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 */
vmbus_setevent(struct vmbus_channel * channel)42 static void vmbus_setevent(struct vmbus_channel *channel)
43 {
44 struct hv_monitor_page *monitorpage;
45
46 if (channel->offermsg.monitor_allocated) {
47 /* Each u32 represents 32 channels */
48 sync_set_bit(channel->offermsg.child_relid & 31,
49 (unsigned long *) vmbus_connection.send_int_page +
50 (channel->offermsg.child_relid >> 5));
51
52 /* Get the child to parent monitor page */
53 monitorpage = vmbus_connection.monitor_pages[1];
54
55 sync_set_bit(channel->monitor_bit,
56 (unsigned long *)&monitorpage->trigger_group
57 [channel->monitor_grp].pending);
58
59 } else {
60 vmbus_set_event(channel);
61 }
62 }
63
64 /*
65 * vmbus_open - Open the specified channel.
66 */
vmbus_open(struct vmbus_channel * newchannel,u32 send_ringbuffer_size,u32 recv_ringbuffer_size,void * userdata,u32 userdatalen,void (* onchannelcallback)(void * context),void * context)67 int vmbus_open(struct vmbus_channel *newchannel, u32 send_ringbuffer_size,
68 u32 recv_ringbuffer_size, void *userdata, u32 userdatalen,
69 void (*onchannelcallback)(void *context), void *context)
70 {
71 struct vmbus_channel_open_channel *open_msg;
72 struct vmbus_channel_msginfo *open_info = NULL;
73 void *in, *out;
74 unsigned long flags;
75 int ret, err = 0;
76 struct page *page;
77
78 spin_lock_irqsave(&newchannel->lock, flags);
79 if (newchannel->state == CHANNEL_OPEN_STATE) {
80 newchannel->state = CHANNEL_OPENING_STATE;
81 } else {
82 spin_unlock_irqrestore(&newchannel->lock, flags);
83 return -EINVAL;
84 }
85 spin_unlock_irqrestore(&newchannel->lock, flags);
86
87 newchannel->onchannel_callback = onchannelcallback;
88 newchannel->channel_callback_context = context;
89
90 /* Allocate the ring buffer */
91 page = alloc_pages_node(cpu_to_node(newchannel->target_cpu),
92 GFP_KERNEL|__GFP_ZERO,
93 get_order(send_ringbuffer_size +
94 recv_ringbuffer_size));
95
96 if (!page)
97 out = (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO,
98 get_order(send_ringbuffer_size +
99 recv_ringbuffer_size));
100 else
101 out = (void *)page_address(page);
102
103 if (!out) {
104 err = -ENOMEM;
105 goto error0;
106 }
107
108 in = (void *)((unsigned long)out + send_ringbuffer_size);
109
110 newchannel->ringbuffer_pages = out;
111 newchannel->ringbuffer_pagecount = (send_ringbuffer_size +
112 recv_ringbuffer_size) >> PAGE_SHIFT;
113
114 ret = hv_ringbuffer_init(
115 &newchannel->outbound, out, send_ringbuffer_size);
116
117 if (ret != 0) {
118 err = ret;
119 goto error0;
120 }
121
122 ret = hv_ringbuffer_init(
123 &newchannel->inbound, in, recv_ringbuffer_size);
124 if (ret != 0) {
125 err = ret;
126 goto error0;
127 }
128
129
130 /* Establish the gpadl for the ring buffer */
131 newchannel->ringbuffer_gpadlhandle = 0;
132
133 ret = vmbus_establish_gpadl(newchannel,
134 newchannel->outbound.ring_buffer,
135 send_ringbuffer_size +
136 recv_ringbuffer_size,
137 &newchannel->ringbuffer_gpadlhandle);
138
139 if (ret != 0) {
140 err = ret;
141 goto error0;
142 }
143
144 /* Create and init the channel open message */
145 open_info = kmalloc(sizeof(*open_info) +
146 sizeof(struct vmbus_channel_open_channel),
147 GFP_KERNEL);
148 if (!open_info) {
149 err = -ENOMEM;
150 goto error_gpadl;
151 }
152
153 init_completion(&open_info->waitevent);
154
155 open_msg = (struct vmbus_channel_open_channel *)open_info->msg;
156 open_msg->header.msgtype = CHANNELMSG_OPENCHANNEL;
157 open_msg->openid = newchannel->offermsg.child_relid;
158 open_msg->child_relid = newchannel->offermsg.child_relid;
159 open_msg->ringbuffer_gpadlhandle = newchannel->ringbuffer_gpadlhandle;
160 open_msg->downstream_ringbuffer_pageoffset = send_ringbuffer_size >>
161 PAGE_SHIFT;
162 open_msg->target_vp = newchannel->target_vp;
163
164 if (userdatalen > MAX_USER_DEFINED_BYTES) {
