Loading...
Note: File does not exist in v3.1.
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/interrupt.h>
25#include <linux/sched.h>
26#include <linux/wait.h>
27#include <linux/mm.h>
28#include <linux/slab.h>
29#include <linux/list.h>
30#include <linux/module.h>
31#include <linux/completion.h>
32#include <linux/delay.h>
33#include <linux/hyperv.h>
34#include <asm/mshyperv.h>
35
36#include "hyperv_vmbus.h"
37
38static void init_vp_index(struct vmbus_channel *channel, u16 dev_type);
39
40static const struct vmbus_device vmbus_devs[] = {
41 /* IDE */
42 { .dev_type = HV_IDE,
43 HV_IDE_GUID,
44 .perf_device = true,
45 },
46
47 /* SCSI */
48 { .dev_type = HV_SCSI,
49 HV_SCSI_GUID,
50 .perf_device = true,
51 },
52
53 /* Fibre Channel */
54 { .dev_type = HV_FC,
55 HV_SYNTHFC_GUID,
56 .perf_device = true,
57 },
58
59 /* Synthetic NIC */
60 { .dev_type = HV_NIC,
61 HV_NIC_GUID,
62 .perf_device = true,
63 },
64
65 /* Network Direct */
66 { .dev_type = HV_ND,
67 HV_ND_GUID,
68 .perf_device = true,
69 },
70
71 /* PCIE */
72 { .dev_type = HV_PCIE,
73 HV_PCIE_GUID,
74 .perf_device = false,
75 },
76
77 /* Synthetic Frame Buffer */
78 { .dev_type = HV_FB,
79 HV_SYNTHVID_GUID,
80 .perf_device = false,
81 },
82
83 /* Synthetic Keyboard */
84 { .dev_type = HV_KBD,
85 HV_KBD_GUID,
86 .perf_device = false,
87 },
88
89 /* Synthetic MOUSE */
90 { .dev_type = HV_MOUSE,
91 HV_MOUSE_GUID,
92 .perf_device = false,
93 },
94
95 /* KVP */
96 { .dev_type = HV_KVP,
97 HV_KVP_GUID,
98 .perf_device = false,
99 },
100
101 /* Time Synch */
102 { .dev_type = HV_TS,
103 HV_TS_GUID,
104 .perf_device = false,
105 },
106
107 /* Heartbeat */
108 { .dev_type = HV_HB,
109 HV_HEART_BEAT_GUID,
110 .perf_device = false,
111 },
112
113 /* Shutdown */
114 { .dev_type = HV_SHUTDOWN,
115 HV_SHUTDOWN_GUID,
116 .perf_device = false,
117 },
118
119 /* File copy */
120 { .dev_type = HV_FCOPY,
121 HV_FCOPY_GUID,
122 .perf_device = false,
123 },
124
125 /* Backup */
126 { .dev_type = HV_BACKUP,
127 HV_VSS_GUID,
128 .perf_device = false,
129 },
130
131 /* Dynamic Memory */
132 { .dev_type = HV_DM,
133 HV_DM_GUID,
134 .perf_device = false,
135 },
136
137 /* Unknown GUID */
138 { .dev_type = HV_UNKNOWN,
139 .perf_device = false,
140 },
141};
142
143static const struct {
144 uuid_le guid;
145} vmbus_unsupported_devs[] = {
146 { HV_AVMA1_GUID },
147 { HV_AVMA2_GUID },
148 { HV_RDV_GUID },
149};
150
151/*
152 * The rescinded channel may be blocked waiting for a response from the host;
153 * take care of that.
154 */
155static void vmbus_rescind_cleanup(struct vmbus_channel *channel)
156{
157 struct vmbus_channel_msginfo *msginfo;
158 unsigned long flags;
159
160
161 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
162 channel->rescind = true;
163 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
164 msglistentry) {
165
166 if (msginfo->waiting_channel == channel) {
167 complete(&msginfo->waitevent);
168 break;
169 }
170 }
171 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
172}
173
174static bool is_unsupported_vmbus_devs(const uuid_le *guid)
175{
176 int i;
177
178 for (i = 0; i < ARRAY_SIZE(vmbus_unsupported_devs); i++)
179 if (!uuid_le_cmp(*guid, vmbus_unsupported_devs[i].guid))
180 return true;
181 return false;
182}
183
184static u16 hv_get_dev_type(const struct vmbus_channel *channel)
185{
186 const uuid_le *guid = &channel->offermsg.offer.if_type;
187 u16 i;
188
189 if (is_hvsock_channel(channel) || is_unsupported_vmbus_devs(guid))
190 return HV_UNKNOWN;
191
192 for (i = HV_IDE; i < HV_UNKNOWN; i++) {
193 if (!uuid_le_cmp(*guid, vmbus_devs[i].guid))
194 return i;
195 }
196 pr_info("Unknown GUID: %pUl\n", guid);
197 return i;
198}
199
200/**
201 * vmbus_prep_negotiate_resp() - Create default response for Hyper-V Negotiate message
202 * @icmsghdrp: Pointer to msg header structure
203 * @icmsg_negotiate: Pointer to negotiate message structure
204 * @buf: Raw buffer channel data
205 *
206 * @icmsghdrp is of type &struct icmsg_hdr.
207 * Set up and fill in default negotiate response message.
208 *
209 * The fw_version and fw_vercnt specifies the framework version that
210 * we can support.
211 *
212 * The srv_version and srv_vercnt specifies the service
213 * versions we can support.
