Merge branch 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[sfrench/cifs-2.6.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 <linux/sched/task_stack.h>
38
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 <linux/efi.h>
45 #include <linux/random.h>
46 #include "hyperv_vmbus.h"
47
48 struct vmbus_dynid {
49         struct list_head node;
50         struct hv_vmbus_device_id id;
51 };
52
53 static struct acpi_device  *hv_acpi_dev;
54
55 static struct completion probe_event;
56
57 static int hyperv_cpuhp_online;
58
59 static void *hv_panic_page;
60
61 static int hyperv_panic_event(struct notifier_block *nb, unsigned long val,
62                               void *args)
63 {
64         struct pt_regs *regs;
65
66         regs = current_pt_regs();
67
68         hyperv_report_panic(regs, val);
69         return NOTIFY_DONE;
70 }
71
72 static int hyperv_die_event(struct notifier_block *nb, unsigned long val,
73                             void *args)
74 {
75         struct die_args *die = (struct die_args *)args;
76         struct pt_regs *regs = die->regs;
77
78         hyperv_report_panic(regs, val);
79         return NOTIFY_DONE;
80 }
81
82 static struct notifier_block hyperv_die_block = {
83         .notifier_call = hyperv_die_event,
84 };
85 static struct notifier_block hyperv_panic_block = {
86         .notifier_call = hyperv_panic_event,
87 };
88
89 static const char *fb_mmio_name = "fb_range";
90 static struct resource *fb_mmio;
91 static struct resource *hyperv_mmio;
92 static DEFINE_SEMAPHORE(hyperv_mmio_lock);
93
94 static int vmbus_exists(void)
95 {
96         if (hv_acpi_dev == NULL)
97                 return -ENODEV;
98
99         return 0;
100 }
101
102 #define VMBUS_ALIAS_LEN ((sizeof((struct hv_vmbus_device_id *)0)->guid) * 2)
103 static void print_alias_name(struct hv_device *hv_dev, char *alias_name)
104 {
105         int i;
106         for (i = 0; i < VMBUS_ALIAS_LEN; i += 2)
107                 sprintf(&alias_name[i], "%02x", hv_dev->dev_type.b[i/2]);
108 }
109
110 static u8 channel_monitor_group(const struct vmbus_channel *channel)
111 {
112         return (u8)channel->offermsg.monitorid / 32;
113 }
114
115 static u8 channel_monitor_offset(const struct vmbus_channel *channel)
116 {
117         return (u8)channel->offermsg.monitorid % 32;
118 }
119
120 static u32 channel_pending(const struct vmbus_channel *channel,
121                            const struct hv_monitor_page *monitor_page)
122 {
123         u8 monitor_group = channel_monitor_group(channel);
124
125         return monitor_page->trigger_group[monitor_group].pending;
126 }
127
128 static u32 channel_latency(const struct vmbus_channel *channel,
129                            const struct hv_monitor_page *monitor_page)
130 {
131         u8 monitor_group = channel_monitor_group(channel);
132         u8 monitor_offset = channel_monitor_offset(channel);
133
134         return monitor_page->latency[monitor_group][monitor_offset];
135 }
136
137 static u32 channel_conn_id(struct vmbus_channel *channel,
138                            struct hv_monitor_page *monitor_page)
139 {
140         u8 monitor_group = channel_monitor_group(channel);
141         u8 monitor_offset = channel_monitor_offset(channel);
142         return monitor_page->parameter[monitor_group][monitor_offset].connectionid.u.id;
143 }
144
145 static ssize_t id_show(struct device *dev, struct device_attribute *dev_attr,
146                        char *buf)
147 {
148         struct hv_device *hv_dev = device_to_hv_device(dev);
149
150         if (!hv_dev->channel)
151                 return -ENODEV;
152         return sprintf(buf, "%d\n", hv_dev->channel->offermsg.child_relid);
153 }
154 static DEVICE_ATTR_RO(id);
155
156 static ssize_t state_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->state);
164 }
165 static DEVICE_ATTR_RO(state);
166
167 static ssize_t monitor_id_show(struct device *dev,
168                                struct device_attribute *dev_attr, 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->offermsg.monitorid);
175 }
176 static DEVICE_ATTR_RO(monitor_id);
177
178 static ssize_t class_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, "{%pUl}\n",
186                        hv_dev->channel->offermsg.offer.if_type.b);
187 }
188 static DEVICE_ATTR_RO(class_id);
189
190 static ssize_t device_id_show(struct device *dev,
191                               struct device_attribute *dev_attr, char *buf)
192 {
193         struct hv_device *hv_dev = device_to_hv_device(dev);
194
195         if (!hv_dev->channel)
196                 return -ENODEV;
197         return sprintf(buf, "{%pUl}\n",
198                        hv_dev->channel->offermsg.offer.if_instance.b);
199 }
200 static DEVICE_ATTR_RO(device_id);
201
202 static ssize_t modalias_show(struct device *dev,
203                              struct device_attribute *dev_attr, char *buf)
204 {
205         struct hv_device *hv_dev = device_to_hv_device(dev);
206         char alias_name[VMBUS_ALIAS_LEN + 1];
207
208         print_alias_name(hv_dev, alias_name);
209         return sprintf(buf, "vmbus:%s\n", alias_name);
210 }
211 static DEVICE_ATTR_RO(modalias);
212
213 #ifdef CONFIG_NUMA
214 static ssize_t numa_node_show(struct device *dev,
215                               struct device_attribute *attr, char *buf)
216 {
217         struct hv_device *hv_dev = device_to_hv_device(dev);
218
219         if (!hv_dev->channel)
220                 return -ENODEV;
221
222         return sprintf(buf, "%d\n", hv_dev->channel->numa_node);
223 }
224 static DEVICE_ATTR_RO(numa_node);
225 #endif
226
227 static ssize_t server_monitor_pending_show(struct device *dev,
228                                            struct device_attribute *dev_attr,
229                                            char *buf)
230 {
231         struct hv_device *hv_dev = device_to_hv_device(dev);
232
233         if (!hv_dev->channel)
234                 return -ENODEV;
235         return sprintf(buf, "%d\n",
236                        channel_pending(hv_dev->channel,
237                                        vmbus_connection.monitor_pages[1]));
238 }
239 static DEVICE_ATTR_RO(server_monitor_pending);
240
241 static ssize_t client_monitor_pending_show(struct device *dev,
242                                            struct device_attribute *dev_attr,
243                                            char *buf)
244 {
245         struct hv_device *hv_dev = device_to_hv_device(dev);
246
247         if (!hv_dev->channel)
248                 return -ENODEV;
249         return sprintf(buf, "%d\n",
250                        channel_pending(hv_dev->channel,
251                                        vmbus_connection.monitor_pages[1]));
252 }
253 static DEVICE_ATTR_RO(client_monitor_pending);
254
255 static ssize_t server_monitor_latency_show(struct device *dev,
256                                            struct device_attribute *dev_attr,
257                                            char *buf)
258 {
259         struct hv_device *hv_dev = device_to_hv_device(dev);
260
261         if (!hv_dev->channel)
262                 return -ENODEV;
263         return sprintf(buf, "%d\n",
264                        channel_latency(hv_dev->channel,
265                                        vmbus_connection.monitor_pages[0]));
266 }
267 static DEVICE_ATTR_RO(server_monitor_latency);
268
269 static ssize_t client_monitor_latency_show(struct device *dev,
270                                            struct device_attribute *dev_attr,
271                                            char *buf)
272 {
273         struct hv_device *hv_dev = device_to_hv_device(dev);
274
275         if (!hv_dev->channel)
276                 return -ENODEV;
277         return sprintf(buf, "%d\n",
278                        channel_latency(hv_dev->channel,
279                                        vmbus_connection.monitor_pages[1]));
280 }
281 static DEVICE_ATTR_RO(client_monitor_latency);
282
283 static ssize_t server_monitor_conn_id_show(struct device *dev,
284                                            struct device_attribute *dev_attr,
285                                            char *buf)
286 {
287         struct hv_device *hv_dev = device_to_hv_device(dev);
288
289         if (!hv_dev->channel)
290                 return -ENODEV;
291         return sprintf(buf, "%d\n",
292                        channel_conn_id(hv_dev->channel,
293                                        vmbus_connection.monitor_pages[0]));
294 }
295 static DEVICE_ATTR_RO(server_monitor_conn_id);
296
297 static ssize_t client_monitor_conn_id_show(struct device *dev,
