drivers/hv/vmbus_drv.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (c) 2009, Microsoft Corporation.
   4  *
   5  * Authors:
   6  *   Haiyang Zhang <haiyangz@microsoft.com>
   7  *   Hank Janssen  <hjanssen@microsoft.com>
   8  *   K. Y. Srinivasan <kys@microsoft.com>
   9  */
  10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  11
  12 #include <linux/init.h>
  13 #include <linux/module.h>
  14 #include <linux/device.h>
  15 #include <linux/interrupt.h>
  16 #include <linux/sysctl.h>
  17 #include <linux/slab.h>
  18 #include <linux/acpi.h>
  19 #include <linux/completion.h>
  20 #include <linux/hyperv.h>
  21 #include <linux/kernel_stat.h>
  22 #include <linux/clockchips.h>
  23 #include <linux/cpu.h>
  24 #include <linux/sched/task_stack.h>
  25
  26 #include <asm/mshyperv.h>
  27 #include <linux/notifier.h>
  28 #include <linux/ptrace.h>
  29 #include <linux/screen_info.h>
  30 #include <linux/kdebug.h>
  31 #include <linux/efi.h>
  32 #include <linux/random.h>
  33 #include <clocksource/hyperv_timer.h>
  34 #include "hyperv_vmbus.h"
  35
  36 struct vmbus_dynid {
  37         struct list_head node;
  38         struct hv_vmbus_device_id id;
  39 };
  40
  41 static struct acpi_device  *hv_acpi_dev;
  42
  43 static struct completion probe_event;
  44
  45 static int hyperv_cpuhp_online;
  46
  47 static void *hv_panic_page;
  48
  49 static int hyperv_panic_event(struct notifier_block *nb, unsigned long val,
  50                               void *args)
  51 {
  52         struct pt_regs *regs;
  53
  54         regs = current_pt_regs();
  55
  56         hyperv_report_panic(regs, val);
  57         return NOTIFY_DONE;
  58 }
  59
  60 static int hyperv_die_event(struct notifier_block *nb, unsigned long val,
  61                             void *args)
  62 {
  63         struct die_args *die = (struct die_args *)args;
  64         struct pt_regs *regs = die->regs;
  65
  66         hyperv_report_panic(regs, val);
  67         return NOTIFY_DONE;
  68 }
  69
  70 static struct notifier_block hyperv_die_block = {
  71         .notifier_call = hyperv_die_event,
  72 };
  73 static struct notifier_block hyperv_panic_block = {
  74         .notifier_call = hyperv_panic_event,
  75 };
  76
  77 static const char *fb_mmio_name = "fb_range";
  78 static struct resource *fb_mmio;
  79 static struct resource *hyperv_mmio;
  80 static DEFINE_SEMAPHORE(hyperv_mmio_lock);
  81
  82 static int vmbus_exists(void)
  83 {
  84         if (hv_acpi_dev == NULL)
  85                 return -ENODEV;
  86
  87         return 0;
  88 }
  89
  90 #define VMBUS_ALIAS_LEN ((sizeof((struct hv_vmbus_device_id *)0)->guid) * 2)
  91 static void print_alias_name(struct hv_device *hv_dev, char *alias_name)
  92 {
  93         int i;
  94         for (i = 0; i < VMBUS_ALIAS_LEN; i += 2)
  95                 sprintf(&alias_name[i], "%02x", hv_dev->dev_type.b[i/2]);
  96 }
  97
  98 static u8 channel_monitor_group(const struct vmbus_channel *channel)
  99 {
 100         return (u8)channel->offermsg.monitorid / 32;
 101 }
 102
 103 static u8 channel_monitor_offset(const struct vmbus_channel *channel)
 104 {
 105         return (u8)channel->offermsg.monitorid % 32;
 106 }
 107
 108 static u32 channel_pending(const struct vmbus_channel *channel,
 109                            const struct hv_monitor_page *monitor_page)
 110 {
 111         u8 monitor_group = channel_monitor_group(channel);
 112
 113         return monitor_page->trigger_group[monitor_group].pending;
 114 }
 115
 116 static u32 channel_latency(const struct vmbus_channel *channel,
 117                            const struct hv_monitor_page *monitor_page)
 118 {
 119         u8 monitor_group = channel_monitor_group(channel);
 120         u8 monitor_offset = channel_monitor_offset(channel);
 121
 122         return monitor_page->latency[monitor_group][monitor_offset];
 123 }
 124
 125 static u32 channel_conn_id(struct vmbus_channel *channel,
 126                            struct hv_monitor_page *monitor_page)
 127 {
 128         u8 monitor_group = channel_monitor_group(channel);
 129         u8 monitor_offset = channel_monitor_offset(channel);
 130         return monitor_page->parameter[monitor_group][monitor_offset].connectionid.u.id;
 131 }
 132
 133 static ssize_t id_show(struct device *dev, struct device_attribute *dev_attr,
 134                        char *buf)
 135 {
 136         struct hv_device *hv_dev = device_to_hv_device(dev);
 137
 138         if (!hv_dev->channel)
 139                 return -ENODEV;
 140         return sprintf(buf, "%d\n", hv_dev->channel->offermsg.child_relid);
 141 }
 142 static DEVICE_ATTR_RO(id);
 143
 144 static ssize_t state_show(struct device *dev, struct device_attribute *dev_attr,
 145                           char *buf)
 146 {
 147         struct hv_device *hv_dev = device_to_hv_device(dev);
 148
 149         if (!hv_dev->channel)
 150                 return -ENODEV;
 151         return sprintf(buf, "%d\n", hv_dev->channel->state);
 152 }
 153 static DEVICE_ATTR_RO(state);
 154
 155 static ssize_t monitor_id_show(struct device *dev,
 156                                struct device_attribute *dev_attr, char *buf)
 157 {
 158         struct hv_device *hv_dev = device_to_hv_device(dev);
 159
 160         if (!hv_dev->channel)
 161                 return -ENODEV;
 162         return sprintf(buf, "%d\n", hv_dev->channel->offermsg.monitorid);
 163 }
 164 static DEVICE_ATTR_RO(monitor_id);
 165
 166 static ssize_t class_id_show(struct device *dev,
 167                                struct device_attribute *dev_attr, char *buf)
 168 {
 169         struct hv_device *hv_dev = device_to_hv_device(dev);
 170
 171         if (!hv_dev->channel)
 172                 return -ENODEV;
 173         return sprintf(buf, "{%pUl}\n",
 174                        hv_dev->channel->offermsg.offer.if_type.b);
 175 }
 176 static DEVICE_ATTR_RO(class_id);
 177
 178 static ssize_t device_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_instance.b);
 187 }
 188 static DEVICE_ATTR_RO(device_id);
 189
 190 static ssize_t modalias_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         char alias_name[VMBUS_ALIAS_LEN + 1];
 195
 196         print_alias_name(hv_dev, alias_name);
 197         return sprintf(buf, "vmbus:%s\n", alias_name);
 198 }
 199 static DEVICE_ATTR_RO(modalias);
 200
 201 #ifdef CONFIG_NUMA
 202 static ssize_t numa_node_show(struct device *dev,
 203                               struct device_attribute *attr, char *buf)
 204 {
 205         struct hv_device *hv_dev = device_to_hv_device(dev);
 206
 207         if (!hv_dev->channel)
 208                 return -ENODEV;
 209
 210         return sprintf(buf, "%d\n", hv_dev->channel->numa_node);
 211 }
 212 static DEVICE_ATTR_RO(numa_node);
 213 #endif
 214
 215 static ssize_t server_monitor_pending_show(struct device *dev,
 216                                            struct device_attribute *dev_attr,
 217                                            char *buf)
 218 {
 219         struct hv_device *hv_dev = device_to_hv_device(dev);
 220
 221         if (!hv_dev->channel)
 222                 return -ENODEV;
 223         return sprintf(buf, "%d\n",
 224                        channel_pending(hv_dev->channel,
 225                                        vmbus_connection.monitor_pages[0]));
 226 }
 227 static DEVICE_ATTR_RO(server_monitor_pending);
 228
 229 static ssize_t client_monitor_pending_show(struct device *dev,
 230                                            struct device_attribute *dev_attr,
 231                                            char *buf)
 232 {
 233         struct hv_device *hv_dev = device_to_hv_device(dev);
 234
 235         if (!hv_dev->channel)
 236                 return -ENODEV;
 237         return sprintf(buf, "%d\n",
 238                        channel_pending(hv_dev->channel,
 239                                        vmbus_connection.monitor_pages[1]));
 240 }
 241 static DEVICE_ATTR_RO(client_monitor_pending);
 242
 243 static ssize_t server_monitor_latency_show(struct device *dev,
 244                                            struct device_attribute *dev_attr,
 245                                            char *buf)
 246 {
 247         struct hv_device *hv_dev = device_to_hv_device(dev);
 248
 249         if (!hv_dev->channel)
 250                 return -ENODEV;
 251         return sprintf(buf, "%d\n",
 252                        channel_latency(hv_dev->channel,
 253                                        vmbus_connection.monitor_pages[0]));
 254 }
 255 static DEVICE_ATTR_RO(server_monitor_latency);
 256
 257 static ssize_t client_monitor_latency_show(struct device *dev,
 258                                            struct device_attribute *dev_attr,
 259                                            char *buf)
 260 {
 261         struct hv_device *hv_dev = device_to_hv_device(dev);
 262
 263         if (!hv_dev->channel)
 264                 return -ENODEV;
 265         return sprintf(buf, "%d\n",
 266                        channel_latency(hv_dev->channel,
 267                                        vmbus_connection.monitor_pages[1]));
 268 }
 269 static DEVICE_ATTR_RO(client_monitor_latency);
 270
 271 static ssize_t server_monitor_conn_id_show(struct device *dev,
 272                                            struct device_attribute *dev_attr,
 273                                            char *buf)
 274 {
 275         struct hv_device *hv_dev = device_to_hv_device(dev);
 276
 277         if (!hv_dev->channel)
 278                 return -ENODEV;
 279         return sprintf(buf, "%d\n",
 280                        channel_conn_id(hv_dev->channel,
 281                                        vmbus_connection.monitor_pages[0]));
 282 }
 283 static DEVICE_ATTR_RO(server_monitor_conn_id);
 284
 285 static ssize_t client_monitor_conn_id_show(struct device *dev,
 286                                            struct device_attribute *dev_attr,
 287                                            char *buf)
 288 {
 289         struct hv_device *hv_dev = device_to_hv_device(dev);
 290
 291         if (!hv_dev->channel)
 292                 return -ENODEV;
 293         return sprintf(buf, "%d\n",
 294                        channel_conn_id(hv_dev->channel,
 295                                        vmbus_connection.monitor_pages[1]));
 296 }
 297 static DEVICE_ATTR_RO(client_monitor_conn_id);
 298
 299 static ssize_t out_intr_mask_show(struct device *dev,
 300                                   struct device_attribute *dev_attr, char *buf)
 301 {
 302         struct hv_device *hv_dev = device_to_hv_device(dev);
 303         struct hv_ring_buffer_debug_info outbound;
 304         int ret;
 305
 306         if (!hv_dev->channel)
 307                 return -ENODEV;
 308
 309         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 310                                           &outbound);
 311         if (ret < 0)
 312                 return ret;
 313
 314         return sprintf(buf, "%d\n", outbound.current_interrupt_mask);
 315 }
 316 static DEVICE_ATTR_RO(out_intr_mask);
 317
 318 static ssize_t out_read_index_show(struct device *dev,
 319                                    struct device_attribute *dev_attr, char *buf)
