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 guid_t 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 guid_t *guid)
175{
176 int i;
177
178 for (i = 0; i < ARRAY_SIZE(vmbus_unsupported_devs); i++)
179 if (guid_equal(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 guid_t *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 (guid_equal(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 Negotiate message
202 * @icmsghdrp: Pointer to msg header structure
203 * @buf: Raw buffer channel data
204 * @fw_version: The framework versions we can support.
205 * @fw_vercnt: The size of @fw_version.
206 * @srv_version: The service versions we can support.
207 * @srv_vercnt: The size of @srv_version.
208 * @nego_fw_version: The selected framework version.
209 * @nego_srv_version: The selected service version.
210 *
211 * Note: Versions are given in decreasing order.
212 *
213 * Set up and fill in default negotiate response message.
214 * Mainly used by Hyper-V drivers.
215 */
216bool vmbus_prep_negotiate_resp(struct icmsg_hdr *icmsghdrp,
217 u8 *buf, const int *fw_version, int fw_vercnt,
218 const int *srv_version, int srv_vercnt,
219 int *nego_fw_version, int *nego_srv_version)
220{
221 int icframe_major, icframe_minor;
222 int icmsg_major, icmsg_minor;
223 int fw_major, fw_minor;
224 int srv_major, srv_minor;
225 int i, j;
226 bool found_match = false;
227 struct icmsg_negotiate *negop;
228
229 icmsghdrp->icmsgsize = 0x10;
230 negop = (struct icmsg_negotiate *)&buf[
231 sizeof(struct vmbuspipe_hdr) +
232 sizeof(struct icmsg_hdr)];
233
234 icframe_major = negop->icframe_vercnt;
235 icframe_minor = 0;
236
237 icmsg_major = negop->icmsg_vercnt;
238 icmsg_minor = 0;
239
240 /*
241 * Select the framework version number we will
242 * support.
243 */
244
245 for (i = 0; i < fw_vercnt; i++) {
246 fw_major = (fw_version[i] >> 16);
247 fw_minor = (fw_version[i] & 0xFFFF);
248
249 for (j = 0; j < negop->icframe_vercnt; j++) {
250 if ((negop->icversion_data[j].major == fw_major) &&
251 (negop->icversion_data[j].minor == fw_minor)) {
252 icframe_major = negop->icversion_data[j].major;
253 icframe_minor = negop->icversion_data[j].minor;
254 found_match = true;
255 break;
256 }
257 }
258
259 if (found_match)
260 break;
261 }
262
263 if (!found_match)
264 goto fw_error;
265
266 found_match = false;
267
268 for (i = 0; i < srv_vercnt; i++) {
269 srv_major = (srv_version[i] >> 16);
270 srv_minor = (srv_version[i] & 0xFFFF);
271
272 for (j = negop->icframe_vercnt;
273 (j < negop->icframe_vercnt + negop->icmsg_vercnt);
274 j++) {
275
276 if ((negop->icversion_data[j].major == srv_major) &&
277 (negop->icversion_data[j].minor == srv_minor)) {
278
279 icmsg_major = negop->icversion_data[j].major;
280 icmsg_minor = negop->icversion_data[j].minor;
281 found_match = true;
282 break;
283 }
284 }
285
286 if (found_match)
287 break;
288 }
289
290 /*
291 * Respond with the framework and service
292 * version numbers we can support.
293 */
294
295fw_error:
296 if (!found_match) {
297 negop->icframe_vercnt = 0;
298 negop->icmsg_vercnt = 0;
299 } else {
300 negop->icframe_vercnt = 1;
301 negop->icmsg_vercnt = 1;
302 }
303
304 if (nego_fw_version)
305 *nego_fw_version = (icframe_major << 16) | icframe_minor;
306
307 if (nego_srv_version)
308 *nego_srv_version = (icmsg_major << 16) | icmsg_minor;
309
310 negop->icversion_data[0].major = icframe_major;
311 negop->icversion_data[0].minor = icframe_minor;
312 negop->icversion_data[1].major = icmsg_major;
313 negop->icversion_data[1].minor = icmsg_minor;
314 return found_match;
315}
316
317EXPORT_SYMBOL_GPL(vmbus_prep_negotiate_resp);
318
319/*
320 * alloc_channel - Allocate and initialize a vmbus channel object
321 */
322static struct vmbus_channel *alloc_channel(void)
323{
324 struct vmbus_channel *channel;
325
326 channel = kzalloc(sizeof(*channel), GFP_ATOMIC);
327 if (!channel)
328 return NULL;
329
330 spin_lock_init(&channel->lock);
331 init_completion(&channel->rescind_event);
332
333 INIT_LIST_HEAD(&channel->sc_list);
334 INIT_LIST_HEAD(&channel->percpu_list);
335
336 tasklet_init(&channel->callback_event,
337 vmbus_on_event, (unsigned long)channel);
338
339 return channel;
340}
341
342/*
343 * free_channel - Release the resources used by the vmbus channel object
