xref: /linux/drivers/net/virtio_net.c (revision 5499b45190237ca90dd2ac86395cf464fe1f4cc7)
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 
29 static int napi_weight = 128;
30 module_param(napi_weight, int, 0444);
31 
32 static int csum = 1, gso = 1;
33 module_param(csum, bool, 0444);
34 module_param(gso, bool, 0444);
35 
36 /* FIXME: MTU in config. */
37 #define MAX_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
38 #define GOOD_COPY_LEN	128
39 
40 #define VIRTNET_SEND_COMMAND_SG_MAX    2
41 
42 struct virtnet_info
43 {
44 	struct virtio_device *vdev;
45 	struct virtqueue *rvq, *svq, *cvq;
46 	struct net_device *dev;
47 	struct napi_struct napi;
48 	unsigned int status;
49 
50 	/* Number of input buffers, and max we've ever had. */
51 	unsigned int num, max;
52 
53 	/* I like... big packets and I cannot lie! */
54 	bool big_packets;
55 
56 	/* Host will merge rx buffers for big packets (shake it! shake it!) */
57 	bool mergeable_rx_bufs;
58 
59 	/* Receive & send queues. */
60 	struct sk_buff_head recv;
61 	struct sk_buff_head send;
62 
63 	/* Work struct for refilling if we run low on memory. */
64 	struct delayed_work refill;
65 
66 	/* Chain pages by the private ptr. */
67 	struct page *pages;
68 };
69 
70 struct skb_vnet_hdr {
71 	union {
72 		struct virtio_net_hdr hdr;
73 		struct virtio_net_hdr_mrg_rxbuf mhdr;
74 	};
75 	unsigned int num_sg;
76 };
77 
78 static inline struct skb_vnet_hdr *skb_vnet_hdr(struct sk_buff *skb)
79 {
80 	return (struct skb_vnet_hdr *)skb->cb;
81 }
82 
83 static void give_a_page(struct virtnet_info *vi, struct page *page)
84 {
85 	page->private = (unsigned long)vi->pages;
86 	vi->pages = page;
87 }
88 
89 static void trim_pages(struct virtnet_info *vi, struct sk_buff *skb)
90 {
91 	unsigned int i;
92 
93 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
94 		give_a_page(vi, skb_shinfo(skb)->frags[i].page);
95 	skb_shinfo(skb)->nr_frags = 0;
96 	skb->data_len = 0;
97 }
98 
99 static struct page *get_a_page(struct virtnet_info *vi, gfp_t gfp_mask)
100 {
101 	struct page *p = vi->pages;
102 
103 	if (p)
104 		vi->pages = (struct page *)p->private;
105 	else
106 		p = alloc_page(gfp_mask);
107 	return p;
108 }
109 
110 static void skb_xmit_done(struct virtqueue *svq)
111 {
112 	struct virtnet_info *vi = svq->vdev->priv;
113 
114 	/* Suppress further interrupts. */
115 	svq->vq_ops->disable_cb(svq);
116 
117 	/* We were probably waiting for more output buffers. */
118 	netif_wake_queue(vi->dev);
119 }
120 
121 static void receive_skb(struct net_device *dev, struct sk_buff *skb,
122 			unsigned len)
123 {
124 	struct virtnet_info *vi = netdev_priv(dev);
125 	struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
126 	int err;
127 	int i;
128 
129 	if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
130 		pr_debug("%s: short packet %i\n", dev->name, len);
131 		dev->stats.rx_length_errors++;
132 		goto drop;
133 	}
134 
135 	if (vi->mergeable_rx_bufs) {
136 		unsigned int copy;
137 		char *p = page_address(skb_shinfo(skb)->frags[0].page);
138 
139 		if (len > PAGE_SIZE)
140 			len = PAGE_SIZE;
141 		len -= sizeof(struct virtio_net_hdr_mrg_rxbuf);
142 
143 		memcpy(&hdr->mhdr, p, sizeof(hdr->mhdr));
144 		p += sizeof(hdr->mhdr);
145 
146 		copy = len;
147 		if (copy > skb_tailroom(skb))
148 			copy = skb_tailroom(skb);
149 
150 		memcpy(skb_put(skb, copy), p, copy);
151 
152 		len -= copy;
153 
154 		if (!len) {
155 			give_a_page(vi, skb_shinfo(skb)->frags[0].page);
156 			skb_shinfo(skb)->nr_frags--;
157 		} else {
158 			skb_shinfo(skb)->frags[0].page_offset +=
159 				sizeof(hdr->mhdr) + copy;
160 			skb_shinfo(skb)->frags[0].size = len;
