1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (c) 2009, Microsoft Corporation.
6 * Haiyang Zhang <haiyangz@microsoft.com>
7 * Hank Janssen <hjanssen@microsoft.com>
9 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11 #include <linux/init.h>
12 #include <linux/atomic.h>
13 #include <linux/module.h>
14 #include <linux/highmem.h>
15 #include <linux/device.h>
17 #include <linux/delay.h>
18 #include <linux/netdevice.h>
19 #include <linux/inetdevice.h>
20 #include <linux/etherdevice.h>
21 #include <linux/pci.h>
22 #include <linux/skbuff.h>
23 #include <linux/if_vlan.h>
25 #include <linux/slab.h>
26 #include <linux/rtnetlink.h>
27 #include <linux/netpoll.h>
28 #include <linux/bpf.h>
31 #include <net/route.h>
33 #include <net/pkt_sched.h>
34 #include <net/checksum.h>
35 #include <net/ip6_checksum.h>
37 #include "hyperv_net.h"
39 #define RING_SIZE_MIN 64
40 #define RETRY_US_LO 5000
41 #define RETRY_US_HI 10000
42 #define RETRY_MAX 2000 /* >10 sec */
44 #define LINKCHANGE_INT (2 * HZ)
45 #define VF_TAKEOVER_INT (HZ / 10)
47 static unsigned int ring_size __ro_after_init = 128;
48 module_param(ring_size, uint, 0444);
49 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
50 unsigned int netvsc_ring_bytes __ro_after_init;
52 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
53 NETIF_MSG_LINK | NETIF_MSG_IFUP |
54 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
57 static int debug = -1;
58 module_param(debug, int, 0444);
59 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
61 static LIST_HEAD(netvsc_dev_list);
63 static void netvsc_change_rx_flags(struct net_device *net, int change)
65 struct net_device_context *ndev_ctx = netdev_priv(net);
66 struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
72 if (change & IFF_PROMISC) {
73 inc = (net->flags & IFF_PROMISC) ? 1 : -1;
74 dev_set_promiscuity(vf_netdev, inc);
77 if (change & IFF_ALLMULTI) {
78 inc = (net->flags & IFF_ALLMULTI) ? 1 : -1;
79 dev_set_allmulti(vf_netdev, inc);
83 static void netvsc_set_rx_mode(struct net_device *net)
85 struct net_device_context *ndev_ctx = netdev_priv(net);
86 struct net_device *vf_netdev;
87 struct netvsc_device *nvdev;
90 vf_netdev = rcu_dereference(ndev_ctx->vf_netdev);
92 dev_uc_sync(vf_netdev, net);
93 dev_mc_sync(vf_netdev, net);
96 nvdev = rcu_dereference(ndev_ctx->nvdev);
98 rndis_filter_update(nvdev);
102 static void netvsc_tx_enable(struct netvsc_device *nvscdev,
103 struct net_device *ndev)
105 nvscdev->tx_disable = false;
106 virt_wmb(); /* ensure queue wake up mechanism is on */
108 netif_tx_wake_all_queues(ndev);
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);
131 netvsc_tx_enable(nvdev, net);
135 /* Setting synthetic device up transparently sets
136 * slave as up. If open fails, then slave will be
137 * still be offline (and not used).
139 ret = dev_open(vf_netdev, NULL);
142 "unable to open slave: %s: %d\n",
143 vf_netdev->name, ret);
148 static int netvsc_wait_until_empty(struct netvsc_device *nvdev)
150 unsigned int retry = 0;
153 /* Ensure pending bytes in ring are read */
157 for (i = 0; i < nvdev->num_chn; i++) {
158 struct vmbus_channel *chn
159 = nvdev->chan_table[i].channel;
164 /* make sure receive not running now */
165 napi_synchronize(&nvdev->chan_table[i].napi);
167 aread = hv_get_bytes_to_read(&chn->inbound);
171 aread = hv_get_bytes_to_read(&chn->outbound);
179 if (++retry > RETRY_MAX)
182 usleep_range(RETRY_US_LO, RETRY_US_HI);
186 static void netvsc_tx_disable(struct netvsc_device *nvscdev,
187 struct net_device *ndev)
190 nvscdev->tx_disable = true;
191 virt_wmb(); /* ensure txq will not wake up after stop */
194 netif_tx_disable(ndev);
197 static int netvsc_close(struct net_device *net)
199 struct net_device_context *net_device_ctx = netdev_priv(net);
200 struct net_device *vf_netdev
201 = rtnl_dereference(net_device_ctx->vf_netdev);
202 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
205 netvsc_tx_disable(nvdev, net);
207 /* No need to close rndis filter if it is removed already */
211 ret = rndis_filter_close(nvdev);
213 netdev_err(net, "unable to close device (ret %d).\n", ret);
217 ret = netvsc_wait_until_empty(nvdev);
219 netdev_err(net, "Ring buffer not empty after closing rndis\n");
222 dev_close(vf_netdev);
227 static inline void *init_ppi_data(struct rndis_message *msg,
228 u32 ppi_size, u32 pkt_type)
230 struct rndis_packet *rndis_pkt = &msg->msg.pkt;
231 struct rndis_per_packet_info *ppi;
233 rndis_pkt->data_offset += ppi_size;
234 ppi = (void *)rndis_pkt + rndis_pkt->per_pkt_info_offset
235 + rndis_pkt->per_pkt_info_len;
237 ppi->size = ppi_size;
238 ppi->type = pkt_type;
240 ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
242 rndis_pkt->per_pkt_info_len += ppi_size;
247 /* Azure hosts don't support non-TCP port numbers in hashing for fragmented
248 * packets. We can use ethtool to change UDP hash level when necessary.
250 static inline u32 netvsc_get_hash(
252 const struct net_device_context *ndc)
254 struct flow_keys flow;
255 u32 hash, pkt_proto = 0;
256 static u32 hashrnd __read_mostly;
258 net_get_random_once(&hashrnd, sizeof(hashrnd));
260 if (!skb_flow_dissect_flow_keys(skb, &flow, 0))
263 switch (flow.basic.ip_proto) {
265 if (flow.basic.n_proto == htons(ETH_P_IP))
266 pkt_proto = HV_TCP4_L4HASH;
267 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
268 pkt_proto = HV_TCP6_L4HASH;
273 if (flow.basic.n_proto == htons(ETH_P_IP))
274 pkt_proto = HV_UDP4_L4HASH;
275 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
276 pkt_proto = HV_UDP6_L4HASH;
281 if (pkt_proto & ndc->l4_hash) {
282 return skb_get_hash(skb);
284 if (flow.basic.n_proto == htons(ETH_P_IP))
285 hash = jhash2((u32 *)&flow.addrs.v4addrs, 2, hashrnd);
286 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
287 hash = jhash2((u32 *)&flow.addrs.v6addrs, 8, hashrnd);
291 __skb_set_sw_hash(skb, hash, false);
297 static inline int netvsc_get_tx_queue(struct net_device *ndev,
298 struct sk_buff *skb, int old_idx)
300 const struct net_device_context *ndc = netdev_priv(ndev);
301 struct sock *sk = skb->sk;
304 q_idx = ndc->tx_table[netvsc_get_hash(skb, ndc) &
305 (VRSS_SEND_TAB_SIZE - 1)];
307 /* If queue index changed record the new value */
308 if (q_idx != old_idx &&
309 sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
310 sk_tx_queue_set(sk, q_idx);
316 * Select queue for transmit.
318 * If a valid queue has already been assigned, then use that.
319 * Otherwise compute tx queue based on hash and the send table.
321 * This is basically similar to default (netdev_pick_tx) with the added step
322 * of using the host send_table when no other queue has been assigned.
324 * TODO support XPS - but get_xps_queue not exported
326 static u16 netvsc_pick_tx(struct net_device *ndev, struct sk_buff *skb)
328 int q_idx = sk_tx_queue_get(skb->sk);
330 if (q_idx < 0 || skb->ooo_okay || q_idx >= ndev->real_num_tx_queues) {
331 /* If forwarding a packet, we use the recorded queue when
332 * available for better cache locality.
334 if (skb_rx_queue_recorded(skb))
335 q_idx = skb_get_rx_queue(skb);
337 q_idx = netvsc_get_tx_queue(ndev, skb, q_idx);
343 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
344 struct net_device *sb_dev)
346 struct net_device_context *ndc = netdev_priv(ndev);
347 struct net_device *vf_netdev;
351 vf_netdev = rcu_dereference(ndc->vf_netdev);
353 const struct net_device_ops *vf_ops = vf_netdev->netdev_ops;
355 if (vf_ops->ndo_select_queue)
356 txq = vf_ops->ndo_select_queue(vf_netdev, skb, sb_dev);
358 txq = netdev_pick_tx(vf_netdev, skb, NULL);
360 /* Record the queue selected by VF so that it can be
361 * used for common case where VF has more queues than
362 * the synthetic device.
