xref: /openbmc/linux/drivers/net/hyperv/netvsc_drv.c (revision aa5b395b)
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 			ipv6_hdr(skb)->payload_len = 0;
642 			tcp_hdr(skb)->check =
643 				~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
644 						 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
645 		}
646 		lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
647 		lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
648 	} else if (skb->ip_summed == CHECKSUM_PARTIAL) {
649 		if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
650 			struct ndis_tcp_ip_checksum_info *csum_info;
651 
652 			rndis_msg_size += NDIS_CSUM_PPI_SIZE;
653 			csum_info = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
654 						  TCPIP_CHKSUM_PKTINFO);
655 
656 			csum_info->value = 0;
657 			csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
658 
659 			if (skb->protocol == htons(ETH_P_IP)) {
660 				csum_info->transmit.is_ipv4 = 1;
661 
662 				if (ip_hdr(skb)->protocol == IPPROTO_TCP)
663 					csum_info->transmit.tcp_checksum = 1;
664 				else
665 					csum_info->transmit.udp_checksum = 1;
666 			} else {
667 				csum_info->transmit.is_ipv6 = 1;
668 
669 				if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
670 					csum_info->transmit.tcp_checksum = 1;
671 				else
672 					csum_info->transmit.udp_checksum = 1;
673 			}
674 		} else {
675 			/* Can't do offload of this type of checksum */
676 			if (skb_checksum_help(skb))
677 				goto drop;
678 		}
679 	}
680 
681 	/* Start filling in the page buffers with the rndis hdr */
682 	rndis_msg->msg_len += rndis_msg_size;
683 	packet->total_data_buflen = rndis_msg->msg_len;
684 	packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
685 					       skb, packet, pb);
686 
687 	/* timestamp packet in software */
688 	skb_tx_timestamp(skb);
689 
690 	ret = netvsc_send(net, packet, rndis_msg, pb, skb, xdp_tx);
691 	if (likely(ret == 0))
692 		return NETDEV_TX_OK;
693 
694 	if (ret == -EAGAIN) {
695 		++net_device_ctx->eth_stats.tx_busy;
696 		return NETDEV_TX_BUSY;
697 	}
698 
699 	if (ret == -ENOSPC)
700 		++net_device_ctx->eth_stats.tx_no_space;
701 
702 drop:
703 	dev_kfree_skb_any(skb);
704 	net->stats.tx_dropped++;
705 
706 	return NETDEV_TX_OK;
707 
708 no_memory:
709 	++net_device_ctx->eth_stats.tx_no_memory;
710 	goto drop;
711 }
712 
713 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *ndev)
714 {
715 	return netvsc_xmit(skb, ndev, false);
716 }
717 
718 /*
719  * netvsc_linkstatus_callback - Link up/down notification
720  */
721 void netvsc_linkstatus_callback(struct net_device *net,
722 				struct rndis_message *resp)
723 {
724 	struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
725 	struct net_device_context *ndev_ctx = netdev_priv(net);
726 	struct netvsc_reconfig *event;
727 	unsigned long flags;
728 
729 	/* Update the physical link speed when changing to another vSwitch */
730 	if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
731 		u32 speed;
732 
733 		speed = *(u32 *)((void *)indicate
734 				 + indicate->status_buf_offset) / 10000;
735 		ndev_ctx->speed = speed;
736 		return;
737 	}
738 
739 	/* Handle these link change statuses below */
740 	if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
741 	    indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
742 	    indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
743 		return;
744 
745 	if (net->reg_state != NETREG_REGISTERED)
746 		return;
747 
748 	event = kzalloc(sizeof(*event), GFP_ATOMIC);
749 	if (!event)
750 		return;
751 	event->event = indicate->status;
752 
753 	spin_lock_irqsave(&ndev_ctx->lock, flags);
754 	list_add_tail(&event->list, &ndev_ctx->reconfig_events);
755 	spin_unlock_irqrestore(&ndev_ctx->lock, flags);
756 
757 	schedule_delayed_work(&ndev_ctx->dwork, 0);
758 }
759 
760 static void netvsc_xdp_xmit(struct sk_buff *skb, struct net_device *ndev)
761 {
762 	int rc;
763 
764 	skb->queue_mapping = skb_get_rx_queue(skb);
765 	__skb_push(skb, ETH_HLEN);
766 
767 	rc = netvsc_xmit(skb, ndev, true);
768 
769 	if (dev_xmit_complete(rc))
770 		return;
771 
772 	dev_kfree_skb_any(skb);
773 	ndev->stats.tx_dropped++;
774 }
775 
776 static void netvsc_comp_ipcsum(struct sk_buff *skb)
777 {
778 	struct iphdr *iph = (struct iphdr *)skb->data;
779 
780 	iph->check = 0;
781 	iph->check = ip_fast_csum(iph, iph->ihl);
782 }
783 
784 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
785 					     struct netvsc_channel *nvchan,
786 					     struct xdp_buff *xdp)
787 {
788 	struct napi_struct *napi = &nvchan->napi;
789 	const struct ndis_pkt_8021q_info *vlan = nvchan->rsc.vlan;
790 	const struct ndis_tcp_ip_checksum_info *csum_info =
791 						nvchan->rsc.csum_info;
792 	const u32 *hash_info = nvchan->rsc.hash_info;
793 	struct sk_buff *skb;
794 	void *xbuf = xdp->data_hard_start;
795 	int i;
796 
797 	if (xbuf) {
798 		unsigned int hdroom = xdp->data - xdp->data_hard_start;
799 		unsigned int xlen = xdp->data_end - xdp->data;
800 		unsigned int frag_size = netvsc_xdp_fraglen(hdroom + xlen);
801 
802 		skb = build_skb(xbuf, frag_size);
803 
804 		if (!skb) {
805 			__free_page(virt_to_page(xbuf));
806 			return NULL;
807 		}
808 
809 		skb_reserve(skb, hdroom);
810 		skb_put(skb, xlen);
811 		skb->dev = napi->dev;
812 	} else {
813 		skb = napi_alloc_skb(napi, nvchan->rsc.pktlen);
814 
815 		if (!skb)
816 			return NULL;
817 
818 		/* Copy to skb. This copy is needed here since the memory
819 		 * pointed by hv_netvsc_packet cannot be deallocated.
820 		 */
821 		for (i = 0; i < nvchan->rsc.cnt; i++)
822 			skb_put_data(skb, nvchan->rsc.data[i],
823 				     nvchan->rsc.len[i]);
824 	}
825 
826 	skb->protocol = eth_type_trans(skb, net);
827 
828 	/* skb is already created with CHECKSUM_NONE */
829 	skb_checksum_none_assert(skb);
830 
831 	/* Incoming packets may have IP header checksum verified by the host.
832 	 * They may not have IP header checksum computed after coalescing.
833 	 * We compute it here if the flags are set, because on Linux, the IP
834 	 * checksum is always checked.
835 	 */
836 	if (csum_info && csum_info->receive.ip_checksum_value_invalid &&
837 	    csum_info->receive.ip_checksum_succeeded &&
838 	    skb->protocol == htons(ETH_P_IP))
839 		netvsc_comp_ipcsum(skb);
840 
841 	/* Do L4 checksum offload if enabled and present. */
842 	if (csum_info && (net->features & NETIF_F_RXCSUM)) {
843 		if (csum_info->receive.tcp_checksum_succeeded ||
844 		    csum_info->receive.udp_checksum_succeeded)
845 			skb->ip_summed = CHECKSUM_UNNECESSARY;
846 	}
847 
848 	if (hash_info && (net->features & NETIF_F_RXHASH))
849 		skb_set_hash(skb, *hash_info, PKT_HASH_TYPE_L4);
850 
851 	if (vlan) {
852 		u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT) |
853 			(vlan->cfi ? VLAN_CFI_MASK : 0);
854 
855 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
856 				       vlan_tci);
857 	}
858 
859 	return skb;
860 }
861 
862 /*
863  * netvsc_recv_callback -  Callback when we receive a packet from the
864  * "wire" on the specified device.
865  */
866 int netvsc_recv_callback(struct net_device *net,
867 			 struct netvsc_device *net_device,
868 			 struct netvsc_channel *nvchan)
869 {
870 	struct net_device_context *net_device_ctx = netdev_priv(net);
871 	struct vmbus_channel *channel = nvchan->channel;
872 	u16 q_idx = channel->offermsg.offer.sub_channel_index;
873 	struct sk_buff *skb;
874 	struct netvsc_stats *rx_stats = &nvchan->rx_stats;
875 	struct xdp_buff xdp;
876 	u32 act;
877 
878 	if (net->reg_state != NETREG_REGISTERED)
879 		return NVSP_STAT_FAIL;
880 
881 	act = netvsc_run_xdp(net, nvchan, &xdp);
882 
883 	if (act != XDP_PASS && act != XDP_TX) {
884 		u64_stats_update_begin(&rx_stats->syncp);
885 		rx_stats->xdp_drop++;
886 		u64_stats_update_end(&rx_stats->syncp);
887 
888 		return NVSP_STAT_SUCCESS; /* consumed by XDP */
889 	}
890 
891 	/* Allocate a skb - TODO direct I/O to pages? */
892 	skb = netvsc_alloc_recv_skb(net, nvchan, &xdp);
893 
894 	if (unlikely(!skb)) {
895 		++net_device_ctx->eth_stats.rx_no_memory;
896 		return NVSP_STAT_FAIL;
897 	}
898 
899 	skb_record_rx_queue(skb, q_idx);
900 
901 	/*
902 	 * Even if injecting the packet, record the statistics
903 	 * on the synthetic device because modifying the VF device
904 	 * statistics will not work correctly.