161 			skb->data_len += len;
162 			skb->len += len;
163 		}
164 
165 		while (--hdr->mhdr.num_buffers) {
166 			struct sk_buff *nskb;
167 
168 			i = skb_shinfo(skb)->nr_frags;
169 			if (i >= MAX_SKB_FRAGS) {
170 				pr_debug("%s: packet too long %d\n", dev->name,
171 					 len);
172 				dev->stats.rx_length_errors++;
173 				goto drop;
174 			}
175 
176 			nskb = vi->rvq->vq_ops->get_buf(vi->rvq, &len);
177 			if (!nskb) {
178 				pr_debug("%s: rx error: %d buffers missing\n",
179 					 dev->name, hdr->mhdr.num_buffers);
180 				dev->stats.rx_length_errors++;
181 				goto drop;
182 			}
183 
184 			__skb_unlink(nskb, &vi->recv);
185 			vi->num--;
186 
187 			skb_shinfo(skb)->frags[i] = skb_shinfo(nskb)->frags[0];
188 			skb_shinfo(nskb)->nr_frags = 0;
189 			kfree_skb(nskb);
190 
191 			if (len > PAGE_SIZE)
192 				len = PAGE_SIZE;
193 
194 			skb_shinfo(skb)->frags[i].size = len;
195 			skb_shinfo(skb)->nr_frags++;
196 			skb->data_len += len;
197 			skb->len += len;
198 		}
199 	} else {
200 		len -= sizeof(hdr->hdr);
201 
202 		if (len <= MAX_PACKET_LEN)
203 			trim_pages(vi, skb);
204 
205 		err = pskb_trim(skb, len);
206 		if (err) {
207 			pr_debug("%s: pskb_trim failed %i %d\n", dev->name,
208 				 len, err);
209 			dev->stats.rx_dropped++;
210 			goto drop;
211 		}
212 	}
213 
214 	skb->truesize += skb->data_len;
215 	dev->stats.rx_bytes += skb->len;
216 	dev->stats.rx_packets++;
217 
218 	if (hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
219 		pr_debug("Needs csum!\n");
220 		if (!skb_partial_csum_set(skb,
221 					  hdr->hdr.csum_start,
222 					  hdr->hdr.csum_offset))
223 			goto frame_err;
224 	}
225 
226 	skb->protocol = eth_type_trans(skb, dev);
227 	pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
228 		 ntohs(skb->protocol), skb->len, skb->pkt_type);
229 
230 	if (hdr->hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
231 		pr_debug("GSO!\n");
232 		switch (hdr->hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
233 		case VIRTIO_NET_HDR_GSO_TCPV4:
234 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
235 			break;
236 		case VIRTIO_NET_HDR_GSO_UDP:
237 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
238 			break;
239 		case VIRTIO_NET_HDR_GSO_TCPV6:
240 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
241 			break;
242 		default:
243 			if (net_ratelimit())
244 				printk(KERN_WARNING "%s: bad gso type %u.\n",
245 				       dev->name, hdr->hdr.gso_type);
246 			goto frame_err;
247 		}
248 
249 		if (hdr->hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
250 			skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
251 
252 		skb_shinfo(skb)->gso_size = hdr->hdr.gso_size;
253 		if (skb_shinfo(skb)->gso_size == 0) {
254 			if (net_ratelimit())
255 				printk(KERN_WARNING "%s: zero gso size.\n",
256 				       dev->name);
257 			goto frame_err;
258 		}
259 
260 		/* Header must be checked, and gso_segs computed. */
261 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
262 		skb_shinfo(skb)->gso_segs = 0;
263 	}
264 
265 	netif_receive_skb(skb);
266 	return;
267 
268 frame_err:
269 	dev->stats.rx_frame_errors++;
270 drop:
271 	dev_kfree_skb(skb);
272 }
273 
274 static bool try_fill_recv_maxbufs(struct virtnet_info *vi, gfp_t gfp)
275 {
276 	struct sk_buff *skb;
277 	struct scatterlist sg[2+MAX_SKB_FRAGS];
278 	int num, err, i;
279 	bool oom = false;
280 
281 	sg_init_table(sg, 2+MAX_SKB_FRAGS);
282 	do {
283 		struct skb_vnet_hdr *hdr;
284 
285 		skb = netdev_alloc_skb_ip_align(vi->dev, MAX_PACKET_LEN);
286 		if (unlikely(!skb)) {
287 			oom = true;
288 			break;
289 		}
290 
291 		skb_put(skb, MAX_PACKET_LEN);
