Loading...
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 */
41static 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 */
66int 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, t, err = 0;
75
76 spin_lock_irqsave(&newchannel->sc_lock, flags);
77 if (newchannel->state == CHANNEL_OPEN_STATE) {
78 newchannel->state = CHANNEL_OPENING_STATE;
79 } else {
80 spin_unlock_irqrestore(&newchannel->sc_lock, flags);
81 return -EINVAL;
82 }
83 spin_unlock_irqrestore(&newchannel->sc_lock, flags);
84
85 newchannel->onchannel_callback = onchannelcallback;
86 newchannel->channel_callback_context = context;
87
88 /* Allocate the ring buffer */
89 out = (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO,
90 get_order(send_ringbuffer_size + recv_ringbuffer_size));
91
92 if (!out)
93 return -ENOMEM;
94
95
96 in = (void *)((unsigned long)out + send_ringbuffer_size);
97
98 newchannel->ringbuffer_pages = out;
99 newchannel->ringbuffer_pagecount = (send_ringbuffer_size +
100 recv_ringbuffer_size) >> PAGE_SHIFT;
101
102 ret = hv_ringbuffer_init(
103 &newchannel->outbound, out, send_ringbuffer_size);
104
105 if (ret != 0) {
106 err = ret;
107 goto error0;
108 }
109
110 ret = hv_ringbuffer_init(
111 &newchannel->inbound, in, recv_ringbuffer_size);
112 if (ret != 0) {
113 err = ret;
114 goto error0;
115 }
116
117
118 /* Establish the gpadl for the ring buffer */
119 newchannel->ringbuffer_gpadlhandle = 0;
120
121 ret = vmbus_establish_gpadl(newchannel,
122 newchannel->outbound.ring_buffer,
123 send_ringbuffer_size +
124 recv_ringbuffer_size,
125 &newchannel->ringbuffer_gpadlhandle);
126
127 if (ret != 0) {
128 err = ret;
129 goto error0;
130 }
131
132 /* Create and init the channel open message */
133 open_info = kmalloc(sizeof(*open_info) +
134 sizeof(struct vmbus_channel_open_channel),
135 GFP_KERNEL);
136 if (!open_info) {
137 err = -ENOMEM;
138 goto error0;
139 }
140
141 init_completion(&open_info->waitevent);
142
143 open_msg = (struct vmbus_channel_open_channel *)open_info->msg;
144 open_msg->header.msgtype = CHANNELMSG_OPENCHANNEL;
145 open_msg->openid = newchannel->offermsg.child_relid;
146 open_msg->child_relid = newchannel->offermsg.child_relid;
147 open_msg->ringbuffer_gpadlhandle = newchannel->ringbuffer_gpadlhandle;
148 open_msg->downstream_ringbuffer_pageoffset = send_ringbuffer_size >>
149 PAGE_SHIFT;
150 open_msg->target_vp = newchannel->target_vp;
151
152 if (userdatalen > MAX_USER_DEFINED_BYTES) {
153 err = -EINVAL;
154 goto error0;
155 }
156
157 if (userdatalen)
158 memcpy(open_msg->userdata, userdata, userdatalen);
159
160 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
161 list_add_tail(&open_info->msglistentry,
162 &vmbus_connection.chn_msg_list);
163 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
164
165 ret = vmbus_post_msg(open_msg,
166 sizeof(struct vmbus_channel_open_channel));
167
168 if (ret != 0)
169 goto error1;
170
171 t = wait_for_completion_timeout(&open_info->waitevent, 5*HZ);
172 if (t == 0) {
173 err = -ETIMEDOUT;
174 goto error1;
175 }
176
177
178 if (open_info->response.open_result.status)
179 err = open_info->response.open_result.status;
180
181 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
182 list_del(&open_info->msglistentry);
183 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
184
185 if (err == 0)
186 newchannel->state = CHANNEL_OPENED_STATE;
187
188 kfree(open_info);
189 return err;
190
191error1:
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
196error0:
197 free_pages((unsigned long)out,
198 get_order(send_ringbuffer_size + recv_ringbuffer_size));
199 kfree(open_info);
200 return err;
201}
202EXPORT_SYMBOL_GPL(vmbus_open);
203
204/*
205 * create_gpadl_header - Creates a gpadl for the specified buffer
206 */
207static int create_gpadl_header(void *kbuffer, u32 size,
208 struct vmbus_channel_msginfo **msginfo,
209 u32 *messagecount)
210{
211 int i;
212 int pagecount;
213 struct vmbus_channel_gpadl_header *gpadl_header;
214 struct vmbus_channel_gpadl_body *gpadl_body;
215 struct vmbus_channel_msginfo *msgheader;
216 struct vmbus_channel_msginfo *msgbody = NULL;
217 u32 msgsize;
218
219 int pfnsum, pfncount, pfnleft, pfncurr, pfnsize;
220
221 pagecount = size >> PAGE_SHIFT;
222
223 /* do we need a gpadl body msg */
224 pfnsize = MAX_SIZE_CHANNEL_MESSAGE -
225 sizeof(struct vmbus_channel_gpadl_header) -
226 sizeof(struct gpa_range);
227 pfncount = pfnsize / sizeof(u64);
228
229 if (pagecount > pfncount) {
230 /* we need a gpadl body */
231 /* fill in the header */
232 msgsize = sizeof(struct vmbus_channel_msginfo) +
233 sizeof(struct vmbus_channel_gpadl_header) +
234 sizeof(struct gpa_range) + pfncount * sizeof(u64);
235 msgheader = kzalloc(msgsize, GFP_KERNEL);
236 if (!msgheader)
237 goto nomem;
238
239 INIT_LIST_HEAD(&msgheader->submsglist);
240 msgheader->msgsize = msgsize;
241
242 gpadl_header = (struct vmbus_channel_gpadl_header *)
243 msgheader->msg;
244 gpadl_header->rangecount = 1;
245 gpadl_header->range_buflen = sizeof(struct gpa_range) +
246 pagecount * sizeof(u64);
247 gpadl_header->range[0].byte_offset = 0;
248 gpadl_header->range[0].byte_count = size;
249 for (i = 0; i < pfncount; i++)
250 gpadl_header->range[0].pfn_array[i] = slow_virt_to_phys(
251 kbuffer + PAGE_SIZE * i) >> PAGE_SHIFT;
252 *msginfo = msgheader;
253 *messagecount = 1;
254
255 pfnsum = pfncount;
256 pfnleft = pagecount - pfncount;
257
258 /* how many pfns can we fit */
259 pfnsize = MAX_SIZE_CHANNEL_MESSAGE -
260 sizeof(struct vmbus_channel_gpadl_body);
261 pfncount = pfnsize / sizeof(u64);
262
263 /* fill in the body */
264 while (pfnleft) {
265 if (pfnleft > pfncount)
266 pfncurr = pfncount;
267 else
268 pfncurr = pfnleft;
269
270 msgsize = sizeof(struct vmbus_channel_msginfo) +
271 sizeof(struct vmbus_channel_gpadl_body) +
272 pfncurr * sizeof(u64);
273 msgbody = kzalloc(msgsize, GFP_KERNEL);
274
275 if (!msgbody) {
276 struct vmbus_channel_msginfo *pos = NULL;
277 struct vmbus_channel_msginfo *tmp = NULL;
278 /*
279 * Free up all the allocated messages.
280 */
281 list_for_each_entry_safe(pos, tmp,
282 &msgheader->submsglist,
283 msglistentry) {
284
285 list_del(&pos->msglistentry);
286 kfree(pos);
287 }
288
289 goto nomem;
290 }
291
292 msgbody->msgsize = msgsize;
293 (*messagecount)++;
294 gpadl_body =
295 (struct vmbus_channel_gpadl_body *)msgbody->msg;
296
297 /*
298 * Gpadl is u32 and we are using a pointer which could
299 * be 64-bit
300 * This is governed by the guest/host protocol and
301 * so the hypervisor gurantees that this is ok.