344 */
345static void free_channel(struct vmbus_channel *channel)
346{
347 tasklet_kill(&channel->callback_event);
348
349 kobject_put(&channel->kobj);
350}
351
352static void percpu_channel_enq(void *arg)
353{
354 struct vmbus_channel *channel = arg;
355 struct hv_per_cpu_context *hv_cpu
356 = this_cpu_ptr(hv_context.cpu_context);
357
358 list_add_tail_rcu(&channel->percpu_list, &hv_cpu->chan_list);
359}
360
361static void percpu_channel_deq(void *arg)
362{
363 struct vmbus_channel *channel = arg;
364
365 list_del_rcu(&channel->percpu_list);
366}
367
368
369static void vmbus_release_relid(u32 relid)
370{
371 struct vmbus_channel_relid_released msg;
372 int ret;
373
374 memset(&msg, 0, sizeof(struct vmbus_channel_relid_released));
375 msg.child_relid = relid;
376 msg.header.msgtype = CHANNELMSG_RELID_RELEASED;
377 ret = vmbus_post_msg(&msg, sizeof(struct vmbus_channel_relid_released),
378 true);
379
380 trace_vmbus_release_relid(&msg, ret);
381}
382
383void hv_process_channel_removal(struct vmbus_channel *channel)
384{
385 struct vmbus_channel *primary_channel;
386 unsigned long flags;
387
388 BUG_ON(!mutex_is_locked(&vmbus_connection.channel_mutex));
389 BUG_ON(!channel->rescind);
390
391 if (channel->target_cpu != get_cpu()) {
392 put_cpu();
393 smp_call_function_single(channel->target_cpu,
394 percpu_channel_deq, channel, true);
395 } else {
396 percpu_channel_deq(channel);
397 put_cpu();
398 }
399
400 if (channel->primary_channel == NULL) {
401 list_del(&channel->listentry);
402
403 primary_channel = channel;
404 } else {
405 primary_channel = channel->primary_channel;
406 spin_lock_irqsave(&primary_channel->lock, flags);
407 list_del(&channel->sc_list);
408 spin_unlock_irqrestore(&primary_channel->lock, flags);
409 }
410
411 /*
412 * We need to free the bit for init_vp_index() to work in the case
413 * of sub-channel, when we reload drivers like hv_netvsc.
414 */
415 if (channel->affinity_policy == HV_LOCALIZED)
416 cpumask_clear_cpu(channel->target_cpu,
417 &primary_channel->alloced_cpus_in_node);
418
419 vmbus_release_relid(channel->offermsg.child_relid);
420
421 free_channel(channel);
422}
423
424void vmbus_free_channels(void)
425{
426 struct vmbus_channel *channel, *tmp;
427
428 list_for_each_entry_safe(channel, tmp, &vmbus_connection.chn_list,
429 listentry) {
430 /* hv_process_channel_removal() needs this */
431 channel->rescind = true;
432
433 vmbus_device_unregister(channel->device_obj);
434 }
435}
436
437/* Note: the function can run concurrently for primary/sub channels. */
438static void vmbus_add_channel_work(struct work_struct *work)
439{
440 struct vmbus_channel *newchannel =
441 container_of(work, struct vmbus_channel, add_channel_work);
442 struct vmbus_channel *primary_channel = newchannel->primary_channel;
443 unsigned long flags;
444 u16 dev_type;
445 int ret;
446
447 dev_type = hv_get_dev_type(newchannel);
448
449 init_vp_index(newchannel, dev_type);
450
451 if (newchannel->target_cpu != get_cpu()) {
452 put_cpu();
453 smp_call_function_single(newchannel->target_cpu,
454 percpu_channel_enq,
455 newchannel, true);
456 } else {
457 percpu_channel_enq(newchannel);
458 put_cpu();
459 }
460
461 /*
462 * This state is used to indicate a successful open
463 * so that when we do close the channel normally, we
464 * can cleanup properly.
465 */
466 newchannel->state = CHANNEL_OPEN_STATE;
467
468 if (primary_channel != NULL) {
469 /* newchannel is a sub-channel. */
470 struct hv_device *dev = primary_channel->device_obj;
471
472 if (vmbus_add_channel_kobj(dev, newchannel))
473 goto err_deq_chan;
474
475 if (primary_channel->sc_creation_callback != NULL)
476 primary_channel->sc_creation_callback(newchannel);
477
478 newchannel->probe_done = true;
479 return;
480 }
481
482 /*
483 * Start the process of binding the primary channel to the driver
484 */
485 newchannel->device_obj = vmbus_device_create(
486 &newchannel->offermsg.offer.if_type,
487 &newchannel->offermsg.offer.if_instance,
488 newchannel);
489 if (!newchannel->device_obj)
490 goto err_deq_chan;
491
492 newchannel->device_obj->device_id = dev_type;
493 /*
494 * Add the new device to the bus. This will kick off device-driver
495 * binding which eventually invokes the device driver's AddDevice()
496 * method.
