1/*
2 * Copyright (c) 2009, Microsoft Corporation.
3 *
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
11 * more details.
12 *
13 * You should have received a copy of the GNU General Public License along with
14 * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
15 * Place - Suite 330, Boston, MA 02111-1307 USA.
16 *
17 * Authors:
18 * Haiyang Zhang <haiyangz@microsoft.com>
19 * Hank Janssen <hjanssen@microsoft.com>
20 * K. Y. Srinivasan <kys@microsoft.com>
21 *
22 */
23#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
24
25#include <linux/init.h>
26#include <linux/module.h>
27#include <linux/device.h>
28#include <linux/interrupt.h>
29#include <linux/sysctl.h>
30#include <linux/slab.h>
31#include <linux/acpi.h>
32#include <linux/completion.h>
33#include <linux/hyperv.h>
34#include <linux/kernel_stat.h>
35#include <linux/clockchips.h>
36#include <linux/cpu.h>
37#include <linux/sched/task_stack.h>
38
39#include <asm/mshyperv.h>
40#include <linux/notifier.h>
41#include <linux/ptrace.h>
42#include <linux/screen_info.h>
43#include <linux/kdebug.h>
44#include <linux/efi.h>
45#include <linux/random.h>
46#include "hyperv_vmbus.h"
47
48struct vmbus_dynid {
49 struct list_head node;
50 struct hv_vmbus_device_id id;
51};
52
53static struct acpi_device *hv_acpi_dev;
54
55static struct completion probe_event;
56
57static int hyperv_cpuhp_online;
58
59static void *hv_panic_page;
60
61static int hyperv_panic_event(struct notifier_block *nb, unsigned long val,
62 void *args)
63{
64 struct pt_regs *regs;
65
66 regs = current_pt_regs();
67
68 hyperv_report_panic(regs, val);
69 return NOTIFY_DONE;
70}
71
72static int hyperv_die_event(struct notifier_block *nb, unsigned long val,
73 void *args)
74{
75 struct die_args *die = (struct die_args *)args;
76 struct pt_regs *regs = die->regs;
77
78 hyperv_report_panic(regs, val);
79 return NOTIFY_DONE;
80}
81
82static struct notifier_block hyperv_die_block = {
83 .notifier_call = hyperv_die_event,
84};
85static struct notifier_block hyperv_panic_block = {
86 .notifier_call = hyperv_panic_event,
87};
88
89static const char *fb_mmio_name = "fb_range";
90static struct resource *fb_mmio;
91static struct resource *hyperv_mmio;
92static DEFINE_SEMAPHORE(hyperv_mmio_lock);
93
94static int vmbus_exists(void)
95{
96 if (hv_acpi_dev == NULL)
97 return -ENODEV;
98
99 return 0;
100}
101
102#define VMBUS_ALIAS_LEN ((sizeof((struct hv_vmbus_device_id *)0)->guid) * 2)
103static void print_alias_name(struct hv_device *hv_dev, char *alias_name)
104{
105 int i;
106 for (i = 0; i < VMBUS_ALIAS_LEN; i += 2)
107 sprintf(&alias_name[i], "%02x", hv_dev->dev_type.b[i/2]);
108}
109
110static u8 channel_monitor_group(const struct vmbus_channel *channel)
111{
112 return (u8)channel->offermsg.monitorid / 32;
113}
114
115static u8 channel_monitor_offset(const struct vmbus_channel *channel)
116{
117 return (u8)channel->offermsg.monitorid % 32;
118}
119
120static u32 channel_pending(const struct vmbus_channel *channel,
121 const struct hv_monitor_page *monitor_page)
122{
123 u8 monitor_group = channel_monitor_group(channel);
124
125 return monitor_page->trigger_group[monitor_group].pending;
126}
127
128static u32 channel_latency(const struct vmbus_channel *channel,
129 const struct hv_monitor_page *monitor_page)
130{
131 u8 monitor_group = channel_monitor_group(channel);
132 u8 monitor_offset = channel_monitor_offset(channel);
133
134 return monitor_page->latency[monitor_group][monitor_offset];
135}
136
137static u32 channel_conn_id(struct vmbus_channel *channel,
138 struct hv_monitor_page *monitor_page)
139{
140 u8 monitor_group = channel_monitor_group(channel);
141 u8 monitor_offset = channel_monitor_offset(channel);
142 return monitor_page->parameter[monitor_group][monitor_offset].connectionid.u.id;
143}
144
145static ssize_t id_show(struct device *dev, struct device_attribute *dev_attr,
146 char *buf)
147{
148 struct hv_device *hv_dev = device_to_hv_device(dev);
149
150 if (!hv_dev->channel)
151 return -ENODEV;
152 return sprintf(buf, "%d\n", hv_dev->channel->offermsg.child_relid);
153}
154static DEVICE_ATTR_RO(id);
155
156static ssize_t state_show(struct device *dev, struct device_attribute *dev_attr,
157 char *buf)
158{
159 struct hv_device *hv_dev = device_to_hv_device(dev);
160
161 if (!hv_dev->channel)
162 return -ENODEV;
163 return sprintf(buf, "%d\n", hv_dev->channel->state);
164}
165static DEVICE_ATTR_RO(state);
166
167static ssize_t monitor_id_show(struct device *dev,
168 struct device_attribute *dev_attr, char *buf)
169{
170 struct hv_device *hv_dev = device_to_hv_device(dev);
171
172 if (!hv_dev->channel)
173 return -ENODEV;
174 return sprintf(buf, "%d\n", hv_dev->channel->offermsg.monitorid);
175}
176static DEVICE_ATTR_RO(monitor_id);
177
178static ssize_t class_id_show(struct device *dev,
179 struct device_attribute *dev_attr, char *buf)
180{
181 struct hv_device *hv_dev = device_to_hv_device(dev);
182
183 if (!hv_dev->channel)
184 return -ENODEV;
185 return sprintf(buf, "{%pUl}\n",
186 hv_dev->channel->offermsg.offer.if_type.b);
187}
188static DEVICE_ATTR_RO(class_id);
189
190static ssize_t device_id_show(struct device *dev,
191 struct device_attribute *dev_attr, char *buf)
192{
193 struct hv_device *hv_dev = device_to_hv_device(dev);
194
195 if (!hv_dev->channel)
196 return -ENODEV;
197 return sprintf(buf, "{%pUl}\n",
198 hv_dev->channel->offermsg.offer.if_instance.b);
199}
200static DEVICE_ATTR_RO(device_id);
201
202static ssize_t modalias_show(struct device *dev,
203 struct device_attribute *dev_attr, char *buf)
204{
205 struct hv_device *hv_dev = device_to_hv_device(dev);
206 char alias_name[VMBUS_ALIAS_LEN + 1];
207
208 print_alias_name(hv_dev, alias_name);
209 return sprintf(buf, "vmbus:%s\n", alias_name);
210}
211static DEVICE_ATTR_RO(modalias);
212
213#ifdef CONFIG_NUMA
214static ssize_t numa_node_show(struct device *dev,
215 struct device_attribute *attr, char *buf)
216{
217 struct hv_device *hv_dev = device_to_hv_device(dev);
218
219 if (!hv_dev->channel)
220 return -ENODEV;
221
222 return sprintf(buf, "%d\n", hv_dev->channel->numa_node);
223}
224static DEVICE_ATTR_RO(numa_node);
225#endif
226
227static ssize_t server_monitor_pending_show(struct device *dev,
228 struct device_attribute *dev_attr,
229 char *buf)
230{
231 struct hv_device *hv_dev = device_to_hv_device(dev);
232
233 if (!hv_dev->channel)
234 return -ENODEV;
235 return sprintf(buf, "%d\n",
236 channel_pending(hv_dev->channel,
237 vmbus_connection.monitor_pages[0]));
238}
239static DEVICE_ATTR_RO(server_monitor_pending);
240
241static ssize_t client_monitor_pending_show(struct device *dev,
242 struct device_attribute *dev_attr,
243 char *buf)
244{
245 struct hv_device *hv_dev = device_to_hv_device(dev);
246
247 if (!hv_dev->channel)
248 return -ENODEV;
249 return sprintf(buf, "%d\n",
250 channel_pending(hv_dev->channel,
251 vmbus_connection.monitor_pages[1]));
252}
253static DEVICE_ATTR_RO(client_monitor_pending);
254
255static ssize_t server_monitor_latency_show(struct device *dev,
256 struct device_attribute *dev_attr,
257 char *buf)
258{
259 struct hv_device *hv_dev = device_to_hv_device(dev);
260
261 if (!hv_dev->channel)