165 err = -EINVAL;
166 goto error_gpadl;
167 }
168
169 if (userdatalen)
170 memcpy(open_msg->userdata, userdata, userdatalen);
171
172 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
173 list_add_tail(&open_info->msglistentry,
174 &vmbus_connection.chn_msg_list);
175 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
176
177 ret = vmbus_post_msg(open_msg,
178 sizeof(struct vmbus_channel_open_channel));
179
180 if (ret != 0) {
181 err = ret;
182 goto error1;
183 }
184
185 wait_for_completion(&open_info->waitevent);
186
187 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
188 list_del(&open_info->msglistentry);
189 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
190
191 if (open_info->response.open_result.status) {
192 err = -EAGAIN;
193 goto error_gpadl;
194 }
195
196 newchannel->state = CHANNEL_OPENED_STATE;
197 kfree(open_info);
198 return 0;
199
200 error1:
201 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
202 list_del(&open_info->msglistentry);
203 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
204
205 error_gpadl:
206 vmbus_teardown_gpadl(newchannel, newchannel->ringbuffer_gpadlhandle);
207
208 error0:
209 free_pages((unsigned long)out,
210 get_order(send_ringbuffer_size + recv_ringbuffer_size));
211 kfree(open_info);
212 newchannel->state = CHANNEL_OPEN_STATE;
213 return err;
214 }
215 EXPORT_SYMBOL_GPL(vmbus_open);
216
217 /*
218 * create_gpadl_header - Creates a gpadl for the specified buffer
219 */
create_gpadl_header(void * kbuffer,u32 size,struct vmbus_channel_msginfo ** msginfo,u32 * messagecount)220 static int create_gpadl_header(void *kbuffer, u32 size,
221 struct vmbus_channel_msginfo **msginfo,
222 u32 *messagecount)
223 {
224 int i;
225 int pagecount;
226 struct vmbus_channel_gpadl_header *gpadl_header;
227 struct vmbus_channel_gpadl_body *gpadl_body;
228 struct vmbus_channel_msginfo *msgheader;
229 struct vmbus_channel_msginfo *msgbody = NULL;
230 u32 msgsize;
231
232 int pfnsum, pfncount, pfnleft, pfncurr, pfnsize;
233
234 pagecount = size >> PAGE_SHIFT;
235
236 /* do we need a gpadl body msg */
237 pfnsize = MAX_SIZE_CHANNEL_MESSAGE -
238 sizeof(struct vmbus_channel_gpadl_header) -
239 sizeof(struct gpa_range);
240 pfncount = pfnsize / sizeof(u64);
241
242 if (pagecount > pfncount) {
243 /* we need a gpadl body */
244 /* fill in the header */
245 msgsize = sizeof(struct vmbus_channel_msginfo) +
246 sizeof(struct vmbus_channel_gpadl_header) +
247 sizeof(struct gpa_range) + pfncount * sizeof(u64);
248 msgheader = kzalloc(msgsize, GFP_KERNEL);
249 if (!msgheader)
250 goto nomem;
251
252 INIT_LIST_HEAD(&msgheader->submsglist);
253 msgheader->msgsize = msgsize;
254
255 gpadl_header = (struct vmbus_channel_gpadl_header *)
256 msgheader->msg;
257 gpadl_header->rangecount = 1;
258 gpadl_header->range_buflen = sizeof(struct gpa_range) +
259 pagecount * sizeof(u64);
260 gpadl_header->range[0].byte_offset = 0;
261 gpadl_header->range[0].byte_count = size;
262 for (i = 0; i < pfncount; i++)
263 gpadl_header->range[0].pfn_array[i] = slow_virt_to_phys(
264 kbuffer + PAGE_SIZE * i) >> PAGE_SHIFT;
265 *msginfo = msgheader;
266 *messagecount = 1;
267
268 pfnsum = pfncount;
269 pfnleft = pagecount - pfncount;
270
271 /* how many pfns can we fit */
272 pfnsize = MAX_SIZE_CHANNEL_MESSAGE -
273 sizeof(struct vmbus_channel_gpadl_body);
274 pfncount = pfnsize / sizeof(u64);
275
276 /* fill in the body */
277 while (pfnleft) {
278 if (pfnleft > pfncount)
279 pfncurr = pfncount;
280 else
281 pfncurr = pfnleft;
282
283 msgsize = sizeof(struct vmbus_channel_msginfo) +
284 sizeof(struct vmbus_channel_gpadl_body) +
285 pfncurr * sizeof(u64);
286 msgbody = kzalloc(msgsize, GFP_KERNEL);
287
288 if (!msgbody) {
289 struct vmbus_channel_msginfo *pos = NULL;
290 struct vmbus_channel_msginfo *tmp = NULL;
291 /*
292 * Free up all the allocated messages.
