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 <linux/sched/task_stack.h>
  38
  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 struct vmbus_dynid {
  49         struct list_head node;
  50         struct hv_vmbus_device_id id;
  51 };
  52
  53 static struct acpi_device  *hv_acpi_dev;
  54
  55 static struct completion probe_event;
  56
  57 static int hyperv_cpuhp_online;
  58
  59 static void *hv_panic_page;
  60
  61 static int hyperv_panic_event(struct notifier_block *nb, unsigned long val,
  62                               void *args)
  63 {
  64         struct pt_regs *regs;
  65
  66         regs = current_pt_regs();
  67
  68         hyperv_report_panic(regs, val);
  69         return NOTIFY_DONE;
  70 }
  71
  72 static int hyperv_die_event(struct notifier_block *nb, unsigned long val,
  73                             void *args)
  74 {
  75         struct die_args *die = (struct die_args *)args;
  76         struct pt_regs *regs = die->regs;
  77
  78         hyperv_report_panic(regs, val);
  79         return NOTIFY_DONE;
  80 }
  81
  82 static struct notifier_block hyperv_die_block = {
  83         .notifier_call = hyperv_die_event,
  84 };
  85 static struct notifier_block hyperv_panic_block = {
  86         .notifier_call = hyperv_panic_event,
  87 };
  88
  89 static const char *fb_mmio_name = "fb_range";
  90 static struct resource *fb_mmio;
  91 static struct resource *hyperv_mmio;
  92 static DEFINE_SEMAPHORE(hyperv_mmio_lock);
  93
  94 static int vmbus_exists(void)
  95 {
  96         if (hv_acpi_dev == NULL)
  97                 return -ENODEV;
  98
  99         return 0;
 100 }
 101
 102 #define VMBUS_ALIAS_LEN ((sizeof((struct hv_vmbus_device_id *)0)->guid) * 2)
 103 static void print_alias_name(struct hv_device *hv_dev, char *alias_name)
 104 {
 105         int i;
 106         for (i = 0; i < VMBUS_ALIAS_LEN; i += 2)
 107                 sprintf(&alias_name[i], "%02x", hv_dev->dev_type.b[i/2]);
 108 }
 109
 110 static u8 channel_monitor_group(const struct vmbus_channel *channel)
 111 {
 112         return (u8)channel->offermsg.monitorid / 32;
 113 }
 114
 115 static u8 channel_monitor_offset(const struct vmbus_channel *channel)
 116 {
 117         return (u8)channel->offermsg.monitorid % 32;
 118 }
 119
 120 static u32 channel_pending(const struct vmbus_channel *channel,
 121                            const struct hv_monitor_page *monitor_page)
 122 {
 123         u8 monitor_group = channel_monitor_group(channel);
 124
 125         return monitor_page->trigger_group[monitor_group].pending;
 126 }
 127
 128 static u32 channel_latency(const struct vmbus_channel *channel,
 129                            const struct hv_monitor_page *monitor_page)
 130 {
 131         u8 monitor_group = channel_monitor_group(channel);
 132         u8 monitor_offset = channel_monitor_offset(channel);
 133
 134         return monitor_page->latency[monitor_group][monitor_offset];
 135 }
 136
 137 static u32 channel_conn_id(struct vmbus_channel *channel,
 138                            struct hv_monitor_page *monitor_page)
 139 {
 140         u8 monitor_group = channel_monitor_group(channel);
 141         u8 monitor_offset = channel_monitor_offset(channel);
 142         return monitor_page->parameter[monitor_group][monitor_offset].connectionid.u.id;
 143 }
 144
 145 static ssize_t id_show(struct device *dev, struct device_attribute *dev_attr,
 146                        char *buf)
 147 {
 148         struct hv_device *hv_dev = device_to_hv_device(dev);
 149
 150         if (!hv_dev->channel)
 151                 return -ENODEV;
 152         return sprintf(buf, "%d\n", hv_dev->channel->offermsg.child_relid);
 153 }
 154 static DEVICE_ATTR_RO(id);
 155
 156 static ssize_t state_show(struct device *dev, struct device_attribute *dev_attr,
 157                           char *buf)
 158 {
 159         struct hv_device *hv_dev = device_to_hv_device(dev);
 160
 161         if (!hv_dev->channel)
 162                 return -ENODEV;
 163         return sprintf(buf, "%d\n", hv_dev->channel->state);
 164 }
 165 static DEVICE_ATTR_RO(state);
 166
 167 static ssize_t monitor_id_show(struct device *dev,
 168                                struct device_attribute *dev_attr, char *buf)
 169 {
 170         struct hv_device *hv_dev = device_to_hv_device(dev);
 171
 172         if (!hv_dev->channel)
 173                 return -ENODEV;
 174         return sprintf(buf, "%d\n", hv_dev->channel->offermsg.monitorid);
 175 }
 176 static DEVICE_ATTR_RO(monitor_id);
 177
 178 static ssize_t class_id_show(struct device *dev,
 179                                struct device_attribute *dev_attr, char *buf)
 180 {
 181         struct hv_device *hv_dev = device_to_hv_device(dev);
 182
 183         if (!hv_dev->channel)
 184                 return -ENODEV;
 185         return sprintf(buf, "{%pUl}\n",
 186                        hv_dev->channel->offermsg.offer.if_type.b);
 187 }
 188 static DEVICE_ATTR_RO(class_id);
 189
 190 static ssize_t device_id_show(struct device *dev,
 191                               struct device_attribute *dev_attr, char *buf)
 192 {
 193         struct hv_device *hv_dev = device_to_hv_device(dev);
 194
 195         if (!hv_dev->channel)
 196                 return -ENODEV;
 197         return sprintf(buf, "{%pUl}\n",
 198                        hv_dev->channel->offermsg.offer.if_instance.b);
 199 }
 200 static DEVICE_ATTR_RO(device_id);
 201
 202 static ssize_t modalias_show(struct device *dev,
 203                              struct device_attribute *dev_attr, char *buf)
 204 {
 205         struct hv_device *hv_dev = device_to_hv_device(dev);
 206         char alias_name[VMBUS_ALIAS_LEN + 1];
 207
 208         print_alias_name(hv_dev, alias_name);
 209         return sprintf(buf, "vmbus:%s\n", alias_name);
 210 }
 211 static DEVICE_ATTR_RO(modalias);
 212
 213 #ifdef CONFIG_NUMA
 214 static ssize_t numa_node_show(struct device *dev,
 215                               struct device_attribute *attr, char *buf)
 216 {
 217         struct hv_device *hv_dev = device_to_hv_device(dev);
 218
 219         if (!hv_dev->channel)
 220                 return -ENODEV;
 221
 222         return sprintf(buf, "%d\n", hv_dev->channel->numa_node);
 223 }
 224 static DEVICE_ATTR_RO(numa_node);
 225 #endif
 226
 227 static ssize_t server_monitor_pending_show(struct device *dev,
 228                                            struct device_attribute *dev_attr,
 229                                            char *buf)
 230 {
 231         struct hv_device *hv_dev = device_to_hv_device(dev);
 232
 233         if (!hv_dev->channel)
 234                 return -ENODEV;
 235         return sprintf(buf, "%d\n",
 236                        channel_pending(hv_dev->channel,
 237                                        vmbus_connection.monitor_pages[0]));
 238 }
 239 static DEVICE_ATTR_RO(server_monitor_pending);
 240
 241 static ssize_t client_monitor_pending_show(struct device *dev,
 242                                            struct device_attribute *dev_attr,
 243                                            char *buf)
 244 {
 245         struct hv_device *hv_dev = device_to_hv_device(dev);
 246
 247         if (!hv_dev->channel)
 248                 return -ENODEV;
 249         return sprintf(buf, "%d\n",
 250                        channel_pending(hv_dev->channel,
 251                                        vmbus_connection.monitor_pages[1]));
 252 }
 253 static DEVICE_ATTR_RO(client_monitor_pending);
 254
 255 static ssize_t server_monitor_latency_show(struct device *dev,
 256                                            struct device_attribute *dev_attr,
 257                                            char *buf)
 258 {
 259         struct hv_device *hv_dev = device_to_hv_device(dev);
 260
 261         if (!hv_dev->channel)
 262                 return -ENODEV;
 263         return sprintf(buf, "%d\n",
 264                        channel_latency(hv_dev->channel,
 265                                        vmbus_connection.monitor_pages[0]));
 266 }
 267 static DEVICE_ATTR_RO(server_monitor_latency);
 268
 269 static ssize_t client_monitor_latency_show(struct device *dev,
 270                                            struct device_attribute *dev_attr,
 271                                            char *buf)
 272 {
 273         struct hv_device *hv_dev = device_to_hv_device(dev);
 274
 275         if (!hv_dev->channel)
 276                 return -ENODEV;
 277         return sprintf(buf, "%d\n",
 278                        channel_latency(hv_dev->channel,
 279                                        vmbus_connection.monitor_pages[1]));
 280 }
 281 static DEVICE_ATTR_RO(client_monitor_latency);
 282
 283 static ssize_t server_monitor_conn_id_show(struct device *dev,
 284                                            struct device_attribute *dev_attr,
 285                                            char *buf)
 286 {
 287         struct hv_device *hv_dev = device_to_hv_device(dev);
 288
 289         if (!hv_dev->channel)
 290                 return -ENODEV;
 291         return sprintf(buf, "%d\n",
 292                        channel_conn_id(hv_dev->channel,
 293                                        vmbus_connection.monitor_pages[0]));
 294 }
 295 static DEVICE_ATTR_RO(server_monitor_conn_id);
 296
 297 static ssize_t client_monitor_conn_id_show(struct device *dev,
 298                                            struct device_attribute *dev_attr,
 299                                            char *buf)
 300 {
 301         struct hv_device *hv_dev = device_to_hv_device(dev);
 302
 303         if (!hv_dev->channel)
 304                 return -ENODEV;
 305         return sprintf(buf, "%d\n",
 306                        channel_conn_id(hv_dev->channel,
 307                                        vmbus_connection.monitor_pages[1]));
 308 }
 309 static DEVICE_ATTR_RO(client_monitor_conn_id);
 310
 311 static ssize_t out_intr_mask_show(struct device *dev,
 312                                   struct device_attribute *dev_attr, char *buf)
 313 {
 314         struct hv_device *hv_dev = device_to_hv_device(dev);
 315         struct hv_ring_buffer_debug_info outbound;
 316         int ret;
 317
 318         if (!hv_dev->channel)
 319                 return -ENODEV;
 320
 321         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 322                                           &outbound);
 323         if (ret < 0)
 324                 return ret;
 325
 326         return sprintf(buf, "%d\n", outbound.current_interrupt_mask);
 327 }
 328 static DEVICE_ATTR_RO(out_intr_mask);
 329
 330 static ssize_t out_read_index_show(struct device *dev,
 331                                    struct device_attribute *dev_attr, char *buf)
 332 {
 333         struct hv_device *hv_dev = device_to_hv_device(dev);
 334         struct hv_ring_buffer_debug_info outbound;
 335         int ret;
 336
 337         if (!hv_dev->channel)
 338                 return -ENODEV;
 339
