xref: /linux/drivers/net/hyperv/netvsc_drv.c (revision 37744feebc086908fd89760650f458ab19071750)
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 netdev_tx_t netvsc_start_xmit(struct sk_buff *skb,
711 				     struct net_device *ndev)
712 {
713 	return netvsc_xmit(skb, ndev, false);
714 }
715 
716 /*
717  * netvsc_linkstatus_callback - Link up/down notification
718  */
719 void netvsc_linkstatus_callback(struct net_device *net,
720 				struct rndis_message *resp)
721 {
722 	struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
723 	struct net_device_context *ndev_ctx = netdev_priv(net);
724 	struct netvsc_reconfig *event;
725 	unsigned long flags;
726 
727 	/* Update the physical link speed when changing to another vSwitch */
728 	if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
729 		u32 speed;
730 
731 		speed = *(u32 *)((void *)indicate
732 				 + indicate->status_buf_offset) / 10000;
733 		ndev_ctx->speed = speed;
734 		return;
735 	}
736 
737 	/* Handle these link change statuses below */
738 	if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
739 	    indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
740 	    indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
741 		return;
742 
743 	if (net->reg_state != NETREG_REGISTERED)
744 		return;
745 
746 	event = kzalloc(sizeof(*event), GFP_ATOMIC);
747 	if (!event)
748 		return;
749 	event->event = indicate->status;
750 
751 	spin_lock_irqsave(&ndev_ctx->lock, flags);
752 	list_add_tail(&event->list, &ndev_ctx->reconfig_events);
753 	spin_unlock_irqrestore(&ndev_ctx->lock, flags);
754 
755 	schedule_delayed_work(&ndev_ctx->dwork, 0);
756 }
757 
758 static void netvsc_xdp_xmit(struct sk_buff *skb, struct net_device *ndev)
759 {
760 	int rc;
761 
762 	skb->queue_mapping = skb_get_rx_queue(skb);
763 	__skb_push(skb, ETH_HLEN);
764 
765 	rc = netvsc_xmit(skb, ndev, true);
766 
767 	if (dev_xmit_complete(rc))
768 		return;
769 
770 	dev_kfree_skb_any(skb);
771 	ndev->stats.tx_dropped++;
772 }
773 
774 static void netvsc_comp_ipcsum(struct sk_buff *skb)
775 {
776 	struct iphdr *iph = (struct iphdr *)skb->data;
777 
778 	iph->check = 0;
779 	iph->check = ip_fast_csum(iph, iph->ihl);
780 }
781 
782 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
783 					     struct netvsc_channel *nvchan,
784 					     struct xdp_buff *xdp)
785 {
786 	struct napi_struct *napi = &nvchan->napi;
787 	const struct ndis_pkt_8021q_info *vlan = nvchan->rsc.vlan;
788 	const struct ndis_tcp_ip_checksum_info *csum_info =
789 						nvchan->rsc.csum_info;
790 	const u32 *hash_info = nvchan->rsc.hash_info;
791 	struct sk_buff *skb;
792 	void *xbuf = xdp->data_hard_start;
793 	int i;
794 
795 	if (xbuf) {
796 		unsigned int hdroom = xdp->data - xdp->data_hard_start;
797 		unsigned int xlen = xdp->data_end - xdp->data;
798 		unsigned int frag_size = netvsc_xdp_fraglen(hdroom + xlen);
799 
800 		skb = build_skb(xbuf, frag_size);
801 
802 		if (!skb) {
803 			__free_page(virt_to_page(xbuf));
804 			return NULL;
805 		}
806 
807 		skb_reserve(skb, hdroom);
808 		skb_put(skb, xlen);
809 		skb->dev = napi->dev;
810 	} else {
811 		skb = napi_alloc_skb(napi, nvchan->rsc.pktlen);
812 
813 		if (!skb)
814 			return NULL;
815 
816 		/* Copy to skb. This copy is needed here since the memory
817 		 * pointed by hv_netvsc_packet cannot be deallocated.
818 		 */
819 		for (i = 0; i < nvchan->rsc.cnt; i++)
820 			skb_put_data(skb, nvchan->rsc.data[i],
821 				     nvchan->rsc.len[i]);
822 	}
823 
824 	skb->protocol = eth_type_trans(skb, net);
825 
826 	/* skb is already created with CHECKSUM_NONE */
827 	skb_checksum_none_assert(skb);
828 
829 	/* Incoming packets may have IP header checksum verified by the host.
830 	 * They may not have IP header checksum computed after coalescing.
831 	 * We compute it here if the flags are set, because on Linux, the IP
832 	 * checksum is always checked.
833 	 */
834 	if (csum_info && csum_info->receive.ip_checksum_value_invalid &&
835 	    csum_info->receive.ip_checksum_succeeded &&
836 	    skb->protocol == htons(ETH_P_IP))
837 		netvsc_comp_ipcsum(skb);
838 
839 	/* Do L4 checksum offload if enabled and present. */
840 	if (csum_info && (net->features & NETIF_F_RXCSUM)) {
841 		if (csum_info->receive.tcp_checksum_succeeded ||
842 		    csum_info->receive.udp_checksum_succeeded)
843 			skb->ip_summed = CHECKSUM_UNNECESSARY;
844 	}
845 
846 	if (hash_info && (net->features & NETIF_F_RXHASH))
847 		skb_set_hash(skb, *hash_info, PKT_HASH_TYPE_L4);
848 
849 	if (vlan) {
850 		u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT) |
851 			(vlan->cfi ? VLAN_CFI_MASK : 0);
852 
853 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
854 				       vlan_tci);
855 	}
856 
857 	return skb;
858 }
859 
860 /*
861  * netvsc_recv_callback -  Callback when we receive a packet from the
862  * "wire" on the specified device.
863  */
864 int netvsc_recv_callback(struct net_device *net,
865 			 struct netvsc_device *net_device,
866 			 struct netvsc_channel *nvchan)
867 {
868 	struct net_device_context *net_device_ctx = netdev_priv(net);
869 	struct vmbus_channel *channel = nvchan->channel;
870 	u16 q_idx = channel->offermsg.offer.sub_channel_index;
871 	struct sk_buff *skb;
872 	struct netvsc_stats *rx_stats = &nvchan->rx_stats;
873 	struct xdp_buff xdp;
874 	u32 act;
875 
876 	if (net->reg_state != NETREG_REGISTERED)
877 		return NVSP_STAT_FAIL;
878 
879 	act = netvsc_run_xdp(net, nvchan, &xdp);
880 
881 	if (act != XDP_PASS && act != XDP_TX) {
882 		u64_stats_update_begin(&rx_stats->syncp);
883 		rx_stats->xdp_drop++;
884 		u64_stats_update_end(&rx_stats->syncp);
885 
886 		return NVSP_STAT_SUCCESS; /* consumed by XDP */
887 	}
888 
889 	/* Allocate a skb - TODO direct I/O to pages? */
890 	skb = netvsc_alloc_recv_skb(net, nvchan, &xdp);
891 
892 	if (unlikely(!skb)) {
893 		++net_device_ctx->eth_stats.rx_no_memory;
894 		return NVSP_STAT_FAIL;
895 	}
896 
897 	skb_record_rx_queue(skb, q_idx);
898 
899 	/*
900 	 * Even if injecting the packet, record the statistics
901 	 * on the synthetic device because modifying the VF device
902 	 * statistics will not work correctly.
903 	 */
904 	u64_stats_update_begin(&rx_stats->syncp);
905 	rx_stats->packets++;
906 	rx_stats->bytes += nvchan->rsc.pktlen;
907 
908 	if (skb->pkt_type == PACKET_BROADCAST)
909 		++rx_stats->broadcast;
910 	else if (skb->pkt_type == PACKET_MULTICAST)
911 		++rx_stats->multicast;
912 	u64_stats_update_end(&rx_stats->syncp);
913 
914 	if (act == XDP_TX) {
915 		netvsc_xdp_xmit(skb, net);
916 		return NVSP_STAT_SUCCESS;
917 	}
918 
919 	napi_gro_receive(&nvchan->napi, skb);
920 	return NVSP_STAT_SUCCESS;
921 }
922 
923 static void netvsc_get_drvinfo(struct net_device *net,
924 			       struct ethtool_drvinfo *info)
925 {
926 	strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
927 	strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
928 }
929 
930 static void netvsc_get_channels(struct net_device *net,
931 				struct ethtool_channels *channel)
932 {
933 	struct net_device_context *net_device_ctx = netdev_priv(net);
934 	struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
935 
936 	if (nvdev) {
937 		channel->max_combined	= nvdev->max_chn;
938 		channel->combined_count = nvdev->num_chn;
939 	}
940 }
941 
942 /* Alloc struct netvsc_device_info, and initialize it from either existing
943  * struct netvsc_device, or from default values.