 320 {
 321         struct hv_device *hv_dev = device_to_hv_device(dev);
 322         struct hv_ring_buffer_debug_info outbound;
 323         int ret;
 324
 325         if (!hv_dev->channel)
 326                 return -ENODEV;
 327
 328         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 329                                           &outbound);
 330         if (ret < 0)
 331                 return ret;
 332         return sprintf(buf, "%d\n", outbound.current_read_index);
 333 }
 334 static DEVICE_ATTR_RO(out_read_index);
 335
 336 static ssize_t out_write_index_show(struct device *dev,
 337                                     struct device_attribute *dev_attr,
 338                                     char *buf)
 339 {
 340         struct hv_device *hv_dev = device_to_hv_device(dev);
 341         struct hv_ring_buffer_debug_info outbound;
 342         int ret;
 343
 344         if (!hv_dev->channel)
 345                 return -ENODEV;
 346
 347         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 348                                           &outbound);
 349         if (ret < 0)
 350                 return ret;
 351         return sprintf(buf, "%d\n", outbound.current_write_index);
 352 }
 353 static DEVICE_ATTR_RO(out_write_index);
 354
 355 static ssize_t out_read_bytes_avail_show(struct device *dev,
 356                                          struct device_attribute *dev_attr,
 357                                          char *buf)
 358 {
 359         struct hv_device *hv_dev = device_to_hv_device(dev);
 360         struct hv_ring_buffer_debug_info outbound;
 361         int ret;
 362
 363         if (!hv_dev->channel)
 364                 return -ENODEV;
 365
 366         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 367                                           &outbound);
 368         if (ret < 0)
 369                 return ret;
 370         return sprintf(buf, "%d\n", outbound.bytes_avail_toread);
 371 }
 372 static DEVICE_ATTR_RO(out_read_bytes_avail);
 373
 374 static ssize_t out_write_bytes_avail_show(struct device *dev,
 375                                           struct device_attribute *dev_attr,
 376                                           char *buf)
 377 {
 378         struct hv_device *hv_dev = device_to_hv_device(dev);
 379         struct hv_ring_buffer_debug_info outbound;
 380         int ret;
 381
 382         if (!hv_dev->channel)
 383                 return -ENODEV;
 384
 385         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 386                                           &outbound);
 387         if (ret < 0)
 388                 return ret;
 389         return sprintf(buf, "%d\n", outbound.bytes_avail_towrite);
 390 }
 391 static DEVICE_ATTR_RO(out_write_bytes_avail);
 392
 393 static ssize_t in_intr_mask_show(struct device *dev,
 394                                  struct device_attribute *dev_attr, char *buf)
 395 {
 396         struct hv_device *hv_dev = device_to_hv_device(dev);
 397         struct hv_ring_buffer_debug_info inbound;
 398         int ret;
 399
 400         if (!hv_dev->channel)
 401                 return -ENODEV;
 402
 403         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 404         if (ret < 0)
 405                 return ret;
 406
 407         return sprintf(buf, "%d\n", inbound.current_interrupt_mask);
 408 }
 409 static DEVICE_ATTR_RO(in_intr_mask);
 410
 411 static ssize_t in_read_index_show(struct device *dev,
 412                                   struct device_attribute *dev_attr, char *buf)
 413 {
 414         struct hv_device *hv_dev = device_to_hv_device(dev);
 415         struct hv_ring_buffer_debug_info inbound;
 416         int ret;
 417
 418         if (!hv_dev->channel)
 419                 return -ENODEV;
 420
 421         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 422         if (ret < 0)
 423                 return ret;
 424
 425         return sprintf(buf, "%d\n", inbound.current_read_index);
 426 }
 427 static DEVICE_ATTR_RO(in_read_index);
 428
 429 static ssize_t in_write_index_show(struct device *dev,
 430                                    struct device_attribute *dev_attr, char *buf)
 431 {
 432         struct hv_device *hv_dev = device_to_hv_device(dev);
 433         struct hv_ring_buffer_debug_info inbound;
 434         int ret;
 435
 436         if (!hv_dev->channel)
 437                 return -ENODEV;
 438
 439         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 440         if (ret < 0)
 441                 return ret;
 442
 443         return sprintf(buf, "%d\n", inbound.current_write_index);
 444 }
 445 static DEVICE_ATTR_RO(in_write_index);
 446
 447 static ssize_t in_read_bytes_avail_show(struct device *dev,
 448                                         struct device_attribute *dev_attr,
 449                                         char *buf)
 450 {
 451         struct hv_device *hv_dev = device_to_hv_device(dev);
 452         struct hv_ring_buffer_debug_info inbound;
 453         int ret;
 454
 455         if (!hv_dev->channel)
 456                 return -ENODEV;
 457
 458         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 459         if (ret < 0)
 460                 return ret;
 461
 462         return sprintf(buf, "%d\n", inbound.bytes_avail_toread);
 463 }
 464 static DEVICE_ATTR_RO(in_read_bytes_avail);
 465
 466 static ssize_t in_write_bytes_avail_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 hv_ring_buffer_debug_info inbound;
 472         int ret;
 473
 474         if (!hv_dev->channel)
 475                 return -ENODEV;
 476
 477         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 478         if (ret < 0)
 479                 return ret;
 480
 481         return sprintf(buf, "%d\n", inbound.bytes_avail_towrite);
 482 }
 483 static DEVICE_ATTR_RO(in_write_bytes_avail);
 484
 485 static ssize_t channel_vp_mapping_show(struct device *dev,
 486                                        struct device_attribute *dev_attr,
 487                                        char *buf)
 488 {
 489         struct hv_device *hv_dev = device_to_hv_device(dev);
 490         struct vmbus_channel *channel = hv_dev->channel, *cur_sc;
 491         unsigned long flags;
 492         int buf_size = PAGE_SIZE, n_written, tot_written;
 493         struct list_head *cur;
 494
 495         if (!channel)
 496                 return -ENODEV;
 497
 498         tot_written = snprintf(buf, buf_size, "%u:%u\n",
 499                 channel->offermsg.child_relid, channel->target_cpu);
 500
 501         spin_lock_irqsave(&channel->lock, flags);
 502
 503         list_for_each(cur, &channel->sc_list) {
 504                 if (tot_written >= buf_size - 1)
 505                         break;
 506
 507                 cur_sc = list_entry(cur, struct vmbus_channel, sc_list);
 508                 n_written = scnprintf(buf + tot_written,
 509                                      buf_size - tot_written,
 510                                      "%u:%u\n",
 511                                      cur_sc->offermsg.child_relid,
 512                                      cur_sc->target_cpu);
 513                 tot_written += n_written;
 514         }
 515
 516         spin_unlock_irqrestore(&channel->lock, flags);
 517
 518         return tot_written;
 519 }
 520 static DEVICE_ATTR_RO(channel_vp_mapping);
 521
 522 static ssize_t vendor_show(struct device *dev,
 523                            struct device_attribute *dev_attr,
 524                            char *buf)
 525 {
 526         struct hv_device *hv_dev = device_to_hv_device(dev);
 527         return sprintf(buf, "0x%x\n", hv_dev->vendor_id);
 528 }
 529 static DEVICE_ATTR_RO(vendor);
 530
 531 static ssize_t device_show(struct device *dev,
 532                            struct device_attribute *dev_attr,
 533                            char *buf)
 534 {
 535         struct hv_device *hv_dev = device_to_hv_device(dev);
 536         return sprintf(buf, "0x%x\n", hv_dev->device_id);
 537 }
 538 static DEVICE_ATTR_RO(device);
 539
 540 static ssize_t driver_override_store(struct device *dev,
 541                                      struct device_attribute *attr,
 542                                      const char *buf, size_t count)
 543 {
 544         struct hv_device *hv_dev = device_to_hv_device(dev);
 545         char *driver_override, *old, *cp;
 546
 547         /* We need to keep extra room for a newline */
 548         if (count >= (PAGE_SIZE - 1))
 549                 return -EINVAL;
 550
 551         driver_override = kstrndup(buf, count, GFP_KERNEL);
 552         if (!driver_override)
 553                 return -ENOMEM;
 554
 555         cp = strchr(driver_override, '\n');
 556         if (cp)
 557                 *cp = '\0';
 558
 559         device_lock(dev);
 560         old = hv_dev->driver_override;
 561         if (strlen(driver_override)) {
 562                 hv_dev->driver_override = driver_override;
 563         } else {
 564                 kfree(driver_override);
 565                 hv_dev->driver_override = NULL;
 566         }
 567         device_unlock(dev);
 568
 569         kfree(old);
 570
 571         return count;
 572 }
 573
 574 static ssize_t driver_override_show(struct device *dev,
 575                                     struct device_attribute *attr, char *buf)
 576 {
 577         struct hv_device *hv_dev = device_to_hv_device(dev);
 578         ssize_t len;
 579
 580         device_lock(dev);
 581         len = snprintf(buf, PAGE_SIZE, "%s\n", hv_dev->driver_override);
 582         device_unlock(dev);
 583
 584         return len;
 585 }
 586 static DEVICE_ATTR_RW(driver_override);
 587
 588 /* Set up per device attributes in /sys/bus/vmbus/devices/<bus device> */
 589 static struct attribute *vmbus_dev_attrs[] = {
 590         &dev_attr_id.attr,
 591         &dev_attr_state.attr,
 592         &dev_attr_monitor_id.attr,
 593         &dev_attr_class_id.attr,
 594         &dev_attr_device_id.attr,
 595         &dev_attr_modalias.attr,
 596 #ifdef CONFIG_NUMA
 597         &dev_attr_numa_node.attr,
 598 #endif
 599         &dev_attr_server_monitor_pending.attr,
 600         &dev_attr_client_monitor_pending.attr,
 601         &dev_attr_server_monitor_latency.attr,
 602         &dev_attr_client_monitor_latency.attr,
 603         &dev_attr_server_monitor_conn_id.attr,
 604         &dev_attr_client_monitor_conn_id.attr,
 605         &dev_attr_out_intr_mask.attr,
 606         &dev_attr_out_read_index.attr,
 607         &dev_attr_out_write_index.attr,
 608         &dev_attr_out_read_bytes_avail.attr,
 609         &dev_attr_out_write_bytes_avail.attr,
 610         &dev_attr_in_intr_mask.attr,
 611         &dev_attr_in_read_index.attr,
 612         &dev_attr_in_write_index.attr,
 613         &dev_attr_in_read_bytes_avail.attr,
 614         &dev_attr_in_write_bytes_avail.attr,
 615         &dev_attr_channel_vp_mapping.attr,
 616         &dev_attr_vendor.attr,
 617         &dev_attr_device.attr,
 618         &dev_attr_driver_override.attr,
 619         NULL,
 620 };
 621
 622 /*
 623  * Device-level attribute_group callback function. Returns the permission for
 624  * each attribute, and returns 0 if an attribute is not visible.
