xref: /linux/drivers/hv/vmbus_drv.c (revision cdae65fc43f28b6508d85fa242264f3bc5c9a5c7)
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/platform_device.h>
16 #include <linux/interrupt.h>
17 #include <linux/sysctl.h>
18 #include <linux/slab.h>
19 #include <linux/acpi.h>
20 #include <linux/completion.h>
21 #include <linux/hyperv.h>
22 #include <linux/kernel_stat.h>
23 #include <linux/of_address.h>
24 #include <linux/clockchips.h>
25 #include <linux/cpu.h>
26 #include <linux/sched/isolation.h>
27 #include <linux/sched/task_stack.h>
28 #include <linux/smpboot.h>
29 
30 #include <linux/delay.h>
31 #include <linux/panic_notifier.h>
32 #include <linux/ptrace.h>
33 #include <linux/sysfb.h>
34 #include <linux/efi.h>
35 #include <linux/kernel.h>
36 #include <linux/syscore_ops.h>
37 #include <linux/dma-map-ops.h>
38 #include <linux/pci.h>
39 #include <linux/export.h>
40 #include <clocksource/hyperv_timer.h>
41 #include <asm/mshyperv.h>
42 #include "hyperv_vmbus.h"
43 
44 struct vmbus_dynid {
45 	struct list_head node;
46 	struct hv_vmbus_device_id id;
47 };
48 
49 /* VMBus Root Device */
50 static struct device  *vmbus_root_device;
51 
52 static int hyperv_cpuhp_online;
53 
54 static DEFINE_PER_CPU(long, vmbus_evt);
55 
56 /* Values parsed from ACPI DSDT */
57 int vmbus_irq;
58 int vmbus_interrupt;
59 
60 /*
61  * If the Confidential VMBus is used, the data on the "wire" is not
62  * visible to either the host or the hypervisor.
63  */
64 static bool is_confidential;
65 
66 bool vmbus_is_confidential(void)
67 {
68 	return is_confidential;
69 }
70 EXPORT_SYMBOL_GPL(vmbus_is_confidential);
71 
72 static bool skip_vmbus_unload;
73 
74 /*
75  * Allow a VMBus framebuffer driver to specify that in the case of a panic,
76  * it will do the VMbus unload operation once it has flushed any dirty
77  * portions of the framebuffer to the Hyper-V host.
78  */
79 void vmbus_set_skip_unload(bool skip)
80 {
81 	skip_vmbus_unload = skip;
82 }
83 EXPORT_SYMBOL_GPL(vmbus_set_skip_unload);
84 
85 /*
86  * The panic notifier below is responsible solely for unloading the
87  * vmbus connection, which is necessary in a panic event.
88  */
89 static int hv_panic_vmbus_unload(struct notifier_block *nb, unsigned long val,
90 			      void *args)
91 {
92 	if (!skip_vmbus_unload)
93 		vmbus_initiate_unload(true);
94 
95 	return NOTIFY_DONE;
96 }
97 static struct notifier_block hyperv_panic_vmbus_unload_block = {
98 	.notifier_call	= hv_panic_vmbus_unload,
99 	.priority	= INT_MIN + 1, /* almost the latest one to execute */
100 };
101 
102 static const char *fb_mmio_name = "fb_range";
103 static struct resource *fb_mmio;
104 static struct resource *hyperv_mmio;
105 static DEFINE_MUTEX(hyperv_mmio_lock);
106 
107 struct device *hv_get_vmbus_root_device(void)
108 {
109 	return vmbus_root_device;
110 }
111 EXPORT_SYMBOL_GPL(hv_get_vmbus_root_device);
112 
113 bool hv_vmbus_exists(void)
114 {
115 	return vmbus_root_device != NULL;
116 }
117 EXPORT_SYMBOL_GPL(hv_vmbus_exists);
118 
119 static u8 channel_monitor_group(const struct vmbus_channel *channel)
120 {
121 	return (u8)channel->offermsg.monitorid / 32;
122 }
123 
124 static u8 channel_monitor_offset(const struct vmbus_channel *channel)
125 {
126 	return (u8)channel->offermsg.monitorid % 32;
127 }
128 
129 static u32 channel_pending(const struct vmbus_channel *channel,
130 			   const struct hv_monitor_page *monitor_page)
131 {
132 	u8 monitor_group = channel_monitor_group(channel);
133 
134 	return monitor_page->trigger_group[monitor_group].pending;
135 }
136 
137 static u32 channel_latency(const struct vmbus_channel *channel,
138 			   const struct hv_monitor_page *monitor_page)
139 {
140 	u8 monitor_group = channel_monitor_group(channel);
141 	u8 monitor_offset = channel_monitor_offset(channel);
142 
143 	return monitor_page->latency[monitor_group][monitor_offset];
144 }
145 
146 static u32 channel_conn_id(struct vmbus_channel *channel,
147 			   struct hv_monitor_page *monitor_page)
148 {
149 	u8 monitor_group = channel_monitor_group(channel);
150 	u8 monitor_offset = channel_monitor_offset(channel);
151 
152 	return monitor_page->parameter[monitor_group][monitor_offset].connectionid.u.id;
153 }
154 
155 static ssize_t id_show(struct device *dev, struct device_attribute *dev_attr,
156 		       char *buf)
157 {
158 	struct hv_device *hv_dev = device_to_hv_device(dev);
159 
160 	if (!hv_dev->channel)
161 		return -ENODEV;
162 	return sysfs_emit(buf, "%d\n", hv_dev->channel->offermsg.child_relid);
163 }
164 static DEVICE_ATTR_RO(id);
165 
166 static ssize_t state_show(struct device *dev, struct device_attribute *dev_attr,
167 			  char *buf)
168 {
169 	struct hv_device *hv_dev = device_to_hv_device(dev);
170 
171 	if (!hv_dev->channel)
172 		return -ENODEV;
173 	return sysfs_emit(buf, "%d\n", hv_dev->channel->state);
174 }
175 static DEVICE_ATTR_RO(state);
176 
177 static ssize_t monitor_id_show(struct device *dev,
178 			       struct device_attribute *dev_attr, char *buf)
179 {
180 	struct hv_device *hv_dev = device_to_hv_device(dev);
181 
182 	if (!hv_dev->channel)
183 		return -ENODEV;
184 	return sysfs_emit(buf, "%d\n", hv_dev->channel->offermsg.monitorid);
185 }
186 static DEVICE_ATTR_RO(monitor_id);
187 
188 static ssize_t class_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 sysfs_emit(buf, "{%pUl}\n",
196 			  &hv_dev->channel->offermsg.offer.if_type);
197 }
198 static DEVICE_ATTR_RO(class_id);
199 
200 static ssize_t device_id_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 
205 	if (!hv_dev->channel)
206 		return -ENODEV;
207 	return sysfs_emit(buf, "{%pUl}\n",
208 			  &hv_dev->channel->offermsg.offer.if_instance);
209 }
210 static DEVICE_ATTR_RO(device_id);
211 
212 static ssize_t modalias_show(struct device *dev,
213 			     struct device_attribute *dev_attr, char *buf)
214 {
215 	struct hv_device *hv_dev = device_to_hv_device(dev);
216 
217 	return sysfs_emit(buf, "vmbus:%*phN\n", UUID_SIZE, &hv_dev->dev_type);
218 }
219 static DEVICE_ATTR_RO(modalias);
220 
221 #ifdef CONFIG_NUMA
222 static ssize_t numa_node_show(struct device *dev,
223 			      struct device_attribute *attr, char *buf)
224 {
225 	struct hv_device *hv_dev = device_to_hv_device(dev);
226 
227 	if (!hv_dev->channel)
228 		return -ENODEV;
229 
230 	return sysfs_emit(buf, "%d\n", cpu_to_node(hv_dev->channel->target_cpu));
231 }
232 static DEVICE_ATTR_RO(numa_node);
233 #endif
234 
235 static ssize_t server_monitor_pending_show(struct device *dev,
236 					   struct device_attribute *dev_attr,
237 					   char *buf)
238 {
239 	struct hv_device *hv_dev = device_to_hv_device(dev);
240 
241 	if (!hv_dev->channel)
242 		return -ENODEV;
243 	return sysfs_emit(buf, "%d\n", channel_pending(hv_dev->channel,
244 			  vmbus_connection.monitor_pages[0]));
245 }
246 static DEVICE_ATTR_RO(server_monitor_pending);
247 
248 static ssize_t client_monitor_pending_show(struct device *dev,
249 					   struct device_attribute *dev_attr,
250 					   char *buf)
251 {
252 	struct hv_device *hv_dev = device_to_hv_device(dev);
253 
254 	if (!hv_dev->channel)
255 		return -ENODEV;
256 	return sysfs_emit(buf, "%d\n", channel_pending(hv_dev->channel,
257 			  vmbus_connection.monitor_pages[1]));
258 }
259 static DEVICE_ATTR_RO(client_monitor_pending);
260 
261 static ssize_t server_monitor_latency_show(struct device *dev,
262 					   struct device_attribute *dev_attr,
263 					   char *buf)
264 {
265 	struct hv_device *hv_dev = device_to_hv_device(dev);
266 
267 	if (!hv_dev->channel)
268 		return -ENODEV;
269 	return sysfs_emit(buf, "%d\n", channel_latency(hv_dev->channel,
270 			  vmbus_connection.monitor_pages[0]));
271 }
272 static DEVICE_ATTR_RO(server_monitor_latency);
273 
274 static ssize_t client_monitor_latency_show(struct device *dev,
275 					   struct device_attribute *dev_attr,
276 					   char *buf)
277 {
278 	struct hv_device *hv_dev = device_to_hv_device(dev);
279 
280 	if (!hv_dev->channel)
281 		return -ENODEV;
282 	return sysfs_emit(buf, "%d\n", channel_latency(hv_dev->channel,
283 			  vmbus_connection.monitor_pages[1]));
284 }
285 static DEVICE_ATTR_RO(client_monitor_latency);
286 
287 static ssize_t server_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 sysfs_emit(buf, "%d\n", channel_conn_id(hv_dev->channel,
296 			  vmbus_connection.monitor_pages[0]));
297 }
298 static DEVICE_ATTR_RO(server_monitor_conn_id);
299 
300 static ssize_t client_monitor_conn_id_show(struct device *dev,
301 					   struct device_attribute *dev_attr,
302 					   char *buf)
303 {
304 	struct hv_device *hv_dev = device_to_hv_device(dev);
305 
306 	if (!hv_dev->channel)
307 		return -ENODEV;
308 	return sysfs_emit(buf, "%d\n", channel_conn_id(hv_dev->channel,
309 			  vmbus_connection.monitor_pages[1]));
310 }
311 static DEVICE_ATTR_RO(client_monitor_conn_id);
312 
313 static ssize_t out_intr_mask_show(struct device *dev,
314 				  struct device_attribute *dev_attr, char *buf)
315 {
316 	struct hv_device *hv_dev = device_to_hv_device(dev);
317 	struct hv_ring_buffer_debug_info outbound;
318 	int ret;
319 
320 	if (!hv_dev->channel)
321 		return -ENODEV;
322 
323 	ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
324 					  &outbound);
325 	if (ret < 0)
326 		return ret;
327 
328 	return sysfs_emit(buf, "%d\n", outbound.current_interrupt_mask);
329 }
330 static DEVICE_ATTR_RO(out_intr_mask);
331 
332 static ssize_t out_read_index_show(struct device *dev,
333 				   struct device_attribute *dev_attr, char *buf)
334 {
335 	struct hv_device *hv_dev = device_to_hv_device(dev);
336 	struct hv_ring_buffer_debug_info outbound;
337 	int ret;
338 
339 	if (!hv_dev->channel)
340 		return -ENODEV;
341 
342 	ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
343 					  &outbound);
344 	if (ret < 0)
345 		return ret;
346 	return sysfs_emit(buf, "%u\n", outbound.current_read_index);
347 }
348 static DEVICE_ATTR_RO(out_read_index);
349 
350 static ssize_t out_write_index_show(struct device *dev,
351 				    struct device_attribute *dev_attr,
352 				    char *buf)
353 {
354 	struct hv_device *hv_dev = device_to_hv_device(dev);
355 	struct hv_ring_buffer_debug_info outbound;
356 	int ret;
357 
358 	if (!hv_dev->channel)
359 		return -ENODEV;
360 
361 	ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
362 					  &outbound);
363 	if (ret < 0)
364 		return ret;
365 	return sysfs_emit(buf, "%u\n", outbound.current_write_index);
366 }
367 static DEVICE_ATTR_RO(out_write_index);
368 
369 static ssize_t out_read_bytes_avail_show(struct device *dev,
370 					 struct device_attribute *dev_attr,
371 					 char *buf)
372 {
373 	struct hv_device *hv_dev = device_to_hv_device(dev);
374 	struct hv_ring_buffer_debug_info outbound;
375 	int ret;
376 
377 	if (!hv_dev->channel)
378 		return -ENODEV;
379 
380 	ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
381 					  &outbound);
382 	if (ret < 0)
383 		return ret;
384 	return sysfs_emit(buf, "%d\n", outbound.bytes_avail_toread);
385 }
386 static DEVICE_ATTR_RO(out_read_bytes_avail);
387 
388 static ssize_t out_write_bytes_avail_show(struct device *dev,
389 					  struct device_attribute *dev_attr,
390 					  char *buf)
391 {
392 	struct hv_device *hv_dev = device_to_hv_device(dev);
393 	struct hv_ring_buffer_debug_info outbound;
394 	int ret;
395 
396 	if (!hv_dev->channel)
397 		return -ENODEV;
398 
399 	ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
400 					  &outbound);
401 	if (ret < 0)
402 		return ret;
403 	return sysfs_emit(buf, "%d\n", outbound.bytes_avail_towrite);
404 }
405 static DEVICE_ATTR_RO(out_write_bytes_avail);
406 
407 static ssize_t in_intr_mask_show(struct device *dev,
408 				 struct device_attribute *dev_attr, char *buf)
409 {
410 	struct hv_device *hv_dev = device_to_hv_device(dev);
411 	struct hv_ring_buffer_debug_info inbound;
412 	int ret;
413 
414 	if (!hv_dev->channel)
415 		return -ENODEV;
416 
417 	ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
418 	if (ret < 0)
419 		return ret;
420 
421 	return sysfs_emit(buf, "%d\n", inbound.current_interrupt_mask);
422 }
423 static DEVICE_ATTR_RO(in_intr_mask);
424 
425 static ssize_t in_read_index_show(struct device *dev,
426 				  struct device_attribute *dev_attr, char *buf)
427 {
428 	struct hv_device *hv_dev = device_to_hv_device(dev);
429 	struct hv_ring_buffer_debug_info inbound;
430 	int ret;
431 
432 	if (!hv_dev->channel)
433 		return -ENODEV;
434 
435 	ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
436 	if (ret < 0)
