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 bool csum = true, gso = true;
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
42 #define VIRTNET_DRIVER_VERSION "1.0.0"
44 struct virtnet_stats {
45 struct u64_stats_sync tx_syncp;
46 struct u64_stats_sync rx_syncp;
55 struct virtio_device *vdev;
56 struct virtqueue *rvq, *svq, *cvq;
57 struct net_device *dev;
58 struct napi_struct napi;
61 /* Number of input buffers, and max we've ever had. */
62 unsigned int num, max;
64 /* I like... big packets and I cannot lie! */
67 /* Host will merge rx buffers for big packets (shake it! shake it!) */
68 bool mergeable_rx_bufs;
70 /* enable config space updates */
73 /* Active statistics */
74 struct virtnet_stats __percpu *stats;
76 /* Work struct for refilling if we run low on memory. */
77 struct delayed_work refill;
79 /* Work struct for config space updates */
80 struct work_struct config_work;
82 /* Lock for config space updates */
83 struct mutex config_lock;
85 /* Chain pages by the private ptr. */
88 /* fragments + linear part + virtio header */
89 struct scatterlist rx_sg[MAX_SKB_FRAGS + 2];
90 struct scatterlist tx_sg[MAX_SKB_FRAGS + 2];
95 struct virtio_net_hdr hdr;
96 struct virtio_net_hdr_mrg_rxbuf mhdr;
100 struct padded_vnet_hdr {
101 struct virtio_net_hdr hdr;
103 * virtio_net_hdr should be in a separated sg buffer because of a
104 * QEMU bug, and data sg buffer shares same page with this header sg.
105 * This padding makes next sg 16 byte aligned after virtio_net_hdr.
110 static inline struct skb_vnet_hdr *skb_vnet_hdr(struct sk_buff *skb)
112 return (struct skb_vnet_hdr *)skb->cb;
116 * private is used to chain pages for big packets, put the whole
117 * most recent used list in the beginning for reuse
119 static void give_pages(struct virtnet_info *vi, struct page *page)
123 /* Find end of list, sew whole thing into vi->pages. */
124 for (end = page; end->private; end = (struct page *)end->private);
125 end->private = (unsigned long)vi->pages;
129 static struct page *get_a_page(struct virtnet_info *vi, gfp_t gfp_mask)
131 struct page *p = vi->pages;
134 vi->pages = (struct page *)p->private;
135 /* clear private here, it is used to chain pages */
138 p = alloc_page(gfp_mask);
142 static void skb_xmit_done(struct virtqueue *svq)
144 struct virtnet_info *vi = svq->vdev->priv;
146 /* Suppress further interrupts. */
147 virtqueue_disable_cb(svq);
149 /* We were probably waiting for more output buffers. */
150 netif_wake_queue(vi->dev);
153 static void set_skb_frag(struct sk_buff *skb, struct page *page,
154 unsigned int offset, unsigned int *len)
156 int size = min((unsigned)PAGE_SIZE - offset, *len);
157 int i = skb_shinfo(skb)->nr_frags;
159 __skb_fill_page_desc(skb, i, page, offset, size);
161 skb->data_len += size;
163 skb->truesize += PAGE_SIZE;
164 skb_shinfo(skb)->nr_frags++;
168 /* Called from bottom half context */
169 static struct sk_buff *page_to_skb(struct virtnet_info *vi,
170 struct page *page, unsigned int len)
173 struct skb_vnet_hdr *hdr;
174 unsigned int copy, hdr_len, offset;
177 p = page_address(page);
179 /* copy small packet so we can reuse these pages for small data */
180 skb = netdev_alloc_skb_ip_align(vi->dev, GOOD_COPY_LEN);
184 hdr = skb_vnet_hdr(skb);
186 if (vi->mergeable_rx_bufs) {
187 hdr_len = sizeof hdr->mhdr;
190 hdr_len = sizeof hdr->hdr;
191 offset = sizeof(struct padded_vnet_hdr);
194 memcpy(hdr, p, hdr_len);
200 if (copy > skb_tailroom(skb))
201 copy = skb_tailroom(skb);
202 memcpy(skb_put(skb, copy), p, copy);
208 * Verify that we can indeed put this data into a skb.
