xref: /openbmc/linux/drivers/net/hyperv/netvsc_drv.c (revision 2cf1c348)
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 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 /* Azure hosts don't support non-TCP port numbers in hashing for fragmented
246  * packets. We can use ethtool to change UDP hash level when necessary.
247  */
248 static inline u32 netvsc_get_hash(
249 	struct sk_buff *skb,
250 	const struct net_device_context *ndc)
251 {
252 	struct flow_keys flow;
253 	u32 hash, pkt_proto = 0;
254 	static u32 hashrnd __read_mostly;
255 
256 	net_get_random_once(&hashrnd, sizeof(hashrnd));
257 
258 	if (!skb_flow_dissect_flow_keys(skb, &flow, 0))
259 		return 0;
260 
261 	switch (flow.basic.ip_proto) {
262 	case IPPROTO_TCP:
263 		if (flow.basic.n_proto == htons(ETH_P_IP))
264 			pkt_proto = HV_TCP4_L4HASH;
265 		else if (flow.basic.n_proto == htons(ETH_P_IPV6))
266 			pkt_proto = HV_TCP6_L4HASH;
267 
268 		break;
269 
270 	case IPPROTO_UDP:
271 		if (flow.basic.n_proto == htons(ETH_P_IP))
272 			pkt_proto = HV_UDP4_L4HASH;
273 		else if (flow.basic.n_proto == htons(ETH_P_IPV6))
274 			pkt_proto = HV_UDP6_L4HASH;
275 
276 		break;
277 	}
278 
279 	if (pkt_proto & ndc->l4_hash) {
280 		return skb_get_hash(skb);
281 	} else {
282 		if (flow.basic.n_proto == htons(ETH_P_IP))
283 			hash = jhash2((u32 *)&flow.addrs.v4addrs, 2, hashrnd);
284 		else if (flow.basic.n_proto == htons(ETH_P_IPV6))
285 			hash = jhash2((u32 *)&flow.addrs.v6addrs, 8, hashrnd);
286 		else
287 			return 0;
288 
289 		__skb_set_sw_hash(skb, hash, false);
290 	}
291 
292 	return hash;
293 }
294 
295 static inline int netvsc_get_tx_queue(struct net_device *ndev,
296 				      struct sk_buff *skb, int old_idx)
297 {
298 	const struct net_device_context *ndc = netdev_priv(ndev);
299 	struct sock *sk = skb->sk;
300 	int q_idx;
301 
302 	q_idx = ndc->tx_table[netvsc_get_hash(skb, ndc) &
303 			      (VRSS_SEND_TAB_SIZE - 1)];
304 
305 	/* If queue index changed record the new value */
306 	if (q_idx != old_idx &&
307 	    sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
308 		sk_tx_queue_set(sk, q_idx);
309 
310 	return q_idx;
311 }
312 
313 /*
314  * Select queue for transmit.
315  *
316  * If a valid queue has already been assigned, then use that.
317  * Otherwise compute tx queue based on hash and the send table.
318  *
319  * This is basically similar to default (netdev_pick_tx) with the added step
320  * of using the host send_table when no other queue has been assigned.
321  *
322  * TODO support XPS - but get_xps_queue not exported
323  */
324 static u16 netvsc_pick_tx(struct net_device *ndev, struct sk_buff *skb)
325 {
326 	int q_idx = sk_tx_queue_get(skb->sk);
327 
328 	if (q_idx < 0 || skb->ooo_okay || q_idx >= ndev->real_num_tx_queues) {
329 		/* If forwarding a packet, we use the recorded queue when
330 		 * available for better cache locality.
331 		 */
332 		if (skb_rx_queue_recorded(skb))
333 			q_idx = skb_get_rx_queue(skb);
334 		else
335 			q_idx = netvsc_get_tx_queue(ndev, skb, q_idx);
336 	}
337 
338 	return q_idx;
339 }
340 
341 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
342 			       struct net_device *sb_dev)
343 {
344 	struct net_device_context *ndc = netdev_priv(ndev);
345 	struct net_device *vf_netdev;
346 	u16 txq;
347 
348 	rcu_read_lock();
349 	vf_netdev = rcu_dereference(ndc->vf_netdev);
350 	if (vf_netdev) {
351 		const struct net_device_ops *vf_ops = vf_netdev->netdev_ops;
352 
353 		if (vf_ops->ndo_select_queue)
354 			txq = vf_ops->ndo_select_queue(vf_netdev, skb, sb_dev);
355 		else
356 			txq = netdev_pick_tx(vf_netdev, skb, NULL);
357 
358 		/* Record the queue selected by VF so that it can be
359 		 * used for common case where VF has more queues than
360 		 * the synthetic device.
361 		 */
362 		qdisc_skb_cb(skb)->slave_dev_queue_mapping = txq;
363 	} else {
364 		txq = netvsc_pick_tx(ndev, skb);
365 	}
366 	rcu_read_unlock();
367 
368 	while (txq >= ndev->real_num_tx_queues)
369 		txq -= ndev->real_num_tx_queues;
370 
371 	return txq;
372 }
373 
374 static u32 fill_pg_buf(unsigned long hvpfn, u32 offset, u32 len,
375 		       struct hv_page_buffer *pb)
376 {
377 	int j = 0;
378 
379 	hvpfn += offset >> HV_HYP_PAGE_SHIFT;
380 	offset = offset & ~HV_HYP_PAGE_MASK;
381 
382 	while (len > 0) {
383 		unsigned long bytes;
384 
385 		bytes = HV_HYP_PAGE_SIZE - offset;
386 		if (bytes > len)
387 			bytes = len;
388 		pb[j].pfn = hvpfn;
389 		pb[j].offset = offset;
390 		pb[j].len = bytes;
391 
392 		offset += bytes;
393 		len -= bytes;
394 
395 		if (offset == HV_HYP_PAGE_SIZE && len) {
396 			hvpfn++;
397 			offset = 0;
398 			j++;
399 		}
400 	}
401 
402 	return j + 1;
403 }
404 
405 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
406 			   struct hv_netvsc_packet *packet,
407 			   struct hv_page_buffer *pb)
408 {
409 	u32 slots_used = 0;
410 	char *data = skb->data;
411 	int frags = skb_shinfo(skb)->nr_frags;
412 	int i;
413 
414 	/* The packet is laid out thus:
415 	 * 1. hdr: RNDIS header and PPI
416 	 * 2. skb linear data
417 	 * 3. skb fragment data
418 	 */
419 	slots_used += fill_pg_buf(virt_to_hvpfn(hdr),
420 				  offset_in_hvpage(hdr),
421 				  len,
422 				  &pb[slots_used]);
423 
424 	packet->rmsg_size = len;
425 	packet->rmsg_pgcnt = slots_used;
426 
427 	slots_used += fill_pg_buf(virt_to_hvpfn(data),
428 				  offset_in_hvpage(data),
429 				  skb_headlen(skb),
430 				  &pb[slots_used]);
431 
432 	for (i = 0; i < frags; i++) {
433 		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
434 
435 		slots_used += fill_pg_buf(page_to_hvpfn(skb_frag_page(frag)),
436 					  skb_frag_off(frag),
437 					  skb_frag_size(frag),
438 					  &pb[slots_used]);
439 	}
440 	return slots_used;
441 }
442 
443 static int count_skb_frag_slots(struct sk_buff *skb)
444 {
445 	int i, frags = skb_shinfo(skb)->nr_frags;
446 	int pages = 0;
447 
448 	for (i = 0; i < frags; i++) {
449 		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
450 		unsigned long size = skb_frag_size(frag);
451 		unsigned long offset = skb_frag_off(frag);
452 
453 		/* Skip unused frames from start of page */
454 		offset &= ~HV_HYP_PAGE_MASK;
455 		pages += HVPFN_UP(offset + size);
456 	}
457 	return pages;
458 }
459 
460 static int netvsc_get_slots(struct sk_buff *skb)
461 {
462 	char *data = skb->data;
463 	unsigned int offset = offset_in_hvpage(data);
464 	unsigned int len = skb_headlen(skb);
465 	int slots;
466 	int frag_slots;
467 
468 	slots = DIV_ROUND_UP(offset + len, HV_HYP_PAGE_SIZE);
469 	frag_slots = count_skb_frag_slots(skb);
470 	return slots + frag_slots;
471 }
472 
473 static u32 net_checksum_info(struct sk_buff *skb)
474 {
475 	if (skb->protocol == htons(ETH_P_IP)) {
476 		struct iphdr *ip = ip_hdr(skb);
477 
478 		if (ip->protocol == IPPROTO_TCP)
479 			return TRANSPORT_INFO_IPV4_TCP;
480 		else if (ip->protocol == IPPROTO_UDP)
481 			return TRANSPORT_INFO_IPV4_UDP;
482 	} else {
483 		struct ipv6hdr *ip6 = ipv6_hdr(skb);
484 
485 		if (ip6->nexthdr == IPPROTO_TCP)
486 			return TRANSPORT_INFO_IPV6_TCP;
487 		else if (ip6->nexthdr == IPPROTO_UDP)
488 			return TRANSPORT_INFO_IPV6_UDP;
489 	}
490 
491 	return TRANSPORT_INFO_NOT_IP;
492 }
493 
494 /* Send skb on the slave VF device. */
495 static int netvsc_vf_xmit(struct net_device *net, struct net_device *vf_netdev,
496 			  struct sk_buff *skb)
497 {
498 	struct net_device_context *ndev_ctx = netdev_priv(net);
499 	unsigned int len = skb->len;
500 	int rc;
501 
502 	skb->dev = vf_netdev;
503 	skb_record_rx_queue(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
504 
505 	rc = dev_queue_xmit(skb);
506 	if (likely(rc == NET_XMIT_SUCCESS || rc == NET_XMIT_CN)) {
507 		struct netvsc_vf_pcpu_stats *pcpu_stats
508 			= this_cpu_ptr(ndev_ctx->vf_stats);
509 
510 		u64_stats_update_begin(&pcpu_stats->syncp);
511 		pcpu_stats->tx_packets++;
512 		pcpu_stats->tx_bytes += len;
513 		u64_stats_update_end(&pcpu_stats->syncp);
514 	} else {
515 		this_cpu_inc(ndev_ctx->vf_stats->tx_dropped);
516 	}
517 
518 	return rc;
519 }
520 
521 static int netvsc_xmit(struct sk_buff *skb, struct net_device *net, bool xdp_tx)
522 {
523 	struct net_device_context *net_device_ctx = netdev_priv(net);
524 	struct hv_netvsc_packet *packet = NULL;
525 	int ret;
526 	unsigned int num_data_pgs;
527 	struct rndis_message *rndis_msg;
528 	struct net_device *vf_netdev;
529 	u32 rndis_msg_size;
530 	u32 hash;
531 	struct hv_page_buffer pb[MAX_PAGE_BUFFER_COUNT];
532 
533 	/* If VF is present and up then redirect packets to it.
534 	 * Skip the VF if it is marked down or has no carrier.
535 	 * If netpoll is in uses, then VF can not be used either.