437 		return ret;
438 
439 	return sysfs_emit(buf, "%d\n", inbound.current_read_index);
440 }
441 static DEVICE_ATTR_RO(in_read_index);
442 
443 static ssize_t in_write_index_show(struct device *dev,
444 				   struct device_attribute *dev_attr, char *buf)
445 {
446 	struct hv_device *hv_dev = device_to_hv_device(dev);
447 	struct hv_ring_buffer_debug_info inbound;
448 	int ret;
449 
450 	if (!hv_dev->channel)
451 		return -ENODEV;
452 
453 	ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
454 	if (ret < 0)
455 		return ret;
456 
457 	return sysfs_emit(buf, "%d\n", inbound.current_write_index);
458 }
459 static DEVICE_ATTR_RO(in_write_index);
460 
461 static ssize_t in_read_bytes_avail_show(struct device *dev,
462 					struct device_attribute *dev_attr,
463 					char *buf)
464 {
465 	struct hv_device *hv_dev = device_to_hv_device(dev);
466 	struct hv_ring_buffer_debug_info inbound;
467 	int ret;
468 
469 	if (!hv_dev->channel)
470 		return -ENODEV;
471 
472 	ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
473 	if (ret < 0)
474 		return ret;
475 
476 	return sysfs_emit(buf, "%d\n", inbound.bytes_avail_toread);
477 }
478 static DEVICE_ATTR_RO(in_read_bytes_avail);
479 
480 static ssize_t in_write_bytes_avail_show(struct device *dev,
481 					 struct device_attribute *dev_attr,
482 					 char *buf)
483 {
484 	struct hv_device *hv_dev = device_to_hv_device(dev);
485 	struct hv_ring_buffer_debug_info inbound;
486 	int ret;
487 
488 	if (!hv_dev->channel)
489 		return -ENODEV;
490 
491 	ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
492 	if (ret < 0)
493 		return ret;
494 
495 	return sysfs_emit(buf, "%d\n", inbound.bytes_avail_towrite);
496 }
497 static DEVICE_ATTR_RO(in_write_bytes_avail);
498 
499 static ssize_t channel_vp_mapping_show(struct device *dev,
500 				       struct device_attribute *dev_attr,
501 				       char *buf)
502 {
503 	struct hv_device *hv_dev = device_to_hv_device(dev);
504 	struct vmbus_channel *channel = hv_dev->channel, *cur_sc;
505 	int n_written;
506 	struct list_head *cur;
507 
508 	if (!channel)
509 		return -ENODEV;
510 
511 	mutex_lock(&vmbus_connection.channel_mutex);
512 
513 	n_written = sysfs_emit(buf, "%u:%u\n",
514 			       channel->offermsg.child_relid,
515 			       channel->target_cpu);
516 
517 	list_for_each(cur, &channel->sc_list) {
518 
519 		cur_sc = list_entry(cur, struct vmbus_channel, sc_list);
520 		n_written += sysfs_emit_at(buf, n_written, "%u:%u\n",
521 					  cur_sc->offermsg.child_relid,
522 					  cur_sc->target_cpu);
523 	}
524 
525 	mutex_unlock(&vmbus_connection.channel_mutex);
526 
527 	return n_written;
528 }
529 static DEVICE_ATTR_RO(channel_vp_mapping);
530 
531 static ssize_t vendor_show(struct device *dev,
532 			   struct device_attribute *dev_attr,
533 			   char *buf)
534 {
535 	struct hv_device *hv_dev = device_to_hv_device(dev);
536 
537 	return sysfs_emit(buf, "0x%x\n", hv_dev->vendor_id);
538 }
539 static DEVICE_ATTR_RO(vendor);
540 
541 static ssize_t device_show(struct device *dev,
542 			   struct device_attribute *dev_attr,
543 			   char *buf)
544 {
545 	struct hv_device *hv_dev = device_to_hv_device(dev);
546 
547 	return sysfs_emit(buf, "0x%x\n", hv_dev->device_id);
548 }
549 static DEVICE_ATTR_RO(device);
550 
551 static ssize_t driver_override_store(struct device *dev,
552 				     struct device_attribute *attr,
553 				     const char *buf, size_t count)
554 {
555 	struct hv_device *hv_dev = device_to_hv_device(dev);
556 	int ret;
557 
558 	ret = driver_set_override(dev, &hv_dev->driver_override, buf, count);
559 	if (ret)
560 		return ret;
561 
562 	return count;
563 }
564 
565 static ssize_t driver_override_show(struct device *dev,
566 				    struct device_attribute *attr, char *buf)
567 {
568 	struct hv_device *hv_dev = device_to_hv_device(dev);
569 	ssize_t len;
570 
571 	device_lock(dev);
572 	len = sysfs_emit(buf, "%s\n", hv_dev->driver_override);
573 	device_unlock(dev);
574 
575 	return len;
576 }
577 static DEVICE_ATTR_RW(driver_override);
578 
579 /* Set up per device attributes in /sys/bus/vmbus/devices/<bus device> */
580 static struct attribute *vmbus_dev_attrs[] = {
581 	&dev_attr_id.attr,
582 	&dev_attr_state.attr,
583 	&dev_attr_monitor_id.attr,
584 	&dev_attr_class_id.attr,
585 	&dev_attr_device_id.attr,
586 	&dev_attr_modalias.attr,
587 #ifdef CONFIG_NUMA
588 	&dev_attr_numa_node.attr,
589 #endif
590 	&dev_attr_server_monitor_pending.attr,
591 	&dev_attr_client_monitor_pending.attr,
592 	&dev_attr_server_monitor_latency.attr,
593 	&dev_attr_client_monitor_latency.attr,
594 	&dev_attr_server_monitor_conn_id.attr,
595 	&dev_attr_client_monitor_conn_id.attr,
596 	&dev_attr_out_intr_mask.attr,
597 	&dev_attr_out_read_index.attr,
598 	&dev_attr_out_write_index.attr,
599 	&dev_attr_out_read_bytes_avail.attr,
600 	&dev_attr_out_write_bytes_avail.attr,
601 	&dev_attr_in_intr_mask.attr,
602 	&dev_attr_in_read_index.attr,
603 	&dev_attr_in_write_index.attr,
604 	&dev_attr_in_read_bytes_avail.attr,
605 	&dev_attr_in_write_bytes_avail.attr,
606 	&dev_attr_channel_vp_mapping.attr,
607 	&dev_attr_vendor.attr,
608 	&dev_attr_device.attr,
609 	&dev_attr_driver_override.attr,
610 	NULL,
611 };
612 
613 /*
614  * Device-level attribute_group callback function. Returns the permission for
615  * each attribute, and returns 0 if an attribute is not visible.
616  */
617 static umode_t vmbus_dev_attr_is_visible(struct kobject *kobj,
618 					 struct attribute *attr, int idx)
619 {
620 	struct device *dev = kobj_to_dev(kobj);
621 	const struct hv_device *hv_dev = device_to_hv_device(dev);
622 
623 	/* Hide the monitor attributes if the monitor mechanism is not used. */
624 	if (!hv_dev->channel->offermsg.monitor_allocated &&
625 	    (attr == &dev_attr_monitor_id.attr ||
626 	     attr == &dev_attr_server_monitor_pending.attr ||
627 	     attr == &dev_attr_client_monitor_pending.attr ||
628 	     attr == &dev_attr_server_monitor_latency.attr ||
629 	     attr == &dev_attr_client_monitor_latency.attr ||
630 	     attr == &dev_attr_server_monitor_conn_id.attr ||
631 	     attr == &dev_attr_client_monitor_conn_id.attr))
632 		return 0;
633 
634 	return attr->mode;
635 }
636 
637 static const struct attribute_group vmbus_dev_group = {
638 	.attrs = vmbus_dev_attrs,
639 	.is_visible = vmbus_dev_attr_is_visible
640 };
641 __ATTRIBUTE_GROUPS(vmbus_dev);
642 
643 /* Set up the attribute for /sys/bus/vmbus/hibernation */
644 static ssize_t hibernation_show(const struct bus_type *bus, char *buf)
645 {
646 	return sprintf(buf, "%d\n", !!hv_is_hibernation_supported());
647 }
648 
649 static BUS_ATTR_RO(hibernation);
650 
651 static struct attribute *vmbus_bus_attrs[] = {
652 	&bus_attr_hibernation.attr,
653 	NULL,
654 };
655 static const struct attribute_group vmbus_bus_group = {
656 	.attrs = vmbus_bus_attrs,
657 };
658 __ATTRIBUTE_GROUPS(vmbus_bus);
659 
660 /*
661  * vmbus_uevent - add uevent for our device
662  *
663  * This routine is invoked when a device is added or removed on the vmbus to
664  * generate a uevent to udev in the userspace. The udev will then look at its
665  * rule and the uevent generated here to load the appropriate driver
666  *
667  * The alias string will be of the form vmbus:guid where guid is the string
668  * representation of the device guid (each byte of the guid will be
669  * represented with two hex characters.
670  */
671 static int vmbus_uevent(const struct device *device, struct kobj_uevent_env *env)
672 {
673 	const struct hv_device *dev = device_to_hv_device(device);
674 	const char *format = "MODALIAS=vmbus:%*phN";
675 
676 	return add_uevent_var(env, format, UUID_SIZE, &dev->dev_type);
677 }
678 
679 static const struct hv_vmbus_device_id *
680 hv_vmbus_dev_match(const struct hv_vmbus_device_id *id, const guid_t *guid)
681 {
682 	if (id == NULL)
683 		return NULL; /* empty device table */
684 
685 	for (; !guid_is_null(&id->guid); id++)
686 		if (guid_equal(&id->guid, guid))
687 			return id;
688 
689 	return NULL;
690 }
691 
692 static const struct hv_vmbus_device_id *
693 hv_vmbus_dynid_match(struct hv_driver *drv, const guid_t *guid)
694 {
695 	const struct hv_vmbus_device_id *id = NULL;
696 	struct vmbus_dynid *dynid;
697 
698 	spin_lock(&drv->dynids.lock);
699 	list_for_each_entry(dynid, &drv->dynids.list, node) {
700 		if (guid_equal(&dynid->id.guid, guid)) {
701 			id = &dynid->id;
702 			break;
703 		}
704 	}
705 	spin_unlock(&drv->dynids.lock);
706 
707 	return id;
708 }
709 
710 static const struct hv_vmbus_device_id vmbus_device_null;
711 
712 /*
713  * Return a matching hv_vmbus_device_id pointer.
714  * If there is no match, return NULL.
715  */
716 static const struct hv_vmbus_device_id *hv_vmbus_get_id(const struct hv_driver *drv,
717 							struct hv_device *dev)
718 {
719 	const guid_t *guid = &dev->dev_type;
720 	const struct hv_vmbus_device_id *id;
721 
722 	/* When driver_override is set, only bind to the matching driver */
723 	if (dev->driver_override && strcmp(dev->driver_override, drv->name))
724 		return NULL;
725 
726 	/* Look at the dynamic ids first, before the static ones */
727 	id = hv_vmbus_dynid_match((struct hv_driver *)drv, guid);
728 	if (!id)
729 		id = hv_vmbus_dev_match(drv->id_table, guid);
730 
731 	/* driver_override will always match, send a dummy id */
732 	if (!id && dev->driver_override)
733 		id = &vmbus_device_null;
734 
735 	return id;
736 }
737 
738 /* vmbus_add_dynid - add a new device ID to this driver and re-probe devices
739  *
740  * This function can race with vmbus_device_register(). This function is
741  * typically running on a user thread in response to writing to the "new_id"
742  * sysfs entry for a driver. vmbus_device_register() is running on a
743  * workqueue thread in response to the Hyper-V host offering a device to the
744  * guest. This function calls driver_attach(), which looks for an existing
745  * device matching the new id, and attaches the driver to which the new id
746  * has been assigned. vmbus_device_register() calls device_register(), which
747  * looks for a driver that matches the device being registered. If both
748  * operations are running simultaneously, the device driver probe function runs
749  * on whichever thread establishes the linkage between the driver and device.
750  *
751  * In most cases, it doesn't matter which thread runs the driver probe
752  * function. But if vmbus_device_register() does not find a matching driver,
753  * it proceeds to create the "channels" subdirectory and numbered per-channel
754  * subdirectory in sysfs. While that multi-step creation is in progress, this
755  * function could run the driver probe function. If the probe function checks
756  * for, or operates on, entries in the "channels" subdirectory, including by
757  * calling hv_create_ring_sysfs(), the operation may or may not succeed
758  * depending on the race. The race can't create a kernel failure in VMBus
759  * or device subsystem code, but probe functions in VMBus drivers doing such
760  * operations must be prepared for the failure case.
761  */
762 static int vmbus_add_dynid(struct hv_driver *drv, guid_t *guid)
763 {
764 	struct vmbus_dynid *dynid;
765 
766 	dynid = kzalloc_obj(*dynid);
767 	if (!dynid)
768 		return -ENOMEM;
769 
770 	dynid->id.guid = *guid;
771 
772 	spin_lock(&drv->dynids.lock);
773 	list_add_tail(&dynid->node, &drv->dynids.list);
774 	spin_unlock(&drv->dynids.lock);
775 
776 	return driver_attach(&drv->driver);
777 }
778 
779 static void vmbus_free_dynids(struct hv_driver *drv)
780 {
781 	struct vmbus_dynid *dynid, *n;
782 
783 	spin_lock(&drv->dynids.lock);
784 	list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
785 		list_del(&dynid->node);
786 		kfree(dynid);
787 	}
788 	spin_unlock(&drv->dynids.lock);
789 }
790 
791 /*
792  * store_new_id - sysfs frontend to vmbus_add_dynid()
793  *
794  * Allow GUIDs to be added to an existing driver via sysfs.
795  */
796 static ssize_t new_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 	guid_t guid;
801 	ssize_t retval;
802 
803 	retval = guid_parse(buf, &guid);
804 	if (retval)
805 		return retval;
806 
807 	if (hv_vmbus_dynid_match(drv, &guid))
808 		return -EEXIST;
809 
810 	retval = vmbus_add_dynid(drv, &guid);
811 	if (retval)
812 		return retval;
813 	return count;
814 }
815 static DRIVER_ATTR_WO(new_id);
816 
817 /*
818  * store_remove_id - remove a PCI device ID from this driver
819  *
820  * Removes a dynamic pci device ID to this driver.
