drivers/hv/vmbus_drv.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (c) 2009, Microsoft Corporation.
   4  *
   5  * Authors:
   6  *   Haiyang Zhang <haiyangz@microsoft.com>
   7  *   Hank Janssen  <hjanssen@microsoft.com>
   8  *   K. Y. Srinivasan <kys@microsoft.com>
   9  */
  10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  11
  12 #include <linux/init.h>
  13 #include <linux/module.h>
  14 #include <linux/device.h>
  15 #include <linux/interrupt.h>
  16 #include <linux/sysctl.h>
  17 #include <linux/slab.h>
  18 #include <linux/acpi.h>
  19 #include <linux/completion.h>
  20 #include <linux/hyperv.h>
  21 #include <linux/kernel_stat.h>
  22 #include <linux/clockchips.h>
  23 #include <linux/cpu.h>
  24 #include <linux/sched/task_stack.h>
  25
  26 #include <linux/delay.h>
  27 #include <linux/notifier.h>
  28 #include <linux/ptrace.h>
  29 #include <linux/screen_info.h>
  30 #include <linux/kdebug.h>
  31 #include <linux/efi.h>
  32 #include <linux/random.h>
  33 #include <linux/kernel.h>
  34 #include <linux/syscore_ops.h>
  35 #include <clocksource/hyperv_timer.h>
  36 #include "hyperv_vmbus.h"
  37
  38 struct vmbus_dynid {
  39         struct list_head node;
  40         struct hv_vmbus_device_id id;
  41 };
  42
  43 static struct acpi_device  *hv_acpi_dev;
  44
  45 static struct completion probe_event;
  46
  47 static int hyperv_cpuhp_online;
  48
  49 static void *hv_panic_page;
  50
  51 /* Values parsed from ACPI DSDT */
  52 static int vmbus_irq;
  53 int vmbus_interrupt;
  54
  55 /*
  56  * Boolean to control whether to report panic messages over Hyper-V.
  57  *
  58  * It can be set via /proc/sys/kernel/hyperv_record_panic_msg
  59  */
  60 static int sysctl_record_panic_msg = 1;
  61
  62 static int hyperv_report_reg(void)
  63 {
  64         return !sysctl_record_panic_msg || !hv_panic_page;
  65 }
  66
  67 static int hyperv_panic_event(struct notifier_block *nb, unsigned long val,
  68                               void *args)
  69 {
  70         struct pt_regs *regs;
  71
  72         vmbus_initiate_unload(true);
  73
  74         /*
  75          * Hyper-V should be notified only once about a panic.  If we will be
  76          * doing hyperv_report_panic_msg() later with kmsg data, don't do
  77          * the notification here.
  78          */
  79         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE
  80             && hyperv_report_reg()) {
  81                 regs = current_pt_regs();
  82                 hyperv_report_panic(regs, val, false);
  83         }
  84         return NOTIFY_DONE;
  85 }
  86
  87 static int hyperv_die_event(struct notifier_block *nb, unsigned long val,
  88                             void *args)
  89 {
  90         struct die_args *die = args;
  91         struct pt_regs *regs = die->regs;
  92
  93         /* Don't notify Hyper-V if the die event is other than oops */
  94         if (val != DIE_OOPS)
  95                 return NOTIFY_DONE;
  96
  97         /*
  98          * Hyper-V should be notified only once about a panic.  If we will be
  99          * doing hyperv_report_panic_msg() later with kmsg data, don't do
 100          * the notification here.
 101          */
 102         if (hyperv_report_reg())
 103                 hyperv_report_panic(regs, val, true);
 104         return NOTIFY_DONE;
 105 }
 106
 107 static struct notifier_block hyperv_die_block = {
 108         .notifier_call = hyperv_die_event,
 109 };
 110 static struct notifier_block hyperv_panic_block = {
 111         .notifier_call = hyperv_panic_event,
 112 };
 113
 114 static const char *fb_mmio_name = "fb_range";
 115 static struct resource *fb_mmio;
 116 static struct resource *hyperv_mmio;
 117 static DEFINE_MUTEX(hyperv_mmio_lock);
 118
 119 static int vmbus_exists(void)
 120 {
 121         if (hv_acpi_dev == NULL)
 122                 return -ENODEV;
 123
 124         return 0;
 125 }
 126
 127 static u8 channel_monitor_group(const struct vmbus_channel *channel)
 128 {
 129         return (u8)channel->offermsg.monitorid / 32;
 130 }
 131
 132 static u8 channel_monitor_offset(const struct vmbus_channel *channel)
 133 {
 134         return (u8)channel->offermsg.monitorid % 32;
 135 }
 136
 137 static u32 channel_pending(const struct vmbus_channel *channel,
 138                            const struct hv_monitor_page *monitor_page)
 139 {
 140         u8 monitor_group = channel_monitor_group(channel);
 141
 142         return monitor_page->trigger_group[monitor_group].pending;
 143 }
 144
 145 static u32 channel_latency(const struct vmbus_channel *channel,
 146                            const struct hv_monitor_page *monitor_page)
 147 {
 148         u8 monitor_group = channel_monitor_group(channel);
 149         u8 monitor_offset = channel_monitor_offset(channel);
 150
 151         return monitor_page->latency[monitor_group][monitor_offset];
 152 }
 153
 154 static u32 channel_conn_id(struct vmbus_channel *channel,
 155                            struct hv_monitor_page *monitor_page)
 156 {
 157         u8 monitor_group = channel_monitor_group(channel);
 158         u8 monitor_offset = channel_monitor_offset(channel);
 159
 160         return monitor_page->parameter[monitor_group][monitor_offset].connectionid.u.id;
 161 }
 162
 163 static ssize_t id_show(struct device *dev, struct device_attribute *dev_attr,
 164                        char *buf)
 165 {
 166         struct hv_device *hv_dev = device_to_hv_device(dev);
 167
 168         if (!hv_dev->channel)
 169                 return -ENODEV;
 170         return sprintf(buf, "%d\n", hv_dev->channel->offermsg.child_relid);
 171 }
 172 static DEVICE_ATTR_RO(id);
 173
 174 static ssize_t state_show(struct device *dev, struct device_attribute *dev_attr,
 175                           char *buf)
 176 {
 177         struct hv_device *hv_dev = device_to_hv_device(dev);
 178
 179         if (!hv_dev->channel)
 180                 return -ENODEV;
 181         return sprintf(buf, "%d\n", hv_dev->channel->state);
 182 }
 183 static DEVICE_ATTR_RO(state);
 184
 185 static ssize_t monitor_id_show(struct device *dev,
 186                                struct device_attribute *dev_attr, char *buf)
 187 {
 188         struct hv_device *hv_dev = device_to_hv_device(dev);
 189
 190         if (!hv_dev->channel)
 191                 return -ENODEV;
 192         return sprintf(buf, "%d\n", hv_dev->channel->offermsg.monitorid);
 193 }
 194 static DEVICE_ATTR_RO(monitor_id);
 195
 196 static ssize_t class_id_show(struct device *dev,
 197                                struct device_attribute *dev_attr, char *buf)
 198 {
 199         struct hv_device *hv_dev = device_to_hv_device(dev);
 200
 201         if (!hv_dev->channel)
 202                 return -ENODEV;
 203         return sprintf(buf, "{%pUl}\n",
 204                        &hv_dev->channel->offermsg.offer.if_type);
 205 }
 206 static DEVICE_ATTR_RO(class_id);
 207
 208 static ssize_t device_id_show(struct device *dev,
 209                               struct device_attribute *dev_attr, char *buf)
 210 {
 211         struct hv_device *hv_dev = device_to_hv_device(dev);
 212
 213         if (!hv_dev->channel)
 214                 return -ENODEV;
 215         return sprintf(buf, "{%pUl}\n",
 216                        &hv_dev->channel->offermsg.offer.if_instance);
 217 }
 218 static DEVICE_ATTR_RO(device_id);
 219
 220 static ssize_t modalias_show(struct device *dev,
 221                              struct device_attribute *dev_attr, char *buf)
 222 {
 223         struct hv_device *hv_dev = device_to_hv_device(dev);
 224
 225         return sprintf(buf, "vmbus:%*phN\n", UUID_SIZE, &hv_dev->dev_type);
 226 }
 227 static DEVICE_ATTR_RO(modalias);
 228
 229 #ifdef CONFIG_NUMA
 230 static ssize_t numa_node_show(struct device *dev,
 231                               struct device_attribute *attr, char *buf)
 232 {
 233         struct hv_device *hv_dev = device_to_hv_device(dev);
 234
 235         if (!hv_dev->channel)
 236                 return -ENODEV;
 237
 238         return sprintf(buf, "%d\n", cpu_to_node(hv_dev->channel->target_cpu));
 239 }
 240 static DEVICE_ATTR_RO(numa_node);
 241 #endif
 242
 243 static ssize_t server_monitor_pending_show(struct device *dev,
 244                                            struct device_attribute *dev_attr,
 245                                            char *buf)
 246 {
 247         struct hv_device *hv_dev = device_to_hv_device(dev);
 248
 249         if (!hv_dev->channel)
 250                 return -ENODEV;
 251         return sprintf(buf, "%d\n",
 252                        channel_pending(hv_dev->channel,
 253                                        vmbus_connection.monitor_pages[0]));
 254 }
 255 static DEVICE_ATTR_RO(server_monitor_pending);
 256
 257 static ssize_t client_monitor_pending_show(struct device *dev,
 258                                            struct device_attribute *dev_attr,
 259                                            char *buf)
 260 {
 261         struct hv_device *hv_dev = device_to_hv_device(dev);
 262
 263         if (!hv_dev->channel)
 264                 return -ENODEV;
 265         return sprintf(buf, "%d\n",
 266                        channel_pending(hv_dev->channel,
 267                                        vmbus_connection.monitor_pages[1]));
 268 }
 269 static DEVICE_ATTR_RO(client_monitor_pending);
 270
 271 static ssize_t server_monitor_latency_show(struct device *dev,
 272                                            struct device_attribute *dev_attr,
 273                                            char *buf)
 274 {
 275         struct hv_device *hv_dev = device_to_hv_device(dev);
 276
 277         if (!hv_dev->channel)
 278                 return -ENODEV;
 279         return sprintf(buf, "%d\n",
 280                        channel_latency(hv_dev->channel,
 281                                        vmbus_connection.monitor_pages[0]));
 282 }
 283 static DEVICE_ATTR_RO(server_monitor_latency);
 284
 285 static ssize_t client_monitor_latency_show(struct device *dev,
 286                                            struct device_attribute *dev_attr,
 287                                            char *buf)
 288 {
 289         struct hv_device *hv_dev = device_to_hv_device(dev);
 290
 291         if (!hv_dev->channel)
 292                 return -ENODEV;
 293         return sprintf(buf, "%d\n",
 294                        channel_latency(hv_dev->channel,
 295                                        vmbus_connection.monitor_pages[1]));
 296 }
 297 static DEVICE_ATTR_RO(client_monitor_latency);
 298
 299 static ssize_t server_monitor_conn_id_show(struct device *dev,
 300                                            struct device_attribute *dev_attr,
 301                                            char *buf)
 302 {
 303         struct hv_device *hv_dev = device_to_hv_device(dev);
 304
 305         if (!hv_dev->channel)
 306                 return -ENODEV;
 307         return sprintf(buf, "%d\n",
 308                        channel_conn_id(hv_dev->channel,
 309                                        vmbus_connection.monitor_pages[0]));
 310 }
 311 static DEVICE_ATTR_RO(server_monitor_conn_id);
 312
 313 static ssize_t client_monitor_conn_id_show(struct device *dev,
 314                                            struct device_attribute *dev_attr,
 315                                            char *buf)
 316 {
 317         struct hv_device *hv_dev = device_to_hv_device(dev);
 318
 319         if (!hv_dev->channel)
 320                 return -ENODEV;
 321         return sprintf(buf, "%d\n",
 322                        channel_conn_id(hv_dev->channel,
 323                                        vmbus_connection.monitor_pages[1]));
 324 }
 325 static DEVICE_ATTR_RO(client_monitor_conn_id);
 326
 327 static ssize_t out_intr_mask_show(struct device *dev,
 328                                   struct device_attribute *dev_attr, char *buf)
 329 {
 330         struct hv_device *hv_dev = device_to_hv_device(dev);
 331         struct hv_ring_buffer_debug_info outbound;
 332         int ret;
 333
 334         if (!hv_dev->channel)
 335                 return -ENODEV;
 336
 337         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 338                                           &outbound);
 339         if (ret < 0)
 340                 return ret;
 341
 342         return sprintf(buf, "%d\n", outbound.current_interrupt_mask);
 343 }
 344 static DEVICE_ATTR_RO(out_intr_mask);
 345
 346 static ssize_t out_read_index_show(struct device *dev,
 347                                    struct device_attribute *dev_attr, char *buf)
 348 {
 349         struct hv_device *hv_dev = device_to_hv_device(dev);
 350         struct hv_ring_buffer_debug_info outbound;
 351         int ret;
 352
 353         if (!hv_dev->channel)
 354                 return -ENODEV;
 355
 356         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 357                                           &outbound);
 358         if (ret < 0)
 359                 return ret;
 360         return sprintf(buf, "%d\n", outbound.current_read_index);
 361 }
