xref: /openbmc/linux/drivers/net/hyperv/netvsc_drv.c (revision 3381df0954199458fa3993db72fb427f0ed1e43b)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (c) 2009, Microsoft Corporation.
4  *
5  * Authors:
6  *   Haiyang Zhang <haiyangz@microsoft.com>
7  *   Hank Janssen  <hjanssen@microsoft.com>
8  */
9 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10 
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>
16 #include <linux/io.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>
24 #include <linux/in.h>
25 #include <linux/slab.h>
26 #include <linux/rtnetlink.h>
27 #include <linux/netpoll.h>
28 #include <linux/bpf.h>
29 
30 #include <net/arp.h>
31 #include <net/route.h>
32 #include <net/sock.h>
33 #include <net/pkt_sched.h>
34 #include <net/checksum.h>
35 #include <net/ip6_checksum.h>
36 
37 #include "hyperv_net.h"
38 
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 */
43 
44 #define LINKCHANGE_INT (2 * HZ)
45 #define VF_TAKEOVER_INT (HZ / 10)
46 
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;
51 
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 |
55 				NETIF_MSG_TX_ERR;
56 
57 static int debug = -1;
58 module_param(debug, int, 0444);
59 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
60 
61 static LIST_HEAD(netvsc_dev_list);
62 
63 static void netvsc_change_rx_flags(struct net_device *net, int change)
64 {
65 	struct net_device_context *ndev_ctx = netdev_priv(net);
66 	struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
67 	int inc;
68 
69 	if (!vf_netdev)
70 		return;
71 
72 	if (change & IFF_PROMISC) {
73 		inc = (net->flags & IFF_PROMISC) ? 1 : -1;
74 		dev_set_promiscuity(vf_netdev, inc);
75 	}
76 
77 	if (change & IFF_ALLMULTI) {
78 		inc = (net->flags & IFF_ALLMULTI) ? 1 : -1;
79 		dev_set_allmulti(vf_netdev, inc);
80 	}
81 }
82 
83 static void netvsc_set_rx_mode(struct net_device *net)
84 {
85 	struct net_device_context *ndev_ctx = netdev_priv(net);
86 	struct net_device *vf_netdev;
87 	struct netvsc_device *nvdev;
88 
89 	rcu_read_lock();
90 	vf_netdev = rcu_dereference(ndev_ctx->vf_netdev);
91 	if (vf_netdev) {
92 		dev_uc_sync(vf_netdev, net);
93 		dev_mc_sync(vf_netdev, net);
94 	}
95 
96 	nvdev = rcu_dereference(ndev_ctx->nvdev);
97 	if (nvdev)
98 		rndis_filter_update(nvdev);
99 	rcu_read_unlock();
100 }
101 
102 static void netvsc_tx_enable(struct netvsc_device *nvscdev,
103 			     struct net_device *ndev)
104 {
105 	nvscdev->tx_disable = false;
106 	virt_wmb(); /* ensure queue wake up mechanism is on */
107 
108 	netif_tx_wake_all_queues(ndev);
109 }
110 
111 static int netvsc_open(struct net_device *net)
112 {
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;
117 	int ret = 0;
118 
119 	netif_carrier_off(net);
120 
121 	/* Open up the device */
122 	ret = rndis_filter_open(nvdev);
123 	if (ret != 0) {
124 		netdev_err(net, "unable to open device (ret %d).\n", ret);
125 		return ret;
126 	}
127 
128 	rdev = nvdev->extension;
129 	if (!rdev->link_state) {
130 		netif_carrier_on(net);
131 		netvsc_tx_enable(nvdev, net);
132 	}
133 
134 	if (vf_netdev) {
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).
138 		 */
139 		ret = dev_open(vf_netdev, NULL);
140 		if (ret)
141 			netdev_warn(net,
142 				    "unable to open slave: %s: %d\n",
143 				    vf_netdev->name, ret);
144 	}
145 	return 0;
146 }
147 
148 static int netvsc_wait_until_empty(struct netvsc_device *nvdev)
149 {
150 	unsigned int retry = 0;
151 	int i;
152 
153 	/* Ensure pending bytes in ring are read */
154 	for (;;) {
155 		u32 aread = 0;
156 
157 		for (i = 0; i < nvdev->num_chn; i++) {
158 			struct vmbus_channel *chn
159 				= nvdev->chan_table[i].channel;
160 
161 			if (!chn)
162 				continue;
163 
164 			/* make sure receive not running now */
165 			napi_synchronize(&nvdev->chan_table[i].napi);
166 
167 			aread = hv_get_bytes_to_read(&chn->inbound);
168 			if (aread)
169 				break;
170 
171 			aread = hv_get_bytes_to_read(&chn->outbound);
172 			if (aread)
173 				break;
174 		}
175 
176 		if (aread == 0)
177 			return 0;
178 
179 		if (++retry > RETRY_MAX)
180 			return -ETIMEDOUT;
181 
182 		usleep_range(RETRY_US_LO, RETRY_US_HI);
183 	}
184 }
185 
186 static void netvsc_tx_disable(struct netvsc_device *nvscdev,
187 			      struct net_device *ndev)
188 {
189 	if (nvscdev) {
190 		nvscdev->tx_disable = true;
191 		virt_wmb(); /* ensure txq will not wake up after stop */
192 	}
193 
194 	netif_tx_disable(ndev);
195 }
196 
197 static int netvsc_close(struct net_device *net)
198 {
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);
203 	int ret;
204 
205 	netvsc_tx_disable(nvdev, net);
206 
207 	/* No need to close rndis filter if it is removed already */
208 	if (!nvdev)
209 		return 0;
210 
211 	ret = rndis_filter_close(nvdev);
212 	if (ret != 0) {
213 		netdev_err(net, "unable to close device (ret %d).\n", ret);
214 		return ret;
215 	}
216 
217 	ret = netvsc_wait_until_empty(nvdev);
218 	if (ret)
219 		netdev_err(net, "Ring buffer not empty after closing rndis\n");
220 
221 	if (vf_netdev)
222 		dev_close(vf_netdev);
223 
224 	return ret;
225 }
226 
227 static inline void *init_ppi_data(struct rndis_message *msg,
228 				  u32 ppi_size, u32 pkt_type)
229 {
230 	struct rndis_packet *rndis_pkt = &msg->msg.pkt;
231 	struct rndis_per_packet_info *ppi;
232 
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;
236 
237 	ppi->size = ppi_size;
238 	ppi->type = pkt_type;
239 	ppi->internal = 0;
240 	ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
241 
242 	rndis_pkt->per_pkt_info_len += ppi_size;
243 
244 	return ppi + 1;
245 }
246 
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.
249  */
250 static inline u32 netvsc_get_hash(
251 	struct sk_buff *skb,
252 	const struct net_device_context *ndc)
253 {
254 	struct flow_keys flow;
255 	u32 hash, pkt_proto = 0;
256 	static u32 hashrnd __read_mostly;
257 
258 	net_get_random_once(&hashrnd, sizeof(hashrnd));
259 
260 	if (!skb_flow_dissect_flow_keys(skb, &flow, 0))
261 		return 0;
262 
263 	switch (flow.basic.ip_proto) {
264 	case IPPROTO_TCP:
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;
269 
270 		break;
271 
272 	case IPPROTO_UDP:
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;
277 
278 		break;
279 	}
280 
281 	if (pkt_proto & ndc->l4_hash) {
282 		return skb_get_hash(skb);
283 	} else {
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);
288 		else
289 			return 0;
290 
291 		__skb_set_sw_hash(skb, hash, false);
292 	}
293 
294 	return hash;
295 }
296 
297 static inline int netvsc_get_tx_queue(struct net_device *ndev,
298 				      struct sk_buff *skb, int old_idx)
299 {
300 	const struct net_device_context *ndc = netdev_priv(ndev);
301 	struct sock *sk = skb->sk;
302 	int q_idx;
303 
304 	q_idx = ndc->tx_table[netvsc_get_hash(skb, ndc) &
305 			      (VRSS_SEND_TAB_SIZE - 1)];
306 
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);
311 
312 	return q_idx;
313 }
314 
315 /*
316  * Select queue for transmit.
317  *
318  * If a valid queue has already been assigned, then use that.
319  * Otherwise compute tx queue based on hash and the send table.
320  *
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.
323  *
324  * TODO support XPS - but get_xps_queue not exported
325  */
326 static u16 netvsc_pick_tx(struct net_device *ndev, struct sk_buff *skb)
327 {
328 	int q_idx = sk_tx_queue_get(skb->sk);
329 
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.
333 		 */
334 		if (skb_rx_queue_recorded(skb))
335 			q_idx = skb_get_rx_queue(skb);
336 		else
337 			q_idx = netvsc_get_tx_queue(ndev, skb, q_idx);
338 	}
339 
340 	return q_idx;
341 }
342 
343 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
344 			       struct net_device *sb_dev)
345 {
346 	struct net_device_context *ndc = netdev_priv(ndev);
347 	struct net_device *vf_netdev;
348 	u16 txq;
349 
350 	rcu_read_lock();
351 	vf_netdev = rcu_dereference(ndc->vf_netdev);
352 	if (vf_netdev) {
353 		const struct net_device_ops *vf_ops = vf_netdev->netdev_ops;
354 
355 		if (vf_ops->ndo_select_queue)
356 			txq = vf_ops->ndo_select_queue(vf_netdev, skb, sb_dev);
357 		else
358 			txq = netdev_pick_tx(vf_netdev, skb, NULL);
359 
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.
363 		 */
364 		qdisc_skb_cb(skb)->slave_dev_queue_mapping = txq;
365 	} else {
366 		txq = netvsc_pick_tx(ndev, skb);
367 	}
368 	rcu_read_unlock();
369 
370 	while (unlikely(txq >= ndev->real_num_tx_queues))
371 		txq -= ndev->real_num_tx_queues;
372 
373 	return txq;
374 }
375 
376 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
377 		       struct hv_page_buffer *pb)
378 {
379 	int j = 0;
380 
381 	/* Deal with compound pages by ignoring unused part
382 	 * of the page.