214 *
215 * Versions are given in decreasing order.
216 *
217 * nego_fw_version and nego_srv_version store the selected protocol versions.
218 *
219 * Mainly used by Hyper-V drivers.
220 */
221bool vmbus_prep_negotiate_resp(struct icmsg_hdr *icmsghdrp,
222 u8 *buf, const int *fw_version, int fw_vercnt,
223 const int *srv_version, int srv_vercnt,
224 int *nego_fw_version, int *nego_srv_version)
225{
226 int icframe_major, icframe_minor;
227 int icmsg_major, icmsg_minor;
228 int fw_major, fw_minor;
229 int srv_major, srv_minor;
230 int i, j;
231 bool found_match = false;
232 struct icmsg_negotiate *negop;
233
234 icmsghdrp->icmsgsize = 0x10;
235 negop = (struct icmsg_negotiate *)&buf[
236 sizeof(struct vmbuspipe_hdr) +
237 sizeof(struct icmsg_hdr)];
238
239 icframe_major = negop->icframe_vercnt;
240 icframe_minor = 0;
241
242 icmsg_major = negop->icmsg_vercnt;
243 icmsg_minor = 0;
244
245 /*
246 * Select the framework version number we will
247 * support.
248 */
249
250 for (i = 0; i < fw_vercnt; i++) {
251 fw_major = (fw_version[i] >> 16);
252 fw_minor = (fw_version[i] & 0xFFFF);
253
254 for (j = 0; j < negop->icframe_vercnt; j++) {
255 if ((negop->icversion_data[j].major == fw_major) &&
256 (negop->icversion_data[j].minor == fw_minor)) {
257 icframe_major = negop->icversion_data[j].major;
258 icframe_minor = negop->icversion_data[j].minor;
259 found_match = true;
260 break;
261 }
262 }
263
264 if (found_match)
265 break;
266 }
267
268 if (!found_match)
269 goto fw_error;
270
271 found_match = false;
272
273 for (i = 0; i < srv_vercnt; i++) {
274 srv_major = (srv_version[i] >> 16);
275 srv_minor = (srv_version[i] & 0xFFFF);
276
277 for (j = negop->icframe_vercnt;
278 (j < negop->icframe_vercnt + negop->icmsg_vercnt);
279 j++) {
280
281 if ((negop->icversion_data[j].major == srv_major) &&
282 (negop->icversion_data[j].minor == srv_minor)) {
283
284 icmsg_major = negop->icversion_data[j].major;
285 icmsg_minor = negop->icversion_data[j].minor;
286 found_match = true;
287 break;
288 }
289 }
290
291 if (found_match)
292 break;
293 }
294
295 /*
296 * Respond with the framework and service
297 * version numbers we can support.
298 */
299
300fw_error:
301 if (!found_match) {
302 negop->icframe_vercnt = 0;
303 negop->icmsg_vercnt = 0;
304 } else {
305 negop->icframe_vercnt = 1;
306 negop->icmsg_vercnt = 1;
307 }
308
309 if (nego_fw_version)
310 *nego_fw_version = (icframe_major << 16) | icframe_minor;
311
312 if (nego_srv_version)
313 *nego_srv_version = (icmsg_major << 16) | icmsg_minor;
314
315 negop->icversion_data[0].major = icframe_major;
316 negop->icversion_data[0].minor = icframe_minor;
317 negop->icversion_data[1].major = icmsg_major;
318 negop->icversion_data[1].minor = icmsg_minor;
319 return found_match;
320}
321
322EXPORT_SYMBOL_GPL(vmbus_prep_negotiate_resp);
323
324/*
325 * alloc_channel - Allocate and initialize a vmbus channel object
326 */
327static struct vmbus_channel *alloc_channel(void)
328{
329 struct vmbus_channel *channel;
330
331 channel = kzalloc(sizeof(*channel), GFP_ATOMIC);
332 if (!channel)
333 return NULL;
334
335 spin_lock_init(&channel->lock);
336 init_completion(&channel->rescind_event);
337
338 INIT_LIST_HEAD(&channel->sc_list);
339 INIT_LIST_HEAD(&channel->percpu_list);
340
341 tasklet_init(&channel->callback_event,
342 vmbus_on_event, (unsigned long)channel);
343
344 return channel;
345}
346
347/*
348 * free_channel - Release the resources used by the vmbus channel object
349 */
350static void free_channel(struct vmbus_channel *channel)
351{
352 tasklet_kill(&channel->callback_event);
353
354 kobject_put(&channel->kobj);
355}
356
357static void percpu_channel_enq(void *arg)
358{
359 struct vmbus_channel *channel = arg;
360 struct hv_per_cpu_context *hv_cpu
361 = this_cpu_ptr(hv_context.cpu_context);
362
363 list_add_tail_rcu(&channel->percpu_list, &hv_cpu->chan_list);
364}
365
366static void percpu_channel_deq(void *arg)
367{
368 struct vmbus_channel *channel = arg;
369
370 list_del_rcu(&channel->percpu_list);
371}
372
373
374static void vmbus_release_relid(u32 relid)
375{
376 struct vmbus_channel_relid_released msg;
377 int ret;
378
379 memset(&msg, 0, sizeof(struct vmbus_channel_relid_released));
380 msg.child_relid = relid;
381 msg.header.msgtype = CHANNELMSG_RELID_RELEASED;
382 ret = vmbus_post_msg(&msg, sizeof(struct vmbus_channel_relid_released),
383 true);
384
385 trace_vmbus_release_relid(&msg, ret);
386}
387
388void hv_process_channel_removal(u32 relid)
389{
390 unsigned long flags;
391 struct vmbus_channel *primary_channel, *channel;
392
393 BUG_ON(!mutex_is_locked(&vmbus_connection.channel_mutex));
394
395 /*
396 * Make sure channel is valid as we may have raced.