 362 static DEVICE_ATTR_RO(out_read_index);
 363
 364 static ssize_t out_write_index_show(struct device *dev,
 365                                     struct device_attribute *dev_attr,
 366                                     char *buf)
 367 {
 368         struct hv_device *hv_dev = device_to_hv_device(dev);
 369         struct hv_ring_buffer_debug_info outbound;
 370         int ret;
 371
 372         if (!hv_dev->channel)
 373                 return -ENODEV;
 374
 375         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 376                                           &outbound);
 377         if (ret < 0)
 378                 return ret;
 379         return sprintf(buf, "%d\n", outbound.current_write_index);
 380 }
 381 static DEVICE_ATTR_RO(out_write_index);
 382
 383 static ssize_t out_read_bytes_avail_show(struct device *dev,
 384                                          struct device_attribute *dev_attr,
 385                                          char *buf)
 386 {
 387         struct hv_device *hv_dev = device_to_hv_device(dev);
 388         struct hv_ring_buffer_debug_info outbound;
 389         int ret;
 390
 391         if (!hv_dev->channel)
 392                 return -ENODEV;
 393
 394         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 395                                           &outbound);
 396         if (ret < 0)
 397                 return ret;
 398         return sprintf(buf, "%d\n", outbound.bytes_avail_toread);
 399 }
 400 static DEVICE_ATTR_RO(out_read_bytes_avail);
 401
 402 static ssize_t out_write_bytes_avail_show(struct device *dev,
 403                                           struct device_attribute *dev_attr,
 404                                           char *buf)
 405 {
 406         struct hv_device *hv_dev = device_to_hv_device(dev);
 407         struct hv_ring_buffer_debug_info outbound;
 408         int ret;
 409
 410         if (!hv_dev->channel)
 411                 return -ENODEV;
 412
 413         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 414                                           &outbound);
 415         if (ret < 0)
 416                 return ret;
 417         return sprintf(buf, "%d\n", outbound.bytes_avail_towrite);
 418 }
 419 static DEVICE_ATTR_RO(out_write_bytes_avail);
 420
 421 static ssize_t in_intr_mask_show(struct device *dev,
 422                                  struct device_attribute *dev_attr, char *buf)
 423 {
 424         struct hv_device *hv_dev = device_to_hv_device(dev);
 425         struct hv_ring_buffer_debug_info inbound;
 426         int ret;
 427
 428         if (!hv_dev->channel)
 429                 return -ENODEV;
 430
 431         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 432         if (ret < 0)
 433                 return ret;
 434
 435         return sprintf(buf, "%d\n", inbound.current_interrupt_mask);
 436 }
 437 static DEVICE_ATTR_RO(in_intr_mask);
 438
 439 static ssize_t in_read_index_show(struct device *dev,
 440                                   struct device_attribute *dev_attr, char *buf)
 441 {
 442         struct hv_device *hv_dev = device_to_hv_device(dev);
 443         struct hv_ring_buffer_debug_info inbound;
 444         int ret;
 445
 446         if (!hv_dev->channel)
 447                 return -ENODEV;
 448
 449         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 450         if (ret < 0)
 451                 return ret;
 452
 453         return sprintf(buf, "%d\n", inbound.current_read_index);
 454 }
 455 static DEVICE_ATTR_RO(in_read_index);
 456
 457 static ssize_t in_write_index_show(struct device *dev,
 458                                    struct device_attribute *dev_attr, char *buf)
 459 {
 460         struct hv_device *hv_dev = device_to_hv_device(dev);
 461         struct hv_ring_buffer_debug_info inbound;
 462         int ret;
 463
 464         if (!hv_dev->channel)
 465                 return -ENODEV;
 466
 467         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 468         if (ret < 0)
 469                 return ret;
 470
 471         return sprintf(buf, "%d\n", inbound.current_write_index);
 472 }
 473 static DEVICE_ATTR_RO(in_write_index);
 474
 475 static ssize_t in_read_bytes_avail_show(struct device *dev,
 476                                         struct device_attribute *dev_attr,
 477                                         char *buf)
 478 {
 479         struct hv_device *hv_dev = device_to_hv_device(dev);
 480         struct hv_ring_buffer_debug_info inbound;
 481         int ret;
 482
 483         if (!hv_dev->channel)
 484                 return -ENODEV;
 485
 486         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 487         if (ret < 0)
 488                 return ret;
 489
 490         return sprintf(buf, "%d\n", inbound.bytes_avail_toread);
 491 }
 492 static DEVICE_ATTR_RO(in_read_bytes_avail);
 493
 494 static ssize_t in_write_bytes_avail_show(struct device *dev,
 495                                          struct device_attribute *dev_attr,
 496                                          char *buf)
 497 {
 498         struct hv_device *hv_dev = device_to_hv_device(dev);
 499         struct hv_ring_buffer_debug_info inbound;
 500         int ret;
 501
 502         if (!hv_dev->channel)
 503                 return -ENODEV;
 504
 505         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 506         if (ret < 0)
 507                 return ret;
 508
 509         return sprintf(buf, "%d\n", inbound.bytes_avail_towrite);
 510 }
 511 static DEVICE_ATTR_RO(in_write_bytes_avail);
 512
 513 static ssize_t channel_vp_mapping_show(struct device *dev,
 514                                        struct device_attribute *dev_attr,
 515                                        char *buf)
 516 {
 517         struct hv_device *hv_dev = device_to_hv_device(dev);
 518         struct vmbus_channel *channel = hv_dev->channel, *cur_sc;
 519         int buf_size = PAGE_SIZE, n_written, tot_written;
 520         struct list_head *cur;
 521
 522         if (!channel)
 523                 return -ENODEV;
 524
 525         mutex_lock(&vmbus_connection.channel_mutex);
 526
 527         tot_written = snprintf(buf, buf_size, "%u:%u\n",
 528                 channel->offermsg.child_relid, channel->target_cpu);
 529
 530         list_for_each(cur, &channel->sc_list) {
 531                 if (tot_written >= buf_size - 1)
 532                         break;
 533
 534                 cur_sc = list_entry(cur, struct vmbus_channel, sc_list);
 535                 n_written = scnprintf(buf + tot_written,
 536                                      buf_size - tot_written,
 537                                      "%u:%u\n",
 538                                      cur_sc->offermsg.child_relid,
 539                                      cur_sc->target_cpu);
 540                 tot_written += n_written;
 541         }
 542
 543         mutex_unlock(&vmbus_connection.channel_mutex);
 544
 545         return tot_written;
 546 }
 547 static DEVICE_ATTR_RO(channel_vp_mapping);
 548
 549 static ssize_t vendor_show(struct device *dev,
 550                            struct device_attribute *dev_attr,
 551                            char *buf)
 552 {
 553         struct hv_device *hv_dev = device_to_hv_device(dev);
 554
 555         return sprintf(buf, "0x%x\n", hv_dev->vendor_id);
 556 }
 557 static DEVICE_ATTR_RO(vendor);
 558
 559 static ssize_t device_show(struct device *dev,
 560                            struct device_attribute *dev_attr,
 561                            char *buf)
 562 {
 563         struct hv_device *hv_dev = device_to_hv_device(dev);
 564
 565         return sprintf(buf, "0x%x\n", hv_dev->device_id);
 566 }
 567 static DEVICE_ATTR_RO(device);
 568
 569 static ssize_t driver_override_store(struct device *dev,
 570                                      struct device_attribute *attr,
 571                                      const char *buf, size_t count)
 572 {
 573         struct hv_device *hv_dev = device_to_hv_device(dev);
 574         char *driver_override, *old, *cp;
 575
 576         /* We need to keep extra room for a newline */
 577         if (count >= (PAGE_SIZE - 1))
 578                 return -EINVAL;
 579
 580         driver_override = kstrndup(buf, count, GFP_KERNEL);
 581         if (!driver_override)
 582                 return -ENOMEM;
 583
 584         cp = strchr(driver_override, '\n');
 585         if (cp)
 586                 *cp = '\0';
 587
 588         device_lock(dev);
 589         old = hv_dev->driver_override;
 590         if (strlen(driver_override)) {
 591                 hv_dev->driver_override = driver_override;
 592         } else {
 593                 kfree(driver_override);
 594                 hv_dev->driver_override = NULL;
 595         }
 596         device_unlock(dev);
 597
 598         kfree(old);
 599
 600         return count;
 601 }
 602
 603 static ssize_t driver_override_show(struct device *dev,
 604                                     struct device_attribute *attr, char *buf)
 605 {
 606         struct hv_device *hv_dev = device_to_hv_device(dev);
 607         ssize_t len;
 608
 609         device_lock(dev);
 610         len = snprintf(buf, PAGE_SIZE, "%s\n", hv_dev->driver_override);
 611         device_unlock(dev);
 612
 613         return len;
 614 }
 615 static DEVICE_ATTR_RW(driver_override);
 616
 617 /* Set up per device attributes in /sys/bus/vmbus/devices/<bus device> */
 618 static struct attribute *vmbus_dev_attrs[] = {
 619         &dev_attr_id.attr,
 620         &dev_attr_state.attr,
 621         &dev_attr_monitor_id.attr,
 622         &dev_attr_class_id.attr,
 623         &dev_attr_device_id.attr,
 624         &dev_attr_modalias.attr,
 625 #ifdef CONFIG_NUMA
 626         &dev_attr_numa_node.attr,
 627 #endif
 628         &dev_attr_server_monitor_pending.attr,
 629         &dev_attr_client_monitor_pending.attr,
 630         &dev_attr_server_monitor_latency.attr,
 631         &dev_attr_client_monitor_latency.attr,
 632         &dev_attr_server_monitor_conn_id.attr,
 633         &dev_attr_client_monitor_conn_id.attr,
 634         &dev_attr_out_intr_mask.attr,
 635         &dev_attr_out_read_index.attr,
 636         &dev_attr_out_write_index.attr,
 637         &dev_attr_out_read_bytes_avail.attr,
 638         &dev_attr_out_write_bytes_avail.attr,
 639         &dev_attr_in_intr_mask.attr,
 640         &dev_attr_in_read_index.attr,
 641         &dev_attr_in_write_index.attr,
 642         &dev_attr_in_read_bytes_avail.attr,
 643         &dev_attr_in_write_bytes_avail.attr,
 644         &dev_attr_channel_vp_mapping.attr,
 645         &dev_attr_vendor.attr,
 646         &dev_attr_device.attr,
 647         &dev_attr_driver_override.attr,
 648         NULL,
 649 };
 650
 651 /*
 652  * Device-level attribute_group callback function. Returns the permission for
 653  * each attribute, and returns 0 if an attribute is not visible.
 654  */
 655 static umode_t vmbus_dev_attr_is_visible(struct kobject *kobj,
 656                                          struct attribute *attr, int idx)
 657 {
 658         struct device *dev = kobj_to_dev(kobj);
 659         const struct hv_device *hv_dev = device_to_hv_device(dev);
 660
 661         /* Hide the monitor attributes if the monitor mechanism is not used. */
 662         if (!hv_dev->channel->offermsg.monitor_allocated &&
 663             (attr == &dev_attr_monitor_id.attr ||
 664              attr == &dev_attr_server_monitor_pending.attr ||
 665              attr == &dev_attr_client_monitor_pending.attr ||
 666              attr == &dev_attr_server_monitor_latency.attr ||
 667              attr == &dev_attr_client_monitor_latency.attr ||
 668              attr == &dev_attr_server_monitor_conn_id.attr ||
 669              attr == &dev_attr_client_monitor_conn_id.attr))
 670                 return 0;
 671
 672         return attr->mode;
 673 }
 674
 675 static const struct attribute_group vmbus_dev_group = {
 676         .attrs = vmbus_dev_attrs,
 677         .is_visible = vmbus_dev_attr_is_visible
 678 };
 679 __ATTRIBUTE_GROUPS(vmbus_dev);
 680
 681 /* Set up the attribute for /sys/bus/vmbus/hibernation */
 682 static ssize_t hibernation_show(struct bus_type *bus, char *buf)
 683 {
 684         return sprintf(buf, "%d\n", !!hv_is_hibernation_supported());
 685 }
 686
 687 static BUS_ATTR_RO(hibernation);
 688
 689 static struct attribute *vmbus_bus_attrs[] = {
 690         &bus_attr_hibernation.attr,
 691         NULL,
 692 };
 693 static const struct attribute_group vmbus_bus_group = {
 694         .attrs = vmbus_bus_attrs,
 695 };
 696 __ATTRIBUTE_GROUPS(vmbus_bus);
 697
 698 /*
 699  * vmbus_uevent - add uevent for our device
 700  *
 701  * This routine is invoked when a device is added or removed on the vmbus to
 702  * generate a uevent to udev in the userspace. The udev will then look at its
 703  * rule and the uevent generated here to load the appropriate driver
 704  *
 705  * The alias string will be of the form vmbus:guid where guid is the string
 706  * representation of the device guid (each byte of the guid will be
 707  * represented with two hex characters.