905 	 */
906 	u64_stats_update_begin(&rx_stats->syncp);
907 	rx_stats->packets++;
908 	rx_stats->bytes += nvchan->rsc.pktlen;
909 
910 	if (skb->pkt_type == PACKET_BROADCAST)
911 		++rx_stats->broadcast;
912 	else if (skb->pkt_type == PACKET_MULTICAST)
913 		++rx_stats->multicast;
914 	u64_stats_update_end(&rx_stats->syncp);
915 
916 	if (act == XDP_TX) {
917 		netvsc_xdp_xmit(skb, net);
918 		return NVSP_STAT_SUCCESS;
919 	}
920 
921 	napi_gro_receive(&nvchan->napi, skb);
922 	return NVSP_STAT_SUCCESS;
923 }
924 
925 static void netvsc_get_drvinfo(struct net_device *net,
926 			       struct ethtool_drvinfo *info)
927 {
928 	strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
929 	strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
930 }
931 
932 static void netvsc_get_channels(struct net_device *net,
933 				struct ethtool_channels *channel)
934 {
935 	struct net_device_context *net_device_ctx = netdev_priv(net);
936 	struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
937 
938 	if (nvdev) {
939 		channel->max_combined	= nvdev->max_chn;
940 		channel->combined_count = nvdev->num_chn;
941 	}
942 }
943 
944 /* Alloc struct netvsc_device_info, and initialize it from either existing
945  * struct netvsc_device, or from default values.
946  */
947 static
948 struct netvsc_device_info *netvsc_devinfo_get(struct netvsc_device *nvdev)
949 {
950 	struct netvsc_device_info *dev_info;
951 	struct bpf_prog *prog;
952 
953 	dev_info = kzalloc(sizeof(*dev_info), GFP_ATOMIC);
954 
955 	if (!dev_info)
956 		return NULL;
957 
958 	if (nvdev) {
959 		ASSERT_RTNL();
960 
961 		dev_info->num_chn = nvdev->num_chn;
962 		dev_info->send_sections = nvdev->send_section_cnt;
963 		dev_info->send_section_size = nvdev->send_section_size;
964 		dev_info->recv_sections = nvdev->recv_section_cnt;
965 		dev_info->recv_section_size = nvdev->recv_section_size;
966 
967 		memcpy(dev_info->rss_key, nvdev->extension->rss_key,
968 		       NETVSC_HASH_KEYLEN);
969 
970 		prog = netvsc_xdp_get(nvdev);
971 		if (prog) {
972 			bpf_prog_inc(prog);
973 			dev_info->bprog = prog;
974 		}
975 	} else {
976 		dev_info->num_chn = VRSS_CHANNEL_DEFAULT;
977 		dev_info->send_sections = NETVSC_DEFAULT_TX;
978 		dev_info->send_section_size = NETVSC_SEND_SECTION_SIZE;
979 		dev_info->recv_sections = NETVSC_DEFAULT_RX;
980 		dev_info->recv_section_size = NETVSC_RECV_SECTION_SIZE;
981 	}
982 
983 	return dev_info;
984 }
985 
986 /* Free struct netvsc_device_info */
987 static void netvsc_devinfo_put(struct netvsc_device_info *dev_info)
988 {
989 	if (dev_info->bprog) {
990 		ASSERT_RTNL();
991 		bpf_prog_put(dev_info->bprog);
992 	}
993 
994 	kfree(dev_info);
995 }
996 
997 static int netvsc_detach(struct net_device *ndev,
998 			 struct netvsc_device *nvdev)
999 {
1000 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
1001 	struct hv_device *hdev = ndev_ctx->device_ctx;
1002 	int ret;
1003 
1004 	/* Don't try continuing to try and setup sub channels */
1005 	if (cancel_work_sync(&nvdev->subchan_work))
1006 		nvdev->num_chn = 1;
1007 
1008 	netvsc_xdp_set(ndev, NULL, NULL, nvdev);
1009 
1010 	/* If device was up (receiving) then shutdown */
1011 	if (netif_running(ndev)) {
1012 		netvsc_tx_disable(nvdev, ndev);
1013 
1014 		ret = rndis_filter_close(nvdev);
1015 		if (ret) {
1016 			netdev_err(ndev,
1017 				   "unable to close device (ret %d).\n", ret);
1018 			return ret;
1019 		}
1020 
1021 		ret = netvsc_wait_until_empty(nvdev);
1022 		if (ret) {
1023 			netdev_err(ndev,
1024 				   "Ring buffer not empty after closing rndis\n");
1025 			return ret;
1026 		}
1027 	}
1028 
1029 	netif_device_detach(ndev);
1030 
1031 	rndis_filter_device_remove(hdev, nvdev);
1032 
1033 	return 0;
1034 }
1035 
1036 static int netvsc_attach(struct net_device *ndev,
1037 			 struct netvsc_device_info *dev_info)
1038 {
1039 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
1040 	struct hv_device *hdev = ndev_ctx->device_ctx;
1041 	struct netvsc_device *nvdev;
1042 	struct rndis_device *rdev;
1043 	struct bpf_prog *prog;
1044 	int ret = 0;
1045 
1046 	nvdev = rndis_filter_device_add(hdev, dev_info);
1047 	if (IS_ERR(nvdev))
1048 		return PTR_ERR(nvdev);
1049 
1050 	if (nvdev->num_chn > 1) {
1051 		ret = rndis_set_subchannel(ndev, nvdev, dev_info);
1052 
1053 		/* if unavailable, just proceed with one queue */
1054 		if (ret) {
1055 			nvdev->max_chn = 1;
1056 			nvdev->num_chn = 1;
1057 		}
1058 	}
1059 
1060 	prog = dev_info->bprog;
1061 	if (prog) {
1062 		ret = netvsc_xdp_set(ndev, prog, NULL, nvdev);
1063 		if (ret)
1064 			goto err1;
1065 	}
1066 
1067 	/* In any case device is now ready */
1068 	netif_device_attach(ndev);
1069 
1070 	/* Note: enable and attach happen when sub-channels setup */
1071 	netif_carrier_off(ndev);
1072 
1073 	if (netif_running(ndev)) {
1074 		ret = rndis_filter_open(nvdev);
1075 		if (ret)
1076 			goto err2;
1077 
1078 		rdev = nvdev->extension;
1079 		if (!rdev->link_state)
1080 			netif_carrier_on(ndev);
1081 	}
1082 
1083 	return 0;
1084 
1085 err2:
1086 	netif_device_detach(ndev);
1087 
1088 err1:
1089 	rndis_filter_device_remove(hdev, nvdev);
1090 
1091 	return ret;
1092 }
1093 
1094 static int netvsc_set_channels(struct net_device *net,
1095 			       struct ethtool_channels *channels)
1096 {
1097 	struct net_device_context *net_device_ctx = netdev_priv(net);
1098 	struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
1099 	unsigned int orig, count = channels->combined_count;
1100 	struct netvsc_device_info *device_info;
1101 	int ret;
1102 
1103 	/* We do not support separate count for rx, tx, or other */
1104 	if (count == 0 ||
1105 	    channels->rx_count || channels->tx_count || channels->other_count)
1106 		return -EINVAL;
1107 
1108 	if (!nvdev || nvdev->destroy)
1109 		return -ENODEV;
1110 
1111 	if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
1112 		return -EINVAL;
1113 
1114 	if (count > nvdev->max_chn)
1115 		return -EINVAL;
1116 
1117 	orig = nvdev->num_chn;
1118 
1119 	device_info = netvsc_devinfo_get(nvdev);
1120 
1121 	if (!device_info)
1122 		return -ENOMEM;
1123 
1124 	device_info->num_chn = count;
1125 
1126 	ret = netvsc_detach(net, nvdev);
1127 	if (ret)
1128 		goto out;
1129 
1130 	ret = netvsc_attach(net, device_info);
1131 	if (ret) {
1132 		device_info->num_chn = orig;
1133 		if (netvsc_attach(net, device_info))
1134 			netdev_err(net, "restoring channel setting failed\n");
1135 	}
1136 
1137 out:
1138 	netvsc_devinfo_put(device_info);
1139 	return ret;
1140 }
1141 
1142 static bool
1143 netvsc_validate_ethtool_ss_cmd(const struct ethtool_link_ksettings *cmd)
1144 {
1145 	struct ethtool_link_ksettings diff1 = *cmd;
1146 	struct ethtool_link_ksettings diff2 = {};
1147 
1148 	diff1.base.speed = 0;
1149 	diff1.base.duplex = 0;
1150 	/* advertising and cmd are usually set */
1151 	ethtool_link_ksettings_zero_link_mode(&diff1, advertising);
1152 	diff1.base.cmd = 0;
1153 	/* We set port to PORT_OTHER */
1154 	diff2.base.port = PORT_OTHER;
1155 
1156 	return !memcmp(&diff1, &diff2, sizeof(diff1));
1157 }
1158 
1159 static void netvsc_init_settings(struct net_device *dev)
1160 {
1161 	struct net_device_context *ndc = netdev_priv(dev);
1162 
1163 	ndc->l4_hash = HV_DEFAULT_L4HASH;
1164 
1165 	ndc->speed = SPEED_UNKNOWN;
1166 	ndc->duplex = DUPLEX_FULL;
1167 
1168 	dev->features = NETIF_F_LRO;
1169 }
1170 
1171 static int netvsc_get_link_ksettings(struct net_device *dev,
1172 				     struct ethtool_link_ksettings *cmd)
1173 {
1174 	struct net_device_context *ndc = netdev_priv(dev);
1175 
1176 	cmd->base.speed = ndc->speed;
1177 	cmd->base.duplex = ndc->duplex;
1178 	cmd->base.port = PORT_OTHER;
1179 
1180 	return 0;
1181 }
1182 
1183 static int netvsc_set_link_ksettings(struct net_device *dev,
1184 				     const struct ethtool_link_ksettings *cmd)
1185 {
1186 	struct net_device_context *ndc = netdev_priv(dev);
1187 	u32 speed;
1188 
1189 	speed = cmd->base.speed;
1190 	if (!ethtool_validate_speed(speed) ||
1191 	    !ethtool_validate_duplex(cmd->base.duplex) ||
1192 	    !netvsc_validate_ethtool_ss_cmd(cmd))
1193 		return -EINVAL;
