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);
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;
249 ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
251 rndis_pkt->per_pkt_info_len += ppi_size;
256 /* Azure hosts don't support non-TCP port numbers in hashing for fragmented
257 * packets. We can use ethtool to change UDP hash level when necessary.
259 static inline u32 netvsc_get_hash(
261 const struct net_device_context *ndc)
263 struct flow_keys flow;
264 u32 hash, pkt_proto = 0;
265 static u32 hashrnd __read_mostly;
267 net_get_random_once(&hashrnd, sizeof(hashrnd));
269 if (!skb_flow_dissect_flow_keys(skb, &flow, 0))
272 switch (flow.basic.ip_proto) {
274 if (flow.basic.n_proto == htons(ETH_P_IP))
275 pkt_proto = HV_TCP4_L4HASH;
276 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
277 pkt_proto = HV_TCP6_L4HASH;
282 if (flow.basic.n_proto == htons(ETH_P_IP))
283 pkt_proto = HV_UDP4_L4HASH;
284 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
285 pkt_proto = HV_UDP6_L4HASH;
290 if (pkt_proto & ndc->l4_hash) {
291 return skb_get_hash(skb);
293 if (flow.basic.n_proto == htons(ETH_P_IP))
294 hash = jhash2((u32 *)&flow.addrs.v4addrs, 2, hashrnd);
295 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
296 hash = jhash2((u32 *)&flow.addrs.v6addrs, 8, hashrnd);
300 __skb_set_sw_hash(skb, hash, false);
306 static inline int netvsc_get_tx_queue(struct net_device *ndev,
307 struct sk_buff *skb, int old_idx)
309 const struct net_device_context *ndc = netdev_priv(ndev);
310 struct sock *sk = skb->sk;
313 q_idx = ndc->tx_table[netvsc_get_hash(skb, ndc) &
314 (VRSS_SEND_TAB_SIZE - 1)];
316 /* If queue index changed record the new value */
317 if (q_idx != old_idx &&
318 sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
319 sk_tx_queue_set(sk, q_idx);
325 * Select queue for transmit.
327 * If a valid queue has already been assigned, then use that.
328 * Otherwise compute tx queue based on hash and the send table.
330 * This is basically similar to default (__netdev_pick_tx) with the added step
331 * of using the host send_table when no other queue has been assigned.
333 * TODO support XPS - but get_xps_queue not exported
335 static u16 netvsc_pick_tx(struct net_device *ndev, struct sk_buff *skb)
337 int q_idx = sk_tx_queue_get(skb->sk);
339 if (q_idx < 0 || skb->ooo_okay || q_idx >= ndev->real_num_tx_queues) {
340 /* If forwarding a packet, we use the recorded queue when
341 * available for better cache locality.
343 if (skb_rx_queue_recorded(skb))
344 q_idx = skb_get_rx_queue(skb);
346 q_idx = netvsc_get_tx_queue(ndev, skb, q_idx);
352 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
353 struct net_device *sb_dev,
354 select_queue_fallback_t fallback)
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,
369 txq = fallback(vf_netdev, skb, NULL);
371 /* Record the queue selected by VF so that it can be
372 * used for common case where VF has more queues than
373 * the synthetic device.
375 qdisc_skb_cb(skb)->slave_dev_queue_mapping = txq;
377 txq = netvsc_pick_tx(ndev, skb);
381 while (txq >= ndev->real_num_tx_queues)
382 txq -= ndev->real_num_tx_queues;
387 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
388 struct hv_page_buffer *pb)
392 /* Deal with compund pages by ignoring unused part
395 page += (offset >> PAGE_SHIFT);
396 offset &= ~PAGE_MASK;
401 bytes = PAGE_SIZE - offset;
404 pb[j].pfn = page_to_pfn(page);
405 pb[j].offset = offset;
411 if (offset == PAGE_SIZE && len) {
421 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
422 struct hv_netvsc_packet *packet,
423 struct hv_page_buffer *pb)
426 char *data = skb->data;
427 int frags = skb_shinfo(skb)->nr_frags;
430 /* The packet is laid out thus:
431 * 1. hdr: RNDIS header and PPI
433 * 3. skb fragment data
435 slots_used += fill_pg_buf(virt_to_page(hdr),
437 len, &pb[slots_used]);
439 packet->rmsg_size = len;
440 packet->rmsg_pgcnt = slots_used;
442 slots_used += fill_pg_buf(virt_to_page(data),
443 offset_in_page(data),
444 skb_headlen(skb), &pb[slots_used]);
446 for (i = 0; i < frags; i++) {
447 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
449 slots_used += fill_pg_buf(skb_frag_page(frag),
451 skb_frag_size(frag), &pb[slots_used]);
456 static int count_skb_frag_slots(struct sk_buff *skb)
458 int i, frags = skb_shinfo(skb)->nr_frags;
461 for (i = 0; i < frags; i++) {
462 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
463 unsigned long size = skb_frag_size(frag);
464 unsigned long offset = frag->page_offset;
466 /* Skip unused frames from start of page */
467 offset &= ~PAGE_MASK;
468 pages += PFN_UP(offset + size);
473 static int netvsc_get_slots(struct sk_buff *skb)
475 char *data = skb->data;
476 unsigned int offset = offset_in_page(data);
477 unsigned int len = skb_headlen(skb);
481 slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
482 frag_slots = count_skb_frag_slots(skb);
483 return slots + frag_slots;
486 static u32 net_checksum_info(struct sk_buff *skb)
488 if (skb->protocol == htons(ETH_P_IP)) {
489 struct iphdr *ip = ip_hdr(skb);
491 if (ip->protocol == IPPROTO_TCP)
492 return TRANSPORT_INFO_IPV4_TCP;
493 else if (ip->protocol == IPPROTO_UDP)
494 return TRANSPORT_INFO_IPV4_UDP;
496 struct ipv6hdr *ip6 = ipv6_hdr(skb);
498 if (ip6->nexthdr == IPPROTO_TCP)
499 return TRANSPORT_INFO_IPV6_TCP;
500 else if (ip6->nexthdr == IPPROTO_UDP)
501 return TRANSPORT_INFO_IPV6_UDP;
504 return TRANSPORT_INFO_NOT_IP;
507 /* Send skb on the slave VF device. */
508 static int netvsc_vf_xmit(struct net_device *net, struct net_device *vf_netdev,
511 struct net_device_context *ndev_ctx = netdev_priv(net);
512 unsigned int len = skb->len;
515 skb->dev = vf_netdev;
516 skb_record_rx_queue(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
518 rc = dev_queue_xmit(skb);
519 if (likely(rc == NET_XMIT_SUCCESS || rc == NET_XMIT_CN)) {
520 struct netvsc_vf_pcpu_stats *pcpu_stats
521 = this_cpu_ptr(ndev_ctx->vf_stats);
523 u64_stats_update_begin(&pcpu_stats->syncp);
524 pcpu_stats->tx_packets++;
525 pcpu_stats->tx_bytes += len;
526 u64_stats_update_end(&pcpu_stats->syncp);
528 this_cpu_inc(ndev_ctx->vf_stats->tx_dropped);
534 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
536 struct net_device_context *net_device_ctx = netdev_priv(net);
537 struct hv_netvsc_packet *packet = NULL;
539 unsigned int num_data_pgs;
540 struct rndis_message *rndis_msg;
541 struct net_device *vf_netdev;
544 struct hv_page_buffer pb[MAX_PAGE_BUFFER_COUNT];
546 /* If VF is present and up then redirect packets to it.
