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_MUTEX(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         hv_debug_rm_dev_dir(hv_dev);
 964
 965         mutex_lock(&vmbus_connection.channel_mutex);
 966         hv_process_channel_removal(channel);
 967         mutex_unlock(&vmbus_connection.channel_mutex);
 968         kfree(hv_dev);
 969 }
 970
 971 /*
 972  * Note: we must use SET_NOIRQ_SYSTEM_SLEEP_PM_OPS rather than
 973  * SET_SYSTEM_SLEEP_PM_OPS: see the comment before vmbus_bus_pm.
 974  */
 975 static const struct dev_pm_ops vmbus_pm = {
 976         SET_NOIRQ_SYSTEM_SLEEP_PM_OPS(vmbus_suspend, vmbus_resume)
 977 };
 978
 979 /* The one and only one */
 980 static struct bus_type  hv_bus = {
 981         .name =         "vmbus",
 982         .match =                vmbus_match,
 983         .shutdown =             vmbus_shutdown,
 984         .remove =               vmbus_remove,
 985         .probe =                vmbus_probe,
 986         .uevent =               vmbus_uevent,
 987         .dev_groups =           vmbus_dev_groups,
 988         .drv_groups =           vmbus_drv_groups,
 989         .pm =                   &vmbus_pm,
 990 };
 991
 992 struct onmessage_work_context {
 993         struct work_struct work;
 994         struct hv_message msg;
 995 };
 996
 997 static void vmbus_onmessage_work(struct work_struct *work)
 998 {
 999         struct onmessage_work_context *ctx;
1000
1001         /* Do not process messages if we're in DISCONNECTED state */
1002         if (vmbus_connection.conn_state == DISCONNECTED)
1003                 return;
1004
1005         ctx = container_of(work, struct onmessage_work_context,
1006                            work);
1007         vmbus_onmessage(&ctx->msg);
1008         kfree(ctx);
1009 }
1010
1011 void vmbus_on_msg_dpc(unsigned long data)
1012 {
1013         struct hv_per_cpu_context *hv_cpu = (void *)data;
1014         void *page_addr = hv_cpu->synic_message_page;
1015         struct hv_message *msg = (struct hv_message *)page_addr +
1016                                   VMBUS_MESSAGE_SINT;
1017         struct vmbus_channel_message_header *hdr;
1018         const struct vmbus_channel_message_table_entry *entry;
1019         struct onmessage_work_context *ctx;
1020         u32 message_type = msg->header.message_type;
1021
1022         if (message_type == HVMSG_NONE)
1023                 /* no msg */
1024                 return;
1025
1026         hdr = (struct vmbus_channel_message_header *)msg->u.payload;
1027
1028         trace_vmbus_on_msg_dpc(hdr);
1029
1030         if (hdr->msgtype >= CHANNELMSG_COUNT) {
1031                 WARN_ONCE(1, "unknown msgtype=%d\n", hdr->msgtype);
1032                 goto msg_handled;
1033         }
1034
1035         entry = &channel_message_table[hdr->msgtype];
1036         if (entry->handler_type == VMHT_BLOCKING) {
1037                 ctx = kmalloc(sizeof(*ctx), GFP_ATOMIC);
1038                 if (ctx == NULL)
1039                         return;
1040
1041                 INIT_WORK(&ctx->work, vmbus_onmessage_work);
1042                 memcpy(&ctx->msg, msg, sizeof(*msg));
1043
1044                 /*
1045                  * The host can generate a rescind message while we
1046                  * may still be handling the original offer. We deal with
1047                  * this condition by ensuring the processing is done on the
1048                  * same CPU.
1049                  */
1050                 switch (hdr->msgtype) {
1051                 case CHANNELMSG_RESCIND_CHANNELOFFER:
1052                         /*
1053                          * If we are handling the rescind message;
1054                          * schedule the work on the global work queue.
1055                          */
1056                         schedule_work_on(vmbus_connection.connect_cpu,
1057                                          &ctx->work);
1058                         break;
1059
1060                 case CHANNELMSG_OFFERCHANNEL:
1061                         atomic_inc(&vmbus_connection.offer_in_progress);
1062                         queue_work_on(vmbus_connection.connect_cpu,
1063                                       vmbus_connection.work_queue,
1064                                       &ctx->work);
1065                         break;
1066
1067                 default:
1068                         queue_work(vmbus_connection.work_queue, &ctx->work);
1069                 }
1070         } else
1071                 entry->message_handler(hdr);
1072
1073 msg_handled:
1074         vmbus_signal_eom(msg, message_type);
1075 }
1076
1077 #ifdef CONFIG_PM_SLEEP
1078 /*
1079  * Fake RESCIND_CHANNEL messages to clean up hv_sock channels by force for
1080  * hibernation, because hv_sock connections can not persist across hibernation.
1081  */
1082 static void vmbus_force_channel_rescinded(struct vmbus_channel *channel)
1083 {
1084         struct onmessage_work_context *ctx;
1085         struct vmbus_channel_rescind_offer *rescind;
1086
1087         WARN_ON(!is_hvsock_channel(channel));
1088
1089         /*
1090          * sizeof(*ctx) is small and the allocation should really not fail,
1091          * otherwise the state of the hv_sock connections ends up in limbo.
1092          */
1093         ctx = kzalloc(sizeof(*ctx), GFP_KERNEL | __GFP_NOFAIL);
1094
1095         /*
1096          * So far, these are not really used by Linux. Just set them to the
1097          * reasonable values conforming to the definitions of the fields.
1098          */
1099         ctx->msg.header.message_type = 1;
1100         ctx->msg.header.payload_size = sizeof(*rescind);
1101
1102         /* These values are actually used by Linux. */
1103         rescind = (struct vmbus_channel_rescind_offer *)ctx->msg.u.payload;
1104         rescind->header.msgtype = CHANNELMSG_RESCIND_CHANNELOFFER;
1105         rescind->child_relid = channel->offermsg.child_relid;
1106
1107         INIT_WORK(&ctx->work, vmbus_onmessage_work);
1108
1109         queue_work_on(vmbus_connection.connect_cpu,
1110                       vmbus_connection.work_queue,
1111                       &ctx->work);
1112 }
1113 #endif /* CONFIG_PM_SLEEP */
1114
1115 /*
1116  * Direct callback for channels using other deferred processing
1117  */
1118 static void vmbus_channel_isr(struct vmbus_channel *channel)
1119 {
1120         void (*callback_fn)(void *);
1121
1122         callback_fn = READ_ONCE(channel->onchannel_callback);
1123         if (likely(callback_fn != NULL))
1124                 (*callback_fn)(channel->channel_callback_context);
1125 }
1126
1127 /*
1128  * Schedule all channels with events pending
1129  */
1130 static void vmbus_chan_sched(struct hv_per_cpu_context *hv_cpu)
1131 {
1132         unsigned long *recv_int_page;
1133         u32 maxbits, relid;
1134
1135         if (vmbus_proto_version < VERSION_WIN8) {
1136                 maxbits = MAX_NUM_CHANNELS_SUPPORTED;
1137                 recv_int_page = vmbus_connection.recv_int_page;
1138         } else {
1139                 /*
1140                  * When the host is win8 and beyond, the event page
1141                  * can be directly checked to get the id of the channel
1142                  * that has the interrupt pending.
