2 * Copyright (c) 2009, Microsoft Corporation.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * You should have received a copy of the GNU General Public License along with
14 * this program; if not, see <http://www.gnu.org/licenses/>.
17 * Haiyang Zhang <haiyangz@microsoft.com>
18 * Hank Janssen <hjanssen@microsoft.com>
20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
22 #include <linux/init.h>
23 #include <linux/atomic.h>
24 #include <linux/module.h>
25 #include <linux/highmem.h>
26 #include <linux/device.h>
28 #include <linux/delay.h>
29 #include <linux/netdevice.h>
30 #include <linux/inetdevice.h>
31 #include <linux/etherdevice.h>
32 #include <linux/skbuff.h>
33 #include <linux/if_vlan.h>
35 #include <linux/slab.h>
36 #include <linux/rtnetlink.h>
37 #include <linux/netpoll.h>
38 #include <linux/reciprocal_div.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, S_IRUGO);
59 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
60 unsigned int netvsc_ring_bytes __ro_after_init;
61 struct reciprocal_value netvsc_ring_reciprocal __ro_after_init;
63 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
64 NETIF_MSG_LINK | NETIF_MSG_IFUP |
65 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
68 static int debug = -1;
69 module_param(debug, int, S_IRUGO);
70 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
72 static void netvsc_change_rx_flags(struct net_device *net, int change)
74 struct net_device_context *ndev_ctx = netdev_priv(net);
75 struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
81 if (change & IFF_PROMISC) {
82 inc = (net->flags & IFF_PROMISC) ? 1 : -1;
83 dev_set_promiscuity(vf_netdev, inc);
86 if (change & IFF_ALLMULTI) {
87 inc = (net->flags & IFF_ALLMULTI) ? 1 : -1;
88 dev_set_allmulti(vf_netdev, inc);
92 static void netvsc_set_rx_mode(struct net_device *net)
94 struct net_device_context *ndev_ctx = netdev_priv(net);
95 struct net_device *vf_netdev;
96 struct netvsc_device *nvdev;
99 vf_netdev = rcu_dereference(ndev_ctx->vf_netdev);
101 dev_uc_sync(vf_netdev, net);
102 dev_mc_sync(vf_netdev, net);
105 nvdev = rcu_dereference(ndev_ctx->nvdev);
107 rndis_filter_update(nvdev);
111 static int netvsc_open(struct net_device *net)
113 struct net_device_context *ndev_ctx = netdev_priv(net);
114 struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
115 struct netvsc_device *nvdev = rtnl_dereference(ndev_ctx->nvdev);
116 struct rndis_device *rdev;
119 netif_carrier_off(net);
121 /* Open up the device */
122 ret = rndis_filter_open(nvdev);
124 netdev_err(net, "unable to open device (ret %d).\n", ret);
128 rdev = nvdev->extension;
129 if (!rdev->link_state)
130 netif_carrier_on(net);
133 /* Setting synthetic device up transparently sets
134 * slave as up. If open fails, then slave will be
135 * still be offline (and not used).
137 ret = dev_open(vf_netdev);
140 "unable to open slave: %s: %d\n",
141 vf_netdev->name, ret);
146 static int netvsc_wait_until_empty(struct netvsc_device *nvdev)
148 unsigned int retry = 0;
151 /* Ensure pending bytes in ring are read */
155 for (i = 0; i < nvdev->num_chn; i++) {
156 struct vmbus_channel *chn
157 = nvdev->chan_table[i].channel;
162 /* make sure receive not running now */
163 napi_synchronize(&nvdev->chan_table[i].napi);
165 aread = hv_get_bytes_to_read(&chn->inbound);
169 aread = hv_get_bytes_to_read(&chn->outbound);
177 if (++retry > RETRY_MAX)
180 usleep_range(RETRY_US_LO, RETRY_US_HI);
184 static int netvsc_close(struct net_device *net)
186 struct net_device_context *net_device_ctx = netdev_priv(net);
187 struct net_device *vf_netdev
188 = rtnl_dereference(net_device_ctx->vf_netdev);
189 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
192 netif_tx_disable(net);
194 /* No need to close rndis filter if it is removed already */
198 ret = rndis_filter_close(nvdev);
200 netdev_err(net, "unable to close device (ret %d).\n", ret);
204 ret = netvsc_wait_until_empty(nvdev);
206 netdev_err(net, "Ring buffer not empty after closing rndis\n");
209 dev_close(vf_netdev);
214 static inline void *init_ppi_data(struct rndis_message *msg,
215 u32 ppi_size, u32 pkt_type)
217 struct rndis_packet *rndis_pkt = &msg->msg.pkt;
218 struct rndis_per_packet_info *ppi;
220 rndis_pkt->data_offset += ppi_size;
221 ppi = (void *)rndis_pkt + rndis_pkt->per_pkt_info_offset
222 + rndis_pkt->per_pkt_info_len;
224 ppi->size = ppi_size;
225 ppi->type = pkt_type;
226 ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
228 rndis_pkt->per_pkt_info_len += ppi_size;
233 /* Azure hosts don't support non-TCP port numbers in hashing for fragmented
234 * packets. We can use ethtool to change UDP hash level when necessary.
236 static inline u32 netvsc_get_hash(
238 const struct net_device_context *ndc)
240 struct flow_keys flow;
241 u32 hash, pkt_proto = 0;
242 static u32 hashrnd __read_mostly;
244 net_get_random_once(&hashrnd, sizeof(hashrnd));
246 if (!skb_flow_dissect_flow_keys(skb, &flow, 0))
249 switch (flow.basic.ip_proto) {
251 if (flow.basic.n_proto == htons(ETH_P_IP))
252 pkt_proto = HV_TCP4_L4HASH;
253 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
254 pkt_proto = HV_TCP6_L4HASH;
259 if (flow.basic.n_proto == htons(ETH_P_IP))
260 pkt_proto = HV_UDP4_L4HASH;
261 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
262 pkt_proto = HV_UDP6_L4HASH;
267 if (pkt_proto & ndc->l4_hash) {
268 return skb_get_hash(skb);
270 if (flow.basic.n_proto == htons(ETH_P_IP))
271 hash = jhash2((u32 *)&flow.addrs.v4addrs, 2, hashrnd);
272 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
273 hash = jhash2((u32 *)&flow.addrs.v6addrs, 8, hashrnd);
277 skb_set_hash(skb, hash, PKT_HASH_TYPE_L3);
283 static inline int netvsc_get_tx_queue(struct net_device *ndev,
284 struct sk_buff *skb, int old_idx)
286 const struct net_device_context *ndc = netdev_priv(ndev);
287 struct sock *sk = skb->sk;
290 q_idx = ndc->tx_table[netvsc_get_hash(skb, ndc) &
291 (VRSS_SEND_TAB_SIZE - 1)];
293 /* If queue index changed record the new value */
294 if (q_idx != old_idx &&
295 sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
296 sk_tx_queue_set(sk, q_idx);
302 * Select queue for transmit.
304 * If a valid queue has already been assigned, then use that.
305 * Otherwise compute tx queue based on hash and the send table.
