OSDN Git Service

clk: at91: fix masterck name
[uclinux-h8/linux.git] / drivers / hv / vmbus_drv.c
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
48 struct vmbus_dynid {
49         struct list_head node;
50         struct hv_vmbus_device_id id;
51 };
52
53 static struct acpi_device  *hv_acpi_dev;
54
55 static struct completion probe_event;
56
57 static int hyperv_cpuhp_online;
58
59 static void *hv_panic_page;
60
61 static int hyperv_panic_event(struct notifier_block *nb, unsigned long val,
62                               void *args)
63 {
64         struct pt_regs *regs;
65
66         regs = current_pt_regs();
67
68         hyperv_report_panic(regs, val);
69         return NOTIFY_DONE;
70 }
71
72 static int hyperv_die_event(struct notifier_block *nb, unsigned long val,
73                             void *args)
74 {
75         struct die_args *die = (struct die_args *)args;
76         struct pt_regs *regs = die->regs;
77
78         hyperv_report_panic(regs, val);
79         return NOTIFY_DONE;
80 }
81
82 static struct notifier_block hyperv_die_block = {
83         .notifier_call = hyperv_die_event,
84 };
85 static struct notifier_block hyperv_panic_block = {
86         .notifier_call = hyperv_panic_event,
87 };
88
89 static const char *fb_mmio_name = "fb_range";
90 static struct resource *fb_mmio;
91 static struct resource *hyperv_mmio;
92 static DEFINE_SEMAPHORE(hyperv_mmio_lock);
93
94 static int vmbus_exists(void)
95 {
96         if (hv_acpi_dev == NULL)
97                 return -ENODEV;
98
99         return 0;
100 }
101
102 #define VMBUS_ALIAS_LEN ((sizeof((struct hv_vmbus_device_id *)0)->guid) * 2)
103 static void print_alias_name(struct hv_device *hv_dev, char *alias_name)
104 {
105         int i;
106         for (i = 0; i < VMBUS_ALIAS_LEN; i += 2)
107                 sprintf(&alias_name[i], "%02x", hv_dev->dev_type.b[i/2]);
108 }
109
110 static u8 channel_monitor_group(const struct vmbus_channel *channel)
111 {
112         return (u8)channel->offermsg.monitorid / 32;
113 }
114
115 static u8 channel_monitor_offset(const struct vmbus_channel *channel)
116 {
117         return (u8)channel->offermsg.monitorid % 32;
118 }
119
120 static 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
128 static 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
137 static 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
145 static 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 }
154 static DEVICE_ATTR_RO(id);
155
156 static 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 }
165 static DEVICE_ATTR_RO(state);
166
167 static 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 }
176 static DEVICE_ATTR_RO(monitor_id);
177
178 static 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 }
188 static DEVICE_ATTR_RO(class_id);
189
190 static 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 }
200 static DEVICE_ATTR_RO(device_id);
201
202 static 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 }
211 static DEVICE_ATTR_RO(modalias);
212
213 #ifdef CONFIG_NUMA
214 static 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 }
224 static DEVICE_ATTR_RO(numa_node);
225 #endif
226
227 static 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[1]));
238 }
239 static DEVICE_ATTR_RO(server_monitor_pending);
240
241 static 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 }
253 static DEVICE_ATTR_RO(client_monitor_pending);
254
255 static 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 }
267 static DEVICE_ATTR_RO(server_monitor_latency);
268
269 static 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 }
281 static DEVICE_ATTR_RO(client_monitor_latency);
282
283 static 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 }
295 static DEVICE_ATTR_RO(server_monitor_conn_id);
296
297 static 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 }
309 static DEVICE_ATTR_RO(client_monitor_conn_id);
310
311 static 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
317         if (!hv_dev->channel)
318                 return -ENODEV;
319         if (hv_dev->channel->state != CHANNEL_OPENED_STATE)
320                 return -EINVAL;
321         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
322         return sprintf(buf, "%d\n", outbound.current_interrupt_mask);
323 }
324 static DEVICE_ATTR_RO(out_intr_mask);
325
326 static ssize_t out_read_index_show(struct device *dev,
327                                    struct device_attribute *dev_attr, char *buf)
328 {
329         struct hv_device *hv_dev = device_to_hv_device(dev);
330         struct hv_ring_buffer_debug_info outbound;
331
332         if (!hv_dev->channel)
333                 return -ENODEV;
334         if (hv_dev->channel->state != CHANNEL_OPENED_STATE)
335                 return -EINVAL;
336         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
337         return sprintf(buf, "%d\n", outbound.current_read_index);
338 }
339 static DEVICE_ATTR_RO(out_read_index);
340
341 static ssize_t out_write_index_show(struct device *dev,
342                                     struct device_attribute *dev_attr,
343                                     char *buf)
344 {
345         struct hv_device *hv_dev = device_to_hv_device(dev);
346         struct hv_ring_buffer_debug_info outbound;
347
348         if (!hv_dev->channel)
349                 return -ENODEV;
350         if (hv_dev->channel->state != CHANNEL_OPENED_STATE)
351                 return -EINVAL;
352         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
353         return sprintf(buf, "%d\n", outbound.current_write_index);
354 }
355 static DEVICE_ATTR_RO(out_write_index);
356
357 static ssize_t out_read_bytes_avail_show(struct device *dev,
358                                          struct device_attribute *dev_attr,
359                                          char *buf)
360 {
361         struct hv_device *hv_dev = device_to_hv_device(dev);
362         struct hv_ring_buffer_debug_info outbound;
363
364         if (!hv_dev->channel)
365                 return -ENODEV;
366         if (hv_dev->channel->state != CHANNEL_OPENED_STATE)
367                 return -EINVAL;
368         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
369         return sprintf(buf, "%d\n", outbound.bytes_avail_toread);
370 }
371 static DEVICE_ATTR_RO(out_read_bytes_avail);
372
373 static ssize_t out_write_bytes_avail_show(struct device *dev,
374                                           struct device_attribute *dev_attr,
375                                           char *buf)
376 {
377         struct hv_device *hv_dev = device_to_hv_device(dev);
378         struct hv_ring_buffer_debug_info outbound;
379
380         if (!hv_dev->channel)
381                 return -ENODEV;
382         if (hv_dev->channel->state != CHANNEL_OPENED_STATE)
383                 return -EINVAL;
384         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
385         return sprintf(buf, "%d\n", outbound.bytes_avail_towrite);
386 }
387 static DEVICE_ATTR_RO(out_write_bytes_avail);
388
389 static ssize_t in_intr_mask_show(struct device *dev,
390                                  struct device_attribute *dev_attr, char *buf)
391 {
392         struct hv_device *hv_dev = device_to_hv_device(dev);
393         struct hv_ring_buffer_debug_info inbound;
394
395         if (!hv_dev->channel)
396                 return -ENODEV;
397         if (hv_dev->channel->state != CHANNEL_OPENED_STATE)
398                 return -EINVAL;
399         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
400         return sprintf(buf, "%d\n", inbound.current_interrupt_mask);
401 }
402 static DEVICE_ATTR_RO(in_intr_mask);
403
404 static ssize_t in_read_index_show(struct device *dev,
405                                   struct device_attribute *dev_attr, char *buf)
406 {
407         struct hv_device *hv_dev = device_to_hv_device(dev);
408         struct hv_ring_buffer_debug_info inbound;
409
410         if (!hv_dev->channel)
411                 return -ENODEV;
412         if (hv_dev->channel->state != CHANNEL_OPENED_STATE)
413                 return -EINVAL;
414         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
415         return sprintf(buf, "%d\n", inbound.current_read_index);
416 }
417 static DEVICE_ATTR_RO(in_read_index);
418
419 static ssize_t in_write_index_show(struct device *dev,
420                                    struct device_attribute *dev_attr, char *buf)
421 {
422         struct hv_device *hv_dev = device_to_hv_device(dev);
423         struct hv_ring_buffer_debug_info inbound;
424
425         if (!hv_dev->channel)
426                 return -ENODEV;
427         if (hv_dev->channel->state != CHANNEL_OPENED_STATE)
428                 return -EINVAL;
429         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
430         return sprintf(buf, "%d\n", inbound.current_write_index);
