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/kernel.h>
  35 #include <linux/syscore_ops.h>
  36 #include <clocksource/hyperv_timer.h>
  37 #include "hyperv_vmbus.h"
  38
  39 struct vmbus_dynid {
  40         struct list_head node;
  41         struct hv_vmbus_device_id id;
  42 };
  43
  44 static struct acpi_device  *hv_acpi_dev;
  45
  46 static struct completion probe_event;
  47
  48 static int hyperv_cpuhp_online;
  49
  50 static void *hv_panic_page;
  51
  52 /*
  53  * Boolean to control whether to report panic messages over Hyper-V.
  54  *
  55  * It can be set via /proc/sys/kernel/hyperv/record_panic_msg
  56  */
  57 static int sysctl_record_panic_msg = 1;
  58
  59 static int hyperv_report_reg(void)
  60 {
  61         return !sysctl_record_panic_msg || !hv_panic_page;
  62 }
  63
  64 static int hyperv_panic_event(struct notifier_block *nb, unsigned long val,
  65                               void *args)
  66 {
  67         struct pt_regs *regs;
  68
  69         vmbus_initiate_unload(true);
  70
  71         /*
  72          * Hyper-V should be notified only once about a panic.  If we will be
  73          * doing hyperv_report_panic_msg() later with kmsg data, don't do
  74          * the notification here.
  75          */
  76         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE
  77             && hyperv_report_reg()) {
  78                 regs = current_pt_regs();
  79                 hyperv_report_panic(regs, val, false);
  80         }
  81         return NOTIFY_DONE;
  82 }
  83
  84 static int hyperv_die_event(struct notifier_block *nb, unsigned long val,
  85                             void *args)
  86 {
  87         struct die_args *die = (struct die_args *)args;
  88         struct pt_regs *regs = die->regs;
  89
  90         /*
  91          * Hyper-V should be notified only once about a panic.  If we will be
  92          * doing hyperv_report_panic_msg() later with kmsg data, don't do
  93          * the notification here.
  94          */
  95         if (hyperv_report_reg())
  96                 hyperv_report_panic(regs, val, true);
  97         return NOTIFY_DONE;
  98 }
  99
 100 static struct notifier_block hyperv_die_block = {
 101         .notifier_call = hyperv_die_event,
 102 };
 103 static struct notifier_block hyperv_panic_block = {
 104         .notifier_call = hyperv_panic_event,
 105 };
 106
 107 static const char *fb_mmio_name = "fb_range";
 108 static struct resource *fb_mmio;
 109 static struct resource *hyperv_mmio;
 110 static DEFINE_SEMAPHORE(hyperv_mmio_lock);
 111
 112 static int vmbus_exists(void)
 113 {
 114         if (hv_acpi_dev == NULL)
 115                 return -ENODEV;
 116
 117         return 0;
 118 }
 119
 120 #define VMBUS_ALIAS_LEN ((sizeof((struct hv_vmbus_device_id *)0)->guid) * 2)
 121 static void print_alias_name(struct hv_device *hv_dev, char *alias_name)
 122 {
 123         int i;
 124         for (i = 0; i < VMBUS_ALIAS_LEN; i += 2)
 125                 sprintf(&alias_name[i], "%02x", hv_dev->dev_type.b[i/2]);
 126 }
 127
 128 static u8 channel_monitor_group(const struct vmbus_channel *channel)
 129 {
 130         return (u8)channel->offermsg.monitorid / 32;
 131 }
 132
 133 static u8 channel_monitor_offset(const struct vmbus_channel *channel)
 134 {
 135         return (u8)channel->offermsg.monitorid % 32;
 136 }
 137
 138 static u32 channel_pending(const struct vmbus_channel *channel,
 139                            const struct hv_monitor_page *monitor_page)
 140 {
 141         u8 monitor_group = channel_monitor_group(channel);
 142
 143         return monitor_page->trigger_group[monitor_group].pending;
 144 }
 145
 146 static u32 channel_latency(const struct vmbus_channel *channel,
 147                            const struct hv_monitor_page *monitor_page)
 148 {
 149         u8 monitor_group = channel_monitor_group(channel);
 150         u8 monitor_offset = channel_monitor_offset(channel);
 151
 152         return monitor_page->latency[monitor_group][monitor_offset];
 153 }
 154
 155 static u32 channel_conn_id(struct vmbus_channel *channel,
 156                            struct hv_monitor_page *monitor_page)
 157 {
 158         u8 monitor_group = channel_monitor_group(channel);
 159         u8 monitor_offset = channel_monitor_offset(channel);
 160         return monitor_page->parameter[monitor_group][monitor_offset].connectionid.u.id;
