]> git.proxmox.com Git - mirror_ubuntu-hirsute-kernel.git/blob - drivers/hv/vmbus_drv.c
projects / mirror_ubuntu-hirsute-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge tag 'notifications-20200601' of git://git.kernel.org/pub/scm/linux/kernel/git...
[mirror_ubuntu-hirsute-kernel.git] / drivers / hv / vmbus_drv.c
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Copyright (c) 2009, Microsoft Corporation.
4 *
5 * Authors:
6 * Haiyang Zhang <haiyangz@microsoft.com>
7 * Hank Janssen <hjanssen@microsoft.com>
8 * K. Y. Srinivasan <kys@microsoft.com>
9 */
10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11
12 #include <linux/init.h>
13 #include <linux/module.h>
14 #include <linux/device.h>
15 #include <linux/interrupt.h>
16 #include <linux/sysctl.h>
17 #include <linux/slab.h>
18 #include <linux/acpi.h>
19 #include <linux/completion.h>
20 #include <linux/hyperv.h>
21 #include <linux/kernel_stat.h>
22 #include <linux/clockchips.h>
23 #include <linux/cpu.h>
24 #include <linux/sched/task_stack.h>
25
26 #include <asm/mshyperv.h>
27 #include <linux/delay.h>
28 #include <linux/notifier.h>
29 #include <linux/ptrace.h>
30 #include <linux/screen_info.h>
31 #include <linux/kdebug.h>
32 #include <linux/efi.h>
33 #include <linux/random.h>
34 #include <linux/kernel.h>
35 #include <linux/syscore_ops.h>
36 #include <clocksource/hyperv_timer.h>
37 #include "hyperv_vmbus.h"
38
39 struct vmbus_dynid {
40 struct list_head node;
41 struct hv_vmbus_device_id id;
42 };
43
44 static struct acpi_device *hv_acpi_dev;
45
46 static struct completion probe_event;
47
48 static int hyperv_cpuhp_online;
49
50 static void *hv_panic_page;
51
52 /*
53 * Boolean to control whether to report panic messages over Hyper-V.
54 *
55 * It can be set via /proc/sys/kernel/hyperv/record_panic_msg
56 */
57 static int sysctl_record_panic_msg = 1;
58
59 static int hyperv_report_reg(void)
60 {
61 return !sysctl_record_panic_msg || !hv_panic_page;
62 }
63
64 static int hyperv_panic_event(struct notifier_block *nb, unsigned long val,
65 void *args)
66 {
67 struct pt_regs *regs;
68
69 vmbus_initiate_unload(true);
70
71 /*
72 * Hyper-V should be notified only once about a panic. If we will be
73 * doing hyperv_report_panic_msg() later with kmsg data, don't do
74 * the notification here.
75 */
76 if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE
77 && hyperv_report_reg()) {
78 regs = current_pt_regs();
79 hyperv_report_panic(regs, val, false);
80 }
81 return NOTIFY_DONE;
82 }
83
84 static int hyperv_die_event(struct notifier_block *nb, unsigned long val,
85 void *args)
86 {
87 struct die_args *die = (struct die_args *)args;
88 struct pt_regs *regs = die->regs;
89
90 /*
91 * Hyper-V should be notified only once about a panic. If we will be
92 * doing hyperv_report_panic_msg() later with kmsg data, don't do
93 * the notification here.
94 */
95 if (hyperv_report_reg())
96 hyperv_report_panic(regs, val, true);
97 return NOTIFY_DONE;
98 }
99
100 static struct notifier_block hyperv_die_block = {
101 .notifier_call = hyperv_die_event,
102 };
103 static struct notifier_block hyperv_panic_block = {
104 .notifier_call = hyperv_panic_event,
105 };
106
107 static const char *fb_mmio_name = "fb_range";
108 static struct resource *fb_mmio;
109 static struct resource *hyperv_mmio;
110 static DEFINE_MUTEX(hyperv_mmio_lock);
111
112 static int vmbus_exists(void)
113 {
114 if (hv_acpi_dev == NULL)
115 return -ENODEV;
116
117 return 0;
118 }
119
120 static u8 channel_monitor_group(const struct vmbus_channel *channel)
121 {
122 return (u8)channel->offermsg.monitorid / 32;
123 }
124
125 static u8 channel_monitor_offset(const struct vmbus_channel *channel)
126 {
127 return (u8)channel->offermsg.monitorid % 32;
128 }
129
130 static u32 channel_pending(const struct vmbus_channel *channel,
131 const struct hv_monitor_page *monitor_page)
132 {
133 u8 monitor_group = channel_monitor_group(channel);
134
135 return monitor_page->trigger_group[monitor_group].pending;
136 }
137
138 static u32 channel_latency(const struct vmbus_channel *channel,
139 const struct hv_monitor_page *monitor_page)
140 {
141 u8 monitor_group = channel_monitor_group(channel);
142 u8 monitor_offset = channel_monitor_offset(channel);
143
144 return monitor_page->latency[monitor_group][monitor_offset];
145 }
146
147 static u32 channel_conn_id(struct vmbus_channel *channel,
148 struct hv_monitor_page *monitor_page)
149 {
150 u8 monitor_group = channel_monitor_group(channel);
151 u8 monitor_offset = channel_monitor_offset(channel);
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 sprintf(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 sprintf(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 sprintf(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 sprintf(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 sprintf(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 sprintf(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 sprintf(buf, "%d\n", hv_dev->channel->numa_node);
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 sprintf(buf, "%d\n",
244 channel_pending(hv_dev->channel,
245 vmbus_connection.monitor_pages[0]));
246 }
247 static DEVICE_ATTR_RO(server_monitor_pending);
248
249 static ssize_t client_monitor_pending_show(struct device *dev,
250 struct device_attribute *dev_attr,
251 char *buf)
252 {
253 struct hv_device *hv_dev = device_to_hv_device(dev);
254
255 if (!hv_dev->channel)
256 return -ENODEV;
257 return sprintf(buf, "%d\n",
258 channel_pending(hv_dev->channel,
259 vmbus_connection.monitor_pages[1]));
260 }
261 static DEVICE_ATTR_RO(client_monitor_pending);
262
263 static ssize_t server_monitor_latency_show(struct device *dev,
264 struct device_attribute *dev_attr,
265 char *buf)
266 {
267 struct hv_device *hv_dev = device_to_hv_device(dev);
268
269 if (!hv_dev->channel)
270 return -ENODEV;
271 return sprintf(buf, "%d\n",
272 channel_latency(hv_dev->channel,
273 vmbus_connection.monitor_pages[0]));
274 }
275 static DEVICE_ATTR_RO(server_monitor_latency);
276
277 static ssize_t client_monitor_latency_show(struct device *dev,
278 struct device_attribute *dev_attr,
279 char *buf)
280 {
281 struct hv_device *hv_dev = device_to_hv_device(dev);
282
283 if (!hv_dev->channel)
284 return -ENODEV;
285 return sprintf(buf, "%d\n",
286 channel_latency(hv_dev->channel,
287 vmbus_connection.monitor_pages[1]));
288 }
289 static DEVICE_ATTR_RO(client_monitor_latency);
290
291 static ssize_t server_monitor_conn_id_show(struct device *dev,
292 struct device_attribute *dev_attr,
293 char *buf)
294 {
295 struct hv_device *hv_dev = device_to_hv_device(dev);
296
297 if (!hv_dev->channel)
298 return -ENODEV;
299 return sprintf(buf, "%d\n",
300 channel_conn_id(hv_dev->channel,
301 vmbus_connection.monitor_pages[0]));
302 }
303 static DEVICE_ATTR_RO(server_monitor_conn_id);
304
305 static ssize_t client_monitor_conn_id_show(struct device *dev,
306 struct device_attribute *dev_attr,
307 char *buf)
308 {
309 struct hv_device *hv_dev = device_to_hv_device(dev);
310
311 if (!hv_dev->channel)
312 return -ENODEV;
313 return sprintf(buf, "%d\n",
314 channel_conn_id(hv_dev->channel,
315 vmbus_connection.monitor_pages[1]));
316 }
317 static DEVICE_ATTR_RO(client_monitor_conn_id);
318
319 static ssize_t out_intr_mask_show(struct device *dev,
320 struct device_attribute *dev_attr, char *buf)
321 {
322 struct hv_device *hv_dev = device_to_hv_device(dev);
323 struct hv_ring_buffer_debug_info outbound;
324 int ret;
325
326 if (!hv_dev->channel)
327 return -ENODEV;
328
329 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
330 &outbound);
331 if (ret < 0)
332 return ret;
333
334 return sprintf(buf, "%d\n", outbound.current_interrupt_mask);
335 }
336 static DEVICE_ATTR_RO(out_intr_mask);
337
338 static ssize_t out_read_index_show(struct device *dev,
339 struct device_attribute *dev_attr, char *buf)
340 {
341 struct hv_device *hv_dev = device_to_hv_device(dev);
342 struct hv_ring_buffer_debug_info outbound;
343 int ret;
344
345 if (!hv_dev->channel)
346 return -ENODEV;
347
348 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
349 &outbound);
350 if (ret < 0)
351 return ret;
352 return sprintf(buf, "%d\n", outbound.current_read_index);
353 }
354 static DEVICE_ATTR_RO(out_read_index);
355
356 static ssize_t out_write_index_show(struct device *dev,
357 struct device_attribute *dev_attr,
358 char *buf)
359 {
360 struct hv_device *hv_dev = device_to_hv_device(dev);
361 struct hv_ring_buffer_debug_info outbound;
362 int ret;
363
364 if (!hv_dev->channel)
365 return -ENODEV;
366
367 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
368 &outbound);
369 if (ret < 0)
370 return ret;
371 return sprintf(buf, "%d\n", outbound.current_write_index);
372 }
373 static DEVICE_ATTR_RO(out_write_index);
374
375 static ssize_t out_read_bytes_avail_show(struct device *dev,
376 struct device_attribute *dev_attr,
377 char *buf)
378 {
379 struct hv_device *hv_dev = device_to_hv_device(dev);
380 struct hv_ring_buffer_debug_info outbound;
381 int ret;
382
383 if (!hv_dev->channel)
384 return -ENODEV;
385
386 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
387 &outbound);
388 if (ret < 0)
389 return ret;
390 return sprintf(buf, "%d\n", outbound.bytes_avail_toread);
391 }
392 static DEVICE_ATTR_RO(out_read_bytes_avail);
393
394 static ssize_t out_write_bytes_avail_show(struct device *dev,
395 struct device_attribute *dev_attr,
396 char *buf)
397 {
398 struct hv_device *hv_dev = device_to_hv_device(dev);
399 struct hv_ring_buffer_debug_info outbound;
400 int ret;
401
402 if (!hv_dev->channel)
403 return -ENODEV;
404
405 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
406 &outbound);
407 if (ret < 0)
408 return ret;
409 return sprintf(buf, "%d\n", outbound.bytes_avail_towrite);
410 }
411 static DEVICE_ATTR_RO(out_write_bytes_avail);
412
413 static ssize_t in_intr_mask_show(struct device *dev,
414 struct device_attribute *dev_attr, char *buf)
415 {
416 struct hv_device *hv_dev = device_to_hv_device(dev);
417 struct hv_ring_buffer_debug_info inbound;
418 int ret;
419
420 if (!hv_dev->channel)
421 return -ENODEV;
422
423 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
424 if (ret < 0)
425 return ret;
426
427 return sprintf(buf, "%d\n", inbound.current_interrupt_mask);
428 }
429 static DEVICE_ATTR_RO(in_intr_mask);
430
