hdr;
358 	int err;
359 
360 	skb = netdev_alloc_skb_ip_align(vi->dev, MAX_PACKET_LEN);
361 	if (unlikely(!skb))
362 		return -ENOMEM;
363 
364 	skb_put(skb, MAX_PACKET_LEN);
365 
366 	hdr = skb_vnet_hdr(skb);
367 	sg_set_buf(vi->rx_sg, &hdr->hdr, sizeof hdr->hdr);
368 
369 	skb_to_sgvec(skb, vi->rx_sg + 1, 0, skb->len);
370 
371 	err = virtqueue_add_buf_gfp(vi->rvq, vi->rx_sg, 0, 2, skb, gfp);
372 	if (err < 0)
373 		dev_kfree_skb(skb);
374 
375 	return err;
376 }
377 
378 static int add_recvbuf_big(struct virtnet_info *vi, gfp_t gfp)
379 {
380 	struct page *first, *list = NULL;
381 	char *p;
382 	int i, err, offset;
383 
384 	/* page in vi->rx_sg[MAX_SKB_FRAGS + 1] is list tail */
385 	for (i = MAX_SKB_FRAGS + 1; i > 1; --i) {
386 		first = get_a_page(vi, gfp);
387 		if (!first) {
388 			if (list)
389 				give_pages(vi, list);
390 			return -ENOMEM;
391 		}
392 		sg_set_buf(&vi->rx_sg[i], page_address(first), PAGE_SIZE);
393 
394 		/* chain new page in list head to match sg */
395 		first->private = (unsigned long)list;
396 		list = first;
397 	}
398 
399 	first = get_a_page(vi, gfp);
400 	if (!first) {
401 		give_pages(vi, list);
402 		return -ENOMEM;
403 	}
404 	p = page_address(first);
405 
406 	/* vi->rx_sg[0], vi->rx_sg[1] share the same page */
407 	/* a separated vi->rx_sg[0] for virtio_net_hdr only due to QEMU bug */
408 	sg_set_buf(&vi->rx_sg[0], p, sizeof(struct virtio_net_hdr));
409 
410 	/* vi->rx_sg[1] for data packet, from offset */
411 	offset = sizeof(struct padded_vnet_hdr);
412 	sg_set_buf(&vi->rx_sg[1], p + offset, PAGE_SIZE - offset);
413 
414 	/* chain first in list head */
415 	first->private = (unsigned long)list;
416 	err = virtqueue_add_buf_gfp(vi->rvq, vi->rx_sg, 0, MAX_SKB_FRAGS + 2,
417 				    first, gfp);
418 	if (err < 0)
419 		give_pages(vi, first);
420 
421 	return err;
422 }
423 
424 static int add_recvbuf_mergeable(struct virtnet_info *vi, gfp_t gfp)
425 {
426 	struct page *page;
427 	int err;
428 
429 	page = get_a_page(vi, gfp);
430 	if (!page)
431 		return -ENOMEM;
432 
433 	sg_init_one(vi->rx_sg, page_address(page), PAGE_SIZE);
434 
435 	err = virtqueue_add_buf_gfp(vi->rvq, vi->rx_sg, 0, 1, page, gfp);
436 	if (err < 0)
437 		give_pages(vi, page);
438 
439 	return err;
440 }
441 
442 /* Returns false if we couldn't fill entirely (OOM). */
443 static bool try_fill_recv(struct virtnet_info *vi, gfp_t gfp)
444 {
445 	int err;
446 	bool oom;
447 
448 	do {
449 		if (vi->mergeable_rx_bufs)
450 			err = add_recvbuf_mergeable(vi, gfp);
451 		else if (vi->big_packets)
452 			err = add_recvbuf_big(vi, gfp);
453 		else
454 			err = add_recvbuf_small(vi, gfp);
455 
456 		oom = err == -ENOMEM;
457 		if (err < 0)
458 			break;
459 		++vi->num;
460 	} while (err > 0);
461 	if (unlikely(vi->num > vi->max))
462 		vi->max = vi->num;
463 	virtqueue_kick(vi->rvq);
464 	return !oom;
465 }
466 
467 static void skb_recv_done(struct virtqueue *rvq)
468 {
469 	struct virtnet_info *vi = rvq->vdev->priv;
470 	/* Schedule NAPI, Suppress further interrupts if successful. */
471 	if (napi_schedule_prep(&vi->napi)) {
472 		virtqueue_disable_cb(rvq);
473 		__napi_schedule(&vi->napi);
474 	}
475 }
476 
477 static void virtnet_napi_enable(struct virtnet_info *vi)
478 {
479 	napi_enable(&vi->napi);
480 
481 	/* If all buffers were filled by other side before we napi_enabled, we
482 	 * won't get another interrupt, so process any outstanding packets
483 	 * now.  virtnet_poll wants re-enable the queue, so we disable here.
