xref: /openbmc/linux/drivers/hv/channel.c (revision b96fc2f3)
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/slab.h>
28 #include <linux/module.h>
29 #include <linux/hyperv.h>
30 #include <linux/uio.h>
31 
32 #include "hyperv_vmbus.h"
33 
34 #define NUM_PAGES_SPANNED(addr, len) \
35 ((PAGE_ALIGN(addr + len) >> PAGE_SHIFT) - (addr >> PAGE_SHIFT))
36 
37 /*
38  * vmbus_setevent- Trigger an event notification on the specified
39  * channel.
40  */
41 static void vmbus_setevent(struct vmbus_channel *channel)
42 {
43 	struct hv_monitor_page *monitorpage;
44 
45 	if (channel->offermsg.monitor_allocated) {
46 		/* Each u32 represents 32 channels */
47 		sync_set_bit(channel->offermsg.child_relid & 31,
48 			(unsigned long *) vmbus_connection.send_int_page +
49 			(channel->offermsg.child_relid >> 5));
50 
51 		/* Get the child to parent monitor page */
52 		monitorpage = vmbus_connection.monitor_pages[1];
53 
54 		sync_set_bit(channel->monitor_bit,
55 			(unsigned long *)&monitorpage->trigger_group
56 					[channel->monitor_grp].pending);
57 
58 	} else {
59 		vmbus_set_event(channel);
60 	}
61 }
62 
63 /*
64  * vmbus_open - Open the specified channel.
65  */
66 int vmbus_open(struct vmbus_channel *newchannel, u32 send_ringbuffer_size,
67 		     u32 recv_ringbuffer_size, void *userdata, u32 userdatalen,
68 		     void (*onchannelcallback)(void *context), void *context)
69 {
70 	struct vmbus_channel_open_channel *open_msg;
71 	struct vmbus_channel_msginfo *open_info = NULL;
72 	void *in, *out;
73 	unsigned long flags;
74 	int ret, err = 0;
75 	unsigned long t;
76 	struct page *page;
77 
78 	spin_lock_irqsave(&newchannel->lock, flags);
79 	if (newchannel->state == CHANNEL_OPEN_STATE) {
80 		newchannel->state = CHANNEL_OPENING_STATE;
81 	} else {
82 		spin_unlock_irqrestore(&newchannel->lock, flags);
83 		return -EINVAL;
84 	}
85 	spin_unlock_irqrestore(&newchannel->lock, flags);
86 
87 	newchannel->onchannel_callback = onchannelcallback;
88 	newchannel->channel_callback_context = context;
89 
90 	/* Allocate the ring buffer */
91 	page = alloc_pages_node(cpu_to_node(newchannel->target_cpu),
92 				GFP_KERNEL|__GFP_ZERO,
93 				get_order(send_ringbuffer_size +
94 				recv_ringbuffer_size));
95 
96 	if (!page)
97 		out = (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO,
98 					       get_order(send_ringbuffer_size +
99 					       recv_ringbuffer_size));
100 	else
101 		out = (void *)page_address(page);
102 
103 	if (!out) {
104 		err = -ENOMEM;
105 		goto error0;
106 	}
107 
108 	in = (void *)((unsigned long)out + send_ringbuffer_size);
109 
110 	newchannel->ringbuffer_pages = out;
111 	newchannel->ringbuffer_pagecount = (send_ringbuffer_size +
112 					   recv_ringbuffer_size) >> PAGE_SHIFT;
113 
114 	ret = hv_ringbuffer_init(
115 		&newchannel->outbound, out, send_ringbuffer_size);
116 
117 	if (ret != 0) {
118 		err = ret;
119 		goto error0;
120 	}
121 
122 	ret = hv_ringbuffer_init(
123 		&newchannel->inbound, in, recv_ringbuffer_size);
124 	if (ret != 0) {
125 		err = ret;
126 		goto error0;
127 	}
128 
129 
130 	/* Establish the gpadl for the ring buffer */
131 	newchannel->ringbuffer_gpadlhandle = 0;
132 
133 	ret = vmbus_establish_gpadl(newchannel,
134 					 newchannel->outbound.ring_buffer,
135 					 send_ringbuffer_size +
136 					 recv_ringbuffer_size,
137 					 &newchannel->ringbuffer_gpadlhandle);
138 
139 	if (ret != 0) {
140 		err = ret;
141 		goto error0;
142 	}
143 
144 	/* Create and init the channel open message */
145 	open_info = kmalloc(sizeof(*open_info) +
146 			   sizeof(struct vmbus_channel_open_channel),
147 			   GFP_KERNEL);
148 	if (!open_info) {
149 		err = -ENOMEM;
150 		goto error_gpadl;
151 	}
152 
153 	init_completion(&open_info->waitevent);
154 
155 	open_msg = (struct vmbus_channel_open_channel *)open_info->msg;
156 	open_msg->header.msgtype = CHANNELMSG_OPENCHANNEL;
157 	open_msg->openid = newchannel->offermsg.child_relid;
158 	open_msg->child_relid = newchannel->offermsg.child_relid;
159 	open_msg->ringbuffer_gpadlhandle = newchannel->ringbuffer_gpadlhandle;
160 	open_msg->downstream_ringbuffer_pageoffset = send_ringbuffer_size >>
161 						  PAGE_SHIFT;