292 
293 		hdr = skb_vnet_hdr(skb);
294 		sg_set_buf(sg, &hdr->hdr, sizeof(hdr->hdr));
295 
296 		if (vi->big_packets) {
297 			for (i = 0; i < MAX_SKB_FRAGS; i++) {
298 				skb_frag_t *f = &skb_shinfo(skb)->frags[i];
299 				f->page = get_a_page(vi, gfp);
300 				if (!f->page)
301 					break;
302 
303 				f->page_offset = 0;
304 				f->size = PAGE_SIZE;
305 
306 				skb->data_len += PAGE_SIZE;
307 				skb->len += PAGE_SIZE;
308 
309 				skb_shinfo(skb)->nr_frags++;
310 			}
311 		}
312 
313 		num = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
314 		skb_queue_head(&vi->recv, skb);
315 
316 		err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, num, skb);
317 		if (err < 0) {
318 			skb_unlink(skb, &vi->recv);
319 			trim_pages(vi, skb);
320 			kfree_skb(skb);
321 			break;
322 		}
323 		vi->num++;
324 	} while (err >= num);
325 	if (unlikely(vi->num > vi->max))
326 		vi->max = vi->num;
327 	vi->rvq->vq_ops->kick(vi->rvq);
328 	return !oom;
329 }
330 
331 /* Returns false if we couldn't fill entirely (OOM). */
332 static bool try_fill_recv(struct virtnet_info *vi, gfp_t gfp)
333 {
334 	struct sk_buff *skb;
335 	struct scatterlist sg[1];
336 	int err;
337 	bool oom = false;
338 
339 	if (!vi->mergeable_rx_bufs)
340 		return try_fill_recv_maxbufs(vi, gfp);
341 
342 	do {
343 		skb_frag_t *f;
344 
345 		skb = netdev_alloc_skb_ip_align(vi->dev, GOOD_COPY_LEN);
346 		if (unlikely(!skb)) {
347 			oom = true;
348 			break;
349 		}
350 
351 		f = &skb_shinfo(skb)->frags[0];
352 		f->page = get_a_page(vi, gfp);
353 		if (!f->page) {
354 			oom = true;
355 			kfree_skb(skb);
356 			break;
357 		}
358 
359 		f->page_offset = 0;
360 		f->size = PAGE_SIZE;
361 
362 		skb_shinfo(skb)->nr_frags++;
363 
364 		sg_init_one(sg, page_address(f->page), PAGE_SIZE);
365 		skb_queue_head(&vi->recv, skb);
366 
367 		err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, 1, skb);
368 		if (err < 0) {
369 			skb_unlink(skb, &vi->recv);
370 			kfree_skb(skb);
371 			break;
372 		}
373 		vi->num++;
374 	} while (err > 0);
375 	if (unlikely(vi->num > vi->max))
376 		vi->max = vi->num;
377 	vi->rvq->vq_ops->kick(vi->rvq);
378 	return !oom;
379 }
380 
381 static void skb_recv_done(struct virtqueue *rvq)
382 {
383 	struct virtnet_info *vi = rvq->vdev->priv;
384 	/* Schedule NAPI, Suppress further interrupts if successful. */
385 	if (napi_schedule_prep(&vi->napi)) {
386 		rvq->vq_ops->disable_cb(rvq);
387 		__napi_schedule(&vi->napi);
388 	}
389 }
390 
391 static void refill_work(struct work_struct *work)
392 {
393 	struct virtnet_info *vi;
394 	bool still_empty;
395 
396 	vi = container_of(work, struct virtnet_info, refill.work);
397 	napi_disable(&vi->napi);
398 	still_empty = !try_fill_recv(vi, GFP_KERNEL);
399 	napi_enable(&vi->napi);
400 
401 	/* In theory, this can happen: if we don't get any buffers in
402 	 * we will *never* try to fill again. */
403 	if (still_empty)
404 		schedule_delayed_work(&vi->refill, HZ/2);
405 }
406 
407 static int virtnet_poll(struct napi_struct *napi, int budget)
408 {
409 	struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi);
410 	struct sk_buff *skb = NULL;
411 	unsigned int len, received = 0;
412 
413 again:
414 	while (received < budget &&
415 	       (skb = vi->rvq->vq_ops->get_buf(vi->rvq, &len)) != NULL) {
416 		__skb_unlink(skb, &vi->recv);
417 		receive_skb(vi->dev, skb, len);
418 		vi->num--;
419 		received++;
420 	}
421 
422 	if (vi->num < vi->max / 2) {
423 		if (!try_fill_recv(vi, GFP_ATOMIC))
424 			schedule_delayed_work(&vi->refill, 0);
425 	}
426 
427 	/* Out of packets? */
428 	if (received < budget) {
429 		napi_complete(napi);
