Drivers: hv: vmbus: avoid wait_for_completion() on crash
[cascardo/linux.git] / drivers / hv / vmbus_drv.c
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  *   K. Y. Srinivasan <kys@microsoft.com>
21  *
22  */
23 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
24
25 #include <linux/init.h>
26 #include <linux/module.h>
27 #include <linux/device.h>
28 #include <linux/interrupt.h>
29 #include <linux/sysctl.h>
30 #include <linux/slab.h>
31 #include <linux/acpi.h>
32 #include <linux/completion.h>
33 #include <linux/hyperv.h>
34 #include <linux/kernel_stat.h>
35 #include <linux/clockchips.h>
36 #include <linux/cpu.h>
37 #include <asm/hyperv.h>
38 #include <asm/hypervisor.h>
39 #include <asm/mshyperv.h>
40 #include <linux/notifier.h>
41 #include <linux/ptrace.h>
42 #include <linux/screen_info.h>
43 #include <linux/kdebug.h>
44 #include "hyperv_vmbus.h"
45
46 static struct acpi_device  *hv_acpi_dev;
47
48 static struct tasklet_struct msg_dpc;
49 static struct completion probe_event;
50
51
52 static void hyperv_report_panic(struct pt_regs *regs)
53 {
54         static bool panic_reported;
55
56         /*
57          * We prefer to report panic on 'die' chain as we have proper
58          * registers to report, but if we miss it (e.g. on BUG()) we need
59          * to report it on 'panic'.
60          */
61         if (panic_reported)
62                 return;
63         panic_reported = true;
64
65         wrmsrl(HV_X64_MSR_CRASH_P0, regs->ip);
66         wrmsrl(HV_X64_MSR_CRASH_P1, regs->ax);
67         wrmsrl(HV_X64_MSR_CRASH_P2, regs->bx);
68         wrmsrl(HV_X64_MSR_CRASH_P3, regs->cx);
69         wrmsrl(HV_X64_MSR_CRASH_P4, regs->dx);
70
71         /*
72          * Let Hyper-V know there is crash data available
73          */
74         wrmsrl(HV_X64_MSR_CRASH_CTL, HV_CRASH_CTL_CRASH_NOTIFY);
75 }
76
77 static int hyperv_panic_event(struct notifier_block *nb, unsigned long val,
78                               void *args)
79 {
80         struct pt_regs *regs;
81
82         regs = current_pt_regs();
83
84         hyperv_report_panic(regs);
85         return NOTIFY_DONE;
86 }
87
88 static int hyperv_die_event(struct notifier_block *nb, unsigned long val,
89                             void *args)
90 {
91         struct die_args *die = (struct die_args *)args;
92         struct pt_regs *regs = die->regs;
93
94         hyperv_report_panic(regs);
95         return NOTIFY_DONE;
96 }
97
98 static struct notifier_block hyperv_die_block = {
99         .notifier_call = hyperv_die_event,
100 };
101 static struct notifier_block hyperv_panic_block = {
102         .notifier_call = hyperv_panic_event,
103 };
104
105 struct resource *hyperv_mmio;
106
107 static int vmbus_exists(void)
108 {
109         if (hv_acpi_dev == NULL)
110                 return -ENODEV;
111
112         return 0;
113 }
114
115 #define VMBUS_ALIAS_LEN ((sizeof((struct hv_vmbus_device_id *)0)->guid) * 2)
116 static void print_alias_name(struct hv_device *hv_dev, char *alias_name)
117 {
118         int i;
119         for (i = 0; i < VMBUS_ALIAS_LEN; i += 2)
120                 sprintf(&alias_name[i], "%02x", hv_dev->dev_type.b[i/2]);
121 }
122
123 static u8 channel_monitor_group(struct vmbus_channel *channel)
124 {
125         return (u8)channel->offermsg.monitorid / 32;
126 }
127
128 static u8 channel_monitor_offset(struct vmbus_channel *channel)
129 {
130         return (u8)channel->offermsg.monitorid % 32;
131 }
132
133 static u32 channel_pending(struct vmbus_channel *channel,
134                            struct hv_monitor_page *monitor_page)
135 {
136         u8 monitor_group = channel_monitor_group(channel);
137         return monitor_page->trigger_group[monitor_group].pending;
138 }
139
140 static u32 channel_latency(struct vmbus_channel *channel,
141                            struct hv_monitor_page *monitor_page)
142 {
143         u8 monitor_group = channel_monitor_group(channel);
144         u8 monitor_offset = channel_monitor_offset(channel);
145         return monitor_page->latency[monitor_group][monitor_offset];
146 }
147
148 static u32 channel_conn_id(struct vmbus_channel *channel,
149                            struct hv_monitor_page *monitor_page)
150 {
151         u8 monitor_group = channel_monitor_group(channel);
152         u8 monitor_offset = channel_monitor_offset(channel);
153         return monitor_page->parameter[monitor_group][monitor_offset].connectionid.u.id;
154 }
155
156 static ssize_t id_show(struct device *dev, struct device_attribute *dev_attr,
157                        char *buf)
158 {
159         struct hv_device *hv_dev = device_to_hv_device(dev);
160
161         if (!hv_dev->channel)
162                 return -ENODEV;
163         return sprintf(buf, "%d\n", hv_dev->channel->offermsg.child_relid);
164 }
165 static DEVICE_ATTR_RO(id);
166
167 static ssize_t state_show(struct device *dev, struct device_attribute *dev_attr,
168                           char *buf)
169 {
170         struct hv_device *hv_dev = device_to_hv_device(dev);
171
172         if (!hv_dev->channel)
173                 return -ENODEV;
174         return sprintf(buf, "%d\n", hv_dev->channel->state);
175 }
176 static DEVICE_ATTR_RO(state);
177
178 static ssize_t monitor_id_show(struct device *dev,
179                                struct device_attribute *dev_attr, char *buf)
180 {
181         struct hv_device *hv_dev = device_to_hv_device(dev);
182
183         if (!hv_dev->channel)
184                 return -ENODEV;
185         return sprintf(buf, "%d\n", hv_dev->channel->offermsg.monitorid);
186 }
187 static DEVICE_ATTR_RO(monitor_id);
188
189 static ssize_t class_id_show(struct device *dev,
190                                struct device_attribute *dev_attr, char *buf)
191 {
192         struct hv_device *hv_dev = device_to_hv_device(dev);
193
194         if (!hv_dev->channel)
195                 return -ENODEV;
196         return sprintf(buf, "{%pUl}\n",
197                        hv_dev->channel->offermsg.offer.if_type.b);
198 }
199 static DEVICE_ATTR_RO(class_id);
200
