xref: /linux/drivers/hv/channel.c (revision bf80eef2212a1e8451df13b52533f4bc31bb4f8e)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (c) 2009, Microsoft Corporation.
4  *
5  * Authors:
6  *   Haiyang Zhang <haiyangz@microsoft.com>
7  *   Hank Janssen  <hjanssen@microsoft.com>
8  */
9 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10 
11 #include <linux/kernel.h>
12 #include <linux/sched.h>
13 #include <linux/wait.h>
14 #include <linux/mm.h>
15 #include <linux/slab.h>
16 #include <linux/module.h>
17 #include <linux/hyperv.h>
18 #include <linux/uio.h>
19 #include <linux/interrupt.h>
20 #include <linux/set_memory.h>
21 #include <asm/page.h>
22 #include <asm/mshyperv.h>
23 
24 #include "hyperv_vmbus.h"
25 
26 /*
27  * hv_gpadl_size - Return the real size of a gpadl, the size that Hyper-V uses
28  *
29  * For BUFFER gpadl, Hyper-V uses the exact same size as the guest does.
30  *
31  * For RING gpadl, in each ring, the guest uses one PAGE_SIZE as the header
32  * (because of the alignment requirement), however, the hypervisor only
33  * uses the first HV_HYP_PAGE_SIZE as the header, therefore leaving a
34  * (PAGE_SIZE - HV_HYP_PAGE_SIZE) gap. And since there are two rings in a
35  * ringbuffer, the total size for a RING gpadl that Hyper-V uses is the
36  * total size that the guest uses minus twice of the gap size.
37  */
38 static inline u32 hv_gpadl_size(enum hv_gpadl_type type, u32 size)
39 {
40 	switch (type) {
41 	case HV_GPADL_BUFFER:
42 		return size;
43 	case HV_GPADL_RING:
44 		/* The size of a ringbuffer must be page-aligned */
45 		BUG_ON(size % PAGE_SIZE);
46 		/*
47 		 * Two things to notice here:
48 		 * 1) We're processing two ring buffers as a unit
49 		 * 2) We're skipping any space larger than HV_HYP_PAGE_SIZE in
50 		 * the first guest-size page of each of the two ring buffers.
51 		 * So we effectively subtract out two guest-size pages, and add
52 		 * back two Hyper-V size pages.
53 		 */
54 		return size - 2 * (PAGE_SIZE - HV_HYP_PAGE_SIZE);
55 	}
56 	BUG();
57 	return 0;
58 }
59 
60 /*
61  * hv_ring_gpadl_send_hvpgoffset - Calculate the send offset (in unit of
62  *                                 HV_HYP_PAGE) in a ring gpadl based on the
63  *                                 offset in the guest
64  *
65  * @offset: the offset (in bytes) where the send ringbuffer starts in the
66  *               virtual address space of the guest
67  */
68 static inline u32 hv_ring_gpadl_send_hvpgoffset(u32 offset)
69 {
70 
71 	/*
72 	 * For RING gpadl, in each ring, the guest uses one PAGE_SIZE as the
73 	 * header (because of the alignment requirement), however, the
74 	 * hypervisor only uses the first HV_HYP_PAGE_SIZE as the header,
75 	 * therefore leaving a (PAGE_SIZE - HV_HYP_PAGE_SIZE) gap.
76 	 *
77 	 * And to calculate the effective send offset in gpadl, we need to
78 	 * substract this gap.
79 	 */
80 	return (offset - (PAGE_SIZE - HV_HYP_PAGE_SIZE)) >> HV_HYP_PAGE_SHIFT;
81 }
82 
83 /*
84  * hv_gpadl_hvpfn - Return the Hyper-V page PFN of the @i th Hyper-V page in
85  *                  the gpadl
86  *
87  * @type: the type of the gpadl
88  * @kbuffer: the pointer to the gpadl in the guest
89  * @size: the total size (in bytes) of the gpadl
90  * @send_offset: the offset (in bytes) where the send ringbuffer starts in the
91  *               virtual address space of the guest
92  * @i: the index
93  */
94 static inline u64 hv_gpadl_hvpfn(enum hv_gpadl_type type, void *kbuffer,
95 				 u32 size, u32 send_offset, int i)
96 {
97 	int send_idx = hv_ring_gpadl_send_hvpgoffset(send_offset);
98 	unsigned long delta = 0UL;
99 
100 	switch (type) {
101 	case HV_GPADL_BUFFER:
102 		break;
103 	case HV_GPADL_RING:
104 		if (i == 0)
105 			delta = 0;
106 		else if (i <= send_idx)
107 			delta = PAGE_SIZE - HV_HYP_PAGE_SIZE;
108 		else
109 			delta = 2 * (PAGE_SIZE - HV_HYP_PAGE_SIZE);
110 		break;
111 	default:
112 		BUG();
113 		break;
114 	}
115 
116 	return virt_to_hvpfn(kbuffer + delta + (HV_HYP_PAGE_SIZE * i));
117 }
118 
119 /*
120  * vmbus_setevent- Trigger an event notification on the specified
121  * channel.
122  */
123 void vmbus_setevent(struct vmbus_channel *channel)
124 {
125 	struct hv_monitor_page *monitorpage;
126 
127 	trace_vmbus_setevent(channel);
128 
129 	/*
130 	 * For channels marked as in "low latency" mode
131 	 * bypass the monitor page mechanism.