383 	 */
384 	page += (offset >> PAGE_SHIFT);
385 	offset &= ~PAGE_MASK;
386 
387 	while (len > 0) {
388 		unsigned long bytes;
389 
390 		bytes = PAGE_SIZE - offset;
391 		if (bytes > len)
392 			bytes = len;
393 		pb[j].pfn = page_to_pfn(page);
394 		pb[j].offset = offset;
395 		pb[j].len = bytes;
396 
397 		offset += bytes;
398 		len -= bytes;
399 
400 		if (offset == PAGE_SIZE && len) {
401 			page++;
402 			offset = 0;
403 			j++;
404 		}
405 	}
406 
407 	return j + 1;
408 }
409 
410 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
411 			   struct hv_netvsc_packet *packet,
412 			   struct hv_page_buffer *pb)
413 {
414 	u32 slots_used = 0;
415 	char *data = skb->data;
416 	int frags = skb_shinfo(skb)->nr_frags;
417 	int i;
418 
419 	/* The packet is laid out thus:
420 	 * 1. hdr: RNDIS header and PPI
421 	 * 2. skb linear data
422 	 * 3. skb fragment data
423 	 */
424 	slots_used += fill_pg_buf(virt_to_page(hdr),
425 				  offset_in_page(hdr),
426 				  len, &pb[slots_used]);
427 
428 	packet->rmsg_size = len;
429 	packet->rmsg_pgcnt = slots_used;
430 
431 	slots_used += fill_pg_buf(virt_to_page(data),
432 				offset_in_page(data),
433 				skb_headlen(skb), &pb[slots_used]);
434 
435 	for (i = 0; i < frags; i++) {
436 		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
437 
438 		slots_used += fill_pg_buf(skb_frag_page(frag),
439 					skb_frag_off(frag),
440 					skb_frag_size(frag), &pb[slots_used]);
441 	}
442 	return slots_used;
443 }
444 
445 static int count_skb_frag_slots(struct sk_buff *skb)
446 {
447 	int i, frags = skb_shinfo(skb)->nr_frags;
448 	int pages = 0;
449 
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);
454 
455 		/* Skip unused frames from start of page */
456 		offset &= ~PAGE_MASK;
457 		pages += PFN_UP(offset + size);
458 	}
459 	return pages;
460 }
461 
462 static int netvsc_get_slots(struct sk_buff *skb)
463 {
464 	char *data = skb->data;
465 	unsigned int offset = offset_in_page(data);
466 	unsigned int len = skb_headlen(skb);
467 	int slots;
468 	int frag_slots;
469 
470 	slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
471 	frag_slots = count_skb_frag_slots(skb);
472 	return slots + frag_slots;
473 }
474 
475 static u32 net_checksum_info(struct sk_buff *skb)
476 {
477 	if (skb->protocol == htons(ETH_P_IP)) {
478 		struct iphdr *ip = ip_hdr(skb);
479 
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;
484 	} else {
485 		struct ipv6hdr *ip6 = ipv6_hdr(skb);
486 
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;
491 	}
492 
493 	return TRANSPORT_INFO_NOT_IP;
494 }
495 
496 /* Send skb on the slave VF device. */
497 static int netvsc_vf_xmit(struct net_device *net, struct net_device *vf_netdev,
498 			  struct sk_buff *skb)
499 {
500 	struct net_device_context *ndev_ctx = netdev_priv(net);
501 	unsigned int len = skb->len;
502 	int rc;
503 
504 	skb->dev = vf_netdev;
505 	skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
506 
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);
511 
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);
516 	} else {
517 		this_cpu_inc(ndev_ctx->vf_stats->tx_dropped);
518 	}
519 
520 	return rc;
521 }
522 
523 static int netvsc_xmit(struct sk_buff *skb, struct net_device *net, bool xdp_tx)
524 {
525 	struct net_device_context *net_device_ctx = netdev_priv(net);
526 	struct hv_netvsc_packet *packet = NULL;
527 	int ret;
528 	unsigned int num_data_pgs;
529 	struct rndis_message *rndis_msg;
530 	struct net_device *vf_netdev;
531 	u32 rndis_msg_size;
532 	u32 hash;
533 	struct hv_page_buffer pb[MAX_PAGE_BUFFER_COUNT];
534 
535 	/* if VF is present and up then redirect packets
536 	 * already called with rcu_read_lock_bh
537 	 */
538 	vf_netdev = rcu_dereference_bh(net_device_ctx->vf_netdev);
539 	if (vf_netdev && netif_running(vf_netdev) &&
540 	    !netpoll_tx_running(net))
541 		return netvsc_vf_xmit(net, vf_netdev, skb);
542 
543 	/* We will atmost need two pages to describe the rndis
544 	 * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
545 	 * of pages in a single packet. If skb is scattered around
546 	 * more pages we try linearizing it.
547 	 */
548 
549 	num_data_pgs = netvsc_get_slots(skb) + 2;
550 
551 	if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
552 		++net_device_ctx->eth_stats.tx_scattered;
553 
554 		if (skb_linearize(skb))
555 			goto no_memory;
556 
557 		num_data_pgs = netvsc_get_slots(skb) + 2;
558 		if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
559 			++net_device_ctx->eth_stats.tx_too_big;
560 			goto drop;
561 		}
562 	}
563 
564 	/*
565 	 * Place the rndis header in the skb head room and
566 	 * the skb->cb will be used for hv_netvsc_packet
567 	 * structure.
568 	 */
569 	ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
570 	if (ret)
571 		goto no_memory;
572 
573 	/* Use the skb control buffer for building up the packet */
574 	BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
575 			sizeof_field(struct sk_buff, cb));
576 	packet = (struct hv_netvsc_packet *)skb->cb;
577 
578 	packet->q_idx = skb_get_queue_mapping(skb);
579 
580 	packet->total_data_buflen = skb->len;
581 	packet->total_bytes = skb->len;
582 	packet->total_packets = 1;
583 
584 	rndis_msg = (struct rndis_message *)skb->head;
585 
586 	/* Add the rndis header */
587 	rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
588 	rndis_msg->msg_len = packet->total_data_buflen;
589 
590 	rndis_msg->msg.pkt = (struct rndis_packet) {
591 		.data_offset = sizeof(struct rndis_packet),
592 		.data_len = packet->total_data_buflen,
593 		.per_pkt_info_offset = sizeof(struct rndis_packet),
594 	};
595 
596 	rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
597 
598 	hash = skb_get_hash_raw(skb);
599 	if (hash != 0 && net->real_num_tx_queues > 1) {
600 		u32 *hash_info;
601 
602 		rndis_msg_size += NDIS_HASH_PPI_SIZE;
603 		hash_info = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
604 					  NBL_HASH_VALUE);
605 		*hash_info = hash;
606 	}
607 
608 	if (skb_vlan_tag_present(skb)) {
609 		struct ndis_pkt_8021q_info *vlan;
610 
611 		rndis_msg_size += NDIS_VLAN_PPI_SIZE;
612 		vlan = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
613 				     IEEE_8021Q_INFO);
614 
615 		vlan->value = 0;
616 		vlan->vlanid = skb_vlan_tag_get_id(skb);
617 		vlan->cfi = skb_vlan_tag_get_cfi(skb);
618 		vlan->pri = skb_vlan_tag_get_prio(skb);
619 	}
620 
621 	if (skb_is_gso(skb)) {
622 		struct ndis_tcp_lso_info *lso_info;
623 
624 		rndis_msg_size += NDIS_LSO_PPI_SIZE;
625 		lso_info = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
626 					 TCP_LARGESEND_PKTINFO);
627 
628 		lso_info->value = 0;
629 		lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
630 		if (skb->protocol == htons(ETH_P_IP)) {
631 			lso_info->lso_v2_transmit.ip_version =
632 				NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
633 			ip_hdr(skb)->tot_len = 0;
634 			ip_hdr(skb)->check = 0;
635 			tcp_hdr(skb)->check =
636 				~csum_tcpudp_magic(ip_hdr(skb)->saddr,
637 						   ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
638 		} else {
639 			lso_info->lso_v2_transmit.ip_version =
640 				NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
641 			tcp_v6_gso_csum_prep(skb);
642 		}
643 		lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
644 		lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
645 	} else if (skb->ip_summed == CHECKSUM_PARTIAL) {
646 		if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
647 			struct ndis_tcp_ip_checksum_info *csum_info;
648 
649 			rndis_msg_size += NDIS_CSUM_PPI_SIZE;
650 			csum_info = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
651 						  TCPIP_CHKSUM_PKTINFO);
652 
653 			csum_info->value = 0;
654 			csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
655 
656 			if (skb->protocol == htons(ETH_P_IP)) {
657 				csum_info->transmit.is_ipv4 = 1;
658 
659 				if (ip_hdr(skb)->protocol == IPPROTO_TCP)
660 					csum_info->transmit.tcp_checksum = 1;
661 				else
662 					csum_info->transmit.udp_checksum = 1;
663 			} else {
664 				csum_info->transmit.is_ipv6 = 1;
665 
666 				if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
667 					csum_info->transmit.tcp_checksum = 1;
668 				else
669 					csum_info->transmit.udp_checksum = 1;
670 			}
671 		} else {
672 			/* Can't do offload of this type of checksum */
673 			if (skb_checksum_help(skb))
674 				goto drop;
675 		}
676 	}
677 
678 	/* Start filling in the page buffers with the rndis hdr */
679 	rndis_msg->msg_len += rndis_msg_size;
680 	packet->total_data_buflen = rndis_msg->msg_len;
681 	packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
682 					       skb, packet, pb);
683 
684 	/* timestamp packet in software */
685 	skb_tx_timestamp(skb);
686 
687 	ret = netvsc_send(net, packet, rndis_msg, pb, skb, xdp_tx);
688 	if (likely(ret == 0))
689 		return NETDEV_TX_OK;
690 
691 	if (ret == -EAGAIN) {
692 		++net_device_ctx->eth_stats.tx_busy;
693 		return NETDEV_TX_BUSY;
694 	}
695 
696 	if (ret == -ENOSPC)
697 		++net_device_ctx->eth_stats.tx_no_space;
698 
699 drop:
700 	dev_kfree_skb_any(skb);
701 	net->stats.tx_dropped++;
702 
703 	return NETDEV_TX_OK;
704 
705 no_memory:
706 	++net_device_ctx->eth_stats.tx_no_memory;
707 	goto drop;
708 }
709 
710 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *ndev)
711 {
712 	return netvsc_xmit(skb, ndev, false);
713 }
714 
715 /*
716  * netvsc_linkstatus_callback - Link up/down notification
717  */
718 void netvsc_linkstatus_callback(struct net_device *net,
719 				struct rndis_message *resp)
720 {
721 	struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
722 	struct net_device_context *ndev_ctx = netdev_priv(net);
723 	struct netvsc_reconfig *event;
724 	unsigned long flags;
725 
726 	/* Update the physical link speed when changing to another vSwitch */
727 	if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
728 		u32 speed;
729 
730 		speed = *(u32 *)((void *)indicate
731 				 + indicate->status_buf_offset) / 10000;
732 		ndev_ctx->speed = speed;
733 		return;
734 	}
735 
736 	/* Handle these link change statuses below */
737 	if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
738 	    indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
739 	    indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
740 		return;
741 
742 	if (net->reg_state != NETREG_REGISTERED)
743 		return;
744 
745 	event = kzalloc(sizeof(*event), GFP_ATOMIC);
746 	if (!event)
747 		return;
748 	event->event = indicate->status;
749 
750 	spin_lock_irqsave(&ndev_ctx->lock, flags);
751 	list_add_tail(&event->list, &ndev_ctx->reconfig_events);
752 	spin_unlock_irqrestore(&ndev_ctx->lock, flags);
753 
754 	schedule_delayed_work(&ndev_ctx->dwork, 0);
755 }
756 
757 static void netvsc_xdp_xmit(struct sk_buff *skb, struct net_device *ndev)
758 {
759 	int rc;
760 
761 	skb->queue_mapping = skb_get_rx_queue(skb);
762 	__skb_push(skb, ETH_HLEN);
763 
764 	rc = netvsc_xmit(skb, ndev, true);
765 
766 	if (dev_xmit_complete(rc))
767 		return;
768 
769 	dev_kfree_skb_any(skb);
770 	ndev->stats.tx_dropped++;
771 }
772 
773 static void netvsc_comp_ipcsum(struct sk_buff *skb)
774 {
775 	struct iphdr *iph = (struct iphdr *)skb->data;
776 
777 	iph->check = 0;
778 	iph->check = ip_fast_csum(iph, iph->ihl);
779 }
780 
781 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
782 					     struct netvsc_channel *nvchan,
783 					     struct xdp_buff *xdp)
784 {
785 	struct napi_struct *napi = &nvchan->napi;
786 	const struct ndis_pkt_8021q_info *vlan = nvchan->rsc.vlan;
787 	const struct ndis_tcp_ip_checksum_info *csum_info =
788 						nvchan->rsc.csum_info;
789 	const u32 *hash_info = nvchan->rsc.hash_info;
790 	struct sk_buff *skb;
791 	void *xbuf = xdp->data_hard_start;
792 	int i;
793 
794 	if (xbuf) {
795 		unsigned int hdroom = xdp->data - xdp->data_hard_start;
796 		unsigned int xlen = xdp->data_end - xdp->data;
797 		unsigned int frag_size = netvsc_xdp_fraglen(hdroom + xlen);
798 
799 		skb = build_skb(xbuf, frag_size);
800 
801 		if (!skb) {
802 			__free_page(virt_to_page(xbuf));
803 			return NULL;
804 		}
805 
806 		skb_reserve(skb, hdroom);
807 		skb_put(skb, xlen);
808 		skb->dev = napi->dev;
809 	} else {
810 		skb = napi_alloc_skb(napi, nvchan->rsc.pktlen);
811 
812 		if (!skb)
813 			return NULL;
814 
815 		/* Copy to skb. This copy is needed here since the memory
816 		 * pointed by hv_netvsc_packet cannot be deallocated.