 340         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 341                                           &outbound);
 342         if (ret < 0)
 343                 return ret;
 344         return sprintf(buf, "%d\n", outbound.current_read_index);
 345 }
 346 static DEVICE_ATTR_RO(out_read_index);
 347
 348 static ssize_t out_write_index_show(struct device *dev,
 349                                     struct device_attribute *dev_attr,
 350                                     char *buf)
 351 {
 352         struct hv_device *hv_dev = device_to_hv_device(dev);
 353         struct hv_ring_buffer_debug_info outbound;
 354         int ret;
 355
 356         if (!hv_dev->channel)
 357                 return -ENODEV;
 358
 359         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 360                                           &outbound);
 361         if (ret < 0)
 362                 return ret;
 363         return sprintf(buf, "%d\n", outbound.current_write_index);
 364 }
 365 static DEVICE_ATTR_RO(out_write_index);
 366
 367 static ssize_t out_read_bytes_avail_show(struct device *dev,
 368                                          struct device_attribute *dev_attr,
 369                                          char *buf)
 370 {
 371         struct hv_device *hv_dev = device_to_hv_device(dev);
 372         struct hv_ring_buffer_debug_info outbound;
 373         int ret;
 374
 375         if (!hv_dev->channel)
 376                 return -ENODEV;
 377
 378         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 379                                           &outbound);
 380         if (ret < 0)
 381                 return ret;
 382         return sprintf(buf, "%d\n", outbound.bytes_avail_toread);
 383 }
 384 static DEVICE_ATTR_RO(out_read_bytes_avail);
 385
 386 static ssize_t out_write_bytes_avail_show(struct device *dev,
 387                                           struct device_attribute *dev_attr,
 388                                           char *buf)
 389 {
 390         struct hv_device *hv_dev = device_to_hv_device(dev);
 391         struct hv_ring_buffer_debug_info outbound;
 392         int ret;
 393
 394         if (!hv_dev->channel)
 395                 return -ENODEV;
 396
 397         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 398                                           &outbound);
 399         if (ret < 0)
 400                 return ret;
 401         return sprintf(buf, "%d\n", outbound.bytes_avail_towrite);
 402 }
 403 static DEVICE_ATTR_RO(out_write_bytes_avail);
 404
 405 static ssize_t in_intr_mask_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         int ret;
 411
 412         if (!hv_dev->channel)
 413                 return -ENODEV;
 414
 415         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 416         if (ret < 0)
 417                 return ret;
 418
 419         return sprintf(buf, "%d\n", inbound.current_interrupt_mask);
 420 }
 421 static DEVICE_ATTR_RO(in_intr_mask);
 422
 423 static ssize_t in_read_index_show(struct device *dev,
 424                                   struct device_attribute *dev_attr, char *buf)
 425 {
 426         struct hv_device *hv_dev = device_to_hv_device(dev);
 427         struct hv_ring_buffer_debug_info inbound;
 428         int ret;
 429
 430         if (!hv_dev->channel)
 431                 return -ENODEV;
 432
 433         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 434         if (ret < 0)
 435                 return ret;
 436
 437         return sprintf(buf, "%d\n", inbound.current_read_index);
 438 }
 439 static DEVICE_ATTR_RO(in_read_index);
 440
 441 static ssize_t in_write_index_show(struct device *dev,
 442                                    struct device_attribute *dev_attr, char *buf)
 443 {
 444         struct hv_device *hv_dev = device_to_hv_device(dev);
 445         struct hv_ring_buffer_debug_info inbound;
 446         int ret;
 447
 448         if (!hv_dev->channel)
 449                 return -ENODEV;
 450
 451         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 452         if (ret < 0)
 453                 return ret;
 454
 455         return sprintf(buf, "%d\n", inbound.current_write_index);
 456 }
 457 static DEVICE_ATTR_RO(in_write_index);
 458
 459 static ssize_t in_read_bytes_avail_show(struct device *dev,
 460                                         struct device_attribute *dev_attr,
 461                                         char *buf)
 462 {
 463         struct hv_device *hv_dev = device_to_hv_device(dev);
 464         struct hv_ring_buffer_debug_info inbound;
 465         int ret;
 466
 467         if (!hv_dev->channel)
 468                 return -ENODEV;
 469
 470         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 471         if (ret < 0)
 472                 return ret;
 473
 474         return sprintf(buf, "%d\n", inbound.bytes_avail_toread);
 475 }
 476 static DEVICE_ATTR_RO(in_read_bytes_avail);
 477
 478 static ssize_t in_write_bytes_avail_show(struct device *dev,
 479                                          struct device_attribute *dev_attr,
 480                                          char *buf)
 481 {
 482         struct hv_device *hv_dev = device_to_hv_device(dev);
 483         struct hv_ring_buffer_debug_info inbound;
 484         int ret;
 485
 486         if (!hv_dev->channel)
 487                 return -ENODEV;
 488
 489         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 490         if (ret < 0)
 491                 return ret;
 492
 493         return sprintf(buf, "%d\n", inbound.bytes_avail_towrite);
 494 }
 495 static DEVICE_ATTR_RO(in_write_bytes_avail);
 496
 497 static ssize_t channel_vp_mapping_show(struct device *dev,
 498                                        struct device_attribute *dev_attr,
 499                                        char *buf)
 500 {
 501         struct hv_device *hv_dev = device_to_hv_device(dev);
 502         struct vmbus_channel *channel = hv_dev->channel, *cur_sc;
 503         unsigned long flags;
 504         int buf_size = PAGE_SIZE, n_written, tot_written;
 505         struct list_head *cur;
 506
 507         if (!channel)
 508                 return -ENODEV;
 509
 510         tot_written = snprintf(buf, buf_size, "%u:%u\n",
 511                 channel->offermsg.child_relid, channel->target_cpu);
 512
 513         spin_lock_irqsave(&channel->lock, flags);
 514
 515         list_for_each(cur, &channel->sc_list) {
 516                 if (tot_written >= buf_size - 1)
 517                         break;
 518
 519                 cur_sc = list_entry(cur, struct vmbus_channel, sc_list);
 520                 n_written = scnprintf(buf + tot_written,
 521                                      buf_size - tot_written,
 522                                      "%u:%u\n",
 523                                      cur_sc->offermsg.child_relid,
 524                                      cur_sc->target_cpu);
 525                 tot_written += n_written;
 526         }
 527
 528         spin_unlock_irqrestore(&channel->lock, flags);
 529
 530         return tot_written;
 531 }
 532 static DEVICE_ATTR_RO(channel_vp_mapping);
 533
 534 static ssize_t vendor_show(struct device *dev,
 535                            struct device_attribute *dev_attr,
 536                            char *buf)
 537 {
 538         struct hv_device *hv_dev = device_to_hv_device(dev);
 539         return sprintf(buf, "0x%x\n", hv_dev->vendor_id);
 540 }
 541 static DEVICE_ATTR_RO(vendor);
 542
 543 static ssize_t device_show(struct device *dev,
 544                            struct device_attribute *dev_attr,
 545                            char *buf)
 546 {
 547         struct hv_device *hv_dev = device_to_hv_device(dev);
 548         return sprintf(buf, "0x%x\n", hv_dev->device_id);
 549 }
 550 static DEVICE_ATTR_RO(device);
 551
 552 static ssize_t driver_override_store(struct device *dev,
 553                                      struct device_attribute *attr,
 554                                      const char *buf, size_t count)
 555 {
 556         struct hv_device *hv_dev = device_to_hv_device(dev);
 557         char *driver_override, *old, *cp;
 558
 559         /* We need to keep extra room for a newline */
 560         if (count >= (PAGE_SIZE - 1))
 561                 return -EINVAL;
 562
 563         driver_override = kstrndup(buf, count, GFP_KERNEL);
 564         if (!driver_override)
 565                 return -ENOMEM;
 566
 567         cp = strchr(driver_override, '\n');
 568         if (cp)
 569                 *cp = '\0';
 570
 571         device_lock(dev);
 572         old = hv_dev->driver_override;
 573         if (strlen(driver_override)) {
 574                 hv_dev->driver_override = driver_override;
 575         } else {
 576                 kfree(driver_override);
 577                 hv_dev->driver_override = NULL;
 578         }
 579         device_unlock(dev);
 580
 581         kfree(old);
 582
 583         return count;
 584 }
 585
 586 static ssize_t driver_override_show(struct device *dev,
 587                                     struct device_attribute *attr, char *buf)
 588 {
 589         struct hv_device *hv_dev = device_to_hv_device(dev);
 590         ssize_t len;
 591
 592         device_lock(dev);
 593         len = snprintf(buf, PAGE_SIZE, "%s\n", hv_dev->driver_override);
 594         device_unlock(dev);
 595
 596         return len;
 597 }
 598 static DEVICE_ATTR_RW(driver_override);
 599
 600 /* Set up per device attributes in /sys/bus/vmbus/devices/<bus device> */
 601 static struct attribute *vmbus_dev_attrs[] = {
 602         &dev_attr_id.attr,
 603         &dev_attr_state.attr,
 604         &dev_attr_monitor_id.attr,
 605         &dev_attr_class_id.attr,
 606         &dev_attr_device_id.attr,
 607         &dev_attr_modalias.attr,
 608 #ifdef CONFIG_NUMA
 609         &dev_attr_numa_node.attr,
 610 #endif
 611         &dev_attr_server_monitor_pending.attr,
 612         &dev_attr_client_monitor_pending.attr,
 613         &dev_attr_server_monitor_latency.attr,
 614         &dev_attr_client_monitor_latency.attr,
 615         &dev_attr_server_monitor_conn_id.attr,
 616         &dev_attr_client_monitor_conn_id.attr,
 617         &dev_attr_out_intr_mask.attr,
 618         &dev_attr_out_read_index.attr,
 619         &dev_attr_out_write_index.attr,
 620         &dev_attr_out_read_bytes_avail.attr,
 621         &dev_attr_out_write_bytes_avail.attr,
 622         &dev_attr_in_intr_mask.attr,
 623         &dev_attr_in_read_index.attr,
 624         &dev_attr_in_write_index.attr,
 625         &dev_attr_in_read_bytes_avail.attr,
 626         &dev_attr_in_write_bytes_avail.attr,
 627         &dev_attr_channel_vp_mapping.attr,
 628         &dev_attr_vendor.attr,
 629         &dev_attr_device.attr,
 630         &dev_attr_driver_override.attr,
 631         NULL,
 632 };
 633
 634 /*
 635  * Device-level attribute_group callback function. Returns the permission for
 636  * each attribute, and returns 0 if an attribute is not visible.