430 		if (unlikely(!vi->rvq->vq_ops->enable_cb(vi->rvq)) &&
431 		    napi_schedule_prep(napi)) {
432 			vi->rvq->vq_ops->disable_cb(vi->rvq);
433 			__napi_schedule(napi);
434 			goto again;
435 		}
436 	}
437 
438 	return received;
439 }
440 
441 static unsigned int free_old_xmit_skbs(struct virtnet_info *vi)
442 {
443 	struct sk_buff *skb;
444 	unsigned int len, tot_sgs = 0;
445 
446 	while ((skb = vi->svq->vq_ops->get_buf(vi->svq, &len)) != NULL) {
447 		pr_debug("Sent skb %p\n", skb);
448 		__skb_unlink(skb, &vi->send);
449 		vi->dev->stats.tx_bytes += skb->len;
450 		vi->dev->stats.tx_packets++;
451 		tot_sgs += skb_vnet_hdr(skb)->num_sg;
452 		dev_kfree_skb_any(skb);
453 	}
454 	return tot_sgs;
455 }
456 
457 static int xmit_skb(struct virtnet_info *vi, struct sk_buff *skb)
458 {
459 	struct scatterlist sg[2+MAX_SKB_FRAGS];
460 	struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
461 	const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
462 
463 	sg_init_table(sg, 2+MAX_SKB_FRAGS);
464 
465 	pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
466 
467 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
468 		hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
469 		hdr->hdr.csum_start = skb->csum_start - skb_headroom(skb);
470 		hdr->hdr.csum_offset = skb->csum_offset;
471 	} else {
472 		hdr->hdr.flags = 0;
473 		hdr->hdr.csum_offset = hdr->hdr.csum_start = 0;
474 	}
475 
476 	if (skb_is_gso(skb)) {
477 		hdr->hdr.hdr_len = skb_headlen(skb);
478 		hdr->hdr.gso_size = skb_shinfo(skb)->gso_size;
479 		if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
480 			hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
481 		else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
482 			hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
483 		else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
484 			hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
485 		else
486 			BUG();
487 		if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
488 			hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
489 	} else {
490 		hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
491 		hdr->hdr.gso_size = hdr->hdr.hdr_len = 0;
492 	}
493 
494 	hdr->mhdr.num_buffers = 0;
495 
496 	/* Encode metadata header at front. */
497 	if (vi->mergeable_rx_bufs)
498 		sg_set_buf(sg, &hdr->mhdr, sizeof(hdr->mhdr));
499 	else
500 		sg_set_buf(sg, &hdr->hdr, sizeof(hdr->hdr));
501 
502 	hdr->num_sg = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
503 	return vi->svq->vq_ops->add_buf(vi->svq, sg, hdr->num_sg, 0, skb);
504 }
505 
506 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
507 {
508 	struct virtnet_info *vi = netdev_priv(dev);
509 	int capacity;
510 
511 again:
512 	/* Free up any pending old buffers before queueing new ones. */
513 	free_old_xmit_skbs(vi);
514 
515 	/* Try to transmit */
516 	capacity = xmit_skb(vi, skb);
517 
518 	/* This can happen with OOM and indirect buffers. */
519 	if (unlikely(capacity < 0)) {
520 		netif_stop_queue(dev);
521 		dev_warn(&dev->dev, "Unexpected full queue\n");
522 		if (unlikely(!vi->svq->vq_ops->enable_cb(vi->svq))) {
523 			vi->svq->vq_ops->disable_cb(vi->svq);
524 			netif_start_queue(dev);
525 			goto again;
526 		}
527 		return NETDEV_TX_BUSY;
528 	}
529 	vi->svq->vq_ops->kick(vi->svq);
530 
531 	/*
532 	 * Put new one in send queue.  You'd expect we'd need this before
533 	 * xmit_skb calls add_buf(), since the callback can be triggered
534 	 * immediately after that.  But since the callback just triggers
535 	 * another call back here, normal network xmit locking prevents the
536 	 * race.