821  */
822 static ssize_t remove_id_store(struct device_driver *driver, const char *buf,
823 			       size_t count)
824 {
825 	struct hv_driver *drv = drv_to_hv_drv(driver);
826 	struct vmbus_dynid *dynid, *n;
827 	guid_t guid;
828 	ssize_t retval;
829 
830 	retval = guid_parse(buf, &guid);
831 	if (retval)
832 		return retval;
833 
834 	retval = -ENODEV;
835 	spin_lock(&drv->dynids.lock);
836 	list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
837 		struct hv_vmbus_device_id *id = &dynid->id;
838 
839 		if (guid_equal(&id->guid, &guid)) {
840 			list_del(&dynid->node);
841 			kfree(dynid);
842 			retval = count;
843 			break;
844 		}
845 	}
846 	spin_unlock(&drv->dynids.lock);
847 
848 	return retval;
849 }
850 static DRIVER_ATTR_WO(remove_id);
851 
852 static struct attribute *vmbus_drv_attrs[] = {
853 	&driver_attr_new_id.attr,
854 	&driver_attr_remove_id.attr,
855 	NULL,
856 };
857 ATTRIBUTE_GROUPS(vmbus_drv);
858 
859 
860 /*
861  * vmbus_match - Attempt to match the specified device to the specified driver
862  */
863 static int vmbus_match(struct device *device, const struct device_driver *driver)
864 {
865 	const struct hv_driver *drv = drv_to_hv_drv(driver);
866 	struct hv_device *hv_dev = device_to_hv_device(device);
867 
868 	/* The hv_sock driver handles all hv_sock offers. */
869 	if (is_hvsock_channel(hv_dev->channel))
870 		return drv->hvsock;
871 
872 	if (hv_vmbus_get_id(drv, hv_dev))
873 		return 1;
874 
875 	return 0;
876 }
877 
878 /*
879  * vmbus_probe - Add the new vmbus's child device
880  */
881 static int vmbus_probe(struct device *child_device)
882 {
883 	int ret = 0;
884 	struct hv_driver *drv =
885 			drv_to_hv_drv(child_device->driver);
886 	struct hv_device *dev = device_to_hv_device(child_device);
887 	const struct hv_vmbus_device_id *dev_id;
888 
889 	dev_id = hv_vmbus_get_id(drv, dev);
890 	if (drv->probe) {
891 		ret = drv->probe(dev, dev_id);
892 		if (ret != 0)
893 			pr_err("probe failed for device %s (%d)\n",
894 			       dev_name(child_device), ret);
895 
896 	} else {
897 		pr_err("probe not set for driver %s\n",
898 		       dev_name(child_device));
899 		ret = -ENODEV;
900 	}
901 	return ret;
902 }
903 
904 /*
905  * vmbus_dma_configure -- Configure DMA coherence for VMbus device
906  */
907 static int vmbus_dma_configure(struct device *child_device)
908 {
909 	/*
910 	 * On ARM64, propagate the DMA coherence setting from the top level
911 	 * VMbus ACPI device to the child VMbus device being added here.
912 	 * On x86/x64 coherence is assumed and these calls have no effect.
913 	 */
914 	hv_setup_dma_ops(child_device,
915 		device_get_dma_attr(vmbus_root_device) == DEV_DMA_COHERENT);
916 	return 0;
917 }
918 
919 /*
920  * vmbus_remove - Remove a vmbus device
921  */
922 static void vmbus_remove(struct device *child_device)
923 {
924 	struct hv_driver *drv;
925 	struct hv_device *dev = device_to_hv_device(child_device);
926 
927 	if (child_device->driver) {
928 		drv = drv_to_hv_drv(child_device->driver);
929 		if (drv->remove)
930 			drv->remove(dev);
931 	}
932 }
933 
934 /*
935  * vmbus_shutdown - Shutdown a vmbus device
936  */
937 static void vmbus_shutdown(struct device *child_device)
938 {
939 	struct hv_driver *drv;
940 	struct hv_device *dev = device_to_hv_device(child_device);
941 
942 
943 	/* The device may not be attached yet */
944 	if (!child_device->driver)
945 		return;
946 
947 	drv = drv_to_hv_drv(child_device->driver);
948 
949 	if (drv->shutdown)
950 		drv->shutdown(dev);
951 }
952 
953 #ifdef CONFIG_PM_SLEEP
954 /*
955  * vmbus_suspend - Suspend a vmbus device
956  */
957 static int vmbus_suspend(struct device *child_device)
958 {
959 	struct hv_driver *drv;
960 	struct hv_device *dev = device_to_hv_device(child_device);
961 
962 	/* The device may not be attached yet */
963 	if (!child_device->driver)
964 		return 0;
965 
966 	drv = drv_to_hv_drv(child_device->driver);
967 	if (!drv->suspend)
968 		return -EOPNOTSUPP;
969 
970 	return drv->suspend(dev);
971 }
972 
973 /*
974  * vmbus_resume - Resume a vmbus device
975  */
976 static int vmbus_resume(struct device *child_device)
977 {
978 	struct hv_driver *drv;
979 	struct hv_device *dev = device_to_hv_device(child_device);
980 
981 	/* The device may not be attached yet */
982 	if (!child_device->driver)
983 		return 0;
984 
985 	drv = drv_to_hv_drv(child_device->driver);
986 	if (!drv->resume)
987 		return -EOPNOTSUPP;
988 
989 	return drv->resume(dev);
990 }
991 #else
992 #define vmbus_suspend NULL
993 #define vmbus_resume NULL
994 #endif /* CONFIG_PM_SLEEP */
995 
996 /*
997  * vmbus_device_release - Final callback release of the vmbus child device
998  */
999 static void vmbus_device_release(struct device *device)
1000 {
1001 	struct hv_device *hv_dev = device_to_hv_device(device);
1002 	struct vmbus_channel *channel = hv_dev->channel;
1003 
1004 	hv_debug_rm_dev_dir(hv_dev);
1005 
1006 	mutex_lock(&vmbus_connection.channel_mutex);
1007 	hv_process_channel_removal(channel);
1008 	mutex_unlock(&vmbus_connection.channel_mutex);
1009 	kfree(hv_dev);
1010 }
1011 
1012 /*
1013  * Note: we must use the "noirq" ops: see the comment before vmbus_bus_pm.
1014  *
1015  * suspend_noirq/resume_noirq are set to NULL to support Suspend-to-Idle: we
1016  * shouldn't suspend the vmbus devices upon Suspend-to-Idle, otherwise there
1017  * is no way to wake up a Generation-2 VM.
1018  *
1019  * The other 4 ops are for hibernation.
1020  */
1021 
1022 static const struct dev_pm_ops vmbus_pm = {
1023 	.suspend_noirq	= NULL,
1024 	.resume_noirq	= NULL,
1025 	.freeze_noirq	= vmbus_suspend,
1026 	.thaw_noirq	= vmbus_resume,
1027 	.poweroff_noirq	= vmbus_suspend,
1028 	.restore_noirq	= vmbus_resume,
1029 };
1030 
1031 /* The one and only one */
1032 static const struct bus_type  hv_bus = {
1033 	.name =		"vmbus",
1034 	.match =		vmbus_match,
1035 	.shutdown =		vmbus_shutdown,
1036 	.remove =		vmbus_remove,
1037 	.probe =		vmbus_probe,
1038 	.uevent =		vmbus_uevent,
1039 	.dma_configure =	vmbus_dma_configure,
1040 	.dev_groups =		vmbus_dev_groups,
1041 	.drv_groups =		vmbus_drv_groups,
1042 	.bus_groups =		vmbus_bus_groups,
1043 	.pm =			&vmbus_pm,
1044 };
1045 
1046 struct onmessage_work_context {
1047 	struct work_struct work;
1048 	struct {
1049 		struct hv_message_header header;
1050 		u8 payload[];
1051 	} msg;
1052 };
1053 
1054 static void vmbus_onmessage_work(struct work_struct *work)
1055 {
1056 	struct onmessage_work_context *ctx;
1057 
1058 	/* Do not process messages if we're in DISCONNECTED state */
1059 	if (vmbus_connection.conn_state == DISCONNECTED)
1060 		return;
1061 
1062 	ctx = container_of(work, struct onmessage_work_context,
1063 			   work);
1064 	vmbus_onmessage((struct vmbus_channel_message_header *)
1065 			&ctx->msg.payload);
1066 	kfree(ctx);
1067 }
1068 
1069 static void __vmbus_on_msg_dpc(void *message_page_addr)
1070 {
1071 	struct hv_message msg_copy, *msg;
1072 	struct vmbus_channel_message_header *hdr;
1073 	enum vmbus_channel_message_type msgtype;
1074 	const struct vmbus_channel_message_table_entry *entry;
1075 	struct onmessage_work_context *ctx;
1076 	__u8 payload_size;
1077 	u32 message_type;
1078 
1079 	if (!message_page_addr)
1080 		return;
1081 	msg = (struct hv_message *)message_page_addr + VMBUS_MESSAGE_SINT;
1082 
1083 	/*
1084 	 * 'enum vmbus_channel_message_type' is supposed to always be 'u32' as
1085 	 * it is being used in 'struct vmbus_channel_message_header' definition
1086 	 * which is supposed to match hypervisor ABI.
1087 	 */
1088 	BUILD_BUG_ON(sizeof(enum vmbus_channel_message_type) != sizeof(u32));
1089 
1090 	/*
1091 	 * Since the message is in memory shared with the host, an erroneous or
1092 	 * malicious Hyper-V could modify the message while vmbus_on_msg_dpc()
1093 	 * or individual message handlers are executing; to prevent this, copy
1094 	 * the message into private memory.
1095 	 */
1096 	memcpy(&msg_copy, msg, sizeof(struct hv_message));
1097 
1098 	message_type = msg_copy.header.message_type;
1099 	if (message_type == HVMSG_NONE)
1100 		/* no msg */
1101 		return;
1102 
1103 	hdr = (struct vmbus_channel_message_header *)msg_copy.u.payload;
1104 	msgtype = hdr->msgtype;
1105 
1106 	trace_vmbus_on_msg_dpc(hdr);
1107 
1108 	if (msgtype >= CHANNELMSG_COUNT) {
1109 		WARN_ONCE(1, "unknown msgtype=%d\n", msgtype);
1110 		goto msg_handled;
1111 	}
1112 
1113 	payload_size = msg_copy.header.payload_size;
1114 	if (payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT) {
1115 		WARN_ONCE(1, "payload size is too large (%d)\n", payload_size);
1116 		goto msg_handled;
1117 	}
1118 
1119 	entry = &channel_message_table[msgtype];
1120 
1121 	if (!entry->message_handler)
1122 		goto msg_handled;
1123 
1124 	if (payload_size < entry->min_payload_len) {
1125 		WARN_ONCE(1, "message too short: msgtype=%d len=%d\n", msgtype, payload_size);
1126 		goto msg_handled;
1127 	}
1128 
1129 	if (entry->handler_type	== VMHT_BLOCKING) {
1130 		ctx = kmalloc_flex(*ctx, msg.payload, payload_size, GFP_ATOMIC);
1131 		if (ctx == NULL)
1132 			return;
1133 
1134 		INIT_WORK(&ctx->work, vmbus_onmessage_work);
1135 		ctx->msg.header = msg_copy.header;
1136 		memcpy(&ctx->msg.payload, msg_copy.u.payload, payload_size);
1137 
1138 		/*
1139 		 * The host can generate a rescind message while we
1140 		 * may still be handling the original offer. We deal with
1141 		 * this condition by relying on the synchronization provided
1142 		 * by offer_in_progress and by channel_mutex.  See also the
1143 		 * inline comments in vmbus_onoffer_rescind().
1144 		 */
1145 		switch (msgtype) {
1146 		case CHANNELMSG_RESCIND_CHANNELOFFER:
1147 			/*
1148 			 * If we are handling the rescind message;
1149 			 * schedule the work on the global work queue.
1150 			 *
1151 			 * The OFFER message and the RESCIND message should
1152 			 * not be handled by the same serialized work queue,
1153 			 * because the OFFER handler may call vmbus_open(),
1154 			 * which tries to open the channel by sending an
1155 			 * OPEN_CHANNEL message to the host and waits for
1156 			 * the host's response; however, if the host has
1157 			 * rescinded the channel before it receives the
1158 			 * OPEN_CHANNEL message, the host just silently
1159 			 * ignores the OPEN_CHANNEL message; as a result,
1160 			 * the guest's OFFER handler hangs for ever, if we
1161 			 * handle the RESCIND message in the same serialized
1162 			 * work queue: the RESCIND handler can not start to
1163 			 * run before the OFFER handler finishes.
1164 			 */
1165 			if (vmbus_connection.ignore_any_offer_msg)
1166 				break;
1167 			queue_work(vmbus_connection.rescind_work_queue, &ctx->work);
1168 			break;
1169 
1170 		case CHANNELMSG_OFFERCHANNEL:
1171 			/*
1172 			 * The host sends the offer message of a given channel
1173 			 * before sending the rescind message of the same
1174 			 * channel.  These messages are sent to the guest's
1175 			 * connect CPU; the guest then starts processing them
1176 			 * in the tasklet handler on this CPU:
1177 			 *
1178 			 * VMBUS_CONNECT_CPU
1179 			 *
1180 			 * [vmbus_on_msg_dpc()]
1181 			 * atomic_inc()  // CHANNELMSG_OFFERCHANNEL
1182 			 * queue_work()
1183 			 * ...
1184 			 * [vmbus_on_msg_dpc()]
1185 			 * schedule_work()  // CHANNELMSG_RESCIND_CHANNELOFFER
1186 			 *
1187 			 * We rely on the memory-ordering properties of the
1188 			 * queue_work() and schedule_work() primitives, which
1189 			 * guarantee that the atomic increment will be visible
1190 			 * to the CPUs which will execute the offer & rescind
1191 			 * works by the time these works will start execution.
1192 			 */
1193 			if (vmbus_connection.ignore_any_offer_msg)
1194 				break;
1195 			atomic_inc(&vmbus_connection.offer_in_progress);
1196 			fallthrough;
1197 
1198 		default:
1199 			queue_work(vmbus_connection.work_queue, &ctx->work);
1200 		}
1201 	} else
1202 		entry->message_handler(hdr);
1203 
1204 msg_handled:
1205 	vmbus_signal_eom(msg, message_type);
1206 }
1207 
1208 void vmbus_on_msg_dpc(unsigned long data)
1209 {
1210 	struct hv_per_cpu_context *hv_cpu = (void *)data;
1211 
1212 	__vmbus_on_msg_dpc(hv_cpu->hyp_synic_message_page);
1213 	__vmbus_on_msg_dpc(hv_cpu->para_synic_message_page);
1214 }
1215 
1216 #ifdef CONFIG_PM_SLEEP
1217 /*
1218  * Fake RESCIND_CHANNEL messages to clean up hv_sock channels by force for
1219  * hibernation, because hv_sock connections can not persist across hibernation.
1220  */
1221 static void vmbus_force_channel_rescinded(struct vmbus_channel *channel)
1222 {
1223 	struct onmessage_work_context *ctx;
1224 	struct vmbus_channel_rescind_offer *rescind;
1225 
1226 	WARN_ON(!is_hvsock_channel(channel));
1227 
1228 	/*
1229 	 * Allocation size is small and the allocation should really not fail,
1230 	 * otherwise the state of the hv_sock connections ends up in limbo.
1231 	 */
1232 	ctx = kzalloc(sizeof(*ctx) + sizeof(*rescind),
1233 		      GFP_KERNEL | __GFP_NOFAIL);
1234 
1235 	/*
1236 	 * So far, these are not really used by Linux. Just set them to the
1237 	 * reasonable values conforming to the definitions of the fields.
