2 * Copyright (c) 2009, Microsoft Corporation.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * You should have received a copy of the GNU General Public License along with
14 * this program; if not, see <http://www.gnu.org/licenses/>.
17 * Haiyang Zhang <haiyangz@microsoft.com>
18 * Hank Janssen <hjanssen@microsoft.com>
20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
22 #include <linux/init.h>
23 #include <linux/atomic.h>
24 #include <linux/module.h>
25 #include <linux/highmem.h>
26 #include <linux/device.h>
28 #include <linux/delay.h>
29 #include <linux/netdevice.h>
30 #include <linux/inetdevice.h>
31 #include <linux/etherdevice.h>
32 #include <linux/skbuff.h>
33 #include <linux/if_vlan.h>
35 #include <linux/slab.h>
37 #include <net/route.h>
39 #include <net/pkt_sched.h>
40 #include <net/checksum.h>
41 #include <net/ip6_checksum.h>
43 #include "hyperv_net.h"
45 #define RING_SIZE_MIN 64
46 #define LINKCHANGE_INT (2 * HZ)
48 static int ring_size = 128;
49 module_param(ring_size, int, S_IRUGO);
50 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
52 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
53 NETIF_MSG_LINK | NETIF_MSG_IFUP |
54 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
57 static int debug = -1;
58 module_param(debug, int, S_IRUGO);
59 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
61 static void netvsc_set_multicast_list(struct net_device *net)
63 struct net_device_context *net_device_ctx = netdev_priv(net);
64 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
66 rndis_filter_update(nvdev);
69 static int netvsc_open(struct net_device *net)
71 struct net_device_context *ndev_ctx = netdev_priv(net);
72 struct netvsc_device *nvdev = rtnl_dereference(ndev_ctx->nvdev);
73 struct rndis_device *rdev;
76 netif_carrier_off(net);
78 /* Open up the device */
79 ret = rndis_filter_open(nvdev);
81 netdev_err(net, "unable to open device (ret %d).\n", ret);
85 netif_tx_wake_all_queues(net);
87 rdev = nvdev->extension;
88 if (!rdev->link_state && !ndev_ctx->datapath)
89 netif_carrier_on(net);
94 static int netvsc_close(struct net_device *net)
96 struct net_device_context *net_device_ctx = netdev_priv(net);
97 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
99 u32 aread, i, msec = 10, retry = 0, retry_max = 20;
100 struct vmbus_channel *chn;
102 netif_tx_disable(net);
104 ret = rndis_filter_close(nvdev);
106 netdev_err(net, "unable to close device (ret %d).\n", ret);
110 /* Ensure pending bytes in ring are read */
113 for (i = 0; i < nvdev->num_chn; i++) {
114 chn = nvdev->chan_table[i].channel;
118 aread = hv_get_bytes_to_read(&chn->inbound);
122 aread = hv_get_bytes_to_read(&chn->outbound);
128 if (retry > retry_max || aread == 0)
138 netdev_err(net, "Ring buffer not empty after closing rndis\n");
145 static void *init_ppi_data(struct rndis_message *msg, u32 ppi_size,
148 struct rndis_packet *rndis_pkt;
149 struct rndis_per_packet_info *ppi;
151 rndis_pkt = &msg->msg.pkt;
152 rndis_pkt->data_offset += ppi_size;
154 ppi = (struct rndis_per_packet_info *)((void *)rndis_pkt +
155 rndis_pkt->per_pkt_info_offset + rndis_pkt->per_pkt_info_len);
157 ppi->size = ppi_size;
158 ppi->type = pkt_type;
159 ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
161 rndis_pkt->per_pkt_info_len += ppi_size;
166 /* Azure hosts don't support non-TCP port numbers in hashing yet. We compute
167 * hash for non-TCP traffic with only IP numbers.
169 static inline u32 netvsc_get_hash(struct sk_buff *skb, struct sock *sk)
171 struct flow_keys flow;
173 static u32 hashrnd __read_mostly;
175 net_get_random_once(&hashrnd, sizeof(hashrnd));
177 if (!skb_flow_dissect_flow_keys(skb, &flow, 0))
180 if (flow.basic.ip_proto == IPPROTO_TCP) {
181 return skb_get_hash(skb);
183 if (flow.basic.n_proto == htons(ETH_P_IP))
184 hash = jhash2((u32 *)&flow.addrs.v4addrs, 2, hashrnd);
185 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
186 hash = jhash2((u32 *)&flow.addrs.v6addrs, 8, hashrnd);
190 skb_set_hash(skb, hash, PKT_HASH_TYPE_L3);
196 static inline int netvsc_get_tx_queue(struct net_device *ndev,
197 struct sk_buff *skb, int old_idx)
199 const struct net_device_context *ndc = netdev_priv(ndev);
200 struct sock *sk = skb->sk;
203 q_idx = ndc->tx_send_table[netvsc_get_hash(skb, sk) &
204 (VRSS_SEND_TAB_SIZE - 1)];
206 /* If queue index changed record the new value */
207 if (q_idx != old_idx &&
208 sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
209 sk_tx_queue_set(sk, q_idx);
215 * Select queue for transmit.
217 * If a valid queue has already been assigned, then use that.
218 * Otherwise compute tx queue based on hash and the send table.
220 * This is basically similar to default (__netdev_pick_tx) with the added step
221 * of using the host send_table when no other queue has been assigned.
