drivers/hv/vmbus_drv.c

   1 /*
   2  * Copyright (c) 2009, Microsoft Corporation.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms and conditions of the GNU General Public License,
   6  * version 2, as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  11  * more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along with
  14  * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
  15  * Place - Suite 330, Boston, MA 02111-1307 USA.
  16  *
  17  * Authors:
  18  *   Haiyang Zhang <haiyangz@microsoft.com>
  19  *   Hank Janssen  <hjanssen@microsoft.com>
  20  *   K. Y. Srinivasan <kys@microsoft.com>
  21  *
  22  */
  23 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  24
  25 #include <linux/init.h>
  26 #include <linux/module.h>
  27 #include <linux/device.h>
  28 #include <linux/interrupt.h>
  29 #include <linux/sysctl.h>
  30 #include <linux/slab.h>
  31 #include <linux/acpi.h>
  32 #include <linux/completion.h>
  33 #include <linux/hyperv.h>
  34 #include <linux/kernel_stat.h>
  35 #include <linux/clockchips.h>
  36 #include <linux/cpu.h>
  37 #include <asm/hyperv.h>
  38 #include <asm/hypervisor.h>
  39 #include <asm/mshyperv.h>
  40 #include <linux/notifier.h>
  41 #include <linux/ptrace.h>
  42 #include <linux/screen_info.h>
  43 #include <linux/kdebug.h>
  44 #include <linux/efi.h>
  45 #include <linux/random.h>
  46 #include "hyperv_vmbus.h"
  47
  48 static struct acpi_device  *hv_acpi_dev;
  49
  50 static struct completion probe_event;
  51
  52
  53 static void hyperv_report_panic(struct pt_regs *regs)
  54 {
  55         static bool panic_reported;
  56
  57         /*
  58          * We prefer to report panic on 'die' chain as we have proper
  59          * registers to report, but if we miss it (e.g. on BUG()) we need
  60          * to report it on 'panic'.
  61          */
  62         if (panic_reported)
  63                 return;
  64         panic_reported = true;
  65
  66         wrmsrl(HV_X64_MSR_CRASH_P0, regs->ip);
  67         wrmsrl(HV_X64_MSR_CRASH_P1, regs->ax);
  68         wrmsrl(HV_X64_MSR_CRASH_P2, regs->bx);
  69         wrmsrl(HV_X64_MSR_CRASH_P3, regs->cx);
  70         wrmsrl(HV_X64_MSR_CRASH_P4, regs->dx);
  71
  72         /*
  73          * Let Hyper-V know there is crash data available
  74          */
  75         wrmsrl(HV_X64_MSR_CRASH_CTL, HV_CRASH_CTL_CRASH_NOTIFY);
  76 }
  77
  78 static int hyperv_panic_event(struct notifier_block *nb, unsigned long val,
  79                               void *args)
  80 {
  81         struct pt_regs *regs;
  82
  83         regs = current_pt_regs();
  84
  85         hyperv_report_panic(regs);
  86         return NOTIFY_DONE;
  87 }
  88
  89 static int hyperv_die_event(struct notifier_block *nb, unsigned long val,
  90                             void *args)
  91 {
  92         struct die_args *die = (struct die_args *)args;
  93         struct pt_regs *regs = die->regs;
  94
  95         hyperv_report_panic(regs);
  96         return NOTIFY_DONE;
  97 }
  98
  99 static struct notifier_block hyperv_die_block = {
 100         .notifier_call = hyperv_die_event,
 101 };
 102 static struct notifier_block hyperv_panic_block = {
 103         .notifier_call = hyperv_panic_event,
 104 };
 105
 106 static const char *fb_mmio_name = "fb_range";
 107 static struct resource *fb_mmio;
 108 static struct resource *hyperv_mmio;
 109 static DEFINE_SEMAPHORE(hyperv_mmio_lock);
 110
 111 static int vmbus_exists(void)
 112 {
 113         if (hv_acpi_dev == NULL)
 114                 return -ENODEV;
 115
 116         return 0;
 117 }
 118
 119 #define VMBUS_ALIAS_LEN ((sizeof((struct hv_vmbus_device_id *)0)->guid) * 2)
 120 static void print_alias_name(struct hv_device *hv_dev, char *alias_name)
 121 {
 122         int i;
 123         for (i = 0; i < VMBUS_ALIAS_LEN; i += 2)
 124                 sprintf(&alias_name[i], "%02x", hv_dev->dev_type.b[i/2]);
 125 }
 126
 127 static u8 channel_monitor_group(struct vmbus_channel *channel)
 128 {
 129         return (u8)channel->offermsg.monitorid / 32;
 130 }
 131
 132 static u8 channel_monitor_offset(struct vmbus_channel *channel)
 133 {
 134         return (u8)channel->offermsg.monitorid % 32;
 135 }
 136
 137 static u32 channel_pending(struct vmbus_channel *channel,
 138                            struct hv_monitor_page *monitor_page)
 139 {
 140         u8 monitor_group = channel_monitor_group(channel);
 141         return monitor_page->trigger_group[monitor_group].pending;
 142 }
 143
 144 static u32 channel_latency(struct vmbus_channel *channel,
 145                            struct hv_monitor_page *monitor_page)
 146 {
 147         u8 monitor_group = channel_monitor_group(channel);
 148         u8 monitor_offset = channel_monitor_offset(channel);
 149         return monitor_page->latency[monitor_group][monitor_offset];
 150 }
 151
 152 static u32 channel_conn_id(struct vmbus_channel *channel,
 153                            struct hv_monitor_page *monitor_page)
 154 {
 155         u8 monitor_group = channel_monitor_group(channel);
 156         u8 monitor_offset = channel_monitor_offset(channel);
 157         return monitor_page->parameter[monitor_group][monitor_offset].connectionid.u.id;
 158 }
 159
 160 static ssize_t id_show(struct device *dev, struct device_attribute *dev_attr,
 161                        char *buf)
 162 {
 163         struct hv_device *hv_dev = device_to_hv_device(dev);
 164
 165         if (!hv_dev->channel)
 166                 return -ENODEV;
 167         return sprintf(buf, "%d\n", hv_dev->channel->offermsg.child_relid);
 168 }
 169 static DEVICE_ATTR_RO(id);
 170
 171 static ssize_t state_show(struct device *dev, struct device_attribute *dev_attr,
 172                           char *buf)
 173 {
 174         struct hv_device *hv_dev = device_to_hv_device(dev);
 175
 176         if (!hv_dev->channel)
 177                 return -ENODEV;
 178         return sprintf(buf, "%d\n", hv_dev->channel->state);
 179 }
 180 static DEVICE_ATTR_RO(state);
 181
 182 static ssize_t monitor_id_show(struct device *dev,
 183                                struct device_attribute *dev_attr, char *buf)
 184 {
 185         struct hv_device *hv_dev = device_to_hv_device(dev);
 186
 187         if (!hv_dev->channel)
 188                 return -ENODEV;
 189         return sprintf(buf, "%d\n", hv_dev->channel->offermsg.monitorid);
 190 }
 191 static DEVICE_ATTR_RO(monitor_id);
