xref: /linux/drivers/hv/channel.c (revision c0c914eca7f251c70facc37dfebeaf176601918d)
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 #include <linux/interrupt.h>
32 
33 #include "hyperv_vmbus.h"
34 
35 #define NUM_PAGES_SPANNED(addr, len) \
36 ((PAGE_ALIGN(addr + len) >> PAGE_SHIFT) - (addr >> PAGE_SHIFT))
37 
38 /*
39  * vmbus_setevent- Trigger an event notification on the specified
40  * channel.
41  */
42 static void vmbus_setevent(struct vmbus_channel *channel)
43 {
44 	struct hv_monitor_page *monitorpage;
45 
46 	if (channel->offermsg.monitor_allocated) {
47 		/* Each u32 represents 32 channels */
48 		sync_set_bit(channel->offermsg.child_relid & 31,
49 			(unsigned long *) vmbus_connection.send_int_page +
50 			(channel->offermsg.child_relid >> 5));
51 
52 		/* Get the child to parent monitor page */
53 		monitorpage = vmbus_connection.monitor_pages[1];
54 
55 		sync_set_bit(channel->monitor_bit,
56 			(unsigned long *)&monitorpage->trigger_group
57 					[channel->monitor_grp].pending);
58 
59 	} else {
60 		vmbus_set_event(channel);
61 	}
62 }
63 
64 /*
65  * vmbus_open - Open the specified channel.
66  */
67 int vmbus_open(struct vmbus_channel *newchannel, u32 send_ringbuffer_size,
68 		     u32 recv_ringbuffer_size, void *userdata, u32 userdatalen,
69 		     void (*onchannelcallback)(void *context), void *context)
70 {
71 	struct vmbus_channel_open_channel *open_msg;
72 	struct vmbus_channel_msginfo *open_info = NULL;
73 	void *in, *out;
74 	unsigned long flags;
75 	int ret, err = 0;
76 	unsigned long t;
77 	struct page *page;
78 
79 	spin_lock_irqsave(&newchannel->lock, flags);
80 	if (newchannel->state == CHANNEL_OPEN_STATE) {
81 		newchannel->state = CHANNEL_OPENING_STATE;
82 	} else {
83 		spin_unlock_irqrestore(&newchannel->lock, flags);
84 		return -EINVAL;
85 	}
86 	spin_unlock_irqrestore(&newchannel->lock, flags);
87 
88 	newchannel->onchannel_callback = onchannelcallback;
89 	newchannel->channel_callback_context = context;
90 
91 	/* Allocate the ring buffer */
92 	page = alloc_pages_node(cpu_to_node(newchannel->target_cpu),
93 				GFP_KERNEL|__GFP_ZERO,
94 				get_order(send_ringbuffer_size +
95 				recv_ringbuffer_size));
96 
97 	if (!page)
98 		out = (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO,
99 					       get_order(send_ringbuffer_size +
100 					       recv_ringbuffer_size));
101 	else
102 		out = (void *)page_address(page);
103 
104 	if (!out) {
105 		err = -ENOMEM;
106 		goto error0;
107 	}
108 
109 	in = (void *)((unsigned long)out + send_ringbuffer_size);
110 
111 	newchannel->ringbuffer_pages = out;
112 	newchannel->ringbuffer_pagecount = (send_ringbuffer_size +
113 					   recv_ringbuffer_size) >> PAGE_SHIFT;
114 
115 	ret = hv_ringbuffer_init(
116 		&newchannel->outbound, out, send_ringbuffer_size);
117 
118 	if (ret != 0) {
119 		err = ret;
120 		goto error0;
121 	}
122 
123 	ret = hv_ringbuffer_init(
124 		&newchannel->inbound, in, recv_ringbuffer_size);
125 	if (ret != 0) {
126 		err = ret;
127 		goto error0;
128 	}
129 
130 
131 	/* Establish the gpadl for the ring buffer */
132 	newchannel->ringbuffer_gpadlhandle = 0;
133 
134 	ret = vmbus_establish_gpadl(newchannel,
135 					 newchannel->outbound.ring_buffer,
136 					 send_ringbuffer_size +
137 					 recv_ringbuffer_size,
138 					 &newchannel->ringbuffer_gpadlhandle);
139 
140 	if (ret != 0) {
141 		err = ret;
142 		goto error0;
143 	}
144 
145 	/* Create and init the channel open message */
146 	open_info = kmalloc(sizeof(*open_info) +
147 			   sizeof(struct vmbus_channel_open_channel),
148 			   GFP_KERNEL);
149 	if (!open_info) {
150 		err = -ENOMEM;
151 		goto error_gpadl;
152 	}
153 
154 	init_completion(&open_info->waitevent);
155 
156 	open_msg = (struct vmbus_channel_open_channel *)open_info->msg;
157 	open_msg->header.msgtype = CHANNELMSG_OPENCHANNEL;
158 	open_msg->openid = newchannel->offermsg.child_relid;
159 	open_msg->child_relid = newchannel->offermsg.child_relid;
160 	open_msg->ringbuffer_gpadlhandle = newchannel->ringbuffer_gpadlhandle;
161 	open_msg->downstream_ringbuffer_pageoffset = send_ringbuffer_size >>
162 						  PAGE_SHIFT;
