xref: /openbmc/linux/drivers/net/hyperv/netvsc_drv.c (revision 83146efc)
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/ethtool.h>
14 #include <linux/module.h>
15 #include <linux/highmem.h>
16 #include <linux/device.h>
17 #include <linux/io.h>
18 #include <linux/delay.h>
19 #include <linux/netdevice.h>
20 #include <linux/inetdevice.h>
21 #include <linux/etherdevice.h>
22 #include <linux/pci.h>
23 #include <linux/skbuff.h>
24 #include <linux/if_vlan.h>
25 #include <linux/in.h>
26 #include <linux/slab.h>
27 #include <linux/rtnetlink.h>
28 #include <linux/netpoll.h>
29 #include <linux/bpf.h>
30 
31 #include <net/arp.h>
32 #include <net/route.h>
33 #include <net/sock.h>
34 #include <net/pkt_sched.h>
35 #include <net/checksum.h>
36 #include <net/ip6_checksum.h>
37 
38 #include "hyperv_net.h"
39 
40 #define RING_SIZE_MIN	64
41 
42 #define LINKCHANGE_INT (2 * HZ)
43 #define VF_TAKEOVER_INT (HZ / 10)
44 
45 static unsigned int ring_size __ro_after_init = 128;
46 module_param(ring_size, uint, 0444);
47 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of 4K pages)");
48 unsigned int netvsc_ring_bytes __ro_after_init;
49 
50 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
51 				NETIF_MSG_LINK | NETIF_MSG_IFUP |
52 				NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
53 				NETIF_MSG_TX_ERR;
54 
55 static int debug = -1;
56 module_param(debug, int, 0444);
57 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
58 
59 static LIST_HEAD(netvsc_dev_list);
60 
61 static void netvsc_change_rx_flags(struct net_device *net, int change)
62 {
63 	struct net_device_context *ndev_ctx = netdev_priv(net);
64 	struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
65 	int inc;
66 
67 	if (!vf_netdev)
68 		return;
69 
70 	if (change & IFF_PROMISC) {
71 		inc = (net->flags & IFF_PROMISC) ? 1 : -1;
72 		dev_set_promiscuity(vf_netdev, inc);
73 	}
74 
75 	if (change & IFF_ALLMULTI) {
76 		inc = (net->flags & IFF_ALLMULTI) ? 1 : -1;
77 		dev_set_allmulti(vf_netdev, inc);
78 	}
79 }
80 
81 static void netvsc_set_rx_mode(struct net_device *net)
82 {
83 	struct net_device_context *ndev_ctx = netdev_priv(net);
84 	struct net_device *vf_netdev;
85 	struct netvsc_device *nvdev;
86 
87 	rcu_read_lock();
88 	vf_netdev = rcu_dereference(ndev_ctx->vf_netdev);
89 	if (vf_netdev) {
90 		dev_uc_sync(vf_netdev, net);
91 		dev_mc_sync(vf_netdev, net);
92 	}
93 
94 	nvdev = rcu_dereference(ndev_ctx->nvdev);
95 	if (nvdev)
96 		rndis_filter_update(nvdev);
97 	rcu_read_unlock();
98 }
99 
100 static void netvsc_tx_enable(struct netvsc_device *nvscdev,
101 			     struct net_device *ndev)
102 {
103 	nvscdev->tx_disable = false;
104 	virt_wmb(); /* ensure queue wake up mechanism is on */
105 
106 	netif_tx_wake_all_queues(ndev);
107 }
108 
109 static int netvsc_open(struct net_device *net)
110 {
111 	struct net_device_context *ndev_ctx = netdev_priv(net);
112 	struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
113 	struct netvsc_device *nvdev = rtnl_dereference(ndev_ctx->nvdev);
114 	struct rndis_device *rdev;
115 	int ret = 0;
116 
117 	netif_carrier_off(net);
118 
119 	/* Open up the device */
120 	ret = rndis_filter_open(nvdev);
121 	if (ret != 0) {
122 		netdev_err(net, "unable to open device (ret %d).\n", ret);
123 		return ret;
124 	}
125 
126 	rdev = nvdev->extension;
127 	if (!rdev->link_state) {
128 		netif_carrier_on(net);
129 		netvsc_tx_enable(nvdev, net);
130 	}
131 
132 	if (vf_netdev) {
133 		/* Setting synthetic device up transparently sets
134 		 * slave as up. If open fails, then slave will be
135 		 * still be offline (and not used).
136 		 */
137 		ret = dev_open(vf_netdev, NULL);
138 		if (ret)
139 			netdev_warn(net,
140 				    "unable to open slave: %s: %d\n",
141 				    vf_netdev->name, ret);
142 	}
143 	return 0;
144 }
145 
146 static int netvsc_wait_until_empty(struct netvsc_device *nvdev)
147 {
148 	unsigned int retry = 0;
149 	int i;
150 
151 	/* Ensure pending bytes in ring are read */
152 	for (;;) {
153 		u32 aread = 0;
154 
155 		for (i = 0; i < nvdev->num_chn; i++) {
156 			struct vmbus_channel *chn
157 				= nvdev->chan_table[i].channel;
158 
159 			if (!chn)
160 				continue;
161 
162 			/* make sure receive not running now */
163 			napi_synchronize(&nvdev->chan_table[i].napi);
164 
165 			aread = hv_get_bytes_to_read(&chn->inbound);
166 			if (aread)
167 				break;
168 
169 			aread = hv_get_bytes_to_read(&chn->outbound);
170 			if (aread)
171 				break;
172 		}
173 
174 		if (aread == 0)
175 			return 0;
176 
177 		if (++retry > RETRY_MAX)
178 			return -ETIMEDOUT;
179 
180 		usleep_range(RETRY_US_LO, RETRY_US_HI);
181 	}
182 }
183 
184 static void netvsc_tx_disable(struct netvsc_device *nvscdev,
185 			      struct net_device *ndev)
186 {
187 	if (nvscdev) {
188 		nvscdev->tx_disable = true;
189 		virt_wmb(); /* ensure txq will not wake up after stop */
190 	}
191 
192 	netif_tx_disable(ndev);
193 }
194 
195 static int netvsc_close(struct net_device *net)
196 {
197 	struct net_device_context *net_device_ctx = netdev_priv(net);
198 	struct net_device *vf_netdev
199 		= rtnl_dereference(net_device_ctx->vf_netdev);
200 	struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
201 	int ret;
202 
203 	netvsc_tx_disable(nvdev, net);
204 
205 	/* No need to close rndis filter if it is removed already */
206 	if (!nvdev)
207 		return 0;
208 
209 	ret = rndis_filter_close(nvdev);
210 	if (ret != 0) {
211 		netdev_err(net, "unable to close device (ret %d).\n", ret);
212 		return ret;
213 	}
214 
215 	ret = netvsc_wait_until_empty(nvdev);
216 	if (ret)
217 		netdev_err(net, "Ring buffer not empty after closing rndis\n");
218 
219 	if (vf_netdev)
220 		dev_close(vf_netdev);
221 
222 	return ret;
223 }
224 
225 static inline void *init_ppi_data(struct rndis_message *msg,
226 				  u32 ppi_size, u32 pkt_type)
227 {
228 	struct rndis_packet *rndis_pkt = &msg->msg.pkt;
229 	struct rndis_per_packet_info *ppi;
230 
231 	rndis_pkt->data_offset += ppi_size;
232 	ppi = (void *)rndis_pkt + rndis_pkt->per_pkt_info_offset
233 		+ rndis_pkt->per_pkt_info_len;
234 
235 	ppi->size = ppi_size;
236 	ppi->type = pkt_type;
237 	ppi->internal = 0;
238 	ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
239 
240 	rndis_pkt->per_pkt_info_len += ppi_size;
241 
242 	return ppi + 1;
243 }
244 
245 static inline int netvsc_get_tx_queue(struct net_device *ndev,
246 				      struct sk_buff *skb, int old_idx)
247 {
248 	const struct net_device_context *ndc = netdev_priv(ndev);
249 	struct sock *sk = skb->sk;
250 	int q_idx;
251 
252 	q_idx = ndc->tx_table[netvsc_get_hash(skb, ndc) &
253 			      (VRSS_SEND_TAB_SIZE - 1)];
254 
255 	/* If queue index changed record the new value */
256 	if (q_idx != old_idx &&
257 	    sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
258 		sk_tx_queue_set(sk, q_idx);
259 
260 	return q_idx;
261 }
262 
263 /*
264  * Select queue for transmit.
265  *
266  * If a valid queue has already been assigned, then use that.
267  * Otherwise compute tx queue based on hash and the send table.
268  *
269  * This is basically similar to default (netdev_pick_tx) with the added step
270  * of using the host send_table when no other queue has been assigned.
271  *
272  * TODO support XPS - but get_xps_queue not exported
273  */
274 static u16 netvsc_pick_tx(struct net_device *ndev, struct sk_buff *skb)
275 {
276 	int q_idx = sk_tx_queue_get(skb->sk);
277 
278 	if (q_idx < 0 || skb->ooo_okay || q_idx >= ndev->real_num_tx_queues) {
279 		/* If forwarding a packet, we use the recorded queue when
280 		 * available for better cache locality.
281 		 */
282 		if (skb_rx_queue_recorded(skb))
283 			q_idx = skb_get_rx_queue(skb);
284 		else
285 			q_idx = netvsc_get_tx_queue(ndev, skb, q_idx);
286 	}
287 
288 	return q_idx;
289 }
290 
291 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
292 			       struct net_device *sb_dev)
293 {
294 	struct net_device_context *ndc = netdev_priv(ndev);
295 	struct net_device *vf_netdev;
296 	u16 txq;
297 
298 	rcu_read_lock();
299 	vf_netdev = rcu_dereference(ndc->vf_netdev);
300 	if (vf_netdev) {
301 		const struct net_device_ops *vf_ops = vf_netdev->netdev_ops;
302 
303 		if (vf_ops->ndo_select_queue)
304 			txq = vf_ops->ndo_select_queue(vf_netdev, skb, sb_dev);
305 		else
306 			txq = netdev_pick_tx(vf_netdev, skb, NULL);
307 
308 		/* Record the queue selected by VF so that it can be
309 		 * used for common case where VF has more queues than
310 		 * the synthetic device.