944  */
945 static
946 struct netvsc_device_info *netvsc_devinfo_get(struct netvsc_device *nvdev)
947 {
948 	struct netvsc_device_info *dev_info;
949 	struct bpf_prog *prog;
950 
951 	dev_info = kzalloc(sizeof(*dev_info), GFP_ATOMIC);
952 
953 	if (!dev_info)
954 		return NULL;
955 
956 	if (nvdev) {
957 		ASSERT_RTNL();
958 
959 		dev_info->num_chn = nvdev->num_chn;
960 		dev_info->send_sections = nvdev->send_section_cnt;
961 		dev_info->send_section_size = nvdev->send_section_size;
962 		dev_info->recv_sections = nvdev->recv_section_cnt;
963 		dev_info->recv_section_size = nvdev->recv_section_size;
964 
965 		memcpy(dev_info->rss_key, nvdev->extension->rss_key,
966 		       NETVSC_HASH_KEYLEN);
967 
968 		prog = netvsc_xdp_get(nvdev);
969 		if (prog) {
970 			bpf_prog_inc(prog);
971 			dev_info->bprog = prog;
972 		}
973 	} else {
974 		dev_info->num_chn = VRSS_CHANNEL_DEFAULT;
975 		dev_info->send_sections = NETVSC_DEFAULT_TX;
976 		dev_info->send_section_size = NETVSC_SEND_SECTION_SIZE;
977 		dev_info->recv_sections = NETVSC_DEFAULT_RX;
978 		dev_info->recv_section_size = NETVSC_RECV_SECTION_SIZE;
979 	}
980 
981 	return dev_info;
982 }
983 
984 /* Free struct netvsc_device_info */
985 static void netvsc_devinfo_put(struct netvsc_device_info *dev_info)
986 {
987 	if (dev_info->bprog) {
988 		ASSERT_RTNL();
989 		bpf_prog_put(dev_info->bprog);
990 	}
991 
992 	kfree(dev_info);
993 }
994 
995 static int netvsc_detach(struct net_device *ndev,
996 			 struct netvsc_device *nvdev)
997 {
998 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
999 	struct hv_device *hdev = ndev_ctx->device_ctx;
1000 	int ret;
1001 
1002 	/* Don't try continuing to try and setup sub channels */
1003 	if (cancel_work_sync(&nvdev->subchan_work))
1004 		nvdev->num_chn = 1;
1005 
1006 	netvsc_xdp_set(ndev, NULL, NULL, nvdev);
1007 
1008 	/* If device was up (receiving) then shutdown */
1009 	if (netif_running(ndev)) {
1010 		netvsc_tx_disable(nvdev, ndev);
1011 
1012 		ret = rndis_filter_close(nvdev);
1013 		if (ret) {
1014 			netdev_err(ndev,
1015 				   "unable to close device (ret %d).\n", ret);
1016 			return ret;
1017 		}
1018 
1019 		ret = netvsc_wait_until_empty(nvdev);
1020 		if (ret) {
1021 			netdev_err(ndev,
1022 				   "Ring buffer not empty after closing rndis\n");
1023 			return ret;
1024 		}
1025 	}
1026 
1027 	netif_device_detach(ndev);
1028 
1029 	rndis_filter_device_remove(hdev, nvdev);
1030 
1031 	return 0;
1032 }
1033 
1034 static int netvsc_attach(struct net_device *ndev,
1035 			 struct netvsc_device_info *dev_info)
1036 {
1037 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
1038 	struct hv_device *hdev = ndev_ctx->device_ctx;
1039 	struct netvsc_device *nvdev;
1040 	struct rndis_device *rdev;
1041 	struct bpf_prog *prog;
1042 	int ret = 0;
1043 
1044 	nvdev = rndis_filter_device_add(hdev, dev_info);
1045 	if (IS_ERR(nvdev))
1046 		return PTR_ERR(nvdev);
1047 
1048 	if (nvdev->num_chn > 1) {
1049 		ret = rndis_set_subchannel(ndev, nvdev, dev_info);
1050 
1051 		/* if unavailable, just proceed with one queue */
1052 		if (ret) {
1053 			nvdev->max_chn = 1;
1054 			nvdev->num_chn = 1;
1055 		}
1056 	}
1057 
1058 	prog = dev_info->bprog;
1059 	if (prog) {
1060 		bpf_prog_inc(prog);
1061 		ret = netvsc_xdp_set(ndev, prog, NULL, nvdev);
1062 		if (ret) {
1063 			bpf_prog_put(prog);
1064 			goto err1;
1065 		}
1066 	}
1067 
1068 	/* In any case device is now ready */
1069 	nvdev->tx_disable = false;
1070 	netif_device_attach(ndev);
1071 
1072 	/* Note: enable and attach happen when sub-channels setup */
1073 	netif_carrier_off(ndev);
1074 
1075 	if (netif_running(ndev)) {
1076 		ret = rndis_filter_open(nvdev);
1077 		if (ret)
1078 			goto err2;
1079 
1080 		rdev = nvdev->extension;
1081 		if (!rdev->link_state)
1082 			netif_carrier_on(ndev);
1083 	}
1084 
1085 	return 0;
1086 
1087 err2:
1088 	netif_device_detach(ndev);
1089 
1090 err1:
1091 	rndis_filter_device_remove(hdev, nvdev);
1092 
1093 	return ret;
1094 }
1095 
1096 static int netvsc_set_channels(struct net_device *net,
1097 			       struct ethtool_channels *channels)
1098 {
1099 	struct net_device_context *net_device_ctx = netdev_priv(net);
1100 	struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
1101 	unsigned int orig, count = channels->combined_count;
1102 	struct netvsc_device_info *device_info;
1103 	int ret;
1104 
1105 	/* We do not support separate count for rx, tx, or other */
1106 	if (count == 0 ||
1107 	    channels->rx_count || channels->tx_count || channels->other_count)
1108 		return -EINVAL;
1109 
1110 	if (!nvdev || nvdev->destroy)
1111 		return -ENODEV;
1112 
1113 	if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
1114 		return -EINVAL;
1115 
1116 	if (count > nvdev->max_chn)
1117 		return -EINVAL;
1118 
1119 	orig = nvdev->num_chn;
1120 
1121 	device_info = netvsc_devinfo_get(nvdev);
1122 
1123 	if (!device_info)
1124 		return -ENOMEM;
1125 
1126 	device_info->num_chn = count;
1127 
1128 	ret = netvsc_detach(net, nvdev);
1129 	if (ret)
1130 		goto out;
1131 
1132 	ret = netvsc_attach(net, device_info);
1133 	if (ret) {
1134 		device_info->num_chn = orig;
1135 		if (netvsc_attach(net, device_info))
1136 			netdev_err(net, "restoring channel setting failed\n");
1137 	}
1138 
1139 out:
1140 	netvsc_devinfo_put(device_info);
1141 	return ret;
1142 }
1143 
1144 static void netvsc_init_settings(struct net_device *dev)
1145 {
1146 	struct net_device_context *ndc = netdev_priv(dev);
1147 
1148 	ndc->l4_hash = HV_DEFAULT_L4HASH;
1149 
1150 	ndc->speed = SPEED_UNKNOWN;
1151 	ndc->duplex = DUPLEX_FULL;
1152 
1153 	dev->features = NETIF_F_LRO;
1154 }
1155 
1156 static int netvsc_get_link_ksettings(struct net_device *dev,
1157 				     struct ethtool_link_ksettings *cmd)
1158 {
1159 	struct net_device_context *ndc = netdev_priv(dev);
1160 	struct net_device *vf_netdev;
1161 
1162 	vf_netdev = rtnl_dereference(ndc->vf_netdev);
1163 
1164 	if (vf_netdev)
1165 		return __ethtool_get_link_ksettings(vf_netdev, cmd);
1166 
1167 	cmd->base.speed = ndc->speed;
1168 	cmd->base.duplex = ndc->duplex;
1169 	cmd->base.port = PORT_OTHER;
1170 
1171 	return 0;
1172 }
1173 
1174 static int netvsc_set_link_ksettings(struct net_device *dev,
1175 				     const struct ethtool_link_ksettings *cmd)
1176 {
1177 	struct net_device_context *ndc = netdev_priv(dev);
1178 	struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1179 
1180 	if (vf_netdev) {
1181 		if (!vf_netdev->ethtool_ops->set_link_ksettings)
1182 			return -EOPNOTSUPP;
1183 
1184 		return vf_netdev->ethtool_ops->set_link_ksettings(vf_netdev,
1185 								  cmd);
1186 	}
1187 
1188 	return ethtool_virtdev_set_link_ksettings(dev, cmd,
1189 						  &ndc->speed, &ndc->duplex);
1190 }
1191 
1192 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
1193 {
1194 	struct net_device_context *ndevctx = netdev_priv(ndev);
1195 	struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1196 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1197 	int orig_mtu = ndev->mtu;
1198 	struct netvsc_device_info *device_info;
1199 	int ret = 0;
1200 
1201 	if (!nvdev || nvdev->destroy)
1202 		return -ENODEV;
1203 
1204 	device_info = netvsc_devinfo_get(nvdev);
1205 
1206 	if (!device_info)
1207 		return -ENOMEM;
1208 
1209 	/* Change MTU of underlying VF netdev first. */
1210 	if (vf_netdev) {
1211 		ret = dev_set_mtu(vf_netdev, mtu);
1212 		if (ret)
1213 			goto out;
1214 	}
1215 
1216 	ret = netvsc_detach(ndev, nvdev);
1217 	if (ret)
1218 		goto rollback_vf;
1219 
1220 	ndev->mtu = mtu;
1221 
1222 	ret = netvsc_attach(ndev, device_info);
1223 	if (!ret)
1224 		goto out;
1225 
1226 	/* Attempt rollback to original MTU */
1227 	ndev->mtu = orig_mtu;
1228 
1229 	if (netvsc_attach(ndev, device_info))
1230 		netdev_err(ndev, "restoring mtu failed\n");
