xref: /linux/drivers/net/hyperv/netvsc.c (revision 2d87650a3bf1b80f7d0d150ee1af3f8a89e5b7aa)
1 /*
2  * Copyright (c) 2009, Microsoft Corporation.
3  *
4  * This program is free software; you can redistribute it and/or modify it
5  * under the terms and conditions of the GNU General Public License,
6  * version 2, as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope it will be useful, but WITHOUT
9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
11  * more details.
12  *
13  * You should have received a copy of the GNU General Public License along with
14  * this program; if not, see <http://www.gnu.org/licenses/>.
15  *
16  * Authors:
17  *   Haiyang Zhang <haiyangz@microsoft.com>
18  *   Hank Janssen  <hjanssen@microsoft.com>
19  */
20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21 
22 #include <linux/kernel.h>
23 #include <linux/sched.h>
24 #include <linux/wait.h>
25 #include <linux/mm.h>
26 #include <linux/delay.h>
27 #include <linux/io.h>
28 #include <linux/slab.h>
29 #include <linux/netdevice.h>
30 #include <linux/if_ether.h>
31 
32 #include "hyperv_net.h"
33 
34 
35 static struct netvsc_device *alloc_net_device(struct hv_device *device)
36 {
37 	struct netvsc_device *net_device;
38 	struct net_device *ndev = hv_get_drvdata(device);
39 
40 	net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
41 	if (!net_device)
42 		return NULL;
43 
44 	init_waitqueue_head(&net_device->wait_drain);
45 	net_device->start_remove = false;
46 	net_device->destroy = false;
47 	net_device->dev = device;
48 	net_device->ndev = ndev;
49 
50 	hv_set_drvdata(device, net_device);
51 	return net_device;
52 }
53 
54 static struct netvsc_device *get_outbound_net_device(struct hv_device *device)
55 {
56 	struct netvsc_device *net_device;
57 
58 	net_device = hv_get_drvdata(device);
59 	if (net_device && net_device->destroy)
60 		net_device = NULL;
61 
62 	return net_device;
63 }
64 
65 static struct netvsc_device *get_inbound_net_device(struct hv_device *device)
66 {
67 	struct netvsc_device *net_device;
68 
69 	net_device = hv_get_drvdata(device);
70 
71 	if (!net_device)
72 		goto get_in_err;
73 
74 	if (net_device->destroy &&
75 		atomic_read(&net_device->num_outstanding_sends) == 0)
76 		net_device = NULL;
77 
78 get_in_err:
79 	return net_device;
80 }
81 
82 
83 static int netvsc_destroy_recv_buf(struct netvsc_device *net_device)
84 {
85 	struct nvsp_message *revoke_packet;
86 	int ret = 0;
87 	struct net_device *ndev = net_device->ndev;
88 
89 	/*
90 	 * If we got a section count, it means we received a
91 	 * SendReceiveBufferComplete msg (ie sent
92 	 * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
93 	 * to send a revoke msg here
94 	 */
95 	if (net_device->recv_section_cnt) {
96 		/* Send the revoke receive buffer */
97 		revoke_packet = &net_device->revoke_packet;
98 		memset(revoke_packet, 0, sizeof(struct nvsp_message));
99 
100 		revoke_packet->hdr.msg_type =
101 			NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
102 		revoke_packet->msg.v1_msg.