201 static ssize_t device_id_show(struct device *dev,
202                               struct device_attribute *dev_attr, char *buf)
203 {
204         struct hv_device *hv_dev = device_to_hv_device(dev);
205
206         if (!hv_dev->channel)
207                 return -ENODEV;
208         return sprintf(buf, "{%pUl}\n",
209                        hv_dev->channel->offermsg.offer.if_instance.b);
210 }
211 static DEVICE_ATTR_RO(device_id);
212
213 static ssize_t modalias_show(struct device *dev,
214                              struct device_attribute *dev_attr, char *buf)
215 {
216         struct hv_device *hv_dev = device_to_hv_device(dev);
217         char alias_name[VMBUS_ALIAS_LEN + 1];
218
219         print_alias_name(hv_dev, alias_name);
220         return sprintf(buf, "vmbus:%s\n", alias_name);
221 }
222 static DEVICE_ATTR_RO(modalias);
223
224 static ssize_t server_monitor_pending_show(struct device *dev,
225                                            struct device_attribute *dev_attr,
226                                            char *buf)
227 {
228         struct hv_device *hv_dev = device_to_hv_device(dev);
229
230         if (!hv_dev->channel)
231                 return -ENODEV;
232         return sprintf(buf, "%d\n",
233                        channel_pending(hv_dev->channel,
234                                        vmbus_connection.monitor_pages[1]));
235 }
236 static DEVICE_ATTR_RO(server_monitor_pending);
237
238 static ssize_t client_monitor_pending_show(struct device *dev,
239                                            struct device_attribute *dev_attr,
240                                            char *buf)
241 {
242         struct hv_device *hv_dev = device_to_hv_device(dev);
243
244         if (!hv_dev->channel)
245                 return -ENODEV;
246         return sprintf(buf, "%d\n",
247                        channel_pending(hv_dev->channel,
248                                        vmbus_connection.monitor_pages[1]));
249 }
250 static DEVICE_ATTR_RO(client_monitor_pending);
251
252 static ssize_t server_monitor_latency_show(struct device *dev,
253                                            struct device_attribute *dev_attr,
254                                            char *buf)
255 {
256         struct hv_device *hv_dev = device_to_hv_device(dev);
257
258         if (!hv_dev->channel)
259                 return -ENODEV;
260         return sprintf(buf, "%d\n",
261                        channel_latency(hv_dev->channel,
262                                        vmbus_connection.monitor_pages[0]));
263 }
264 static DEVICE_ATTR_RO(server_monitor_latency);
265
266 static ssize_t client_monitor_latency_show(struct device *dev,
267                                            struct device_attribute *dev_attr,
268                                            char *buf)
269 {
270         struct hv_device *hv_dev = device_to_hv_device(dev);
271
272         if (!hv_dev->channel)
273                 return -ENODEV;
274         return sprintf(buf, "%d\n",
275                        channel_latency(hv_dev->channel,
276                                        vmbus_connection.monitor_pages[1]));
277 }
278 static DEVICE_ATTR_RO(client_monitor_latency);
279
280 static ssize_t server_monitor_conn_id_show(struct device *dev,
281                                            struct device_attribute *dev_attr,
282                                            char *buf)
283 {
284         struct hv_device *hv_dev = device_to_hv_device(dev);
285
286         if (!hv_dev->channel)
287                 return -ENODEV;
288         return sprintf(buf, "%d\n",
289                        channel_conn_id(hv_dev->channel,
290                                        vmbus_connection.monitor_pages[0]));
291 }
292 static DEVICE_ATTR_RO(server_monitor_conn_id);
293
294 static ssize_t client_monitor_conn_id_show(struct device *dev,
295                                            struct device_attribute *dev_attr,
296                                            char *buf)
297 {
298         struct hv_device *hv_dev = device_to_hv_device(dev);
299
300         if (!hv_dev->channel)
301                 return -ENODEV;
302         return sprintf(buf, "%d\n",
303                        channel_conn_id(hv_dev->channel,
304                                        vmbus_connection.monitor_pages[1]));
305 }
306 static DEVICE_ATTR_RO(client_monitor_conn_id);
307
308 static ssize_t out_intr_mask_show(struct device *dev,
309                                   struct device_attribute *dev_attr, char *buf)
310 {
311         struct hv_device *hv_dev = device_to_hv_device(dev);
312         struct hv_ring_buffer_debug_info outbound;
313
314         if (!hv_dev->channel)
315                 return -ENODEV;
316         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
317         return sprintf(buf, "%d\n", outbound.current_interrupt_mask);
318 }
319 static DEVICE_ATTR_RO(out_intr_mask);
320
321 static ssize_t out_read_index_show(struct device *dev,
322                                    struct device_attribute *dev_attr, char *buf)
323 {
324         struct hv_device *hv_dev = device_to_hv_device(dev);
325         struct hv_ring_buffer_debug_info outbound;
326
327         if (!hv_dev->channel)
328                 return -ENODEV;
329         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
330         return sprintf(buf, "%d\n", outbound.current_read_index);
331 }
332 static DEVICE_ATTR_RO(out_read_index);
333
334 static ssize_t out_write_index_show(struct device *dev,
335                                     struct device_attribute *dev_attr,
336                                     char *buf)
337 {
338         struct hv_device *hv_dev = device_to_hv_device(dev);
339         struct hv_ring_buffer_debug_info outbound;
340
341         if (!hv_dev->channel)
342                 return -ENODEV;
343         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
344         return sprintf(buf, "%d\n", outbound.current_write_index);
345 }
346 static DEVICE_ATTR_RO(out_write_index);
347
348 static ssize_t out_read_bytes_avail_show(struct device *dev,