 708  */
 709 static int vmbus_uevent(struct device *device, struct kobj_uevent_env *env)
 710 {
 711         struct hv_device *dev = device_to_hv_device(device);
 712         const char *format = "MODALIAS=vmbus:%*phN";
 713
 714         return add_uevent_var(env, format, UUID_SIZE, &dev->dev_type);
 715 }
 716
 717 static const struct hv_vmbus_device_id *
 718 hv_vmbus_dev_match(const struct hv_vmbus_device_id *id, const guid_t *guid)
 719 {
 720         if (id == NULL)
 721                 return NULL; /* empty device table */
 722
 723         for (; !guid_is_null(&id->guid); id++)
 724                 if (guid_equal(&id->guid, guid))
 725                         return id;
 726
 727         return NULL;
 728 }
 729
 730 static const struct hv_vmbus_device_id *
 731 hv_vmbus_dynid_match(struct hv_driver *drv, const guid_t *guid)
 732 {
 733         const struct hv_vmbus_device_id *id = NULL;
 734         struct vmbus_dynid *dynid;
 735
 736         spin_lock(&drv->dynids.lock);
 737         list_for_each_entry(dynid, &drv->dynids.list, node) {
 738                 if (guid_equal(&dynid->id.guid, guid)) {
 739                         id = &dynid->id;
 740                         break;
 741                 }
 742         }
 743         spin_unlock(&drv->dynids.lock);
 744
 745         return id;
 746 }
 747
 748 static const struct hv_vmbus_device_id vmbus_device_null;
 749
 750 /*
 751  * Return a matching hv_vmbus_device_id pointer.
 752  * If there is no match, return NULL.
 753  */
 754 static const struct hv_vmbus_device_id *hv_vmbus_get_id(struct hv_driver *drv,
 755                                                         struct hv_device *dev)
 756 {
 757         const guid_t *guid = &dev->dev_type;
 758         const struct hv_vmbus_device_id *id;
 759
 760         /* When driver_override is set, only bind to the matching driver */
 761         if (dev->driver_override && strcmp(dev->driver_override, drv->name))
 762                 return NULL;
 763
 764         /* Look at the dynamic ids first, before the static ones */
 765         id = hv_vmbus_dynid_match(drv, guid);
 766         if (!id)
 767                 id = hv_vmbus_dev_match(drv->id_table, guid);
 768
 769         /* driver_override will always match, send a dummy id */
 770         if (!id && dev->driver_override)
 771                 id = &vmbus_device_null;
 772
 773         return id;
 774 }
 775
 776 /* vmbus_add_dynid - add a new device ID to this driver and re-probe devices */
 777 static int vmbus_add_dynid(struct hv_driver *drv, guid_t *guid)
 778 {
 779         struct vmbus_dynid *dynid;
 780
 781         dynid = kzalloc(sizeof(*dynid), GFP_KERNEL);
 782         if (!dynid)
 783                 return -ENOMEM;
 784
 785         dynid->id.guid = *guid;
 786
 787         spin_lock(&drv->dynids.lock);
 788         list_add_tail(&dynid->node, &drv->dynids.list);
 789         spin_unlock(&drv->dynids.lock);
 790
 791         return driver_attach(&drv->driver);
 792 }
 793
 794 static void vmbus_free_dynids(struct hv_driver *drv)
 795 {
 796         struct vmbus_dynid *dynid, *n;
 797
 798         spin_lock(&drv->dynids.lock);
 799         list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
 800                 list_del(&dynid->node);
 801                 kfree(dynid);
 802         }
 803         spin_unlock(&drv->dynids.lock);
 804 }
 805
 806 /*
 807  * store_new_id - sysfs frontend to vmbus_add_dynid()
 808  *
 809  * Allow GUIDs to be added to an existing driver via sysfs.
 810  */
 811 static ssize_t new_id_store(struct device_driver *driver, const char *buf,
 812                             size_t count)
 813 {
 814         struct hv_driver *drv = drv_to_hv_drv(driver);
 815         guid_t guid;
 816         ssize_t retval;
 817
 818         retval = guid_parse(buf, &guid);
 819         if (retval)
 820                 return retval;
 821
 822         if (hv_vmbus_dynid_match(drv, &guid))
 823                 return -EEXIST;
 824
 825         retval = vmbus_add_dynid(drv, &guid);
 826         if (retval)
 827                 return retval;
 828         return count;
 829 }
 830 static DRIVER_ATTR_WO(new_id);
 831
 832 /*
 833  * store_remove_id - remove a PCI device ID from this driver
 834  *
 835  * Removes a dynamic pci device ID to this driver.
 836  */
 837 static ssize_t remove_id_store(struct device_driver *driver, const char *buf,
 838                                size_t count)
 839 {
 840         struct hv_driver *drv = drv_to_hv_drv(driver);
 841         struct vmbus_dynid *dynid, *n;
 842         guid_t guid;
 843         ssize_t retval;
 844
 845         retval = guid_parse(buf, &guid);
 846         if (retval)
 847                 return retval;
 848
 849         retval = -ENODEV;
 850         spin_lock(&drv->dynids.lock);
 851         list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
 852                 struct hv_vmbus_device_id *id = &dynid->id;
 853
 854                 if (guid_equal(&id->guid, &guid)) {
 855                         list_del(&dynid->node);
 856                         kfree(dynid);
 857                         retval = count;
 858                         break;
 859                 }
 860         }
 861         spin_unlock(&drv->dynids.lock);
 862
 863         return retval;
 864 }
 865 static DRIVER_ATTR_WO(remove_id);
 866
 867 static struct attribute *vmbus_drv_attrs[] = {
 868         &driver_attr_new_id.attr,
 869         &driver_attr_remove_id.attr,
 870         NULL,
 871 };
 872 ATTRIBUTE_GROUPS(vmbus_drv);
 873
 874
 875 /*
 876  * vmbus_match - Attempt to match the specified device to the specified driver
 877  */
 878 static int vmbus_match(struct device *device, struct device_driver *driver)
 879 {
 880         struct hv_driver *drv = drv_to_hv_drv(driver);
 881         struct hv_device *hv_dev = device_to_hv_device(device);
 882
 883         /* The hv_sock driver handles all hv_sock offers. */
 884         if (is_hvsock_channel(hv_dev->channel))
 885                 return drv->hvsock;
 886
 887         if (hv_vmbus_get_id(drv, hv_dev))
 888                 return 1;
 889
 890         return 0;
 891 }
 892
 893 /*
 894  * vmbus_probe - Add the new vmbus's child device
 895  */
 896 static int vmbus_probe(struct device *child_device)
 897 {
 898         int ret = 0;
 899         struct hv_driver *drv =
 900                         drv_to_hv_drv(child_device->driver);
 901         struct hv_device *dev = device_to_hv_device(child_device);
 902         const struct hv_vmbus_device_id *dev_id;
 903
 904         dev_id = hv_vmbus_get_id(drv, dev);
 905         if (drv->probe) {
 906                 ret = drv->probe(dev, dev_id);
 907                 if (ret != 0)
 908                         pr_err("probe failed for device %s (%d)\n",
 909                                dev_name(child_device), ret);
 910
 911         } else {
 912                 pr_err("probe not set for driver %s\n",
 913                        dev_name(child_device));
 914                 ret = -ENODEV;
 915         }
 916         return ret;
 917 }
 918
 919 /*
 920  * vmbus_remove - Remove a vmbus device
 921  */
 922 static int vmbus_remove(struct device *child_device)
 923 {
 924         struct hv_driver *drv;
 925         struct hv_device *dev = device_to_hv_device(child_device);
 926
 927         if (child_device->driver) {
 928                 drv = drv_to_hv_drv(child_device->driver);
 929                 if (drv->remove)
 930                         drv->remove(dev);
 931         }
 932
 933         return 0;
 934 }
 935
 936
 937 /*
 938  * vmbus_shutdown - Shutdown a vmbus device
 939  */
 940 static void vmbus_shutdown(struct device *child_device)
 941 {
 942         struct hv_driver *drv;
 943         struct hv_device *dev = device_to_hv_device(child_device);
 944
 945
 946         /* The device may not be attached yet */
 947         if (!child_device->driver)
 948                 return;
 949
 950         drv = drv_to_hv_drv(child_device->driver);
 951
 952         if (drv->shutdown)
 953                 drv->shutdown(dev);
 954 }
 955
 956 #ifdef CONFIG_PM_SLEEP
 957 /*
 958  * vmbus_suspend - Suspend a vmbus device
 959  */
 960 static int vmbus_suspend(struct device *child_device)
 961 {
 962         struct hv_driver *drv;
 963         struct hv_device *dev = device_to_hv_device(child_device);
 964
 965         /* The device may not be attached yet */
 966         if (!child_device->driver)
 967                 return 0;
 968
 969         drv = drv_to_hv_drv(child_device->driver);
 970         if (!drv->suspend)
 971                 return -EOPNOTSUPP;
 972
 973         return drv->suspend(dev);
 974 }
 975
 976 /*
 977  * vmbus_resume - Resume a vmbus device
 978  */
 979 static int vmbus_resume(struct device *child_device)
 980 {
 981         struct hv_driver *drv;
 982         struct hv_device *dev = device_to_hv_device(child_device);
 983
 984         /* The device may not be attached yet */
 985         if (!child_device->driver)
 986                 return 0;
 987
 988         drv = drv_to_hv_drv(child_device->driver);
 989         if (!drv->resume)
 990                 return -EOPNOTSUPP;
 991
 992         return drv->resume(dev);
 993 }
 994 #else
 995 #define vmbus_suspend NULL
 996 #define vmbus_resume NULL
 997 #endif /* CONFIG_PM_SLEEP */
 998
 999 /*
1000  * vmbus_device_release - Final callback release of the vmbus child device
1001  */
1002 static void vmbus_device_release(struct device *device)
1003 {
1004         struct hv_device *hv_dev = device_to_hv_device(device);
1005         struct vmbus_channel *channel = hv_dev->channel;
1006
1007         hv_debug_rm_dev_dir(hv_dev);
1008
1009         mutex_lock(&vmbus_connection.channel_mutex);
1010         hv_process_channel_removal(channel);
1011         mutex_unlock(&vmbus_connection.channel_mutex);
1012         kfree(hv_dev);
1013 }
1014
1015 /*
1016  * Note: we must use the "noirq" ops: see the comment before vmbus_bus_pm.
1017  *
1018  * suspend_noirq/resume_noirq are set to NULL to support Suspend-to-Idle: we
1019  * shouldn't suspend the vmbus devices upon Suspend-to-Idle, otherwise there
1020  * is no way to wake up a Generation-2 VM.
1021  *
1022  * The other 4 ops are for hibernation.