1143                  */
1144                 void *page_addr = hv_cpu->synic_event_page;
1145                 union hv_synic_event_flags *event
1146                         = (union hv_synic_event_flags *)page_addr +
1147                                                  VMBUS_MESSAGE_SINT;
1148
1149                 maxbits = HV_EVENT_FLAGS_COUNT;
1150                 recv_int_page = event->flags;
1151         }
1152
1153         if (unlikely(!recv_int_page))
1154                 return;
1155
1156         for_each_set_bit(relid, recv_int_page, maxbits) {
1157                 struct vmbus_channel *channel;
1158
1159                 if (!sync_test_and_clear_bit(relid, recv_int_page))
1160                         continue;
1161
1162                 /* Special case - vmbus channel protocol msg */
1163                 if (relid == 0)
1164                         continue;
1165
1166                 rcu_read_lock();
1167
1168                 /* Find channel based on relid */
1169                 list_for_each_entry_rcu(channel, &hv_cpu->chan_list, percpu_list) {
1170                         if (channel->offermsg.child_relid != relid)
1171                                 continue;
1172
1173                         if (channel->rescind)
1174                                 continue;
1175
1176                         trace_vmbus_chan_sched(channel);
1177
1178                         ++channel->interrupts;
1179
1180                         switch (channel->callback_mode) {
1181                         case HV_CALL_ISR:
1182                                 vmbus_channel_isr(channel);
1183                                 break;
1184
1185                         case HV_CALL_BATCHED:
1186                                 hv_begin_read(&channel->inbound);
1187                                 /* fallthrough */
1188                         case HV_CALL_DIRECT:
1189                                 tasklet_schedule(&channel->callback_event);
1190                         }
1191                 }
1192
1193                 rcu_read_unlock();
1194         }
1195 }
1196
1197 static void vmbus_isr(void)
1198 {
1199         struct hv_per_cpu_context *hv_cpu
1200                 = this_cpu_ptr(hv_context.cpu_context);
1201         void *page_addr = hv_cpu->synic_event_page;
1202         struct hv_message *msg;
1203         union hv_synic_event_flags *event;
1204         bool handled = false;
1205
1206         if (unlikely(page_addr == NULL))
1207                 return;
1208
1209         event = (union hv_synic_event_flags *)page_addr +
1210                                          VMBUS_MESSAGE_SINT;
1211         /*
1212          * Check for events before checking for messages. This is the order
1213          * in which events and messages are checked in Windows guests on
1214          * Hyper-V, and the Windows team suggested we do the same.
1215          */
1216
1217         if ((vmbus_proto_version == VERSION_WS2008) ||
1218                 (vmbus_proto_version == VERSION_WIN7)) {
1219
1220                 /* Since we are a child, we only need to check bit 0 */
1221                 if (sync_test_and_clear_bit(0, event->flags))
1222                         handled = true;
1223         } else {
1224                 /*
1225                  * Our host is win8 or above. The signaling mechanism
1226                  * has changed and we can directly look at the event page.
1227                  * If bit n is set then we have an interrup on the channel
1228                  * whose id is n.
1229                  */
1230                 handled = true;
1231         }
1232
1233         if (handled)
1234                 vmbus_chan_sched(hv_cpu);
1235
1236         page_addr = hv_cpu->synic_message_page;
1237         msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
1238
1239         /* Check if there are actual msgs to be processed */
1240         if (msg->header.message_type != HVMSG_NONE) {
1241                 if (msg->header.message_type == HVMSG_TIMER_EXPIRED) {
1242                         hv_stimer0_isr();
1243                         vmbus_signal_eom(msg, HVMSG_TIMER_EXPIRED);
1244                 } else
1245                         tasklet_schedule(&hv_cpu->msg_dpc);
1246         }
1247
1248         add_interrupt_randomness(HYPERVISOR_CALLBACK_VECTOR, 0);
1249 }
1250
1251 /*
1252  * Boolean to control whether to report panic messages over Hyper-V.
1253  *
1254  * It can be set via /proc/sys/kernel/hyperv/record_panic_msg
1255  */
1256 static int sysctl_record_panic_msg = 1;
1257
1258 /*
1259  * Callback from kmsg_dump. Grab as much as possible from the end of the kmsg
1260  * buffer and call into Hyper-V to transfer the data.
1261  */
1262 static void hv_kmsg_dump(struct kmsg_dumper *dumper,
1263                          enum kmsg_dump_reason reason)
1264 {
1265         size_t bytes_written;
1266         phys_addr_t panic_pa;
1267
1268         /* We are only interested in panics. */
1269         if ((reason != KMSG_DUMP_PANIC) || (!sysctl_record_panic_msg))
1270                 return;
1271
1272         panic_pa = virt_to_phys(hv_panic_page);
1273
1274         /*
1275          * Write dump contents to the page. No need to synchronize; panic should
1276          * be single-threaded.
1277          */
1278         kmsg_dump_get_buffer(dumper, true, hv_panic_page, HV_HYP_PAGE_SIZE,
1279                              &bytes_written);
1280         if (bytes_written)
1281                 hyperv_report_panic_msg(panic_pa, bytes_written);
1282 }
1283
1284 static struct kmsg_dumper hv_kmsg_dumper = {
1285         .dump = hv_kmsg_dump,
1286 };
1287
1288 static struct ctl_table_header *hv_ctl_table_hdr;
1289
1290 /*
1291  * sysctl option to allow the user to control whether kmsg data should be
1292  * reported to Hyper-V on panic.
1293  */
1294 static struct ctl_table hv_ctl_table[] = {
1295         {
1296                 .procname       = "hyperv_record_panic_msg",
1297                 .data           = &sysctl_record_panic_msg,
1298                 .maxlen         = sizeof(int),
1299                 .mode           = 0644,
1300                 .proc_handler   = proc_dointvec_minmax,
1301                 .extra1         = SYSCTL_ZERO,
1302                 .extra2         = SYSCTL_ONE
1303         },
1304         {}
1305 };
1306
1307 static struct ctl_table hv_root_table[] = {
1308         {
1309                 .procname       = "kernel",
1310                 .mode           = 0555,
1311                 .child          = hv_ctl_table
1312         },
1313         {}
1314 };
1315
1316 /*
1317  * vmbus_bus_init -Main vmbus driver initialization routine.
