xref: /linux/drivers/hv/vmbus_drv.c (revision 905e46acd3272d04566fec49afbd7ad9e2ed9ae3)
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/hyperv.h>
40 #include <asm/hypervisor.h>
41 #include <asm/mshyperv.h>
42 #include <linux/notifier.h>
43 #include <linux/ptrace.h>
44 #include <linux/screen_info.h>
45 #include <linux/kdebug.h>
46 #include <linux/efi.h>
47 #include <linux/random.h>
48 #include "hyperv_vmbus.h"
49 
50 struct vmbus_dynid {
51 	struct list_head node;
52 	struct hv_vmbus_device_id id;
53 };
54 
55 static struct acpi_device  *hv_acpi_dev;
56 
57 static struct completion probe_event;
58 
59 static int hyperv_cpuhp_online;
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);
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);
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(struct vmbus_channel *channel)
111 {
112 	return (u8)channel->offermsg.monitorid / 32;
113 }
114 
115 static u8 channel_monitor_offset(struct vmbus_channel *channel)
116 {
117 	return (u8)channel->offermsg.monitorid % 32;
118 }
119 
120 static u32 channel_pending(struct vmbus_channel *channel,
121 			   struct hv_monitor_page *monitor_page)
122 {
123 	u8 monitor_group = channel_monitor_group(channel);
124 	return monitor_page->trigger_group[monitor_group].pending;
125 }
126 
127 static u32 channel_latency(struct vmbus_channel *channel,
128 			   struct hv_monitor_page *monitor_page)
129 {
130 	u8 monitor_group = channel_monitor_group(channel);
131 	u8 monitor_offset = channel_monitor_offset(channel);
132 	return monitor_page->latency[monitor_group][monitor_offset];
133 }
134 
135 static u32 channel_conn_id(struct vmbus_channel *channel,
136 			   struct hv_monitor_page *monitor_page)
137 {
138 	u8 monitor_group = channel_monitor_group(channel);
139 	u8 monitor_offset = channel_monitor_offset(channel);
140 	return monitor_page->parameter[monitor_group][monitor_offset].connectionid.u.id;
141 }
142 
143 static ssize_t id_show(struct device *dev, struct device_attribute *dev_attr,
144 		       char *buf)
145 {
146 	struct hv_device *hv_dev = device_to_hv_device(dev);
147 
148 	if (!hv_dev->channel)
149 		return -ENODEV;
150 	return sprintf(buf, "%d\n", hv_dev->channel->offermsg.child_relid);
151 }
152 static DEVICE_ATTR_RO(id);
153 
154 static ssize_t state_show(struct device *dev, struct device_attribute *dev_attr,
155 			  char *buf)
156 {
157 	struct hv_device *hv_dev = device_to_hv_device(dev);
158 
159 	if (!hv_dev->channel)
160 		return -ENODEV;
161 	return sprintf(buf, "%d\n", hv_dev->channel->state);
162 }
163 static DEVICE_ATTR_RO(state);
164 
165 static ssize_t monitor_id_show(struct device *dev,
166 			       struct device_attribute *dev_attr, char *buf)
167 {
168 	struct hv_device *hv_dev = device_to_hv_device(dev);
169 
170 	if (!hv_dev->channel)
171 		return -ENODEV;
172 	return sprintf(buf, "%d\n", hv_dev->channel->offermsg.monitorid);
173 }
174 static DEVICE_ATTR_RO(monitor_id);
175 
176 static ssize_t class_id_show(struct device *dev,
177 			       struct device_attribute *dev_attr, char *buf)
178 {
179 	struct hv_device *hv_dev = device_to_hv_device(dev);
180 
181 	if (!hv_dev->channel)
182 		return -ENODEV;
183 	return sprintf(buf, "{%pUl}\n",
184 		       hv_dev->channel->offermsg.offer.if_type.b);
185 }
186 static DEVICE_ATTR_RO(class_id);
187 
188 static ssize_t device_id_show(struct device *dev,
189 			      struct device_attribute *dev_attr, char *buf)
190 {
191 	struct hv_device *hv_dev = device_to_hv_device(dev);
192 
193 	if (!hv_dev->channel)
194 		return -ENODEV;
195 	return sprintf(buf, "{%pUl}\n",
196 		       hv_dev->channel->offermsg.offer.if_instance.b);
197 }
198 static DEVICE_ATTR_RO(device_id);
199 
200 static ssize_t modalias_show(struct device *dev,
201 			     struct device_attribute *dev_attr, char *buf)
202 {
203 	struct hv_device *hv_dev = device_to_hv_device(dev);
204 	char alias_name[VMBUS_ALIAS_LEN + 1];
205 
206 	print_alias_name(hv_dev, alias_name);
207 	return sprintf(buf, "vmbus:%s\n", alias_name);
208 }
209 static DEVICE_ATTR_RO(modalias);
210 
211 static ssize_t server_monitor_pending_show(struct device *dev,
212 					   struct device_attribute *dev_attr,
213 					   char *buf)
214 {
215 	struct hv_device *hv_dev = device_to_hv_device(dev);
216 
217 	if (!hv_dev->channel)
218 		return -ENODEV;
219 	return sprintf(buf, "%d\n",
220 		       channel_pending(hv_dev->channel,
221 				       vmbus_connection.monitor_pages[1]));
222 }
223 static DEVICE_ATTR_RO(server_monitor_pending);
224 
225 static ssize_t client_monitor_pending_show(struct device *dev,
226 					   struct device_attribute *dev_attr,
227 					   char *buf)
228 {
229 	struct hv_device *hv_dev = device_to_hv_device(dev);
230 
231 	if (!hv_dev->channel)
232 		return -ENODEV;
233 	return sprintf(buf, "%d\n",
234 		       channel_pending(hv_dev->channel,
235 				       vmbus_connection.monitor_pages[1]));
236 }
237 static DEVICE_ATTR_RO(client_monitor_pending);
238 
239 static ssize_t server_monitor_latency_show(struct device *dev,
240 					   struct device_attribute *dev_attr,
241 					   char *buf)
242 {
243 	struct hv_device *hv_dev = device_to_hv_device(dev);
244 
245 	if (!hv_dev->channel)
246 		return -ENODEV;
247 	return sprintf(buf, "%d\n",
248 		       channel_latency(hv_dev->channel,
249 				       vmbus_connection.monitor_pages[0]));
250 }
251 static DEVICE_ATTR_RO(server_monitor_latency);
252 
253 static ssize_t client_monitor_latency_show(struct device *dev,
254 					   struct device_attribute *dev_attr,
255 					   char *buf)
256 {
257 	struct hv_device *hv_dev = device_to_hv_device(dev);
258 
259 	if (!hv_dev->channel)
260 		return -ENODEV;
261 	return sprintf(buf, "%d\n",