163 	open_msg->target_vp = newchannel->target_vp;
164 
165 	if (userdatalen > MAX_USER_DEFINED_BYTES) {
166 		err = -EINVAL;
167 		goto error_gpadl;
168 	}
169 
170 	if (userdatalen)
171 		memcpy(open_msg->userdata, userdata, userdatalen);
172 
173 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
174 	list_add_tail(&open_info->msglistentry,
175 		      &vmbus_connection.chn_msg_list);
176 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
177 
178 	ret = vmbus_post_msg(open_msg,
179 			       sizeof(struct vmbus_channel_open_channel));
180 
181 	if (ret != 0) {
182 		err = ret;
183 		goto error1;
184 	}
185 
186 	t = wait_for_completion_timeout(&open_info->waitevent, 5*HZ);
187 	if (t == 0) {
188 		err = -ETIMEDOUT;
189 		goto error1;
190 	}
191 
192 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
193 	list_del(&open_info->msglistentry);
194 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
195 
196 	if (open_info->response.open_result.status) {
197 		err = -EAGAIN;
198 		goto error_gpadl;
199 	}
200 
201 	newchannel->state = CHANNEL_OPENED_STATE;
202 	kfree(open_info);
203 	return 0;
204 
205 error1:
206 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
207 	list_del(&open_info->msglistentry);
208 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
209 
210 error_gpadl:
211 	vmbus_teardown_gpadl(newchannel, newchannel->ringbuffer_gpadlhandle);
212 
213 error0:
214 	free_pages((unsigned long)out,
215 		get_order(send_ringbuffer_size + recv_ringbuffer_size));
216 	kfree(open_info);
217 	newchannel->state = CHANNEL_OPEN_STATE;
218 	return err;
219 }
220 EXPORT_SYMBOL_GPL(vmbus_open);
221 
222 /*
223  * create_gpadl_header - Creates a gpadl for the specified buffer
224  */
225 static int create_gpadl_header(void *kbuffer, u32 size,
226 					 struct vmbus_channel_msginfo **msginfo,
227 					 u32 *messagecount)
228 {
229 	int i;
230 	int pagecount;
231 	struct vmbus_channel_gpadl_header *gpadl_header;
232 	struct vmbus_channel_gpadl_body *gpadl_body;
233 	struct vmbus_channel_msginfo *msgheader;
234 	struct vmbus_channel_msginfo *msgbody = NULL;
235 	u32 msgsize;
236 
237 	int pfnsum, pfncount, pfnleft, pfncurr, pfnsize;
238 
239 	pagecount = size >> PAGE_SHIFT;
240 
241 	/* do we need a gpadl body msg */
242 	pfnsize = MAX_SIZE_CHANNEL_MESSAGE -
243 		  sizeof(struct vmbus_channel_gpadl_header) -
244 		  sizeof(struct gpa_range);
245 	pfncount = pfnsize / sizeof(u64);
246 
247 	if (pagecount > pfncount) {
248 		/* we need a gpadl body */
249 		/* fill in the header */
250 		msgsize = sizeof(struct vmbus_channel_msginfo) +
251 			  sizeof(struct vmbus_channel_gpadl_header) +
252 			  sizeof(struct gpa_range) + pfncount * sizeof(u64);
253 		msgheader =  kzalloc(msgsize, GFP_KERNEL);
254 		if (!msgheader)
255 			goto nomem;
256 
257 		INIT_LIST_HEAD(&msgheader->submsglist);
258 		msgheader->msgsize = msgsize;
259 
260 		gpadl_header = (struct vmbus_channel_gpadl_header *)
261 			msgheader->msg;
262 		gpadl_header->rangecount = 1;
263 		gpadl_header->range_buflen = sizeof(struct gpa_range) +
264 					 pagecount * sizeof(u64);
265 		gpadl_header->range[0].byte_offset = 0;
266 		gpadl_header->range[0].byte_count = size;
267 		for (i = 0; i < pfncount; i++)
268 			gpadl_header->range[0].pfn_array[i] = slow_virt_to_phys(
269 				kbuffer + PAGE_SIZE * i) >> PAGE_SHIFT;
270 		*msginfo = msgheader;
271 		*messagecount = 1;
272 
273 		pfnsum = pfncount;
274 		pfnleft = pagecount - pfncount;
275 
276 		/* how many pfns can we fit */
277 		pfnsize = MAX_SIZE_CHANNEL_MESSAGE -
278 			  sizeof(struct vmbus_channel_gpadl_body);
279 		pfncount = pfnsize / sizeof(u64);
280 
281 		/* fill in the body */
282 		while (pfnleft) {
283 			if (pfnleft > pfncount)
284 				pfncurr = pfncount;
285 			else
286 				pfncurr = pfnleft;
287 
288 			msgsize = sizeof(struct vmbus_channel_msginfo) +
289 				  sizeof(struct vmbus_channel_gpadl_body) +
290 				  pfncurr * sizeof(u64);
291 			msgbody = kzalloc(msgsize, GFP_KERNEL);
292 
293 			if (!msgbody) {
294 				struct vmbus_channel_msginfo *pos = NULL;
295 				struct vmbus_channel_msginfo *tmp = NULL;
296 				/*
297 				 * Free up all the allocated messages.