1238 	 */
1239 	ctx->msg.header.message_type = 1;
1240 	ctx->msg.header.payload_size = sizeof(*rescind);
1241 
1242 	/* These values are actually used by Linux. */
1243 	rescind = (struct vmbus_channel_rescind_offer *)ctx->msg.payload;
1244 	rescind->header.msgtype = CHANNELMSG_RESCIND_CHANNELOFFER;
1245 	rescind->child_relid = channel->offermsg.child_relid;
1246 
1247 	INIT_WORK(&ctx->work, vmbus_onmessage_work);
1248 
1249 	queue_work(vmbus_connection.work_queue, &ctx->work);
1250 }
1251 #endif /* CONFIG_PM_SLEEP */
1252 
1253 /*
1254  * Schedule all channels with events pending.
1255  * The event page can be directly checked to get the id of
1256  * the channel that has the interrupt pending.
1257  */
1258 static void vmbus_chan_sched(void *event_page_addr)
1259 {
1260 	unsigned long *recv_int_page;
1261 	u32 maxbits, relid;
1262 	union hv_synic_event_flags *event;
1263 
1264 	if (!event_page_addr)
1265 		return;
1266 	event = (union hv_synic_event_flags *)event_page_addr + VMBUS_MESSAGE_SINT;
1267 
1268 	maxbits = READ_ONCE(vmbus_connection.relid_hiwater) + 1;
1269 	recv_int_page = event->flags;
1270 
1271 	if (unlikely(!recv_int_page))
1272 		return;
1273 
1274 	for_each_set_bit(relid, recv_int_page, maxbits) {
1275 		void (*callback_fn)(void *context);
1276 		struct vmbus_channel *channel;
1277 
1278 		if (!sync_test_and_clear_bit(relid, recv_int_page))
1279 			continue;
1280 
1281 		/* Special case - vmbus channel protocol msg */
1282 		if (relid == 0)
1283 			continue;
1284 
1285 		/*
1286 		 * Pairs with the kfree_rcu() in vmbus_chan_release().
1287 		 * Guarantees that the channel data structure doesn't
1288 		 * get freed while the channel pointer below is being
1289 		 * dereferenced.
1290 		 */
1291 		rcu_read_lock();
1292 
1293 		/* Find channel based on relid */
1294 		channel = relid2channel(relid);
1295 		if (channel == NULL)
1296 			goto sched_unlock_rcu;
1297 
1298 		if (channel->rescind)
1299 			goto sched_unlock_rcu;
1300 
1301 		/*
1302 		 * Make sure that the ring buffer data structure doesn't get
1303 		 * freed while we dereference the ring buffer pointer.  Test
1304 		 * for the channel's onchannel_callback being NULL within a
1305 		 * sched_lock critical section.  See also the inline comments
1306 		 * in vmbus_reset_channel_cb().
1307 		 */
1308 		spin_lock(&channel->sched_lock);
1309 
1310 		callback_fn = channel->onchannel_callback;
1311 		if (unlikely(callback_fn == NULL))
1312 			goto sched_unlock;
1313 
1314 		trace_vmbus_chan_sched(channel);
1315 
1316 		++channel->interrupts;
1317 
1318 		switch (channel->callback_mode) {
1319 		case HV_CALL_ISR:
1320 			(*callback_fn)(channel->channel_callback_context);
1321 			break;
1322 
1323 		case HV_CALL_BATCHED:
1324 			hv_begin_read(&channel->inbound);
1325 			fallthrough;
1326 		case HV_CALL_DIRECT:
1327 			tasklet_schedule(&channel->callback_event);
1328 		}
1329 
1330 sched_unlock:
1331 		spin_unlock(&channel->sched_lock);
1332 sched_unlock_rcu:
1333 		rcu_read_unlock();
1334 	}
1335 }
1336 
1337 static void vmbus_message_sched(struct hv_per_cpu_context *hv_cpu, void *message_page_addr)
1338 {
1339 	struct hv_message *msg;
1340 
1341 	if (!message_page_addr)
1342 		return;
1343 	msg = (struct hv_message *)message_page_addr + VMBUS_MESSAGE_SINT;
1344 
1345 	/* Check if there are actual msgs to be processed */
1346 	if (msg->header.message_type != HVMSG_NONE) {
1347 		if (msg->header.message_type == HVMSG_TIMER_EXPIRED) {
1348 			hv_stimer0_isr();
1349 			vmbus_signal_eom(msg, HVMSG_TIMER_EXPIRED);
1350 		} else {
1351 			tasklet_schedule(&hv_cpu->msg_dpc);
1352 		}
1353 	}
1354 }
1355 
1356 static void __vmbus_isr(void)
1357 {
1358 	struct hv_per_cpu_context *hv_cpu
1359 		= this_cpu_ptr(hv_context.cpu_context);
1360 
1361 	vmbus_chan_sched(hv_cpu->hyp_synic_event_page);
1362 	vmbus_chan_sched(hv_cpu->para_synic_event_page);
1363 
1364 	vmbus_message_sched(hv_cpu, hv_cpu->hyp_synic_message_page);
1365 	vmbus_message_sched(hv_cpu, hv_cpu->para_synic_message_page);
1366 }
1367 
1368 static DEFINE_PER_CPU(bool, vmbus_irq_pending);
1369 static DEFINE_PER_CPU(struct task_struct *, vmbus_irqd);
1370 
1371 static void vmbus_irqd_wake(void)
1372 {
1373 	struct task_struct *tsk = __this_cpu_read(vmbus_irqd);
1374 
1375 	__this_cpu_write(vmbus_irq_pending, true);
1376 	wake_up_process(tsk);
1377 }
1378 
1379 static void vmbus_irqd_setup(unsigned int cpu)
1380 {
1381 	sched_set_fifo(current);
1382 }
1383 
1384 static int vmbus_irqd_should_run(unsigned int cpu)
1385 {
1386 	return __this_cpu_read(vmbus_irq_pending);
1387 }
1388 
1389 static void run_vmbus_irqd(unsigned int cpu)
1390 {
1391 	__this_cpu_write(vmbus_irq_pending, false);
1392 	__vmbus_isr();
1393 }
1394 
1395 static bool vmbus_irq_initialized;
1396 
1397 static struct smp_hotplug_thread vmbus_irq_threads = {
1398 	.store                  = &vmbus_irqd,
1399 	.setup			= vmbus_irqd_setup,
1400 	.thread_should_run      = vmbus_irqd_should_run,
1401 	.thread_fn              = run_vmbus_irqd,
1402 	.thread_comm            = "vmbus_irq/%u",
1403 };
1404 
1405 void vmbus_isr(void)
1406 {
1407 	if (IS_ENABLED(CONFIG_PREEMPT_RT)) {
1408 		vmbus_irqd_wake();
1409 	} else {
1410 		lockdep_hardirq_threaded();
1411 		__vmbus_isr();
1412 	}
1413 }
1414 EXPORT_SYMBOL_FOR_MODULES(vmbus_isr, "mshv_vtl");
1415 
1416 static irqreturn_t vmbus_percpu_isr(int irq, void *dev_id)
1417 {
1418 	vmbus_isr();
1419 	return IRQ_HANDLED;
1420 }
1421 
1422 static void vmbus_percpu_work(struct work_struct *work)
1423 {
1424 	unsigned int cpu = smp_processor_id();
1425 
1426 	hv_synic_init(cpu);
1427 }
1428 
1429 static int vmbus_alloc_synic_and_connect(void)
1430 {
1431 	int ret, cpu;
1432 	struct work_struct __percpu *works;
1433 
1434 	ret = hv_synic_alloc();
1435 	if (ret < 0)
1436 		goto err_alloc;
1437 
1438 	works = alloc_percpu(struct work_struct);
1439 	if (!works) {
1440 		ret = -ENOMEM;
1441 		goto err_alloc;
1442 	}
1443 
1444 	/*
1445 	 * Initialize the per-cpu interrupt state and stimer state.
1446 	 * Then connect to the host.
1447 	 */
1448 	cpus_read_lock();
1449 	for_each_online_cpu(cpu) {
1450 		struct work_struct *work = per_cpu_ptr(works, cpu);
1451 
1452 		INIT_WORK(work, vmbus_percpu_work);
1453 		schedule_work_on(cpu, work);
1454 	}
1455 
1456 	for_each_online_cpu(cpu)
1457 		flush_work(per_cpu_ptr(works, cpu));
1458 
1459 	/* Register the callbacks for possible CPU online/offline'ing */
1460 	ret = cpuhp_setup_state_nocalls_cpuslocked(CPUHP_AP_ONLINE_DYN, "hyperv/vmbus:online",
1461 						   hv_synic_init, hv_synic_cleanup);
1462 	cpus_read_unlock();
1463 	free_percpu(works);
1464 	if (ret < 0)
1465 		goto err_alloc;
1466 	hyperv_cpuhp_online = ret;
1467 
1468 	ret = vmbus_connect();
1469 	if (ret)
1470 		goto err_connect;
1471 	return 0;
1472 
1473 err_connect:
1474 	cpuhp_remove_state(hyperv_cpuhp_online);
1475 	return -ENODEV;
1476 err_alloc:
1477 	hv_synic_free();
1478 	return -ENOMEM;
1479 }
1480 
1481 /*
1482  * vmbus_bus_init -Main vmbus driver initialization routine.
1483  *
1484  * Here, we
1485  *	- initialize the vmbus driver context
1486  *	- invoke the vmbus hv main init routine
1487  *	- retrieve the channel offers
1488  */
1489 static int vmbus_bus_init(void)
1490 {
1491 	int ret;
1492 
1493 	ret = hv_init();
1494 	if (ret != 0) {
1495 		pr_err("Unable to initialize the hypervisor - 0x%x\n", ret);
1496 		return ret;
1497 	}
1498 
1499 	ret = bus_register(&hv_bus);
1500 	if (ret)
1501 		return ret;
1502 
1503 	/*
1504 	 * VMbus interrupts are best modeled as per-cpu interrupts. If
1505 	 * on an architecture with support for per-cpu IRQs (e.g. ARM64),
1506 	 * allocate a per-cpu IRQ using standard Linux kernel functionality.
1507 	 * If not on such an architecture (e.g., x86/x64), then rely on
1508 	 * code in the arch-specific portion of the code tree to connect
1509 	 * the VMbus interrupt handler.
1510 	 */
1511 
1512 	if (IS_ENABLED(CONFIG_PREEMPT_RT) && !vmbus_irq_initialized) {
1513 		ret = smpboot_register_percpu_thread(&vmbus_irq_threads);
1514 		if (ret)
1515 			goto err_kthread;
1516 		vmbus_irq_initialized = true;
1517 	}
1518 
1519 	if (vmbus_irq == -1) {
1520 		hv_setup_vmbus_handler(vmbus_isr);
1521 	} else {
1522 		ret = request_percpu_irq(vmbus_irq, vmbus_percpu_isr,
1523 				"Hyper-V VMbus", &vmbus_evt);
1524 		if (ret) {
1525 			pr_err("Can't request Hyper-V VMbus IRQ %d, Err %d",
1526 					vmbus_irq, ret);
1527 			goto err_setup;
1528 		}
1529 	}
1530 
1531 	/*
1532 	 * Cache the value as getting it involves a VM exit on x86(_64), and
1533 	 * doing that on each VP while initializing SynIC's wastes time.
1534 	 */
1535 	is_confidential = ms_hyperv.confidential_vmbus_available;
1536 	if (is_confidential)
1537 		pr_info("Establishing connection to the confidential VMBus\n");
1538 	hv_para_set_sint_proxy(!is_confidential);
1539 	ret = vmbus_alloc_synic_and_connect();
1540 	if (ret)
1541 		goto err_connect;
1542 
1543 	/*
1544 	 * Always register the vmbus unload panic notifier because we
1545 	 * need to shut the VMbus channel connection on panic.
1546 	 */
1547 	atomic_notifier_chain_register(&panic_notifier_list,
1548 			       &hyperv_panic_vmbus_unload_block);
1549 
1550 	vmbus_request_offers();
1551 
1552 	return 0;
1553 
1554 err_connect:
1555 	if (vmbus_irq == -1)
1556 		hv_remove_vmbus_handler();
1557 	else
1558 		free_percpu_irq(vmbus_irq, &vmbus_evt);
1559 err_setup:
1560 	if (IS_ENABLED(CONFIG_PREEMPT_RT) && vmbus_irq_initialized) {
1561 		smpboot_unregister_percpu_thread(&vmbus_irq_threads);
1562 		vmbus_irq_initialized = false;
1563 	}
1564 err_kthread:
1565 	bus_unregister(&hv_bus);
1566 	return ret;
1567 }
1568 
1569 /**
1570  * __vmbus_driver_register() - Register a vmbus's driver
1571  * @hv_driver: Pointer to driver structure you want to register
1572  * @owner: owner module of the drv
1573  * @mod_name: module name string
1574  *
1575  * Registers the given driver with Linux through the 'driver_register()' call
1576  * and sets up the hyper-v vmbus handling for this driver.
1577  * It will return the state of the 'driver_register()' call.
1578  *
1579  */
1580 int __vmbus_driver_register(struct hv_driver *hv_driver, struct module *owner, const char *mod_name)
1581 {
1582 	int ret;
1583 
1584 	if (!hv_vmbus_exists())
1585 		return -ENODEV;
1586 
1587 	pr_info("registering driver %s\n", hv_driver->name);
1588 
1589 	hv_driver->driver.name = hv_driver->name;
1590 	hv_driver->driver.owner = owner;
1591 	hv_driver->driver.mod_name = mod_name;
1592 	hv_driver->driver.bus = &hv_bus;
1593 
1594 	spin_lock_init(&hv_driver->dynids.lock);
1595 	INIT_LIST_HEAD(&hv_driver->dynids.list);
1596 
1597 	ret = driver_register(&hv_driver->driver);
1598 
1599 	return ret;
1600 }
1601 EXPORT_SYMBOL_GPL(__vmbus_driver_register);
1602 
1603 /**
1604  * vmbus_driver_unregister() - Unregister a vmbus's driver
1605  * @hv_driver: Pointer to driver structure you want to
1606  *             un-register
1607  *
1608  * Un-register the given driver that was previous registered with a call to
1609  * vmbus_driver_register()
1610  */
1611 void vmbus_driver_unregister(struct hv_driver *hv_driver)
1612 {
1613 	if (hv_vmbus_exists()) {
1614 		pr_info("unregistering driver %s\n", hv_driver->name);
1615 		driver_unregister(&hv_driver->driver);
1616 		vmbus_free_dynids(hv_driver);
1617 	}
1618 }
1619 EXPORT_SYMBOL_GPL(vmbus_driver_unregister);
1620 
1621 
1622 /*
1623  * Called when last reference to channel is gone.