1194 
1195 	ndc->speed = speed;
1196 	ndc->duplex = cmd->base.duplex;
1197 
1198 	return 0;
1199 }
1200 
1201 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
1202 {
1203 	struct net_device_context *ndevctx = netdev_priv(ndev);
1204 	struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1205 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1206 	int orig_mtu = ndev->mtu;
1207 	struct netvsc_device_info *device_info;
1208 	int ret = 0;
1209 
1210 	if (!nvdev || nvdev->destroy)
1211 		return -ENODEV;
1212 
1213 	device_info = netvsc_devinfo_get(nvdev);
1214 
1215 	if (!device_info)
1216 		return -ENOMEM;
1217 
1218 	/* Change MTU of underlying VF netdev first. */
1219 	if (vf_netdev) {
1220 		ret = dev_set_mtu(vf_netdev, mtu);
1221 		if (ret)
1222 			goto out;
1223 	}
1224 
1225 	ret = netvsc_detach(ndev, nvdev);
1226 	if (ret)
1227 		goto rollback_vf;
1228 
1229 	ndev->mtu = mtu;
1230 
1231 	ret = netvsc_attach(ndev, device_info);
1232 	if (!ret)
1233 		goto out;
1234 
1235 	/* Attempt rollback to original MTU */
1236 	ndev->mtu = orig_mtu;
1237 
1238 	if (netvsc_attach(ndev, device_info))
1239 		netdev_err(ndev, "restoring mtu failed\n");
1240 rollback_vf:
1241 	if (vf_netdev)
1242 		dev_set_mtu(vf_netdev, orig_mtu);
1243 
1244 out:
1245 	netvsc_devinfo_put(device_info);
1246 	return ret;
1247 }
1248 
1249 static void netvsc_get_vf_stats(struct net_device *net,
1250 				struct netvsc_vf_pcpu_stats *tot)
1251 {
1252 	struct net_device_context *ndev_ctx = netdev_priv(net);
1253 	int i;
1254 
1255 	memset(tot, 0, sizeof(*tot));
1256 
1257 	for_each_possible_cpu(i) {
1258 		const struct netvsc_vf_pcpu_stats *stats
1259 			= per_cpu_ptr(ndev_ctx->vf_stats, i);
1260 		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1261 		unsigned int start;
1262 
1263 		do {
1264 			start = u64_stats_fetch_begin_irq(&stats->syncp);
1265 			rx_packets = stats->rx_packets;
1266 			tx_packets = stats->tx_packets;
1267 			rx_bytes = stats->rx_bytes;
1268 			tx_bytes = stats->tx_bytes;
1269 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1270 
1271 		tot->rx_packets += rx_packets;
1272 		tot->tx_packets += tx_packets;
1273 		tot->rx_bytes   += rx_bytes;
1274 		tot->tx_bytes   += tx_bytes;
1275 		tot->tx_dropped += stats->tx_dropped;
1276 	}
1277 }
1278 
1279 static void netvsc_get_pcpu_stats(struct net_device *net,
1280 				  struct netvsc_ethtool_pcpu_stats *pcpu_tot)
1281 {
1282 	struct net_device_context *ndev_ctx = netdev_priv(net);
1283 	struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1284 	int i;
1285 
1286 	/* fetch percpu stats of vf */
1287 	for_each_possible_cpu(i) {
1288 		const struct netvsc_vf_pcpu_stats *stats =
1289 			per_cpu_ptr(ndev_ctx->vf_stats, i);
1290 		struct netvsc_ethtool_pcpu_stats *this_tot = &pcpu_tot[i];
1291 		unsigned int start;
1292 
1293 		do {
1294 			start = u64_stats_fetch_begin_irq(&stats->syncp);
1295 			this_tot->vf_rx_packets = stats->rx_packets;
1296 			this_tot->vf_tx_packets = stats->tx_packets;
1297 			this_tot->vf_rx_bytes = stats->rx_bytes;
1298 			this_tot->vf_tx_bytes = stats->tx_bytes;
1299 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1300 		this_tot->rx_packets = this_tot->vf_rx_packets;
1301 		this_tot->tx_packets = this_tot->vf_tx_packets;
1302 		this_tot->rx_bytes   = this_tot->vf_rx_bytes;
1303 		this_tot->tx_bytes   = this_tot->vf_tx_bytes;
1304 	}
1305 
1306 	/* fetch percpu stats of netvsc */
1307 	for (i = 0; i < nvdev->num_chn; i++) {
1308 		const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1309 		const struct netvsc_stats *stats;
1310 		struct netvsc_ethtool_pcpu_stats *this_tot =
1311 			&pcpu_tot[nvchan->channel->target_cpu];
1312 		u64 packets, bytes;
1313 		unsigned int start;
1314 
1315 		stats = &nvchan->tx_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->tx_bytes	+= bytes;
1323 		this_tot->tx_packets	+= packets;
1324 
1325 		stats = &nvchan->rx_stats;
1326 		do {
1327 			start = u64_stats_fetch_begin_irq(&stats->syncp);
1328 			packets = stats->packets;
1329 			bytes = stats->bytes;
1330 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1331 
1332 		this_tot->rx_bytes	+= bytes;
1333 		this_tot->rx_packets	+= packets;
1334 	}
1335 }
1336 
1337 static void netvsc_get_stats64(struct net_device *net,
1338 			       struct rtnl_link_stats64 *t)
1339 {
1340 	struct net_device_context *ndev_ctx = netdev_priv(net);
1341 	struct netvsc_device *nvdev;
1342 	struct netvsc_vf_pcpu_stats vf_tot;
1343 	int i;
1344 
1345 	rcu_read_lock();
1346 
1347 	nvdev = rcu_dereference(ndev_ctx->nvdev);
1348 	if (!nvdev)
1349 		goto out;
1350 
1351 	netdev_stats_to_stats64(t, &net->stats);
1352 
1353 	netvsc_get_vf_stats(net, &vf_tot);
1354 	t->rx_packets += vf_tot.rx_packets;
1355 	t->tx_packets += vf_tot.tx_packets;
1356 	t->rx_bytes   += vf_tot.rx_bytes;
1357 	t->tx_bytes   += vf_tot.tx_bytes;
1358 	t->tx_dropped += vf_tot.tx_dropped;
1359 
1360 	for (i = 0; i < nvdev->num_chn; i++) {
1361 		const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1362 		const struct netvsc_stats *stats;
1363 		u64 packets, bytes, multicast;
1364 		unsigned int start;
1365 
1366 		stats = &nvchan->tx_stats;
1367 		do {
1368 			start = u64_stats_fetch_begin_irq(&stats->syncp);
1369 			packets = stats->packets;
1370 			bytes = stats->bytes;
1371 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1372 
1373 		t->tx_bytes	+= bytes;
1374 		t->tx_packets	+= packets;
1375 
1376 		stats = &nvchan->rx_stats;
1377 		do {
1378 			start = u64_stats_fetch_begin_irq(&stats->syncp);
1379 			packets = stats->packets;
1380 			bytes = stats->bytes;
1381 			multicast = stats->multicast + stats->broadcast;
1382 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1383 
1384 		t->rx_bytes	+= bytes;
1385 		t->rx_packets	+= packets;
1386 		t->multicast	+= multicast;
1387 	}
1388 out:
1389 	rcu_read_unlock();
1390 }
1391 
1392 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1393 {
1394 	struct net_device_context *ndc = netdev_priv(ndev);
1395 	struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1396 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1397 	struct sockaddr *addr = p;
1398 	int err;
1399 
1400 	err = eth_prepare_mac_addr_change(ndev, p);
1401 	if (err)
1402 		return err;
1403 
1404 	if (!nvdev)
1405 		return -ENODEV;
1406 
1407 	if (vf_netdev) {
1408 		err = dev_set_mac_address(vf_netdev, addr, NULL);
1409 		if (err)
1410 			return err;
1411 	}
1412 
1413 	err = rndis_filter_set_device_mac(nvdev, addr->sa_data);
1414 	if (!err) {
1415 		eth_commit_mac_addr_change(ndev, p);
1416 	} else if (vf_netdev) {
1417 		/* rollback change on VF */
1418 		memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN);
1419 		dev_set_mac_address(vf_netdev, addr, NULL);
1420 	}
1421 
1422 	return err;
1423 }
1424 
1425 static const struct {
1426 	char name[ETH_GSTRING_LEN];
1427 	u16 offset;
1428 } netvsc_stats[] = {
1429 	{ "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1430 	{ "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1431 	{ "tx_no_space",  offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1432 	{ "tx_too_big",	  offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1433 	{ "tx_busy",	  offsetof(struct netvsc_ethtool_stats, tx_busy) },
1434 	{ "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) },
1435 	{ "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) },
1436 	{ "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) },
1437 	{ "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) },
1438 	{ "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) },
1439 }, pcpu_stats[] = {
1440 	{ "cpu%u_rx_packets",
1441 		offsetof(struct netvsc_ethtool_pcpu_stats, rx_packets) },
1442 	{ "cpu%u_rx_bytes",
1443 		offsetof(struct netvsc_ethtool_pcpu_stats, rx_bytes) },
1444 	{ "cpu%u_tx_packets",
1445 		offsetof(struct netvsc_ethtool_pcpu_stats, tx_packets) },
1446 	{ "cpu%u_tx_bytes",
1447 		offsetof(struct netvsc_ethtool_pcpu_stats, tx_bytes) },
1448 	{ "cpu%u_vf_rx_packets",
1449 		offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_packets) },
1450 	{ "cpu%u_vf_rx_bytes",