 192
 193 static ssize_t class_id_show(struct device *dev,
 194                                struct device_attribute *dev_attr, char *buf)
 195 {
 196         struct hv_device *hv_dev = device_to_hv_device(dev);
 197
 198         if (!hv_dev->channel)
 199                 return -ENODEV;
 200         return sprintf(buf, "{%pUl}\n",
 201                        hv_dev->channel->offermsg.offer.if_type.b);
 202 }
 203 static DEVICE_ATTR_RO(class_id);
 204
 205 static ssize_t device_id_show(struct device *dev,
 206                               struct device_attribute *dev_attr, char *buf)
 207 {
 208         struct hv_device *hv_dev = device_to_hv_device(dev);
 209
 210         if (!hv_dev->channel)
 211                 return -ENODEV;
 212         return sprintf(buf, "{%pUl}\n",
 213                        hv_dev->channel->offermsg.offer.if_instance.b);
 214 }
 215 static DEVICE_ATTR_RO(device_id);
 216
 217 static ssize_t modalias_show(struct device *dev,
 218                              struct device_attribute *dev_attr, char *buf)
 219 {
 220         struct hv_device *hv_dev = device_to_hv_device(dev);
 221         char alias_name[VMBUS_ALIAS_LEN + 1];
 222
 223         print_alias_name(hv_dev, alias_name);
 224         return sprintf(buf, "vmbus:%s\n", alias_name);
 225 }
 226 static DEVICE_ATTR_RO(modalias);
 227
 228 static ssize_t server_monitor_pending_show(struct device *dev,
 229                                            struct device_attribute *dev_attr,
 230                                            char *buf)
 231 {
 232         struct hv_device *hv_dev = device_to_hv_device(dev);
 233
 234         if (!hv_dev->channel)
 235                 return -ENODEV;
 236         return sprintf(buf, "%d\n",
 237                        channel_pending(hv_dev->channel,
 238                                        vmbus_connection.monitor_pages[1]));
 239 }
 240 static DEVICE_ATTR_RO(server_monitor_pending);
 241
 242 static ssize_t client_monitor_pending_show(struct device *dev,
 243                                            struct device_attribute *dev_attr,
 244                                            char *buf)
 245 {
 246         struct hv_device *hv_dev = device_to_hv_device(dev);
 247
 248         if (!hv_dev->channel)
 249                 return -ENODEV;
 250         return sprintf(buf, "%d\n",
 251                        channel_pending(hv_dev->channel,
 252                                        vmbus_connection.monitor_pages[1]));
 253 }
 254 static DEVICE_ATTR_RO(client_monitor_pending);
 255
 256 static ssize_t server_monitor_latency_show(struct device *dev,
 257                                            struct device_attribute *dev_attr,
 258                                            char *buf)
 259 {
 260         struct hv_device *hv_dev = device_to_hv_device(dev);
 261
 262         if (!hv_dev->channel)
 263                 return -ENODEV;
 264         return sprintf(buf, "%d\n",
 265                        channel_latency(hv_dev->channel,
 266                                        vmbus_connection.monitor_pages[0]));
 267 }
 268 static DEVICE_ATTR_RO(server_monitor_latency);
 269
 270 static ssize_t client_monitor_latency_show(struct device *dev,
 271                                            struct device_attribute *dev_attr,
 272                                            char *buf)
 273 {
 274         struct hv_device *hv_dev = device_to_hv_device(dev);
 275
 276         if (!hv_dev->channel)
 277                 return -ENODEV;
 278         return sprintf(buf, "%d\n",
 279                        channel_latency(hv_dev->channel,
 280                                        vmbus_connection.monitor_pages[1]));
 281 }
 282 static DEVICE_ATTR_RO(client_monitor_latency);
 283
 284 static ssize_t server_monitor_conn_id_show(struct device *dev,
 285                                            struct device_attribute *dev_attr,
 286                                            char *buf)
 287 {
 288         struct hv_device *hv_dev = device_to_hv_device(dev);
 289
 290         if (!hv_dev->channel)
 291                 return -ENODEV;
 292         return sprintf(buf, "%d\n",
 293                        channel_conn_id(hv_dev->channel,
 294                                        vmbus_connection.monitor_pages[0]));
 295 }
 296 static DEVICE_ATTR_RO(server_monitor_conn_id);
 297
 298 static ssize_t client_monitor_conn_id_show(struct device *dev,
 299                                            struct device_attribute *dev_attr,
 300                                            char *buf)
 301 {
 302         struct hv_device *hv_dev = device_to_hv_device(dev);
 303
 304         if (!hv_dev->channel)
 305                 return -ENODEV;
 306         return sprintf(buf, "%d\n",
 307                        channel_conn_id(hv_dev->channel,
 308                                        vmbus_connection.monitor_pages[1]));
 309 }
 310 static DEVICE_ATTR_RO(client_monitor_conn_id);
 311
 312 static ssize_t out_intr_mask_show(struct device *dev,
 313                                   struct device_attribute *dev_attr, char *buf)
 314 {
 315         struct hv_device *hv_dev = device_to_hv_device(dev);
 316         struct hv_ring_buffer_debug_info outbound;
 317
 318         if (!hv_dev->channel)
 319                 return -ENODEV;
 320         if (hv_dev->channel->state != CHANNEL_OPENED_STATE)
 321                 return -EINVAL;
 322         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
 323         return sprintf(buf, "%d\n", outbound.current_interrupt_mask);
 324 }
 325 static DEVICE_ATTR_RO(out_intr_mask);
 326
 327 static ssize_t out_read_index_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
 333         if (!hv_dev->channel)
 334                 return -ENODEV;
 335         if (hv_dev->channel->state != CHANNEL_OPENED_STATE)
 336                 return -EINVAL;
 337         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
 338         return sprintf(buf, "%d\n", outbound.current_read_index);
 339 }
 340 static DEVICE_ATTR_RO(out_read_index);
 341
 342 static ssize_t out_write_index_show(struct device *dev,
 343                                     struct device_attribute *dev_attr,
 344                                     char *buf)
 345 {
 346         struct hv_device *hv_dev = device_to_hv_device(dev);
 347         struct hv_ring_buffer_debug_info outbound;
 348
 349         if (!hv_dev->channel)