431 }
432 static DEVICE_ATTR_RO(in_write_index);
433
434 static ssize_t in_read_bytes_avail_show(struct device *dev,
435                                         struct device_attribute *dev_attr,
436                                         char *buf)
437 {
438         struct hv_device *hv_dev = device_to_hv_device(dev);
439         struct hv_ring_buffer_debug_info inbound;
440
441         if (!hv_dev->channel)
442                 return -ENODEV;
443         if (hv_dev->channel->state != CHANNEL_OPENED_STATE)
444                 return -EINVAL;
445         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
446         return sprintf(buf, "%d\n", inbound.bytes_avail_toread);
447 }
448 static DEVICE_ATTR_RO(in_read_bytes_avail);
449
450 static ssize_t in_write_bytes_avail_show(struct device *dev,
451                                          struct device_attribute *dev_attr,
452                                          char *buf)
453 {
454         struct hv_device *hv_dev = device_to_hv_device(dev);
455         struct hv_ring_buffer_debug_info inbound;
456
457         if (!hv_dev->channel)
458                 return -ENODEV;
459         if (hv_dev->channel->state != CHANNEL_OPENED_STATE)
460                 return -EINVAL;
461         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
462         return sprintf(buf, "%d\n", inbound.bytes_avail_towrite);
463 }
464 static DEVICE_ATTR_RO(in_write_bytes_avail);
465
466 static ssize_t channel_vp_mapping_show(struct device *dev,
467                                        struct device_attribute *dev_attr,
468                                        char *buf)
469 {
470         struct hv_device *hv_dev = device_to_hv_device(dev);
471         struct vmbus_channel *channel = hv_dev->channel, *cur_sc;
472         unsigned long flags;
473         int buf_size = PAGE_SIZE, n_written, tot_written;
474         struct list_head *cur;
475
476         if (!channel)
477                 return -ENODEV;
478
479         tot_written = snprintf(buf, buf_size, "%u:%u\n",
480                 channel->offermsg.child_relid, channel->target_cpu);
481
482         spin_lock_irqsave(&channel->lock, flags);
483
484         list_for_each(cur, &channel->sc_list) {
485                 if (tot_written >= buf_size - 1)
486                         break;
487
488                 cur_sc = list_entry(cur, struct vmbus_channel, sc_list);
489                 n_written = scnprintf(buf + tot_written,
490                                      buf_size - tot_written,
491                                      "%u:%u\n",
492                                      cur_sc->offermsg.child_relid,
493                                      cur_sc->target_cpu);
494                 tot_written += n_written;
495         }
496
497         spin_unlock_irqrestore(&channel->lock, flags);
498
499         return tot_written;
500 }
501 static DEVICE_ATTR_RO(channel_vp_mapping);
502
503 static ssize_t vendor_show(struct device *dev,
504                            struct device_attribute *dev_attr,
505                            char *buf)
506 {
507         struct hv_device *hv_dev = device_to_hv_device(dev);
508         return sprintf(buf, "0x%x\n", hv_dev->vendor_id);
509 }
510 static DEVICE_ATTR_RO(vendor);
511
512 static ssize_t device_show(struct device *dev,
513                            struct device_attribute *dev_attr,
514                            char *buf)
515 {
516         struct hv_device *hv_dev = device_to_hv_device(dev);
517         return sprintf(buf, "0x%x\n", hv_dev->device_id);
518 }
519 static DEVICE_ATTR_RO(device);
520
521 static ssize_t driver_override_store(struct device *dev,
522                                      struct device_attribute *attr,
523                                      const char *buf, size_t count)
524 {
525         struct hv_device *hv_dev = device_to_hv_device(dev);
526         char *driver_override, *old, *cp;
527
528         /* We need to keep extra room for a newline */
529         if (count >= (PAGE_SIZE - 1))
530                 return -EINVAL;
531
532         driver_override = kstrndup(buf, count, GFP_KERNEL);
533         if (!driver_override)
534                 return -ENOMEM;
535
536         cp = strchr(driver_override, '\n');
537         if (cp)
538                 *cp = '\0';
539
540         device_lock(dev);
541         old = hv_dev->driver_override;
542         if (strlen(driver_override)) {
543                 hv_dev->driver_override = driver_override;
544         } else {
545                 kfree(driver_override);
546                 hv_dev->driver_override = NULL;
547         }
548         device_unlock(dev);
549
550         kfree(old);
551
552         return count;
553 }
554
555 static ssize_t driver_override_show(struct device *dev,
556                                     struct device_attribute *attr, char *buf)
557 {
558         struct hv_device *hv_dev = device_to_hv_device(dev);
559         ssize_t len;
560
561         device_lock(dev);
562         len = snprintf(buf, PAGE_SIZE, "%s\n", hv_dev->driver_override);
563         device_unlock(dev);
564
565         return len;
566 }
567 static DEVICE_ATTR_RW(driver_override);
568
569 /* Set up per device attributes in /sys/bus/vmbus/devices/<bus device> */
570 static struct attribute *vmbus_dev_attrs[] = {
571         &dev_attr_id.attr,
572         &dev_attr_state.attr,
573         &dev_attr_monitor_id.attr,
574         &dev_attr_class_id.attr,
575         &dev_attr_device_id.attr,
576         &dev_attr_modalias.attr,
577 #ifdef CONFIG_NUMA
578         &dev_attr_numa_node.attr,
579 #endif
580         &dev_attr_server_monitor_pending.attr,
581         &dev_attr_client_monitor_pending.attr,
582         &dev_attr_server_monitor_latency.attr,
583         &dev_attr_client_monitor_latency.attr,
584         &dev_attr_server_monitor_conn_id.attr,
585         &dev_attr_client_monitor_conn_id.attr,
586         &dev_attr_out_intr_mask.attr,
587         &dev_attr_out_read_index.attr,
588         &dev_attr_out_write_index.attr,
589         &dev_attr_out_read_bytes_avail.attr,
590         &dev_attr_out_write_bytes_avail.attr,
591         &dev_attr_in_intr_mask.attr,
592         &dev_attr_in_read_index.attr,
593         &dev_attr_in_write_index.attr,
594         &dev_attr_in_read_bytes_avail.attr,
595         &dev_attr_in_write_bytes_avail.attr,
596         &dev_attr_channel_vp_mapping.attr,
597         &dev_attr_vendor.attr,
598         &dev_attr_device.attr,
599         &dev_attr_driver_override.attr,
600         NULL,
601 };
602 ATTRIBUTE_GROUPS(vmbus_dev);
603
604 /*
605  * vmbus_uevent - add uevent for our device
606  *
607  * This routine is invoked when a device is added or removed on the vmbus to
608  * generate a uevent to udev in the userspace. The udev will then look at its
609  * rule and the uevent generated here to load the appropriate driver
610  *
611  * The alias string will be of the form vmbus:guid where guid is the string
612  * representation of the device guid (each byte of the guid will be
613  * represented with two hex characters.
614  */
615 static int vmbus_uevent(struct device *device, struct kobj_uevent_env *env)
616 {
617         struct hv_device *dev = device_to_hv_device(device);
618         int ret;
619         char alias_name[VMBUS_ALIAS_LEN + 1];
620
621         print_alias_name(dev, alias_name);
622         ret = add_uevent_var(env, "MODALIAS=vmbus:%s", alias_name);
623         return ret;
624 }
625
626 static const uuid_le null_guid;
627
628 static inline bool is_null_guid(const uuid_le *guid)
629 {
630         if (uuid_le_cmp(*guid, null_guid))
631                 return false;
632         return true;
633 }
634
635 static const struct hv_vmbus_device_id *
636 hv_vmbus_dev_match(const struct hv_vmbus_device_id *id, const uuid_le *guid)
637
638 {
639         if (id == NULL)
640                 return NULL; /* empty device table */
641
642         for (; !is_null_guid(&id->guid); id++)
643                 if (!uuid_le_cmp(id->guid, *guid))
644                         return id;
645
646         return NULL;
647 }
648
649 static const struct hv_vmbus_device_id *
650 hv_vmbus_dynid_match(struct hv_driver *drv, const uuid_le *guid)
651 {
652         const struct hv_vmbus_device_id *id = NULL;
653         struct vmbus_dynid *dynid;
654
655         spin_lock(&drv->dynids.lock);
656         list_for_each_entry(dynid, &drv->dynids.list, node) {
657                 if (!uuid_le_cmp(dynid->id.guid, *guid)) {
658                         id = &dynid->id;
659                         break;
660                 }
661         }
662         spin_unlock(&drv->dynids.lock);
663
664         return id;
665 }
666
667 static const struct hv_vmbus_device_id vmbus_device_null = {
668         .guid = NULL_UUID_LE,
669 };
670
671 /*
672  * Return a matching hv_vmbus_device_id pointer.