1318  *
1319  * Here, we
1320  *      - initialize the vmbus driver context
1321  *      - invoke the vmbus hv main init routine
1322  *      - retrieve the channel offers
1323  */
1324 static int vmbus_bus_init(void)
1325 {
1326         int ret;
1327
1328         /* Hypervisor initialization...setup hypercall page..etc */
1329         ret = hv_init();
1330         if (ret != 0) {
1331                 pr_err("Unable to initialize the hypervisor - 0x%x\n", ret);
1332                 return ret;
1333         }
1334
1335         ret = bus_register(&hv_bus);
1336         if (ret)
1337                 return ret;
1338
1339         hv_setup_vmbus_irq(vmbus_isr);
1340
1341         ret = hv_synic_alloc();
1342         if (ret)
1343                 goto err_alloc;
1344
1345         /*
1346          * Initialize the per-cpu interrupt state and stimer state.
1347          * Then connect to the host.
1348          */
1349         ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "hyperv/vmbus:online",
1350                                 hv_synic_init, hv_synic_cleanup);
1351         if (ret < 0)
1352                 goto err_cpuhp;
1353         hyperv_cpuhp_online = ret;
1354
1355         ret = vmbus_connect();
1356         if (ret)
1357                 goto err_connect;
1358
1359         /*
1360          * Only register if the crash MSRs are available
1361          */
1362         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1363                 u64 hyperv_crash_ctl;
1364                 /*
1365                  * Sysctl registration is not fatal, since by default
1366                  * reporting is enabled.
1367                  */
1368                 hv_ctl_table_hdr = register_sysctl_table(hv_root_table);
1369                 if (!hv_ctl_table_hdr)
1370                         pr_err("Hyper-V: sysctl table register error");
1371
1372                 /*
1373                  * Register for panic kmsg callback only if the right
1374                  * capability is supported by the hypervisor.
1375                  */
1376                 hv_get_crash_ctl(hyperv_crash_ctl);
1377                 if (hyperv_crash_ctl & HV_CRASH_CTL_CRASH_NOTIFY_MSG) {
1378                         hv_panic_page = (void *)hv_alloc_hyperv_zeroed_page();
1379                         if (hv_panic_page) {
1380                                 ret = kmsg_dump_register(&hv_kmsg_dumper);
1381                                 if (ret)
1382                                         pr_err("Hyper-V: kmsg dump register "
1383                                                 "error 0x%x\n", ret);
1384                         } else
1385                                 pr_err("Hyper-V: panic message page memory "
1386                                         "allocation failed");
1387                 }
1388
1389                 register_die_notifier(&hyperv_die_block);
1390                 atomic_notifier_chain_register(&panic_notifier_list,
1391                                                &hyperv_panic_block);
1392         }
1393
1394         vmbus_request_offers();
1395
1396         return 0;
1397
1398 err_connect:
1399         cpuhp_remove_state(hyperv_cpuhp_online);
1400 err_cpuhp:
1401         hv_synic_free();
1402 err_alloc:
1403         hv_remove_vmbus_irq();
1404
1405         bus_unregister(&hv_bus);
1406         hv_free_hyperv_page((unsigned long)hv_panic_page);
1407         unregister_sysctl_table(hv_ctl_table_hdr);
1408         hv_ctl_table_hdr = NULL;
1409         return ret;
1410 }
1411
1412 /**
1413  * __vmbus_child_driver_register() - Register a vmbus's driver
1414  * @hv_driver: Pointer to driver structure you want to register
1415  * @owner: owner module of the drv
1416  * @mod_name: module name string
1417  *
1418  * Registers the given driver with Linux through the 'driver_register()' call
1419  * and sets up the hyper-v vmbus handling for this driver.
1420  * It will return the state of the 'driver_register()' call.
1421  *
1422  */
1423 int __vmbus_driver_register(struct hv_driver *hv_driver, struct module *owner, const char *mod_name)
1424 {
1425         int ret;
1426
1427         pr_info("registering driver %s\n", hv_driver->name);
1428
1429         ret = vmbus_exists();
1430         if (ret < 0)
1431                 return ret;
1432
1433         hv_driver->driver.name = hv_driver->name;
1434         hv_driver->driver.owner = owner;
1435         hv_driver->driver.mod_name = mod_name;
1436         hv_driver->driver.bus = &hv_bus;
1437
1438         spin_lock_init(&hv_driver->dynids.lock);
1439         INIT_LIST_HEAD(&hv_driver->dynids.list);
1440
1441         ret = driver_register(&hv_driver->driver);
1442
1443         return ret;
1444 }
1445 EXPORT_SYMBOL_GPL(__vmbus_driver_register);
1446
1447 /**
1448  * vmbus_driver_unregister() - Unregister a vmbus's driver
1449  * @hv_driver: Pointer to driver structure you want to
1450  *             un-register
1451  *
1452  * Un-register the given driver that was previous registered with a call to
1453  * vmbus_driver_register()
1454  */
1455 void vmbus_driver_unregister(struct hv_driver *hv_driver)
1456 {
1457         pr_info("unregistering driver %s\n", hv_driver->name);
1458
1459         if (!vmbus_exists()) {
1460                 driver_unregister(&hv_driver->driver);
1461                 vmbus_free_dynids(hv_driver);
1462         }
1463 }
1464 EXPORT_SYMBOL_GPL(vmbus_driver_unregister);
1465
1466
1467 /*
1468  * Called when last reference to channel is gone.