223 * TODO support XPS - but get_xps_queue not exported
225 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
226 void *accel_priv, select_queue_fallback_t fallback)
228 unsigned int num_tx_queues = ndev->real_num_tx_queues;
229 int q_idx = sk_tx_queue_get(skb->sk);
231 if (q_idx < 0 || skb->ooo_okay) {
232 /* If forwarding a packet, we use the recorded queue when
233 * available for better cache locality.
235 if (skb_rx_queue_recorded(skb))
236 q_idx = skb_get_rx_queue(skb);
238 q_idx = netvsc_get_tx_queue(ndev, skb, q_idx);
241 while (unlikely(q_idx >= num_tx_queues))
242 q_idx -= num_tx_queues;
247 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
248 struct hv_page_buffer *pb)
252 /* Deal with compund pages by ignoring unused part
255 page += (offset >> PAGE_SHIFT);
256 offset &= ~PAGE_MASK;
261 bytes = PAGE_SIZE - offset;
264 pb[j].pfn = page_to_pfn(page);
265 pb[j].offset = offset;
271 if (offset == PAGE_SIZE && len) {
281 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
282 struct hv_netvsc_packet *packet,
283 struct hv_page_buffer **page_buf)
285 struct hv_page_buffer *pb = *page_buf;
287 char *data = skb->data;
288 int frags = skb_shinfo(skb)->nr_frags;
291 /* The packet is laid out thus:
292 * 1. hdr: RNDIS header and PPI
294 * 3. skb fragment data
297 slots_used += fill_pg_buf(virt_to_page(hdr),
299 len, &pb[slots_used]);
301 packet->rmsg_size = len;
302 packet->rmsg_pgcnt = slots_used;
304 slots_used += fill_pg_buf(virt_to_page(data),
305 offset_in_page(data),
306 skb_headlen(skb), &pb[slots_used]);
308 for (i = 0; i < frags; i++) {
309 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
311 slots_used += fill_pg_buf(skb_frag_page(frag),
313 skb_frag_size(frag), &pb[slots_used]);
318 /* Estimate number of page buffers neede to transmit
319 * Need at most 2 for RNDIS header plus skb body and fragments.
321 static unsigned int netvsc_get_slots(const struct sk_buff *skb)
323 return PFN_UP(offset_in_page(skb->data) + skb_headlen(skb))
324 + skb_shinfo(skb)->nr_frags
328 static u32 net_checksum_info(struct sk_buff *skb)
330 if (skb->protocol == htons(ETH_P_IP)) {
331 struct iphdr *ip = ip_hdr(skb);
333 if (ip->protocol == IPPROTO_TCP)
334 return TRANSPORT_INFO_IPV4_TCP;
335 else if (ip->protocol == IPPROTO_UDP)
336 return TRANSPORT_INFO_IPV4_UDP;
338 struct ipv6hdr *ip6 = ipv6_hdr(skb);
340 if (ip6->nexthdr == IPPROTO_TCP)
341 return TRANSPORT_INFO_IPV6_TCP;
342 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
343 return TRANSPORT_INFO_IPV6_UDP;
346 return TRANSPORT_INFO_NOT_IP;
349 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
351 struct net_device_context *net_device_ctx = netdev_priv(net);
352 struct hv_netvsc_packet *packet = NULL;
354 unsigned int num_data_pgs;
355 struct rndis_message *rndis_msg;
356 struct rndis_packet *rndis_pkt;
358 struct rndis_per_packet_info *ppi;
360 struct hv_page_buffer page_buf[MAX_PAGE_BUFFER_COUNT];
361 struct hv_page_buffer *pb = page_buf;
363 /* We can only transmit MAX_PAGE_BUFFER_COUNT number
364 * of pages in a single packet. If skb is scattered around
365 * more pages we try linearizing it.
367 num_data_pgs = netvsc_get_slots(skb);
368 if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
369 ++net_device_ctx->eth_stats.tx_scattered;
371 if (skb_linearize(skb))
374 num_data_pgs = netvsc_get_slots(skb);
375 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
376 ++net_device_ctx->eth_stats.tx_too_big;
382 * Place the rndis header in the skb head room and
383 * the skb->cb will be used for hv_netvsc_packet
386 ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
390 /* Use the skb control buffer for building up the packet */
391 BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
392 FIELD_SIZEOF(struct sk_buff, cb));
393 packet = (struct hv_netvsc_packet *)skb->cb;
395 packet->q_idx = skb_get_queue_mapping(skb);
397 packet->total_data_buflen = skb->len;
398 packet->total_bytes = skb->len;
399 packet->total_packets = 1;
401 rndis_msg = (struct rndis_message *)skb->head;
403 memset(rndis_msg, 0, RNDIS_AND_PPI_SIZE);
405 /* Add the rndis header */
406 rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
407 rndis_msg->msg_len = packet->total_data_buflen;
408 rndis_pkt = &rndis_msg->msg.pkt;
409 rndis_pkt->data_offset = sizeof(struct rndis_packet);
410 rndis_pkt->data_len = packet->total_data_buflen;
411 rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
413 rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
415 hash = skb_get_hash_raw(skb);
416 if (hash != 0 && net->real_num_tx_queues > 1) {
417 rndis_msg_size += NDIS_HASH_PPI_SIZE;
418 ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
420 *(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
423 if (skb_vlan_tag_present(skb)) {
424 struct ndis_pkt_8021q_info *vlan;
426 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
427 ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
429 vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
431 vlan->vlanid = skb->vlan_tci & VLAN_VID_MASK;
432 vlan->pri = (skb->vlan_tci & VLAN_PRIO_MASK) >>
436 if (skb_is_gso(skb)) {
437 struct ndis_tcp_lso_info *lso_info;
439 rndis_msg_size += NDIS_LSO_PPI_SIZE;
440 ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
441 TCP_LARGESEND_PKTINFO);