817 		 */
818 		for (i = 0; i < nvchan->rsc.cnt; i++)
819 			skb_put_data(skb, nvchan->rsc.data[i],
820 				     nvchan->rsc.len[i]);
821 	}
822 
823 	skb->protocol = eth_type_trans(skb, net);
824 
825 	/* skb is already created with CHECKSUM_NONE */
826 	skb_checksum_none_assert(skb);
827 
828 	/* Incoming packets may have IP header checksum verified by the host.
829 	 * They may not have IP header checksum computed after coalescing.
830 	 * We compute it here if the flags are set, because on Linux, the IP
831 	 * checksum is always checked.
832 	 */
833 	if (csum_info && csum_info->receive.ip_checksum_value_invalid &&
834 	    csum_info->receive.ip_checksum_succeeded &&
835 	    skb->protocol == htons(ETH_P_IP))
836 		netvsc_comp_ipcsum(skb);
837 
838 	/* Do L4 checksum offload if enabled and present. */
839 	if (csum_info && (net->features & NETIF_F_RXCSUM)) {
840 		if (csum_info->receive.tcp_checksum_succeeded ||
841 		    csum_info->receive.udp_checksum_succeeded)
842 			skb->ip_summed = CHECKSUM_UNNECESSARY;
843 	}
844 
845 	if (hash_info && (net->features & NETIF_F_RXHASH))
846 		skb_set_hash(skb, *hash_info, PKT_HASH_TYPE_L4);
847 
848 	if (vlan) {
849 		u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT) |
850 			(vlan->cfi ? VLAN_CFI_MASK : 0);
851 
852 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
853 				       vlan_tci);
854 	}
855 
856 	return skb;
857 }
858 
859 /*
860  * netvsc_recv_callback -  Callback when we receive a packet from the
861  * "wire" on the specified device.
862  */
863 int netvsc_recv_callback(struct net_device *net,
864 			 struct netvsc_device *net_device,
865 			 struct netvsc_channel *nvchan)
866 {
867 	struct net_device_context *net_device_ctx = netdev_priv(net);
868 	struct vmbus_channel *channel = nvchan->channel;
869 	u16 q_idx = channel->offermsg.offer.sub_channel_index;
870 	struct sk_buff *skb;
871 	struct netvsc_stats *rx_stats = &nvchan->rx_stats;
872 	struct xdp_buff xdp;
873 	u32 act;
874 
875 	if (net->reg_state != NETREG_REGISTERED)
876 		return NVSP_STAT_FAIL;
877 
878 	act = netvsc_run_xdp(net, nvchan, &xdp);
879 
880 	if (act != XDP_PASS && act != XDP_TX) {
881 		u64_stats_update_begin(&rx_stats->syncp);
882 		rx_stats->xdp_drop++;
883 		u64_stats_update_end(&rx_stats->syncp);
884 
885 		return NVSP_STAT_SUCCESS; /* consumed by XDP */
886 	}
887 
888 	/* Allocate a skb - TODO direct I/O to pages? */
889 	skb = netvsc_alloc_recv_skb(net, nvchan, &xdp);
890 
891 	if (unlikely(!skb)) {
892 		++net_device_ctx->eth_stats.rx_no_memory;
893 		return NVSP_STAT_FAIL;
894 	}
895 
896 	skb_record_rx_queue(skb, q_idx);
897 
898 	/*
899 	 * Even if injecting the packet, record the statistics
900 	 * on the synthetic device because modifying the VF device
901 	 * statistics will not work correctly.
902 	 */
903 	u64_stats_update_begin(&rx_stats->syncp);
904 	rx_stats->packets++;
905 	rx_stats->bytes += nvchan->rsc.pktlen;
906 
907 	if (skb->pkt_type == PACKET_BROADCAST)
908 		++rx_stats->broadcast;
909 	else if (skb->pkt_type == PACKET_MULTICAST)
910 		++rx_stats->multicast;
911 	u64_stats_update_end(&rx_stats->syncp);
912 
913 	if (act == XDP_TX) {
914 		netvsc_xdp_xmit(skb, net);
915 		return NVSP_STAT_SUCCESS;
916 	}
917 
918 	napi_gro_receive(&nvchan->napi, skb);
919 	return NVSP_STAT_SUCCESS;
920 }
921 
922 static void netvsc_get_drvinfo(struct net_device *net,
923 			       struct ethtool_drvinfo *info)
924 {
925 	strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
926 	strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
927 }
928 
929 static void netvsc_get_channels(struct net_device *net,
930 				struct ethtool_channels *channel)
931 {
932 	struct net_device_context *net_device_ctx = netdev_priv(net);
933 	struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
934 
935 	if (nvdev) {
936 		channel->max_combined	= nvdev->max_chn;
937 		channel->combined_count = nvdev->num_chn;
938 	}
939 }
940 
941 /* Alloc struct netvsc_device_info, and initialize it from either existing
942  * struct netvsc_device, or from default values.
943  */
944 static
945 struct netvsc_device_info *netvsc_devinfo_get(struct netvsc_device *nvdev)
946 {
947 	struct netvsc_device_info *dev_info;
948 	struct bpf_prog *prog;
949 
950 	dev_info = kzalloc(sizeof(*dev_info), GFP_ATOMIC);
951 
952 	if (!dev_info)
953 		return NULL;
954 
955 	if (nvdev) {
956 		ASSERT_RTNL();
957 
958 		dev_info->num_chn = nvdev->num_chn;
959 		dev_info->send_sections = nvdev->send_section_cnt;
960 		dev_info->send_section_size = nvdev->send_section_size;
961 		dev_info->recv_sections = nvdev->recv_section_cnt;
962 		dev_info->recv_section_size = nvdev->recv_section_size;
963 
964 		memcpy(dev_info->rss_key, nvdev->extension->rss_key,
965 		       NETVSC_HASH_KEYLEN);
966 
967 		prog = netvsc_xdp_get(nvdev);
968 		if (prog) {
969 			bpf_prog_inc(prog);
970 			dev_info->bprog = prog;
971 		}
972 	} else {
973 		dev_info->num_chn = VRSS_CHANNEL_DEFAULT;
974 		dev_info->send_sections = NETVSC_DEFAULT_TX;
975 		dev_info->send_section_size = NETVSC_SEND_SECTION_SIZE;
976 		dev_info->recv_sections = NETVSC_DEFAULT_RX;
977 		dev_info->recv_section_size = NETVSC_RECV_SECTION_SIZE;
978 	}
979 
980 	return dev_info;
981 }
982 
983 /* Free struct netvsc_device_info */
984 static void netvsc_devinfo_put(struct netvsc_device_info *dev_info)
985 {
986 	if (dev_info->bprog) {
987 		ASSERT_RTNL();
988 		bpf_prog_put(dev_info->bprog);
989 	}
990 
991 	kfree(dev_info);
992 }
993 
994 static int netvsc_detach(struct net_device *ndev,
995 			 struct netvsc_device *nvdev)
996 {
997 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
998 	struct hv_device *hdev = ndev_ctx->device_ctx;
999 	int ret;
1000 
1001 	/* Don't try continuing to try and setup sub channels */
1002 	if (cancel_work_sync(&nvdev->subchan_work))
1003 		nvdev->num_chn = 1;
1004 
1005 	netvsc_xdp_set(ndev, NULL, NULL, nvdev);
1006 
1007 	/* If device was up (receiving) then shutdown */
1008 	if (netif_running(ndev)) {
1009 		netvsc_tx_disable(nvdev, ndev);
1010 
1011 		ret = rndis_filter_close(nvdev);
1012 		if (ret) {
1013 			netdev_err(ndev,
1014 				   "unable to close device (ret %d).\n", ret);
1015 			return ret;
1016 		}
1017 
1018 		ret = netvsc_wait_until_empty(nvdev);
1019 		if (ret) {
1020 			netdev_err(ndev,
1021 				   "Ring buffer not empty after closing rndis\n");
1022 			return ret;
1023 		}
1024 	}
1025 
1026 	netif_device_detach(ndev);
1027 
1028 	rndis_filter_device_remove(hdev, nvdev);
1029 
1030 	return 0;
1031 }
1032 
1033 static int netvsc_attach(struct net_device *ndev,
1034 			 struct netvsc_device_info *dev_info)
1035 {
1036 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
1037 	struct hv_device *hdev = ndev_ctx->device_ctx;
1038 	struct netvsc_device *nvdev;
1039 	struct rndis_device *rdev;
1040 	struct bpf_prog *prog;
1041 	int ret = 0;
1042 
1043 	nvdev = rndis_filter_device_add(hdev, dev_info);
1044 	if (IS_ERR(nvdev))
1045 		return PTR_ERR(nvdev);
1046 
1047 	if (nvdev->num_chn > 1) {
1048 		ret = rndis_set_subchannel(ndev, nvdev, dev_info);
1049 
1050 		/* if unavailable, just proceed with one queue */
1051 		if (ret) {
1052 			nvdev->max_chn = 1;
1053 			nvdev->num_chn = 1;
1054 		}
1055 	}
1056 
1057 	prog = dev_info->bprog;
1058 	if (prog) {
1059 		bpf_prog_inc(prog);
1060 		ret = netvsc_xdp_set(ndev, prog, NULL, nvdev);
1061 		if (ret) {
1062 			bpf_prog_put(prog);
1063 			goto err1;
1064 		}
1065 	}
1066 
1067 	/* In any case device is now ready */
1068 	nvdev->tx_disable = false;
1069 	netif_device_attach(ndev);
1070 
1071 	/* Note: enable and attach happen when sub-channels setup */
1072 	netif_carrier_off(ndev);
1073 
1074 	if (netif_running(ndev)) {
1075 		ret = rndis_filter_open(nvdev);
1076 		if (ret)
1077 			goto err2;
1078 
1079 		rdev = nvdev->extension;
1080 		if (!rdev->link_state)
1081 			netif_carrier_on(ndev);
1082 	}
1083 
1084 	return 0;
1085 
1086 err2:
1087 	netif_device_detach(ndev);
1088 
1089 err1:
1090 	rndis_filter_device_remove(hdev, nvdev);
1091 
1092 	return ret;
1093 }
1094 
1095 static int netvsc_set_channels(struct net_device *net,
1096 			       struct ethtool_channels *channels)
1097 {
1098 	struct net_device_context *net_device_ctx = netdev_priv(net);
1099 	struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
1100 	unsigned int orig, count = channels->combined_count;
1101 	struct netvsc_device_info *device_info;
1102 	int ret;
1103 
1104 	/* We do not support separate count for rx, tx, or other */
1105 	if (count == 0 ||
1106 	    channels->rx_count || channels->tx_count || channels->other_count)
1107 		return -EINVAL;
1108 
1109 	if (!nvdev || nvdev->destroy)
1110 		return -ENODEV;
1111 
1112 	if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
1113 		return -EINVAL;
1114 
1115 	if (count > nvdev->max_chn)
1116 		return -EINVAL;
1117 
1118 	orig = nvdev->num_chn;
1119 
1120 	device_info = netvsc_devinfo_get(nvdev);
1121 
1122 	if (!device_info)
1123 		return -ENOMEM;
1124 
1125 	device_info->num_chn = count;
1126 
1127 	ret = netvsc_detach(net, nvdev);
1128 	if (ret)
1129 		goto out;
1130 
1131 	ret = netvsc_attach(net, device_info);
1132 	if (ret) {
1133 		device_info->num_chn = orig;
1134 		if (netvsc_attach(net, device_info))
1135 			netdev_err(net, "restoring channel setting failed\n");
1136 	}
1137 
1138 out:
1139 	netvsc_devinfo_put(device_info);
1140 	return ret;
1141 }
1142 
1143 static void netvsc_init_settings(struct net_device *dev)
1144 {
1145 	struct net_device_context *ndc = netdev_priv(dev);
1146 