132 	 */
133 	if (channel->offermsg.monitor_allocated && !channel->low_latency) {
134 		vmbus_send_interrupt(channel->offermsg.child_relid);
135 
136 		/* Get the child to parent monitor page */
137 		monitorpage = vmbus_connection.monitor_pages[1];
138 
139 		sync_set_bit(channel->monitor_bit,
140 			(unsigned long *)&monitorpage->trigger_group
141 					[channel->monitor_grp].pending);
142 
143 	} else {
144 		vmbus_set_event(channel);
145 	}
146 }
147 EXPORT_SYMBOL_GPL(vmbus_setevent);
148 
149 /* vmbus_free_ring - drop mapping of ring buffer */
150 void vmbus_free_ring(struct vmbus_channel *channel)
151 {
152 	hv_ringbuffer_cleanup(&channel->outbound);
153 	hv_ringbuffer_cleanup(&channel->inbound);
154 
155 	if (channel->ringbuffer_page) {
156 		__free_pages(channel->ringbuffer_page,
157 			     get_order(channel->ringbuffer_pagecount
158 				       << PAGE_SHIFT));
159 		channel->ringbuffer_page = NULL;
160 	}
161 }
162 EXPORT_SYMBOL_GPL(vmbus_free_ring);
163 
164 /* vmbus_alloc_ring - allocate and map pages for ring buffer */
165 int vmbus_alloc_ring(struct vmbus_channel *newchannel,
166 		     u32 send_size, u32 recv_size)
167 {
168 	struct page *page;
169 	int order;
170 
171 	if (send_size % PAGE_SIZE || recv_size % PAGE_SIZE)
172 		return -EINVAL;
173 
174 	/* Allocate the ring buffer */
175 	order = get_order(send_size + recv_size);
176 	page = alloc_pages_node(cpu_to_node(newchannel->target_cpu),
177 				GFP_KERNEL|__GFP_ZERO, order);
178 
179 	if (!page)
180 		page = alloc_pages(GFP_KERNEL|__GFP_ZERO, order);
181 
182 	if (!page)
183 		return -ENOMEM;
184 
185 	newchannel->ringbuffer_page = page;
186 	newchannel->ringbuffer_pagecount = (send_size + recv_size) >> PAGE_SHIFT;
187 	newchannel->ringbuffer_send_offset = send_size >> PAGE_SHIFT;
188 
189 	return 0;
190 }
191 EXPORT_SYMBOL_GPL(vmbus_alloc_ring);
192 
193 /* Used for Hyper-V Socket: a guest client's connect() to the host */
194 int vmbus_send_tl_connect_request(const guid_t *shv_guest_servie_id,
195 				  const guid_t *shv_host_servie_id)
196 {
197 	struct vmbus_channel_tl_connect_request conn_msg;
198 	int ret;
199 
200 	memset(&conn_msg, 0, sizeof(conn_msg));
201 	conn_msg.header.msgtype = CHANNELMSG_TL_CONNECT_REQUEST;
202 	conn_msg.guest_endpoint_id = *shv_guest_servie_id;
203 	conn_msg.host_service_id = *shv_host_servie_id;
204 
205 	ret = vmbus_post_msg(&conn_msg, sizeof(conn_msg), true);
206 
207 	trace_vmbus_send_tl_connect_request(&conn_msg, ret);
208 
209 	return ret;
210 }
211 EXPORT_SYMBOL_GPL(vmbus_send_tl_connect_request);
212 
213 static int send_modifychannel_without_ack(struct vmbus_channel *channel, u32 target_vp)
214 {
215 	struct vmbus_channel_modifychannel msg;
216 	int ret;
217 
218 	memset(&msg, 0, sizeof(msg));
219 	msg.header.msgtype = CHANNELMSG_MODIFYCHANNEL;
220 	msg.child_relid = channel->offermsg.child_relid;
221 	msg.target_vp = target_vp;
222 
223 	ret = vmbus_post_msg(&msg, sizeof(msg), true);
224 	trace_vmbus_send_modifychannel(&msg, ret);
225 
226 	return ret;
227 }
228 
229 static int send_modifychannel_with_ack(struct vmbus_channel *channel, u32 target_vp)
230 {
231 	struct vmbus_channel_modifychannel *msg;
232 	struct vmbus_channel_msginfo *info;
233 	unsigned long flags;
234 	int ret;
235 
236 	info = kzalloc(sizeof(struct vmbus_channel_msginfo) +
237 				sizeof(struct vmbus_channel_modifychannel),
238 		       GFP_KERNEL);
239 	if (!info)
240 		return -ENOMEM;
241 
242 	init_completion(&info->waitevent);
243 	info->waiting_channel = channel;
244 
245 	msg = (struct vmbus_channel_modifychannel *)info->msg;
246 	msg->header.msgtype = CHANNELMSG_MODIFYCHANNEL;
247 	msg->child_relid = channel->offermsg.child_relid;
248 	msg->target_vp = target_vp;
249 
250 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
251 	list_add_tail(&info->msglistentry, &vmbus_connection.chn_msg_list);
252 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
253 
254 	ret = vmbus_post_msg(msg, sizeof(*msg), true);
255 	trace_vmbus_send_modifychannel(msg, ret);
256 	if (ret != 0) {
257 		spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
258 		list_del(&info->msglistentry);
259 		spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
260 		goto free_info;
261 	}
262 
263 	/*
264 	 * Release channel_mutex; otherwise, vmbus_onoffer_rescind() could block on
265 	 * the mutex and be unable to signal the completion.
266 	 *
267 	 * See the caller target_cpu_store() for information about the usage of the
268 	 * mutex.
269 	 */
270 	mutex_unlock(&vmbus_connection.channel_mutex);
271 	wait_for_completion(&info->waitevent);
272 	mutex_lock(&vmbus_connection.channel_mutex);
273 
274 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
275 	list_del(&info->msglistentry);
276 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
277 
278 	if (info->response.modify_response.status)
279 		ret = -EAGAIN;
280 
281 free_info:
282 	kfree(info);
283 	return ret;
284 }
285 
286 /*
287  * Set/change the vCPU (@target_vp) the channel (@child_relid) will interrupt.
288  *
289  * CHANNELMSG_MODIFYCHANNEL messages are aynchronous.  When VMbus version 5.3
290  * or later is negotiated, Hyper-V always sends an ACK in response to such a
291  * message.  For VMbus version 5.2 and earlier, it never sends an ACK.  With-
292  * out an ACK, we can not know when the host will stop interrupting the "old"
293  * vCPU and start interrupting the "new" vCPU for the given channel.
294  *
295  * The CHANNELMSG_MODIFYCHANNEL message type is supported since VMBus version
296  * VERSION_WIN10_V4_1.