 350                 return -ENODEV;
 351         if (hv_dev->channel->state != CHANNEL_OPENED_STATE)
 352                 return -EINVAL;
 353         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
 354         return sprintf(buf, "%d\n", outbound.current_write_index);
 355 }
 356 static DEVICE_ATTR_RO(out_write_index);
 357
 358 static ssize_t out_read_bytes_avail_show(struct device *dev,
 359                                          struct device_attribute *dev_attr,
 360                                          char *buf)
 361 {
 362         struct hv_device *hv_dev = device_to_hv_device(dev);
 363         struct hv_ring_buffer_debug_info outbound;
 364
 365         if (!hv_dev->channel)
 366                 return -ENODEV;
 367         if (hv_dev->channel->state != CHANNEL_OPENED_STATE)
 368                 return -EINVAL;
 369         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
 370         return sprintf(buf, "%d\n", outbound.bytes_avail_toread);
 371 }
 372 static DEVICE_ATTR_RO(out_read_bytes_avail);
 373
 374 static ssize_t out_write_bytes_avail_show(struct device *dev,
 375                                           struct device_attribute *dev_attr,
 376                                           char *buf)
 377 {
 378         struct hv_device *hv_dev = device_to_hv_device(dev);
 379         struct hv_ring_buffer_debug_info outbound;
 380
 381         if (!hv_dev->channel)
 382                 return -ENODEV;
 383         if (hv_dev->channel->state != CHANNEL_OPENED_STATE)
 384                 return -EINVAL;
 385         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
 386         return sprintf(buf, "%d\n", outbound.bytes_avail_towrite);
 387 }
 388 static DEVICE_ATTR_RO(out_write_bytes_avail);
 389
 390 static ssize_t in_intr_mask_show(struct device *dev,
 391                                  struct device_attribute *dev_attr, char *buf)
 392 {
 393         struct hv_device *hv_dev = device_to_hv_device(dev);
 394         struct hv_ring_buffer_debug_info inbound;
 395
 396         if (!hv_dev->channel)
 397                 return -ENODEV;
 398         if (hv_dev->channel->state != CHANNEL_OPENED_STATE)
 399                 return -EINVAL;
 400         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 401         return sprintf(buf, "%d\n", inbound.current_interrupt_mask);
 402 }
 403 static DEVICE_ATTR_RO(in_intr_mask);
 404
 405 static ssize_t in_read_index_show(struct device *dev,
 406                                   struct device_attribute *dev_attr, char *buf)
 407 {
 408         struct hv_device *hv_dev = device_to_hv_device(dev);
 409         struct hv_ring_buffer_debug_info inbound;
 410
 411         if (!hv_dev->channel)
 412                 return -ENODEV;
 413         if (hv_dev->channel->state != CHANNEL_OPENED_STATE)
 414                 return -EINVAL;
 415         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 416         return sprintf(buf, "%d\n", inbound.current_read_index);
 417 }
 418 static DEVICE_ATTR_RO(in_read_index);
 419
 420 static ssize_t in_write_index_show(struct device *dev,
 421                                    struct device_attribute *dev_attr, char *buf)
 422 {
 423         struct hv_device *hv_dev = device_to_hv_device(dev);
 424         struct hv_ring_buffer_debug_info inbound;
 425
 426         if (!hv_dev->channel)
 427                 return -ENODEV;
 428         if (hv_dev->channel->state != CHANNEL_OPENED_STATE)
 429                 return -EINVAL;
 430         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 431         return sprintf(buf, "%d\n", inbound.current_write_index);
 432 }
 433 static DEVICE_ATTR_RO(in_write_index);
 434
 435 static ssize_t in_read_bytes_avail_show(struct device *dev,
 436                                         struct device_attribute *dev_attr,
 437                                         char *buf)
 438 {
 439         struct hv_device *hv_dev = device_to_hv_device(dev);
 440         struct hv_ring_buffer_debug_info inbound;
 441
 442         if (!hv_dev->channel)
 443                 return -ENODEV;
 444         if (hv_dev->channel->state != CHANNEL_OPENED_STATE)
 445                 return -EINVAL;
 446         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 447         return sprintf(buf, "%d\n", inbound.bytes_avail_toread);
 448 }
 449 static DEVICE_ATTR_RO(in_read_bytes_avail);
 450
 451 static ssize_t in_write_bytes_avail_show(struct device *dev,
 452                                          struct device_attribute *dev_attr,
 453                                          char *buf)
 454 {
 455         struct hv_device *hv_dev = device_to_hv_device(dev);
 456         struct hv_ring_buffer_debug_info inbound;
 457
 458         if (!hv_dev->channel)
 459                 return -ENODEV;
 460         if (hv_dev->channel->state != CHANNEL_OPENED_STATE)
 461                 return -EINVAL;
 462         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 463         return sprintf(buf, "%d\n", inbound.bytes_avail_towrite);
 464 }
 465 static DEVICE_ATTR_RO(in_write_bytes_avail);
 466
 467 static ssize_t channel_vp_mapping_show(struct device *dev,
 468                                        struct device_attribute *dev_attr,
 469                                        char *buf)
 470 {
 471         struct hv_device *hv_dev = device_to_hv_device(dev);
 472         struct vmbus_channel *channel = hv_dev->channel, *cur_sc;
 473         unsigned long flags;
 474         int buf_size = PAGE_SIZE, n_written, tot_written;
 475         struct list_head *cur;
 476
 477         if (!channel)
 478                 return -ENODEV;
 479
 480         tot_written = snprintf(buf, buf_size, "%u:%u\n",
 481                 channel->offermsg.child_relid, channel->target_cpu);
 482
 483         spin_lock_irqsave(&channel->lock, flags);
 484
 485         list_for_each(cur, &channel->sc_list) {
 486                 if (tot_written >= buf_size - 1)
 487                         break;
 488
 489                 cur_sc = list_entry(cur, struct vmbus_channel, sc_list);
 490                 n_written = scnprintf(buf + tot_written,
 491                                      buf_size - tot_written,
 492                                      "%u:%u\n",
 493                                      cur_sc->offermsg.child_relid,
 494                                      cur_sc->target_cpu);
 495                 tot_written += n_written;
