xref: /openbmc/linux/drivers/net/hyperv/netvsc_drv.c (revision dc6a81c3)
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 		bpf_prog_inc(prog);
1063 		ret = netvsc_xdp_set(ndev, prog, NULL, nvdev);
1064 		if (ret) {
1065 			bpf_prog_put(prog);
1066 			goto err1;
1067 		}
1068 	}
1069 
1070 	/* In any case device is now ready */
1071 	netif_device_attach(ndev);
1072 
1073 	/* Note: enable and attach happen when sub-channels setup */
1074 	netif_carrier_off(ndev);
1075 
1076 	if (netif_running(ndev)) {
1077 		ret = rndis_filter_open(nvdev);
1078 		if (ret)
1079 			goto err2;
1080 
1081 		rdev = nvdev->extension;
1082 		if (!rdev->link_state)
1083 			netif_carrier_on(ndev);
1084 	}
1085 
1086 	return 0;
1087 
1088 err2:
1089 	netif_device_detach(ndev);
1090 
1091 err1:
1092 	rndis_filter_device_remove(hdev, nvdev);
1093 
1094 	return ret;
1095 }
1096 
1097 static int netvsc_set_channels(struct net_device *net,
1098 			       struct ethtool_channels *channels)
1099 {
1100 	struct net_device_context *net_device_ctx = netdev_priv(net);
1101 	struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
1102 	unsigned int orig, count = channels->combined_count;
1103 	struct netvsc_device_info *device_info;
1104 	int ret;
1105 
1106 	/* We do not support separate count for rx, tx, or other */
1107 	if (count == 0 ||
1108 	    channels->rx_count || channels->tx_count || channels->other_count)
1109 		return -EINVAL;
1110 
1111 	if (!nvdev || nvdev->destroy)
1112 		return -ENODEV;
1113 
1114 	if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
1115 		return -EINVAL;
1116 
1117 	if (count > nvdev->max_chn)
1118 		return -EINVAL;
1119 
1120 	orig = nvdev->num_chn;
1121 
1122 	device_info = netvsc_devinfo_get(nvdev);
1123 
1124 	if (!device_info)
1125 		return -ENOMEM;
1126 
1127 	device_info->num_chn = count;
1128 
1129 	ret = netvsc_detach(net, nvdev);
1130 	if (ret)
1131 		goto out;
1132 
1133 	ret = netvsc_attach(net, device_info);
1134 	if (ret) {
1135 		device_info->num_chn = orig;
1136 		if (netvsc_attach(net, device_info))
1137 			netdev_err(net, "restoring channel setting failed\n");
1138 	}
1139 
1140 out:
1141 	netvsc_devinfo_put(device_info);
1142 	return ret;
1143 }
1144 
1145 static bool
1146 netvsc_validate_ethtool_ss_cmd(const struct ethtool_link_ksettings *cmd)
1147 {
1148 	struct ethtool_link_ksettings diff1 = *cmd;
1149 	struct ethtool_link_ksettings diff2 = {};
1150 
1151 	diff1.base.speed = 0;
1152 	diff1.base.duplex = 0;
1153 	/* advertising and cmd are usually set */
1154 	ethtool_link_ksettings_zero_link_mode(&diff1, advertising);
1155 	diff1.base.cmd = 0;
1156 	/* We set port to PORT_OTHER */
1157 	diff2.base.port = PORT_OTHER;
1158 
1159 	return !memcmp(&diff1, &diff2, sizeof(diff1));
1160 }
1161 
1162 static void netvsc_init_settings(struct net_device *dev)
1163 {
1164 	struct net_device_context *ndc = netdev_priv(dev);
1165 
1166 	ndc->l4_hash = HV_DEFAULT_L4HASH;
1167 
1168 	ndc->speed = SPEED_UNKNOWN;
1169 	ndc->duplex = DUPLEX_FULL;
1170 
1171 	dev->features = NETIF_F_LRO;
1172 }
1173 
1174 static int netvsc_get_link_ksettings(struct net_device *dev,
1175 				     struct ethtool_link_ksettings *cmd)
1176 {
1177 	struct net_device_context *ndc = netdev_priv(dev);
1178 
1179 	cmd->base.speed = ndc->speed;
1180 	cmd->base.duplex = ndc->duplex;
1181 	cmd->base.port = PORT_OTHER;
1182 
1183 	return 0;
1184 }
1185 
1186 static int netvsc_set_link_ksettings(struct net_device *dev,
1187 				     const struct ethtool_link_ksettings *cmd)
1188 {
1189 	struct net_device_context *ndc = netdev_priv(dev);
1190 	u32 speed;
1191 
1192 	speed = cmd->base.speed;
1193 	if (!ethtool_validate_speed(speed) ||
1194 	    !ethtool_validate_duplex(cmd->base.duplex) ||
1195 	    !netvsc_validate_ethtool_ss_cmd(cmd))
1196 		return -EINVAL;
1197 
1198 	ndc->speed = speed;
1199 	ndc->duplex = cmd->base.duplex;
1200 
1201 	return 0;
1202 }
1203 
1204 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
1205 {
1206 	struct net_device_context *ndevctx = netdev_priv(ndev);
1207 	struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1208 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1209 	int orig_mtu = ndev->mtu;
1210 	struct netvsc_device_info *device_info;
1211 	int ret = 0;
1212 
1213 	if (!nvdev || nvdev->destroy)
1214 		return -ENODEV;
1215 
1216 	device_info = netvsc_devinfo_get(nvdev);
1217 
1218 	if (!device_info)
1219 		return -ENOMEM;
1220 
1221 	/* Change MTU of underlying VF netdev first. */
1222 	if (vf_netdev) {
1223 		ret = dev_set_mtu(vf_netdev, mtu);
1224 		if (ret)
1225 			goto out;
1226 	}
1227 
1228 	ret = netvsc_detach(ndev, nvdev);
1229 	if (ret)
1230 		goto rollback_vf;
1231 
1232 	ndev->mtu = mtu;
1233 
1234 	ret = netvsc_attach(ndev, device_info);
1235 	if (!ret)
1236 		goto out;
1237 
1238 	/* Attempt rollback to original MTU */
1239 	ndev->mtu = orig_mtu;
1240 
1241 	if (netvsc_attach(ndev, device_info))
1242 		netdev_err(ndev, "restoring mtu failed\n");
1243 rollback_vf:
1244 	if (vf_netdev)
1245 		dev_set_mtu(vf_netdev, orig_mtu);
1246 
1247 out:
1248 	netvsc_devinfo_put(device_info);
1249 	return ret;
1250 }
1251 
1252 static void netvsc_get_vf_stats(struct net_device *net,
1253 				struct netvsc_vf_pcpu_stats *tot)
1254 {
1255 	struct net_device_context *ndev_ctx = netdev_priv(net);
1256 	int i;
1257 
1258 	memset(tot, 0, sizeof(*tot));
1259 
1260 	for_each_possible_cpu(i) {
1261 		const struct netvsc_vf_pcpu_stats *stats
1262 			= per_cpu_ptr(ndev_ctx->vf_stats, i);
1263 		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1264 		unsigned int start;
1265 
1266 		do {
1267 			start = u64_stats_fetch_begin_irq(&stats->syncp);
1268 			rx_packets = stats->rx_packets;
1269 			tx_packets = stats->tx_packets;
1270 			rx_bytes = stats->rx_bytes;
1271 			tx_bytes = stats->tx_bytes;
1272 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1273 
1274 		tot->rx_packets += rx_packets;
1275 		tot->tx_packets += tx_packets;
1276 		tot->rx_bytes   += rx_bytes;
1277 		tot->tx_bytes   += tx_bytes;
1278 		tot->tx_dropped += stats->tx_dropped;
1279 	}
1280 }
1281 
1282 static void netvsc_get_pcpu_stats(struct net_device *net,
1283 				  struct netvsc_ethtool_pcpu_stats *pcpu_tot)
1284 {
1285 	struct net_device_context *ndev_ctx = netdev_priv(net);