297  */
298 int vmbus_send_modifychannel(struct vmbus_channel *channel, u32 target_vp)
299 {
300 	if (vmbus_proto_version >= VERSION_WIN10_V5_3)
301 		return send_modifychannel_with_ack(channel, target_vp);
302 	return send_modifychannel_without_ack(channel, target_vp);
303 }
304 EXPORT_SYMBOL_GPL(vmbus_send_modifychannel);
305 
306 /*
307  * create_gpadl_header - Creates a gpadl for the specified buffer
308  */
309 static int create_gpadl_header(enum hv_gpadl_type type, void *kbuffer,
310 			       u32 size, u32 send_offset,
311 			       struct vmbus_channel_msginfo **msginfo)
312 {
313 	int i;
314 	int pagecount;
315 	struct vmbus_channel_gpadl_header *gpadl_header;
316 	struct vmbus_channel_gpadl_body *gpadl_body;
317 	struct vmbus_channel_msginfo *msgheader;
318 	struct vmbus_channel_msginfo *msgbody = NULL;
319 	u32 msgsize;
320 
321 	int pfnsum, pfncount, pfnleft, pfncurr, pfnsize;
322 
323 	pagecount = hv_gpadl_size(type, size) >> HV_HYP_PAGE_SHIFT;
324 
325 	/* do we need a gpadl body msg */
326 	pfnsize = MAX_SIZE_CHANNEL_MESSAGE -
327 		  sizeof(struct vmbus_channel_gpadl_header) -
328 		  sizeof(struct gpa_range);
329 	pfncount = pfnsize / sizeof(u64);
330 
331 	if (pagecount > pfncount) {
332 		/* we need a gpadl body */
333 		/* fill in the header */
334 		msgsize = sizeof(struct vmbus_channel_msginfo) +
335 			  sizeof(struct vmbus_channel_gpadl_header) +
336 			  sizeof(struct gpa_range) + pfncount * sizeof(u64);
337 		msgheader =  kzalloc(msgsize, GFP_KERNEL);
338 		if (!msgheader)
339 			goto nomem;
340 
341 		INIT_LIST_HEAD(&msgheader->submsglist);
342 		msgheader->msgsize = msgsize;
343 
344 		gpadl_header = (struct vmbus_channel_gpadl_header *)
345 			msgheader->msg;
346 		gpadl_header->rangecount = 1;
347 		gpadl_header->range_buflen = sizeof(struct gpa_range) +
348 					 pagecount * sizeof(u64);
349 		gpadl_header->range[0].byte_offset = 0;
350 		gpadl_header->range[0].byte_count = hv_gpadl_size(type, size);
351 		for (i = 0; i < pfncount; i++)
352 			gpadl_header->range[0].pfn_array[i] = hv_gpadl_hvpfn(
353 				type, kbuffer, size, send_offset, i);
354 		*msginfo = msgheader;
355 
356 		pfnsum = pfncount;
357 		pfnleft = pagecount - pfncount;
358 
359 		/* how many pfns can we fit */
360 		pfnsize = MAX_SIZE_CHANNEL_MESSAGE -
361 			  sizeof(struct vmbus_channel_gpadl_body);
362 		pfncount = pfnsize / sizeof(u64);
363 
364 		/* fill in the body */
365 		while (pfnleft) {
366 			if (pfnleft > pfncount)
367 				pfncurr = pfncount;
368 			else
369 				pfncurr = pfnleft;
370 
371 			msgsize = sizeof(struct vmbus_channel_msginfo) +
372 				  sizeof(struct vmbus_channel_gpadl_body) +
373 				  pfncurr * sizeof(u64);
374 			msgbody = kzalloc(msgsize, GFP_KERNEL);
375 
376 			if (!msgbody) {
377 				struct vmbus_channel_msginfo *pos = NULL;
378 				struct vmbus_channel_msginfo *tmp = NULL;
379 				/*
380 				 * Free up all the allocated messages.
381 				 */
382 				list_for_each_entry_safe(pos, tmp,
383 					&msgheader->submsglist,
384 					msglistentry) {
385 
386 					list_del(&pos->msglistentry);
387 					kfree(pos);
388 				}
389 
390 				goto nomem;
391 			}
392 
393 			msgbody->msgsize = msgsize;
394 			gpadl_body =
395 				(struct vmbus_channel_gpadl_body *)msgbody->msg;
396 
397 			/*
398 			 * Gpadl is u32 and we are using a pointer which could
399 			 * be 64-bit
400 			 * This is governed by the guest/host protocol and
401 			 * so the hypervisor guarantees that this is ok.
402 			 */
403 			for (i = 0; i < pfncurr; i++)
404 				gpadl_body->pfn[i] = hv_gpadl_hvpfn(type,
405 					kbuffer, size, send_offset, pfnsum + i);
406 
407 			/* add to msg header */
408 			list_add_tail(&msgbody->msglistentry,
409 				      &msgheader->submsglist);
410 			pfnsum += pfncurr;
411 			pfnleft -= pfncurr;
412 		}
413 	} else {
414 		/* everything fits in a header */
415 		msgsize = sizeof(struct vmbus_channel_msginfo) +
416 			  sizeof(struct vmbus_channel_gpadl_header) +
417 			  sizeof(struct gpa_range) + pagecount * sizeof(u64);
418 		msgheader = kzalloc(msgsize, GFP_KERNEL);
419 		if (msgheader == NULL)
420 			goto nomem;
421 
422 		INIT_LIST_HEAD(&msgheader->submsglist);
423 		msgheader->msgsize = msgsize;
424 
425 		gpadl_header = (struct vmbus_channel_gpadl_header *)
426 			msgheader->msg;
427 		gpadl_header->rangecount = 1;
428 		gpadl_header->range_buflen = sizeof(struct gpa_range) +
429 					 pagecount * sizeof(u64);
430 		gpadl_header->range[0].byte_offset = 0;
431 		gpadl_header->range[0].byte_count = hv_gpadl_size(type, size);
432 		for (i = 0; i < pagecount; i++)
433 			gpadl_header->range[0].pfn_array[i] = hv_gpadl_hvpfn(
434 				type, kbuffer, size, send_offset, i);
435 
436 		*msginfo = msgheader;
437 	}
438 
439 	return 0;
440 nomem:
441 	kfree(msgheader);
442 	kfree(msgbody);
443 	return -ENOMEM;
444 }
445 
446 /*
447  * __vmbus_establish_gpadl - Establish a GPADL for a buffer or ringbuffer
448  *
449  * @channel: a channel
450  * @type: the type of the corresponding GPADL, only meaningful for the guest.
451  * @kbuffer: from kmalloc or vmalloc
452  * @size: page-size multiple
453  * @send_offset: the offset (in bytes) where the send ring buffer starts,
454  *              should be 0 for BUFFER type gpadl
455  * @gpadl_handle: some funky thing
456  */
457 static int __vmbus_establish_gpadl(struct vmbus_channel *channel,
458 				   enum hv_gpadl_type type, void *kbuffer,
459 				   u32 size, u32 send_offset,
460 				   struct vmbus_gpadl *gpadl)
461 {
462 	struct vmbus_channel_gpadl_header *gpadlmsg;
463 	struct vmbus_channel_gpadl_body *gpadl_body;
464 	struct vmbus_channel_msginfo *msginfo = NULL;
465 	struct vmbus_channel_msginfo *submsginfo, *tmp;
466 	struct list_head *curr;
467 	u32 next_gpadl_handle;
468 	unsigned long flags;
469 	int ret = 0;
470 
471 	next_gpadl_handle =
472 		(atomic_inc_return(&vmbus_connection.next_gpadl_handle) - 1);
473 
474 	ret = create_gpadl_header(type, kbuffer, size, send_offset, &msginfo);
475 	if (ret)
476 		return ret;
477 
478 	ret = set_memory_decrypted((unsigned long)kbuffer,
479 				   PFN_UP(size));
480 	if (ret) {
481 		dev_warn(&channel->device_obj->device,
482 			 "Failed to set host visibility for new GPADL %d.\n",
483 			 ret);
484 		return ret;
485 	}
486 
487 	init_completion(&msginfo->waitevent);
488 	msginfo->waiting_channel = channel;
489 
490 	gpadlmsg = (struct vmbus_channel_gpadl_header *)msginfo->msg;
491 	gpadlmsg->header.msgtype = CHANNELMSG_GPADL_HEADER;
492 	gpadlmsg->child_relid = channel->offermsg.child_relid;
493 	gpadlmsg->gpadl = next_gpadl_handle;
494 
495 
496 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
497 	list_add_tail(&msginfo->msglistentry,
498 		      &vmbus_connection.chn_msg_list);
499 
500 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
501 
502 	if (channel->rescind) {