1231 rollback_vf:
1232 	if (vf_netdev)
1233 		dev_set_mtu(vf_netdev, orig_mtu);
1234 
1235 out:
1236 	netvsc_devinfo_put(device_info);
1237 	return ret;
1238 }
1239 
1240 static void netvsc_get_vf_stats(struct net_device *net,
1241 				struct netvsc_vf_pcpu_stats *tot)
1242 {
1243 	struct net_device_context *ndev_ctx = netdev_priv(net);
1244 	int i;
1245 
1246 	memset(tot, 0, sizeof(*tot));
1247 
1248 	for_each_possible_cpu(i) {
1249 		const struct netvsc_vf_pcpu_stats *stats
1250 			= per_cpu_ptr(ndev_ctx->vf_stats, i);
1251 		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1252 		unsigned int start;
1253 
1254 		do {
1255 			start = u64_stats_fetch_begin_irq(&stats->syncp);
1256 			rx_packets = stats->rx_packets;
1257 			tx_packets = stats->tx_packets;
1258 			rx_bytes = stats->rx_bytes;
1259 			tx_bytes = stats->tx_bytes;
1260 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1261 
1262 		tot->rx_packets += rx_packets;
1263 		tot->tx_packets += tx_packets;
1264 		tot->rx_bytes   += rx_bytes;
1265 		tot->tx_bytes   += tx_bytes;
1266 		tot->tx_dropped += stats->tx_dropped;
1267 	}
1268 }
1269 
1270 static void netvsc_get_pcpu_stats(struct net_device *net,
1271 				  struct netvsc_ethtool_pcpu_stats *pcpu_tot)
1272 {
1273 	struct net_device_context *ndev_ctx = netdev_priv(net);
1274 	struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1275 	int i;
1276 
1277 	/* fetch percpu stats of vf */
1278 	for_each_possible_cpu(i) {
1279 		const struct netvsc_vf_pcpu_stats *stats =
1280 			per_cpu_ptr(ndev_ctx->vf_stats, i);
1281 		struct netvsc_ethtool_pcpu_stats *this_tot = &pcpu_tot[i];
1282 		unsigned int start;
1283 
1284 		do {
1285 			start = u64_stats_fetch_begin_irq(&stats->syncp);
1286 			this_tot->vf_rx_packets = stats->rx_packets;
1287 			this_tot->vf_tx_packets = stats->tx_packets;
1288 			this_tot->vf_rx_bytes = stats->rx_bytes;
1289 			this_tot->vf_tx_bytes = stats->tx_bytes;
1290 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1291 		this_tot->rx_packets = this_tot->vf_rx_packets;
1292 		this_tot->tx_packets = this_tot->vf_tx_packets;
1293 		this_tot->rx_bytes   = this_tot->vf_rx_bytes;
1294 		this_tot->tx_bytes   = this_tot->vf_tx_bytes;
1295 	}
1296 
1297 	/* fetch percpu stats of netvsc */
1298 	for (i = 0; i < nvdev->num_chn; i++) {
1299 		const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1300 		const struct netvsc_stats *stats;
1301 		struct netvsc_ethtool_pcpu_stats *this_tot =
1302 			&pcpu_tot[nvchan->channel->target_cpu];
1303 		u64 packets, bytes;
1304 		unsigned int start;
1305 
1306 		stats = &nvchan->tx_stats;
1307 		do {
1308 			start = u64_stats_fetch_begin_irq(&stats->syncp);
1309 			packets = stats->packets;
1310 			bytes = stats->bytes;
1311 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1312 
1313 		this_tot->tx_bytes	+= bytes;
1314 		this_tot->tx_packets	+= packets;
1315 
1316 		stats = &nvchan->rx_stats;
1317 		do {
1318 			start = u64_stats_fetch_begin_irq(&stats->syncp);
1319 			packets = stats->packets;
1320 			bytes = stats->bytes;
1321 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1322 
1323 		this_tot->rx_bytes	+= bytes;
1324 		this_tot->rx_packets	+= packets;
1325 	}
1326 }
1327 
1328 static void netvsc_get_stats64(struct net_device *net,
1329 			       struct rtnl_link_stats64 *t)
1330 {
1331 	struct net_device_context *ndev_ctx = netdev_priv(net);
1332 	struct netvsc_device *nvdev;
1333 	struct netvsc_vf_pcpu_stats vf_tot;
1334 	int i;
1335 
1336 	rcu_read_lock();
1337 
1338 	nvdev = rcu_dereference(ndev_ctx->nvdev);
1339 	if (!nvdev)
1340 		goto out;
1341 
1342 	netdev_stats_to_stats64(t, &net->stats);
1343 
1344 	netvsc_get_vf_stats(net, &vf_tot);
1345 	t->rx_packets += vf_tot.rx_packets;
1346 	t->tx_packets += vf_tot.tx_packets;
1347 	t->rx_bytes   += vf_tot.rx_bytes;
1348 	t->tx_bytes   += vf_tot.tx_bytes;
1349 	t->tx_dropped += vf_tot.tx_dropped;
1350 
1351 	for (i = 0; i < nvdev->num_chn; i++) {
1352 		const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1353 		const struct netvsc_stats *stats;
1354 		u64 packets, bytes, multicast;
1355 		unsigned int start;
1356 
1357 		stats = &nvchan->tx_stats;
1358 		do {
1359 			start = u64_stats_fetch_begin_irq(&stats->syncp);
1360 			packets = stats->packets;
1361 			bytes = stats->bytes;
1362 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1363 
1364 		t->tx_bytes	+= bytes;
1365 		t->tx_packets	+= packets;
1366 
1367 		stats = &nvchan->rx_stats;
1368 		do {
1369 			start = u64_stats_fetch_begin_irq(&stats->syncp);
1370 			packets = stats->packets;
1371 			bytes = stats->bytes;
1372 			multicast = stats->multicast + stats->broadcast;
1373 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1374 
1375 		t->rx_bytes	+= bytes;
1376 		t->rx_packets	+= packets;
1377 		t->multicast	+= multicast;
1378 	}
1379 out:
1380 	rcu_read_unlock();
1381 }
1382 
1383 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1384 {
1385 	struct net_device_context *ndc = netdev_priv(ndev);
1386 	struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1387 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1388 	struct sockaddr *addr = p;
1389 	int err;
1390 
1391 	err = eth_prepare_mac_addr_change(ndev, p);
1392 	if (err)
1393 		return err;
1394 
1395 	if (!nvdev)
1396 		return -ENODEV;
1397 
1398 	if (vf_netdev) {
1399 		err = dev_set_mac_address(vf_netdev, addr, NULL);
1400 		if (err)
1401 			return err;
1402 	}
1403 
1404 	err = rndis_filter_set_device_mac(nvdev, addr->sa_data);
1405 	if (!err) {
1406 		eth_commit_mac_addr_change(ndev, p);
1407 	} else if (vf_netdev) {
1408 		/* rollback change on VF */
1409 		memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN);
1410 		dev_set_mac_address(vf_netdev, addr, NULL);
1411 	}
1412 
1413 	return err;
1414 }
1415 
1416 static const struct {
1417 	char name[ETH_GSTRING_LEN];
1418 	u16 offset;
1419 } netvsc_stats[] = {
1420 	{ "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1421 	{ "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1422 	{ "tx_no_space",  offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1423 	{ "tx_too_big",	  offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1424 	{ "tx_busy",	  offsetof(struct netvsc_ethtool_stats, tx_busy) },
1425 	{ "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) },
1426 	{ "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) },
1427 	{ "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) },
1428 	{ "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) },
1429 	{ "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) },
1430 }, pcpu_stats[] = {
1431 	{ "cpu%u_rx_packets",
1432 		offsetof(struct netvsc_ethtool_pcpu_stats, rx_packets) },
1433 	{ "cpu%u_rx_bytes",
1434 		offsetof(struct netvsc_ethtool_pcpu_stats, rx_bytes) },
1435 	{ "cpu%u_tx_packets",
1436 		offsetof(struct netvsc_ethtool_pcpu_stats, tx_packets) },
1437 	{ "cpu%u_tx_bytes",
1438 		offsetof(struct netvsc_ethtool_pcpu_stats, tx_bytes) },
1439 	{ "cpu%u_vf_rx_packets",
1440 		offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_packets) },
1441 	{ "cpu%u_vf_rx_bytes",
1442 		offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_bytes) },
1443 	{ "cpu%u_vf_tx_packets",
1444 		offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_packets) },
1445 	{ "cpu%u_vf_tx_bytes",
1446 		offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_bytes) },
1447 }, vf_stats[] = {
1448 	{ "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) },
1449 	{ "vf_rx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) },
1450 	{ "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) },
1451 	{ "vf_tx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) },
1452 	{ "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) },