364 qdisc_skb_cb(skb)->slave_dev_queue_mapping = txq;
366 txq = netvsc_pick_tx(ndev, skb);
370 while (txq >= ndev->real_num_tx_queues)
371 txq -= ndev->real_num_tx_queues;
376 static u32 fill_pg_buf(unsigned long hvpfn, u32 offset, u32 len,
377 struct hv_page_buffer *pb)
381 hvpfn += offset >> HV_HYP_PAGE_SHIFT;
382 offset = offset & ~HV_HYP_PAGE_MASK;
387 bytes = HV_HYP_PAGE_SIZE - offset;
391 pb[j].offset = offset;
397 if (offset == HV_HYP_PAGE_SIZE && len) {
407 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
408 struct hv_netvsc_packet *packet,
409 struct hv_page_buffer *pb)
412 char *data = skb->data;
413 int frags = skb_shinfo(skb)->nr_frags;
416 /* The packet is laid out thus:
417 * 1. hdr: RNDIS header and PPI
419 * 3. skb fragment data
421 slots_used += fill_pg_buf(virt_to_hvpfn(hdr),
422 offset_in_hvpage(hdr),
426 packet->rmsg_size = len;
427 packet->rmsg_pgcnt = slots_used;
429 slots_used += fill_pg_buf(virt_to_hvpfn(data),
430 offset_in_hvpage(data),
434 for (i = 0; i < frags; i++) {
435 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
437 slots_used += fill_pg_buf(page_to_hvpfn(skb_frag_page(frag)),
445 static int count_skb_frag_slots(struct sk_buff *skb)
447 int i, frags = skb_shinfo(skb)->nr_frags;
450 for (i = 0; i < frags; i++) {
451 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
452 unsigned long size = skb_frag_size(frag);
453 unsigned long offset = skb_frag_off(frag);
455 /* Skip unused frames from start of page */
456 offset &= ~HV_HYP_PAGE_MASK;
457 pages += HVPFN_UP(offset + size);
462 static int netvsc_get_slots(struct sk_buff *skb)
464 char *data = skb->data;
465 unsigned int offset = offset_in_hvpage(data);
466 unsigned int len = skb_headlen(skb);
470 slots = DIV_ROUND_UP(offset + len, HV_HYP_PAGE_SIZE);
471 frag_slots = count_skb_frag_slots(skb);
472 return slots + frag_slots;
475 static u32 net_checksum_info(struct sk_buff *skb)
477 if (skb->protocol == htons(ETH_P_IP)) {
478 struct iphdr *ip = ip_hdr(skb);
480 if (ip->protocol == IPPROTO_TCP)
481 return TRANSPORT_INFO_IPV4_TCP;
482 else if (ip->protocol == IPPROTO_UDP)
483 return TRANSPORT_INFO_IPV4_UDP;
485 struct ipv6hdr *ip6 = ipv6_hdr(skb);
487 if (ip6->nexthdr == IPPROTO_TCP)
488 return TRANSPORT_INFO_IPV6_TCP;
489 else if (ip6->nexthdr == IPPROTO_UDP)
490 return TRANSPORT_INFO_IPV6_UDP;
493 return TRANSPORT_INFO_NOT_IP;
496 /* Send skb on the slave VF device. */
497 static int netvsc_vf_xmit(struct net_device *net, struct net_device *vf_netdev,
500 struct net_device_context *ndev_ctx = netdev_priv(net);
501 unsigned int len = skb->len;
504 skb->dev = vf_netdev;
505 skb_record_rx_queue(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
507 rc = dev_queue_xmit(skb);
508 if (likely(rc == NET_XMIT_SUCCESS || rc == NET_XMIT_CN)) {
509 struct netvsc_vf_pcpu_stats *pcpu_stats
510 = this_cpu_ptr(ndev_ctx->vf_stats);
512 u64_stats_update_begin(&pcpu_stats->syncp);
513 pcpu_stats->tx_packets++;
514 pcpu_stats->tx_bytes += len;
515 u64_stats_update_end(&pcpu_stats->syncp);
517 this_cpu_inc(ndev_ctx->vf_stats->tx_dropped);
523 static int netvsc_xmit(struct sk_buff *skb, struct net_device *net, bool xdp_tx)
525 struct net_device_context *net_device_ctx = netdev_priv(net);
526 struct hv_netvsc_packet *packet = NULL;
528 unsigned int num_data_pgs;
529 struct rndis_message *rndis_msg;
530 struct net_device *vf_netdev;
533 struct hv_page_buffer pb[MAX_PAGE_BUFFER_COUNT];
535 /* If VF is present and up then redirect packets to it.
536 * Skip the VF if it is marked down or has no carrier.
537 * If netpoll is in uses, then VF can not be used either.
539 vf_netdev = rcu_dereference_bh(net_device_ctx->vf_netdev);
540 if (vf_netdev && netif_running(vf_netdev) &&
541 netif_carrier_ok(vf_netdev) && !netpoll_tx_running(net))
542 return netvsc_vf_xmit(net, vf_netdev, skb);
544 /* We will atmost need two pages to describe the rndis
545 * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
546 * of pages in a single packet. If skb is scattered around
547 * more pages we try linearizing it.
550 num_data_pgs = netvsc_get_slots(skb) + 2;
552 if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
553 ++net_device_ctx->eth_stats.tx_scattered;
555 if (skb_linearize(skb))
558 num_data_pgs = netvsc_get_slots(skb) + 2;
559 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
560 ++net_device_ctx->eth_stats.tx_too_big;
566 * Place the rndis header in the skb head room and
567 * the skb->cb will be used for hv_netvsc_packet
570 ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
574 /* Use the skb control buffer for building up the packet */
575 BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
576 sizeof_field(struct sk_buff, cb));
577 packet = (struct hv_netvsc_packet *)skb->cb;
579 packet->q_idx = skb_get_queue_mapping(skb);
581 packet->total_data_buflen = skb->len;
582 packet->total_bytes = skb->len;
583 packet->total_packets = 1;
585 rndis_msg = (struct rndis_message *)skb->head;
587 /* Add the rndis header */
588 rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
589 rndis_msg->msg_len = packet->total_data_buflen;
591 rndis_msg->msg.pkt = (struct rndis_packet) {
592 .data_offset = sizeof(struct rndis_packet),
593 .data_len = packet->total_data_buflen,
594 .per_pkt_info_offset = sizeof(struct rndis_packet),
597 rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
599 hash = skb_get_hash_raw(skb);
600 if (hash != 0 && net->real_num_tx_queues > 1) {
603 rndis_msg_size += NDIS_HASH_PPI_SIZE;
604 hash_info = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
609 /* When using AF_PACKET we need to drop VLAN header from
610 * the frame and update the SKB to allow the HOST OS
611 * to transmit the 802.1Q packet
613 if (skb->protocol == htons(ETH_P_8021Q)) {
616 skb_reset_mac_header(skb);
617 if (eth_type_vlan(eth_hdr(skb)->h_proto)) {
618 if (unlikely(__skb_vlan_pop(skb, &vlan_tci) != 0)) {
619 ++net_device_ctx->eth_stats.vlan_error;
623 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
624 /* Update the NDIS header pkt lengths */
625 packet->total_data_buflen -= VLAN_HLEN;
626 packet->total_bytes -= VLAN_HLEN;
627 rndis_msg->msg_len = packet->total_data_buflen;
628 rndis_msg->msg.pkt.data_len = packet->total_data_buflen;
632 if (skb_vlan_tag_present(skb)) {
633 struct ndis_pkt_8021q_info *vlan;
635 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
636 vlan = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
640 vlan->vlanid = skb_vlan_tag_get_id(skb);
641 vlan->cfi = skb_vlan_tag_get_cfi(skb);
642 vlan->pri = skb_vlan_tag_get_prio(skb);
645 if (skb_is_gso(skb)) {
646 struct ndis_tcp_lso_info *lso_info;
648 rndis_msg_size += NDIS_LSO_PPI_SIZE;
649 lso_info = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
650 TCP_LARGESEND_PKTINFO);
653 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
654 if (skb->protocol == htons(ETH_P_IP)) {
655 lso_info->lso_v2_transmit.ip_version =
656 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
657 ip_hdr(skb)->tot_len = 0;
658 ip_hdr(skb)->check = 0;
659 tcp_hdr(skb)->check =
660 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
661 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
663 lso_info->lso_v2_transmit.ip_version =
664 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
665 tcp_v6_gso_csum_prep(skb);
667 lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
668 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
669 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
670 if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
671 struct ndis_tcp_ip_checksum_info *csum_info;