302 */
303 for (i = 0; i < pfncurr; i++)
304 gpadl_body->pfn[i] = slow_virt_to_phys(
305 kbuffer + PAGE_SIZE * (pfnsum + i)) >>
306 PAGE_SHIFT;
307
308 /* add to msg header */
309 list_add_tail(&msgbody->msglistentry,
310 &msgheader->submsglist);
311 pfnsum += pfncurr;
312 pfnleft -= pfncurr;
313 }
314 } else {
315 /* everything fits in a header */
316 msgsize = sizeof(struct vmbus_channel_msginfo) +
317 sizeof(struct vmbus_channel_gpadl_header) +
318 sizeof(struct gpa_range) + pagecount * sizeof(u64);
319 msgheader = kzalloc(msgsize, GFP_KERNEL);
320 if (msgheader == NULL)
321 goto nomem;
322 msgheader->msgsize = msgsize;
323
324 gpadl_header = (struct vmbus_channel_gpadl_header *)
325 msgheader->msg;
326 gpadl_header->rangecount = 1;
327 gpadl_header->range_buflen = sizeof(struct gpa_range) +
328 pagecount * sizeof(u64);
329 gpadl_header->range[0].byte_offset = 0;
330 gpadl_header->range[0].byte_count = size;
331 for (i = 0; i < pagecount; i++)
332 gpadl_header->range[0].pfn_array[i] = slow_virt_to_phys(
333 kbuffer + PAGE_SIZE * i) >> PAGE_SHIFT;
334
335 *msginfo = msgheader;
336 *messagecount = 1;
337 }
338
339 return 0;
340nomem:
341 kfree(msgheader);
342 kfree(msgbody);
343 return -ENOMEM;
344}
345
346/*
347 * vmbus_establish_gpadl - Estabish a GPADL for the specified buffer
348 *
349 * @channel: a channel
350 * @kbuffer: from kmalloc or vmalloc
351 * @size: page-size multiple
352 * @gpadl_handle: some funky thing
353 */
354int vmbus_establish_gpadl(struct vmbus_channel *channel, void *kbuffer,
355 u32 size, u32 *gpadl_handle)
356{
357 struct vmbus_channel_gpadl_header *gpadlmsg;
358 struct vmbus_channel_gpadl_body *gpadl_body;
359 struct vmbus_channel_msginfo *msginfo = NULL;
360 struct vmbus_channel_msginfo *submsginfo;
361 u32 msgcount;
362 struct list_head *curr;
363 u32 next_gpadl_handle;
364 unsigned long flags;
365 int ret = 0;
366 int t;
367
368 next_gpadl_handle = atomic_read(&vmbus_connection.next_gpadl_handle);
369 atomic_inc(&vmbus_connection.next_gpadl_handle);
370
371 ret = create_gpadl_header(kbuffer, size, &msginfo, &msgcount);
372 if (ret)
373 return ret;
374
375 init_completion(&msginfo->waitevent);
376
377 gpadlmsg = (struct vmbus_channel_gpadl_header *)msginfo->msg;
378 gpadlmsg->header.msgtype = CHANNELMSG_GPADL_HEADER;
379 gpadlmsg->child_relid = channel->offermsg.child_relid;
380 gpadlmsg->gpadl = next_gpadl_handle;
381
382
383 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
384 list_add_tail(&msginfo->msglistentry,
385 &vmbus_connection.chn_msg_list);
386
387 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
388
389 ret = vmbus_post_msg(gpadlmsg, msginfo->msgsize -
390 sizeof(*msginfo));
391 if (ret != 0)
392 goto cleanup;
393
394 if (msgcount > 1) {
395 list_for_each(curr, &msginfo->submsglist) {
396
397 submsginfo = (struct vmbus_channel_msginfo *)curr;
398 gpadl_body =
399 (struct vmbus_channel_gpadl_body *)submsginfo->msg;
400
401 gpadl_body->header.msgtype =
402 CHANNELMSG_GPADL_BODY;
403 gpadl_body->gpadl = next_gpadl_handle;
404
405 ret = vmbus_post_msg(gpadl_body,
406 submsginfo->msgsize -
407 sizeof(*submsginfo));
408 if (ret != 0)
409 goto cleanup;
410
411 }
412 }
413 t = wait_for_completion_timeout(&msginfo->waitevent, 5*HZ);
414 BUG_ON(t == 0);
415
416
417 /* At this point, we received the gpadl created msg */
418 *gpadl_handle = gpadlmsg->gpadl;
419
420cleanup:
421 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
422 list_del(&msginfo->msglistentry);
423 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
424
425 kfree(msginfo);
426 return ret;
427}
428EXPORT_SYMBOL_GPL(vmbus_establish_gpadl);
429
430/*
431 * vmbus_teardown_gpadl -Teardown the specified GPADL handle
432 */
433int vmbus_teardown_gpadl(struct vmbus_channel *channel, u32 gpadl_handle)
434{
435 struct vmbus_channel_gpadl_teardown *msg;
436 struct vmbus_channel_msginfo *info;
437 unsigned long flags;
438 int ret, t;
439
440 info = kmalloc(sizeof(*info) +
441 sizeof(struct vmbus_channel_gpadl_teardown), GFP_KERNEL);
442 if (!info)
443 return -ENOMEM;
444
445 init_completion(&info->waitevent);
446
447 msg = (struct vmbus_channel_gpadl_teardown *)info->msg;
448
449 msg->header.msgtype = CHANNELMSG_GPADL_TEARDOWN;
450 msg->child_relid = channel->offermsg.child_relid;
451 msg->gpadl = gpadl_handle;
452
453 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
454 list_add_tail(&info->msglistentry,
455 &vmbus_connection.chn_msg_list);
456 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
457 ret = vmbus_post_msg(msg,
458 sizeof(struct vmbus_channel_gpadl_teardown));
459
460 BUG_ON(ret != 0);
461 t = wait_for_completion_timeout(&info->waitevent, 5*HZ);
462 BUG_ON(t == 0);
463
464 /* Received a torndown response */
465 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
466 list_del(&info->msglistentry);
467 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
468
469 kfree(info);
470 return ret;
471}
472EXPORT_SYMBOL_GPL(vmbus_teardown_gpadl);
473
474static void vmbus_close_internal(struct vmbus_channel *channel)
475{
476 struct vmbus_channel_close_channel *msg;
477 int ret;
478 unsigned long flags;
479
480 channel->state = CHANNEL_OPEN_STATE;
481 channel->sc_creation_callback = NULL;
482 /* Stop callback and cancel the timer asap */
483 spin_lock_irqsave(&channel->inbound_lock, flags);
484 channel->onchannel_callback = NULL;
485 spin_unlock_irqrestore(&channel->inbound_lock, flags);
486
487 /* Send a closing message */
488
489 msg = &channel->close_msg.msg;
490
491 msg->header.msgtype = CHANNELMSG_CLOSECHANNEL;
492 msg->child_relid = channel->offermsg.child_relid;
493
494 ret = vmbus_post_msg(msg, sizeof(struct vmbus_channel_close_channel));
495
496 BUG_ON(ret != 0);
497 /* Tear down the gpadl for the channel's ring buffer */
498 if (channel->ringbuffer_gpadlhandle)
499 vmbus_teardown_gpadl(channel,
500 channel->ringbuffer_gpadlhandle);
501
502 /* Cleanup the ring buffers for this channel */
503 hv_ringbuffer_cleanup(&channel->outbound);
504 hv_ringbuffer_cleanup(&channel->inbound);
505
506 free_pages((unsigned long)channel->ringbuffer_pages,
507 get_order(channel->ringbuffer_pagecount * PAGE_SIZE));
508
509
510}
511
512/*
513 * vmbus_close - Close the specified channel
514 */
515void vmbus_close(struct vmbus_channel *channel)
516{
517 struct list_head *cur, *tmp;
518 struct vmbus_channel *cur_channel;
519
520 if (channel->primary_channel != NULL) {
521 /*
522 * We will only close sub-channels when
523 * the primary is closed.
524 */
525 return;
526 }
527 /*
528 * Close all the sub-channels first and then close the
529 * primary channel.
530 */
531 list_for_each_safe(cur, tmp, &channel->sc_list) {
532 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
533 if (cur_channel->state != CHANNEL_OPENED_STATE)
534 continue;
535 vmbus_close_internal(cur_channel);
536 }
537 /*
538 * Now close the primary.
539 */
540 vmbus_close_internal(channel);
541}
542EXPORT_SYMBOL_GPL(vmbus_close);
543
544/**
545 * vmbus_sendpacket() - Send the specified buffer on the given channel
546 * @channel: Pointer to vmbus_channel structure.
547 * @buffer: Pointer to the buffer you want to receive the data into.
548 * @bufferlen: Maximum size of what the the buffer will hold
549 * @requestid: Identifier of the request
550 * @type: Type of packet that is being send e.g. negotiate, time
551 * packet etc.
552 *
553 * Sends data in @buffer directly to hyper-v via the vmbus
554 * This will send the data unparsed to hyper-v.
555 *
556 * Mainly used by Hyper-V drivers.
557 */
558int vmbus_sendpacket(struct vmbus_channel *channel, void *buffer,
559 u32 bufferlen, u64 requestid,
560 enum vmbus_packet_type type, u32 flags)
561{
562 struct vmpacket_descriptor desc;
563 u32 packetlen = sizeof(struct vmpacket_descriptor) + bufferlen;
564 u32 packetlen_aligned = ALIGN(packetlen, sizeof(u64));
565 struct kvec bufferlist[3];
566 u64 aligned_data = 0;
567 int ret;
568 bool signal = false;
569
570
571 /* Setup the descriptor */
572 desc.type = type; /* VmbusPacketTypeDataInBand; */
573 desc.flags = flags; /* VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED; */
574 /* in 8-bytes granularity */
575 desc.offset8 = sizeof(struct vmpacket_descriptor) >> 3;
576 desc.len8 = (u16)(packetlen_aligned >> 3);
577 desc.trans_id = requestid;
578
579 bufferlist[0].iov_base = &desc;
580 bufferlist[0].iov_len = sizeof(struct vmpacket_descriptor);
581 bufferlist[1].iov_base = buffer;
582 bufferlist[1].iov_len = bufferlen;
583 bufferlist[2].iov_base = &aligned_data;
584 bufferlist[2].iov_len = (packetlen_aligned - packetlen);
585
586 ret = hv_ringbuffer_write(&channel->outbound, bufferlist, 3, &signal);
587
588 if (ret == 0 && signal)
589 vmbus_setevent(channel);
590
591 return ret;
592}
593EXPORT_SYMBOL(vmbus_sendpacket);
594
595/*
596 * vmbus_sendpacket_pagebuffer - Send a range of single-page buffer
597 * packets using a GPADL Direct packet type.
