xref: /openbmc/linux/drivers/net/virtio_net.c (revision 78c99ba1)
1 /* A simple 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 	/* The skb we couldn't send because buffers were full. */
51 	struct sk_buff *last_xmit_skb;
52 
53 	/* If we need to free in a timer, this is it. */
54 	struct timer_list xmit_free_timer;
55 
56 	/* Number of input buffers, and max we've ever had. */
57 	unsigned int num, max;
58 
59 	/* For cleaning up after transmission. */
60 	struct tasklet_struct tasklet;
61 	bool free_in_tasklet;
62 
63 	/* I like... big packets and I cannot lie! */
64 	bool big_packets;
65 
66 	/* Host will merge rx buffers for big packets (shake it! shake it!) */
67 	bool mergeable_rx_bufs;
68 
69 	/* Receive & send queues. */
70 	struct sk_buff_head recv;
71 	struct sk_buff_head send;
72 
73 	/* Chain pages by the private ptr. */
74 	struct page *pages;
75 };
76 
77 static inline void *skb_vnet_hdr(struct sk_buff *skb)
78 {
79 	return (struct virtio_net_hdr *)skb->cb;
80 }
81 
82 static void give_a_page(struct virtnet_info *vi, struct page *page)
83 {
84 	page->private = (unsigned long)vi->pages;
85 	vi->pages = page;
86 }
87 
88 static void trim_pages(struct virtnet_info *vi, struct sk_buff *skb)
89 {
90 	unsigned int i;
91 
92 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
93 		give_a_page(vi, skb_shinfo(skb)->frags[i].page);
94 	skb_shinfo(skb)->nr_frags = 0;
95 	skb->data_len = 0;
96 }
97 
98 static struct page *get_a_page(struct virtnet_info *vi, gfp_t gfp_mask)
99 {
100 	struct page *p = vi->pages;
101 
102 	if (p)
103 		vi->pages = (struct page *)p->private;
104 	else
105 		p = alloc_page(gfp_mask);
106 	return p;
107 }
108 
109 static void skb_xmit_done(struct virtqueue *svq)
110 {
111 	struct virtnet_info *vi = svq->vdev->priv;
112 
113 	/* Suppress further interrupts. */
114 	svq->vq_ops->disable_cb(svq);
115 
116 	/* We were probably waiting for more output buffers. */
117 	netif_wake_queue(vi->dev);
118 
119 	/* Make sure we re-xmit last_xmit_skb: if there are no more packets
120 	 * queued, start_xmit won't be called. */
121 	tasklet_schedule(&vi->tasklet);
122 }
123 
124 static void receive_skb(struct net_device *dev, struct sk_buff *skb,
125 			unsigned len)
126 {
127 	struct virtnet_info *vi = netdev_priv(dev);
128 	struct virtio_net_hdr *hdr = skb_vnet_hdr(skb);
129 	int err;
130 	int i;
131 
132 	if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
133 		pr_debug("%s: short packet %i\n", dev->name, len);
134 		dev->stats.rx_length_errors++;
135 		goto drop;
136 	}
137 
138 	if (vi->mergeable_rx_bufs) {
139 		struct virtio_net_hdr_mrg_rxbuf *mhdr = skb_vnet_hdr(skb);
140 		unsigned int copy;
141 		char *p = page_address(skb_shinfo(skb)->frags[0].page);
142 
143 		if (len > PAGE_SIZE)
144 			len = PAGE_SIZE;
145 		len -= sizeof(struct virtio_net_hdr_mrg_rxbuf);
146 
147 		memcpy(hdr, p, sizeof(*mhdr));
148 		p += sizeof(*mhdr);
149 
150 		copy = len;
151 		if (copy > skb_tailroom(skb))
152 			copy = skb_tailroom(skb);
153 
154 		memcpy(skb_put(skb, copy), p, copy);
155 
156 		len -= copy;
157 
158 		if (!len) {
159 			give_a_page(vi, skb_shinfo(skb)->frags[0].page);
160 			skb_shinfo(skb)->nr_frags--;
161 		} else {
162 			skb_shinfo(skb)->frags[0].page_offset +=
163 				sizeof(*mhdr) + copy;
164 			skb_shinfo(skb)->frags[0].size = len;
165 			skb->data_len += len;
166 			skb->len += len;
167 		}
168 
169 		while (--mhdr->num_buffers) {
170 			struct sk_buff *nskb;
171 
