Merge branches 'pm-cpufreq' and 'pm-cpuidle'
[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 /* Set up per device attributes in /sys/bus/vmbus/devices/<bus device> */
481 static struct attribute *vmbus_attrs[] = {
482         &dev_attr_id.attr,
483         &dev_attr_state.attr,
484         &dev_attr_monitor_id.attr,
485         &dev_attr_class_id.attr,
486         &dev_attr_device_id.attr,
487         &dev_attr_modalias.attr,
488         &dev_attr_server_monitor_pending.attr,
489         &dev_attr_client_monitor_pending.attr,
490         &dev_attr_server_monitor_latency.attr,
491         &dev_attr_client_monitor_latency.attr,
492         &dev_attr_server_monitor_conn_id.attr,
493         &dev_attr_client_monitor_conn_id.attr,
494         &dev_attr_out_intr_mask.attr,
495         &dev_attr_out_read_index.attr,
496         &dev_attr_out_write_index.attr,
497         &dev_attr_out_read_bytes_avail.attr,
498         &dev_attr_out_write_bytes_avail.attr,
499         &dev_attr_in_intr_mask.attr,
500         &dev_attr_in_read_index.attr,
501         &dev_attr_in_write_index.attr,
502         &dev_attr_in_read_bytes_avail.attr,
503         &dev_attr_in_write_bytes_avail.attr,
504         &dev_attr_channel_vp_mapping.attr,
505         NULL,
506 };
507 ATTRIBUTE_GROUPS(vmbus);
508
509 /*
510  * vmbus_uevent - add uevent for our device
511  *
512  * This routine is invoked when a device is added or removed on the vmbus to
513  * generate a uevent to udev in the userspace. The udev will then look at its
514  * rule and the uevent generated here to load the appropriate driver
515  *
516  * The alias string will be of the form vmbus:guid where guid is the string
517  * representation of the device guid (each byte of the guid will be
518  * represented with two hex characters.
519  */
520 static int vmbus_uevent(struct device *device, struct kobj_uevent_env *env)
521 {
522         struct hv_device *dev = device_to_hv_device(device);
523         int ret;
524         char alias_name[VMBUS_ALIAS_LEN + 1];
525
526         print_alias_name(dev, alias_name);
527         ret = add_uevent_var(env, "MODALIAS=vmbus:%s", alias_name);
528         return ret;
529 }
530
531 static const uuid_le null_guid;
532
533 static inline bool is_null_guid(const uuid_le *guid)
534 {
535         if (uuid_le_cmp(*guid, null_guid))
536                 return false;
537         return true;
538 }
539
540 /*
541  * Return a matching hv_vmbus_device_id pointer.
542  * If there is no match, return NULL.
543  */
544 static const struct hv_vmbus_device_id *hv_vmbus_get_id(
545                                         const struct hv_vmbus_device_id *id,
546                                         const uuid_le *guid)
547 {
548         for (; !is_null_guid(&id->guid); id++)
549                 if (!uuid_le_cmp(id->guid, *guid))
550                         return id;
551
552         return NULL;
553 }
554
555
556
557 /*
558  * vmbus_match - Attempt to match the specified device to the specified driver
559  */
560 static int vmbus_match(struct device *device, struct device_driver *driver)
561 {
562         struct hv_driver *drv = drv_to_hv_drv(driver);
563         struct hv_device *hv_dev = device_to_hv_device(device);
564
565         if (hv_vmbus_get_id(drv->id_table, &hv_dev->dev_type))
566                 return 1;
567
568         return 0;
569 }
570
571 /*
572  * vmbus_probe - Add the new vmbus's child device
573  */
574 static int vmbus_probe(struct device *child_device)
575 {
576         int ret = 0;
577         struct hv_driver *drv =
578                         drv_to_hv_drv(child_device->driver);
579         struct hv_device *dev = device_to_hv_device(child_device);
580         const struct hv_vmbus_device_id *dev_id;
581
582         dev_id = hv_vmbus_get_id(drv->id_table, &dev->dev_type);
583         if (drv->probe) {
584                 ret = drv->probe(dev, dev_id);
585                 if (ret != 0)
586                         pr_err("probe failed for device %s (%d)\n",
587                                dev_name(child_device), ret);
588
589         } else {
590                 pr_err("probe not set for driver %s\n",
591                        dev_name(child_device));
592                 ret = -ENODEV;
593         }
594         return ret;
595 }
596
597 /*
598  * vmbus_remove - Remove a vmbus device
599  */
600 static int vmbus_remove(struct device *child_device)