598 */
599int vmbus_sendpacket_pagebuffer(struct vmbus_channel *channel,
600 struct hv_page_buffer pagebuffers[],
601 u32 pagecount, void *buffer, u32 bufferlen,
602 u64 requestid)
603{
604 int ret;
605 int i;
606 struct vmbus_channel_packet_page_buffer desc;
607 u32 descsize;
608 u32 packetlen;
609 u32 packetlen_aligned;
610 struct kvec bufferlist[3];
611 u64 aligned_data = 0;
612 bool signal = false;
613
614 if (pagecount > MAX_PAGE_BUFFER_COUNT)
615 return -EINVAL;
616
617
618 /*
619 * Adjust the size down since vmbus_channel_packet_page_buffer is the
620 * largest size we support
621 */
622 descsize = sizeof(struct vmbus_channel_packet_page_buffer) -
623 ((MAX_PAGE_BUFFER_COUNT - pagecount) *
624 sizeof(struct hv_page_buffer));
625 packetlen = descsize + bufferlen;
626 packetlen_aligned = ALIGN(packetlen, sizeof(u64));
627
628 /* Setup the descriptor */
629 desc.type = VM_PKT_DATA_USING_GPA_DIRECT;
630 desc.flags = VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED;
631 desc.dataoffset8 = descsize >> 3; /* in 8-bytes grandularity */
632 desc.length8 = (u16)(packetlen_aligned >> 3);
633 desc.transactionid = requestid;
634 desc.rangecount = pagecount;
635
636 for (i = 0; i < pagecount; i++) {
637 desc.range[i].len = pagebuffers[i].len;
638 desc.range[i].offset = pagebuffers[i].offset;
639 desc.range[i].pfn = pagebuffers[i].pfn;
640 }
641
642 bufferlist[0].iov_base = &desc;
643 bufferlist[0].iov_len = descsize;
644 bufferlist[1].iov_base = buffer;
645 bufferlist[1].iov_len = bufferlen;
646 bufferlist[2].iov_base = &aligned_data;
647 bufferlist[2].iov_len = (packetlen_aligned - packetlen);
648
649 ret = hv_ringbuffer_write(&channel->outbound, bufferlist, 3, &signal);
650
651 if (ret == 0 && signal)
652 vmbus_setevent(channel);
653
654 return ret;
655}
656EXPORT_SYMBOL_GPL(vmbus_sendpacket_pagebuffer);
657
658/*
659 * vmbus_sendpacket_multipagebuffer - Send a multi-page buffer packet
660 * using a GPADL Direct packet type.
661 */
662int vmbus_sendpacket_multipagebuffer(struct vmbus_channel *channel,
663 struct hv_multipage_buffer *multi_pagebuffer,
664 void *buffer, u32 bufferlen, u64 requestid)
665{
666 int ret;
667 struct vmbus_channel_packet_multipage_buffer desc;
668 u32 descsize;
669 u32 packetlen;
670 u32 packetlen_aligned;
671 struct kvec bufferlist[3];
672 u64 aligned_data = 0;
673 bool signal = false;
674 u32 pfncount = NUM_PAGES_SPANNED(multi_pagebuffer->offset,
675 multi_pagebuffer->len);
676
677
678 if ((pfncount < 0) || (pfncount > MAX_MULTIPAGE_BUFFER_COUNT))
679 return -EINVAL;
680
681 /*
682 * Adjust the size down since vmbus_channel_packet_multipage_buffer is
683 * the largest size we support
684 */
685 descsize = sizeof(struct vmbus_channel_packet_multipage_buffer) -
686 ((MAX_MULTIPAGE_BUFFER_COUNT - pfncount) *
687 sizeof(u64));
688 packetlen = descsize + bufferlen;
689 packetlen_aligned = ALIGN(packetlen, sizeof(u64));
690
691
692 /* Setup the descriptor */
693 desc.type = VM_PKT_DATA_USING_GPA_DIRECT;
694 desc.flags = VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED;
695 desc.dataoffset8 = descsize >> 3; /* in 8-bytes grandularity */
696 desc.length8 = (u16)(packetlen_aligned >> 3);
697 desc.transactionid = requestid;
698 desc.rangecount = 1;
699
700 desc.range.len = multi_pagebuffer->len;
701 desc.range.offset = multi_pagebuffer->offset;
702
703 memcpy(desc.range.pfn_array, multi_pagebuffer->pfn_array,
704 pfncount * sizeof(u64));
705
706 bufferlist[0].iov_base = &desc;
707 bufferlist[0].iov_len = descsize;
708 bufferlist[1].iov_base = buffer;
709 bufferlist[1].iov_len = bufferlen;
710 bufferlist[2].iov_base = &aligned_data;
711 bufferlist[2].iov_len = (packetlen_aligned - packetlen);
712
713 ret = hv_ringbuffer_write(&channel->outbound, bufferlist, 3, &signal);
714
715 if (ret == 0 && signal)
716 vmbus_setevent(channel);
717
718 return ret;
719}
720EXPORT_SYMBOL_GPL(vmbus_sendpacket_multipagebuffer);
721
722/**
723 * vmbus_recvpacket() - Retrieve the user packet on the specified channel
724 * @channel: Pointer to vmbus_channel structure.
725 * @buffer: Pointer to the buffer you want to receive the data into.
726 * @bufferlen: Maximum size of what the the buffer will hold
727 * @buffer_actual_len: The actual size of the data after it was received
728 * @requestid: Identifier of the request
729 *
730 * Receives directly from the hyper-v vmbus and puts the data it received
731 * into Buffer. This will receive the data unparsed from hyper-v.
732 *
733 * Mainly used by Hyper-V drivers.
734 */
735int vmbus_recvpacket(struct vmbus_channel *channel, void *buffer,
736 u32 bufferlen, u32 *buffer_actual_len, u64 *requestid)
737{
738 struct vmpacket_descriptor desc;
739 u32 packetlen;
740 u32 userlen;
741 int ret;
742 bool signal = false;
743
744 *buffer_actual_len = 0;
745 *requestid = 0;
746
747
748 ret = hv_ringbuffer_peek(&channel->inbound, &desc,
749 sizeof(struct vmpacket_descriptor));
750 if (ret != 0)
751 return 0;
752
753 packetlen = desc.len8 << 3;
754 userlen = packetlen - (desc.offset8 << 3);
755
756 *buffer_actual_len = userlen;
757
758 if (userlen > bufferlen) {
759
760 pr_err("Buffer too small - got %d needs %d\n",
761 bufferlen, userlen);
762 return -ETOOSMALL;
763 }
764
765 *requestid = desc.trans_id;
766
767 /* Copy over the packet to the user buffer */
768 ret = hv_ringbuffer_read(&channel->inbound, buffer, userlen,
769 (desc.offset8 << 3), &signal);
770
771 if (signal)
772 vmbus_setevent(channel);
773
774 return 0;
775}
776EXPORT_SYMBOL(vmbus_recvpacket);
777
778/*
779 * vmbus_recvpacket_raw - Retrieve the raw packet on the specified channel
780 */
781int vmbus_recvpacket_raw(struct vmbus_channel *channel, void *buffer,
782 u32 bufferlen, u32 *buffer_actual_len,
783 u64 *requestid)
784{
785 struct vmpacket_descriptor desc;
786 u32 packetlen;
787 int ret;
788 bool signal = false;
789
790 *buffer_actual_len = 0;
791 *requestid = 0;
792
793
794 ret = hv_ringbuffer_peek(&channel->inbound, &desc,
795 sizeof(struct vmpacket_descriptor));
796 if (ret != 0)
797 return 0;
798
799
800 packetlen = desc.len8 << 3;
801
802 *buffer_actual_len = packetlen;
803
804 if (packetlen > bufferlen) {
805 pr_err("Buffer too small - needed %d bytes but "
806 "got space for only %d bytes\n",
807 packetlen, bufferlen);
808 return -ENOBUFS;
809 }
810
811 *requestid = desc.trans_id;
812
813 /* Copy over the entire packet to the user buffer */
814 ret = hv_ringbuffer_read(&channel->inbound, buffer, packetlen, 0,
815 &signal);
816
817 if (signal)
818 vmbus_setevent(channel);
819
820 return ret;
821}
822EXPORT_SYMBOL_GPL(vmbus_recvpacket_raw);
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (c) 2009, Microsoft Corporation.
4 *
5 * Authors:
6 * Haiyang Zhang <haiyangz@microsoft.com>
7 * Hank Janssen <hjanssen@microsoft.com>
8 */
9#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10
11#include <linux/kernel.h>
12#include <linux/sched.h>
13#include <linux/wait.h>
14#include <linux/mm.h>
15#include <linux/slab.h>
16#include <linux/module.h>
17#include <linux/hyperv.h>
18#include <linux/uio.h>
19#include <linux/interrupt.h>
20#include <linux/set_memory.h>
21#include <asm/page.h>
22#include <asm/mshyperv.h>
23
24#include "hyperv_vmbus.h"
25
26/*
27 * hv_gpadl_size - Return the real size of a gpadl, the size that Hyper-V uses
28 *
29 * For BUFFER gpadl, Hyper-V uses the exact same size as the guest does.
30 *
31 * For RING gpadl, in each ring, the guest uses one PAGE_SIZE as the header
32 * (because of the alignment requirement), however, the hypervisor only
33 * uses the first HV_HYP_PAGE_SIZE as the header, therefore leaving a
34 * (PAGE_SIZE - HV_HYP_PAGE_SIZE) gap. And since there are two rings in a
35 * ringbuffer, the total size for a RING gpadl that Hyper-V uses is the
36 * total size that the guest uses minus twice of the gap size.
37 */
38static inline u32 hv_gpadl_size(enum hv_gpadl_type type, u32 size)
39{
40 switch (type) {
41 case HV_GPADL_BUFFER:
42 return size;
43 case HV_GPADL_RING:
44 /* The size of a ringbuffer must be page-aligned */
45 BUG_ON(size % PAGE_SIZE);
46 /*
47 * Two things to notice here:
48 * 1) We're processing two ring buffers as a unit
49 * 2) We're skipping any space larger than HV_HYP_PAGE_SIZE in
50 * the first guest-size page of each of the two ring buffers.
51 * So we effectively subtract out two guest-size pages, and add
52 * back two Hyper-V size pages.