1453 };
1454 
1455 #define NETVSC_GLOBAL_STATS_LEN	ARRAY_SIZE(netvsc_stats)
1456 #define NETVSC_VF_STATS_LEN	ARRAY_SIZE(vf_stats)
1457 
1458 /* statistics per queue (rx/tx packets/bytes) */
1459 #define NETVSC_PCPU_STATS_LEN (num_present_cpus() * ARRAY_SIZE(pcpu_stats))
1460 
1461 /* 5 statistics per queue (rx/tx packets/bytes, rx xdp_drop) */
1462 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 5)
1463 
1464 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1465 {
1466 	struct net_device_context *ndc = netdev_priv(dev);
1467 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1468 
1469 	if (!nvdev)
1470 		return -ENODEV;
1471 
1472 	switch (string_set) {
1473 	case ETH_SS_STATS:
1474 		return NETVSC_GLOBAL_STATS_LEN
1475 			+ NETVSC_VF_STATS_LEN
1476 			+ NETVSC_QUEUE_STATS_LEN(nvdev)
1477 			+ NETVSC_PCPU_STATS_LEN;
1478 	default:
1479 		return -EINVAL;
1480 	}
1481 }
1482 
1483 static void netvsc_get_ethtool_stats(struct net_device *dev,
1484 				     struct ethtool_stats *stats, u64 *data)
1485 {
1486 	struct net_device_context *ndc = netdev_priv(dev);
1487 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1488 	const void *nds = &ndc->eth_stats;
1489 	const struct netvsc_stats *qstats;
1490 	struct netvsc_vf_pcpu_stats sum;
1491 	struct netvsc_ethtool_pcpu_stats *pcpu_sum;
1492 	unsigned int start;
1493 	u64 packets, bytes;
1494 	u64 xdp_drop;
1495 	int i, j, cpu;
1496 
1497 	if (!nvdev)
1498 		return;
1499 
1500 	for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1501 		data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1502 
1503 	netvsc_get_vf_stats(dev, &sum);
1504 	for (j = 0; j < NETVSC_VF_STATS_LEN; j++)
1505 		data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset);
1506 
1507 	for (j = 0; j < nvdev->num_chn; j++) {
1508 		qstats = &nvdev->chan_table[j].tx_stats;
1509 
1510 		do {
1511 			start = u64_stats_fetch_begin_irq(&qstats->syncp);
1512 			packets = qstats->packets;
1513 			bytes = qstats->bytes;
1514 		} while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1515 		data[i++] = packets;
1516 		data[i++] = bytes;
1517 
1518 		qstats = &nvdev->chan_table[j].rx_stats;
1519 		do {
1520 			start = u64_stats_fetch_begin_irq(&qstats->syncp);
1521 			packets = qstats->packets;
1522 			bytes = qstats->bytes;
1523 			xdp_drop = qstats->xdp_drop;
1524 		} while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1525 		data[i++] = packets;
1526 		data[i++] = bytes;
1527 		data[i++] = xdp_drop;
1528 	}
1529 
1530 	pcpu_sum = kvmalloc_array(num_possible_cpus(),
1531 				  sizeof(struct netvsc_ethtool_pcpu_stats),
1532 				  GFP_KERNEL);
1533 	netvsc_get_pcpu_stats(dev, pcpu_sum);
1534 	for_each_present_cpu(cpu) {
1535 		struct netvsc_ethtool_pcpu_stats *this_sum = &pcpu_sum[cpu];
1536 
1537 		for (j = 0; j < ARRAY_SIZE(pcpu_stats); j++)
1538 			data[i++] = *(u64 *)((void *)this_sum
1539 					     + pcpu_stats[j].offset);
1540 	}
1541 	kvfree(pcpu_sum);
1542 }
1543 
1544 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1545 {
1546 	struct net_device_context *ndc = netdev_priv(dev);
1547 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1548 	u8 *p = data;
1549 	int i, cpu;
1550 
1551 	if (!nvdev)
1552 		return;
1553 
1554 	switch (stringset) {
1555 	case ETH_SS_STATS:
1556 		for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++) {
1557 			memcpy(p, netvsc_stats[i].name, ETH_GSTRING_LEN);
1558 			p += ETH_GSTRING_LEN;
1559 		}
1560 
1561 		for (i = 0; i < ARRAY_SIZE(vf_stats); i++) {
1562 			memcpy(p, vf_stats[i].name, ETH_GSTRING_LEN);
1563 			p += ETH_GSTRING_LEN;
1564 		}
1565 
1566 		for (i = 0; i < nvdev->num_chn; i++) {
1567 			sprintf(p, "tx_queue_%u_packets", i);
1568 			p += ETH_GSTRING_LEN;
1569 			sprintf(p, "tx_queue_%u_bytes", i);
1570 			p += ETH_GSTRING_LEN;
1571 			sprintf(p, "rx_queue_%u_packets", i);
1572 			p += ETH_GSTRING_LEN;
1573 			sprintf(p, "rx_queue_%u_bytes", i);
1574 			p += ETH_GSTRING_LEN;
1575 			sprintf(p, "rx_queue_%u_xdp_drop", i);
1576 			p += ETH_GSTRING_LEN;
1577 		}
1578 
1579 		for_each_present_cpu(cpu) {
1580 			for (i = 0; i < ARRAY_SIZE(pcpu_stats); i++) {
1581 				sprintf(p, pcpu_stats[i].name, cpu);
1582 				p += ETH_GSTRING_LEN;
1583 			}
1584 		}
1585 
1586 		break;
1587 	}
1588 }
1589 
1590 static int
1591 netvsc_get_rss_hash_opts(struct net_device_context *ndc,
1592 			 struct ethtool_rxnfc *info)
1593 {
1594 	const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3;
1595 
1596 	info->data = RXH_IP_SRC | RXH_IP_DST;
1597 
1598 	switch (info->flow_type) {
1599 	case TCP_V4_FLOW:
1600 		if (ndc->l4_hash & HV_TCP4_L4HASH)
1601 			info->data |= l4_flag;
1602 
1603 		break;
1604 
1605 	case TCP_V6_FLOW:
1606 		if (ndc->l4_hash & HV_TCP6_L4HASH)
1607 			info->data |= l4_flag;
1608 
1609 		break;
1610 
1611 	case UDP_V4_FLOW:
1612 		if (ndc->l4_hash & HV_UDP4_L4HASH)
1613 			info->data |= l4_flag;
1614 
1615 		break;
1616 
1617 	case UDP_V6_FLOW:
1618 		if (ndc->l4_hash & HV_UDP6_L4HASH)
1619 			info->data |= l4_flag;
1620 
1621 		break;
1622 
1623 	case IPV4_FLOW:
1624 	case IPV6_FLOW:
1625 		break;
1626 	default:
1627 		info->data = 0;
1628 		break;
1629 	}
1630 
1631 	return 0;
1632 }
1633 
1634 static int
1635 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1636 		 u32 *rules)
1637 {
1638 	struct net_device_context *ndc = netdev_priv(dev);
1639 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1640 
1641 	if (!nvdev)
1642 		return -ENODEV;
1643 
1644 	switch (info->cmd) {
1645 	case ETHTOOL_GRXRINGS:
1646 		info->data = nvdev->num_chn;
1647 		return 0;
1648 
1649 	case ETHTOOL_GRXFH:
1650 		return netvsc_get_rss_hash_opts(ndc, info);
1651 	}
1652 	return -EOPNOTSUPP;
1653 }
1654 
1655 static int netvsc_set_rss_hash_opts(struct net_device_context *ndc,
1656 				    struct ethtool_rxnfc *info)
1657 {
1658 	if (info->data == (RXH_IP_SRC | RXH_IP_DST |
1659 			   RXH_L4_B_0_1 | RXH_L4_B_2_3)) {
1660 		switch (info->flow_type) {
1661 		case TCP_V4_FLOW:
1662 			ndc->l4_hash |= HV_TCP4_L4HASH;
1663 			break;
1664 
1665 		case TCP_V6_FLOW:
1666 			ndc->l4_hash |= HV_TCP6_L4HASH;
1667 			break;
1668 
1669 		case UDP_V4_FLOW:
1670 			ndc->l4_hash |= HV_UDP4_L4HASH;
1671 			break;
1672 
1673 		case UDP_V6_FLOW:
1674 			ndc->l4_hash |= HV_UDP6_L4HASH;
1675 			break;
1676 
1677 		default:
1678 			return -EOPNOTSUPP;
1679 		}
1680 
1681 		return 0;
1682 	}
1683 
1684 	if (info->data == (RXH_IP_SRC | RXH_IP_DST)) {
1685 		switch (info->flow_type) {
1686 		case TCP_V4_FLOW:
1687 			ndc->l4_hash &= ~HV_TCP4_L4HASH;
1688 			break;
1689 
1690 		case TCP_V6_FLOW:
1691 			ndc->l4_hash &= ~HV_TCP6_L4HASH;
1692 			break;
1693 
1694 		case UDP_V4_FLOW:
1695 			ndc->l4_hash &= ~HV_UDP4_L4HASH;
1696 			break;
1697 
1698 		case UDP_V6_FLOW:
1699 			ndc->l4_hash &= ~HV_UDP6_L4HASH;
1700 			break;
1701 
1702 		default:
1703 			return -EOPNOTSUPP;
1704 		}
1705 
1706 		return 0;
1707 	}
1708 
1709 	return -EOPNOTSUPP;
1710 }
1711 
1712 static int
1713 netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info)
1714 {
1715 	struct net_device_context *ndc = netdev_priv(ndev);
1716 
1717 	if (info->cmd == ETHTOOL_SRXFH)
1718 		return netvsc_set_rss_hash_opts(ndc, info);
1719 
1720 	return -EOPNOTSUPP;
1721 }
1722 
1723 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1724 {
1725 	return NETVSC_HASH_KEYLEN;
1726 }
1727 
1728 static u32 netvsc_rss_indir_size(struct net_device *dev)
1729 {
1730 	return ITAB_NUM;
1731 }
1732 
1733 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1734 			   u8 *hfunc)
1735 {
1736 	struct net_device_context *ndc = netdev_priv(dev);
1737 	struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1738 	struct rndis_device *rndis_dev;
1739 	int i;
1740 
1741 	if (!ndev)
1742 		return -ENODEV;
1743 
1744 	if (hfunc)