103 		revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
104 
105 		ret = vmbus_sendpacket(net_device->dev->channel,
106 				       revoke_packet,
107 				       sizeof(struct nvsp_message),
108 				       (unsigned long)revoke_packet,
109 				       VM_PKT_DATA_INBAND, 0);
110 		/*
111 		 * If we failed here, we might as well return and
112 		 * have a leak rather than continue and a bugchk
113 		 */
114 		if (ret != 0) {
115 			netdev_err(ndev, "unable to send "
116 				"revoke receive buffer to netvsp\n");
117 			return ret;
118 		}
119 	}
120 
121 	/* Teardown the gpadl on the vsp end */
122 	if (net_device->recv_buf_gpadl_handle) {
123 		ret = vmbus_teardown_gpadl(net_device->dev->channel,
124 			   net_device->recv_buf_gpadl_handle);
125 
126 		/* If we failed here, we might as well return and have a leak
127 		 * rather than continue and a bugchk
128 		 */
129 		if (ret != 0) {
130 			netdev_err(ndev,
131 				   "unable to teardown receive buffer's gpadl\n");
132 			return ret;
133 		}
134 		net_device->recv_buf_gpadl_handle = 0;
135 	}
136 
137 	if (net_device->recv_buf) {
138 		/* Free up the receive buffer */
139 		free_pages((unsigned long)net_device->recv_buf,
140 			get_order(net_device->recv_buf_size));
141 		net_device->recv_buf = NULL;
142 	}
143 
144 	if (net_device->recv_section) {
145 		net_device->recv_section_cnt = 0;
146 		kfree(net_device->recv_section);
147 		net_device->recv_section = NULL;
148 	}
149 
150 	return ret;
151 }
152 
153 static int netvsc_init_recv_buf(struct hv_device *device)
154 {
155 	int ret = 0;
156 	int t;
157 	struct netvsc_device *net_device;
158 	struct nvsp_message *init_packet;
159 	struct net_device *ndev;
160 
161 	net_device = get_outbound_net_device(device);
162 	if (!net_device)
163 		return -ENODEV;
164 	ndev = net_device->ndev;
165 
166 	net_device->recv_buf =
167 		(void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO,
168 				get_order(net_device->recv_buf_size));
169 	if (!net_device->recv_buf) {
170 		netdev_err(ndev, "unable to allocate receive "
171 			"buffer of size %d\n", net_device->recv_buf_size);
172 		ret = -ENOMEM;
173 		goto cleanup;
174 	}
175 
176 	/*
177 	 * Establish the gpadl handle for this buffer on this
178 	 * channel.  Note: This call uses the vmbus connection rather
179 	 * than the channel to establish the gpadl handle.
180 	 */
181 	ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
182 				    net_device->recv_buf_size,
183 				    &net_device->recv_buf_gpadl_handle);
184 	if (ret != 0) {
185 		netdev_err(ndev,
186 			"unable to establish receive buffer's gpadl\n");
187 		goto cleanup;
188 	}
189 
190 
191 	/* Notify the NetVsp of the gpadl handle */
192 	init_packet = &net_device->channel_init_pkt;
193 
194 	memset(init_packet, 0, sizeof(struct nvsp_message));
195 
196 	init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
197 	init_packet->msg.v1_msg.send_recv_buf.
198 		gpadl_handle = net_device->recv_buf_gpadl_handle;
199 	init_packet->msg.v1_msg.
200 		send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
201 
202 	/* Send the gpadl notification request */
203 	ret = vmbus_sendpacket(device->channel, init_packet,
204 			       sizeof(struct nvsp_message),
205 			       (unsigned long)init_packet,
206 			       VM_PKT_DATA_INBAND,
207 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
208 	if (ret != 0) {
209 		netdev_err(ndev,
210 			"unable to send receive buffer's gpadl to netvsp\n");
211 		goto cleanup;
212 	}
213 
214 	t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
215 	BUG_ON(t == 0);
216 
217 
218 	/* Check the response */
219 	if (init_packet->msg.v1_msg.
220 	    send_recv_buf_complete.status != NVSP_STAT_SUCCESS) {
221 		netdev_err(ndev, "Unable to complete receive buffer "
222 			   "initialization with NetVsp - status %d\n",
223 			   init_packet->msg.v1_msg.
224 			   send_recv_buf_complete.status);
225 		ret = -EINVAL;
226 		goto cleanup;
227 	}
228 
229 	/* Parse the response */
230 
231 	net_device->recv_section_cnt = init_packet->msg.