298 				 */
299 				list_for_each_entry_safe(pos, tmp,
300 					&msgheader->submsglist,
301 					msglistentry) {
302 
303 					list_del(&pos->msglistentry);
304 					kfree(pos);
305 				}
306 
307 				goto nomem;
308 			}
309 
310 			msgbody->msgsize = msgsize;
311 			(*messagecount)++;
312 			gpadl_body =
313 				(struct vmbus_channel_gpadl_body *)msgbody->msg;
314 
315 			/*
316 			 * Gpadl is u32 and we are using a pointer which could
317 			 * be 64-bit
318 			 * This is governed by the guest/host protocol and
319 			 * so the hypervisor gurantees that this is ok.
320 			 */
321 			for (i = 0; i < pfncurr; i++)
322 				gpadl_body->pfn[i] = slow_virt_to_phys(
323 					kbuffer + PAGE_SIZE * (pfnsum + i)) >>
324 					PAGE_SHIFT;
325 
326 			/* add to msg header */
327 			list_add_tail(&msgbody->msglistentry,
328 				      &msgheader->submsglist);
329 			pfnsum += pfncurr;
330 			pfnleft -= pfncurr;
331 		}
332 	} else {
333 		/* everything fits in a header */
334 		msgsize = sizeof(struct vmbus_channel_msginfo) +
335 			  sizeof(struct vmbus_channel_gpadl_header) +
336 			  sizeof(struct gpa_range) + pagecount * sizeof(u64);
337 		msgheader = kzalloc(msgsize, GFP_KERNEL);
338 		if (msgheader == NULL)
339 			goto nomem;
340 		msgheader->msgsize = msgsize;
341 
342 		gpadl_header = (struct vmbus_channel_gpadl_header *)
343 			msgheader->msg;
344 		gpadl_header->rangecount = 1;
345 		gpadl_header->range_buflen = sizeof(struct gpa_range) +
346 					 pagecount * sizeof(u64);
347 		gpadl_header->range[0].byte_offset = 0;
348 		gpadl_header->range[0].byte_count = size;
349 		for (i = 0; i < pagecount; i++)
350 			gpadl_header->range[0].pfn_array[i] = slow_virt_to_phys(
351 				kbuffer + PAGE_SIZE * i) >> PAGE_SHIFT;
352 
353 		*msginfo = msgheader;
354 		*messagecount = 1;
355 	}
356 
357 	return 0;
358 nomem:
359 	kfree(msgheader);
360 	kfree(msgbody);
361 	return -ENOMEM;
362 }
363 
364 /*
365  * vmbus_establish_gpadl - Estabish a GPADL for the specified buffer
366  *
367  * @channel: a channel
368  * @kbuffer: from kmalloc or vmalloc
369  * @size: page-size multiple
370  * @gpadl_handle: some funky thing
371  */
372 int vmbus_establish_gpadl(struct vmbus_channel *channel, void *kbuffer,
373 			       u32 size, u32 *gpadl_handle)
374 {
375 	struct vmbus_channel_gpadl_header *gpadlmsg;
376 	struct vmbus_channel_gpadl_body *gpadl_body;
377 	struct vmbus_channel_msginfo *msginfo = NULL;
378 	struct vmbus_channel_msginfo *submsginfo;
379 	u32 msgcount;
380 	struct list_head *curr;
381 	u32 next_gpadl_handle;
382 	unsigned long flags;
383 	int ret = 0;
384 
385 	next_gpadl_handle =
386 		(atomic_inc_return(&vmbus_connection.next_gpadl_handle) - 1);
387 
388 	ret = create_gpadl_header(kbuffer, size, &msginfo, &msgcount);
389 	if (ret)
390 		return ret;
391 
392 	init_completion(&msginfo->waitevent);
393 
394 	gpadlmsg = (struct vmbus_channel_gpadl_header *)msginfo->msg;
395 	gpadlmsg->header.msgtype = CHANNELMSG_GPADL_HEADER;
396 	gpadlmsg->child_relid = channel->offermsg.child_relid;
397 	gpadlmsg->gpadl = next_gpadl_handle;
398 
399 
400 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
401 	list_add_tail(&msginfo->msglistentry,
402 		      &vmbus_connection.chn_msg_list);