293 */
294 list_for_each_entry_safe(pos, tmp,
295 &msgheader->submsglist,
296 msglistentry) {
297
298 list_del(&pos->msglistentry);
299 kfree(pos);
300 }
301
302 goto nomem;
303 }
304
305 msgbody->msgsize = msgsize;
306 (*messagecount)++;
307 gpadl_body =
308 (struct vmbus_channel_gpadl_body *)msgbody->msg;
309
310 /*
311 * Gpadl is u32 and we are using a pointer which could
312 * be 64-bit
313 * This is governed by the guest/host protocol and
314 * so the hypervisor gurantees that this is ok.
315 */
316 for (i = 0; i < pfncurr; i++)
317 gpadl_body->pfn[i] = slow_virt_to_phys(
318 kbuffer + PAGE_SIZE * (pfnsum + i)) >>
319 PAGE_SHIFT;
320
321 /* add to msg header */
322 list_add_tail(&msgbody->msglistentry,
323 &msgheader->submsglist);
324 pfnsum += pfncurr;
325 pfnleft -= pfncurr;
326 }
327 } else {
328 /* everything fits in a header */
329 msgsize = sizeof(struct vmbus_channel_msginfo) +
330 sizeof(struct vmbus_channel_gpadl_header) +
331 sizeof(struct gpa_range) + pagecount * sizeof(u64);
332 msgheader = kzalloc(msgsize, GFP_KERNEL);
333 if (msgheader == NULL)
334 goto nomem;
335 msgheader->msgsize = msgsize;
336
337 gpadl_header = (struct vmbus_channel_gpadl_header *)
338 msgheader->msg;
339 gpadl_header->rangecount = 1;
340 gpadl_header->range_buflen = sizeof(struct gpa_range) +
341 pagecount * sizeof(u64);
342 gpadl_header->range[0].byte_offset = 0;
343 gpadl_header->range[0].byte_count = size;
344 for (i = 0; i < pagecount; i++)
345 gpadl_header->range[0].pfn_array[i] = slow_virt_to_phys(
346 kbuffer + PAGE_SIZE * i) >> PAGE_SHIFT;
347
348 *msginfo = msgheader;
349 *messagecount = 1;
350 }
351
352 return 0;
353 nomem:
354 kfree(msgheader);
355 kfree(msgbody);
356 return -ENOMEM;
357 }
358
359 /*
360 * vmbus_establish_gpadl - Estabish a GPADL for the specified buffer
361 *
362 * @channel: a channel
363 * @kbuffer: from kmalloc or vmalloc
364 * @size: page-size multiple
365 * @gpadl_handle: some funky thing
366 */
vmbus_establish_gpadl(struct vmbus_channel * channel,void * kbuffer,u32 size,u32 * gpadl_handle)367 int vmbus_establish_gpadl(struct vmbus_channel *channel, void *kbuffer,
368 u32 size, u32 *gpadl_handle)
369 {
370 struct vmbus_channel_gpadl_header *gpadlmsg;
371 struct vmbus_channel_gpadl_body *gpadl_body;
372 struct vmbus_channel_msginfo *msginfo = NULL;
373 struct vmbus_channel_msginfo *submsginfo, *tmp;
374 u32 msgcount;
375 struct list_head *curr;
376 u32 next_gpadl_handle;
377 unsigned long flags;
378 int ret = 0;
379
380 next_gpadl_handle =
381 (atomic_inc_return(&vmbus_connection.next_gpadl_handle) - 1);
382
383 ret = create_gpadl_header(kbuffer, size, &msginfo, &msgcount);
384 if (ret)
385 return ret;
386
387 init_completion(&msginfo->waitevent);
388
389 gpadlmsg = (struct vmbus_channel_gpadl_header *)msginfo->msg;
390 gpadlmsg->header.msgtype = CHANNELMSG_GPADL_HEADER;
391 gpadlmsg->child_relid = channel->offermsg.child_relid;
392 gpadlmsg->gpadl = next_gpadl_handle;
393
394
395 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
396 list_add_tail(&msginfo->msglistentry,
397 &vmbus_connection.chn_msg_list);
398
399 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
400
401 ret = vmbus_post_msg(gpadlmsg, msginfo->msgsize -
402 sizeof(*msginfo));
403 if (ret != 0)
404 goto cleanup;
405
406 if (msgcount > 1) {
407 list_for_each(curr, &msginfo->submsglist) {
408
409 submsginfo = (struct vmbus_channel_msginfo *)curr;
410 gpadl_body =
411 (struct vmbus_channel_gpadl_body *)submsginfo->msg;
412
413 gpadl_body->header.msgtype =
414 CHANNELMSG_GPADL_BODY;
415 gpadl_body->gpadl = next_gpadl_handle;
416
417 ret = vmbus_post_msg(gpadl_body,
418 submsginfo->msgsize -