397 */
398 channel = relid2channel(relid);
399 if (!channel)
400 return;
401
402 BUG_ON(!channel->rescind);
403 if (channel->target_cpu != get_cpu()) {
404 put_cpu();
405 smp_call_function_single(channel->target_cpu,
406 percpu_channel_deq, channel, true);
407 } else {
408 percpu_channel_deq(channel);
409 put_cpu();
410 }
411
412 if (channel->primary_channel == NULL) {
413 list_del(&channel->listentry);
414
415 primary_channel = channel;
416 } else {
417 primary_channel = channel->primary_channel;
418 spin_lock_irqsave(&primary_channel->lock, flags);
419 list_del(&channel->sc_list);
420 primary_channel->num_sc--;
421 spin_unlock_irqrestore(&primary_channel->lock, flags);
422 }
423
424 /*
425 * We need to free the bit for init_vp_index() to work in the case
426 * of sub-channel, when we reload drivers like hv_netvsc.
427 */
428 if (channel->affinity_policy == HV_LOCALIZED)
429 cpumask_clear_cpu(channel->target_cpu,
430 &primary_channel->alloced_cpus_in_node);
431
432 vmbus_release_relid(relid);
433
434 free_channel(channel);
435}
436
437void vmbus_free_channels(void)
438{
439 struct vmbus_channel *channel, *tmp;
440
441 list_for_each_entry_safe(channel, tmp, &vmbus_connection.chn_list,
442 listentry) {
443 /* hv_process_channel_removal() needs this */
444 channel->rescind = true;
445
446 vmbus_device_unregister(channel->device_obj);
447 }
448}
449
450/*
451 * vmbus_process_offer - Process the offer by creating a channel/device
452 * associated with this offer
453 */
454static void vmbus_process_offer(struct vmbus_channel *newchannel)
455{
456 struct vmbus_channel *channel;
457 bool fnew = true;
458 unsigned long flags;
459 u16 dev_type;
460 int ret;
461
462 /* Make sure this is a new offer */
463 mutex_lock(&vmbus_connection.channel_mutex);
464
465 /*
466 * Now that we have acquired the channel_mutex,
467 * we can release the potentially racing rescind thread.
468 */
469 atomic_dec(&vmbus_connection.offer_in_progress);
470
471 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
472 if (!uuid_le_cmp(channel->offermsg.offer.if_type,
473 newchannel->offermsg.offer.if_type) &&
474 !uuid_le_cmp(channel->offermsg.offer.if_instance,
475 newchannel->offermsg.offer.if_instance)) {
476 fnew = false;
477 break;
478 }
479 }
480
481 if (fnew)
482 list_add_tail(&newchannel->listentry,
483 &vmbus_connection.chn_list);
484
485 mutex_unlock(&vmbus_connection.channel_mutex);
486
487 if (!fnew) {
488 /*
489 * Check to see if this is a sub-channel.
490 */
491 if (newchannel->offermsg.offer.sub_channel_index != 0) {
492 /*
493 * Process the sub-channel.
494 */
495 newchannel->primary_channel = channel;
496 spin_lock_irqsave(&channel->lock, flags);
497 list_add_tail(&newchannel->sc_list, &channel->sc_list);
498 channel->num_sc++;
499 spin_unlock_irqrestore(&channel->lock, flags);
500 } else {
501 goto err_free_chan;
502 }
503 }
504
505 dev_type = hv_get_dev_type(newchannel);
506
507 init_vp_index(newchannel, dev_type);
508
509 if (newchannel->target_cpu != get_cpu()) {
510 put_cpu();
511 smp_call_function_single(newchannel->target_cpu,
512 percpu_channel_enq,
513 newchannel, true);
514 } else {
515 percpu_channel_enq(newchannel);
516 put_cpu();
517 }
518
519 /*
520 * This state is used to indicate a successful open
521 * so that when we do close the channel normally, we
522 * can cleanup properly
523 */
524 newchannel->state = CHANNEL_OPEN_STATE;
525
526 if (!fnew) {
527 struct hv_device *dev
528 = newchannel->primary_channel->device_obj;
529
530 if (vmbus_add_channel_kobj(dev, newchannel)) {
531 atomic_dec(&vmbus_connection.offer_in_progress);
532 goto err_free_chan;
533 }
534
535 if (channel->sc_creation_callback != NULL)
536 channel->sc_creation_callback(newchannel);
537 newchannel->probe_done = true;
538 return;
539 }
540
541 /*
542 * Start the process of binding this offer to the driver
543 * We need to set the DeviceObject field before calling
544 * vmbus_child_dev_add()
545 */
546 newchannel->device_obj = vmbus_device_create(
547 &newchannel->offermsg.offer.if_type,
548 &newchannel->offermsg.offer.if_instance,
549 newchannel);
550 if (!newchannel->device_obj)
551 goto err_deq_chan;
552
553 newchannel->device_obj->device_id = dev_type;
554 /*
555 * Add the new device to the bus. This will kick off device-driver
556 * binding which eventually invokes the device driver's AddDevice()
557 * method.
