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>
41 #include "hyperv_net.h"
43 #define RING_SIZE_MIN 64
44 #define LINKCHANGE_INT (2 * HZ)
46 static int ring_size = 128;
47 module_param(ring_size, int, S_IRUGO);
48 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
50 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
51 NETIF_MSG_LINK | NETIF_MSG_IFUP |
52 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
55 static int debug = -1;
56 module_param(debug, int, S_IRUGO);
57 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
59 static void do_set_multicast(struct work_struct *w)
61 struct net_device_context *ndevctx =
62 container_of(w, struct net_device_context, work);
63 struct hv_device *device_obj = ndevctx->device_ctx;
64 struct net_device *ndev = hv_get_drvdata(device_obj);
65 struct netvsc_device *nvdev = rcu_dereference(ndevctx->nvdev);
66 struct rndis_device *rdev;
71 rdev = nvdev->extension;
75 if (ndev->flags & IFF_PROMISC)
76 rndis_filter_set_packet_filter(rdev,
77 NDIS_PACKET_TYPE_PROMISCUOUS);
79 rndis_filter_set_packet_filter(rdev,
80 NDIS_PACKET_TYPE_BROADCAST |
81 NDIS_PACKET_TYPE_ALL_MULTICAST |
82 NDIS_PACKET_TYPE_DIRECTED);
85 static void netvsc_set_multicast_list(struct net_device *net)
87 struct net_device_context *net_device_ctx = netdev_priv(net);
89 schedule_work(&net_device_ctx->work);
92 static int netvsc_open(struct net_device *net)
94 struct netvsc_device *nvdev = net_device_to_netvsc_device(net);
95 struct rndis_device *rdev;
98 netif_carrier_off(net);
100 /* Open up the device */
101 ret = rndis_filter_open(nvdev);
103 netdev_err(net, "unable to open device (ret %d).\n", ret);
107 netif_tx_wake_all_queues(net);
109 rdev = nvdev->extension;
110 if (!rdev->link_state)
111 netif_carrier_on(net);
116 static int netvsc_close(struct net_device *net)
118 struct net_device_context *net_device_ctx = netdev_priv(net);
119 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
121 u32 aread, awrite, i, msec = 10, retry = 0, retry_max = 20;
122 struct vmbus_channel *chn;
124 netif_tx_disable(net);
126 /* Make sure netvsc_set_multicast_list doesn't re-enable filter! */
127 cancel_work_sync(&net_device_ctx->work);
128 ret = rndis_filter_close(nvdev);
130 netdev_err(net, "unable to close device (ret %d).\n", ret);
134 /* Ensure pending bytes in ring are read */
137 for (i = 0; i < nvdev->num_chn; i++) {
138 chn = nvdev->chan_table[i].channel;
142 hv_get_ringbuffer_availbytes(&chn->inbound, &aread,
148 hv_get_ringbuffer_availbytes(&chn->outbound, &aread,
156 if (retry > retry_max || aread == 0)
166 netdev_err(net, "Ring buffer not empty after closing rndis\n");
173 static void *init_ppi_data(struct rndis_message *msg, u32 ppi_size,
176 struct rndis_packet *rndis_pkt;
177 struct rndis_per_packet_info *ppi;
179 rndis_pkt = &msg->msg.pkt;
180 rndis_pkt->data_offset += ppi_size;
182 ppi = (struct rndis_per_packet_info *)((void *)rndis_pkt +
183 rndis_pkt->per_pkt_info_offset + rndis_pkt->per_pkt_info_len);
185 ppi->size = ppi_size;
186 ppi->type = pkt_type;
187 ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
189 rndis_pkt->per_pkt_info_len += ppi_size;
194 /* Azure hosts don't support non-TCP port numbers in hashing yet. We compute
195 * hash for non-TCP traffic with only IP numbers.
197 static inline u32 netvsc_get_hash(struct sk_buff *skb, struct sock *sk)
199 struct flow_keys flow;
201 static u32 hashrnd __read_mostly;
203 net_get_random_once(&hashrnd, sizeof(hashrnd));
205 if (!skb_flow_dissect_flow_keys(skb, &flow, 0))
208 if (flow.basic.ip_proto == IPPROTO_TCP) {
209 return skb_get_hash(skb);
211 if (flow.basic.n_proto == htons(ETH_P_IP))
212 hash = jhash2((u32 *)&flow.addrs.v4addrs, 2, hashrnd);
213 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
214 hash = jhash2((u32 *)&flow.addrs.v6addrs, 8, hashrnd);
218 skb_set_hash(skb, hash, PKT_HASH_TYPE_L3);
224 static inline int netvsc_get_tx_queue(struct net_device *ndev,
225 struct sk_buff *skb, int old_idx)
227 const struct net_device_context *ndc = netdev_priv(ndev);
228 struct sock *sk = skb->sk;
231 q_idx = ndc->tx_send_table[netvsc_get_hash(skb, sk) &
232 (VRSS_SEND_TAB_SIZE - 1)];
234 /* If queue index changed record the new value */
235 if (q_idx != old_idx &&
236 sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
237 sk_tx_queue_set(sk, q_idx);
243 * Select queue for transmit.
245 * If a valid queue has already been assigned, then use that.
246 * Otherwise compute tx queue based on hash and the send table.
248 * This is basically similar to default (__netdev_pick_tx) with the added step
249 * of using the host send_table when no other queue has been assigned.
