xref: /openbmc/linux/drivers/hv/channel.c (revision 82003e04)
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 	/*
47 	 * For channels marked as in "low latency" mode
48 	 * bypass the monitor page mechanism.
49 	 */
50 	if ((channel->offermsg.monitor_allocated) &&
51 	    (!channel->low_latency)) {
52 		/* Each u32 represents 32 channels */
53 		sync_set_bit(channel->offermsg.child_relid & 31,
54 			(unsigned long *) vmbus_connection.send_int_page +
55 			(channel->offermsg.child_relid >> 5));
56 
57 		/* Get the child to parent monitor page */
58 		monitorpage = vmbus_connection.monitor_pages[1];
59 
60 		sync_set_bit(channel->monitor_bit,
61 			(unsigned long *)&monitorpage->trigger_group
62 					[channel->monitor_grp].pending);
63 
64 	} else {
65 		vmbus_set_event(channel);
66 	}
67 }
68 
69 /*
70  * vmbus_open - Open the specified channel.
71  */
72 int vmbus_open(struct vmbus_channel *newchannel, u32 send_ringbuffer_size,
73 		     u32 recv_ringbuffer_size, void *userdata, u32 userdatalen,
74 		     void (*onchannelcallback)(void *context), void *context)
75 {
76 	struct vmbus_channel_open_channel *open_msg;
77 	struct vmbus_channel_msginfo *open_info = NULL;
78 	unsigned long flags;
79 	int ret, err = 0;
80 	struct page *page;
81 
82 	if (send_ringbuffer_size % PAGE_SIZE ||
83 	    recv_ringbuffer_size % PAGE_SIZE)
84 		return -EINVAL;
85 
86 	spin_lock_irqsave(&newchannel->lock, flags);
87 	if (newchannel->state == CHANNEL_OPEN_STATE) {
88 		newchannel->state = CHANNEL_OPENING_STATE;
89 	} else {
90 		spin_unlock_irqrestore(&newchannel->lock, flags);
91 		return -EINVAL;
92 	}
93 	spin_unlock_irqrestore(&newchannel->lock, flags);
94 
95 	newchannel->onchannel_callback = onchannelcallback;
96 	newchannel->channel_callback_context = context;
97 
98 	/* Allocate the ring buffer */
99 	page = alloc_pages_node(cpu_to_node(newchannel->target_cpu),
100 				GFP_KERNEL|__GFP_ZERO,
101 				get_order(send_ringbuffer_size +
102 				recv_ringbuffer_size));
103 
104 	if (!page)
105 		page = alloc_pages(GFP_KERNEL|__GFP_ZERO,
106 				   get_order(send_ringbuffer_size +
107 					     recv_ringbuffer_size));
108 
109 	if (!page) {
110 		err = -ENOMEM;
111 		goto error_set_chnstate;
112 	}
113 
114 	newchannel->ringbuffer_pages = page_address(page);
115 	newchannel->ringbuffer_pagecount = (send_ringbuffer_size +
116 					   recv_ringbuffer_size) >> PAGE_SHIFT;
117 
118 	ret = hv_ringbuffer_init(&newchannel->outbound, page,
119 				 send_ringbuffer_size >> PAGE_SHIFT);
120 
121 	if (ret != 0) {
122 		err = ret;
123 		goto error_free_pages;
124 	}
125 
126 	ret = hv_ringbuffer_init(&newchannel->inbound,
127 				 &page[send_ringbuffer_size >> PAGE_SHIFT],
128 				 recv_ringbuffer_size >> PAGE_SHIFT);
129 	if (ret != 0) {
130 		err = ret;
131 		goto error_free_pages;
132 	}
133 
134 
135 	/* Establish the gpadl for the ring buffer */
136 	newchannel->ringbuffer_gpadlhandle = 0;
137 
138 	ret = vmbus_establish_gpadl(newchannel,
139 				    page_address(page),
140 				    send_ringbuffer_size +
141 				    recv_ringbuffer_size,
142 				    &newchannel->ringbuffer_gpadlhandle);
143 
144 	if (ret != 0) {
145 		err = ret;
146 		goto error_free_pages;
147 	}
148 
149 	/* Create and init the channel open message */
150 	open_info = kmalloc(sizeof(*open_info) +
151 			   sizeof(struct vmbus_channel_open_channel),
152 			   GFP_KERNEL);
153 	if (!open_info) {
154 		err = -ENOMEM;
155 		goto error_free_gpadl;
156 	}
157 
158 	init_completion(&open_info->waitevent);
159 
160 	open_msg = (struct vmbus_channel_open_channel *)open_info->msg;
161 	open_msg->header.msgtype = CHANNELMSG_OPENCHANNEL;
162 	open_msg->openid = newchannel->offermsg.child_relid;
