virtio-net: pass gfp to add_buf
[linux-2.6.git] / drivers / net / virtio_net.c
1 /* A network driver using virtio.
2  *
3  * Copyright 2007 Rusty Russell <rusty@rustcorp.com.au> IBM Corporation
4  *
5  * This program is free software; you can redistribute it and/or modify
6  * it under the terms of the GNU General Public License as published by
7  * the Free Software Foundation; either version 2 of the License, or
8  * (at your option) any later version.
9  *
10  * This program is distributed in the hope that it will be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License for more details.
14  *
15  * You should have received a copy of the GNU General Public License
16  * along with this program; if not, write to the Free Software
17  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
18  */
19 //#define DEBUG
20 #include <linux/netdevice.h>
21 #include <linux/etherdevice.h>
22 #include <linux/ethtool.h>
23 #include <linux/module.h>
24 #include <linux/virtio.h>
25 #include <linux/virtio_net.h>
26 #include <linux/scatterlist.h>
27 #include <linux/if_vlan.h>
28 #include <linux/slab.h>
29
30 static int napi_weight = 128;
31 module_param(napi_weight, int, 0444);
32
33 static int csum = 1, gso = 1;
34 module_param(csum, bool, 0444);
35 module_param(gso, bool, 0444);
36
37 /* FIXME: MTU in config. */
38 #define MAX_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
39 #define GOOD_COPY_LEN   128
40
41 #define VIRTNET_SEND_COMMAND_SG_MAX    2
42
43 struct virtnet_info {
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         /* Work struct for refilling if we run low on memory. */
60         struct delayed_work refill;
61
62         /* Chain pages by the private ptr. */
63         struct page *pages;
64
65         /* fragments + linear part + virtio header */
66         struct scatterlist rx_sg[MAX_SKB_FRAGS + 2];
67         struct scatterlist tx_sg[MAX_SKB_FRAGS + 2];
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 struct padded_vnet_hdr {
79         struct virtio_net_hdr hdr;
80         /*
81          * virtio_net_hdr should be in a separated sg buffer because of a
82          * QEMU bug, and data sg buffer shares same page with this header sg.
83          * This padding makes next sg 16 byte aligned after virtio_net_hdr.
84          */
85         char padding[6];
86 };
87
88 static inline struct skb_vnet_hdr *skb_vnet_hdr(struct sk_buff *skb)
89 {
90         return (struct skb_vnet_hdr *)skb->cb;
91 }
92
93 /*
94  * private is used to chain pages for big packets, put the whole
95  * most recent used list in the beginning for reuse
96  */
97 static void give_pages(struct virtnet_info *vi, struct page *page)
98 {
99         struct page *end;
100
101         /* Find end of list, sew whole thing into vi->pages. */
102         for (end = page; end->private; end = (struct page *)end->private);
103         end->private = (unsigned long)vi->pages;
104         vi->pages = page;
105 }
106
107 static struct page *get_a_page(struct virtnet_info *vi, gfp_t gfp_mask)
108 {
109         struct page *p = vi->pages;
110
111         if (p) {
112                 vi->pages = (struct page *)p->private;
113                 /* clear private here, it is used to chain pages */
114                 p->private = 0;
115         } else
116                 p = alloc_page(gfp_mask);
117         return p;
118 }
119
120 static void skb_xmit_done(struct virtqueue *svq)
121 {
122         struct virtnet_info *vi = svq->vdev->priv;
123
124         /* Suppress further interrupts. */
125         virtqueue_disable_cb(svq);
126
127         /* We were probably waiting for more output buffers. */
128         netif_wake_queue(vi->dev);
129 }
130
131 static void set_skb_frag(struct sk_buff *skb, struct page *page,
132                          unsigned int offset, unsigned int *len)
133 {
134         int i = skb_shinfo(skb)->nr_frags;
135         skb_frag_t *f;
136
137         f = &skb_shinfo(skb)->frags[i];
138         f->size = min((unsigned)PAGE_SIZE - offset, *len);
139         f->page_offset = offset;
140         f->page = page;
141
142         skb->data_len += f->size;
143         skb->len += f->size;
144         skb_shinfo(skb)->nr_frags++;
145         *len -= f->size;
146 }
147
148 static struct sk_buff *page_to_skb(struct virtnet_info *vi,
149                                    struct page *page, unsigned int len)
150 {
151         struct sk_buff *skb;
152         struct skb_vnet_hdr *hdr;
153         unsigned int copy, hdr_len, offset;
154         char *p;
155
156         p = page_address(page);
157
158         /* copy small packet so we can reuse these pages for small data */
159         skb = netdev_alloc_skb_ip_align(vi->dev, GOOD_COPY_LEN);
160         if (unlikely(!skb))
161                 return NULL;
162
163         hdr = skb_vnet_hdr(skb);
164
165         if (vi->mergeable_rx_bufs) {