558 */
559 ret = vmbus_device_register(newchannel->device_obj);
560
561 if (ret != 0) {
562 pr_err("unable to add child device object (relid %d)\n",
563 newchannel->offermsg.child_relid);
564 kfree(newchannel->device_obj);
565 goto err_deq_chan;
566 }
567
568 newchannel->probe_done = true;
569 return;
570
571err_deq_chan:
572 mutex_lock(&vmbus_connection.channel_mutex);
573 list_del(&newchannel->listentry);
574 mutex_unlock(&vmbus_connection.channel_mutex);
575
576 if (newchannel->target_cpu != get_cpu()) {
577 put_cpu();
578 smp_call_function_single(newchannel->target_cpu,
579 percpu_channel_deq, newchannel, true);
580 } else {
581 percpu_channel_deq(newchannel);
582 put_cpu();
583 }
584
585 vmbus_release_relid(newchannel->offermsg.child_relid);
586
587err_free_chan:
588 free_channel(newchannel);
589}
590
591/*
592 * We use this state to statically distribute the channel interrupt load.
593 */
594static int next_numa_node_id;
595
596/*
597 * Starting with Win8, we can statically distribute the incoming
598 * channel interrupt load by binding a channel to VCPU.
599 * We distribute the interrupt loads to one or more NUMA nodes based on
600 * the channel's affinity_policy.
601 *
602 * For pre-win8 hosts or non-performance critical channels we assign the
603 * first CPU in the first NUMA node.
604 */
605static void init_vp_index(struct vmbus_channel *channel, u16 dev_type)
606{
607 u32 cur_cpu;
608 bool perf_chn = vmbus_devs[dev_type].perf_device;
609 struct vmbus_channel *primary = channel->primary_channel;
610 int next_node;
611 struct cpumask available_mask;
612 struct cpumask *alloced_mask;
613
614 if ((vmbus_proto_version == VERSION_WS2008) ||
615 (vmbus_proto_version == VERSION_WIN7) || (!perf_chn)) {
616 /*
617 * Prior to win8, all channel interrupts are
618 * delivered on cpu 0.
619 * Also if the channel is not a performance critical
620 * channel, bind it to cpu 0.
621 */
622 channel->numa_node = 0;
623 channel->target_cpu = 0;
624 channel->target_vp = hv_cpu_number_to_vp_number(0);
625 return;
626 }
627
628 /*
629 * Based on the channel affinity policy, we will assign the NUMA
630 * nodes.
631 */
632
633 if ((channel->affinity_policy == HV_BALANCED) || (!primary)) {
634 while (true) {
635 next_node = next_numa_node_id++;
636 if (next_node == nr_node_ids) {
637 next_node = next_numa_node_id = 0;
638 continue;
639 }
640 if (cpumask_empty(cpumask_of_node(next_node)))
641 continue;
642 break;
643 }
644 channel->numa_node = next_node;
645 primary = channel;
646 }
647 alloced_mask = &hv_context.hv_numa_map[primary->numa_node];
648
649 if (cpumask_weight(alloced_mask) ==
650 cpumask_weight(cpumask_of_node(primary->numa_node))) {
651 /*
652 * We have cycled through all the CPUs in the node;
653 * reset the alloced map.
654 */
655 cpumask_clear(alloced_mask);
656 }
657
658 cpumask_xor(&available_mask, alloced_mask,
659 cpumask_of_node(primary->numa_node));
660
661 cur_cpu = -1;
662
663 if (primary->affinity_policy == HV_LOCALIZED) {
664 /*
665 * Normally Hyper-V host doesn't create more subchannels
666 * than there are VCPUs on the node but it is possible when not
667 * all present VCPUs on the node are initialized by guest.
668 * Clear the alloced_cpus_in_node to start over.
669 */
670 if (cpumask_equal(&primary->alloced_cpus_in_node,
671 cpumask_of_node(primary->numa_node)))
672 cpumask_clear(&primary->alloced_cpus_in_node);
673 }
674
675 while (true) {
676 cur_cpu = cpumask_next(cur_cpu, &available_mask);
677 if (cur_cpu >= nr_cpu_ids) {
678 cur_cpu = -1;
679 cpumask_copy(&available_mask,
680 cpumask_of_node(primary->numa_node));
681 continue;
682 }
683
684 if (primary->affinity_policy == HV_LOCALIZED) {
685 /*
686 * NOTE: in the case of sub-channel, we clear the
687 * sub-channel related bit(s) in
688 * primary->alloced_cpus_in_node in
689 * hv_process_channel_removal(), so when we
690 * reload drivers like hv_netvsc in SMP guest, here
691 * we're able to re-allocate
692 * bit from primary->alloced_cpus_in_node.
693 */
694 if (!cpumask_test_cpu(cur_cpu,
695 &primary->alloced_cpus_in_node)) {
696 cpumask_set_cpu(cur_cpu,
697 &primary->alloced_cpus_in_node);
698 cpumask_set_cpu(cur_cpu, alloced_mask);
699 break;
700 }
701 } else {
702 cpumask_set_cpu(cur_cpu, alloced_mask);
703 break;
704 }
705 }
706
707 channel->target_cpu = cur_cpu;
708 channel->target_vp = hv_cpu_number_to_vp_number(cur_cpu);
709}
710
711static void vmbus_wait_for_unload(void)
712{
713 int cpu;
714 void *page_addr;
715 struct hv_message *msg;
716 struct vmbus_channel_message_header *hdr;
717 u32 message_type;
718
719 /*
720 * CHANNELMSG_UNLOAD_RESPONSE is always delivered to the CPU which was
721 * used for initial contact or to CPU0 depending on host version. When
722 * we're crashing on a different CPU let's hope that IRQ handler on
723 * the cpu which receives CHANNELMSG_UNLOAD_RESPONSE is still
724 * functional and vmbus_unload_response() will complete
725 * vmbus_connection.unload_event. If not, the last thing we can do is
726 * read message pages for all CPUs directly.
