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/panic_notifier.h>
  29 #include <linux/ptrace.h>
  30 #include <linux/screen_info.h>
  31 #include <linux/kdebug.h>
  32 #include <linux/efi.h>
  33 #include <linux/random.h>
  34 #include <linux/kernel.h>
  35 #include <linux/syscore_ops.h>
  36 #include <linux/dma-map-ops.h>
  37 #include <clocksource/hyperv_timer.h>
  38 #include "hyperv_vmbus.h"
  39
  40 struct vmbus_dynid {
  41         struct list_head node;
  42         struct hv_vmbus_device_id id;
  43 };
  44
  45 static struct acpi_device  *hv_acpi_dev;
  46
  47 static struct completion probe_event;
  48
  49 static int hyperv_cpuhp_online;
  50
  51 static void *hv_panic_page;
  52
  53 static long __percpu *vmbus_evt;
  54
  55 /* Values parsed from ACPI DSDT */
  56 int vmbus_irq;
  57 int vmbus_interrupt;
  58
  59 /*
  60  * Boolean to control whether to report panic messages over Hyper-V.
  61  *
  62  * It can be set via /proc/sys/kernel/hyperv_record_panic_msg
  63  */
  64 static int sysctl_record_panic_msg = 1;
  65
  66 static int hyperv_report_reg(void)
  67 {
  68         return !sysctl_record_panic_msg || !hv_panic_page;
  69 }
  70
  71 static int hyperv_panic_event(struct notifier_block *nb, unsigned long val,
  72                               void *args)
  73 {
  74         struct pt_regs *regs;
  75
  76         vmbus_initiate_unload(true);
  77
  78         /*
  79          * Hyper-V should be notified only once about a panic.  If we will be
  80          * doing hyperv_report_panic_msg() later with kmsg data, don't do
  81          * the notification here.
  82          */
  83         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE
  84             && hyperv_report_reg()) {
  85                 regs = current_pt_regs();
  86                 hyperv_report_panic(regs, val, false);
  87         }
  88         return NOTIFY_DONE;
  89 }
  90
  91 static int hyperv_die_event(struct notifier_block *nb, unsigned long val,
  92                             void *args)
  93 {
  94         struct die_args *die = args;
  95         struct pt_regs *regs = die->regs;
  96
  97         /* Don't notify Hyper-V if the die event is other than oops */
  98         if (val != DIE_OOPS)
  99                 return NOTIFY_DONE;
 100
 101         /*
 102          * Hyper-V should be notified only once about a panic.  If we will be
 103          * doing hyperv_report_panic_msg() later with kmsg data, don't do
 104          * the notification here.
 105          */
 106         if (hyperv_report_reg())
 107                 hyperv_report_panic(regs, val, true);
 108         return NOTIFY_DONE;
 109 }
 110
 111 static struct notifier_block hyperv_die_block = {
 112         .notifier_call = hyperv_die_event,
 113 };
 114 static struct notifier_block hyperv_panic_block = {
 115         .notifier_call = hyperv_panic_event,
 116 };
 117
 118 static const char *fb_mmio_name = "fb_range";
 119 static struct resource *fb_mmio;
 120 static struct resource *hyperv_mmio;
 121 static DEFINE_MUTEX(hyperv_mmio_lock);
 122
 123 static int vmbus_exists(void)
 124 {
 125         if (hv_acpi_dev == NULL)
 126                 return -ENODEV;
 127
 128         return 0;
 129 }
 130
 131 static u8 channel_monitor_group(const struct vmbus_channel *channel)
 132 {
 133         return (u8)channel->offermsg.monitorid / 32;
 134 }
 135
 136 static u8 channel_monitor_offset(const struct vmbus_channel *channel)
 137 {
 138         return (u8)channel->offermsg.monitorid % 32;
 139 }
 140
 141 static u32 channel_pending(const struct vmbus_channel *channel,
 142                            const struct hv_monitor_page *monitor_page)
 143 {
 144         u8 monitor_group = channel_monitor_group(channel);
 145
 146         return monitor_page->trigger_group[monitor_group].pending;
 147 }
 148
 149 static u32 channel_latency(const struct vmbus_channel *channel,
 150                            const struct hv_monitor_page *monitor_page)
 151 {
 152         u8 monitor_group = channel_monitor_group(channel);
 153         u8 monitor_offset = channel_monitor_offset(channel);
 154
 155         return monitor_page->latency[monitor_group][monitor_offset];
 156 }
 157
 158 static u32 channel_conn_id(struct vmbus_channel *channel,
 159                            struct hv_monitor_page *monitor_page)
 160 {
 161         u8 monitor_group = channel_monitor_group(channel);
 162         u8 monitor_offset = channel_monitor_offset(channel);
 163
 164         return monitor_page->parameter[monitor_group][monitor_offset].connectionid.u.id;
 165 }
 166
 167 static ssize_t id_show(struct device *dev, struct device_attribute *dev_attr,
 168                        char *buf)
 169 {
 170         struct hv_device *hv_dev = device_to_hv_device(dev);
 171
 172         if (!hv_dev->channel)
 173                 return -ENODEV;
 174         return sprintf(buf, "%d\n", hv_dev->channel->offermsg.child_relid);
 175 }
 176 static DEVICE_ATTR_RO(id);
 177
 178 static ssize_t state_show(struct device *dev, struct device_attribute *dev_attr,
 179                           char *buf)
 180 {
 181         struct hv_device *hv_dev = device_to_hv_device(dev);
 182
 183         if (!hv_dev->channel)
 184                 return -ENODEV;
 185         return sprintf(buf, "%d\n", hv_dev->channel->state);
 186 }
 187 static DEVICE_ATTR_RO(state);
 188
 189 static ssize_t monitor_id_show(struct device *dev,
 190                                struct device_attribute *dev_attr, char *buf)
 191 {
 192         struct hv_device *hv_dev = device_to_hv_device(dev);
 193
 194         if (!hv_dev->channel)
 195                 return -ENODEV;
 196         return sprintf(buf, "%d\n", hv_dev->channel->offermsg.monitorid);
 197 }
 198 static DEVICE_ATTR_RO(monitor_id);
 199
 200 static ssize_t class_id_show(struct device *dev,
 201                                struct device_attribute *dev_attr, char *buf)
 202 {
 203         struct hv_device *hv_dev = device_to_hv_device(dev);
 204
 205         if (!hv_dev->channel)
 206                 return -ENODEV;
 207         return sprintf(buf, "{%pUl}\n",
 208                        &hv_dev->channel->offermsg.offer.if_type);
 209 }
 210 static DEVICE_ATTR_RO(class_id);
 211
 212 static ssize_t device_id_show(struct device *dev,
 213                               struct device_attribute *dev_attr, char *buf)
 214 {
 215         struct hv_device *hv_dev = device_to_hv_device(dev);
 216
 217         if (!hv_dev->channel)
 218                 return -ENODEV;
 219         return sprintf(buf, "{%pUl}\n",
 220                        &hv_dev->channel->offermsg.offer.if_instance);
 221 }
 222 static DEVICE_ATTR_RO(device_id);
 223
 224 static ssize_t modalias_show(struct device *dev,
 225                              struct device_attribute *dev_attr, char *buf)
 226 {
 227         struct hv_device *hv_dev = device_to_hv_device(dev);
 228
 229         return sprintf(buf, "vmbus:%*phN\n", UUID_SIZE, &hv_dev->dev_type);
 230 }
 231 static DEVICE_ATTR_RO(modalias);
 232
 233 #ifdef CONFIG_NUMA
 234 static ssize_t numa_node_show(struct device *dev,
 235                               struct device_attribute *attr, char *buf)
 236 {
 237         struct hv_device *hv_dev = device_to_hv_device(dev);
 238
 239         if (!hv_dev->channel)
 240                 return -ENODEV;
 241
 242         return sprintf(buf, "%d\n", cpu_to_node(hv_dev->channel->target_cpu));
 243 }
 244 static DEVICE_ATTR_RO(numa_node);
 245 #endif
 246
 247 static ssize_t server_monitor_pending_show(struct device *dev,
 248                                            struct device_attribute *dev_attr,
 249                                            char *buf)
 250 {
 251         struct hv_device *hv_dev = device_to_hv_device(dev);
 252
 253         if (!hv_dev->channel)
 254                 return -ENODEV;
 255         return sprintf(buf, "%d\n",
 256                        channel_pending(hv_dev->channel,
 257                                        vmbus_connection.monitor_pages[0]));
 258 }
 259 static DEVICE_ATTR_RO(server_monitor_pending);
 260
 261 static ssize_t client_monitor_pending_show(struct device *dev,
 262                                            struct device_attribute *dev_attr,
 263                                            char *buf)
 264 {
 265         struct hv_device *hv_dev = device_to_hv_device(dev);
 266
 267         if (!hv_dev->channel)
 268                 return -ENODEV;
 269         return sprintf(buf, "%d\n",
 270                        channel_pending(hv_dev->channel,
 271                                        vmbus_connection.monitor_pages[1]));
 272 }
 273 static DEVICE_ATTR_RO(client_monitor_pending);
 274
 275 static ssize_t server_monitor_latency_show(struct device *dev,
 276                                            struct device_attribute *dev_attr,
 277                                            char *buf)
 278 {
 279         struct hv_device *hv_dev = device_to_hv_device(dev);
 280
 281         if (!hv_dev->channel)
 282                 return -ENODEV;
 283         return sprintf(buf, "%d\n",
 284                        channel_latency(hv_dev->channel,
 285                                        vmbus_connection.monitor_pages[0]));
 286 }
 287 static DEVICE_ATTR_RO(server_monitor_latency);
 288
 289 static ssize_t client_monitor_latency_show(struct device *dev,
 290                                            struct device_attribute *dev_attr,
 291                                            char *buf)
 292 {
 293         struct hv_device *hv_dev = device_to_hv_device(dev);
 294
 295         if (!hv_dev->channel)
 296                 return -ENODEV;
 297         return sprintf(buf, "%d\n",
 298                        channel_latency(hv_dev->channel,
 299                                        vmbus_connection.monitor_pages[1]));
 300 }
 301 static DEVICE_ATTR_RO(client_monitor_latency);
 302
 303 static ssize_t server_monitor_conn_id_show(struct device *dev,
 304                                            struct device_attribute *dev_attr,
 305                                            char *buf)
 306 {
 307         struct hv_device *hv_dev = device_to_hv_device(dev);
 308
 309         if (!hv_dev->channel)
 310                 return -ENODEV;
 311         return sprintf(buf, "%d\n",
 312                        channel_conn_id(hv_dev->channel,
 313                                        vmbus_connection.monitor_pages[0]));
 314 }
 315 static DEVICE_ATTR_RO(server_monitor_conn_id);
 316
 317 static ssize_t client_monitor_conn_id_show(struct device *dev,
 318                                            struct device_attribute *dev_attr,
 319                                            char *buf)
 320 {
 321         struct hv_device *hv_dev = device_to_hv_device(dev);
 322
 323         if (!hv_dev->channel)
 324                 return -ENODEV;
 325         return sprintf(buf, "%d\n",
 326                        channel_conn_id(hv_dev->channel,
 327                                        vmbus_connection.monitor_pages[1]));
 328 }
 329 static DEVICE_ATTR_RO(client_monitor_conn_id);
 330
 331 static ssize_t out_intr_mask_show(struct device *dev,
 332                                   struct device_attribute *dev_attr, char *buf)
 333 {
 334         struct hv_device *hv_dev = device_to_hv_device(dev);
 335         struct hv_ring_buffer_debug_info outbound;
 336         int ret;
 337
 338         if (!hv_dev->channel)
 339                 return -ENODEV;
 340
 341         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 342                                           &outbound);
 343         if (ret < 0)
 344                 return ret;
 345
 346         return sprintf(buf, "%d\n", outbound.current_interrupt_mask);
 347 }
 348 static DEVICE_ATTR_RO(out_intr_mask);
 349
 350 static ssize_t out_read_index_show(struct device *dev,
 351                                    struct device_attribute *dev_attr, char *buf)
 352 {
 353         struct hv_device *hv_dev = device_to_hv_device(dev);
 354         struct hv_ring_buffer_debug_info outbound;
 355         int ret;
 356
 357         if (!hv_dev->channel)
 358                 return -ENODEV;
 359
 360         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 361                                           &outbound);
 362         if (ret < 0)
 363                 return ret;
 364         return sprintf(buf, "%d\n", outbound.current_read_index);
 365 }
 366 static DEVICE_ATTR_RO(out_read_index);
 367
 368 static ssize_t out_write_index_show(struct device *dev,
 369                                     struct device_attribute *dev_attr,
 370                                     char *buf)
 371 {