1624  */
1625 static void vmbus_chan_release(struct kobject *kobj)
1626 {
1627 	struct vmbus_channel *channel
1628 		= container_of(kobj, struct vmbus_channel, kobj);
1629 
1630 	kfree_rcu(channel, rcu);
1631 }
1632 
1633 struct vmbus_chan_attribute {
1634 	struct attribute attr;
1635 	ssize_t (*show)(struct vmbus_channel *chan, char *buf);
1636 	ssize_t (*store)(struct vmbus_channel *chan,
1637 			 const char *buf, size_t count);
1638 };
1639 #define VMBUS_CHAN_ATTR(_name, _mode, _show, _store) \
1640 	struct vmbus_chan_attribute chan_attr_##_name \
1641 		= __ATTR(_name, _mode, _show, _store)
1642 #define VMBUS_CHAN_ATTR_RW(_name) \
1643 	struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RW(_name)
1644 #define VMBUS_CHAN_ATTR_RO(_name) \
1645 	struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RO(_name)
1646 #define VMBUS_CHAN_ATTR_WO(_name) \
1647 	struct vmbus_chan_attribute chan_attr_##_name = __ATTR_WO(_name)
1648 
1649 static ssize_t vmbus_chan_attr_show(struct kobject *kobj,
1650 				    struct attribute *attr, char *buf)
1651 {
1652 	const struct vmbus_chan_attribute *attribute
1653 		= container_of(attr, struct vmbus_chan_attribute, attr);
1654 	struct vmbus_channel *chan
1655 		= container_of(kobj, struct vmbus_channel, kobj);
1656 
1657 	if (!attribute->show)
1658 		return -EIO;
1659 
1660 	return attribute->show(chan, buf);
1661 }
1662 
1663 static ssize_t vmbus_chan_attr_store(struct kobject *kobj,
1664 				     struct attribute *attr, const char *buf,
1665 				     size_t count)
1666 {
1667 	const struct vmbus_chan_attribute *attribute
1668 		= container_of(attr, struct vmbus_chan_attribute, attr);
1669 	struct vmbus_channel *chan
1670 		= container_of(kobj, struct vmbus_channel, kobj);
1671 
1672 	if (!attribute->store)
1673 		return -EIO;
1674 
1675 	return attribute->store(chan, buf, count);
1676 }
1677 
1678 static const struct sysfs_ops vmbus_chan_sysfs_ops = {
1679 	.show = vmbus_chan_attr_show,
1680 	.store = vmbus_chan_attr_store,
1681 };
1682 
1683 static ssize_t out_mask_show(struct vmbus_channel *channel, char *buf)
1684 {
1685 	struct hv_ring_buffer_info *rbi = &channel->outbound;
1686 	ssize_t ret;
1687 
1688 	mutex_lock(&rbi->ring_buffer_mutex);
1689 	if (!rbi->ring_buffer) {
1690 		mutex_unlock(&rbi->ring_buffer_mutex);
1691 		return -EINVAL;
1692 	}
1693 
1694 	ret = sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1695 	mutex_unlock(&rbi->ring_buffer_mutex);
1696 	return ret;
1697 }
1698 static VMBUS_CHAN_ATTR_RO(out_mask);
1699 
1700 static ssize_t in_mask_show(struct vmbus_channel *channel, char *buf)
1701 {
1702 	struct hv_ring_buffer_info *rbi = &channel->inbound;
1703 	ssize_t ret;
1704 
1705 	mutex_lock(&rbi->ring_buffer_mutex);
1706 	if (!rbi->ring_buffer) {
1707 		mutex_unlock(&rbi->ring_buffer_mutex);
1708 		return -EINVAL;
1709 	}
1710 
1711 	ret = sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1712 	mutex_unlock(&rbi->ring_buffer_mutex);
1713 	return ret;
1714 }
1715 static VMBUS_CHAN_ATTR_RO(in_mask);
1716 
1717 static ssize_t read_avail_show(struct vmbus_channel *channel, char *buf)
1718 {
1719 	struct hv_ring_buffer_info *rbi = &channel->inbound;
1720 	ssize_t ret;
1721 
1722 	mutex_lock(&rbi->ring_buffer_mutex);
1723 	if (!rbi->ring_buffer) {
1724 		mutex_unlock(&rbi->ring_buffer_mutex);
1725 		return -EINVAL;
1726 	}
1727 
1728 	ret = sprintf(buf, "%u\n", hv_get_bytes_to_read(rbi));
1729 	mutex_unlock(&rbi->ring_buffer_mutex);
1730 	return ret;
1731 }
1732 static VMBUS_CHAN_ATTR_RO(read_avail);
1733 
1734 static ssize_t write_avail_show(struct vmbus_channel *channel, char *buf)
1735 {
1736 	struct hv_ring_buffer_info *rbi = &channel->outbound;
1737 	ssize_t ret;
1738 
1739 	mutex_lock(&rbi->ring_buffer_mutex);
1740 	if (!rbi->ring_buffer) {
1741 		mutex_unlock(&rbi->ring_buffer_mutex);
1742 		return -EINVAL;
1743 	}
1744 
1745 	ret = sprintf(buf, "%u\n", hv_get_bytes_to_write(rbi));
1746 	mutex_unlock(&rbi->ring_buffer_mutex);
1747 	return ret;
1748 }
1749 static VMBUS_CHAN_ATTR_RO(write_avail);
1750 
1751 static ssize_t target_cpu_show(struct vmbus_channel *channel, char *buf)
1752 {
1753 	return sprintf(buf, "%u\n", channel->target_cpu);
1754 }
1755 
1756 int vmbus_channel_set_cpu(struct vmbus_channel *channel, u32 target_cpu)
1757 {
1758 	u32 origin_cpu;
1759 	int ret = 0;
1760 
1761 	lockdep_assert_cpus_held();
1762 	lockdep_assert_held(&vmbus_connection.channel_mutex);
1763 
1764 	if (vmbus_proto_version < VERSION_WIN10_V4_1)
1765 		return -EIO;
1766 
1767 	/* Validate target_cpu for the cpumask_test_cpu() operation below. */
1768 	if (target_cpu >= nr_cpumask_bits)
1769 		return -EINVAL;
1770 
1771 	if (!cpumask_test_cpu(target_cpu, housekeeping_cpumask(HK_TYPE_MANAGED_IRQ)))
1772 		return -EINVAL;
1773 
1774 	if (!cpu_online(target_cpu))
1775 		return -EINVAL;
1776 
1777 	/*
1778 	 * Synchronizes vmbus_channel_set_cpu() and channel closure:
1779 	 *
1780 	 * { Initially: state = CHANNEL_OPENED }
1781 	 *
1782 	 * CPU1				CPU2
1783 	 *
1784 	 * [vmbus_channel_set_cpu()]	[vmbus_disconnect_ring()]
1785 	 *
1786 	 * LOCK channel_mutex		LOCK channel_mutex
1787 	 * LOAD r1 = state		LOAD r2 = state
1788 	 * IF (r1 == CHANNEL_OPENED)	IF (r2 == CHANNEL_OPENED)
1789 	 *   SEND MODIFYCHANNEL		  STORE state = CHANNEL_OPEN
1790 	 *   [...]			  SEND CLOSECHANNEL
1791 	 * UNLOCK channel_mutex		UNLOCK channel_mutex
1792 	 *
1793 	 * Forbids: r1 == r2 == CHANNEL_OPENED (i.e., CPU1's LOCK precedes
1794 	 * 		CPU2's LOCK) && CPU2's SEND precedes CPU1's SEND
1795 	 *
1796 	 * Note.  The host processes the channel messages "sequentially", in
1797 	 * the order in which they are received on a per-partition basis.
1798 	 */
1799 
1800 	/*
1801 	 * Hyper-V will ignore MODIFYCHANNEL messages for "non-open" channels;
1802 	 * avoid sending the message and fail here for such channels.
1803 	 */
1804 	if (channel->state != CHANNEL_OPENED_STATE) {
1805 		ret = -EIO;
1806 		goto end;
1807 	}
1808 
1809 	origin_cpu = channel->target_cpu;
1810 	if (target_cpu == origin_cpu)
1811 		goto end;
1812 
1813 	if (vmbus_send_modifychannel(channel,
1814 				     hv_cpu_number_to_vp_number(target_cpu))) {
1815 		ret = -EIO;
1816 		goto end;
1817 	}
1818 
1819 	/*
1820 	 * For version before VERSION_WIN10_V5_3, the following warning holds:
1821 	 *
1822 	 * Warning.  At this point, there is *no* guarantee that the host will
1823 	 * have successfully processed the vmbus_send_modifychannel() request.
1824 	 * See the header comment of vmbus_send_modifychannel() for more info.
1825 	 *
1826 	 * Lags in the processing of the above vmbus_send_modifychannel() can
1827 	 * result in missed interrupts if the "old" target CPU is taken offline
1828 	 * before Hyper-V starts sending interrupts to the "new" target CPU.
1829 	 * But apart from this offlining scenario, the code tolerates such
1830 	 * lags.  It will function correctly even if a channel interrupt comes
1831 	 * in on a CPU that is different from the channel target_cpu value.
1832 	 */
1833 
1834 	channel->target_cpu = target_cpu;
1835 
1836 	/* See init_vp_index(). */
1837 	if (hv_is_perf_channel(channel))
1838 		hv_update_allocated_cpus(origin_cpu, target_cpu);
1839 
1840 	/* Currently set only for storvsc channels. */
1841 	if (channel->change_target_cpu_callback) {
1842 		(*channel->change_target_cpu_callback)(channel,
1843 				origin_cpu, target_cpu);
1844 	}
1845 
1846 end:
1847 	return ret;
1848 }
1849 
1850 static ssize_t target_cpu_store(struct vmbus_channel *channel,
1851 				const char *buf, size_t count)
1852 {
1853 	u32 target_cpu;
1854 	ssize_t ret;
1855 
1856 	if (sscanf(buf, "%u", &target_cpu) != 1)
1857 		return -EIO;
1858 
1859 	cpus_read_lock();
1860 	mutex_lock(&vmbus_connection.channel_mutex);
1861 	ret = vmbus_channel_set_cpu(channel, target_cpu);
1862 	mutex_unlock(&vmbus_connection.channel_mutex);
1863 	cpus_read_unlock();
1864 
1865 	return ret ?: count;
1866 }
1867 static VMBUS_CHAN_ATTR(cpu, 0644, target_cpu_show, target_cpu_store);
1868 
1869 static ssize_t channel_pending_show(struct vmbus_channel *channel,
1870 				    char *buf)
1871 {
1872 	return sprintf(buf, "%d\n",
1873 		       channel_pending(channel,
1874 				       vmbus_connection.monitor_pages[1]));
1875 }
1876 static VMBUS_CHAN_ATTR(pending, 0444, channel_pending_show, NULL);
1877 
1878 static ssize_t channel_latency_show(struct vmbus_channel *channel,
1879 				    char *buf)
1880 {
1881 	return sprintf(buf, "%d\n",
1882 		       channel_latency(channel,
1883 				       vmbus_connection.monitor_pages[1]));
1884 }
1885 static VMBUS_CHAN_ATTR(latency, 0444, channel_latency_show, NULL);
1886 
1887 static ssize_t channel_interrupts_show(struct vmbus_channel *channel, char *buf)
1888 {
1889 	return sprintf(buf, "%llu\n", channel->interrupts);
1890 }
1891 static VMBUS_CHAN_ATTR(interrupts, 0444, channel_interrupts_show, NULL);
1892 
1893 static ssize_t channel_events_show(struct vmbus_channel *channel, char *buf)
1894 {
1895 	return sprintf(buf, "%llu\n", channel->sig_events);
1896 }
1897 static VMBUS_CHAN_ATTR(events, 0444, channel_events_show, NULL);
1898 
1899 static ssize_t channel_intr_in_full_show(struct vmbus_channel *channel,
1900 					 char *buf)
1901 {
1902 	return sprintf(buf, "%llu\n",
1903 		       (unsigned long long)channel->intr_in_full);
1904 }
1905 static VMBUS_CHAN_ATTR(intr_in_full, 0444, channel_intr_in_full_show, NULL);
1906 
1907 static ssize_t channel_intr_out_empty_show(struct vmbus_channel *channel,
1908 					   char *buf)
1909 {
1910 	return sprintf(buf, "%llu\n",
1911 		       (unsigned long long)channel->intr_out_empty);
1912 }
1913 static VMBUS_CHAN_ATTR(intr_out_empty, 0444, channel_intr_out_empty_show, NULL);
1914 
1915 static ssize_t channel_out_full_first_show(struct vmbus_channel *channel,
1916 					   char *buf)
1917 {
1918 	return sprintf(buf, "%llu\n",
1919 		       (unsigned long long)channel->out_full_first);
1920 }
1921 static VMBUS_CHAN_ATTR(out_full_first, 0444, channel_out_full_first_show, NULL);
1922 
1923 static ssize_t channel_out_full_total_show(struct vmbus_channel *channel,
1924 					   char *buf)
1925 {
1926 	return sprintf(buf, "%llu\n",
1927 		       (unsigned long long)channel->out_full_total);
1928 }
1929 static VMBUS_CHAN_ATTR(out_full_total, 0444, channel_out_full_total_show, NULL);
1930 
1931 static ssize_t subchannel_monitor_id_show(struct vmbus_channel *channel,
1932 					  char *buf)
1933 {
1934 	return sprintf(buf, "%u\n", channel->offermsg.monitorid);
1935 }
1936 static VMBUS_CHAN_ATTR(monitor_id, 0444, subchannel_monitor_id_show, NULL);
1937 
1938 static ssize_t subchannel_id_show(struct vmbus_channel *channel,
1939 				  char *buf)
1940 {
1941 	return sprintf(buf, "%u\n",
1942 		       channel->offermsg.offer.sub_channel_index);
1943 }
1944 static VMBUS_CHAN_ATTR_RO(subchannel_id);
1945 
1946 static int hv_mmap_ring_buffer_wrapper(struct file *filp, struct kobject *kobj,
1947 				       const struct bin_attribute *attr,
1948 				       struct vm_area_struct *vma)
1949 {
1950 	struct vmbus_channel *channel = container_of(kobj, struct vmbus_channel, kobj);
1951 	struct vm_area_desc desc;
1952 	int err;
1953 
1954 	/*
1955 	 * hv_(create|remove)_ring_sysfs implementation ensures that
1956 	 * mmap_prepare_ring_buffer is not NULL.