172 			i = skb_shinfo(skb)->nr_frags;
173 			if (i >= MAX_SKB_FRAGS) {
174 				pr_debug("%s: packet too long %d\n", dev->name,
175 					 len);
176 				dev->stats.rx_length_errors++;
177 				goto drop;
178 			}
179 
180 			nskb = vi->rvq->vq_ops->get_buf(vi->rvq, &len);
181 			if (!nskb) {
182 				pr_debug("%s: rx error: %d buffers missing\n",
183 					 dev->name, mhdr->num_buffers);
184 				dev->stats.rx_length_errors++;
185 				goto drop;
186 			}
187 
188 			__skb_unlink(nskb, &vi->recv);
189 			vi->num--;
190 
191 			skb_shinfo(skb)->frags[i] = skb_shinfo(nskb)->frags[0];
192 			skb_shinfo(nskb)->nr_frags = 0;
193 			kfree_skb(nskb);
194 
195 			if (len > PAGE_SIZE)
196 				len = PAGE_SIZE;
197 
198 			skb_shinfo(skb)->frags[i].size = len;
199 			skb_shinfo(skb)->nr_frags++;
200 			skb->data_len += len;
201 			skb->len += len;
202 		}
203 	} else {
204 		len -= sizeof(struct virtio_net_hdr);
205 
206 		if (len <= MAX_PACKET_LEN)
207 			trim_pages(vi, skb);
208 
209 		err = pskb_trim(skb, len);
210 		if (err) {
211 			pr_debug("%s: pskb_trim failed %i %d\n", dev->name,
212 				 len, err);
213 			dev->stats.rx_dropped++;
214 			goto drop;
215 		}
216 	}
217 
218 	skb->truesize += skb->data_len;
219 	dev->stats.rx_bytes += skb->len;
220 	dev->stats.rx_packets++;
221 
222 	if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
223 		pr_debug("Needs csum!\n");
224 		if (!skb_partial_csum_set(skb,hdr->csum_start,hdr->csum_offset))
225 			goto frame_err;
226 	}
227 
228 	skb->protocol = eth_type_trans(skb, dev);
229 	pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
230 		 ntohs(skb->protocol), skb->len, skb->pkt_type);
231 
232 	if (hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
233 		pr_debug("GSO!\n");
234 		switch (hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
235 		case VIRTIO_NET_HDR_GSO_TCPV4:
236 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
237 			break;
238 		case VIRTIO_NET_HDR_GSO_UDP:
239 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
240 			break;
241 		case VIRTIO_NET_HDR_GSO_TCPV6:
242 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
243 			break;
244 		default:
245 			if (net_ratelimit())
246 				printk(KERN_WARNING "%s: bad gso type %u.\n",
247 				       dev->name, hdr->gso_type);
248 			goto frame_err;
249 		}
250 
251 		if (hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
252 			skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
253 
254 		skb_shinfo(skb)->gso_size = hdr->gso_size;
255 		if (skb_shinfo(skb)->gso_size == 0) {
256 			if (net_ratelimit())
257 				printk(KERN_WARNING "%s: zero gso size.\n",
258 				       dev->name);
259 			goto frame_err;
260 		}
261 
262 		/* Header must be checked, and gso_segs computed. */
263 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
264 		skb_shinfo(skb)->gso_segs = 0;
265 	}
266 
267 	netif_receive_skb(skb);
268 	return;
269 
270 frame_err:
271 	dev->stats.rx_frame_errors++;
272 drop:
273 	dev_kfree_skb(skb);
274 }
275 
276 static void try_fill_recv_maxbufs(struct virtnet_info *vi)
277 {
278 	struct sk_buff *skb;
279 	struct scatterlist sg[2+MAX_SKB_FRAGS];
280 	int num, err, i;
281 
282 	sg_init_table(sg, 2+MAX_SKB_FRAGS);
283 	for (;;) {
284 		struct virtio_net_hdr *hdr;
285 
286 		skb = netdev_alloc_skb(vi->dev, MAX_PACKET_LEN);
287 		if (unlikely(!skb))
288 			break;
289 
290 		skb_put(skb, MAX_PACKET_LEN);
291 
292 		hdr = skb_vnet_hdr(skb);
293 		sg_set_buf(sg, hdr, sizeof(*hdr));
294 