431 static ssize_t in_read_index_show(struct device *dev,
432 struct device_attribute *dev_attr, char *buf)
433 {
434 struct hv_device *hv_dev = device_to_hv_device(dev);
435 struct hv_ring_buffer_debug_info inbound;
436 int ret;
437
438 if (!hv_dev->channel)
439 return -ENODEV;
440
441 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
442 if (ret < 0)
443 return ret;
444
445 return sprintf(buf, "%d\n", inbound.current_read_index);
446 }
447 static DEVICE_ATTR_RO(in_read_index);
448
449 static ssize_t in_write_index_show(struct device *dev,
450 struct device_attribute *dev_attr, char *buf)
451 {
452 struct hv_device *hv_dev = device_to_hv_device(dev);
453 struct hv_ring_buffer_debug_info inbound;
454 int ret;
455
456 if (!hv_dev->channel)
457 return -ENODEV;
458
459 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
460 if (ret < 0)
461 return ret;
462
463 return sprintf(buf, "%d\n", inbound.current_write_index);
464 }
465 static DEVICE_ATTR_RO(in_write_index);
466
467 static ssize_t in_read_bytes_avail_show(struct device *dev,
468 struct device_attribute *dev_attr,
469 char *buf)
470 {
471 struct hv_device *hv_dev = device_to_hv_device(dev);
472 struct hv_ring_buffer_debug_info inbound;
473 int ret;
474
475 if (!hv_dev->channel)
476 return -ENODEV;
477
478 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
479 if (ret < 0)
480 return ret;
481
482 return sprintf(buf, "%d\n", inbound.bytes_avail_toread);
483 }
484 static DEVICE_ATTR_RO(in_read_bytes_avail);
485
486 static ssize_t in_write_bytes_avail_show(struct device *dev,
487 struct device_attribute *dev_attr,
488 char *buf)
489 {
490 struct hv_device *hv_dev = device_to_hv_device(dev);
491 struct hv_ring_buffer_debug_info inbound;
492 int ret;
493
494 if (!hv_dev->channel)
495 return -ENODEV;
496
497 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
498 if (ret < 0)
499 return ret;
500
501 return sprintf(buf, "%d\n", inbound.bytes_avail_towrite);
502 }
503 static DEVICE_ATTR_RO(in_write_bytes_avail);
504
505 static ssize_t channel_vp_mapping_show(struct device *dev,
506 struct device_attribute *dev_attr,
507 char *buf)
508 {
509 struct hv_device *hv_dev = device_to_hv_device(dev);
510 struct vmbus_channel *channel = hv_dev->channel, *cur_sc;
511 unsigned long flags;
512 int buf_size = PAGE_SIZE, n_written, tot_written;
513 struct list_head *cur;
514
515 if (!channel)
516 return -ENODEV;
517
518 tot_written = snprintf(buf, buf_size, "%u:%u\n",
519 channel->offermsg.child_relid, channel->target_cpu);
520
521 spin_lock_irqsave(&channel->lock, flags);
522
523 list_for_each(cur, &channel->sc_list) {
524 if (tot_written >= buf_size - 1)
525 break;
526
527 cur_sc = list_entry(cur, struct vmbus_channel, sc_list);
528 n_written = scnprintf(buf + tot_written,
529 buf_size - tot_written,
530 "%u:%u\n",
531 cur_sc->offermsg.child_relid,
532 cur_sc->target_cpu);
533 tot_written += n_written;
534 }
535
536 spin_unlock_irqrestore(&channel->lock, flags);
537
538 return tot_written;
539 }
540 static DEVICE_ATTR_RO(channel_vp_mapping);
541
542 static ssize_t vendor_show(struct device *dev,
543 struct device_attribute *dev_attr,
544 char *buf)
545 {
546 struct hv_device *hv_dev = device_to_hv_device(dev);
547 return sprintf(buf, "0x%x\n", hv_dev->vendor_id);
548 }
549 static DEVICE_ATTR_RO(vendor);
550
551 static ssize_t device_show(struct device *dev,
552 struct device_attribute *dev_attr,
553 char *buf)
554 {
555 struct hv_device *hv_dev = device_to_hv_device(dev);
556 return sprintf(buf, "0x%x\n", hv_dev->device_id);
557 }
558 static DEVICE_ATTR_RO(device);
559
560 static ssize_t driver_override_store(struct device *dev,
561 struct device_attribute *attr,
562 const char *buf, size_t count)
563 {
564 struct hv_device *hv_dev = device_to_hv_device(dev);
565 char *driver_override, *old, *cp;
566
567 /* We need to keep extra room for a newline */
568 if (count >= (PAGE_SIZE - 1))
569 return -EINVAL;
570
571 driver_override = kstrndup(buf, count, GFP_KERNEL);
572 if (!driver_override)
573 return -ENOMEM;
574
575 cp = strchr(driver_override, '\n');
576 if (cp)
577 *cp = '\0';
578
579 device_lock(dev);
580 old = hv_dev->driver_override;
581 if (strlen(driver_override)) {
582 hv_dev->driver_override = driver_override;
583 } else {
584 kfree(driver_override);
585 hv_dev->driver_override = NULL;
586 }
587 device_unlock(dev);
588
589 kfree(old);
590
591 return count;
592 }
593
594 static ssize_t driver_override_show(struct device *dev,
595 struct device_attribute *attr, char *buf)
596 {
597 struct hv_device *hv_dev = device_to_hv_device(dev);
598 ssize_t len;
599
600 device_lock(dev);
601 len = snprintf(buf, PAGE_SIZE, "%s\n", hv_dev->driver_override);
602 device_unlock(dev);
603
604 return len;
605 }
606 static DEVICE_ATTR_RW(driver_override);
607
608 /* Set up per device attributes in /sys/bus/vmbus/devices/<bus device> */
609 static struct attribute *vmbus_dev_attrs[] = {
610 &dev_attr_id.attr,
611 &dev_attr_state.attr,
612 &dev_attr_monitor_id.attr,
613 &dev_attr_class_id.attr,
614 &dev_attr_device_id.attr,
615 &dev_attr_modalias.attr,
616 #ifdef CONFIG_NUMA
617 &dev_attr_numa_node.attr,
618 #endif
619 &dev_attr_server_monitor_pending.attr,
620 &dev_attr_client_monitor_pending.attr,
621 &dev_attr_server_monitor_latency.attr,
622 &dev_attr_client_monitor_latency.attr,
623 &dev_attr_server_monitor_conn_id.attr,
624 &dev_attr_client_monitor_conn_id.attr,
625 &dev_attr_out_intr_mask.attr,
626 &dev_attr_out_read_index.attr,
627 &dev_attr_out_write_index.attr,
628 &dev_attr_out_read_bytes_avail.attr,
629 &dev_attr_out_write_bytes_avail.attr,
630 &dev_attr_in_intr_mask.attr,
631 &dev_attr_in_read_index.attr,
632 &dev_attr_in_write_index.attr,
633 &dev_attr_in_read_bytes_avail.attr,
634 &dev_attr_in_write_bytes_avail.attr,
635 &dev_attr_channel_vp_mapping.attr,
636 &dev_attr_vendor.attr,
637 &dev_attr_device.attr,
638 &dev_attr_driver_override.attr,
639 NULL,
640 };
641
642 /*
643 * Device-level attribute_group callback function. Returns the permission for
644 * each attribute, and returns 0 if an attribute is not visible.
645 */
646 static umode_t vmbus_dev_attr_is_visible(struct kobject *kobj,
647 struct attribute *attr, int idx)
648 {
649 struct device *dev = kobj_to_dev(kobj);
650 const struct hv_device *hv_dev = device_to_hv_device(dev);
651
652 /* Hide the monitor attributes if the monitor mechanism is not used. */
653 if (!hv_dev->channel->offermsg.monitor_allocated &&
654 (attr == &dev_attr_monitor_id.attr ||
655 attr == &dev_attr_server_monitor_pending.attr ||
656 attr == &dev_attr_client_monitor_pending.attr ||
657 attr == &dev_attr_server_monitor_latency.attr ||
658 attr == &dev_attr_client_monitor_latency.attr ||
659 attr == &dev_attr_server_monitor_conn_id.attr ||
660 attr == &dev_attr_client_monitor_conn_id.attr))
661 return 0;
662
663 return attr->mode;
664 }
665
666 static const struct attribute_group vmbus_dev_group = {
667 .attrs = vmbus_dev_attrs,
668 .is_visible = vmbus_dev_attr_is_visible
669 };
670 __ATTRIBUTE_GROUPS(vmbus_dev);
671
672 /*
673 * vmbus_uevent - add uevent for our device
674 *
675 * This routine is invoked when a device is added or removed on the vmbus to
676 * generate a uevent to udev in the userspace. The udev will then look at its
677 * rule and the uevent generated here to load the appropriate driver
678 *
679 * The alias string will be of the form vmbus:guid where guid is the string
680 * representation of the device guid (each byte of the guid will be
681 * represented with two hex characters.
682 */
683 static int vmbus_uevent(struct device *device, struct kobj_uevent_env *env)
684 {
685 struct hv_device *dev = device_to_hv_device(device);
686 const char *format = "MODALIAS=vmbus:%*phN";
687
688 return add_uevent_var(env, format, UUID_SIZE, &dev->dev_type);
689 }
690
691 static const struct hv_vmbus_device_id *
692 hv_vmbus_dev_match(const struct hv_vmbus_device_id *id, const guid_t *guid)
693 {
694 if (id == NULL)
695 return NULL; /* empty device table */
696
697 for (; !guid_is_null(&id->guid); id++)
698 if (guid_equal(&id->guid, guid))
699 return id;
700
701 return NULL;
702 }
703
704 static const struct hv_vmbus_device_id *
705 hv_vmbus_dynid_match(struct hv_driver *drv, const guid_t *guid)
706 {
707 const struct hv_vmbus_device_id *id = NULL;
708 struct vmbus_dynid *dynid;
709
710 spin_lock(&drv->dynids.lock);
711 list_for_each_entry(dynid, &drv->dynids.list, node) {
712 if (guid_equal(&dynid->id.guid, guid)) {
713 id = &dynid->id;
714 break;
715 }
716 }
717 spin_unlock(&drv->dynids.lock);
718
719 return id;
720 }
721
722 static const struct hv_vmbus_device_id vmbus_device_null;
723
724 /*
725 * Return a matching hv_vmbus_device_id pointer.
726 * If there is no match, return NULL.
727 */
728 static const struct hv_vmbus_device_id *hv_vmbus_get_id(struct hv_driver *drv,
729 struct hv_device *dev)
730 {
731 const guid_t *guid = &dev->dev_type;
732 const struct hv_vmbus_device_id *id;
733
734 /* When driver_override is set, only bind to the matching driver */
735 if (dev->driver_override && strcmp(dev->driver_override, drv->name))
736 return NULL;
737
738 /* Look at the dynamic ids first, before the static ones */
739 id = hv_vmbus_dynid_match(drv, guid);
740 if (!id)
741 id = hv_vmbus_dev_match(drv->id_table, guid);
742
743 /* driver_override will always match, send a dummy id */
744 if (!id && dev->driver_override)
745 id = &vmbus_device_null;
746
747 return id;
748 }
749
750 /* vmbus_add_dynid - add a new device ID to this driver and re-probe devices */
751 static int vmbus_add_dynid(struct hv_driver *drv, guid_t *guid)
752 {
753 struct vmbus_dynid *dynid;
754
755 dynid = kzalloc(sizeof(*dynid), GFP_KERNEL);
756 if (!dynid)
757 return -ENOMEM;
758
759 dynid->id.guid = *guid;
760
761 spin_lock(&drv->dynids.lock);
762 list_add_tail(&dynid->node, &drv->dynids.list);
763 spin_unlock(&drv->dynids.lock);
764
765 return driver_attach(&drv->driver);
766 }
767
768 static void vmbus_free_dynids(struct hv_driver *drv)
769 {
770 struct vmbus_dynid *dynid, *n;
771
772 spin_lock(&drv->dynids.lock);
773 list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
774 list_del(&dynid->node);
775 kfree(dynid);
776 }
777 spin_unlock(&drv->dynids.lock);
778 }
779
780 /*
781 * store_new_id - sysfs frontend to vmbus_add_dynid()
782 *
783 * Allow GUIDs to be added to an existing driver via sysfs.