484 	 * We synchronize against interrupts via NAPI_STATE_SCHED */
485 	if (napi_schedule_prep(&vi->napi)) {
486 		virtqueue_disable_cb(vi->rvq);
487 		__napi_schedule(&vi->napi);
488 	}
489 }
490 
491 static void refill_work(struct work_struct *work)
492 {
493 	struct virtnet_info *vi;
494 	bool still_empty;
495 
496 	vi = container_of(work, struct virtnet_info, refill.work);
497 	napi_disable(&vi->napi);
498 	still_empty = !try_fill_recv(vi, GFP_KERNEL);
499 	virtnet_napi_enable(vi);
500 
501 	/* In theory, this can happen: if we don't get any buffers in
502 	 * we will *never* try to fill again. */
503 	if (still_empty)
504 		schedule_delayed_work(&vi->refill, HZ/2);
505 }
506 
507 static int virtnet_poll(struct napi_struct *napi, int budget)
508 {
509 	struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi);
510 	void *buf;
511 	unsigned int len, received = 0;
512 
513 again:
514 	while (received < budget &&
515 	       (buf = virtqueue_get_buf(vi->rvq, &len)) != NULL) {
516 		receive_buf(vi->dev, buf, len);
517 		--vi->num;
518 		received++;
519 	}
520 
521 	if (vi->num < vi->max / 2) {
522 		if (!try_fill_recv(vi, GFP_ATOMIC))
523 			schedule_delayed_work(&vi->refill, 0);
524 	}
525 
526 	/* Out of packets? */
527 	if (received < budget) {
528 		napi_complete(napi);
529 		if (unlikely(!virtqueue_enable_cb(vi->rvq)) &&
530 		    napi_schedule_prep(napi)) {
531 			virtqueue_disable_cb(vi->rvq);
532 			__napi_schedule(napi);
533 			goto again;
534 		}
535 	}
536 
537 	return received;
538 }
539 
540 static unsigned int free_old_xmit_skbs(struct virtnet_info *vi)
541 {
542 	struct sk_buff *skb;
543 	unsigned int len, tot_sgs = 0;
544 	struct virtnet_stats __percpu *stats = this_cpu_ptr(vi->stats);
545 
546 	while ((skb = virtqueue_get_buf(vi->svq, &len)) != NULL) {
547 		pr_debug("Sent skb %p\n", skb);
548 
549 		u64_stats_update_begin(&stats->syncp);
550 		stats->tx_bytes += skb->len;
551 		stats->tx_packets++;
552 		u64_stats_update_end(&stats->syncp);
553 
554 		tot_sgs += skb_vnet_hdr(skb)->num_sg;
555 		dev_kfree_skb_any(skb);
556 	}
557 	return tot_sgs;
558 }
559 
560 static int xmit_skb(struct virtnet_info *vi, struct sk_buff *skb)
561 {
562 	struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
563 	const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
564 
565 	pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
566 
567 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
568 		hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
569 		hdr->hdr.csum_start = skb_checksum_start_offset(skb);
570 		hdr->hdr.csum_offset = skb->csum_offset;
571 	} else {
572 		hdr->hdr.flags = 0;
573 		hdr->hdr.csum_offset = hdr->hdr.csum_start = 0;
574 	}
575 
576 	if (skb_is_gso(skb)) {
577 		hdr->hdr.hdr_len = skb_headlen(skb);
578 		hdr->hdr.gso_size = skb_shinfo(skb)->gso_size;
579 		if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
580 			hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
581 		else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
582 			hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
583 		else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
584 			hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
585 		else
586 			BUG();
587 		if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
588 			hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
589 	} else {
590 		hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
591 		hdr->hdr.gso_size = hdr->hdr.hdr_len = 0;
592 	}
593 
594 	hdr->mhdr.num_buffers = 0;
595 
596 	/* Encode metadata header at front. */
597 	if (vi->mergeable_rx_bufs)
598 		sg_set_buf(vi->tx_sg, &hdr->mhdr, sizeof hdr->mhdr);
599 	else
600 		sg_set_buf(vi->tx_sg, &hdr->hdr, sizeof hdr->hdr);
601 