1957 	 */
1958 	compat_set_desc_from_vma(&desc, filp, vma);
1959 	err = channel->mmap_prepare_ring_buffer(channel, &desc);
1960 	if (err)
1961 		return err;
1962 
1963 	return __compat_vma_mmap(&desc, vma);
1964 }
1965 
1966 static struct bin_attribute chan_attr_ring_buffer = {
1967 	.attr = {
1968 		.name = "ring",
1969 		.mode = 0600,
1970 	},
1971 	.mmap = hv_mmap_ring_buffer_wrapper,
1972 };
1973 static struct attribute *vmbus_chan_attrs[] = {
1974 	&chan_attr_out_mask.attr,
1975 	&chan_attr_in_mask.attr,
1976 	&chan_attr_read_avail.attr,
1977 	&chan_attr_write_avail.attr,
1978 	&chan_attr_cpu.attr,
1979 	&chan_attr_pending.attr,
1980 	&chan_attr_latency.attr,
1981 	&chan_attr_interrupts.attr,
1982 	&chan_attr_events.attr,
1983 	&chan_attr_intr_in_full.attr,
1984 	&chan_attr_intr_out_empty.attr,
1985 	&chan_attr_out_full_first.attr,
1986 	&chan_attr_out_full_total.attr,
1987 	&chan_attr_monitor_id.attr,
1988 	&chan_attr_subchannel_id.attr,
1989 	NULL
1990 };
1991 
1992 static const struct bin_attribute *vmbus_chan_bin_attrs[] = {
1993 	&chan_attr_ring_buffer,
1994 	NULL
1995 };
1996 
1997 /*
1998  * Channel-level attribute_group callback function. Returns the permission for
1999  * each attribute, and returns 0 if an attribute is not visible.
2000  */
2001 static umode_t vmbus_chan_attr_is_visible(struct kobject *kobj,
2002 					  struct attribute *attr, int idx)
2003 {
2004 	const struct vmbus_channel *channel =
2005 		container_of(kobj, struct vmbus_channel, kobj);
2006 
2007 	/* Hide the monitor attributes if the monitor mechanism is not used. */
2008 	if (!channel->offermsg.monitor_allocated &&
2009 	    (attr == &chan_attr_pending.attr ||
2010 	     attr == &chan_attr_latency.attr ||
2011 	     attr == &chan_attr_monitor_id.attr))
2012 		return 0;
2013 
2014 	return attr->mode;
2015 }
2016 
2017 static umode_t vmbus_chan_bin_attr_is_visible(struct kobject *kobj,
2018 					      const struct bin_attribute *attr, int idx)
2019 {
2020 	const struct vmbus_channel *channel =
2021 		container_of(kobj, struct vmbus_channel, kobj);
2022 
2023 	/* Hide ring attribute if channel's ring_sysfs_visible is set to false */
2024 	if (attr ==  &chan_attr_ring_buffer && !channel->ring_sysfs_visible)
2025 		return 0;
2026 
2027 	return attr->attr.mode;
2028 }
2029 
2030 static size_t vmbus_chan_bin_size(struct kobject *kobj,
2031 				  const struct bin_attribute *bin_attr, int a)
2032 {
2033 	const struct vmbus_channel *channel =
2034 		container_of(kobj, struct vmbus_channel, kobj);
2035 
2036 	return channel->ringbuffer_pagecount << PAGE_SHIFT;
2037 }
2038 
2039 static const struct attribute_group vmbus_chan_group = {
2040 	.attrs = vmbus_chan_attrs,
2041 	.bin_attrs = vmbus_chan_bin_attrs,
2042 	.is_visible = vmbus_chan_attr_is_visible,
2043 	.is_bin_visible = vmbus_chan_bin_attr_is_visible,
2044 	.bin_size = vmbus_chan_bin_size,
2045 };
2046 
2047 static const struct kobj_type vmbus_chan_ktype = {
2048 	.sysfs_ops = &vmbus_chan_sysfs_ops,
2049 	.release = vmbus_chan_release,
2050 };
2051 
2052 /**
2053  * hv_create_ring_sysfs() - create "ring" sysfs entry corresponding to ring buffers for a channel.
2054  * @channel: Pointer to vmbus_channel structure
2055  * @hv_mmap_prepare_ring_buffer: function pointer for initializing the function to be called on mmap
2056  *                       channel's "ring" sysfs node, which is for the ring buffer of that channel.
2057  *                       Function pointer is of below type:
2058  *                       int (*hv_mmap_prepare_ring_buffer)(struct vmbus_channel *channel,
2059  *                                                          struct vm_area_desc *desc))
2060  *                       This has a pointer to the channel and a pointer to vm_area_desc,
2061  *                       used for mmap_prepare, as arguments.
2062  *
2063  * Sysfs node for ring buffer of a channel is created along with other fields, however its
2064  * visibility is disabled by default. Sysfs creation needs to be controlled when the use-case
2065  * is running.
2066  * For example, HV_NIC device is used either by uio_hv_generic or hv_netvsc at any given point of
2067  * time, and "ring" sysfs is needed only when uio_hv_generic is bound to that device. To avoid
2068  * exposing the ring buffer by default, this function is responsible to enable visibility of
2069  * ring for userspace to use.
2070  * Note: Race conditions can happen with userspace and it is not encouraged to create new
2071  * use-cases for this. This was added to maintain backward compatibility, while solving
2072  * one of the race conditions in uio_hv_generic while creating sysfs. See comments with
2073  * vmbus_add_dynid() and vmbus_device_register().
2074  *
2075  * Returns 0 on success or error code on failure.
2076  */
2077 int hv_create_ring_sysfs(struct vmbus_channel *channel,
2078 			 int (*hv_mmap_prepare_ring_buffer)(struct vmbus_channel *channel,
2079 							    struct vm_area_desc *desc))
2080 {
2081 	struct kobject *kobj = &channel->kobj;
2082 
2083 	channel->mmap_prepare_ring_buffer = hv_mmap_prepare_ring_buffer;
2084 	channel->ring_sysfs_visible = true;
2085 
2086 	return sysfs_update_group(kobj, &vmbus_chan_group);
2087 }
2088 EXPORT_SYMBOL_GPL(hv_create_ring_sysfs);
2089 
2090 /**
2091  * hv_remove_ring_sysfs() - remove ring sysfs entry corresponding to ring buffers for a channel.
2092  * @channel: Pointer to vmbus_channel structure
2093  *
2094  * Hide "ring" sysfs for a channel by changing its is_visible attribute and updating sysfs group.
2095  *
2096  * Returns 0 on success or error code on failure.
2097  */
2098 int hv_remove_ring_sysfs(struct vmbus_channel *channel)
2099 {
2100 	struct kobject *kobj = &channel->kobj;
2101 	int ret;
2102 
2103 	channel->ring_sysfs_visible = false;
2104 	ret = sysfs_update_group(kobj, &vmbus_chan_group);
2105 	channel->mmap_prepare_ring_buffer = NULL;
2106 	return ret;
2107 }
2108 EXPORT_SYMBOL_GPL(hv_remove_ring_sysfs);
2109 
2110 /*
2111  * vmbus_add_channel_kobj - setup a sub-directory under device/channels
2112  */
2113 int vmbus_add_channel_kobj(struct hv_device *dev, struct vmbus_channel *channel)
2114 {
2115 	const struct device *device = &dev->device;
2116 	struct kobject *kobj = &channel->kobj;
2117 	u32 relid = channel->offermsg.child_relid;
2118 	int ret;
2119 
2120 	kobj->kset = dev->channels_kset;
2121 	ret = kobject_init_and_add(kobj, &vmbus_chan_ktype, NULL,
2122 				   "%u", relid);
2123 	if (ret) {
2124 		kobject_put(kobj);
2125 		return ret;
2126 	}
2127 
2128 	ret = sysfs_create_group(kobj, &vmbus_chan_group);
2129 
2130 	if (ret) {
2131 		/*
2132 		 * The calling functions' error handling paths will cleanup the
2133 		 * empty channel directory.
2134 		 */
2135 		kobject_put(kobj);
2136 		dev_err(device, "Unable to set up channel sysfs files\n");
2137 		return ret;
2138 	}
2139 
2140 	kobject_uevent(kobj, KOBJ_ADD);
2141 
2142 	return 0;
2143 }
2144 
2145 /*
2146  * vmbus_remove_channel_attr_group - remove the channel's attribute group
2147  */
2148 void vmbus_remove_channel_attr_group(struct vmbus_channel *channel)
2149 {
2150 	sysfs_remove_group(&channel->kobj, &vmbus_chan_group);
2151 }
2152 
2153 /*
2154  * vmbus_device_create - Creates and registers a new child device
2155  * on the vmbus.
2156  */
2157 struct hv_device *vmbus_device_create(const guid_t *type,
2158 				      const guid_t *instance,
2159 				      struct vmbus_channel *channel)
2160 {
2161 	struct hv_device *child_device_obj;
2162 
2163 	child_device_obj = kzalloc_obj(struct hv_device);
2164 	if (!child_device_obj) {
2165 		pr_err("Unable to allocate device object for child device\n");
2166 		return NULL;
2167 	}
2168 
2169 	child_device_obj->channel = channel;
2170 	guid_copy(&child_device_obj->dev_type, type);
2171 	guid_copy(&child_device_obj->dev_instance, instance);
2172 	child_device_obj->vendor_id = PCI_VENDOR_ID_MICROSOFT;
2173 
2174 	return child_device_obj;
2175 }
2176 
2177 /*
2178  * vmbus_device_register - Register the child device
2179  */
2180 int vmbus_device_register(struct hv_device *child_device_obj)
2181 {
2182 	struct kobject *kobj = &child_device_obj->device.kobj;
2183 	int ret;
2184 
2185 	dev_set_name(&child_device_obj->device, "%pUl",
2186 		     &child_device_obj->channel->offermsg.offer.if_instance);
2187 
2188 	child_device_obj->device.bus = &hv_bus;
2189 	child_device_obj->device.parent = vmbus_root_device;
2190 	child_device_obj->device.release = vmbus_device_release;
2191 
2192 	child_device_obj->device.dma_parms = &child_device_obj->dma_parms;
2193 	child_device_obj->device.dma_mask = &child_device_obj->dma_mask;
2194 	dma_set_mask(&child_device_obj->device, DMA_BIT_MASK(64));
2195 
2196 	/*
2197 	 * Register with the LDM. This will kick off the driver/device
2198 	 * binding...which will eventually call vmbus_match() and vmbus_probe()
2199 	 */
2200 	ret = device_register(&child_device_obj->device);
2201 	if (ret) {
2202 		pr_err("Unable to register child device\n");
2203 		put_device(&child_device_obj->device);
2204 		return ret;
2205 	}
2206 
2207 	/*
2208 	 * If device_register() found a driver to assign to the device, the
2209 	 * driver's probe function has already run at this point. If that
2210 	 * probe function accesses or operates on the "channels" subdirectory
2211 	 * in sysfs, those operations will have failed because the "channels"
2212 	 * subdirectory doesn't exist until the code below runs. Or if the
2213 	 * probe function creates a /dev entry, a user space program could
2214 	 * find and open the /dev entry, and then create a race by accessing
2215 	 * the "channels" subdirectory while the creation steps are in progress
2216 	 * here. The race can't result in a kernel failure, but the user space
2217 	 * program may get an error in accessing "channels" or its
2218 	 * subdirectories. See also comments with vmbus_add_dynid() about a
2219 	 * related race condition.
2220 	 */
2221 	child_device_obj->channels_kset = kset_create_and_add("channels",
2222 							      NULL, kobj);
2223 	if (!child_device_obj->channels_kset) {
2224 		ret = -ENOMEM;
2225 		goto err_dev_unregister;
2226 	}
2227 
2228 	ret = vmbus_add_channel_kobj(child_device_obj,
2229 				     child_device_obj->channel);
2230 	if (ret) {
2231 		pr_err("Unable to register primary channel\n");
2232 		goto err_kset_unregister;
2233 	}
2234 	hv_debug_add_dev_dir(child_device_obj);
2235 
2236 	return 0;
2237 
2238 err_kset_unregister:
2239 	kset_unregister(child_device_obj->channels_kset);
2240 
2241 err_dev_unregister:
2242 	device_unregister(&child_device_obj->device);
2243 	return ret;
2244 }
2245 
2246 /*
2247  * vmbus_device_unregister - Remove the specified child device
2248  * from the vmbus.
2249  */
2250 void vmbus_device_unregister(struct hv_device *device_obj)
2251 {
2252 	pr_debug("child device %s unregistered\n",
2253 		dev_name(&device_obj->device));
2254 
2255 	kset_unregister(device_obj->channels_kset);
2256 
2257 	/*
2258 	 * Kick off the process of unregistering the device.
2259 	 * This will call vmbus_remove() and eventually vmbus_device_release()
2260 	 */
2261 	device_unregister(&device_obj->device);
2262 }
2263 EXPORT_SYMBOL_GPL(vmbus_device_unregister);
2264 
2265 #ifdef CONFIG_ACPI
2266 /*
2267  * VMBUS is an acpi enumerated device. Get the information we
2268  * need from DSDT.
2269  */
2270 static acpi_status vmbus_walk_resources(struct acpi_resource *res, void *ctx)
2271 {
2272 	resource_size_t start = 0;
2273 	resource_size_t end = 0;
2274 	struct resource *new_res;
2275 	struct resource **old_res = &hyperv_mmio;
2276 	struct resource **prev_res = NULL;
2277 	struct resource r;
2278 
2279 	switch (res->type) {
2280 
2281 	/*
2282 	 * "Address" descriptors are for bus windows. Ignore
2283 	 * "memory" descriptors, which are for registers on
2284 	 * devices.
2285 	 */
2286 	case ACPI_RESOURCE_TYPE_ADDRESS32:
2287 		start = res->data.address32.address.minimum;
2288 		end = res->data.address32.address.maximum;
2289 		break;
2290 
2291 	case ACPI_RESOURCE_TYPE_ADDRESS64:
2292 		start = res->data.address64.address.minimum;
2293 		end = res->data.address64.address.maximum;
2294 		break;
2295 
2296 	/*
2297 	 * The IRQ information is needed only on ARM64, which Hyper-V
2298 	 * sets up in the extended format. IRQ information is present
2299 	 * on x86/x64 in the non-extended format but it is not used by
2300 	 * Linux. So don't bother checking for the non-extended format.
2301 	 */
2302 	case ACPI_RESOURCE_TYPE_EXTENDED_IRQ:
2303 		if (!acpi_dev_resource_interrupt(res, 0, &r)) {
2304 			pr_err("Unable to parse Hyper-V ACPI interrupt\n");
2305 			return AE_ERROR;
2306 		}
2307 		/* ARM64 INTID for VMbus */
2308 		vmbus_interrupt = res->data.extended_irq.interrupts[0];
2309 		/* Linux IRQ number */
2310 		vmbus_irq = r.start;
2311 		return AE_OK;
2312 
2313 	default:
2314 		/* Unused resource type */
2315 		return AE_OK;
2316 
2317 	}
2318 	/*
2319 	 * Ignore ranges that are below 1MB, as they're not
2320 	 * necessary or useful here.
2321 	 */
2322 	if (end < 0x100000)
2323 		return AE_OK;
2324 
2325 	new_res = kzalloc_obj(*new_res, GFP_ATOMIC);
2326 	if (!new_res)
2327 		return AE_NO_MEMORY;
2328 
2329 	/* If this range overlaps the virtual TPM, truncate it. */
2330 	if (end >= VTPM_BASE_ADDRESS && start < VTPM_BASE_ADDRESS)
2331 		end = VTPM_BASE_ADDRESS - 1;
2332 
2333 	new_res->name = "hyperv mmio";
2334 	new_res->flags = IORESOURCE_MEM;
2335 	new_res->start = start;
2336 	new_res->end = end;
2337 
2338 	/*
2339 	 * If two ranges are adjacent, merge them.
