xref: /linux/drivers/net/hyperv/netvsc.c (revision d91517839e5d95adc0cf4b28caa7af62a71de526)
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 		vfree(net_device->recv_buf);
140 		net_device->recv_buf = NULL;
141 	}
142 
143 	if (net_device->recv_section) {
144 		net_device->recv_section_cnt = 0;
145 		kfree(net_device->recv_section);
146 		net_device->recv_section = NULL;
147 	}
148 
149 	return ret;
150 }
151 
152 static int netvsc_init_recv_buf(struct hv_device *device)
153 {
154 	int ret = 0;
155 	int t;
156 	struct netvsc_device *net_device;
157 	struct nvsp_message *init_packet;
158 	struct net_device *ndev;
159 
160 	net_device = get_outbound_net_device(device);
161 	if (!net_device)
162 		return -ENODEV;
163 	ndev = net_device->ndev;
164 
165 	net_device->recv_buf = vzalloc(net_device->recv_buf_size);
166 	if (!net_device->recv_buf) {
167 		netdev_err(ndev, "unable to allocate receive "
168 			"buffer of size %d\n", net_device->recv_buf_size);
169 		ret = -ENOMEM;
170 		goto cleanup;
171 	}
172 
173 	/*
174 	 * Establish the gpadl handle for this buffer on this
175 	 * channel.  Note: This call uses the vmbus connection rather
176 	 * than the channel to establish the gpadl handle.
177 	 */
178 	ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
179 				    net_device->recv_buf_size,
180 				    &net_device->recv_buf_gpadl_handle);
181 	if (ret != 0) {
182 		netdev_err(ndev,
183 			"unable to establish receive buffer's gpadl\n");
184 		goto cleanup;
185 	}
186 
187 
188 	/* Notify the NetVsp of the gpadl handle */
189 	init_packet = &net_device->channel_init_pkt;
190 
191 	memset(init_packet, 0, sizeof(struct nvsp_message));
192 
193 	init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
194 	init_packet->msg.v1_msg.send_recv_buf.
195 		gpadl_handle = net_device->recv_buf_gpadl_handle;
196 	init_packet->msg.v1_msg.
197 		send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
198 
199 	/* Send the gpadl notification request */
200 	ret = vmbus_sendpacket(device->channel, init_packet,
201 			       sizeof(struct nvsp_message),
202 			       (unsigned long)init_packet,
203 			       VM_PKT_DATA_INBAND,
204 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
205 	if (ret != 0) {
206 		netdev_err(ndev,
207 			"unable to send receive buffer's gpadl to netvsp\n");
208 		goto cleanup;
209 	}
210 
211 	t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
212 	BUG_ON(t == 0);
213 
214 
215 	/* Check the response */
216 	if (init_packet->msg.v1_msg.
217 	    send_recv_buf_complete.status != NVSP_STAT_SUCCESS) {
218 		netdev_err(ndev, "Unable to complete receive buffer "
219 			   "initialization with NetVsp - status %d\n",
220 			   init_packet->msg.v1_msg.
221 			   send_recv_buf_complete.status);
222 		ret = -EINVAL;
223 		goto cleanup;
224 	}
225 
226 	/* Parse the response */
227 
228 	net_device->recv_section_cnt = init_packet->msg.