673  * If there is no match, return NULL.
674  */
675 static const struct hv_vmbus_device_id *hv_vmbus_get_id(struct hv_driver *drv,
676                                                         struct hv_device *dev)
677 {
678         const uuid_le *guid = &dev->dev_type;
679         const struct hv_vmbus_device_id *id;
680
681         /* When driver_override is set, only bind to the matching driver */
682         if (dev->driver_override && strcmp(dev->driver_override, drv->name))
683                 return NULL;
684
685         /* Look at the dynamic ids first, before the static ones */
686         id = hv_vmbus_dynid_match(drv, guid);
687         if (!id)
688                 id = hv_vmbus_dev_match(drv->id_table, guid);
689
690         /* driver_override will always match, send a dummy id */
691         if (!id && dev->driver_override)
692                 id = &vmbus_device_null;
693
694         return id;
695 }
696
697 /* vmbus_add_dynid - add a new device ID to this driver and re-probe devices */
698 static int vmbus_add_dynid(struct hv_driver *drv, uuid_le *guid)
699 {
700         struct vmbus_dynid *dynid;
701
702         dynid = kzalloc(sizeof(*dynid), GFP_KERNEL);
703         if (!dynid)
704                 return -ENOMEM;
705
706         dynid->id.guid = *guid;
707
708         spin_lock(&drv->dynids.lock);
709         list_add_tail(&dynid->node, &drv->dynids.list);
710         spin_unlock(&drv->dynids.lock);
711
712         return driver_attach(&drv->driver);
713 }
714
715 static void vmbus_free_dynids(struct hv_driver *drv)
716 {
717         struct vmbus_dynid *dynid, *n;
718
719         spin_lock(&drv->dynids.lock);
720         list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
721                 list_del(&dynid->node);
722                 kfree(dynid);
723         }
724         spin_unlock(&drv->dynids.lock);
725 }
726
727 /*
728  * store_new_id - sysfs frontend to vmbus_add_dynid()
729  *
730  * Allow GUIDs to be added to an existing driver via sysfs.
731  */
732 static ssize_t new_id_store(struct device_driver *driver, const char *buf,
733                             size_t count)
734 {
735         struct hv_driver *drv = drv_to_hv_drv(driver);
736         uuid_le guid;
737         ssize_t retval;
738
739         retval = uuid_le_to_bin(buf, &guid);
740         if (retval)
741                 return retval;
742
743         if (hv_vmbus_dynid_match(drv, &guid))
744                 return -EEXIST;
745
746         retval = vmbus_add_dynid(drv, &guid);
747         if (retval)
748                 return retval;
749         return count;
750 }
751 static DRIVER_ATTR_WO(new_id);
752
753 /*
754  * store_remove_id - remove a PCI device ID from this driver
755  *
756  * Removes a dynamic pci device ID to this driver.
757  */
758 static ssize_t remove_id_store(struct device_driver *driver, const char *buf,
759                                size_t count)
760 {
761         struct hv_driver *drv = drv_to_hv_drv(driver);
762         struct vmbus_dynid *dynid, *n;
763         uuid_le guid;
764         ssize_t retval;
765
766         retval = uuid_le_to_bin(buf, &guid);
767         if (retval)
768                 return retval;
769
770         retval = -ENODEV;
771         spin_lock(&drv->dynids.lock);
772         list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
773                 struct hv_vmbus_device_id *id = &dynid->id;
774
775                 if (!uuid_le_cmp(id->guid, guid)) {
776                         list_del(&dynid->node);
777                         kfree(dynid);
778                         retval = count;
779                         break;
780                 }
781         }
782         spin_unlock(&drv->dynids.lock);
783
784         return retval;
785 }
786 static DRIVER_ATTR_WO(remove_id);
787
788 static struct attribute *vmbus_drv_attrs[] = {
789         &driver_attr_new_id.attr,
790         &driver_attr_remove_id.attr,
791         NULL,
792 };
793 ATTRIBUTE_GROUPS(vmbus_drv);
794
795
796 /*
797  * vmbus_match - Attempt to match the specified device to the specified driver
798  */
799 static int vmbus_match(struct device *device, struct device_driver *driver)
800 {
801         struct hv_driver *drv = drv_to_hv_drv(driver);
802         struct hv_device *hv_dev = device_to_hv_device(device);
803
804         /* The hv_sock driver handles all hv_sock offers. */
805         if (is_hvsock_channel(hv_dev->channel))
806                 return drv->hvsock;
807
808         if (hv_vmbus_get_id(drv, hv_dev))
809                 return 1;
810
811         return 0;
812 }
813
814 /*
815  * vmbus_probe - Add the new vmbus's child device
816  */
817 static int vmbus_probe(struct device *child_device)
818 {
819         int ret = 0;
820         struct hv_driver *drv =
821                         drv_to_hv_drv(child_device->driver);
822         struct hv_device *dev = device_to_hv_device(child_device);
823         const struct hv_vmbus_device_id *dev_id;
824
825         dev_id = hv_vmbus_get_id(drv, dev);
826         if (drv->probe) {
827                 ret = drv->probe(dev, dev_id);
828                 if (ret != 0)
829                         pr_err("probe failed for device %s (%d)\n",
830                                dev_name(child_device), ret);
831
832         } else {
833                 pr_err("probe not set for driver %s\n",
834                        dev_name(child_device));
835                 ret = -ENODEV;
836         }
837         return ret;
838 }
839
840 /*
841  * vmbus_remove - Remove a vmbus device
842  */
843 static int vmbus_remove(struct device *child_device)
844 {
845         struct hv_driver *drv;
846         struct hv_device *dev = device_to_hv_device(child_device);
847
848         if (child_device->driver) {
849                 drv = drv_to_hv_drv(child_device->driver);
850                 if (drv->remove)
851                         drv->remove(dev);
852         }
853
854         return 0;
855 }
856
857
858 /*
859  * vmbus_shutdown - Shutdown a vmbus device
860  */
861 static void vmbus_shutdown(struct device *child_device)
862 {
863         struct hv_driver *drv;
864         struct hv_device *dev = device_to_hv_device(child_device);
865
866
867         /* The device may not be attached yet */
868         if (!child_device->driver)
869                 return;
870
871         drv = drv_to_hv_drv(child_device->driver);
872
873         if (drv->shutdown)
874                 drv->shutdown(dev);
875 }
876
877
878 /*
879  * vmbus_device_release - Final callback release of the vmbus child device
880  */
881 static void vmbus_device_release(struct device *device)
882 {
883         struct hv_device *hv_dev = device_to_hv_device(device);
884         struct vmbus_channel *channel = hv_dev->channel;
885
886         mutex_lock(&vmbus_connection.channel_mutex);
887         hv_process_channel_removal(channel);
888         mutex_unlock(&vmbus_connection.channel_mutex);
889         kfree(hv_dev);
890 }
891
892 /* The one and only one */
893 static struct bus_type  hv_bus = {
894         .name =         "vmbus",
895         .match =                vmbus_match,
896         .shutdown =             vmbus_shutdown,
897         .remove =               vmbus_remove,
898         .probe =                vmbus_probe,
899         .uevent =               vmbus_uevent,
900         .dev_groups =           vmbus_dev_groups,
901         .drv_groups =           vmbus_drv_groups,
902 };
903
904 struct onmessage_work_context {
905         struct work_struct work;
906         struct hv_message msg;
907 };
908
909 static void vmbus_onmessage_work(struct work_struct *work)
910 {
911         struct onmessage_work_context *ctx;
912
913         /* Do not process messages if we're in DISCONNECTED state */
914         if (vmbus_connection.conn_state == DISCONNECTED)
915                 return;
916
917         ctx = container_of(work, struct onmessage_work_context,
918                            work);
919         vmbus_onmessage(&ctx->msg);
920         kfree(ctx);
921 }
922
923 static void hv_process_timer_expiration(struct hv_message *msg,
924                                         struct hv_per_cpu_context *hv_cpu)
925 {
926         struct clock_event_device *dev = hv_cpu->clk_evt;
927
928         if (dev->event_handler)