311 		 */
312 		qdisc_skb_cb(skb)->slave_dev_queue_mapping = txq;
313 	} else {
314 		txq = netvsc_pick_tx(ndev, skb);
315 	}
316 	rcu_read_unlock();
317 
318 	while (txq >= ndev->real_num_tx_queues)
319 		txq -= ndev->real_num_tx_queues;
320 
321 	return txq;
322 }
323 
324 static u32 fill_pg_buf(unsigned long hvpfn, u32 offset, u32 len,
325 		       struct hv_page_buffer *pb)
326 {
327 	int j = 0;
328 
329 	hvpfn += offset >> HV_HYP_PAGE_SHIFT;
330 	offset = offset & ~HV_HYP_PAGE_MASK;
331 
332 	while (len > 0) {
333 		unsigned long bytes;
334 
335 		bytes = HV_HYP_PAGE_SIZE - offset;
336 		if (bytes > len)
337 			bytes = len;
338 		pb[j].pfn = hvpfn;
339 		pb[j].offset = offset;
340 		pb[j].len = bytes;
341 
342 		offset += bytes;
343 		len -= bytes;
344 
345 		if (offset == HV_HYP_PAGE_SIZE && len) {
346 			hvpfn++;
347 			offset = 0;
348 			j++;
349 		}
350 	}
351 
352 	return j + 1;
353 }
354 
355 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
356 			   struct hv_netvsc_packet *packet,
357 			   struct hv_page_buffer *pb)
358 {
359 	u32 slots_used = 0;
360 	char *data = skb->data;
361 	int frags = skb_shinfo(skb)->nr_frags;
362 	int i;
363 
364 	/* The packet is laid out thus:
365 	 * 1. hdr: RNDIS header and PPI
366 	 * 2. skb linear data
367 	 * 3. skb fragment data
368 	 */
369 	slots_used += fill_pg_buf(virt_to_hvpfn(hdr),
370 				  offset_in_hvpage(hdr),
371 				  len,
372 				  &pb[slots_used]);
373 
374 	packet->rmsg_size = len;
375 	packet->rmsg_pgcnt = slots_used;
376 
377 	slots_used += fill_pg_buf(virt_to_hvpfn(data),
378 				  offset_in_hvpage(data),
379 				  skb_headlen(skb),
380 				  &pb[slots_used]);
381 
382 	for (i = 0; i < frags; i++) {
383 		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
384 
385 		slots_used += fill_pg_buf(page_to_hvpfn(skb_frag_page(frag)),
386 					  skb_frag_off(frag),
387 					  skb_frag_size(frag),
388 					  &pb[slots_used]);
389 	}
390 	return slots_used;
391 }
392 
393 static int count_skb_frag_slots(struct sk_buff *skb)
394 {
395 	int i, frags = skb_shinfo(skb)->nr_frags;
396 	int pages = 0;
397 
398 	for (i = 0; i < frags; i++) {
399 		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
400 		unsigned long size = skb_frag_size(frag);
401 		unsigned long offset = skb_frag_off(frag);
402 
403 		/* Skip unused frames from start of page */
404 		offset &= ~HV_HYP_PAGE_MASK;
405 		pages += HVPFN_UP(offset + size);
406 	}
407 	return pages;
408 }
409 
410 static int netvsc_get_slots(struct sk_buff *skb)
411 {
412 	char *data = skb->data;
413 	unsigned int offset = offset_in_hvpage(data);
414 	unsigned int len = skb_headlen(skb);
415 	int slots;
416 	int frag_slots;
417 
418 	slots = DIV_ROUND_UP(offset + len, HV_HYP_PAGE_SIZE);
419 	frag_slots = count_skb_frag_slots(skb);
420 	return slots + frag_slots;
421 }
422 
423 static u32 net_checksum_info(struct sk_buff *skb)
424 {
425 	if (skb->protocol == htons(ETH_P_IP)) {
426 		struct iphdr *ip = ip_hdr(skb);
427 
428 		if (ip->protocol == IPPROTO_TCP)
429 			return TRANSPORT_INFO_IPV4_TCP;
430 		else if (ip->protocol == IPPROTO_UDP)
431 			return TRANSPORT_INFO_IPV4_UDP;
432 	} else {
433 		struct ipv6hdr *ip6 = ipv6_hdr(skb);
434 
435 		if (ip6->nexthdr == IPPROTO_TCP)
436 			return TRANSPORT_INFO_IPV6_TCP;
437 		else if (ip6->nexthdr == IPPROTO_UDP)
438 			return TRANSPORT_INFO_IPV6_UDP;
439 	}
440 
441 	return TRANSPORT_INFO_NOT_IP;
442 }
443 
444 /* Send skb on the slave VF device. */
445 static int netvsc_vf_xmit(struct net_device *net, struct net_device *vf_netdev,
446 			  struct sk_buff *skb)
447 {
448 	struct net_device_context *ndev_ctx = netdev_priv(net);
449 	unsigned int len = skb->len;
450 	int rc;
451 
452 	skb->dev = vf_netdev;
453 	skb_record_rx_queue(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
454 
455 	rc = dev_queue_xmit(skb);
456 	if (likely(rc == NET_XMIT_SUCCESS || rc == NET_XMIT_CN)) {
457 		struct netvsc_vf_pcpu_stats *pcpu_stats
458 			= this_cpu_ptr(ndev_ctx->vf_stats);
459 
460 		u64_stats_update_begin(&pcpu_stats->syncp);
461 		pcpu_stats->tx_packets++;
462 		pcpu_stats->tx_bytes += len;
463 		u64_stats_update_end(&pcpu_stats->syncp);
464 	} else {
465 		this_cpu_inc(ndev_ctx->vf_stats->tx_dropped);
466 	}
467 
468 	return rc;
469 }
470 
471 static int netvsc_xmit(struct sk_buff *skb, struct net_device *net, bool xdp_tx)
472 {
473 	struct net_device_context *net_device_ctx = netdev_priv(net);
474 	struct hv_netvsc_packet *packet = NULL;
475 	int ret;
476 	unsigned int num_data_pgs;
477 	struct rndis_message *rndis_msg;
478 	struct net_device *vf_netdev;
479 	u32 rndis_msg_size;
480 	u32 hash;
481 	struct hv_page_buffer pb[MAX_PAGE_BUFFER_COUNT];
482 
483 	/* If VF is present and up then redirect packets to it.
484 	 * Skip the VF if it is marked down or has no carrier.
485 	 * If netpoll is in uses, then VF can not be used either.
486 	 */
487 	vf_netdev = rcu_dereference_bh(net_device_ctx->vf_netdev);
488 	if (vf_netdev && netif_running(vf_netdev) &&
489 	    netif_carrier_ok(vf_netdev) && !netpoll_tx_running(net) &&
490 	    net_device_ctx->data_path_is_vf)
491 		return netvsc_vf_xmit(net, vf_netdev, skb);
492 
493 	/* We will atmost need two pages to describe the rndis
494 	 * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
495 	 * of pages in a single packet. If skb is scattered around
496 	 * more pages we try linearizing it.
497 	 */
498 
499 	num_data_pgs = netvsc_get_slots(skb) + 2;
500 
501 	if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
502 		++net_device_ctx->eth_stats.tx_scattered;
503 
504 		if (skb_linearize(skb))
505 			goto no_memory;
506 
507 		num_data_pgs = netvsc_get_slots(skb) + 2;
508 		if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
509 			++net_device_ctx->eth_stats.tx_too_big;
510 			goto drop;
511 		}
512 	}
513 
514 	/*
515 	 * Place the rndis header in the skb head room and
516 	 * the skb->cb will be used for hv_netvsc_packet
517 	 * structure.
518 	 */
519 	ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
520 	if (ret)
521 		goto no_memory;
522 
523 	/* Use the skb control buffer for building up the packet */
524 	BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
525 			sizeof_field(struct sk_buff, cb));
526 	packet = (struct hv_netvsc_packet *)skb->cb;
527 
528 	packet->q_idx = skb_get_queue_mapping(skb);
529 
530 	packet->total_data_buflen = skb->len;
531 	packet->total_bytes = skb->len;
532 	packet->total_packets = 1;
533 
534 	rndis_msg = (struct rndis_message *)skb->head;
535 
536 	/* Add the rndis header */
537 	rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
538 	rndis_msg->msg_len = packet->total_data_buflen;
539 
540 	rndis_msg->msg.pkt = (struct rndis_packet) {
541 		.data_offset = sizeof(struct rndis_packet),
542 		.data_len = packet->total_data_buflen,
543 		.per_pkt_info_offset = sizeof(struct rndis_packet),
544 	};
545 
546 	rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
547 
548 	hash = skb_get_hash_raw(skb);
549 	if (hash != 0 && net->real_num_tx_queues > 1) {
550 		u32 *hash_info;
551 
552 		rndis_msg_size += NDIS_HASH_PPI_SIZE;
553 		hash_info = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
554 					  NBL_HASH_VALUE);
555 		*hash_info = hash;
556 	}
557 
558 	/* When using AF_PACKET we need to drop VLAN header from
559 	 * the frame and update the SKB to allow the HOST OS
560 	 * to transmit the 802.1Q packet
561 	 */
562 	if (skb->protocol == htons(ETH_P_8021Q)) {
563 		u16 vlan_tci;
564 
565 		skb_reset_mac_header(skb);
566 		if (eth_type_vlan(eth_hdr(skb)->h_proto)) {
567 			if (unlikely(__skb_vlan_pop(skb, &vlan_tci) != 0)) {
568 				++net_device_ctx->eth_stats.vlan_error;
569 				goto drop;
570 			}
571 
572 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
573 			/* Update the NDIS header pkt lengths */
574 			packet->total_data_buflen -= VLAN_HLEN;
575 			packet->total_bytes -= VLAN_HLEN;
576 			rndis_msg->msg_len = packet->total_data_buflen;
577 			rndis_msg->msg.pkt.data_len = packet->total_data_buflen;
578 		}
579 	}
580 
581 	if (skb_vlan_tag_present(skb)) {
582 		struct ndis_pkt_8021q_info *vlan;
583 
584 		rndis_msg_size += NDIS_VLAN_PPI_SIZE;
585 		vlan = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
586 				     IEEE_8021Q_INFO);
587 
588 		vlan->value = 0;
589 		vlan->vlanid = skb_vlan_tag_get_id(skb);
590 		vlan->cfi = skb_vlan_tag_get_cfi(skb);
591 		vlan->pri = skb_vlan_tag_get_prio(skb);
592 	}
593 
594 	if (skb_is_gso(skb)) {
595 		struct ndis_tcp_lso_info *lso_info;
596 
597 		rndis_msg_size += NDIS_LSO_PPI_SIZE;
598 		lso_info = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
599 					 TCP_LARGESEND_PKTINFO);
600 
601 		lso_info->value = 0;
602 		lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
603 		if (skb->protocol == htons(ETH_P_IP)) {
604 			lso_info->lso_v2_transmit.ip_version =
605 				NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
606 			ip_hdr(skb)->tot_len = 0;
607 			ip_hdr(skb)->check = 0;
608 			tcp_hdr(skb)->check =
609 				~csum_tcpudp_magic(ip_hdr(skb)->saddr,
610 						   ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
611 		} else {
612 			lso_info->lso_v2_transmit.ip_version =
613 				NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
614 			tcp_v6_gso_csum_prep(skb);
615 		}
616 		lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
617 		lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
618 	} else if (skb->ip_summed == CHECKSUM_PARTIAL) {
619 		if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
620 			struct ndis_tcp_ip_checksum_info *csum_info;
621 
622 			rndis_msg_size += NDIS_CSUM_PPI_SIZE;
623 			csum_info = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
624 						  TCPIP_CHKSUM_PKTINFO);
625 
626 			csum_info->value = 0;
627 			csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
628 
629 			if (skb->protocol == htons(ETH_P_IP)) {
630 				csum_info->transmit.is_ipv4 = 1;
631 
632 				if (ip_hdr(skb)->protocol == IPPROTO_TCP)
633 					csum_info->transmit.tcp_checksum = 1;
634 				else
635 					csum_info->transmit.udp_checksum = 1;
636 			} else {
637 				csum_info->transmit.is_ipv6 = 1;
638 
639 				if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
640 					csum_info->transmit.tcp_checksum = 1;
641 				else
642 					csum_info->transmit.udp_checksum = 1;
643 			}
644 		} else {
645 			/* Can't do offload of this type of checksum */
646 			if (skb_checksum_help(skb))
647 				goto drop;
648 		}
649 	}
650 
651 	/* Start filling in the page buffers with the rndis hdr */
652 	rndis_msg->msg_len += rndis_msg_size;
653 	packet->total_data_buflen = rndis_msg->msg_len;
654 	packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
655 					       skb, packet, pb);
656 
657 	/* timestamp packet in software */
658 	skb_tx_timestamp(skb);
659 
660 	ret = netvsc_send(net, packet, rndis_msg, pb, skb, xdp_tx);
661 	if (likely(ret == 0))
662 		return NETDEV_TX_OK;
663 
664 	if (ret == -EAGAIN) {
665 		++net_device_ctx->eth_stats.tx_busy;
666 		return NETDEV_TX_BUSY;
667 	}
668 
669 	if (ret == -ENOSPC)
670 		++net_device_ctx->eth_stats.tx_no_space;
671 
672 drop:
673 	dev_kfree_skb_any(skb);
674 	net->stats.tx_dropped++;
675 
676 	return NETDEV_TX_OK;
677 
678 no_memory:
679 	++net_device_ctx->eth_stats.tx_no_memory;
680 	goto drop;
681 }
682 
683 static netdev_tx_t netvsc_start_xmit(struct sk_buff *skb,
684 				     struct net_device *ndev)
685 {
686 	return netvsc_xmit(skb, ndev, false);
687 }
688 
689 /*
690  * netvsc_linkstatus_callback - Link up/down notification
691  */
692 void netvsc_linkstatus_callback(struct net_device *net,
693 				struct rndis_message *resp,
694 				void *data, u32 data_buflen)
695 {
696 	struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
697 	struct net_device_context *ndev_ctx = netdev_priv(net);
698 	struct netvsc_reconfig *event;
699 	unsigned long flags;
700 
701 	/* Ensure the packet is big enough to access its fields */
702 	if (resp->msg_len - RNDIS_HEADER_SIZE < sizeof(struct rndis_indicate_status)) {
703 		netdev_err(net, "invalid rndis_indicate_status packet, len: %u\n",
704 			   resp->msg_len);
705 		return;
706 	}
707 
708 	/* Copy the RNDIS indicate status into nvchan->recv_buf */
709 	memcpy(indicate, data + RNDIS_HEADER_SIZE, sizeof(*indicate));
710 
711 	/* Update the physical link speed when changing to another vSwitch */
712 	if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
713 		u32 speed;
714 
715 		/* Validate status_buf_offset and status_buflen.