162 	open_msg->target_vp = newchannel->target_vp;
163 
164 	if (userdatalen > MAX_USER_DEFINED_BYTES) {
165 		err = -EINVAL;
166 		goto error_gpadl;
167 	}
168 
169 	if (userdatalen)
170 		memcpy(open_msg->userdata, userdata, userdatalen);
171 
172 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
173 	list_add_tail(&open_info->msglistentry,
174 		      &vmbus_connection.chn_msg_list);
175 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
176 
177 	ret = vmbus_post_msg(open_msg,
178 			       sizeof(struct vmbus_channel_open_channel));
179 
180 	if (ret != 0) {
181 		err = ret;
182 		goto error1;
183 	}
184 
185 	t = wait_for_completion_timeout(&open_info->waitevent, 5*HZ);
186 	if (t == 0) {
187 		err = -ETIMEDOUT;
188 		goto error1;
189 	}
190 
191 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
192 	list_del(&open_info->msglistentry);
193 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
194 
195 	if (open_info->response.open_result.status) {
196 		err = -EAGAIN;
197 		goto error_gpadl;
198 	}
199 
200 	newchannel->state = CHANNEL_OPENED_STATE;
201 	kfree(open_info);
202 	return 0;
203 
204 error1:
205 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
206 	list_del(&open_info->msglistentry);
207 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
208 
209 error_gpadl:
210 	vmbus_teardown_gpadl(newchannel, newchannel->ringbuffer_gpadlhandle);
211 
212 error0:
213 	free_pages((unsigned long)out,
214 		get_order(send_ringbuffer_size + recv_ringbuffer_size));
215 	kfree(open_info);
216 	newchannel->state = CHANNEL_OPEN_STATE;
217 	return err;
218 }
219 EXPORT_SYMBOL_GPL(vmbus_open);
220 
221 /*
222  * create_gpadl_header - Creates a gpadl for the specified buffer
223  */
224 static int create_gpadl_header(void *kbuffer, u32 size,
225 					 struct vmbus_channel_msginfo **msginfo,
226 					 u32 *messagecount)
227 {
228 	int i;
229 	int pagecount;
230 	struct vmbus_channel_gpadl_header *gpadl_header;
231 	struct vmbus_channel_gpadl_body *gpadl_body;
232 	struct vmbus_channel_msginfo *msgheader;
233 	struct vmbus_channel_msginfo *msgbody = NULL;
234 	u32 msgsize;
235 
236 	int pfnsum, pfncount, pfnleft, pfncurr, pfnsize;
237 
238 	pagecount = size >> PAGE_SHIFT;
239 
240 	/* do we need a gpadl body msg */
241 	pfnsize = MAX_SIZE_CHANNEL_MESSAGE -
242 		  sizeof(struct vmbus_channel_gpadl_header) -
243 		  sizeof(struct gpa_range);
244 	pfncount = pfnsize / sizeof(u64);
245 
246 	if (pagecount > pfncount) {
247 		/* we need a gpadl body */
248 		/* fill in the header */
249 		msgsize = sizeof(struct vmbus_channel_msginfo) +
250 			  sizeof(struct vmbus_channel_gpadl_header) +
251 			  sizeof(struct gpa_range) + pfncount * sizeof(u64);
252 		msgheader =  kzalloc(msgsize, GFP_KERNEL);
253 		if (!msgheader)
254 			goto nomem;
255 
256 		INIT_LIST_HEAD(&msgheader->submsglist);
257 		msgheader->msgsize = msgsize;
258 
259 		gpadl_header = (struct vmbus_channel_gpadl_header *)
260 			msgheader->msg;
261 		gpadl_header->rangecount = 1;
262 		gpadl_header->range_buflen = sizeof(struct gpa_range) +
263 					 pagecount * sizeof(u64);
264 		gpadl_header->range[0].byte_offset = 0;
265 		gpadl_header->range[0].byte_count = size;
266 		for (i = 0; i < pfncount; i++)
267 			gpadl_header->range[0].pfn_array[i] = slow_virt_to_phys(
268 				kbuffer + PAGE_SIZE * i) >> PAGE_SHIFT;
269 		*msginfo = msgheader;
270 		*messagecount = 1;
271 
272 		pfnsum = pfncount;
273 		pfnleft = pagecount - pfncount;
274 
275 		/* how many pfns can we fit */
276 		pfnsize = MAX_SIZE_CHANNEL_MESSAGE -
277 			  sizeof(struct vmbus_channel_gpadl_body);
278 		pfncount = pfnsize / sizeof(u64);
279 
280 		/* fill in the body */
281 		while (pfnleft) {
282 			if (pfnleft > pfncount)
283 				pfncurr = pfncount;
284 			else
285 				pfncurr = pfnleft;
286 
287 			msgsize = sizeof(struct vmbus_channel_msginfo) +
288 				  sizeof(struct vmbus_channel_gpadl_body) +
289 				  pfncurr * sizeof(u64);
290 			msgbody = kzalloc(msgsize, GFP_KERNEL);
291 
292 			if (!msgbody) {
293 				struct vmbus_channel_msginfo *pos = NULL;
294 				struct vmbus_channel_msginfo *tmp = NULL;
295 				/*
296 				 * Free up all the allocated messages.