537 	 */
538 	__skb_queue_head(&vi->send, skb);
539 
540 	/* Don't wait up for transmitted skbs to be freed. */
541 	skb_orphan(skb);
542 	nf_reset(skb);
543 
544 	/* Apparently nice girls don't return TX_BUSY; stop the queue
545 	 * before it gets out of hand.  Naturally, this wastes entries. */
546 	if (capacity < 2+MAX_SKB_FRAGS) {
547 		netif_stop_queue(dev);
548 		if (unlikely(!vi->svq->vq_ops->enable_cb(vi->svq))) {
549 			/* More just got used, free them then recheck. */
550 			capacity += free_old_xmit_skbs(vi);
551 			if (capacity >= 2+MAX_SKB_FRAGS) {
552 				netif_start_queue(dev);
553 				vi->svq->vq_ops->disable_cb(vi->svq);
554 			}
555 		}
556 	}
557 
558 	return NETDEV_TX_OK;
559 }
560 
561 static int virtnet_set_mac_address(struct net_device *dev, void *p)
562 {
563 	struct virtnet_info *vi = netdev_priv(dev);
564 	struct virtio_device *vdev = vi->vdev;
565 	int ret;
566 
567 	ret = eth_mac_addr(dev, p);
568 	if (ret)
569 		return ret;
570 
571 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
572 		vdev->config->set(vdev, offsetof(struct virtio_net_config, mac),
573 		                  dev->dev_addr, dev->addr_len);
574 
575 	return 0;
576 }
577 
578 #ifdef CONFIG_NET_POLL_CONTROLLER
579 static void virtnet_netpoll(struct net_device *dev)
580 {
581 	struct virtnet_info *vi = netdev_priv(dev);
582 
583 	napi_schedule(&vi->napi);
584 }
585 #endif
586 
587 static int virtnet_open(struct net_device *dev)
588 {
589 	struct virtnet_info *vi = netdev_priv(dev);
590 
591 	napi_enable(&vi->napi);
592 
593 	/* If all buffers were filled by other side before we napi_enabled, we
594 	 * won't get another interrupt, so process any outstanding packets
595 	 * now.  virtnet_poll wants re-enable the queue, so we disable here.
596 	 * We synchronize against interrupts via NAPI_STATE_SCHED */
597 	if (napi_schedule_prep(&vi->napi)) {
598 		vi->rvq->vq_ops->disable_cb(vi->rvq);
599 		__napi_schedule(&vi->napi);
600 	}
601 	return 0;
602 }
603 
604 /*
605  * Send command via the control virtqueue and check status.  Commands
606  * supported by the hypervisor, as indicated by feature bits, should
607  * never fail unless improperly formated.
608  */
609 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
610 				 struct scatterlist *data, int out, int in)
611 {
612 	struct scatterlist *s, sg[VIRTNET_SEND_COMMAND_SG_MAX + 2];
613 	struct virtio_net_ctrl_hdr ctrl;
614 	virtio_net_ctrl_ack status = ~0;
615 	unsigned int tmp;
616 	int i;
617 
618 	/* Caller should know better */
619 	BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ||
620 		(out + in > VIRTNET_SEND_COMMAND_SG_MAX));
621 
622 	out++; /* Add header */
623 	in++; /* Add return status */
624 
625 	ctrl.class = class;
626 	ctrl.cmd = cmd;
627 
628 	sg_init_table(sg, out + in);
629 
630 	sg_set_buf(&sg[0], &ctrl, sizeof(ctrl));
631 	for_each_sg(data, s, out + in - 2, i)
632 		sg_set_buf(&sg[i + 1], sg_virt(s), s->length);
633 	sg_set_buf(&sg[out + in - 1], &status, sizeof(status));
634 
635 	BUG_ON(vi->cvq->vq_ops->add_buf(vi->cvq, sg, out, in, vi) < 0);
636 
637 	vi->cvq->vq_ops->kick(vi->cvq);
638 
639 	/*
640 	 * Spin for a response, the kick causes an ioport write, trapping
641 	 * into the hypervisor, so the request should be handled immediately.
