xref: /openbmc/linux/drivers/net/virtio_net.c (revision a09d2831)
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 	try_fill_recv(vi, GFP_KERNEL);
399 	still_empty = (vi->num == 0);
400 	napi_enable(&vi->napi);
401 
402 	/* In theory, this can happen: if we don't get any buffers in
403 	 * we will *never* try to fill again. */
404 	if (still_empty)
405 		schedule_delayed_work(&vi->refill, HZ/2);
406 }
407 
408 static int virtnet_poll(struct napi_struct *napi, int budget)
409 {
410 	struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi);
411 	struct sk_buff *skb = NULL;
412 	unsigned int len, received = 0;
413 
414 again:
415 	while (received < budget &&
416 	       (skb = vi->rvq->vq_ops->get_buf(vi->rvq, &len)) != NULL) {
417 		__skb_unlink(skb, &vi->recv);
418 		receive_skb(vi->dev, skb, len);
419 		vi->num--;
420 		received++;
421 	}
422 
423 	if (vi->num < vi->max / 2) {
424 		if (!try_fill_recv(vi, GFP_ATOMIC))
425 			schedule_delayed_work(&vi->refill, 0);
426 	}
427 
428 	/* Out of packets? */
429 	if (received < budget) {
430 		napi_complete(napi);
431 		if (unlikely(!vi->rvq->vq_ops->enable_cb(vi->rvq)) &&
432 		    napi_schedule_prep(napi)) {
433 			vi->rvq->vq_ops->disable_cb(vi->rvq);
434 			__napi_schedule(napi);
435 			goto again;
436 		}
437 	}
438 
439 	return received;
440 }
441 
442 static unsigned int free_old_xmit_skbs(struct virtnet_info *vi)
443 {
444 	struct sk_buff *skb;
445 	unsigned int len, tot_sgs = 0;
446 
447 	while ((skb = vi->svq->vq_ops->get_buf(vi->svq, &len)) != NULL) {
448 		pr_debug("Sent skb %p\n", skb);
449 		__skb_unlink(skb, &vi->send);
450 		vi->dev->stats.tx_bytes += skb->len;
451 		vi->dev->stats.tx_packets++;
452 		tot_sgs += skb_vnet_hdr(skb)->num_sg;
453 		dev_kfree_skb_any(skb);
454 	}
455 	return tot_sgs;
456 }
457 
458 static int xmit_skb(struct virtnet_info *vi, struct sk_buff *skb)
459 {
460 	struct scatterlist sg[2+MAX_SKB_FRAGS];
461 	struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
462 	const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
463 
464 	sg_init_table(sg, 2+MAX_SKB_FRAGS);
465 
466 	pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
467 
468 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
469 		hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
470 		hdr->hdr.csum_start = skb->csum_start - skb_headroom(skb);
471 		hdr->hdr.csum_offset = skb->csum_offset;
472 	} else {
473 		hdr->hdr.flags = 0;
474 		hdr->hdr.csum_offset = hdr->hdr.csum_start = 0;
475 	}
476 
477 	if (skb_is_gso(skb)) {
478 		hdr->hdr.hdr_len = skb_headlen(skb);
479 		hdr->hdr.gso_size = skb_shinfo(skb)->gso_size;
480 		if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
481 			hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
482 		else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
483 			hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
484 		else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
485 			hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
486 		else
487 			BUG();
488 		if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
489 			hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
490 	} else {
491 		hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
492 		hdr->hdr.gso_size = hdr->hdr.hdr_len = 0;
493 	}
494 
495 	hdr->mhdr.num_buffers = 0;
496 
497 	/* Encode metadata header at front. */
498 	if (vi->mergeable_rx_bufs)
499 		sg_set_buf(sg, &hdr->mhdr, sizeof(hdr->mhdr));
500 	else
501 		sg_set_buf(sg, &hdr->hdr, sizeof(hdr->hdr));
502 
503 	hdr->num_sg = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
504 	return vi->svq->vq_ops->add_buf(vi->svq, sg, hdr->num_sg, 0, skb);
505 }
506 
507 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
508 {
509 	struct virtnet_info *vi = netdev_priv(dev);
510 	int capacity;
511 
512 again:
513 	/* Free up any pending old buffers before queueing new ones. */
514 	free_old_xmit_skbs(vi);
515 
516 	/* Try to transmit */
517 	capacity = xmit_skb(vi, skb);
518 
519 	/* This can happen with OOM and indirect buffers. */
520 	if (unlikely(capacity < 0)) {
521 		netif_stop_queue(dev);
522 		dev_warn(&dev->dev, "Unexpected full queue\n");
523 		if (unlikely(!vi->svq->vq_ops->enable_cb(vi->svq))) {
524 			vi->svq->vq_ops->disable_cb(vi->svq);
525 			netif_start_queue(dev);
526 			goto again;
527 		}
528 		return NETDEV_TX_BUSY;
529 	}
530 	vi->svq->vq_ops->kick(vi->svq);
531 
532 	/*
533 	 * Put new one in send queue.  You'd expect we'd need this before
534 	 * xmit_skb calls add_buf(), since the callback can be triggered
535 	 * immediately after that.  But since the callback just triggers
536 	 * another call back here, normal network xmit locking prevents the
537 	 * race.