295 		if (vi->big_packets) {
296 			for (i = 0; i < MAX_SKB_FRAGS; i++) {
297 				skb_frag_t *f = &skb_shinfo(skb)->frags[i];
298 				f->page = get_a_page(vi, GFP_ATOMIC);
299 				if (!f->page)
300 					break;
301 
302 				f->page_offset = 0;
303 				f->size = PAGE_SIZE;
304 
305 				skb->data_len += PAGE_SIZE;
306 				skb->len += PAGE_SIZE;
307 
308 				skb_shinfo(skb)->nr_frags++;
309 			}
310 		}
311 
312 		num = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
313 		skb_queue_head(&vi->recv, skb);
314 
315 		err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, num, skb);
316 		if (err) {
317 			skb_unlink(skb, &vi->recv);
318 			trim_pages(vi, skb);
319 			kfree_skb(skb);
320 			break;
321 		}
322 		vi->num++;
323 	}
324 	if (unlikely(vi->num > vi->max))
325 		vi->max = vi->num;
326 	vi->rvq->vq_ops->kick(vi->rvq);
327 }
328 
329 static void try_fill_recv(struct virtnet_info *vi)
330 {
331 	struct sk_buff *skb;
332 	struct scatterlist sg[1];
333 	int err;
334 
335 	if (!vi->mergeable_rx_bufs) {
336 		try_fill_recv_maxbufs(vi);
337 		return;
338 	}
339 
340 	for (;;) {
341 		skb_frag_t *f;
342 
343 		skb = netdev_alloc_skb(vi->dev, GOOD_COPY_LEN + NET_IP_ALIGN);
344 		if (unlikely(!skb))
345 			break;
346 
347 		skb_reserve(skb, NET_IP_ALIGN);
348 
349 		f = &skb_shinfo(skb)->frags[0];
350 		f->page = get_a_page(vi, GFP_ATOMIC);
351 		if (!f->page) {
352 			kfree_skb(skb);
353 			break;
354 		}
355 
356 		f->page_offset = 0;
357 		f->size = PAGE_SIZE;
358 
359 		skb_shinfo(skb)->nr_frags++;
360 
361 		sg_init_one(sg, page_address(f->page), PAGE_SIZE);
362 		skb_queue_head(&vi->recv, skb);
363 
364 		err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, 1, skb);
365 		if (err) {
366 			skb_unlink(skb, &vi->recv);
367 			kfree_skb(skb);
368 			break;
369 		}
370 		vi->num++;
371 	}
372 	if (unlikely(vi->num > vi->max))
373 		vi->max = vi->num;
374 	vi->rvq->vq_ops->kick(vi->rvq);
375 }
376 
377 static void skb_recv_done(struct virtqueue *rvq)
378 {
379 	struct virtnet_info *vi = rvq->vdev->priv;
380 	/* Schedule NAPI, Suppress further interrupts if successful. */
381 	if (napi_schedule_prep(&vi->napi)) {
382 		rvq->vq_ops->disable_cb(rvq);
383 		__napi_schedule(&vi->napi);
384 	}
385 }
386 
387 static int virtnet_poll(struct napi_struct *napi, int budget)
388 {
389 	struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi);
390 	struct sk_buff *skb = NULL;
391 	unsigned int len, received = 0;
392 
393 again:
394 	while (received < budget &&
395 	       (skb = vi->rvq->vq_ops->get_buf(vi->rvq, &len)) != NULL) {
396 		__skb_unlink(skb, &vi->recv);
397 		receive_skb(vi->dev, skb, len);
398 		vi->num--;
399 		received++;
400 	}
401 
402 	/* FIXME: If we oom and completely run out of inbufs, we need
403 	 * to start a timer trying to fill more. */
404 	if (vi->num < vi->max / 2)
405 		try_fill_recv(vi);
406 
407 	/* Out of packets? */
408 	if (received < budget) {
409 		napi_complete(napi);
410 		if (unlikely(!vi->rvq->vq_ops->enable_cb(vi->rvq))
411 		    && napi_schedule_prep(napi)) {
412 			vi->rvq->vq_ops->disable_cb(vi->rvq);
413 			__napi_schedule(napi);
414 			goto again;
415 		}
416 	}
417 
418 	return received;
419 }
420 
421 static void free_old_xmit_skbs(struct virtnet_info *vi)
422 {
423 	struct sk_buff *skb;
424 	unsigned int len;
425 
426 	while ((skb = vi->svq->vq_ops->get_buf(vi->svq, &len)) != NULL) {
427 		pr_debug("Sent skb %p\n", skb);
428 		__skb_unlink(skb, &vi->send);