307 * This is basically similar to default (__netdev_pick_tx) with the added step
308 * of using the host send_table when no other queue has been assigned.
310 * TODO support XPS - but get_xps_queue not exported
312 static u16 netvsc_pick_tx(struct net_device *ndev, struct sk_buff *skb)
314 int q_idx = sk_tx_queue_get(skb->sk);
316 if (q_idx < 0 || skb->ooo_okay || q_idx >= ndev->real_num_tx_queues) {
317 /* If forwarding a packet, we use the recorded queue when
318 * available for better cache locality.
320 if (skb_rx_queue_recorded(skb))
321 q_idx = skb_get_rx_queue(skb);
323 q_idx = netvsc_get_tx_queue(ndev, skb, q_idx);
329 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
331 select_queue_fallback_t fallback)
333 struct net_device_context *ndc = netdev_priv(ndev);
334 struct net_device *vf_netdev;
338 vf_netdev = rcu_dereference(ndc->vf_netdev);
340 const struct net_device_ops *vf_ops = vf_netdev->netdev_ops;
342 if (vf_ops->ndo_select_queue)
343 txq = vf_ops->ndo_select_queue(vf_netdev, skb,
344 accel_priv, fallback);
346 txq = fallback(vf_netdev, skb);
348 /* Record the queue selected by VF so that it can be
349 * used for common case where VF has more queues than
350 * the synthetic device.
352 qdisc_skb_cb(skb)->slave_dev_queue_mapping = txq;
354 txq = netvsc_pick_tx(ndev, skb);
358 while (unlikely(txq >= ndev->real_num_tx_queues))
359 txq -= ndev->real_num_tx_queues;
364 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
365 struct hv_page_buffer *pb)
369 /* Deal with compund pages by ignoring unused part
372 page += (offset >> PAGE_SHIFT);
373 offset &= ~PAGE_MASK;
378 bytes = PAGE_SIZE - offset;
381 pb[j].pfn = page_to_pfn(page);
382 pb[j].offset = offset;
388 if (offset == PAGE_SIZE && len) {
398 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
399 struct hv_netvsc_packet *packet,
400 struct hv_page_buffer *pb)
403 char *data = skb->data;
404 int frags = skb_shinfo(skb)->nr_frags;
407 /* The packet is laid out thus:
408 * 1. hdr: RNDIS header and PPI
410 * 3. skb fragment data
412 slots_used += fill_pg_buf(virt_to_page(hdr),
414 len, &pb[slots_used]);
416 packet->rmsg_size = len;
417 packet->rmsg_pgcnt = slots_used;
419 slots_used += fill_pg_buf(virt_to_page(data),
420 offset_in_page(data),
421 skb_headlen(skb), &pb[slots_used]);
423 for (i = 0; i < frags; i++) {
424 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
426 slots_used += fill_pg_buf(skb_frag_page(frag),
428 skb_frag_size(frag), &pb[slots_used]);
433 static int count_skb_frag_slots(struct sk_buff *skb)
435 int i, frags = skb_shinfo(skb)->nr_frags;
438 for (i = 0; i < frags; i++) {
439 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
440 unsigned long size = skb_frag_size(frag);
441 unsigned long offset = frag->page_offset;
443 /* Skip unused frames from start of page */
444 offset &= ~PAGE_MASK;
445 pages += PFN_UP(offset + size);
450 static int netvsc_get_slots(struct sk_buff *skb)
452 char *data = skb->data;
453 unsigned int offset = offset_in_page(data);
454 unsigned int len = skb_headlen(skb);
458 slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
459 frag_slots = count_skb_frag_slots(skb);
460 return slots + frag_slots;
463 static u32 net_checksum_info(struct sk_buff *skb)
465 if (skb->protocol == htons(ETH_P_IP)) {
466 struct iphdr *ip = ip_hdr(skb);
468 if (ip->protocol == IPPROTO_TCP)
469 return TRANSPORT_INFO_IPV4_TCP;
470 else if (ip->protocol == IPPROTO_UDP)
471 return TRANSPORT_INFO_IPV4_UDP;
473 struct ipv6hdr *ip6 = ipv6_hdr(skb);
475 if (ip6->nexthdr == IPPROTO_TCP)
476 return TRANSPORT_INFO_IPV6_TCP;
477 else if (ip6->nexthdr == IPPROTO_UDP)
478 return TRANSPORT_INFO_IPV6_UDP;
481 return TRANSPORT_INFO_NOT_IP;
484 /* Send skb on the slave VF device. */
485 static int netvsc_vf_xmit(struct net_device *net, struct net_device *vf_netdev,
488 struct net_device_context *ndev_ctx = netdev_priv(net);
489 unsigned int len = skb->len;
492 skb->dev = vf_netdev;
493 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
495 rc = dev_queue_xmit(skb);
496 if (likely(rc == NET_XMIT_SUCCESS || rc == NET_XMIT_CN)) {
497 struct netvsc_vf_pcpu_stats *pcpu_stats
498 = this_cpu_ptr(ndev_ctx->vf_stats);
500 u64_stats_update_begin(&pcpu_stats->syncp);
501 pcpu_stats->tx_packets++;
502 pcpu_stats->tx_bytes += len;
503 u64_stats_update_end(&pcpu_stats->syncp);
505 this_cpu_inc(ndev_ctx->vf_stats->tx_dropped);
511 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
513 struct net_device_context *net_device_ctx = netdev_priv(net);
514 struct hv_netvsc_packet *packet = NULL;
516 unsigned int num_data_pgs;
517 struct rndis_message *rndis_msg;
518 struct net_device *vf_netdev;
521 struct hv_page_buffer pb[MAX_PAGE_BUFFER_COUNT];
523 /* if VF is present and up then redirect packets
524 * already called with rcu_read_lock_bh
526 vf_netdev = rcu_dereference_bh(net_device_ctx->vf_netdev);
527 if (vf_netdev && netif_running(vf_netdev) &&
528 !netpoll_tx_running(net))
529 return netvsc_vf_xmit(net, vf_netdev, skb);
531 /* We will atmost need two pages to describe the rndis
532 * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
533 * of pages in a single packet. If skb is scattered around
534 * more pages we try linearizing it.