163 	open_msg->child_relid = newchannel->offermsg.child_relid;
164 	open_msg->ringbuffer_gpadlhandle = newchannel->ringbuffer_gpadlhandle;
165 	open_msg->downstream_ringbuffer_pageoffset = send_ringbuffer_size >>
166 						  PAGE_SHIFT;
167 	open_msg->target_vp = newchannel->target_vp;
168 
169 	if (userdatalen > MAX_USER_DEFINED_BYTES) {
170 		err = -EINVAL;
171 		goto error_free_gpadl;
172 	}
173 
174 	if (userdatalen)
175 		memcpy(open_msg->userdata, userdata, userdatalen);
176 
177 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
178 	list_add_tail(&open_info->msglistentry,
179 		      &vmbus_connection.chn_msg_list);
180 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
181 
182 	ret = vmbus_post_msg(open_msg,
183 			       sizeof(struct vmbus_channel_open_channel));
184 
185 	if (ret != 0) {
186 		err = ret;
187 		goto error_clean_msglist;
188 	}
189 
190 	wait_for_completion(&open_info->waitevent);
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_free_gpadl;
199 	}
200 
201 	newchannel->state = CHANNEL_OPENED_STATE;
202 	kfree(open_info);
203 	return 0;
204 
205 error_clean_msglist:
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_free_gpadl:
211 	vmbus_teardown_gpadl(newchannel, newchannel->ringbuffer_gpadlhandle);
212 	kfree(open_info);
213 error_free_pages:
214 	hv_ringbuffer_cleanup(&newchannel->outbound);
215 	hv_ringbuffer_cleanup(&newchannel->inbound);
216 	__free_pages(page,
217 		     get_order(send_ringbuffer_size + recv_ringbuffer_size));
218 error_set_chnstate:
219 	newchannel->state = CHANNEL_OPEN_STATE;
220 	return err;
221 }
222 EXPORT_SYMBOL_GPL(vmbus_open);
223 
224 /* Used for Hyper-V Socket: a guest client's connect() to the host */
225 int vmbus_send_tl_connect_request(const uuid_le *shv_guest_servie_id,
226 				  const uuid_le *shv_host_servie_id)
227 {
228 	struct vmbus_channel_tl_connect_request conn_msg;
229 
230 	memset(&conn_msg, 0, sizeof(conn_msg));
231 	conn_msg.header.msgtype = CHANNELMSG_TL_CONNECT_REQUEST;
232 	conn_msg.guest_endpoint_id = *shv_guest_servie_id;
233 	conn_msg.host_service_id = *shv_host_servie_id;
234 
235 	return vmbus_post_msg(&conn_msg, sizeof(conn_msg));
236 }
237 EXPORT_SYMBOL_GPL(vmbus_send_tl_connect_request);
238 
239 /*
240  * create_gpadl_header - Creates a gpadl for the specified buffer
241  */
242 static int create_gpadl_header(void *kbuffer, u32 size,
243 			       struct vmbus_channel_msginfo **msginfo)
244 {
245 	int i;
246 	int pagecount;
247 	struct vmbus_channel_gpadl_header *gpadl_header;
248 	struct vmbus_channel_gpadl_body *gpadl_body;
249 	struct vmbus_channel_msginfo *msgheader;
250 	struct vmbus_channel_msginfo *msgbody = NULL;
251 	u32 msgsize;
252 
253 	int pfnsum, pfncount, pfnleft, pfncurr, pfnsize;
254 
255 	pagecount = size >> PAGE_SHIFT;
256 
257 	/* do we need a gpadl body msg */
258 	pfnsize = MAX_SIZE_CHANNEL_MESSAGE -
259 		  sizeof(struct vmbus_channel_gpadl_header) -
260 		  sizeof(struct gpa_range);
261 	pfncount = pfnsize / sizeof(u64);
262 
263 	if (pagecount > pfncount) {
264 		/* we need a gpadl body */
265 		/* fill in the header */
266 		msgsize = sizeof(struct vmbus_channel_msginfo) +
267 			  sizeof(struct vmbus_channel_gpadl_header) +
268 			  sizeof(struct gpa_range) + pfncount * sizeof(u64);
269 		msgheader =  kzalloc(msgsize, GFP_KERNEL);
270 		if (!msgheader)
271 			goto nomem;
272 
273 		INIT_LIST_HEAD(&msgheader->submsglist);
274 		msgheader->msgsize = msgsize;
275 
276 		gpadl_header = (struct vmbus_channel_gpadl_header *)
277 			msgheader->msg;
278 		gpadl_header->rangecount = 1;
279 		gpadl_header->range_buflen = sizeof(struct gpa_range) +
280 					 pagecount * sizeof(u64);
281 		gpadl_header->range[0].byte_offset = 0;
282 		gpadl_header->range[0].byte_count = size;
