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/pci.h>
33 #include <linux/skbuff.h>
34 #include <linux/if_vlan.h>
36 #include <linux/slab.h>
37 #include <linux/rtnetlink.h>
38 #include <linux/netpoll.h>
41 #include <net/route.h>
43 #include <net/pkt_sched.h>
44 #include <net/checksum.h>
45 #include <net/ip6_checksum.h>
47 #include "hyperv_net.h"
49 #define RING_SIZE_MIN 64
50 #define RETRY_US_LO 5000
51 #define RETRY_US_HI 10000
52 #define RETRY_MAX 2000 /* >10 sec */
54 #define LINKCHANGE_INT (2 * HZ)
55 #define VF_TAKEOVER_INT (HZ / 10)
57 static unsigned int ring_size __ro_after_init = 128;
58 module_param(ring_size, uint, 0444);
59 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
60 unsigned int netvsc_ring_bytes __ro_after_init;
62 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
63 NETIF_MSG_LINK | NETIF_MSG_IFUP |
64 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
67 static int debug = -1;
68 module_param(debug, int, 0444);
69 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
71 static LIST_HEAD(netvsc_dev_list);
73 static void netvsc_change_rx_flags(struct net_device *net, int change)
75 struct net_device_context *ndev_ctx = netdev_priv(net);
76 struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
82 if (change & IFF_PROMISC) {
83 inc = (net->flags & IFF_PROMISC) ? 1 : -1;
84 dev_set_promiscuity(vf_netdev, inc);
87 if (change & IFF_ALLMULTI) {
88 inc = (net->flags & IFF_ALLMULTI) ? 1 : -1;
89 dev_set_allmulti(vf_netdev, inc);
93 static void netvsc_set_rx_mode(struct net_device *net)
95 struct net_device_context *ndev_ctx = netdev_priv(net);
96 struct net_device *vf_netdev;
97 struct netvsc_device *nvdev;
100 vf_netdev = rcu_dereference(ndev_ctx->vf_netdev);
102 dev_uc_sync(vf_netdev, net);
103 dev_mc_sync(vf_netdev, net);
106 nvdev = rcu_dereference(ndev_ctx->nvdev);
108 rndis_filter_update(nvdev);
112 static void netvsc_tx_enable(struct netvsc_device *nvscdev,
113 struct net_device *ndev)
115 nvscdev->tx_disable = false;
116 virt_wmb(); /* ensure queue wake up mechanism is on */
118 netif_tx_wake_all_queues(ndev);
121 static int netvsc_open(struct net_device *net)
123 struct net_device_context *ndev_ctx = netdev_priv(net);
124 struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
125 struct netvsc_device *nvdev = rtnl_dereference(ndev_ctx->nvdev);
126 struct rndis_device *rdev;
129 netif_carrier_off(net);
131 /* Open up the device */
132 ret = rndis_filter_open(nvdev);
134 netdev_err(net, "unable to open device (ret %d).\n", ret);
138 rdev = nvdev->extension;
139 if (!rdev->link_state) {
140 netif_carrier_on(net);
141 netvsc_tx_enable(nvdev, net);
145 /* Setting synthetic device up transparently sets
146 * slave as up. If open fails, then slave will be
147 * still be offline (and not used).
149 ret = dev_open(vf_netdev, NULL);
152 "unable to open slave: %s: %d\n",
153 vf_netdev->name, ret);
158 static int netvsc_wait_until_empty(struct netvsc_device *nvdev)
160 unsigned int retry = 0;
163 /* Ensure pending bytes in ring are read */
167 for (i = 0; i < nvdev->num_chn; i++) {
168 struct vmbus_channel *chn
169 = nvdev->chan_table[i].channel;
174 /* make sure receive not running now */
175 napi_synchronize(&nvdev->chan_table[i].napi);
177 aread = hv_get_bytes_to_read(&chn->inbound);
181 aread = hv_get_bytes_to_read(&chn->outbound);
189 if (++retry > RETRY_MAX)
192 usleep_range(RETRY_US_LO, RETRY_US_HI);
196 static void netvsc_tx_disable(struct netvsc_device *nvscdev,
197 struct net_device *ndev)
200 nvscdev->tx_disable = true;
201 virt_wmb(); /* ensure txq will not wake up after stop */
204 netif_tx_disable(ndev);
207 static int netvsc_close(struct net_device *net)
209 struct net_device_context *net_device_ctx = netdev_priv(net);
210 struct net_device *vf_netdev
211 = rtnl_dereference(net_device_ctx->vf_netdev);
212 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
215 netvsc_tx_disable(nvdev, net);
217 /* No need to close rndis filter if it is removed already */
221 ret = rndis_filter_close(nvdev);
223 netdev_err(net, "unable to close device (ret %d).\n", ret);
227 ret = netvsc_wait_until_empty(nvdev);
229 netdev_err(net, "Ring buffer not empty after closing rndis\n");
232 dev_close(vf_netdev);
237 static inline void *init_ppi_data(struct rndis_message *msg,
238 u32 ppi_size, u32 pkt_type)
240 struct rndis_packet *rndis_pkt = &msg->msg.pkt;
241 struct rndis_per_packet_info *ppi;
243 rndis_pkt->data_offset += ppi_size;
244 ppi = (void *)rndis_pkt + rndis_pkt->per_pkt_info_offset
245 + rndis_pkt->per_pkt_info_len;
247 ppi->size = ppi_size;
248 ppi->type = pkt_type;
250 ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
252 rndis_pkt->per_pkt_info_len += ppi_size;
257 /* Azure hosts don't support non-TCP port numbers in hashing for fragmented
258 * packets. We can use ethtool to change UDP hash level when necessary.
260 static inline u32 netvsc_get_hash(
262 const struct net_device_context *ndc)
264 struct flow_keys flow;
265 u32 hash, pkt_proto = 0;
266 static u32 hashrnd __read_mostly;
268 net_get_random_once(&hashrnd, sizeof(hashrnd));
270 if (!skb_flow_dissect_flow_keys(skb, &flow, 0))
273 switch (flow.basic.ip_proto) {
275 if (flow.basic.n_proto == htons(ETH_P_IP))
276 pkt_proto = HV_TCP4_L4HASH;
277 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
278 pkt_proto = HV_TCP6_L4HASH;
283 if (flow.basic.n_proto == htons(ETH_P_IP))
284 pkt_proto = HV_UDP4_L4HASH;
285 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
286 pkt_proto = HV_UDP6_L4HASH;
291 if (pkt_proto & ndc->l4_hash) {
292 return skb_get_hash(skb);
294 if (flow.basic.n_proto == htons(ETH_P_IP))
295 hash = jhash2((u32 *)&flow.addrs.v4addrs, 2, hashrnd);
296 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
297 hash = jhash2((u32 *)&flow.addrs.v6addrs, 8, hashrnd);
301 skb_set_hash(skb, hash, PKT_HASH_TYPE_L3);
307 static inline int netvsc_get_tx_queue(struct net_device *ndev,
308 struct sk_buff *skb, int old_idx)
310 const struct net_device_context *ndc = netdev_priv(ndev);
311 struct sock *sk = skb->sk;
314 q_idx = ndc->tx_table[netvsc_get_hash(skb, ndc) &
315 (VRSS_SEND_TAB_SIZE - 1)];
317 /* If queue index changed record the new value */
318 if (q_idx != old_idx &&
319 sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
320 sk_tx_queue_set(sk, q_idx);
326 * Select queue for transmit.
328 * If a valid queue has already been assigned, then use that.
329 * Otherwise compute tx queue based on hash and the send table.
331 * This is basically similar to default (netdev_pick_tx) with the added step
332 * of using the host send_table when no other queue has been assigned.
334 * TODO support XPS - but get_xps_queue not exported
336 static u16 netvsc_pick_tx(struct net_device *ndev, struct sk_buff *skb)
338 int q_idx = sk_tx_queue_get(skb->sk);
340 if (q_idx < 0 || skb->ooo_okay || q_idx >= ndev->real_num_tx_queues) {
341 /* If forwarding a packet, we use the recorded queue when
342 * available for better cache locality.