673 rndis_msg_size += NDIS_CSUM_PPI_SIZE;
674 csum_info = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
675 TCPIP_CHKSUM_PKTINFO);
677 csum_info->value = 0;
678 csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
680 if (skb->protocol == htons(ETH_P_IP)) {
681 csum_info->transmit.is_ipv4 = 1;
683 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
684 csum_info->transmit.tcp_checksum = 1;
686 csum_info->transmit.udp_checksum = 1;
688 csum_info->transmit.is_ipv6 = 1;
690 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
691 csum_info->transmit.tcp_checksum = 1;
693 csum_info->transmit.udp_checksum = 1;
696 /* Can't do offload of this type of checksum */
697 if (skb_checksum_help(skb))
702 /* Start filling in the page buffers with the rndis hdr */
703 rndis_msg->msg_len += rndis_msg_size;
704 packet->total_data_buflen = rndis_msg->msg_len;
705 packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
708 /* timestamp packet in software */
709 skb_tx_timestamp(skb);
711 ret = netvsc_send(net, packet, rndis_msg, pb, skb, xdp_tx);
712 if (likely(ret == 0))
715 if (ret == -EAGAIN) {
716 ++net_device_ctx->eth_stats.tx_busy;
717 return NETDEV_TX_BUSY;
721 ++net_device_ctx->eth_stats.tx_no_space;
724 dev_kfree_skb_any(skb);
725 net->stats.tx_dropped++;
730 ++net_device_ctx->eth_stats.tx_no_memory;
734 static netdev_tx_t netvsc_start_xmit(struct sk_buff *skb,
735 struct net_device *ndev)
737 return netvsc_xmit(skb, ndev, false);
741 * netvsc_linkstatus_callback - Link up/down notification
743 void netvsc_linkstatus_callback(struct net_device *net,
744 struct rndis_message *resp)
746 struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
747 struct net_device_context *ndev_ctx = netdev_priv(net);
748 struct netvsc_reconfig *event;
751 /* Ensure the packet is big enough to access its fields */
752 if (resp->msg_len - RNDIS_HEADER_SIZE < sizeof(struct rndis_indicate_status)) {
753 netdev_err(net, "invalid rndis_indicate_status packet, len: %u\n",
758 /* Update the physical link speed when changing to another vSwitch */
759 if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
762 speed = *(u32 *)((void *)indicate
763 + indicate->status_buf_offset) / 10000;
764 ndev_ctx->speed = speed;
768 /* Handle these link change statuses below */
769 if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
770 indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
771 indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
774 if (net->reg_state != NETREG_REGISTERED)
777 event = kzalloc(sizeof(*event), GFP_ATOMIC);
780 event->event = indicate->status;
782 spin_lock_irqsave(&ndev_ctx->lock, flags);
783 list_add_tail(&event->list, &ndev_ctx->reconfig_events);
784 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
786 schedule_delayed_work(&ndev_ctx->dwork, 0);
789 static void netvsc_xdp_xmit(struct sk_buff *skb, struct net_device *ndev)
793 skb->queue_mapping = skb_get_rx_queue(skb);
794 __skb_push(skb, ETH_HLEN);
796 rc = netvsc_xmit(skb, ndev, true);
798 if (dev_xmit_complete(rc))
801 dev_kfree_skb_any(skb);
802 ndev->stats.tx_dropped++;
805 static void netvsc_comp_ipcsum(struct sk_buff *skb)
807 struct iphdr *iph = (struct iphdr *)skb->data;
810 iph->check = ip_fast_csum(iph, iph->ihl);
813 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
814 struct netvsc_channel *nvchan,
815 struct xdp_buff *xdp)
817 struct napi_struct *napi = &nvchan->napi;
818 const struct ndis_pkt_8021q_info *vlan = nvchan->rsc.vlan;
819 const struct ndis_tcp_ip_checksum_info *csum_info =
820 nvchan->rsc.csum_info;
821 const u32 *hash_info = nvchan->rsc.hash_info;
823 void *xbuf = xdp->data_hard_start;
827 unsigned int hdroom = xdp->data - xdp->data_hard_start;
828 unsigned int xlen = xdp->data_end - xdp->data;
829 unsigned int frag_size = xdp->frame_sz;
831 skb = build_skb(xbuf, frag_size);
834 __free_page(virt_to_page(xbuf));
838 skb_reserve(skb, hdroom);
840 skb->dev = napi->dev;
842 skb = napi_alloc_skb(napi, nvchan->rsc.pktlen);
847 /* Copy to skb. This copy is needed here since the memory
848 * pointed by hv_netvsc_packet cannot be deallocated.
850 for (i = 0; i < nvchan->rsc.cnt; i++)
851 skb_put_data(skb, nvchan->rsc.data[i],
855 skb->protocol = eth_type_trans(skb, net);
857 /* skb is already created with CHECKSUM_NONE */
858 skb_checksum_none_assert(skb);
860 /* Incoming packets may have IP header checksum verified by the host.
861 * They may not have IP header checksum computed after coalescing.
862 * We compute it here if the flags are set, because on Linux, the IP
863 * checksum is always checked.
865 if (csum_info && csum_info->receive.ip_checksum_value_invalid &&
866 csum_info->receive.ip_checksum_succeeded &&
867 skb->protocol == htons(ETH_P_IP))
868 netvsc_comp_ipcsum(skb);
870 /* Do L4 checksum offload if enabled and present. */
871 if (csum_info && (net->features & NETIF_F_RXCSUM)) {
872 if (csum_info->receive.tcp_checksum_succeeded ||
873 csum_info->receive.udp_checksum_succeeded)
874 skb->ip_summed = CHECKSUM_UNNECESSARY;
877 if (hash_info && (net->features & NETIF_F_RXHASH))
878 skb_set_hash(skb, *hash_info, PKT_HASH_TYPE_L4);
881 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT) |
882 (vlan->cfi ? VLAN_CFI_MASK : 0);
884 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
892 * netvsc_recv_callback - Callback when we receive a packet from the
893 * "wire" on the specified device.
895 int netvsc_recv_callback(struct net_device *net,
896 struct netvsc_device *net_device,
897 struct netvsc_channel *nvchan)
899 struct net_device_context *net_device_ctx = netdev_priv(net);
900 struct vmbus_channel *channel = nvchan->channel;
901 u16 q_idx = channel->offermsg.offer.sub_channel_index;
903 struct netvsc_stats *rx_stats = &nvchan->rx_stats;
907 if (net->reg_state != NETREG_REGISTERED)
908 return NVSP_STAT_FAIL;
910 act = netvsc_run_xdp(net, nvchan, &xdp);
912 if (act != XDP_PASS && act != XDP_TX) {
913 u64_stats_update_begin(&rx_stats->syncp);
914 rx_stats->xdp_drop++;
915 u64_stats_update_end(&rx_stats->syncp);
917 return NVSP_STAT_SUCCESS; /* consumed by XDP */
920 /* Allocate a skb - TODO direct I/O to pages? */
921 skb = netvsc_alloc_recv_skb(net, nvchan, &xdp);
923 if (unlikely(!skb)) {
924 ++net_device_ctx->eth_stats.rx_no_memory;
925 return NVSP_STAT_FAIL;
928 skb_record_rx_queue(skb, q_idx);
931 * Even if injecting the packet, record the statistics
932 * on the synthetic device because modifying the VF device
933 * statistics will not work correctly.
935 u64_stats_update_begin(&rx_stats->syncp);
937 rx_stats->bytes += nvchan->rsc.pktlen;
939 if (skb->pkt_type == PACKET_BROADCAST)
940 ++rx_stats->broadcast;
941 else if (skb->pkt_type == PACKET_MULTICAST)
942 ++rx_stats->multicast;
943 u64_stats_update_end(&rx_stats->syncp);
946 netvsc_xdp_xmit(skb, net);
947 return NVSP_STAT_SUCCESS;
950 napi_gro_receive(&nvchan->napi, skb);
951 return NVSP_STAT_SUCCESS;
954 static void netvsc_get_drvinfo(struct net_device *net,
955 struct ethtool_drvinfo *info)
957 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
958 strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
961 static void netvsc_get_channels(struct net_device *net,
962 struct ethtool_channels *channel)
964 struct net_device_context *net_device_ctx = netdev_priv(net);
965 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
968 channel->max_combined = nvdev->max_chn;
969 channel->combined_count = nvdev->num_chn;
973 /* Alloc struct netvsc_device_info, and initialize it from either existing
974 * struct netvsc_device, or from default values.