2340 	 */
2341 	do {
2342 		if (!*old_res) {
2343 			*old_res = new_res;
2344 			break;
2345 		}
2346 
2347 		if (((*old_res)->end + 1) == new_res->start) {
2348 			(*old_res)->end = new_res->end;
2349 			kfree(new_res);
2350 			break;
2351 		}
2352 
2353 		if ((*old_res)->start == new_res->end + 1) {
2354 			(*old_res)->start = new_res->start;
2355 			kfree(new_res);
2356 			break;
2357 		}
2358 
2359 		if ((*old_res)->start > new_res->end) {
2360 			new_res->sibling = *old_res;
2361 			if (prev_res)
2362 				(*prev_res)->sibling = new_res;
2363 			*old_res = new_res;
2364 			break;
2365 		}
2366 
2367 		prev_res = old_res;
2368 		old_res = &(*old_res)->sibling;
2369 
2370 	} while (1);
2371 
2372 	return AE_OK;
2373 }
2374 #endif
2375 
2376 static void vmbus_mmio_remove(void)
2377 {
2378 	struct resource *cur_res;
2379 	struct resource *next_res;
2380 
2381 	if (hyperv_mmio) {
2382 		if (fb_mmio) {
2383 			__release_region(hyperv_mmio, fb_mmio->start,
2384 					 resource_size(fb_mmio));
2385 			fb_mmio = NULL;
2386 		}
2387 
2388 		for (cur_res = hyperv_mmio; cur_res; cur_res = next_res) {
2389 			next_res = cur_res->sibling;
2390 			kfree(cur_res);
2391 		}
2392 	}
2393 }
2394 
2395 static void __maybe_unused vmbus_reserve_fb(void)
2396 {
2397 	resource_size_t start = 0, size;
2398 	resource_size_t low_mmio_base;
2399 	struct pci_dev *pdev;
2400 
2401 	if (efi_enabled(EFI_BOOT)) {
2402 		/* Gen2 VM: get FB base from EFI framebuffer */
2403 		if (IS_ENABLED(CONFIG_SYSFB)) {
2404 			start = sysfb_primary_display.screen.lfb_base;
2405 			size = max_t(__u32, sysfb_primary_display.screen.lfb_size, 0x800000);
2406 
2407 			low_mmio_base = hyperv_mmio->start;
2408 			if (!low_mmio_base || upper_32_bits(low_mmio_base) ||
2409 			    (start && start < low_mmio_base)) {
2410 				pr_warn("Unexpected low mmio base %pa\n", &low_mmio_base);
2411 			} else {
2412 				/*
2413 				 * If the kdump/kexec or CVM kernel's lfb_base
2414 				 * is 0, fall back to the low mmio base.
2415 				 */
2416 				if (!start)
2417 					start = low_mmio_base;
2418 				/*
2419 				 * Reserve half of the space below 4GB for high
2420 				 * resolutions, but cap the reservation to 128MB.
2421 				 */
2422 				size = min((SZ_4G - start) / 2, SZ_128M);
2423 			}
2424 		}
2425 	} else {
2426 		/* Gen1 VM: get FB base from PCI */
2427 		pdev = pci_get_device(PCI_VENDOR_ID_MICROSOFT,
2428 				      PCI_DEVICE_ID_HYPERV_VIDEO, NULL);
2429 		if (!pdev)
2430 			return;
2431 
2432 		if (pdev->resource[0].flags & IORESOURCE_MEM) {
2433 			start = pci_resource_start(pdev, 0);
2434 			size = pci_resource_len(pdev, 0);
2435 		}
2436 
2437 		/*
2438 		 * Release the PCI device so hyperv_drm driver can grab it
2439 		 * later.
2440 		 */
2441 		pci_dev_put(pdev);
2442 	}
2443 
2444 	if (!start) {
2445 		pr_warn("Unexpected framebuffer mmio base of zero\n");
2446 		return;
2447 	}
2448 
2449 	/*
2450 	 * Make a claim for the frame buffer in the resource tree under the
2451 	 * first node, which will be the one below 4GB.  The length seems to
2452 	 * be underreported, particularly in a Generation 1 VM.  So start out
2453 	 * reserving a larger area and make it smaller until it succeeds.
2454 	 */
2455 	for (; !fb_mmio && (size >= 0x100000); size >>= 1)
2456 		fb_mmio = __request_region(hyperv_mmio, start, size, fb_mmio_name, 0);
2457 
2458 	pr_info("hv_mmio=%pR,%pR fb=%pR\n", hyperv_mmio, hyperv_mmio->sibling, fb_mmio);
2459 }
2460 
2461 /**
2462  * vmbus_allocate_mmio() - Pick a memory-mapped I/O range.
2463  * @new:		If successful, supplied a pointer to the
2464  *			allocated MMIO space.
2465  * @device_obj:		Identifies the caller
2466  * @min:		Minimum guest physical address of the
2467  *			allocation
2468  * @max:		Maximum guest physical address
2469  * @size:		Size of the range to be allocated
2470  * @align:		Alignment of the range to be allocated
2471  * @fb_overlap_ok:	Whether this allocation can be allowed
2472  *			to overlap the video frame buffer.
2473  *
2474  * This function walks the resources granted to VMBus by the
2475  * _CRS object in the ACPI namespace underneath the parent
2476  * "bridge" whether that's a root PCI bus in the Generation 1
2477  * case or a Module Device in the Generation 2 case.  It then
2478  * attempts to allocate from the global MMIO pool in a way that
2479  * matches the constraints supplied in these parameters and by
2480  * that _CRS.
2481  *
2482  * Return: 0 on success, -errno on failure
2483  */
2484 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
2485 			resource_size_t min, resource_size_t max,
2486 			resource_size_t size, resource_size_t align,
2487 			bool fb_overlap_ok)
2488 {
2489 	struct resource *iter, *shadow;
2490 	resource_size_t range_min, range_max, start, end;
2491 	const char *dev_n = dev_name(&device_obj->device);
2492 	int retval;
2493 
2494 	retval = -ENXIO;
2495 	mutex_lock(&hyperv_mmio_lock);
2496 
2497 	/*
2498 	 * If overlaps with frame buffers are allowed, then first attempt to
2499 	 * make the allocation from within the reserved region.  Because it
2500 	 * is already reserved, no shadow allocation is necessary.
2501 	 */
2502 	if (fb_overlap_ok && fb_mmio && !(min > fb_mmio->end) &&
2503 	    !(max < fb_mmio->start)) {
2504 
2505 		range_min = fb_mmio->start;
2506 		range_max = fb_mmio->end;
2507 		start = (range_min + align - 1) & ~(align - 1);
2508 		for (; start + size - 1 <= range_max; start += align) {
2509 			*new = request_mem_region_exclusive(start, size, dev_n);
2510 			if (*new) {
2511 				retval = 0;
2512 				goto exit;
2513 			}
2514 		}
2515 	}
2516 
2517 	for (iter = hyperv_mmio; iter; iter = iter->sibling) {
2518 		if ((iter->start >= max) || (iter->end <= min))
2519 			continue;
2520 
2521 		range_min = iter->start;
2522 		range_max = iter->end;
2523 		start = (range_min + align - 1) & ~(align - 1);
2524 		for (; start + size - 1 <= range_max; start += align) {
2525 			end = start + size - 1;
2526 
2527 			/* Skip the whole fb_mmio region if not fb_overlap_ok */
2528 			if (!fb_overlap_ok && fb_mmio &&
2529 			    (((start >= fb_mmio->start) && (start <= fb_mmio->end)) ||
2530 			     ((end >= fb_mmio->start) && (end <= fb_mmio->end))))
2531 				continue;
2532 
2533 			shadow = __request_region(iter, start, size, NULL,
2534 						  IORESOURCE_BUSY);
2535 			if (!shadow)
2536 				continue;
2537 
2538 			*new = request_mem_region_exclusive(start, size, dev_n);
2539 			if (*new) {
2540 				shadow->name = (char *)*new;
2541 				retval = 0;
2542 				goto exit;
2543 			}
2544 
2545 			__release_region(iter, start, size);
2546 		}
2547 	}
2548 
2549 exit:
2550 	mutex_unlock(&hyperv_mmio_lock);
2551 	return retval;
2552 }
2553 EXPORT_SYMBOL_GPL(vmbus_allocate_mmio);
2554 
2555 /**
2556  * vmbus_free_mmio() - Free a memory-mapped I/O range.
2557  * @start:		Base address of region to release.
2558  * @size:		Size of the range to be allocated
2559  *
2560  * This function releases anything requested by
2561  * vmbus_mmio_allocate().
2562  */
2563 void vmbus_free_mmio(resource_size_t start, resource_size_t size)
2564 {
2565 	struct resource *iter;
2566 
2567 	mutex_lock(&hyperv_mmio_lock);
2568 
2569 	/*
2570 	 * If all bytes of the MMIO range to be released are within the
2571 	 * special case fb_mmio shadow region, skip releasing the shadow
2572 	 * region since no corresponding __request_region() was done
2573 	 * in vmbus_allocate_mmio().
2574 	 */
2575 	if (fb_mmio && start >= fb_mmio->start &&
2576 	    (start + size - 1 <= fb_mmio->end))
2577 		goto skip_shadow_release;
2578 
2579 	for (iter = hyperv_mmio; iter; iter = iter->sibling) {
2580 		if ((iter->start >= start + size) || (iter->end <= start))
2581 			continue;
2582 
2583 		__release_region(iter, start, size);
2584 	}
2585 
2586 skip_shadow_release:
2587 	release_mem_region(start, size);
2588 	mutex_unlock(&hyperv_mmio_lock);
2589 
2590 }
2591 EXPORT_SYMBOL_GPL(vmbus_free_mmio);
2592 
2593 #ifdef CONFIG_ACPI
2594 static int vmbus_acpi_add(struct platform_device *pdev)
2595 {
2596 	acpi_status result;
2597 	int ret_val = -ENODEV;
2598 	struct acpi_device *ancestor;
2599 	struct acpi_device *device = ACPI_COMPANION(&pdev->dev);
2600 
2601 	vmbus_root_device = &device->dev;
2602 
2603 	/*
2604 	 * Older versions of Hyper-V for ARM64 fail to include the _CCA
2605 	 * method on the top level VMbus device in the DSDT. But devices
2606 	 * are hardware coherent in all current Hyper-V use cases, so fix
2607 	 * up the ACPI device to behave as if _CCA is present and indicates
2608 	 * hardware coherence.
2609 	 */
2610 	ACPI_COMPANION_SET(&device->dev, device);
2611 	if (IS_ENABLED(CONFIG_ACPI_CCA_REQUIRED) &&
2612 	    device_get_dma_attr(&device->dev) == DEV_DMA_NOT_SUPPORTED) {
2613 		pr_info("No ACPI _CCA found; assuming coherent device I/O\n");
2614 		device->flags.cca_seen = true;
2615 		device->flags.coherent_dma = true;
2616 	}
2617 
2618 	result = acpi_walk_resources(device->handle, METHOD_NAME__CRS,
2619 					vmbus_walk_resources, NULL);
2620 
2621 	if (ACPI_FAILURE(result))
2622 		goto acpi_walk_err;
2623 	/*
2624 	 * Some ancestor of the vmbus acpi device (Gen1 or Gen2
2625 	 * firmware) is the VMOD that has the mmio ranges. Get that.
2626 	 */
2627 	for (ancestor = acpi_dev_parent(device);
2628 	     ancestor && ancestor->handle != ACPI_ROOT_OBJECT;
2629 	     ancestor = acpi_dev_parent(ancestor)) {
2630 		result = acpi_walk_resources(ancestor->handle, METHOD_NAME__CRS,
2631 					     vmbus_walk_resources, NULL);
2632 
2633 		if (ACPI_FAILURE(result))
2634 			continue;
2635 		if (hyperv_mmio) {
2636 			vmbus_reserve_fb();
2637 			break;
2638 		}
2639 	}
2640 	ret_val = 0;
2641 
2642 acpi_walk_err:
2643 	if (ret_val)
2644 		vmbus_mmio_remove();
2645 	return ret_val;
2646 }
2647 #else
2648 static int vmbus_acpi_add(struct platform_device *pdev)
2649 {
2650 	return 0;
2651 }
2652 #endif
2653 #ifndef HYPERVISOR_CALLBACK_VECTOR
2654 static int vmbus_set_irq(struct platform_device *pdev)
2655 {
2656 	struct irq_data *data;
2657 	int irq;
2658 	irq_hw_number_t hwirq;
2659 
2660 	irq = platform_get_irq(pdev, 0);
2661 	/* platform_get_irq() may not return 0. */
2662 	if (irq < 0)
2663 		return irq;
2664 
2665 	data = irq_get_irq_data(irq);
2666 	if (!data) {
2667 		pr_err("No interrupt data for VMBus virq %d\n", irq);
2668 		return -ENODEV;
2669 	}
2670 	hwirq = irqd_to_hwirq(data);
2671 
2672 	vmbus_irq = irq;
2673 	vmbus_interrupt = hwirq;
2674 	pr_debug("VMBus virq %d, hwirq %d\n", vmbus_irq, vmbus_interrupt);
2675 
2676 	return 0;
2677 }
2678 #endif
2679 
2680 static int vmbus_device_add(struct platform_device *pdev)
2681 {
2682 	struct resource **cur_res = &hyperv_mmio;
2683 	struct of_range range;
2684 	struct of_range_parser parser;
2685 	struct device_node *np = pdev->dev.of_node;
2686 	int ret;
2687 
2688 	vmbus_root_device = &pdev->dev;
2689 
2690 	ret = of_range_parser_init(&parser, np);
2691 	if (ret)
2692 		return ret;
2693 
2694 #ifndef HYPERVISOR_CALLBACK_VECTOR
2695 	ret = vmbus_set_irq(pdev);
2696 	if (ret)
2697 		return ret;
2698 #endif
2699 	for_each_of_range(&parser, &range) {
2700 		struct resource *res;
2701 
2702 		res = kzalloc_obj(*res);
2703 		if (!res) {
2704 			vmbus_mmio_remove();
2705 			return -ENOMEM;
2706 		}
2707 
2708 		res->name = "hyperv mmio";
2709 		res->flags = range.flags;
2710 		res->start = range.cpu_addr;
2711 		res->end = range.cpu_addr + range.size;
2712 
2713 		*cur_res = res;
2714 		cur_res = &res->sibling;
2715 	}
2716 
2717 	return ret;
2718 }
2719 
2720 static int vmbus_platform_driver_probe(struct platform_device *pdev)
2721 {
2722 	if (acpi_disabled)
2723 		return vmbus_device_add(pdev);
2724 	else
2725 		return vmbus_acpi_add(pdev);
2726 }
2727 
2728 static void vmbus_platform_driver_remove(struct platform_device *pdev)
2729 {
2730 	vmbus_mmio_remove();
2731 }
2732 
2733 #ifdef CONFIG_PM_SLEEP
2734 static int vmbus_bus_suspend(struct device *dev)
2735 {
2736 	struct hv_per_cpu_context *hv_cpu = per_cpu_ptr(
2737 			hv_context.cpu_context, VMBUS_CONNECT_CPU);
2738 	struct vmbus_channel *channel, *sc;
2739 
2740 	tasklet_disable(&hv_cpu->msg_dpc);
2741 	vmbus_connection.ignore_any_offer_msg = true;
2742 	/* The tasklet_enable() takes care of providing a memory barrier */
2743 	tasklet_enable(&hv_cpu->msg_dpc);
2744 
2745 	/* Drain all the workqueues as we are in suspend */
2746 	drain_workqueue(vmbus_connection.rescind_work_queue);
2747 	drain_workqueue(vmbus_connection.work_queue);
2748 	drain_workqueue(vmbus_connection.handle_primary_chan_wq);
2749 	drain_workqueue(vmbus_connection.handle_sub_chan_wq);
2750 
2751 	mutex_lock(&vmbus_connection.channel_mutex);
2752 	list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
2753 		if (!is_hvsock_channel(channel))
2754 			continue;
2755 
2756 		vmbus_force_channel_rescinded(channel);
2757 	}
2758 	mutex_unlock(&vmbus_connection.channel_mutex);
2759 
2760 	/*
2761 	 * Wait until all the sub-channels and hv_sock channels have been
2762 	 * cleaned up. Sub-channels should be destroyed upon suspend, otherwise
2763 	 * they would conflict with the new sub-channels that will be created
2764 	 * in the resume path. hv_sock channels should also be destroyed, but
2765 	 * a hv_sock channel of an established hv_sock connection can not be
2766 	 * really destroyed since it may still be referenced by the userspace
2767 	 * application, so we just force the hv_sock channel to be rescinded
2768 	 * by vmbus_force_channel_rescinded(), and the userspace application
2769 	 * will thoroughly destroy the channel after hibernation.