344 if (skb_rx_queue_recorded(skb))
345 q_idx = skb_get_rx_queue(skb);
347 q_idx = netvsc_get_tx_queue(ndev, skb, q_idx);
353 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
354 struct net_device *sb_dev)
356 struct net_device_context *ndc = netdev_priv(ndev);
357 struct net_device *vf_netdev;
361 vf_netdev = rcu_dereference(ndc->vf_netdev);
363 const struct net_device_ops *vf_ops = vf_netdev->netdev_ops;
365 if (vf_ops->ndo_select_queue)
366 txq = vf_ops->ndo_select_queue(vf_netdev, skb, sb_dev);
368 txq = netdev_pick_tx(vf_netdev, skb, NULL);
370 /* Record the queue selected by VF so that it can be
371 * used for common case where VF has more queues than
372 * the synthetic device.
374 qdisc_skb_cb(skb)->slave_dev_queue_mapping = txq;
376 txq = netvsc_pick_tx(ndev, skb);
380 while (unlikely(txq >= ndev->real_num_tx_queues))
381 txq -= ndev->real_num_tx_queues;
386 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
387 struct hv_page_buffer *pb)
391 /* Deal with compound pages by ignoring unused part
394 page += (offset >> PAGE_SHIFT);
395 offset &= ~PAGE_MASK;
400 bytes = PAGE_SIZE - offset;
403 pb[j].pfn = page_to_pfn(page);
404 pb[j].offset = offset;
410 if (offset == PAGE_SIZE && len) {
420 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
421 struct hv_netvsc_packet *packet,
422 struct hv_page_buffer *pb)
425 char *data = skb->data;
426 int frags = skb_shinfo(skb)->nr_frags;
429 /* The packet is laid out thus:
430 * 1. hdr: RNDIS header and PPI
432 * 3. skb fragment data
434 slots_used += fill_pg_buf(virt_to_page(hdr),
436 len, &pb[slots_used]);
438 packet->rmsg_size = len;
439 packet->rmsg_pgcnt = slots_used;
441 slots_used += fill_pg_buf(virt_to_page(data),
442 offset_in_page(data),
443 skb_headlen(skb), &pb[slots_used]);
445 for (i = 0; i < frags; i++) {
446 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
448 slots_used += fill_pg_buf(skb_frag_page(frag),
450 skb_frag_size(frag), &pb[slots_used]);
455 static int count_skb_frag_slots(struct sk_buff *skb)
457 int i, frags = skb_shinfo(skb)->nr_frags;
460 for (i = 0; i < frags; i++) {
461 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
462 unsigned long size = skb_frag_size(frag);
463 unsigned long offset = frag->page_offset;
465 /* Skip unused frames from start of page */
466 offset &= ~PAGE_MASK;
467 pages += PFN_UP(offset + size);
472 static int netvsc_get_slots(struct sk_buff *skb)
474 char *data = skb->data;
475 unsigned int offset = offset_in_page(data);
476 unsigned int len = skb_headlen(skb);
480 slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
481 frag_slots = count_skb_frag_slots(skb);
482 return slots + frag_slots;
485 static u32 net_checksum_info(struct sk_buff *skb)
487 if (skb->protocol == htons(ETH_P_IP)) {
488 struct iphdr *ip = ip_hdr(skb);
490 if (ip->protocol == IPPROTO_TCP)
491 return TRANSPORT_INFO_IPV4_TCP;
492 else if (ip->protocol == IPPROTO_UDP)
493 return TRANSPORT_INFO_IPV4_UDP;
495 struct ipv6hdr *ip6 = ipv6_hdr(skb);
497 if (ip6->nexthdr == IPPROTO_TCP)
498 return TRANSPORT_INFO_IPV6_TCP;
499 else if (ip6->nexthdr == IPPROTO_UDP)
500 return TRANSPORT_INFO_IPV6_UDP;
503 return TRANSPORT_INFO_NOT_IP;
506 /* Send skb on the slave VF device. */
507 static int netvsc_vf_xmit(struct net_device *net, struct net_device *vf_netdev,
510 struct net_device_context *ndev_ctx = netdev_priv(net);
511 unsigned int len = skb->len;
514 skb->dev = vf_netdev;
515 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
517 rc = dev_queue_xmit(skb);
518 if (likely(rc == NET_XMIT_SUCCESS || rc == NET_XMIT_CN)) {
519 struct netvsc_vf_pcpu_stats *pcpu_stats
520 = this_cpu_ptr(ndev_ctx->vf_stats);
522 u64_stats_update_begin(&pcpu_stats->syncp);
523 pcpu_stats->tx_packets++;
524 pcpu_stats->tx_bytes += len;
525 u64_stats_update_end(&pcpu_stats->syncp);
527 this_cpu_inc(ndev_ctx->vf_stats->tx_dropped);
533 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
535 struct net_device_context *net_device_ctx = netdev_priv(net);
536 struct hv_netvsc_packet *packet = NULL;
538 unsigned int num_data_pgs;
539 struct rndis_message *rndis_msg;
540 struct net_device *vf_netdev;
543 struct hv_page_buffer pb[MAX_PAGE_BUFFER_COUNT];
545 /* if VF is present and up then redirect packets
546 * already called with rcu_read_lock_bh
548 vf_netdev = rcu_dereference_bh(net_device_ctx->vf_netdev);
549 if (vf_netdev && netif_running(vf_netdev) &&
550 !netpoll_tx_running(net))
551 return netvsc_vf_xmit(net, vf_netdev, skb);
553 /* We will atmost need two pages to describe the rndis
554 * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
555 * of pages in a single packet. If skb is scattered around
556 * more pages we try linearizing it.
559 num_data_pgs = netvsc_get_slots(skb) + 2;
561 if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
562 ++net_device_ctx->eth_stats.tx_scattered;
564 if (skb_linearize(skb))
567 num_data_pgs = netvsc_get_slots(skb) + 2;
568 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
569 ++net_device_ctx->eth_stats.tx_too_big;
575 * Place the rndis header in the skb head room and
576 * the skb->cb will be used for hv_netvsc_packet
579 ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
583 /* Use the skb control buffer for building up the packet */
584 BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
585 FIELD_SIZEOF(struct sk_buff, cb));
586 packet = (struct hv_netvsc_packet *)skb->cb;
588 packet->q_idx = skb_get_queue_mapping(skb);
590 packet->total_data_buflen = skb->len;
591 packet->total_bytes = skb->len;
592 packet->total_packets = 1;
594 rndis_msg = (struct rndis_message *)skb->head;
596 /* Add the rndis header */
597 rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
598 rndis_msg->msg_len = packet->total_data_buflen;
600 rndis_msg->msg.pkt = (struct rndis_packet) {
601 .data_offset = sizeof(struct rndis_packet),
602 .data_len = packet->total_data_buflen,
603 .per_pkt_info_offset = sizeof(struct rndis_packet),
606 rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
608 hash = skb_get_hash_raw(skb);
609 if (hash != 0 && net->real_num_tx_queues > 1) {
612 rndis_msg_size += NDIS_HASH_PPI_SIZE;
613 hash_info = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
618 if (skb_vlan_tag_present(skb)) {
619 struct ndis_pkt_8021q_info *vlan;
621 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
622 vlan = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
626 vlan->vlanid = skb_vlan_tag_get_id(skb);
627 vlan->cfi = skb_vlan_tag_get_cfi(skb);
628 vlan->pri = skb_vlan_tag_get_prio(skb);
631 if (skb_is_gso(skb)) {
632 struct ndis_tcp_lso_info *lso_info;
634 rndis_msg_size += NDIS_LSO_PPI_SIZE;
635 lso_info = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
636 TCP_LARGESEND_PKTINFO);
639 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
640 if (skb->protocol == htons(ETH_P_IP)) {
641 lso_info->lso_v2_transmit.ip_version =