1451 		offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_bytes) },
1452 	{ "cpu%u_vf_tx_packets",
1453 		offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_packets) },
1454 	{ "cpu%u_vf_tx_bytes",
1455 		offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_bytes) },
1456 }, vf_stats[] = {
1457 	{ "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) },
1458 	{ "vf_rx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) },
1459 	{ "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) },
1460 	{ "vf_tx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) },
1461 	{ "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) },
1462 };
1463 
1464 #define NETVSC_GLOBAL_STATS_LEN	ARRAY_SIZE(netvsc_stats)
1465 #define NETVSC_VF_STATS_LEN	ARRAY_SIZE(vf_stats)
1466 
1467 /* statistics per queue (rx/tx packets/bytes) */
1468 #define NETVSC_PCPU_STATS_LEN (num_present_cpus() * ARRAY_SIZE(pcpu_stats))
1469 
1470 /* 5 statistics per queue (rx/tx packets/bytes, rx xdp_drop) */
1471 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 5)
1472 
1473 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1474 {
1475 	struct net_device_context *ndc = netdev_priv(dev);
1476 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1477 
1478 	if (!nvdev)
1479 		return -ENODEV;
1480 
1481 	switch (string_set) {
1482 	case ETH_SS_STATS:
1483 		return NETVSC_GLOBAL_STATS_LEN
1484 			+ NETVSC_VF_STATS_LEN
1485 			+ NETVSC_QUEUE_STATS_LEN(nvdev)
1486 			+ NETVSC_PCPU_STATS_LEN;
1487 	default:
1488 		return -EINVAL;
1489 	}
1490 }
1491 
1492 static void netvsc_get_ethtool_stats(struct net_device *dev,
1493 				     struct ethtool_stats *stats, u64 *data)
1494 {
1495 	struct net_device_context *ndc = netdev_priv(dev);
1496 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1497 	const void *nds = &ndc->eth_stats;
1498 	const struct netvsc_stats *qstats;
1499 	struct netvsc_vf_pcpu_stats sum;
1500 	struct netvsc_ethtool_pcpu_stats *pcpu_sum;
1501 	unsigned int start;
1502 	u64 packets, bytes;
1503 	u64 xdp_drop;
1504 	int i, j, cpu;
1505 
1506 	if (!nvdev)
1507 		return;
1508 
1509 	for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1510 		data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1511 
1512 	netvsc_get_vf_stats(dev, &sum);
1513 	for (j = 0; j < NETVSC_VF_STATS_LEN; j++)
1514 		data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset);
1515 
1516 	for (j = 0; j < nvdev->num_chn; j++) {
1517 		qstats = &nvdev->chan_table[j].tx_stats;
1518 
1519 		do {
1520 			start = u64_stats_fetch_begin_irq(&qstats->syncp);
1521 			packets = qstats->packets;
1522 			bytes = qstats->bytes;
1523 		} while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1524 		data[i++] = packets;
1525 		data[i++] = bytes;
1526 
1527 		qstats = &nvdev->chan_table[j].rx_stats;
1528 		do {
1529 			start = u64_stats_fetch_begin_irq(&qstats->syncp);
1530 			packets = qstats->packets;
1531 			bytes = qstats->bytes;
1532 			xdp_drop = qstats->xdp_drop;
1533 		} while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1534 		data[i++] = packets;
1535 		data[i++] = bytes;
1536 		data[i++] = xdp_drop;
1537 	}
1538 
1539 	pcpu_sum = kvmalloc_array(num_possible_cpus(),
1540 				  sizeof(struct netvsc_ethtool_pcpu_stats),
1541 				  GFP_KERNEL);
1542 	netvsc_get_pcpu_stats(dev, pcpu_sum);
1543 	for_each_present_cpu(cpu) {
1544 		struct netvsc_ethtool_pcpu_stats *this_sum = &pcpu_sum[cpu];
1545 
1546 		for (j = 0; j < ARRAY_SIZE(pcpu_stats); j++)
1547 			data[i++] = *(u64 *)((void *)this_sum
1548 					     + pcpu_stats[j].offset);
1549 	}
1550 	kvfree(pcpu_sum);
1551 }
1552 
1553 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1554 {
1555 	struct net_device_context *ndc = netdev_priv(dev);
1556 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1557 	u8 *p = data;
1558 	int i, cpu;
1559 
1560 	if (!nvdev)
1561 		return;
1562 
1563 	switch (stringset) {
1564 	case ETH_SS_STATS:
1565 		for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++) {
1566 			memcpy(p, netvsc_stats[i].name, ETH_GSTRING_LEN);
1567 			p += ETH_GSTRING_LEN;
1568 		}
1569 
1570 		for (i = 0; i < ARRAY_SIZE(vf_stats); i++) {
1571 			memcpy(p, vf_stats[i].name, ETH_GSTRING_LEN);
1572 			p += ETH_GSTRING_LEN;
1573 		}
1574 
1575 		for (i = 0; i < nvdev->num_chn; i++) {
1576 			sprintf(p, "tx_queue_%u_packets", i);
1577 			p += ETH_GSTRING_LEN;
1578 			sprintf(p, "tx_queue_%u_bytes", i);
1579 			p += ETH_GSTRING_LEN;
1580 			sprintf(p, "rx_queue_%u_packets", i);
1581 			p += ETH_GSTRING_LEN;
1582 			sprintf(p, "rx_queue_%u_bytes", i);
1583 			p += ETH_GSTRING_LEN;
1584 			sprintf(p, "rx_queue_%u_xdp_drop", i);
1585 			p += ETH_GSTRING_LEN;
1586 		}
1587 
1588 		for_each_present_cpu(cpu) {
1589 			for (i = 0; i < ARRAY_SIZE(pcpu_stats); i++) {
1590 				sprintf(p, pcpu_stats[i].name, cpu);
1591 				p += ETH_GSTRING_LEN;
1592 			}
1593 		}
1594 
1595 		break;
1596 	}
1597 }
1598 
1599 static int
1600 netvsc_get_rss_hash_opts(struct net_device_context *ndc,
1601 			 struct ethtool_rxnfc *info)
1602 {
1603 	const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3;
1604 
1605 	info->data = RXH_IP_SRC | RXH_IP_DST;
1606 
1607 	switch (info->flow_type) {
1608 	case TCP_V4_FLOW:
1609 		if (ndc->l4_hash & HV_TCP4_L4HASH)
1610 			info->data |= l4_flag;
1611 
1612 		break;
1613 
1614 	case TCP_V6_FLOW:
1615 		if (ndc->l4_hash & HV_TCP6_L4HASH)
1616 			info->data |= l4_flag;
1617 
1618 		break;
1619 
1620 	case UDP_V4_FLOW:
1621 		if (ndc->l4_hash & HV_UDP4_L4HASH)
1622 			info->data |= l4_flag;
1623 
1624 		break;
1625 
1626 	case UDP_V6_FLOW:
1627 		if (ndc->l4_hash & HV_UDP6_L4HASH)
1628 			info->data |= l4_flag;
1629 
1630 		break;
1631 
1632 	case IPV4_FLOW:
1633 	case IPV6_FLOW:
1634 		break;
1635 	default:
1636 		info->data = 0;
1637 		break;
1638 	}
1639 
1640 	return 0;
1641 }
1642 
1643 static int
1644 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1645 		 u32 *rules)
1646 {
1647 	struct net_device_context *ndc = netdev_priv(dev);
1648 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1649 
1650 	if (!nvdev)
1651 		return -ENODEV;
1652 
1653 	switch (info->cmd) {
1654 	case ETHTOOL_GRXRINGS:
1655 		info->data = nvdev->num_chn;
1656 		return 0;
1657 
1658 	case ETHTOOL_GRXFH:
1659 		return netvsc_get_rss_hash_opts(ndc, info);
1660 	}
1661 	return -EOPNOTSUPP;
1662 }
1663 
1664 static int netvsc_set_rss_hash_opts(struct net_device_context *ndc,
1665 				    struct ethtool_rxnfc *info)
1666 {
1667 	if (info->data == (RXH_IP_SRC | RXH_IP_DST |
1668 			   RXH_L4_B_0_1 | RXH_L4_B_2_3)) {
1669 		switch (info->flow_type) {
1670 		case TCP_V4_FLOW:
1671 			ndc->l4_hash |= HV_TCP4_L4HASH;
1672 			break;
1673 
1674 		case TCP_V6_FLOW:
1675 			ndc->l4_hash |= HV_TCP6_L4HASH;
1676 			break;
1677 
1678 		case UDP_V4_FLOW:
1679 			ndc->l4_hash |= HV_UDP4_L4HASH;
1680 			break;
1681 
1682 		case UDP_V6_FLOW:
1683 			ndc->l4_hash |= HV_UDP6_L4HASH;
1684 			break;
1685 
1686 		default:
1687 			return -EOPNOTSUPP;
1688 		}
1689 
1690 		return 0;
1691 	}
1692 
1693 	if (info->data == (RXH_IP_SRC | RXH_IP_DST)) {
1694 		switch (info->flow_type) {
1695 		case TCP_V4_FLOW:
1696 			ndc->l4_hash &= ~HV_TCP4_L4HASH;
1697 			break;
1698 
1699 		case TCP_V6_FLOW:
1700 			ndc->l4_hash &= ~HV_TCP6_L4HASH;
1701 			break;
1702 
1703 		case UDP_V4_FLOW:
1704 			ndc->l4_hash &= ~HV_UDP4_L4HASH;
1705 			break;
1706 
1707 		case UDP_V6_FLOW:
1708 			ndc->l4_hash &= ~HV_UDP6_L4HASH;
1709 			break;
1710 
1711 		default:
1712 			return -EOPNOTSUPP;
1713 		}
1714 
1715 		return 0;
1716 	}
1717 
1718 	return -EOPNOTSUPP;
1719 }
1720 
1721 static int
1722 netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info)
1723 {
1724 	struct net_device_context *ndc = netdev_priv(ndev);
1725 
1726 	if (info->cmd == ETHTOOL_SRXFH)
1727 		return netvsc_set_rss_hash_opts(ndc, info);
1728 
1729 	return -EOPNOTSUPP;
1730 }
1731 
1732 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1733 {
1734 	return NETVSC_HASH_KEYLEN;
1735 }
1736 
1737 static u32 netvsc_rss_indir_size(struct net_device *dev)
1738 {
1739 	return ITAB_NUM;
1740 }
1741 