784 */
785 static ssize_t new_id_store(struct device_driver *driver, const char *buf,
786 size_t count)
787 {
788 struct hv_driver *drv = drv_to_hv_drv(driver);
789 guid_t guid;
790 ssize_t retval;
791
792 retval = guid_parse(buf, &guid);
793 if (retval)
794 return retval;
795
796 if (hv_vmbus_dynid_match(drv, &guid))
797 return -EEXIST;
798
799 retval = vmbus_add_dynid(drv, &guid);
800 if (retval)
801 return retval;
802 return count;
803 }
804 static DRIVER_ATTR_WO(new_id);
805
806 /*
807 * store_remove_id - remove a PCI device ID from this driver
808 *
809 * Removes a dynamic pci device ID to this driver.
810 */
811 static ssize_t remove_id_store(struct device_driver *driver, const char *buf,
812 size_t count)
813 {
814 struct hv_driver *drv = drv_to_hv_drv(driver);
815 struct vmbus_dynid *dynid, *n;
816 guid_t guid;
817 ssize_t retval;
818
819 retval = guid_parse(buf, &guid);
820 if (retval)
821 return retval;
822
823 retval = -ENODEV;
824 spin_lock(&drv->dynids.lock);
825 list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
826 struct hv_vmbus_device_id *id = &dynid->id;
827
828 if (guid_equal(&id->guid, &guid)) {
829 list_del(&dynid->node);
830 kfree(dynid);
831 retval = count;
832 break;
833 }
834 }
835 spin_unlock(&drv->dynids.lock);
836
837 return retval;
838 }
839 static DRIVER_ATTR_WO(remove_id);
840
841 static struct attribute *vmbus_drv_attrs[] = {
842 &driver_attr_new_id.attr,
843 &driver_attr_remove_id.attr,
844 NULL,
845 };
846 ATTRIBUTE_GROUPS(vmbus_drv);
847
848
849 /*
850 * vmbus_match - Attempt to match the specified device to the specified driver
851 */
852 static int vmbus_match(struct device *device, struct device_driver *driver)
853 {
854 struct hv_driver *drv = drv_to_hv_drv(driver);
855 struct hv_device *hv_dev = device_to_hv_device(device);
856
857 /* The hv_sock driver handles all hv_sock offers. */
858 if (is_hvsock_channel(hv_dev->channel))
859 return drv->hvsock;
860
861 if (hv_vmbus_get_id(drv, hv_dev))
862 return 1;
863
864 return 0;
865 }
866
867 /*
868 * vmbus_probe - Add the new vmbus's child device
869 */
870 static int vmbus_probe(struct device *child_device)
871 {
872 int ret = 0;
873 struct hv_driver *drv =
874 drv_to_hv_drv(child_device->driver);
875 struct hv_device *dev = device_to_hv_device(child_device);
876 const struct hv_vmbus_device_id *dev_id;
877
878 dev_id = hv_vmbus_get_id(drv, dev);
879 if (drv->probe) {
880 ret = drv->probe(dev, dev_id);
881 if (ret != 0)
882 pr_err("probe failed for device %s (%d)\n",
883 dev_name(child_device), ret);
884
885 } else {
886 pr_err("probe not set for driver %s\n",
887 dev_name(child_device));
888 ret = -ENODEV;
889 }
890 return ret;
891 }
892
893 /*
894 * vmbus_remove - Remove a vmbus device
895 */
896 static int vmbus_remove(struct device *child_device)
897 {
898 struct hv_driver *drv;
899 struct hv_device *dev = device_to_hv_device(child_device);
900
901 if (child_device->driver) {
902 drv = drv_to_hv_drv(child_device->driver);
903 if (drv->remove)
904 drv->remove(dev);
905 }
906
907 return 0;
908 }
909
910
911 /*
912 * vmbus_shutdown - Shutdown a vmbus device
913 */
914 static void vmbus_shutdown(struct device *child_device)
915 {
916 struct hv_driver *drv;
917 struct hv_device *dev = device_to_hv_device(child_device);
918
919
920 /* The device may not be attached yet */
921 if (!child_device->driver)
922 return;
923
924 drv = drv_to_hv_drv(child_device->driver);
925
926 if (drv->shutdown)
927 drv->shutdown(dev);
928 }
929
930 #ifdef CONFIG_PM_SLEEP
931 /*
932 * vmbus_suspend - Suspend a vmbus device
933 */
934 static int vmbus_suspend(struct device *child_device)
935 {
936 struct hv_driver *drv;
937 struct hv_device *dev = device_to_hv_device(child_device);
938
939 /* The device may not be attached yet */
940 if (!child_device->driver)
941 return 0;
942
943 drv = drv_to_hv_drv(child_device->driver);
944 if (!drv->suspend)
945 return -EOPNOTSUPP;
946
947 return drv->suspend(dev);
948 }
949
950 /*
951 * vmbus_resume - Resume a vmbus device
952 */
953 static int vmbus_resume(struct device *child_device)
954 {
955 struct hv_driver *drv;
956 struct hv_device *dev = device_to_hv_device(child_device);
957
958 /* The device may not be attached yet */
959 if (!child_device->driver)
960 return 0;
961
962 drv = drv_to_hv_drv(child_device->driver);
963 if (!drv->resume)
964 return -EOPNOTSUPP;
965
966 return drv->resume(dev);
967 }
968 #else
969 #define vmbus_suspend NULL
970 #define vmbus_resume NULL
971 #endif /* CONFIG_PM_SLEEP */
972
973 /*
974 * vmbus_device_release - Final callback release of the vmbus child device
975 */
976 static void vmbus_device_release(struct device *device)
977 {
978 struct hv_device *hv_dev = device_to_hv_device(device);
979 struct vmbus_channel *channel = hv_dev->channel;
980
981 hv_debug_rm_dev_dir(hv_dev);
982
983 mutex_lock(&vmbus_connection.channel_mutex);
984 hv_process_channel_removal(channel);
985 mutex_unlock(&vmbus_connection.channel_mutex);
986 kfree(hv_dev);
987 }
988
989 /*
990 * Note: we must use the "noirq" ops: see the comment before vmbus_bus_pm.
991 *
992 * suspend_noirq/resume_noirq are set to NULL to support Suspend-to-Idle: we
993 * shouldn't suspend the vmbus devices upon Suspend-to-Idle, otherwise there
994 * is no way to wake up a Generation-2 VM.
995 *
996 * The other 4 ops are for hibernation.
997 */
998
999 static const struct dev_pm_ops vmbus_pm = {
1000 .suspend_noirq = NULL,
1001 .resume_noirq = NULL,
1002 .freeze_noirq = vmbus_suspend,
1003 .thaw_noirq = vmbus_resume,
1004 .poweroff_noirq = vmbus_suspend,
1005 .restore_noirq = vmbus_resume,
1006 };
1007
1008 /* The one and only one */
1009 static struct bus_type hv_bus = {
1010 .name = "vmbus",
1011 .match = vmbus_match,
1012 .shutdown = vmbus_shutdown,
1013 .remove = vmbus_remove,
1014 .probe = vmbus_probe,
1015 .uevent = vmbus_uevent,
1016 .dev_groups = vmbus_dev_groups,
1017 .drv_groups = vmbus_drv_groups,
1018 .pm = &vmbus_pm,
1019 };
1020
1021 struct onmessage_work_context {
1022 struct work_struct work;
1023 struct {
1024 struct hv_message_header header;
1025 u8 payload[];
1026 } msg;
1027 };
1028
1029 static void vmbus_onmessage_work(struct work_struct *work)
1030 {
1031 struct onmessage_work_context *ctx;
1032
1033 /* Do not process messages if we're in DISCONNECTED state */
1034 if (vmbus_connection.conn_state == DISCONNECTED)
1035 return;
1036
1037 ctx = container_of(work, struct onmessage_work_context,
1038 work);
1039 vmbus_onmessage((struct vmbus_channel_message_header *)
1040 &ctx->msg.payload);
1041 kfree(ctx);
1042 }
1043
1044 void vmbus_on_msg_dpc(unsigned long data)
1045 {
1046 struct hv_per_cpu_context *hv_cpu = (void *)data;
1047 void *page_addr = hv_cpu->synic_message_page;
1048 struct hv_message *msg = (struct hv_message *)page_addr +
1049 VMBUS_MESSAGE_SINT;
1050 struct vmbus_channel_message_header *hdr;
1051 const struct vmbus_channel_message_table_entry *entry;
1052 struct onmessage_work_context *ctx;
1053 u32 message_type = msg->header.message_type;
1054
1055 /*
1056 * 'enum vmbus_channel_message_type' is supposed to always be 'u32' as
1057 * it is being used in 'struct vmbus_channel_message_header' definition
1058 * which is supposed to match hypervisor ABI.
1059 */
1060 BUILD_BUG_ON(sizeof(enum vmbus_channel_message_type) != sizeof(u32));
1061
1062 if (message_type == HVMSG_NONE)
1063 /* no msg */
1064 return;
1065
1066 hdr = (struct vmbus_channel_message_header *)msg->u.payload;
1067
1068 trace_vmbus_on_msg_dpc(hdr);
1069
1070 if (hdr->msgtype >= CHANNELMSG_COUNT) {
1071 WARN_ONCE(1, "unknown msgtype=%d\n", hdr->msgtype);
1072 goto msg_handled;
1073 }
1074
1075 if (msg->header.payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT) {
1076 WARN_ONCE(1, "payload size is too large (%d)\n",
1077 msg->header.payload_size);
1078 goto msg_handled;
1079 }
1080
1081 entry = &channel_message_table[hdr->msgtype];
1082
1083 if (!entry->message_handler)
1084 goto msg_handled;
1085
1086 if (msg->header.payload_size < entry->min_payload_len) {
1087 WARN_ONCE(1, "message too short: msgtype=%d len=%d\n",
1088 hdr->msgtype, msg->header.payload_size);
1089 goto msg_handled;
1090 }
1091
1092 if (entry->handler_type == VMHT_BLOCKING) {
1093 ctx = kmalloc(sizeof(*ctx) + msg->header.payload_size,
1094 GFP_ATOMIC);
1095 if (ctx == NULL)
1096 return;
1097
1098 INIT_WORK(&ctx->work, vmbus_onmessage_work);
1099 memcpy(&ctx->msg, msg, sizeof(msg->header) +
1100 msg->header.payload_size);
1101
1102 /*
1103 * The host can generate a rescind message while we
1104 * may still be handling the original offer. We deal with
1105 * this condition by relying on the synchronization provided
1106 * by offer_in_progress and by channel_mutex. See also the
1107 * inline comments in vmbus_onoffer_rescind().
1108 */
1109 switch (hdr->msgtype) {
1110 case CHANNELMSG_RESCIND_CHANNELOFFER:
1111 /*
1112 * If we are handling the rescind message;
1113 * schedule the work on the global work queue.