642 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
643 ip_hdr(skb)->tot_len = 0;
644 ip_hdr(skb)->check = 0;
645 tcp_hdr(skb)->check =
646 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
647 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
649 lso_info->lso_v2_transmit.ip_version =
650 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
651 ipv6_hdr(skb)->payload_len = 0;
652 tcp_hdr(skb)->check =
653 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
654 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
656 lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
657 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
658 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
659 if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
660 struct ndis_tcp_ip_checksum_info *csum_info;
662 rndis_msg_size += NDIS_CSUM_PPI_SIZE;
663 csum_info = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
664 TCPIP_CHKSUM_PKTINFO);
666 csum_info->value = 0;
667 csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
669 if (skb->protocol == htons(ETH_P_IP)) {
670 csum_info->transmit.is_ipv4 = 1;
672 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
673 csum_info->transmit.tcp_checksum = 1;
675 csum_info->transmit.udp_checksum = 1;
677 csum_info->transmit.is_ipv6 = 1;
679 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
680 csum_info->transmit.tcp_checksum = 1;
682 csum_info->transmit.udp_checksum = 1;
685 /* Can't do offload of this type of checksum */
686 if (skb_checksum_help(skb))
691 /* Start filling in the page buffers with the rndis hdr */
692 rndis_msg->msg_len += rndis_msg_size;
693 packet->total_data_buflen = rndis_msg->msg_len;
694 packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
697 /* timestamp packet in software */
698 skb_tx_timestamp(skb);
700 ret = netvsc_send(net, packet, rndis_msg, pb, skb);
701 if (likely(ret == 0))
704 if (ret == -EAGAIN) {
705 ++net_device_ctx->eth_stats.tx_busy;
706 return NETDEV_TX_BUSY;
710 ++net_device_ctx->eth_stats.tx_no_space;
713 dev_kfree_skb_any(skb);
714 net->stats.tx_dropped++;
719 ++net_device_ctx->eth_stats.tx_no_memory;
724 * netvsc_linkstatus_callback - Link up/down notification
726 void netvsc_linkstatus_callback(struct net_device *net,
727 struct rndis_message *resp)
729 struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
730 struct net_device_context *ndev_ctx = netdev_priv(net);
731 struct netvsc_reconfig *event;
734 /* Update the physical link speed when changing to another vSwitch */
735 if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
738 speed = *(u32 *)((void *)indicate
739 + indicate->status_buf_offset) / 10000;
740 ndev_ctx->speed = speed;
744 /* Handle these link change statuses below */
745 if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
746 indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
747 indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
750 if (net->reg_state != NETREG_REGISTERED)
753 event = kzalloc(sizeof(*event), GFP_ATOMIC);
756 event->event = indicate->status;
758 spin_lock_irqsave(&ndev_ctx->lock, flags);
759 list_add_tail(&event->list, &ndev_ctx->reconfig_events);
760 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
762 schedule_delayed_work(&ndev_ctx->dwork, 0);
765 static void netvsc_comp_ipcsum(struct sk_buff *skb)
767 struct iphdr *iph = (struct iphdr *)skb->data;
770 iph->check = ip_fast_csum(iph, iph->ihl);
773 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
774 struct netvsc_channel *nvchan)
776 struct napi_struct *napi = &nvchan->napi;
777 const struct ndis_pkt_8021q_info *vlan = nvchan->rsc.vlan;
778 const struct ndis_tcp_ip_checksum_info *csum_info =
779 nvchan->rsc.csum_info;
783 skb = napi_alloc_skb(napi, nvchan->rsc.pktlen);
788 * Copy to skb. This copy is needed here since the memory pointed by
789 * hv_netvsc_packet cannot be deallocated
791 for (i = 0; i < nvchan->rsc.cnt; i++)
792 skb_put_data(skb, nvchan->rsc.data[i], nvchan->rsc.len[i]);
794 skb->protocol = eth_type_trans(skb, net);
796 /* skb is already created with CHECKSUM_NONE */
797 skb_checksum_none_assert(skb);
799 /* Incoming packets may have IP header checksum verified by the host.
800 * They may not have IP header checksum computed after coalescing.
801 * We compute it here if the flags are set, because on Linux, the IP
802 * checksum is always checked.
804 if (csum_info && csum_info->receive.ip_checksum_value_invalid &&
805 csum_info->receive.ip_checksum_succeeded &&
806 skb->protocol == htons(ETH_P_IP))
807 netvsc_comp_ipcsum(skb);
809 /* Do L4 checksum offload if enabled and present.
811 if (csum_info && (net->features & NETIF_F_RXCSUM)) {
812 if (csum_info->receive.tcp_checksum_succeeded ||
813 csum_info->receive.udp_checksum_succeeded)
814 skb->ip_summed = CHECKSUM_UNNECESSARY;
818 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT) |
819 (vlan->cfi ? VLAN_CFI_MASK : 0);
821 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
829 * netvsc_recv_callback - Callback when we receive a packet from the
830 * "wire" on the specified device.
832 int netvsc_recv_callback(struct net_device *net,
833 struct netvsc_device *net_device,
834 struct netvsc_channel *nvchan)
836 struct net_device_context *net_device_ctx = netdev_priv(net);
837 struct vmbus_channel *channel = nvchan->channel;
838 u16 q_idx = channel->offermsg.offer.sub_channel_index;
840 struct netvsc_stats *rx_stats;
842 if (net->reg_state != NETREG_REGISTERED)
843 return NVSP_STAT_FAIL;
845 /* Allocate a skb - TODO direct I/O to pages? */
846 skb = netvsc_alloc_recv_skb(net, nvchan);
848 if (unlikely(!skb)) {
849 ++net_device_ctx->eth_stats.rx_no_memory;
851 return NVSP_STAT_FAIL;
854 skb_record_rx_queue(skb, q_idx);
857 * Even if injecting the packet, record the statistics
858 * on the synthetic device because modifying the VF device
859 * statistics will not work correctly.
861 rx_stats = &nvchan->rx_stats;
862 u64_stats_update_begin(&rx_stats->syncp);
864 rx_stats->bytes += nvchan->rsc.pktlen;
866 if (skb->pkt_type == PACKET_BROADCAST)
867 ++rx_stats->broadcast;
868 else if (skb->pkt_type == PACKET_MULTICAST)
869 ++rx_stats->multicast;
870 u64_stats_update_end(&rx_stats->syncp);
872 napi_gro_receive(&nvchan->napi, skb);
873 return NVSP_STAT_SUCCESS;
876 static void netvsc_get_drvinfo(struct net_device *net,
877 struct ethtool_drvinfo *info)
879 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
880 strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
883 static void netvsc_get_channels(struct net_device *net,
884 struct ethtool_channels *channel)
886 struct net_device_context *net_device_ctx = netdev_priv(net);
887 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
890 channel->max_combined = nvdev->max_chn;
891 channel->combined_count = nvdev->num_chn;
895 /* Alloc struct netvsc_device_info, and initialize it from either existing
896 * struct netvsc_device, or from default values.