229 		v1_msg.send_recv_buf_complete.num_sections;
230 
231 	net_device->recv_section = kmemdup(
232 		init_packet->msg.v1_msg.send_recv_buf_complete.sections,
233 		net_device->recv_section_cnt *
234 		sizeof(struct nvsp_1_receive_buffer_section),
235 		GFP_KERNEL);
236 	if (net_device->recv_section == NULL) {
237 		ret = -EINVAL;
238 		goto cleanup;
239 	}
240 
241 	/*
242 	 * For 1st release, there should only be 1 section that represents the
243 	 * entire receive buffer
244 	 */
245 	if (net_device->recv_section_cnt != 1 ||
246 	    net_device->recv_section->offset != 0) {
247 		ret = -EINVAL;
248 		goto cleanup;
249 	}
250 
251 	goto exit;
252 
253 cleanup:
254 	netvsc_destroy_recv_buf(net_device);
255 
256 exit:
257 	return ret;
258 }
259 
260 
261 /* Negotiate NVSP protocol version */
262 static int negotiate_nvsp_ver(struct hv_device *device,
263 			      struct netvsc_device *net_device,
264 			      struct nvsp_message *init_packet,
265 			      u32 nvsp_ver)
266 {
267 	int ret, t;
268 
269 	memset(init_packet, 0, sizeof(struct nvsp_message));
270 	init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
271 	init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
272 	init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
273 
274 	/* Send the init request */
275 	ret = vmbus_sendpacket(device->channel, init_packet,
276 			       sizeof(struct nvsp_message),
277 			       (unsigned long)init_packet,
278 			       VM_PKT_DATA_INBAND,
279 			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
280 
281 	if (ret != 0)
282 		return ret;
283 
284 	t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
285 
286 	if (t == 0)
287 		return -ETIMEDOUT;
288 
289 	if (init_packet->msg.init_msg.init_complete.status !=
290 	    NVSP_STAT_SUCCESS)
291 		return -EINVAL;
292 
293 	if (nvsp_ver != NVSP_PROTOCOL_VERSION_2)
294 		return 0;
295 
296 	/* NVSPv2 only: Send NDIS config */
297 	memset(init_packet, 0, sizeof(struct nvsp_message));
298 	init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
299 	init_packet->msg.v2_msg.send_ndis_config.mtu = net_device->ndev->mtu;
300 	init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
301 
302 	ret = vmbus_sendpacket(device->channel, init_packet,
303 				sizeof(struct nvsp_message),
304 				(unsigned long)init_packet,
305 				VM_PKT_DATA_INBAND, 0);
306 
307 	return ret;
308 }
309 
310 static int netvsc_connect_vsp(struct hv_device *device)
311 {
312 	int ret;
313 	struct netvsc_device *net_device;
314 	struct nvsp_message *init_packet;
315 	int ndis_version;
316 	struct net_device *ndev;
317 
318 	net_device = get_outbound_net_device(device);
319 	if (!net_device)
320 		return -ENODEV;
321 	ndev = net_device->ndev;
322 
323 	init_packet = &net_device->channel_init_pkt;
324 
325 	/* Negotiate the latest NVSP protocol supported */
326 	if (negotiate_nvsp_ver(device, net_device, init_packet,
327 			       NVSP_PROTOCOL_VERSION_2) == 0) {
328 		net_device->nvsp_version = NVSP_PROTOCOL_VERSION_2;
329 	} else if (negotiate_nvsp_ver(device, net_device, init_packet,
330 				    NVSP_PROTOCOL_VERSION_1) == 0) {
331 		net_device->nvsp_version = NVSP_PROTOCOL_VERSION_1;
332 	} else {
333 		ret = -EPROTO;
334 		goto cleanup;
335 	}
336 
337 	pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
338 
339 	/* Send the ndis version */
340 	memset(init_packet, 0, sizeof(struct nvsp_message));
341 
342 	ndis_version = 0x00050001;
343 
344 	init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
345 	init_packet->msg.v1_msg.
346 		send_ndis_ver.ndis_major_ver =
347 				(ndis_version & 0xFFFF0000) >> 16;
348 	init_packet->msg.v1_msg.
