xref: /openbmc/linux/drivers/hv/vmbus_drv.c (revision dc6a81c3)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (c) 2009, Microsoft Corporation.
4  *
5  * Authors:
6  *   Haiyang Zhang <haiyangz@microsoft.com>
7  *   Hank Janssen  <hjanssen@microsoft.com>
8  *   K. Y. Srinivasan <kys@microsoft.com>
9  */
10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11 
12 #include <linux/init.h>
13 #include <linux/module.h>
14 #include <linux/device.h>
15 #include <linux/interrupt.h>
16 #include <linux/sysctl.h>
17 #include <linux/slab.h>
18 #include <linux/acpi.h>
19 #include <linux/completion.h>
20 #include <linux/hyperv.h>
21 #include <linux/kernel_stat.h>
22 #include <linux/clockchips.h>
23 #include <linux/cpu.h>
24 #include <linux/sched/task_stack.h>
25 
26 #include <asm/mshyperv.h>
27 #include <linux/delay.h>
28 #include <linux/notifier.h>
29 #include <linux/ptrace.h>
30 #include <linux/screen_info.h>
31 #include <linux/kdebug.h>
32 #include <linux/efi.h>
33 #include <linux/random.h>
34 #include <linux/syscore_ops.h>
35 #include <clocksource/hyperv_timer.h>
36 #include "hyperv_vmbus.h"
37 
38 struct vmbus_dynid {
39 	struct list_head node;
40 	struct hv_vmbus_device_id id;
41 };
42 
43 static struct acpi_device  *hv_acpi_dev;
44 
45 static struct completion probe_event;
46 
47 static int hyperv_cpuhp_online;
48 
49 static void *hv_panic_page;
50 
51 static int hyperv_panic_event(struct notifier_block *nb, unsigned long val,
52 			      void *args)
53 {
54 	struct pt_regs *regs;
55 
56 	regs = current_pt_regs();
57 
58 	hyperv_report_panic(regs, val);
59 	return NOTIFY_DONE;
60 }
61 
62 static int hyperv_die_event(struct notifier_block *nb, unsigned long val,
63 			    void *args)
64 {
65 	struct die_args *die = (struct die_args *)args;
66 	struct pt_regs *regs = die->regs;
67 
68 	hyperv_report_panic(regs, val);
69 	return NOTIFY_DONE;
70 }
71 
72 static struct notifier_block hyperv_die_block = {
73 	.notifier_call = hyperv_die_event,
74 };
75 static struct notifier_block hyperv_panic_block = {
76 	.notifier_call = hyperv_panic_event,
77 };
78 
79 static const char *fb_mmio_name = "fb_range";
80 static struct resource *fb_mmio;
81 static struct resource *hyperv_mmio;
82 static DEFINE_MUTEX(hyperv_mmio_lock);
83 
84 static int vmbus_exists(void)
85 {
86 	if (hv_acpi_dev == NULL)
87 		return -ENODEV;
88 
89 	return 0;
90 }
91 
92 #define VMBUS_ALIAS_LEN ((sizeof((struct hv_vmbus_device_id *)0)->guid) * 2)
93 static void print_alias_name(struct hv_device *hv_dev, char *alias_name)
94 {
95 	int i;
96 	for (i = 0; i < VMBUS_ALIAS_LEN; i += 2)
97 		sprintf(&alias_name[i], "%02x", hv_dev->dev_type.b[i/2]);
98 }
99 
100 static u8 channel_monitor_group(const struct vmbus_channel *channel)
101 {
102 	return (u8)channel->offermsg.monitorid / 32;
103 }
104 
105 static u8 channel_monitor_offset(const struct vmbus_channel *channel)
106 {
107 	return (u8)channel->offermsg.monitorid % 32;
108 }
109 
110 static u32 channel_pending(const struct vmbus_channel *channel,
111 			   const struct hv_monitor_page *monitor_page)
112 {
113 	u8 monitor_group = channel_monitor_group(channel);
114 
115 	return monitor_page->trigger_group[monitor_group].pending;
116 }
117 
118 static u32 channel_latency(const struct vmbus_channel *channel,
119 			   const struct hv_monitor_page *monitor_page)
120 {
121 	u8 monitor_group = channel_monitor_group(channel);
122 	u8 monitor_offset = channel_monitor_offset(channel);
123 
124 	return monitor_page->latency[monitor_group][monitor_offset];
125 }
126 
127 static u32 channel_conn_id(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->parameter[monitor_group][monitor_offset].connectionid.u.id;
133 }
134 
135 static ssize_t id_show(struct device *dev, struct device_attribute *dev_attr,
136 		       char *buf)
137 {
138 	struct hv_device *hv_dev = device_to_hv_device(dev);
139 
140 	if (!hv_dev->channel)
141 		return -ENODEV;
142 	return sprintf(buf, "%d\n", hv_dev->channel->offermsg.child_relid);
143 }
144 static DEVICE_ATTR_RO(id);
145 
146 static ssize_t state_show(struct device *dev, struct device_attribute *dev_attr,
147 			  char *buf)
148 {
149 	struct hv_device *hv_dev = device_to_hv_device(dev);
150 
151 	if (!hv_dev->channel)
152 		return -ENODEV;
153 	return sprintf(buf, "%d\n", hv_dev->channel->state);
154 }
155 static DEVICE_ATTR_RO(state);
156 
157 static ssize_t monitor_id_show(struct device *dev,
158 			       struct device_attribute *dev_attr, char *buf)
159 {
160 	struct hv_device *hv_dev = device_to_hv_device(dev);
161 
162 	if (!hv_dev->channel)
163 		return -ENODEV;
164 	return sprintf(buf, "%d\n", hv_dev->channel->offermsg.monitorid);
165 }
166 static DEVICE_ATTR_RO(monitor_id);
167 
168 static ssize_t class_id_show(struct device *dev,
169 			       struct device_attribute *dev_attr, char *buf)
170 {
171 	struct hv_device *hv_dev = device_to_hv_device(dev);
172 
173 	if (!hv_dev->channel)
174 		return -ENODEV;
175 	return sprintf(buf, "{%pUl}\n",
176 		       hv_dev->channel->offermsg.offer.if_type.b);
177 }
178 static DEVICE_ATTR_RO(class_id);
179 
180 static ssize_t device_id_show(struct device *dev,
181 			      struct device_attribute *dev_attr, char *buf)
182 {
183 	struct hv_device *hv_dev = device_to_hv_device(dev);
184 
185 	if (!hv_dev->channel)
186 		return -ENODEV;
187 	return sprintf(buf, "{%pUl}\n",
188 		       hv_dev->channel->offermsg.offer.if_instance.b);
189 }
190 static DEVICE_ATTR_RO(device_id);
191 
192 static ssize_t modalias_show(struct device *dev,
193 			     struct device_attribute *dev_attr, char *buf)
194 {
195 	struct hv_device *hv_dev = device_to_hv_device(dev);
196 	char alias_name[VMBUS_ALIAS_LEN + 1];
197 
198 	print_alias_name(hv_dev, alias_name);
199 	return sprintf(buf, "vmbus:%s\n", alias_name);
200 }
201 static DEVICE_ATTR_RO(modalias);
202 
203 #ifdef CONFIG_NUMA
204 static ssize_t numa_node_show(struct device *dev,
205 			      struct device_attribute *attr, char *buf)
206 {
207 	struct hv_device *hv_dev = device_to_hv_device(dev);
208 
209 	if (!hv_dev->channel)
210 		return -ENODEV;
211 
212 	return sprintf(buf, "%d\n", hv_dev->channel->numa_node);
213 }
214 static DEVICE_ATTR_RO(numa_node);
215 #endif
216 
217 static ssize_t server_monitor_pending_show(struct device *dev,
218 					   struct device_attribute *dev_attr,
219 					   char *buf)
220 {
221 	struct hv_device *hv_dev = device_to_hv_device(dev);
222 
223 	if (!hv_dev->channel)
224 		return -ENODEV;
225 	return sprintf(buf, "%d\n",
226 		       channel_pending(hv_dev->channel,
227 				       vmbus_connection.monitor_pages[0]));
228 }
229 static DEVICE_ATTR_RO(server_monitor_pending);
230 
231 static ssize_t client_monitor_pending_show(struct device *dev,
232 					   struct device_attribute *dev_attr,
233 					   char *buf)
234 {
235 	struct hv_device *hv_dev = device_to_hv_device(dev);
236 
237 	if (!hv_dev->channel)
238 		return -ENODEV;
239 	return sprintf(buf, "%d\n",
240 		       channel_pending(hv_dev->channel,
241 				       vmbus_connection.monitor_pages[1]));
242 }
243 static DEVICE_ATTR_RO(client_monitor_pending);
244 
245 static ssize_t server_monitor_latency_show(struct device *dev,
246 					   struct device_attribute *dev_attr,
247 					   char *buf)
248 {
249 	struct hv_device *hv_dev = device_to_hv_device(dev);
250 
251 	if (!hv_dev->channel)
252 		return -ENODEV;
253 	return sprintf(buf, "%d\n",
254 		       channel_latency(hv_dev->channel,
255 				       vmbus_connection.monitor_pages[0]));
256 }
257 static DEVICE_ATTR_RO(server_monitor_latency);
258 
259 static ssize_t client_monitor_latency_show(struct device *dev,
260 					   struct device_attribute *dev_attr,
261 					   char *buf)
262 {
263 	struct hv_device *hv_dev = device_to_hv_device(dev);
264 
265 	if (!hv_dev->channel)
266 		return -ENODEV;
267 	return sprintf(buf, "%d\n",
268 		       channel_latency(hv_dev->channel,
269 				       vmbus_connection.monitor_pages[1]));
270 }
271 static DEVICE_ATTR_RO(client_monitor_latency);
272 
273 static ssize_t server_monitor_conn_id_show(struct device *dev,
274 					   struct device_attribute *dev_attr,
275 					   char *buf)
276 {
277 	struct hv_device *hv_dev = device_to_hv_device(dev);
278 
279 	if (!hv_dev->channel)
280 		return -ENODEV;
281 	return sprintf(buf, "%d\n",
282 		       channel_conn_id(hv_dev->channel,
283 				       vmbus_connection.monitor_pages[0]));
284 }
285 static DEVICE_ATTR_RO(server_monitor_conn_id);
286 
287 static ssize_t client_monitor_conn_id_show(struct device *dev,
288 					   struct device_attribute *dev_attr,
289 					   char *buf)
290 {
291 	struct hv_device *hv_dev = device_to_hv_device(dev);
292 
293 	if (!hv_dev->channel)
294 		return -ENODEV;
295 	return sprintf(buf, "%d\n",
296 		       channel_conn_id(hv_dev->channel,
297 				       vmbus_connection.monitor_pages[1]));
298 }
299 static DEVICE_ATTR_RO(client_monitor_conn_id);
300 
301 static ssize_t out_intr_mask_show(struct device *dev,
302 				  struct device_attribute *dev_attr, char *buf)
303 {
304 	struct hv_device *hv_dev = device_to_hv_device(dev);
305 	struct hv_ring_buffer_debug_info outbound;
306 	int ret;
307 
308 	if (!hv_dev->channel)
309 		return -ENODEV;
310 
311 	ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
312 					  &outbound);