547 * Skip the VF if it is marked down or has no carrier.
548 * If netpoll is in uses, then VF can not be used either.
550 vf_netdev = rcu_dereference_bh(net_device_ctx->vf_netdev);
551 if (vf_netdev && netif_running(vf_netdev) &&
552 netif_carrier_ok(vf_netdev) && !netpoll_tx_running(net))
553 return netvsc_vf_xmit(net, vf_netdev, skb);
555 /* We will atmost need two pages to describe the rndis
556 * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
557 * of pages in a single packet. If skb is scattered around
558 * more pages we try linearizing it.
561 num_data_pgs = netvsc_get_slots(skb) + 2;
563 if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
564 ++net_device_ctx->eth_stats.tx_scattered;
566 if (skb_linearize(skb))
569 num_data_pgs = netvsc_get_slots(skb) + 2;
570 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
571 ++net_device_ctx->eth_stats.tx_too_big;
577 * Place the rndis header in the skb head room and
578 * the skb->cb will be used for hv_netvsc_packet
581 ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
585 /* Use the skb control buffer for building up the packet */
586 BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
587 FIELD_SIZEOF(struct sk_buff, cb));
588 packet = (struct hv_netvsc_packet *)skb->cb;
590 packet->q_idx = skb_get_queue_mapping(skb);
592 packet->total_data_buflen = skb->len;
593 packet->total_bytes = skb->len;
594 packet->total_packets = 1;
596 rndis_msg = (struct rndis_message *)skb->head;
598 /* Add the rndis header */
599 rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
600 rndis_msg->msg_len = packet->total_data_buflen;
602 rndis_msg->msg.pkt = (struct rndis_packet) {
603 .data_offset = sizeof(struct rndis_packet),
604 .data_len = packet->total_data_buflen,
605 .per_pkt_info_offset = sizeof(struct rndis_packet),
608 rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
610 hash = skb_get_hash_raw(skb);
611 if (hash != 0 && net->real_num_tx_queues > 1) {
614 rndis_msg_size += NDIS_HASH_PPI_SIZE;
615 hash_info = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
620 if (skb_vlan_tag_present(skb)) {
621 struct ndis_pkt_8021q_info *vlan;
623 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
624 vlan = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
628 vlan->vlanid = skb->vlan_tci & VLAN_VID_MASK;
629 vlan->pri = (skb->vlan_tci & VLAN_PRIO_MASK) >>
633 if (skb_is_gso(skb)) {
634 struct ndis_tcp_lso_info *lso_info;
636 rndis_msg_size += NDIS_LSO_PPI_SIZE;
637 lso_info = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
638 TCP_LARGESEND_PKTINFO);
641 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
642 if (skb->protocol == htons(ETH_P_IP)) {
643 lso_info->lso_v2_transmit.ip_version =
644 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
645 ip_hdr(skb)->tot_len = 0;
646 ip_hdr(skb)->check = 0;
647 tcp_hdr(skb)->check =
648 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
649 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
651 lso_info->lso_v2_transmit.ip_version =
652 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
653 ipv6_hdr(skb)->payload_len = 0;
654 tcp_hdr(skb)->check =
655 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
656 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
658 lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
659 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
660 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
661 if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
662 struct ndis_tcp_ip_checksum_info *csum_info;
664 rndis_msg_size += NDIS_CSUM_PPI_SIZE;
665 csum_info = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
666 TCPIP_CHKSUM_PKTINFO);
668 csum_info->value = 0;
669 csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
671 if (skb->protocol == htons(ETH_P_IP)) {
672 csum_info->transmit.is_ipv4 = 1;
674 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
675 csum_info->transmit.tcp_checksum = 1;
677 csum_info->transmit.udp_checksum = 1;
679 csum_info->transmit.is_ipv6 = 1;
681 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
682 csum_info->transmit.tcp_checksum = 1;
684 csum_info->transmit.udp_checksum = 1;
687 /* Can't do offload of this type of checksum */
688 if (skb_checksum_help(skb))
693 /* Start filling in the page buffers with the rndis hdr */
694 rndis_msg->msg_len += rndis_msg_size;
695 packet->total_data_buflen = rndis_msg->msg_len;
696 packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
699 /* timestamp packet in software */
700 skb_tx_timestamp(skb);
702 ret = netvsc_send(net, packet, rndis_msg, pb, skb);
703 if (likely(ret == 0))
706 if (ret == -EAGAIN) {
707 ++net_device_ctx->eth_stats.tx_busy;
708 return NETDEV_TX_BUSY;
712 ++net_device_ctx->eth_stats.tx_no_space;
715 dev_kfree_skb_any(skb);
716 net->stats.tx_dropped++;
721 ++net_device_ctx->eth_stats.tx_no_memory;
726 * netvsc_linkstatus_callback - Link up/down notification
728 void netvsc_linkstatus_callback(struct net_device *net,
729 struct rndis_message *resp)
731 struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
732 struct net_device_context *ndev_ctx = netdev_priv(net);
733 struct netvsc_reconfig *event;
736 /* Update the physical link speed when changing to another vSwitch */
737 if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
740 speed = *(u32 *)((void *)indicate
741 + indicate->status_buf_offset) / 10000;
742 ndev_ctx->speed = speed;
746 /* Handle these link change statuses below */
747 if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
748 indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
749 indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
752 if (net->reg_state != NETREG_REGISTERED)
755 event = kzalloc(sizeof(*event), GFP_ATOMIC);
758 event->event = indicate->status;
760 spin_lock_irqsave(&ndev_ctx->lock, flags);
761 list_add_tail(&event->list, &ndev_ctx->reconfig_events);
762 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
764 schedule_delayed_work(&ndev_ctx->dwork, 0);
767 static void netvsc_comp_ipcsum(struct sk_buff *skb)
769 struct iphdr *iph = (struct iphdr *)skb->data;
772 iph->check = ip_fast_csum(iph, iph->ihl);
775 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
776 struct napi_struct *napi,
777 const struct ndis_tcp_ip_checksum_info *csum_info,
778 const struct ndis_pkt_8021q_info *vlan,
779 void *data, u32 buflen)
783 skb = napi_alloc_skb(napi, buflen);
788 * Copy to skb. This copy is needed here since the memory pointed by
789 * hv_netvsc_packet cannot be deallocated
791 skb_put_data(skb, data, buflen);
793 skb->protocol = eth_type_trans(skb, net);
795 /* skb is already created with CHECKSUM_NONE */
796 skb_checksum_none_assert(skb);
798 /* Incoming packets may have IP header checksum verified by the host.
799 * They may not have IP header checksum computed after coalescing.
800 * We compute it here if the flags are set, because on Linux, the IP
801 * checksum is always checked.
803 if (csum_info && csum_info->receive.ip_checksum_value_invalid &&
804 csum_info->receive.ip_checksum_succeeded &&
805 skb->protocol == htons(ETH_P_IP))
806 netvsc_comp_ipcsum(skb);
808 /* Do L4 checksum offload if enabled and present. */
809 if (csum_info && (net->features & NETIF_F_RXCSUM)) {
810 if (csum_info->receive.tcp_checksum_succeeded ||
811 csum_info->receive.udp_checksum_succeeded)
812 skb->ip_summed = CHECKSUM_UNNECESSARY;
816 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT);
818 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
826 * netvsc_recv_callback - Callback when we receive a packet from the
827 * "wire" on the specified device.