298                                            struct device_attribute *dev_attr,
299                                            char *buf)
300 {
301         struct hv_device *hv_dev = device_to_hv_device(dev);
302
303         if (!hv_dev->channel)
304                 return -ENODEV;
305         return sprintf(buf, "%d\n",
306                        channel_conn_id(hv_dev->channel,
307                                        vmbus_connection.monitor_pages[1]));
308 }
309 static DEVICE_ATTR_RO(client_monitor_conn_id);
310
311 static ssize_t out_intr_mask_show(struct device *dev,
312                                   struct device_attribute *dev_attr, char *buf)
313 {
314         struct hv_device *hv_dev = device_to_hv_device(dev);
315         struct hv_ring_buffer_debug_info outbound;
316         int ret;
317
318         if (!hv_dev->channel)
319                 return -ENODEV;
320
321         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
322                                           &outbound);
323         if (ret < 0)
324                 return ret;
325
326         return sprintf(buf, "%d\n", outbound.current_interrupt_mask);
327 }
328 static DEVICE_ATTR_RO(out_intr_mask);
329
330 static ssize_t out_read_index_show(struct device *dev,
331                                    struct device_attribute *dev_attr, char *buf)
332 {
333         struct hv_device *hv_dev = device_to_hv_device(dev);
334         struct hv_ring_buffer_debug_info outbound;
335         int ret;
336
337         if (!hv_dev->channel)
338                 return -ENODEV;
339
340         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
341                                           &outbound);
342         if (ret < 0)
343                 return ret;
344         return sprintf(buf, "%d\n", outbound.current_read_index);
345 }
346 static DEVICE_ATTR_RO(out_read_index);
347
348 static ssize_t out_write_index_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         int ret;
355
356         if (!hv_dev->channel)
357                 return -ENODEV;
358
359         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
360                                           &outbound);
361         if (ret < 0)
362                 return ret;
363         return sprintf(buf, "%d\n", outbound.current_write_index);
364 }
365 static DEVICE_ATTR_RO(out_write_index);
366
367 static ssize_t out_read_bytes_avail_show(struct device *dev,
368                                          struct device_attribute *dev_attr,
369                                          char *buf)
370 {
371         struct hv_device *hv_dev = device_to_hv_device(dev);
372         struct hv_ring_buffer_debug_info outbound;
373         int ret;
374
375         if (!hv_dev->channel)
376                 return -ENODEV;
377
378         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
379                                           &outbound);
380         if (ret < 0)
381                 return ret;
382         return sprintf(buf, "%d\n", outbound.bytes_avail_toread);
383 }
384 static DEVICE_ATTR_RO(out_read_bytes_avail);
385
386 static ssize_t out_write_bytes_avail_show(struct device *dev,
387                                           struct device_attribute *dev_attr,
388                                           char *buf)
389 {
390         struct hv_device *hv_dev = device_to_hv_device(dev);
391         struct hv_ring_buffer_debug_info outbound;
392         int ret;
393
394         if (!hv_dev->channel)
395                 return -ENODEV;
396
397         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
398                                           &outbound);
399         if (ret < 0)
400                 return ret;
401         return sprintf(buf, "%d\n", outbound.bytes_avail_towrite);
402 }
403 static DEVICE_ATTR_RO(out_write_bytes_avail);
404
405 static ssize_t in_intr_mask_show(struct device *dev,
406                                  struct device_attribute *dev_attr, char *buf)
407 {
408         struct hv_device *hv_dev = device_to_hv_device(dev);
409         struct hv_ring_buffer_debug_info inbound;
410         int ret;
411
412         if (!hv_dev->channel)
413                 return -ENODEV;
414
415         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
416         if (ret < 0)
417                 return ret;
418
419         return sprintf(buf, "%d\n", inbound.current_interrupt_mask);
420 }
421 static DEVICE_ATTR_RO(in_intr_mask);
422
423 static ssize_t in_read_index_show(struct device *dev,
424                                   struct device_attribute *dev_attr, char *buf)
425 {
426         struct hv_device *hv_dev = device_to_hv_device(dev);
427         struct hv_ring_buffer_debug_info inbound;
428         int ret;
429
430         if (!hv_dev->channel)
431                 return -ENODEV;
432
433         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
434         if (ret < 0)
435                 return ret;
436
437         return sprintf(buf, "%d\n", inbound.current_read_index);
438 }
439 static DEVICE_ATTR_RO(in_read_index);
440
441 static ssize_t in_write_index_show(struct device *dev,
442                                    struct device_attribute *dev_attr, char *buf)
443 {
444         struct hv_device *hv_dev = device_to_hv_device(dev);
445         struct hv_ring_buffer_debug_info inbound;
446         int ret;
447
448         if (!hv_dev->channel)
449                 return -ENODEV;
450
451         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
452         if (ret < 0)
453                 return ret;
454
455         return sprintf(buf, "%d\n", inbound.current_write_index);
456 }
457 static DEVICE_ATTR_RO(in_write_index);
458
459 static ssize_t in_read_bytes_avail_show(struct device *dev,
460                                         struct device_attribute *dev_attr,
461                                         char *buf)
462 {
463         struct hv_device *hv_dev = device_to_hv_device(dev);
464         struct hv_ring_buffer_debug_info inbound;
465         int ret;
466
467         if (!hv_dev->channel)
468                 return -ENODEV;
469
470         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
471         if (ret < 0)
472                 return ret;
473
474         return sprintf(buf, "%d\n", inbound.bytes_avail_toread);
475 }
476 static DEVICE_ATTR_RO(in_read_bytes_avail);
477
478 static ssize_t in_write_bytes_avail_show(struct device *dev,
479                                          struct device_attribute *dev_attr,
480                                          char *buf)
481 {
482         struct hv_device *hv_dev = device_to_hv_device(dev);
483         struct hv_ring_buffer_debug_info inbound;
484         int ret;
485
486         if (!hv_dev->channel)
487                 return -ENODEV;
488
489         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
490         if (ret < 0)
491                 return ret;
492
493         return sprintf(buf, "%d\n", inbound.bytes_avail_towrite);
494 }
495 static DEVICE_ATTR_RO(in_write_bytes_avail);
496
497 static ssize_t channel_vp_mapping_show(struct device *dev,
498                                        struct device_attribute *dev_attr,
499                                        char *buf)
500 {
501         struct hv_device *hv_dev = device_to_hv_device(dev);
502         struct vmbus_channel *channel = hv_dev->channel, *cur_sc;
503         unsigned long flags;
504         int buf_size = PAGE_SIZE, n_written, tot_written;
505         struct list_head *cur;
506
507         if (!channel)
508                 return -ENODEV;
509
510         tot_written = snprintf(buf, buf_size, "%u:%u\n",
511                 channel->offermsg.child_relid, channel->target_cpu);
512
513         spin_lock_irqsave(&channel->lock, flags);
514
515         list_for_each(cur, &channel->sc_list) {
516                 if (tot_written >= buf_size - 1)
517                         break;
518
519                 cur_sc = list_entry(cur, struct vmbus_channel, sc_list);
520                 n_written = scnprintf(buf + tot_written,
521                                      buf_size - tot_written,
522                                      "%u:%u\n",
523                                      cur_sc->offermsg.child_relid,
524                                      cur_sc->target_cpu);
525                 tot_written += n_written;
526         }
527
528         spin_unlock_irqrestore(&channel->lock, flags);
529
530         return tot_written;
531 }
532 static DEVICE_ATTR_RO(channel_vp_mapping);
533
534 static ssize_t vendor_show(struct device *dev,