 161 }
 162
 163 static ssize_t id_show(struct device *dev, struct device_attribute *dev_attr,
 164                        char *buf)
 165 {
 166         struct hv_device *hv_dev = device_to_hv_device(dev);
 167
 168         if (!hv_dev->channel)
 169                 return -ENODEV;
 170         return sprintf(buf, "%d\n", hv_dev->channel->offermsg.child_relid);
 171 }
 172 static DEVICE_ATTR_RO(id);
 173
 174 static ssize_t state_show(struct device *dev, struct device_attribute *dev_attr,
 175                           char *buf)
 176 {
 177         struct hv_device *hv_dev = device_to_hv_device(dev);
 178
 179         if (!hv_dev->channel)
 180                 return -ENODEV;
 181         return sprintf(buf, "%d\n", hv_dev->channel->state);
 182 }
 183 static DEVICE_ATTR_RO(state);
 184
 185 static ssize_t monitor_id_show(struct device *dev,
 186                                struct device_attribute *dev_attr, char *buf)
 187 {
 188         struct hv_device *hv_dev = device_to_hv_device(dev);
 189
 190         if (!hv_dev->channel)
 191                 return -ENODEV;
 192         return sprintf(buf, "%d\n", hv_dev->channel->offermsg.monitorid);
 193 }
 194 static DEVICE_ATTR_RO(monitor_id);
 195
 196 static ssize_t class_id_show(struct device *dev,
 197                                struct device_attribute *dev_attr, char *buf)
 198 {
 199         struct hv_device *hv_dev = device_to_hv_device(dev);
 200
 201         if (!hv_dev->channel)
 202                 return -ENODEV;
 203         return sprintf(buf, "{%pUl}\n",
 204                        hv_dev->channel->offermsg.offer.if_type.b);
 205 }
 206 static DEVICE_ATTR_RO(class_id);
 207
 208 static ssize_t device_id_show(struct device *dev,
 209                               struct device_attribute *dev_attr, char *buf)
 210 {
 211         struct hv_device *hv_dev = device_to_hv_device(dev);
 212
 213         if (!hv_dev->channel)
 214                 return -ENODEV;
 215         return sprintf(buf, "{%pUl}\n",
 216                        hv_dev->channel->offermsg.offer.if_instance.b);
 217 }
 218 static DEVICE_ATTR_RO(device_id);
 219
 220 static ssize_t modalias_show(struct device *dev,
 221                              struct device_attribute *dev_attr, char *buf)
 222 {
 223         struct hv_device *hv_dev = device_to_hv_device(dev);
 224         char alias_name[VMBUS_ALIAS_LEN + 1];
 225
 226         print_alias_name(hv_dev, alias_name);
 227         return sprintf(buf, "vmbus:%s\n", alias_name);
 228 }
 229 static DEVICE_ATTR_RO(modalias);
 230
 231 #ifdef CONFIG_NUMA
 232 static ssize_t numa_node_show(struct device *dev,
 233                               struct device_attribute *attr, char *buf)
 234 {
 235         struct hv_device *hv_dev = device_to_hv_device(dev);
 236
 237         if (!hv_dev->channel)
 238                 return -ENODEV;
 239
 240         return sprintf(buf, "%d\n", hv_dev->channel->numa_node);
 241 }
 242 static DEVICE_ATTR_RO(numa_node);
 243 #endif
 244
 245 static ssize_t server_monitor_pending_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_pending(hv_dev->channel,
 255                                        vmbus_connection.monitor_pages[0]));
 256 }
 257 static DEVICE_ATTR_RO(server_monitor_pending);
 258
 259 static ssize_t client_monitor_pending_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_pending(hv_dev->channel,
 269                                        vmbus_connection.monitor_pages[1]));
 270 }
 271 static DEVICE_ATTR_RO(client_monitor_pending);
 272
 273 static ssize_t server_monitor_latency_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_latency(hv_dev->channel,
 283                                        vmbus_connection.monitor_pages[0]));
 284 }
 285 static DEVICE_ATTR_RO(server_monitor_latency);
 286
 287 static ssize_t client_monitor_latency_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_latency(hv_dev->channel,
 297                                        vmbus_connection.monitor_pages[1]));
 298 }
 299 static DEVICE_ATTR_RO(client_monitor_latency);
 300
 301 static ssize_t server_monitor_conn_id_show(struct device *dev,
 302                                            struct device_attribute *dev_attr,
 303                                            char *buf)
 304 {
 305         struct hv_device *hv_dev = device_to_hv_device(dev);
 306
 307         if (!hv_dev->channel)