1147 	ndc->l4_hash = HV_DEFAULT_L4HASH;
1148 
1149 	ndc->speed = SPEED_UNKNOWN;
1150 	ndc->duplex = DUPLEX_FULL;
1151 
1152 	dev->features = NETIF_F_LRO;
1153 }
1154 
1155 static int netvsc_get_link_ksettings(struct net_device *dev,
1156 				     struct ethtool_link_ksettings *cmd)
1157 {
1158 	struct net_device_context *ndc = netdev_priv(dev);
1159 	struct net_device *vf_netdev;
1160 
1161 	vf_netdev = rtnl_dereference(ndc->vf_netdev);
1162 
1163 	if (vf_netdev)
1164 		return __ethtool_get_link_ksettings(vf_netdev, cmd);
1165 
1166 	cmd->base.speed = ndc->speed;
1167 	cmd->base.duplex = ndc->duplex;
1168 	cmd->base.port = PORT_OTHER;
1169 
1170 	return 0;
1171 }
1172 
1173 static int netvsc_set_link_ksettings(struct net_device *dev,
1174 				     const struct ethtool_link_ksettings *cmd)
1175 {
1176 	struct net_device_context *ndc = netdev_priv(dev);
1177 	struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1178 
1179 	if (vf_netdev) {
1180 		if (!vf_netdev->ethtool_ops->set_link_ksettings)
1181 			return -EOPNOTSUPP;
1182 
1183 		return vf_netdev->ethtool_ops->set_link_ksettings(vf_netdev,
1184 								  cmd);
1185 	}
1186 
1187 	return ethtool_virtdev_set_link_ksettings(dev, cmd,
1188 						  &ndc->speed, &ndc->duplex);
1189 }
1190 
1191 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
1192 {
1193 	struct net_device_context *ndevctx = netdev_priv(ndev);
1194 	struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1195 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1196 	int orig_mtu = ndev->mtu;
1197 	struct netvsc_device_info *device_info;
1198 	int ret = 0;
1199 
1200 	if (!nvdev || nvdev->destroy)
1201 		return -ENODEV;
1202 
1203 	device_info = netvsc_devinfo_get(nvdev);
1204 
1205 	if (!device_info)
1206 		return -ENOMEM;
1207 
1208 	/* Change MTU of underlying VF netdev first. */
1209 	if (vf_netdev) {
1210 		ret = dev_set_mtu(vf_netdev, mtu);
1211 		if (ret)
1212 			goto out;
1213 	}
1214 
1215 	ret = netvsc_detach(ndev, nvdev);
1216 	if (ret)
1217 		goto rollback_vf;
1218 
1219 	ndev->mtu = mtu;
1220 
1221 	ret = netvsc_attach(ndev, device_info);
1222 	if (!ret)
1223 		goto out;
1224 
1225 	/* Attempt rollback to original MTU */
1226 	ndev->mtu = orig_mtu;
1227 
1228 	if (netvsc_attach(ndev, device_info))
1229 		netdev_err(ndev, "restoring mtu failed\n");
1230 rollback_vf:
1231 	if (vf_netdev)
1232 		dev_set_mtu(vf_netdev, orig_mtu);
1233 
1234 out:
1235 	netvsc_devinfo_put(device_info);
1236 	return ret;
1237 }
1238 
1239 static void netvsc_get_vf_stats(struct net_device *net,
1240 				struct netvsc_vf_pcpu_stats *tot)
1241 {
1242 	struct net_device_context *ndev_ctx = netdev_priv(net);
1243 	int i;
1244 
1245 	memset(tot, 0, sizeof(*tot));
1246 
1247 	for_each_possible_cpu(i) {
1248 		const struct netvsc_vf_pcpu_stats *stats
1249 			= per_cpu_ptr(ndev_ctx->vf_stats, i);
1250 		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1251 		unsigned int start;
1252 
1253 		do {
1254 			start = u64_stats_fetch_begin_irq(&stats->syncp);
1255 			rx_packets = stats->rx_packets;
1256 			tx_packets = stats->tx_packets;
1257 			rx_bytes = stats->rx_bytes;
1258 			tx_bytes = stats->tx_bytes;
1259 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1260 
1261 		tot->rx_packets += rx_packets;
1262 		tot->tx_packets += tx_packets;
1263 		tot->rx_bytes   += rx_bytes;
1264 		tot->tx_bytes   += tx_bytes;
1265 		tot->tx_dropped += stats->tx_dropped;
1266 	}
1267 }
1268 
1269 static void netvsc_get_pcpu_stats(struct net_device *net,
1270 				  struct netvsc_ethtool_pcpu_stats *pcpu_tot)
1271 {
1272 	struct net_device_context *ndev_ctx = netdev_priv(net);
1273 	struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1274 	int i;
1275 
1276 	/* fetch percpu stats of vf */
1277 	for_each_possible_cpu(i) {
1278 		const struct netvsc_vf_pcpu_stats *stats =
1279 			per_cpu_ptr(ndev_ctx->vf_stats, i);
1280 		struct netvsc_ethtool_pcpu_stats *this_tot = &pcpu_tot[i];
1281 		unsigned int start;
1282 
1283 		do {
1284 			start = u64_stats_fetch_begin_irq(&stats->syncp);
1285 			this_tot->vf_rx_packets = stats->rx_packets;
1286 			this_tot->vf_tx_packets = stats->tx_packets;
1287 			this_tot->vf_rx_bytes = stats->rx_bytes;
1288 			this_tot->vf_tx_bytes = stats->tx_bytes;
1289 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1290 		this_tot->rx_packets = this_tot->vf_rx_packets;
1291 		this_tot->tx_packets = this_tot->vf_tx_packets;
1292 		this_tot->rx_bytes   = this_tot->vf_rx_bytes;
1293 		this_tot->tx_bytes   = this_tot->vf_tx_bytes;
1294 	}
1295 
1296 	/* fetch percpu stats of netvsc */
1297 	for (i = 0; i < nvdev->num_chn; i++) {
1298 		const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1299 		const struct netvsc_stats *stats;
1300 		struct netvsc_ethtool_pcpu_stats *this_tot =
1301 			&pcpu_tot[nvchan->channel->target_cpu];
1302 		u64 packets, bytes;
1303 		unsigned int start;
1304 
1305 		stats = &nvchan->tx_stats;
1306 		do {
1307 			start = u64_stats_fetch_begin_irq(&stats->syncp);
1308 			packets = stats->packets;
1309 			bytes = stats->bytes;
1310 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1311 
1312 		this_tot->tx_bytes	+= bytes;
1313 		this_tot->tx_packets	+= packets;
1314 
1315 		stats = &nvchan->rx_stats;
1316 		do {
1317 			start = u64_stats_fetch_begin_irq(&stats->syncp);
1318 			packets = stats->packets;
1319 			bytes = stats->bytes;
1320 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1321 
1322 		this_tot->rx_bytes	+= bytes;
1323 		this_tot->rx_packets	+= packets;
1324 	}
1325 }
1326 
1327 static void netvsc_get_stats64(struct net_device *net,
1328 			       struct rtnl_link_stats64 *t)
1329 {
1330 	struct net_device_context *ndev_ctx = netdev_priv(net);
1331 	struct netvsc_device *nvdev;
1332 	struct netvsc_vf_pcpu_stats vf_tot;
1333 	int i;
1334 
1335 	rcu_read_lock();
1336 
1337 	nvdev = rcu_dereference(ndev_ctx->nvdev);
1338 	if (!nvdev)
1339 		goto out;
1340 
1341 	netdev_stats_to_stats64(t, &net->stats);
1342 
1343 	netvsc_get_vf_stats(net, &vf_tot);
1344 	t->rx_packets += vf_tot.rx_packets;
1345 	t->tx_packets += vf_tot.tx_packets;
1346 	t->rx_bytes   += vf_tot.rx_bytes;
1347 	t->tx_bytes   += vf_tot.tx_bytes;
1348 	t->tx_dropped += vf_tot.tx_dropped;
1349 
1350 	for (i = 0; i < nvdev->num_chn; i++) {
1351 		const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1352 		const struct netvsc_stats *stats;
1353 		u64 packets, bytes, multicast;
1354 		unsigned int start;
1355 
1356 		stats = &nvchan->tx_stats;
1357 		do {
1358 			start = u64_stats_fetch_begin_irq(&stats->syncp);
1359 			packets = stats->packets;
1360 			bytes = stats->bytes;
1361 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1362 
1363 		t->tx_bytes	+= bytes;
1364 		t->tx_packets	+= packets;
1365 
1366 		stats = &nvchan->rx_stats;
1367 		do {
1368 			start = u64_stats_fetch_begin_irq(&stats->syncp);
1369 			packets = stats->packets;
1370 			bytes = stats->bytes;
1371 			multicast = stats->multicast + stats->broadcast;
1372 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1373 
1374 		t->rx_bytes	+= bytes;
1375 		t->rx_packets	+= packets;
1376 		t->multicast	+= multicast;
1377 	}
1378 out:
1379 	rcu_read_unlock();
1380 }
1381 
1382 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1383 {
1384 	struct net_device_context *ndc = netdev_priv(ndev);
1385 	struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1386 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1387 	struct sockaddr *addr = p;
1388 	int err;
1389 
1390 	err = eth_prepare_mac_addr_change(ndev, p);
1391 	if (err)
1392 		return err;
1393 
1394 	if (!nvdev)
1395 		return -ENODEV;
1396 
1397 	if (vf_netdev) {
1398 		err = dev_set_mac_address(vf_netdev, addr, NULL);
1399 		if (err)
1400 			return err;
1401 	}
1402 
1403 	err = rndis_filter_set_device_mac(nvdev, addr->sa_data);
1404 	if (!err) {
1405 		eth_commit_mac_addr_change(ndev, p);
1406 	} else if (vf_netdev) {
1407 		/* rollback change on VF */
1408 		memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN);
1409 		dev_set_mac_address(vf_netdev, addr, NULL);
1410 	}
1411 
1412 	return err;
1413 }
1414 
1415 static const struct {
1416 	char name[ETH_GSTRING_LEN];
1417 	u16 offset;
1418 } netvsc_stats[] = {
1419 	{ "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1420 	{ "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1421 	{ "tx_no_space",  offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1422 	{ "tx_too_big",	  offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1423 	{ "tx_busy",	  offsetof(struct netvsc_ethtool_stats, tx_busy) },
1424 	{ "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) },
1425 	{ "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) },
1426 	{ "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) },
1427 	{ "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) },
1428 	{ "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) },
1429 }, pcpu_stats[] = {
1430 	{ "cpu%u_rx_packets",
1431 		offsetof(struct netvsc_ethtool_pcpu_stats, rx_packets) },
1432 	{ "cpu%u_rx_bytes",
1433 		offsetof(struct netvsc_ethtool_pcpu_stats, rx_bytes) },
1434 	{ "cpu%u_tx_packets",
1435 		offsetof(struct netvsc_ethtool_pcpu_stats, tx_packets) },
1436 	{ "cpu%u_tx_bytes",
1437 		offsetof(struct netvsc_ethtool_pcpu_stats, tx_bytes) },
1438 	{ "cpu%u_vf_rx_packets",
1439 		offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_packets) },
1440 	{ "cpu%u_vf_rx_bytes",
1441 		offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_bytes) },
1442 	{ "cpu%u_vf_tx_packets",