349                                          struct device_attribute *dev_attr,
350                                          char *buf)
351 {
352         struct hv_device *hv_dev = device_to_hv_device(dev);
353         struct hv_ring_buffer_debug_info outbound;
354
355         if (!hv_dev->channel)
356                 return -ENODEV;
357         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
358         return sprintf(buf, "%d\n", outbound.bytes_avail_toread);
359 }
360 static DEVICE_ATTR_RO(out_read_bytes_avail);
361
362 static ssize_t out_write_bytes_avail_show(struct device *dev,
363                                           struct device_attribute *dev_attr,
364                                           char *buf)
365 {
366         struct hv_device *hv_dev = device_to_hv_device(dev);
367         struct hv_ring_buffer_debug_info outbound;
368
369         if (!hv_dev->channel)
370                 return -ENODEV;
371         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
372         return sprintf(buf, "%d\n", outbound.bytes_avail_towrite);
373 }
374 static DEVICE_ATTR_RO(out_write_bytes_avail);
375
376 static ssize_t in_intr_mask_show(struct device *dev,
377                                  struct device_attribute *dev_attr, char *buf)
378 {
379         struct hv_device *hv_dev = device_to_hv_device(dev);
380         struct hv_ring_buffer_debug_info inbound;
381
382         if (!hv_dev->channel)
383                 return -ENODEV;
384         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
385         return sprintf(buf, "%d\n", inbound.current_interrupt_mask);
386 }
387 static DEVICE_ATTR_RO(in_intr_mask);
388
389 static ssize_t in_read_index_show(struct device *dev,
390                                   struct device_attribute *dev_attr, char *buf)
391 {
392         struct hv_device *hv_dev = device_to_hv_device(dev);
393         struct hv_ring_buffer_debug_info inbound;
394
395         if (!hv_dev->channel)
396                 return -ENODEV;
397         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
398         return sprintf(buf, "%d\n", inbound.current_read_index);
399 }
400 static DEVICE_ATTR_RO(in_read_index);
401
402 static ssize_t in_write_index_show(struct device *dev,
403                                    struct device_attribute *dev_attr, char *buf)
404 {
405         struct hv_device *hv_dev = device_to_hv_device(dev);
406         struct hv_ring_buffer_debug_info inbound;
407
408         if (!hv_dev->channel)
409                 return -ENODEV;
410         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
411         return sprintf(buf, "%d\n", inbound.current_write_index);
412 }
413 static DEVICE_ATTR_RO(in_write_index);
414
415 static ssize_t in_read_bytes_avail_show(struct device *dev,
416                                         struct device_attribute *dev_attr,
417                                         char *buf)
418 {
419         struct hv_device *hv_dev = device_to_hv_device(dev);
420         struct hv_ring_buffer_debug_info inbound;
421
422         if (!hv_dev->channel)
423                 return -ENODEV;
424         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
425         return sprintf(buf, "%d\n", inbound.bytes_avail_toread);
426 }
427 static DEVICE_ATTR_RO(in_read_bytes_avail);
428
429 static ssize_t in_write_bytes_avail_show(struct device *dev,
430                                          struct device_attribute *dev_attr,
431                                          char *buf)
432 {
433         struct hv_device *hv_dev = device_to_hv_device(dev);
434         struct hv_ring_buffer_debug_info inbound;
435
436         if (!hv_dev->channel)
437                 return -ENODEV;
438         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
439         return sprintf(buf, "%d\n", inbound.bytes_avail_towrite);
440 }
441 static DEVICE_ATTR_RO(in_write_bytes_avail);
442
443 static ssize_t channel_vp_mapping_show(struct device *dev,
444                                        struct device_attribute *dev_attr,
445                                        char *buf)
446 {
447         struct hv_device *hv_dev = device_to_hv_device(dev);
448         struct vmbus_channel *channel = hv_dev->channel, *cur_sc;
449         unsigned long flags;
450         int buf_size = PAGE_SIZE, n_written, tot_written;
451         struct list_head *cur;
452
453         if (!channel)
454                 return -ENODEV;
455
456         tot_written = snprintf(buf, buf_size, "%u:%u\n",
457                 channel->offermsg.child_relid, channel->target_cpu);
458
459         spin_lock_irqsave(&channel->lock, flags);
460
461         list_for_each(cur, &channel->sc_list) {
462                 if (tot_written >= buf_size - 1)
463                         break;
464
465                 cur_sc = list_entry(cur, struct vmbus_channel, sc_list);
466                 n_written = scnprintf(buf + tot_written,
467                                      buf_size - tot_written,
468                                      "%u:%u\n",
469                                      cur_sc->offermsg.child_relid,
470                                      cur_sc->target_cpu);
471                 tot_written += n_written;
472         }
473
474         spin_unlock_irqrestore(&channel->lock, flags);
475
476         return tot_written;
477 }
478 static DEVICE_ATTR_RO(channel_vp_mapping);
479
480 static ssize_t vendor_show(struct device *dev,
481                            struct device_attribute *dev_attr,
482                            char *buf)
483 {
484         struct hv_device *hv_dev = device_to_hv_device(dev);
485         return sprintf(buf, "0x%x\n", hv_dev->vendor_id);
486 }
487 static DEVICE_ATTR_RO(vendor);
488
489 static ssize_t device_show(struct device *dev,
490                            struct device_attribute *dev_attr,
491                            char *buf)
492 {
493         struct hv_device *hv_dev = device_to_hv_device(dev);
494         return sprintf(buf, "0x%x\n", hv_dev->device_id);
495 }
496 static DEVICE_ATTR_RO(device);