297 				 */
298 				list_for_each_entry_safe(pos, tmp,
299 					&msgheader->submsglist,
300 					msglistentry) {
301 
302 					list_del(&pos->msglistentry);
303 					kfree(pos);
304 				}
305 
306 				goto nomem;
307 			}
308 
309 			msgbody->msgsize = msgsize;
310 			(*messagecount)++;
311 			gpadl_body =
312 				(struct vmbus_channel_gpadl_body *)msgbody->msg;
313 
314 			/*
315 			 * Gpadl is u32 and we are using a pointer which could
316 			 * be 64-bit
317 			 * This is governed by the guest/host protocol and
318 			 * so the hypervisor gurantees that this is ok.
319 			 */
320 			for (i = 0; i < pfncurr; i++)
321 				gpadl_body->pfn[i] = slow_virt_to_phys(
322 					kbuffer + PAGE_SIZE * (pfnsum + i)) >>
323 					PAGE_SHIFT;
324 
325 			/* add to msg header */
326 			list_add_tail(&msgbody->msglistentry,
327 				      &msgheader->submsglist);
328 			pfnsum += pfncurr;
329 			pfnleft -= pfncurr;
330 		}
331 	} else {
332 		/* everything fits in a header */
333 		msgsize = sizeof(struct vmbus_channel_msginfo) +
334 			  sizeof(struct vmbus_channel_gpadl_header) +
335 			  sizeof(struct gpa_range) + pagecount * sizeof(u64);
336 		msgheader = kzalloc(msgsize, GFP_KERNEL);
337 		if (msgheader == NULL)
338 			goto nomem;
339 		msgheader->msgsize = msgsize;
340 
341 		gpadl_header = (struct vmbus_channel_gpadl_header *)
342 			msgheader->msg;
343 		gpadl_header->rangecount = 1;
344 		gpadl_header->range_buflen = sizeof(struct gpa_range) +
345 					 pagecount * sizeof(u64);
346 		gpadl_header->range[0].byte_offset = 0;
347 		gpadl_header->range[0].byte_count = size;
348 		for (i = 0; i < pagecount; i++)
349 			gpadl_header->range[0].pfn_array[i] = slow_virt_to_phys(
350 				kbuffer + PAGE_SIZE * i) >> PAGE_SHIFT;
351 
352 		*msginfo = msgheader;
353 		*messagecount = 1;
354 	}
355 
356 	return 0;
357 nomem:
358 	kfree(msgheader);
359 	kfree(msgbody);
360 	return -ENOMEM;
361 }
362 
363 /*
364  * vmbus_establish_gpadl - Estabish a GPADL for the specified buffer
365  *
366  * @channel: a channel
367  * @kbuffer: from kmalloc or vmalloc
368  * @size: page-size multiple
369  * @gpadl_handle: some funky thing
370  */
371 int vmbus_establish_gpadl(struct vmbus_channel *channel, void *kbuffer,
372 			       u32 size, u32 *gpadl_handle)
373 {
374 	struct vmbus_channel_gpadl_header *gpadlmsg;
375 	struct vmbus_channel_gpadl_body *gpadl_body;
376 	struct vmbus_channel_msginfo *msginfo = NULL;
377 	struct vmbus_channel_msginfo *submsginfo;
378 	u32 msgcount;
379 	struct list_head *curr;
380 	u32 next_gpadl_handle;
381 	unsigned long flags;
382 	int ret = 0;
383 
384 	next_gpadl_handle =
385 		(atomic_inc_return(&vmbus_connection.next_gpadl_handle) - 1);
386 
387 	ret = create_gpadl_header(kbuffer, size, &msginfo, &msgcount);
388 	if (ret)
389 		return ret;
390 
391 	init_completion(&msginfo->waitevent);
392 
393 	gpadlmsg = (struct vmbus_channel_gpadl_header *)msginfo->msg;
394 	gpadlmsg->header.msgtype = CHANNELMSG_GPADL_HEADER;
395 	gpadlmsg->child_relid = channel->offermsg.child_relid;
396 	gpadlmsg->gpadl = next_gpadl_handle;
397 
398 
399 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
400 	list_add_tail(&msginfo->msglistentry,
401 		      &vmbus_connection.chn_msg_list);
402 
403 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
404 
405 	ret = vmbus_post_msg(gpadlmsg, msginfo->msgsize -
406 			       sizeof(*msginfo));
407 	if (ret != 0)
408 		goto cleanup;