232 		v1_msg.send_recv_buf_complete.num_sections;
233 
234 	net_device->recv_section = kmemdup(
235 		init_packet->msg.v1_msg.send_recv_buf_complete.sections,
236 		net_device->recv_section_cnt *
237 		sizeof(struct nvsp_1_receive_buffer_section),
238 		GFP_KERNEL);
239 	if (net_device->recv_section == NULL) {
240 		ret = -EINVAL;
241 		goto cleanup;
242 	}
243 
244 	/*
245 	 * For 1st release, there should only be 1 section that represents the
246 	 * entire receive buffer
247 	 */
248 	if (net_device->recv_section_cnt != 1 ||
249 	    net_device->recv_section->offset != 0) {
250 		ret = -EINVAL;
251 		goto cleanup;
252 	}
253 
254 	goto exit;
255 
256 cleanup:
257 	netvsc_destroy_recv_buf(net_device);
258 
259 exit:
260 	return ret;
261 }
262 
263 
264 /* Negotiate NVSP protocol version */
265 static int negotiate_nvsp_ver(struct hv_device *device,
266 			      struct netvsc_device *net_device,
267 			      struct nvsp_message *init_packet,
268 			      u32 nvsp_ver)
269 {
270 	int ret, t;
271 
272 	memset(init_packet, 0, sizeof(struct nvsp_message));
273 	init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
274 	init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
275 	init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
276 
277 	/* Send the init request */
278 	ret = vmbus_sendpacket(device->channel, init_packet,
279 			       sizeof(struct nvsp_message),
280 			       (unsigned long)init_packet,
281 			       VM_PKT_DATA_INBAND,
282 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
283 
284 	if (ret != 0)
285 		return ret;
286 
287 	t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
288 
289 	if (t == 0)
290 		return -ETIMEDOUT;
291 
292 	if (init_packet->msg.init_msg.init_complete.status !=
293 	    NVSP_STAT_SUCCESS)
294 		return -EINVAL;
295 
296 	if (nvsp_ver != NVSP_PROTOCOL_VERSION_2)
297 		return 0;
298 
299 	/* NVSPv2 only: Send NDIS config */
300 	memset(init_packet, 0, sizeof(struct nvsp_message));
301 	init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
302 	init_packet->msg.v2_msg.send_ndis_config.mtu = net_device->ndev->mtu;
303 	init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
304 
305 	ret = vmbus_sendpacket(device->channel, init_packet,
306 				sizeof(struct nvsp_message),
307 				(unsigned long)init_packet,
308 				VM_PKT_DATA_INBAND, 0);
309 
310 	return ret;
311 }
312 
313 static int netvsc_connect_vsp(struct hv_device *device)
314 {
315 	int ret;
316 	struct netvsc_device *net_device;
317 	struct nvsp_message *init_packet;
318 	int ndis_version;
319 	struct net_device *ndev;
320 
321 	net_device = get_outbound_net_device(device);
322 	if (!net_device)
323 		return -ENODEV;
324 	ndev = net_device->ndev;
325 
326 	init_packet = &net_device->channel_init_pkt;
327 
328 	/* Negotiate the latest NVSP protocol supported */
329 	if (negotiate_nvsp_ver(device, net_device, init_packet,
330 			       NVSP_PROTOCOL_VERSION_2) == 0) {
331 		net_device->nvsp_version = NVSP_PROTOCOL_VERSION_2;
332 	} else if (negotiate_nvsp_ver(device, net_device, init_packet,
333 				    NVSP_PROTOCOL_VERSION_1) == 0) {
334 		net_device->nvsp_version = NVSP_PROTOCOL_VERSION_1;
335 	} else {
336 		ret = -EPROTO;
337 		goto cleanup;
338 	}
339 
340 	pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
341 
342 	/* Send the ndis version */
343 	memset(init_packet, 0, sizeof(struct nvsp_message));
344 
345 	ndis_version = 0x00050001;
346 
347 	init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
348 	init_packet->msg.v1_msg.
349 		send_ndis_ver.ndis_major_ver =
350 				(ndis_version & 0xFFFF0000) >> 16;
351 	init_packet->msg.v1_msg.
352 		send_ndis_ver.ndis_minor_ver =
353 				ndis_version & 0xFFFF;
354 
355 	/* Send the init request */
356 	ret = vmbus_sendpacket(device->channel, init_packet,
357 				sizeof(struct nvsp_message),
358 				(unsigned long)init_packet,
359 				VM_PKT_DATA_INBAND, 0);
360 	if (ret != 0)
361 		goto cleanup;
362 
363 	/* Post the big receive buffer to NetVSP */
364 	ret = netvsc_init_recv_buf(device);
365 
366 cleanup:
367 	return ret;
368 }
369 
370 static void netvsc_disconnect_vsp(struct netvsc_device *net_device)
371 {
372 	netvsc_destroy_recv_buf(net_device);
373 }
374 
375 /*
376  * netvsc_device_remove - Callback when the root bus device is removed
377  */
378 int netvsc_device_remove(struct hv_device *device)
379 {
380 	struct netvsc_device *net_device;
381 	struct hv_netvsc_packet *netvsc_packet, *pos;
382 	unsigned long flags;
383 
384 	net_device = hv_get_drvdata(device);
385 
386 	netvsc_disconnect_vsp(net_device);
387 
388 	/*
389 	 * Since we have already drained, we don't need to busy wait
390 	 * as was done in final_release_stor_device()
391 	 * Note that we cannot set the ext pointer to NULL until
392 	 * we have drained - to drain the outgoing packets, we need to
393 	 * allow incoming packets.