829 int netvsc_recv_callback(struct net_device *net,
830 struct netvsc_device *net_device,
831 struct vmbus_channel *channel,
833 const struct ndis_tcp_ip_checksum_info *csum_info,
834 const struct ndis_pkt_8021q_info *vlan)
836 struct net_device_context *net_device_ctx = netdev_priv(net);
837 u16 q_idx = channel->offermsg.offer.sub_channel_index;
838 struct netvsc_channel *nvchan = &net_device->chan_table[q_idx];
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->napi,
847 csum_info, vlan, data, len);
848 if (unlikely(!skb)) {
849 ++net_device_ctx->eth_stats.rx_no_memory;
850 return NVSP_STAT_FAIL;
853 skb_record_rx_queue(skb, q_idx);
856 * Even if injecting the packet, record the statistics
857 * on the synthetic device because modifying the VF device
858 * statistics will not work correctly.
860 rx_stats = &nvchan->rx_stats;
861 u64_stats_update_begin(&rx_stats->syncp);
863 rx_stats->bytes += len;
865 if (skb->pkt_type == PACKET_BROADCAST)
866 ++rx_stats->broadcast;
867 else if (skb->pkt_type == PACKET_MULTICAST)
868 ++rx_stats->multicast;
869 u64_stats_update_end(&rx_stats->syncp);
871 napi_gro_receive(&nvchan->napi, skb);
872 return NVSP_STAT_SUCCESS;
875 static void netvsc_get_drvinfo(struct net_device *net,
876 struct ethtool_drvinfo *info)
878 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
879 strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
882 static void netvsc_get_channels(struct net_device *net,
883 struct ethtool_channels *channel)
885 struct net_device_context *net_device_ctx = netdev_priv(net);
886 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
889 channel->max_combined = nvdev->max_chn;
890 channel->combined_count = nvdev->num_chn;
894 /* Alloc struct netvsc_device_info, and initialize it from either existing
895 * struct netvsc_device, or from default values.
897 static struct netvsc_device_info *netvsc_devinfo_get
898 (struct netvsc_device *nvdev)
900 struct netvsc_device_info *dev_info;
902 dev_info = kzalloc(sizeof(*dev_info), GFP_ATOMIC);
908 dev_info->num_chn = nvdev->num_chn;
909 dev_info->send_sections = nvdev->send_section_cnt;
910 dev_info->send_section_size = nvdev->send_section_size;
911 dev_info->recv_sections = nvdev->recv_section_cnt;
912 dev_info->recv_section_size = nvdev->recv_section_size;
914 memcpy(dev_info->rss_key, nvdev->extension->rss_key,
917 dev_info->num_chn = VRSS_CHANNEL_DEFAULT;
918 dev_info->send_sections = NETVSC_DEFAULT_TX;
919 dev_info->send_section_size = NETVSC_SEND_SECTION_SIZE;
920 dev_info->recv_sections = NETVSC_DEFAULT_RX;
921 dev_info->recv_section_size = NETVSC_RECV_SECTION_SIZE;
927 static int netvsc_detach(struct net_device *ndev,
928 struct netvsc_device *nvdev)
930 struct net_device_context *ndev_ctx = netdev_priv(ndev);
931 struct hv_device *hdev = ndev_ctx->device_ctx;
934 /* Don't try continuing to try and setup sub channels */
935 if (cancel_work_sync(&nvdev->subchan_work))
938 /* If device was up (receiving) then shutdown */
939 if (netif_running(ndev)) {
940 netvsc_tx_disable(nvdev, ndev);
942 ret = rndis_filter_close(nvdev);
945 "unable to close device (ret %d).\n", ret);
949 ret = netvsc_wait_until_empty(nvdev);
952 "Ring buffer not empty after closing rndis\n");
957 netif_device_detach(ndev);
959 rndis_filter_device_remove(hdev, nvdev);
964 static int netvsc_attach(struct net_device *ndev,
965 struct netvsc_device_info *dev_info)
967 struct net_device_context *ndev_ctx = netdev_priv(ndev);
968 struct hv_device *hdev = ndev_ctx->device_ctx;
969 struct netvsc_device *nvdev;
970 struct rndis_device *rdev;
973 nvdev = rndis_filter_device_add(hdev, dev_info);
975 return PTR_ERR(nvdev);
977 if (nvdev->num_chn > 1) {
978 ret = rndis_set_subchannel(ndev, nvdev, dev_info);
980 /* if unavailable, just proceed with one queue */
987 /* In any case device is now ready */
988 nvdev->tx_disable = false;
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 netif_device_detach(ndev);
1009 rndis_filter_device_remove(hdev, nvdev);
1014 static int netvsc_set_channels(struct net_device *net,
1015 struct ethtool_channels *channels)
1017 struct net_device_context *net_device_ctx = netdev_priv(net);
1018 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
1019 unsigned int orig, count = channels->combined_count;
1020 struct netvsc_device_info *device_info;
1023 /* We do not support separate count for rx, tx, or other */
1025 channels->rx_count || channels->tx_count || channels->other_count)
1028 if (!nvdev || nvdev->destroy)
1031 if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
1034 if (count > nvdev->max_chn)
1037 orig = nvdev->num_chn;
1039 device_info = netvsc_devinfo_get(nvdev);
1044 device_info->num_chn = count;
1046 ret = netvsc_detach(net, nvdev);
1050 ret = netvsc_attach(net, device_info);
1052 device_info->num_chn = orig;
1053 if (netvsc_attach(net, device_info))
1054 netdev_err(net, "restoring channel setting failed\n");
1063 netvsc_validate_ethtool_ss_cmd(const struct ethtool_link_ksettings *cmd)
1065 struct ethtool_link_ksettings diff1 = *cmd;
1066 struct ethtool_link_ksettings diff2 = {};
1068 diff1.base.speed = 0;
1069 diff1.base.duplex = 0;
1070 /* advertising and cmd are usually set */
1071 ethtool_link_ksettings_zero_link_mode(&diff1, advertising);
1073 /* We set port to PORT_OTHER */
1074 diff2.base.port = PORT_OTHER;
1076 return !memcmp(&diff1, &diff2, sizeof(diff1));
1079 static void netvsc_init_settings(struct net_device *dev)
1081 struct net_device_context *ndc = netdev_priv(dev);
1083 ndc->l4_hash = HV_DEFAULT_L4HASH;
1085 ndc->speed = SPEED_UNKNOWN;
1086 ndc->duplex = DUPLEX_FULL;
1089 static int netvsc_get_link_ksettings(struct net_device *dev,
1090 struct ethtool_link_ksettings *cmd)
1092 struct net_device_context *ndc = netdev_priv(dev);
1094 cmd->base.speed = ndc->speed;
1095 cmd->base.duplex = ndc->duplex;
1096 cmd->base.port = PORT_OTHER;
1101 static int netvsc_set_link_ksettings(struct net_device *dev,
1102 const struct ethtool_link_ksettings *cmd)
1104 struct net_device_context *ndc = netdev_priv(dev);
1107 speed = cmd->base.speed;
1108 if (!ethtool_validate_speed(speed) ||
1109 !ethtool_validate_duplex(cmd->base.duplex) ||
1110 !netvsc_validate_ethtool_ss_cmd(cmd))
1114 ndc->duplex = cmd->base.duplex;
1119 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
1121 struct net_device_context *ndevctx = netdev_priv(ndev);
1122 struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1123 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1124 int orig_mtu = ndev->mtu;
1125 struct netvsc_device_info *device_info;
1128 if (!nvdev || nvdev->destroy)
1131 device_info = netvsc_devinfo_get(nvdev);