166                 hdr_len = sizeof hdr->mhdr;
167                 offset = hdr_len;
168         } else {
169                 hdr_len = sizeof hdr->hdr;
170                 offset = sizeof(struct padded_vnet_hdr);
171         }
172
173         memcpy(hdr, p, hdr_len);
174
175         len -= hdr_len;
176         p += offset;
177
178         copy = len;
179         if (copy > skb_tailroom(skb))
180                 copy = skb_tailroom(skb);
181         memcpy(skb_put(skb, copy), p, copy);
182
183         len -= copy;
184         offset += copy;
185
186         while (len) {
187                 set_skb_frag(skb, page, offset, &len);
188                 page = (struct page *)page->private;
189                 offset = 0;
190         }
191
192         if (page)
193                 give_pages(vi, page);
194
195         return skb;
196 }
197
198 static int receive_mergeable(struct virtnet_info *vi, struct sk_buff *skb)
199 {
200         struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
201         struct page *page;
202         int num_buf, i, len;
203
204         num_buf = hdr->mhdr.num_buffers;
205         while (--num_buf) {
206                 i = skb_shinfo(skb)->nr_frags;
207                 if (i >= MAX_SKB_FRAGS) {
208                         pr_debug("%s: packet too long\n", skb->dev->name);
209                         skb->dev->stats.rx_length_errors++;
210                         return -EINVAL;
211                 }
212
213                 page = virtqueue_get_buf(vi->rvq, &len);
214                 if (!page) {
215                         pr_debug("%s: rx error: %d buffers missing\n",
216                                  skb->dev->name, hdr->mhdr.num_buffers);
217                         skb->dev->stats.rx_length_errors++;
218                         return -EINVAL;
219                 }
220                 if (len > PAGE_SIZE)
221                         len = PAGE_SIZE;
222
223                 set_skb_frag(skb, page, 0, &len);
224
225                 --vi->num;
226         }
227         return 0;
228 }
229
230 static void receive_buf(struct net_device *dev, void *buf, unsigned int len)
231 {
232         struct virtnet_info *vi = netdev_priv(dev);
233         struct sk_buff *skb;
234         struct page *page;
235         struct skb_vnet_hdr *hdr;
236
237         if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
238                 pr_debug("%s: short packet %i\n", dev->name, len);
239                 dev->stats.rx_length_errors++;
240                 if (vi->mergeable_rx_bufs || vi->big_packets)
241                         give_pages(vi, buf);
242                 else
243                         dev_kfree_skb(buf);
244                 return;
245         }
246
247         if (!vi->mergeable_rx_bufs && !vi->big_packets) {
248                 skb = buf;
249                 len -= sizeof(struct virtio_net_hdr);
250                 skb_trim(skb, len);
251         } else {
252                 page = buf;
253                 skb = page_to_skb(vi, page, len);
254                 if (unlikely(!skb)) {
255                         dev->stats.rx_dropped++;
256                         give_pages(vi, page);
257                         return;
258                 }
259                 if (vi->mergeable_rx_bufs)
260                         if (receive_mergeable(vi, skb)) {
261                                 dev_kfree_skb(skb);
262                                 return;
263                         }
264         }
265
266         hdr = skb_vnet_hdr(skb);
267         skb->truesize += skb->data_len;
268         dev->stats.rx_bytes += skb->len;
269         dev->stats.rx_packets++;
270
271         if (hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
272                 pr_debug("Needs csum!\n");
273                 if (!skb_partial_csum_set(skb,
274                                           hdr->hdr.csum_start,
275                                           hdr->hdr.csum_offset))
276                         goto frame_err;
277         }
278
279         skb->protocol = eth_type_trans(skb, dev);
280         pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
281                  ntohs(skb->protocol), skb->len, skb->pkt_type);
282
283         if (hdr->hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
284                 pr_debug("GSO!\n");
285                 switch (hdr->hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
286                 case VIRTIO_NET_HDR_GSO_TCPV4:
287                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
288                         break;
289                 case VIRTIO_NET_HDR_GSO_UDP:
290                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
291                         break;
292                 case VIRTIO_NET_HDR_GSO_TCPV6:
293                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