283 		for (i = 0; i < pfncount; i++)
284 			gpadl_header->range[0].pfn_array[i] = slow_virt_to_phys(
285 				kbuffer + PAGE_SIZE * i) >> PAGE_SHIFT;
286 		*msginfo = msgheader;
287 
288 		pfnsum = pfncount;
289 		pfnleft = pagecount - pfncount;
290 
291 		/* how many pfns can we fit */
292 		pfnsize = MAX_SIZE_CHANNEL_MESSAGE -
293 			  sizeof(struct vmbus_channel_gpadl_body);
294 		pfncount = pfnsize / sizeof(u64);
295 
296 		/* fill in the body */
297 		while (pfnleft) {
298 			if (pfnleft > pfncount)
299 				pfncurr = pfncount;
300 			else
301 				pfncurr = pfnleft;
302 
303 			msgsize = sizeof(struct vmbus_channel_msginfo) +
304 				  sizeof(struct vmbus_channel_gpadl_body) +
305 				  pfncurr * sizeof(u64);
306 			msgbody = kzalloc(msgsize, GFP_KERNEL);
307 
308 			if (!msgbody) {
309 				struct vmbus_channel_msginfo *pos = NULL;
310 				struct vmbus_channel_msginfo *tmp = NULL;
311 				/*
312 				 * Free up all the allocated messages.
313 				 */
314 				list_for_each_entry_safe(pos, tmp,
315 					&msgheader->submsglist,
316 					msglistentry) {
317 
318 					list_del(&pos->msglistentry);
319 					kfree(pos);
320 				}
321 
322 				goto nomem;
323 			}
324 
325 			msgbody->msgsize = msgsize;
326 			gpadl_body =
327 				(struct vmbus_channel_gpadl_body *)msgbody->msg;
328 
329 			/*
330 			 * Gpadl is u32 and we are using a pointer which could
331 			 * be 64-bit
332 			 * This is governed by the guest/host protocol and
333 			 * so the hypervisor gurantees that this is ok.
334 			 */
335 			for (i = 0; i < pfncurr; i++)
336 				gpadl_body->pfn[i] = slow_virt_to_phys(
337 					kbuffer + PAGE_SIZE * (pfnsum + i)) >>
338 					PAGE_SHIFT;
339 
340 			/* add to msg header */
341 			list_add_tail(&msgbody->msglistentry,
342 				      &msgheader->submsglist);
343 			pfnsum += pfncurr;
344 			pfnleft -= pfncurr;
345 		}
346 	} else {
347 		/* everything fits in a header */
348 		msgsize = sizeof(struct vmbus_channel_msginfo) +
349 			  sizeof(struct vmbus_channel_gpadl_header) +
350 			  sizeof(struct gpa_range) + pagecount * sizeof(u64);
351 		msgheader = kzalloc(msgsize, GFP_KERNEL);
352 		if (msgheader == NULL)
353 			goto nomem;
354 
355 		INIT_LIST_HEAD(&msgheader->submsglist);
356 		msgheader->msgsize = msgsize;
357 
358 		gpadl_header = (struct vmbus_channel_gpadl_header *)
359 			msgheader->msg;
360 		gpadl_header->rangecount = 1;
361 		gpadl_header->range_buflen = sizeof(struct gpa_range) +
362 					 pagecount * sizeof(u64);
363 		gpadl_header->range[0].byte_offset = 0;
364 		gpadl_header->range[0].byte_count = size;
365 		for (i = 0; i < pagecount; i++)
366 			gpadl_header->range[0].pfn_array[i] = slow_virt_to_phys(
367 				kbuffer + PAGE_SIZE * i) >> PAGE_SHIFT;
368 
369 		*msginfo = msgheader;
370 	}
371 
372 	return 0;
373 nomem:
374 	kfree(msgheader);
375 	kfree(msgbody);
376 	return -ENOMEM;
377 }
378 
379 /*
380  * vmbus_establish_gpadl - Estabish a GPADL for the specified buffer
381  *
382  * @channel: a channel
383  * @kbuffer: from kmalloc or vmalloc
384  * @size: page-size multiple
385  * @gpadl_handle: some funky thing
386  */
387 int vmbus_establish_gpadl(struct vmbus_channel *channel, void *kbuffer,
388 			       u32 size, u32 *gpadl_handle)
389 {
390 	struct vmbus_channel_gpadl_header *gpadlmsg;
391 	struct vmbus_channel_gpadl_body *gpadl_body;
392 	struct vmbus_channel_msginfo *msginfo = NULL;
393 	struct vmbus_channel_msginfo *submsginfo, *tmp;
394 	struct list_head *curr;
395 	u32 next_gpadl_handle;
396 	unsigned long flags;
397 	int ret = 0;
398 
399 	next_gpadl_handle =
400 		(atomic_inc_return(&vmbus_connection.next_gpadl_handle) - 1);
401 
402 	ret = create_gpadl_header(kbuffer, size, &msginfo);