313 	if (ret < 0)
314 		return ret;
315 
316 	return sprintf(buf, "%d\n", outbound.current_interrupt_mask);
317 }
318 static DEVICE_ATTR_RO(out_intr_mask);
319 
320 static ssize_t out_read_index_show(struct device *dev,
321 				   struct device_attribute *dev_attr, char *buf)
322 {
323 	struct hv_device *hv_dev = device_to_hv_device(dev);
324 	struct hv_ring_buffer_debug_info outbound;
325 	int ret;
326 
327 	if (!hv_dev->channel)
328 		return -ENODEV;
329 
330 	ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
331 					  &outbound);
332 	if (ret < 0)
333 		return ret;
334 	return sprintf(buf, "%d\n", outbound.current_read_index);
335 }
336 static DEVICE_ATTR_RO(out_read_index);
337 
338 static ssize_t out_write_index_show(struct device *dev,
339 				    struct device_attribute *dev_attr,
340 				    char *buf)
341 {
342 	struct hv_device *hv_dev = device_to_hv_device(dev);
343 	struct hv_ring_buffer_debug_info outbound;
344 	int ret;
345 
346 	if (!hv_dev->channel)
347 		return -ENODEV;
348 
349 	ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
350 					  &outbound);
351 	if (ret < 0)
352 		return ret;
353 	return sprintf(buf, "%d\n", outbound.current_write_index);
354 }
355 static DEVICE_ATTR_RO(out_write_index);
356 
357 static ssize_t out_read_bytes_avail_show(struct device *dev,
358 					 struct device_attribute *dev_attr,
359 					 char *buf)
360 {
361 	struct hv_device *hv_dev = device_to_hv_device(dev);
362 	struct hv_ring_buffer_debug_info outbound;
363 	int ret;
364 
365 	if (!hv_dev->channel)
366 		return -ENODEV;
367 
368 	ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
369 					  &outbound);
370 	if (ret < 0)
371 		return ret;
372 	return sprintf(buf, "%d\n", outbound.bytes_avail_toread);
373 }
374 static DEVICE_ATTR_RO(out_read_bytes_avail);
375 
376 static ssize_t out_write_bytes_avail_show(struct device *dev,
377 					  struct device_attribute *dev_attr,
378 					  char *buf)
379 {
380 	struct hv_device *hv_dev = device_to_hv_device(dev);
381 	struct hv_ring_buffer_debug_info outbound;
382 	int ret;
383 
384 	if (!hv_dev->channel)
385 		return -ENODEV;
386 
387 	ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
388 					  &outbound);
389 	if (ret < 0)
390 		return ret;
391 	return sprintf(buf, "%d\n", outbound.bytes_avail_towrite);
392 }
393 static DEVICE_ATTR_RO(out_write_bytes_avail);
394 
395 static ssize_t in_intr_mask_show(struct device *dev,
396 				 struct device_attribute *dev_attr, char *buf)
397 {
398 	struct hv_device *hv_dev = device_to_hv_device(dev);
399 	struct hv_ring_buffer_debug_info inbound;
400 	int ret;
401 
402 	if (!hv_dev->channel)
403 		return -ENODEV;
404 
405 	ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
406 	if (ret < 0)
407 		return ret;
408 
409 	return sprintf(buf, "%d\n", inbound.current_interrupt_mask);
410 }
411 static DEVICE_ATTR_RO(in_intr_mask);
412 
413 static ssize_t in_read_index_show(struct device *dev,
414 				  struct device_attribute *dev_attr, char *buf)
415 {
416 	struct hv_device *hv_dev = device_to_hv_device(dev);
417 	struct hv_ring_buffer_debug_info inbound;
418 	int ret;
419 
420 	if (!hv_dev->channel)
421 		return -ENODEV;
422 
423 	ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
424 	if (ret < 0)
425 		return ret;
426 
427 	return sprintf(buf, "%d\n", inbound.current_read_index);
428 }
429 static DEVICE_ATTR_RO(in_read_index);
430 
431 static ssize_t in_write_index_show(struct device *dev,
432 				   struct device_attribute *dev_attr, char *buf)
433 {
434 	struct hv_device *hv_dev = device_to_hv_device(dev);
435 	struct hv_ring_buffer_debug_info inbound;
436 	int ret;
437 
438 	if (!hv_dev->channel)
439 		return -ENODEV;
440 
441 	ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
442 	if (ret < 0)
443 		return ret;
444 
445 	return sprintf(buf, "%d\n", inbound.current_write_index);
446 }
447 static DEVICE_ATTR_RO(in_write_index);
448 
449 static ssize_t in_read_bytes_avail_show(struct device *dev,
450 					struct device_attribute *dev_attr,
451 					char *buf)
452 {
453 	struct hv_device *hv_dev = device_to_hv_device(dev);
454 	struct hv_ring_buffer_debug_info inbound;
455 	int ret;
456 
457 	if (!hv_dev->channel)
458 		return -ENODEV;
459 
460 	ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
461 	if (ret < 0)
462 		return ret;
463 
464 	return sprintf(buf, "%d\n", inbound.bytes_avail_toread);
465 }
466 static DEVICE_ATTR_RO(in_read_bytes_avail);
467 
468 static ssize_t in_write_bytes_avail_show(struct device *dev,
469 					 struct device_attribute *dev_attr,
470 					 char *buf)
471 {
472 	struct hv_device *hv_dev = device_to_hv_device(dev);
473 	struct hv_ring_buffer_debug_info inbound;
474 	int ret;
475 
476 	if (!hv_dev->channel)
477 		return -ENODEV;
478 
479 	ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
480 	if (ret < 0)
481 		return ret;
482 
483 	return sprintf(buf, "%d\n", inbound.bytes_avail_towrite);
484 }
485 static DEVICE_ATTR_RO(in_write_bytes_avail);
486 
487 static ssize_t channel_vp_mapping_show(struct device *dev,
488 				       struct device_attribute *dev_attr,
489 				       char *buf)
490 {
491 	struct hv_device *hv_dev = device_to_hv_device(dev);
492 	struct vmbus_channel *channel = hv_dev->channel, *cur_sc;
493 	unsigned long flags;
494 	int buf_size = PAGE_SIZE, n_written, tot_written;
495 	struct list_head *cur;
496 
497 	if (!channel)
498 		return -ENODEV;
499 
500 	tot_written = snprintf(buf, buf_size, "%u:%u\n",
501 		channel->offermsg.child_relid, channel->target_cpu);
502 
503 	spin_lock_irqsave(&channel->lock, flags);
504 
505 	list_for_each(cur, &channel->sc_list) {
506 		if (tot_written >= buf_size - 1)
507 			break;
508 
509 		cur_sc = list_entry(cur, struct vmbus_channel, sc_list);
510 		n_written = scnprintf(buf + tot_written,
511 				     buf_size - tot_written,
512 				     "%u:%u\n",
513 				     cur_sc->offermsg.child_relid,
514 				     cur_sc->target_cpu);
515 		tot_written += n_written;
516 	}
517 
518 	spin_unlock_irqrestore(&channel->lock, flags);
519 
520 	return tot_written;
521 }
522 static DEVICE_ATTR_RO(channel_vp_mapping);
523 
524 static ssize_t vendor_show(struct device *dev,
525 			   struct device_attribute *dev_attr,
526 			   char *buf)
527 {
528 	struct hv_device *hv_dev = device_to_hv_device(dev);
529 	return sprintf(buf, "0x%x\n", hv_dev->vendor_id);
530 }
531 static DEVICE_ATTR_RO(vendor);
532 
533 static ssize_t device_show(struct device *dev,
534 			   struct device_attribute *dev_attr,
535 			   char *buf)
536 {
537 	struct hv_device *hv_dev = device_to_hv_device(dev);
538 	return sprintf(buf, "0x%x\n", hv_dev->device_id);
539 }
540 static DEVICE_ATTR_RO(device);
541 
542 static ssize_t driver_override_store(struct device *dev,
543 				     struct device_attribute *attr,
544 				     const char *buf, size_t count)
545 {
546 	struct hv_device *hv_dev = device_to_hv_device(dev);
547 	char *driver_override, *old, *cp;
548 
549 	/* We need to keep extra room for a newline */
550 	if (count >= (PAGE_SIZE - 1))
551 		return -EINVAL;
552 
553 	driver_override = kstrndup(buf, count, GFP_KERNEL);
554 	if (!driver_override)
555 		return -ENOMEM;
556 
557 	cp = strchr(driver_override, '\n');
558 	if (cp)
559 		*cp = '\0';
560 
561 	device_lock(dev);
562 	old = hv_dev->driver_override;
563 	if (strlen(driver_override)) {
564 		hv_dev->driver_override = driver_override;
565 	} else {
566 		kfree(driver_override);
567 		hv_dev->driver_override = NULL;
568 	}
569 	device_unlock(dev);
570 
571 	kfree(old);
572 
573 	return count;
574 }
575 
576 static ssize_t driver_override_show(struct device *dev,
577 				    struct device_attribute *attr, char *buf)
578 {
579 	struct hv_device *hv_dev = device_to_hv_device(dev);
580 	ssize_t len;
581 
582 	device_lock(dev);
583 	len = snprintf(buf, PAGE_SIZE, "%s\n", hv_dev->driver_override);
584 	device_unlock(dev);
585 
586 	return len;
587 }
588 static DEVICE_ATTR_RW(driver_override);
589 
590 /* Set up per device attributes in /sys/bus/vmbus/devices/<bus device> */
591 static struct attribute *vmbus_dev_attrs[] = {
592 	&dev_attr_id.attr,
593 	&dev_attr_state.attr,
594 	&dev_attr_monitor_id.attr,
595 	&dev_attr_class_id.attr,
596 	&dev_attr_device_id.attr,
597 	&dev_attr_modalias.attr,
598 #ifdef CONFIG_NUMA
599 	&dev_attr_numa_node.attr,
600 #endif
601 	&dev_attr_server_monitor_pending.attr,
602 	&dev_attr_client_monitor_pending.attr,
603 	&dev_attr_server_monitor_latency.attr,
604 	&dev_attr_client_monitor_latency.attr,
605 	&dev_attr_server_monitor_conn_id.attr,
606 	&dev_attr_client_monitor_conn_id.attr,
607 	&dev_attr_out_intr_mask.attr,
608 	&dev_attr_out_read_index.attr,
609 	&dev_attr_out_write_index.attr,
610 	&dev_attr_out_read_bytes_avail.attr,
611 	&dev_attr_out_write_bytes_avail.attr,
612 	&dev_attr_in_intr_mask.attr,
613 	&dev_attr_in_read_index.attr,
614 	&dev_attr_in_write_index.attr,
615 	&dev_attr_in_read_bytes_avail.attr,
616 	&dev_attr_in_write_bytes_avail.attr,
617 	&dev_attr_channel_vp_mapping.attr,
618 	&dev_attr_vendor.attr,
619 	&dev_attr_device.attr,
620 	&dev_attr_driver_override.attr,
621 	NULL,
622 };
623 
624 /*
625  * Device-level attribute_group callback function. Returns the permission for
626  * each attribute, and returns 0 if an attribute is not visible.