262 return -ENODEV;
263 return sprintf(buf, "%d\n",
264 channel_latency(hv_dev->channel,
265 vmbus_connection.monitor_pages[0]));
266}
267static DEVICE_ATTR_RO(server_monitor_latency);
268
269static ssize_t client_monitor_latency_show(struct device *dev,
270 struct device_attribute *dev_attr,
271 char *buf)
272{
273 struct hv_device *hv_dev = device_to_hv_device(dev);
274
275 if (!hv_dev->channel)
276 return -ENODEV;
277 return sprintf(buf, "%d\n",
278 channel_latency(hv_dev->channel,
279 vmbus_connection.monitor_pages[1]));
280}
281static DEVICE_ATTR_RO(client_monitor_latency);
282
283static ssize_t server_monitor_conn_id_show(struct device *dev,
284 struct device_attribute *dev_attr,
285 char *buf)
286{
287 struct hv_device *hv_dev = device_to_hv_device(dev);
288
289 if (!hv_dev->channel)
290 return -ENODEV;
291 return sprintf(buf, "%d\n",
292 channel_conn_id(hv_dev->channel,
293 vmbus_connection.monitor_pages[0]));
294}
295static DEVICE_ATTR_RO(server_monitor_conn_id);
296
297static ssize_t client_monitor_conn_id_show(struct device *dev,
298 struct device_attribute *dev_attr,
299 char *buf)
300{
301 struct hv_device *hv_dev = device_to_hv_device(dev);
302
303 if (!hv_dev->channel)
304 return -ENODEV;
305 return sprintf(buf, "%d\n",
306 channel_conn_id(hv_dev->channel,
307 vmbus_connection.monitor_pages[1]));
308}
309static DEVICE_ATTR_RO(client_monitor_conn_id);
310
311static ssize_t out_intr_mask_show(struct device *dev,
312 struct device_attribute *dev_attr, char *buf)
313{
314 struct hv_device *hv_dev = device_to_hv_device(dev);
315 struct hv_ring_buffer_debug_info outbound;
316 int ret;
317
318 if (!hv_dev->channel)
319 return -ENODEV;
320
321 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
322 &outbound);
323 if (ret < 0)
324 return ret;
325
326 return sprintf(buf, "%d\n", outbound.current_interrupt_mask);
327}
328static DEVICE_ATTR_RO(out_intr_mask);
329
330static ssize_t out_read_index_show(struct device *dev,
331 struct device_attribute *dev_attr, char *buf)
332{
333 struct hv_device *hv_dev = device_to_hv_device(dev);
334 struct hv_ring_buffer_debug_info outbound;
335 int ret;
336
337 if (!hv_dev->channel)
338 return -ENODEV;
339
340 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
341 &outbound);
342 if (ret < 0)
343 return ret;
344 return sprintf(buf, "%d\n", outbound.current_read_index);
345}
346static DEVICE_ATTR_RO(out_read_index);
347
348static ssize_t out_write_index_show(struct device *dev,
349 struct device_attribute *dev_attr,
350 char *buf)
351{
352 struct hv_device *hv_dev = device_to_hv_device(dev);
353 struct hv_ring_buffer_debug_info outbound;
354 int ret;
355
356 if (!hv_dev->channel)
357 return -ENODEV;
358
359 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
360 &outbound);
361 if (ret < 0)
362 return ret;
363 return sprintf(buf, "%d\n", outbound.current_write_index);
364}
365static DEVICE_ATTR_RO(out_write_index);
366
367static ssize_t out_read_bytes_avail_show(struct device *dev,
368 struct device_attribute *dev_attr,
369 char *buf)
370{
371 struct hv_device *hv_dev = device_to_hv_device(dev);
372 struct hv_ring_buffer_debug_info outbound;
373 int ret;
374
375 if (!hv_dev->channel)
376 return -ENODEV;
377
378 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
379 &outbound);
380 if (ret < 0)
381 return ret;
382 return sprintf(buf, "%d\n", outbound.bytes_avail_toread);
383}
384static DEVICE_ATTR_RO(out_read_bytes_avail);
385
386static ssize_t out_write_bytes_avail_show(struct device *dev,
387 struct device_attribute *dev_attr,
388 char *buf)
389{
390 struct hv_device *hv_dev = device_to_hv_device(dev);
391 struct hv_ring_buffer_debug_info outbound;
392 int ret;
393
394 if (!hv_dev->channel)
395 return -ENODEV;
396
397 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
398 &outbound);
399 if (ret < 0)
400 return ret;
401 return sprintf(buf, "%d\n", outbound.bytes_avail_towrite);
402}
403static DEVICE_ATTR_RO(out_write_bytes_avail);
404
405static ssize_t in_intr_mask_show(struct device *dev,
406 struct device_attribute *dev_attr, char *buf)
407{
408 struct hv_device *hv_dev = device_to_hv_device(dev);
409 struct hv_ring_buffer_debug_info inbound;
410 int ret;
411
412 if (!hv_dev->channel)
413 return -ENODEV;
414
415 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
416 if (ret < 0)
417 return ret;
418
419 return sprintf(buf, "%d\n", inbound.current_interrupt_mask);
420}
421static DEVICE_ATTR_RO(in_intr_mask);
422
423static ssize_t in_read_index_show(struct device *dev,
424 struct device_attribute *dev_attr, char *buf)
425{
426 struct hv_device *hv_dev = device_to_hv_device(dev);
427 struct hv_ring_buffer_debug_info inbound;
428 int ret;
429
430 if (!hv_dev->channel)
431 return -ENODEV;
432
433 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
434 if (ret < 0)
435 return ret;
436
437 return sprintf(buf, "%d\n", inbound.current_read_index);
438}
439static DEVICE_ATTR_RO(in_read_index);
440
441static ssize_t in_write_index_show(struct device *dev,
442 struct device_attribute *dev_attr, char *buf)
443{
444 struct hv_device *hv_dev = device_to_hv_device(dev);
445 struct hv_ring_buffer_debug_info inbound;
446 int ret;
447
448 if (!hv_dev->channel)
449 return -ENODEV;
450
451 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
452 if (ret < 0)
453 return ret;
454
455 return sprintf(buf, "%d\n", inbound.current_write_index);
456}
457static DEVICE_ATTR_RO(in_write_index);
458
459static ssize_t in_read_bytes_avail_show(struct device *dev,
460 struct device_attribute *dev_attr,
461 char *buf)
462{
463 struct hv_device *hv_dev = device_to_hv_device(dev);
464 struct hv_ring_buffer_debug_info inbound;
465 int ret;
466
467 if (!hv_dev->channel)
468 return -ENODEV;
469
470 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
471 if (ret < 0)
472 return ret;
473
474 return sprintf(buf, "%d\n", inbound.bytes_avail_toread);
475}
476static DEVICE_ATTR_RO(in_read_bytes_avail);
477
478static ssize_t in_write_bytes_avail_show(struct device *dev,
479 struct device_attribute *dev_attr,
480 char *buf)
481{
482 struct hv_device *hv_dev = device_to_hv_device(dev);
483 struct hv_ring_buffer_debug_info inbound;
484 int ret;
485
486 if (!hv_dev->channel)
487 return -ENODEV;
488
489 ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
490 if (ret < 0)
491 return ret;
492
493 return sprintf(buf, "%d\n", inbound.bytes_avail_towrite);
494}
495static DEVICE_ATTR_RO(in_write_bytes_avail);
496
497static ssize_t channel_vp_mapping_show(struct device *dev,
498 struct device_attribute *dev_attr,
499 char *buf)
500{
501 struct hv_device *hv_dev = device_to_hv_device(dev);
502 struct vmbus_channel *channel = hv_dev->channel, *cur_sc;
503 unsigned long flags;
504 int buf_size = PAGE_SIZE, n_written, tot_written;
505 struct list_head *cur;
506
507 if (!channel)
508 return -ENODEV;
509
510 tot_written = snprintf(buf, buf_size, "%u:%u\n",
511 channel->offermsg.child_relid, channel->target_cpu);
512
513 spin_lock_irqsave(&channel->lock, flags);
514
515 list_for_each(cur, &channel->sc_list) {
516 if (tot_written >= buf_size - 1)
517 break;
518
519 cur_sc = list_entry(cur, struct vmbus_channel, sc_list);