601 {
602         struct hv_driver *drv;
603         struct hv_device *dev = device_to_hv_device(child_device);
604
605         if (child_device->driver) {
606                 drv = drv_to_hv_drv(child_device->driver);
607                 if (drv->remove)
608                         drv->remove(dev);
609         }
610
611         return 0;
612 }
613
614
615 /*
616  * vmbus_shutdown - Shutdown a vmbus device
617  */
618 static void vmbus_shutdown(struct device *child_device)
619 {
620         struct hv_driver *drv;
621         struct hv_device *dev = device_to_hv_device(child_device);
622
623
624         /* The device may not be attached yet */
625         if (!child_device->driver)
626                 return;
627
628         drv = drv_to_hv_drv(child_device->driver);
629
630         if (drv->shutdown)
631                 drv->shutdown(dev);
632
633         return;
634 }
635
636
637 /*
638  * vmbus_device_release - Final callback release of the vmbus child device
639  */
640 static void vmbus_device_release(struct device *device)
641 {
642         struct hv_device *hv_dev = device_to_hv_device(device);
643         struct vmbus_channel *channel = hv_dev->channel;
644
645         hv_process_channel_removal(channel,
646                                    channel->offermsg.child_relid);
647         kfree(hv_dev);
648
649 }
650
651 /* The one and only one */
652 static struct bus_type  hv_bus = {
653         .name =         "vmbus",
654         .match =                vmbus_match,
655         .shutdown =             vmbus_shutdown,
656         .remove =               vmbus_remove,
657         .probe =                vmbus_probe,
658         .uevent =               vmbus_uevent,
659         .dev_groups =           vmbus_groups,
660 };
661
662 struct onmessage_work_context {
663         struct work_struct work;
664         struct hv_message msg;
665 };
666
667 static void vmbus_onmessage_work(struct work_struct *work)
668 {
669         struct onmessage_work_context *ctx;
670
671         /* Do not process messages if we're in DISCONNECTED state */
672         if (vmbus_connection.conn_state == DISCONNECTED)
673                 return;
674
675         ctx = container_of(work, struct onmessage_work_context,
676                            work);
677         vmbus_onmessage(&ctx->msg);
678         kfree(ctx);
679 }
680
681 static void hv_process_timer_expiration(struct hv_message *msg, int cpu)
682 {
683         struct clock_event_device *dev = hv_context.clk_evt[cpu];
684
685         if (dev->event_handler)
686                 dev->event_handler(dev);
687
688         msg->header.message_type = HVMSG_NONE;
689
690         /*
691          * Make sure the write to MessageType (ie set to
692          * HVMSG_NONE) happens before we read the
693          * MessagePending and EOMing. Otherwise, the EOMing
694          * will not deliver any more messages since there is
695          * no empty slot
696          */
697         mb();
698
699         if (msg->header.message_flags.msg_pending) {
700                 /*
701                  * This will cause message queue rescan to
702                  * possibly deliver another msg from the
703                  * hypervisor
704                  */
705                 wrmsrl(HV_X64_MSR_EOM, 0);
706         }
707 }
708
709 static void vmbus_on_msg_dpc(unsigned long data)
710 {
711         int cpu = smp_processor_id();
712         void *page_addr = hv_context.synic_message_page[cpu];
713         struct hv_message *msg = (struct hv_message *)page_addr +
714                                   VMBUS_MESSAGE_SINT;
715         struct vmbus_channel_message_header *hdr;
716         struct vmbus_channel_message_table_entry *entry;
717         struct onmessage_work_context *ctx;
718
719         while (1) {
720                 if (msg->header.message_type == HVMSG_NONE)
721                         /* no msg */
722                         break;
723
724                 hdr = (struct vmbus_channel_message_header *)msg->u.payload;
725
726                 if (hdr->msgtype >= CHANNELMSG_COUNT) {
727                         WARN_ONCE(1, "unknown msgtype=%d\n", hdr->msgtype);
728                         goto msg_handled;
729                 }
730
731                 entry = &channel_message_table[hdr->msgtype];
732                 if (entry->handler_type == VMHT_BLOCKING) {
733                         ctx = kmalloc(sizeof(*ctx), GFP_ATOMIC);