536 	 */
537 	vf_netdev = rcu_dereference_bh(net_device_ctx->vf_netdev);
538 	if (vf_netdev && netif_running(vf_netdev) &&
539 	    netif_carrier_ok(vf_netdev) && !netpoll_tx_running(net) &&
540 	    net_device_ctx->data_path_is_vf)
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 	/* When using AF_PACKET we need to drop VLAN header from
609 	 * the frame and update the SKB to allow the HOST OS
610 	 * to transmit the 802.1Q packet
611 	 */
612 	if (skb->protocol == htons(ETH_P_8021Q)) {
613 		u16 vlan_tci;
614 
615 		skb_reset_mac_header(skb);
616 		if (eth_type_vlan(eth_hdr(skb)->h_proto)) {
617 			if (unlikely(__skb_vlan_pop(skb, &vlan_tci) != 0)) {
618 				++net_device_ctx->eth_stats.vlan_error;
619 				goto drop;
620 			}
621 
622 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
623 			/* Update the NDIS header pkt lengths */
624 			packet->total_data_buflen -= VLAN_HLEN;
625 			packet->total_bytes -= VLAN_HLEN;
626 			rndis_msg->msg_len = packet->total_data_buflen;
627 			rndis_msg->msg.pkt.data_len = packet->total_data_buflen;
628 		}
629 	}
630 
631 	if (skb_vlan_tag_present(skb)) {
632 		struct ndis_pkt_8021q_info *vlan;
633 
634 		rndis_msg_size += NDIS_VLAN_PPI_SIZE;
635 		vlan = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
636 				     IEEE_8021Q_INFO);
637 
638 		vlan->value = 0;
639 		vlan->vlanid = skb_vlan_tag_get_id(skb);
640 		vlan->cfi = skb_vlan_tag_get_cfi(skb);
641 		vlan->pri = skb_vlan_tag_get_prio(skb);
642 	}
643 
644 	if (skb_is_gso(skb)) {
645 		struct ndis_tcp_lso_info *lso_info;
646 
647 		rndis_msg_size += NDIS_LSO_PPI_SIZE;
648 		lso_info = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
649 					 TCP_LARGESEND_PKTINFO);
650 
651 		lso_info->value = 0;
652 		lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
653 		if (skb->protocol == htons(ETH_P_IP)) {
654 			lso_info->lso_v2_transmit.ip_version =
655 				NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
656 			ip_hdr(skb)->tot_len = 0;
657 			ip_hdr(skb)->check = 0;
658 			tcp_hdr(skb)->check =
659 				~csum_tcpudp_magic(ip_hdr(skb)->saddr,
660 						   ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
661 		} else {
662 			lso_info->lso_v2_transmit.ip_version =
663 				NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
664 			tcp_v6_gso_csum_prep(skb);
665 		}
666 		lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
667 		lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
668 	} else if (skb->ip_summed == CHECKSUM_PARTIAL) {
669 		if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
670 			struct ndis_tcp_ip_checksum_info *csum_info;
671 
672 			rndis_msg_size += NDIS_CSUM_PPI_SIZE;
673 			csum_info = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
674 						  TCPIP_CHKSUM_PKTINFO);
675 
676 			csum_info->value = 0;
677 			csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
678 
679 			if (skb->protocol == htons(ETH_P_IP)) {
680 				csum_info->transmit.is_ipv4 = 1;
681 
682 				if (ip_hdr(skb)->protocol == IPPROTO_TCP)
683 					csum_info->transmit.tcp_checksum = 1;
684 				else
685 					csum_info->transmit.udp_checksum = 1;
686 			} else {
687 				csum_info->transmit.is_ipv6 = 1;
688 
689 				if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
690 					csum_info->transmit.tcp_checksum = 1;
691 				else
692 					csum_info->transmit.udp_checksum = 1;
693 			}
694 		} else {
695 			/* Can't do offload of this type of checksum */
696 			if (skb_checksum_help(skb))
697 				goto drop;
698 		}
699 	}
700 
701 	/* Start filling in the page buffers with the rndis hdr */
702 	rndis_msg->msg_len += rndis_msg_size;
703 	packet->total_data_buflen = rndis_msg->msg_len;
704 	packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
705 					       skb, packet, pb);
706 
707 	/* timestamp packet in software */
708 	skb_tx_timestamp(skb);
709 
710 	ret = netvsc_send(net, packet, rndis_msg, pb, skb, xdp_tx);
711 	if (likely(ret == 0))
712 		return NETDEV_TX_OK;
713 
714 	if (ret == -EAGAIN) {
715 		++net_device_ctx->eth_stats.tx_busy;
716 		return NETDEV_TX_BUSY;
717 	}
718 
719 	if (ret == -ENOSPC)
720 		++net_device_ctx->eth_stats.tx_no_space;
721 
722 drop:
723 	dev_kfree_skb_any(skb);
724 	net->stats.tx_dropped++;
725 
726 	return NETDEV_TX_OK;
727 
728 no_memory:
729 	++net_device_ctx->eth_stats.tx_no_memory;
730 	goto drop;
731 }
732 
733 static netdev_tx_t netvsc_start_xmit(struct sk_buff *skb,
734 				     struct net_device *ndev)
735 {
736 	return netvsc_xmit(skb, ndev, false);
737 }
738 
739 /*
740  * netvsc_linkstatus_callback - Link up/down notification
741  */
742 void netvsc_linkstatus_callback(struct net_device *net,
743 				struct rndis_message *resp,
744 				void *data, u32 data_buflen)
745 {
746 	struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
747 	struct net_device_context *ndev_ctx = netdev_priv(net);
748 	struct netvsc_reconfig *event;
749 	unsigned long flags;
750 
751 	/* Ensure the packet is big enough to access its fields */
752 	if (resp->msg_len - RNDIS_HEADER_SIZE < sizeof(struct rndis_indicate_status)) {
753 		netdev_err(net, "invalid rndis_indicate_status packet, len: %u\n",
754 			   resp->msg_len);
755 		return;
756 	}
757 
758 	/* Copy the RNDIS indicate status into nvchan->recv_buf */
759 	memcpy(indicate, data + RNDIS_HEADER_SIZE, sizeof(*indicate));
760 
761 	/* Update the physical link speed when changing to another vSwitch */
762 	if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
763 		u32 speed;
764 
765 		/* Validate status_buf_offset and status_buflen.
766 		 *
767 		 * Certain (pre-Fe) implementations of Hyper-V's vSwitch didn't account
768 		 * for the status buffer field in resp->msg_len; perform the validation
769 		 * using data_buflen (>= resp->msg_len).
770 		 */
771 		if (indicate->status_buflen < sizeof(speed) ||
772 		    indicate->status_buf_offset < sizeof(*indicate) ||
773 		    data_buflen - RNDIS_HEADER_SIZE < indicate->status_buf_offset ||
774 		    data_buflen - RNDIS_HEADER_SIZE - indicate->status_buf_offset
775 				< indicate->status_buflen) {
776 			netdev_err(net, "invalid rndis_indicate_status packet\n");
777 			return;
778 		}
779 
780 		speed = *(u32 *)(data + RNDIS_HEADER_SIZE + indicate->status_buf_offset) / 10000;
781 		ndev_ctx->speed = speed;
782 		return;
783 	}
784 
785 	/* Handle these link change statuses below */
786 	if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
787 	    indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
788 	    indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
789 		return;
790 
791 	if (net->reg_state != NETREG_REGISTERED)
792 		return;
793 
794 	event = kzalloc(sizeof(*event), GFP_ATOMIC);
795 	if (!event)
796 		return;
797 	event->event = indicate->status;
798 
799 	spin_lock_irqsave(&ndev_ctx->lock, flags);
800 	list_add_tail(&event->list, &ndev_ctx->reconfig_events);
801 	spin_unlock_irqrestore(&ndev_ctx->lock, flags);
802 
803 	schedule_delayed_work(&ndev_ctx->dwork, 0);
804 }
805 
806 /* This function should only be called after skb_record_rx_queue() */
807 static void netvsc_xdp_xmit(struct sk_buff *skb, struct net_device *ndev)
808 {
809 	int rc;
810 
811 	skb->queue_mapping = skb_get_rx_queue(skb);
812 	__skb_push(skb, ETH_HLEN);
813 
814 	rc = netvsc_xmit(skb, ndev, true);
815 
816 	if (dev_xmit_complete(rc))
817 		return;
818 
819 	dev_kfree_skb_any(skb);
820 	ndev->stats.tx_dropped++;
821 }
822 
823 static void netvsc_comp_ipcsum(struct sk_buff *skb)
824 {
825 	struct iphdr *iph = (struct iphdr *)skb->data;
826 
827 	iph->check = 0;
828 	iph->check = ip_fast_csum(iph, iph->ihl);
829 }
830 
831 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
832 					     struct netvsc_channel *nvchan,
833 					     struct xdp_buff *xdp)
834 {
835 	struct napi_struct *napi = &nvchan->napi;
836 	const struct ndis_pkt_8021q_info *vlan = &nvchan->rsc.vlan;
837 	const struct ndis_tcp_ip_checksum_info *csum_info =
838 						&nvchan->rsc.csum_info;
839 	const u32 *hash_info = &nvchan->rsc.hash_info;
840 	u8 ppi_flags = nvchan->rsc.ppi_flags;
841 	struct sk_buff *skb;
842 	void *xbuf = xdp->data_hard_start;
843 	int i;
844 
845 	if (xbuf) {
846 		unsigned int hdroom = xdp->data - xdp->data_hard_start;
847 		unsigned int xlen = xdp->data_end - xdp->data;
848 		unsigned int frag_size = xdp->frame_sz;
849 
850 		skb = build_skb(xbuf, frag_size);
851 
852 		if (!skb) {
853 			__free_page(virt_to_page(xbuf));
854 			return NULL;
855 		}
856 
857 		skb_reserve(skb, hdroom);
858 		skb_put(skb, xlen);
859 		skb->dev = napi->dev;
860 	} else {
861 		skb = napi_alloc_skb(napi, nvchan->rsc.pktlen);
862 
863 		if (!skb)
864 			return NULL;
865 
866 		/* Copy to skb. This copy is needed here since the memory
867 		 * pointed by hv_netvsc_packet cannot be deallocated.
868 		 */
869 		for (i = 0; i < nvchan->rsc.cnt; i++)
870 			skb_put_data(skb, nvchan->rsc.data[i],
871 				     nvchan->rsc.len[i]);
872 	}
873 
874 	skb->protocol = eth_type_trans(skb, net);
875 
876 	/* skb is already created with CHECKSUM_NONE */
877 	skb_checksum_none_assert(skb);
878 
879 	/* Incoming packets may have IP header checksum verified by the host.
880 	 * They may not have IP header checksum computed after coalescing.
881 	 * We compute it here if the flags are set, because on Linux, the IP
882 	 * checksum is always checked.
883 	 */
884 	if ((ppi_flags & NVSC_RSC_CSUM_INFO) && csum_info->receive.ip_checksum_value_invalid &&
885 	    csum_info->receive.ip_checksum_succeeded &&
886 	    skb->protocol == htons(ETH_P_IP)) {
887 		/* Check that there is enough space to hold the IP header. */
888 		if (skb_headlen(skb) < sizeof(struct iphdr)) {
889 			kfree_skb(skb);
890 			return NULL;
891 		}
892 		netvsc_comp_ipcsum(skb);
893 	}
894 
895 	/* Do L4 checksum offload if enabled and present. */
896 	if ((ppi_flags & NVSC_RSC_CSUM_INFO) && (net->features & NETIF_F_RXCSUM)) {
897 		if (csum_info->receive.tcp_checksum_succeeded ||
898 		    csum_info->receive.udp_checksum_succeeded)
899 			skb->ip_summed = CHECKSUM_UNNECESSARY;
900 	}
901 
902 	if ((ppi_flags & NVSC_RSC_HASH_INFO) && (net->features & NETIF_F_RXHASH))
903 		skb_set_hash(skb, *hash_info, PKT_HASH_TYPE_L4);
904 
905 	if (ppi_flags & NVSC_RSC_VLAN) {
906 		u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT) |
907 			(vlan->cfi ? VLAN_CFI_MASK : 0);
908 
909 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
910 				       vlan_tci);
911 	}
912 
913 	return skb;
914 }
915 
916 /*
917  * netvsc_recv_callback -  Callback when we receive a packet from the
918  * "wire" on the specified device.