251 * TODO support XPS - but get_xps_queue not exported
253 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
254 void *accel_priv, select_queue_fallback_t fallback)
256 unsigned int num_tx_queues = ndev->real_num_tx_queues;
257 int q_idx = sk_tx_queue_get(skb->sk);
259 if (q_idx < 0 || skb->ooo_okay) {
260 /* If forwarding a packet, we use the recorded queue when
261 * available for better cache locality.
263 if (skb_rx_queue_recorded(skb))
264 q_idx = skb_get_rx_queue(skb);
266 q_idx = netvsc_get_tx_queue(ndev, skb, q_idx);
269 while (unlikely(q_idx >= num_tx_queues))
270 q_idx -= num_tx_queues;
275 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
276 struct hv_page_buffer *pb)
280 /* Deal with compund pages by ignoring unused part
283 page += (offset >> PAGE_SHIFT);
284 offset &= ~PAGE_MASK;
289 bytes = PAGE_SIZE - offset;
292 pb[j].pfn = page_to_pfn(page);
293 pb[j].offset = offset;
299 if (offset == PAGE_SIZE && len) {
309 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
310 struct hv_netvsc_packet *packet,
311 struct hv_page_buffer **page_buf)
313 struct hv_page_buffer *pb = *page_buf;
315 char *data = skb->data;
316 int frags = skb_shinfo(skb)->nr_frags;
319 /* The packet is laid out thus:
320 * 1. hdr: RNDIS header and PPI
322 * 3. skb fragment data
325 slots_used += fill_pg_buf(virt_to_page(hdr),
327 len, &pb[slots_used]);
329 packet->rmsg_size = len;
330 packet->rmsg_pgcnt = slots_used;
332 slots_used += fill_pg_buf(virt_to_page(data),
333 offset_in_page(data),
334 skb_headlen(skb), &pb[slots_used]);
336 for (i = 0; i < frags; i++) {
337 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
339 slots_used += fill_pg_buf(skb_frag_page(frag),
341 skb_frag_size(frag), &pb[slots_used]);
346 static int count_skb_frag_slots(struct sk_buff *skb)
348 int i, frags = skb_shinfo(skb)->nr_frags;
351 for (i = 0; i < frags; i++) {
352 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
353 unsigned long size = skb_frag_size(frag);
354 unsigned long offset = frag->page_offset;
356 /* Skip unused frames from start of page */
357 offset &= ~PAGE_MASK;
358 pages += PFN_UP(offset + size);
363 static int netvsc_get_slots(struct sk_buff *skb)
365 char *data = skb->data;
366 unsigned int offset = offset_in_page(data);
367 unsigned int len = skb_headlen(skb);
371 slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
372 frag_slots = count_skb_frag_slots(skb);
373 return slots + frag_slots;
376 static u32 net_checksum_info(struct sk_buff *skb)
378 if (skb->protocol == htons(ETH_P_IP)) {
379 struct iphdr *ip = ip_hdr(skb);
381 if (ip->protocol == IPPROTO_TCP)
382 return TRANSPORT_INFO_IPV4_TCP;
383 else if (ip->protocol == IPPROTO_UDP)
384 return TRANSPORT_INFO_IPV4_UDP;
386 struct ipv6hdr *ip6 = ipv6_hdr(skb);
388 if (ip6->nexthdr == IPPROTO_TCP)
389 return TRANSPORT_INFO_IPV6_TCP;
390 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
391 return TRANSPORT_INFO_IPV6_UDP;
394 return TRANSPORT_INFO_NOT_IP;
397 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
399 struct net_device_context *net_device_ctx = netdev_priv(net);
400 struct hv_netvsc_packet *packet = NULL;
402 unsigned int num_data_pgs;
403 struct rndis_message *rndis_msg;
404 struct rndis_packet *rndis_pkt;
406 struct rndis_per_packet_info *ppi;
408 struct hv_page_buffer page_buf[MAX_PAGE_BUFFER_COUNT];
409 struct hv_page_buffer *pb = page_buf;
411 /* We will atmost need two pages to describe the rndis
412 * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
413 * of pages in a single packet. If skb is scattered around
414 * more pages we try linearizing it.