1469  */
1470 static void vmbus_chan_release(struct kobject *kobj)
1471 {
1472         struct vmbus_channel *channel
1473                 = container_of(kobj, struct vmbus_channel, kobj);
1474
1475         kfree_rcu(channel, rcu);
1476 }
1477
1478 struct vmbus_chan_attribute {
1479         struct attribute attr;
1480         ssize_t (*show)(struct vmbus_channel *chan, char *buf);
1481         ssize_t (*store)(struct vmbus_channel *chan,
1482                          const char *buf, size_t count);
1483 };
1484 #define VMBUS_CHAN_ATTR(_name, _mode, _show, _store) \
1485         struct vmbus_chan_attribute chan_attr_##_name \
1486                 = __ATTR(_name, _mode, _show, _store)
1487 #define VMBUS_CHAN_ATTR_RW(_name) \
1488         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RW(_name)
1489 #define VMBUS_CHAN_ATTR_RO(_name) \
1490         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RO(_name)
1491 #define VMBUS_CHAN_ATTR_WO(_name) \
1492         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_WO(_name)
1493
1494 static ssize_t vmbus_chan_attr_show(struct kobject *kobj,
1495                                     struct attribute *attr, char *buf)
1496 {
1497         const struct vmbus_chan_attribute *attribute
1498                 = container_of(attr, struct vmbus_chan_attribute, attr);
1499         struct vmbus_channel *chan
1500                 = container_of(kobj, struct vmbus_channel, kobj);
1501
1502         if (!attribute->show)
1503                 return -EIO;
1504
1505         return attribute->show(chan, buf);
1506 }
1507
1508 static const struct sysfs_ops vmbus_chan_sysfs_ops = {
1509         .show = vmbus_chan_attr_show,
1510 };
1511
1512 static ssize_t out_mask_show(struct vmbus_channel *channel, char *buf)
1513 {
1514         struct hv_ring_buffer_info *rbi = &channel->outbound;
1515         ssize_t ret;
1516
1517         mutex_lock(&rbi->ring_buffer_mutex);
1518         if (!rbi->ring_buffer) {
1519                 mutex_unlock(&rbi->ring_buffer_mutex);
1520                 return -EINVAL;
1521         }
1522
1523         ret = sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1524         mutex_unlock(&rbi->ring_buffer_mutex);
1525         return ret;
1526 }
1527 static VMBUS_CHAN_ATTR_RO(out_mask);
1528
1529 static ssize_t in_mask_show(struct vmbus_channel *channel, char *buf)
1530 {
1531         struct hv_ring_buffer_info *rbi = &channel->inbound;
1532         ssize_t ret;
1533
1534         mutex_lock(&rbi->ring_buffer_mutex);
1535         if (!rbi->ring_buffer) {
1536                 mutex_unlock(&rbi->ring_buffer_mutex);
1537                 return -EINVAL;
1538         }
1539
1540         ret = sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1541         mutex_unlock(&rbi->ring_buffer_mutex);
1542         return ret;
1543 }
1544 static VMBUS_CHAN_ATTR_RO(in_mask);
1545
1546 static ssize_t read_avail_show(struct vmbus_channel *channel, char *buf)
1547 {
1548         struct hv_ring_buffer_info *rbi = &channel->inbound;
1549         ssize_t ret;
1550
1551         mutex_lock(&rbi->ring_buffer_mutex);
1552         if (!rbi->ring_buffer) {
1553                 mutex_unlock(&rbi->ring_buffer_mutex);
1554                 return -EINVAL;
1555         }
1556
1557         ret = sprintf(buf, "%u\n", hv_get_bytes_to_read(rbi));
1558         mutex_unlock(&rbi->ring_buffer_mutex);
1559         return ret;
1560 }
1561 static VMBUS_CHAN_ATTR_RO(read_avail);
1562
1563 static ssize_t write_avail_show(struct vmbus_channel *channel, char *buf)
1564 {
1565         struct hv_ring_buffer_info *rbi = &channel->outbound;
1566         ssize_t ret;
1567
1568         mutex_lock(&rbi->ring_buffer_mutex);
1569         if (!rbi->ring_buffer) {
1570                 mutex_unlock(&rbi->ring_buffer_mutex);
1571                 return -EINVAL;
1572         }
1573
1574         ret = sprintf(buf, "%u\n", hv_get_bytes_to_write(rbi));
1575         mutex_unlock(&rbi->ring_buffer_mutex);
1576         return ret;
1577 }
1578 static VMBUS_CHAN_ATTR_RO(write_avail);
1579
1580 static ssize_t show_target_cpu(struct vmbus_channel *channel, char *buf)
1581 {
1582         return sprintf(buf, "%u\n", channel->target_cpu);
1583 }
1584 static VMBUS_CHAN_ATTR(cpu, S_IRUGO, show_target_cpu, NULL);
1585
1586 static ssize_t channel_pending_show(struct vmbus_channel *channel,
1587                                     char *buf)
1588 {
1589         return sprintf(buf, "%d\n",
1590                        channel_pending(channel,
1591                                        vmbus_connection.monitor_pages[1]));
1592 }
1593 static VMBUS_CHAN_ATTR(pending, S_IRUGO, channel_pending_show, NULL);
1594
1595 static ssize_t channel_latency_show(struct vmbus_channel *channel,
1596                                     char *buf)
1597 {
1598         return sprintf(buf, "%d\n",
1599                        channel_latency(channel,
1600                                        vmbus_connection.monitor_pages[1]));
1601 }
1602 static VMBUS_CHAN_ATTR(latency, S_IRUGO, channel_latency_show, NULL);
1603
1604 static ssize_t channel_interrupts_show(struct vmbus_channel *channel, char *buf)
1605 {
1606         return sprintf(buf, "%llu\n", channel->interrupts);
1607 }
1608 static VMBUS_CHAN_ATTR(interrupts, S_IRUGO, channel_interrupts_show, NULL);
1609
1610 static ssize_t channel_events_show(struct vmbus_channel *channel, char *buf)
1611 {
1612         return sprintf(buf, "%llu\n", channel->sig_events);
1613 }
1614 static VMBUS_CHAN_ATTR(events, S_IRUGO, channel_events_show, NULL);
1615
1616 static ssize_t channel_intr_in_full_show(struct vmbus_channel *channel,
1617                                          char *buf)
1618 {
1619         return sprintf(buf, "%llu\n",
1620                        (unsigned long long)channel->intr_in_full);
1621 }
1622 static VMBUS_CHAN_ATTR(intr_in_full, 0444, channel_intr_in_full_show, NULL);
1623
1624 static ssize_t channel_intr_out_empty_show(struct vmbus_channel *channel,
1625                                            char *buf)
1626 {
1627         return sprintf(buf, "%llu\n",
1628                        (unsigned long long)channel->intr_out_empty);
1629 }
1630 static VMBUS_CHAN_ATTR(intr_out_empty, 0444, channel_intr_out_empty_show, NULL);
1631
1632 static ssize_t channel_out_full_first_show(struct vmbus_channel *channel,
1633                                            char *buf)
1634 {
1635         return sprintf(buf, "%llu\n",
1636                        (unsigned long long)channel->out_full_first);
1637 }
1638 static VMBUS_CHAN_ATTR(out_full_first, 0444, channel_out_full_first_show, NULL);
1639
1640 static ssize_t channel_out_full_total_show(struct vmbus_channel *channel,
1641                                            char *buf)
1642 {
1643         return sprintf(buf, "%llu\n",
1644                        (unsigned long long)channel->out_full_total);
1645 }
1646 static VMBUS_CHAN_ATTR(out_full_total, 0444, channel_out_full_total_show, NULL);
1647
1648 static ssize_t subchannel_monitor_id_show(struct vmbus_channel *channel,
1649                                           char *buf)
1650 {
1651         return sprintf(buf, "%u\n", channel->offermsg.monitorid);
1652 }
1653 static VMBUS_CHAN_ATTR(monitor_id, S_IRUGO, subchannel_monitor_id_show, NULL);
1654
1655 static ssize_t subchannel_id_show(struct vmbus_channel *channel,
1656                                   char *buf)
1657 {
1658         return sprintf(buf, "%u\n",
1659                        channel->offermsg.offer.sub_channel_index);
1660 }
1661 static VMBUS_CHAN_ATTR_RO(subchannel_id);
1662
1663 static struct attribute *vmbus_chan_attrs[] = {
1664         &chan_attr_out_mask.attr,
1665         &chan_attr_in_mask.attr,
1666         &chan_attr_read_avail.attr,
1667         &chan_attr_write_avail.attr,
1668         &chan_attr_cpu.attr,
1669         &chan_attr_pending.attr,
1670         &chan_attr_latency.attr,
1671         &chan_attr_interrupts.attr,
1672         &chan_attr_events.attr,
1673         &chan_attr_intr_in_full.attr,
1674         &chan_attr_intr_out_empty.attr,
1675         &chan_attr_out_full_first.attr,
1676         &chan_attr_out_full_total.attr,
1677         &chan_attr_monitor_id.attr,
1678         &chan_attr_subchannel_id.attr,
1679         NULL
1680 };
1681
1682 /*
1683  * Channel-level attribute_group callback function. Returns the permission for
1684  * each attribute, and returns 0 if an attribute is not visible.