 637  */
 638 static umode_t vmbus_dev_attr_is_visible(struct kobject *kobj,
 639                                          struct attribute *attr, int idx)
 640 {
 641         struct device *dev = kobj_to_dev(kobj);
 642         const struct hv_device *hv_dev = device_to_hv_device(dev);
 643
 644         /* Hide the monitor attributes if the monitor mechanism is not used. */
 645         if (!hv_dev->channel->offermsg.monitor_allocated &&
 646             (attr == &dev_attr_monitor_id.attr ||
 647              attr == &dev_attr_server_monitor_pending.attr ||
 648              attr == &dev_attr_client_monitor_pending.attr ||
 649              attr == &dev_attr_server_monitor_latency.attr ||
 650              attr == &dev_attr_client_monitor_latency.attr ||
 651              attr == &dev_attr_server_monitor_conn_id.attr ||
 652              attr == &dev_attr_client_monitor_conn_id.attr))
 653                 return 0;
 654
 655         return attr->mode;
 656 }
 657
 658 static const struct attribute_group vmbus_dev_group = {
 659         .attrs = vmbus_dev_attrs,
 660         .is_visible = vmbus_dev_attr_is_visible
 661 };
 662 __ATTRIBUTE_GROUPS(vmbus_dev);
 663
 664 /*
 665  * vmbus_uevent - add uevent for our device
 666  *
 667  * This routine is invoked when a device is added or removed on the vmbus to
 668  * generate a uevent to udev in the userspace. The udev will then look at its
 669  * rule and the uevent generated here to load the appropriate driver
 670  *
 671  * The alias string will be of the form vmbus:guid where guid is the string
 672  * representation of the device guid (each byte of the guid will be
 673  * represented with two hex characters.
 674  */
 675 static int vmbus_uevent(struct device *device, struct kobj_uevent_env *env)
 676 {
 677         struct hv_device *dev = device_to_hv_device(device);
 678         int ret;
 679         char alias_name[VMBUS_ALIAS_LEN + 1];
 680
 681         print_alias_name(dev, alias_name);
 682         ret = add_uevent_var(env, "MODALIAS=vmbus:%s", alias_name);
 683         return ret;
 684 }
 685
 686 static const struct hv_vmbus_device_id *
 687 hv_vmbus_dev_match(const struct hv_vmbus_device_id *id, const guid_t *guid)
 688 {
 689         if (id == NULL)
 690                 return NULL; /* empty device table */
 691
 692         for (; !guid_is_null(&id->guid); id++)
 693                 if (guid_equal(&id->guid, guid))
 694                         return id;
 695
 696         return NULL;
 697 }
 698
 699 static const struct hv_vmbus_device_id *
 700 hv_vmbus_dynid_match(struct hv_driver *drv, const guid_t *guid)
 701 {
 702         const struct hv_vmbus_device_id *id = NULL;
 703         struct vmbus_dynid *dynid;
 704
 705         spin_lock(&drv->dynids.lock);
 706         list_for_each_entry(dynid, &drv->dynids.list, node) {
 707                 if (guid_equal(&dynid->id.guid, guid)) {
 708                         id = &dynid->id;
 709                         break;
 710                 }
 711         }
 712         spin_unlock(&drv->dynids.lock);
 713
 714         return id;
 715 }
 716
 717 static const struct hv_vmbus_device_id vmbus_device_null;
 718
 719 /*
 720  * Return a matching hv_vmbus_device_id pointer.
 721  * If there is no match, return NULL.
 722  */
 723 static const struct hv_vmbus_device_id *hv_vmbus_get_id(struct hv_driver *drv,
 724                                                         struct hv_device *dev)
 725 {
 726         const guid_t *guid = &dev->dev_type;
 727         const struct hv_vmbus_device_id *id;
 728
 729         /* When driver_override is set, only bind to the matching driver */
 730         if (dev->driver_override && strcmp(dev->driver_override, drv->name))
 731                 return NULL;
 732
 733         /* Look at the dynamic ids first, before the static ones */
 734         id = hv_vmbus_dynid_match(drv, guid);
 735         if (!id)
 736                 id = hv_vmbus_dev_match(drv->id_table, guid);
 737
 738         /* driver_override will always match, send a dummy id */
 739         if (!id && dev->driver_override)
 740                 id = &vmbus_device_null;
 741
 742         return id;
 743 }
 744
 745 /* vmbus_add_dynid - add a new device ID to this driver and re-probe devices */
 746 static int vmbus_add_dynid(struct hv_driver *drv, guid_t *guid)
 747 {
 748         struct vmbus_dynid *dynid;
 749
 750         dynid = kzalloc(sizeof(*dynid), GFP_KERNEL);
 751         if (!dynid)
 752                 return -ENOMEM;
 753
 754         dynid->id.guid = *guid;
 755
 756         spin_lock(&drv->dynids.lock);
 757         list_add_tail(&dynid->node, &drv->dynids.list);
 758         spin_unlock(&drv->dynids.lock);
 759
 760         return driver_attach(&drv->driver);
 761 }
 762
 763 static void vmbus_free_dynids(struct hv_driver *drv)
 764 {
 765         struct vmbus_dynid *dynid, *n;
 766
 767         spin_lock(&drv->dynids.lock);
 768         list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
 769                 list_del(&dynid->node);
 770                 kfree(dynid);
 771         }
 772         spin_unlock(&drv->dynids.lock);
 773 }
 774
 775 /*
 776  * store_new_id - sysfs frontend to vmbus_add_dynid()
 777  *
 778  * Allow GUIDs to be added to an existing driver via sysfs.
 779  */
 780 static ssize_t new_id_store(struct device_driver *driver, const char *buf,
 781                             size_t count)
 782 {
 783         struct hv_driver *drv = drv_to_hv_drv(driver);
 784         guid_t guid;
 785         ssize_t retval;
 786
 787         retval = guid_parse(buf, &guid);
 788         if (retval)
 789                 return retval;
 790
 791         if (hv_vmbus_dynid_match(drv, &guid))
 792                 return -EEXIST;
 793
 794         retval = vmbus_add_dynid(drv, &guid);
 795         if (retval)
 796                 return retval;
 797         return count;
 798 }
 799 static DRIVER_ATTR_WO(new_id);
 800
 801 /*
 802  * store_remove_id - remove a PCI device ID from this driver
 803  *
 804  * Removes a dynamic pci device ID to this driver.