977 struct netvsc_device_info *netvsc_devinfo_get(struct netvsc_device *nvdev)
979 struct netvsc_device_info *dev_info;
980 struct bpf_prog *prog;
982 dev_info = kzalloc(sizeof(*dev_info), GFP_ATOMIC);
990 dev_info->num_chn = nvdev->num_chn;
991 dev_info->send_sections = nvdev->send_section_cnt;
992 dev_info->send_section_size = nvdev->send_section_size;
993 dev_info->recv_sections = nvdev->recv_section_cnt;
994 dev_info->recv_section_size = nvdev->recv_section_size;
996 memcpy(dev_info->rss_key, nvdev->extension->rss_key,
999 prog = netvsc_xdp_get(nvdev);
1002 dev_info->bprog = prog;
1005 dev_info->num_chn = VRSS_CHANNEL_DEFAULT;
1006 dev_info->send_sections = NETVSC_DEFAULT_TX;
1007 dev_info->send_section_size = NETVSC_SEND_SECTION_SIZE;
1008 dev_info->recv_sections = NETVSC_DEFAULT_RX;
1009 dev_info->recv_section_size = NETVSC_RECV_SECTION_SIZE;
1015 /* Free struct netvsc_device_info */
1016 static void netvsc_devinfo_put(struct netvsc_device_info *dev_info)
1018 if (dev_info->bprog) {
1020 bpf_prog_put(dev_info->bprog);
1026 static int netvsc_detach(struct net_device *ndev,
1027 struct netvsc_device *nvdev)
1029 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1030 struct hv_device *hdev = ndev_ctx->device_ctx;
1033 /* Don't try continuing to try and setup sub channels */
1034 if (cancel_work_sync(&nvdev->subchan_work))
1037 netvsc_xdp_set(ndev, NULL, NULL, nvdev);
1039 /* If device was up (receiving) then shutdown */
1040 if (netif_running(ndev)) {
1041 netvsc_tx_disable(nvdev, ndev);
1043 ret = rndis_filter_close(nvdev);
1046 "unable to close device (ret %d).\n", ret);
1050 ret = netvsc_wait_until_empty(nvdev);
1053 "Ring buffer not empty after closing rndis\n");
1058 netif_device_detach(ndev);
1060 rndis_filter_device_remove(hdev, nvdev);
1065 static int netvsc_attach(struct net_device *ndev,
1066 struct netvsc_device_info *dev_info)
1068 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1069 struct hv_device *hdev = ndev_ctx->device_ctx;
1070 struct netvsc_device *nvdev;
1071 struct rndis_device *rdev;
1072 struct bpf_prog *prog;
1075 nvdev = rndis_filter_device_add(hdev, dev_info);
1077 return PTR_ERR(nvdev);
1079 if (nvdev->num_chn > 1) {
1080 ret = rndis_set_subchannel(ndev, nvdev, dev_info);
1082 /* if unavailable, just proceed with one queue */
1089 prog = dev_info->bprog;
1092 ret = netvsc_xdp_set(ndev, prog, NULL, nvdev);
1099 /* In any case device is now ready */
1100 nvdev->tx_disable = false;
1101 netif_device_attach(ndev);
1103 /* Note: enable and attach happen when sub-channels setup */
1104 netif_carrier_off(ndev);
1106 if (netif_running(ndev)) {
1107 ret = rndis_filter_open(nvdev);
1111 rdev = nvdev->extension;
1112 if (!rdev->link_state)
1113 netif_carrier_on(ndev);
1119 netif_device_detach(ndev);
1122 rndis_filter_device_remove(hdev, nvdev);
1127 static int netvsc_set_channels(struct net_device *net,
1128 struct ethtool_channels *channels)
1130 struct net_device_context *net_device_ctx = netdev_priv(net);
1131 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
1132 unsigned int orig, count = channels->combined_count;
1133 struct netvsc_device_info *device_info;
1136 /* We do not support separate count for rx, tx, or other */
1138 channels->rx_count || channels->tx_count || channels->other_count)
1141 if (!nvdev || nvdev->destroy)
1144 if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
1147 if (count > nvdev->max_chn)
1150 orig = nvdev->num_chn;
1152 device_info = netvsc_devinfo_get(nvdev);
1157 device_info->num_chn = count;
1159 ret = netvsc_detach(net, nvdev);
1163 ret = netvsc_attach(net, device_info);
1165 device_info->num_chn = orig;
1166 if (netvsc_attach(net, device_info))
1167 netdev_err(net, "restoring channel setting failed\n");
1171 netvsc_devinfo_put(device_info);
1175 static void netvsc_init_settings(struct net_device *dev)
1177 struct net_device_context *ndc = netdev_priv(dev);
1179 ndc->l4_hash = HV_DEFAULT_L4HASH;
1181 ndc->speed = SPEED_UNKNOWN;
1182 ndc->duplex = DUPLEX_FULL;
1184 dev->features = NETIF_F_LRO;
1187 static int netvsc_get_link_ksettings(struct net_device *dev,
1188 struct ethtool_link_ksettings *cmd)
1190 struct net_device_context *ndc = netdev_priv(dev);
1191 struct net_device *vf_netdev;
1193 vf_netdev = rtnl_dereference(ndc->vf_netdev);
1196 return __ethtool_get_link_ksettings(vf_netdev, cmd);
1198 cmd->base.speed = ndc->speed;
1199 cmd->base.duplex = ndc->duplex;
1200 cmd->base.port = PORT_OTHER;
1205 static int netvsc_set_link_ksettings(struct net_device *dev,
1206 const struct ethtool_link_ksettings *cmd)
1208 struct net_device_context *ndc = netdev_priv(dev);
1209 struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1212 if (!vf_netdev->ethtool_ops->set_link_ksettings)
1215 return vf_netdev->ethtool_ops->set_link_ksettings(vf_netdev,
1219 return ethtool_virtdev_set_link_ksettings(dev, cmd,
1220 &ndc->speed, &ndc->duplex);
1223 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
1225 struct net_device_context *ndevctx = netdev_priv(ndev);
1226 struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1227 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1228 int orig_mtu = ndev->mtu;
1229 struct netvsc_device_info *device_info;
1232 if (!nvdev || nvdev->destroy)
1235 device_info = netvsc_devinfo_get(nvdev);
1240 /* Change MTU of underlying VF netdev first. */
1242 ret = dev_set_mtu(vf_netdev, mtu);
1247 ret = netvsc_detach(ndev, nvdev);
1253 ret = netvsc_attach(ndev, device_info);
1257 /* Attempt rollback to original MTU */
1258 ndev->mtu = orig_mtu;
1260 if (netvsc_attach(ndev, device_info))
1261 netdev_err(ndev, "restoring mtu failed\n");
1264 dev_set_mtu(vf_netdev, orig_mtu);
1267 netvsc_devinfo_put(device_info);
1271 static void netvsc_get_vf_stats(struct net_device *net,
1272 struct netvsc_vf_pcpu_stats *tot)
1274 struct net_device_context *ndev_ctx = netdev_priv(net);
1277 memset(tot, 0, sizeof(*tot));
1279 for_each_possible_cpu(i) {
1280 const struct netvsc_vf_pcpu_stats *stats
1281 = per_cpu_ptr(ndev_ctx->vf_stats, i);
1282 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1286 start = u64_stats_fetch_begin_irq(&stats->syncp);
1287 rx_packets = stats->rx_packets;
1288 tx_packets = stats->tx_packets;
1289 rx_bytes = stats->rx_bytes;
1290 tx_bytes = stats->tx_bytes;
1291 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1293 tot->rx_packets += rx_packets;
1294 tot->tx_packets += tx_packets;
1295 tot->rx_bytes += rx_bytes;
1296 tot->tx_bytes += tx_bytes;
1297 tot->tx_dropped += stats->tx_dropped;
1301 static void netvsc_get_pcpu_stats(struct net_device *net,
1302 struct netvsc_ethtool_pcpu_stats *pcpu_tot)
1304 struct net_device_context *ndev_ctx = netdev_priv(net);
1305 struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1308 /* fetch percpu stats of vf */
1309 for_each_possible_cpu(i) {
1310 const struct netvsc_vf_pcpu_stats *stats =
1311 per_cpu_ptr(ndev_ctx->vf_stats, i);
1312 struct netvsc_ethtool_pcpu_stats *this_tot = &pcpu_tot[i];
1316 start = u64_stats_fetch_begin_irq(&stats->syncp);
1317 this_tot->vf_rx_packets = stats->rx_packets;
1318 this_tot->vf_tx_packets = stats->tx_packets;
1319 this_tot->vf_rx_bytes = stats->rx_bytes;
1320 this_tot->vf_tx_bytes = stats->tx_bytes;
1321 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1322 this_tot->rx_packets = this_tot->vf_rx_packets;
1323 this_tot->tx_packets = this_tot->vf_tx_packets;
1324 this_tot->rx_bytes = this_tot->vf_rx_bytes;
1325 this_tot->tx_bytes = this_tot->vf_tx_bytes;
1328 /* fetch percpu stats of netvsc */
1329 for (i = 0; i < nvdev->num_chn; i++) {
1330 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1331 const struct netvsc_stats *stats;
1332 struct netvsc_ethtool_pcpu_stats *this_tot =
1333 &pcpu_tot[nvchan->channel->target_cpu];
1337 stats = &nvchan->tx_stats;
1339 start = u64_stats_fetch_begin_irq(&stats->syncp);
1340 packets = stats->packets;
1341 bytes = stats->bytes;
1342 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1344 this_tot->tx_bytes += bytes;
1345 this_tot->tx_packets += packets;
1347 stats = &nvchan->rx_stats;
1349 start = u64_stats_fetch_begin_irq(&stats->syncp);
1350 packets = stats->packets;
1351 bytes = stats->bytes;
1352 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1354 this_tot->rx_bytes += bytes;
1355 this_tot->rx_packets += packets;
1359 static void netvsc_get_stats64(struct net_device *net,
1360 struct rtnl_link_stats64 *t)
1362 struct net_device_context *ndev_ctx = netdev_priv(net);
1363 struct netvsc_device *nvdev;
1364 struct netvsc_vf_pcpu_stats vf_tot;
1369 nvdev = rcu_dereference(ndev_ctx->nvdev);
1373 netdev_stats_to_stats64(t, &net->stats);
1375 netvsc_get_vf_stats(net, &vf_tot);
1376 t->rx_packets += vf_tot.rx_packets;
1377 t->tx_packets += vf_tot.tx_packets;
1378 t->rx_bytes += vf_tot.rx_bytes;
1379 t->tx_bytes += vf_tot.tx_bytes;
1380 t->tx_dropped += vf_tot.tx_dropped;
1382 for (i = 0; i < nvdev->num_chn; i++) {
1383 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1384 const struct netvsc_stats *stats;
1385 u64 packets, bytes, multicast;
1388 stats = &nvchan->tx_stats;
1390 start = u64_stats_fetch_begin_irq(&stats->syncp);
1391 packets = stats->packets;
1392 bytes = stats->bytes;