1114 *
1115 * The OFFER message and the RESCIND message should
1116 * not be handled by the same serialized work queue,
1117 * because the OFFER handler may call vmbus_open(),
1118 * which tries to open the channel by sending an
1119 * OPEN_CHANNEL message to the host and waits for
1120 * the host's response; however, if the host has
1121 * rescinded the channel before it receives the
1122 * OPEN_CHANNEL message, the host just silently
1123 * ignores the OPEN_CHANNEL message; as a result,
1124 * the guest's OFFER handler hangs for ever, if we
1125 * handle the RESCIND message in the same serialized
1126 * work queue: the RESCIND handler can not start to
1127 * run before the OFFER handler finishes.
1128 */
1129 schedule_work(&ctx->work);
1130 break;
1131
1132 case CHANNELMSG_OFFERCHANNEL:
1133 /*
1134 * The host sends the offer message of a given channel
1135 * before sending the rescind message of the same
1136 * channel. These messages are sent to the guest's
1137 * connect CPU; the guest then starts processing them
1138 * in the tasklet handler on this CPU:
1139 *
1140 * VMBUS_CONNECT_CPU
1141 *
1142 * [vmbus_on_msg_dpc()]
1143 * atomic_inc() // CHANNELMSG_OFFERCHANNEL
1144 * queue_work()
1145 * ...
1146 * [vmbus_on_msg_dpc()]
1147 * schedule_work() // CHANNELMSG_RESCIND_CHANNELOFFER
1148 *
1149 * We rely on the memory-ordering properties of the
1150 * queue_work() and schedule_work() primitives, which
1151 * guarantee that the atomic increment will be visible
1152 * to the CPUs which will execute the offer & rescind
1153 * works by the time these works will start execution.
1154 */
1155 atomic_inc(&vmbus_connection.offer_in_progress);
1156 fallthrough;
1157
1158 default:
1159 queue_work(vmbus_connection.work_queue, &ctx->work);
1160 }
1161 } else
1162 entry->message_handler(hdr);
1163
1164 msg_handled:
1165 vmbus_signal_eom(msg, message_type);
1166 }
1167
1168 #ifdef CONFIG_PM_SLEEP
1169 /*
1170 * Fake RESCIND_CHANNEL messages to clean up hv_sock channels by force for
1171 * hibernation, because hv_sock connections can not persist across hibernation.
1172 */
1173 static void vmbus_force_channel_rescinded(struct vmbus_channel *channel)
1174 {
1175 struct onmessage_work_context *ctx;
1176 struct vmbus_channel_rescind_offer *rescind;
1177
1178 WARN_ON(!is_hvsock_channel(channel));
1179
1180 /*
1181 * Allocation size is small and the allocation should really not fail,
1182 * otherwise the state of the hv_sock connections ends up in limbo.
1183 */
1184 ctx = kzalloc(sizeof(*ctx) + sizeof(*rescind),
1185 GFP_KERNEL | __GFP_NOFAIL);
1186
1187 /*
1188 * So far, these are not really used by Linux. Just set them to the
1189 * reasonable values conforming to the definitions of the fields.
1190 */
1191 ctx->msg.header.message_type = 1;
1192 ctx->msg.header.payload_size = sizeof(*rescind);
1193
1194 /* These values are actually used by Linux. */
1195 rescind = (struct vmbus_channel_rescind_offer *)ctx->msg.payload;
1196 rescind->header.msgtype = CHANNELMSG_RESCIND_CHANNELOFFER;
1197 rescind->child_relid = channel->offermsg.child_relid;
1198
1199 INIT_WORK(&ctx->work, vmbus_onmessage_work);
1200
1201 queue_work(vmbus_connection.work_queue, &ctx->work);
1202 }
1203 #endif /* CONFIG_PM_SLEEP */
1204
1205 /*
1206 * Schedule all channels with events pending
1207 */
1208 static void vmbus_chan_sched(struct hv_per_cpu_context *hv_cpu)
1209 {
1210 unsigned long *recv_int_page;
1211 u32 maxbits, relid;
1212
1213 if (vmbus_proto_version < VERSION_WIN8) {
1214 maxbits = MAX_NUM_CHANNELS_SUPPORTED;
1215 recv_int_page = vmbus_connection.recv_int_page;
1216 } else {
1217 /*
1218 * When the host is win8 and beyond, the event page
1219 * can be directly checked to get the id of the channel
1220 * that has the interrupt pending.
1221 */
1222 void *page_addr = hv_cpu->synic_event_page;
1223 union hv_synic_event_flags *event
1224 = (union hv_synic_event_flags *)page_addr +
1225 VMBUS_MESSAGE_SINT;
1226
1227 maxbits = HV_EVENT_FLAGS_COUNT;
1228 recv_int_page = event->flags;
1229 }
1230
1231 if (unlikely(!recv_int_page))
1232 return;
1233
1234 for_each_set_bit(relid, recv_int_page, maxbits) {
1235 void (*callback_fn)(void *context);
1236 struct vmbus_channel *channel;
1237
1238 if (!sync_test_and_clear_bit(relid, recv_int_page))
1239 continue;
1240
1241 /* Special case - vmbus channel protocol msg */
1242 if (relid == 0)
1243 continue;
1244
1245 /*
1246 * Pairs with the kfree_rcu() in vmbus_chan_release().
1247 * Guarantees that the channel data structure doesn't
1248 * get freed while the channel pointer below is being
1249 * dereferenced.
1250 */
1251 rcu_read_lock();
1252
1253 /* Find channel based on relid */
1254 channel = relid2channel(relid);
1255 if (channel == NULL)
1256 goto sched_unlock_rcu;
1257
1258 if (channel->rescind)
1259 goto sched_unlock_rcu;
1260
1261 /*
1262 * Make sure that the ring buffer data structure doesn't get
1263 * freed while we dereference the ring buffer pointer. Test
1264 * for the channel's onchannel_callback being NULL within a
1265 * sched_lock critical section. See also the inline comments
1266 * in vmbus_reset_channel_cb().
1267 */
1268 spin_lock(&channel->sched_lock);
1269
1270 callback_fn = channel->onchannel_callback;
1271 if (unlikely(callback_fn == NULL))
1272 goto sched_unlock;
1273
1274 trace_vmbus_chan_sched(channel);
1275
1276 ++channel->interrupts;
1277
1278 switch (channel->callback_mode) {
1279 case HV_CALL_ISR:
1280 (*callback_fn)(channel->channel_callback_context);
1281 break;
1282
1283 case HV_CALL_BATCHED:
1284 hv_begin_read(&channel->inbound);
1285 fallthrough;
1286 case HV_CALL_DIRECT:
1287 tasklet_schedule(&channel->callback_event);
1288 }
1289
1290 sched_unlock:
1291 spin_unlock(&channel->sched_lock);
1292 sched_unlock_rcu:
1293 rcu_read_unlock();
1294 }
1295 }
1296
1297 static void vmbus_isr(void)
1298 {
1299 struct hv_per_cpu_context *hv_cpu
1300 = this_cpu_ptr(hv_context.cpu_context);
1301 void *page_addr = hv_cpu->synic_event_page;
1302 struct hv_message *msg;
1303 union hv_synic_event_flags *event;
1304 bool handled = false;
1305
1306 if (unlikely(page_addr == NULL))
1307 return;
1308
1309 event = (union hv_synic_event_flags *)page_addr +
1310 VMBUS_MESSAGE_SINT;
1311 /*
1312 * Check for events before checking for messages. This is the order
1313 * in which events and messages are checked in Windows guests on
1314 * Hyper-V, and the Windows team suggested we do the same.
1315 */
1316
1317 if ((vmbus_proto_version == VERSION_WS2008) ||
1318 (vmbus_proto_version == VERSION_WIN7)) {
1319
1320 /* Since we are a child, we only need to check bit 0 */
1321 if (sync_test_and_clear_bit(0, event->flags))
1322 handled = true;
1323 } else {
1324 /*
1325 * Our host is win8 or above. The signaling mechanism
1326 * has changed and we can directly look at the event page.
1327 * If bit n is set then we have an interrup on the channel
1328 * whose id is n.
1329 */
1330 handled = true;
1331 }
1332
1333 if (handled)
1334 vmbus_chan_sched(hv_cpu);
1335
1336 page_addr = hv_cpu->synic_message_page;
1337 msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
1338
1339 /* Check if there are actual msgs to be processed */
1340 if (msg->header.message_type != HVMSG_NONE) {
1341 if (msg->header.message_type == HVMSG_TIMER_EXPIRED) {
1342 hv_stimer0_isr();
1343 vmbus_signal_eom(msg, HVMSG_TIMER_EXPIRED);
1344 } else
1345 tasklet_schedule(&hv_cpu->msg_dpc);
1346 }
1347
1348 add_interrupt_randomness(HYPERVISOR_CALLBACK_VECTOR, 0);
1349 }
1350
1351 /*
1352 * Callback from kmsg_dump. Grab as much as possible from the end of the kmsg
1353 * buffer and call into Hyper-V to transfer the data.
1354 */
1355 static void hv_kmsg_dump(struct kmsg_dumper *dumper,
1356 enum kmsg_dump_reason reason)
1357 {
1358 size_t bytes_written;
1359 phys_addr_t panic_pa;
1360
1361 /* We are only interested in panics. */
1362 if ((reason != KMSG_DUMP_PANIC) || (!sysctl_record_panic_msg))
1363 return;
1364
1365 panic_pa = virt_to_phys(hv_panic_page);
1366
1367 /*
1368 * Write dump contents to the page. No need to synchronize; panic should
1369 * be single-threaded.
1370 */
1371 kmsg_dump_get_buffer(dumper, true, hv_panic_page, HV_HYP_PAGE_SIZE,
1372 &bytes_written);
1373 if (bytes_written)
1374 hyperv_report_panic_msg(panic_pa, bytes_written);
1375 }
1376
1377 static struct kmsg_dumper hv_kmsg_dumper = {
1378 .dump = hv_kmsg_dump,
1379 };
1380
1381 static struct ctl_table_header *hv_ctl_table_hdr;
1382
1383 /*
1384 * sysctl option to allow the user to control whether kmsg data should be
1385 * reported to Hyper-V on panic.
1386 */
1387 static struct ctl_table hv_ctl_table[] = {
1388 {
1389 .procname = "hyperv_record_panic_msg",
1390 .data = &sysctl_record_panic_msg,
1391 .maxlen = sizeof(int),
1392 .mode = 0644,
1393 .proc_handler = proc_dointvec_minmax,
1394 .extra1 = SYSCTL_ZERO,
1395 .extra2 = SYSCTL_ONE
1396 },
1397 {}
1398 };
1399
1400 static struct ctl_table hv_root_table[] = {
1401 {
1402 .procname = "kernel",
1403 .mode = 0555,
1404 .child = hv_ctl_table
1405 },
1406 {}
1407 };
1408
1409 /*
1410 * vmbus_bus_init -Main vmbus driver initialization routine.
1411 *
1412 * Here, we
1413 * - initialize the vmbus driver context
1414 * - invoke the vmbus hv main init routine
1415 * - retrieve the channel offers
1416 */
1417 static int vmbus_bus_init(void)
1418 {
1419 int ret;
1420
1421 ret = hv_init();
1422 if (ret != 0) {
1423 pr_err("Unable to initialize the hypervisor - 0x%x\n", ret);
1424 return ret;
1425 }
1426
1427 ret = bus_register(&hv_bus);
1428 if (ret)
1429 return ret;
1430
1431 hv_setup_vmbus_irq(vmbus_isr);
1432
1433 ret = hv_synic_alloc();
1434 if (ret)
1435 goto err_alloc;
1436
1437 /*
1438 * Initialize the per-cpu interrupt state and stimer state.