1443 		offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_packets) },
1444 	{ "cpu%u_vf_tx_bytes",
1445 		offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_bytes) },
1446 }, vf_stats[] = {
1447 	{ "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) },
1448 	{ "vf_rx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) },
1449 	{ "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) },
1450 	{ "vf_tx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) },
1451 	{ "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) },
1452 };
1453 
1454 #define NETVSC_GLOBAL_STATS_LEN	ARRAY_SIZE(netvsc_stats)
1455 #define NETVSC_VF_STATS_LEN	ARRAY_SIZE(vf_stats)
1456 
1457 /* statistics per queue (rx/tx packets/bytes) */
1458 #define NETVSC_PCPU_STATS_LEN (num_present_cpus() * ARRAY_SIZE(pcpu_stats))
1459 
1460 /* 5 statistics per queue (rx/tx packets/bytes, rx xdp_drop) */
1461 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 5)
1462 
1463 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1464 {
1465 	struct net_device_context *ndc = netdev_priv(dev);
1466 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1467 
1468 	if (!nvdev)
1469 		return -ENODEV;
1470 
1471 	switch (string_set) {
1472 	case ETH_SS_STATS:
1473 		return NETVSC_GLOBAL_STATS_LEN
1474 			+ NETVSC_VF_STATS_LEN
1475 			+ NETVSC_QUEUE_STATS_LEN(nvdev)
1476 			+ NETVSC_PCPU_STATS_LEN;
1477 	default:
1478 		return -EINVAL;
1479 	}
1480 }
1481 
1482 static void netvsc_get_ethtool_stats(struct net_device *dev,
1483 				     struct ethtool_stats *stats, u64 *data)
1484 {
1485 	struct net_device_context *ndc = netdev_priv(dev);
1486 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1487 	const void *nds = &ndc->eth_stats;
1488 	const struct netvsc_stats *qstats;
1489 	struct netvsc_vf_pcpu_stats sum;
1490 	struct netvsc_ethtool_pcpu_stats *pcpu_sum;
1491 	unsigned int start;
1492 	u64 packets, bytes;
1493 	u64 xdp_drop;
1494 	int i, j, cpu;
1495 
1496 	if (!nvdev)
1497 		return;
1498 
1499 	for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1500 		data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1501 
1502 	netvsc_get_vf_stats(dev, &sum);
1503 	for (j = 0; j < NETVSC_VF_STATS_LEN; j++)
1504 		data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset);
1505 
1506 	for (j = 0; j < nvdev->num_chn; j++) {
1507 		qstats = &nvdev->chan_table[j].tx_stats;
1508 
1509 		do {
1510 			start = u64_stats_fetch_begin_irq(&qstats->syncp);
1511 			packets = qstats->packets;
1512 			bytes = qstats->bytes;
1513 		} while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1514 		data[i++] = packets;
1515 		data[i++] = bytes;
1516 
1517 		qstats = &nvdev->chan_table[j].rx_stats;
1518 		do {
1519 			start = u64_stats_fetch_begin_irq(&qstats->syncp);
1520 			packets = qstats->packets;
1521 			bytes = qstats->bytes;
1522 			xdp_drop = qstats->xdp_drop;
1523 		} while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1524 		data[i++] = packets;
1525 		data[i++] = bytes;
1526 		data[i++] = xdp_drop;
1527 	}
1528 
1529 	pcpu_sum = kvmalloc_array(num_possible_cpus(),
1530 				  sizeof(struct netvsc_ethtool_pcpu_stats),
1531 				  GFP_KERNEL);
1532 	netvsc_get_pcpu_stats(dev, pcpu_sum);
1533 	for_each_present_cpu(cpu) {
1534 		struct netvsc_ethtool_pcpu_stats *this_sum = &pcpu_sum[cpu];
1535 
1536 		for (j = 0; j < ARRAY_SIZE(pcpu_stats); j++)
1537 			data[i++] = *(u64 *)((void *)this_sum
1538 					     + pcpu_stats[j].offset);
1539 	}
1540 	kvfree(pcpu_sum);
1541 }
1542 
1543 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1544 {
1545 	struct net_device_context *ndc = netdev_priv(dev);
1546 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1547 	u8 *p = data;
1548 	int i, cpu;
1549 
1550 	if (!nvdev)
1551 		return;
1552 
1553 	switch (stringset) {
1554 	case ETH_SS_STATS:
1555 		for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++) {
1556 			memcpy(p, netvsc_stats[i].name, ETH_GSTRING_LEN);
1557 			p += ETH_GSTRING_LEN;
1558 		}
1559 
1560 		for (i = 0; i < ARRAY_SIZE(vf_stats); i++) {
1561 			memcpy(p, vf_stats[i].name, ETH_GSTRING_LEN);
1562 			p += ETH_GSTRING_LEN;
1563 		}
1564 
1565 		for (i = 0; i < nvdev->num_chn; i++) {
1566 			sprintf(p, "tx_queue_%u_packets", i);
1567 			p += ETH_GSTRING_LEN;
1568 			sprintf(p, "tx_queue_%u_bytes", i);
1569 			p += ETH_GSTRING_LEN;
1570 			sprintf(p, "rx_queue_%u_packets", i);
1571 			p += ETH_GSTRING_LEN;
1572 			sprintf(p, "rx_queue_%u_bytes", i);
1573 			p += ETH_GSTRING_LEN;
1574 			sprintf(p, "rx_queue_%u_xdp_drop", i);
1575 			p += ETH_GSTRING_LEN;
1576 		}
1577 
1578 		for_each_present_cpu(cpu) {
1579 			for (i = 0; i < ARRAY_SIZE(pcpu_stats); i++) {
1580 				sprintf(p, pcpu_stats[i].name, cpu);
1581 				p += ETH_GSTRING_LEN;
1582 			}
1583 		}
1584 
1585 		break;
1586 	}
1587 }
1588 
1589 static int
1590 netvsc_get_rss_hash_opts(struct net_device_context *ndc,
1591 			 struct ethtool_rxnfc *info)
1592 {
1593 	const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3;
1594 
1595 	info->data = RXH_IP_SRC | RXH_IP_DST;
1596 
1597 	switch (info->flow_type) {
1598 	case TCP_V4_FLOW:
1599 		if (ndc->l4_hash & HV_TCP4_L4HASH)
1600 			info->data |= l4_flag;
1601 
1602 		break;
1603 
1604 	case TCP_V6_FLOW:
1605 		if (ndc->l4_hash & HV_TCP6_L4HASH)
1606 			info->data |= l4_flag;
1607 
1608 		break;
1609 
1610 	case UDP_V4_FLOW:
1611 		if (ndc->l4_hash & HV_UDP4_L4HASH)
1612 			info->data |= l4_flag;
1613 
1614 		break;
1615 
1616 	case UDP_V6_FLOW:
1617 		if (ndc->l4_hash & HV_UDP6_L4HASH)
1618 			info->data |= l4_flag;
1619 
1620 		break;
1621 
1622 	case IPV4_FLOW:
1623 	case IPV6_FLOW:
1624 		break;
1625 	default:
1626 		info->data = 0;
1627 		break;
1628 	}
1629 
1630 	return 0;
1631 }
1632 
1633 static int
1634 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1635 		 u32 *rules)
1636 {
1637 	struct net_device_context *ndc = netdev_priv(dev);
1638 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1639 
1640 	if (!nvdev)
1641 		return -ENODEV;
1642 
1643 	switch (info->cmd) {
1644 	case ETHTOOL_GRXRINGS:
1645 		info->data = nvdev->num_chn;
1646 		return 0;
1647 
1648 	case ETHTOOL_GRXFH:
1649 		return netvsc_get_rss_hash_opts(ndc, info);
1650 	}
1651 	return -EOPNOTSUPP;
1652 }
1653 
1654 static int netvsc_set_rss_hash_opts(struct net_device_context *ndc,
1655 				    struct ethtool_rxnfc *info)
1656 {
1657 	if (info->data == (RXH_IP_SRC | RXH_IP_DST |
1658 			   RXH_L4_B_0_1 | RXH_L4_B_2_3)) {
1659 		switch (info->flow_type) {
1660 		case TCP_V4_FLOW:
1661 			ndc->l4_hash |= HV_TCP4_L4HASH;
1662 			break;
1663 
1664 		case TCP_V6_FLOW:
1665 			ndc->l4_hash |= HV_TCP6_L4HASH;
1666 			break;
1667 
1668 		case UDP_V4_FLOW:
1669 			ndc->l4_hash |= HV_UDP4_L4HASH;
1670 			break;
1671 
1672 		case UDP_V6_FLOW:
1673 			ndc->l4_hash |= HV_UDP6_L4HASH;
1674 			break;
1675 
1676 		default:
1677 			return -EOPNOTSUPP;
1678 		}
1679 
1680 		return 0;
1681 	}
1682 
1683 	if (info->data == (RXH_IP_SRC | RXH_IP_DST)) {
1684 		switch (info->flow_type) {
1685 		case TCP_V4_FLOW:
1686 			ndc->l4_hash &= ~HV_TCP4_L4HASH;
1687 			break;
1688 
1689 		case TCP_V6_FLOW:
1690 			ndc->l4_hash &= ~HV_TCP6_L4HASH;
1691 			break;
1692 
1693 		case UDP_V4_FLOW:
1694 			ndc->l4_hash &= ~HV_UDP4_L4HASH;
1695 			break;
1696 
1697 		case UDP_V6_FLOW:
1698 			ndc->l4_hash &= ~HV_UDP6_L4HASH;
1699 			break;
1700 
1701 		default:
1702 			return -EOPNOTSUPP;
1703 		}
1704 
1705 		return 0;
1706 	}
1707 
1708 	return -EOPNOTSUPP;
1709 }
1710 
1711 static int
1712 netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info)
1713 {
1714 	struct net_device_context *ndc = netdev_priv(ndev);
1715 
1716 	if (info->cmd == ETHTOOL_SRXFH)
1717 		return netvsc_set_rss_hash_opts(ndc, info);
1718 
1719 	return -EOPNOTSUPP;
1720 }
1721 
1722 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1723 {
1724 	return NETVSC_HASH_KEYLEN;
1725 }
1726 
1727 static u32 netvsc_rss_indir_size(struct net_device *dev)
1728 {
1729 	return ITAB_NUM;
1730 }
1731 
1732 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1733 			   u8 *hfunc)
1734 {
1735 	struct net_device_context *ndc = netdev_priv(dev);
1736 	struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1737 	struct rndis_device *rndis_dev;
1738 	int i;
1739 
1740 	if (!ndev)
1741 		return -ENODEV;
1742 
1743 	if (hfunc)
1744 		*hfunc = ETH_RSS_HASH_TOP;	/* Toeplitz */
1745 
1746 	rndis_dev = ndev->extension;
1747 	if (indir) {
1748 		for (i = 0; i < ITAB_NUM; i++)
1749 			indir[i] = ndc->rx_table[i];
1750 	}
1751 
1752 	if (key)
1753 		memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1754 
1755 	return 0;
1756 }
1757 
1758 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1759 			   const u8 *key, const u8 hfunc)
1760 {
1761 	struct net_device_context *ndc = netdev_priv(dev);
1762 	struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1763 	struct rndis_device *rndis_dev;
1764 	int i;
1765 
1766 	if (!ndev)
1767 		return -ENODEV;
1768 
1769 	if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1770 		return -EOPNOTSUPP;
1771 
1772 	rndis_dev = ndev->extension;
1773 	if (indir) {
1774 		for (i = 0; i < ITAB_NUM; i++)
1775 			if (indir[i] >= ndev->num_chn)
1776 				return -EINVAL;
1777 
1778 		for (i = 0; i < ITAB_NUM; i++)
1779 			ndc->rx_table[i] = indir[i];
1780 	}
1781 
1782 	if (!key) {
1783 		if (!indir)
1784 			return 0;
1785 
1786 		key = rndis_dev->rss_key;
1787 	}
1788 
1789 	return rndis_filter_set_rss_param(rndis_dev, key);
1790 }
1791 
1792 /* Hyper-V RNDIS protocol does not have ring in the HW sense.