 625  */
 626 static umode_t vmbus_dev_attr_is_visible(struct kobject *kobj,
 627                                          struct attribute *attr, int idx)
 628 {
 629         struct device *dev = kobj_to_dev(kobj);
 630         const struct hv_device *hv_dev = device_to_hv_device(dev);
 631
 632         /* Hide the monitor attributes if the monitor mechanism is not used. */
 633         if (!hv_dev->channel->offermsg.monitor_allocated &&
 634             (attr == &dev_attr_monitor_id.attr ||
 635              attr == &dev_attr_server_monitor_pending.attr ||
 636              attr == &dev_attr_client_monitor_pending.attr ||
 637              attr == &dev_attr_server_monitor_latency.attr ||
 638              attr == &dev_attr_client_monitor_latency.attr ||
 639              attr == &dev_attr_server_monitor_conn_id.attr ||
 640              attr == &dev_attr_client_monitor_conn_id.attr))
 641                 return 0;
 642
 643         return attr->mode;
 644 }
 645
 646 static const struct attribute_group vmbus_dev_group = {
 647         .attrs = vmbus_dev_attrs,
 648         .is_visible = vmbus_dev_attr_is_visible
 649 };
 650 __ATTRIBUTE_GROUPS(vmbus_dev);
 651
 652 /*
 653  * vmbus_uevent - add uevent for our device
 654  *
 655  * This routine is invoked when a device is added or removed on the vmbus to
 656  * generate a uevent to udev in the userspace. The udev will then look at its
 657  * rule and the uevent generated here to load the appropriate driver
 658  *
 659  * The alias string will be of the form vmbus:guid where guid is the string
 660  * representation of the device guid (each byte of the guid will be
 661  * represented with two hex characters.
 662  */
 663 static int vmbus_uevent(struct device *device, struct kobj_uevent_env *env)
 664 {
 665         struct hv_device *dev = device_to_hv_device(device);
 666         int ret;
 667         char alias_name[VMBUS_ALIAS_LEN + 1];
 668
 669         print_alias_name(dev, alias_name);
 670         ret = add_uevent_var(env, "MODALIAS=vmbus:%s", alias_name);
 671         return ret;
 672 }
 673
 674 static const struct hv_vmbus_device_id *
 675 hv_vmbus_dev_match(const struct hv_vmbus_device_id *id, const guid_t *guid)
 676 {
 677         if (id == NULL)
 678                 return NULL; /* empty device table */
 679
 680         for (; !guid_is_null(&id->guid); id++)
 681                 if (guid_equal(&id->guid, guid))
 682                         return id;
 683
 684         return NULL;
 685 }
 686
 687 static const struct hv_vmbus_device_id *
 688 hv_vmbus_dynid_match(struct hv_driver *drv, const guid_t *guid)
 689 {
 690         const struct hv_vmbus_device_id *id = NULL;
 691         struct vmbus_dynid *dynid;
 692
 693         spin_lock(&drv->dynids.lock);
 694         list_for_each_entry(dynid, &drv->dynids.list, node) {
 695                 if (guid_equal(&dynid->id.guid, guid)) {
 696                         id = &dynid->id;
 697                         break;
 698                 }
 699         }
 700         spin_unlock(&drv->dynids.lock);
 701
 702         return id;
 703 }
 704
 705 static const struct hv_vmbus_device_id vmbus_device_null;
 706
 707 /*
 708  * Return a matching hv_vmbus_device_id pointer.
 709  * If there is no match, return NULL.
 710  */
 711 static const struct hv_vmbus_device_id *hv_vmbus_get_id(struct hv_driver *drv,
 712                                                         struct hv_device *dev)
 713 {
 714         const guid_t *guid = &dev->dev_type;
 715         const struct hv_vmbus_device_id *id;
 716
 717         /* When driver_override is set, only bind to the matching driver */
 718         if (dev->driver_override && strcmp(dev->driver_override, drv->name))
 719                 return NULL;
 720
 721         /* Look at the dynamic ids first, before the static ones */
 722         id = hv_vmbus_dynid_match(drv, guid);
 723         if (!id)
 724                 id = hv_vmbus_dev_match(drv->id_table, guid);
 725
 726         /* driver_override will always match, send a dummy id */
 727         if (!id && dev->driver_override)
 728                 id = &vmbus_device_null;
 729
 730         return id;
 731 }
 732
 733 /* vmbus_add_dynid - add a new device ID to this driver and re-probe devices */
 734 static int vmbus_add_dynid(struct hv_driver *drv, guid_t *guid)
 735 {
 736         struct vmbus_dynid *dynid;
 737
 738         dynid = kzalloc(sizeof(*dynid), GFP_KERNEL);
 739         if (!dynid)
 740                 return -ENOMEM;
 741
 742         dynid->id.guid = *guid;
 743
 744         spin_lock(&drv->dynids.lock);
 745         list_add_tail(&dynid->node, &drv->dynids.list);
 746         spin_unlock(&drv->dynids.lock);
 747
 748         return driver_attach(&drv->driver);
 749 }
 750
 751 static void vmbus_free_dynids(struct hv_driver *drv)
 752 {
 753         struct vmbus_dynid *dynid, *n;
 754
 755         spin_lock(&drv->dynids.lock);
 756         list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
 757                 list_del(&dynid->node);
 758                 kfree(dynid);
 759         }
 760         spin_unlock(&drv->dynids.lock);
 761 }
 762
 763 /*
 764  * store_new_id - sysfs frontend to vmbus_add_dynid()
 765  *
 766  * Allow GUIDs to be added to an existing driver via sysfs.
 767  */
 768 static ssize_t new_id_store(struct device_driver *driver, const char *buf,
 769                             size_t count)
 770 {
 771         struct hv_driver *drv = drv_to_hv_drv(driver);
 772         guid_t guid;
 773         ssize_t retval;
 774
 775         retval = guid_parse(buf, &guid);
 776         if (retval)
 777                 return retval;
 778
 779         if (hv_vmbus_dynid_match(drv, &guid))
 780                 return -EEXIST;
 781
 782         retval = vmbus_add_dynid(drv, &guid);
 783         if (retval)
 784                 return retval;
 785         return count;
 786 }
 787 static DRIVER_ATTR_WO(new_id);
 788
 789 /*
 790  * store_remove_id - remove a PCI device ID from this driver
 791  *
 792  * Removes a dynamic pci device ID to this driver.