262 		       channel_latency(hv_dev->channel,
263 				       vmbus_connection.monitor_pages[1]));
264 }
265 static DEVICE_ATTR_RO(client_monitor_latency);
266 
267 static ssize_t server_monitor_conn_id_show(struct device *dev,
268 					   struct device_attribute *dev_attr,
269 					   char *buf)
270 {
271 	struct hv_device *hv_dev = device_to_hv_device(dev);
272 
273 	if (!hv_dev->channel)
274 		return -ENODEV;
275 	return sprintf(buf, "%d\n",
276 		       channel_conn_id(hv_dev->channel,
277 				       vmbus_connection.monitor_pages[0]));
278 }
279 static DEVICE_ATTR_RO(server_monitor_conn_id);
280 
281 static ssize_t client_monitor_conn_id_show(struct device *dev,
282 					   struct device_attribute *dev_attr,
283 					   char *buf)
284 {
285 	struct hv_device *hv_dev = device_to_hv_device(dev);
286 
287 	if (!hv_dev->channel)
288 		return -ENODEV;
289 	return sprintf(buf, "%d\n",
290 		       channel_conn_id(hv_dev->channel,
291 				       vmbus_connection.monitor_pages[1]));
292 }
293 static DEVICE_ATTR_RO(client_monitor_conn_id);
294 
295 static ssize_t out_intr_mask_show(struct device *dev,
296 				  struct device_attribute *dev_attr, char *buf)
297 {
298 	struct hv_device *hv_dev = device_to_hv_device(dev);
299 	struct hv_ring_buffer_debug_info outbound;
300 
301 	if (!hv_dev->channel)
302 		return -ENODEV;
303 	hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
304 	return sprintf(buf, "%d\n", outbound.current_interrupt_mask);
305 }
306 static DEVICE_ATTR_RO(out_intr_mask);
307 
308 static ssize_t out_read_index_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_read_index);
318 }
319 static DEVICE_ATTR_RO(out_read_index);
320 
321 static ssize_t out_write_index_show(struct device *dev,
322 				    struct device_attribute *dev_attr,
323 				    char *buf)
324 {
325 	struct hv_device *hv_dev = device_to_hv_device(dev);
326 	struct hv_ring_buffer_debug_info outbound;
327 
328 	if (!hv_dev->channel)
329 		return -ENODEV;
330 	hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
331 	return sprintf(buf, "%d\n", outbound.current_write_index);
332 }
333 static DEVICE_ATTR_RO(out_write_index);
334 
335 static ssize_t out_read_bytes_avail_show(struct device *dev,
336 					 struct device_attribute *dev_attr,
337 					 char *buf)
338 {
339 	struct hv_device *hv_dev = device_to_hv_device(dev);
340 	struct hv_ring_buffer_debug_info outbound;
341 
342 	if (!hv_dev->channel)
343 		return -ENODEV;
344 	hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
345 	return sprintf(buf, "%d\n", outbound.bytes_avail_toread);
346 }
347 static DEVICE_ATTR_RO(out_read_bytes_avail);
348 
349 static ssize_t out_write_bytes_avail_show(struct device *dev,
350 					  struct device_attribute *dev_attr,
351 					  char *buf)
352 {
353 	struct hv_device *hv_dev = device_to_hv_device(dev);
354 	struct hv_ring_buffer_debug_info outbound;
355 
356 	if (!hv_dev->channel)
357 		return -ENODEV;
358 	hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
359 	return sprintf(buf, "%d\n", outbound.bytes_avail_towrite);
360 }
361 static DEVICE_ATTR_RO(out_write_bytes_avail);
362 
363 static ssize_t in_intr_mask_show(struct device *dev,
364 				 struct device_attribute *dev_attr, char *buf)
365 {
366 	struct hv_device *hv_dev = device_to_hv_device(dev);
367 	struct hv_ring_buffer_debug_info inbound;
368 
369 	if (!hv_dev->channel)
370 		return -ENODEV;
371 	hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
372 	return sprintf(buf, "%d\n", inbound.current_interrupt_mask);
373 }
374 static DEVICE_ATTR_RO(in_intr_mask);
375 
376 static ssize_t in_read_index_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_read_index);
386 }
387 static DEVICE_ATTR_RO(in_read_index);
388 
389 static ssize_t in_write_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_write_index);
399 }
400 static DEVICE_ATTR_RO(in_write_index);
401 
402 static ssize_t in_read_bytes_avail_show(struct device *dev,
403 					struct device_attribute *dev_attr,
404 					char *buf)
405 {
406 	struct hv_device *hv_dev = device_to_hv_device(dev);
407 	struct hv_ring_buffer_debug_info inbound;
408 
409 	if (!hv_dev->channel)
410 		return -ENODEV;
411 	hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
412 	return sprintf(buf, "%d\n", inbound.bytes_avail_toread);
413 }
414 static DEVICE_ATTR_RO(in_read_bytes_avail);
415 
416 static ssize_t in_write_bytes_avail_show(struct device *dev,
417 					 struct device_attribute *dev_attr,
418 					 char *buf)
419 {
420 	struct hv_device *hv_dev = device_to_hv_device(dev);
421 	struct hv_ring_buffer_debug_info inbound;
422 
423 	if (!hv_dev->channel)
424 		return -ENODEV;
425 	hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
426 	return sprintf(buf, "%d\n", inbound.bytes_avail_towrite);
427 }
428 static DEVICE_ATTR_RO(in_write_bytes_avail);
429 
430 static ssize_t channel_vp_mapping_show(struct device *dev,
431 				       struct device_attribute *dev_attr,
432 				       char *buf)
433 {
434 	struct hv_device *hv_dev = device_to_hv_device(dev);
435 	struct vmbus_channel *channel = hv_dev->channel, *cur_sc;
436 	unsigned long flags;
437 	int buf_size = PAGE_SIZE, n_written, tot_written;
438 	struct list_head *cur;
439 
440 	if (!channel)
441 		return -ENODEV;
442 
443 	tot_written = snprintf(buf, buf_size, "%u:%u\n",
444 		channel->offermsg.child_relid, channel->target_cpu);
445 
446 	spin_lock_irqsave(&channel->lock, flags);
447 
448 	list_for_each(cur, &channel->sc_list) {
449 		if (tot_written >= buf_size - 1)
450 			break;
451 
452 		cur_sc = list_entry(cur, struct vmbus_channel, sc_list);
453 		n_written = scnprintf(buf + tot_written,
454 				     buf_size - tot_written,