627  */
628 static umode_t vmbus_dev_attr_is_visible(struct kobject *kobj,
629 					 struct attribute *attr, int idx)
630 {
631 	struct device *dev = kobj_to_dev(kobj);
632 	const struct hv_device *hv_dev = device_to_hv_device(dev);
633 
634 	/* Hide the monitor attributes if the monitor mechanism is not used. */
635 	if (!hv_dev->channel->offermsg.monitor_allocated &&
636 	    (attr == &dev_attr_monitor_id.attr ||
637 	     attr == &dev_attr_server_monitor_pending.attr ||
638 	     attr == &dev_attr_client_monitor_pending.attr ||
639 	     attr == &dev_attr_server_monitor_latency.attr ||
640 	     attr == &dev_attr_client_monitor_latency.attr ||
641 	     attr == &dev_attr_server_monitor_conn_id.attr ||
642 	     attr == &dev_attr_client_monitor_conn_id.attr))
643 		return 0;
644 
645 	return attr->mode;
646 }
647 
648 static const struct attribute_group vmbus_dev_group = {
649 	.attrs = vmbus_dev_attrs,
650 	.is_visible = vmbus_dev_attr_is_visible
651 };
652 __ATTRIBUTE_GROUPS(vmbus_dev);
653 
654 /*
655  * vmbus_uevent - add uevent for our device
656  *
657  * This routine is invoked when a device is added or removed on the vmbus to
658  * generate a uevent to udev in the userspace. The udev will then look at its
659  * rule and the uevent generated here to load the appropriate driver
660  *
661  * The alias string will be of the form vmbus:guid where guid is the string
662  * representation of the device guid (each byte of the guid will be
663  * represented with two hex characters.
664  */
665 static int vmbus_uevent(struct device *device, struct kobj_uevent_env *env)
666 {
667 	struct hv_device *dev = device_to_hv_device(device);
668 	int ret;
669 	char alias_name[VMBUS_ALIAS_LEN + 1];
670 
671 	print_alias_name(dev, alias_name);
672 	ret = add_uevent_var(env, "MODALIAS=vmbus:%s", alias_name);
673 	return ret;
674 }
675 
676 static const struct hv_vmbus_device_id *
677 hv_vmbus_dev_match(const struct hv_vmbus_device_id *id, const guid_t *guid)
678 {
679 	if (id == NULL)
680 		return NULL; /* empty device table */
681 
682 	for (; !guid_is_null(&id->guid); id++)
683 		if (guid_equal(&id->guid, guid))
684 			return id;
685 
686 	return NULL;
687 }
688 
689 static const struct hv_vmbus_device_id *
690 hv_vmbus_dynid_match(struct hv_driver *drv, const guid_t *guid)
691 {
692 	const struct hv_vmbus_device_id *id = NULL;
693 	struct vmbus_dynid *dynid;
694 
695 	spin_lock(&drv->dynids.lock);
696 	list_for_each_entry(dynid, &drv->dynids.list, node) {
697 		if (guid_equal(&dynid->id.guid, guid)) {
698 			id = &dynid->id;
699 			break;
700 		}
701 	}
702 	spin_unlock(&drv->dynids.lock);
703 
704 	return id;
705 }
706 
707 static const struct hv_vmbus_device_id vmbus_device_null;
708 
709 /*
710  * Return a matching hv_vmbus_device_id pointer.
711  * If there is no match, return NULL.
712  */
713 static const struct hv_vmbus_device_id *hv_vmbus_get_id(struct hv_driver *drv,
714 							struct hv_device *dev)
715 {
716 	const guid_t *guid = &dev->dev_type;
717 	const struct hv_vmbus_device_id *id;
718 
719 	/* When driver_override is set, only bind to the matching driver */
720 	if (dev->driver_override && strcmp(dev->driver_override, drv->name))
721 		return NULL;
722 
723 	/* Look at the dynamic ids first, before the static ones */
724 	id = hv_vmbus_dynid_match(drv, guid);
725 	if (!id)
726 		id = hv_vmbus_dev_match(drv->id_table, guid);
727 
728 	/* driver_override will always match, send a dummy id */
729 	if (!id && dev->driver_override)
730 		id = &vmbus_device_null;
731 
732 	return id;
733 }
734 
735 /* vmbus_add_dynid - add a new device ID to this driver and re-probe devices */
736 static int vmbus_add_dynid(struct hv_driver *drv, guid_t *guid)
737 {
738 	struct vmbus_dynid *dynid;
739 
740 	dynid = kzalloc(sizeof(*dynid), GFP_KERNEL);
741 	if (!dynid)
742 		return -ENOMEM;
743 
744 	dynid->id.guid = *guid;
745 
746 	spin_lock(&drv->dynids.lock);
747 	list_add_tail(&dynid->node, &drv->dynids.list);
748 	spin_unlock(&drv->dynids.lock);
749 
750 	return driver_attach(&drv->driver);
751 }
752 
753 static void vmbus_free_dynids(struct hv_driver *drv)
754 {
755 	struct vmbus_dynid *dynid, *n;
756 
757 	spin_lock(&drv->dynids.lock);
758 	list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
759 		list_del(&dynid->node);
760 		kfree(dynid);
761 	}
762 	spin_unlock(&drv->dynids.lock);
763 }
764 
765 /*
766  * store_new_id - sysfs frontend to vmbus_add_dynid()
767  *
768  * Allow GUIDs to be added to an existing driver via sysfs.
769  */
770 static ssize_t new_id_store(struct device_driver *driver, const char *buf,
771 			    size_t count)
772 {
773 	struct hv_driver *drv = drv_to_hv_drv(driver);
774 	guid_t guid;
775 	ssize_t retval;
776 
777 	retval = guid_parse(buf, &guid);
778 	if (retval)
779 		return retval;
780 
781 	if (hv_vmbus_dynid_match(drv, &guid))
782 		return -EEXIST;
783 
784 	retval = vmbus_add_dynid(drv, &guid);
785 	if (retval)
786 		return retval;
787 	return count;
788 }
789 static DRIVER_ATTR_WO(new_id);
790 
791 /*
792  * store_remove_id - remove a PCI device ID from this driver
793  *
794  * Removes a dynamic pci device ID to this driver.
795  */
796 static ssize_t remove_id_store(struct device_driver *driver, const char *buf,
797 			       size_t count)
798 {
799 	struct hv_driver *drv = drv_to_hv_drv(driver);
800 	struct vmbus_dynid *dynid, *n;
801 	guid_t guid;
802 	ssize_t retval;
803 
804 	retval = guid_parse(buf, &guid);
805 	if (retval)
806 		return retval;
807 
808 	retval = -ENODEV;
809 	spin_lock(&drv->dynids.lock);
810 	list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
811 		struct hv_vmbus_device_id *id = &dynid->id;
812 
813 		if (guid_equal(&id->guid, &guid)) {
814 			list_del(&dynid->node);
815 			kfree(dynid);
816 			retval = count;
817 			break;
818 		}
819 	}
820 	spin_unlock(&drv->dynids.lock);
821 
822 	return retval;
823 }
824 static DRIVER_ATTR_WO(remove_id);
825 
826 static struct attribute *vmbus_drv_attrs[] = {
827 	&driver_attr_new_id.attr,
828 	&driver_attr_remove_id.attr,
829 	NULL,
830 };
831 ATTRIBUTE_GROUPS(vmbus_drv);
832 
833 
834 /*
835  * vmbus_match - Attempt to match the specified device to the specified driver
836  */
837 static int vmbus_match(struct device *device, struct device_driver *driver)
838 {
839 	struct hv_driver *drv = drv_to_hv_drv(driver);
840 	struct hv_device *hv_dev = device_to_hv_device(device);
841 
842 	/* The hv_sock driver handles all hv_sock offers. */
843 	if (is_hvsock_channel(hv_dev->channel))
844 		return drv->hvsock;
845 
846 	if (hv_vmbus_get_id(drv, hv_dev))
847 		return 1;
848 
849 	return 0;
850 }
851 
852 /*
853  * vmbus_probe - Add the new vmbus's child device
854  */
855 static int vmbus_probe(struct device *child_device)
856 {
857 	int ret = 0;
858 	struct hv_driver *drv =
859 			drv_to_hv_drv(child_device->driver);
860 	struct hv_device *dev = device_to_hv_device(child_device);
861 	const struct hv_vmbus_device_id *dev_id;
862 
863 	dev_id = hv_vmbus_get_id(drv, dev);
864 	if (drv->probe) {
865 		ret = drv->probe(dev, dev_id);
866 		if (ret != 0)
867 			pr_err("probe failed for device %s (%d)\n",
868 			       dev_name(child_device), ret);
869 
870 	} else {
871 		pr_err("probe not set for driver %s\n",
872 		       dev_name(child_device));
873 		ret = -ENODEV;
874 	}
875 	return ret;
876 }
877 
878 /*
879  * vmbus_remove - Remove a vmbus device
880  */
881 static int vmbus_remove(struct device *child_device)
882 {
883 	struct hv_driver *drv;
884 	struct hv_device *dev = device_to_hv_device(child_device);
885 
886 	if (child_device->driver) {
887 		drv = drv_to_hv_drv(child_device->driver);
888 		if (drv->remove)
889 			drv->remove(dev);
890 	}
891 
892 	return 0;
893 }
894 
895 
896 /*
897  * vmbus_shutdown - Shutdown a vmbus device
898  */
899 static void vmbus_shutdown(struct device *child_device)
900 {
901 	struct hv_driver *drv;
902 	struct hv_device *dev = device_to_hv_device(child_device);
903 
904 
905 	/* The device may not be attached yet */
906 	if (!child_device->driver)
907 		return;
908 
909 	drv = drv_to_hv_drv(child_device->driver);
910 
911 	if (drv->shutdown)
912 		drv->shutdown(dev);
913 }
914 
915 #ifdef CONFIG_PM_SLEEP
916 /*
917  * vmbus_suspend - Suspend a vmbus device
918  */
919 static int vmbus_suspend(struct device *child_device)
920 {
921 	struct hv_driver *drv;
922 	struct hv_device *dev = device_to_hv_device(child_device);
923 
924 	/* The device may not be attached yet */
925 	if (!child_device->driver)
926 		return 0;
927 
928 	drv = drv_to_hv_drv(child_device->driver);
929 	if (!drv->suspend)
930 		return -EOPNOTSUPP;
931 
932 	return drv->suspend(dev);
933 }
934 
935 /*
936  * vmbus_resume - Resume a vmbus device
937  */
938 static int vmbus_resume(struct device *child_device)
939 {
940 	struct hv_driver *drv;
941 	struct hv_device *dev = device_to_hv_device(child_device);
942 
943 	/* The device may not be attached yet */
944 	if (!child_device->driver)
945 		return 0;
946 
947 	drv = drv_to_hv_drv(child_device->driver);
948 	if (!drv->resume)
949 		return -EOPNOTSUPP;
950 
951 	return drv->resume(dev);
952 }
953 #endif /* CONFIG_PM_SLEEP */
954 
955 /*
956  * vmbus_device_release - Final callback release of the vmbus child device
957  */
958 static void vmbus_device_release(struct device *device)
959 {
960 	struct hv_device *hv_dev = device_to_hv_device(device);
961 	struct vmbus_channel *channel = hv_dev->channel;
962 
963 	hv_debug_rm_dev_dir(hv_dev);
964 
965 	mutex_lock(&vmbus_connection.channel_mutex);
966 	hv_process_channel_removal(channel);
967 	mutex_unlock(&vmbus_connection.channel_mutex);
968 	kfree(hv_dev);
969 }
970 
971 /*
972  * Note: we must use SET_NOIRQ_SYSTEM_SLEEP_PM_OPS rather than
973  * SET_SYSTEM_SLEEP_PM_OPS: see the comment before vmbus_bus_pm.
