1 /* A network driver using virtio.
3 * Copyright 2007 Rusty Russell <rusty@rustcorp.com.au> IBM Corporation
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.
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.
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
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>
30 static int napi_weight = 128;
31 module_param(napi_weight, int, 0444);
33 static int csum = 1, gso = 1;
34 module_param(csum, bool, 0444);
35 module_param(gso, bool, 0444);
37 /* FIXME: MTU in config. */
38 #define MAX_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
39 #define GOOD_COPY_LEN 128
41 #define VIRTNET_SEND_COMMAND_SG_MAX 2
43 struct virtnet_stats {
44 struct u64_stats_sync syncp;
53 struct virtio_device *vdev;
54 struct virtqueue *rvq, *svq, *cvq;
55 struct net_device *dev;
56 struct napi_struct napi;
59 /* Number of input buffers, and max we've ever had. */
60 unsigned int num, max;
62 /* I like... big packets and I cannot lie! */
65 /* Host will merge rx buffers for big packets (shake it! shake it!) */
66 bool mergeable_rx_bufs;
68 /* Active statistics */
69 struct virtnet_stats __percpu *stats;
71 /* Work struct for refilling if we run low on memory. */
72 struct delayed_work refill;
74 /* Chain pages by the private ptr. */
77 /* fragments + linear part + virtio header */
78 struct scatterlist rx_sg[MAX_SKB_FRAGS + 2];
79 struct scatterlist tx_sg[MAX_SKB_FRAGS + 2];
84 struct virtio_net_hdr hdr;
85 struct virtio_net_hdr_mrg_rxbuf mhdr;
90 struct padded_vnet_hdr {
91 struct virtio_net_hdr hdr;
93 * virtio_net_hdr should be in a separated sg buffer because of a
94 * QEMU bug, and data sg buffer shares same page with this header sg.
95 * This padding makes next sg 16 byte aligned after virtio_net_hdr.
100 static inline struct skb_vnet_hdr *skb_vnet_hdr(struct sk_buff *skb)
102 return (struct skb_vnet_hdr *)skb->cb;
106 * private is used to chain pages for big packets, put the whole
107 * most recent used list in the beginning for reuse
109 static void give_pages(struct virtnet_info *vi, struct page *page)
113 /* Find end of list, sew whole thing into vi->pages. */
114 for (end = page; end->private; end = (struct page *)end->private);
115 end->private = (unsigned long)vi->pages;
119 static struct page *get_a_page(struct virtnet_info *vi, gfp_t gfp_mask)
121 struct page *p = vi->pages;
124 vi->pages = (struct page *)p->private;
125 /* clear private here, it is used to chain pages */
128 p = alloc_page(gfp_mask);
132 static void skb_xmit_done(struct virtqueue *svq)
134 struct virtnet_info *vi = svq->vdev->priv;
136 /* Suppress further interrupts. */
137 virtqueue_disable_cb(svq);
139 /* We were probably waiting for more output buffers. */
140 netif_wake_queue(vi->dev);
143 static void set_skb_frag(struct sk_buff *skb, struct page *page,
144 unsigned int offset, unsigned int *len)
146 int size = min((unsigned)PAGE_SIZE - offset, *len);
147 int i = skb_shinfo(skb)->nr_frags;
149 __skb_fill_page_desc(skb, i, page, offset, size);
151 skb->data_len += size;
153 skb->truesize += PAGE_SIZE;
154 skb_shinfo(skb)->nr_frags++;
158 static struct sk_buff *page_to_skb(struct virtnet_info *vi,
159 struct page *page, unsigned int len)
162 struct skb_vnet_hdr *hdr;
163 unsigned int copy, hdr_len, offset;
166 p = page_address(page);
168 /* copy small packet so we can reuse these pages for small data */
169 skb = netdev_alloc_skb_ip_align(vi->dev, GOOD_COPY_LEN);
173 hdr = skb_vnet_hdr(skb);
175 if (vi->mergeable_rx_bufs) {
176 hdr_len = sizeof hdr->mhdr;
179 hdr_len = sizeof hdr->hdr;
180 offset = sizeof(struct padded_vnet_hdr);
183 memcpy(hdr, p, hdr_len);
189 if (copy > skb_tailroom(skb))
190 copy = skb_tailroom(skb);
191 memcpy(skb_put(skb, copy), p, copy);
197 * Verify that we can indeed put this data into a skb.