929                 dev->event_handler(dev);
930
931         vmbus_signal_eom(msg, HVMSG_TIMER_EXPIRED);
932 }
933
934 void vmbus_on_msg_dpc(unsigned long data)
935 {
936         struct hv_per_cpu_context *hv_cpu = (void *)data;
937         void *page_addr = hv_cpu->synic_message_page;
938         struct hv_message *msg = (struct hv_message *)page_addr +
939                                   VMBUS_MESSAGE_SINT;
940         struct vmbus_channel_message_header *hdr;
941         const struct vmbus_channel_message_table_entry *entry;
942         struct onmessage_work_context *ctx;
943         u32 message_type = msg->header.message_type;
944
945         if (message_type == HVMSG_NONE)
946                 /* no msg */
947                 return;
948
949         hdr = (struct vmbus_channel_message_header *)msg->u.payload;
950
951         trace_vmbus_on_msg_dpc(hdr);
952
953         if (hdr->msgtype >= CHANNELMSG_COUNT) {
954                 WARN_ONCE(1, "unknown msgtype=%d\n", hdr->msgtype);
955                 goto msg_handled;
956         }
957
958         entry = &channel_message_table[hdr->msgtype];
959         if (entry->handler_type == VMHT_BLOCKING) {
960                 ctx = kmalloc(sizeof(*ctx), GFP_ATOMIC);
961                 if (ctx == NULL)
962                         return;
963
964                 INIT_WORK(&ctx->work, vmbus_onmessage_work);
965                 memcpy(&ctx->msg, msg, sizeof(*msg));
966
967                 /*
968                  * The host can generate a rescind message while we
969                  * may still be handling the original offer. We deal with
970                  * this condition by ensuring the processing is done on the
971                  * same CPU.
972                  */
973                 switch (hdr->msgtype) {
974                 case CHANNELMSG_RESCIND_CHANNELOFFER:
975                         /*
976                          * If we are handling the rescind message;
977                          * schedule the work on the global work queue.
978                          */
979                         schedule_work_on(vmbus_connection.connect_cpu,
980                                          &ctx->work);
981                         break;
982
983                 case CHANNELMSG_OFFERCHANNEL:
984                         atomic_inc(&vmbus_connection.offer_in_progress);
985                         queue_work_on(vmbus_connection.connect_cpu,
986                                       vmbus_connection.work_queue,
987                                       &ctx->work);
988                         break;
989
990                 default:
991                         queue_work(vmbus_connection.work_queue, &ctx->work);
992                 }
993         } else
994                 entry->message_handler(hdr);
995
996 msg_handled:
997         vmbus_signal_eom(msg, message_type);
998 }
999
1000
1001 /*
1002  * Direct callback for channels using other deferred processing
1003  */
1004 static void vmbus_channel_isr(struct vmbus_channel *channel)
1005 {
1006         void (*callback_fn)(void *);
1007
1008         callback_fn = READ_ONCE(channel->onchannel_callback);
1009         if (likely(callback_fn != NULL))
1010                 (*callback_fn)(channel->channel_callback_context);
1011 }
1012
1013 /*
1014  * Schedule all channels with events pending
1015  */
1016 static void vmbus_chan_sched(struct hv_per_cpu_context *hv_cpu)
1017 {
1018         unsigned long *recv_int_page;
1019         u32 maxbits, relid;
1020
1021         if (vmbus_proto_version < VERSION_WIN8) {
1022                 maxbits = MAX_NUM_CHANNELS_SUPPORTED;
1023                 recv_int_page = vmbus_connection.recv_int_page;
1024         } else {
1025                 /*
1026                  * When the host is win8 and beyond, the event page
1027                  * can be directly checked to get the id of the channel
1028                  * that has the interrupt pending.
1029                  */
1030                 void *page_addr = hv_cpu->synic_event_page;
1031                 union hv_synic_event_flags *event
1032                         = (union hv_synic_event_flags *)page_addr +
1033                                                  VMBUS_MESSAGE_SINT;
1034
1035                 maxbits = HV_EVENT_FLAGS_COUNT;
1036                 recv_int_page = event->flags;
1037         }
1038
1039         if (unlikely(!recv_int_page))
1040                 return;
1041
1042         for_each_set_bit(relid, recv_int_page, maxbits) {
1043                 struct vmbus_channel *channel;
1044
1045                 if (!sync_test_and_clear_bit(relid, recv_int_page))
1046                         continue;
1047
1048                 /* Special case - vmbus channel protocol msg */
1049                 if (relid == 0)
1050                         continue;
1051
1052                 rcu_read_lock();
1053
1054                 /* Find channel based on relid */
1055                 list_for_each_entry_rcu(channel, &hv_cpu->chan_list, percpu_list) {
1056                         if (channel->offermsg.child_relid != relid)
1057                                 continue;
1058
1059                         if (channel->rescind)
1060                                 continue;
1061
1062                         trace_vmbus_chan_sched(channel);
1063
1064                         ++channel->interrupts;
1065
1066                         switch (channel->callback_mode) {
1067                         case HV_CALL_ISR:
1068                                 vmbus_channel_isr(channel);
1069                                 break;
1070
1071                         case HV_CALL_BATCHED:
1072                                 hv_begin_read(&channel->inbound);
1073                                 /* fallthrough */
1074                         case HV_CALL_DIRECT:
1075                                 tasklet_schedule(&channel->callback_event);
1076                         }
1077                 }
1078
1079                 rcu_read_unlock();
1080         }
1081 }
1082
1083 static void vmbus_isr(void)
1084 {
1085         struct hv_per_cpu_context *hv_cpu
1086                 = this_cpu_ptr(hv_context.cpu_context);
1087         void *page_addr = hv_cpu->synic_event_page;
1088         struct hv_message *msg;
1089         union hv_synic_event_flags *event;
1090         bool handled = false;
1091
1092         if (unlikely(page_addr == NULL))
1093                 return;
1094
1095         event = (union hv_synic_event_flags *)page_addr +
1096                                          VMBUS_MESSAGE_SINT;
1097         /*
1098          * Check for events before checking for messages. This is the order
1099          * in which events and messages are checked in Windows guests on
1100          * Hyper-V, and the Windows team suggested we do the same.
1101          */
1102
1103         if ((vmbus_proto_version == VERSION_WS2008) ||
1104                 (vmbus_proto_version == VERSION_WIN7)) {
1105
1106                 /* Since we are a child, we only need to check bit 0 */
1107                 if (sync_test_and_clear_bit(0, event->flags))
1108                         handled = true;
1109         } else {
1110                 /*
1111                  * Our host is win8 or above. The signaling mechanism
1112                  * has changed and we can directly look at the event page.
1113                  * If bit n is set then we have an interrup on the channel
1114                  * whose id is n.
1115                  */
1116                 handled = true;
1117         }
1118
1119         if (handled)
1120                 vmbus_chan_sched(hv_cpu);
1121
1122         page_addr = hv_cpu->synic_message_page;
1123         msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
1124
1125         /* Check if there are actual msgs to be processed */
1126         if (msg->header.message_type != HVMSG_NONE) {
1127                 if (msg->header.message_type == HVMSG_TIMER_EXPIRED)
1128                         hv_process_timer_expiration(msg, hv_cpu);
1129                 else
1130                         tasklet_schedule(&hv_cpu->msg_dpc);
1131         }
1132
1133         add_interrupt_randomness(HYPERVISOR_CALLBACK_VECTOR, 0);
1134 }
1135
1136 /*
1137  * Boolean to control whether to report panic messages over Hyper-V.