209 * This is here to handle cases when the device erroneously
210 * tries to receive more than is possible. This is usually
211 * the case of a broken device.
213 if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) {
215 pr_debug("%s: too much data\n", skb->dev->name);
221 set_skb_frag(skb, page, offset, &len);
222 page = (struct page *)page->private;
227 give_pages(vi, page);
232 static int receive_mergeable(struct virtnet_info *vi, struct sk_buff *skb)
234 struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
238 num_buf = hdr->mhdr.num_buffers;
240 i = skb_shinfo(skb)->nr_frags;
241 if (i >= MAX_SKB_FRAGS) {
242 pr_debug("%s: packet too long\n", skb->dev->name);
243 skb->dev->stats.rx_length_errors++;
246 page = virtqueue_get_buf(vi->rvq, &len);
248 pr_debug("%s: rx error: %d buffers missing\n",
249 skb->dev->name, hdr->mhdr.num_buffers);
250 skb->dev->stats.rx_length_errors++;
257 set_skb_frag(skb, page, 0, &len);
264 static void receive_buf(struct net_device *dev, void *buf, unsigned int len)
266 struct virtnet_info *vi = netdev_priv(dev);
267 struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
270 struct skb_vnet_hdr *hdr;
272 if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
273 pr_debug("%s: short packet %i\n", dev->name, len);
274 dev->stats.rx_length_errors++;
275 if (vi->mergeable_rx_bufs || vi->big_packets)
282 if (!vi->mergeable_rx_bufs && !vi->big_packets) {
284 len -= sizeof(struct virtio_net_hdr);
288 skb = page_to_skb(vi, page, len);
289 if (unlikely(!skb)) {
290 dev->stats.rx_dropped++;
291 give_pages(vi, page);
294 if (vi->mergeable_rx_bufs)
295 if (receive_mergeable(vi, skb)) {
301 hdr = skb_vnet_hdr(skb);
303 u64_stats_update_begin(&stats->rx_syncp);
304 stats->rx_bytes += skb->len;
306 u64_stats_update_end(&stats->rx_syncp);
308 if (hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
309 pr_debug("Needs csum!\n");
310 if (!skb_partial_csum_set(skb,
312 hdr->hdr.csum_offset))
314 } else if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) {
315 skb->ip_summed = CHECKSUM_UNNECESSARY;
318 skb->protocol = eth_type_trans(skb, dev);
319 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
320 ntohs(skb->protocol), skb->len, skb->pkt_type);
322 if (hdr->hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
324 switch (hdr->hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
325 case VIRTIO_NET_HDR_GSO_TCPV4:
326 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
328 case VIRTIO_NET_HDR_GSO_UDP:
329 skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
331 case VIRTIO_NET_HDR_GSO_TCPV6:
332 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
336 printk(KERN_WARNING "%s: bad gso type %u.\n",
337 dev->name, hdr->hdr.gso_type);
341 if (hdr->hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
342 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
344 skb_shinfo(skb)->gso_size = hdr->hdr.gso_size;
345 if (skb_shinfo(skb)->gso_size == 0) {
347 printk(KERN_WARNING "%s: zero gso size.\n",
352 /* Header must be checked, and gso_segs computed. */