403 	if (ret)
404 		return ret;
405 
406 	init_completion(&msginfo->waitevent);
407 
408 	gpadlmsg = (struct vmbus_channel_gpadl_header *)msginfo->msg;
409 	gpadlmsg->header.msgtype = CHANNELMSG_GPADL_HEADER;
410 	gpadlmsg->child_relid = channel->offermsg.child_relid;
411 	gpadlmsg->gpadl = next_gpadl_handle;
412 
413 
414 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
415 	list_add_tail(&msginfo->msglistentry,
416 		      &vmbus_connection.chn_msg_list);
417 
418 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
419 
420 	ret = vmbus_post_msg(gpadlmsg, msginfo->msgsize -
421 			       sizeof(*msginfo));
422 	if (ret != 0)
423 		goto cleanup;
424 
425 	list_for_each(curr, &msginfo->submsglist) {
426 		submsginfo = (struct vmbus_channel_msginfo *)curr;
427 		gpadl_body =
428 			(struct vmbus_channel_gpadl_body *)submsginfo->msg;
429 
430 		gpadl_body->header.msgtype =
431 			CHANNELMSG_GPADL_BODY;
432 		gpadl_body->gpadl = next_gpadl_handle;
433 
434 		ret = vmbus_post_msg(gpadl_body,
435 				     submsginfo->msgsize -
436 				     sizeof(*submsginfo));
437 		if (ret != 0)
438 			goto cleanup;
439 
440 	}
441 	wait_for_completion(&msginfo->waitevent);
442 
443 	/* At this point, we received the gpadl created msg */
444 	*gpadl_handle = gpadlmsg->gpadl;
445 
446 cleanup:
447 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
448 	list_del(&msginfo->msglistentry);
449 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
450 	list_for_each_entry_safe(submsginfo, tmp, &msginfo->submsglist,
451 				 msglistentry) {
452 		kfree(submsginfo);
453 	}
454 
455 	kfree(msginfo);
456 	return ret;
457 }
458 EXPORT_SYMBOL_GPL(vmbus_establish_gpadl);
459 
460 /*
461  * vmbus_teardown_gpadl -Teardown the specified GPADL handle
462  */
463 int vmbus_teardown_gpadl(struct vmbus_channel *channel, u32 gpadl_handle)
464 {
465 	struct vmbus_channel_gpadl_teardown *msg;
466 	struct vmbus_channel_msginfo *info;
467 	unsigned long flags;
468 	int ret;
469 
470 	info = kmalloc(sizeof(*info) +
471 		       sizeof(struct vmbus_channel_gpadl_teardown), GFP_KERNEL);
472 	if (!info)
473 		return -ENOMEM;
474 
475 	init_completion(&info->waitevent);
476 
477 	msg = (struct vmbus_channel_gpadl_teardown *)info->msg;
478 
479 	msg->header.msgtype = CHANNELMSG_GPADL_TEARDOWN;
480 	msg->child_relid = channel->offermsg.child_relid;
481 	msg->gpadl = gpadl_handle;
482 
483 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
484 	list_add_tail(&info->msglistentry,
485 		      &vmbus_connection.chn_msg_list);
486 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
487 	ret = vmbus_post_msg(msg,
488 			       sizeof(struct vmbus_channel_gpadl_teardown));
489 
490 	if (ret)
491 		goto post_msg_err;
492 
493 	wait_for_completion(&info->waitevent);
494 
495 post_msg_err:
496 	spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
497 	list_del(&info->msglistentry);
498 	spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
499 
500 	kfree(info);
501 	return ret;
502 }
503 EXPORT_SYMBOL_GPL(vmbus_teardown_gpadl);
504 
505 static void reset_channel_cb(void *arg)
506 {
507 	struct vmbus_channel *channel = arg;
508 
509 	channel->onchannel_callback = NULL;
510 }
511 
512 static int vmbus_close_internal(struct vmbus_channel *channel)
513 {
514 	struct vmbus_channel_close_channel *msg;
515 	int ret;
516 
517 	/*
518 	 * process_chn_event(), running in the tasklet, can race
519 	 * with vmbus_close_internal() in the case of SMP guest, e.g., when
520 	 * the former is accessing channel->inbound.ring_buffer, the latter
521 	 * could be freeing the ring_buffer pages.
522 	 *
523 	 * To resolve the race, we can serialize them by disabling the
524 	 * tasklet when the latter is running here.