734                         if (ctx == NULL)
735                                 continue;
736
737                         INIT_WORK(&ctx->work, vmbus_onmessage_work);
738                         memcpy(&ctx->msg, msg, sizeof(*msg));
739
740                         queue_work(vmbus_connection.work_queue, &ctx->work);
741                 } else
742                         entry->message_handler(hdr);
743
744 msg_handled:
745                 msg->header.message_type = HVMSG_NONE;
746
747                 /*
748                  * Make sure the write to MessageType (ie set to
749                  * HVMSG_NONE) happens before we read the
750                  * MessagePending and EOMing. Otherwise, the EOMing
751                  * will not deliver any more messages since there is
752                  * no empty slot
753                  */
754                 mb();
755
756                 if (msg->header.message_flags.msg_pending) {
757                         /*
758                          * This will cause message queue rescan to
759                          * possibly deliver another msg from the
760                          * hypervisor
761                          */
762                         wrmsrl(HV_X64_MSR_EOM, 0);
763                 }
764         }
765 }
766
767 static void vmbus_isr(void)
768 {
769         int cpu = smp_processor_id();
770         void *page_addr;
771         struct hv_message *msg;
772         union hv_synic_event_flags *event;
773         bool handled = false;
774
775         page_addr = hv_context.synic_event_page[cpu];
776         if (page_addr == NULL)
777                 return;
778
779         event = (union hv_synic_event_flags *)page_addr +
780                                          VMBUS_MESSAGE_SINT;
781         /*
782          * Check for events before checking for messages. This is the order
783          * in which events and messages are checked in Windows guests on
784          * Hyper-V, and the Windows team suggested we do the same.
785          */
786
787         if ((vmbus_proto_version == VERSION_WS2008) ||
788                 (vmbus_proto_version == VERSION_WIN7)) {
789
790                 /* Since we are a child, we only need to check bit 0 */
791                 if (sync_test_and_clear_bit(0,
792                         (unsigned long *) &event->flags32[0])) {
793                         handled = true;
794                 }
795         } else {
796                 /*
797                  * Our host is win8 or above. The signaling mechanism
798                  * has changed and we can directly look at the event page.
799                  * If bit n is set then we have an interrup on the channel
800                  * whose id is n.
801                  */
802                 handled = true;
803         }
804
805         if (handled)
806                 tasklet_schedule(hv_context.event_dpc[cpu]);
807
808
809         page_addr = hv_context.synic_message_page[cpu];
810         msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
811
812         /* Check if there are actual msgs to be processed */
813         if (msg->header.message_type != HVMSG_NONE) {
814                 if (msg->header.message_type == HVMSG_TIMER_EXPIRED)
815                         hv_process_timer_expiration(msg, cpu);
816                 else
817                         tasklet_schedule(&msg_dpc);
818         }
819 }
820
821
822 /*
823  * vmbus_bus_init -Main vmbus driver initialization routine.
824  *
825  * Here, we
826  *      - initialize the vmbus driver context
827  *      - invoke the vmbus hv main init routine
828  *      - retrieve the channel offers
829  */
830 static int vmbus_bus_init(void)
831 {
832         int ret;
833
834         /* Hypervisor initialization...setup hypercall page..etc */
835         ret = hv_init();
836         if (ret != 0) {
837                 pr_err("Unable to initialize the hypervisor - 0x%x\n", ret);
838                 return ret;
839         }
840
841         tasklet_init(&msg_dpc, vmbus_on_msg_dpc, 0);
842
843         ret = bus_register(&hv_bus);
844         if (ret)
845                 goto err_cleanup;
846
847         hv_setup_vmbus_irq(vmbus_isr);
848
849         ret = hv_synic_alloc();
850         if (ret)
851                 goto err_alloc;
852         /*
853          * Initialize the per-cpu interrupt state and
854          * connect to the host.
