drivers/hv/vmbus_drv.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (c) 2009, Microsoft Corporation.
   4  *
   5  * Authors:
   6  *   Haiyang Zhang <haiyangz@microsoft.com>
   7  *   Hank Janssen  <hjanssen@microsoft.com>
   8  *   K. Y. Srinivasan <kys@microsoft.com>
   9  */
  10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  11
  12 #include <linux/init.h>
  13 #include <linux/module.h>
  14 #include <linux/device.h>
  15 #include <linux/interrupt.h>
  16 #include <linux/sysctl.h>
  17 #include <linux/slab.h>
  18 #include <linux/acpi.h>
  19 #include <linux/completion.h>
  20 #include <linux/hyperv.h>
  21 #include <linux/kernel_stat.h>
  22 #include <linux/clockchips.h>
  23 #include <linux/cpu.h>
  24 #include <linux/sched/task_stack.h>
  25
  26 #include <asm/mshyperv.h>
  27 #include <linux/delay.h>
  28 #include <linux/notifier.h>
  29 #include <linux/ptrace.h>
  30 #include <linux/screen_info.h>
  31 #include <linux/kdebug.h>
  32 #include <linux/efi.h>
  33 #include <linux/random.h>
  34 #include <linux/syscore_ops.h>
  35 #include <clocksource/hyperv_timer.h>
  36 #include "hyperv_vmbus.h"
  37
  38 struct vmbus_dynid {
  39         struct list_head node;
  40         struct hv_vmbus_device_id id;
  41 };
  42
  43 static struct acpi_device  *hv_acpi_dev;
  44
  45 static struct completion probe_event;
  46
  47 static int hyperv_cpuhp_online;
  48
  49 static void *hv_panic_page;
  50
  51 static int hyperv_panic_event(struct notifier_block *nb, unsigned long val,
  52                               void *args)
  53 {
  54         struct pt_regs *regs;
  55
  56         regs = current_pt_regs();
  57
  58         hyperv_report_panic(regs, val);
  59         return NOTIFY_DONE;
  60 }
  61
  62 static int hyperv_die_event(struct notifier_block *nb, unsigned long val,
  63                             void *args)
  64 {
  65         struct die_args *die = (struct die_args *)args;
  66         struct pt_regs *regs = die->regs;
  67
  68         hyperv_report_panic(regs, val);
  69         return NOTIFY_DONE;
  70 }
  71
  72 static struct notifier_block hyperv_die_block = {
  73         .notifier_call = hyperv_die_event,
  74 };
  75 static struct notifier_block hyperv_panic_block = {
  76         .notifier_call = hyperv_panic_event,
  77 };
  78
  79 static const char *fb_mmio_name = "fb_range";
  80 static struct resource *fb_mmio;
  81 static struct resource *hyperv_mmio;
  82 static DEFINE_SEMAPHORE(hyperv_mmio_lock);
  83
  84 static int vmbus_exists(void)
  85 {
  86         if (hv_acpi_dev == NULL)
  87                 return -ENODEV;
  88
  89         return 0;
  90 }
  91
  92 #define VMBUS_ALIAS_LEN ((sizeof((struct hv_vmbus_device_id *)0)->guid) * 2)
  93 static void print_alias_name(struct hv_device *hv_dev, char *alias_name)
  94 {
  95         int i;
  96         for (i = 0; i < VMBUS_ALIAS_LEN; i += 2)
  97                 sprintf(&alias_name[i], "%02x", hv_dev->dev_type.b[i/2]);
  98 }
  99
 100 static u8 channel_monitor_group(const struct vmbus_channel *channel)
 101 {
 102         return (u8)channel->offermsg.monitorid / 32;
 103 }
 104
 105 static u8 channel_monitor_offset(const struct vmbus_channel *channel)
 106 {
 107         return (u8)channel->offermsg.monitorid % 32;
 108 }
 109
 110 static u32 channel_pending(const struct vmbus_channel *channel,
 111                            const struct hv_monitor_page *monitor_page)
 112 {
 113         u8 monitor_group = channel_monitor_group(channel);
 114
 115         return monitor_page->trigger_group[monitor_group].pending;
 116 }
 117
 118 static u32 channel_latency(const struct vmbus_channel *channel,
 119                            const struct hv_monitor_page *monitor_page)
 120 {
 121         u8 monitor_group = channel_monitor_group(channel);
 122         u8 monitor_offset = channel_monitor_offset(channel);
 123
 124         return monitor_page->latency[monitor_group][monitor_offset];
 125 }
 126
 127 static u32 channel_conn_id(struct vmbus_channel *channel,
 128                            struct hv_monitor_page *monitor_page)
 129 {
 130         u8 monitor_group = channel_monitor_group(channel);
 131         u8 monitor_offset = channel_monitor_offset(channel);
 132         return monitor_page->parameter[monitor_group][monitor_offset].connectionid.u.id;
 133 }
 134
 135 static ssize_t id_show(struct device *dev, struct device_attribute *dev_attr,
 136                        char *buf)
 137 {
 138         struct hv_device *hv_dev = device_to_hv_device(dev);
 139
 140         if (!hv_dev->channel)
 141                 return -ENODEV;
 142         return sprintf(buf, "%d\n", hv_dev->channel->offermsg.child_relid);
 143 }
 144 static DEVICE_ATTR_RO(id);
 145
 146 static ssize_t state_show(struct device *dev, struct device_attribute *dev_attr,
 147                           char *buf)
 148 {
 149         struct hv_device *hv_dev = device_to_hv_device(dev);
 150
 151         if (!hv_dev->channel)
 152                 return -ENODEV;
 153         return sprintf(buf, "%d\n", hv_dev->channel->state);
 154 }
 155 static DEVICE_ATTR_RO(state);
 156
 157 static ssize_t monitor_id_show(struct device *dev,
 158                                struct device_attribute *dev_attr, char *buf)
 159 {
 160         struct hv_device *hv_dev = device_to_hv_device(dev);
 161
 162         if (!hv_dev->channel)
 163                 return -ENODEV;
 164         return sprintf(buf, "%d\n", hv_dev->channel->offermsg.monitorid);
 165 }
 166 static DEVICE_ATTR_RO(monitor_id);
 167
 168 static ssize_t class_id_show(struct device *dev,
 169                                struct device_attribute *dev_attr, char *buf)
 170 {
 171         struct hv_device *hv_dev = device_to_hv_device(dev);
 172
 173         if (!hv_dev->channel)
 174                 return -ENODEV;
 175         return sprintf(buf, "{%pUl}\n",
 176                        hv_dev->channel->offermsg.offer.if_type.b);
 177 }
 178 static DEVICE_ATTR_RO(class_id);
 179
 180 static ssize_t device_id_show(struct device *dev,
 181                               struct device_attribute *dev_attr, char *buf)
 182 {
 183         struct hv_device *hv_dev = device_to_hv_device(dev);
 184
 185         if (!hv_dev->channel)
 186                 return -ENODEV;
 187         return sprintf(buf, "{%pUl}\n",
 188                        hv_dev->channel->offermsg.offer.if_instance.b);
 189 }
 190 static DEVICE_ATTR_RO(device_id);
 191
 192 static ssize_t modalias_show(struct device *dev,
 193                              struct device_attribute *dev_attr, char *buf)
 194 {
 195         struct hv_device *hv_dev = device_to_hv_device(dev);
 196         char alias_name[VMBUS_ALIAS_LEN + 1];
 197
 198         print_alias_name(hv_dev, alias_name);
 199         return sprintf(buf, "vmbus:%s\n", alias_name);
 200 }
 201 static DEVICE_ATTR_RO(modalias);
 202
 203 #ifdef CONFIG_NUMA
 204 static ssize_t numa_node_show(struct device *dev,
 205                               struct device_attribute *attr, char *buf)
 206 {
 207         struct hv_device *hv_dev = device_to_hv_device(dev);
 208
 209         if (!hv_dev->channel)
 210                 return -ENODEV;
 211
 212         return sprintf(buf, "%d\n", hv_dev->channel->numa_node);
 213 }
 214 static DEVICE_ATTR_RO(numa_node);
 215 #endif
 216
 217 static ssize_t server_monitor_pending_show(struct device *dev,
 218                                            struct device_attribute *dev_attr,
 219                                            char *buf)
 220 {
 221         struct hv_device *hv_dev = device_to_hv_device(dev);
 222
 223         if (!hv_dev->channel)
 224                 return -ENODEV;
 225         return sprintf(buf, "%d\n",
 226                        channel_pending(hv_dev->channel,
 227                                        vmbus_connection.monitor_pages[0]));
 228 }
 229 static DEVICE_ATTR_RO(server_monitor_pending);
 230
 231 static ssize_t client_monitor_pending_show(struct device *dev,
 232                                            struct device_attribute *dev_attr,
 233                                            char *buf)
 234 {
 235         struct hv_device *hv_dev = device_to_hv_device(dev);
 236
 237         if (!hv_dev->channel)
 238                 return -ENODEV;
 239         return sprintf(buf, "%d\n",
 240                        channel_pending(hv_dev->channel,
 241                                        vmbus_connection.monitor_pages[1]));
 242 }
 243 static DEVICE_ATTR_RO(client_monitor_pending);
 244
 245 static ssize_t server_monitor_latency_show(struct device *dev,
 246                                            struct device_attribute *dev_attr,
 247                                            char *buf)
 248 {
 249         struct hv_device *hv_dev = device_to_hv_device(dev);
 250
 251         if (!hv_dev->channel)
 252                 return -ENODEV;
 253         return sprintf(buf, "%d\n",
 254                        channel_latency(hv_dev->channel,
 255                                        vmbus_connection.monitor_pages[0]));
 256 }
 257 static DEVICE_ATTR_RO(server_monitor_latency);
 258
 259 static ssize_t client_monitor_latency_show(struct device *dev,
 260                                            struct device_attribute *dev_attr,
 261                                            char *buf)
 262 {
 263         struct hv_device *hv_dev = device_to_hv_device(dev);
 264
 265         if (!hv_dev->channel)
 266                 return -ENODEV;
 267         return sprintf(buf, "%d\n",
 268                        channel_latency(hv_dev->channel,
 269                                        vmbus_connection.monitor_pages[1]));
 270 }
 271 static DEVICE_ATTR_RO(client_monitor_latency);
 272
 273 static ssize_t server_monitor_conn_id_show(struct device *dev,
 274                                            struct device_attribute *dev_attr,
 275                                            char *buf)
 276 {
 277         struct hv_device *hv_dev = device_to_hv_device(dev);
 278
 279         if (!hv_dev->channel)
 280                 return -ENODEV;
 281         return sprintf(buf, "%d\n",
 282                        channel_conn_id(hv_dev->channel,
 283                                        vmbus_connection.monitor_pages[0]));
 284 }
 285 static DEVICE_ATTR_RO(server_monitor_conn_id);
 286
 287 static ssize_t client_monitor_conn_id_show(struct device *dev,
 288                                            struct device_attribute *dev_attr,
 289                                            char *buf)
 290 {
 291         struct hv_device *hv_dev = device_to_hv_device(dev);
 292
 293         if (!hv_dev->channel)
 294                 return -ENODEV;
 295         return sprintf(buf, "%d\n",
 296                        channel_conn_id(hv_dev->channel,
 297                                        vmbus_connection.monitor_pages[1]));
 298 }
 299 static DEVICE_ATTR_RO(client_monitor_conn_id);
 300
 301 static ssize_t out_intr_mask_show(struct device *dev,
 302                                   struct device_attribute *dev_attr, char *buf)
 303 {
 304         struct hv_device *hv_dev = device_to_hv_device(dev);
 305         struct hv_ring_buffer_debug_info outbound;
 306         int ret;
 307
 308         if (!hv_dev->channel)
 309                 return -ENODEV;
 310
 311         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 312                                           &outbound);
 313         if (ret < 0)
 314                 return ret;
 315
 316         return sprintf(buf, "%d\n", outbound.current_interrupt_mask);
 317 }
 318 static DEVICE_ATTR_RO(out_intr_mask);