394 	 */
395 
396 	spin_lock_irqsave(&device->channel->inbound_lock, flags);
397 	hv_set_drvdata(device, NULL);
398 	spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
399 
400 	/*
401 	 * At this point, no one should be accessing net_device
402 	 * except in here
403 	 */
404 	dev_notice(&device->device, "net device safe to remove\n");
405 
406 	/* Now, we can close the channel safely */
407 	vmbus_close(device->channel);
408 
409 	/* Release all resources */
410 	list_for_each_entry_safe(netvsc_packet, pos,
411 				 &net_device->recv_pkt_list, list_ent) {
412 		list_del(&netvsc_packet->list_ent);
413 		kfree(netvsc_packet);
414 	}
415 
416 	kfree(net_device);
417 	return 0;
418 }
419 
420 
421 #define RING_AVAIL_PERCENT_HIWATER 20
422 #define RING_AVAIL_PERCENT_LOWATER 10
423 
424 /*
425  * Get the percentage of available bytes to write in the ring.
426  * The return value is in range from 0 to 100.
427  */
428 static inline u32 hv_ringbuf_avail_percent(
429 		struct hv_ring_buffer_info *ring_info)
430 {
431 	u32 avail_read, avail_write;
432 
433 	hv_get_ringbuffer_availbytes(ring_info, &avail_read, &avail_write);
434 
435 	return avail_write * 100 / ring_info->ring_datasize;
436 }
437 
438 static void netvsc_send_completion(struct hv_device *device,
439 				   struct vmpacket_descriptor *packet)
440 {
441 	struct netvsc_device *net_device;
442 	struct nvsp_message *nvsp_packet;
443 	struct hv_netvsc_packet *nvsc_packet;
444 	struct net_device *ndev;
445 
446 	net_device = get_inbound_net_device(device);
447 	if (!net_device)
448 		return;
449 	ndev = net_device->ndev;
450 
451 	nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
452 			(packet->offset8 << 3));
453 
454 	if ((nvsp_packet->hdr.msg_type == NVSP_MSG_TYPE_INIT_COMPLETE) ||
455 	    (nvsp_packet->hdr.msg_type ==
456 	     NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE) ||
457 	    (nvsp_packet->hdr.msg_type ==
458 	     NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE)) {
459 		/* Copy the response back */
460 		memcpy(&net_device->channel_init_pkt, nvsp_packet,
461 		       sizeof(struct nvsp_message));
462 		complete(&net_device->channel_init_wait);
463 	} else if (nvsp_packet->hdr.msg_type ==
464 		   NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE) {
465 		int num_outstanding_sends;
466 
467 		/* Get the send context */
468 		nvsc_packet = (struct hv_netvsc_packet *)(unsigned long)
469 			packet->trans_id;
470 
471 		/* Notify the layer above us */
472 		if (nvsc_packet)
473 			nvsc_packet->completion.send.send_completion(
474 				nvsc_packet->completion.send.