409 
410 	if (msgcount > 1) {
411 		list_for_each(curr, &msginfo->submsglist) {
412 
413 			submsginfo = (struct vmbus_channel_msginfo *)curr;
414 			gpadl_body =
415 			     (struct vmbus_channel_gpadl_body *)submsginfo->msg;
416 
417 			gpadl_body->header.msgtype =
418 				CHANNELMSG_GPADL_BODY;
419 			gpadl_body->gpadl = next_gpadl_handle;
420 
421 			ret = vmbus_post_msg(gpadl_body,
422 					       submsginfo->msgsize -
423 					       sizeof(*submsginfo));
424 			if (ret != 0)
425 				goto cleanup;
426 
427 		}
428 	}
429 	wait_for_completion(&msginfo->waitevent);
430 
431 	/* At this point, we received the gpadl created msg */
432 	*gpadl_handle = gpadlmsg->gpadl;
433 
434 cleanup:
435 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
436 	list_del(&msginfo->msglistentry);
437 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
438 
439 	kfree(msginfo);
440 	return ret;
441 }
442 EXPORT_SYMBOL_GPL(vmbus_establish_gpadl);
443 
444 /*
445  * vmbus_teardown_gpadl -Teardown the specified GPADL handle
446  */
447 int vmbus_teardown_gpadl(struct vmbus_channel *channel, u32 gpadl_handle)
448 {
449 	struct vmbus_channel_gpadl_teardown *msg;
450 	struct vmbus_channel_msginfo *info;
451 	unsigned long flags;
452 	int ret;
453 
454 	info = kmalloc(sizeof(*info) +
455 		       sizeof(struct vmbus_channel_gpadl_teardown), GFP_KERNEL);
456 	if (!info)
457 		return -ENOMEM;
458 
459 	init_completion(&info->waitevent);
460 
461 	msg = (struct vmbus_channel_gpadl_teardown *)info->msg;
462 
463 	msg->header.msgtype = CHANNELMSG_GPADL_TEARDOWN;
464 	msg->child_relid = channel->offermsg.child_relid;
465 	msg->gpadl = gpadl_handle;
466 
467 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
468 	list_add_tail(&info->msglistentry,
469 		      &vmbus_connection.chn_msg_list);
470 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
471 	ret = vmbus_post_msg(msg,
472 			       sizeof(struct vmbus_channel_gpadl_teardown));
473 
474 	if (ret)
475 		goto post_msg_err;
476 
477 	wait_for_completion(&info->waitevent);
478 
479 post_msg_err:
480 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
481 	list_del(&info->msglistentry);
482 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
483 
484 	kfree(info);
485 	return ret;
486 }
487 EXPORT_SYMBOL_GPL(vmbus_teardown_gpadl);
488 
489 static void reset_channel_cb(void *arg)
490 {
491 	struct vmbus_channel *channel = arg;
492 
493 	channel->onchannel_callback = NULL;
494 }
495 
496 static int vmbus_close_internal(struct vmbus_channel *channel)
497 {
498 	struct vmbus_channel_close_channel *msg;
499 	int ret;
500 
501 	channel->state = CHANNEL_OPEN_STATE;
502 	channel->sc_creation_callback = NULL;
503 	/* Stop callback and cancel the timer asap */
504 	if (channel->target_cpu != get_cpu()) {
505 		put_cpu();
506 		smp_call_function_single(channel->target_cpu, reset_channel_cb,
507 					 channel, true);
508 	} else {
509 		reset_channel_cb(channel);
510 		put_cpu();
511 	}
512 
513 	/* Send a closing message */
514 
515 	msg = &channel->close_msg.msg;
516 
517 	msg->header.msgtype = CHANNELMSG_CLOSECHANNEL;
518 	msg->child_relid = channel->offermsg.child_relid;
519 
520 	ret = vmbus_post_msg(msg, sizeof(struct vmbus_channel_close_channel));
521 
522 	if (ret) {
523 		pr_err("Close failed: close post msg return is %d\n", ret);
524 		/*
525 		 * If we failed to post the close msg,
526 		 * it is perhaps better to leak memory.