294                         break;
295                 default:
296                         if (net_ratelimit())
297                                 printk(KERN_WARNING "%s: bad gso type %u.\n",
298                                        dev->name, hdr->hdr.gso_type);
299                         goto frame_err;
300                 }
301
302                 if (hdr->hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
303                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
304
305                 skb_shinfo(skb)->gso_size = hdr->hdr.gso_size;
306                 if (skb_shinfo(skb)->gso_size == 0) {
307                         if (net_ratelimit())
308                                 printk(KERN_WARNING "%s: zero gso size.\n",
309                                        dev->name);
310                         goto frame_err;
311                 }
312
313                 /* Header must be checked, and gso_segs computed. */
314                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
315                 skb_shinfo(skb)->gso_segs = 0;
316         }
317
318         netif_receive_skb(skb);
319         return;
320
321 frame_err:
322         dev->stats.rx_frame_errors++;
323         dev_kfree_skb(skb);
324 }
325
326 static int add_recvbuf_small(struct virtnet_info *vi, gfp_t gfp)
327 {
328         struct sk_buff *skb;
329         struct skb_vnet_hdr *hdr;
330         int err;
331
332         skb = netdev_alloc_skb_ip_align(vi->dev, MAX_PACKET_LEN);
333         if (unlikely(!skb))
334                 return -ENOMEM;
335
336         skb_put(skb, MAX_PACKET_LEN);
337
338         hdr = skb_vnet_hdr(skb);
339         sg_set_buf(vi->rx_sg, &hdr->hdr, sizeof hdr->hdr);
340
341         skb_to_sgvec(skb, vi->rx_sg + 1, 0, skb->len);
342
343         err = virtqueue_add_buf_gfp(vi->rvq, vi->rx_sg, 0, 2, skb, gfp);
344         if (err < 0)
345                 dev_kfree_skb(skb);
346
347         return err;
348 }
349
350 static int add_recvbuf_big(struct virtnet_info *vi, gfp_t gfp)
351 {
352         struct page *first, *list = NULL;
353         char *p;
354         int i, err, offset;
355
356         /* page in vi->rx_sg[MAX_SKB_FRAGS + 1] is list tail */
357         for (i = MAX_SKB_FRAGS + 1; i > 1; --i) {
358                 first = get_a_page(vi, gfp);
359                 if (!first) {
360                         if (list)
361                                 give_pages(vi, list);
362                         return -ENOMEM;
363                 }
364                 sg_set_buf(&vi->rx_sg[i], page_address(first), PAGE_SIZE);
365
366                 /* chain new page in list head to match sg */
367                 first->private = (unsigned long)list;
368                 list = first;
369         }
370
371         first = get_a_page(vi, gfp);
372         if (!first) {
373                 give_pages(vi, list);
374                 return -ENOMEM;
375         }
376         p = page_address(first);
377
378         /* vi->rx_sg[0], vi->rx_sg[1] share the same page */
379         /* a separated vi->rx_sg[0] for virtio_net_hdr only due to QEMU bug */
380         sg_set_buf(&vi->rx_sg[0], p, sizeof(struct virtio_net_hdr));
381
382         /* vi->rx_sg[1] for data packet, from offset */
383         offset = sizeof(struct padded_vnet_hdr);
384         sg_set_buf(&vi->rx_sg[1], p + offset, PAGE_SIZE - offset);
385
386         /* chain first in list head */
387         first->private = (unsigned long)list;
388         err = virtqueue_add_buf_gfp(vi->rvq, vi->rx_sg, 0, MAX_SKB_FRAGS + 2,
389                                     first, gfp);
390         if (err < 0)
391                 give_pages(vi, first);
392
393         return err;
394 }
395
396 static int add_recvbuf_mergeable(struct virtnet_info *vi, gfp_t gfp)
397 {
398         struct page *page;
399         int err;
400
401         page = get_a_page(vi, gfp);
402         if (!page)
403                 return -ENOMEM;
404
405         sg_init_one(vi->rx_sg, page_address(page), PAGE_SIZE);
406
407         err = virtqueue_add_buf_gfp(vi->rvq, vi->rx_sg, 0, 1, page, gfp);
408         if (err < 0)
409                 give_pages(vi, page);
410
411         return err;
412 }
413
414 /* Returns false if we couldn't fill entirely (OOM). */
415 static bool try_fill_recv(struct virtnet_info *vi, gfp_t gfp)
416 {
417         int err;
418         bool oom = false;
419
420         do {
421                 if (vi->mergeable_rx_bufs)
422                         err = add_recvbuf_mergeable(vi, gfp);
423                 else if (vi->big_packets)
424                         err = add_recvbuf_big(vi, gfp);
425                 else
426                         err = add_recvbuf_small(vi, gfp);
427
428                 if (err < 0) {
429                         oom = true;
430                         break;
431                 }
432                 ++vi->num;
