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 <linux/delay.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 <linux/kernel.h>
  34 #include <linux/syscore_ops.h>
  35 #include <clocksource/hyperv_timer.h>
  36 #include "hyperv_vmbus.h"
  37
  38 struct vmbus_dynid {
  39         struct list_head node;
  40         struct hv_vmbus_device_id id;
  41 };
  42
  43 static struct acpi_device  *hv_acpi_dev;
  44
  45 static struct completion probe_event;
  46
  47 static int hyperv_cpuhp_online;
  48
  49 static void *hv_panic_page;
  50
  51 /*
  52  * Boolean to control whether to report panic messages over Hyper-V.
  53  *
  54  * It can be set via /proc/sys/kernel/hyperv/record_panic_msg
  55  */
  56 static int sysctl_record_panic_msg = 1;
  57
  58 static int hyperv_report_reg(void)
  59 {
  60         return !sysctl_record_panic_msg || !hv_panic_page;
  61 }
  62
  63 static int hyperv_panic_event(struct notifier_block *nb, unsigned long val,
  64                               void *args)
  65 {
  66         struct pt_regs *regs;
  67
  68         vmbus_initiate_unload(true);
  69
  70         /*
  71          * Hyper-V should be notified only once about a panic.  If we will be
  72          * doing hyperv_report_panic_msg() later with kmsg data, don't do
  73          * the notification here.
  74          */
  75         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE
  76             && hyperv_report_reg()) {
  77                 regs = current_pt_regs();
  78                 hyperv_report_panic(regs, val, false);
  79         }
  80         return NOTIFY_DONE;
  81 }
  82
  83 static int hyperv_die_event(struct notifier_block *nb, unsigned long val,
  84                             void *args)
  85 {
  86         struct die_args *die = args;
  87         struct pt_regs *regs = die->regs;
  88
  89         /* Don't notify Hyper-V if the die event is other than oops */
  90         if (val != DIE_OOPS)
  91                 return NOTIFY_DONE;
  92
  93         /*
  94          * Hyper-V should be notified only once about a panic.  If we will be
  95          * doing hyperv_report_panic_msg() later with kmsg data, don't do
  96          * the notification here.
  97          */
  98         if (hyperv_report_reg())
  99                 hyperv_report_panic(regs, val, true);
 100         return NOTIFY_DONE;
 101 }
 102
 103 static struct notifier_block hyperv_die_block = {
 104         .notifier_call = hyperv_die_event,
 105 };
 106 static struct notifier_block hyperv_panic_block = {
 107         .notifier_call = hyperv_panic_event,
 108 };
 109
 110 static const char *fb_mmio_name = "fb_range";
 111 static struct resource *fb_mmio;
 112 static struct resource *hyperv_mmio;
 113 static DEFINE_MUTEX(hyperv_mmio_lock);
 114
 115 static int vmbus_exists(void)
 116 {
 117         if (hv_acpi_dev == NULL)
 118                 return -ENODEV;
 119
 120         return 0;
 121 }
 122
 123 static u8 channel_monitor_group(const struct vmbus_channel *channel)
 124 {
 125         return (u8)channel->offermsg.monitorid / 32;
 126 }
 127
 128 static u8 channel_monitor_offset(const struct vmbus_channel *channel)
 129 {
 130         return (u8)channel->offermsg.monitorid % 32;
 131 }
 132
 133 static u32 channel_pending(const struct vmbus_channel *channel,
 134                            const struct hv_monitor_page *monitor_page)
 135 {
 136         u8 monitor_group = channel_monitor_group(channel);
 137
 138         return monitor_page->trigger_group[monitor_group].pending;
 139 }
 140
 141 static u32 channel_latency(const struct vmbus_channel *channel,
 142                            const struct hv_monitor_page *monitor_page)
 143 {
 144         u8 monitor_group = channel_monitor_group(channel);
 145         u8 monitor_offset = channel_monitor_offset(channel);
 146
 147         return monitor_page->latency[monitor_group][monitor_offset];
 148 }
 149
 150 static u32 channel_conn_id(struct vmbus_channel *channel,
 151                            struct hv_monitor_page *monitor_page)
 152 {
 153         u8 monitor_group = channel_monitor_group(channel);
 154         u8 monitor_offset = channel_monitor_offset(channel);
 155         return monitor_page->parameter[monitor_group][monitor_offset].connectionid.u.id;
 156 }
 157
 158 static ssize_t id_show(struct device *dev, struct device_attribute *dev_attr,
 159                        char *buf)
 160 {
 161         struct hv_device *hv_dev = device_to_hv_device(dev);
 162
 163         if (!hv_dev->channel)
 164                 return -ENODEV;
 165         return sprintf(buf, "%d\n", hv_dev->channel->offermsg.child_relid);
 166 }
 167 static DEVICE_ATTR_RO(id);
 168
 169 static ssize_t state_show(struct device *dev, struct device_attribute *dev_attr,
 170                           char *buf)
 171 {
 172         struct hv_device *hv_dev = device_to_hv_device(dev);
 173
 174         if (!hv_dev->channel)
 175                 return -ENODEV;
 176         return sprintf(buf, "%d\n", hv_dev->channel->state);
 177 }
 178 static DEVICE_ATTR_RO(state);
 179
 180 static ssize_t monitor_id_show(struct device *dev,
 181                                struct device_attribute *dev_attr, char *buf)
 182 {
 183         struct hv_device *hv_dev = device_to_hv_device(dev);
 184
 185         if (!hv_dev->channel)
 186                 return -ENODEV;
 187         return sprintf(buf, "%d\n", hv_dev->channel->offermsg.monitorid);
 188 }
 189 static DEVICE_ATTR_RO(monitor_id);
 190
 191 static ssize_t class_id_show(struct device *dev,
 192                                struct device_attribute *dev_attr, char *buf)
 193 {
 194         struct hv_device *hv_dev = device_to_hv_device(dev);
 195
 196         if (!hv_dev->channel)
 197                 return -ENODEV;
 198         return sprintf(buf, "{%pUl}\n",
 199                        &hv_dev->channel->offermsg.offer.if_type);
 200 }
 201 static DEVICE_ATTR_RO(class_id);
 202
 203 static ssize_t device_id_show(struct device *dev,
 204                               struct device_attribute *dev_attr, char *buf)
 205 {
 206         struct hv_device *hv_dev = device_to_hv_device(dev);
 207
 208         if (!hv_dev->channel)
 209                 return -ENODEV;
 210         return sprintf(buf, "{%pUl}\n",
 211                        &hv_dev->channel->offermsg.offer.if_instance);
 212 }
 213 static DEVICE_ATTR_RO(device_id);
 214
 215 static ssize_t modalias_show(struct device *dev,
 216                              struct device_attribute *dev_attr, char *buf)
 217 {
 218         struct hv_device *hv_dev = device_to_hv_device(dev);
 219
 220         return sprintf(buf, "vmbus:%*phN\n", UUID_SIZE, &hv_dev->dev_type);
 221 }
 222 static DEVICE_ATTR_RO(modalias);
 223
 224 #ifdef CONFIG_NUMA
 225 static ssize_t numa_node_show(struct device *dev,
 226                               struct device_attribute *attr, char *buf)
 227 {
 228         struct hv_device *hv_dev = device_to_hv_device(dev);
 229
 230         if (!hv_dev->channel)
 231                 return -ENODEV;
 232
 233         return sprintf(buf, "%d\n", cpu_to_node(hv_dev->channel->target_cpu));
 234 }
 235 static DEVICE_ATTR_RO(numa_node);
 236 #endif
 237
 238 static ssize_t server_monitor_pending_show(struct device *dev,
 239                                            struct device_attribute *dev_attr,
 240                                            char *buf)
 241 {
 242         struct hv_device *hv_dev = device_to_hv_device(dev);
 243
 244         if (!hv_dev->channel)
 245                 return -ENODEV;
 246         return sprintf(buf, "%d\n",
 247                        channel_pending(hv_dev->channel,
 248                                        vmbus_connection.monitor_pages[0]));
 249 }
 250 static DEVICE_ATTR_RO(server_monitor_pending);
 251
 252 static ssize_t client_monitor_pending_show(struct device *dev,
 253                                            struct device_attribute *dev_attr,
 254                                            char *buf)
 255 {
 256         struct hv_device *hv_dev = device_to_hv_device(dev);
 257
 258         if (!hv_dev->channel)
 259                 return -ENODEV;
 260         return sprintf(buf, "%d\n",
 261                        channel_pending(hv_dev->channel,
 262                                        vmbus_connection.monitor_pages[1]));
 263 }
 264 static DEVICE_ATTR_RO(client_monitor_pending);
 265
 266 static ssize_t server_monitor_latency_show(struct device *dev,
 267                                            struct device_attribute *dev_attr,
 268                                            char *buf)
 269 {
 270         struct hv_device *hv_dev = device_to_hv_device(dev);
 271
 272         if (!hv_dev->channel)
 273                 return -ENODEV;
 274         return sprintf(buf, "%d\n",
 275                        channel_latency(hv_dev->channel,
 276                                        vmbus_connection.monitor_pages[0]));
 277 }
 278 static DEVICE_ATTR_RO(server_monitor_latency);
 279
 280 static ssize_t client_monitor_latency_show(struct device *dev,
 281                                            struct device_attribute *dev_attr,
 282                                            char *buf)
 283 {
 284         struct hv_device *hv_dev = device_to_hv_device(dev);
 285
 286         if (!hv_dev->channel)
 287                 return -ENODEV;
 288         return sprintf(buf, "%d\n",
 289                        channel_latency(hv_dev->channel,
 290                                        vmbus_connection.monitor_pages[1]));
 291 }
 292 static DEVICE_ATTR_RO(client_monitor_latency);
 293
 294 static ssize_t server_monitor_conn_id_show(struct device *dev,
 295                                            struct device_attribute *dev_attr,
 296                                            char *buf)
 297 {
 298         struct hv_device *hv_dev = device_to_hv_device(dev);
 299
 300         if (!hv_dev->channel)
 301                 return -ENODEV;
 302         return sprintf(buf, "%d\n",
 303                        channel_conn_id(hv_dev->channel,
 304                                        vmbus_connection.monitor_pages[0]));
 305 }
 306 static DEVICE_ATTR_RO(server_monitor_conn_id);
 307
 308 static ssize_t client_monitor_conn_id_show(struct device *dev,
 309                                            struct device_attribute *dev_attr,
 310                                            char *buf)
 311 {
 312         struct hv_device *hv_dev = device_to_hv_device(dev);
 313
 314         if (!hv_dev->channel)
 315                 return -ENODEV;
 316         return sprintf(buf, "%d\n",
 317                        channel_conn_id(hv_dev->channel,
 318                                        vmbus_connection.monitor_pages[1]));
 319 }
 320 static DEVICE_ATTR_RO(client_monitor_conn_id);
 321
 322 static ssize_t out_intr_mask_show(struct device *dev,
 323                                   struct device_attribute *dev_attr, char *buf)
 324 {
 325         struct hv_device *hv_dev = device_to_hv_device(dev);
 326         struct hv_ring_buffer_debug_info outbound;
 327         int ret;
 328
 329         if (!hv_dev->channel)
 330                 return -ENODEV;
 331
 332         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 333                                           &outbound);
 334         if (ret < 0)
 335                 return ret;
 336
 337         return sprintf(buf, "%d\n", outbound.current_interrupt_mask);
 338 }
 339 static DEVICE_ATTR_RO(out_intr_mask);
 340
 341 static ssize_t out_read_index_show(struct device *dev,
 342                                    struct device_attribute *dev_attr, char *buf)
 343 {
 344         struct hv_device *hv_dev = device_to_hv_device(dev);
 345         struct hv_ring_buffer_debug_info outbound;
 346         int ret;
 347
 348         if (!hv_dev->channel)
 349                 return -ENODEV;
 350
 351         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 352                                           &outbound);
 353         if (ret < 0)
 354                 return ret;
 355         return sprintf(buf, "%d\n", outbound.current_read_index);
 356 }
 357 static DEVICE_ATTR_RO(out_read_index);
 358
 359 static ssize_t out_write_index_show(struct device *dev,
 360                                     struct device_attribute *dev_attr,
 361                                     char *buf)