1685  */
1686 static umode_t vmbus_chan_attr_is_visible(struct kobject *kobj,
1687                                           struct attribute *attr, int idx)
1688 {
1689         const struct vmbus_channel *channel =
1690                 container_of(kobj, struct vmbus_channel, kobj);
1691
1692         /* Hide the monitor attributes if the monitor mechanism is not used. */
1693         if (!channel->offermsg.monitor_allocated &&
1694             (attr == &chan_attr_pending.attr ||
1695              attr == &chan_attr_latency.attr ||
1696              attr == &chan_attr_monitor_id.attr))
1697                 return 0;
1698
1699         return attr->mode;
1700 }
1701
1702 static struct attribute_group vmbus_chan_group = {
1703         .attrs = vmbus_chan_attrs,
1704         .is_visible = vmbus_chan_attr_is_visible
1705 };
1706
1707 static struct kobj_type vmbus_chan_ktype = {
1708         .sysfs_ops = &vmbus_chan_sysfs_ops,
1709         .release = vmbus_chan_release,
1710 };
1711
1712 /*
1713  * vmbus_add_channel_kobj - setup a sub-directory under device/channels
1714  */
1715 int vmbus_add_channel_kobj(struct hv_device *dev, struct vmbus_channel *channel)
1716 {
1717         const struct device *device = &dev->device;
1718         struct kobject *kobj = &channel->kobj;
1719         u32 relid = channel->offermsg.child_relid;
1720         int ret;
1721
1722         kobj->kset = dev->channels_kset;
1723         ret = kobject_init_and_add(kobj, &vmbus_chan_ktype, NULL,
1724                                    "%u", relid);
1725         if (ret)
1726                 return ret;
1727
1728         ret = sysfs_create_group(kobj, &vmbus_chan_group);
1729
1730         if (ret) {
1731                 /*
1732                  * The calling functions' error handling paths will cleanup the
1733                  * empty channel directory.
1734                  */
1735                 dev_err(device, "Unable to set up channel sysfs files\n");
1736                 return ret;
1737         }
1738
1739         kobject_uevent(kobj, KOBJ_ADD);
1740
1741         return 0;
1742 }
1743
1744 /*
1745  * vmbus_remove_channel_attr_group - remove the channel's attribute group
1746  */
1747 void vmbus_remove_channel_attr_group(struct vmbus_channel *channel)
1748 {
1749         sysfs_remove_group(&channel->kobj, &vmbus_chan_group);
1750 }
1751
1752 /*
1753  * vmbus_device_create - Creates and registers a new child device
1754  * on the vmbus.
1755  */
1756 struct hv_device *vmbus_device_create(const guid_t *type,
1757                                       const guid_t *instance,
1758                                       struct vmbus_channel *channel)
1759 {
1760         struct hv_device *child_device_obj;
1761
1762         child_device_obj = kzalloc(sizeof(struct hv_device), GFP_KERNEL);
1763         if (!child_device_obj) {
1764                 pr_err("Unable to allocate device object for child device\n");
1765                 return NULL;
1766         }
1767
1768         child_device_obj->channel = channel;
1769         guid_copy(&child_device_obj->dev_type, type);
1770         guid_copy(&child_device_obj->dev_instance, instance);
1771         child_device_obj->vendor_id = 0x1414; /* MSFT vendor ID */
1772
1773         return child_device_obj;
1774 }
1775
1776 /*
1777  * vmbus_device_register - Register the child device
1778  */
1779 int vmbus_device_register(struct hv_device *child_device_obj)
1780 {
1781         struct kobject *kobj = &child_device_obj->device.kobj;
1782         int ret;
1783
1784         dev_set_name(&child_device_obj->device, "%pUl",
1785                      child_device_obj->channel->offermsg.offer.if_instance.b);
1786
1787         child_device_obj->device.bus = &hv_bus;
1788         child_device_obj->device.parent = &hv_acpi_dev->dev;
1789         child_device_obj->device.release = vmbus_device_release;
1790
1791         /*
1792          * Register with the LDM. This will kick off the driver/device
1793          * binding...which will eventually call vmbus_match() and vmbus_probe()
1794          */
1795         ret = device_register(&child_device_obj->device);
1796         if (ret) {
1797                 pr_err("Unable to register child device\n");
1798                 return ret;
1799         }
1800
1801         child_device_obj->channels_kset = kset_create_and_add("channels",
1802                                                               NULL, kobj);
1803         if (!child_device_obj->channels_kset) {
1804                 ret = -ENOMEM;
1805                 goto err_dev_unregister;
1806         }
1807
1808         ret = vmbus_add_channel_kobj(child_device_obj,
1809                                      child_device_obj->channel);
1810         if (ret) {
1811                 pr_err("Unable to register primary channeln");
1812                 goto err_kset_unregister;
1813         }
1814         hv_debug_add_dev_dir(child_device_obj);
1815
1816         return 0;
1817
1818 err_kset_unregister:
1819         kset_unregister(child_device_obj->channels_kset);
1820
1821 err_dev_unregister:
1822         device_unregister(&child_device_obj->device);
1823         return ret;
1824 }
1825
1826 /*
1827  * vmbus_device_unregister - Remove the specified child device
1828  * from the vmbus.
1829  */
1830 void vmbus_device_unregister(struct hv_device *device_obj)
1831 {
1832         pr_debug("child device %s unregistered\n",
1833                 dev_name(&device_obj->device));
1834
1835         kset_unregister(device_obj->channels_kset);
1836
1837         /*
1838          * Kick off the process of unregistering the device.
1839          * This will call vmbus_remove() and eventually vmbus_device_release()
1840          */
1841         device_unregister(&device_obj->device);
1842 }
1843
1844
1845 /*
1846  * VMBUS is an acpi enumerated device. Get the information we
1847  * need from DSDT.