53 */
54 return size - 2 * (PAGE_SIZE - HV_HYP_PAGE_SIZE);
55 }
56 BUG();
57 return 0;
58}
59
60/*
61 * hv_ring_gpadl_send_hvpgoffset - Calculate the send offset (in unit of
62 * HV_HYP_PAGE) in a ring gpadl based on the
63 * offset in the guest
64 *
65 * @offset: the offset (in bytes) where the send ringbuffer starts in the
66 * virtual address space of the guest
67 */
68static inline u32 hv_ring_gpadl_send_hvpgoffset(u32 offset)
69{
70
71 /*
72 * For RING gpadl, in each ring, the guest uses one PAGE_SIZE as the
73 * header (because of the alignment requirement), however, the
74 * hypervisor only uses the first HV_HYP_PAGE_SIZE as the header,
75 * therefore leaving a (PAGE_SIZE - HV_HYP_PAGE_SIZE) gap.
76 *
77 * And to calculate the effective send offset in gpadl, we need to
78 * substract this gap.
79 */
80 return (offset - (PAGE_SIZE - HV_HYP_PAGE_SIZE)) >> HV_HYP_PAGE_SHIFT;
81}
82
83/*
84 * hv_gpadl_hvpfn - Return the Hyper-V page PFN of the @i th Hyper-V page in
85 * the gpadl
86 *
87 * @type: the type of the gpadl
88 * @kbuffer: the pointer to the gpadl in the guest
89 * @size: the total size (in bytes) of the gpadl
90 * @send_offset: the offset (in bytes) where the send ringbuffer starts in the
91 * virtual address space of the guest
92 * @i: the index
93 */
94static inline u64 hv_gpadl_hvpfn(enum hv_gpadl_type type, void *kbuffer,
95 u32 size, u32 send_offset, int i)
96{
97 int send_idx = hv_ring_gpadl_send_hvpgoffset(send_offset);
98 unsigned long delta = 0UL;
99
100 switch (type) {
101 case HV_GPADL_BUFFER:
102 break;
103 case HV_GPADL_RING:
104 if (i == 0)
105 delta = 0;
106 else if (i <= send_idx)
107 delta = PAGE_SIZE - HV_HYP_PAGE_SIZE;
108 else
109 delta = 2 * (PAGE_SIZE - HV_HYP_PAGE_SIZE);
110 break;
111 default:
112 BUG();
113 break;
114 }
115
116 return virt_to_hvpfn(kbuffer + delta + (HV_HYP_PAGE_SIZE * i));
117}
118
119/*
120 * vmbus_setevent- Trigger an event notification on the specified
121 * channel.
122 */
123void vmbus_setevent(struct vmbus_channel *channel)
124{
125 struct hv_monitor_page *monitorpage;
126
127 trace_vmbus_setevent(channel);
128
129 /*
130 * For channels marked as in "low latency" mode
131 * bypass the monitor page mechanism.
132 */
133 if (channel->offermsg.monitor_allocated && !channel->low_latency) {
134 vmbus_send_interrupt(channel->offermsg.child_relid);
135
136 /* Get the child to parent monitor page */
137 monitorpage = vmbus_connection.monitor_pages[1];
138
139 sync_set_bit(channel->monitor_bit,
140 (unsigned long *)&monitorpage->trigger_group
141 [channel->monitor_grp].pending);
142
143 } else {
144 vmbus_set_event(channel);
145 }
146}
147EXPORT_SYMBOL_GPL(vmbus_setevent);
148
149/* vmbus_free_ring - drop mapping of ring buffer */
150void vmbus_free_ring(struct vmbus_channel *channel)
151{
152 hv_ringbuffer_cleanup(&channel->outbound);
153 hv_ringbuffer_cleanup(&channel->inbound);
154
155 if (channel->ringbuffer_page) {
156 /* In a CoCo VM leak the memory if it didn't get re-encrypted */
157 if (!channel->ringbuffer_gpadlhandle.decrypted)
158 __free_pages(channel->ringbuffer_page,
159 get_order(channel->ringbuffer_pagecount
160 << PAGE_SHIFT));
161 channel->ringbuffer_page = NULL;
162 }
163}
164EXPORT_SYMBOL_GPL(vmbus_free_ring);
165
166/* vmbus_alloc_ring - allocate and map pages for ring buffer */
167int vmbus_alloc_ring(struct vmbus_channel *newchannel,
168 u32 send_size, u32 recv_size)
169{
170 struct page *page;
171 int order;
172
173 if (send_size % PAGE_SIZE || recv_size % PAGE_SIZE)
174 return -EINVAL;
175
176 /* Allocate the ring buffer */
177 order = get_order(send_size + recv_size);
178 page = alloc_pages_node(cpu_to_node(newchannel->target_cpu),
179 GFP_KERNEL|__GFP_ZERO, order);
180
181 if (!page)
182 page = alloc_pages(GFP_KERNEL|__GFP_ZERO, order);
183
184 if (!page)
185 return -ENOMEM;
186
187 newchannel->ringbuffer_page = page;
188 newchannel->ringbuffer_pagecount = (send_size + recv_size) >> PAGE_SHIFT;
189 newchannel->ringbuffer_send_offset = send_size >> PAGE_SHIFT;
190
191 return 0;
192}
193EXPORT_SYMBOL_GPL(vmbus_alloc_ring);
194
195/* Used for Hyper-V Socket: a guest client's connect() to the host */
196int vmbus_send_tl_connect_request(const guid_t *shv_guest_servie_id,
197 const guid_t *shv_host_servie_id)
198{
199 struct vmbus_channel_tl_connect_request conn_msg;
200 int ret;
201
202 memset(&conn_msg, 0, sizeof(conn_msg));
203 conn_msg.header.msgtype = CHANNELMSG_TL_CONNECT_REQUEST;
204 conn_msg.guest_endpoint_id = *shv_guest_servie_id;
205 conn_msg.host_service_id = *shv_host_servie_id;
206
207 ret = vmbus_post_msg(&conn_msg, sizeof(conn_msg), true);
208
209 trace_vmbus_send_tl_connect_request(&conn_msg, ret);
210
211 return ret;
212}
213EXPORT_SYMBOL_GPL(vmbus_send_tl_connect_request);
214
215static int send_modifychannel_without_ack(struct vmbus_channel *channel, u32 target_vp)
216{
217 struct vmbus_channel_modifychannel msg;
218 int ret;
219
220 memset(&msg, 0, sizeof(msg));
221 msg.header.msgtype = CHANNELMSG_MODIFYCHANNEL;
222 msg.child_relid = channel->offermsg.child_relid;
223 msg.target_vp = target_vp;
224
225 ret = vmbus_post_msg(&msg, sizeof(msg), true);
226 trace_vmbus_send_modifychannel(&msg, ret);
227
228 return ret;
229}
230
231static int send_modifychannel_with_ack(struct vmbus_channel *channel, u32 target_vp)
232{
233 struct vmbus_channel_modifychannel *msg;
234 struct vmbus_channel_msginfo *info;
235 unsigned long flags;
236 int ret;
237
238 info = kzalloc(sizeof(struct vmbus_channel_msginfo) +
239 sizeof(struct vmbus_channel_modifychannel),
240 GFP_KERNEL);
241 if (!info)
242 return -ENOMEM;
243
244 init_completion(&info->waitevent);
245 info->waiting_channel = channel;
246
247 msg = (struct vmbus_channel_modifychannel *)info->msg;
248 msg->header.msgtype = CHANNELMSG_MODIFYCHANNEL;
249 msg->child_relid = channel->offermsg.child_relid;
250 msg->target_vp = target_vp;
251
252 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
253 list_add_tail(&info->msglistentry, &vmbus_connection.chn_msg_list);
254 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
255
256 ret = vmbus_post_msg(msg, sizeof(*msg), true);
257 trace_vmbus_send_modifychannel(msg, ret);
258 if (ret != 0) {
259 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
260 list_del(&info->msglistentry);
261 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
262 goto free_info;
263 }
264
265 /*
266 * Release channel_mutex; otherwise, vmbus_onoffer_rescind() could block on
267 * the mutex and be unable to signal the completion.
268 *
269 * See the caller target_cpu_store() for information about the usage of the
270 * mutex.
271 */
272 mutex_unlock(&vmbus_connection.channel_mutex);
273 wait_for_completion(&info->waitevent);
274 mutex_lock(&vmbus_connection.channel_mutex);
275
276 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
277 list_del(&info->msglistentry);
278 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
279
280 if (info->response.modify_response.status)
281 ret = -EAGAIN;
282
283free_info:
284 kfree(info);
285 return ret;
286}
287
288/*
289 * Set/change the vCPU (@target_vp) the channel (@child_relid) will interrupt.
290 *
291 * CHANNELMSG_MODIFYCHANNEL messages are aynchronous. When VMbus version 5.3
292 * or later is negotiated, Hyper-V always sends an ACK in response to such a
293 * message. For VMbus version 5.2 and earlier, it never sends an ACK. With-
294 * out an ACK, we can not know when the host will stop interrupting the "old"
295 * vCPU and start interrupting the "new" vCPU for the given channel.
296 *
297 * The CHANNELMSG_MODIFYCHANNEL message type is supported since VMBus version
298 * VERSION_WIN10_V4_1.