1742 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1743 			   u8 *hfunc)
1744 {
1745 	struct net_device_context *ndc = netdev_priv(dev);
1746 	struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1747 	struct rndis_device *rndis_dev;
1748 	int i;
1749 
1750 	if (!ndev)
1751 		return -ENODEV;
1752 
1753 	if (hfunc)
1754 		*hfunc = ETH_RSS_HASH_TOP;	/* Toeplitz */
1755 
1756 	rndis_dev = ndev->extension;
1757 	if (indir) {
1758 		for (i = 0; i < ITAB_NUM; i++)
1759 			indir[i] = ndc->rx_table[i];
1760 	}
1761 
1762 	if (key)
1763 		memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1764 
1765 	return 0;
1766 }
1767 
1768 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1769 			   const u8 *key, const u8 hfunc)
1770 {
1771 	struct net_device_context *ndc = netdev_priv(dev);
1772 	struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1773 	struct rndis_device *rndis_dev;
1774 	int i;
1775 
1776 	if (!ndev)
1777 		return -ENODEV;
1778 
1779 	if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1780 		return -EOPNOTSUPP;
1781 
1782 	rndis_dev = ndev->extension;
1783 	if (indir) {
1784 		for (i = 0; i < ITAB_NUM; i++)
1785 			if (indir[i] >= ndev->num_chn)
1786 				return -EINVAL;
1787 
1788 		for (i = 0; i < ITAB_NUM; i++)
1789 			ndc->rx_table[i] = indir[i];
1790 	}
1791 
1792 	if (!key) {
1793 		if (!indir)
1794 			return 0;
1795 
1796 		key = rndis_dev->rss_key;
1797 	}
1798 
1799 	return rndis_filter_set_rss_param(rndis_dev, key);
1800 }
1801 
1802 /* Hyper-V RNDIS protocol does not have ring in the HW sense.
1803  * It does have pre-allocated receive area which is divided into sections.
1804  */
1805 static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
1806 				   struct ethtool_ringparam *ring)
1807 {
1808 	u32 max_buf_size;
1809 
1810 	ring->rx_pending = nvdev->recv_section_cnt;
1811 	ring->tx_pending = nvdev->send_section_cnt;
1812 
1813 	if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
1814 		max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
1815 	else
1816 		max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1817 
1818 	ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
1819 	ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
1820 		/ nvdev->send_section_size;
1821 }
1822 
1823 static void netvsc_get_ringparam(struct net_device *ndev,
1824 				 struct ethtool_ringparam *ring)
1825 {
1826 	struct net_device_context *ndevctx = netdev_priv(ndev);
1827 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1828 
1829 	if (!nvdev)
1830 		return;
1831 
1832 	__netvsc_get_ringparam(nvdev, ring);
1833 }
1834 
1835 static int netvsc_set_ringparam(struct net_device *ndev,
1836 				struct ethtool_ringparam *ring)
1837 {
1838 	struct net_device_context *ndevctx = netdev_priv(ndev);
1839 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1840 	struct netvsc_device_info *device_info;
1841 	struct ethtool_ringparam orig;
1842 	u32 new_tx, new_rx;
1843 	int ret = 0;
1844 
1845 	if (!nvdev || nvdev->destroy)
1846 		return -ENODEV;
1847 
1848 	memset(&orig, 0, sizeof(orig));
1849 	__netvsc_get_ringparam(nvdev, &orig);
1850 
1851 	new_tx = clamp_t(u32, ring->tx_pending,
1852 			 NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
1853 	new_rx = clamp_t(u32, ring->rx_pending,
1854 			 NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
1855 
1856 	if (new_tx == orig.tx_pending &&
1857 	    new_rx == orig.rx_pending)
1858 		return 0;	 /* no change */
1859 
1860 	device_info = netvsc_devinfo_get(nvdev);
1861 
1862 	if (!device_info)
1863 		return -ENOMEM;
1864 
1865 	device_info->send_sections = new_tx;
1866 	device_info->recv_sections = new_rx;
1867 
1868 	ret = netvsc_detach(ndev, nvdev);
1869 	if (ret)
1870 		goto out;
1871 
1872 	ret = netvsc_attach(ndev, device_info);
1873 	if (ret) {
1874 		device_info->send_sections = orig.tx_pending;
1875 		device_info->recv_sections = orig.rx_pending;
1876 
1877 		if (netvsc_attach(ndev, device_info))
1878 			netdev_err(ndev, "restoring ringparam failed");
1879 	}
1880 
1881 out:
1882 	netvsc_devinfo_put(device_info);
1883 	return ret;
1884 }
1885 
1886 static netdev_features_t netvsc_fix_features(struct net_device *ndev,
1887 					     netdev_features_t features)
1888 {
1889 	struct net_device_context *ndevctx = netdev_priv(ndev);
1890 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1891 
1892 	if (!nvdev || nvdev->destroy)
1893 		return features;
1894 
1895 	if ((features & NETIF_F_LRO) && netvsc_xdp_get(nvdev)) {
1896 		features ^= NETIF_F_LRO;
1897 		netdev_info(ndev, "Skip LRO - unsupported with XDP\n");
1898 	}
1899 
1900 	return features;
1901 }
1902 
1903 static int netvsc_set_features(struct net_device *ndev,
1904 			       netdev_features_t features)
1905 {
1906 	netdev_features_t change = features ^ ndev->features;
1907 	struct net_device_context *ndevctx = netdev_priv(ndev);
1908 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1909 	struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1910 	struct ndis_offload_params offloads;
1911 	int ret = 0;
1912 
1913 	if (!nvdev || nvdev->destroy)
1914 		return -ENODEV;
1915 
1916 	if (!(change & NETIF_F_LRO))
1917 		goto syncvf;
1918 
1919 	memset(&offloads, 0, sizeof(struct ndis_offload_params));
1920 
1921 	if (features & NETIF_F_LRO) {
1922 		offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1923 		offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1924 	} else {
1925 		offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1926 		offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1927 	}
1928 
1929 	ret = rndis_filter_set_offload_params(ndev, nvdev, &offloads);
1930 
1931 	if (ret) {
1932 		features ^= NETIF_F_LRO;
1933 		ndev->features = features;
1934 	}
1935 
1936 syncvf:
1937 	if (!vf_netdev)
1938 		return ret;
1939 
1940 	vf_netdev->wanted_features = features;
1941 	netdev_update_features(vf_netdev);
1942 
1943 	return ret;
1944 }
1945 
1946 static u32 netvsc_get_msglevel(struct net_device *ndev)
1947 {
1948 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
1949 
1950 	return ndev_ctx->msg_enable;
1951 }
1952 
1953 static void netvsc_set_msglevel(struct net_device *ndev, u32 val)
1954 {
1955 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
1956 
1957 	ndev_ctx->msg_enable = val;
1958 }
1959 
1960 static const struct ethtool_ops ethtool_ops = {
1961 	.get_drvinfo	= netvsc_get_drvinfo,
1962 	.get_msglevel	= netvsc_get_msglevel,
1963 	.set_msglevel	= netvsc_set_msglevel,
1964 	.get_link	= ethtool_op_get_link,
1965 	.get_ethtool_stats = netvsc_get_ethtool_stats,
1966 	.get_sset_count = netvsc_get_sset_count,
1967 	.get_strings	= netvsc_get_strings,
1968 	.get_channels   = netvsc_get_channels,
1969 	.set_channels   = netvsc_set_channels,
1970 	.get_ts_info	= ethtool_op_get_ts_info,
1971 	.get_rxnfc	= netvsc_get_rxnfc,
1972 	.set_rxnfc	= netvsc_set_rxnfc,
1973 	.get_rxfh_key_size = netvsc_get_rxfh_key_size,
1974 	.get_rxfh_indir_size = netvsc_rss_indir_size,
1975 	.get_rxfh	= netvsc_get_rxfh,
1976 	.set_rxfh	= netvsc_set_rxfh,
1977 	.get_link_ksettings = netvsc_get_link_ksettings,
1978 	.set_link_ksettings = netvsc_set_link_ksettings,
1979 	.get_ringparam	= netvsc_get_ringparam,
1980 	.set_ringparam	= netvsc_set_ringparam,
1981 };
1982 
1983 static const struct net_device_ops device_ops = {
1984 	.ndo_open =			netvsc_open,
1985 	.ndo_stop =			netvsc_close,
1986 	.ndo_start_xmit =		netvsc_start_xmit,
1987 	.ndo_change_rx_flags =		netvsc_change_rx_flags,
1988 	.ndo_set_rx_mode =		netvsc_set_rx_mode,
1989 	.ndo_fix_features =		netvsc_fix_features,
1990 	.ndo_set_features =		netvsc_set_features,
1991 	.ndo_change_mtu =		netvsc_change_mtu,
1992 	.ndo_validate_addr =		eth_validate_addr,
1993 	.ndo_set_mac_address =		netvsc_set_mac_addr,
1994 	.ndo_select_queue =		netvsc_select_queue,
1995 	.ndo_get_stats64 =		netvsc_get_stats64,
1996 	.ndo_bpf =			netvsc_bpf,
1997 };
1998 
1999 /*
2000  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
2001  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
2002  * present send GARP packet to network peers with netif_notify_peers().