497
498 /* Set up per device attributes in /sys/bus/vmbus/devices/<bus device> */
499 static struct attribute *vmbus_attrs[] = {
500         &dev_attr_id.attr,
501         &dev_attr_state.attr,
502         &dev_attr_monitor_id.attr,
503         &dev_attr_class_id.attr,
504         &dev_attr_device_id.attr,
505         &dev_attr_modalias.attr,
506         &dev_attr_server_monitor_pending.attr,
507         &dev_attr_client_monitor_pending.attr,
508         &dev_attr_server_monitor_latency.attr,
509         &dev_attr_client_monitor_latency.attr,
510         &dev_attr_server_monitor_conn_id.attr,
511         &dev_attr_client_monitor_conn_id.attr,
512         &dev_attr_out_intr_mask.attr,
513         &dev_attr_out_read_index.attr,
514         &dev_attr_out_write_index.attr,
515         &dev_attr_out_read_bytes_avail.attr,
516         &dev_attr_out_write_bytes_avail.attr,
517         &dev_attr_in_intr_mask.attr,
518         &dev_attr_in_read_index.attr,
519         &dev_attr_in_write_index.attr,
520         &dev_attr_in_read_bytes_avail.attr,
521         &dev_attr_in_write_bytes_avail.attr,
522         &dev_attr_channel_vp_mapping.attr,
523         &dev_attr_vendor.attr,
524         &dev_attr_device.attr,
525         NULL,
526 };
527 ATTRIBUTE_GROUPS(vmbus);
528
529 /*
530  * vmbus_uevent - add uevent for our device
531  *
532  * This routine is invoked when a device is added or removed on the vmbus to
533  * generate a uevent to udev in the userspace. The udev will then look at its
534  * rule and the uevent generated here to load the appropriate driver
535  *
536  * The alias string will be of the form vmbus:guid where guid is the string
537  * representation of the device guid (each byte of the guid will be
538  * represented with two hex characters.
539  */
540 static int vmbus_uevent(struct device *device, struct kobj_uevent_env *env)
541 {
542         struct hv_device *dev = device_to_hv_device(device);
543         int ret;
544         char alias_name[VMBUS_ALIAS_LEN + 1];
545
546         print_alias_name(dev, alias_name);
547         ret = add_uevent_var(env, "MODALIAS=vmbus:%s", alias_name);
548         return ret;
549 }
550
551 static const uuid_le null_guid;
552
553 static inline bool is_null_guid(const uuid_le *guid)
554 {
555         if (uuid_le_cmp(*guid, null_guid))
556                 return false;
557         return true;
558 }
559
560 /*
561  * Return a matching hv_vmbus_device_id pointer.
562  * If there is no match, return NULL.
563  */
564 static const struct hv_vmbus_device_id *hv_vmbus_get_id(
565                                         const struct hv_vmbus_device_id *id,
566                                         const uuid_le *guid)
567 {
568         for (; !is_null_guid(&id->guid); id++)
569                 if (!uuid_le_cmp(id->guid, *guid))
570                         return id;
571
572         return NULL;
573 }
574
575
576
577 /*
578  * vmbus_match - Attempt to match the specified device to the specified driver
579  */
580 static int vmbus_match(struct device *device, struct device_driver *driver)
581 {
582         struct hv_driver *drv = drv_to_hv_drv(driver);
583         struct hv_device *hv_dev = device_to_hv_device(device);
584
585         /* The hv_sock driver handles all hv_sock offers. */
586         if (is_hvsock_channel(hv_dev->channel))
587                 return drv->hvsock;
588
589         if (hv_vmbus_get_id(drv->id_table, &hv_dev->dev_type))
590                 return 1;
591
592         return 0;
593 }
594
595 /*
596  * vmbus_probe - Add the new vmbus's child device
597  */
598 static int vmbus_probe(struct device *child_device)
599 {
600         int ret = 0;
601         struct hv_driver *drv =
602                         drv_to_hv_drv(child_device->driver);
603         struct hv_device *dev = device_to_hv_device(child_device);
604         const struct hv_vmbus_device_id *dev_id;
605
606         dev_id = hv_vmbus_get_id(drv->id_table, &dev->dev_type);
607         if (drv->probe) {
608                 ret = drv->probe(dev, dev_id);
609                 if (ret != 0)
610                         pr_err("probe failed for device %s (%d)\n",
611                                dev_name(child_device), ret);
612
613         } else {
614                 pr_err("probe not set for driver %s\n",
615                        dev_name(child_device));
616                 ret = -ENODEV;
617         }
618         return ret;
619 }
620
621 /*
622  * vmbus_remove - Remove a vmbus device
623  */
624 static int vmbus_remove(struct device *child_device)
625 {
626         struct hv_driver *drv;
627         struct hv_device *dev = device_to_hv_device(child_device);
628
629         if (child_device->driver) {
630                 drv = drv_to_hv_drv(child_device->driver);
631                 if (drv->remove)
632                         drv->remove(dev);
633         }
634
635         return 0;
636 }
637
638
639 /*
640  * vmbus_shutdown - Shutdown a vmbus device
641  */
642 static void vmbus_shutdown(struct device *child_device)
643 {
644         struct hv_driver *drv;
645         struct hv_device *dev = device_to_hv_device(child_device);
646
647
648         /* The device may not be attached yet */
649         if (!child_device->driver)
650                 return;
651
652         drv = drv_to_hv_drv(child_device->driver);
653
654         if (drv->shutdown)
655                 drv->shutdown(dev);
656
657         return;
658 }
659
660
661 /*
662  * vmbus_device_release - Final callback release of the vmbus child device
663  */
664 static void vmbus_device_release(struct device *device)
665 {
666         struct hv_device *hv_dev = device_to_hv_device(device);
667         struct vmbus_channel *channel = hv_dev->channel;
668
669         hv_process_channel_removal(channel,
670                                    channel->offermsg.child_relid);
671         kfree(hv_dev);
672
673 }
674
675 /* The one and only one */
676 static struct bus_type  hv_bus = {
677         .name =         "vmbus",
678         .match =                vmbus_match,