855          */
856         on_each_cpu(hv_synic_init, NULL, 1);
857         ret = vmbus_connect();
858         if (ret)
859                 goto err_connect;
860
861         if (vmbus_proto_version > VERSION_WIN7)
862                 cpu_hotplug_disable();
863
864         /*
865          * Only register if the crash MSRs are available
866          */
867         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
868                 register_die_notifier(&hyperv_die_block);
869                 atomic_notifier_chain_register(&panic_notifier_list,
870                                                &hyperv_panic_block);
871         }
872
873         vmbus_request_offers();
874
875         return 0;
876
877 err_connect:
878         on_each_cpu(hv_synic_cleanup, NULL, 1);
879 err_alloc:
880         hv_synic_free();
881         hv_remove_vmbus_irq();
882
883         bus_unregister(&hv_bus);
884
885 err_cleanup:
886         hv_cleanup();
887
888         return ret;
889 }
890
891 /**
892  * __vmbus_child_driver_register() - Register a vmbus's driver
893  * @hv_driver: Pointer to driver structure you want to register
894  * @owner: owner module of the drv
895  * @mod_name: module name string
896  *
897  * Registers the given driver with Linux through the 'driver_register()' call
898  * and sets up the hyper-v vmbus handling for this driver.
899  * It will return the state of the 'driver_register()' call.
900  *
901  */
902 int __vmbus_driver_register(struct hv_driver *hv_driver, struct module *owner, const char *mod_name)
903 {
904         int ret;
905
906         pr_info("registering driver %s\n", hv_driver->name);
907
908         ret = vmbus_exists();
909         if (ret < 0)
910                 return ret;
911
912         hv_driver->driver.name = hv_driver->name;
913         hv_driver->driver.owner = owner;
914         hv_driver->driver.mod_name = mod_name;
915         hv_driver->driver.bus = &hv_bus;
916
917         ret = driver_register(&hv_driver->driver);
918
919         return ret;
920 }
921 EXPORT_SYMBOL_GPL(__vmbus_driver_register);
922
923 /**
924  * vmbus_driver_unregister() - Unregister a vmbus's driver
925  * @hv_driver: Pointer to driver structure you want to
926  *             un-register
927  *
928  * Un-register the given driver that was previous registered with a call to
929  * vmbus_driver_register()
930  */
931 void vmbus_driver_unregister(struct hv_driver *hv_driver)
932 {
933         pr_info("unregistering driver %s\n", hv_driver->name);
934
935         if (!vmbus_exists())
936                 driver_unregister(&hv_driver->driver);
937 }
938 EXPORT_SYMBOL_GPL(vmbus_driver_unregister);
939
940 /*
941  * vmbus_device_create - Creates and registers a new child device
942  * on the vmbus.
943  */
944 struct hv_device *vmbus_device_create(const uuid_le *type,
945                                       const uuid_le *instance,
946                                       struct vmbus_channel *channel)
947 {
948         struct hv_device *child_device_obj;
949
950         child_device_obj = kzalloc(sizeof(struct hv_device), GFP_KERNEL);
951         if (!child_device_obj) {
952                 pr_err("Unable to allocate device object for child device\n");
953                 return NULL;
954         }
955
956         child_device_obj->channel = channel;
957         memcpy(&child_device_obj->dev_type, type, sizeof(uuid_le));
958         memcpy(&child_device_obj->dev_instance, instance,
959                sizeof(uuid_le));
960
961
962         return child_device_obj;
963 }
964
965 /*
966  * vmbus_device_register - Register the child device
967  */
968 int vmbus_device_register(struct hv_device *child_device_obj)
969 {
970         int ret = 0;
971
972         dev_set_name(&child_device_obj->device, "vmbus_%d",
973                      child_device_obj->channel->id);
974
975         child_device_obj->device.bus = &hv_bus;
976         child_device_obj->device.parent = &hv_acpi_dev->dev;
977         child_device_obj->device.release = vmbus_device_release;
978
979         /*
980          * Register with the LDM. This will kick off the driver/device
981          * binding...which will eventually call vmbus_match() and vmbus_probe()
982          */
983         ret = device_register(&child_device_obj->device);
984
985         if (ret)
986                 pr_err("Unable to register child device\n");
987         else
988                 pr_debug("child device %s registered\n",
989                         dev_name(&child_device_obj->device));
990
991         return ret;
992 }
993
994 /*
995  * vmbus_device_unregister - Remove the specified child device
996  * from the vmbus.