1023  */
1024
1025 static const struct dev_pm_ops vmbus_pm = {
1026         .suspend_noirq  = NULL,
1027         .resume_noirq   = NULL,
1028         .freeze_noirq   = vmbus_suspend,
1029         .thaw_noirq     = vmbus_resume,
1030         .poweroff_noirq = vmbus_suspend,
1031         .restore_noirq  = vmbus_resume,
1032 };
1033
1034 /* The one and only one */
1035 static struct bus_type  hv_bus = {
1036         .name =         "vmbus",
1037         .match =                vmbus_match,
1038         .shutdown =             vmbus_shutdown,
1039         .remove =               vmbus_remove,
1040         .probe =                vmbus_probe,
1041         .uevent =               vmbus_uevent,
1042         .dev_groups =           vmbus_dev_groups,
1043         .drv_groups =           vmbus_drv_groups,
1044         .bus_groups =           vmbus_bus_groups,
1045         .pm =                   &vmbus_pm,
1046 };
1047
1048 struct onmessage_work_context {
1049         struct work_struct work;
1050         struct {
1051                 struct hv_message_header header;
1052                 u8 payload[];
1053         } msg;
1054 };
1055
1056 static void vmbus_onmessage_work(struct work_struct *work)
1057 {
1058         struct onmessage_work_context *ctx;
1059
1060         /* Do not process messages if we're in DISCONNECTED state */
1061         if (vmbus_connection.conn_state == DISCONNECTED)
1062                 return;
1063
1064         ctx = container_of(work, struct onmessage_work_context,
1065                            work);
1066         vmbus_onmessage((struct vmbus_channel_message_header *)
1067                         &ctx->msg.payload);
1068         kfree(ctx);
1069 }
1070
1071 void vmbus_on_msg_dpc(unsigned long data)
1072 {
1073         struct hv_per_cpu_context *hv_cpu = (void *)data;
1074         void *page_addr = hv_cpu->synic_message_page;
1075         struct hv_message msg_copy, *msg = (struct hv_message *)page_addr +
1076                                   VMBUS_MESSAGE_SINT;
1077         struct vmbus_channel_message_header *hdr;
1078         enum vmbus_channel_message_type msgtype;
1079         const struct vmbus_channel_message_table_entry *entry;
1080         struct onmessage_work_context *ctx;
1081         __u8 payload_size;
1082         u32 message_type;
1083
1084         /*
1085          * 'enum vmbus_channel_message_type' is supposed to always be 'u32' as
1086          * it is being used in 'struct vmbus_channel_message_header' definition
1087          * which is supposed to match hypervisor ABI.
1088          */
1089         BUILD_BUG_ON(sizeof(enum vmbus_channel_message_type) != sizeof(u32));
1090
1091         /*
1092          * Since the message is in memory shared with the host, an erroneous or
1093          * malicious Hyper-V could modify the message while vmbus_on_msg_dpc()
1094          * or individual message handlers are executing; to prevent this, copy
1095          * the message into private memory.
1096          */
1097         memcpy(&msg_copy, msg, sizeof(struct hv_message));
1098
1099         message_type = msg_copy.header.message_type;
1100         if (message_type == HVMSG_NONE)
1101                 /* no msg */
1102                 return;
1103
1104         hdr = (struct vmbus_channel_message_header *)msg_copy.u.payload;
1105         msgtype = hdr->msgtype;
1106
1107         trace_vmbus_on_msg_dpc(hdr);
1108
1109         if (msgtype >= CHANNELMSG_COUNT) {
1110                 WARN_ONCE(1, "unknown msgtype=%d\n", msgtype);
1111                 goto msg_handled;
1112         }
1113
1114         payload_size = msg_copy.header.payload_size;
1115         if (payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT) {
1116                 WARN_ONCE(1, "payload size is too large (%d)\n", payload_size);
1117                 goto msg_handled;
1118         }
1119
1120         entry = &channel_message_table[msgtype];
1121
1122         if (!entry->message_handler)
1123                 goto msg_handled;
1124
1125         if (payload_size < entry->min_payload_len) {
1126                 WARN_ONCE(1, "message too short: msgtype=%d len=%d\n", msgtype, payload_size);
1127                 goto msg_handled;
1128         }
1129
1130         if (entry->handler_type == VMHT_BLOCKING) {
1131                 ctx = kmalloc(sizeof(*ctx) + payload_size, GFP_ATOMIC);
1132                 if (ctx == NULL)
1133                         return;
1134
1135                 INIT_WORK(&ctx->work, vmbus_onmessage_work);
1136                 memcpy(&ctx->msg, &msg_copy, sizeof(msg->header) + payload_size);
1137
1138                 /*
1139                  * The host can generate a rescind message while we
1140                  * may still be handling the original offer. We deal with
1141                  * this condition by relying on the synchronization provided
1142                  * by offer_in_progress and by channel_mutex.  See also the
1143                  * inline comments in vmbus_onoffer_rescind().
1144                  */
1145                 switch (msgtype) {
1146                 case CHANNELMSG_RESCIND_CHANNELOFFER:
1147                         /*
1148                          * If we are handling the rescind message;
1149                          * schedule the work on the global work queue.
1150                          *
1151                          * The OFFER message and the RESCIND message should
1152                          * not be handled by the same serialized work queue,
1153                          * because the OFFER handler may call vmbus_open(),
1154                          * which tries to open the channel by sending an
1155                          * OPEN_CHANNEL message to the host and waits for
1156                          * the host's response; however, if the host has
1157                          * rescinded the channel before it receives the
1158                          * OPEN_CHANNEL message, the host just silently
1159                          * ignores the OPEN_CHANNEL message; as a result,
1160                          * the guest's OFFER handler hangs for ever, if we
1161                          * handle the RESCIND message in the same serialized
1162                          * work queue: the RESCIND handler can not start to
1163                          * run before the OFFER handler finishes.
1164                          */
1165                         schedule_work(&ctx->work);
1166                         break;
1167
1168                 case CHANNELMSG_OFFERCHANNEL:
1169                         /*
1170                          * The host sends the offer message of a given channel
1171                          * before sending the rescind message of the same
1172                          * channel.  These messages are sent to the guest's
1173                          * connect CPU; the guest then starts processing them
1174                          * in the tasklet handler on this CPU:
1175                          *
1176                          * VMBUS_CONNECT_CPU
1177                          *
1178                          * [vmbus_on_msg_dpc()]
1179                          * atomic_inc()  // CHANNELMSG_OFFERCHANNEL
1180                          * queue_work()
1181                          * ...
1182                          * [vmbus_on_msg_dpc()]
1183                          * schedule_work()  // CHANNELMSG_RESCIND_CHANNELOFFER
1184                          *
1185                          * We rely on the memory-ordering properties of the
1186                          * queue_work() and schedule_work() primitives, which
1187                          * guarantee that the atomic increment will be visible
1188                          * to the CPUs which will execute the offer & rescind
1189                          * works by the time these works will start execution.
1190                          */
1191                         atomic_inc(&vmbus_connection.offer_in_progress);
1192                         fallthrough;
1193
1194                 default:
1195                         queue_work(vmbus_connection.work_queue, &ctx->work);
1196                 }
1197         } else
1198                 entry->message_handler(hdr);
1199
1200 msg_handled:
1201         vmbus_signal_eom(msg, message_type);
1202 }
1203
1204 #ifdef CONFIG_PM_SLEEP
1205 /*
1206  * Fake RESCIND_CHANNEL messages to clean up hv_sock channels by force for
1207  * hibernation, because hv_sock connections can not persist across hibernation.
1208  */
1209 static void vmbus_force_channel_rescinded(struct vmbus_channel *channel)
1210 {
1211         struct onmessage_work_context *ctx;
1212         struct vmbus_channel_rescind_offer *rescind;
1213
1214         WARN_ON(!is_hvsock_channel(channel));
1215
1216         /*
1217          * Allocation size is small and the allocation should really not fail,
1218          * otherwise the state of the hv_sock connections ends up in limbo.
1219          */
1220         ctx = kzalloc(sizeof(*ctx) + sizeof(*rescind),
1221                       GFP_KERNEL | __GFP_NOFAIL);
1222
1223         /*
1224          * So far, these are not really used by Linux. Just set them to the
1225          * reasonable values conforming to the definitions of the fields.
1226          */
1227         ctx->msg.header.message_type = 1;
1228         ctx->msg.header.payload_size = sizeof(*rescind);
1229
1230         /* These values are actually used by Linux. */
1231         rescind = (struct vmbus_channel_rescind_offer *)ctx->msg.payload;
1232         rescind->header.msgtype = CHANNELMSG_RESCIND_CHANNELOFFER;
1233         rescind->child_relid = channel->offermsg.child_relid;
1234
1235         INIT_WORK(&ctx->work, vmbus_onmessage_work);
1236
1237         queue_work(vmbus_connection.work_queue, &ctx->work);
1238 }
1239 #endif /* CONFIG_PM_SLEEP */
1240
1241 /*
1242  * Schedule all channels with events pending
1243  */
1244 static void vmbus_chan_sched(struct hv_per_cpu_context *hv_cpu)
1245 {
1246         unsigned long *recv_int_page;
1247         u32 maxbits, relid;
1248
1249         if (vmbus_proto_version < VERSION_WIN8) {
1250                 maxbits = MAX_NUM_CHANNELS_SUPPORTED;
1251                 recv_int_page = vmbus_connection.recv_int_page;
1252         } else {
1253                 /*
1254                  * When the host is win8 and beyond, the event page
1255                  * can be directly checked to get the id of the channel
1256                  * that has the interrupt pending.
1257                  */
1258                 void *page_addr = hv_cpu->synic_event_page;
1259                 union hv_synic_event_flags *event
1260                         = (union hv_synic_event_flags *)page_addr +
1261                                                  VMBUS_MESSAGE_SINT;
1262
1263                 maxbits = HV_EVENT_FLAGS_COUNT;
1264                 recv_int_page = event->flags;
1265         }
1266
1267         if (unlikely(!recv_int_page))
1268                 return;
1269
1270         for_each_set_bit(relid, recv_int_page, maxbits) {
1271                 void (*callback_fn)(void *context);
1272                 struct vmbus_channel *channel;
1273
1274                 if (!sync_test_and_clear_bit(relid, recv_int_page))
1275                         continue;
1276
1277                 /* Special case - vmbus channel protocol msg */
1278                 if (relid == 0)
1279                         continue;
1280
1281                 /*
1282                  * Pairs with the kfree_rcu() in vmbus_chan_release().
1283                  * Guarantees that the channel data structure doesn't
1284                  * get freed while the channel pointer below is being
1285                  * dereferenced.
1286                  */
1287                 rcu_read_lock();
1288
1289                 /* Find channel based on relid */
1290                 channel = relid2channel(relid);
1291                 if (channel == NULL)
1292                         goto sched_unlock_rcu;
1293
1294                 if (channel->rescind)
1295                         goto sched_unlock_rcu;
1296
1297                 /*
1298                  * Make sure that the ring buffer data structure doesn't get
1299                  * freed while we dereference the ring buffer pointer.  Test
1300                  * for the channel's onchannel_callback being NULL within a
1301                  * sched_lock critical section.  See also the inline comments
1302                  * in vmbus_reset_channel_cb().
1303                  */
1304                 spin_lock(&channel->sched_lock);
1305
1306                 callback_fn = channel->onchannel_callback;
1307                 if (unlikely(callback_fn == NULL))
1308                         goto sched_unlock;
1309
1310                 trace_vmbus_chan_sched(channel);
1311
1312                 ++channel->interrupts;
1313
1314                 switch (channel->callback_mode) {
1315                 case HV_CALL_ISR:
1316                         (*callback_fn)(channel->channel_callback_context);
1317                         break;
1318
1319                 case HV_CALL_BATCHED:
1320                         hv_begin_read(&channel->inbound);
1321                         fallthrough;
1322                 case HV_CALL_DIRECT:
1323                         tasklet_schedule(&channel->callback_event);
1324                 }
1325
1326 sched_unlock:
1327                 spin_unlock(&channel->sched_lock);
1328 sched_unlock_rcu:
1329                 rcu_read_unlock();
1330         }
1331 }
1332
1333 static void vmbus_isr(void)
1334 {
1335         struct hv_per_cpu_context *hv_cpu
1336                 = this_cpu_ptr(hv_context.cpu_context);
1337         void *page_addr = hv_cpu->synic_event_page;
1338         struct hv_message *msg;
1339         union hv_synic_event_flags *event;
1340         bool handled = false;
1341
1342         if (unlikely(page_addr == NULL))
1343                 return;
1344
1345         event = (union hv_synic_event_flags *)page_addr +
1346                                          VMBUS_MESSAGE_SINT;
1347         /*
1348          * Check for events before checking for messages. This is the order
1349          * in which events and messages are checked in Windows guests on
1350          * Hyper-V, and the Windows team suggested we do the same.
1351          */
1352
1353         if ((vmbus_proto_version == VERSION_WS2008) ||
1354                 (vmbus_proto_version == VERSION_WIN7)) {
1355
1356                 /* Since we are a child, we only need to check bit 0 */
1357                 if (sync_test_and_clear_bit(0, event->flags))
1358                         handled = true;
1359         } else {
1360                 /*
1361                  * Our host is win8 or above. The signaling mechanism
1362                  * has changed and we can directly look at the event page.
1363                  * If bit n is set then we have an interrup on the channel
1364                  * whose id is n.