2770 	 *
2771 	 * Note: the counter nr_chan_close_on_suspend may never go above 0 if
2772 	 * the VM has no sub-channel and hv_sock channel, e.g. a 1-vCPU VM.
2773 	 */
2774 	if (atomic_read(&vmbus_connection.nr_chan_close_on_suspend) > 0)
2775 		wait_for_completion(&vmbus_connection.ready_for_suspend_event);
2776 
2777 	mutex_lock(&vmbus_connection.channel_mutex);
2778 
2779 	list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
2780 		/*
2781 		 * Remove the channel from the array of channels and invalidate
2782 		 * the channel's relid.  Upon resume, vmbus_onoffer() will fix
2783 		 * up the relid (and other fields, if necessary) and add the
2784 		 * channel back to the array.
2785 		 */
2786 		vmbus_channel_unmap_relid(channel);
2787 		channel->offermsg.child_relid = INVALID_RELID;
2788 
2789 		if (is_hvsock_channel(channel)) {
2790 			if (!channel->rescind) {
2791 				pr_err("hv_sock channel not rescinded!\n");
2792 				WARN_ON_ONCE(1);
2793 			}
2794 			continue;
2795 		}
2796 
2797 		list_for_each_entry(sc, &channel->sc_list, sc_list) {
2798 			pr_err("Sub-channel not deleted!\n");
2799 			WARN_ON_ONCE(1);
2800 		}
2801 	}
2802 
2803 	mutex_unlock(&vmbus_connection.channel_mutex);
2804 
2805 	vmbus_initiate_unload(false);
2806 
2807 	return 0;
2808 }
2809 
2810 static int vmbus_bus_resume(struct device *dev)
2811 {
2812 	struct vmbus_channel *channel;
2813 	struct vmbus_channel_msginfo *msginfo;
2814 	size_t msgsize;
2815 	int ret;
2816 
2817 	vmbus_connection.ignore_any_offer_msg = false;
2818 
2819 	/*
2820 	 * We only use the 'vmbus_proto_version', which was in use before
2821 	 * hibernation, to re-negotiate with the host.
2822 	 */
2823 	if (!vmbus_proto_version) {
2824 		pr_err("Invalid proto version = 0x%x\n", vmbus_proto_version);
2825 		return -EINVAL;
2826 	}
2827 
2828 	msgsize = sizeof(*msginfo) +
2829 		  sizeof(struct vmbus_channel_initiate_contact);
2830 
2831 	msginfo = kzalloc(msgsize, GFP_KERNEL);
2832 
2833 	if (msginfo == NULL)
2834 		return -ENOMEM;
2835 
2836 	ret = vmbus_negotiate_version(msginfo, vmbus_proto_version);
2837 
2838 	kfree(msginfo);
2839 
2840 	if (ret != 0)
2841 		return ret;
2842 
2843 	vmbus_request_offers();
2844 
2845 	mutex_lock(&vmbus_connection.channel_mutex);
2846 	list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
2847 		if (channel->offermsg.child_relid != INVALID_RELID)
2848 			continue;
2849 
2850 		/* hvsock channels are not expected to be present. */
2851 		if (is_hvsock_channel(channel))
2852 			continue;
2853 
2854 		pr_err("channel %pUl/%pUl not present after resume.\n",
2855 		       &channel->offermsg.offer.if_type,
2856 		       &channel->offermsg.offer.if_instance);
2857 		/* ToDo: Cleanup these channels here */
2858 	}
2859 	mutex_unlock(&vmbus_connection.channel_mutex);
2860 
2861 	/* Reset the event for the next suspend. */
2862 	reinit_completion(&vmbus_connection.ready_for_suspend_event);
2863 
2864 	return 0;
2865 }
2866 #else
2867 #define vmbus_bus_suspend NULL
2868 #define vmbus_bus_resume NULL
2869 #endif /* CONFIG_PM_SLEEP */
2870 
2871 static const __maybe_unused struct of_device_id vmbus_of_match[] = {
2872 	{
2873 		.compatible = "microsoft,vmbus",
2874 	},
2875 	{
2876 		/* sentinel */
2877 	},
2878 };
2879 MODULE_DEVICE_TABLE(of, vmbus_of_match);
2880 
2881 static const __maybe_unused struct acpi_device_id vmbus_acpi_device_ids[] = {
2882 	{"VMBUS", 0},
2883 	{"VMBus", 0},
2884 	{"", 0},
2885 };
2886 MODULE_DEVICE_TABLE(acpi, vmbus_acpi_device_ids);
2887 
2888 /*
2889  * Note: we must use the "no_irq" ops, otherwise hibernation can not work with
2890  * PCI device assignment, because "pci_dev_pm_ops" uses the "noirq" ops: in
2891  * the resume path, the pci "noirq" restore op runs before "non-noirq" op (see
2892  * resume_target_kernel() -> dpm_resume_start(), and hibernation_restore() ->
2893  * dpm_resume_end()). This means vmbus_bus_resume() and the pci-hyperv's
2894  * resume callback must also run via the "noirq" ops.
2895  *
2896  * Set suspend_noirq/resume_noirq to NULL for Suspend-to-Idle: see the comment
2897  * earlier in this file before vmbus_pm.
2898  */
2899 
2900 static const struct dev_pm_ops vmbus_bus_pm = {
2901 	.suspend_noirq	= NULL,
2902 	.resume_noirq	= NULL,
2903 	.freeze_noirq	= vmbus_bus_suspend,
2904 	.thaw_noirq	= vmbus_bus_resume,
2905 	.poweroff_noirq	= vmbus_bus_suspend,
2906 	.restore_noirq	= vmbus_bus_resume
2907 };
2908 
2909 static struct platform_driver vmbus_platform_driver = {
2910 	.probe = vmbus_platform_driver_probe,
2911 	.remove = vmbus_platform_driver_remove,
2912 	.driver = {
2913 		.name = "vmbus",
2914 		.acpi_match_table = ACPI_PTR(vmbus_acpi_device_ids),
2915 		.of_match_table = of_match_ptr(vmbus_of_match),
2916 		.pm = &vmbus_bus_pm,
2917 		.probe_type = PROBE_FORCE_SYNCHRONOUS,
2918 	}
2919 };
2920 
2921 static void hv_kexec_handler(void)
2922 {
2923 	vmbus_initiate_unload(false);
2924 	/* Make sure conn_state is set as hv_synic_cleanup checks for it */
2925 	mb();
2926 	cpuhp_remove_state(hyperv_cpuhp_online);
2927 };
2928 
2929 static void hv_crash_handler(struct pt_regs *regs)
2930 {
2931 	int cpu;
2932 
2933 	if (!skip_vmbus_unload)
2934 		vmbus_initiate_unload(true);
2935 	/*
2936 	 * In crash handler we can't schedule synic cleanup for all CPUs,
2937 	 * doing the cleanup for current CPU only. This should be sufficient
2938 	 * for kdump.
2939 	 */
2940 	cpu = smp_processor_id();
2941 	hv_stimer_cleanup(cpu);
2942 	hv_hyp_synic_disable_regs(cpu);
2943 };
2944 
2945 static int hv_synic_suspend(void *data)
2946 {
2947 	/*
2948 	 * When we reach here, all the non-boot CPUs have been offlined.
2949 	 * If we're in a legacy configuration where stimer Direct Mode is
2950 	 * not enabled, the stimers on the non-boot CPUs have been unbound
2951 	 * in hv_synic_cleanup() -> hv_stimer_legacy_cleanup() ->
2952 	 * hv_stimer_cleanup() -> clockevents_unbind_device().
2953 	 *
2954 	 * hv_synic_suspend() only runs on CPU0 with interrupts disabled.
2955 	 * Here we do not call hv_stimer_legacy_cleanup() on CPU0 because:
2956 	 * 1) it's unnecessary as interrupts remain disabled between
2957 	 * syscore_suspend() and syscore_resume(): see create_image() and
2958 	 * resume_target_kernel()
2959 	 * 2) the stimer on CPU0 is automatically disabled later by
2960 	 * syscore_suspend() -> timekeeping_suspend() -> tick_suspend() -> ...
2961 	 * -> clockevents_shutdown() -> ... -> hv_ce_shutdown()
2962 	 * 3) a warning would be triggered if we call
2963 	 * clockevents_unbind_device(), which may sleep, in an
2964 	 * interrupts-disabled context.
2965 	 */
2966 
2967 	hv_hyp_synic_disable_regs(0);
2968 
2969 	return 0;
2970 }
2971 
2972 static void hv_synic_resume(void *data)
2973 {
2974 	hv_hyp_synic_enable_regs(0);
2975 
2976 	/*
2977 	 * Note: we don't need to call hv_stimer_init(0), because the timer
2978 	 * on CPU0 is not unbound in hv_synic_suspend(), and the timer is
2979 	 * automatically re-enabled in timekeeping_resume().
2980 	 */
2981 }
2982 
2983 /* The callbacks run only on CPU0, with irqs_disabled. */
2984 static const struct syscore_ops hv_synic_syscore_ops = {
2985 	.suspend = hv_synic_suspend,
2986 	.resume = hv_synic_resume,
2987 };
2988 
2989 static struct syscore hv_synic_syscore = {
2990 	.ops = &hv_synic_syscore_ops,
2991 };
2992 
2993 static int __init hv_acpi_init(void)
2994 {
2995 	int ret;
2996 
2997 	if (!hv_is_hyperv_initialized())
2998 		return -ENODEV;
2999 
3000 	if (hv_root_partition() && !hv_nested)
3001 		return 0;
3002 
3003 	/*
3004 	 * Get ACPI resources first.
3005 	 */
3006 	ret = platform_driver_register(&vmbus_platform_driver);
3007 	if (ret)
3008 		return ret;
3009 
3010 	if (!vmbus_root_device) {
3011 		ret = -ENODEV;
3012 		goto cleanup;
3013 	}
3014 
3015 	/*
3016 	 * If we're on an architecture with a hardcoded hypervisor
3017 	 * vector (i.e. x86/x64), override the VMbus interrupt found
3018 	 * in the ACPI tables. Ensure vmbus_irq is not set since the
3019 	 * normal Linux IRQ mechanism is not used in this case.
3020 	 */
3021 #ifdef HYPERVISOR_CALLBACK_VECTOR
3022 	vmbus_interrupt = HYPERVISOR_CALLBACK_VECTOR;
3023 	vmbus_irq = -1;
3024 #endif
3025 
3026 	hv_debug_init();
3027 
3028 	ret = vmbus_bus_init();
3029 	if (ret)
3030 		goto cleanup;
3031 
3032 	hv_setup_kexec_handler(hv_kexec_handler);
3033 	hv_setup_crash_handler(hv_crash_handler);
3034 
3035 	register_syscore(&hv_synic_syscore);
3036 
3037 	return 0;
3038 
3039 cleanup:
3040 	platform_driver_unregister(&vmbus_platform_driver);
3041 	vmbus_root_device = NULL;
3042 	return ret;
3043 }
3044 
3045 static void __exit vmbus_exit(void)
3046 {
3047 	int cpu;
3048 
3049 	unregister_syscore(&hv_synic_syscore);
3050 
3051 	hv_remove_kexec_handler();
3052 	hv_remove_crash_handler();
3053 	vmbus_connection.conn_state = DISCONNECTED;
3054 	hv_stimer_global_cleanup();
3055 	vmbus_disconnect();
3056 	if (vmbus_irq == -1)
3057 		hv_remove_vmbus_handler();
3058 	else
3059 		free_percpu_irq(vmbus_irq, &vmbus_evt);
3060 	if (IS_ENABLED(CONFIG_PREEMPT_RT) && vmbus_irq_initialized) {
3061 		smpboot_unregister_percpu_thread(&vmbus_irq_threads);
3062 		vmbus_irq_initialized = false;
3063 	}
3064 	for_each_online_cpu(cpu) {
3065 		struct hv_per_cpu_context *hv_cpu
3066 			= per_cpu_ptr(hv_context.cpu_context, cpu);
3067 
3068 		tasklet_kill(&hv_cpu->msg_dpc);
3069 	}
3070 	hv_debug_rm_all_dir();
3071 
3072 	vmbus_free_channels();
3073 	kfree(vmbus_connection.channels);
3074 
3075 	/*
3076 	 * The vmbus panic notifier is always registered, hence we should
3077 	 * also unconditionally unregister it here as well.
3078 	 */
3079 	atomic_notifier_chain_unregister(&panic_notifier_list,
3080 					&hyperv_panic_vmbus_unload_block);
3081 
3082 	bus_unregister(&hv_bus);
3083 
3084 	cpuhp_remove_state(hyperv_cpuhp_online);
3085 	hv_synic_free();
3086 	platform_driver_unregister(&vmbus_platform_driver);
3087 }
3088 
3089 
3090 MODULE_LICENSE("GPL");
3091 MODULE_DESCRIPTION("Microsoft Hyper-V VMBus Driver");
3092 
3093 subsys_initcall(hv_acpi_init);
3094 module_exit(vmbus_exit);
3095