503 		ret = -ENODEV;
504 		goto cleanup;
505 	}
506 
507 	ret = vmbus_post_msg(gpadlmsg, msginfo->msgsize -
508 			     sizeof(*msginfo), true);
509 
510 	trace_vmbus_establish_gpadl_header(gpadlmsg, ret);
511 
512 	if (ret != 0)
513 		goto cleanup;
514 
515 	list_for_each(curr, &msginfo->submsglist) {
516 		submsginfo = (struct vmbus_channel_msginfo *)curr;
517 		gpadl_body =
518 			(struct vmbus_channel_gpadl_body *)submsginfo->msg;
519 
520 		gpadl_body->header.msgtype =
521 			CHANNELMSG_GPADL_BODY;
522 		gpadl_body->gpadl = next_gpadl_handle;
523 
524 		ret = vmbus_post_msg(gpadl_body,
525 				     submsginfo->msgsize - sizeof(*submsginfo),
526 				     true);
527 
528 		trace_vmbus_establish_gpadl_body(gpadl_body, ret);
529 
530 		if (ret != 0)
531 			goto cleanup;
532 
533 	}
534 	wait_for_completion(&msginfo->waitevent);
535 
536 	if (msginfo->response.gpadl_created.creation_status != 0) {
537 		pr_err("Failed to establish GPADL: err = 0x%x\n",
538 		       msginfo->response.gpadl_created.creation_status);
539 
540 		ret = -EDQUOT;
541 		goto cleanup;
542 	}
543 
544 	if (channel->rescind) {
545 		ret = -ENODEV;
546 		goto cleanup;
547 	}
548 
549 	/* At this point, we received the gpadl created msg */
550 	gpadl->gpadl_handle = gpadlmsg->gpadl;
551 	gpadl->buffer = kbuffer;
552 	gpadl->size = size;
553 
554 
555 cleanup:
556 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
557 	list_del(&msginfo->msglistentry);
558 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
559 	list_for_each_entry_safe(submsginfo, tmp, &msginfo->submsglist,
560 				 msglistentry) {
561 		kfree(submsginfo);
562 	}
563 
564 	kfree(msginfo);
565 
566 	if (ret)
567 		set_memory_encrypted((unsigned long)kbuffer,
568 				     PFN_UP(size));
569 
570 	return ret;
571 }
572 
573 /*
574  * vmbus_establish_gpadl - Establish a GPADL for the specified buffer
575  *
576  * @channel: a channel
577  * @kbuffer: from kmalloc or vmalloc
578  * @size: page-size multiple
579  * @gpadl_handle: some funky thing
580  */
581 int vmbus_establish_gpadl(struct vmbus_channel *channel, void *kbuffer,
582 			  u32 size, struct vmbus_gpadl *gpadl)
583 {
584 	return __vmbus_establish_gpadl(channel, HV_GPADL_BUFFER, kbuffer, size,
585 				       0U, gpadl);
586 }
587 EXPORT_SYMBOL_GPL(vmbus_establish_gpadl);
588 
589 /**
590  * request_arr_init - Allocates memory for the requestor array. Each slot
591  * keeps track of the next available slot in the array. Initially, each
592  * slot points to the next one (as in a Linked List). The last slot
593  * does not point to anything, so its value is U64_MAX by default.
594  * @size The size of the array
595  */
596 static u64 *request_arr_init(u32 size)
597 {
598 	int i;
599 	u64 *req_arr;
600 
601 	req_arr = kcalloc(size, sizeof(u64), GFP_KERNEL);
602 	if (!req_arr)
603 		return NULL;
604 
605 	for (i = 0; i < size - 1; i++)
606 		req_arr[i] = i + 1;
607 
608 	/* Last slot (no more available slots) */
609 	req_arr[i] = U64_MAX;
610 
611 	return req_arr;
612 }
613 
614 /*
615  * vmbus_alloc_requestor - Initializes @rqstor's fields.
616  * Index 0 is the first free slot
617  * @size: Size of the requestor array
618  */
619 static int vmbus_alloc_requestor(struct vmbus_requestor *rqstor, u32 size)
620 {
621 	u64 *rqst_arr;
622 	unsigned long *bitmap;
623 
624 	rqst_arr = request_arr_init(size);
625 	if (!rqst_arr)
626 		return -ENOMEM;
627 
628 	bitmap = bitmap_zalloc(size, GFP_KERNEL);
629 	if (!bitmap) {
630 		kfree(rqst_arr);
631 		return -ENOMEM;
632 	}
633 
634 	rqstor->req_arr = rqst_arr;
635 	rqstor->req_bitmap = bitmap;
636 	rqstor->size = size;
637 	rqstor->next_request_id = 0;
638 	spin_lock_init(&rqstor->req_lock);
639 
640 	return 0;
641 }
642 
643 /*
644  * vmbus_free_requestor - Frees memory allocated for @rqstor
645  * @rqstor: Pointer to the requestor struct
646  */
647 static void vmbus_free_requestor(struct vmbus_requestor *rqstor)
648 {
649 	kfree(rqstor->req_arr);
650 	bitmap_free(rqstor->req_bitmap);
651 }
652 
653 static int __vmbus_open(struct vmbus_channel *newchannel,
654 		       void *userdata, u32 userdatalen,
655 		       void (*onchannelcallback)(void *context), void *context)
656 {
657 	struct vmbus_channel_open_channel *open_msg;
658 	struct vmbus_channel_msginfo *open_info = NULL;
659 	struct page *page = newchannel->ringbuffer_page;
660 	u32 send_pages, recv_pages;
661 	unsigned long flags;
662 	int err;
663 
664 	if (userdatalen > MAX_USER_DEFINED_BYTES)
665 		return -EINVAL;
666 
667 	send_pages = newchannel->ringbuffer_send_offset;
668 	recv_pages = newchannel->ringbuffer_pagecount - send_pages;
669 
670 	if (newchannel->state != CHANNEL_OPEN_STATE)
671 		return -EINVAL;
672 
673 	/* Create and init requestor */
674 	if (newchannel->rqstor_size) {
675 		if (vmbus_alloc_requestor(&newchannel->requestor, newchannel->rqstor_size))
676 			return -ENOMEM;
677 	}
678 
679 	newchannel->state = CHANNEL_OPENING_STATE;
680 	newchannel->onchannel_callback = onchannelcallback;
681 	newchannel->channel_callback_context = context;
682 
683 	if (!newchannel->max_pkt_size)
684 		newchannel->max_pkt_size = VMBUS_DEFAULT_MAX_PKT_SIZE;
685 
686 	/* Establish the gpadl for the ring buffer */
687 	newchannel->ringbuffer_gpadlhandle.gpadl_handle = 0;
688 
689 	err = __vmbus_establish_gpadl(newchannel, HV_GPADL_RING,
690 				      page_address(newchannel->ringbuffer_page),
691 				      (send_pages + recv_pages) << PAGE_SHIFT,
692 				      newchannel->ringbuffer_send_offset << PAGE_SHIFT,
693 				      &newchannel->ringbuffer_gpadlhandle);
694 	if (err)
695 		goto error_clean_ring;
696 
697 	err = hv_ringbuffer_init(&newchannel->outbound,
698 				 page, send_pages, 0);
699 	if (err)
700 		goto error_free_gpadl;
701 
702 	err = hv_ringbuffer_init(&newchannel->inbound, &page[send_pages],
703 				 recv_pages, newchannel->max_pkt_size);
704 	if (err)
705 		goto error_free_gpadl;
706 
707 	/* Create and init the channel open message */
708 	open_info = kzalloc(sizeof(*open_info) +
709 			   sizeof(struct vmbus_channel_open_channel),
710 			   GFP_KERNEL);
711 	if (!open_info) {
712 		err = -ENOMEM;
713 		goto error_free_gpadl;
714 	}
715 
716 	init_completion(&open_info->waitevent);
717 	open_info->waiting_channel = newchannel;
718 
719 	open_msg = (struct vmbus_channel_open_channel *)open_info->msg;
720 	open_msg->header.msgtype = CHANNELMSG_OPENCHANNEL;
721 	open_msg->openid = newchannel->offermsg.child_relid;
722 	open_msg->child_relid = newchannel->offermsg.child_relid;
723 	open_msg->ringbuffer_gpadlhandle
724 		= newchannel->ringbuffer_gpadlhandle.gpadl_handle;
725 	/*
726 	 * The unit of ->downstream_ringbuffer_pageoffset is HV_HYP_PAGE and
727 	 * the unit of ->ringbuffer_send_offset (i.e. send_pages) is PAGE, so
728 	 * here we calculate it into HV_HYP_PAGE.