 362 {
 363         struct hv_device *hv_dev = device_to_hv_device(dev);
 364         struct hv_ring_buffer_debug_info outbound;
 365         int ret;
 366
 367         if (!hv_dev->channel)
 368                 return -ENODEV;
 369
 370         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 371                                           &outbound);
 372         if (ret < 0)
 373                 return ret;
 374         return sprintf(buf, "%d\n", outbound.current_write_index);
 375 }
 376 static DEVICE_ATTR_RO(out_write_index);
 377
 378 static ssize_t out_read_bytes_avail_show(struct device *dev,
 379                                          struct device_attribute *dev_attr,
 380                                          char *buf)
 381 {
 382         struct hv_device *hv_dev = device_to_hv_device(dev);
 383         struct hv_ring_buffer_debug_info outbound;
 384         int ret;
 385
 386         if (!hv_dev->channel)
 387                 return -ENODEV;
 388
 389         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 390                                           &outbound);
 391         if (ret < 0)
 392                 return ret;
 393         return sprintf(buf, "%d\n", outbound.bytes_avail_toread);
 394 }
 395 static DEVICE_ATTR_RO(out_read_bytes_avail);
 396
 397 static ssize_t out_write_bytes_avail_show(struct device *dev,
 398                                           struct device_attribute *dev_attr,
 399                                           char *buf)
 400 {
 401         struct hv_device *hv_dev = device_to_hv_device(dev);
 402         struct hv_ring_buffer_debug_info outbound;
 403         int ret;
 404
 405         if (!hv_dev->channel)
 406                 return -ENODEV;
 407
 408         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 409                                           &outbound);
 410         if (ret < 0)
 411                 return ret;
 412         return sprintf(buf, "%d\n", outbound.bytes_avail_towrite);
 413 }
 414 static DEVICE_ATTR_RO(out_write_bytes_avail);
 415
 416 static ssize_t in_intr_mask_show(struct device *dev,
 417                                  struct device_attribute *dev_attr, char *buf)
 418 {
 419         struct hv_device *hv_dev = device_to_hv_device(dev);
 420         struct hv_ring_buffer_debug_info inbound;
 421         int ret;
 422
 423         if (!hv_dev->channel)
 424                 return -ENODEV;
 425
 426         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 427         if (ret < 0)
 428                 return ret;
 429
 430         return sprintf(buf, "%d\n", inbound.current_interrupt_mask);
 431 }
 432 static DEVICE_ATTR_RO(in_intr_mask);
 433
 434 static ssize_t in_read_index_show(struct device *dev,
 435                                   struct device_attribute *dev_attr, char *buf)
 436 {
 437         struct hv_device *hv_dev = device_to_hv_device(dev);
 438         struct hv_ring_buffer_debug_info inbound;
 439         int ret;
 440
 441         if (!hv_dev->channel)
 442                 return -ENODEV;
 443
 444         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 445         if (ret < 0)
 446                 return ret;
 447
 448         return sprintf(buf, "%d\n", inbound.current_read_index);
 449 }
 450 static DEVICE_ATTR_RO(in_read_index);
 451
 452 static ssize_t in_write_index_show(struct device *dev,
 453                                    struct device_attribute *dev_attr, char *buf)
 454 {
 455         struct hv_device *hv_dev = device_to_hv_device(dev);
 456         struct hv_ring_buffer_debug_info inbound;
 457         int ret;
 458
 459         if (!hv_dev->channel)
 460                 return -ENODEV;
 461
 462         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 463         if (ret < 0)
 464                 return ret;
 465
 466         return sprintf(buf, "%d\n", inbound.current_write_index);
 467 }
 468 static DEVICE_ATTR_RO(in_write_index);
 469
 470 static ssize_t in_read_bytes_avail_show(struct device *dev,
 471                                         struct device_attribute *dev_attr,
 472                                         char *buf)
 473 {
 474         struct hv_device *hv_dev = device_to_hv_device(dev);
 475         struct hv_ring_buffer_debug_info inbound;
 476         int ret;
 477
 478         if (!hv_dev->channel)
 479                 return -ENODEV;
 480
 481         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 482         if (ret < 0)
 483                 return ret;
 484
 485         return sprintf(buf, "%d\n", inbound.bytes_avail_toread);
 486 }
 487 static DEVICE_ATTR_RO(in_read_bytes_avail);
 488
 489 static ssize_t in_write_bytes_avail_show(struct device *dev,
 490                                          struct device_attribute *dev_attr,
 491                                          char *buf)
 492 {
 493         struct hv_device *hv_dev = device_to_hv_device(dev);
 494         struct hv_ring_buffer_debug_info inbound;
 495         int ret;
 496
 497         if (!hv_dev->channel)
 498                 return -ENODEV;
 499
 500         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 501         if (ret < 0)
 502                 return ret;
 503
 504         return sprintf(buf, "%d\n", inbound.bytes_avail_towrite);
 505 }
 506 static DEVICE_ATTR_RO(in_write_bytes_avail);
 507
 508 static ssize_t channel_vp_mapping_show(struct device *dev,
 509                                        struct device_attribute *dev_attr,
 510                                        char *buf)
 511 {
 512         struct hv_device *hv_dev = device_to_hv_device(dev);
 513         struct vmbus_channel *channel = hv_dev->channel, *cur_sc;
 514         int buf_size = PAGE_SIZE, n_written, tot_written;
 515         struct list_head *cur;
 516
 517         if (!channel)
 518                 return -ENODEV;
 519
 520         mutex_lock(&vmbus_connection.channel_mutex);
 521
 522         tot_written = snprintf(buf, buf_size, "%u:%u\n",
 523                 channel->offermsg.child_relid, channel->target_cpu);
 524
 525         list_for_each(cur, &channel->sc_list) {
 526                 if (tot_written >= buf_size - 1)
 527                         break;
 528
 529                 cur_sc = list_entry(cur, struct vmbus_channel, sc_list);
 530                 n_written = scnprintf(buf + tot_written,
 531                                      buf_size - tot_written,
 532                                      "%u:%u\n",
 533                                      cur_sc->offermsg.child_relid,
 534                                      cur_sc->target_cpu);
 535                 tot_written += n_written;
 536         }
 537
 538         mutex_unlock(&vmbus_connection.channel_mutex);
 539
 540         return tot_written;
 541 }
 542 static DEVICE_ATTR_RO(channel_vp_mapping);
 543
 544 static ssize_t vendor_show(struct device *dev,
 545                            struct device_attribute *dev_attr,
 546                            char *buf)
 547 {
 548         struct hv_device *hv_dev = device_to_hv_device(dev);
 549         return sprintf(buf, "0x%x\n", hv_dev->vendor_id);
 550 }
 551 static DEVICE_ATTR_RO(vendor);
 552
 553 static ssize_t device_show(struct device *dev,
 554                            struct device_attribute *dev_attr,
 555                            char *buf)
 556 {
 557         struct hv_device *hv_dev = device_to_hv_device(dev);
 558         return sprintf(buf, "0x%x\n", hv_dev->device_id);
 559 }
 560 static DEVICE_ATTR_RO(device);
 561
 562 static ssize_t driver_override_store(struct device *dev,
 563                                      struct device_attribute *attr,
 564                                      const char *buf, size_t count)
 565 {
 566         struct hv_device *hv_dev = device_to_hv_device(dev);
 567         char *driver_override, *old, *cp;
 568
 569         /* We need to keep extra room for a newline */
 570         if (count >= (PAGE_SIZE - 1))
 571                 return -EINVAL;
 572
 573         driver_override = kstrndup(buf, count, GFP_KERNEL);
 574         if (!driver_override)
 575                 return -ENOMEM;
 576
 577         cp = strchr(driver_override, '\n');
 578         if (cp)
 579                 *cp = '\0';
 580
 581         device_lock(dev);
 582         old = hv_dev->driver_override;
 583         if (strlen(driver_override)) {
 584                 hv_dev->driver_override = driver_override;
 585         } else {
 586                 kfree(driver_override);
 587                 hv_dev->driver_override = NULL;
 588         }
 589         device_unlock(dev);
 590
 591         kfree(old);
 592
 593         return count;
 594 }
 595
 596 static ssize_t driver_override_show(struct device *dev,
 597                                     struct device_attribute *attr, char *buf)
 598 {
 599         struct hv_device *hv_dev = device_to_hv_device(dev);
 600         ssize_t len;
 601
 602         device_lock(dev);
 603         len = snprintf(buf, PAGE_SIZE, "%s\n", hv_dev->driver_override);
 604         device_unlock(dev);
 605
 606         return len;
 607 }
 608 static DEVICE_ATTR_RW(driver_override);
 609
 610 /* Set up per device attributes in /sys/bus/vmbus/devices/<bus device> */
 611 static struct attribute *vmbus_dev_attrs[] = {
 612         &dev_attr_id.attr,
 613         &dev_attr_state.attr,
 614         &dev_attr_monitor_id.attr,
 615         &dev_attr_class_id.attr,
 616         &dev_attr_device_id.attr,
 617         &dev_attr_modalias.attr,
 618 #ifdef CONFIG_NUMA
 619         &dev_attr_numa_node.attr,
 620 #endif
 621         &dev_attr_server_monitor_pending.attr,
 622         &dev_attr_client_monitor_pending.attr,
 623         &dev_attr_server_monitor_latency.attr,
 624         &dev_attr_client_monitor_latency.attr,
 625         &dev_attr_server_monitor_conn_id.attr,
 626         &dev_attr_client_monitor_conn_id.attr,
 627         &dev_attr_out_intr_mask.attr,
 628         &dev_attr_out_read_index.attr,
 629         &dev_attr_out_write_index.attr,
 630         &dev_attr_out_read_bytes_avail.attr,
 631         &dev_attr_out_write_bytes_avail.attr,
 632         &dev_attr_in_intr_mask.attr,
 633         &dev_attr_in_read_index.attr,
 634         &dev_attr_in_write_index.attr,
 635         &dev_attr_in_read_bytes_avail.attr,
 636         &dev_attr_in_write_bytes_avail.attr,
 637         &dev_attr_channel_vp_mapping.attr,
 638         &dev_attr_vendor.attr,
 639         &dev_attr_device.attr,
 640         &dev_attr_driver_override.attr,
 641         NULL,
 642 };
 643
 644 /*
 645  * Device-level attribute_group callback function. Returns the permission for
 646  * each attribute, and returns 0 if an attribute is not visible.
 647  */
 648 static umode_t vmbus_dev_attr_is_visible(struct kobject *kobj,
 649                                          struct attribute *attr, int idx)
 650 {
 651         struct device *dev = kobj_to_dev(kobj);
 652         const struct hv_device *hv_dev = device_to_hv_device(dev);
 653
 654         /* Hide the monitor attributes if the monitor mechanism is not used. */
 655         if (!hv_dev->channel->offermsg.monitor_allocated &&
 656             (attr == &dev_attr_monitor_id.attr ||
 657              attr == &dev_attr_server_monitor_pending.attr ||
 658              attr == &dev_attr_client_monitor_pending.attr ||
 659              attr == &dev_attr_server_monitor_latency.attr ||
 660              attr == &dev_attr_client_monitor_latency.attr ||
 661              attr == &dev_attr_server_monitor_conn_id.attr ||
 662              attr == &dev_attr_client_monitor_conn_id.attr))
 663                 return 0;
 664
 665         return attr->mode;
 666 }
 667
 668 static const struct attribute_group vmbus_dev_group = {
 669         .attrs = vmbus_dev_attrs,
 670         .is_visible = vmbus_dev_attr_is_visible
 671 };
 672 __ATTRIBUTE_GROUPS(vmbus_dev);
 673
 674 /*
 675  * vmbus_uevent - add uevent for our device
 676  *
 677  * This routine is invoked when a device is added or removed on the vmbus to
 678  * generate a uevent to udev in the userspace. The udev will then look at its
 679  * rule and the uevent generated here to load the appropriate driver
 680  *
 681  * The alias string will be of the form vmbus:guid where guid is the string
 682  * representation of the device guid (each byte of the guid will be
 683  * represented with two hex characters.