919  */
920 int netvsc_recv_callback(struct net_device *net,
921 			 struct netvsc_device *net_device,
922 			 struct netvsc_channel *nvchan)
923 {
924 	struct net_device_context *net_device_ctx = netdev_priv(net);
925 	struct vmbus_channel *channel = nvchan->channel;
926 	u16 q_idx = channel->offermsg.offer.sub_channel_index;
927 	struct sk_buff *skb;
928 	struct netvsc_stats *rx_stats = &nvchan->rx_stats;
929 	struct xdp_buff xdp;
930 	u32 act;
931 
932 	if (net->reg_state != NETREG_REGISTERED)
933 		return NVSP_STAT_FAIL;
934 
935 	act = netvsc_run_xdp(net, nvchan, &xdp);
936 
937 	if (act != XDP_PASS && act != XDP_TX) {
938 		u64_stats_update_begin(&rx_stats->syncp);
939 		rx_stats->xdp_drop++;
940 		u64_stats_update_end(&rx_stats->syncp);
941 
942 		return NVSP_STAT_SUCCESS; /* consumed by XDP */
943 	}
944 
945 	/* Allocate a skb - TODO direct I/O to pages? */
946 	skb = netvsc_alloc_recv_skb(net, nvchan, &xdp);
947 
948 	if (unlikely(!skb)) {
949 		++net_device_ctx->eth_stats.rx_no_memory;
950 		return NVSP_STAT_FAIL;
951 	}
952 
953 	skb_record_rx_queue(skb, q_idx);
954 
955 	/*
956 	 * Even if injecting the packet, record the statistics
957 	 * on the synthetic device because modifying the VF device
958 	 * statistics will not work correctly.
959 	 */
960 	u64_stats_update_begin(&rx_stats->syncp);
961 	rx_stats->packets++;
962 	rx_stats->bytes += nvchan->rsc.pktlen;
963 
964 	if (skb->pkt_type == PACKET_BROADCAST)
965 		++rx_stats->broadcast;
966 	else if (skb->pkt_type == PACKET_MULTICAST)
967 		++rx_stats->multicast;
968 	u64_stats_update_end(&rx_stats->syncp);
969 
970 	if (act == XDP_TX) {
971 		netvsc_xdp_xmit(skb, net);
972 		return NVSP_STAT_SUCCESS;
973 	}
974 
975 	napi_gro_receive(&nvchan->napi, skb);
976 	return NVSP_STAT_SUCCESS;
977 }
978 
979 static void netvsc_get_drvinfo(struct net_device *net,
980 			       struct ethtool_drvinfo *info)
981 {
982 	strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
983 	strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
984 }
985 
986 static void netvsc_get_channels(struct net_device *net,
987 				struct ethtool_channels *channel)
988 {
989 	struct net_device_context *net_device_ctx = netdev_priv(net);
990 	struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
991 
992 	if (nvdev) {
993 		channel->max_combined	= nvdev->max_chn;
994 		channel->combined_count = nvdev->num_chn;
995 	}
996 }
997 
998 /* Alloc struct netvsc_device_info, and initialize it from either existing
999  * struct netvsc_device, or from default values.
1000  */
1001 static
1002 struct netvsc_device_info *netvsc_devinfo_get(struct netvsc_device *nvdev)
1003 {
1004 	struct netvsc_device_info *dev_info;
1005 	struct bpf_prog *prog;
1006 
1007 	dev_info = kzalloc(sizeof(*dev_info), GFP_ATOMIC);
1008 
1009 	if (!dev_info)
1010 		return NULL;
1011 
1012 	if (nvdev) {
1013 		ASSERT_RTNL();
1014 
1015 		dev_info->num_chn = nvdev->num_chn;
1016 		dev_info->send_sections = nvdev->send_section_cnt;
1017 		dev_info->send_section_size = nvdev->send_section_size;
1018 		dev_info->recv_sections = nvdev->recv_section_cnt;
1019 		dev_info->recv_section_size = nvdev->recv_section_size;
1020 
1021 		memcpy(dev_info->rss_key, nvdev->extension->rss_key,
1022 		       NETVSC_HASH_KEYLEN);
1023 
1024 		prog = netvsc_xdp_get(nvdev);
1025 		if (prog) {
1026 			bpf_prog_inc(prog);
1027 			dev_info->bprog = prog;
1028 		}
1029 	} else {
1030 		dev_info->num_chn = VRSS_CHANNEL_DEFAULT;
1031 		dev_info->send_sections = NETVSC_DEFAULT_TX;
1032 		dev_info->send_section_size = NETVSC_SEND_SECTION_SIZE;
1033 		dev_info->recv_sections = NETVSC_DEFAULT_RX;
1034 		dev_info->recv_section_size = NETVSC_RECV_SECTION_SIZE;
1035 	}
1036 
1037 	return dev_info;
1038 }
1039 
1040 /* Free struct netvsc_device_info */
1041 static void netvsc_devinfo_put(struct netvsc_device_info *dev_info)
1042 {
1043 	if (dev_info->bprog) {
1044 		ASSERT_RTNL();
1045 		bpf_prog_put(dev_info->bprog);
1046 	}
1047 
1048 	kfree(dev_info);
1049 }
1050 
1051 static int netvsc_detach(struct net_device *ndev,
1052 			 struct netvsc_device *nvdev)
1053 {
1054 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
1055 	struct hv_device *hdev = ndev_ctx->device_ctx;
1056 	int ret;
1057 
1058 	/* Don't try continuing to try and setup sub channels */
1059 	if (cancel_work_sync(&nvdev->subchan_work))
1060 		nvdev->num_chn = 1;
1061 
1062 	netvsc_xdp_set(ndev, NULL, NULL, nvdev);
1063 
1064 	/* If device was up (receiving) then shutdown */
1065 	if (netif_running(ndev)) {
1066 		netvsc_tx_disable(nvdev, ndev);
1067 
1068 		ret = rndis_filter_close(nvdev);
1069 		if (ret) {
1070 			netdev_err(ndev,
1071 				   "unable to close device (ret %d).\n", ret);
1072 			return ret;
1073 		}
1074 
1075 		ret = netvsc_wait_until_empty(nvdev);
1076 		if (ret) {
1077 			netdev_err(ndev,
1078 				   "Ring buffer not empty after closing rndis\n");
1079 			return ret;
1080 		}
1081 	}
1082 
1083 	netif_device_detach(ndev);
1084 
1085 	rndis_filter_device_remove(hdev, nvdev);
1086 
1087 	return 0;
1088 }
1089 
1090 static int netvsc_attach(struct net_device *ndev,
1091 			 struct netvsc_device_info *dev_info)
1092 {
1093 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
1094 	struct hv_device *hdev = ndev_ctx->device_ctx;
1095 	struct netvsc_device *nvdev;
1096 	struct rndis_device *rdev;
1097 	struct bpf_prog *prog;
1098 	int ret = 0;
1099 
1100 	nvdev = rndis_filter_device_add(hdev, dev_info);
1101 	if (IS_ERR(nvdev))
1102 		return PTR_ERR(nvdev);
1103 
1104 	if (nvdev->num_chn > 1) {
1105 		ret = rndis_set_subchannel(ndev, nvdev, dev_info);
1106 
1107 		/* if unavailable, just proceed with one queue */
1108 		if (ret) {
1109 			nvdev->max_chn = 1;
1110 			nvdev->num_chn = 1;
1111 		}
1112 	}
1113 
1114 	prog = dev_info->bprog;
1115 	if (prog) {
1116 		bpf_prog_inc(prog);
1117 		ret = netvsc_xdp_set(ndev, prog, NULL, nvdev);
1118 		if (ret) {
1119 			bpf_prog_put(prog);
1120 			goto err1;
1121 		}
1122 	}
1123 
1124 	/* In any case device is now ready */
1125 	nvdev->tx_disable = false;
1126 	netif_device_attach(ndev);
1127 
1128 	/* Note: enable and attach happen when sub-channels setup */
1129 	netif_carrier_off(ndev);
1130 
1131 	if (netif_running(ndev)) {
1132 		ret = rndis_filter_open(nvdev);
1133 		if (ret)
1134 			goto err2;
1135 
1136 		rdev = nvdev->extension;
1137 		if (!rdev->link_state)
1138 			netif_carrier_on(ndev);
1139 	}
1140 
1141 	return 0;
1142 
1143 err2:
1144 	netif_device_detach(ndev);
1145 
1146 err1:
1147 	rndis_filter_device_remove(hdev, nvdev);
1148 
1149 	return ret;
1150 }
1151 
1152 static int netvsc_set_channels(struct net_device *net,
1153 			       struct ethtool_channels *channels)
1154 {
1155 	struct net_device_context *net_device_ctx = netdev_priv(net);
1156 	struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
1157 	unsigned int orig, count = channels->combined_count;
1158 	struct netvsc_device_info *device_info;
1159 	int ret;
1160 
1161 	/* We do not support separate count for rx, tx, or other */
1162 	if (count == 0 ||
1163 	    channels->rx_count || channels->tx_count || channels->other_count)
1164 		return -EINVAL;
1165 
1166 	if (!nvdev || nvdev->destroy)
1167 		return -ENODEV;
1168 
1169 	if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
1170 		return -EINVAL;
1171 
1172 	if (count > nvdev->max_chn)
1173 		return -EINVAL;
1174 
1175 	orig = nvdev->num_chn;
1176 
1177 	device_info = netvsc_devinfo_get(nvdev);
1178 
1179 	if (!device_info)
1180 		return -ENOMEM;
1181 
1182 	device_info->num_chn = count;
1183 
1184 	ret = netvsc_detach(net, nvdev);
1185 	if (ret)
1186 		goto out;
1187 
1188 	ret = netvsc_attach(net, device_info);
1189 	if (ret) {
1190 		device_info->num_chn = orig;
1191 		if (netvsc_attach(net, device_info))
1192 			netdev_err(net, "restoring channel setting failed\n");
1193 	}
1194 
1195 out:
1196 	netvsc_devinfo_put(device_info);
1197 	return ret;
1198 }
1199 
1200 static void netvsc_init_settings(struct net_device *dev)
1201 {
1202 	struct net_device_context *ndc = netdev_priv(dev);
1203 
1204 	ndc->l4_hash = HV_DEFAULT_L4HASH;
1205 
1206 	ndc->speed = SPEED_UNKNOWN;
1207 	ndc->duplex = DUPLEX_FULL;
1208 
1209 	dev->features = NETIF_F_LRO;
1210 }
1211 
1212 static int netvsc_get_link_ksettings(struct net_device *dev,
1213 				     struct ethtool_link_ksettings *cmd)
1214 {
1215 	struct net_device_context *ndc = netdev_priv(dev);
1216 	struct net_device *vf_netdev;
1217 
1218 	vf_netdev = rtnl_dereference(ndc->vf_netdev);
1219 
1220 	if (vf_netdev)
1221 		return __ethtool_get_link_ksettings(vf_netdev, cmd);
1222 
1223 	cmd->base.speed = ndc->speed;
1224 	cmd->base.duplex = ndc->duplex;
1225 	cmd->base.port = PORT_OTHER;
1226 
1227 	return 0;
1228 }
1229 
1230 static int netvsc_set_link_ksettings(struct net_device *dev,
1231 				     const struct ethtool_link_ksettings *cmd)
1232 {
1233 	struct net_device_context *ndc = netdev_priv(dev);
1234 	struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1235 
1236 	if (vf_netdev) {