1286 	struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1287 	int i;
1288 
1289 	/* fetch percpu stats of vf */
1290 	for_each_possible_cpu(i) {
1291 		const struct netvsc_vf_pcpu_stats *stats =
1292 			per_cpu_ptr(ndev_ctx->vf_stats, i);
1293 		struct netvsc_ethtool_pcpu_stats *this_tot = &pcpu_tot[i];
1294 		unsigned int start;
1295 
1296 		do {
1297 			start = u64_stats_fetch_begin_irq(&stats->syncp);
1298 			this_tot->vf_rx_packets = stats->rx_packets;
1299 			this_tot->vf_tx_packets = stats->tx_packets;
1300 			this_tot->vf_rx_bytes = stats->rx_bytes;
1301 			this_tot->vf_tx_bytes = stats->tx_bytes;
1302 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1303 		this_tot->rx_packets = this_tot->vf_rx_packets;
1304 		this_tot->tx_packets = this_tot->vf_tx_packets;
1305 		this_tot->rx_bytes   = this_tot->vf_rx_bytes;
1306 		this_tot->tx_bytes   = this_tot->vf_tx_bytes;
1307 	}
1308 
1309 	/* fetch percpu stats of netvsc */
1310 	for (i = 0; i < nvdev->num_chn; i++) {
1311 		const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1312 		const struct netvsc_stats *stats;
1313 		struct netvsc_ethtool_pcpu_stats *this_tot =
1314 			&pcpu_tot[nvchan->channel->target_cpu];
1315 		u64 packets, bytes;
1316 		unsigned int start;
1317 
1318 		stats = &nvchan->tx_stats;
1319 		do {
1320 			start = u64_stats_fetch_begin_irq(&stats->syncp);
1321 			packets = stats->packets;
1322 			bytes = stats->bytes;
1323 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1324 
1325 		this_tot->tx_bytes	+= bytes;
1326 		this_tot->tx_packets	+= packets;
1327 
1328 		stats = &nvchan->rx_stats;
1329 		do {
1330 			start = u64_stats_fetch_begin_irq(&stats->syncp);
1331 			packets = stats->packets;
1332 			bytes = stats->bytes;
1333 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1334 
1335 		this_tot->rx_bytes	+= bytes;
1336 		this_tot->rx_packets	+= packets;
1337 	}
1338 }
1339 
1340 static void netvsc_get_stats64(struct net_device *net,
1341 			       struct rtnl_link_stats64 *t)
1342 {
1343 	struct net_device_context *ndev_ctx = netdev_priv(net);
1344 	struct netvsc_device *nvdev;
1345 	struct netvsc_vf_pcpu_stats vf_tot;
1346 	int i;
1347 
1348 	rcu_read_lock();
1349 
1350 	nvdev = rcu_dereference(ndev_ctx->nvdev);
1351 	if (!nvdev)
1352 		goto out;
1353 
1354 	netdev_stats_to_stats64(t, &net->stats);
1355 
1356 	netvsc_get_vf_stats(net, &vf_tot);
1357 	t->rx_packets += vf_tot.rx_packets;
1358 	t->tx_packets += vf_tot.tx_packets;
1359 	t->rx_bytes   += vf_tot.rx_bytes;
1360 	t->tx_bytes   += vf_tot.tx_bytes;
1361 	t->tx_dropped += vf_tot.tx_dropped;
1362 
1363 	for (i = 0; i < nvdev->num_chn; i++) {
1364 		const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1365 		const struct netvsc_stats *stats;
1366 		u64 packets, bytes, multicast;
1367 		unsigned int start;
1368 
1369 		stats = &nvchan->tx_stats;
1370 		do {
1371 			start = u64_stats_fetch_begin_irq(&stats->syncp);
1372 			packets = stats->packets;
1373 			bytes = stats->bytes;
1374 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1375 
1376 		t->tx_bytes	+= bytes;
1377 		t->tx_packets	+= packets;
1378 
1379 		stats = &nvchan->rx_stats;
1380 		do {
1381 			start = u64_stats_fetch_begin_irq(&stats->syncp);
1382 			packets = stats->packets;
1383 			bytes = stats->bytes;
1384 			multicast = stats->multicast + stats->broadcast;
1385 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1386 
1387 		t->rx_bytes	+= bytes;
1388 		t->rx_packets	+= packets;
1389 		t->multicast	+= multicast;
1390 	}
1391 out:
1392 	rcu_read_unlock();
1393 }
1394 
1395 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1396 {
1397 	struct net_device_context *ndc = netdev_priv(ndev);
1398 	struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1399 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1400 	struct sockaddr *addr = p;
1401 	int err;
1402 
1403 	err = eth_prepare_mac_addr_change(ndev, p);
1404 	if (err)
1405 		return err;
1406 
1407 	if (!nvdev)
1408 		return -ENODEV;
1409 
1410 	if (vf_netdev) {
1411 		err = dev_set_mac_address(vf_netdev, addr, NULL);
1412 		if (err)
1413 			return err;
1414 	}
1415 
1416 	err = rndis_filter_set_device_mac(nvdev, addr->sa_data);
1417 	if (!err) {
1418 		eth_commit_mac_addr_change(ndev, p);
1419 	} else if (vf_netdev) {
1420 		/* rollback change on VF */
1421 		memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN);
1422 		dev_set_mac_address(vf_netdev, addr, NULL);
1423 	}
1424 
1425 	return err;
1426 }
1427 
1428 static const struct {
1429 	char name[ETH_GSTRING_LEN];
1430 	u16 offset;
1431 } netvsc_stats[] = {
1432 	{ "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1433 	{ "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1434 	{ "tx_no_space",  offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1435 	{ "tx_too_big",	  offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1436 	{ "tx_busy",	  offsetof(struct netvsc_ethtool_stats, tx_busy) },
1437 	{ "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) },
1438 	{ "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) },
1439 	{ "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) },
1440 	{ "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) },
1441 	{ "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) },
1442 }, pcpu_stats[] = {
1443 	{ "cpu%u_rx_packets",
1444 		offsetof(struct netvsc_ethtool_pcpu_stats, rx_packets) },
1445 	{ "cpu%u_rx_bytes",
1446 		offsetof(struct netvsc_ethtool_pcpu_stats, rx_bytes) },
1447 	{ "cpu%u_tx_packets",
1448 		offsetof(struct netvsc_ethtool_pcpu_stats, tx_packets) },
1449 	{ "cpu%u_tx_bytes",
1450 		offsetof(struct netvsc_ethtool_pcpu_stats, tx_bytes) },
1451 	{ "cpu%u_vf_rx_packets",
1452 		offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_packets) },
1453 	{ "cpu%u_vf_rx_bytes",
1454 		offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_bytes) },
1455 	{ "cpu%u_vf_tx_packets",
1456 		offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_packets) },
1457 	{ "cpu%u_vf_tx_bytes",
1458 		offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_bytes) },
1459 }, vf_stats[] = {
1460 	{ "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) },
1461 	{ "vf_rx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) },