403 
404 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
405 
406 	ret = vmbus_post_msg(gpadlmsg, msginfo->msgsize -
407 			       sizeof(*msginfo));
408 	if (ret != 0)
409 		goto cleanup;
410 
411 	if (msgcount > 1) {
412 		list_for_each(curr, &msginfo->submsglist) {
413 
414 			submsginfo = (struct vmbus_channel_msginfo *)curr;
415 			gpadl_body =
416 			     (struct vmbus_channel_gpadl_body *)submsginfo->msg;
417 
418 			gpadl_body->header.msgtype =
419 				CHANNELMSG_GPADL_BODY;
420 			gpadl_body->gpadl = next_gpadl_handle;
421 
422 			ret = vmbus_post_msg(gpadl_body,
423 					       submsginfo->msgsize -
424 					       sizeof(*submsginfo));
425 			if (ret != 0)
426 				goto cleanup;
427 
428 		}
429 	}
430 	wait_for_completion(&msginfo->waitevent);
431 
432 	/* At this point, we received the gpadl created msg */
433 	*gpadl_handle = gpadlmsg->gpadl;
434 
435 cleanup:
436 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
437 	list_del(&msginfo->msglistentry);
438 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
439 
440 	kfree(msginfo);
441 	return ret;
442 }
443 EXPORT_SYMBOL_GPL(vmbus_establish_gpadl);
444 
445 /*
446  * vmbus_teardown_gpadl -Teardown the specified GPADL handle
447  */
448 int vmbus_teardown_gpadl(struct vmbus_channel *channel, u32 gpadl_handle)
449 {
450 	struct vmbus_channel_gpadl_teardown *msg;
451 	struct vmbus_channel_msginfo *info;
452 	unsigned long flags;
453 	int ret;
454 
455 	info = kmalloc(sizeof(*info) +
456 		       sizeof(struct vmbus_channel_gpadl_teardown), GFP_KERNEL);
457 	if (!info)
458 		return -ENOMEM;
459 
460 	init_completion(&info->waitevent);
461 
462 	msg = (struct vmbus_channel_gpadl_teardown *)info->msg;
463 
464 	msg->header.msgtype = CHANNELMSG_GPADL_TEARDOWN;
465 	msg->child_relid = channel->offermsg.child_relid;
466 	msg->gpadl = gpadl_handle;
467 
468 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
469 	list_add_tail(&info->msglistentry,
470 		      &vmbus_connection.chn_msg_list);
471 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
472 	ret = vmbus_post_msg(msg,
473 			       sizeof(struct vmbus_channel_gpadl_teardown));
474 
475 	if (ret)
476 		goto post_msg_err;
477 
478 	wait_for_completion(&info->waitevent);
479 
480 post_msg_err:
481 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
482 	list_del(&info->msglistentry);
483 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
484 
485 	kfree(info);
486 	return ret;
487 }
488 EXPORT_SYMBOL_GPL(vmbus_teardown_gpadl);
489 
490 static void reset_channel_cb(void *arg)
491 {
492 	struct vmbus_channel *channel = arg;
493 
494 	channel->onchannel_callback = NULL;
495 }
496 
497 static int vmbus_close_internal(struct vmbus_channel *channel)
498 {
499 	struct vmbus_channel_close_channel *msg;
500 	struct tasklet_struct *tasklet;
501 	int ret;
502 
503 	/*
504 	 * process_chn_event(), running in the tasklet, can race
505 	 * with vmbus_close_internal() in the case of SMP guest, e.g., when
506 	 * the former is accessing channel->inbound.ring_buffer, the latter
507 	 * could be freeing the ring_buffer pages.
508 	 *
509 	 * To resolve the race, we can serialize them by disabling the
510 	 * tasklet when the latter is running here.