729 	 */
730 	open_msg->downstream_ringbuffer_pageoffset =
731 		hv_ring_gpadl_send_hvpgoffset(send_pages << PAGE_SHIFT);
732 	open_msg->target_vp = hv_cpu_number_to_vp_number(newchannel->target_cpu);
733 
734 	if (userdatalen)
735 		memcpy(open_msg->userdata, userdata, userdatalen);
736 
737 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
738 	list_add_tail(&open_info->msglistentry,
739 		      &vmbus_connection.chn_msg_list);
740 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
741 
742 	if (newchannel->rescind) {
743 		err = -ENODEV;
744 		goto error_clean_msglist;
745 	}
746 
747 	err = vmbus_post_msg(open_msg,
748 			     sizeof(struct vmbus_channel_open_channel), true);
749 
750 	trace_vmbus_open(open_msg, err);
751 
752 	if (err != 0)
753 		goto error_clean_msglist;
754 
755 	wait_for_completion(&open_info->waitevent);
756 
757 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
758 	list_del(&open_info->msglistentry);
759 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
760 
761 	if (newchannel->rescind) {
762 		err = -ENODEV;
763 		goto error_free_info;
764 	}
765 
766 	if (open_info->response.open_result.status) {
767 		err = -EAGAIN;
768 		goto error_free_info;
769 	}
770 
771 	newchannel->state = CHANNEL_OPENED_STATE;
772 	kfree(open_info);
773 	return 0;
774 
775 error_clean_msglist:
776 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
777 	list_del(&open_info->msglistentry);
778 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
779 error_free_info:
780 	kfree(open_info);
781 error_free_gpadl:
782 	vmbus_teardown_gpadl(newchannel, &newchannel->ringbuffer_gpadlhandle);
783 error_clean_ring:
784 	hv_ringbuffer_cleanup(&newchannel->outbound);
785 	hv_ringbuffer_cleanup(&newchannel->inbound);
786 	vmbus_free_requestor(&newchannel->requestor);
787 	newchannel->state = CHANNEL_OPEN_STATE;
788 	return err;
789 }
790 
791 /*
792  * vmbus_connect_ring - Open the channel but reuse ring buffer
793  */
794 int vmbus_connect_ring(struct vmbus_channel *newchannel,
795 		       void (*onchannelcallback)(void *context), void *context)
796 {
797 	return  __vmbus_open(newchannel, NULL, 0, onchannelcallback, context);
798 }
799 EXPORT_SYMBOL_GPL(vmbus_connect_ring);
800 
801 /*
802  * vmbus_open - Open the specified channel.
803  */
804 int vmbus_open(struct vmbus_channel *newchannel,
805 	       u32 send_ringbuffer_size, u32 recv_ringbuffer_size,
806 	       void *userdata, u32 userdatalen,
807 	       void (*onchannelcallback)(void *context), void *context)
808 {
809 	int err;
810 
811 	err = vmbus_alloc_ring(newchannel, send_ringbuffer_size,
812 			       recv_ringbuffer_size);
813 	if (err)
814 		return err;
815 
816 	err = __vmbus_open(newchannel, userdata, userdatalen,
817 			   onchannelcallback, context);
818 	if (err)
819 		vmbus_free_ring(newchannel);
820 
821 	return err;
822 }
823 EXPORT_SYMBOL_GPL(vmbus_open);
824 
825 /*
826  * vmbus_teardown_gpadl -Teardown the specified GPADL handle
827  */
828 int vmbus_teardown_gpadl(struct vmbus_channel *channel, struct vmbus_gpadl *gpadl)
829 {
830 	struct vmbus_channel_gpadl_teardown *msg;
831 	struct vmbus_channel_msginfo *info;
832 	unsigned long flags;
833 	int ret;
834 
835 	info = kzalloc(sizeof(*info) +
836 		       sizeof(struct vmbus_channel_gpadl_teardown), GFP_KERNEL);
837 	if (!info)
838 		return -ENOMEM;
839 
840 	init_completion(&info->waitevent);
841 	info->waiting_channel = channel;
842 
843 	msg = (struct vmbus_channel_gpadl_teardown *)info->msg;
844 
845 	msg->header.msgtype = CHANNELMSG_GPADL_TEARDOWN;
846 	msg->child_relid = channel->offermsg.child_relid;
847 	msg->gpadl = gpadl->gpadl_handle;
848 
849 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
850 	list_add_tail(&info->msglistentry,
851 		      &vmbus_connection.chn_msg_list);
852 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
853 
854 	if (channel->rescind)
855 		goto post_msg_err;
856 
857 	ret = vmbus_post_msg(msg, sizeof(struct vmbus_channel_gpadl_teardown),
858 			     true);
859 
860 	trace_vmbus_teardown_gpadl(msg, ret);
861 
862 	if (ret)
863 		goto post_msg_err;
864 
865 	wait_for_completion(&info->waitevent);
866 
867 	gpadl->gpadl_handle = 0;
868 
869 post_msg_err:
870 	/*
871 	 * If the channel has been rescinded;
872 	 * we will be awakened by the rescind
873 	 * handler; set the error code to zero so we don't leak memory.