455 				     "%u:%u\n",
456 				     cur_sc->offermsg.child_relid,
457 				     cur_sc->target_cpu);
458 		tot_written += n_written;
459 	}
460 
461 	spin_unlock_irqrestore(&channel->lock, flags);
462 
463 	return tot_written;
464 }
465 static DEVICE_ATTR_RO(channel_vp_mapping);
466 
467 static ssize_t vendor_show(struct device *dev,
468 			   struct device_attribute *dev_attr,
469 			   char *buf)
470 {
471 	struct hv_device *hv_dev = device_to_hv_device(dev);
472 	return sprintf(buf, "0x%x\n", hv_dev->vendor_id);
473 }
474 static DEVICE_ATTR_RO(vendor);
475 
476 static ssize_t device_show(struct device *dev,
477 			   struct device_attribute *dev_attr,
478 			   char *buf)
479 {
480 	struct hv_device *hv_dev = device_to_hv_device(dev);
481 	return sprintf(buf, "0x%x\n", hv_dev->device_id);
482 }
483 static DEVICE_ATTR_RO(device);
484 
485 /* Set up per device attributes in /sys/bus/vmbus/devices/<bus device> */
486 static struct attribute *vmbus_dev_attrs[] = {
487 	&dev_attr_id.attr,
488 	&dev_attr_state.attr,
489 	&dev_attr_monitor_id.attr,
490 	&dev_attr_class_id.attr,
491 	&dev_attr_device_id.attr,
492 	&dev_attr_modalias.attr,
493 	&dev_attr_server_monitor_pending.attr,
494 	&dev_attr_client_monitor_pending.attr,
495 	&dev_attr_server_monitor_latency.attr,
496 	&dev_attr_client_monitor_latency.attr,
497 	&dev_attr_server_monitor_conn_id.attr,
498 	&dev_attr_client_monitor_conn_id.attr,
499 	&dev_attr_out_intr_mask.attr,
500 	&dev_attr_out_read_index.attr,
501 	&dev_attr_out_write_index.attr,
502 	&dev_attr_out_read_bytes_avail.attr,
503 	&dev_attr_out_write_bytes_avail.attr,
504 	&dev_attr_in_intr_mask.attr,
505 	&dev_attr_in_read_index.attr,
506 	&dev_attr_in_write_index.attr,
507 	&dev_attr_in_read_bytes_avail.attr,
508 	&dev_attr_in_write_bytes_avail.attr,
509 	&dev_attr_channel_vp_mapping.attr,
510 	&dev_attr_vendor.attr,
511 	&dev_attr_device.attr,
512 	NULL,
513 };
514 ATTRIBUTE_GROUPS(vmbus_dev);
515 
516 /*
517  * vmbus_uevent - add uevent for our device
518  *
519  * This routine is invoked when a device is added or removed on the vmbus to
520  * generate a uevent to udev in the userspace. The udev will then look at its
521  * rule and the uevent generated here to load the appropriate driver
522  *
523  * The alias string will be of the form vmbus:guid where guid is the string
524  * representation of the device guid (each byte of the guid will be
525  * represented with two hex characters.
526  */
527 static int vmbus_uevent(struct device *device, struct kobj_uevent_env *env)
528 {
529 	struct hv_device *dev = device_to_hv_device(device);
530 	int ret;
531 	char alias_name[VMBUS_ALIAS_LEN + 1];
532 
533 	print_alias_name(dev, alias_name);
534 	ret = add_uevent_var(env, "MODALIAS=vmbus:%s", alias_name);
535 	return ret;
536 }
537 
538 static const uuid_le null_guid;
539 
540 static inline bool is_null_guid(const uuid_le *guid)
541 {
542 	if (uuid_le_cmp(*guid, null_guid))
543 		return false;
544 	return true;
545 }
546 
547 /*
548  * Return a matching hv_vmbus_device_id pointer.
549  * If there is no match, return NULL.
550  */
551 static const struct hv_vmbus_device_id *hv_vmbus_get_id(struct hv_driver *drv,
552 					const uuid_le *guid)
553 {
554 	const struct hv_vmbus_device_id *id = NULL;
555 	struct vmbus_dynid *dynid;
556 
557 	/* Look at the dynamic ids first, before the static ones */
558 	spin_lock(&drv->dynids.lock);
559 	list_for_each_entry(dynid, &drv->dynids.list, node) {
560 		if (!uuid_le_cmp(dynid->id.guid, *guid)) {
561 			id = &dynid->id;
562 			break;
563 		}
564 	}
565 	spin_unlock(&drv->dynids.lock);
566 
567 	if (id)
568 		return id;
569 
570 	id = drv->id_table;
571 	if (id == NULL)
572 		return NULL; /* empty device table */
573 
574 	for (; !is_null_guid(&id->guid); id++)
575 		if (!uuid_le_cmp(id->guid, *guid))
576 			return id;
577 
578 	return NULL;
579 }
580 
581 /* vmbus_add_dynid - add a new device ID to this driver and re-probe devices */
582 static int vmbus_add_dynid(struct hv_driver *drv, uuid_le *guid)
583 {
584 	struct vmbus_dynid *dynid;
585 
586 	dynid = kzalloc(sizeof(*dynid), GFP_KERNEL);
587 	if (!dynid)
588 		return -ENOMEM;
589 
590 	dynid->id.guid = *guid;
591 
592 	spin_lock(&drv->dynids.lock);
593 	list_add_tail(&dynid->node, &drv->dynids.list);
594 	spin_unlock(&drv->dynids.lock);
595 
596 	return driver_attach(&drv->driver);
597 }
598 
599 static void vmbus_free_dynids(struct hv_driver *drv)
600 {
601 	struct vmbus_dynid *dynid, *n;
602 
603 	spin_lock(&drv->dynids.lock);
604 	list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
605 		list_del(&dynid->node);
606 		kfree(dynid);
607 	}
608 	spin_unlock(&drv->dynids.lock);
609 }
610 
611 /* Parse string of form: 1b4e28ba-2fa1-11d2-883f-b9a761bde3f */
612 static int get_uuid_le(const char *str, uuid_le *uu)
613 {
614 	unsigned int b[16];
615 	int i;
616 
617 	if (strlen(str) < 37)
618 		return -1;
619 
620 	for (i = 0; i < 36; i++) {
621 		switch (i) {
622 		case 8: case 13: case 18: case 23:
623 			if (str[i] != '-')
624 				return -1;
625 			break;
626 		default:
627 			if (!isxdigit(str[i]))
628 				return -1;
629 		}
630 	}
631 
632 	/* unparse little endian output byte order */
633 	if (sscanf(str,
634 		   "%2x%2x%2x%2x-%2x%2x-%2x%2x-%2x%2x-%2x%2x%2x%2x%2x%2x",
635 		   &b[3], &b[2], &b[1], &b[0],
636 		   &b[5], &b[4], &b[7], &b[6], &b[8], &b[9],
637 		   &b[10], &b[11], &b[12], &b[13], &b[14], &b[15]) != 16)
638 		return -1;
639 
640 	for (i = 0; i < 16; i++)
641 		uu->b[i] = b[i];
642 	return 0;
643 }
644 
645 /*
646  * store_new_id - sysfs frontend to vmbus_add_dynid()
647  *
648  * Allow GUIDs to be added to an existing driver via sysfs.