 308                 return -ENODEV;
 309         return sprintf(buf, "%d\n",
 310                        channel_conn_id(hv_dev->channel,
 311                                        vmbus_connection.monitor_pages[0]));
 312 }
 313 static DEVICE_ATTR_RO(server_monitor_conn_id);
 314
 315 static ssize_t client_monitor_conn_id_show(struct device *dev,
 316                                            struct device_attribute *dev_attr,
 317                                            char *buf)
 318 {
 319         struct hv_device *hv_dev = device_to_hv_device(dev);
 320
 321         if (!hv_dev->channel)
 322                 return -ENODEV;
 323         return sprintf(buf, "%d\n",
 324                        channel_conn_id(hv_dev->channel,
 325                                        vmbus_connection.monitor_pages[1]));
 326 }
 327 static DEVICE_ATTR_RO(client_monitor_conn_id);
 328
 329 static ssize_t out_intr_mask_show(struct device *dev,
 330                                   struct device_attribute *dev_attr, char *buf)
 331 {
 332         struct hv_device *hv_dev = device_to_hv_device(dev);
 333         struct hv_ring_buffer_debug_info outbound;
 334         int ret;
 335
 336         if (!hv_dev->channel)
 337                 return -ENODEV;
 338
 339         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 340                                           &outbound);
 341         if (ret < 0)
 342                 return ret;
 343
 344         return sprintf(buf, "%d\n", outbound.current_interrupt_mask);
 345 }
 346 static DEVICE_ATTR_RO(out_intr_mask);
 347
 348 static ssize_t out_read_index_show(struct device *dev,
 349                                    struct device_attribute *dev_attr, char *buf)
 350 {
 351         struct hv_device *hv_dev = device_to_hv_device(dev);
 352         struct hv_ring_buffer_debug_info outbound;
 353         int ret;
 354
 355         if (!hv_dev->channel)
 356                 return -ENODEV;
 357
 358         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 359                                           &outbound);
 360         if (ret < 0)
 361                 return ret;
 362         return sprintf(buf, "%d\n", outbound.current_read_index);
 363 }
 364 static DEVICE_ATTR_RO(out_read_index);
 365
 366 static ssize_t out_write_index_show(struct device *dev,
 367                                     struct device_attribute *dev_attr,
 368                                     char *buf)
 369 {
 370         struct hv_device *hv_dev = device_to_hv_device(dev);
 371         struct hv_ring_buffer_debug_info outbound;
 372         int ret;
 373
 374         if (!hv_dev->channel)
 375                 return -ENODEV;
 376
 377         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 378                                           &outbound);
 379         if (ret < 0)
 380                 return ret;
 381         return sprintf(buf, "%d\n", outbound.current_write_index);
 382 }
 383 static DEVICE_ATTR_RO(out_write_index);
 384
 385 static ssize_t out_read_bytes_avail_show(struct device *dev,
 386                                          struct device_attribute *dev_attr,
 387                                          char *buf)
 388 {
 389         struct hv_device *hv_dev = device_to_hv_device(dev);
 390         struct hv_ring_buffer_debug_info outbound;
 391         int ret;
 392
 393         if (!hv_dev->channel)
 394                 return -ENODEV;
 395
 396         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 397                                           &outbound);
 398         if (ret < 0)
 399                 return ret;
 400         return sprintf(buf, "%d\n", outbound.bytes_avail_toread);
 401 }
 402 static DEVICE_ATTR_RO(out_read_bytes_avail);
 403
 404 static ssize_t out_write_bytes_avail_show(struct device *dev,
 405                                           struct device_attribute *dev_attr,
 406                                           char *buf)
 407 {
 408         struct hv_device *hv_dev = device_to_hv_device(dev);
 409         struct hv_ring_buffer_debug_info outbound;
 410         int ret;
 411
 412         if (!hv_dev->channel)
 413                 return -ENODEV;
 414
 415         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 416                                           &outbound);
 417         if (ret < 0)
 418                 return ret;