 372         struct hv_device *hv_dev = device_to_hv_device(dev);
 373         struct hv_ring_buffer_debug_info outbound;
 374         int ret;
 375
 376         if (!hv_dev->channel)
 377                 return -ENODEV;
 378
 379         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 380                                           &outbound);
 381         if (ret < 0)
 382                 return ret;
 383         return sprintf(buf, "%d\n", outbound.current_write_index);
 384 }
 385 static DEVICE_ATTR_RO(out_write_index);
 386
 387 static ssize_t out_read_bytes_avail_show(struct device *dev,
 388                                          struct device_attribute *dev_attr,
 389                                          char *buf)
 390 {
 391         struct hv_device *hv_dev = device_to_hv_device(dev);
 392         struct hv_ring_buffer_debug_info outbound;
 393         int ret;
 394
 395         if (!hv_dev->channel)
 396                 return -ENODEV;
 397
 398         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 399                                           &outbound);
 400         if (ret < 0)
 401                 return ret;
 402         return sprintf(buf, "%d\n", outbound.bytes_avail_toread);
 403 }
 404 static DEVICE_ATTR_RO(out_read_bytes_avail);
 405
 406 static ssize_t out_write_bytes_avail_show(struct device *dev,
 407                                           struct device_attribute *dev_attr,
 408                                           char *buf)
 409 {
 410         struct hv_device *hv_dev = device_to_hv_device(dev);
 411         struct hv_ring_buffer_debug_info outbound;
 412         int ret;
 413
 414         if (!hv_dev->channel)
 415                 return -ENODEV;
 416
 417         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 418                                           &outbound);
 419         if (ret < 0)
 420                 return ret;
 421         return sprintf(buf, "%d\n", outbound.bytes_avail_towrite);
 422 }
 423 static DEVICE_ATTR_RO(out_write_bytes_avail);
 424
 425 static ssize_t in_intr_mask_show(struct device *dev,
 426                                  struct device_attribute *dev_attr, char *buf)
 427 {
 428         struct hv_device *hv_dev = device_to_hv_device(dev);
 429         struct hv_ring_buffer_debug_info inbound;
 430         int ret;
 431
 432         if (!hv_dev->channel)
 433                 return -ENODEV;
 434
 435         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 436         if (ret < 0)
 437                 return ret;
 438
 439         return sprintf(buf, "%d\n", inbound.current_interrupt_mask);
 440 }
 441 static DEVICE_ATTR_RO(in_intr_mask);
 442
 443 static ssize_t in_read_index_show(struct device *dev,
 444                                   struct device_attribute *dev_attr, char *buf)
 445 {
 446         struct hv_device *hv_dev = device_to_hv_device(dev);
 447         struct hv_ring_buffer_debug_info inbound;
 448         int ret;
 449
 450         if (!hv_dev->channel)
 451                 return -ENODEV;
 452
 453         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 454         if (ret < 0)
 455                 return ret;
 456
 457         return sprintf(buf, "%d\n", inbound.current_read_index);
 458 }
 459 static DEVICE_ATTR_RO(in_read_index);
 460
 461 static ssize_t in_write_index_show(struct device *dev,
 462                                    struct device_attribute *dev_attr, char *buf)
 463 {
 464         struct hv_device *hv_dev = device_to_hv_device(dev);
 465         struct hv_ring_buffer_debug_info inbound;
 466         int ret;
 467
 468         if (!hv_dev->channel)
 469                 return -ENODEV;
 470
 471         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 472         if (ret < 0)
 473                 return ret;
 474
 475         return sprintf(buf, "%d\n", inbound.current_write_index);
 476 }
 477 static DEVICE_ATTR_RO(in_write_index);
 478
 479 static ssize_t in_read_bytes_avail_show(struct device *dev,
 480                                         struct device_attribute *dev_attr,
 481                                         char *buf)
 482 {
 483         struct hv_device *hv_dev = device_to_hv_device(dev);
 484         struct hv_ring_buffer_debug_info inbound;
 485         int ret;
 486
 487         if (!hv_dev->channel)
 488                 return -ENODEV;
 489
 490         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 491         if (ret < 0)
 492                 return ret;
 493
 494         return sprintf(buf, "%d\n", inbound.bytes_avail_toread);
 495 }
 496 static DEVICE_ATTR_RO(in_read_bytes_avail);
 497
 498 static ssize_t in_write_bytes_avail_show(struct device *dev,
 499                                          struct device_attribute *dev_attr,
 500                                          char *buf)
 501 {
 502         struct hv_device *hv_dev = device_to_hv_device(dev);
 503         struct hv_ring_buffer_debug_info inbound;
 504         int ret;
 505
 506         if (!hv_dev->channel)
 507                 return -ENODEV;
 508
 509         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 510         if (ret < 0)
 511                 return ret;
 512
 513         return sprintf(buf, "%d\n", inbound.bytes_avail_towrite);
 514 }
 515 static DEVICE_ATTR_RO(in_write_bytes_avail);
 516
 517 static ssize_t channel_vp_mapping_show(struct device *dev,
 518                                        struct device_attribute *dev_attr,
 519                                        char *buf)
 520 {
 521         struct hv_device *hv_dev = device_to_hv_device(dev);
 522         struct vmbus_channel *channel = hv_dev->channel, *cur_sc;
 523         int buf_size = PAGE_SIZE, n_written, tot_written;
 524         struct list_head *cur;
 525
 526         if (!channel)
 527                 return -ENODEV;
 528
 529         mutex_lock(&vmbus_connection.channel_mutex);
 530
 531         tot_written = snprintf(buf, buf_size, "%u:%u\n",
 532                 channel->offermsg.child_relid, channel->target_cpu);
 533
 534         list_for_each(cur, &channel->sc_list) {
 535                 if (tot_written >= buf_size - 1)
 536                         break;
 537
 538                 cur_sc = list_entry(cur, struct vmbus_channel, sc_list);
 539                 n_written = scnprintf(buf + tot_written,
 540                                      buf_size - tot_written,
 541                                      "%u:%u\n",
 542                                      cur_sc->offermsg.child_relid,
 543                                      cur_sc->target_cpu);
 544                 tot_written += n_written;
 545         }
 546
 547         mutex_unlock(&vmbus_connection.channel_mutex);
 548
 549         return tot_written;
 550 }
 551 static DEVICE_ATTR_RO(channel_vp_mapping);
 552
 553 static ssize_t vendor_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
 559         return sprintf(buf, "0x%x\n", hv_dev->vendor_id);
 560 }
 561 static DEVICE_ATTR_RO(vendor);
 562
 563 static ssize_t device_show(struct device *dev,
 564                            struct device_attribute *dev_attr,
 565                            char *buf)
 566 {
 567         struct hv_device *hv_dev = device_to_hv_device(dev);
 568
 569         return sprintf(buf, "0x%x\n", hv_dev->device_id);
 570 }
 571 static DEVICE_ATTR_RO(device);
 572
 573 static ssize_t driver_override_store(struct device *dev,
 574                                      struct device_attribute *attr,
 575                                      const char *buf, size_t count)
 576 {
 577         struct hv_device *hv_dev = device_to_hv_device(dev);
 578         char *driver_override, *old, *cp;
 579
 580         /* We need to keep extra room for a newline */
 581         if (count >= (PAGE_SIZE - 1))
 582                 return -EINVAL;
 583
 584         driver_override = kstrndup(buf, count, GFP_KERNEL);
 585         if (!driver_override)
 586                 return -ENOMEM;
 587
 588         cp = strchr(driver_override, '\n');
 589         if (cp)
 590                 *cp = '\0';
 591
 592         device_lock(dev);
 593         old = hv_dev->driver_override;
 594         if (strlen(driver_override)) {
 595                 hv_dev->driver_override = driver_override;
 596         } else {
 597                 kfree(driver_override);
 598                 hv_dev->driver_override = NULL;
 599         }
 600         device_unlock(dev);
 601
 602         kfree(old);
 603
 604         return count;
 605 }
 606
 607 static ssize_t driver_override_show(struct device *dev,
 608                                     struct device_attribute *attr, char *buf)
 609 {
 610         struct hv_device *hv_dev = device_to_hv_device(dev);
 611         ssize_t len;
 612
 613         device_lock(dev);
 614         len = snprintf(buf, PAGE_SIZE, "%s\n", hv_dev->driver_override);
 615         device_unlock(dev);
 616
 617         return len;
 618 }
 619 static DEVICE_ATTR_RW(driver_override);
 620
 621 /* Set up per device attributes in /sys/bus/vmbus/devices/<bus device> */
 622 static struct attribute *vmbus_dev_attrs[] = {
 623         &dev_attr_id.attr,
 624         &dev_attr_state.attr,
 625         &dev_attr_monitor_id.attr,
 626         &dev_attr_class_id.attr,
 627         &dev_attr_device_id.attr,
 628         &dev_attr_modalias.attr,
 629 #ifdef CONFIG_NUMA
 630         &dev_attr_numa_node.attr,
 631 #endif
 632         &dev_attr_server_monitor_pending.attr,
 633         &dev_attr_client_monitor_pending.attr,
 634         &dev_attr_server_monitor_latency.attr,
 635         &dev_attr_client_monitor_latency.attr,
 636         &dev_attr_server_monitor_conn_id.attr,
 637         &dev_attr_client_monitor_conn_id.attr,
 638         &dev_attr_out_intr_mask.attr,
 639         &dev_attr_out_read_index.attr,
 640         &dev_attr_out_write_index.attr,
 641         &dev_attr_out_read_bytes_avail.attr,
 642         &dev_attr_out_write_bytes_avail.attr,
 643         &dev_attr_in_intr_mask.attr,
 644         &dev_attr_in_read_index.attr,
 645         &dev_attr_in_write_index.attr,
 646         &dev_attr_in_read_bytes_avail.attr,
 647         &dev_attr_in_write_bytes_avail.attr,
 648         &dev_attr_channel_vp_mapping.attr,
 649         &dev_attr_vendor.attr,
 650         &dev_attr_device.attr,
 651         &dev_attr_driver_override.attr,
 652         NULL,
 653 };
 654
 655 /*
 656  * Device-level attribute_group callback function. Returns the permission for
 657  * each attribute, and returns 0 if an attribute is not visible.
 658  */
 659 static umode_t vmbus_dev_attr_is_visible(struct kobject *kobj,
 660                                          struct attribute *attr, int idx)
 661 {
 662         struct device *dev = kobj_to_dev(kobj);
 663         const struct hv_device *hv_dev = device_to_hv_device(dev);
 664
 665         /* Hide the monitor attributes if the monitor mechanism is not used. */
 666         if (!hv_dev->channel->offermsg.monitor_allocated &&
 667             (attr == &dev_attr_monitor_id.attr ||
 668              attr == &dev_attr_server_monitor_pending.attr ||
 669              attr == &dev_attr_client_monitor_pending.attr ||
 670              attr == &dev_attr_server_monitor_latency.attr ||
 671              attr == &dev_attr_client_monitor_latency.attr ||
 672              attr == &dev_attr_server_monitor_conn_id.attr ||
 673              attr == &dev_attr_client_monitor_conn_id.attr))
 674                 return 0;
 675
 676         return attr->mode;
 677 }
 678
 679 static const struct attribute_group vmbus_dev_group = {
 680         .attrs = vmbus_dev_attrs,
 681         .is_visible = vmbus_dev_attr_is_visible
 682 };
 683 __ATTRIBUTE_GROUPS(vmbus_dev);
 684
 685 /* Set up the attribute for /sys/bus/vmbus/hibernation */
 686 static ssize_t hibernation_show(struct bus_type *bus, char *buf)
 687 {
 688         return sprintf(buf, "%d\n", !!hv_is_hibernation_supported());
 689 }
 690
 691 static BUS_ATTR_RO(hibernation);
 692
 693 static struct attribute *vmbus_bus_attrs[] = {
 694         &bus_attr_hibernation.attr,
 695         NULL,
 696 };
 697 static const struct attribute_group vmbus_bus_group = {
 698         .attrs = vmbus_bus_attrs,
 699 };
 700 __ATTRIBUTE_GROUPS(vmbus_bus);
 701
 702 /*
 703  * vmbus_uevent - add uevent for our device
 704  *
 705  * This routine is invoked when a device is added or removed on the vmbus to
 706  * generate a uevent to udev in the userspace. The udev will then look at its
 707  * rule and the uevent generated here to load the appropriate driver
 708  *
 709  * The alias string will be of the form vmbus:guid where guid is the string
 710  * representation of the device guid (each byte of the guid will be
 711  * represented with two hex characters.