443 lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
446 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
447 if (skb->protocol == htons(ETH_P_IP)) {
448 lso_info->lso_v2_transmit.ip_version =
449 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
450 ip_hdr(skb)->tot_len = 0;
451 ip_hdr(skb)->check = 0;
452 tcp_hdr(skb)->check =
453 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
454 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
456 lso_info->lso_v2_transmit.ip_version =
457 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
458 ipv6_hdr(skb)->payload_len = 0;
459 tcp_hdr(skb)->check =
460 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
461 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
463 lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
464 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
465 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
466 if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
467 struct ndis_tcp_ip_checksum_info *csum_info;
469 rndis_msg_size += NDIS_CSUM_PPI_SIZE;
470 ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
471 TCPIP_CHKSUM_PKTINFO);
473 csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
476 csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
478 if (skb->protocol == htons(ETH_P_IP)) {
479 csum_info->transmit.is_ipv4 = 1;
481 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
482 csum_info->transmit.tcp_checksum = 1;
484 csum_info->transmit.udp_checksum = 1;
486 csum_info->transmit.is_ipv6 = 1;
488 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
489 csum_info->transmit.tcp_checksum = 1;
491 csum_info->transmit.udp_checksum = 1;
494 /* Can't do offload of this type of checksum */
495 if (skb_checksum_help(skb))
500 /* Start filling in the page buffers with the rndis hdr */
501 rndis_msg->msg_len += rndis_msg_size;
502 packet->total_data_buflen = rndis_msg->msg_len;
503 packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
506 /* timestamp packet in software */
507 skb_tx_timestamp(skb);
509 ret = netvsc_send(net_device_ctx, packet, rndis_msg, &pb, skb);
510 if (likely(ret == 0))
513 if (ret == -EAGAIN) {
514 ++net_device_ctx->eth_stats.tx_busy;
515 return NETDEV_TX_BUSY;
519 ++net_device_ctx->eth_stats.tx_no_space;
522 dev_kfree_skb_any(skb);
523 net->stats.tx_dropped++;
528 ++net_device_ctx->eth_stats.tx_no_memory;
532 * netvsc_linkstatus_callback - Link up/down notification
534 void netvsc_linkstatus_callback(struct hv_device *device_obj,
535 struct rndis_message *resp)
537 struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
538 struct net_device *net;
539 struct net_device_context *ndev_ctx;
540 struct netvsc_reconfig *event;
543 net = hv_get_drvdata(device_obj);
548 ndev_ctx = netdev_priv(net);
550 /* Update the physical link speed when changing to another vSwitch */
551 if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
554 speed = *(u32 *)((void *)indicate + indicate->
555 status_buf_offset) / 10000;
556 ndev_ctx->speed = speed;
560 /* Handle these link change statuses below */
561 if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
562 indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
563 indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
566 if (net->reg_state != NETREG_REGISTERED)
569 event = kzalloc(sizeof(*event), GFP_ATOMIC);
572 event->event = indicate->status;
574 spin_lock_irqsave(&ndev_ctx->lock, flags);
575 list_add_tail(&event->list, &ndev_ctx->reconfig_events);
576 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
578 schedule_delayed_work(&ndev_ctx->dwork, 0);
581 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
582 struct napi_struct *napi,
583 const struct ndis_tcp_ip_checksum_info *csum_info,
584 const struct ndis_pkt_8021q_info *vlan,
585 void *data, u32 buflen)
589 skb = napi_alloc_skb(napi, buflen);
594 * Copy to skb. This copy is needed here since the memory pointed by
595 * hv_netvsc_packet cannot be deallocated
597 skb_put_data(skb, data, buflen);
599 skb->protocol = eth_type_trans(skb, net);
601 /* skb is already created with CHECKSUM_NONE */
602 skb_checksum_none_assert(skb);
605 * In Linux, the IP checksum is always checked.
606 * Do L4 checksum offload if enabled and present.
608 if (csum_info && (net->features & NETIF_F_RXCSUM)) {
609 if (csum_info->receive.tcp_checksum_succeeded ||
610 csum_info->receive.udp_checksum_succeeded)
611 skb->ip_summed = CHECKSUM_UNNECESSARY;
615 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT);
617 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
625 * netvsc_recv_callback - Callback when we receive a packet from the
626 * "wire" on the specified device.
628 int netvsc_recv_callback(struct net_device *net,
629 struct vmbus_channel *channel,
631 const struct ndis_tcp_ip_checksum_info *csum_info,
632 const struct ndis_pkt_8021q_info *vlan)
634 struct net_device_context *net_device_ctx = netdev_priv(net);
635 struct netvsc_device *net_device;
636 u16 q_idx = channel->offermsg.offer.sub_channel_index;
637 struct netvsc_channel *nvchan;
638 struct net_device *vf_netdev;
640 struct netvsc_stats *rx_stats;
642 if (net->reg_state != NETREG_REGISTERED)
643 return NVSP_STAT_FAIL;
646 * If necessary, inject this packet into the VF interface.