475 				send_completion_ctx);
476 
477 		num_outstanding_sends =
478 			atomic_dec_return(&net_device->num_outstanding_sends);
479 
480 		if (net_device->destroy && num_outstanding_sends == 0)
481 			wake_up(&net_device->wait_drain);
482 
483 		if (netif_queue_stopped(ndev) && !net_device->start_remove &&
484 			(hv_ringbuf_avail_percent(&device->channel->outbound)
485 			> RING_AVAIL_PERCENT_HIWATER ||
486 			num_outstanding_sends < 1))
487 				netif_wake_queue(ndev);
488 	} else {
489 		netdev_err(ndev, "Unknown send completion packet type- "
490 			   "%d received!!\n", nvsp_packet->hdr.msg_type);
491 	}
492 
493 }
494 
495 int netvsc_send(struct hv_device *device,
496 			struct hv_netvsc_packet *packet)
497 {
498 	struct netvsc_device *net_device;
499 	int ret = 0;
500 	struct nvsp_message sendMessage;
501 	struct net_device *ndev;
502 	u64 req_id;
503 
504 	net_device = get_outbound_net_device(device);
505 	if (!net_device)
506 		return -ENODEV;
507 	ndev = net_device->ndev;
508 
509 	sendMessage.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
510 	if (packet->is_data_pkt) {
511 		/* 0 is RMC_DATA; */
512 		sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 0;
513 	} else {
514 		/* 1 is RMC_CONTROL; */
515 		sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 1;
516 	}
517 
518 	/* Not using send buffer section */
519 	sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_index =
520 		0xFFFFFFFF;
521 	sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_size = 0;
522 
523 	if (packet->completion.send.send_completion)
524 		req_id = (ulong)packet;
525 	else
526 		req_id = 0;
527 
528 	if (packet->page_buf_cnt) {
529 		ret = vmbus_sendpacket_pagebuffer(device->channel,
530 						  packet->page_buf,
531 						  packet->page_buf_cnt,
532 						  &sendMessage,
533 						  sizeof(struct nvsp_message),
534 						  req_id);
535 	} else {
536 		ret = vmbus_sendpacket(device->channel, &sendMessage,
537 				sizeof(struct nvsp_message),
538 				req_id,
539 				VM_PKT_DATA_INBAND,
540 				VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
541 	}
542 
543 	if (ret == 0) {
544 		atomic_inc(&net_device->num_outstanding_sends);
545 		if (hv_ringbuf_avail_percent(&device->channel->outbound) <
546 			RING_AVAIL_PERCENT_LOWATER) {
547 			netif_stop_queue(ndev);
548 			if (atomic_read(&net_device->
549 				num_outstanding_sends) < 1)
550 				netif_wake_queue(ndev);
551 		}
552 	} else if (ret == -EAGAIN) {
553 		netif_stop_queue(ndev);
554 		if (atomic_read(&net_device->num_outstanding_sends) < 1) {
555 			netif_wake_queue(ndev);
556 			ret = -ENOSPC;
557 		}
558 	} else {
559 		netdev_err(ndev, "Unable to send packet %p ret %d\n",
560 			   packet, ret);
561 	}
562 
563 	return ret;
564 }
565 
566 static void netvsc_send_recv_completion(struct hv_device *device,
567 					u64 transaction_id, u32 status)
568 {
569 	struct nvsp_message recvcompMessage;
570 	int retries = 0;
571 	int ret;
572 	struct net_device *ndev;
573 	struct netvsc_device *net_device = hv_get_drvdata(device);
574 
575 	ndev = net_device->ndev;
576 
577 	recvcompMessage.hdr.msg_type =
578 				NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
579 
580 	recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status = status;
581 
582 retry_send_cmplt:
583 	/* Send the completion */
584 	ret = vmbus_sendpacket(device->channel, &recvcompMessage,
585 			       sizeof(struct nvsp_message), transaction_id,
586 			       VM_PKT_COMP, 0);
587 	if (ret == 0) {
588 		/* success */
589 		/* no-op */
590 	} else if (ret == -EAGAIN) {
591 		/* no more room...wait a bit and attempt to retry 3 times */
592 		retries++;
593 		netdev_err(ndev, "unable to send receive completion pkt"
594 			" (tid %llx)...retrying %d\n", transaction_id, retries);
595 
596 		if (retries < 4) {
597 			udelay(100);
598 			goto retry_send_cmplt;
599 		} else {
600 			netdev_err(ndev, "unable to send receive "
601 				"completion pkt (tid %llx)...give up retrying\n",
602 				transaction_id);
603 		}
604 	} else {
605 		netdev_err(ndev, "unable to send receive "
606 			"completion pkt - %llx\n", transaction_id);
607 	}
608 }
609 
610 /* Send a receive completion packet to RNDIS device (ie NetVsp) */
611 static void netvsc_receive_completion(void *context)
612 {
613 	struct hv_netvsc_packet *packet = context;
614 	struct hv_device *device = packet->device;
615 	struct netvsc_device *net_device;
616 	u64 transaction_id = 0;
617 	bool fsend_receive_comp = false;
618 	unsigned long flags;
619 	struct net_device *ndev;
620 	u32 status = NVSP_STAT_NONE;
621 
622 	/*
623 	 * Even though it seems logical to do a GetOutboundNetDevice() here to
624 	 * send out receive completion, we are using GetInboundNetDevice()
625 	 * since we may have disable outbound traffic already.