 805  */
 806 static ssize_t remove_id_store(struct device_driver *driver, const char *buf,
 807                                size_t count)
 808 {
 809         struct hv_driver *drv = drv_to_hv_drv(driver);
 810         struct vmbus_dynid *dynid, *n;
 811         guid_t guid;
 812         ssize_t retval;
 813
 814         retval = guid_parse(buf, &guid);
 815         if (retval)
 816                 return retval;
 817
 818         retval = -ENODEV;
 819         spin_lock(&drv->dynids.lock);
 820         list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
 821                 struct hv_vmbus_device_id *id = &dynid->id;
 822
 823                 if (guid_equal(&id->guid, &guid)) {
 824                         list_del(&dynid->node);
 825                         kfree(dynid);
 826                         retval = count;
 827                         break;
 828                 }
 829         }
 830         spin_unlock(&drv->dynids.lock);
 831
 832         return retval;
 833 }
 834 static DRIVER_ATTR_WO(remove_id);
 835
 836 static struct attribute *vmbus_drv_attrs[] = {
 837         &driver_attr_new_id.attr,
 838         &driver_attr_remove_id.attr,
 839         NULL,
 840 };
 841 ATTRIBUTE_GROUPS(vmbus_drv);
 842
 843
 844 /*
 845  * vmbus_match - Attempt to match the specified device to the specified driver
 846  */
 847 static int vmbus_match(struct device *device, struct device_driver *driver)
 848 {
 849         struct hv_driver *drv = drv_to_hv_drv(driver);
 850         struct hv_device *hv_dev = device_to_hv_device(device);
 851
 852         /* The hv_sock driver handles all hv_sock offers. */
 853         if (is_hvsock_channel(hv_dev->channel))
 854                 return drv->hvsock;
 855
 856         if (hv_vmbus_get_id(drv, hv_dev))
 857                 return 1;
 858
 859         return 0;
 860 }
 861
 862 /*
 863  * vmbus_probe - Add the new vmbus's child device
 864  */
 865 static int vmbus_probe(struct device *child_device)
 866 {
 867         int ret = 0;
 868         struct hv_driver *drv =
 869                         drv_to_hv_drv(child_device->driver);
 870         struct hv_device *dev = device_to_hv_device(child_device);
 871         const struct hv_vmbus_device_id *dev_id;
 872
 873         dev_id = hv_vmbus_get_id(drv, dev);
 874         if (drv->probe) {
 875                 ret = drv->probe(dev, dev_id);
 876                 if (ret != 0)
 877                         pr_err("probe failed for device %s (%d)\n",
 878                                dev_name(child_device), ret);
 879
 880         } else {
 881                 pr_err("probe not set for driver %s\n",
 882                        dev_name(child_device));
 883                 ret = -ENODEV;
 884         }
 885         return ret;
 886 }
 887
 888 /*
 889  * vmbus_remove - Remove a vmbus device
 890  */
 891 static int vmbus_remove(struct device *child_device)
 892 {
 893         struct hv_driver *drv;
 894         struct hv_device *dev = device_to_hv_device(child_device);
 895
 896         if (child_device->driver) {
 897                 drv = drv_to_hv_drv(child_device->driver);
 898                 if (drv->remove)
 899                         drv->remove(dev);
 900         }
 901
 902         return 0;
 903 }
 904
 905
 906 /*
 907  * vmbus_shutdown - Shutdown a vmbus device
 908  */
 909 static void vmbus_shutdown(struct device *child_device)
 910 {
 911         struct hv_driver *drv;
 912         struct hv_device *dev = device_to_hv_device(child_device);
 913
 914
 915         /* The device may not be attached yet */
 916         if (!child_device->driver)
 917                 return;
 918
 919         drv = drv_to_hv_drv(child_device->driver);
 920
 921         if (drv->shutdown)
 922                 drv->shutdown(dev);
 923 }
 924
 925
 926 /*
 927  * vmbus_device_release - Final callback release of the vmbus child device
 928  */
 929 static void vmbus_device_release(struct device *device)
 930 {
 931         struct hv_device *hv_dev = device_to_hv_device(device);
 932         struct vmbus_channel *channel = hv_dev->channel;
 933
 934         mutex_lock(&vmbus_connection.channel_mutex);
 935         hv_process_channel_removal(channel);
 936         mutex_unlock(&vmbus_connection.channel_mutex);
 937         kfree(hv_dev);
 938 }
 939
 940 /* The one and only one */
 941 static struct bus_type  hv_bus = {
 942         .name =         "vmbus",
 943         .match =                vmbus_match,
 944         .shutdown =             vmbus_shutdown,
 945         .remove =               vmbus_remove,
 946         .probe =                vmbus_probe,
 947         .uevent =               vmbus_uevent,
 948         .dev_groups =           vmbus_dev_groups,
 949         .drv_groups =           vmbus_drv_groups,
 950 };
 951
 952 struct onmessage_work_context {
 953         struct work_struct work;
 954         struct hv_message msg;
 955 };
 956
 957 static void vmbus_onmessage_work(struct work_struct *work)
 958 {
 959         struct onmessage_work_context *ctx;
 960
 961         /* Do not process messages if we're in DISCONNECTED state */
 962         if (vmbus_connection.conn_state == DISCONNECTED)
 963                 return;
 964
 965         ctx = container_of(work, struct onmessage_work_context,
 966                            work);
 967         vmbus_onmessage(&ctx->msg);
 968         kfree(ctx);
 969 }
 970
 971 static void hv_process_timer_expiration(struct hv_message *msg,
 972                                         struct hv_per_cpu_context *hv_cpu)
 973 {
 974         struct clock_event_device *dev = hv_cpu->clk_evt;
 975
 976         if (dev->event_handler)
 977                 dev->event_handler(dev);
 978
 979         vmbus_signal_eom(msg, HVMSG_TIMER_EXPIRED);
 980 }
 981
 982 void vmbus_on_msg_dpc(unsigned long data)
 983 {
 984         struct hv_per_cpu_context *hv_cpu = (void *)data;
 985         void *page_addr = hv_cpu->synic_message_page;
 986         struct hv_message *msg = (struct hv_message *)page_addr +
 987                                   VMBUS_MESSAGE_SINT;
 988         struct vmbus_channel_message_header *hdr;
 989         const struct vmbus_channel_message_table_entry *entry;
 990         struct onmessage_work_context *ctx;
 991         u32 message_type = msg->header.message_type;
 992
 993         if (message_type == HVMSG_NONE)
 994                 /* no msg */
 995                 return;
 996
 997         hdr = (struct vmbus_channel_message_header *)msg->u.payload;
 998
 999         trace_vmbus_on_msg_dpc(hdr);
1000
1001         if (hdr->msgtype >= CHANNELMSG_COUNT) {
1002                 WARN_ONCE(1, "unknown msgtype=%d\n", hdr->msgtype);
1003                 goto msg_handled;
1004         }
1005
1006         entry = &channel_message_table[hdr->msgtype];
1007         if (entry->handler_type == VMHT_BLOCKING) {
1008                 ctx = kmalloc(sizeof(*ctx), GFP_ATOMIC);
1009                 if (ctx == NULL)
1010                         return;
1011
1012                 INIT_WORK(&ctx->work, vmbus_onmessage_work);
1013                 memcpy(&ctx->msg, msg, sizeof(*msg));
1014
1015                 /*
1016                  * The host can generate a rescind message while we
1017                  * may still be handling the original offer. We deal with
1018                  * this condition by ensuring the processing is done on the
1019                  * same CPU.
1020                  */
1021                 switch (hdr->msgtype) {
1022                 case CHANNELMSG_RESCIND_CHANNELOFFER:
1023                         /*
1024                          * If we are handling the rescind message;
1025                          * schedule the work on the global work queue.
1026                          */
1027                         schedule_work_on(vmbus_connection.connect_cpu,
1028                                          &ctx->work);
1029                         break;
1030
1031                 case CHANNELMSG_OFFERCHANNEL:
1032                         atomic_inc(&vmbus_connection.offer_in_progress);
1033                         queue_work_on(vmbus_connection.connect_cpu,
1034                                       vmbus_connection.work_queue,
1035                                       &ctx->work);
1036                         break;
1037
1038                 default:
1039                         queue_work(vmbus_connection.work_queue, &ctx->work);
1040                 }
1041         } else
1042                 entry->message_handler(hdr);
1043
1044 msg_handled:
1045         vmbus_signal_eom(msg, message_type);
1046 }
1047
1048
1049 /*
1050  * Direct callback for channels using other deferred processing
1051  */
1052 static void vmbus_channel_isr(struct vmbus_channel *channel)
1053 {
1054         void (*callback_fn)(void *);
1055
1056         callback_fn = READ_ONCE(channel->onchannel_callback);
1057         if (likely(callback_fn != NULL))
1058                 (*callback_fn)(channel->channel_callback_context);
1059 }
1060
1061 /*
1062  * Schedule all channels with events pending
1063  */
1064 static void vmbus_chan_sched(struct hv_per_cpu_context *hv_cpu)
1065 {
1066         unsigned long *recv_int_page;
1067         u32 maxbits, relid;
1068
1069         if (vmbus_proto_version < VERSION_WIN8) {
1070                 maxbits = MAX_NUM_CHANNELS_SUPPORTED;
1071                 recv_int_page = vmbus_connection.recv_int_page;
1072         } else {
1073                 /*
1074                  * When the host is win8 and beyond, the event page
1075                  * can be directly checked to get the id of the channel
1076                  * that has the interrupt pending.
1077                  */
1078                 void *page_addr = hv_cpu->synic_event_page;
1079                 union hv_synic_event_flags *event
1080                         = (union hv_synic_event_flags *)page_addr +
1081                                                  VMBUS_MESSAGE_SINT;
1082
1083                 maxbits = HV_EVENT_FLAGS_COUNT;
1084                 recv_int_page = event->flags;
1085         }
1086
1087         if (unlikely(!recv_int_page))
1088                 return;
1089
1090         for_each_set_bit(relid, recv_int_page, maxbits) {
1091                 struct vmbus_channel *channel;
1092
1093                 if (!sync_test_and_clear_bit(relid, recv_int_page))
1094                         continue;
1095
1096                 /* Special case - vmbus channel protocol msg */
1097                 if (relid == 0)
1098                         continue;
1099
1100                 rcu_read_lock();
1101
1102                 /* Find channel based on relid */
1103                 list_for_each_entry_rcu(channel, &hv_cpu->chan_list, percpu_list) {
1104                         if (channel->offermsg.child_relid != relid)
1105                                 continue;
1106
1107                         if (channel->rescind)
1108                                 continue;
1109
1110                         trace_vmbus_chan_sched(channel);
1111
1112                         ++channel->interrupts;
1113
1114                         switch (channel->callback_mode) {
1115                         case HV_CALL_ISR:
1116                                 vmbus_channel_isr(channel);
1117                                 break;
1118
1119                         case HV_CALL_BATCHED:
1120                                 hv_begin_read(&channel->inbound);
1121                                 /* fallthrough */
1122                         case HV_CALL_DIRECT:
1123                                 tasklet_schedule(&channel->callback_event);
1124                         }
1125                 }
1126
1127                 rcu_read_unlock();
1128         }
1129 }
1130
1131 static void vmbus_isr(void)
1132 {
1133         struct hv_per_cpu_context *hv_cpu
1134                 = this_cpu_ptr(hv_context.cpu_context);
1135         void *page_addr = hv_cpu->synic_event_page;
1136         struct hv_message *msg;
1137         union hv_synic_event_flags *event;
1138         bool handled = false;
1139
1140         if (unlikely(page_addr == NULL))
1141                 return;
1142
1143         event = (union hv_synic_event_flags *)page_addr +
1144                                          VMBUS_MESSAGE_SINT;
1145         /*
1146          * Check for events before checking for messages. This is the order
1147          * in which events and messages are checked in Windows guests on
1148          * Hyper-V, and the Windows team suggested we do the same.