419 sizeof(*submsginfo));
420 if (ret != 0)
421 goto cleanup;
422
423 }
424 }
425 wait_for_completion(&msginfo->waitevent);
426
427 /* At this point, we received the gpadl created msg */
428 *gpadl_handle = gpadlmsg->gpadl;
429
430 cleanup:
431 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
432 list_del(&msginfo->msglistentry);
433 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
434
435 if (msgcount > 1) {
436 list_for_each_entry_safe(submsginfo, tmp, &msginfo->submsglist,
437 msglistentry) {
438 kfree(submsginfo);
439 }
440 }
441
442 kfree(msginfo);
443 return ret;
444 }
445 EXPORT_SYMBOL_GPL(vmbus_establish_gpadl);
446
447 /*
448 * vmbus_teardown_gpadl -Teardown the specified GPADL handle
449 */
vmbus_teardown_gpadl(struct vmbus_channel * channel,u32 gpadl_handle)450 int vmbus_teardown_gpadl(struct vmbus_channel *channel, u32 gpadl_handle)
451 {
452 struct vmbus_channel_gpadl_teardown *msg;
453 struct vmbus_channel_msginfo *info;
454 unsigned long flags;
455 int ret;
456
457 info = kmalloc(sizeof(*info) +
458 sizeof(struct vmbus_channel_gpadl_teardown), GFP_KERNEL);
459 if (!info)
460 return -ENOMEM;
461
462 init_completion(&info->waitevent);
463
464 msg = (struct vmbus_channel_gpadl_teardown *)info->msg;
465
466 msg->header.msgtype = CHANNELMSG_GPADL_TEARDOWN;
467 msg->child_relid = channel->offermsg.child_relid;
468 msg->gpadl = gpadl_handle;
469
470 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
471 list_add_tail(&info->msglistentry,
472 &vmbus_connection.chn_msg_list);
473 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
474 ret = vmbus_post_msg(msg,
475 sizeof(struct vmbus_channel_gpadl_teardown));
476
477 if (ret)
478 goto post_msg_err;
479
480 wait_for_completion(&info->waitevent);
481
482 post_msg_err:
483 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
484 list_del(&info->msglistentry);
485 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
486
487 kfree(info);
488 return ret;
489 }
490 EXPORT_SYMBOL_GPL(vmbus_teardown_gpadl);
491
reset_channel_cb(void * arg)492 static void reset_channel_cb(void *arg)
493 {
494 struct vmbus_channel *channel = arg;
495
496 channel->onchannel_callback = NULL;
497 }
498
vmbus_close_internal(struct vmbus_channel * channel)499 static int vmbus_close_internal(struct vmbus_channel *channel)
500 {
501 struct vmbus_channel_close_channel *msg;
502 struct tasklet_struct *tasklet;
503 int ret;
504
505 /*
506 * process_chn_event(), running in the tasklet, can race
507 * with vmbus_close_internal() in the case of SMP guest, e.g., when
508 * the former is accessing channel->inbound.ring_buffer, the latter
509 * could be freeing the ring_buffer pages.
510 *
511 * To resolve the race, we can serialize them by disabling the
512 * tasklet when the latter is running here.
513 */
514 tasklet = hv_context.event_dpc[channel->target_cpu];
515 tasklet_disable(tasklet);
516
517 channel->state = CHANNEL_OPEN_STATE;
518 channel->sc_creation_callback = NULL;
519 /* Stop callback and cancel the timer asap */
520 if (channel->target_cpu != get_cpu()) {
521 put_cpu();
522 smp_call_function_single(channel->target_cpu, reset_channel_cb,
523 channel, true);
524 } else {
525 reset_channel_cb(channel);
526 put_cpu();
527 }
528
529 /* Send a closing message */
530
531 msg = &channel->close_msg.msg;
532
533 msg->header.msgtype = CHANNELMSG_CLOSECHANNEL;
534 msg->child_relid = channel->offermsg.child_relid;
535
536 ret = vmbus_post_msg(msg, sizeof(struct vmbus_channel_close_channel));
537
538 if (ret) {
539 pr_err("Close failed: close post msg return is %d\n", ret);
540 /*
541 * If we failed to post the close msg,
542 * it is perhaps better to leak memory.