716 		 *
717 		 * Certain (pre-Fe) implementations of Hyper-V's vSwitch didn't account
718 		 * for the status buffer field in resp->msg_len; perform the validation
719 		 * using data_buflen (>= resp->msg_len).
720 		 */
721 		if (indicate->status_buflen < sizeof(speed) ||
722 		    indicate->status_buf_offset < sizeof(*indicate) ||
723 		    data_buflen - RNDIS_HEADER_SIZE < indicate->status_buf_offset ||
724 		    data_buflen - RNDIS_HEADER_SIZE - indicate->status_buf_offset
725 				< indicate->status_buflen) {
726 			netdev_err(net, "invalid rndis_indicate_status packet\n");
727 			return;
728 		}
729 
730 		speed = *(u32 *)(data + RNDIS_HEADER_SIZE + indicate->status_buf_offset) / 10000;
731 		ndev_ctx->speed = speed;
732 		return;
733 	}
734 
735 	/* Handle these link change statuses below */
736 	if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
737 	    indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
738 	    indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
739 		return;
740 
741 	if (net->reg_state != NETREG_REGISTERED)
742 		return;
743 
744 	event = kzalloc(sizeof(*event), GFP_ATOMIC);
745 	if (!event)
746 		return;
747 	event->event = indicate->status;
748 
749 	spin_lock_irqsave(&ndev_ctx->lock, flags);
750 	list_add_tail(&event->list, &ndev_ctx->reconfig_events);
751 	spin_unlock_irqrestore(&ndev_ctx->lock, flags);
752 
753 	schedule_delayed_work(&ndev_ctx->dwork, 0);
754 }
755 
756 /* This function should only be called after skb_record_rx_queue() */
757 void netvsc_xdp_xmit(struct sk_buff *skb, struct net_device *ndev)
758 {
759 	int rc;
760 
761 	skb->queue_mapping = skb_get_rx_queue(skb);
762 	__skb_push(skb, ETH_HLEN);
763 
764 	rc = netvsc_xmit(skb, ndev, true);
765 
766 	if (dev_xmit_complete(rc))
767 		return;
768 
769 	dev_kfree_skb_any(skb);
770 	ndev->stats.tx_dropped++;
771 }
772 
773 static void netvsc_comp_ipcsum(struct sk_buff *skb)
774 {
775 	struct iphdr *iph = (struct iphdr *)skb->data;
776 
777 	iph->check = 0;
778 	iph->check = ip_fast_csum(iph, iph->ihl);
779 }
780 
781 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
782 					     struct netvsc_channel *nvchan,
783 					     struct xdp_buff *xdp)
784 {
785 	struct napi_struct *napi = &nvchan->napi;
786 	const struct ndis_pkt_8021q_info *vlan = &nvchan->rsc.vlan;
787 	const struct ndis_tcp_ip_checksum_info *csum_info =
788 						&nvchan->rsc.csum_info;
789 	const u32 *hash_info = &nvchan->rsc.hash_info;
790 	u8 ppi_flags = nvchan->rsc.ppi_flags;
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 = xdp->frame_sz;
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 ((ppi_flags & NVSC_RSC_CSUM_INFO) && csum_info->receive.ip_checksum_value_invalid &&
835 	    csum_info->receive.ip_checksum_succeeded &&
836 	    skb->protocol == htons(ETH_P_IP)) {
837 		/* Check that there is enough space to hold the IP header. */
838 		if (skb_headlen(skb) < sizeof(struct iphdr)) {
839 			kfree_skb(skb);
840 			return NULL;
841 		}
842 		netvsc_comp_ipcsum(skb);
843 	}
844 
845 	/* Do L4 checksum offload if enabled and present. */
846 	if ((ppi_flags & NVSC_RSC_CSUM_INFO) && (net->features & NETIF_F_RXCSUM)) {
847 		if (csum_info->receive.tcp_checksum_succeeded ||
848 		    csum_info->receive.udp_checksum_succeeded)
849 			skb->ip_summed = CHECKSUM_UNNECESSARY;
850 	}
851 
852 	if ((ppi_flags & NVSC_RSC_HASH_INFO) && (net->features & NETIF_F_RXHASH))
853 		skb_set_hash(skb, *hash_info, PKT_HASH_TYPE_L4);
854 
855 	if (ppi_flags & NVSC_RSC_VLAN) {
856 		u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT) |
857 			(vlan->cfi ? VLAN_CFI_MASK : 0);
858 
859 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
860 				       vlan_tci);
861 	}
862 
863 	return skb;
864 }
865 
866 /*
867  * netvsc_recv_callback -  Callback when we receive a packet from the
868  * "wire" on the specified device.
869  */
870 int netvsc_recv_callback(struct net_device *net,
871 			 struct netvsc_device *net_device,
872 			 struct netvsc_channel *nvchan)
873 {
874 	struct net_device_context *net_device_ctx = netdev_priv(net);
875 	struct vmbus_channel *channel = nvchan->channel;
876 	u16 q_idx = channel->offermsg.offer.sub_channel_index;
877 	struct sk_buff *skb;
878 	struct netvsc_stats_rx *rx_stats = &nvchan->rx_stats;
879 	struct xdp_buff xdp;
880 	u32 act;
881 
882 	if (net->reg_state != NETREG_REGISTERED)
883 		return NVSP_STAT_FAIL;
884 
885 	act = netvsc_run_xdp(net, nvchan, &xdp);
886 
887 	if (act == XDP_REDIRECT)
888 		return NVSP_STAT_SUCCESS;
889 
890 	if (act != XDP_PASS && act != XDP_TX) {
891 		u64_stats_update_begin(&rx_stats->syncp);
892 		rx_stats->xdp_drop++;
893 		u64_stats_update_end(&rx_stats->syncp);
894 
895 		return NVSP_STAT_SUCCESS; /* consumed by XDP */
896 	}
897 
898 	/* Allocate a skb - TODO direct I/O to pages? */
899 	skb = netvsc_alloc_recv_skb(net, nvchan, &xdp);
900 
901 	if (unlikely(!skb)) {
902 		++net_device_ctx->eth_stats.rx_no_memory;
903 		return NVSP_STAT_FAIL;
904 	}
905 
906 	skb_record_rx_queue(skb, q_idx);
907 
908 	/*
909 	 * Even if injecting the packet, record the statistics
910 	 * on the synthetic device because modifying the VF device
911 	 * statistics will not work correctly.
912 	 */
913 	u64_stats_update_begin(&rx_stats->syncp);
914 	if (act == XDP_TX)
915 		rx_stats->xdp_tx++;
916 
917 	rx_stats->packets++;
918 	rx_stats->bytes += nvchan->rsc.pktlen;
919 
920 	if (skb->pkt_type == PACKET_BROADCAST)
921 		++rx_stats->broadcast;
922 	else if (skb->pkt_type == PACKET_MULTICAST)
923 		++rx_stats->multicast;
924 	u64_stats_update_end(&rx_stats->syncp);
925 
926 	if (act == XDP_TX) {
927 		netvsc_xdp_xmit(skb, net);
928 		return NVSP_STAT_SUCCESS;
929 	}
930 
931 	napi_gro_receive(&nvchan->napi, skb);
932 	return NVSP_STAT_SUCCESS;
933 }
934 
935 static void netvsc_get_drvinfo(struct net_device *net,
936 			       struct ethtool_drvinfo *info)
937 {
938 	strscpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
939 	strscpy(info->fw_version, "N/A", sizeof(info->fw_version));
940 }
941 
942 static void netvsc_get_channels(struct net_device *net,
943 				struct ethtool_channels *channel)
944 {
945 	struct net_device_context *net_device_ctx = netdev_priv(net);
946 	struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
947 
948 	if (nvdev) {
949 		channel->max_combined	= nvdev->max_chn;
950 		channel->combined_count = nvdev->num_chn;
951 	}
952 }
953 
954 /* Alloc struct netvsc_device_info, and initialize it from either existing
955  * struct netvsc_device, or from default values.
956  */
957 static
958 struct netvsc_device_info *netvsc_devinfo_get(struct netvsc_device *nvdev)
959 {
960 	struct netvsc_device_info *dev_info;
961 	struct bpf_prog *prog;
962 
963 	dev_info = kzalloc(sizeof(*dev_info), GFP_ATOMIC);
964 
965 	if (!dev_info)
966 		return NULL;
967 
968 	if (nvdev) {
969 		ASSERT_RTNL();
970 
971 		dev_info->num_chn = nvdev->num_chn;
972 		dev_info->send_sections = nvdev->send_section_cnt;
973 		dev_info->send_section_size = nvdev->send_section_size;
974 		dev_info->recv_sections = nvdev->recv_section_cnt;
975 		dev_info->recv_section_size = nvdev->recv_section_size;
976 
977 		memcpy(dev_info->rss_key, nvdev->extension->rss_key,
978 		       NETVSC_HASH_KEYLEN);
979 
980 		prog = netvsc_xdp_get(nvdev);
981 		if (prog) {
982 			bpf_prog_inc(prog);
983 			dev_info->bprog = prog;
984 		}
985 	} else {
986 		dev_info->num_chn = VRSS_CHANNEL_DEFAULT;
987 		dev_info->send_sections = NETVSC_DEFAULT_TX;
988 		dev_info->send_section_size = NETVSC_SEND_SECTION_SIZE;
989 		dev_info->recv_sections = NETVSC_DEFAULT_RX;
990 		dev_info->recv_section_size = NETVSC_RECV_SECTION_SIZE;
991 	}
992 
993 	return dev_info;
994 }
995 
996 /* Free struct netvsc_device_info */
997 static void netvsc_devinfo_put(struct netvsc_device_info *dev_info)
998 {
999 	if (dev_info->bprog) {
1000 		ASSERT_RTNL();
1001 		bpf_prog_put(dev_info->bprog);
1002 	}
1003 
1004 	kfree(dev_info);
1005 }
1006 
1007 static int netvsc_detach(struct net_device *ndev,
1008 			 struct netvsc_device *nvdev)
1009 {
1010 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
1011 	struct hv_device *hdev = ndev_ctx->device_ctx;
1012 	int ret;
1013 
1014 	/* Don't try continuing to try and setup sub channels */
1015 	if (cancel_work_sync(&nvdev->subchan_work))
1016 		nvdev->num_chn = 1;
1017 
1018 	netvsc_xdp_set(ndev, NULL, NULL, nvdev);
1019 
1020 	/* If device was up (receiving) then shutdown */
1021 	if (netif_running(ndev)) {
1022 		netvsc_tx_disable(nvdev, ndev);
1023 
1024 		ret = rndis_filter_close(nvdev);
1025 		if (ret) {
1026 			netdev_err(ndev,
1027 				   "unable to close device (ret %d).\n", ret);
1028 			return ret;
1029 		}
1030 
1031 		ret = netvsc_wait_until_empty(nvdev);
1032 		if (ret) {
1033 			netdev_err(ndev,
1034 				   "Ring buffer not empty after closing rndis\n");
1035 			return ret;
1036 		}
1037 	}
1038 
1039 	netif_device_detach(ndev);
1040 
1041 	rndis_filter_device_remove(hdev, nvdev);
1042 
1043 	return 0;
1044 }
1045 
1046 static int netvsc_attach(struct net_device *ndev,
1047 			 struct netvsc_device_info *dev_info)
1048 {
1049 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
1050 	struct hv_device *hdev = ndev_ctx->device_ctx;
1051 	struct netvsc_device *nvdev;
1052 	struct rndis_device *rdev;
1053 	struct bpf_prog *prog;
1054 	int ret = 0;
1055 
1056 	nvdev = rndis_filter_device_add(hdev, dev_info);
1057 	if (IS_ERR(nvdev))
1058 		return PTR_ERR(nvdev);