997  */
998 void vmbus_device_unregister(struct hv_device *device_obj)
999 {
1000         pr_debug("child device %s unregistered\n",
1001                 dev_name(&device_obj->device));
1002
1003         /*
1004          * Kick off the process of unregistering the device.
1005          * This will call vmbus_remove() and eventually vmbus_device_release()
1006          */
1007         device_unregister(&device_obj->device);
1008 }
1009
1010
1011 /*
1012  * VMBUS is an acpi enumerated device. Get the information we
1013  * need from DSDT.
1014  */
1015 #define VTPM_BASE_ADDRESS 0xfed40000
1016 static acpi_status vmbus_walk_resources(struct acpi_resource *res, void *ctx)
1017 {
1018         resource_size_t start = 0;
1019         resource_size_t end = 0;
1020         struct resource *new_res;
1021         struct resource **old_res = &hyperv_mmio;
1022         struct resource **prev_res = NULL;
1023
1024         switch (res->type) {
1025
1026         /*
1027          * "Address" descriptors are for bus windows. Ignore
1028          * "memory" descriptors, which are for registers on
1029          * devices.
1030          */
1031         case ACPI_RESOURCE_TYPE_ADDRESS32:
1032                 start = res->data.address32.address.minimum;
1033                 end = res->data.address32.address.maximum;
1034                 break;
1035
1036         case ACPI_RESOURCE_TYPE_ADDRESS64:
1037                 start = res->data.address64.address.minimum;
1038                 end = res->data.address64.address.maximum;
1039                 break;
1040
1041         default:
1042                 /* Unused resource type */
1043                 return AE_OK;
1044
1045         }
1046         /*
1047          * Ignore ranges that are below 1MB, as they're not
1048          * necessary or useful here.
1049          */
1050         if (end < 0x100000)
1051                 return AE_OK;
1052
1053         new_res = kzalloc(sizeof(*new_res), GFP_ATOMIC);
1054         if (!new_res)
1055                 return AE_NO_MEMORY;
1056
1057         /* If this range overlaps the virtual TPM, truncate it. */
1058         if (end > VTPM_BASE_ADDRESS && start < VTPM_BASE_ADDRESS)
1059                 end = VTPM_BASE_ADDRESS;
1060
1061         new_res->name = "hyperv mmio";
1062         new_res->flags = IORESOURCE_MEM;
1063         new_res->start = start;
1064         new_res->end = end;
1065
1066         /*
1067          * Stick ranges from higher in address space at the front of the list.
1068          * If two ranges are adjacent, merge them.
1069          */
1070         do {
1071                 if (!*old_res) {
1072                         *old_res = new_res;
1073                         break;
1074                 }
1075
1076                 if (((*old_res)->end + 1) == new_res->start) {
1077                         (*old_res)->end = new_res->end;
1078                         kfree(new_res);
1079                         break;
1080                 }
1081
1082                 if ((*old_res)->start == new_res->end + 1) {
1083                         (*old_res)->start = new_res->start;
1084                         kfree(new_res);
1085                         break;
1086                 }
1087
1088                 if ((*old_res)->end < new_res->start) {
1089                         new_res->sibling = *old_res;
1090                         if (prev_res)
1091                                 (*prev_res)->sibling = new_res;
1092                         *old_res = new_res;
1093                         break;
1094                 }
1095
1096                 prev_res = old_res;
1097                 old_res = &(*old_res)->sibling;
1098
1099         } while (1);
1100
1101         return AE_OK;
1102 }
1103
1104 static int vmbus_acpi_remove(struct acpi_device *device)
1105 {
1106         struct resource *cur_res;
1107         struct resource *next_res;
1108
1109         if (hyperv_mmio) {
1110                 for (cur_res = hyperv_mmio; cur_res; cur_res = next_res) {
1111                         next_res = cur_res->sibling;
1112                         kfree(cur_res);
1113                 }
1114         }
1115
1116         return 0;
1117 }
1118
1119 /**
1120  * vmbus_allocate_mmio() - Pick a memory-mapped I/O range.
1121  * @new:                If successful, supplied a pointer to the
1122  *                      allocated MMIO space.