 419         return sprintf(buf, "%d\n", outbound.bytes_avail_towrite);
 420 }
 421 static DEVICE_ATTR_RO(out_write_bytes_avail);
 422
 423 static ssize_t in_intr_mask_show(struct device *dev,
 424                                  struct device_attribute *dev_attr, char *buf)
 425 {
 426         struct hv_device *hv_dev = device_to_hv_device(dev);
 427         struct hv_ring_buffer_debug_info inbound;
 428         int ret;
 429
 430         if (!hv_dev->channel)
 431                 return -ENODEV;
 432
 433         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 434         if (ret < 0)
 435                 return ret;
 436
 437         return sprintf(buf, "%d\n", inbound.current_interrupt_mask);
 438 }
 439 static DEVICE_ATTR_RO(in_intr_mask);
 440
 441 static ssize_t in_read_index_show(struct device *dev,
 442                                   struct device_attribute *dev_attr, char *buf)
 443 {
 444         struct hv_device *hv_dev = device_to_hv_device(dev);
 445         struct hv_ring_buffer_debug_info inbound;
 446         int ret;
 447
 448         if (!hv_dev->channel)
 449                 return -ENODEV;
 450
 451         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 452         if (ret < 0)
 453                 return ret;
 454
 455         return sprintf(buf, "%d\n", inbound.current_read_index);
 456 }
 457 static DEVICE_ATTR_RO(in_read_index);
 458
 459 static ssize_t in_write_index_show(struct device *dev,
 460                                    struct device_attribute *dev_attr, char *buf)
 461 {
 462         struct hv_device *hv_dev = device_to_hv_device(dev);
 463         struct hv_ring_buffer_debug_info inbound;
 464         int ret;
 465
 466         if (!hv_dev->channel)
 467                 return -ENODEV;
 468
 469         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 470         if (ret < 0)
 471                 return ret;
 472
 473         return sprintf(buf, "%d\n", inbound.current_write_index);
 474 }
 475 static DEVICE_ATTR_RO(in_write_index);
 476
 477 static ssize_t in_read_bytes_avail_show(struct device *dev,
 478                                         struct device_attribute *dev_attr,
 479                                         char *buf)
 480 {
 481         struct hv_device *hv_dev = device_to_hv_device(dev);
 482         struct hv_ring_buffer_debug_info inbound;
 483         int ret;
 484
 485         if (!hv_dev->channel)
 486                 return -ENODEV;
 487
 488         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 489         if (ret < 0)
 490                 return ret;
 491
 492         return sprintf(buf, "%d\n", inbound.bytes_avail_toread);
 493 }
 494 static DEVICE_ATTR_RO(in_read_bytes_avail);
 495
 496 static ssize_t in_write_bytes_avail_show(struct device *dev,
 497                                          struct device_attribute *dev_attr,
 498                                          char *buf)
 499 {
 500         struct hv_device *hv_dev = device_to_hv_device(dev);
 501         struct hv_ring_buffer_debug_info inbound;
 502         int ret;
 503
 504         if (!hv_dev->channel)
 505                 return -ENODEV;
 506
 507         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 508         if (ret < 0)
 509                 return ret;
 510
 511         return sprintf(buf, "%d\n", inbound.bytes_avail_towrite);
 512 }
 513 static DEVICE_ATTR_RO(in_write_bytes_avail);
 514
 515 static ssize_t channel_vp_mapping_show(struct device *dev,
 516                                        struct device_attribute *dev_attr,
 517                                        char *buf)
 518 {
 519         struct hv_device *hv_dev = device_to_hv_device(dev);
 520         struct vmbus_channel *channel = hv_dev->channel, *cur_sc;
 521         unsigned long flags;
 522         int buf_size = PAGE_SIZE, n_written, tot_written;
 523         struct list_head *cur;
 524
 525         if (!channel)
 526                 return -ENODEV;
 527
 528         tot_written = snprintf(buf, buf_size, "%u:%u\n",
 529                 channel->offermsg.child_relid, channel->target_cpu);
 530
 531         spin_lock_irqsave(&channel->lock, flags);
 532
 533         list_for_each(cur, &channel->sc_list) {
 534                 if (tot_written >= buf_size - 1)
 535                         break;
 536
 537                 cur_sc = list_entry(cur, struct vmbus_channel, sc_list);