898 static struct netvsc_device_info *netvsc_devinfo_get
899 (struct netvsc_device *nvdev)
901 struct netvsc_device_info *dev_info;
903 dev_info = kzalloc(sizeof(*dev_info), GFP_ATOMIC);
909 dev_info->num_chn = nvdev->num_chn;
910 dev_info->send_sections = nvdev->send_section_cnt;
911 dev_info->send_section_size = nvdev->send_section_size;
912 dev_info->recv_sections = nvdev->recv_section_cnt;
913 dev_info->recv_section_size = nvdev->recv_section_size;
915 memcpy(dev_info->rss_key, nvdev->extension->rss_key,
918 dev_info->num_chn = VRSS_CHANNEL_DEFAULT;
919 dev_info->send_sections = NETVSC_DEFAULT_TX;
920 dev_info->send_section_size = NETVSC_SEND_SECTION_SIZE;
921 dev_info->recv_sections = NETVSC_DEFAULT_RX;
922 dev_info->recv_section_size = NETVSC_RECV_SECTION_SIZE;
928 static int netvsc_detach(struct net_device *ndev,
929 struct netvsc_device *nvdev)
931 struct net_device_context *ndev_ctx = netdev_priv(ndev);
932 struct hv_device *hdev = ndev_ctx->device_ctx;
935 /* Don't try continuing to try and setup sub channels */
936 if (cancel_work_sync(&nvdev->subchan_work))
939 /* If device was up (receiving) then shutdown */
940 if (netif_running(ndev)) {
941 netvsc_tx_disable(nvdev, ndev);
943 ret = rndis_filter_close(nvdev);
946 "unable to close device (ret %d).\n", ret);
950 ret = netvsc_wait_until_empty(nvdev);
953 "Ring buffer not empty after closing rndis\n");
958 netif_device_detach(ndev);
960 rndis_filter_device_remove(hdev, nvdev);
965 static int netvsc_attach(struct net_device *ndev,
966 struct netvsc_device_info *dev_info)
968 struct net_device_context *ndev_ctx = netdev_priv(ndev);
969 struct hv_device *hdev = ndev_ctx->device_ctx;
970 struct netvsc_device *nvdev;
971 struct rndis_device *rdev;
974 nvdev = rndis_filter_device_add(hdev, dev_info);
976 return PTR_ERR(nvdev);
978 if (nvdev->num_chn > 1) {
979 ret = rndis_set_subchannel(ndev, nvdev, dev_info);
981 /* if unavailable, just proceed with one queue */
988 /* In any case device is now ready */
989 netif_device_attach(ndev);
991 /* Note: enable and attach happen when sub-channels setup */
992 netif_carrier_off(ndev);
994 if (netif_running(ndev)) {
995 ret = rndis_filter_open(nvdev);
999 rdev = nvdev->extension;
1000 if (!rdev->link_state)
1001 netif_carrier_on(ndev);
1007 static int netvsc_set_channels(struct net_device *net,
1008 struct ethtool_channels *channels)
1010 struct net_device_context *net_device_ctx = netdev_priv(net);
1011 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
1012 unsigned int orig, count = channels->combined_count;
1013 struct netvsc_device_info *device_info;
1016 /* We do not support separate count for rx, tx, or other */
1018 channels->rx_count || channels->tx_count || channels->other_count)
1021 if (!nvdev || nvdev->destroy)
1024 if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
1027 if (count > nvdev->max_chn)
1030 orig = nvdev->num_chn;
1032 device_info = netvsc_devinfo_get(nvdev);
1037 device_info->num_chn = count;
1039 ret = netvsc_detach(net, nvdev);
1043 ret = netvsc_attach(net, device_info);
1045 device_info->num_chn = orig;
1046 if (netvsc_attach(net, device_info))
1047 netdev_err(net, "restoring channel setting failed\n");
1056 netvsc_validate_ethtool_ss_cmd(const struct ethtool_link_ksettings *cmd)
1058 struct ethtool_link_ksettings diff1 = *cmd;
1059 struct ethtool_link_ksettings diff2 = {};
1061 diff1.base.speed = 0;
1062 diff1.base.duplex = 0;
1063 /* advertising and cmd are usually set */
1064 ethtool_link_ksettings_zero_link_mode(&diff1, advertising);
1066 /* We set port to PORT_OTHER */
1067 diff2.base.port = PORT_OTHER;
1069 return !memcmp(&diff1, &diff2, sizeof(diff1));
1072 static void netvsc_init_settings(struct net_device *dev)
1074 struct net_device_context *ndc = netdev_priv(dev);
1076 ndc->l4_hash = HV_DEFAULT_L4HASH;
1078 ndc->speed = SPEED_UNKNOWN;
1079 ndc->duplex = DUPLEX_FULL;
1081 dev->features = NETIF_F_LRO;
1084 static int netvsc_get_link_ksettings(struct net_device *dev,
1085 struct ethtool_link_ksettings *cmd)
1087 struct net_device_context *ndc = netdev_priv(dev);
1089 cmd->base.speed = ndc->speed;
1090 cmd->base.duplex = ndc->duplex;
1091 cmd->base.port = PORT_OTHER;
1096 static int netvsc_set_link_ksettings(struct net_device *dev,
1097 const struct ethtool_link_ksettings *cmd)
1099 struct net_device_context *ndc = netdev_priv(dev);
1102 speed = cmd->base.speed;
1103 if (!ethtool_validate_speed(speed) ||
1104 !ethtool_validate_duplex(cmd->base.duplex) ||
1105 !netvsc_validate_ethtool_ss_cmd(cmd))
1109 ndc->duplex = cmd->base.duplex;
1114 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
1116 struct net_device_context *ndevctx = netdev_priv(ndev);
1117 struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1118 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1119 int orig_mtu = ndev->mtu;
1120 struct netvsc_device_info *device_info;
1123 if (!nvdev || nvdev->destroy)
1126 device_info = netvsc_devinfo_get(nvdev);
1131 /* Change MTU of underlying VF netdev first. */
1133 ret = dev_set_mtu(vf_netdev, mtu);
1138 ret = netvsc_detach(ndev, nvdev);
1144 ret = netvsc_attach(ndev, device_info);
1148 /* Attempt rollback to original MTU */
1149 ndev->mtu = orig_mtu;
1151 if (netvsc_attach(ndev, device_info))
1152 netdev_err(ndev, "restoring mtu failed\n");
1155 dev_set_mtu(vf_netdev, orig_mtu);
1162 static void netvsc_get_vf_stats(struct net_device *net,
1163 struct netvsc_vf_pcpu_stats *tot)
1165 struct net_device_context *ndev_ctx = netdev_priv(net);
1168 memset(tot, 0, sizeof(*tot));
1170 for_each_possible_cpu(i) {
1171 const struct netvsc_vf_pcpu_stats *stats
1172 = per_cpu_ptr(ndev_ctx->vf_stats, i);
1173 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1177 start = u64_stats_fetch_begin_irq(&stats->syncp);
1178 rx_packets = stats->rx_packets;
1179 tx_packets = stats->tx_packets;
1180 rx_bytes = stats->rx_bytes;
1181 tx_bytes = stats->tx_bytes;
1182 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1184 tot->rx_packets += rx_packets;
1185 tot->tx_packets += tx_packets;
1186 tot->rx_bytes += rx_bytes;
1187 tot->tx_bytes += tx_bytes;
1188 tot->tx_dropped += stats->tx_dropped;
1192 static void netvsc_get_pcpu_stats(struct net_device *net,
1193 struct netvsc_ethtool_pcpu_stats *pcpu_tot)
1195 struct net_device_context *ndev_ctx = netdev_priv(net);
1196 struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1199 /* fetch percpu stats of vf */
1200 for_each_possible_cpu(i) {
1201 const struct netvsc_vf_pcpu_stats *stats =
1202 per_cpu_ptr(ndev_ctx->vf_stats, i);
1203 struct netvsc_ethtool_pcpu_stats *this_tot = &pcpu_tot[i];
1207 start = u64_stats_fetch_begin_irq(&stats->syncp);
1208 this_tot->vf_rx_packets = stats->rx_packets;
1209 this_tot->vf_tx_packets = stats->tx_packets;
1210 this_tot->vf_rx_bytes = stats->rx_bytes;
1211 this_tot->vf_tx_bytes = stats->tx_bytes;
1212 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1213 this_tot->rx_packets = this_tot->vf_rx_packets;
1214 this_tot->tx_packets = this_tot->vf_tx_packets;
1215 this_tot->rx_bytes = this_tot->vf_rx_bytes;
1216 this_tot->tx_bytes = this_tot->vf_tx_bytes;
1219 /* fetch percpu stats of netvsc */
1220 for (i = 0; i < nvdev->num_chn; i++) {
1221 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1222 const struct netvsc_stats *stats;
1223 struct netvsc_ethtool_pcpu_stats *this_tot =
1224 &pcpu_tot[nvchan->channel->target_cpu];
1228 stats = &nvchan->tx_stats;
1230 start = u64_stats_fetch_begin_irq(&stats->syncp);
1231 packets = stats->packets;
1232 bytes = stats->bytes;