1123  * @device_obj:         Identifies the caller
1124  * @min:                Minimum guest physical address of the
1125  *                      allocation
1126  * @max:                Maximum guest physical address
1127  * @size:               Size of the range to be allocated
1128  * @align:              Alignment of the range to be allocated
1129  * @fb_overlap_ok:      Whether this allocation can be allowed
1130  *                      to overlap the video frame buffer.
1131  *
1132  * This function walks the resources granted to VMBus by the
1133  * _CRS object in the ACPI namespace underneath the parent
1134  * "bridge" whether that's a root PCI bus in the Generation 1
1135  * case or a Module Device in the Generation 2 case.  It then
1136  * attempts to allocate from the global MMIO pool in a way that
1137  * matches the constraints supplied in these parameters and by
1138  * that _CRS.
1139  *
1140  * Return: 0 on success, -errno on failure
1141  */
1142 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
1143                         resource_size_t min, resource_size_t max,
1144                         resource_size_t size, resource_size_t align,
1145                         bool fb_overlap_ok)
1146 {
1147         struct resource *iter;
1148         resource_size_t range_min, range_max, start, local_min, local_max;
1149         const char *dev_n = dev_name(&device_obj->device);
1150         u32 fb_end = screen_info.lfb_base + (screen_info.lfb_size << 1);
1151         int i;
1152
1153         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
1154                 if ((iter->start >= max) || (iter->end <= min))
1155                         continue;
1156
1157                 range_min = iter->start;
1158                 range_max = iter->end;
1159
1160                 /* If this range overlaps the frame buffer, split it into
1161                    two tries. */
1162                 for (i = 0; i < 2; i++) {
1163                         local_min = range_min;
1164                         local_max = range_max;
1165                         if (fb_overlap_ok || (range_min >= fb_end) ||
1166                             (range_max <= screen_info.lfb_base)) {
1167                                 i++;
1168                         } else {
1169                                 if ((range_min <= screen_info.lfb_base) &&
1170                                     (range_max >= screen_info.lfb_base)) {
1171                                         /*
1172                                          * The frame buffer is in this window,
1173                                          * so trim this into the part that
1174                                          * preceeds the frame buffer.
1175                                          */
1176                                         local_max = screen_info.lfb_base - 1;
1177                                         range_min = fb_end;
1178                                 } else {
1179                                         range_min = fb_end;
1180                                         continue;
1181                                 }
1182                         }
1183
1184                         start = (local_min + align - 1) & ~(align - 1);
1185                         for (; start + size - 1 <= local_max; start += align) {
1186                                 *new = request_mem_region_exclusive(start, size,
1187                                                                     dev_n);
1188                                 if (*new)
1189                                         return 0;
1190                         }
1191                 }
1192         }
1193
1194         return -ENXIO;
1195 }
1196 EXPORT_SYMBOL_GPL(vmbus_allocate_mmio);
1197
1198 /**
1199  * vmbus_cpu_number_to_vp_number() - Map CPU to VP.
1200  * @cpu_number: CPU number in Linux terms
1201  *
1202  * This function returns the mapping between the Linux processor
1203  * number and the hypervisor's virtual processor number, useful
1204  * in making hypercalls and such that talk about specific
1205  * processors.
1206  *
1207  * Return: Virtual processor number in Hyper-V terms
1208  */
1209 int vmbus_cpu_number_to_vp_number(int cpu_number)
1210 {
1211         return hv_context.vp_index[cpu_number];
1212 }
1213 EXPORT_SYMBOL_GPL(vmbus_cpu_number_to_vp_number);
1214
1215 static int vmbus_acpi_add(struct acpi_device *device)
1216 {
1217         acpi_status result;
1218         int ret_val = -ENODEV;
1219         struct acpi_device *ancestor;
1220
1221         hv_acpi_dev = device;
1222
1223         result = acpi_walk_resources(device->handle, METHOD_NAME__CRS,
1224                                         vmbus_walk_resources, NULL);
1225
1226         if (ACPI_FAILURE(result))
1227                 goto acpi_walk_err;
1228         /*
1229          * Some ancestor of the vmbus acpi device (Gen1 or Gen2
1230          * firmware) is the VMOD that has the mmio ranges. Get that.