679         .shutdown =             vmbus_shutdown,
680         .remove =               vmbus_remove,
681         .probe =                vmbus_probe,
682         .uevent =               vmbus_uevent,
683         .dev_groups =           vmbus_groups,
684 };
685
686 struct onmessage_work_context {
687         struct work_struct work;
688         struct hv_message msg;
689 };
690
691 static void vmbus_onmessage_work(struct work_struct *work)
692 {
693         struct onmessage_work_context *ctx;
694
695         /* Do not process messages if we're in DISCONNECTED state */
696         if (vmbus_connection.conn_state == DISCONNECTED)
697                 return;
698
699         ctx = container_of(work, struct onmessage_work_context,
700                            work);
701         vmbus_onmessage(&ctx->msg);
702         kfree(ctx);
703 }
704
705 static void hv_process_timer_expiration(struct hv_message *msg, int cpu)
706 {
707         struct clock_event_device *dev = hv_context.clk_evt[cpu];
708
709         if (dev->event_handler)
710                 dev->event_handler(dev);
711
712         msg->header.message_type = HVMSG_NONE;
713
714         /*
715          * Make sure the write to MessageType (ie set to
716          * HVMSG_NONE) happens before we read the
717          * MessagePending and EOMing. Otherwise, the EOMing
718          * will not deliver any more messages since there is
719          * no empty slot
720          */
721         mb();
722
723         if (msg->header.message_flags.msg_pending) {
724                 /*
725                  * This will cause message queue rescan to
726                  * possibly deliver another msg from the
727                  * hypervisor
728                  */
729                 wrmsrl(HV_X64_MSR_EOM, 0);
730         }
731 }
732
733 static void vmbus_on_msg_dpc(unsigned long data)
734 {
735         int cpu = smp_processor_id();
736         void *page_addr = hv_context.synic_message_page[cpu];
737         struct hv_message *msg = (struct hv_message *)page_addr +
738                                   VMBUS_MESSAGE_SINT;
739         struct vmbus_channel_message_header *hdr;
740         struct vmbus_channel_message_table_entry *entry;
741         struct onmessage_work_context *ctx;
742
743         if (msg->header.message_type == HVMSG_NONE)
744                 /* no msg */
745                 return;
746
747         hdr = (struct vmbus_channel_message_header *)msg->u.payload;
748
749         if (hdr->msgtype >= CHANNELMSG_COUNT) {
750                 WARN_ONCE(1, "unknown msgtype=%d\n", hdr->msgtype);
751                 goto msg_handled;
752         }
753
754         entry = &channel_message_table[hdr->msgtype];
755         if (entry->handler_type == VMHT_BLOCKING) {
756                 ctx = kmalloc(sizeof(*ctx), GFP_ATOMIC);
757                 if (ctx == NULL)
758                         return;
759
760                 INIT_WORK(&ctx->work, vmbus_onmessage_work);
761                 memcpy(&ctx->msg, msg, sizeof(*msg));
762
763                 queue_work(vmbus_connection.work_queue, &ctx->work);
764         } else
765                 entry->message_handler(hdr);
766
767 msg_handled:
768         msg->header.message_type = HVMSG_NONE;
769
770         /*
771          * Make sure the write to MessageType (ie set to
772          * HVMSG_NONE) happens before we read the
773          * MessagePending and EOMing. Otherwise, the EOMing
774          * will not deliver any more messages since there is
775          * no empty slot
776          */
777         mb();
778
779         if (msg->header.message_flags.msg_pending) {
780                 /*
781                  * This will cause message queue rescan to
782                  * possibly deliver another msg from the
783                  * hypervisor
784                  */
785                 wrmsrl(HV_X64_MSR_EOM, 0);
786         }
787 }
788
789 static void vmbus_isr(void)
790 {
791         int cpu = smp_processor_id();
792         void *page_addr;
793         struct hv_message *msg;
794         union hv_synic_event_flags *event;
795         bool handled = false;
796
797         page_addr = hv_context.synic_event_page[cpu];
798         if (page_addr == NULL)
799                 return;
800
801         event = (union hv_synic_event_flags *)page_addr +
802                                          VMBUS_MESSAGE_SINT;
803         /*
804          * Check for events before checking for messages. This is the order
805          * in which events and messages are checked in Windows guests on
806          * Hyper-V, and the Windows team suggested we do the same.
807          */
808
809         if ((vmbus_proto_version == VERSION_WS2008) ||
810                 (vmbus_proto_version == VERSION_WIN7)) {
811
812                 /* Since we are a child, we only need to check bit 0 */
813                 if (sync_test_and_clear_bit(0,
814                         (unsigned long *) &event->flags32[0])) {
815                         handled = true;
816                 }
817         } else {
818                 /*
819                  * Our host is win8 or above. The signaling mechanism
820                  * has changed and we can directly look at the event page.
821                  * If bit n is set then we have an interrup on the channel
822                  * whose id is n.
823                  */
824                 handled = true;
825         }
826
827         if (handled)
828                 tasklet_schedule(hv_context.event_dpc[cpu]);
829
830
831         page_addr = hv_context.synic_message_page[cpu];
832         msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
833
834         /* Check if there are actual msgs to be processed */
835         if (msg->header.message_type != HVMSG_NONE) {
836                 if (msg->header.message_type == HVMSG_TIMER_EXPIRED)
837                         hv_process_timer_expiration(msg, cpu);
838                 else
839                         tasklet_schedule(&msg_dpc);
840         }
841 }
842
843
844 /*
845  * vmbus_bus_init -Main vmbus driver initialization routine.