349 		send_ndis_ver.ndis_minor_ver =
350 				ndis_version & 0xFFFF;
351 
352 	/* Send the init request */
353 	ret = vmbus_sendpacket(device->channel, init_packet,
354 				sizeof(struct nvsp_message),
355 				(unsigned long)init_packet,
356 				VM_PKT_DATA_INBAND, 0);
357 	if (ret != 0)
358 		goto cleanup;
359 
360 	/* Post the big receive buffer to NetVSP */
361 	ret = netvsc_init_recv_buf(device);
362 
363 cleanup:
364 	return ret;
365 }
366 
367 static void netvsc_disconnect_vsp(struct netvsc_device *net_device)
368 {
369 	netvsc_destroy_recv_buf(net_device);
370 }
371 
372 /*
373  * netvsc_device_remove - Callback when the root bus device is removed
374  */
375 int netvsc_device_remove(struct hv_device *device)
376 {
377 	struct netvsc_device *net_device;
378 	struct hv_netvsc_packet *netvsc_packet, *pos;
379 	unsigned long flags;
380 
381 	net_device = hv_get_drvdata(device);
382 
383 	netvsc_disconnect_vsp(net_device);
384 
385 	/*
386 	 * Since we have already drained, we don't need to busy wait
387 	 * as was done in final_release_stor_device()
388 	 * Note that we cannot set the ext pointer to NULL until
389 	 * we have drained - to drain the outgoing packets, we need to
390 	 * allow incoming packets.
391 	 */
392 
393 	spin_lock_irqsave(&device->channel->inbound_lock, flags);
394 	hv_set_drvdata(device, NULL);
395 	spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
396 
397 	/*
398 	 * At this point, no one should be accessing net_device
399 	 * except in here
400 	 */
401 	dev_notice(&device->device, "net device safe to remove\n");
402 
403 	/* Now, we can close the channel safely */
404 	vmbus_close(device->channel);
405 
406 	/* Release all resources */
407 	list_for_each_entry_safe(netvsc_packet, pos,
408 				 &net_device->recv_pkt_list, list_ent) {
409 		list_del(&netvsc_packet->list_ent);
410 		kfree(netvsc_packet);
411 	}
412 
413 	kfree(net_device);
414 	return 0;
415 }
416 
417 
418 #define RING_AVAIL_PERCENT_HIWATER 20
419 #define RING_AVAIL_PERCENT_LOWATER 10
420 
421 /*
422  * Get the percentage of available bytes to write in the ring.
423  * The return value is in range from 0 to 100.
424  */
425 static inline u32 hv_ringbuf_avail_percent(
426 		struct hv_ring_buffer_info *ring_info)
427 {
428 	u32 avail_read, avail_write;
429 
430 	hv_get_ringbuffer_availbytes(ring_info, &avail_read, &avail_write);
431 
432 	return avail_write * 100 / ring_info->ring_datasize;
433 }
434 
435 static void netvsc_send_completion(struct hv_device *device,
436 				   struct vmpacket_descriptor *packet)
437 {
438 	struct netvsc_device *net_device;
439 	struct nvsp_message *nvsp_packet;
440 	struct hv_netvsc_packet *nvsc_packet;
441 	struct net_device *ndev;
442 
443 	net_device = get_inbound_net_device(device);
444 	if (!net_device)
445 		return;
446 	ndev = net_device->ndev;
447 
448 	nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
449 			(packet->offset8 << 3));
450 
451 	if ((nvsp_packet->hdr.msg_type == NVSP_MSG_TYPE_INIT_COMPLETE) ||
452 	    (nvsp_packet->hdr.msg_type ==
453 	     NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE) ||
454 	    (nvsp_packet->hdr.msg_type ==
455 	     NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE)) {
456 		/* Copy the response back */
457 		memcpy(&net_device->channel_init_pkt, nvsp_packet,
458 		       sizeof(struct nvsp_message));
459 		complete(&net_device->channel_init_wait);
460 	} else if (nvsp_packet->hdr.msg_type ==
461 		   NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE) {
462 		int num_outstanding_sends;
463 
464 		/* Get the send context */
465 		nvsc_packet = (struct hv_netvsc_packet *)(unsigned long)
466 			packet->trans_id;
467 
468 		/* Notify the layer above us */
469 		if (nvsc_packet)
470 			nvsc_packet->completion.send.send_completion(
471 				nvsc_packet->completion.send.