1059 
1060 	if (nvdev->num_chn > 1) {
1061 		ret = rndis_set_subchannel(ndev, nvdev, dev_info);
1062 
1063 		/* if unavailable, just proceed with one queue */
1064 		if (ret) {
1065 			nvdev->max_chn = 1;
1066 			nvdev->num_chn = 1;
1067 		}
1068 	}
1069 
1070 	prog = dev_info->bprog;
1071 	if (prog) {
1072 		bpf_prog_inc(prog);
1073 		ret = netvsc_xdp_set(ndev, prog, NULL, nvdev);
1074 		if (ret) {
1075 			bpf_prog_put(prog);
1076 			goto err1;
1077 		}
1078 	}
1079 
1080 	/* In any case device is now ready */
1081 	nvdev->tx_disable = false;
1082 	netif_device_attach(ndev);
1083 
1084 	/* Note: enable and attach happen when sub-channels setup */
1085 	netif_carrier_off(ndev);
1086 
1087 	if (netif_running(ndev)) {
1088 		ret = rndis_filter_open(nvdev);
1089 		if (ret)
1090 			goto err2;
1091 
1092 		rdev = nvdev->extension;
1093 		if (!rdev->link_state)
1094 			netif_carrier_on(ndev);
1095 	}
1096 
1097 	return 0;
1098 
1099 err2:
1100 	netif_device_detach(ndev);
1101 
1102 err1:
1103 	rndis_filter_device_remove(hdev, nvdev);
1104 
1105 	return ret;
1106 }
1107 
1108 static int netvsc_set_channels(struct net_device *net,
1109 			       struct ethtool_channels *channels)
1110 {
1111 	struct net_device_context *net_device_ctx = netdev_priv(net);
1112 	struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
1113 	unsigned int orig, count = channels->combined_count;
1114 	struct netvsc_device_info *device_info;
1115 	int ret;
1116 
1117 	/* We do not support separate count for rx, tx, or other */
1118 	if (count == 0 ||
1119 	    channels->rx_count || channels->tx_count || channels->other_count)
1120 		return -EINVAL;
1121 
1122 	if (!nvdev || nvdev->destroy)
1123 		return -ENODEV;
1124 
1125 	if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
1126 		return -EINVAL;
1127 
1128 	if (count > nvdev->max_chn)
1129 		return -EINVAL;
1130 
1131 	orig = nvdev->num_chn;
1132 
1133 	device_info = netvsc_devinfo_get(nvdev);
1134 
1135 	if (!device_info)
1136 		return -ENOMEM;
1137 
1138 	device_info->num_chn = count;
1139 
1140 	ret = netvsc_detach(net, nvdev);
1141 	if (ret)
1142 		goto out;
1143 
1144 	ret = netvsc_attach(net, device_info);
1145 	if (ret) {
1146 		device_info->num_chn = orig;
1147 		if (netvsc_attach(net, device_info))
1148 			netdev_err(net, "restoring channel setting failed\n");
1149 	}
1150 
1151 out:
1152 	netvsc_devinfo_put(device_info);
1153 	return ret;
1154 }
1155 
1156 static void netvsc_init_settings(struct net_device *dev)
1157 {
1158 	struct net_device_context *ndc = netdev_priv(dev);
1159 
1160 	ndc->l4_hash = HV_DEFAULT_L4HASH;
1161 
1162 	ndc->speed = SPEED_UNKNOWN;
1163 	ndc->duplex = DUPLEX_FULL;
1164 
1165 	dev->features = NETIF_F_LRO;
1166 }
1167 
1168 static int netvsc_get_link_ksettings(struct net_device *dev,
1169 				     struct ethtool_link_ksettings *cmd)
1170 {
1171 	struct net_device_context *ndc = netdev_priv(dev);
1172 	struct net_device *vf_netdev;
1173 
1174 	vf_netdev = rtnl_dereference(ndc->vf_netdev);
1175 
1176 	if (vf_netdev)
1177 		return __ethtool_get_link_ksettings(vf_netdev, cmd);
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 	struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1191 
1192 	if (vf_netdev) {
1193 		if (!vf_netdev->ethtool_ops->set_link_ksettings)
1194 			return -EOPNOTSUPP;
1195 
1196 		return vf_netdev->ethtool_ops->set_link_ksettings(vf_netdev,
1197 								  cmd);
1198 	}
1199 
1200 	return ethtool_virtdev_set_link_ksettings(dev, cmd,
1201 						  &ndc->speed, &ndc->duplex);
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(&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(&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(&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(&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_tx *tx_stats;
1313 		const struct netvsc_stats_rx *rx_stats;
1314 		struct netvsc_ethtool_pcpu_stats *this_tot =
1315 			&pcpu_tot[nvchan->channel->target_cpu];
1316 		u64 packets, bytes;
1317 		unsigned int start;
1318 
1319 		tx_stats = &nvchan->tx_stats;
1320 		do {
1321 			start = u64_stats_fetch_begin(&tx_stats->syncp);
1322 			packets = tx_stats->packets;
1323 			bytes = tx_stats->bytes;
1324 		} while (u64_stats_fetch_retry(&tx_stats->syncp, start));
1325 
1326 		this_tot->tx_bytes	+= bytes;
1327 		this_tot->tx_packets	+= packets;
1328 
1329 		rx_stats = &nvchan->rx_stats;
1330 		do {
1331 			start = u64_stats_fetch_begin(&rx_stats->syncp);
1332 			packets = rx_stats->packets;
1333 			bytes = rx_stats->bytes;
1334 		} while (u64_stats_fetch_retry(&rx_stats->syncp, start));
1335 
1336 		this_tot->rx_bytes	+= bytes;
1337 		this_tot->rx_packets	+= packets;
1338 	}
1339 }
1340 
1341 static void netvsc_get_stats64(struct net_device *net,
1342 			       struct rtnl_link_stats64 *t)
1343 {
1344 	struct net_device_context *ndev_ctx = netdev_priv(net);
1345 	struct netvsc_device *nvdev;
1346 	struct netvsc_vf_pcpu_stats vf_tot;
1347 	int i;
1348 
1349 	rcu_read_lock();
1350 
1351 	nvdev = rcu_dereference(ndev_ctx->nvdev);
1352 	if (!nvdev)
1353 		goto out;
1354 
1355 	netdev_stats_to_stats64(t, &net->stats);
1356 
1357 	netvsc_get_vf_stats(net, &vf_tot);
1358 	t->rx_packets += vf_tot.rx_packets;
1359 	t->tx_packets += vf_tot.tx_packets;
1360 	t->rx_bytes   += vf_tot.rx_bytes;
1361 	t->tx_bytes   += vf_tot.tx_bytes;
1362 	t->tx_dropped += vf_tot.tx_dropped;
1363 
1364 	for (i = 0; i < nvdev->num_chn; i++) {
1365 		const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1366 		const struct netvsc_stats_tx *tx_stats;
1367 		const struct netvsc_stats_rx *rx_stats;
1368 		u64 packets, bytes, multicast;
1369 		unsigned int start;
1370 
1371 		tx_stats = &nvchan->tx_stats;
1372 		do {
1373 			start = u64_stats_fetch_begin(&tx_stats->syncp);
1374 			packets = tx_stats->packets;
1375 			bytes = tx_stats->bytes;
1376 		} while (u64_stats_fetch_retry(&tx_stats->syncp, start));
1377 
1378 		t->tx_bytes	+= bytes;
1379 		t->tx_packets	+= packets;
1380 
1381 		rx_stats = &nvchan->rx_stats;
1382 		do {
1383 			start = u64_stats_fetch_begin(&rx_stats->syncp);
1384 			packets = rx_stats->packets;
1385 			bytes = rx_stats->bytes;
1386 			multicast = rx_stats->multicast + rx_stats->broadcast;
1387 		} while (u64_stats_fetch_retry(&rx_stats->syncp, start));
1388 
1389 		t->rx_bytes	+= bytes;
1390 		t->rx_packets	+= packets;
1391 		t->multicast	+= multicast;
1392 	}
1393 out:
1394 	rcu_read_unlock();
1395 }
1396 
1397 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1398 {
1399 	struct net_device_context *ndc = netdev_priv(ndev);
1400 	struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1401 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1402 	struct sockaddr *addr = p;
1403 	int err;
1404 
1405 	err = eth_prepare_mac_addr_change(ndev, p);
1406 	if (err)
1407 		return err;
1408 
1409 	if (!nvdev)
1410 		return -ENODEV;
1411 
1412 	if (vf_netdev) {
1413 		err = dev_set_mac_address(vf_netdev, addr, NULL);
1414 		if (err)
1415 			return err;
1416 	}
1417 
1418 	err = rndis_filter_set_device_mac(nvdev, addr->sa_data);
1419 	if (!err) {
1420 		eth_commit_mac_addr_change(ndev, p);
1421 	} else if (vf_netdev) {
1422 		/* rollback change on VF */
1423 		memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN);
1424 		dev_set_mac_address(vf_netdev, addr, NULL);
1425 	}
1426 
1427 	return err;
1428 }
1429 
1430 static const struct {
1431 	char name[ETH_GSTRING_LEN];
1432 	u16 offset;
1433 } netvsc_stats[] = {
1434 	{ "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1435 	{ "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1436 	{ "tx_no_space",  offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1437 	{ "tx_too_big",	  offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1438 	{ "tx_busy",	  offsetof(struct netvsc_ethtool_stats, tx_busy) },
1439 	{ "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) },
1440 	{ "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) },
1441 	{ "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) },
1442 	{ "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) },
1443 	{ "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) },
1444 	{ "vlan_error", offsetof(struct netvsc_ethtool_stats, vlan_error) },
1445 }, pcpu_stats[] = {
1446 	{ "cpu%u_rx_packets",
1447 		offsetof(struct netvsc_ethtool_pcpu_stats, rx_packets) },
1448 	{ "cpu%u_rx_bytes",
1449 		offsetof(struct netvsc_ethtool_pcpu_stats, rx_bytes) },
1450 	{ "cpu%u_tx_packets",
1451 		offsetof(struct netvsc_ethtool_pcpu_stats, tx_packets) },
1452 	{ "cpu%u_tx_bytes",
1453 		offsetof(struct netvsc_ethtool_pcpu_stats, tx_bytes) },
1454 	{ "cpu%u_vf_rx_packets",
1455 		offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_packets) },
1456 	{ "cpu%u_vf_rx_bytes",
1457 		offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_bytes) },
1458 	{ "cpu%u_vf_tx_packets",
1459 		offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_packets) },
1460 	{ "cpu%u_vf_tx_bytes",
1461 		offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_bytes) },
1462 }, vf_stats[] = {
1463 	{ "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) },
1464 	{ "vf_rx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) },
1465 	{ "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) },
1466 	{ "vf_tx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) },
1467 	{ "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) },
1468 };
1469 