198 * This is here to handle cases when the device erroneously
199 * tries to receive more than is possible. This is usually
200 * the case of a broken device.
202 if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) {
204 pr_debug("%s: too much data\n", skb->dev->name);
210 set_skb_frag(skb, page, offset, &len);
211 page = (struct page *)page->private;
216 give_pages(vi, page);
221 static int receive_mergeable(struct virtnet_info *vi, struct sk_buff *skb)
223 struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
227 num_buf = hdr->mhdr.num_buffers;
229 i = skb_shinfo(skb)->nr_frags;
230 if (i >= MAX_SKB_FRAGS) {
231 pr_debug("%s: packet too long\n", skb->dev->name);
232 skb->dev->stats.rx_length_errors++;
235 page = virtqueue_get_buf(vi->rvq, &len);
237 pr_debug("%s: rx error: %d buffers missing\n",
238 skb->dev->name, hdr->mhdr.num_buffers);
239 skb->dev->stats.rx_length_errors++;
246 set_skb_frag(skb, page, 0, &len);
253 static void receive_buf(struct net_device *dev, void *buf, unsigned int len)
255 struct virtnet_info *vi = netdev_priv(dev);
256 struct virtnet_stats __percpu *stats = this_cpu_ptr(vi->stats);
259 struct skb_vnet_hdr *hdr;
261 if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
262 pr_debug("%s: short packet %i\n", dev->name, len);
263 dev->stats.rx_length_errors++;
264 if (vi->mergeable_rx_bufs || vi->big_packets)
271 if (!vi->mergeable_rx_bufs && !vi->big_packets) {
273 len -= sizeof(struct virtio_net_hdr);
277 skb = page_to_skb(vi, page, len);
278 if (unlikely(!skb)) {
279 dev->stats.rx_dropped++;
280 give_pages(vi, page);
283 if (vi->mergeable_rx_bufs)
284 if (receive_mergeable(vi, skb)) {
290 hdr = skb_vnet_hdr(skb);
292 u64_stats_update_begin(&stats->syncp);
293 stats->rx_bytes += skb->len;
295 u64_stats_update_end(&stats->syncp);
297 if (hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
298 pr_debug("Needs csum!\n");
299 if (!skb_partial_csum_set(skb,
301 hdr->hdr.csum_offset))
303 } else if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) {
304 skb->ip_summed = CHECKSUM_UNNECESSARY;
307 skb->protocol = eth_type_trans(skb, dev);
308 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
309 ntohs(skb->protocol), skb->len, skb->pkt_type);
311 if (hdr->hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
313 switch (hdr->hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
314 case VIRTIO_NET_HDR_GSO_TCPV4:
315 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
317 case VIRTIO_NET_HDR_GSO_UDP:
318 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
320 case VIRTIO_NET_HDR_GSO_TCPV6:
321 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
325 printk(KERN_WARNING "%s: bad gso type %u.\n",
326 dev->name, hdr->hdr.gso_type);
330 if (hdr->hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
331 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
333 skb_shinfo(skb)->gso_size = hdr->hdr.gso_size;
334 if (skb_shinfo(skb)->gso_size == 0) {
336 printk(KERN_WARNING "%s: zero gso size.\n",
341 /* Header must be checked, and gso_segs computed. */
342 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
343 skb_shinfo(skb)->gso_segs = 0;
346 netif_receive_skb(skb);
350 dev->stats.rx_frame_errors++;
354 static int add_recvbuf_small(struct virtnet_info *vi, gfp_t gfp)
357 struct skb_vnet_hdr *hdr;
360 skb = netdev_alloc_skb_ip_align(vi->dev, MAX_PACKET_LEN);
364 skb_put(skb, MAX_PACKET_LEN);
366 hdr = skb_vnet_hdr(skb);
367 sg_set_buf(vi->rx_sg, &hdr->hdr, sizeof hdr->hdr);
369 skb_to_sgvec(skb, vi->rx_sg + 1, 0, skb->len);
371 err = virtqueue_add_buf_gfp(vi->rvq, vi->rx_sg, 0, 2, skb, gfp);
378 static int add_recvbuf_big(struct virtnet_info *vi, gfp_t gfp)
380 struct page *first, *list = NULL;
384 /* page in vi->rx_sg[MAX_SKB_FRAGS + 1] is list tail */
385 for (i = MAX_SKB_FRAGS + 1; i > 1; --i) {
386 first = get_a_page(vi, gfp);
389 give_pages(vi, list);
392 sg_set_buf(&vi->rx_sg[i], page_address(first), PAGE_SIZE);
394 /* chain new page in list head to match sg */