874 	 */
875 	if (channel->rescind)
876 		ret = 0;
877 
878 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
879 	list_del(&info->msglistentry);
880 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
881 
882 	kfree(info);
883 
884 	ret = set_memory_encrypted((unsigned long)gpadl->buffer,
885 				   PFN_UP(gpadl->size));
886 	if (ret)
887 		pr_warn("Fail to set mem host visibility in GPADL teardown %d.\n", ret);
888 
889 	return ret;
890 }
891 EXPORT_SYMBOL_GPL(vmbus_teardown_gpadl);
892 
893 void vmbus_reset_channel_cb(struct vmbus_channel *channel)
894 {
895 	unsigned long flags;
896 
897 	/*
898 	 * vmbus_on_event(), running in the per-channel tasklet, can race
899 	 * with vmbus_close_internal() in the case of SMP guest, e.g., when
900 	 * the former is accessing channel->inbound.ring_buffer, the latter
901 	 * could be freeing the ring_buffer pages, so here we must stop it
902 	 * first.
903 	 *
904 	 * vmbus_chan_sched() might call the netvsc driver callback function
905 	 * that ends up scheduling NAPI work that accesses the ring buffer.
906 	 * At this point, we have to ensure that any such work is completed
907 	 * and that the channel ring buffer is no longer being accessed, cf.
908 	 * the calls to napi_disable() in netvsc_device_remove().
909 	 */
910 	tasklet_disable(&channel->callback_event);
911 
912 	/* See the inline comments in vmbus_chan_sched(). */
913 	spin_lock_irqsave(&channel->sched_lock, flags);
914 	channel->onchannel_callback = NULL;
915 	spin_unlock_irqrestore(&channel->sched_lock, flags);
916 
917 	channel->sc_creation_callback = NULL;
918 
919 	/* Re-enable tasklet for use on re-open */
920 	tasklet_enable(&channel->callback_event);
921 }
922 
923 static int vmbus_close_internal(struct vmbus_channel *channel)
924 {
925 	struct vmbus_channel_close_channel *msg;
926 	int ret;
927 
928 	vmbus_reset_channel_cb(channel);
929 
930 	/*
931 	 * In case a device driver's probe() fails (e.g.,
932 	 * util_probe() -> vmbus_open() returns -ENOMEM) and the device is
933 	 * rescinded later (e.g., we dynamically disable an Integrated Service
934 	 * in Hyper-V Manager), the driver's remove() invokes vmbus_close():
935 	 * here we should skip most of the below cleanup work.
936 	 */
937 	if (channel->state != CHANNEL_OPENED_STATE)
938 		return -EINVAL;
939 
940 	channel->state = CHANNEL_OPEN_STATE;
941 
942 	/* Send a closing message */
943 
944 	msg = &channel->close_msg.msg;
945 
946 	msg->header.msgtype = CHANNELMSG_CLOSECHANNEL;
947 	msg->child_relid = channel->offermsg.child_relid;
948 
949 	ret = vmbus_post_msg(msg, sizeof(struct vmbus_channel_close_channel),
950 			     true);
951 
952 	trace_vmbus_close_internal(msg, ret);
953 
954 	if (ret) {
955 		pr_err("Close failed: close post msg return is %d\n", ret);
956 		/*
957 		 * If we failed to post the close msg,
958 		 * it is perhaps better to leak memory.
959 		 */
960 	}
961 
962 	/* Tear down the gpadl for the channel's ring buffer */
963 	else if (channel->ringbuffer_gpadlhandle.gpadl_handle) {
964 		ret = vmbus_teardown_gpadl(channel, &channel->ringbuffer_gpadlhandle);
965 		if (ret) {
966 			pr_err("Close failed: teardown gpadl return %d\n", ret);
967 			/*
968 			 * If we failed to teardown gpadl,
969 			 * it is perhaps better to leak memory.
970 			 */
971 		}
972 	}
973 
974 	if (!ret)
975 		vmbus_free_requestor(&channel->requestor);
976 
977 	return ret;
978 }
979 
980 /* disconnect ring - close all channels */
981 int vmbus_disconnect_ring(struct vmbus_channel *channel)
982 {
983 	struct vmbus_channel *cur_channel, *tmp;
984 	int ret;
985 
986 	if (channel->primary_channel != NULL)
987 		return -EINVAL;
988 
989 	list_for_each_entry_safe(cur_channel, tmp, &channel->sc_list, sc_list) {
990 		if (cur_channel->rescind)
991 			wait_for_completion(&cur_channel->rescind_event);
992 
993 		mutex_lock(&vmbus_connection.channel_mutex);
994 		if (vmbus_close_internal(cur_channel) == 0) {
995 			vmbus_free_ring(cur_channel);
996 
997 			if (cur_channel->rescind)
998 				hv_process_channel_removal(cur_channel);
999 		}
1000 		mutex_unlock(&vmbus_connection.channel_mutex);
1001 	}
1002 
1003 	/*
1004 	 * Now close the primary.
1005 	 */
1006 	mutex_lock(&vmbus_connection.channel_mutex);
1007 	ret = vmbus_close_internal(channel);
1008 	mutex_unlock(&vmbus_connection.channel_mutex);
1009 
1010 	return ret;
1011 }
1012 EXPORT_SYMBOL_GPL(vmbus_disconnect_ring);
1013 
1014 /*
1015  * vmbus_close - Close the specified channel
1016  */
1017 void vmbus_close(struct vmbus_channel *channel)
1018 {
1019 	if (vmbus_disconnect_ring(channel) == 0)
1020 		vmbus_free_ring(channel);
1021 }
1022 EXPORT_SYMBOL_GPL(vmbus_close);
1023 
1024 /**
1025  * vmbus_sendpacket_getid() - Send the specified buffer on the given channel
1026  * @channel: Pointer to vmbus_channel structure
1027  * @buffer: Pointer to the buffer you want to send the data from.
1028  * @bufferlen: Maximum size of what the buffer holds.
1029  * @requestid: Identifier of the request
1030  * @trans_id: Identifier of the transaction associated to this request, if
1031  *            the send is successful; undefined, otherwise.
1032  * @type: Type of packet that is being sent e.g. negotiate, time
1033  *	  packet etc.
1034  * @flags: 0 or VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED
1035  *
1036  * Sends data in @buffer directly to Hyper-V via the vmbus.
1037  * This will send the data unparsed to Hyper-V.
1038  *
1039  * Mainly used by Hyper-V drivers.