497 */
498 ret = vmbus_device_register(newchannel->device_obj);
499
500 if (ret != 0) {
501 pr_err("unable to add child device object (relid %d)\n",
502 newchannel->offermsg.child_relid);
503 kfree(newchannel->device_obj);
504 goto err_deq_chan;
505 }
506
507 newchannel->probe_done = true;
508 return;
509
510err_deq_chan:
511 mutex_lock(&vmbus_connection.channel_mutex);
512
513 /*
514 * We need to set the flag, otherwise
515 * vmbus_onoffer_rescind() can be blocked.
516 */
517 newchannel->probe_done = true;
518
519 if (primary_channel == NULL) {
520 list_del(&newchannel->listentry);
521 } else {
522 spin_lock_irqsave(&primary_channel->lock, flags);
523 list_del(&newchannel->sc_list);
524 spin_unlock_irqrestore(&primary_channel->lock, flags);
525 }
526
527 mutex_unlock(&vmbus_connection.channel_mutex);
528
529 if (newchannel->target_cpu != get_cpu()) {
530 put_cpu();
531 smp_call_function_single(newchannel->target_cpu,
532 percpu_channel_deq,
533 newchannel, true);
534 } else {
535 percpu_channel_deq(newchannel);
536 put_cpu();
537 }
538
539 vmbus_release_relid(newchannel->offermsg.child_relid);
540
541 free_channel(newchannel);
542}
543
544/*
545 * vmbus_process_offer - Process the offer by creating a channel/device
546 * associated with this offer
547 */
548static void vmbus_process_offer(struct vmbus_channel *newchannel)
549{
550 struct vmbus_channel *channel;
551 struct workqueue_struct *wq;
552 unsigned long flags;
553 bool fnew = true;
554
555 mutex_lock(&vmbus_connection.channel_mutex);
556
557 /*
558 * Now that we have acquired the channel_mutex,
559 * we can release the potentially racing rescind thread.
560 */
561 atomic_dec(&vmbus_connection.offer_in_progress);
562
563 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
564 if (guid_equal(&channel->offermsg.offer.if_type,
565 &newchannel->offermsg.offer.if_type) &&
566 guid_equal(&channel->offermsg.offer.if_instance,
567 &newchannel->offermsg.offer.if_instance)) {
568 fnew = false;
569 break;
570 }
571 }
572
573 if (fnew)
574 list_add_tail(&newchannel->listentry,
575 &vmbus_connection.chn_list);
576 else {
577 /*
578 * Check to see if this is a valid sub-channel.
579 */
580 if (newchannel->offermsg.offer.sub_channel_index == 0) {
581 mutex_unlock(&vmbus_connection.channel_mutex);
582 /*
583 * Don't call free_channel(), because newchannel->kobj
584 * is not initialized yet.
585 */
586 kfree(newchannel);
587 WARN_ON_ONCE(1);
588 return;
589 }
590 /*
591 * Process the sub-channel.
592 */
593 newchannel->primary_channel = channel;
594 spin_lock_irqsave(&channel->lock, flags);
595 list_add_tail(&newchannel->sc_list, &channel->sc_list);
596 spin_unlock_irqrestore(&channel->lock, flags);
597 }
598
599 mutex_unlock(&vmbus_connection.channel_mutex);
600
601 /*
602 * vmbus_process_offer() mustn't call channel->sc_creation_callback()
603 * directly for sub-channels, because sc_creation_callback() ->
604 * vmbus_open() may never get the host's response to the
605 * OPEN_CHANNEL message (the host may rescind a channel at any time,
606 * e.g. in the case of hot removing a NIC), and vmbus_onoffer_rescind()
607 * may not wake up the vmbus_open() as it's blocked due to a non-zero
608 * vmbus_connection.offer_in_progress, and finally we have a deadlock.
609 *
610 * The above is also true for primary channels, if the related device
611 * drivers use sync probing mode by default.
612 *
613 * And, usually the handling of primary channels and sub-channels can
614 * depend on each other, so we should offload them to different
615 * workqueues to avoid possible deadlock, e.g. in sync-probing mode,
616 * NIC1's netvsc_subchan_work() can race with NIC2's netvsc_probe() ->
617 * rtnl_lock(), and causes deadlock: the former gets the rtnl_lock
618 * and waits for all the sub-channels to appear, but the latter
619 * can't get the rtnl_lock and this blocks the handling of
620 * sub-channels.
621 */
622 INIT_WORK(&newchannel->add_channel_work, vmbus_add_channel_work);
623 wq = fnew ? vmbus_connection.handle_primary_chan_wq :
624 vmbus_connection.handle_sub_chan_wq;
625 queue_work(wq, &newchannel->add_channel_work);
626}
627
628/*
629 * We use this state to statically distribute the channel interrupt load.
630 */
631static int next_numa_node_id;
632/*
633 * init_vp_index() accesses global variables like next_numa_node_id, and
634 * it can run concurrently for primary channels and sub-channels: see
635 * vmbus_process_offer(), so we need the lock to protect the global
636 * variables.
637 */
638static DEFINE_SPINLOCK(bind_channel_to_cpu_lock);
639
640/*
641 * Starting with Win8, we can statically distribute the incoming
642 * channel interrupt load by binding a channel to VCPU.
643 * We distribute the interrupt loads to one or more NUMA nodes based on
644 * the channel's affinity_policy.