1138  *
1139  * It can be set via /proc/sys/kernel/hyperv/record_panic_msg
1140  */
1141 static int sysctl_record_panic_msg = 1;
1142
1143 /*
1144  * Callback from kmsg_dump. Grab as much as possible from the end of the kmsg
1145  * buffer and call into Hyper-V to transfer the data.
1146  */
1147 static void hv_kmsg_dump(struct kmsg_dumper *dumper,
1148                          enum kmsg_dump_reason reason)
1149 {
1150         size_t bytes_written;
1151         phys_addr_t panic_pa;
1152
1153         /* We are only interested in panics. */
1154         if ((reason != KMSG_DUMP_PANIC) || (!sysctl_record_panic_msg))
1155                 return;
1156
1157         panic_pa = virt_to_phys(hv_panic_page);
1158
1159         /*
1160          * Write dump contents to the page. No need to synchronize; panic should
1161          * be single-threaded.
1162          */
1163         kmsg_dump_get_buffer(dumper, true, hv_panic_page, PAGE_SIZE,
1164                              &bytes_written);
1165         if (bytes_written)
1166                 hyperv_report_panic_msg(panic_pa, bytes_written);
1167 }
1168
1169 static struct kmsg_dumper hv_kmsg_dumper = {
1170         .dump = hv_kmsg_dump,
1171 };
1172
1173 static struct ctl_table_header *hv_ctl_table_hdr;
1174 static int zero;
1175 static int one = 1;
1176
1177 /*
1178  * sysctl option to allow the user to control whether kmsg data should be
1179  * reported to Hyper-V on panic.
1180  */
1181 static struct ctl_table hv_ctl_table[] = {
1182         {
1183                 .procname       = "hyperv_record_panic_msg",
1184                 .data           = &sysctl_record_panic_msg,
1185                 .maxlen         = sizeof(int),
1186                 .mode           = 0644,
1187                 .proc_handler   = proc_dointvec_minmax,
1188                 .extra1         = &zero,
1189                 .extra2         = &one
1190         },
1191         {}
1192 };
1193
1194 static struct ctl_table hv_root_table[] = {
1195         {
1196                 .procname       = "kernel",
1197                 .mode           = 0555,
1198                 .child          = hv_ctl_table
1199         },
1200         {}
1201 };
1202
1203 /*
1204  * vmbus_bus_init -Main vmbus driver initialization routine.
1205  *
1206  * Here, we
1207  *      - initialize the vmbus driver context
1208  *      - invoke the vmbus hv main init routine
1209  *      - retrieve the channel offers
1210  */
1211 static int vmbus_bus_init(void)
1212 {
1213         int ret;
1214
1215         /* Hypervisor initialization...setup hypercall page..etc */
1216         ret = hv_init();
1217         if (ret != 0) {
1218                 pr_err("Unable to initialize the hypervisor - 0x%x\n", ret);
1219                 return ret;
1220         }
1221
1222         ret = bus_register(&hv_bus);
1223         if (ret)
1224                 return ret;
1225
1226         hv_setup_vmbus_irq(vmbus_isr);
1227
1228         ret = hv_synic_alloc();
1229         if (ret)
1230                 goto err_alloc;
1231         /*
1232          * Initialize the per-cpu interrupt state and
1233          * connect to the host.
1234          */
1235         ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "hyperv/vmbus:online",
1236                                 hv_synic_init, hv_synic_cleanup);
1237         if (ret < 0)
1238                 goto err_alloc;
1239         hyperv_cpuhp_online = ret;
1240
1241         ret = vmbus_connect();
1242         if (ret)
1243                 goto err_connect;
1244
1245         /*
1246          * Only register if the crash MSRs are available
1247          */
1248         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1249                 u64 hyperv_crash_ctl;
1250                 /*
1251                  * Sysctl registration is not fatal, since by default
1252                  * reporting is enabled.
1253                  */
1254                 hv_ctl_table_hdr = register_sysctl_table(hv_root_table);
1255                 if (!hv_ctl_table_hdr)
1256                         pr_err("Hyper-V: sysctl table register error");
1257
1258                 /*
1259                  * Register for panic kmsg callback only if the right
1260                  * capability is supported by the hypervisor.
1261                  */
1262                 hv_get_crash_ctl(hyperv_crash_ctl);
1263                 if (hyperv_crash_ctl & HV_CRASH_CTL_CRASH_NOTIFY_MSG) {
1264                         hv_panic_page = (void *)get_zeroed_page(GFP_KERNEL);
1265                         if (hv_panic_page) {
1266                                 ret = kmsg_dump_register(&hv_kmsg_dumper);
1267                                 if (ret)
1268                                         pr_err("Hyper-V: kmsg dump register "
1269                                                 "error 0x%x\n", ret);
1270                         } else
1271                                 pr_err("Hyper-V: panic message page memory "
1272                                         "allocation failed");
1273                 }
1274
1275                 register_die_notifier(&hyperv_die_block);
1276                 atomic_notifier_chain_register(&panic_notifier_list,
1277                                                &hyperv_panic_block);
1278         }
1279
1280         vmbus_request_offers();
1281
1282         return 0;
1283
1284 err_connect:
1285         cpuhp_remove_state(hyperv_cpuhp_online);
1286 err_alloc:
1287         hv_synic_free();
1288         hv_remove_vmbus_irq();
1289
1290         bus_unregister(&hv_bus);
1291         free_page((unsigned long)hv_panic_page);
1292         unregister_sysctl_table(hv_ctl_table_hdr);
1293         hv_ctl_table_hdr = NULL;
1294         return ret;
1295 }
1296
1297 /**
1298  * __vmbus_child_driver_register() - Register a vmbus's driver
1299  * @hv_driver: Pointer to driver structure you want to register
1300  * @owner: owner module of the drv
1301  * @mod_name: module name string
1302  *
1303  * Registers the given driver with Linux through the 'driver_register()' call
1304  * and sets up the hyper-v vmbus handling for this driver.
1305  * It will return the state of the 'driver_register()' call.
1306  *
1307  */
1308 int __vmbus_driver_register(struct hv_driver *hv_driver, struct module *owner, const char *mod_name)
1309 {
1310         int ret;
1311
1312         pr_info("registering driver %s\n", hv_driver->name);
1313
1314         ret = vmbus_exists();
1315         if (ret < 0)
1316                 return ret;
1317
1318         hv_driver->driver.name = hv_driver->name;
1319         hv_driver->driver.owner = owner;
1320         hv_driver->driver.mod_name = mod_name;
1321         hv_driver->driver.bus = &hv_bus;
1322
1323         spin_lock_init(&hv_driver->dynids.lock);
1324         INIT_LIST_HEAD(&hv_driver->dynids.list);
1325
1326         ret = driver_register(&hv_driver->driver);
1327
1328         return ret;
1329 }
1330 EXPORT_SYMBOL_GPL(__vmbus_driver_register);
1331
1332 /**
1333  * vmbus_driver_unregister() - Unregister a vmbus's driver
1334  * @hv_driver: Pointer to driver structure you want to
1335  *             un-register
1336  *
1337  * Un-register the given driver that was previous registered with a call to
1338  * vmbus_driver_register()
1339  */
1340 void vmbus_driver_unregister(struct hv_driver *hv_driver)
1341 {
1342         pr_info("unregistering driver %s\n", hv_driver->name);
1343
1344         if (!vmbus_exists()) {
1345                 driver_unregister(&hv_driver->driver);
1346                 vmbus_free_dynids(hv_driver);
1347         }
1348 }
1349 EXPORT_SYMBOL_GPL(vmbus_driver_unregister);
1350
1351
1352 /*
1353  * Called when last reference to channel is gone.