1470 #define NETVSC_GLOBAL_STATS_LEN	ARRAY_SIZE(netvsc_stats)
1471 #define NETVSC_VF_STATS_LEN	ARRAY_SIZE(vf_stats)
1472 
1473 /* statistics per queue (rx/tx packets/bytes) */
1474 #define NETVSC_PCPU_STATS_LEN (num_present_cpus() * ARRAY_SIZE(pcpu_stats))
1475 
1476 /* 8 statistics per queue (rx/tx packets/bytes, XDP actions) */
1477 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 8)
1478 
1479 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1480 {
1481 	struct net_device_context *ndc = netdev_priv(dev);
1482 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1483 
1484 	if (!nvdev)
1485 		return -ENODEV;
1486 
1487 	switch (string_set) {
1488 	case ETH_SS_STATS:
1489 		return NETVSC_GLOBAL_STATS_LEN
1490 			+ NETVSC_VF_STATS_LEN
1491 			+ NETVSC_QUEUE_STATS_LEN(nvdev)
1492 			+ NETVSC_PCPU_STATS_LEN;
1493 	default:
1494 		return -EINVAL;
1495 	}
1496 }
1497 
1498 static void netvsc_get_ethtool_stats(struct net_device *dev,
1499 				     struct ethtool_stats *stats, u64 *data)
1500 {
1501 	struct net_device_context *ndc = netdev_priv(dev);
1502 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1503 	const void *nds = &ndc->eth_stats;
1504 	const struct netvsc_stats_tx *tx_stats;
1505 	const struct netvsc_stats_rx *rx_stats;
1506 	struct netvsc_vf_pcpu_stats sum;
1507 	struct netvsc_ethtool_pcpu_stats *pcpu_sum;
1508 	unsigned int start;
1509 	u64 packets, bytes;
1510 	u64 xdp_drop;
1511 	u64 xdp_redirect;
1512 	u64 xdp_tx;
1513 	u64 xdp_xmit;
1514 	int i, j, cpu;
1515 
1516 	if (!nvdev)
1517 		return;
1518 
1519 	for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1520 		data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1521 
1522 	netvsc_get_vf_stats(dev, &sum);
1523 	for (j = 0; j < NETVSC_VF_STATS_LEN; j++)
1524 		data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset);
1525 
1526 	for (j = 0; j < nvdev->num_chn; j++) {
1527 		tx_stats = &nvdev->chan_table[j].tx_stats;
1528 
1529 		do {
1530 			start = u64_stats_fetch_begin(&tx_stats->syncp);
1531 			packets = tx_stats->packets;
1532 			bytes = tx_stats->bytes;
1533 			xdp_xmit = tx_stats->xdp_xmit;
1534 		} while (u64_stats_fetch_retry(&tx_stats->syncp, start));
1535 		data[i++] = packets;
1536 		data[i++] = bytes;
1537 		data[i++] = xdp_xmit;
1538 
1539 		rx_stats = &nvdev->chan_table[j].rx_stats;
1540 		do {
1541 			start = u64_stats_fetch_begin(&rx_stats->syncp);
1542 			packets = rx_stats->packets;
1543 			bytes = rx_stats->bytes;
1544 			xdp_drop = rx_stats->xdp_drop;
1545 			xdp_redirect = rx_stats->xdp_redirect;
1546 			xdp_tx = rx_stats->xdp_tx;
1547 		} while (u64_stats_fetch_retry(&rx_stats->syncp, start));
1548 		data[i++] = packets;
1549 		data[i++] = bytes;
1550 		data[i++] = xdp_drop;
1551 		data[i++] = xdp_redirect;
1552 		data[i++] = xdp_tx;
1553 	}
1554 
1555 	pcpu_sum = kvmalloc_array(num_possible_cpus(),
1556 				  sizeof(struct netvsc_ethtool_pcpu_stats),
1557 				  GFP_KERNEL);
1558 	if (!pcpu_sum)
1559 		return;
1560 
1561 	netvsc_get_pcpu_stats(dev, pcpu_sum);
1562 	for_each_present_cpu(cpu) {
1563 		struct netvsc_ethtool_pcpu_stats *this_sum = &pcpu_sum[cpu];
1564 
1565 		for (j = 0; j < ARRAY_SIZE(pcpu_stats); j++)
1566 			data[i++] = *(u64 *)((void *)this_sum
1567 					     + pcpu_stats[j].offset);
1568 	}
1569 	kvfree(pcpu_sum);
1570 }
1571 
1572 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1573 {
1574 	struct net_device_context *ndc = netdev_priv(dev);
1575 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1576 	u8 *p = data;
1577 	int i, cpu;
1578 
1579 	if (!nvdev)
1580 		return;
1581 
1582 	switch (stringset) {
1583 	case ETH_SS_STATS:
1584 		for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++)
1585 			ethtool_sprintf(&p, netvsc_stats[i].name);
1586 
1587 		for (i = 0; i < ARRAY_SIZE(vf_stats); i++)
1588 			ethtool_sprintf(&p, vf_stats[i].name);
1589 
1590 		for (i = 0; i < nvdev->num_chn; i++) {
1591 			ethtool_sprintf(&p, "tx_queue_%u_packets", i);
1592 			ethtool_sprintf(&p, "tx_queue_%u_bytes", i);
1593 			ethtool_sprintf(&p, "tx_queue_%u_xdp_xmit", i);
1594 			ethtool_sprintf(&p, "rx_queue_%u_packets", i);
1595 			ethtool_sprintf(&p, "rx_queue_%u_bytes", i);
1596 			ethtool_sprintf(&p, "rx_queue_%u_xdp_drop", i);
1597 			ethtool_sprintf(&p, "rx_queue_%u_xdp_redirect", i);
1598 			ethtool_sprintf(&p, "rx_queue_%u_xdp_tx", i);
1599 		}
1600 
1601 		for_each_present_cpu(cpu) {
1602 			for (i = 0; i < ARRAY_SIZE(pcpu_stats); i++)
1603 				ethtool_sprintf(&p, pcpu_stats[i].name, cpu);
1604 		}
1605 
1606 		break;
1607 	}
1608 }
1609 
1610 static int
1611 netvsc_get_rss_hash_opts(struct net_device_context *ndc,
1612 			 struct ethtool_rxnfc *info)
1613 {
1614 	const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3;
1615 
1616 	info->data = RXH_IP_SRC | RXH_IP_DST;
1617 
1618 	switch (info->flow_type) {
1619 	case TCP_V4_FLOW:
1620 		if (ndc->l4_hash & HV_TCP4_L4HASH)
1621 			info->data |= l4_flag;
1622 
1623 		break;
1624 
1625 	case TCP_V6_FLOW:
1626 		if (ndc->l4_hash & HV_TCP6_L4HASH)
1627 			info->data |= l4_flag;
1628 
1629 		break;
1630 
1631 	case UDP_V4_FLOW:
1632 		if (ndc->l4_hash & HV_UDP4_L4HASH)
1633 			info->data |= l4_flag;
1634 
1635 		break;
1636 
1637 	case UDP_V6_FLOW:
1638 		if (ndc->l4_hash & HV_UDP6_L4HASH)
1639 			info->data |= l4_flag;
1640 
1641 		break;
1642 
1643 	case IPV4_FLOW:
1644 	case IPV6_FLOW:
1645 		break;
1646 	default:
1647 		info->data = 0;
1648 		break;
1649 	}
1650 
1651 	return 0;
1652 }
1653 
1654 static int
1655 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1656 		 u32 *rules)
1657 {
1658 	struct net_device_context *ndc = netdev_priv(dev);
1659 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1660 
1661 	if (!nvdev)
1662 		return -ENODEV;
1663 
1664 	switch (info->cmd) {
1665 	case ETHTOOL_GRXRINGS:
1666 		info->data = nvdev->num_chn;
1667 		return 0;
1668 
1669 	case ETHTOOL_GRXFH:
1670 		return netvsc_get_rss_hash_opts(ndc, info);
1671 	}
1672 	return -EOPNOTSUPP;
1673 }
1674 
1675 static int netvsc_set_rss_hash_opts(struct net_device_context *ndc,
1676 				    struct ethtool_rxnfc *info)
1677 {
1678 	if (info->data == (RXH_IP_SRC | RXH_IP_DST |
1679 			   RXH_L4_B_0_1 | RXH_L4_B_2_3)) {
1680 		switch (info->flow_type) {
1681 		case TCP_V4_FLOW:
1682 			ndc->l4_hash |= HV_TCP4_L4HASH;
1683 			break;
1684 
1685 		case TCP_V6_FLOW:
1686 			ndc->l4_hash |= HV_TCP6_L4HASH;
1687 			break;
1688 
1689 		case UDP_V4_FLOW:
1690 			ndc->l4_hash |= HV_UDP4_L4HASH;
1691 			break;
1692 
1693 		case UDP_V6_FLOW:
1694 			ndc->l4_hash |= HV_UDP6_L4HASH;
1695 			break;
1696 
1697 		default:
1698 			return -EOPNOTSUPP;
1699 		}
1700 
1701 		return 0;
1702 	}
1703 
1704 	if (info->data == (RXH_IP_SRC | RXH_IP_DST)) {
1705 		switch (info->flow_type) {
1706 		case TCP_V4_FLOW:
1707 			ndc->l4_hash &= ~HV_TCP4_L4HASH;
1708 			break;
1709 
1710 		case TCP_V6_FLOW:
1711 			ndc->l4_hash &= ~HV_TCP6_L4HASH;
1712 			break;
1713 
1714 		case UDP_V4_FLOW:
1715 			ndc->l4_hash &= ~HV_UDP4_L4HASH;
1716 			break;
1717 
1718 		case UDP_V6_FLOW:
1719 			ndc->l4_hash &= ~HV_UDP6_L4HASH;
1720 			break;
1721 
1722 		default:
1723 			return -EOPNOTSUPP;
1724 		}
1725 
1726 		return 0;
1727 	}
1728 
1729 	return -EOPNOTSUPP;
1730 }
1731 
1732 static int
1733 netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info)
1734 {
1735 	struct net_device_context *ndc = netdev_priv(ndev);
1736 
1737 	if (info->cmd == ETHTOOL_SRXFH)
1738 		return netvsc_set_rss_hash_opts(ndc, info);
1739 
1740 	return -EOPNOTSUPP;
1741 }
1742 
1743 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1744 {
1745 	return NETVSC_HASH_KEYLEN;
1746 }
1747 
1748 static u32 netvsc_rss_indir_size(struct net_device *dev)
1749 {
1750 	struct net_device_context *ndc = netdev_priv(dev);
1751 
1752 	return ndc->rx_table_sz;
1753 }
1754 
1755 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1756 			   u8 *hfunc)
1757 {
1758 	struct net_device_context *ndc = netdev_priv(dev);
1759 	struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1760 	struct rndis_device *rndis_dev;
1761 	int i;
1762 
1763 	if (!ndev)
1764 		return -ENODEV;
1765 
1766 	if (hfunc)
1767 		*hfunc = ETH_RSS_HASH_TOP;	/* Toeplitz */
1768 
1769 	rndis_dev = ndev->extension;
1770 	if (indir) {
1771 		for (i = 0; i < ndc->rx_table_sz; i++)
1772 			indir[i] = ndc->rx_table[i];
1773 	}
1774 
1775 	if (key)
1776 		memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1777 
1778 	return 0;
1779 }
1780 
1781 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1782 			   const u8 *key, const u8 hfunc)
1783 {
1784 	struct net_device_context *ndc = netdev_priv(dev);
1785 	struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1786 	struct rndis_device *rndis_dev;
1787 	int i;
1788 
1789 	if (!ndev)
1790 		return -ENODEV;
1791 
1792 	if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1793 		return -EOPNOTSUPP;
1794 
1795 	rndis_dev = ndev->extension;
1796 	if (indir) {
1797 		for (i = 0; i < ndc->rx_table_sz; i++)
1798 			if (indir[i] >= ndev->num_chn)
1799 				return -EINVAL;
1800 
1801 		for (i = 0; i < ndc->rx_table_sz; i++)
1802 			ndc->rx_table[i] = indir[i];
1803 	}
1804 
1805 	if (!key) {
1806 		if (!indir)
1807 			return 0;
1808 
1809 		key = rndis_dev->rss_key;
1810 	}
1811 
1812 	return rndis_filter_set_rss_param(rndis_dev, key);
1813 }
1814 
1815 /* Hyper-V RNDIS protocol does not have ring in the HW sense.
1816  * It does have pre-allocated receive area which is divided into sections.