645 *
646 * For pre-win8 hosts or non-performance critical channels we assign the
647 * first CPU in the first NUMA node.
648 */
649static void init_vp_index(struct vmbus_channel *channel, u16 dev_type)
650{
651 u32 cur_cpu;
652 bool perf_chn = vmbus_devs[dev_type].perf_device;
653 struct vmbus_channel *primary = channel->primary_channel;
654 int next_node;
655 cpumask_var_t available_mask;
656 struct cpumask *alloced_mask;
657
658 if ((vmbus_proto_version == VERSION_WS2008) ||
659 (vmbus_proto_version == VERSION_WIN7) || (!perf_chn) ||
660 !alloc_cpumask_var(&available_mask, GFP_KERNEL)) {
661 /*
662 * Prior to win8, all channel interrupts are
663 * delivered on cpu 0.
664 * Also if the channel is not a performance critical
665 * channel, bind it to cpu 0.
666 * In case alloc_cpumask_var() fails, bind it to cpu 0.
667 */
668 channel->numa_node = 0;
669 channel->target_cpu = 0;
670 channel->target_vp = hv_cpu_number_to_vp_number(0);
671 return;
672 }
673
674 spin_lock(&bind_channel_to_cpu_lock);
675
676 /*
677 * Based on the channel affinity policy, we will assign the NUMA
678 * nodes.
679 */
680
681 if ((channel->affinity_policy == HV_BALANCED) || (!primary)) {
682 while (true) {
683 next_node = next_numa_node_id++;
684 if (next_node == nr_node_ids) {
685 next_node = next_numa_node_id = 0;
686 continue;
687 }
688 if (cpumask_empty(cpumask_of_node(next_node)))
689 continue;
690 break;
691 }
692 channel->numa_node = next_node;
693 primary = channel;
694 }
695 alloced_mask = &hv_context.hv_numa_map[primary->numa_node];
696
697 if (cpumask_weight(alloced_mask) ==
698 cpumask_weight(cpumask_of_node(primary->numa_node))) {
699 /*
700 * We have cycled through all the CPUs in the node;
701 * reset the alloced map.
702 */
703 cpumask_clear(alloced_mask);
704 }
705
706 cpumask_xor(available_mask, alloced_mask,
707 cpumask_of_node(primary->numa_node));
708
709 cur_cpu = -1;
710
711 if (primary->affinity_policy == HV_LOCALIZED) {
712 /*
713 * Normally Hyper-V host doesn't create more subchannels
714 * than there are VCPUs on the node but it is possible when not
715 * all present VCPUs on the node are initialized by guest.
716 * Clear the alloced_cpus_in_node to start over.
717 */
718 if (cpumask_equal(&primary->alloced_cpus_in_node,
719 cpumask_of_node(primary->numa_node)))
720 cpumask_clear(&primary->alloced_cpus_in_node);
721 }
722
723 while (true) {
724 cur_cpu = cpumask_next(cur_cpu, available_mask);
725 if (cur_cpu >= nr_cpu_ids) {
726 cur_cpu = -1;
727 cpumask_copy(available_mask,
728 cpumask_of_node(primary->numa_node));
729 continue;
730 }
731
732 if (primary->affinity_policy == HV_LOCALIZED) {
733 /*
734 * NOTE: in the case of sub-channel, we clear the
735 * sub-channel related bit(s) in
736 * primary->alloced_cpus_in_node in
737 * hv_process_channel_removal(), so when we
738 * reload drivers like hv_netvsc in SMP guest, here
739 * we're able to re-allocate
740 * bit from primary->alloced_cpus_in_node.
741 */
742 if (!cpumask_test_cpu(cur_cpu,
743 &primary->alloced_cpus_in_node)) {
744 cpumask_set_cpu(cur_cpu,
745 &primary->alloced_cpus_in_node);
746 cpumask_set_cpu(cur_cpu, alloced_mask);
747 break;
748 }
749 } else {
750 cpumask_set_cpu(cur_cpu, alloced_mask);
751 break;
752 }
753 }
754
755 channel->target_cpu = cur_cpu;
756 channel->target_vp = hv_cpu_number_to_vp_number(cur_cpu);
757
758 spin_unlock(&bind_channel_to_cpu_lock);
759
760 free_cpumask_var(available_mask);
761}
762
763static void vmbus_wait_for_unload(void)
764{
765 int cpu;
766 void *page_addr;
767 struct hv_message *msg;
768 struct vmbus_channel_message_header *hdr;
769 u32 message_type;
770
771 /*
772 * CHANNELMSG_UNLOAD_RESPONSE is always delivered to the CPU which was
773 * used for initial contact or to CPU0 depending on host version. When
774 * we're crashing on a different CPU let's hope that IRQ handler on
775 * the cpu which receives CHANNELMSG_UNLOAD_RESPONSE is still
776 * functional and vmbus_unload_response() will complete
777 * vmbus_connection.unload_event. If not, the last thing we can do is
778 * read message pages for all CPUs directly.