417 num_data_pgs = netvsc_get_slots(skb) + 2;
419 if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
420 ++net_device_ctx->eth_stats.tx_scattered;
422 if (skb_linearize(skb))
425 num_data_pgs = netvsc_get_slots(skb) + 2;
426 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
427 ++net_device_ctx->eth_stats.tx_too_big;
433 * Place the rndis header in the skb head room and
434 * the skb->cb will be used for hv_netvsc_packet
437 ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
441 /* Use the skb control buffer for building up the packet */
442 BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
443 FIELD_SIZEOF(struct sk_buff, cb));
444 packet = (struct hv_netvsc_packet *)skb->cb;
446 packet->q_idx = skb_get_queue_mapping(skb);
448 packet->total_data_buflen = skb->len;
449 packet->total_bytes = skb->len;
450 packet->total_packets = 1;
452 rndis_msg = (struct rndis_message *)skb->head;
454 memset(rndis_msg, 0, RNDIS_AND_PPI_SIZE);
456 /* Add the rndis header */
457 rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
458 rndis_msg->msg_len = packet->total_data_buflen;
459 rndis_pkt = &rndis_msg->msg.pkt;
460 rndis_pkt->data_offset = sizeof(struct rndis_packet);
461 rndis_pkt->data_len = packet->total_data_buflen;
462 rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
464 rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
466 hash = skb_get_hash_raw(skb);
467 if (hash != 0 && net->real_num_tx_queues > 1) {
468 rndis_msg_size += NDIS_HASH_PPI_SIZE;
469 ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
471 *(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
474 if (skb_vlan_tag_present(skb)) {
475 struct ndis_pkt_8021q_info *vlan;
477 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
478 ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
480 vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
482 vlan->vlanid = skb->vlan_tci & VLAN_VID_MASK;
483 vlan->pri = (skb->vlan_tci & VLAN_PRIO_MASK) >>
487 if (skb_is_gso(skb)) {
488 struct ndis_tcp_lso_info *lso_info;
490 rndis_msg_size += NDIS_LSO_PPI_SIZE;
491 ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
492 TCP_LARGESEND_PKTINFO);
494 lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
497 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
498 if (skb->protocol == htons(ETH_P_IP)) {
499 lso_info->lso_v2_transmit.ip_version =
500 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
501 ip_hdr(skb)->tot_len = 0;
502 ip_hdr(skb)->check = 0;
503 tcp_hdr(skb)->check =
504 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
505 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
507 lso_info->lso_v2_transmit.ip_version =
508 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
509 ipv6_hdr(skb)->payload_len = 0;
510 tcp_hdr(skb)->check =
511 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
512 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
514 lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
515 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
516 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
517 if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
518 struct ndis_tcp_ip_checksum_info *csum_info;
520 rndis_msg_size += NDIS_CSUM_PPI_SIZE;
521 ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
522 TCPIP_CHKSUM_PKTINFO);
524 csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
527 csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
529 if (skb->protocol == htons(ETH_P_IP)) {
530 csum_info->transmit.is_ipv4 = 1;
532 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
533 csum_info->transmit.tcp_checksum = 1;
535 csum_info->transmit.udp_checksum = 1;
537 csum_info->transmit.is_ipv6 = 1;
539 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
540 csum_info->transmit.tcp_checksum = 1;
542 csum_info->transmit.udp_checksum = 1;
545 /* Can't do offload of this type of checksum */
546 if (skb_checksum_help(skb))
551 /* Start filling in the page buffers with the rndis hdr */
552 rndis_msg->msg_len += rndis_msg_size;
553 packet->total_data_buflen = rndis_msg->msg_len;
554 packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
557 /* timestamp packet in software */
558 skb_tx_timestamp(skb);
559 ret = netvsc_send(net_device_ctx->device_ctx, packet,
560 rndis_msg, &pb, skb);
561 if (likely(ret == 0))
564 if (ret == -EAGAIN) {
565 ++net_device_ctx->eth_stats.tx_busy;
566 return NETDEV_TX_BUSY;
570 ++net_device_ctx->eth_stats.tx_no_space;
573 dev_kfree_skb_any(skb);
574 net->stats.tx_dropped++;
579 ++net_device_ctx->eth_stats.tx_no_memory;
583 * netvsc_linkstatus_callback - Link up/down notification
585 void netvsc_linkstatus_callback(struct hv_device *device_obj,
586 struct rndis_message *resp)
588 struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
589 struct net_device *net;
590 struct net_device_context *ndev_ctx;
591 struct netvsc_reconfig *event;
594 net = hv_get_drvdata(device_obj);
599 ndev_ctx = netdev_priv(net);
601 /* Update the physical link speed when changing to another vSwitch */
602 if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
605 speed = *(u32 *)((void *)indicate + indicate->
606 status_buf_offset) / 10000;
607 ndev_ctx->speed = speed;
611 /* Handle these link change statuses below */
612 if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
613 indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
614 indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
617 if (net->reg_state != NETREG_REGISTERED)
620 event = kzalloc(sizeof(*event), GFP_ATOMIC);
623 event->event = indicate->status;
625 spin_lock_irqsave(&ndev_ctx->lock, flags);
626 list_add_tail(&event->list, &ndev_ctx->reconfig_events);
627 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
629 schedule_delayed_work(&ndev_ctx->dwork, 0);
632 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
633 struct napi_struct *napi,
634 const struct ndis_tcp_ip_checksum_info *csum_info,
635 const struct ndis_pkt_8021q_info *vlan,
636 void *data, u32 buflen)
640 skb = napi_alloc_skb(napi, buflen);
645 * Copy to skb. This copy is needed here since the memory pointed by
646 * hv_netvsc_packet cannot be deallocated
648 memcpy(skb_put(skb, buflen), data, buflen);
650 skb->protocol = eth_type_trans(skb, net);
652 /* skb is already created with CHECKSUM_NONE */
653 skb_checksum_none_assert(skb);
656 * In Linux, the IP checksum is always checked.
657 * Do L4 checksum offload if enabled and present.
659 if (csum_info && (net->features & NETIF_F_RXCSUM)) {
660 if (csum_info->receive.tcp_checksum_succeeded ||
661 csum_info->receive.udp_checksum_succeeded)
662 skb->ip_summed = CHECKSUM_UNNECESSARY;
666 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT);
668 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
676 * netvsc_recv_callback - Callback when we receive a packet from the
677 * "wire" on the specified device.