1817  */
1818 static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
1819 				   struct ethtool_ringparam *ring)
1820 {
1821 	u32 max_buf_size;
1822 
1823 	ring->rx_pending = nvdev->recv_section_cnt;
1824 	ring->tx_pending = nvdev->send_section_cnt;
1825 
1826 	if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
1827 		max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
1828 	else
1829 		max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1830 
1831 	ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
1832 	ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
1833 		/ nvdev->send_section_size;
1834 }
1835 
1836 static void netvsc_get_ringparam(struct net_device *ndev,
1837 				 struct ethtool_ringparam *ring,
1838 				 struct kernel_ethtool_ringparam *kernel_ring,
1839 				 struct netlink_ext_ack *extack)
1840 {
1841 	struct net_device_context *ndevctx = netdev_priv(ndev);
1842 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1843 
1844 	if (!nvdev)
1845 		return;
1846 
1847 	__netvsc_get_ringparam(nvdev, ring);
1848 }
1849 
1850 static int netvsc_set_ringparam(struct net_device *ndev,
1851 				struct ethtool_ringparam *ring,
1852 				struct kernel_ethtool_ringparam *kernel_ring,
1853 				struct netlink_ext_ack *extack)
1854 {
1855 	struct net_device_context *ndevctx = netdev_priv(ndev);
1856 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1857 	struct netvsc_device_info *device_info;
1858 	struct ethtool_ringparam orig;
1859 	u32 new_tx, new_rx;
1860 	int ret = 0;
1861 
1862 	if (!nvdev || nvdev->destroy)
1863 		return -ENODEV;
1864 
1865 	memset(&orig, 0, sizeof(orig));
1866 	__netvsc_get_ringparam(nvdev, &orig);
1867 
1868 	new_tx = clamp_t(u32, ring->tx_pending,
1869 			 NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
1870 	new_rx = clamp_t(u32, ring->rx_pending,
1871 			 NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
1872 
1873 	if (new_tx == orig.tx_pending &&
1874 	    new_rx == orig.rx_pending)
1875 		return 0;	 /* no change */
1876 
1877 	device_info = netvsc_devinfo_get(nvdev);
1878 
1879 	if (!device_info)
1880 		return -ENOMEM;
1881 
1882 	device_info->send_sections = new_tx;
1883 	device_info->recv_sections = new_rx;
1884 
1885 	ret = netvsc_detach(ndev, nvdev);
1886 	if (ret)
1887 		goto out;
1888 
1889 	ret = netvsc_attach(ndev, device_info);
1890 	if (ret) {
1891 		device_info->send_sections = orig.tx_pending;
1892 		device_info->recv_sections = orig.rx_pending;
1893 
1894 		if (netvsc_attach(ndev, device_info))
1895 			netdev_err(ndev, "restoring ringparam failed");
1896 	}
1897 
1898 out:
1899 	netvsc_devinfo_put(device_info);
1900 	return ret;
1901 }
1902 
1903 static netdev_features_t netvsc_fix_features(struct net_device *ndev,
1904 					     netdev_features_t features)
1905 {
1906 	struct net_device_context *ndevctx = netdev_priv(ndev);
1907 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1908 
1909 	if (!nvdev || nvdev->destroy)
1910 		return features;
1911 
1912 	if ((features & NETIF_F_LRO) && netvsc_xdp_get(nvdev)) {
1913 		features ^= NETIF_F_LRO;
1914 		netdev_info(ndev, "Skip LRO - unsupported with XDP\n");
1915 	}
1916 
1917 	return features;
1918 }
1919 
1920 static int netvsc_set_features(struct net_device *ndev,
1921 			       netdev_features_t features)
1922 {
1923 	netdev_features_t change = features ^ ndev->features;
1924 	struct net_device_context *ndevctx = netdev_priv(ndev);
1925 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1926 	struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1927 	struct ndis_offload_params offloads;
1928 	int ret = 0;
1929 
1930 	if (!nvdev || nvdev->destroy)
1931 		return -ENODEV;
1932 
1933 	if (!(change & NETIF_F_LRO))
1934 		goto syncvf;
1935 
1936 	memset(&offloads, 0, sizeof(struct ndis_offload_params));
1937 
1938 	if (features & NETIF_F_LRO) {
1939 		offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1940 		offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1941 	} else {
1942 		offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1943 		offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1944 	}
1945 
1946 	ret = rndis_filter_set_offload_params(ndev, nvdev, &offloads);
1947 
1948 	if (ret) {
1949 		features ^= NETIF_F_LRO;
1950 		ndev->features = features;
1951 	}
1952 
1953 syncvf:
1954 	if (!vf_netdev)
1955 		return ret;
1956 
1957 	vf_netdev->wanted_features = features;
1958 	netdev_update_features(vf_netdev);
1959 
1960 	return ret;
1961 }
1962 
1963 static int netvsc_get_regs_len(struct net_device *netdev)
1964 {
1965 	return VRSS_SEND_TAB_SIZE * sizeof(u32);
1966 }
1967 
1968 static void netvsc_get_regs(struct net_device *netdev,
1969 			    struct ethtool_regs *regs, void *p)
1970 {
1971 	struct net_device_context *ndc = netdev_priv(netdev);
1972 	u32 *regs_buff = p;
1973 
1974 	/* increase the version, if buffer format is changed. */
1975 	regs->version = 1;
1976 
1977 	memcpy(regs_buff, ndc->tx_table, VRSS_SEND_TAB_SIZE * sizeof(u32));
1978 }
1979 
1980 static u32 netvsc_get_msglevel(struct net_device *ndev)
1981 {
1982 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
1983 
1984 	return ndev_ctx->msg_enable;
1985 }
1986 
1987 static void netvsc_set_msglevel(struct net_device *ndev, u32 val)
1988 {
1989 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
1990 
1991 	ndev_ctx->msg_enable = val;
1992 }
1993 
1994 static const struct ethtool_ops ethtool_ops = {
1995 	.get_drvinfo	= netvsc_get_drvinfo,
1996 	.get_regs_len	= netvsc_get_regs_len,
1997 	.get_regs	= netvsc_get_regs,
1998 	.get_msglevel	= netvsc_get_msglevel,
1999 	.set_msglevel	= netvsc_set_msglevel,
2000 	.get_link	= ethtool_op_get_link,
2001 	.get_ethtool_stats = netvsc_get_ethtool_stats,
2002 	.get_sset_count = netvsc_get_sset_count,
2003 	.get_strings	= netvsc_get_strings,
2004 	.get_channels   = netvsc_get_channels,
2005 	.set_channels   = netvsc_set_channels,
2006 	.get_ts_info	= ethtool_op_get_ts_info,
2007 	.get_rxnfc	= netvsc_get_rxnfc,
2008 	.set_rxnfc	= netvsc_set_rxnfc,
2009 	.get_rxfh_key_size = netvsc_get_rxfh_key_size,
2010 	.get_rxfh_indir_size = netvsc_rss_indir_size,
2011 	.get_rxfh	= netvsc_get_rxfh,
2012 	.set_rxfh	= netvsc_set_rxfh,
2013 	.get_link_ksettings = netvsc_get_link_ksettings,
2014 	.set_link_ksettings = netvsc_set_link_ksettings,
2015 	.get_ringparam	= netvsc_get_ringparam,
2016 	.set_ringparam	= netvsc_set_ringparam,
2017 };
2018 
2019 static const struct net_device_ops device_ops = {
2020 	.ndo_open =			netvsc_open,
2021 	.ndo_stop =			netvsc_close,
2022 	.ndo_start_xmit =		netvsc_start_xmit,
2023 	.ndo_change_rx_flags =		netvsc_change_rx_flags,
2024 	.ndo_set_rx_mode =		netvsc_set_rx_mode,
2025 	.ndo_fix_features =		netvsc_fix_features,
2026 	.ndo_set_features =		netvsc_set_features,
2027 	.ndo_change_mtu =		netvsc_change_mtu,
2028 	.ndo_validate_addr =		eth_validate_addr,
2029 	.ndo_set_mac_address =		netvsc_set_mac_addr,
2030 	.ndo_select_queue =		netvsc_select_queue,
2031 	.ndo_get_stats64 =		netvsc_get_stats64,
2032 	.ndo_bpf =			netvsc_bpf,
2033 	.ndo_xdp_xmit =			netvsc_ndoxdp_xmit,
2034 };
2035 
2036 /*
2037  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
2038  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
2039  * present send GARP packet to network peers with netif_notify_peers().
2040  */
2041 static void netvsc_link_change(struct work_struct *w)
2042 {
2043 	struct net_device_context *ndev_ctx =
2044 		container_of(w, struct net_device_context, dwork.work);
2045 	struct hv_device *device_obj = ndev_ctx->device_ctx;
2046 	struct net_device *net = hv_get_drvdata(device_obj);
2047 	unsigned long flags, next_reconfig, delay;
2048 	struct netvsc_reconfig *event = NULL;
2049 	struct netvsc_device *net_device;
2050 	struct rndis_device *rdev;
2051 	bool reschedule = false;
2052 
2053 	/* if changes are happening, comeback later */
2054 	if (!rtnl_trylock()) {
2055 		schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
2056 		return;
2057 	}
2058 
2059 	net_device = rtnl_dereference(ndev_ctx->nvdev);
2060 	if (!net_device)
2061 		goto out_unlock;
2062 
2063 	rdev = net_device->extension;
2064 
2065 	next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
2066 	if (time_is_after_jiffies(next_reconfig)) {
2067 		/* link_watch only sends one notification with current state
2068 		 * per second, avoid doing reconfig more frequently. Handle
2069 		 * wrap around.
2070 		 */
2071 		delay = next_reconfig - jiffies;
2072 		delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
2073 		schedule_delayed_work(&ndev_ctx->dwork, delay);
2074 		goto out_unlock;
2075 	}
2076 	ndev_ctx->last_reconfig = jiffies;
2077 
2078 	spin_lock_irqsave(&ndev_ctx->lock, flags);
2079 	if (!list_empty(&ndev_ctx->reconfig_events)) {
2080 		event = list_first_entry(&ndev_ctx->reconfig_events,
2081 					 struct netvsc_reconfig, list);
2082 		list_del(&event->list);
2083 		reschedule = !list_empty(&ndev_ctx->reconfig_events);
2084 	}
2085 	spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2086 
2087 	if (!event)
2088 		goto out_unlock;
2089 
2090 	switch (event->event) {
2091 		/* Only the following events are possible due to the check in
2092 		 * netvsc_linkstatus_callback()
2093 		 */
2094 	case RNDIS_STATUS_MEDIA_CONNECT:
2095 		if (rdev->link_state) {
2096 			rdev->link_state = false;
2097 			netif_carrier_on(net);
2098 			netvsc_tx_enable(net_device, net);
2099 		} else {
2100 			__netdev_notify_peers(net);
2101 		}
2102 		kfree(event);
2103 		break;
2104 	case RNDIS_STATUS_MEDIA_DISCONNECT:
2105 		if (!rdev->link_state) {
2106 			rdev->link_state = true;
2107 			netif_carrier_off(net);
2108 			netvsc_tx_disable(net_device, net);
2109 		}
2110 		kfree(event);
2111 		break;
2112 	case RNDIS_STATUS_NETWORK_CHANGE:
2113 		/* Only makes sense if carrier is present */
2114 		if (!rdev->link_state) {
2115 			rdev->link_state = true;
2116 			netif_carrier_off(net);
2117 			netvsc_tx_disable(net_device, net);
2118 			event->event = RNDIS_STATUS_MEDIA_CONNECT;
2119 			spin_lock_irqsave(&ndev_ctx->lock, flags);
2120 			list_add(&event->list, &ndev_ctx->reconfig_events);
2121 			spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2122 			reschedule = true;
2123 		}
2124 		break;
2125 	}
2126 
2127 	rtnl_unlock();
2128 
2129 	/* link_watch only sends one notification with current state per
2130 	 * second, handle next reconfig event in 2 seconds.
2131 	 */
2132 	if (reschedule)
2133 		schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
2134 
2135 	return;
2136 
2137 out_unlock:
2138 	rtnl_unlock();
2139 }
2140 
2141 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
2142 {
2143 	struct net_device_context *net_device_ctx;
2144 	struct net_device *dev;
2145 
2146 	dev = netdev_master_upper_dev_get(vf_netdev);
2147 	if (!dev || dev->netdev_ops != &device_ops)
2148 		return NULL;	/* not a netvsc device */
2149 
2150 	net_device_ctx = netdev_priv(dev);
2151 	if (!rtnl_dereference(net_device_ctx->nvdev))
2152 		return NULL;	/* device is removed */
2153 
2154 	return dev;
2155 }
2156 
2157 /* Called when VF is injecting data into network stack.
2158  * Change the associated network device from VF to netvsc.