647 * On Hyper-V, multicast and brodcast packets are only delivered
648 * to the synthetic interface (after subjecting these to
649 * policy filters on the host). Deliver these via the VF
650 * interface in the guest.
653 net_device = rcu_dereference(net_device_ctx->nvdev);
654 if (unlikely(!net_device))
657 nvchan = &net_device->chan_table[q_idx];
658 vf_netdev = rcu_dereference(net_device_ctx->vf_netdev);
659 if (vf_netdev && (vf_netdev->flags & IFF_UP))
662 /* Allocate a skb - TODO direct I/O to pages? */
663 skb = netvsc_alloc_recv_skb(net, &nvchan->napi,
664 csum_info, vlan, data, len);
665 if (unlikely(!skb)) {
667 ++net->stats.rx_dropped;
669 return NVSP_STAT_FAIL;
672 if (net != vf_netdev)
673 skb_record_rx_queue(skb, q_idx);
676 * Even if injecting the packet, record the statistics
677 * on the synthetic device because modifying the VF device
678 * statistics will not work correctly.
680 rx_stats = &nvchan->rx_stats;
681 u64_stats_update_begin(&rx_stats->syncp);
683 rx_stats->bytes += len;
685 if (skb->pkt_type == PACKET_BROADCAST)
686 ++rx_stats->broadcast;
687 else if (skb->pkt_type == PACKET_MULTICAST)
688 ++rx_stats->multicast;
689 u64_stats_update_end(&rx_stats->syncp);
691 napi_gro_receive(&nvchan->napi, skb);
697 static void netvsc_get_drvinfo(struct net_device *net,
698 struct ethtool_drvinfo *info)
700 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
701 strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
704 static void netvsc_get_channels(struct net_device *net,
705 struct ethtool_channels *channel)
707 struct net_device_context *net_device_ctx = netdev_priv(net);
708 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
711 channel->max_combined = nvdev->max_chn;
712 channel->combined_count = nvdev->num_chn;
716 static int netvsc_set_queues(struct net_device *net, struct hv_device *dev,
719 struct netvsc_device_info device_info;
720 struct netvsc_device *net_device;
723 memset(&device_info, 0, sizeof(device_info));
724 device_info.num_chn = num_chn;
725 device_info.ring_size = ring_size;
726 device_info.max_num_vrss_chns = num_chn;
728 ret = netif_set_real_num_tx_queues(net, num_chn);
732 ret = netif_set_real_num_rx_queues(net, num_chn);
736 net_device = rndis_filter_device_add(dev, &device_info);
737 return PTR_ERR_OR_ZERO(net_device);
740 static int netvsc_set_channels(struct net_device *net,
741 struct ethtool_channels *channels)
743 struct net_device_context *net_device_ctx = netdev_priv(net);
744 struct hv_device *dev = net_device_ctx->device_ctx;
745 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
746 unsigned int count = channels->combined_count;
750 /* We do not support separate count for rx, tx, or other */
752 channels->rx_count || channels->tx_count || channels->other_count)
755 if (count > net->num_tx_queues || count > VRSS_CHANNEL_MAX)
758 if (!nvdev || nvdev->destroy)
761 if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
764 if (count > nvdev->max_chn)
767 was_opened = rndis_filter_opened(nvdev);
769 rndis_filter_close(nvdev);
771 rndis_filter_device_remove(dev, nvdev);
773 ret = netvsc_set_queues(net, dev, count);
775 nvdev->num_chn = count;
777 netvsc_set_queues(net, dev, nvdev->num_chn);
779 nvdev = rtnl_dereference(net_device_ctx->nvdev);
781 rndis_filter_open(nvdev);
783 /* We may have missed link change notifications */
784 net_device_ctx->last_reconfig = 0;
785 schedule_delayed_work(&net_device_ctx->dwork, 0);
791 netvsc_validate_ethtool_ss_cmd(const struct ethtool_link_ksettings *cmd)
793 struct ethtool_link_ksettings diff1 = *cmd;
794 struct ethtool_link_ksettings diff2 = {};
796 diff1.base.speed = 0;
797 diff1.base.duplex = 0;
798 /* advertising and cmd are usually set */
799 ethtool_link_ksettings_zero_link_mode(&diff1, advertising);
801 /* We set port to PORT_OTHER */
802 diff2.base.port = PORT_OTHER;
804 return !memcmp(&diff1, &diff2, sizeof(diff1));
807 static void netvsc_init_settings(struct net_device *dev)
809 struct net_device_context *ndc = netdev_priv(dev);
811 ndc->speed = SPEED_UNKNOWN;
812 ndc->duplex = DUPLEX_FULL;
815 static int netvsc_get_link_ksettings(struct net_device *dev,
816 struct ethtool_link_ksettings *cmd)
818 struct net_device_context *ndc = netdev_priv(dev);
820 cmd->base.speed = ndc->speed;
821 cmd->base.duplex = ndc->duplex;
822 cmd->base.port = PORT_OTHER;
827 static int netvsc_set_link_ksettings(struct net_device *dev,
828 const struct ethtool_link_ksettings *cmd)