1233 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1235 this_tot->tx_bytes += bytes;
1236 this_tot->tx_packets += packets;
1238 stats = &nvchan->rx_stats;
1240 start = u64_stats_fetch_begin_irq(&stats->syncp);
1241 packets = stats->packets;
1242 bytes = stats->bytes;
1243 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1245 this_tot->rx_bytes += bytes;
1246 this_tot->rx_packets += packets;
1250 static void netvsc_get_stats64(struct net_device *net,
1251 struct rtnl_link_stats64 *t)
1253 struct net_device_context *ndev_ctx = netdev_priv(net);
1254 struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1255 struct netvsc_vf_pcpu_stats vf_tot;
1261 netdev_stats_to_stats64(t, &net->stats);
1263 netvsc_get_vf_stats(net, &vf_tot);
1264 t->rx_packets += vf_tot.rx_packets;
1265 t->tx_packets += vf_tot.tx_packets;
1266 t->rx_bytes += vf_tot.rx_bytes;
1267 t->tx_bytes += vf_tot.tx_bytes;
1268 t->tx_dropped += vf_tot.tx_dropped;
1270 for (i = 0; i < nvdev->num_chn; i++) {
1271 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1272 const struct netvsc_stats *stats;
1273 u64 packets, bytes, multicast;
1276 stats = &nvchan->tx_stats;
1278 start = u64_stats_fetch_begin_irq(&stats->syncp);
1279 packets = stats->packets;
1280 bytes = stats->bytes;
1281 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1283 t->tx_bytes += bytes;
1284 t->tx_packets += packets;
1286 stats = &nvchan->rx_stats;
1288 start = u64_stats_fetch_begin_irq(&stats->syncp);
1289 packets = stats->packets;
1290 bytes = stats->bytes;
1291 multicast = stats->multicast + stats->broadcast;
1292 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1294 t->rx_bytes += bytes;
1295 t->rx_packets += packets;
1296 t->multicast += multicast;
1300 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1302 struct net_device_context *ndc = netdev_priv(ndev);
1303 struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1304 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1305 struct sockaddr *addr = p;
1308 err = eth_prepare_mac_addr_change(ndev, p);
1316 err = dev_set_mac_address(vf_netdev, addr, NULL);
1321 err = rndis_filter_set_device_mac(nvdev, addr->sa_data);
1323 eth_commit_mac_addr_change(ndev, p);
1324 } else if (vf_netdev) {
1325 /* rollback change on VF */
1326 memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN);
1327 dev_set_mac_address(vf_netdev, addr, NULL);
1333 static const struct {
1334 char name[ETH_GSTRING_LEN];
1336 } netvsc_stats[] = {
1337 { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1338 { "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1339 { "tx_no_space", offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1340 { "tx_too_big", offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1341 { "tx_busy", offsetof(struct netvsc_ethtool_stats, tx_busy) },
1342 { "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) },
1343 { "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) },
1344 { "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) },
1345 { "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) },
1346 { "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) },
1348 { "cpu%u_rx_packets",
1349 offsetof(struct netvsc_ethtool_pcpu_stats, rx_packets) },
1351 offsetof(struct netvsc_ethtool_pcpu_stats, rx_bytes) },
1352 { "cpu%u_tx_packets",
1353 offsetof(struct netvsc_ethtool_pcpu_stats, tx_packets) },
1355 offsetof(struct netvsc_ethtool_pcpu_stats, tx_bytes) },
1356 { "cpu%u_vf_rx_packets",
1357 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_packets) },
1358 { "cpu%u_vf_rx_bytes",
1359 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_bytes) },
1360 { "cpu%u_vf_tx_packets",
1361 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_packets) },
1362 { "cpu%u_vf_tx_bytes",
1363 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_bytes) },
1365 { "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) },
1366 { "vf_rx_bytes", offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) },
1367 { "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) },
1368 { "vf_tx_bytes", offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) },
1369 { "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) },
1372 #define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats)
1373 #define NETVSC_VF_STATS_LEN ARRAY_SIZE(vf_stats)
1375 /* statistics per queue (rx/tx packets/bytes) */
1376 #define NETVSC_PCPU_STATS_LEN (num_present_cpus() * ARRAY_SIZE(pcpu_stats))
1378 /* 4 statistics per queue (rx/tx packets/bytes) */
1379 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 4)
1381 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1383 struct net_device_context *ndc = netdev_priv(dev);
1384 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1389 switch (string_set) {
1391 return NETVSC_GLOBAL_STATS_LEN
1392 + NETVSC_VF_STATS_LEN
1393 + NETVSC_QUEUE_STATS_LEN(nvdev)
1394 + NETVSC_PCPU_STATS_LEN;
1400 static void netvsc_get_ethtool_stats(struct net_device *dev,
1401 struct ethtool_stats *stats, u64 *data)
1403 struct net_device_context *ndc = netdev_priv(dev);
1404 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1405 const void *nds = &ndc->eth_stats;
1406 const struct netvsc_stats *qstats;
1407 struct netvsc_vf_pcpu_stats sum;
1408 struct netvsc_ethtool_pcpu_stats *pcpu_sum;
1416 for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1417 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1419 netvsc_get_vf_stats(dev, &sum);
1420 for (j = 0; j < NETVSC_VF_STATS_LEN; j++)
1421 data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset);
1423 for (j = 0; j < nvdev->num_chn; j++) {
1424 qstats = &nvdev->chan_table[j].tx_stats;
1427 start = u64_stats_fetch_begin_irq(&qstats->syncp);
1428 packets = qstats->packets;
1429 bytes = qstats->bytes;
1430 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1431 data[i++] = packets;
1434 qstats = &nvdev->chan_table[j].rx_stats;
1436 start = u64_stats_fetch_begin_irq(&qstats->syncp);
1437 packets = qstats->packets;
1438 bytes = qstats->bytes;
1439 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1440 data[i++] = packets;
1444 pcpu_sum = kvmalloc_array(num_possible_cpus(),
1445 sizeof(struct netvsc_ethtool_pcpu_stats),
1447 netvsc_get_pcpu_stats(dev, pcpu_sum);
1448 for_each_present_cpu(cpu) {
1449 struct netvsc_ethtool_pcpu_stats *this_sum = &pcpu_sum[cpu];
1451 for (j = 0; j < ARRAY_SIZE(pcpu_stats); j++)
1452 data[i++] = *(u64 *)((void *)this_sum
1453 + pcpu_stats[j].offset);
1458 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1460 struct net_device_context *ndc = netdev_priv(dev);
1461 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1468 switch (stringset) {
1470 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++) {
1471 memcpy(p, netvsc_stats[i].name, ETH_GSTRING_LEN);
1472 p += ETH_GSTRING_LEN;
1475 for (i = 0; i < ARRAY_SIZE(vf_stats); i++) {
1476 memcpy(p, vf_stats[i].name, ETH_GSTRING_LEN);
1477 p += ETH_GSTRING_LEN;
1480 for (i = 0; i < nvdev->num_chn; i++) {
1481 sprintf(p, "tx_queue_%u_packets", i);
1482 p += ETH_GSTRING_LEN;
1483 sprintf(p, "tx_queue_%u_bytes", i);
1484 p += ETH_GSTRING_LEN;
1485 sprintf(p, "rx_queue_%u_packets", i);
1486 p += ETH_GSTRING_LEN;
1487 sprintf(p, "rx_queue_%u_bytes", i);
1488 p += ETH_GSTRING_LEN;
1491 for_each_present_cpu(cpu) {
1492 for (i = 0; i < ARRAY_SIZE(pcpu_stats); i++) {
1493 sprintf(p, pcpu_stats[i].name, cpu);
1494 p += ETH_GSTRING_LEN;
1503 netvsc_get_rss_hash_opts(struct net_device_context *ndc,
1504 struct ethtool_rxnfc *info)