1462 	{ "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) },
1463 	{ "vf_tx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) },
1464 	{ "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) },
1465 };
1466 
1467 #define NETVSC_GLOBAL_STATS_LEN	ARRAY_SIZE(netvsc_stats)
1468 #define NETVSC_VF_STATS_LEN	ARRAY_SIZE(vf_stats)
1469 
1470 /* statistics per queue (rx/tx packets/bytes) */
1471 #define NETVSC_PCPU_STATS_LEN (num_present_cpus() * ARRAY_SIZE(pcpu_stats))
1472 
1473 /* 5 statistics per queue (rx/tx packets/bytes, rx xdp_drop) */
1474 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 5)
1475 
1476 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1477 {
1478 	struct net_device_context *ndc = netdev_priv(dev);
1479 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1480 
1481 	if (!nvdev)
1482 		return -ENODEV;
1483 
1484 	switch (string_set) {
1485 	case ETH_SS_STATS:
1486 		return NETVSC_GLOBAL_STATS_LEN
1487 			+ NETVSC_VF_STATS_LEN
1488 			+ NETVSC_QUEUE_STATS_LEN(nvdev)
1489 			+ NETVSC_PCPU_STATS_LEN;
1490 	default:
1491 		return -EINVAL;
1492 	}
1493 }
1494 
1495 static void netvsc_get_ethtool_stats(struct net_device *dev,
1496 				     struct ethtool_stats *stats, u64 *data)
1497 {
1498 	struct net_device_context *ndc = netdev_priv(dev);
1499 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1500 	const void *nds = &ndc->eth_stats;
1501 	const struct netvsc_stats *qstats;
1502 	struct netvsc_vf_pcpu_stats sum;
1503 	struct netvsc_ethtool_pcpu_stats *pcpu_sum;
1504 	unsigned int start;
1505 	u64 packets, bytes;
1506 	u64 xdp_drop;
1507 	int i, j, cpu;
1508 
1509 	if (!nvdev)
1510 		return;
1511 
1512 	for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1513 		data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1514 
1515 	netvsc_get_vf_stats(dev, &sum);
1516 	for (j = 0; j < NETVSC_VF_STATS_LEN; j++)
1517 		data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset);
1518 
1519 	for (j = 0; j < nvdev->num_chn; j++) {
1520 		qstats = &nvdev->chan_table[j].tx_stats;
1521 
1522 		do {
1523 			start = u64_stats_fetch_begin_irq(&qstats->syncp);
1524 			packets = qstats->packets;
1525 			bytes = qstats->bytes;
1526 		} while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1527 		data[i++] = packets;
1528 		data[i++] = bytes;
1529 
1530 		qstats = &nvdev->chan_table[j].rx_stats;
1531 		do {
1532 			start = u64_stats_fetch_begin_irq(&qstats->syncp);
1533 			packets = qstats->packets;
1534 			bytes = qstats->bytes;
1535 			xdp_drop = qstats->xdp_drop;
1536 		} while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1537 		data[i++] = packets;
1538 		data[i++] = bytes;
1539 		data[i++] = xdp_drop;
1540 	}
1541 
1542 	pcpu_sum = kvmalloc_array(num_possible_cpus(),
1543 				  sizeof(struct netvsc_ethtool_pcpu_stats),
1544 				  GFP_KERNEL);
1545 	netvsc_get_pcpu_stats(dev, pcpu_sum);
1546 	for_each_present_cpu(cpu) {
1547 		struct netvsc_ethtool_pcpu_stats *this_sum = &pcpu_sum[cpu];
1548 
1549 		for (j = 0; j < ARRAY_SIZE(pcpu_stats); j++)
1550 			data[i++] = *(u64 *)((void *)this_sum
1551 					     + pcpu_stats[j].offset);
1552 	}
1553 	kvfree(pcpu_sum);
1554 }
1555 
1556 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1557 {
1558 	struct net_device_context *ndc = netdev_priv(dev);
1559 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1560 	u8 *p = data;
1561 	int i, cpu;
1562 
1563 	if (!nvdev)
1564 		return;
1565 
1566 	switch (stringset) {
1567 	case ETH_SS_STATS:
1568 		for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++) {
1569 			memcpy(p, netvsc_stats[i].name, ETH_GSTRING_LEN);
1570 			p += ETH_GSTRING_LEN;
1571 		}
1572 
1573 		for (i = 0; i < ARRAY_SIZE(vf_stats); i++) {
1574 			memcpy(p, vf_stats[i].name, ETH_GSTRING_LEN);
1575 			p += ETH_GSTRING_LEN;
1576 		}
1577 
1578 		for (i = 0; i < nvdev->num_chn; i++) {
1579 			sprintf(p, "tx_queue_%u_packets", i);
1580 			p += ETH_GSTRING_LEN;
1581 			sprintf(p, "tx_queue_%u_bytes", i);
1582 			p += ETH_GSTRING_LEN;
1583 			sprintf(p, "rx_queue_%u_packets", i);
1584 			p += ETH_GSTRING_LEN;
1585 			sprintf(p, "rx_queue_%u_bytes", i);
1586 			p += ETH_GSTRING_LEN;
1587 			sprintf(p, "rx_queue_%u_xdp_drop", i);
1588 			p += ETH_GSTRING_LEN;
1589 		}
1590 
1591 		for_each_present_cpu(cpu) {
1592 			for (i = 0; i < ARRAY_SIZE(pcpu_stats); i++) {
1593 				sprintf(p, pcpu_stats[i].name, cpu);
1594 				p += ETH_GSTRING_LEN;
1595 			}
1596 		}
1597 
1598 		break;
1599 	}
1600 }
1601 
1602 static int
1603 netvsc_get_rss_hash_opts(struct net_device_context *ndc,
1604 			 struct ethtool_rxnfc *info)
1605 {
1606 	const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3;
1607 
1608 	info->data = RXH_IP_SRC | RXH_IP_DST;
1609 
1610 	switch (info->flow_type) {
1611 	case TCP_V4_FLOW:
1612 		if (ndc->l4_hash & HV_TCP4_L4HASH)
1613 			info->data |= l4_flag;
1614 
1615 		break;
1616 
1617 	case TCP_V6_FLOW:
1618 		if (ndc->l4_hash & HV_TCP6_L4HASH)
1619 			info->data |= l4_flag;
1620 
1621 		break;
1622 
1623 	case UDP_V4_FLOW:
1624 		if (ndc->l4_hash & HV_UDP4_L4HASH)
1625 			info->data |= l4_flag;
1626 
1627 		break;
1628 
1629 	case UDP_V6_FLOW:
1630 		if (ndc->l4_hash & HV_UDP6_L4HASH)
1631 			info->data |= l4_flag;
1632 
1633 		break;
1634 
1635 	case IPV4_FLOW:
1636 	case IPV6_FLOW:
1637 		break;
1638 	default:
1639 		info->data = 0;
1640 		break;
1641 	}
1642 
1643 	return 0;
1644 }
1645 
1646 static int
1647 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1648 		 u32 *rules)
1649 {
1650 	struct net_device_context *ndc = netdev_priv(dev);
1651 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1652 
1653 	if (!nvdev)
1654 		return -ENODEV;
1655 
1656 	switch (info->cmd) {
1657 	case ETHTOOL_GRXRINGS:
1658 		info->data = nvdev->num_chn;
1659 		return 0;
1660 
1661 	case ETHTOOL_GRXFH:
1662 		return netvsc_get_rss_hash_opts(ndc, info);
1663 	}
1664 	return -EOPNOTSUPP;
1665 }
1666 
1667 static int netvsc_set_rss_hash_opts(struct net_device_context *ndc,
1668 				    struct ethtool_rxnfc *info)
1669 {
1670 	if (info->data == (RXH_IP_SRC | RXH_IP_DST |
1671 			   RXH_L4_B_0_1 | RXH_L4_B_2_3)) {
1672 		switch (info->flow_type) {
1673 		case TCP_V4_FLOW:
1674 			ndc->l4_hash |= HV_TCP4_L4HASH;
1675 			break;
1676 