649  */
650 static ssize_t new_id_store(struct device_driver *driver, const char *buf,
651 			    size_t count)
652 {
653 	struct hv_driver *drv = drv_to_hv_drv(driver);
654 	uuid_le guid = NULL_UUID_LE;
655 	ssize_t retval;
656 
657 	if (get_uuid_le(buf, &guid) != 0)
658 		return -EINVAL;
659 
660 	if (hv_vmbus_get_id(drv, &guid))
661 		return -EEXIST;
662 
663 	retval = vmbus_add_dynid(drv, &guid);
664 	if (retval)
665 		return retval;
666 	return count;
667 }
668 static DRIVER_ATTR_WO(new_id);
669 
670 /*
671  * store_remove_id - remove a PCI device ID from this driver
672  *
673  * Removes a dynamic pci device ID to this driver.
674  */
675 static ssize_t remove_id_store(struct device_driver *driver, const char *buf,
676 			       size_t count)
677 {
678 	struct hv_driver *drv = drv_to_hv_drv(driver);
679 	struct vmbus_dynid *dynid, *n;
680 	uuid_le guid = NULL_UUID_LE;
681 	size_t retval = -ENODEV;
682 
683 	if (get_uuid_le(buf, &guid))
684 		return -EINVAL;
685 
686 	spin_lock(&drv->dynids.lock);
687 	list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
688 		struct hv_vmbus_device_id *id = &dynid->id;
689 
690 		if (!uuid_le_cmp(id->guid, guid)) {
691 			list_del(&dynid->node);
692 			kfree(dynid);
693 			retval = count;
694 			break;
695 		}
696 	}
697 	spin_unlock(&drv->dynids.lock);
698 
699 	return retval;
700 }
701 static DRIVER_ATTR_WO(remove_id);
702 
703 static struct attribute *vmbus_drv_attrs[] = {
704 	&driver_attr_new_id.attr,
705 	&driver_attr_remove_id.attr,
706 	NULL,
707 };
708 ATTRIBUTE_GROUPS(vmbus_drv);
709 
710 
711 /*
712  * vmbus_match - Attempt to match the specified device to the specified driver
713  */
714 static int vmbus_match(struct device *device, struct device_driver *driver)
715 {
716 	struct hv_driver *drv = drv_to_hv_drv(driver);
717 	struct hv_device *hv_dev = device_to_hv_device(device);
718 
719 	/* The hv_sock driver handles all hv_sock offers. */
720 	if (is_hvsock_channel(hv_dev->channel))
721 		return drv->hvsock;
722 
723 	if (hv_vmbus_get_id(drv, &hv_dev->dev_type))
724 		return 1;
725 
726 	return 0;
727 }
728 
729 /*
730  * vmbus_probe - Add the new vmbus's child device
731  */
732 static int vmbus_probe(struct device *child_device)
733 {
734 	int ret = 0;
735 	struct hv_driver *drv =
736 			drv_to_hv_drv(child_device->driver);
737 	struct hv_device *dev = device_to_hv_device(child_device);
738 	const struct hv_vmbus_device_id *dev_id;
739 
740 	dev_id = hv_vmbus_get_id(drv, &dev->dev_type);
741 	if (drv->probe) {
742 		ret = drv->probe(dev, dev_id);
743 		if (ret != 0)
744 			pr_err("probe failed for device %s (%d)\n",
745 			       dev_name(child_device), ret);
746 
747 	} else {
748 		pr_err("probe not set for driver %s\n",
749 		       dev_name(child_device));
750 		ret = -ENODEV;
751 	}
752 	return ret;
753 }
754 
755 /*
756  * vmbus_remove - Remove a vmbus device
757  */
758 static int vmbus_remove(struct device *child_device)
759 {
760 	struct hv_driver *drv;
761 	struct hv_device *dev = device_to_hv_device(child_device);
762 
763 	if (child_device->driver) {
764 		drv = drv_to_hv_drv(child_device->driver);
765 		if (drv->remove)
766 			drv->remove(dev);
767 	}
768 
769 	return 0;
770 }
771 
772 
773 /*
774  * vmbus_shutdown - Shutdown a vmbus device
775  */
776 static void vmbus_shutdown(struct device *child_device)
777 {
778 	struct hv_driver *drv;
779 	struct hv_device *dev = device_to_hv_device(child_device);
780 
781 
782 	/* The device may not be attached yet */
783 	if (!child_device->driver)
784 		return;
785 
786 	drv = drv_to_hv_drv(child_device->driver);
787 
788 	if (drv->shutdown)
789 		drv->shutdown(dev);
790 }
791 
792 
793 /*
794  * vmbus_device_release - Final callback release of the vmbus child device
795  */
796 static void vmbus_device_release(struct device *device)
797 {
798 	struct hv_device *hv_dev = device_to_hv_device(device);
799 	struct vmbus_channel *channel = hv_dev->channel;
800 
801 	hv_process_channel_removal(channel,
802 				   channel->offermsg.child_relid);
803 	kfree(hv_dev);
804 
805 }
806 
807 /* The one and only one */
808 static struct bus_type  hv_bus = {
809 	.name =		"vmbus",
810 	.match =		vmbus_match,
811 	.shutdown =		vmbus_shutdown,
812 	.remove =		vmbus_remove,
813 	.probe =		vmbus_probe,
814 	.uevent =		vmbus_uevent,
815 	.dev_groups =		vmbus_dev_groups,
816 	.drv_groups =		vmbus_drv_groups,
817 };
818 
819 struct onmessage_work_context {
820 	struct work_struct work;
821 	struct hv_message msg;
822 };
823 
824 static void vmbus_onmessage_work(struct work_struct *work)
825 {
826 	struct onmessage_work_context *ctx;
827 
828 	/* Do not process messages if we're in DISCONNECTED state */
829 	if (vmbus_connection.conn_state == DISCONNECTED)
830 		return;
831 
832 	ctx = container_of(work, struct onmessage_work_context,
833 			   work);
834 	vmbus_onmessage(&ctx->msg);
835 	kfree(ctx);
836 }
837 
838 static void hv_process_timer_expiration(struct hv_message *msg,
839 					struct hv_per_cpu_context *hv_cpu)
840 {
841 	struct clock_event_device *dev = hv_cpu->clk_evt;
842 
843 	if (dev->event_handler)
844 		dev->event_handler(dev);
845 
846 	vmbus_signal_eom(msg, HVMSG_TIMER_EXPIRED);
847 }
848 
849 void vmbus_on_msg_dpc(unsigned long data)
850 {
851 	struct hv_per_cpu_context *hv_cpu = (void *)data;
852 	void *page_addr = hv_cpu->synic_message_page;
853 	struct hv_message *msg = (struct hv_message *)page_addr +
854 				  VMBUS_MESSAGE_SINT;
855 	struct vmbus_channel_message_header *hdr;
856 	const struct vmbus_channel_message_table_entry *entry;
857 	struct onmessage_work_context *ctx;
858 	u32 message_type = msg->header.message_type;
859 
860 	if (message_type == HVMSG_NONE)
861 		/* no msg */
862 		return;
863 
864 	hdr = (struct vmbus_channel_message_header *)msg->u.payload;
865 
866 	if (hdr->msgtype >= CHANNELMSG_COUNT) {
867 		WARN_ONCE(1, "unknown msgtype=%d\n", hdr->msgtype);
868 		goto msg_handled;
869 	}
870 
871 	entry = &channel_message_table[hdr->msgtype];
872 	if (entry->handler_type	== VMHT_BLOCKING) {
873 		ctx = kmalloc(sizeof(*ctx), GFP_ATOMIC);
874 		if (ctx == NULL)
875 			return;
876 
877 		INIT_WORK(&ctx->work, vmbus_onmessage_work);
878 		memcpy(&ctx->msg, msg, sizeof(*msg));
879 
880 		queue_work(vmbus_connection.work_queue, &ctx->work);
881 	} else
882 		entry->message_handler(hdr);
883 
884 msg_handled:
885 	vmbus_signal_eom(msg, message_type);
886 }
887 
888 
889 /*
890  * Direct callback for channels using other deferred processing
891  */
892 static void vmbus_channel_isr(struct vmbus_channel *channel)
893 {
894 	void (*callback_fn)(void *);
895 
896 	callback_fn = READ_ONCE(channel->onchannel_callback);
897 	if (likely(callback_fn != NULL))
898 		(*callback_fn)(channel->channel_callback_context);
899 }
900 
901 /*
902  * Schedule all channels with events pending
903  */
904 static void vmbus_chan_sched(struct hv_per_cpu_context *hv_cpu)
905 {
906 	unsigned long *recv_int_page;
907 	u32 maxbits, relid;
908 
909 	if (vmbus_proto_version < VERSION_WIN8) {
910 		maxbits = MAX_NUM_CHANNELS_SUPPORTED;
911 		recv_int_page = vmbus_connection.recv_int_page;
912 	} else {
913 		/*
914 		 * When the host is win8 and beyond, the event page
915 		 * can be directly checked to get the id of the channel
916 		 * that has the interrupt pending.