535                            struct device_attribute *dev_attr,
536                            char *buf)
537 {
538         struct hv_device *hv_dev = device_to_hv_device(dev);
539         return sprintf(buf, "0x%x\n", hv_dev->vendor_id);
540 }
541 static DEVICE_ATTR_RO(vendor);
542
543 static ssize_t device_show(struct device *dev,
544                            struct device_attribute *dev_attr,
545                            char *buf)
546 {
547         struct hv_device *hv_dev = device_to_hv_device(dev);
548         return sprintf(buf, "0x%x\n", hv_dev->device_id);
549 }
550 static DEVICE_ATTR_RO(device);
551
552 static ssize_t driver_override_store(struct device *dev,
553                                      struct device_attribute *attr,
554                                      const char *buf, size_t count)
555 {
556         struct hv_device *hv_dev = device_to_hv_device(dev);
557         char *driver_override, *old, *cp;
558
559         /* We need to keep extra room for a newline */
560         if (count >= (PAGE_SIZE - 1))
561                 return -EINVAL;
562
563         driver_override = kstrndup(buf, count, GFP_KERNEL);
564         if (!driver_override)
565                 return -ENOMEM;
566
567         cp = strchr(driver_override, '\n');
568         if (cp)
569                 *cp = '\0';
570
571         device_lock(dev);
572         old = hv_dev->driver_override;
573         if (strlen(driver_override)) {
574                 hv_dev->driver_override = driver_override;
575         } else {
576                 kfree(driver_override);
577                 hv_dev->driver_override = NULL;
578         }
579         device_unlock(dev);
580
581         kfree(old);
582
583         return count;
584 }
585
586 static ssize_t driver_override_show(struct device *dev,
587                                     struct device_attribute *attr, char *buf)
588 {
589         struct hv_device *hv_dev = device_to_hv_device(dev);
590         ssize_t len;
591
592         device_lock(dev);
593         len = snprintf(buf, PAGE_SIZE, "%s\n", hv_dev->driver_override);
594         device_unlock(dev);
595
596         return len;
597 }
598 static DEVICE_ATTR_RW(driver_override);
599
600 /* Set up per device attributes in /sys/bus/vmbus/devices/<bus device> */
601 static struct attribute *vmbus_dev_attrs[] = {
602         &dev_attr_id.attr,
603         &dev_attr_state.attr,
604         &dev_attr_monitor_id.attr,
605         &dev_attr_class_id.attr,
606         &dev_attr_device_id.attr,
607         &dev_attr_modalias.attr,
608 #ifdef CONFIG_NUMA
609         &dev_attr_numa_node.attr,
610 #endif
611         &dev_attr_server_monitor_pending.attr,
612         &dev_attr_client_monitor_pending.attr,
613         &dev_attr_server_monitor_latency.attr,
614         &dev_attr_client_monitor_latency.attr,
615         &dev_attr_server_monitor_conn_id.attr,
616         &dev_attr_client_monitor_conn_id.attr,
617         &dev_attr_out_intr_mask.attr,
618         &dev_attr_out_read_index.attr,
619         &dev_attr_out_write_index.attr,
620         &dev_attr_out_read_bytes_avail.attr,
621         &dev_attr_out_write_bytes_avail.attr,
622         &dev_attr_in_intr_mask.attr,
623         &dev_attr_in_read_index.attr,
624         &dev_attr_in_write_index.attr,
625         &dev_attr_in_read_bytes_avail.attr,
626         &dev_attr_in_write_bytes_avail.attr,
627         &dev_attr_channel_vp_mapping.attr,
628         &dev_attr_vendor.attr,
629         &dev_attr_device.attr,
630         &dev_attr_driver_override.attr,
631         NULL,
632 };
633 ATTRIBUTE_GROUPS(vmbus_dev);
634
635 /*
636  * vmbus_uevent - add uevent for our device
637  *
638  * This routine is invoked when a device is added or removed on the vmbus to
639  * generate a uevent to udev in the userspace. The udev will then look at its
640  * rule and the uevent generated here to load the appropriate driver
641  *
642  * The alias string will be of the form vmbus:guid where guid is the string
643  * representation of the device guid (each byte of the guid will be
644  * represented with two hex characters.
645  */
646 static int vmbus_uevent(struct device *device, struct kobj_uevent_env *env)
647 {
648         struct hv_device *dev = device_to_hv_device(device);
649         int ret;
650         char alias_name[VMBUS_ALIAS_LEN + 1];
651
652         print_alias_name(dev, alias_name);
653         ret = add_uevent_var(env, "MODALIAS=vmbus:%s", alias_name);
654         return ret;
655 }
656
657 static const uuid_le null_guid;
658
659 static inline bool is_null_guid(const uuid_le *guid)
660 {
661         if (uuid_le_cmp(*guid, null_guid))
662                 return false;
663         return true;
664 }
665
666 static const struct hv_vmbus_device_id *
667 hv_vmbus_dev_match(const struct hv_vmbus_device_id *id, const uuid_le *guid)
668
669 {
670         if (id == NULL)
671                 return NULL; /* empty device table */
672
673         for (; !is_null_guid(&id->guid); id++)
674                 if (!uuid_le_cmp(id->guid, *guid))
675                         return id;
676
677         return NULL;
678 }
679
680 static const struct hv_vmbus_device_id *
681 hv_vmbus_dynid_match(struct hv_driver *drv, const uuid_le *guid)
682 {
683         const struct hv_vmbus_device_id *id = NULL;
684         struct vmbus_dynid *dynid;
685
686         spin_lock(&drv->dynids.lock);
687         list_for_each_entry(dynid, &drv->dynids.list, node) {
688                 if (!uuid_le_cmp(dynid->id.guid, *guid)) {
689                         id = &dynid->id;
690                         break;
691                 }
692         }
693         spin_unlock(&drv->dynids.lock);
694
695         return id;
696 }
697
698 static const struct hv_vmbus_device_id vmbus_device_null = {
699         .guid = NULL_UUID_LE,
700 };
701
702 /*
703  * Return a matching hv_vmbus_device_id pointer.
704  * If there is no match, return NULL.
705  */
706 static const struct hv_vmbus_device_id *hv_vmbus_get_id(struct hv_driver *drv,
707                                                         struct hv_device *dev)
708 {
709         const uuid_le *guid = &dev->dev_type;
710         const struct hv_vmbus_device_id *id;
711
712         /* When driver_override is set, only bind to the matching driver */
713         if (dev->driver_override && strcmp(dev->driver_override, drv->name))
714                 return NULL;
715
716         /* Look at the dynamic ids first, before the static ones */
717         id = hv_vmbus_dynid_match(drv, guid);
718         if (!id)
719                 id = hv_vmbus_dev_match(drv->id_table, guid);
720
721         /* driver_override will always match, send a dummy id */
722         if (!id && dev->driver_override)
723                 id = &vmbus_device_null;
724
725         return id;
726 }
727
728 /* vmbus_add_dynid - add a new device ID to this driver and re-probe devices */
729 static int vmbus_add_dynid(struct hv_driver *drv, uuid_le *guid)
730 {
731         struct vmbus_dynid *dynid;
732
733         dynid = kzalloc(sizeof(*dynid), GFP_KERNEL);
734         if (!dynid)
735                 return -ENOMEM;
736
737         dynid->id.guid = *guid;
738
739         spin_lock(&drv->dynids.lock);
740         list_add_tail(&dynid->node, &drv->dynids.list);
741         spin_unlock(&drv->dynids.lock);
742
743         return driver_attach(&drv->driver);
744 }
745
746 static void vmbus_free_dynids(struct hv_driver *drv)
747 {
748         struct vmbus_dynid *dynid, *n;
749
750         spin_lock(&drv->dynids.lock);
751         list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
752                 list_del(&dynid->node);
753                 kfree(dynid);
754         }
755         spin_unlock(&drv->dynids.lock);
756 }
757
758 /*
759  * store_new_id - sysfs frontend to vmbus_add_dynid()
760  *
761  * Allow GUIDs to be added to an existing driver via sysfs.
762  */
763 static ssize_t new_id_store(struct device_driver *driver, const char *buf,
764                             size_t count)
765 {
766         struct hv_driver *drv = drv_to_hv_drv(driver);
767         uuid_le guid;
768         ssize_t retval;
769
770         retval = uuid_le_to_bin(buf, &guid);
771         if (retval)
772                 return retval;
773
774         if (hv_vmbus_dynid_match(drv, &guid))
775                 return -EEXIST;
776
777         retval = vmbus_add_dynid(drv, &guid);
778         if (retval)
779                 return retval;
780         return count;
781 }
782 static DRIVER_ATTR_WO(new_id);
783
784 /*
785  * store_remove_id - remove a PCI device ID from this driver
786  *
787  * Removes a dynamic pci device ID to this driver.
