xref: /linux/drivers/net/hyperv/netvsc.c (revision a508da6cc0093171833efb8376b00473f24221b9)
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, write to the Free Software Foundation, Inc., 59 Temple
15  * Place - Suite 330, Boston, MA 02111-1307 USA.
16  *
17  * Authors:
18  *   Haiyang Zhang <haiyangz@microsoft.com>
19  *   Hank Janssen  <hjanssen@microsoft.com>
20  */
21 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
22 
23 #include <linux/kernel.h>
24 #include <linux/sched.h>
25 #include <linux/wait.h>
26 #include <linux/mm.h>
27 #include <linux/delay.h>
28 #include <linux/io.h>
29 #include <linux/slab.h>
30 #include <linux/netdevice.h>
31 #include <linux/if_ether.h>
32 
33 #include "hyperv_net.h"
34 
35 
36 static struct netvsc_device *alloc_net_device(struct hv_device *device)
37 {
38 	struct netvsc_device *net_device;
39 	struct net_device *ndev = hv_get_drvdata(device);
40 
41 	net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
42 	if (!net_device)
43 		return NULL;
44 
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 	spin_lock_irqsave(&device->channel->inbound_lock, flags);
386 	net_device->destroy = true;
387 	spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
388 
389 	/* Wait for all send completions */
390 	while (atomic_read(&net_device->num_outstanding_sends)) {
391 		dev_info(&device->device,
392 			"waiting for %d requests to complete...\n",
393 			atomic_read(&net_device->num_outstanding_sends));
394 		udelay(100);
395 	}
396 
397 	netvsc_disconnect_vsp(net_device);
398 
399 	/*
400 	 * Since we have already drained, we don't need to busy wait
401 	 * as was done in final_release_stor_device()
402 	 * Note that we cannot set the ext pointer to NULL until
403 	 * we have drained - to drain the outgoing packets, we need to
404 	 * allow incoming packets.
405 	 */
406 
407 	spin_lock_irqsave(&device->channel->inbound_lock, flags);
408 	hv_set_drvdata(device, NULL);
409 	spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
410 
411 	/*
412 	 * At this point, no one should be accessing net_device
413 	 * except in here
414 	 */
415 	dev_notice(&device->device, "net device safe to remove\n");
416 
417 	/* Now, we can close the channel safely */
418 	vmbus_close(device->channel);
419 
420 	/* Release all resources */
421 	list_for_each_entry_safe(netvsc_packet, pos,
422 				 &net_device->recv_pkt_list, list_ent) {
423 		list_del(&netvsc_packet->list_ent);
424 		kfree(netvsc_packet);
425 	}
426 
427 	kfree(net_device);
428 	return 0;
429 }
430 
431 
432 #define RING_AVAIL_PERCENT_HIWATER 20
433 #define RING_AVAIL_PERCENT_LOWATER 10
434 
435 /*
436  * Get the percentage of available bytes to write in the ring.
437  * The return value is in range from 0 to 100.
438  */
439 static inline u32 hv_ringbuf_avail_percent(
440 		struct hv_ring_buffer_info *ring_info)
441 {
442 	u32 avail_read, avail_write;
443 
444 	hv_get_ringbuffer_availbytes(ring_info, &avail_read, &avail_write);
445 
446 	return avail_write * 100 / ring_info->ring_datasize;
447 }
448 
449 static void netvsc_send_completion(struct hv_device *device,
450 				   struct vmpacket_descriptor *packet)
451 {
452 	struct netvsc_device *net_device;
453 	struct nvsp_message *nvsp_packet;
454 	struct hv_netvsc_packet *nvsc_packet;
455 	struct net_device *ndev;
456 
457 	net_device = get_inbound_net_device(device);
458 	if (!net_device)
459 		return;
460 	ndev = net_device->ndev;
461 
462 	nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
463 			(packet->offset8 << 3));
464 
465 	if ((nvsp_packet->hdr.msg_type == NVSP_MSG_TYPE_INIT_COMPLETE) ||
466 	    (nvsp_packet->hdr.msg_type ==
467 	     NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE) ||
468 	    (nvsp_packet->hdr.msg_type ==
469 	     NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE)) {
470 		/* Copy the response back */
471 		memcpy(&net_device->channel_init_pkt, nvsp_packet,
472 		       sizeof(struct nvsp_message));
473 		complete(&net_device->channel_init_wait);
474 	} else if (nvsp_packet->hdr.msg_type ==
475 		   NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE) {
476 		int num_outstanding_sends;
477 
478 		/* Get the send context */
479 		nvsc_packet = (struct hv_netvsc_packet *)(unsigned long)
480 			packet->trans_id;
481 
482 		/* Notify the layer above us */
483 		nvsc_packet->completion.send.send_completion(
484 			nvsc_packet->completion.send.send_completion_ctx);