917 		 */
918 		void *page_addr = hv_cpu->synic_event_page;
919 		union hv_synic_event_flags *event
920 			= (union hv_synic_event_flags *)page_addr +
921 						 VMBUS_MESSAGE_SINT;
922 
923 		maxbits = HV_EVENT_FLAGS_COUNT;
924 		recv_int_page = event->flags;
925 	}
926 
927 	if (unlikely(!recv_int_page))
928 		return;
929 
930 	for_each_set_bit(relid, recv_int_page, maxbits) {
931 		struct vmbus_channel *channel;
932 
933 		if (!sync_test_and_clear_bit(relid, recv_int_page))
934 			continue;
935 
936 		/* Special case - vmbus channel protocol msg */
937 		if (relid == 0)
938 			continue;
939 
940 		rcu_read_lock();
941 
942 		/* Find channel based on relid */
943 		list_for_each_entry_rcu(channel, &hv_cpu->chan_list, percpu_list) {
944 			if (channel->offermsg.child_relid != relid)
945 				continue;
946 
947 			switch (channel->callback_mode) {
948 			case HV_CALL_ISR:
949 				vmbus_channel_isr(channel);
950 				break;
951 
952 			case HV_CALL_BATCHED:
953 				hv_begin_read(&channel->inbound);
954 				/* fallthrough */
955 			case HV_CALL_DIRECT:
956 				tasklet_schedule(&channel->callback_event);
957 			}
958 		}
959 
960 		rcu_read_unlock();
961 	}
962 }
963 
964 static void vmbus_isr(void)
965 {
966 	struct hv_per_cpu_context *hv_cpu
967 		= this_cpu_ptr(hv_context.cpu_context);
968 	void *page_addr = hv_cpu->synic_event_page;
969 	struct hv_message *msg;
970 	union hv_synic_event_flags *event;
971 	bool handled = false;
972 
973 	if (unlikely(page_addr == NULL))
974 		return;
975 
976 	event = (union hv_synic_event_flags *)page_addr +
977 					 VMBUS_MESSAGE_SINT;
978 	/*
979 	 * Check for events before checking for messages. This is the order
980 	 * in which events and messages are checked in Windows guests on
981 	 * Hyper-V, and the Windows team suggested we do the same.
982 	 */
983 
984 	if ((vmbus_proto_version == VERSION_WS2008) ||
985 		(vmbus_proto_version == VERSION_WIN7)) {
986 
987 		/* Since we are a child, we only need to check bit 0 */
988 		if (sync_test_and_clear_bit(0, event->flags))
989 			handled = true;
990 	} else {
991 		/*
992 		 * Our host is win8 or above. The signaling mechanism
993 		 * has changed and we can directly look at the event page.
994 		 * If bit n is set then we have an interrup on the channel
995 		 * whose id is n.
996 		 */
997 		handled = true;
998 	}
999 
1000 	if (handled)
1001 		vmbus_chan_sched(hv_cpu);
1002 
1003 	page_addr = hv_cpu->synic_message_page;
1004 	msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
1005 
1006 	/* Check if there are actual msgs to be processed */
1007 	if (msg->header.message_type != HVMSG_NONE) {
1008 		if (msg->header.message_type == HVMSG_TIMER_EXPIRED)
1009 			hv_process_timer_expiration(msg, hv_cpu);
1010 		else
1011 			tasklet_schedule(&hv_cpu->msg_dpc);
1012 	}
1013 
1014 	add_interrupt_randomness(HYPERVISOR_CALLBACK_VECTOR, 0);
1015 }
1016 
1017 
1018 /*
1019  * vmbus_bus_init -Main vmbus driver initialization routine.
1020  *
1021  * Here, we
1022  *	- initialize the vmbus driver context
1023  *	- invoke the vmbus hv main init routine
1024  *	- retrieve the channel offers
1025  */
1026 static int vmbus_bus_init(void)
1027 {
1028 	int ret;
1029 
1030 	/* Hypervisor initialization...setup hypercall page..etc */
1031 	ret = hv_init();
1032 	if (ret != 0) {
1033 		pr_err("Unable to initialize the hypervisor - 0x%x\n", ret);
1034 		return ret;
1035 	}
1036 
1037 	ret = bus_register(&hv_bus);
1038 	if (ret)
1039 		return ret;
1040 
1041 	hv_setup_vmbus_irq(vmbus_isr);
1042 
1043 	ret = hv_synic_alloc();
1044 	if (ret)
1045 		goto err_alloc;
1046 	/*
1047 	 * Initialize the per-cpu interrupt state and
1048 	 * connect to the host.