2003  */
2004 static void netvsc_link_change(struct work_struct *w)
2005 {
2006 	struct net_device_context *ndev_ctx =
2007 		container_of(w, struct net_device_context, dwork.work);
2008 	struct hv_device *device_obj = ndev_ctx->device_ctx;
2009 	struct net_device *net = hv_get_drvdata(device_obj);
2010 	struct netvsc_device *net_device;
2011 	struct rndis_device *rdev;
2012 	struct netvsc_reconfig *event = NULL;
2013 	bool notify = false, reschedule = false;
2014 	unsigned long flags, next_reconfig, delay;
2015 
2016 	/* if changes are happening, comeback later */
2017 	if (!rtnl_trylock()) {
2018 		schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
2019 		return;
2020 	}
2021 
2022 	net_device = rtnl_dereference(ndev_ctx->nvdev);
2023 	if (!net_device)
2024 		goto out_unlock;
2025 
2026 	rdev = net_device->extension;
2027 
2028 	next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
2029 	if (time_is_after_jiffies(next_reconfig)) {
2030 		/* link_watch only sends one notification with current state
2031 		 * per second, avoid doing reconfig more frequently. Handle
2032 		 * wrap around.
2033 		 */
2034 		delay = next_reconfig - jiffies;
2035 		delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
2036 		schedule_delayed_work(&ndev_ctx->dwork, delay);
2037 		goto out_unlock;
2038 	}
2039 	ndev_ctx->last_reconfig = jiffies;
2040 
2041 	spin_lock_irqsave(&ndev_ctx->lock, flags);
2042 	if (!list_empty(&ndev_ctx->reconfig_events)) {
2043 		event = list_first_entry(&ndev_ctx->reconfig_events,
2044 					 struct netvsc_reconfig, list);
2045 		list_del(&event->list);
2046 		reschedule = !list_empty(&ndev_ctx->reconfig_events);
2047 	}
2048 	spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2049 
2050 	if (!event)
2051 		goto out_unlock;
2052 
2053 	switch (event->event) {
2054 		/* Only the following events are possible due to the check in
2055 		 * netvsc_linkstatus_callback()
2056 		 */
2057 	case RNDIS_STATUS_MEDIA_CONNECT:
2058 		if (rdev->link_state) {
2059 			rdev->link_state = false;
2060 			netif_carrier_on(net);
2061 			netvsc_tx_enable(net_device, net);
2062 		} else {
2063 			notify = true;
2064 		}
2065 		kfree(event);
2066 		break;
2067 	case RNDIS_STATUS_MEDIA_DISCONNECT:
2068 		if (!rdev->link_state) {
2069 			rdev->link_state = true;
2070 			netif_carrier_off(net);
2071 			netvsc_tx_disable(net_device, net);
2072 		}
2073 		kfree(event);
2074 		break;
2075 	case RNDIS_STATUS_NETWORK_CHANGE:
2076 		/* Only makes sense if carrier is present */
2077 		if (!rdev->link_state) {
2078 			rdev->link_state = true;
2079 			netif_carrier_off(net);
2080 			netvsc_tx_disable(net_device, net);
2081 			event->event = RNDIS_STATUS_MEDIA_CONNECT;
2082 			spin_lock_irqsave(&ndev_ctx->lock, flags);
2083 			list_add(&event->list, &ndev_ctx->reconfig_events);
2084 			spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2085 			reschedule = true;
2086 		}
2087 		break;
2088 	}
2089 
2090 	rtnl_unlock();
2091 
2092 	if (notify)
2093 		netdev_notify_peers(net);
2094 
2095 	/* link_watch only sends one notification with current state per
2096 	 * second, handle next reconfig event in 2 seconds.
2097 	 */
2098 	if (reschedule)
2099 		schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
2100 
2101 	return;
2102 
2103 out_unlock:
2104 	rtnl_unlock();
2105 }
2106 
2107 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
2108 {
2109 	struct net_device_context *net_device_ctx;
2110 	struct net_device *dev;
2111 
2112 	dev = netdev_master_upper_dev_get(vf_netdev);
2113 	if (!dev || dev->netdev_ops != &device_ops)
2114 		return NULL;	/* not a netvsc device */
2115 
2116 	net_device_ctx = netdev_priv(dev);
2117 	if (!rtnl_dereference(net_device_ctx->nvdev))
2118 		return NULL;	/* device is removed */
2119 
2120 	return dev;
2121 }
2122 
2123 /* Called when VF is injecting data into network stack.
2124  * Change the associated network device from VF to netvsc.
2125  * note: already called with rcu_read_lock
2126  */
2127 static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
2128 {
2129 	struct sk_buff *skb = *pskb;
2130 	struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data);
2131 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
2132 	struct netvsc_vf_pcpu_stats *pcpu_stats
2133 		 = this_cpu_ptr(ndev_ctx->vf_stats);
2134 
2135 	skb = skb_share_check(skb, GFP_ATOMIC);
2136 	if (unlikely(!skb))
2137 		return RX_HANDLER_CONSUMED;
2138 
2139 	*pskb = skb;
2140 
2141 	skb->dev = ndev;
2142 
2143 	u64_stats_update_begin(&pcpu_stats->syncp);
2144 	pcpu_stats->rx_packets++;
2145 	pcpu_stats->rx_bytes += skb->len;
2146 	u64_stats_update_end(&pcpu_stats->syncp);
2147 
2148 	return RX_HANDLER_ANOTHER;
2149 }
2150 
2151 static int netvsc_vf_join(struct net_device *vf_netdev,
2152 			  struct net_device *ndev)
2153 {
2154 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
2155 	int ret;
2156 
2157 	ret = netdev_rx_handler_register(vf_netdev,
2158 					 netvsc_vf_handle_frame, ndev);
2159 	if (ret != 0) {
2160 		netdev_err(vf_netdev,
2161 			   "can not register netvsc VF receive handler (err = %d)\n",
2162 			   ret);
2163 		goto rx_handler_failed;
2164 	}
2165 
2166 	ret = netdev_master_upper_dev_link(vf_netdev, ndev,
2167 					   NULL, NULL, NULL);
2168 	if (ret != 0) {
2169 		netdev_err(vf_netdev,
2170 			   "can not set master device %s (err = %d)\n",
2171 			   ndev->name, ret);
2172 		goto upper_link_failed;
2173 	}
2174 
2175 	/* set slave flag before open to prevent IPv6 addrconf */
2176 	vf_netdev->flags |= IFF_SLAVE;
2177 
2178 	schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
2179 
2180 	call_netdevice_notifiers(NETDEV_JOIN, vf_netdev);
2181 
2182 	netdev_info(vf_netdev, "joined to %s\n", ndev->name);
2183 	return 0;
2184 
2185 upper_link_failed:
2186 	netdev_rx_handler_unregister(vf_netdev);
2187 rx_handler_failed:
2188 	return ret;
2189 }
2190 
2191 static void __netvsc_vf_setup(struct net_device *ndev,
2192 			      struct net_device *vf_netdev)
2193 {
2194 	int ret;
2195 
2196 	/* Align MTU of VF with master */
2197 	ret = dev_set_mtu(vf_netdev, ndev->mtu);
2198 	if (ret)
2199 		netdev_warn(vf_netdev,
2200 			    "unable to change mtu to %u\n", ndev->mtu);
2201 
2202 	/* set multicast etc flags on VF */
2203 	dev_change_flags(vf_netdev, ndev->flags | IFF_SLAVE, NULL);
2204 
2205 	/* sync address list from ndev to VF */
2206 	netif_addr_lock_bh(ndev);
2207 	dev_uc_sync(vf_netdev, ndev);
2208 	dev_mc_sync(vf_netdev, ndev);
2209 	netif_addr_unlock_bh(ndev);
2210 
2211 	if (netif_running(ndev)) {
2212 		ret = dev_open(vf_netdev, NULL);
2213 		if (ret)
2214 			netdev_warn(vf_netdev,
2215 				    "unable to open: %d\n", ret);
2216 	}
2217 }
2218 
2219 /* Setup VF as slave of the synthetic device.