1237 		if (!vf_netdev->ethtool_ops->set_link_ksettings)
1238 			return -EOPNOTSUPP;
1239 
1240 		return vf_netdev->ethtool_ops->set_link_ksettings(vf_netdev,
1241 								  cmd);
1242 	}
1243 
1244 	return ethtool_virtdev_set_link_ksettings(dev, cmd,
1245 						  &ndc->speed, &ndc->duplex);
1246 }
1247 
1248 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
1249 {
1250 	struct net_device_context *ndevctx = netdev_priv(ndev);
1251 	struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1252 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1253 	int orig_mtu = ndev->mtu;
1254 	struct netvsc_device_info *device_info;
1255 	int ret = 0;
1256 
1257 	if (!nvdev || nvdev->destroy)
1258 		return -ENODEV;
1259 
1260 	device_info = netvsc_devinfo_get(nvdev);
1261 
1262 	if (!device_info)
1263 		return -ENOMEM;
1264 
1265 	/* Change MTU of underlying VF netdev first. */
1266 	if (vf_netdev) {
1267 		ret = dev_set_mtu(vf_netdev, mtu);
1268 		if (ret)
1269 			goto out;
1270 	}
1271 
1272 	ret = netvsc_detach(ndev, nvdev);
1273 	if (ret)
1274 		goto rollback_vf;
1275 
1276 	ndev->mtu = mtu;
1277 
1278 	ret = netvsc_attach(ndev, device_info);
1279 	if (!ret)
1280 		goto out;
1281 
1282 	/* Attempt rollback to original MTU */
1283 	ndev->mtu = orig_mtu;
1284 
1285 	if (netvsc_attach(ndev, device_info))
1286 		netdev_err(ndev, "restoring mtu failed\n");
1287 rollback_vf:
1288 	if (vf_netdev)
1289 		dev_set_mtu(vf_netdev, orig_mtu);
1290 
1291 out:
1292 	netvsc_devinfo_put(device_info);
1293 	return ret;
1294 }
1295 
1296 static void netvsc_get_vf_stats(struct net_device *net,
1297 				struct netvsc_vf_pcpu_stats *tot)
1298 {
1299 	struct net_device_context *ndev_ctx = netdev_priv(net);
1300 	int i;
1301 
1302 	memset(tot, 0, sizeof(*tot));
1303 
1304 	for_each_possible_cpu(i) {
1305 		const struct netvsc_vf_pcpu_stats *stats
1306 			= per_cpu_ptr(ndev_ctx->vf_stats, i);
1307 		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1308 		unsigned int start;
1309 
1310 		do {
1311 			start = u64_stats_fetch_begin_irq(&stats->syncp);
1312 			rx_packets = stats->rx_packets;
1313 			tx_packets = stats->tx_packets;
1314 			rx_bytes = stats->rx_bytes;
1315 			tx_bytes = stats->tx_bytes;
1316 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1317 
1318 		tot->rx_packets += rx_packets;
1319 		tot->tx_packets += tx_packets;
1320 		tot->rx_bytes   += rx_bytes;
1321 		tot->tx_bytes   += tx_bytes;
1322 		tot->tx_dropped += stats->tx_dropped;
1323 	}
1324 }
1325 
1326 static void netvsc_get_pcpu_stats(struct net_device *net,
1327 				  struct netvsc_ethtool_pcpu_stats *pcpu_tot)
1328 {
1329 	struct net_device_context *ndev_ctx = netdev_priv(net);
1330 	struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1331 	int i;
1332 
1333 	/* fetch percpu stats of vf */
1334 	for_each_possible_cpu(i) {
1335 		const struct netvsc_vf_pcpu_stats *stats =
1336 			per_cpu_ptr(ndev_ctx->vf_stats, i);
1337 		struct netvsc_ethtool_pcpu_stats *this_tot = &pcpu_tot[i];
1338 		unsigned int start;
1339 
1340 		do {
1341 			start = u64_stats_fetch_begin_irq(&stats->syncp);
1342 			this_tot->vf_rx_packets = stats->rx_packets;
1343 			this_tot->vf_tx_packets = stats->tx_packets;
1344 			this_tot->vf_rx_bytes = stats->rx_bytes;
1345 			this_tot->vf_tx_bytes = stats->tx_bytes;
1346 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1347 		this_tot->rx_packets = this_tot->vf_rx_packets;
1348 		this_tot->tx_packets = this_tot->vf_tx_packets;
1349 		this_tot->rx_bytes   = this_tot->vf_rx_bytes;
1350 		this_tot->tx_bytes   = this_tot->vf_tx_bytes;
1351 	}
1352 
1353 	/* fetch percpu stats of netvsc */
1354 	for (i = 0; i < nvdev->num_chn; i++) {
1355 		const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1356 		const struct netvsc_stats *stats;
1357 		struct netvsc_ethtool_pcpu_stats *this_tot =
1358 			&pcpu_tot[nvchan->channel->target_cpu];
1359 		u64 packets, bytes;
1360 		unsigned int start;
1361 
1362 		stats = &nvchan->tx_stats;
1363 		do {
1364 			start = u64_stats_fetch_begin_irq(&stats->syncp);
1365 			packets = stats->packets;
1366 			bytes = stats->bytes;
1367 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1368 
1369 		this_tot->tx_bytes	+= bytes;
1370 		this_tot->tx_packets	+= packets;
1371 
1372 		stats = &nvchan->rx_stats;
1373 		do {
1374 			start = u64_stats_fetch_begin_irq(&stats->syncp);
1375 			packets = stats->packets;
1376 			bytes = stats->bytes;
1377 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1378 
1379 		this_tot->rx_bytes	+= bytes;
1380 		this_tot->rx_packets	+= packets;
1381 	}
1382 }
1383 
1384 static void netvsc_get_stats64(struct net_device *net,
1385 			       struct rtnl_link_stats64 *t)
1386 {
1387 	struct net_device_context *ndev_ctx = netdev_priv(net);
1388 	struct netvsc_device *nvdev;
1389 	struct netvsc_vf_pcpu_stats vf_tot;
1390 	int i;
1391 
1392 	rcu_read_lock();
1393 
1394 	nvdev = rcu_dereference(ndev_ctx->nvdev);
1395 	if (!nvdev)
1396 		goto out;
1397 
1398 	netdev_stats_to_stats64(t, &net->stats);
1399 
1400 	netvsc_get_vf_stats(net, &vf_tot);
1401 	t->rx_packets += vf_tot.rx_packets;
1402 	t->tx_packets += vf_tot.tx_packets;
1403 	t->rx_bytes   += vf_tot.rx_bytes;
1404 	t->tx_bytes   += vf_tot.tx_bytes;
1405 	t->tx_dropped += vf_tot.tx_dropped;
1406 
1407 	for (i = 0; i < nvdev->num_chn; i++) {
1408 		const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1409 		const struct netvsc_stats *stats;
1410 		u64 packets, bytes, multicast;
1411 		unsigned int start;
1412 
1413 		stats = &nvchan->tx_stats;
1414 		do {
1415 			start = u64_stats_fetch_begin_irq(&stats->syncp);
1416 			packets = stats->packets;
1417 			bytes = stats->bytes;
1418 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1419 
1420 		t->tx_bytes	+= bytes;
1421 		t->tx_packets	+= packets;
1422 
1423 		stats = &nvchan->rx_stats;
1424 		do {
1425 			start = u64_stats_fetch_begin_irq(&stats->syncp);
1426 			packets = stats->packets;
1427 			bytes = stats->bytes;
1428 			multicast = stats->multicast + stats->broadcast;
1429 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1430 
1431 		t->rx_bytes	+= bytes;
1432 		t->rx_packets	+= packets;
1433 		t->multicast	+= multicast;
1434 	}
1435 out:
1436 	rcu_read_unlock();
1437 }
1438 
1439 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1440 {
1441 	struct net_device_context *ndc = netdev_priv(ndev);
1442 	struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1443 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1444 	struct sockaddr *addr = p;
1445 	int err;
1446 
1447 	err = eth_prepare_mac_addr_change(ndev, p);
1448 	if (err)
1449 		return err;
1450 
1451 	if (!nvdev)
1452 		return -ENODEV;
1453 
1454 	if (vf_netdev) {
1455 		err = dev_set_mac_address(vf_netdev, addr, NULL);
1456 		if (err)
1457 			return err;
1458 	}
1459 
1460 	err = rndis_filter_set_device_mac(nvdev, addr->sa_data);
1461 	if (!err) {
1462 		eth_commit_mac_addr_change(ndev, p);
1463 	} else if (vf_netdev) {
1464 		/* rollback change on VF */
1465 		memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN);
1466 		dev_set_mac_address(vf_netdev, addr, NULL);
1467 	}
1468 
1469 	return err;
1470 }
1471 
1472 static const struct {
1473 	char name[ETH_GSTRING_LEN];
1474 	u16 offset;
1475 } netvsc_stats[] = {
1476 	{ "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1477 	{ "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1478 	{ "tx_no_space",  offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1479 	{ "tx_too_big",	  offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1480 	{ "tx_busy",	  offsetof(struct netvsc_ethtool_stats, tx_busy) },
1481 	{ "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) },
1482 	{ "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) },
1483 	{ "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) },
1484 	{ "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) },
1485 	{ "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) },
1486 	{ "vlan_error", offsetof(struct netvsc_ethtool_stats, vlan_error) },
1487 }, pcpu_stats[] = {
1488 	{ "cpu%u_rx_packets",
1489 		offsetof(struct netvsc_ethtool_pcpu_stats, rx_packets) },
1490 	{ "cpu%u_rx_bytes",
1491 		offsetof(struct netvsc_ethtool_pcpu_stats, rx_bytes) },
1492 	{ "cpu%u_tx_packets",
1493 		offsetof(struct netvsc_ethtool_pcpu_stats, tx_packets) },
1494 	{ "cpu%u_tx_bytes",
1495 		offsetof(struct netvsc_ethtool_pcpu_stats, tx_bytes) },
1496 	{ "cpu%u_vf_rx_packets",
1497 		offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_packets) },
1498 	{ "cpu%u_vf_rx_bytes",
1499 		offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_bytes) },
1500 	{ "cpu%u_vf_tx_packets",
1501 		offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_packets) },
1502 	{ "cpu%u_vf_tx_bytes",
1503 		offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_bytes) },
1504 }, vf_stats[] = {
1505 	{ "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) },
1506 	{ "vf_rx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) },