 712  */
 713 static int vmbus_uevent(struct device *device, struct kobj_uevent_env *env)
 714 {
 715         struct hv_device *dev = device_to_hv_device(device);
 716         const char *format = "MODALIAS=vmbus:%*phN";
 717
 718         return add_uevent_var(env, format, UUID_SIZE, &dev->dev_type);
 719 }
 720
 721 static const struct hv_vmbus_device_id *
 722 hv_vmbus_dev_match(const struct hv_vmbus_device_id *id, const guid_t *guid)
 723 {
 724         if (id == NULL)
 725                 return NULL; /* empty device table */
 726
 727         for (; !guid_is_null(&id->guid); id++)
 728                 if (guid_equal(&id->guid, guid))
 729                         return id;
 730
 731         return NULL;
 732 }
 733
 734 static const struct hv_vmbus_device_id *
 735 hv_vmbus_dynid_match(struct hv_driver *drv, const guid_t *guid)
 736 {
 737         const struct hv_vmbus_device_id *id = NULL;
 738         struct vmbus_dynid *dynid;
 739
 740         spin_lock(&drv->dynids.lock);
 741         list_for_each_entry(dynid, &drv->dynids.list, node) {
 742                 if (guid_equal(&dynid->id.guid, guid)) {
 743                         id = &dynid->id;
 744                         break;
 745                 }
 746         }
 747         spin_unlock(&drv->dynids.lock);
 748
 749         return id;
 750 }
 751
 752 static const struct hv_vmbus_device_id vmbus_device_null;
 753
 754 /*
 755  * Return a matching hv_vmbus_device_id pointer.
 756  * If there is no match, return NULL.
 757  */
 758 static const struct hv_vmbus_device_id *hv_vmbus_get_id(struct hv_driver *drv,
 759                                                         struct hv_device *dev)
 760 {
 761         const guid_t *guid = &dev->dev_type;
 762         const struct hv_vmbus_device_id *id;
 763
 764         /* When driver_override is set, only bind to the matching driver */
 765         if (dev->driver_override && strcmp(dev->driver_override, drv->name))
 766                 return NULL;
 767
 768         /* Look at the dynamic ids first, before the static ones */
 769         id = hv_vmbus_dynid_match(drv, guid);
 770         if (!id)
 771                 id = hv_vmbus_dev_match(drv->id_table, guid);
 772
 773         /* driver_override will always match, send a dummy id */
 774         if (!id && dev->driver_override)
 775                 id = &vmbus_device_null;
 776
 777         return id;
 778 }
 779
 780 /* vmbus_add_dynid - add a new device ID to this driver and re-probe devices */
 781 static int vmbus_add_dynid(struct hv_driver *drv, guid_t *guid)
 782 {
 783         struct vmbus_dynid *dynid;
 784
 785         dynid = kzalloc(sizeof(*dynid), GFP_KERNEL);
 786         if (!dynid)
 787                 return -ENOMEM;
 788
 789         dynid->id.guid = *guid;
 790
 791         spin_lock(&drv->dynids.lock);
 792         list_add_tail(&dynid->node, &drv->dynids.list);
 793         spin_unlock(&drv->dynids.lock);
 794
 795         return driver_attach(&drv->driver);
 796 }
 797
 798 static void vmbus_free_dynids(struct hv_driver *drv)
 799 {
 800         struct vmbus_dynid *dynid, *n;
 801
 802         spin_lock(&drv->dynids.lock);
 803         list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
 804                 list_del(&dynid->node);
 805                 kfree(dynid);
 806         }
 807         spin_unlock(&drv->dynids.lock);
 808 }
 809
 810 /*
 811  * store_new_id - sysfs frontend to vmbus_add_dynid()
 812  *
 813  * Allow GUIDs to be added to an existing driver via sysfs.
 814  */
 815 static ssize_t new_id_store(struct device_driver *driver, const char *buf,
 816                             size_t count)
 817 {
 818         struct hv_driver *drv = drv_to_hv_drv(driver);
 819         guid_t guid;
 820         ssize_t retval;
 821
 822         retval = guid_parse(buf, &guid);
 823         if (retval)
 824                 return retval;
 825
 826         if (hv_vmbus_dynid_match(drv, &guid))
 827                 return -EEXIST;
 828
 829         retval = vmbus_add_dynid(drv, &guid);
 830         if (retval)
 831                 return retval;
 832         return count;
 833 }
 834 static DRIVER_ATTR_WO(new_id);
 835
 836 /*
 837  * store_remove_id - remove a PCI device ID from this driver
 838  *
 839  * Removes a dynamic pci device ID to this driver.
 840  */
 841 static ssize_t remove_id_store(struct device_driver *driver, const char *buf,
 842                                size_t count)
 843 {
 844         struct hv_driver *drv = drv_to_hv_drv(driver);
 845         struct vmbus_dynid *dynid, *n;
 846         guid_t guid;
 847         ssize_t retval;
 848
 849         retval = guid_parse(buf, &guid);
 850         if (retval)
 851                 return retval;
 852
 853         retval = -ENODEV;
 854         spin_lock(&drv->dynids.lock);
 855         list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
 856                 struct hv_vmbus_device_id *id = &dynid->id;
 857
 858                 if (guid_equal(&id->guid, &guid)) {
 859                         list_del(&dynid->node);
 860                         kfree(dynid);
 861                         retval = count;
 862                         break;
 863                 }
 864         }
 865         spin_unlock(&drv->dynids.lock);
 866
 867         return retval;
 868 }
 869 static DRIVER_ATTR_WO(remove_id);
 870
 871 static struct attribute *vmbus_drv_attrs[] = {
 872         &driver_attr_new_id.attr,
 873         &driver_attr_remove_id.attr,
 874         NULL,
 875 };
 876 ATTRIBUTE_GROUPS(vmbus_drv);
 877
 878
 879 /*
 880  * vmbus_match - Attempt to match the specified device to the specified driver
 881  */
 882 static int vmbus_match(struct device *device, struct device_driver *driver)
 883 {
 884         struct hv_driver *drv = drv_to_hv_drv(driver);
 885         struct hv_device *hv_dev = device_to_hv_device(device);
 886
 887         /* The hv_sock driver handles all hv_sock offers. */
 888         if (is_hvsock_channel(hv_dev->channel))
 889                 return drv->hvsock;
 890
 891         if (hv_vmbus_get_id(drv, hv_dev))
 892                 return 1;
 893
 894         return 0;
 895 }
 896
 897 /*
 898  * vmbus_probe - Add the new vmbus's child device
 899  */
 900 static int vmbus_probe(struct device *child_device)
 901 {
 902         int ret = 0;
 903         struct hv_driver *drv =
 904                         drv_to_hv_drv(child_device->driver);
 905         struct hv_device *dev = device_to_hv_device(child_device);
 906         const struct hv_vmbus_device_id *dev_id;
 907
 908         dev_id = hv_vmbus_get_id(drv, dev);
 909         if (drv->probe) {
 910                 ret = drv->probe(dev, dev_id);
 911                 if (ret != 0)
 912                         pr_err("probe failed for device %s (%d)\n",
 913                                dev_name(child_device), ret);
 914
 915         } else {
 916                 pr_err("probe not set for driver %s\n",
 917                        dev_name(child_device));
 918                 ret = -ENODEV;
 919         }
 920         return ret;
 921 }
 922
 923 /*
 924  * vmbus_remove - Remove a vmbus device
 925  */
 926 static void vmbus_remove(struct device *child_device)
 927 {
 928         struct hv_driver *drv;
 929         struct hv_device *dev = device_to_hv_device(child_device);
 930
 931         if (child_device->driver) {
 932                 drv = drv_to_hv_drv(child_device->driver);
 933                 if (drv->remove)
 934                         drv->remove(dev);
 935         }
 936 }
 937
 938 /*
 939  * vmbus_shutdown - Shutdown a vmbus device
 940  */
 941 static void vmbus_shutdown(struct device *child_device)
 942 {
 943         struct hv_driver *drv;
 944         struct hv_device *dev = device_to_hv_device(child_device);
 945
 946
 947         /* The device may not be attached yet */
 948         if (!child_device->driver)
 949                 return;
 950
 951         drv = drv_to_hv_drv(child_device->driver);
 952
 953         if (drv->shutdown)
 954                 drv->shutdown(dev);
 955 }
 956
 957 #ifdef CONFIG_PM_SLEEP
 958 /*
 959  * vmbus_suspend - Suspend a vmbus device
 960  */
 961 static int vmbus_suspend(struct device *child_device)
 962 {
 963         struct hv_driver *drv;
 964         struct hv_device *dev = device_to_hv_device(child_device);
 965
 966         /* The device may not be attached yet */
 967         if (!child_device->driver)
 968                 return 0;
 969
 970         drv = drv_to_hv_drv(child_device->driver);
 971         if (!drv->suspend)
 972                 return -EOPNOTSUPP;
 973
 974         return drv->suspend(dev);
 975 }
 976
 977 /*
 978  * vmbus_resume - Resume a vmbus device
 979  */
 980 static int vmbus_resume(struct device *child_device)
 981 {
 982         struct hv_driver *drv;
 983         struct hv_device *dev = device_to_hv_device(child_device);
 984
 985         /* The device may not be attached yet */
 986         if (!child_device->driver)
 987                 return 0;
 988
 989         drv = drv_to_hv_drv(child_device->driver);
 990         if (!drv->resume)
 991                 return -EOPNOTSUPP;
 992
 993         return drv->resume(dev);
 994 }
 995 #else
 996 #define vmbus_suspend NULL
 997 #define vmbus_resume NULL
 998 #endif /* CONFIG_PM_SLEEP */
 999
1000 /*
1001  * vmbus_device_release - Final callback release of the vmbus child device
1002  */
1003 static void vmbus_device_release(struct device *device)
1004 {
1005         struct hv_device *hv_dev = device_to_hv_device(device);
1006         struct vmbus_channel *channel = hv_dev->channel;
1007
1008         hv_debug_rm_dev_dir(hv_dev);
1009
1010         mutex_lock(&vmbus_connection.channel_mutex);
1011         hv_process_channel_removal(channel);
1012         mutex_unlock(&vmbus_connection.channel_mutex);
1013         kfree(hv_dev);
1014 }
1015
1016 /*
1017  * Note: we must use the "noirq" ops: see the comment before vmbus_bus_pm.
1018  *
1019  * suspend_noirq/resume_noirq are set to NULL to support Suspend-to-Idle: we
1020  * shouldn't suspend the vmbus devices upon Suspend-to-Idle, otherwise there
1021  * is no way to wake up a Generation-2 VM.
1022  *
1023  * The other 4 ops are for hibernation.
1024  */
1025
1026 static const struct dev_pm_ops vmbus_pm = {
1027         .suspend_noirq  = NULL,
1028         .resume_noirq   = NULL,
1029         .freeze_noirq   = vmbus_suspend,
1030         .thaw_noirq     = vmbus_resume,
1031         .poweroff_noirq = vmbus_suspend,
1032         .restore_noirq  = vmbus_resume,
1033 };
1034
1035 /* The one and only one */
1036 static struct bus_type  hv_bus = {
1037         .name =         "vmbus",
1038         .match =                vmbus_match,
1039         .shutdown =             vmbus_shutdown,
1040         .remove =               vmbus_remove,
1041         .probe =                vmbus_probe,
1042         .uevent =               vmbus_uevent,
1043         .dev_groups =           vmbus_dev_groups,
1044         .drv_groups =           vmbus_drv_groups,
1045         .bus_groups =           vmbus_bus_groups,
1046         .pm =                   &vmbus_pm,
1047 };
1048
1049 struct onmessage_work_context {
1050         struct work_struct work;
1051         struct {
1052                 struct hv_message_header header;
1053                 u8 payload[];
1054         } msg;
1055 };
1056
1057 static void vmbus_onmessage_work(struct work_struct *work)
1058 {
1059         struct onmessage_work_context *ctx;
1060
1061         /* Do not process messages if we're in DISCONNECTED state */
1062         if (vmbus_connection.conn_state == DISCONNECTED)
1063                 return;
1064
1065         ctx = container_of(work, struct onmessage_work_context,
1066                            work);
1067         vmbus_onmessage((struct vmbus_channel_message_header *)
1068                         &ctx->msg.payload);
1069         kfree(ctx);
1070 }
1071
1072 void vmbus_on_msg_dpc(unsigned long data)
1073 {
1074         struct hv_per_cpu_context *hv_cpu = (void *)data;
1075         void *page_addr = hv_cpu->synic_message_page;
1076         struct hv_message msg_copy, *msg = (struct hv_message *)page_addr +
1077                                   VMBUS_MESSAGE_SINT;
1078         struct vmbus_channel_message_header *hdr;
1079         enum vmbus_channel_message_type msgtype;
1080         const struct vmbus_channel_message_table_entry *entry;
1081         struct onmessage_work_context *ctx;
1082         __u8 payload_size;
1083         u32 message_type;
1084
1085         /*
1086          * 'enum vmbus_channel_message_type' is supposed to always be 'u32' as
1087          * it is being used in 'struct vmbus_channel_message_header' definition
1088          * which is supposed to match hypervisor ABI.
1089          */
1090         BUILD_BUG_ON(sizeof(enum vmbus_channel_message_type) != sizeof(u32));
1091
1092         /*
1093          * Since the message is in memory shared with the host, an erroneous or
1094          * malicious Hyper-V could modify the message while vmbus_on_msg_dpc()
1095          * or individual message handlers are executing; to prevent this, copy
1096          * the message into private memory.