1365                  */
1366                 handled = true;
1367         }
1368
1369         if (handled)
1370                 vmbus_chan_sched(hv_cpu);
1371
1372         page_addr = hv_cpu->synic_message_page;
1373         msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
1374
1375         /* Check if there are actual msgs to be processed */
1376         if (msg->header.message_type != HVMSG_NONE) {
1377                 if (msg->header.message_type == HVMSG_TIMER_EXPIRED) {
1378                         hv_stimer0_isr();
1379                         vmbus_signal_eom(msg, HVMSG_TIMER_EXPIRED);
1380                 } else
1381                         tasklet_schedule(&hv_cpu->msg_dpc);
1382         }
1383
1384         add_interrupt_randomness(hv_get_vector(), 0);
1385 }
1386
1387 /*
1388  * Callback from kmsg_dump. Grab as much as possible from the end of the kmsg
1389  * buffer and call into Hyper-V to transfer the data.
1390  */
1391 static void hv_kmsg_dump(struct kmsg_dumper *dumper,
1392                          enum kmsg_dump_reason reason)
1393 {
1394         size_t bytes_written;
1395         phys_addr_t panic_pa;
1396
1397         /* We are only interested in panics. */
1398         if ((reason != KMSG_DUMP_PANIC) || (!sysctl_record_panic_msg))
1399                 return;
1400
1401         panic_pa = virt_to_phys(hv_panic_page);
1402
1403         /*
1404          * Write dump contents to the page. No need to synchronize; panic should
1405          * be single-threaded.
1406          */
1407         kmsg_dump_get_buffer(dumper, false, hv_panic_page, HV_HYP_PAGE_SIZE,
1408                              &bytes_written);
1409         if (bytes_written)
1410                 hyperv_report_panic_msg(panic_pa, bytes_written);
1411 }
1412
1413 static struct kmsg_dumper hv_kmsg_dumper = {
1414         .dump = hv_kmsg_dump,
1415 };
1416
1417 static void hv_kmsg_dump_register(void)
1418 {
1419         int ret;
1420
1421         hv_panic_page = hv_alloc_hyperv_zeroed_page();
1422         if (!hv_panic_page) {
1423                 pr_err("Hyper-V: panic message page memory allocation failed\n");
1424                 return;
1425         }
1426
1427         ret = kmsg_dump_register(&hv_kmsg_dumper);
1428         if (ret) {
1429                 pr_err("Hyper-V: kmsg dump register error 0x%x\n", ret);
1430                 hv_free_hyperv_page((unsigned long)hv_panic_page);
1431                 hv_panic_page = NULL;
1432         }
1433 }
1434
1435 static struct ctl_table_header *hv_ctl_table_hdr;
1436
1437 /*
1438  * sysctl option to allow the user to control whether kmsg data should be
1439  * reported to Hyper-V on panic.
1440  */
1441 static struct ctl_table hv_ctl_table[] = {
1442         {
1443                 .procname       = "hyperv_record_panic_msg",
1444                 .data           = &sysctl_record_panic_msg,
1445                 .maxlen         = sizeof(int),
1446                 .mode           = 0644,
1447                 .proc_handler   = proc_dointvec_minmax,
1448                 .extra1         = SYSCTL_ZERO,
1449                 .extra2         = SYSCTL_ONE
1450         },
1451         {}
1452 };
1453
1454 static struct ctl_table hv_root_table[] = {
1455         {
1456                 .procname       = "kernel",
1457                 .mode           = 0555,
1458                 .child          = hv_ctl_table
1459         },
1460         {}
1461 };
1462
1463 /*
1464  * vmbus_bus_init -Main vmbus driver initialization routine.
1465  *
1466  * Here, we
1467  *      - initialize the vmbus driver context
1468  *      - invoke the vmbus hv main init routine
1469  *      - retrieve the channel offers
1470  */
1471 static int vmbus_bus_init(void)
1472 {
1473         int ret;
1474
1475         ret = hv_init();
1476         if (ret != 0) {
1477                 pr_err("Unable to initialize the hypervisor - 0x%x\n", ret);
1478                 return ret;
1479         }
1480
1481         ret = bus_register(&hv_bus);
1482         if (ret)
1483                 return ret;
1484
1485         ret = hv_setup_vmbus_irq(vmbus_irq, vmbus_isr);
1486         if (ret)
1487                 goto err_setup;
1488
1489         ret = hv_synic_alloc();
1490         if (ret)
1491                 goto err_alloc;
1492
1493         /*
1494          * Initialize the per-cpu interrupt state and stimer state.
1495          * Then connect to the host.
1496          */
1497         ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "hyperv/vmbus:online",
1498                                 hv_synic_init, hv_synic_cleanup);
1499         if (ret < 0)
1500                 goto err_cpuhp;
1501         hyperv_cpuhp_online = ret;
1502
1503         ret = vmbus_connect();
1504         if (ret)
1505                 goto err_connect;
1506
1507         /*
1508          * Only register if the crash MSRs are available
1509          */
1510         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1511                 u64 hyperv_crash_ctl;
1512                 /*
1513                  * Sysctl registration is not fatal, since by default
1514                  * reporting is enabled.
1515                  */
1516                 hv_ctl_table_hdr = register_sysctl_table(hv_root_table);
1517                 if (!hv_ctl_table_hdr)
1518                         pr_err("Hyper-V: sysctl table register error");
1519
1520                 /*
1521                  * Register for panic kmsg callback only if the right
1522                  * capability is supported by the hypervisor.
1523                  */
1524                 hv_get_crash_ctl(hyperv_crash_ctl);
1525                 if (hyperv_crash_ctl & HV_CRASH_CTL_CRASH_NOTIFY_MSG)
1526                         hv_kmsg_dump_register();
1527
1528                 register_die_notifier(&hyperv_die_block);
1529         }
1530
1531         /*
1532          * Always register the panic notifier because we need to unload
1533          * the VMbus channel connection to prevent any VMbus
1534          * activity after the VM panics.
1535          */
1536         atomic_notifier_chain_register(&panic_notifier_list,
1537                                &hyperv_panic_block);
1538
1539         vmbus_request_offers();
1540
1541         return 0;
1542
1543 err_connect:
1544         cpuhp_remove_state(hyperv_cpuhp_online);
1545 err_cpuhp:
1546         hv_synic_free();
1547 err_alloc:
1548         hv_remove_vmbus_irq();
1549 err_setup:
1550         bus_unregister(&hv_bus);
1551         unregister_sysctl_table(hv_ctl_table_hdr);
1552         hv_ctl_table_hdr = NULL;
1553         return ret;
1554 }
1555
1556 /**
1557  * __vmbus_child_driver_register() - Register a vmbus's driver
1558  * @hv_driver: Pointer to driver structure you want to register
1559  * @owner: owner module of the drv
1560  * @mod_name: module name string
1561  *
1562  * Registers the given driver with Linux through the 'driver_register()' call
1563  * and sets up the hyper-v vmbus handling for this driver.
1564  * It will return the state of the 'driver_register()' call.
1565  *
1566  */
1567 int __vmbus_driver_register(struct hv_driver *hv_driver, struct module *owner, const char *mod_name)
1568 {
1569         int ret;
1570
1571         pr_info("registering driver %s\n", hv_driver->name);
1572
1573         ret = vmbus_exists();
1574         if (ret < 0)
1575                 return ret;
1576
1577         hv_driver->driver.name = hv_driver->name;
1578         hv_driver->driver.owner = owner;
1579         hv_driver->driver.mod_name = mod_name;
1580         hv_driver->driver.bus = &hv_bus;
1581
1582         spin_lock_init(&hv_driver->dynids.lock);
1583         INIT_LIST_HEAD(&hv_driver->dynids.list);
1584
1585         ret = driver_register(&hv_driver->driver);
1586
1587         return ret;
1588 }
1589 EXPORT_SYMBOL_GPL(__vmbus_driver_register);
1590
1591 /**
1592  * vmbus_driver_unregister() - Unregister a vmbus's driver
1593  * @hv_driver: Pointer to driver structure you want to
1594  *             un-register
1595  *
1596  * Un-register the given driver that was previous registered with a call to
1597  * vmbus_driver_register()
1598  */
1599 void vmbus_driver_unregister(struct hv_driver *hv_driver)
1600 {
1601         pr_info("unregistering driver %s\n", hv_driver->name);
1602
1603         if (!vmbus_exists()) {
1604                 driver_unregister(&hv_driver->driver);
1605                 vmbus_free_dynids(hv_driver);
1606         }
1607 }
1608 EXPORT_SYMBOL_GPL(vmbus_driver_unregister);
1609
1610
1611 /*
1612  * Called when last reference to channel is gone.
1613  */
1614 static void vmbus_chan_release(struct kobject *kobj)
1615 {
1616         struct vmbus_channel *channel
1617                 = container_of(kobj, struct vmbus_channel, kobj);
1618
1619         kfree_rcu(channel, rcu);
1620 }
1621
1622 struct vmbus_chan_attribute {
1623         struct attribute attr;
1624         ssize_t (*show)(struct vmbus_channel *chan, char *buf);
1625         ssize_t (*store)(struct vmbus_channel *chan,
1626                          const char *buf, size_t count);
1627 };
1628 #define VMBUS_CHAN_ATTR(_name, _mode, _show, _store) \
1629         struct vmbus_chan_attribute chan_attr_##_name \
1630                 = __ATTR(_name, _mode, _show, _store)
1631 #define VMBUS_CHAN_ATTR_RW(_name) \
1632         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RW(_name)
1633 #define VMBUS_CHAN_ATTR_RO(_name) \
1634         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RO(_name)
1635 #define VMBUS_CHAN_ATTR_WO(_name) \
1636         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_WO(_name)
1637
1638 static ssize_t vmbus_chan_attr_show(struct kobject *kobj,
1639                                     struct attribute *attr, char *buf)
1640 {
1641         const struct vmbus_chan_attribute *attribute
1642                 = container_of(attr, struct vmbus_chan_attribute, attr);
1643         struct vmbus_channel *chan
1644                 = container_of(kobj, struct vmbus_channel, kobj);
1645
1646         if (!attribute->show)
1647                 return -EIO;
1648
1649         return attribute->show(chan, buf);
1650 }
1651
1652 static ssize_t vmbus_chan_attr_store(struct kobject *kobj,
1653                                      struct attribute *attr, const char *buf,
1654                                      size_t count)
1655 {
1656         const struct vmbus_chan_attribute *attribute
1657                 = container_of(attr, struct vmbus_chan_attribute, attr);
1658         struct vmbus_channel *chan
1659                 = container_of(kobj, struct vmbus_channel, kobj);
1660
1661         if (!attribute->store)
1662                 return -EIO;
1663
1664         return attribute->store(chan, buf, count);
1665 }
1666
1667 static const struct sysfs_ops vmbus_chan_sysfs_ops = {
1668         .show = vmbus_chan_attr_show,
1669         .store = vmbus_chan_attr_store,
1670 };
1671
1672 static ssize_t out_mask_show(struct vmbus_channel *channel, char *buf)
1673 {
1674         struct hv_ring_buffer_info *rbi = &channel->outbound;
1675         ssize_t ret;
1676
1677         mutex_lock(&rbi->ring_buffer_mutex);
1678         if (!rbi->ring_buffer) {
1679                 mutex_unlock(&rbi->ring_buffer_mutex);
1680                 return -EINVAL;
1681         }
1682
1683         ret = sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1684         mutex_unlock(&rbi->ring_buffer_mutex);
1685         return ret;
1686 }
1687 static VMBUS_CHAN_ATTR_RO(out_mask);
1688
1689 static ssize_t in_mask_show(struct vmbus_channel *channel, char *buf)
1690 {
1691         struct hv_ring_buffer_info *rbi = &channel->inbound;
1692         ssize_t ret;
1693
1694         mutex_lock(&rbi->ring_buffer_mutex);
1695         if (!rbi->ring_buffer) {
1696                 mutex_unlock(&rbi->ring_buffer_mutex);
1697                 return -EINVAL;
1698         }
1699
1700         ret = sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1701         mutex_unlock(&rbi->ring_buffer_mutex);
1702         return ret;
1703 }
1704 static VMBUS_CHAN_ATTR_RO(in_mask);
1705
1706 static ssize_t read_avail_show(struct vmbus_channel *channel, char *buf)
1707 {
1708         struct hv_ring_buffer_info *rbi = &channel->inbound;
1709         ssize_t ret;
1710
1711         mutex_lock(&rbi->ring_buffer_mutex);
1712         if (!rbi->ring_buffer) {
1713                 mutex_unlock(&rbi->ring_buffer_mutex);
1714                 return -EINVAL;
1715         }
1716
1717         ret = sprintf(buf, "%u\n", hv_get_bytes_to_read(rbi));
1718         mutex_unlock(&rbi->ring_buffer_mutex);
1719         return ret;
1720 }
1721 static VMBUS_CHAN_ATTR_RO(read_avail);
1722
1723 static ssize_t write_avail_show(struct vmbus_channel *channel, char *buf)
1724 {
1725         struct hv_ring_buffer_info *rbi = &channel->outbound;
1726         ssize_t ret;
1727
1728         mutex_lock(&rbi->ring_buffer_mutex);
1729         if (!rbi->ring_buffer) {
1730                 mutex_unlock(&rbi->ring_buffer_mutex);
1731                 return -EINVAL;
1732         }
1733
1734         ret = sprintf(buf, "%u\n", hv_get_bytes_to_write(rbi));
1735         mutex_unlock(&rbi->ring_buffer_mutex);
1736         return ret;
1737 }
1738 static VMBUS_CHAN_ATTR_RO(write_avail);
1739
1740 static ssize_t target_cpu_show(struct vmbus_channel *channel, char *buf)
1741 {
1742         return sprintf(buf, "%u\n", channel->target_cpu);
1743 }
1744 static ssize_t target_cpu_store(struct vmbus_channel *channel,
1745                                 const char *buf, size_t count)
1746 {
1747         u32 target_cpu, origin_cpu;
1748         ssize_t ret = count;
1749
1750         if (vmbus_proto_version < VERSION_WIN10_V4_1)
1751                 return -EIO;
1752
1753         if (sscanf(buf, "%uu", &target_cpu) != 1)
1754                 return -EIO;
1755
1756         /* Validate target_cpu for the cpumask_test_cpu() operation below. */
1757         if (target_cpu >= nr_cpumask_bits)
1758                 return -EINVAL;
1759
1760         /* No CPUs should come up or down during this. */
1761         cpus_read_lock();
1762
1763         if (!cpu_online(target_cpu)) {
1764                 cpus_read_unlock();
1765                 return -EINVAL;
1766         }
1767
1768         /*
1769          * Synchronizes target_cpu_store() and channel closure:
1770          *
1771          * { Initially: state = CHANNEL_OPENED }
1772          *
1773          * CPU1                         CPU2
1774          *
1775          * [target_cpu_store()]         [vmbus_disconnect_ring()]
1776          *
1777          * LOCK channel_mutex           LOCK channel_mutex
1778          * LOAD r1 = state              LOAD r2 = state
1779          * IF (r1 == CHANNEL_OPENED)    IF (r2 == CHANNEL_OPENED)
1780          *   SEND MODIFYCHANNEL           STORE state = CHANNEL_OPEN
1781          *   [...]                        SEND CLOSECHANNEL
1782          * UNLOCK channel_mutex         UNLOCK channel_mutex
1783          *
1784          * Forbids: r1 == r2 == CHANNEL_OPENED (i.e., CPU1's LOCK precedes
1785          *              CPU2's LOCK) && CPU2's SEND precedes CPU1's SEND
1786          *
1787          * Note.  The host processes the channel messages "sequentially", in
1788          * the order in which they are received on a per-partition basis.