433         } while (err > 0);
434         if (unlikely(vi->num > vi->max))
435                 vi->max = vi->num;
436         virtqueue_kick(vi->rvq);
437         return !oom;
438 }
439
440 static void skb_recv_done(struct virtqueue *rvq)
441 {
442         struct virtnet_info *vi = rvq->vdev->priv;
443         /* Schedule NAPI, Suppress further interrupts if successful. */
444         if (napi_schedule_prep(&vi->napi)) {
445                 virtqueue_disable_cb(rvq);
446                 __napi_schedule(&vi->napi);
447         }
448 }
449
450 static void refill_work(struct work_struct *work)
451 {
452         struct virtnet_info *vi;
453         bool still_empty;
454
455         vi = container_of(work, struct virtnet_info, refill.work);
456         napi_disable(&vi->napi);
457         still_empty = !try_fill_recv(vi, GFP_KERNEL);
458         napi_enable(&vi->napi);
459
460         /* In theory, this can happen: if we don't get any buffers in
461          * we will *never* try to fill again. */
462         if (still_empty)
463                 schedule_delayed_work(&vi->refill, HZ/2);
464 }
465
466 static int virtnet_poll(struct napi_struct *napi, int budget)
467 {
468         struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi);
469         void *buf;
470         unsigned int len, received = 0;
471
472 again:
473         while (received < budget &&
474                (buf = virtqueue_get_buf(vi->rvq, &len)) != NULL) {
475                 receive_buf(vi->dev, buf, len);
476                 --vi->num;
477                 received++;
478         }
479
480         if (vi->num < vi->max / 2) {
481                 if (!try_fill_recv(vi, GFP_ATOMIC))
482                         schedule_delayed_work(&vi->refill, 0);
483         }
484
485         /* Out of packets? */
486         if (received < budget) {
487                 napi_complete(napi);
488                 if (unlikely(!virtqueue_enable_cb(vi->rvq)) &&
489                     napi_schedule_prep(napi)) {
490                         virtqueue_disable_cb(vi->rvq);
491                         __napi_schedule(napi);
492                         goto again;
493                 }
494         }
495
496         return received;
497 }
498
499 static unsigned int free_old_xmit_skbs(struct virtnet_info *vi)
500 {
501         struct sk_buff *skb;
502         unsigned int len, tot_sgs = 0;
503
504         while ((skb = virtqueue_get_buf(vi->svq, &len)) != NULL) {
505                 pr_debug("Sent skb %p\n", skb);
506                 vi->dev->stats.tx_bytes += skb->len;
507                 vi->dev->stats.tx_packets++;
508                 tot_sgs += skb_vnet_hdr(skb)->num_sg;
509                 dev_kfree_skb_any(skb);
510         }
511         return tot_sgs;
512 }
513
514 static int xmit_skb(struct virtnet_info *vi, struct sk_buff *skb)
515 {
516         struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
517         const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
518
519         pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
520
521         if (skb->ip_summed == CHECKSUM_PARTIAL) {
522                 hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
523                 hdr->hdr.csum_start = skb->csum_start - skb_headroom(skb);
524                 hdr->hdr.csum_offset = skb->csum_offset;
525         } else {
526                 hdr->hdr.flags = 0;
527                 hdr->hdr.csum_offset = hdr->hdr.csum_start = 0;
528         }
529
530         if (skb_is_gso(skb)) {
531                 hdr->hdr.hdr_len = skb_headlen(skb);
532                 hdr->hdr.gso_size = skb_shinfo(skb)->gso_size;
533                 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
534                         hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
535                 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
536                         hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
537                 else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
538                         hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
539                 else
540                         BUG();
541                 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
542                         hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
543         } else {
544                 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
545                 hdr->hdr.gso_size = hdr->hdr.hdr_len = 0;
546         }
547
548         hdr->mhdr.num_buffers = 0;
549
550         /* Encode metadata header at front. */
551         if (vi->mergeable_rx_bufs)
552                 sg_set_buf(vi->tx_sg, &hdr->mhdr, sizeof hdr->mhdr);
553         else
554                 sg_set_buf(vi->tx_sg, &hdr->hdr, sizeof hdr->hdr);
555
556         hdr->num_sg = skb_to_sgvec(skb, vi->tx_sg + 1, 0, skb->len) + 1;