626 	 */
627 	net_device = get_inbound_net_device(device);
628 	if (!net_device)
629 		return;
630 	ndev = net_device->ndev;
631 
632 	/* Overloading use of the lock. */
633 	spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
634 
635 	if (packet->status != NVSP_STAT_SUCCESS)
636 		packet->xfer_page_pkt->status = NVSP_STAT_FAIL;
637 
638 	packet->xfer_page_pkt->count--;
639 
640 	/*
641 	 * Last one in the line that represent 1 xfer page packet.
642 	 * Return the xfer page packet itself to the freelist
643 	 */
644 	if (packet->xfer_page_pkt->count == 0) {
645 		fsend_receive_comp = true;
646 		transaction_id = packet->completion.recv.recv_completion_tid;
647 		status = packet->xfer_page_pkt->status;
648 		list_add_tail(&packet->xfer_page_pkt->list_ent,
649 			      &net_device->recv_pkt_list);
650 
651 	}
652 
653 	/* Put the packet back */
654 	list_add_tail(&packet->list_ent, &net_device->recv_pkt_list);
655 	spin_unlock_irqrestore(&net_device->recv_pkt_list_lock, flags);
656 
657 	/* Send a receive completion for the xfer page packet */
658 	if (fsend_receive_comp)
659 		netvsc_send_recv_completion(device, transaction_id, status);
660 
661 }
662 
663 static void netvsc_receive(struct hv_device *device,
664 			    struct vmpacket_descriptor *packet)
665 {
666 	struct netvsc_device *net_device;
667 	struct vmtransfer_page_packet_header *vmxferpage_packet;
668 	struct nvsp_message *nvsp_packet;
669 	struct hv_netvsc_packet *netvsc_packet = NULL;
670 	/* struct netvsc_driver *netvscDriver; */
671 	struct xferpage_packet *xferpage_packet = NULL;
672 	int i;
673 	int count = 0;
674 	unsigned long flags;
675 	struct net_device *ndev;
676 
677 	LIST_HEAD(listHead);
678 
679 	net_device = get_inbound_net_device(device);
680 	if (!net_device)
681 		return;
682 	ndev = net_device->ndev;
683 
684 	/*
685 	 * All inbound packets other than send completion should be xfer page
686 	 * packet
687 	 */
688 	if (packet->type != VM_PKT_DATA_USING_XFER_PAGES) {
689 		netdev_err(ndev, "Unknown packet type received - %d\n",
690 			   packet->type);
691 		return;
692 	}
693 
694 	nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
695 			(packet->offset8 << 3));
696 
697 	/* Make sure this is a valid nvsp packet */
698 	if (nvsp_packet->hdr.msg_type !=
699 	    NVSP_MSG1_TYPE_SEND_RNDIS_PKT) {
700 		netdev_err(ndev, "Unknown nvsp packet type received-"
701 			" %d\n", nvsp_packet->hdr.msg_type);
702 		return;
703 	}
704 
705 	vmxferpage_packet = (struct vmtransfer_page_packet_header *)packet;
706 
707 	if (vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID) {
708 		netdev_err(ndev, "Invalid xfer page set id - "
709 			   "expecting %x got %x\n", NETVSC_RECEIVE_BUFFER_ID,
710 			   vmxferpage_packet->xfer_pageset_id);
711 		return;
712 	}
713 
714 	/*
715 	 * Grab free packets (range count + 1) to represent this xfer
716 	 * page packet. +1 to represent the xfer page packet itself.
717 	 * We grab it here so that we know exactly how many we can
718 	 * fulfil
719 	 */
720 	spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
721 	while (!list_empty(&net_device->recv_pkt_list)) {
722 		list_move_tail(net_device->recv_pkt_list.next, &listHead);
723 		if (++count == vmxferpage_packet->range_cnt + 1)
724 			break;
725 	}
726 	spin_unlock_irqrestore(&net_device->recv_pkt_list_lock, flags);
727 
728 	/*
729 	 * We need at least 2 netvsc pkts (1 to represent the xfer
730 	 * page and at least 1 for the range) i.e. we can handled
731 	 * some of the xfer page packet ranges...