 496         }
 497
 498         spin_unlock_irqrestore(&channel->lock, flags);
 499
 500         return tot_written;
 501 }
 502 static DEVICE_ATTR_RO(channel_vp_mapping);
 503
 504 static ssize_t vendor_show(struct device *dev,
 505                            struct device_attribute *dev_attr,
 506                            char *buf)
 507 {
 508         struct hv_device *hv_dev = device_to_hv_device(dev);
 509         return sprintf(buf, "0x%x\n", hv_dev->vendor_id);
 510 }
 511 static DEVICE_ATTR_RO(vendor);
 512
 513 static ssize_t device_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         return sprintf(buf, "0x%x\n", hv_dev->device_id);
 519 }
 520 static DEVICE_ATTR_RO(device);
 521
 522 /* Set up per device attributes in /sys/bus/vmbus/devices/<bus device> */
 523 static struct attribute *vmbus_attrs[] = {
 524         &dev_attr_id.attr,
 525         &dev_attr_state.attr,
 526         &dev_attr_monitor_id.attr,
 527         &dev_attr_class_id.attr,
 528         &dev_attr_device_id.attr,
 529         &dev_attr_modalias.attr,
 530         &dev_attr_server_monitor_pending.attr,
 531         &dev_attr_client_monitor_pending.attr,
 532         &dev_attr_server_monitor_latency.attr,
 533         &dev_attr_client_monitor_latency.attr,
 534         &dev_attr_server_monitor_conn_id.attr,
 535         &dev_attr_client_monitor_conn_id.attr,
 536         &dev_attr_out_intr_mask.attr,
 537         &dev_attr_out_read_index.attr,
 538         &dev_attr_out_write_index.attr,
 539         &dev_attr_out_read_bytes_avail.attr,
 540         &dev_attr_out_write_bytes_avail.attr,
 541         &dev_attr_in_intr_mask.attr,
 542         &dev_attr_in_read_index.attr,
 543         &dev_attr_in_write_index.attr,
 544         &dev_attr_in_read_bytes_avail.attr,
 545         &dev_attr_in_write_bytes_avail.attr,
 546         &dev_attr_channel_vp_mapping.attr,
 547         &dev_attr_vendor.attr,
 548         &dev_attr_device.attr,
 549         NULL,
 550 };
 551 ATTRIBUTE_GROUPS(vmbus);
 552
 553 /*
 554  * vmbus_uevent - add uevent for our device
 555  *
 556  * This routine is invoked when a device is added or removed on the vmbus to
 557  * generate a uevent to udev in the userspace. The udev will then look at its
 558  * rule and the uevent generated here to load the appropriate driver
 559  *
 560  * The alias string will be of the form vmbus:guid where guid is the string
 561  * representation of the device guid (each byte of the guid will be
 562  * represented with two hex characters.
 563  */
 564 static int vmbus_uevent(struct device *device, struct kobj_uevent_env *env)
 565 {
 566         struct hv_device *dev = device_to_hv_device(device);
 567         int ret;
 568         char alias_name[VMBUS_ALIAS_LEN + 1];
 569
 570         print_alias_name(dev, alias_name);
 571         ret = add_uevent_var(env, "MODALIAS=vmbus:%s", alias_name);
 572         return ret;
 573 }
 574
 575 static const uuid_le null_guid;
 576
 577 static inline bool is_null_guid(const uuid_le *guid)
 578 {
 579         if (uuid_le_cmp(*guid, null_guid))
 580                 return false;
 581         return true;
 582 }
 583
 584 /*
 585  * Return a matching hv_vmbus_device_id pointer.
 586  * If there is no match, return NULL.
 587  */
 588 static const struct hv_vmbus_device_id *hv_vmbus_get_id(
 589                                         const struct hv_vmbus_device_id *id,
 590                                         const uuid_le *guid)
 591 {
 592         for (; !is_null_guid(&id->guid); id++)
 593                 if (!uuid_le_cmp(id->guid, *guid))
 594                         return id;
 595
 596         return NULL;
 597 }
 598
 599
 600
 601 /*
 602  * vmbus_match - Attempt to match the specified device to the specified driver
 603  */
 604 static int vmbus_match(struct device *device, struct device_driver *driver)
 605 {
 606         struct hv_driver *drv = drv_to_hv_drv(driver);
 607         struct hv_device *hv_dev = device_to_hv_device(device);
 608
 609         /* The hv_sock driver handles all hv_sock offers. */
 610         if (is_hvsock_channel(hv_dev->channel))
 611                 return drv->hvsock;
 612
 613         if (hv_vmbus_get_id(drv->id_table, &hv_dev->dev_type))
 614                 return 1;
 615
 616         return 0;
 617 }
 618
 619 /*
 620  * vmbus_probe - Add the new vmbus's child device
 621  */
 622 static int vmbus_probe(struct device *child_device)
 623 {
 624         int ret = 0;
 625         struct hv_driver *drv =
 626                         drv_to_hv_drv(child_device->driver);
 627         struct hv_device *dev = device_to_hv_device(child_device);
 628         const struct hv_vmbus_device_id *dev_id;
 629
 630         dev_id = hv_vmbus_get_id(drv->id_table, &dev->dev_type);
 631         if (drv->probe) {
 632                 ret = drv->probe(dev, dev_id);
 633                 if (ret != 0)
 634                         pr_err("probe failed for device %s (%d)\n",
 635                                dev_name(child_device), ret);
 636
 637         } else {
 638                 pr_err("probe not set for driver %s\n",
 639                        dev_name(child_device));
 640                 ret = -ENODEV;
 641         }
 642         return ret;
 643 }
 644
 645 /*
 646  * vmbus_remove - Remove a vmbus device
 647  */
 648 static int vmbus_remove(struct device *child_device)
 649 {
 650         struct hv_driver *drv;
 651         struct hv_device *dev = device_to_hv_device(child_device);
 652
 653         if (child_device->driver) {
 654                 drv = drv_to_hv_drv(child_device->driver);
 655                 if (drv->remove)
 656                         drv->remove(dev);
 657         }
 658
 659         return 0;
 660 }
 661
 662
 663 /*
 664  * vmbus_shutdown - Shutdown a vmbus device
 665  */
 666 static void vmbus_shutdown(struct device *child_device)
 667 {
 668         struct hv_driver *drv;
 669         struct hv_device *dev = device_to_hv_device(child_device);
 670
 671
 672         /* The device may not be attached yet */
 673         if (!child_device->driver)
 674                 return;
 675
 676         drv = drv_to_hv_drv(child_device->driver);
 677
 678         if (drv->shutdown)
 679                 drv->shutdown(dev);
 680
 681         return;
 682 }
 683
 684
 685 /*
 686  * vmbus_device_release - Final callback release of the vmbus child device
 687  */