779 */
780 while (1) {
781 if (completion_done(&vmbus_connection.unload_event))
782 break;
783
784 for_each_online_cpu(cpu) {
785 struct hv_per_cpu_context *hv_cpu
786 = per_cpu_ptr(hv_context.cpu_context, cpu);
787
788 page_addr = hv_cpu->synic_message_page;
789 msg = (struct hv_message *)page_addr
790 + VMBUS_MESSAGE_SINT;
791
792 message_type = READ_ONCE(msg->header.message_type);
793 if (message_type == HVMSG_NONE)
794 continue;
795
796 hdr = (struct vmbus_channel_message_header *)
797 msg->u.payload;
798
799 if (hdr->msgtype == CHANNELMSG_UNLOAD_RESPONSE)
800 complete(&vmbus_connection.unload_event);
801
802 vmbus_signal_eom(msg, message_type);
803 }
804
805 mdelay(10);
806 }
807
808 /*
809 * We're crashing and already got the UNLOAD_RESPONSE, cleanup all
810 * maybe-pending messages on all CPUs to be able to receive new
811 * messages after we reconnect.
812 */
813 for_each_online_cpu(cpu) {
814 struct hv_per_cpu_context *hv_cpu
815 = per_cpu_ptr(hv_context.cpu_context, cpu);
816
817 page_addr = hv_cpu->synic_message_page;
818 msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
819 msg->header.message_type = HVMSG_NONE;
820 }
821}
822
823/*
824 * vmbus_unload_response - Handler for the unload response.
825 */
826static void vmbus_unload_response(struct vmbus_channel_message_header *hdr)
827{
828 /*
829 * This is a global event; just wakeup the waiting thread.
830 * Once we successfully unload, we can cleanup the monitor state.
831 */
832 complete(&vmbus_connection.unload_event);
833}
834
835void vmbus_initiate_unload(bool crash)
836{
837 struct vmbus_channel_message_header hdr;
838
839 /* Pre-Win2012R2 hosts don't support reconnect */
840 if (vmbus_proto_version < VERSION_WIN8_1)
841 return;
842
843 init_completion(&vmbus_connection.unload_event);
844 memset(&hdr, 0, sizeof(struct vmbus_channel_message_header));
845 hdr.msgtype = CHANNELMSG_UNLOAD;
846 vmbus_post_msg(&hdr, sizeof(struct vmbus_channel_message_header),
847 !crash);
848
849 /*
850 * vmbus_initiate_unload() is also called on crash and the crash can be
851 * happening in an interrupt context, where scheduling is impossible.
852 */
853 if (!crash)
854 wait_for_completion(&vmbus_connection.unload_event);
855 else
856 vmbus_wait_for_unload();
857}
858
859/*
860 * vmbus_onoffer - Handler for channel offers from vmbus in parent partition.
861 *
862 */
863static void vmbus_onoffer(struct vmbus_channel_message_header *hdr)
864{
865 struct vmbus_channel_offer_channel *offer;
866 struct vmbus_channel *newchannel;
867
868 offer = (struct vmbus_channel_offer_channel *)hdr;
869
870 trace_vmbus_onoffer(offer);
871
872 /* Allocate the channel object and save this offer. */
873 newchannel = alloc_channel();
874 if (!newchannel) {
875 vmbus_release_relid(offer->child_relid);
876 atomic_dec(&vmbus_connection.offer_in_progress);
877 pr_err("Unable to allocate channel object\n");
878 return;
879 }
880
881 /*
882 * Setup state for signalling the host.
883 */
884 newchannel->sig_event = VMBUS_EVENT_CONNECTION_ID;
885
886 if (vmbus_proto_version != VERSION_WS2008) {
887 newchannel->is_dedicated_interrupt =
888 (offer->is_dedicated_interrupt != 0);
889 newchannel->sig_event = offer->connection_id;
890 }
891
892 memcpy(&newchannel->offermsg, offer,
893 sizeof(struct vmbus_channel_offer_channel));
894 newchannel->monitor_grp = (u8)offer->monitorid / 32;
895 newchannel->monitor_bit = (u8)offer->monitorid % 32;
896
897 vmbus_process_offer(newchannel);
898}
899
900/*
901 * vmbus_onoffer_rescind - Rescind offer handler.
902 *
903 * We queue a work item to process this offer synchronously
904 */
905static void vmbus_onoffer_rescind(struct vmbus_channel_message_header *hdr)
906{
907 struct vmbus_channel_rescind_offer *rescind;
908 struct vmbus_channel *channel;
909 struct device *dev;
910
911 rescind = (struct vmbus_channel_rescind_offer *)hdr;
912
913 trace_vmbus_onoffer_rescind(rescind);
914
915 /*
916 * The offer msg and the corresponding rescind msg
917 * from the host are guranteed to be ordered -
918 * offer comes in first and then the rescind.
919 * Since we process these events in work elements,
920 * and with preemption, we may end up processing
921 * the events out of order. Given that we handle these
922 * work elements on the same CPU, this is possible only
923 * in the case of preemption. In any case wait here
924 * until the offer processing has moved beyond the
925 * point where the channel is discoverable.