543 */
544 goto out;
545 }
546
547 /* Tear down the gpadl for the channel's ring buffer */
548 if (channel->ringbuffer_gpadlhandle) {
549 ret = vmbus_teardown_gpadl(channel,
550 channel->ringbuffer_gpadlhandle);
551 if (ret) {
552 pr_err("Close failed: teardown gpadl return %d\n", ret);
553 /*
554 * If we failed to teardown gpadl,
555 * it is perhaps better to leak memory.
556 */
557 goto out;
558 }
559 }
560
561 /* Cleanup the ring buffers for this channel */
562 hv_ringbuffer_cleanup(&channel->outbound);
563 hv_ringbuffer_cleanup(&channel->inbound);
564
565 free_pages((unsigned long)channel->ringbuffer_pages,
566 get_order(channel->ringbuffer_pagecount * PAGE_SIZE));
567
568 out:
569 tasklet_enable(tasklet);
570
571 return ret;
572 }
573
574 /*
575 * vmbus_close - Close the specified channel
576 */
vmbus_close(struct vmbus_channel * channel)577 void vmbus_close(struct vmbus_channel *channel)
578 {
579 struct list_head *cur, *tmp;
580 struct vmbus_channel *cur_channel;
581
582 if (channel->primary_channel != NULL) {
583 /*
584 * We will only close sub-channels when
585 * the primary is closed.
586 */
587 return;
588 }
589 /*
590 * Close all the sub-channels first and then close the
591 * primary channel.
592 */
593 list_for_each_safe(cur, tmp, &channel->sc_list) {
594 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
595 if (cur_channel->state != CHANNEL_OPENED_STATE)
596 continue;
597 vmbus_close_internal(cur_channel);
598 }
599 /*
600 * Now close the primary.
601 */
602 vmbus_close_internal(channel);
603 }
604 EXPORT_SYMBOL_GPL(vmbus_close);
605
vmbus_sendpacket_ctl(struct vmbus_channel * channel,void * buffer,u32 bufferlen,u64 requestid,enum vmbus_packet_type type,u32 flags,bool kick_q)606 int vmbus_sendpacket_ctl(struct vmbus_channel *channel, void *buffer,
607 u32 bufferlen, u64 requestid,
608 enum vmbus_packet_type type, u32 flags, bool kick_q)
609 {
610 struct vmpacket_descriptor desc;
611 u32 packetlen = sizeof(struct vmpacket_descriptor) + bufferlen;
612 u32 packetlen_aligned = ALIGN(packetlen, sizeof(u64));
613 struct kvec bufferlist[3];
614 u64 aligned_data = 0;
615 int ret;
616 bool signal = false;
617 int num_vecs = ((bufferlen != 0) ? 3 : 1);
618
619
620 /* Setup the descriptor */
621 desc.type = type; /* VmbusPacketTypeDataInBand; */
622 desc.flags = flags; /* VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED; */
623 /* in 8-bytes granularity */
624 desc.offset8 = sizeof(struct vmpacket_descriptor) >> 3;
625 desc.len8 = (u16)(packetlen_aligned >> 3);
626 desc.trans_id = requestid;
627
628 bufferlist[0].iov_base = &desc;
629 bufferlist[0].iov_len = sizeof(struct vmpacket_descriptor);
630 bufferlist[1].iov_base = buffer;
631 bufferlist[1].iov_len = bufferlen;
632 bufferlist[2].iov_base = &aligned_data;
633 bufferlist[2].iov_len = (packetlen_aligned - packetlen);
634
635 ret = hv_ringbuffer_write(&channel->outbound, bufferlist, num_vecs,
636 &signal);
637
638 /*
639 * Signalling the host is conditional on many factors:
640 * 1. The ring state changed from being empty to non-empty.
641 * This is tracked by the variable "signal".
642 * 2. The variable kick_q tracks if more data will be placed
643 * on the ring. We will not signal if more data is
644 * to be placed.
645 *
646 * Based on the channel signal state, we will decide
647 * which signaling policy will be applied.
648 *
649 * If we cannot write to the ring-buffer; signal the host
650 * even if we may not have written anything. This is a rare
651 * enough condition that it should not matter.
652 */
653
654 if (channel->signal_policy)
655 signal = true;
656 else
657 kick_q = true;
658
659 if (((ret == 0) && kick_q && signal) || (ret))
660 vmbus_setevent(channel);
661
662 return ret;
663 }
664 EXPORT_SYMBOL(vmbus_sendpacket_ctl);
665
666 /**
667 * vmbus_sendpacket() - Send the specified buffer on the given channel
668 * @channel: Pointer to vmbus_channel structure.