1136 /* Change MTU of underlying VF netdev first. */
1138 ret = dev_set_mtu(vf_netdev, mtu);
1143 ret = netvsc_detach(ndev, nvdev);
1149 ret = netvsc_attach(ndev, device_info);
1153 /* Attempt rollback to original MTU */
1154 ndev->mtu = orig_mtu;
1156 if (netvsc_attach(ndev, device_info))
1157 netdev_err(ndev, "restoring mtu failed\n");
1160 dev_set_mtu(vf_netdev, orig_mtu);
1167 static void netvsc_get_vf_stats(struct net_device *net,
1168 struct netvsc_vf_pcpu_stats *tot)
1170 struct net_device_context *ndev_ctx = netdev_priv(net);
1173 memset(tot, 0, sizeof(*tot));
1175 for_each_possible_cpu(i) {
1176 const struct netvsc_vf_pcpu_stats *stats
1177 = per_cpu_ptr(ndev_ctx->vf_stats, i);
1178 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1182 start = u64_stats_fetch_begin_irq(&stats->syncp);
1183 rx_packets = stats->rx_packets;
1184 tx_packets = stats->tx_packets;
1185 rx_bytes = stats->rx_bytes;
1186 tx_bytes = stats->tx_bytes;
1187 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1189 tot->rx_packets += rx_packets;
1190 tot->tx_packets += tx_packets;
1191 tot->rx_bytes += rx_bytes;
1192 tot->tx_bytes += tx_bytes;
1193 tot->tx_dropped += stats->tx_dropped;
1197 static void netvsc_get_pcpu_stats(struct net_device *net,
1198 struct netvsc_ethtool_pcpu_stats *pcpu_tot)
1200 struct net_device_context *ndev_ctx = netdev_priv(net);
1201 struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1204 /* fetch percpu stats of vf */
1205 for_each_possible_cpu(i) {
1206 const struct netvsc_vf_pcpu_stats *stats =
1207 per_cpu_ptr(ndev_ctx->vf_stats, i);
1208 struct netvsc_ethtool_pcpu_stats *this_tot = &pcpu_tot[i];
1212 start = u64_stats_fetch_begin_irq(&stats->syncp);
1213 this_tot->vf_rx_packets = stats->rx_packets;
1214 this_tot->vf_tx_packets = stats->tx_packets;
1215 this_tot->vf_rx_bytes = stats->rx_bytes;
1216 this_tot->vf_tx_bytes = stats->tx_bytes;
1217 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1218 this_tot->rx_packets = this_tot->vf_rx_packets;
1219 this_tot->tx_packets = this_tot->vf_tx_packets;
1220 this_tot->rx_bytes = this_tot->vf_rx_bytes;
1221 this_tot->tx_bytes = this_tot->vf_tx_bytes;
1224 /* fetch percpu stats of netvsc */
1225 for (i = 0; i < nvdev->num_chn; i++) {
1226 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1227 const struct netvsc_stats *stats;
1228 struct netvsc_ethtool_pcpu_stats *this_tot =
1229 &pcpu_tot[nvchan->channel->target_cpu];
1233 stats = &nvchan->tx_stats;
1235 start = u64_stats_fetch_begin_irq(&stats->syncp);
1236 packets = stats->packets;
1237 bytes = stats->bytes;
1238 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1240 this_tot->tx_bytes += bytes;
1241 this_tot->tx_packets += packets;
1243 stats = &nvchan->rx_stats;
1245 start = u64_stats_fetch_begin_irq(&stats->syncp);
1246 packets = stats->packets;
1247 bytes = stats->bytes;
1248 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1250 this_tot->rx_bytes += bytes;
1251 this_tot->rx_packets += packets;
1255 static void netvsc_get_stats64(struct net_device *net,
1256 struct rtnl_link_stats64 *t)
1258 struct net_device_context *ndev_ctx = netdev_priv(net);
1259 struct netvsc_device *nvdev;
1260 struct netvsc_vf_pcpu_stats vf_tot;
1265 nvdev = rcu_dereference(ndev_ctx->nvdev);
1269 netdev_stats_to_stats64(t, &net->stats);
1271 netvsc_get_vf_stats(net, &vf_tot);
1272 t->rx_packets += vf_tot.rx_packets;
1273 t->tx_packets += vf_tot.tx_packets;
1274 t->rx_bytes += vf_tot.rx_bytes;
1275 t->tx_bytes += vf_tot.tx_bytes;
1276 t->tx_dropped += vf_tot.tx_dropped;
1278 for (i = 0; i < nvdev->num_chn; i++) {
1279 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1280 const struct netvsc_stats *stats;
1281 u64 packets, bytes, multicast;
1284 stats = &nvchan->tx_stats;
1286 start = u64_stats_fetch_begin_irq(&stats->syncp);
1287 packets = stats->packets;
1288 bytes = stats->bytes;
1289 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1291 t->tx_bytes += bytes;
1292 t->tx_packets += packets;
1294 stats = &nvchan->rx_stats;
1296 start = u64_stats_fetch_begin_irq(&stats->syncp);
1297 packets = stats->packets;
1298 bytes = stats->bytes;
1299 multicast = stats->multicast + stats->broadcast;
1300 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1302 t->rx_bytes += bytes;
1303 t->rx_packets += packets;
1304 t->multicast += multicast;
1310 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1312 struct net_device_context *ndc = netdev_priv(ndev);
1313 struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1314 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1315 struct sockaddr *addr = p;
1318 err = eth_prepare_mac_addr_change(ndev, p);
1326 err = dev_set_mac_address(vf_netdev, addr);
1331 err = rndis_filter_set_device_mac(nvdev, addr->sa_data);
1333 eth_commit_mac_addr_change(ndev, p);
1334 } else if (vf_netdev) {
1335 /* rollback change on VF */
1336 memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN);
1337 dev_set_mac_address(vf_netdev, addr);
1343 static const struct {
1344 char name[ETH_GSTRING_LEN];
1346 } netvsc_stats[] = {
1347 { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1348 { "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1349 { "tx_no_space", offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1350 { "tx_too_big", offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1351 { "tx_busy", offsetof(struct netvsc_ethtool_stats, tx_busy) },
1352 { "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) },
1353 { "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) },
1354 { "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) },
1355 { "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) },
1356 { "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) },
1358 { "cpu%u_rx_packets",
1359 offsetof(struct netvsc_ethtool_pcpu_stats, rx_packets) },
1361 offsetof(struct netvsc_ethtool_pcpu_stats, rx_bytes) },
1362 { "cpu%u_tx_packets",
1363 offsetof(struct netvsc_ethtool_pcpu_stats, tx_packets) },
1365 offsetof(struct netvsc_ethtool_pcpu_stats, tx_bytes) },
1366 { "cpu%u_vf_rx_packets",
1367 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_packets) },
1368 { "cpu%u_vf_rx_bytes",
1369 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_bytes) },
1370 { "cpu%u_vf_tx_packets",
1371 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_packets) },
1372 { "cpu%u_vf_tx_bytes",
1373 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_bytes) },
1375 { "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) },
1376 { "vf_rx_bytes", offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) },