1040  */
1041 int vmbus_sendpacket_getid(struct vmbus_channel *channel, void *buffer,
1042 			   u32 bufferlen, u64 requestid, u64 *trans_id,
1043 			   enum vmbus_packet_type type, u32 flags)
1044 {
1045 	struct vmpacket_descriptor desc;
1046 	u32 packetlen = sizeof(struct vmpacket_descriptor) + bufferlen;
1047 	u32 packetlen_aligned = ALIGN(packetlen, sizeof(u64));
1048 	struct kvec bufferlist[3];
1049 	u64 aligned_data = 0;
1050 	int num_vecs = ((bufferlen != 0) ? 3 : 1);
1051 
1052 
1053 	/* Setup the descriptor */
1054 	desc.type = type; /* VmbusPacketTypeDataInBand; */
1055 	desc.flags = flags; /* VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED; */
1056 	/* in 8-bytes granularity */
1057 	desc.offset8 = sizeof(struct vmpacket_descriptor) >> 3;
1058 	desc.len8 = (u16)(packetlen_aligned >> 3);
1059 	desc.trans_id = VMBUS_RQST_ERROR; /* will be updated in hv_ringbuffer_write() */
1060 
1061 	bufferlist[0].iov_base = &desc;
1062 	bufferlist[0].iov_len = sizeof(struct vmpacket_descriptor);
1063 	bufferlist[1].iov_base = buffer;
1064 	bufferlist[1].iov_len = bufferlen;
1065 	bufferlist[2].iov_base = &aligned_data;
1066 	bufferlist[2].iov_len = (packetlen_aligned - packetlen);
1067 
1068 	return hv_ringbuffer_write(channel, bufferlist, num_vecs, requestid, trans_id);
1069 }
1070 EXPORT_SYMBOL(vmbus_sendpacket_getid);
1071 
1072 /**
1073  * vmbus_sendpacket() - Send the specified buffer on the given channel
1074  * @channel: Pointer to vmbus_channel structure
1075  * @buffer: Pointer to the buffer you want to send the data from.
1076  * @bufferlen: Maximum size of what the buffer holds.
1077  * @requestid: Identifier of the request
1078  * @type: Type of packet that is being sent e.g. negotiate, time
1079  *	  packet etc.
1080  * @flags: 0 or VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED
1081  *
1082  * Sends data in @buffer directly to Hyper-V via the vmbus.
1083  * This will send the data unparsed to Hyper-V.
1084  *
1085  * Mainly used by Hyper-V drivers.
1086  */
1087 int vmbus_sendpacket(struct vmbus_channel *channel, void *buffer,
1088 		     u32 bufferlen, u64 requestid,
1089 		     enum vmbus_packet_type type, u32 flags)
1090 {
1091 	return vmbus_sendpacket_getid(channel, buffer, bufferlen,
1092 				      requestid, NULL, type, flags);
1093 }
1094 EXPORT_SYMBOL(vmbus_sendpacket);
1095 
1096 /*
1097  * vmbus_sendpacket_pagebuffer - Send a range of single-page buffer
1098  * packets using a GPADL Direct packet type. This interface allows you
1099  * to control notifying the host. This will be useful for sending
1100  * batched data. Also the sender can control the send flags
1101  * explicitly.
1102  */
1103 int vmbus_sendpacket_pagebuffer(struct vmbus_channel *channel,
1104 				struct hv_page_buffer pagebuffers[],
1105 				u32 pagecount, void *buffer, u32 bufferlen,
1106 				u64 requestid)
1107 {
1108 	int i;
1109 	struct vmbus_channel_packet_page_buffer desc;
1110 	u32 descsize;
1111 	u32 packetlen;
1112 	u32 packetlen_aligned;
1113 	struct kvec bufferlist[3];
1114 	u64 aligned_data = 0;
1115 
1116 	if (pagecount > MAX_PAGE_BUFFER_COUNT)
1117 		return -EINVAL;
1118 
1119 	/*
1120 	 * Adjust the size down since vmbus_channel_packet_page_buffer is the
1121 	 * largest size we support
1122 	 */
1123 	descsize = sizeof(struct vmbus_channel_packet_page_buffer) -
1124 			  ((MAX_PAGE_BUFFER_COUNT - pagecount) *
1125 			  sizeof(struct hv_page_buffer));
1126 	packetlen = descsize + bufferlen;
1127 	packetlen_aligned = ALIGN(packetlen, sizeof(u64));
1128 
1129 	/* Setup the descriptor */
1130 	desc.type = VM_PKT_DATA_USING_GPA_DIRECT;
1131 	desc.flags = VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED;
1132 	desc.dataoffset8 = descsize >> 3; /* in 8-bytes granularity */
1133 	desc.length8 = (u16)(packetlen_aligned >> 3);
1134 	desc.transactionid = VMBUS_RQST_ERROR; /* will be updated in hv_ringbuffer_write() */
1135 	desc.reserved = 0;
1136 	desc.rangecount = pagecount;
1137 
1138 	for (i = 0; i < pagecount; i++) {
1139 		desc.range[i].len = pagebuffers[i].len;
1140 		desc.range[i].offset = pagebuffers[i].offset;
1141 		desc.range[i].pfn	 = pagebuffers[i].pfn;
1142 	}
1143 
1144 	bufferlist[0].iov_base = &desc;
1145 	bufferlist[0].iov_len = descsize;
1146 	bufferlist[1].iov_base = buffer;
1147 	bufferlist[1].iov_len = bufferlen;
1148 	bufferlist[2].iov_base = &aligned_data;
1149 	bufferlist[2].iov_len = (packetlen_aligned - packetlen);
1150 
1151 	return hv_ringbuffer_write(channel, bufferlist, 3, requestid, NULL);
1152 }
1153 EXPORT_SYMBOL_GPL(vmbus_sendpacket_pagebuffer);
1154 
1155 /*
1156  * vmbus_sendpacket_multipagebuffer - Send a multi-page buffer packet
1157  * using a GPADL Direct packet type.
1158  * The buffer includes the vmbus descriptor.
1159  */
1160 int vmbus_sendpacket_mpb_desc(struct vmbus_channel *channel,
1161 			      struct vmbus_packet_mpb_array *desc,
1162 			      u32 desc_size,
1163 			      void *buffer, u32 bufferlen, u64 requestid)
1164 {
1165 	u32 packetlen;
1166 	u32 packetlen_aligned;
1167 	struct kvec bufferlist[3];
1168 	u64 aligned_data = 0;
1169 
1170 	packetlen = desc_size + bufferlen;
1171 	packetlen_aligned = ALIGN(packetlen, sizeof(u64));
1172 
1173 	/* Setup the descriptor */
1174 	desc->type = VM_PKT_DATA_USING_GPA_DIRECT;
1175 	desc->flags = VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED;
1176 	desc->dataoffset8 = desc_size >> 3; /* in 8-bytes granularity */
1177 	desc->length8 = (u16)(packetlen_aligned >> 3);
1178 	desc->transactionid = VMBUS_RQST_ERROR; /* will be updated in hv_ringbuffer_write() */
1179 	desc->reserved = 0;
1180 	desc->rangecount = 1;
1181 
1182 	bufferlist[0].iov_base = desc;
1183 	bufferlist[0].iov_len = desc_size;
1184 	bufferlist[1].iov_base = buffer;
1185 	bufferlist[1].iov_len = bufferlen;
1186 	bufferlist[2].iov_base = &aligned_data;
1187 	bufferlist[2].iov_len = (packetlen_aligned - packetlen);
1188 
1189 	return hv_ringbuffer_write(channel, bufferlist, 3, requestid, NULL);
1190 }
1191 EXPORT_SYMBOL_GPL(vmbus_sendpacket_mpb_desc);
1192 
1193 /**
1194  * __vmbus_recvpacket() - Retrieve the user packet on the specified channel
1195  * @channel: Pointer to vmbus_channel structure
1196  * @buffer: Pointer to the buffer you want to receive the data into.