1745 		*hfunc = ETH_RSS_HASH_TOP;	/* Toeplitz */
1746 
1747 	rndis_dev = ndev->extension;
1748 	if (indir) {
1749 		for (i = 0; i < ITAB_NUM; i++)
1750 			indir[i] = ndc->rx_table[i];
1751 	}
1752 
1753 	if (key)
1754 		memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1755 
1756 	return 0;
1757 }
1758 
1759 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1760 			   const u8 *key, const u8 hfunc)
1761 {
1762 	struct net_device_context *ndc = netdev_priv(dev);
1763 	struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1764 	struct rndis_device *rndis_dev;
1765 	int i;
1766 
1767 	if (!ndev)
1768 		return -ENODEV;
1769 
1770 	if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1771 		return -EOPNOTSUPP;
1772 
1773 	rndis_dev = ndev->extension;
1774 	if (indir) {
1775 		for (i = 0; i < ITAB_NUM; i++)
1776 			if (indir[i] >= ndev->num_chn)
1777 				return -EINVAL;
1778 
1779 		for (i = 0; i < ITAB_NUM; i++)
1780 			ndc->rx_table[i] = indir[i];
1781 	}
1782 
1783 	if (!key) {
1784 		if (!indir)
1785 			return 0;
1786 
1787 		key = rndis_dev->rss_key;
1788 	}
1789 
1790 	return rndis_filter_set_rss_param(rndis_dev, key);
1791 }
1792 
1793 /* Hyper-V RNDIS protocol does not have ring in the HW sense.
1794  * It does have pre-allocated receive area which is divided into sections.
1795  */
1796 static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
1797 				   struct ethtool_ringparam *ring)
1798 {
1799 	u32 max_buf_size;
1800 
1801 	ring->rx_pending = nvdev->recv_section_cnt;
1802 	ring->tx_pending = nvdev->send_section_cnt;
1803 
1804 	if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
1805 		max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
1806 	else
1807 		max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1808 
1809 	ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
1810 	ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
1811 		/ nvdev->send_section_size;
1812 }
1813 
1814 static void netvsc_get_ringparam(struct net_device *ndev,
1815 				 struct ethtool_ringparam *ring)
1816 {
1817 	struct net_device_context *ndevctx = netdev_priv(ndev);
1818 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1819 
1820 	if (!nvdev)
1821 		return;
1822 
1823 	__netvsc_get_ringparam(nvdev, ring);
1824 }
1825 
1826 static int netvsc_set_ringparam(struct net_device *ndev,
1827 				struct ethtool_ringparam *ring)
1828 {
1829 	struct net_device_context *ndevctx = netdev_priv(ndev);
1830 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1831 	struct netvsc_device_info *device_info;
1832 	struct ethtool_ringparam orig;
1833 	u32 new_tx, new_rx;
1834 	int ret = 0;
1835 
1836 	if (!nvdev || nvdev->destroy)
1837 		return -ENODEV;
1838 
1839 	memset(&orig, 0, sizeof(orig));
1840 	__netvsc_get_ringparam(nvdev, &orig);
1841 
1842 	new_tx = clamp_t(u32, ring->tx_pending,
1843 			 NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
1844 	new_rx = clamp_t(u32, ring->rx_pending,
1845 			 NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
1846 
1847 	if (new_tx == orig.tx_pending &&
1848 	    new_rx == orig.rx_pending)
1849 		return 0;	 /* no change */
1850 
1851 	device_info = netvsc_devinfo_get(nvdev);
1852 
1853 	if (!device_info)
1854 		return -ENOMEM;
1855 
1856 	device_info->send_sections = new_tx;
1857 	device_info->recv_sections = new_rx;
1858 
1859 	ret = netvsc_detach(ndev, nvdev);
1860 	if (ret)
1861 		goto out;
1862 
1863 	ret = netvsc_attach(ndev, device_info);
1864 	if (ret) {
1865 		device_info->send_sections = orig.tx_pending;
1866 		device_info->recv_sections = orig.rx_pending;
1867 
1868 		if (netvsc_attach(ndev, device_info))
1869 			netdev_err(ndev, "restoring ringparam failed");
1870 	}
1871 
1872 out:
1873 	netvsc_devinfo_put(device_info);
1874 	return ret;
1875 }
1876 
1877 static netdev_features_t netvsc_fix_features(struct net_device *ndev,
1878 					     netdev_features_t features)
1879 {
1880 	struct net_device_context *ndevctx = netdev_priv(ndev);
1881 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1882 
1883 	if (!nvdev || nvdev->destroy)
1884 		return features;
1885 
1886 	if ((features & NETIF_F_LRO) && netvsc_xdp_get(nvdev)) {
1887 		features ^= NETIF_F_LRO;
1888 		netdev_info(ndev, "Skip LRO - unsupported with XDP\n");
1889 	}
1890 
1891 	return features;
1892 }
1893 
1894 static int netvsc_set_features(struct net_device *ndev,
1895 			       netdev_features_t features)
1896 {
1897 	netdev_features_t change = features ^ ndev->features;
1898 	struct net_device_context *ndevctx = netdev_priv(ndev);
1899 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1900 	struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1901 	struct ndis_offload_params offloads;
1902 	int ret = 0;
1903 
1904 	if (!nvdev || nvdev->destroy)
1905 		return -ENODEV;
1906 
1907 	if (!(change & NETIF_F_LRO))
1908 		goto syncvf;
1909 
1910 	memset(&offloads, 0, sizeof(struct ndis_offload_params));
1911 
1912 	if (features & NETIF_F_LRO) {
1913 		offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1914 		offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1915 	} else {
1916 		offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1917 		offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1918 	}
1919 
1920 	ret = rndis_filter_set_offload_params(ndev, nvdev, &offloads);
1921 
1922 	if (ret) {
1923 		features ^= NETIF_F_LRO;
1924 		ndev->features = features;
1925 	}
1926 
1927 syncvf:
1928 	if (!vf_netdev)
1929 		return ret;
1930 
1931 	vf_netdev->wanted_features = features;
1932 	netdev_update_features(vf_netdev);
1933 
1934 	return ret;
1935 }
1936 
1937 static u32 netvsc_get_msglevel(struct net_device *ndev)
1938 {
1939 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
1940 
1941 	return ndev_ctx->msg_enable;
1942 }
1943 
1944 static void netvsc_set_msglevel(struct net_device *ndev, u32 val)
1945 {
1946 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
1947 
1948 	ndev_ctx->msg_enable = val;
1949 }
1950 
1951 static const struct ethtool_ops ethtool_ops = {
1952 	.get_drvinfo	= netvsc_get_drvinfo,
1953 	.get_msglevel	= netvsc_get_msglevel,
1954 	.set_msglevel	= netvsc_set_msglevel,
1955 	.get_link	= ethtool_op_get_link,
1956 	.get_ethtool_stats = netvsc_get_ethtool_stats,
1957 	.get_sset_count = netvsc_get_sset_count,
1958 	.get_strings	= netvsc_get_strings,
1959 	.get_channels   = netvsc_get_channels,
1960 	.set_channels   = netvsc_set_channels,
1961 	.get_ts_info	= ethtool_op_get_ts_info,
1962 	.get_rxnfc	= netvsc_get_rxnfc,
1963 	.set_rxnfc	= netvsc_set_rxnfc,
1964 	.get_rxfh_key_size = netvsc_get_rxfh_key_size,
1965 	.get_rxfh_indir_size = netvsc_rss_indir_size,
1966 	.get_rxfh	= netvsc_get_rxfh,
1967 	.set_rxfh	= netvsc_set_rxfh,
1968 	.get_link_ksettings = netvsc_get_link_ksettings,
1969 	.set_link_ksettings = netvsc_set_link_ksettings,
1970 	.get_ringparam	= netvsc_get_ringparam,
1971 	.set_ringparam	= netvsc_set_ringparam,
1972 };
1973 
1974 static const struct net_device_ops device_ops = {
1975 	.ndo_open =			netvsc_open,
1976 	.ndo_stop =			netvsc_close,
1977 	.ndo_start_xmit =		netvsc_start_xmit,
1978 	.ndo_change_rx_flags =		netvsc_change_rx_flags,
1979 	.ndo_set_rx_mode =		netvsc_set_rx_mode,
1980 	.ndo_fix_features =		netvsc_fix_features,
1981 	.ndo_set_features =		netvsc_set_features,
1982 	.ndo_change_mtu =		netvsc_change_mtu,
1983 	.ndo_validate_addr =		eth_validate_addr,
1984 	.ndo_set_mac_address =		netvsc_set_mac_addr,
1985 	.ndo_select_queue =		netvsc_select_queue,
1986 	.ndo_get_stats64 =		netvsc_get_stats64,
1987 	.ndo_bpf =			netvsc_bpf,
1988 };
1989 
1990 /*
1991  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1992  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1993  * present send GARP packet to network peers with netif_notify_peers().