537 num_data_pgs = netvsc_get_slots(skb) + 2;
539 if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
540 ++net_device_ctx->eth_stats.tx_scattered;
542 if (skb_linearize(skb))
545 num_data_pgs = netvsc_get_slots(skb) + 2;
546 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
547 ++net_device_ctx->eth_stats.tx_too_big;
553 * Place the rndis header in the skb head room and
554 * the skb->cb will be used for hv_netvsc_packet
557 ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
561 /* Use the skb control buffer for building up the packet */
562 BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
563 FIELD_SIZEOF(struct sk_buff, cb));
564 packet = (struct hv_netvsc_packet *)skb->cb;
566 packet->q_idx = skb_get_queue_mapping(skb);
568 packet->total_data_buflen = skb->len;
569 packet->total_bytes = skb->len;
570 packet->total_packets = 1;
572 rndis_msg = (struct rndis_message *)skb->head;
574 /* Add the rndis header */
575 rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
576 rndis_msg->msg_len = packet->total_data_buflen;
578 rndis_msg->msg.pkt = (struct rndis_packet) {
579 .data_offset = sizeof(struct rndis_packet),
580 .data_len = packet->total_data_buflen,
581 .per_pkt_info_offset = sizeof(struct rndis_packet),
584 rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
586 hash = skb_get_hash_raw(skb);
587 if (hash != 0 && net->real_num_tx_queues > 1) {
590 rndis_msg_size += NDIS_HASH_PPI_SIZE;
591 hash_info = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
596 if (skb_vlan_tag_present(skb)) {
597 struct ndis_pkt_8021q_info *vlan;
599 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
600 vlan = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
604 vlan->vlanid = skb->vlan_tci & VLAN_VID_MASK;
605 vlan->pri = (skb->vlan_tci & VLAN_PRIO_MASK) >>
609 if (skb_is_gso(skb)) {
610 struct ndis_tcp_lso_info *lso_info;
612 rndis_msg_size += NDIS_LSO_PPI_SIZE;
613 lso_info = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
614 TCP_LARGESEND_PKTINFO);
617 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
618 if (skb->protocol == htons(ETH_P_IP)) {
619 lso_info->lso_v2_transmit.ip_version =
620 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
621 ip_hdr(skb)->tot_len = 0;
622 ip_hdr(skb)->check = 0;
623 tcp_hdr(skb)->check =
624 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
625 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
627 lso_info->lso_v2_transmit.ip_version =
628 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
629 ipv6_hdr(skb)->payload_len = 0;
630 tcp_hdr(skb)->check =
631 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
632 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
634 lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
635 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
636 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
637 if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
638 struct ndis_tcp_ip_checksum_info *csum_info;
640 rndis_msg_size += NDIS_CSUM_PPI_SIZE;
641 csum_info = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
642 TCPIP_CHKSUM_PKTINFO);
644 csum_info->value = 0;
645 csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
647 if (skb->protocol == htons(ETH_P_IP)) {
648 csum_info->transmit.is_ipv4 = 1;
650 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
651 csum_info->transmit.tcp_checksum = 1;
653 csum_info->transmit.udp_checksum = 1;
655 csum_info->transmit.is_ipv6 = 1;
657 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
658 csum_info->transmit.tcp_checksum = 1;
660 csum_info->transmit.udp_checksum = 1;
663 /* Can't do offload of this type of checksum */
664 if (skb_checksum_help(skb))
669 /* Start filling in the page buffers with the rndis hdr */
670 rndis_msg->msg_len += rndis_msg_size;
671 packet->total_data_buflen = rndis_msg->msg_len;
672 packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
675 /* timestamp packet in software */
676 skb_tx_timestamp(skb);
678 ret = netvsc_send(net, packet, rndis_msg, pb, skb);
679 if (likely(ret == 0))
682 if (ret == -EAGAIN) {
683 ++net_device_ctx->eth_stats.tx_busy;
684 return NETDEV_TX_BUSY;
688 ++net_device_ctx->eth_stats.tx_no_space;
691 dev_kfree_skb_any(skb);
692 net->stats.tx_dropped++;
697 ++net_device_ctx->eth_stats.tx_no_memory;
702 * netvsc_linkstatus_callback - Link up/down notification
704 void netvsc_linkstatus_callback(struct net_device *net,
705 struct rndis_message *resp)
707 struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
708 struct net_device_context *ndev_ctx = netdev_priv(net);
709 struct netvsc_reconfig *event;
712 /* Update the physical link speed when changing to another vSwitch */
713 if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
716 speed = *(u32 *)((void *)indicate
717 + indicate->status_buf_offset) / 10000;
718 ndev_ctx->speed = speed;
722 /* Handle these link change statuses below */
723 if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
724 indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
725 indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
728 if (net->reg_state != NETREG_REGISTERED)
731 event = kzalloc(sizeof(*event), GFP_ATOMIC);
734 event->event = indicate->status;
736 spin_lock_irqsave(&ndev_ctx->lock, flags);
737 list_add_tail(&event->list, &ndev_ctx->reconfig_events);
738 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
740 schedule_delayed_work(&ndev_ctx->dwork, 0);
743 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
744 struct napi_struct *napi,
745 const struct ndis_tcp_ip_checksum_info *csum_info,
746 const struct ndis_pkt_8021q_info *vlan,
747 void *data, u32 buflen)
751 skb = napi_alloc_skb(napi, buflen);
756 * Copy to skb. This copy is needed here since the memory pointed by
757 * hv_netvsc_packet cannot be deallocated
759 skb_put_data(skb, data, buflen);
761 skb->protocol = eth_type_trans(skb, net);
763 /* skb is already created with CHECKSUM_NONE */
764 skb_checksum_none_assert(skb);
767 * In Linux, the IP checksum is always checked.
768 * Do L4 checksum offload if enabled and present.
770 if (csum_info && (net->features & NETIF_F_RXCSUM)) {
771 if (csum_info->receive.tcp_checksum_succeeded ||
772 csum_info->receive.udp_checksum_succeeded)
773 skb->ip_summed = CHECKSUM_UNNECESSARY;
777 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT);
779 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
787 * netvsc_recv_callback - Callback when we receive a packet from the
788 * "wire" on the specified device.
790 int netvsc_recv_callback(struct net_device *net,
791 struct netvsc_device *net_device,
792 struct vmbus_channel *channel,
794 const struct ndis_tcp_ip_checksum_info *csum_info,
795 const struct ndis_pkt_8021q_info *vlan)
797 struct net_device_context *net_device_ctx = netdev_priv(net);
798 u16 q_idx = channel->offermsg.offer.sub_channel_index;
799 struct netvsc_channel *nvchan = &net_device->chan_table[q_idx];
801 struct netvsc_stats *rx_stats;
803 if (net->reg_state != NETREG_REGISTERED)
804 return NVSP_STAT_FAIL;
806 /* Allocate a skb - TODO direct I/O to pages? */
807 skb = netvsc_alloc_recv_skb(net, &nvchan->napi,
808 csum_info, vlan, data, len);
809 if (unlikely(!skb)) {
810 ++net_device_ctx->eth_stats.rx_no_memory;
812 return NVSP_STAT_FAIL;
815 skb_record_rx_queue(skb, q_idx);
818 * Even if injecting the packet, record the statistics
819 * on the synthetic device because modifying the VF device
820 * statistics will not work correctly.