527 		 */
528 		return ret;
529 	}
530 
531 	/* Tear down the gpadl for the channel's ring buffer */
532 	if (channel->ringbuffer_gpadlhandle) {
533 		ret = vmbus_teardown_gpadl(channel,
534 					   channel->ringbuffer_gpadlhandle);
535 		if (ret) {
536 			pr_err("Close failed: teardown gpadl return %d\n", ret);
537 			/*
538 			 * If we failed to teardown gpadl,
539 			 * it is perhaps better to leak memory.
540 			 */
541 			return ret;
542 		}
543 	}
544 
545 	/* Cleanup the ring buffers for this channel */
546 	hv_ringbuffer_cleanup(&channel->outbound);
547 	hv_ringbuffer_cleanup(&channel->inbound);
548 
549 	free_pages((unsigned long)channel->ringbuffer_pages,
550 		get_order(channel->ringbuffer_pagecount * PAGE_SIZE));
551 
552 	/*
553 	 * If the channel has been rescinded; process device removal.
554 	 */
555 	if (channel->rescind)
556 		hv_process_channel_removal(channel,
557 					   channel->offermsg.child_relid);
558 	return ret;
559 }
560 
561 /*
562  * vmbus_close - Close the specified channel
563  */
564 void vmbus_close(struct vmbus_channel *channel)
565 {
566 	struct list_head *cur, *tmp;
567 	struct vmbus_channel *cur_channel;
568 
569 	if (channel->primary_channel != NULL) {
570 		/*
571 		 * We will only close sub-channels when
572 		 * the primary is closed.
573 		 */
574 		return;
575 	}
576 	/*
577 	 * Close all the sub-channels first and then close the
578 	 * primary channel.
579 	 */
580 	list_for_each_safe(cur, tmp, &channel->sc_list) {
581 		cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
582 		if (cur_channel->state != CHANNEL_OPENED_STATE)
583 			continue;
584 		vmbus_close_internal(cur_channel);
585 	}
586 	/*
587 	 * Now close the primary.
588 	 */
589 	vmbus_close_internal(channel);
590 }
591 EXPORT_SYMBOL_GPL(vmbus_close);
592 
593 int vmbus_sendpacket_ctl(struct vmbus_channel *channel, void *buffer,
594 			   u32 bufferlen, u64 requestid,
595 			   enum vmbus_packet_type type, u32 flags, bool kick_q)
596 {
597 	struct vmpacket_descriptor desc;
598 	u32 packetlen = sizeof(struct vmpacket_descriptor) + bufferlen;
599 	u32 packetlen_aligned = ALIGN(packetlen, sizeof(u64));
600 	struct kvec bufferlist[3];
601 	u64 aligned_data = 0;
602 	int ret;
603 	bool signal = false;
604 	int num_vecs = ((bufferlen != 0) ? 3 : 1);
605 
606 
607 	/* Setup the descriptor */
608 	desc.type = type; /* VmbusPacketTypeDataInBand; */
609 	desc.flags = flags; /* VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED; */
610 	/* in 8-bytes granularity */
611 	desc.offset8 = sizeof(struct vmpacket_descriptor) >> 3;
612 	desc.len8 = (u16)(packetlen_aligned >> 3);
613 	desc.trans_id = requestid;
614 
615 	bufferlist[0].iov_base = &desc;
616 	bufferlist[0].iov_len = sizeof(struct vmpacket_descriptor);
617 	bufferlist[1].iov_base = buffer;
618 	bufferlist[1].iov_len = bufferlen;
619 	bufferlist[2].iov_base = &aligned_data;
620 	bufferlist[2].iov_len = (packetlen_aligned - packetlen);
621 
622 	ret = hv_ringbuffer_write(&channel->outbound, bufferlist, num_vecs,
623 				  &signal);
624 
625 	/*
626 	 * Signalling the host is conditional on many factors:
627 	 * 1. The ring state changed from being empty to non-empty.
628 	 *    This is tracked by the variable "signal".