538 	 */
539 	__skb_queue_head(&vi->send, skb);
540 
541 	/* Don't wait up for transmitted skbs to be freed. */
542 	skb_orphan(skb);
543 	nf_reset(skb);
544 
545 	/* Apparently nice girls don't return TX_BUSY; stop the queue
546 	 * before it gets out of hand.  Naturally, this wastes entries. */
547 	if (capacity < 2+MAX_SKB_FRAGS) {
548 		netif_stop_queue(dev);
549 		if (unlikely(!vi->svq->vq_ops->enable_cb(vi->svq))) {
550 			/* More just got used, free them then recheck. */
551 			capacity += free_old_xmit_skbs(vi);
552 			if (capacity >= 2+MAX_SKB_FRAGS) {
553 				netif_start_queue(dev);
554 				vi->svq->vq_ops->disable_cb(vi->svq);
555 			}
556 		}
557 	}
558 
559 	return NETDEV_TX_OK;
560 }
561 
562 static int virtnet_set_mac_address(struct net_device *dev, void *p)
563 {
564 	struct virtnet_info *vi = netdev_priv(dev);
565 	struct virtio_device *vdev = vi->vdev;
566 	int ret;
567 
568 	ret = eth_mac_addr(dev, p);
569 	if (ret)
570 		return ret;
571 
572 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
573 		vdev->config->set(vdev, offsetof(struct virtio_net_config, mac),
574 		                  dev->dev_addr, dev->addr_len);
575 
576 	return 0;
577 }
578 
579 #ifdef CONFIG_NET_POLL_CONTROLLER
580 static void virtnet_netpoll(struct net_device *dev)
581 {
582 	struct virtnet_info *vi = netdev_priv(dev);
583 
584 	napi_schedule(&vi->napi);
585 }
586 #endif
587 
588 static int virtnet_open(struct net_device *dev)
589 {
590 	struct virtnet_info *vi = netdev_priv(dev);
591 
592 	napi_enable(&vi->napi);
593 
594 	/* If all buffers were filled by other side before we napi_enabled, we
595 	 * won't get another interrupt, so process any outstanding packets
596 	 * now.  virtnet_poll wants re-enable the queue, so we disable here.
597 	 * We synchronize against interrupts via NAPI_STATE_SCHED */
598 	if (napi_schedule_prep(&vi->napi)) {
599 		vi->rvq->vq_ops->disable_cb(vi->rvq);
600 		__napi_schedule(&vi->napi);
601 	}
602 	return 0;
603 }
604 
605 /*
606  * Send command via the control virtqueue and check status.  Commands
607  * supported by the hypervisor, as indicated by feature bits, should
608  * never fail unless improperly formated.
609  */
610 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
611 				 struct scatterlist *data, int out, int in)
612 {
613 	struct scatterlist *s, sg[VIRTNET_SEND_COMMAND_SG_MAX + 2];
614 	struct virtio_net_ctrl_hdr ctrl;
615 	virtio_net_ctrl_ack status = ~0;
616 	unsigned int tmp;
617 	int i;
618 
619 	/* Caller should know better */
620 	BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ||
621 		(out + in > VIRTNET_SEND_COMMAND_SG_MAX));
622 
623 	out++; /* Add header */
624 	in++; /* Add return status */
625 
626 	ctrl.class = class;
627 	ctrl.cmd = cmd;
628 
629 	sg_init_table(sg, out + in);
630 
631 	sg_set_buf(&sg[0], &ctrl, sizeof(ctrl));
632 	for_each_sg(data, s, out + in - 2, i)
633 		sg_set_buf(&sg[i + 1], sg_virt(s), s->length);
634 	sg_set_buf(&sg[out + in - 1], &status, sizeof(status));
635 
636 	BUG_ON(vi->cvq->vq_ops->add_buf(vi->cvq, sg, out, in, vi) < 0);
637 
638 	vi->cvq->vq_ops->kick(vi->cvq);
639 
640 	/*
641 	 * Spin for a response, the kick causes an ioport write, trapping
642 	 * into the hypervisor, so the request should be handled immediately.