1097          */
1098         memcpy(&msg_copy, msg, sizeof(struct hv_message));
1099
1100         message_type = msg_copy.header.message_type;
1101         if (message_type == HVMSG_NONE)
1102                 /* no msg */
1103                 return;
1104
1105         hdr = (struct vmbus_channel_message_header *)msg_copy.u.payload;
1106         msgtype = hdr->msgtype;
1107
1108         trace_vmbus_on_msg_dpc(hdr);
1109
1110         if (msgtype >= CHANNELMSG_COUNT) {
1111                 WARN_ONCE(1, "unknown msgtype=%d\n", msgtype);
1112                 goto msg_handled;
1113         }
1114
1115         payload_size = msg_copy.header.payload_size;
1116         if (payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT) {
1117                 WARN_ONCE(1, "payload size is too large (%d)\n", payload_size);
1118                 goto msg_handled;
1119         }
1120
1121         entry = &channel_message_table[msgtype];
1122
1123         if (!entry->message_handler)
1124                 goto msg_handled;
1125
1126         if (payload_size < entry->min_payload_len) {
1127                 WARN_ONCE(1, "message too short: msgtype=%d len=%d\n", msgtype, payload_size);
1128                 goto msg_handled;
1129         }
1130
1131         if (entry->handler_type == VMHT_BLOCKING) {
1132                 ctx = kmalloc(sizeof(*ctx) + payload_size, GFP_ATOMIC);
1133                 if (ctx == NULL)
1134                         return;
1135
1136                 INIT_WORK(&ctx->work, vmbus_onmessage_work);
1137                 memcpy(&ctx->msg, &msg_copy, sizeof(msg->header) + payload_size);
1138
1139                 /*
1140                  * The host can generate a rescind message while we
1141                  * may still be handling the original offer. We deal with
1142                  * this condition by relying on the synchronization provided
1143                  * by offer_in_progress and by channel_mutex.  See also the
1144                  * inline comments in vmbus_onoffer_rescind().
1145                  */
1146                 switch (msgtype) {
1147                 case CHANNELMSG_RESCIND_CHANNELOFFER:
1148                         /*
1149                          * If we are handling the rescind message;
1150                          * schedule the work on the global work queue.
1151                          *
1152                          * The OFFER message and the RESCIND message should
1153                          * not be handled by the same serialized work queue,
1154                          * because the OFFER handler may call vmbus_open(),
1155                          * which tries to open the channel by sending an
1156                          * OPEN_CHANNEL message to the host and waits for
1157                          * the host's response; however, if the host has
1158                          * rescinded the channel before it receives the
1159                          * OPEN_CHANNEL message, the host just silently
1160                          * ignores the OPEN_CHANNEL message; as a result,
1161                          * the guest's OFFER handler hangs for ever, if we
1162                          * handle the RESCIND message in the same serialized
1163                          * work queue: the RESCIND handler can not start to
1164                          * run before the OFFER handler finishes.
1165                          */
1166                         schedule_work(&ctx->work);
1167                         break;
1168
1169                 case CHANNELMSG_OFFERCHANNEL:
1170                         /*
1171                          * The host sends the offer message of a given channel
1172                          * before sending the rescind message of the same
1173                          * channel.  These messages are sent to the guest's
1174                          * connect CPU; the guest then starts processing them
1175                          * in the tasklet handler on this CPU:
1176                          *
1177                          * VMBUS_CONNECT_CPU
1178                          *
1179                          * [vmbus_on_msg_dpc()]
1180                          * atomic_inc()  // CHANNELMSG_OFFERCHANNEL
1181                          * queue_work()
1182                          * ...
1183                          * [vmbus_on_msg_dpc()]
1184                          * schedule_work()  // CHANNELMSG_RESCIND_CHANNELOFFER
1185                          *
1186                          * We rely on the memory-ordering properties of the
1187                          * queue_work() and schedule_work() primitives, which
1188                          * guarantee that the atomic increment will be visible
1189                          * to the CPUs which will execute the offer & rescind
1190                          * works by the time these works will start execution.
1191                          */
1192                         atomic_inc(&vmbus_connection.offer_in_progress);
1193                         fallthrough;
1194
1195                 default:
1196                         queue_work(vmbus_connection.work_queue, &ctx->work);
1197                 }
1198         } else
1199                 entry->message_handler(hdr);
1200
1201 msg_handled:
1202         vmbus_signal_eom(msg, message_type);
1203 }
1204
1205 #ifdef CONFIG_PM_SLEEP
1206 /*
1207  * Fake RESCIND_CHANNEL messages to clean up hv_sock channels by force for
1208  * hibernation, because hv_sock connections can not persist across hibernation.
1209  */
1210 static void vmbus_force_channel_rescinded(struct vmbus_channel *channel)
1211 {
1212         struct onmessage_work_context *ctx;
1213         struct vmbus_channel_rescind_offer *rescind;
1214
1215         WARN_ON(!is_hvsock_channel(channel));
1216
1217         /*
1218          * Allocation size is small and the allocation should really not fail,
1219          * otherwise the state of the hv_sock connections ends up in limbo.
1220          */
1221         ctx = kzalloc(sizeof(*ctx) + sizeof(*rescind),
1222                       GFP_KERNEL | __GFP_NOFAIL);
1223
1224         /*
1225          * So far, these are not really used by Linux. Just set them to the
1226          * reasonable values conforming to the definitions of the fields.
1227          */
1228         ctx->msg.header.message_type = 1;
1229         ctx->msg.header.payload_size = sizeof(*rescind);
1230
1231         /* These values are actually used by Linux. */
1232         rescind = (struct vmbus_channel_rescind_offer *)ctx->msg.payload;
1233         rescind->header.msgtype = CHANNELMSG_RESCIND_CHANNELOFFER;
1234         rescind->child_relid = channel->offermsg.child_relid;
1235
1236         INIT_WORK(&ctx->work, vmbus_onmessage_work);
1237
1238         queue_work(vmbus_connection.work_queue, &ctx->work);
1239 }
1240 #endif /* CONFIG_PM_SLEEP */
1241
1242 /*
1243  * Schedule all channels with events pending
1244  */
1245 static void vmbus_chan_sched(struct hv_per_cpu_context *hv_cpu)
1246 {
1247         unsigned long *recv_int_page;
1248         u32 maxbits, relid;
1249
1250         if (vmbus_proto_version < VERSION_WIN8) {
1251                 maxbits = MAX_NUM_CHANNELS_SUPPORTED;
1252                 recv_int_page = vmbus_connection.recv_int_page;
1253         } else {
1254                 /*
1255                  * When the host is win8 and beyond, the event page
1256                  * can be directly checked to get the id of the channel
1257                  * that has the interrupt pending.
1258                  */
1259                 void *page_addr = hv_cpu->synic_event_page;
1260                 union hv_synic_event_flags *event
1261                         = (union hv_synic_event_flags *)page_addr +
1262                                                  VMBUS_MESSAGE_SINT;
1263
1264                 maxbits = HV_EVENT_FLAGS_COUNT;
1265                 recv_int_page = event->flags;
1266         }
1267
1268         if (unlikely(!recv_int_page))
1269                 return;
1270
1271         for_each_set_bit(relid, recv_int_page, maxbits) {
1272                 void (*callback_fn)(void *context);
1273                 struct vmbus_channel *channel;
1274
1275                 if (!sync_test_and_clear_bit(relid, recv_int_page))
1276                         continue;
1277
1278                 /* Special case - vmbus channel protocol msg */
1279                 if (relid == 0)
1280                         continue;
1281
1282                 /*
1283                  * Pairs with the kfree_rcu() in vmbus_chan_release().
1284                  * Guarantees that the channel data structure doesn't
1285                  * get freed while the channel pointer below is being
1286                  * dereferenced.
1287                  */
1288                 rcu_read_lock();
1289
1290                 /* Find channel based on relid */
1291                 channel = relid2channel(relid);
1292                 if (channel == NULL)
1293                         goto sched_unlock_rcu;
1294
1295                 if (channel->rescind)
1296                         goto sched_unlock_rcu;
1297
1298                 /*
1299                  * Make sure that the ring buffer data structure doesn't get
1300                  * freed while we dereference the ring buffer pointer.  Test
1301                  * for the channel's onchannel_callback being NULL within a
1302                  * sched_lock critical section.  See also the inline comments
1303                  * in vmbus_reset_channel_cb().
1304                  */
1305                 spin_lock(&channel->sched_lock);
1306
1307                 callback_fn = channel->onchannel_callback;
1308                 if (unlikely(callback_fn == NULL))
1309                         goto sched_unlock;
1310
1311                 trace_vmbus_chan_sched(channel);
1312
1313                 ++channel->interrupts;
1314
1315                 switch (channel->callback_mode) {
1316                 case HV_CALL_ISR:
1317                         (*callback_fn)(channel->channel_callback_context);
1318                         break;
1319
1320                 case HV_CALL_BATCHED:
1321                         hv_begin_read(&channel->inbound);
1322                         fallthrough;
1323                 case HV_CALL_DIRECT:
1324                         tasklet_schedule(&channel->callback_event);
1325                 }
1326
1327 sched_unlock:
1328                 spin_unlock(&channel->sched_lock);
1329 sched_unlock_rcu:
1330                 rcu_read_unlock();
1331         }
1332 }
1333
1334 static void vmbus_isr(void)
1335 {
1336         struct hv_per_cpu_context *hv_cpu
1337                 = this_cpu_ptr(hv_context.cpu_context);
1338         void *page_addr = hv_cpu->synic_event_page;
1339         struct hv_message *msg;
1340         union hv_synic_event_flags *event;
1341         bool handled = false;
1342
1343         if (unlikely(page_addr == NULL))
1344                 return;
1345
1346         event = (union hv_synic_event_flags *)page_addr +
1347                                          VMBUS_MESSAGE_SINT;
1348         /*
1349          * Check for events before checking for messages. This is the order
1350          * in which events and messages are checked in Windows guests on
1351          * Hyper-V, and the Windows team suggested we do the same.
1352          */
1353
1354         if ((vmbus_proto_version == VERSION_WS2008) ||
1355                 (vmbus_proto_version == VERSION_WIN7)) {
1356
1357                 /* Since we are a child, we only need to check bit 0 */
1358                 if (sync_test_and_clear_bit(0, event->flags))
1359                         handled = true;
1360         } else {
1361                 /*
1362                  * Our host is win8 or above. The signaling mechanism
1363                  * has changed and we can directly look at the event page.
1364                  * If bit n is set then we have an interrup on the channel
1365                  * whose id is n.
1366                  */
1367                 handled = true;
1368         }
1369
1370         if (handled)
1371                 vmbus_chan_sched(hv_cpu);
1372
1373         page_addr = hv_cpu->synic_message_page;
1374         msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
1375
1376         /* Check if there are actual msgs to be processed */
1377         if (msg->header.message_type != HVMSG_NONE) {
1378                 if (msg->header.message_type == HVMSG_TIMER_EXPIRED) {
1379                         hv_stimer0_isr();
1380                         vmbus_signal_eom(msg, HVMSG_TIMER_EXPIRED);
1381                 } else
1382                         tasklet_schedule(&hv_cpu->msg_dpc);
1383         }
1384
1385         add_interrupt_randomness(vmbus_interrupt);
1386 }
1387
1388 static irqreturn_t vmbus_percpu_isr(int irq, void *dev_id)
1389 {
1390         vmbus_isr();
1391         return IRQ_HANDLED;
1392 }
1393
1394 /*
1395  * Callback from kmsg_dump. Grab as much as possible from the end of the kmsg
1396  * buffer and call into Hyper-V to transfer the data.
1397  */
1398 static void hv_kmsg_dump(struct kmsg_dumper *dumper,
1399                          enum kmsg_dump_reason reason)
1400 {
1401         struct kmsg_dump_iter iter;
1402         size_t bytes_written;
1403
1404         /* We are only interested in panics. */
1405         if ((reason != KMSG_DUMP_PANIC) || (!sysctl_record_panic_msg))
1406                 return;
1407
1408         /*
1409          * Write dump contents to the page. No need to synchronize; panic should
1410          * be single-threaded.
1411          */
1412         kmsg_dump_rewind(&iter);
1413         kmsg_dump_get_buffer(&iter, false, hv_panic_page, HV_HYP_PAGE_SIZE,
1414                              &bytes_written);
1415         if (!bytes_written)
1416                 return;
1417         /*
1418          * P3 to contain the physical address of the panic page & P4 to
1419          * contain the size of the panic data in that page. Rest of the
1420          * registers are no-op when the NOTIFY_MSG flag is set.
1421          */
1422         hv_set_register(HV_REGISTER_CRASH_P0, 0);
1423         hv_set_register(HV_REGISTER_CRASH_P1, 0);
1424         hv_set_register(HV_REGISTER_CRASH_P2, 0);
1425         hv_set_register(HV_REGISTER_CRASH_P3, virt_to_phys(hv_panic_page));
1426         hv_set_register(HV_REGISTER_CRASH_P4, bytes_written);
1427
1428         /*
1429          * Let Hyper-V know there is crash data available along with
1430          * the panic message.