1049 	 */
1050 	ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "x86/hyperv:online",
1051 				hv_synic_init, hv_synic_cleanup);
1052 	if (ret < 0)
1053 		goto err_alloc;
1054 	hyperv_cpuhp_online = ret;
1055 
1056 	ret = vmbus_connect();
1057 	if (ret)
1058 		goto err_connect;
1059 
1060 	/*
1061 	 * Only register if the crash MSRs are available
1062 	 */
1063 	if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1064 		register_die_notifier(&hyperv_die_block);
1065 		atomic_notifier_chain_register(&panic_notifier_list,
1066 					       &hyperv_panic_block);
1067 	}
1068 
1069 	vmbus_request_offers();
1070 
1071 	return 0;
1072 
1073 err_connect:
1074 	cpuhp_remove_state(hyperv_cpuhp_online);
1075 err_alloc:
1076 	hv_synic_free();
1077 	hv_remove_vmbus_irq();
1078 
1079 	bus_unregister(&hv_bus);
1080 
1081 	return ret;
1082 }
1083 
1084 /**
1085  * __vmbus_child_driver_register() - Register a vmbus's driver
1086  * @hv_driver: Pointer to driver structure you want to register
1087  * @owner: owner module of the drv
1088  * @mod_name: module name string
1089  *
1090  * Registers the given driver with Linux through the 'driver_register()' call
1091  * and sets up the hyper-v vmbus handling for this driver.
1092  * It will return the state of the 'driver_register()' call.
1093  *
1094  */
1095 int __vmbus_driver_register(struct hv_driver *hv_driver, struct module *owner, const char *mod_name)
1096 {
1097 	int ret;
1098 
1099 	pr_info("registering driver %s\n", hv_driver->name);
1100 
1101 	ret = vmbus_exists();
1102 	if (ret < 0)
1103 		return ret;
1104 
1105 	hv_driver->driver.name = hv_driver->name;
1106 	hv_driver->driver.owner = owner;
1107 	hv_driver->driver.mod_name = mod_name;
1108 	hv_driver->driver.bus = &hv_bus;
1109 
1110 	spin_lock_init(&hv_driver->dynids.lock);
1111 	INIT_LIST_HEAD(&hv_driver->dynids.list);
1112 
1113 	ret = driver_register(&hv_driver->driver);
1114 
1115 	return ret;
1116 }
1117 EXPORT_SYMBOL_GPL(__vmbus_driver_register);
1118 
1119 /**
1120  * vmbus_driver_unregister() - Unregister a vmbus's driver
1121  * @hv_driver: Pointer to driver structure you want to
1122  *             un-register
1123  *
1124  * Un-register the given driver that was previous registered with a call to
1125  * vmbus_driver_register()
1126  */
1127 void vmbus_driver_unregister(struct hv_driver *hv_driver)
1128 {
1129 	pr_info("unregistering driver %s\n", hv_driver->name);
1130 
1131 	if (!vmbus_exists()) {
1132 		driver_unregister(&hv_driver->driver);
1133 		vmbus_free_dynids(hv_driver);
1134 	}
1135 }
1136 EXPORT_SYMBOL_GPL(vmbus_driver_unregister);
1137 
1138 /*
1139  * vmbus_device_create - Creates and registers a new child device
1140  * on the vmbus.
1141  */
1142 struct hv_device *vmbus_device_create(const uuid_le *type,
1143 				      const uuid_le *instance,
1144 				      struct vmbus_channel *channel)
1145 {
1146 	struct hv_device *child_device_obj;
1147 
1148 	child_device_obj = kzalloc(sizeof(struct hv_device), GFP_KERNEL);
1149 	if (!child_device_obj) {
1150 		pr_err("Unable to allocate device object for child device\n");
1151 		return NULL;
1152 	}
1153 
1154 	child_device_obj->channel = channel;
1155 	memcpy(&child_device_obj->dev_type, type, sizeof(uuid_le));
1156 	memcpy(&child_device_obj->dev_instance, instance,
1157 	       sizeof(uuid_le));
1158 	child_device_obj->vendor_id = 0x1414; /* MSFT vendor ID */
1159 
1160 
1161 	return child_device_obj;
1162 }
1163 
1164 /*
1165  * vmbus_device_register - Register the child device
1166  */
1167 int vmbus_device_register(struct hv_device *child_device_obj)
1168 {
1169 	int ret = 0;
1170 
1171 	dev_set_name(&child_device_obj->device, "%pUl",
1172 		     child_device_obj->channel->offermsg.offer.if_instance.b);
1173 
1174 	child_device_obj->device.bus = &hv_bus;
1175 	child_device_obj->device.parent = &hv_acpi_dev->dev;
1176 	child_device_obj->device.release = vmbus_device_release;
1177 
1178 	/*
1179 	 * Register with the LDM. This will kick off the driver/device
1180 	 * binding...which will eventually call vmbus_match() and vmbus_probe()
1181 	 */
1182 	ret = device_register(&child_device_obj->device);
1183 
1184 	if (ret)
1185 		pr_err("Unable to register child device\n");
1186 	else
1187 		pr_debug("child device %s registered\n",
1188 			dev_name(&child_device_obj->device));
1189 
1190 	return ret;
1191 }
1192 
1193 /*
1194  * vmbus_device_unregister - Remove the specified child device
1195  * from the vmbus.
1196  */
1197 void vmbus_device_unregister(struct hv_device *device_obj)
1198 {
1199 	pr_debug("child device %s unregistered\n",
1200 		dev_name(&device_obj->device));
1201 
1202 	/*
1203 	 * Kick off the process of unregistering the device.
1204 	 * This will call vmbus_remove() and eventually vmbus_device_release()
1205 	 */
1206 	device_unregister(&device_obj->device);
1207 }
1208 
1209 
1210 /*
1211  * VMBUS is an acpi enumerated device. Get the information we
1212  * need from DSDT.
1213  */
1214 #define VTPM_BASE_ADDRESS 0xfed40000
1215 static acpi_status vmbus_walk_resources(struct acpi_resource *res, void *ctx)
1216 {
1217 	resource_size_t start = 0;
1218 	resource_size_t end = 0;
1219 	struct resource *new_res;
1220 	struct resource **old_res = &hyperv_mmio;
1221 	struct resource **prev_res = NULL;
1222 
1223 	switch (res->type) {
1224 
1225 	/*
1226 	 * "Address" descriptors are for bus windows. Ignore
1227 	 * "memory" descriptors, which are for registers on
1228 	 * devices.
1229 	 */
1230 	case ACPI_RESOURCE_TYPE_ADDRESS32:
1231 		start = res->data.address32.address.minimum;
1232 		end = res->data.address32.address.maximum;
1233 		break;
1234 
1235 	case ACPI_RESOURCE_TYPE_ADDRESS64:
1236 		start = res->data.address64.address.minimum;
1237 		end = res->data.address64.address.maximum;
1238 		break;
1239 
1240 	default:
1241 		/* Unused resource type */
1242 		return AE_OK;
1243 
1244 	}
1245 	/*
1246 	 * Ignore ranges that are below 1MB, as they're not
1247 	 * necessary or useful here.