485 
486 		num_outstanding_sends =
487 			atomic_dec_return(&net_device->num_outstanding_sends);
488 
489 		if (netif_queue_stopped(ndev) && !net_device->start_remove &&
490 			(hv_ringbuf_avail_percent(&device->channel->outbound)
491 			> RING_AVAIL_PERCENT_HIWATER ||
492 			num_outstanding_sends < 1))
493 				netif_wake_queue(ndev);
494 	} else {
495 		netdev_err(ndev, "Unknown send completion packet type- "
496 			   "%d received!!\n", nvsp_packet->hdr.msg_type);
497 	}
498 
499 }
500 
501 int netvsc_send(struct hv_device *device,
502 			struct hv_netvsc_packet *packet)
503 {
504 	struct netvsc_device *net_device;
505 	int ret = 0;
506 	struct nvsp_message sendMessage;
507 	struct net_device *ndev;
508 
509 	net_device = get_outbound_net_device(device);
510 	if (!net_device)
511 		return -ENODEV;
512 	ndev = net_device->ndev;
513 
514 	sendMessage.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
515 	if (packet->is_data_pkt) {
516 		/* 0 is RMC_DATA; */
517 		sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 0;
518 	} else {
519 		/* 1 is RMC_CONTROL; */
520 		sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 1;
521 	}
522 
523 	/* Not using send buffer section */
524 	sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_index =
525 		0xFFFFFFFF;
526 	sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_size = 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 						  (unsigned long)packet);
535 	} else {
536 		ret = vmbus_sendpacket(device->channel, &sendMessage,
537 				sizeof(struct nvsp_message),
538 				(unsigned long)packet,
539 				VM_PKT_DATA_INBAND,
540 				VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
541 
542 	}
543 
544 	if (ret == 0) {
545 		atomic_inc(&net_device->num_outstanding_sends);
546 		if (hv_ringbuf_avail_percent(&device->channel->outbound) <
547 			RING_AVAIL_PERCENT_LOWATER) {
548 			netif_stop_queue(ndev);
549 			if (atomic_read(&net_device->
550 				num_outstanding_sends) < 1)
551 				netif_wake_queue(ndev);
552 		}
553 	} else if (ret == -EAGAIN) {
554 		netif_stop_queue(ndev);
555 		if (atomic_read(&net_device->num_outstanding_sends) < 1) {
556 			netif_wake_queue(ndev);
557 			ret = -ENOSPC;
558 		}
559 	} else {
560 		netdev_err(ndev, "Unable to send packet %p ret %d\n",
561 			   packet, ret);
562 	}
563 
564 	return ret;
565 }
566 
567 static void netvsc_send_recv_completion(struct hv_device *device,
568 					u64 transaction_id)
569 {
570 	struct nvsp_message recvcompMessage;
571 	int retries = 0;
572 	int ret;
573 	struct net_device *ndev;
574 	struct netvsc_device *net_device = hv_get_drvdata(device);
575 
576 	ndev = net_device->ndev;
577 
578 	recvcompMessage.hdr.msg_type =
579 				NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
580 
581 	/* FIXME: Pass in the status */
582 	recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status =
583 		NVSP_STAT_SUCCESS;
584 
585 retry_send_cmplt:
586 	/* Send the completion */
587 	ret = vmbus_sendpacket(device->channel, &recvcompMessage,
588 			       sizeof(struct nvsp_message), transaction_id,
589 			       VM_PKT_COMP, 0);
590 	if (ret == 0) {
591 		/* success */
592 		/* no-op */
593 	} else if (ret == -EAGAIN) {
594 		/* no more room...wait a bit and attempt to retry 3 times */
595 		retries++;
596 		netdev_err(ndev, "unable to send receive completion pkt"
597 			" (tid %llx)...retrying %d\n", transaction_id, retries);
598 
599 		if (retries < 4) {
600 			udelay(100);
601 			goto retry_send_cmplt;
602 		} else {
603 			netdev_err(ndev, "unable to send receive "
604 				"completion pkt (tid %llx)...give up retrying\n",
605 				transaction_id);
606 		}
607 	} else {
608 		netdev_err(ndev, "unable to send receive "
609 			"completion pkt - %llx\n", transaction_id);
610 	}
611 }
612 
613 /* Send a receive completion packet to RNDIS device (ie NetVsp) */
614 static void netvsc_receive_completion(void *context)
615 {
616 	struct hv_netvsc_packet *packet = context;
617 	struct hv_device *device = (struct hv_device *)packet->device;
618 	struct netvsc_device *net_device;
619 	u64 transaction_id = 0;
620 	bool fsend_receive_comp = false;
621 	unsigned long flags;
622 	struct net_device *ndev;
623 
624 	/*
625 	 * Even though it seems logical to do a GetOutboundNetDevice() here to
626 	 * send out receive completion, we are using GetInboundNetDevice()
627 	 * since we may have disable outbound traffic already.