1431          */
1432         hv_set_register(HV_REGISTER_CRASH_CTL,
1433                (HV_CRASH_CTL_CRASH_NOTIFY | HV_CRASH_CTL_CRASH_NOTIFY_MSG));
1434 }
1435
1436 static struct kmsg_dumper hv_kmsg_dumper = {
1437         .dump = hv_kmsg_dump,
1438 };
1439
1440 static void hv_kmsg_dump_register(void)
1441 {
1442         int ret;
1443
1444         hv_panic_page = hv_alloc_hyperv_zeroed_page();
1445         if (!hv_panic_page) {
1446                 pr_err("Hyper-V: panic message page memory allocation failed\n");
1447                 return;
1448         }
1449
1450         ret = kmsg_dump_register(&hv_kmsg_dumper);
1451         if (ret) {
1452                 pr_err("Hyper-V: kmsg dump register error 0x%x\n", ret);
1453                 hv_free_hyperv_page((unsigned long)hv_panic_page);
1454                 hv_panic_page = NULL;
1455         }
1456 }
1457
1458 static struct ctl_table_header *hv_ctl_table_hdr;
1459
1460 /*
1461  * sysctl option to allow the user to control whether kmsg data should be
1462  * reported to Hyper-V on panic.
1463  */
1464 static struct ctl_table hv_ctl_table[] = {
1465         {
1466                 .procname       = "hyperv_record_panic_msg",
1467                 .data           = &sysctl_record_panic_msg,
1468                 .maxlen         = sizeof(int),
1469                 .mode           = 0644,
1470                 .proc_handler   = proc_dointvec_minmax,
1471                 .extra1         = SYSCTL_ZERO,
1472                 .extra2         = SYSCTL_ONE
1473         },
1474         {}
1475 };
1476
1477 static struct ctl_table hv_root_table[] = {
1478         {
1479                 .procname       = "kernel",
1480                 .mode           = 0555,
1481                 .child          = hv_ctl_table
1482         },
1483         {}
1484 };
1485
1486 /*
1487  * vmbus_bus_init -Main vmbus driver initialization routine.
1488  *
1489  * Here, we
1490  *      - initialize the vmbus driver context
1491  *      - invoke the vmbus hv main init routine
1492  *      - retrieve the channel offers
1493  */
1494 static int vmbus_bus_init(void)
1495 {
1496         int ret;
1497
1498         ret = hv_init();
1499         if (ret != 0) {
1500                 pr_err("Unable to initialize the hypervisor - 0x%x\n", ret);
1501                 return ret;
1502         }
1503
1504         ret = bus_register(&hv_bus);
1505         if (ret)
1506                 return ret;
1507
1508         /*
1509          * VMbus interrupts are best modeled as per-cpu interrupts. If
1510          * on an architecture with support for per-cpu IRQs (e.g. ARM64),
1511          * allocate a per-cpu IRQ using standard Linux kernel functionality.
1512          * If not on such an architecture (e.g., x86/x64), then rely on
1513          * code in the arch-specific portion of the code tree to connect
1514          * the VMbus interrupt handler.
1515          */
1516
1517         if (vmbus_irq == -1) {
1518                 hv_setup_vmbus_handler(vmbus_isr);
1519         } else {
1520                 vmbus_evt = alloc_percpu(long);
1521                 ret = request_percpu_irq(vmbus_irq, vmbus_percpu_isr,
1522                                 "Hyper-V VMbus", vmbus_evt);
1523                 if (ret) {
1524                         pr_err("Can't request Hyper-V VMbus IRQ %d, Err %d",
1525                                         vmbus_irq, ret);
1526                         free_percpu(vmbus_evt);
1527                         goto err_setup;
1528                 }
1529         }
1530
1531         ret = hv_synic_alloc();
1532         if (ret)
1533                 goto err_alloc;
1534
1535         /*
1536          * Initialize the per-cpu interrupt state and stimer state.
1537          * Then connect to the host.
1538          */
1539         ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "hyperv/vmbus:online",
1540                                 hv_synic_init, hv_synic_cleanup);
1541         if (ret < 0)
1542                 goto err_cpuhp;
1543         hyperv_cpuhp_online = ret;
1544
1545         ret = vmbus_connect();
1546         if (ret)
1547                 goto err_connect;
1548
1549         /*
1550          * Only register if the crash MSRs are available
1551          */
1552         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1553                 u64 hyperv_crash_ctl;
1554                 /*
1555                  * Sysctl registration is not fatal, since by default
1556                  * reporting is enabled.
1557                  */
1558                 hv_ctl_table_hdr = register_sysctl_table(hv_root_table);
1559                 if (!hv_ctl_table_hdr)
1560                         pr_err("Hyper-V: sysctl table register error");
1561
1562                 /*
1563                  * Register for panic kmsg callback only if the right
1564                  * capability is supported by the hypervisor.
1565                  */
1566                 hyperv_crash_ctl = hv_get_register(HV_REGISTER_CRASH_CTL);
1567                 if (hyperv_crash_ctl & HV_CRASH_CTL_CRASH_NOTIFY_MSG)
1568                         hv_kmsg_dump_register();
1569
1570                 register_die_notifier(&hyperv_die_block);
1571         }
1572
1573         /*
1574          * Always register the panic notifier because we need to unload
1575          * the VMbus channel connection to prevent any VMbus
1576          * activity after the VM panics.
1577          */
1578         atomic_notifier_chain_register(&panic_notifier_list,
1579                                &hyperv_panic_block);
1580
1581         vmbus_request_offers();
1582
1583         return 0;
1584
1585 err_connect:
1586         cpuhp_remove_state(hyperv_cpuhp_online);
1587 err_cpuhp:
1588         hv_synic_free();
1589 err_alloc:
1590         if (vmbus_irq == -1) {
1591                 hv_remove_vmbus_handler();
1592         } else {
1593                 free_percpu_irq(vmbus_irq, vmbus_evt);
1594                 free_percpu(vmbus_evt);
1595         }
1596 err_setup:
1597         bus_unregister(&hv_bus);
1598         unregister_sysctl_table(hv_ctl_table_hdr);
1599         hv_ctl_table_hdr = NULL;
1600         return ret;
1601 }
1602
1603 /**
1604  * __vmbus_child_driver_register() - Register a vmbus's driver
1605  * @hv_driver: Pointer to driver structure you want to register
1606  * @owner: owner module of the drv
1607  * @mod_name: module name string
1608  *
1609  * Registers the given driver with Linux through the 'driver_register()' call
1610  * and sets up the hyper-v vmbus handling for this driver.
1611  * It will return the state of the 'driver_register()' call.
1612  *
1613  */
1614 int __vmbus_driver_register(struct hv_driver *hv_driver, struct module *owner, const char *mod_name)
1615 {
1616         int ret;
1617
1618         pr_info("registering driver %s\n", hv_driver->name);
1619
1620         ret = vmbus_exists();
1621         if (ret < 0)
1622                 return ret;
1623
1624         hv_driver->driver.name = hv_driver->name;
1625         hv_driver->driver.owner = owner;
1626         hv_driver->driver.mod_name = mod_name;
1627         hv_driver->driver.bus = &hv_bus;
1628
1629         spin_lock_init(&hv_driver->dynids.lock);
1630         INIT_LIST_HEAD(&hv_driver->dynids.list);
1631
1632         ret = driver_register(&hv_driver->driver);
1633
1634         return ret;
1635 }
1636 EXPORT_SYMBOL_GPL(__vmbus_driver_register);
1637
1638 /**
1639  * vmbus_driver_unregister() - Unregister a vmbus's driver
1640  * @hv_driver: Pointer to driver structure you want to
1641  *             un-register
1642  *
1643  * Un-register the given driver that was previous registered with a call to
1644  * vmbus_driver_register()
1645  */
1646 void vmbus_driver_unregister(struct hv_driver *hv_driver)
1647 {
1648         pr_info("unregistering driver %s\n", hv_driver->name);
1649
1650         if (!vmbus_exists()) {
1651                 driver_unregister(&hv_driver->driver);
1652                 vmbus_free_dynids(hv_driver);
1653         }
1654 }
1655 EXPORT_SYMBOL_GPL(vmbus_driver_unregister);
1656
1657
1658 /*
1659  * Called when last reference to channel is gone.
1660  */
1661 static void vmbus_chan_release(struct kobject *kobj)
1662 {
1663         struct vmbus_channel *channel
1664                 = container_of(kobj, struct vmbus_channel, kobj);
1665
1666         kfree_rcu(channel, rcu);
1667 }
1668
1669 struct vmbus_chan_attribute {
1670         struct attribute attr;
1671         ssize_t (*show)(struct vmbus_channel *chan, char *buf);
1672         ssize_t (*store)(struct vmbus_channel *chan,
1673                          const char *buf, size_t count);
1674 };
1675 #define VMBUS_CHAN_ATTR(_name, _mode, _show, _store) \
1676         struct vmbus_chan_attribute chan_attr_##_name \
1677                 = __ATTR(_name, _mode, _show, _store)
1678 #define VMBUS_CHAN_ATTR_RW(_name) \
1679         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RW(_name)
1680 #define VMBUS_CHAN_ATTR_RO(_name) \
1681         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RO(_name)
1682 #define VMBUS_CHAN_ATTR_WO(_name) \
1683         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_WO(_name)
1684
1685 static ssize_t vmbus_chan_attr_show(struct kobject *kobj,
1686                                     struct attribute *attr, char *buf)
1687 {
1688         const struct vmbus_chan_attribute *attribute
1689                 = container_of(attr, struct vmbus_chan_attribute, attr);
1690         struct vmbus_channel *chan
1691                 = container_of(kobj, struct vmbus_channel, kobj);
1692
1693         if (!attribute->show)
1694                 return -EIO;
1695
1696         return attribute->show(chan, buf);
1697 }
1698
1699 static ssize_t vmbus_chan_attr_store(struct kobject *kobj,
1700                                      struct attribute *attr, const char *buf,
1701                                      size_t count)
1702 {
1703         const struct vmbus_chan_attribute *attribute
1704                 = container_of(attr, struct vmbus_chan_attribute, attr);
1705         struct vmbus_channel *chan
1706                 = container_of(kobj, struct vmbus_channel, kobj);
1707
1708         if (!attribute->store)
1709                 return -EIO;
1710
1711         return attribute->store(chan, buf, count);
1712 }
1713
1714 static const struct sysfs_ops vmbus_chan_sysfs_ops = {
1715         .show = vmbus_chan_attr_show,
1716         .store = vmbus_chan_attr_store,
1717 };
1718
1719 static ssize_t out_mask_show(struct vmbus_channel *channel, char *buf)
1720 {
1721         struct hv_ring_buffer_info *rbi = &channel->outbound;
1722         ssize_t ret;
1723
1724         mutex_lock(&rbi->ring_buffer_mutex);
1725         if (!rbi->ring_buffer) {
1726                 mutex_unlock(&rbi->ring_buffer_mutex);
1727                 return -EINVAL;
1728         }
1729
1730         ret = sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1731         mutex_unlock(&rbi->ring_buffer_mutex);
1732         return ret;
1733 }
1734 static VMBUS_CHAN_ATTR_RO(out_mask);
1735
1736 static ssize_t in_mask_show(struct vmbus_channel *channel, char *buf)
1737 {
1738         struct hv_ring_buffer_info *rbi = &channel->inbound;
1739         ssize_t ret;
1740
1741         mutex_lock(&rbi->ring_buffer_mutex);
1742         if (!rbi->ring_buffer) {
1743                 mutex_unlock(&rbi->ring_buffer_mutex);
1744                 return -EINVAL;
1745         }
1746
1747         ret = sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1748         mutex_unlock(&rbi->ring_buffer_mutex);
1749         return ret;
1750 }
1751 static VMBUS_CHAN_ATTR_RO(in_mask);
1752
1753 static ssize_t read_avail_show(struct vmbus_channel *channel, char *buf)
1754 {
1755         struct hv_ring_buffer_info *rbi = &channel->inbound;
1756         ssize_t ret;
1757
1758         mutex_lock(&rbi->ring_buffer_mutex);
1759         if (!rbi->ring_buffer) {
1760                 mutex_unlock(&rbi->ring_buffer_mutex);
1761                 return -EINVAL;
1762         }
1763
1764         ret = sprintf(buf, "%u\n", hv_get_bytes_to_read(rbi));
1765         mutex_unlock(&rbi->ring_buffer_mutex);
1766         return ret;
1767 }
1768 static VMBUS_CHAN_ATTR_RO(read_avail);
1769
1770 static ssize_t write_avail_show(struct vmbus_channel *channel, char *buf)
1771 {
1772         struct hv_ring_buffer_info *rbi = &channel->outbound;
1773         ssize_t ret;
1774
1775         mutex_lock(&rbi->ring_buffer_mutex);
1776         if (!rbi->ring_buffer) {
1777                 mutex_unlock(&rbi->ring_buffer_mutex);
1778                 return -EINVAL;
1779         }
1780
1781         ret = sprintf(buf, "%u\n", hv_get_bytes_to_write(rbi));
1782         mutex_unlock(&rbi->ring_buffer_mutex);
1783         return ret;
1784 }
1785 static VMBUS_CHAN_ATTR_RO(write_avail);
1786
1787 static ssize_t target_cpu_show(struct vmbus_channel *channel, char *buf)
1788 {
1789         return sprintf(buf, "%u\n", channel->target_cpu);
1790 }
1791 static ssize_t target_cpu_store(struct vmbus_channel *channel,
1792                                 const char *buf, size_t count)
1793 {
1794         u32 target_cpu, origin_cpu;
1795         ssize_t ret = count;
1796
1797         if (vmbus_proto_version < VERSION_WIN10_V4_1)
1798                 return -EIO;
1799
1800         if (sscanf(buf, "%uu", &target_cpu) != 1)
1801                 return -EIO;
1802
1803         /* Validate target_cpu for the cpumask_test_cpu() operation below. */
1804         if (target_cpu >= nr_cpumask_bits)
1805                 return -EINVAL;
1806
1807         /* No CPUs should come up or down during this. */
1808         cpus_read_lock();
1809
1810         if (!cpu_online(target_cpu)) {
1811                 cpus_read_unlock();
1812                 return -EINVAL;
1813         }
1814
1815         /*
1816          * Synchronizes target_cpu_store() and channel closure:
1817          *
1818          * { Initially: state = CHANNEL_OPENED }
1819          *
1820          * CPU1                         CPU2
1821          *
1822          * [target_cpu_store()]         [vmbus_disconnect_ring()]
1823          *
1824          * LOCK channel_mutex           LOCK channel_mutex
1825          * LOAD r1 = state              LOAD r2 = state
1826          * IF (r1 == CHANNEL_OPENED)    IF (r2 == CHANNEL_OPENED)
1827          *   SEND MODIFYCHANNEL           STORE state = CHANNEL_OPEN
1828          *   [...]                        SEND CLOSECHANNEL
1829          * UNLOCK channel_mutex         UNLOCK channel_mutex
1830          *
1831          * Forbids: r1 == r2 == CHANNEL_OPENED (i.e., CPU1's LOCK precedes
1832          *              CPU2's LOCK) && CPU2's SEND precedes CPU1's SEND
1833          *
1834          * Note.  The host processes the channel messages "sequentially", in
1835          * the order in which they are received on a per-partition basis.