 319
 320 static ssize_t out_read_index_show(struct device *dev,
 321                                    struct device_attribute *dev_attr, char *buf)
 322 {
 323         struct hv_device *hv_dev = device_to_hv_device(dev);
 324         struct hv_ring_buffer_debug_info outbound;
 325         int ret;
 326
 327         if (!hv_dev->channel)
 328                 return -ENODEV;
 329
 330         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 331                                           &outbound);
 332         if (ret < 0)
 333                 return ret;
 334         return sprintf(buf, "%d\n", outbound.current_read_index);
 335 }
 336 static DEVICE_ATTR_RO(out_read_index);
 337
 338 static ssize_t out_write_index_show(struct device *dev,
 339                                     struct device_attribute *dev_attr,
 340                                     char *buf)
 341 {
 342         struct hv_device *hv_dev = device_to_hv_device(dev);
 343         struct hv_ring_buffer_debug_info outbound;
 344         int ret;
 345
 346         if (!hv_dev->channel)
 347                 return -ENODEV;
 348
 349         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 350                                           &outbound);
 351         if (ret < 0)
 352                 return ret;
 353         return sprintf(buf, "%d\n", outbound.current_write_index);
 354 }
 355 static DEVICE_ATTR_RO(out_write_index);
 356
 357 static ssize_t out_read_bytes_avail_show(struct device *dev,
 358                                          struct device_attribute *dev_attr,
 359                                          char *buf)
 360 {
 361         struct hv_device *hv_dev = device_to_hv_device(dev);
 362         struct hv_ring_buffer_debug_info outbound;
 363         int ret;
 364
 365         if (!hv_dev->channel)
 366                 return -ENODEV;
 367
 368         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 369                                           &outbound);
 370         if (ret < 0)
 371                 return ret;
 372         return sprintf(buf, "%d\n", outbound.bytes_avail_toread);
 373 }
 374 static DEVICE_ATTR_RO(out_read_bytes_avail);
 375
 376 static ssize_t out_write_bytes_avail_show(struct device *dev,
 377                                           struct device_attribute *dev_attr,
 378                                           char *buf)
 379 {
 380         struct hv_device *hv_dev = device_to_hv_device(dev);
 381         struct hv_ring_buffer_debug_info outbound;
 382         int ret;
 383
 384         if (!hv_dev->channel)
 385                 return -ENODEV;
 386
 387         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 388                                           &outbound);
 389         if (ret < 0)
 390                 return ret;
 391         return sprintf(buf, "%d\n", outbound.bytes_avail_towrite);
 392 }
 393 static DEVICE_ATTR_RO(out_write_bytes_avail);
 394
 395 static ssize_t in_intr_mask_show(struct device *dev,
 396                                  struct device_attribute *dev_attr, char *buf)
 397 {
 398         struct hv_device *hv_dev = device_to_hv_device(dev);
 399         struct hv_ring_buffer_debug_info inbound;
 400         int ret;
 401
 402         if (!hv_dev->channel)
 403                 return -ENODEV;
 404
 405         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 406         if (ret < 0)
 407                 return ret;
 408
 409         return sprintf(buf, "%d\n", inbound.current_interrupt_mask);
 410 }
 411 static DEVICE_ATTR_RO(in_intr_mask);
 412
 413 static ssize_t in_read_index_show(struct device *dev,
 414                                   struct device_attribute *dev_attr, char *buf)
 415 {
 416         struct hv_device *hv_dev = device_to_hv_device(dev);
 417         struct hv_ring_buffer_debug_info inbound;
 418         int ret;
 419
 420         if (!hv_dev->channel)
 421                 return -ENODEV;
 422
 423         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 424         if (ret < 0)
 425                 return ret;
 426
 427         return sprintf(buf, "%d\n", inbound.current_read_index);
 428 }
 429 static DEVICE_ATTR_RO(in_read_index);
 430
 431 static ssize_t in_write_index_show(struct device *dev,
 432                                    struct device_attribute *dev_attr, char *buf)
 433 {
 434         struct hv_device *hv_dev = device_to_hv_device(dev);
 435         struct hv_ring_buffer_debug_info inbound;
 436         int ret;
 437
 438         if (!hv_dev->channel)
 439                 return -ENODEV;
 440
 441         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 442         if (ret < 0)
 443                 return ret;
 444
 445         return sprintf(buf, "%d\n", inbound.current_write_index);
 446 }
 447 static DEVICE_ATTR_RO(in_write_index);
 448
 449 static ssize_t in_read_bytes_avail_show(struct device *dev,
 450                                         struct device_attribute *dev_attr,
 451                                         char *buf)
 452 {
 453         struct hv_device *hv_dev = device_to_hv_device(dev);
 454         struct hv_ring_buffer_debug_info inbound;
 455         int ret;
 456
 457         if (!hv_dev->channel)
 458                 return -ENODEV;
 459
 460         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 461         if (ret < 0)
 462                 return ret;
 463
 464         return sprintf(buf, "%d\n", inbound.bytes_avail_toread);
 465 }
 466 static DEVICE_ATTR_RO(in_read_bytes_avail);
 467
 468 static ssize_t in_write_bytes_avail_show(struct device *dev,
 469                                          struct device_attribute *dev_attr,
 470                                          char *buf)
 471 {
 472         struct hv_device *hv_dev = device_to_hv_device(dev);
 473         struct hv_ring_buffer_debug_info inbound;
 474         int ret;
 475
 476         if (!hv_dev->channel)
 477                 return -ENODEV;
 478
 479         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 480         if (ret < 0)
 481                 return ret;
 482
 483         return sprintf(buf, "%d\n", inbound.bytes_avail_towrite);
 484 }
 485 static DEVICE_ATTR_RO(in_write_bytes_avail);
 486
 487 static ssize_t channel_vp_mapping_show(struct device *dev,
 488                                        struct device_attribute *dev_attr,
 489                                        char *buf)
 490 {
 491         struct hv_device *hv_dev = device_to_hv_device(dev);
 492         struct vmbus_channel *channel = hv_dev->channel, *cur_sc;
 493         unsigned long flags;
 494         int buf_size = PAGE_SIZE, n_written, tot_written;
 495         struct list_head *cur;
 496
 497         if (!channel)
 498                 return -ENODEV;
 499
 500         tot_written = snprintf(buf, buf_size, "%u:%u\n",
 501                 channel->offermsg.child_relid, channel->target_cpu);
 502
 503         spin_lock_irqsave(&channel->lock, flags);
 504
 505         list_for_each(cur, &channel->sc_list) {
 506                 if (tot_written >= buf_size - 1)
 507                         break;
 508
 509                 cur_sc = list_entry(cur, struct vmbus_channel, sc_list);
 510                 n_written = scnprintf(buf + tot_written,
 511                                      buf_size - tot_written,
 512                                      "%u:%u\n",
 513                                      cur_sc->offermsg.child_relid,
 514                                      cur_sc->target_cpu);
 515                 tot_written += n_written;
 516         }
 517
 518         spin_unlock_irqrestore(&channel->lock, flags);
 519
 520         return tot_written;
 521 }
 522 static DEVICE_ATTR_RO(channel_vp_mapping);
 523
 524 static ssize_t vendor_show(struct device *dev,
 525                            struct device_attribute *dev_attr,
 526                            char *buf)
 527 {
 528         struct hv_device *hv_dev = device_to_hv_device(dev);
 529         return sprintf(buf, "0x%x\n", hv_dev->vendor_id);
 530 }
 531 static DEVICE_ATTR_RO(vendor);
 532
 533 static ssize_t device_show(struct device *dev,
 534                            struct device_attribute *dev_attr,
 535                            char *buf)
 536 {
 537         struct hv_device *hv_dev = device_to_hv_device(dev);
 538         return sprintf(buf, "0x%x\n", hv_dev->device_id);
 539 }
 540 static DEVICE_ATTR_RO(device);
 541
 542 static ssize_t driver_override_store(struct device *dev,
 543                                      struct device_attribute *attr,
 544                                      const char *buf, size_t count)
 545 {
 546         struct hv_device *hv_dev = device_to_hv_device(dev);
 547         char *driver_override, *old, *cp;
 548
 549         /* We need to keep extra room for a newline */
 550         if (count >= (PAGE_SIZE - 1))
 551                 return -EINVAL;
 552
 553         driver_override = kstrndup(buf, count, GFP_KERNEL);
 554         if (!driver_override)
 555                 return -ENOMEM;
 556
 557         cp = strchr(driver_override, '\n');
 558         if (cp)
 559                 *cp = '\0';
 560
 561         device_lock(dev);
 562         old = hv_dev->driver_override;
 563         if (strlen(driver_override)) {
 564                 hv_dev->driver_override = driver_override;
 565         } else {
 566                 kfree(driver_override);
 567                 hv_dev->driver_override = NULL;
 568         }
 569         device_unlock(dev);
 570
 571         kfree(old);
 572
 573         return count;
 574 }
 575
 576 static ssize_t driver_override_show(struct device *dev,
 577                                     struct device_attribute *attr, char *buf)
 578 {
 579         struct hv_device *hv_dev = device_to_hv_device(dev);
 580         ssize_t len;
 581
 582         device_lock(dev);
 583         len = snprintf(buf, PAGE_SIZE, "%s\n", hv_dev->driver_override);
 584         device_unlock(dev);
 585
 586         return len;
 587 }
 588 static DEVICE_ATTR_RW(driver_override);
 589
 590 /* Set up per device attributes in /sys/bus/vmbus/devices/<bus device> */
 591 static struct attribute *vmbus_dev_attrs[] = {
 592         &dev_attr_id.attr,
 593         &dev_attr_state.attr,
 594         &dev_attr_monitor_id.attr,
 595         &dev_attr_class_id.attr,
 596         &dev_attr_device_id.attr,
 597         &dev_attr_modalias.attr,
 598 #ifdef CONFIG_NUMA
 599         &dev_attr_numa_node.attr,
 600 #endif
 601         &dev_attr_server_monitor_pending.attr,
 602         &dev_attr_client_monitor_pending.attr,
 603         &dev_attr_server_monitor_latency.attr,
 604         &dev_attr_client_monitor_latency.attr,
 605         &dev_attr_server_monitor_conn_id.attr,
 606         &dev_attr_client_monitor_conn_id.attr,
 607         &dev_attr_out_intr_mask.attr,
 608         &dev_attr_out_read_index.attr,
 609         &dev_attr_out_write_index.attr,
 610         &dev_attr_out_read_bytes_avail.attr,
 611         &dev_attr_out_write_bytes_avail.attr,
 612         &dev_attr_in_intr_mask.attr,
 613         &dev_attr_in_read_index.attr,
 614         &dev_attr_in_write_index.attr,
 615         &dev_attr_in_read_bytes_avail.attr,
 616         &dev_attr_in_write_bytes_avail.attr,
 617         &dev_attr_channel_vp_mapping.attr,
 618         &dev_attr_vendor.attr,
 619         &dev_attr_device.attr,
 620         &dev_attr_driver_override.attr,
 621         NULL,
 622 };
 623
 624 /*
 625  * Device-level attribute_group callback function. Returns the permission for
 626  * each attribute, and returns 0 if an attribute is not visible.