1793  * It does have pre-allocated receive area which is divided into sections.
1794  */
1795 static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
1796 				   struct ethtool_ringparam *ring)
1797 {
1798 	u32 max_buf_size;
1799 
1800 	ring->rx_pending = nvdev->recv_section_cnt;
1801 	ring->tx_pending = nvdev->send_section_cnt;
1802 
1803 	if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
1804 		max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
1805 	else
1806 		max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1807 
1808 	ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
1809 	ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
1810 		/ nvdev->send_section_size;
1811 }
1812 
1813 static void netvsc_get_ringparam(struct net_device *ndev,
1814 				 struct ethtool_ringparam *ring)
1815 {
1816 	struct net_device_context *ndevctx = netdev_priv(ndev);
1817 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1818 
1819 	if (!nvdev)
1820 		return;
1821 
1822 	__netvsc_get_ringparam(nvdev, ring);
1823 }
1824 
1825 static int netvsc_set_ringparam(struct net_device *ndev,
1826 				struct ethtool_ringparam *ring)
1827 {
1828 	struct net_device_context *ndevctx = netdev_priv(ndev);
1829 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1830 	struct netvsc_device_info *device_info;
1831 	struct ethtool_ringparam orig;
1832 	u32 new_tx, new_rx;
1833 	int ret = 0;
1834 
1835 	if (!nvdev || nvdev->destroy)
1836 		return -ENODEV;
1837 
1838 	memset(&orig, 0, sizeof(orig));
1839 	__netvsc_get_ringparam(nvdev, &orig);
1840 
1841 	new_tx = clamp_t(u32, ring->tx_pending,
1842 			 NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
1843 	new_rx = clamp_t(u32, ring->rx_pending,
1844 			 NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
1845 
1846 	if (new_tx == orig.tx_pending &&
1847 	    new_rx == orig.rx_pending)
1848 		return 0;	 /* no change */
1849 
1850 	device_info = netvsc_devinfo_get(nvdev);
1851 
1852 	if (!device_info)
1853 		return -ENOMEM;
1854 
1855 	device_info->send_sections = new_tx;
1856 	device_info->recv_sections = new_rx;
1857 
1858 	ret = netvsc_detach(ndev, nvdev);
1859 	if (ret)
1860 		goto out;
1861 
1862 	ret = netvsc_attach(ndev, device_info);
1863 	if (ret) {
1864 		device_info->send_sections = orig.tx_pending;
1865 		device_info->recv_sections = orig.rx_pending;
1866 
1867 		if (netvsc_attach(ndev, device_info))
1868 			netdev_err(ndev, "restoring ringparam failed");
1869 	}
1870 
1871 out:
1872 	netvsc_devinfo_put(device_info);
1873 	return ret;
1874 }
1875 
1876 static netdev_features_t netvsc_fix_features(struct net_device *ndev,
1877 					     netdev_features_t features)
1878 {
1879 	struct net_device_context *ndevctx = netdev_priv(ndev);
1880 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1881 
1882 	if (!nvdev || nvdev->destroy)
1883 		return features;
1884 
1885 	if ((features & NETIF_F_LRO) && netvsc_xdp_get(nvdev)) {
1886 		features ^= NETIF_F_LRO;
1887 		netdev_info(ndev, "Skip LRO - unsupported with XDP\n");
1888 	}
1889 
1890 	return features;
1891 }
1892 
1893 static int netvsc_set_features(struct net_device *ndev,
1894 			       netdev_features_t features)
1895 {
1896 	netdev_features_t change = features ^ ndev->features;
1897 	struct net_device_context *ndevctx = netdev_priv(ndev);
1898 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1899 	struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1900 	struct ndis_offload_params offloads;
1901 	int ret = 0;
1902 
1903 	if (!nvdev || nvdev->destroy)
1904 		return -ENODEV;
1905 
1906 	if (!(change & NETIF_F_LRO))
1907 		goto syncvf;
1908 
1909 	memset(&offloads, 0, sizeof(struct ndis_offload_params));
1910 
1911 	if (features & NETIF_F_LRO) {
1912 		offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1913 		offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1914 	} else {
1915 		offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1916 		offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1917 	}
1918 
1919 	ret = rndis_filter_set_offload_params(ndev, nvdev, &offloads);
1920 
1921 	if (ret) {
1922 		features ^= NETIF_F_LRO;
1923 		ndev->features = features;
1924 	}
1925 
1926 syncvf:
1927 	if (!vf_netdev)
1928 		return ret;
1929 
1930 	vf_netdev->wanted_features = features;
1931 	netdev_update_features(vf_netdev);
1932 
1933 	return ret;
1934 }
1935 
1936 static u32 netvsc_get_msglevel(struct net_device *ndev)
1937 {
1938 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
1939 
1940 	return ndev_ctx->msg_enable;
1941 }
1942 
1943 static void netvsc_set_msglevel(struct net_device *ndev, u32 val)
1944 {
1945 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
1946 
1947 	ndev_ctx->msg_enable = val;
1948 }
1949 
1950 static const struct ethtool_ops ethtool_ops = {
1951 	.get_drvinfo	= netvsc_get_drvinfo,
1952 	.get_msglevel	= netvsc_get_msglevel,
1953 	.set_msglevel	= netvsc_set_msglevel,
1954 	.get_link	= ethtool_op_get_link,
1955 	.get_ethtool_stats = netvsc_get_ethtool_stats,
1956 	.get_sset_count = netvsc_get_sset_count,
1957 	.get_strings	= netvsc_get_strings,
1958 	.get_channels   = netvsc_get_channels,
1959 	.set_channels   = netvsc_set_channels,
1960 	.get_ts_info	= ethtool_op_get_ts_info,
1961 	.get_rxnfc	= netvsc_get_rxnfc,
1962 	.set_rxnfc	= netvsc_set_rxnfc,
1963 	.get_rxfh_key_size = netvsc_get_rxfh_key_size,
1964 	.get_rxfh_indir_size = netvsc_rss_indir_size,
1965 	.get_rxfh	= netvsc_get_rxfh,
1966 	.set_rxfh	= netvsc_set_rxfh,
1967 	.get_link_ksettings = netvsc_get_link_ksettings,
1968 	.set_link_ksettings = netvsc_set_link_ksettings,
1969 	.get_ringparam	= netvsc_get_ringparam,
1970 	.set_ringparam	= netvsc_set_ringparam,
1971 };
1972 
1973 static const struct net_device_ops device_ops = {
1974 	.ndo_open =			netvsc_open,
1975 	.ndo_stop =			netvsc_close,
1976 	.ndo_start_xmit =		netvsc_start_xmit,
1977 	.ndo_change_rx_flags =		netvsc_change_rx_flags,
1978 	.ndo_set_rx_mode =		netvsc_set_rx_mode,
1979 	.ndo_fix_features =		netvsc_fix_features,
1980 	.ndo_set_features =		netvsc_set_features,
1981 	.ndo_change_mtu =		netvsc_change_mtu,
1982 	.ndo_validate_addr =		eth_validate_addr,
1983 	.ndo_set_mac_address =		netvsc_set_mac_addr,
1984 	.ndo_select_queue =		netvsc_select_queue,
1985 	.ndo_get_stats64 =		netvsc_get_stats64,
1986 	.ndo_bpf =			netvsc_bpf,
1987 };
1988 
1989 /*
1990  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1991  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1992  * present send GARP packet to network peers with netif_notify_peers().
1993  */
1994 static void netvsc_link_change(struct work_struct *w)
1995 {
1996 	struct net_device_context *ndev_ctx =
1997 		container_of(w, struct net_device_context, dwork.work);
1998 	struct hv_device *device_obj = ndev_ctx->device_ctx;
1999 	struct net_device *net = hv_get_drvdata(device_obj);
2000 	struct netvsc_device *net_device;
2001 	struct rndis_device *rdev;
2002 	struct netvsc_reconfig *event = NULL;
2003 	bool notify = false, reschedule = false;
2004 	unsigned long flags, next_reconfig, delay;
2005 
2006 	/* if changes are happening, comeback later */
2007 	if (!rtnl_trylock()) {
2008 		schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
2009 		return;
2010 	}
2011 
2012 	net_device = rtnl_dereference(ndev_ctx->nvdev);
2013 	if (!net_device)
2014 		goto out_unlock;
2015 
2016 	rdev = net_device->extension;
2017 
2018 	next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
2019 	if (time_is_after_jiffies(next_reconfig)) {
2020 		/* link_watch only sends one notification with current state
2021 		 * per second, avoid doing reconfig more frequently. Handle
2022 		 * wrap around.
2023 		 */
2024 		delay = next_reconfig - jiffies;
2025 		delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
2026 		schedule_delayed_work(&ndev_ctx->dwork, delay);
2027 		goto out_unlock;
2028 	}
2029 	ndev_ctx->last_reconfig = jiffies;
2030 
2031 	spin_lock_irqsave(&ndev_ctx->lock, flags);
2032 	if (!list_empty(&ndev_ctx->reconfig_events)) {
2033 		event = list_first_entry(&ndev_ctx->reconfig_events,
2034 					 struct netvsc_reconfig, list);
2035 		list_del(&event->list);
2036 		reschedule = !list_empty(&ndev_ctx->reconfig_events);
2037 	}
2038 	spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2039 
2040 	if (!event)
2041 		goto out_unlock;
2042 
2043 	switch (event->event) {
2044 		/* Only the following events are possible due to the check in
2045 		 * netvsc_linkstatus_callback()
2046 		 */
2047 	case RNDIS_STATUS_MEDIA_CONNECT:
2048 		if (rdev->link_state) {
2049 			rdev->link_state = false;
2050 			netif_carrier_on(net);
2051 			netvsc_tx_enable(net_device, net);
2052 		} else {
2053 			notify = true;
2054 		}
2055 		kfree(event);
2056 		break;
2057 	case RNDIS_STATUS_MEDIA_DISCONNECT:
2058 		if (!rdev->link_state) {
2059 			rdev->link_state = true;
2060 			netif_carrier_off(net);
2061 			netvsc_tx_disable(net_device, net);
2062 		}
2063 		kfree(event);
2064 		break;
2065 	case RNDIS_STATUS_NETWORK_CHANGE:
2066 		/* Only makes sense if carrier is present */
2067 		if (!rdev->link_state) {
2068 			rdev->link_state = true;
2069 			netif_carrier_off(net);
2070 			netvsc_tx_disable(net_device, net);
2071 			event->event = RNDIS_STATUS_MEDIA_CONNECT;
2072 			spin_lock_irqsave(&ndev_ctx->lock, flags);
2073 			list_add(&event->list, &ndev_ctx->reconfig_events);
2074 			spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2075 			reschedule = true;
2076 		}
2077 		break;
2078 	}
2079 
2080 	rtnl_unlock();
2081 
2082 	if (notify)
2083 		netdev_notify_peers(net);
2084 
2085 	/* link_watch only sends one notification with current state per
2086 	 * second, handle next reconfig event in 2 seconds.