732 	 */
733 	if (count < 2) {
734 		netdev_err(ndev, "Got only %d netvsc pkt...needed "
735 			"%d pkts. Dropping this xfer page packet completely!\n",
736 			count, vmxferpage_packet->range_cnt + 1);
737 
738 		/* Return it to the freelist */
739 		spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
740 		for (i = count; i != 0; i--) {
741 			list_move_tail(listHead.next,
742 				       &net_device->recv_pkt_list);
743 		}
744 		spin_unlock_irqrestore(&net_device->recv_pkt_list_lock,
745 				       flags);
746 
747 		netvsc_send_recv_completion(device,
748 					    vmxferpage_packet->d.trans_id,
749 					    NVSP_STAT_FAIL);
750 
751 		return;
752 	}
753 
754 	/* Remove the 1st packet to represent the xfer page packet itself */
755 	xferpage_packet = (struct xferpage_packet *)listHead.next;
756 	list_del(&xferpage_packet->list_ent);
757 	xferpage_packet->status = NVSP_STAT_SUCCESS;
758 
759 	/* This is how much we can satisfy */
760 	xferpage_packet->count = count - 1;
761 
762 	if (xferpage_packet->count != vmxferpage_packet->range_cnt) {
763 		netdev_err(ndev, "Needed %d netvsc pkts to satisfy "
764 			"this xfer page...got %d\n",
765 			vmxferpage_packet->range_cnt, xferpage_packet->count);
766 	}
767 
768 	/* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
769 	for (i = 0; i < (count - 1); i++) {
770 		netvsc_packet = (struct hv_netvsc_packet *)listHead.next;
771 		list_del(&netvsc_packet->list_ent);
772 
773 		/* Initialize the netvsc packet */
774 		netvsc_packet->status = NVSP_STAT_SUCCESS;
775 		netvsc_packet->xfer_page_pkt = xferpage_packet;
776 		netvsc_packet->completion.recv.recv_completion =
777 					netvsc_receive_completion;
778 		netvsc_packet->completion.recv.recv_completion_ctx =
779 					netvsc_packet;
780 		netvsc_packet->device = device;
781 		/* Save this so that we can send it back */
782 		netvsc_packet->completion.recv.recv_completion_tid =
783 					vmxferpage_packet->d.trans_id;
784 
785 		netvsc_packet->data = (void *)((unsigned long)net_device->
786 			recv_buf + vmxferpage_packet->ranges[i].byte_offset);
787 		netvsc_packet->total_data_buflen =
788 					vmxferpage_packet->ranges[i].byte_count;
789 
790 		/* Pass it to the upper layer */
791 		rndis_filter_receive(device, netvsc_packet);
792 
793 		netvsc_receive_completion(netvsc_packet->
794 				completion.recv.recv_completion_ctx);
795 	}
796 
797 }
798 
799 static void netvsc_channel_cb(void *context)
800 {
801 	int ret;
802 	struct hv_device *device = context;
803 	struct netvsc_device *net_device;
804 	u32 bytes_recvd;
805 	u64 request_id;
806 	unsigned char *packet;
807 	struct vmpacket_descriptor *desc;
808 	unsigned char *buffer;
809 	int bufferlen = NETVSC_PACKET_SIZE;
810 	struct net_device *ndev;
811 
812 	packet = kzalloc(NETVSC_PACKET_SIZE * sizeof(unsigned char),
813 			 GFP_ATOMIC);
814 	if (!packet)
815 		return;
816 	buffer = packet;
817 
818 	net_device = get_inbound_net_device(device);
819 	if (!net_device)
820 		goto out;
821 	ndev = net_device->ndev;
822 
823 	do {
824 		ret = vmbus_recvpacket_raw(device->channel, buffer, bufferlen,
825 					   &bytes_recvd, &request_id);
826 		if (ret == 0) {
827 			if (bytes_recvd > 0) {
828 				desc = (struct vmpacket_descriptor *)buffer;
829 				switch (desc->type) {
830 				case VM_PKT_COMP:
831 					netvsc_send_completion(device, desc);
832 					break;
833 
834 				case VM_PKT_DATA_USING_XFER_PAGES:
835 					netvsc_receive(device, desc);