511 	 */
512 	tasklet = hv_context.event_dpc[channel->target_cpu];
513 	tasklet_disable(tasklet);
514 
515 	/*
516 	 * In case a device driver's probe() fails (e.g.,
517 	 * util_probe() -> vmbus_open() returns -ENOMEM) and the device is
518 	 * rescinded later (e.g., we dynamically disble an Integrated Service
519 	 * in Hyper-V Manager), the driver's remove() invokes vmbus_close():
520 	 * here we should skip most of the below cleanup work.
521 	 */
522 	if (channel->state != CHANNEL_OPENED_STATE) {
523 		ret = -EINVAL;
524 		goto out;
525 	}
526 
527 	channel->state = CHANNEL_OPEN_STATE;
528 	channel->sc_creation_callback = NULL;
529 	/* Stop callback and cancel the timer asap */
530 	if (channel->target_cpu != get_cpu()) {
531 		put_cpu();
532 		smp_call_function_single(channel->target_cpu, reset_channel_cb,
533 					 channel, true);
534 	} else {
535 		reset_channel_cb(channel);
536 		put_cpu();
537 	}
538 
539 	/* Send a closing message */
540 
541 	msg = &channel->close_msg.msg;
542 
543 	msg->header.msgtype = CHANNELMSG_CLOSECHANNEL;
544 	msg->child_relid = channel->offermsg.child_relid;
545 
546 	ret = vmbus_post_msg(msg, sizeof(struct vmbus_channel_close_channel));
547 
548 	if (ret) {
549 		pr_err("Close failed: close post msg return is %d\n", ret);
550 		/*
551 		 * If we failed to post the close msg,
552 		 * it is perhaps better to leak memory.
553 		 */
554 		goto out;
555 	}
556 
557 	/* Tear down the gpadl for the channel's ring buffer */
558 	if (channel->ringbuffer_gpadlhandle) {
559 		ret = vmbus_teardown_gpadl(channel,
560 					   channel->ringbuffer_gpadlhandle);
561 		if (ret) {
562 			pr_err("Close failed: teardown gpadl return %d\n", ret);
563 			/*
564 			 * If we failed to teardown gpadl,
565 			 * it is perhaps better to leak memory.
566 			 */
567 			goto out;
568 		}
569 	}
570 
571 	/* Cleanup the ring buffers for this channel */
572 	hv_ringbuffer_cleanup(&channel->outbound);
573 	hv_ringbuffer_cleanup(&channel->inbound);
574 
575 	free_pages((unsigned long)channel->ringbuffer_pages,
576 		get_order(channel->ringbuffer_pagecount * PAGE_SIZE));
577 
578 out:
579 	tasklet_enable(tasklet);
580 
581 	return ret;
582 }
583 
584 /*
585  * vmbus_close - Close the specified channel
586  */
587 void vmbus_close(struct vmbus_channel *channel)
588 {
589 	struct list_head *cur, *tmp;
590 	struct vmbus_channel *cur_channel;
591 
592 	if (channel->primary_channel != NULL) {
593 		/*
594 		 * We will only close sub-channels when
595 		 * the primary is closed.
596 		 */
597 		return;
598 	}
599 	/*
600 	 * Close all the sub-channels first and then close the
601 	 * primary channel.
602 	 */
603 	list_for_each_safe(cur, tmp, &channel->sc_list) {
604 		cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
605 		if (cur_channel->state != CHANNEL_OPENED_STATE)
606 			continue;
607 		vmbus_close_internal(cur_channel);
608 	}
609 	/*
610 	 * Now close the primary.
611 	 */
612 	vmbus_close_internal(channel);
613 }
614 EXPORT_SYMBOL_GPL(vmbus_close);
615 
616 int vmbus_sendpacket_ctl(struct vmbus_channel *channel, void *buffer,
617 			   u32 bufferlen, u64 requestid,
618 			   enum vmbus_packet_type type, u32 flags, bool kick_q)
619 {
620 	struct vmpacket_descriptor desc;
621 	u32 packetlen = sizeof(struct vmpacket_descriptor) + bufferlen;
622 	u32 packetlen_aligned = ALIGN(packetlen, sizeof(u64));
623 	struct kvec bufferlist[3];
624 	u64 aligned_data = 0;
625 	int ret;
626 	bool signal = false;
627 	int num_vecs = ((bufferlen != 0) ? 3 : 1);
628 
629 
630 	/* Setup the descriptor */
631 	desc.type = type; /* VmbusPacketTypeDataInBand; */
632 	desc.flags = flags; /* VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED; */
633 	/* in 8-bytes granularity */
634 	desc.offset8 = sizeof(struct vmpacket_descriptor) >> 3;
635 	desc.len8 = (u16)(packetlen_aligned >> 3);
636 	desc.trans_id = requestid;
637 
638 	bufferlist[0].iov_base = &desc;
639 	bufferlist[0].iov_len = sizeof(struct vmpacket_descriptor);
640 	bufferlist[1].iov_base = buffer;
641 	bufferlist[1].iov_len = bufferlen;
642 	bufferlist[2].iov_base = &aligned_data;
643 	bufferlist[2].iov_len = (packetlen_aligned - packetlen);
644 
645 	ret = hv_ringbuffer_write(&channel->outbound, bufferlist, num_vecs,
646 				  &signal);
647 
648 	/*
649 	 * Signalling the host is conditional on many factors:
650 	 * 1. The ring state changed from being empty to non-empty.