 688 static void vmbus_device_release(struct device *device)
 689 {
 690         struct hv_device *hv_dev = device_to_hv_device(device);
 691         struct vmbus_channel *channel = hv_dev->channel;
 692
 693         hv_process_channel_removal(channel,
 694                                    channel->offermsg.child_relid);
 695         kfree(hv_dev);
 696
 697 }
 698
 699 /* The one and only one */
 700 static struct bus_type  hv_bus = {
 701         .name =         "vmbus",
 702         .match =                vmbus_match,
 703         .shutdown =             vmbus_shutdown,
 704         .remove =               vmbus_remove,
 705         .probe =                vmbus_probe,
 706         .uevent =               vmbus_uevent,
 707         .dev_groups =           vmbus_groups,
 708 };
 709
 710 struct onmessage_work_context {
 711         struct work_struct work;
 712         struct hv_message msg;
 713 };
 714
 715 static void vmbus_onmessage_work(struct work_struct *work)
 716 {
 717         struct onmessage_work_context *ctx;
 718
 719         /* Do not process messages if we're in DISCONNECTED state */
 720         if (vmbus_connection.conn_state == DISCONNECTED)
 721                 return;
 722
 723         ctx = container_of(work, struct onmessage_work_context,
 724                            work);
 725         vmbus_onmessage(&ctx->msg);
 726         kfree(ctx);
 727 }
 728
 729 static void hv_process_timer_expiration(struct hv_message *msg, int cpu)
 730 {
 731         struct clock_event_device *dev = hv_context.clk_evt[cpu];
 732
 733         if (dev->event_handler)
 734                 dev->event_handler(dev);
 735
 736         vmbus_signal_eom(msg, HVMSG_TIMER_EXPIRED);
 737 }
 738
 739 void vmbus_on_msg_dpc(unsigned long data)
 740 {
 741         int cpu = smp_processor_id();
 742         void *page_addr = hv_context.synic_message_page[cpu];
 743         struct hv_message *msg = (struct hv_message *)page_addr +
 744                                   VMBUS_MESSAGE_SINT;
 745         struct vmbus_channel_message_header *hdr;
 746         struct vmbus_channel_message_table_entry *entry;
 747         struct onmessage_work_context *ctx;
 748         u32 message_type = msg->header.message_type;
 749
 750         if (message_type == HVMSG_NONE)
 751                 /* no msg */
 752                 return;
 753
 754         hdr = (struct vmbus_channel_message_header *)msg->u.payload;
 755
 756         if (hdr->msgtype >= CHANNELMSG_COUNT) {
 757                 WARN_ONCE(1, "unknown msgtype=%d\n", hdr->msgtype);
 758                 goto msg_handled;
 759         }
 760
 761         entry = &channel_message_table[hdr->msgtype];
 762         if (entry->handler_type == VMHT_BLOCKING) {
 763                 ctx = kmalloc(sizeof(*ctx), GFP_ATOMIC);
 764                 if (ctx == NULL)
 765                         return;
 766
 767                 INIT_WORK(&ctx->work, vmbus_onmessage_work);
 768                 memcpy(&ctx->msg, msg, sizeof(*msg));
 769
 770                 queue_work(vmbus_connection.work_queue, &ctx->work);
 771         } else
 772                 entry->message_handler(hdr);
 773
 774 msg_handled:
 775         vmbus_signal_eom(msg, message_type);
 776 }
 777
 778 static void vmbus_isr(void)
 779 {
 780         int cpu = smp_processor_id();
 781         void *page_addr;
 782         struct hv_message *msg;
 783         union hv_synic_event_flags *event;
 784         bool handled = false;
 785
 786         page_addr = hv_context.synic_event_page[cpu];
 787         if (page_addr == NULL)
 788                 return;
 789
 790         event = (union hv_synic_event_flags *)page_addr +
 791                                          VMBUS_MESSAGE_SINT;
 792         /*
 793          * Check for events before checking for messages. This is the order
 794          * in which events and messages are checked in Windows guests on
 795          * Hyper-V, and the Windows team suggested we do the same.
 796          */
 797
 798         if ((vmbus_proto_version == VERSION_WS2008) ||
 799                 (vmbus_proto_version == VERSION_WIN7)) {
 800
 801                 /* Since we are a child, we only need to check bit 0 */
 802                 if (sync_test_and_clear_bit(0,
 803                         (unsigned long *) &event->flags32[0])) {
 804                         handled = true;
 805                 }
 806         } else {
 807                 /*
 808                  * Our host is win8 or above. The signaling mechanism
 809                  * has changed and we can directly look at the event page.
 810                  * If bit n is set then we have an interrup on the channel
 811                  * whose id is n.
 812                  */
 813                 handled = true;
 814         }
 815
 816         if (handled)
 817                 tasklet_schedule(hv_context.event_dpc[cpu]);
 818
 819
 820         page_addr = hv_context.synic_message_page[cpu];
 821         msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
 822
 823         /* Check if there are actual msgs to be processed */
 824         if (msg->header.message_type != HVMSG_NONE) {
 825                 if (msg->header.message_type == HVMSG_TIMER_EXPIRED)
 826                         hv_process_timer_expiration(msg, cpu);
 827                 else
 828                         tasklet_schedule(hv_context.msg_dpc[cpu]);
 829         }
 830
 831         add_interrupt_randomness(HYPERVISOR_CALLBACK_VECTOR, 0);
 832 }
 833
 834
 835 /*
 836  * vmbus_bus_init -Main vmbus driver initialization routine.
 837  *
 838  * Here, we
 839  *      - initialize the vmbus driver context
 840  *      - invoke the vmbus hv main init routine
 841  *      - retrieve the channel offers
 842  */
 843 static int vmbus_bus_init(void)
 844 {
 845         int ret;
 846
 847         /* Hypervisor initialization...setup hypercall page..etc */
 848         ret = hv_init();
 849         if (ret != 0) {
 850                 pr_err("Unable to initialize the hypervisor - 0x%x\n", ret);
 851                 return ret;
 852         }
 853
 854         ret = bus_register(&hv_bus);
 855         if (ret)
 856                 goto err_cleanup;
 857
 858         hv_setup_vmbus_irq(vmbus_isr);
 859
 860         ret = hv_synic_alloc();
 861         if (ret)
 862                 goto err_alloc;
 863         /*
 864          * Initialize the per-cpu interrupt state and
 865          * connect to the host.