1149          */
1150
1151         if ((vmbus_proto_version == VERSION_WS2008) ||
1152                 (vmbus_proto_version == VERSION_WIN7)) {
1153
1154                 /* Since we are a child, we only need to check bit 0 */
1155                 if (sync_test_and_clear_bit(0, event->flags))
1156                         handled = true;
1157         } else {
1158                 /*
1159                  * Our host is win8 or above. The signaling mechanism
1160                  * has changed and we can directly look at the event page.
1161                  * If bit n is set then we have an interrup on the channel
1162                  * whose id is n.
1163                  */
1164                 handled = true;
1165         }
1166
1167         if (handled)
1168                 vmbus_chan_sched(hv_cpu);
1169
1170         page_addr = hv_cpu->synic_message_page;
1171         msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
1172
1173         /* Check if there are actual msgs to be processed */
1174         if (msg->header.message_type != HVMSG_NONE) {
1175                 if (msg->header.message_type == HVMSG_TIMER_EXPIRED)
1176                         hv_process_timer_expiration(msg, hv_cpu);
1177                 else
1178                         tasklet_schedule(&hv_cpu->msg_dpc);
1179         }
1180
1181         add_interrupt_randomness(HYPERVISOR_CALLBACK_VECTOR, 0);
1182 }
1183
1184 /*
1185  * Boolean to control whether to report panic messages over Hyper-V.
1186  *
1187  * It can be set via /proc/sys/kernel/hyperv/record_panic_msg
1188  */
1189 static int sysctl_record_panic_msg = 1;
1190
1191 /*
1192  * Callback from kmsg_dump. Grab as much as possible from the end of the kmsg
1193  * buffer and call into Hyper-V to transfer the data.
1194  */
1195 static void hv_kmsg_dump(struct kmsg_dumper *dumper,
1196                          enum kmsg_dump_reason reason)
1197 {
1198         size_t bytes_written;
1199         phys_addr_t panic_pa;
1200
1201         /* We are only interested in panics. */
1202         if ((reason != KMSG_DUMP_PANIC) || (!sysctl_record_panic_msg))
1203                 return;
1204
1205         panic_pa = virt_to_phys(hv_panic_page);
1206
1207         /*
1208          * Write dump contents to the page. No need to synchronize; panic should
1209          * be single-threaded.
1210          */
1211         kmsg_dump_get_buffer(dumper, true, hv_panic_page, PAGE_SIZE,
1212                              &bytes_written);
1213         if (bytes_written)
1214                 hyperv_report_panic_msg(panic_pa, bytes_written);
1215 }
1216
1217 static struct kmsg_dumper hv_kmsg_dumper = {
1218         .dump = hv_kmsg_dump,
1219 };
1220
1221 static struct ctl_table_header *hv_ctl_table_hdr;
1222 static int zero;
1223 static int one = 1;
1224
1225 /*
1226  * sysctl option to allow the user to control whether kmsg data should be
1227  * reported to Hyper-V on panic.
1228  */
1229 static struct ctl_table hv_ctl_table[] = {
1230         {
1231                 .procname       = "hyperv_record_panic_msg",
1232                 .data           = &sysctl_record_panic_msg,
1233                 .maxlen         = sizeof(int),
1234                 .mode           = 0644,
1235                 .proc_handler   = proc_dointvec_minmax,
1236                 .extra1         = &zero,
1237                 .extra2         = &one
1238         },
1239         {}
1240 };
1241
1242 static struct ctl_table hv_root_table[] = {
1243         {
1244                 .procname       = "kernel",
1245                 .mode           = 0555,
1246                 .child          = hv_ctl_table
1247         },
1248         {}
1249 };
1250
1251 /*
1252  * vmbus_bus_init -Main vmbus driver initialization routine.
1253  *
1254  * Here, we
1255  *      - initialize the vmbus driver context
1256  *      - invoke the vmbus hv main init routine
1257  *      - retrieve the channel offers
1258  */
1259 static int vmbus_bus_init(void)
1260 {
1261         int ret;
1262
1263         /* Hypervisor initialization...setup hypercall page..etc */
1264         ret = hv_init();
1265         if (ret != 0) {
1266                 pr_err("Unable to initialize the hypervisor - 0x%x\n", ret);
1267                 return ret;
1268         }
1269
1270         ret = bus_register(&hv_bus);
1271         if (ret)
1272                 return ret;
1273
1274         hv_setup_vmbus_irq(vmbus_isr);
1275
1276         ret = hv_synic_alloc();
1277         if (ret)
1278                 goto err_alloc;
1279         /*
1280          * Initialize the per-cpu interrupt state and
1281          * connect to the host.
1282          */
1283         ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "hyperv/vmbus:online",
1284                                 hv_synic_init, hv_synic_cleanup);
1285         if (ret < 0)
1286                 goto err_alloc;
1287         hyperv_cpuhp_online = ret;
1288
1289         ret = vmbus_connect();
1290         if (ret)
1291                 goto err_connect;
1292
1293         /*
1294          * Only register if the crash MSRs are available
1295          */
1296         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1297                 u64 hyperv_crash_ctl;
1298                 /*
1299                  * Sysctl registration is not fatal, since by default
1300                  * reporting is enabled.
1301                  */
1302                 hv_ctl_table_hdr = register_sysctl_table(hv_root_table);
1303                 if (!hv_ctl_table_hdr)
1304                         pr_err("Hyper-V: sysctl table register error");
1305
1306                 /*
1307                  * Register for panic kmsg callback only if the right
1308                  * capability is supported by the hypervisor.
1309                  */
1310                 hv_get_crash_ctl(hyperv_crash_ctl);
1311                 if (hyperv_crash_ctl & HV_CRASH_CTL_CRASH_NOTIFY_MSG) {
1312                         hv_panic_page = (void *)get_zeroed_page(GFP_KERNEL);
1313                         if (hv_panic_page) {
1314                                 ret = kmsg_dump_register(&hv_kmsg_dumper);
1315                                 if (ret)
1316                                         pr_err("Hyper-V: kmsg dump register "
1317                                                 "error 0x%x\n", ret);
1318                         } else
1319                                 pr_err("Hyper-V: panic message page memory "
1320                                         "allocation failed");
1321                 }
1322
1323                 register_die_notifier(&hyperv_die_block);
1324                 atomic_notifier_chain_register(&panic_notifier_list,
1325                                                &hyperv_panic_block);
1326         }
1327
1328         vmbus_request_offers();
1329
1330         return 0;
1331
1332 err_connect:
1333         cpuhp_remove_state(hyperv_cpuhp_online);
1334 err_alloc:
1335         hv_synic_free();
1336         hv_remove_vmbus_irq();
1337
1338         bus_unregister(&hv_bus);
1339         free_page((unsigned long)hv_panic_page);
1340         unregister_sysctl_table(hv_ctl_table_hdr);
1341         hv_ctl_table_hdr = NULL;
1342         return ret;
1343 }
1344
1345 /**
1346  * __vmbus_child_driver_register() - Register a vmbus's driver
1347  * @hv_driver: Pointer to driver structure you want to register
1348  * @owner: owner module of the drv
1349  * @mod_name: module name string
1350  *
1351  * Registers the given driver with Linux through the 'driver_register()' call
1352  * and sets up the hyper-v vmbus handling for this driver.
1353  * It will return the state of the 'driver_register()' call.
1354  *
1355  */
1356 int __vmbus_driver_register(struct hv_driver *hv_driver, struct module *owner, const char *mod_name)
1357 {
1358         int ret;
1359
1360         pr_info("registering driver %s\n", hv_driver->name);
1361
1362         ret = vmbus_exists();
1363         if (ret < 0)
1364                 return ret;
1365
1366         hv_driver->driver.name = hv_driver->name;
1367         hv_driver->driver.owner = owner;
1368         hv_driver->driver.mod_name = mod_name;
1369         hv_driver->driver.bus = &hv_bus;
1370
1371         spin_lock_init(&hv_driver->dynids.lock);
1372         INIT_LIST_HEAD(&hv_driver->dynids.list);
1373
1374         ret = driver_register(&hv_driver->driver);
1375
1376         return ret;
1377 }
1378 EXPORT_SYMBOL_GPL(__vmbus_driver_register);
1379
1380 /**
1381  * vmbus_driver_unregister() - Unregister a vmbus's driver
1382  * @hv_driver: Pointer to driver structure you want to
1383  *             un-register
1384  *
1385  * Un-register the given driver that was previous registered with a call to
1386  * vmbus_driver_register()
1387  */
1388 void vmbus_driver_unregister(struct hv_driver *hv_driver)
1389 {
1390         pr_info("unregistering driver %s\n", hv_driver->name);
1391
1392         if (!vmbus_exists()) {
1393                 driver_unregister(&hv_driver->driver);
1394                 vmbus_free_dynids(hv_driver);
1395         }
1396 }
1397 EXPORT_SYMBOL_GPL(vmbus_driver_unregister);
1398
1399
1400 /*
1401  * Called when last reference to channel is gone.