846  *
847  * Here, we
848  *      - initialize the vmbus driver context
849  *      - invoke the vmbus hv main init routine
850  *      - retrieve the channel offers
851  */
852 static int vmbus_bus_init(void)
853 {
854         int ret;
855
856         /* Hypervisor initialization...setup hypercall page..etc */
857         ret = hv_init();
858         if (ret != 0) {
859                 pr_err("Unable to initialize the hypervisor - 0x%x\n", ret);
860                 return ret;
861         }
862
863         tasklet_init(&msg_dpc, vmbus_on_msg_dpc, 0);
864
865         ret = bus_register(&hv_bus);
866         if (ret)
867                 goto err_cleanup;
868
869         hv_setup_vmbus_irq(vmbus_isr);
870
871         ret = hv_synic_alloc();
872         if (ret)
873                 goto err_alloc;
874         /*
875          * Initialize the per-cpu interrupt state and
876          * connect to the host.
877          */
878         on_each_cpu(hv_synic_init, NULL, 1);
879         ret = vmbus_connect();
880         if (ret)
881                 goto err_connect;
882
883         if (vmbus_proto_version > VERSION_WIN7)
884                 cpu_hotplug_disable();
885
886         /*
887          * Only register if the crash MSRs are available
888          */
889         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
890                 register_die_notifier(&hyperv_die_block);
891                 atomic_notifier_chain_register(&panic_notifier_list,
892                                                &hyperv_panic_block);
893         }
894
895         vmbus_request_offers();
896
897         return 0;
898
899 err_connect:
900         on_each_cpu(hv_synic_cleanup, NULL, 1);
901 err_alloc:
902         hv_synic_free();
903         hv_remove_vmbus_irq();
904
905         bus_unregister(&hv_bus);
906
907 err_cleanup:
908         hv_cleanup();
909
910         return ret;
911 }
912
913 /**
914  * __vmbus_child_driver_register() - Register a vmbus's driver
915  * @hv_driver: Pointer to driver structure you want to register
916  * @owner: owner module of the drv
917  * @mod_name: module name string
918  *
919  * Registers the given driver with Linux through the 'driver_register()' call
920  * and sets up the hyper-v vmbus handling for this driver.
921  * It will return the state of the 'driver_register()' call.
922  *
923  */
924 int __vmbus_driver_register(struct hv_driver *hv_driver, struct module *owner, const char *mod_name)
925 {
926         int ret;
927
928         pr_info("registering driver %s\n", hv_driver->name);
929
930         ret = vmbus_exists();
931         if (ret < 0)
932                 return ret;
933
934         hv_driver->driver.name = hv_driver->name;
935         hv_driver->driver.owner = owner;
936         hv_driver->driver.mod_name = mod_name;
937         hv_driver->driver.bus = &hv_bus;
938
939         ret = driver_register(&hv_driver->driver);
940
941         return ret;
942 }
943 EXPORT_SYMBOL_GPL(__vmbus_driver_register);
944
945 /**
946  * vmbus_driver_unregister() - Unregister a vmbus's driver
947  * @hv_driver: Pointer to driver structure you want to
948  *             un-register
949  *
950  * Un-register the given driver that was previous registered with a call to
951  * vmbus_driver_register()
952  */
953 void vmbus_driver_unregister(struct hv_driver *hv_driver)
954 {
955         pr_info("unregistering driver %s\n", hv_driver->name);
956
957         if (!vmbus_exists())
958                 driver_unregister(&hv_driver->driver);
959 }
960 EXPORT_SYMBOL_GPL(vmbus_driver_unregister);
961
962 /*
963  * vmbus_device_create - Creates and registers a new child device
964  * on the vmbus.
965  */
966 struct hv_device *vmbus_device_create(const uuid_le *type,
967                                       const uuid_le *instance,
968                                       struct vmbus_channel *channel)
969 {
970         struct hv_device *child_device_obj;
971
972         child_device_obj = kzalloc(sizeof(struct hv_device), GFP_KERNEL);
973         if (!child_device_obj) {
974                 pr_err("Unable to allocate device object for child device\n");
975                 return NULL;
976         }
977
978         child_device_obj->channel = channel;
979         memcpy(&child_device_obj->dev_type, type, sizeof(uuid_le));
980         memcpy(&child_device_obj->dev_instance, instance,
981                sizeof(uuid_le));
982         child_device_obj->vendor_id = 0x1414; /* MSFT vendor ID */
983
984
985         return child_device_obj;
986 }
987
988 /*
989  * vmbus_device_register - Register the child device
990  */
991 int vmbus_device_register(struct hv_device *child_device_obj)
992 {
993         int ret = 0;
994
995         dev_set_name(&child_device_obj->device, "vmbus_%d",
996                      child_device_obj->channel->id);
997
998         child_device_obj->device.bus = &hv_bus;
999         child_device_obj->device.parent = &hv_acpi_dev->dev;
1000         child_device_obj->device.release = vmbus_device_release;
1001
1002         /*
1003          * Register with the LDM. This will kick off the driver/device
1004          * binding...which will eventually call vmbus_match() and vmbus_probe()
1005          */
1006         ret = device_register(&child_device_obj->device);
1007
1008         if (ret)
1009                 pr_err("Unable to register child device\n");
1010         else
1011                 pr_debug("child device %s registered\n",
1012                         dev_name(&child_device_obj->device));
1013
1014         return ret;
1015 }
1016
1017 /*
1018  * vmbus_device_unregister - Remove the specified child device
1019  * from the vmbus.
1020  */
1021 void vmbus_device_unregister(struct hv_device *device_obj)
1022 {
1023         pr_debug("child device %s unregistered\n",
1024                 dev_name(&device_obj->device));
1025
1026         /*
1027          * Kick off the process of unregistering the device.
1028          * This will call vmbus_remove() and eventually vmbus_device_release()
1029          */
1030         device_unregister(&device_obj->device);
1031 }
1032
1033
1034 /*
1035  * VMBUS is an acpi enumerated device. Get the information we
1036  * need from DSDT.