788  */
789 static ssize_t remove_id_store(struct device_driver *driver, const char *buf,
790                                size_t count)
791 {
792         struct hv_driver *drv = drv_to_hv_drv(driver);
793         struct vmbus_dynid *dynid, *n;
794         uuid_le guid;
795         ssize_t retval;
796
797         retval = uuid_le_to_bin(buf, &guid);
798         if (retval)
799                 return retval;
800
801         retval = -ENODEV;
802         spin_lock(&drv->dynids.lock);
803         list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
804                 struct hv_vmbus_device_id *id = &dynid->id;
805
806                 if (!uuid_le_cmp(id->guid, guid)) {
807                         list_del(&dynid->node);
808                         kfree(dynid);
809                         retval = count;
810                         break;
811                 }
812         }
813         spin_unlock(&drv->dynids.lock);
814
815         return retval;
816 }
817 static DRIVER_ATTR_WO(remove_id);
818
819 static struct attribute *vmbus_drv_attrs[] = {
820         &driver_attr_new_id.attr,
821         &driver_attr_remove_id.attr,
822         NULL,
823 };
824 ATTRIBUTE_GROUPS(vmbus_drv);
825
826
827 /*
828  * vmbus_match - Attempt to match the specified device to the specified driver
829  */
830 static int vmbus_match(struct device *device, struct device_driver *driver)
831 {
832         struct hv_driver *drv = drv_to_hv_drv(driver);
833         struct hv_device *hv_dev = device_to_hv_device(device);
834
835         /* The hv_sock driver handles all hv_sock offers. */
836         if (is_hvsock_channel(hv_dev->channel))
837                 return drv->hvsock;
838
839         if (hv_vmbus_get_id(drv, hv_dev))
840                 return 1;
841
842         return 0;
843 }
844
845 /*
846  * vmbus_probe - Add the new vmbus's child device
847  */
848 static int vmbus_probe(struct device *child_device)
849 {
850         int ret = 0;
851         struct hv_driver *drv =
852                         drv_to_hv_drv(child_device->driver);
853         struct hv_device *dev = device_to_hv_device(child_device);
854         const struct hv_vmbus_device_id *dev_id;
855
856         dev_id = hv_vmbus_get_id(drv, dev);
857         if (drv->probe) {
858                 ret = drv->probe(dev, dev_id);
859                 if (ret != 0)
860                         pr_err("probe failed for device %s (%d)\n",
861                                dev_name(child_device), ret);
862
863         } else {
864                 pr_err("probe not set for driver %s\n",
865                        dev_name(child_device));
866                 ret = -ENODEV;
867         }
868         return ret;
869 }
870
871 /*
872  * vmbus_remove - Remove a vmbus device
873  */
874 static int vmbus_remove(struct device *child_device)
875 {
876         struct hv_driver *drv;
877         struct hv_device *dev = device_to_hv_device(child_device);
878
879         if (child_device->driver) {
880                 drv = drv_to_hv_drv(child_device->driver);
881                 if (drv->remove)
882                         drv->remove(dev);
883         }
884
885         return 0;
886 }
887
888
889 /*
890  * vmbus_shutdown - Shutdown a vmbus device
891  */
892 static void vmbus_shutdown(struct device *child_device)
893 {
894         struct hv_driver *drv;
895         struct hv_device *dev = device_to_hv_device(child_device);
896
897
898         /* The device may not be attached yet */
899         if (!child_device->driver)
900                 return;
901
902         drv = drv_to_hv_drv(child_device->driver);
903
904         if (drv->shutdown)
905                 drv->shutdown(dev);
906 }
907
908
909 /*
910  * vmbus_device_release - Final callback release of the vmbus child device
911  */
912 static void vmbus_device_release(struct device *device)
913 {
914         struct hv_device *hv_dev = device_to_hv_device(device);
915         struct vmbus_channel *channel = hv_dev->channel;
916
917         mutex_lock(&vmbus_connection.channel_mutex);
918         hv_process_channel_removal(channel);
919         mutex_unlock(&vmbus_connection.channel_mutex);
920         kfree(hv_dev);
921 }
922
923 /* The one and only one */
924 static struct bus_type  hv_bus = {
925         .name =         "vmbus",
926         .match =                vmbus_match,
927         .shutdown =             vmbus_shutdown,
928         .remove =               vmbus_remove,
929         .probe =                vmbus_probe,
930         .uevent =               vmbus_uevent,
931         .dev_groups =           vmbus_dev_groups,
932         .drv_groups =           vmbus_drv_groups,
933 };
934
935 struct onmessage_work_context {
936         struct work_struct work;
937         struct hv_message msg;
938 };
939
940 static void vmbus_onmessage_work(struct work_struct *work)
941 {
942         struct onmessage_work_context *ctx;
943
944         /* Do not process messages if we're in DISCONNECTED state */
945         if (vmbus_connection.conn_state == DISCONNECTED)
946                 return;
947
948         ctx = container_of(work, struct onmessage_work_context,
949                            work);
950         vmbus_onmessage(&ctx->msg);
951         kfree(ctx);
952 }
953
954 static void hv_process_timer_expiration(struct hv_message *msg,
955                                         struct hv_per_cpu_context *hv_cpu)
956 {
957         struct clock_event_device *dev = hv_cpu->clk_evt;
958
959         if (dev->event_handler)
960                 dev->event_handler(dev);
961
962         vmbus_signal_eom(msg, HVMSG_TIMER_EXPIRED);
963 }
964
965 void vmbus_on_msg_dpc(unsigned long data)
966 {
967         struct hv_per_cpu_context *hv_cpu = (void *)data;
968         void *page_addr = hv_cpu->synic_message_page;
969         struct hv_message *msg = (struct hv_message *)page_addr +
970                                   VMBUS_MESSAGE_SINT;
971         struct vmbus_channel_message_header *hdr;
972         const struct vmbus_channel_message_table_entry *entry;
973         struct onmessage_work_context *ctx;
974         u32 message_type = msg->header.message_type;
975
976         if (message_type == HVMSG_NONE)
977                 /* no msg */
978                 return;
979
980         hdr = (struct vmbus_channel_message_header *)msg->u.payload;
981
982         trace_vmbus_on_msg_dpc(hdr);
983
984         if (hdr->msgtype >= CHANNELMSG_COUNT) {
985                 WARN_ONCE(1, "unknown msgtype=%d\n", hdr->msgtype);
986                 goto msg_handled;
987         }
988
989         entry = &channel_message_table[hdr->msgtype];
990         if (entry->handler_type == VMHT_BLOCKING) {
991                 ctx = kmalloc(sizeof(*ctx), GFP_ATOMIC);
992                 if (ctx == NULL)
993                         return;
994
995                 INIT_WORK(&ctx->work, vmbus_onmessage_work);
996                 memcpy(&ctx->msg, msg, sizeof(*msg));
997
998                 /*
999                  * The host can generate a rescind message while we
1000                  * may still be handling the original offer. We deal with
1001                  * this condition by ensuring the processing is done on the
1002                  * same CPU.
1003                  */
1004                 switch (hdr->msgtype) {
1005                 case CHANNELMSG_RESCIND_CHANNELOFFER:
1006                         /*
1007                          * If we are handling the rescind message;
1008                          * schedule the work on the global work queue.
1009                          */
1010                         schedule_work_on(vmbus_connection.connect_cpu,
1011                                          &ctx->work);
1012                         break;
1013
1014                 case CHANNELMSG_OFFERCHANNEL:
1015                         atomic_inc(&vmbus_connection.offer_in_progress);
1016                         queue_work_on(vmbus_connection.connect_cpu,
1017                                       vmbus_connection.work_queue,
1018                                       &ctx->work);
1019                         break;
1020
1021                 default:
1022                         queue_work(vmbus_connection.work_queue, &ctx->work);
1023                 }
1024         } else
1025                 entry->message_handler(hdr);
1026
1027 msg_handled:
1028         vmbus_signal_eom(msg, message_type);
1029 }
1030
1031
1032 /*
1033  * Direct callback for channels using other deferred processing
1034  */
1035 static void vmbus_channel_isr(struct vmbus_channel *channel)
1036 {
1037         void (*callback_fn)(void *);
1038
1039         callback_fn = READ_ONCE(channel->onchannel_callback);
1040         if (likely(callback_fn != NULL))
1041                 (*callback_fn)(channel->channel_callback_context);
1042 }
1043
1044 /*
1045  * Schedule all channels with events pending
1046  */
1047 static void vmbus_chan_sched(struct hv_per_cpu_context *hv_cpu)
1048 {
1049         unsigned long *recv_int_page;
1050         u32 maxbits, relid;
1051
1052         if (vmbus_proto_version < VERSION_WIN8) {
1053                 maxbits = MAX_NUM_CHANNELS_SUPPORTED;
1054                 recv_int_page = vmbus_connection.recv_int_page;
1055         } else {
1056                 /*
1057                  * When the host is win8 and beyond, the event page
1058                  * can be directly checked to get the id of the channel
1059                  * that has the interrupt pending.