822 rx_stats = &nvchan->rx_stats;
823 u64_stats_update_begin(&rx_stats->syncp);
825 rx_stats->bytes += len;
827 if (skb->pkt_type == PACKET_BROADCAST)
828 ++rx_stats->broadcast;
829 else if (skb->pkt_type == PACKET_MULTICAST)
830 ++rx_stats->multicast;
831 u64_stats_update_end(&rx_stats->syncp);
833 napi_gro_receive(&nvchan->napi, skb);
837 static void netvsc_get_drvinfo(struct net_device *net,
838 struct ethtool_drvinfo *info)
840 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
841 strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
844 static void netvsc_get_channels(struct net_device *net,
845 struct ethtool_channels *channel)
847 struct net_device_context *net_device_ctx = netdev_priv(net);
848 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
851 channel->max_combined = nvdev->max_chn;
852 channel->combined_count = nvdev->num_chn;
856 static int netvsc_detach(struct net_device *ndev,
857 struct netvsc_device *nvdev)
859 struct net_device_context *ndev_ctx = netdev_priv(ndev);
860 struct hv_device *hdev = ndev_ctx->device_ctx;
863 /* Don't try continuing to try and setup sub channels */
864 if (cancel_work_sync(&nvdev->subchan_work))
867 /* If device was up (receiving) then shutdown */
868 if (netif_running(ndev)) {
869 netif_tx_disable(ndev);
871 ret = rndis_filter_close(nvdev);
874 "unable to close device (ret %d).\n", ret);
878 ret = netvsc_wait_until_empty(nvdev);
881 "Ring buffer not empty after closing rndis\n");
886 netif_device_detach(ndev);
888 rndis_filter_device_remove(hdev, nvdev);
893 static int netvsc_attach(struct net_device *ndev,
894 struct netvsc_device_info *dev_info)
896 struct net_device_context *ndev_ctx = netdev_priv(ndev);
897 struct hv_device *hdev = ndev_ctx->device_ctx;
898 struct netvsc_device *nvdev;
899 struct rndis_device *rdev;
902 nvdev = rndis_filter_device_add(hdev, dev_info);
904 return PTR_ERR(nvdev);
906 /* Note: enable and attach happen when sub-channels setup */
908 netif_carrier_off(ndev);
910 if (netif_running(ndev)) {
911 ret = rndis_filter_open(nvdev);
915 rdev = nvdev->extension;
916 if (!rdev->link_state)
917 netif_carrier_on(ndev);
923 static int netvsc_set_channels(struct net_device *net,
924 struct ethtool_channels *channels)
926 struct net_device_context *net_device_ctx = netdev_priv(net);
927 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
928 unsigned int orig, count = channels->combined_count;
929 struct netvsc_device_info device_info;
932 /* We do not support separate count for rx, tx, or other */
934 channels->rx_count || channels->tx_count || channels->other_count)
937 if (!nvdev || nvdev->destroy)
940 if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
943 if (count > nvdev->max_chn)
946 orig = nvdev->num_chn;
948 memset(&device_info, 0, sizeof(device_info));
949 device_info.num_chn = count;
950 device_info.send_sections = nvdev->send_section_cnt;
951 device_info.send_section_size = nvdev->send_section_size;
952 device_info.recv_sections = nvdev->recv_section_cnt;
953 device_info.recv_section_size = nvdev->recv_section_size;
955 ret = netvsc_detach(net, nvdev);
959 ret = netvsc_attach(net, &device_info);
961 device_info.num_chn = orig;
962 if (netvsc_attach(net, &device_info))
963 netdev_err(net, "restoring channel setting failed\n");
970 netvsc_validate_ethtool_ss_cmd(const struct ethtool_link_ksettings *cmd)
972 struct ethtool_link_ksettings diff1 = *cmd;
973 struct ethtool_link_ksettings diff2 = {};
975 diff1.base.speed = 0;
976 diff1.base.duplex = 0;
977 /* advertising and cmd are usually set */
978 ethtool_link_ksettings_zero_link_mode(&diff1, advertising);
980 /* We set port to PORT_OTHER */
981 diff2.base.port = PORT_OTHER;
983 return !memcmp(&diff1, &diff2, sizeof(diff1));
986 static void netvsc_init_settings(struct net_device *dev)
988 struct net_device_context *ndc = netdev_priv(dev);
990 ndc->l4_hash = HV_DEFAULT_L4HASH;
992 ndc->speed = SPEED_UNKNOWN;
993 ndc->duplex = DUPLEX_FULL;
996 static int netvsc_get_link_ksettings(struct net_device *dev,
997 struct ethtool_link_ksettings *cmd)
999 struct net_device_context *ndc = netdev_priv(dev);
1001 cmd->base.speed = ndc->speed;
1002 cmd->base.duplex = ndc->duplex;
1003 cmd->base.port = PORT_OTHER;
1008 static int netvsc_set_link_ksettings(struct net_device *dev,
1009 const struct ethtool_link_ksettings *cmd)
1011 struct net_device_context *ndc = netdev_priv(dev);
1014 speed = cmd->base.speed;
1015 if (!ethtool_validate_speed(speed) ||
1016 !ethtool_validate_duplex(cmd->base.duplex) ||
1017 !netvsc_validate_ethtool_ss_cmd(cmd))
1021 ndc->duplex = cmd->base.duplex;
1026 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
1028 struct net_device_context *ndevctx = netdev_priv(ndev);
1029 struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1030 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1031 int orig_mtu = ndev->mtu;
1032 struct netvsc_device_info device_info;
1035 if (!nvdev || nvdev->destroy)
1038 /* Change MTU of underlying VF netdev first. */
1040 ret = dev_set_mtu(vf_netdev, mtu);
1045 memset(&device_info, 0, sizeof(device_info));
1046 device_info.num_chn = nvdev->num_chn;
1047 device_info.send_sections = nvdev->send_section_cnt;
1048 device_info.send_section_size = nvdev->send_section_size;
1049 device_info.recv_sections = nvdev->recv_section_cnt;
1050 device_info.recv_section_size = nvdev->recv_section_size;
1052 ret = netvsc_detach(ndev, nvdev);
1058 ret = netvsc_attach(ndev, &device_info);
1065 /* Attempt rollback to original MTU */
1066 ndev->mtu = orig_mtu;
1068 if (netvsc_attach(ndev, &device_info))
1069 netdev_err(ndev, "restoring mtu failed\n");
1072 dev_set_mtu(vf_netdev, orig_mtu);
1077 static void netvsc_get_vf_stats(struct net_device *net,
1078 struct netvsc_vf_pcpu_stats *tot)
1080 struct net_device_context *ndev_ctx = netdev_priv(net);
1083 memset(tot, 0, sizeof(*tot));
1085 for_each_possible_cpu(i) {
1086 const struct netvsc_vf_pcpu_stats *stats
1087 = per_cpu_ptr(ndev_ctx->vf_stats, i);
1088 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1092 start = u64_stats_fetch_begin_irq(&stats->syncp);
1093 rx_packets = stats->rx_packets;
1094 tx_packets = stats->tx_packets;
1095 rx_bytes = stats->rx_bytes;
1096 tx_bytes = stats->tx_bytes;
1097 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1099 tot->rx_packets += rx_packets;
1100 tot->tx_packets += tx_packets;
1101 tot->rx_bytes += rx_bytes;
1102 tot->tx_bytes += tx_bytes;
1103 tot->tx_dropped += stats->tx_dropped;
1107 static void netvsc_get_stats64(struct net_device *net,
1108 struct rtnl_link_stats64 *t)
1110 struct net_device_context *ndev_ctx = netdev_priv(net);
1111 struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1112 struct netvsc_vf_pcpu_stats vf_tot;
1118 netdev_stats_to_stats64(t, &net->stats);
1120 netvsc_get_vf_stats(net, &vf_tot);
1121 t->rx_packets += vf_tot.rx_packets;
1122 t->tx_packets += vf_tot.tx_packets;
1123 t->rx_bytes += vf_tot.rx_bytes;
1124 t->tx_bytes += vf_tot.tx_bytes;
1125 t->tx_dropped += vf_tot.tx_dropped;
1127 for (i = 0; i < nvdev->num_chn; i++) {
1128 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1129 const struct netvsc_stats *stats;
1130 u64 packets, bytes, multicast;
1133 stats = &nvchan->tx_stats;
1135 start = u64_stats_fetch_begin_irq(&stats->syncp);
1136 packets = stats->packets;
1137 bytes = stats->bytes;
1138 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1140 t->tx_bytes += bytes;
1141 t->tx_packets += packets;
1143 stats = &nvchan->rx_stats;
1145 start = u64_stats_fetch_begin_irq(&stats->syncp);
1146 packets = stats->packets;
1147 bytes = stats->bytes;
1148 multicast = stats->multicast + stats->broadcast;