836 					break;
837 
838 				default:
839 					netdev_err(ndev,
840 						   "unhandled packet type %d, "
841 						   "tid %llx len %d\n",
842 						   desc->type, request_id,
843 						   bytes_recvd);
844 					break;
845 				}
846 
847 				/* reset */
848 				if (bufferlen > NETVSC_PACKET_SIZE) {
849 					kfree(buffer);
850 					buffer = packet;
851 					bufferlen = NETVSC_PACKET_SIZE;
852 				}
853 			} else {
854 				/* reset */
855 				if (bufferlen > NETVSC_PACKET_SIZE) {
856 					kfree(buffer);
857 					buffer = packet;
858 					bufferlen = NETVSC_PACKET_SIZE;
859 				}
860 
861 				break;
862 			}
863 		} else if (ret == -ENOBUFS) {
864 			/* Handle large packet */
865 			buffer = kmalloc(bytes_recvd, GFP_ATOMIC);
866 			if (buffer == NULL) {
867 				/* Try again next time around */
868 				netdev_err(ndev,
869 					   "unable to allocate buffer of size "
870 					   "(%d)!!\n", bytes_recvd);
871 				break;
872 			}
873 
874 			bufferlen = bytes_recvd;
875 		}
876 	} while (1);
877 
878 out:
879 	kfree(buffer);
880 	return;
881 }
882 
883 /*
884  * netvsc_device_add - Callback when the device belonging to this
885  * driver is added
886  */
887 int netvsc_device_add(struct hv_device *device, void *additional_info)
888 {
889 	int ret = 0;
890 	int i;
891 	int ring_size =
892 	((struct netvsc_device_info *)additional_info)->ring_size;
893 	struct netvsc_device *net_device;
894 	struct hv_netvsc_packet *packet, *pos;
895 	struct net_device *ndev;
896 
897 	net_device = alloc_net_device(device);
898 	if (!net_device) {
899 		ret = -ENOMEM;
900 		goto cleanup;
901 	}
902 
903 	/*
904 	 * Coming into this function, struct net_device * is
905 	 * registered as the driver private data.
906 	 * In alloc_net_device(), we register struct netvsc_device *
907 	 * as the driver private data and stash away struct net_device *
908 	 * in struct netvsc_device *.
909 	 */
910 	ndev = net_device->ndev;
911 
912 	/* Initialize the NetVSC channel extension */
913 	net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
914 	spin_lock_init(&net_device->recv_pkt_list_lock);
915 
916 	INIT_LIST_HEAD(&net_device->recv_pkt_list);
917 
918 	for (i = 0; i < NETVSC_RECEIVE_PACKETLIST_COUNT; i++) {
919 		packet = kzalloc(sizeof(struct hv_netvsc_packet), GFP_KERNEL);
920 		if (!packet)
921 			break;
922 
923 		list_add_tail(&packet->list_ent,
924 			      &net_device->recv_pkt_list);
925 	}
926 	init_completion(&net_device->channel_init_wait);
927 
928 	/* Open the channel */
929 	ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
930 			 ring_size * PAGE_SIZE, NULL, 0,
931 			 netvsc_channel_cb, device);
932 
933 	if (ret != 0) {
934 		netdev_err(ndev, "unable to open channel: %d\n", ret);
935 		goto cleanup;
936 	}
937 
938 	/* Channel is opened */
939 	pr_info("hv_netvsc channel opened successfully\n");
940 
941 	/* Connect with the NetVsp */
942 	ret = netvsc_connect_vsp(device);
943 	if (ret != 0) {
944 		netdev_err(ndev,
945 			"unable to connect to NetVSP - %d\n", ret);
946 		goto close;
947 	}
948 
949 	return ret;
950 
951 close:
952 	/* Now, we can close the channel safely */
953 	vmbus_close(device->channel);
954 
955 cleanup:
956 
957 	if (net_device) {
958 		list_for_each_entry_safe(packet, pos,
959 					 &net_device->recv_pkt_list,
960 					 list_ent) {
961 			list_del(&packet->list_ent);
962 			kfree(packet);
963 		}
964 
965 		kfree(net_device);
966 	}
967 
968 	return ret;
969 }
970