1836          */
1837         mutex_lock(&vmbus_connection.channel_mutex);
1838
1839         /*
1840          * Hyper-V will ignore MODIFYCHANNEL messages for "non-open" channels;
1841          * avoid sending the message and fail here for such channels.
1842          */
1843         if (channel->state != CHANNEL_OPENED_STATE) {
1844                 ret = -EIO;
1845                 goto cpu_store_unlock;
1846         }
1847
1848         origin_cpu = channel->target_cpu;
1849         if (target_cpu == origin_cpu)
1850                 goto cpu_store_unlock;
1851
1852         if (vmbus_send_modifychannel(channel,
1853                                      hv_cpu_number_to_vp_number(target_cpu))) {
1854                 ret = -EIO;
1855                 goto cpu_store_unlock;
1856         }
1857
1858         /*
1859          * For version before VERSION_WIN10_V5_3, the following warning holds:
1860          *
1861          * Warning.  At this point, there is *no* guarantee that the host will
1862          * have successfully processed the vmbus_send_modifychannel() request.
1863          * See the header comment of vmbus_send_modifychannel() for more info.
1864          *
1865          * Lags in the processing of the above vmbus_send_modifychannel() can
1866          * result in missed interrupts if the "old" target CPU is taken offline
1867          * before Hyper-V starts sending interrupts to the "new" target CPU.
1868          * But apart from this offlining scenario, the code tolerates such
1869          * lags.  It will function correctly even if a channel interrupt comes
1870          * in on a CPU that is different from the channel target_cpu value.
1871          */
1872
1873         channel->target_cpu = target_cpu;
1874
1875         /* See init_vp_index(). */
1876         if (hv_is_perf_channel(channel))
1877                 hv_update_alloced_cpus(origin_cpu, target_cpu);
1878
1879         /* Currently set only for storvsc channels. */
1880         if (channel->change_target_cpu_callback) {
1881                 (*channel->change_target_cpu_callback)(channel,
1882                                 origin_cpu, target_cpu);
1883         }
1884
1885 cpu_store_unlock:
1886         mutex_unlock(&vmbus_connection.channel_mutex);
1887         cpus_read_unlock();
1888         return ret;
1889 }
1890 static VMBUS_CHAN_ATTR(cpu, 0644, target_cpu_show, target_cpu_store);
1891
1892 static ssize_t channel_pending_show(struct vmbus_channel *channel,
1893                                     char *buf)
1894 {
1895         return sprintf(buf, "%d\n",
1896                        channel_pending(channel,
1897                                        vmbus_connection.monitor_pages[1]));
1898 }
1899 static VMBUS_CHAN_ATTR(pending, 0444, channel_pending_show, NULL);
1900
1901 static ssize_t channel_latency_show(struct vmbus_channel *channel,
1902                                     char *buf)
1903 {
1904         return sprintf(buf, "%d\n",
1905                        channel_latency(channel,
1906                                        vmbus_connection.monitor_pages[1]));
1907 }
1908 static VMBUS_CHAN_ATTR(latency, 0444, channel_latency_show, NULL);
1909
1910 static ssize_t channel_interrupts_show(struct vmbus_channel *channel, char *buf)
1911 {
1912         return sprintf(buf, "%llu\n", channel->interrupts);
1913 }
1914 static VMBUS_CHAN_ATTR(interrupts, 0444, channel_interrupts_show, NULL);
1915
1916 static ssize_t channel_events_show(struct vmbus_channel *channel, char *buf)
1917 {
1918         return sprintf(buf, "%llu\n", channel->sig_events);
1919 }
1920 static VMBUS_CHAN_ATTR(events, 0444, channel_events_show, NULL);
1921
1922 static ssize_t channel_intr_in_full_show(struct vmbus_channel *channel,
1923                                          char *buf)
1924 {
1925         return sprintf(buf, "%llu\n",
1926                        (unsigned long long)channel->intr_in_full);
1927 }
1928 static VMBUS_CHAN_ATTR(intr_in_full, 0444, channel_intr_in_full_show, NULL);
1929
1930 static ssize_t channel_intr_out_empty_show(struct vmbus_channel *channel,
1931                                            char *buf)
1932 {
1933         return sprintf(buf, "%llu\n",
1934                        (unsigned long long)channel->intr_out_empty);
1935 }
1936 static VMBUS_CHAN_ATTR(intr_out_empty, 0444, channel_intr_out_empty_show, NULL);
1937
1938 static ssize_t channel_out_full_first_show(struct vmbus_channel *channel,
1939                                            char *buf)
1940 {
1941         return sprintf(buf, "%llu\n",
1942                        (unsigned long long)channel->out_full_first);
1943 }
1944 static VMBUS_CHAN_ATTR(out_full_first, 0444, channel_out_full_first_show, NULL);
1945
1946 static ssize_t channel_out_full_total_show(struct vmbus_channel *channel,
1947                                            char *buf)
1948 {
1949         return sprintf(buf, "%llu\n",
1950                        (unsigned long long)channel->out_full_total);
1951 }
1952 static VMBUS_CHAN_ATTR(out_full_total, 0444, channel_out_full_total_show, NULL);
1953
1954 static ssize_t subchannel_monitor_id_show(struct vmbus_channel *channel,
1955                                           char *buf)
1956 {
1957         return sprintf(buf, "%u\n", channel->offermsg.monitorid);
1958 }
1959 static VMBUS_CHAN_ATTR(monitor_id, 0444, subchannel_monitor_id_show, NULL);
1960
1961 static ssize_t subchannel_id_show(struct vmbus_channel *channel,
1962                                   char *buf)
1963 {
1964         return sprintf(buf, "%u\n",
1965                        channel->offermsg.offer.sub_channel_index);
1966 }
1967 static VMBUS_CHAN_ATTR_RO(subchannel_id);
1968
1969 static struct attribute *vmbus_chan_attrs[] = {
1970         &chan_attr_out_mask.attr,
1971         &chan_attr_in_mask.attr,
1972         &chan_attr_read_avail.attr,
1973         &chan_attr_write_avail.attr,
1974         &chan_attr_cpu.attr,
1975         &chan_attr_pending.attr,
1976         &chan_attr_latency.attr,
1977         &chan_attr_interrupts.attr,
1978         &chan_attr_events.attr,
1979         &chan_attr_intr_in_full.attr,
1980         &chan_attr_intr_out_empty.attr,
1981         &chan_attr_out_full_first.attr,
1982         &chan_attr_out_full_total.attr,
1983         &chan_attr_monitor_id.attr,
1984         &chan_attr_subchannel_id.attr,
1985         NULL
1986 };
1987
1988 /*
1989  * Channel-level attribute_group callback function. Returns the permission for
1990  * each attribute, and returns 0 if an attribute is not visible.
1991  */
1992 static umode_t vmbus_chan_attr_is_visible(struct kobject *kobj,
1993                                           struct attribute *attr, int idx)
1994 {
1995         const struct vmbus_channel *channel =
1996                 container_of(kobj, struct vmbus_channel, kobj);
1997
1998         /* Hide the monitor attributes if the monitor mechanism is not used. */
1999         if (!channel->offermsg.monitor_allocated &&
2000             (attr == &chan_attr_pending.attr ||
2001              attr == &chan_attr_latency.attr ||
2002              attr == &chan_attr_monitor_id.attr))
2003                 return 0;
2004
2005         return attr->mode;
2006 }
2007
2008 static struct attribute_group vmbus_chan_group = {
2009         .attrs = vmbus_chan_attrs,
2010         .is_visible = vmbus_chan_attr_is_visible
2011 };
2012
2013 static struct kobj_type vmbus_chan_ktype = {
2014         .sysfs_ops = &vmbus_chan_sysfs_ops,
2015         .release = vmbus_chan_release,
2016 };
2017
2018 /*
2019  * vmbus_add_channel_kobj - setup a sub-directory under device/channels
2020  */
2021 int vmbus_add_channel_kobj(struct hv_device *dev, struct vmbus_channel *channel)
2022 {
2023         const struct device *device = &dev->device;
2024         struct kobject *kobj = &channel->kobj;
2025         u32 relid = channel->offermsg.child_relid;
2026         int ret;
2027
2028         kobj->kset = dev->channels_kset;
2029         ret = kobject_init_and_add(kobj, &vmbus_chan_ktype, NULL,
2030                                    "%u", relid);
2031         if (ret)
2032                 return ret;
2033
2034         ret = sysfs_create_group(kobj, &vmbus_chan_group);
2035
2036         if (ret) {
2037                 /*
2038                  * The calling functions' error handling paths will cleanup the
2039                  * empty channel directory.
2040                  */
2041                 dev_err(device, "Unable to set up channel sysfs files\n");
2042                 return ret;
2043         }
2044
2045         kobject_uevent(kobj, KOBJ_ADD);
2046
2047         return 0;
2048 }
2049
2050 /*
2051  * vmbus_remove_channel_attr_group - remove the channel's attribute group
2052  */
2053 void vmbus_remove_channel_attr_group(struct vmbus_channel *channel)
2054 {
2055         sysfs_remove_group(&channel->kobj, &vmbus_chan_group);
2056 }
2057
2058 /*
2059  * vmbus_device_create - Creates and registers a new child device
2060  * on the vmbus.