1037  */
1038 #define VTPM_BASE_ADDRESS 0xfed40000
1039 static acpi_status vmbus_walk_resources(struct acpi_resource *res, void *ctx)
1040 {
1041         resource_size_t start = 0;
1042         resource_size_t end = 0;
1043         struct resource *new_res;
1044         struct resource **old_res = &hyperv_mmio;
1045         struct resource **prev_res = NULL;
1046
1047         switch (res->type) {
1048
1049         /*
1050          * "Address" descriptors are for bus windows. Ignore
1051          * "memory" descriptors, which are for registers on
1052          * devices.
1053          */
1054         case ACPI_RESOURCE_TYPE_ADDRESS32:
1055                 start = res->data.address32.address.minimum;
1056                 end = res->data.address32.address.maximum;
1057                 break;
1058
1059         case ACPI_RESOURCE_TYPE_ADDRESS64:
1060                 start = res->data.address64.address.minimum;
1061                 end = res->data.address64.address.maximum;
1062                 break;
1063
1064         default:
1065                 /* Unused resource type */
1066                 return AE_OK;
1067
1068         }
1069         /*
1070          * Ignore ranges that are below 1MB, as they're not
1071          * necessary or useful here.
1072          */
1073         if (end < 0x100000)
1074                 return AE_OK;
1075
1076         new_res = kzalloc(sizeof(*new_res), GFP_ATOMIC);
1077         if (!new_res)
1078                 return AE_NO_MEMORY;
1079
1080         /* If this range overlaps the virtual TPM, truncate it. */
1081         if (end > VTPM_BASE_ADDRESS && start < VTPM_BASE_ADDRESS)
1082                 end = VTPM_BASE_ADDRESS;
1083
1084         new_res->name = "hyperv mmio";
1085         new_res->flags = IORESOURCE_MEM;
1086         new_res->start = start;
1087         new_res->end = end;
1088
1089         /*
1090          * Stick ranges from higher in address space at the front of the list.
1091          * If two ranges are adjacent, merge them.
1092          */
1093         do {
1094                 if (!*old_res) {
1095                         *old_res = new_res;
1096                         break;
1097                 }
1098
1099                 if (((*old_res)->end + 1) == new_res->start) {
1100                         (*old_res)->end = new_res->end;
1101                         kfree(new_res);
1102                         break;
1103                 }
1104
1105                 if ((*old_res)->start == new_res->end + 1) {
1106                         (*old_res)->start = new_res->start;
1107                         kfree(new_res);
1108                         break;
1109                 }
1110
1111                 if ((*old_res)->end < new_res->start) {
1112                         new_res->sibling = *old_res;
1113                         if (prev_res)
1114                                 (*prev_res)->sibling = new_res;
1115                         *old_res = new_res;
1116                         break;
1117                 }
1118
1119                 prev_res = old_res;
1120                 old_res = &(*old_res)->sibling;
1121
1122         } while (1);
1123
1124         return AE_OK;
1125 }
1126
1127 static int vmbus_acpi_remove(struct acpi_device *device)
1128 {
1129         struct resource *cur_res;
1130         struct resource *next_res;
1131
1132         if (hyperv_mmio) {
1133                 for (cur_res = hyperv_mmio; cur_res; cur_res = next_res) {
1134                         next_res = cur_res->sibling;
1135                         kfree(cur_res);
1136                 }
1137         }
1138
1139         return 0;
1140 }
1141
1142 /**
1143  * vmbus_allocate_mmio() - Pick a memory-mapped I/O range.
1144  * @new:                If successful, supplied a pointer to the
1145  *                      allocated MMIO space.
1146  * @device_obj:         Identifies the caller
1147  * @min:                Minimum guest physical address of the
1148  *                      allocation
1149  * @max:                Maximum guest physical address
1150  * @size:               Size of the range to be allocated
1151  * @align:              Alignment of the range to be allocated
1152  * @fb_overlap_ok:      Whether this allocation can be allowed
1153  *                      to overlap the video frame buffer.
1154  *
1155  * This function walks the resources granted to VMBus by the
1156  * _CRS object in the ACPI namespace underneath the parent
1157  * "bridge" whether that's a root PCI bus in the Generation 1
1158  * case or a Module Device in the Generation 2 case.  It then
1159  * attempts to allocate from the global MMIO pool in a way that
1160  * matches the constraints supplied in these parameters and by
1161  * that _CRS.
1162  *
1163  * Return: 0 on success, -errno on failure
1164  */
1165 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
1166                         resource_size_t min, resource_size_t max,
1167                         resource_size_t size, resource_size_t align,
1168                         bool fb_overlap_ok)
1169 {
1170         struct resource *iter;
1171         resource_size_t range_min, range_max, start, local_min, local_max;
1172         const char *dev_n = dev_name(&device_obj->device);
1173         u32 fb_end = screen_info.lfb_base + (screen_info.lfb_size << 1);
1174         int i;
1175
1176         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
1177                 if ((iter->start >= max) || (iter->end <= min))
1178                         continue;
1179
1180                 range_min = iter->start;
1181                 range_max = iter->end;
1182
1183                 /* If this range overlaps the frame buffer, split it into
1184                    two tries. */
1185                 for (i = 0; i < 2; i++) {
1186                         local_min = range_min;
1187                         local_max = range_max;
1188                         if (fb_overlap_ok || (range_min >= fb_end) ||
1189                             (range_max <= screen_info.lfb_base)) {
1190                                 i++;
1191                         } else {
1192                                 if ((range_min <= screen_info.lfb_base) &&
1193                                     (range_max >= screen_info.lfb_base)) {
1194                                         /*
1195                                          * The frame buffer is in this window,
1196                                          * so trim this into the part that
1197                                          * preceeds the frame buffer.
1198                                          */
1199                                         local_max = screen_info.lfb_base - 1;
1200                                         range_min = fb_end;
1201                                 } else {
1202                                         range_min = fb_end;
1203                                         continue;
1204                                 }
1205                         }
1206
1207                         start = (local_min + align - 1) & ~(align - 1);
1208                         for (; start + size - 1 <= local_max; start += align) {
1209                                 *new = request_mem_region_exclusive(start, size,
1210                                                                     dev_n);
1211                                 if (*new)
1212                                         return 0;
1213                         }
1214                 }
1215         }
1216
1217         return -ENXIO;
1218 }
1219 EXPORT_SYMBOL_GPL(vmbus_allocate_mmio);
1220
1221 /**
1222  * vmbus_cpu_number_to_vp_number() - Map CPU to VP.