353 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
354 skb_shinfo(skb)->gso_segs = 0;
357 netif_receive_skb(skb);
361 dev->stats.rx_frame_errors++;
365 static int add_recvbuf_small(struct virtnet_info *vi, gfp_t gfp)
368 struct skb_vnet_hdr *hdr;
371 skb = __netdev_alloc_skb_ip_align(vi->dev, MAX_PACKET_LEN, gfp);
375 skb_put(skb, MAX_PACKET_LEN);
377 hdr = skb_vnet_hdr(skb);
378 sg_set_buf(vi->rx_sg, &hdr->hdr, sizeof hdr->hdr);
380 skb_to_sgvec(skb, vi->rx_sg + 1, 0, skb->len);
382 err = virtqueue_add_buf(vi->rvq, vi->rx_sg, 0, 2, skb, gfp);
389 static int add_recvbuf_big(struct virtnet_info *vi, gfp_t gfp)
391 struct page *first, *list = NULL;
395 /* page in vi->rx_sg[MAX_SKB_FRAGS + 1] is list tail */
396 for (i = MAX_SKB_FRAGS + 1; i > 1; --i) {
397 first = get_a_page(vi, gfp);
400 give_pages(vi, list);
403 sg_set_buf(&vi->rx_sg[i], page_address(first), PAGE_SIZE);
405 /* chain new page in list head to match sg */
406 first->private = (unsigned long)list;
410 first = get_a_page(vi, gfp);
412 give_pages(vi, list);
415 p = page_address(first);
417 /* vi->rx_sg[0], vi->rx_sg[1] share the same page */
418 /* a separated vi->rx_sg[0] for virtio_net_hdr only due to QEMU bug */
419 sg_set_buf(&vi->rx_sg[0], p, sizeof(struct virtio_net_hdr));
421 /* vi->rx_sg[1] for data packet, from offset */
422 offset = sizeof(struct padded_vnet_hdr);
423 sg_set_buf(&vi->rx_sg[1], p + offset, PAGE_SIZE - offset);
425 /* chain first in list head */
426 first->private = (unsigned long)list;
427 err = virtqueue_add_buf(vi->rvq, vi->rx_sg, 0, MAX_SKB_FRAGS + 2,
430 give_pages(vi, first);
435 static int add_recvbuf_mergeable(struct virtnet_info *vi, gfp_t gfp)
440 page = get_a_page(vi, gfp);
444 sg_init_one(vi->rx_sg, page_address(page), PAGE_SIZE);
446 err = virtqueue_add_buf(vi->rvq, vi->rx_sg, 0, 1, page, gfp);
448 give_pages(vi, page);
454 * Returns false if we couldn't fill entirely (OOM).
456 * Normally run in the receive path, but can also be run from ndo_open
457 * before we're receiving packets, or from refill_work which is
458 * careful to disable receiving (using napi_disable).
460 static bool try_fill_recv(struct virtnet_info *vi, gfp_t gfp)
466 if (vi->mergeable_rx_bufs)
467 err = add_recvbuf_mergeable(vi, gfp);
468 else if (vi->big_packets)
469 err = add_recvbuf_big(vi, gfp);
471 err = add_recvbuf_small(vi, gfp);
473 oom = err == -ENOMEM;
477 } while (vi->rvq->num_free);
479 if (unlikely(vi->num > vi->max))
481 virtqueue_kick(vi->rvq);
485 static void skb_recv_done(struct virtqueue *rvq)
487 struct virtnet_info *vi = rvq->vdev->priv;
488 /* Schedule NAPI, Suppress further interrupts if successful. */
489 if (napi_schedule_prep(&vi->napi)) {
490 virtqueue_disable_cb(rvq);
491 __napi_schedule(&vi->napi);
495 static void virtnet_napi_enable(struct virtnet_info *vi)
497 napi_enable(&vi->napi);
499 /* If all buffers were filled by other side before we napi_enabled, we
500 * won't get another interrupt, so process any outstanding packets
501 * now. virtnet_poll wants re-enable the queue, so we disable here.