651 	 *    This is tracked by the variable "signal".
652 	 * 2. The variable kick_q tracks if more data will be placed
653 	 *    on the ring. We will not signal if more data is
654 	 *    to be placed.
655 	 *
656 	 * Based on the channel signal state, we will decide
657 	 * which signaling policy will be applied.
658 	 *
659 	 * If we cannot write to the ring-buffer; signal the host
660 	 * even if we may not have written anything. This is a rare
661 	 * enough condition that it should not matter.
662 	 */
663 
664 	if (channel->signal_policy)
665 		signal = true;
666 	else
667 		kick_q = true;
668 
669 	if (((ret == 0) && kick_q && signal) || (ret))
670 		vmbus_setevent(channel);
671 
672 	return ret;
673 }
674 EXPORT_SYMBOL(vmbus_sendpacket_ctl);
675 
676 /**
677  * vmbus_sendpacket() - Send the specified buffer on the given channel
678  * @channel: Pointer to vmbus_channel structure.
679  * @buffer: Pointer to the buffer you want to receive the data into.
680  * @bufferlen: Maximum size of what the the buffer will hold
681  * @requestid: Identifier of the request
682  * @type: Type of packet that is being send e.g. negotiate, time
683  * packet etc.
684  *
685  * Sends data in @buffer directly to hyper-v via the vmbus
686  * This will send the data unparsed to hyper-v.
687  *
688  * Mainly used by Hyper-V drivers.
689  */
690 int vmbus_sendpacket(struct vmbus_channel *channel, void *buffer,
691 			   u32 bufferlen, u64 requestid,
692 			   enum vmbus_packet_type type, u32 flags)
693 {
694 	return vmbus_sendpacket_ctl(channel, buffer, bufferlen, requestid,
695 				    type, flags, true);
696 }
697 EXPORT_SYMBOL(vmbus_sendpacket);
698 
699 /*
700  * vmbus_sendpacket_pagebuffer_ctl - Send a range of single-page buffer
701  * packets using a GPADL Direct packet type. This interface allows you
702  * to control notifying the host. This will be useful for sending
703  * batched data. Also the sender can control the send flags
704  * explicitly.
705  */
706 int vmbus_sendpacket_pagebuffer_ctl(struct vmbus_channel *channel,
707 				     struct hv_page_buffer pagebuffers[],
708 				     u32 pagecount, void *buffer, u32 bufferlen,
709 				     u64 requestid,
710 				     u32 flags,
711 				     bool kick_q)
712 {
713 	int ret;
714 	int i;
715 	struct vmbus_channel_packet_page_buffer desc;
716 	u32 descsize;
717 	u32 packetlen;
718 	u32 packetlen_aligned;
719 	struct kvec bufferlist[3];
720 	u64 aligned_data = 0;
721 	bool signal = false;
722 
723 	if (pagecount > MAX_PAGE_BUFFER_COUNT)
724 		return -EINVAL;
725 
726 
727 	/*
728 	 * Adjust the size down since vmbus_channel_packet_page_buffer is the
729 	 * largest size we support
730 	 */
731 	descsize = sizeof(struct vmbus_channel_packet_page_buffer) -
732 			  ((MAX_PAGE_BUFFER_COUNT - pagecount) *
733 			  sizeof(struct hv_page_buffer));
734 	packetlen = descsize + bufferlen;
735 	packetlen_aligned = ALIGN(packetlen, sizeof(u64));
736 
737 	/* Setup the descriptor */
738 	desc.type = VM_PKT_DATA_USING_GPA_DIRECT;
739 	desc.flags = flags;
740 	desc.dataoffset8 = descsize >> 3; /* in 8-bytes grandularity */
741 	desc.length8 = (u16)(packetlen_aligned >> 3);
742 	desc.transactionid = requestid;
743 	desc.rangecount = pagecount;
744 
745 	for (i = 0; i < pagecount; i++) {
746 		desc.range[i].len = pagebuffers[i].len;
747 		desc.range[i].offset = pagebuffers[i].offset;
748 		desc.range[i].pfn	 = pagebuffers[i].pfn;
749 	}
750 
751 	bufferlist[0].iov_base = &desc;
752 	bufferlist[0].iov_len = descsize;
753 	bufferlist[1].iov_base = buffer;
754 	bufferlist[1].iov_len = bufferlen;
755 	bufferlist[2].iov_base = &aligned_data;
756 	bufferlist[2].iov_len = (packetlen_aligned - packetlen);
757 
758 	ret = hv_ringbuffer_write(&channel->outbound, bufferlist, 3, &signal);
759 
760 	/*
761 	 * Signalling the host is conditional on many factors:
762 	 * 1. The ring state changed from being empty to non-empty.