1994  */
1995 static void netvsc_link_change(struct work_struct *w)
1996 {
1997 	struct net_device_context *ndev_ctx =
1998 		container_of(w, struct net_device_context, dwork.work);
1999 	struct hv_device *device_obj = ndev_ctx->device_ctx;
2000 	struct net_device *net = hv_get_drvdata(device_obj);
2001 	struct netvsc_device *net_device;
2002 	struct rndis_device *rdev;
2003 	struct netvsc_reconfig *event = NULL;
2004 	bool notify = false, reschedule = false;
2005 	unsigned long flags, next_reconfig, delay;
2006 
2007 	/* if changes are happening, comeback later */
2008 	if (!rtnl_trylock()) {
2009 		schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
2010 		return;
2011 	}
2012 
2013 	net_device = rtnl_dereference(ndev_ctx->nvdev);
2014 	if (!net_device)
2015 		goto out_unlock;
2016 
2017 	rdev = net_device->extension;
2018 
2019 	next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
2020 	if (time_is_after_jiffies(next_reconfig)) {
2021 		/* link_watch only sends one notification with current state
2022 		 * per second, avoid doing reconfig more frequently. Handle
2023 		 * wrap around.
2024 		 */
2025 		delay = next_reconfig - jiffies;
2026 		delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
2027 		schedule_delayed_work(&ndev_ctx->dwork, delay);
2028 		goto out_unlock;
2029 	}
2030 	ndev_ctx->last_reconfig = jiffies;
2031 
2032 	spin_lock_irqsave(&ndev_ctx->lock, flags);
2033 	if (!list_empty(&ndev_ctx->reconfig_events)) {
2034 		event = list_first_entry(&ndev_ctx->reconfig_events,
2035 					 struct netvsc_reconfig, list);
2036 		list_del(&event->list);
2037 		reschedule = !list_empty(&ndev_ctx->reconfig_events);
2038 	}
2039 	spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2040 
2041 	if (!event)
2042 		goto out_unlock;
2043 
2044 	switch (event->event) {
2045 		/* Only the following events are possible due to the check in
2046 		 * netvsc_linkstatus_callback()
2047 		 */
2048 	case RNDIS_STATUS_MEDIA_CONNECT:
2049 		if (rdev->link_state) {
2050 			rdev->link_state = false;
2051 			netif_carrier_on(net);
2052 			netvsc_tx_enable(net_device, net);
2053 		} else {
2054 			notify = true;
2055 		}
2056 		kfree(event);
2057 		break;
2058 	case RNDIS_STATUS_MEDIA_DISCONNECT:
2059 		if (!rdev->link_state) {
2060 			rdev->link_state = true;
2061 			netif_carrier_off(net);
2062 			netvsc_tx_disable(net_device, net);
2063 		}
2064 		kfree(event);
2065 		break;
2066 	case RNDIS_STATUS_NETWORK_CHANGE:
2067 		/* Only makes sense if carrier is present */
2068 		if (!rdev->link_state) {
2069 			rdev->link_state = true;
2070 			netif_carrier_off(net);
2071 			netvsc_tx_disable(net_device, net);
2072 			event->event = RNDIS_STATUS_MEDIA_CONNECT;
2073 			spin_lock_irqsave(&ndev_ctx->lock, flags);
2074 			list_add(&event->list, &ndev_ctx->reconfig_events);
2075 			spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2076 			reschedule = true;
2077 		}
2078 		break;
2079 	}
2080 
2081 	rtnl_unlock();
2082 
2083 	if (notify)
2084 		netdev_notify_peers(net);
2085 
2086 	/* link_watch only sends one notification with current state per
2087 	 * second, handle next reconfig event in 2 seconds.
2088 	 */
2089 	if (reschedule)
2090 		schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
2091 
2092 	return;
2093 
2094 out_unlock:
2095 	rtnl_unlock();
2096 }
2097 
2098 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
2099 {
2100 	struct net_device_context *net_device_ctx;
2101 	struct net_device *dev;
2102 
2103 	dev = netdev_master_upper_dev_get(vf_netdev);
2104 	if (!dev || dev->netdev_ops != &device_ops)
2105 		return NULL;	/* not a netvsc device */
2106 
2107 	net_device_ctx = netdev_priv(dev);
2108 	if (!rtnl_dereference(net_device_ctx->nvdev))
2109 		return NULL;	/* device is removed */
2110 
2111 	return dev;
2112 }
2113 
2114 /* Called when VF is injecting data into network stack.
2115  * Change the associated network device from VF to netvsc.
2116  * note: already called with rcu_read_lock
2117  */
2118 static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
2119 {
2120 	struct sk_buff *skb = *pskb;
2121 	struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data);
2122 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
2123 	struct netvsc_vf_pcpu_stats *pcpu_stats
2124 		 = this_cpu_ptr(ndev_ctx->vf_stats);
2125 
2126 	skb = skb_share_check(skb, GFP_ATOMIC);
2127 	if (unlikely(!skb))
2128 		return RX_HANDLER_CONSUMED;
2129 
2130 	*pskb = skb;
2131 
2132 	skb->dev = ndev;
2133 
2134 	u64_stats_update_begin(&pcpu_stats->syncp);
2135 	pcpu_stats->rx_packets++;
2136 	pcpu_stats->rx_bytes += skb->len;
2137 	u64_stats_update_end(&pcpu_stats->syncp);
2138 
2139 	return RX_HANDLER_ANOTHER;
2140 }
2141 
2142 static int netvsc_vf_join(struct net_device *vf_netdev,
2143 			  struct net_device *ndev)
2144 {
2145 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
2146 	int ret;
2147 
2148 	ret = netdev_rx_handler_register(vf_netdev,
2149 					 netvsc_vf_handle_frame, ndev);
2150 	if (ret != 0) {
2151 		netdev_err(vf_netdev,
2152 			   "can not register netvsc VF receive handler (err = %d)\n",
2153 			   ret);
2154 		goto rx_handler_failed;
2155 	}
2156 
2157 	ret = netdev_master_upper_dev_link(vf_netdev, ndev,
2158 					   NULL, NULL, NULL);
2159 	if (ret != 0) {
2160 		netdev_err(vf_netdev,
2161 			   "can not set master device %s (err = %d)\n",
2162 			   ndev->name, ret);
2163 		goto upper_link_failed;
2164 	}
2165 
2166 	/* set slave flag before open to prevent IPv6 addrconf */
2167 	vf_netdev->flags |= IFF_SLAVE;
2168 
2169 	schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
2170 
2171 	call_netdevice_notifiers(NETDEV_JOIN, vf_netdev);
2172 
2173 	netdev_info(vf_netdev, "joined to %s\n", ndev->name);
2174 	return 0;
2175 
2176 upper_link_failed:
2177 	netdev_rx_handler_unregister(vf_netdev);
2178 rx_handler_failed:
2179 	return ret;
2180 }
2181 
2182 static void __netvsc_vf_setup(struct net_device *ndev,
2183 			      struct net_device *vf_netdev)
2184 {
2185 	int ret;
2186 
2187 	/* Align MTU of VF with master */
2188 	ret = dev_set_mtu(vf_netdev, ndev->mtu);
2189 	if (ret)
2190 		netdev_warn(vf_netdev,
2191 			    "unable to change mtu to %u\n", ndev->mtu);
2192 
2193 	/* set multicast etc flags on VF */
2194 	dev_change_flags(vf_netdev, ndev->flags | IFF_SLAVE, NULL);
2195 
2196 	/* sync address list from ndev to VF */
2197 	netif_addr_lock_bh(ndev);
2198 	dev_uc_sync(vf_netdev, ndev);
2199 	dev_mc_sync(vf_netdev, ndev);
2200 	netif_addr_unlock_bh(ndev);
2201 
2202 	if (netif_running(ndev)) {
2203 		ret = dev_open(vf_netdev, NULL);
2204 		if (ret)
2205 			netdev_warn(vf_netdev,
2206 				    "unable to open: %d\n", ret);
2207 	}
2208 }
2209 
2210 /* Setup VF as slave of the synthetic device.