679 int netvsc_recv_callback(struct net_device *net,
680 struct vmbus_channel *channel,
682 const struct ndis_tcp_ip_checksum_info *csum_info,
683 const struct ndis_pkt_8021q_info *vlan)
685 struct net_device_context *net_device_ctx = netdev_priv(net);
686 struct netvsc_device *net_device;
687 u16 q_idx = channel->offermsg.offer.sub_channel_index;
688 struct netvsc_channel *nvchan;
689 struct net_device *vf_netdev;
691 struct netvsc_stats *rx_stats;
693 if (net->reg_state != NETREG_REGISTERED)
694 return NVSP_STAT_FAIL;
697 * If necessary, inject this packet into the VF interface.
698 * On Hyper-V, multicast and brodcast packets are only delivered
699 * to the synthetic interface (after subjecting these to
700 * policy filters on the host). Deliver these via the VF
701 * interface in the guest.
704 net_device = rcu_dereference(net_device_ctx->nvdev);
705 if (unlikely(!net_device))
708 nvchan = &net_device->chan_table[q_idx];
709 vf_netdev = rcu_dereference(net_device_ctx->vf_netdev);
710 if (vf_netdev && (vf_netdev->flags & IFF_UP))
713 /* Allocate a skb - TODO direct I/O to pages? */
714 skb = netvsc_alloc_recv_skb(net, &nvchan->napi,
715 csum_info, vlan, data, len);
716 if (unlikely(!skb)) {
718 ++net->stats.rx_dropped;
720 return NVSP_STAT_FAIL;
723 if (net != vf_netdev)
724 skb_record_rx_queue(skb, q_idx);
727 * Even if injecting the packet, record the statistics
728 * on the synthetic device because modifying the VF device
729 * statistics will not work correctly.
731 rx_stats = &nvchan->rx_stats;
732 u64_stats_update_begin(&rx_stats->syncp);
734 rx_stats->bytes += len;
736 if (skb->pkt_type == PACKET_BROADCAST)
737 ++rx_stats->broadcast;
738 else if (skb->pkt_type == PACKET_MULTICAST)
739 ++rx_stats->multicast;
740 u64_stats_update_end(&rx_stats->syncp);
742 napi_gro_receive(&nvchan->napi, skb);
748 static void netvsc_get_drvinfo(struct net_device *net,
749 struct ethtool_drvinfo *info)
751 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
752 strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
755 static void netvsc_get_channels(struct net_device *net,
756 struct ethtool_channels *channel)
758 struct net_device_context *net_device_ctx = netdev_priv(net);
759 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
762 channel->max_combined = nvdev->max_chn;
763 channel->combined_count = nvdev->num_chn;
767 static int netvsc_set_queues(struct net_device *net, struct hv_device *dev,
770 struct netvsc_device_info device_info;
773 memset(&device_info, 0, sizeof(device_info));
774 device_info.num_chn = num_chn;
775 device_info.ring_size = ring_size;
776 device_info.max_num_vrss_chns = num_chn;
778 ret = rndis_filter_device_add(dev, &device_info);
782 ret = netif_set_real_num_tx_queues(net, num_chn);
786 ret = netif_set_real_num_rx_queues(net, num_chn);
791 static int netvsc_set_channels(struct net_device *net,
792 struct ethtool_channels *channels)
794 struct net_device_context *net_device_ctx = netdev_priv(net);
795 struct hv_device *dev = net_device_ctx->device_ctx;
796 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
797 unsigned int count = channels->combined_count;
801 /* We do not support separate count for rx, tx, or other */
803 channels->rx_count || channels->tx_count || channels->other_count)
806 if (count > net->num_tx_queues || count > net->num_rx_queues)
809 if (!nvdev || nvdev->destroy)
812 if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
815 if (count > nvdev->max_chn)
818 was_running = netif_running(net);
820 ret = netvsc_close(net);
825 rndis_filter_device_remove(dev, nvdev);
827 ret = netvsc_set_queues(net, dev, count);
829 nvdev->num_chn = count;
831 netvsc_set_queues(net, dev, nvdev->num_chn);
834 ret = netvsc_open(net);
836 /* We may have missed link change notifications */
837 schedule_delayed_work(&net_device_ctx->dwork, 0);
843 netvsc_validate_ethtool_ss_cmd(const struct ethtool_link_ksettings *cmd)
845 struct ethtool_link_ksettings diff1 = *cmd;
846 struct ethtool_link_ksettings diff2 = {};
848 diff1.base.speed = 0;
849 diff1.base.duplex = 0;
850 /* advertising and cmd are usually set */
851 ethtool_link_ksettings_zero_link_mode(&diff1, advertising);
853 /* We set port to PORT_OTHER */
854 diff2.base.port = PORT_OTHER;
856 return !memcmp(&diff1, &diff2, sizeof(diff1));
859 static void netvsc_init_settings(struct net_device *dev)
861 struct net_device_context *ndc = netdev_priv(dev);
863 ndc->speed = SPEED_UNKNOWN;
864 ndc->duplex = DUPLEX_FULL;
867 static int netvsc_get_link_ksettings(struct net_device *dev,
868 struct ethtool_link_ksettings *cmd)
870 struct net_device_context *ndc = netdev_priv(dev);
872 cmd->base.speed = ndc->speed;
873 cmd->base.duplex = ndc->duplex;
874 cmd->base.port = PORT_OTHER;
879 static int netvsc_set_link_ksettings(struct net_device *dev,
880 const struct ethtool_link_ksettings *cmd)
882 struct net_device_context *ndc = netdev_priv(dev);
885 speed = cmd->base.speed;
886 if (!ethtool_validate_speed(speed) ||
887 !ethtool_validate_duplex(cmd->base.duplex) ||
888 !netvsc_validate_ethtool_ss_cmd(cmd))
892 ndc->duplex = cmd->base.duplex;
897 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
899 struct net_device_context *ndevctx = netdev_priv(ndev);
900 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
901 struct hv_device *hdev = ndevctx->device_ctx;
902 struct netvsc_device_info device_info;
906 if (!nvdev || nvdev->destroy)
909 was_running = netif_running(ndev);
911 ret = netvsc_close(ndev);
916 memset(&device_info, 0, sizeof(device_info));
917 device_info.ring_size = ring_size;
918 device_info.num_chn = nvdev->num_chn;
919 device_info.max_num_vrss_chns = nvdev->num_chn;
921 rndis_filter_device_remove(hdev, nvdev);
923 /* 'nvdev' has been freed in rndis_filter_device_remove() ->
924 * netvsc_device_remove () -> free_netvsc_device().