727 */
728 while (1) {
729 if (completion_done(&vmbus_connection.unload_event))
730 break;
731
732 for_each_online_cpu(cpu) {
733 struct hv_per_cpu_context *hv_cpu
734 = per_cpu_ptr(hv_context.cpu_context, cpu);
735
736 page_addr = hv_cpu->synic_message_page;
737 msg = (struct hv_message *)page_addr
738 + VMBUS_MESSAGE_SINT;
739
740 message_type = READ_ONCE(msg->header.message_type);
741 if (message_type == HVMSG_NONE)
742 continue;
743
744 hdr = (struct vmbus_channel_message_header *)
745 msg->u.payload;
746
747 if (hdr->msgtype == CHANNELMSG_UNLOAD_RESPONSE)
748 complete(&vmbus_connection.unload_event);
749
750 vmbus_signal_eom(msg, message_type);
751 }
752
753 mdelay(10);
754 }
755
756 /*
757 * We're crashing and already got the UNLOAD_RESPONSE, cleanup all
758 * maybe-pending messages on all CPUs to be able to receive new
759 * messages after we reconnect.
760 */
761 for_each_online_cpu(cpu) {
762 struct hv_per_cpu_context *hv_cpu
763 = per_cpu_ptr(hv_context.cpu_context, cpu);
764
765 page_addr = hv_cpu->synic_message_page;
766 msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
767 msg->header.message_type = HVMSG_NONE;
768 }
769}
770
771/*
772 * vmbus_unload_response - Handler for the unload response.
773 */
774static void vmbus_unload_response(struct vmbus_channel_message_header *hdr)
775{
776 /*
777 * This is a global event; just wakeup the waiting thread.
778 * Once we successfully unload, we can cleanup the monitor state.
779 */
780 complete(&vmbus_connection.unload_event);
781}
782
783void vmbus_initiate_unload(bool crash)
784{
785 struct vmbus_channel_message_header hdr;
786
787 /* Pre-Win2012R2 hosts don't support reconnect */
788 if (vmbus_proto_version < VERSION_WIN8_1)
789 return;
790
791 init_completion(&vmbus_connection.unload_event);
792 memset(&hdr, 0, sizeof(struct vmbus_channel_message_header));
793 hdr.msgtype = CHANNELMSG_UNLOAD;
794 vmbus_post_msg(&hdr, sizeof(struct vmbus_channel_message_header),
795 !crash);
796
797 /*
798 * vmbus_initiate_unload() is also called on crash and the crash can be
799 * happening in an interrupt context, where scheduling is impossible.
800 */
801 if (!crash)
802 wait_for_completion(&vmbus_connection.unload_event);
803 else
804 vmbus_wait_for_unload();
805}
806
807/*
808 * vmbus_onoffer - Handler for channel offers from vmbus in parent partition.
809 *
810 */
811static void vmbus_onoffer(struct vmbus_channel_message_header *hdr)
812{
813 struct vmbus_channel_offer_channel *offer;
814 struct vmbus_channel *newchannel;
815
816 offer = (struct vmbus_channel_offer_channel *)hdr;
817
818 trace_vmbus_onoffer(offer);
819
820 /* Allocate the channel object and save this offer. */
821 newchannel = alloc_channel();
822 if (!newchannel) {
823 vmbus_release_relid(offer->child_relid);
824 atomic_dec(&vmbus_connection.offer_in_progress);
825 pr_err("Unable to allocate channel object\n");
826 return;
827 }
828
829 /*
830 * Setup state for signalling the host.
831 */
832 newchannel->sig_event = VMBUS_EVENT_CONNECTION_ID;
833
834 if (vmbus_proto_version != VERSION_WS2008) {
835 newchannel->is_dedicated_interrupt =
836 (offer->is_dedicated_interrupt != 0);
837 newchannel->sig_event = offer->connection_id;
838 }
839
840 memcpy(&newchannel->offermsg, offer,
841 sizeof(struct vmbus_channel_offer_channel));
842 newchannel->monitor_grp = (u8)offer->monitorid / 32;
843 newchannel->monitor_bit = (u8)offer->monitorid % 32;
844
845 vmbus_process_offer(newchannel);
846}
847
848/*
849 * vmbus_onoffer_rescind - Rescind offer handler.
850 *
851 * We queue a work item to process this offer synchronously
852 */
853static void vmbus_onoffer_rescind(struct vmbus_channel_message_header *hdr)
854{
855 struct vmbus_channel_rescind_offer *rescind;
856 struct vmbus_channel *channel;
857 struct device *dev;
858
859 rescind = (struct vmbus_channel_rescind_offer *)hdr;
860
861 trace_vmbus_onoffer_rescind(rescind);
862
863 /*
864 * The offer msg and the corresponding rescind msg
865 * from the host are guranteed to be ordered -
866 * offer comes in first and then the rescind.
867 * Since we process these events in work elements,
868 * and with preemption, we may end up processing
869 * the events out of order. Given that we handle these
870 * work elements on the same CPU, this is possible only
871 * in the case of preemption. In any case wait here
872 * until the offer processing has moved beyond the
873 * point where the channel is discoverable.