629 	 * 2. The variable kick_q tracks if more data will be placed
630 	 *    on the ring. We will not signal if more data is
631 	 *    to be placed.
632 	 *
633 	 * If we cannot write to the ring-buffer; signal the host
634 	 * even if we may not have written anything. This is a rare
635 	 * enough condition that it should not matter.
636 	 */
637 	if (((ret == 0) && kick_q && signal) || (ret))
638 		vmbus_setevent(channel);
639 
640 	return ret;
641 }
642 EXPORT_SYMBOL(vmbus_sendpacket_ctl);
643 
644 /**
645  * vmbus_sendpacket() - Send the specified buffer on the given channel
646  * @channel: Pointer to vmbus_channel structure.
647  * @buffer: Pointer to the buffer you want to receive the data into.
648  * @bufferlen: Maximum size of what the the buffer will hold
649  * @requestid: Identifier of the request
650  * @type: Type of packet that is being send e.g. negotiate, time
651  * packet etc.
652  *
653  * Sends data in @buffer directly to hyper-v via the vmbus
654  * This will send the data unparsed to hyper-v.
655  *
656  * Mainly used by Hyper-V drivers.
657  */
658 int vmbus_sendpacket(struct vmbus_channel *channel, void *buffer,
659 			   u32 bufferlen, u64 requestid,
660 			   enum vmbus_packet_type type, u32 flags)
661 {
662 	return vmbus_sendpacket_ctl(channel, buffer, bufferlen, requestid,
663 				    type, flags, true);
664 }
665 EXPORT_SYMBOL(vmbus_sendpacket);
666 
667 /*
668  * vmbus_sendpacket_pagebuffer_ctl - Send a range of single-page buffer
669  * packets using a GPADL Direct packet type. This interface allows you
670  * to control notifying the host. This will be useful for sending
671  * batched data. Also the sender can control the send flags
672  * explicitly.
673  */
674 int vmbus_sendpacket_pagebuffer_ctl(struct vmbus_channel *channel,
675 				     struct hv_page_buffer pagebuffers[],
676 				     u32 pagecount, void *buffer, u32 bufferlen,
677 				     u64 requestid,
678 				     u32 flags,
679 				     bool kick_q)
680 {
681 	int ret;
682 	int i;
683 	struct vmbus_channel_packet_page_buffer desc;
684 	u32 descsize;
685 	u32 packetlen;
686 	u32 packetlen_aligned;
687 	struct kvec bufferlist[3];
688 	u64 aligned_data = 0;
689 	bool signal = false;
690 
691 	if (pagecount > MAX_PAGE_BUFFER_COUNT)
692 		return -EINVAL;
693 
694 
695 	/*
696 	 * Adjust the size down since vmbus_channel_packet_page_buffer is the
697 	 * largest size we support
698 	 */
699 	descsize = sizeof(struct vmbus_channel_packet_page_buffer) -
700 			  ((MAX_PAGE_BUFFER_COUNT - pagecount) *
701 			  sizeof(struct hv_page_buffer));
702 	packetlen = descsize + bufferlen;
703 	packetlen_aligned = ALIGN(packetlen, sizeof(u64));
704 
705 	/* Setup the descriptor */
706 	desc.type = VM_PKT_DATA_USING_GPA_DIRECT;
707 	desc.flags = flags;
708 	desc.dataoffset8 = descsize >> 3; /* in 8-bytes grandularity */
709 	desc.length8 = (u16)(packetlen_aligned >> 3);
710 	desc.transactionid = requestid;
711 	desc.rangecount = pagecount;
712 
713 	for (i = 0; i < pagecount; i++) {
714 		desc.range[i].len = pagebuffers[i].len;
715 		desc.range[i].offset = pagebuffers[i].offset;
716 		desc.range[i].pfn	 = pagebuffers[i].pfn;
717 	}
718 
719 	bufferlist[0].iov_base = &desc;
720 	bufferlist[0].iov_len = descsize;
721 	bufferlist[1].iov_base = buffer;
722 	bufferlist[1].iov_len = bufferlen;
723 	bufferlist[2].iov_base = &aligned_data;
724 	bufferlist[2].iov_len = (packetlen_aligned - packetlen);
725 
726 	ret = hv_ringbuffer_write(&channel->outbound, bufferlist, 3, &signal);
727 
728 	/*
729 	 * Signalling the host is conditional on many factors:
730 	 * 1. The ring state changed from being empty to non-empty.
731 	 *    This is tracked by the variable "signal".
732 	 * 2. The variable kick_q tracks if more data will be placed
733 	 *    on the ring. We will not signal if more data is
734 	 *    to be placed.