1789          */
1790         mutex_lock(&vmbus_connection.channel_mutex);
1791
1792         /*
1793          * Hyper-V will ignore MODIFYCHANNEL messages for "non-open" channels;
1794          * avoid sending the message and fail here for such channels.
1795          */
1796         if (channel->state != CHANNEL_OPENED_STATE) {
1797                 ret = -EIO;
1798                 goto cpu_store_unlock;
1799         }
1800
1801         origin_cpu = channel->target_cpu;
1802         if (target_cpu == origin_cpu)
1803                 goto cpu_store_unlock;
1804
1805         if (vmbus_send_modifychannel(channel->offermsg.child_relid,
1806                                      hv_cpu_number_to_vp_number(target_cpu))) {
1807                 ret = -EIO;
1808                 goto cpu_store_unlock;
1809         }
1810
1811         /*
1812          * Warning.  At this point, there is *no* guarantee that the host will
1813          * have successfully processed the vmbus_send_modifychannel() request.
1814          * See the header comment of vmbus_send_modifychannel() for more info.
1815          *
1816          * Lags in the processing of the above vmbus_send_modifychannel() can
1817          * result in missed interrupts if the "old" target CPU is taken offline
1818          * before Hyper-V starts sending interrupts to the "new" target CPU.
1819          * But apart from this offlining scenario, the code tolerates such
1820          * lags.  It will function correctly even if a channel interrupt comes
1821          * in on a CPU that is different from the channel target_cpu value.
1822          */
1823
1824         channel->target_cpu = target_cpu;
1825
1826         /* See init_vp_index(). */
1827         if (hv_is_perf_channel(channel))
1828                 hv_update_alloced_cpus(origin_cpu, target_cpu);
1829
1830         /* Currently set only for storvsc channels. */
1831         if (channel->change_target_cpu_callback) {
1832                 (*channel->change_target_cpu_callback)(channel,
1833                                 origin_cpu, target_cpu);
1834         }
1835
1836 cpu_store_unlock:
1837         mutex_unlock(&vmbus_connection.channel_mutex);
1838         cpus_read_unlock();
1839         return ret;
1840 }
1841 static VMBUS_CHAN_ATTR(cpu, 0644, target_cpu_show, target_cpu_store);
1842
1843 static ssize_t channel_pending_show(struct vmbus_channel *channel,
1844                                     char *buf)
1845 {
1846         return sprintf(buf, "%d\n",
1847                        channel_pending(channel,
1848                                        vmbus_connection.monitor_pages[1]));
1849 }
1850 static VMBUS_CHAN_ATTR(pending, 0444, channel_pending_show, NULL);
1851
1852 static ssize_t channel_latency_show(struct vmbus_channel *channel,
1853                                     char *buf)
1854 {
1855         return sprintf(buf, "%d\n",
1856                        channel_latency(channel,
1857                                        vmbus_connection.monitor_pages[1]));
1858 }
1859 static VMBUS_CHAN_ATTR(latency, 0444, channel_latency_show, NULL);
1860
1861 static ssize_t channel_interrupts_show(struct vmbus_channel *channel, char *buf)
1862 {
1863         return sprintf(buf, "%llu\n", channel->interrupts);
1864 }
1865 static VMBUS_CHAN_ATTR(interrupts, 0444, channel_interrupts_show, NULL);
1866
1867 static ssize_t channel_events_show(struct vmbus_channel *channel, char *buf)
1868 {
1869         return sprintf(buf, "%llu\n", channel->sig_events);
1870 }
1871 static VMBUS_CHAN_ATTR(events, 0444, channel_events_show, NULL);
1872
1873 static ssize_t channel_intr_in_full_show(struct vmbus_channel *channel,
1874                                          char *buf)
1875 {
1876         return sprintf(buf, "%llu\n",
1877                        (unsigned long long)channel->intr_in_full);
1878 }
1879 static VMBUS_CHAN_ATTR(intr_in_full, 0444, channel_intr_in_full_show, NULL);
1880
1881 static ssize_t channel_intr_out_empty_show(struct vmbus_channel *channel,
1882                                            char *buf)
1883 {
1884         return sprintf(buf, "%llu\n",
1885                        (unsigned long long)channel->intr_out_empty);
1886 }
1887 static VMBUS_CHAN_ATTR(intr_out_empty, 0444, channel_intr_out_empty_show, NULL);
1888
1889 static ssize_t channel_out_full_first_show(struct vmbus_channel *channel,
1890                                            char *buf)
1891 {
1892         return sprintf(buf, "%llu\n",
1893                        (unsigned long long)channel->out_full_first);
1894 }
1895 static VMBUS_CHAN_ATTR(out_full_first, 0444, channel_out_full_first_show, NULL);
1896
1897 static ssize_t channel_out_full_total_show(struct vmbus_channel *channel,
1898                                            char *buf)
1899 {
1900         return sprintf(buf, "%llu\n",
1901                        (unsigned long long)channel->out_full_total);
1902 }
1903 static VMBUS_CHAN_ATTR(out_full_total, 0444, channel_out_full_total_show, NULL);
1904
1905 static ssize_t subchannel_monitor_id_show(struct vmbus_channel *channel,
1906                                           char *buf)
1907 {
1908         return sprintf(buf, "%u\n", channel->offermsg.monitorid);
1909 }
1910 static VMBUS_CHAN_ATTR(monitor_id, 0444, subchannel_monitor_id_show, NULL);
1911
1912 static ssize_t subchannel_id_show(struct vmbus_channel *channel,
1913                                   char *buf)
1914 {
1915         return sprintf(buf, "%u\n",
1916                        channel->offermsg.offer.sub_channel_index);
1917 }
1918 static VMBUS_CHAN_ATTR_RO(subchannel_id);
1919
1920 static struct attribute *vmbus_chan_attrs[] = {
1921         &chan_attr_out_mask.attr,
1922         &chan_attr_in_mask.attr,
1923         &chan_attr_read_avail.attr,
1924         &chan_attr_write_avail.attr,
1925         &chan_attr_cpu.attr,
1926         &chan_attr_pending.attr,
1927         &chan_attr_latency.attr,
1928         &chan_attr_interrupts.attr,
1929         &chan_attr_events.attr,
1930         &chan_attr_intr_in_full.attr,
1931         &chan_attr_intr_out_empty.attr,
1932         &chan_attr_out_full_first.attr,
1933         &chan_attr_out_full_total.attr,
1934         &chan_attr_monitor_id.attr,
1935         &chan_attr_subchannel_id.attr,
1936         NULL
1937 };
1938
1939 /*
1940  * Channel-level attribute_group callback function. Returns the permission for
1941  * each attribute, and returns 0 if an attribute is not visible.
1942  */
1943 static umode_t vmbus_chan_attr_is_visible(struct kobject *kobj,
1944                                           struct attribute *attr, int idx)
1945 {
1946         const struct vmbus_channel *channel =
1947                 container_of(kobj, struct vmbus_channel, kobj);
1948
1949         /* Hide the monitor attributes if the monitor mechanism is not used. */
1950         if (!channel->offermsg.monitor_allocated &&
1951             (attr == &chan_attr_pending.attr ||
1952              attr == &chan_attr_latency.attr ||
1953              attr == &chan_attr_monitor_id.attr))
1954                 return 0;
1955
1956         return attr->mode;
1957 }
1958
1959 static struct attribute_group vmbus_chan_group = {
1960         .attrs = vmbus_chan_attrs,
1961         .is_visible = vmbus_chan_attr_is_visible
1962 };
1963
1964 static struct kobj_type vmbus_chan_ktype = {
1965         .sysfs_ops = &vmbus_chan_sysfs_ops,
1966         .release = vmbus_chan_release,
1967 };
1968
1969 /*
1970  * vmbus_add_channel_kobj - setup a sub-directory under device/channels
1971  */
1972 int vmbus_add_channel_kobj(struct hv_device *dev, struct vmbus_channel *channel)
1973 {
1974         const struct device *device = &dev->device;
1975         struct kobject *kobj = &channel->kobj;
1976         u32 relid = channel->offermsg.child_relid;
1977         int ret;
1978
1979         kobj->kset = dev->channels_kset;
1980         ret = kobject_init_and_add(kobj, &vmbus_chan_ktype, NULL,
1981                                    "%u", relid);
1982         if (ret)
1983                 return ret;
1984
1985         ret = sysfs_create_group(kobj, &vmbus_chan_group);
1986
1987         if (ret) {
1988                 /*
1989                  * The calling functions' error handling paths will cleanup the
1990                  * empty channel directory.
1991                  */
1992                 dev_err(device, "Unable to set up channel sysfs files\n");
1993                 return ret;
1994         }
1995
1996         kobject_uevent(kobj, KOBJ_ADD);
1997
1998         return 0;
1999 }
2000
2001 /*
2002  * vmbus_remove_channel_attr_group - remove the channel's attribute group
2003  */
2004 void vmbus_remove_channel_attr_group(struct vmbus_channel *channel)
2005 {
2006         sysfs_remove_group(&channel->kobj, &vmbus_chan_group);
2007 }
2008
2009 /*
2010  * vmbus_device_create - Creates and registers a new child device
2011  * on the vmbus.