 627  */
 628 static umode_t vmbus_dev_attr_is_visible(struct kobject *kobj,
 629                                          struct attribute *attr, int idx)
 630 {
 631         struct device *dev = kobj_to_dev(kobj);
 632         const struct hv_device *hv_dev = device_to_hv_device(dev);
 633
 634         /* Hide the monitor attributes if the monitor mechanism is not used. */
 635         if (!hv_dev->channel->offermsg.monitor_allocated &&
 636             (attr == &dev_attr_monitor_id.attr ||
 637              attr == &dev_attr_server_monitor_pending.attr ||
 638              attr == &dev_attr_client_monitor_pending.attr ||
 639              attr == &dev_attr_server_monitor_latency.attr ||
 640              attr == &dev_attr_client_monitor_latency.attr ||
 641              attr == &dev_attr_server_monitor_conn_id.attr ||
 642              attr == &dev_attr_client_monitor_conn_id.attr))
 643                 return 0;
 644
 645         return attr->mode;
 646 }
 647
 648 static const struct attribute_group vmbus_dev_group = {
 649         .attrs = vmbus_dev_attrs,
 650         .is_visible = vmbus_dev_attr_is_visible
 651 };
 652 __ATTRIBUTE_GROUPS(vmbus_dev);
 653
 654 /*
 655  * vmbus_uevent - add uevent for our device
 656  *
 657  * This routine is invoked when a device is added or removed on the vmbus to
 658  * generate a uevent to udev in the userspace. The udev will then look at its
 659  * rule and the uevent generated here to load the appropriate driver
 660  *
 661  * The alias string will be of the form vmbus:guid where guid is the string
 662  * representation of the device guid (each byte of the guid will be
 663  * represented with two hex characters.
 664  */
 665 static int vmbus_uevent(struct device *device, struct kobj_uevent_env *env)
 666 {
 667         struct hv_device *dev = device_to_hv_device(device);
 668         int ret;
 669         char alias_name[VMBUS_ALIAS_LEN + 1];
 670
 671         print_alias_name(dev, alias_name);
 672         ret = add_uevent_var(env, "MODALIAS=vmbus:%s", alias_name);
 673         return ret;
 674 }
 675
 676 static const struct hv_vmbus_device_id *
 677 hv_vmbus_dev_match(const struct hv_vmbus_device_id *id, const guid_t *guid)
 678 {
 679         if (id == NULL)
 680                 return NULL; /* empty device table */
 681
 682         for (; !guid_is_null(&id->guid); id++)
 683                 if (guid_equal(&id->guid, guid))
 684                         return id;
 685
 686         return NULL;
 687 }
 688
 689 static const struct hv_vmbus_device_id *
 690 hv_vmbus_dynid_match(struct hv_driver *drv, const guid_t *guid)
 691 {
 692         const struct hv_vmbus_device_id *id = NULL;
 693         struct vmbus_dynid *dynid;
 694
 695         spin_lock(&drv->dynids.lock);
 696         list_for_each_entry(dynid, &drv->dynids.list, node) {
 697                 if (guid_equal(&dynid->id.guid, guid)) {
 698                         id = &dynid->id;
 699                         break;
 700                 }
 701         }
 702         spin_unlock(&drv->dynids.lock);
 703
 704         return id;
 705 }
 706
 707 static const struct hv_vmbus_device_id vmbus_device_null;
 708
 709 /*
 710  * Return a matching hv_vmbus_device_id pointer.
 711  * If there is no match, return NULL.
 712  */
 713 static const struct hv_vmbus_device_id *hv_vmbus_get_id(struct hv_driver *drv,
 714                                                         struct hv_device *dev)
 715 {
 716         const guid_t *guid = &dev->dev_type;
 717         const struct hv_vmbus_device_id *id;
 718
 719         /* When driver_override is set, only bind to the matching driver */
 720         if (dev->driver_override && strcmp(dev->driver_override, drv->name))
 721                 return NULL;
 722
 723         /* Look at the dynamic ids first, before the static ones */
 724         id = hv_vmbus_dynid_match(drv, guid);
 725         if (!id)
 726                 id = hv_vmbus_dev_match(drv->id_table, guid);
 727
 728         /* driver_override will always match, send a dummy id */
 729         if (!id && dev->driver_override)
 730                 id = &vmbus_device_null;
 731
 732         return id;
 733 }
 734
 735 /* vmbus_add_dynid - add a new device ID to this driver and re-probe devices */
 736 static int vmbus_add_dynid(struct hv_driver *drv, guid_t *guid)
 737 {
 738         struct vmbus_dynid *dynid;
 739
 740         dynid = kzalloc(sizeof(*dynid), GFP_KERNEL);
 741         if (!dynid)
 742                 return -ENOMEM;
 743
 744         dynid->id.guid = *guid;
 745
 746         spin_lock(&drv->dynids.lock);
 747         list_add_tail(&dynid->node, &drv->dynids.list);
 748         spin_unlock(&drv->dynids.lock);
 749
 750         return driver_attach(&drv->driver);
 751 }
 752
 753 static void vmbus_free_dynids(struct hv_driver *drv)
 754 {
 755         struct vmbus_dynid *dynid, *n;
 756
 757         spin_lock(&drv->dynids.lock);
 758         list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
 759                 list_del(&dynid->node);
 760                 kfree(dynid);
 761         }
 762         spin_unlock(&drv->dynids.lock);
 763 }
 764
 765 /*
 766  * store_new_id - sysfs frontend to vmbus_add_dynid()
 767  *
 768  * Allow GUIDs to be added to an existing driver via sysfs.
 769  */
 770 static ssize_t new_id_store(struct device_driver *driver, const char *buf,
 771                             size_t count)
 772 {
 773         struct hv_driver *drv = drv_to_hv_drv(driver);
 774         guid_t guid;
 775         ssize_t retval;
 776
 777         retval = guid_parse(buf, &guid);
 778         if (retval)
 779                 return retval;
 780
 781         if (hv_vmbus_dynid_match(drv, &guid))
 782                 return -EEXIST;
 783
 784         retval = vmbus_add_dynid(drv, &guid);
 785         if (retval)
 786                 return retval;
 787         return count;
 788 }
 789 static DRIVER_ATTR_WO(new_id);
 790
 791 /*
 792  * store_remove_id - remove a PCI device ID from this driver
 793  *
 794  * Removes a dynamic pci device ID to this driver.
 795  */
 796 static ssize_t remove_id_store(struct device_driver *driver, const char *buf,
 797                                size_t count)
 798 {
 799         struct hv_driver *drv = drv_to_hv_drv(driver);
 800         struct vmbus_dynid *dynid, *n;
 801         guid_t guid;
 802         ssize_t retval;
 803
 804         retval = guid_parse(buf, &guid);
 805         if (retval)
 806                 return retval;
 807
 808         retval = -ENODEV;
 809         spin_lock(&drv->dynids.lock);
 810         list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
 811                 struct hv_vmbus_device_id *id = &dynid->id;
 812
 813                 if (guid_equal(&id->guid, &guid)) {
 814                         list_del(&dynid->node);
 815                         kfree(dynid);
 816                         retval = count;
 817                         break;
 818                 }
 819         }
 820         spin_unlock(&drv->dynids.lock);
 821
 822         return retval;
 823 }
 824 static DRIVER_ATTR_WO(remove_id);
 825
 826 static struct attribute *vmbus_drv_attrs[] = {
 827         &driver_attr_new_id.attr,
 828         &driver_attr_remove_id.attr,
 829         NULL,
 830 };
 831 ATTRIBUTE_GROUPS(vmbus_drv);
 832
 833
 834 /*
 835  * vmbus_match - Attempt to match the specified device to the specified driver
 836  */
 837 static int vmbus_match(struct device *device, struct device_driver *driver)
 838 {
 839         struct hv_driver *drv = drv_to_hv_drv(driver);
 840         struct hv_device *hv_dev = device_to_hv_device(device);
 841
 842         /* The hv_sock driver handles all hv_sock offers. */
 843         if (is_hvsock_channel(hv_dev->channel))
 844                 return drv->hvsock;
 845
 846         if (hv_vmbus_get_id(drv, hv_dev))
 847                 return 1;
 848
 849         return 0;
 850 }
 851
 852 /*
 853  * vmbus_probe - Add the new vmbus's child device
 854  */
 855 static int vmbus_probe(struct device *child_device)
 856 {
 857         int ret = 0;
 858         struct hv_driver *drv =
 859                         drv_to_hv_drv(child_device->driver);
 860         struct hv_device *dev = device_to_hv_device(child_device);
 861         const struct hv_vmbus_device_id *dev_id;
 862
 863         dev_id = hv_vmbus_get_id(drv, dev);
 864         if (drv->probe) {
 865                 ret = drv->probe(dev, dev_id);
 866                 if (ret != 0)
 867                         pr_err("probe failed for device %s (%d)\n",
 868                                dev_name(child_device), ret);
 869
 870         } else {
 871                 pr_err("probe not set for driver %s\n",
 872                        dev_name(child_device));
 873                 ret = -ENODEV;
 874         }
 875         return ret;
 876 }
 877
 878 /*
 879  * vmbus_remove - Remove a vmbus device
 880  */
 881 static int vmbus_remove(struct device *child_device)
 882 {
 883         struct hv_driver *drv;
 884         struct hv_device *dev = device_to_hv_device(child_device);
 885
 886         if (child_device->driver) {
 887                 drv = drv_to_hv_drv(child_device->driver);
 888                 if (drv->remove)
 889                         drv->remove(dev);
 890         }
 891
 892         return 0;
 893 }
 894
 895
 896 /*
 897  * vmbus_shutdown - Shutdown a vmbus device
 898  */
 899 static void vmbus_shutdown(struct device *child_device)
 900 {
 901         struct hv_driver *drv;
 902         struct hv_device *dev = device_to_hv_device(child_device);
 903
 904
 905         /* The device may not be attached yet */
 906         if (!child_device->driver)
 907                 return;
 908
 909         drv = drv_to_hv_drv(child_device->driver);
 910
 911         if (drv->shutdown)
 912                 drv->shutdown(dev);
 913 }
 914
 915 #ifdef CONFIG_PM_SLEEP
 916 /*
 917  * vmbus_suspend - Suspend a vmbus device
 918  */
 919 static int vmbus_suspend(struct device *child_device)
 920 {
 921         struct hv_driver *drv;
 922         struct hv_device *dev = device_to_hv_device(child_device);
 923
 924         /* The device may not be attached yet */
 925         if (!child_device->driver)
 926                 return 0;
 927
 928         drv = drv_to_hv_drv(child_device->driver);
 929         if (!drv->suspend)
 930                 return -EOPNOTSUPP;
 931
 932         return drv->suspend(dev);
 933 }
 934
 935 /*
 936  * vmbus_resume - Resume a vmbus device
 937  */
 938 static int vmbus_resume(struct device *child_device)
 939 {
 940         struct hv_driver *drv;
 941         struct hv_device *dev = device_to_hv_device(child_device);
 942
 943         /* The device may not be attached yet */
 944         if (!child_device->driver)
 945                 return 0;
 946
 947         drv = drv_to_hv_drv(child_device->driver);
 948         if (!drv->resume)
 949                 return -EOPNOTSUPP;
 950
 951         return drv->resume(dev);
 952 }
 953 #endif /* CONFIG_PM_SLEEP */
 954
 955 /*
 956  * vmbus_device_release - Final callback release of the vmbus child device
 957  */
 958 static void vmbus_device_release(struct device *device)
 959 {
 960         struct hv_device *hv_dev = device_to_hv_device(device);
 961         struct vmbus_channel *channel = hv_dev->channel;
 962
 963         mutex_lock(&vmbus_connection.channel_mutex);
 964         hv_process_channel_removal(channel);
 965         mutex_unlock(&vmbus_connection.channel_mutex);
 966         kfree(hv_dev);
 967 }
 968
 969 /*
 970  * Note: we must use SET_NOIRQ_SYSTEM_SLEEP_PM_OPS rather than
 971  * SET_SYSTEM_SLEEP_PM_OPS: see the comment before vmbus_bus_pm.