974  */
975 static const struct dev_pm_ops vmbus_pm = {
976 	SET_NOIRQ_SYSTEM_SLEEP_PM_OPS(vmbus_suspend, vmbus_resume)
977 };
978 
979 /* The one and only one */
980 static struct bus_type  hv_bus = {
981 	.name =		"vmbus",
982 	.match =		vmbus_match,
983 	.shutdown =		vmbus_shutdown,
984 	.remove =		vmbus_remove,
985 	.probe =		vmbus_probe,
986 	.uevent =		vmbus_uevent,
987 	.dev_groups =		vmbus_dev_groups,
988 	.drv_groups =		vmbus_drv_groups,
989 	.pm =			&vmbus_pm,
990 };
991 
992 struct onmessage_work_context {
993 	struct work_struct work;
994 	struct hv_message msg;
995 };
996 
997 static void vmbus_onmessage_work(struct work_struct *work)
998 {
999 	struct onmessage_work_context *ctx;
1000 
1001 	/* Do not process messages if we're in DISCONNECTED state */
1002 	if (vmbus_connection.conn_state == DISCONNECTED)
1003 		return;
1004 
1005 	ctx = container_of(work, struct onmessage_work_context,
1006 			   work);
1007 	vmbus_onmessage(&ctx->msg);
1008 	kfree(ctx);
1009 }
1010 
1011 void vmbus_on_msg_dpc(unsigned long data)
1012 {
1013 	struct hv_per_cpu_context *hv_cpu = (void *)data;
1014 	void *page_addr = hv_cpu->synic_message_page;
1015 	struct hv_message *msg = (struct hv_message *)page_addr +
1016 				  VMBUS_MESSAGE_SINT;
1017 	struct vmbus_channel_message_header *hdr;
1018 	const struct vmbus_channel_message_table_entry *entry;
1019 	struct onmessage_work_context *ctx;
1020 	u32 message_type = msg->header.message_type;
1021 
1022 	if (message_type == HVMSG_NONE)
1023 		/* no msg */
1024 		return;
1025 
1026 	hdr = (struct vmbus_channel_message_header *)msg->u.payload;
1027 
1028 	trace_vmbus_on_msg_dpc(hdr);
1029 
1030 	if (hdr->msgtype >= CHANNELMSG_COUNT) {
1031 		WARN_ONCE(1, "unknown msgtype=%d\n", hdr->msgtype);
1032 		goto msg_handled;
1033 	}
1034 
1035 	entry = &channel_message_table[hdr->msgtype];
1036 
1037 	if (!entry->message_handler)
1038 		goto msg_handled;
1039 
1040 	if (entry->handler_type	== VMHT_BLOCKING) {
1041 		ctx = kmalloc(sizeof(*ctx), GFP_ATOMIC);
1042 		if (ctx == NULL)
1043 			return;
1044 
1045 		INIT_WORK(&ctx->work, vmbus_onmessage_work);
1046 		memcpy(&ctx->msg, msg, sizeof(*msg));
1047 
1048 		/*
1049 		 * The host can generate a rescind message while we
1050 		 * may still be handling the original offer. We deal with
1051 		 * this condition by ensuring the processing is done on the
1052 		 * same CPU.
1053 		 */
1054 		switch (hdr->msgtype) {
1055 		case CHANNELMSG_RESCIND_CHANNELOFFER:
1056 			/*
1057 			 * If we are handling the rescind message;
1058 			 * schedule the work on the global work queue.
1059 			 */
1060 			schedule_work_on(vmbus_connection.connect_cpu,
1061 					 &ctx->work);
1062 			break;
1063 
1064 		case CHANNELMSG_OFFERCHANNEL:
1065 			atomic_inc(&vmbus_connection.offer_in_progress);
1066 			queue_work_on(vmbus_connection.connect_cpu,
1067 				      vmbus_connection.work_queue,
1068 				      &ctx->work);
1069 			break;
1070 
1071 		default:
1072 			queue_work(vmbus_connection.work_queue, &ctx->work);
1073 		}
1074 	} else
1075 		entry->message_handler(hdr);
1076 
1077 msg_handled:
1078 	vmbus_signal_eom(msg, message_type);
1079 }
1080 
1081 #ifdef CONFIG_PM_SLEEP
1082 /*
1083  * Fake RESCIND_CHANNEL messages to clean up hv_sock channels by force for
1084  * hibernation, because hv_sock connections can not persist across hibernation.
1085  */
1086 static void vmbus_force_channel_rescinded(struct vmbus_channel *channel)
1087 {
1088 	struct onmessage_work_context *ctx;
1089 	struct vmbus_channel_rescind_offer *rescind;
1090 
1091 	WARN_ON(!is_hvsock_channel(channel));
1092 
1093 	/*
1094 	 * sizeof(*ctx) is small and the allocation should really not fail,
1095 	 * otherwise the state of the hv_sock connections ends up in limbo.
1096 	 */
1097 	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL | __GFP_NOFAIL);
1098 
1099 	/*
1100 	 * So far, these are not really used by Linux. Just set them to the
1101 	 * reasonable values conforming to the definitions of the fields.
1102 	 */
1103 	ctx->msg.header.message_type = 1;
1104 	ctx->msg.header.payload_size = sizeof(*rescind);
1105 
1106 	/* These values are actually used by Linux. */
1107 	rescind = (struct vmbus_channel_rescind_offer *)ctx->msg.u.payload;
1108 	rescind->header.msgtype = CHANNELMSG_RESCIND_CHANNELOFFER;
1109 	rescind->child_relid = channel->offermsg.child_relid;
1110 
1111 	INIT_WORK(&ctx->work, vmbus_onmessage_work);
1112 
1113 	queue_work_on(vmbus_connection.connect_cpu,
1114 		      vmbus_connection.work_queue,
1115 		      &ctx->work);
1116 }
1117 #endif /* CONFIG_PM_SLEEP */
1118 
1119 /*
1120  * Direct callback for channels using other deferred processing
1121  */
1122 static void vmbus_channel_isr(struct vmbus_channel *channel)
1123 {
1124 	void (*callback_fn)(void *);
1125 
1126 	callback_fn = READ_ONCE(channel->onchannel_callback);
1127 	if (likely(callback_fn != NULL))
1128 		(*callback_fn)(channel->channel_callback_context);
1129 }
1130 
1131 /*
1132  * Schedule all channels with events pending
1133  */
1134 static void vmbus_chan_sched(struct hv_per_cpu_context *hv_cpu)
1135 {
1136 	unsigned long *recv_int_page;
1137 	u32 maxbits, relid;
1138 
1139 	if (vmbus_proto_version < VERSION_WIN8) {
1140 		maxbits = MAX_NUM_CHANNELS_SUPPORTED;
1141 		recv_int_page = vmbus_connection.recv_int_page;
1142 	} else {
1143 		/*
1144 		 * When the host is win8 and beyond, the event page
1145 		 * can be directly checked to get the id of the channel
1146 		 * that has the interrupt pending.
1147 		 */
1148 		void *page_addr = hv_cpu->synic_event_page;
1149 		union hv_synic_event_flags *event
1150 			= (union hv_synic_event_flags *)page_addr +
1151 						 VMBUS_MESSAGE_SINT;
1152 
1153 		maxbits = HV_EVENT_FLAGS_COUNT;
1154 		recv_int_page = event->flags;
1155 	}
1156 
1157 	if (unlikely(!recv_int_page))
1158 		return;
1159 
1160 	for_each_set_bit(relid, recv_int_page, maxbits) {
1161 		struct vmbus_channel *channel;
1162 
1163 		if (!sync_test_and_clear_bit(relid, recv_int_page))
1164 			continue;
1165 
1166 		/* Special case - vmbus channel protocol msg */
1167 		if (relid == 0)
1168 			continue;
1169 
1170 		rcu_read_lock();
1171 
1172 		/* Find channel based on relid */
1173 		list_for_each_entry_rcu(channel, &hv_cpu->chan_list, percpu_list) {
1174 			if (channel->offermsg.child_relid != relid)
1175 				continue;
1176 
1177 			if (channel->rescind)
1178 				continue;
1179 
1180 			trace_vmbus_chan_sched(channel);
1181 
1182 			++channel->interrupts;
1183 
1184 			switch (channel->callback_mode) {
1185 			case HV_CALL_ISR:
1186 				vmbus_channel_isr(channel);
1187 				break;
1188 
1189 			case HV_CALL_BATCHED:
1190 				hv_begin_read(&channel->inbound);
1191 				/* fallthrough */
1192 			case HV_CALL_DIRECT:
1193 				tasklet_schedule(&channel->callback_event);
1194 			}
1195 		}
1196 
1197 		rcu_read_unlock();
1198 	}
1199 }
1200 
1201 static void vmbus_isr(void)
1202 {
1203 	struct hv_per_cpu_context *hv_cpu
1204 		= this_cpu_ptr(hv_context.cpu_context);
1205 	void *page_addr = hv_cpu->synic_event_page;
1206 	struct hv_message *msg;
1207 	union hv_synic_event_flags *event;
1208 	bool handled = false;
1209 
1210 	if (unlikely(page_addr == NULL))
1211 		return;
1212 
1213 	event = (union hv_synic_event_flags *)page_addr +
1214 					 VMBUS_MESSAGE_SINT;
1215 	/*
1216 	 * Check for events before checking for messages. This is the order
1217 	 * in which events and messages are checked in Windows guests on
1218 	 * Hyper-V, and the Windows team suggested we do the same.
1219 	 */
1220 
1221 	if ((vmbus_proto_version == VERSION_WS2008) ||
1222 		(vmbus_proto_version == VERSION_WIN7)) {
1223 
1224 		/* Since we are a child, we only need to check bit 0 */
1225 		if (sync_test_and_clear_bit(0, event->flags))
1226 			handled = true;
1227 	} else {
1228 		/*
1229 		 * Our host is win8 or above. The signaling mechanism
1230 		 * has changed and we can directly look at the event page.
1231 		 * If bit n is set then we have an interrup on the channel
1232 		 * whose id is n.
1233 		 */
1234 		handled = true;
1235 	}
1236 
1237 	if (handled)
1238 		vmbus_chan_sched(hv_cpu);
1239 
1240 	page_addr = hv_cpu->synic_message_page;
1241 	msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
1242 
1243 	/* Check if there are actual msgs to be processed */
1244 	if (msg->header.message_type != HVMSG_NONE) {
1245 		if (msg->header.message_type == HVMSG_TIMER_EXPIRED) {
1246 			hv_stimer0_isr();
1247 			vmbus_signal_eom(msg, HVMSG_TIMER_EXPIRED);
1248 		} else
1249 			tasklet_schedule(&hv_cpu->msg_dpc);
1250 	}
1251 
1252 	add_interrupt_randomness(HYPERVISOR_CALLBACK_VECTOR, 0);
1253 }
1254 
1255 /*
1256  * Boolean to control whether to report panic messages over Hyper-V.
1257  *
1258  * It can be set via /proc/sys/kernel/hyperv/record_panic_msg
1259  */
1260 static int sysctl_record_panic_msg = 1;
1261 
1262 /*
1263  * Callback from kmsg_dump. Grab as much as possible from the end of the kmsg
1264  * buffer and call into Hyper-V to transfer the data.
1265  */
1266 static void hv_kmsg_dump(struct kmsg_dumper *dumper,
1267 			 enum kmsg_dump_reason reason)
1268 {
1269 	size_t bytes_written;
1270 	phys_addr_t panic_pa;
1271 
1272 	/* We are only interested in panics. */
1273 	if ((reason != KMSG_DUMP_PANIC) || (!sysctl_record_panic_msg))
1274 		return;
1275 
1276 	panic_pa = virt_to_phys(hv_panic_page);
1277 
1278 	/*
1279 	 * Write dump contents to the page. No need to synchronize; panic should
1280 	 * be single-threaded.