557         return virtqueue_add_buf(vi->svq, vi->tx_sg, hdr->num_sg,
558                                         0, skb);
559 }
560
561 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
562 {
563         struct virtnet_info *vi = netdev_priv(dev);
564         int capacity;
565
566 again:
567         /* Free up any pending old buffers before queueing new ones. */
568         free_old_xmit_skbs(vi);
569
570         /* Try to transmit */
571         capacity = xmit_skb(vi, skb);
572
573         /* This can happen with OOM and indirect buffers. */
574         if (unlikely(capacity < 0)) {
575                 netif_stop_queue(dev);
576                 dev_warn(&dev->dev, "Unexpected full queue\n");
577                 if (unlikely(!virtqueue_enable_cb(vi->svq))) {
578                         virtqueue_disable_cb(vi->svq);
579                         netif_start_queue(dev);
580                         goto again;
581                 }
582                 return NETDEV_TX_BUSY;
583         }
584         virtqueue_kick(vi->svq);
585
586         /* Don't wait up for transmitted skbs to be freed. */
587         skb_orphan(skb);
588         nf_reset(skb);
589
590         /* Apparently nice girls don't return TX_BUSY; stop the queue
591          * before it gets out of hand.  Naturally, this wastes entries. */
592         if (capacity < 2+MAX_SKB_FRAGS) {
593                 netif_stop_queue(dev);
594                 if (unlikely(!virtqueue_enable_cb(vi->svq))) {
595                         /* More just got used, free them then recheck. */
596                         capacity += free_old_xmit_skbs(vi);
597                         if (capacity >= 2+MAX_SKB_FRAGS) {
598                                 netif_start_queue(dev);
599                                 virtqueue_disable_cb(vi->svq);
600                         }
601                 }
602         }
603
604         return NETDEV_TX_OK;
605 }
606
607 static int virtnet_set_mac_address(struct net_device *dev, void *p)
608 {
609         struct virtnet_info *vi = netdev_priv(dev);
610         struct virtio_device *vdev = vi->vdev;
611         int ret;
612
613         ret = eth_mac_addr(dev, p);
614         if (ret)
615                 return ret;
616
617         if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
618                 vdev->config->set(vdev, offsetof(struct virtio_net_config, mac),
619                                   dev->dev_addr, dev->addr_len);
620
621         return 0;
622 }
623
624 #ifdef CONFIG_NET_POLL_CONTROLLER
625 static void virtnet_netpoll(struct net_device *dev)
626 {
627         struct virtnet_info *vi = netdev_priv(dev);
628
629         napi_schedule(&vi->napi);
630 }
631 #endif
632
633 static int virtnet_open(struct net_device *dev)
634 {
635         struct virtnet_info *vi = netdev_priv(dev);
636
637         napi_enable(&vi->napi);
638
639         /* If all buffers were filled by other side before we napi_enabled, we
640          * won't get another interrupt, so process any outstanding packets
641          * now.  virtnet_poll wants re-enable the queue, so we disable here.
642          * We synchronize against interrupts via NAPI_STATE_SCHED */
643         if (napi_schedule_prep(&vi->napi)) {
644                 virtqueue_disable_cb(vi->rvq);
645                 __napi_schedule(&vi->napi);
646         }
647         return 0;
648 }
649
650 /*
651  * Send command via the control virtqueue and check status.  Commands
652  * supported by the hypervisor, as indicated by feature bits, should
653  * never fail unless improperly formated.
654  */
655 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
656                                  struct scatterlist *data, int out, int in)
657 {
658         struct scatterlist *s, sg[VIRTNET_SEND_COMMAND_SG_MAX + 2];
659         struct virtio_net_ctrl_hdr ctrl;
660         virtio_net_ctrl_ack status = ~0;
661         unsigned int tmp;
662         int i;
663
664         /* Caller should know better */
665         BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ||
666                 (out + in > VIRTNET_SEND_COMMAND_SG_MAX));
667
668         out++; /* Add header */
669         in++; /* Add return status */
670
671         ctrl.class = class;
672         ctrl.cmd = cmd;
673
674         sg_init_table(sg, out + in);
675
676         sg_set_buf(&sg[0], &ctrl, sizeof(ctrl));
677         for_each_sg(data, s, out + in - 2, i)
678                 sg_set_buf(&sg[i + 1], sg_virt(s), s->length);
679         sg_set_buf(&sg[out + in - 1], &status, sizeof(status));
680
681         BUG_ON(virtqueue_add_buf(vi->cvq, sg, out, in, vi) < 0);
682
683         virtqueue_kick(vi->cvq);
684
685         /*
686          * Spin for a response, the kick causes an ioport write, trapping
687          * into the hypervisor, so the request should be handled immediately.