926 */
927
928 while (atomic_read(&vmbus_connection.offer_in_progress) != 0) {
929 /*
930 * We wait here until any channel offer is currently
931 * being processed.
932 */
933 msleep(1);
934 }
935
936 mutex_lock(&vmbus_connection.channel_mutex);
937 channel = relid2channel(rescind->child_relid);
938 mutex_unlock(&vmbus_connection.channel_mutex);
939
940 if (channel == NULL) {
941 /*
942 * We failed in processing the offer message;
943 * we would have cleaned up the relid in that
944 * failure path.
945 */
946 return;
947 }
948
949 /*
950 * Before setting channel->rescind in vmbus_rescind_cleanup(), we
951 * should make sure the channel callback is not running any more.
952 */
953 vmbus_reset_channel_cb(channel);
954
955 /*
956 * Now wait for offer handling to complete.
957 */
958 vmbus_rescind_cleanup(channel);
959 while (READ_ONCE(channel->probe_done) == false) {
960 /*
961 * We wait here until any channel offer is currently
962 * being processed.
963 */
964 msleep(1);
965 }
966
967 /*
968 * At this point, the rescind handling can proceed safely.
969 */
970
971 if (channel->device_obj) {
972 if (channel->chn_rescind_callback) {
973 channel->chn_rescind_callback(channel);
974 return;
975 }
976 /*
977 * We will have to unregister this device from the
978 * driver core.
979 */
980 dev = get_device(&channel->device_obj->device);
981 if (dev) {
982 vmbus_device_unregister(channel->device_obj);
983 put_device(dev);
984 }
985 }
986 if (channel->primary_channel != NULL) {
987 /*
988 * Sub-channel is being rescinded. Following is the channel
989 * close sequence when initiated from the driveri (refer to
990 * vmbus_close() for details):
991 * 1. Close all sub-channels first
992 * 2. Then close the primary channel.
993 */
994 mutex_lock(&vmbus_connection.channel_mutex);
995 if (channel->state == CHANNEL_OPEN_STATE) {
996 /*
997 * The channel is currently not open;
998 * it is safe for us to cleanup the channel.
999 */
1000 hv_process_channel_removal(channel);
1001 } else {
1002 complete(&channel->rescind_event);
1003 }
1004 mutex_unlock(&vmbus_connection.channel_mutex);
1005 }
1006}
1007
1008void vmbus_hvsock_device_unregister(struct vmbus_channel *channel)
1009{
1010 BUG_ON(!is_hvsock_channel(channel));
1011
1012 /* We always get a rescind msg when a connection is closed. */
1013 while (!READ_ONCE(channel->probe_done) || !READ_ONCE(channel->rescind))
1014 msleep(1);
1015
1016 vmbus_device_unregister(channel->device_obj);
1017}
1018EXPORT_SYMBOL_GPL(vmbus_hvsock_device_unregister);
1019
1020
1021/*
1022 * vmbus_onoffers_delivered -
1023 * This is invoked when all offers have been delivered.
1024 *
1025 * Nothing to do here.
1026 */
1027static void vmbus_onoffers_delivered(
1028 struct vmbus_channel_message_header *hdr)
1029{
1030}
1031
1032/*
1033 * vmbus_onopen_result - Open result handler.
1034 *
1035 * This is invoked when we received a response to our channel open request.
1036 * Find the matching request, copy the response and signal the requesting
1037 * thread.
1038 */
1039static void vmbus_onopen_result(struct vmbus_channel_message_header *hdr)
1040{
1041 struct vmbus_channel_open_result *result;
1042 struct vmbus_channel_msginfo *msginfo;
1043 struct vmbus_channel_message_header *requestheader;
1044 struct vmbus_channel_open_channel *openmsg;
1045 unsigned long flags;
1046
1047 result = (struct vmbus_channel_open_result *)hdr;
1048
1049 trace_vmbus_onopen_result(result);
1050
1051 /*
1052 * Find the open msg, copy the result and signal/unblock the wait event
1053 */
1054 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1055
1056 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1057 msglistentry) {
1058 requestheader =
1059 (struct vmbus_channel_message_header *)msginfo->msg;
1060
1061 if (requestheader->msgtype == CHANNELMSG_OPENCHANNEL) {
1062 openmsg =
1063 (struct vmbus_channel_open_channel *)msginfo->msg;
1064 if (openmsg->child_relid == result->child_relid &&
1065 openmsg->openid == result->openid) {
1066 memcpy(&msginfo->response.open_result,
1067 result,
1068 sizeof(
1069 struct vmbus_channel_open_result));
1070 complete(&msginfo->waitevent);
1071 break;
1072 }
1073 }
1074 }
1075 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1076}
1077
1078/*
1079 * vmbus_ongpadl_created - GPADL created handler.
1080 *
1081 * This is invoked when we received a response to our gpadl create request.
1082 * Find the matching request, copy the response and signal the requesting
1083 * thread.