 866          */
 867         on_each_cpu(hv_synic_init, NULL, 1);
 868         ret = vmbus_connect();
 869         if (ret)
 870                 goto err_connect;
 871
 872         if (vmbus_proto_version > VERSION_WIN7)
 873                 cpu_hotplug_disable();
 874
 875         /*
 876          * Only register if the crash MSRs are available
 877          */
 878         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
 879                 register_die_notifier(&hyperv_die_block);
 880                 atomic_notifier_chain_register(&panic_notifier_list,
 881                                                &hyperv_panic_block);
 882         }
 883
 884         vmbus_request_offers();
 885
 886         return 0;
 887
 888 err_connect:
 889         on_each_cpu(hv_synic_cleanup, NULL, 1);
 890 err_alloc:
 891         hv_synic_free();
 892         hv_remove_vmbus_irq();
 893
 894         bus_unregister(&hv_bus);
 895
 896 err_cleanup:
 897         hv_cleanup(false);
 898
 899         return ret;
 900 }
 901
 902 /**
 903  * __vmbus_child_driver_register() - Register a vmbus's driver
 904  * @hv_driver: Pointer to driver structure you want to register
 905  * @owner: owner module of the drv
 906  * @mod_name: module name string
 907  *
 908  * Registers the given driver with Linux through the 'driver_register()' call
 909  * and sets up the hyper-v vmbus handling for this driver.
 910  * It will return the state of the 'driver_register()' call.
 911  *
 912  */
 913 int __vmbus_driver_register(struct hv_driver *hv_driver, struct module *owner, const char *mod_name)
 914 {
 915         int ret;
 916
 917         pr_info("registering driver %s\n", hv_driver->name);
 918
 919         ret = vmbus_exists();
 920         if (ret < 0)
 921                 return ret;
 922
 923         hv_driver->driver.name = hv_driver->name;
 924         hv_driver->driver.owner = owner;
 925         hv_driver->driver.mod_name = mod_name;
 926         hv_driver->driver.bus = &hv_bus;
 927
 928         ret = driver_register(&hv_driver->driver);
 929
 930         return ret;
 931 }
 932 EXPORT_SYMBOL_GPL(__vmbus_driver_register);
 933
 934 /**
 935  * vmbus_driver_unregister() - Unregister a vmbus's driver
 936  * @hv_driver: Pointer to driver structure you want to
 937  *             un-register
 938  *
 939  * Un-register the given driver that was previous registered with a call to
 940  * vmbus_driver_register()
 941  */
 942 void vmbus_driver_unregister(struct hv_driver *hv_driver)
 943 {
 944         pr_info("unregistering driver %s\n", hv_driver->name);
 945
 946         if (!vmbus_exists())
 947                 driver_unregister(&hv_driver->driver);
 948 }
 949 EXPORT_SYMBOL_GPL(vmbus_driver_unregister);
 950
 951 /*
 952  * vmbus_device_create - Creates and registers a new child device
 953  * on the vmbus.
 954  */
 955 struct hv_device *vmbus_device_create(const uuid_le *type,
 956                                       const uuid_le *instance,
 957                                       struct vmbus_channel *channel)
 958 {
 959         struct hv_device *child_device_obj;
 960
 961         child_device_obj = kzalloc(sizeof(struct hv_device), GFP_KERNEL);
 962         if (!child_device_obj) {
 963                 pr_err("Unable to allocate device object for child device\n");
 964                 return NULL;
 965         }
 966
 967         child_device_obj->channel = channel;
 968         memcpy(&child_device_obj->dev_type, type, sizeof(uuid_le));
 969         memcpy(&child_device_obj->dev_instance, instance,
 970                sizeof(uuid_le));
 971         child_device_obj->vendor_id = 0x1414; /* MSFT vendor ID */
 972
 973
 974         return child_device_obj;
 975 }
 976
 977 /*
 978  * vmbus_device_register - Register the child device
 979  */
 980 int vmbus_device_register(struct hv_device *child_device_obj)
 981 {
 982         int ret = 0;
 983
 984         dev_set_name(&child_device_obj->device, "%pUl",
 985                      child_device_obj->channel->offermsg.offer.if_instance.b);
 986
 987         child_device_obj->device.bus = &hv_bus;
 988         child_device_obj->device.parent = &hv_acpi_dev->dev;
 989         child_device_obj->device.release = vmbus_device_release;
 990
 991         /*
 992          * Register with the LDM. This will kick off the driver/device
 993          * binding...which will eventually call vmbus_match() and vmbus_probe()
 994          */
 995         ret = device_register(&child_device_obj->device);
 996
 997         if (ret)
 998                 pr_err("Unable to register child device\n");
 999         else
1000                 pr_debug("child device %s registered\n",
1001                         dev_name(&child_device_obj->device));
1002
1003         return ret;
1004 }
1005
1006 /*
1007  * vmbus_device_unregister - Remove the specified child device
1008  * from the vmbus.
1009  */
1010 void vmbus_device_unregister(struct hv_device *device_obj)
1011 {
1012         pr_debug("child device %s unregistered\n",
1013                 dev_name(&device_obj->device));
1014
1015         /*
1016          * Kick off the process of unregistering the device.
1017          * This will call vmbus_remove() and eventually vmbus_device_release()
1018          */
1019         device_unregister(&device_obj->device);
1020 }
1021
1022
1023 /*
1024  * VMBUS is an acpi enumerated device. Get the information we
1025  * need from DSDT.
1026  */
1027 #define VTPM_BASE_ADDRESS 0xfed40000
1028 static acpi_status vmbus_walk_resources(struct acpi_resource *res, void *ctx)
1029 {
1030         resource_size_t start = 0;
1031         resource_size_t end = 0;
1032         struct resource *new_res;
1033         struct resource **old_res = &hyperv_mmio;
1034         struct resource **prev_res = NULL;
1035
1036         switch (res->type) {
1037
1038         /*
1039          * "Address" descriptors are for bus windows. Ignore
1040          * "memory" descriptors, which are for registers on
1041          * devices.
1042          */
1043         case ACPI_RESOURCE_TYPE_ADDRESS32:
1044                 start = res->data.address32.address.minimum;
1045                 end = res->data.address32.address.maximum;
1046                 break;
1047
1048         case ACPI_RESOURCE_TYPE_ADDRESS64:
1049                 start = res->data.address64.address.minimum;
1050                 end = res->data.address64.address.maximum;
1051                 break;
1052
1053         default:
1054                 /* Unused resource type */
1055                 return AE_OK;
1056
1057         }
1058         /*
1059          * Ignore ranges that are below 1MB, as they're not
1060          * necessary or useful here.