299 */
300int vmbus_send_modifychannel(struct vmbus_channel *channel, u32 target_vp)
301{
302 if (vmbus_proto_version >= VERSION_WIN10_V5_3)
303 return send_modifychannel_with_ack(channel, target_vp);
304 return send_modifychannel_without_ack(channel, target_vp);
305}
306EXPORT_SYMBOL_GPL(vmbus_send_modifychannel);
307
308/*
309 * create_gpadl_header - Creates a gpadl for the specified buffer
310 */
311static int create_gpadl_header(enum hv_gpadl_type type, void *kbuffer,
312 u32 size, u32 send_offset,
313 struct vmbus_channel_msginfo **msginfo)
314{
315 int i;
316 int pagecount;
317 struct vmbus_channel_gpadl_header *gpadl_header;
318 struct vmbus_channel_gpadl_body *gpadl_body;
319 struct vmbus_channel_msginfo *msgheader;
320 struct vmbus_channel_msginfo *msgbody = NULL;
321 u32 msgsize;
322
323 int pfnsum, pfncount, pfnleft, pfncurr, pfnsize;
324
325 pagecount = hv_gpadl_size(type, size) >> HV_HYP_PAGE_SHIFT;
326
327 pfnsize = MAX_SIZE_CHANNEL_MESSAGE -
328 sizeof(struct vmbus_channel_gpadl_header) -
329 sizeof(struct gpa_range);
330 pfncount = umin(pagecount, pfnsize / sizeof(u64));
331
332 msgsize = sizeof(struct vmbus_channel_msginfo) +
333 sizeof(struct vmbus_channel_gpadl_header) +
334 sizeof(struct gpa_range) + pfncount * sizeof(u64);
335 msgheader = kzalloc(msgsize, GFP_KERNEL);
336 if (!msgheader)
337 return -ENOMEM;
338
339 INIT_LIST_HEAD(&msgheader->submsglist);
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 = hv_gpadl_size(type, size);
349 for (i = 0; i < pfncount; i++)
350 gpadl_header->range[0].pfn_array[i] = hv_gpadl_hvpfn(
351 type, kbuffer, size, send_offset, i);
352 *msginfo = msgheader;
353
354 pfnsum = pfncount;
355 pfnleft = pagecount - pfncount;
356
357 /* how many pfns can we fit in a body message */
358 pfnsize = MAX_SIZE_CHANNEL_MESSAGE -
359 sizeof(struct vmbus_channel_gpadl_body);
360 pfncount = pfnsize / sizeof(u64);
361
362 /*
363 * If pfnleft is zero, everything fits in the header and no body
364 * messages are needed
365 */
366 while (pfnleft) {
367 pfncurr = umin(pfncount, pfnleft);
368 msgsize = sizeof(struct vmbus_channel_msginfo) +
369 sizeof(struct vmbus_channel_gpadl_body) +
370 pfncurr * sizeof(u64);
371 msgbody = kzalloc(msgsize, GFP_KERNEL);
372
373 if (!msgbody) {
374 struct vmbus_channel_msginfo *pos = NULL;
375 struct vmbus_channel_msginfo *tmp = NULL;
376 /*
377 * Free up all the allocated messages.
378 */
379 list_for_each_entry_safe(pos, tmp,
380 &msgheader->submsglist,
381 msglistentry) {
382
383 list_del(&pos->msglistentry);
384 kfree(pos);
385 }
386 kfree(msgheader);
387 return -ENOMEM;
388 }
389
390 msgbody->msgsize = msgsize;
391 gpadl_body = (struct vmbus_channel_gpadl_body *)msgbody->msg;
392
393 /*
394 * Gpadl is u32 and we are using a pointer which could
395 * be 64-bit
396 * This is governed by the guest/host protocol and
397 * so the hypervisor guarantees that this is ok.
398 */
399 for (i = 0; i < pfncurr; i++)
400 gpadl_body->pfn[i] = hv_gpadl_hvpfn(type,
401 kbuffer, size, send_offset, pfnsum + i);
402
403 /* add to msg header */
404 list_add_tail(&msgbody->msglistentry, &msgheader->submsglist);
405 pfnsum += pfncurr;
406 pfnleft -= pfncurr;
407 }
408
409 return 0;
410}
411
412/*
413 * __vmbus_establish_gpadl - Establish a GPADL for a buffer or ringbuffer
414 *
415 * @channel: a channel
416 * @type: the type of the corresponding GPADL, only meaningful for the guest.
417 * @kbuffer: from kmalloc or vmalloc
418 * @size: page-size multiple
419 * @send_offset: the offset (in bytes) where the send ring buffer starts,
420 * should be 0 for BUFFER type gpadl
421 * @gpadl_handle: some funky thing
422 */
423static int __vmbus_establish_gpadl(struct vmbus_channel *channel,
424 enum hv_gpadl_type type, void *kbuffer,
425 u32 size, u32 send_offset,
426 struct vmbus_gpadl *gpadl)
427{
428 struct vmbus_channel_gpadl_header *gpadlmsg;
429 struct vmbus_channel_gpadl_body *gpadl_body;
430 struct vmbus_channel_msginfo *msginfo = NULL;
431 struct vmbus_channel_msginfo *submsginfo, *tmp;
432 struct list_head *curr;
433 u32 next_gpadl_handle;
434 unsigned long flags;
435 int ret = 0;
436
437 next_gpadl_handle =
438 (atomic_inc_return(&vmbus_connection.next_gpadl_handle) - 1);
439
440 ret = create_gpadl_header(type, kbuffer, size, send_offset, &msginfo);
441 if (ret) {
442 gpadl->decrypted = false;
443 return ret;
444 }
445
446 /*
447 * Set the "decrypted" flag to true for the set_memory_decrypted()
448 * success case. In the failure case, the encryption state of the
449 * memory is unknown. Leave "decrypted" as true to ensure the
450 * memory will be leaked instead of going back on the free list.
451 */
452 gpadl->decrypted = true;
453 ret = set_memory_decrypted((unsigned long)kbuffer,
454 PFN_UP(size));
455 if (ret) {
456 dev_warn(&channel->device_obj->device,
457 "Failed to set host visibility for new GPADL %d.\n",
458 ret);
459 return ret;
460 }
461
462 init_completion(&msginfo->waitevent);
463 msginfo->waiting_channel = channel;
464
465 gpadlmsg = (struct vmbus_channel_gpadl_header *)msginfo->msg;
466 gpadlmsg->header.msgtype = CHANNELMSG_GPADL_HEADER;
467 gpadlmsg->child_relid = channel->offermsg.child_relid;
468 gpadlmsg->gpadl = next_gpadl_handle;
469
470
471 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
472 list_add_tail(&msginfo->msglistentry,
473 &vmbus_connection.chn_msg_list);
474
475 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
476
477 if (channel->rescind) {
478 ret = -ENODEV;
479 goto cleanup;
480 }
481
482 ret = vmbus_post_msg(gpadlmsg, msginfo->msgsize -
483 sizeof(*msginfo), true);
484
485 trace_vmbus_establish_gpadl_header(gpadlmsg, ret);
486
487 if (ret != 0)
488 goto cleanup;
489
490 list_for_each(curr, &msginfo->submsglist) {
491 submsginfo = (struct vmbus_channel_msginfo *)curr;
492 gpadl_body =
493 (struct vmbus_channel_gpadl_body *)submsginfo->msg;
494
495 gpadl_body->header.msgtype =
496 CHANNELMSG_GPADL_BODY;
497 gpadl_body->gpadl = next_gpadl_handle;
498
499 ret = vmbus_post_msg(gpadl_body,
500 submsginfo->msgsize - sizeof(*submsginfo),
501 true);
502
503 trace_vmbus_establish_gpadl_body(gpadl_body, ret);
504
505 if (ret != 0)
506 goto cleanup;
507
508 }
509 wait_for_completion(&msginfo->waitevent);
510
511 if (msginfo->response.gpadl_created.creation_status != 0) {
512 pr_err("Failed to establish GPADL: err = 0x%x\n",
513 msginfo->response.gpadl_created.creation_status);
514
515 ret = -EDQUOT;
516 goto cleanup;
517 }
518
519 if (channel->rescind) {
520 ret = -ENODEV;
521 goto cleanup;
522 }
523
524 /* At this point, we received the gpadl created msg */
525 gpadl->gpadl_handle = gpadlmsg->gpadl;
526 gpadl->buffer = kbuffer;
527 gpadl->size = size;
528
529
530cleanup:
531 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
532 list_del(&msginfo->msglistentry);
533 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
534 list_for_each_entry_safe(submsginfo, tmp, &msginfo->submsglist,
535 msglistentry) {
536 kfree(submsginfo);
537 }
538
539 kfree(msginfo);
540
541 if (ret) {
542 /*
543 * If set_memory_encrypted() fails, the decrypted flag is
544 * left as true so the memory is leaked instead of being
545 * put back on the free list.
546 */
547 if (!set_memory_encrypted((unsigned long)kbuffer, PFN_UP(size)))
548 gpadl->decrypted = false;
549 }
550
551 return ret;
552}
553
554/*
555 * vmbus_establish_gpadl - Establish a GPADL for the specified buffer
556 *
557 * @channel: a channel
558 * @kbuffer: from kmalloc or vmalloc
559 * @size: page-size multiple
560 * @gpadl_handle: some funky thing
561 */
562int vmbus_establish_gpadl(struct vmbus_channel *channel, void *kbuffer,
563 u32 size, struct vmbus_gpadl *gpadl)
564{
565 return __vmbus_establish_gpadl(channel, HV_GPADL_BUFFER, kbuffer, size,
566 0U, gpadl);
567}
568EXPORT_SYMBOL_GPL(vmbus_establish_gpadl);
569
570/**
571 * request_arr_init - Allocates memory for the requestor array. Each slot
572 * keeps track of the next available slot in the array. Initially, each
573 * slot points to the next one (as in a Linked List). The last slot
574 * does not point to anything, so its value is U64_MAX by default.
575 * @size The size of the array
576 */
577static u64 *request_arr_init(u32 size)
578{
579 int i;
580 u64 *req_arr;
581
582 req_arr = kcalloc(size, sizeof(u64), GFP_KERNEL);
583 if (!req_arr)
584 return NULL;
585
586 for (i = 0; i < size - 1; i++)
587 req_arr[i] = i + 1;
588
589 /* Last slot (no more available slots) */
590 req_arr[i] = U64_MAX;
591
592 return req_arr;
593}
594
595/*
596 * vmbus_alloc_requestor - Initializes @rqstor's fields.