520 n_written = scnprintf(buf + tot_written,
521 buf_size - tot_written,
522 "%u:%u\n",
523 cur_sc->offermsg.child_relid,
524 cur_sc->target_cpu);
525 tot_written += n_written;
526 }
527
528 spin_unlock_irqrestore(&channel->lock, flags);
529
530 return tot_written;
531}
532static DEVICE_ATTR_RO(channel_vp_mapping);
533
534static ssize_t vendor_show(struct device *dev,
535 struct device_attribute *dev_attr,
536 char *buf)
537{
538 struct hv_device *hv_dev = device_to_hv_device(dev);
539 return sprintf(buf, "0x%x\n", hv_dev->vendor_id);
540}
541static DEVICE_ATTR_RO(vendor);
542
543static ssize_t device_show(struct device *dev,
544 struct device_attribute *dev_attr,
545 char *buf)
546{
547 struct hv_device *hv_dev = device_to_hv_device(dev);
548 return sprintf(buf, "0x%x\n", hv_dev->device_id);
549}
550static DEVICE_ATTR_RO(device);
551
552static ssize_t driver_override_store(struct device *dev,
553 struct device_attribute *attr,
554 const char *buf, size_t count)
555{
556 struct hv_device *hv_dev = device_to_hv_device(dev);
557 char *driver_override, *old, *cp;
558
559 /* We need to keep extra room for a newline */
560 if (count >= (PAGE_SIZE - 1))
561 return -EINVAL;
562
563 driver_override = kstrndup(buf, count, GFP_KERNEL);
564 if (!driver_override)
565 return -ENOMEM;
566
567 cp = strchr(driver_override, '\n');
568 if (cp)
569 *cp = '\0';
570
571 device_lock(dev);
572 old = hv_dev->driver_override;
573 if (strlen(driver_override)) {
574 hv_dev->driver_override = driver_override;
575 } else {
576 kfree(driver_override);
577 hv_dev->driver_override = NULL;
578 }
579 device_unlock(dev);
580
581 kfree(old);
582
583 return count;
584}
585
586static ssize_t driver_override_show(struct device *dev,
587 struct device_attribute *attr, char *buf)
588{
589 struct hv_device *hv_dev = device_to_hv_device(dev);
590 ssize_t len;
591
592 device_lock(dev);
593 len = snprintf(buf, PAGE_SIZE, "%s\n", hv_dev->driver_override);
594 device_unlock(dev);
595
596 return len;
597}
598static DEVICE_ATTR_RW(driver_override);
599
600/* Set up per device attributes in /sys/bus/vmbus/devices/<bus device> */
601static struct attribute *vmbus_dev_attrs[] = {
602 &dev_attr_id.attr,
603 &dev_attr_state.attr,
604 &dev_attr_monitor_id.attr,
605 &dev_attr_class_id.attr,
606 &dev_attr_device_id.attr,
607 &dev_attr_modalias.attr,
608#ifdef CONFIG_NUMA
609 &dev_attr_numa_node.attr,
610#endif
611 &dev_attr_server_monitor_pending.attr,
612 &dev_attr_client_monitor_pending.attr,
613 &dev_attr_server_monitor_latency.attr,
614 &dev_attr_client_monitor_latency.attr,
615 &dev_attr_server_monitor_conn_id.attr,
616 &dev_attr_client_monitor_conn_id.attr,
617 &dev_attr_out_intr_mask.attr,
618 &dev_attr_out_read_index.attr,
619 &dev_attr_out_write_index.attr,
620 &dev_attr_out_read_bytes_avail.attr,
621 &dev_attr_out_write_bytes_avail.attr,
622 &dev_attr_in_intr_mask.attr,
623 &dev_attr_in_read_index.attr,
624 &dev_attr_in_write_index.attr,
625 &dev_attr_in_read_bytes_avail.attr,
626 &dev_attr_in_write_bytes_avail.attr,
627 &dev_attr_channel_vp_mapping.attr,
628 &dev_attr_vendor.attr,
629 &dev_attr_device.attr,
630 &dev_attr_driver_override.attr,
631 NULL,
632};
633ATTRIBUTE_GROUPS(vmbus_dev);
634
635/*
636 * vmbus_uevent - add uevent for our device
637 *
638 * This routine is invoked when a device is added or removed on the vmbus to
639 * generate a uevent to udev in the userspace. The udev will then look at its
640 * rule and the uevent generated here to load the appropriate driver
641 *
642 * The alias string will be of the form vmbus:guid where guid is the string
643 * representation of the device guid (each byte of the guid will be
644 * represented with two hex characters.
645 */
646static int vmbus_uevent(struct device *device, struct kobj_uevent_env *env)
647{
648 struct hv_device *dev = device_to_hv_device(device);
649 int ret;
650 char alias_name[VMBUS_ALIAS_LEN + 1];
651
652 print_alias_name(dev, alias_name);
653 ret = add_uevent_var(env, "MODALIAS=vmbus:%s", alias_name);
654 return ret;
655}
656
657static const struct hv_vmbus_device_id *
658hv_vmbus_dev_match(const struct hv_vmbus_device_id *id, const guid_t *guid)
659{
660 if (id == NULL)
661 return NULL; /* empty device table */
662
663 for (; !guid_is_null(&id->guid); id++)
664 if (guid_equal(&id->guid, guid))
665 return id;
666
667 return NULL;
668}
669
670static const struct hv_vmbus_device_id *
671hv_vmbus_dynid_match(struct hv_driver *drv, const guid_t *guid)
672{
673 const struct hv_vmbus_device_id *id = NULL;
674 struct vmbus_dynid *dynid;
675
676 spin_lock(&drv->dynids.lock);
677 list_for_each_entry(dynid, &drv->dynids.list, node) {
678 if (guid_equal(&dynid->id.guid, guid)) {
679 id = &dynid->id;
680 break;
681 }
682 }
683 spin_unlock(&drv->dynids.lock);
684
685 return id;
686}
687
688static const struct hv_vmbus_device_id vmbus_device_null;
689
690/*
691 * Return a matching hv_vmbus_device_id pointer.
692 * If there is no match, return NULL.
693 */
694static const struct hv_vmbus_device_id *hv_vmbus_get_id(struct hv_driver *drv,
695 struct hv_device *dev)
696{
697 const guid_t *guid = &dev->dev_type;
698 const struct hv_vmbus_device_id *id;
699
700 /* When driver_override is set, only bind to the matching driver */
701 if (dev->driver_override && strcmp(dev->driver_override, drv->name))
702 return NULL;
703
704 /* Look at the dynamic ids first, before the static ones */
705 id = hv_vmbus_dynid_match(drv, guid);
706 if (!id)
707 id = hv_vmbus_dev_match(drv->id_table, guid);
708
709 /* driver_override will always match, send a dummy id */
710 if (!id && dev->driver_override)
711 id = &vmbus_device_null;
712
713 return id;
714}
715
716/* vmbus_add_dynid - add a new device ID to this driver and re-probe devices */
717static int vmbus_add_dynid(struct hv_driver *drv, guid_t *guid)
718{
719 struct vmbus_dynid *dynid;
720
721 dynid = kzalloc(sizeof(*dynid), GFP_KERNEL);
722 if (!dynid)
723 return -ENOMEM;
724
725 dynid->id.guid = *guid;
726
727 spin_lock(&drv->dynids.lock);
728 list_add_tail(&dynid->node, &drv->dynids.list);
729 spin_unlock(&drv->dynids.lock);
730
731 return driver_attach(&drv->driver);
732}
733
734static void vmbus_free_dynids(struct hv_driver *drv)
735{
736 struct vmbus_dynid *dynid, *n;
737
738 spin_lock(&drv->dynids.lock);
739 list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
740 list_del(&dynid->node);
741 kfree(dynid);
742 }
743 spin_unlock(&drv->dynids.lock);
744}
745
746/*
747 * store_new_id - sysfs frontend to vmbus_add_dynid()
748 *
749 * Allow GUIDs to be added to an existing driver via sysfs.
750 */
751static ssize_t new_id_store(struct device_driver *driver, const char *buf,
752 size_t count)
753{
754 struct hv_driver *drv = drv_to_hv_drv(driver);
755 guid_t guid;
756 ssize_t retval;
757
758 retval = guid_parse(buf, &guid);
759 if (retval)
760 return retval;
761
762 if (hv_vmbus_dynid_match(drv, &guid))
763 return -EEXIST;
764
765 retval = vmbus_add_dynid(drv, &guid);
766 if (retval)
767 return retval;
768 return count;
769}
770static DRIVER_ATTR_WO(new_id);
771
772/*
773 * store_remove_id - remove a PCI device ID from this driver
774 *
775 * Removes a dynamic pci device ID to this driver.