1377 { "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) },
1378 { "vf_tx_bytes", offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) },
1379 { "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) },
1382 #define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats)
1383 #define NETVSC_VF_STATS_LEN ARRAY_SIZE(vf_stats)
1385 /* statistics per queue (rx/tx packets/bytes) */
1386 #define NETVSC_PCPU_STATS_LEN (num_present_cpus() * ARRAY_SIZE(pcpu_stats))
1388 /* 4 statistics per queue (rx/tx packets/bytes) */
1389 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 4)
1391 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1393 struct net_device_context *ndc = netdev_priv(dev);
1394 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1399 switch (string_set) {
1401 return NETVSC_GLOBAL_STATS_LEN
1402 + NETVSC_VF_STATS_LEN
1403 + NETVSC_QUEUE_STATS_LEN(nvdev)
1404 + NETVSC_PCPU_STATS_LEN;
1410 static void netvsc_get_ethtool_stats(struct net_device *dev,
1411 struct ethtool_stats *stats, u64 *data)
1413 struct net_device_context *ndc = netdev_priv(dev);
1414 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1415 const void *nds = &ndc->eth_stats;
1416 const struct netvsc_stats *qstats;
1417 struct netvsc_vf_pcpu_stats sum;
1418 struct netvsc_ethtool_pcpu_stats *pcpu_sum;
1426 for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1427 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1429 netvsc_get_vf_stats(dev, &sum);
1430 for (j = 0; j < NETVSC_VF_STATS_LEN; j++)
1431 data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset);
1433 for (j = 0; j < nvdev->num_chn; j++) {
1434 qstats = &nvdev->chan_table[j].tx_stats;
1437 start = u64_stats_fetch_begin_irq(&qstats->syncp);
1438 packets = qstats->packets;
1439 bytes = qstats->bytes;
1440 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1441 data[i++] = packets;
1444 qstats = &nvdev->chan_table[j].rx_stats;
1446 start = u64_stats_fetch_begin_irq(&qstats->syncp);
1447 packets = qstats->packets;
1448 bytes = qstats->bytes;
1449 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1450 data[i++] = packets;
1454 pcpu_sum = kvmalloc_array(num_possible_cpus(),
1455 sizeof(struct netvsc_ethtool_pcpu_stats),
1457 netvsc_get_pcpu_stats(dev, pcpu_sum);
1458 for_each_present_cpu(cpu) {
1459 struct netvsc_ethtool_pcpu_stats *this_sum = &pcpu_sum[cpu];
1461 for (j = 0; j < ARRAY_SIZE(pcpu_stats); j++)
1462 data[i++] = *(u64 *)((void *)this_sum
1463 + pcpu_stats[j].offset);
1468 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1470 struct net_device_context *ndc = netdev_priv(dev);
1471 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1478 switch (stringset) {
1480 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++) {
1481 memcpy(p, netvsc_stats[i].name, ETH_GSTRING_LEN);
1482 p += ETH_GSTRING_LEN;
1485 for (i = 0; i < ARRAY_SIZE(vf_stats); i++) {
1486 memcpy(p, vf_stats[i].name, ETH_GSTRING_LEN);
1487 p += ETH_GSTRING_LEN;
1490 for (i = 0; i < nvdev->num_chn; i++) {
1491 sprintf(p, "tx_queue_%u_packets", i);
1492 p += ETH_GSTRING_LEN;
1493 sprintf(p, "tx_queue_%u_bytes", i);
1494 p += ETH_GSTRING_LEN;
1495 sprintf(p, "rx_queue_%u_packets", i);
1496 p += ETH_GSTRING_LEN;
1497 sprintf(p, "rx_queue_%u_bytes", i);
1498 p += ETH_GSTRING_LEN;
1501 for_each_present_cpu(cpu) {
1502 for (i = 0; i < ARRAY_SIZE(pcpu_stats); i++) {
1503 sprintf(p, pcpu_stats[i].name, cpu);
1504 p += ETH_GSTRING_LEN;
1513 netvsc_get_rss_hash_opts(struct net_device_context *ndc,
1514 struct ethtool_rxnfc *info)
1516 const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3;
1518 info->data = RXH_IP_SRC | RXH_IP_DST;
1520 switch (info->flow_type) {
1522 if (ndc->l4_hash & HV_TCP4_L4HASH)
1523 info->data |= l4_flag;
1528 if (ndc->l4_hash & HV_TCP6_L4HASH)
1529 info->data |= l4_flag;
1534 if (ndc->l4_hash & HV_UDP4_L4HASH)
1535 info->data |= l4_flag;
1540 if (ndc->l4_hash & HV_UDP6_L4HASH)
1541 info->data |= l4_flag;
1557 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1560 struct net_device_context *ndc = netdev_priv(dev);
1561 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1566 switch (info->cmd) {
1567 case ETHTOOL_GRXRINGS:
1568 info->data = nvdev->num_chn;
1572 return netvsc_get_rss_hash_opts(ndc, info);
1577 static int netvsc_set_rss_hash_opts(struct net_device_context *ndc,
1578 struct ethtool_rxnfc *info)
1580 if (info->data == (RXH_IP_SRC | RXH_IP_DST |
1581 RXH_L4_B_0_1 | RXH_L4_B_2_3)) {
1582 switch (info->flow_type) {
1584 ndc->l4_hash |= HV_TCP4_L4HASH;
1588 ndc->l4_hash |= HV_TCP6_L4HASH;
1592 ndc->l4_hash |= HV_UDP4_L4HASH;
1596 ndc->l4_hash |= HV_UDP6_L4HASH;
1606 if (info->data == (RXH_IP_SRC | RXH_IP_DST)) {
1607 switch (info->flow_type) {
1609 ndc->l4_hash &= ~HV_TCP4_L4HASH;
1613 ndc->l4_hash &= ~HV_TCP6_L4HASH;
1617 ndc->l4_hash &= ~HV_UDP4_L4HASH;
1621 ndc->l4_hash &= ~HV_UDP6_L4HASH;
1635 netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info)
1637 struct net_device_context *ndc = netdev_priv(ndev);
1639 if (info->cmd == ETHTOOL_SRXFH)
1640 return netvsc_set_rss_hash_opts(ndc, info);
1645 #ifdef CONFIG_NET_POLL_CONTROLLER
1646 static void netvsc_poll_controller(struct net_device *dev)
1648 struct net_device_context *ndc = netdev_priv(dev);
1649 struct netvsc_device *ndev;
1653 ndev = rcu_dereference(ndc->nvdev);
1655 for (i = 0; i < ndev->num_chn; i++) {
1656 struct netvsc_channel *nvchan = &ndev->chan_table[i];
1658 napi_schedule(&nvchan->napi);
1665 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1667 return NETVSC_HASH_KEYLEN;
1670 static u32 netvsc_rss_indir_size(struct net_device *dev)
1675 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1678 struct net_device_context *ndc = netdev_priv(dev);
1679 struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1680 struct rndis_device *rndis_dev;
1687 *hfunc = ETH_RSS_HASH_TOP; /* Toeplitz */
1689 rndis_dev = ndev->extension;
1691 for (i = 0; i < ITAB_NUM; i++)
1692 indir[i] = ndc->rx_table[i];
1696 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1701 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1702 const u8 *key, const u8 hfunc)
1704 struct net_device_context *ndc = netdev_priv(dev);
1705 struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1706 struct rndis_device *rndis_dev;
1712 if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1715 rndis_dev = ndev->extension;
1717 for (i = 0; i < ITAB_NUM; i++)
1718 if (indir[i] >= ndev->num_chn)
1721 for (i = 0; i < ITAB_NUM; i++)
1722 ndc->rx_table[i] = indir[i];
1729 key = rndis_dev->rss_key;
1732 return rndis_filter_set_rss_param(rndis_dev, key);
1735 /* Hyper-V RNDIS protocol does not have ring in the HW sense.