1354  */
1355 static void vmbus_chan_release(struct kobject *kobj)
1356 {
1357         struct vmbus_channel *channel
1358                 = container_of(kobj, struct vmbus_channel, kobj);
1359
1360         kfree_rcu(channel, rcu);
1361 }
1362
1363 struct vmbus_chan_attribute {
1364         struct attribute attr;
1365         ssize_t (*show)(const struct vmbus_channel *chan, char *buf);
1366         ssize_t (*store)(struct vmbus_channel *chan,
1367                          const char *buf, size_t count);
1368 };
1369 #define VMBUS_CHAN_ATTR(_name, _mode, _show, _store) \
1370         struct vmbus_chan_attribute chan_attr_##_name \
1371                 = __ATTR(_name, _mode, _show, _store)
1372 #define VMBUS_CHAN_ATTR_RW(_name) \
1373         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RW(_name)
1374 #define VMBUS_CHAN_ATTR_RO(_name) \
1375         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RO(_name)
1376 #define VMBUS_CHAN_ATTR_WO(_name) \
1377         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_WO(_name)
1378
1379 static ssize_t vmbus_chan_attr_show(struct kobject *kobj,
1380                                     struct attribute *attr, char *buf)
1381 {
1382         const struct vmbus_chan_attribute *attribute
1383                 = container_of(attr, struct vmbus_chan_attribute, attr);
1384         const struct vmbus_channel *chan
1385                 = container_of(kobj, struct vmbus_channel, kobj);
1386
1387         if (!attribute->show)
1388                 return -EIO;
1389
1390         if (chan->state != CHANNEL_OPENED_STATE)
1391                 return -EINVAL;
1392
1393         return attribute->show(chan, buf);
1394 }
1395
1396 static const struct sysfs_ops vmbus_chan_sysfs_ops = {
1397         .show = vmbus_chan_attr_show,
1398 };
1399
1400 static ssize_t out_mask_show(const struct vmbus_channel *channel, char *buf)
1401 {
1402         const struct hv_ring_buffer_info *rbi = &channel->outbound;
1403
1404         return sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1405 }
1406 static VMBUS_CHAN_ATTR_RO(out_mask);
1407
1408 static ssize_t in_mask_show(const struct vmbus_channel *channel, char *buf)
1409 {
1410         const struct hv_ring_buffer_info *rbi = &channel->inbound;
1411
1412         return sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1413 }
1414 static VMBUS_CHAN_ATTR_RO(in_mask);
1415
1416 static ssize_t read_avail_show(const struct vmbus_channel *channel, char *buf)
1417 {
1418         const struct hv_ring_buffer_info *rbi = &channel->inbound;
1419
1420         return sprintf(buf, "%u\n", hv_get_bytes_to_read(rbi));
1421 }
1422 static VMBUS_CHAN_ATTR_RO(read_avail);
1423
1424 static ssize_t write_avail_show(const struct vmbus_channel *channel, char *buf)
1425 {
1426         const struct hv_ring_buffer_info *rbi = &channel->outbound;
1427
1428         return sprintf(buf, "%u\n", hv_get_bytes_to_write(rbi));
1429 }
1430 static VMBUS_CHAN_ATTR_RO(write_avail);
1431
1432 static ssize_t show_target_cpu(const struct vmbus_channel *channel, char *buf)
1433 {
1434         return sprintf(buf, "%u\n", channel->target_cpu);
1435 }
1436 static VMBUS_CHAN_ATTR(cpu, S_IRUGO, show_target_cpu, NULL);
1437
1438 static ssize_t channel_pending_show(const struct vmbus_channel *channel,
1439                                     char *buf)
1440 {
1441         return sprintf(buf, "%d\n",
1442                        channel_pending(channel,
1443                                        vmbus_connection.monitor_pages[1]));
1444 }
1445 static VMBUS_CHAN_ATTR(pending, S_IRUGO, channel_pending_show, NULL);
1446
1447 static ssize_t channel_latency_show(const struct vmbus_channel *channel,
1448                                     char *buf)
1449 {
1450         return sprintf(buf, "%d\n",
1451                        channel_latency(channel,
1452                                        vmbus_connection.monitor_pages[1]));
1453 }
1454 static VMBUS_CHAN_ATTR(latency, S_IRUGO, channel_latency_show, NULL);
1455
1456 static ssize_t channel_interrupts_show(const struct vmbus_channel *channel, char *buf)
1457 {
1458         return sprintf(buf, "%llu\n", channel->interrupts);
1459 }
1460 static VMBUS_CHAN_ATTR(interrupts, S_IRUGO, channel_interrupts_show, NULL);
1461
1462 static ssize_t channel_events_show(const struct vmbus_channel *channel, char *buf)
1463 {
1464         return sprintf(buf, "%llu\n", channel->sig_events);
1465 }
1466 static VMBUS_CHAN_ATTR(events, S_IRUGO, channel_events_show, NULL);
1467
1468 static ssize_t subchannel_monitor_id_show(const struct vmbus_channel *channel,
1469                                           char *buf)
1470 {
1471         return sprintf(buf, "%u\n", channel->offermsg.monitorid);
1472 }
1473 static VMBUS_CHAN_ATTR(monitor_id, S_IRUGO, subchannel_monitor_id_show, NULL);
1474
1475 static ssize_t subchannel_id_show(const struct vmbus_channel *channel,
1476                                   char *buf)
1477 {
1478         return sprintf(buf, "%u\n",
1479                        channel->offermsg.offer.sub_channel_index);
1480 }
1481 static VMBUS_CHAN_ATTR_RO(subchannel_id);
1482
1483 static struct attribute *vmbus_chan_attrs[] = {
1484         &chan_attr_out_mask.attr,
1485         &chan_attr_in_mask.attr,
1486         &chan_attr_read_avail.attr,
1487         &chan_attr_write_avail.attr,
1488         &chan_attr_cpu.attr,
1489         &chan_attr_pending.attr,
1490         &chan_attr_latency.attr,
1491         &chan_attr_interrupts.attr,
1492         &chan_attr_events.attr,
1493         &chan_attr_monitor_id.attr,
1494         &chan_attr_subchannel_id.attr,
1495         NULL
1496 };
1497
1498 static struct kobj_type vmbus_chan_ktype = {
1499         .sysfs_ops = &vmbus_chan_sysfs_ops,
1500         .release = vmbus_chan_release,
1501         .default_attrs = vmbus_chan_attrs,
1502 };
1503
1504 /*
1505  * vmbus_add_channel_kobj - setup a sub-directory under device/channels
1506  */
1507 int vmbus_add_channel_kobj(struct hv_device *dev, struct vmbus_channel *channel)
1508 {
1509         struct kobject *kobj = &channel->kobj;
1510         u32 relid = channel->offermsg.child_relid;
1511         int ret;
1512
1513         kobj->kset = dev->channels_kset;
1514         ret = kobject_init_and_add(kobj, &vmbus_chan_ktype, NULL,
1515                                    "%u", relid);
1516         if (ret)
1517                 return ret;
1518
1519         kobject_uevent(kobj, KOBJ_ADD);
1520
1521         return 0;
1522 }
1523
1524 /*
1525  * vmbus_device_create - Creates and registers a new child device
1526  * on the vmbus.