597 * Index 0 is the first free slot
598 * @size: Size of the requestor array
599 */
600static int vmbus_alloc_requestor(struct vmbus_requestor *rqstor, u32 size)
601{
602 u64 *rqst_arr;
603 unsigned long *bitmap;
604
605 rqst_arr = request_arr_init(size);
606 if (!rqst_arr)
607 return -ENOMEM;
608
609 bitmap = bitmap_zalloc(size, GFP_KERNEL);
610 if (!bitmap) {
611 kfree(rqst_arr);
612 return -ENOMEM;
613 }
614
615 rqstor->req_arr = rqst_arr;
616 rqstor->req_bitmap = bitmap;
617 rqstor->size = size;
618 rqstor->next_request_id = 0;
619 spin_lock_init(&rqstor->req_lock);
620
621 return 0;
622}
623
624/*
625 * vmbus_free_requestor - Frees memory allocated for @rqstor
626 * @rqstor: Pointer to the requestor struct
627 */
628static void vmbus_free_requestor(struct vmbus_requestor *rqstor)
629{
630 kfree(rqstor->req_arr);
631 bitmap_free(rqstor->req_bitmap);
632}
633
634static int __vmbus_open(struct vmbus_channel *newchannel,
635 void *userdata, u32 userdatalen,
636 void (*onchannelcallback)(void *context), void *context)
637{
638 struct vmbus_channel_open_channel *open_msg;
639 struct vmbus_channel_msginfo *open_info = NULL;
640 struct page *page = newchannel->ringbuffer_page;
641 u32 send_pages, recv_pages;
642 unsigned long flags;
643 int err;
644
645 if (userdatalen > MAX_USER_DEFINED_BYTES)
646 return -EINVAL;
647
648 send_pages = newchannel->ringbuffer_send_offset;
649 recv_pages = newchannel->ringbuffer_pagecount - send_pages;
650
651 if (newchannel->state != CHANNEL_OPEN_STATE)
652 return -EINVAL;
653
654 /* Create and init requestor */
655 if (newchannel->rqstor_size) {
656 if (vmbus_alloc_requestor(&newchannel->requestor, newchannel->rqstor_size))
657 return -ENOMEM;
658 }
659
660 newchannel->state = CHANNEL_OPENING_STATE;
661 newchannel->onchannel_callback = onchannelcallback;
662 newchannel->channel_callback_context = context;
663
664 if (!newchannel->max_pkt_size)
665 newchannel->max_pkt_size = VMBUS_DEFAULT_MAX_PKT_SIZE;
666
667 /* Establish the gpadl for the ring buffer */
668 newchannel->ringbuffer_gpadlhandle.gpadl_handle = 0;
669
670 err = __vmbus_establish_gpadl(newchannel, HV_GPADL_RING,
671 page_address(newchannel->ringbuffer_page),
672 (send_pages + recv_pages) << PAGE_SHIFT,
673 newchannel->ringbuffer_send_offset << PAGE_SHIFT,
674 &newchannel->ringbuffer_gpadlhandle);
675 if (err)
676 goto error_clean_ring;
677
678 err = hv_ringbuffer_init(&newchannel->outbound,
679 page, send_pages, 0);
680 if (err)
681 goto error_free_gpadl;
682
683 err = hv_ringbuffer_init(&newchannel->inbound, &page[send_pages],
684 recv_pages, newchannel->max_pkt_size);
685 if (err)
686 goto error_free_gpadl;
687
688 /* Create and init the channel open message */
689 open_info = kzalloc(sizeof(*open_info) +
690 sizeof(struct vmbus_channel_open_channel),
691 GFP_KERNEL);
692 if (!open_info) {
693 err = -ENOMEM;
694 goto error_free_gpadl;
695 }
696
697 init_completion(&open_info->waitevent);
698 open_info->waiting_channel = newchannel;
699
700 open_msg = (struct vmbus_channel_open_channel *)open_info->msg;
701 open_msg->header.msgtype = CHANNELMSG_OPENCHANNEL;
702 open_msg->openid = newchannel->offermsg.child_relid;
703 open_msg->child_relid = newchannel->offermsg.child_relid;
704 open_msg->ringbuffer_gpadlhandle
705 = newchannel->ringbuffer_gpadlhandle.gpadl_handle;
706 /*
707 * The unit of ->downstream_ringbuffer_pageoffset is HV_HYP_PAGE and
708 * the unit of ->ringbuffer_send_offset (i.e. send_pages) is PAGE, so
709 * here we calculate it into HV_HYP_PAGE.
710 */
711 open_msg->downstream_ringbuffer_pageoffset =
712 hv_ring_gpadl_send_hvpgoffset(send_pages << PAGE_SHIFT);
713 open_msg->target_vp = hv_cpu_number_to_vp_number(newchannel->target_cpu);
714
715 if (userdatalen)
716 memcpy(open_msg->userdata, userdata, userdatalen);
717
718 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
719 list_add_tail(&open_info->msglistentry,
720 &vmbus_connection.chn_msg_list);
721 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
722
723 if (newchannel->rescind) {
724 err = -ENODEV;
725 goto error_clean_msglist;
726 }
727
728 err = vmbus_post_msg(open_msg,
729 sizeof(struct vmbus_channel_open_channel), true);
730
731 trace_vmbus_open(open_msg, err);
732
733 if (err != 0)
734 goto error_clean_msglist;
735
736 wait_for_completion(&open_info->waitevent);
737
738 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
739 list_del(&open_info->msglistentry);
740 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
741
742 if (newchannel->rescind) {
743 err = -ENODEV;
744 goto error_free_info;
745 }
746
747 if (open_info->response.open_result.status) {
748 err = -EAGAIN;
749 goto error_free_info;
750 }
751
752 newchannel->state = CHANNEL_OPENED_STATE;
753 kfree(open_info);
754 return 0;
755
756error_clean_msglist:
757 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
758 list_del(&open_info->msglistentry);
759 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
760error_free_info:
761 kfree(open_info);
762error_free_gpadl:
763 vmbus_teardown_gpadl(newchannel, &newchannel->ringbuffer_gpadlhandle);
764error_clean_ring:
765 hv_ringbuffer_cleanup(&newchannel->outbound);
766 hv_ringbuffer_cleanup(&newchannel->inbound);
767 vmbus_free_requestor(&newchannel->requestor);
768 newchannel->state = CHANNEL_OPEN_STATE;
769 return err;
770}
771
772/*
773 * vmbus_connect_ring - Open the channel but reuse ring buffer
774 */
775int vmbus_connect_ring(struct vmbus_channel *newchannel,
776 void (*onchannelcallback)(void *context), void *context)
777{
778 return __vmbus_open(newchannel, NULL, 0, onchannelcallback, context);
779}
780EXPORT_SYMBOL_GPL(vmbus_connect_ring);
781
782/*
783 * vmbus_open - Open the specified channel.
784 */
785int vmbus_open(struct vmbus_channel *newchannel,
786 u32 send_ringbuffer_size, u32 recv_ringbuffer_size,
787 void *userdata, u32 userdatalen,
788 void (*onchannelcallback)(void *context), void *context)
789{
790 int err;
791
792 err = vmbus_alloc_ring(newchannel, send_ringbuffer_size,
793 recv_ringbuffer_size);
794 if (err)
795 return err;
796
797 err = __vmbus_open(newchannel, userdata, userdatalen,
798 onchannelcallback, context);
799 if (err)
800 vmbus_free_ring(newchannel);
801
802 return err;
803}
804EXPORT_SYMBOL_GPL(vmbus_open);
805
806/*
807 * vmbus_teardown_gpadl -Teardown the specified GPADL handle
808 */
809int vmbus_teardown_gpadl(struct vmbus_channel *channel, struct vmbus_gpadl *gpadl)
810{
811 struct vmbus_channel_gpadl_teardown *msg;
812 struct vmbus_channel_msginfo *info;
813 unsigned long flags;
814 int ret;
815
816 info = kzalloc(sizeof(*info) +
817 sizeof(struct vmbus_channel_gpadl_teardown), GFP_KERNEL);
818 if (!info)
819 return -ENOMEM;
820
821 init_completion(&info->waitevent);
822 info->waiting_channel = channel;
823
824 msg = (struct vmbus_channel_gpadl_teardown *)info->msg;
825
826 msg->header.msgtype = CHANNELMSG_GPADL_TEARDOWN;
827 msg->child_relid = channel->offermsg.child_relid;
828 msg->gpadl = gpadl->gpadl_handle;
829
830 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
831 list_add_tail(&info->msglistentry,
832 &vmbus_connection.chn_msg_list);
833 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
834
835 if (channel->rescind)
836 goto post_msg_err;
837
838 ret = vmbus_post_msg(msg, sizeof(struct vmbus_channel_gpadl_teardown),
839 true);
840
841 trace_vmbus_teardown_gpadl(msg, ret);
842
843 if (ret)
844 goto post_msg_err;
845
846 wait_for_completion(&info->waitevent);
847
848 gpadl->gpadl_handle = 0;
849
850post_msg_err:
851 /*
852 * If the channel has been rescinded;
853 * we will be awakened by the rescind
854 * handler; set the error code to zero so we don't leak memory.
855 */
856 if (channel->rescind)
857 ret = 0;
858
859 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
860 list_del(&info->msglistentry);
861 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
862
863 kfree(info);
864
865 ret = set_memory_encrypted((unsigned long)gpadl->buffer,
866 PFN_UP(gpadl->size));
867 if (ret)
868 pr_warn("Fail to set mem host visibility in GPADL teardown %d.\n", ret);
869
870 gpadl->decrypted = ret;
871
872 return ret;
873}
874EXPORT_SYMBOL_GPL(vmbus_teardown_gpadl);
875
876void vmbus_reset_channel_cb(struct vmbus_channel *channel)
877{
878 unsigned long flags;
879
880 /*
881 * vmbus_on_event(), running in the per-channel tasklet, can race
882 * with vmbus_close_internal() in the case of SMP guest, e.g., when
883 * the former is accessing channel->inbound.ring_buffer, the latter
884 * could be freeing the ring_buffer pages, so here we must stop it
885 * first.