502 * We synchronize against interrupts via NAPI_STATE_SCHED */
503 if (napi_schedule_prep(&vi->napi)) {
504 virtqueue_disable_cb(vi->rvq);
506 __napi_schedule(&vi->napi);
511 static void refill_work(struct work_struct *work)
513 struct virtnet_info *vi;
516 vi = container_of(work, struct virtnet_info, refill.work);
517 napi_disable(&vi->napi);
518 still_empty = !try_fill_recv(vi, GFP_KERNEL);
519 virtnet_napi_enable(vi);
521 /* In theory, this can happen: if we don't get any buffers in
522 * we will *never* try to fill again. */
524 schedule_delayed_work(&vi->refill, HZ/2);
527 static int virtnet_poll(struct napi_struct *napi, int budget)
529 struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi);
531 unsigned int len, received = 0;
534 while (received < budget &&
535 (buf = virtqueue_get_buf(vi->rvq, &len)) != NULL) {
536 receive_buf(vi->dev, buf, len);
541 if (vi->num < vi->max / 2) {
542 if (!try_fill_recv(vi, GFP_ATOMIC))
543 schedule_delayed_work(&vi->refill, 0);
546 /* Out of packets? */
547 if (received < budget) {
549 if (unlikely(!virtqueue_enable_cb(vi->rvq)) &&
550 napi_schedule_prep(napi)) {
551 virtqueue_disable_cb(vi->rvq);
552 __napi_schedule(napi);
560 static void free_old_xmit_skbs(struct virtnet_info *vi)
564 struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
566 while ((skb = virtqueue_get_buf(vi->svq, &len)) != NULL) {
567 pr_debug("Sent skb %p\n", skb);
569 u64_stats_update_begin(&stats->tx_syncp);
570 stats->tx_bytes += skb->len;
572 u64_stats_update_end(&stats->tx_syncp);
574 dev_kfree_skb_any(skb);
578 static int xmit_skb(struct virtnet_info *vi, struct sk_buff *skb)
580 struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
581 const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
584 pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
586 if (skb->ip_summed == CHECKSUM_PARTIAL) {
587 hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
588 hdr->hdr.csum_start = skb_checksum_start_offset(skb);
589 hdr->hdr.csum_offset = skb->csum_offset;
592 hdr->hdr.csum_offset = hdr->hdr.csum_start = 0;
595 if (skb_is_gso(skb)) {
596 hdr->hdr.hdr_len = skb_headlen(skb);
597 hdr->hdr.gso_size = skb_shinfo(skb)->gso_size;
598 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
599 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
600 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
601 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
602 else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
603 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
606 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
607 hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
609 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
610 hdr->hdr.gso_size = hdr->hdr.hdr_len = 0;
613 hdr->mhdr.num_buffers = 0;
615 /* Encode metadata header at front. */
616 if (vi->mergeable_rx_bufs)
617 sg_set_buf(vi->tx_sg, &hdr->mhdr, sizeof hdr->mhdr);
619 sg_set_buf(vi->tx_sg, &hdr->hdr, sizeof hdr->hdr);
621 num_sg = skb_to_sgvec(skb, vi->tx_sg + 1, 0, skb->len) + 1;
622 return virtqueue_add_buf(vi->svq, vi->tx_sg, num_sg,
626 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
628 struct virtnet_info *vi = netdev_priv(dev);
631 /* Free up any pending old buffers before queueing new ones. */
632 free_old_xmit_skbs(vi);
634 /* Try to transmit */
635 err = xmit_skb(vi, skb);
637 /* This should not happen! */
639 dev->stats.tx_fifo_errors++;
642 "Unexpected TX queue failure: %d\n", err);
643 dev->stats.tx_dropped++;