874 */
875
876 while (atomic_read(&vmbus_connection.offer_in_progress) != 0) {
877 /*
878 * We wait here until any channel offer is currently
879 * being processed.
880 */
881 msleep(1);
882 }
883
884 mutex_lock(&vmbus_connection.channel_mutex);
885 channel = relid2channel(rescind->child_relid);
886 mutex_unlock(&vmbus_connection.channel_mutex);
887
888 if (channel == NULL) {
889 /*
890 * We failed in processing the offer message;
891 * we would have cleaned up the relid in that
892 * failure path.
893 */
894 return;
895 }
896
897 /*
898 * Now wait for offer handling to complete.
899 */
900 vmbus_rescind_cleanup(channel);
901 while (READ_ONCE(channel->probe_done) == false) {
902 /*
903 * We wait here until any channel offer is currently
904 * being processed.
905 */
906 msleep(1);
907 }
908
909 /*
910 * At this point, the rescind handling can proceed safely.
911 */
912
913 if (channel->device_obj) {
914 if (channel->chn_rescind_callback) {
915 channel->chn_rescind_callback(channel);
916 return;
917 }
918 /*
919 * We will have to unregister this device from the
920 * driver core.
921 */
922 dev = get_device(&channel->device_obj->device);
923 if (dev) {
924 vmbus_device_unregister(channel->device_obj);
925 put_device(dev);
926 }
927 }
928 if (channel->primary_channel != NULL) {
929 /*
930 * Sub-channel is being rescinded. Following is the channel
931 * close sequence when initiated from the driveri (refer to
932 * vmbus_close() for details):
933 * 1. Close all sub-channels first
934 * 2. Then close the primary channel.
935 */
936 mutex_lock(&vmbus_connection.channel_mutex);
937 if (channel->state == CHANNEL_OPEN_STATE) {
938 /*
939 * The channel is currently not open;
940 * it is safe for us to cleanup the channel.
941 */
942 hv_process_channel_removal(rescind->child_relid);
943 } else {
944 complete(&channel->rescind_event);
945 }
946 mutex_unlock(&vmbus_connection.channel_mutex);
947 }
948}
949
950void vmbus_hvsock_device_unregister(struct vmbus_channel *channel)
951{
952 BUG_ON(!is_hvsock_channel(channel));
953
954 /* We always get a rescind msg when a connection is closed. */
955 while (!READ_ONCE(channel->probe_done) || !READ_ONCE(channel->rescind))
956 msleep(1);
957
958 vmbus_device_unregister(channel->device_obj);
959}
960EXPORT_SYMBOL_GPL(vmbus_hvsock_device_unregister);
961
962
963/*
964 * vmbus_onoffers_delivered -
965 * This is invoked when all offers have been delivered.
966 *
967 * Nothing to do here.
968 */
969static void vmbus_onoffers_delivered(
970 struct vmbus_channel_message_header *hdr)
971{
972}
973
974/*
975 * vmbus_onopen_result - Open result handler.
976 *
977 * This is invoked when we received a response to our channel open request.
978 * Find the matching request, copy the response and signal the requesting
979 * thread.
980 */
981static void vmbus_onopen_result(struct vmbus_channel_message_header *hdr)
982{
983 struct vmbus_channel_open_result *result;
984 struct vmbus_channel_msginfo *msginfo;
985 struct vmbus_channel_message_header *requestheader;
986 struct vmbus_channel_open_channel *openmsg;
987 unsigned long flags;
988
989 result = (struct vmbus_channel_open_result *)hdr;
990
991 trace_vmbus_onopen_result(result);
992
993 /*
994 * Find the open msg, copy the result and signal/unblock the wait event
995 */
996 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
997
998 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
999 msglistentry) {
1000 requestheader =
1001 (struct vmbus_channel_message_header *)msginfo->msg;
1002
1003 if (requestheader->msgtype == CHANNELMSG_OPENCHANNEL) {
1004 openmsg =
1005 (struct vmbus_channel_open_channel *)msginfo->msg;
1006 if (openmsg->child_relid == result->child_relid &&
1007 openmsg->openid == result->openid) {
1008 memcpy(&msginfo->response.open_result,
1009 result,
1010 sizeof(
1011 struct vmbus_channel_open_result));
1012 complete(&msginfo->waitevent);
1013 break;
1014 }
1015 }
1016 }
1017 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1018}
1019
1020/*
1021 * vmbus_ongpadl_created - GPADL created handler.
1022 *
1023 * This is invoked when we received a response to our gpadl create request.
1024 * Find the matching request, copy the response and signal the requesting
1025 * thread.
1026 */
1027static void vmbus_ongpadl_created(struct vmbus_channel_message_header *hdr)
1028{
1029 struct vmbus_channel_gpadl_created *gpadlcreated;
1030 struct vmbus_channel_msginfo *msginfo;
1031 struct vmbus_channel_message_header *requestheader;
1032 struct vmbus_channel_gpadl_header *gpadlheader;
1033 unsigned long flags;
1034
1035 gpadlcreated = (struct vmbus_channel_gpadl_created *)hdr;
1036
1037 trace_vmbus_ongpadl_created(gpadlcreated);
1038
1039 /*
1040 * Find the establish msg, copy the result and signal/unblock the wait
1041 * event
1042 */
1043 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1044
1045 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1046 msglistentry) {
1047 requestheader =
1048 (struct vmbus_channel_message_header *)msginfo->msg;
1049
1050 if (requestheader->msgtype == CHANNELMSG_GPADL_HEADER) {
1051 gpadlheader =
1052 (struct vmbus_channel_gpadl_header *)requestheader;
1053
1054 if ((gpadlcreated->child_relid ==
1055 gpadlheader->child_relid) &&
1056 (gpadlcreated->gpadl == gpadlheader->gpadl)) {
1057 memcpy(&msginfo->response.gpadl_created,
1058 gpadlcreated,
1059 sizeof(
1060 struct vmbus_channel_gpadl_created));
1061 complete(&msginfo->waitevent);
1062 break;
1063 }
1064 }
1065 }
1066 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1067}
1068
1069/*
1070 * vmbus_ongpadl_torndown - GPADL torndown handler.