1848  */
1849 #define VTPM_BASE_ADDRESS 0xfed40000
1850 static acpi_status vmbus_walk_resources(struct acpi_resource *res, void *ctx)
1851 {
1852         resource_size_t start = 0;
1853         resource_size_t end = 0;
1854         struct resource *new_res;
1855         struct resource **old_res = &hyperv_mmio;
1856         struct resource **prev_res = NULL;
1857
1858         switch (res->type) {
1859
1860         /*
1861          * "Address" descriptors are for bus windows. Ignore
1862          * "memory" descriptors, which are for registers on
1863          * devices.
1864          */
1865         case ACPI_RESOURCE_TYPE_ADDRESS32:
1866                 start = res->data.address32.address.minimum;
1867                 end = res->data.address32.address.maximum;
1868                 break;
1869
1870         case ACPI_RESOURCE_TYPE_ADDRESS64:
1871                 start = res->data.address64.address.minimum;
1872                 end = res->data.address64.address.maximum;
1873                 break;
1874
1875         default:
1876                 /* Unused resource type */
1877                 return AE_OK;
1878
1879         }
1880         /*
1881          * Ignore ranges that are below 1MB, as they're not
1882          * necessary or useful here.
1883          */
1884         if (end < 0x100000)
1885                 return AE_OK;
1886
1887         new_res = kzalloc(sizeof(*new_res), GFP_ATOMIC);
1888         if (!new_res)
1889                 return AE_NO_MEMORY;
1890
1891         /* If this range overlaps the virtual TPM, truncate it. */
1892         if (end > VTPM_BASE_ADDRESS && start < VTPM_BASE_ADDRESS)
1893                 end = VTPM_BASE_ADDRESS;
1894
1895         new_res->name = "hyperv mmio";
1896         new_res->flags = IORESOURCE_MEM;
1897         new_res->start = start;
1898         new_res->end = end;
1899
1900         /*
1901          * If two ranges are adjacent, merge them.
1902          */
1903         do {
1904                 if (!*old_res) {
1905                         *old_res = new_res;
1906                         break;
1907                 }
1908
1909                 if (((*old_res)->end + 1) == new_res->start) {
1910                         (*old_res)->end = new_res->end;
1911                         kfree(new_res);
1912                         break;
1913                 }
1914
1915                 if ((*old_res)->start == new_res->end + 1) {
1916                         (*old_res)->start = new_res->start;
1917                         kfree(new_res);
1918                         break;
1919                 }
1920
1921                 if ((*old_res)->start > new_res->end) {
1922                         new_res->sibling = *old_res;
1923                         if (prev_res)
1924                                 (*prev_res)->sibling = new_res;
1925                         *old_res = new_res;
1926                         break;
1927                 }
1928
1929                 prev_res = old_res;
1930                 old_res = &(*old_res)->sibling;
1931
1932         } while (1);
1933
1934         return AE_OK;
1935 }
1936
1937 static int vmbus_acpi_remove(struct acpi_device *device)
1938 {
1939         struct resource *cur_res;
1940         struct resource *next_res;
1941
1942         if (hyperv_mmio) {
1943                 if (fb_mmio) {
1944                         __release_region(hyperv_mmio, fb_mmio->start,
1945                                          resource_size(fb_mmio));
1946                         fb_mmio = NULL;
1947                 }
1948
1949                 for (cur_res = hyperv_mmio; cur_res; cur_res = next_res) {
1950                         next_res = cur_res->sibling;
1951                         kfree(cur_res);
1952                 }
1953         }
1954
1955         return 0;
1956 }
1957
1958 static void vmbus_reserve_fb(void)
1959 {
1960         int size;
1961         /*
1962          * Make a claim for the frame buffer in the resource tree under the
1963          * first node, which will be the one below 4GB.  The length seems to
1964          * be underreported, particularly in a Generation 1 VM.  So start out
1965          * reserving a larger area and make it smaller until it succeeds.
1966          */
1967
1968         if (screen_info.lfb_base) {
1969                 if (efi_enabled(EFI_BOOT))
1970                         size = max_t(__u32, screen_info.lfb_size, 0x800000);
1971                 else
1972                         size = max_t(__u32, screen_info.lfb_size, 0x4000000);
1973
1974                 for (; !fb_mmio && (size >= 0x100000); size >>= 1) {
1975                         fb_mmio = __request_region(hyperv_mmio,
1976                                                    screen_info.lfb_base, size,
1977                                                    fb_mmio_name, 0);
1978                 }
1979         }
1980 }
1981
1982 /**
1983  * vmbus_allocate_mmio() - Pick a memory-mapped I/O range.
1984  * @new:                If successful, supplied a pointer to the
1985  *                      allocated MMIO space.
1986  * @device_obj:         Identifies the caller
1987  * @min:                Minimum guest physical address of the
1988  *                      allocation
1989  * @max:                Maximum guest physical address
1990  * @size:               Size of the range to be allocated
1991  * @align:              Alignment of the range to be allocated
1992  * @fb_overlap_ok:      Whether this allocation can be allowed
1993  *                      to overlap the video frame buffer.
1994  *
1995  * This function walks the resources granted to VMBus by the
1996  * _CRS object in the ACPI namespace underneath the parent
1997  * "bridge" whether that's a root PCI bus in the Generation 1
1998  * case or a Module Device in the Generation 2 case.  It then
1999  * attempts to allocate from the global MMIO pool in a way that
2000  * matches the constraints supplied in these parameters and by
2001  * that _CRS.
2002  *
2003  * Return: 0 on success, -errno on failure
2004  */
2005 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
2006                         resource_size_t min, resource_size_t max,
2007                         resource_size_t size, resource_size_t align,
2008                         bool fb_overlap_ok)
2009 {
2010         struct resource *iter, *shadow;
2011         resource_size_t range_min, range_max, start;
2012         const char *dev_n = dev_name(&device_obj->device);
2013         int retval;
2014
2015         retval = -ENXIO;
2016         mutex_lock(&hyperv_mmio_lock);
2017
2018         /*
2019          * If overlaps with frame buffers are allowed, then first attempt to
2020          * make the allocation from within the reserved region.  Because it
2021          * is already reserved, no shadow allocation is necessary.