830 struct net_device_context *ndc = netdev_priv(dev);
833 speed = cmd->base.speed;
834 if (!ethtool_validate_speed(speed) ||
835 !ethtool_validate_duplex(cmd->base.duplex) ||
836 !netvsc_validate_ethtool_ss_cmd(cmd))
840 ndc->duplex = cmd->base.duplex;
845 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
847 struct net_device_context *ndevctx = netdev_priv(ndev);
848 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
849 struct hv_device *hdev = ndevctx->device_ctx;
850 int orig_mtu = ndev->mtu;
851 struct netvsc_device_info device_info;
855 if (!nvdev || nvdev->destroy)
858 netif_device_detach(ndev);
859 was_opened = rndis_filter_opened(nvdev);
861 rndis_filter_close(nvdev);
863 memset(&device_info, 0, sizeof(device_info));
864 device_info.ring_size = ring_size;
865 device_info.num_chn = nvdev->num_chn;
866 device_info.max_num_vrss_chns = nvdev->num_chn;
868 rndis_filter_device_remove(hdev, nvdev);
872 nvdev = rndis_filter_device_add(hdev, &device_info);
874 ret = PTR_ERR(nvdev);
876 /* Attempt rollback to original MTU */
877 ndev->mtu = orig_mtu;
878 rndis_filter_device_add(hdev, &device_info);
882 rndis_filter_open(nvdev);
884 netif_device_attach(ndev);
886 /* We may have missed link change notifications */
887 schedule_delayed_work(&ndevctx->dwork, 0);
892 static void netvsc_get_stats64(struct net_device *net,
893 struct rtnl_link_stats64 *t)
895 struct net_device_context *ndev_ctx = netdev_priv(net);
896 struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
902 for (i = 0; i < nvdev->num_chn; i++) {
903 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
904 const struct netvsc_stats *stats;
905 u64 packets, bytes, multicast;
908 stats = &nvchan->tx_stats;
910 start = u64_stats_fetch_begin_irq(&stats->syncp);
911 packets = stats->packets;
912 bytes = stats->bytes;
913 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
915 t->tx_bytes += bytes;
916 t->tx_packets += packets;
918 stats = &nvchan->rx_stats;
920 start = u64_stats_fetch_begin_irq(&stats->syncp);
921 packets = stats->packets;
922 bytes = stats->bytes;
923 multicast = stats->multicast + stats->broadcast;
924 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
926 t->rx_bytes += bytes;
927 t->rx_packets += packets;
928 t->multicast += multicast;
931 t->tx_dropped = net->stats.tx_dropped;
932 t->tx_errors = net->stats.tx_errors;
934 t->rx_dropped = net->stats.rx_dropped;
935 t->rx_errors = net->stats.rx_errors;
938 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
940 struct sockaddr *addr = p;
941 char save_adr[ETH_ALEN];
942 unsigned char save_aatype;
945 memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
946 save_aatype = ndev->addr_assign_type;
948 err = eth_mac_addr(ndev, p);
952 err = rndis_filter_set_device_mac(ndev, addr->sa_data);
954 /* roll back to saved MAC */
955 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
956 ndev->addr_assign_type = save_aatype;
962 static const struct {
963 char name[ETH_GSTRING_LEN];
966 { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
967 { "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
968 { "tx_no_space", offsetof(struct netvsc_ethtool_stats, tx_no_space) },
969 { "tx_too_big", offsetof(struct netvsc_ethtool_stats, tx_too_big) },
970 { "tx_busy", offsetof(struct netvsc_ethtool_stats, tx_busy) },
973 #define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats)
975 /* 4 statistics per queue (rx/tx packets/bytes) */
976 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 4)
978 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
980 struct net_device_context *ndc = netdev_priv(dev);
981 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
986 switch (string_set) {
988 return NETVSC_GLOBAL_STATS_LEN + NETVSC_QUEUE_STATS_LEN(nvdev);
994 static void netvsc_get_ethtool_stats(struct net_device *dev,
995 struct ethtool_stats *stats, u64 *data)
997 struct net_device_context *ndc = netdev_priv(dev);
998 struct netvsc_device *nvdev = rcu_dereference(ndc->nvdev);
999 const void *nds = &ndc->eth_stats;
1000 const struct netvsc_stats *qstats;
1008 for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1009 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1011 for (j = 0; j < nvdev->num_chn; j++) {
1012 qstats = &nvdev->chan_table[j].tx_stats;
1015 start = u64_stats_fetch_begin_irq(&qstats->syncp);
1016 packets = qstats->packets;
1017 bytes = qstats->bytes;
1018 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1019 data[i++] = packets;