 538                 n_written = scnprintf(buf + tot_written,
 539                                      buf_size - tot_written,
 540                                      "%u:%u\n",
 541                                      cur_sc->offermsg.child_relid,
 542                                      cur_sc->target_cpu);
 543                 tot_written += n_written;
 544         }
 545
 546         spin_unlock_irqrestore(&channel->lock, flags);
 547
 548         return tot_written;
 549 }
 550 static DEVICE_ATTR_RO(channel_vp_mapping);
 551
 552 static ssize_t vendor_show(struct device *dev,
 553                            struct device_attribute *dev_attr,
 554                            char *buf)
 555 {
 556         struct hv_device *hv_dev = device_to_hv_device(dev);
 557         return sprintf(buf, "0x%x\n", hv_dev->vendor_id);
 558 }
 559 static DEVICE_ATTR_RO(vendor);
 560
 561 static ssize_t device_show(struct device *dev,
 562                            struct device_attribute *dev_attr,
 563                            char *buf)
 564 {
 565         struct hv_device *hv_dev = device_to_hv_device(dev);
 566         return sprintf(buf, "0x%x\n", hv_dev->device_id);
 567 }
 568 static DEVICE_ATTR_RO(device);
 569
 570 static ssize_t driver_override_store(struct device *dev,
 571                                      struct device_attribute *attr,
 572                                      const char *buf, size_t count)
 573 {
 574         struct hv_device *hv_dev = device_to_hv_device(dev);
 575         char *driver_override, *old, *cp;
 576
 577         /* We need to keep extra room for a newline */
 578         if (count >= (PAGE_SIZE - 1))
 579                 return -EINVAL;
 580
 581         driver_override = kstrndup(buf, count, GFP_KERNEL);
 582         if (!driver_override)
 583                 return -ENOMEM;
 584
 585         cp = strchr(driver_override, '\n');
 586         if (cp)
 587                 *cp = '\0';
 588
 589         device_lock(dev);
 590         old = hv_dev->driver_override;
 591         if (strlen(driver_override)) {
 592                 hv_dev->driver_override = driver_override;
 593         } else {
 594                 kfree(driver_override);
 595                 hv_dev->driver_override = NULL;
 596         }
 597         device_unlock(dev);
 598
 599         kfree(old);
 600
 601         return count;
 602 }
 603
 604 static ssize_t driver_override_show(struct device *dev,
 605                                     struct device_attribute *attr, char *buf)
 606 {
 607         struct hv_device *hv_dev = device_to_hv_device(dev);
 608         ssize_t len;
 609
 610         device_lock(dev);
 611         len = snprintf(buf, PAGE_SIZE, "%s\n", hv_dev->driver_override);
 612         device_unlock(dev);
 613
 614         return len;
 615 }
 616 static DEVICE_ATTR_RW(driver_override);
 617
 618 /* Set up per device attributes in /sys/bus/vmbus/devices/<bus device> */
 619 static struct attribute *vmbus_dev_attrs[] = {
 620         &dev_attr_id.attr,
 621         &dev_attr_state.attr,
 622         &dev_attr_monitor_id.attr,
 623         &dev_attr_class_id.attr,
 624         &dev_attr_device_id.attr,
 625         &dev_attr_modalias.attr,
 626 #ifdef CONFIG_NUMA
 627         &dev_attr_numa_node.attr,
 628 #endif
 629         &dev_attr_server_monitor_pending.attr,
 630         &dev_attr_client_monitor_pending.attr,
 631         &dev_attr_server_monitor_latency.attr,
 632         &dev_attr_client_monitor_latency.attr,
 633         &dev_attr_server_monitor_conn_id.attr,
 634         &dev_attr_client_monitor_conn_id.attr,
 635         &dev_attr_out_intr_mask.attr,
 636         &dev_attr_out_read_index.attr,
 637         &dev_attr_out_write_index.attr,
 638         &dev_attr_out_read_bytes_avail.attr,
 639         &dev_attr_out_write_bytes_avail.attr,
 640         &dev_attr_in_intr_mask.attr,
 641         &dev_attr_in_read_index.attr,
 642         &dev_attr_in_write_index.attr,
 643         &dev_attr_in_read_bytes_avail.attr,
 644         &dev_attr_in_write_bytes_avail.attr,
 645         &dev_attr_channel_vp_mapping.attr,
 646         &dev_attr_vendor.attr,
 647         &dev_attr_device.attr,
 648         &dev_attr_driver_override.attr,
 649         NULL,
 650 };
 651
 652 /*
 653  * Device-level attribute_group callback function. Returns the permission for
 654  * each attribute, and returns 0 if an attribute is not visible.