1060                  */
1061                 void *page_addr = hv_cpu->synic_event_page;
1062                 union hv_synic_event_flags *event
1063                         = (union hv_synic_event_flags *)page_addr +
1064                                                  VMBUS_MESSAGE_SINT;
1065
1066                 maxbits = HV_EVENT_FLAGS_COUNT;
1067                 recv_int_page = event->flags;
1068         }
1069
1070         if (unlikely(!recv_int_page))
1071                 return;
1072
1073         for_each_set_bit(relid, recv_int_page, maxbits) {
1074                 struct vmbus_channel *channel;
1075
1076                 if (!sync_test_and_clear_bit(relid, recv_int_page))
1077                         continue;
1078
1079                 /* Special case - vmbus channel protocol msg */
1080                 if (relid == 0)
1081                         continue;
1082
1083                 rcu_read_lock();
1084
1085                 /* Find channel based on relid */
1086                 list_for_each_entry_rcu(channel, &hv_cpu->chan_list, percpu_list) {
1087                         if (channel->offermsg.child_relid != relid)
1088                                 continue;
1089
1090                         if (channel->rescind)
1091                                 continue;
1092
1093                         trace_vmbus_chan_sched(channel);
1094
1095                         ++channel->interrupts;
1096
1097                         switch (channel->callback_mode) {
1098                         case HV_CALL_ISR:
1099                                 vmbus_channel_isr(channel);
1100                                 break;
1101
1102                         case HV_CALL_BATCHED:
1103                                 hv_begin_read(&channel->inbound);
1104                                 /* fallthrough */
1105                         case HV_CALL_DIRECT:
1106                                 tasklet_schedule(&channel->callback_event);
1107                         }
1108                 }
1109
1110                 rcu_read_unlock();
1111         }
1112 }
1113
1114 static void vmbus_isr(void)
1115 {
1116         struct hv_per_cpu_context *hv_cpu
1117                 = this_cpu_ptr(hv_context.cpu_context);
1118         void *page_addr = hv_cpu->synic_event_page;
1119         struct hv_message *msg;
1120         union hv_synic_event_flags *event;
1121         bool handled = false;
1122
1123         if (unlikely(page_addr == NULL))
1124                 return;
1125
1126         event = (union hv_synic_event_flags *)page_addr +
1127                                          VMBUS_MESSAGE_SINT;
1128         /*
1129          * Check for events before checking for messages. This is the order
1130          * in which events and messages are checked in Windows guests on
1131          * Hyper-V, and the Windows team suggested we do the same.
1132          */
1133
1134         if ((vmbus_proto_version == VERSION_WS2008) ||
1135                 (vmbus_proto_version == VERSION_WIN7)) {
1136
1137                 /* Since we are a child, we only need to check bit 0 */
1138                 if (sync_test_and_clear_bit(0, event->flags))
1139                         handled = true;
1140         } else {
1141                 /*
1142                  * Our host is win8 or above. The signaling mechanism
1143                  * has changed and we can directly look at the event page.
1144                  * If bit n is set then we have an interrup on the channel
1145                  * whose id is n.
1146                  */
1147                 handled = true;
1148         }
1149
1150         if (handled)
1151                 vmbus_chan_sched(hv_cpu);
1152
1153         page_addr = hv_cpu->synic_message_page;
1154         msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
1155
1156         /* Check if there are actual msgs to be processed */
1157         if (msg->header.message_type != HVMSG_NONE) {
1158                 if (msg->header.message_type == HVMSG_TIMER_EXPIRED)
1159                         hv_process_timer_expiration(msg, hv_cpu);
1160                 else
1161                         tasklet_schedule(&hv_cpu->msg_dpc);
1162         }
1163
1164         add_interrupt_randomness(HYPERVISOR_CALLBACK_VECTOR, 0);
1165 }
1166
1167 /*
1168  * Boolean to control whether to report panic messages over Hyper-V.
1169  *
1170  * It can be set via /proc/sys/kernel/hyperv/record_panic_msg
1171  */
1172 static int sysctl_record_panic_msg = 1;
1173
1174 /*
1175  * Callback from kmsg_dump. Grab as much as possible from the end of the kmsg
1176  * buffer and call into Hyper-V to transfer the data.
1177  */
1178 static void hv_kmsg_dump(struct kmsg_dumper *dumper,
1179                          enum kmsg_dump_reason reason)
1180 {
1181         size_t bytes_written;
1182         phys_addr_t panic_pa;
1183
1184         /* We are only interested in panics. */
1185         if ((reason != KMSG_DUMP_PANIC) || (!sysctl_record_panic_msg))
1186                 return;
1187
1188         panic_pa = virt_to_phys(hv_panic_page);
1189
1190         /*
1191          * Write dump contents to the page. No need to synchronize; panic should
1192          * be single-threaded.
1193          */
1194         kmsg_dump_get_buffer(dumper, true, hv_panic_page, PAGE_SIZE,
1195                              &bytes_written);
1196         if (bytes_written)
1197                 hyperv_report_panic_msg(panic_pa, bytes_written);
1198 }
1199
1200 static struct kmsg_dumper hv_kmsg_dumper = {
1201         .dump = hv_kmsg_dump,
1202 };
1203
1204 static struct ctl_table_header *hv_ctl_table_hdr;
1205 static int zero;
1206 static int one = 1;
1207
1208 /*
1209  * sysctl option to allow the user to control whether kmsg data should be
1210  * reported to Hyper-V on panic.
1211  */
1212 static struct ctl_table hv_ctl_table[] = {
1213         {
1214                 .procname       = "hyperv_record_panic_msg",
1215                 .data           = &sysctl_record_panic_msg,
1216                 .maxlen         = sizeof(int),
1217                 .mode           = 0644,
1218                 .proc_handler   = proc_dointvec_minmax,
1219                 .extra1         = &zero,
1220                 .extra2         = &one
1221         },
1222         {}
1223 };
1224
1225 static struct ctl_table hv_root_table[] = {
1226         {
1227                 .procname       = "kernel",
1228                 .mode           = 0555,
1229                 .child          = hv_ctl_table
1230         },
1231         {}
1232 };
1233
1234 /*
1235  * vmbus_bus_init -Main vmbus driver initialization routine.
1236  *
1237  * Here, we
1238  *      - initialize the vmbus driver context
1239  *      - invoke the vmbus hv main init routine
1240  *      - retrieve the channel offers
1241  */
1242 static int vmbus_bus_init(void)
1243 {
1244         int ret;
1245
1246         /* Hypervisor initialization...setup hypercall page..etc */
1247         ret = hv_init();
1248         if (ret != 0) {
1249                 pr_err("Unable to initialize the hypervisor - 0x%x\n", ret);
1250                 return ret;
1251         }
1252
1253         ret = bus_register(&hv_bus);
1254         if (ret)
1255                 return ret;
1256
1257         hv_setup_vmbus_irq(vmbus_isr);
1258
1259         ret = hv_synic_alloc();
1260         if (ret)
1261                 goto err_alloc;
1262         /*
1263          * Initialize the per-cpu interrupt state and
1264          * connect to the host.
1265          */
1266         ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "hyperv/vmbus:online",
1267                                 hv_synic_init, hv_synic_cleanup);
1268         if (ret < 0)
1269                 goto err_alloc;
1270         hyperv_cpuhp_online = ret;
1271
1272         ret = vmbus_connect();
1273         if (ret)
1274                 goto err_connect;
1275
1276         /*
1277          * Only register if the crash MSRs are available
1278          */
1279         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1280                 u64 hyperv_crash_ctl;
1281                 /*
1282                  * Sysctl registration is not fatal, since by default
1283                  * reporting is enabled.
1284                  */
1285                 hv_ctl_table_hdr = register_sysctl_table(hv_root_table);
1286                 if (!hv_ctl_table_hdr)
1287                         pr_err("Hyper-V: sysctl table register error");
1288
1289                 /*
1290                  * Register for panic kmsg callback only if the right
1291                  * capability is supported by the hypervisor.
1292                  */
1293                 hv_get_crash_ctl(hyperv_crash_ctl);
1294                 if (hyperv_crash_ctl & HV_CRASH_CTL_CRASH_NOTIFY_MSG) {
1295                         hv_panic_page = (void *)get_zeroed_page(GFP_KERNEL);
1296                         if (hv_panic_page) {
1297                                 ret = kmsg_dump_register(&hv_kmsg_dumper);
1298                                 if (ret)
1299                                         pr_err("Hyper-V: kmsg dump register "
1300                                                 "error 0x%x\n", ret);
1301                         } else
1302                                 pr_err("Hyper-V: panic message page memory "
1303                                         "allocation failed");
1304                 }
1305
1306                 register_die_notifier(&hyperv_die_block);
1307                 atomic_notifier_chain_register(&panic_notifier_list,
1308                                                &hyperv_panic_block);
1309         }
1310
1311         vmbus_request_offers();
1312
1313         return 0;
1314
1315 err_connect:
1316         cpuhp_remove_state(hyperv_cpuhp_online);
1317 err_alloc:
1318         hv_synic_free();
1319         hv_remove_vmbus_irq();
1320
1321         bus_unregister(&hv_bus);
1322         free_page((unsigned long)hv_panic_page);
1323         unregister_sysctl_table(hv_ctl_table_hdr);
1324         hv_ctl_table_hdr = NULL;
1325         return ret;
1326 }
1327
1328 /**
1329  * __vmbus_child_driver_register() - Register a vmbus's driver
1330  * @hv_driver: Pointer to driver structure you want to register
1331  * @owner: owner module of the drv
1332  * @mod_name: module name string
1333  *
1334  * Registers the given driver with Linux through the 'driver_register()' call
1335  * and sets up the hyper-v vmbus handling for this driver.