 793  */
 794 static ssize_t remove_id_store(struct device_driver *driver, const char *buf,
 795                                size_t count)
 796 {
 797         struct hv_driver *drv = drv_to_hv_drv(driver);
 798         struct vmbus_dynid *dynid, *n;
 799         guid_t guid;
 800         ssize_t retval;
 801
 802         retval = guid_parse(buf, &guid);
 803         if (retval)
 804                 return retval;
 805
 806         retval = -ENODEV;
 807         spin_lock(&drv->dynids.lock);
 808         list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
 809                 struct hv_vmbus_device_id *id = &dynid->id;
 810
 811                 if (guid_equal(&id->guid, &guid)) {
 812                         list_del(&dynid->node);
 813                         kfree(dynid);
 814                         retval = count;
 815                         break;
 816                 }
 817         }
 818         spin_unlock(&drv->dynids.lock);
 819
 820         return retval;
 821 }
 822 static DRIVER_ATTR_WO(remove_id);
 823
 824 static struct attribute *vmbus_drv_attrs[] = {
 825         &driver_attr_new_id.attr,
 826         &driver_attr_remove_id.attr,
 827         NULL,
 828 };
 829 ATTRIBUTE_GROUPS(vmbus_drv);
 830
 831
 832 /*
 833  * vmbus_match - Attempt to match the specified device to the specified driver
 834  */
 835 static int vmbus_match(struct device *device, struct device_driver *driver)
 836 {
 837         struct hv_driver *drv = drv_to_hv_drv(driver);
 838         struct hv_device *hv_dev = device_to_hv_device(device);
 839
 840         /* The hv_sock driver handles all hv_sock offers. */
 841         if (is_hvsock_channel(hv_dev->channel))
 842                 return drv->hvsock;
 843
 844         if (hv_vmbus_get_id(drv, hv_dev))
 845                 return 1;
 846
 847         return 0;
 848 }
 849
 850 /*
 851  * vmbus_probe - Add the new vmbus's child device
 852  */
 853 static int vmbus_probe(struct device *child_device)
 854 {
 855         int ret = 0;
 856         struct hv_driver *drv =
 857                         drv_to_hv_drv(child_device->driver);
 858         struct hv_device *dev = device_to_hv_device(child_device);
 859         const struct hv_vmbus_device_id *dev_id;
 860
 861         dev_id = hv_vmbus_get_id(drv, dev);
 862         if (drv->probe) {
 863                 ret = drv->probe(dev, dev_id);
 864                 if (ret != 0)
 865                         pr_err("probe failed for device %s (%d)\n",
 866                                dev_name(child_device), ret);
 867
 868         } else {
 869                 pr_err("probe not set for driver %s\n",
 870                        dev_name(child_device));
 871                 ret = -ENODEV;
 872         }
 873         return ret;
 874 }
 875
 876 /*
 877  * vmbus_remove - Remove a vmbus device
 878  */
 879 static int vmbus_remove(struct device *child_device)
 880 {
 881         struct hv_driver *drv;
 882         struct hv_device *dev = device_to_hv_device(child_device);
 883
 884         if (child_device->driver) {
 885                 drv = drv_to_hv_drv(child_device->driver);
 886                 if (drv->remove)
 887                         drv->remove(dev);
 888         }
 889
 890         return 0;
 891 }
 892
 893
 894 /*
 895  * vmbus_shutdown - Shutdown a vmbus device
 896  */
 897 static void vmbus_shutdown(struct device *child_device)
 898 {
 899         struct hv_driver *drv;
 900         struct hv_device *dev = device_to_hv_device(child_device);
 901
 902
 903         /* The device may not be attached yet */
 904         if (!child_device->driver)
 905                 return;
 906
 907         drv = drv_to_hv_drv(child_device->driver);
 908
 909         if (drv->shutdown)
 910                 drv->shutdown(dev);
 911 }
 912
 913
 914 /*
 915  * vmbus_device_release - Final callback release of the vmbus child device
 916  */
 917 static void vmbus_device_release(struct device *device)
 918 {
 919         struct hv_device *hv_dev = device_to_hv_device(device);
 920         struct vmbus_channel *channel = hv_dev->channel;
 921
 922         mutex_lock(&vmbus_connection.channel_mutex);
 923         hv_process_channel_removal(channel);
 924         mutex_unlock(&vmbus_connection.channel_mutex);
 925         kfree(hv_dev);
 926 }
 927
 928 /* The one and only one */
 929 static struct bus_type  hv_bus = {
 930         .name =         "vmbus",
 931         .match =                vmbus_match,
 932         .shutdown =             vmbus_shutdown,
 933         .remove =               vmbus_remove,
 934         .probe =                vmbus_probe,
 935         .uevent =               vmbus_uevent,
 936         .dev_groups =           vmbus_dev_groups,
 937         .drv_groups =           vmbus_drv_groups,
 938 };
 939
 940 struct onmessage_work_context {
 941         struct work_struct work;
 942         struct hv_message msg;
 943 };
 944
 945 static void vmbus_onmessage_work(struct work_struct *work)
 946 {
 947         struct onmessage_work_context *ctx;
 948
 949         /* Do not process messages if we're in DISCONNECTED state */
 950         if (vmbus_connection.conn_state == DISCONNECTED)
 951                 return;
 952
 953         ctx = container_of(work, struct onmessage_work_context,
 954                            work);
 955         vmbus_onmessage(&ctx->msg);
 956         kfree(ctx);
 957 }
 958
 959 void vmbus_on_msg_dpc(unsigned long data)
 960 {
 961         struct hv_per_cpu_context *hv_cpu = (void *)data;
 962         void *page_addr = hv_cpu->synic_message_page;
 963         struct hv_message *msg = (struct hv_message *)page_addr +
 964                                   VMBUS_MESSAGE_SINT;
 965         struct vmbus_channel_message_header *hdr;
 966         const struct vmbus_channel_message_table_entry *entry;
 967         struct onmessage_work_context *ctx;
 968         u32 message_type = msg->header.message_type;
 969
 970         if (message_type == HVMSG_NONE)
 971                 /* no msg */
 972                 return;
 973
 974         hdr = (struct vmbus_channel_message_header *)msg->u.payload;
 975
 976         trace_vmbus_on_msg_dpc(hdr);
 977
 978         if (hdr->msgtype >= CHANNELMSG_COUNT) {
 979                 WARN_ONCE(1, "unknown msgtype=%d\n", hdr->msgtype);
 980                 goto msg_handled;
 981         }
 982
 983         entry = &channel_message_table[hdr->msgtype];
 984         if (entry->handler_type == VMHT_BLOCKING) {
 985                 ctx = kmalloc(sizeof(*ctx), GFP_ATOMIC);
 986                 if (ctx == NULL)
 987                         return;
 988
 989                 INIT_WORK(&ctx->work, vmbus_onmessage_work);
 990                 memcpy(&ctx->msg, msg, sizeof(*msg));
 991
 992                 /*
 993                  * The host can generate a rescind message while we
 994                  * may still be handling the original offer. We deal with
 995                  * this condition by ensuring the processing is done on the
 996                  * same CPU.
 997                  */
 998                 switch (hdr->msgtype) {
 999                 case CHANNELMSG_RESCIND_CHANNELOFFER:
1000                         /*
1001                          * If we are handling the rescind message;
1002                          * schedule the work on the global work queue.
1003                          */
1004                         schedule_work_on(vmbus_connection.connect_cpu,
1005                                          &ctx->work);
1006                         break;
1007
1008                 case CHANNELMSG_OFFERCHANNEL:
1009                         atomic_inc(&vmbus_connection.offer_in_progress);
1010                         queue_work_on(vmbus_connection.connect_cpu,
1011                                       vmbus_connection.work_queue,
1012                                       &ctx->work);
1013                         break;
1014
1015                 default:
1016                         queue_work(vmbus_connection.work_queue, &ctx->work);
1017                 }
1018         } else
1019                 entry->message_handler(hdr);
1020
1021 msg_handled:
1022         vmbus_signal_eom(msg, message_type);
1023 }
1024
1025
1026 /*
1027  * Direct callback for channels using other deferred processing
1028  */
1029 static void vmbus_channel_isr(struct vmbus_channel *channel)
1030 {
1031         void (*callback_fn)(void *);
1032
1033         callback_fn = READ_ONCE(channel->onchannel_callback);
1034         if (likely(callback_fn != NULL))
1035                 (*callback_fn)(channel->channel_callback_context);
1036 }
1037
1038 /*
1039  * Schedule all channels with events pending
1040  */
1041 static void vmbus_chan_sched(struct hv_per_cpu_context *hv_cpu)
1042 {
1043         unsigned long *recv_int_page;
1044         u32 maxbits, relid;
1045
1046         if (vmbus_proto_version < VERSION_WIN8) {
1047                 maxbits = MAX_NUM_CHANNELS_SUPPORTED;
1048                 recv_int_page = vmbus_connection.recv_int_page;
1049         } else {
1050                 /*
1051                  * When the host is win8 and beyond, the event page
1052                  * can be directly checked to get the id of the channel
1053                  * that has the interrupt pending.
1054                  */
1055                 void *page_addr = hv_cpu->synic_event_page;
1056                 union hv_synic_event_flags *event
1057                         = (union hv_synic_event_flags *)page_addr +
1058                                                  VMBUS_MESSAGE_SINT;
1059
1060                 maxbits = HV_EVENT_FLAGS_COUNT;
1061                 recv_int_page = event->flags;
1062         }
1063
1064         if (unlikely(!recv_int_page))
1065                 return;
1066
1067         for_each_set_bit(relid, recv_int_page, maxbits) {
1068                 struct vmbus_channel *channel;
1069
1070                 if (!sync_test_and_clear_bit(relid, recv_int_page))
1071                         continue;
1072
1073                 /* Special case - vmbus channel protocol msg */
1074                 if (relid == 0)
1075                         continue;
1076
1077                 rcu_read_lock();
1078
1079                 /* Find channel based on relid */
1080                 list_for_each_entry_rcu(channel, &hv_cpu->chan_list, percpu_list) {
1081                         if (channel->offermsg.child_relid != relid)
1082                                 continue;
1083
1084                         if (channel->rescind)
1085                                 continue;
1086
1087                         trace_vmbus_chan_sched(channel);
1088
1089                         ++channel->interrupts;
1090
1091                         switch (channel->callback_mode) {
1092                         case HV_CALL_ISR:
1093                                 vmbus_channel_isr(channel);
1094                                 break;
1095
1096                         case HV_CALL_BATCHED:
1097                                 hv_begin_read(&channel->inbound);
1098                                 /* fallthrough */
1099                         case HV_CALL_DIRECT:
1100                                 tasklet_schedule(&channel->callback_event);
1101                         }
1102                 }
1103
1104                 rcu_read_unlock();
1105         }
1106 }
1107
1108 static void vmbus_isr(void)
1109 {
1110         struct hv_per_cpu_context *hv_cpu
1111                 = this_cpu_ptr(hv_context.cpu_context);
1112         void *page_addr = hv_cpu->synic_event_page;
1113         struct hv_message *msg;
1114         union hv_synic_event_flags *event;
1115         bool handled = false;
1116
1117         if (unlikely(page_addr == NULL))
1118                 return;
1119
1120         event = (union hv_synic_event_flags *)page_addr +
1121                                          VMBUS_MESSAGE_SINT;
1122         /*
1123          * Check for events before checking for messages. This is the order
1124          * in which events and messages are checked in Windows guests on
1125          * Hyper-V, and the Windows team suggested we do the same.