 684  */
 685 static int vmbus_uevent(struct device *device, struct kobj_uevent_env *env)
 686 {
 687         struct hv_device *dev = device_to_hv_device(device);
 688         const char *format = "MODALIAS=vmbus:%*phN";
 689
 690         return add_uevent_var(env, format, UUID_SIZE, &dev->dev_type);
 691 }
 692
 693 static const struct hv_vmbus_device_id *
 694 hv_vmbus_dev_match(const struct hv_vmbus_device_id *id, const guid_t *guid)
 695 {
 696         if (id == NULL)
 697                 return NULL; /* empty device table */
 698
 699         for (; !guid_is_null(&id->guid); id++)
 700                 if (guid_equal(&id->guid, guid))
 701                         return id;
 702
 703         return NULL;
 704 }
 705
 706 static const struct hv_vmbus_device_id *
 707 hv_vmbus_dynid_match(struct hv_driver *drv, const guid_t *guid)
 708 {
 709         const struct hv_vmbus_device_id *id = NULL;
 710         struct vmbus_dynid *dynid;
 711
 712         spin_lock(&drv->dynids.lock);
 713         list_for_each_entry(dynid, &drv->dynids.list, node) {
 714                 if (guid_equal(&dynid->id.guid, guid)) {
 715                         id = &dynid->id;
 716                         break;
 717                 }
 718         }
 719         spin_unlock(&drv->dynids.lock);
 720
 721         return id;
 722 }
 723
 724 static const struct hv_vmbus_device_id vmbus_device_null;
 725
 726 /*
 727  * Return a matching hv_vmbus_device_id pointer.
 728  * If there is no match, return NULL.
 729  */
 730 static const struct hv_vmbus_device_id *hv_vmbus_get_id(struct hv_driver *drv,
 731                                                         struct hv_device *dev)
 732 {
 733         const guid_t *guid = &dev->dev_type;
 734         const struct hv_vmbus_device_id *id;
 735
 736         /* When driver_override is set, only bind to the matching driver */
 737         if (dev->driver_override && strcmp(dev->driver_override, drv->name))
 738                 return NULL;
 739
 740         /* Look at the dynamic ids first, before the static ones */
 741         id = hv_vmbus_dynid_match(drv, guid);
 742         if (!id)
 743                 id = hv_vmbus_dev_match(drv->id_table, guid);
 744
 745         /* driver_override will always match, send a dummy id */
 746         if (!id && dev->driver_override)
 747                 id = &vmbus_device_null;
 748
 749         return id;
 750 }
 751
 752 /* vmbus_add_dynid - add a new device ID to this driver and re-probe devices */
 753 static int vmbus_add_dynid(struct hv_driver *drv, guid_t *guid)
 754 {
 755         struct vmbus_dynid *dynid;
 756
 757         dynid = kzalloc(sizeof(*dynid), GFP_KERNEL);
 758         if (!dynid)
 759                 return -ENOMEM;
 760
 761         dynid->id.guid = *guid;
 762
 763         spin_lock(&drv->dynids.lock);
 764         list_add_tail(&dynid->node, &drv->dynids.list);
 765         spin_unlock(&drv->dynids.lock);
 766
 767         return driver_attach(&drv->driver);
 768 }
 769
 770 static void vmbus_free_dynids(struct hv_driver *drv)
 771 {
 772         struct vmbus_dynid *dynid, *n;
 773
 774         spin_lock(&drv->dynids.lock);
 775         list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
 776                 list_del(&dynid->node);
 777                 kfree(dynid);
 778         }
 779         spin_unlock(&drv->dynids.lock);
 780 }
 781
 782 /*
 783  * store_new_id - sysfs frontend to vmbus_add_dynid()
 784  *
 785  * Allow GUIDs to be added to an existing driver via sysfs.
 786  */
 787 static ssize_t new_id_store(struct device_driver *driver, const char *buf,
 788                             size_t count)
 789 {
 790         struct hv_driver *drv = drv_to_hv_drv(driver);
 791         guid_t guid;
 792         ssize_t retval;
 793
 794         retval = guid_parse(buf, &guid);
 795         if (retval)
 796                 return retval;
 797
 798         if (hv_vmbus_dynid_match(drv, &guid))
 799                 return -EEXIST;
 800
 801         retval = vmbus_add_dynid(drv, &guid);
 802         if (retval)
 803                 return retval;
 804         return count;
 805 }
 806 static DRIVER_ATTR_WO(new_id);
 807
 808 /*
 809  * store_remove_id - remove a PCI device ID from this driver
 810  *
 811  * Removes a dynamic pci device ID to this driver.
 812  */
 813 static ssize_t remove_id_store(struct device_driver *driver, const char *buf,
 814                                size_t count)
 815 {
 816         struct hv_driver *drv = drv_to_hv_drv(driver);
 817         struct vmbus_dynid *dynid, *n;
 818         guid_t guid;
 819         ssize_t retval;
 820
 821         retval = guid_parse(buf, &guid);
 822         if (retval)
 823                 return retval;
 824
 825         retval = -ENODEV;
 826         spin_lock(&drv->dynids.lock);
 827         list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
 828                 struct hv_vmbus_device_id *id = &dynid->id;
 829
 830                 if (guid_equal(&id->guid, &guid)) {
 831                         list_del(&dynid->node);
 832                         kfree(dynid);
 833                         retval = count;
 834                         break;
 835                 }
 836         }
 837         spin_unlock(&drv->dynids.lock);
 838
 839         return retval;
 840 }
 841 static DRIVER_ATTR_WO(remove_id);
 842
 843 static struct attribute *vmbus_drv_attrs[] = {
 844         &driver_attr_new_id.attr,
 845         &driver_attr_remove_id.attr,
 846         NULL,
 847 };
 848 ATTRIBUTE_GROUPS(vmbus_drv);
 849
 850
 851 /*
 852  * vmbus_match - Attempt to match the specified device to the specified driver
 853  */
 854 static int vmbus_match(struct device *device, struct device_driver *driver)
 855 {
 856         struct hv_driver *drv = drv_to_hv_drv(driver);
 857         struct hv_device *hv_dev = device_to_hv_device(device);
 858
 859         /* The hv_sock driver handles all hv_sock offers. */
 860         if (is_hvsock_channel(hv_dev->channel))
 861                 return drv->hvsock;
 862
 863         if (hv_vmbus_get_id(drv, hv_dev))
 864                 return 1;
 865
 866         return 0;
 867 }
 868
 869 /*
 870  * vmbus_probe - Add the new vmbus's child device
 871  */
 872 static int vmbus_probe(struct device *child_device)
 873 {
 874         int ret = 0;
 875         struct hv_driver *drv =
 876                         drv_to_hv_drv(child_device->driver);
 877         struct hv_device *dev = device_to_hv_device(child_device);
 878         const struct hv_vmbus_device_id *dev_id;
 879
 880         dev_id = hv_vmbus_get_id(drv, dev);
 881         if (drv->probe) {
 882                 ret = drv->probe(dev, dev_id);
 883                 if (ret != 0)
 884                         pr_err("probe failed for device %s (%d)\n",
 885                                dev_name(child_device), ret);
 886
 887         } else {
 888                 pr_err("probe not set for driver %s\n",
 889                        dev_name(child_device));
 890                 ret = -ENODEV;
 891         }
 892         return ret;
 893 }
 894
 895 /*
 896  * vmbus_remove - Remove a vmbus device
 897  */
 898 static int vmbus_remove(struct device *child_device)
 899 {
 900         struct hv_driver *drv;
 901         struct hv_device *dev = device_to_hv_device(child_device);
 902
 903         if (child_device->driver) {
 904                 drv = drv_to_hv_drv(child_device->driver);
 905                 if (drv->remove)
 906                         drv->remove(dev);
 907         }
 908
 909         return 0;
 910 }
 911
 912
 913 /*
 914  * vmbus_shutdown - Shutdown a vmbus device
 915  */
 916 static void vmbus_shutdown(struct device *child_device)
 917 {
 918         struct hv_driver *drv;
 919         struct hv_device *dev = device_to_hv_device(child_device);
 920
 921
 922         /* The device may not be attached yet */
 923         if (!child_device->driver)
 924                 return;
 925
 926         drv = drv_to_hv_drv(child_device->driver);
 927
 928         if (drv->shutdown)
 929                 drv->shutdown(dev);
 930 }
 931
 932 #ifdef CONFIG_PM_SLEEP
 933 /*
 934  * vmbus_suspend - Suspend a vmbus device
 935  */
 936 static int vmbus_suspend(struct device *child_device)
 937 {
 938         struct hv_driver *drv;
 939         struct hv_device *dev = device_to_hv_device(child_device);
 940
 941         /* The device may not be attached yet */
 942         if (!child_device->driver)
 943                 return 0;
 944
 945         drv = drv_to_hv_drv(child_device->driver);
 946         if (!drv->suspend)
 947                 return -EOPNOTSUPP;
 948
 949         return drv->suspend(dev);
 950 }
 951
 952 /*
 953  * vmbus_resume - Resume a vmbus device
 954  */
 955 static int vmbus_resume(struct device *child_device)
 956 {
 957         struct hv_driver *drv;
 958         struct hv_device *dev = device_to_hv_device(child_device);
 959
 960         /* The device may not be attached yet */
 961         if (!child_device->driver)
 962                 return 0;
 963
 964         drv = drv_to_hv_drv(child_device->driver);
 965         if (!drv->resume)
 966                 return -EOPNOTSUPP;
 967
 968         return drv->resume(dev);
 969 }
 970 #else
 971 #define vmbus_suspend NULL
 972 #define vmbus_resume NULL
 973 #endif /* CONFIG_PM_SLEEP */
 974
 975 /*
 976  * vmbus_device_release - Final callback release of the vmbus child device
 977  */
 978 static void vmbus_device_release(struct device *device)
 979 {
 980         struct hv_device *hv_dev = device_to_hv_device(device);
 981         struct vmbus_channel *channel = hv_dev->channel;
 982
 983         hv_debug_rm_dev_dir(hv_dev);
 984
 985         mutex_lock(&vmbus_connection.channel_mutex);
 986         hv_process_channel_removal(channel);
 987         mutex_unlock(&vmbus_connection.channel_mutex);
 988         kfree(hv_dev);
 989 }
 990
 991 /*
 992  * Note: we must use the "noirq" ops: see the comment before vmbus_bus_pm.
 993  *
 994  * suspend_noirq/resume_noirq are set to NULL to support Suspend-to-Idle: we
 995  * shouldn't suspend the vmbus devices upon Suspend-to-Idle, otherwise there
 996  * is no way to wake up a Generation-2 VM.
 997  *
 998  * The other 4 ops are for hibernation.
 999  */
1000
1001 static const struct dev_pm_ops vmbus_pm = {
1002         .suspend_noirq  = NULL,
1003         .resume_noirq   = NULL,
1004         .freeze_noirq   = vmbus_suspend,
1005         .thaw_noirq     = vmbus_resume,
1006         .poweroff_noirq = vmbus_suspend,
1007         .restore_noirq  = vmbus_resume,
1008 };
1009
1010 /* The one and only one */
1011 static struct bus_type  hv_bus = {
1012         .name =         "vmbus",
1013         .match =                vmbus_match,
1014         .shutdown =             vmbus_shutdown,
1015         .remove =               vmbus_remove,
1016         .probe =                vmbus_probe,
1017         .uevent =               vmbus_uevent,
1018         .dev_groups =           vmbus_dev_groups,
1019         .drv_groups =           vmbus_drv_groups,
1020         .pm =                   &vmbus_pm,
1021 };
1022
1023 struct onmessage_work_context {
1024         struct work_struct work;
1025         struct {
1026                 struct hv_message_header header;
1027                 u8 payload[];
1028         } msg;
1029 };
1030
1031 static void vmbus_onmessage_work(struct work_struct *work)
1032 {
1033         struct onmessage_work_context *ctx;
1034
1035         /* Do not process messages if we're in DISCONNECTED state */
1036         if (vmbus_connection.conn_state == DISCONNECTED)
1037                 return;
1038
1039         ctx = container_of(work, struct onmessage_work_context,
1040                            work);
1041         vmbus_onmessage((struct vmbus_channel_message_header *)
1042                         &ctx->msg.payload);
1043         kfree(ctx);
1044 }
1045
1046 void vmbus_on_msg_dpc(unsigned long data)
1047 {
1048         struct hv_per_cpu_context *hv_cpu = (void *)data;
1049         void *page_addr = hv_cpu->synic_message_page;
1050         struct hv_message *msg = (struct hv_message *)page_addr +
1051                                   VMBUS_MESSAGE_SINT;
1052         struct vmbus_channel_message_header *hdr;
1053         const struct vmbus_channel_message_table_entry *entry;
1054         struct onmessage_work_context *ctx;
1055         u32 message_type = msg->header.message_type;
1056
1057         /*
1058          * 'enum vmbus_channel_message_type' is supposed to always be 'u32' as
1059          * it is being used in 'struct vmbus_channel_message_header' definition
1060          * which is supposed to match hypervisor ABI.