1022 qstats = &nvdev->chan_table[j].rx_stats;
1024 start = u64_stats_fetch_begin_irq(&qstats->syncp);
1025 packets = qstats->packets;
1026 bytes = qstats->bytes;
1027 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1028 data[i++] = packets;
1033 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1035 struct net_device_context *ndc = netdev_priv(dev);
1036 struct netvsc_device *nvdev = rcu_dereference(ndc->nvdev);
1043 switch (stringset) {
1045 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++)
1046 memcpy(p + i * ETH_GSTRING_LEN,
1047 netvsc_stats[i].name, ETH_GSTRING_LEN);
1049 p += i * ETH_GSTRING_LEN;
1050 for (i = 0; i < nvdev->num_chn; i++) {
1051 sprintf(p, "tx_queue_%u_packets", i);
1052 p += ETH_GSTRING_LEN;
1053 sprintf(p, "tx_queue_%u_bytes", i);
1054 p += ETH_GSTRING_LEN;
1055 sprintf(p, "rx_queue_%u_packets", i);
1056 p += ETH_GSTRING_LEN;
1057 sprintf(p, "rx_queue_%u_bytes", i);
1058 p += ETH_GSTRING_LEN;
1066 netvsc_get_rss_hash_opts(struct netvsc_device *nvdev,
1067 struct ethtool_rxnfc *info)
1069 info->data = RXH_IP_SRC | RXH_IP_DST;
1071 switch (info->flow_type) {
1074 info->data |= RXH_L4_B_0_1 | RXH_L4_B_2_3;
1090 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1093 struct net_device_context *ndc = netdev_priv(dev);
1094 struct netvsc_device *nvdev = rcu_dereference(ndc->nvdev);
1099 switch (info->cmd) {
1100 case ETHTOOL_GRXRINGS:
1101 info->data = nvdev->num_chn;
1105 return netvsc_get_rss_hash_opts(nvdev, info);
1110 #ifdef CONFIG_NET_POLL_CONTROLLER
1111 static void netvsc_poll_controller(struct net_device *dev)
1113 struct net_device_context *ndc = netdev_priv(dev);
1114 struct netvsc_device *ndev;
1118 ndev = rcu_dereference(ndc->nvdev);
1120 for (i = 0; i < ndev->num_chn; i++) {
1121 struct netvsc_channel *nvchan = &ndev->chan_table[i];
1123 napi_schedule(&nvchan->napi);
1130 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1132 return NETVSC_HASH_KEYLEN;
1135 static u32 netvsc_rss_indir_size(struct net_device *dev)
1140 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1143 struct net_device_context *ndc = netdev_priv(dev);
1144 struct netvsc_device *ndev = rcu_dereference(ndc->nvdev);
1145 struct rndis_device *rndis_dev;
1152 *hfunc = ETH_RSS_HASH_TOP; /* Toeplitz */
1154 rndis_dev = ndev->extension;
1156 for (i = 0; i < ITAB_NUM; i++)
1157 indir[i] = rndis_dev->ind_table[i];
1161 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1166 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1167 const u8 *key, const u8 hfunc)
1169 struct net_device_context *ndc = netdev_priv(dev);
1170 struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1171 struct rndis_device *rndis_dev;
1177 if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1180 rndis_dev = ndev->extension;
1182 for (i = 0; i < ITAB_NUM; i++)
1183 if (indir[i] >= VRSS_CHANNEL_MAX)
1186 for (i = 0; i < ITAB_NUM; i++)
1187 rndis_dev->ind_table[i] = indir[i];
1194 key = rndis_dev->rss_key;
1197 return rndis_filter_set_rss_param(rndis_dev, key, ndev->num_chn);
1200 static const struct ethtool_ops ethtool_ops = {
1201 .get_drvinfo = netvsc_get_drvinfo,
1202 .get_link = ethtool_op_get_link,
1203 .get_ethtool_stats = netvsc_get_ethtool_stats,
1204 .get_sset_count = netvsc_get_sset_count,
1205 .get_strings = netvsc_get_strings,
1206 .get_channels = netvsc_get_channels,
1207 .set_channels = netvsc_set_channels,
1208 .get_ts_info = ethtool_op_get_ts_info,
1209 .get_rxnfc = netvsc_get_rxnfc,
1210 .get_rxfh_key_size = netvsc_get_rxfh_key_size,
1211 .get_rxfh_indir_size = netvsc_rss_indir_size,
1212 .get_rxfh = netvsc_get_rxfh,
1213 .set_rxfh = netvsc_set_rxfh,
1214 .get_link_ksettings = netvsc_get_link_ksettings,
1215 .set_link_ksettings = netvsc_set_link_ksettings,
1218 static const struct net_device_ops device_ops = {
1219 .ndo_open = netvsc_open,
1220 .ndo_stop = netvsc_close,
1221 .ndo_start_xmit = netvsc_start_xmit,
1222 .ndo_set_rx_mode = netvsc_set_multicast_list,
1223 .ndo_change_mtu = netvsc_change_mtu,
1224 .ndo_validate_addr = eth_validate_addr,
1225 .ndo_set_mac_address = netvsc_set_mac_addr,
1226 .ndo_select_queue = netvsc_select_queue,
1227 .ndo_get_stats64 = netvsc_get_stats64,
1228 #ifdef CONFIG_NET_POLL_CONTROLLER
1229 .ndo_poll_controller = netvsc_poll_controller,
1234 * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1235 * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1236 * present send GARP packet to network peers with netif_notify_peers().