776 */
777static ssize_t remove_id_store(struct device_driver *driver, const char *buf,
778 size_t count)
779{
780 struct hv_driver *drv = drv_to_hv_drv(driver);
781 struct vmbus_dynid *dynid, *n;
782 guid_t guid;
783 ssize_t retval;
784
785 retval = guid_parse(buf, &guid);
786 if (retval)
787 return retval;
788
789 retval = -ENODEV;
790 spin_lock(&drv->dynids.lock);
791 list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
792 struct hv_vmbus_device_id *id = &dynid->id;
793
794 if (guid_equal(&id->guid, &guid)) {
795 list_del(&dynid->node);
796 kfree(dynid);
797 retval = count;
798 break;
799 }
800 }
801 spin_unlock(&drv->dynids.lock);
802
803 return retval;
804}
805static DRIVER_ATTR_WO(remove_id);
806
807static struct attribute *vmbus_drv_attrs[] = {
808 &driver_attr_new_id.attr,
809 &driver_attr_remove_id.attr,
810 NULL,
811};
812ATTRIBUTE_GROUPS(vmbus_drv);
813
814
815/*
816 * vmbus_match - Attempt to match the specified device to the specified driver
817 */
818static int vmbus_match(struct device *device, struct device_driver *driver)
819{
820 struct hv_driver *drv = drv_to_hv_drv(driver);
821 struct hv_device *hv_dev = device_to_hv_device(device);
822
823 /* The hv_sock driver handles all hv_sock offers. */
824 if (is_hvsock_channel(hv_dev->channel))
825 return drv->hvsock;
826
827 if (hv_vmbus_get_id(drv, hv_dev))
828 return 1;
829
830 return 0;
831}
832
833/*
834 * vmbus_probe - Add the new vmbus's child device
835 */
836static int vmbus_probe(struct device *child_device)
837{
838 int ret = 0;
839 struct hv_driver *drv =
840 drv_to_hv_drv(child_device->driver);
841 struct hv_device *dev = device_to_hv_device(child_device);
842 const struct hv_vmbus_device_id *dev_id;
843
844 dev_id = hv_vmbus_get_id(drv, dev);
845 if (drv->probe) {
846 ret = drv->probe(dev, dev_id);
847 if (ret != 0)
848 pr_err("probe failed for device %s (%d)\n",
849 dev_name(child_device), ret);
850
851 } else {
852 pr_err("probe not set for driver %s\n",
853 dev_name(child_device));
854 ret = -ENODEV;
855 }
856 return ret;
857}
858
859/*
860 * vmbus_remove - Remove a vmbus device
861 */
862static int vmbus_remove(struct device *child_device)
863{
864 struct hv_driver *drv;
865 struct hv_device *dev = device_to_hv_device(child_device);
866
867 if (child_device->driver) {
868 drv = drv_to_hv_drv(child_device->driver);
869 if (drv->remove)
870 drv->remove(dev);
871 }
872
873 return 0;
874}
875
876
877/*
878 * vmbus_shutdown - Shutdown a vmbus device
879 */
880static void vmbus_shutdown(struct device *child_device)
881{
882 struct hv_driver *drv;
883 struct hv_device *dev = device_to_hv_device(child_device);
884
885
886 /* The device may not be attached yet */
887 if (!child_device->driver)
888 return;
889
890 drv = drv_to_hv_drv(child_device->driver);
891
892 if (drv->shutdown)
893 drv->shutdown(dev);
894}
895
896
897/*
898 * vmbus_device_release - Final callback release of the vmbus child device
899 */
900static void vmbus_device_release(struct device *device)
901{
902 struct hv_device *hv_dev = device_to_hv_device(device);
903 struct vmbus_channel *channel = hv_dev->channel;
904
905 mutex_lock(&vmbus_connection.channel_mutex);
906 hv_process_channel_removal(channel);
907 mutex_unlock(&vmbus_connection.channel_mutex);
908 kfree(hv_dev);
909}
910
911/* The one and only one */
912static struct bus_type hv_bus = {
913 .name = "vmbus",
914 .match = vmbus_match,
915 .shutdown = vmbus_shutdown,
916 .remove = vmbus_remove,
917 .probe = vmbus_probe,
918 .uevent = vmbus_uevent,
919 .dev_groups = vmbus_dev_groups,
920 .drv_groups = vmbus_drv_groups,
921};
922
923struct onmessage_work_context {
924 struct work_struct work;
925 struct hv_message msg;
926};
927
928static void vmbus_onmessage_work(struct work_struct *work)
929{
930 struct onmessage_work_context *ctx;
931
932 /* Do not process messages if we're in DISCONNECTED state */
933 if (vmbus_connection.conn_state == DISCONNECTED)
934 return;
935
936 ctx = container_of(work, struct onmessage_work_context,
937 work);
938 vmbus_onmessage(&ctx->msg);
939 kfree(ctx);
940}
941
942static void hv_process_timer_expiration(struct hv_message *msg,
943 struct hv_per_cpu_context *hv_cpu)
944{
945 struct clock_event_device *dev = hv_cpu->clk_evt;
946
947 if (dev->event_handler)
948 dev->event_handler(dev);
949
950 vmbus_signal_eom(msg, HVMSG_TIMER_EXPIRED);
951}
952
953void vmbus_on_msg_dpc(unsigned long data)
954{
955 struct hv_per_cpu_context *hv_cpu = (void *)data;
956 void *page_addr = hv_cpu->synic_message_page;
957 struct hv_message *msg = (struct hv_message *)page_addr +
958 VMBUS_MESSAGE_SINT;
959 struct vmbus_channel_message_header *hdr;
960 const struct vmbus_channel_message_table_entry *entry;
961 struct onmessage_work_context *ctx;
962 u32 message_type = msg->header.message_type;
963
964 if (message_type == HVMSG_NONE)
965 /* no msg */
966 return;
967
968 hdr = (struct vmbus_channel_message_header *)msg->u.payload;
969
970 trace_vmbus_on_msg_dpc(hdr);
971
972 if (hdr->msgtype >= CHANNELMSG_COUNT) {
973 WARN_ONCE(1, "unknown msgtype=%d\n", hdr->msgtype);
974 goto msg_handled;
975 }
976
977 entry = &channel_message_table[hdr->msgtype];
978 if (entry->handler_type == VMHT_BLOCKING) {
979 ctx = kmalloc(sizeof(*ctx), GFP_ATOMIC);
980 if (ctx == NULL)
981 return;
982
983 INIT_WORK(&ctx->work, vmbus_onmessage_work);
984 memcpy(&ctx->msg, msg, sizeof(*msg));
985
986 /*
987 * The host can generate a rescind message while we
988 * may still be handling the original offer. We deal with
989 * this condition by ensuring the processing is done on the
990 * same CPU.
991 */
992 switch (hdr->msgtype) {
993 case CHANNELMSG_RESCIND_CHANNELOFFER:
994 /*
995 * If we are handling the rescind message;
996 * schedule the work on the global work queue.
997 */
998 schedule_work_on(vmbus_connection.connect_cpu,
999 &ctx->work);
1000 break;
1001
1002 case CHANNELMSG_OFFERCHANNEL:
1003 atomic_inc(&vmbus_connection.offer_in_progress);
1004 queue_work_on(vmbus_connection.connect_cpu,
1005 vmbus_connection.work_queue,
1006 &ctx->work);
1007 break;
1008
1009 default:
1010 queue_work(vmbus_connection.work_queue, &ctx->work);
1011 }
1012 } else
1013 entry->message_handler(hdr);
1014
1015msg_handled:
1016 vmbus_signal_eom(msg, message_type);
1017}
1018
1019
1020/*
1021 * Direct callback for channels using other deferred processing
1022 */
1023static void vmbus_channel_isr(struct vmbus_channel *channel)
1024{
1025 void (*callback_fn)(void *);
1026
1027 callback_fn = READ_ONCE(channel->onchannel_callback);
1028 if (likely(callback_fn != NULL))
1029 (*callback_fn)(channel->channel_callback_context);
1030}
1031
1032/*
1033 * Schedule all channels with events pending
1034 */
1035static void vmbus_chan_sched(struct hv_per_cpu_context *hv_cpu)
1036{
1037 unsigned long *recv_int_page;
1038 u32 maxbits, relid;
1039
1040 if (vmbus_proto_version < VERSION_WIN8) {
1041 maxbits = MAX_NUM_CHANNELS_SUPPORTED;
1042 recv_int_page = vmbus_connection.recv_int_page;
1043 } else {
1044 /*
1045 * When the host is win8 and beyond, the event page
1046 * can be directly checked to get the id of the channel
1047 * that has the interrupt pending.