1084 */
1085static void vmbus_ongpadl_created(struct vmbus_channel_message_header *hdr)
1086{
1087 struct vmbus_channel_gpadl_created *gpadlcreated;
1088 struct vmbus_channel_msginfo *msginfo;
1089 struct vmbus_channel_message_header *requestheader;
1090 struct vmbus_channel_gpadl_header *gpadlheader;
1091 unsigned long flags;
1092
1093 gpadlcreated = (struct vmbus_channel_gpadl_created *)hdr;
1094
1095 trace_vmbus_ongpadl_created(gpadlcreated);
1096
1097 /*
1098 * Find the establish msg, copy the result and signal/unblock the wait
1099 * event
1100 */
1101 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1102
1103 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1104 msglistentry) {
1105 requestheader =
1106 (struct vmbus_channel_message_header *)msginfo->msg;
1107
1108 if (requestheader->msgtype == CHANNELMSG_GPADL_HEADER) {
1109 gpadlheader =
1110 (struct vmbus_channel_gpadl_header *)requestheader;
1111
1112 if ((gpadlcreated->child_relid ==
1113 gpadlheader->child_relid) &&
1114 (gpadlcreated->gpadl == gpadlheader->gpadl)) {
1115 memcpy(&msginfo->response.gpadl_created,
1116 gpadlcreated,
1117 sizeof(
1118 struct vmbus_channel_gpadl_created));
1119 complete(&msginfo->waitevent);
1120 break;
1121 }
1122 }
1123 }
1124 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1125}
1126
1127/*
1128 * vmbus_ongpadl_torndown - GPADL torndown handler.
1129 *
1130 * This is invoked when we received a response to our gpadl teardown request.
1131 * Find the matching request, copy the response and signal the requesting
1132 * thread.
1133 */
1134static void vmbus_ongpadl_torndown(
1135 struct vmbus_channel_message_header *hdr)
1136{
1137 struct vmbus_channel_gpadl_torndown *gpadl_torndown;
1138 struct vmbus_channel_msginfo *msginfo;
1139 struct vmbus_channel_message_header *requestheader;
1140 struct vmbus_channel_gpadl_teardown *gpadl_teardown;
1141 unsigned long flags;
1142
1143 gpadl_torndown = (struct vmbus_channel_gpadl_torndown *)hdr;
1144
1145 trace_vmbus_ongpadl_torndown(gpadl_torndown);
1146
1147 /*
1148 * Find the open msg, copy the result and signal/unblock the wait event
1149 */
1150 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1151
1152 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1153 msglistentry) {
1154 requestheader =
1155 (struct vmbus_channel_message_header *)msginfo->msg;
1156
1157 if (requestheader->msgtype == CHANNELMSG_GPADL_TEARDOWN) {
1158 gpadl_teardown =
1159 (struct vmbus_channel_gpadl_teardown *)requestheader;
1160
1161 if (gpadl_torndown->gpadl == gpadl_teardown->gpadl) {
1162 memcpy(&msginfo->response.gpadl_torndown,
1163 gpadl_torndown,
1164 sizeof(
1165 struct vmbus_channel_gpadl_torndown));
1166 complete(&msginfo->waitevent);
1167 break;
1168 }
1169 }
1170 }
1171 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1172}
1173
1174/*
1175 * vmbus_onversion_response - Version response handler
1176 *
1177 * This is invoked when we received a response to our initiate contact request.
1178 * Find the matching request, copy the response and signal the requesting
1179 * thread.
1180 */
1181static void vmbus_onversion_response(
1182 struct vmbus_channel_message_header *hdr)
1183{
1184 struct vmbus_channel_msginfo *msginfo;
1185 struct vmbus_channel_message_header *requestheader;
1186 struct vmbus_channel_version_response *version_response;
1187 unsigned long flags;
1188
1189 version_response = (struct vmbus_channel_version_response *)hdr;
1190
1191 trace_vmbus_onversion_response(version_response);
1192
1193 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
1194
1195 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
1196 msglistentry) {
1197 requestheader =
1198 (struct vmbus_channel_message_header *)msginfo->msg;
1199
1200 if (requestheader->msgtype ==
1201 CHANNELMSG_INITIATE_CONTACT) {
1202 memcpy(&msginfo->response.version_response,
1203 version_response,
1204 sizeof(struct vmbus_channel_version_response));
1205 complete(&msginfo->waitevent);
1206 }
1207 }
1208 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
1209}
1210
1211/* Channel message dispatch table */
1212const struct vmbus_channel_message_table_entry
1213channel_message_table[CHANNELMSG_COUNT] = {
1214 { CHANNELMSG_INVALID, 0, NULL },
1215 { CHANNELMSG_OFFERCHANNEL, 0, vmbus_onoffer },
1216 { CHANNELMSG_RESCIND_CHANNELOFFER, 0, vmbus_onoffer_rescind },
1217 { CHANNELMSG_REQUESTOFFERS, 0, NULL },
1218 { CHANNELMSG_ALLOFFERS_DELIVERED, 1, vmbus_onoffers_delivered },
1219 { CHANNELMSG_OPENCHANNEL, 0, NULL },
1220 { CHANNELMSG_OPENCHANNEL_RESULT, 1, vmbus_onopen_result },
1221 { CHANNELMSG_CLOSECHANNEL, 0, NULL },
1222 { CHANNELMSG_GPADL_HEADER, 0, NULL },
1223 { CHANNELMSG_GPADL_BODY, 0, NULL },
1224 { CHANNELMSG_GPADL_CREATED, 1, vmbus_ongpadl_created },
1225 { CHANNELMSG_GPADL_TEARDOWN, 0, NULL },
1226 { CHANNELMSG_GPADL_TORNDOWN, 1, vmbus_ongpadl_torndown },
1227 { CHANNELMSG_RELID_RELEASED, 0, NULL },
1228 { CHANNELMSG_INITIATE_CONTACT, 0, NULL },
1229 { CHANNELMSG_VERSION_RESPONSE, 1, vmbus_onversion_response },
1230 { CHANNELMSG_UNLOAD, 0, NULL },
1231 { CHANNELMSG_UNLOAD_RESPONSE, 1, vmbus_unload_response },
1232 { CHANNELMSG_18, 0, NULL },
1233 { CHANNELMSG_19, 0, NULL },
1234 { CHANNELMSG_20, 0, NULL },
1235 { CHANNELMSG_TL_CONNECT_REQUEST, 0, NULL },
1236};
1237
1238/*
1239 * vmbus_onmessage - Handler for channel protocol messages.