688          */
689         while (!virtqueue_get_buf(vi->cvq, &tmp))
690                 cpu_relax();
691
692         return status == VIRTIO_NET_OK;
693 }
694
695 static int virtnet_close(struct net_device *dev)
696 {
697         struct virtnet_info *vi = netdev_priv(dev);
698
699         napi_disable(&vi->napi);
700
701         return 0;
702 }
703
704 static int virtnet_set_tx_csum(struct net_device *dev, u32 data)
705 {
706         struct virtnet_info *vi = netdev_priv(dev);
707         struct virtio_device *vdev = vi->vdev;
708
709         if (data && !virtio_has_feature(vdev, VIRTIO_NET_F_CSUM))
710                 return -ENOSYS;
711
712         return ethtool_op_set_tx_hw_csum(dev, data);
713 }
714
715 static void virtnet_set_rx_mode(struct net_device *dev)
716 {
717         struct virtnet_info *vi = netdev_priv(dev);
718         struct scatterlist sg[2];
719         u8 promisc, allmulti;
720         struct virtio_net_ctrl_mac *mac_data;
721         struct netdev_hw_addr *ha;
722         int uc_count;
723         int mc_count;
724         void *buf;
725         int i;
726
727         /* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */
728         if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
729                 return;
730
731         promisc = ((dev->flags & IFF_PROMISC) != 0);
732         allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
733
734         sg_init_one(sg, &promisc, sizeof(promisc));
735
736         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
737                                   VIRTIO_NET_CTRL_RX_PROMISC,
738                                   sg, 1, 0))
739                 dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
740                          promisc ? "en" : "dis");
741
742         sg_init_one(sg, &allmulti, sizeof(allmulti));
743
744         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
745                                   VIRTIO_NET_CTRL_RX_ALLMULTI,
746                                   sg, 1, 0))
747                 dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
748                          allmulti ? "en" : "dis");
749
750         uc_count = netdev_uc_count(dev);
751         mc_count = netdev_mc_count(dev);
752         /* MAC filter - use one buffer for both lists */
753         buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
754                       (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
755         mac_data = buf;
756         if (!buf) {
757                 dev_warn(&dev->dev, "No memory for MAC address buffer\n");
758                 return;
759         }
760
761         sg_init_table(sg, 2);
762
763         /* Store the unicast list and count in the front of the buffer */
764         mac_data->entries = uc_count;
765         i = 0;
766         netdev_for_each_uc_addr(ha, dev)
767                 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
768
769         sg_set_buf(&sg[0], mac_data,
770                    sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
771
772         /* multicast list and count fill the end */
773         mac_data = (void *)&mac_data->macs[uc_count][0];
774
775         mac_data->entries = mc_count;
776         i = 0;
777         netdev_for_each_mc_addr(ha, dev)
778                 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
779
780         sg_set_buf(&sg[1], mac_data,
781                    sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
782
783         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
784                                   VIRTIO_NET_CTRL_MAC_TABLE_SET,
785                                   sg, 2, 0))
786                 dev_warn(&dev->dev, "Failed to set MAC fitler table.\n");
787
788         kfree(buf);
789 }
790
791 static void virtnet_vlan_rx_add_vid(struct net_device *dev, u16 vid)
792 {
793         struct virtnet_info *vi = netdev_priv(dev);
794         struct scatterlist sg;
795
796         sg_init_one(&sg, &vid, sizeof(vid));
797
798         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
799                                   VIRTIO_NET_CTRL_VLAN_ADD, &sg, 1, 0))
800                 dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
801 }
802
803 static void virtnet_vlan_rx_kill_vid(struct net_device *dev, u16 vid)
804 {
805         struct virtnet_info *vi = netdev_priv(dev);
806         struct scatterlist sg;
807
808         sg_init_one(&sg, &vid, sizeof(vid));
809
810         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
811                                   VIRTIO_NET_CTRL_VLAN_DEL, &sg, 1, 0))
812                 dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
813 }
814
815 static const struct ethtool_ops virtnet_ethtool_ops = {
816         .set_tx_csum = virtnet_set_tx_csum,
817         .set_sg = ethtool_op_set_sg,
818         .set_tso = ethtool_op_set_tso,
819         .set_ufo = ethtool_op_set_ufo,