1048 */
1049 void *page_addr = hv_cpu->synic_event_page;
1050 union hv_synic_event_flags *event
1051 = (union hv_synic_event_flags *)page_addr +
1052 VMBUS_MESSAGE_SINT;
1053
1054 maxbits = HV_EVENT_FLAGS_COUNT;
1055 recv_int_page = event->flags;
1056 }
1057
1058 if (unlikely(!recv_int_page))
1059 return;
1060
1061 for_each_set_bit(relid, recv_int_page, maxbits) {
1062 struct vmbus_channel *channel;
1063
1064 if (!sync_test_and_clear_bit(relid, recv_int_page))
1065 continue;
1066
1067 /* Special case - vmbus channel protocol msg */
1068 if (relid == 0)
1069 continue;
1070
1071 rcu_read_lock();
1072
1073 /* Find channel based on relid */
1074 list_for_each_entry_rcu(channel, &hv_cpu->chan_list, percpu_list) {
1075 if (channel->offermsg.child_relid != relid)
1076 continue;
1077
1078 if (channel->rescind)
1079 continue;
1080
1081 trace_vmbus_chan_sched(channel);
1082
1083 ++channel->interrupts;
1084
1085 switch (channel->callback_mode) {
1086 case HV_CALL_ISR:
1087 vmbus_channel_isr(channel);
1088 break;
1089
1090 case HV_CALL_BATCHED:
1091 hv_begin_read(&channel->inbound);
1092 /* fallthrough */
1093 case HV_CALL_DIRECT:
1094 tasklet_schedule(&channel->callback_event);
1095 }
1096 }
1097
1098 rcu_read_unlock();
1099 }
1100}
1101
1102static void vmbus_isr(void)
1103{
1104 struct hv_per_cpu_context *hv_cpu
1105 = this_cpu_ptr(hv_context.cpu_context);
1106 void *page_addr = hv_cpu->synic_event_page;
1107 struct hv_message *msg;
1108 union hv_synic_event_flags *event;
1109 bool handled = false;
1110
1111 if (unlikely(page_addr == NULL))
1112 return;
1113
1114 event = (union hv_synic_event_flags *)page_addr +
1115 VMBUS_MESSAGE_SINT;
1116 /*
1117 * Check for events before checking for messages. This is the order
1118 * in which events and messages are checked in Windows guests on
1119 * Hyper-V, and the Windows team suggested we do the same.
1120 */
1121
1122 if ((vmbus_proto_version == VERSION_WS2008) ||
1123 (vmbus_proto_version == VERSION_WIN7)) {
1124
1125 /* Since we are a child, we only need to check bit 0 */
1126 if (sync_test_and_clear_bit(0, event->flags))
1127 handled = true;
1128 } else {
1129 /*
1130 * Our host is win8 or above. The signaling mechanism
1131 * has changed and we can directly look at the event page.
1132 * If bit n is set then we have an interrup on the channel
1133 * whose id is n.
1134 */
1135 handled = true;
1136 }
1137
1138 if (handled)
1139 vmbus_chan_sched(hv_cpu);
1140
1141 page_addr = hv_cpu->synic_message_page;
1142 msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
1143
1144 /* Check if there are actual msgs to be processed */
1145 if (msg->header.message_type != HVMSG_NONE) {
1146 if (msg->header.message_type == HVMSG_TIMER_EXPIRED)
1147 hv_process_timer_expiration(msg, hv_cpu);
1148 else
1149 tasklet_schedule(&hv_cpu->msg_dpc);
1150 }
1151
1152 add_interrupt_randomness(HYPERVISOR_CALLBACK_VECTOR, 0);
1153}
1154
1155/*
1156 * Boolean to control whether to report panic messages over Hyper-V.
1157 *
1158 * It can be set via /proc/sys/kernel/hyperv/record_panic_msg
1159 */
1160static int sysctl_record_panic_msg = 1;
1161
1162/*
1163 * Callback from kmsg_dump. Grab as much as possible from the end of the kmsg
1164 * buffer and call into Hyper-V to transfer the data.
1165 */
1166static void hv_kmsg_dump(struct kmsg_dumper *dumper,
1167 enum kmsg_dump_reason reason)
1168{
1169 size_t bytes_written;
1170 phys_addr_t panic_pa;
1171
1172 /* We are only interested in panics. */
1173 if ((reason != KMSG_DUMP_PANIC) || (!sysctl_record_panic_msg))
1174 return;
1175
1176 panic_pa = virt_to_phys(hv_panic_page);
1177
1178 /*
1179 * Write dump contents to the page. No need to synchronize; panic should
1180 * be single-threaded.
1181 */
1182 kmsg_dump_get_buffer(dumper, true, hv_panic_page, PAGE_SIZE,
1183 &bytes_written);
1184 if (bytes_written)
1185 hyperv_report_panic_msg(panic_pa, bytes_written);
1186}
1187
1188static struct kmsg_dumper hv_kmsg_dumper = {
1189 .dump = hv_kmsg_dump,
1190};
1191
1192static struct ctl_table_header *hv_ctl_table_hdr;
1193static int zero;
1194static int one = 1;
1195
1196/*
1197 * sysctl option to allow the user to control whether kmsg data should be
1198 * reported to Hyper-V on panic.
1199 */
1200static struct ctl_table hv_ctl_table[] = {
1201 {
1202 .procname = "hyperv_record_panic_msg",
1203 .data = &sysctl_record_panic_msg,
1204 .maxlen = sizeof(int),
1205 .mode = 0644,
1206 .proc_handler = proc_dointvec_minmax,
1207 .extra1 = &zero,
1208 .extra2 = &one
1209 },
1210 {}
1211};
1212
1213static struct ctl_table hv_root_table[] = {
1214 {
1215 .procname = "kernel",
1216 .mode = 0555,
1217 .child = hv_ctl_table
1218 },
1219 {}
1220};
1221
1222/*
1223 * vmbus_bus_init -Main vmbus driver initialization routine.
1224 *
1225 * Here, we
1226 * - initialize the vmbus driver context
1227 * - invoke the vmbus hv main init routine
1228 * - retrieve the channel offers
1229 */
1230static int vmbus_bus_init(void)
1231{
1232 int ret;
1233
1234 /* Hypervisor initialization...setup hypercall page..etc */
1235 ret = hv_init();
1236 if (ret != 0) {
1237 pr_err("Unable to initialize the hypervisor - 0x%x\n", ret);
1238 return ret;
1239 }
1240
1241 ret = bus_register(&hv_bus);
1242 if (ret)
1243 return ret;
1244
1245 hv_setup_vmbus_irq(vmbus_isr);
1246
1247 ret = hv_synic_alloc();
1248 if (ret)
1249 goto err_alloc;
1250 /*
1251 * Initialize the per-cpu interrupt state and
1252 * connect to the host.
1253 */
1254 ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "hyperv/vmbus:online",
1255 hv_synic_init, hv_synic_cleanup);
1256 if (ret < 0)
1257 goto err_alloc;
1258 hyperv_cpuhp_online = ret;
1259
1260 ret = vmbus_connect();
1261 if (ret)
1262 goto err_connect;
1263
1264 /*
1265 * Only register if the crash MSRs are available
1266 */
1267 if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1268 u64 hyperv_crash_ctl;
1269 /*
1270 * Sysctl registration is not fatal, since by default
1271 * reporting is enabled.
1272 */
1273 hv_ctl_table_hdr = register_sysctl_table(hv_root_table);
1274 if (!hv_ctl_table_hdr)
1275 pr_err("Hyper-V: sysctl table register error");
1276
1277 /*
1278 * Register for panic kmsg callback only if the right
1279 * capability is supported by the hypervisor.
1280 */
1281 hv_get_crash_ctl(hyperv_crash_ctl);
1282 if (hyperv_crash_ctl & HV_CRASH_CTL_CRASH_NOTIFY_MSG) {
1283 hv_panic_page = (void *)get_zeroed_page(GFP_KERNEL);
1284 if (hv_panic_page) {
1285 ret = kmsg_dump_register(&hv_kmsg_dumper);
1286 if (ret)
1287 pr_err("Hyper-V: kmsg dump register "
1288 "error 0x%x\n", ret);
1289 } else
1290 pr_err("Hyper-V: panic message page memory "
1291 "allocation failed");
1292 }
1293
1294 register_die_notifier(&hyperv_die_block);
1295 atomic_notifier_chain_register(&panic_notifier_list,
1296 &hyperv_panic_block);
1297 }
1298
1299 vmbus_request_offers();
1300
1301 return 0;
1302
1303err_connect:
1304 cpuhp_remove_state(hyperv_cpuhp_online);
1305err_alloc:
1306 hv_synic_free();
1307 hv_remove_vmbus_irq();
1308
1309 bus_unregister(&hv_bus);
1310 free_page((unsigned long)hv_panic_page);
1311 unregister_sysctl_table(hv_ctl_table_hdr);
1312 hv_ctl_table_hdr = NULL;
1313 return ret;
1314}
1315
1316/**
1317 * __vmbus_child_driver_register() - Register a vmbus's driver
1318 * @hv_driver: Pointer to driver structure you want to register
1319 * @owner: owner module of the drv
1320 * @mod_name: module name string
1321 *
1322 * Registers the given driver with Linux through the 'driver_register()' call
1323 * and sets up the hyper-v vmbus handling for this driver.
1324 * It will return the state of the 'driver_register()' call.