1736 * It does have pre-allocated receive area which is divided into sections.
1738 static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
1739 struct ethtool_ringparam *ring)
1743 ring->rx_pending = nvdev->recv_section_cnt;
1744 ring->tx_pending = nvdev->send_section_cnt;
1746 if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
1747 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
1749 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1751 ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
1752 ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
1753 / nvdev->send_section_size;
1756 static void netvsc_get_ringparam(struct net_device *ndev,
1757 struct ethtool_ringparam *ring)
1759 struct net_device_context *ndevctx = netdev_priv(ndev);
1760 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1765 __netvsc_get_ringparam(nvdev, ring);
1768 static int netvsc_set_ringparam(struct net_device *ndev,
1769 struct ethtool_ringparam *ring)
1771 struct net_device_context *ndevctx = netdev_priv(ndev);
1772 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1773 struct netvsc_device_info *device_info;
1774 struct ethtool_ringparam orig;
1778 if (!nvdev || nvdev->destroy)
1781 memset(&orig, 0, sizeof(orig));
1782 __netvsc_get_ringparam(nvdev, &orig);
1784 new_tx = clamp_t(u32, ring->tx_pending,
1785 NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
1786 new_rx = clamp_t(u32, ring->rx_pending,
1787 NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
1789 if (new_tx == orig.tx_pending &&
1790 new_rx == orig.rx_pending)
1791 return 0; /* no change */
1793 device_info = netvsc_devinfo_get(nvdev);
1798 device_info->send_sections = new_tx;
1799 device_info->recv_sections = new_rx;
1801 ret = netvsc_detach(ndev, nvdev);
1805 ret = netvsc_attach(ndev, device_info);
1807 device_info->send_sections = orig.tx_pending;
1808 device_info->recv_sections = orig.rx_pending;
1810 if (netvsc_attach(ndev, device_info))
1811 netdev_err(ndev, "restoring ringparam failed");
1819 static u32 netvsc_get_msglevel(struct net_device *ndev)
1821 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1823 return ndev_ctx->msg_enable;
1826 static void netvsc_set_msglevel(struct net_device *ndev, u32 val)
1828 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1830 ndev_ctx->msg_enable = val;
1833 static const struct ethtool_ops ethtool_ops = {
1834 .get_drvinfo = netvsc_get_drvinfo,
1835 .get_msglevel = netvsc_get_msglevel,
1836 .set_msglevel = netvsc_set_msglevel,
1837 .get_link = ethtool_op_get_link,
1838 .get_ethtool_stats = netvsc_get_ethtool_stats,
1839 .get_sset_count = netvsc_get_sset_count,
1840 .get_strings = netvsc_get_strings,
1841 .get_channels = netvsc_get_channels,
1842 .set_channels = netvsc_set_channels,
1843 .get_ts_info = ethtool_op_get_ts_info,
1844 .get_rxnfc = netvsc_get_rxnfc,
1845 .set_rxnfc = netvsc_set_rxnfc,
1846 .get_rxfh_key_size = netvsc_get_rxfh_key_size,
1847 .get_rxfh_indir_size = netvsc_rss_indir_size,
1848 .get_rxfh = netvsc_get_rxfh,
1849 .set_rxfh = netvsc_set_rxfh,
1850 .get_link_ksettings = netvsc_get_link_ksettings,
1851 .set_link_ksettings = netvsc_set_link_ksettings,
1852 .get_ringparam = netvsc_get_ringparam,
1853 .set_ringparam = netvsc_set_ringparam,
1856 static const struct net_device_ops device_ops = {
1857 .ndo_open = netvsc_open,
1858 .ndo_stop = netvsc_close,
1859 .ndo_start_xmit = netvsc_start_xmit,
1860 .ndo_change_rx_flags = netvsc_change_rx_flags,
1861 .ndo_set_rx_mode = netvsc_set_rx_mode,
1862 .ndo_change_mtu = netvsc_change_mtu,
1863 .ndo_validate_addr = eth_validate_addr,
1864 .ndo_set_mac_address = netvsc_set_mac_addr,
1865 .ndo_select_queue = netvsc_select_queue,
1866 .ndo_get_stats64 = netvsc_get_stats64,
1867 #ifdef CONFIG_NET_POLL_CONTROLLER
1868 .ndo_poll_controller = netvsc_poll_controller,
1873 * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1874 * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1875 * present send GARP packet to network peers with netif_notify_peers().
1877 static void netvsc_link_change(struct work_struct *w)
1879 struct net_device_context *ndev_ctx =
1880 container_of(w, struct net_device_context, dwork.work);
1881 struct hv_device *device_obj = ndev_ctx->device_ctx;
1882 struct net_device *net = hv_get_drvdata(device_obj);
1883 struct netvsc_device *net_device;
1884 struct rndis_device *rdev;
1885 struct netvsc_reconfig *event = NULL;
1886 bool notify = false, reschedule = false;
1887 unsigned long flags, next_reconfig, delay;
1889 /* if changes are happening, comeback later */
1890 if (!rtnl_trylock()) {
1891 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1895 net_device = rtnl_dereference(ndev_ctx->nvdev);
1899 rdev = net_device->extension;
1901 next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1902 if (time_is_after_jiffies(next_reconfig)) {
1903 /* link_watch only sends one notification with current state
1904 * per second, avoid doing reconfig more frequently. Handle
1907 delay = next_reconfig - jiffies;
1908 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1909 schedule_delayed_work(&ndev_ctx->dwork, delay);
1912 ndev_ctx->last_reconfig = jiffies;
1914 spin_lock_irqsave(&ndev_ctx->lock, flags);
1915 if (!list_empty(&ndev_ctx->reconfig_events)) {
1916 event = list_first_entry(&ndev_ctx->reconfig_events,
1917 struct netvsc_reconfig, list);
1918 list_del(&event->list);
1919 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1921 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1926 switch (event->event) {
1927 /* Only the following events are possible due to the check in
1928 * netvsc_linkstatus_callback()
1930 case RNDIS_STATUS_MEDIA_CONNECT:
1931 if (rdev->link_state) {
1932 rdev->link_state = false;
1933 netif_carrier_on(net);
1934 netvsc_tx_enable(net_device, net);
1940 case RNDIS_STATUS_MEDIA_DISCONNECT:
1941 if (!rdev->link_state) {
1942 rdev->link_state = true;
1943 netif_carrier_off(net);
1944 netvsc_tx_disable(net_device, net);
1948 case RNDIS_STATUS_NETWORK_CHANGE:
1949 /* Only makes sense if carrier is present */
1950 if (!rdev->link_state) {
1951 rdev->link_state = true;
1952 netif_carrier_off(net);
1953 netvsc_tx_disable(net_device, net);
1954 event->event = RNDIS_STATUS_MEDIA_CONNECT;
1955 spin_lock_irqsave(&ndev_ctx->lock, flags);
1956 list_add(&event->list, &ndev_ctx->reconfig_events);
1957 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1966 netdev_notify_peers(net);
1968 /* link_watch only sends one notification with current state per
1969 * second, handle next reconfig event in 2 seconds.