1061          */
1062         BUILD_BUG_ON(sizeof(enum vmbus_channel_message_type) != sizeof(u32));
1063
1064         if (message_type == HVMSG_NONE)
1065                 /* no msg */
1066                 return;
1067
1068         hdr = (struct vmbus_channel_message_header *)msg->u.payload;
1069
1070         trace_vmbus_on_msg_dpc(hdr);
1071
1072         if (hdr->msgtype >= CHANNELMSG_COUNT) {
1073                 WARN_ONCE(1, "unknown msgtype=%d\n", hdr->msgtype);
1074                 goto msg_handled;
1075         }
1076
1077         if (msg->header.payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT) {
1078                 WARN_ONCE(1, "payload size is too large (%d)\n",
1079                           msg->header.payload_size);
1080                 goto msg_handled;
1081         }
1082
1083         entry = &channel_message_table[hdr->msgtype];
1084
1085         if (!entry->message_handler)
1086                 goto msg_handled;
1087
1088         if (msg->header.payload_size < entry->min_payload_len) {
1089                 WARN_ONCE(1, "message too short: msgtype=%d len=%d\n",
1090                           hdr->msgtype, msg->header.payload_size);
1091                 goto msg_handled;
1092         }
1093
1094         if (entry->handler_type == VMHT_BLOCKING) {
1095                 ctx = kmalloc(sizeof(*ctx) + msg->header.payload_size,
1096                               GFP_ATOMIC);
1097                 if (ctx == NULL)
1098                         return;
1099
1100                 INIT_WORK(&ctx->work, vmbus_onmessage_work);
1101                 memcpy(&ctx->msg, msg, sizeof(msg->header) +
1102                        msg->header.payload_size);
1103
1104                 /*
1105                  * The host can generate a rescind message while we
1106                  * may still be handling the original offer. We deal with
1107                  * this condition by relying on the synchronization provided
1108                  * by offer_in_progress and by channel_mutex.  See also the
1109                  * inline comments in vmbus_onoffer_rescind().
1110                  */
1111                 switch (hdr->msgtype) {
1112                 case CHANNELMSG_RESCIND_CHANNELOFFER:
1113                         /*
1114                          * If we are handling the rescind message;
1115                          * schedule the work on the global work queue.
1116                          *
1117                          * The OFFER message and the RESCIND message should
1118                          * not be handled by the same serialized work queue,
1119                          * because the OFFER handler may call vmbus_open(),
1120                          * which tries to open the channel by sending an
1121                          * OPEN_CHANNEL message to the host and waits for
1122                          * the host's response; however, if the host has
1123                          * rescinded the channel before it receives the
1124                          * OPEN_CHANNEL message, the host just silently
1125                          * ignores the OPEN_CHANNEL message; as a result,
1126                          * the guest's OFFER handler hangs for ever, if we
1127                          * handle the RESCIND message in the same serialized
1128                          * work queue: the RESCIND handler can not start to
1129                          * run before the OFFER handler finishes.
1130                          */
1131                         schedule_work(&ctx->work);
1132                         break;
1133
1134                 case CHANNELMSG_OFFERCHANNEL:
1135                         /*
1136                          * The host sends the offer message of a given channel
1137                          * before sending the rescind message of the same
1138                          * channel.  These messages are sent to the guest's
1139                          * connect CPU; the guest then starts processing them
1140                          * in the tasklet handler on this CPU:
1141                          *
1142                          * VMBUS_CONNECT_CPU
1143                          *
1144                          * [vmbus_on_msg_dpc()]
1145                          * atomic_inc()  // CHANNELMSG_OFFERCHANNEL
1146                          * queue_work()
1147                          * ...
1148                          * [vmbus_on_msg_dpc()]
1149                          * schedule_work()  // CHANNELMSG_RESCIND_CHANNELOFFER
1150                          *
1151                          * We rely on the memory-ordering properties of the
1152                          * queue_work() and schedule_work() primitives, which
1153                          * guarantee that the atomic increment will be visible
1154                          * to the CPUs which will execute the offer & rescind
1155                          * works by the time these works will start execution.
1156                          */
1157                         atomic_inc(&vmbus_connection.offer_in_progress);
1158                         fallthrough;
1159
1160                 default:
1161                         queue_work(vmbus_connection.work_queue, &ctx->work);
1162                 }
1163         } else
1164                 entry->message_handler(hdr);
1165
1166 msg_handled:
1167         vmbus_signal_eom(msg, message_type);
1168 }
1169
1170 #ifdef CONFIG_PM_SLEEP
1171 /*
1172  * Fake RESCIND_CHANNEL messages to clean up hv_sock channels by force for
1173  * hibernation, because hv_sock connections can not persist across hibernation.
1174  */
1175 static void vmbus_force_channel_rescinded(struct vmbus_channel *channel)
1176 {
1177         struct onmessage_work_context *ctx;
1178         struct vmbus_channel_rescind_offer *rescind;
1179
1180         WARN_ON(!is_hvsock_channel(channel));
1181
1182         /*
1183          * Allocation size is small and the allocation should really not fail,
1184          * otherwise the state of the hv_sock connections ends up in limbo.
1185          */
1186         ctx = kzalloc(sizeof(*ctx) + sizeof(*rescind),
1187                       GFP_KERNEL | __GFP_NOFAIL);
1188
1189         /*
1190          * So far, these are not really used by Linux. Just set them to the
1191          * reasonable values conforming to the definitions of the fields.
1192          */
1193         ctx->msg.header.message_type = 1;
1194         ctx->msg.header.payload_size = sizeof(*rescind);
1195
1196         /* These values are actually used by Linux. */
1197         rescind = (struct vmbus_channel_rescind_offer *)ctx->msg.payload;
1198         rescind->header.msgtype = CHANNELMSG_RESCIND_CHANNELOFFER;
1199         rescind->child_relid = channel->offermsg.child_relid;
1200
1201         INIT_WORK(&ctx->work, vmbus_onmessage_work);
1202
1203         queue_work(vmbus_connection.work_queue, &ctx->work);
1204 }
1205 #endif /* CONFIG_PM_SLEEP */
1206
1207 /*
1208  * Schedule all channels with events pending
1209  */
1210 static void vmbus_chan_sched(struct hv_per_cpu_context *hv_cpu)
1211 {
1212         unsigned long *recv_int_page;
1213         u32 maxbits, relid;
1214
1215         if (vmbus_proto_version < VERSION_WIN8) {
1216                 maxbits = MAX_NUM_CHANNELS_SUPPORTED;
1217                 recv_int_page = vmbus_connection.recv_int_page;
1218         } else {
1219                 /*
1220                  * When the host is win8 and beyond, the event page
1221                  * can be directly checked to get the id of the channel
1222                  * that has the interrupt pending.
1223                  */
1224                 void *page_addr = hv_cpu->synic_event_page;
1225                 union hv_synic_event_flags *event
1226                         = (union hv_synic_event_flags *)page_addr +
1227                                                  VMBUS_MESSAGE_SINT;
1228
1229                 maxbits = HV_EVENT_FLAGS_COUNT;
1230                 recv_int_page = event->flags;
1231         }
1232
1233         if (unlikely(!recv_int_page))
1234                 return;
1235
1236         for_each_set_bit(relid, recv_int_page, maxbits) {
1237                 void (*callback_fn)(void *context);
1238                 struct vmbus_channel *channel;
1239
1240                 if (!sync_test_and_clear_bit(relid, recv_int_page))
1241                         continue;
1242
1243                 /* Special case - vmbus channel protocol msg */
1244                 if (relid == 0)
1245                         continue;
1246
1247                 /*
1248                  * Pairs with the kfree_rcu() in vmbus_chan_release().
1249                  * Guarantees that the channel data structure doesn't
1250                  * get freed while the channel pointer below is being
1251                  * dereferenced.
1252                  */
1253                 rcu_read_lock();
1254
1255                 /* Find channel based on relid */
1256                 channel = relid2channel(relid);
1257                 if (channel == NULL)
1258                         goto sched_unlock_rcu;
1259
1260                 if (channel->rescind)
1261                         goto sched_unlock_rcu;
1262
1263                 /*
1264                  * Make sure that the ring buffer data structure doesn't get
1265                  * freed while we dereference the ring buffer pointer.  Test
1266                  * for the channel's onchannel_callback being NULL within a
1267                  * sched_lock critical section.  See also the inline comments
1268                  * in vmbus_reset_channel_cb().
1269                  */
1270                 spin_lock(&channel->sched_lock);
1271
1272                 callback_fn = channel->onchannel_callback;
1273                 if (unlikely(callback_fn == NULL))
1274                         goto sched_unlock;
1275
1276                 trace_vmbus_chan_sched(channel);
1277
1278                 ++channel->interrupts;
1279
1280                 switch (channel->callback_mode) {
1281                 case HV_CALL_ISR:
1282                         (*callback_fn)(channel->channel_callback_context);
1283                         break;
1284
1285                 case HV_CALL_BATCHED:
1286                         hv_begin_read(&channel->inbound);
1287                         fallthrough;
1288                 case HV_CALL_DIRECT:
1289                         tasklet_schedule(&channel->callback_event);
1290                 }
1291
1292 sched_unlock:
1293                 spin_unlock(&channel->sched_lock);
1294 sched_unlock_rcu:
1295                 rcu_read_unlock();
1296         }
1297 }
1298
1299 static void vmbus_isr(void)
1300 {
1301         struct hv_per_cpu_context *hv_cpu
1302                 = this_cpu_ptr(hv_context.cpu_context);
1303         void *page_addr = hv_cpu->synic_event_page;
1304         struct hv_message *msg;
1305         union hv_synic_event_flags *event;
1306         bool handled = false;
1307
1308         if (unlikely(page_addr == NULL))
1309                 return;
1310
1311         event = (union hv_synic_event_flags *)page_addr +
1312                                          VMBUS_MESSAGE_SINT;
1313         /*
1314          * Check for events before checking for messages. This is the order
1315          * in which events and messages are checked in Windows guests on
1316          * Hyper-V, and the Windows team suggested we do the same.
1317          */
1318
1319         if ((vmbus_proto_version == VERSION_WS2008) ||
1320                 (vmbus_proto_version == VERSION_WIN7)) {
1321
1322                 /* Since we are a child, we only need to check bit 0 */
1323                 if (sync_test_and_clear_bit(0, event->flags))
1324                         handled = true;
1325         } else {
1326                 /*
1327                  * Our host is win8 or above. The signaling mechanism
1328                  * has changed and we can directly look at the event page.
1329                  * If bit n is set then we have an interrup on the channel
1330                  * whose id is n.
1331                  */
1332                 handled = true;
1333         }
1334
1335         if (handled)
1336                 vmbus_chan_sched(hv_cpu);
1337
1338         page_addr = hv_cpu->synic_message_page;
1339         msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
1340
1341         /* Check if there are actual msgs to be processed */
1342         if (msg->header.message_type != HVMSG_NONE) {
1343                 if (msg->header.message_type == HVMSG_TIMER_EXPIRED) {
1344                         hv_stimer0_isr();
1345                         vmbus_signal_eom(msg, HVMSG_TIMER_EXPIRED);
1346                 } else
1347                         tasklet_schedule(&hv_cpu->msg_dpc);
1348         }
1349
1350         add_interrupt_randomness(HYPERVISOR_CALLBACK_VECTOR, 0);
1351 }
1352
1353 /*
1354  * Callback from kmsg_dump. Grab as much as possible from the end of the kmsg
1355  * buffer and call into Hyper-V to transfer the data.