1325 *
1326 */
1327int __vmbus_driver_register(struct hv_driver *hv_driver, struct module *owner, const char *mod_name)
1328{
1329 int ret;
1330
1331 pr_info("registering driver %s\n", hv_driver->name);
1332
1333 ret = vmbus_exists();
1334 if (ret < 0)
1335 return ret;
1336
1337 hv_driver->driver.name = hv_driver->name;
1338 hv_driver->driver.owner = owner;
1339 hv_driver->driver.mod_name = mod_name;
1340 hv_driver->driver.bus = &hv_bus;
1341
1342 spin_lock_init(&hv_driver->dynids.lock);
1343 INIT_LIST_HEAD(&hv_driver->dynids.list);
1344
1345 ret = driver_register(&hv_driver->driver);
1346
1347 return ret;
1348}
1349EXPORT_SYMBOL_GPL(__vmbus_driver_register);
1350
1351/**
1352 * vmbus_driver_unregister() - Unregister a vmbus's driver
1353 * @hv_driver: Pointer to driver structure you want to
1354 * un-register
1355 *
1356 * Un-register the given driver that was previous registered with a call to
1357 * vmbus_driver_register()
1358 */
1359void vmbus_driver_unregister(struct hv_driver *hv_driver)
1360{
1361 pr_info("unregistering driver %s\n", hv_driver->name);
1362
1363 if (!vmbus_exists()) {
1364 driver_unregister(&hv_driver->driver);
1365 vmbus_free_dynids(hv_driver);
1366 }
1367}
1368EXPORT_SYMBOL_GPL(vmbus_driver_unregister);
1369
1370
1371/*
1372 * Called when last reference to channel is gone.
1373 */
1374static void vmbus_chan_release(struct kobject *kobj)
1375{
1376 struct vmbus_channel *channel
1377 = container_of(kobj, struct vmbus_channel, kobj);
1378
1379 kfree_rcu(channel, rcu);
1380}
1381
1382struct vmbus_chan_attribute {
1383 struct attribute attr;
1384 ssize_t (*show)(const struct vmbus_channel *chan, char *buf);
1385 ssize_t (*store)(struct vmbus_channel *chan,
1386 const char *buf, size_t count);
1387};
1388#define VMBUS_CHAN_ATTR(_name, _mode, _show, _store) \
1389 struct vmbus_chan_attribute chan_attr_##_name \
1390 = __ATTR(_name, _mode, _show, _store)
1391#define VMBUS_CHAN_ATTR_RW(_name) \
1392 struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RW(_name)
1393#define VMBUS_CHAN_ATTR_RO(_name) \
1394 struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RO(_name)
1395#define VMBUS_CHAN_ATTR_WO(_name) \
1396 struct vmbus_chan_attribute chan_attr_##_name = __ATTR_WO(_name)
1397
1398static ssize_t vmbus_chan_attr_show(struct kobject *kobj,
1399 struct attribute *attr, char *buf)
1400{
1401 const struct vmbus_chan_attribute *attribute
1402 = container_of(attr, struct vmbus_chan_attribute, attr);
1403 const struct vmbus_channel *chan
1404 = container_of(kobj, struct vmbus_channel, kobj);
1405
1406 if (!attribute->show)
1407 return -EIO;
1408
1409 if (chan->state != CHANNEL_OPENED_STATE)
1410 return -EINVAL;
1411
1412 return attribute->show(chan, buf);
1413}
1414
1415static const struct sysfs_ops vmbus_chan_sysfs_ops = {
1416 .show = vmbus_chan_attr_show,
1417};
1418
1419static ssize_t out_mask_show(const struct vmbus_channel *channel, char *buf)
1420{
1421 const struct hv_ring_buffer_info *rbi = &channel->outbound;
1422
1423 return sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1424}
1425static VMBUS_CHAN_ATTR_RO(out_mask);
1426
1427static ssize_t in_mask_show(const struct vmbus_channel *channel, char *buf)
1428{
1429 const struct hv_ring_buffer_info *rbi = &channel->inbound;
1430
1431 return sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1432}
1433static VMBUS_CHAN_ATTR_RO(in_mask);
1434
1435static ssize_t read_avail_show(const struct vmbus_channel *channel, char *buf)
1436{
1437 const struct hv_ring_buffer_info *rbi = &channel->inbound;
1438
1439 return sprintf(buf, "%u\n", hv_get_bytes_to_read(rbi));
1440}
1441static VMBUS_CHAN_ATTR_RO(read_avail);
1442
1443static ssize_t write_avail_show(const struct vmbus_channel *channel, char *buf)
1444{
1445 const struct hv_ring_buffer_info *rbi = &channel->outbound;
1446
1447 return sprintf(buf, "%u\n", hv_get_bytes_to_write(rbi));
1448}
1449static VMBUS_CHAN_ATTR_RO(write_avail);
1450
1451static ssize_t show_target_cpu(const struct vmbus_channel *channel, char *buf)
1452{
1453 return sprintf(buf, "%u\n", channel->target_cpu);
1454}
1455static VMBUS_CHAN_ATTR(cpu, S_IRUGO, show_target_cpu, NULL);
1456
1457static ssize_t channel_pending_show(const struct vmbus_channel *channel,
1458 char *buf)
1459{
1460 return sprintf(buf, "%d\n",
1461 channel_pending(channel,
1462 vmbus_connection.monitor_pages[1]));
1463}
1464static VMBUS_CHAN_ATTR(pending, S_IRUGO, channel_pending_show, NULL);
1465
1466static ssize_t channel_latency_show(const struct vmbus_channel *channel,
1467 char *buf)
1468{
1469 return sprintf(buf, "%d\n",
1470 channel_latency(channel,
1471 vmbus_connection.monitor_pages[1]));
1472}
1473static VMBUS_CHAN_ATTR(latency, S_IRUGO, channel_latency_show, NULL);
1474
1475static ssize_t channel_interrupts_show(const struct vmbus_channel *channel, char *buf)
1476{
1477 return sprintf(buf, "%llu\n", channel->interrupts);
1478}
1479static VMBUS_CHAN_ATTR(interrupts, S_IRUGO, channel_interrupts_show, NULL);
1480
1481static ssize_t channel_events_show(const struct vmbus_channel *channel, char *buf)
1482{
1483 return sprintf(buf, "%llu\n", channel->sig_events);
1484}
1485static VMBUS_CHAN_ATTR(events, S_IRUGO, channel_events_show, NULL);
1486
1487static ssize_t channel_intr_in_full_show(const struct vmbus_channel *channel,
1488 char *buf)
1489{
1490 return sprintf(buf, "%llu\n",
1491 (unsigned long long)channel->intr_in_full);
1492}
1493static VMBUS_CHAN_ATTR(intr_in_full, 0444, channel_intr_in_full_show, NULL);
1494
1495static ssize_t channel_intr_out_empty_show(const struct vmbus_channel *channel,
1496 char *buf)
1497{
1498 return sprintf(buf, "%llu\n",
1499 (unsigned long long)channel->intr_out_empty);
1500}
1501static VMBUS_CHAN_ATTR(intr_out_empty, 0444, channel_intr_out_empty_show, NULL);
1502
1503static ssize_t channel_out_full_first_show(const struct vmbus_channel *channel,
1504 char *buf)
1505{
1506 return sprintf(buf, "%llu\n",
1507 (unsigned long long)channel->out_full_first);
1508}
1509static VMBUS_CHAN_ATTR(out_full_first, 0444, channel_out_full_first_show, NULL);
1510
1511static ssize_t channel_out_full_total_show(const struct vmbus_channel *channel,
1512 char *buf)
1513{
1514 return sprintf(buf, "%llu\n",
1515 (unsigned long long)channel->out_full_total);
1516}
1517static VMBUS_CHAN_ATTR(out_full_total, 0444, channel_out_full_total_show, NULL);
1518
1519static ssize_t subchannel_monitor_id_show(const struct vmbus_channel *channel,
1520 char *buf)
1521{
1522 return sprintf(buf, "%u\n", channel->offermsg.monitorid);
1523}
1524static VMBUS_CHAN_ATTR(monitor_id, S_IRUGO, subchannel_monitor_id_show, NULL);
1525
1526static ssize_t subchannel_id_show(const struct vmbus_channel *channel,
1527 char *buf)
1528{
1529 return sprintf(buf, "%u\n",
1530 channel->offermsg.offer.sub_channel_index);
1531}
1532static VMBUS_CHAN_ATTR_RO(subchannel_id);
1533
1534static struct attribute *vmbus_chan_attrs[] = {
1535 &chan_attr_out_mask.attr,
1536 &chan_attr_in_mask.attr,
1537 &chan_attr_read_avail.attr,
1538 &chan_attr_write_avail.attr,
1539 &chan_attr_cpu.attr,
1540 &chan_attr_pending.attr,
1541 &chan_attr_latency.attr,
1542 &chan_attr_interrupts.attr,
1543 &chan_attr_events.attr,
1544 &chan_attr_intr_in_full.attr,
1545 &chan_attr_intr_out_empty.attr,
1546 &chan_attr_out_full_first.attr,
1547 &chan_attr_out_full_total.attr,
1548 &chan_attr_monitor_id.attr,
1549 &chan_attr_subchannel_id.attr,
1550 NULL
1551};
1552
1553static struct kobj_type vmbus_chan_ktype = {
1554 .sysfs_ops = &vmbus_chan_sysfs_ops,
1555 .release = vmbus_chan_release,
1556 .default_attrs = vmbus_chan_attrs,
1557};
1558
1559/*
1560 * vmbus_add_channel_kobj - setup a sub-directory under device/channels
1561 */
1562int vmbus_add_channel_kobj(struct hv_device *dev, struct vmbus_channel *channel)
1563{
1564 struct kobject *kobj = &channel->kobj;
1565 u32 relid = channel->offermsg.child_relid;
1566 int ret;
1567
1568 kobj->kset = dev->channels_kset;
1569 ret = kobject_init_and_add(kobj, &vmbus_chan_ktype, NULL,
1570 "%u", relid);
1571 if (ret)
1572 return ret;
1573
1574 kobject_uevent(kobj, KOBJ_ADD);
1575
1576 return 0;
1577}
1578
1579/*
1580 * vmbus_device_create - Creates and registers a new child device
1581 * on the vmbus.