1071 *
1072 * This is invoked when we received a response to our gpadl teardown request.
1073 * Find the matching request, copy the response and signal the requesting
1074 * thread.
1075 */
1076static void vmbus_ongpadl_torndown(
1077 struct vmbus_channel_message_header *hdr)
1078{
1079 struct vmbus_channel_gpadl_torndown *gpadl_torndown;
1080 struct vmbus_channel_msginfo *msginfo;
1081 struct vmbus_channel_message_header *requestheader;
1082 struct vmbus_channel_gpadl_teardown *gpadl_teardown;
1083 unsigned long flags;
1084
1085 gpadl_torndown = (struct vmbus_channel_gpadl_torndown *)hdr;
1086
1087 trace_vmbus_ongpadl_torndown(gpadl_torndown);
1088
1089 /*
1090 * Find the open msg, copy the result and signal/unblock the wait event
1091 */
1092 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1093
1094 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1095 msglistentry) {
1096 requestheader =
1097 (struct vmbus_channel_message_header *)msginfo->msg;
1098
1099 if (requestheader->msgtype == CHANNELMSG_GPADL_TEARDOWN) {
1100 gpadl_teardown =
1101 (struct vmbus_channel_gpadl_teardown *)requestheader;
1102
1103 if (gpadl_torndown->gpadl == gpadl_teardown->gpadl) {
1104 memcpy(&msginfo->response.gpadl_torndown,
1105 gpadl_torndown,
1106 sizeof(
1107 struct vmbus_channel_gpadl_torndown));
1108 complete(&msginfo->waitevent);
1109 break;
1110 }
1111 }
1112 }
1113 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1114}
1115
1116/*
1117 * vmbus_onversion_response - Version response handler
1118 *
1119 * This is invoked when we received a response to our initiate contact request.
1120 * Find the matching request, copy the response and signal the requesting
1121 * thread.
1122 */
1123static void vmbus_onversion_response(
1124 struct vmbus_channel_message_header *hdr)
1125{
1126 struct vmbus_channel_msginfo *msginfo;
1127 struct vmbus_channel_message_header *requestheader;
1128 struct vmbus_channel_version_response *version_response;
1129 unsigned long flags;
1130
1131 version_response = (struct vmbus_channel_version_response *)hdr;
1132
1133 trace_vmbus_onversion_response(version_response);
1134
1135 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1136
1137 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1138 msglistentry) {
1139 requestheader =
1140 (struct vmbus_channel_message_header *)msginfo->msg;
1141
1142 if (requestheader->msgtype ==
1143 CHANNELMSG_INITIATE_CONTACT) {
1144 memcpy(&msginfo->response.version_response,
1145 version_response,
1146 sizeof(struct vmbus_channel_version_response));
1147 complete(&msginfo->waitevent);
1148 }
1149 }
1150 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1151}
1152
1153/* Channel message dispatch table */
1154const struct vmbus_channel_message_table_entry
1155channel_message_table[CHANNELMSG_COUNT] = {
1156 { CHANNELMSG_INVALID, 0, NULL },
1157 { CHANNELMSG_OFFERCHANNEL, 0, vmbus_onoffer },
1158 { CHANNELMSG_RESCIND_CHANNELOFFER, 0, vmbus_onoffer_rescind },
1159 { CHANNELMSG_REQUESTOFFERS, 0, NULL },
1160 { CHANNELMSG_ALLOFFERS_DELIVERED, 1, vmbus_onoffers_delivered },
1161 { CHANNELMSG_OPENCHANNEL, 0, NULL },
1162 { CHANNELMSG_OPENCHANNEL_RESULT, 1, vmbus_onopen_result },
1163 { CHANNELMSG_CLOSECHANNEL, 0, NULL },
1164 { CHANNELMSG_GPADL_HEADER, 0, NULL },
1165 { CHANNELMSG_GPADL_BODY, 0, NULL },
1166 { CHANNELMSG_GPADL_CREATED, 1, vmbus_ongpadl_created },
1167 { CHANNELMSG_GPADL_TEARDOWN, 0, NULL },
1168 { CHANNELMSG_GPADL_TORNDOWN, 1, vmbus_ongpadl_torndown },
1169 { CHANNELMSG_RELID_RELEASED, 0, NULL },
1170 { CHANNELMSG_INITIATE_CONTACT, 0, NULL },
1171 { CHANNELMSG_VERSION_RESPONSE, 1, vmbus_onversion_response },
1172 { CHANNELMSG_UNLOAD, 0, NULL },
1173 { CHANNELMSG_UNLOAD_RESPONSE, 1, vmbus_unload_response },
1174 { CHANNELMSG_18, 0, NULL },
1175 { CHANNELMSG_19, 0, NULL },
1176 { CHANNELMSG_20, 0, NULL },
1177 { CHANNELMSG_TL_CONNECT_REQUEST, 0, NULL },
1178};
1179
1180/*
1181 * vmbus_onmessage - Handler for channel protocol messages.