1248 	 */
1249 	if (end < 0x100000)
1250 		return AE_OK;
1251 
1252 	new_res = kzalloc(sizeof(*new_res), GFP_ATOMIC);
1253 	if (!new_res)
1254 		return AE_NO_MEMORY;
1255 
1256 	/* If this range overlaps the virtual TPM, truncate it. */
1257 	if (end > VTPM_BASE_ADDRESS && start < VTPM_BASE_ADDRESS)
1258 		end = VTPM_BASE_ADDRESS;
1259 
1260 	new_res->name = "hyperv mmio";
1261 	new_res->flags = IORESOURCE_MEM;
1262 	new_res->start = start;
1263 	new_res->end = end;
1264 
1265 	/*
1266 	 * If two ranges are adjacent, merge them.
1267 	 */
1268 	do {
1269 		if (!*old_res) {
1270 			*old_res = new_res;
1271 			break;
1272 		}
1273 
1274 		if (((*old_res)->end + 1) == new_res->start) {
1275 			(*old_res)->end = new_res->end;
1276 			kfree(new_res);
1277 			break;
1278 		}
1279 
1280 		if ((*old_res)->start == new_res->end + 1) {
1281 			(*old_res)->start = new_res->start;
1282 			kfree(new_res);
1283 			break;
1284 		}
1285 
1286 		if ((*old_res)->start > new_res->end) {
1287 			new_res->sibling = *old_res;
1288 			if (prev_res)
1289 				(*prev_res)->sibling = new_res;
1290 			*old_res = new_res;
1291 			break;
1292 		}
1293 
1294 		prev_res = old_res;
1295 		old_res = &(*old_res)->sibling;
1296 
1297 	} while (1);
1298 
1299 	return AE_OK;
1300 }
1301 
1302 static int vmbus_acpi_remove(struct acpi_device *device)
1303 {
1304 	struct resource *cur_res;
1305 	struct resource *next_res;
1306 
1307 	if (hyperv_mmio) {
1308 		if (fb_mmio) {
1309 			__release_region(hyperv_mmio, fb_mmio->start,
1310 					 resource_size(fb_mmio));
1311 			fb_mmio = NULL;
1312 		}
1313 
1314 		for (cur_res = hyperv_mmio; cur_res; cur_res = next_res) {
1315 			next_res = cur_res->sibling;
1316 			kfree(cur_res);
1317 		}
1318 	}
1319 
1320 	return 0;
1321 }
1322 
1323 static void vmbus_reserve_fb(void)
1324 {
1325 	int size;
1326 	/*
1327 	 * Make a claim for the frame buffer in the resource tree under the
1328 	 * first node, which will be the one below 4GB.  The length seems to
1329 	 * be underreported, particularly in a Generation 1 VM.  So start out
1330 	 * reserving a larger area and make it smaller until it succeeds.
1331 	 */
1332 
1333 	if (screen_info.lfb_base) {
1334 		if (efi_enabled(EFI_BOOT))
1335 			size = max_t(__u32, screen_info.lfb_size, 0x800000);
1336 		else
1337 			size = max_t(__u32, screen_info.lfb_size, 0x4000000);
1338 
1339 		for (; !fb_mmio && (size >= 0x100000); size >>= 1) {
1340 			fb_mmio = __request_region(hyperv_mmio,
1341 						   screen_info.lfb_base, size,
1342 						   fb_mmio_name, 0);
1343 		}
1344 	}
1345 }
1346 
1347 /**
1348  * vmbus_allocate_mmio() - Pick a memory-mapped I/O range.
1349  * @new:		If successful, supplied a pointer to the
1350  *			allocated MMIO space.
1351  * @device_obj:		Identifies the caller
1352  * @min:		Minimum guest physical address of the
1353  *			allocation
1354  * @max:		Maximum guest physical address
1355  * @size:		Size of the range to be allocated
1356  * @align:		Alignment of the range to be allocated
1357  * @fb_overlap_ok:	Whether this allocation can be allowed
1358  *			to overlap the video frame buffer.
1359  *
1360  * This function walks the resources granted to VMBus by the
1361  * _CRS object in the ACPI namespace underneath the parent
1362  * "bridge" whether that's a root PCI bus in the Generation 1
1363  * case or a Module Device in the Generation 2 case.  It then
1364  * attempts to allocate from the global MMIO pool in a way that
1365  * matches the constraints supplied in these parameters and by
1366  * that _CRS.
1367  *
1368  * Return: 0 on success, -errno on failure
1369  */
1370 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
1371 			resource_size_t min, resource_size_t max,
1372 			resource_size_t size, resource_size_t align,
1373 			bool fb_overlap_ok)
1374 {
1375 	struct resource *iter, *shadow;
1376 	resource_size_t range_min, range_max, start;
1377 	const char *dev_n = dev_name(&device_obj->device);
1378 	int retval;
1379 
1380 	retval = -ENXIO;
1381 	down(&hyperv_mmio_lock);
1382 
1383 	/*
1384 	 * If overlaps with frame buffers are allowed, then first attempt to
1385 	 * make the allocation from within the reserved region.  Because it
1386 	 * is already reserved, no shadow allocation is necessary.
1387 	 */
1388 	if (fb_overlap_ok && fb_mmio && !(min > fb_mmio->end) &&
1389 	    !(max < fb_mmio->start)) {
1390 
1391 		range_min = fb_mmio->start;
1392 		range_max = fb_mmio->end;
1393 		start = (range_min + align - 1) & ~(align - 1);
1394 		for (; start + size - 1 <= range_max; start += align) {
1395 			*new = request_mem_region_exclusive(start, size, dev_n);
1396 			if (*new) {
1397 				retval = 0;
1398 				goto exit;
1399 			}
1400 		}
1401 	}
1402 
1403 	for (iter = hyperv_mmio; iter; iter = iter->sibling) {
1404 		if ((iter->start >= max) || (iter->end <= min))
1405 			continue;
1406 
1407 		range_min = iter->start;
1408 		range_max = iter->end;
1409 		start = (range_min + align - 1) & ~(align - 1);
1410 		for (; start + size - 1 <= range_max; start += align) {
1411 			shadow = __request_region(iter, start, size, NULL,
1412 						  IORESOURCE_BUSY);
1413 			if (!shadow)
1414 				continue;
1415 
1416 			*new = request_mem_region_exclusive(start, size, dev_n);
1417 			if (*new) {
1418 				shadow->name = (char *)*new;
1419 				retval = 0;
1420 				goto exit;
1421 			}
1422 
1423 			__release_region(iter, start, size);
1424 		}
1425 	}
1426 
1427 exit:
1428 	up(&hyperv_mmio_lock);
1429 	return retval;
1430 }
1431 EXPORT_SYMBOL_GPL(vmbus_allocate_mmio);
1432 
1433 /**
1434  * vmbus_free_mmio() - Free a memory-mapped I/O range.
1435  * @start:		Base address of region to release.
1436  * @size:		Size of the range to be allocated
1437  *
1438  * This function releases anything requested by
1439  * vmbus_mmio_allocate().