1507 	{ "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) },
1508 	{ "vf_tx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) },
1509 	{ "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) },
1510 };
1511 
1512 #define NETVSC_GLOBAL_STATS_LEN	ARRAY_SIZE(netvsc_stats)
1513 #define NETVSC_VF_STATS_LEN	ARRAY_SIZE(vf_stats)
1514 
1515 /* statistics per queue (rx/tx packets/bytes) */
1516 #define NETVSC_PCPU_STATS_LEN (num_present_cpus() * ARRAY_SIZE(pcpu_stats))
1517 
1518 /* 5 statistics per queue (rx/tx packets/bytes, rx xdp_drop) */
1519 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 5)
1520 
1521 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1522 {
1523 	struct net_device_context *ndc = netdev_priv(dev);
1524 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1525 
1526 	if (!nvdev)
1527 		return -ENODEV;
1528 
1529 	switch (string_set) {
1530 	case ETH_SS_STATS:
1531 		return NETVSC_GLOBAL_STATS_LEN
1532 			+ NETVSC_VF_STATS_LEN
1533 			+ NETVSC_QUEUE_STATS_LEN(nvdev)
1534 			+ NETVSC_PCPU_STATS_LEN;
1535 	default:
1536 		return -EINVAL;
1537 	}
1538 }
1539 
1540 static void netvsc_get_ethtool_stats(struct net_device *dev,
1541 				     struct ethtool_stats *stats, u64 *data)
1542 {
1543 	struct net_device_context *ndc = netdev_priv(dev);
1544 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1545 	const void *nds = &ndc->eth_stats;
1546 	const struct netvsc_stats *qstats;
1547 	struct netvsc_vf_pcpu_stats sum;
1548 	struct netvsc_ethtool_pcpu_stats *pcpu_sum;
1549 	unsigned int start;
1550 	u64 packets, bytes;
1551 	u64 xdp_drop;
1552 	int i, j, cpu;
1553 
1554 	if (!nvdev)
1555 		return;
1556 
1557 	for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1558 		data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1559 
1560 	netvsc_get_vf_stats(dev, &sum);
1561 	for (j = 0; j < NETVSC_VF_STATS_LEN; j++)
1562 		data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset);
1563 
1564 	for (j = 0; j < nvdev->num_chn; j++) {
1565 		qstats = &nvdev->chan_table[j].tx_stats;
1566 
1567 		do {
1568 			start = u64_stats_fetch_begin_irq(&qstats->syncp);
1569 			packets = qstats->packets;
1570 			bytes = qstats->bytes;
1571 		} while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1572 		data[i++] = packets;
1573 		data[i++] = bytes;
1574 
1575 		qstats = &nvdev->chan_table[j].rx_stats;
1576 		do {
1577 			start = u64_stats_fetch_begin_irq(&qstats->syncp);
1578 			packets = qstats->packets;
1579 			bytes = qstats->bytes;
1580 			xdp_drop = qstats->xdp_drop;
1581 		} while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1582 		data[i++] = packets;
1583 		data[i++] = bytes;
1584 		data[i++] = xdp_drop;
1585 	}
1586 
1587 	pcpu_sum = kvmalloc_array(num_possible_cpus(),
1588 				  sizeof(struct netvsc_ethtool_pcpu_stats),
1589 				  GFP_KERNEL);
1590 	netvsc_get_pcpu_stats(dev, pcpu_sum);
1591 	for_each_present_cpu(cpu) {
1592 		struct netvsc_ethtool_pcpu_stats *this_sum = &pcpu_sum[cpu];
1593 
1594 		for (j = 0; j < ARRAY_SIZE(pcpu_stats); j++)
1595 			data[i++] = *(u64 *)((void *)this_sum
1596 					     + pcpu_stats[j].offset);
1597 	}
1598 	kvfree(pcpu_sum);
1599 }
1600 
1601 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1602 {
1603 	struct net_device_context *ndc = netdev_priv(dev);
1604 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1605 	u8 *p = data;
1606 	int i, cpu;
1607 
1608 	if (!nvdev)
1609 		return;
1610 
1611 	switch (stringset) {
1612 	case ETH_SS_STATS:
1613 		for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++)
1614 			ethtool_sprintf(&p, netvsc_stats[i].name);
1615 
1616 		for (i = 0; i < ARRAY_SIZE(vf_stats); i++)
1617 			ethtool_sprintf(&p, vf_stats[i].name);
1618 
1619 		for (i = 0; i < nvdev->num_chn; i++) {
1620 			ethtool_sprintf(&p, "tx_queue_%u_packets", i);
1621 			ethtool_sprintf(&p, "tx_queue_%u_bytes", i);
1622 			ethtool_sprintf(&p, "rx_queue_%u_packets", i);
1623 			ethtool_sprintf(&p, "rx_queue_%u_bytes", i);
1624 			ethtool_sprintf(&p, "rx_queue_%u_xdp_drop", i);
1625 		}
1626 
1627 		for_each_present_cpu(cpu) {
1628 			for (i = 0; i < ARRAY_SIZE(pcpu_stats); i++)
1629 				ethtool_sprintf(&p, pcpu_stats[i].name, cpu);
1630 		}
1631 
1632 		break;
1633 	}
1634 }
1635 
1636 static int
1637 netvsc_get_rss_hash_opts(struct net_device_context *ndc,
1638 			 struct ethtool_rxnfc *info)
1639 {
1640 	const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3;
1641 
1642 	info->data = RXH_IP_SRC | RXH_IP_DST;
1643 
1644 	switch (info->flow_type) {
1645 	case TCP_V4_FLOW:
1646 		if (ndc->l4_hash & HV_TCP4_L4HASH)
1647 			info->data |= l4_flag;
1648 
1649 		break;
1650 
1651 	case TCP_V6_FLOW:
1652 		if (ndc->l4_hash & HV_TCP6_L4HASH)
1653 			info->data |= l4_flag;
1654 
1655 		break;
1656 
1657 	case UDP_V4_FLOW:
1658 		if (ndc->l4_hash & HV_UDP4_L4HASH)
1659 			info->data |= l4_flag;
1660 
1661 		break;
1662 
1663 	case UDP_V6_FLOW:
1664 		if (ndc->l4_hash & HV_UDP6_L4HASH)
1665 			info->data |= l4_flag;
1666 
1667 		break;
1668 
1669 	case IPV4_FLOW:
1670 	case IPV6_FLOW:
1671 		break;
1672 	default:
1673 		info->data = 0;
1674 		break;
1675 	}
1676 
1677 	return 0;
1678 }
1679 
1680 static int
1681 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1682 		 u32 *rules)
1683 {
1684 	struct net_device_context *ndc = netdev_priv(dev);
1685 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1686 
1687 	if (!nvdev)
1688 		return -ENODEV;
1689 
1690 	switch (info->cmd) {
1691 	case ETHTOOL_GRXRINGS:
1692 		info->data = nvdev->num_chn;
1693 		return 0;
1694 
1695 	case ETHTOOL_GRXFH:
1696 		return netvsc_get_rss_hash_opts(ndc, info);
1697 	}
1698 	return -EOPNOTSUPP;
1699 }
1700 
1701 static int netvsc_set_rss_hash_opts(struct net_device_context *ndc,
1702 				    struct ethtool_rxnfc *info)
1703 {
1704 	if (info->data == (RXH_IP_SRC | RXH_IP_DST |
1705 			   RXH_L4_B_0_1 | RXH_L4_B_2_3)) {
1706 		switch (info->flow_type) {
1707 		case TCP_V4_FLOW:
1708 			ndc->l4_hash |= HV_TCP4_L4HASH;
1709 			break;
1710 
1711 		case TCP_V6_FLOW:
1712 			ndc->l4_hash |= HV_TCP6_L4HASH;
1713 			break;
1714 
1715 		case UDP_V4_FLOW:
1716 			ndc->l4_hash |= HV_UDP4_L4HASH;
1717 			break;
1718 
1719 		case UDP_V6_FLOW:
1720 			ndc->l4_hash |= HV_UDP6_L4HASH;
1721 			break;
1722 
1723 		default:
1724 			return -EOPNOTSUPP;
1725 		}
1726 
1727 		return 0;
1728 	}
1729 
1730 	if (info->data == (RXH_IP_SRC | RXH_IP_DST)) {
1731 		switch (info->flow_type) {
1732 		case TCP_V4_FLOW:
1733 			ndc->l4_hash &= ~HV_TCP4_L4HASH;
1734 			break;
1735 
1736 		case TCP_V6_FLOW:
1737 			ndc->l4_hash &= ~HV_TCP6_L4HASH;
1738 			break;
1739 
1740 		case UDP_V4_FLOW:
1741 			ndc->l4_hash &= ~HV_UDP4_L4HASH;
1742 			break;
1743 
1744 		case UDP_V6_FLOW:
1745 			ndc->l4_hash &= ~HV_UDP6_L4HASH;
1746 			break;
1747 
1748 		default:
1749 			return -EOPNOTSUPP;
1750 		}
1751 
1752 		return 0;
1753 	}
1754 
1755 	return -EOPNOTSUPP;
1756 }
1757 
1758 static int
1759 netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info)
1760 {
1761 	struct net_device_context *ndc = netdev_priv(ndev);
1762 
1763 	if (info->cmd == ETHTOOL_SRXFH)
1764 		return netvsc_set_rss_hash_opts(ndc, info);
1765 
1766 	return -EOPNOTSUPP;
1767 }
1768 
1769 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1770 {
1771 	return NETVSC_HASH_KEYLEN;
1772 }
1773 
1774 static u32 netvsc_rss_indir_size(struct net_device *dev)
1775 {
1776 	return ITAB_NUM;
1777 }
1778 
1779 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1780 			   u8 *hfunc)
1781 {
1782 	struct net_device_context *ndc = netdev_priv(dev);
1783 	struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1784 	struct rndis_device *rndis_dev;
1785 	int i;
1786 
1787 	if (!ndev)
1788 		return -ENODEV;
1789 
1790 	if (hfunc)
1791 		*hfunc = ETH_RSS_HASH_TOP;	/* Toeplitz */
1792 
1793 	rndis_dev = ndev->extension;
1794 	if (indir) {
1795 		for (i = 0; i < ITAB_NUM; i++)
1796 			indir[i] = ndc->rx_table[i];
1797 	}
1798 
1799 	if (key)
1800 		memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1801 
1802 	return 0;
1803 }
1804 
1805 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1806 			   const u8 *key, const u8 hfunc)
1807 {
1808 	struct net_device_context *ndc = netdev_priv(dev);
1809 	struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1810 	struct rndis_device *rndis_dev;
1811 	int i;
1812 
1813 	if (!ndev)
1814 		return -ENODEV;
1815 
1816 	if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1817 		return -EOPNOTSUPP;
1818 
1819 	rndis_dev = ndev->extension;
1820 	if (indir) {
1821 		for (i = 0; i < ITAB_NUM; i++)
1822 			if (indir[i] >= ndev->num_chn)
1823 				return -EINVAL;
1824 
1825 		for (i = 0; i < ITAB_NUM; i++)
1826 			ndc->rx_table[i] = indir[i];
1827 	}
1828 
1829 	if (!key) {
1830 		if (!indir)
1831 			return 0;
1832 
1833 		key = rndis_dev->rss_key;
1834 	}
1835 
1836 	return rndis_filter_set_rss_param(rndis_dev, key);
1837 }
1838 
1839 /* Hyper-V RNDIS protocol does not have ring in the HW sense.
1840  * It does have pre-allocated receive area which is divided into sections.