1336  * It will return the state of the 'driver_register()' call.
1337  *
1338  */
1339 int __vmbus_driver_register(struct hv_driver *hv_driver, struct module *owner, const char *mod_name)
1340 {
1341         int ret;
1342
1343         pr_info("registering driver %s\n", hv_driver->name);
1344
1345         ret = vmbus_exists();
1346         if (ret < 0)
1347                 return ret;
1348
1349         hv_driver->driver.name = hv_driver->name;
1350         hv_driver->driver.owner = owner;
1351         hv_driver->driver.mod_name = mod_name;
1352         hv_driver->driver.bus = &hv_bus;
1353
1354         spin_lock_init(&hv_driver->dynids.lock);
1355         INIT_LIST_HEAD(&hv_driver->dynids.list);
1356
1357         ret = driver_register(&hv_driver->driver);
1358
1359         return ret;
1360 }
1361 EXPORT_SYMBOL_GPL(__vmbus_driver_register);
1362
1363 /**
1364  * vmbus_driver_unregister() - Unregister a vmbus's driver
1365  * @hv_driver: Pointer to driver structure you want to
1366  *             un-register
1367  *
1368  * Un-register the given driver that was previous registered with a call to
1369  * vmbus_driver_register()
1370  */
1371 void vmbus_driver_unregister(struct hv_driver *hv_driver)
1372 {
1373         pr_info("unregistering driver %s\n", hv_driver->name);
1374
1375         if (!vmbus_exists()) {
1376                 driver_unregister(&hv_driver->driver);
1377                 vmbus_free_dynids(hv_driver);
1378         }
1379 }
1380 EXPORT_SYMBOL_GPL(vmbus_driver_unregister);
1381
1382
1383 /*
1384  * Called when last reference to channel is gone.
1385  */
1386 static void vmbus_chan_release(struct kobject *kobj)
1387 {
1388         struct vmbus_channel *channel
1389                 = container_of(kobj, struct vmbus_channel, kobj);
1390
1391         kfree_rcu(channel, rcu);
1392 }
1393
1394 struct vmbus_chan_attribute {
1395         struct attribute attr;
1396         ssize_t (*show)(const struct vmbus_channel *chan, char *buf);
1397         ssize_t (*store)(struct vmbus_channel *chan,
1398                          const char *buf, size_t count);
1399 };
1400 #define VMBUS_CHAN_ATTR(_name, _mode, _show, _store) \
1401         struct vmbus_chan_attribute chan_attr_##_name \
1402                 = __ATTR(_name, _mode, _show, _store)
1403 #define VMBUS_CHAN_ATTR_RW(_name) \
1404         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RW(_name)
1405 #define VMBUS_CHAN_ATTR_RO(_name) \
1406         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RO(_name)
1407 #define VMBUS_CHAN_ATTR_WO(_name) \
1408         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_WO(_name)
1409
1410 static ssize_t vmbus_chan_attr_show(struct kobject *kobj,
1411                                     struct attribute *attr, char *buf)
1412 {
1413         const struct vmbus_chan_attribute *attribute
1414                 = container_of(attr, struct vmbus_chan_attribute, attr);
1415         const struct vmbus_channel *chan
1416                 = container_of(kobj, struct vmbus_channel, kobj);
1417
1418         if (!attribute->show)
1419                 return -EIO;
1420
1421         if (chan->state != CHANNEL_OPENED_STATE)
1422                 return -EINVAL;
1423
1424         return attribute->show(chan, buf);
1425 }
1426
1427 static const struct sysfs_ops vmbus_chan_sysfs_ops = {
1428         .show = vmbus_chan_attr_show,
1429 };
1430
1431 static ssize_t out_mask_show(const struct vmbus_channel *channel, char *buf)
1432 {
1433         const struct hv_ring_buffer_info *rbi = &channel->outbound;
1434
1435         return sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1436 }
1437 static VMBUS_CHAN_ATTR_RO(out_mask);
1438
1439 static ssize_t in_mask_show(const struct vmbus_channel *channel, char *buf)
1440 {
1441         const struct hv_ring_buffer_info *rbi = &channel->inbound;
1442
1443         return sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1444 }
1445 static VMBUS_CHAN_ATTR_RO(in_mask);
1446
1447 static ssize_t read_avail_show(const struct vmbus_channel *channel, char *buf)
1448 {
1449         const struct hv_ring_buffer_info *rbi = &channel->inbound;
1450
1451         return sprintf(buf, "%u\n", hv_get_bytes_to_read(rbi));
1452 }
1453 static VMBUS_CHAN_ATTR_RO(read_avail);
1454
1455 static ssize_t write_avail_show(const struct vmbus_channel *channel, char *buf)
1456 {
1457         const struct hv_ring_buffer_info *rbi = &channel->outbound;
1458
1459         return sprintf(buf, "%u\n", hv_get_bytes_to_write(rbi));
1460 }
1461 static VMBUS_CHAN_ATTR_RO(write_avail);
1462
1463 static ssize_t show_target_cpu(const struct vmbus_channel *channel, char *buf)
1464 {
1465         return sprintf(buf, "%u\n", channel->target_cpu);
1466 }
1467 static VMBUS_CHAN_ATTR(cpu, S_IRUGO, show_target_cpu, NULL);
1468
1469 static ssize_t channel_pending_show(const struct vmbus_channel *channel,
1470                                     char *buf)
1471 {
1472         return sprintf(buf, "%d\n",
1473                        channel_pending(channel,
1474                                        vmbus_connection.monitor_pages[1]));
1475 }
1476 static VMBUS_CHAN_ATTR(pending, S_IRUGO, channel_pending_show, NULL);
1477
1478 static ssize_t channel_latency_show(const struct vmbus_channel *channel,
1479                                     char *buf)
1480 {
1481         return sprintf(buf, "%d\n",
1482                        channel_latency(channel,
1483                                        vmbus_connection.monitor_pages[1]));
1484 }
1485 static VMBUS_CHAN_ATTR(latency, S_IRUGO, channel_latency_show, NULL);
1486
1487 static ssize_t channel_interrupts_show(const struct vmbus_channel *channel, char *buf)
1488 {
1489         return sprintf(buf, "%llu\n", channel->interrupts);
1490 }
1491 static VMBUS_CHAN_ATTR(interrupts, S_IRUGO, channel_interrupts_show, NULL);
1492
1493 static ssize_t channel_events_show(const struct vmbus_channel *channel, char *buf)
1494 {
1495         return sprintf(buf, "%llu\n", channel->sig_events);
1496 }
1497 static VMBUS_CHAN_ATTR(events, S_IRUGO, channel_events_show, NULL);
1498
1499 static ssize_t subchannel_monitor_id_show(const struct vmbus_channel *channel,
1500                                           char *buf)
1501 {
1502         return sprintf(buf, "%u\n", channel->offermsg.monitorid);
1503 }
1504 static VMBUS_CHAN_ATTR(monitor_id, S_IRUGO, subchannel_monitor_id_show, NULL);
1505
1506 static ssize_t subchannel_id_show(const struct vmbus_channel *channel,
1507                                   char *buf)
1508 {
1509         return sprintf(buf, "%u\n",
1510                        channel->offermsg.offer.sub_channel_index);
1511 }
1512 static VMBUS_CHAN_ATTR_RO(subchannel_id);
1513
1514 static struct attribute *vmbus_chan_attrs[] = {
1515         &chan_attr_out_mask.attr,
1516         &chan_attr_in_mask.attr,
1517         &chan_attr_read_avail.attr,
1518         &chan_attr_write_avail.attr,
1519         &chan_attr_cpu.attr,
1520         &chan_attr_pending.attr,
1521         &chan_attr_latency.attr,
1522         &chan_attr_interrupts.attr,
1523         &chan_attr_events.attr,
1524         &chan_attr_monitor_id.attr,
1525         &chan_attr_subchannel_id.attr,
1526         NULL
1527 };
1528
1529 static struct kobj_type vmbus_chan_ktype = {
1530         .sysfs_ops = &vmbus_chan_sysfs_ops,
1531         .release = vmbus_chan_release,
1532         .default_attrs = vmbus_chan_attrs,
1533 };
1534
1535 /*
1536  * vmbus_add_channel_kobj - setup a sub-directory under device/channels
1537  */
1538 int vmbus_add_channel_kobj(struct hv_device *dev, struct vmbus_channel *channel)
1539 {
1540         struct kobject *kobj = &channel->kobj;
1541         u32 relid = channel->offermsg.child_relid;
1542         int ret;
1543
1544         kobj->kset = dev->channels_kset;
1545         ret = kobject_init_and_add(kobj, &vmbus_chan_ktype, NULL,
1546                                    "%u", relid);
1547         if (ret)
1548                 return ret;
1549
1550         kobject_uevent(kobj, KOBJ_ADD);
1551
1552         return 0;
1553 }
1554
1555 /*
1556  * vmbus_device_create - Creates and registers a new child device
1557  * on the vmbus.