1149 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1151 t->rx_bytes += bytes;
1152 t->rx_packets += packets;
1153 t->multicast += multicast;
1157 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1159 struct net_device_context *ndc = netdev_priv(ndev);
1160 struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1161 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1162 struct sockaddr *addr = p;
1165 err = eth_prepare_mac_addr_change(ndev, p);
1173 err = dev_set_mac_address(vf_netdev, addr);
1178 err = rndis_filter_set_device_mac(nvdev, addr->sa_data);
1180 eth_commit_mac_addr_change(ndev, p);
1181 } else if (vf_netdev) {
1182 /* rollback change on VF */
1183 memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN);
1184 dev_set_mac_address(vf_netdev, addr);
1190 static const struct {
1191 char name[ETH_GSTRING_LEN];
1193 } netvsc_stats[] = {
1194 { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1195 { "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1196 { "tx_no_space", offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1197 { "tx_too_big", offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1198 { "tx_busy", offsetof(struct netvsc_ethtool_stats, tx_busy) },
1199 { "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) },
1200 { "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) },
1201 { "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) },
1202 { "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) },
1203 { "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) },
1205 { "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) },
1206 { "vf_rx_bytes", offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) },
1207 { "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) },
1208 { "vf_tx_bytes", offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) },
1209 { "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) },
1212 #define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats)
1213 #define NETVSC_VF_STATS_LEN ARRAY_SIZE(vf_stats)
1215 /* 4 statistics per queue (rx/tx packets/bytes) */
1216 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 4)
1218 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1220 struct net_device_context *ndc = netdev_priv(dev);
1221 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1226 switch (string_set) {
1228 return NETVSC_GLOBAL_STATS_LEN
1229 + NETVSC_VF_STATS_LEN
1230 + NETVSC_QUEUE_STATS_LEN(nvdev);
1236 static void netvsc_get_ethtool_stats(struct net_device *dev,
1237 struct ethtool_stats *stats, u64 *data)
1239 struct net_device_context *ndc = netdev_priv(dev);
1240 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1241 const void *nds = &ndc->eth_stats;
1242 const struct netvsc_stats *qstats;
1243 struct netvsc_vf_pcpu_stats sum;
1251 for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1252 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1254 netvsc_get_vf_stats(dev, &sum);
1255 for (j = 0; j < NETVSC_VF_STATS_LEN; j++)
1256 data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset);
1258 for (j = 0; j < nvdev->num_chn; j++) {
1259 qstats = &nvdev->chan_table[j].tx_stats;
1262 start = u64_stats_fetch_begin_irq(&qstats->syncp);
1263 packets = qstats->packets;
1264 bytes = qstats->bytes;
1265 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1266 data[i++] = packets;
1269 qstats = &nvdev->chan_table[j].rx_stats;
1271 start = u64_stats_fetch_begin_irq(&qstats->syncp);
1272 packets = qstats->packets;
1273 bytes = qstats->bytes;
1274 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1275 data[i++] = packets;
1280 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1282 struct net_device_context *ndc = netdev_priv(dev);
1283 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1290 switch (stringset) {
1292 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++) {
1293 memcpy(p, netvsc_stats[i].name, ETH_GSTRING_LEN);
1294 p += ETH_GSTRING_LEN;
1297 for (i = 0; i < ARRAY_SIZE(vf_stats); i++) {
1298 memcpy(p, vf_stats[i].name, ETH_GSTRING_LEN);
1299 p += ETH_GSTRING_LEN;
1302 for (i = 0; i < nvdev->num_chn; i++) {
1303 sprintf(p, "tx_queue_%u_packets", i);
1304 p += ETH_GSTRING_LEN;
1305 sprintf(p, "tx_queue_%u_bytes", i);
1306 p += ETH_GSTRING_LEN;
1307 sprintf(p, "rx_queue_%u_packets", i);
1308 p += ETH_GSTRING_LEN;
1309 sprintf(p, "rx_queue_%u_bytes", i);
1310 p += ETH_GSTRING_LEN;
1318 netvsc_get_rss_hash_opts(struct net_device_context *ndc,
1319 struct ethtool_rxnfc *info)
1321 const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3;
1323 info->data = RXH_IP_SRC | RXH_IP_DST;
1325 switch (info->flow_type) {
1327 if (ndc->l4_hash & HV_TCP4_L4HASH)
1328 info->data |= l4_flag;
1333 if (ndc->l4_hash & HV_TCP6_L4HASH)
1334 info->data |= l4_flag;
1339 if (ndc->l4_hash & HV_UDP4_L4HASH)
1340 info->data |= l4_flag;
1345 if (ndc->l4_hash & HV_UDP6_L4HASH)
1346 info->data |= l4_flag;
1362 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1365 struct net_device_context *ndc = netdev_priv(dev);
1366 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1371 switch (info->cmd) {
1372 case ETHTOOL_GRXRINGS:
1373 info->data = nvdev->num_chn;
1377 return netvsc_get_rss_hash_opts(ndc, info);
1382 static int netvsc_set_rss_hash_opts(struct net_device_context *ndc,
1383 struct ethtool_rxnfc *info)
1385 if (info->data == (RXH_IP_SRC | RXH_IP_DST |
1386 RXH_L4_B_0_1 | RXH_L4_B_2_3)) {
1387 switch (info->flow_type) {
1389 ndc->l4_hash |= HV_TCP4_L4HASH;
1393 ndc->l4_hash |= HV_TCP6_L4HASH;
1397 ndc->l4_hash |= HV_UDP4_L4HASH;
1401 ndc->l4_hash |= HV_UDP6_L4HASH;
1411 if (info->data == (RXH_IP_SRC | RXH_IP_DST)) {
1412 switch (info->flow_type) {
1414 ndc->l4_hash &= ~HV_TCP4_L4HASH;
1418 ndc->l4_hash &= ~HV_TCP6_L4HASH;
1422 ndc->l4_hash &= ~HV_UDP4_L4HASH;
1426 ndc->l4_hash &= ~HV_UDP6_L4HASH;
1440 netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info)
1442 struct net_device_context *ndc = netdev_priv(ndev);
1444 if (info->cmd == ETHTOOL_SRXFH)
1445 return netvsc_set_rss_hash_opts(ndc, info);
1450 #ifdef CONFIG_NET_POLL_CONTROLLER
1451 static void netvsc_poll_controller(struct net_device *dev)
1453 struct net_device_context *ndc = netdev_priv(dev);
1454 struct netvsc_device *ndev;
1458 ndev = rcu_dereference(ndc->nvdev);
1460 for (i = 0; i < ndev->num_chn; i++) {
1461 struct netvsc_channel *nvchan = &ndev->chan_table[i];
1463 napi_schedule(&nvchan->napi);
1470 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1472 return NETVSC_HASH_KEYLEN;
1475 static u32 netvsc_rss_indir_size(struct net_device *dev)
1480 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1483 struct net_device_context *ndc = netdev_priv(dev);
1484 struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1485 struct rndis_device *rndis_dev;
1492 *hfunc = ETH_RSS_HASH_TOP; /* Toeplitz */
1494 rndis_dev = ndev->extension;
1496 for (i = 0; i < ITAB_NUM; i++)
1497 indir[i] = rndis_dev->rx_table[i];
1501 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1506 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1507 const u8 *key, const u8 hfunc)
1509 struct net_device_context *ndc = netdev_priv(dev);
1510 struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1511 struct rndis_device *rndis_dev;
1517 if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1520 rndis_dev = ndev->extension;
1522 for (i = 0; i < ITAB_NUM; i++)
1523 if (indir[i] >= ndev->num_chn)
1526 for (i = 0; i < ITAB_NUM; i++)
1527 rndis_dev->rx_table[i] = indir[i];
1534 key = rndis_dev->rss_key;
1537 return rndis_filter_set_rss_param(rndis_dev, key);
1540 /* Hyper-V RNDIS protocol does not have ring in the HW sense.