1402  */
1403 static void vmbus_chan_release(struct kobject *kobj)
1404 {
1405         struct vmbus_channel *channel
1406                 = container_of(kobj, struct vmbus_channel, kobj);
1407
1408         kfree_rcu(channel, rcu);
1409 }
1410
1411 struct vmbus_chan_attribute {
1412         struct attribute attr;
1413         ssize_t (*show)(struct vmbus_channel *chan, char *buf);
1414         ssize_t (*store)(struct vmbus_channel *chan,
1415                          const char *buf, size_t count);
1416 };
1417 #define VMBUS_CHAN_ATTR(_name, _mode, _show, _store) \
1418         struct vmbus_chan_attribute chan_attr_##_name \
1419                 = __ATTR(_name, _mode, _show, _store)
1420 #define VMBUS_CHAN_ATTR_RW(_name) \
1421         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RW(_name)
1422 #define VMBUS_CHAN_ATTR_RO(_name) \
1423         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RO(_name)
1424 #define VMBUS_CHAN_ATTR_WO(_name) \
1425         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_WO(_name)
1426
1427 static ssize_t vmbus_chan_attr_show(struct kobject *kobj,
1428                                     struct attribute *attr, char *buf)
1429 {
1430         const struct vmbus_chan_attribute *attribute
1431                 = container_of(attr, struct vmbus_chan_attribute, attr);
1432         struct vmbus_channel *chan
1433                 = container_of(kobj, struct vmbus_channel, kobj);
1434
1435         if (!attribute->show)
1436                 return -EIO;
1437
1438         return attribute->show(chan, buf);
1439 }
1440
1441 static const struct sysfs_ops vmbus_chan_sysfs_ops = {
1442         .show = vmbus_chan_attr_show,
1443 };
1444
1445 static ssize_t out_mask_show(struct vmbus_channel *channel, char *buf)
1446 {
1447         struct hv_ring_buffer_info *rbi = &channel->outbound;
1448         ssize_t ret;
1449
1450         mutex_lock(&rbi->ring_buffer_mutex);
1451         if (!rbi->ring_buffer) {
1452                 mutex_unlock(&rbi->ring_buffer_mutex);
1453                 return -EINVAL;
1454         }
1455
1456         ret = sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1457         mutex_unlock(&rbi->ring_buffer_mutex);
1458         return ret;
1459 }
1460 static VMBUS_CHAN_ATTR_RO(out_mask);
1461
1462 static ssize_t in_mask_show(struct vmbus_channel *channel, char *buf)
1463 {
1464         struct hv_ring_buffer_info *rbi = &channel->inbound;
1465         ssize_t ret;
1466
1467         mutex_lock(&rbi->ring_buffer_mutex);
1468         if (!rbi->ring_buffer) {
1469                 mutex_unlock(&rbi->ring_buffer_mutex);
1470                 return -EINVAL;
1471         }
1472
1473         ret = sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1474         mutex_unlock(&rbi->ring_buffer_mutex);
1475         return ret;
1476 }
1477 static VMBUS_CHAN_ATTR_RO(in_mask);
1478
1479 static ssize_t read_avail_show(struct vmbus_channel *channel, char *buf)
1480 {
1481         struct hv_ring_buffer_info *rbi = &channel->inbound;
1482         ssize_t ret;
1483
1484         mutex_lock(&rbi->ring_buffer_mutex);
1485         if (!rbi->ring_buffer) {
1486                 mutex_unlock(&rbi->ring_buffer_mutex);
1487                 return -EINVAL;
1488         }
1489
1490         ret = sprintf(buf, "%u\n", hv_get_bytes_to_read(rbi));
1491         mutex_unlock(&rbi->ring_buffer_mutex);
1492         return ret;
1493 }
1494 static VMBUS_CHAN_ATTR_RO(read_avail);
1495
1496 static ssize_t write_avail_show(struct vmbus_channel *channel, char *buf)
1497 {
1498         struct hv_ring_buffer_info *rbi = &channel->outbound;
1499         ssize_t ret;
1500
1501         mutex_lock(&rbi->ring_buffer_mutex);
1502         if (!rbi->ring_buffer) {
1503                 mutex_unlock(&rbi->ring_buffer_mutex);
1504                 return -EINVAL;
1505         }
1506
1507         ret = sprintf(buf, "%u\n", hv_get_bytes_to_write(rbi));
1508         mutex_unlock(&rbi->ring_buffer_mutex);
1509         return ret;
1510 }
1511 static VMBUS_CHAN_ATTR_RO(write_avail);
1512
1513 static ssize_t show_target_cpu(struct vmbus_channel *channel, char *buf)
1514 {
1515         return sprintf(buf, "%u\n", channel->target_cpu);
1516 }
1517 static VMBUS_CHAN_ATTR(cpu, S_IRUGO, show_target_cpu, NULL);
1518
1519 static ssize_t channel_pending_show(struct vmbus_channel *channel,
1520                                     char *buf)
1521 {
1522         return sprintf(buf, "%d\n",
1523                        channel_pending(channel,
1524                                        vmbus_connection.monitor_pages[1]));
1525 }
1526 static VMBUS_CHAN_ATTR(pending, S_IRUGO, channel_pending_show, NULL);
1527
1528 static ssize_t channel_latency_show(struct vmbus_channel *channel,
1529                                     char *buf)
1530 {
1531         return sprintf(buf, "%d\n",
1532                        channel_latency(channel,
1533                                        vmbus_connection.monitor_pages[1]));
1534 }
1535 static VMBUS_CHAN_ATTR(latency, S_IRUGO, channel_latency_show, NULL);
1536
1537 static ssize_t channel_interrupts_show(struct vmbus_channel *channel, char *buf)
1538 {
1539         return sprintf(buf, "%llu\n", channel->interrupts);
1540 }
1541 static VMBUS_CHAN_ATTR(interrupts, S_IRUGO, channel_interrupts_show, NULL);
1542
1543 static ssize_t channel_events_show(struct vmbus_channel *channel, char *buf)
1544 {
1545         return sprintf(buf, "%llu\n", channel->sig_events);
1546 }
1547 static VMBUS_CHAN_ATTR(events, S_IRUGO, channel_events_show, NULL);
1548
1549 static ssize_t channel_intr_in_full_show(struct vmbus_channel *channel,
1550                                          char *buf)
1551 {
1552         return sprintf(buf, "%llu\n",
1553                        (unsigned long long)channel->intr_in_full);
1554 }
1555 static VMBUS_CHAN_ATTR(intr_in_full, 0444, channel_intr_in_full_show, NULL);
1556
1557 static ssize_t channel_intr_out_empty_show(struct vmbus_channel *channel,
1558                                            char *buf)
1559 {
1560         return sprintf(buf, "%llu\n",
1561                        (unsigned long long)channel->intr_out_empty);
1562 }
1563 static VMBUS_CHAN_ATTR(intr_out_empty, 0444, channel_intr_out_empty_show, NULL);
1564
1565 static ssize_t channel_out_full_first_show(struct vmbus_channel *channel,
1566                                            char *buf)
1567 {
1568         return sprintf(buf, "%llu\n",
1569                        (unsigned long long)channel->out_full_first);
1570 }
1571 static VMBUS_CHAN_ATTR(out_full_first, 0444, channel_out_full_first_show, NULL);
1572
1573 static ssize_t channel_out_full_total_show(struct vmbus_channel *channel,
1574                                            char *buf)
1575 {
1576         return sprintf(buf, "%llu\n",
1577                        (unsigned long long)channel->out_full_total);
1578 }
1579 static VMBUS_CHAN_ATTR(out_full_total, 0444, channel_out_full_total_show, NULL);
1580
1581 static ssize_t subchannel_monitor_id_show(struct vmbus_channel *channel,
1582                                           char *buf)
1583 {
1584         return sprintf(buf, "%u\n", channel->offermsg.monitorid);
1585 }
1586 static VMBUS_CHAN_ATTR(monitor_id, S_IRUGO, subchannel_monitor_id_show, NULL);
1587
1588 static ssize_t subchannel_id_show(struct vmbus_channel *channel,
1589                                   char *buf)
1590 {
1591         return sprintf(buf, "%u\n",
1592                        channel->offermsg.offer.sub_channel_index);
1593 }
1594 static VMBUS_CHAN_ATTR_RO(subchannel_id);
1595
1596 static struct attribute *vmbus_chan_attrs[] = {
1597         &chan_attr_out_mask.attr,
1598         &chan_attr_in_mask.attr,
1599         &chan_attr_read_avail.attr,
1600         &chan_attr_write_avail.attr,
1601         &chan_attr_cpu.attr,
1602         &chan_attr_pending.attr,
1603         &chan_attr_latency.attr,
1604         &chan_attr_interrupts.attr,
1605         &chan_attr_events.attr,
1606         &chan_attr_intr_in_full.attr,
1607         &chan_attr_intr_out_empty.attr,
1608         &chan_attr_out_full_first.attr,
1609         &chan_attr_out_full_total.attr,
1610         &chan_attr_monitor_id.attr,
1611         &chan_attr_subchannel_id.attr,
1612         NULL
1613 };
1614
1615 /*
1616  * Channel-level attribute_group callback function. Returns the permission for
1617  * each attribute, and returns 0 if an attribute is not visible.