1582 */
1583struct hv_device *vmbus_device_create(const guid_t *type,
1584 const guid_t *instance,
1585 struct vmbus_channel *channel)
1586{
1587 struct hv_device *child_device_obj;
1588
1589 child_device_obj = kzalloc(sizeof(struct hv_device), GFP_KERNEL);
1590 if (!child_device_obj) {
1591 pr_err("Unable to allocate device object for child device\n");
1592 return NULL;
1593 }
1594
1595 child_device_obj->channel = channel;
1596 guid_copy(&child_device_obj->dev_type, type);
1597 guid_copy(&child_device_obj->dev_instance, instance);
1598 child_device_obj->vendor_id = 0x1414; /* MSFT vendor ID */
1599
1600 return child_device_obj;
1601}
1602
1603/*
1604 * vmbus_device_register - Register the child device
1605 */
1606int vmbus_device_register(struct hv_device *child_device_obj)
1607{
1608 struct kobject *kobj = &child_device_obj->device.kobj;
1609 int ret;
1610
1611 dev_set_name(&child_device_obj->device, "%pUl",
1612 child_device_obj->channel->offermsg.offer.if_instance.b);
1613
1614 child_device_obj->device.bus = &hv_bus;
1615 child_device_obj->device.parent = &hv_acpi_dev->dev;
1616 child_device_obj->device.release = vmbus_device_release;
1617
1618 /*
1619 * Register with the LDM. This will kick off the driver/device
1620 * binding...which will eventually call vmbus_match() and vmbus_probe()
1621 */
1622 ret = device_register(&child_device_obj->device);
1623 if (ret) {
1624 pr_err("Unable to register child device\n");
1625 return ret;
1626 }
1627
1628 child_device_obj->channels_kset = kset_create_and_add("channels",
1629 NULL, kobj);
1630 if (!child_device_obj->channels_kset) {
1631 ret = -ENOMEM;
1632 goto err_dev_unregister;
1633 }
1634
1635 ret = vmbus_add_channel_kobj(child_device_obj,
1636 child_device_obj->channel);
1637 if (ret) {
1638 pr_err("Unable to register primary channeln");
1639 goto err_kset_unregister;
1640 }
1641
1642 return 0;
1643
1644err_kset_unregister:
1645 kset_unregister(child_device_obj->channels_kset);
1646
1647err_dev_unregister:
1648 device_unregister(&child_device_obj->device);
1649 return ret;
1650}
1651
1652/*
1653 * vmbus_device_unregister - Remove the specified child device
1654 * from the vmbus.
1655 */
1656void vmbus_device_unregister(struct hv_device *device_obj)
1657{
1658 pr_debug("child device %s unregistered\n",
1659 dev_name(&device_obj->device));
1660
1661 kset_unregister(device_obj->channels_kset);
1662
1663 /*
1664 * Kick off the process of unregistering the device.
1665 * This will call vmbus_remove() and eventually vmbus_device_release()
1666 */
1667 device_unregister(&device_obj->device);
1668}
1669
1670
1671/*
1672 * VMBUS is an acpi enumerated device. Get the information we
1673 * need from DSDT.
1674 */
1675#define VTPM_BASE_ADDRESS 0xfed40000
1676static acpi_status vmbus_walk_resources(struct acpi_resource *res, void *ctx)
1677{
1678 resource_size_t start = 0;
1679 resource_size_t end = 0;
1680 struct resource *new_res;
1681 struct resource **old_res = &hyperv_mmio;
1682 struct resource **prev_res = NULL;
1683
1684 switch (res->type) {
1685
1686 /*
1687 * "Address" descriptors are for bus windows. Ignore
1688 * "memory" descriptors, which are for registers on
1689 * devices.
1690 */
1691 case ACPI_RESOURCE_TYPE_ADDRESS32:
1692 start = res->data.address32.address.minimum;
1693 end = res->data.address32.address.maximum;
1694 break;
1695
1696 case ACPI_RESOURCE_TYPE_ADDRESS64:
1697 start = res->data.address64.address.minimum;
1698 end = res->data.address64.address.maximum;
1699 break;
1700
1701 default:
1702 /* Unused resource type */
1703 return AE_OK;
1704
1705 }
1706 /*
1707 * Ignore ranges that are below 1MB, as they're not
1708 * necessary or useful here.
1709 */
1710 if (end < 0x100000)
1711 return AE_OK;
1712
1713 new_res = kzalloc(sizeof(*new_res), GFP_ATOMIC);
1714 if (!new_res)
1715 return AE_NO_MEMORY;
1716
1717 /* If this range overlaps the virtual TPM, truncate it. */
1718 if (end > VTPM_BASE_ADDRESS && start < VTPM_BASE_ADDRESS)
1719 end = VTPM_BASE_ADDRESS;
1720
1721 new_res->name = "hyperv mmio";
1722 new_res->flags = IORESOURCE_MEM;
1723 new_res->start = start;
1724 new_res->end = end;
1725
1726 /*
1727 * If two ranges are adjacent, merge them.
1728 */
1729 do {
1730 if (!*old_res) {
1731 *old_res = new_res;
1732 break;
1733 }
1734
1735 if (((*old_res)->end + 1) == new_res->start) {
1736 (*old_res)->end = new_res->end;
1737 kfree(new_res);
1738 break;
1739 }
1740
1741 if ((*old_res)->start == new_res->end + 1) {
1742 (*old_res)->start = new_res->start;
1743 kfree(new_res);
1744 break;
1745 }
1746
1747 if ((*old_res)->start > new_res->end) {
1748 new_res->sibling = *old_res;
1749 if (prev_res)
1750 (*prev_res)->sibling = new_res;
1751 *old_res = new_res;
1752 break;
1753 }
1754
1755 prev_res = old_res;
1756 old_res = &(*old_res)->sibling;
1757
1758 } while (1);
1759
1760 return AE_OK;
1761}
1762
1763static int vmbus_acpi_remove(struct acpi_device *device)
1764{
1765 struct resource *cur_res;
1766 struct resource *next_res;
1767
1768 if (hyperv_mmio) {
1769 if (fb_mmio) {
1770 __release_region(hyperv_mmio, fb_mmio->start,
1771 resource_size(fb_mmio));
1772 fb_mmio = NULL;
1773 }
1774
1775 for (cur_res = hyperv_mmio; cur_res; cur_res = next_res) {
1776 next_res = cur_res->sibling;
1777 kfree(cur_res);
1778 }
1779 }
1780
1781 return 0;
1782}
1783
1784static void vmbus_reserve_fb(void)
1785{
1786 int size;
1787 /*
1788 * Make a claim for the frame buffer in the resource tree under the
1789 * first node, which will be the one below 4GB. The length seems to
1790 * be underreported, particularly in a Generation 1 VM. So start out
1791 * reserving a larger area and make it smaller until it succeeds.
1792 */
1793
1794 if (screen_info.lfb_base) {
1795 if (efi_enabled(EFI_BOOT))
1796 size = max_t(__u32, screen_info.lfb_size, 0x800000);
1797 else
1798 size = max_t(__u32, screen_info.lfb_size, 0x4000000);
1799
1800 for (; !fb_mmio && (size >= 0x100000); size >>= 1) {
1801 fb_mmio = __request_region(hyperv_mmio,
1802 screen_info.lfb_base, size,
1803 fb_mmio_name, 0);
1804 }
1805 }
1806}
1807
1808/**
1809 * vmbus_allocate_mmio() - Pick a memory-mapped I/O range.
1810 * @new: If successful, supplied a pointer to the
1811 * allocated MMIO space.
1812 * @device_obj: Identifies the caller
1813 * @min: Minimum guest physical address of the
1814 * allocation
1815 * @max: Maximum guest physical address
1816 * @size: Size of the range to be allocated
1817 * @align: Alignment of the range to be allocated
1818 * @fb_overlap_ok: Whether this allocation can be allowed
1819 * to overlap the video frame buffer.