395 first->private = (unsigned long)list;
399 first = get_a_page(vi, gfp);
401 give_pages(vi, list);
404 p = page_address(first);
406 /* vi->rx_sg[0], vi->rx_sg[1] share the same page */
407 /* a separated vi->rx_sg[0] for virtio_net_hdr only due to QEMU bug */
408 sg_set_buf(&vi->rx_sg[0], p, sizeof(struct virtio_net_hdr));
410 /* vi->rx_sg[1] for data packet, from offset */
411 offset = sizeof(struct padded_vnet_hdr);
412 sg_set_buf(&vi->rx_sg[1], p + offset, PAGE_SIZE - offset);
414 /* chain first in list head */
415 first->private = (unsigned long)list;
416 err = virtqueue_add_buf_gfp(vi->rvq, vi->rx_sg, 0, MAX_SKB_FRAGS + 2,
419 give_pages(vi, first);
424 static int add_recvbuf_mergeable(struct virtnet_info *vi, gfp_t gfp)
429 page = get_a_page(vi, gfp);
433 sg_init_one(vi->rx_sg, page_address(page), PAGE_SIZE);
435 err = virtqueue_add_buf_gfp(vi->rvq, vi->rx_sg, 0, 1, page, gfp);
437 give_pages(vi, page);
442 /* Returns false if we couldn't fill entirely (OOM). */
443 static bool try_fill_recv(struct virtnet_info *vi, gfp_t gfp)
449 if (vi->mergeable_rx_bufs)
450 err = add_recvbuf_mergeable(vi, gfp);
451 else if (vi->big_packets)
452 err = add_recvbuf_big(vi, gfp);
454 err = add_recvbuf_small(vi, gfp);
456 oom = err == -ENOMEM;
461 if (unlikely(vi->num > vi->max))
463 virtqueue_kick(vi->rvq);
467 static void skb_recv_done(struct virtqueue *rvq)
469 struct virtnet_info *vi = rvq->vdev->priv;
470 /* Schedule NAPI, Suppress further interrupts if successful. */
471 if (napi_schedule_prep(&vi->napi)) {
472 virtqueue_disable_cb(rvq);
473 __napi_schedule(&vi->napi);
477 static void virtnet_napi_enable(struct virtnet_info *vi)
479 napi_enable(&vi->napi);
481 /* If all buffers were filled by other side before we napi_enabled, we
482 * won't get another interrupt, so process any outstanding packets
483 * now. virtnet_poll wants re-enable the queue, so we disable here.
484 * We synchronize against interrupts via NAPI_STATE_SCHED */
485 if (napi_schedule_prep(&vi->napi)) {
486 virtqueue_disable_cb(vi->rvq);
487 __napi_schedule(&vi->napi);
491 static void refill_work(struct work_struct *work)
493 struct virtnet_info *vi;
496 vi = container_of(work, struct virtnet_info, refill.work);
497 napi_disable(&vi->napi);
498 still_empty = !try_fill_recv(vi, GFP_KERNEL);
499 virtnet_napi_enable(vi);
501 /* In theory, this can happen: if we don't get any buffers in
502 * we will *never* try to fill again. */
504 schedule_delayed_work(&vi->refill, HZ/2);
507 static int virtnet_poll(struct napi_struct *napi, int budget)
509 struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi);
511 unsigned int r, len, received = 0;
514 while (received < budget &&
515 (buf = virtqueue_get_buf(vi->rvq, &len)) != NULL) {
516 receive_buf(vi->dev, buf, len);
521 if (vi->num < vi->max / 2) {
522 if (!try_fill_recv(vi, GFP_ATOMIC))
523 schedule_delayed_work(&vi->refill, 0);
526 /* Out of packets? */
527 if (received < budget) {
528 r = virtqueue_enable_cb_prepare(vi->rvq);
530 if (unlikely(virtqueue_poll(vi->rvq, r)) &&
531 napi_schedule_prep(napi)) {
532 virtqueue_disable_cb(vi->rvq);
533 __napi_schedule(napi);
541 static unsigned int free_old_xmit_skbs(struct virtnet_info *vi)
544 unsigned int len, tot_sgs = 0;
545 struct virtnet_stats __percpu *stats = this_cpu_ptr(vi->stats);
547 while ((skb = virtqueue_get_buf(vi->svq, &len)) != NULL) {
548 pr_debug("Sent skb %p\n", skb);
550 u64_stats_update_begin(&stats->syncp);
551 stats->tx_bytes += skb->len;
553 u64_stats_update_end(&stats->syncp);
555 tot_sgs += skb_vnet_hdr(skb)->num_sg;
556 dev_kfree_skb_any(skb);
561 static int xmit_skb(struct virtnet_info *vi, struct sk_buff *skb)
563 struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
564 const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
566 pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
568 if (skb->ip_summed == CHECKSUM_PARTIAL) {
569 hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
570 hdr->hdr.csum_start = skb_checksum_start_offset(skb);