1527  */
1528 struct hv_device *vmbus_device_create(const uuid_le *type,
1529                                       const uuid_le *instance,
1530                                       struct vmbus_channel *channel)
1531 {
1532         struct hv_device *child_device_obj;
1533
1534         child_device_obj = kzalloc(sizeof(struct hv_device), GFP_KERNEL);
1535         if (!child_device_obj) {
1536                 pr_err("Unable to allocate device object for child device\n");
1537                 return NULL;
1538         }
1539
1540         child_device_obj->channel = channel;
1541         memcpy(&child_device_obj->dev_type, type, sizeof(uuid_le));
1542         memcpy(&child_device_obj->dev_instance, instance,
1543                sizeof(uuid_le));
1544         child_device_obj->vendor_id = 0x1414; /* MSFT vendor ID */
1545
1546
1547         return child_device_obj;
1548 }
1549
1550 /*
1551  * vmbus_device_register - Register the child device
1552  */
1553 int vmbus_device_register(struct hv_device *child_device_obj)
1554 {
1555         struct kobject *kobj = &child_device_obj->device.kobj;
1556         int ret;
1557
1558         dev_set_name(&child_device_obj->device, "%pUl",
1559                      child_device_obj->channel->offermsg.offer.if_instance.b);
1560
1561         child_device_obj->device.bus = &hv_bus;
1562         child_device_obj->device.parent = &hv_acpi_dev->dev;
1563         child_device_obj->device.release = vmbus_device_release;
1564
1565         /*
1566          * Register with the LDM. This will kick off the driver/device
1567          * binding...which will eventually call vmbus_match() and vmbus_probe()
1568          */
1569         ret = device_register(&child_device_obj->device);
1570         if (ret) {
1571                 pr_err("Unable to register child device\n");
1572                 return ret;
1573         }
1574
1575         child_device_obj->channels_kset = kset_create_and_add("channels",
1576                                                               NULL, kobj);
1577         if (!child_device_obj->channels_kset) {
1578                 ret = -ENOMEM;
1579                 goto err_dev_unregister;
1580         }
1581
1582         ret = vmbus_add_channel_kobj(child_device_obj,
1583                                      child_device_obj->channel);
1584         if (ret) {
1585                 pr_err("Unable to register primary channeln");
1586                 goto err_kset_unregister;
1587         }
1588
1589         return 0;
1590
1591 err_kset_unregister:
1592         kset_unregister(child_device_obj->channels_kset);
1593
1594 err_dev_unregister:
1595         device_unregister(&child_device_obj->device);
1596         return ret;
1597 }
1598
1599 /*
1600  * vmbus_device_unregister - Remove the specified child device
1601  * from the vmbus.
1602  */
1603 void vmbus_device_unregister(struct hv_device *device_obj)
1604 {
1605         pr_debug("child device %s unregistered\n",
1606                 dev_name(&device_obj->device));
1607
1608         kset_unregister(device_obj->channels_kset);
1609
1610         /*
1611          * Kick off the process of unregistering the device.
1612          * This will call vmbus_remove() and eventually vmbus_device_release()
1613          */
1614         device_unregister(&device_obj->device);
1615 }
1616
1617
1618 /*
1619  * VMBUS is an acpi enumerated device. Get the information we
1620  * need from DSDT.
1621  */
1622 #define VTPM_BASE_ADDRESS 0xfed40000
1623 static acpi_status vmbus_walk_resources(struct acpi_resource *res, void *ctx)
1624 {
1625         resource_size_t start = 0;
1626         resource_size_t end = 0;
1627         struct resource *new_res;
1628         struct resource **old_res = &hyperv_mmio;
1629         struct resource **prev_res = NULL;
1630
1631         switch (res->type) {
1632
1633         /*
1634          * "Address" descriptors are for bus windows. Ignore
1635          * "memory" descriptors, which are for registers on
1636          * devices.
1637          */
1638         case ACPI_RESOURCE_TYPE_ADDRESS32:
1639                 start = res->data.address32.address.minimum;
1640                 end = res->data.address32.address.maximum;
1641                 break;
1642
1643         case ACPI_RESOURCE_TYPE_ADDRESS64:
1644                 start = res->data.address64.address.minimum;
1645                 end = res->data.address64.address.maximum;
1646                 break;
1647
1648         default:
1649                 /* Unused resource type */
1650                 return AE_OK;
1651
1652         }
1653         /*
1654          * Ignore ranges that are below 1MB, as they're not
1655          * necessary or useful here.
1656          */
1657         if (end < 0x100000)
1658                 return AE_OK;
1659
1660         new_res = kzalloc(sizeof(*new_res), GFP_ATOMIC);
1661         if (!new_res)
1662                 return AE_NO_MEMORY;
1663
1664         /* If this range overlaps the virtual TPM, truncate it. */
1665         if (end > VTPM_BASE_ADDRESS && start < VTPM_BASE_ADDRESS)
1666                 end = VTPM_BASE_ADDRESS;
1667
1668         new_res->name = "hyperv mmio";
1669         new_res->flags = IORESOURCE_MEM;
1670         new_res->start = start;
1671         new_res->end = end;
1672
1673         /*
1674          * If two ranges are adjacent, merge them.
1675          */
1676         do {
1677                 if (!*old_res) {
1678                         *old_res = new_res;
1679                         break;
1680                 }
1681
1682                 if (((*old_res)->end + 1) == new_res->start) {
1683                         (*old_res)->end = new_res->end;
1684                         kfree(new_res);
1685                         break;
1686                 }
1687
1688                 if ((*old_res)->start == new_res->end + 1) {
1689                         (*old_res)->start = new_res->start;
1690                         kfree(new_res);
1691                         break;
1692                 }
1693
1694                 if ((*old_res)->start > new_res->end) {
1695                         new_res->sibling = *old_res;
1696                         if (prev_res)
1697                                 (*prev_res)->sibling = new_res;
1698                         *old_res = new_res;
1699                         break;
1700                 }
1701
1702                 prev_res = old_res;
1703                 old_res = &(*old_res)->sibling;
1704
1705         } while (1);
1706
1707         return AE_OK;
1708 }
1709
1710 static int vmbus_acpi_remove(struct acpi_device *device)
1711 {
1712         struct resource *cur_res;
1713         struct resource *next_res;
1714
1715         if (hyperv_mmio) {
1716                 if (fb_mmio) {
1717                         __release_region(hyperv_mmio, fb_mmio->start,
1718                                          resource_size(fb_mmio));
1719                         fb_mmio = NULL;
1720                 }
1721
1722                 for (cur_res = hyperv_mmio; cur_res; cur_res = next_res) {
1723                         next_res = cur_res->sibling;
1724                         kfree(cur_res);
1725                 }
1726         }
1727
1728         return 0;
1729 }
1730
1731 static void vmbus_reserve_fb(void)
1732 {
1733         int size;
1734         /*
1735          * Make a claim for the frame buffer in the resource tree under the
1736          * first node, which will be the one below 4GB.  The length seems to
1737          * be underreported, particularly in a Generation 1 VM.  So start out
1738          * reserving a larger area and make it smaller until it succeeds.
1739          */
1740
1741         if (screen_info.lfb_base) {
1742                 if (efi_enabled(EFI_BOOT))
1743                         size = max_t(__u32, screen_info.lfb_size, 0x800000);
1744                 else
1745                         size = max_t(__u32, screen_info.lfb_size, 0x4000000);
1746
1747                 for (; !fb_mmio && (size >= 0x100000); size >>= 1) {
1748                         fb_mmio = __request_region(hyperv_mmio,
1749                                                    screen_info.lfb_base, size,
1750                                                    fb_mmio_name, 0);
1751                 }
1752         }
1753 }
1754
1755 /**
1756  * vmbus_allocate_mmio() - Pick a memory-mapped I/O range.
1757  * @new:                If successful, supplied a pointer to the
1758  *                      allocated MMIO space.
1759  * @device_obj:         Identifies the caller
1760  * @min:                Minimum guest physical address of the
1761  *                      allocation
1762  * @max:                Maximum guest physical address
1763  * @size:               Size of the range to be allocated
1764  * @align:              Alignment of the range to be allocated
1765  * @fb_overlap_ok:      Whether this allocation can be allowed
1766  *                      to overlap the video frame buffer.
1767  *
1768  * This function walks the resources granted to VMBus by the
1769  * _CRS object in the ACPI namespace underneath the parent
1770  * "bridge" whether that's a root PCI bus in the Generation 1
1771  * case or a Module Device in the Generation 2 case.  It then
1772  * attempts to allocate from the global MMIO pool in a way that
1773  * matches the constraints supplied in these parameters and by
1774  * that _CRS.
1775  *
1776  * Return: 0 on success, -errno on failure
1777  */
1778 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
1779                         resource_size_t min, resource_size_t max,
1780                         resource_size_t size, resource_size_t align,
1781                         bool fb_overlap_ok)
1782 {
1783         struct resource *iter, *shadow;
1784         resource_size_t range_min, range_max, start;
1785         const char *dev_n = dev_name(&device_obj->device);
1786         int retval;
1787
1788         retval = -ENXIO;
1789         down(&hyperv_mmio_lock);
1790
1791         /*
1792          * If overlaps with frame buffers are allowed, then first attempt to
1793          * make the allocation from within the reserved region.  Because it
1794          * is already reserved, no shadow allocation is necessary.
