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
36static 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
54static 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
65static 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
78get_in_err:
79	return net_device;
80}
81
82
83static 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
153static 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
256cleanup:
257	netvsc_destroy_recv_buf(net_device);
258
259exit:
260	return ret;
261}
262
263
264/* Negotiate NVSP protocol version */
265static 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
313static 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
366cleanup:
367	return ret;
368}
369
370static 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 */
378int 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
431static void netvsc_send_completion(struct hv_device *device,
432				   struct vmpacket_descriptor *packet)
433{
434	struct netvsc_device *net_device;
435	struct nvsp_message *nvsp_packet;
436	struct hv_netvsc_packet *nvsc_packet;
437	struct net_device *ndev;
438
439	net_device = get_inbound_net_device(device);
440	if (!net_device)
441		return;
442	ndev = net_device->ndev;
443
444	nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
445			(packet->offset8 << 3));
446
447	if ((nvsp_packet->hdr.msg_type == NVSP_MSG_TYPE_INIT_COMPLETE) ||
448	    (nvsp_packet->hdr.msg_type ==
449	     NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE) ||
450	    (nvsp_packet->hdr.msg_type ==
451	     NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE)) {
452		/* Copy the response back */
453		memcpy(&net_device->channel_init_pkt, nvsp_packet,
454		       sizeof(struct nvsp_message));
455		complete(&net_device->channel_init_wait);
456	} else if (nvsp_packet->hdr.msg_type ==
457		   NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE) {
458		/* Get the send context */
459		nvsc_packet = (struct hv_netvsc_packet *)(unsigned long)
460			packet->trans_id;
461
462		/* Notify the layer above us */
463		nvsc_packet->completion.send.send_completion(
464			nvsc_packet->completion.send.send_completion_ctx);
465
466		atomic_dec(&net_device->num_outstanding_sends);
467
468		if (netif_queue_stopped(ndev) && !net_device->start_remove)
469			netif_wake_queue(ndev);
470	} else {
471		netdev_err(ndev, "Unknown send completion packet type- "
472			   "%d received!!\n", nvsp_packet->hdr.msg_type);
473	}
474
475}
476
477int netvsc_send(struct hv_device *device,
478			struct hv_netvsc_packet *packet)
479{
480	struct netvsc_device *net_device;
481	int ret = 0;
482	struct nvsp_message sendMessage;
483	struct net_device *ndev;
484
485	net_device = get_outbound_net_device(device);
486	if (!net_device)
487		return -ENODEV;
488	ndev = net_device->ndev;
489
490	sendMessage.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
491	if (packet->is_data_pkt) {
492		/* 0 is RMC_DATA; */
493		sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 0;
494	} else {
495		/* 1 is RMC_CONTROL; */
496		sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 1;
497	}
498
499	/* Not using send buffer section */
500	sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_index =
501		0xFFFFFFFF;
502	sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_size = 0;
503
504	if (packet->page_buf_cnt) {
505		ret = vmbus_sendpacket_pagebuffer(device->channel,
506						  packet->page_buf,
507						  packet->page_buf_cnt,
508						  &sendMessage,
509						  sizeof(struct nvsp_message),
510						  (unsigned long)packet);
511	} else {
512		ret = vmbus_sendpacket(device->channel, &sendMessage,
513				sizeof(struct nvsp_message),
514				(unsigned long)packet,
515				VM_PKT_DATA_INBAND,
516				VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
517
518	}
519
520	if (ret == 0) {
521		atomic_inc(&net_device->num_outstanding_sends);
522	} else if (ret == -EAGAIN) {
523		netif_stop_queue(ndev);
524		if (atomic_read(&net_device->num_outstanding_sends) < 1)
525			netif_wake_queue(ndev);
526	} else {
527		netdev_err(ndev, "Unable to send packet %p ret %d\n",
528			   packet, ret);
529	}
530
531	return ret;
532}
533
534static void netvsc_send_recv_completion(struct hv_device *device,
535					u64 transaction_id)
536{
537	struct nvsp_message recvcompMessage;
538	int retries = 0;
539	int ret;
540	struct net_device *ndev;
541	struct netvsc_device *net_device = hv_get_drvdata(device);