1541 * It does have pre-allocated receive area which is divided into sections.
1543 static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
1544 struct ethtool_ringparam *ring)
1548 ring->rx_pending = nvdev->recv_section_cnt;
1549 ring->tx_pending = nvdev->send_section_cnt;
1551 if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
1552 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
1554 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1556 ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
1557 ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
1558 / nvdev->send_section_size;
1561 static void netvsc_get_ringparam(struct net_device *ndev,
1562 struct ethtool_ringparam *ring)
1564 struct net_device_context *ndevctx = netdev_priv(ndev);
1565 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1570 __netvsc_get_ringparam(nvdev, ring);
1573 static int netvsc_set_ringparam(struct net_device *ndev,
1574 struct ethtool_ringparam *ring)
1576 struct net_device_context *ndevctx = netdev_priv(ndev);
1577 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1578 struct netvsc_device_info device_info;
1579 struct ethtool_ringparam orig;
1583 if (!nvdev || nvdev->destroy)
1586 memset(&orig, 0, sizeof(orig));
1587 __netvsc_get_ringparam(nvdev, &orig);
1589 new_tx = clamp_t(u32, ring->tx_pending,
1590 NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
1591 new_rx = clamp_t(u32, ring->rx_pending,
1592 NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
1594 if (new_tx == orig.tx_pending &&
1595 new_rx == orig.rx_pending)
1596 return 0; /* no change */
1598 memset(&device_info, 0, sizeof(device_info));
1599 device_info.num_chn = nvdev->num_chn;
1600 device_info.send_sections = new_tx;
1601 device_info.send_section_size = nvdev->send_section_size;
1602 device_info.recv_sections = new_rx;
1603 device_info.recv_section_size = nvdev->recv_section_size;
1605 ret = netvsc_detach(ndev, nvdev);
1609 ret = netvsc_attach(ndev, &device_info);
1611 device_info.send_sections = orig.tx_pending;
1612 device_info.recv_sections = orig.rx_pending;
1614 if (netvsc_attach(ndev, &device_info))
1615 netdev_err(ndev, "restoring ringparam failed");
1621 static const struct ethtool_ops ethtool_ops = {
1622 .get_drvinfo = netvsc_get_drvinfo,
1623 .get_link = ethtool_op_get_link,
1624 .get_ethtool_stats = netvsc_get_ethtool_stats,
1625 .get_sset_count = netvsc_get_sset_count,
1626 .get_strings = netvsc_get_strings,
1627 .get_channels = netvsc_get_channels,
1628 .set_channels = netvsc_set_channels,
1629 .get_ts_info = ethtool_op_get_ts_info,
1630 .get_rxnfc = netvsc_get_rxnfc,
1631 .set_rxnfc = netvsc_set_rxnfc,
1632 .get_rxfh_key_size = netvsc_get_rxfh_key_size,
1633 .get_rxfh_indir_size = netvsc_rss_indir_size,
1634 .get_rxfh = netvsc_get_rxfh,
1635 .set_rxfh = netvsc_set_rxfh,
1636 .get_link_ksettings = netvsc_get_link_ksettings,
1637 .set_link_ksettings = netvsc_set_link_ksettings,
1638 .get_ringparam = netvsc_get_ringparam,
1639 .set_ringparam = netvsc_set_ringparam,
1642 static const struct net_device_ops device_ops = {
1643 .ndo_open = netvsc_open,
1644 .ndo_stop = netvsc_close,
1645 .ndo_start_xmit = netvsc_start_xmit,
1646 .ndo_change_rx_flags = netvsc_change_rx_flags,
1647 .ndo_set_rx_mode = netvsc_set_rx_mode,
1648 .ndo_change_mtu = netvsc_change_mtu,
1649 .ndo_validate_addr = eth_validate_addr,
1650 .ndo_set_mac_address = netvsc_set_mac_addr,
1651 .ndo_select_queue = netvsc_select_queue,
1652 .ndo_get_stats64 = netvsc_get_stats64,
1653 #ifdef CONFIG_NET_POLL_CONTROLLER
1654 .ndo_poll_controller = netvsc_poll_controller,
1659 * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1660 * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1661 * present send GARP packet to network peers with netif_notify_peers().