571 hdr->hdr.csum_offset = skb->csum_offset;
574 hdr->hdr.csum_offset = hdr->hdr.csum_start = 0;
577 if (skb_is_gso(skb)) {
578 hdr->hdr.hdr_len = skb_headlen(skb);
579 hdr->hdr.gso_size = skb_shinfo(skb)->gso_size;
580 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
581 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
582 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
583 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
584 else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
585 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
588 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
589 hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
591 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
592 hdr->hdr.gso_size = hdr->hdr.hdr_len = 0;
595 hdr->mhdr.num_buffers = 0;
597 /* Encode metadata header at front. */
598 if (vi->mergeable_rx_bufs)
599 sg_set_buf(vi->tx_sg, &hdr->mhdr, sizeof hdr->mhdr);
601 sg_set_buf(vi->tx_sg, &hdr->hdr, sizeof hdr->hdr);
603 hdr->num_sg = skb_to_sgvec(skb, vi->tx_sg + 1, 0, skb->len) + 1;
604 return virtqueue_add_buf(vi->svq, vi->tx_sg, hdr->num_sg,
608 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
610 struct virtnet_info *vi = netdev_priv(dev);
613 /* Free up any pending old buffers before queueing new ones. */
614 free_old_xmit_skbs(vi);
616 /* Try to transmit */
617 capacity = xmit_skb(vi, skb);
619 /* This can happen with OOM and indirect buffers. */
620 if (unlikely(capacity < 0)) {
621 if (net_ratelimit()) {
622 if (likely(capacity == -ENOMEM)) {
624 "TX queue failure: out of memory\n");
626 dev->stats.tx_fifo_errors++;
628 "Unexpected TX queue failure: %d\n",
632 dev->stats.tx_dropped++;
636 virtqueue_kick(vi->svq);
638 /* Don't wait up for transmitted skbs to be freed. */
642 /* Apparently nice girls don't return TX_BUSY; stop the queue
643 * before it gets out of hand. Naturally, this wastes entries. */
644 if (capacity < 2+MAX_SKB_FRAGS) {
645 netif_stop_queue(dev);
646 if (unlikely(!virtqueue_enable_cb_delayed(vi->svq))) {
647 /* More just got used, free them then recheck. */
648 capacity += free_old_xmit_skbs(vi);
649 if (capacity >= 2+MAX_SKB_FRAGS) {
650 netif_start_queue(dev);
651 virtqueue_disable_cb(vi->svq);
659 static int virtnet_set_mac_address(struct net_device *dev, void *p)
661 struct virtnet_info *vi = netdev_priv(dev);
662 struct virtio_device *vdev = vi->vdev;
665 ret = eth_mac_addr(dev, p);
669 if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
670 vdev->config->set(vdev, offsetof(struct virtio_net_config, mac),
671 dev->dev_addr, dev->addr_len);
676 static struct rtnl_link_stats64 *virtnet_stats(struct net_device *dev,
677 struct rtnl_link_stats64 *tot)
679 struct virtnet_info *vi = netdev_priv(dev);
683 for_each_possible_cpu(cpu) {
684 struct virtnet_stats __percpu *stats
685 = per_cpu_ptr(vi->stats, cpu);
686 u64 tpackets, tbytes, rpackets, rbytes;
689 start = u64_stats_fetch_begin(&stats->syncp);
690 tpackets = stats->tx_packets;
691 tbytes = stats->tx_bytes;
692 rpackets = stats->rx_packets;
693 rbytes = stats->rx_bytes;
694 } while (u64_stats_fetch_retry(&stats->syncp, start));
696 tot->rx_packets += rpackets;
697 tot->tx_packets += tpackets;
698 tot->rx_bytes += rbytes;
699 tot->tx_bytes += tbytes;
702 tot->tx_dropped = dev->stats.tx_dropped;
703 tot->rx_dropped = dev->stats.rx_dropped;
704 tot->rx_length_errors = dev->stats.rx_length_errors;
705 tot->rx_frame_errors = dev->stats.rx_frame_errors;
710 #ifdef CONFIG_NET_POLL_CONTROLLER
711 static void virtnet_netpoll(struct net_device *dev)
713 struct virtnet_info *vi = netdev_priv(dev);
715 napi_schedule(&vi->napi);
719 static int virtnet_open(struct net_device *dev)
721 struct virtnet_info *vi = netdev_priv(dev);
723 virtnet_napi_enable(vi);
728 * Send command via the control virtqueue and check status. Commands
729 * supported by the hypervisor, as indicated by feature bits, should
730 * never fail unless improperly formated.