602 	hdr->num_sg = skb_to_sgvec(skb, vi->tx_sg + 1, 0, skb->len) + 1;
603 	return virtqueue_add_buf(vi->svq, vi->tx_sg, hdr->num_sg,
604 					0, skb);
605 }
606 
607 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
608 {
609 	struct virtnet_info *vi = netdev_priv(dev);
610 	int capacity;
611 
612 	/* Free up any pending old buffers before queueing new ones. */
613 	free_old_xmit_skbs(vi);
614 
615 	/* Try to transmit */
616 	capacity = xmit_skb(vi, skb);
617 
618 	/* This can happen with OOM and indirect buffers. */
619 	if (unlikely(capacity < 0)) {
620 		if (net_ratelimit()) {
621 			if (likely(capacity == -ENOMEM)) {
622 				dev_warn(&dev->dev,
623 					 "TX queue failure: out of memory\n");
624 			} else {
625 				dev->stats.tx_fifo_errors++;
626 				dev_warn(&dev->dev,
627 					 "Unexpected TX queue failure: %d\n",
628 					 capacity);
629 			}
630 		}
631 		dev->stats.tx_dropped++;
632 		kfree_skb(skb);
633 		return NETDEV_TX_OK;
634 	}
635 	virtqueue_kick(vi->svq);
636 
637 	/* Don't wait up for transmitted skbs to be freed. */
638 	skb_orphan(skb);
639 	nf_reset(skb);
640 
641 	/* Apparently nice girls don't return TX_BUSY; stop the queue
642 	 * before it gets out of hand.  Naturally, this wastes entries. */
643 	if (capacity < 2+MAX_SKB_FRAGS) {
644 		netif_stop_queue(dev);
645 		if (unlikely(!virtqueue_enable_cb_delayed(vi->svq))) {
646 			/* More just got used, free them then recheck. */
647 			capacity += free_old_xmit_skbs(vi);
648 			if (capacity >= 2+MAX_SKB_FRAGS) {
649 				netif_start_queue(dev);
650 				virtqueue_disable_cb(vi->svq);
651 			}
652 		}
653 	}
654 
655 	return NETDEV_TX_OK;
656 }
657 
658 static int virtnet_set_mac_address(struct net_device *dev, void *p)
659 {
660 	struct virtnet_info *vi = netdev_priv(dev);
661 	struct virtio_device *vdev = vi->vdev;
662 	int ret;
663 
664 	ret = eth_mac_addr(dev, p);
665 	if (ret)
666 		return ret;
667 
668 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
669 		vdev->config->set(vdev, offsetof(struct virtio_net_config, mac),
670 		                  dev->dev_addr, dev->addr_len);
671 
672 	return 0;
673 }
674 
675 static struct rtnl_link_stats64 *virtnet_stats(struct net_device *dev,
676 					       struct rtnl_link_stats64 *tot)
677 {
678 	struct virtnet_info *vi = netdev_priv(dev);
679 	int cpu;
680 	unsigned int start;
681 
682 	for_each_possible_cpu(cpu) {
683 		struct virtnet_stats __percpu *stats
684 			= per_cpu_ptr(vi->stats, cpu);
685 		u64 tpackets, tbytes, rpackets, rbytes;
686 
687 		do {
688 			start = u64_stats_fetch_begin(&stats->syncp);
689 			tpackets = stats->tx_packets;
690 			tbytes   = stats->tx_bytes;
691 			rpackets = stats->rx_packets;
692 			rbytes   = stats->rx_bytes;
693 		} while (u64_stats_fetch_retry(&stats->syncp, start));
694 
695 		tot->rx_packets += rpackets;
696 		tot->tx_packets += tpackets;
697 		tot->rx_bytes   += rbytes;
698 		tot->tx_bytes   += tbytes;
699 	}
700 
701 	tot->tx_dropped = dev->stats.tx_dropped;
702 	tot->rx_dropped = dev->stats.rx_dropped;
703 	tot->rx_length_errors = dev->stats.rx_length_errors;
704 	tot->rx_frame_errors = dev->stats.rx_frame_errors;
705 
706 	return tot;
707 }
708 
709 #ifdef CONFIG_NET_POLL_CONTROLLER
710 static void virtnet_netpoll(struct net_device *dev)
711 {
712 	struct virtnet_info *vi = netdev_priv(dev);
713 
714 	napi_schedule(&vi->napi);
715 }
716 #endif
717 
718 static int virtnet_open(struct net_device *dev)
719 {
720 	struct virtnet_info *vi = netdev_priv(dev);
721 
722 	virtnet_napi_enable(vi);
723 	return 0;
724 }
725 
726 /*
727  * Send command via the control virtqueue and check status.  Commands
728  * supported by the hypervisor, as indicated by feature bits, should
729  * never fail unless improperly formated.