1356  */
1357 static void hv_kmsg_dump(struct kmsg_dumper *dumper,
1358                          enum kmsg_dump_reason reason)
1359 {
1360         size_t bytes_written;
1361         phys_addr_t panic_pa;
1362
1363         /* We are only interested in panics. */
1364         if ((reason != KMSG_DUMP_PANIC) || (!sysctl_record_panic_msg))
1365                 return;
1366
1367         panic_pa = virt_to_phys(hv_panic_page);
1368
1369         /*
1370          * Write dump contents to the page. No need to synchronize; panic should
1371          * be single-threaded.
1372          */
1373         kmsg_dump_get_buffer(dumper, false, hv_panic_page, HV_HYP_PAGE_SIZE,
1374                              &bytes_written);
1375         if (bytes_written)
1376                 hyperv_report_panic_msg(panic_pa, bytes_written);
1377 }
1378
1379 static struct kmsg_dumper hv_kmsg_dumper = {
1380         .dump = hv_kmsg_dump,
1381 };
1382
1383 static struct ctl_table_header *hv_ctl_table_hdr;
1384
1385 /*
1386  * sysctl option to allow the user to control whether kmsg data should be
1387  * reported to Hyper-V on panic.
1388  */
1389 static struct ctl_table hv_ctl_table[] = {
1390         {
1391                 .procname       = "hyperv_record_panic_msg",
1392                 .data           = &sysctl_record_panic_msg,
1393                 .maxlen         = sizeof(int),
1394                 .mode           = 0644,
1395                 .proc_handler   = proc_dointvec_minmax,
1396                 .extra1         = SYSCTL_ZERO,
1397                 .extra2         = SYSCTL_ONE
1398         },
1399         {}
1400 };
1401
1402 static struct ctl_table hv_root_table[] = {
1403         {
1404                 .procname       = "kernel",
1405                 .mode           = 0555,
1406                 .child          = hv_ctl_table
1407         },
1408         {}
1409 };
1410
1411 /*
1412  * vmbus_bus_init -Main vmbus driver initialization routine.
1413  *
1414  * Here, we
1415  *      - initialize the vmbus driver context
1416  *      - invoke the vmbus hv main init routine
1417  *      - retrieve the channel offers
1418  */
1419 static int vmbus_bus_init(void)
1420 {
1421         int ret;
1422
1423         ret = hv_init();
1424         if (ret != 0) {
1425                 pr_err("Unable to initialize the hypervisor - 0x%x\n", ret);
1426                 return ret;
1427         }
1428
1429         ret = bus_register(&hv_bus);
1430         if (ret)
1431                 return ret;
1432
1433         hv_setup_vmbus_irq(vmbus_isr);
1434
1435         ret = hv_synic_alloc();
1436         if (ret)
1437                 goto err_alloc;
1438
1439         /*
1440          * Initialize the per-cpu interrupt state and stimer state.
1441          * Then connect to the host.
1442          */
1443         ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "hyperv/vmbus:online",
1444                                 hv_synic_init, hv_synic_cleanup);
1445         if (ret < 0)
1446                 goto err_cpuhp;
1447         hyperv_cpuhp_online = ret;
1448
1449         ret = vmbus_connect();
1450         if (ret)
1451                 goto err_connect;
1452
1453         /*
1454          * Only register if the crash MSRs are available
1455          */
1456         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1457                 u64 hyperv_crash_ctl;
1458                 /*
1459                  * Sysctl registration is not fatal, since by default
1460                  * reporting is enabled.
1461                  */
1462                 hv_ctl_table_hdr = register_sysctl_table(hv_root_table);
1463                 if (!hv_ctl_table_hdr)
1464                         pr_err("Hyper-V: sysctl table register error");
1465
1466                 /*
1467                  * Register for panic kmsg callback only if the right
1468                  * capability is supported by the hypervisor.
1469                  */
1470                 hv_get_crash_ctl(hyperv_crash_ctl);
1471                 if (hyperv_crash_ctl & HV_CRASH_CTL_CRASH_NOTIFY_MSG) {
1472                         hv_panic_page = (void *)hv_alloc_hyperv_zeroed_page();
1473                         if (hv_panic_page) {
1474                                 ret = kmsg_dump_register(&hv_kmsg_dumper);
1475                                 if (ret) {
1476                                         pr_err("Hyper-V: kmsg dump register "
1477                                                 "error 0x%x\n", ret);
1478                                         hv_free_hyperv_page(
1479                                             (unsigned long)hv_panic_page);
1480                                         hv_panic_page = NULL;
1481                                 }
1482                         } else
1483                                 pr_err("Hyper-V: panic message page memory "
1484                                         "allocation failed");
1485                 }
1486
1487                 register_die_notifier(&hyperv_die_block);
1488         }
1489
1490         /*
1491          * Always register the panic notifier because we need to unload
1492          * the VMbus channel connection to prevent any VMbus
1493          * activity after the VM panics.
1494          */
1495         atomic_notifier_chain_register(&panic_notifier_list,
1496                                &hyperv_panic_block);
1497
1498         vmbus_request_offers();
1499
1500         return 0;
1501
1502 err_connect:
1503         cpuhp_remove_state(hyperv_cpuhp_online);
1504 err_cpuhp:
1505         hv_synic_free();
1506 err_alloc:
1507         hv_remove_vmbus_irq();
1508
1509         bus_unregister(&hv_bus);
1510         unregister_sysctl_table(hv_ctl_table_hdr);
1511         hv_ctl_table_hdr = NULL;
1512         return ret;
1513 }
1514
1515 /**
1516  * __vmbus_child_driver_register() - Register a vmbus's driver
1517  * @hv_driver: Pointer to driver structure you want to register
1518  * @owner: owner module of the drv
1519  * @mod_name: module name string
1520  *
1521  * Registers the given driver with Linux through the 'driver_register()' call
1522  * and sets up the hyper-v vmbus handling for this driver.
1523  * It will return the state of the 'driver_register()' call.
1524  *
1525  */
1526 int __vmbus_driver_register(struct hv_driver *hv_driver, struct module *owner, const char *mod_name)
1527 {
1528         int ret;
1529
1530         pr_info("registering driver %s\n", hv_driver->name);
1531
1532         ret = vmbus_exists();
1533         if (ret < 0)
1534                 return ret;
1535
1536         hv_driver->driver.name = hv_driver->name;
1537         hv_driver->driver.owner = owner;
1538         hv_driver->driver.mod_name = mod_name;
1539         hv_driver->driver.bus = &hv_bus;
1540
1541         spin_lock_init(&hv_driver->dynids.lock);
1542         INIT_LIST_HEAD(&hv_driver->dynids.list);
1543
1544         ret = driver_register(&hv_driver->driver);
1545
1546         return ret;
1547 }
1548 EXPORT_SYMBOL_GPL(__vmbus_driver_register);
1549
1550 /**
1551  * vmbus_driver_unregister() - Unregister a vmbus's driver
1552  * @hv_driver: Pointer to driver structure you want to
1553  *             un-register
1554  *
1555  * Un-register the given driver that was previous registered with a call to
1556  * vmbus_driver_register()
1557  */
1558 void vmbus_driver_unregister(struct hv_driver *hv_driver)
1559 {
1560         pr_info("unregistering driver %s\n", hv_driver->name);
1561
1562         if (!vmbus_exists()) {
1563                 driver_unregister(&hv_driver->driver);
1564                 vmbus_free_dynids(hv_driver);
1565         }
1566 }
1567 EXPORT_SYMBOL_GPL(vmbus_driver_unregister);
1568
1569
1570 /*
1571  * Called when last reference to channel is gone.
1572  */
1573 static void vmbus_chan_release(struct kobject *kobj)
1574 {
1575         struct vmbus_channel *channel
1576                 = container_of(kobj, struct vmbus_channel, kobj);
1577
1578         kfree_rcu(channel, rcu);
1579 }
1580
1581 struct vmbus_chan_attribute {
1582         struct attribute attr;
1583         ssize_t (*show)(struct vmbus_channel *chan, char *buf);
1584         ssize_t (*store)(struct vmbus_channel *chan,
1585                          const char *buf, size_t count);
1586 };
1587 #define VMBUS_CHAN_ATTR(_name, _mode, _show, _store) \
1588         struct vmbus_chan_attribute chan_attr_##_name \
1589                 = __ATTR(_name, _mode, _show, _store)
1590 #define VMBUS_CHAN_ATTR_RW(_name) \
1591         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RW(_name)
1592 #define VMBUS_CHAN_ATTR_RO(_name) \
1593         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RO(_name)
1594 #define VMBUS_CHAN_ATTR_WO(_name) \
1595         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_WO(_name)
1596
1597 static ssize_t vmbus_chan_attr_show(struct kobject *kobj,
1598                                     struct attribute *attr, char *buf)
1599 {
1600         const struct vmbus_chan_attribute *attribute
1601                 = container_of(attr, struct vmbus_chan_attribute, attr);
1602         struct vmbus_channel *chan
1603                 = container_of(kobj, struct vmbus_channel, kobj);
1604
1605         if (!attribute->show)
1606                 return -EIO;
1607
1608         return attribute->show(chan, buf);
1609 }
1610
1611 static ssize_t vmbus_chan_attr_store(struct kobject *kobj,
1612                                      struct attribute *attr, const char *buf,
1613                                      size_t count)
1614 {
1615         const struct vmbus_chan_attribute *attribute
1616                 = container_of(attr, struct vmbus_chan_attribute, attr);
1617         struct vmbus_channel *chan
1618                 = container_of(kobj, struct vmbus_channel, kobj);
1619
1620         if (!attribute->store)
1621                 return -EIO;
1622
1623         return attribute->store(chan, buf, count);
1624 }
1625
1626 static const struct sysfs_ops vmbus_chan_sysfs_ops = {
1627         .show = vmbus_chan_attr_show,
1628         .store = vmbus_chan_attr_store,
1629 };
1630
1631 static ssize_t out_mask_show(struct vmbus_channel *channel, char *buf)
1632 {
1633         struct hv_ring_buffer_info *rbi = &channel->outbound;
1634         ssize_t ret;
1635
1636         mutex_lock(&rbi->ring_buffer_mutex);
1637         if (!rbi->ring_buffer) {
1638                 mutex_unlock(&rbi->ring_buffer_mutex);
1639                 return -EINVAL;
1640         }
1641
1642         ret = sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1643         mutex_unlock(&rbi->ring_buffer_mutex);
1644         return ret;
1645 }
1646 static VMBUS_CHAN_ATTR_RO(out_mask);
1647
1648 static ssize_t in_mask_show(struct vmbus_channel *channel, char *buf)
1649 {
1650         struct hv_ring_buffer_info *rbi = &channel->inbound;
1651         ssize_t ret;
1652
1653         mutex_lock(&rbi->ring_buffer_mutex);
1654         if (!rbi->ring_buffer) {
1655                 mutex_unlock(&rbi->ring_buffer_mutex);
1656                 return -EINVAL;
1657         }
1658
1659         ret = sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1660         mutex_unlock(&rbi->ring_buffer_mutex);
1661         return ret;
1662 }
1663 static VMBUS_CHAN_ATTR_RO(in_mask);
1664
1665 static ssize_t read_avail_show(struct vmbus_channel *channel, char *buf)
1666 {
1667         struct hv_ring_buffer_info *rbi = &channel->inbound;
1668         ssize_t ret;
1669
1670         mutex_lock(&rbi->ring_buffer_mutex);
1671         if (!rbi->ring_buffer) {
1672                 mutex_unlock(&rbi->ring_buffer_mutex);
1673                 return -EINVAL;
1674         }
1675
1676         ret = sprintf(buf, "%u\n", hv_get_bytes_to_read(rbi));
1677         mutex_unlock(&rbi->ring_buffer_mutex);
1678         return ret;
1679 }
1680 static VMBUS_CHAN_ATTR_RO(read_avail);
1681
1682 static ssize_t write_avail_show(struct vmbus_channel *channel, char *buf)
1683 {
1684         struct hv_ring_buffer_info *rbi = &channel->outbound;
1685         ssize_t ret;
1686
1687         mutex_lock(&rbi->ring_buffer_mutex);
1688         if (!rbi->ring_buffer) {
1689                 mutex_unlock(&rbi->ring_buffer_mutex);
1690                 return -EINVAL;
1691         }
1692
1693         ret = sprintf(buf, "%u\n", hv_get_bytes_to_write(rbi));
1694         mutex_unlock(&rbi->ring_buffer_mutex);
1695         return ret;
1696 }
1697 static VMBUS_CHAN_ATTR_RO(write_avail);
1698
1699 static ssize_t target_cpu_show(struct vmbus_channel *channel, char *buf)
1700 {
1701         return sprintf(buf, "%u\n", channel->target_cpu);
1702 }
1703 static ssize_t target_cpu_store(struct vmbus_channel *channel,
1704                                 const char *buf, size_t count)
1705 {
1706         u32 target_cpu, origin_cpu;
1707         ssize_t ret = count;
1708
1709         if (vmbus_proto_version < VERSION_WIN10_V4_1)
1710                 return -EIO;
1711
1712         if (sscanf(buf, "%uu", &target_cpu) != 1)
1713                 return -EIO;
1714
1715         /* Validate target_cpu for the cpumask_test_cpu() operation below. */
1716         if (target_cpu >= nr_cpumask_bits)
1717                 return -EINVAL;
1718
1719         /* No CPUs should come up or down during this. */
1720         cpus_read_lock();
1721
1722         if (!cpu_online(target_cpu)) {
1723                 cpus_read_unlock();
1724                 return -EINVAL;
1725         }
1726
1727         /*
1728          * Synchronizes target_cpu_store() and channel closure:
1729          *
1730          * { Initially: state = CHANNEL_OPENED }
1731          *
1732          * CPU1                         CPU2
1733          *
1734          * [target_cpu_store()]         [vmbus_disconnect_ring()]
1735          *
1736          * LOCK channel_mutex           LOCK channel_mutex
1737          * LOAD r1 = state              LOAD r2 = state
1738          * IF (r1 == CHANNEL_OPENED)    IF (r2 == CHANNEL_OPENED)
1739          *   SEND MODIFYCHANNEL           STORE state = CHANNEL_OPEN
1740          *   [...]                        SEND CLOSECHANNEL
1741          * UNLOCK channel_mutex         UNLOCK channel_mutex
1742          *
1743          * Forbids: r1 == r2 == CHANNEL_OPENED (i.e., CPU1's LOCK precedes
1744          *              CPU2's LOCK) && CPU2's SEND precedes CPU1's SEND
1745          *
1746          * Note.  The host processes the channel messages "sequentially", in
1747          * the order in which they are received on a per-partition basis.