2061  */
2062 struct hv_device *vmbus_device_create(const guid_t *type,
2063                                       const guid_t *instance,
2064                                       struct vmbus_channel *channel)
2065 {
2066         struct hv_device *child_device_obj;
2067
2068         child_device_obj = kzalloc(sizeof(struct hv_device), GFP_KERNEL);
2069         if (!child_device_obj) {
2070                 pr_err("Unable to allocate device object for child device\n");
2071                 return NULL;
2072         }
2073
2074         child_device_obj->channel = channel;
2075         guid_copy(&child_device_obj->dev_type, type);
2076         guid_copy(&child_device_obj->dev_instance, instance);
2077         child_device_obj->vendor_id = 0x1414; /* MSFT vendor ID */
2078
2079         return child_device_obj;
2080 }
2081
2082 static u64 vmbus_dma_mask = DMA_BIT_MASK(64);
2083 /*
2084  * vmbus_device_register - Register the child device
2085  */
2086 int vmbus_device_register(struct hv_device *child_device_obj)
2087 {
2088         struct kobject *kobj = &child_device_obj->device.kobj;
2089         int ret;
2090
2091         dev_set_name(&child_device_obj->device, "%pUl",
2092                      &child_device_obj->channel->offermsg.offer.if_instance);
2093
2094         child_device_obj->device.bus = &hv_bus;
2095         child_device_obj->device.parent = &hv_acpi_dev->dev;
2096         child_device_obj->device.release = vmbus_device_release;
2097
2098         /*
2099          * Register with the LDM. This will kick off the driver/device
2100          * binding...which will eventually call vmbus_match() and vmbus_probe()
2101          */
2102         ret = device_register(&child_device_obj->device);
2103         if (ret) {
2104                 pr_err("Unable to register child device\n");
2105                 return ret;
2106         }
2107
2108         child_device_obj->channels_kset = kset_create_and_add("channels",
2109                                                               NULL, kobj);
2110         if (!child_device_obj->channels_kset) {
2111                 ret = -ENOMEM;
2112                 goto err_dev_unregister;
2113         }
2114
2115         ret = vmbus_add_channel_kobj(child_device_obj,
2116                                      child_device_obj->channel);
2117         if (ret) {
2118                 pr_err("Unable to register primary channeln");
2119                 goto err_kset_unregister;
2120         }
2121         hv_debug_add_dev_dir(child_device_obj);
2122
2123         child_device_obj->device.dma_mask = &vmbus_dma_mask;
2124         child_device_obj->device.dma_parms = &child_device_obj->dma_parms;
2125         return 0;
2126
2127 err_kset_unregister:
2128         kset_unregister(child_device_obj->channels_kset);
2129
2130 err_dev_unregister:
2131         device_unregister(&child_device_obj->device);
2132         return ret;
2133 }
2134
2135 /*
2136  * vmbus_device_unregister - Remove the specified child device
2137  * from the vmbus.
2138  */
2139 void vmbus_device_unregister(struct hv_device *device_obj)
2140 {
2141         pr_debug("child device %s unregistered\n",
2142                 dev_name(&device_obj->device));
2143
2144         kset_unregister(device_obj->channels_kset);
2145
2146         /*
2147          * Kick off the process of unregistering the device.
2148          * This will call vmbus_remove() and eventually vmbus_device_release()
2149          */
2150         device_unregister(&device_obj->device);
2151 }
2152
2153
2154 /*
2155  * VMBUS is an acpi enumerated device. Get the information we
2156  * need from DSDT.
2157  */
2158 #define VTPM_BASE_ADDRESS 0xfed40000
2159 static acpi_status vmbus_walk_resources(struct acpi_resource *res, void *ctx)
2160 {
2161         resource_size_t start = 0;
2162         resource_size_t end = 0;
2163         struct resource *new_res;
2164         struct resource **old_res = &hyperv_mmio;
2165         struct resource **prev_res = NULL;
2166         struct resource r;
2167
2168         switch (res->type) {
2169
2170         /*
2171          * "Address" descriptors are for bus windows. Ignore
2172          * "memory" descriptors, which are for registers on
2173          * devices.
2174          */
2175         case ACPI_RESOURCE_TYPE_ADDRESS32:
2176                 start = res->data.address32.address.minimum;
2177                 end = res->data.address32.address.maximum;
2178                 break;
2179
2180         case ACPI_RESOURCE_TYPE_ADDRESS64:
2181                 start = res->data.address64.address.minimum;
2182                 end = res->data.address64.address.maximum;
2183                 break;
2184
2185         /*
2186          * The IRQ information is needed only on ARM64, which Hyper-V
2187          * sets up in the extended format. IRQ information is present
2188          * on x86/x64 in the non-extended format but it is not used by
2189          * Linux. So don't bother checking for the non-extended format.
2190          */
2191         case ACPI_RESOURCE_TYPE_EXTENDED_IRQ:
2192                 if (!acpi_dev_resource_interrupt(res, 0, &r)) {
2193                         pr_err("Unable to parse Hyper-V ACPI interrupt\n");
2194                         return AE_ERROR;
2195                 }
2196                 /* ARM64 INTID for VMbus */
2197                 vmbus_interrupt = res->data.extended_irq.interrupts[0];
2198                 /* Linux IRQ number */
2199                 vmbus_irq = r.start;
2200                 return AE_OK;
2201
2202         default:
2203                 /* Unused resource type */
2204                 return AE_OK;
2205
2206         }
2207         /*
2208          * Ignore ranges that are below 1MB, as they're not
2209          * necessary or useful here.
2210          */
2211         if (end < 0x100000)
2212                 return AE_OK;
2213
2214         new_res = kzalloc(sizeof(*new_res), GFP_ATOMIC);
2215         if (!new_res)
2216                 return AE_NO_MEMORY;
2217
2218         /* If this range overlaps the virtual TPM, truncate it. */
2219         if (end > VTPM_BASE_ADDRESS && start < VTPM_BASE_ADDRESS)
2220                 end = VTPM_BASE_ADDRESS;
2221
2222         new_res->name = "hyperv mmio";
2223         new_res->flags = IORESOURCE_MEM;
2224         new_res->start = start;
2225         new_res->end = end;
2226
2227         /*
2228          * If two ranges are adjacent, merge them.
2229          */
2230         do {
2231                 if (!*old_res) {
2232                         *old_res = new_res;
2233                         break;
2234                 }
2235
2236                 if (((*old_res)->end + 1) == new_res->start) {
2237                         (*old_res)->end = new_res->end;
2238                         kfree(new_res);
2239                         break;
2240                 }
2241
2242                 if ((*old_res)->start == new_res->end + 1) {
2243                         (*old_res)->start = new_res->start;
2244                         kfree(new_res);
2245                         break;
2246                 }
2247
2248                 if ((*old_res)->start > new_res->end) {
2249                         new_res->sibling = *old_res;
2250                         if (prev_res)
2251                                 (*prev_res)->sibling = new_res;
2252                         *old_res = new_res;
2253                         break;
2254                 }
2255
2256                 prev_res = old_res;
2257                 old_res = &(*old_res)->sibling;
2258
2259         } while (1);
2260
2261         return AE_OK;
2262 }
2263
2264 static int vmbus_acpi_remove(struct acpi_device *device)
2265 {
2266         struct resource *cur_res;
2267         struct resource *next_res;
2268
2269         if (hyperv_mmio) {
2270                 if (fb_mmio) {
2271                         __release_region(hyperv_mmio, fb_mmio->start,
2272                                          resource_size(fb_mmio));
2273                         fb_mmio = NULL;
2274                 }
2275
2276                 for (cur_res = hyperv_mmio; cur_res; cur_res = next_res) {
2277                         next_res = cur_res->sibling;
2278                         kfree(cur_res);
2279                 }
2280         }
2281
2282         return 0;
2283 }
2284
2285 static void vmbus_reserve_fb(void)
2286 {
2287         int size;
2288         /*
2289          * Make a claim for the frame buffer in the resource tree under the
2290          * first node, which will be the one below 4GB.  The length seems to
2291          * be underreported, particularly in a Generation 1 VM.  So start out
2292          * reserving a larger area and make it smaller until it succeeds.
2293          */
2294
2295         if (screen_info.lfb_base) {
2296                 if (efi_enabled(EFI_BOOT))
2297                         size = max_t(__u32, screen_info.lfb_size, 0x800000);
2298                 else
2299                         size = max_t(__u32, screen_info.lfb_size, 0x4000000);
2300
2301                 for (; !fb_mmio && (size >= 0x100000); size >>= 1) {
2302                         fb_mmio = __request_region(hyperv_mmio,
2303                                                    screen_info.lfb_base, size,
2304                                                    fb_mmio_name, 0);
2305                 }
2306         }
2307 }
2308
2309 /**
2310  * vmbus_allocate_mmio() - Pick a memory-mapped I/O range.
2311  * @new:                If successful, supplied a pointer to the
2312  *                      allocated MMIO space.
2313  * @device_obj:         Identifies the caller
2314  * @min:                Minimum guest physical address of the
2315  *                      allocation
2316  * @max:                Maximum guest physical address
2317  * @size:               Size of the range to be allocated
2318  * @align:              Alignment of the range to be allocated
2319  * @fb_overlap_ok:      Whether this allocation can be allowed
2320  *                      to overlap the video frame buffer.
2321  *
2322  * This function walks the resources granted to VMBus by the
2323  * _CRS object in the ACPI namespace underneath the parent
2324  * "bridge" whether that's a root PCI bus in the Generation 1
2325  * case or a Module Device in the Generation 2 case.  It then
2326  * attempts to allocate from the global MMIO pool in a way that
2327  * matches the constraints supplied in these parameters and by
2328  * that _CRS.
2329  *
2330  * Return: 0 on success, -errno on failure
2331  */
2332 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
2333                         resource_size_t min, resource_size_t max,
2334                         resource_size_t size, resource_size_t align,
2335                         bool fb_overlap_ok)
2336 {
2337         struct resource *iter, *shadow;
2338         resource_size_t range_min, range_max, start;
2339         const char *dev_n = dev_name(&device_obj->device);
2340         int retval;
2341
2342         retval = -ENXIO;
2343         mutex_lock(&hyperv_mmio_lock);
2344
2345         /*
2346          * If overlaps with frame buffers are allowed, then first attempt to
2347          * make the allocation from within the reserved region.  Because it
2348          * is already reserved, no shadow allocation is necessary.
2349          */
2350         if (fb_overlap_ok && fb_mmio && !(min > fb_mmio->end) &&
2351             !(max < fb_mmio->start)) {
2352
2353                 range_min = fb_mmio->start;
2354                 range_max = fb_mmio->end;
2355                 start = (range_min + align - 1) & ~(align - 1);
2356                 for (; start + size - 1 <= range_max; start += align) {
2357                         *new = request_mem_region_exclusive(start, size, dev_n);
2358                         if (*new) {
2359                                 retval = 0;
2360                                 goto exit;
2361                         }
2362                 }
2363         }
2364
2365         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
2366                 if ((iter->start >= max) || (iter->end <= min))
2367                         continue;
2368
2369                 range_min = iter->start;
2370                 range_max = iter->end;
2371                 start = (range_min + align - 1) & ~(align - 1);
2372                 for (; start + size - 1 <= range_max; start += align) {
2373                         shadow = __request_region(iter, start, size, NULL,
2374                                                   IORESOURCE_BUSY);
2375                         if (!shadow)
2376                                 continue;
2377
2378                         *new = request_mem_region_exclusive(start, size, dev_n);
2379                         if (*new) {
2380                                 shadow->name = (char *)*new;
2381                                 retval = 0;
2382                                 goto exit;
2383                         }
2384
2385                         __release_region(iter, start, size);
2386                 }
2387         }
2388
2389 exit:
2390         mutex_unlock(&hyperv_mmio_lock);
2391         return retval;
2392 }
2393 EXPORT_SYMBOL_GPL(vmbus_allocate_mmio);
2394
2395 /**
2396  * vmbus_free_mmio() - Free a memory-mapped I/O range.
2397  * @start:              Base address of region to release.
2398  * @size:               Size of the range to be allocated
2399  *
2400  * This function releases anything requested by
2401  * vmbus_mmio_allocate().
2402  */
2403 void vmbus_free_mmio(resource_size_t start, resource_size_t size)
2404 {
2405         struct resource *iter;
2406
2407         mutex_lock(&hyperv_mmio_lock);
2408         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
2409                 if ((iter->start >= start + size) || (iter->end <= start))
2410                         continue;
2411
2412                 __release_region(iter, start, size);
2413         }
2414         release_mem_region(start, size);
2415         mutex_unlock(&hyperv_mmio_lock);
2416
2417 }
2418 EXPORT_SYMBOL_GPL(vmbus_free_mmio);
2419
2420 static int vmbus_acpi_add(struct acpi_device *device)
2421 {
2422         acpi_status result;
2423         int ret_val = -ENODEV;
2424         struct acpi_device *ancestor;
2425
2426         hv_acpi_dev = device;
2427
2428         result = acpi_walk_resources(device->handle, METHOD_NAME__CRS,
2429                                         vmbus_walk_resources, NULL);
2430
2431         if (ACPI_FAILURE(result))
2432                 goto acpi_walk_err;
2433         /*
2434          * Some ancestor of the vmbus acpi device (Gen1 or Gen2
2435          * firmware) is the VMOD that has the mmio ranges. Get that.
2436          */
2437         for (ancestor = device->parent; ancestor; ancestor = ancestor->parent) {
2438                 result = acpi_walk_resources(ancestor->handle, METHOD_NAME__CRS,
2439                                              vmbus_walk_resources, NULL);
2440
2441                 if (ACPI_FAILURE(result))
2442                         continue;
2443                 if (hyperv_mmio) {
2444                         vmbus_reserve_fb();
2445                         break;
2446                 }
2447         }
2448         ret_val = 0;
2449
2450 acpi_walk_err:
2451         complete(&probe_event);
2452         if (ret_val)
2453                 vmbus_acpi_remove(device);
2454         return ret_val;
2455 }
2456
2457 #ifdef CONFIG_PM_SLEEP
2458 static int vmbus_bus_suspend(struct device *dev)
2459 {
2460         struct vmbus_channel *channel, *sc;
2461
2462         while (atomic_read(&vmbus_connection.offer_in_progress) != 0) {
2463                 /*
2464                  * We wait here until the completion of any channel
2465                  * offers that are currently in progress.