542
543	ndev = net_device->ndev;
544
545	recvcompMessage.hdr.msg_type =
546				NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
547
548	/* FIXME: Pass in the status */
549	recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status =
550		NVSP_STAT_SUCCESS;
551
552retry_send_cmplt:
553	/* Send the completion */
554	ret = vmbus_sendpacket(device->channel, &recvcompMessage,
555			       sizeof(struct nvsp_message), transaction_id,
556			       VM_PKT_COMP, 0);
557	if (ret == 0) {
558		/* success */
559		/* no-op */
560	} else if (ret == -EAGAIN) {
561		/* no more room...wait a bit and attempt to retry 3 times */
562		retries++;
563		netdev_err(ndev, "unable to send receive completion pkt"
564			" (tid %llx)...retrying %d\n", transaction_id, retries);
565
566		if (retries < 4) {
567			udelay(100);
568			goto retry_send_cmplt;
569		} else {
570			netdev_err(ndev, "unable to send receive "
571				"completion pkt (tid %llx)...give up retrying\n",
572				transaction_id);
573		}
574	} else {
575		netdev_err(ndev, "unable to send receive "
576			"completion pkt - %llx\n", transaction_id);
577	}
578}
579
580/* Send a receive completion packet to RNDIS device (ie NetVsp) */
581static void netvsc_receive_completion(void *context)
582{
583	struct hv_netvsc_packet *packet = context;
584	struct hv_device *device = (struct hv_device *)packet->device;
585	struct netvsc_device *net_device;
586	u64 transaction_id = 0;
587	bool fsend_receive_comp = false;
588	unsigned long flags;
589	struct net_device *ndev;
590
591	/*
592	 * Even though it seems logical to do a GetOutboundNetDevice() here to
593	 * send out receive completion, we are using GetInboundNetDevice()
594	 * since we may have disable outbound traffic already.
595	 */
596	net_device = get_inbound_net_device(device);
597	if (!net_device)
598		return;
599	ndev = net_device->ndev;
600
601	/* Overloading use of the lock. */
602	spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
603
604	packet->xfer_page_pkt->count--;
605
606	/*
607	 * Last one in the line that represent 1 xfer page packet.
608	 * Return the xfer page packet itself to the freelist
609	 */
610	if (packet->xfer_page_pkt->count == 0) {
611		fsend_receive_comp = true;
612		transaction_id = packet->completion.recv.recv_completion_tid;
613		list_add_tail(&packet->xfer_page_pkt->list_ent,
614			      &net_device->recv_pkt_list);
615
616	}
617
618	/* Put the packet back */
619	list_add_tail(&packet->list_ent, &net_device->recv_pkt_list);
620	spin_unlock_irqrestore(&net_device->recv_pkt_list_lock, flags);
621
622	/* Send a receive completion for the xfer page packet */
623	if (fsend_receive_comp)
624		netvsc_send_recv_completion(device, transaction_id);
625
626}
627
628static void netvsc_receive(struct hv_device *device,
629			    struct vmpacket_descriptor *packet)
630{
631	struct netvsc_device *net_device;
632	struct vmtransfer_page_packet_header *vmxferpage_packet;
633	struct nvsp_message *nvsp_packet;
634	struct hv_netvsc_packet *netvsc_packet = NULL;
635	/* struct netvsc_driver *netvscDriver; */
636	struct xferpage_packet *xferpage_packet = NULL;
637	int i;
638	int count = 0;
639	unsigned long flags;
640	struct net_device *ndev;
641
642	LIST_HEAD(listHead);
643
644	net_device = get_inbound_net_device(device);
645	if (!net_device)
646		return;
647	ndev = net_device->ndev;
648
649	/*
650	 * All inbound packets other than send completion should be xfer page
651	 * packet
652	 */
653	if (packet->type != VM_PKT_DATA_USING_XFER_PAGES) {
654		netdev_err(ndev, "Unknown packet type received - %d\n",
655			   packet->type);
656		return;
657	}
658
659	nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
660			(packet->offset8 << 3));
661
662	/* Make sure this is a valid nvsp packet */
663	if (nvsp_packet->hdr.msg_type !=
664	    NVSP_MSG1_TYPE_SEND_RNDIS_PKT) {
665		netdev_err(ndev, "Unknown nvsp packet type received-"
666			" %d\n", nvsp_packet->hdr.msg_type);
667		return;
668	}
669
670	vmxferpage_packet = (struct vmtransfer_page_packet_header *)packet;
671
672	if (vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID) {
673		netdev_err(ndev, "Invalid xfer page set id - "
674			   "expecting %x got %x\n", NETVSC_RECEIVE_BUFFER_ID,
675			   vmxferpage_packet->xfer_pageset_id);
676		return;
677	}
678
679	/*
680	 * Grab free packets (range count + 1) to represent this xfer
681	 * page packet. +1 to represent the xfer page packet itself.