1677 		case TCP_V6_FLOW:
1678 			ndc->l4_hash |= HV_TCP6_L4HASH;
1679 			break;
1680 
1681 		case UDP_V4_FLOW:
1682 			ndc->l4_hash |= HV_UDP4_L4HASH;
1683 			break;
1684 
1685 		case UDP_V6_FLOW:
1686 			ndc->l4_hash |= HV_UDP6_L4HASH;
1687 			break;
1688 
1689 		default:
1690 			return -EOPNOTSUPP;
1691 		}
1692 
1693 		return 0;
1694 	}
1695 
1696 	if (info->data == (RXH_IP_SRC | RXH_IP_DST)) {
1697 		switch (info->flow_type) {
1698 		case TCP_V4_FLOW:
1699 			ndc->l4_hash &= ~HV_TCP4_L4HASH;
1700 			break;
1701 
1702 		case TCP_V6_FLOW:
1703 			ndc->l4_hash &= ~HV_TCP6_L4HASH;
1704 			break;
1705 
1706 		case UDP_V4_FLOW:
1707 			ndc->l4_hash &= ~HV_UDP4_L4HASH;
1708 			break;
1709 
1710 		case UDP_V6_FLOW:
1711 			ndc->l4_hash &= ~HV_UDP6_L4HASH;
1712 			break;
1713 
1714 		default:
1715 			return -EOPNOTSUPP;
1716 		}
1717 
1718 		return 0;
1719 	}
1720 
1721 	return -EOPNOTSUPP;
1722 }
1723 
1724 static int
1725 netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info)
1726 {
1727 	struct net_device_context *ndc = netdev_priv(ndev);
1728 
1729 	if (info->cmd == ETHTOOL_SRXFH)
1730 		return netvsc_set_rss_hash_opts(ndc, info);
1731 
1732 	return -EOPNOTSUPP;
1733 }
1734 
1735 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1736 {
1737 	return NETVSC_HASH_KEYLEN;
1738 }
1739 
1740 static u32 netvsc_rss_indir_size(struct net_device *dev)
1741 {
1742 	return ITAB_NUM;
1743 }
1744 
1745 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1746 			   u8 *hfunc)
1747 {
1748 	struct net_device_context *ndc = netdev_priv(dev);
1749 	struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1750 	struct rndis_device *rndis_dev;
1751 	int i;
1752 
1753 	if (!ndev)
1754 		return -ENODEV;
1755 
1756 	if (hfunc)
1757 		*hfunc = ETH_RSS_HASH_TOP;	/* Toeplitz */
1758 
1759 	rndis_dev = ndev->extension;
1760 	if (indir) {
1761 		for (i = 0; i < ITAB_NUM; i++)
1762 			indir[i] = ndc->rx_table[i];
1763 	}
1764 
1765 	if (key)
1766 		memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1767 
1768 	return 0;
1769 }
1770 
1771 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1772 			   const u8 *key, const u8 hfunc)
1773 {
1774 	struct net_device_context *ndc = netdev_priv(dev);
1775 	struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1776 	struct rndis_device *rndis_dev;
1777 	int i;
1778 
1779 	if (!ndev)
1780 		return -ENODEV;
1781 
1782 	if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1783 		return -EOPNOTSUPP;
1784 
1785 	rndis_dev = ndev->extension;
1786 	if (indir) {
1787 		for (i = 0; i < ITAB_NUM; i++)
1788 			if (indir[i] >= ndev->num_chn)
1789 				return -EINVAL;
1790 
1791 		for (i = 0; i < ITAB_NUM; i++)
1792 			ndc->rx_table[i] = indir[i];
1793 	}
1794 
1795 	if (!key) {
1796 		if (!indir)
1797 			return 0;
1798 
1799 		key = rndis_dev->rss_key;
1800 	}
1801 
1802 	return rndis_filter_set_rss_param(rndis_dev, key);
1803 }
1804 
1805 /* Hyper-V RNDIS protocol does not have ring in the HW sense.
1806  * It does have pre-allocated receive area which is divided into sections.
1807  */
1808 static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
1809 				   struct ethtool_ringparam *ring)
1810 {
1811 	u32 max_buf_size;
1812 
1813 	ring->rx_pending = nvdev->recv_section_cnt;
1814 	ring->tx_pending = nvdev->send_section_cnt;
1815 
1816 	if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
1817 		max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
1818 	else
1819 		max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1820 
1821 	ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
1822 	ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
1823 		/ nvdev->send_section_size;
1824 }
1825 
1826 static void netvsc_get_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 
1832 	if (!nvdev)
1833 		return;
1834 
1835 	__netvsc_get_ringparam(nvdev, ring);
1836 }
1837 
1838 static int netvsc_set_ringparam(struct net_device *ndev,
1839 				struct ethtool_ringparam *ring)
1840 {
1841 	struct net_device_context *ndevctx = netdev_priv(ndev);
1842 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1843 	struct netvsc_device_info *device_info;
1844 	struct ethtool_ringparam orig;
1845 	u32 new_tx, new_rx;
1846 	int ret = 0;
1847 
1848 	if (!nvdev || nvdev->destroy)
1849 		return -ENODEV;
1850 
1851 	memset(&orig, 0, sizeof(orig));
1852 	__netvsc_get_ringparam(nvdev, &orig);
1853 
1854 	new_tx = clamp_t(u32, ring->tx_pending,
1855 			 NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
1856 	new_rx = clamp_t(u32, ring->rx_pending,
1857 			 NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
1858 
1859 	if (new_tx == orig.tx_pending &&
1860 	    new_rx == orig.rx_pending)
1861 		return 0;	 /* no change */
1862 
1863 	device_info = netvsc_devinfo_get(nvdev);
1864 
1865 	if (!device_info)
1866 		return -ENOMEM;
1867 
1868 	device_info->send_sections = new_tx;
1869 	device_info->recv_sections = new_rx;
1870 
1871 	ret = netvsc_detach(ndev, nvdev);
1872 	if (ret)
1873 		goto out;
1874 
1875 	ret = netvsc_attach(ndev, device_info);
1876 	if (ret) {
1877 		device_info->send_sections = orig.tx_pending;
1878 		device_info->recv_sections = orig.rx_pending;
1879 
1880 		if (netvsc_attach(ndev, device_info))
1881 			netdev_err(ndev, "restoring ringparam failed");
1882 	}
1883 
1884 out:
1885 	netvsc_devinfo_put(device_info);
1886 	return ret;
1887 }
1888 
1889 static netdev_features_t netvsc_fix_features(struct net_device *ndev,
1890 					     netdev_features_t features)
1891 {
1892 	struct net_device_context *ndevctx = netdev_priv(ndev);
1893 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1894 
1895 	if (!nvdev || nvdev->destroy)
1896 		return features;
1897 
1898 	if ((features & NETIF_F_LRO) && netvsc_xdp_get(nvdev)) {
1899 		features ^= NETIF_F_LRO;
1900 		netdev_info(ndev, "Skip LRO - unsupported with XDP\n");
1901 	}
1902 
1903 	return features;
1904 }
1905 
1906 static int netvsc_set_features(struct net_device *ndev,
1907 			       netdev_features_t features)
1908 {
1909 	netdev_features_t change = features ^ ndev->features;
1910 	struct net_device_context *ndevctx = netdev_priv(ndev);
1911 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1912 	struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1913 	struct ndis_offload_params offloads;