1506 const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3;
1508 info->data = RXH_IP_SRC | RXH_IP_DST;
1510 switch (info->flow_type) {
1512 if (ndc->l4_hash & HV_TCP4_L4HASH)
1513 info->data |= l4_flag;
1518 if (ndc->l4_hash & HV_TCP6_L4HASH)
1519 info->data |= l4_flag;
1524 if (ndc->l4_hash & HV_UDP4_L4HASH)
1525 info->data |= l4_flag;
1530 if (ndc->l4_hash & HV_UDP6_L4HASH)
1531 info->data |= l4_flag;
1547 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1550 struct net_device_context *ndc = netdev_priv(dev);
1551 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1556 switch (info->cmd) {
1557 case ETHTOOL_GRXRINGS:
1558 info->data = nvdev->num_chn;
1562 return netvsc_get_rss_hash_opts(ndc, info);
1567 static int netvsc_set_rss_hash_opts(struct net_device_context *ndc,
1568 struct ethtool_rxnfc *info)
1570 if (info->data == (RXH_IP_SRC | RXH_IP_DST |
1571 RXH_L4_B_0_1 | RXH_L4_B_2_3)) {
1572 switch (info->flow_type) {
1574 ndc->l4_hash |= HV_TCP4_L4HASH;
1578 ndc->l4_hash |= HV_TCP6_L4HASH;
1582 ndc->l4_hash |= HV_UDP4_L4HASH;
1586 ndc->l4_hash |= HV_UDP6_L4HASH;
1596 if (info->data == (RXH_IP_SRC | RXH_IP_DST)) {
1597 switch (info->flow_type) {
1599 ndc->l4_hash &= ~HV_TCP4_L4HASH;
1603 ndc->l4_hash &= ~HV_TCP6_L4HASH;
1607 ndc->l4_hash &= ~HV_UDP4_L4HASH;
1611 ndc->l4_hash &= ~HV_UDP6_L4HASH;
1625 netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info)
1627 struct net_device_context *ndc = netdev_priv(ndev);
1629 if (info->cmd == ETHTOOL_SRXFH)
1630 return netvsc_set_rss_hash_opts(ndc, info);
1635 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1637 return NETVSC_HASH_KEYLEN;
1640 static u32 netvsc_rss_indir_size(struct net_device *dev)
1645 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1648 struct net_device_context *ndc = netdev_priv(dev);
1649 struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1650 struct rndis_device *rndis_dev;
1657 *hfunc = ETH_RSS_HASH_TOP; /* Toeplitz */
1659 rndis_dev = ndev->extension;
1661 for (i = 0; i < ITAB_NUM; i++)
1662 indir[i] = rndis_dev->rx_table[i];
1666 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1671 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1672 const u8 *key, const u8 hfunc)
1674 struct net_device_context *ndc = netdev_priv(dev);
1675 struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1676 struct rndis_device *rndis_dev;
1682 if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1685 rndis_dev = ndev->extension;
1687 for (i = 0; i < ITAB_NUM; i++)
1688 if (indir[i] >= ndev->num_chn)
1691 for (i = 0; i < ITAB_NUM; i++)
1692 rndis_dev->rx_table[i] = indir[i];
1699 key = rndis_dev->rss_key;
1702 return rndis_filter_set_rss_param(rndis_dev, key);
1705 /* Hyper-V RNDIS protocol does not have ring in the HW sense.
1706 * It does have pre-allocated receive area which is divided into sections.
1708 static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
1709 struct ethtool_ringparam *ring)
1713 ring->rx_pending = nvdev->recv_section_cnt;
1714 ring->tx_pending = nvdev->send_section_cnt;
1716 if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
1717 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
1719 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1721 ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
1722 ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
1723 / nvdev->send_section_size;
1726 static void netvsc_get_ringparam(struct net_device *ndev,
1727 struct ethtool_ringparam *ring)
1729 struct net_device_context *ndevctx = netdev_priv(ndev);
1730 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1735 __netvsc_get_ringparam(nvdev, ring);
1738 static int netvsc_set_ringparam(struct net_device *ndev,
1739 struct ethtool_ringparam *ring)
1741 struct net_device_context *ndevctx = netdev_priv(ndev);
1742 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1743 struct netvsc_device_info *device_info;
1744 struct ethtool_ringparam orig;
1748 if (!nvdev || nvdev->destroy)
1751 memset(&orig, 0, sizeof(orig));
1752 __netvsc_get_ringparam(nvdev, &orig);
1754 new_tx = clamp_t(u32, ring->tx_pending,
1755 NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
1756 new_rx = clamp_t(u32, ring->rx_pending,
1757 NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
1759 if (new_tx == orig.tx_pending &&
1760 new_rx == orig.rx_pending)
1761 return 0; /* no change */
1763 device_info = netvsc_devinfo_get(nvdev);
1768 device_info->send_sections = new_tx;
1769 device_info->recv_sections = new_rx;
1771 ret = netvsc_detach(ndev, nvdev);
1775 ret = netvsc_attach(ndev, device_info);
1777 device_info->send_sections = orig.tx_pending;
1778 device_info->recv_sections = orig.rx_pending;
1780 if (netvsc_attach(ndev, device_info))
1781 netdev_err(ndev, "restoring ringparam failed");
1789 static int netvsc_set_features(struct net_device *ndev,
1790 netdev_features_t features)
1792 netdev_features_t change = features ^ ndev->features;
1793 struct net_device_context *ndevctx = netdev_priv(ndev);
1794 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1795 struct ndis_offload_params offloads;
1797 if (!nvdev || nvdev->destroy)
1800 if (!(change & NETIF_F_LRO))
1803 memset(&offloads, 0, sizeof(struct ndis_offload_params));
1805 if (features & NETIF_F_LRO) {
1806 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1807 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1809 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1810 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1813 return rndis_filter_set_offload_params(ndev, nvdev, &offloads);
1816 static u32 netvsc_get_msglevel(struct net_device *ndev)
1818 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1820 return ndev_ctx->msg_enable;
1823 static void netvsc_set_msglevel(struct net_device *ndev, u32 val)
1825 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1827 ndev_ctx->msg_enable = val;
1830 static const struct ethtool_ops ethtool_ops = {
1831 .get_drvinfo = netvsc_get_drvinfo,
1832 .get_msglevel = netvsc_get_msglevel,
1833 .set_msglevel = netvsc_set_msglevel,
1834 .get_link = ethtool_op_get_link,
1835 .get_ethtool_stats = netvsc_get_ethtool_stats,
1836 .get_sset_count = netvsc_get_sset_count,
1837 .get_strings = netvsc_get_strings,
1838 .get_channels = netvsc_get_channels,
1839 .set_channels = netvsc_set_channels,
1840 .get_ts_info = ethtool_op_get_ts_info,
1841 .get_rxnfc = netvsc_get_rxnfc,
1842 .set_rxnfc = netvsc_set_rxnfc,
1843 .get_rxfh_key_size = netvsc_get_rxfh_key_size,
1844 .get_rxfh_indir_size = netvsc_rss_indir_size,
1845 .get_rxfh = netvsc_get_rxfh,
1846 .set_rxfh = netvsc_set_rxfh,
1847 .get_link_ksettings = netvsc_get_link_ksettings,
1848 .set_link_ksettings = netvsc_set_link_ksettings,
1849 .get_ringparam = netvsc_get_ringparam,
1850 .set_ringparam = netvsc_set_ringparam,
1853 static const struct net_device_ops device_ops = {
1854 .ndo_open = netvsc_open,
1855 .ndo_stop = netvsc_close,
1856 .ndo_start_xmit = netvsc_start_xmit,
1857 .ndo_change_rx_flags = netvsc_change_rx_flags,
1858 .ndo_set_rx_mode = netvsc_set_rx_mode,
1859 .ndo_set_features = netvsc_set_features,
1860 .ndo_change_mtu = netvsc_change_mtu,
1861 .ndo_validate_addr = eth_validate_addr,
1862 .ndo_set_mac_address = netvsc_set_mac_addr,
1863 .ndo_select_queue = netvsc_select_queue,
1864 .ndo_get_stats64 = netvsc_get_stats64,
1868 * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1869 * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1870 * present send GARP packet to network peers with netif_notify_peers().