2087 	 */
2088 	if (reschedule)
2089 		schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
2090 
2091 	return;
2092 
2093 out_unlock:
2094 	rtnl_unlock();
2095 }
2096 
2097 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
2098 {
2099 	struct net_device_context *net_device_ctx;
2100 	struct net_device *dev;
2101 
2102 	dev = netdev_master_upper_dev_get(vf_netdev);
2103 	if (!dev || dev->netdev_ops != &device_ops)
2104 		return NULL;	/* not a netvsc device */
2105 
2106 	net_device_ctx = netdev_priv(dev);
2107 	if (!rtnl_dereference(net_device_ctx->nvdev))
2108 		return NULL;	/* device is removed */
2109 
2110 	return dev;
2111 }
2112 
2113 /* Called when VF is injecting data into network stack.
2114  * Change the associated network device from VF to netvsc.
2115  * note: already called with rcu_read_lock
2116  */
2117 static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
2118 {
2119 	struct sk_buff *skb = *pskb;
2120 	struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data);
2121 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
2122 	struct netvsc_vf_pcpu_stats *pcpu_stats
2123 		 = this_cpu_ptr(ndev_ctx->vf_stats);
2124 
2125 	skb = skb_share_check(skb, GFP_ATOMIC);
2126 	if (unlikely(!skb))
2127 		return RX_HANDLER_CONSUMED;
2128 
2129 	*pskb = skb;
2130 
2131 	skb->dev = ndev;
2132 
2133 	u64_stats_update_begin(&pcpu_stats->syncp);
2134 	pcpu_stats->rx_packets++;
2135 	pcpu_stats->rx_bytes += skb->len;
2136 	u64_stats_update_end(&pcpu_stats->syncp);
2137 
2138 	return RX_HANDLER_ANOTHER;
2139 }
2140 
2141 static int netvsc_vf_join(struct net_device *vf_netdev,
2142 			  struct net_device *ndev)
2143 {
2144 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
2145 	int ret;
2146 
2147 	ret = netdev_rx_handler_register(vf_netdev,
2148 					 netvsc_vf_handle_frame, ndev);
2149 	if (ret != 0) {
2150 		netdev_err(vf_netdev,
2151 			   "can not register netvsc VF receive handler (err = %d)\n",
2152 			   ret);
2153 		goto rx_handler_failed;
2154 	}
2155 
2156 	ret = netdev_master_upper_dev_link(vf_netdev, ndev,
2157 					   NULL, NULL, NULL);
2158 	if (ret != 0) {
2159 		netdev_err(vf_netdev,
2160 			   "can not set master device %s (err = %d)\n",
2161 			   ndev->name, ret);
2162 		goto upper_link_failed;
2163 	}
2164 
2165 	/* set slave flag before open to prevent IPv6 addrconf */
2166 	vf_netdev->flags |= IFF_SLAVE;
2167 
2168 	schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
2169 
2170 	call_netdevice_notifiers(NETDEV_JOIN, vf_netdev);
2171 
2172 	netdev_info(vf_netdev, "joined to %s\n", ndev->name);
2173 	return 0;
2174 
2175 upper_link_failed:
2176 	netdev_rx_handler_unregister(vf_netdev);
2177 rx_handler_failed:
2178 	return ret;
2179 }
2180 
2181 static void __netvsc_vf_setup(struct net_device *ndev,
2182 			      struct net_device *vf_netdev)
2183 {
2184 	int ret;
2185 
2186 	/* Align MTU of VF with master */
2187 	ret = dev_set_mtu(vf_netdev, ndev->mtu);
2188 	if (ret)
2189 		netdev_warn(vf_netdev,
2190 			    "unable to change mtu to %u\n", ndev->mtu);
2191 
2192 	/* set multicast etc flags on VF */
2193 	dev_change_flags(vf_netdev, ndev->flags | IFF_SLAVE, NULL);
2194 
2195 	/* sync address list from ndev to VF */
2196 	netif_addr_lock_bh(ndev);
2197 	dev_uc_sync(vf_netdev, ndev);
2198 	dev_mc_sync(vf_netdev, ndev);
2199 	netif_addr_unlock_bh(ndev);
2200 
2201 	if (netif_running(ndev)) {
2202 		ret = dev_open(vf_netdev, NULL);
2203 		if (ret)
2204 			netdev_warn(vf_netdev,
2205 				    "unable to open: %d\n", ret);
2206 	}
2207 }
2208 
2209 /* Setup VF as slave of the synthetic device.
2210  * Runs in workqueue to avoid recursion in netlink callbacks.
2211  */
2212 static void netvsc_vf_setup(struct work_struct *w)
2213 {
2214 	struct net_device_context *ndev_ctx
2215 		= container_of(w, struct net_device_context, vf_takeover.work);
2216 	struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2217 	struct net_device *vf_netdev;
2218 
2219 	if (!rtnl_trylock()) {
2220 		schedule_delayed_work(&ndev_ctx->vf_takeover, 0);
2221 		return;
2222 	}
2223 
2224 	vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2225 	if (vf_netdev)
2226 		__netvsc_vf_setup(ndev, vf_netdev);
2227 
2228 	rtnl_unlock();
2229 }
2230 
2231 /* Find netvsc by VF serial number.
2232  * The PCI hyperv controller records the serial number as the slot kobj name.
2233  */
2234 static struct net_device *get_netvsc_byslot(const struct net_device *vf_netdev)
2235 {
2236 	struct device *parent = vf_netdev->dev.parent;
2237 	struct net_device_context *ndev_ctx;
2238 	struct pci_dev *pdev;
2239 	u32 serial;
2240 
2241 	if (!parent || !dev_is_pci(parent))
2242 		return NULL; /* not a PCI device */
2243 
2244 	pdev = to_pci_dev(parent);
2245 	if (!pdev->slot) {
2246 		netdev_notice(vf_netdev, "no PCI slot information\n");
2247 		return NULL;
2248 	}
2249 
2250 	if (kstrtou32(pci_slot_name(pdev->slot), 10, &serial)) {
2251 		netdev_notice(vf_netdev, "Invalid vf serial:%s\n",
2252 			      pci_slot_name(pdev->slot));
2253 		return NULL;
2254 	}
2255 
2256 	list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
2257 		if (!ndev_ctx->vf_alloc)
2258 			continue;
2259 
2260 		if (ndev_ctx->vf_serial == serial)
2261 			return hv_get_drvdata(ndev_ctx->device_ctx);
2262 	}
2263 
2264 	netdev_notice(vf_netdev,
2265 		      "no netdev found for vf serial:%u\n", serial);
2266 	return NULL;
2267 }
2268 
2269 static int netvsc_register_vf(struct net_device *vf_netdev)
2270 {
2271 	struct net_device_context *net_device_ctx;
2272 	struct netvsc_device *netvsc_dev;
2273 	struct bpf_prog *prog;
2274 	struct net_device *ndev;
2275 	int ret;
2276 
2277 	if (vf_netdev->addr_len != ETH_ALEN)
2278 		return NOTIFY_DONE;
2279 
2280 	ndev = get_netvsc_byslot(vf_netdev);
2281 	if (!ndev)
2282 		return NOTIFY_DONE;
2283 
2284 	net_device_ctx = netdev_priv(ndev);
2285 	netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2286 	if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
2287 		return NOTIFY_DONE;
2288 
2289 	/* if synthetic interface is a different namespace,
2290 	 * then move the VF to that namespace; join will be
2291 	 * done again in that context.
2292 	 */
2293 	if (!net_eq(dev_net(ndev), dev_net(vf_netdev))) {
2294 		ret = dev_change_net_namespace(vf_netdev,
2295 					       dev_net(ndev), "eth%d");
2296 		if (ret)
2297 			netdev_err(vf_netdev,
2298 				   "could not move to same namespace as %s: %d\n",
2299 				   ndev->name, ret);
2300 		else
2301 			netdev_info(vf_netdev,
2302 				    "VF moved to namespace with: %s\n",
2303 				    ndev->name);
2304 		return NOTIFY_DONE;
2305 	}
2306 
2307 	netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
2308 
2309 	if (netvsc_vf_join(vf_netdev, ndev) != 0)
2310 		return NOTIFY_DONE;
2311 
2312 	dev_hold(vf_netdev);
2313 	rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
2314 
2315 	vf_netdev->wanted_features = ndev->features;
2316 	netdev_update_features(vf_netdev);
2317 
2318 	prog = netvsc_xdp_get(netvsc_dev);
2319 	netvsc_vf_setxdp(vf_netdev, prog);
2320 
2321 	return NOTIFY_OK;
2322 }
2323 
2324 /* VF up/down change detected, schedule to change data path */
2325 static int netvsc_vf_changed(struct net_device *vf_netdev)
2326 {
2327 	struct net_device_context *net_device_ctx;
2328 	struct netvsc_device *netvsc_dev;
2329 	struct net_device *ndev;
2330 	bool vf_is_up = netif_running(vf_netdev);
2331 
2332 	ndev = get_netvsc_byref(vf_netdev);
2333 	if (!ndev)
2334 		return NOTIFY_DONE;
2335 
2336 	net_device_ctx = netdev_priv(ndev);
2337 	netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2338 	if (!netvsc_dev)
2339 		return NOTIFY_DONE;
2340 
2341 	netvsc_switch_datapath(ndev, vf_is_up);
2342 	netdev_info(ndev, "Data path switched %s VF: %s\n",
2343 		    vf_is_up ? "to" : "from", vf_netdev->name);
2344 
2345 	return NOTIFY_OK;
2346 }
2347 
2348 static int netvsc_unregister_vf(struct net_device *vf_netdev)
2349 {
2350 	struct net_device *ndev;
2351 	struct net_device_context *net_device_ctx;
2352 
2353 	ndev = get_netvsc_byref(vf_netdev);
2354 	if (!ndev)
2355 		return NOTIFY_DONE;
2356 
2357 	net_device_ctx = netdev_priv(ndev);
2358 	cancel_delayed_work_sync(&net_device_ctx->vf_takeover);
2359 
2360 	netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
2361 
2362 	netvsc_vf_setxdp(vf_netdev, NULL);
2363 
2364 	netdev_rx_handler_unregister(vf_netdev);
2365 	netdev_upper_dev_unlink(vf_netdev, ndev);
2366 	RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
2367 	dev_put(vf_netdev);
2368 
2369 	return NOTIFY_OK;
2370 }
2371 
2372 static int netvsc_probe(struct hv_device *dev,
2373 			const struct hv_vmbus_device_id *dev_id)
2374 {
2375 	struct net_device *net = NULL;
2376 	struct net_device_context *net_device_ctx;
2377 	struct netvsc_device_info *device_info = NULL;
2378 	struct netvsc_device *nvdev;
2379 	int ret = -ENOMEM;
2380 
2381 	net = alloc_etherdev_mq(sizeof(struct net_device_context),
2382 				VRSS_CHANNEL_MAX);
2383 	if (!net)
2384 		goto no_net;
2385 
2386 	netif_carrier_off(net);
2387 
2388 	netvsc_init_settings(net);
2389 
2390 	net_device_ctx = netdev_priv(net);
2391 	net_device_ctx->device_ctx = dev;
2392 	net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
2393 	if (netif_msg_probe(net_device_ctx))
2394 		netdev_dbg(net, "netvsc msg_enable: %d\n",
2395 			   net_device_ctx->msg_enable);
2396 
2397 	hv_set_drvdata(dev, net);
2398 
2399 	INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
2400 
2401 	spin_lock_init(&net_device_ctx->lock);
2402 	INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
2403 	INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup);
2404 
2405 	net_device_ctx->vf_stats
2406 		= netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats);
2407 	if (!net_device_ctx->vf_stats)
2408 		goto no_stats;
2409 
2410 	net->netdev_ops = &device_ops;
2411 	net->ethtool_ops = &ethtool_ops;
2412 	SET_NETDEV_DEV(net, &dev->device);
2413 
2414 	/* We always need headroom for rndis header */
2415 	net->needed_headroom = RNDIS_AND_PPI_SIZE;
2416 
2417 	/* Initialize the number of queues to be 1, we may change it if more
2418 	 * channels are offered later.