1914 	int ret = 0;
1915 
1916 	if (!nvdev || nvdev->destroy)
1917 		return -ENODEV;
1918 
1919 	if (!(change & NETIF_F_LRO))
1920 		goto syncvf;
1921 
1922 	memset(&offloads, 0, sizeof(struct ndis_offload_params));
1923 
1924 	if (features & NETIF_F_LRO) {
1925 		offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1926 		offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1927 	} else {
1928 		offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1929 		offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1930 	}
1931 
1932 	ret = rndis_filter_set_offload_params(ndev, nvdev, &offloads);
1933 
1934 	if (ret) {
1935 		features ^= NETIF_F_LRO;
1936 		ndev->features = features;
1937 	}
1938 
1939 syncvf:
1940 	if (!vf_netdev)
1941 		return ret;
1942 
1943 	vf_netdev->wanted_features = features;
1944 	netdev_update_features(vf_netdev);
1945 
1946 	return ret;
1947 }
1948 
1949 static u32 netvsc_get_msglevel(struct net_device *ndev)
1950 {
1951 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
1952 
1953 	return ndev_ctx->msg_enable;
1954 }
1955 
1956 static void netvsc_set_msglevel(struct net_device *ndev, u32 val)
1957 {
1958 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
1959 
1960 	ndev_ctx->msg_enable = val;
1961 }
1962 
1963 static const struct ethtool_ops ethtool_ops = {
1964 	.get_drvinfo	= netvsc_get_drvinfo,
1965 	.get_msglevel	= netvsc_get_msglevel,
1966 	.set_msglevel	= netvsc_set_msglevel,
1967 	.get_link	= ethtool_op_get_link,
1968 	.get_ethtool_stats = netvsc_get_ethtool_stats,
1969 	.get_sset_count = netvsc_get_sset_count,
1970 	.get_strings	= netvsc_get_strings,
1971 	.get_channels   = netvsc_get_channels,
1972 	.set_channels   = netvsc_set_channels,
1973 	.get_ts_info	= ethtool_op_get_ts_info,
1974 	.get_rxnfc	= netvsc_get_rxnfc,
1975 	.set_rxnfc	= netvsc_set_rxnfc,
1976 	.get_rxfh_key_size = netvsc_get_rxfh_key_size,
1977 	.get_rxfh_indir_size = netvsc_rss_indir_size,
1978 	.get_rxfh	= netvsc_get_rxfh,
1979 	.set_rxfh	= netvsc_set_rxfh,
1980 	.get_link_ksettings = netvsc_get_link_ksettings,
1981 	.set_link_ksettings = netvsc_set_link_ksettings,
1982 	.get_ringparam	= netvsc_get_ringparam,
1983 	.set_ringparam	= netvsc_set_ringparam,
1984 };
1985 
1986 static const struct net_device_ops device_ops = {
1987 	.ndo_open =			netvsc_open,
1988 	.ndo_stop =			netvsc_close,
1989 	.ndo_start_xmit =		netvsc_start_xmit,
1990 	.ndo_change_rx_flags =		netvsc_change_rx_flags,
1991 	.ndo_set_rx_mode =		netvsc_set_rx_mode,
1992 	.ndo_fix_features =		netvsc_fix_features,
1993 	.ndo_set_features =		netvsc_set_features,
1994 	.ndo_change_mtu =		netvsc_change_mtu,
1995 	.ndo_validate_addr =		eth_validate_addr,
1996 	.ndo_set_mac_address =		netvsc_set_mac_addr,
1997 	.ndo_select_queue =		netvsc_select_queue,
1998 	.ndo_get_stats64 =		netvsc_get_stats64,
1999 	.ndo_bpf =			netvsc_bpf,
2000 };
2001 
2002 /*
2003  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
2004  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
2005  * present send GARP packet to network peers with netif_notify_peers().
2006  */
2007 static void netvsc_link_change(struct work_struct *w)
2008 {
2009 	struct net_device_context *ndev_ctx =
2010 		container_of(w, struct net_device_context, dwork.work);
2011 	struct hv_device *device_obj = ndev_ctx->device_ctx;
2012 	struct net_device *net = hv_get_drvdata(device_obj);
2013 	struct netvsc_device *net_device;
2014 	struct rndis_device *rdev;
2015 	struct netvsc_reconfig *event = NULL;
2016 	bool notify = false, reschedule = false;
2017 	unsigned long flags, next_reconfig, delay;
2018 
2019 	/* if changes are happening, comeback later */
2020 	if (!rtnl_trylock()) {
2021 		schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
2022 		return;
2023 	}
2024 
2025 	net_device = rtnl_dereference(ndev_ctx->nvdev);
2026 	if (!net_device)
2027 		goto out_unlock;
2028 
2029 	rdev = net_device->extension;
2030 
2031 	next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
2032 	if (time_is_after_jiffies(next_reconfig)) {
2033 		/* link_watch only sends one notification with current state
2034 		 * per second, avoid doing reconfig more frequently. Handle
2035 		 * wrap around.
2036 		 */
2037 		delay = next_reconfig - jiffies;
2038 		delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
2039 		schedule_delayed_work(&ndev_ctx->dwork, delay);
2040 		goto out_unlock;
2041 	}
2042 	ndev_ctx->last_reconfig = jiffies;
2043 
2044 	spin_lock_irqsave(&ndev_ctx->lock, flags);
2045 	if (!list_empty(&ndev_ctx->reconfig_events)) {
2046 		event = list_first_entry(&ndev_ctx->reconfig_events,
2047 					 struct netvsc_reconfig, list);
2048 		list_del(&event->list);
2049 		reschedule = !list_empty(&ndev_ctx->reconfig_events);
2050 	}
2051 	spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2052 
2053 	if (!event)
2054 		goto out_unlock;
2055 
2056 	switch (event->event) {
2057 		/* Only the following events are possible due to the check in
2058 		 * netvsc_linkstatus_callback()
2059 		 */
2060 	case RNDIS_STATUS_MEDIA_CONNECT:
2061 		if (rdev->link_state) {
2062 			rdev->link_state = false;
2063 			netif_carrier_on(net);
2064 			netvsc_tx_enable(net_device, net);
2065 		} else {
2066 			notify = true;
2067 		}
2068 		kfree(event);
2069 		break;
2070 	case RNDIS_STATUS_MEDIA_DISCONNECT:
2071 		if (!rdev->link_state) {
2072 			rdev->link_state = true;
2073 			netif_carrier_off(net);
2074 			netvsc_tx_disable(net_device, net);
2075 		}
2076 		kfree(event);
2077 		break;
2078 	case RNDIS_STATUS_NETWORK_CHANGE:
2079 		/* Only makes sense if carrier is present */
2080 		if (!rdev->link_state) {
2081 			rdev->link_state = true;
2082 			netif_carrier_off(net);
2083 			netvsc_tx_disable(net_device, net);
2084 			event->event = RNDIS_STATUS_MEDIA_CONNECT;
2085 			spin_lock_irqsave(&ndev_ctx->lock, flags);
2086 			list_add(&event->list, &ndev_ctx->reconfig_events);
2087 			spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2088 			reschedule = true;
2089 		}
2090 		break;
2091 	}
2092 
2093 	rtnl_unlock();
2094 
2095 	if (notify)
2096 		netdev_notify_peers(net);
2097 
2098 	/* link_watch only sends one notification with current state per
2099 	 * second, handle next reconfig event in 2 seconds.