1223  * @cpu_number: CPU number in Linux terms
1224  *
1225  * This function returns the mapping between the Linux processor
1226  * number and the hypervisor's virtual processor number, useful
1227  * in making hypercalls and such that talk about specific
1228  * processors.
1229  *
1230  * Return: Virtual processor number in Hyper-V terms
1231  */
1232 int vmbus_cpu_number_to_vp_number(int cpu_number)
1233 {
1234         return hv_context.vp_index[cpu_number];
1235 }
1236 EXPORT_SYMBOL_GPL(vmbus_cpu_number_to_vp_number);
1237
1238 static int vmbus_acpi_add(struct acpi_device *device)
1239 {
1240         acpi_status result;
1241         int ret_val = -ENODEV;
1242         struct acpi_device *ancestor;
1243
1244         hv_acpi_dev = device;
1245
1246         result = acpi_walk_resources(device->handle, METHOD_NAME__CRS,
1247                                         vmbus_walk_resources, NULL);
1248
1249         if (ACPI_FAILURE(result))
1250                 goto acpi_walk_err;
1251         /*
1252          * Some ancestor of the vmbus acpi device (Gen1 or Gen2
1253          * firmware) is the VMOD that has the mmio ranges. Get that.
1254          */
1255         for (ancestor = device->parent; ancestor; ancestor = ancestor->parent) {
1256                 result = acpi_walk_resources(ancestor->handle, METHOD_NAME__CRS,
1257                                              vmbus_walk_resources, NULL);
1258
1259                 if (ACPI_FAILURE(result))
1260                         continue;
1261                 if (hyperv_mmio)
1262                         break;
1263         }
1264         ret_val = 0;
1265
1266 acpi_walk_err:
1267         complete(&probe_event);
1268         if (ret_val)
1269                 vmbus_acpi_remove(device);
1270         return ret_val;
1271 }
1272
1273 static const struct acpi_device_id vmbus_acpi_device_ids[] = {
1274         {"VMBUS", 0},
1275         {"VMBus", 0},
1276         {"", 0},
1277 };
1278 MODULE_DEVICE_TABLE(acpi, vmbus_acpi_device_ids);
1279
1280 static struct acpi_driver vmbus_acpi_driver = {
1281         .name = "vmbus",
1282         .ids = vmbus_acpi_device_ids,
1283         .ops = {
1284                 .add = vmbus_acpi_add,
1285                 .remove = vmbus_acpi_remove,
1286         },
1287 };
1288
1289 static void hv_kexec_handler(void)
1290 {
1291         int cpu;
1292
1293         hv_synic_clockevents_cleanup();
1294         vmbus_initiate_unload(false);
1295         for_each_online_cpu(cpu)
1296                 smp_call_function_single(cpu, hv_synic_cleanup, NULL, 1);
1297         hv_cleanup();
1298 };
1299
1300 static void hv_crash_handler(struct pt_regs *regs)
1301 {
1302         vmbus_initiate_unload(true);
1303         /*
1304          * In crash handler we can't schedule synic cleanup for all CPUs,
1305          * doing the cleanup for current CPU only. This should be sufficient
1306          * for kdump.
1307          */
1308         hv_synic_cleanup(NULL);
1309         hv_cleanup();
1310 };
1311
1312 static int __init hv_acpi_init(void)
1313 {
1314         int ret, t;
1315
1316         if (x86_hyper != &x86_hyper_ms_hyperv)
1317                 return -ENODEV;
1318
1319         init_completion(&probe_event);
1320
1321         /*
1322          * Get ACPI resources first.
1323          */
1324         ret = acpi_bus_register_driver(&vmbus_acpi_driver);
1325
1326         if (ret)
1327                 return ret;
1328
1329         t = wait_for_completion_timeout(&probe_event, 5*HZ);
1330         if (t == 0) {
1331                 ret = -ETIMEDOUT;
1332                 goto cleanup;
1333         }
1334
1335         ret = vmbus_bus_init();
1336         if (ret)
1337                 goto cleanup;
1338
1339         hv_setup_kexec_handler(hv_kexec_handler);
1340         hv_setup_crash_handler(hv_crash_handler);
1341
1342         return 0;
1343
1344 cleanup:
1345         acpi_bus_unregister_driver(&vmbus_acpi_driver);
1346         hv_acpi_dev = NULL;
1347         return ret;
1348 }
1349
1350 static void __exit vmbus_exit(void)
1351 {
1352         int cpu;
1353
1354         hv_remove_kexec_handler();
1355         hv_remove_crash_handler();
1356         vmbus_connection.conn_state = DISCONNECTED;
1357         hv_synic_clockevents_cleanup();
1358         vmbus_disconnect();
1359         hv_remove_vmbus_irq();
1360         tasklet_kill(&msg_dpc);
1361         vmbus_free_channels();
1362         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1363                 unregister_die_notifier(&hyperv_die_block);
1364                 atomic_notifier_chain_unregister(&panic_notifier_list,
1365                                                  &hyperv_panic_block);
1366         }
1367         bus_unregister(&hv_bus);
1368         hv_cleanup();
1369         for_each_online_cpu(cpu) {
1370                 tasklet_kill(hv_context.event_dpc[cpu]);
1371                 smp_call_function_single(cpu, hv_synic_cleanup, NULL, 1);
1372         }
1373         hv_synic_free();
1374         acpi_bus_unregister_driver(&vmbus_acpi_driver);
1375         if (vmbus_proto_version > VERSION_WIN7)
1376                 cpu_hotplug_enable();
1377 }
1378
1379
1380 MODULE_LICENSE("GPL");
1381
1382 subsys_initcall(hv_acpi_init);
1383 module_exit(vmbus_exit);