643 	 */
644 	while (!vi->cvq->vq_ops->get_buf(vi->cvq, &tmp))
645 		cpu_relax();
646 
647 	return status == VIRTIO_NET_OK;
648 }
649 
650 static int virtnet_close(struct net_device *dev)
651 {
652 	struct virtnet_info *vi = netdev_priv(dev);
653 
654 	napi_disable(&vi->napi);
655 
656 	return 0;
657 }
658 
659 static int virtnet_set_tx_csum(struct net_device *dev, u32 data)
660 {
661 	struct virtnet_info *vi = netdev_priv(dev);
662 	struct virtio_device *vdev = vi->vdev;
663 
664 	if (data && !virtio_has_feature(vdev, VIRTIO_NET_F_CSUM))
665 		return -ENOSYS;
666 
667 	return ethtool_op_set_tx_hw_csum(dev, data);
668 }
669 
670 static void virtnet_set_rx_mode(struct net_device *dev)
671 {
672 	struct virtnet_info *vi = netdev_priv(dev);
673 	struct scatterlist sg[2];
674 	u8 promisc, allmulti;
675 	struct virtio_net_ctrl_mac *mac_data;
676 	struct dev_addr_list *addr;
677 	struct netdev_hw_addr *ha;
678 	void *buf;
679 	int i;
680 
681 	/* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */
682 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
683 		return;
684 
685 	promisc = ((dev->flags & IFF_PROMISC) != 0);
686 	allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
687 
688 	sg_init_one(sg, &promisc, sizeof(promisc));
689 
690 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
691 				  VIRTIO_NET_CTRL_RX_PROMISC,
692 				  sg, 1, 0))
693 		dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
694 			 promisc ? "en" : "dis");
695 
696 	sg_init_one(sg, &allmulti, sizeof(allmulti));
697 
698 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
699 				  VIRTIO_NET_CTRL_RX_ALLMULTI,
700 				  sg, 1, 0))
701 		dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
702 			 allmulti ? "en" : "dis");
703 
704 	/* MAC filter - use one buffer for both lists */
705 	mac_data = buf = kzalloc(((dev->uc.count + dev->mc_count) * ETH_ALEN) +
706 				 (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
707 	if (!buf) {
708 		dev_warn(&dev->dev, "No memory for MAC address buffer\n");
709 		return;
710 	}
711 
712 	sg_init_table(sg, 2);
713 
714 	/* Store the unicast list and count in the front of the buffer */
715 	mac_data->entries = dev->uc.count;
716 	i = 0;
717 	list_for_each_entry(ha, &dev->uc.list, list)
718 		memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
719 
720 	sg_set_buf(&sg[0], mac_data,
721 		   sizeof(mac_data->entries) + (dev->uc.count * ETH_ALEN));
722 
723 	/* multicast list and count fill the end */
724 	mac_data = (void *)&mac_data->macs[dev->uc.count][0];
725 
726 	mac_data->entries = dev->mc_count;
727 	addr = dev->mc_list;
728 	for (i = 0; i < dev->mc_count; i++, addr = addr->next)
729 		memcpy(&mac_data->macs[i][0], addr->da_addr, ETH_ALEN);
730 
731 	sg_set_buf(&sg[1], mac_data,
732 		   sizeof(mac_data->entries) + (dev->mc_count * ETH_ALEN));
733 
734 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
735 				  VIRTIO_NET_CTRL_MAC_TABLE_SET,
736 				  sg, 2, 0))
737 		dev_warn(&dev->dev, "Failed to set MAC fitler table.\n");
738 
739 	kfree(buf);
740 }
741 
742 static void virtnet_vlan_rx_add_vid(struct net_device *dev, u16 vid)
743 {
744 	struct virtnet_info *vi = netdev_priv(dev);
745 	struct scatterlist sg;
746 
747 	sg_init_one(&sg, &vid, sizeof(vid));
748 
749 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
750 				  VIRTIO_NET_CTRL_VLAN_ADD, &sg, 1, 0))
751 		dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
752 }
753 
754 static void virtnet_vlan_rx_kill_vid(struct net_device *dev, u16 vid)
755 {
756 	struct virtnet_info *vi = netdev_priv(dev);
757 	struct scatterlist sg;
758 
759 	sg_init_one(&sg, &vid, sizeof(vid));
760 
761 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
762 				  VIRTIO_NET_CTRL_VLAN_DEL, &sg, 1, 0))
763 		dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
764 }
765 
766 static const struct ethtool_ops virtnet_ethtool_ops = {
767 	.set_tx_csum = virtnet_set_tx_csum,
768 	.set_sg = ethtool_op_set_sg,