1281 	 */
1282 	kmsg_dump_get_buffer(dumper, true, hv_panic_page, HV_HYP_PAGE_SIZE,
1283 			     &bytes_written);
1284 	if (bytes_written)
1285 		hyperv_report_panic_msg(panic_pa, bytes_written);
1286 }
1287 
1288 static struct kmsg_dumper hv_kmsg_dumper = {
1289 	.dump = hv_kmsg_dump,
1290 };
1291 
1292 static struct ctl_table_header *hv_ctl_table_hdr;
1293 
1294 /*
1295  * sysctl option to allow the user to control whether kmsg data should be
1296  * reported to Hyper-V on panic.
1297  */
1298 static struct ctl_table hv_ctl_table[] = {
1299 	{
1300 		.procname       = "hyperv_record_panic_msg",
1301 		.data           = &sysctl_record_panic_msg,
1302 		.maxlen         = sizeof(int),
1303 		.mode           = 0644,
1304 		.proc_handler   = proc_dointvec_minmax,
1305 		.extra1		= SYSCTL_ZERO,
1306 		.extra2		= SYSCTL_ONE
1307 	},
1308 	{}
1309 };
1310 
1311 static struct ctl_table hv_root_table[] = {
1312 	{
1313 		.procname	= "kernel",
1314 		.mode		= 0555,
1315 		.child		= hv_ctl_table
1316 	},
1317 	{}
1318 };
1319 
1320 /*
1321  * vmbus_bus_init -Main vmbus driver initialization routine.
1322  *
1323  * Here, we
1324  *	- initialize the vmbus driver context
1325  *	- invoke the vmbus hv main init routine
1326  *	- retrieve the channel offers
1327  */
1328 static int vmbus_bus_init(void)
1329 {
1330 	int ret;
1331 
1332 	/* Hypervisor initialization...setup hypercall page..etc */
1333 	ret = hv_init();
1334 	if (ret != 0) {
1335 		pr_err("Unable to initialize the hypervisor - 0x%x\n", ret);
1336 		return ret;
1337 	}
1338 
1339 	ret = bus_register(&hv_bus);
1340 	if (ret)
1341 		return ret;
1342 
1343 	hv_setup_vmbus_irq(vmbus_isr);
1344 
1345 	ret = hv_synic_alloc();
1346 	if (ret)
1347 		goto err_alloc;
1348 
1349 	/*
1350 	 * Initialize the per-cpu interrupt state and stimer state.
1351 	 * Then connect to the host.
1352 	 */
1353 	ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "hyperv/vmbus:online",
1354 				hv_synic_init, hv_synic_cleanup);
1355 	if (ret < 0)
1356 		goto err_cpuhp;
1357 	hyperv_cpuhp_online = ret;
1358 
1359 	ret = vmbus_connect();
1360 	if (ret)
1361 		goto err_connect;
1362 
1363 	/*
1364 	 * Only register if the crash MSRs are available
1365 	 */
1366 	if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1367 		u64 hyperv_crash_ctl;
1368 		/*
1369 		 * Sysctl registration is not fatal, since by default
1370 		 * reporting is enabled.
1371 		 */
1372 		hv_ctl_table_hdr = register_sysctl_table(hv_root_table);
1373 		if (!hv_ctl_table_hdr)
1374 			pr_err("Hyper-V: sysctl table register error");
1375 
1376 		/*
1377 		 * Register for panic kmsg callback only if the right
1378 		 * capability is supported by the hypervisor.
1379 		 */
1380 		hv_get_crash_ctl(hyperv_crash_ctl);
1381 		if (hyperv_crash_ctl & HV_CRASH_CTL_CRASH_NOTIFY_MSG) {
1382 			hv_panic_page = (void *)hv_alloc_hyperv_zeroed_page();
1383 			if (hv_panic_page) {
1384 				ret = kmsg_dump_register(&hv_kmsg_dumper);
1385 				if (ret)
1386 					pr_err("Hyper-V: kmsg dump register "
1387 						"error 0x%x\n", ret);
1388 			} else
1389 				pr_err("Hyper-V: panic message page memory "
1390 					"allocation failed");
1391 		}
1392 
1393 		register_die_notifier(&hyperv_die_block);
1394 		atomic_notifier_chain_register(&panic_notifier_list,
1395 					       &hyperv_panic_block);
1396 	}
1397 
1398 	vmbus_request_offers();
1399 
1400 	return 0;
1401 
1402 err_connect:
1403 	cpuhp_remove_state(hyperv_cpuhp_online);
1404 err_cpuhp:
1405 	hv_synic_free();
1406 err_alloc:
1407 	hv_remove_vmbus_irq();
1408 
1409 	bus_unregister(&hv_bus);
1410 	hv_free_hyperv_page((unsigned long)hv_panic_page);
1411 	unregister_sysctl_table(hv_ctl_table_hdr);
1412 	hv_ctl_table_hdr = NULL;
1413 	return ret;
1414 }
1415 
1416 /**
1417  * __vmbus_child_driver_register() - Register a vmbus's driver
1418  * @hv_driver: Pointer to driver structure you want to register
1419  * @owner: owner module of the drv
1420  * @mod_name: module name string
1421  *
1422  * Registers the given driver with Linux through the 'driver_register()' call
1423  * and sets up the hyper-v vmbus handling for this driver.
1424  * It will return the state of the 'driver_register()' call.
1425  *
1426  */
1427 int __vmbus_driver_register(struct hv_driver *hv_driver, struct module *owner, const char *mod_name)
1428 {
1429 	int ret;
1430 
1431 	pr_info("registering driver %s\n", hv_driver->name);
1432 
1433 	ret = vmbus_exists();
1434 	if (ret < 0)
1435 		return ret;
1436 
1437 	hv_driver->driver.name = hv_driver->name;
1438 	hv_driver->driver.owner = owner;
1439 	hv_driver->driver.mod_name = mod_name;
1440 	hv_driver->driver.bus = &hv_bus;
1441 
1442 	spin_lock_init(&hv_driver->dynids.lock);
1443 	INIT_LIST_HEAD(&hv_driver->dynids.list);
1444 
1445 	ret = driver_register(&hv_driver->driver);
1446 
1447 	return ret;
1448 }
1449 EXPORT_SYMBOL_GPL(__vmbus_driver_register);
1450 
1451 /**
1452  * vmbus_driver_unregister() - Unregister a vmbus's driver
1453  * @hv_driver: Pointer to driver structure you want to
1454  *             un-register
1455  *
1456  * Un-register the given driver that was previous registered with a call to
1457  * vmbus_driver_register()
1458  */
1459 void vmbus_driver_unregister(struct hv_driver *hv_driver)
1460 {
1461 	pr_info("unregistering driver %s\n", hv_driver->name);
1462 
1463 	if (!vmbus_exists()) {
1464 		driver_unregister(&hv_driver->driver);
1465 		vmbus_free_dynids(hv_driver);
1466 	}
1467 }
1468 EXPORT_SYMBOL_GPL(vmbus_driver_unregister);
1469 
1470 
1471 /*
1472  * Called when last reference to channel is gone.
1473  */
1474 static void vmbus_chan_release(struct kobject *kobj)
1475 {
1476 	struct vmbus_channel *channel
1477 		= container_of(kobj, struct vmbus_channel, kobj);
1478 
1479 	kfree_rcu(channel, rcu);
1480 }
1481 
1482 struct vmbus_chan_attribute {
1483 	struct attribute attr;
1484 	ssize_t (*show)(struct vmbus_channel *chan, char *buf);
1485 	ssize_t (*store)(struct vmbus_channel *chan,
1486 			 const char *buf, size_t count);
1487 };
1488 #define VMBUS_CHAN_ATTR(_name, _mode, _show, _store) \
1489 	struct vmbus_chan_attribute chan_attr_##_name \
1490 		= __ATTR(_name, _mode, _show, _store)
1491 #define VMBUS_CHAN_ATTR_RW(_name) \
1492 	struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RW(_name)
1493 #define VMBUS_CHAN_ATTR_RO(_name) \
1494 	struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RO(_name)
1495 #define VMBUS_CHAN_ATTR_WO(_name) \
1496 	struct vmbus_chan_attribute chan_attr_##_name = __ATTR_WO(_name)
1497 
1498 static ssize_t vmbus_chan_attr_show(struct kobject *kobj,
1499 				    struct attribute *attr, char *buf)
1500 {
1501 	const struct vmbus_chan_attribute *attribute
1502 		= container_of(attr, struct vmbus_chan_attribute, attr);
1503 	struct vmbus_channel *chan
1504 		= container_of(kobj, struct vmbus_channel, kobj);
1505 
1506 	if (!attribute->show)
1507 		return -EIO;
1508 
1509 	return attribute->show(chan, buf);
1510 }
1511 
1512 static const struct sysfs_ops vmbus_chan_sysfs_ops = {
1513 	.show = vmbus_chan_attr_show,
1514 };
1515 
1516 static ssize_t out_mask_show(struct vmbus_channel *channel, char *buf)
1517 {
1518 	struct hv_ring_buffer_info *rbi = &channel->outbound;
1519 	ssize_t ret;
1520 
1521 	mutex_lock(&rbi->ring_buffer_mutex);
1522 	if (!rbi->ring_buffer) {
1523 		mutex_unlock(&rbi->ring_buffer_mutex);
1524 		return -EINVAL;
1525 	}
1526 
1527 	ret = sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1528 	mutex_unlock(&rbi->ring_buffer_mutex);
1529 	return ret;
1530 }
1531 static VMBUS_CHAN_ATTR_RO(out_mask);
1532 
1533 static ssize_t in_mask_show(struct vmbus_channel *channel, char *buf)
1534 {
1535 	struct hv_ring_buffer_info *rbi = &channel->inbound;
1536 	ssize_t ret;
1537 
1538 	mutex_lock(&rbi->ring_buffer_mutex);
1539 	if (!rbi->ring_buffer) {
1540 		mutex_unlock(&rbi->ring_buffer_mutex);
1541 		return -EINVAL;
1542 	}
1543 
1544 	ret = sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1545 	mutex_unlock(&rbi->ring_buffer_mutex);
1546 	return ret;
1547 }
1548 static VMBUS_CHAN_ATTR_RO(in_mask);
1549 
1550 static ssize_t read_avail_show(struct vmbus_channel *channel, char *buf)
1551 {
1552 	struct hv_ring_buffer_info *rbi = &channel->inbound;
1553 	ssize_t ret;
1554 
1555 	mutex_lock(&rbi->ring_buffer_mutex);
1556 	if (!rbi->ring_buffer) {
1557 		mutex_unlock(&rbi->ring_buffer_mutex);
1558 		return -EINVAL;
1559 	}
1560 
1561 	ret = sprintf(buf, "%u\n", hv_get_bytes_to_read(rbi));
1562 	mutex_unlock(&rbi->ring_buffer_mutex);
1563 	return ret;
1564 }
1565 static VMBUS_CHAN_ATTR_RO(read_avail);
1566 
1567 static ssize_t write_avail_show(struct vmbus_channel *channel, char *buf)
1568 {
1569 	struct hv_ring_buffer_info *rbi = &channel->outbound;
1570 	ssize_t ret;
1571 
1572 	mutex_lock(&rbi->ring_buffer_mutex);
1573 	if (!rbi->ring_buffer) {
1574 		mutex_unlock(&rbi->ring_buffer_mutex);
1575 		return -EINVAL;
1576 	}
1577 
1578 	ret = sprintf(buf, "%u\n", hv_get_bytes_to_write(rbi));