1558  */
1559 struct hv_device *vmbus_device_create(const uuid_le *type,
1560                                       const uuid_le *instance,
1561                                       struct vmbus_channel *channel)
1562 {
1563         struct hv_device *child_device_obj;
1564
1565         child_device_obj = kzalloc(sizeof(struct hv_device), GFP_KERNEL);
1566         if (!child_device_obj) {
1567                 pr_err("Unable to allocate device object for child device\n");
1568                 return NULL;
1569         }
1570
1571         child_device_obj->channel = channel;
1572         memcpy(&child_device_obj->dev_type, type, sizeof(uuid_le));
1573         memcpy(&child_device_obj->dev_instance, instance,
1574                sizeof(uuid_le));
1575         child_device_obj->vendor_id = 0x1414; /* MSFT vendor ID */
1576
1577
1578         return child_device_obj;
1579 }
1580
1581 /*
1582  * vmbus_device_register - Register the child device
1583  */
1584 int vmbus_device_register(struct hv_device *child_device_obj)
1585 {
1586         struct kobject *kobj = &child_device_obj->device.kobj;
1587         int ret;
1588
1589         dev_set_name(&child_device_obj->device, "%pUl",
1590                      child_device_obj->channel->offermsg.offer.if_instance.b);
1591
1592         child_device_obj->device.bus = &hv_bus;
1593         child_device_obj->device.parent = &hv_acpi_dev->dev;
1594         child_device_obj->device.release = vmbus_device_release;
1595
1596         /*
1597          * Register with the LDM. This will kick off the driver/device
1598          * binding...which will eventually call vmbus_match() and vmbus_probe()
1599          */
1600         ret = device_register(&child_device_obj->device);
1601         if (ret) {
1602                 pr_err("Unable to register child device\n");
1603                 return ret;
1604         }
1605
1606         child_device_obj->channels_kset = kset_create_and_add("channels",
1607                                                               NULL, kobj);
1608         if (!child_device_obj->channels_kset) {
1609                 ret = -ENOMEM;
1610                 goto err_dev_unregister;
1611         }
1612
1613         ret = vmbus_add_channel_kobj(child_device_obj,
1614                                      child_device_obj->channel);
1615         if (ret) {
1616                 pr_err("Unable to register primary channeln");
1617                 goto err_kset_unregister;
1618         }
1619
1620         return 0;
1621
1622 err_kset_unregister:
1623         kset_unregister(child_device_obj->channels_kset);
1624
1625 err_dev_unregister:
1626         device_unregister(&child_device_obj->device);
1627         return ret;
1628 }
1629
1630 /*
1631  * vmbus_device_unregister - Remove the specified child device
1632  * from the vmbus.
1633  */
1634 void vmbus_device_unregister(struct hv_device *device_obj)
1635 {
1636         pr_debug("child device %s unregistered\n",
1637                 dev_name(&device_obj->device));
1638
1639         kset_unregister(device_obj->channels_kset);
1640
1641         /*
1642          * Kick off the process of unregistering the device.
1643          * This will call vmbus_remove() and eventually vmbus_device_release()
1644          */
1645         device_unregister(&device_obj->device);
1646 }
1647
1648
1649 /*
1650  * VMBUS is an acpi enumerated device. Get the information we
1651  * need from DSDT.
1652  */
1653 #define VTPM_BASE_ADDRESS 0xfed40000
1654 static acpi_status vmbus_walk_resources(struct acpi_resource *res, void *ctx)
1655 {
1656         resource_size_t start = 0;
1657         resource_size_t end = 0;
1658         struct resource *new_res;
1659         struct resource **old_res = &hyperv_mmio;
1660         struct resource **prev_res = NULL;
1661
1662         switch (res->type) {
1663
1664         /*
1665          * "Address" descriptors are for bus windows. Ignore
1666          * "memory" descriptors, which are for registers on
1667          * devices.
1668          */
1669         case ACPI_RESOURCE_TYPE_ADDRESS32:
1670                 start = res->data.address32.address.minimum;
1671                 end = res->data.address32.address.maximum;
1672                 break;
1673
1674         case ACPI_RESOURCE_TYPE_ADDRESS64:
1675                 start = res->data.address64.address.minimum;
1676                 end = res->data.address64.address.maximum;
1677                 break;
1678
1679         default:
1680                 /* Unused resource type */
1681                 return AE_OK;
1682
1683         }
1684         /*
1685          * Ignore ranges that are below 1MB, as they're not
1686          * necessary or useful here.
1687          */
1688         if (end < 0x100000)
1689                 return AE_OK;
1690
1691         new_res = kzalloc(sizeof(*new_res), GFP_ATOMIC);
1692         if (!new_res)
1693                 return AE_NO_MEMORY;
1694
1695         /* If this range overlaps the virtual TPM, truncate it. */
1696         if (end > VTPM_BASE_ADDRESS && start < VTPM_BASE_ADDRESS)
1697                 end = VTPM_BASE_ADDRESS;
1698
1699         new_res->name = "hyperv mmio";
1700         new_res->flags = IORESOURCE_MEM;
1701         new_res->start = start;
1702         new_res->end = end;
1703
1704         /*
1705          * If two ranges are adjacent, merge them.
1706          */
1707         do {
1708                 if (!*old_res) {
1709                         *old_res = new_res;
1710                         break;
1711                 }
1712
1713                 if (((*old_res)->end + 1) == new_res->start) {
1714                         (*old_res)->end = new_res->end;
1715                         kfree(new_res);
1716                         break;
1717                 }
1718
1719                 if ((*old_res)->start == new_res->end + 1) {
1720                         (*old_res)->start = new_res->start;
1721                         kfree(new_res);
1722                         break;
1723                 }
1724
1725                 if ((*old_res)->start > new_res->end) {
1726                         new_res->sibling = *old_res;
1727                         if (prev_res)
1728                                 (*prev_res)->sibling = new_res;
1729                         *old_res = new_res;
1730                         break;
1731                 }
1732
1733                 prev_res = old_res;
1734                 old_res = &(*old_res)->sibling;
1735
1736         } while (1);
1737
1738         return AE_OK;
1739 }
1740
1741 static int vmbus_acpi_remove(struct acpi_device *device)
1742 {
1743         struct resource *cur_res;
1744         struct resource *next_res;
1745
1746         if (hyperv_mmio) {
1747                 if (fb_mmio) {
1748                         __release_region(hyperv_mmio, fb_mmio->start,
1749                                          resource_size(fb_mmio));
1750                         fb_mmio = NULL;
1751                 }
1752
1753                 for (cur_res = hyperv_mmio; cur_res; cur_res = next_res) {
1754                         next_res = cur_res->sibling;
1755                         kfree(cur_res);
1756                 }
1757         }
1758
1759         return 0;
1760 }
1761
1762 static void vmbus_reserve_fb(void)
1763 {
1764         int size;
1765         /*
1766          * Make a claim for the frame buffer in the resource tree under the
1767          * first node, which will be the one below 4GB.  The length seems to
1768          * be underreported, particularly in a Generation 1 VM.  So start out
1769          * reserving a larger area and make it smaller until it succeeds.
1770          */
1771
1772         if (screen_info.lfb_base) {
1773                 if (efi_enabled(EFI_BOOT))
1774                         size = max_t(__u32, screen_info.lfb_size, 0x800000);
1775                 else
1776                         size = max_t(__u32, screen_info.lfb_size, 0x4000000);
1777
1778                 for (; !fb_mmio && (size >= 0x100000); size >>= 1) {
1779                         fb_mmio = __request_region(hyperv_mmio,
1780                                                    screen_info.lfb_base, size,
1781                                                    fb_mmio_name, 0);
1782                 }
1783         }
1784 }
1785
1786 /**
1787  * vmbus_allocate_mmio() - Pick a memory-mapped I/O range.
1788  * @new:                If successful, supplied a pointer to the
1789  *                      allocated MMIO space.
1790  * @device_obj:         Identifies the caller
1791  * @min:                Minimum guest physical address of the
1792  *                      allocation
1793  * @max:                Maximum guest physical address
1794  * @size:               Size of the range to be allocated
1795  * @align:              Alignment of the range to be allocated
1796  * @fb_overlap_ok:      Whether this allocation can be allowed
1797  *                      to overlap the video frame buffer.
1798  *
1799  * This function walks the resources granted to VMBus by the
1800  * _CRS object in the ACPI namespace underneath the parent
1801  * "bridge" whether that's a root PCI bus in the Generation 1
1802  * case or a Module Device in the Generation 2 case.  It then
1803  * attempts to allocate from the global MMIO pool in a way that
1804  * matches the constraints supplied in these parameters and by
1805  * that _CRS.