925 * We mustn't access it before it's re-created in
926 * rndis_filter_device_add() -> netvsc_device_add().
931 rndis_filter_device_add(hdev, &device_info);
934 ret = netvsc_open(ndev);
936 /* We may have missed link change notifications */
937 schedule_delayed_work(&ndevctx->dwork, 0);
942 static void netvsc_get_stats64(struct net_device *net,
943 struct rtnl_link_stats64 *t)
945 struct net_device_context *ndev_ctx = netdev_priv(net);
946 struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
952 for (i = 0; i < nvdev->num_chn; i++) {
953 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
954 const struct netvsc_stats *stats;
955 u64 packets, bytes, multicast;
958 stats = &nvchan->tx_stats;
960 start = u64_stats_fetch_begin_irq(&stats->syncp);
961 packets = stats->packets;
962 bytes = stats->bytes;
963 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
965 t->tx_bytes += bytes;
966 t->tx_packets += packets;
968 stats = &nvchan->rx_stats;
970 start = u64_stats_fetch_begin_irq(&stats->syncp);
971 packets = stats->packets;
972 bytes = stats->bytes;
973 multicast = stats->multicast + stats->broadcast;
974 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
976 t->rx_bytes += bytes;
977 t->rx_packets += packets;
978 t->multicast += multicast;
981 t->tx_dropped = net->stats.tx_dropped;
982 t->tx_errors = net->stats.tx_errors;
984 t->rx_dropped = net->stats.rx_dropped;
985 t->rx_errors = net->stats.rx_errors;
988 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
990 struct sockaddr *addr = p;
991 char save_adr[ETH_ALEN];
992 unsigned char save_aatype;
995 memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
996 save_aatype = ndev->addr_assign_type;
998 err = eth_mac_addr(ndev, p);
1002 err = rndis_filter_set_device_mac(ndev, addr->sa_data);
1004 /* roll back to saved MAC */
1005 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
1006 ndev->addr_assign_type = save_aatype;
1012 static const struct {
1013 char name[ETH_GSTRING_LEN];
1015 } netvsc_stats[] = {
1016 { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1017 { "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1018 { "tx_no_space", offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1019 { "tx_too_big", offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1020 { "tx_busy", offsetof(struct netvsc_ethtool_stats, tx_busy) },
1023 #define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats)
1025 /* 4 statistics per queue (rx/tx packets/bytes) */
1026 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 4)
1028 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1030 struct net_device_context *ndc = netdev_priv(dev);
1031 struct netvsc_device *nvdev = rcu_dereference(ndc->nvdev);
1036 switch (string_set) {
1038 return NETVSC_GLOBAL_STATS_LEN + NETVSC_QUEUE_STATS_LEN(nvdev);
1044 static void netvsc_get_ethtool_stats(struct net_device *dev,
1045 struct ethtool_stats *stats, u64 *data)
1047 struct net_device_context *ndc = netdev_priv(dev);
1048 struct netvsc_device *nvdev = rcu_dereference(ndc->nvdev);
1049 const void *nds = &ndc->eth_stats;
1050 const struct netvsc_stats *qstats;
1058 for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1059 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1061 for (j = 0; j < nvdev->num_chn; j++) {
1062 qstats = &nvdev->chan_table[j].tx_stats;
1065 start = u64_stats_fetch_begin_irq(&qstats->syncp);
1066 packets = qstats->packets;
1067 bytes = qstats->bytes;
1068 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1069 data[i++] = packets;
1072 qstats = &nvdev->chan_table[j].rx_stats;
1074 start = u64_stats_fetch_begin_irq(&qstats->syncp);
1075 packets = qstats->packets;
1076 bytes = qstats->bytes;
1077 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1078 data[i++] = packets;
1083 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1085 struct net_device_context *ndc = netdev_priv(dev);
1086 struct netvsc_device *nvdev = rcu_dereference(ndc->nvdev);
1093 switch (stringset) {
1095 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++)
1096 memcpy(p + i * ETH_GSTRING_LEN,
1097 netvsc_stats[i].name, ETH_GSTRING_LEN);
1099 p += i * ETH_GSTRING_LEN;
1100 for (i = 0; i < nvdev->num_chn; i++) {
1101 sprintf(p, "tx_queue_%u_packets", i);
1102 p += ETH_GSTRING_LEN;
1103 sprintf(p, "tx_queue_%u_bytes", i);
1104 p += ETH_GSTRING_LEN;
1105 sprintf(p, "rx_queue_%u_packets", i);
1106 p += ETH_GSTRING_LEN;
1107 sprintf(p, "rx_queue_%u_bytes", i);
1108 p += ETH_GSTRING_LEN;
1116 netvsc_get_rss_hash_opts(struct netvsc_device *nvdev,
1117 struct ethtool_rxnfc *info)
1119 info->data = RXH_IP_SRC | RXH_IP_DST;
1121 switch (info->flow_type) {
1124 info->data |= RXH_L4_B_0_1 | RXH_L4_B_2_3;
1140 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1143 struct net_device_context *ndc = netdev_priv(dev);
1144 struct netvsc_device *nvdev = rcu_dereference(ndc->nvdev);
1149 switch (info->cmd) {
1150 case ETHTOOL_GRXRINGS:
1151 info->data = nvdev->num_chn;
1155 return netvsc_get_rss_hash_opts(nvdev, info);
1160 #ifdef CONFIG_NET_POLL_CONTROLLER
1161 static void netvsc_poll_controller(struct net_device *net)
1163 /* As netvsc_start_xmit() works synchronous we don't have to
1164 * trigger anything here.