429 		vi->dev->stats.tx_bytes += skb->len;
430 		vi->dev->stats.tx_packets++;
431 		kfree_skb(skb);
432 	}
433 }
434 
435 /* If the virtio transport doesn't always notify us when all in-flight packets
436  * are consumed, we fall back to using this function on a timer to free them. */
437 static void xmit_free(unsigned long data)
438 {
439 	struct virtnet_info *vi = (void *)data;
440 
441 	netif_tx_lock(vi->dev);
442 
443 	free_old_xmit_skbs(vi);
444 
445 	if (!skb_queue_empty(&vi->send))
446 		mod_timer(&vi->xmit_free_timer, jiffies + (HZ/10));
447 
448 	netif_tx_unlock(vi->dev);
449 }
450 
451 static int xmit_skb(struct virtnet_info *vi, struct sk_buff *skb)
452 {
453 	int num, err;
454 	struct scatterlist sg[2+MAX_SKB_FRAGS];
455 	struct virtio_net_hdr_mrg_rxbuf *mhdr = skb_vnet_hdr(skb);
456 	struct virtio_net_hdr *hdr = skb_vnet_hdr(skb);
457 	const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
458 
459 	sg_init_table(sg, 2+MAX_SKB_FRAGS);
460 
461 	pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
462 
463 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
464 		hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
465 		hdr->csum_start = skb->csum_start - skb_headroom(skb);
466 		hdr->csum_offset = skb->csum_offset;
467 	} else {
468 		hdr->flags = 0;
469 		hdr->csum_offset = hdr->csum_start = 0;
470 	}
471 
472 	if (skb_is_gso(skb)) {
473 		hdr->hdr_len = skb_transport_header(skb) - skb->data;
474 		hdr->gso_size = skb_shinfo(skb)->gso_size;
475 		if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
476 			hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
477 		else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
478 			hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
479 		else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
480 			hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
481 		else
482 			BUG();
483 		if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
484 			hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
485 	} else {
486 		hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
487 		hdr->gso_size = hdr->hdr_len = 0;
488 	}
489 
490 	mhdr->num_buffers = 0;
491 
492 	/* Encode metadata header at front. */
493 	if (vi->mergeable_rx_bufs)
494 		sg_set_buf(sg, mhdr, sizeof(*mhdr));
495 	else
496 		sg_set_buf(sg, hdr, sizeof(*hdr));
497 
498 	num = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
499 
500 	err = vi->svq->vq_ops->add_buf(vi->svq, sg, num, 0, skb);
501 	if (!err && !vi->free_in_tasklet)
502 		mod_timer(&vi->xmit_free_timer, jiffies + (HZ/10));
503 
504 	return err;
505 }
506 
507 static void xmit_tasklet(unsigned long data)
508 {
509 	struct virtnet_info *vi = (void *)data;
510 
511 	netif_tx_lock_bh(vi->dev);
512 	if (vi->last_xmit_skb && xmit_skb(vi, vi->last_xmit_skb) == 0) {
513 		vi->svq->vq_ops->kick(vi->svq);
514 		vi->last_xmit_skb = NULL;
515 	}
516 	if (vi->free_in_tasklet)
517 		free_old_xmit_skbs(vi);
518 	netif_tx_unlock_bh(vi->dev);
519 }
520 
521 static int start_xmit(struct sk_buff *skb, struct net_device *dev)
522 {
523 	struct virtnet_info *vi = netdev_priv(dev);
524 
525 again:
526 	/* Free up any pending old buffers before queueing new ones. */
527 	free_old_xmit_skbs(vi);
528 
529 	/* If we has a buffer left over from last time, send it now. */
530 	if (unlikely(vi->last_xmit_skb) &&
531 	    xmit_skb(vi, vi->last_xmit_skb) != 0)
532 		goto stop_queue;
533 
534 	vi->last_xmit_skb = NULL;
535 