1240 *
1241 * This is invoked in the vmbus worker thread context.
1242 */
1243void vmbus_onmessage(void *context)
1244{
1245 struct hv_message *msg = context;
1246 struct vmbus_channel_message_header *hdr;
1247 int size;
1248
1249 hdr = (struct vmbus_channel_message_header *)msg->u.payload;
1250 size = msg->header.payload_size;
1251
1252 trace_vmbus_on_message(hdr);
1253
1254 if (hdr->msgtype >= CHANNELMSG_COUNT) {
1255 pr_err("Received invalid channel message type %d size %d\n",
1256 hdr->msgtype, size);
1257 print_hex_dump_bytes("", DUMP_PREFIX_NONE,
1258 (unsigned char *)msg->u.payload, size);
1259 return;
1260 }
1261
1262 if (channel_message_table[hdr->msgtype].message_handler)
1263 channel_message_table[hdr->msgtype].message_handler(hdr);
1264 else
1265 pr_err("Unhandled channel message type %d\n", hdr->msgtype);
1266}
1267
1268/*
1269 * vmbus_request_offers - Send a request to get all our pending offers.
1270 */
1271int vmbus_request_offers(void)
1272{
1273 struct vmbus_channel_message_header *msg;
1274 struct vmbus_channel_msginfo *msginfo;
1275 int ret;
1276
1277 msginfo = kmalloc(sizeof(*msginfo) +
1278 sizeof(struct vmbus_channel_message_header),
1279 GFP_KERNEL);
1280 if (!msginfo)
1281 return -ENOMEM;
1282
1283 msg = (struct vmbus_channel_message_header *)msginfo->msg;
1284
1285 msg->msgtype = CHANNELMSG_REQUESTOFFERS;
1286
1287 ret = vmbus_post_msg(msg, sizeof(struct vmbus_channel_message_header),
1288 true);
1289
1290 trace_vmbus_request_offers(ret);
1291
1292 if (ret != 0) {
1293 pr_err("Unable to request offers - %d\n", ret);
1294
1295 goto cleanup;
1296 }
1297
1298cleanup:
1299 kfree(msginfo);
1300
1301 return ret;
1302}
1303
1304static void invoke_sc_cb(struct vmbus_channel *primary_channel)
1305{
1306 struct list_head *cur, *tmp;
1307 struct vmbus_channel *cur_channel;
1308
1309 if (primary_channel->sc_creation_callback == NULL)
1310 return;
1311
1312 list_for_each_safe(cur, tmp, &primary_channel->sc_list) {
1313 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
1314
1315 primary_channel->sc_creation_callback(cur_channel);
1316 }
1317}
1318
1319void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel,
1320 void (*sc_cr_cb)(struct vmbus_channel *new_sc))
1321{
1322 primary_channel->sc_creation_callback = sc_cr_cb;
1323}
1324EXPORT_SYMBOL_GPL(vmbus_set_sc_create_callback);
1325
1326bool vmbus_are_subchannels_present(struct vmbus_channel *primary)
1327{
1328 bool ret;
1329
1330 ret = !list_empty(&primary->sc_list);
1331
1332 if (ret) {
1333 /*
1334 * Invoke the callback on sub-channel creation.
1335 * This will present a uniform interface to the
1336 * clients.
1337 */
1338 invoke_sc_cb(primary);
1339 }
1340
1341 return ret;
1342}
1343EXPORT_SYMBOL_GPL(vmbus_are_subchannels_present);
1344
1345void vmbus_set_chn_rescind_callback(struct vmbus_channel *channel,
1346 void (*chn_rescind_cb)(struct vmbus_channel *))
1347{
1348 channel->chn_rescind_callback = chn_rescind_cb;
1349}
1350EXPORT_SYMBOL_GPL(vmbus_set_chn_rescind_callback);
1351