1231          */
1232         for (ancestor = device->parent; ancestor; ancestor = ancestor->parent) {
1233                 result = acpi_walk_resources(ancestor->handle, METHOD_NAME__CRS,
1234                                              vmbus_walk_resources, NULL);
1235
1236                 if (ACPI_FAILURE(result))
1237                         continue;
1238                 if (hyperv_mmio)
1239                         break;
1240         }
1241         ret_val = 0;
1242
1243 acpi_walk_err:
1244         complete(&probe_event);
1245         if (ret_val)
1246                 vmbus_acpi_remove(device);
1247         return ret_val;
1248 }
1249
1250 static const struct acpi_device_id vmbus_acpi_device_ids[] = {
1251         {"VMBUS", 0},
1252         {"VMBus", 0},
1253         {"", 0},
1254 };
1255 MODULE_DEVICE_TABLE(acpi, vmbus_acpi_device_ids);
1256
1257 static struct acpi_driver vmbus_acpi_driver = {
1258         .name = "vmbus",
1259         .ids = vmbus_acpi_device_ids,
1260         .ops = {
1261                 .add = vmbus_acpi_add,
1262                 .remove = vmbus_acpi_remove,
1263         },
1264 };
1265
1266 static void hv_kexec_handler(void)
1267 {
1268         int cpu;
1269
1270         hv_synic_clockevents_cleanup();
1271         vmbus_initiate_unload();
1272         for_each_online_cpu(cpu)
1273                 smp_call_function_single(cpu, hv_synic_cleanup, NULL, 1);
1274         hv_cleanup();
1275 };
1276
1277 static void hv_crash_handler(struct pt_regs *regs)
1278 {
1279         vmbus_initiate_unload();
1280         /*
1281          * In crash handler we can't schedule synic cleanup for all CPUs,
1282          * doing the cleanup for current CPU only. This should be sufficient
1283          * for kdump.
1284          */
1285         hv_synic_cleanup(NULL);
1286         hv_cleanup();
1287 };
1288
1289 static int __init hv_acpi_init(void)
1290 {
1291         int ret, t;
1292
1293         if (x86_hyper != &x86_hyper_ms_hyperv)
1294                 return -ENODEV;
1295
1296         init_completion(&probe_event);
1297
1298         /*
1299          * Get ACPI resources first.
1300          */
1301         ret = acpi_bus_register_driver(&vmbus_acpi_driver);
1302
1303         if (ret)
1304                 return ret;
1305
1306         t = wait_for_completion_timeout(&probe_event, 5*HZ);
1307         if (t == 0) {
1308                 ret = -ETIMEDOUT;
1309                 goto cleanup;
1310         }
1311
1312         ret = vmbus_bus_init();
1313         if (ret)
1314                 goto cleanup;
1315
1316         hv_setup_kexec_handler(hv_kexec_handler);
1317         hv_setup_crash_handler(hv_crash_handler);
1318
1319         return 0;
1320
1321 cleanup:
1322         acpi_bus_unregister_driver(&vmbus_acpi_driver);
1323         hv_acpi_dev = NULL;
1324         return ret;
1325 }
1326
1327 static void __exit vmbus_exit(void)
1328 {
1329         int cpu;
1330
1331         hv_remove_kexec_handler();
1332         hv_remove_crash_handler();
1333         vmbus_connection.conn_state = DISCONNECTED;
1334         hv_synic_clockevents_cleanup();
1335         vmbus_disconnect();
1336         hv_remove_vmbus_irq();
1337         tasklet_kill(&msg_dpc);
1338         vmbus_free_channels();
1339         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1340                 unregister_die_notifier(&hyperv_die_block);
1341                 atomic_notifier_chain_unregister(&panic_notifier_list,
1342                                                  &hyperv_panic_block);
1343         }
1344         bus_unregister(&hv_bus);
1345         hv_cleanup();
1346         for_each_online_cpu(cpu) {
1347                 tasklet_kill(hv_context.event_dpc[cpu]);
1348                 smp_call_function_single(cpu, hv_synic_cleanup, NULL, 1);
1349         }
1350         hv_synic_free();
1351         acpi_bus_unregister_driver(&vmbus_acpi_driver);
1352         if (vmbus_proto_version > VERSION_WIN7)
1353                 cpu_hotplug_enable();
1354 }
1355
1356
1357 MODULE_LICENSE("GPL");
1358
1359 subsys_initcall(hv_acpi_init);
1360 module_exit(vmbus_exit);