1618  */
1619 static umode_t vmbus_chan_attr_is_visible(struct kobject *kobj,
1620                                           struct attribute *attr, int idx)
1621 {
1622         const struct vmbus_channel *channel =
1623                 container_of(kobj, struct vmbus_channel, kobj);
1624
1625         /* Hide the monitor attributes if the monitor mechanism is not used. */
1626         if (!channel->offermsg.monitor_allocated &&
1627             (attr == &chan_attr_pending.attr ||
1628              attr == &chan_attr_latency.attr ||
1629              attr == &chan_attr_monitor_id.attr))
1630                 return 0;
1631
1632         return attr->mode;
1633 }
1634
1635 static struct attribute_group vmbus_chan_group = {
1636         .attrs = vmbus_chan_attrs,
1637         .is_visible = vmbus_chan_attr_is_visible
1638 };
1639
1640 static struct kobj_type vmbus_chan_ktype = {
1641         .sysfs_ops = &vmbus_chan_sysfs_ops,
1642         .release = vmbus_chan_release,
1643 };
1644
1645 /*
1646  * vmbus_add_channel_kobj - setup a sub-directory under device/channels
1647  */
1648 int vmbus_add_channel_kobj(struct hv_device *dev, struct vmbus_channel *channel)
1649 {
1650         const struct device *device = &dev->device;
1651         struct kobject *kobj = &channel->kobj;
1652         u32 relid = channel->offermsg.child_relid;
1653         int ret;
1654
1655         kobj->kset = dev->channels_kset;
1656         ret = kobject_init_and_add(kobj, &vmbus_chan_ktype, NULL,
1657                                    "%u", relid);
1658         if (ret)
1659                 return ret;
1660
1661         ret = sysfs_create_group(kobj, &vmbus_chan_group);
1662
1663         if (ret) {
1664                 /*
1665                  * The calling functions' error handling paths will cleanup the
1666                  * empty channel directory.
1667                  */
1668                 dev_err(device, "Unable to set up channel sysfs files\n");
1669                 return ret;
1670         }
1671
1672         kobject_uevent(kobj, KOBJ_ADD);
1673
1674         return 0;
1675 }
1676
1677 /*
1678  * vmbus_remove_channel_attr_group - remove the channel's attribute group
1679  */
1680 void vmbus_remove_channel_attr_group(struct vmbus_channel *channel)
1681 {
1682         sysfs_remove_group(&channel->kobj, &vmbus_chan_group);
1683 }
1684
1685 /*
1686  * vmbus_device_create - Creates and registers a new child device
1687  * on the vmbus.
1688  */
1689 struct hv_device *vmbus_device_create(const guid_t *type,
1690                                       const guid_t *instance,
1691                                       struct vmbus_channel *channel)
1692 {
1693         struct hv_device *child_device_obj;
1694
1695         child_device_obj = kzalloc(sizeof(struct hv_device), GFP_KERNEL);
1696         if (!child_device_obj) {
1697                 pr_err("Unable to allocate device object for child device\n");
1698                 return NULL;
1699         }
1700
1701         child_device_obj->channel = channel;
1702         guid_copy(&child_device_obj->dev_type, type);
1703         guid_copy(&child_device_obj->dev_instance, instance);
1704         child_device_obj->vendor_id = 0x1414; /* MSFT vendor ID */
1705
1706         return child_device_obj;
1707 }
1708
1709 /*
1710  * vmbus_device_register - Register the child device
1711  */
1712 int vmbus_device_register(struct hv_device *child_device_obj)
1713 {
1714         struct kobject *kobj = &child_device_obj->device.kobj;
1715         int ret;
1716
1717         dev_set_name(&child_device_obj->device, "%pUl",
1718                      child_device_obj->channel->offermsg.offer.if_instance.b);
1719
1720         child_device_obj->device.bus = &hv_bus;
1721         child_device_obj->device.parent = &hv_acpi_dev->dev;
1722         child_device_obj->device.release = vmbus_device_release;
1723
1724         /*
1725          * Register with the LDM. This will kick off the driver/device
1726          * binding...which will eventually call vmbus_match() and vmbus_probe()
1727          */
1728         ret = device_register(&child_device_obj->device);
1729         if (ret) {
1730                 pr_err("Unable to register child device\n");
1731                 return ret;
1732         }
1733
1734         child_device_obj->channels_kset = kset_create_and_add("channels",
1735                                                               NULL, kobj);
1736         if (!child_device_obj->channels_kset) {
1737                 ret = -ENOMEM;
1738                 goto err_dev_unregister;
1739         }
1740
1741         ret = vmbus_add_channel_kobj(child_device_obj,
1742                                      child_device_obj->channel);
1743         if (ret) {
1744                 pr_err("Unable to register primary channeln");
1745                 goto err_kset_unregister;
1746         }
1747
1748         return 0;
1749
1750 err_kset_unregister:
1751         kset_unregister(child_device_obj->channels_kset);
1752
1753 err_dev_unregister:
1754         device_unregister(&child_device_obj->device);
1755         return ret;
1756 }
1757
1758 /*
1759  * vmbus_device_unregister - Remove the specified child device
1760  * from the vmbus.
1761  */
1762 void vmbus_device_unregister(struct hv_device *device_obj)
1763 {
1764         pr_debug("child device %s unregistered\n",
1765                 dev_name(&device_obj->device));
1766
1767         kset_unregister(device_obj->channels_kset);
1768
1769         /*
1770          * Kick off the process of unregistering the device.
1771          * This will call vmbus_remove() and eventually vmbus_device_release()
1772          */
1773         device_unregister(&device_obj->device);
1774 }
1775
1776
1777 /*
1778  * VMBUS is an acpi enumerated device. Get the information we
1779  * need from DSDT.
1780  */
1781 #define VTPM_BASE_ADDRESS 0xfed40000
1782 static acpi_status vmbus_walk_resources(struct acpi_resource *res, void *ctx)
1783 {
1784         resource_size_t start = 0;
1785         resource_size_t end = 0;
1786         struct resource *new_res;
1787         struct resource **old_res = &hyperv_mmio;
1788         struct resource **prev_res = NULL;
1789
1790         switch (res->type) {
1791
1792         /*
1793          * "Address" descriptors are for bus windows. Ignore
1794          * "memory" descriptors, which are for registers on
1795          * devices.
1796          */
1797         case ACPI_RESOURCE_TYPE_ADDRESS32:
1798                 start = res->data.address32.address.minimum;
1799                 end = res->data.address32.address.maximum;
1800                 break;
1801
1802         case ACPI_RESOURCE_TYPE_ADDRESS64:
1803                 start = res->data.address64.address.minimum;
1804                 end = res->data.address64.address.maximum;
1805                 break;
1806
1807         default:
1808                 /* Unused resource type */
1809                 return AE_OK;
1810
1811         }
1812         /*
1813          * Ignore ranges that are below 1MB, as they're not
1814          * necessary or useful here.
1815          */
1816         if (end < 0x100000)
1817                 return AE_OK;
1818
1819         new_res = kzalloc(sizeof(*new_res), GFP_ATOMIC);
1820         if (!new_res)
1821                 return AE_NO_MEMORY;
1822
1823         /* If this range overlaps the virtual TPM, truncate it. */
1824         if (end > VTPM_BASE_ADDRESS && start < VTPM_BASE_ADDRESS)
1825                 end = VTPM_BASE_ADDRESS;
1826
1827         new_res->name = "hyperv mmio";
1828         new_res->flags = IORESOURCE_MEM;
1829         new_res->start = start;
1830         new_res->end = end;
1831
1832         /*
1833          * If two ranges are adjacent, merge them.
1834          */
1835         do {
1836                 if (!*old_res) {
1837                         *old_res = new_res;
1838                         break;
1839                 }
1840
1841                 if (((*old_res)->end + 1) == new_res->start) {
1842                         (*old_res)->end = new_res->end;
1843                         kfree(new_res);
1844                         break;
1845                 }
1846
1847                 if ((*old_res)->start == new_res->end + 1) {
1848                         (*old_res)->start = new_res->start;
1849                         kfree(new_res);
1850                         break;
1851                 }
1852
1853                 if ((*old_res)->start > new_res->end) {
1854                         new_res->sibling = *old_res;
1855                         if (prev_res)
1856                                 (*prev_res)->sibling = new_res;
1857                         *old_res = new_res;
1858                         break;
1859                 }
1860
1861                 prev_res = old_res;
1862                 old_res = &(*old_res)->sibling;
1863
1864         } while (1);
1865
1866         return AE_OK;
1867 }
1868
1869 static int vmbus_acpi_remove(struct acpi_device *device)
1870 {
1871         struct resource *cur_res;
1872         struct resource *next_res;
1873
1874         if (hyperv_mmio) {
1875                 if (fb_mmio) {
1876                         __release_region(hyperv_mmio, fb_mmio->start,
1877                                          resource_size(fb_mmio));
1878                         fb_mmio = NULL;
1879                 }
1880
1881                 for (cur_res = hyperv_mmio; cur_res; cur_res = next_res) {
1882                         next_res = cur_res->sibling;
1883                         kfree(cur_res);
1884                 }
1885         }
1886
1887         return 0;
1888 }
1889
1890 static void vmbus_reserve_fb(void)
1891 {
1892         int size;
1893         /*
1894          * Make a claim for the frame buffer in the resource tree under the
1895          * first node, which will be the one below 4GB.  The length seems to
1896          * be underreported, particularly in a Generation 1 VM.  So start out
1897          * reserving a larger area and make it smaller until it succeeds.