732 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
733 struct scatterlist *data, int out, int in)
735 struct scatterlist *s, sg[VIRTNET_SEND_COMMAND_SG_MAX + 2];
736 struct virtio_net_ctrl_hdr ctrl;
737 virtio_net_ctrl_ack status = ~0;
741 /* Caller should know better */
742 BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ||
743 (out + in > VIRTNET_SEND_COMMAND_SG_MAX));
745 out++; /* Add header */
746 in++; /* Add return status */
751 sg_init_table(sg, out + in);
753 sg_set_buf(&sg[0], &ctrl, sizeof(ctrl));
754 for_each_sg(data, s, out + in - 2, i)
755 sg_set_buf(&sg[i + 1], sg_virt(s), s->length);
756 sg_set_buf(&sg[out + in - 1], &status, sizeof(status));
758 BUG_ON(virtqueue_add_buf(vi->cvq, sg, out, in, vi) < 0);
760 virtqueue_kick(vi->cvq);
763 * Spin for a response, the kick causes an ioport write, trapping
764 * into the hypervisor, so the request should be handled immediately.
766 while (!virtqueue_get_buf(vi->cvq, &tmp))
769 return status == VIRTIO_NET_OK;
772 static int virtnet_close(struct net_device *dev)
774 struct virtnet_info *vi = netdev_priv(dev);
776 napi_disable(&vi->napi);
781 static void virtnet_set_rx_mode(struct net_device *dev)
783 struct virtnet_info *vi = netdev_priv(dev);
784 struct scatterlist sg[2];
785 u8 promisc, allmulti;
786 struct virtio_net_ctrl_mac *mac_data;
787 struct netdev_hw_addr *ha;
793 /* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */
794 if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
797 promisc = ((dev->flags & IFF_PROMISC) != 0);
798 allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
800 sg_init_one(sg, &promisc, sizeof(promisc));
802 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
803 VIRTIO_NET_CTRL_RX_PROMISC,
805 dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
806 promisc ? "en" : "dis");
808 sg_init_one(sg, &allmulti, sizeof(allmulti));
810 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
811 VIRTIO_NET_CTRL_RX_ALLMULTI,
813 dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
814 allmulti ? "en" : "dis");
816 uc_count = netdev_uc_count(dev);
817 mc_count = netdev_mc_count(dev);
818 /* MAC filter - use one buffer for both lists */
819 buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
820 (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
823 dev_warn(&dev->dev, "No memory for MAC address buffer\n");
827 sg_init_table(sg, 2);
829 /* Store the unicast list and count in the front of the buffer */
830 mac_data->entries = uc_count;
832 netdev_for_each_uc_addr(ha, dev)
833 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
835 sg_set_buf(&sg[0], mac_data,
836 sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
838 /* multicast list and count fill the end */
839 mac_data = (void *)&mac_data->macs[uc_count][0];
841 mac_data->entries = mc_count;
843 netdev_for_each_mc_addr(ha, dev)
844 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
846 sg_set_buf(&sg[1], mac_data,
847 sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
849 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
850 VIRTIO_NET_CTRL_MAC_TABLE_SET,
852 dev_warn(&dev->dev, "Failed to set MAC fitler table.\n");
857 static void virtnet_vlan_rx_add_vid(struct net_device *dev, u16 vid)
859 struct virtnet_info *vi = netdev_priv(dev);
860 struct scatterlist sg;
862 sg_init_one(&sg, &vid, sizeof(vid));
864 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
865 VIRTIO_NET_CTRL_VLAN_ADD, &sg, 1, 0))
866 dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
869 static void virtnet_vlan_rx_kill_vid(struct net_device *dev, u16 vid)
871 struct virtnet_info *vi = netdev_priv(dev);
872 struct scatterlist sg;
874 sg_init_one(&sg, &vid, sizeof(vid));
876 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
877 VIRTIO_NET_CTRL_VLAN_DEL, &sg, 1, 0))
878 dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