 972  */
 973 static const struct dev_pm_ops vmbus_pm = {
 974         SET_NOIRQ_SYSTEM_SLEEP_PM_OPS(vmbus_suspend, vmbus_resume)
 975 };
 976
 977 /* The one and only one */
 978 static struct bus_type  hv_bus = {
 979         .name =         "vmbus",
 980         .match =                vmbus_match,
 981         .shutdown =             vmbus_shutdown,
 982         .remove =               vmbus_remove,
 983         .probe =                vmbus_probe,
 984         .uevent =               vmbus_uevent,
 985         .dev_groups =           vmbus_dev_groups,
 986         .drv_groups =           vmbus_drv_groups,
 987         .pm =                   &vmbus_pm,
 988 };
 989
 990 struct onmessage_work_context {
 991         struct work_struct work;
 992         struct hv_message msg;
 993 };
 994
 995 static void vmbus_onmessage_work(struct work_struct *work)
 996 {
 997         struct onmessage_work_context *ctx;
 998
 999         /* Do not process messages if we're in DISCONNECTED state */
1000         if (vmbus_connection.conn_state == DISCONNECTED)
1001                 return;
1002
1003         ctx = container_of(work, struct onmessage_work_context,
1004                            work);
1005         vmbus_onmessage(&ctx->msg);
1006         kfree(ctx);
1007 }
1008
1009 void vmbus_on_msg_dpc(unsigned long data)
1010 {
1011         struct hv_per_cpu_context *hv_cpu = (void *)data;
1012         void *page_addr = hv_cpu->synic_message_page;
1013         struct hv_message *msg = (struct hv_message *)page_addr +
1014                                   VMBUS_MESSAGE_SINT;
1015         struct vmbus_channel_message_header *hdr;
1016         const struct vmbus_channel_message_table_entry *entry;
1017         struct onmessage_work_context *ctx;
1018         u32 message_type = msg->header.message_type;
1019
1020         if (message_type == HVMSG_NONE)
1021                 /* no msg */
1022                 return;
1023
1024         hdr = (struct vmbus_channel_message_header *)msg->u.payload;
1025
1026         trace_vmbus_on_msg_dpc(hdr);
1027
1028         if (hdr->msgtype >= CHANNELMSG_COUNT) {
1029                 WARN_ONCE(1, "unknown msgtype=%d\n", hdr->msgtype);
1030                 goto msg_handled;
1031         }
1032
1033         entry = &channel_message_table[hdr->msgtype];
1034         if (entry->handler_type == VMHT_BLOCKING) {
1035                 ctx = kmalloc(sizeof(*ctx), GFP_ATOMIC);
1036                 if (ctx == NULL)
1037                         return;
1038
1039                 INIT_WORK(&ctx->work, vmbus_onmessage_work);
1040                 memcpy(&ctx->msg, msg, sizeof(*msg));
1041
1042                 /*
1043                  * The host can generate a rescind message while we
1044                  * may still be handling the original offer. We deal with
1045                  * this condition by ensuring the processing is done on the
1046                  * same CPU.
1047                  */
1048                 switch (hdr->msgtype) {
1049                 case CHANNELMSG_RESCIND_CHANNELOFFER:
1050                         /*
1051                          * If we are handling the rescind message;
1052                          * schedule the work on the global work queue.
1053                          */
1054                         schedule_work_on(vmbus_connection.connect_cpu,
1055                                          &ctx->work);
1056                         break;
1057
1058                 case CHANNELMSG_OFFERCHANNEL:
1059                         atomic_inc(&vmbus_connection.offer_in_progress);
1060                         queue_work_on(vmbus_connection.connect_cpu,
1061                                       vmbus_connection.work_queue,
1062                                       &ctx->work);
1063                         break;
1064
1065                 default:
1066                         queue_work(vmbus_connection.work_queue, &ctx->work);
1067                 }
1068         } else
1069                 entry->message_handler(hdr);
1070
1071 msg_handled:
1072         vmbus_signal_eom(msg, message_type);
1073 }
1074
1075 #ifdef CONFIG_PM_SLEEP
1076 /*
1077  * Fake RESCIND_CHANNEL messages to clean up hv_sock channels by force for
1078  * hibernation, because hv_sock connections can not persist across hibernation.
1079  */
1080 static void vmbus_force_channel_rescinded(struct vmbus_channel *channel)
1081 {
1082         struct onmessage_work_context *ctx;
1083         struct vmbus_channel_rescind_offer *rescind;
1084
1085         WARN_ON(!is_hvsock_channel(channel));
1086
1087         /*
1088          * sizeof(*ctx) is small and the allocation should really not fail,
1089          * otherwise the state of the hv_sock connections ends up in limbo.
1090          */
1091         ctx = kzalloc(sizeof(*ctx), GFP_KERNEL | __GFP_NOFAIL);
1092
1093         /*
1094          * So far, these are not really used by Linux. Just set them to the
1095          * reasonable values conforming to the definitions of the fields.
1096          */
1097         ctx->msg.header.message_type = 1;
1098         ctx->msg.header.payload_size = sizeof(*rescind);
1099
1100         /* These values are actually used by Linux. */
1101         rescind = (struct vmbus_channel_rescind_offer *)ctx->msg.u.payload;
1102         rescind->header.msgtype = CHANNELMSG_RESCIND_CHANNELOFFER;
1103         rescind->child_relid = channel->offermsg.child_relid;
1104
1105         INIT_WORK(&ctx->work, vmbus_onmessage_work);
1106
1107         queue_work_on(vmbus_connection.connect_cpu,
1108                       vmbus_connection.work_queue,
1109                       &ctx->work);
1110 }
1111 #endif /* CONFIG_PM_SLEEP */
1112
1113 /*
1114  * Direct callback for channels using other deferred processing
1115  */
1116 static void vmbus_channel_isr(struct vmbus_channel *channel)
1117 {
1118         void (*callback_fn)(void *);
1119
1120         callback_fn = READ_ONCE(channel->onchannel_callback);
1121         if (likely(callback_fn != NULL))
1122                 (*callback_fn)(channel->channel_callback_context);
1123 }
1124
1125 /*
1126  * Schedule all channels with events pending
1127  */
1128 static void vmbus_chan_sched(struct hv_per_cpu_context *hv_cpu)
1129 {
1130         unsigned long *recv_int_page;
1131         u32 maxbits, relid;
1132
1133         if (vmbus_proto_version < VERSION_WIN8) {
1134                 maxbits = MAX_NUM_CHANNELS_SUPPORTED;
1135                 recv_int_page = vmbus_connection.recv_int_page;
1136         } else {
1137                 /*
1138                  * When the host is win8 and beyond, the event page
1139                  * can be directly checked to get the id of the channel
1140                  * that has the interrupt pending.
1141                  */
1142                 void *page_addr = hv_cpu->synic_event_page;
1143                 union hv_synic_event_flags *event
1144                         = (union hv_synic_event_flags *)page_addr +
1145                                                  VMBUS_MESSAGE_SINT;
1146
1147                 maxbits = HV_EVENT_FLAGS_COUNT;
1148                 recv_int_page = event->flags;
1149         }
1150
1151         if (unlikely(!recv_int_page))
1152                 return;
1153
1154         for_each_set_bit(relid, recv_int_page, maxbits) {
1155                 struct vmbus_channel *channel;
1156
1157                 if (!sync_test_and_clear_bit(relid, recv_int_page))
1158                         continue;
1159
1160                 /* Special case - vmbus channel protocol msg */
1161                 if (relid == 0)
1162                         continue;
1163
1164                 rcu_read_lock();
1165
1166                 /* Find channel based on relid */
1167                 list_for_each_entry_rcu(channel, &hv_cpu->chan_list, percpu_list) {
1168                         if (channel->offermsg.child_relid != relid)
1169                                 continue;
1170
1171                         if (channel->rescind)
1172                                 continue;
1173
1174                         trace_vmbus_chan_sched(channel);
1175
1176                         ++channel->interrupts;
1177
1178                         switch (channel->callback_mode) {
1179                         case HV_CALL_ISR:
1180                                 vmbus_channel_isr(channel);
1181                                 break;
1182
1183                         case HV_CALL_BATCHED:
1184                                 hv_begin_read(&channel->inbound);
1185                                 /* fallthrough */
1186                         case HV_CALL_DIRECT:
1187                                 tasklet_schedule(&channel->callback_event);
1188                         }
1189                 }
1190
1191                 rcu_read_unlock();
1192         }
1193 }
1194
1195 static void vmbus_isr(void)
1196 {
1197         struct hv_per_cpu_context *hv_cpu
1198                 = this_cpu_ptr(hv_context.cpu_context);
1199         void *page_addr = hv_cpu->synic_event_page;
1200         struct hv_message *msg;
1201         union hv_synic_event_flags *event;
1202         bool handled = false;
1203
1204         if (unlikely(page_addr == NULL))
1205                 return;
1206
1207         event = (union hv_synic_event_flags *)page_addr +
1208                                          VMBUS_MESSAGE_SINT;
1209         /*
1210          * Check for events before checking for messages. This is the order
1211          * in which events and messages are checked in Windows guests on
1212          * Hyper-V, and the Windows team suggested we do the same.
1213          */
1214
1215         if ((vmbus_proto_version == VERSION_WS2008) ||
1216                 (vmbus_proto_version == VERSION_WIN7)) {
1217
1218                 /* Since we are a child, we only need to check bit 0 */
1219                 if (sync_test_and_clear_bit(0, event->flags))
1220                         handled = true;
1221         } else {
1222                 /*
1223                  * Our host is win8 or above. The signaling mechanism
1224                  * has changed and we can directly look at the event page.
1225                  * If bit n is set then we have an interrup on the channel
1226                  * whose id is n.