1126          */
1127
1128         if ((vmbus_proto_version == VERSION_WS2008) ||
1129                 (vmbus_proto_version == VERSION_WIN7)) {
1130
1131                 /* Since we are a child, we only need to check bit 0 */
1132                 if (sync_test_and_clear_bit(0, event->flags))
1133                         handled = true;
1134         } else {
1135                 /*
1136                  * Our host is win8 or above. The signaling mechanism
1137                  * has changed and we can directly look at the event page.
1138                  * If bit n is set then we have an interrup on the channel
1139                  * whose id is n.
1140                  */
1141                 handled = true;
1142         }
1143
1144         if (handled)
1145                 vmbus_chan_sched(hv_cpu);
1146
1147         page_addr = hv_cpu->synic_message_page;
1148         msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
1149
1150         /* Check if there are actual msgs to be processed */
1151         if (msg->header.message_type != HVMSG_NONE) {
1152                 if (msg->header.message_type == HVMSG_TIMER_EXPIRED) {
1153                         hv_stimer0_isr();
1154                         vmbus_signal_eom(msg, HVMSG_TIMER_EXPIRED);
1155                 } else
1156                         tasklet_schedule(&hv_cpu->msg_dpc);
1157         }
1158
1159         add_interrupt_randomness(HYPERVISOR_CALLBACK_VECTOR, 0);
1160 }
1161
1162 /*
1163  * Boolean to control whether to report panic messages over Hyper-V.
1164  *
1165  * It can be set via /proc/sys/kernel/hyperv/record_panic_msg
1166  */
1167 static int sysctl_record_panic_msg = 1;
1168
1169 /*
1170  * Callback from kmsg_dump. Grab as much as possible from the end of the kmsg
1171  * buffer and call into Hyper-V to transfer the data.
1172  */
1173 static void hv_kmsg_dump(struct kmsg_dumper *dumper,
1174                          enum kmsg_dump_reason reason)
1175 {
1176         size_t bytes_written;
1177         phys_addr_t panic_pa;
1178
1179         /* We are only interested in panics. */
1180         if ((reason != KMSG_DUMP_PANIC) || (!sysctl_record_panic_msg))
1181                 return;
1182
1183         panic_pa = virt_to_phys(hv_panic_page);
1184
1185         /*
1186          * Write dump contents to the page. No need to synchronize; panic should
1187          * be single-threaded.
1188          */
1189         kmsg_dump_get_buffer(dumper, true, hv_panic_page, PAGE_SIZE,
1190                              &bytes_written);
1191         if (bytes_written)
1192                 hyperv_report_panic_msg(panic_pa, bytes_written);
1193 }
1194
1195 static struct kmsg_dumper hv_kmsg_dumper = {
1196         .dump = hv_kmsg_dump,
1197 };
1198
1199 static struct ctl_table_header *hv_ctl_table_hdr;
1200 static int zero;
1201 static int one = 1;
1202
1203 /*
1204  * sysctl option to allow the user to control whether kmsg data should be
1205  * reported to Hyper-V on panic.
1206  */
1207 static struct ctl_table hv_ctl_table[] = {
1208         {
1209                 .procname       = "hyperv_record_panic_msg",
1210                 .data           = &sysctl_record_panic_msg,
1211                 .maxlen         = sizeof(int),
1212                 .mode           = 0644,
1213                 .proc_handler   = proc_dointvec_minmax,
1214                 .extra1         = &zero,
1215                 .extra2         = &one
1216         },
1217         {}
1218 };
1219
1220 static struct ctl_table hv_root_table[] = {
1221         {
1222                 .procname       = "kernel",
1223                 .mode           = 0555,
1224                 .child          = hv_ctl_table
1225         },
1226         {}
1227 };
1228
1229 /*
1230  * vmbus_bus_init -Main vmbus driver initialization routine.
1231  *
1232  * Here, we
1233  *      - initialize the vmbus driver context
1234  *      - invoke the vmbus hv main init routine
1235  *      - retrieve the channel offers
1236  */
1237 static int vmbus_bus_init(void)
1238 {
1239         int ret;
1240
1241         /* Hypervisor initialization...setup hypercall page..etc */
1242         ret = hv_init();
1243         if (ret != 0) {
1244                 pr_err("Unable to initialize the hypervisor - 0x%x\n", ret);
1245                 return ret;
1246         }
1247
1248         ret = bus_register(&hv_bus);
1249         if (ret)
1250                 return ret;
1251
1252         hv_setup_vmbus_irq(vmbus_isr);
1253
1254         ret = hv_synic_alloc();
1255         if (ret)
1256                 goto err_alloc;
1257
1258         ret = hv_stimer_alloc(VMBUS_MESSAGE_SINT);
1259         if (ret < 0)
1260                 goto err_alloc;
1261
1262         /*
1263          * Initialize the per-cpu interrupt state and stimer state.
1264          * Then connect to the host.
1265          */
1266         ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "hyperv/vmbus:online",
1267                                 hv_synic_init, hv_synic_cleanup);
1268         if (ret < 0)
1269                 goto err_cpuhp;
1270         hyperv_cpuhp_online = ret;
1271
1272         ret = vmbus_connect();
1273         if (ret)
1274                 goto err_connect;
1275
1276         /*
1277          * Only register if the crash MSRs are available
1278          */
1279         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1280                 u64 hyperv_crash_ctl;
1281                 /*
1282                  * Sysctl registration is not fatal, since by default
1283                  * reporting is enabled.
1284                  */
1285                 hv_ctl_table_hdr = register_sysctl_table(hv_root_table);
1286                 if (!hv_ctl_table_hdr)
1287                         pr_err("Hyper-V: sysctl table register error");
1288
1289                 /*
1290                  * Register for panic kmsg callback only if the right
1291                  * capability is supported by the hypervisor.
1292                  */
1293                 hv_get_crash_ctl(hyperv_crash_ctl);
1294                 if (hyperv_crash_ctl & HV_CRASH_CTL_CRASH_NOTIFY_MSG) {
1295                         hv_panic_page = (void *)get_zeroed_page(GFP_KERNEL);
1296                         if (hv_panic_page) {
1297                                 ret = kmsg_dump_register(&hv_kmsg_dumper);
1298                                 if (ret)
1299                                         pr_err("Hyper-V: kmsg dump register "
1300                                                 "error 0x%x\n", ret);
1301                         } else
1302                                 pr_err("Hyper-V: panic message page memory "
1303                                         "allocation failed");
1304                 }
1305
1306                 register_die_notifier(&hyperv_die_block);
1307                 atomic_notifier_chain_register(&panic_notifier_list,
1308                                                &hyperv_panic_block);
1309         }
1310
1311         vmbus_request_offers();
1312
1313         return 0;
1314
1315 err_connect:
1316         cpuhp_remove_state(hyperv_cpuhp_online);
1317 err_cpuhp:
1318         hv_stimer_free();
1319 err_alloc:
1320         hv_synic_free();
1321         hv_remove_vmbus_irq();
1322
1323         bus_unregister(&hv_bus);
1324         free_page((unsigned long)hv_panic_page);
1325         unregister_sysctl_table(hv_ctl_table_hdr);
1326         hv_ctl_table_hdr = NULL;
1327         return ret;
1328 }
1329
1330 /**
1331  * __vmbus_child_driver_register() - Register a vmbus's driver
1332  * @hv_driver: Pointer to driver structure you want to register
1333  * @owner: owner module of the drv
1334  * @mod_name: module name string
1335  *
1336  * Registers the given driver with Linux through the 'driver_register()' call
1337  * and sets up the hyper-v vmbus handling for this driver.
1338  * It will return the state of the 'driver_register()' call.
1339  *
1340  */
1341 int __vmbus_driver_register(struct hv_driver *hv_driver, struct module *owner, const char *mod_name)
1342 {
1343         int ret;
1344
1345         pr_info("registering driver %s\n", hv_driver->name);
1346
1347         ret = vmbus_exists();
1348         if (ret < 0)
1349                 return ret;
1350
1351         hv_driver->driver.name = hv_driver->name;
1352         hv_driver->driver.owner = owner;
1353         hv_driver->driver.mod_name = mod_name;
1354         hv_driver->driver.bus = &hv_bus;
1355
1356         spin_lock_init(&hv_driver->dynids.lock);
1357         INIT_LIST_HEAD(&hv_driver->dynids.list);
1358
1359         ret = driver_register(&hv_driver->driver);
1360
1361         return ret;
1362 }
1363 EXPORT_SYMBOL_GPL(__vmbus_driver_register);
1364
1365 /**
1366  * vmbus_driver_unregister() - Unregister a vmbus's driver
1367  * @hv_driver: Pointer to driver structure you want to
1368  *             un-register
1369  *
1370  * Un-register the given driver that was previous registered with a call to
1371  * vmbus_driver_register()
1372  */
1373 void vmbus_driver_unregister(struct hv_driver *hv_driver)
1374 {
1375         pr_info("unregistering driver %s\n", hv_driver->name);
1376
1377         if (!vmbus_exists()) {
1378                 driver_unregister(&hv_driver->driver);
1379                 vmbus_free_dynids(hv_driver);
1380         }
1381 }
1382 EXPORT_SYMBOL_GPL(vmbus_driver_unregister);
1383
1384
1385 /*
1386  * Called when last reference to channel is gone.