536 	/* Put new one in send queue and do transmit */
537 	if (likely(skb)) {
538 		__skb_queue_head(&vi->send, skb);
539 		if (xmit_skb(vi, skb) != 0) {
540 			vi->last_xmit_skb = skb;
541 			skb = NULL;
542 			goto stop_queue;
543 		}
544 	}
545 done:
546 	vi->svq->vq_ops->kick(vi->svq);
547 	return NETDEV_TX_OK;
548 
549 stop_queue:
550 	pr_debug("%s: virtio not prepared to send\n", dev->name);
551 	netif_stop_queue(dev);
552 
553 	/* Activate callback for using skbs: if this returns false it
554 	 * means some were used in the meantime. */
555 	if (unlikely(!vi->svq->vq_ops->enable_cb(vi->svq))) {
556 		vi->svq->vq_ops->disable_cb(vi->svq);
557 		netif_start_queue(dev);
558 		goto again;
559 	}
560 	if (skb) {
561 		/* Drop this skb: we only queue one. */
562 		vi->dev->stats.tx_dropped++;
563 		kfree_skb(skb);
564 	}
565 	goto done;
566 }
567 
568 static int virtnet_set_mac_address(struct net_device *dev, void *p)
569 {
570 	struct virtnet_info *vi = netdev_priv(dev);
571 	struct virtio_device *vdev = vi->vdev;
572 	int ret;
573 
574 	ret = eth_mac_addr(dev, p);
575 	if (ret)
576 		return ret;
577 
578 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
579 		vdev->config->set(vdev, offsetof(struct virtio_net_config, mac),
580 		                  dev->dev_addr, dev->addr_len);
581 
582 	return 0;
583 }
584 
585 #ifdef CONFIG_NET_POLL_CONTROLLER
586 static void virtnet_netpoll(struct net_device *dev)
587 {
588 	struct virtnet_info *vi = netdev_priv(dev);
589 
590 	napi_schedule(&vi->napi);
591 }
592 #endif
593 
594 static int virtnet_open(struct net_device *dev)
595 {
596 	struct virtnet_info *vi = netdev_priv(dev);
597 
598 	napi_enable(&vi->napi);
599 
600 	/* If all buffers were filled by other side before we napi_enabled, we
601 	 * won't get another interrupt, so process any outstanding packets
602 	 * now.  virtnet_poll wants re-enable the queue, so we disable here.
603 	 * We synchronize against interrupts via NAPI_STATE_SCHED */
604 	if (napi_schedule_prep(&vi->napi)) {
605 		vi->rvq->vq_ops->disable_cb(vi->rvq);
606 		__napi_schedule(&vi->napi);
607 	}
608 	return 0;
609 }
610 
611 /*
612  * Send command via the control virtqueue and check status.  Commands
613  * supported by the hypervisor, as indicated by feature bits, should
614  * never fail unless improperly formated.
615  */
616 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
617 				 struct scatterlist *data, int out, int in)
618 {
619 	struct scatterlist *s, sg[VIRTNET_SEND_COMMAND_SG_MAX + 2];
620 	struct virtio_net_ctrl_hdr ctrl;
621 	virtio_net_ctrl_ack status = ~0;
622 	unsigned int tmp;
623 	int i;
624 
625 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)) {
626 		BUG();  /* Caller should know better */
627 		return false;
628 	}
629 
630 	BUG_ON(out + in > VIRTNET_SEND_COMMAND_SG_MAX);
631 
632 	out++; /* Add header */
633 	in++; /* Add return status */
634 
635 	ctrl.class = class;
636 	ctrl.cmd = cmd;
637 
638 	sg_init_table(sg, out + in);
639 
640 	sg_set_buf(&sg[0], &ctrl, sizeof(ctrl));
641 	for_each_sg(data, s, out + in - 2, i)
642 		sg_set_buf(&sg[i + 1], sg_virt(s), s->length);
643 	sg_set_buf(&sg[out + in - 1], &status, sizeof(status));
644 
645 	if (vi->cvq->vq_ops->add_buf(vi->cvq, sg, out, in, vi) != 0)
646 		BUG();
647 
648 	vi->cvq->vq_ops->kick(vi->cvq);
649 
650 	/*
651 	 * Spin for a response, the kick causes an ioport write, trapping
652 	 * into the hypervisor, so the request should be handled immediately.