1841  */
1842 static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
1843 				   struct ethtool_ringparam *ring)
1844 {
1845 	u32 max_buf_size;
1846 
1847 	ring->rx_pending = nvdev->recv_section_cnt;
1848 	ring->tx_pending = nvdev->send_section_cnt;
1849 
1850 	if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
1851 		max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
1852 	else
1853 		max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1854 
1855 	ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
1856 	ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
1857 		/ nvdev->send_section_size;
1858 }
1859 
1860 static void netvsc_get_ringparam(struct net_device *ndev,
1861 				 struct ethtool_ringparam *ring,
1862 				 struct kernel_ethtool_ringparam *kernel_ring,
1863 				 struct netlink_ext_ack *extack)
1864 {
1865 	struct net_device_context *ndevctx = netdev_priv(ndev);
1866 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1867 
1868 	if (!nvdev)
1869 		return;
1870 
1871 	__netvsc_get_ringparam(nvdev, ring);
1872 }
1873 
1874 static int netvsc_set_ringparam(struct net_device *ndev,
1875 				struct ethtool_ringparam *ring,
1876 				struct kernel_ethtool_ringparam *kernel_ring,
1877 				struct netlink_ext_ack *extack)
1878 {
1879 	struct net_device_context *ndevctx = netdev_priv(ndev);
1880 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1881 	struct netvsc_device_info *device_info;
1882 	struct ethtool_ringparam orig;
1883 	u32 new_tx, new_rx;
1884 	int ret = 0;
1885 
1886 	if (!nvdev || nvdev->destroy)
1887 		return -ENODEV;
1888 
1889 	memset(&orig, 0, sizeof(orig));
1890 	__netvsc_get_ringparam(nvdev, &orig);
1891 
1892 	new_tx = clamp_t(u32, ring->tx_pending,
1893 			 NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
1894 	new_rx = clamp_t(u32, ring->rx_pending,
1895 			 NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
1896 
1897 	if (new_tx == orig.tx_pending &&
1898 	    new_rx == orig.rx_pending)
1899 		return 0;	 /* no change */
1900 
1901 	device_info = netvsc_devinfo_get(nvdev);
1902 
1903 	if (!device_info)
1904 		return -ENOMEM;
1905 
1906 	device_info->send_sections = new_tx;
1907 	device_info->recv_sections = new_rx;
1908 
1909 	ret = netvsc_detach(ndev, nvdev);
1910 	if (ret)
1911 		goto out;
1912 
1913 	ret = netvsc_attach(ndev, device_info);
1914 	if (ret) {
1915 		device_info->send_sections = orig.tx_pending;
1916 		device_info->recv_sections = orig.rx_pending;
1917 
1918 		if (netvsc_attach(ndev, device_info))
1919 			netdev_err(ndev, "restoring ringparam failed");
1920 	}
1921 
1922 out:
1923 	netvsc_devinfo_put(device_info);
1924 	return ret;
1925 }
1926 
1927 static netdev_features_t netvsc_fix_features(struct net_device *ndev,
1928 					     netdev_features_t features)
1929 {
1930 	struct net_device_context *ndevctx = netdev_priv(ndev);
1931 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1932 
1933 	if (!nvdev || nvdev->destroy)
1934 		return features;
1935 
1936 	if ((features & NETIF_F_LRO) && netvsc_xdp_get(nvdev)) {
1937 		features ^= NETIF_F_LRO;
1938 		netdev_info(ndev, "Skip LRO - unsupported with XDP\n");
1939 	}
1940 
1941 	return features;
1942 }
1943 
1944 static int netvsc_set_features(struct net_device *ndev,
1945 			       netdev_features_t features)
1946 {
1947 	netdev_features_t change = features ^ ndev->features;
1948 	struct net_device_context *ndevctx = netdev_priv(ndev);
1949 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1950 	struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1951 	struct ndis_offload_params offloads;
1952 	int ret = 0;
1953 
1954 	if (!nvdev || nvdev->destroy)
1955 		return -ENODEV;
1956 
1957 	if (!(change & NETIF_F_LRO))
1958 		goto syncvf;
1959 
1960 	memset(&offloads, 0, sizeof(struct ndis_offload_params));
1961 
1962 	if (features & NETIF_F_LRO) {
1963 		offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1964 		offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1965 	} else {
1966 		offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1967 		offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1968 	}
1969 
1970 	ret = rndis_filter_set_offload_params(ndev, nvdev, &offloads);
1971 
1972 	if (ret) {
1973 		features ^= NETIF_F_LRO;
1974 		ndev->features = features;
1975 	}
1976 
1977 syncvf:
1978 	if (!vf_netdev)
1979 		return ret;
1980 
1981 	vf_netdev->wanted_features = features;
1982 	netdev_update_features(vf_netdev);
1983 
1984 	return ret;
1985 }
1986 
1987 static int netvsc_get_regs_len(struct net_device *netdev)
1988 {
1989 	return VRSS_SEND_TAB_SIZE * sizeof(u32);
1990 }
1991 
1992 static void netvsc_get_regs(struct net_device *netdev,
1993 			    struct ethtool_regs *regs, void *p)
1994 {
1995 	struct net_device_context *ndc = netdev_priv(netdev);
1996 	u32 *regs_buff = p;
1997 
1998 	/* increase the version, if buffer format is changed. */
1999 	regs->version = 1;
2000 
2001 	memcpy(regs_buff, ndc->tx_table, VRSS_SEND_TAB_SIZE * sizeof(u32));
2002 }
2003 
2004 static u32 netvsc_get_msglevel(struct net_device *ndev)
2005 {
2006 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
2007 
2008 	return ndev_ctx->msg_enable;
2009 }
2010 
2011 static void netvsc_set_msglevel(struct net_device *ndev, u32 val)
2012 {
2013 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
2014 
2015 	ndev_ctx->msg_enable = val;
2016 }
2017 
2018 static const struct ethtool_ops ethtool_ops = {
2019 	.get_drvinfo	= netvsc_get_drvinfo,
2020 	.get_regs_len	= netvsc_get_regs_len,
2021 	.get_regs	= netvsc_get_regs,
2022 	.get_msglevel	= netvsc_get_msglevel,
2023 	.set_msglevel	= netvsc_set_msglevel,
2024 	.get_link	= ethtool_op_get_link,
2025 	.get_ethtool_stats = netvsc_get_ethtool_stats,
2026 	.get_sset_count = netvsc_get_sset_count,
2027 	.get_strings	= netvsc_get_strings,
2028 	.get_channels   = netvsc_get_channels,
2029 	.set_channels   = netvsc_set_channels,
2030 	.get_ts_info	= ethtool_op_get_ts_info,
2031 	.get_rxnfc	= netvsc_get_rxnfc,
2032 	.set_rxnfc	= netvsc_set_rxnfc,
2033 	.get_rxfh_key_size = netvsc_get_rxfh_key_size,
2034 	.get_rxfh_indir_size = netvsc_rss_indir_size,
2035 	.get_rxfh	= netvsc_get_rxfh,
2036 	.set_rxfh	= netvsc_set_rxfh,
2037 	.get_link_ksettings = netvsc_get_link_ksettings,
2038 	.set_link_ksettings = netvsc_set_link_ksettings,
2039 	.get_ringparam	= netvsc_get_ringparam,
2040 	.set_ringparam	= netvsc_set_ringparam,
2041 };
2042 
2043 static const struct net_device_ops device_ops = {
2044 	.ndo_open =			netvsc_open,
2045 	.ndo_stop =			netvsc_close,
2046 	.ndo_start_xmit =		netvsc_start_xmit,
2047 	.ndo_change_rx_flags =		netvsc_change_rx_flags,
2048 	.ndo_set_rx_mode =		netvsc_set_rx_mode,
2049 	.ndo_fix_features =		netvsc_fix_features,
2050 	.ndo_set_features =		netvsc_set_features,
2051 	.ndo_change_mtu =		netvsc_change_mtu,
2052 	.ndo_validate_addr =		eth_validate_addr,
2053 	.ndo_set_mac_address =		netvsc_set_mac_addr,
2054 	.ndo_select_queue =		netvsc_select_queue,
2055 	.ndo_get_stats64 =		netvsc_get_stats64,
2056 	.ndo_bpf =			netvsc_bpf,
2057 };
2058 
2059 /*
2060  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
2061  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
2062  * present send GARP packet to network peers with netif_notify_peers().
2063  */
2064 static void netvsc_link_change(struct work_struct *w)
2065 {
2066 	struct net_device_context *ndev_ctx =
2067 		container_of(w, struct net_device_context, dwork.work);
2068 	struct hv_device *device_obj = ndev_ctx->device_ctx;
2069 	struct net_device *net = hv_get_drvdata(device_obj);
2070 	unsigned long flags, next_reconfig, delay;
2071 	struct netvsc_reconfig *event = NULL;
2072 	struct netvsc_device *net_device;
2073 	struct rndis_device *rdev;
2074 	bool reschedule = false;
2075 
2076 	/* if changes are happening, comeback later */
2077 	if (!rtnl_trylock()) {
2078 		schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
2079 		return;
2080 	}
2081 
2082 	net_device = rtnl_dereference(ndev_ctx->nvdev);
2083 	if (!net_device)
2084 		goto out_unlock;
2085 
2086 	rdev = net_device->extension;
2087 
2088 	next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
2089 	if (time_is_after_jiffies(next_reconfig)) {
2090 		/* link_watch only sends one notification with current state
2091 		 * per second, avoid doing reconfig more frequently. Handle
2092 		 * wrap around.
2093 		 */
2094 		delay = next_reconfig - jiffies;
2095 		delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
2096 		schedule_delayed_work(&ndev_ctx->dwork, delay);
2097 		goto out_unlock;
2098 	}
2099 	ndev_ctx->last_reconfig = jiffies;
2100 
2101 	spin_lock_irqsave(&ndev_ctx->lock, flags);
2102 	if (!list_empty(&ndev_ctx->reconfig_events)) {
2103 		event = list_first_entry(&ndev_ctx->reconfig_events,
2104 					 struct netvsc_reconfig, list);
2105 		list_del(&event->list);
2106 		reschedule = !list_empty(&ndev_ctx->reconfig_events);
2107 	}
2108 	spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2109 
2110 	if (!event)
2111 		goto out_unlock;
2112 
2113 	switch (event->event) {
2114 		/* Only the following events are possible due to the check in
2115 		 * netvsc_linkstatus_callback()
2116 		 */
2117 	case RNDIS_STATUS_MEDIA_CONNECT:
2118 		if (rdev->link_state) {
2119 			rdev->link_state = false;
2120 			netif_carrier_on(net);
2121 			netvsc_tx_enable(net_device, net);
2122 		} else {
2123 			__netdev_notify_peers(net);
2124 		}
2125 		kfree(event);
2126 		break;
2127 	case RNDIS_STATUS_MEDIA_DISCONNECT:
2128 		if (!rdev->link_state) {
2129 			rdev->link_state = true;
2130 			netif_carrier_off(net);
2131 			netvsc_tx_disable(net_device, net);
2132 		}
2133 		kfree(event);
2134 		break;
2135 	case RNDIS_STATUS_NETWORK_CHANGE:
2136 		/* Only makes sense if carrier is present */
2137 		if (!rdev->link_state) {
2138 			rdev->link_state = true;
2139 			netif_carrier_off(net);
2140 			netvsc_tx_disable(net_device, net);
2141 			event->event = RNDIS_STATUS_MEDIA_CONNECT;
2142 			spin_lock_irqsave(&ndev_ctx->lock, flags);
2143 			list_add(&event->list, &ndev_ctx->reconfig_events);
2144 			spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2145 			reschedule = true;
2146 		}
2147 		break;
2148 	}
2149 
2150 	rtnl_unlock();
2151 
2152 	/* link_watch only sends one notification with current state per
2153 	 * second, handle next reconfig event in 2 seconds.
2154 	 */
2155 	if (reschedule)
2156 		schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
2157 
2158 	return;
2159 
2160 out_unlock:
2161 	rtnl_unlock();
2162 }
2163 
2164 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
2165 {
2166 	struct net_device_context *net_device_ctx;
2167 	struct net_device *dev;
2168 
2169 	dev = netdev_master_upper_dev_get(vf_netdev);
2170 	if (!dev || dev->netdev_ops != &device_ops)
2171 		return NULL;	/* not a netvsc device */
2172 
2173 	net_device_ctx = netdev_priv(dev);
2174 	if (!rtnl_dereference(net_device_ctx->nvdev))
2175 		return NULL;	/* device is removed */
2176 
2177 	return dev;
2178 }
2179 
2180 /* Called when VF is injecting data into network stack.
2181  * Change the associated network device from VF to netvsc.