1579 	mutex_unlock(&rbi->ring_buffer_mutex);
1580 	return ret;
1581 }
1582 static VMBUS_CHAN_ATTR_RO(write_avail);
1583 
1584 static ssize_t show_target_cpu(struct vmbus_channel *channel, char *buf)
1585 {
1586 	return sprintf(buf, "%u\n", channel->target_cpu);
1587 }
1588 static VMBUS_CHAN_ATTR(cpu, S_IRUGO, show_target_cpu, NULL);
1589 
1590 static ssize_t channel_pending_show(struct vmbus_channel *channel,
1591 				    char *buf)
1592 {
1593 	return sprintf(buf, "%d\n",
1594 		       channel_pending(channel,
1595 				       vmbus_connection.monitor_pages[1]));
1596 }
1597 static VMBUS_CHAN_ATTR(pending, S_IRUGO, channel_pending_show, NULL);
1598 
1599 static ssize_t channel_latency_show(struct vmbus_channel *channel,
1600 				    char *buf)
1601 {
1602 	return sprintf(buf, "%d\n",
1603 		       channel_latency(channel,
1604 				       vmbus_connection.monitor_pages[1]));
1605 }
1606 static VMBUS_CHAN_ATTR(latency, S_IRUGO, channel_latency_show, NULL);
1607 
1608 static ssize_t channel_interrupts_show(struct vmbus_channel *channel, char *buf)
1609 {
1610 	return sprintf(buf, "%llu\n", channel->interrupts);
1611 }
1612 static VMBUS_CHAN_ATTR(interrupts, S_IRUGO, channel_interrupts_show, NULL);
1613 
1614 static ssize_t channel_events_show(struct vmbus_channel *channel, char *buf)
1615 {
1616 	return sprintf(buf, "%llu\n", channel->sig_events);
1617 }
1618 static VMBUS_CHAN_ATTR(events, S_IRUGO, channel_events_show, NULL);
1619 
1620 static ssize_t channel_intr_in_full_show(struct vmbus_channel *channel,
1621 					 char *buf)
1622 {
1623 	return sprintf(buf, "%llu\n",
1624 		       (unsigned long long)channel->intr_in_full);
1625 }
1626 static VMBUS_CHAN_ATTR(intr_in_full, 0444, channel_intr_in_full_show, NULL);
1627 
1628 static ssize_t channel_intr_out_empty_show(struct vmbus_channel *channel,
1629 					   char *buf)
1630 {
1631 	return sprintf(buf, "%llu\n",
1632 		       (unsigned long long)channel->intr_out_empty);
1633 }
1634 static VMBUS_CHAN_ATTR(intr_out_empty, 0444, channel_intr_out_empty_show, NULL);
1635 
1636 static ssize_t channel_out_full_first_show(struct vmbus_channel *channel,
1637 					   char *buf)
1638 {
1639 	return sprintf(buf, "%llu\n",
1640 		       (unsigned long long)channel->out_full_first);
1641 }
1642 static VMBUS_CHAN_ATTR(out_full_first, 0444, channel_out_full_first_show, NULL);
1643 
1644 static ssize_t channel_out_full_total_show(struct vmbus_channel *channel,
1645 					   char *buf)
1646 {
1647 	return sprintf(buf, "%llu\n",
1648 		       (unsigned long long)channel->out_full_total);
1649 }
1650 static VMBUS_CHAN_ATTR(out_full_total, 0444, channel_out_full_total_show, NULL);
1651 
1652 static ssize_t subchannel_monitor_id_show(struct vmbus_channel *channel,
1653 					  char *buf)
1654 {
1655 	return sprintf(buf, "%u\n", channel->offermsg.monitorid);
1656 }
1657 static VMBUS_CHAN_ATTR(monitor_id, S_IRUGO, subchannel_monitor_id_show, NULL);
1658 
1659 static ssize_t subchannel_id_show(struct vmbus_channel *channel,
1660 				  char *buf)
1661 {
1662 	return sprintf(buf, "%u\n",
1663 		       channel->offermsg.offer.sub_channel_index);
1664 }
1665 static VMBUS_CHAN_ATTR_RO(subchannel_id);
1666 
1667 static struct attribute *vmbus_chan_attrs[] = {
1668 	&chan_attr_out_mask.attr,
1669 	&chan_attr_in_mask.attr,
1670 	&chan_attr_read_avail.attr,
1671 	&chan_attr_write_avail.attr,
1672 	&chan_attr_cpu.attr,
1673 	&chan_attr_pending.attr,
1674 	&chan_attr_latency.attr,
1675 	&chan_attr_interrupts.attr,
1676 	&chan_attr_events.attr,
1677 	&chan_attr_intr_in_full.attr,
1678 	&chan_attr_intr_out_empty.attr,
1679 	&chan_attr_out_full_first.attr,
1680 	&chan_attr_out_full_total.attr,
1681 	&chan_attr_monitor_id.attr,
1682 	&chan_attr_subchannel_id.attr,
1683 	NULL
1684 };
1685 
1686 /*
1687  * Channel-level attribute_group callback function. Returns the permission for
1688  * each attribute, and returns 0 if an attribute is not visible.
1689  */
1690 static umode_t vmbus_chan_attr_is_visible(struct kobject *kobj,
1691 					  struct attribute *attr, int idx)
1692 {
1693 	const struct vmbus_channel *channel =
1694 		container_of(kobj, struct vmbus_channel, kobj);
1695 
1696 	/* Hide the monitor attributes if the monitor mechanism is not used. */
1697 	if (!channel->offermsg.monitor_allocated &&
1698 	    (attr == &chan_attr_pending.attr ||
1699 	     attr == &chan_attr_latency.attr ||
1700 	     attr == &chan_attr_monitor_id.attr))
1701 		return 0;
1702 
1703 	return attr->mode;
1704 }
1705 
1706 static struct attribute_group vmbus_chan_group = {
1707 	.attrs = vmbus_chan_attrs,
1708 	.is_visible = vmbus_chan_attr_is_visible
1709 };
1710 
1711 static struct kobj_type vmbus_chan_ktype = {
1712 	.sysfs_ops = &vmbus_chan_sysfs_ops,
1713 	.release = vmbus_chan_release,
1714 };
1715 
1716 /*
1717  * vmbus_add_channel_kobj - setup a sub-directory under device/channels
1718  */
1719 int vmbus_add_channel_kobj(struct hv_device *dev, struct vmbus_channel *channel)
1720 {
1721 	const struct device *device = &dev->device;
1722 	struct kobject *kobj = &channel->kobj;
1723 	u32 relid = channel->offermsg.child_relid;
1724 	int ret;
1725 
1726 	kobj->kset = dev->channels_kset;
1727 	ret = kobject_init_and_add(kobj, &vmbus_chan_ktype, NULL,
1728 				   "%u", relid);
1729 	if (ret)
1730 		return ret;
1731 
1732 	ret = sysfs_create_group(kobj, &vmbus_chan_group);
1733 
1734 	if (ret) {
1735 		/*
1736 		 * The calling functions' error handling paths will cleanup the
1737 		 * empty channel directory.
1738 		 */
1739 		dev_err(device, "Unable to set up channel sysfs files\n");
1740 		return ret;
1741 	}
1742 
1743 	kobject_uevent(kobj, KOBJ_ADD);
1744 
1745 	return 0;
1746 }
1747 
1748 /*
1749  * vmbus_remove_channel_attr_group - remove the channel's attribute group
1750  */
1751 void vmbus_remove_channel_attr_group(struct vmbus_channel *channel)
1752 {
1753 	sysfs_remove_group(&channel->kobj, &vmbus_chan_group);
1754 }
1755 
1756 /*
1757  * vmbus_device_create - Creates and registers a new child device
1758  * on the vmbus.
1759  */
1760 struct hv_device *vmbus_device_create(const guid_t *type,
1761 				      const guid_t *instance,
1762 				      struct vmbus_channel *channel)
1763 {
1764 	struct hv_device *child_device_obj;
1765 
1766 	child_device_obj = kzalloc(sizeof(struct hv_device), GFP_KERNEL);
1767 	if (!child_device_obj) {
1768 		pr_err("Unable to allocate device object for child device\n");
1769 		return NULL;
1770 	}
1771 
1772 	child_device_obj->channel = channel;
1773 	guid_copy(&child_device_obj->dev_type, type);
1774 	guid_copy(&child_device_obj->dev_instance, instance);
1775 	child_device_obj->vendor_id = 0x1414; /* MSFT vendor ID */
1776 
1777 	return child_device_obj;
1778 }
1779 
1780 /*
1781  * vmbus_device_register - Register the child device
1782  */
1783 int vmbus_device_register(struct hv_device *child_device_obj)
1784 {
1785 	struct kobject *kobj = &child_device_obj->device.kobj;
1786 	int ret;
1787 
1788 	dev_set_name(&child_device_obj->device, "%pUl",
1789 		     child_device_obj->channel->offermsg.offer.if_instance.b);
1790 
1791 	child_device_obj->device.bus = &hv_bus;
1792 	child_device_obj->device.parent = &hv_acpi_dev->dev;
1793 	child_device_obj->device.release = vmbus_device_release;
1794 
1795 	/*
1796 	 * Register with the LDM. This will kick off the driver/device
1797 	 * binding...which will eventually call vmbus_match() and vmbus_probe()
1798 	 */
1799 	ret = device_register(&child_device_obj->device);
1800 	if (ret) {
1801 		pr_err("Unable to register child device\n");
1802 		return ret;
1803 	}
1804 
1805 	child_device_obj->channels_kset = kset_create_and_add("channels",
1806 							      NULL, kobj);
1807 	if (!child_device_obj->channels_kset) {
1808 		ret = -ENOMEM;
1809 		goto err_dev_unregister;
1810 	}
1811 
1812 	ret = vmbus_add_channel_kobj(child_device_obj,
1813 				     child_device_obj->channel);
1814 	if (ret) {
1815 		pr_err("Unable to register primary channeln");
1816 		goto err_kset_unregister;
1817 	}
1818 	hv_debug_add_dev_dir(child_device_obj);
1819 
1820 	return 0;
1821 
1822 err_kset_unregister:
1823 	kset_unregister(child_device_obj->channels_kset);
1824 
1825 err_dev_unregister:
1826 	device_unregister(&child_device_obj->device);
1827 	return ret;
1828 }
1829 
1830 /*
1831  * vmbus_device_unregister - Remove the specified child device
1832  * from the vmbus.
1833  */
1834 void vmbus_device_unregister(struct hv_device *device_obj)
1835 {
1836 	pr_debug("child device %s unregistered\n",
1837 		dev_name(&device_obj->device));
1838 
1839 	kset_unregister(device_obj->channels_kset);
1840 
1841 	/*
1842 	 * Kick off the process of unregistering the device.
1843 	 * This will call vmbus_remove() and eventually vmbus_device_release()
1844 	 */
1845 	device_unregister(&device_obj->device);
1846 }
1847 
1848 
1849 /*
1850  * VMBUS is an acpi enumerated device. Get the information we
1851  * need from DSDT.