525 	 */
526 	hv_event_tasklet_disable(channel);
527 
528 	/*
529 	 * In case a device driver's probe() fails (e.g.,
530 	 * util_probe() -> vmbus_open() returns -ENOMEM) and the device is
531 	 * rescinded later (e.g., we dynamically disble an Integrated Service
532 	 * in Hyper-V Manager), the driver's remove() invokes vmbus_close():
533 	 * here we should skip most of the below cleanup work.
534 	 */
535 	if (channel->state != CHANNEL_OPENED_STATE) {
536 		ret = -EINVAL;
537 		goto out;
538 	}
539 
540 	channel->state = CHANNEL_OPEN_STATE;
541 	channel->sc_creation_callback = NULL;
542 	/* Stop callback and cancel the timer asap */
543 	if (channel->target_cpu != get_cpu()) {
544 		put_cpu();
545 		smp_call_function_single(channel->target_cpu, reset_channel_cb,
546 					 channel, true);
547 	} else {
548 		reset_channel_cb(channel);
549 		put_cpu();
550 	}
551 
552 	/* Send a closing message */
553 
554 	msg = &channel->close_msg.msg;
555 
556 	msg->header.msgtype = CHANNELMSG_CLOSECHANNEL;
557 	msg->child_relid = channel->offermsg.child_relid;
558 
559 	ret = vmbus_post_msg(msg, sizeof(struct vmbus_channel_close_channel));
560 
561 	if (ret) {
562 		pr_err("Close failed: close post msg return is %d\n", ret);
563 		/*
564 		 * If we failed to post the close msg,
565 		 * it is perhaps better to leak memory.
566 		 */
567 		goto out;
568 	}
569 
570 	/* Tear down the gpadl for the channel's ring buffer */
571 	if (channel->ringbuffer_gpadlhandle) {
572 		ret = vmbus_teardown_gpadl(channel,
573 					   channel->ringbuffer_gpadlhandle);
574 		if (ret) {
575 			pr_err("Close failed: teardown gpadl return %d\n", ret);
576 			/*
577 			 * If we failed to teardown gpadl,
578 			 * it is perhaps better to leak memory.
579 			 */
580 			goto out;
581 		}
582 	}
583 
584 	/* Cleanup the ring buffers for this channel */
585 	hv_ringbuffer_cleanup(&channel->outbound);
586 	hv_ringbuffer_cleanup(&channel->inbound);
587 
588 	free_pages((unsigned long)channel->ringbuffer_pages,
589 		get_order(channel->ringbuffer_pagecount * PAGE_SIZE));
590 
591 out:
592 	hv_event_tasklet_enable(channel);
593 
594 	return ret;
595 }
596 
597 /*
598  * vmbus_close - Close the specified channel
599  */
600 void vmbus_close(struct vmbus_channel *channel)
601 {
602 	struct list_head *cur, *tmp;
603 	struct vmbus_channel *cur_channel;
604 
605 	if (channel->primary_channel != NULL) {
606 		/*
607 		 * We will only close sub-channels when
608 		 * the primary is closed.
609 		 */
610 		return;
611 	}
612 	/*
613 	 * Close all the sub-channels first and then close the
614 	 * primary channel.
615 	 */
616 	list_for_each_safe(cur, tmp, &channel->sc_list) {
617 		cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
618 		if (cur_channel->state != CHANNEL_OPENED_STATE)
619 			continue;
620 		vmbus_close_internal(cur_channel);
621 	}
622 	/*
623 	 * Now close the primary.
624 	 */
625 	vmbus_close_internal(channel);
626 }
627 EXPORT_SYMBOL_GPL(vmbus_close);
628 
629 int vmbus_sendpacket_ctl(struct vmbus_channel *channel, void *buffer,
630 			   u32 bufferlen, u64 requestid,
631 			   enum vmbus_packet_type type, u32 flags, bool kick_q)
632 {
633 	struct vmpacket_descriptor desc;
634 	u32 packetlen = sizeof(struct vmpacket_descriptor) + bufferlen;
635 	u32 packetlen_aligned = ALIGN(packetlen, sizeof(u64));
636 	struct kvec bufferlist[3];
637 	u64 aligned_data = 0;
638 	int ret;
639 	bool signal = false;
640 	bool lock = channel->acquire_ring_lock;
641 	int num_vecs = ((bufferlen != 0) ? 3 : 1);
642 
643 
644 	/* Setup the descriptor */
645 	desc.type = type; /* VmbusPacketTypeDataInBand; */
646 	desc.flags = flags; /* VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED; */
647 	/* in 8-bytes granularity */
648 	desc.offset8 = sizeof(struct vmpacket_descriptor) >> 3;
649 	desc.len8 = (u16)(packetlen_aligned >> 3);
650 	desc.trans_id = requestid;
651 
652 	bufferlist[0].iov_base = &desc;
653 	bufferlist[0].iov_len = sizeof(struct vmpacket_descriptor);
654 	bufferlist[1].iov_base = buffer;
655 	bufferlist[1].iov_len = bufferlen;
656 	bufferlist[2].iov_base = &aligned_data;
657 	bufferlist[2].iov_len = (packetlen_aligned - packetlen);
658 
659 	ret = hv_ringbuffer_write(&channel->outbound, bufferlist, num_vecs,
660 				  &signal, lock, channel->signal_policy);
661 
662 	/*
663 	 * Signalling the host is conditional on many factors:
664 	 * 1. The ring state changed from being empty to non-empty.