2182  * note: already called with rcu_read_lock
2183  */
2184 static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
2185 {
2186 	struct sk_buff *skb = *pskb;
2187 	struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data);
2188 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
2189 	struct netvsc_vf_pcpu_stats *pcpu_stats
2190 		 = this_cpu_ptr(ndev_ctx->vf_stats);
2191 
2192 	skb = skb_share_check(skb, GFP_ATOMIC);
2193 	if (unlikely(!skb))
2194 		return RX_HANDLER_CONSUMED;
2195 
2196 	*pskb = skb;
2197 
2198 	skb->dev = ndev;
2199 
2200 	u64_stats_update_begin(&pcpu_stats->syncp);
2201 	pcpu_stats->rx_packets++;
2202 	pcpu_stats->rx_bytes += skb->len;
2203 	u64_stats_update_end(&pcpu_stats->syncp);
2204 
2205 	return RX_HANDLER_ANOTHER;
2206 }
2207 
2208 static int netvsc_vf_join(struct net_device *vf_netdev,
2209 			  struct net_device *ndev)
2210 {
2211 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
2212 	int ret;
2213 
2214 	ret = netdev_rx_handler_register(vf_netdev,
2215 					 netvsc_vf_handle_frame, ndev);
2216 	if (ret != 0) {
2217 		netdev_err(vf_netdev,
2218 			   "can not register netvsc VF receive handler (err = %d)\n",
2219 			   ret);
2220 		goto rx_handler_failed;
2221 	}
2222 
2223 	ret = netdev_master_upper_dev_link(vf_netdev, ndev,
2224 					   NULL, NULL, NULL);
2225 	if (ret != 0) {
2226 		netdev_err(vf_netdev,
2227 			   "can not set master device %s (err = %d)\n",
2228 			   ndev->name, ret);
2229 		goto upper_link_failed;
2230 	}
2231 
2232 	/* set slave flag before open to prevent IPv6 addrconf */
2233 	vf_netdev->flags |= IFF_SLAVE;
2234 
2235 	schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
2236 
2237 	call_netdevice_notifiers(NETDEV_JOIN, vf_netdev);
2238 
2239 	netdev_info(vf_netdev, "joined to %s\n", ndev->name);
2240 	return 0;
2241 
2242 upper_link_failed:
2243 	netdev_rx_handler_unregister(vf_netdev);
2244 rx_handler_failed:
2245 	return ret;
2246 }
2247 
2248 static void __netvsc_vf_setup(struct net_device *ndev,
2249 			      struct net_device *vf_netdev)
2250 {
2251 	int ret;
2252 
2253 	/* Align MTU of VF with master */
2254 	ret = dev_set_mtu(vf_netdev, ndev->mtu);
2255 	if (ret)
2256 		netdev_warn(vf_netdev,
2257 			    "unable to change mtu to %u\n", ndev->mtu);
2258 
2259 	/* set multicast etc flags on VF */
2260 	dev_change_flags(vf_netdev, ndev->flags | IFF_SLAVE, NULL);
2261 
2262 	/* sync address list from ndev to VF */
2263 	netif_addr_lock_bh(ndev);
2264 	dev_uc_sync(vf_netdev, ndev);
2265 	dev_mc_sync(vf_netdev, ndev);
2266 	netif_addr_unlock_bh(ndev);
2267 
2268 	if (netif_running(ndev)) {
2269 		ret = dev_open(vf_netdev, NULL);
2270 		if (ret)
2271 			netdev_warn(vf_netdev,
2272 				    "unable to open: %d\n", ret);
2273 	}
2274 }
2275 
2276 /* Setup VF as slave of the synthetic device.
2277  * Runs in workqueue to avoid recursion in netlink callbacks.
2278  */
2279 static void netvsc_vf_setup(struct work_struct *w)
2280 {
2281 	struct net_device_context *ndev_ctx
2282 		= container_of(w, struct net_device_context, vf_takeover.work);
2283 	struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2284 	struct net_device *vf_netdev;
2285 
2286 	if (!rtnl_trylock()) {
2287 		schedule_delayed_work(&ndev_ctx->vf_takeover, 0);
2288 		return;
2289 	}
2290 
2291 	vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2292 	if (vf_netdev)
2293 		__netvsc_vf_setup(ndev, vf_netdev);
2294 
2295 	rtnl_unlock();
2296 }
2297 
2298 /* Find netvsc by VF serial number.
2299  * The PCI hyperv controller records the serial number as the slot kobj name.
2300  */
2301 static struct net_device *get_netvsc_byslot(const struct net_device *vf_netdev)
2302 {
2303 	struct device *parent = vf_netdev->dev.parent;
2304 	struct net_device_context *ndev_ctx;
2305 	struct net_device *ndev;
2306 	struct pci_dev *pdev;
2307 	u32 serial;
2308 
2309 	if (!parent || !dev_is_pci(parent))
2310 		return NULL; /* not a PCI device */
2311 
2312 	pdev = to_pci_dev(parent);
2313 	if (!pdev->slot) {
2314 		netdev_notice(vf_netdev, "no PCI slot information\n");
2315 		return NULL;
2316 	}
2317 
2318 	if (kstrtou32(pci_slot_name(pdev->slot), 10, &serial)) {
2319 		netdev_notice(vf_netdev, "Invalid vf serial:%s\n",
2320 			      pci_slot_name(pdev->slot));
2321 		return NULL;
2322 	}
2323 
2324 	list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
2325 		if (!ndev_ctx->vf_alloc)
2326 			continue;
2327 
2328 		if (ndev_ctx->vf_serial != serial)
2329 			continue;
2330 
2331 		ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2332 		if (ndev->addr_len != vf_netdev->addr_len ||
2333 		    memcmp(ndev->perm_addr, vf_netdev->perm_addr,
2334 			   ndev->addr_len) != 0)
2335 			continue;
2336 
2337 		return ndev;
2338 
2339 	}
2340 
2341 	netdev_notice(vf_netdev,
2342 		      "no netdev found for vf serial:%u\n", serial);
2343 	return NULL;
2344 }
2345 
2346 static int netvsc_register_vf(struct net_device *vf_netdev)
2347 {
2348 	struct net_device_context *net_device_ctx;
2349 	struct netvsc_device *netvsc_dev;
2350 	struct bpf_prog *prog;
2351 	struct net_device *ndev;
2352 	int ret;
2353 
2354 	if (vf_netdev->addr_len != ETH_ALEN)
2355 		return NOTIFY_DONE;
2356 
2357 	ndev = get_netvsc_byslot(vf_netdev);
2358 	if (!ndev)
2359 		return NOTIFY_DONE;
2360 
2361 	net_device_ctx = netdev_priv(ndev);
2362 	netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2363 	if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
2364 		return NOTIFY_DONE;
2365 
2366 	/* if synthetic interface is a different namespace,
2367 	 * then move the VF to that namespace; join will be
2368 	 * done again in that context.
2369 	 */
2370 	if (!net_eq(dev_net(ndev), dev_net(vf_netdev))) {
2371 		ret = dev_change_net_namespace(vf_netdev,
2372 					       dev_net(ndev), "eth%d");
2373 		if (ret)
2374 			netdev_err(vf_netdev,
2375 				   "could not move to same namespace as %s: %d\n",
2376 				   ndev->name, ret);
2377 		else
2378 			netdev_info(vf_netdev,
2379 				    "VF moved to namespace with: %s\n",
2380 				    ndev->name);
2381 		return NOTIFY_DONE;
2382 	}
2383 
2384 	netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
2385 
2386 	if (netvsc_vf_join(vf_netdev, ndev) != 0)
2387 		return NOTIFY_DONE;
2388 
2389 	dev_hold(vf_netdev);
2390 	rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
2391 
2392 	if (ndev->needed_headroom < vf_netdev->needed_headroom)
2393 		ndev->needed_headroom = vf_netdev->needed_headroom;
2394 
2395 	vf_netdev->wanted_features = ndev->features;
2396 	netdev_update_features(vf_netdev);
2397 
2398 	prog = netvsc_xdp_get(netvsc_dev);
2399 	netvsc_vf_setxdp(vf_netdev, prog);
2400 
2401 	return NOTIFY_OK;
2402 }
2403 
2404 /* Change the data path when VF UP/DOWN/CHANGE are detected.
2405  *
2406  * Typically a UP or DOWN event is followed by a CHANGE event, so
2407  * net_device_ctx->data_path_is_vf is used to cache the current data path
2408  * to avoid the duplicate call of netvsc_switch_datapath() and the duplicate
2409  * message.
2410  *
2411  * During hibernation, if a VF NIC driver (e.g. mlx5) preserves the network
2412  * interface, there is only the CHANGE event and no UP or DOWN event.
2413  */
2414 static int netvsc_vf_changed(struct net_device *vf_netdev, unsigned long event)
2415 {
2416 	struct net_device_context *net_device_ctx;
2417 	struct netvsc_device *netvsc_dev;
2418 	struct net_device *ndev;
2419 	bool vf_is_up = false;
2420 	int ret;
2421 
2422 	if (event != NETDEV_GOING_DOWN)
2423 		vf_is_up = netif_running(vf_netdev);
2424 
2425 	ndev = get_netvsc_byref(vf_netdev);
2426 	if (!ndev)
2427 		return NOTIFY_DONE;
2428 
2429 	net_device_ctx = netdev_priv(ndev);
2430 	netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2431 	if (!netvsc_dev)
2432 		return NOTIFY_DONE;
2433 
2434 	if (net_device_ctx->data_path_is_vf == vf_is_up)
2435 		return NOTIFY_OK;
2436 
2437 	ret = netvsc_switch_datapath(ndev, vf_is_up);
2438 
2439 	if (ret) {
2440 		netdev_err(ndev,
2441 			   "Data path failed to switch %s VF: %s, err: %d\n",
2442 			   vf_is_up ? "to" : "from", vf_netdev->name, ret);
2443 		return NOTIFY_DONE;
2444 	} else {
2445 		netdev_info(ndev, "Data path switched %s VF: %s\n",
2446 			    vf_is_up ? "to" : "from", vf_netdev->name);
2447 	}
2448 
2449 	return NOTIFY_OK;
2450 }
2451 
2452 static int netvsc_unregister_vf(struct net_device *vf_netdev)
2453 {
2454 	struct net_device *ndev;
2455 	struct net_device_context *net_device_ctx;
2456 
2457 	ndev = get_netvsc_byref(vf_netdev);
2458 	if (!ndev)
2459 		return NOTIFY_DONE;
2460 
2461 	net_device_ctx = netdev_priv(ndev);
2462 	cancel_delayed_work_sync(&net_device_ctx->vf_takeover);
2463 
2464 	netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
2465 
2466 	netvsc_vf_setxdp(vf_netdev, NULL);
2467 
2468 	netdev_rx_handler_unregister(vf_netdev);
2469 	netdev_upper_dev_unlink(vf_netdev, ndev);
2470 	RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
2471 	dev_put(vf_netdev);
2472 
2473 	ndev->needed_headroom = RNDIS_AND_PPI_SIZE;
2474 
2475 	return NOTIFY_OK;
2476 }
2477 
2478 static int netvsc_probe(struct hv_device *dev,
2479 			const struct hv_vmbus_device_id *dev_id)
2480 {
2481 	struct net_device *net = NULL;
2482 	struct net_device_context *net_device_ctx;
2483 	struct netvsc_device_info *device_info = NULL;
2484 	struct netvsc_device *nvdev;
2485 	int ret = -ENOMEM;
2486 
2487 	net = alloc_etherdev_mq(sizeof(struct net_device_context),
2488 				VRSS_CHANNEL_MAX);
2489 	if (!net)
2490 		goto no_net;
2491 
2492 	netif_carrier_off(net);
2493 
2494 	netvsc_init_settings(net);
2495 
2496 	net_device_ctx = netdev_priv(net);
2497 	net_device_ctx->device_ctx = dev;
2498 	net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
2499 	if (netif_msg_probe(net_device_ctx))
2500 		netdev_dbg(net, "netvsc msg_enable: %d\n",
2501 			   net_device_ctx->msg_enable);
2502 
2503 	hv_set_drvdata(dev, net);
2504 
2505 	INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
2506 
2507 	spin_lock_init(&net_device_ctx->lock);
2508 	INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
2509 	INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup);
2510 
2511 	net_device_ctx->vf_stats
2512 		= netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats);
2513 	if (!net_device_ctx->vf_stats)
2514 		goto no_stats;
2515 
2516 	net->netdev_ops = &device_ops;
2517 	net->ethtool_ops = &ethtool_ops;
2518 	SET_NETDEV_DEV(net, &dev->device);
2519 	dma_set_min_align_mask(&dev->device, HV_HYP_PAGE_SIZE - 1);
2520 
2521 	/* We always need headroom for rndis header */
2522 	net->needed_headroom = RNDIS_AND_PPI_SIZE;
2523 
2524 	/* Initialize the number of queues to be 1, we may change it if more
2525 	 * channels are offered later.