647 virtqueue_kick(vi->svq);
649 /* Don't wait up for transmitted skbs to be freed. */
653 /* Apparently nice girls don't return TX_BUSY; stop the queue
654 * before it gets out of hand. Naturally, this wastes entries. */
655 if (vi->svq->num_free < 2+MAX_SKB_FRAGS) {
656 netif_stop_queue(dev);
657 if (unlikely(!virtqueue_enable_cb_delayed(vi->svq))) {
658 /* More just got used, free them then recheck. */
659 free_old_xmit_skbs(vi);
660 if (vi->svq->num_free >= 2+MAX_SKB_FRAGS) {
661 netif_start_queue(dev);
662 virtqueue_disable_cb(vi->svq);
670 static int virtnet_set_mac_address(struct net_device *dev, void *p)
672 struct virtnet_info *vi = netdev_priv(dev);
673 struct virtio_device *vdev = vi->vdev;
676 ret = eth_mac_addr(dev, p);
680 if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
681 vdev->config->set(vdev, offsetof(struct virtio_net_config, mac),
682 dev->dev_addr, dev->addr_len);
687 static struct rtnl_link_stats64 *virtnet_stats(struct net_device *dev,
688 struct rtnl_link_stats64 *tot)
690 struct virtnet_info *vi = netdev_priv(dev);
694 for_each_possible_cpu(cpu) {
695 struct virtnet_stats *stats = per_cpu_ptr(vi->stats, cpu);
696 u64 tpackets, tbytes, rpackets, rbytes;
699 start = u64_stats_fetch_begin_bh(&stats->tx_syncp);
700 tpackets = stats->tx_packets;
701 tbytes = stats->tx_bytes;
702 } while (u64_stats_fetch_retry_bh(&stats->tx_syncp, start));
705 start = u64_stats_fetch_begin_bh(&stats->rx_syncp);
706 rpackets = stats->rx_packets;
707 rbytes = stats->rx_bytes;
708 } while (u64_stats_fetch_retry_bh(&stats->rx_syncp, start));
710 tot->rx_packets += rpackets;
711 tot->tx_packets += tpackets;
712 tot->rx_bytes += rbytes;
713 tot->tx_bytes += tbytes;
716 tot->tx_dropped = dev->stats.tx_dropped;
717 tot->tx_fifo_errors = dev->stats.tx_fifo_errors;
718 tot->rx_dropped = dev->stats.rx_dropped;
719 tot->rx_length_errors = dev->stats.rx_length_errors;
720 tot->rx_frame_errors = dev->stats.rx_frame_errors;
725 #ifdef CONFIG_NET_POLL_CONTROLLER
726 static void virtnet_netpoll(struct net_device *dev)
728 struct virtnet_info *vi = netdev_priv(dev);
730 napi_schedule(&vi->napi);
734 static int virtnet_open(struct net_device *dev)
736 struct virtnet_info *vi = netdev_priv(dev);
738 /* Make sure we have some buffers: if oom use wq. */
739 if (!try_fill_recv(vi, GFP_KERNEL))
740 schedule_delayed_work(&vi->refill, 0);
742 virtnet_napi_enable(vi);
747 * Send command via the control virtqueue and check status. Commands
748 * supported by the hypervisor, as indicated by feature bits, should
749 * never fail unless improperly formated.
751 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
752 struct scatterlist *data, int out, int in)
754 struct scatterlist *s, sg[VIRTNET_SEND_COMMAND_SG_MAX + 2];
755 struct virtio_net_ctrl_hdr ctrl;
756 virtio_net_ctrl_ack status = ~0;
760 /* Caller should know better */
761 BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ||
762 (out + in > VIRTNET_SEND_COMMAND_SG_MAX));
764 out++; /* Add header */
765 in++; /* Add return status */
770 sg_init_table(sg, out + in);
772 sg_set_buf(&sg[0], &ctrl, sizeof(ctrl));
773 for_each_sg(data, s, out + in - 2, i)
774 sg_set_buf(&sg[i + 1], sg_virt(s), s->length);
775 sg_set_buf(&sg[out + in - 1], &status, sizeof(status));
777 BUG_ON(virtqueue_add_buf(vi->cvq, sg, out, in, vi, GFP_ATOMIC) < 0);
779 virtqueue_kick(vi->cvq);
782 * Spin for a response, the kick causes an ioport write, trapping
783 * into the hypervisor, so the request should be handled immediately.