1238 static void netvsc_link_change(struct work_struct *w)
1240 struct net_device_context *ndev_ctx =
1241 container_of(w, struct net_device_context, dwork.work);
1242 struct hv_device *device_obj = ndev_ctx->device_ctx;
1243 struct net_device *net = hv_get_drvdata(device_obj);
1244 struct netvsc_device *net_device;
1245 struct rndis_device *rdev;
1246 struct netvsc_reconfig *event = NULL;
1247 bool notify = false, reschedule = false;
1248 unsigned long flags, next_reconfig, delay;
1251 net_device = rtnl_dereference(ndev_ctx->nvdev);
1255 rdev = net_device->extension;
1257 next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1258 if (time_is_after_jiffies(next_reconfig)) {
1259 /* link_watch only sends one notification with current state
1260 * per second, avoid doing reconfig more frequently. Handle
1263 delay = next_reconfig - jiffies;
1264 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1265 schedule_delayed_work(&ndev_ctx->dwork, delay);
1268 ndev_ctx->last_reconfig = jiffies;
1270 spin_lock_irqsave(&ndev_ctx->lock, flags);
1271 if (!list_empty(&ndev_ctx->reconfig_events)) {
1272 event = list_first_entry(&ndev_ctx->reconfig_events,
1273 struct netvsc_reconfig, list);
1274 list_del(&event->list);
1275 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1277 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1282 switch (event->event) {
1283 /* Only the following events are possible due to the check in
1284 * netvsc_linkstatus_callback()
1286 case RNDIS_STATUS_MEDIA_CONNECT:
1287 if (rdev->link_state) {
1288 rdev->link_state = false;
1289 if (!ndev_ctx->datapath)
1290 netif_carrier_on(net);
1291 netif_tx_wake_all_queues(net);
1297 case RNDIS_STATUS_MEDIA_DISCONNECT:
1298 if (!rdev->link_state) {
1299 rdev->link_state = true;
1300 netif_carrier_off(net);
1301 netif_tx_stop_all_queues(net);
1305 case RNDIS_STATUS_NETWORK_CHANGE:
1306 /* Only makes sense if carrier is present */
1307 if (!rdev->link_state) {
1308 rdev->link_state = true;
1309 netif_carrier_off(net);
1310 netif_tx_stop_all_queues(net);
1311 event->event = RNDIS_STATUS_MEDIA_CONNECT;
1312 spin_lock_irqsave(&ndev_ctx->lock, flags);
1313 list_add(&event->list, &ndev_ctx->reconfig_events);
1314 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1323 netdev_notify_peers(net);
1325 /* link_watch only sends one notification with current state per
1326 * second, handle next reconfig event in 2 seconds.
1329 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1337 static struct net_device *get_netvsc_bymac(const u8 *mac)
1339 struct net_device *dev;
1343 for_each_netdev(&init_net, dev) {
1344 if (dev->netdev_ops != &device_ops)
1345 continue; /* not a netvsc device */
1347 if (ether_addr_equal(mac, dev->perm_addr))
1354 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
1356 struct net_device *dev;
1360 for_each_netdev(&init_net, dev) {
1361 struct net_device_context *net_device_ctx;
1363 if (dev->netdev_ops != &device_ops)
1364 continue; /* not a netvsc device */
1366 net_device_ctx = netdev_priv(dev);
1367 if (!rtnl_dereference(net_device_ctx->nvdev))
1368 continue; /* device is removed */
1370 if (rtnl_dereference(net_device_ctx->vf_netdev) == vf_netdev)
1371 return dev; /* a match */
1377 static int netvsc_register_vf(struct net_device *vf_netdev)
1379 struct net_device *ndev;
1380 struct net_device_context *net_device_ctx;
1381 struct netvsc_device *netvsc_dev;
1383 if (vf_netdev->addr_len != ETH_ALEN)
1387 * We will use the MAC address to locate the synthetic interface to
1388 * associate with the VF interface. If we don't find a matching
1389 * synthetic interface, move on.
1391 ndev = get_netvsc_bymac(vf_netdev->perm_addr);
1395 net_device_ctx = netdev_priv(ndev);
1396 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
1397 if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
1400 netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
1402 * Take a reference on the module.
1404 try_module_get(THIS_MODULE);
1406 dev_hold(vf_netdev);
1407 rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
1411 static int netvsc_vf_up(struct net_device *vf_netdev)
1413 struct net_device *ndev;
1414 struct netvsc_device *netvsc_dev;
1415 struct net_device_context *net_device_ctx;
1417 ndev = get_netvsc_byref(vf_netdev);
1421 net_device_ctx = netdev_priv(ndev);
1422 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
1424 netdev_info(ndev, "VF up: %s\n", vf_netdev->name);
1427 * Open the device before switching data path.
1429 rndis_filter_open(netvsc_dev);
1432 * notify the host to switch the data path.