669 * @buffer: Pointer to the buffer you want to receive the data into.
670 * @bufferlen: Maximum size of what the the buffer will hold
671 * @requestid: Identifier of the request
672 * @type: Type of packet that is being send e.g. negotiate, time
673 * packet etc.
674 *
675 * Sends data in @buffer directly to hyper-v via the vmbus
676 * This will send the data unparsed to hyper-v.
677 *
678 * Mainly used by Hyper-V drivers.
679 */
vmbus_sendpacket(struct vmbus_channel * channel,void * buffer,u32 bufferlen,u64 requestid,enum vmbus_packet_type type,u32 flags)680 int vmbus_sendpacket(struct vmbus_channel *channel, void *buffer,
681 u32 bufferlen, u64 requestid,
682 enum vmbus_packet_type type, u32 flags)
683 {
684 return vmbus_sendpacket_ctl(channel, buffer, bufferlen, requestid,
685 type, flags, true);
686 }
687 EXPORT_SYMBOL(vmbus_sendpacket);
688
689 /*
690 * vmbus_sendpacket_pagebuffer_ctl - Send a range of single-page buffer
691 * packets using a GPADL Direct packet type. This interface allows you
692 * to control notifying the host. This will be useful for sending
693 * batched data. Also the sender can control the send flags
694 * explicitly.
695 */
vmbus_sendpacket_pagebuffer_ctl(struct vmbus_channel * channel,struct hv_page_buffer pagebuffers[],u32 pagecount,void * buffer,u32 bufferlen,u64 requestid,u32 flags,bool kick_q)696 int vmbus_sendpacket_pagebuffer_ctl(struct vmbus_channel *channel,
697 struct hv_page_buffer pagebuffers[],
698 u32 pagecount, void *buffer, u32 bufferlen,
699 u64 requestid,
700 u32 flags,
701 bool kick_q)
702 {
703 int ret;
704 int i;
705 struct vmbus_channel_packet_page_buffer desc;
706 u32 descsize;
707 u32 packetlen;
708 u32 packetlen_aligned;
709 struct kvec bufferlist[3];
710 u64 aligned_data = 0;
711 bool signal = false;
712
713 if (pagecount > MAX_PAGE_BUFFER_COUNT)
714 return -EINVAL;
715
716
717 /*
718 * Adjust the size down since vmbus_channel_packet_page_buffer is the
719 * largest size we support
720 */
721 descsize = sizeof(struct vmbus_channel_packet_page_buffer) -
722 ((MAX_PAGE_BUFFER_COUNT - pagecount) *
723 sizeof(struct hv_page_buffer));
724 packetlen = descsize + bufferlen;
725 packetlen_aligned = ALIGN(packetlen, sizeof(u64));
726
727 /* Setup the descriptor */
728 desc.type = VM_PKT_DATA_USING_GPA_DIRECT;
729 desc.flags = flags;
730 desc.dataoffset8 = descsize >> 3; /* in 8-bytes grandularity */
731 desc.length8 = (u16)(packetlen_aligned >> 3);
732 desc.transactionid = requestid;
733 desc.rangecount = pagecount;
734
735 for (i = 0; i < pagecount; i++) {
736 desc.range[i].len = pagebuffers[i].len;
737 desc.range[i].offset = pagebuffers[i].offset;
738 desc.range[i].pfn = pagebuffers[i].pfn;
739 }
740
741 bufferlist[0].iov_base = &desc;
742 bufferlist[0].iov_len = descsize;
743 bufferlist[1].iov_base = buffer;
744 bufferlist[1].iov_len = bufferlen;
745 bufferlist[2].iov_base = &aligned_data;
746 bufferlist[2].iov_len = (packetlen_aligned - packetlen);
747
748 ret = hv_ringbuffer_write(&channel->outbound, bufferlist, 3, &signal);
749
750 /*
751 * Signalling the host is conditional on many factors:
752 * 1. The ring state changed from being empty to non-empty.
753 * This is tracked by the variable "signal".
754 * 2. The variable kick_q tracks if more data will be placed
755 * on the ring. We will not signal if more data is
756 * to be placed.
757 *
758 * Based on the channel signal state, we will decide
759 * which signaling policy will be applied.
760 *
761 * If we cannot write to the ring-buffer; signal the host
762 * even if we may not have written anything. This is a rare
763 * enough condition that it should not matter.
764 */
765
766 if (channel->signal_policy)
767 signal = true;
768 else
769 kick_q = true;
770
771 if (((ret == 0) && kick_q && signal) || (ret))
772 vmbus_setevent(channel);
773
774 return ret;
775 }
776 EXPORT_SYMBOL_GPL(vmbus_sendpacket_pagebuffer_ctl);
777
778 /*
779 * vmbus_sendpacket_pagebuffer - Send a range of single-page buffer
780 * packets using a GPADL Direct packet type.