1439 * Then connect to the host.
1440 */
1441 ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "hyperv/vmbus:online",
1442 hv_synic_init, hv_synic_cleanup);
1443 if (ret < 0)
1444 goto err_cpuhp;
1445 hyperv_cpuhp_online = ret;
1446
1447 ret = vmbus_connect();
1448 if (ret)
1449 goto err_connect;
1450
1451 /*
1452 * Only register if the crash MSRs are available
1453 */
1454 if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1455 u64 hyperv_crash_ctl;
1456 /*
1457 * Sysctl registration is not fatal, since by default
1458 * reporting is enabled.
1459 */
1460 hv_ctl_table_hdr = register_sysctl_table(hv_root_table);
1461 if (!hv_ctl_table_hdr)
1462 pr_err("Hyper-V: sysctl table register error");
1463
1464 /*
1465 * Register for panic kmsg callback only if the right
1466 * capability is supported by the hypervisor.
1467 */
1468 hv_get_crash_ctl(hyperv_crash_ctl);
1469 if (hyperv_crash_ctl & HV_CRASH_CTL_CRASH_NOTIFY_MSG) {
1470 hv_panic_page = (void *)hv_alloc_hyperv_zeroed_page();
1471 if (hv_panic_page) {
1472 ret = kmsg_dump_register(&hv_kmsg_dumper);
1473 if (ret) {
1474 pr_err("Hyper-V: kmsg dump register "
1475 "error 0x%x\n", ret);
1476 hv_free_hyperv_page(
1477 (unsigned long)hv_panic_page);
1478 hv_panic_page = NULL;
1479 }
1480 } else
1481 pr_err("Hyper-V: panic message page memory "
1482 "allocation failed");
1483 }
1484
1485 register_die_notifier(&hyperv_die_block);
1486 }
1487
1488 /*
1489 * Always register the panic notifier because we need to unload
1490 * the VMbus channel connection to prevent any VMbus
1491 * activity after the VM panics.
1492 */
1493 atomic_notifier_chain_register(&panic_notifier_list,
1494 &hyperv_panic_block);
1495
1496 vmbus_request_offers();
1497
1498 return 0;
1499
1500 err_connect:
1501 cpuhp_remove_state(hyperv_cpuhp_online);
1502 err_cpuhp:
1503 hv_synic_free();
1504 err_alloc:
1505 hv_remove_vmbus_irq();
1506
1507 bus_unregister(&hv_bus);
1508 unregister_sysctl_table(hv_ctl_table_hdr);
1509 hv_ctl_table_hdr = NULL;
1510 return ret;
1511 }
1512
1513 /**
1514 * __vmbus_child_driver_register() - Register a vmbus's driver
1515 * @hv_driver: Pointer to driver structure you want to register
1516 * @owner: owner module of the drv
1517 * @mod_name: module name string
1518 *
1519 * Registers the given driver with Linux through the 'driver_register()' call
1520 * and sets up the hyper-v vmbus handling for this driver.
1521 * It will return the state of the 'driver_register()' call.
1522 *
1523 */
1524 int __vmbus_driver_register(struct hv_driver *hv_driver, struct module *owner, const char *mod_name)
1525 {
1526 int ret;
1527
1528 pr_info("registering driver %s\n", hv_driver->name);
1529
1530 ret = vmbus_exists();
1531 if (ret < 0)
1532 return ret;
1533
1534 hv_driver->driver.name = hv_driver->name;
1535 hv_driver->driver.owner = owner;
1536 hv_driver->driver.mod_name = mod_name;
1537 hv_driver->driver.bus = &hv_bus;
1538
1539 spin_lock_init(&hv_driver->dynids.lock);
1540 INIT_LIST_HEAD(&hv_driver->dynids.list);
1541
1542 ret = driver_register(&hv_driver->driver);
1543
1544 return ret;
1545 }
1546 EXPORT_SYMBOL_GPL(__vmbus_driver_register);
1547
1548 /**
1549 * vmbus_driver_unregister() - Unregister a vmbus's driver
1550 * @hv_driver: Pointer to driver structure you want to
1551 * un-register
1552 *
1553 * Un-register the given driver that was previous registered with a call to
1554 * vmbus_driver_register()
1555 */
1556 void vmbus_driver_unregister(struct hv_driver *hv_driver)
1557 {
1558 pr_info("unregistering driver %s\n", hv_driver->name);
1559
1560 if (!vmbus_exists()) {
1561 driver_unregister(&hv_driver->driver);
1562 vmbus_free_dynids(hv_driver);
1563 }
1564 }
1565 EXPORT_SYMBOL_GPL(vmbus_driver_unregister);
1566
1567
1568 /*
1569 * Called when last reference to channel is gone.
1570 */
1571 static void vmbus_chan_release(struct kobject *kobj)
1572 {
1573 struct vmbus_channel *channel
1574 = container_of(kobj, struct vmbus_channel, kobj);
1575
1576 kfree_rcu(channel, rcu);
1577 }
1578
1579 struct vmbus_chan_attribute {
1580 struct attribute attr;
1581 ssize_t (*show)(struct vmbus_channel *chan, char *buf);
1582 ssize_t (*store)(struct vmbus_channel *chan,
1583 const char *buf, size_t count);
1584 };
1585 #define VMBUS_CHAN_ATTR(_name, _mode, _show, _store) \
1586 struct vmbus_chan_attribute chan_attr_##_name \
1587 = __ATTR(_name, _mode, _show, _store)
1588 #define VMBUS_CHAN_ATTR_RW(_name) \
1589 struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RW(_name)
1590 #define VMBUS_CHAN_ATTR_RO(_name) \
1591 struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RO(_name)
1592 #define VMBUS_CHAN_ATTR_WO(_name) \
1593 struct vmbus_chan_attribute chan_attr_##_name = __ATTR_WO(_name)
1594
1595 static ssize_t vmbus_chan_attr_show(struct kobject *kobj,
1596 struct attribute *attr, char *buf)
1597 {
1598 const struct vmbus_chan_attribute *attribute
1599 = container_of(attr, struct vmbus_chan_attribute, attr);
1600 struct vmbus_channel *chan
1601 = container_of(kobj, struct vmbus_channel, kobj);
1602
1603 if (!attribute->show)
1604 return -EIO;
1605
1606 return attribute->show(chan, buf);
1607 }
1608
1609 static ssize_t vmbus_chan_attr_store(struct kobject *kobj,
1610 struct attribute *attr, const char *buf,
1611 size_t count)
1612 {
1613 const struct vmbus_chan_attribute *attribute
1614 = container_of(attr, struct vmbus_chan_attribute, attr);
1615 struct vmbus_channel *chan
1616 = container_of(kobj, struct vmbus_channel, kobj);
1617
1618 if (!attribute->store)
1619 return -EIO;
1620
1621 return attribute->store(chan, buf, count);
1622 }
1623
1624 static const struct sysfs_ops vmbus_chan_sysfs_ops = {
1625 .show = vmbus_chan_attr_show,
1626 .store = vmbus_chan_attr_store,
1627 };
1628
1629 static ssize_t out_mask_show(struct vmbus_channel *channel, char *buf)
1630 {
1631 struct hv_ring_buffer_info *rbi = &channel->outbound;
1632 ssize_t ret;
1633
1634 mutex_lock(&rbi->ring_buffer_mutex);
1635 if (!rbi->ring_buffer) {
1636 mutex_unlock(&rbi->ring_buffer_mutex);
1637 return -EINVAL;
1638 }
1639
1640 ret = sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1641 mutex_unlock(&rbi->ring_buffer_mutex);
1642 return ret;
1643 }
1644 static VMBUS_CHAN_ATTR_RO(out_mask);
1645
1646 static ssize_t in_mask_show(struct vmbus_channel *channel, char *buf)
1647 {
1648 struct hv_ring_buffer_info *rbi = &channel->inbound;
1649 ssize_t ret;
1650
1651 mutex_lock(&rbi->ring_buffer_mutex);
1652 if (!rbi->ring_buffer) {
1653 mutex_unlock(&rbi->ring_buffer_mutex);
1654 return -EINVAL;
1655 }
1656
1657 ret = sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1658 mutex_unlock(&rbi->ring_buffer_mutex);
1659 return ret;
1660 }
1661 static VMBUS_CHAN_ATTR_RO(in_mask);
1662
1663 static ssize_t read_avail_show(struct vmbus_channel *channel, char *buf)
1664 {
1665 struct hv_ring_buffer_info *rbi = &channel->inbound;
1666 ssize_t ret;
1667
1668 mutex_lock(&rbi->ring_buffer_mutex);
1669 if (!rbi->ring_buffer) {
1670 mutex_unlock(&rbi->ring_buffer_mutex);
1671 return -EINVAL;
1672 }
1673
1674 ret = sprintf(buf, "%u\n", hv_get_bytes_to_read(rbi));
1675 mutex_unlock(&rbi->ring_buffer_mutex);
1676 return ret;
1677 }
1678 static VMBUS_CHAN_ATTR_RO(read_avail);
1679
1680 static ssize_t write_avail_show(struct vmbus_channel *channel, char *buf)
1681 {
1682 struct hv_ring_buffer_info *rbi = &channel->outbound;
1683 ssize_t ret;
1684
1685 mutex_lock(&rbi->ring_buffer_mutex);
1686 if (!rbi->ring_buffer) {
1687 mutex_unlock(&rbi->ring_buffer_mutex);
1688 return -EINVAL;
1689 }
1690
1691 ret = sprintf(buf, "%u\n", hv_get_bytes_to_write(rbi));
1692 mutex_unlock(&rbi->ring_buffer_mutex);
1693 return ret;
1694 }
1695 static VMBUS_CHAN_ATTR_RO(write_avail);
1696
1697 static ssize_t target_cpu_show(struct vmbus_channel *channel, char *buf)
1698 {
1699 return sprintf(buf, "%u\n", channel->target_cpu);
1700 }
1701 static ssize_t target_cpu_store(struct vmbus_channel *channel,
1702 const char *buf, size_t count)
1703 {
1704 u32 target_cpu, origin_cpu;
1705 ssize_t ret = count;
1706
1707 if (vmbus_proto_version < VERSION_WIN10_V4_1)
1708 return -EIO;
1709
1710 if (sscanf(buf, "%uu", &target_cpu) != 1)
1711 return -EIO;
1712
1713 /* Validate target_cpu for the cpumask_test_cpu() operation below. */
1714 if (target_cpu >= nr_cpumask_bits)
1715 return -EINVAL;
1716
1717 /* No CPUs should come up or down during this. */
1718 cpus_read_lock();
1719
1720 if (!cpumask_test_cpu(target_cpu, cpu_online_mask)) {
1721 cpus_read_unlock();
1722 return -EINVAL;
1723 }
1724
1725 /*
1726 * Synchronizes target_cpu_store() and channel closure:
1727 *
1728 * { Initially: state = CHANNEL_OPENED }
1729 *
1730 * CPU1 CPU2
1731 *
1732 * [target_cpu_store()] [vmbus_disconnect_ring()]
1733 *
1734 * LOCK channel_mutex LOCK channel_mutex
1735 * LOAD r1 = state LOAD r2 = state
1736 * IF (r1 == CHANNEL_OPENED) IF (r2 == CHANNEL_OPENED)
1737 * SEND MODIFYCHANNEL STORE state = CHANNEL_OPEN
1738 * [...] SEND CLOSECHANNEL
1739 * UNLOCK channel_mutex UNLOCK channel_mutex
1740 *
1741 * Forbids: r1 == r2 == CHANNEL_OPENED (i.e., CPU1's LOCK precedes
1742 * CPU2's LOCK) && CPU2's SEND precedes CPU1's SEND
1743 *
1744 * Note. The host processes the channel messages "sequentially", in
1745 * the order in which they are received on a per-partition basis.