2526 	 */
2527 	netif_set_real_num_tx_queues(net, 1);
2528 	netif_set_real_num_rx_queues(net, 1);
2529 
2530 	/* Notify the netvsc driver of the new device */
2531 	device_info = netvsc_devinfo_get(NULL);
2532 
2533 	if (!device_info) {
2534 		ret = -ENOMEM;
2535 		goto devinfo_failed;
2536 	}
2537 
2538 	nvdev = rndis_filter_device_add(dev, device_info);
2539 	if (IS_ERR(nvdev)) {
2540 		ret = PTR_ERR(nvdev);
2541 		netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
2542 		goto rndis_failed;
2543 	}
2544 
2545 	eth_hw_addr_set(net, device_info->mac_adr);
2546 
2547 	/* We must get rtnl lock before scheduling nvdev->subchan_work,
2548 	 * otherwise netvsc_subchan_work() can get rtnl lock first and wait
2549 	 * all subchannels to show up, but that may not happen because
2550 	 * netvsc_probe() can't get rtnl lock and as a result vmbus_onoffer()
2551 	 * -> ... -> device_add() -> ... -> __device_attach() can't get
2552 	 * the device lock, so all the subchannels can't be processed --
2553 	 * finally netvsc_subchan_work() hangs forever.
2554 	 */
2555 	rtnl_lock();
2556 
2557 	if (nvdev->num_chn > 1)
2558 		schedule_work(&nvdev->subchan_work);
2559 
2560 	/* hw_features computed in rndis_netdev_set_hwcaps() */
2561 	net->features = net->hw_features |
2562 		NETIF_F_HIGHDMA | NETIF_F_HW_VLAN_CTAG_TX |
2563 		NETIF_F_HW_VLAN_CTAG_RX;
2564 	net->vlan_features = net->features;
2565 
2566 	netdev_lockdep_set_classes(net);
2567 
2568 	/* MTU range: 68 - 1500 or 65521 */
2569 	net->min_mtu = NETVSC_MTU_MIN;
2570 	if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
2571 		net->max_mtu = NETVSC_MTU - ETH_HLEN;
2572 	else
2573 		net->max_mtu = ETH_DATA_LEN;
2574 
2575 	nvdev->tx_disable = false;
2576 
2577 	ret = register_netdevice(net);
2578 	if (ret != 0) {
2579 		pr_err("Unable to register netdev.\n");
2580 		goto register_failed;
2581 	}
2582 
2583 	list_add(&net_device_ctx->list, &netvsc_dev_list);
2584 	rtnl_unlock();
2585 
2586 	netvsc_devinfo_put(device_info);
2587 	return 0;
2588 
2589 register_failed:
2590 	rtnl_unlock();
2591 	rndis_filter_device_remove(dev, nvdev);
2592 rndis_failed:
2593 	netvsc_devinfo_put(device_info);
2594 devinfo_failed:
2595 	free_percpu(net_device_ctx->vf_stats);
2596 no_stats:
2597 	hv_set_drvdata(dev, NULL);
2598 	free_netdev(net);
2599 no_net:
2600 	return ret;
2601 }
2602 
2603 static int netvsc_remove(struct hv_device *dev)
2604 {
2605 	struct net_device_context *ndev_ctx;
2606 	struct net_device *vf_netdev, *net;
2607 	struct netvsc_device *nvdev;
2608 
2609 	net = hv_get_drvdata(dev);
2610 	if (net == NULL) {
2611 		dev_err(&dev->device, "No net device to remove\n");
2612 		return 0;
2613 	}
2614 
2615 	ndev_ctx = netdev_priv(net);
2616 
2617 	cancel_delayed_work_sync(&ndev_ctx->dwork);
2618 
2619 	rtnl_lock();
2620 	nvdev = rtnl_dereference(ndev_ctx->nvdev);
2621 	if (nvdev) {
2622 		cancel_work_sync(&nvdev->subchan_work);
2623 		netvsc_xdp_set(net, NULL, NULL, nvdev);
2624 	}
2625 
2626 	/*
2627 	 * Call to the vsc driver to let it know that the device is being
2628 	 * removed. Also blocks mtu and channel changes.
2629 	 */
2630 	vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2631 	if (vf_netdev)
2632 		netvsc_unregister_vf(vf_netdev);
2633 
2634 	if (nvdev)
2635 		rndis_filter_device_remove(dev, nvdev);
2636 
2637 	unregister_netdevice(net);
2638 	list_del(&ndev_ctx->list);
2639 
2640 	rtnl_unlock();
2641 
2642 	hv_set_drvdata(dev, NULL);
2643 
2644 	free_percpu(ndev_ctx->vf_stats);
2645 	free_netdev(net);
2646 	return 0;
2647 }
2648 
2649 static int netvsc_suspend(struct hv_device *dev)
2650 {
2651 	struct net_device_context *ndev_ctx;
2652 	struct netvsc_device *nvdev;
2653 	struct net_device *net;
2654 	int ret;
2655 
2656 	net = hv_get_drvdata(dev);
2657 
2658 	ndev_ctx = netdev_priv(net);
2659 	cancel_delayed_work_sync(&ndev_ctx->dwork);
2660 
2661 	rtnl_lock();
2662 
2663 	nvdev = rtnl_dereference(ndev_ctx->nvdev);
2664 	if (nvdev == NULL) {
2665 		ret = -ENODEV;
2666 		goto out;
2667 	}
2668 
2669 	/* Save the current config info */
2670 	ndev_ctx->saved_netvsc_dev_info = netvsc_devinfo_get(nvdev);
2671 
2672 	ret = netvsc_detach(net, nvdev);
2673 out:
2674 	rtnl_unlock();
2675 
2676 	return ret;
2677 }
2678 
2679 static int netvsc_resume(struct hv_device *dev)
2680 {
2681 	struct net_device *net = hv_get_drvdata(dev);
2682 	struct net_device_context *net_device_ctx;
2683 	struct netvsc_device_info *device_info;
2684 	int ret;
2685 
2686 	rtnl_lock();
2687 
2688 	net_device_ctx = netdev_priv(net);
2689 
2690 	/* Reset the data path to the netvsc NIC before re-opening the vmbus
2691 	 * channel. Later netvsc_netdev_event() will switch the data path to
2692 	 * the VF upon the UP or CHANGE event.
2693 	 */
2694 	net_device_ctx->data_path_is_vf = false;
2695 	device_info = net_device_ctx->saved_netvsc_dev_info;
2696 
2697 	ret = netvsc_attach(net, device_info);
2698 
2699 	netvsc_devinfo_put(device_info);
2700 	net_device_ctx->saved_netvsc_dev_info = NULL;
2701 
2702 	rtnl_unlock();
2703 
2704 	return ret;
2705 }
2706 static const struct hv_vmbus_device_id id_table[] = {
2707 	/* Network guid */
2708 	{ HV_NIC_GUID, },
2709 	{ },
2710 };
2711 
2712 MODULE_DEVICE_TABLE(vmbus, id_table);
2713 
2714 /* The one and only one */
2715 static struct  hv_driver netvsc_drv = {
2716 	.name = KBUILD_MODNAME,
2717 	.id_table = id_table,
2718 	.probe = netvsc_probe,
2719 	.remove = netvsc_remove,
2720 	.suspend = netvsc_suspend,
2721 	.resume = netvsc_resume,
2722 	.driver = {
2723 		.probe_type = PROBE_FORCE_SYNCHRONOUS,
2724 	},
2725 };
2726 
2727 /*
2728  * On Hyper-V, every VF interface is matched with a corresponding
2729  * synthetic interface. The synthetic interface is presented first
2730  * to the guest. When the corresponding VF instance is registered,
2731  * we will take care of switching the data path.
2732  */
2733 static int netvsc_netdev_event(struct notifier_block *this,
2734 			       unsigned long event, void *ptr)
2735 {
2736 	struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2737 
2738 	/* Skip our own events */
2739 	if (event_dev->netdev_ops == &device_ops)
2740 		return NOTIFY_DONE;
2741 
2742 	/* Avoid non-Ethernet type devices */
2743 	if (event_dev->type != ARPHRD_ETHER)
2744 		return NOTIFY_DONE;
2745 
2746 	/* Avoid Vlan dev with same MAC registering as VF */
2747 	if (is_vlan_dev(event_dev))
2748 		return NOTIFY_DONE;
2749 
2750 	/* Avoid Bonding master dev with same MAC registering as VF */
2751 	if (netif_is_bond_master(event_dev))
2752 		return NOTIFY_DONE;
2753 
2754 	switch (event) {
2755 	case NETDEV_REGISTER:
2756 		return netvsc_register_vf(event_dev);
2757 	case NETDEV_UNREGISTER:
2758 		return netvsc_unregister_vf(event_dev);
2759 	case NETDEV_UP:
2760 	case NETDEV_DOWN:
2761 	case NETDEV_CHANGE:
2762 	case NETDEV_GOING_DOWN:
2763 		return netvsc_vf_changed(event_dev, event);
2764 	default:
2765 		return NOTIFY_DONE;
2766 	}
2767 }
2768 
2769 static struct notifier_block netvsc_netdev_notifier = {
2770 	.notifier_call = netvsc_netdev_event,
2771 };
2772 
2773 static void __exit netvsc_drv_exit(void)
2774 {
2775 	unregister_netdevice_notifier(&netvsc_netdev_notifier);
2776 	vmbus_driver_unregister(&netvsc_drv);
2777 }
2778 
2779 static int __init netvsc_drv_init(void)
2780 {
2781 	int ret;
2782 
2783 	if (ring_size < RING_SIZE_MIN) {
2784 		ring_size = RING_SIZE_MIN;
2785 		pr_info("Increased ring_size to %u (min allowed)\n",
2786 			ring_size);
2787 	}
2788 	netvsc_ring_bytes = ring_size * PAGE_SIZE;
2789 
2790 	ret = vmbus_driver_register(&netvsc_drv);
2791 	if (ret)
2792 		return ret;
2793 
2794 	register_netdevice_notifier(&netvsc_netdev_notifier);
2795 	return 0;
2796 }
2797 
2798 MODULE_LICENSE("GPL");
2799 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
2800 
2801 module_init(netvsc_drv_init);
2802 module_exit(netvsc_drv_exit);
2803