1440  */
1441 void vmbus_free_mmio(resource_size_t start, resource_size_t size)
1442 {
1443 	struct resource *iter;
1444 
1445 	down(&hyperv_mmio_lock);
1446 	for (iter = hyperv_mmio; iter; iter = iter->sibling) {
1447 		if ((iter->start >= start + size) || (iter->end <= start))
1448 			continue;
1449 
1450 		__release_region(iter, start, size);
1451 	}
1452 	release_mem_region(start, size);
1453 	up(&hyperv_mmio_lock);
1454 
1455 }
1456 EXPORT_SYMBOL_GPL(vmbus_free_mmio);
1457 
1458 /**
1459  * vmbus_cpu_number_to_vp_number() - Map CPU to VP.
1460  * @cpu_number: CPU number in Linux terms
1461  *
1462  * This function returns the mapping between the Linux processor
1463  * number and the hypervisor's virtual processor number, useful
1464  * in making hypercalls and such that talk about specific
1465  * processors.
1466  *
1467  * Return: Virtual processor number in Hyper-V terms
1468  */
1469 int vmbus_cpu_number_to_vp_number(int cpu_number)
1470 {
1471 	return hv_context.vp_index[cpu_number];
1472 }
1473 EXPORT_SYMBOL_GPL(vmbus_cpu_number_to_vp_number);
1474 
1475 static int vmbus_acpi_add(struct acpi_device *device)
1476 {
1477 	acpi_status result;
1478 	int ret_val = -ENODEV;
1479 	struct acpi_device *ancestor;
1480 
1481 	hv_acpi_dev = device;
1482 
1483 	result = acpi_walk_resources(device->handle, METHOD_NAME__CRS,
1484 					vmbus_walk_resources, NULL);
1485 
1486 	if (ACPI_FAILURE(result))
1487 		goto acpi_walk_err;
1488 	/*
1489 	 * Some ancestor of the vmbus acpi device (Gen1 or Gen2
1490 	 * firmware) is the VMOD that has the mmio ranges. Get that.
1491 	 */
1492 	for (ancestor = device->parent; ancestor; ancestor = ancestor->parent) {
1493 		result = acpi_walk_resources(ancestor->handle, METHOD_NAME__CRS,
1494 					     vmbus_walk_resources, NULL);
1495 
1496 		if (ACPI_FAILURE(result))
1497 			continue;
1498 		if (hyperv_mmio) {
1499 			vmbus_reserve_fb();
1500 			break;
1501 		}
1502 	}
1503 	ret_val = 0;
1504 
1505 acpi_walk_err:
1506 	complete(&probe_event);
1507 	if (ret_val)
1508 		vmbus_acpi_remove(device);
1509 	return ret_val;
1510 }
1511 
1512 static const struct acpi_device_id vmbus_acpi_device_ids[] = {
1513 	{"VMBUS", 0},
1514 	{"VMBus", 0},
1515 	{"", 0},
1516 };
1517 MODULE_DEVICE_TABLE(acpi, vmbus_acpi_device_ids);
1518 
1519 static struct acpi_driver vmbus_acpi_driver = {
1520 	.name = "vmbus",
1521 	.ids = vmbus_acpi_device_ids,
1522 	.ops = {
1523 		.add = vmbus_acpi_add,
1524 		.remove = vmbus_acpi_remove,
1525 	},
1526 };
1527 
1528 static void hv_kexec_handler(void)
1529 {
1530 	hv_synic_clockevents_cleanup();
1531 	vmbus_initiate_unload(false);
1532 	vmbus_connection.conn_state = DISCONNECTED;
1533 	/* Make sure conn_state is set as hv_synic_cleanup checks for it */
1534 	mb();
1535 	cpuhp_remove_state(hyperv_cpuhp_online);
1536 	hyperv_cleanup();
1537 };
1538 
1539 static void hv_crash_handler(struct pt_regs *regs)
1540 {
1541 	vmbus_initiate_unload(true);
1542 	/*
1543 	 * In crash handler we can't schedule synic cleanup for all CPUs,
1544 	 * doing the cleanup for current CPU only. This should be sufficient
1545 	 * for kdump.
1546 	 */
1547 	vmbus_connection.conn_state = DISCONNECTED;
1548 	hv_synic_cleanup(smp_processor_id());
1549 	hyperv_cleanup();
1550 };
1551 
1552 static int __init hv_acpi_init(void)
1553 {
1554 	int ret, t;
1555 
1556 	if (x86_hyper != &x86_hyper_ms_hyperv)
1557 		return -ENODEV;
1558 
1559 	init_completion(&probe_event);
1560 
1561 	/*
1562 	 * Get ACPI resources first.
1563 	 */
1564 	ret = acpi_bus_register_driver(&vmbus_acpi_driver);
1565 
1566 	if (ret)
1567 		return ret;
1568 
1569 	t = wait_for_completion_timeout(&probe_event, 5*HZ);
1570 	if (t == 0) {
1571 		ret = -ETIMEDOUT;
1572 		goto cleanup;
1573 	}
1574 
1575 	ret = vmbus_bus_init();
1576 	if (ret)
1577 		goto cleanup;
1578 
1579 	hv_setup_kexec_handler(hv_kexec_handler);
1580 	hv_setup_crash_handler(hv_crash_handler);
1581 
1582 	return 0;
1583 
1584 cleanup:
1585 	acpi_bus_unregister_driver(&vmbus_acpi_driver);
1586 	hv_acpi_dev = NULL;
1587 	return ret;
1588 }
1589 
1590 static void __exit vmbus_exit(void)
1591 {
1592 	int cpu;
1593 
1594 	hv_remove_kexec_handler();
1595 	hv_remove_crash_handler();
1596 	vmbus_connection.conn_state = DISCONNECTED;
1597 	hv_synic_clockevents_cleanup();
1598 	vmbus_disconnect();
1599 	hv_remove_vmbus_irq();
1600 	for_each_online_cpu(cpu) {
1601 		struct hv_per_cpu_context *hv_cpu
1602 			= per_cpu_ptr(hv_context.cpu_context, cpu);
1603 
1604 		tasklet_kill(&hv_cpu->msg_dpc);
1605 	}
1606 	vmbus_free_channels();
1607 
1608 	if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1609 		unregister_die_notifier(&hyperv_die_block);
1610 		atomic_notifier_chain_unregister(&panic_notifier_list,
1611 						 &hyperv_panic_block);
1612 	}
1613 	bus_unregister(&hv_bus);
1614 
1615 	cpuhp_remove_state(hyperv_cpuhp_online);
1616 	hv_synic_free();
1617 	acpi_bus_unregister_driver(&vmbus_acpi_driver);
1618 }
1619 
1620 
1621 MODULE_LICENSE("GPL");
1622 
1623 subsys_initcall(hv_acpi_init);
1624 module_exit(vmbus_exit);
1625