1663 static void netvsc_link_change(struct work_struct *w)
1665 struct net_device_context *ndev_ctx =
1666 container_of(w, struct net_device_context, dwork.work);
1667 struct hv_device *device_obj = ndev_ctx->device_ctx;
1668 struct net_device *net = hv_get_drvdata(device_obj);
1669 struct netvsc_device *net_device;
1670 struct rndis_device *rdev;
1671 struct netvsc_reconfig *event = NULL;
1672 bool notify = false, reschedule = false;
1673 unsigned long flags, next_reconfig, delay;
1675 /* if changes are happening, comeback later */
1676 if (!rtnl_trylock()) {
1677 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1681 net_device = rtnl_dereference(ndev_ctx->nvdev);
1685 rdev = net_device->extension;
1687 next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1688 if (time_is_after_jiffies(next_reconfig)) {
1689 /* link_watch only sends one notification with current state
1690 * per second, avoid doing reconfig more frequently. Handle
1693 delay = next_reconfig - jiffies;
1694 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1695 schedule_delayed_work(&ndev_ctx->dwork, delay);
1698 ndev_ctx->last_reconfig = jiffies;
1700 spin_lock_irqsave(&ndev_ctx->lock, flags);
1701 if (!list_empty(&ndev_ctx->reconfig_events)) {
1702 event = list_first_entry(&ndev_ctx->reconfig_events,
1703 struct netvsc_reconfig, list);
1704 list_del(&event->list);
1705 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1707 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1712 switch (event->event) {
1713 /* Only the following events are possible due to the check in
1714 * netvsc_linkstatus_callback()
1716 case RNDIS_STATUS_MEDIA_CONNECT:
1717 if (rdev->link_state) {
1718 rdev->link_state = false;
1719 netif_carrier_on(net);
1720 netif_tx_wake_all_queues(net);
1726 case RNDIS_STATUS_MEDIA_DISCONNECT:
1727 if (!rdev->link_state) {
1728 rdev->link_state = true;
1729 netif_carrier_off(net);
1730 netif_tx_stop_all_queues(net);
1734 case RNDIS_STATUS_NETWORK_CHANGE:
1735 /* Only makes sense if carrier is present */
1736 if (!rdev->link_state) {
1737 rdev->link_state = true;
1738 netif_carrier_off(net);
1739 netif_tx_stop_all_queues(net);
1740 event->event = RNDIS_STATUS_MEDIA_CONNECT;
1741 spin_lock_irqsave(&ndev_ctx->lock, flags);
1742 list_add(&event->list, &ndev_ctx->reconfig_events);
1743 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1752 netdev_notify_peers(net);
1754 /* link_watch only sends one notification with current state per
1755 * second, handle next reconfig event in 2 seconds.
1758 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1766 static struct net_device *get_netvsc_bymac(const u8 *mac)
1768 struct net_device *dev;
1772 for_each_netdev(&init_net, dev) {
1773 if (dev->netdev_ops != &device_ops)
1774 continue; /* not a netvsc device */
1776 if (ether_addr_equal(mac, dev->perm_addr))
1783 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
1785 struct net_device *dev;
1789 for_each_netdev(&init_net, dev) {
1790 struct net_device_context *net_device_ctx;
1792 if (dev->netdev_ops != &device_ops)
1793 continue; /* not a netvsc device */
1795 net_device_ctx = netdev_priv(dev);
1796 if (!rtnl_dereference(net_device_ctx->nvdev))
1797 continue; /* device is removed */
1799 if (rtnl_dereference(net_device_ctx->vf_netdev) == vf_netdev)
1800 return dev; /* a match */
1806 /* Called when VF is injecting data into network stack.
1807 * Change the associated network device from VF to netvsc.
1808 * note: already called with rcu_read_lock
1810 static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
1812 struct sk_buff *skb = *pskb;
1813 struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data);
1814 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1815 struct netvsc_vf_pcpu_stats *pcpu_stats
1816 = this_cpu_ptr(ndev_ctx->vf_stats);
1820 u64_stats_update_begin(&pcpu_stats->syncp);
1821 pcpu_stats->rx_packets++;
1822 pcpu_stats->rx_bytes += skb->len;
1823 u64_stats_update_end(&pcpu_stats->syncp);
1825 return RX_HANDLER_ANOTHER;
1828 static int netvsc_vf_join(struct net_device *vf_netdev,
1829 struct net_device *ndev)
1831 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1834 ret = netdev_rx_handler_register(vf_netdev,
1835 netvsc_vf_handle_frame, ndev);
1837 netdev_err(vf_netdev,
1838 "can not register netvsc VF receive handler (err = %d)\n",
1840 goto rx_handler_failed;
1843 ret = netdev_upper_dev_link(vf_netdev, ndev, NULL);
1845 netdev_err(vf_netdev,
1846 "can not set master device %s (err = %d)\n",
1848 goto upper_link_failed;
1851 /* set slave flag before open to prevent IPv6 addrconf */
1852 vf_netdev->flags |= IFF_SLAVE;
1854 schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
1856 call_netdevice_notifiers(NETDEV_JOIN, vf_netdev);
1858 netdev_info(vf_netdev, "joined to %s\n", ndev->name);
1862 netdev_rx_handler_unregister(vf_netdev);
1867 static void __netvsc_vf_setup(struct net_device *ndev,
1868 struct net_device *vf_netdev)
1872 /* Align MTU of VF with master */
1873 ret = dev_set_mtu(vf_netdev, ndev->mtu);
1875 netdev_warn(vf_netdev,
1876 "unable to change mtu to %u\n", ndev->mtu);
1878 /* set multicast etc flags on VF */
1879 dev_change_flags(vf_netdev, ndev->flags | IFF_SLAVE);
1881 /* sync address list from ndev to VF */
1882 netif_addr_lock_bh(ndev);
1883 dev_uc_sync(vf_netdev, ndev);
1884 dev_mc_sync(vf_netdev, ndev);
1885 netif_addr_unlock_bh(ndev);
1887 if (netif_running(ndev)) {
1888 ret = dev_open(vf_netdev);
1890 netdev_warn(vf_netdev,
1891 "unable to open: %d\n", ret);
1895 /* Setup VF as slave of the synthetic device.
1896 * Runs in workqueue to avoid recursion in netlink callbacks.
1898 static void netvsc_vf_setup(struct work_struct *w)
1900 struct net_device_context *ndev_ctx
1901 = container_of(w, struct net_device_context, vf_takeover.work);
1902 struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx);
1903 struct net_device *vf_netdev;
1905 if (!rtnl_trylock()) {
1906 schedule_delayed_work(&ndev_ctx->vf_takeover, 0);
1910 vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
1912 __netvsc_vf_setup(ndev, vf_netdev);
1917 static int netvsc_register_vf(struct net_device *vf_netdev)
1919 struct net_device *ndev;
1920 struct net_device_context *net_device_ctx;
1921 struct netvsc_device *netvsc_dev;
1923 if (vf_netdev->addr_len != ETH_ALEN)
1927 * We will use the MAC address to locate the synthetic interface to
1928 * associate with the VF interface. If we don't find a matching
1929 * synthetic interface, move on.