665 	 *    This is tracked by the variable "signal".
666 	 * 2. The variable kick_q tracks if more data will be placed
667 	 *    on the ring. We will not signal if more data is
668 	 *    to be placed.
669 	 *
670 	 * Based on the channel signal state, we will decide
671 	 * which signaling policy will be applied.
672 	 *
673 	 * If we cannot write to the ring-buffer; signal the host
674 	 * even if we may not have written anything. This is a rare
675 	 * enough condition that it should not matter.
676 	 * NOTE: in this case, the hvsock channel is an exception, because
677 	 * it looks the host side's hvsock implementation has a throttling
678 	 * mechanism which can hurt the performance otherwise.
679 	 */
680 
681 	if (((ret == 0) && kick_q && signal) ||
682 	    (ret && !is_hvsock_channel(channel)))
683 		vmbus_setevent(channel);
684 
685 	return ret;
686 }
687 EXPORT_SYMBOL(vmbus_sendpacket_ctl);
688 
689 /**
690  * vmbus_sendpacket() - Send the specified buffer on the given channel
691  * @channel: Pointer to vmbus_channel structure.
692  * @buffer: Pointer to the buffer you want to receive the data into.
693  * @bufferlen: Maximum size of what the the buffer will hold
694  * @requestid: Identifier of the request
695  * @type: Type of packet that is being send e.g. negotiate, time
696  * packet etc.
697  *
698  * Sends data in @buffer directly to hyper-v via the vmbus
699  * This will send the data unparsed to hyper-v.
700  *
701  * Mainly used by Hyper-V drivers.
702  */
703 int vmbus_sendpacket(struct vmbus_channel *channel, void *buffer,
704 			   u32 bufferlen, u64 requestid,
705 			   enum vmbus_packet_type type, u32 flags)
706 {
707 	return vmbus_sendpacket_ctl(channel, buffer, bufferlen, requestid,
708 				    type, flags, true);
709 }
710 EXPORT_SYMBOL(vmbus_sendpacket);
711 
712 /*
713  * vmbus_sendpacket_pagebuffer_ctl - Send a range of single-page buffer
714  * packets using a GPADL Direct packet type. This interface allows you
715  * to control notifying the host. This will be useful for sending
716  * batched data. Also the sender can control the send flags
717  * explicitly.
718  */
719 int vmbus_sendpacket_pagebuffer_ctl(struct vmbus_channel *channel,
720 				     struct hv_page_buffer pagebuffers[],
721 				     u32 pagecount, void *buffer, u32 bufferlen,
722 				     u64 requestid,
723 				     u32 flags,
724 				     bool kick_q)
725 {
726 	int ret;
727 	int i;
728 	struct vmbus_channel_packet_page_buffer desc;
729 	u32 descsize;
730 	u32 packetlen;
731 	u32 packetlen_aligned;
732 	struct kvec bufferlist[3];
733 	u64 aligned_data = 0;
734 	bool signal = false;
735 	bool lock = channel->acquire_ring_lock;
736 
737 	if (pagecount > MAX_PAGE_BUFFER_COUNT)
738 		return -EINVAL;
739 
740 
741 	/*
742 	 * Adjust the size down since vmbus_channel_packet_page_buffer is the
743 	 * largest size we support
744 	 */
745 	descsize = sizeof(struct vmbus_channel_packet_page_buffer) -
746 			  ((MAX_PAGE_BUFFER_COUNT - pagecount) *
747 			  sizeof(struct hv_page_buffer));
748 	packetlen = descsize + bufferlen;
749 	packetlen_aligned = ALIGN(packetlen, sizeof(u64));
750 
751 	/* Setup the descriptor */
752 	desc.type = VM_PKT_DATA_USING_GPA_DIRECT;
753 	desc.flags = flags;
754 	desc.dataoffset8 = descsize >> 3; /* in 8-bytes grandularity */
755 	desc.length8 = (u16)(packetlen_aligned >> 3);
756 	desc.transactionid = requestid;
757 	desc.rangecount = pagecount;
758 
759 	for (i = 0; i < pagecount; i++) {
760 		desc.range[i].len = pagebuffers[i].len;
761 		desc.range[i].offset = pagebuffers[i].offset;
762 		desc.range[i].pfn	 = pagebuffers[i].pfn;
763 	}
764 
765 	bufferlist[0].iov_base = &desc;
766 	bufferlist[0].iov_len = descsize;
767 	bufferlist[1].iov_base = buffer;
768 	bufferlist[1].iov_len = bufferlen;
769 	bufferlist[2].iov_base = &aligned_data;
770 	bufferlist[2].iov_len = (packetlen_aligned - packetlen);
771 
772 	ret = hv_ringbuffer_write(&channel->outbound, bufferlist, 3,
773 				  &signal, lock, channel->signal_policy);
774 
775 	/*
776 	 * Signalling the host is conditional on many factors:
777 	 * 1. The ring state changed from being empty to non-empty.