2100 	 */
2101 	if (reschedule)
2102 		schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
2103 
2104 	return;
2105 
2106 out_unlock:
2107 	rtnl_unlock();
2108 }
2109 
2110 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
2111 {
2112 	struct net_device_context *net_device_ctx;
2113 	struct net_device *dev;
2114 
2115 	dev = netdev_master_upper_dev_get(vf_netdev);
2116 	if (!dev || dev->netdev_ops != &device_ops)
2117 		return NULL;	/* not a netvsc device */
2118 
2119 	net_device_ctx = netdev_priv(dev);
2120 	if (!rtnl_dereference(net_device_ctx->nvdev))
2121 		return NULL;	/* device is removed */
2122 
2123 	return dev;
2124 }
2125 
2126 /* Called when VF is injecting data into network stack.
2127  * Change the associated network device from VF to netvsc.
2128  * note: already called with rcu_read_lock
2129  */
2130 static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
2131 {
2132 	struct sk_buff *skb = *pskb;
2133 	struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data);
2134 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
2135 	struct netvsc_vf_pcpu_stats *pcpu_stats
2136 		 = this_cpu_ptr(ndev_ctx->vf_stats);
2137 
2138 	skb = skb_share_check(skb, GFP_ATOMIC);
2139 	if (unlikely(!skb))
2140 		return RX_HANDLER_CONSUMED;
2141 
2142 	*pskb = skb;
2143 
2144 	skb->dev = ndev;
2145 
2146 	u64_stats_update_begin(&pcpu_stats->syncp);
2147 	pcpu_stats->rx_packets++;
2148 	pcpu_stats->rx_bytes += skb->len;
2149 	u64_stats_update_end(&pcpu_stats->syncp);
2150 
2151 	return RX_HANDLER_ANOTHER;
2152 }
2153 
2154 static int netvsc_vf_join(struct net_device *vf_netdev,
2155 			  struct net_device *ndev)
2156 {
2157 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
2158 	int ret;
2159 
2160 	ret = netdev_rx_handler_register(vf_netdev,
2161 					 netvsc_vf_handle_frame, ndev);
2162 	if (ret != 0) {
2163 		netdev_err(vf_netdev,
2164 			   "can not register netvsc VF receive handler (err = %d)\n",
2165 			   ret);
2166 		goto rx_handler_failed;
2167 	}
2168 
2169 	ret = netdev_master_upper_dev_link(vf_netdev, ndev,
2170 					   NULL, NULL, NULL);
2171 	if (ret != 0) {
2172 		netdev_err(vf_netdev,
2173 			   "can not set master device %s (err = %d)\n",
2174 			   ndev->name, ret);
2175 		goto upper_link_failed;
2176 	}
2177 
2178 	/* set slave flag before open to prevent IPv6 addrconf */
2179 	vf_netdev->flags |= IFF_SLAVE;
2180 
2181 	schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
2182 
2183 	call_netdevice_notifiers(NETDEV_JOIN, vf_netdev);
2184 
2185 	netdev_info(vf_netdev, "joined to %s\n", ndev->name);
2186 	return 0;
2187 
2188 upper_link_failed:
2189 	netdev_rx_handler_unregister(vf_netdev);
2190 rx_handler_failed:
2191 	return ret;
2192 }
2193 
2194 static void __netvsc_vf_setup(struct net_device *ndev,
2195 			      struct net_device *vf_netdev)
2196 {
2197 	int ret;
2198 
2199 	/* Align MTU of VF with master */
2200 	ret = dev_set_mtu(vf_netdev, ndev->mtu);
2201 	if (ret)
2202 		netdev_warn(vf_netdev,
2203 			    "unable to change mtu to %u\n", ndev->mtu);
2204 
2205 	/* set multicast etc flags on VF */
2206 	dev_change_flags(vf_netdev, ndev->flags | IFF_SLAVE, NULL);
2207 
2208 	/* sync address list from ndev to VF */
2209 	netif_addr_lock_bh(ndev);
2210 	dev_uc_sync(vf_netdev, ndev);
2211 	dev_mc_sync(vf_netdev, ndev);
2212 	netif_addr_unlock_bh(ndev);
2213 
2214 	if (netif_running(ndev)) {
2215 		ret = dev_open(vf_netdev, NULL);
2216 		if (ret)
2217 			netdev_warn(vf_netdev,
2218 				    "unable to open: %d\n", ret);
2219 	}
2220 }
2221 
2222 /* Setup VF as slave of the synthetic device.
2223  * Runs in workqueue to avoid recursion in netlink callbacks.
2224  */
2225 static void netvsc_vf_setup(struct work_struct *w)
2226 {
2227 	struct net_device_context *ndev_ctx
2228 		= container_of(w, struct net_device_context, vf_takeover.work);
2229 	struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2230 	struct net_device *vf_netdev;
2231 
2232 	if (!rtnl_trylock()) {
2233 		schedule_delayed_work(&ndev_ctx->vf_takeover, 0);
2234 		return;
2235 	}
2236 
2237 	vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2238 	if (vf_netdev)
2239 		__netvsc_vf_setup(ndev, vf_netdev);
2240 
2241 	rtnl_unlock();
2242 }
2243 
2244 /* Find netvsc by VF serial number.
2245  * The PCI hyperv controller records the serial number as the slot kobj name.
2246  */
2247 static struct net_device *get_netvsc_byslot(const struct net_device *vf_netdev)
2248 {
2249 	struct device *parent = vf_netdev->dev.parent;
2250 	struct net_device_context *ndev_ctx;
2251 	struct pci_dev *pdev;
2252 	u32 serial;
2253 
2254 	if (!parent || !dev_is_pci(parent))
2255 		return NULL; /* not a PCI device */
2256 
2257 	pdev = to_pci_dev(parent);
2258 	if (!pdev->slot) {
2259 		netdev_notice(vf_netdev, "no PCI slot information\n");
2260 		return NULL;
2261 	}
2262 
2263 	if (kstrtou32(pci_slot_name(pdev->slot), 10, &serial)) {
2264 		netdev_notice(vf_netdev, "Invalid vf serial:%s\n",
2265 			      pci_slot_name(pdev->slot));
2266 		return NULL;
2267 	}
2268 
2269 	list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
2270 		if (!ndev_ctx->vf_alloc)
2271 			continue;
2272 
2273 		if (ndev_ctx->vf_serial == serial)
2274 			return hv_get_drvdata(ndev_ctx->device_ctx);
2275 	}
2276 
2277 	netdev_notice(vf_netdev,
2278 		      "no netdev found for vf serial:%u\n", serial);
2279 	return NULL;
2280 }
2281 
2282 static int netvsc_register_vf(struct net_device *vf_netdev)
2283 {
2284 	struct net_device_context *net_device_ctx;
2285 	struct netvsc_device *netvsc_dev;
2286 	struct bpf_prog *prog;
2287 	struct net_device *ndev;
2288 	int ret;
2289 
2290 	if (vf_netdev->addr_len != ETH_ALEN)
2291 		return NOTIFY_DONE;
2292 
2293 	ndev = get_netvsc_byslot(vf_netdev);
2294 	if (!ndev)
2295 		return NOTIFY_DONE;
2296 
2297 	net_device_ctx = netdev_priv(ndev);
2298 	netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2299 	if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
2300 		return NOTIFY_DONE;
2301 
2302 	/* if synthetic interface is a different namespace,
2303 	 * then move the VF to that namespace; join will be
2304 	 * done again in that context.