2220  * Runs in workqueue to avoid recursion in netlink callbacks.
2221  */
2222 static void netvsc_vf_setup(struct work_struct *w)
2223 {
2224 	struct net_device_context *ndev_ctx
2225 		= container_of(w, struct net_device_context, vf_takeover.work);
2226 	struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2227 	struct net_device *vf_netdev;
2228 
2229 	if (!rtnl_trylock()) {
2230 		schedule_delayed_work(&ndev_ctx->vf_takeover, 0);
2231 		return;
2232 	}
2233 
2234 	vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2235 	if (vf_netdev)
2236 		__netvsc_vf_setup(ndev, vf_netdev);
2237 
2238 	rtnl_unlock();
2239 }
2240 
2241 /* Find netvsc by VF serial number.
2242  * The PCI hyperv controller records the serial number as the slot kobj name.
2243  */
2244 static struct net_device *get_netvsc_byslot(const struct net_device *vf_netdev)
2245 {
2246 	struct device *parent = vf_netdev->dev.parent;
2247 	struct net_device_context *ndev_ctx;
2248 	struct pci_dev *pdev;
2249 	u32 serial;
2250 
2251 	if (!parent || !dev_is_pci(parent))
2252 		return NULL; /* not a PCI device */
2253 
2254 	pdev = to_pci_dev(parent);
2255 	if (!pdev->slot) {
2256 		netdev_notice(vf_netdev, "no PCI slot information\n");
2257 		return NULL;
2258 	}
2259 
2260 	if (kstrtou32(pci_slot_name(pdev->slot), 10, &serial)) {
2261 		netdev_notice(vf_netdev, "Invalid vf serial:%s\n",
2262 			      pci_slot_name(pdev->slot));
2263 		return NULL;
2264 	}
2265 
2266 	list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
2267 		if (!ndev_ctx->vf_alloc)
2268 			continue;
2269 
2270 		if (ndev_ctx->vf_serial == serial)
2271 			return hv_get_drvdata(ndev_ctx->device_ctx);
2272 	}
2273 
2274 	netdev_notice(vf_netdev,
2275 		      "no netdev found for vf serial:%u\n", serial);
2276 	return NULL;
2277 }
2278 
2279 static int netvsc_register_vf(struct net_device *vf_netdev)
2280 {
2281 	struct net_device_context *net_device_ctx;
2282 	struct netvsc_device *netvsc_dev;
2283 	struct bpf_prog *prog;
2284 	struct net_device *ndev;
2285 	int ret;
2286 
2287 	if (vf_netdev->addr_len != ETH_ALEN)
2288 		return NOTIFY_DONE;
2289 
2290 	ndev = get_netvsc_byslot(vf_netdev);
2291 	if (!ndev)
2292 		return NOTIFY_DONE;
2293 
2294 	net_device_ctx = netdev_priv(ndev);
2295 	netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2296 	if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
2297 		return NOTIFY_DONE;
2298 
2299 	/* if synthetic interface is a different namespace,
2300 	 * then move the VF to that namespace; join will be
2301 	 * done again in that context.
2302 	 */
2303 	if (!net_eq(dev_net(ndev), dev_net(vf_netdev))) {
2304 		ret = dev_change_net_namespace(vf_netdev,
2305 					       dev_net(ndev), "eth%d");
2306 		if (ret)
2307 			netdev_err(vf_netdev,
2308 				   "could not move to same namespace as %s: %d\n",
2309 				   ndev->name, ret);
2310 		else
2311 			netdev_info(vf_netdev,
2312 				    "VF moved to namespace with: %s\n",
2313 				    ndev->name);
2314 		return NOTIFY_DONE;
2315 	}
2316 
2317 	netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
2318 
2319 	if (netvsc_vf_join(vf_netdev, ndev) != 0)
2320 		return NOTIFY_DONE;
2321 
2322 	dev_hold(vf_netdev);
2323 	rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
2324 
2325 	vf_netdev->wanted_features = ndev->features;
2326 	netdev_update_features(vf_netdev);
2327 
2328 	prog = netvsc_xdp_get(netvsc_dev);
2329 	netvsc_vf_setxdp(vf_netdev, prog);
2330 
2331 	return NOTIFY_OK;
2332 }
2333 
2334 /* VF up/down change detected, schedule to change data path */
2335 static int netvsc_vf_changed(struct net_device *vf_netdev)
2336 {
2337 	struct net_device_context *net_device_ctx;
2338 	struct netvsc_device *netvsc_dev;
2339 	struct net_device *ndev;
2340 	bool vf_is_up = netif_running(vf_netdev);
2341 
2342 	ndev = get_netvsc_byref(vf_netdev);
2343 	if (!ndev)
2344 		return NOTIFY_DONE;
2345 
2346 	net_device_ctx = netdev_priv(ndev);
2347 	netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2348 	if (!netvsc_dev)
2349 		return NOTIFY_DONE;
2350 
2351 	netvsc_switch_datapath(ndev, vf_is_up);
2352 	netdev_info(ndev, "Data path switched %s VF: %s\n",
2353 		    vf_is_up ? "to" : "from", vf_netdev->name);
2354 
2355 	return NOTIFY_OK;
2356 }
2357 
2358 static int netvsc_unregister_vf(struct net_device *vf_netdev)
2359 {
2360 	struct net_device *ndev;
2361 	struct net_device_context *net_device_ctx;
2362 
2363 	ndev = get_netvsc_byref(vf_netdev);
2364 	if (!ndev)
2365 		return NOTIFY_DONE;
2366 
2367 	net_device_ctx = netdev_priv(ndev);
2368 	cancel_delayed_work_sync(&net_device_ctx->vf_takeover);
2369 
2370 	netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
2371 
2372 	netvsc_vf_setxdp(vf_netdev, NULL);
2373 
2374 	netdev_rx_handler_unregister(vf_netdev);
2375 	netdev_upper_dev_unlink(vf_netdev, ndev);
2376 	RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
2377 	dev_put(vf_netdev);
2378 
2379 	return NOTIFY_OK;
2380 }
2381 
2382 static int netvsc_probe(struct hv_device *dev,
2383 			const struct hv_vmbus_device_id *dev_id)
2384 {
2385 	struct net_device *net = NULL;
2386 	struct net_device_context *net_device_ctx;
2387 	struct netvsc_device_info *device_info = NULL;
2388 	struct netvsc_device *nvdev;
2389 	int ret = -ENOMEM;
2390 
2391 	net = alloc_etherdev_mq(sizeof(struct net_device_context),
2392 				VRSS_CHANNEL_MAX);
2393 	if (!net)
2394 		goto no_net;
2395 
2396 	netif_carrier_off(net);
2397 
2398 	netvsc_init_settings(net);
2399 
2400 	net_device_ctx = netdev_priv(net);
2401 	net_device_ctx->device_ctx = dev;
2402 	net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
2403 	if (netif_msg_probe(net_device_ctx))
2404 		netdev_dbg(net, "netvsc msg_enable: %d\n",
2405 			   net_device_ctx->msg_enable);
2406 
2407 	hv_set_drvdata(dev, net);
2408 
2409 	INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
2410 
2411 	spin_lock_init(&net_device_ctx->lock);
2412 	INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
2413 	INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup);
2414 
2415 	net_device_ctx->vf_stats
2416 		= netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats);
2417 	if (!net_device_ctx->vf_stats)
2418 		goto no_stats;
2419 
2420 	net->netdev_ops = &device_ops;
2421 	net->ethtool_ops = &ethtool_ops;
2422 	SET_NETDEV_DEV(net, &dev->device);
2423 
2424 	/* We always need headroom for rndis header */
2425 	net->needed_headroom = RNDIS_AND_PPI_SIZE;
2426 
2427 	/* Initialize the number of queues to be 1, we may change it if more
2428 	 * channels are offered later.