1393 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1395 t->tx_bytes += bytes;
1396 t->tx_packets += packets;
1398 stats = &nvchan->rx_stats;
1400 start = u64_stats_fetch_begin_irq(&stats->syncp);
1401 packets = stats->packets;
1402 bytes = stats->bytes;
1403 multicast = stats->multicast + stats->broadcast;
1404 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1406 t->rx_bytes += bytes;
1407 t->rx_packets += packets;
1408 t->multicast += multicast;
1414 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1416 struct net_device_context *ndc = netdev_priv(ndev);
1417 struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1418 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1419 struct sockaddr *addr = p;
1422 err = eth_prepare_mac_addr_change(ndev, p);
1430 err = dev_set_mac_address(vf_netdev, addr, NULL);
1435 err = rndis_filter_set_device_mac(nvdev, addr->sa_data);
1437 eth_commit_mac_addr_change(ndev, p);
1438 } else if (vf_netdev) {
1439 /* rollback change on VF */
1440 memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN);
1441 dev_set_mac_address(vf_netdev, addr, NULL);
1447 static const struct {
1448 char name[ETH_GSTRING_LEN];
1450 } netvsc_stats[] = {
1451 { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1452 { "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1453 { "tx_no_space", offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1454 { "tx_too_big", offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1455 { "tx_busy", offsetof(struct netvsc_ethtool_stats, tx_busy) },
1456 { "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) },
1457 { "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) },
1458 { "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) },
1459 { "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) },
1460 { "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) },
1461 { "vlan_error", offsetof(struct netvsc_ethtool_stats, vlan_error) },
1463 { "cpu%u_rx_packets",
1464 offsetof(struct netvsc_ethtool_pcpu_stats, rx_packets) },
1466 offsetof(struct netvsc_ethtool_pcpu_stats, rx_bytes) },
1467 { "cpu%u_tx_packets",
1468 offsetof(struct netvsc_ethtool_pcpu_stats, tx_packets) },
1470 offsetof(struct netvsc_ethtool_pcpu_stats, tx_bytes) },
1471 { "cpu%u_vf_rx_packets",
1472 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_packets) },
1473 { "cpu%u_vf_rx_bytes",
1474 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_bytes) },
1475 { "cpu%u_vf_tx_packets",
1476 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_packets) },
1477 { "cpu%u_vf_tx_bytes",
1478 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_bytes) },
1480 { "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) },
1481 { "vf_rx_bytes", offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) },
1482 { "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) },
1483 { "vf_tx_bytes", offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) },
1484 { "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) },
1487 #define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats)
1488 #define NETVSC_VF_STATS_LEN ARRAY_SIZE(vf_stats)
1490 /* statistics per queue (rx/tx packets/bytes) */
1491 #define NETVSC_PCPU_STATS_LEN (num_present_cpus() * ARRAY_SIZE(pcpu_stats))
1493 /* 5 statistics per queue (rx/tx packets/bytes, rx xdp_drop) */
1494 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 5)
1496 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1498 struct net_device_context *ndc = netdev_priv(dev);
1499 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1504 switch (string_set) {
1506 return NETVSC_GLOBAL_STATS_LEN
1507 + NETVSC_VF_STATS_LEN
1508 + NETVSC_QUEUE_STATS_LEN(nvdev)
1509 + NETVSC_PCPU_STATS_LEN;
1515 static void netvsc_get_ethtool_stats(struct net_device *dev,
1516 struct ethtool_stats *stats, u64 *data)
1518 struct net_device_context *ndc = netdev_priv(dev);
1519 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1520 const void *nds = &ndc->eth_stats;
1521 const struct netvsc_stats *qstats;
1522 struct netvsc_vf_pcpu_stats sum;
1523 struct netvsc_ethtool_pcpu_stats *pcpu_sum;
1532 for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1533 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1535 netvsc_get_vf_stats(dev, &sum);
1536 for (j = 0; j < NETVSC_VF_STATS_LEN; j++)
1537 data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset);
1539 for (j = 0; j < nvdev->num_chn; j++) {
1540 qstats = &nvdev->chan_table[j].tx_stats;
1543 start = u64_stats_fetch_begin_irq(&qstats->syncp);
1544 packets = qstats->packets;
1545 bytes = qstats->bytes;
1546 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1547 data[i++] = packets;
1550 qstats = &nvdev->chan_table[j].rx_stats;
1552 start = u64_stats_fetch_begin_irq(&qstats->syncp);
1553 packets = qstats->packets;
1554 bytes = qstats->bytes;
1555 xdp_drop = qstats->xdp_drop;
1556 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1557 data[i++] = packets;
1559 data[i++] = xdp_drop;
1562 pcpu_sum = kvmalloc_array(num_possible_cpus(),
1563 sizeof(struct netvsc_ethtool_pcpu_stats),
1568 netvsc_get_pcpu_stats(dev, pcpu_sum);
1569 for_each_present_cpu(cpu) {
1570 struct netvsc_ethtool_pcpu_stats *this_sum = &pcpu_sum[cpu];
1572 for (j = 0; j < ARRAY_SIZE(pcpu_stats); j++)
1573 data[i++] = *(u64 *)((void *)this_sum
1574 + pcpu_stats[j].offset);
1579 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1581 struct net_device_context *ndc = netdev_priv(dev);
1582 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1589 switch (stringset) {
1591 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++) {
1592 memcpy(p, netvsc_stats[i].name, ETH_GSTRING_LEN);
1593 p += ETH_GSTRING_LEN;
1596 for (i = 0; i < ARRAY_SIZE(vf_stats); i++) {
1597 memcpy(p, vf_stats[i].name, ETH_GSTRING_LEN);
1598 p += ETH_GSTRING_LEN;
1601 for (i = 0; i < nvdev->num_chn; i++) {
1602 sprintf(p, "tx_queue_%u_packets", i);
1603 p += ETH_GSTRING_LEN;
1604 sprintf(p, "tx_queue_%u_bytes", i);
1605 p += ETH_GSTRING_LEN;
1606 sprintf(p, "rx_queue_%u_packets", i);
1607 p += ETH_GSTRING_LEN;
1608 sprintf(p, "rx_queue_%u_bytes", i);
1609 p += ETH_GSTRING_LEN;
1610 sprintf(p, "rx_queue_%u_xdp_drop", i);
1611 p += ETH_GSTRING_LEN;
1614 for_each_present_cpu(cpu) {
1615 for (i = 0; i < ARRAY_SIZE(pcpu_stats); i++) {
1616 sprintf(p, pcpu_stats[i].name, cpu);
1617 p += ETH_GSTRING_LEN;
1626 netvsc_get_rss_hash_opts(struct net_device_context *ndc,
1627 struct ethtool_rxnfc *info)
1629 const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3;
1631 info->data = RXH_IP_SRC | RXH_IP_DST;
1633 switch (info->flow_type) {
1635 if (ndc->l4_hash & HV_TCP4_L4HASH)
1636 info->data |= l4_flag;
1641 if (ndc->l4_hash & HV_TCP6_L4HASH)
1642 info->data |= l4_flag;
1647 if (ndc->l4_hash & HV_UDP4_L4HASH)
1648 info->data |= l4_flag;
1653 if (ndc->l4_hash & HV_UDP6_L4HASH)
1654 info->data |= l4_flag;
1670 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1673 struct net_device_context *ndc = netdev_priv(dev);
1674 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1679 switch (info->cmd) {
1680 case ETHTOOL_GRXRINGS:
1681 info->data = nvdev->num_chn;
1685 return netvsc_get_rss_hash_opts(ndc, info);
1690 static int netvsc_set_rss_hash_opts(struct net_device_context *ndc,
1691 struct ethtool_rxnfc *info)
1693 if (info->data == (RXH_IP_SRC | RXH_IP_DST |
1694 RXH_L4_B_0_1 | RXH_L4_B_2_3)) {
1695 switch (info->flow_type) {
1697 ndc->l4_hash |= HV_TCP4_L4HASH;
1701 ndc->l4_hash |= HV_TCP6_L4HASH;
1705 ndc->l4_hash |= HV_UDP4_L4HASH;
1709 ndc->l4_hash |= HV_UDP6_L4HASH;
1719 if (info->data == (RXH_IP_SRC | RXH_IP_DST)) {
1720 switch (info->flow_type) {
1722 ndc->l4_hash &= ~HV_TCP4_L4HASH;
1726 ndc->l4_hash &= ~HV_TCP6_L4HASH;
1730 ndc->l4_hash &= ~HV_UDP4_L4HASH;
1734 ndc->l4_hash &= ~HV_UDP6_L4HASH;
1748 netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info)
1750 struct net_device_context *ndc = netdev_priv(ndev);
1752 if (info->cmd == ETHTOOL_SRXFH)
1753 return netvsc_set_rss_hash_opts(ndc, info);
1758 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1760 return NETVSC_HASH_KEYLEN;
1763 static u32 netvsc_rss_indir_size(struct net_device *dev)
1768 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1771 struct net_device_context *ndc = netdev_priv(dev);
1772 struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1773 struct rndis_device *rndis_dev;
1780 *hfunc = ETH_RSS_HASH_TOP; /* Toeplitz */
1782 rndis_dev = ndev->extension;
1784 for (i = 0; i < ITAB_NUM; i++)
1785 indir[i] = ndc->rx_table[i];
1789 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1794 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1795 const u8 *key, const u8 hfunc)
1797 struct net_device_context *ndc = netdev_priv(dev);
1798 struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1799 struct rndis_device *rndis_dev;
1805 if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1808 rndis_dev = ndev->extension;
1810 for (i = 0; i < ITAB_NUM; i++)
1811 if (indir[i] >= ndev->num_chn)
1814 for (i = 0; i < ITAB_NUM; i++)
1815 ndc->rx_table[i] = indir[i];
1822 key = rndis_dev->rss_key;
1825 return rndis_filter_set_rss_param(rndis_dev, key);
1828 /* Hyper-V RNDIS protocol does not have ring in the HW sense.