769 	.set_tso = ethtool_op_set_tso,
770 	.set_ufo = ethtool_op_set_ufo,
771 	.get_link = ethtool_op_get_link,
772 };
773 
774 #define MIN_MTU 68
775 #define MAX_MTU 65535
776 
777 static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
778 {
779 	if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
780 		return -EINVAL;
781 	dev->mtu = new_mtu;
782 	return 0;
783 }
784 
785 static const struct net_device_ops virtnet_netdev = {
786 	.ndo_open            = virtnet_open,
787 	.ndo_stop   	     = virtnet_close,
788 	.ndo_start_xmit      = start_xmit,
789 	.ndo_validate_addr   = eth_validate_addr,
790 	.ndo_set_mac_address = virtnet_set_mac_address,
791 	.ndo_set_rx_mode     = virtnet_set_rx_mode,
792 	.ndo_change_mtu	     = virtnet_change_mtu,
793 	.ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
794 	.ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
795 #ifdef CONFIG_NET_POLL_CONTROLLER
796 	.ndo_poll_controller = virtnet_netpoll,
797 #endif
798 };
799 
800 static void virtnet_update_status(struct virtnet_info *vi)
801 {
802 	u16 v;
803 
804 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS))
805 		return;
806 
807 	vi->vdev->config->get(vi->vdev,
808 			      offsetof(struct virtio_net_config, status),
809 			      &v, sizeof(v));
810 
811 	/* Ignore unknown (future) status bits */
812 	v &= VIRTIO_NET_S_LINK_UP;
813 
814 	if (vi->status == v)
815 		return;
816 
817 	vi->status = v;
818 
819 	if (vi->status & VIRTIO_NET_S_LINK_UP) {
820 		netif_carrier_on(vi->dev);
821 		netif_wake_queue(vi->dev);
822 	} else {
823 		netif_carrier_off(vi->dev);
824 		netif_stop_queue(vi->dev);
825 	}
826 }
827 
828 static void virtnet_config_changed(struct virtio_device *vdev)
829 {
830 	struct virtnet_info *vi = vdev->priv;
831 
832 	virtnet_update_status(vi);
833 }
834 
835 static int virtnet_probe(struct virtio_device *vdev)
836 {
837 	int err;
838 	struct net_device *dev;
839 	struct virtnet_info *vi;
840 	struct virtqueue *vqs[3];
841 	vq_callback_t *callbacks[] = { skb_recv_done, skb_xmit_done, NULL};
842 	const char *names[] = { "input", "output", "control" };
843 	int nvqs;
844 
845 	/* Allocate ourselves a network device with room for our info */
846 	dev = alloc_etherdev(sizeof(struct virtnet_info));
847 	if (!dev)
848 		return -ENOMEM;
849 
850 	/* Set up network device as normal. */
851 	dev->netdev_ops = &virtnet_netdev;
852 	dev->features = NETIF_F_HIGHDMA;
853 	SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
854 	SET_NETDEV_DEV(dev, &vdev->dev);
855 
856 	/* Do we support "hardware" checksums? */
857 	if (csum && virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
858 		/* This opens up the world of extra features. */
859 		dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
860 		if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
861 			dev->features |= NETIF_F_TSO | NETIF_F_UFO
862 				| NETIF_F_TSO_ECN | NETIF_F_TSO6;
863 		}
864 		/* Individual feature bits: what can host handle? */
865 		if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
866 			dev->features |= NETIF_F_TSO;
867 		if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
868 			dev->features |= NETIF_F_TSO6;
869 		if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
870 			dev->features |= NETIF_F_TSO_ECN;
871 		if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
872 			dev->features |= NETIF_F_UFO;
873 	}
874 
875 	/* Configuration may specify what MAC to use.  Otherwise random. */
876 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
877 		vdev->config->get(vdev,
878 				  offsetof(struct virtio_net_config, mac),
879 				  dev->dev_addr, dev->addr_len);
880 	} else
881 		random_ether_addr(dev->dev_addr);
882 
883 	/* Set up our device-specific information */
884 	vi = netdev_priv(dev);
885 	netif_napi_add(dev, &vi->napi, virtnet_poll, napi_weight);
886 	vi->dev = dev;
887 	vi->vdev = vdev;
888 	vdev->priv = vi;
889 	vi->pages = NULL;