628 	 */
629 	net_device = get_inbound_net_device(device);
630 	if (!net_device)
631 		return;
632 	ndev = net_device->ndev;
633 
634 	/* Overloading use of the lock. */
635 	spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
636 
637 	packet->xfer_page_pkt->count--;
638 
639 	/*
640 	 * Last one in the line that represent 1 xfer page packet.
641 	 * Return the xfer page packet itself to the freelist
642 	 */
643 	if (packet->xfer_page_pkt->count == 0) {
644 		fsend_receive_comp = true;
645 		transaction_id = packet->completion.recv.recv_completion_tid;
646 		list_add_tail(&packet->xfer_page_pkt->list_ent,
647 			      &net_device->recv_pkt_list);
648 
649 	}
650 
651 	/* Put the packet back */
652 	list_add_tail(&packet->list_ent, &net_device->recv_pkt_list);
653 	spin_unlock_irqrestore(&net_device->recv_pkt_list_lock, flags);
654 
655 	/* Send a receive completion for the xfer page packet */
656 	if (fsend_receive_comp)
657 		netvsc_send_recv_completion(device, transaction_id);
658 
659 }
660 
661 static void netvsc_receive(struct hv_device *device,
662 			    struct vmpacket_descriptor *packet)
663 {
664 	struct netvsc_device *net_device;
665 	struct vmtransfer_page_packet_header *vmxferpage_packet;
666 	struct nvsp_message *nvsp_packet;
667 	struct hv_netvsc_packet *netvsc_packet = NULL;
668 	/* struct netvsc_driver *netvscDriver; */
669 	struct xferpage_packet *xferpage_packet = NULL;
670 	int i;
671 	int count = 0;
672 	unsigned long flags;
673 	struct net_device *ndev;
674 
675 	LIST_HEAD(listHead);
676 
677 	net_device = get_inbound_net_device(device);
678 	if (!net_device)
679 		return;
680 	ndev = net_device->ndev;
681 
682 	/*
683 	 * All inbound packets other than send completion should be xfer page
684 	 * packet
685 	 */
686 	if (packet->type != VM_PKT_DATA_USING_XFER_PAGES) {
687 		netdev_err(ndev, "Unknown packet type received - %d\n",
688 			   packet->type);
689 		return;
690 	}
691 
692 	nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
693 			(packet->offset8 << 3));
694 
695 	/* Make sure this is a valid nvsp packet */
696 	if (nvsp_packet->hdr.msg_type !=
697 	    NVSP_MSG1_TYPE_SEND_RNDIS_PKT) {
698 		netdev_err(ndev, "Unknown nvsp packet type received-"
699 			" %d\n", nvsp_packet->hdr.msg_type);
700 		return;
701 	}
702 
703 	vmxferpage_packet = (struct vmtransfer_page_packet_header *)packet;
704 
705 	if (vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID) {
706 		netdev_err(ndev, "Invalid xfer page set id - "
707 			   "expecting %x got %x\n", NETVSC_RECEIVE_BUFFER_ID,
708 			   vmxferpage_packet->xfer_pageset_id);
709 		return;
710 	}
711 
712 	/*
713 	 * Grab free packets (range count + 1) to represent this xfer
714 	 * page packet. +1 to represent the xfer page packet itself.