472 				send_completion_ctx);
473 
474 		num_outstanding_sends =
475 			atomic_dec_return(&net_device->num_outstanding_sends);
476 
477 		if (net_device->destroy && num_outstanding_sends == 0)
478 			wake_up(&net_device->wait_drain);
479 
480 		if (netif_queue_stopped(ndev) && !net_device->start_remove &&
481 			(hv_ringbuf_avail_percent(&device->channel->outbound)
482 			> RING_AVAIL_PERCENT_HIWATER ||
483 			num_outstanding_sends < 1))
484 				netif_wake_queue(ndev);
485 	} else {
486 		netdev_err(ndev, "Unknown send completion packet type- "
487 			   "%d received!!\n", nvsp_packet->hdr.msg_type);
488 	}
489 
490 }
491 
492 int netvsc_send(struct hv_device *device,
493 			struct hv_netvsc_packet *packet)
494 {
495 	struct netvsc_device *net_device;
496 	int ret = 0;
497 	struct nvsp_message sendMessage;
498 	struct net_device *ndev;
499 	u64 req_id;
500 
501 	net_device = get_outbound_net_device(device);
502 	if (!net_device)
503 		return -ENODEV;
504 	ndev = net_device->ndev;
505 
506 	sendMessage.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
507 	if (packet->is_data_pkt) {
508 		/* 0 is RMC_DATA; */
509 		sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 0;
510 	} else {
511 		/* 1 is RMC_CONTROL; */
512 		sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 1;
513 	}
514 
515 	/* Not using send buffer section */
516 	sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_index =
517 		0xFFFFFFFF;
518 	sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_size = 0;
519 
520 	if (packet->completion.send.send_completion)
521 		req_id = (ulong)packet;
522 	else
523 		req_id = 0;
524 
525 	if (packet->page_buf_cnt) {
526 		ret = vmbus_sendpacket_pagebuffer(device->channel,
527 						  packet->page_buf,
528 						  packet->page_buf_cnt,
529 						  &sendMessage,
530 						  sizeof(struct nvsp_message),
531 						  req_id);
532 	} else {
533 		ret = vmbus_sendpacket(device->channel, &sendMessage,
534 				sizeof(struct nvsp_message),
535 				req_id,
536 				VM_PKT_DATA_INBAND,
537 				VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
538 	}
539 
540 	if (ret == 0) {
541 		atomic_inc(&net_device->num_outstanding_sends);
542 		if (hv_ringbuf_avail_percent(&device->channel->outbound) <
543 			RING_AVAIL_PERCENT_LOWATER) {
544 			netif_stop_queue(ndev);
545 			if (atomic_read(&net_device->
546 				num_outstanding_sends) < 1)
547 				netif_wake_queue(ndev);
548 		}
549 	} else if (ret == -EAGAIN) {
550 		netif_stop_queue(ndev);
551 		if (atomic_read(&net_device->num_outstanding_sends) < 1) {
552 			netif_wake_queue(ndev);
553 			ret = -ENOSPC;
554 		}
555 	} else {
556 		netdev_err(ndev, "Unable to send packet %p ret %d\n",
557 			   packet, ret);
558 	}
559 
560 	return ret;
561 }
562 
563 static void netvsc_send_recv_completion(struct hv_device *device,
564 					u64 transaction_id, u32 status)
565 {
566 	struct nvsp_message recvcompMessage;
567 	int retries = 0;
568 	int ret;
569 	struct net_device *ndev;
570 	struct netvsc_device *net_device = hv_get_drvdata(device);
571 
572 	ndev = net_device->ndev;
573 
574 	recvcompMessage.hdr.msg_type =
575 				NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
576 
577 	recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status = status;
578 
579 retry_send_cmplt:
580 	/* Send the completion */
581 	ret = vmbus_sendpacket(device->channel, &recvcompMessage,
582 			       sizeof(struct nvsp_message), transaction_id,
583 			       VM_PKT_COMP, 0);
584 	if (ret == 0) {
585 		/* success */
586 		/* no-op */
587 	} else if (ret == -EAGAIN) {
588 		/* no more room...wait a bit and attempt to retry 3 times */
589 		retries++;
590 		netdev_err(ndev, "unable to send receive completion pkt"
591 			" (tid %llx)...retrying %d\n", transaction_id, retries);
592 
593 		if (retries < 4) {
594 			udelay(100);
595 			goto retry_send_cmplt;