1227                  */
1228                 handled = true;
1229         }
1230
1231         if (handled)
1232                 vmbus_chan_sched(hv_cpu);
1233
1234         page_addr = hv_cpu->synic_message_page;
1235         msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
1236
1237         /* Check if there are actual msgs to be processed */
1238         if (msg->header.message_type != HVMSG_NONE) {
1239                 if (msg->header.message_type == HVMSG_TIMER_EXPIRED) {
1240                         hv_stimer0_isr();
1241                         vmbus_signal_eom(msg, HVMSG_TIMER_EXPIRED);
1242                 } else
1243                         tasklet_schedule(&hv_cpu->msg_dpc);
1244         }
1245
1246         add_interrupt_randomness(HYPERVISOR_CALLBACK_VECTOR, 0);
1247 }
1248
1249 /*
1250  * Boolean to control whether to report panic messages over Hyper-V.
1251  *
1252  * It can be set via /proc/sys/kernel/hyperv/record_panic_msg
1253  */
1254 static int sysctl_record_panic_msg = 1;
1255
1256 /*
1257  * Callback from kmsg_dump. Grab as much as possible from the end of the kmsg
1258  * buffer and call into Hyper-V to transfer the data.
1259  */
1260 static void hv_kmsg_dump(struct kmsg_dumper *dumper,
1261                          enum kmsg_dump_reason reason)
1262 {
1263         size_t bytes_written;
1264         phys_addr_t panic_pa;
1265
1266         /* We are only interested in panics. */
1267         if ((reason != KMSG_DUMP_PANIC) || (!sysctl_record_panic_msg))
1268                 return;
1269
1270         panic_pa = virt_to_phys(hv_panic_page);
1271
1272         /*
1273          * Write dump contents to the page. No need to synchronize; panic should
1274          * be single-threaded.
1275          */
1276         kmsg_dump_get_buffer(dumper, true, hv_panic_page, PAGE_SIZE,
1277                              &bytes_written);
1278         if (bytes_written)
1279                 hyperv_report_panic_msg(panic_pa, bytes_written);
1280 }
1281
1282 static struct kmsg_dumper hv_kmsg_dumper = {
1283         .dump = hv_kmsg_dump,
1284 };
1285
1286 static struct ctl_table_header *hv_ctl_table_hdr;
1287
1288 /*
1289  * sysctl option to allow the user to control whether kmsg data should be
1290  * reported to Hyper-V on panic.
1291  */
1292 static struct ctl_table hv_ctl_table[] = {
1293         {
1294                 .procname       = "hyperv_record_panic_msg",
1295                 .data           = &sysctl_record_panic_msg,
1296                 .maxlen         = sizeof(int),
1297                 .mode           = 0644,
1298                 .proc_handler   = proc_dointvec_minmax,
1299                 .extra1         = SYSCTL_ZERO,
1300                 .extra2         = SYSCTL_ONE
1301         },
1302         {}
1303 };
1304
1305 static struct ctl_table hv_root_table[] = {
1306         {
1307                 .procname       = "kernel",
1308                 .mode           = 0555,
1309                 .child          = hv_ctl_table
1310         },
1311         {}
1312 };
1313
1314 /*
1315  * vmbus_bus_init -Main vmbus driver initialization routine.
1316  *
1317  * Here, we
1318  *      - initialize the vmbus driver context
1319  *      - invoke the vmbus hv main init routine
1320  *      - retrieve the channel offers
1321  */
1322 static int vmbus_bus_init(void)
1323 {
1324         int ret;
1325
1326         /* Hypervisor initialization...setup hypercall page..etc */
1327         ret = hv_init();
1328         if (ret != 0) {
1329                 pr_err("Unable to initialize the hypervisor - 0x%x\n", ret);
1330                 return ret;
1331         }
1332
1333         ret = bus_register(&hv_bus);
1334         if (ret)
1335                 return ret;
1336
1337         hv_setup_vmbus_irq(vmbus_isr);
1338
1339         ret = hv_synic_alloc();
1340         if (ret)
1341                 goto err_alloc;
1342
1343         ret = hv_stimer_alloc(VMBUS_MESSAGE_SINT);
1344         if (ret < 0)
1345                 goto err_alloc;
1346
1347         /*
1348          * Initialize the per-cpu interrupt state and stimer state.
1349          * Then connect to the host.
1350          */
1351         ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "hyperv/vmbus:online",
1352                                 hv_synic_init, hv_synic_cleanup);
1353         if (ret < 0)
1354                 goto err_cpuhp;
1355         hyperv_cpuhp_online = ret;
1356
1357         ret = vmbus_connect();
1358         if (ret)
1359                 goto err_connect;
1360
1361         /*
1362          * Only register if the crash MSRs are available
1363          */
1364         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1365                 u64 hyperv_crash_ctl;
1366                 /*
1367                  * Sysctl registration is not fatal, since by default
1368                  * reporting is enabled.
1369                  */
1370                 hv_ctl_table_hdr = register_sysctl_table(hv_root_table);
1371                 if (!hv_ctl_table_hdr)
1372                         pr_err("Hyper-V: sysctl table register error");
1373
1374                 /*
1375                  * Register for panic kmsg callback only if the right
1376                  * capability is supported by the hypervisor.
1377                  */
1378                 hv_get_crash_ctl(hyperv_crash_ctl);
1379                 if (hyperv_crash_ctl & HV_CRASH_CTL_CRASH_NOTIFY_MSG) {
1380                         hv_panic_page = (void *)get_zeroed_page(GFP_KERNEL);
1381                         if (hv_panic_page) {
1382                                 ret = kmsg_dump_register(&hv_kmsg_dumper);
1383                                 if (ret)
1384                                         pr_err("Hyper-V: kmsg dump register "
1385                                                 "error 0x%x\n", ret);
1386                         } else
1387                                 pr_err("Hyper-V: panic message page memory "
1388                                         "allocation failed");
1389                 }
1390
1391                 register_die_notifier(&hyperv_die_block);
1392                 atomic_notifier_chain_register(&panic_notifier_list,
1393                                                &hyperv_panic_block);
1394         }
1395
1396         vmbus_request_offers();
1397
1398         return 0;
1399
1400 err_connect:
1401         cpuhp_remove_state(hyperv_cpuhp_online);
1402 err_cpuhp:
1403         hv_stimer_free();
1404 err_alloc:
1405         hv_synic_free();
1406         hv_remove_vmbus_irq();
1407
1408         bus_unregister(&hv_bus);
1409         free_page((unsigned long)hv_panic_page);
1410         unregister_sysctl_table(hv_ctl_table_hdr);
1411         hv_ctl_table_hdr = NULL;
1412         return ret;
1413 }
1414
1415 /**
1416  * __vmbus_child_driver_register() - Register a vmbus's driver
1417  * @hv_driver: Pointer to driver structure you want to register
1418  * @owner: owner module of the drv
1419  * @mod_name: module name string
1420  *
1421  * Registers the given driver with Linux through the 'driver_register()' call
1422  * and sets up the hyper-v vmbus handling for this driver.
1423  * It will return the state of the 'driver_register()' call.
1424  *
1425  */
1426 int __vmbus_driver_register(struct hv_driver *hv_driver, struct module *owner, const char *mod_name)
1427 {
1428         int ret;
1429
1430         pr_info("registering driver %s\n", hv_driver->name);
1431
1432         ret = vmbus_exists();
1433         if (ret < 0)
1434                 return ret;
1435
1436         hv_driver->driver.name = hv_driver->name;
1437         hv_driver->driver.owner = owner;
1438         hv_driver->driver.mod_name = mod_name;
1439         hv_driver->driver.bus = &hv_bus;
1440
1441         spin_lock_init(&hv_driver->dynids.lock);
1442         INIT_LIST_HEAD(&hv_driver->dynids.list);
1443
1444         ret = driver_register(&hv_driver->driver);
1445
1446         return ret;
1447 }
1448 EXPORT_SYMBOL_GPL(__vmbus_driver_register);
1449
1450 /**
1451  * vmbus_driver_unregister() - Unregister a vmbus's driver
1452  * @hv_driver: Pointer to driver structure you want to
1453  *             un-register
1454  *
1455  * Un-register the given driver that was previous registered with a call to
1456  * vmbus_driver_register()
1457  */
1458 void vmbus_driver_unregister(struct hv_driver *hv_driver)
1459 {
1460         pr_info("unregistering driver %s\n", hv_driver->name);
1461
1462         if (!vmbus_exists()) {
1463                 driver_unregister(&hv_driver->driver);
1464                 vmbus_free_dynids(hv_driver);
1465         }
1466 }
1467 EXPORT_SYMBOL_GPL(vmbus_driver_unregister);
1468
1469
1470 /*
1471  * Called when last reference to channel is gone.