881 static void virtnet_get_ringparam(struct net_device *dev,
882 struct ethtool_ringparam *ring)
884 struct virtnet_info *vi = netdev_priv(dev);
886 ring->rx_max_pending = virtqueue_get_vring_size(vi->rvq);
887 ring->tx_max_pending = virtqueue_get_vring_size(vi->svq);
888 ring->rx_pending = ring->rx_max_pending;
889 ring->tx_pending = ring->tx_max_pending;
893 static const struct ethtool_ops virtnet_ethtool_ops = {
894 .get_link = ethtool_op_get_link,
895 .get_ringparam = virtnet_get_ringparam,
899 #define MAX_MTU 65535
901 static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
903 if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
909 static const struct net_device_ops virtnet_netdev = {
910 .ndo_open = virtnet_open,
911 .ndo_stop = virtnet_close,
912 .ndo_start_xmit = start_xmit,
913 .ndo_validate_addr = eth_validate_addr,
914 .ndo_set_mac_address = virtnet_set_mac_address,
915 .ndo_set_rx_mode = virtnet_set_rx_mode,
916 .ndo_change_mtu = virtnet_change_mtu,
917 .ndo_get_stats64 = virtnet_stats,
918 .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
919 .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
920 #ifdef CONFIG_NET_POLL_CONTROLLER
921 .ndo_poll_controller = virtnet_netpoll,
925 static void virtnet_update_status(struct virtnet_info *vi)
929 if (virtio_config_val(vi->vdev, VIRTIO_NET_F_STATUS,
930 offsetof(struct virtio_net_config, status),
934 /* Ignore unknown (future) status bits */
935 v &= VIRTIO_NET_S_LINK_UP;
942 if (vi->status & VIRTIO_NET_S_LINK_UP) {
943 netif_carrier_on(vi->dev);
944 netif_wake_queue(vi->dev);
946 netif_carrier_off(vi->dev);
947 netif_stop_queue(vi->dev);
951 static void virtnet_config_changed(struct virtio_device *vdev)
953 struct virtnet_info *vi = vdev->priv;
955 virtnet_update_status(vi);
958 static int virtnet_probe(struct virtio_device *vdev)
961 struct net_device *dev;
962 struct virtnet_info *vi;
963 struct virtqueue *vqs[3];
964 vq_callback_t *callbacks[] = { skb_recv_done, skb_xmit_done, NULL};
965 const char *names[] = { "input", "output", "control" };
968 /* Allocate ourselves a network device with room for our info */
969 dev = alloc_etherdev(sizeof(struct virtnet_info));
973 /* Set up network device as normal. */
974 dev->priv_flags |= IFF_UNICAST_FLT;
975 dev->netdev_ops = &virtnet_netdev;
976 dev->features = NETIF_F_HIGHDMA;
978 SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
979 SET_NETDEV_DEV(dev, &vdev->dev);
981 /* Do we support "hardware" checksums? */
982 if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
983 /* This opens up the world of extra features. */
984 dev->hw_features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
986 dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
988 if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
989 dev->hw_features |= NETIF_F_TSO | NETIF_F_UFO
990 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
992 /* Individual feature bits: what can host handle? */
993 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
994 dev->hw_features |= NETIF_F_TSO;
995 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
996 dev->hw_features |= NETIF_F_TSO6;
997 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
998 dev->hw_features |= NETIF_F_TSO_ECN;
999 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
1000 dev->hw_features |= NETIF_F_UFO;
1003 dev->features |= dev->hw_features & (NETIF_F_ALL_TSO|NETIF_F_UFO);
1004 /* (!csum && gso) case will be fixed by register_netdev() */
1007 /* Configuration may specify what MAC to use. Otherwise random. */
1008 if (virtio_config_val_len(vdev, VIRTIO_NET_F_MAC,
1009 offsetof(struct virtio_net_config, mac),
1010 dev->dev_addr, dev->addr_len) < 0)
1011 random_ether_addr(dev->dev_addr);
1013 /* Set up our device-specific information */
1014 vi = netdev_priv(dev);