2022          */
2023         if (fb_overlap_ok && fb_mmio && !(min > fb_mmio->end) &&
2024             !(max < fb_mmio->start)) {
2025
2026                 range_min = fb_mmio->start;
2027                 range_max = fb_mmio->end;
2028                 start = (range_min + align - 1) & ~(align - 1);
2029                 for (; start + size - 1 <= range_max; start += align) {
2030                         *new = request_mem_region_exclusive(start, size, dev_n);
2031                         if (*new) {
2032                                 retval = 0;
2033                                 goto exit;
2034                         }
2035                 }
2036         }
2037
2038         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
2039                 if ((iter->start >= max) || (iter->end <= min))
2040                         continue;
2041
2042                 range_min = iter->start;
2043                 range_max = iter->end;
2044                 start = (range_min + align - 1) & ~(align - 1);
2045                 for (; start + size - 1 <= range_max; start += align) {
2046                         shadow = __request_region(iter, start, size, NULL,
2047                                                   IORESOURCE_BUSY);
2048                         if (!shadow)
2049                                 continue;
2050
2051                         *new = request_mem_region_exclusive(start, size, dev_n);
2052                         if (*new) {
2053                                 shadow->name = (char *)*new;
2054                                 retval = 0;
2055                                 goto exit;
2056                         }
2057
2058                         __release_region(iter, start, size);
2059                 }
2060         }
2061
2062 exit:
2063         mutex_unlock(&hyperv_mmio_lock);
2064         return retval;
2065 }
2066 EXPORT_SYMBOL_GPL(vmbus_allocate_mmio);
2067
2068 /**
2069  * vmbus_free_mmio() - Free a memory-mapped I/O range.
2070  * @start:              Base address of region to release.
2071  * @size:               Size of the range to be allocated
2072  *
2073  * This function releases anything requested by
2074  * vmbus_mmio_allocate().
2075  */
2076 void vmbus_free_mmio(resource_size_t start, resource_size_t size)
2077 {
2078         struct resource *iter;
2079
2080         mutex_lock(&hyperv_mmio_lock);
2081         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
2082                 if ((iter->start >= start + size) || (iter->end <= start))
2083                         continue;
2084
2085                 __release_region(iter, start, size);
2086         }
2087         release_mem_region(start, size);
2088         mutex_unlock(&hyperv_mmio_lock);
2089
2090 }
2091 EXPORT_SYMBOL_GPL(vmbus_free_mmio);
2092
2093 static int vmbus_acpi_add(struct acpi_device *device)
2094 {
2095         acpi_status result;
2096         int ret_val = -ENODEV;
2097         struct acpi_device *ancestor;
2098
2099         hv_acpi_dev = device;
2100
2101         result = acpi_walk_resources(device->handle, METHOD_NAME__CRS,
2102                                         vmbus_walk_resources, NULL);
2103
2104         if (ACPI_FAILURE(result))
2105                 goto acpi_walk_err;
2106         /*
2107          * Some ancestor of the vmbus acpi device (Gen1 or Gen2
2108          * firmware) is the VMOD that has the mmio ranges. Get that.
2109          */
2110         for (ancestor = device->parent; ancestor; ancestor = ancestor->parent) {
2111                 result = acpi_walk_resources(ancestor->handle, METHOD_NAME__CRS,
2112                                              vmbus_walk_resources, NULL);
2113
2114                 if (ACPI_FAILURE(result))
2115                         continue;
2116                 if (hyperv_mmio) {
2117                         vmbus_reserve_fb();
2118                         break;
2119                 }
2120         }
2121         ret_val = 0;
2122
2123 acpi_walk_err:
2124         complete(&probe_event);
2125         if (ret_val)
2126                 vmbus_acpi_remove(device);
2127         return ret_val;
2128 }
2129
2130 #ifdef CONFIG_PM_SLEEP
2131 static int vmbus_bus_suspend(struct device *dev)
2132 {
2133         struct vmbus_channel *channel, *sc;
2134         unsigned long flags;
2135
2136         while (atomic_read(&vmbus_connection.offer_in_progress) != 0) {
2137                 /*
2138                  * We wait here until the completion of any channel
2139                  * offers that are currently in progress.
2140                  */
2141                 msleep(1);
2142         }
2143
2144         mutex_lock(&vmbus_connection.channel_mutex);
2145         list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
2146                 if (!is_hvsock_channel(channel))
2147                         continue;
2148
2149                 vmbus_force_channel_rescinded(channel);
2150         }
2151         mutex_unlock(&vmbus_connection.channel_mutex);
2152
2153         /*
2154          * Wait until all the sub-channels and hv_sock channels have been
2155          * cleaned up. Sub-channels should be destroyed upon suspend, otherwise
2156          * they would conflict with the new sub-channels that will be created
2157          * in the resume path. hv_sock channels should also be destroyed, but
2158          * a hv_sock channel of an established hv_sock connection can not be
2159          * really destroyed since it may still be referenced by the userspace
2160          * application, so we just force the hv_sock channel to be rescinded
2161          * by vmbus_force_channel_rescinded(), and the userspace application
2162          * will thoroughly destroy the channel after hibernation.
2163          *
2164          * Note: the counter nr_chan_close_on_suspend may never go above 0 if
2165          * the VM has no sub-channel and hv_sock channel, e.g. a 1-vCPU VM.
2166          */
2167         if (atomic_read(&vmbus_connection.nr_chan_close_on_suspend) > 0)
2168                 wait_for_completion(&vmbus_connection.ready_for_suspend_event);
2169
2170         WARN_ON(atomic_read(&vmbus_connection.nr_chan_fixup_on_resume) != 0);
2171
2172         mutex_lock(&vmbus_connection.channel_mutex);
2173
2174         list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
2175                 /*
2176                  * Invalidate the field. Upon resume, vmbus_onoffer() will fix
2177                  * up the field, and the other fields (if necessary).
2178                  */
2179                 channel->offermsg.child_relid = INVALID_RELID;
2180
2181                 if (is_hvsock_channel(channel)) {
2182                         if (!channel->rescind) {
2183                                 pr_err("hv_sock channel not rescinded!\n");
2184                                 WARN_ON_ONCE(1);
2185                         }
2186                         continue;
2187                 }
2188
2189                 spin_lock_irqsave(&channel->lock, flags);
2190                 list_for_each_entry(sc, &channel->sc_list, sc_list) {
2191                         pr_err("Sub-channel not deleted!\n");
2192                         WARN_ON_ONCE(1);
2193                 }
2194                 spin_unlock_irqrestore(&channel->lock, flags);
2195
2196                 atomic_inc(&vmbus_connection.nr_chan_fixup_on_resume);
2197         }
2198
2199         mutex_unlock(&vmbus_connection.channel_mutex);
2200
2201         vmbus_initiate_unload(false);
2202
2203         vmbus_connection.conn_state = DISCONNECTED;
2204
2205         /* Reset the event for the next resume. */
2206         reinit_completion(&vmbus_connection.ready_for_resume_event);
2207
2208         return 0;
2209 }
2210
2211 static int vmbus_bus_resume(struct device *dev)
2212 {
2213         struct vmbus_channel_msginfo *msginfo;
2214         size_t msgsize;
2215         int ret;
2216
2217         /*
2218          * We only use the 'vmbus_proto_version', which was in use before
2219          * hibernation, to re-negotiate with the host.