1472  */
1473 static void vmbus_chan_release(struct kobject *kobj)
1474 {
1475         struct vmbus_channel *channel
1476                 = container_of(kobj, struct vmbus_channel, kobj);
1477
1478         kfree_rcu(channel, rcu);
1479 }
1480
1481 struct vmbus_chan_attribute {
1482         struct attribute attr;
1483         ssize_t (*show)(struct vmbus_channel *chan, char *buf);
1484         ssize_t (*store)(struct vmbus_channel *chan,
1485                          const char *buf, size_t count);
1486 };
1487 #define VMBUS_CHAN_ATTR(_name, _mode, _show, _store) \
1488         struct vmbus_chan_attribute chan_attr_##_name \
1489                 = __ATTR(_name, _mode, _show, _store)
1490 #define VMBUS_CHAN_ATTR_RW(_name) \
1491         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RW(_name)
1492 #define VMBUS_CHAN_ATTR_RO(_name) \
1493         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RO(_name)
1494 #define VMBUS_CHAN_ATTR_WO(_name) \
1495         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_WO(_name)
1496
1497 static ssize_t vmbus_chan_attr_show(struct kobject *kobj,
1498                                     struct attribute *attr, char *buf)
1499 {
1500         const struct vmbus_chan_attribute *attribute
1501                 = container_of(attr, struct vmbus_chan_attribute, attr);
1502         struct vmbus_channel *chan
1503                 = container_of(kobj, struct vmbus_channel, kobj);
1504
1505         if (!attribute->show)
1506                 return -EIO;
1507
1508         return attribute->show(chan, buf);
1509 }
1510
1511 static const struct sysfs_ops vmbus_chan_sysfs_ops = {
1512         .show = vmbus_chan_attr_show,
1513 };
1514
1515 static ssize_t out_mask_show(struct vmbus_channel *channel, char *buf)
1516 {
1517         struct hv_ring_buffer_info *rbi = &channel->outbound;
1518         ssize_t ret;
1519
1520         mutex_lock(&rbi->ring_buffer_mutex);
1521         if (!rbi->ring_buffer) {
1522                 mutex_unlock(&rbi->ring_buffer_mutex);
1523                 return -EINVAL;
1524         }
1525
1526         ret = sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1527         mutex_unlock(&rbi->ring_buffer_mutex);
1528         return ret;
1529 }
1530 static VMBUS_CHAN_ATTR_RO(out_mask);
1531
1532 static ssize_t in_mask_show(struct vmbus_channel *channel, char *buf)
1533 {
1534         struct hv_ring_buffer_info *rbi = &channel->inbound;
1535         ssize_t ret;
1536
1537         mutex_lock(&rbi->ring_buffer_mutex);
1538         if (!rbi->ring_buffer) {
1539                 mutex_unlock(&rbi->ring_buffer_mutex);
1540                 return -EINVAL;
1541         }
1542
1543         ret = sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1544         mutex_unlock(&rbi->ring_buffer_mutex);
1545         return ret;
1546 }
1547 static VMBUS_CHAN_ATTR_RO(in_mask);
1548
1549 static ssize_t read_avail_show(struct vmbus_channel *channel, char *buf)
1550 {
1551         struct hv_ring_buffer_info *rbi = &channel->inbound;
1552         ssize_t ret;
1553
1554         mutex_lock(&rbi->ring_buffer_mutex);
1555         if (!rbi->ring_buffer) {
1556                 mutex_unlock(&rbi->ring_buffer_mutex);
1557                 return -EINVAL;
1558         }
1559
1560         ret = sprintf(buf, "%u\n", hv_get_bytes_to_read(rbi));
1561         mutex_unlock(&rbi->ring_buffer_mutex);
1562         return ret;
1563 }
1564 static VMBUS_CHAN_ATTR_RO(read_avail);
1565
1566 static ssize_t write_avail_show(struct vmbus_channel *channel, char *buf)
1567 {
1568         struct hv_ring_buffer_info *rbi = &channel->outbound;
1569         ssize_t ret;
1570
1571         mutex_lock(&rbi->ring_buffer_mutex);
1572         if (!rbi->ring_buffer) {
1573                 mutex_unlock(&rbi->ring_buffer_mutex);
1574                 return -EINVAL;
1575         }
1576
1577         ret = sprintf(buf, "%u\n", hv_get_bytes_to_write(rbi));
1578         mutex_unlock(&rbi->ring_buffer_mutex);
1579         return ret;
1580 }
1581 static VMBUS_CHAN_ATTR_RO(write_avail);
1582
1583 static ssize_t show_target_cpu(struct vmbus_channel *channel, char *buf)
1584 {
1585         return sprintf(buf, "%u\n", channel->target_cpu);
1586 }
1587 static VMBUS_CHAN_ATTR(cpu, S_IRUGO, show_target_cpu, NULL);
1588
1589 static ssize_t channel_pending_show(struct vmbus_channel *channel,
1590                                     char *buf)
1591 {
1592         return sprintf(buf, "%d\n",
1593                        channel_pending(channel,
1594                                        vmbus_connection.monitor_pages[1]));
1595 }
1596 static VMBUS_CHAN_ATTR(pending, S_IRUGO, channel_pending_show, NULL);
1597
1598 static ssize_t channel_latency_show(struct vmbus_channel *channel,
1599                                     char *buf)
1600 {
1601         return sprintf(buf, "%d\n",
1602                        channel_latency(channel,
1603                                        vmbus_connection.monitor_pages[1]));
1604 }
1605 static VMBUS_CHAN_ATTR(latency, S_IRUGO, channel_latency_show, NULL);
1606
1607 static ssize_t channel_interrupts_show(struct vmbus_channel *channel, char *buf)
1608 {
1609         return sprintf(buf, "%llu\n", channel->interrupts);
1610 }
1611 static VMBUS_CHAN_ATTR(interrupts, S_IRUGO, channel_interrupts_show, NULL);
1612
1613 static ssize_t channel_events_show(struct vmbus_channel *channel, char *buf)
1614 {
1615         return sprintf(buf, "%llu\n", channel->sig_events);
1616 }
1617 static VMBUS_CHAN_ATTR(events, S_IRUGO, channel_events_show, NULL);
1618
1619 static ssize_t channel_intr_in_full_show(struct vmbus_channel *channel,
1620                                          char *buf)
1621 {
1622         return sprintf(buf, "%llu\n",
1623                        (unsigned long long)channel->intr_in_full);
1624 }
1625 static VMBUS_CHAN_ATTR(intr_in_full, 0444, channel_intr_in_full_show, NULL);
1626
1627 static ssize_t channel_intr_out_empty_show(struct vmbus_channel *channel,
1628                                            char *buf)
1629 {
1630         return sprintf(buf, "%llu\n",
1631                        (unsigned long long)channel->intr_out_empty);
1632 }
1633 static VMBUS_CHAN_ATTR(intr_out_empty, 0444, channel_intr_out_empty_show, NULL);
1634
1635 static ssize_t channel_out_full_first_show(struct vmbus_channel *channel,
1636                                            char *buf)
1637 {
1638         return sprintf(buf, "%llu\n",
1639                        (unsigned long long)channel->out_full_first);
1640 }
1641 static VMBUS_CHAN_ATTR(out_full_first, 0444, channel_out_full_first_show, NULL);
1642
1643 static ssize_t channel_out_full_total_show(struct vmbus_channel *channel,
1644                                            char *buf)
1645 {
1646         return sprintf(buf, "%llu\n",
1647                        (unsigned long long)channel->out_full_total);
1648 }
1649 static VMBUS_CHAN_ATTR(out_full_total, 0444, channel_out_full_total_show, NULL);
1650
1651 static ssize_t subchannel_monitor_id_show(struct vmbus_channel *channel,
1652                                           char *buf)
1653 {
1654         return sprintf(buf, "%u\n", channel->offermsg.monitorid);
1655 }
1656 static VMBUS_CHAN_ATTR(monitor_id, S_IRUGO, subchannel_monitor_id_show, NULL);
1657
1658 static ssize_t subchannel_id_show(struct vmbus_channel *channel,
1659                                   char *buf)
1660 {
1661         return sprintf(buf, "%u\n",
1662                        channel->offermsg.offer.sub_channel_index);
1663 }
1664 static VMBUS_CHAN_ATTR_RO(subchannel_id);
1665
1666 static struct attribute *vmbus_chan_attrs[] = {
1667         &chan_attr_out_mask.attr,
1668         &chan_attr_in_mask.attr,
1669         &chan_attr_read_avail.attr,
1670         &chan_attr_write_avail.attr,
1671         &chan_attr_cpu.attr,
1672         &chan_attr_pending.attr,
1673         &chan_attr_latency.attr,
1674         &chan_attr_interrupts.attr,
1675         &chan_attr_events.attr,
1676         &chan_attr_intr_in_full.attr,
1677         &chan_attr_intr_out_empty.attr,
1678         &chan_attr_out_full_first.attr,
1679         &chan_attr_out_full_total.attr,
1680         &chan_attr_monitor_id.attr,
1681         &chan_attr_subchannel_id.attr,
1682         NULL
1683 };
1684
1685 /*
1686  * Channel-level attribute_group callback function. Returns the permission for
1687  * each attribute, and returns 0 if an attribute is not visible.
1688  */
1689 static umode_t vmbus_chan_attr_is_visible(struct kobject *kobj,
1690                                           struct attribute *attr, int idx)
1691 {
1692         const struct vmbus_channel *channel =
1693                 container_of(kobj, struct vmbus_channel, kobj);
1694
1695         /* Hide the monitor attributes if the monitor mechanism is not used. */
1696         if (!channel->offermsg.monitor_allocated &&
1697             (attr == &chan_attr_pending.attr ||
1698              attr == &chan_attr_latency.attr ||
1699              attr == &chan_attr_monitor_id.attr))
1700                 return 0;
1701
1702         return attr->mode;
1703 }
1704
1705 static struct attribute_group vmbus_chan_group = {
1706         .attrs = vmbus_chan_attrs,
1707         .is_visible = vmbus_chan_attr_is_visible
1708 };
1709
1710 static struct kobj_type vmbus_chan_ktype = {
1711         .sysfs_ops = &vmbus_chan_sysfs_ops,
1712         .release = vmbus_chan_release,
1713 };
1714
1715 /*
1716  * vmbus_add_channel_kobj - setup a sub-directory under device/channels
1717  */
1718 int vmbus_add_channel_kobj(struct hv_device *dev, struct vmbus_channel *channel)
1719 {
1720         const struct device *device = &dev->device;
1721         struct kobject *kobj = &channel->kobj;
1722         u32 relid = channel->offermsg.child_relid;
1723         int ret;
1724
1725         kobj->kset = dev->channels_kset;
1726         ret = kobject_init_and_add(kobj, &vmbus_chan_ktype, NULL,
1727                                    "%u", relid);
1728         if (ret)
1729                 return ret;
1730
1731         ret = sysfs_create_group(kobj, &vmbus_chan_group);
1732
1733         if (ret) {
1734                 /*
1735                  * The calling functions' error handling paths will cleanup the
1736                  * empty channel directory.
1737                  */
1738                 dev_err(device, "Unable to set up channel sysfs files\n");
1739                 return ret;
1740         }
1741
1742         kobject_uevent(kobj, KOBJ_ADD);
1743
1744         return 0;
1745 }
1746
1747 /*
1748  * vmbus_remove_channel_attr_group - remove the channel's attribute group
1749  */
1750 void vmbus_remove_channel_attr_group(struct vmbus_channel *channel)
1751 {
1752         sysfs_remove_group(&channel->kobj, &vmbus_chan_group);
1753 }
1754
1755 /*
1756  * vmbus_device_create - Creates and registers a new child device
1757  * on the vmbus.
1758  */
1759 struct hv_device *vmbus_device_create(const guid_t *type,
1760                                       const guid_t *instance,
1761                                       struct vmbus_channel *channel)
1762 {
1763         struct hv_device *child_device_obj;
1764
1765         child_device_obj = kzalloc(sizeof(struct hv_device), GFP_KERNEL);
1766         if (!child_device_obj) {
1767                 pr_err("Unable to allocate device object for child device\n");
1768                 return NULL;
1769         }
1770
1771         child_device_obj->channel = channel;
1772         guid_copy(&child_device_obj->dev_type, type);
1773         guid_copy(&child_device_obj->dev_instance, instance);
1774         child_device_obj->vendor_id = 0x1414; /* MSFT vendor ID */
1775
1776         return child_device_obj;
1777 }
1778
1779 /*
1780  * vmbus_device_register - Register the child device
1781  */
1782 int vmbus_device_register(struct hv_device *child_device_obj)
1783 {
1784         struct kobject *kobj = &child_device_obj->device.kobj;
1785         int ret;
1786
1787         dev_set_name(&child_device_obj->device, "%pUl",
1788                      child_device_obj->channel->offermsg.offer.if_instance.b);
1789
1790         child_device_obj->device.bus = &hv_bus;
1791         child_device_obj->device.parent = &hv_acpi_dev->dev;
1792         child_device_obj->device.release = vmbus_device_release;
1793
1794         /*
1795          * Register with the LDM. This will kick off the driver/device
1796          * binding...which will eventually call vmbus_match() and vmbus_probe()
1797          */
1798         ret = device_register(&child_device_obj->device);
1799         if (ret) {
1800                 pr_err("Unable to register child device\n");
1801                 return ret;
1802         }
1803
1804         child_device_obj->channels_kset = kset_create_and_add("channels",
1805                                                               NULL, kobj);
1806         if (!child_device_obj->channels_kset) {
1807                 ret = -ENOMEM;
1808                 goto err_dev_unregister;
1809         }
1810
1811         ret = vmbus_add_channel_kobj(child_device_obj,
1812                                      child_device_obj->channel);
1813         if (ret) {
1814                 pr_err("Unable to register primary channeln");
1815                 goto err_kset_unregister;
1816         }
1817
1818         return 0;
1819
1820 err_kset_unregister:
1821         kset_unregister(child_device_obj->channels_kset);
1822
1823 err_dev_unregister:
1824         device_unregister(&child_device_obj->device);
1825         return ret;
1826 }
1827
1828 /*
1829  * vmbus_device_unregister - Remove the specified child device
1830  * from the vmbus.