2159  * note: already called with rcu_read_lock
2160  */
2161 static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
2162 {
2163 	struct sk_buff *skb = *pskb;
2164 	struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data);
2165 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
2166 	struct netvsc_vf_pcpu_stats *pcpu_stats
2167 		 = this_cpu_ptr(ndev_ctx->vf_stats);
2168 
2169 	skb = skb_share_check(skb, GFP_ATOMIC);
2170 	if (unlikely(!skb))
2171 		return RX_HANDLER_CONSUMED;
2172 
2173 	*pskb = skb;
2174 
2175 	skb->dev = ndev;
2176 
2177 	u64_stats_update_begin(&pcpu_stats->syncp);
2178 	pcpu_stats->rx_packets++;
2179 	pcpu_stats->rx_bytes += skb->len;
2180 	u64_stats_update_end(&pcpu_stats->syncp);
2181 
2182 	return RX_HANDLER_ANOTHER;
2183 }
2184 
2185 static int netvsc_vf_join(struct net_device *vf_netdev,
2186 			  struct net_device *ndev)
2187 {
2188 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
2189 	int ret;
2190 
2191 	ret = netdev_rx_handler_register(vf_netdev,
2192 					 netvsc_vf_handle_frame, ndev);
2193 	if (ret != 0) {
2194 		netdev_err(vf_netdev,
2195 			   "can not register netvsc VF receive handler (err = %d)\n",
2196 			   ret);
2197 		goto rx_handler_failed;
2198 	}
2199 
2200 	ret = netdev_master_upper_dev_link(vf_netdev, ndev,
2201 					   NULL, NULL, NULL);
2202 	if (ret != 0) {
2203 		netdev_err(vf_netdev,
2204 			   "can not set master device %s (err = %d)\n",
2205 			   ndev->name, ret);
2206 		goto upper_link_failed;
2207 	}
2208 
2209 	schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
2210 
2211 	call_netdevice_notifiers(NETDEV_JOIN, vf_netdev);
2212 
2213 	netdev_info(vf_netdev, "joined to %s\n", ndev->name);
2214 	return 0;
2215 
2216 upper_link_failed:
2217 	netdev_rx_handler_unregister(vf_netdev);
2218 rx_handler_failed:
2219 	return ret;
2220 }
2221 
2222 static void __netvsc_vf_setup(struct net_device *ndev,
2223 			      struct net_device *vf_netdev)
2224 {
2225 	int ret;
2226 
2227 	/* Align MTU of VF with master */
2228 	ret = dev_set_mtu(vf_netdev, ndev->mtu);
2229 	if (ret)
2230 		netdev_warn(vf_netdev,
2231 			    "unable to change mtu to %u\n", ndev->mtu);
2232 
2233 	/* set multicast etc flags on VF */
2234 	dev_change_flags(vf_netdev, ndev->flags | IFF_SLAVE, NULL);
2235 
2236 	/* sync address list from ndev to VF */
2237 	netif_addr_lock_bh(ndev);
2238 	dev_uc_sync(vf_netdev, ndev);
2239 	dev_mc_sync(vf_netdev, ndev);
2240 	netif_addr_unlock_bh(ndev);
2241 
2242 	if (netif_running(ndev)) {
2243 		ret = dev_open(vf_netdev, NULL);
2244 		if (ret)
2245 			netdev_warn(vf_netdev,
2246 				    "unable to open: %d\n", ret);
2247 	}
2248 }
2249 
2250 /* Setup VF as slave of the synthetic device.
2251  * Runs in workqueue to avoid recursion in netlink callbacks.
2252  */
2253 static void netvsc_vf_setup(struct work_struct *w)
2254 {
2255 	struct net_device_context *ndev_ctx
2256 		= container_of(w, struct net_device_context, vf_takeover.work);
2257 	struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2258 	struct net_device *vf_netdev;
2259 
2260 	if (!rtnl_trylock()) {
2261 		schedule_delayed_work(&ndev_ctx->vf_takeover, 0);
2262 		return;
2263 	}
2264 
2265 	vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2266 	if (vf_netdev)
2267 		__netvsc_vf_setup(ndev, vf_netdev);
2268 
2269 	rtnl_unlock();
2270 }
2271 
2272 /* Find netvsc by VF serial number.
2273  * The PCI hyperv controller records the serial number as the slot kobj name.
2274  */
2275 static struct net_device *get_netvsc_byslot(const struct net_device *vf_netdev)
2276 {
2277 	struct device *parent = vf_netdev->dev.parent;
2278 	struct net_device_context *ndev_ctx;
2279 	struct net_device *ndev;
2280 	struct pci_dev *pdev;
2281 	u32 serial;
2282 
2283 	if (!parent || !dev_is_pci(parent))
2284 		return NULL; /* not a PCI device */
2285 
2286 	pdev = to_pci_dev(parent);
2287 	if (!pdev->slot) {
2288 		netdev_notice(vf_netdev, "no PCI slot information\n");
2289 		return NULL;
2290 	}
2291 
2292 	if (kstrtou32(pci_slot_name(pdev->slot), 10, &serial)) {
2293 		netdev_notice(vf_netdev, "Invalid vf serial:%s\n",
2294 			      pci_slot_name(pdev->slot));
2295 		return NULL;
2296 	}
2297 
2298 	list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
2299 		if (!ndev_ctx->vf_alloc)
2300 			continue;
2301 
2302 		if (ndev_ctx->vf_serial != serial)
2303 			continue;
2304 
2305 		ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2306 		if (ndev->addr_len != vf_netdev->addr_len ||
2307 		    memcmp(ndev->perm_addr, vf_netdev->perm_addr,
2308 			   ndev->addr_len) != 0)
2309 			continue;
2310 
2311 		return ndev;
2312 
2313 	}
2314 
2315 	/* Fallback path to check synthetic vf with help of mac addr.
2316 	 * Because this function can be called before vf_netdev is
2317 	 * initialized (NETDEV_POST_INIT) when its perm_addr has not been copied
2318 	 * from dev_addr, also try to match to its dev_addr.
2319 	 * Note: On Hyper-V and Azure, it's not possible to set a MAC address
2320 	 * on a VF that matches to the MAC of a unrelated NETVSC device.
2321 	 */
2322 	list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
2323 		ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2324 		if (ether_addr_equal(vf_netdev->perm_addr, ndev->perm_addr) ||
2325 		    ether_addr_equal(vf_netdev->dev_addr, ndev->perm_addr))
2326 			return ndev;
2327 	}
2328 
2329 	netdev_notice(vf_netdev,
2330 		      "no netdev found for vf serial:%u\n", serial);
2331 	return NULL;
2332 }
2333 
2334 static int netvsc_prepare_bonding(struct net_device *vf_netdev)
2335 {
2336 	struct net_device *ndev;
2337 
2338 	ndev = get_netvsc_byslot(vf_netdev);
2339 	if (!ndev)
2340 		return NOTIFY_DONE;
2341 
2342 	/* set slave flag before open to prevent IPv6 addrconf */
2343 	vf_netdev->flags |= IFF_SLAVE;
2344 	return NOTIFY_DONE;
2345 }
2346 
2347 static int netvsc_register_vf(struct net_device *vf_netdev)
2348 {
2349 	struct net_device_context *net_device_ctx;
2350 	struct netvsc_device *netvsc_dev;
2351 	struct bpf_prog *prog;
2352 	struct net_device *ndev;
2353 	int ret;
2354 
2355 	if (vf_netdev->addr_len != ETH_ALEN)
2356 		return NOTIFY_DONE;
2357 
2358 	ndev = get_netvsc_byslot(vf_netdev);
2359 	if (!ndev)
2360 		return NOTIFY_DONE;
2361 
2362 	net_device_ctx = netdev_priv(ndev);
2363 	netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2364 	if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
2365 		return NOTIFY_DONE;
2366 
2367 	/* if synthetic interface is a different namespace,
2368 	 * then move the VF to that namespace; join will be
2369 	 * done again in that context.
2370 	 */
2371 	if (!net_eq(dev_net(ndev), dev_net(vf_netdev))) {
2372 		ret = dev_change_net_namespace(vf_netdev,
2373 					       dev_net(ndev), "eth%d");
2374 		if (ret)
2375 			netdev_err(vf_netdev,
2376 				   "could not move to same namespace as %s: %d\n",
2377 				   ndev->name, ret);
2378 		else
2379 			netdev_info(vf_netdev,
2380 				    "VF moved to namespace with: %s\n",
2381 				    ndev->name);
2382 		return NOTIFY_DONE;
2383 	}
2384 
2385 	netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
2386 
2387 	if (netvsc_vf_join(vf_netdev, ndev) != 0)
2388 		return NOTIFY_DONE;
2389 
2390 	dev_hold(vf_netdev);
2391 	rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
2392 
2393 	if (ndev->needed_headroom < vf_netdev->needed_headroom)
2394 		ndev->needed_headroom = vf_netdev->needed_headroom;
2395 
2396 	vf_netdev->wanted_features = ndev->features;
2397 	netdev_update_features(vf_netdev);
2398 
2399 	prog = netvsc_xdp_get(netvsc_dev);
2400 	netvsc_vf_setxdp(vf_netdev, prog);
2401 
2402 	return NOTIFY_OK;
2403 }
2404 
2405 /* Change the data path when VF UP/DOWN/CHANGE are detected.
2406  *
2407  * Typically a UP or DOWN event is followed by a CHANGE event, so
2408  * net_device_ctx->data_path_is_vf is used to cache the current data path
2409  * to avoid the duplicate call of netvsc_switch_datapath() and the duplicate
2410  * message.
2411  *
2412  * During hibernation, if a VF NIC driver (e.g. mlx5) preserves the network
2413  * interface, there is only the CHANGE event and no UP or DOWN event.
2414  */
2415 static int netvsc_vf_changed(struct net_device *vf_netdev, unsigned long event)
2416 {
2417 	struct net_device_context *net_device_ctx;
2418 	struct netvsc_device *netvsc_dev;
2419 	struct net_device *ndev;
2420 	bool vf_is_up = false;
2421 	int ret;
2422 
2423 	if (event != NETDEV_GOING_DOWN)
2424 		vf_is_up = netif_running(vf_netdev);
2425 
2426 	ndev = get_netvsc_byref(vf_netdev);
2427 	if (!ndev)
2428 		return NOTIFY_DONE;
2429 
2430 	net_device_ctx = netdev_priv(ndev);
2431 	netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2432 	if (!netvsc_dev)
2433 		return NOTIFY_DONE;
2434 
2435 	if (net_device_ctx->data_path_is_vf == vf_is_up)
2436 		return NOTIFY_OK;
2437 
2438 	if (vf_is_up && !net_device_ctx->vf_alloc) {
2439 		netdev_info(ndev, "Waiting for the VF association from host\n");
2440 		wait_for_completion(&net_device_ctx->vf_add);
2441 	}
2442 
2443 	ret = netvsc_switch_datapath(ndev, vf_is_up);
2444 
2445 	if (ret) {
2446 		netdev_err(ndev,
2447 			   "Data path failed to switch %s VF: %s, err: %d\n",
2448 			   vf_is_up ? "to" : "from", vf_netdev->name, ret);
2449 		return NOTIFY_DONE;
2450 	} else {
2451 		netdev_info(ndev, "Data path switched %s VF: %s\n",
2452 			    vf_is_up ? "to" : "from", vf_netdev->name);
2453 	}
2454 
2455 	return NOTIFY_OK;
2456 }
2457 
2458 static int netvsc_unregister_vf(struct net_device *vf_netdev)
2459 {
2460 	struct net_device *ndev;
2461 	struct net_device_context *net_device_ctx;
2462 
2463 	ndev = get_netvsc_byref(vf_netdev);
2464 	if (!ndev)
2465 		return NOTIFY_DONE;
2466 
2467 	net_device_ctx = netdev_priv(ndev);
2468 	cancel_delayed_work_sync(&net_device_ctx->vf_takeover);
2469 
2470 	netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
2471 
2472 	netvsc_vf_setxdp(vf_netdev, NULL);
2473 
2474 	reinit_completion(&net_device_ctx->vf_add);
2475 	netdev_rx_handler_unregister(vf_netdev);
2476 	netdev_upper_dev_unlink(vf_netdev, ndev);
2477 	RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
2478 	dev_put(vf_netdev);
2479 
2480 	ndev->needed_headroom = RNDIS_AND_PPI_SIZE;
2481 
2482 	return NOTIFY_OK;
2483 }
2484 
2485 static int netvsc_probe(struct hv_device *dev,
2486 			const struct hv_vmbus_device_id *dev_id)
2487 {
2488 	struct net_device *net = NULL;
2489 	struct net_device_context *net_device_ctx;
2490 	struct netvsc_device_info *device_info = NULL;
2491 	struct netvsc_device *nvdev;
2492 	int ret = -ENOMEM;
2493 
2494 	net = alloc_etherdev_mq(sizeof(struct net_device_context),
2495 				VRSS_CHANNEL_MAX);
2496 	if (!net)
2497 		goto no_net;
2498 
2499 	netif_carrier_off(net);
2500 
2501 	netvsc_init_settings(net);
2502 
2503 	net_device_ctx = netdev_priv(net);
2504 	net_device_ctx->device_ctx = dev;
2505 	net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
2506 	if (netif_msg_probe(net_device_ctx))
2507 		netdev_dbg(net, "netvsc msg_enable: %d\n",
2508 			   net_device_ctx->msg_enable);
2509 
2510 	hv_set_drvdata(dev, net);
2511 
2512 	INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
2513 
2514 	init_completion(&net_device_ctx->vf_add);
2515 	spin_lock_init(&net_device_ctx->lock);
2516 	INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
2517 	INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup);
2518 
2519 	net_device_ctx->vf_stats
2520 		= netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats);
2521 	if (!net_device_ctx->vf_stats)
2522 		goto no_stats;
2523 
2524 	net->netdev_ops = &device_ops;
2525 	net->ethtool_ops = &ethtool_ops;
2526 	SET_NETDEV_DEV(net, &dev->device);
2527 	dma_set_min_align_mask(&dev->device, HV_HYP_PAGE_SIZE - 1);
2528 
2529 	/* We always need headroom for rndis header */
2530 	net->needed_headroom = RNDIS_AND_PPI_SIZE;
2531 
2532 	/* Initialize the number of queues to be 1, we may change it if more
2533 	 * channels are offered later.