2012  */
2013 struct hv_device *vmbus_device_create(const guid_t *type,
2014                                       const guid_t *instance,
2015                                       struct vmbus_channel *channel)
2016 {
2017         struct hv_device *child_device_obj;
2018
2019         child_device_obj = kzalloc(sizeof(struct hv_device), GFP_KERNEL);
2020         if (!child_device_obj) {
2021                 pr_err("Unable to allocate device object for child device\n");
2022                 return NULL;
2023         }
2024
2025         child_device_obj->channel = channel;
2026         guid_copy(&child_device_obj->dev_type, type);
2027         guid_copy(&child_device_obj->dev_instance, instance);
2028         child_device_obj->vendor_id = 0x1414; /* MSFT vendor ID */
2029
2030         return child_device_obj;
2031 }
2032
2033 /*
2034  * vmbus_device_register - Register the child device
2035  */
2036 int vmbus_device_register(struct hv_device *child_device_obj)
2037 {
2038         struct kobject *kobj = &child_device_obj->device.kobj;
2039         int ret;
2040
2041         dev_set_name(&child_device_obj->device, "%pUl",
2042                      &child_device_obj->channel->offermsg.offer.if_instance);
2043
2044         child_device_obj->device.bus = &hv_bus;
2045         child_device_obj->device.parent = &hv_acpi_dev->dev;
2046         child_device_obj->device.release = vmbus_device_release;
2047
2048         /*
2049          * Register with the LDM. This will kick off the driver/device
2050          * binding...which will eventually call vmbus_match() and vmbus_probe()
2051          */
2052         ret = device_register(&child_device_obj->device);
2053         if (ret) {
2054                 pr_err("Unable to register child device\n");
2055                 return ret;
2056         }
2057
2058         child_device_obj->channels_kset = kset_create_and_add("channels",
2059                                                               NULL, kobj);
2060         if (!child_device_obj->channels_kset) {
2061                 ret = -ENOMEM;
2062                 goto err_dev_unregister;
2063         }
2064
2065         ret = vmbus_add_channel_kobj(child_device_obj,
2066                                      child_device_obj->channel);
2067         if (ret) {
2068                 pr_err("Unable to register primary channeln");
2069                 goto err_kset_unregister;
2070         }
2071         hv_debug_add_dev_dir(child_device_obj);
2072
2073         return 0;
2074
2075 err_kset_unregister:
2076         kset_unregister(child_device_obj->channels_kset);
2077
2078 err_dev_unregister:
2079         device_unregister(&child_device_obj->device);
2080         return ret;
2081 }
2082
2083 /*
2084  * vmbus_device_unregister - Remove the specified child device
2085  * from the vmbus.
2086  */
2087 void vmbus_device_unregister(struct hv_device *device_obj)
2088 {
2089         pr_debug("child device %s unregistered\n",
2090                 dev_name(&device_obj->device));
2091
2092         kset_unregister(device_obj->channels_kset);
2093
2094         /*
2095          * Kick off the process of unregistering the device.
2096          * This will call vmbus_remove() and eventually vmbus_device_release()
2097          */
2098         device_unregister(&device_obj->device);
2099 }
2100
2101
2102 /*
2103  * VMBUS is an acpi enumerated device. Get the information we
2104  * need from DSDT.
2105  */
2106 #define VTPM_BASE_ADDRESS 0xfed40000
2107 static acpi_status vmbus_walk_resources(struct acpi_resource *res, void *ctx)
2108 {
2109         resource_size_t start = 0;
2110         resource_size_t end = 0;
2111         struct resource *new_res;
2112         struct resource **old_res = &hyperv_mmio;
2113         struct resource **prev_res = NULL;
2114         struct resource r;
2115
2116         switch (res->type) {
2117
2118         /*
2119          * "Address" descriptors are for bus windows. Ignore
2120          * "memory" descriptors, which are for registers on
2121          * devices.
2122          */
2123         case ACPI_RESOURCE_TYPE_ADDRESS32:
2124                 start = res->data.address32.address.minimum;
2125                 end = res->data.address32.address.maximum;
2126                 break;
2127
2128         case ACPI_RESOURCE_TYPE_ADDRESS64:
2129                 start = res->data.address64.address.minimum;
2130                 end = res->data.address64.address.maximum;
2131                 break;
2132
2133         /*
2134          * The IRQ information is needed only on ARM64, which Hyper-V
2135          * sets up in the extended format. IRQ information is present
2136          * on x86/x64 in the non-extended format but it is not used by
2137          * Linux. So don't bother checking for the non-extended format.
2138          */
2139         case ACPI_RESOURCE_TYPE_EXTENDED_IRQ:
2140                 if (!acpi_dev_resource_interrupt(res, 0, &r)) {
2141                         pr_err("Unable to parse Hyper-V ACPI interrupt\n");
2142                         return AE_ERROR;
2143                 }
2144                 /* ARM64 INTID for VMbus */
2145                 vmbus_interrupt = res->data.extended_irq.interrupts[0];
2146                 /* Linux IRQ number */
2147                 vmbus_irq = r.start;
2148                 return AE_OK;
2149
2150         default:
2151                 /* Unused resource type */
2152                 return AE_OK;
2153
2154         }
2155         /*
2156          * Ignore ranges that are below 1MB, as they're not
2157          * necessary or useful here.
2158          */
2159         if (end < 0x100000)
2160                 return AE_OK;
2161
2162         new_res = kzalloc(sizeof(*new_res), GFP_ATOMIC);
2163         if (!new_res)
2164                 return AE_NO_MEMORY;
2165
2166         /* If this range overlaps the virtual TPM, truncate it. */
2167         if (end > VTPM_BASE_ADDRESS && start < VTPM_BASE_ADDRESS)
2168                 end = VTPM_BASE_ADDRESS;
2169
2170         new_res->name = "hyperv mmio";
2171         new_res->flags = IORESOURCE_MEM;
2172         new_res->start = start;
2173         new_res->end = end;
2174
2175         /*
2176          * If two ranges are adjacent, merge them.
2177          */
2178         do {
2179                 if (!*old_res) {
2180                         *old_res = new_res;
2181                         break;
2182                 }
2183
2184                 if (((*old_res)->end + 1) == new_res->start) {
2185                         (*old_res)->end = new_res->end;
2186                         kfree(new_res);
2187                         break;
2188                 }
2189
2190                 if ((*old_res)->start == new_res->end + 1) {
2191                         (*old_res)->start = new_res->start;
2192                         kfree(new_res);
2193                         break;
2194                 }
2195
2196                 if ((*old_res)->start > new_res->end) {
2197                         new_res->sibling = *old_res;
2198                         if (prev_res)
2199                                 (*prev_res)->sibling = new_res;
2200                         *old_res = new_res;
2201                         break;
2202                 }
2203
2204                 prev_res = old_res;
2205                 old_res = &(*old_res)->sibling;
2206
2207         } while (1);
2208
2209         return AE_OK;
2210 }
2211
2212 static int vmbus_acpi_remove(struct acpi_device *device)
2213 {
2214         struct resource *cur_res;
2215         struct resource *next_res;
2216
2217         if (hyperv_mmio) {
2218                 if (fb_mmio) {
2219                         __release_region(hyperv_mmio, fb_mmio->start,
2220                                          resource_size(fb_mmio));
2221                         fb_mmio = NULL;
2222                 }
2223
2224                 for (cur_res = hyperv_mmio; cur_res; cur_res = next_res) {
2225                         next_res = cur_res->sibling;
2226                         kfree(cur_res);
2227                 }
2228         }
2229
2230         return 0;
2231 }
2232
2233 static void vmbus_reserve_fb(void)
2234 {
2235         int size;
2236         /*
2237          * Make a claim for the frame buffer in the resource tree under the
2238          * first node, which will be the one below 4GB.  The length seems to
2239          * be underreported, particularly in a Generation 1 VM.  So start out
2240          * reserving a larger area and make it smaller until it succeeds.
2241          */
2242
2243         if (screen_info.lfb_base) {
2244                 if (efi_enabled(EFI_BOOT))
2245                         size = max_t(__u32, screen_info.lfb_size, 0x800000);
2246                 else
2247                         size = max_t(__u32, screen_info.lfb_size, 0x4000000);
2248
2249                 for (; !fb_mmio && (size >= 0x100000); size >>= 1) {
2250                         fb_mmio = __request_region(hyperv_mmio,
2251                                                    screen_info.lfb_base, size,
2252                                                    fb_mmio_name, 0);
2253                 }
2254         }
2255 }
2256
2257 /**
2258  * vmbus_allocate_mmio() - Pick a memory-mapped I/O range.
2259  * @new:                If successful, supplied a pointer to the
2260  *                      allocated MMIO space.
2261  * @device_obj:         Identifies the caller
2262  * @min:                Minimum guest physical address of the
2263  *                      allocation
2264  * @max:                Maximum guest physical address
2265  * @size:               Size of the range to be allocated
2266  * @align:              Alignment of the range to be allocated
2267  * @fb_overlap_ok:      Whether this allocation can be allowed
2268  *                      to overlap the video frame buffer.
2269  *
2270  * This function walks the resources granted to VMBus by the
2271  * _CRS object in the ACPI namespace underneath the parent
2272  * "bridge" whether that's a root PCI bus in the Generation 1
2273  * case or a Module Device in the Generation 2 case.  It then
2274  * attempts to allocate from the global MMIO pool in a way that
2275  * matches the constraints supplied in these parameters and by
2276  * that _CRS.
2277  *
2278  * Return: 0 on success, -errno on failure
2279  */
2280 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
2281                         resource_size_t min, resource_size_t max,
2282                         resource_size_t size, resource_size_t align,
2283                         bool fb_overlap_ok)
2284 {
2285         struct resource *iter, *shadow;
2286         resource_size_t range_min, range_max, start;
2287         const char *dev_n = dev_name(&device_obj->device);
2288         int retval;
2289
2290         retval = -ENXIO;
2291         mutex_lock(&hyperv_mmio_lock);
2292
2293         /*
2294          * If overlaps with frame buffers are allowed, then first attempt to
2295          * make the allocation from within the reserved region.  Because it
2296          * is already reserved, no shadow allocation is necessary.
2297          */
2298         if (fb_overlap_ok && fb_mmio && !(min > fb_mmio->end) &&
2299             !(max < fb_mmio->start)) {
2300
2301                 range_min = fb_mmio->start;
2302                 range_max = fb_mmio->end;
2303                 start = (range_min + align - 1) & ~(align - 1);
2304                 for (; start + size - 1 <= range_max; start += align) {
2305                         *new = request_mem_region_exclusive(start, size, dev_n);
2306                         if (*new) {
2307                                 retval = 0;
2308                                 goto exit;
2309                         }
2310                 }
2311         }
2312
2313         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
2314                 if ((iter->start >= max) || (iter->end <= min))
2315                         continue;
2316
2317                 range_min = iter->start;
2318                 range_max = iter->end;
2319                 start = (range_min + align - 1) & ~(align - 1);
2320                 for (; start + size - 1 <= range_max; start += align) {
2321                         shadow = __request_region(iter, start, size, NULL,
2322                                                   IORESOURCE_BUSY);
2323                         if (!shadow)
2324                                 continue;
2325
2326                         *new = request_mem_region_exclusive(start, size, dev_n);
2327                         if (*new) {
2328                                 shadow->name = (char *)*new;
2329                                 retval = 0;
2330                                 goto exit;
2331                         }
2332
2333                         __release_region(iter, start, size);
2334                 }
2335         }
2336
2337 exit:
2338         mutex_unlock(&hyperv_mmio_lock);
2339         return retval;
2340 }
2341 EXPORT_SYMBOL_GPL(vmbus_allocate_mmio);
2342
2343 /**
2344  * vmbus_free_mmio() - Free a memory-mapped I/O range.
2345  * @start:              Base address of region to release.
2346  * @size:               Size of the range to be allocated
2347  *
2348  * This function releases anything requested by
2349  * vmbus_mmio_allocate().
2350  */
2351 void vmbus_free_mmio(resource_size_t start, resource_size_t size)
2352 {
2353         struct resource *iter;
2354
2355         mutex_lock(&hyperv_mmio_lock);
2356         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
2357                 if ((iter->start >= start + size) || (iter->end <= start))
2358                         continue;
2359
2360                 __release_region(iter, start, size);
2361         }
2362         release_mem_region(start, size);
2363         mutex_unlock(&hyperv_mmio_lock);
2364
2365 }
2366 EXPORT_SYMBOL_GPL(vmbus_free_mmio);
2367
2368 static int vmbus_acpi_add(struct acpi_device *device)
2369 {
2370         acpi_status result;
2371         int ret_val = -ENODEV;
2372         struct acpi_device *ancestor;
2373
2374         hv_acpi_dev = device;
2375
2376         result = acpi_walk_resources(device->handle, METHOD_NAME__CRS,
2377                                         vmbus_walk_resources, NULL);
2378
2379         if (ACPI_FAILURE(result))
2380                 goto acpi_walk_err;
2381         /*
2382          * Some ancestor of the vmbus acpi device (Gen1 or Gen2
2383          * firmware) is the VMOD that has the mmio ranges. Get that.