682	 * We grab it here so that we know exactly how many we can
683	 * fulfil
684	 */
685	spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
686	while (!list_empty(&net_device->recv_pkt_list)) {
687		list_move_tail(net_device->recv_pkt_list.next, &listHead);
688		if (++count == vmxferpage_packet->range_cnt + 1)
689			break;
690	}
691	spin_unlock_irqrestore(&net_device->recv_pkt_list_lock, flags);
692
693	/*
694	 * We need at least 2 netvsc pkts (1 to represent the xfer
695	 * page and at least 1 for the range) i.e. we can handled
696	 * some of the xfer page packet ranges...
697	 */
698	if (count < 2) {
699		netdev_err(ndev, "Got only %d netvsc pkt...needed "
700			"%d pkts. Dropping this xfer page packet completely!\n",
701			count, vmxferpage_packet->range_cnt + 1);
702
703		/* Return it to the freelist */
704		spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
705		for (i = count; i != 0; i--) {
706			list_move_tail(listHead.next,
707				       &net_device->recv_pkt_list);
708		}
709		spin_unlock_irqrestore(&net_device->recv_pkt_list_lock,
710				       flags);
711
712		netvsc_send_recv_completion(device,
713					    vmxferpage_packet->d.trans_id);
714
715		return;
716	}
717
718	/* Remove the 1st packet to represent the xfer page packet itself */
719	xferpage_packet = (struct xferpage_packet *)listHead.next;
720	list_del(&xferpage_packet->list_ent);
721
722	/* This is how much we can satisfy */
723	xferpage_packet->count = count - 1;
724
725	if (xferpage_packet->count != vmxferpage_packet->range_cnt) {
726		netdev_err(ndev, "Needed %d netvsc pkts to satisfy "
727			"this xfer page...got %d\n",
728			vmxferpage_packet->range_cnt, xferpage_packet->count);
729	}
730
731	/* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
732	for (i = 0; i < (count - 1); i++) {
733		netvsc_packet = (struct hv_netvsc_packet *)listHead.next;
734		list_del(&netvsc_packet->list_ent);
735
736		/* Initialize the netvsc packet */
737		netvsc_packet->xfer_page_pkt = xferpage_packet;
738		netvsc_packet->completion.recv.recv_completion =
739					netvsc_receive_completion;
740		netvsc_packet->completion.recv.recv_completion_ctx =
741					netvsc_packet;
742		netvsc_packet->device = device;
743		/* Save this so that we can send it back */
744		netvsc_packet->completion.recv.recv_completion_tid =
745					vmxferpage_packet->d.trans_id;
746
747		netvsc_packet->data = (void *)((unsigned long)net_device->
748			recv_buf + vmxferpage_packet->ranges[i].byte_offset);
749		netvsc_packet->total_data_buflen =
750					vmxferpage_packet->ranges[i].byte_count;
751
752		/* Pass it to the upper layer */
753		rndis_filter_receive(device, netvsc_packet);
754
755		netvsc_receive_completion(netvsc_packet->
756				completion.recv.recv_completion_ctx);
757	}
758
759}
760
761static void netvsc_channel_cb(void *context)
762{
763	int ret;
764	struct hv_device *device = context;
765	struct netvsc_device *net_device;
766	u32 bytes_recvd;
767	u64 request_id;
768	unsigned char *packet;
769	struct vmpacket_descriptor *desc;
770	unsigned char *buffer;
771	int bufferlen = NETVSC_PACKET_SIZE;
772	struct net_device *ndev;
773
774	packet = kzalloc(NETVSC_PACKET_SIZE * sizeof(unsigned char),
775			 GFP_ATOMIC);
776	if (!packet)
777		return;
778	buffer = packet;
779
780	net_device = get_inbound_net_device(device);
781	if (!net_device)
782		goto out;
783	ndev = net_device->ndev;
784
785	do {
786		ret = vmbus_recvpacket_raw(device->channel, buffer, bufferlen,
787					   &bytes_recvd, &request_id);
788		if (ret == 0) {
789			if (bytes_recvd > 0) {
790				desc = (struct vmpacket_descriptor *)buffer;