 655  */
 656 static umode_t vmbus_dev_attr_is_visible(struct kobject *kobj,
 657                                          struct attribute *attr, int idx)
 658 {
 659         struct device *dev = kobj_to_dev(kobj);
 660         const struct hv_device *hv_dev = device_to_hv_device(dev);
 661
 662         /* Hide the monitor attributes if the monitor mechanism is not used. */
 663         if (!hv_dev->channel->offermsg.monitor_allocated &&
 664             (attr == &dev_attr_monitor_id.attr ||
 665              attr == &dev_attr_server_monitor_pending.attr ||
 666              attr == &dev_attr_client_monitor_pending.attr ||
 667              attr == &dev_attr_server_monitor_latency.attr ||
 668              attr == &dev_attr_client_monitor_latency.attr ||
 669              attr == &dev_attr_server_monitor_conn_id.attr ||
 670              attr == &dev_attr_client_monitor_conn_id.attr))
 671                 return 0;
 672
 673         return attr->mode;
 674 }
 675
 676 static const struct attribute_group vmbus_dev_group = {
 677         .attrs = vmbus_dev_attrs,
 678         .is_visible = vmbus_dev_attr_is_visible
 679 };
 680 __ATTRIBUTE_GROUPS(vmbus_dev);
 681
 682 /*
 683  * vmbus_uevent - add uevent for our device
 684  *
 685  * This routine is invoked when a device is added or removed on the vmbus to
 686  * generate a uevent to udev in the userspace. The udev will then look at its
 687  * rule and the uevent generated here to load the appropriate driver
 688  *
 689  * The alias string will be of the form vmbus:guid where guid is the string
 690  * representation of the device guid (each byte of the guid will be
 691  * represented with two hex characters.
 692  */
 693 static int vmbus_uevent(struct device *device, struct kobj_uevent_env *env)
 694 {
 695         struct hv_device *dev = device_to_hv_device(device);
 696         int ret;
 697         char alias_name[VMBUS_ALIAS_LEN + 1];
 698
 699         print_alias_name(dev, alias_name);
 700         ret = add_uevent_var(env, "MODALIAS=vmbus:%s", alias_name);
 701         return ret;
 702 }
 703
 704 static const struct hv_vmbus_device_id *
 705 hv_vmbus_dev_match(const struct hv_vmbus_device_id *id, const guid_t *guid)
 706 {
 707         if (id == NULL)
 708                 return NULL; /* empty device table */
 709
 710         for (; !guid_is_null(&id->guid); id++)
 711                 if (guid_equal(&id->guid, guid))
 712                         return id;
 713
 714         return NULL;
 715 }
 716
 717 static const struct hv_vmbus_device_id *
 718 hv_vmbus_dynid_match(struct hv_driver *drv, const guid_t *guid)
 719 {
 720         const struct hv_vmbus_device_id *id = NULL;
 721         struct vmbus_dynid *dynid;
 722
 723         spin_lock(&drv->dynids.lock);
 724         list_for_each_entry(dynid, &drv->dynids.list, node) {
 725                 if (guid_equal(&dynid->id.guid, guid)) {
 726                         id = &dynid->id;
 727                         break;
 728                 }
 729         }
 730         spin_unlock(&drv->dynids.lock);
 731
 732         return id;
 733 }
 734
 735 static const struct hv_vmbus_device_id vmbus_device_null;
 736
 737 /*
 738  * Return a matching hv_vmbus_device_id pointer.