1434 netvsc_switch_datapath(ndev, true);
1435 netdev_info(ndev, "Data path switched to VF: %s\n", vf_netdev->name);
1437 netif_carrier_off(ndev);
1439 /* Now notify peers through VF device. */
1440 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, vf_netdev);
1445 static int netvsc_vf_down(struct net_device *vf_netdev)
1447 struct net_device *ndev;
1448 struct netvsc_device *netvsc_dev;
1449 struct net_device_context *net_device_ctx;
1451 ndev = get_netvsc_byref(vf_netdev);
1455 net_device_ctx = netdev_priv(ndev);
1456 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
1458 netdev_info(ndev, "VF down: %s\n", vf_netdev->name);
1459 netvsc_switch_datapath(ndev, false);
1460 netdev_info(ndev, "Data path switched from VF: %s\n", vf_netdev->name);
1461 rndis_filter_close(netvsc_dev);
1462 netif_carrier_on(ndev);
1464 /* Now notify peers through netvsc device. */
1465 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, ndev);
1470 static int netvsc_unregister_vf(struct net_device *vf_netdev)
1472 struct net_device *ndev;
1473 struct net_device_context *net_device_ctx;
1475 ndev = get_netvsc_byref(vf_netdev);
1479 net_device_ctx = netdev_priv(ndev);
1481 netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
1483 RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
1485 module_put(THIS_MODULE);
1489 static int netvsc_probe(struct hv_device *dev,
1490 const struct hv_vmbus_device_id *dev_id)
1492 struct net_device *net = NULL;
1493 struct net_device_context *net_device_ctx;
1494 struct netvsc_device_info device_info;
1495 struct netvsc_device *nvdev;
1498 net = alloc_etherdev_mq(sizeof(struct net_device_context),
1503 netif_carrier_off(net);
1505 netvsc_init_settings(net);
1507 net_device_ctx = netdev_priv(net);
1508 net_device_ctx->device_ctx = dev;
1509 net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
1510 if (netif_msg_probe(net_device_ctx))
1511 netdev_dbg(net, "netvsc msg_enable: %d\n",
1512 net_device_ctx->msg_enable);
1514 hv_set_drvdata(dev, net);
1516 INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
1518 spin_lock_init(&net_device_ctx->lock);
1519 INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
1521 net->netdev_ops = &device_ops;
1522 net->ethtool_ops = ðtool_ops;
1523 SET_NETDEV_DEV(net, &dev->device);
1525 /* We always need headroom for rndis header */
1526 net->needed_headroom = RNDIS_AND_PPI_SIZE;
1528 /* Notify the netvsc driver of the new device */
1529 memset(&device_info, 0, sizeof(device_info));
1530 device_info.ring_size = ring_size;
1531 device_info.num_chn = VRSS_CHANNEL_DEFAULT;
1533 nvdev = rndis_filter_device_add(dev, &device_info);
1534 if (IS_ERR(nvdev)) {
1535 ret = PTR_ERR(nvdev);
1536 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
1538 hv_set_drvdata(dev, NULL);
1541 memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
1543 /* hw_features computed in rndis_filter_device_add */
1544 net->features = net->hw_features |
1545 NETIF_F_HIGHDMA | NETIF_F_SG |
1546 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
1547 net->vlan_features = net->features;
1549 netif_set_real_num_tx_queues(net, nvdev->num_chn);
1550 netif_set_real_num_rx_queues(net, nvdev->num_chn);
1552 netdev_lockdep_set_classes(net);
1554 /* MTU range: 68 - 1500 or 65521 */
1555 net->min_mtu = NETVSC_MTU_MIN;
1556 if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
1557 net->max_mtu = NETVSC_MTU - ETH_HLEN;
1559 net->max_mtu = ETH_DATA_LEN;
1561 ret = register_netdev(net);
1563 pr_err("Unable to register netdev.\n");
1564 rndis_filter_device_remove(dev, nvdev);
1571 static int netvsc_remove(struct hv_device *dev)
1573 struct net_device *net;
1574 struct net_device_context *ndev_ctx;
1576 net = hv_get_drvdata(dev);
1579 dev_err(&dev->device, "No net device to remove\n");
1583 ndev_ctx = netdev_priv(net);
1585 netif_device_detach(net);
1587 cancel_delayed_work_sync(&ndev_ctx->dwork);
1590 * Call to the vsc driver to let it know that the device is being
1591 * removed. Also blocks mtu and channel changes.
1594 rndis_filter_device_remove(dev,
1595 rtnl_dereference(ndev_ctx->nvdev));
1598 unregister_netdev(net);
1600 hv_set_drvdata(dev, NULL);
1606 static const struct hv_vmbus_device_id id_table[] = {
1612 MODULE_DEVICE_TABLE(vmbus, id_table);
1614 /* The one and only one */
1615 static struct hv_driver netvsc_drv = {
1616 .name = KBUILD_MODNAME,
1617 .id_table = id_table,
1618 .probe = netvsc_probe,
1619 .remove = netvsc_remove,
1623 * On Hyper-V, every VF interface is matched with a corresponding
1624 * synthetic interface. The synthetic interface is presented first
1625 * to the guest. When the corresponding VF instance is registered,
1626 * we will take care of switching the data path.
1628 static int netvsc_netdev_event(struct notifier_block *this,
1629 unsigned long event, void *ptr)
1631 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
1633 /* Skip our own events */
1634 if (event_dev->netdev_ops == &device_ops)
1637 /* Avoid non-Ethernet type devices */
1638 if (event_dev->type != ARPHRD_ETHER)
1641 /* Avoid Vlan dev with same MAC registering as VF */
1642 if (is_vlan_dev(event_dev))
1645 /* Avoid Bonding master dev with same MAC registering as VF */
1646 if ((event_dev->priv_flags & IFF_BONDING) &&
1647 (event_dev->flags & IFF_MASTER))
1651 case NETDEV_REGISTER:
1652 return netvsc_register_vf(event_dev);
1653 case NETDEV_UNREGISTER:
1654 return netvsc_unregister_vf(event_dev);
1656 return netvsc_vf_up(event_dev);
1658 return netvsc_vf_down(event_dev);
1664 static struct notifier_block netvsc_netdev_notifier = {
1665 .notifier_call = netvsc_netdev_event,
1668 static void __exit netvsc_drv_exit(void)
1670 unregister_netdevice_notifier(&netvsc_netdev_notifier);
1671 vmbus_driver_unregister(&netvsc_drv);
1674 static int __init netvsc_drv_init(void)
1678 if (ring_size < RING_SIZE_MIN) {
1679 ring_size = RING_SIZE_MIN;
1680 pr_info("Increased ring_size to %d (min allowed)\n",
1683 ret = vmbus_driver_register(&netvsc_drv);
1688 register_netdevice_notifier(&netvsc_netdev_notifier);
1692 MODULE_LICENSE("GPL");
1693 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
1695 module_init(netvsc_drv_init);
1696 module_exit(netvsc_drv_exit);