1169 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1171 return NETVSC_HASH_KEYLEN;
1174 static u32 netvsc_rss_indir_size(struct net_device *dev)
1179 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1182 struct net_device_context *ndc = netdev_priv(dev);
1183 struct netvsc_device *ndev = rcu_dereference(ndc->nvdev);
1184 struct rndis_device *rndis_dev;
1191 *hfunc = ETH_RSS_HASH_TOP; /* Toeplitz */
1193 rndis_dev = ndev->extension;
1195 for (i = 0; i < ITAB_NUM; i++)
1196 indir[i] = rndis_dev->ind_table[i];
1200 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1205 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1206 const u8 *key, const u8 hfunc)
1208 struct net_device_context *ndc = netdev_priv(dev);
1209 struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1210 struct rndis_device *rndis_dev;
1216 if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1219 rndis_dev = ndev->extension;
1221 for (i = 0; i < ITAB_NUM; i++)
1222 if (indir[i] >= dev->num_rx_queues)
1225 for (i = 0; i < ITAB_NUM; i++)
1226 rndis_dev->ind_table[i] = indir[i];
1233 key = rndis_dev->rss_key;
1236 return rndis_filter_set_rss_param(rndis_dev, key, ndev->num_chn);
1239 static const struct ethtool_ops ethtool_ops = {
1240 .get_drvinfo = netvsc_get_drvinfo,
1241 .get_link = ethtool_op_get_link,
1242 .get_ethtool_stats = netvsc_get_ethtool_stats,
1243 .get_sset_count = netvsc_get_sset_count,
1244 .get_strings = netvsc_get_strings,
1245 .get_channels = netvsc_get_channels,
1246 .set_channels = netvsc_set_channels,
1247 .get_ts_info = ethtool_op_get_ts_info,
1248 .get_rxnfc = netvsc_get_rxnfc,
1249 .get_rxfh_key_size = netvsc_get_rxfh_key_size,
1250 .get_rxfh_indir_size = netvsc_rss_indir_size,
1251 .get_rxfh = netvsc_get_rxfh,
1252 .set_rxfh = netvsc_set_rxfh,
1253 .get_link_ksettings = netvsc_get_link_ksettings,
1254 .set_link_ksettings = netvsc_set_link_ksettings,
1257 static const struct net_device_ops device_ops = {
1258 .ndo_open = netvsc_open,
1259 .ndo_stop = netvsc_close,
1260 .ndo_start_xmit = netvsc_start_xmit,
1261 .ndo_set_rx_mode = netvsc_set_multicast_list,
1262 .ndo_change_mtu = netvsc_change_mtu,
1263 .ndo_validate_addr = eth_validate_addr,
1264 .ndo_set_mac_address = netvsc_set_mac_addr,
1265 .ndo_select_queue = netvsc_select_queue,
1266 .ndo_get_stats64 = netvsc_get_stats64,
1267 #ifdef CONFIG_NET_POLL_CONTROLLER
1268 .ndo_poll_controller = netvsc_poll_controller,
1273 * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1274 * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1275 * present send GARP packet to network peers with netif_notify_peers().