642 	 */
643 	while (!vi->cvq->vq_ops->get_buf(vi->cvq, &tmp))
644 		cpu_relax();
645 
646 	return status == VIRTIO_NET_OK;
647 }
648 
649 static int virtnet_close(struct net_device *dev)
650 {
651 	struct virtnet_info *vi = netdev_priv(dev);
652 
653 	napi_disable(&vi->napi);
654 
655 	return 0;
656 }
657 
658 static int virtnet_set_tx_csum(struct net_device *dev, u32 data)
659 {
660 	struct virtnet_info *vi = netdev_priv(dev);
661 	struct virtio_device *vdev = vi->vdev;
662 
663 	if (data && !virtio_has_feature(vdev, VIRTIO_NET_F_CSUM))
664 		return -ENOSYS;
665 
666 	return ethtool_op_set_tx_hw_csum(dev, data);
667 }
668 
669 static void virtnet_set_rx_mode(struct net_device *dev)
670 {
671 	struct virtnet_info *vi = netdev_priv(dev);
672 	struct scatterlist sg[2];
673 	u8 promisc, allmulti;
674 	struct virtio_net_ctrl_mac *mac_data;
675 	struct dev_addr_list *addr;
676 	struct netdev_hw_addr *ha;
677 	void *buf;
678 	int i;
679 
680 	/* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */
681 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
682 		return;
683 
684 	promisc = ((dev->flags & IFF_PROMISC) != 0);
685 	allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
686 
687 	sg_init_one(sg, &promisc, sizeof(promisc));
688 
689 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
690 				  VIRTIO_NET_CTRL_RX_PROMISC,
691 				  sg, 1, 0))
692 		dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
693 			 promisc ? "en" : "dis");
694 
695 	sg_init_one(sg, &allmulti, sizeof(allmulti));
696 
697 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
698 				  VIRTIO_NET_CTRL_RX_ALLMULTI,
699 				  sg, 1, 0))
700 		dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
701 			 allmulti ? "en" : "dis");
702 
703 	/* MAC filter - use one buffer for both lists */
704 	mac_data = buf = kzalloc(((dev->uc.count + dev->mc_count) * ETH_ALEN) +
705 				 (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
706 	if (!buf) {
707 		dev_warn(&dev->dev, "No memory for MAC address buffer\n");
708 		return;
709 	}
710 
711 	sg_init_table(sg, 2);
712 
713 	/* Store the unicast list and count in the front of the buffer */
714 	mac_data->entries = dev->uc.count;
715 	i = 0;
716 	list_for_each_entry(ha, &dev->uc.list, list)
717 		memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
718 
719 	sg_set_buf(&sg[0], mac_data,
720 		   sizeof(mac_data->entries) + (dev->uc.count * ETH_ALEN));
721 
722 	/* multicast list and count fill the end */
723 	mac_data = (void *)&mac_data->macs[dev->uc.count][0];
724 
725 	mac_data->entries = dev->mc_count;
726 	addr = dev->mc_list;
727 	for (i = 0; i < dev->mc_count; i++, addr = addr->next)
728 		memcpy(&mac_data->macs[i][0], addr->da_addr, ETH_ALEN);
729 
730 	sg_set_buf(&sg[1], mac_data,
731 		   sizeof(mac_data->entries) + (dev->mc_count * ETH_ALEN));
732 
733 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
734 				  VIRTIO_NET_CTRL_MAC_TABLE_SET,
735 				  sg, 2, 0))
736 		dev_warn(&dev->dev, "Failed to set MAC fitler table.\n");
737 
738 	kfree(buf);
739 }
740 
741 static void virtnet_vlan_rx_add_vid(struct net_device *dev, u16 vid)
742 {
743 	struct virtnet_info *vi = netdev_priv(dev);
744 	struct scatterlist sg;
745 
746 	sg_init_one(&sg, &vid, sizeof(vid));
747 
748 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
749 				  VIRTIO_NET_CTRL_VLAN_ADD, &sg, 1, 0))
750 		dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
751 }
752 
753 static void virtnet_vlan_rx_kill_vid(struct net_device *dev, u16 vid)
754 {
755 	struct virtnet_info *vi = netdev_priv(dev);
756 	struct scatterlist sg;
757 
758 	sg_init_one(&sg, &vid, sizeof(vid));
759 
760 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
761 				  VIRTIO_NET_CTRL_VLAN_DEL, &sg, 1, 0))
762 		dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
763 }
764 
765 static const struct ethtool_ops virtnet_ethtool_ops = {
766 	.set_tx_csum = virtnet_set_tx_csum,
767 	.set_sg = ethtool_op_set_sg,