715 	 * We grab it here so that we know exactly how many we can
716 	 * fulfil
717 	 */
718 	spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
719 	while (!list_empty(&net_device->recv_pkt_list)) {
720 		list_move_tail(net_device->recv_pkt_list.next, &listHead);
721 		if (++count == vmxferpage_packet->range_cnt + 1)
722 			break;
723 	}
724 	spin_unlock_irqrestore(&net_device->recv_pkt_list_lock, flags);
725 
726 	/*
727 	 * We need at least 2 netvsc pkts (1 to represent the xfer
728 	 * page and at least 1 for the range) i.e. we can handled
729 	 * some of the xfer page packet ranges...
730 	 */
731 	if (count < 2) {
732 		netdev_err(ndev, "Got only %d netvsc pkt...needed "
733 			"%d pkts. Dropping this xfer page packet completely!\n",
734 			count, vmxferpage_packet->range_cnt + 1);
735 
736 		/* Return it to the freelist */
737 		spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
738 		for (i = count; i != 0; i--) {
739 			list_move_tail(listHead.next,
740 				       &net_device->recv_pkt_list);
741 		}
742 		spin_unlock_irqrestore(&net_device->recv_pkt_list_lock,
743 				       flags);
744 
745 		netvsc_send_recv_completion(device,
746 					    vmxferpage_packet->d.trans_id);
747 
748 		return;
749 	}
750 
751 	/* Remove the 1st packet to represent the xfer page packet itself */
752 	xferpage_packet = (struct xferpage_packet *)listHead.next;
753 	list_del(&xferpage_packet->list_ent);
754 
755 	/* This is how much we can satisfy */
756 	xferpage_packet->count = count - 1;
757 
758 	if (xferpage_packet->count != vmxferpage_packet->range_cnt) {
759 		netdev_err(ndev, "Needed %d netvsc pkts to satisfy "
760 			"this xfer page...got %d\n",
761 			vmxferpage_packet->range_cnt, xferpage_packet->count);
762 	}
763 
764 	/* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
765 	for (i = 0; i < (count - 1); i++) {
766 		netvsc_packet = (struct hv_netvsc_packet *)listHead.next;
767 		list_del(&netvsc_packet->list_ent);
768 
769 		/* Initialize the netvsc packet */
770 		netvsc_packet->xfer_page_pkt = xferpage_packet;
771 		netvsc_packet->completion.recv.recv_completion =
772 					netvsc_receive_completion;
773 		netvsc_packet->completion.recv.recv_completion_ctx =
774 					netvsc_packet;
775 		netvsc_packet->device = device;
776 		/* Save this so that we can send it back */
777 		netvsc_packet->completion.recv.recv_completion_tid =
778 					vmxferpage_packet->d.trans_id;
779 
780 		netvsc_packet->data = (void *)((unsigned long)net_device->
781 			recv_buf + vmxferpage_packet->ranges[i].byte_offset);
782 		netvsc_packet->total_data_buflen =
783 					vmxferpage_packet->ranges[i].byte_count;
784 
785 		/* Pass it to the upper layer */
786 		rndis_filter_receive(device, netvsc_packet);
787 
788 		netvsc_receive_completion(netvsc_packet->
789 				completion.recv.recv_completion_ctx);
790 	}
791 
792 }
793 
794 static void netvsc_channel_cb(void *context)
795 {
796 	int ret;
797 	struct hv_device *device = context;
798 	struct netvsc_device *net_device;
799 	u32 bytes_recvd;
800 	u64 request_id;
801 	unsigned char *packet;
802 	struct vmpacket_descriptor *desc;
803 	unsigned char *buffer;
804 	int bufferlen = NETVSC_PACKET_SIZE;
805 	struct net_device *ndev;
806 
807 	packet = kzalloc(NETVSC_PACKET_SIZE * sizeof(unsigned char),
808 			 GFP_ATOMIC);
809 	if (!packet)
810 		return;
811 	buffer = packet;
812 
813 	net_device = get_inbound_net_device(device);
814 	if (!net_device)
815 		goto out;
816 	ndev = net_device->ndev;
817 
818 	do {
819 		ret = vmbus_recvpacket_raw(device->channel, buffer, bufferlen,
820 					   &bytes_recvd, &request_id);
821 		if (ret == 0) {
822 			if (bytes_recvd > 0) {
823 				desc = (struct vmpacket_descriptor *)buffer;