1197  * @bufferlen: Maximum size of what the buffer can hold.
1198  * @buffer_actual_len: The actual size of the data after it was received.
1199  * @requestid: Identifier of the request
1200  * @raw: true means keep the vmpacket_descriptor header in the received data.
1201  *
1202  * Receives directly from the hyper-v vmbus and puts the data it received
1203  * into Buffer. This will receive the data unparsed from hyper-v.
1204  *
1205  * Mainly used by Hyper-V drivers.
1206  */
1207 static inline int
1208 __vmbus_recvpacket(struct vmbus_channel *channel, void *buffer,
1209 		   u32 bufferlen, u32 *buffer_actual_len, u64 *requestid,
1210 		   bool raw)
1211 {
1212 	return hv_ringbuffer_read(channel, buffer, bufferlen,
1213 				  buffer_actual_len, requestid, raw);
1214 
1215 }
1216 
1217 int vmbus_recvpacket(struct vmbus_channel *channel, void *buffer,
1218 		     u32 bufferlen, u32 *buffer_actual_len,
1219 		     u64 *requestid)
1220 {
1221 	return __vmbus_recvpacket(channel, buffer, bufferlen,
1222 				  buffer_actual_len, requestid, false);
1223 }
1224 EXPORT_SYMBOL(vmbus_recvpacket);
1225 
1226 /*
1227  * vmbus_recvpacket_raw - Retrieve the raw packet on the specified channel
1228  */
1229 int vmbus_recvpacket_raw(struct vmbus_channel *channel, void *buffer,
1230 			      u32 bufferlen, u32 *buffer_actual_len,
1231 			      u64 *requestid)
1232 {
1233 	return __vmbus_recvpacket(channel, buffer, bufferlen,
1234 				  buffer_actual_len, requestid, true);
1235 }
1236 EXPORT_SYMBOL_GPL(vmbus_recvpacket_raw);
1237 
1238 /*
1239  * vmbus_next_request_id - Returns a new request id. It is also
1240  * the index at which the guest memory address is stored.
1241  * Uses a spin lock to avoid race conditions.
1242  * @channel: Pointer to the VMbus channel struct
1243  * @rqst_add: Guest memory address to be stored in the array
1244  */
1245 u64 vmbus_next_request_id(struct vmbus_channel *channel, u64 rqst_addr)
1246 {
1247 	struct vmbus_requestor *rqstor = &channel->requestor;
1248 	unsigned long flags;
1249 	u64 current_id;
1250 
1251 	/* Check rqstor has been initialized */
1252 	if (!channel->rqstor_size)
1253 		return VMBUS_NO_RQSTOR;
1254 
1255 	lock_requestor(channel, flags);
1256 	current_id = rqstor->next_request_id;
1257 
1258 	/* Requestor array is full */
1259 	if (current_id >= rqstor->size) {
1260 		unlock_requestor(channel, flags);
1261 		return VMBUS_RQST_ERROR;
1262 	}
1263 
1264 	rqstor->next_request_id = rqstor->req_arr[current_id];
1265 	rqstor->req_arr[current_id] = rqst_addr;
1266 
1267 	/* The already held spin lock provides atomicity */
1268 	bitmap_set(rqstor->req_bitmap, current_id, 1);
1269 
1270 	unlock_requestor(channel, flags);
1271 
1272 	/*
1273 	 * Cannot return an ID of 0, which is reserved for an unsolicited
1274 	 * message from Hyper-V; Hyper-V does not acknowledge (respond to)
1275 	 * VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED requests with ID of
1276 	 * 0 sent by the guest.
1277 	 */
1278 	return current_id + 1;
1279 }
1280 EXPORT_SYMBOL_GPL(vmbus_next_request_id);
1281 
1282 /* As in vmbus_request_addr_match() but without the requestor lock */
1283 u64 __vmbus_request_addr_match(struct vmbus_channel *channel, u64 trans_id,
1284 			       u64 rqst_addr)
1285 {
1286 	struct vmbus_requestor *rqstor = &channel->requestor;
1287 	u64 req_addr;
1288 
1289 	/* Check rqstor has been initialized */
1290 	if (!channel->rqstor_size)
1291 		return VMBUS_NO_RQSTOR;
1292 
1293 	/* Hyper-V can send an unsolicited message with ID of 0 */
1294 	if (!trans_id)
1295 		return VMBUS_RQST_ERROR;
1296 
1297 	/* Data corresponding to trans_id is stored at trans_id - 1 */
1298 	trans_id--;
1299 
1300 	/* Invalid trans_id */
1301 	if (trans_id >= rqstor->size || !test_bit(trans_id, rqstor->req_bitmap))
1302 		return VMBUS_RQST_ERROR;
1303 
1304 	req_addr = rqstor->req_arr[trans_id];
1305 	if (rqst_addr == VMBUS_RQST_ADDR_ANY || req_addr == rqst_addr) {
1306 		rqstor->req_arr[trans_id] = rqstor->next_request_id;
1307 		rqstor->next_request_id = trans_id;
1308 
1309 		/* The already held spin lock provides atomicity */
1310 		bitmap_clear(rqstor->req_bitmap, trans_id, 1);
1311 	}
1312 
1313 	return req_addr;
1314 }
1315 EXPORT_SYMBOL_GPL(__vmbus_request_addr_match);
1316 
1317 /*
1318  * vmbus_request_addr_match - Clears/removes @trans_id from the @channel's
1319  * requestor, provided the memory address stored at @trans_id equals @rqst_addr
1320  * (or provided @rqst_addr matches the sentinel value VMBUS_RQST_ADDR_ANY).
1321  *
1322  * Returns the memory address stored at @trans_id, or VMBUS_RQST_ERROR if
1323  * @trans_id is not contained in the requestor.
1324  *
1325  * Acquires and releases the requestor spin lock.
1326  */
1327 u64 vmbus_request_addr_match(struct vmbus_channel *channel, u64 trans_id,
1328 			     u64 rqst_addr)
1329 {
1330 	unsigned long flags;
1331 	u64 req_addr;
1332 
1333 	lock_requestor(channel, flags);
1334 	req_addr = __vmbus_request_addr_match(channel, trans_id, rqst_addr);
1335 	unlock_requestor(channel, flags);
1336 
1337 	return req_addr;
1338 }
1339 EXPORT_SYMBOL_GPL(vmbus_request_addr_match);
1340 
1341 /*
1342  * vmbus_request_addr - Returns the memory address stored at @trans_id
1343  * in @rqstor. Uses a spin lock to avoid race conditions.
1344  * @channel: Pointer to the VMbus channel struct
1345  * @trans_id: Request id sent back from Hyper-V. Becomes the requestor's
1346  * next request id.
1347  */
1348 u64 vmbus_request_addr(struct vmbus_channel *channel, u64 trans_id)
1349 {
1350 	return vmbus_request_addr_match(channel, trans_id, VMBUS_RQST_ADDR_ANY);
1351 }
1352 EXPORT_SYMBOL_GPL(vmbus_request_addr);
1353