781 */
vmbus_sendpacket_pagebuffer(struct vmbus_channel * channel,struct hv_page_buffer pagebuffers[],u32 pagecount,void * buffer,u32 bufferlen,u64 requestid)782 int vmbus_sendpacket_pagebuffer(struct vmbus_channel *channel,
783 struct hv_page_buffer pagebuffers[],
784 u32 pagecount, void *buffer, u32 bufferlen,
785 u64 requestid)
786 {
787 u32 flags = VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED;
788 return vmbus_sendpacket_pagebuffer_ctl(channel, pagebuffers, pagecount,
789 buffer, bufferlen, requestid,
790 flags, true);
791
792 }
793 EXPORT_SYMBOL_GPL(vmbus_sendpacket_pagebuffer);
794
795 /*
796 * vmbus_sendpacket_multipagebuffer - Send a multi-page buffer packet
797 * using a GPADL Direct packet type.
798 * The buffer includes the vmbus descriptor.
799 */
vmbus_sendpacket_mpb_desc(struct vmbus_channel * channel,struct vmbus_packet_mpb_array * desc,u32 desc_size,void * buffer,u32 bufferlen,u64 requestid)800 int vmbus_sendpacket_mpb_desc(struct vmbus_channel *channel,
801 struct vmbus_packet_mpb_array *desc,
802 u32 desc_size,
803 void *buffer, u32 bufferlen, u64 requestid)
804 {
805 int ret;
806 u32 packetlen;
807 u32 packetlen_aligned;
808 struct kvec bufferlist[3];
809 u64 aligned_data = 0;
810 bool signal = false;
811
812 packetlen = desc_size + bufferlen;
813 packetlen_aligned = ALIGN(packetlen, sizeof(u64));
814
815 /* Setup the descriptor */
816 desc->type = VM_PKT_DATA_USING_GPA_DIRECT;
817 desc->flags = VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED;
818 desc->dataoffset8 = desc_size >> 3; /* in 8-bytes grandularity */
819 desc->length8 = (u16)(packetlen_aligned >> 3);
820 desc->transactionid = requestid;
821 desc->rangecount = 1;
822
823 bufferlist[0].iov_base = desc;
824 bufferlist[0].iov_len = desc_size;
825 bufferlist[1].iov_base = buffer;
826 bufferlist[1].iov_len = bufferlen;
827 bufferlist[2].iov_base = &aligned_data;
828 bufferlist[2].iov_len = (packetlen_aligned - packetlen);
829
830 ret = hv_ringbuffer_write(&channel->outbound, bufferlist, 3, &signal);
831
832 if (ret == 0 && signal)
833 vmbus_setevent(channel);
834
835 return ret;
836 }
837 EXPORT_SYMBOL_GPL(vmbus_sendpacket_mpb_desc);
838
839 /*
840 * vmbus_sendpacket_multipagebuffer - Send a multi-page buffer packet
841 * using a GPADL Direct packet type.
842 */
vmbus_sendpacket_multipagebuffer(struct vmbus_channel * channel,struct hv_multipage_buffer * multi_pagebuffer,void * buffer,u32 bufferlen,u64 requestid)843 int vmbus_sendpacket_multipagebuffer(struct vmbus_channel *channel,
844 struct hv_multipage_buffer *multi_pagebuffer,
845 void *buffer, u32 bufferlen, u64 requestid)
846 {
847 int ret;
848 struct vmbus_channel_packet_multipage_buffer desc;
849 u32 descsize;
850 u32 packetlen;
851 u32 packetlen_aligned;
852 struct kvec bufferlist[3];
853 u64 aligned_data = 0;
854 bool signal = false;
855 u32 pfncount = NUM_PAGES_SPANNED(multi_pagebuffer->offset,
856 multi_pagebuffer->len);
857
858 if (pfncount > MAX_MULTIPAGE_BUFFER_COUNT)
859 return -EINVAL;
860
861 /*
862 * Adjust the size down since vmbus_channel_packet_multipage_buffer is
863 * the largest size we support
864 */
865 descsize = sizeof(struct vmbus_channel_packet_multipage_buffer) -
866 ((MAX_MULTIPAGE_BUFFER_COUNT - pfncount) *
867 sizeof(u64));
868 packetlen = descsize + bufferlen;
869 packetlen_aligned = ALIGN(packetlen, sizeof(u64));
870
871
872 /* Setup the descriptor */
873 desc.type = VM_PKT_DATA_USING_GPA_DIRECT;
874 desc.flags = VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED;
875 desc.dataoffset8 = descsize >> 3; /* in 8-bytes grandularity */
876 desc.length8 = (u16)(packetlen_aligned >> 3);
877 desc.transactionid = requestid;
878 desc.rangecount = 1;
879
880 desc.range.len = multi_pagebuffer->len;
881 desc.range.offset = multi_pagebuffer->offset;
882
883 memcpy(desc.range.pfn_array, multi_pagebuffer->pfn_array,
884 pfncount * sizeof(u64));
885
886 bufferlist[0].iov_base = &desc;
887 bufferlist[0].iov_len = descsize;
888 bufferlist[1].iov_base = buffer;
889 bufferlist[1].iov_len = bufferlen;
890 bufferlist[2].iov_base = &aligned_data;
891 bufferlist[2].iov_len = (packetlen_aligned - packetlen);
892
893 ret = hv_ringbuffer_write(&channel->outbound, bufferlist, 3, &signal);
894
895 if (ret == 0 && signal)
896 vmbus_setevent(channel);
897
898 return ret;
899 }
900 EXPORT_SYMBOL_GPL(vmbus_sendpacket_multipagebuffer);
901
902 /**
903 * vmbus_recvpacket() - Retrieve the user packet on the specified channel
904 * @channel: Pointer to vmbus_channel structure.