730  */
731 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
732 				 struct scatterlist *data, int out, int in)
733 {
734 	struct scatterlist *s, sg[VIRTNET_SEND_COMMAND_SG_MAX + 2];
735 	struct virtio_net_ctrl_hdr ctrl;
736 	virtio_net_ctrl_ack status = ~0;
737 	unsigned int tmp;
738 	int i;
739 
740 	/* Caller should know better */
741 	BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ||
742 		(out + in > VIRTNET_SEND_COMMAND_SG_MAX));
743 
744 	out++; /* Add header */
745 	in++; /* Add return status */
746 
747 	ctrl.class = class;
748 	ctrl.cmd = cmd;
749 
750 	sg_init_table(sg, out + in);
751 
752 	sg_set_buf(&sg[0], &ctrl, sizeof(ctrl));
753 	for_each_sg(data, s, out + in - 2, i)
754 		sg_set_buf(&sg[i + 1], sg_virt(s), s->length);
755 	sg_set_buf(&sg[out + in - 1], &status, sizeof(status));
756 
757 	BUG_ON(virtqueue_add_buf(vi->cvq, sg, out, in, vi) < 0);
758 
759 	virtqueue_kick(vi->cvq);
760 
761 	/*
762 	 * Spin for a response, the kick causes an ioport write, trapping
763 	 * into the hypervisor, so the request should be handled immediately.
764 	 */
765 	while (!virtqueue_get_buf(vi->cvq, &tmp))
766 		cpu_relax();
767 
768 	return status == VIRTIO_NET_OK;
769 }
770 
771 static int virtnet_close(struct net_device *dev)
772 {
773 	struct virtnet_info *vi = netdev_priv(dev);
774 
775 	napi_disable(&vi->napi);
776 
777 	return 0;
778 }
779 
780 static void virtnet_set_rx_mode(struct net_device *dev)
781 {
782 	struct virtnet_info *vi = netdev_priv(dev);
783 	struct scatterlist sg[2];
784 	u8 promisc, allmulti;
785 	struct virtio_net_ctrl_mac *mac_data;
786 	struct netdev_hw_addr *ha;
787 	int uc_count;
788 	int mc_count;
789 	void *buf;
790 	int i;
791 
792 	/* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */
793 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
794 		return;
795 
796 	promisc = ((dev->flags & IFF_PROMISC) != 0);
797 	allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
798 
799 	sg_init_one(sg, &promisc, sizeof(promisc));
800 
801 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
802 				  VIRTIO_NET_CTRL_RX_PROMISC,
803 				  sg, 1, 0))
804 		dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
805 			 promisc ? "en" : "dis");
806 
807 	sg_init_one(sg, &allmulti, sizeof(allmulti));
808 
809 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
810 				  VIRTIO_NET_CTRL_RX_ALLMULTI,
811 				  sg, 1, 0))
812 		dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
813 			 allmulti ? "en" : "dis");
814 
815 	uc_count = netdev_uc_count(dev);
816 	mc_count = netdev_mc_count(dev);
817 	/* MAC filter - use one buffer for both lists */
818 	buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
819 		      (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
820 	mac_data = buf;
821 	if (!buf) {
822 		dev_warn(&dev->dev, "No memory for MAC address buffer\n");
823 		return;
824 	}
825 
826 	sg_init_table(sg, 2);
827 
828 	/* Store the unicast list and count in the front of the buffer */
829 	mac_data->entries = uc_count;
830 	i = 0;
831 	netdev_for_each_uc_addr(ha, dev)
832 		memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
833 
834 	sg_set_buf(&sg[0], mac_data,
835 		   sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
836 
837 	/* multicast list and count fill the end */
838 	mac_data = (void *)&mac_data->macs[uc_count][0];
839 
840 	mac_data->entries = mc_count;
841 	i = 0;
842 	netdev_for_each_mc_addr(ha, dev)
843 		memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
844 
845 	sg_set_buf(&sg[1], mac_data,