1387  */
1388 static void vmbus_chan_release(struct kobject *kobj)
1389 {
1390         struct vmbus_channel *channel
1391                 = container_of(kobj, struct vmbus_channel, kobj);
1392
1393         kfree_rcu(channel, rcu);
1394 }
1395
1396 struct vmbus_chan_attribute {
1397         struct attribute attr;
1398         ssize_t (*show)(struct vmbus_channel *chan, char *buf);
1399         ssize_t (*store)(struct vmbus_channel *chan,
1400                          const char *buf, size_t count);
1401 };
1402 #define VMBUS_CHAN_ATTR(_name, _mode, _show, _store) \
1403         struct vmbus_chan_attribute chan_attr_##_name \
1404                 = __ATTR(_name, _mode, _show, _store)
1405 #define VMBUS_CHAN_ATTR_RW(_name) \
1406         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RW(_name)
1407 #define VMBUS_CHAN_ATTR_RO(_name) \
1408         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RO(_name)
1409 #define VMBUS_CHAN_ATTR_WO(_name) \
1410         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_WO(_name)
1411
1412 static ssize_t vmbus_chan_attr_show(struct kobject *kobj,
1413                                     struct attribute *attr, char *buf)
1414 {
1415         const struct vmbus_chan_attribute *attribute
1416                 = container_of(attr, struct vmbus_chan_attribute, attr);
1417         struct vmbus_channel *chan
1418                 = container_of(kobj, struct vmbus_channel, kobj);
1419
1420         if (!attribute->show)
1421                 return -EIO;
1422
1423         return attribute->show(chan, buf);
1424 }
1425
1426 static const struct sysfs_ops vmbus_chan_sysfs_ops = {
1427         .show = vmbus_chan_attr_show,
1428 };
1429
1430 static ssize_t out_mask_show(struct vmbus_channel *channel, char *buf)
1431 {
1432         struct hv_ring_buffer_info *rbi = &channel->outbound;
1433         ssize_t ret;
1434
1435         mutex_lock(&rbi->ring_buffer_mutex);
1436         if (!rbi->ring_buffer) {
1437                 mutex_unlock(&rbi->ring_buffer_mutex);
1438                 return -EINVAL;
1439         }
1440
1441         ret = sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1442         mutex_unlock(&rbi->ring_buffer_mutex);
1443         return ret;
1444 }
1445 static VMBUS_CHAN_ATTR_RO(out_mask);
1446
1447 static ssize_t in_mask_show(struct vmbus_channel *channel, char *buf)
1448 {
1449         struct hv_ring_buffer_info *rbi = &channel->inbound;
1450         ssize_t ret;
1451
1452         mutex_lock(&rbi->ring_buffer_mutex);
1453         if (!rbi->ring_buffer) {
1454                 mutex_unlock(&rbi->ring_buffer_mutex);
1455                 return -EINVAL;
1456         }
1457
1458         ret = sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1459         mutex_unlock(&rbi->ring_buffer_mutex);
1460         return ret;
1461 }
1462 static VMBUS_CHAN_ATTR_RO(in_mask);
1463
1464 static ssize_t read_avail_show(struct vmbus_channel *channel, char *buf)
1465 {
1466         struct hv_ring_buffer_info *rbi = &channel->inbound;
1467         ssize_t ret;
1468
1469         mutex_lock(&rbi->ring_buffer_mutex);
1470         if (!rbi->ring_buffer) {
1471                 mutex_unlock(&rbi->ring_buffer_mutex);
1472                 return -EINVAL;
1473         }
1474
1475         ret = sprintf(buf, "%u\n", hv_get_bytes_to_read(rbi));
1476         mutex_unlock(&rbi->ring_buffer_mutex);
1477         return ret;
1478 }
1479 static VMBUS_CHAN_ATTR_RO(read_avail);
1480
1481 static ssize_t write_avail_show(struct vmbus_channel *channel, char *buf)
1482 {
1483         struct hv_ring_buffer_info *rbi = &channel->outbound;
1484         ssize_t ret;
1485
1486         mutex_lock(&rbi->ring_buffer_mutex);
1487         if (!rbi->ring_buffer) {
1488                 mutex_unlock(&rbi->ring_buffer_mutex);
1489                 return -EINVAL;
1490         }
1491
1492         ret = sprintf(buf, "%u\n", hv_get_bytes_to_write(rbi));
1493         mutex_unlock(&rbi->ring_buffer_mutex);
1494         return ret;
1495 }
1496 static VMBUS_CHAN_ATTR_RO(write_avail);
1497
1498 static ssize_t show_target_cpu(struct vmbus_channel *channel, char *buf)
1499 {
1500         return sprintf(buf, "%u\n", channel->target_cpu);
1501 }
1502 static VMBUS_CHAN_ATTR(cpu, S_IRUGO, show_target_cpu, NULL);
1503
1504 static ssize_t channel_pending_show(struct vmbus_channel *channel,
1505                                     char *buf)
1506 {
1507         return sprintf(buf, "%d\n",
1508                        channel_pending(channel,
1509                                        vmbus_connection.monitor_pages[1]));
1510 }
1511 static VMBUS_CHAN_ATTR(pending, S_IRUGO, channel_pending_show, NULL);
1512
1513 static ssize_t channel_latency_show(struct vmbus_channel *channel,
1514                                     char *buf)
1515 {
1516         return sprintf(buf, "%d\n",
1517                        channel_latency(channel,
1518                                        vmbus_connection.monitor_pages[1]));
1519 }
1520 static VMBUS_CHAN_ATTR(latency, S_IRUGO, channel_latency_show, NULL);
1521
1522 static ssize_t channel_interrupts_show(struct vmbus_channel *channel, char *buf)
1523 {
1524         return sprintf(buf, "%llu\n", channel->interrupts);
1525 }
1526 static VMBUS_CHAN_ATTR(interrupts, S_IRUGO, channel_interrupts_show, NULL);
1527
1528 static ssize_t channel_events_show(struct vmbus_channel *channel, char *buf)
1529 {
1530         return sprintf(buf, "%llu\n", channel->sig_events);
1531 }
1532 static VMBUS_CHAN_ATTR(events, S_IRUGO, channel_events_show, NULL);
1533
1534 static ssize_t channel_intr_in_full_show(struct vmbus_channel *channel,
1535                                          char *buf)
1536 {
1537         return sprintf(buf, "%llu\n",
1538                        (unsigned long long)channel->intr_in_full);
1539 }
1540 static VMBUS_CHAN_ATTR(intr_in_full, 0444, channel_intr_in_full_show, NULL);
1541
1542 static ssize_t channel_intr_out_empty_show(struct vmbus_channel *channel,
1543                                            char *buf)
1544 {
1545         return sprintf(buf, "%llu\n",
1546                        (unsigned long long)channel->intr_out_empty);
1547 }
1548 static VMBUS_CHAN_ATTR(intr_out_empty, 0444, channel_intr_out_empty_show, NULL);
1549
1550 static ssize_t channel_out_full_first_show(struct vmbus_channel *channel,
1551                                            char *buf)
1552 {
1553         return sprintf(buf, "%llu\n",
1554                        (unsigned long long)channel->out_full_first);
1555 }
1556 static VMBUS_CHAN_ATTR(out_full_first, 0444, channel_out_full_first_show, NULL);
1557
1558 static ssize_t channel_out_full_total_show(struct vmbus_channel *channel,
1559                                            char *buf)
1560 {
1561         return sprintf(buf, "%llu\n",
1562                        (unsigned long long)channel->out_full_total);
1563 }
1564 static VMBUS_CHAN_ATTR(out_full_total, 0444, channel_out_full_total_show, NULL);
1565
1566 static ssize_t subchannel_monitor_id_show(struct vmbus_channel *channel,
1567                                           char *buf)
1568 {
1569         return sprintf(buf, "%u\n", channel->offermsg.monitorid);
1570 }
1571 static VMBUS_CHAN_ATTR(monitor_id, S_IRUGO, subchannel_monitor_id_show, NULL);
1572
1573 static ssize_t subchannel_id_show(struct vmbus_channel *channel,
1574                                   char *buf)
1575 {
1576         return sprintf(buf, "%u\n",
1577                        channel->offermsg.offer.sub_channel_index);
1578 }
1579 static VMBUS_CHAN_ATTR_RO(subchannel_id);
1580
1581 static struct attribute *vmbus_chan_attrs[] = {
1582         &chan_attr_out_mask.attr,
1583         &chan_attr_in_mask.attr,
1584         &chan_attr_read_avail.attr,
1585         &chan_attr_write_avail.attr,
1586         &chan_attr_cpu.attr,
1587         &chan_attr_pending.attr,
1588         &chan_attr_latency.attr,
1589         &chan_attr_interrupts.attr,
1590         &chan_attr_events.attr,
1591         &chan_attr_intr_in_full.attr,
1592         &chan_attr_intr_out_empty.attr,
1593         &chan_attr_out_full_first.attr,
1594         &chan_attr_out_full_total.attr,
1595         &chan_attr_monitor_id.attr,
1596         &chan_attr_subchannel_id.attr,
1597         NULL
1598 };
1599
1600 /*
1601  * Channel-level attribute_group callback function. Returns the permission for
1602  * each attribute, and returns 0 if an attribute is not visible.