785 while (!virtqueue_get_buf(vi->cvq, &tmp))
788 return status == VIRTIO_NET_OK;
791 static void virtnet_ack_link_announce(struct virtnet_info *vi)
794 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE,
795 VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL,
797 dev_warn(&vi->dev->dev, "Failed to ack link announce.\n");
801 static int virtnet_close(struct net_device *dev)
803 struct virtnet_info *vi = netdev_priv(dev);
805 /* Make sure refill_work doesn't re-enable napi! */
806 cancel_delayed_work_sync(&vi->refill);
807 napi_disable(&vi->napi);
812 static void virtnet_set_rx_mode(struct net_device *dev)
814 struct virtnet_info *vi = netdev_priv(dev);
815 struct scatterlist sg[2];
816 u8 promisc, allmulti;
817 struct virtio_net_ctrl_mac *mac_data;
818 struct netdev_hw_addr *ha;
824 /* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */
825 if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
828 promisc = ((dev->flags & IFF_PROMISC) != 0);
829 allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
831 sg_init_one(sg, &promisc, sizeof(promisc));
833 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
834 VIRTIO_NET_CTRL_RX_PROMISC,
836 dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
837 promisc ? "en" : "dis");
839 sg_init_one(sg, &allmulti, sizeof(allmulti));
841 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
842 VIRTIO_NET_CTRL_RX_ALLMULTI,
844 dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
845 allmulti ? "en" : "dis");
847 uc_count = netdev_uc_count(dev);
848 mc_count = netdev_mc_count(dev);
849 /* MAC filter - use one buffer for both lists */
850 buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
851 (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
854 dev_warn(&dev->dev, "No memory for MAC address buffer\n");
858 sg_init_table(sg, 2);
860 /* Store the unicast list and count in the front of the buffer */
861 mac_data->entries = uc_count;
863 netdev_for_each_uc_addr(ha, dev)
864 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
866 sg_set_buf(&sg[0], mac_data,
867 sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
869 /* multicast list and count fill the end */
870 mac_data = (void *)&mac_data->macs[uc_count][0];
872 mac_data->entries = mc_count;
874 netdev_for_each_mc_addr(ha, dev)
875 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
877 sg_set_buf(&sg[1], mac_data,
878 sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
880 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
881 VIRTIO_NET_CTRL_MAC_TABLE_SET,
883 dev_warn(&dev->dev, "Failed to set MAC fitler table.\n");
888 static int virtnet_vlan_rx_add_vid(struct net_device *dev, u16 vid)
890 struct virtnet_info *vi = netdev_priv(dev);
891 struct scatterlist sg;
893 sg_init_one(&sg, &vid, sizeof(vid));
895 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
896 VIRTIO_NET_CTRL_VLAN_ADD, &sg, 1, 0))
897 dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
901 static int virtnet_vlan_rx_kill_vid(struct net_device *dev, u16 vid)
903 struct virtnet_info *vi = netdev_priv(dev);
904 struct scatterlist sg;
906 sg_init_one(&sg, &vid, sizeof(vid));
908 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
909 VIRTIO_NET_CTRL_VLAN_DEL, &sg, 1, 0))
910 dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
914 static void virtnet_get_ringparam(struct net_device *dev,
915 struct ethtool_ringparam *ring)
917 struct virtnet_info *vi = netdev_priv(dev);
919 ring->rx_max_pending = virtqueue_get_vring_size(vi->rvq);
920 ring->tx_max_pending = virtqueue_get_vring_size(vi->svq);
921 ring->rx_pending = ring->rx_max_pending;
922 ring->tx_pending = ring->tx_max_pending;
927 static void virtnet_get_drvinfo(struct net_device *dev,
928 struct ethtool_drvinfo *info)
930 struct virtnet_info *vi = netdev_priv(dev);
931 struct virtio_device *vdev = vi->vdev;
933 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
934 strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version));
935 strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info));
939 static const struct ethtool_ops virtnet_ethtool_ops = {
940 .get_drvinfo = virtnet_get_drvinfo,
941 .get_link = ethtool_op_get_link,
942 .get_ringparam = virtnet_get_ringparam,
946 #define MAX_MTU 65535
948 static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
950 if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