1746 */
1747 mutex_lock(&vmbus_connection.channel_mutex);
1748
1749 /*
1750 * Hyper-V will ignore MODIFYCHANNEL messages for "non-open" channels;
1751 * avoid sending the message and fail here for such channels.
1752 */
1753 if (channel->state != CHANNEL_OPENED_STATE) {
1754 ret = -EIO;
1755 goto cpu_store_unlock;
1756 }
1757
1758 origin_cpu = channel->target_cpu;
1759 if (target_cpu == origin_cpu)
1760 goto cpu_store_unlock;
1761
1762 if (vmbus_send_modifychannel(channel->offermsg.child_relid,
1763 hv_cpu_number_to_vp_number(target_cpu))) {
1764 ret = -EIO;
1765 goto cpu_store_unlock;
1766 }
1767
1768 /*
1769 * Warning. At this point, there is *no* guarantee that the host will
1770 * have successfully processed the vmbus_send_modifychannel() request.
1771 * See the header comment of vmbus_send_modifychannel() for more info.
1772 *
1773 * Lags in the processing of the above vmbus_send_modifychannel() can
1774 * result in missed interrupts if the "old" target CPU is taken offline
1775 * before Hyper-V starts sending interrupts to the "new" target CPU.
1776 * But apart from this offlining scenario, the code tolerates such
1777 * lags. It will function correctly even if a channel interrupt comes
1778 * in on a CPU that is different from the channel target_cpu value.
1779 */
1780
1781 channel->target_cpu = target_cpu;
1782 channel->target_vp = hv_cpu_number_to_vp_number(target_cpu);
1783 channel->numa_node = cpu_to_node(target_cpu);
1784
1785 /* See init_vp_index(). */
1786 if (hv_is_perf_channel(channel))
1787 hv_update_alloced_cpus(origin_cpu, target_cpu);
1788
1789 /* Currently set only for storvsc channels. */
1790 if (channel->change_target_cpu_callback) {
1791 (*channel->change_target_cpu_callback)(channel,
1792 origin_cpu, target_cpu);
1793 }
1794
1795 cpu_store_unlock:
1796 mutex_unlock(&vmbus_connection.channel_mutex);
1797 cpus_read_unlock();
1798 return ret;
1799 }
1800 static VMBUS_CHAN_ATTR(cpu, 0644, target_cpu_show, target_cpu_store);
1801
1802 static ssize_t channel_pending_show(struct vmbus_channel *channel,
1803 char *buf)
1804 {
1805 return sprintf(buf, "%d\n",
1806 channel_pending(channel,
1807 vmbus_connection.monitor_pages[1]));
1808 }
1809 static VMBUS_CHAN_ATTR(pending, S_IRUGO, channel_pending_show, NULL);
1810
1811 static ssize_t channel_latency_show(struct vmbus_channel *channel,
1812 char *buf)
1813 {
1814 return sprintf(buf, "%d\n",
1815 channel_latency(channel,
1816 vmbus_connection.monitor_pages[1]));
1817 }
1818 static VMBUS_CHAN_ATTR(latency, S_IRUGO, channel_latency_show, NULL);
1819
1820 static ssize_t channel_interrupts_show(struct vmbus_channel *channel, char *buf)
1821 {
1822 return sprintf(buf, "%llu\n", channel->interrupts);
1823 }
1824 static VMBUS_CHAN_ATTR(interrupts, S_IRUGO, channel_interrupts_show, NULL);
1825
1826 static ssize_t channel_events_show(struct vmbus_channel *channel, char *buf)
1827 {
1828 return sprintf(buf, "%llu\n", channel->sig_events);
1829 }
1830 static VMBUS_CHAN_ATTR(events, S_IRUGO, channel_events_show, NULL);
1831
1832 static ssize_t channel_intr_in_full_show(struct vmbus_channel *channel,
1833 char *buf)
1834 {
1835 return sprintf(buf, "%llu\n",
1836 (unsigned long long)channel->intr_in_full);
1837 }
1838 static VMBUS_CHAN_ATTR(intr_in_full, 0444, channel_intr_in_full_show, NULL);
1839
1840 static ssize_t channel_intr_out_empty_show(struct vmbus_channel *channel,
1841 char *buf)
1842 {
1843 return sprintf(buf, "%llu\n",
1844 (unsigned long long)channel->intr_out_empty);
1845 }
1846 static VMBUS_CHAN_ATTR(intr_out_empty, 0444, channel_intr_out_empty_show, NULL);
1847
1848 static ssize_t channel_out_full_first_show(struct vmbus_channel *channel,
1849 char *buf)
1850 {
1851 return sprintf(buf, "%llu\n",
1852 (unsigned long long)channel->out_full_first);
1853 }
1854 static VMBUS_CHAN_ATTR(out_full_first, 0444, channel_out_full_first_show, NULL);
1855
1856 static ssize_t channel_out_full_total_show(struct vmbus_channel *channel,
1857 char *buf)
1858 {
1859 return sprintf(buf, "%llu\n",
1860 (unsigned long long)channel->out_full_total);
1861 }
1862 static VMBUS_CHAN_ATTR(out_full_total, 0444, channel_out_full_total_show, NULL);
1863
1864 static ssize_t subchannel_monitor_id_show(struct vmbus_channel *channel,
1865 char *buf)
1866 {
1867 return sprintf(buf, "%u\n", channel->offermsg.monitorid);
1868 }
1869 static VMBUS_CHAN_ATTR(monitor_id, S_IRUGO, subchannel_monitor_id_show, NULL);
1870
1871 static ssize_t subchannel_id_show(struct vmbus_channel *channel,
1872 char *buf)
1873 {
1874 return sprintf(buf, "%u\n",
1875 channel->offermsg.offer.sub_channel_index);
1876 }
1877 static VMBUS_CHAN_ATTR_RO(subchannel_id);
1878
1879 static struct attribute *vmbus_chan_attrs[] = {
1880 &chan_attr_out_mask.attr,
1881 &chan_attr_in_mask.attr,
1882 &chan_attr_read_avail.attr,
1883 &chan_attr_write_avail.attr,
1884 &chan_attr_cpu.attr,
1885 &chan_attr_pending.attr,
1886 &chan_attr_latency.attr,
1887 &chan_attr_interrupts.attr,
1888 &chan_attr_events.attr,
1889 &chan_attr_intr_in_full.attr,
1890 &chan_attr_intr_out_empty.attr,
1891 &chan_attr_out_full_first.attr,
1892 &chan_attr_out_full_total.attr,
1893 &chan_attr_monitor_id.attr,
1894 &chan_attr_subchannel_id.attr,
1895 NULL
1896 };
1897
1898 /*
1899 * Channel-level attribute_group callback function. Returns the permission for
1900 * each attribute, and returns 0 if an attribute is not visible.
1901 */
1902 static umode_t vmbus_chan_attr_is_visible(struct kobject *kobj,
1903 struct attribute *attr, int idx)
1904 {
1905 const struct vmbus_channel *channel =
1906 container_of(kobj, struct vmbus_channel, kobj);
1907
1908 /* Hide the monitor attributes if the monitor mechanism is not used. */
1909 if (!channel->offermsg.monitor_allocated &&
1910 (attr == &chan_attr_pending.attr ||
1911 attr == &chan_attr_latency.attr ||
1912 attr == &chan_attr_monitor_id.attr))
1913 return 0;
1914
1915 return attr->mode;
1916 }
1917
1918 static struct attribute_group vmbus_chan_group = {
1919 .attrs = vmbus_chan_attrs,
1920 .is_visible = vmbus_chan_attr_is_visible
1921 };
1922
1923 static struct kobj_type vmbus_chan_ktype = {
1924 .sysfs_ops = &vmbus_chan_sysfs_ops,
1925 .release = vmbus_chan_release,
1926 };
1927
1928 /*
1929 * vmbus_add_channel_kobj - setup a sub-directory under device/channels
1930 */
1931 int vmbus_add_channel_kobj(struct hv_device *dev, struct vmbus_channel *channel)
1932 {
1933 const struct device *device = &dev->device;
1934 struct kobject *kobj = &channel->kobj;
1935 u32 relid = channel->offermsg.child_relid;
1936 int ret;
1937
1938 kobj->kset = dev->channels_kset;
1939 ret = kobject_init_and_add(kobj, &vmbus_chan_ktype, NULL,
1940 "%u", relid);
1941 if (ret)
1942 return ret;
1943
1944 ret = sysfs_create_group(kobj, &vmbus_chan_group);
1945
1946 if (ret) {
1947 /*
1948 * The calling functions' error handling paths will cleanup the
1949 * empty channel directory.
1950 */
1951 dev_err(device, "Unable to set up channel sysfs files\n");
1952 return ret;
1953 }
1954
1955 kobject_uevent(kobj, KOBJ_ADD);
1956
1957 return 0;
1958 }
1959
1960 /*
1961 * vmbus_remove_channel_attr_group - remove the channel's attribute group
1962 */
1963 void vmbus_remove_channel_attr_group(struct vmbus_channel *channel)
1964 {
1965 sysfs_remove_group(&channel->kobj, &vmbus_chan_group);
1966 }
1967
1968 /*
1969 * vmbus_device_create - Creates and registers a new child device
1970 * on the vmbus.
1971 */
1972 struct hv_device *vmbus_device_create(const guid_t *type,
1973 const guid_t *instance,
1974 struct vmbus_channel *channel)
1975 {
1976 struct hv_device *child_device_obj;
1977
1978 child_device_obj = kzalloc(sizeof(struct hv_device), GFP_KERNEL);
1979 if (!child_device_obj) {
1980 pr_err("Unable to allocate device object for child device\n");
1981 return NULL;
1982 }
1983
1984 child_device_obj->channel = channel;
1985 guid_copy(&child_device_obj->dev_type, type);
1986 guid_copy(&child_device_obj->dev_instance, instance);
1987 child_device_obj->vendor_id = 0x1414; /* MSFT vendor ID */
1988
1989 return child_device_obj;
1990 }
1991
1992 /*
1993 * vmbus_device_register - Register the child device
1994 */
1995 int vmbus_device_register(struct hv_device *child_device_obj)
1996 {
1997 struct kobject *kobj = &child_device_obj->device.kobj;
1998 int ret;
1999
2000 dev_set_name(&child_device_obj->device, "%pUl",
2001 &child_device_obj->channel->offermsg.offer.if_instance);
2002
2003 child_device_obj->device.bus = &hv_bus;
2004 child_device_obj->device.parent = &hv_acpi_dev->dev;
2005 child_device_obj->device.release = vmbus_device_release;
2006
2007 /*
2008 * Register with the LDM. This will kick off the driver/device
2009 * binding...which will eventually call vmbus_match() and vmbus_probe()
2010 */
2011 ret = device_register(&child_device_obj->device);
2012 if (ret) {
2013 pr_err("Unable to register child device\n");
2014 return ret;
2015 }
2016
2017 child_device_obj->channels_kset = kset_create_and_add("channels",
2018 NULL, kobj);
2019 if (!child_device_obj->channels_kset) {
2020 ret = -ENOMEM;
2021 goto err_dev_unregister;
2022 }
2023
2024 ret = vmbus_add_channel_kobj(child_device_obj,
2025 child_device_obj->channel);
2026 if (ret) {
2027 pr_err("Unable to register primary channeln");
2028 goto err_kset_unregister;
2029 }
2030 hv_debug_add_dev_dir(child_device_obj);
2031
2032 return 0;
2033
2034 err_kset_unregister:
2035 kset_unregister(child_device_obj->channels_kset);
2036
2037 err_dev_unregister:
2038 device_unregister(&child_device_obj->device);
2039 return ret;
2040 }
2041
2042 /*
2043 * vmbus_device_unregister - Remove the specified child device
2044 * from the vmbus.