1748          */
1749         mutex_lock(&vmbus_connection.channel_mutex);
1750
1751         /*
1752          * Hyper-V will ignore MODIFYCHANNEL messages for "non-open" channels;
1753          * avoid sending the message and fail here for such channels.
1754          */
1755         if (channel->state != CHANNEL_OPENED_STATE) {
1756                 ret = -EIO;
1757                 goto cpu_store_unlock;
1758         }
1759
1760         origin_cpu = channel->target_cpu;
1761         if (target_cpu == origin_cpu)
1762                 goto cpu_store_unlock;
1763
1764         if (vmbus_send_modifychannel(channel->offermsg.child_relid,
1765                                      hv_cpu_number_to_vp_number(target_cpu))) {
1766                 ret = -EIO;
1767                 goto cpu_store_unlock;
1768         }
1769
1770         /*
1771          * Warning.  At this point, there is *no* guarantee that the host will
1772          * have successfully processed the vmbus_send_modifychannel() request.
1773          * See the header comment of vmbus_send_modifychannel() for more info.
1774          *
1775          * Lags in the processing of the above vmbus_send_modifychannel() can
1776          * result in missed interrupts if the "old" target CPU is taken offline
1777          * before Hyper-V starts sending interrupts to the "new" target CPU.
1778          * But apart from this offlining scenario, the code tolerates such
1779          * lags.  It will function correctly even if a channel interrupt comes
1780          * in on a CPU that is different from the channel target_cpu value.
1781          */
1782
1783         channel->target_cpu = target_cpu;
1784
1785         /* See init_vp_index(). */
1786         if (hv_is_perf_channel(channel))
1787                 hv_update_alloced_cpus(origin_cpu, target_cpu);
1788
1789         /* Currently set only for storvsc channels. */
1790         if (channel->change_target_cpu_callback) {
1791                 (*channel->change_target_cpu_callback)(channel,
1792                                 origin_cpu, target_cpu);
1793         }
1794
1795 cpu_store_unlock:
1796         mutex_unlock(&vmbus_connection.channel_mutex);
1797         cpus_read_unlock();
1798         return ret;
1799 }
1800 static VMBUS_CHAN_ATTR(cpu, 0644, target_cpu_show, target_cpu_store);
1801
1802 static ssize_t channel_pending_show(struct vmbus_channel *channel,
1803                                     char *buf)
1804 {
1805         return sprintf(buf, "%d\n",
1806                        channel_pending(channel,
1807                                        vmbus_connection.monitor_pages[1]));
1808 }
1809 static VMBUS_CHAN_ATTR(pending, S_IRUGO, channel_pending_show, NULL);
1810
1811 static ssize_t channel_latency_show(struct vmbus_channel *channel,
1812                                     char *buf)
1813 {
1814         return sprintf(buf, "%d\n",
1815                        channel_latency(channel,
1816                                        vmbus_connection.monitor_pages[1]));
1817 }
1818 static VMBUS_CHAN_ATTR(latency, S_IRUGO, channel_latency_show, NULL);
1819
1820 static ssize_t channel_interrupts_show(struct vmbus_channel *channel, char *buf)
1821 {
1822         return sprintf(buf, "%llu\n", channel->interrupts);
1823 }
1824 static VMBUS_CHAN_ATTR(interrupts, S_IRUGO, channel_interrupts_show, NULL);
1825
1826 static ssize_t channel_events_show(struct vmbus_channel *channel, char *buf)
1827 {
1828         return sprintf(buf, "%llu\n", channel->sig_events);
1829 }
1830 static VMBUS_CHAN_ATTR(events, S_IRUGO, channel_events_show, NULL);
1831
1832 static ssize_t channel_intr_in_full_show(struct vmbus_channel *channel,
1833                                          char *buf)
1834 {
1835         return sprintf(buf, "%llu\n",
1836                        (unsigned long long)channel->intr_in_full);
1837 }
1838 static VMBUS_CHAN_ATTR(intr_in_full, 0444, channel_intr_in_full_show, NULL);
1839
1840 static ssize_t channel_intr_out_empty_show(struct vmbus_channel *channel,
1841                                            char *buf)
1842 {
1843         return sprintf(buf, "%llu\n",
1844                        (unsigned long long)channel->intr_out_empty);
1845 }
1846 static VMBUS_CHAN_ATTR(intr_out_empty, 0444, channel_intr_out_empty_show, NULL);
1847
1848 static ssize_t channel_out_full_first_show(struct vmbus_channel *channel,
1849                                            char *buf)
1850 {
1851         return sprintf(buf, "%llu\n",
1852                        (unsigned long long)channel->out_full_first);
1853 }
1854 static VMBUS_CHAN_ATTR(out_full_first, 0444, channel_out_full_first_show, NULL);
1855
1856 static ssize_t channel_out_full_total_show(struct vmbus_channel *channel,
1857                                            char *buf)
1858 {
1859         return sprintf(buf, "%llu\n",
1860                        (unsigned long long)channel->out_full_total);
1861 }
1862 static VMBUS_CHAN_ATTR(out_full_total, 0444, channel_out_full_total_show, NULL);
1863
1864 static ssize_t subchannel_monitor_id_show(struct vmbus_channel *channel,
1865                                           char *buf)
1866 {
1867         return sprintf(buf, "%u\n", channel->offermsg.monitorid);
1868 }
1869 static VMBUS_CHAN_ATTR(monitor_id, S_IRUGO, subchannel_monitor_id_show, NULL);
1870
1871 static ssize_t subchannel_id_show(struct vmbus_channel *channel,
1872                                   char *buf)
1873 {
1874         return sprintf(buf, "%u\n",
1875                        channel->offermsg.offer.sub_channel_index);
1876 }
1877 static VMBUS_CHAN_ATTR_RO(subchannel_id);
1878
1879 static struct attribute *vmbus_chan_attrs[] = {
1880         &chan_attr_out_mask.attr,
1881         &chan_attr_in_mask.attr,
1882         &chan_attr_read_avail.attr,
1883         &chan_attr_write_avail.attr,
1884         &chan_attr_cpu.attr,
1885         &chan_attr_pending.attr,
1886         &chan_attr_latency.attr,
1887         &chan_attr_interrupts.attr,
1888         &chan_attr_events.attr,
1889         &chan_attr_intr_in_full.attr,
1890         &chan_attr_intr_out_empty.attr,
1891         &chan_attr_out_full_first.attr,
1892         &chan_attr_out_full_total.attr,
1893         &chan_attr_monitor_id.attr,
1894         &chan_attr_subchannel_id.attr,
1895         NULL
1896 };
1897
1898 /*
1899  * Channel-level attribute_group callback function. Returns the permission for
1900  * each attribute, and returns 0 if an attribute is not visible.
1901  */
1902 static umode_t vmbus_chan_attr_is_visible(struct kobject *kobj,
1903                                           struct attribute *attr, int idx)
1904 {
1905         const struct vmbus_channel *channel =
1906                 container_of(kobj, struct vmbus_channel, kobj);
1907
1908         /* Hide the monitor attributes if the monitor mechanism is not used. */
1909         if (!channel->offermsg.monitor_allocated &&
1910             (attr == &chan_attr_pending.attr ||
1911              attr == &chan_attr_latency.attr ||
1912              attr == &chan_attr_monitor_id.attr))
1913                 return 0;
1914
1915         return attr->mode;
1916 }
1917
1918 static struct attribute_group vmbus_chan_group = {
1919         .attrs = vmbus_chan_attrs,
1920         .is_visible = vmbus_chan_attr_is_visible
1921 };
1922
1923 static struct kobj_type vmbus_chan_ktype = {
1924         .sysfs_ops = &vmbus_chan_sysfs_ops,
1925         .release = vmbus_chan_release,
1926 };
1927
1928 /*
1929  * vmbus_add_channel_kobj - setup a sub-directory under device/channels
1930  */
1931 int vmbus_add_channel_kobj(struct hv_device *dev, struct vmbus_channel *channel)
1932 {
1933         const struct device *device = &dev->device;
1934         struct kobject *kobj = &channel->kobj;
1935         u32 relid = channel->offermsg.child_relid;
1936         int ret;
1937
1938         kobj->kset = dev->channels_kset;
1939         ret = kobject_init_and_add(kobj, &vmbus_chan_ktype, NULL,
1940                                    "%u", relid);
1941         if (ret)
1942                 return ret;
1943
1944         ret = sysfs_create_group(kobj, &vmbus_chan_group);
1945
1946         if (ret) {
1947                 /*
1948                  * The calling functions' error handling paths will cleanup the
1949                  * empty channel directory.
1950                  */
1951                 dev_err(device, "Unable to set up channel sysfs files\n");
1952                 return ret;
1953         }
1954
1955         kobject_uevent(kobj, KOBJ_ADD);
1956
1957         return 0;
1958 }
1959
1960 /*
1961  * vmbus_remove_channel_attr_group - remove the channel's attribute group
1962  */
1963 void vmbus_remove_channel_attr_group(struct vmbus_channel *channel)
1964 {
1965         sysfs_remove_group(&channel->kobj, &vmbus_chan_group);
1966 }
1967
1968 /*
1969  * vmbus_device_create - Creates and registers a new child device
1970  * on the vmbus.