1972 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1980 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
1982 struct net_device_context *net_device_ctx;
1983 struct net_device *dev;
1985 dev = netdev_master_upper_dev_get(vf_netdev);
1986 if (!dev || dev->netdev_ops != &device_ops)
1987 return NULL; /* not a netvsc device */
1989 net_device_ctx = netdev_priv(dev);
1990 if (!rtnl_dereference(net_device_ctx->nvdev))
1991 return NULL; /* device is removed */
1996 /* Called when VF is injecting data into network stack.
1997 * Change the associated network device from VF to netvsc.
1998 * note: already called with rcu_read_lock
2000 static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
2002 struct sk_buff *skb = *pskb;
2003 struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data);
2004 struct net_device_context *ndev_ctx = netdev_priv(ndev);
2005 struct netvsc_vf_pcpu_stats *pcpu_stats
2006 = this_cpu_ptr(ndev_ctx->vf_stats);
2008 skb = skb_share_check(skb, GFP_ATOMIC);
2010 return RX_HANDLER_CONSUMED;
2016 u64_stats_update_begin(&pcpu_stats->syncp);
2017 pcpu_stats->rx_packets++;
2018 pcpu_stats->rx_bytes += skb->len;
2019 u64_stats_update_end(&pcpu_stats->syncp);
2021 return RX_HANDLER_ANOTHER;
2024 static int netvsc_vf_join(struct net_device *vf_netdev,
2025 struct net_device *ndev)
2027 struct net_device_context *ndev_ctx = netdev_priv(ndev);
2030 ret = netdev_rx_handler_register(vf_netdev,
2031 netvsc_vf_handle_frame, ndev);
2033 netdev_err(vf_netdev,
2034 "can not register netvsc VF receive handler (err = %d)\n",
2036 goto rx_handler_failed;
2039 ret = netdev_master_upper_dev_link(vf_netdev, ndev,
2042 netdev_err(vf_netdev,
2043 "can not set master device %s (err = %d)\n",
2045 goto upper_link_failed;
2048 /* set slave flag before open to prevent IPv6 addrconf */
2049 vf_netdev->flags |= IFF_SLAVE;
2051 schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
2053 call_netdevice_notifiers(NETDEV_JOIN, vf_netdev);
2055 netdev_info(vf_netdev, "joined to %s\n", ndev->name);
2059 netdev_rx_handler_unregister(vf_netdev);
2064 static void __netvsc_vf_setup(struct net_device *ndev,
2065 struct net_device *vf_netdev)
2069 /* Align MTU of VF with master */
2070 ret = dev_set_mtu(vf_netdev, ndev->mtu);
2072 netdev_warn(vf_netdev,
2073 "unable to change mtu to %u\n", ndev->mtu);
2075 /* set multicast etc flags on VF */
2076 dev_change_flags(vf_netdev, ndev->flags | IFF_SLAVE);
2078 /* sync address list from ndev to VF */
2079 netif_addr_lock_bh(ndev);
2080 dev_uc_sync(vf_netdev, ndev);
2081 dev_mc_sync(vf_netdev, ndev);
2082 netif_addr_unlock_bh(ndev);
2084 if (netif_running(ndev)) {
2085 ret = dev_open(vf_netdev);
2087 netdev_warn(vf_netdev,
2088 "unable to open: %d\n", ret);
2092 /* Setup VF as slave of the synthetic device.
2093 * Runs in workqueue to avoid recursion in netlink callbacks.
2095 static void netvsc_vf_setup(struct work_struct *w)
2097 struct net_device_context *ndev_ctx
2098 = container_of(w, struct net_device_context, vf_takeover.work);
2099 struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2100 struct net_device *vf_netdev;
2102 if (!rtnl_trylock()) {
2103 schedule_delayed_work(&ndev_ctx->vf_takeover, 0);
2107 vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2109 __netvsc_vf_setup(ndev, vf_netdev);
2114 /* Find netvsc by VF serial number.
2115 * The PCI hyperv controller records the serial number as the slot kobj name.
2117 static struct net_device *get_netvsc_byslot(const struct net_device *vf_netdev)
2119 struct device *parent = vf_netdev->dev.parent;
2120 struct net_device_context *ndev_ctx;
2121 struct pci_dev *pdev;
2124 if (!parent || !dev_is_pci(parent))
2125 return NULL; /* not a PCI device */
2127 pdev = to_pci_dev(parent);
2129 netdev_notice(vf_netdev, "no PCI slot information\n");
2133 if (kstrtou32(pci_slot_name(pdev->slot), 10, &serial)) {
2134 netdev_notice(vf_netdev, "Invalid vf serial:%s\n",
2135 pci_slot_name(pdev->slot));
2139 list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
2140 if (!ndev_ctx->vf_alloc)
2143 if (ndev_ctx->vf_serial == serial)
2144 return hv_get_drvdata(ndev_ctx->device_ctx);
2147 netdev_notice(vf_netdev,
2148 "no netdev found for vf serial:%u\n", serial);
2152 static int netvsc_register_vf(struct net_device *vf_netdev)
2154 struct net_device_context *net_device_ctx;
2155 struct netvsc_device *netvsc_dev;
2156 struct net_device *ndev;
2159 if (vf_netdev->addr_len != ETH_ALEN)
2162 ndev = get_netvsc_byslot(vf_netdev);
2166 net_device_ctx = netdev_priv(ndev);
2167 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2168 if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
2171 /* if syntihetic interface is a different namespace,
2172 * then move the VF to that namespace; join will be
2173 * done again in that context.