735 	 *
736 	 * If we cannot write to the ring-buffer; signal the host
737 	 * even if we may not have written anything. This is a rare
738 	 * enough condition that it should not matter.
739 	 */
740 	if (((ret == 0) && kick_q && signal) || (ret))
741 		vmbus_setevent(channel);
742 
743 	return ret;
744 }
745 EXPORT_SYMBOL_GPL(vmbus_sendpacket_pagebuffer_ctl);
746 
747 /*
748  * vmbus_sendpacket_pagebuffer - Send a range of single-page buffer
749  * packets using a GPADL Direct packet type.
750  */
751 int vmbus_sendpacket_pagebuffer(struct vmbus_channel *channel,
752 				     struct hv_page_buffer pagebuffers[],
753 				     u32 pagecount, void *buffer, u32 bufferlen,
754 				     u64 requestid)
755 {
756 	u32 flags = VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED;
757 	return vmbus_sendpacket_pagebuffer_ctl(channel, pagebuffers, pagecount,
758 					       buffer, bufferlen, requestid,
759 					       flags, true);
760 
761 }
762 EXPORT_SYMBOL_GPL(vmbus_sendpacket_pagebuffer);
763 
764 /*
765  * vmbus_sendpacket_multipagebuffer - Send a multi-page buffer packet
766  * using a GPADL Direct packet type.
767  * The buffer includes the vmbus descriptor.
768  */
769 int vmbus_sendpacket_mpb_desc(struct vmbus_channel *channel,
770 			      struct vmbus_packet_mpb_array *desc,
771 			      u32 desc_size,
772 			      void *buffer, u32 bufferlen, u64 requestid)
773 {
774 	int ret;
775 	u32 packetlen;
776 	u32 packetlen_aligned;
777 	struct kvec bufferlist[3];
778 	u64 aligned_data = 0;
779 	bool signal = false;
780 
781 	packetlen = desc_size + bufferlen;
782 	packetlen_aligned = ALIGN(packetlen, sizeof(u64));
783 
784 	/* Setup the descriptor */
785 	desc->type = VM_PKT_DATA_USING_GPA_DIRECT;
786 	desc->flags = VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED;
787 	desc->dataoffset8 = desc_size >> 3; /* in 8-bytes grandularity */
788 	desc->length8 = (u16)(packetlen_aligned >> 3);
789 	desc->transactionid = requestid;
790 	desc->rangecount = 1;
791 
792 	bufferlist[0].iov_base = desc;
793 	bufferlist[0].iov_len = desc_size;
794 	bufferlist[1].iov_base = buffer;
795 	bufferlist[1].iov_len = bufferlen;
796 	bufferlist[2].iov_base = &aligned_data;
797 	bufferlist[2].iov_len = (packetlen_aligned - packetlen);
798 
799 	ret = hv_ringbuffer_write(&channel->outbound, bufferlist, 3, &signal);
800 
801 	if (ret == 0 && signal)
802 		vmbus_setevent(channel);
803 
804 	return ret;
805 }
806 EXPORT_SYMBOL_GPL(vmbus_sendpacket_mpb_desc);
807 
808 /*
809  * vmbus_sendpacket_multipagebuffer - Send a multi-page buffer packet
810  * using a GPADL Direct packet type.
811  */
812 int vmbus_sendpacket_multipagebuffer(struct vmbus_channel *channel,
813 				struct hv_multipage_buffer *multi_pagebuffer,
814 				void *buffer, u32 bufferlen, u64 requestid)
815 {
816 	int ret;
817 	struct vmbus_channel_packet_multipage_buffer desc;
818 	u32 descsize;
819 	u32 packetlen;
820 	u32 packetlen_aligned;
821 	struct kvec bufferlist[3];
822 	u64 aligned_data = 0;
823 	bool signal = false;
824 	u32 pfncount = NUM_PAGES_SPANNED(multi_pagebuffer->offset,
825 					 multi_pagebuffer->len);
826 
827 	if (pfncount > MAX_MULTIPAGE_BUFFER_COUNT)
828 		return -EINVAL;
829 
830 	/*
831 	 * Adjust the size down since vmbus_channel_packet_multipage_buffer is
832 	 * the largest size we support
833 	 */
834 	descsize = sizeof(struct vmbus_channel_packet_multipage_buffer) -
835 			  ((MAX_MULTIPAGE_BUFFER_COUNT - pfncount) *
836 			  sizeof(u64));
837 	packetlen = descsize + bufferlen;
838 	packetlen_aligned = ALIGN(packetlen, sizeof(u64));
839 
840 
841 	/* Setup the descriptor */
842 	desc.type = VM_PKT_DATA_USING_GPA_DIRECT;