2045 */
2046 void vmbus_device_unregister(struct hv_device *device_obj)
2047 {
2048 pr_debug("child device %s unregistered\n",
2049 dev_name(&device_obj->device));
2050
2051 kset_unregister(device_obj->channels_kset);
2052
2053 /*
2054 * Kick off the process of unregistering the device.
2055 * This will call vmbus_remove() and eventually vmbus_device_release()
2056 */
2057 device_unregister(&device_obj->device);
2058 }
2059
2060
2061 /*
2062 * VMBUS is an acpi enumerated device. Get the information we
2063 * need from DSDT.
2064 */
2065 #define VTPM_BASE_ADDRESS 0xfed40000
2066 static acpi_status vmbus_walk_resources(struct acpi_resource *res, void *ctx)
2067 {
2068 resource_size_t start = 0;
2069 resource_size_t end = 0;
2070 struct resource *new_res;
2071 struct resource **old_res = &hyperv_mmio;
2072 struct resource **prev_res = NULL;
2073
2074 switch (res->type) {
2075
2076 /*
2077 * "Address" descriptors are for bus windows. Ignore
2078 * "memory" descriptors, which are for registers on
2079 * devices.
2080 */
2081 case ACPI_RESOURCE_TYPE_ADDRESS32:
2082 start = res->data.address32.address.minimum;
2083 end = res->data.address32.address.maximum;
2084 break;
2085
2086 case ACPI_RESOURCE_TYPE_ADDRESS64:
2087 start = res->data.address64.address.minimum;
2088 end = res->data.address64.address.maximum;
2089 break;
2090
2091 default:
2092 /* Unused resource type */
2093 return AE_OK;
2094
2095 }
2096 /*
2097 * Ignore ranges that are below 1MB, as they're not
2098 * necessary or useful here.
2099 */
2100 if (end < 0x100000)
2101 return AE_OK;
2102
2103 new_res = kzalloc(sizeof(*new_res), GFP_ATOMIC);
2104 if (!new_res)
2105 return AE_NO_MEMORY;
2106
2107 /* If this range overlaps the virtual TPM, truncate it. */
2108 if (end > VTPM_BASE_ADDRESS && start < VTPM_BASE_ADDRESS)
2109 end = VTPM_BASE_ADDRESS;
2110
2111 new_res->name = "hyperv mmio";
2112 new_res->flags = IORESOURCE_MEM;
2113 new_res->start = start;
2114 new_res->end = end;
2115
2116 /*
2117 * If two ranges are adjacent, merge them.
2118 */
2119 do {
2120 if (!*old_res) {
2121 *old_res = new_res;
2122 break;
2123 }
2124
2125 if (((*old_res)->end + 1) == new_res->start) {
2126 (*old_res)->end = new_res->end;
2127 kfree(new_res);
2128 break;
2129 }
2130
2131 if ((*old_res)->start == new_res->end + 1) {
2132 (*old_res)->start = new_res->start;
2133 kfree(new_res);
2134 break;
2135 }
2136
2137 if ((*old_res)->start > new_res->end) {
2138 new_res->sibling = *old_res;
2139 if (prev_res)
2140 (*prev_res)->sibling = new_res;
2141 *old_res = new_res;
2142 break;
2143 }
2144
2145 prev_res = old_res;
2146 old_res = &(*old_res)->sibling;
2147
2148 } while (1);
2149
2150 return AE_OK;
2151 }
2152
2153 static int vmbus_acpi_remove(struct acpi_device *device)
2154 {
2155 struct resource *cur_res;
2156 struct resource *next_res;
2157
2158 if (hyperv_mmio) {
2159 if (fb_mmio) {
2160 __release_region(hyperv_mmio, fb_mmio->start,
2161 resource_size(fb_mmio));
2162 fb_mmio = NULL;
2163 }
2164
2165 for (cur_res = hyperv_mmio; cur_res; cur_res = next_res) {
2166 next_res = cur_res->sibling;
2167 kfree(cur_res);
2168 }
2169 }
2170
2171 return 0;
2172 }
2173
2174 static void vmbus_reserve_fb(void)
2175 {
2176 int size;
2177 /*
2178 * Make a claim for the frame buffer in the resource tree under the
2179 * first node, which will be the one below 4GB. The length seems to
2180 * be underreported, particularly in a Generation 1 VM. So start out
2181 * reserving a larger area and make it smaller until it succeeds.
2182 */
2183
2184 if (screen_info.lfb_base) {
2185 if (efi_enabled(EFI_BOOT))
2186 size = max_t(__u32, screen_info.lfb_size, 0x800000);
2187 else
2188 size = max_t(__u32, screen_info.lfb_size, 0x4000000);
2189
2190 for (; !fb_mmio && (size >= 0x100000); size >>= 1) {
2191 fb_mmio = __request_region(hyperv_mmio,
2192 screen_info.lfb_base, size,
2193 fb_mmio_name, 0);
2194 }
2195 }
2196 }
2197
2198 /**
2199 * vmbus_allocate_mmio() - Pick a memory-mapped I/O range.
2200 * @new: If successful, supplied a pointer to the
2201 * allocated MMIO space.
2202 * @device_obj: Identifies the caller
2203 * @min: Minimum guest physical address of the
2204 * allocation
2205 * @max: Maximum guest physical address
2206 * @size: Size of the range to be allocated
2207 * @align: Alignment of the range to be allocated
2208 * @fb_overlap_ok: Whether this allocation can be allowed
2209 * to overlap the video frame buffer.
2210 *
2211 * This function walks the resources granted to VMBus by the
2212 * _CRS object in the ACPI namespace underneath the parent
2213 * "bridge" whether that's a root PCI bus in the Generation 1
2214 * case or a Module Device in the Generation 2 case. It then
2215 * attempts to allocate from the global MMIO pool in a way that
2216 * matches the constraints supplied in these parameters and by
2217 * that _CRS.
2218 *
2219 * Return: 0 on success, -errno on failure
2220 */
2221 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
2222 resource_size_t min, resource_size_t max,
2223 resource_size_t size, resource_size_t align,
2224 bool fb_overlap_ok)
2225 {
2226 struct resource *iter, *shadow;
2227 resource_size_t range_min, range_max, start;
2228 const char *dev_n = dev_name(&device_obj->device);
2229 int retval;
2230
2231 retval = -ENXIO;
2232 mutex_lock(&hyperv_mmio_lock);
2233
2234 /*
2235 * If overlaps with frame buffers are allowed, then first attempt to
2236 * make the allocation from within the reserved region. Because it
2237 * is already reserved, no shadow allocation is necessary.
2238 */
2239 if (fb_overlap_ok && fb_mmio && !(min > fb_mmio->end) &&
2240 !(max < fb_mmio->start)) {
2241
2242 range_min = fb_mmio->start;
2243 range_max = fb_mmio->end;
2244 start = (range_min + align - 1) & ~(align - 1);
2245 for (; start + size - 1 <= range_max; start += align) {
2246 *new = request_mem_region_exclusive(start, size, dev_n);
2247 if (*new) {
2248 retval = 0;
2249 goto exit;
2250 }
2251 }
2252 }
2253
2254 for (iter = hyperv_mmio; iter; iter = iter->sibling) {
2255 if ((iter->start >= max) || (iter->end <= min))
2256 continue;
2257
2258 range_min = iter->start;
2259 range_max = iter->end;
2260 start = (range_min + align - 1) & ~(align - 1);
2261 for (; start + size - 1 <= range_max; start += align) {
2262 shadow = __request_region(iter, start, size, NULL,
2263 IORESOURCE_BUSY);
2264 if (!shadow)
2265 continue;
2266
2267 *new = request_mem_region_exclusive(start, size, dev_n);
2268 if (*new) {
2269 shadow->name = (char *)*new;
2270 retval = 0;
2271 goto exit;
2272 }
2273
2274 __release_region(iter, start, size);
2275 }
2276 }
2277
2278 exit:
2279 mutex_unlock(&hyperv_mmio_lock);
2280 return retval;
2281 }
2282 EXPORT_SYMBOL_GPL(vmbus_allocate_mmio);
2283
2284 /**
2285 * vmbus_free_mmio() - Free a memory-mapped I/O range.
2286 * @start: Base address of region to release.
2287 * @size: Size of the range to be allocated
2288 *
2289 * This function releases anything requested by
2290 * vmbus_mmio_allocate().
2291 */
2292 void vmbus_free_mmio(resource_size_t start, resource_size_t size)
2293 {
2294 struct resource *iter;
2295
2296 mutex_lock(&hyperv_mmio_lock);
2297 for (iter = hyperv_mmio; iter; iter = iter->sibling) {
2298 if ((iter->start >= start + size) || (iter->end <= start))
2299 continue;
2300
2301 __release_region(iter, start, size);
2302 }
2303 release_mem_region(start, size);
2304 mutex_unlock(&hyperv_mmio_lock);
2305
2306 }
2307 EXPORT_SYMBOL_GPL(vmbus_free_mmio);
2308
2309 static int vmbus_acpi_add(struct acpi_device *device)
2310 {
2311 acpi_status result;
2312 int ret_val = -ENODEV;
2313 struct acpi_device *ancestor;
2314
2315 hv_acpi_dev = device;
2316
2317 result = acpi_walk_resources(device->handle, METHOD_NAME__CRS,
2318 vmbus_walk_resources, NULL);
2319
2320 if (ACPI_FAILURE(result))
2321 goto acpi_walk_err;
2322 /*
2323 * Some ancestor of the vmbus acpi device (Gen1 or Gen2
2324 * firmware) is the VMOD that has the mmio ranges. Get that.
2325 */
2326 for (ancestor = device->parent; ancestor; ancestor = ancestor->parent) {
2327 result = acpi_walk_resources(ancestor->handle, METHOD_NAME__CRS,
2328 vmbus_walk_resources, NULL);
2329
2330 if (ACPI_FAILURE(result))
2331 continue;
2332 if (hyperv_mmio) {
2333 vmbus_reserve_fb();
2334 break;
2335 }
2336 }
2337 ret_val = 0;
2338
2339 acpi_walk_err:
2340 complete(&probe_event);
2341 if (ret_val)
2342 vmbus_acpi_remove(device);
2343 return ret_val;
2344 }
2345
2346 #ifdef CONFIG_PM_SLEEP
2347 static int vmbus_bus_suspend(struct device *dev)
2348 {
2349 struct vmbus_channel *channel, *sc;
2350 unsigned long flags;
2351
2352 while (atomic_read(&vmbus_connection.offer_in_progress) != 0) {
2353 /*
2354 * We wait here until the completion of any channel
2355 * offers that are currently in progress.