1829 * It does have pre-allocated receive area which is divided into sections.
1831 static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
1832 struct ethtool_ringparam *ring)
1836 ring->rx_pending = nvdev->recv_section_cnt;
1837 ring->tx_pending = nvdev->send_section_cnt;
1839 if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
1840 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
1842 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1844 ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
1845 ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
1846 / nvdev->send_section_size;
1849 static void netvsc_get_ringparam(struct net_device *ndev,
1850 struct ethtool_ringparam *ring)
1852 struct net_device_context *ndevctx = netdev_priv(ndev);
1853 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1858 __netvsc_get_ringparam(nvdev, ring);
1861 static int netvsc_set_ringparam(struct net_device *ndev,
1862 struct ethtool_ringparam *ring)
1864 struct net_device_context *ndevctx = netdev_priv(ndev);
1865 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1866 struct netvsc_device_info *device_info;
1867 struct ethtool_ringparam orig;
1871 if (!nvdev || nvdev->destroy)
1874 memset(&orig, 0, sizeof(orig));
1875 __netvsc_get_ringparam(nvdev, &orig);
1877 new_tx = clamp_t(u32, ring->tx_pending,
1878 NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
1879 new_rx = clamp_t(u32, ring->rx_pending,
1880 NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
1882 if (new_tx == orig.tx_pending &&
1883 new_rx == orig.rx_pending)
1884 return 0; /* no change */
1886 device_info = netvsc_devinfo_get(nvdev);
1891 device_info->send_sections = new_tx;
1892 device_info->recv_sections = new_rx;
1894 ret = netvsc_detach(ndev, nvdev);
1898 ret = netvsc_attach(ndev, device_info);
1900 device_info->send_sections = orig.tx_pending;
1901 device_info->recv_sections = orig.rx_pending;
1903 if (netvsc_attach(ndev, device_info))
1904 netdev_err(ndev, "restoring ringparam failed");
1908 netvsc_devinfo_put(device_info);
1912 static netdev_features_t netvsc_fix_features(struct net_device *ndev,
1913 netdev_features_t features)
1915 struct net_device_context *ndevctx = netdev_priv(ndev);
1916 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1918 if (!nvdev || nvdev->destroy)
1921 if ((features & NETIF_F_LRO) && netvsc_xdp_get(nvdev)) {
1922 features ^= NETIF_F_LRO;
1923 netdev_info(ndev, "Skip LRO - unsupported with XDP\n");
1929 static int netvsc_set_features(struct net_device *ndev,
1930 netdev_features_t features)
1932 netdev_features_t change = features ^ ndev->features;
1933 struct net_device_context *ndevctx = netdev_priv(ndev);
1934 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1935 struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1936 struct ndis_offload_params offloads;
1939 if (!nvdev || nvdev->destroy)
1942 if (!(change & NETIF_F_LRO))
1945 memset(&offloads, 0, sizeof(struct ndis_offload_params));
1947 if (features & NETIF_F_LRO) {
1948 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1949 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1951 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1952 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1955 ret = rndis_filter_set_offload_params(ndev, nvdev, &offloads);
1958 features ^= NETIF_F_LRO;
1959 ndev->features = features;
1966 vf_netdev->wanted_features = features;
1967 netdev_update_features(vf_netdev);
1972 static int netvsc_get_regs_len(struct net_device *netdev)
1974 return VRSS_SEND_TAB_SIZE * sizeof(u32);
1977 static void netvsc_get_regs(struct net_device *netdev,
1978 struct ethtool_regs *regs, void *p)
1980 struct net_device_context *ndc = netdev_priv(netdev);
1983 /* increase the version, if buffer format is changed. */
1986 memcpy(regs_buff, ndc->tx_table, VRSS_SEND_TAB_SIZE * sizeof(u32));
1989 static u32 netvsc_get_msglevel(struct net_device *ndev)
1991 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1993 return ndev_ctx->msg_enable;
1996 static void netvsc_set_msglevel(struct net_device *ndev, u32 val)
1998 struct net_device_context *ndev_ctx = netdev_priv(ndev);
2000 ndev_ctx->msg_enable = val;
2003 static const struct ethtool_ops ethtool_ops = {
2004 .get_drvinfo = netvsc_get_drvinfo,
2005 .get_regs_len = netvsc_get_regs_len,
2006 .get_regs = netvsc_get_regs,
2007 .get_msglevel = netvsc_get_msglevel,
2008 .set_msglevel = netvsc_set_msglevel,
2009 .get_link = ethtool_op_get_link,
2010 .get_ethtool_stats = netvsc_get_ethtool_stats,
2011 .get_sset_count = netvsc_get_sset_count,
2012 .get_strings = netvsc_get_strings,
2013 .get_channels = netvsc_get_channels,
2014 .set_channels = netvsc_set_channels,
2015 .get_ts_info = ethtool_op_get_ts_info,
2016 .get_rxnfc = netvsc_get_rxnfc,
2017 .set_rxnfc = netvsc_set_rxnfc,
2018 .get_rxfh_key_size = netvsc_get_rxfh_key_size,
2019 .get_rxfh_indir_size = netvsc_rss_indir_size,
2020 .get_rxfh = netvsc_get_rxfh,
2021 .set_rxfh = netvsc_set_rxfh,
2022 .get_link_ksettings = netvsc_get_link_ksettings,
2023 .set_link_ksettings = netvsc_set_link_ksettings,
2024 .get_ringparam = netvsc_get_ringparam,
2025 .set_ringparam = netvsc_set_ringparam,
2028 static const struct net_device_ops device_ops = {
2029 .ndo_open = netvsc_open,
2030 .ndo_stop = netvsc_close,
2031 .ndo_start_xmit = netvsc_start_xmit,
2032 .ndo_change_rx_flags = netvsc_change_rx_flags,
2033 .ndo_set_rx_mode = netvsc_set_rx_mode,
2034 .ndo_fix_features = netvsc_fix_features,
2035 .ndo_set_features = netvsc_set_features,
2036 .ndo_change_mtu = netvsc_change_mtu,
2037 .ndo_validate_addr = eth_validate_addr,
2038 .ndo_set_mac_address = netvsc_set_mac_addr,
2039 .ndo_select_queue = netvsc_select_queue,
2040 .ndo_get_stats64 = netvsc_get_stats64,
2041 .ndo_bpf = netvsc_bpf,
2045 * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
2046 * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
2047 * present send GARP packet to network peers with netif_notify_peers().
2049 static void netvsc_link_change(struct work_struct *w)
2051 struct net_device_context *ndev_ctx =
2052 container_of(w, struct net_device_context, dwork.work);
2053 struct hv_device *device_obj = ndev_ctx->device_ctx;
2054 struct net_device *net = hv_get_drvdata(device_obj);
2055 struct netvsc_device *net_device;
2056 struct rndis_device *rdev;
2057 struct netvsc_reconfig *event = NULL;
2058 bool notify = false, reschedule = false;
2059 unsigned long flags, next_reconfig, delay;
2061 /* if changes are happening, comeback later */
2062 if (!rtnl_trylock()) {
2063 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
2067 net_device = rtnl_dereference(ndev_ctx->nvdev);
2071 rdev = net_device->extension;
2073 next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
2074 if (time_is_after_jiffies(next_reconfig)) {
2075 /* link_watch only sends one notification with current state
2076 * per second, avoid doing reconfig more frequently. Handle
2079 delay = next_reconfig - jiffies;
2080 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
2081 schedule_delayed_work(&ndev_ctx->dwork, delay);
2084 ndev_ctx->last_reconfig = jiffies;
2086 spin_lock_irqsave(&ndev_ctx->lock, flags);
2087 if (!list_empty(&ndev_ctx->reconfig_events)) {
2088 event = list_first_entry(&ndev_ctx->reconfig_events,
2089 struct netvsc_reconfig, list);
2090 list_del(&event->list);
2091 reschedule = !list_empty(&ndev_ctx->reconfig_events);
2093 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2098 switch (event->event) {
2099 /* Only the following events are possible due to the check in
2100 * netvsc_linkstatus_callback()
2102 case RNDIS_STATUS_MEDIA_CONNECT:
2103 if (rdev->link_state) {
2104 rdev->link_state = false;
2105 netif_carrier_on(net);
2106 netvsc_tx_enable(net_device, net);
2112 case RNDIS_STATUS_MEDIA_DISCONNECT:
2113 if (!rdev->link_state) {
2114 rdev->link_state = true;
2115 netif_carrier_off(net);
2116 netvsc_tx_disable(net_device, net);
2120 case RNDIS_STATUS_NETWORK_CHANGE:
2121 /* Only makes sense if carrier is present */
2122 if (!rdev->link_state) {
2123 rdev->link_state = true;
2124 netif_carrier_off(net);
2125 netvsc_tx_disable(net_device, net);
2126 event->event = RNDIS_STATUS_MEDIA_CONNECT;
2127 spin_lock_irqsave(&ndev_ctx->lock, flags);
2128 list_add(&event->list, &ndev_ctx->reconfig_events);
2129 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2138 netdev_notify_peers(net);
2140 /* link_watch only sends one notification with current state per
2141 * second, handle next reconfig event in 2 seconds.