2419 	 */
2420 	netif_set_real_num_tx_queues(net, 1);
2421 	netif_set_real_num_rx_queues(net, 1);
2422 
2423 	/* Notify the netvsc driver of the new device */
2424 	device_info = netvsc_devinfo_get(NULL);
2425 
2426 	if (!device_info) {
2427 		ret = -ENOMEM;
2428 		goto devinfo_failed;
2429 	}
2430 
2431 	nvdev = rndis_filter_device_add(dev, device_info);
2432 	if (IS_ERR(nvdev)) {
2433 		ret = PTR_ERR(nvdev);
2434 		netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
2435 		goto rndis_failed;
2436 	}
2437 
2438 	memcpy(net->dev_addr, device_info->mac_adr, ETH_ALEN);
2439 
2440 	/* We must get rtnl lock before scheduling nvdev->subchan_work,
2441 	 * otherwise netvsc_subchan_work() can get rtnl lock first and wait
2442 	 * all subchannels to show up, but that may not happen because
2443 	 * netvsc_probe() can't get rtnl lock and as a result vmbus_onoffer()
2444 	 * -> ... -> device_add() -> ... -> __device_attach() can't get
2445 	 * the device lock, so all the subchannels can't be processed --
2446 	 * finally netvsc_subchan_work() hangs forever.
2447 	 */
2448 	rtnl_lock();
2449 
2450 	if (nvdev->num_chn > 1)
2451 		schedule_work(&nvdev->subchan_work);
2452 
2453 	/* hw_features computed in rndis_netdev_set_hwcaps() */
2454 	net->features = net->hw_features |
2455 		NETIF_F_HIGHDMA | NETIF_F_HW_VLAN_CTAG_TX |
2456 		NETIF_F_HW_VLAN_CTAG_RX;
2457 	net->vlan_features = net->features;
2458 
2459 	/* MTU range: 68 - 1500 or 65521 */
2460 	net->min_mtu = NETVSC_MTU_MIN;
2461 	if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
2462 		net->max_mtu = NETVSC_MTU - ETH_HLEN;
2463 	else
2464 		net->max_mtu = ETH_DATA_LEN;
2465 
2466 	nvdev->tx_disable = false;
2467 
2468 	ret = register_netdevice(net);
2469 	if (ret != 0) {
2470 		pr_err("Unable to register netdev.\n");
2471 		goto register_failed;
2472 	}
2473 
2474 	list_add(&net_device_ctx->list, &netvsc_dev_list);
2475 	rtnl_unlock();
2476 
2477 	netvsc_devinfo_put(device_info);
2478 	return 0;
2479 
2480 register_failed:
2481 	rtnl_unlock();
2482 	rndis_filter_device_remove(dev, nvdev);
2483 rndis_failed:
2484 	netvsc_devinfo_put(device_info);
2485 devinfo_failed:
2486 	free_percpu(net_device_ctx->vf_stats);
2487 no_stats:
2488 	hv_set_drvdata(dev, NULL);
2489 	free_netdev(net);
2490 no_net:
2491 	return ret;
2492 }
2493 
2494 static int netvsc_remove(struct hv_device *dev)
2495 {
2496 	struct net_device_context *ndev_ctx;
2497 	struct net_device *vf_netdev, *net;
2498 	struct netvsc_device *nvdev;
2499 
2500 	net = hv_get_drvdata(dev);
2501 	if (net == NULL) {
2502 		dev_err(&dev->device, "No net device to remove\n");
2503 		return 0;
2504 	}
2505 
2506 	ndev_ctx = netdev_priv(net);
2507 
2508 	cancel_delayed_work_sync(&ndev_ctx->dwork);
2509 
2510 	rtnl_lock();
2511 	nvdev = rtnl_dereference(ndev_ctx->nvdev);
2512 	if (nvdev) {
2513 		cancel_work_sync(&nvdev->subchan_work);
2514 		netvsc_xdp_set(net, NULL, NULL, nvdev);
2515 	}
2516 
2517 	/*
2518 	 * Call to the vsc driver to let it know that the device is being
2519 	 * removed. Also blocks mtu and channel changes.
2520 	 */
2521 	vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2522 	if (vf_netdev)
2523 		netvsc_unregister_vf(vf_netdev);
2524 
2525 	if (nvdev)
2526 		rndis_filter_device_remove(dev, nvdev);
2527 
2528 	unregister_netdevice(net);
2529 	list_del(&ndev_ctx->list);
2530 
2531 	rtnl_unlock();
2532 
2533 	hv_set_drvdata(dev, NULL);
2534 
2535 	free_percpu(ndev_ctx->vf_stats);
2536 	free_netdev(net);
2537 	return 0;
2538 }
2539 
2540 static int netvsc_suspend(struct hv_device *dev)
2541 {
2542 	struct net_device_context *ndev_ctx;
2543 	struct net_device *vf_netdev, *net;
2544 	struct netvsc_device *nvdev;
2545 	int ret;
2546 
2547 	net = hv_get_drvdata(dev);
2548 
2549 	ndev_ctx = netdev_priv(net);
2550 	cancel_delayed_work_sync(&ndev_ctx->dwork);
2551 
2552 	rtnl_lock();
2553 
2554 	nvdev = rtnl_dereference(ndev_ctx->nvdev);
2555 	if (nvdev == NULL) {
2556 		ret = -ENODEV;
2557 		goto out;
2558 	}
2559 
2560 	vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2561 	if (vf_netdev)
2562 		netvsc_unregister_vf(vf_netdev);
2563 
2564 	/* Save the current config info */
2565 	ndev_ctx->saved_netvsc_dev_info = netvsc_devinfo_get(nvdev);
2566 
2567 	ret = netvsc_detach(net, nvdev);
2568 out:
2569 	rtnl_unlock();
2570 
2571 	return ret;
2572 }
2573 
2574 static int netvsc_resume(struct hv_device *dev)
2575 {
2576 	struct net_device *net = hv_get_drvdata(dev);
2577 	struct net_device_context *net_device_ctx;
2578 	struct netvsc_device_info *device_info;
2579 	int ret;
2580 
2581 	rtnl_lock();
2582 
2583 	net_device_ctx = netdev_priv(net);
2584 	device_info = net_device_ctx->saved_netvsc_dev_info;
2585 
2586 	ret = netvsc_attach(net, device_info);
2587 
2588 	netvsc_devinfo_put(device_info);
2589 	net_device_ctx->saved_netvsc_dev_info = NULL;
2590 
2591 	rtnl_unlock();
2592 
2593 	return ret;
2594 }
2595 static const struct hv_vmbus_device_id id_table[] = {
2596 	/* Network guid */
2597 	{ HV_NIC_GUID, },
2598 	{ },
2599 };
2600 
2601 MODULE_DEVICE_TABLE(vmbus, id_table);
2602 
2603 /* The one and only one */
2604 static struct  hv_driver netvsc_drv = {
2605 	.name = KBUILD_MODNAME,
2606 	.id_table = id_table,
2607 	.probe = netvsc_probe,
2608 	.remove = netvsc_remove,
2609 	.suspend = netvsc_suspend,
2610 	.resume = netvsc_resume,
2611 	.driver = {
2612 		.probe_type = PROBE_FORCE_SYNCHRONOUS,
2613 	},
2614 };
2615 
2616 /*
2617  * On Hyper-V, every VF interface is matched with a corresponding
2618  * synthetic interface. The synthetic interface is presented first
2619  * to the guest. When the corresponding VF instance is registered,
2620  * we will take care of switching the data path.
2621  */
2622 static int netvsc_netdev_event(struct notifier_block *this,
2623 			       unsigned long event, void *ptr)
2624 {
2625 	struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2626 
2627 	/* Skip our own events */
2628 	if (event_dev->netdev_ops == &device_ops)
2629 		return NOTIFY_DONE;
2630 
2631 	/* Avoid non-Ethernet type devices */
2632 	if (event_dev->type != ARPHRD_ETHER)
2633 		return NOTIFY_DONE;
2634 
2635 	/* Avoid Vlan dev with same MAC registering as VF */
2636 	if (is_vlan_dev(event_dev))
2637 		return NOTIFY_DONE;
2638 
2639 	/* Avoid Bonding master dev with same MAC registering as VF */
2640 	if ((event_dev->priv_flags & IFF_BONDING) &&
2641 	    (event_dev->flags & IFF_MASTER))
2642 		return NOTIFY_DONE;
2643 
2644 	switch (event) {
2645 	case NETDEV_REGISTER:
2646 		return netvsc_register_vf(event_dev);
2647 	case NETDEV_UNREGISTER:
2648 		return netvsc_unregister_vf(event_dev);
2649 	case NETDEV_UP:
2650 	case NETDEV_DOWN:
2651 		return netvsc_vf_changed(event_dev);
2652 	default:
2653 		return NOTIFY_DONE;
2654 	}
2655 }
2656 
2657 static struct notifier_block netvsc_netdev_notifier = {
2658 	.notifier_call = netvsc_netdev_event,
2659 };
2660 
2661 static void __exit netvsc_drv_exit(void)
2662 {
2663 	unregister_netdevice_notifier(&netvsc_netdev_notifier);
2664 	vmbus_driver_unregister(&netvsc_drv);
2665 }
2666 
2667 static int __init netvsc_drv_init(void)
2668 {
2669 	int ret;
2670 
2671 	if (ring_size < RING_SIZE_MIN) {
2672 		ring_size = RING_SIZE_MIN;
2673 		pr_info("Increased ring_size to %u (min allowed)\n",
2674 			ring_size);
2675 	}
2676 	netvsc_ring_bytes = ring_size * PAGE_SIZE;
2677 
2678 	ret = vmbus_driver_register(&netvsc_drv);
2679 	if (ret)
2680 		return ret;
2681 
2682 	register_netdevice_notifier(&netvsc_netdev_notifier);
2683 	return 0;
2684 }
2685 
2686 MODULE_LICENSE("GPL");
2687 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
2688 
2689 module_init(netvsc_drv_init);
2690 module_exit(netvsc_drv_exit);
2691