846 		   sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
847 
848 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
849 				  VIRTIO_NET_CTRL_MAC_TABLE_SET,
850 				  sg, 2, 0))
851 		dev_warn(&dev->dev, "Failed to set MAC fitler table.\n");
852 
853 	kfree(buf);
854 }
855 
856 static void virtnet_vlan_rx_add_vid(struct net_device *dev, u16 vid)
857 {
858 	struct virtnet_info *vi = netdev_priv(dev);
859 	struct scatterlist sg;
860 
861 	sg_init_one(&sg, &vid, sizeof(vid));
862 
863 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
864 				  VIRTIO_NET_CTRL_VLAN_ADD, &sg, 1, 0))
865 		dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
866 }
867 
868 static void virtnet_vlan_rx_kill_vid(struct net_device *dev, u16 vid)
869 {
870 	struct virtnet_info *vi = netdev_priv(dev);
871 	struct scatterlist sg;
872 
873 	sg_init_one(&sg, &vid, sizeof(vid));
874 
875 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
876 				  VIRTIO_NET_CTRL_VLAN_DEL, &sg, 1, 0))
877 		dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
878 }
879 
880 static void virtnet_get_ringparam(struct net_device *dev,
881 				struct ethtool_ringparam *ring)
882 {
883 	struct virtnet_info *vi = netdev_priv(dev);
884 
885 	ring->rx_max_pending = virtqueue_get_vring_size(vi->rvq);
886 	ring->tx_max_pending = virtqueue_get_vring_size(vi->svq);
887 	ring->rx_pending = ring->rx_max_pending;
888 	ring->tx_pending = ring->tx_max_pending;
889 
890 }
891 
892 static const struct ethtool_ops virtnet_ethtool_ops = {
893 	.get_link = ethtool_op_get_link,
894 	.get_ringparam = virtnet_get_ringparam,
895 };
896 
897 #define MIN_MTU 68
898 #define MAX_MTU 65535
899 
900 static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
901 {
902 	if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
903 		return -EINVAL;
904 	dev->mtu = new_mtu;
905 	return 0;
906 }
907 
908 static const struct net_device_ops virtnet_netdev = {
909 	.ndo_open            = virtnet_open,
910 	.ndo_stop   	     = virtnet_close,
911 	.ndo_start_xmit      = start_xmit,
912 	.ndo_validate_addr   = eth_validate_addr,
913 	.ndo_set_mac_address = virtnet_set_mac_address,
914 	.ndo_set_rx_mode     = virtnet_set_rx_mode,
915 	.ndo_change_mtu	     = virtnet_change_mtu,
916 	.ndo_get_stats64     = virtnet_stats,
917 	.ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
918 	.ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
919 #ifdef CONFIG_NET_POLL_CONTROLLER
920 	.ndo_poll_controller = virtnet_netpoll,
921 #endif
922 };
923 
924 static void virtnet_update_status(struct virtnet_info *vi)
925 {
926 	u16 v;
927 
928 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS))
929 		return;
930 
931 	vi->vdev->config->get(vi->vdev,
932 			      offsetof(struct virtio_net_config, status),
933 			      &v, sizeof(v));
934 
935 	/* Ignore unknown (future) status bits */
936 	v &= VIRTIO_NET_S_LINK_UP;
937 
938 	if (vi->status == v)
939 		return;
940 
941 	vi->status = v;
942 
943 	if (vi->status & VIRTIO_NET_S_LINK_UP) {
944 		netif_carrier_on(vi->dev);
945 		netif_wake_queue(vi->dev);
946 	} else {
947 		netif_carrier_off(vi->dev);
948 		netif_stop_queue(vi->dev);
949 	}
950 }
951 
952 static void virtnet_config_changed(struct virtio_device *vdev)
953 {
954 	struct virtnet_info *vi = vdev->priv;
955 
956 	virtnet_update_status(vi);
957 }
958 
959 static int virtnet_probe(struct virtio_device *vdev)
960 {
961 	int err;
962 	struct net_device *dev;
963 	struct virtnet_info *vi;
964 	struct virtqueue *vqs[3];