2305 	 */
2306 	if (!net_eq(dev_net(ndev), dev_net(vf_netdev))) {
2307 		ret = dev_change_net_namespace(vf_netdev,
2308 					       dev_net(ndev), "eth%d");
2309 		if (ret)
2310 			netdev_err(vf_netdev,
2311 				   "could not move to same namespace as %s: %d\n",
2312 				   ndev->name, ret);
2313 		else
2314 			netdev_info(vf_netdev,
2315 				    "VF moved to namespace with: %s\n",
2316 				    ndev->name);
2317 		return NOTIFY_DONE;
2318 	}
2319 
2320 	netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
2321 
2322 	if (netvsc_vf_join(vf_netdev, ndev) != 0)
2323 		return NOTIFY_DONE;
2324 
2325 	dev_hold(vf_netdev);
2326 	rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
2327 
2328 	vf_netdev->wanted_features = ndev->features;
2329 	netdev_update_features(vf_netdev);
2330 
2331 	prog = netvsc_xdp_get(netvsc_dev);
2332 	netvsc_vf_setxdp(vf_netdev, prog);
2333 
2334 	return NOTIFY_OK;
2335 }
2336 
2337 /* VF up/down change detected, schedule to change data path */
2338 static int netvsc_vf_changed(struct net_device *vf_netdev)
2339 {
2340 	struct net_device_context *net_device_ctx;
2341 	struct netvsc_device *netvsc_dev;
2342 	struct net_device *ndev;
2343 	bool vf_is_up = netif_running(vf_netdev);
2344 
2345 	ndev = get_netvsc_byref(vf_netdev);
2346 	if (!ndev)
2347 		return NOTIFY_DONE;
2348 
2349 	net_device_ctx = netdev_priv(ndev);
2350 	netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2351 	if (!netvsc_dev)
2352 		return NOTIFY_DONE;
2353 
2354 	netvsc_switch_datapath(ndev, vf_is_up);
2355 	netdev_info(ndev, "Data path switched %s VF: %s\n",
2356 		    vf_is_up ? "to" : "from", vf_netdev->name);
2357 
2358 	return NOTIFY_OK;
2359 }
2360 
2361 static int netvsc_unregister_vf(struct net_device *vf_netdev)
2362 {
2363 	struct net_device *ndev;
2364 	struct net_device_context *net_device_ctx;
2365 
2366 	ndev = get_netvsc_byref(vf_netdev);
2367 	if (!ndev)
2368 		return NOTIFY_DONE;
2369 
2370 	net_device_ctx = netdev_priv(ndev);
2371 	cancel_delayed_work_sync(&net_device_ctx->vf_takeover);
2372 
2373 	netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
2374 
2375 	netvsc_vf_setxdp(vf_netdev, NULL);
2376 
2377 	netdev_rx_handler_unregister(vf_netdev);
2378 	netdev_upper_dev_unlink(vf_netdev, ndev);
2379 	RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
2380 	dev_put(vf_netdev);
2381 
2382 	return NOTIFY_OK;
2383 }
2384 
2385 static int netvsc_probe(struct hv_device *dev,
2386 			const struct hv_vmbus_device_id *dev_id)
2387 {
2388 	struct net_device *net = NULL;
2389 	struct net_device_context *net_device_ctx;
2390 	struct netvsc_device_info *device_info = NULL;
2391 	struct netvsc_device *nvdev;
2392 	int ret = -ENOMEM;
2393 
2394 	net = alloc_etherdev_mq(sizeof(struct net_device_context),
2395 				VRSS_CHANNEL_MAX);
2396 	if (!net)
2397 		goto no_net;
2398 
2399 	netif_carrier_off(net);
2400 
2401 	netvsc_init_settings(net);
2402 
2403 	net_device_ctx = netdev_priv(net);
2404 	net_device_ctx->device_ctx = dev;
2405 	net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
2406 	if (netif_msg_probe(net_device_ctx))
2407 		netdev_dbg(net, "netvsc msg_enable: %d\n",
2408 			   net_device_ctx->msg_enable);
2409 
2410 	hv_set_drvdata(dev, net);
2411 
2412 	INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
2413 
2414 	spin_lock_init(&net_device_ctx->lock);
2415 	INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
2416 	INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup);
2417 
2418 	net_device_ctx->vf_stats
2419 		= netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats);
2420 	if (!net_device_ctx->vf_stats)
2421 		goto no_stats;
2422 
2423 	net->netdev_ops = &device_ops;
2424 	net->ethtool_ops = &ethtool_ops;
2425 	SET_NETDEV_DEV(net, &dev->device);
2426 
2427 	/* We always need headroom for rndis header */
2428 	net->needed_headroom = RNDIS_AND_PPI_SIZE;
2429 
2430 	/* Initialize the number of queues to be 1, we may change it if more
2431 	 * channels are offered later.
2432 	 */
2433 	netif_set_real_num_tx_queues(net, 1);
2434 	netif_set_real_num_rx_queues(net, 1);
2435 
2436 	/* Notify the netvsc driver of the new device */
2437 	device_info = netvsc_devinfo_get(NULL);
2438 
2439 	if (!device_info) {
2440 		ret = -ENOMEM;
2441 		goto devinfo_failed;
2442 	}
2443 
2444 	nvdev = rndis_filter_device_add(dev, device_info);
2445 	if (IS_ERR(nvdev)) {
2446 		ret = PTR_ERR(nvdev);
2447 		netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
2448 		goto rndis_failed;
2449 	}
2450 
2451 	memcpy(net->dev_addr, device_info->mac_adr, ETH_ALEN);
2452 
2453 	/* We must get rtnl lock before scheduling nvdev->subchan_work,
2454 	 * otherwise netvsc_subchan_work() can get rtnl lock first and wait
2455 	 * all subchannels to show up, but that may not happen because
2456 	 * netvsc_probe() can't get rtnl lock and as a result vmbus_onoffer()
2457 	 * -> ... -> device_add() -> ... -> __device_attach() can't get
2458 	 * the device lock, so all the subchannels can't be processed --
2459 	 * finally netvsc_subchan_work() hangs forever.