1277 static void netvsc_link_change(struct work_struct *w)
1279 struct net_device_context *ndev_ctx =
1280 container_of(w, struct net_device_context, dwork.work);
1281 struct hv_device *device_obj = ndev_ctx->device_ctx;
1282 struct net_device *net = hv_get_drvdata(device_obj);
1283 struct netvsc_device *net_device;
1284 struct rndis_device *rdev;
1285 struct netvsc_reconfig *event = NULL;
1286 bool notify = false, reschedule = false;
1287 unsigned long flags, next_reconfig, delay;
1290 net_device = rtnl_dereference(ndev_ctx->nvdev);
1294 rdev = net_device->extension;
1296 next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1297 if (time_is_after_jiffies(next_reconfig)) {
1298 /* link_watch only sends one notification with current state
1299 * per second, avoid doing reconfig more frequently. Handle
1302 delay = next_reconfig - jiffies;
1303 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1304 schedule_delayed_work(&ndev_ctx->dwork, delay);
1307 ndev_ctx->last_reconfig = jiffies;
1309 spin_lock_irqsave(&ndev_ctx->lock, flags);
1310 if (!list_empty(&ndev_ctx->reconfig_events)) {
1311 event = list_first_entry(&ndev_ctx->reconfig_events,
1312 struct netvsc_reconfig, list);
1313 list_del(&event->list);
1314 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1316 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1321 switch (event->event) {
1322 /* Only the following events are possible due to the check in
1323 * netvsc_linkstatus_callback()
1325 case RNDIS_STATUS_MEDIA_CONNECT:
1326 if (rdev->link_state) {
1327 rdev->link_state = false;
1328 netif_carrier_on(net);
1329 netif_tx_wake_all_queues(net);
1335 case RNDIS_STATUS_MEDIA_DISCONNECT:
1336 if (!rdev->link_state) {
1337 rdev->link_state = true;
1338 netif_carrier_off(net);
1339 netif_tx_stop_all_queues(net);
1343 case RNDIS_STATUS_NETWORK_CHANGE:
1344 /* Only makes sense if carrier is present */
1345 if (!rdev->link_state) {
1346 rdev->link_state = true;
1347 netif_carrier_off(net);
1348 netif_tx_stop_all_queues(net);
1349 event->event = RNDIS_STATUS_MEDIA_CONNECT;
1350 spin_lock_irqsave(&ndev_ctx->lock, flags);
1351 list_add(&event->list, &ndev_ctx->reconfig_events);
1352 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1361 netdev_notify_peers(net);
1363 /* link_watch only sends one notification with current state per
1364 * second, handle next reconfig event in 2 seconds.
1367 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1375 static struct net_device *get_netvsc_bymac(const u8 *mac)
1377 struct net_device *dev;
1381 for_each_netdev(&init_net, dev) {
1382 if (dev->netdev_ops != &device_ops)
1383 continue; /* not a netvsc device */
1385 if (ether_addr_equal(mac, dev->perm_addr))
1392 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
1394 struct net_device *dev;
1398 for_each_netdev(&init_net, dev) {
1399 struct net_device_context *net_device_ctx;
1401 if (dev->netdev_ops != &device_ops)
1402 continue; /* not a netvsc device */
1404 net_device_ctx = netdev_priv(dev);
1405 if (net_device_ctx->nvdev == NULL)
1406 continue; /* device is removed */
1408 if (rtnl_dereference(net_device_ctx->vf_netdev) == vf_netdev)
1409 return dev; /* a match */
1415 static int netvsc_register_vf(struct net_device *vf_netdev)
1417 struct net_device *ndev;
1418 struct net_device_context *net_device_ctx;
1419 struct netvsc_device *netvsc_dev;
1421 if (vf_netdev->addr_len != ETH_ALEN)
1425 * We will use the MAC address to locate the synthetic interface to
1426 * associate with the VF interface. If we don't find a matching
1427 * synthetic interface, move on.
1429 ndev = get_netvsc_bymac(vf_netdev->perm_addr);
1433 net_device_ctx = netdev_priv(ndev);
1434 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
1435 if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
1438 netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
1440 * Take a reference on the module.
1442 try_module_get(THIS_MODULE);
1444 dev_hold(vf_netdev);
1445 rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
1449 static int netvsc_vf_up(struct net_device *vf_netdev)
1451 struct net_device *ndev;
1452 struct netvsc_device *netvsc_dev;
1453 struct net_device_context *net_device_ctx;
1455 ndev = get_netvsc_byref(vf_netdev);
1459 net_device_ctx = netdev_priv(ndev);
1460 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
1462 netdev_info(ndev, "VF up: %s\n", vf_netdev->name);
1465 * Open the device before switching data path.
1467 rndis_filter_open(netvsc_dev);
1470 * notify the host to switch the data path.
1472 netvsc_switch_datapath(ndev, true);
1473 netdev_info(ndev, "Data path switched to VF: %s\n", vf_netdev->name);
1475 netif_carrier_off(ndev);
1477 /* Now notify peers through VF device. */
1478 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, vf_netdev);
1483 static int netvsc_vf_down(struct net_device *vf_netdev)
1485 struct net_device *ndev;
1486 struct netvsc_device *netvsc_dev;
1487 struct net_device_context *net_device_ctx;
1489 ndev = get_netvsc_byref(vf_netdev);
1493 net_device_ctx = netdev_priv(ndev);
1494 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
1496 netdev_info(ndev, "VF down: %s\n", vf_netdev->name);
1497 netvsc_switch_datapath(ndev, false);
1498 netdev_info(ndev, "Data path switched from VF: %s\n", vf_netdev->name);
1499 rndis_filter_close(netvsc_dev);
1500 netif_carrier_on(ndev);
1502 /* Now notify peers through netvsc device. */