2466                  */
2467                 usleep_range(1000, 2000);
2468         }
2469
2470         mutex_lock(&vmbus_connection.channel_mutex);
2471         list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
2472                 if (!is_hvsock_channel(channel))
2473                         continue;
2474
2475                 vmbus_force_channel_rescinded(channel);
2476         }
2477         mutex_unlock(&vmbus_connection.channel_mutex);
2478
2479         /*
2480          * Wait until all the sub-channels and hv_sock channels have been
2481          * cleaned up. Sub-channels should be destroyed upon suspend, otherwise
2482          * they would conflict with the new sub-channels that will be created
2483          * in the resume path. hv_sock channels should also be destroyed, but
2484          * a hv_sock channel of an established hv_sock connection can not be
2485          * really destroyed since it may still be referenced by the userspace
2486          * application, so we just force the hv_sock channel to be rescinded
2487          * by vmbus_force_channel_rescinded(), and the userspace application
2488          * will thoroughly destroy the channel after hibernation.
2489          *
2490          * Note: the counter nr_chan_close_on_suspend may never go above 0 if
2491          * the VM has no sub-channel and hv_sock channel, e.g. a 1-vCPU VM.
2492          */
2493         if (atomic_read(&vmbus_connection.nr_chan_close_on_suspend) > 0)
2494                 wait_for_completion(&vmbus_connection.ready_for_suspend_event);
2495
2496         if (atomic_read(&vmbus_connection.nr_chan_fixup_on_resume) != 0) {
2497                 pr_err("Can not suspend due to a previous failed resuming\n");
2498                 return -EBUSY;
2499         }
2500
2501         mutex_lock(&vmbus_connection.channel_mutex);
2502
2503         list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
2504                 /*
2505                  * Remove the channel from the array of channels and invalidate
2506                  * the channel's relid.  Upon resume, vmbus_onoffer() will fix
2507                  * up the relid (and other fields, if necessary) and add the
2508                  * channel back to the array.
2509                  */
2510                 vmbus_channel_unmap_relid(channel);
2511                 channel->offermsg.child_relid = INVALID_RELID;
2512
2513                 if (is_hvsock_channel(channel)) {
2514                         if (!channel->rescind) {
2515                                 pr_err("hv_sock channel not rescinded!\n");
2516                                 WARN_ON_ONCE(1);
2517                         }
2518                         continue;
2519                 }
2520
2521                 list_for_each_entry(sc, &channel->sc_list, sc_list) {
2522                         pr_err("Sub-channel not deleted!\n");
2523                         WARN_ON_ONCE(1);
2524                 }
2525
2526                 atomic_inc(&vmbus_connection.nr_chan_fixup_on_resume);
2527         }
2528
2529         mutex_unlock(&vmbus_connection.channel_mutex);
2530
2531         vmbus_initiate_unload(false);
2532
2533         /* Reset the event for the next resume. */
2534         reinit_completion(&vmbus_connection.ready_for_resume_event);
2535
2536         return 0;
2537 }
2538
2539 static int vmbus_bus_resume(struct device *dev)
2540 {
2541         struct vmbus_channel_msginfo *msginfo;
2542         size_t msgsize;
2543         int ret;
2544
2545         /*
2546          * We only use the 'vmbus_proto_version', which was in use before
2547          * hibernation, to re-negotiate with the host.
2548          */
2549         if (!vmbus_proto_version) {
2550                 pr_err("Invalid proto version = 0x%x\n", vmbus_proto_version);
2551                 return -EINVAL;
2552         }
2553
2554         msgsize = sizeof(*msginfo) +
2555                   sizeof(struct vmbus_channel_initiate_contact);
2556
2557         msginfo = kzalloc(msgsize, GFP_KERNEL);
2558
2559         if (msginfo == NULL)
2560                 return -ENOMEM;
2561
2562         ret = vmbus_negotiate_version(msginfo, vmbus_proto_version);
2563
2564         kfree(msginfo);
2565
2566         if (ret != 0)
2567                 return ret;
2568
2569         WARN_ON(atomic_read(&vmbus_connection.nr_chan_fixup_on_resume) == 0);
2570
2571         vmbus_request_offers();
2572
2573         if (wait_for_completion_timeout(
2574                 &vmbus_connection.ready_for_resume_event, 10 * HZ) == 0)
2575                 pr_err("Some vmbus device is missing after suspending?\n");
2576
2577         /* Reset the event for the next suspend. */
2578         reinit_completion(&vmbus_connection.ready_for_suspend_event);
2579
2580         return 0;
2581 }
2582 #else
2583 #define vmbus_bus_suspend NULL
2584 #define vmbus_bus_resume NULL
2585 #endif /* CONFIG_PM_SLEEP */
2586
2587 static const struct acpi_device_id vmbus_acpi_device_ids[] = {
2588         {"VMBUS", 0},
2589         {"VMBus", 0},
2590         {"", 0},
2591 };
2592 MODULE_DEVICE_TABLE(acpi, vmbus_acpi_device_ids);
2593
2594 /*
2595  * Note: we must use the "no_irq" ops, otherwise hibernation can not work with
2596  * PCI device assignment, because "pci_dev_pm_ops" uses the "noirq" ops: in
2597  * the resume path, the pci "noirq" restore op runs before "non-noirq" op (see
2598  * resume_target_kernel() -> dpm_resume_start(), and hibernation_restore() ->
2599  * dpm_resume_end()). This means vmbus_bus_resume() and the pci-hyperv's
2600  * resume callback must also run via the "noirq" ops.
2601  *
2602  * Set suspend_noirq/resume_noirq to NULL for Suspend-to-Idle: see the comment
2603  * earlier in this file before vmbus_pm.
2604  */
2605
2606 static const struct dev_pm_ops vmbus_bus_pm = {
2607         .suspend_noirq  = NULL,
2608         .resume_noirq   = NULL,
2609         .freeze_noirq   = vmbus_bus_suspend,
2610         .thaw_noirq     = vmbus_bus_resume,
2611         .poweroff_noirq = vmbus_bus_suspend,
2612         .restore_noirq  = vmbus_bus_resume
2613 };
2614
2615 static struct acpi_driver vmbus_acpi_driver = {
2616         .name = "vmbus",
2617         .ids = vmbus_acpi_device_ids,
2618         .ops = {
2619                 .add = vmbus_acpi_add,
2620                 .remove = vmbus_acpi_remove,
2621         },
2622         .drv.pm = &vmbus_bus_pm,
2623 };
2624
2625 static void hv_kexec_handler(void)
2626 {
2627         hv_stimer_global_cleanup();
2628         vmbus_initiate_unload(false);
2629         /* Make sure conn_state is set as hv_synic_cleanup checks for it */
2630         mb();
2631         cpuhp_remove_state(hyperv_cpuhp_online);
2632 };
2633
2634 static void hv_crash_handler(struct pt_regs *regs)
2635 {
2636         int cpu;
2637
2638         vmbus_initiate_unload(true);
2639         /*
2640          * In crash handler we can't schedule synic cleanup for all CPUs,
2641          * doing the cleanup for current CPU only. This should be sufficient
2642          * for kdump.
2643          */
2644         cpu = smp_processor_id();
2645         hv_stimer_cleanup(cpu);
2646         hv_synic_disable_regs(cpu);
2647 };
2648
2649 static int hv_synic_suspend(void)
2650 {
2651         /*
2652          * When we reach here, all the non-boot CPUs have been offlined.
2653          * If we're in a legacy configuration where stimer Direct Mode is
2654          * not enabled, the stimers on the non-boot CPUs have been unbound
2655          * in hv_synic_cleanup() -> hv_stimer_legacy_cleanup() ->
2656          * hv_stimer_cleanup() -> clockevents_unbind_device().
2657          *
2658          * hv_synic_suspend() only runs on CPU0 with interrupts disabled.
2659          * Here we do not call hv_stimer_legacy_cleanup() on CPU0 because:
2660          * 1) it's unnecessary as interrupts remain disabled between
2661          * syscore_suspend() and syscore_resume(): see create_image() and
2662          * resume_target_kernel()
2663          * 2) the stimer on CPU0 is automatically disabled later by
2664          * syscore_suspend() -> timekeeping_suspend() -> tick_suspend() -> ...
2665          * -> clockevents_shutdown() -> ... -> hv_ce_shutdown()
2666          * 3) a warning would be triggered if we call
2667          * clockevents_unbind_device(), which may sleep, in an
2668          * interrupts-disabled context.
2669          */
2670
2671         hv_synic_disable_regs(0);
2672
2673         return 0;
2674 }
2675
2676 static void hv_synic_resume(void)
2677 {
2678         hv_synic_enable_regs(0);
2679
2680         /*
2681          * Note: we don't need to call hv_stimer_init(0), because the timer
2682          * on CPU0 is not unbound in hv_synic_suspend(), and the timer is
2683          * automatically re-enabled in timekeeping_resume().
2684          */
2685 }
2686
2687 /* The callbacks run only on CPU0, with irqs_disabled. */
2688 static struct syscore_ops hv_synic_syscore_ops = {
2689         .suspend = hv_synic_suspend,
2690         .resume = hv_synic_resume,
2691 };
2692
2693 static int __init hv_acpi_init(void)
2694 {
2695         int ret, t;
2696
2697         if (!hv_is_hyperv_initialized())
2698                 return -ENODEV;
2699
2700         if (hv_root_partition)
2701                 return 0;
2702
2703         init_completion(&probe_event);
2704
2705         /*
2706          * Get ACPI resources first.
2707          */
2708         ret = acpi_bus_register_driver(&vmbus_acpi_driver);
2709
2710         if (ret)
2711                 return ret;
2712
2713         t = wait_for_completion_timeout(&probe_event, 5*HZ);
2714         if (t == 0) {
2715                 ret = -ETIMEDOUT;
2716                 goto cleanup;
2717         }
2718
2719         /*
2720          * If we're on an architecture with a hardcoded hypervisor
2721          * vector (i.e. x86/x64), override the VMbus interrupt found
2722          * in the ACPI tables. Ensure vmbus_irq is not set since the
2723          * normal Linux IRQ mechanism is not used in this case.
2724          */
2725 #ifdef HYPERVISOR_CALLBACK_VECTOR
2726         vmbus_interrupt = HYPERVISOR_CALLBACK_VECTOR;
2727         vmbus_irq = -1;
2728 #endif
2729
2730         hv_debug_init();
2731
2732         ret = vmbus_bus_init();
2733         if (ret)
2734                 goto cleanup;
2735
2736         hv_setup_kexec_handler(hv_kexec_handler);
2737         hv_setup_crash_handler(hv_crash_handler);
2738
2739         register_syscore_ops(&hv_synic_syscore_ops);
2740
2741         return 0;
2742
2743 cleanup:
2744         acpi_bus_unregister_driver(&vmbus_acpi_driver);
2745         hv_acpi_dev = NULL;
2746         return ret;
2747 }
2748
2749 static void __exit vmbus_exit(void)
2750 {
2751         int cpu;
2752
2753         unregister_syscore_ops(&hv_synic_syscore_ops);
2754
2755         hv_remove_kexec_handler();
2756         hv_remove_crash_handler();
2757         vmbus_connection.conn_state = DISCONNECTED;
2758         hv_stimer_global_cleanup();
2759         vmbus_disconnect();
2760         if (vmbus_irq == -1) {
2761                 hv_remove_vmbus_handler();
2762         } else {
2763                 free_percpu_irq(vmbus_irq, vmbus_evt);
2764                 free_percpu(vmbus_evt);
2765         }
2766         for_each_online_cpu(cpu) {
2767                 struct hv_per_cpu_context *hv_cpu
2768                         = per_cpu_ptr(hv_context.cpu_context, cpu);
2769
2770                 tasklet_kill(&hv_cpu->msg_dpc);
2771         }
2772         hv_debug_rm_all_dir();
2773
2774         vmbus_free_channels();
2775         kfree(vmbus_connection.channels);
2776
2777         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
2778                 kmsg_dump_unregister(&hv_kmsg_dumper);
2779                 unregister_die_notifier(&hyperv_die_block);
2780                 atomic_notifier_chain_unregister(&panic_notifier_list,
2781                                                  &hyperv_panic_block);
2782         }
2783
2784         free_page((unsigned long)hv_panic_page);
2785         unregister_sysctl_table(hv_ctl_table_hdr);
2786         hv_ctl_table_hdr = NULL;
2787         bus_unregister(&hv_bus);
2788
2789         cpuhp_remove_state(hyperv_cpuhp_online);
2790         hv_synic_free();
2791         acpi_bus_unregister_driver(&vmbus_acpi_driver);
2792 }
2793
2794
2795 MODULE_LICENSE("GPL");
2796 MODULE_DESCRIPTION("Microsoft Hyper-V VMBus Driver");
2797
2798 subsys_initcall(hv_acpi_init);
2799 module_exit(vmbus_exit);