1872 static void netvsc_link_change(struct work_struct *w)
1874 struct net_device_context *ndev_ctx =
1875 container_of(w, struct net_device_context, dwork.work);
1876 struct hv_device *device_obj = ndev_ctx->device_ctx;
1877 struct net_device *net = hv_get_drvdata(device_obj);
1878 struct netvsc_device *net_device;
1879 struct rndis_device *rdev;
1880 struct netvsc_reconfig *event = NULL;
1881 bool notify = false, reschedule = false;
1882 unsigned long flags, next_reconfig, delay;
1884 /* if changes are happening, comeback later */
1885 if (!rtnl_trylock()) {
1886 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1890 net_device = rtnl_dereference(ndev_ctx->nvdev);
1894 rdev = net_device->extension;
1896 next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1897 if (time_is_after_jiffies(next_reconfig)) {
1898 /* link_watch only sends one notification with current state
1899 * per second, avoid doing reconfig more frequently. Handle
1902 delay = next_reconfig - jiffies;
1903 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1904 schedule_delayed_work(&ndev_ctx->dwork, delay);
1907 ndev_ctx->last_reconfig = jiffies;
1909 spin_lock_irqsave(&ndev_ctx->lock, flags);
1910 if (!list_empty(&ndev_ctx->reconfig_events)) {
1911 event = list_first_entry(&ndev_ctx->reconfig_events,
1912 struct netvsc_reconfig, list);
1913 list_del(&event->list);
1914 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1916 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1921 switch (event->event) {
1922 /* Only the following events are possible due to the check in
1923 * netvsc_linkstatus_callback()
1925 case RNDIS_STATUS_MEDIA_CONNECT:
1926 if (rdev->link_state) {
1927 rdev->link_state = false;
1928 netif_carrier_on(net);
1929 netvsc_tx_enable(net_device, net);
1935 case RNDIS_STATUS_MEDIA_DISCONNECT:
1936 if (!rdev->link_state) {
1937 rdev->link_state = true;
1938 netif_carrier_off(net);
1939 netvsc_tx_disable(net_device, net);
1943 case RNDIS_STATUS_NETWORK_CHANGE:
1944 /* Only makes sense if carrier is present */
1945 if (!rdev->link_state) {
1946 rdev->link_state = true;
1947 netif_carrier_off(net);
1948 netvsc_tx_disable(net_device, net);
1949 event->event = RNDIS_STATUS_MEDIA_CONNECT;
1950 spin_lock_irqsave(&ndev_ctx->lock, flags);
1951 list_add(&event->list, &ndev_ctx->reconfig_events);
1952 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1961 netdev_notify_peers(net);
1963 /* link_watch only sends one notification with current state per
1964 * second, handle next reconfig event in 2 seconds.
1967 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1975 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
1977 struct net_device_context *net_device_ctx;
1978 struct net_device *dev;
1980 dev = netdev_master_upper_dev_get(vf_netdev);
1981 if (!dev || dev->netdev_ops != &device_ops)
1982 return NULL; /* not a netvsc device */
1984 net_device_ctx = netdev_priv(dev);
1985 if (!rtnl_dereference(net_device_ctx->nvdev))
1986 return NULL; /* device is removed */
1991 /* Called when VF is injecting data into network stack.
1992 * Change the associated network device from VF to netvsc.
1993 * note: already called with rcu_read_lock
1995 static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
1997 struct sk_buff *skb = *pskb;
1998 struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data);
1999 struct net_device_context *ndev_ctx = netdev_priv(ndev);
2000 struct netvsc_vf_pcpu_stats *pcpu_stats
2001 = this_cpu_ptr(ndev_ctx->vf_stats);
2005 u64_stats_update_begin(&pcpu_stats->syncp);
2006 pcpu_stats->rx_packets++;
2007 pcpu_stats->rx_bytes += skb->len;
2008 u64_stats_update_end(&pcpu_stats->syncp);
2010 return RX_HANDLER_ANOTHER;
2013 static int netvsc_vf_join(struct net_device *vf_netdev,
2014 struct net_device *ndev)
2016 struct net_device_context *ndev_ctx = netdev_priv(ndev);
2019 ret = netdev_rx_handler_register(vf_netdev,
2020 netvsc_vf_handle_frame, ndev);
2022 netdev_err(vf_netdev,
2023 "can not register netvsc VF receive handler (err = %d)\n",
2025 goto rx_handler_failed;
2028 ret = netdev_master_upper_dev_link(vf_netdev, ndev,
2031 netdev_err(vf_netdev,
2032 "can not set master device %s (err = %d)\n",
2034 goto upper_link_failed;
2037 /* set slave flag before open to prevent IPv6 addrconf */
2038 vf_netdev->flags |= IFF_SLAVE;
2040 schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
2042 call_netdevice_notifiers(NETDEV_JOIN, vf_netdev);
2044 netdev_info(vf_netdev, "joined to %s\n", ndev->name);
2048 netdev_rx_handler_unregister(vf_netdev);
2053 static void __netvsc_vf_setup(struct net_device *ndev,
2054 struct net_device *vf_netdev)
2058 /* Align MTU of VF with master */
2059 ret = dev_set_mtu(vf_netdev, ndev->mtu);
2061 netdev_warn(vf_netdev,
2062 "unable to change mtu to %u\n", ndev->mtu);
2064 /* set multicast etc flags on VF */
2065 dev_change_flags(vf_netdev, ndev->flags | IFF_SLAVE, NULL);
2067 /* sync address list from ndev to VF */
2068 netif_addr_lock_bh(ndev);
2069 dev_uc_sync(vf_netdev, ndev);
2070 dev_mc_sync(vf_netdev, ndev);
2071 netif_addr_unlock_bh(ndev);
2073 if (netif_running(ndev)) {
2074 ret = dev_open(vf_netdev, NULL);
2076 netdev_warn(vf_netdev,
2077 "unable to open: %d\n", ret);
2081 /* Setup VF as slave of the synthetic device.
2082 * Runs in workqueue to avoid recursion in netlink callbacks.
2084 static void netvsc_vf_setup(struct work_struct *w)
2086 struct net_device_context *ndev_ctx
2087 = container_of(w, struct net_device_context, vf_takeover.work);
2088 struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2089 struct net_device *vf_netdev;
2091 if (!rtnl_trylock()) {
2092 schedule_delayed_work(&ndev_ctx->vf_takeover, 0);
2096 vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2098 __netvsc_vf_setup(ndev, vf_netdev);
2103 /* Find netvsc by VF serial number.
2104 * The PCI hyperv controller records the serial number as the slot kobj name.
2106 static struct net_device *get_netvsc_byslot(const struct net_device *vf_netdev)
2108 struct device *parent = vf_netdev->dev.parent;
2109 struct net_device_context *ndev_ctx;
2110 struct pci_dev *pdev;
2113 if (!parent || !dev_is_pci(parent))
2114 return NULL; /* not a PCI device */
2116 pdev = to_pci_dev(parent);
2118 netdev_notice(vf_netdev, "no PCI slot information\n");
2122 if (kstrtou32(pci_slot_name(pdev->slot), 10, &serial)) {
2123 netdev_notice(vf_netdev, "Invalid vf serial:%s\n",
2124 pci_slot_name(pdev->slot));
2128 list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
2129 if (!ndev_ctx->vf_alloc)
2132 if (ndev_ctx->vf_serial == serial)
2133 return hv_get_drvdata(ndev_ctx->device_ctx);
2136 netdev_notice(vf_netdev,
2137 "no netdev found for vf serial:%u\n", serial);
2141 static int netvsc_register_vf(struct net_device *vf_netdev)
2143 struct net_device_context *net_device_ctx;
2144 struct netvsc_device *netvsc_dev;
2145 struct net_device *ndev;
2148 if (vf_netdev->addr_len != ETH_ALEN)
2151 ndev = get_netvsc_byslot(vf_netdev);
2155 net_device_ctx = netdev_priv(ndev);
2156 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2157 if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
2160 /* if synthetic interface is a different namespace,
2161 * then move the VF to that namespace; join will be
2162 * done again in that context.