1971  */
1972 struct hv_device *vmbus_device_create(const guid_t *type,
1973                                       const guid_t *instance,
1974                                       struct vmbus_channel *channel)
1975 {
1976         struct hv_device *child_device_obj;
1977
1978         child_device_obj = kzalloc(sizeof(struct hv_device), GFP_KERNEL);
1979         if (!child_device_obj) {
1980                 pr_err("Unable to allocate device object for child device\n");
1981                 return NULL;
1982         }
1983
1984         child_device_obj->channel = channel;
1985         guid_copy(&child_device_obj->dev_type, type);
1986         guid_copy(&child_device_obj->dev_instance, instance);
1987         child_device_obj->vendor_id = 0x1414; /* MSFT vendor ID */
1988
1989         return child_device_obj;
1990 }
1991
1992 /*
1993  * vmbus_device_register - Register the child device
1994  */
1995 int vmbus_device_register(struct hv_device *child_device_obj)
1996 {
1997         struct kobject *kobj = &child_device_obj->device.kobj;
1998         int ret;
1999
2000         dev_set_name(&child_device_obj->device, "%pUl",
2001                      &child_device_obj->channel->offermsg.offer.if_instance);
2002
2003         child_device_obj->device.bus = &hv_bus;
2004         child_device_obj->device.parent = &hv_acpi_dev->dev;
2005         child_device_obj->device.release = vmbus_device_release;
2006
2007         /*
2008          * Register with the LDM. This will kick off the driver/device
2009          * binding...which will eventually call vmbus_match() and vmbus_probe()
2010          */
2011         ret = device_register(&child_device_obj->device);
2012         if (ret) {
2013                 pr_err("Unable to register child device\n");
2014                 return ret;
2015         }
2016
2017         child_device_obj->channels_kset = kset_create_and_add("channels",
2018                                                               NULL, kobj);
2019         if (!child_device_obj->channels_kset) {
2020                 ret = -ENOMEM;
2021                 goto err_dev_unregister;
2022         }
2023
2024         ret = vmbus_add_channel_kobj(child_device_obj,
2025                                      child_device_obj->channel);
2026         if (ret) {
2027                 pr_err("Unable to register primary channeln");
2028                 goto err_kset_unregister;
2029         }
2030         hv_debug_add_dev_dir(child_device_obj);
2031
2032         return 0;
2033
2034 err_kset_unregister:
2035         kset_unregister(child_device_obj->channels_kset);
2036
2037 err_dev_unregister:
2038         device_unregister(&child_device_obj->device);
2039         return ret;
2040 }
2041
2042 /*
2043  * vmbus_device_unregister - Remove the specified child device
2044  * from the vmbus.
2045  */
2046 void vmbus_device_unregister(struct hv_device *device_obj)
2047 {
2048         pr_debug("child device %s unregistered\n",
2049                 dev_name(&device_obj->device));
2050
2051         kset_unregister(device_obj->channels_kset);
2052
2053         /*
2054          * Kick off the process of unregistering the device.
2055          * This will call vmbus_remove() and eventually vmbus_device_release()
2056          */
2057         device_unregister(&device_obj->device);
2058 }
2059
2060
2061 /*
2062  * VMBUS is an acpi enumerated device. Get the information we
2063  * need from DSDT.
2064  */
2065 #define VTPM_BASE_ADDRESS 0xfed40000
2066 static acpi_status vmbus_walk_resources(struct acpi_resource *res, void *ctx)
2067 {
2068         resource_size_t start = 0;
2069         resource_size_t end = 0;
2070         struct resource *new_res;
2071         struct resource **old_res = &hyperv_mmio;
2072         struct resource **prev_res = NULL;
2073
2074         switch (res->type) {
2075
2076         /*
2077          * "Address" descriptors are for bus windows. Ignore
2078          * "memory" descriptors, which are for registers on
2079          * devices.
2080          */
2081         case ACPI_RESOURCE_TYPE_ADDRESS32:
2082                 start = res->data.address32.address.minimum;
2083                 end = res->data.address32.address.maximum;
2084                 break;
2085
2086         case ACPI_RESOURCE_TYPE_ADDRESS64:
2087                 start = res->data.address64.address.minimum;
2088                 end = res->data.address64.address.maximum;
2089                 break;
2090
2091         default:
2092                 /* Unused resource type */
2093                 return AE_OK;
2094
2095         }
2096         /*
2097          * Ignore ranges that are below 1MB, as they're not
2098          * necessary or useful here.
2099          */
2100         if (end < 0x100000)
2101                 return AE_OK;
2102
2103         new_res = kzalloc(sizeof(*new_res), GFP_ATOMIC);
2104         if (!new_res)
2105                 return AE_NO_MEMORY;
2106
2107         /* If this range overlaps the virtual TPM, truncate it. */
2108         if (end > VTPM_BASE_ADDRESS && start < VTPM_BASE_ADDRESS)
2109                 end = VTPM_BASE_ADDRESS;
2110
2111         new_res->name = "hyperv mmio";
2112         new_res->flags = IORESOURCE_MEM;
2113         new_res->start = start;
2114         new_res->end = end;
2115
2116         /*
2117          * If two ranges are adjacent, merge them.
2118          */
2119         do {
2120                 if (!*old_res) {
2121                         *old_res = new_res;
2122                         break;
2123                 }
2124
2125                 if (((*old_res)->end + 1) == new_res->start) {
2126                         (*old_res)->end = new_res->end;
2127                         kfree(new_res);
2128                         break;
2129                 }
2130
2131                 if ((*old_res)->start == new_res->end + 1) {
2132                         (*old_res)->start = new_res->start;
2133                         kfree(new_res);
2134                         break;
2135                 }
2136
2137                 if ((*old_res)->start > new_res->end) {
2138                         new_res->sibling = *old_res;
2139                         if (prev_res)
2140                                 (*prev_res)->sibling = new_res;
2141                         *old_res = new_res;
2142                         break;
2143                 }
2144
2145                 prev_res = old_res;
2146                 old_res = &(*old_res)->sibling;
2147
2148         } while (1);
2149
2150         return AE_OK;
2151 }
2152
2153 static int vmbus_acpi_remove(struct acpi_device *device)
2154 {
2155         struct resource *cur_res;
2156         struct resource *next_res;
2157
2158         if (hyperv_mmio) {
2159                 if (fb_mmio) {
2160                         __release_region(hyperv_mmio, fb_mmio->start,
2161                                          resource_size(fb_mmio));
2162                         fb_mmio = NULL;
2163                 }
2164
2165                 for (cur_res = hyperv_mmio; cur_res; cur_res = next_res) {
2166                         next_res = cur_res->sibling;
2167                         kfree(cur_res);
2168                 }
2169         }
2170
2171         return 0;
2172 }
2173
2174 static void vmbus_reserve_fb(void)
2175 {
2176         int size;
2177         /*
2178          * Make a claim for the frame buffer in the resource tree under the
2179          * first node, which will be the one below 4GB.  The length seems to
2180          * be underreported, particularly in a Generation 1 VM.  So start out
2181          * reserving a larger area and make it smaller until it succeeds.
2182          */
2183
2184         if (screen_info.lfb_base) {
2185                 if (efi_enabled(EFI_BOOT))
2186                         size = max_t(__u32, screen_info.lfb_size, 0x800000);
2187                 else
2188                         size = max_t(__u32, screen_info.lfb_size, 0x4000000);
2189
2190                 for (; !fb_mmio && (size >= 0x100000); size >>= 1) {
2191                         fb_mmio = __request_region(hyperv_mmio,
2192                                                    screen_info.lfb_base, size,
2193                                                    fb_mmio_name, 0);
2194                 }
2195         }
2196 }
2197
2198 /**
2199  * vmbus_allocate_mmio() - Pick a memory-mapped I/O range.
2200  * @new:                If successful, supplied a pointer to the
2201  *                      allocated MMIO space.
2202  * @device_obj:         Identifies the caller
2203  * @min:                Minimum guest physical address of the
2204  *                      allocation
2205  * @max:                Maximum guest physical address
2206  * @size:               Size of the range to be allocated
2207  * @align:              Alignment of the range to be allocated
2208  * @fb_overlap_ok:      Whether this allocation can be allowed
2209  *                      to overlap the video frame buffer.
2210  *
2211  * This function walks the resources granted to VMBus by the
2212  * _CRS object in the ACPI namespace underneath the parent
2213  * "bridge" whether that's a root PCI bus in the Generation 1
2214  * case or a Module Device in the Generation 2 case.  It then
2215  * attempts to allocate from the global MMIO pool in a way that
2216  * matches the constraints supplied in these parameters and by
2217  * that _CRS.
2218  *
2219  * Return: 0 on success, -errno on failure
2220  */
2221 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
2222                         resource_size_t min, resource_size_t max,
2223                         resource_size_t size, resource_size_t align,
2224                         bool fb_overlap_ok)
2225 {
2226         struct resource *iter, *shadow;
2227         resource_size_t range_min, range_max, start;
2228         const char *dev_n = dev_name(&device_obj->device);
2229         int retval;
2230
2231         retval = -ENXIO;
2232         mutex_lock(&hyperv_mmio_lock);
2233
2234         /*
2235          * If overlaps with frame buffers are allowed, then first attempt to
2236          * make the allocation from within the reserved region.  Because it
2237          * is already reserved, no shadow allocation is necessary.
2238          */
2239         if (fb_overlap_ok && fb_mmio && !(min > fb_mmio->end) &&
2240             !(max < fb_mmio->start)) {
2241
2242                 range_min = fb_mmio->start;
2243                 range_max = fb_mmio->end;
2244                 start = (range_min + align - 1) & ~(align - 1);
2245                 for (; start + size - 1 <= range_max; start += align) {
2246                         *new = request_mem_region_exclusive(start, size, dev_n);
2247                         if (*new) {
2248                                 retval = 0;
2249                                 goto exit;
2250                         }
2251                 }
2252         }
2253
2254         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
2255                 if ((iter->start >= max) || (iter->end <= min))
2256                         continue;
2257
2258                 range_min = iter->start;
2259                 range_max = iter->end;
2260                 start = (range_min + align - 1) & ~(align - 1);
2261                 for (; start + size - 1 <= range_max; start += align) {
2262                         shadow = __request_region(iter, start, size, NULL,
2263                                                   IORESOURCE_BUSY);
2264                         if (!shadow)
2265                                 continue;
2266
2267                         *new = request_mem_region_exclusive(start, size, dev_n);
2268                         if (*new) {
2269                                 shadow->name = (char *)*new;
2270                                 retval = 0;
2271                                 goto exit;
2272                         }
2273
2274                         __release_region(iter, start, size);
2275                 }
2276         }
2277
2278 exit:
2279         mutex_unlock(&hyperv_mmio_lock);
2280         return retval;
2281 }
2282 EXPORT_SYMBOL_GPL(vmbus_allocate_mmio);
2283
2284 /**
2285  * vmbus_free_mmio() - Free a memory-mapped I/O range.
2286  * @start:              Base address of region to release.
2287  * @size:               Size of the range to be allocated
2288  *
2289  * This function releases anything requested by
2290  * vmbus_mmio_allocate().
2291  */
2292 void vmbus_free_mmio(resource_size_t start, resource_size_t size)
2293 {
2294         struct resource *iter;
2295
2296         mutex_lock(&hyperv_mmio_lock);
2297         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
2298                 if ((iter->start >= start + size) || (iter->end <= start))
2299                         continue;
2300
2301                 __release_region(iter, start, size);
2302         }
2303         release_mem_region(start, size);
2304         mutex_unlock(&hyperv_mmio_lock);
2305
2306 }
2307 EXPORT_SYMBOL_GPL(vmbus_free_mmio);
2308
2309 static int vmbus_acpi_add(struct acpi_device *device)
2310 {
2311         acpi_status result;
2312         int ret_val = -ENODEV;
2313         struct acpi_device *ancestor;
2314
2315         hv_acpi_dev = device;
2316
2317         result = acpi_walk_resources(device->handle, METHOD_NAME__CRS,
2318                                         vmbus_walk_resources, NULL);
2319
2320         if (ACPI_FAILURE(result))
2321                 goto acpi_walk_err;
2322         /*
2323          * Some ancestor of the vmbus acpi device (Gen1 or Gen2
2324          * firmware) is the VMOD that has the mmio ranges. Get that.
2325          */
2326         for (ancestor = device->parent; ancestor; ancestor = ancestor->parent) {
2327                 result = acpi_walk_resources(ancestor->handle, METHOD_NAME__CRS,
2328                                              vmbus_walk_resources, NULL);
2329
2330                 if (ACPI_FAILURE(result))
2331                         continue;
2332                 if (hyperv_mmio) {
2333                         vmbus_reserve_fb();
2334                         break;
2335                 }
2336         }
2337         ret_val = 0;
2338
2339 acpi_walk_err:
2340         complete(&probe_event);
2341         if (ret_val)
2342                 vmbus_acpi_remove(device);
2343         return ret_val;
2344 }
2345
2346 #ifdef CONFIG_PM_SLEEP
2347 static int vmbus_bus_suspend(struct device *dev)
2348 {
2349         struct vmbus_channel *channel, *sc;
2350
2351         while (atomic_read(&vmbus_connection.offer_in_progress) != 0) {
2352                 /*
2353                  * We wait here until the completion of any channel
2354                  * offers that are currently in progress.