2384          */
2385         for (ancestor = device->parent; ancestor; ancestor = ancestor->parent) {
2386                 result = acpi_walk_resources(ancestor->handle, METHOD_NAME__CRS,
2387                                              vmbus_walk_resources, NULL);
2388
2389                 if (ACPI_FAILURE(result))
2390                         continue;
2391                 if (hyperv_mmio) {
2392                         vmbus_reserve_fb();
2393                         break;
2394                 }
2395         }
2396         ret_val = 0;
2397
2398 acpi_walk_err:
2399         complete(&probe_event);
2400         if (ret_val)
2401                 vmbus_acpi_remove(device);
2402         return ret_val;
2403 }
2404
2405 #ifdef CONFIG_PM_SLEEP
2406 static int vmbus_bus_suspend(struct device *dev)
2407 {
2408         struct vmbus_channel *channel, *sc;
2409
2410         while (atomic_read(&vmbus_connection.offer_in_progress) != 0) {
2411                 /*
2412                  * We wait here until the completion of any channel
2413                  * offers that are currently in progress.
2414                  */
2415                 usleep_range(1000, 2000);
2416         }
2417
2418         mutex_lock(&vmbus_connection.channel_mutex);
2419         list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
2420                 if (!is_hvsock_channel(channel))
2421                         continue;
2422
2423                 vmbus_force_channel_rescinded(channel);
2424         }
2425         mutex_unlock(&vmbus_connection.channel_mutex);
2426
2427         /*
2428          * Wait until all the sub-channels and hv_sock channels have been
2429          * cleaned up. Sub-channels should be destroyed upon suspend, otherwise
2430          * they would conflict with the new sub-channels that will be created
2431          * in the resume path. hv_sock channels should also be destroyed, but
2432          * a hv_sock channel of an established hv_sock connection can not be
2433          * really destroyed since it may still be referenced by the userspace
2434          * application, so we just force the hv_sock channel to be rescinded
2435          * by vmbus_force_channel_rescinded(), and the userspace application
2436          * will thoroughly destroy the channel after hibernation.
2437          *
2438          * Note: the counter nr_chan_close_on_suspend may never go above 0 if
2439          * the VM has no sub-channel and hv_sock channel, e.g. a 1-vCPU VM.
2440          */
2441         if (atomic_read(&vmbus_connection.nr_chan_close_on_suspend) > 0)
2442                 wait_for_completion(&vmbus_connection.ready_for_suspend_event);
2443
2444         if (atomic_read(&vmbus_connection.nr_chan_fixup_on_resume) != 0) {
2445                 pr_err("Can not suspend due to a previous failed resuming\n");
2446                 return -EBUSY;
2447         }
2448
2449         mutex_lock(&vmbus_connection.channel_mutex);
2450
2451         list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
2452                 /*
2453                  * Remove the channel from the array of channels and invalidate
2454                  * the channel's relid.  Upon resume, vmbus_onoffer() will fix
2455                  * up the relid (and other fields, if necessary) and add the
2456                  * channel back to the array.
2457                  */
2458                 vmbus_channel_unmap_relid(channel);
2459                 channel->offermsg.child_relid = INVALID_RELID;
2460
2461                 if (is_hvsock_channel(channel)) {
2462                         if (!channel->rescind) {
2463                                 pr_err("hv_sock channel not rescinded!\n");
2464                                 WARN_ON_ONCE(1);
2465                         }
2466                         continue;
2467                 }
2468
2469                 list_for_each_entry(sc, &channel->sc_list, sc_list) {
2470                         pr_err("Sub-channel not deleted!\n");
2471                         WARN_ON_ONCE(1);
2472                 }
2473
2474                 atomic_inc(&vmbus_connection.nr_chan_fixup_on_resume);
2475         }
2476
2477         mutex_unlock(&vmbus_connection.channel_mutex);
2478
2479         vmbus_initiate_unload(false);
2480
2481         /* Reset the event for the next resume. */
2482         reinit_completion(&vmbus_connection.ready_for_resume_event);
2483
2484         return 0;
2485 }
2486
2487 static int vmbus_bus_resume(struct device *dev)
2488 {
2489         struct vmbus_channel_msginfo *msginfo;
2490         size_t msgsize;
2491         int ret;
2492
2493         /*
2494          * We only use the 'vmbus_proto_version', which was in use before
2495          * hibernation, to re-negotiate with the host.
2496          */
2497         if (!vmbus_proto_version) {
2498                 pr_err("Invalid proto version = 0x%x\n", vmbus_proto_version);
2499                 return -EINVAL;
2500         }
2501
2502         msgsize = sizeof(*msginfo) +
2503                   sizeof(struct vmbus_channel_initiate_contact);
2504
2505         msginfo = kzalloc(msgsize, GFP_KERNEL);
2506
2507         if (msginfo == NULL)
2508                 return -ENOMEM;
2509
2510         ret = vmbus_negotiate_version(msginfo, vmbus_proto_version);
2511
2512         kfree(msginfo);
2513
2514         if (ret != 0)
2515                 return ret;
2516
2517         WARN_ON(atomic_read(&vmbus_connection.nr_chan_fixup_on_resume) == 0);
2518
2519         vmbus_request_offers();
2520
2521         if (wait_for_completion_timeout(
2522                 &vmbus_connection.ready_for_resume_event, 10 * HZ) == 0)
2523                 pr_err("Some vmbus device is missing after suspending?\n");
2524
2525         /* Reset the event for the next suspend. */
2526         reinit_completion(&vmbus_connection.ready_for_suspend_event);
2527
2528         return 0;
2529 }
2530 #else
2531 #define vmbus_bus_suspend NULL
2532 #define vmbus_bus_resume NULL
2533 #endif /* CONFIG_PM_SLEEP */
2534
2535 static const struct acpi_device_id vmbus_acpi_device_ids[] = {
2536         {"VMBUS", 0},
2537         {"VMBus", 0},
2538         {"", 0},
2539 };
2540 MODULE_DEVICE_TABLE(acpi, vmbus_acpi_device_ids);
2541
2542 /*
2543  * Note: we must use the "no_irq" ops, otherwise hibernation can not work with
2544  * PCI device assignment, because "pci_dev_pm_ops" uses the "noirq" ops: in
2545  * the resume path, the pci "noirq" restore op runs before "non-noirq" op (see
2546  * resume_target_kernel() -> dpm_resume_start(), and hibernation_restore() ->
2547  * dpm_resume_end()). This means vmbus_bus_resume() and the pci-hyperv's
2548  * resume callback must also run via the "noirq" ops.
2549  *
2550  * Set suspend_noirq/resume_noirq to NULL for Suspend-to-Idle: see the comment
2551  * earlier in this file before vmbus_pm.
2552  */
2553
2554 static const struct dev_pm_ops vmbus_bus_pm = {
2555         .suspend_noirq  = NULL,
2556         .resume_noirq   = NULL,
2557         .freeze_noirq   = vmbus_bus_suspend,
2558         .thaw_noirq     = vmbus_bus_resume,
2559         .poweroff_noirq = vmbus_bus_suspend,
2560         .restore_noirq  = vmbus_bus_resume
2561 };
2562
2563 static struct acpi_driver vmbus_acpi_driver = {
2564         .name = "vmbus",
2565         .ids = vmbus_acpi_device_ids,
2566         .ops = {
2567                 .add = vmbus_acpi_add,
2568                 .remove = vmbus_acpi_remove,
2569         },
2570         .drv.pm = &vmbus_bus_pm,
2571 };
2572
2573 static void hv_kexec_handler(void)
2574 {
2575         hv_stimer_global_cleanup();
2576         vmbus_initiate_unload(false);
2577         /* Make sure conn_state is set as hv_synic_cleanup checks for it */
2578         mb();
2579         cpuhp_remove_state(hyperv_cpuhp_online);
2580 };
2581
2582 static void hv_crash_handler(struct pt_regs *regs)
2583 {
2584         int cpu;
2585
2586         vmbus_initiate_unload(true);
2587         /*
2588          * In crash handler we can't schedule synic cleanup for all CPUs,
2589          * doing the cleanup for current CPU only. This should be sufficient
2590          * for kdump.
2591          */
2592         cpu = smp_processor_id();
2593         hv_stimer_cleanup(cpu);
2594         hv_synic_disable_regs(cpu);
2595 };
2596
2597 static int hv_synic_suspend(void)
2598 {
2599         /*
2600          * When we reach here, all the non-boot CPUs have been offlined.
2601          * If we're in a legacy configuration where stimer Direct Mode is
2602          * not enabled, the stimers on the non-boot CPUs have been unbound
2603          * in hv_synic_cleanup() -> hv_stimer_legacy_cleanup() ->
2604          * hv_stimer_cleanup() -> clockevents_unbind_device().
2605          *
2606          * hv_synic_suspend() only runs on CPU0 with interrupts disabled.
2607          * Here we do not call hv_stimer_legacy_cleanup() on CPU0 because:
2608          * 1) it's unnecessary as interrupts remain disabled between
2609          * syscore_suspend() and syscore_resume(): see create_image() and
2610          * resume_target_kernel()
2611          * 2) the stimer on CPU0 is automatically disabled later by
2612          * syscore_suspend() -> timekeeping_suspend() -> tick_suspend() -> ...
2613          * -> clockevents_shutdown() -> ... -> hv_ce_shutdown()
2614          * 3) a warning would be triggered if we call
2615          * clockevents_unbind_device(), which may sleep, in an
2616          * interrupts-disabled context.
2617          */
2618
2619         hv_synic_disable_regs(0);
2620
2621         return 0;
2622 }
2623
2624 static void hv_synic_resume(void)
2625 {
2626         hv_synic_enable_regs(0);
2627
2628         /*
2629          * Note: we don't need to call hv_stimer_init(0), because the timer
2630          * on CPU0 is not unbound in hv_synic_suspend(), and the timer is
2631          * automatically re-enabled in timekeeping_resume().
2632          */
2633 }
2634
2635 /* The callbacks run only on CPU0, with irqs_disabled. */
2636 static struct syscore_ops hv_synic_syscore_ops = {
2637         .suspend = hv_synic_suspend,
2638         .resume = hv_synic_resume,
2639 };
2640
2641 static int __init hv_acpi_init(void)
2642 {
2643         int ret, t;
2644
2645         if (!hv_is_hyperv_initialized())
2646                 return -ENODEV;
2647
2648         if (hv_root_partition)
2649                 return 0;
2650
2651         init_completion(&probe_event);
2652
2653         /*
2654          * Get ACPI resources first.
2655          */
2656         ret = acpi_bus_register_driver(&vmbus_acpi_driver);
2657
2658         if (ret)
2659                 return ret;
2660
2661         t = wait_for_completion_timeout(&probe_event, 5*HZ);
2662         if (t == 0) {
2663                 ret = -ETIMEDOUT;
2664                 goto cleanup;
2665         }
2666         hv_debug_init();
2667
2668         ret = vmbus_bus_init();
2669         if (ret)
2670                 goto cleanup;
2671
2672         hv_setup_kexec_handler(hv_kexec_handler);
2673         hv_setup_crash_handler(hv_crash_handler);
2674
2675         register_syscore_ops(&hv_synic_syscore_ops);
2676
2677         return 0;
2678
2679 cleanup:
2680         acpi_bus_unregister_driver(&vmbus_acpi_driver);
2681         hv_acpi_dev = NULL;
2682         return ret;
2683 }
2684
2685 static void __exit vmbus_exit(void)
2686 {
2687         int cpu;
2688
2689         unregister_syscore_ops(&hv_synic_syscore_ops);
2690
2691         hv_remove_kexec_handler();
2692         hv_remove_crash_handler();
2693         vmbus_connection.conn_state = DISCONNECTED;
2694         hv_stimer_global_cleanup();
2695         vmbus_disconnect();
2696         hv_remove_vmbus_irq();
2697         for_each_online_cpu(cpu) {
2698                 struct hv_per_cpu_context *hv_cpu
2699                         = per_cpu_ptr(hv_context.cpu_context, cpu);
2700
2701                 tasklet_kill(&hv_cpu->msg_dpc);
2702         }
2703         hv_debug_rm_all_dir();
2704
2705         vmbus_free_channels();
2706         kfree(vmbus_connection.channels);
2707
2708         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
2709                 kmsg_dump_unregister(&hv_kmsg_dumper);
2710                 unregister_die_notifier(&hyperv_die_block);
2711                 atomic_notifier_chain_unregister(&panic_notifier_list,
2712                                                  &hyperv_panic_block);
2713         }
2714
2715         free_page((unsigned long)hv_panic_page);
2716         unregister_sysctl_table(hv_ctl_table_hdr);
2717         hv_ctl_table_hdr = NULL;
2718         bus_unregister(&hv_bus);
2719
2720         cpuhp_remove_state(hyperv_cpuhp_online);
2721         hv_synic_free();
2722         acpi_bus_unregister_driver(&vmbus_acpi_driver);
2723 }
2724
2725
2726 MODULE_LICENSE("GPL");
2727 MODULE_DESCRIPTION("Microsoft Hyper-V VMBus Driver");
2728
2729 subsys_initcall(hv_acpi_init);
2730 module_exit(vmbus_exit);