1852  */
1853 #define VTPM_BASE_ADDRESS 0xfed40000
1854 static acpi_status vmbus_walk_resources(struct acpi_resource *res, void *ctx)
1855 {
1856 	resource_size_t start = 0;
1857 	resource_size_t end = 0;
1858 	struct resource *new_res;
1859 	struct resource **old_res = &hyperv_mmio;
1860 	struct resource **prev_res = NULL;
1861 
1862 	switch (res->type) {
1863 
1864 	/*
1865 	 * "Address" descriptors are for bus windows. Ignore
1866 	 * "memory" descriptors, which are for registers on
1867 	 * devices.
1868 	 */
1869 	case ACPI_RESOURCE_TYPE_ADDRESS32:
1870 		start = res->data.address32.address.minimum;
1871 		end = res->data.address32.address.maximum;
1872 		break;
1873 
1874 	case ACPI_RESOURCE_TYPE_ADDRESS64:
1875 		start = res->data.address64.address.minimum;
1876 		end = res->data.address64.address.maximum;
1877 		break;
1878 
1879 	default:
1880 		/* Unused resource type */
1881 		return AE_OK;
1882 
1883 	}
1884 	/*
1885 	 * Ignore ranges that are below 1MB, as they're not
1886 	 * necessary or useful here.
1887 	 */
1888 	if (end < 0x100000)
1889 		return AE_OK;
1890 
1891 	new_res = kzalloc(sizeof(*new_res), GFP_ATOMIC);
1892 	if (!new_res)
1893 		return AE_NO_MEMORY;
1894 
1895 	/* If this range overlaps the virtual TPM, truncate it. */
1896 	if (end > VTPM_BASE_ADDRESS && start < VTPM_BASE_ADDRESS)
1897 		end = VTPM_BASE_ADDRESS;
1898 
1899 	new_res->name = "hyperv mmio";
1900 	new_res->flags = IORESOURCE_MEM;
1901 	new_res->start = start;
1902 	new_res->end = end;
1903 
1904 	/*
1905 	 * If two ranges are adjacent, merge them.
1906 	 */
1907 	do {
1908 		if (!*old_res) {
1909 			*old_res = new_res;
1910 			break;
1911 		}
1912 
1913 		if (((*old_res)->end + 1) == new_res->start) {
1914 			(*old_res)->end = new_res->end;
1915 			kfree(new_res);
1916 			break;
1917 		}
1918 
1919 		if ((*old_res)->start == new_res->end + 1) {
1920 			(*old_res)->start = new_res->start;
1921 			kfree(new_res);
1922 			break;
1923 		}
1924 
1925 		if ((*old_res)->start > new_res->end) {
1926 			new_res->sibling = *old_res;
1927 			if (prev_res)
1928 				(*prev_res)->sibling = new_res;
1929 			*old_res = new_res;
1930 			break;
1931 		}
1932 
1933 		prev_res = old_res;
1934 		old_res = &(*old_res)->sibling;
1935 
1936 	} while (1);
1937 
1938 	return AE_OK;
1939 }
1940 
1941 static int vmbus_acpi_remove(struct acpi_device *device)
1942 {
1943 	struct resource *cur_res;
1944 	struct resource *next_res;
1945 
1946 	if (hyperv_mmio) {
1947 		if (fb_mmio) {
1948 			__release_region(hyperv_mmio, fb_mmio->start,
1949 					 resource_size(fb_mmio));
1950 			fb_mmio = NULL;
1951 		}
1952 
1953 		for (cur_res = hyperv_mmio; cur_res; cur_res = next_res) {
1954 			next_res = cur_res->sibling;
1955 			kfree(cur_res);
1956 		}
1957 	}
1958 
1959 	return 0;
1960 }
1961 
1962 static void vmbus_reserve_fb(void)
1963 {
1964 	int size;
1965 	/*
1966 	 * Make a claim for the frame buffer in the resource tree under the
1967 	 * first node, which will be the one below 4GB.  The length seems to
1968 	 * be underreported, particularly in a Generation 1 VM.  So start out
1969 	 * reserving a larger area and make it smaller until it succeeds.
1970 	 */
1971 
1972 	if (screen_info.lfb_base) {
1973 		if (efi_enabled(EFI_BOOT))
1974 			size = max_t(__u32, screen_info.lfb_size, 0x800000);
1975 		else
1976 			size = max_t(__u32, screen_info.lfb_size, 0x4000000);
1977 
1978 		for (; !fb_mmio && (size >= 0x100000); size >>= 1) {
1979 			fb_mmio = __request_region(hyperv_mmio,
1980 						   screen_info.lfb_base, size,
1981 						   fb_mmio_name, 0);
1982 		}
1983 	}
1984 }
1985 
1986 /**
1987  * vmbus_allocate_mmio() - Pick a memory-mapped I/O range.
1988  * @new:		If successful, supplied a pointer to the
1989  *			allocated MMIO space.
1990  * @device_obj:		Identifies the caller
1991  * @min:		Minimum guest physical address of the
1992  *			allocation
1993  * @max:		Maximum guest physical address
1994  * @size:		Size of the range to be allocated
1995  * @align:		Alignment of the range to be allocated
1996  * @fb_overlap_ok:	Whether this allocation can be allowed
1997  *			to overlap the video frame buffer.
1998  *
1999  * This function walks the resources granted to VMBus by the
2000  * _CRS object in the ACPI namespace underneath the parent
2001  * "bridge" whether that's a root PCI bus in the Generation 1
2002  * case or a Module Device in the Generation 2 case.  It then
2003  * attempts to allocate from the global MMIO pool in a way that
2004  * matches the constraints supplied in these parameters and by
2005  * that _CRS.
2006  *
2007  * Return: 0 on success, -errno on failure
2008  */
2009 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
2010 			resource_size_t min, resource_size_t max,
2011 			resource_size_t size, resource_size_t align,
2012 			bool fb_overlap_ok)
2013 {
2014 	struct resource *iter, *shadow;
2015 	resource_size_t range_min, range_max, start;
2016 	const char *dev_n = dev_name(&device_obj->device);
2017 	int retval;
2018 
2019 	retval = -ENXIO;
2020 	mutex_lock(&hyperv_mmio_lock);
2021 
2022 	/*
2023 	 * If overlaps with frame buffers are allowed, then first attempt to
2024 	 * make the allocation from within the reserved region.  Because it
2025 	 * is already reserved, no shadow allocation is necessary.
2026 	 */
2027 	if (fb_overlap_ok && fb_mmio && !(min > fb_mmio->end) &&
2028 	    !(max < fb_mmio->start)) {
2029 
2030 		range_min = fb_mmio->start;
2031 		range_max = fb_mmio->end;
2032 		start = (range_min + align - 1) & ~(align - 1);
2033 		for (; start + size - 1 <= range_max; start += align) {
2034 			*new = request_mem_region_exclusive(start, size, dev_n);
2035 			if (*new) {
2036 				retval = 0;
2037 				goto exit;
2038 			}
2039 		}
2040 	}
2041 
2042 	for (iter = hyperv_mmio; iter; iter = iter->sibling) {
2043 		if ((iter->start >= max) || (iter->end <= min))
2044 			continue;
2045 
2046 		range_min = iter->start;
2047 		range_max = iter->end;
2048 		start = (range_min + align - 1) & ~(align - 1);
2049 		for (; start + size - 1 <= range_max; start += align) {
2050 			shadow = __request_region(iter, start, size, NULL,
2051 						  IORESOURCE_BUSY);
2052 			if (!shadow)
2053 				continue;
2054 
2055 			*new = request_mem_region_exclusive(start, size, dev_n);
2056 			if (*new) {
2057 				shadow->name = (char *)*new;
2058 				retval = 0;
2059 				goto exit;
2060 			}
2061 
2062 			__release_region(iter, start, size);
2063 		}
2064 	}
2065 
2066 exit:
2067 	mutex_unlock(&hyperv_mmio_lock);
2068 	return retval;
2069 }
2070 EXPORT_SYMBOL_GPL(vmbus_allocate_mmio);
2071 
2072 /**
2073  * vmbus_free_mmio() - Free a memory-mapped I/O range.
2074  * @start:		Base address of region to release.
2075  * @size:		Size of the range to be allocated
2076  *
2077  * This function releases anything requested by
2078  * vmbus_mmio_allocate().
2079  */
2080 void vmbus_free_mmio(resource_size_t start, resource_size_t size)
2081 {
2082 	struct resource *iter;
2083 
2084 	mutex_lock(&hyperv_mmio_lock);
2085 	for (iter = hyperv_mmio; iter; iter = iter->sibling) {
2086 		if ((iter->start >= start + size) || (iter->end <= start))
2087 			continue;
2088 
2089 		__release_region(iter, start, size);
2090 	}
2091 	release_mem_region(start, size);
2092 	mutex_unlock(&hyperv_mmio_lock);
2093 
2094 }
2095 EXPORT_SYMBOL_GPL(vmbus_free_mmio);
2096 
2097 static int vmbus_acpi_add(struct acpi_device *device)
2098 {
2099 	acpi_status result;
2100 	int ret_val = -ENODEV;
2101 	struct acpi_device *ancestor;
2102 
2103 	hv_acpi_dev = device;
2104 
2105 	result = acpi_walk_resources(device->handle, METHOD_NAME__CRS,
2106 					vmbus_walk_resources, NULL);
2107 
2108 	if (ACPI_FAILURE(result))
2109 		goto acpi_walk_err;
2110 	/*
2111 	 * Some ancestor of the vmbus acpi device (Gen1 or Gen2
2112 	 * firmware) is the VMOD that has the mmio ranges. Get that.
2113 	 */
2114 	for (ancestor = device->parent; ancestor; ancestor = ancestor->parent) {
2115 		result = acpi_walk_resources(ancestor->handle, METHOD_NAME__CRS,
2116 					     vmbus_walk_resources, NULL);
2117 
2118 		if (ACPI_FAILURE(result))
2119 			continue;
2120 		if (hyperv_mmio) {
2121 			vmbus_reserve_fb();
2122 			break;
2123 		}
2124 	}
2125 	ret_val = 0;
2126 
2127 acpi_walk_err:
2128 	complete(&probe_event);
2129 	if (ret_val)
2130 		vmbus_acpi_remove(device);
2131 	return ret_val;
2132 }
2133 
2134 #ifdef CONFIG_PM_SLEEP
2135 static int vmbus_bus_suspend(struct device *dev)
2136 {
2137 	struct vmbus_channel *channel, *sc;
2138 	unsigned long flags;
2139 
2140 	while (atomic_read(&vmbus_connection.offer_in_progress) != 0) {
2141 		/*
2142 		 * We wait here until the completion of any channel
2143 		 * offers that are currently in progress.
2144 		 */
2145 		msleep(1);
2146 	}
2147 
2148 	mutex_lock(&vmbus_connection.channel_mutex);
2149 	list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
2150 		if (!is_hvsock_channel(channel))
2151 			continue;
2152 
2153 		vmbus_force_channel_rescinded(channel);
2154 	}
2155 	mutex_unlock(&vmbus_connection.channel_mutex);
2156 
2157 	/*
2158 	 * Wait until all the sub-channels and hv_sock channels have been
2159 	 * cleaned up. Sub-channels should be destroyed upon suspend, otherwise
2160 	 * they would conflict with the new sub-channels that will be created
2161 	 * in the resume path. hv_sock channels should also be destroyed, but
2162 	 * a hv_sock channel of an established hv_sock connection can not be
2163 	 * really destroyed since it may still be referenced by the userspace
2164 	 * application, so we just force the hv_sock channel to be rescinded
2165 	 * by vmbus_force_channel_rescinded(), and the userspace application
2166 	 * will thoroughly destroy the channel after hibernation.