1831  */
1832 void vmbus_device_unregister(struct hv_device *device_obj)
1833 {
1834         pr_debug("child device %s unregistered\n",
1835                 dev_name(&device_obj->device));
1836
1837         kset_unregister(device_obj->channels_kset);
1838
1839         /*
1840          * Kick off the process of unregistering the device.
1841          * This will call vmbus_remove() and eventually vmbus_device_release()
1842          */
1843         device_unregister(&device_obj->device);
1844 }
1845
1846
1847 /*
1848  * VMBUS is an acpi enumerated device. Get the information we
1849  * need from DSDT.
1850  */
1851 #define VTPM_BASE_ADDRESS 0xfed40000
1852 static acpi_status vmbus_walk_resources(struct acpi_resource *res, void *ctx)
1853 {
1854         resource_size_t start = 0;
1855         resource_size_t end = 0;
1856         struct resource *new_res;
1857         struct resource **old_res = &hyperv_mmio;
1858         struct resource **prev_res = NULL;
1859
1860         switch (res->type) {
1861
1862         /*
1863          * "Address" descriptors are for bus windows. Ignore
1864          * "memory" descriptors, which are for registers on
1865          * devices.
1866          */
1867         case ACPI_RESOURCE_TYPE_ADDRESS32:
1868                 start = res->data.address32.address.minimum;
1869                 end = res->data.address32.address.maximum;
1870                 break;
1871
1872         case ACPI_RESOURCE_TYPE_ADDRESS64:
1873                 start = res->data.address64.address.minimum;
1874                 end = res->data.address64.address.maximum;
1875                 break;
1876
1877         default:
1878                 /* Unused resource type */
1879                 return AE_OK;
1880
1881         }
1882         /*
1883          * Ignore ranges that are below 1MB, as they're not
1884          * necessary or useful here.
1885          */
1886         if (end < 0x100000)
1887                 return AE_OK;
1888
1889         new_res = kzalloc(sizeof(*new_res), GFP_ATOMIC);
1890         if (!new_res)
1891                 return AE_NO_MEMORY;
1892
1893         /* If this range overlaps the virtual TPM, truncate it. */
1894         if (end > VTPM_BASE_ADDRESS && start < VTPM_BASE_ADDRESS)
1895                 end = VTPM_BASE_ADDRESS;
1896
1897         new_res->name = "hyperv mmio";
1898         new_res->flags = IORESOURCE_MEM;
1899         new_res->start = start;
1900         new_res->end = end;
1901
1902         /*
1903          * If two ranges are adjacent, merge them.
1904          */
1905         do {
1906                 if (!*old_res) {
1907                         *old_res = new_res;
1908                         break;
1909                 }
1910
1911                 if (((*old_res)->end + 1) == new_res->start) {
1912                         (*old_res)->end = new_res->end;
1913                         kfree(new_res);
1914                         break;
1915                 }
1916
1917                 if ((*old_res)->start == new_res->end + 1) {
1918                         (*old_res)->start = new_res->start;
1919                         kfree(new_res);
1920                         break;
1921                 }
1922
1923                 if ((*old_res)->start > new_res->end) {
1924                         new_res->sibling = *old_res;
1925                         if (prev_res)
1926                                 (*prev_res)->sibling = new_res;
1927                         *old_res = new_res;
1928                         break;
1929                 }
1930
1931                 prev_res = old_res;
1932                 old_res = &(*old_res)->sibling;
1933
1934         } while (1);
1935
1936         return AE_OK;
1937 }
1938
1939 static int vmbus_acpi_remove(struct acpi_device *device)
1940 {
1941         struct resource *cur_res;
1942         struct resource *next_res;
1943
1944         if (hyperv_mmio) {
1945                 if (fb_mmio) {
1946                         __release_region(hyperv_mmio, fb_mmio->start,
1947                                          resource_size(fb_mmio));
1948                         fb_mmio = NULL;
1949                 }
1950
1951                 for (cur_res = hyperv_mmio; cur_res; cur_res = next_res) {
1952                         next_res = cur_res->sibling;
1953                         kfree(cur_res);
1954                 }
1955         }
1956
1957         return 0;
1958 }
1959
1960 static void vmbus_reserve_fb(void)
1961 {
1962         int size;
1963         /*
1964          * Make a claim for the frame buffer in the resource tree under the
1965          * first node, which will be the one below 4GB.  The length seems to
1966          * be underreported, particularly in a Generation 1 VM.  So start out
1967          * reserving a larger area and make it smaller until it succeeds.
1968          */
1969
1970         if (screen_info.lfb_base) {
1971                 if (efi_enabled(EFI_BOOT))
1972                         size = max_t(__u32, screen_info.lfb_size, 0x800000);
1973                 else
1974                         size = max_t(__u32, screen_info.lfb_size, 0x4000000);
1975
1976                 for (; !fb_mmio && (size >= 0x100000); size >>= 1) {
1977                         fb_mmio = __request_region(hyperv_mmio,
1978                                                    screen_info.lfb_base, size,
1979                                                    fb_mmio_name, 0);
1980                 }
1981         }
1982 }
1983
1984 /**
1985  * vmbus_allocate_mmio() - Pick a memory-mapped I/O range.
1986  * @new:                If successful, supplied a pointer to the
1987  *                      allocated MMIO space.
1988  * @device_obj:         Identifies the caller
1989  * @min:                Minimum guest physical address of the
1990  *                      allocation
1991  * @max:                Maximum guest physical address
1992  * @size:               Size of the range to be allocated
1993  * @align:              Alignment of the range to be allocated
1994  * @fb_overlap_ok:      Whether this allocation can be allowed
1995  *                      to overlap the video frame buffer.
1996  *
1997  * This function walks the resources granted to VMBus by the
1998  * _CRS object in the ACPI namespace underneath the parent
1999  * "bridge" whether that's a root PCI bus in the Generation 1
2000  * case or a Module Device in the Generation 2 case.  It then
2001  * attempts to allocate from the global MMIO pool in a way that
2002  * matches the constraints supplied in these parameters and by
2003  * that _CRS.
2004  *
2005  * Return: 0 on success, -errno on failure
2006  */
2007 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
2008                         resource_size_t min, resource_size_t max,
2009                         resource_size_t size, resource_size_t align,
2010                         bool fb_overlap_ok)
2011 {
2012         struct resource *iter, *shadow;
2013         resource_size_t range_min, range_max, start;
2014         const char *dev_n = dev_name(&device_obj->device);
2015         int retval;
2016
2017         retval = -ENXIO;
2018         down(&hyperv_mmio_lock);
2019
2020         /*
2021          * If overlaps with frame buffers are allowed, then first attempt to
2022          * make the allocation from within the reserved region.  Because it
2023          * is already reserved, no shadow allocation is necessary.
2024          */
2025         if (fb_overlap_ok && fb_mmio && !(min > fb_mmio->end) &&
2026             !(max < fb_mmio->start)) {
2027
2028                 range_min = fb_mmio->start;
2029                 range_max = fb_mmio->end;
2030                 start = (range_min + align - 1) & ~(align - 1);
2031                 for (; start + size - 1 <= range_max; start += align) {
2032                         *new = request_mem_region_exclusive(start, size, dev_n);
2033                         if (*new) {
2034                                 retval = 0;
2035                                 goto exit;
2036                         }
2037                 }
2038         }
2039
2040         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
2041                 if ((iter->start >= max) || (iter->end <= min))
2042                         continue;
2043
2044                 range_min = iter->start;
2045                 range_max = iter->end;
2046                 start = (range_min + align - 1) & ~(align - 1);
2047                 for (; start + size - 1 <= range_max; start += align) {
2048                         shadow = __request_region(iter, start, size, NULL,
2049                                                   IORESOURCE_BUSY);
2050                         if (!shadow)
2051                                 continue;
2052
2053                         *new = request_mem_region_exclusive(start, size, dev_n);
2054                         if (*new) {
2055                                 shadow->name = (char *)*new;
2056                                 retval = 0;
2057                                 goto exit;
2058                         }
2059
2060                         __release_region(iter, start, size);
2061                 }
2062         }
2063
2064 exit:
2065         up(&hyperv_mmio_lock);
2066         return retval;
2067 }
2068 EXPORT_SYMBOL_GPL(vmbus_allocate_mmio);
2069
2070 /**
2071  * vmbus_free_mmio() - Free a memory-mapped I/O range.
2072  * @start:              Base address of region to release.
2073  * @size:               Size of the range to be allocated
2074  *
2075  * This function releases anything requested by
2076  * vmbus_mmio_allocate().
2077  */
2078 void vmbus_free_mmio(resource_size_t start, resource_size_t size)
2079 {
2080         struct resource *iter;
2081
2082         down(&hyperv_mmio_lock);
2083         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
2084                 if ((iter->start >= start + size) || (iter->end <= start))
2085                         continue;
2086
2087                 __release_region(iter, start, size);
2088         }
2089         release_mem_region(start, size);
2090         up(&hyperv_mmio_lock);
2091
2092 }
2093 EXPORT_SYMBOL_GPL(vmbus_free_mmio);
2094
2095 static int vmbus_acpi_add(struct acpi_device *device)
2096 {
2097         acpi_status result;
2098         int ret_val = -ENODEV;
2099         struct acpi_device *ancestor;
2100
2101         hv_acpi_dev = device;
2102
2103         result = acpi_walk_resources(device->handle, METHOD_NAME__CRS,
2104                                         vmbus_walk_resources, NULL);
2105
2106         if (ACPI_FAILURE(result))
2107                 goto acpi_walk_err;
2108         /*
2109          * Some ancestor of the vmbus acpi device (Gen1 or Gen2
2110          * firmware) is the VMOD that has the mmio ranges. Get that.
2111          */
2112         for (ancestor = device->parent; ancestor; ancestor = ancestor->parent) {
2113                 result = acpi_walk_resources(ancestor->handle, METHOD_NAME__CRS,
2114                                              vmbus_walk_resources, NULL);
2115
2116                 if (ACPI_FAILURE(result))
2117                         continue;
2118                 if (hyperv_mmio) {
2119                         vmbus_reserve_fb();
2120                         break;
2121                 }
2122         }
2123         ret_val = 0;
2124
2125 acpi_walk_err:
2126         complete(&probe_event);
2127         if (ret_val)
2128                 vmbus_acpi_remove(device);
2129         return ret_val;
2130 }
2131
2132 #ifdef CONFIG_PM_SLEEP
2133 static int vmbus_bus_suspend(struct device *dev)
2134 {
2135         struct vmbus_channel *channel, *sc;
2136         unsigned long flags;
2137
2138         while (atomic_read(&vmbus_connection.offer_in_progress) != 0) {
2139                 /*
2140                  * We wait here until the completion of any channel
2141                  * offers that are currently in progress.