886 *
887 * vmbus_chan_sched() might call the netvsc driver callback function
888 * that ends up scheduling NAPI work that accesses the ring buffer.
889 * At this point, we have to ensure that any such work is completed
890 * and that the channel ring buffer is no longer being accessed, cf.
891 * the calls to napi_disable() in netvsc_device_remove().
892 */
893 tasklet_disable(&channel->callback_event);
894
895 /* See the inline comments in vmbus_chan_sched(). */
896 spin_lock_irqsave(&channel->sched_lock, flags);
897 channel->onchannel_callback = NULL;
898 spin_unlock_irqrestore(&channel->sched_lock, flags);
899
900 channel->sc_creation_callback = NULL;
901
902 /* Re-enable tasklet for use on re-open */
903 tasklet_enable(&channel->callback_event);
904}
905
906static int vmbus_close_internal(struct vmbus_channel *channel)
907{
908 struct vmbus_channel_close_channel *msg;
909 int ret;
910
911 vmbus_reset_channel_cb(channel);
912
913 /*
914 * In case a device driver's probe() fails (e.g.,
915 * util_probe() -> vmbus_open() returns -ENOMEM) and the device is
916 * rescinded later (e.g., we dynamically disable an Integrated Service
917 * in Hyper-V Manager), the driver's remove() invokes vmbus_close():
918 * here we should skip most of the below cleanup work.
919 */
920 if (channel->state != CHANNEL_OPENED_STATE)
921 return -EINVAL;
922
923 channel->state = CHANNEL_OPEN_STATE;
924
925 /* Send a closing message */
926
927 msg = &channel->close_msg.msg;
928
929 msg->header.msgtype = CHANNELMSG_CLOSECHANNEL;
930 msg->child_relid = channel->offermsg.child_relid;
931
932 ret = vmbus_post_msg(msg, sizeof(struct vmbus_channel_close_channel),
933 true);
934
935 trace_vmbus_close_internal(msg, ret);
936
937 if (ret) {
938 pr_err("Close failed: close post msg return is %d\n", ret);
939 /*
940 * If we failed to post the close msg,
941 * it is perhaps better to leak memory.
942 */
943 }
944
945 /* Tear down the gpadl for the channel's ring buffer */
946 else if (channel->ringbuffer_gpadlhandle.gpadl_handle) {
947 ret = vmbus_teardown_gpadl(channel, &channel->ringbuffer_gpadlhandle);
948 if (ret) {
949 pr_err("Close failed: teardown gpadl return %d\n", ret);
950 /*
951 * If we failed to teardown gpadl,
952 * it is perhaps better to leak memory.
953 */
954 }
955 }
956
957 if (!ret)
958 vmbus_free_requestor(&channel->requestor);
959
960 return ret;
961}
962
963/* disconnect ring - close all channels */
964int vmbus_disconnect_ring(struct vmbus_channel *channel)
965{
966 struct vmbus_channel *cur_channel, *tmp;
967 int ret;
968
969 if (channel->primary_channel != NULL)
970 return -EINVAL;
971
972 list_for_each_entry_safe(cur_channel, tmp, &channel->sc_list, sc_list) {
973 if (cur_channel->rescind)
974 wait_for_completion(&cur_channel->rescind_event);
975
976 mutex_lock(&vmbus_connection.channel_mutex);
977 if (vmbus_close_internal(cur_channel) == 0) {
978 vmbus_free_ring(cur_channel);
979
980 if (cur_channel->rescind)
981 hv_process_channel_removal(cur_channel);
982 }
983 mutex_unlock(&vmbus_connection.channel_mutex);
984 }
985
986 /*
987 * Now close the primary.
988 */
989 mutex_lock(&vmbus_connection.channel_mutex);
990 ret = vmbus_close_internal(channel);
991 mutex_unlock(&vmbus_connection.channel_mutex);
992
993 return ret;
994}
995EXPORT_SYMBOL_GPL(vmbus_disconnect_ring);
996
997/*
998 * vmbus_close - Close the specified channel
999 */
1000void vmbus_close(struct vmbus_channel *channel)
1001{
1002 if (vmbus_disconnect_ring(channel) == 0)
1003 vmbus_free_ring(channel);
1004}
1005EXPORT_SYMBOL_GPL(vmbus_close);
1006
1007/**
1008 * vmbus_sendpacket_getid() - Send the specified buffer on the given channel
1009 * @channel: Pointer to vmbus_channel structure
1010 * @buffer: Pointer to the buffer you want to send the data from.
1011 * @bufferlen: Maximum size of what the buffer holds.
1012 * @requestid: Identifier of the request
1013 * @trans_id: Identifier of the transaction associated to this request, if
1014 * the send is successful; undefined, otherwise.
1015 * @type: Type of packet that is being sent e.g. negotiate, time
1016 * packet etc.
1017 * @flags: 0 or VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED
1018 *
1019 * Sends data in @buffer directly to Hyper-V via the vmbus.
1020 * This will send the data unparsed to Hyper-V.
1021 *
1022 * Mainly used by Hyper-V drivers.
1023 */
1024int vmbus_sendpacket_getid(struct vmbus_channel *channel, void *buffer,
1025 u32 bufferlen, u64 requestid, u64 *trans_id,
1026 enum vmbus_packet_type type, u32 flags)
1027{
1028 struct vmpacket_descriptor desc;
1029 u32 packetlen = sizeof(struct vmpacket_descriptor) + bufferlen;
1030 u32 packetlen_aligned = ALIGN(packetlen, sizeof(u64));
1031 struct kvec bufferlist[3];
1032 u64 aligned_data = 0;
1033 int num_vecs = ((bufferlen != 0) ? 3 : 1);
1034
1035
1036 /* Setup the descriptor */
1037 desc.type = type; /* VmbusPacketTypeDataInBand; */
1038 desc.flags = flags; /* VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED; */
1039 /* in 8-bytes granularity */
1040 desc.offset8 = sizeof(struct vmpacket_descriptor) >> 3;
1041 desc.len8 = (u16)(packetlen_aligned >> 3);
1042 desc.trans_id = VMBUS_RQST_ERROR; /* will be updated in hv_ringbuffer_write() */
1043
1044 bufferlist[0].iov_base = &desc;
1045 bufferlist[0].iov_len = sizeof(struct vmpacket_descriptor);
1046 bufferlist[1].iov_base = buffer;
1047 bufferlist[1].iov_len = bufferlen;
1048 bufferlist[2].iov_base = &aligned_data;
1049 bufferlist[2].iov_len = (packetlen_aligned - packetlen);
1050
1051 return hv_ringbuffer_write(channel, bufferlist, num_vecs, requestid, trans_id);
1052}
1053EXPORT_SYMBOL(vmbus_sendpacket_getid);
1054
1055/**
1056 * vmbus_sendpacket() - Send the specified buffer on the given channel
1057 * @channel: Pointer to vmbus_channel structure
1058 * @buffer: Pointer to the buffer you want to send the data from.
1059 * @bufferlen: Maximum size of what the buffer holds.
1060 * @requestid: Identifier of the request
1061 * @type: Type of packet that is being sent e.g. negotiate, time
1062 * packet etc.
1063 * @flags: 0 or VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED
1064 *
1065 * Sends data in @buffer directly to Hyper-V via the vmbus.
1066 * This will send the data unparsed to Hyper-V.
1067 *
1068 * Mainly used by Hyper-V drivers.
1069 */
1070int vmbus_sendpacket(struct vmbus_channel *channel, void *buffer,
1071 u32 bufferlen, u64 requestid,
1072 enum vmbus_packet_type type, u32 flags)
1073{
1074 return vmbus_sendpacket_getid(channel, buffer, bufferlen,
1075 requestid, NULL, type, flags);
1076}
1077EXPORT_SYMBOL(vmbus_sendpacket);
1078
1079/*
1080 * vmbus_sendpacket_pagebuffer - Send a range of single-page buffer
1081 * packets using a GPADL Direct packet type. This interface allows you
1082 * to control notifying the host. This will be useful for sending
1083 * batched data. Also the sender can control the send flags
1084 * explicitly.
1085 */
1086int vmbus_sendpacket_pagebuffer(struct vmbus_channel *channel,
1087 struct hv_page_buffer pagebuffers[],
1088 u32 pagecount, void *buffer, u32 bufferlen,
1089 u64 requestid)
1090{
1091 int i;
1092 struct vmbus_channel_packet_page_buffer desc;
1093 u32 descsize;
1094 u32 packetlen;
1095 u32 packetlen_aligned;
1096 struct kvec bufferlist[3];
1097 u64 aligned_data = 0;
1098
1099 if (pagecount > MAX_PAGE_BUFFER_COUNT)
1100 return -EINVAL;
1101
1102 /*
1103 * Adjust the size down since vmbus_channel_packet_page_buffer is the
1104 * largest size we support
1105 */
1106 descsize = sizeof(struct vmbus_channel_packet_page_buffer) -
1107 ((MAX_PAGE_BUFFER_COUNT - pagecount) *
1108 sizeof(struct hv_page_buffer));
1109 packetlen = descsize + bufferlen;
1110 packetlen_aligned = ALIGN(packetlen, sizeof(u64));
1111
1112 /* Setup the descriptor */
1113 desc.type = VM_PKT_DATA_USING_GPA_DIRECT;
1114 desc.flags = VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED;
1115 desc.dataoffset8 = descsize >> 3; /* in 8-bytes granularity */
1116 desc.length8 = (u16)(packetlen_aligned >> 3);
1117 desc.transactionid = VMBUS_RQST_ERROR; /* will be updated in hv_ringbuffer_write() */
1118 desc.reserved = 0;
1119 desc.rangecount = pagecount;
1120
1121 for (i = 0; i < pagecount; i++) {
1122 desc.range[i].len = pagebuffers[i].len;
1123 desc.range[i].offset = pagebuffers[i].offset;
1124 desc.range[i].pfn = pagebuffers[i].pfn;
1125 }
1126
1127 bufferlist[0].iov_base = &desc;
1128 bufferlist[0].iov_len = descsize;
1129 bufferlist[1].iov_base = buffer;
1130 bufferlist[1].iov_len = bufferlen;
1131 bufferlist[2].iov_base = &aligned_data;
1132 bufferlist[2].iov_len = (packetlen_aligned - packetlen);
1133
1134 return hv_ringbuffer_write(channel, bufferlist, 3, requestid, NULL);
1135}
1136EXPORT_SYMBOL_GPL(vmbus_sendpacket_pagebuffer);
1137
1138/*
1139 * vmbus_sendpacket_multipagebuffer - Send a multi-page buffer packet
1140 * using a GPADL Direct packet type.