890 	INIT_DELAYED_WORK(&vi->refill, refill_work);
891 
892 	/* If we can receive ANY GSO packets, we must allocate large ones. */
893 	if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
894 	    virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
895 	    virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
896 		vi->big_packets = true;
897 
898 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
899 		vi->mergeable_rx_bufs = true;
900 
901 	/* We expect two virtqueues, receive then send,
902 	 * and optionally control. */
903 	nvqs = virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ? 3 : 2;
904 
905 	err = vdev->config->find_vqs(vdev, nvqs, vqs, callbacks, names);
906 	if (err)
907 		goto free;
908 
909 	vi->rvq = vqs[0];
910 	vi->svq = vqs[1];
911 
912 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)) {
913 		vi->cvq = vqs[2];
914 
915 		if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
916 			dev->features |= NETIF_F_HW_VLAN_FILTER;
917 	}
918 
919 	/* Initialize our empty receive and send queues. */
920 	skb_queue_head_init(&vi->recv);
921 	skb_queue_head_init(&vi->send);
922 
923 	err = register_netdev(dev);
924 	if (err) {
925 		pr_debug("virtio_net: registering device failed\n");
926 		goto free_vqs;
927 	}
928 
929 	/* Last of all, set up some receive buffers. */
930 	try_fill_recv(vi, GFP_KERNEL);
931 
932 	/* If we didn't even get one input buffer, we're useless. */
933 	if (vi->num == 0) {
934 		err = -ENOMEM;
935 		goto unregister;
936 	}
937 
938 	vi->status = VIRTIO_NET_S_LINK_UP;
939 	virtnet_update_status(vi);
940 	netif_carrier_on(dev);
941 
942 	pr_debug("virtnet: registered device %s\n", dev->name);
943 	return 0;
944 
945 unregister:
946 	unregister_netdev(dev);
947 	cancel_delayed_work_sync(&vi->refill);
948 free_vqs:
949 	vdev->config->del_vqs(vdev);
950 free:
951 	free_netdev(dev);
952 	return err;
953 }
954 
955 static void __devexit virtnet_remove(struct virtio_device *vdev)
956 {
957 	struct virtnet_info *vi = vdev->priv;
958 	struct sk_buff *skb;
959 
960 	/* Stop all the virtqueues. */
961 	vdev->config->reset(vdev);
962 
963 	/* Free our skbs in send and recv queues, if any. */
964 	while ((skb = __skb_dequeue(&vi->recv)) != NULL) {
965 		kfree_skb(skb);
966 		vi->num--;
967 	}
968 	__skb_queue_purge(&vi->send);
969 
970 	BUG_ON(vi->num != 0);
971 
972 	unregister_netdev(vi->dev);
973 	cancel_delayed_work_sync(&vi->refill);
974 
975 	vdev->config->del_vqs(vi->vdev);
976 
977 	while (vi->pages)
978 		__free_pages(get_a_page(vi, GFP_KERNEL), 0);
979 
980 	free_netdev(vi->dev);
981 }
982 
983 static struct virtio_device_id id_table[] = {
984 	{ VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
985 	{ 0 },
986 };
987 
988 static unsigned int features[] = {
989 	VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
990 	VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
991 	VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
992 	VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
993 	VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO,
994 	VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ,
995 	VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN,
996 };
997 
998 static struct virtio_driver virtio_net_driver = {
999 	.feature_table = features,
1000 	.feature_table_size = ARRAY_SIZE(features),
1001 	.driver.name =	KBUILD_MODNAME,
1002 	.driver.owner =	THIS_MODULE,
1003 	.id_table =	id_table,
1004 	.probe =	virtnet_probe,
1005 	.remove =	__devexit_p(virtnet_remove),
1006 	.config_changed = virtnet_config_changed,
1007 };
1008 
1009 static int __init init(void)
1010 {
1011 	return register_virtio_driver(&virtio_net_driver);
1012 }
1013 
1014 static void __exit fini(void)
1015 {
1016 	unregister_virtio_driver(&virtio_net_driver);
1017 }
1018 module_init(init);
1019 module_exit(fini);
1020 
1021 MODULE_DEVICE_TABLE(virtio, id_table);
1022 MODULE_DESCRIPTION("Virtio network driver");
1023 MODULE_LICENSE("GPL");
1024