596 		} else {
597 			netdev_err(ndev, "unable to send receive "
598 				"completion pkt (tid %llx)...give up retrying\n",
599 				transaction_id);
600 		}
601 	} else {
602 		netdev_err(ndev, "unable to send receive "
603 			"completion pkt - %llx\n", transaction_id);
604 	}
605 }
606 
607 /* Send a receive completion packet to RNDIS device (ie NetVsp) */
608 static void netvsc_receive_completion(void *context)
609 {
610 	struct hv_netvsc_packet *packet = context;
611 	struct hv_device *device = packet->device;
612 	struct netvsc_device *net_device;
613 	u64 transaction_id = 0;
614 	bool fsend_receive_comp = false;
615 	unsigned long flags;
616 	struct net_device *ndev;
617 	u32 status = NVSP_STAT_NONE;
618 
619 	/*
620 	 * Even though it seems logical to do a GetOutboundNetDevice() here to
621 	 * send out receive completion, we are using GetInboundNetDevice()
622 	 * since we may have disable outbound traffic already.
623 	 */
624 	net_device = get_inbound_net_device(device);
625 	if (!net_device)
626 		return;
627 	ndev = net_device->ndev;
628 
629 	/* Overloading use of the lock. */
630 	spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
631 
632 	if (packet->status != NVSP_STAT_SUCCESS)
633 		packet->xfer_page_pkt->status = NVSP_STAT_FAIL;
634 
635 	packet->xfer_page_pkt->count--;
636 
637 	/*
638 	 * Last one in the line that represent 1 xfer page packet.
639 	 * Return the xfer page packet itself to the freelist
640 	 */
641 	if (packet->xfer_page_pkt->count == 0) {
642 		fsend_receive_comp = true;
643 		transaction_id = packet->completion.recv.recv_completion_tid;
644 		status = packet->xfer_page_pkt->status;
645 		list_add_tail(&packet->xfer_page_pkt->list_ent,
646 			      &net_device->recv_pkt_list);
647 
648 	}
649 
650 	/* Put the packet back */
651 	list_add_tail(&packet->list_ent, &net_device->recv_pkt_list);
652 	spin_unlock_irqrestore(&net_device->recv_pkt_list_lock, flags);
653 
654 	/* Send a receive completion for the xfer page packet */
655 	if (fsend_receive_comp)
656 		netvsc_send_recv_completion(device, transaction_id, status);
657 
658 }
659 
660 static void netvsc_receive(struct hv_device *device,
661 			    struct vmpacket_descriptor *packet)
662 {
663 	struct netvsc_device *net_device;
664 	struct vmtransfer_page_packet_header *vmxferpage_packet;
665 	struct nvsp_message *nvsp_packet;
666 	struct hv_netvsc_packet *netvsc_packet = NULL;
667 	/* struct netvsc_driver *netvscDriver; */
668 	struct xferpage_packet *xferpage_packet = NULL;
669 	int i;
670 	int count = 0;
671 	unsigned long flags;
672 	struct net_device *ndev;
673 
674 	LIST_HEAD(listHead);
675 
676 	net_device = get_inbound_net_device(device);
677 	if (!net_device)
678 		return;
679 	ndev = net_device->ndev;
680 
681 	/*
682 	 * All inbound packets other than send completion should be xfer page
683 	 * packet
684 	 */
685 	if (packet->type != VM_PKT_DATA_USING_XFER_PAGES) {
686 		netdev_err(ndev, "Unknown packet type received - %d\n",
687 			   packet->type);
688 		return;
689 	}
690 
691 	nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
692 			(packet->offset8 << 3));
693 
694 	/* Make sure this is a valid nvsp packet */
695 	if (nvsp_packet->hdr.msg_type !=
696 	    NVSP_MSG1_TYPE_SEND_RNDIS_PKT) {
697 		netdev_err(ndev, "Unknown nvsp packet type received-"
698 			" %d\n", nvsp_packet->hdr.msg_type);
699 		return;
700 	}
701 
702 	vmxferpage_packet = (struct vmtransfer_page_packet_header *)packet;
703 
704 	if (vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID) {
705 		netdev_err(ndev, "Invalid xfer page set id - "
706 			   "expecting %x got %x\n", NETVSC_RECEIVE_BUFFER_ID,
707 			   vmxferpage_packet->xfer_pageset_id);
708 		return;
709 	}
710 
711 	/*
712 	 * Grab free packets (range count + 1) to represent this xfer
713 	 * page packet. +1 to represent the xfer page packet itself.