2175 if (!net_eq(dev_net(ndev), dev_net(vf_netdev))) {
2176 ret = dev_change_net_namespace(vf_netdev,
2177 dev_net(ndev), "eth%d");
2179 netdev_err(vf_netdev,
2180 "could not move to same namespace as %s: %d\n",
2183 netdev_info(vf_netdev,
2184 "VF moved to namespace with: %s\n",
2189 netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
2191 if (netvsc_vf_join(vf_netdev, ndev) != 0)
2194 dev_hold(vf_netdev);
2195 rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
2199 /* VF up/down change detected, schedule to change data path */
2200 static int netvsc_vf_changed(struct net_device *vf_netdev)
2202 struct net_device_context *net_device_ctx;
2203 struct netvsc_device *netvsc_dev;
2204 struct net_device *ndev;
2205 bool vf_is_up = netif_running(vf_netdev);
2207 ndev = get_netvsc_byref(vf_netdev);
2211 net_device_ctx = netdev_priv(ndev);
2212 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2216 netvsc_switch_datapath(ndev, vf_is_up);
2217 netdev_info(ndev, "Data path switched %s VF: %s\n",
2218 vf_is_up ? "to" : "from", vf_netdev->name);
2223 static int netvsc_unregister_vf(struct net_device *vf_netdev)
2225 struct net_device *ndev;
2226 struct net_device_context *net_device_ctx;
2228 ndev = get_netvsc_byref(vf_netdev);
2232 net_device_ctx = netdev_priv(ndev);
2233 cancel_delayed_work_sync(&net_device_ctx->vf_takeover);
2235 netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
2237 netdev_rx_handler_unregister(vf_netdev);
2238 netdev_upper_dev_unlink(vf_netdev, ndev);
2239 RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
2245 static int netvsc_probe(struct hv_device *dev,
2246 const struct hv_vmbus_device_id *dev_id)
2248 struct net_device *net = NULL;
2249 struct net_device_context *net_device_ctx;
2250 struct netvsc_device_info *device_info = NULL;
2251 struct netvsc_device *nvdev;
2254 net = alloc_etherdev_mq(sizeof(struct net_device_context),
2259 netif_carrier_off(net);
2261 netvsc_init_settings(net);
2263 net_device_ctx = netdev_priv(net);
2264 net_device_ctx->device_ctx = dev;
2265 net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
2266 if (netif_msg_probe(net_device_ctx))
2267 netdev_dbg(net, "netvsc msg_enable: %d\n",
2268 net_device_ctx->msg_enable);
2270 hv_set_drvdata(dev, net);
2272 INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
2274 spin_lock_init(&net_device_ctx->lock);
2275 INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
2276 INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup);
2278 net_device_ctx->vf_stats
2279 = netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats);
2280 if (!net_device_ctx->vf_stats)
2283 net->netdev_ops = &device_ops;
2284 net->ethtool_ops = ðtool_ops;
2285 SET_NETDEV_DEV(net, &dev->device);
2287 /* We always need headroom for rndis header */
2288 net->needed_headroom = RNDIS_AND_PPI_SIZE;
2290 /* Initialize the number of queues to be 1, we may change it if more
2291 * channels are offered later.
2293 netif_set_real_num_tx_queues(net, 1);
2294 netif_set_real_num_rx_queues(net, 1);
2296 /* Notify the netvsc driver of the new device */
2297 device_info = netvsc_devinfo_get(NULL);
2301 goto devinfo_failed;
2304 nvdev = rndis_filter_device_add(dev, device_info);
2305 if (IS_ERR(nvdev)) {
2306 ret = PTR_ERR(nvdev);
2307 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
2311 memcpy(net->dev_addr, device_info->mac_adr, ETH_ALEN);
2313 /* We must get rtnl lock before scheduling nvdev->subchan_work,
2314 * otherwise netvsc_subchan_work() can get rtnl lock first and wait
2315 * all subchannels to show up, but that may not happen because
2316 * netvsc_probe() can't get rtnl lock and as a result vmbus_onoffer()
2317 * -> ... -> device_add() -> ... -> __device_attach() can't get
2318 * the device lock, so all the subchannels can't be processed --
2319 * finally netvsc_subchan_work() hangs for ever.
2323 if (nvdev->num_chn > 1)
2324 schedule_work(&nvdev->subchan_work);
2326 /* hw_features computed in rndis_netdev_set_hwcaps() */
2327 net->features = net->hw_features |
2328 NETIF_F_HIGHDMA | NETIF_F_SG |
2329 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
2330 net->vlan_features = net->features;
2332 netdev_lockdep_set_classes(net);
2334 /* MTU range: 68 - 1500 or 65521 */
2335 net->min_mtu = NETVSC_MTU_MIN;
2336 if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
2337 net->max_mtu = NETVSC_MTU - ETH_HLEN;
2339 net->max_mtu = ETH_DATA_LEN;
2341 nvdev->tx_disable = false;
2343 ret = register_netdevice(net);
2345 pr_err("Unable to register netdev.\n");
2346 goto register_failed;
2349 list_add(&net_device_ctx->list, &netvsc_dev_list);
2357 rndis_filter_device_remove(dev, nvdev);
2361 free_percpu(net_device_ctx->vf_stats);
2363 hv_set_drvdata(dev, NULL);
2369 static int netvsc_remove(struct hv_device *dev)
2371 struct net_device_context *ndev_ctx;
2372 struct net_device *vf_netdev, *net;
2373 struct netvsc_device *nvdev;
2375 net = hv_get_drvdata(dev);
2377 dev_err(&dev->device, "No net device to remove\n");
2381 ndev_ctx = netdev_priv(net);
2383 cancel_delayed_work_sync(&ndev_ctx->dwork);
2386 nvdev = rtnl_dereference(ndev_ctx->nvdev);
2388 cancel_work_sync(&nvdev->subchan_work);
2391 * Call to the vsc driver to let it know that the device is being
2392 * removed. Also blocks mtu and channel changes.
2394 vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2396 netvsc_unregister_vf(vf_netdev);
2399 rndis_filter_device_remove(dev, nvdev);
2401 unregister_netdevice(net);
2402 list_del(&ndev_ctx->list);
2406 hv_set_drvdata(dev, NULL);
2408 free_percpu(ndev_ctx->vf_stats);
2413 static const struct hv_vmbus_device_id id_table[] = {
2419 MODULE_DEVICE_TABLE(vmbus, id_table);
2421 /* The one and only one */
2422 static struct hv_driver netvsc_drv = {
2423 .name = KBUILD_MODNAME,
2424 .id_table = id_table,
2425 .probe = netvsc_probe,
2426 .remove = netvsc_remove,
2428 .probe_type = PROBE_FORCE_SYNCHRONOUS,
2433 * On Hyper-V, every VF interface is matched with a corresponding
2434 * synthetic interface. The synthetic interface is presented first
2435 * to the guest. When the corresponding VF instance is registered,
2436 * we will take care of switching the data path.
2438 static int netvsc_netdev_event(struct notifier_block *this,
2439 unsigned long event, void *ptr)
2441 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2443 /* Skip our own events */
2444 if (event_dev->netdev_ops == &device_ops)
2447 /* Avoid non-Ethernet type devices */
2448 if (event_dev->type != ARPHRD_ETHER)
2451 /* Avoid Vlan dev with same MAC registering as VF */
2452 if (is_vlan_dev(event_dev))
2455 /* Avoid Bonding master dev with same MAC registering as VF */
2456 if ((event_dev->priv_flags & IFF_BONDING) &&
2457 (event_dev->flags & IFF_MASTER))
2461 case NETDEV_REGISTER:
2462 return netvsc_register_vf(event_dev);
2463 case NETDEV_UNREGISTER:
2464 return netvsc_unregister_vf(event_dev);
2467 return netvsc_vf_changed(event_dev);
2473 static struct notifier_block netvsc_netdev_notifier = {
2474 .notifier_call = netvsc_netdev_event,
2477 static void __exit netvsc_drv_exit(void)
2479 unregister_netdevice_notifier(&netvsc_netdev_notifier);
2480 vmbus_driver_unregister(&netvsc_drv);
2483 static int __init netvsc_drv_init(void)
2487 if (ring_size < RING_SIZE_MIN) {
2488 ring_size = RING_SIZE_MIN;
2489 pr_info("Increased ring_size to %u (min allowed)\n",
2492 netvsc_ring_bytes = ring_size * PAGE_SIZE;
2494 ret = vmbus_driver_register(&netvsc_drv);
2498 register_netdevice_notifier(&netvsc_netdev_notifier);
2502 MODULE_LICENSE("GPL");
2503 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
2505 module_init(netvsc_drv_init);
2506 module_exit(netvsc_drv_exit);