653 	 */
654 	while (!vi->cvq->vq_ops->get_buf(vi->cvq, &tmp))
655 		cpu_relax();
656 
657 	return status == VIRTIO_NET_OK;
658 }
659 
660 static int virtnet_close(struct net_device *dev)
661 {
662 	struct virtnet_info *vi = netdev_priv(dev);
663 
664 	napi_disable(&vi->napi);
665 
666 	return 0;
667 }
668 
669 static int virtnet_set_tx_csum(struct net_device *dev, u32 data)
670 {
671 	struct virtnet_info *vi = netdev_priv(dev);
672 	struct virtio_device *vdev = vi->vdev;
673 
674 	if (data && !virtio_has_feature(vdev, VIRTIO_NET_F_CSUM))
675 		return -ENOSYS;
676 
677 	return ethtool_op_set_tx_hw_csum(dev, data);
678 }
679 
680 static void virtnet_set_rx_mode(struct net_device *dev)
681 {
682 	struct virtnet_info *vi = netdev_priv(dev);
683 	struct scatterlist sg[2];
684 	u8 promisc, allmulti;
685 	struct virtio_net_ctrl_mac *mac_data;
686 	struct dev_addr_list *addr;
687 	void *buf;
688 	int i;
689 
690 	/* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */
691 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
692 		return;
693 
694 	promisc = ((dev->flags & IFF_PROMISC) != 0);
695 	allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
696 
697 	sg_init_one(sg, &promisc, sizeof(promisc));
698 
699 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
700 				  VIRTIO_NET_CTRL_RX_PROMISC,
701 				  sg, 1, 0))
702 		dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
703 			 promisc ? "en" : "dis");
704 
705 	sg_init_one(sg, &allmulti, sizeof(allmulti));
706 
707 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
708 				  VIRTIO_NET_CTRL_RX_ALLMULTI,
709 				  sg, 1, 0))
710 		dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
711 			 allmulti ? "en" : "dis");
712 
713 	/* MAC filter - use one buffer for both lists */
714 	mac_data = buf = kzalloc(((dev->uc_count + dev->mc_count) * ETH_ALEN) +
715 				 (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
716 	if (!buf) {
717 		dev_warn(&dev->dev, "No memory for MAC address buffer\n");
718 		return;
719 	}
720 
721 	sg_init_table(sg, 2);
722 
723 	/* Store the unicast list and count in the front of the buffer */
724 	mac_data->entries = dev->uc_count;
725 	addr = dev->uc_list;
726 	for (i = 0; i < dev->uc_count; i++, addr = addr->next)
727 		memcpy(&mac_data->macs[i][0], addr->da_addr, ETH_ALEN);
728 
729 	sg_set_buf(&sg[0], mac_data,
730 		   sizeof(mac_data->entries) + (dev->uc_count * ETH_ALEN));
731 
732 	/* multicast list and count fill the end */
733 	mac_data = (void *)&mac_data->macs[dev->uc_count][0];
734 
735 	mac_data->entries = dev->mc_count;
736 	addr = dev->mc_list;
737 	for (i = 0; i < dev->mc_count; i++, addr = addr->next)
738 		memcpy(&mac_data->macs[i][0], addr->da_addr, ETH_ALEN);
739 
740 	sg_set_buf(&sg[1], mac_data,
741 		   sizeof(mac_data->entries) + (dev->mc_count * ETH_ALEN));
742 
743 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
744 				  VIRTIO_NET_CTRL_MAC_TABLE_SET,
745 				  sg, 2, 0))
746 		dev_warn(&dev->dev, "Failed to set MAC fitler table.\n");
747 
748 	kfree(buf);
749 }
750 
751 static void virtnet_vlan_rx_add_vid(struct net_device *dev, u16 vid)
752 {
753 	struct virtnet_info *vi = netdev_priv(dev);
754 	struct scatterlist sg;
755 
756 	sg_init_one(&sg, &vid, sizeof(vid));
757 
758 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
759 				  VIRTIO_NET_CTRL_VLAN_ADD, &sg, 1, 0))
760 		dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
761 }
762 
763 static void virtnet_vlan_rx_kill_vid(struct net_device *dev, u16 vid)
764 {
765 	struct virtnet_info *vi = netdev_priv(dev);
766 	struct scatterlist sg;
767 
768 	sg_init_one(&sg, &vid, sizeof(vid));
769 
770 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
771 				  VIRTIO_NET_CTRL_VLAN_DEL, &sg, 1, 0))
772 		dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
773 }
774 
775 static struct ethtool_ops virtnet_ethtool_ops = {
776 	.set_tx_csum = virtnet_set_tx_csum,
777 	.set_sg = ethtool_op_set_sg,
778 	.set_tso = ethtool_op_set_tso,
779 	.get_link = ethtool_op_get_link,
780 };
781 
782 #define MIN_MTU 68
783 #define MAX_MTU 65535
784 
785 static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