1820 *
1821 * This function walks the resources granted to VMBus by the
1822 * _CRS object in the ACPI namespace underneath the parent
1823 * "bridge" whether that's a root PCI bus in the Generation 1
1824 * case or a Module Device in the Generation 2 case. It then
1825 * attempts to allocate from the global MMIO pool in a way that
1826 * matches the constraints supplied in these parameters and by
1827 * that _CRS.
1828 *
1829 * Return: 0 on success, -errno on failure
1830 */
1831int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
1832 resource_size_t min, resource_size_t max,
1833 resource_size_t size, resource_size_t align,
1834 bool fb_overlap_ok)
1835{
1836 struct resource *iter, *shadow;
1837 resource_size_t range_min, range_max, start;
1838 const char *dev_n = dev_name(&device_obj->device);
1839 int retval;
1840
1841 retval = -ENXIO;
1842 down(&hyperv_mmio_lock);
1843
1844 /*
1845 * If overlaps with frame buffers are allowed, then first attempt to
1846 * make the allocation from within the reserved region. Because it
1847 * is already reserved, no shadow allocation is necessary.
1848 */
1849 if (fb_overlap_ok && fb_mmio && !(min > fb_mmio->end) &&
1850 !(max < fb_mmio->start)) {
1851
1852 range_min = fb_mmio->start;
1853 range_max = fb_mmio->end;
1854 start = (range_min + align - 1) & ~(align - 1);
1855 for (; start + size - 1 <= range_max; start += align) {
1856 *new = request_mem_region_exclusive(start, size, dev_n);
1857 if (*new) {
1858 retval = 0;
1859 goto exit;
1860 }
1861 }
1862 }
1863
1864 for (iter = hyperv_mmio; iter; iter = iter->sibling) {
1865 if ((iter->start >= max) || (iter->end <= min))
1866 continue;
1867
1868 range_min = iter->start;
1869 range_max = iter->end;
1870 start = (range_min + align - 1) & ~(align - 1);
1871 for (; start + size - 1 <= range_max; start += align) {
1872 shadow = __request_region(iter, start, size, NULL,
1873 IORESOURCE_BUSY);
1874 if (!shadow)
1875 continue;
1876
1877 *new = request_mem_region_exclusive(start, size, dev_n);
1878 if (*new) {
1879 shadow->name = (char *)*new;
1880 retval = 0;
1881 goto exit;
1882 }
1883
1884 __release_region(iter, start, size);
1885 }
1886 }
1887
1888exit:
1889 up(&hyperv_mmio_lock);
1890 return retval;
1891}
1892EXPORT_SYMBOL_GPL(vmbus_allocate_mmio);
1893
1894/**
1895 * vmbus_free_mmio() - Free a memory-mapped I/O range.
1896 * @start: Base address of region to release.
1897 * @size: Size of the range to be allocated
1898 *
1899 * This function releases anything requested by
1900 * vmbus_mmio_allocate().
1901 */
1902void vmbus_free_mmio(resource_size_t start, resource_size_t size)
1903{
1904 struct resource *iter;
1905
1906 down(&hyperv_mmio_lock);
1907 for (iter = hyperv_mmio; iter; iter = iter->sibling) {
1908 if ((iter->start >= start + size) || (iter->end <= start))
1909 continue;
1910
1911 __release_region(iter, start, size);
1912 }
1913 release_mem_region(start, size);
1914 up(&hyperv_mmio_lock);
1915
1916}
1917EXPORT_SYMBOL_GPL(vmbus_free_mmio);
1918
1919static int vmbus_acpi_add(struct acpi_device *device)
1920{
1921 acpi_status result;
1922 int ret_val = -ENODEV;
1923 struct acpi_device *ancestor;
1924
1925 hv_acpi_dev = device;
1926
1927 result = acpi_walk_resources(device->handle, METHOD_NAME__CRS,
1928 vmbus_walk_resources, NULL);
1929
1930 if (ACPI_FAILURE(result))
1931 goto acpi_walk_err;
1932 /*
1933 * Some ancestor of the vmbus acpi device (Gen1 or Gen2
1934 * firmware) is the VMOD that has the mmio ranges. Get that.
1935 */
1936 for (ancestor = device->parent; ancestor; ancestor = ancestor->parent) {
1937 result = acpi_walk_resources(ancestor->handle, METHOD_NAME__CRS,
1938 vmbus_walk_resources, NULL);
1939
1940 if (ACPI_FAILURE(result))
1941 continue;
1942 if (hyperv_mmio) {
1943 vmbus_reserve_fb();
1944 break;
1945 }
1946 }
1947 ret_val = 0;
1948
1949acpi_walk_err:
1950 complete(&probe_event);
1951 if (ret_val)
1952 vmbus_acpi_remove(device);
1953 return ret_val;
1954}
1955
1956static const struct acpi_device_id vmbus_acpi_device_ids[] = {
1957 {"VMBUS", 0},
1958 {"VMBus", 0},
1959 {"", 0},
1960};
1961MODULE_DEVICE_TABLE(acpi, vmbus_acpi_device_ids);
1962
1963static struct acpi_driver vmbus_acpi_driver = {
1964 .name = "vmbus",
1965 .ids = vmbus_acpi_device_ids,
1966 .ops = {
1967 .add = vmbus_acpi_add,
1968 .remove = vmbus_acpi_remove,
1969 },
1970};
1971
1972static void hv_kexec_handler(void)
1973{
1974 hv_synic_clockevents_cleanup();
1975 vmbus_initiate_unload(false);
1976 vmbus_connection.conn_state = DISCONNECTED;
1977 /* Make sure conn_state is set as hv_synic_cleanup checks for it */
1978 mb();
1979 cpuhp_remove_state(hyperv_cpuhp_online);
1980 hyperv_cleanup();
1981};
1982
1983static void hv_crash_handler(struct pt_regs *regs)
1984{
1985 vmbus_initiate_unload(true);
1986 /*
1987 * In crash handler we can't schedule synic cleanup for all CPUs,
1988 * doing the cleanup for current CPU only. This should be sufficient
1989 * for kdump.
1990 */
1991 vmbus_connection.conn_state = DISCONNECTED;
1992 hv_synic_cleanup(smp_processor_id());
1993 hyperv_cleanup();
1994};
1995
1996static int __init hv_acpi_init(void)
1997{
1998 int ret, t;
1999
2000 if (!hv_is_hyperv_initialized())
2001 return -ENODEV;
2002
2003 init_completion(&probe_event);
2004
2005 /*
2006 * Get ACPI resources first.
2007 */
2008 ret = acpi_bus_register_driver(&vmbus_acpi_driver);
2009
2010 if (ret)
2011 return ret;
2012
2013 t = wait_for_completion_timeout(&probe_event, 5*HZ);
2014 if (t == 0) {
2015 ret = -ETIMEDOUT;
2016 goto cleanup;
2017 }
2018
2019 ret = vmbus_bus_init();
2020 if (ret)
2021 goto cleanup;
2022
2023 hv_setup_kexec_handler(hv_kexec_handler);
2024 hv_setup_crash_handler(hv_crash_handler);
2025
2026 return 0;
2027
2028cleanup:
2029 acpi_bus_unregister_driver(&vmbus_acpi_driver);
2030 hv_acpi_dev = NULL;
2031 return ret;
2032}
2033
2034static void __exit vmbus_exit(void)
2035{
2036 int cpu;
2037
2038 hv_remove_kexec_handler();
2039 hv_remove_crash_handler();
2040 vmbus_connection.conn_state = DISCONNECTED;
2041 hv_synic_clockevents_cleanup();
2042 vmbus_disconnect();
2043 hv_remove_vmbus_irq();
2044 for_each_online_cpu(cpu) {
2045 struct hv_per_cpu_context *hv_cpu
2046 = per_cpu_ptr(hv_context.cpu_context, cpu);
2047
2048 tasklet_kill(&hv_cpu->msg_dpc);
2049 }
2050 vmbus_free_channels();
2051
2052 if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
2053 kmsg_dump_unregister(&hv_kmsg_dumper);
2054 unregister_die_notifier(&hyperv_die_block);
2055 atomic_notifier_chain_unregister(&panic_notifier_list,
2056 &hyperv_panic_block);
2057 }
2058
2059 free_page((unsigned long)hv_panic_page);
2060 unregister_sysctl_table(hv_ctl_table_hdr);
2061 hv_ctl_table_hdr = NULL;
2062 bus_unregister(&hv_bus);
2063
2064 cpuhp_remove_state(hyperv_cpuhp_online);
2065 hv_synic_free();
2066 acpi_bus_unregister_driver(&vmbus_acpi_driver);
2067}
2068
2069
2070MODULE_LICENSE("GPL");
2071
2072subsys_initcall(hv_acpi_init);
2073module_exit(vmbus_exit);
2074