2211  * Runs in workqueue to avoid recursion in netlink callbacks.
2212  */
2213 static void netvsc_vf_setup(struct work_struct *w)
2214 {
2215 	struct net_device_context *ndev_ctx
2216 		= container_of(w, struct net_device_context, vf_takeover.work);
2217 	struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2218 	struct net_device *vf_netdev;
2219 
2220 	if (!rtnl_trylock()) {
2221 		schedule_delayed_work(&ndev_ctx->vf_takeover, 0);
2222 		return;
2223 	}
2224 
2225 	vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2226 	if (vf_netdev)
2227 		__netvsc_vf_setup(ndev, vf_netdev);
2228 
2229 	rtnl_unlock();
2230 }
2231 
2232 /* Find netvsc by VF serial number.
2233  * The PCI hyperv controller records the serial number as the slot kobj name.
2234  */
2235 static struct net_device *get_netvsc_byslot(const struct net_device *vf_netdev)
2236 {
2237 	struct device *parent = vf_netdev->dev.parent;
2238 	struct net_device_context *ndev_ctx;
2239 	struct pci_dev *pdev;
2240 	u32 serial;
2241 
2242 	if (!parent || !dev_is_pci(parent))
2243 		return NULL; /* not a PCI device */
2244 
2245 	pdev = to_pci_dev(parent);
2246 	if (!pdev->slot) {
2247 		netdev_notice(vf_netdev, "no PCI slot information\n");
2248 		return NULL;
2249 	}
2250 
2251 	if (kstrtou32(pci_slot_name(pdev->slot), 10, &serial)) {
2252 		netdev_notice(vf_netdev, "Invalid vf serial:%s\n",
2253 			      pci_slot_name(pdev->slot));
2254 		return NULL;
2255 	}
2256 
2257 	list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
2258 		if (!ndev_ctx->vf_alloc)
2259 			continue;
2260 
2261 		if (ndev_ctx->vf_serial == serial)
2262 			return hv_get_drvdata(ndev_ctx->device_ctx);
2263 	}
2264 
2265 	netdev_notice(vf_netdev,
2266 		      "no netdev found for vf serial:%u\n", serial);
2267 	return NULL;
2268 }
2269 
2270 static int netvsc_register_vf(struct net_device *vf_netdev)
2271 {
2272 	struct net_device_context *net_device_ctx;
2273 	struct netvsc_device *netvsc_dev;
2274 	struct bpf_prog *prog;
2275 	struct net_device *ndev;
2276 	int ret;
2277 
2278 	if (vf_netdev->addr_len != ETH_ALEN)
2279 		return NOTIFY_DONE;
2280 
2281 	ndev = get_netvsc_byslot(vf_netdev);
2282 	if (!ndev)
2283 		return NOTIFY_DONE;
2284 
2285 	net_device_ctx = netdev_priv(ndev);
2286 	netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2287 	if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
2288 		return NOTIFY_DONE;
2289 
2290 	/* if synthetic interface is a different namespace,
2291 	 * then move the VF to that namespace; join will be
2292 	 * done again in that context.
2293 	 */
2294 	if (!net_eq(dev_net(ndev), dev_net(vf_netdev))) {
2295 		ret = dev_change_net_namespace(vf_netdev,
2296 					       dev_net(ndev), "eth%d");
2297 		if (ret)
2298 			netdev_err(vf_netdev,
2299 				   "could not move to same namespace as %s: %d\n",
2300 				   ndev->name, ret);
2301 		else
2302 			netdev_info(vf_netdev,
2303 				    "VF moved to namespace with: %s\n",
2304 				    ndev->name);
2305 		return NOTIFY_DONE;
2306 	}
2307 
2308 	netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
2309 
2310 	if (netvsc_vf_join(vf_netdev, ndev) != 0)
2311 		return NOTIFY_DONE;
2312 
2313 	dev_hold(vf_netdev);
2314 	rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
2315 
2316 	vf_netdev->wanted_features = ndev->features;
2317 	netdev_update_features(vf_netdev);
2318 
2319 	prog = netvsc_xdp_get(netvsc_dev);
2320 	netvsc_vf_setxdp(vf_netdev, prog);
2321 
2322 	return NOTIFY_OK;
2323 }
2324 
2325 /* VF up/down change detected, schedule to change data path */
2326 static int netvsc_vf_changed(struct net_device *vf_netdev)
2327 {
2328 	struct net_device_context *net_device_ctx;
2329 	struct netvsc_device *netvsc_dev;
2330 	struct net_device *ndev;
2331 	bool vf_is_up = netif_running(vf_netdev);
2332 
2333 	ndev = get_netvsc_byref(vf_netdev);
2334 	if (!ndev)
2335 		return NOTIFY_DONE;
2336 
2337 	net_device_ctx = netdev_priv(ndev);
2338 	netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2339 	if (!netvsc_dev)
2340 		return NOTIFY_DONE;
2341 
2342 	netvsc_switch_datapath(ndev, vf_is_up);
2343 	netdev_info(ndev, "Data path switched %s VF: %s\n",
2344 		    vf_is_up ? "to" : "from", vf_netdev->name);
2345 
2346 	return NOTIFY_OK;
2347 }
2348 
2349 static int netvsc_unregister_vf(struct net_device *vf_netdev)
2350 {
2351 	struct net_device *ndev;
2352 	struct net_device_context *net_device_ctx;
2353 
2354 	ndev = get_netvsc_byref(vf_netdev);
2355 	if (!ndev)
2356 		return NOTIFY_DONE;
2357 
2358 	net_device_ctx = netdev_priv(ndev);
2359 	cancel_delayed_work_sync(&net_device_ctx->vf_takeover);
2360 
2361 	netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
2362 
2363 	netvsc_vf_setxdp(vf_netdev, NULL);
2364 
2365 	netdev_rx_handler_unregister(vf_netdev);
2366 	netdev_upper_dev_unlink(vf_netdev, ndev);
2367 	RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
2368 	dev_put(vf_netdev);
2369 
2370 	return NOTIFY_OK;
2371 }
2372 
2373 static int netvsc_probe(struct hv_device *dev,
2374 			const struct hv_vmbus_device_id *dev_id)
2375 {
2376 	struct net_device *net = NULL;
2377 	struct net_device_context *net_device_ctx;
2378 	struct netvsc_device_info *device_info = NULL;
2379 	struct netvsc_device *nvdev;
2380 	int ret = -ENOMEM;
2381 
2382 	net = alloc_etherdev_mq(sizeof(struct net_device_context),
2383 				VRSS_CHANNEL_MAX);
2384 	if (!net)
2385 		goto no_net;
2386 
2387 	netif_carrier_off(net);
2388 
2389 	netvsc_init_settings(net);
2390 
2391 	net_device_ctx = netdev_priv(net);
2392 	net_device_ctx->device_ctx = dev;
2393 	net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
2394 	if (netif_msg_probe(net_device_ctx))
2395 		netdev_dbg(net, "netvsc msg_enable: %d\n",
2396 			   net_device_ctx->msg_enable);
2397 
2398 	hv_set_drvdata(dev, net);
2399 
2400 	INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
2401 
2402 	spin_lock_init(&net_device_ctx->lock);
2403 	INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
2404 	INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup);
2405 
2406 	net_device_ctx->vf_stats
2407 		= netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats);
2408 	if (!net_device_ctx->vf_stats)
2409 		goto no_stats;
2410 
2411 	net->netdev_ops = &device_ops;
2412 	net->ethtool_ops = &ethtool_ops;
2413 	SET_NETDEV_DEV(net, &dev->device);
2414 
2415 	/* We always need headroom for rndis header */
2416 	net->needed_headroom = RNDIS_AND_PPI_SIZE;
2417 
2418 	/* Initialize the number of queues to be 1, we may change it if more
2419 	 * channels are offered later.
2420 	 */
2421 	netif_set_real_num_tx_queues(net, 1);
2422 	netif_set_real_num_rx_queues(net, 1);
2423 
2424 	/* Notify the netvsc driver of the new device */
2425 	device_info = netvsc_devinfo_get(NULL);
2426 
2427 	if (!device_info) {
2428 		ret = -ENOMEM;
2429 		goto devinfo_failed;
2430 	}
2431 
2432 	nvdev = rndis_filter_device_add(dev, device_info);
2433 	if (IS_ERR(nvdev)) {
2434 		ret = PTR_ERR(nvdev);
2435 		netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
2436 		goto rndis_failed;
2437 	}
2438 
2439 	memcpy(net->dev_addr, device_info->mac_adr, ETH_ALEN);
2440 
2441 	/* We must get rtnl lock before scheduling nvdev->subchan_work,
2442 	 * otherwise netvsc_subchan_work() can get rtnl lock first and wait
2443 	 * all subchannels to show up, but that may not happen because
2444 	 * netvsc_probe() can't get rtnl lock and as a result vmbus_onoffer()
2445 	 * -> ... -> device_add() -> ... -> __device_attach() can't get
2446 	 * the device lock, so all the subchannels can't be processed --
2447 	 * finally netvsc_subchan_work() hangs forever.