965 	vq_callback_t *callbacks[] = { skb_recv_done, skb_xmit_done, NULL};
966 	const char *names[] = { "input", "output", "control" };
967 	int nvqs;
968 
969 	/* Allocate ourselves a network device with room for our info */
970 	dev = alloc_etherdev(sizeof(struct virtnet_info));
971 	if (!dev)
972 		return -ENOMEM;
973 
974 	/* Set up network device as normal. */
975 	dev->priv_flags |= IFF_UNICAST_FLT;
976 	dev->netdev_ops = &virtnet_netdev;
977 	dev->features = NETIF_F_HIGHDMA;
978 
979 	SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
980 	SET_NETDEV_DEV(dev, &vdev->dev);
981 
982 	/* Do we support "hardware" checksums? */
983 	if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
984 		/* This opens up the world of extra features. */
985 		dev->hw_features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
986 		if (csum)
987 			dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
988 
989 		if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
990 			dev->hw_features |= NETIF_F_TSO | NETIF_F_UFO
991 				| NETIF_F_TSO_ECN | NETIF_F_TSO6;
992 		}
993 		/* Individual feature bits: what can host handle? */
994 		if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
995 			dev->hw_features |= NETIF_F_TSO;
996 		if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
997 			dev->hw_features |= NETIF_F_TSO6;
998 		if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
999 			dev->hw_features |= NETIF_F_TSO_ECN;
1000 		if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
1001 			dev->hw_features |= NETIF_F_UFO;
1002 
1003 		if (gso)
1004 			dev->features |= dev->hw_features & (NETIF_F_ALL_TSO|NETIF_F_UFO);
1005 		/* (!csum && gso) case will be fixed by register_netdev() */
1006 	}
1007 
1008 	/* Configuration may specify what MAC to use.  Otherwise random. */
1009 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
1010 		vdev->config->get(vdev,
1011 				  offsetof(struct virtio_net_config, mac),
1012 				  dev->dev_addr, dev->addr_len);
1013 	} else
1014 		random_ether_addr(dev->dev_addr);
1015 
1016 	/* Set up our device-specific information */
1017 	vi = netdev_priv(dev);
1018 	netif_napi_add(dev, &vi->napi, virtnet_poll, napi_weight);
1019 	vi->dev = dev;
1020 	vi->vdev = vdev;
1021 	vdev->priv = vi;
1022 	vi->pages = NULL;
1023 	vi->stats = alloc_percpu(struct virtnet_stats);
1024 	err = -ENOMEM;
1025 	if (vi->stats == NULL)
1026 		goto free;
1027 
1028 	INIT_DELAYED_WORK(&vi->refill, refill_work);
1029 	sg_init_table(vi->rx_sg, ARRAY_SIZE(vi->rx_sg));
1030 	sg_init_table(vi->tx_sg, ARRAY_SIZE(vi->tx_sg));
1031 
1032 	/* If we can receive ANY GSO packets, we must allocate large ones. */
1033 	if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
1034 	    virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
1035 	    virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
1036 		vi->big_packets = true;
1037 
1038 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
1039 		vi->mergeable_rx_bufs = true;
1040 
1041 	/* We expect two virtqueues, receive then send,
1042 	 * and optionally control. */
1043 	nvqs = virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ? 3 : 2;
1044 
1045 	err = vdev->config->find_vqs(vdev, nvqs, vqs, callbacks, names);
1046 	if (err)
1047 		goto free_stats;
1048 
1049 	vi->rvq = vqs[0];
1050 	vi->svq = vqs[1];
1051 
1052 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)) {
1053 		vi->cvq = vqs[2];
1054 
1055 		if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
1056 			dev->features |= NETIF_F_HW_VLAN_FILTER;
1057 	}
1058 
1059 	err = register_netdev(dev);