2429 	 */
2430 	netif_set_real_num_tx_queues(net, 1);
2431 	netif_set_real_num_rx_queues(net, 1);
2432 
2433 	/* Notify the netvsc driver of the new device */
2434 	device_info = netvsc_devinfo_get(NULL);
2435 
2436 	if (!device_info) {
2437 		ret = -ENOMEM;
2438 		goto devinfo_failed;
2439 	}
2440 
2441 	nvdev = rndis_filter_device_add(dev, device_info);
2442 	if (IS_ERR(nvdev)) {
2443 		ret = PTR_ERR(nvdev);
2444 		netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
2445 		goto rndis_failed;
2446 	}
2447 
2448 	memcpy(net->dev_addr, device_info->mac_adr, ETH_ALEN);
2449 
2450 	/* We must get rtnl lock before scheduling nvdev->subchan_work,
2451 	 * otherwise netvsc_subchan_work() can get rtnl lock first and wait
2452 	 * all subchannels to show up, but that may not happen because
2453 	 * netvsc_probe() can't get rtnl lock and as a result vmbus_onoffer()
2454 	 * -> ... -> device_add() -> ... -> __device_attach() can't get
2455 	 * the device lock, so all the subchannels can't be processed --
2456 	 * finally netvsc_subchan_work() hangs forever.
2457 	 */
2458 	rtnl_lock();
2459 
2460 	if (nvdev->num_chn > 1)
2461 		schedule_work(&nvdev->subchan_work);
2462 
2463 	/* hw_features computed in rndis_netdev_set_hwcaps() */
2464 	net->features = net->hw_features |
2465 		NETIF_F_HIGHDMA | NETIF_F_HW_VLAN_CTAG_TX |
2466 		NETIF_F_HW_VLAN_CTAG_RX;
2467 	net->vlan_features = net->features;
2468 
2469 	/* MTU range: 68 - 1500 or 65521 */
2470 	net->min_mtu = NETVSC_MTU_MIN;
2471 	if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
2472 		net->max_mtu = NETVSC_MTU - ETH_HLEN;
2473 	else
2474 		net->max_mtu = ETH_DATA_LEN;
2475 
2476 	ret = register_netdevice(net);
2477 	if (ret != 0) {
2478 		pr_err("Unable to register netdev.\n");
2479 		goto register_failed;
2480 	}
2481 
2482 	list_add(&net_device_ctx->list, &netvsc_dev_list);
2483 	rtnl_unlock();
2484 
2485 	netvsc_devinfo_put(device_info);
2486 	return 0;
2487 
2488 register_failed:
2489 	rtnl_unlock();
2490 	rndis_filter_device_remove(dev, nvdev);
2491 rndis_failed:
2492 	netvsc_devinfo_put(device_info);
2493 devinfo_failed:
2494 	free_percpu(net_device_ctx->vf_stats);
2495 no_stats:
2496 	hv_set_drvdata(dev, NULL);
2497 	free_netdev(net);
2498 no_net:
2499 	return ret;
2500 }
2501 
2502 static int netvsc_remove(struct hv_device *dev)
2503 {
2504 	struct net_device_context *ndev_ctx;
2505 	struct net_device *vf_netdev, *net;
2506 	struct netvsc_device *nvdev;
2507 
2508 	net = hv_get_drvdata(dev);
2509 	if (net == NULL) {
2510 		dev_err(&dev->device, "No net device to remove\n");
2511 		return 0;
2512 	}
2513 
2514 	ndev_ctx = netdev_priv(net);
2515 
2516 	cancel_delayed_work_sync(&ndev_ctx->dwork);
2517 
2518 	rtnl_lock();
2519 	nvdev = rtnl_dereference(ndev_ctx->nvdev);
2520 	if (nvdev) {
2521 		cancel_work_sync(&nvdev->subchan_work);
2522 		netvsc_xdp_set(net, NULL, NULL, nvdev);
2523 	}
2524 
2525 	/*
2526 	 * Call to the vsc driver to let it know that the device is being
2527 	 * removed. Also blocks mtu and channel changes.
2528 	 */
2529 	vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2530 	if (vf_netdev)
2531 		netvsc_unregister_vf(vf_netdev);
2532 
2533 	if (nvdev)
2534 		rndis_filter_device_remove(dev, nvdev);
2535 
2536 	unregister_netdevice(net);
2537 	list_del(&ndev_ctx->list);
2538 
2539 	rtnl_unlock();
2540 
2541 	hv_set_drvdata(dev, NULL);
2542 
2543 	free_percpu(ndev_ctx->vf_stats);
2544 	free_netdev(net);
2545 	return 0;
2546 }
2547 
2548 static int netvsc_suspend(struct hv_device *dev)
2549 {
2550 	struct net_device_context *ndev_ctx;
2551 	struct net_device *vf_netdev, *net;
2552 	struct netvsc_device *nvdev;
2553 	int ret;
2554 
2555 	net = hv_get_drvdata(dev);
2556 
2557 	ndev_ctx = netdev_priv(net);
2558 	cancel_delayed_work_sync(&ndev_ctx->dwork);
2559 
2560 	rtnl_lock();
2561 
2562 	nvdev = rtnl_dereference(ndev_ctx->nvdev);
2563 	if (nvdev == NULL) {
2564 		ret = -ENODEV;
2565 		goto out;
2566 	}
2567 
2568 	vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2569 	if (vf_netdev)
2570 		netvsc_unregister_vf(vf_netdev);
2571 
2572 	/* Save the current config info */
2573 	ndev_ctx->saved_netvsc_dev_info = netvsc_devinfo_get(nvdev);
2574 
2575 	ret = netvsc_detach(net, nvdev);
2576 out:
2577 	rtnl_unlock();
2578 
2579 	return ret;
2580 }
2581 
2582 static int netvsc_resume(struct hv_device *dev)
2583 {
2584 	struct net_device *net = hv_get_drvdata(dev);
2585 	struct net_device_context *net_device_ctx;
2586 	struct netvsc_device_info *device_info;
2587 	int ret;
2588 
2589 	rtnl_lock();
2590 
2591 	net_device_ctx = netdev_priv(net);
2592 	device_info = net_device_ctx->saved_netvsc_dev_info;
2593 
2594 	ret = netvsc_attach(net, device_info);
2595 
2596 	netvsc_devinfo_put(device_info);
2597 	net_device_ctx->saved_netvsc_dev_info = NULL;
2598 
2599 	rtnl_unlock();
2600 
2601 	return ret;
2602 }
2603 static const struct hv_vmbus_device_id id_table[] = {
2604 	/* Network guid */
2605 	{ HV_NIC_GUID, },
2606 	{ },
2607 };
2608 
2609 MODULE_DEVICE_TABLE(vmbus, id_table);
2610 
2611 /* The one and only one */
2612 static struct  hv_driver netvsc_drv = {
2613 	.name = KBUILD_MODNAME,
2614 	.id_table = id_table,
2615 	.probe = netvsc_probe,
2616 	.remove = netvsc_remove,
2617 	.suspend = netvsc_suspend,
2618 	.resume = netvsc_resume,
2619 	.driver = {
2620 		.probe_type = PROBE_FORCE_SYNCHRONOUS,
2621 	},
2622 };
2623 
2624 /*
2625  * On Hyper-V, every VF interface is matched with a corresponding
2626  * synthetic interface. The synthetic interface is presented first
2627  * to the guest. When the corresponding VF instance is registered,
2628  * we will take care of switching the data path.
2629  */
2630 static int netvsc_netdev_event(struct notifier_block *this,
2631 			       unsigned long event, void *ptr)
2632 {
2633 	struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2634 
2635 	/* Skip our own events */
2636 	if (event_dev->netdev_ops == &device_ops)
2637 		return NOTIFY_DONE;
2638 
2639 	/* Avoid non-Ethernet type devices */
2640 	if (event_dev->type != ARPHRD_ETHER)
2641 		return NOTIFY_DONE;
2642 
2643 	/* Avoid Vlan dev with same MAC registering as VF */
2644 	if (is_vlan_dev(event_dev))
2645 		return NOTIFY_DONE;
2646 
2647 	/* Avoid Bonding master dev with same MAC registering as VF */
2648 	if ((event_dev->priv_flags & IFF_BONDING) &&
2649 	    (event_dev->flags & IFF_MASTER))
2650 		return NOTIFY_DONE;
2651 
2652 	switch (event) {
2653 	case NETDEV_REGISTER:
2654 		return netvsc_register_vf(event_dev);
2655 	case NETDEV_UNREGISTER:
2656 		return netvsc_unregister_vf(event_dev);
2657 	case NETDEV_UP:
2658 	case NETDEV_DOWN:
2659 		return netvsc_vf_changed(event_dev);
2660 	default:
2661 		return NOTIFY_DONE;
2662 	}
2663 }
2664 
2665 static struct notifier_block netvsc_netdev_notifier = {
2666 	.notifier_call = netvsc_netdev_event,
2667 };
2668 
2669 static void __exit netvsc_drv_exit(void)
2670 {
2671 	unregister_netdevice_notifier(&netvsc_netdev_notifier);
2672 	vmbus_driver_unregister(&netvsc_drv);
2673 }
2674 
2675 static int __init netvsc_drv_init(void)
2676 {
2677 	int ret;
2678 
2679 	if (ring_size < RING_SIZE_MIN) {
2680 		ring_size = RING_SIZE_MIN;
2681 		pr_info("Increased ring_size to %u (min allowed)\n",
2682 			ring_size);
2683 	}
2684 	netvsc_ring_bytes = ring_size * PAGE_SIZE;
2685 
2686 	ret = vmbus_driver_register(&netvsc_drv);
2687 	if (ret)
2688 		return ret;
2689 
2690 	register_netdevice_notifier(&netvsc_netdev_notifier);
2691 	return 0;
2692 }
2693 
2694 MODULE_LICENSE("GPL");
2695 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
2696 
2697 module_init(netvsc_drv_init);
2698 module_exit(netvsc_drv_exit);
2699