768 	.set_tso = ethtool_op_set_tso,
769 	.set_ufo = ethtool_op_set_ufo,
770 	.get_link = ethtool_op_get_link,
771 };
772 
773 #define MIN_MTU 68
774 #define MAX_MTU 65535
775 
776 static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
777 {
778 	if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
779 		return -EINVAL;
780 	dev->mtu = new_mtu;
781 	return 0;
782 }
783 
784 static const struct net_device_ops virtnet_netdev = {
785 	.ndo_open            = virtnet_open,
786 	.ndo_stop   	     = virtnet_close,
787 	.ndo_start_xmit      = start_xmit,
788 	.ndo_validate_addr   = eth_validate_addr,
789 	.ndo_set_mac_address = virtnet_set_mac_address,
790 	.ndo_set_rx_mode     = virtnet_set_rx_mode,
791 	.ndo_change_mtu	     = virtnet_change_mtu,
792 	.ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
793 	.ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
794 #ifdef CONFIG_NET_POLL_CONTROLLER
795 	.ndo_poll_controller = virtnet_netpoll,
796 #endif
797 };
798 
799 static void virtnet_update_status(struct virtnet_info *vi)
800 {
801 	u16 v;
802 
803 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS))
804 		return;
805 
806 	vi->vdev->config->get(vi->vdev,
807 			      offsetof(struct virtio_net_config, status),
808 			      &v, sizeof(v));
809 
810 	/* Ignore unknown (future) status bits */
811 	v &= VIRTIO_NET_S_LINK_UP;
812 
813 	if (vi->status == v)
814 		return;
815 
816 	vi->status = v;
817 
818 	if (vi->status & VIRTIO_NET_S_LINK_UP) {
819 		netif_carrier_on(vi->dev);
820 		netif_wake_queue(vi->dev);
821 	} else {
822 		netif_carrier_off(vi->dev);
823 		netif_stop_queue(vi->dev);
824 	}
825 }
826 
827 static void virtnet_config_changed(struct virtio_device *vdev)
828 {
829 	struct virtnet_info *vi = vdev->priv;
830 
831 	virtnet_update_status(vi);
832 }
833 
834 static int virtnet_probe(struct virtio_device *vdev)
835 {
836 	int err;
837 	struct net_device *dev;
838 	struct virtnet_info *vi;
839 	struct virtqueue *vqs[3];
840 	vq_callback_t *callbacks[] = { skb_recv_done, skb_xmit_done, NULL};
841 	const char *names[] = { "input", "output", "control" };
842 	int nvqs;
843 
844 	/* Allocate ourselves a network device with room for our info */
845 	dev = alloc_etherdev(sizeof(struct virtnet_info));
846 	if (!dev)
847 		return -ENOMEM;
848 
849 	/* Set up network device as normal. */
850 	dev->netdev_ops = &virtnet_netdev;
851 	dev->features = NETIF_F_HIGHDMA;
852 	SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
853 	SET_NETDEV_DEV(dev, &vdev->dev);
854 
855 	/* Do we support "hardware" checksums? */
856 	if (csum && virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
857 		/* This opens up the world of extra features. */
858 		dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
859 		if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
860 			dev->features |= NETIF_F_TSO | NETIF_F_UFO
861 				| NETIF_F_TSO_ECN | NETIF_F_TSO6;
862 		}
863 		/* Individual feature bits: what can host handle? */
864 		if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
865 			dev->features |= NETIF_F_TSO;
866 		if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
867 			dev->features |= NETIF_F_TSO6;
868 		if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
869 			dev->features |= NETIF_F_TSO_ECN;
870 		if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
871 			dev->features |= NETIF_F_UFO;
872 	}
873 
874 	/* Configuration may specify what MAC to use.  Otherwise random. */
875 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
876 		vdev->config->get(vdev,
877 				  offsetof(struct virtio_net_config, mac),
878 				  dev->dev_addr, dev->addr_len);
879 	} else
880 		random_ether_addr(dev->dev_addr);
881 
882 	/* Set up our device-specific information */
883 	vi = netdev_priv(dev);
884 	netif_napi_add(dev, &vi->napi, virtnet_poll, napi_weight);
885 	vi->dev = dev;
886 	vi->vdev = vdev;
887 	vdev->priv = vi;
888 	vi->pages = NULL;