1015 netif_napi_add(dev, &vi->napi, virtnet_poll, napi_weight);
1020 vi->stats = alloc_percpu(struct virtnet_stats);
1022 if (vi->stats == NULL)
1025 INIT_DELAYED_WORK(&vi->refill, refill_work);
1026 sg_init_table(vi->rx_sg, ARRAY_SIZE(vi->rx_sg));
1027 sg_init_table(vi->tx_sg, ARRAY_SIZE(vi->tx_sg));
1029 /* If we can receive ANY GSO packets, we must allocate large ones. */
1030 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
1031 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
1032 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN) ||
1033 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_UFO))
1034 vi->big_packets = true;
1036 if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
1037 vi->mergeable_rx_bufs = true;
1039 /* We expect two virtqueues, receive then send,
1040 * and optionally control. */
1041 nvqs = virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ? 3 : 2;
1043 err = vdev->config->find_vqs(vdev, nvqs, vqs, callbacks, names);
1050 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)) {
1053 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
1054 dev->features |= NETIF_F_HW_VLAN_FILTER;
1057 err = register_netdev(dev);
1059 pr_debug("virtio_net: registering device failed\n");
1063 /* Last of all, set up some receive buffers. */
1064 try_fill_recv(vi, GFP_KERNEL);
1066 /* If we didn't even get one input buffer, we're useless. */
1072 /* Assume link up if device can't report link status,
1073 otherwise get link status from config. */
1074 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
1075 netif_carrier_off(dev);
1076 virtnet_update_status(vi);
1078 vi->status = VIRTIO_NET_S_LINK_UP;
1079 netif_carrier_on(dev);
1082 pr_debug("virtnet: registered device %s\n", dev->name);
1086 unregister_netdev(dev);
1087 cancel_delayed_work_sync(&vi->refill);
1089 vdev->config->del_vqs(vdev);
1091 free_percpu(vi->stats);
1097 static void free_unused_bufs(struct virtnet_info *vi)
1101 buf = virtqueue_detach_unused_buf(vi->svq);
1107 buf = virtqueue_detach_unused_buf(vi->rvq);
1110 if (vi->mergeable_rx_bufs || vi->big_packets)
1111 give_pages(vi, buf);
1116 BUG_ON(vi->num != 0);
1119 static void __devexit virtnet_remove(struct virtio_device *vdev)
1121 struct virtnet_info *vi = vdev->priv;
1123 /* Stop all the virtqueues. */
1124 vdev->config->reset(vdev);
1127 unregister_netdev(vi->dev);
1128 cancel_delayed_work_sync(&vi->refill);
1130 /* Free unused buffers in both send and recv, if any. */
1131 free_unused_bufs(vi);
1133 vdev->config->del_vqs(vi->vdev);
1136 __free_pages(get_a_page(vi, GFP_KERNEL), 0);
1138 free_percpu(vi->stats);
1139 free_netdev(vi->dev);
1142 static struct virtio_device_id id_table[] = {
1143 { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
1147 static unsigned int features[] = {
1148 VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
1149 VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
1150 VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
1151 VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
1152 VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO,
1153 VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ,
1154 VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN,
1157 static struct virtio_driver virtio_net_driver = {
1158 .feature_table = features,
1159 .feature_table_size = ARRAY_SIZE(features),
1160 .driver.name = KBUILD_MODNAME,
1161 .driver.owner = THIS_MODULE,
1162 .id_table = id_table,
1163 .probe = virtnet_probe,
1164 .remove = __devexit_p(virtnet_remove),
1165 .config_changed = virtnet_config_changed,
1168 static int __init init(void)
1170 return register_virtio_driver(&virtio_net_driver);
1173 static void __exit fini(void)
1175 unregister_virtio_driver(&virtio_net_driver);
1180 MODULE_DEVICE_TABLE(virtio, id_table);
1181 MODULE_DESCRIPTION("Virtio network driver");
1182 MODULE_LICENSE("GPL");