1060 	if (err) {
1061 		pr_debug("virtio_net: registering device failed\n");
1062 		goto free_vqs;
1063 	}
1064 
1065 	/* Last of all, set up some receive buffers. */
1066 	try_fill_recv(vi, GFP_KERNEL);
1067 
1068 	/* If we didn't even get one input buffer, we're useless. */
1069 	if (vi->num == 0) {
1070 		err = -ENOMEM;
1071 		goto unregister;
1072 	}
1073 
1074 	/* Assume link up if device can't report link status,
1075 	   otherwise get link status from config. */
1076 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
1077 		netif_carrier_off(dev);
1078 		virtnet_update_status(vi);
1079 	} else {
1080 		vi->status = VIRTIO_NET_S_LINK_UP;
1081 		netif_carrier_on(dev);
1082 	}
1083 
1084 	pr_debug("virtnet: registered device %s\n", dev->name);
1085 	return 0;
1086 
1087 unregister:
1088 	unregister_netdev(dev);
1089 	cancel_delayed_work_sync(&vi->refill);
1090 free_vqs:
1091 	vdev->config->del_vqs(vdev);
1092 free_stats:
1093 	free_percpu(vi->stats);
1094 free:
1095 	free_netdev(dev);
1096 	return err;
1097 }
1098 
1099 static void free_unused_bufs(struct virtnet_info *vi)
1100 {
1101 	void *buf;
1102 	while (1) {
1103 		buf = virtqueue_detach_unused_buf(vi->svq);
1104 		if (!buf)
1105 			break;
1106 		dev_kfree_skb(buf);
1107 	}
1108 	while (1) {
1109 		buf = virtqueue_detach_unused_buf(vi->rvq);
1110 		if (!buf)
1111 			break;
1112 		if (vi->mergeable_rx_bufs || vi->big_packets)
1113 			give_pages(vi, buf);
1114 		else
1115 			dev_kfree_skb(buf);
1116 		--vi->num;
1117 	}
1118 	BUG_ON(vi->num != 0);
1119 }
1120 
1121 static void __devexit virtnet_remove(struct virtio_device *vdev)
1122 {
1123 	struct virtnet_info *vi = vdev->priv;
1124 
1125 	/* Stop all the virtqueues. */
1126 	vdev->config->reset(vdev);
1127 
1128 
1129 	unregister_netdev(vi->dev);
1130 	cancel_delayed_work_sync(&vi->refill);
1131 
1132 	/* Free unused buffers in both send and recv, if any. */
1133 	free_unused_bufs(vi);
1134 
1135 	vdev->config->del_vqs(vi->vdev);
1136 
1137 	while (vi->pages)
1138 		__free_pages(get_a_page(vi, GFP_KERNEL), 0);
1139 
1140 	free_percpu(vi->stats);
1141 	free_netdev(vi->dev);
1142 }
1143 
1144 static struct virtio_device_id id_table[] = {
1145 	{ VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
1146 	{ 0 },
1147 };
1148 
1149 static unsigned int features[] = {
1150 	VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
1151 	VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
1152 	VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
1153 	VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
1154 	VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO,
1155 	VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ,
1156 	VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN,
1157 };
1158 
1159 static struct virtio_driver virtio_net_driver = {
1160 	.feature_table = features,
1161 	.feature_table_size = ARRAY_SIZE(features),
1162 	.driver.name =	KBUILD_MODNAME,
1163 	.driver.owner =	THIS_MODULE,
1164 	.id_table =	id_table,
1165 	.probe =	virtnet_probe,
1166 	.remove =	__devexit_p(virtnet_remove),
1167 	.config_changed = virtnet_config_changed,
1168 };
1169 
1170 static int __init init(void)
1171 {
1172 	return register_virtio_driver(&virtio_net_driver);
1173 }
1174 
1175 static void __exit fini(void)
1176 {
1177 	unregister_virtio_driver(&virtio_net_driver);
1178 }
1179 module_init(init);
1180 module_exit(fini);
1181 
1182 MODULE_DEVICE_TABLE(virtio, id_table);
1183 MODULE_DESCRIPTION("Virtio network driver");
1184 MODULE_LICENSE("GPL");
1185