714 	 * We grab it here so that we know exactly how many we can
715 	 * fulfil
716 	 */
717 	spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
718 	while (!list_empty(&net_device->recv_pkt_list)) {
719 		list_move_tail(net_device->recv_pkt_list.next, &listHead);
720 		if (++count == vmxferpage_packet->range_cnt + 1)
721 			break;
722 	}
723 	spin_unlock_irqrestore(&net_device->recv_pkt_list_lock, flags);
724 
725 	/*
726 	 * We need at least 2 netvsc pkts (1 to represent the xfer
727 	 * page and at least 1 for the range) i.e. we can handled
728 	 * some of the xfer page packet ranges...
729 	 */
730 	if (count < 2) {
731 		netdev_err(ndev, "Got only %d netvsc pkt...needed "
732 			"%d pkts. Dropping this xfer page packet completely!\n",
733 			count, vmxferpage_packet->range_cnt + 1);
734 
735 		/* Return it to the freelist */
736 		spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
737 		for (i = count; i != 0; i--) {
738 			list_move_tail(listHead.next,
739 				       &net_device->recv_pkt_list);
740 		}
741 		spin_unlock_irqrestore(&net_device->recv_pkt_list_lock,
742 				       flags);
743 
744 		netvsc_send_recv_completion(device,
745 					    vmxferpage_packet->d.trans_id,
746 					    NVSP_STAT_FAIL);
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 	xferpage_packet->status = NVSP_STAT_SUCCESS;
755 
756 	/* This is how much we can satisfy */
757 	xferpage_packet->count = count - 1;
758 
759 	if (xferpage_packet->count != vmxferpage_packet->range_cnt) {
760 		netdev_err(ndev, "Needed %d netvsc pkts to satisfy "
761 			"this xfer page...got %d\n",
762 			vmxferpage_packet->range_cnt, xferpage_packet->count);
763 	}
764 
765 	/* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
766 	for (i = 0; i < (count - 1); i++) {
767 		netvsc_packet = (struct hv_netvsc_packet *)listHead.next;
768 		list_del(&netvsc_packet->list_ent);
769 
770 		/* Initialize the netvsc packet */
771 		netvsc_packet->status = NVSP_STAT_SUCCESS;
772 		netvsc_packet->xfer_page_pkt = xferpage_packet;
773 		netvsc_packet->completion.recv.recv_completion =
774 					netvsc_receive_completion;
775 		netvsc_packet->completion.recv.recv_completion_ctx =
776 					netvsc_packet;
777 		netvsc_packet->device = device;
778 		/* Save this so that we can send it back */
779 		netvsc_packet->completion.recv.recv_completion_tid =
780 					vmxferpage_packet->d.trans_id;
781 
782 		netvsc_packet->data = (void *)((unsigned long)net_device->
783 			recv_buf + vmxferpage_packet->ranges[i].byte_offset);
784 		netvsc_packet->total_data_buflen =
785 					vmxferpage_packet->ranges[i].byte_count;
786 
787 		/* Pass it to the upper layer */
788 		rndis_filter_receive(device, netvsc_packet);
789 
790 		netvsc_receive_completion(netvsc_packet->
791 				completion.recv.recv_completion_ctx);
792 	}
793 
794 }
795 
796 static void netvsc_channel_cb(void *context)
797 {
798 	int ret;
799 	struct hv_device *device = context;
800 	struct netvsc_device *net_device;
801 	u32 bytes_recvd;
802 	u64 request_id;
803 	unsigned char *packet;
804 	struct vmpacket_descriptor *desc;
805 	unsigned char *buffer;