956 static const struct net_device_ops virtnet_netdev = {
957 .ndo_open = virtnet_open,
958 .ndo_stop = virtnet_close,
959 .ndo_start_xmit = start_xmit,
960 .ndo_validate_addr = eth_validate_addr,
961 .ndo_set_mac_address = virtnet_set_mac_address,
962 .ndo_set_rx_mode = virtnet_set_rx_mode,
963 .ndo_change_mtu = virtnet_change_mtu,
964 .ndo_get_stats64 = virtnet_stats,
965 .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
966 .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
967 #ifdef CONFIG_NET_POLL_CONTROLLER
968 .ndo_poll_controller = virtnet_netpoll,
972 static void virtnet_config_changed_work(struct work_struct *work)
974 struct virtnet_info *vi =
975 container_of(work, struct virtnet_info, config_work);
978 mutex_lock(&vi->config_lock);
979 if (!vi->config_enable)
982 if (virtio_config_val(vi->vdev, VIRTIO_NET_F_STATUS,
983 offsetof(struct virtio_net_config, status),
987 if (v & VIRTIO_NET_S_ANNOUNCE) {
988 netdev_notify_peers(vi->dev);
989 virtnet_ack_link_announce(vi);
992 /* Ignore unknown (future) status bits */
993 v &= VIRTIO_NET_S_LINK_UP;
1000 if (vi->status & VIRTIO_NET_S_LINK_UP) {
1001 netif_carrier_on(vi->dev);
1002 netif_wake_queue(vi->dev);
1004 netif_carrier_off(vi->dev);
1005 netif_stop_queue(vi->dev);
1008 mutex_unlock(&vi->config_lock);
1011 static void virtnet_config_changed(struct virtio_device *vdev)
1013 struct virtnet_info *vi = vdev->priv;
1015 schedule_work(&vi->config_work);
1018 static int init_vqs(struct virtnet_info *vi)
1020 struct virtqueue *vqs[3];
1021 vq_callback_t *callbacks[] = { skb_recv_done, skb_xmit_done, NULL};
1022 const char *names[] = { "input", "output", "control" };
1025 /* We expect two virtqueues, receive then send,
1026 * and optionally control. */
1027 nvqs = virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ? 3 : 2;
1029 err = vi->vdev->config->find_vqs(vi->vdev, nvqs, vqs, callbacks, names);
1036 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)) {
1039 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
1040 vi->dev->features |= NETIF_F_HW_VLAN_FILTER;
1045 static int virtnet_probe(struct virtio_device *vdev)
1048 struct net_device *dev;
1049 struct virtnet_info *vi;
1051 /* Allocate ourselves a network device with room for our info */
1052 dev = alloc_etherdev(sizeof(struct virtnet_info));
1056 /* Set up network device as normal. */
1057 dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE;
1058 dev->netdev_ops = &virtnet_netdev;
1059 dev->features = NETIF_F_HIGHDMA;
1061 SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
1062 SET_NETDEV_DEV(dev, &vdev->dev);
1064 /* Do we support "hardware" checksums? */
1065 if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
1066 /* This opens up the world of extra features. */
1067 dev->hw_features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1069 dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1071 if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
1072 dev->hw_features |= NETIF_F_TSO | NETIF_F_UFO
1073 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
1075 /* Individual feature bits: what can host handle? */
1076 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
1077 dev->hw_features |= NETIF_F_TSO;
1078 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
1079 dev->hw_features |= NETIF_F_TSO6;
1080 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
1081 dev->hw_features |= NETIF_F_TSO_ECN;
1082 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
1083 dev->hw_features |= NETIF_F_UFO;
1086 dev->features |= dev->hw_features & (NETIF_F_ALL_TSO|NETIF_F_UFO);
1087 /* (!csum && gso) case will be fixed by register_netdev() */
1090 /* Configuration may specify what MAC to use. Otherwise random. */
1091 if (virtio_config_val_len(vdev, VIRTIO_NET_F_MAC,
1092 offsetof(struct virtio_net_config, mac),
1093 dev->dev_addr, dev->addr_len) < 0)
1094 eth_hw_addr_random(dev);
1096 /* Set up our device-specific information */
1097 vi = netdev_priv(dev);
1098 netif_napi_add(dev, &vi->napi, virtnet_poll, napi_weight);
1103 vi->stats = alloc_percpu(struct virtnet_stats);
1105 if (vi->stats == NULL)
1108 INIT_DELAYED_WORK(&vi->refill, refill_work);
1109 mutex_init(&vi->config_lock);
1110 vi->config_enable = true;
1111 INIT_WORK(&vi->config_work, virtnet_config_changed_work);