1061          */
1062         if (end < 0x100000)
1063                 return AE_OK;
1064
1065         new_res = kzalloc(sizeof(*new_res), GFP_ATOMIC);
1066         if (!new_res)
1067                 return AE_NO_MEMORY;
1068
1069         /* If this range overlaps the virtual TPM, truncate it. */
1070         if (end > VTPM_BASE_ADDRESS && start < VTPM_BASE_ADDRESS)
1071                 end = VTPM_BASE_ADDRESS;
1072
1073         new_res->name = "hyperv mmio";
1074         new_res->flags = IORESOURCE_MEM;
1075         new_res->start = start;
1076         new_res->end = end;
1077
1078         /*
1079          * If two ranges are adjacent, merge them.
1080          */
1081         do {
1082                 if (!*old_res) {
1083                         *old_res = new_res;
1084                         break;
1085                 }
1086
1087                 if (((*old_res)->end + 1) == new_res->start) {
1088                         (*old_res)->end = new_res->end;
1089                         kfree(new_res);
1090                         break;
1091                 }
1092
1093                 if ((*old_res)->start == new_res->end + 1) {
1094                         (*old_res)->start = new_res->start;
1095                         kfree(new_res);
1096                         break;
1097                 }
1098
1099                 if ((*old_res)->start > new_res->end) {
1100                         new_res->sibling = *old_res;
1101                         if (prev_res)
1102                                 (*prev_res)->sibling = new_res;
1103                         *old_res = new_res;
1104                         break;
1105                 }
1106
1107                 prev_res = old_res;
1108                 old_res = &(*old_res)->sibling;
1109
1110         } while (1);
1111
1112         return AE_OK;
1113 }
1114
1115 static int vmbus_acpi_remove(struct acpi_device *device)
1116 {
1117         struct resource *cur_res;
1118         struct resource *next_res;
1119
1120         if (hyperv_mmio) {
1121                 if (fb_mmio) {
1122                         __release_region(hyperv_mmio, fb_mmio->start,
1123                                          resource_size(fb_mmio));
1124                         fb_mmio = NULL;
1125                 }
1126
1127                 for (cur_res = hyperv_mmio; cur_res; cur_res = next_res) {
1128                         next_res = cur_res->sibling;
1129                         kfree(cur_res);
1130                 }
1131         }
1132
1133         return 0;
1134 }
1135
1136 static void vmbus_reserve_fb(void)
1137 {
1138         int size;
1139         /*
1140          * Make a claim for the frame buffer in the resource tree under the
1141          * first node, which will be the one below 4GB.  The length seems to
1142          * be underreported, particularly in a Generation 1 VM.  So start out
1143          * reserving a larger area and make it smaller until it succeeds.
1144          */
1145
1146         if (screen_info.lfb_base) {
1147                 if (efi_enabled(EFI_BOOT))
1148                         size = max_t(__u32, screen_info.lfb_size, 0x800000);
1149                 else
1150                         size = max_t(__u32, screen_info.lfb_size, 0x4000000);
1151
1152                 for (; !fb_mmio && (size >= 0x100000); size >>= 1) {
1153                         fb_mmio = __request_region(hyperv_mmio,
1154                                                    screen_info.lfb_base, size,
1155                                                    fb_mmio_name, 0);
1156                 }
1157         }
1158 }
1159
1160 /**
1161  * vmbus_allocate_mmio() - Pick a memory-mapped I/O range.
1162  * @new:                If successful, supplied a pointer to the
1163  *                      allocated MMIO space.
1164  * @device_obj:         Identifies the caller
1165  * @min:                Minimum guest physical address of the
1166  *                      allocation
1167  * @max:                Maximum guest physical address
1168  * @size:               Size of the range to be allocated
1169  * @align:              Alignment of the range to be allocated
1170  * @fb_overlap_ok:      Whether this allocation can be allowed
1171  *                      to overlap the video frame buffer.
1172  *
1173  * This function walks the resources granted to VMBus by the
1174  * _CRS object in the ACPI namespace underneath the parent
1175  * "bridge" whether that's a root PCI bus in the Generation 1
1176  * case or a Module Device in the Generation 2 case.  It then
1177  * attempts to allocate from the global MMIO pool in a way that
1178  * matches the constraints supplied in these parameters and by
1179  * that _CRS.
1180  *
1181  * Return: 0 on success, -errno on failure
1182  */
1183 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
1184                         resource_size_t min, resource_size_t max,
1185                         resource_size_t size, resource_size_t align,
1186                         bool fb_overlap_ok)
1187 {
1188         struct resource *iter, *shadow;
1189         resource_size_t range_min, range_max, start;
1190         const char *dev_n = dev_name(&device_obj->device);
1191         int retval;
1192
1193         retval = -ENXIO;
1194         down(&hyperv_mmio_lock);
1195
1196         /*
1197          * If overlaps with frame buffers are allowed, then first attempt to
1198          * make the allocation from within the reserved region.  Because it
1199          * is already reserved, no shadow allocation is necessary.
1200          */
1201         if (fb_overlap_ok && fb_mmio && !(min > fb_mmio->end) &&
1202             !(max < fb_mmio->start)) {
1203
1204                 range_min = fb_mmio->start;
1205                 range_max = fb_mmio->end;
1206                 start = (range_min + align - 1) & ~(align - 1);
1207                 for (; start + size - 1 <= range_max; start += align) {
1208                         *new = request_mem_region_exclusive(start, size, dev_n);
1209                         if (*new) {
1210                                 retval = 0;
1211                                 goto exit;
1212                         }
1213                 }
1214         }
1215
1216         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
1217                 if ((iter->start >= max) || (iter->end <= min))
1218                         continue;
1219
1220                 range_min = iter->start;
1221                 range_max = iter->end;
1222                 start = (range_min + align - 1) & ~(align - 1);
1223                 for (; start + size - 1 <= range_max; start += align) {
1224                         shadow = __request_region(iter, start, size, NULL,
1225                                                   IORESOURCE_BUSY);
1226                         if (!shadow)
1227                                 continue;
1228
1229                         *new = request_mem_region_exclusive(start, size, dev_n);
1230                         if (*new) {
1231                                 shadow->name = (char *)*new;
1232                                 retval = 0;
1233                                 goto exit;
1234                         }
1235
1236                         __release_region(iter, start, size);
1237                 }
1238         }
1239
1240 exit:
1241         up(&hyperv_mmio_lock);
1242         return retval;
1243 }
1244 EXPORT_SYMBOL_GPL(vmbus_allocate_mmio);
1245
1246 /**
1247  * vmbus_free_mmio() - Free a memory-mapped I/O range.
1248  * @start:              Base address of region to release.