2144 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
2152 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
2154 struct net_device_context *net_device_ctx;
2155 struct net_device *dev;
2157 dev = netdev_master_upper_dev_get(vf_netdev);
2158 if (!dev || dev->netdev_ops != &device_ops)
2159 return NULL; /* not a netvsc device */
2161 net_device_ctx = netdev_priv(dev);
2162 if (!rtnl_dereference(net_device_ctx->nvdev))
2163 return NULL; /* device is removed */
2168 /* Called when VF is injecting data into network stack.
2169 * Change the associated network device from VF to netvsc.
2170 * note: already called with rcu_read_lock
2172 static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
2174 struct sk_buff *skb = *pskb;
2175 struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data);
2176 struct net_device_context *ndev_ctx = netdev_priv(ndev);
2177 struct netvsc_vf_pcpu_stats *pcpu_stats
2178 = this_cpu_ptr(ndev_ctx->vf_stats);
2180 skb = skb_share_check(skb, GFP_ATOMIC);
2182 return RX_HANDLER_CONSUMED;
2188 u64_stats_update_begin(&pcpu_stats->syncp);
2189 pcpu_stats->rx_packets++;
2190 pcpu_stats->rx_bytes += skb->len;
2191 u64_stats_update_end(&pcpu_stats->syncp);
2193 return RX_HANDLER_ANOTHER;
2196 static int netvsc_vf_join(struct net_device *vf_netdev,
2197 struct net_device *ndev)
2199 struct net_device_context *ndev_ctx = netdev_priv(ndev);
2202 ret = netdev_rx_handler_register(vf_netdev,
2203 netvsc_vf_handle_frame, ndev);
2205 netdev_err(vf_netdev,
2206 "can not register netvsc VF receive handler (err = %d)\n",
2208 goto rx_handler_failed;
2211 ret = netdev_master_upper_dev_link(vf_netdev, ndev,
2214 netdev_err(vf_netdev,
2215 "can not set master device %s (err = %d)\n",
2217 goto upper_link_failed;
2220 /* set slave flag before open to prevent IPv6 addrconf */
2221 vf_netdev->flags |= IFF_SLAVE;
2223 schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
2225 call_netdevice_notifiers(NETDEV_JOIN, vf_netdev);
2227 netdev_info(vf_netdev, "joined to %s\n", ndev->name);
2231 netdev_rx_handler_unregister(vf_netdev);
2236 static void __netvsc_vf_setup(struct net_device *ndev,
2237 struct net_device *vf_netdev)
2241 /* Align MTU of VF with master */
2242 ret = dev_set_mtu(vf_netdev, ndev->mtu);
2244 netdev_warn(vf_netdev,
2245 "unable to change mtu to %u\n", ndev->mtu);
2247 /* set multicast etc flags on VF */
2248 dev_change_flags(vf_netdev, ndev->flags | IFF_SLAVE, NULL);
2250 /* sync address list from ndev to VF */
2251 netif_addr_lock_bh(ndev);
2252 dev_uc_sync(vf_netdev, ndev);
2253 dev_mc_sync(vf_netdev, ndev);
2254 netif_addr_unlock_bh(ndev);
2256 if (netif_running(ndev)) {
2257 ret = dev_open(vf_netdev, NULL);
2259 netdev_warn(vf_netdev,
2260 "unable to open: %d\n", ret);
2264 /* Setup VF as slave of the synthetic device.
2265 * Runs in workqueue to avoid recursion in netlink callbacks.
2267 static void netvsc_vf_setup(struct work_struct *w)
2269 struct net_device_context *ndev_ctx
2270 = container_of(w, struct net_device_context, vf_takeover.work);
2271 struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2272 struct net_device *vf_netdev;
2274 if (!rtnl_trylock()) {
2275 schedule_delayed_work(&ndev_ctx->vf_takeover, 0);
2279 vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2281 __netvsc_vf_setup(ndev, vf_netdev);
2286 /* Find netvsc by VF serial number.
2287 * The PCI hyperv controller records the serial number as the slot kobj name.
2289 static struct net_device *get_netvsc_byslot(const struct net_device *vf_netdev)
2291 struct device *parent = vf_netdev->dev.parent;
2292 struct net_device_context *ndev_ctx;
2293 struct net_device *ndev;
2294 struct pci_dev *pdev;
2297 if (!parent || !dev_is_pci(parent))
2298 return NULL; /* not a PCI device */
2300 pdev = to_pci_dev(parent);
2302 netdev_notice(vf_netdev, "no PCI slot information\n");
2306 if (kstrtou32(pci_slot_name(pdev->slot), 10, &serial)) {
2307 netdev_notice(vf_netdev, "Invalid vf serial:%s\n",
2308 pci_slot_name(pdev->slot));
2312 list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
2313 if (!ndev_ctx->vf_alloc)
2316 if (ndev_ctx->vf_serial == serial)
2317 return hv_get_drvdata(ndev_ctx->device_ctx);
2320 /* Fallback path to check synthetic vf with
2323 list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
2324 ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2325 if (ether_addr_equal(vf_netdev->perm_addr, ndev->perm_addr)) {
2326 netdev_notice(vf_netdev,
2327 "falling back to mac addr based matching\n");
2332 netdev_notice(vf_netdev,
2333 "no netdev found for vf serial:%u\n", serial);
2337 static int netvsc_register_vf(struct net_device *vf_netdev)
2339 struct net_device_context *net_device_ctx;
2340 struct netvsc_device *netvsc_dev;
2341 struct bpf_prog *prog;
2342 struct net_device *ndev;
2345 if (vf_netdev->addr_len != ETH_ALEN)
2348 ndev = get_netvsc_byslot(vf_netdev);
2352 net_device_ctx = netdev_priv(ndev);
2353 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2354 if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
2357 /* if synthetic interface is a different namespace,
2358 * then move the VF to that namespace; join will be
2359 * done again in that context.
2361 if (!net_eq(dev_net(ndev), dev_net(vf_netdev))) {
2362 ret = dev_change_net_namespace(vf_netdev,
2363 dev_net(ndev), "eth%d");
2365 netdev_err(vf_netdev,
2366 "could not move to same namespace as %s: %d\n",
2369 netdev_info(vf_netdev,
2370 "VF moved to namespace with: %s\n",
2375 netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
2377 if (netvsc_vf_join(vf_netdev, ndev) != 0)
2380 dev_hold(vf_netdev);
2381 rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
2383 vf_netdev->wanted_features = ndev->features;
2384 netdev_update_features(vf_netdev);
2386 prog = netvsc_xdp_get(netvsc_dev);
2387 netvsc_vf_setxdp(vf_netdev, prog);
2392 /* Change the data path when VF UP/DOWN/CHANGE are detected.
2394 * Typically a UP or DOWN event is followed by a CHANGE event, so
2395 * net_device_ctx->data_path_is_vf is used to cache the current data path
2396 * to avoid the duplicate call of netvsc_switch_datapath() and the duplicate
2399 * During hibernation, if a VF NIC driver (e.g. mlx5) preserves the network
2400 * interface, there is only the CHANGE event and no UP or DOWN event.