791				switch (desc->type) {
792				case VM_PKT_COMP:
793					netvsc_send_completion(device, desc);
794					break;
795
796				case VM_PKT_DATA_USING_XFER_PAGES:
797					netvsc_receive(device, desc);
798					break;
799
800				default:
801					netdev_err(ndev,
802						   "unhandled packet type %d, "
803						   "tid %llx len %d\n",
804						   desc->type, request_id,
805						   bytes_recvd);
806					break;
807				}
808
809				/* reset */
810				if (bufferlen > NETVSC_PACKET_SIZE) {
811					kfree(buffer);
812					buffer = packet;
813					bufferlen = NETVSC_PACKET_SIZE;
814				}
815			} else {
816				/* reset */
817				if (bufferlen > NETVSC_PACKET_SIZE) {
818					kfree(buffer);
819					buffer = packet;
820					bufferlen = NETVSC_PACKET_SIZE;
821				}
822
823				break;
824			}
825		} else if (ret == -ENOBUFS) {
826			/* Handle large packet */
827			buffer = kmalloc(bytes_recvd, GFP_ATOMIC);
828			if (buffer == NULL) {
829				/* Try again next time around */
830				netdev_err(ndev,
831					   "unable to allocate buffer of size "
832					   "(%d)!!\n", bytes_recvd);
833				break;
834			}
835
836			bufferlen = bytes_recvd;
837		}
838	} while (1);
839
840out:
841	kfree(buffer);
842	return;
843}
844
845/*
846 * netvsc_device_add - Callback when the device belonging to this
847 * driver is added
848 */
849int netvsc_device_add(struct hv_device *device, void *additional_info)
850{
851	int ret = 0;
852	int i;
853	int ring_size =
854	((struct netvsc_device_info *)additional_info)->ring_size;
855	struct netvsc_device *net_device;
856	struct hv_netvsc_packet *packet, *pos;
857	struct net_device *ndev;
858
859	net_device = alloc_net_device(device);
860	if (!net_device) {
861		ret = -ENOMEM;
862		goto cleanup;
863	}
864
865	/*
866	 * Coming into this function, struct net_device * is
867	 * registered as the driver private data.
868	 * In alloc_net_device(), we register struct netvsc_device *
869	 * as the driver private data and stash away struct net_device *
870	 * in struct netvsc_device *.
871	 */
872	ndev = net_device->ndev;
873
874	/* Initialize the NetVSC channel extension */
875	net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
876	spin_lock_init(&net_device->recv_pkt_list_lock);
877
878	INIT_LIST_HEAD(&net_device->recv_pkt_list);
879
880	for (i = 0; i < NETVSC_RECEIVE_PACKETLIST_COUNT; i++) {
881		packet = kzalloc(sizeof(struct hv_netvsc_packet) +
882				 (NETVSC_RECEIVE_SG_COUNT *
883				  sizeof(struct hv_page_buffer)), GFP_KERNEL);
884		if (!packet)
885			break;
886
887		list_add_tail(&packet->list_ent,
888			      &net_device->recv_pkt_list);
889	}
890	init_completion(&net_device->channel_init_wait);
891
892	/* Open the channel */
893	ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
894			 ring_size * PAGE_SIZE, NULL, 0,
895			 netvsc_channel_cb, device);
896
897	if (ret != 0) {
898		netdev_err(ndev, "unable to open channel: %d\n", ret);
899		goto cleanup;
900	}
901
902	/* Channel is opened */
903	pr_info("hv_netvsc channel opened successfully\n");
904
905	/* Connect with the NetVsp */
906	ret = netvsc_connect_vsp(device);
907	if (ret != 0) {
908		netdev_err(ndev,
909			"unable to connect to NetVSP - %d\n", ret);
910		goto close;
911	}
912
913	return ret;
914
915close:
916	/* Now, we can close the channel safely */
917	vmbus_close(device->channel);
918
919cleanup:
920
921	if (net_device) {
922		list_for_each_entry_safe(packet, pos,
923					 &net_device->recv_pkt_list,
924					 list_ent) {
925			list_del(&packet->list_ent);
926			kfree(packet);
927		}
928
929		kfree(net_device);
930	}
931
932	return ret;
933}
934