2355                  */
2356                 msleep(1);
2357         }
2358
2359         mutex_lock(&vmbus_connection.channel_mutex);
2360         list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
2361                 if (!is_hvsock_channel(channel))
2362                         continue;
2363
2364                 vmbus_force_channel_rescinded(channel);
2365         }
2366         mutex_unlock(&vmbus_connection.channel_mutex);
2367
2368         /*
2369          * Wait until all the sub-channels and hv_sock channels have been
2370          * cleaned up. Sub-channels should be destroyed upon suspend, otherwise
2371          * they would conflict with the new sub-channels that will be created
2372          * in the resume path. hv_sock channels should also be destroyed, but
2373          * a hv_sock channel of an established hv_sock connection can not be
2374          * really destroyed since it may still be referenced by the userspace
2375          * application, so we just force the hv_sock channel to be rescinded
2376          * by vmbus_force_channel_rescinded(), and the userspace application
2377          * will thoroughly destroy the channel after hibernation.
2378          *
2379          * Note: the counter nr_chan_close_on_suspend may never go above 0 if
2380          * the VM has no sub-channel and hv_sock channel, e.g. a 1-vCPU VM.
2381          */
2382         if (atomic_read(&vmbus_connection.nr_chan_close_on_suspend) > 0)
2383                 wait_for_completion(&vmbus_connection.ready_for_suspend_event);
2384
2385         if (atomic_read(&vmbus_connection.nr_chan_fixup_on_resume) != 0) {
2386                 pr_err("Can not suspend due to a previous failed resuming\n");
2387                 return -EBUSY;
2388         }
2389
2390         mutex_lock(&vmbus_connection.channel_mutex);
2391
2392         list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
2393                 /*
2394                  * Remove the channel from the array of channels and invalidate
2395                  * the channel's relid.  Upon resume, vmbus_onoffer() will fix
2396                  * up the relid (and other fields, if necessary) and add the
2397                  * channel back to the array.
2398                  */
2399                 vmbus_channel_unmap_relid(channel);
2400                 channel->offermsg.child_relid = INVALID_RELID;
2401
2402                 if (is_hvsock_channel(channel)) {
2403                         if (!channel->rescind) {
2404                                 pr_err("hv_sock channel not rescinded!\n");
2405                                 WARN_ON_ONCE(1);
2406                         }
2407                         continue;
2408                 }
2409
2410                 list_for_each_entry(sc, &channel->sc_list, sc_list) {
2411                         pr_err("Sub-channel not deleted!\n");
2412                         WARN_ON_ONCE(1);
2413                 }
2414
2415                 atomic_inc(&vmbus_connection.nr_chan_fixup_on_resume);
2416         }
2417
2418         mutex_unlock(&vmbus_connection.channel_mutex);
2419
2420         vmbus_initiate_unload(false);
2421
2422         /* Reset the event for the next resume. */
2423         reinit_completion(&vmbus_connection.ready_for_resume_event);
2424
2425         return 0;
2426 }
2427
2428 static int vmbus_bus_resume(struct device *dev)
2429 {
2430         struct vmbus_channel_msginfo *msginfo;
2431         size_t msgsize;
2432         int ret;
2433
2434         /*
2435          * We only use the 'vmbus_proto_version', which was in use before
2436          * hibernation, to re-negotiate with the host.
2437          */
2438         if (!vmbus_proto_version) {
2439                 pr_err("Invalid proto version = 0x%x\n", vmbus_proto_version);
2440                 return -EINVAL;
2441         }
2442
2443         msgsize = sizeof(*msginfo) +
2444                   sizeof(struct vmbus_channel_initiate_contact);
2445
2446         msginfo = kzalloc(msgsize, GFP_KERNEL);
2447
2448         if (msginfo == NULL)
2449                 return -ENOMEM;
2450
2451         ret = vmbus_negotiate_version(msginfo, vmbus_proto_version);
2452
2453         kfree(msginfo);
2454
2455         if (ret != 0)
2456                 return ret;
2457
2458         WARN_ON(atomic_read(&vmbus_connection.nr_chan_fixup_on_resume) == 0);
2459
2460         vmbus_request_offers();
2461
2462         if (wait_for_completion_timeout(
2463                 &vmbus_connection.ready_for_resume_event, 10 * HZ) == 0)
2464                 pr_err("Some vmbus device is missing after suspending?\n");
2465
2466         /* Reset the event for the next suspend. */
2467         reinit_completion(&vmbus_connection.ready_for_suspend_event);
2468
2469         return 0;
2470 }
2471 #else
2472 #define vmbus_bus_suspend NULL
2473 #define vmbus_bus_resume NULL
2474 #endif /* CONFIG_PM_SLEEP */
2475
2476 static const struct acpi_device_id vmbus_acpi_device_ids[] = {
2477         {"VMBUS", 0},
2478         {"VMBus", 0},
2479         {"", 0},
2480 };
2481 MODULE_DEVICE_TABLE(acpi, vmbus_acpi_device_ids);
2482
2483 /*
2484  * Note: we must use the "no_irq" ops, otherwise hibernation can not work with
2485  * PCI device assignment, because "pci_dev_pm_ops" uses the "noirq" ops: in
2486  * the resume path, the pci "noirq" restore op runs before "non-noirq" op (see
2487  * resume_target_kernel() -> dpm_resume_start(), and hibernation_restore() ->
2488  * dpm_resume_end()). This means vmbus_bus_resume() and the pci-hyperv's
2489  * resume callback must also run via the "noirq" ops.
2490  *
2491  * Set suspend_noirq/resume_noirq to NULL for Suspend-to-Idle: see the comment
2492  * earlier in this file before vmbus_pm.
2493  */
2494
2495 static const struct dev_pm_ops vmbus_bus_pm = {
2496         .suspend_noirq  = NULL,
2497         .resume_noirq   = NULL,
2498         .freeze_noirq   = vmbus_bus_suspend,
2499         .thaw_noirq     = vmbus_bus_resume,
2500         .poweroff_noirq = vmbus_bus_suspend,
2501         .restore_noirq  = vmbus_bus_resume
2502 };
2503
2504 static struct acpi_driver vmbus_acpi_driver = {
2505         .name = "vmbus",
2506         .ids = vmbus_acpi_device_ids,
2507         .ops = {
2508                 .add = vmbus_acpi_add,
2509                 .remove = vmbus_acpi_remove,
2510         },
2511         .drv.pm = &vmbus_bus_pm,
2512 };
2513
2514 static void hv_kexec_handler(void)
2515 {
2516         hv_stimer_global_cleanup();
2517         vmbus_initiate_unload(false);
2518         /* Make sure conn_state is set as hv_synic_cleanup checks for it */
2519         mb();
2520         cpuhp_remove_state(hyperv_cpuhp_online);
2521         hyperv_cleanup();
2522 };
2523
2524 static void hv_crash_handler(struct pt_regs *regs)
2525 {
2526         int cpu;
2527
2528         vmbus_initiate_unload(true);
2529         /*
2530          * In crash handler we can't schedule synic cleanup for all CPUs,
2531          * doing the cleanup for current CPU only. This should be sufficient
2532          * for kdump.
2533          */
2534         cpu = smp_processor_id();
2535         hv_stimer_cleanup(cpu);
2536         hv_synic_disable_regs(cpu);
2537         hyperv_cleanup();
2538 };
2539
2540 static int hv_synic_suspend(void)
2541 {
2542         /*
2543          * When we reach here, all the non-boot CPUs have been offlined.
2544          * If we're in a legacy configuration where stimer Direct Mode is
2545          * not enabled, the stimers on the non-boot CPUs have been unbound
2546          * in hv_synic_cleanup() -> hv_stimer_legacy_cleanup() ->
2547          * hv_stimer_cleanup() -> clockevents_unbind_device().
2548          *
2549          * hv_synic_suspend() only runs on CPU0 with interrupts disabled.
2550          * Here we do not call hv_stimer_legacy_cleanup() on CPU0 because:
2551          * 1) it's unnecessary as interrupts remain disabled between
2552          * syscore_suspend() and syscore_resume(): see create_image() and
2553          * resume_target_kernel()
2554          * 2) the stimer on CPU0 is automatically disabled later by
2555          * syscore_suspend() -> timekeeping_suspend() -> tick_suspend() -> ...
2556          * -> clockevents_shutdown() -> ... -> hv_ce_shutdown()
2557          * 3) a warning would be triggered if we call
2558          * clockevents_unbind_device(), which may sleep, in an
2559          * interrupts-disabled context.
2560          */
2561
2562         hv_synic_disable_regs(0);
2563
2564         return 0;
2565 }
2566
2567 static void hv_synic_resume(void)
2568 {
2569         hv_synic_enable_regs(0);
2570
2571         /*
2572          * Note: we don't need to call hv_stimer_init(0), because the timer
2573          * on CPU0 is not unbound in hv_synic_suspend(), and the timer is
2574          * automatically re-enabled in timekeeping_resume().
2575          */
2576 }
2577
2578 /* The callbacks run only on CPU0, with irqs_disabled. */
2579 static struct syscore_ops hv_synic_syscore_ops = {
2580         .suspend = hv_synic_suspend,
2581         .resume = hv_synic_resume,
2582 };
2583
2584 static int __init hv_acpi_init(void)
2585 {
2586         int ret, t;
2587
2588         if (!hv_is_hyperv_initialized())
2589                 return -ENODEV;
2590
2591         init_completion(&probe_event);
2592
2593         /*
2594          * Get ACPI resources first.
2595          */
2596         ret = acpi_bus_register_driver(&vmbus_acpi_driver);
2597
2598         if (ret)
2599                 return ret;
2600
2601         t = wait_for_completion_timeout(&probe_event, 5*HZ);
2602         if (t == 0) {
2603                 ret = -ETIMEDOUT;
2604                 goto cleanup;
2605         }
2606         hv_debug_init();
2607
2608         ret = vmbus_bus_init();
2609         if (ret)
2610                 goto cleanup;
2611
2612         hv_setup_kexec_handler(hv_kexec_handler);
2613         hv_setup_crash_handler(hv_crash_handler);
2614
2615         register_syscore_ops(&hv_synic_syscore_ops);
2616
2617         return 0;
2618
2619 cleanup:
2620         acpi_bus_unregister_driver(&vmbus_acpi_driver);
2621         hv_acpi_dev = NULL;
2622         return ret;
2623 }
2624
2625 static void __exit vmbus_exit(void)
2626 {
2627         int cpu;
2628
2629         unregister_syscore_ops(&hv_synic_syscore_ops);
2630
2631         hv_remove_kexec_handler();
2632         hv_remove_crash_handler();
2633         vmbus_connection.conn_state = DISCONNECTED;
2634         hv_stimer_global_cleanup();
2635         vmbus_disconnect();
2636         hv_remove_vmbus_irq();
2637         for_each_online_cpu(cpu) {
2638                 struct hv_per_cpu_context *hv_cpu
2639                         = per_cpu_ptr(hv_context.cpu_context, cpu);
2640
2641                 tasklet_kill(&hv_cpu->msg_dpc);
2642         }
2643         hv_debug_rm_all_dir();
2644
2645         vmbus_free_channels();
2646         kfree(vmbus_connection.channels);
2647
2648         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
2649                 kmsg_dump_unregister(&hv_kmsg_dumper);
2650                 unregister_die_notifier(&hyperv_die_block);
2651                 atomic_notifier_chain_unregister(&panic_notifier_list,
2652                                                  &hyperv_panic_block);
2653         }
2654
2655         free_page((unsigned long)hv_panic_page);
2656         unregister_sysctl_table(hv_ctl_table_hdr);
2657         hv_ctl_table_hdr = NULL;
2658         bus_unregister(&hv_bus);
2659
2660         cpuhp_remove_state(hyperv_cpuhp_online);
2661         hv_synic_free();
2662         acpi_bus_unregister_driver(&vmbus_acpi_driver);
2663 }
2664
2665
2666 MODULE_LICENSE("GPL");
2667 MODULE_DESCRIPTION("Microsoft Hyper-V VMBus Driver");
2668
2669 subsys_initcall(hv_acpi_init);
2670 module_exit(vmbus_exit);