786 {
787 	if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
788 		return -EINVAL;
789 	dev->mtu = new_mtu;
790 	return 0;
791 }
792 
793 static const struct net_device_ops virtnet_netdev = {
794 	.ndo_open            = virtnet_open,
795 	.ndo_stop   	     = virtnet_close,
796 	.ndo_start_xmit      = start_xmit,
797 	.ndo_validate_addr   = eth_validate_addr,
798 	.ndo_set_mac_address = virtnet_set_mac_address,
799 	.ndo_set_rx_mode     = virtnet_set_rx_mode,
800 	.ndo_change_mtu	     = virtnet_change_mtu,
801 	.ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
802 	.ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
803 #ifdef CONFIG_NET_POLL_CONTROLLER
804 	.ndo_poll_controller = virtnet_netpoll,
805 #endif
806 };
807 
808 static void virtnet_update_status(struct virtnet_info *vi)
809 {
810 	u16 v;
811 
812 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS))
813 		return;
814 
815 	vi->vdev->config->get(vi->vdev,
816 			      offsetof(struct virtio_net_config, status),
817 			      &v, sizeof(v));
818 
819 	/* Ignore unknown (future) status bits */
820 	v &= VIRTIO_NET_S_LINK_UP;
821 
822 	if (vi->status == v)
823 		return;
824 
825 	vi->status = v;
826 
827 	if (vi->status & VIRTIO_NET_S_LINK_UP) {
828 		netif_carrier_on(vi->dev);
829 		netif_wake_queue(vi->dev);
830 	} else {
831 		netif_carrier_off(vi->dev);
832 		netif_stop_queue(vi->dev);
833 	}
834 }
835 
836 static void virtnet_config_changed(struct virtio_device *vdev)
837 {
838 	struct virtnet_info *vi = vdev->priv;
839 
840 	virtnet_update_status(vi);
841 }
842 
843 static int virtnet_probe(struct virtio_device *vdev)
844 {
845 	int err;
846 	struct net_device *dev;
847 	struct virtnet_info *vi;
848 
849 	/* Allocate ourselves a network device with room for our info */
850 	dev = alloc_etherdev(sizeof(struct virtnet_info));
851 	if (!dev)
852 		return -ENOMEM;
853 
854 	/* Set up network device as normal. */
855 	dev->netdev_ops = &virtnet_netdev;
856 	dev->features = NETIF_F_HIGHDMA;
857 	SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
858 	SET_NETDEV_DEV(dev, &vdev->dev);
859 
860 	/* Do we support "hardware" checksums? */
861 	if (csum && virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
862 		/* This opens up the world of extra features. */
863 		dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
864 		if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
865 			dev->features |= NETIF_F_TSO | NETIF_F_UFO
866 				| NETIF_F_TSO_ECN | NETIF_F_TSO6;
867 		}
868 		/* Individual feature bits: what can host handle? */
869 		if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
870 			dev->features |= NETIF_F_TSO;
871 		if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
872 			dev->features |= NETIF_F_TSO6;
873 		if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
874 			dev->features |= NETIF_F_TSO_ECN;
875 		if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
876 			dev->features |= NETIF_F_UFO;
877 	}
878 
879 	/* Configuration may specify what MAC to use.  Otherwise random. */
880 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
881 		vdev->config->get(vdev,
882 				  offsetof(struct virtio_net_config, mac),
883 				  dev->dev_addr, dev->addr_len);
884 	} else
885 		random_ether_addr(dev->dev_addr);
886 
887 	/* Set up our device-specific information */
888 	vi = netdev_priv(dev);
889 	netif_napi_add(dev, &vi->napi, virtnet_poll, napi_weight);
890 	vi->dev = dev;
891 	vi->vdev = vdev;
892 	vdev->priv = vi;
893 	vi->pages = NULL;
894 
895 	/* If they give us a callback when all buffers are done, we don't need
896 	 * the timer. */
897 	vi->free_in_tasklet = virtio_has_feature(vdev,VIRTIO_F_NOTIFY_ON_EMPTY);
898 
899 	/* If we can receive ANY GSO packets, we must allocate large ones. */
900 	if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4)
901 	    || virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6)
902 	    || virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
903 		vi->big_packets = true;
904 
905 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
906 		vi->mergeable_rx_bufs = true;
907 