1249  * @size:               Size of the range to be allocated
1250  *
1251  * This function releases anything requested by
1252  * vmbus_mmio_allocate().
1253  */
1254 void vmbus_free_mmio(resource_size_t start, resource_size_t size)
1255 {
1256         struct resource *iter;
1257
1258         down(&hyperv_mmio_lock);
1259         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
1260                 if ((iter->start >= start + size) || (iter->end <= start))
1261                         continue;
1262
1263                 __release_region(iter, start, size);
1264         }
1265         release_mem_region(start, size);
1266         up(&hyperv_mmio_lock);
1267
1268 }
1269 EXPORT_SYMBOL_GPL(vmbus_free_mmio);
1270
1271 /**
1272  * vmbus_cpu_number_to_vp_number() - Map CPU to VP.
1273  * @cpu_number: CPU number in Linux terms
1274  *
1275  * This function returns the mapping between the Linux processor
1276  * number and the hypervisor's virtual processor number, useful
1277  * in making hypercalls and such that talk about specific
1278  * processors.
1279  *
1280  * Return: Virtual processor number in Hyper-V terms
1281  */
1282 int vmbus_cpu_number_to_vp_number(int cpu_number)
1283 {
1284         return hv_context.vp_index[cpu_number];
1285 }
1286 EXPORT_SYMBOL_GPL(vmbus_cpu_number_to_vp_number);
1287
1288 static int vmbus_acpi_add(struct acpi_device *device)
1289 {
1290         acpi_status result;
1291         int ret_val = -ENODEV;
1292         struct acpi_device *ancestor;
1293
1294         hv_acpi_dev = device;
1295
1296         result = acpi_walk_resources(device->handle, METHOD_NAME__CRS,
1297                                         vmbus_walk_resources, NULL);
1298
1299         if (ACPI_FAILURE(result))
1300                 goto acpi_walk_err;
1301         /*
1302          * Some ancestor of the vmbus acpi device (Gen1 or Gen2
1303          * firmware) is the VMOD that has the mmio ranges. Get that.
1304          */
1305         for (ancestor = device->parent; ancestor; ancestor = ancestor->parent) {
1306                 result = acpi_walk_resources(ancestor->handle, METHOD_NAME__CRS,
1307                                              vmbus_walk_resources, NULL);
1308
1309                 if (ACPI_FAILURE(result))
1310                         continue;
1311                 if (hyperv_mmio) {
1312                         vmbus_reserve_fb();
1313                         break;
1314                 }
1315         }
1316         ret_val = 0;
1317
1318 acpi_walk_err:
1319         complete(&probe_event);
1320         if (ret_val)
1321                 vmbus_acpi_remove(device);
1322         return ret_val;
1323 }
1324
1325 static const struct acpi_device_id vmbus_acpi_device_ids[] = {
1326         {"VMBUS", 0},
1327         {"VMBus", 0},
1328         {"", 0},
1329 };
1330 MODULE_DEVICE_TABLE(acpi, vmbus_acpi_device_ids);
1331
1332 static struct acpi_driver vmbus_acpi_driver = {
1333         .name = "vmbus",
1334         .ids = vmbus_acpi_device_ids,
1335         .ops = {
1336                 .add = vmbus_acpi_add,
1337                 .remove = vmbus_acpi_remove,
1338         },
1339 };
1340
1341 static void hv_kexec_handler(void)
1342 {
1343         int cpu;
1344
1345         hv_synic_clockevents_cleanup();
1346         vmbus_initiate_unload(false);
1347         for_each_online_cpu(cpu)
1348                 smp_call_function_single(cpu, hv_synic_cleanup, NULL, 1);
1349         hv_cleanup(false);
1350 };
1351
1352 static void hv_crash_handler(struct pt_regs *regs)
1353 {
1354         vmbus_initiate_unload(true);
1355         /*
1356          * In crash handler we can't schedule synic cleanup for all CPUs,
1357          * doing the cleanup for current CPU only. This should be sufficient
1358          * for kdump.
1359          */
1360         hv_synic_cleanup(NULL);
1361         hv_cleanup(true);
1362 };
1363
1364 static int __init hv_acpi_init(void)
1365 {
1366         int ret, t;
1367
1368         if (x86_hyper != &x86_hyper_ms_hyperv)
1369                 return -ENODEV;
1370
1371         init_completion(&probe_event);
1372
1373         /*
1374          * Get ACPI resources first.
1375          */
1376         ret = acpi_bus_register_driver(&vmbus_acpi_driver);
1377
1378         if (ret)
1379                 return ret;
1380
1381         t = wait_for_completion_timeout(&probe_event, 5*HZ);
1382         if (t == 0) {
1383                 ret = -ETIMEDOUT;
1384                 goto cleanup;
1385         }
1386
1387         ret = vmbus_bus_init();
1388         if (ret)
1389                 goto cleanup;
1390
1391         hv_setup_kexec_handler(hv_kexec_handler);
1392         hv_setup_crash_handler(hv_crash_handler);
1393
1394         return 0;
1395
1396 cleanup:
1397         acpi_bus_unregister_driver(&vmbus_acpi_driver);
1398         hv_acpi_dev = NULL;
1399         return ret;
1400 }
1401
1402 static void __exit vmbus_exit(void)
1403 {
1404         int cpu;
1405
1406         hv_remove_kexec_handler();
1407         hv_remove_crash_handler();
1408         vmbus_connection.conn_state = DISCONNECTED;
1409         hv_synic_clockevents_cleanup();
1410         vmbus_disconnect();
1411         hv_remove_vmbus_irq();
1412         for_each_online_cpu(cpu)
1413                 tasklet_kill(hv_context.msg_dpc[cpu]);
1414         vmbus_free_channels();
1415         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1416                 unregister_die_notifier(&hyperv_die_block);
1417                 atomic_notifier_chain_unregister(&panic_notifier_list,
1418                                                  &hyperv_panic_block);
1419         }
1420         bus_unregister(&hv_bus);
1421         hv_cleanup(false);
1422         for_each_online_cpu(cpu) {
1423                 tasklet_kill(hv_context.event_dpc[cpu]);
1424                 smp_call_function_single(cpu, hv_synic_cleanup, NULL, 1);
1425         }
1426         hv_synic_free();
1427         acpi_bus_unregister_driver(&vmbus_acpi_driver);
1428         if (vmbus_proto_version > VERSION_WIN7)
1429                 cpu_hotplug_enable();
1430 }
1431
1432
1433 MODULE_LICENSE("GPL");
1434
1435 subsys_initcall(hv_acpi_init);
1436 module_exit(vmbus_exit);