1112 sg_init_table(vi->rx_sg, ARRAY_SIZE(vi->rx_sg));
1113 sg_init_table(vi->tx_sg, ARRAY_SIZE(vi->tx_sg));
1115 /* If we can receive ANY GSO packets, we must allocate large ones. */
1116 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
1117 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
1118 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
1119 vi->big_packets = true;
1121 if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
1122 vi->mergeable_rx_bufs = true;
1128 err = register_netdev(dev);
1130 pr_debug("virtio_net: registering device failed\n");
1134 /* Last of all, set up some receive buffers. */
1135 try_fill_recv(vi, GFP_KERNEL);
1137 /* If we didn't even get one input buffer, we're useless. */
1143 /* Assume link up if device can't report link status,
1144 otherwise get link status from config. */
1145 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
1146 netif_carrier_off(dev);
1147 schedule_work(&vi->config_work);
1149 vi->status = VIRTIO_NET_S_LINK_UP;
1150 netif_carrier_on(dev);
1153 pr_debug("virtnet: registered device %s\n", dev->name);
1157 unregister_netdev(dev);
1159 vdev->config->del_vqs(vdev);
1161 free_percpu(vi->stats);
1167 static void free_unused_bufs(struct virtnet_info *vi)
1171 buf = virtqueue_detach_unused_buf(vi->svq);
1177 buf = virtqueue_detach_unused_buf(vi->rvq);
1180 if (vi->mergeable_rx_bufs || vi->big_packets)
1181 give_pages(vi, buf);
1186 BUG_ON(vi->num != 0);
1189 static void remove_vq_common(struct virtnet_info *vi)
1191 vi->vdev->config->reset(vi->vdev);
1193 /* Free unused buffers in both send and recv, if any. */
1194 free_unused_bufs(vi);
1196 vi->vdev->config->del_vqs(vi->vdev);
1199 __free_pages(get_a_page(vi, GFP_KERNEL), 0);
1202 static void __devexit virtnet_remove(struct virtio_device *vdev)
1204 struct virtnet_info *vi = vdev->priv;
1206 /* Prevent config work handler from accessing the device. */
1207 mutex_lock(&vi->config_lock);
1208 vi->config_enable = false;
1209 mutex_unlock(&vi->config_lock);
1211 unregister_netdev(vi->dev);
1213 remove_vq_common(vi);
1215 flush_work(&vi->config_work);
1217 free_percpu(vi->stats);
1218 free_netdev(vi->dev);
1222 static int virtnet_freeze(struct virtio_device *vdev)
1224 struct virtnet_info *vi = vdev->priv;
1226 /* Prevent config work handler from accessing the device */
1227 mutex_lock(&vi->config_lock);
1228 vi->config_enable = false;
1229 mutex_unlock(&vi->config_lock);
1231 netif_device_detach(vi->dev);
1232 cancel_delayed_work_sync(&vi->refill);
1234 if (netif_running(vi->dev))
1235 napi_disable(&vi->napi);
1237 remove_vq_common(vi);
1239 flush_work(&vi->config_work);
1244 static int virtnet_restore(struct virtio_device *vdev)
1246 struct virtnet_info *vi = vdev->priv;
1253 if (netif_running(vi->dev))
1254 virtnet_napi_enable(vi);
1256 netif_device_attach(vi->dev);
1258 if (!try_fill_recv(vi, GFP_KERNEL))
1259 schedule_delayed_work(&vi->refill, 0);
1261 mutex_lock(&vi->config_lock);
1262 vi->config_enable = true;
1263 mutex_unlock(&vi->config_lock);
1269 static struct virtio_device_id id_table[] = {
1270 { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
1274 static unsigned int features[] = {
1275 VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
1276 VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
1277 VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
1278 VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
1279 VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO,
1280 VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ,
1281 VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN,
1282 VIRTIO_NET_F_GUEST_ANNOUNCE,
1285 static struct virtio_driver virtio_net_driver = {
1286 .feature_table = features,
1287 .feature_table_size = ARRAY_SIZE(features),
1288 .driver.name = KBUILD_MODNAME,
1289 .driver.owner = THIS_MODULE,
1290 .id_table = id_table,
1291 .probe = virtnet_probe,
1292 .remove = __devexit_p(virtnet_remove),
1293 .config_changed = virtnet_config_changed,
1295 .freeze = virtnet_freeze,
1296 .restore = virtnet_restore,
1300 static int __init init(void)
1302 return register_virtio_driver(&virtio_net_driver);
1305 static void __exit fini(void)
1307 unregister_virtio_driver(&virtio_net_driver);
1312 MODULE_DEVICE_TABLE(virtio, id_table);
1313 MODULE_DESCRIPTION("Virtio network driver");
1314 MODULE_LICENSE("GPL");