1795          */
1796         if (fb_overlap_ok && fb_mmio && !(min > fb_mmio->end) &&
1797             !(max < fb_mmio->start)) {
1798
1799                 range_min = fb_mmio->start;
1800                 range_max = fb_mmio->end;
1801                 start = (range_min + align - 1) & ~(align - 1);
1802                 for (; start + size - 1 <= range_max; start += align) {
1803                         *new = request_mem_region_exclusive(start, size, dev_n);
1804                         if (*new) {
1805                                 retval = 0;
1806                                 goto exit;
1807                         }
1808                 }
1809         }
1810
1811         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
1812                 if ((iter->start >= max) || (iter->end <= min))
1813                         continue;
1814
1815                 range_min = iter->start;
1816                 range_max = iter->end;
1817                 start = (range_min + align - 1) & ~(align - 1);
1818                 for (; start + size - 1 <= range_max; start += align) {
1819                         shadow = __request_region(iter, start, size, NULL,
1820                                                   IORESOURCE_BUSY);
1821                         if (!shadow)
1822                                 continue;
1823
1824                         *new = request_mem_region_exclusive(start, size, dev_n);
1825                         if (*new) {
1826                                 shadow->name = (char *)*new;
1827                                 retval = 0;
1828                                 goto exit;
1829                         }
1830
1831                         __release_region(iter, start, size);
1832                 }
1833         }
1834
1835 exit:
1836         up(&hyperv_mmio_lock);
1837         return retval;
1838 }
1839 EXPORT_SYMBOL_GPL(vmbus_allocate_mmio);
1840
1841 /**
1842  * vmbus_free_mmio() - Free a memory-mapped I/O range.
1843  * @start:              Base address of region to release.
1844  * @size:               Size of the range to be allocated
1845  *
1846  * This function releases anything requested by
1847  * vmbus_mmio_allocate().
1848  */
1849 void vmbus_free_mmio(resource_size_t start, resource_size_t size)
1850 {
1851         struct resource *iter;
1852
1853         down(&hyperv_mmio_lock);
1854         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
1855                 if ((iter->start >= start + size) || (iter->end <= start))
1856                         continue;
1857
1858                 __release_region(iter, start, size);
1859         }
1860         release_mem_region(start, size);
1861         up(&hyperv_mmio_lock);
1862
1863 }
1864 EXPORT_SYMBOL_GPL(vmbus_free_mmio);
1865
1866 static int vmbus_acpi_add(struct acpi_device *device)
1867 {
1868         acpi_status result;
1869         int ret_val = -ENODEV;
1870         struct acpi_device *ancestor;
1871
1872         hv_acpi_dev = device;
1873
1874         result = acpi_walk_resources(device->handle, METHOD_NAME__CRS,
1875                                         vmbus_walk_resources, NULL);
1876
1877         if (ACPI_FAILURE(result))
1878                 goto acpi_walk_err;
1879         /*
1880          * Some ancestor of the vmbus acpi device (Gen1 or Gen2
1881          * firmware) is the VMOD that has the mmio ranges. Get that.
1882          */
1883         for (ancestor = device->parent; ancestor; ancestor = ancestor->parent) {
1884                 result = acpi_walk_resources(ancestor->handle, METHOD_NAME__CRS,
1885                                              vmbus_walk_resources, NULL);
1886
1887                 if (ACPI_FAILURE(result))
1888                         continue;
1889                 if (hyperv_mmio) {
1890                         vmbus_reserve_fb();
1891                         break;
1892                 }
1893         }
1894         ret_val = 0;
1895
1896 acpi_walk_err:
1897         complete(&probe_event);
1898         if (ret_val)
1899                 vmbus_acpi_remove(device);
1900         return ret_val;
1901 }
1902
1903 static const struct acpi_device_id vmbus_acpi_device_ids[] = {
1904         {"VMBUS", 0},
1905         {"VMBus", 0},
1906         {"", 0},
1907 };
1908 MODULE_DEVICE_TABLE(acpi, vmbus_acpi_device_ids);
1909
1910 static struct acpi_driver vmbus_acpi_driver = {
1911         .name = "vmbus",
1912         .ids = vmbus_acpi_device_ids,
1913         .ops = {
1914                 .add = vmbus_acpi_add,
1915                 .remove = vmbus_acpi_remove,
1916         },
1917 };
1918
1919 static void hv_kexec_handler(void)
1920 {
1921         hv_synic_clockevents_cleanup();
1922         vmbus_initiate_unload(false);
1923         vmbus_connection.conn_state = DISCONNECTED;
1924         /* Make sure conn_state is set as hv_synic_cleanup checks for it */
1925         mb();
1926         cpuhp_remove_state(hyperv_cpuhp_online);
1927         hyperv_cleanup();
1928 };
1929
1930 static void hv_crash_handler(struct pt_regs *regs)
1931 {
1932         vmbus_initiate_unload(true);
1933         /*
1934          * In crash handler we can't schedule synic cleanup for all CPUs,
1935          * doing the cleanup for current CPU only. This should be sufficient
1936          * for kdump.
1937          */
1938         vmbus_connection.conn_state = DISCONNECTED;
1939         hv_synic_cleanup(smp_processor_id());
1940         hyperv_cleanup();
1941 };
1942
1943 static int __init hv_acpi_init(void)
1944 {
1945         int ret, t;
1946
1947         if (!hv_is_hyperv_initialized())
1948                 return -ENODEV;
1949
1950         init_completion(&probe_event);
1951
1952         /*
1953          * Get ACPI resources first.
1954          */
1955         ret = acpi_bus_register_driver(&vmbus_acpi_driver);
1956
1957         if (ret)
1958                 return ret;
1959
1960         t = wait_for_completion_timeout(&probe_event, 5*HZ);
1961         if (t == 0) {
1962                 ret = -ETIMEDOUT;
1963                 goto cleanup;
1964         }
1965
1966         ret = vmbus_bus_init();
1967         if (ret)
1968                 goto cleanup;
1969
1970         hv_setup_kexec_handler(hv_kexec_handler);
1971         hv_setup_crash_handler(hv_crash_handler);
1972
1973         return 0;
1974
1975 cleanup:
1976         acpi_bus_unregister_driver(&vmbus_acpi_driver);
1977         hv_acpi_dev = NULL;
1978         return ret;
1979 }
1980
1981 static void __exit vmbus_exit(void)
1982 {
1983         int cpu;
1984
1985         hv_remove_kexec_handler();
1986         hv_remove_crash_handler();
1987         vmbus_connection.conn_state = DISCONNECTED;
1988         hv_synic_clockevents_cleanup();
1989         vmbus_disconnect();
1990         hv_remove_vmbus_irq();
1991         for_each_online_cpu(cpu) {
1992                 struct hv_per_cpu_context *hv_cpu
1993                         = per_cpu_ptr(hv_context.cpu_context, cpu);
1994
1995                 tasklet_kill(&hv_cpu->msg_dpc);
1996         }
1997         vmbus_free_channels();
1998
1999         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
2000                 kmsg_dump_unregister(&hv_kmsg_dumper);
2001                 unregister_die_notifier(&hyperv_die_block);
2002                 atomic_notifier_chain_unregister(&panic_notifier_list,
2003                                                  &hyperv_panic_block);
2004         }
2005
2006         free_page((unsigned long)hv_panic_page);
2007         unregister_sysctl_table(hv_ctl_table_hdr);
2008         hv_ctl_table_hdr = NULL;
2009         bus_unregister(&hv_bus);
2010
2011         cpuhp_remove_state(hyperv_cpuhp_online);
2012         hv_synic_free();
2013         acpi_bus_unregister_driver(&vmbus_acpi_driver);
2014 }
2015
2016
2017 MODULE_LICENSE("GPL");
2018
2019 subsys_initcall(hv_acpi_init);
2020 module_exit(vmbus_exit);