905 * @buffer: Pointer to the buffer you want to receive the data into.
906 * @bufferlen: Maximum size of what the the buffer will hold
907 * @buffer_actual_len: The actual size of the data after it was received
908 * @requestid: Identifier of the request
909 *
910 * Receives directly from the hyper-v vmbus and puts the data it received
911 * into Buffer. This will receive the data unparsed from hyper-v.
912 *
913 * Mainly used by Hyper-V drivers.
914 */
vmbus_recvpacket(struct vmbus_channel * channel,void * buffer,u32 bufferlen,u32 * buffer_actual_len,u64 * requestid)915 int vmbus_recvpacket(struct vmbus_channel *channel, void *buffer,
916 u32 bufferlen, u32 *buffer_actual_len, u64 *requestid)
917 {
918 struct vmpacket_descriptor desc;
919 u32 packetlen;
920 u32 userlen;
921 int ret;
922 bool signal = false;
923
924 *buffer_actual_len = 0;
925 *requestid = 0;
926
927
928 ret = hv_ringbuffer_peek(&channel->inbound, &desc,
929 sizeof(struct vmpacket_descriptor));
930 if (ret != 0)
931 return 0;
932
933 packetlen = desc.len8 << 3;
934 userlen = packetlen - (desc.offset8 << 3);
935
936 *buffer_actual_len = userlen;
937
938 if (userlen > bufferlen) {
939
940 pr_err("Buffer too small - got %d needs %d\n",
941 bufferlen, userlen);
942 return -ETOOSMALL;
943 }
944
945 *requestid = desc.trans_id;
946
947 /* Copy over the packet to the user buffer */
948 ret = hv_ringbuffer_read(&channel->inbound, buffer, userlen,
949 (desc.offset8 << 3), &signal);
950
951 if (signal)
952 vmbus_setevent(channel);
953
954 return 0;
955 }
956 EXPORT_SYMBOL(vmbus_recvpacket);
957
958 /*
959 * vmbus_recvpacket_raw - Retrieve the raw packet on the specified channel
960 */
vmbus_recvpacket_raw(struct vmbus_channel * channel,void * buffer,u32 bufferlen,u32 * buffer_actual_len,u64 * requestid)961 int vmbus_recvpacket_raw(struct vmbus_channel *channel, void *buffer,
962 u32 bufferlen, u32 *buffer_actual_len,
963 u64 *requestid)
964 {
965 struct vmpacket_descriptor desc;
966 u32 packetlen;
967 int ret;
968 bool signal = false;
969
970 *buffer_actual_len = 0;
971 *requestid = 0;
972
973
974 ret = hv_ringbuffer_peek(&channel->inbound, &desc,
975 sizeof(struct vmpacket_descriptor));
976 if (ret != 0)
977 return 0;
978
979
980 packetlen = desc.len8 << 3;
981
982 *buffer_actual_len = packetlen;
983
984 if (packetlen > bufferlen)
985 return -ENOBUFS;
986
987 *requestid = desc.trans_id;
988
989 /* Copy over the entire packet to the user buffer */
990 ret = hv_ringbuffer_read(&channel->inbound, buffer, packetlen, 0,
991 &signal);
992
993 if (signal)
994 vmbus_setevent(channel);
995
996 return ret;
997 }
998 EXPORT_SYMBOL_GPL(vmbus_recvpacket_raw);
999