1141 * The buffer includes the vmbus descriptor.
1142 */
1143int vmbus_sendpacket_mpb_desc(struct vmbus_channel *channel,
1144 struct vmbus_packet_mpb_array *desc,
1145 u32 desc_size,
1146 void *buffer, u32 bufferlen, u64 requestid)
1147{
1148 u32 packetlen;
1149 u32 packetlen_aligned;
1150 struct kvec bufferlist[3];
1151 u64 aligned_data = 0;
1152
1153 packetlen = desc_size + bufferlen;
1154 packetlen_aligned = ALIGN(packetlen, sizeof(u64));
1155
1156 /* Setup the descriptor */
1157 desc->type = VM_PKT_DATA_USING_GPA_DIRECT;
1158 desc->flags = VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED;
1159 desc->dataoffset8 = desc_size >> 3; /* in 8-bytes granularity */
1160 desc->length8 = (u16)(packetlen_aligned >> 3);
1161 desc->transactionid = VMBUS_RQST_ERROR; /* will be updated in hv_ringbuffer_write() */
1162 desc->reserved = 0;
1163 desc->rangecount = 1;
1164
1165 bufferlist[0].iov_base = desc;
1166 bufferlist[0].iov_len = desc_size;
1167 bufferlist[1].iov_base = buffer;
1168 bufferlist[1].iov_len = bufferlen;
1169 bufferlist[2].iov_base = &aligned_data;
1170 bufferlist[2].iov_len = (packetlen_aligned - packetlen);
1171
1172 return hv_ringbuffer_write(channel, bufferlist, 3, requestid, NULL);
1173}
1174EXPORT_SYMBOL_GPL(vmbus_sendpacket_mpb_desc);
1175
1176/**
1177 * __vmbus_recvpacket() - Retrieve the user packet on the specified channel
1178 * @channel: Pointer to vmbus_channel structure
1179 * @buffer: Pointer to the buffer you want to receive the data into.
1180 * @bufferlen: Maximum size of what the buffer can hold.
1181 * @buffer_actual_len: The actual size of the data after it was received.
1182 * @requestid: Identifier of the request
1183 * @raw: true means keep the vmpacket_descriptor header in the received data.
1184 *
1185 * Receives directly from the hyper-v vmbus and puts the data it received
1186 * into Buffer. This will receive the data unparsed from hyper-v.
1187 *
1188 * Mainly used by Hyper-V drivers.
1189 */
1190static inline int
1191__vmbus_recvpacket(struct vmbus_channel *channel, void *buffer,
1192 u32 bufferlen, u32 *buffer_actual_len, u64 *requestid,
1193 bool raw)
1194{
1195 return hv_ringbuffer_read(channel, buffer, bufferlen,
1196 buffer_actual_len, requestid, raw);
1197
1198}
1199
1200int vmbus_recvpacket(struct vmbus_channel *channel, void *buffer,
1201 u32 bufferlen, u32 *buffer_actual_len,
1202 u64 *requestid)
1203{
1204 return __vmbus_recvpacket(channel, buffer, bufferlen,
1205 buffer_actual_len, requestid, false);
1206}
1207EXPORT_SYMBOL(vmbus_recvpacket);
1208
1209/*
1210 * vmbus_recvpacket_raw - Retrieve the raw packet on the specified channel
1211 */
1212int vmbus_recvpacket_raw(struct vmbus_channel *channel, void *buffer,
1213 u32 bufferlen, u32 *buffer_actual_len,
1214 u64 *requestid)
1215{
1216 return __vmbus_recvpacket(channel, buffer, bufferlen,
1217 buffer_actual_len, requestid, true);
1218}
1219EXPORT_SYMBOL_GPL(vmbus_recvpacket_raw);
1220
1221/*
1222 * vmbus_next_request_id - Returns a new request id. It is also
1223 * the index at which the guest memory address is stored.
1224 * Uses a spin lock to avoid race conditions.
1225 * @channel: Pointer to the VMbus channel struct
1226 * @rqst_add: Guest memory address to be stored in the array
1227 */
1228u64 vmbus_next_request_id(struct vmbus_channel *channel, u64 rqst_addr)
1229{
1230 struct vmbus_requestor *rqstor = &channel->requestor;
1231 unsigned long flags;
1232 u64 current_id;
1233
1234 /* Check rqstor has been initialized */
1235 if (!channel->rqstor_size)
1236 return VMBUS_NO_RQSTOR;
1237
1238 lock_requestor(channel, flags);
1239 current_id = rqstor->next_request_id;
1240
1241 /* Requestor array is full */
1242 if (current_id >= rqstor->size) {
1243 unlock_requestor(channel, flags);
1244 return VMBUS_RQST_ERROR;
1245 }
1246
1247 rqstor->next_request_id = rqstor->req_arr[current_id];
1248 rqstor->req_arr[current_id] = rqst_addr;
1249
1250 /* The already held spin lock provides atomicity */
1251 bitmap_set(rqstor->req_bitmap, current_id, 1);
1252
1253 unlock_requestor(channel, flags);
1254
1255 /*
1256 * Cannot return an ID of 0, which is reserved for an unsolicited
1257 * message from Hyper-V; Hyper-V does not acknowledge (respond to)
1258 * VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED requests with ID of
1259 * 0 sent by the guest.
1260 */
1261 return current_id + 1;
1262}
1263EXPORT_SYMBOL_GPL(vmbus_next_request_id);
1264
1265/* As in vmbus_request_addr_match() but without the requestor lock */
1266u64 __vmbus_request_addr_match(struct vmbus_channel *channel, u64 trans_id,
1267 u64 rqst_addr)
1268{
1269 struct vmbus_requestor *rqstor = &channel->requestor;
1270 u64 req_addr;
1271
1272 /* Check rqstor has been initialized */
1273 if (!channel->rqstor_size)
1274 return VMBUS_NO_RQSTOR;
1275
1276 /* Hyper-V can send an unsolicited message with ID of 0 */
1277 if (!trans_id)
1278 return VMBUS_RQST_ERROR;
1279
1280 /* Data corresponding to trans_id is stored at trans_id - 1 */
1281 trans_id--;
1282
1283 /* Invalid trans_id */
1284 if (trans_id >= rqstor->size || !test_bit(trans_id, rqstor->req_bitmap))
1285 return VMBUS_RQST_ERROR;
1286
1287 req_addr = rqstor->req_arr[trans_id];
1288 if (rqst_addr == VMBUS_RQST_ADDR_ANY || req_addr == rqst_addr) {
1289 rqstor->req_arr[trans_id] = rqstor->next_request_id;
1290 rqstor->next_request_id = trans_id;
1291
1292 /* The already held spin lock provides atomicity */
1293 bitmap_clear(rqstor->req_bitmap, trans_id, 1);
1294 }
1295
1296 return req_addr;
1297}
1298EXPORT_SYMBOL_GPL(__vmbus_request_addr_match);
1299
1300/*
1301 * vmbus_request_addr_match - Clears/removes @trans_id from the @channel's
1302 * requestor, provided the memory address stored at @trans_id equals @rqst_addr
1303 * (or provided @rqst_addr matches the sentinel value VMBUS_RQST_ADDR_ANY).
1304 *
1305 * Returns the memory address stored at @trans_id, or VMBUS_RQST_ERROR if
1306 * @trans_id is not contained in the requestor.
1307 *
1308 * Acquires and releases the requestor spin lock.
1309 */
1310u64 vmbus_request_addr_match(struct vmbus_channel *channel, u64 trans_id,
1311 u64 rqst_addr)
1312{
1313 unsigned long flags;
1314 u64 req_addr;
1315
1316 lock_requestor(channel, flags);
1317 req_addr = __vmbus_request_addr_match(channel, trans_id, rqst_addr);
1318 unlock_requestor(channel, flags);
1319
1320 return req_addr;
1321}
1322EXPORT_SYMBOL_GPL(vmbus_request_addr_match);
1323
1324/*
1325 * vmbus_request_addr - Returns the memory address stored at @trans_id
1326 * in @rqstor. Uses a spin lock to avoid race conditions.
1327 * @channel: Pointer to the VMbus channel struct
1328 * @trans_id: Request id sent back from Hyper-V. Becomes the requestor's
1329 * next request id.
1330 */
1331u64 vmbus_request_addr(struct vmbus_channel *channel, u64 trans_id)
1332{
1333 return vmbus_request_addr_match(channel, trans_id, VMBUS_RQST_ADDR_ANY);
1334}
1335EXPORT_SYMBOL_GPL(vmbus_request_addr);