763 	 *    This is tracked by the variable "signal".
764 	 * 2. The variable kick_q tracks if more data will be placed
765 	 *    on the ring. We will not signal if more data is
766 	 *    to be placed.
767 	 *
768 	 * Based on the channel signal state, we will decide
769 	 * which signaling policy will be applied.
770 	 *
771 	 * If we cannot write to the ring-buffer; signal the host
772 	 * even if we may not have written anything. This is a rare
773 	 * enough condition that it should not matter.
774 	 */
775 
776 	if (channel->signal_policy)
777 		signal = true;
778 	else
779 		kick_q = true;
780 
781 	if (((ret == 0) && kick_q && signal) || (ret))
782 		vmbus_setevent(channel);
783 
784 	return ret;
785 }
786 EXPORT_SYMBOL_GPL(vmbus_sendpacket_pagebuffer_ctl);
787 
788 /*
789  * vmbus_sendpacket_pagebuffer - Send a range of single-page buffer
790  * packets using a GPADL Direct packet type.
791  */
792 int vmbus_sendpacket_pagebuffer(struct vmbus_channel *channel,
793 				     struct hv_page_buffer pagebuffers[],
794 				     u32 pagecount, void *buffer, u32 bufferlen,
795 				     u64 requestid)
796 {
797 	u32 flags = VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED;
798 	return vmbus_sendpacket_pagebuffer_ctl(channel, pagebuffers, pagecount,
799 					       buffer, bufferlen, requestid,
800 					       flags, true);
801 
802 }
803 EXPORT_SYMBOL_GPL(vmbus_sendpacket_pagebuffer);
804 
805 /*
806  * vmbus_sendpacket_multipagebuffer - Send a multi-page buffer packet
807  * using a GPADL Direct packet type.
808  * The buffer includes the vmbus descriptor.
809  */
810 int vmbus_sendpacket_mpb_desc(struct vmbus_channel *channel,
811 			      struct vmbus_packet_mpb_array *desc,
812 			      u32 desc_size,
813 			      void *buffer, u32 bufferlen, u64 requestid)
814 {
815 	int ret;
816 	u32 packetlen;
817 	u32 packetlen_aligned;
818 	struct kvec bufferlist[3];
819 	u64 aligned_data = 0;
820 	bool signal = false;
821 
822 	packetlen = desc_size + bufferlen;
823 	packetlen_aligned = ALIGN(packetlen, sizeof(u64));
824 
825 	/* Setup the descriptor */
826 	desc->type = VM_PKT_DATA_USING_GPA_DIRECT;
827 	desc->flags = VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED;
828 	desc->dataoffset8 = desc_size >> 3; /* in 8-bytes grandularity */
829 	desc->length8 = (u16)(packetlen_aligned >> 3);
830 	desc->transactionid = requestid;
831 	desc->rangecount = 1;
832 
833 	bufferlist[0].iov_base = desc;
834 	bufferlist[0].iov_len = desc_size;
835 	bufferlist[1].iov_base = buffer;
836 	bufferlist[1].iov_len = bufferlen;
837 	bufferlist[2].iov_base = &aligned_data;
838 	bufferlist[2].iov_len = (packetlen_aligned - packetlen);
839 
840 	ret = hv_ringbuffer_write(&channel->outbound, bufferlist, 3, &signal);
841 
842 	if (ret == 0 && signal)
843 		vmbus_setevent(channel);
844 
845 	return ret;
846 }
847 EXPORT_SYMBOL_GPL(vmbus_sendpacket_mpb_desc);
848 
849 /*
850  * vmbus_sendpacket_multipagebuffer - Send a multi-page buffer packet
851  * using a GPADL Direct packet type.