2164 if (!net_eq(dev_net(ndev), dev_net(vf_netdev))) {
2165 ret = dev_change_net_namespace(vf_netdev,
2166 dev_net(ndev), "eth%d");
2168 netdev_err(vf_netdev,
2169 "could not move to same namespace as %s: %d\n",
2172 netdev_info(vf_netdev,
2173 "VF moved to namespace with: %s\n",
2178 netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
2180 if (netvsc_vf_join(vf_netdev, ndev) != 0)
2183 dev_hold(vf_netdev);
2184 rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
2188 /* VF up/down change detected, schedule to change data path */
2189 static int netvsc_vf_changed(struct net_device *vf_netdev)
2191 struct net_device_context *net_device_ctx;
2192 struct netvsc_device *netvsc_dev;
2193 struct net_device *ndev;
2194 bool vf_is_up = netif_running(vf_netdev);
2196 ndev = get_netvsc_byref(vf_netdev);
2200 net_device_ctx = netdev_priv(ndev);
2201 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2205 netvsc_switch_datapath(ndev, vf_is_up);
2206 netdev_info(ndev, "Data path switched %s VF: %s\n",
2207 vf_is_up ? "to" : "from", vf_netdev->name);
2212 static int netvsc_unregister_vf(struct net_device *vf_netdev)
2214 struct net_device *ndev;
2215 struct net_device_context *net_device_ctx;
2217 ndev = get_netvsc_byref(vf_netdev);
2221 net_device_ctx = netdev_priv(ndev);
2222 cancel_delayed_work_sync(&net_device_ctx->vf_takeover);
2224 netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
2226 netdev_rx_handler_unregister(vf_netdev);
2227 netdev_upper_dev_unlink(vf_netdev, ndev);
2228 RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
2234 static int netvsc_probe(struct hv_device *dev,
2235 const struct hv_vmbus_device_id *dev_id)
2237 struct net_device *net = NULL;
2238 struct net_device_context *net_device_ctx;
2239 struct netvsc_device_info *device_info = NULL;
2240 struct netvsc_device *nvdev;
2243 net = alloc_etherdev_mq(sizeof(struct net_device_context),
2248 netif_carrier_off(net);
2250 netvsc_init_settings(net);
2252 net_device_ctx = netdev_priv(net);
2253 net_device_ctx->device_ctx = dev;
2254 net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
2255 if (netif_msg_probe(net_device_ctx))
2256 netdev_dbg(net, "netvsc msg_enable: %d\n",
2257 net_device_ctx->msg_enable);
2259 hv_set_drvdata(dev, net);
2261 INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
2263 spin_lock_init(&net_device_ctx->lock);
2264 INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
2265 INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup);
2267 net_device_ctx->vf_stats
2268 = netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats);
2269 if (!net_device_ctx->vf_stats)
2272 net->netdev_ops = &device_ops;
2273 net->ethtool_ops = ðtool_ops;
2274 SET_NETDEV_DEV(net, &dev->device);
2276 /* We always need headroom for rndis header */
2277 net->needed_headroom = RNDIS_AND_PPI_SIZE;
2279 /* Initialize the number of queues to be 1, we may change it if more
2280 * channels are offered later.
2282 netif_set_real_num_tx_queues(net, 1);
2283 netif_set_real_num_rx_queues(net, 1);
2285 /* Notify the netvsc driver of the new device */
2286 device_info = netvsc_devinfo_get(NULL);
2290 goto devinfo_failed;
2293 nvdev = rndis_filter_device_add(dev, device_info);
2294 if (IS_ERR(nvdev)) {
2295 ret = PTR_ERR(nvdev);
2296 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
2300 memcpy(net->dev_addr, device_info->mac_adr, ETH_ALEN);
2302 /* We must get rtnl lock before scheduling nvdev->subchan_work,
2303 * otherwise netvsc_subchan_work() can get rtnl lock first and wait
2304 * all subchannels to show up, but that may not happen because
2305 * netvsc_probe() can't get rtnl lock and as a result vmbus_onoffer()
2306 * -> ... -> device_add() -> ... -> __device_attach() can't get
2307 * the device lock, so all the subchannels can't be processed --
2308 * finally netvsc_subchan_work() hangs forever.
2312 if (nvdev->num_chn > 1)
2313 schedule_work(&nvdev->subchan_work);
2315 /* hw_features computed in rndis_netdev_set_hwcaps() */
2316 net->features = net->hw_features |
2317 NETIF_F_HIGHDMA | NETIF_F_SG |
2318 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
2319 net->vlan_features = net->features;
2321 netdev_lockdep_set_classes(net);
2323 /* MTU range: 68 - 1500 or 65521 */
2324 net->min_mtu = NETVSC_MTU_MIN;
2325 if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
2326 net->max_mtu = NETVSC_MTU - ETH_HLEN;
2328 net->max_mtu = ETH_DATA_LEN;
2330 ret = register_netdevice(net);
2332 pr_err("Unable to register netdev.\n");
2333 goto register_failed;
2336 list_add(&net_device_ctx->list, &netvsc_dev_list);
2344 rndis_filter_device_remove(dev, nvdev);
2348 free_percpu(net_device_ctx->vf_stats);
2350 hv_set_drvdata(dev, NULL);
2356 static int netvsc_remove(struct hv_device *dev)
2358 struct net_device_context *ndev_ctx;
2359 struct net_device *vf_netdev, *net;
2360 struct netvsc_device *nvdev;
2362 net = hv_get_drvdata(dev);
2364 dev_err(&dev->device, "No net device to remove\n");
2368 ndev_ctx = netdev_priv(net);
2370 cancel_delayed_work_sync(&ndev_ctx->dwork);
2373 nvdev = rtnl_dereference(ndev_ctx->nvdev);
2375 cancel_work_sync(&nvdev->subchan_work);
2378 * Call to the vsc driver to let it know that the device is being
2379 * removed. Also blocks mtu and channel changes.
2381 vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2383 netvsc_unregister_vf(vf_netdev);
2386 rndis_filter_device_remove(dev, nvdev);
2388 unregister_netdevice(net);
2389 list_del(&ndev_ctx->list);
2393 hv_set_drvdata(dev, NULL);
2395 free_percpu(ndev_ctx->vf_stats);
2400 static const struct hv_vmbus_device_id id_table[] = {
2406 MODULE_DEVICE_TABLE(vmbus, id_table);
2408 /* The one and only one */
2409 static struct hv_driver netvsc_drv = {
2410 .name = KBUILD_MODNAME,
2411 .id_table = id_table,
2412 .probe = netvsc_probe,
2413 .remove = netvsc_remove,
2415 .probe_type = PROBE_PREFER_ASYNCHRONOUS,
2420 * On Hyper-V, every VF interface is matched with a corresponding
2421 * synthetic interface. The synthetic interface is presented first
2422 * to the guest. When the corresponding VF instance is registered,
2423 * we will take care of switching the data path.
2425 static int netvsc_netdev_event(struct notifier_block *this,
2426 unsigned long event, void *ptr)
2428 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2430 /* Skip our own events */
2431 if (event_dev->netdev_ops == &device_ops)
2434 /* Avoid non-Ethernet type devices */
2435 if (event_dev->type != ARPHRD_ETHER)
2438 /* Avoid Vlan dev with same MAC registering as VF */
2439 if (is_vlan_dev(event_dev))
2442 /* Avoid Bonding master dev with same MAC registering as VF */
2443 if ((event_dev->priv_flags & IFF_BONDING) &&
2444 (event_dev->flags & IFF_MASTER))
2448 case NETDEV_REGISTER:
2449 return netvsc_register_vf(event_dev);
2450 case NETDEV_UNREGISTER:
2451 return netvsc_unregister_vf(event_dev);
2454 return netvsc_vf_changed(event_dev);
2460 static struct notifier_block netvsc_netdev_notifier = {
2461 .notifier_call = netvsc_netdev_event,
2464 static void __exit netvsc_drv_exit(void)
2466 unregister_netdevice_notifier(&netvsc_netdev_notifier);
2467 vmbus_driver_unregister(&netvsc_drv);
2470 static int __init netvsc_drv_init(void)
2474 if (ring_size < RING_SIZE_MIN) {
2475 ring_size = RING_SIZE_MIN;
2476 pr_info("Increased ring_size to %u (min allowed)\n",
2479 netvsc_ring_bytes = ring_size * PAGE_SIZE;
2481 ret = vmbus_driver_register(&netvsc_drv);
2485 register_netdevice_notifier(&netvsc_netdev_notifier);
2489 MODULE_LICENSE("GPL");
2490 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
2492 module_init(netvsc_drv_init);
2493 module_exit(netvsc_drv_exit);