1898          */
1899
1900         if (screen_info.lfb_base) {
1901                 if (efi_enabled(EFI_BOOT))
1902                         size = max_t(__u32, screen_info.lfb_size, 0x800000);
1903                 else
1904                         size = max_t(__u32, screen_info.lfb_size, 0x4000000);
1905
1906                 for (; !fb_mmio && (size >= 0x100000); size >>= 1) {
1907                         fb_mmio = __request_region(hyperv_mmio,
1908                                                    screen_info.lfb_base, size,
1909                                                    fb_mmio_name, 0);
1910                 }
1911         }
1912 }
1913
1914 /**
1915  * vmbus_allocate_mmio() - Pick a memory-mapped I/O range.
1916  * @new:                If successful, supplied a pointer to the
1917  *                      allocated MMIO space.
1918  * @device_obj:         Identifies the caller
1919  * @min:                Minimum guest physical address of the
1920  *                      allocation
1921  * @max:                Maximum guest physical address
1922  * @size:               Size of the range to be allocated
1923  * @align:              Alignment of the range to be allocated
1924  * @fb_overlap_ok:      Whether this allocation can be allowed
1925  *                      to overlap the video frame buffer.
1926  *
1927  * This function walks the resources granted to VMBus by the
1928  * _CRS object in the ACPI namespace underneath the parent
1929  * "bridge" whether that's a root PCI bus in the Generation 1
1930  * case or a Module Device in the Generation 2 case.  It then
1931  * attempts to allocate from the global MMIO pool in a way that
1932  * matches the constraints supplied in these parameters and by
1933  * that _CRS.
1934  *
1935  * Return: 0 on success, -errno on failure
1936  */
1937 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
1938                         resource_size_t min, resource_size_t max,
1939                         resource_size_t size, resource_size_t align,
1940                         bool fb_overlap_ok)
1941 {
1942         struct resource *iter, *shadow;
1943         resource_size_t range_min, range_max, start;
1944         const char *dev_n = dev_name(&device_obj->device);
1945         int retval;
1946
1947         retval = -ENXIO;
1948         down(&hyperv_mmio_lock);
1949
1950         /*
1951          * If overlaps with frame buffers are allowed, then first attempt to
1952          * make the allocation from within the reserved region.  Because it
1953          * is already reserved, no shadow allocation is necessary.
1954          */
1955         if (fb_overlap_ok && fb_mmio && !(min > fb_mmio->end) &&
1956             !(max < fb_mmio->start)) {
1957
1958                 range_min = fb_mmio->start;
1959                 range_max = fb_mmio->end;
1960                 start = (range_min + align - 1) & ~(align - 1);
1961                 for (; start + size - 1 <= range_max; start += align) {
1962                         *new = request_mem_region_exclusive(start, size, dev_n);
1963                         if (*new) {
1964                                 retval = 0;
1965                                 goto exit;
1966                         }
1967                 }
1968         }
1969
1970         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
1971                 if ((iter->start >= max) || (iter->end <= min))
1972                         continue;
1973
1974                 range_min = iter->start;
1975                 range_max = iter->end;
1976                 start = (range_min + align - 1) & ~(align - 1);
1977                 for (; start + size - 1 <= range_max; start += align) {
1978                         shadow = __request_region(iter, start, size, NULL,
1979                                                   IORESOURCE_BUSY);
1980                         if (!shadow)
1981                                 continue;
1982
1983                         *new = request_mem_region_exclusive(start, size, dev_n);
1984                         if (*new) {
1985                                 shadow->name = (char *)*new;
1986                                 retval = 0;
1987                                 goto exit;
1988                         }
1989
1990                         __release_region(iter, start, size);
1991                 }
1992         }
1993
1994 exit:
1995         up(&hyperv_mmio_lock);
1996         return retval;
1997 }
1998 EXPORT_SYMBOL_GPL(vmbus_allocate_mmio);
1999
2000 /**
2001  * vmbus_free_mmio() - Free a memory-mapped I/O range.
2002  * @start:              Base address of region to release.
2003  * @size:               Size of the range to be allocated
2004  *
2005  * This function releases anything requested by
2006  * vmbus_mmio_allocate().
2007  */
2008 void vmbus_free_mmio(resource_size_t start, resource_size_t size)
2009 {
2010         struct resource *iter;
2011
2012         down(&hyperv_mmio_lock);
2013         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
2014                 if ((iter->start >= start + size) || (iter->end <= start))
2015                         continue;
2016
2017                 __release_region(iter, start, size);
2018         }
2019         release_mem_region(start, size);
2020         up(&hyperv_mmio_lock);
2021
2022 }
2023 EXPORT_SYMBOL_GPL(vmbus_free_mmio);
2024
2025 static int vmbus_acpi_add(struct acpi_device *device)
2026 {
2027         acpi_status result;
2028         int ret_val = -ENODEV;
2029         struct acpi_device *ancestor;
2030
2031         hv_acpi_dev = device;
2032
2033         result = acpi_walk_resources(device->handle, METHOD_NAME__CRS,
2034                                         vmbus_walk_resources, NULL);
2035
2036         if (ACPI_FAILURE(result))
2037                 goto acpi_walk_err;
2038         /*
2039          * Some ancestor of the vmbus acpi device (Gen1 or Gen2
2040          * firmware) is the VMOD that has the mmio ranges. Get that.
2041          */
2042         for (ancestor = device->parent; ancestor; ancestor = ancestor->parent) {
2043                 result = acpi_walk_resources(ancestor->handle, METHOD_NAME__CRS,
2044                                              vmbus_walk_resources, NULL);
2045
2046                 if (ACPI_FAILURE(result))
2047                         continue;
2048                 if (hyperv_mmio) {
2049                         vmbus_reserve_fb();
2050                         break;
2051                 }
2052         }
2053         ret_val = 0;
2054
2055 acpi_walk_err:
2056         complete(&probe_event);
2057         if (ret_val)
2058                 vmbus_acpi_remove(device);
2059         return ret_val;
2060 }
2061
2062 static const struct acpi_device_id vmbus_acpi_device_ids[] = {
2063         {"VMBUS", 0},
2064         {"VMBus", 0},
2065         {"", 0},
2066 };
2067 MODULE_DEVICE_TABLE(acpi, vmbus_acpi_device_ids);
2068
2069 static struct acpi_driver vmbus_acpi_driver = {
2070         .name = "vmbus",
2071         .ids = vmbus_acpi_device_ids,
2072         .ops = {
2073                 .add = vmbus_acpi_add,
2074                 .remove = vmbus_acpi_remove,
2075         },
2076 };
2077
2078 static void hv_kexec_handler(void)
2079 {
2080         hv_synic_clockevents_cleanup();
2081         vmbus_initiate_unload(false);
2082         vmbus_connection.conn_state = DISCONNECTED;
2083         /* Make sure conn_state is set as hv_synic_cleanup checks for it */
2084         mb();
2085         cpuhp_remove_state(hyperv_cpuhp_online);
2086         hyperv_cleanup();
2087 };
2088
2089 static void hv_crash_handler(struct pt_regs *regs)
2090 {
2091         vmbus_initiate_unload(true);
2092         /*
2093          * In crash handler we can't schedule synic cleanup for all CPUs,
2094          * doing the cleanup for current CPU only. This should be sufficient
2095          * for kdump.
2096          */
2097         vmbus_connection.conn_state = DISCONNECTED;
2098         hv_synic_cleanup(smp_processor_id());
2099         hyperv_cleanup();
2100 };
2101
2102 static int __init hv_acpi_init(void)
2103 {
2104         int ret, t;
2105
2106         if (!hv_is_hyperv_initialized())
2107                 return -ENODEV;
2108
2109         init_completion(&probe_event);
2110
2111         /*
2112          * Get ACPI resources first.
2113          */
2114         ret = acpi_bus_register_driver(&vmbus_acpi_driver);
2115
2116         if (ret)
2117                 return ret;
2118
2119         t = wait_for_completion_timeout(&probe_event, 5*HZ);
2120         if (t == 0) {
2121                 ret = -ETIMEDOUT;
2122                 goto cleanup;
2123         }
2124
2125         ret = vmbus_bus_init();
2126         if (ret)
2127                 goto cleanup;
2128
2129         hv_setup_kexec_handler(hv_kexec_handler);
2130         hv_setup_crash_handler(hv_crash_handler);
2131
2132         return 0;
2133
2134 cleanup:
2135         acpi_bus_unregister_driver(&vmbus_acpi_driver);
2136         hv_acpi_dev = NULL;
2137         return ret;
2138 }
2139
2140 static void __exit vmbus_exit(void)
2141 {
2142         int cpu;
2143
2144         hv_remove_kexec_handler();
2145         hv_remove_crash_handler();
2146         vmbus_connection.conn_state = DISCONNECTED;
2147         hv_synic_clockevents_cleanup();
2148         vmbus_disconnect();
2149         hv_remove_vmbus_irq();
2150         for_each_online_cpu(cpu) {
2151                 struct hv_per_cpu_context *hv_cpu
2152                         = per_cpu_ptr(hv_context.cpu_context, cpu);
2153
2154                 tasklet_kill(&hv_cpu->msg_dpc);
2155         }
2156         vmbus_free_channels();
2157
2158         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
2159                 kmsg_dump_unregister(&hv_kmsg_dumper);
2160                 unregister_die_notifier(&hyperv_die_block);
2161                 atomic_notifier_chain_unregister(&panic_notifier_list,
2162                                                  &hyperv_panic_block);
2163         }
2164
2165         free_page((unsigned long)hv_panic_page);
2166         unregister_sysctl_table(hv_ctl_table_hdr);
2167         hv_ctl_table_hdr = NULL;
2168         bus_unregister(&hv_bus);
2169
2170         cpuhp_remove_state(hyperv_cpuhp_online);
2171         hv_synic_free();
2172         acpi_bus_unregister_driver(&vmbus_acpi_driver);
2173 }
2174
2175
2176 MODULE_LICENSE("GPL");
2177
2178 subsys_initcall(hv_acpi_init);
2179 module_exit(vmbus_exit);