908 	/* We expect two virtqueues, receive then send. */
909 	vi->rvq = vdev->config->find_vq(vdev, 0, skb_recv_done);
910 	if (IS_ERR(vi->rvq)) {
911 		err = PTR_ERR(vi->rvq);
912 		goto free;
913 	}
914 
915 	vi->svq = vdev->config->find_vq(vdev, 1, skb_xmit_done);
916 	if (IS_ERR(vi->svq)) {
917 		err = PTR_ERR(vi->svq);
918 		goto free_recv;
919 	}
920 
921 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)) {
922 		vi->cvq = vdev->config->find_vq(vdev, 2, NULL);
923 		if (IS_ERR(vi->cvq)) {
924 			err = PTR_ERR(vi->svq);
925 			goto free_send;
926 		}
927 
928 		if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
929 			dev->features |= NETIF_F_HW_VLAN_FILTER;
930 	}
931 
932 	/* Initialize our empty receive and send queues. */
933 	skb_queue_head_init(&vi->recv);
934 	skb_queue_head_init(&vi->send);
935 
936 	tasklet_init(&vi->tasklet, xmit_tasklet, (unsigned long)vi);
937 
938 	if (!vi->free_in_tasklet)
939 		setup_timer(&vi->xmit_free_timer, xmit_free, (unsigned long)vi);
940 
941 	err = register_netdev(dev);
942 	if (err) {
943 		pr_debug("virtio_net: registering device failed\n");
944 		goto free_ctrl;
945 	}
946 
947 	/* Last of all, set up some receive buffers. */
948 	try_fill_recv(vi);
949 
950 	/* If we didn't even get one input buffer, we're useless. */
951 	if (vi->num == 0) {
952 		err = -ENOMEM;
953 		goto unregister;
954 	}
955 
956 	vi->status = VIRTIO_NET_S_LINK_UP;
957 	virtnet_update_status(vi);
958 	netif_carrier_on(dev);
959 
960 	pr_debug("virtnet: registered device %s\n", dev->name);
961 	return 0;
962 
963 unregister:
964 	unregister_netdev(dev);
965 free_ctrl:
966 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ))
967 		vdev->config->del_vq(vi->cvq);
968 free_send:
969 	vdev->config->del_vq(vi->svq);
970 free_recv:
971 	vdev->config->del_vq(vi->rvq);
972 free:
973 	free_netdev(dev);
974 	return err;
975 }
976 
977 static void virtnet_remove(struct virtio_device *vdev)
978 {
979 	struct virtnet_info *vi = vdev->priv;
980 	struct sk_buff *skb;
981 
982 	/* Stop all the virtqueues. */
983 	vdev->config->reset(vdev);
984 
985 	if (!vi->free_in_tasklet)
986 		del_timer_sync(&vi->xmit_free_timer);
987 
988 	/* Free our skbs in send and recv queues, if any. */
989 	while ((skb = __skb_dequeue(&vi->recv)) != NULL) {
990 		kfree_skb(skb);
991 		vi->num--;
992 	}
993 	__skb_queue_purge(&vi->send);
994 
995 	BUG_ON(vi->num != 0);
996 
997 	vdev->config->del_vq(vi->svq);
998 	vdev->config->del_vq(vi->rvq);
999 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ))
1000 		vdev->config->del_vq(vi->cvq);
1001 	unregister_netdev(vi->dev);
1002 
1003 	while (vi->pages)
1004 		__free_pages(get_a_page(vi, GFP_KERNEL), 0);
1005 
1006 	free_netdev(vi->dev);
1007 }
1008 
1009 static struct virtio_device_id id_table[] = {
1010 	{ VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
1011 	{ 0 },
1012 };
1013 
1014 static unsigned int features[] = {
1015 	VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
1016 	VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
1017 	VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
1018 	VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
1019 	VIRTIO_NET_F_GUEST_ECN, /* We don't yet handle UFO input. */
1020 	VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ,
1021 	VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN,
1022 	VIRTIO_F_NOTIFY_ON_EMPTY,
1023 };
1024 
1025 static struct virtio_driver virtio_net = {
1026 	.feature_table = features,
1027 	.feature_table_size = ARRAY_SIZE(features),
1028 	.driver.name =	KBUILD_MODNAME,
1029 	.driver.owner =	THIS_MODULE,
1030 	.id_table =	id_table,
1031 	.probe =	virtnet_probe,
1032 	.remove =	__devexit_p(virtnet_remove),
1033 	.config_changed = virtnet_config_changed,
1034 };
1035 
1036 static int __init init(void)
1037 {
1038 	return register_virtio_driver(&virtio_net);
1039 }
1040 
1041 static void __exit fini(void)
1042 {
1043 	unregister_virtio_driver(&virtio_net);
1044 }
1045 module_init(init);
1046 module_exit(fini);
1047 
1048 MODULE_DEVICE_TABLE(virtio, id_table);
1049 MODULE_DESCRIPTION("Virtio network driver");
1050 MODULE_LICENSE("GPL");
1051