2460 	 */
2461 	rtnl_lock();
2462 
2463 	if (nvdev->num_chn > 1)
2464 		schedule_work(&nvdev->subchan_work);
2465 
2466 	/* hw_features computed in rndis_netdev_set_hwcaps() */
2467 	net->features = net->hw_features |
2468 		NETIF_F_HIGHDMA | NETIF_F_HW_VLAN_CTAG_TX |
2469 		NETIF_F_HW_VLAN_CTAG_RX;
2470 	net->vlan_features = net->features;
2471 
2472 	/* MTU range: 68 - 1500 or 65521 */
2473 	net->min_mtu = NETVSC_MTU_MIN;
2474 	if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
2475 		net->max_mtu = NETVSC_MTU - ETH_HLEN;
2476 	else
2477 		net->max_mtu = ETH_DATA_LEN;
2478 
2479 	ret = register_netdevice(net);
2480 	if (ret != 0) {
2481 		pr_err("Unable to register netdev.\n");
2482 		goto register_failed;
2483 	}
2484 
2485 	list_add(&net_device_ctx->list, &netvsc_dev_list);
2486 	rtnl_unlock();
2487 
2488 	netvsc_devinfo_put(device_info);
2489 	return 0;
2490 
2491 register_failed:
2492 	rtnl_unlock();
2493 	rndis_filter_device_remove(dev, nvdev);
2494 rndis_failed:
2495 	netvsc_devinfo_put(device_info);
2496 devinfo_failed:
2497 	free_percpu(net_device_ctx->vf_stats);
2498 no_stats:
2499 	hv_set_drvdata(dev, NULL);
2500 	free_netdev(net);
2501 no_net:
2502 	return ret;
2503 }
2504 
2505 static int netvsc_remove(struct hv_device *dev)
2506 {
2507 	struct net_device_context *ndev_ctx;
2508 	struct net_device *vf_netdev, *net;
2509 	struct netvsc_device *nvdev;
2510 
2511 	net = hv_get_drvdata(dev);
2512 	if (net == NULL) {
2513 		dev_err(&dev->device, "No net device to remove\n");
2514 		return 0;
2515 	}
2516 
2517 	ndev_ctx = netdev_priv(net);
2518 
2519 	cancel_delayed_work_sync(&ndev_ctx->dwork);
2520 
2521 	rtnl_lock();
2522 	nvdev = rtnl_dereference(ndev_ctx->nvdev);
2523 	if (nvdev) {
2524 		cancel_work_sync(&nvdev->subchan_work);
2525 		netvsc_xdp_set(net, NULL, NULL, nvdev);
2526 	}
2527 
2528 	/*
2529 	 * Call to the vsc driver to let it know that the device is being
2530 	 * removed. Also blocks mtu and channel changes.
2531 	 */
2532 	vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2533 	if (vf_netdev)
2534 		netvsc_unregister_vf(vf_netdev);
2535 
2536 	if (nvdev)
2537 		rndis_filter_device_remove(dev, nvdev);
2538 
2539 	unregister_netdevice(net);
2540 	list_del(&ndev_ctx->list);
2541 
2542 	rtnl_unlock();
2543 
2544 	hv_set_drvdata(dev, NULL);
2545 
2546 	free_percpu(ndev_ctx->vf_stats);
2547 	free_netdev(net);
2548 	return 0;
2549 }
2550 
2551 static int netvsc_suspend(struct hv_device *dev)
2552 {
2553 	struct net_device_context *ndev_ctx;
2554 	struct net_device *vf_netdev, *net;
2555 	struct netvsc_device *nvdev;
2556 	int ret;
2557 
2558 	net = hv_get_drvdata(dev);
2559 
2560 	ndev_ctx = netdev_priv(net);
2561 	cancel_delayed_work_sync(&ndev_ctx->dwork);
2562 
2563 	rtnl_lock();
2564 
2565 	nvdev = rtnl_dereference(ndev_ctx->nvdev);
2566 	if (nvdev == NULL) {
2567 		ret = -ENODEV;
2568 		goto out;
2569 	}
2570 
2571 	vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2572 	if (vf_netdev)
2573 		netvsc_unregister_vf(vf_netdev);
2574 
2575 	/* Save the current config info */
2576 	ndev_ctx->saved_netvsc_dev_info = netvsc_devinfo_get(nvdev);
2577 
2578 	ret = netvsc_detach(net, nvdev);
2579 out:
2580 	rtnl_unlock();
2581 
2582 	return ret;
2583 }
2584 
2585 static int netvsc_resume(struct hv_device *dev)
2586 {
2587 	struct net_device *net = hv_get_drvdata(dev);
2588 	struct net_device_context *net_device_ctx;
2589 	struct netvsc_device_info *device_info;
2590 	int ret;
2591 
2592 	rtnl_lock();
2593 
2594 	net_device_ctx = netdev_priv(net);
2595 	device_info = net_device_ctx->saved_netvsc_dev_info;
2596 
2597 	ret = netvsc_attach(net, device_info);
2598 
2599 	netvsc_devinfo_put(device_info);
2600 	net_device_ctx->saved_netvsc_dev_info = NULL;
2601 
2602 	rtnl_unlock();
2603 
2604 	return ret;
2605 }
2606 static const struct hv_vmbus_device_id id_table[] = {
2607 	/* Network guid */
2608 	{ HV_NIC_GUID, },
2609 	{ },
2610 };
2611 
2612 MODULE_DEVICE_TABLE(vmbus, id_table);
2613 
2614 /* The one and only one */
2615 static struct  hv_driver netvsc_drv = {
2616 	.name = KBUILD_MODNAME,
2617 	.id_table = id_table,
2618 	.probe = netvsc_probe,
2619 	.remove = netvsc_remove,
2620 	.suspend = netvsc_suspend,
2621 	.resume = netvsc_resume,
2622 	.driver = {
2623 		.probe_type = PROBE_FORCE_SYNCHRONOUS,
2624 	},
2625 };
2626 
2627 /*
2628  * On Hyper-V, every VF interface is matched with a corresponding
2629  * synthetic interface. The synthetic interface is presented first
2630  * to the guest. When the corresponding VF instance is registered,
2631  * we will take care of switching the data path.
2632  */
2633 static int netvsc_netdev_event(struct notifier_block *this,
2634 			       unsigned long event, void *ptr)
2635 {
2636 	struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2637 
2638 	/* Skip our own events */
2639 	if (event_dev->netdev_ops == &device_ops)
2640 		return NOTIFY_DONE;
2641 
2642 	/* Avoid non-Ethernet type devices */
2643 	if (event_dev->type != ARPHRD_ETHER)
2644 		return NOTIFY_DONE;
2645 
2646 	/* Avoid Vlan dev with same MAC registering as VF */
2647 	if (is_vlan_dev(event_dev))
2648 		return NOTIFY_DONE;
2649 
2650 	/* Avoid Bonding master dev with same MAC registering as VF */
2651 	if ((event_dev->priv_flags & IFF_BONDING) &&
2652 	    (event_dev->flags & IFF_MASTER))
2653 		return NOTIFY_DONE;
2654 
2655 	switch (event) {
2656 	case NETDEV_REGISTER:
2657 		return netvsc_register_vf(event_dev);
2658 	case NETDEV_UNREGISTER:
2659 		return netvsc_unregister_vf(event_dev);
2660 	case NETDEV_UP:
2661 	case NETDEV_DOWN:
2662 		return netvsc_vf_changed(event_dev);
2663 	default:
2664 		return NOTIFY_DONE;
2665 	}
2666 }
2667 
2668 static struct notifier_block netvsc_netdev_notifier = {
2669 	.notifier_call = netvsc_netdev_event,
2670 };
2671 
2672 static void __exit netvsc_drv_exit(void)
2673 {
2674 	unregister_netdevice_notifier(&netvsc_netdev_notifier);
2675 	vmbus_driver_unregister(&netvsc_drv);
2676 }
2677 
2678 static int __init netvsc_drv_init(void)
2679 {
2680 	int ret;
2681 
2682 	if (ring_size < RING_SIZE_MIN) {
2683 		ring_size = RING_SIZE_MIN;
2684 		pr_info("Increased ring_size to %u (min allowed)\n",
2685 			ring_size);
2686 	}
2687 	netvsc_ring_bytes = ring_size * PAGE_SIZE;
2688 
2689 	ret = vmbus_driver_register(&netvsc_drv);
2690 	if (ret)
2691 		return ret;
2692 
2693 	register_netdevice_notifier(&netvsc_netdev_notifier);
2694 	return 0;
2695 }
2696 
2697 MODULE_LICENSE("GPL");
2698 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
2699 
2700 module_init(netvsc_drv_init);
2701 module_exit(netvsc_drv_exit);
2702