820         .get_link = ethtool_op_get_link,
821 };
822
823 #define MIN_MTU 68
824 #define MAX_MTU 65535
825
826 static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
827 {
828         if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
829                 return -EINVAL;
830         dev->mtu = new_mtu;
831         return 0;
832 }
833
834 static const struct net_device_ops virtnet_netdev = {
835         .ndo_open            = virtnet_open,
836         .ndo_stop            = virtnet_close,
837         .ndo_start_xmit      = start_xmit,
838         .ndo_validate_addr   = eth_validate_addr,
839         .ndo_set_mac_address = virtnet_set_mac_address,
840         .ndo_set_rx_mode     = virtnet_set_rx_mode,
841         .ndo_change_mtu      = virtnet_change_mtu,
842         .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
843         .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
844 #ifdef CONFIG_NET_POLL_CONTROLLER
845         .ndo_poll_controller = virtnet_netpoll,
846 #endif
847 };
848
849 static void virtnet_update_status(struct virtnet_info *vi)
850 {
851         u16 v;
852
853         if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS))
854                 return;
855
856         vi->vdev->config->get(vi->vdev,
857                               offsetof(struct virtio_net_config, status),
858                               &v, sizeof(v));
859
860         /* Ignore unknown (future) status bits */
861         v &= VIRTIO_NET_S_LINK_UP;
862
863         if (vi->status == v)
864                 return;
865
866         vi->status = v;
867
868         if (vi->status & VIRTIO_NET_S_LINK_UP) {
869                 netif_carrier_on(vi->dev);
870                 netif_wake_queue(vi->dev);
871         } else {
872                 netif_carrier_off(vi->dev);
873                 netif_stop_queue(vi->dev);
874         }
875 }
876
877 static void virtnet_config_changed(struct virtio_device *vdev)
878 {
879         struct virtnet_info *vi = vdev->priv;
880
881         virtnet_update_status(vi);
882 }
883
884 static int virtnet_probe(struct virtio_device *vdev)
885 {
886         int err;
887         struct net_device *dev;
888         struct virtnet_info *vi;
889         struct virtqueue *vqs[3];
890         vq_callback_t *callbacks[] = { skb_recv_done, skb_xmit_done, NULL};
891         const char *names[] = { "input", "output", "control" };
892         int nvqs;
893
894         /* Allocate ourselves a network device with room for our info */
895         dev = alloc_etherdev(sizeof(struct virtnet_info));
896         if (!dev)
897                 return -ENOMEM;
898
899         /* Set up network device as normal. */
900         dev->netdev_ops = &virtnet_netdev;
901         dev->features = NETIF_F_HIGHDMA;
902         SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
903         SET_NETDEV_DEV(dev, &vdev->dev);
904
905         /* Do we support "hardware" checksums? */
906         if (csum && virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
907                 /* This opens up the world of extra features. */
908                 dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
909                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
910                         dev->features |= NETIF_F_TSO | NETIF_F_UFO
911                                 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
912                 }
913                 /* Individual feature bits: what can host handle? */
914                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
915                         dev->features |= NETIF_F_TSO;
916                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
917                         dev->features |= NETIF_F_TSO6;
918                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
919                         dev->features |= NETIF_F_TSO_ECN;
920                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
921                         dev->features |= NETIF_F_UFO;
922         }
923
924         /* Configuration may specify what MAC to use.  Otherwise random. */
925         if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
926                 vdev->config->get(vdev,
927                                   offsetof(struct virtio_net_config, mac),
928                                   dev->dev_addr, dev->addr_len);
929         } else
930                 random_ether_addr(dev->dev_addr);
931
932         /* Set up our device-specific information */
933         vi = netdev_priv(dev);
934         netif_napi_add(dev, &vi->napi, virtnet_poll, napi_weight);
935         vi->dev = dev;
936         vi->vdev = vdev;
937         vdev->priv = vi;
938         vi->pages = NULL;
939         INIT_DELAYED_WORK(&vi->refill, refill_work);
940         sg_init_table(vi->rx_sg, ARRAY_SIZE(vi->rx_sg));
941         sg_init_table(vi->tx_sg, ARRAY_SIZE(vi->tx_sg));
942
943         /* If we can receive ANY GSO packets, we must allocate large ones. */