806 	int bufferlen = NETVSC_PACKET_SIZE;
807 	struct net_device *ndev;
808 
809 	packet = kzalloc(NETVSC_PACKET_SIZE * sizeof(unsigned char),
810 			 GFP_ATOMIC);
811 	if (!packet)
812 		return;
813 	buffer = packet;
814 
815 	net_device = get_inbound_net_device(device);
816 	if (!net_device)
817 		goto out;
818 	ndev = net_device->ndev;
819 
820 	do {
821 		ret = vmbus_recvpacket_raw(device->channel, buffer, bufferlen,
822 					   &bytes_recvd, &request_id);
823 		if (ret == 0) {
824 			if (bytes_recvd > 0) {
825 				desc = (struct vmpacket_descriptor *)buffer;
826 				switch (desc->type) {
827 				case VM_PKT_COMP:
828 					netvsc_send_completion(device, desc);
829 					break;
830 
831 				case VM_PKT_DATA_USING_XFER_PAGES:
832 					netvsc_receive(device, desc);
833 					break;
834 
835 				default:
836 					netdev_err(ndev,
837 						   "unhandled packet type %d, "
838 						   "tid %llx len %d\n",
839 						   desc->type, request_id,
840 						   bytes_recvd);
841 					break;
842 				}
843 
844 				/* reset */
845 				if (bufferlen > NETVSC_PACKET_SIZE) {
846 					kfree(buffer);
847 					buffer = packet;
848 					bufferlen = NETVSC_PACKET_SIZE;
849 				}
850 			} else {
851 				/* reset */
852 				if (bufferlen > NETVSC_PACKET_SIZE) {
853 					kfree(buffer);
854 					buffer = packet;
855 					bufferlen = NETVSC_PACKET_SIZE;
856 				}
857 
858 				break;
859 			}
860 		} else if (ret == -ENOBUFS) {
861 			/* Handle large packet */
862 			buffer = kmalloc(bytes_recvd, GFP_ATOMIC);
863 			if (buffer == NULL) {
864 				/* Try again next time around */
865 				netdev_err(ndev,
866 					   "unable to allocate buffer of size "
867 					   "(%d)!!\n", bytes_recvd);
868 				break;
869 			}
870 
871 			bufferlen = bytes_recvd;
872 		}
873 	} while (1);
874 
875 out:
876 	kfree(buffer);
877 	return;
878 }
879 
880 /*
881  * netvsc_device_add - Callback when the device belonging to this
882  * driver is added
883  */
884 int netvsc_device_add(struct hv_device *device, void *additional_info)
885 {
886 	int ret = 0;
887 	int i;
888 	int ring_size =
889 	((struct netvsc_device_info *)additional_info)->ring_size;
890 	struct netvsc_device *net_device;
891 	struct hv_netvsc_packet *packet, *pos;
892 	struct net_device *ndev;
893 
894 	net_device = alloc_net_device(device);
895 	if (!net_device) {
896 		ret = -ENOMEM;
897 		goto cleanup;
898 	}
899 
900 	/*
901 	 * Coming into this function, struct net_device * is
902 	 * registered as the driver private data.
903 	 * In alloc_net_device(), we register struct netvsc_device *
904 	 * as the driver private data and stash away struct net_device *
905 	 * in struct netvsc_device *.
906 	 */
907 	ndev = net_device->ndev;
908 
909 	/* Initialize the NetVSC channel extension */
910 	net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
911 	spin_lock_init(&net_device->recv_pkt_list_lock);
912 
913 	INIT_LIST_HEAD(&net_device->recv_pkt_list);
914 
915 	for (i = 0; i < NETVSC_RECEIVE_PACKETLIST_COUNT; i++) {
916 		packet = kzalloc(sizeof(struct hv_netvsc_packet), 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