1806  *
1807  * Return: 0 on success, -errno on failure
1808  */
1809 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
1810                         resource_size_t min, resource_size_t max,
1811                         resource_size_t size, resource_size_t align,
1812                         bool fb_overlap_ok)
1813 {
1814         struct resource *iter, *shadow;
1815         resource_size_t range_min, range_max, start;
1816         const char *dev_n = dev_name(&device_obj->device);
1817         int retval;
1818
1819         retval = -ENXIO;
1820         down(&hyperv_mmio_lock);
1821
1822         /*
1823          * If overlaps with frame buffers are allowed, then first attempt to
1824          * make the allocation from within the reserved region.  Because it
1825          * is already reserved, no shadow allocation is necessary.
1826          */
1827         if (fb_overlap_ok && fb_mmio && !(min > fb_mmio->end) &&
1828             !(max < fb_mmio->start)) {
1829
1830                 range_min = fb_mmio->start;
1831                 range_max = fb_mmio->end;
1832                 start = (range_min + align - 1) & ~(align - 1);
1833                 for (; start + size - 1 <= range_max; start += align) {
1834                         *new = request_mem_region_exclusive(start, size, dev_n);
1835                         if (*new) {
1836                                 retval = 0;
1837                                 goto exit;
1838                         }
1839                 }
1840         }
1841
1842         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
1843                 if ((iter->start >= max) || (iter->end <= min))
1844                         continue;
1845
1846                 range_min = iter->start;
1847                 range_max = iter->end;
1848                 start = (range_min + align - 1) & ~(align - 1);
1849                 for (; start + size - 1 <= range_max; start += align) {
1850                         shadow = __request_region(iter, start, size, NULL,
1851                                                   IORESOURCE_BUSY);
1852                         if (!shadow)
1853                                 continue;
1854
1855                         *new = request_mem_region_exclusive(start, size, dev_n);
1856                         if (*new) {
1857                                 shadow->name = (char *)*new;
1858                                 retval = 0;
1859                                 goto exit;
1860                         }
1861
1862                         __release_region(iter, start, size);
1863                 }
1864         }
1865
1866 exit:
1867         up(&hyperv_mmio_lock);
1868         return retval;
1869 }
1870 EXPORT_SYMBOL_GPL(vmbus_allocate_mmio);
1871
1872 /**
1873  * vmbus_free_mmio() - Free a memory-mapped I/O range.
1874  * @start:              Base address of region to release.
1875  * @size:               Size of the range to be allocated
1876  *
1877  * This function releases anything requested by
1878  * vmbus_mmio_allocate().
1879  */
1880 void vmbus_free_mmio(resource_size_t start, resource_size_t size)
1881 {
1882         struct resource *iter;
1883
1884         down(&hyperv_mmio_lock);
1885         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
1886                 if ((iter->start >= start + size) || (iter->end <= start))
1887                         continue;
1888
1889                 __release_region(iter, start, size);
1890         }
1891         release_mem_region(start, size);
1892         up(&hyperv_mmio_lock);
1893
1894 }
1895 EXPORT_SYMBOL_GPL(vmbus_free_mmio);
1896
1897 static int vmbus_acpi_add(struct acpi_device *device)
1898 {
1899         acpi_status result;
1900         int ret_val = -ENODEV;
1901         struct acpi_device *ancestor;
1902
1903         hv_acpi_dev = device;
1904
1905         result = acpi_walk_resources(device->handle, METHOD_NAME__CRS,
1906                                         vmbus_walk_resources, NULL);
1907
1908         if (ACPI_FAILURE(result))
1909                 goto acpi_walk_err;
1910         /*
1911          * Some ancestor of the vmbus acpi device (Gen1 or Gen2
1912          * firmware) is the VMOD that has the mmio ranges. Get that.
1913          */
1914         for (ancestor = device->parent; ancestor; ancestor = ancestor->parent) {
1915                 result = acpi_walk_resources(ancestor->handle, METHOD_NAME__CRS,
1916                                              vmbus_walk_resources, NULL);
1917
1918                 if (ACPI_FAILURE(result))
1919                         continue;
1920                 if (hyperv_mmio) {
1921                         vmbus_reserve_fb();
1922                         break;
1923                 }
1924         }
1925         ret_val = 0;
1926
1927 acpi_walk_err:
1928         complete(&probe_event);
1929         if (ret_val)
1930                 vmbus_acpi_remove(device);
1931         return ret_val;
1932 }
1933
1934 static const struct acpi_device_id vmbus_acpi_device_ids[] = {
1935         {"VMBUS", 0},
1936         {"VMBus", 0},
1937         {"", 0},
1938 };
1939 MODULE_DEVICE_TABLE(acpi, vmbus_acpi_device_ids);
1940
1941 static struct acpi_driver vmbus_acpi_driver = {
1942         .name = "vmbus",
1943         .ids = vmbus_acpi_device_ids,
1944         .ops = {
1945                 .add = vmbus_acpi_add,
1946                 .remove = vmbus_acpi_remove,
1947         },
1948 };
1949
1950 static void hv_kexec_handler(void)
1951 {
1952         hv_synic_clockevents_cleanup();
1953         vmbus_initiate_unload(false);
1954         vmbus_connection.conn_state = DISCONNECTED;
1955         /* Make sure conn_state is set as hv_synic_cleanup checks for it */
1956         mb();
1957         cpuhp_remove_state(hyperv_cpuhp_online);
1958         hyperv_cleanup();
1959 };
1960
1961 static void hv_crash_handler(struct pt_regs *regs)
1962 {
1963         vmbus_initiate_unload(true);
1964         /*
1965          * In crash handler we can't schedule synic cleanup for all CPUs,
1966          * doing the cleanup for current CPU only. This should be sufficient
1967          * for kdump.
1968          */
1969         vmbus_connection.conn_state = DISCONNECTED;
1970         hv_synic_cleanup(smp_processor_id());
1971         hyperv_cleanup();
1972 };
1973
1974 static int __init hv_acpi_init(void)
1975 {
1976         int ret, t;
1977
1978         if (!hv_is_hyperv_initialized())
1979                 return -ENODEV;
1980
1981         init_completion(&probe_event);
1982
1983         /*
1984          * Get ACPI resources first.
1985          */
1986         ret = acpi_bus_register_driver(&vmbus_acpi_driver);
1987
1988         if (ret)
1989                 return ret;
1990
1991         t = wait_for_completion_timeout(&probe_event, 5*HZ);
1992         if (t == 0) {
1993                 ret = -ETIMEDOUT;
1994                 goto cleanup;
1995         }
1996
1997         ret = vmbus_bus_init();
1998         if (ret)
1999                 goto cleanup;
2000
2001         hv_setup_kexec_handler(hv_kexec_handler);
2002         hv_setup_crash_handler(hv_crash_handler);
2003
2004         return 0;
2005
2006 cleanup:
2007         acpi_bus_unregister_driver(&vmbus_acpi_driver);
2008         hv_acpi_dev = NULL;
2009         return ret;
2010 }
2011
2012 static void __exit vmbus_exit(void)
2013 {
2014         int cpu;
2015
2016         hv_remove_kexec_handler();
2017         hv_remove_crash_handler();
2018         vmbus_connection.conn_state = DISCONNECTED;
2019         hv_synic_clockevents_cleanup();
2020         vmbus_disconnect();
2021         hv_remove_vmbus_irq();
2022         for_each_online_cpu(cpu) {
2023                 struct hv_per_cpu_context *hv_cpu
2024                         = per_cpu_ptr(hv_context.cpu_context, cpu);
2025
2026                 tasklet_kill(&hv_cpu->msg_dpc);
2027         }
2028         vmbus_free_channels();
2029
2030         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
2031                 kmsg_dump_unregister(&hv_kmsg_dumper);
2032                 unregister_die_notifier(&hyperv_die_block);
2033                 atomic_notifier_chain_unregister(&panic_notifier_list,
2034                                                  &hyperv_panic_block);
2035         }
2036
2037         free_page((unsigned long)hv_panic_page);
2038         unregister_sysctl_table(hv_ctl_table_hdr);
2039         hv_ctl_table_hdr = NULL;
2040         bus_unregister(&hv_bus);
2041
2042         cpuhp_remove_state(hyperv_cpuhp_online);
2043         hv_synic_free();
2044         acpi_bus_unregister_driver(&vmbus_acpi_driver);
2045 }
2046
2047
2048 MODULE_LICENSE("GPL");
2049
2050 subsys_initcall(hv_acpi_init);
2051 module_exit(vmbus_exit);