778 	 *    This is tracked by the variable "signal".
779 	 * 2. The variable kick_q tracks if more data will be placed
780 	 *    on the ring. We will not signal if more data is
781 	 *    to be placed.
782 	 *
783 	 * Based on the channel signal state, we will decide
784 	 * which signaling policy will be applied.
785 	 *
786 	 * If we cannot write to the ring-buffer; signal the host
787 	 * even if we may not have written anything. This is a rare
788 	 * enough condition that it should not matter.
789 	 */
790 
791 	if (((ret == 0) && kick_q && signal) || (ret))
792 		vmbus_setevent(channel);
793 
794 	return ret;
795 }
796 EXPORT_SYMBOL_GPL(vmbus_sendpacket_pagebuffer_ctl);
797 
798 /*
799  * vmbus_sendpacket_pagebuffer - Send a range of single-page buffer
800  * packets using a GPADL Direct packet type.
801  */
802 int vmbus_sendpacket_pagebuffer(struct vmbus_channel *channel,
803 				     struct hv_page_buffer pagebuffers[],
804 				     u32 pagecount, void *buffer, u32 bufferlen,
805 				     u64 requestid)
806 {
807 	u32 flags = VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED;
808 	return vmbus_sendpacket_pagebuffer_ctl(channel, pagebuffers, pagecount,
809 					       buffer, bufferlen, requestid,
810 					       flags, true);
811 
812 }
813 EXPORT_SYMBOL_GPL(vmbus_sendpacket_pagebuffer);
814 
815 /*
816  * vmbus_sendpacket_multipagebuffer - Send a multi-page buffer packet
817  * using a GPADL Direct packet type.
818  * The buffer includes the vmbus descriptor.
819  */
820 int vmbus_sendpacket_mpb_desc(struct vmbus_channel *channel,
821 			      struct vmbus_packet_mpb_array *desc,
822 			      u32 desc_size,
823 			      void *buffer, u32 bufferlen, u64 requestid)
824 {
825 	int ret;
826 	u32 packetlen;
827 	u32 packetlen_aligned;
828 	struct kvec bufferlist[3];
829 	u64 aligned_data = 0;
830 	bool signal = false;
831 	bool lock = channel->acquire_ring_lock;
832 
833 	packetlen = desc_size + bufferlen;
834 	packetlen_aligned = ALIGN(packetlen, sizeof(u64));
835 
836 	/* Setup the descriptor */
837 	desc->type = VM_PKT_DATA_USING_GPA_DIRECT;
838 	desc->flags = VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED;
839 	desc->dataoffset8 = desc_size >> 3; /* in 8-bytes grandularity */
840 	desc->length8 = (u16)(packetlen_aligned >> 3);
841 	desc->transactionid = requestid;
842 	desc->rangecount = 1;
843 
844 	bufferlist[0].iov_base = desc;
845 	bufferlist[0].iov_len = desc_size;
846 	bufferlist[1].iov_base = buffer;
847 	bufferlist[1].iov_len = bufferlen;
848 	bufferlist[2].iov_base = &aligned_data;
849 	bufferlist[2].iov_len = (packetlen_aligned - packetlen);
850 
851 	ret = hv_ringbuffer_write(&channel->outbound, bufferlist, 3,
852 				  &signal, lock, channel->signal_policy);
853 
854 	if (ret == 0 && signal)
855 		vmbus_setevent(channel);
856 
857 	return ret;
858 }
859 EXPORT_SYMBOL_GPL(vmbus_sendpacket_mpb_desc);
860 
861 /*
862  * vmbus_sendpacket_multipagebuffer - Send a multi-page buffer packet
863  * using a GPADL Direct packet type.