2167 	 *
2168 	 * Note: the counter nr_chan_close_on_suspend may never go above 0 if
2169 	 * the VM has no sub-channel and hv_sock channel, e.g. a 1-vCPU VM.
2170 	 */
2171 	if (atomic_read(&vmbus_connection.nr_chan_close_on_suspend) > 0)
2172 		wait_for_completion(&vmbus_connection.ready_for_suspend_event);
2173 
2174 	WARN_ON(atomic_read(&vmbus_connection.nr_chan_fixup_on_resume) != 0);
2175 
2176 	mutex_lock(&vmbus_connection.channel_mutex);
2177 
2178 	list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
2179 		/*
2180 		 * Invalidate the field. Upon resume, vmbus_onoffer() will fix
2181 		 * up the field, and the other fields (if necessary).
2182 		 */
2183 		channel->offermsg.child_relid = INVALID_RELID;
2184 
2185 		if (is_hvsock_channel(channel)) {
2186 			if (!channel->rescind) {
2187 				pr_err("hv_sock channel not rescinded!\n");
2188 				WARN_ON_ONCE(1);
2189 			}
2190 			continue;
2191 		}
2192 
2193 		spin_lock_irqsave(&channel->lock, flags);
2194 		list_for_each_entry(sc, &channel->sc_list, sc_list) {
2195 			pr_err("Sub-channel not deleted!\n");
2196 			WARN_ON_ONCE(1);
2197 		}
2198 		spin_unlock_irqrestore(&channel->lock, flags);
2199 
2200 		atomic_inc(&vmbus_connection.nr_chan_fixup_on_resume);
2201 	}
2202 
2203 	mutex_unlock(&vmbus_connection.channel_mutex);
2204 
2205 	vmbus_initiate_unload(false);
2206 
2207 	vmbus_connection.conn_state = DISCONNECTED;
2208 
2209 	/* Reset the event for the next resume. */
2210 	reinit_completion(&vmbus_connection.ready_for_resume_event);
2211 
2212 	return 0;
2213 }
2214 
2215 static int vmbus_bus_resume(struct device *dev)
2216 {
2217 	struct vmbus_channel_msginfo *msginfo;
2218 	size_t msgsize;
2219 	int ret;
2220 
2221 	/*
2222 	 * We only use the 'vmbus_proto_version', which was in use before
2223 	 * hibernation, to re-negotiate with the host.
2224 	 */
2225 	if (!vmbus_proto_version) {
2226 		pr_err("Invalid proto version = 0x%x\n", vmbus_proto_version);
2227 		return -EINVAL;
2228 	}
2229 
2230 	msgsize = sizeof(*msginfo) +
2231 		  sizeof(struct vmbus_channel_initiate_contact);
2232 
2233 	msginfo = kzalloc(msgsize, GFP_KERNEL);
2234 
2235 	if (msginfo == NULL)
2236 		return -ENOMEM;
2237 
2238 	ret = vmbus_negotiate_version(msginfo, vmbus_proto_version);
2239 
2240 	kfree(msginfo);
2241 
2242 	if (ret != 0)
2243 		return ret;
2244 
2245 	WARN_ON(atomic_read(&vmbus_connection.nr_chan_fixup_on_resume) == 0);
2246 
2247 	vmbus_request_offers();
2248 
2249 	wait_for_completion(&vmbus_connection.ready_for_resume_event);
2250 
2251 	/* Reset the event for the next suspend. */
2252 	reinit_completion(&vmbus_connection.ready_for_suspend_event);
2253 
2254 	return 0;
2255 }
2256 #endif /* CONFIG_PM_SLEEP */
2257 
2258 static const struct acpi_device_id vmbus_acpi_device_ids[] = {
2259 	{"VMBUS", 0},
2260 	{"VMBus", 0},
2261 	{"", 0},
2262 };
2263 MODULE_DEVICE_TABLE(acpi, vmbus_acpi_device_ids);
2264 
2265 /*
2266  * Note: we must use SET_NOIRQ_SYSTEM_SLEEP_PM_OPS rather than
2267  * SET_SYSTEM_SLEEP_PM_OPS, otherwise NIC SR-IOV can not work, because the
2268  * "pci_dev_pm_ops" uses the "noirq" callbacks: in the resume path, the
2269  * pci "noirq" restore callback runs before "non-noirq" callbacks (see
2270  * resume_target_kernel() -> dpm_resume_start(), and hibernation_restore() ->
2271  * dpm_resume_end()). This means vmbus_bus_resume() and the pci-hyperv's
2272  * resume callback must also run via the "noirq" callbacks.
2273  */
2274 static const struct dev_pm_ops vmbus_bus_pm = {
2275 	SET_NOIRQ_SYSTEM_SLEEP_PM_OPS(vmbus_bus_suspend, vmbus_bus_resume)
2276 };
2277 
2278 static struct acpi_driver vmbus_acpi_driver = {
2279 	.name = "vmbus",
2280 	.ids = vmbus_acpi_device_ids,
2281 	.ops = {
2282 		.add = vmbus_acpi_add,
2283 		.remove = vmbus_acpi_remove,
2284 	},
2285 	.drv.pm = &vmbus_bus_pm,
2286 };
2287 
2288 static void hv_kexec_handler(void)
2289 {
2290 	hv_stimer_global_cleanup();
2291 	vmbus_initiate_unload(false);
2292 	vmbus_connection.conn_state = DISCONNECTED;
2293 	/* Make sure conn_state is set as hv_synic_cleanup checks for it */
2294 	mb();
2295 	cpuhp_remove_state(hyperv_cpuhp_online);
2296 	hyperv_cleanup();
2297 };
2298 
2299 static void hv_crash_handler(struct pt_regs *regs)
2300 {
2301 	int cpu;
2302 
2303 	vmbus_initiate_unload(true);
2304 	/*
2305 	 * In crash handler we can't schedule synic cleanup for all CPUs,
2306 	 * doing the cleanup for current CPU only. This should be sufficient
2307 	 * for kdump.
2308 	 */
2309 	vmbus_connection.conn_state = DISCONNECTED;
2310 	cpu = smp_processor_id();
2311 	hv_stimer_cleanup(cpu);
2312 	hv_synic_disable_regs(cpu);
2313 	hyperv_cleanup();
2314 };
2315 
2316 static int hv_synic_suspend(void)
2317 {
2318 	/*
2319 	 * When we reach here, all the non-boot CPUs have been offlined.
2320 	 * If we're in a legacy configuration where stimer Direct Mode is
2321 	 * not enabled, the stimers on the non-boot CPUs have been unbound
2322 	 * in hv_synic_cleanup() -> hv_stimer_legacy_cleanup() ->
2323 	 * hv_stimer_cleanup() -> clockevents_unbind_device().
2324 	 *
2325 	 * hv_synic_suspend() only runs on CPU0 with interrupts disabled.
2326 	 * Here we do not call hv_stimer_legacy_cleanup() on CPU0 because:
2327 	 * 1) it's unnecessary as interrupts remain disabled between
2328 	 * syscore_suspend() and syscore_resume(): see create_image() and
2329 	 * resume_target_kernel()
2330 	 * 2) the stimer on CPU0 is automatically disabled later by
2331 	 * syscore_suspend() -> timekeeping_suspend() -> tick_suspend() -> ...
2332 	 * -> clockevents_shutdown() -> ... -> hv_ce_shutdown()
2333 	 * 3) a warning would be triggered if we call
2334 	 * clockevents_unbind_device(), which may sleep, in an
2335 	 * interrupts-disabled context.
2336 	 */
2337 
2338 	hv_synic_disable_regs(0);
2339 
2340 	return 0;
2341 }
2342 
2343 static void hv_synic_resume(void)
2344 {
2345 	hv_synic_enable_regs(0);
2346 
2347 	/*
2348 	 * Note: we don't need to call hv_stimer_init(0), because the timer
2349 	 * on CPU0 is not unbound in hv_synic_suspend(), and the timer is
2350 	 * automatically re-enabled in timekeeping_resume().
2351 	 */
2352 }
2353 
2354 /* The callbacks run only on CPU0, with irqs_disabled. */
2355 static struct syscore_ops hv_synic_syscore_ops = {
2356 	.suspend = hv_synic_suspend,
2357 	.resume = hv_synic_resume,
2358 };
2359 
2360 static int __init hv_acpi_init(void)
2361 {
2362 	int ret, t;
2363 
2364 	if (!hv_is_hyperv_initialized())
2365 		return -ENODEV;
2366 
2367 	init_completion(&probe_event);
2368 
2369 	/*
2370 	 * Get ACPI resources first.
2371 	 */
2372 	ret = acpi_bus_register_driver(&vmbus_acpi_driver);
2373 
2374 	if (ret)
2375 		return ret;
2376 
2377 	t = wait_for_completion_timeout(&probe_event, 5*HZ);
2378 	if (t == 0) {
2379 		ret = -ETIMEDOUT;
2380 		goto cleanup;
2381 	}
2382 	hv_debug_init();
2383 
2384 	ret = vmbus_bus_init();
2385 	if (ret)
2386 		goto cleanup;
2387 
2388 	hv_setup_kexec_handler(hv_kexec_handler);
2389 	hv_setup_crash_handler(hv_crash_handler);
2390 
2391 	register_syscore_ops(&hv_synic_syscore_ops);
2392 
2393 	return 0;
2394 
2395 cleanup:
2396 	acpi_bus_unregister_driver(&vmbus_acpi_driver);
2397 	hv_acpi_dev = NULL;
2398 	return ret;
2399 }
2400 
2401 static void __exit vmbus_exit(void)
2402 {
2403 	int cpu;
2404 
2405 	unregister_syscore_ops(&hv_synic_syscore_ops);
2406 
2407 	hv_remove_kexec_handler();
2408 	hv_remove_crash_handler();
2409 	vmbus_connection.conn_state = DISCONNECTED;
2410 	hv_stimer_global_cleanup();
2411 	vmbus_disconnect();
2412 	hv_remove_vmbus_irq();
2413 	for_each_online_cpu(cpu) {
2414 		struct hv_per_cpu_context *hv_cpu
2415 			= per_cpu_ptr(hv_context.cpu_context, cpu);
2416 
2417 		tasklet_kill(&hv_cpu->msg_dpc);
2418 	}
2419 	hv_debug_rm_all_dir();
2420 
2421 	vmbus_free_channels();
2422 
2423 	if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
2424 		kmsg_dump_unregister(&hv_kmsg_dumper);
2425 		unregister_die_notifier(&hyperv_die_block);
2426 		atomic_notifier_chain_unregister(&panic_notifier_list,
2427 						 &hyperv_panic_block);
2428 	}
2429 
2430 	free_page((unsigned long)hv_panic_page);
2431 	unregister_sysctl_table(hv_ctl_table_hdr);
2432 	hv_ctl_table_hdr = NULL;
2433 	bus_unregister(&hv_bus);
2434 
2435 	cpuhp_remove_state(hyperv_cpuhp_online);
2436 	hv_synic_free();
2437 	acpi_bus_unregister_driver(&vmbus_acpi_driver);
2438 }
2439 
2440 
2441 MODULE_LICENSE("GPL");
2442 MODULE_DESCRIPTION("Microsoft Hyper-V VMBus Driver");
2443 
2444 subsys_initcall(hv_acpi_init);
2445 module_exit(vmbus_exit);
2446