2534 	 */
2535 	netif_set_real_num_tx_queues(net, 1);
2536 	netif_set_real_num_rx_queues(net, 1);
2537 
2538 	/* Notify the netvsc driver of the new device */
2539 	device_info = netvsc_devinfo_get(NULL);
2540 
2541 	if (!device_info) {
2542 		ret = -ENOMEM;
2543 		goto devinfo_failed;
2544 	}
2545 
2546 	/* We must get rtnl lock before scheduling nvdev->subchan_work,
2547 	 * otherwise netvsc_subchan_work() can get rtnl lock first and wait
2548 	 * all subchannels to show up, but that may not happen because
2549 	 * netvsc_probe() can't get rtnl lock and as a result vmbus_onoffer()
2550 	 * -> ... -> device_add() -> ... -> __device_attach() can't get
2551 	 * the device lock, so all the subchannels can't be processed --
2552 	 * finally netvsc_subchan_work() hangs forever.
2553 	 *
2554 	 * The rtnl lock also needs to be held before rndis_filter_device_add()
2555 	 * which advertises nvsp_2_vsc_capability / sriov bit, and triggers
2556 	 * VF NIC offering and registering. If VF NIC finished register_netdev()
2557 	 * earlier it may cause name based config failure.
2558 	 */
2559 	rtnl_lock();
2560 
2561 	nvdev = rndis_filter_device_add(dev, device_info);
2562 	if (IS_ERR(nvdev)) {
2563 		ret = PTR_ERR(nvdev);
2564 		netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
2565 		goto rndis_failed;
2566 	}
2567 
2568 	eth_hw_addr_set(net, device_info->mac_adr);
2569 
2570 	if (nvdev->num_chn > 1)
2571 		schedule_work(&nvdev->subchan_work);
2572 
2573 	/* hw_features computed in rndis_netdev_set_hwcaps() */
2574 	net->features = net->hw_features |
2575 		NETIF_F_HIGHDMA | NETIF_F_HW_VLAN_CTAG_TX |
2576 		NETIF_F_HW_VLAN_CTAG_RX;
2577 	net->vlan_features = net->features;
2578 
2579 	netdev_lockdep_set_classes(net);
2580 
2581 	net->xdp_features = NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT |
2582 			    NETDEV_XDP_ACT_NDO_XMIT;
2583 
2584 	/* MTU range: 68 - 1500 or 65521 */
2585 	net->min_mtu = NETVSC_MTU_MIN;
2586 	if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
2587 		net->max_mtu = NETVSC_MTU - ETH_HLEN;
2588 	else
2589 		net->max_mtu = ETH_DATA_LEN;
2590 
2591 	nvdev->tx_disable = false;
2592 
2593 	ret = register_netdevice(net);
2594 	if (ret != 0) {
2595 		pr_err("Unable to register netdev.\n");
2596 		goto register_failed;
2597 	}
2598 
2599 	list_add(&net_device_ctx->list, &netvsc_dev_list);
2600 	rtnl_unlock();
2601 
2602 	netvsc_devinfo_put(device_info);
2603 	return 0;
2604 
2605 register_failed:
2606 	rndis_filter_device_remove(dev, nvdev);
2607 rndis_failed:
2608 	rtnl_unlock();
2609 	netvsc_devinfo_put(device_info);
2610 devinfo_failed:
2611 	free_percpu(net_device_ctx->vf_stats);
2612 no_stats:
2613 	hv_set_drvdata(dev, NULL);
2614 	free_netdev(net);
2615 no_net:
2616 	return ret;
2617 }
2618 
2619 static void netvsc_remove(struct hv_device *dev)
2620 {
2621 	struct net_device_context *ndev_ctx;
2622 	struct net_device *vf_netdev, *net;
2623 	struct netvsc_device *nvdev;
2624 
2625 	net = hv_get_drvdata(dev);
2626 	if (net == NULL) {
2627 		dev_err(&dev->device, "No net device to remove\n");
2628 		return;
2629 	}
2630 
2631 	ndev_ctx = netdev_priv(net);
2632 
2633 	cancel_delayed_work_sync(&ndev_ctx->dwork);
2634 
2635 	rtnl_lock();
2636 	nvdev = rtnl_dereference(ndev_ctx->nvdev);
2637 	if (nvdev) {
2638 		cancel_work_sync(&nvdev->subchan_work);
2639 		netvsc_xdp_set(net, NULL, NULL, nvdev);
2640 	}
2641 
2642 	/*
2643 	 * Call to the vsc driver to let it know that the device is being
2644 	 * removed. Also blocks mtu and channel changes.
2645 	 */
2646 	vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2647 	if (vf_netdev)
2648 		netvsc_unregister_vf(vf_netdev);
2649 
2650 	if (nvdev)
2651 		rndis_filter_device_remove(dev, nvdev);
2652 
2653 	unregister_netdevice(net);
2654 	list_del(&ndev_ctx->list);
2655 
2656 	rtnl_unlock();
2657 
2658 	hv_set_drvdata(dev, NULL);
2659 
2660 	free_percpu(ndev_ctx->vf_stats);
2661 	free_netdev(net);
2662 }
2663 
2664 static int netvsc_suspend(struct hv_device *dev)
2665 {
2666 	struct net_device_context *ndev_ctx;
2667 	struct netvsc_device *nvdev;
2668 	struct net_device *net;
2669 	int ret;
2670 
2671 	net = hv_get_drvdata(dev);
2672 
2673 	ndev_ctx = netdev_priv(net);
2674 	cancel_delayed_work_sync(&ndev_ctx->dwork);
2675 
2676 	rtnl_lock();
2677 
2678 	nvdev = rtnl_dereference(ndev_ctx->nvdev);
2679 	if (nvdev == NULL) {
2680 		ret = -ENODEV;
2681 		goto out;
2682 	}
2683 
2684 	/* Save the current config info */
2685 	ndev_ctx->saved_netvsc_dev_info = netvsc_devinfo_get(nvdev);
2686 	if (!ndev_ctx->saved_netvsc_dev_info) {
2687 		ret = -ENOMEM;
2688 		goto out;
2689 	}
2690 	ret = netvsc_detach(net, nvdev);
2691 out:
2692 	rtnl_unlock();
2693 
2694 	return ret;
2695 }
2696 
2697 static int netvsc_resume(struct hv_device *dev)
2698 {
2699 	struct net_device *net = hv_get_drvdata(dev);
2700 	struct net_device_context *net_device_ctx;
2701 	struct netvsc_device_info *device_info;
2702 	int ret;
2703 
2704 	rtnl_lock();
2705 
2706 	net_device_ctx = netdev_priv(net);
2707 
2708 	/* Reset the data path to the netvsc NIC before re-opening the vmbus
2709 	 * channel. Later netvsc_netdev_event() will switch the data path to
2710 	 * the VF upon the UP or CHANGE event.
2711 	 */
2712 	net_device_ctx->data_path_is_vf = false;
2713 	device_info = net_device_ctx->saved_netvsc_dev_info;
2714 
2715 	ret = netvsc_attach(net, device_info);
2716 
2717 	netvsc_devinfo_put(device_info);
2718 	net_device_ctx->saved_netvsc_dev_info = NULL;
2719 
2720 	rtnl_unlock();
2721 
2722 	return ret;
2723 }
2724 static const struct hv_vmbus_device_id id_table[] = {
2725 	/* Network guid */
2726 	{ HV_NIC_GUID, },
2727 	{ },
2728 };
2729 
2730 MODULE_DEVICE_TABLE(vmbus, id_table);
2731 
2732 /* The one and only one */
2733 static struct  hv_driver netvsc_drv = {
2734 	.name = KBUILD_MODNAME,
2735 	.id_table = id_table,
2736 	.probe = netvsc_probe,
2737 	.remove = netvsc_remove,
2738 	.suspend = netvsc_suspend,
2739 	.resume = netvsc_resume,
2740 	.driver = {
2741 		.probe_type = PROBE_FORCE_SYNCHRONOUS,
2742 	},
2743 };
2744 
2745 /*
2746  * On Hyper-V, every VF interface is matched with a corresponding
2747  * synthetic interface. The synthetic interface is presented first
2748  * to the guest. When the corresponding VF instance is registered,
2749  * we will take care of switching the data path.
2750  */
2751 static int netvsc_netdev_event(struct notifier_block *this,
2752 			       unsigned long event, void *ptr)
2753 {
2754 	struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2755 
2756 	/* Skip our own events */
2757 	if (event_dev->netdev_ops == &device_ops)
2758 		return NOTIFY_DONE;
2759 
2760 	/* Avoid non-Ethernet type devices */
2761 	if (event_dev->type != ARPHRD_ETHER)
2762 		return NOTIFY_DONE;
2763 
2764 	/* Avoid Vlan dev with same MAC registering as VF */
2765 	if (is_vlan_dev(event_dev))
2766 		return NOTIFY_DONE;
2767 
2768 	/* Avoid Bonding master dev with same MAC registering as VF */
2769 	if (netif_is_bond_master(event_dev))
2770 		return NOTIFY_DONE;
2771 
2772 	switch (event) {
2773 	case NETDEV_POST_INIT:
2774 		return netvsc_prepare_bonding(event_dev);
2775 	case NETDEV_REGISTER:
2776 		return netvsc_register_vf(event_dev);
2777 	case NETDEV_UNREGISTER:
2778 		return netvsc_unregister_vf(event_dev);
2779 	case NETDEV_UP:
2780 	case NETDEV_DOWN:
2781 	case NETDEV_CHANGE:
2782 	case NETDEV_GOING_DOWN:
2783 		return netvsc_vf_changed(event_dev, event);
2784 	default:
2785 		return NOTIFY_DONE;
2786 	}
2787 }
2788 
2789 static struct notifier_block netvsc_netdev_notifier = {
2790 	.notifier_call = netvsc_netdev_event,
2791 };
2792 
2793 static void __exit netvsc_drv_exit(void)
2794 {
2795 	unregister_netdevice_notifier(&netvsc_netdev_notifier);
2796 	vmbus_driver_unregister(&netvsc_drv);
2797 }
2798 
2799 static int __init netvsc_drv_init(void)
2800 {
2801 	int ret;
2802 
2803 	if (ring_size < RING_SIZE_MIN) {
2804 		ring_size = RING_SIZE_MIN;
2805 		pr_info("Increased ring_size to %u (min allowed)\n",
2806 			ring_size);
2807 	}
2808 	netvsc_ring_bytes = VMBUS_RING_SIZE(ring_size * 4096);
2809 
2810 	register_netdevice_notifier(&netvsc_netdev_notifier);
2811 
2812 	ret = vmbus_driver_register(&netvsc_drv);
2813 	if (ret)
2814 		goto err_vmbus_reg;
2815 
2816 	return 0;
2817 
2818 err_vmbus_reg:
2819 	unregister_netdevice_notifier(&netvsc_netdev_notifier);
2820 	return ret;
2821 }
2822 
2823 MODULE_LICENSE("GPL");
2824 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
2825 
2826 module_init(netvsc_drv_init);
2827 module_exit(netvsc_drv_exit);
2828