1182 *
1183 * This is invoked in the vmbus worker thread context.
1184 */
1185void vmbus_onmessage(void *context)
1186{
1187 struct hv_message *msg = context;
1188 struct vmbus_channel_message_header *hdr;
1189 int size;
1190
1191 hdr = (struct vmbus_channel_message_header *)msg->u.payload;
1192 size = msg->header.payload_size;
1193
1194 trace_vmbus_on_message(hdr);
1195
1196 if (hdr->msgtype >= CHANNELMSG_COUNT) {
1197 pr_err("Received invalid channel message type %d size %d\n",
1198 hdr->msgtype, size);
1199 print_hex_dump_bytes("", DUMP_PREFIX_NONE,
1200 (unsigned char *)msg->u.payload, size);
1201 return;
1202 }
1203
1204 if (channel_message_table[hdr->msgtype].message_handler)
1205 channel_message_table[hdr->msgtype].message_handler(hdr);
1206 else
1207 pr_err("Unhandled channel message type %d\n", hdr->msgtype);
1208}
1209
1210/*
1211 * vmbus_request_offers - Send a request to get all our pending offers.
1212 */
1213int vmbus_request_offers(void)
1214{
1215 struct vmbus_channel_message_header *msg;
1216 struct vmbus_channel_msginfo *msginfo;
1217 int ret;
1218
1219 msginfo = kmalloc(sizeof(*msginfo) +
1220 sizeof(struct vmbus_channel_message_header),
1221 GFP_KERNEL);
1222 if (!msginfo)
1223 return -ENOMEM;
1224
1225 msg = (struct vmbus_channel_message_header *)msginfo->msg;
1226
1227 msg->msgtype = CHANNELMSG_REQUESTOFFERS;
1228
1229 ret = vmbus_post_msg(msg, sizeof(struct vmbus_channel_message_header),
1230 true);
1231
1232 trace_vmbus_request_offers(ret);
1233
1234 if (ret != 0) {
1235 pr_err("Unable to request offers - %d\n", ret);
1236
1237 goto cleanup;
1238 }
1239
1240cleanup:
1241 kfree(msginfo);
1242
1243 return ret;
1244}
1245
1246/*
1247 * Retrieve the (sub) channel on which to send an outgoing request.
1248 * When a primary channel has multiple sub-channels, we try to
1249 * distribute the load equally amongst all available channels.
1250 */
1251struct vmbus_channel *vmbus_get_outgoing_channel(struct vmbus_channel *primary)
1252{
1253 struct list_head *cur, *tmp;
1254 int cur_cpu;
1255 struct vmbus_channel *cur_channel;
1256 struct vmbus_channel *outgoing_channel = primary;
1257 int next_channel;
1258 int i = 1;
1259
1260 if (list_empty(&primary->sc_list))
1261 return outgoing_channel;
1262
1263 next_channel = primary->next_oc++;
1264
1265 if (next_channel > (primary->num_sc)) {
1266 primary->next_oc = 0;
1267 return outgoing_channel;
1268 }
1269
1270 cur_cpu = hv_cpu_number_to_vp_number(smp_processor_id());
1271 list_for_each_safe(cur, tmp, &primary->sc_list) {
1272 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
1273 if (cur_channel->state != CHANNEL_OPENED_STATE)
1274 continue;
1275
1276 if (cur_channel->target_vp == cur_cpu)
1277 return cur_channel;
1278
1279 if (i == next_channel)
1280 return cur_channel;
1281
1282 i++;
1283 }
1284
1285 return outgoing_channel;
1286}
1287EXPORT_SYMBOL_GPL(vmbus_get_outgoing_channel);
1288
1289static void invoke_sc_cb(struct vmbus_channel *primary_channel)
1290{
1291 struct list_head *cur, *tmp;
1292 struct vmbus_channel *cur_channel;
1293
1294 if (primary_channel->sc_creation_callback == NULL)
1295 return;
1296
1297 list_for_each_safe(cur, tmp, &primary_channel->sc_list) {
1298 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
1299
1300 primary_channel->sc_creation_callback(cur_channel);
1301 }
1302}
1303
1304void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel,
1305 void (*sc_cr_cb)(struct vmbus_channel *new_sc))
1306{
1307 primary_channel->sc_creation_callback = sc_cr_cb;
1308}
1309EXPORT_SYMBOL_GPL(vmbus_set_sc_create_callback);
1310
1311bool vmbus_are_subchannels_present(struct vmbus_channel *primary)
1312{
1313 bool ret;
1314
1315 ret = !list_empty(&primary->sc_list);
1316
1317 if (ret) {
1318 /*
1319 * Invoke the callback on sub-channel creation.
1320 * This will present a uniform interface to the
1321 * clients.
1322 */
1323 invoke_sc_cb(primary);
1324 }
1325
1326 return ret;
1327}
1328EXPORT_SYMBOL_GPL(vmbus_are_subchannels_present);
1329
1330void vmbus_set_chn_rescind_callback(struct vmbus_channel *channel,
1331 void (*chn_rescind_cb)(struct vmbus_channel *))
1332{
1333 channel->chn_rescind_callback = chn_rescind_cb;
1334}
1335EXPORT_SYMBOL_GPL(vmbus_set_chn_rescind_callback);