889 	INIT_DELAYED_WORK(&vi->refill, refill_work);
890 
891 	/* If we can receive ANY GSO packets, we must allocate large ones. */
892 	if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
893 	    virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
894 	    virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
895 		vi->big_packets = true;
896 
897 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
898 		vi->mergeable_rx_bufs = true;
899 
900 	/* We expect two virtqueues, receive then send,
901 	 * and optionally control. */
902 	nvqs = virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ? 3 : 2;
903 
904 	err = vdev->config->find_vqs(vdev, nvqs, vqs, callbacks, names);
905 	if (err)
906 		goto free;
907 
908 	vi->rvq = vqs[0];
909 	vi->svq = vqs[1];
910 
911 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)) {
912 		vi->cvq = vqs[2];
913 
914 		if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
915 			dev->features |= NETIF_F_HW_VLAN_FILTER;
916 	}
917 
918 	/* Initialize our empty receive and send queues. */
919 	skb_queue_head_init(&vi->recv);
920 	skb_queue_head_init(&vi->send);
921 
922 	err = register_netdev(dev);
923 	if (err) {
924 		pr_debug("virtio_net: registering device failed\n");
925 		goto free_vqs;
926 	}
927 
928 	/* Last of all, set up some receive buffers. */
929 	try_fill_recv(vi, GFP_KERNEL);
930 
931 	/* If we didn't even get one input buffer, we're useless. */
932 	if (vi->num == 0) {
933 		err = -ENOMEM;
934 		goto unregister;
935 	}
936 
937 	vi->status = VIRTIO_NET_S_LINK_UP;
938 	virtnet_update_status(vi);
939 	netif_carrier_on(dev);
940 
941 	pr_debug("virtnet: registered device %s\n", dev->name);
942 	return 0;
943 
944 unregister:
945 	unregister_netdev(dev);
946 	cancel_delayed_work_sync(&vi->refill);
947 free_vqs:
948 	vdev->config->del_vqs(vdev);
949 free:
950 	free_netdev(dev);
951 	return err;
952 }
953 
954 static void __devexit virtnet_remove(struct virtio_device *vdev)
955 {
956 	struct virtnet_info *vi = vdev->priv;
957 	struct sk_buff *skb;
958 
959 	/* Stop all the virtqueues. */
960 	vdev->config->reset(vdev);
961 
962 	/* Free our skbs in send and recv queues, if any. */
963 	while ((skb = __skb_dequeue(&vi->recv)) != NULL) {
964 		kfree_skb(skb);
965 		vi->num--;
966 	}
967 	__skb_queue_purge(&vi->send);
968 
969 	BUG_ON(vi->num != 0);
970 
971 	unregister_netdev(vi->dev);
972 	cancel_delayed_work_sync(&vi->refill);
973 
974 	vdev->config->del_vqs(vi->vdev);
975 
976 	while (vi->pages)
977 		__free_pages(get_a_page(vi, GFP_KERNEL), 0);
978 
979 	free_netdev(vi->dev);
980 }
981 
982 static struct virtio_device_id id_table[] = {
983 	{ VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
984 	{ 0 },
985 };
986 
987 static unsigned int features[] = {
988 	VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
989 	VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
990 	VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
991 	VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
992 	VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO,
993 	VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ,
994 	VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN,
995 };
996 
997 static struct virtio_driver virtio_net_driver = {
998 	.feature_table = features,
999 	.feature_table_size = ARRAY_SIZE(features),
1000 	.driver.name =	KBUILD_MODNAME,
1001 	.driver.owner =	THIS_MODULE,
1002 	.id_table =	id_table,
1003 	.probe =	virtnet_probe,
1004 	.remove =	__devexit_p(virtnet_remove),
1005 	.config_changed = virtnet_config_changed,
1006 };
1007 
1008 static int __init init(void)
1009 {
1010 	return register_virtio_driver(&virtio_net_driver);
1011 }
1012 
1013 static void __exit fini(void)
1014 {
1015 	unregister_virtio_driver(&virtio_net_driver);
1016 }
1017 module_init(init);
1018 module_exit(fini);
1019 
1020 MODULE_DEVICE_TABLE(virtio, id_table);
1021 MODULE_DESCRIPTION("Virtio network driver");
1022 MODULE_LICENSE("GPL");
1023