2142                  */
2143                 msleep(1);
2144         }
2145
2146         mutex_lock(&vmbus_connection.channel_mutex);
2147         list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
2148                 if (!is_hvsock_channel(channel))
2149                         continue;
2150
2151                 vmbus_force_channel_rescinded(channel);
2152         }
2153         mutex_unlock(&vmbus_connection.channel_mutex);
2154
2155         /*
2156          * Wait until all the sub-channels and hv_sock channels have been
2157          * cleaned up. Sub-channels should be destroyed upon suspend, otherwise
2158          * they would conflict with the new sub-channels that will be created
2159          * in the resume path. hv_sock channels should also be destroyed, but
2160          * a hv_sock channel of an established hv_sock connection can not be
2161          * really destroyed since it may still be referenced by the userspace
2162          * application, so we just force the hv_sock channel to be rescinded
2163          * by vmbus_force_channel_rescinded(), and the userspace application
2164          * will thoroughly destroy the channel after hibernation.
2165          *
2166          * Note: the counter nr_chan_close_on_suspend may never go above 0 if
2167          * the VM has no sub-channel and hv_sock channel, e.g. a 1-vCPU VM.
2168          */
2169         if (atomic_read(&vmbus_connection.nr_chan_close_on_suspend) > 0)
2170                 wait_for_completion(&vmbus_connection.ready_for_suspend_event);
2171
2172         WARN_ON(atomic_read(&vmbus_connection.nr_chan_fixup_on_resume) != 0);
2173
2174         mutex_lock(&vmbus_connection.channel_mutex);
2175
2176         list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
2177                 /*
2178                  * Invalidate the field. Upon resume, vmbus_onoffer() will fix
2179                  * up the field, and the other fields (if necessary).
2180                  */
2181                 channel->offermsg.child_relid = INVALID_RELID;
2182
2183                 if (is_hvsock_channel(channel)) {
2184                         if (!channel->rescind) {
2185                                 pr_err("hv_sock channel not rescinded!\n");
2186                                 WARN_ON_ONCE(1);
2187                         }
2188                         continue;
2189                 }
2190
2191                 spin_lock_irqsave(&channel->lock, flags);
2192                 list_for_each_entry(sc, &channel->sc_list, sc_list) {
2193                         pr_err("Sub-channel not deleted!\n");
2194                         WARN_ON_ONCE(1);
2195                 }
2196                 spin_unlock_irqrestore(&channel->lock, flags);
2197
2198                 atomic_inc(&vmbus_connection.nr_chan_fixup_on_resume);
2199         }
2200
2201         mutex_unlock(&vmbus_connection.channel_mutex);
2202
2203         vmbus_initiate_unload(false);
2204
2205         vmbus_connection.conn_state = DISCONNECTED;
2206
2207         /* Reset the event for the next resume. */
2208         reinit_completion(&vmbus_connection.ready_for_resume_event);
2209
2210         return 0;
2211 }
2212
2213 static int vmbus_bus_resume(struct device *dev)
2214 {
2215         struct vmbus_channel_msginfo *msginfo;
2216         size_t msgsize;
2217         int ret;
2218
2219         /*
2220          * We only use the 'vmbus_proto_version', which was in use before
2221          * hibernation, to re-negotiate with the host.
2222          */
2223         if (vmbus_proto_version == VERSION_INVAL ||
2224             vmbus_proto_version == 0) {
2225                 pr_err("Invalid proto version = 0x%x\n", vmbus_proto_version);
2226                 return -EINVAL;
2227         }
2228
2229         msgsize = sizeof(*msginfo) +
2230                   sizeof(struct vmbus_channel_initiate_contact);
2231
2232         msginfo = kzalloc(msgsize, GFP_KERNEL);
2233
2234         if (msginfo == NULL)
2235                 return -ENOMEM;
2236
2237         ret = vmbus_negotiate_version(msginfo, vmbus_proto_version);
2238
2239         kfree(msginfo);
2240
2241         if (ret != 0)
2242                 return ret;
2243
2244         WARN_ON(atomic_read(&vmbus_connection.nr_chan_fixup_on_resume) == 0);
2245
2246         vmbus_request_offers();
2247
2248         wait_for_completion(&vmbus_connection.ready_for_resume_event);
2249
2250         /* Reset the event for the next suspend. */
2251         reinit_completion(&vmbus_connection.ready_for_suspend_event);
2252
2253         return 0;
2254 }
2255 #endif /* CONFIG_PM_SLEEP */
2256
2257 static const struct acpi_device_id vmbus_acpi_device_ids[] = {
2258         {"VMBUS", 0},
2259         {"VMBus", 0},
2260         {"", 0},
2261 };
2262 MODULE_DEVICE_TABLE(acpi, vmbus_acpi_device_ids);
2263
2264 /*
2265  * Note: we must use SET_NOIRQ_SYSTEM_SLEEP_PM_OPS rather than
2266  * SET_SYSTEM_SLEEP_PM_OPS, otherwise NIC SR-IOV can not work, because the
2267  * "pci_dev_pm_ops" uses the "noirq" callbacks: in the resume path, the
2268  * pci "noirq" restore callback runs before "non-noirq" callbacks (see
2269  * resume_target_kernel() -> dpm_resume_start(), and hibernation_restore() ->
2270  * dpm_resume_end()). This means vmbus_bus_resume() and the pci-hyperv's
2271  * resume callback must also run via the "noirq" callbacks.
2272  */
2273 static const struct dev_pm_ops vmbus_bus_pm = {
2274         SET_NOIRQ_SYSTEM_SLEEP_PM_OPS(vmbus_bus_suspend, vmbus_bus_resume)
2275 };
2276
2277 static struct acpi_driver vmbus_acpi_driver = {
2278         .name = "vmbus",
2279         .ids = vmbus_acpi_device_ids,
2280         .ops = {
2281                 .add = vmbus_acpi_add,
2282                 .remove = vmbus_acpi_remove,
2283         },
2284         .drv.pm = &vmbus_bus_pm,
2285 };
2286
2287 static void hv_kexec_handler(void)
2288 {
2289         hv_stimer_global_cleanup();
2290         vmbus_initiate_unload(false);
2291         vmbus_connection.conn_state = DISCONNECTED;
2292         /* Make sure conn_state is set as hv_synic_cleanup checks for it */
2293         mb();
2294         cpuhp_remove_state(hyperv_cpuhp_online);
2295         hyperv_cleanup();
2296 };
2297
2298 static void hv_crash_handler(struct pt_regs *regs)
2299 {
2300         int cpu;
2301
2302         vmbus_initiate_unload(true);
2303         /*
2304          * In crash handler we can't schedule synic cleanup for all CPUs,
2305          * doing the cleanup for current CPU only. This should be sufficient
2306          * for kdump.
2307          */
2308         vmbus_connection.conn_state = DISCONNECTED;
2309         cpu = smp_processor_id();
2310         hv_stimer_cleanup(cpu);
2311         hv_synic_cleanup(cpu);
2312         hyperv_cleanup();
2313 };
2314
2315 static int hv_synic_suspend(void)
2316 {
2317         /*
2318          * When we reach here, all the non-boot CPUs have been offlined, and
2319          * the stimers on them have been unbound in hv_synic_cleanup() ->
2320          * hv_stimer_cleanup() -> clockevents_unbind_device().
2321          *
2322          * hv_synic_suspend() only runs on CPU0 with interrupts disabled. Here
2323          * we do not unbind the stimer on CPU0 because: 1) it's unnecessary
2324          * because the interrupts remain disabled between syscore_suspend()
2325          * and syscore_resume(): see create_image() and resume_target_kernel();
2326          * 2) the stimer on CPU0 is automatically disabled later by
2327          * syscore_suspend() -> timekeeping_suspend() -> tick_suspend() -> ...
2328          * -> clockevents_shutdown() -> ... -> hv_ce_shutdown(); 3) a warning
2329          * would be triggered if we call clockevents_unbind_device(), which
2330          * may sleep, in an interrupts-disabled context. So, we intentionally
2331          * don't call hv_stimer_cleanup(0) here.
2332          */
2333
2334         hv_synic_disable_regs(0);
2335
2336         return 0;
2337 }
2338
2339 static void hv_synic_resume(void)
2340 {
2341         hv_synic_enable_regs(0);
2342
2343         /*
2344          * Note: we don't need to call hv_stimer_init(0), because the timer
2345          * on CPU0 is not unbound in hv_synic_suspend(), and the timer is
2346          * automatically re-enabled in timekeeping_resume().
2347          */
2348 }
2349
2350 /* The callbacks run only on CPU0, with irqs_disabled. */
2351 static struct syscore_ops hv_synic_syscore_ops = {
2352         .suspend = hv_synic_suspend,
2353         .resume = hv_synic_resume,
2354 };
2355
2356 static int __init hv_acpi_init(void)
2357 {
2358         int ret, t;
2359
2360         if (!hv_is_hyperv_initialized())
2361                 return -ENODEV;
2362
2363         init_completion(&probe_event);
2364
2365         /*
2366          * Get ACPI resources first.
2367          */
2368         ret = acpi_bus_register_driver(&vmbus_acpi_driver);
2369
2370         if (ret)
2371                 return ret;
2372
2373         t = wait_for_completion_timeout(&probe_event, 5*HZ);
2374         if (t == 0) {
2375                 ret = -ETIMEDOUT;
2376                 goto cleanup;
2377         }
2378
2379         ret = vmbus_bus_init();
2380         if (ret)
2381                 goto cleanup;
2382
2383         hv_setup_kexec_handler(hv_kexec_handler);
2384         hv_setup_crash_handler(hv_crash_handler);
2385
2386         register_syscore_ops(&hv_synic_syscore_ops);
2387
2388         return 0;
2389
2390 cleanup:
2391         acpi_bus_unregister_driver(&vmbus_acpi_driver);
2392         hv_acpi_dev = NULL;
2393         return ret;
2394 }
2395
2396 static void __exit vmbus_exit(void)
2397 {
2398         int cpu;
2399
2400         unregister_syscore_ops(&hv_synic_syscore_ops);
2401
2402         hv_remove_kexec_handler();
2403         hv_remove_crash_handler();
2404         vmbus_connection.conn_state = DISCONNECTED;
2405         hv_stimer_global_cleanup();
2406         vmbus_disconnect();
2407         hv_remove_vmbus_irq();
2408         for_each_online_cpu(cpu) {
2409                 struct hv_per_cpu_context *hv_cpu
2410                         = per_cpu_ptr(hv_context.cpu_context, cpu);
2411
2412                 tasklet_kill(&hv_cpu->msg_dpc);
2413         }
2414         vmbus_free_channels();
2415
2416         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
2417                 kmsg_dump_unregister(&hv_kmsg_dumper);
2418                 unregister_die_notifier(&hyperv_die_block);
2419                 atomic_notifier_chain_unregister(&panic_notifier_list,
2420                                                  &hyperv_panic_block);
2421         }
2422
2423         free_page((unsigned long)hv_panic_page);
2424         unregister_sysctl_table(hv_ctl_table_hdr);
2425         hv_ctl_table_hdr = NULL;
2426         bus_unregister(&hv_bus);
2427
2428         cpuhp_remove_state(hyperv_cpuhp_online);
2429         hv_synic_free();
2430         acpi_bus_unregister_driver(&vmbus_acpi_driver);
2431 }
2432
2433
2434 MODULE_LICENSE("GPL");
2435 MODULE_DESCRIPTION("Microsoft Hyper-V VMBus Driver");
2436
2437 subsys_initcall(hv_acpi_init);
2438 module_exit(vmbus_exit);