864  */
865 int vmbus_sendpacket_multipagebuffer(struct vmbus_channel *channel,
866 				struct hv_multipage_buffer *multi_pagebuffer,
867 				void *buffer, u32 bufferlen, u64 requestid)
868 {
869 	int ret;
870 	struct vmbus_channel_packet_multipage_buffer desc;
871 	u32 descsize;
872 	u32 packetlen;
873 	u32 packetlen_aligned;
874 	struct kvec bufferlist[3];
875 	u64 aligned_data = 0;
876 	bool signal = false;
877 	bool lock = channel->acquire_ring_lock;
878 	u32 pfncount = NUM_PAGES_SPANNED(multi_pagebuffer->offset,
879 					 multi_pagebuffer->len);
880 
881 	if (pfncount > MAX_MULTIPAGE_BUFFER_COUNT)
882 		return -EINVAL;
883 
884 	/*
885 	 * Adjust the size down since vmbus_channel_packet_multipage_buffer is
886 	 * the largest size we support
887 	 */
888 	descsize = sizeof(struct vmbus_channel_packet_multipage_buffer) -
889 			  ((MAX_MULTIPAGE_BUFFER_COUNT - pfncount) *
890 			  sizeof(u64));
891 	packetlen = descsize + bufferlen;
892 	packetlen_aligned = ALIGN(packetlen, sizeof(u64));
893 
894 
895 	/* Setup the descriptor */
896 	desc.type = VM_PKT_DATA_USING_GPA_DIRECT;
897 	desc.flags = VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED;
898 	desc.dataoffset8 = descsize >> 3; /* in 8-bytes grandularity */
899 	desc.length8 = (u16)(packetlen_aligned >> 3);
900 	desc.transactionid = requestid;
901 	desc.rangecount = 1;
902 
903 	desc.range.len = multi_pagebuffer->len;
904 	desc.range.offset = multi_pagebuffer->offset;
905 
906 	memcpy(desc.range.pfn_array, multi_pagebuffer->pfn_array,
907 	       pfncount * sizeof(u64));
908 
909 	bufferlist[0].iov_base = &desc;
910 	bufferlist[0].iov_len = descsize;
911 	bufferlist[1].iov_base = buffer;
912 	bufferlist[1].iov_len = bufferlen;
913 	bufferlist[2].iov_base = &aligned_data;
914 	bufferlist[2].iov_len = (packetlen_aligned - packetlen);
915 
916 	ret = hv_ringbuffer_write(&channel->outbound, bufferlist, 3,
917 				  &signal, lock, channel->signal_policy);
918 
919 	if (ret == 0 && signal)
920 		vmbus_setevent(channel);
921 
922 	return ret;
923 }
924 EXPORT_SYMBOL_GPL(vmbus_sendpacket_multipagebuffer);
925 
926 /**
927  * vmbus_recvpacket() - Retrieve the user packet on the specified channel
928  * @channel: Pointer to vmbus_channel structure.
929  * @buffer: Pointer to the buffer you want to receive the data into.
930  * @bufferlen: Maximum size of what the the buffer will hold
931  * @buffer_actual_len: The actual size of the data after it was received
932  * @requestid: Identifier of the request
933  *
934  * Receives directly from the hyper-v vmbus and puts the data it received
935  * into Buffer. This will receive the data unparsed from hyper-v.
936  *
937  * Mainly used by Hyper-V drivers.
938  */
939 static inline int
940 __vmbus_recvpacket(struct vmbus_channel *channel, void *buffer,
941 		   u32 bufferlen, u32 *buffer_actual_len, u64 *requestid,
942 		   bool raw)
943 {
944 	int ret;
945 	bool signal = false;
946 
947 	ret = hv_ringbuffer_read(&channel->inbound, buffer, bufferlen,
948 				 buffer_actual_len, requestid, &signal, raw);
949 
950 	if (signal)
951 		vmbus_setevent(channel);
952 
953 	return ret;
954 }
955 
956 int vmbus_recvpacket(struct vmbus_channel *channel, void *buffer,
957 		     u32 bufferlen, u32 *buffer_actual_len,
958 		     u64 *requestid)
959 {
960 	return __vmbus_recvpacket(channel, buffer, bufferlen,
961 				  buffer_actual_len, requestid, false);
962 }
963 EXPORT_SYMBOL(vmbus_recvpacket);
964 
965 /*
966  * vmbus_recvpacket_raw - Retrieve the raw packet on the specified channel
967  */
968 int vmbus_recvpacket_raw(struct vmbus_channel *channel, void *buffer,
969 			      u32 bufferlen, u32 *buffer_actual_len,
970 			      u64 *requestid)
971 {
972 	return __vmbus_recvpacket(channel, buffer, bufferlen,
973 				  buffer_actual_len, requestid, true);
974 }
975 EXPORT_SYMBOL_GPL(vmbus_recvpacket_raw);
976