1503 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, ndev);
1508 static int netvsc_unregister_vf(struct net_device *vf_netdev)
1510 struct net_device *ndev;
1511 struct net_device_context *net_device_ctx;
1513 ndev = get_netvsc_byref(vf_netdev);
1517 net_device_ctx = netdev_priv(ndev);
1519 netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
1521 RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
1523 module_put(THIS_MODULE);
1527 static int netvsc_probe(struct hv_device *dev,
1528 const struct hv_vmbus_device_id *dev_id)
1530 struct net_device *net = NULL;
1531 struct net_device_context *net_device_ctx;
1532 struct netvsc_device_info device_info;
1533 struct netvsc_device *nvdev;
1536 net = alloc_etherdev_mq(sizeof(struct net_device_context),
1541 netif_carrier_off(net);
1543 netvsc_init_settings(net);
1545 net_device_ctx = netdev_priv(net);
1546 net_device_ctx->device_ctx = dev;
1547 net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
1548 if (netif_msg_probe(net_device_ctx))
1549 netdev_dbg(net, "netvsc msg_enable: %d\n",
1550 net_device_ctx->msg_enable);
1552 hv_set_drvdata(dev, net);
1554 INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
1555 INIT_WORK(&net_device_ctx->work, do_set_multicast);
1557 spin_lock_init(&net_device_ctx->lock);
1558 INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
1560 net->netdev_ops = &device_ops;
1561 net->ethtool_ops = ðtool_ops;
1562 SET_NETDEV_DEV(net, &dev->device);
1564 /* We always need headroom for rndis header */
1565 net->needed_headroom = RNDIS_AND_PPI_SIZE;
1567 /* Notify the netvsc driver of the new device */
1568 memset(&device_info, 0, sizeof(device_info));
1569 device_info.ring_size = ring_size;
1570 device_info.num_chn = VRSS_CHANNEL_DEFAULT;
1571 ret = rndis_filter_device_add(dev, &device_info);
1573 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
1575 hv_set_drvdata(dev, NULL);
1578 memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
1580 /* hw_features computed in rndis_filter_device_add */
1581 net->features = net->hw_features |
1582 NETIF_F_HIGHDMA | NETIF_F_SG |
1583 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
1584 net->vlan_features = net->features;
1586 /* RCU not necessary here, device not registered */
1587 nvdev = net_device_ctx->nvdev;
1588 netif_set_real_num_tx_queues(net, nvdev->num_chn);
1589 netif_set_real_num_rx_queues(net, nvdev->num_chn);
1591 /* MTU range: 68 - 1500 or 65521 */
1592 net->min_mtu = NETVSC_MTU_MIN;
1593 if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
1594 net->max_mtu = NETVSC_MTU - ETH_HLEN;
1596 net->max_mtu = ETH_DATA_LEN;
1598 ret = register_netdev(net);
1600 pr_err("Unable to register netdev.\n");
1601 rndis_filter_device_remove(dev, nvdev);
1608 static int netvsc_remove(struct hv_device *dev)
1610 struct net_device *net;
1611 struct net_device_context *ndev_ctx;
1613 net = hv_get_drvdata(dev);
1616 dev_err(&dev->device, "No net device to remove\n");
1620 ndev_ctx = netdev_priv(net);
1622 netif_device_detach(net);
1624 cancel_delayed_work_sync(&ndev_ctx->dwork);
1625 cancel_work_sync(&ndev_ctx->work);
1628 * Call to the vsc driver to let it know that the device is being
1629 * removed. Also blocks mtu and channel changes.
1632 rndis_filter_device_remove(dev, ndev_ctx->nvdev);
1635 unregister_netdev(net);
1637 hv_set_drvdata(dev, NULL);
1643 static const struct hv_vmbus_device_id id_table[] = {
1649 MODULE_DEVICE_TABLE(vmbus, id_table);
1651 /* The one and only one */
1652 static struct hv_driver netvsc_drv = {
1653 .name = KBUILD_MODNAME,
1654 .id_table = id_table,
1655 .probe = netvsc_probe,
1656 .remove = netvsc_remove,
1660 * On Hyper-V, every VF interface is matched with a corresponding
1661 * synthetic interface. The synthetic interface is presented first
1662 * to the guest. When the corresponding VF instance is registered,
1663 * we will take care of switching the data path.
1665 static int netvsc_netdev_event(struct notifier_block *this,
1666 unsigned long event, void *ptr)
1668 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
1670 /* Skip our own events */
1671 if (event_dev->netdev_ops == &device_ops)
1674 /* Avoid non-Ethernet type devices */
1675 if (event_dev->type != ARPHRD_ETHER)
1678 /* Avoid Vlan dev with same MAC registering as VF */
1679 if (is_vlan_dev(event_dev))
1682 /* Avoid Bonding master dev with same MAC registering as VF */
1683 if ((event_dev->priv_flags & IFF_BONDING) &&
1684 (event_dev->flags & IFF_MASTER))
1688 case NETDEV_REGISTER:
1689 return netvsc_register_vf(event_dev);
1690 case NETDEV_UNREGISTER:
1691 return netvsc_unregister_vf(event_dev);
1693 return netvsc_vf_up(event_dev);
1695 return netvsc_vf_down(event_dev);
1701 static struct notifier_block netvsc_netdev_notifier = {
1702 .notifier_call = netvsc_netdev_event,
1705 static void __exit netvsc_drv_exit(void)
1707 unregister_netdevice_notifier(&netvsc_netdev_notifier);
1708 vmbus_driver_unregister(&netvsc_drv);
1711 static int __init netvsc_drv_init(void)
1715 if (ring_size < RING_SIZE_MIN) {
1716 ring_size = RING_SIZE_MIN;
1717 pr_info("Increased ring_size to %d (min allowed)\n",
1720 ret = vmbus_driver_register(&netvsc_drv);
1725 register_netdevice_notifier(&netvsc_netdev_notifier);
1729 MODULE_LICENSE("GPL");
1730 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
1732 module_init(netvsc_drv_init);
1733 module_exit(netvsc_drv_exit);