944         if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
945             virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
946             virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
947                 vi->big_packets = true;
948
949         if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
950                 vi->mergeable_rx_bufs = true;
951
952         /* We expect two virtqueues, receive then send,
953          * and optionally control. */
954         nvqs = virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ? 3 : 2;
955
956         err = vdev->config->find_vqs(vdev, nvqs, vqs, callbacks, names);
957         if (err)
958                 goto free;
959
960         vi->rvq = vqs[0];
961         vi->svq = vqs[1];
962
963         if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)) {
964                 vi->cvq = vqs[2];
965
966                 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
967                         dev->features |= NETIF_F_HW_VLAN_FILTER;
968         }
969
970         err = register_netdev(dev);
971         if (err) {
972                 pr_debug("virtio_net: registering device failed\n");
973                 goto free_vqs;
974         }
975
976         /* Last of all, set up some receive buffers. */
977         try_fill_recv(vi, GFP_KERNEL);
978
979         /* If we didn't even get one input buffer, we're useless. */
980         if (vi->num == 0) {
981                 err = -ENOMEM;
982                 goto unregister;
983         }
984
985         vi->status = VIRTIO_NET_S_LINK_UP;
986         virtnet_update_status(vi);
987         netif_carrier_on(dev);
988
989         pr_debug("virtnet: registered device %s\n", dev->name);
990         return 0;
991
992 unregister:
993         unregister_netdev(dev);
994         cancel_delayed_work_sync(&vi->refill);
995 free_vqs:
996         vdev->config->del_vqs(vdev);
997 free:
998         free_netdev(dev);
999         return err;
1000 }
1001
1002 static void free_unused_bufs(struct virtnet_info *vi)
1003 {
1004         void *buf;
1005         while (1) {
1006                 buf = virtqueue_detach_unused_buf(vi->svq);
1007                 if (!buf)
1008                         break;
1009                 dev_kfree_skb(buf);
1010         }
1011         while (1) {
1012                 buf = virtqueue_detach_unused_buf(vi->rvq);
1013                 if (!buf)
1014                         break;
1015                 if (vi->mergeable_rx_bufs || vi->big_packets)
1016                         give_pages(vi, buf);
1017                 else
1018                         dev_kfree_skb(buf);
1019                 --vi->num;
1020         }
1021         BUG_ON(vi->num != 0);
1022 }
1023
1024 static void __devexit virtnet_remove(struct virtio_device *vdev)
1025 {
1026         struct virtnet_info *vi = vdev->priv;
1027
1028         /* Stop all the virtqueues. */
1029         vdev->config->reset(vdev);
1030
1031
1032         unregister_netdev(vi->dev);
1033         cancel_delayed_work_sync(&vi->refill);
1034
1035         /* Free unused buffers in both send and recv, if any. */
1036         free_unused_bufs(vi);
1037
1038         vdev->config->del_vqs(vi->vdev);
1039
1040         while (vi->pages)
1041                 __free_pages(get_a_page(vi, GFP_KERNEL), 0);
1042
1043         free_netdev(vi->dev);
1044 }
1045
1046 static struct virtio_device_id id_table[] = {
1047         { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
1048         { 0 },
1049 };
1050
1051 static unsigned int features[] = {
1052         VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
1053         VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
1054         VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
1055         VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
1056         VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO,
1057         VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ,
1058         VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN,
1059 };
1060
1061 static struct virtio_driver virtio_net_driver = {
1062         .feature_table = features,
1063         .feature_table_size = ARRAY_SIZE(features),
1064         .driver.name =  KBUILD_MODNAME,
1065         .driver.owner = THIS_MODULE,
1066         .id_table =     id_table,
1067         .probe =        virtnet_probe,
1068         .remove =       __devexit_p(virtnet_remove),
1069         .config_changed = virtnet_config_changed,
1070 };
1071
1072 static int __init init(void)
1073 {
1074         return register_virtio_driver(&virtio_net_driver);
1075 }
1076
1077 static void __exit fini(void)
1078 {
1079         unregister_virtio_driver(&virtio_net_driver);
1080 }
1081 module_init(init);
1082 module_exit(fini);
1083
1084 MODULE_DEVICE_TABLE(virtio, id_table);
1085 MODULE_DESCRIPTION("Virtio network driver");
1086 MODULE_LICENSE("GPL");