843 	desc.flags = VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED;
844 	desc.dataoffset8 = descsize >> 3; /* in 8-bytes grandularity */
845 	desc.length8 = (u16)(packetlen_aligned >> 3);
846 	desc.transactionid = requestid;
847 	desc.rangecount = 1;
848 
849 	desc.range.len = multi_pagebuffer->len;
850 	desc.range.offset = multi_pagebuffer->offset;
851 
852 	memcpy(desc.range.pfn_array, multi_pagebuffer->pfn_array,
853 	       pfncount * sizeof(u64));
854 
855 	bufferlist[0].iov_base = &desc;
856 	bufferlist[0].iov_len = descsize;
857 	bufferlist[1].iov_base = buffer;
858 	bufferlist[1].iov_len = bufferlen;
859 	bufferlist[2].iov_base = &aligned_data;
860 	bufferlist[2].iov_len = (packetlen_aligned - packetlen);
861 
862 	ret = hv_ringbuffer_write(&channel->outbound, bufferlist, 3, &signal);
863 
864 	if (ret == 0 && signal)
865 		vmbus_setevent(channel);
866 
867 	return ret;
868 }
869 EXPORT_SYMBOL_GPL(vmbus_sendpacket_multipagebuffer);
870 
871 /**
872  * vmbus_recvpacket() - Retrieve the user packet on the specified channel
873  * @channel: Pointer to vmbus_channel structure.
874  * @buffer: Pointer to the buffer you want to receive the data into.
875  * @bufferlen: Maximum size of what the the buffer will hold
876  * @buffer_actual_len: The actual size of the data after it was received
877  * @requestid: Identifier of the request
878  *
879  * Receives directly from the hyper-v vmbus and puts the data it received
880  * into Buffer. This will receive the data unparsed from hyper-v.
881  *
882  * Mainly used by Hyper-V drivers.
883  */
884 int vmbus_recvpacket(struct vmbus_channel *channel, void *buffer,
885 			u32 bufferlen, u32 *buffer_actual_len, u64 *requestid)
886 {
887 	struct vmpacket_descriptor desc;
888 	u32 packetlen;
889 	u32 userlen;
890 	int ret;
891 	bool signal = false;
892 
893 	*buffer_actual_len = 0;
894 	*requestid = 0;
895 
896 
897 	ret = hv_ringbuffer_peek(&channel->inbound, &desc,
898 			     sizeof(struct vmpacket_descriptor));
899 	if (ret != 0)
900 		return 0;
901 
902 	packetlen = desc.len8 << 3;
903 	userlen = packetlen - (desc.offset8 << 3);
904 
905 	*buffer_actual_len = userlen;
906 
907 	if (userlen > bufferlen) {
908 
909 		pr_err("Buffer too small - got %d needs %d\n",
910 			   bufferlen, userlen);
911 		return -ETOOSMALL;
912 	}
913 
914 	*requestid = desc.trans_id;
915 
916 	/* Copy over the packet to the user buffer */
917 	ret = hv_ringbuffer_read(&channel->inbound, buffer, userlen,
918 			     (desc.offset8 << 3), &signal);
919 
920 	if (signal)
921 		vmbus_setevent(channel);
922 
923 	return 0;
924 }
925 EXPORT_SYMBOL(vmbus_recvpacket);
926 
927 /*
928  * vmbus_recvpacket_raw - Retrieve the raw packet on the specified channel
929  */
930 int vmbus_recvpacket_raw(struct vmbus_channel *channel, void *buffer,
931 			      u32 bufferlen, u32 *buffer_actual_len,
932 			      u64 *requestid)
933 {
934 	struct vmpacket_descriptor desc;
935 	u32 packetlen;
936 	int ret;
937 	bool signal = false;
938 
939 	*buffer_actual_len = 0;
940 	*requestid = 0;
941 
942 
943 	ret = hv_ringbuffer_peek(&channel->inbound, &desc,
944 			     sizeof(struct vmpacket_descriptor));
945 	if (ret != 0)
946 		return 0;
947 
948 
949 	packetlen = desc.len8 << 3;
950 
951 	*buffer_actual_len = packetlen;
952 
953 	if (packetlen > bufferlen)
954 		return -ENOBUFS;
955 
956 	*requestid = desc.trans_id;
957 
958 	/* Copy over the entire packet to the user buffer */
959 	ret = hv_ringbuffer_read(&channel->inbound, buffer, packetlen, 0,
960 				 &signal);
961 
962 	if (signal)
963 		vmbus_setevent(channel);
964 
965 	return ret;
966 }
967 EXPORT_SYMBOL_GPL(vmbus_recvpacket_raw);
968