824 				switch (desc->type) {
825 				case VM_PKT_COMP:
826 					netvsc_send_completion(device, desc);
827 					break;
828 
829 				case VM_PKT_DATA_USING_XFER_PAGES:
830 					netvsc_receive(device, desc);
831 					break;
832 
833 				default:
834 					netdev_err(ndev,
835 						   "unhandled packet type %d, "
836 						   "tid %llx len %d\n",
837 						   desc->type, request_id,
838 						   bytes_recvd);
839 					break;
840 				}
841 
842 				/* reset */
843 				if (bufferlen > NETVSC_PACKET_SIZE) {
844 					kfree(buffer);
845 					buffer = packet;
846 					bufferlen = NETVSC_PACKET_SIZE;
847 				}
848 			} else {
849 				/* reset */
850 				if (bufferlen > NETVSC_PACKET_SIZE) {
851 					kfree(buffer);
852 					buffer = packet;
853 					bufferlen = NETVSC_PACKET_SIZE;
854 				}
855 
856 				break;
857 			}
858 		} else if (ret == -ENOBUFS) {
859 			/* Handle large packet */
860 			buffer = kmalloc(bytes_recvd, GFP_ATOMIC);
861 			if (buffer == NULL) {
862 				/* Try again next time around */
863 				netdev_err(ndev,
864 					   "unable to allocate buffer of size "
865 					   "(%d)!!\n", bytes_recvd);
866 				break;
867 			}
868 
869 			bufferlen = bytes_recvd;
870 		}
871 	} while (1);
872 
873 out:
874 	kfree(buffer);
875 	return;
876 }
877 
878 /*
879  * netvsc_device_add - Callback when the device belonging to this
880  * driver is added
881  */
882 int netvsc_device_add(struct hv_device *device, void *additional_info)
883 {
884 	int ret = 0;
885 	int i;
886 	int ring_size =
887 	((struct netvsc_device_info *)additional_info)->ring_size;
888 	struct netvsc_device *net_device;
889 	struct hv_netvsc_packet *packet, *pos;
890 	struct net_device *ndev;
891 
892 	net_device = alloc_net_device(device);
893 	if (!net_device) {
894 		ret = -ENOMEM;
895 		goto cleanup;
896 	}
897 
898 	/*
899 	 * Coming into this function, struct net_device * is
900 	 * registered as the driver private data.
901 	 * In alloc_net_device(), we register struct netvsc_device *
902 	 * as the driver private data and stash away struct net_device *
903 	 * in struct netvsc_device *.
904 	 */
905 	ndev = net_device->ndev;
906 
907 	/* Initialize the NetVSC channel extension */
908 	net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
909 	spin_lock_init(&net_device->recv_pkt_list_lock);
910 
911 	INIT_LIST_HEAD(&net_device->recv_pkt_list);
912 
913 	for (i = 0; i < NETVSC_RECEIVE_PACKETLIST_COUNT; i++) {
914 		packet = kzalloc(sizeof(struct hv_netvsc_packet) +
915 				 (NETVSC_RECEIVE_SG_COUNT *
916 				  sizeof(struct hv_page_buffer)), GFP_KERNEL);
917 		if (!packet)
918 			break;
919 
920 		list_add_tail(&packet->list_ent,
921 			      &net_device->recv_pkt_list);
922 	}
923 	init_completion(&net_device->channel_init_wait);
924 
925 	/* Open the channel */
926 	ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
927 			 ring_size * PAGE_SIZE, NULL, 0,
928 			 netvsc_channel_cb, device);
929 
930 	if (ret != 0) {
931 		netdev_err(ndev, "unable to open channel: %d\n", ret);
932 		goto cleanup;
933 	}
934 
935 	/* Channel is opened */
936 	pr_info("hv_netvsc channel opened successfully\n");
937 
938 	/* Connect with the NetVsp */
939 	ret = netvsc_connect_vsp(device);
940 	if (ret != 0) {
941 		netdev_err(ndev,
942 			"unable to connect to NetVSP - %d\n", ret);
943 		goto close;
944 	}
945 
946 	return ret;
947 
948 close:
949 	/* Now, we can close the channel safely */
950 	vmbus_close(device->channel);
951 
952 cleanup:
953 
954 	if (net_device) {
955 		list_for_each_entry_safe(packet, pos,
956 					 &net_device->recv_pkt_list,
957 					 list_ent) {
958 			list_del(&packet->list_ent);
959 			kfree(packet);
960 		}
961 
962 		kfree(net_device);
963 	}
964 
965 	return ret;
966 }
967