2402 static int netvsc_vf_changed(struct net_device *vf_netdev)
2404 struct net_device_context *net_device_ctx;
2405 struct netvsc_device *netvsc_dev;
2406 struct net_device *ndev;
2407 bool vf_is_up = netif_running(vf_netdev);
2409 ndev = get_netvsc_byref(vf_netdev);
2413 net_device_ctx = netdev_priv(ndev);
2414 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2418 if (net_device_ctx->data_path_is_vf == vf_is_up)
2420 net_device_ctx->data_path_is_vf = vf_is_up;
2422 if (vf_is_up && !net_device_ctx->vf_alloc) {
2423 netdev_info(ndev, "Waiting for the VF association from host\n");
2424 wait_for_completion(&net_device_ctx->vf_add);
2427 netvsc_switch_datapath(ndev, vf_is_up);
2428 netdev_info(ndev, "Data path switched %s VF: %s\n",
2429 vf_is_up ? "to" : "from", vf_netdev->name);
2434 static int netvsc_unregister_vf(struct net_device *vf_netdev)
2436 struct net_device *ndev;
2437 struct net_device_context *net_device_ctx;
2439 ndev = get_netvsc_byref(vf_netdev);
2443 net_device_ctx = netdev_priv(ndev);
2444 cancel_delayed_work_sync(&net_device_ctx->vf_takeover);
2446 netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
2448 netvsc_vf_setxdp(vf_netdev, NULL);
2450 reinit_completion(&net_device_ctx->vf_add);
2451 netdev_rx_handler_unregister(vf_netdev);
2452 netdev_upper_dev_unlink(vf_netdev, ndev);
2453 RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
2459 static int netvsc_probe(struct hv_device *dev,
2460 const struct hv_vmbus_device_id *dev_id)
2462 struct net_device *net = NULL;
2463 struct net_device_context *net_device_ctx;
2464 struct netvsc_device_info *device_info = NULL;
2465 struct netvsc_device *nvdev;
2468 net = alloc_etherdev_mq(sizeof(struct net_device_context),
2473 netif_carrier_off(net);
2475 netvsc_init_settings(net);
2477 net_device_ctx = netdev_priv(net);
2478 net_device_ctx->device_ctx = dev;
2479 net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
2480 if (netif_msg_probe(net_device_ctx))
2481 netdev_dbg(net, "netvsc msg_enable: %d\n",
2482 net_device_ctx->msg_enable);
2484 hv_set_drvdata(dev, net);
2486 INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
2488 init_completion(&net_device_ctx->vf_add);
2489 spin_lock_init(&net_device_ctx->lock);
2490 INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
2491 INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup);
2493 net_device_ctx->vf_stats
2494 = netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats);
2495 if (!net_device_ctx->vf_stats)
2498 net->netdev_ops = &device_ops;
2499 net->ethtool_ops = ðtool_ops;
2500 SET_NETDEV_DEV(net, &dev->device);
2502 /* We always need headroom for rndis header */
2503 net->needed_headroom = RNDIS_AND_PPI_SIZE;
2505 /* Initialize the number of queues to be 1, we may change it if more
2506 * channels are offered later.
2508 netif_set_real_num_tx_queues(net, 1);
2509 netif_set_real_num_rx_queues(net, 1);
2511 /* Notify the netvsc driver of the new device */
2512 device_info = netvsc_devinfo_get(NULL);
2516 goto devinfo_failed;
2519 nvdev = rndis_filter_device_add(dev, device_info);
2520 if (IS_ERR(nvdev)) {
2521 ret = PTR_ERR(nvdev);
2522 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
2526 memcpy(net->dev_addr, device_info->mac_adr, ETH_ALEN);
2528 /* We must get rtnl lock before scheduling nvdev->subchan_work,
2529 * otherwise netvsc_subchan_work() can get rtnl lock first and wait
2530 * all subchannels to show up, but that may not happen because
2531 * netvsc_probe() can't get rtnl lock and as a result vmbus_onoffer()
2532 * -> ... -> device_add() -> ... -> __device_attach() can't get
2533 * the device lock, so all the subchannels can't be processed --
2534 * finally netvsc_subchan_work() hangs forever.
2538 if (nvdev->num_chn > 1)
2539 schedule_work(&nvdev->subchan_work);
2541 /* hw_features computed in rndis_netdev_set_hwcaps() */
2542 net->features = net->hw_features |
2543 NETIF_F_HIGHDMA | NETIF_F_HW_VLAN_CTAG_TX |
2544 NETIF_F_HW_VLAN_CTAG_RX;
2545 net->vlan_features = net->features;
2547 netdev_lockdep_set_classes(net);
2549 /* MTU range: 68 - 1500 or 65521 */
2550 net->min_mtu = NETVSC_MTU_MIN;
2551 if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
2552 net->max_mtu = NETVSC_MTU - ETH_HLEN;
2554 net->max_mtu = ETH_DATA_LEN;
2556 nvdev->tx_disable = false;
2558 ret = register_netdevice(net);
2560 pr_err("Unable to register netdev.\n");
2561 goto register_failed;
2564 list_add(&net_device_ctx->list, &netvsc_dev_list);
2567 netvsc_devinfo_put(device_info);
2572 rndis_filter_device_remove(dev, nvdev);
2574 netvsc_devinfo_put(device_info);
2576 free_percpu(net_device_ctx->vf_stats);
2578 hv_set_drvdata(dev, NULL);
2584 static int netvsc_remove(struct hv_device *dev)
2586 struct net_device_context *ndev_ctx;
2587 struct net_device *vf_netdev, *net;
2588 struct netvsc_device *nvdev;
2590 net = hv_get_drvdata(dev);
2592 dev_err(&dev->device, "No net device to remove\n");
2596 ndev_ctx = netdev_priv(net);
2598 cancel_delayed_work_sync(&ndev_ctx->dwork);
2601 nvdev = rtnl_dereference(ndev_ctx->nvdev);
2603 cancel_work_sync(&nvdev->subchan_work);
2604 netvsc_xdp_set(net, NULL, NULL, nvdev);
2608 * Call to the vsc driver to let it know that the device is being
2609 * removed. Also blocks mtu and channel changes.
2611 vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2613 netvsc_unregister_vf(vf_netdev);
2616 rndis_filter_device_remove(dev, nvdev);
2618 unregister_netdevice(net);
2619 list_del(&ndev_ctx->list);
2623 hv_set_drvdata(dev, NULL);
2625 free_percpu(ndev_ctx->vf_stats);
2630 static int netvsc_suspend(struct hv_device *dev)
2632 struct net_device_context *ndev_ctx;
2633 struct netvsc_device *nvdev;
2634 struct net_device *net;
2637 net = hv_get_drvdata(dev);
2639 ndev_ctx = netdev_priv(net);
2640 cancel_delayed_work_sync(&ndev_ctx->dwork);
2644 nvdev = rtnl_dereference(ndev_ctx->nvdev);
2645 if (nvdev == NULL) {
2650 /* Save the current config info */
2651 ndev_ctx->saved_netvsc_dev_info = netvsc_devinfo_get(nvdev);
2652 if (!ndev_ctx->saved_netvsc_dev_info) {
2656 ret = netvsc_detach(net, nvdev);
2663 static int netvsc_resume(struct hv_device *dev)
2665 struct net_device *net = hv_get_drvdata(dev);
2666 struct net_device_context *net_device_ctx;
2667 struct netvsc_device_info *device_info;
2672 net_device_ctx = netdev_priv(net);
2674 /* Reset the data path to the netvsc NIC before re-opening the vmbus
2675 * channel. Later netvsc_netdev_event() will switch the data path to
2676 * the VF upon the UP or CHANGE event.
2678 net_device_ctx->data_path_is_vf = false;
2679 device_info = net_device_ctx->saved_netvsc_dev_info;
2681 ret = netvsc_attach(net, device_info);
2683 netvsc_devinfo_put(device_info);
2684 net_device_ctx->saved_netvsc_dev_info = NULL;
2690 static const struct hv_vmbus_device_id id_table[] = {
2696 MODULE_DEVICE_TABLE(vmbus, id_table);
2698 /* The one and only one */
2699 static struct hv_driver netvsc_drv = {
2700 .name = KBUILD_MODNAME,
2701 .id_table = id_table,
2702 .probe = netvsc_probe,
2703 .remove = netvsc_remove,
2704 .suspend = netvsc_suspend,
2705 .resume = netvsc_resume,
2707 .probe_type = PROBE_FORCE_SYNCHRONOUS,
2712 * On Hyper-V, every VF interface is matched with a corresponding
2713 * synthetic interface. The synthetic interface is presented first
2714 * to the guest. When the corresponding VF instance is registered,
2715 * we will take care of switching the data path.
2717 static int netvsc_netdev_event(struct notifier_block *this,
2718 unsigned long event, void *ptr)
2720 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2722 /* Skip our own events */
2723 if (event_dev->netdev_ops == &device_ops)
2726 /* Avoid non-Ethernet type devices */
2727 if (event_dev->type != ARPHRD_ETHER)
2730 /* Avoid Vlan dev with same MAC registering as VF */
2731 if (is_vlan_dev(event_dev))
2734 /* Avoid Bonding master dev with same MAC registering as VF */
2735 if ((event_dev->priv_flags & IFF_BONDING) &&
2736 (event_dev->flags & IFF_MASTER))
2740 case NETDEV_REGISTER:
2741 return netvsc_register_vf(event_dev);
2742 case NETDEV_UNREGISTER:
2743 return netvsc_unregister_vf(event_dev);
2747 return netvsc_vf_changed(event_dev);
2753 static struct notifier_block netvsc_netdev_notifier = {
2754 .notifier_call = netvsc_netdev_event,
2757 static void __exit netvsc_drv_exit(void)
2759 unregister_netdevice_notifier(&netvsc_netdev_notifier);
2760 vmbus_driver_unregister(&netvsc_drv);
2763 static int __init netvsc_drv_init(void)
2767 if (ring_size < RING_SIZE_MIN) {
2768 ring_size = RING_SIZE_MIN;
2769 pr_info("Increased ring_size to %u (min allowed)\n",
2772 netvsc_ring_bytes = ring_size * PAGE_SIZE;
2774 ret = vmbus_driver_register(&netvsc_drv);
2778 register_netdevice_notifier(&netvsc_netdev_notifier);
2782 MODULE_LICENSE("GPL");
2783 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
2785 module_init(netvsc_drv_init);
2786 module_exit(netvsc_drv_exit);