2356 */
2357 msleep(1);
2358 }
2359
2360 mutex_lock(&vmbus_connection.channel_mutex);
2361 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
2362 if (!is_hvsock_channel(channel))
2363 continue;
2364
2365 vmbus_force_channel_rescinded(channel);
2366 }
2367 mutex_unlock(&vmbus_connection.channel_mutex);
2368
2369 /*
2370 * Wait until all the sub-channels and hv_sock channels have been
2371 * cleaned up. Sub-channels should be destroyed upon suspend, otherwise
2372 * they would conflict with the new sub-channels that will be created
2373 * in the resume path. hv_sock channels should also be destroyed, but
2374 * a hv_sock channel of an established hv_sock connection can not be
2375 * really destroyed since it may still be referenced by the userspace
2376 * application, so we just force the hv_sock channel to be rescinded
2377 * by vmbus_force_channel_rescinded(), and the userspace application
2378 * will thoroughly destroy the channel after hibernation.
2379 *
2380 * Note: the counter nr_chan_close_on_suspend may never go above 0 if
2381 * the VM has no sub-channel and hv_sock channel, e.g. a 1-vCPU VM.
2382 */
2383 if (atomic_read(&vmbus_connection.nr_chan_close_on_suspend) > 0)
2384 wait_for_completion(&vmbus_connection.ready_for_suspend_event);
2385
2386 WARN_ON(atomic_read(&vmbus_connection.nr_chan_fixup_on_resume) != 0);
2387
2388 mutex_lock(&vmbus_connection.channel_mutex);
2389
2390 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
2391 /*
2392 * Remove the channel from the array of channels and invalidate
2393 * the channel's relid. Upon resume, vmbus_onoffer() will fix
2394 * up the relid (and other fields, if necessary) and add the
2395 * channel back to the array.
2396 */
2397 vmbus_channel_unmap_relid(channel);
2398 channel->offermsg.child_relid = INVALID_RELID;
2399
2400 if (is_hvsock_channel(channel)) {
2401 if (!channel->rescind) {
2402 pr_err("hv_sock channel not rescinded!\n");
2403 WARN_ON_ONCE(1);
2404 }
2405 continue;
2406 }
2407
2408 spin_lock_irqsave(&channel->lock, flags);
2409 list_for_each_entry(sc, &channel->sc_list, sc_list) {
2410 pr_err("Sub-channel not deleted!\n");
2411 WARN_ON_ONCE(1);
2412 }
2413 spin_unlock_irqrestore(&channel->lock, flags);
2414
2415 atomic_inc(&vmbus_connection.nr_chan_fixup_on_resume);
2416 }
2417
2418 mutex_unlock(&vmbus_connection.channel_mutex);
2419
2420 vmbus_initiate_unload(false);
2421
2422 /* Reset the event for the next resume. */
2423 reinit_completion(&vmbus_connection.ready_for_resume_event);
2424
2425 return 0;
2426 }
2427
2428 static int vmbus_bus_resume(struct device *dev)
2429 {
2430 struct vmbus_channel_msginfo *msginfo;
2431 size_t msgsize;
2432 int ret;
2433
2434 /*
2435 * We only use the 'vmbus_proto_version', which was in use before
2436 * hibernation, to re-negotiate with the host.
2437 */
2438 if (!vmbus_proto_version) {
2439 pr_err("Invalid proto version = 0x%x\n", vmbus_proto_version);
2440 return -EINVAL;
2441 }
2442
2443 msgsize = sizeof(*msginfo) +
2444 sizeof(struct vmbus_channel_initiate_contact);
2445
2446 msginfo = kzalloc(msgsize, GFP_KERNEL);
2447
2448 if (msginfo == NULL)
2449 return -ENOMEM;
2450
2451 ret = vmbus_negotiate_version(msginfo, vmbus_proto_version);
2452
2453 kfree(msginfo);
2454
2455 if (ret != 0)
2456 return ret;
2457
2458 WARN_ON(atomic_read(&vmbus_connection.nr_chan_fixup_on_resume) == 0);
2459
2460 vmbus_request_offers();
2461
2462 wait_for_completion(&vmbus_connection.ready_for_resume_event);
2463
2464 /* Reset the event for the next suspend. */
2465 reinit_completion(&vmbus_connection.ready_for_suspend_event);
2466
2467 return 0;
2468 }
2469 #else
2470 #define vmbus_bus_suspend NULL
2471 #define vmbus_bus_resume NULL
2472 #endif /* CONFIG_PM_SLEEP */
2473
2474 static const struct acpi_device_id vmbus_acpi_device_ids[] = {
2475 {"VMBUS", 0},
2476 {"VMBus", 0},
2477 {"", 0},
2478 };
2479 MODULE_DEVICE_TABLE(acpi, vmbus_acpi_device_ids);
2480
2481 /*
2482 * Note: we must use the "no_irq" ops, otherwise hibernation can not work with
2483 * PCI device assignment, because "pci_dev_pm_ops" uses the "noirq" ops: in
2484 * the resume path, the pci "noirq" restore op runs before "non-noirq" op (see
2485 * resume_target_kernel() -> dpm_resume_start(), and hibernation_restore() ->
2486 * dpm_resume_end()). This means vmbus_bus_resume() and the pci-hyperv's
2487 * resume callback must also run via the "noirq" ops.
2488 *
2489 * Set suspend_noirq/resume_noirq to NULL for Suspend-to-Idle: see the comment
2490 * earlier in this file before vmbus_pm.
2491 */
2492
2493 static const struct dev_pm_ops vmbus_bus_pm = {
2494 .suspend_noirq = NULL,
2495 .resume_noirq = NULL,
2496 .freeze_noirq = vmbus_bus_suspend,
2497 .thaw_noirq = vmbus_bus_resume,
2498 .poweroff_noirq = vmbus_bus_suspend,
2499 .restore_noirq = vmbus_bus_resume
2500 };
2501
2502 static struct acpi_driver vmbus_acpi_driver = {
2503 .name = "vmbus",
2504 .ids = vmbus_acpi_device_ids,
2505 .ops = {
2506 .add = vmbus_acpi_add,
2507 .remove = vmbus_acpi_remove,
2508 },
2509 .drv.pm = &vmbus_bus_pm,
2510 };
2511
2512 static void hv_kexec_handler(void)
2513 {
2514 hv_stimer_global_cleanup();
2515 vmbus_initiate_unload(false);
2516 /* Make sure conn_state is set as hv_synic_cleanup checks for it */
2517 mb();
2518 cpuhp_remove_state(hyperv_cpuhp_online);
2519 hyperv_cleanup();
2520 };
2521
2522 static void hv_crash_handler(struct pt_regs *regs)
2523 {
2524 int cpu;
2525
2526 vmbus_initiate_unload(true);
2527 /*
2528 * In crash handler we can't schedule synic cleanup for all CPUs,
2529 * doing the cleanup for current CPU only. This should be sufficient
2530 * for kdump.
2531 */
2532 cpu = smp_processor_id();
2533 hv_stimer_cleanup(cpu);
2534 hv_synic_disable_regs(cpu);
2535 hyperv_cleanup();
2536 };
2537
2538 static int hv_synic_suspend(void)
2539 {
2540 /*
2541 * When we reach here, all the non-boot CPUs have been offlined.
2542 * If we're in a legacy configuration where stimer Direct Mode is
2543 * not enabled, the stimers on the non-boot CPUs have been unbound
2544 * in hv_synic_cleanup() -> hv_stimer_legacy_cleanup() ->
2545 * hv_stimer_cleanup() -> clockevents_unbind_device().
2546 *
2547 * hv_synic_suspend() only runs on CPU0 with interrupts disabled.
2548 * Here we do not call hv_stimer_legacy_cleanup() on CPU0 because:
2549 * 1) it's unnecessary as interrupts remain disabled between
2550 * syscore_suspend() and syscore_resume(): see create_image() and
2551 * resume_target_kernel()
2552 * 2) the stimer on CPU0 is automatically disabled later by
2553 * syscore_suspend() -> timekeeping_suspend() -> tick_suspend() -> ...
2554 * -> clockevents_shutdown() -> ... -> hv_ce_shutdown()
2555 * 3) a warning would be triggered if we call
2556 * clockevents_unbind_device(), which may sleep, in an
2557 * interrupts-disabled context.
2558 */
2559
2560 hv_synic_disable_regs(0);
2561
2562 return 0;
2563 }
2564
2565 static void hv_synic_resume(void)
2566 {
2567 hv_synic_enable_regs(0);
2568
2569 /*
2570 * Note: we don't need to call hv_stimer_init(0), because the timer
2571 * on CPU0 is not unbound in hv_synic_suspend(), and the timer is
2572 * automatically re-enabled in timekeeping_resume().
2573 */
2574 }
2575
2576 /* The callbacks run only on CPU0, with irqs_disabled. */
2577 static struct syscore_ops hv_synic_syscore_ops = {
2578 .suspend = hv_synic_suspend,
2579 .resume = hv_synic_resume,
2580 };
2581
2582 static int __init hv_acpi_init(void)
2583 {
2584 int ret, t;
2585
2586 if (!hv_is_hyperv_initialized())
2587 return -ENODEV;
2588
2589 init_completion(&probe_event);
2590
2591 /*
2592 * Get ACPI resources first.
2593 */
2594 ret = acpi_bus_register_driver(&vmbus_acpi_driver);
2595
2596 if (ret)
2597 return ret;
2598
2599 t = wait_for_completion_timeout(&probe_event, 5*HZ);
2600 if (t == 0) {
2601 ret = -ETIMEDOUT;
2602 goto cleanup;
2603 }
2604 hv_debug_init();
2605
2606 ret = vmbus_bus_init();
2607 if (ret)
2608 goto cleanup;
2609
2610 hv_setup_kexec_handler(hv_kexec_handler);
2611 hv_setup_crash_handler(hv_crash_handler);
2612
2613 register_syscore_ops(&hv_synic_syscore_ops);
2614
2615 return 0;
2616
2617 cleanup:
2618 acpi_bus_unregister_driver(&vmbus_acpi_driver);
2619 hv_acpi_dev = NULL;
2620 return ret;
2621 }
2622
2623 static void __exit vmbus_exit(void)
2624 {
2625 int cpu;
2626
2627 unregister_syscore_ops(&hv_synic_syscore_ops);
2628
2629 hv_remove_kexec_handler();
2630 hv_remove_crash_handler();
2631 vmbus_connection.conn_state = DISCONNECTED;
2632 hv_stimer_global_cleanup();
2633 vmbus_disconnect();
2634 hv_remove_vmbus_irq();
2635 for_each_online_cpu(cpu) {
2636 struct hv_per_cpu_context *hv_cpu
2637 = per_cpu_ptr(hv_context.cpu_context, cpu);
2638
2639 tasklet_kill(&hv_cpu->msg_dpc);
2640 }
2641 hv_debug_rm_all_dir();
2642
2643 vmbus_free_channels();
2644 kfree(vmbus_connection.channels);
2645
2646 if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
2647 kmsg_dump_unregister(&hv_kmsg_dumper);
2648 unregister_die_notifier(&hyperv_die_block);
2649 atomic_notifier_chain_unregister(&panic_notifier_list,
2650 &hyperv_panic_block);
2651 }
2652
2653 free_page((unsigned long)hv_panic_page);
2654 unregister_sysctl_table(hv_ctl_table_hdr);
2655 hv_ctl_table_hdr = NULL;
2656 bus_unregister(&hv_bus);
2657
2658 cpuhp_remove_state(hyperv_cpuhp_online);
2659 hv_synic_free();
2660 acpi_bus_unregister_driver(&vmbus_acpi_driver);
2661 }
2662
2663
2664 MODULE_LICENSE("GPL");
2665 MODULE_DESCRIPTION("Microsoft Hyper-V VMBus Driver");
2666
2667 subsys_initcall(hv_acpi_init);
2668 module_exit(vmbus_exit);
Ubuntu Hirsute Linux kernel mirror