1931 ndev = get_netvsc_bymac(vf_netdev->perm_addr);
1935 net_device_ctx = netdev_priv(ndev);
1936 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
1937 if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
1940 if (netvsc_vf_join(vf_netdev, ndev) != 0)
1943 netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
1945 dev_hold(vf_netdev);
1946 rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
1950 /* VF up/down change detected, schedule to change data path */
1951 static int netvsc_vf_changed(struct net_device *vf_netdev)
1953 struct net_device_context *net_device_ctx;
1954 struct netvsc_device *netvsc_dev;
1955 struct net_device *ndev;
1956 bool vf_is_up = netif_running(vf_netdev);
1958 ndev = get_netvsc_byref(vf_netdev);
1962 net_device_ctx = netdev_priv(ndev);
1963 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
1967 netvsc_switch_datapath(ndev, vf_is_up);
1968 netdev_info(ndev, "Data path switched %s VF: %s\n",
1969 vf_is_up ? "to" : "from", vf_netdev->name);
1974 static int netvsc_unregister_vf(struct net_device *vf_netdev)
1976 struct net_device *ndev;
1977 struct net_device_context *net_device_ctx;
1979 ndev = get_netvsc_byref(vf_netdev);
1983 net_device_ctx = netdev_priv(ndev);
1984 cancel_delayed_work_sync(&net_device_ctx->vf_takeover);
1986 netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
1988 netdev_rx_handler_unregister(vf_netdev);
1989 netdev_upper_dev_unlink(vf_netdev, ndev);
1990 RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
1996 static int netvsc_probe(struct hv_device *dev,
1997 const struct hv_vmbus_device_id *dev_id)
1999 struct net_device *net = NULL;
2000 struct net_device_context *net_device_ctx;
2001 struct netvsc_device_info device_info;
2002 struct netvsc_device *nvdev;
2005 net = alloc_etherdev_mq(sizeof(struct net_device_context),
2010 netif_carrier_off(net);
2012 netvsc_init_settings(net);
2014 net_device_ctx = netdev_priv(net);
2015 net_device_ctx->device_ctx = dev;
2016 net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
2017 if (netif_msg_probe(net_device_ctx))
2018 netdev_dbg(net, "netvsc msg_enable: %d\n",
2019 net_device_ctx->msg_enable);
2021 hv_set_drvdata(dev, net);
2023 INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
2025 spin_lock_init(&net_device_ctx->lock);
2026 INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
2027 INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup);
2029 net_device_ctx->vf_stats
2030 = netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats);
2031 if (!net_device_ctx->vf_stats)
2034 net->netdev_ops = &device_ops;
2035 net->ethtool_ops = ðtool_ops;
2036 SET_NETDEV_DEV(net, &dev->device);
2038 /* We always need headroom for rndis header */
2039 net->needed_headroom = RNDIS_AND_PPI_SIZE;
2041 /* Initialize the number of queues to be 1, we may change it if more
2042 * channels are offered later.
2044 netif_set_real_num_tx_queues(net, 1);
2045 netif_set_real_num_rx_queues(net, 1);
2047 /* Notify the netvsc driver of the new device */
2048 memset(&device_info, 0, sizeof(device_info));
2049 device_info.num_chn = VRSS_CHANNEL_DEFAULT;
2050 device_info.send_sections = NETVSC_DEFAULT_TX;
2051 device_info.send_section_size = NETVSC_SEND_SECTION_SIZE;
2052 device_info.recv_sections = NETVSC_DEFAULT_RX;
2053 device_info.recv_section_size = NETVSC_RECV_SECTION_SIZE;
2055 nvdev = rndis_filter_device_add(dev, &device_info);
2056 if (IS_ERR(nvdev)) {
2057 ret = PTR_ERR(nvdev);
2058 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
2062 memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
2064 /* hw_features computed in rndis_netdev_set_hwcaps() */
2065 net->features = net->hw_features |
2066 NETIF_F_HIGHDMA | NETIF_F_SG |
2067 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
2068 net->vlan_features = net->features;
2070 netdev_lockdep_set_classes(net);
2072 /* MTU range: 68 - 1500 or 65521 */
2073 net->min_mtu = NETVSC_MTU_MIN;
2074 if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
2075 net->max_mtu = NETVSC_MTU - ETH_HLEN;
2077 net->max_mtu = ETH_DATA_LEN;
2079 ret = register_netdev(net);
2081 pr_err("Unable to register netdev.\n");
2082 goto register_failed;
2088 rndis_filter_device_remove(dev, nvdev);
2090 free_percpu(net_device_ctx->vf_stats);
2092 hv_set_drvdata(dev, NULL);
2098 static int netvsc_remove(struct hv_device *dev)
2100 struct net_device_context *ndev_ctx;
2101 struct net_device *vf_netdev, *net;
2102 struct netvsc_device *nvdev;
2104 net = hv_get_drvdata(dev);
2106 dev_err(&dev->device, "No net device to remove\n");
2110 ndev_ctx = netdev_priv(net);
2112 cancel_delayed_work_sync(&ndev_ctx->dwork);
2115 nvdev = rcu_dereference(ndev_ctx->nvdev);
2118 cancel_work_sync(&nvdev->subchan_work);
2121 * Call to the vsc driver to let it know that the device is being
2122 * removed. Also blocks mtu and channel changes.
2125 vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2127 netvsc_unregister_vf(vf_netdev);
2130 rndis_filter_device_remove(dev, nvdev);
2132 unregister_netdevice(net);
2137 hv_set_drvdata(dev, NULL);
2139 free_percpu(ndev_ctx->vf_stats);
2144 static const struct hv_vmbus_device_id id_table[] = {
2150 MODULE_DEVICE_TABLE(vmbus, id_table);
2152 /* The one and only one */
2153 static struct hv_driver netvsc_drv = {
2154 .name = KBUILD_MODNAME,
2155 .id_table = id_table,
2156 .probe = netvsc_probe,
2157 .remove = netvsc_remove,
2161 * On Hyper-V, every VF interface is matched with a corresponding
2162 * synthetic interface. The synthetic interface is presented first
2163 * to the guest. When the corresponding VF instance is registered,
2164 * we will take care of switching the data path.
2166 static int netvsc_netdev_event(struct notifier_block *this,
2167 unsigned long event, void *ptr)
2169 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2171 /* Skip our own events */
2172 if (event_dev->netdev_ops == &device_ops)
2175 /* Avoid non-Ethernet type devices */
2176 if (event_dev->type != ARPHRD_ETHER)
2179 /* Avoid Vlan dev with same MAC registering as VF */
2180 if (is_vlan_dev(event_dev))
2183 /* Avoid Bonding master dev with same MAC registering as VF */
2184 if ((event_dev->priv_flags & IFF_BONDING) &&
2185 (event_dev->flags & IFF_MASTER))
2189 case NETDEV_REGISTER:
2190 return netvsc_register_vf(event_dev);
2191 case NETDEV_UNREGISTER:
2192 return netvsc_unregister_vf(event_dev);
2195 return netvsc_vf_changed(event_dev);
2201 static struct notifier_block netvsc_netdev_notifier = {
2202 .notifier_call = netvsc_netdev_event,
2205 static void __exit netvsc_drv_exit(void)
2207 unregister_netdevice_notifier(&netvsc_netdev_notifier);
2208 vmbus_driver_unregister(&netvsc_drv);
2211 static int __init netvsc_drv_init(void)
2215 if (ring_size < RING_SIZE_MIN) {
2216 ring_size = RING_SIZE_MIN;
2217 pr_info("Increased ring_size to %u (min allowed)\n",
2220 netvsc_ring_bytes = ring_size * PAGE_SIZE;
2221 netvsc_ring_reciprocal = reciprocal_value(netvsc_ring_bytes);
2223 ret = vmbus_driver_register(&netvsc_drv);
2227 register_netdevice_notifier(&netvsc_netdev_notifier);
2231 MODULE_LICENSE("GPL");
2232 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
2234 module_init(netvsc_drv_init);
2235 module_exit(netvsc_drv_exit);