2220          */
2221         if (!vmbus_proto_version) {
2222                 pr_err("Invalid proto version = 0x%x\n", vmbus_proto_version);
2223                 return -EINVAL;
2224         }
2225
2226         msgsize = sizeof(*msginfo) +
2227                   sizeof(struct vmbus_channel_initiate_contact);
2228
2229         msginfo = kzalloc(msgsize, GFP_KERNEL);
2230
2231         if (msginfo == NULL)
2232                 return -ENOMEM;
2233
2234         ret = vmbus_negotiate_version(msginfo, vmbus_proto_version);
2235
2236         kfree(msginfo);
2237
2238         if (ret != 0)
2239                 return ret;
2240
2241         WARN_ON(atomic_read(&vmbus_connection.nr_chan_fixup_on_resume) == 0);
2242
2243         vmbus_request_offers();
2244
2245         wait_for_completion(&vmbus_connection.ready_for_resume_event);
2246
2247         /* Reset the event for the next suspend. */
2248         reinit_completion(&vmbus_connection.ready_for_suspend_event);
2249
2250         return 0;
2251 }
2252 #endif /* CONFIG_PM_SLEEP */
2253
2254 static const struct acpi_device_id vmbus_acpi_device_ids[] = {
2255         {"VMBUS", 0},
2256         {"VMBus", 0},
2257         {"", 0},
2258 };
2259 MODULE_DEVICE_TABLE(acpi, vmbus_acpi_device_ids);
2260
2261 /*
2262  * Note: we must use SET_NOIRQ_SYSTEM_SLEEP_PM_OPS rather than
2263  * SET_SYSTEM_SLEEP_PM_OPS, otherwise NIC SR-IOV can not work, because the
2264  * "pci_dev_pm_ops" uses the "noirq" callbacks: in the resume path, the
2265  * pci "noirq" restore callback runs before "non-noirq" callbacks (see
2266  * resume_target_kernel() -> dpm_resume_start(), and hibernation_restore() ->
2267  * dpm_resume_end()). This means vmbus_bus_resume() and the pci-hyperv's
2268  * resume callback must also run via the "noirq" callbacks.
2269  */
2270 static const struct dev_pm_ops vmbus_bus_pm = {
2271         SET_NOIRQ_SYSTEM_SLEEP_PM_OPS(vmbus_bus_suspend, vmbus_bus_resume)
2272 };
2273
2274 static struct acpi_driver vmbus_acpi_driver = {
2275         .name = "vmbus",
2276         .ids = vmbus_acpi_device_ids,
2277         .ops = {
2278                 .add = vmbus_acpi_add,
2279                 .remove = vmbus_acpi_remove,
2280         },
2281         .drv.pm = &vmbus_bus_pm,
2282 };
2283
2284 static void hv_kexec_handler(void)
2285 {
2286         hv_stimer_global_cleanup();
2287         vmbus_initiate_unload(false);
2288         vmbus_connection.conn_state = DISCONNECTED;
2289         /* Make sure conn_state is set as hv_synic_cleanup checks for it */
2290         mb();
2291         cpuhp_remove_state(hyperv_cpuhp_online);
2292         hyperv_cleanup();
2293 };
2294
2295 static void hv_crash_handler(struct pt_regs *regs)
2296 {
2297         int cpu;
2298
2299         vmbus_initiate_unload(true);
2300         /*
2301          * In crash handler we can't schedule synic cleanup for all CPUs,
2302          * doing the cleanup for current CPU only. This should be sufficient
2303          * for kdump.
2304          */
2305         vmbus_connection.conn_state = DISCONNECTED;
2306         cpu = smp_processor_id();
2307         hv_stimer_cleanup(cpu);
2308         hv_synic_disable_regs(cpu);
2309         hyperv_cleanup();
2310 };
2311
2312 static int hv_synic_suspend(void)
2313 {
2314         /*
2315          * When we reach here, all the non-boot CPUs have been offlined.
2316          * If we're in a legacy configuration where stimer Direct Mode is
2317          * not enabled, the stimers on the non-boot CPUs have been unbound
2318          * in hv_synic_cleanup() -> hv_stimer_legacy_cleanup() ->
2319          * hv_stimer_cleanup() -> clockevents_unbind_device().
2320          *
2321          * hv_synic_suspend() only runs on CPU0 with interrupts disabled.
2322          * Here we do not call hv_stimer_legacy_cleanup() on CPU0 because:
2323          * 1) it's unnecessary as interrupts remain disabled between
2324          * syscore_suspend() and syscore_resume(): see create_image() and
2325          * 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()
2329          * 3) a warning would be triggered if we call
2330          * clockevents_unbind_device(), which may sleep, in an
2331          * interrupts-disabled context.
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         hv_debug_init();
2379
2380         ret = vmbus_bus_init();
2381         if (ret)
2382                 goto cleanup;
2383
2384         hv_setup_kexec_handler(hv_kexec_handler);
2385         hv_setup_crash_handler(hv_crash_handler);
2386
2387         register_syscore_ops(&hv_synic_syscore_ops);
2388
2389         return 0;
2390
2391 cleanup:
2392         acpi_bus_unregister_driver(&vmbus_acpi_driver);
2393         hv_acpi_dev = NULL;
2394         return ret;
2395 }
2396
2397 static void __exit vmbus_exit(void)
2398 {
2399         int cpu;
2400
2401         unregister_syscore_ops(&hv_synic_syscore_ops);
2402
2403         hv_remove_kexec_handler();
2404         hv_remove_crash_handler();
2405         vmbus_connection.conn_state = DISCONNECTED;
2406         hv_stimer_global_cleanup();
2407         vmbus_disconnect();
2408         hv_remove_vmbus_irq();
2409         for_each_online_cpu(cpu) {
2410                 struct hv_per_cpu_context *hv_cpu
2411                         = per_cpu_ptr(hv_context.cpu_context, cpu);
2412
2413                 tasklet_kill(&hv_cpu->msg_dpc);
2414         }
2415         hv_debug_rm_all_dir();
2416
2417         vmbus_free_channels();
2418
2419         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
2420                 kmsg_dump_unregister(&hv_kmsg_dumper);
2421                 unregister_die_notifier(&hyperv_die_block);
2422                 atomic_notifier_chain_unregister(&panic_notifier_list,
2423                                                  &hyperv_panic_block);
2424         }
2425
2426         free_page((unsigned long)hv_panic_page);
2427         unregister_sysctl_table(hv_ctl_table_hdr);
2428         hv_ctl_table_hdr = NULL;
2429         bus_unregister(&hv_bus);
2430
2431         cpuhp_remove_state(hyperv_cpuhp_online);
2432         hv_synic_free();
2433         acpi_bus_unregister_driver(&vmbus_acpi_driver);
2434 }
2435
2436
2437 MODULE_LICENSE("GPL");
2438 MODULE_DESCRIPTION("Microsoft Hyper-V VMBus Driver");
2439
2440 subsys_initcall(hv_acpi_init);
2441 module_exit(vmbus_exit);