1603  */
1604 static umode_t vmbus_chan_attr_is_visible(struct kobject *kobj,
1605                                           struct attribute *attr, int idx)
1606 {
1607         const struct vmbus_channel *channel =
1608                 container_of(kobj, struct vmbus_channel, kobj);
1609
1610         /* Hide the monitor attributes if the monitor mechanism is not used. */
1611         if (!channel->offermsg.monitor_allocated &&
1612             (attr == &chan_attr_pending.attr ||
1613              attr == &chan_attr_latency.attr ||
1614              attr == &chan_attr_monitor_id.attr))
1615                 return 0;
1616
1617         return attr->mode;
1618 }
1619
1620 static struct attribute_group vmbus_chan_group = {
1621         .attrs = vmbus_chan_attrs,
1622         .is_visible = vmbus_chan_attr_is_visible
1623 };
1624
1625 static struct kobj_type vmbus_chan_ktype = {
1626         .sysfs_ops = &vmbus_chan_sysfs_ops,
1627         .release = vmbus_chan_release,
1628 };
1629
1630 /*
1631  * vmbus_add_channel_kobj - setup a sub-directory under device/channels
1632  */
1633 int vmbus_add_channel_kobj(struct hv_device *dev, struct vmbus_channel *channel)
1634 {
1635         const struct device *device = &dev->device;
1636         struct kobject *kobj = &channel->kobj;
1637         u32 relid = channel->offermsg.child_relid;
1638         int ret;
1639
1640         kobj->kset = dev->channels_kset;
1641         ret = kobject_init_and_add(kobj, &vmbus_chan_ktype, NULL,
1642                                    "%u", relid);
1643         if (ret)
1644                 return ret;
1645
1646         ret = sysfs_create_group(kobj, &vmbus_chan_group);
1647
1648         if (ret) {
1649                 /*
1650                  * The calling functions' error handling paths will cleanup the
1651                  * empty channel directory.
1652                  */
1653                 dev_err(device, "Unable to set up channel sysfs files\n");
1654                 return ret;
1655         }
1656
1657         kobject_uevent(kobj, KOBJ_ADD);
1658
1659         return 0;
1660 }
1661
1662 /*
1663  * vmbus_remove_channel_attr_group - remove the channel's attribute group
1664  */
1665 void vmbus_remove_channel_attr_group(struct vmbus_channel *channel)
1666 {
1667         sysfs_remove_group(&channel->kobj, &vmbus_chan_group);
1668 }
1669
1670 /*
1671  * vmbus_device_create - Creates and registers a new child device
1672  * on the vmbus.
1673  */
1674 struct hv_device *vmbus_device_create(const guid_t *type,
1675                                       const guid_t *instance,
1676                                       struct vmbus_channel *channel)
1677 {
1678         struct hv_device *child_device_obj;
1679
1680         child_device_obj = kzalloc(sizeof(struct hv_device), GFP_KERNEL);
1681         if (!child_device_obj) {
1682                 pr_err("Unable to allocate device object for child device\n");
1683                 return NULL;
1684         }
1685
1686         child_device_obj->channel = channel;
1687         guid_copy(&child_device_obj->dev_type, type);
1688         guid_copy(&child_device_obj->dev_instance, instance);
1689         child_device_obj->vendor_id = 0x1414; /* MSFT vendor ID */
1690
1691         return child_device_obj;
1692 }
1693
1694 /*
1695  * vmbus_device_register - Register the child device
1696  */
1697 int vmbus_device_register(struct hv_device *child_device_obj)
1698 {
1699         struct kobject *kobj = &child_device_obj->device.kobj;
1700         int ret;
1701
1702         dev_set_name(&child_device_obj->device, "%pUl",
1703                      child_device_obj->channel->offermsg.offer.if_instance.b);
1704
1705         child_device_obj->device.bus = &hv_bus;
1706         child_device_obj->device.parent = &hv_acpi_dev->dev;
1707         child_device_obj->device.release = vmbus_device_release;
1708
1709         /*
1710          * Register with the LDM. This will kick off the driver/device
1711          * binding...which will eventually call vmbus_match() and vmbus_probe()
1712          */
1713         ret = device_register(&child_device_obj->device);
1714         if (ret) {
1715                 pr_err("Unable to register child device\n");
1716                 return ret;
1717         }
1718
1719         child_device_obj->channels_kset = kset_create_and_add("channels",
1720                                                               NULL, kobj);
1721         if (!child_device_obj->channels_kset) {
1722                 ret = -ENOMEM;
1723                 goto err_dev_unregister;
1724         }
1725
1726         ret = vmbus_add_channel_kobj(child_device_obj,
1727                                      child_device_obj->channel);
1728         if (ret) {
1729                 pr_err("Unable to register primary channeln");
1730                 goto err_kset_unregister;
1731         }
1732
1733         return 0;
1734
1735 err_kset_unregister:
1736         kset_unregister(child_device_obj->channels_kset);
1737
1738 err_dev_unregister:
1739         device_unregister(&child_device_obj->device);
1740         return ret;
1741 }
1742
1743 /*
1744  * vmbus_device_unregister - Remove the specified child device
1745  * from the vmbus.
1746  */
1747 void vmbus_device_unregister(struct hv_device *device_obj)
1748 {
1749         pr_debug("child device %s unregistered\n",
1750                 dev_name(&device_obj->device));
1751
1752         kset_unregister(device_obj->channels_kset);
1753
1754         /*
1755          * Kick off the process of unregistering the device.
1756          * This will call vmbus_remove() and eventually vmbus_device_release()
1757          */
1758         device_unregister(&device_obj->device);
1759 }
1760
1761
1762 /*
1763  * VMBUS is an acpi enumerated device. Get the information we
1764  * need from DSDT.
1765  */
1766 #define VTPM_BASE_ADDRESS 0xfed40000
1767 static acpi_status vmbus_walk_resources(struct acpi_resource *res, void *ctx)
1768 {
1769         resource_size_t start = 0;
1770         resource_size_t end = 0;
1771         struct resource *new_res;
1772         struct resource **old_res = &hyperv_mmio;
1773         struct resource **prev_res = NULL;
1774
1775         switch (res->type) {
1776
1777         /*
1778          * "Address" descriptors are for bus windows. Ignore
1779          * "memory" descriptors, which are for registers on
1780          * devices.
1781          */
1782         case ACPI_RESOURCE_TYPE_ADDRESS32:
1783                 start = res->data.address32.address.minimum;
1784                 end = res->data.address32.address.maximum;
1785                 break;
1786
1787         case ACPI_RESOURCE_TYPE_ADDRESS64:
1788                 start = res->data.address64.address.minimum;
1789                 end = res->data.address64.address.maximum;
1790                 break;
1791
1792         default:
1793                 /* Unused resource type */
1794                 return AE_OK;
1795
1796         }
1797         /*
1798          * Ignore ranges that are below 1MB, as they're not
1799          * necessary or useful here.
1800          */
1801         if (end < 0x100000)
1802                 return AE_OK;
1803
1804         new_res = kzalloc(sizeof(*new_res), GFP_ATOMIC);
1805         if (!new_res)
1806                 return AE_NO_MEMORY;
1807
1808         /* If this range overlaps the virtual TPM, truncate it. */
1809         if (end > VTPM_BASE_ADDRESS && start < VTPM_BASE_ADDRESS)
1810                 end = VTPM_BASE_ADDRESS;
1811
1812         new_res->name = "hyperv mmio";
1813         new_res->flags = IORESOURCE_MEM;
1814         new_res->start = start;
1815         new_res->end = end;
1816
1817         /*
1818          * If two ranges are adjacent, merge them.
1819          */
1820         do {
1821                 if (!*old_res) {
1822                         *old_res = new_res;
1823                         break;
1824                 }
1825
1826                 if (((*old_res)->end + 1) == new_res->start) {
1827                         (*old_res)->end = new_res->end;
1828                         kfree(new_res);
1829                         break;
1830                 }
1831
1832                 if ((*old_res)->start == new_res->end + 1) {
1833                         (*old_res)->start = new_res->start;
1834                         kfree(new_res);
1835                         break;
1836                 }
1837
1838                 if ((*old_res)->start > new_res->end) {
1839                         new_res->sibling = *old_res;
1840                         if (prev_res)
1841                                 (*prev_res)->sibling = new_res;
1842                         *old_res = new_res;
1843                         break;
1844                 }
1845
1846                 prev_res = old_res;
1847                 old_res = &(*old_res)->sibling;
1848
1849         } while (1);
1850
1851         return AE_OK;
1852 }
1853
1854 static int vmbus_acpi_remove(struct acpi_device *device)
1855 {
1856         struct resource *cur_res;
1857         struct resource *next_res;
1858
1859         if (hyperv_mmio) {
1860                 if (fb_mmio) {
1861                         __release_region(hyperv_mmio, fb_mmio->start,
1862                                          resource_size(fb_mmio));
1863                         fb_mmio = NULL;
1864                 }
1865
1866                 for (cur_res = hyperv_mmio; cur_res; cur_res = next_res) {
1867                         next_res = cur_res->sibling;
1868                         kfree(cur_res);
1869                 }
1870         }
1871
1872         return 0;
1873 }
1874
1875 static void vmbus_reserve_fb(void)
1876 {
1877         int size;
1878         /*
1879          * Make a claim for the frame buffer in the resource tree under the
1880          * first node, which will be the one below 4GB.  The length seems to
1881          * be underreported, particularly in a Generation 1 VM.  So start out
1882          * reserving a larger area and make it smaller until it succeeds.