852  */
853 int vmbus_sendpacket_multipagebuffer(struct vmbus_channel *channel,
854 				struct hv_multipage_buffer *multi_pagebuffer,
855 				void *buffer, u32 bufferlen, u64 requestid)
856 {
857 	int ret;
858 	struct vmbus_channel_packet_multipage_buffer desc;
859 	u32 descsize;
860 	u32 packetlen;
861 	u32 packetlen_aligned;
862 	struct kvec bufferlist[3];
863 	u64 aligned_data = 0;
864 	bool signal = false;
865 	u32 pfncount = NUM_PAGES_SPANNED(multi_pagebuffer->offset,
866 					 multi_pagebuffer->len);
867 
868 	if (pfncount > MAX_MULTIPAGE_BUFFER_COUNT)
869 		return -EINVAL;
870 
871 	/*
872 	 * Adjust the size down since vmbus_channel_packet_multipage_buffer is
873 	 * the largest size we support
874 	 */
875 	descsize = sizeof(struct vmbus_channel_packet_multipage_buffer) -
876 			  ((MAX_MULTIPAGE_BUFFER_COUNT - pfncount) *
877 			  sizeof(u64));
878 	packetlen = descsize + bufferlen;
879 	packetlen_aligned = ALIGN(packetlen, sizeof(u64));
880 
881 
882 	/* Setup the descriptor */
883 	desc.type = VM_PKT_DATA_USING_GPA_DIRECT;
884 	desc.flags = VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED;
885 	desc.dataoffset8 = descsize >> 3; /* in 8-bytes grandularity */
886 	desc.length8 = (u16)(packetlen_aligned >> 3);
887 	desc.transactionid = requestid;
888 	desc.rangecount = 1;
889 
890 	desc.range.len = multi_pagebuffer->len;
891 	desc.range.offset = multi_pagebuffer->offset;
892 
893 	memcpy(desc.range.pfn_array, multi_pagebuffer->pfn_array,
894 	       pfncount * sizeof(u64));
895 
896 	bufferlist[0].iov_base = &desc;
897 	bufferlist[0].iov_len = descsize;
898 	bufferlist[1].iov_base = buffer;
899 	bufferlist[1].iov_len = bufferlen;
900 	bufferlist[2].iov_base = &aligned_data;
901 	bufferlist[2].iov_len = (packetlen_aligned - packetlen);
902 
903 	ret = hv_ringbuffer_write(&channel->outbound, bufferlist, 3, &signal);
904 
905 	if (ret == 0 && signal)
906 		vmbus_setevent(channel);
907 
908 	return ret;
909 }
910 EXPORT_SYMBOL_GPL(vmbus_sendpacket_multipagebuffer);
911 
912 /**
913  * vmbus_recvpacket() - Retrieve the user packet on the specified channel
914  * @channel: Pointer to vmbus_channel structure.
915  * @buffer: Pointer to the buffer you want to receive the data into.
916  * @bufferlen: Maximum size of what the the buffer will hold
917  * @buffer_actual_len: The actual size of the data after it was received
918  * @requestid: Identifier of the request
919  *
920  * Receives directly from the hyper-v vmbus and puts the data it received
921  * into Buffer. This will receive the data unparsed from hyper-v.
922  *
923  * Mainly used by Hyper-V drivers.
924  */
925 static inline int
926 __vmbus_recvpacket(struct vmbus_channel *channel, void *buffer,
927 		   u32 bufferlen, u32 *buffer_actual_len, u64 *requestid,
928 		   bool raw)
929 {
930 	int ret;
931 	bool signal = false;
932 
933 	ret = hv_ringbuffer_read(&channel->inbound, buffer, bufferlen,
934 				 buffer_actual_len, requestid, &signal, raw);
935 
936 	if (signal)
937 		vmbus_setevent(channel);
938 
939 	return ret;
940 }
941 
942 int vmbus_recvpacket(struct vmbus_channel *channel, void *buffer,
943 		     u32 bufferlen, u32 *buffer_actual_len,
944 		     u64 *requestid)
945 {
946 	return __vmbus_recvpacket(channel, buffer, bufferlen,
947 				  buffer_actual_len, requestid, false);
948 }
949 EXPORT_SYMBOL(vmbus_recvpacket);
950 
951 /*
952  * vmbus_recvpacket_raw - Retrieve the raw packet on the specified channel
953  */
954 int vmbus_recvpacket_raw(struct vmbus_channel *channel, void *buffer,
955 			      u32 bufferlen, u32 *buffer_actual_len,
956 			      u64 *requestid)
957 {
958 	return __vmbus_recvpacket(channel, buffer, bufferlen,
959 				  buffer_actual_len, requestid, true);
960 }
961 EXPORT_SYMBOL_GPL(vmbus_recvpacket_raw);
962