2448 	 */
2449 	rtnl_lock();
2450 
2451 	if (nvdev->num_chn > 1)
2452 		schedule_work(&nvdev->subchan_work);
2453 
2454 	/* hw_features computed in rndis_netdev_set_hwcaps() */
2455 	net->features = net->hw_features |
2456 		NETIF_F_HIGHDMA | NETIF_F_HW_VLAN_CTAG_TX |
2457 		NETIF_F_HW_VLAN_CTAG_RX;
2458 	net->vlan_features = net->features;
2459 
2460 	/* MTU range: 68 - 1500 or 65521 */
2461 	net->min_mtu = NETVSC_MTU_MIN;
2462 	if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
2463 		net->max_mtu = NETVSC_MTU - ETH_HLEN;
2464 	else
2465 		net->max_mtu = ETH_DATA_LEN;
2466 
2467 	nvdev->tx_disable = false;
2468 
2469 	ret = register_netdevice(net);
2470 	if (ret != 0) {
2471 		pr_err("Unable to register netdev.\n");
2472 		goto register_failed;
2473 	}
2474 
2475 	list_add(&net_device_ctx->list, &netvsc_dev_list);
2476 	rtnl_unlock();
2477 
2478 	netvsc_devinfo_put(device_info);
2479 	return 0;
2480 
2481 register_failed:
2482 	rtnl_unlock();
2483 	rndis_filter_device_remove(dev, nvdev);
2484 rndis_failed:
2485 	netvsc_devinfo_put(device_info);
2486 devinfo_failed:
2487 	free_percpu(net_device_ctx->vf_stats);
2488 no_stats:
2489 	hv_set_drvdata(dev, NULL);
2490 	free_netdev(net);
2491 no_net:
2492 	return ret;
2493 }
2494 
2495 static int netvsc_remove(struct hv_device *dev)
2496 {
2497 	struct net_device_context *ndev_ctx;
2498 	struct net_device *vf_netdev, *net;
2499 	struct netvsc_device *nvdev;
2500 
2501 	net = hv_get_drvdata(dev);
2502 	if (net == NULL) {
2503 		dev_err(&dev->device, "No net device to remove\n");
2504 		return 0;
2505 	}
2506 
2507 	ndev_ctx = netdev_priv(net);
2508 
2509 	cancel_delayed_work_sync(&ndev_ctx->dwork);
2510 
2511 	rtnl_lock();
2512 	nvdev = rtnl_dereference(ndev_ctx->nvdev);
2513 	if (nvdev) {
2514 		cancel_work_sync(&nvdev->subchan_work);
2515 		netvsc_xdp_set(net, NULL, NULL, nvdev);
2516 	}
2517 
2518 	/*
2519 	 * Call to the vsc driver to let it know that the device is being
2520 	 * removed. Also blocks mtu and channel changes.
2521 	 */
2522 	vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2523 	if (vf_netdev)
2524 		netvsc_unregister_vf(vf_netdev);
2525 
2526 	if (nvdev)
2527 		rndis_filter_device_remove(dev, nvdev);
2528 
2529 	unregister_netdevice(net);
2530 	list_del(&ndev_ctx->list);
2531 
2532 	rtnl_unlock();
2533 
2534 	hv_set_drvdata(dev, NULL);
2535 
2536 	free_percpu(ndev_ctx->vf_stats);
2537 	free_netdev(net);
2538 	return 0;
2539 }
2540 
2541 static int netvsc_suspend(struct hv_device *dev)
2542 {
2543 	struct net_device_context *ndev_ctx;
2544 	struct net_device *vf_netdev, *net;
2545 	struct netvsc_device *nvdev;
2546 	int ret;
2547 
2548 	net = hv_get_drvdata(dev);
2549 
2550 	ndev_ctx = netdev_priv(net);
2551 	cancel_delayed_work_sync(&ndev_ctx->dwork);
2552 
2553 	rtnl_lock();
2554 
2555 	nvdev = rtnl_dereference(ndev_ctx->nvdev);
2556 	if (nvdev == NULL) {
2557 		ret = -ENODEV;
2558 		goto out;
2559 	}
2560 
2561 	vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2562 	if (vf_netdev)
2563 		netvsc_unregister_vf(vf_netdev);
2564 
2565 	/* Save the current config info */
2566 	ndev_ctx->saved_netvsc_dev_info = netvsc_devinfo_get(nvdev);
2567 
2568 	ret = netvsc_detach(net, nvdev);
2569 out:
2570 	rtnl_unlock();
2571 
2572 	return ret;
2573 }
2574 
2575 static int netvsc_resume(struct hv_device *dev)
2576 {
2577 	struct net_device *net = hv_get_drvdata(dev);
2578 	struct net_device_context *net_device_ctx;
2579 	struct netvsc_device_info *device_info;
2580 	int ret;
2581 
2582 	rtnl_lock();
2583 
2584 	net_device_ctx = netdev_priv(net);
2585 	device_info = net_device_ctx->saved_netvsc_dev_info;
2586 
2587 	ret = netvsc_attach(net, device_info);
2588 
2589 	netvsc_devinfo_put(device_info);
2590 	net_device_ctx->saved_netvsc_dev_info = NULL;
2591 
2592 	rtnl_unlock();
2593 
2594 	return ret;
2595 }
2596 static const struct hv_vmbus_device_id id_table[] = {
2597 	/* Network guid */
2598 	{ HV_NIC_GUID, },
2599 	{ },
2600 };
2601 
2602 MODULE_DEVICE_TABLE(vmbus, id_table);
2603 
2604 /* The one and only one */
2605 static struct  hv_driver netvsc_drv = {
2606 	.name = KBUILD_MODNAME,
2607 	.id_table = id_table,
2608 	.probe = netvsc_probe,
2609 	.remove = netvsc_remove,
2610 	.suspend = netvsc_suspend,
2611 	.resume = netvsc_resume,
2612 	.driver = {
2613 		.probe_type = PROBE_FORCE_SYNCHRONOUS,
2614 	},
2615 };
2616 
2617 /*
2618  * On Hyper-V, every VF interface is matched with a corresponding
2619  * synthetic interface. The synthetic interface is presented first
2620  * to the guest. When the corresponding VF instance is registered,
2621  * we will take care of switching the data path.
2622  */
2623 static int netvsc_netdev_event(struct notifier_block *this,
2624 			       unsigned long event, void *ptr)
2625 {
2626 	struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2627 
2628 	/* Skip our own events */
2629 	if (event_dev->netdev_ops == &device_ops)
2630 		return NOTIFY_DONE;
2631 
2632 	/* Avoid non-Ethernet type devices */
2633 	if (event_dev->type != ARPHRD_ETHER)
2634 		return NOTIFY_DONE;
2635 
2636 	/* Avoid Vlan dev with same MAC registering as VF */
2637 	if (is_vlan_dev(event_dev))
2638 		return NOTIFY_DONE;
2639 
2640 	/* Avoid Bonding master dev with same MAC registering as VF */
2641 	if ((event_dev->priv_flags & IFF_BONDING) &&
2642 	    (event_dev->flags & IFF_MASTER))
2643 		return NOTIFY_DONE;
2644 
2645 	switch (event) {
2646 	case NETDEV_REGISTER:
2647 		return netvsc_register_vf(event_dev);
2648 	case NETDEV_UNREGISTER:
2649 		return netvsc_unregister_vf(event_dev);
2650 	case NETDEV_UP:
2651 	case NETDEV_DOWN:
2652 		return netvsc_vf_changed(event_dev);
2653 	default:
2654 		return NOTIFY_DONE;
2655 	}
2656 }
2657 
2658 static struct notifier_block netvsc_netdev_notifier = {
2659 	.notifier_call = netvsc_netdev_event,
2660 };
2661 
2662 static void __exit netvsc_drv_exit(void)
2663 {
2664 	unregister_netdevice_notifier(&netvsc_netdev_notifier);
2665 	vmbus_driver_unregister(&netvsc_drv);
2666 }
2667 
2668 static int __init netvsc_drv_init(void)
2669 {
2670 	int ret;
2671 
2672 	if (ring_size < RING_SIZE_MIN) {
2673 		ring_size = RING_SIZE_MIN;
2674 		pr_info("Increased ring_size to %u (min allowed)\n",
2675 			ring_size);
2676 	}
2677 	netvsc_ring_bytes = ring_size * PAGE_SIZE;
2678 
2679 	ret = vmbus_driver_register(&netvsc_drv);
2680 	if (ret)
2681 		return ret;
2682 
2683 	register_netdevice_notifier(&netvsc_netdev_notifier);
2684 	return 0;
2685 }
2686 
2687 MODULE_LICENSE("GPL");
2688 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
2689 
2690 module_init(netvsc_drv_init);
2691 module_exit(netvsc_drv_exit);
2692