1883          */
1884
1885         if (screen_info.lfb_base) {
1886                 if (efi_enabled(EFI_BOOT))
1887                         size = max_t(__u32, screen_info.lfb_size, 0x800000);
1888                 else
1889                         size = max_t(__u32, screen_info.lfb_size, 0x4000000);
1890
1891                 for (; !fb_mmio && (size >= 0x100000); size >>= 1) {
1892                         fb_mmio = __request_region(hyperv_mmio,
1893                                                    screen_info.lfb_base, size,
1894                                                    fb_mmio_name, 0);
1895                 }
1896         }
1897 }
1898
1899 /**
1900  * vmbus_allocate_mmio() - Pick a memory-mapped I/O range.
1901  * @new:                If successful, supplied a pointer to the
1902  *                      allocated MMIO space.
1903  * @device_obj:         Identifies the caller
1904  * @min:                Minimum guest physical address of the
1905  *                      allocation
1906  * @max:                Maximum guest physical address
1907  * @size:               Size of the range to be allocated
1908  * @align:              Alignment of the range to be allocated
1909  * @fb_overlap_ok:      Whether this allocation can be allowed
1910  *                      to overlap the video frame buffer.
1911  *
1912  * This function walks the resources granted to VMBus by the
1913  * _CRS object in the ACPI namespace underneath the parent
1914  * "bridge" whether that's a root PCI bus in the Generation 1
1915  * case or a Module Device in the Generation 2 case.  It then
1916  * attempts to allocate from the global MMIO pool in a way that
1917  * matches the constraints supplied in these parameters and by
1918  * that _CRS.
1919  *
1920  * Return: 0 on success, -errno on failure
1921  */
1922 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
1923                         resource_size_t min, resource_size_t max,
1924                         resource_size_t size, resource_size_t align,
1925                         bool fb_overlap_ok)
1926 {
1927         struct resource *iter, *shadow;
1928         resource_size_t range_min, range_max, start;
1929         const char *dev_n = dev_name(&device_obj->device);
1930         int retval;
1931
1932         retval = -ENXIO;
1933         down(&hyperv_mmio_lock);
1934
1935         /*
1936          * If overlaps with frame buffers are allowed, then first attempt to
1937          * make the allocation from within the reserved region.  Because it
1938          * is already reserved, no shadow allocation is necessary.
1939          */
1940         if (fb_overlap_ok && fb_mmio && !(min > fb_mmio->end) &&
1941             !(max < fb_mmio->start)) {
1942
1943                 range_min = fb_mmio->start;
1944                 range_max = fb_mmio->end;
1945                 start = (range_min + align - 1) & ~(align - 1);
1946                 for (; start + size - 1 <= range_max; start += align) {
1947                         *new = request_mem_region_exclusive(start, size, dev_n);
1948                         if (*new) {
1949                                 retval = 0;
1950                                 goto exit;
1951                         }
1952                 }
1953         }
1954
1955         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
1956                 if ((iter->start >= max) || (iter->end <= min))
1957                         continue;
1958
1959                 range_min = iter->start;
1960                 range_max = iter->end;
1961                 start = (range_min + align - 1) & ~(align - 1);
1962                 for (; start + size - 1 <= range_max; start += align) {
1963                         shadow = __request_region(iter, start, size, NULL,
1964                                                   IORESOURCE_BUSY);
1965                         if (!shadow)
1966                                 continue;
1967
1968                         *new = request_mem_region_exclusive(start, size, dev_n);
1969                         if (*new) {
1970                                 shadow->name = (char *)*new;
1971                                 retval = 0;
1972                                 goto exit;
1973                         }
1974
1975                         __release_region(iter, start, size);
1976                 }
1977         }
1978
1979 exit:
1980         up(&hyperv_mmio_lock);
1981         return retval;
1982 }
1983 EXPORT_SYMBOL_GPL(vmbus_allocate_mmio);
1984
1985 /**
1986  * vmbus_free_mmio() - Free a memory-mapped I/O range.
1987  * @start:              Base address of region to release.
1988  * @size:               Size of the range to be allocated
1989  *
1990  * This function releases anything requested by
1991  * vmbus_mmio_allocate().
1992  */
1993 void vmbus_free_mmio(resource_size_t start, resource_size_t size)
1994 {
1995         struct resource *iter;
1996
1997         down(&hyperv_mmio_lock);
1998         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
1999                 if ((iter->start >= start + size) || (iter->end <= start))
2000                         continue;
2001
2002                 __release_region(iter, start, size);
2003         }
2004         release_mem_region(start, size);
2005         up(&hyperv_mmio_lock);
2006
2007 }
2008 EXPORT_SYMBOL_GPL(vmbus_free_mmio);
2009
2010 static int vmbus_acpi_add(struct acpi_device *device)
2011 {
2012         acpi_status result;
2013         int ret_val = -ENODEV;
2014         struct acpi_device *ancestor;
2015
2016         hv_acpi_dev = device;
2017
2018         result = acpi_walk_resources(device->handle, METHOD_NAME__CRS,
2019                                         vmbus_walk_resources, NULL);
2020
2021         if (ACPI_FAILURE(result))
2022                 goto acpi_walk_err;
2023         /*
2024          * Some ancestor of the vmbus acpi device (Gen1 or Gen2
2025          * firmware) is the VMOD that has the mmio ranges. Get that.
2026          */
2027         for (ancestor = device->parent; ancestor; ancestor = ancestor->parent) {
2028                 result = acpi_walk_resources(ancestor->handle, METHOD_NAME__CRS,
2029                                              vmbus_walk_resources, NULL);
2030
2031                 if (ACPI_FAILURE(result))
2032                         continue;
2033                 if (hyperv_mmio) {
2034                         vmbus_reserve_fb();
2035                         break;
2036                 }
2037         }
2038         ret_val = 0;
2039
2040 acpi_walk_err:
2041         complete(&probe_event);
2042         if (ret_val)
2043                 vmbus_acpi_remove(device);
2044         return ret_val;
2045 }
2046
2047 static const struct acpi_device_id vmbus_acpi_device_ids[] = {
2048         {"VMBUS", 0},
2049         {"VMBus", 0},
2050         {"", 0},
2051 };
2052 MODULE_DEVICE_TABLE(acpi, vmbus_acpi_device_ids);
2053
2054 static struct acpi_driver vmbus_acpi_driver = {
2055         .name = "vmbus",
2056         .ids = vmbus_acpi_device_ids,
2057         .ops = {
2058                 .add = vmbus_acpi_add,
2059                 .remove = vmbus_acpi_remove,
2060         },
2061 };
2062
2063 static void hv_kexec_handler(void)
2064 {
2065         hv_stimer_global_cleanup();
2066         vmbus_initiate_unload(false);
2067         vmbus_connection.conn_state = DISCONNECTED;
2068         /* Make sure conn_state is set as hv_synic_cleanup checks for it */
2069         mb();
2070         cpuhp_remove_state(hyperv_cpuhp_online);
2071         hyperv_cleanup();
2072 };
2073
2074 static void hv_crash_handler(struct pt_regs *regs)
2075 {
2076         int cpu;
2077
2078         vmbus_initiate_unload(true);
2079         /*
2080          * In crash handler we can't schedule synic cleanup for all CPUs,
2081          * doing the cleanup for current CPU only. This should be sufficient
2082          * for kdump.
2083          */
2084         vmbus_connection.conn_state = DISCONNECTED;
2085         cpu = smp_processor_id();
2086         hv_stimer_cleanup(cpu);
2087         hv_synic_cleanup(cpu);
2088         hyperv_cleanup();
2089 };
2090
2091 static int __init hv_acpi_init(void)
2092 {
2093         int ret, t;
2094
2095         if (!hv_is_hyperv_initialized())
2096                 return -ENODEV;
2097
2098         init_completion(&probe_event);
2099
2100         /*
2101          * Get ACPI resources first.
2102          */
2103         ret = acpi_bus_register_driver(&vmbus_acpi_driver);
2104
2105         if (ret)
2106                 return ret;
2107
2108         t = wait_for_completion_timeout(&probe_event, 5*HZ);
2109         if (t == 0) {
2110                 ret = -ETIMEDOUT;
2111                 goto cleanup;
2112         }
2113
2114         ret = vmbus_bus_init();
2115         if (ret)
2116                 goto cleanup;
2117
2118         hv_setup_kexec_handler(hv_kexec_handler);
2119         hv_setup_crash_handler(hv_crash_handler);
2120
2121         return 0;
2122
2123 cleanup:
2124         acpi_bus_unregister_driver(&vmbus_acpi_driver);
2125         hv_acpi_dev = NULL;
2126         return ret;
2127 }
2128
2129 static void __exit vmbus_exit(void)
2130 {
2131         int cpu;
2132
2133         hv_remove_kexec_handler();
2134         hv_remove_crash_handler();
2135         vmbus_connection.conn_state = DISCONNECTED;
2136         hv_stimer_global_cleanup();
2137         vmbus_disconnect();
2138         hv_remove_vmbus_irq();
2139         for_each_online_cpu(cpu) {
2140                 struct hv_per_cpu_context *hv_cpu
2141                         = per_cpu_ptr(hv_context.cpu_context, cpu);
2142
2143                 tasklet_kill(&hv_cpu->msg_dpc);
2144         }
2145         vmbus_free_channels();
2146
2147         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
2148                 kmsg_dump_unregister(&hv_kmsg_dumper);
2149                 unregister_die_notifier(&hyperv_die_block);
2150                 atomic_notifier_chain_unregister(&panic_notifier_list,
2151                                                  &hyperv_panic_block);
2152         }
2153
2154         free_page((unsigned long)hv_panic_page);
2155         unregister_sysctl_table(hv_ctl_table_hdr);
2156         hv_ctl_table_hdr = NULL;
2157         bus_unregister(&hv_bus);
2158
2159         cpuhp_remove_state(hyperv_cpuhp_online);
2160         hv_synic_free();
2161         acpi_bus_unregister_driver(&vmbus_acpi_driver);
2162 }
2163
2164
2165 MODULE_LICENSE("GPL");
2166
2167 subsys_initcall(hv_acpi_init);
2168 module_exit(vmbus_exit);