 739  * If there is no match, return NULL.
 740  */
 741 static const struct hv_vmbus_device_id *hv_vmbus_get_id(struct hv_driver *drv,
 742                                                         struct hv_device *dev)
 743 {
 744         const guid_t *guid = &dev->dev_type;
 745         const struct hv_vmbus_device_id *id;
 746
 747         /* When driver_override is set, only bind to the matching driver */
 748         if (dev->driver_override && strcmp(dev->driver_override, drv->name))
 749                 return NULL;
 750
 751         /* Look at the dynamic ids first, before the static ones */
 752         id = hv_vmbus_dynid_match(drv, guid);
 753         if (!id)
 754                 id = hv_vmbus_dev_match(drv->id_table, guid);
 755
 756         /* driver_override will always match, send a dummy id */
 757         if (!id && dev->driver_override)
 758                 id = &vmbus_device_null;
 759
 760         return id;
 761 }
 762
 763 /* vmbus_add_dynid - add a new device ID to this driver and re-probe devices */
 764 static int vmbus_add_dynid(struct hv_driver *drv, guid_t *guid)
 765 {
 766         struct vmbus_dynid *dynid;
 767
 768         dynid = kzalloc(sizeof(*dynid), GFP_KERNEL);
 769         if (!dynid)
 770                 return -ENOMEM;
 771
 772         dynid->id.guid = *guid;
 773
 774         spin_lock(&drv->dynids.lock);
 775         list_add_tail(&dynid->node, &drv->dynids.list);
 776         spin_unlock(&drv->dynids.lock);
 777
 778         return driver_attach(&drv->driver);
 779 }
 780
 781 static void vmbus_free_dynids(struct hv_driver *drv)
 782 {
 783         struct vmbus_dynid *dynid, *n;
 784
 785         spin_lock(&drv->dynids.lock);
 786         list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
 787                 list_del(&dynid->node);
 788                 kfree(dynid);
 789         }
 790         spin_unlock(&drv->dynids.lock);
 791 }
 792
 793 /*
 794  * store_new_id - sysfs frontend to vmbus_add_dynid()
 795  *
 796  * Allow GUIDs to be added to an existing driver via sysfs.
 797  */
 798 static ssize_t new_id_store(struct device_driver *driver, const char *buf,
 799                             size_t count)
 800 {
 801         struct hv_driver *drv = drv_to_hv_drv(driver);
 802         guid_t guid;
 803         ssize_t retval;
 804
 805         retval = guid_parse(buf, &guid);
 806         if (retval)
 807                 return retval;
 808
 809         if (hv_vmbus_dynid_match(drv, &guid))
 810                 return -EEXIST;
 811
 812         retval = vmbus_add_dynid(drv, &guid);
 813         if (retval)
 814                 return retval;
 815         return count;
 816 }
 817 static DRIVER_ATTR_WO(new_id);
 818
 819 /*
 820  * store_remove_id - remove a PCI device ID from this driver
 821  *
 822  * Removes a dynamic pci device ID to this driver.
 823  */
 824 static ssize_t remove_id_store(struct device_driver *driver, const char *buf,
 825                                size_t count)
 826 {
 827         struct hv_driver *drv = drv_to_hv_drv(driver);
 828         struct vmbus_dynid *dynid, *n;
 829         guid_t guid;
 830         ssize_t retval;
 831
 832         retval = guid_parse(buf, &guid);
 833         if (retval)
 834                 return retval;
 835
 836         retval = -ENODEV;
 837         spin_lock(&drv->dynids.lock);
 838         list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
 839                 struct hv_vmbus_device_id *id = &dynid->id;
 840
 841                 if (guid_equal(&id->guid, &guid)) {
 842                         list_del(&dynid->node);
 843                         kfree(dynid);
 844                         retval = count;
 845                         break;
 846                 }
 847         }
 848         spin_unlock(&drv->dynids.lock);
 849
 850         return retval;
 851 }
 852 static DRIVER_ATTR_WO(remove_id);
 853
 854 static struct attribute *vmbus_drv_attrs[] = {
 855         &driver_attr_new_id.attr,
 856         &driver_attr_remove_id.attr,
 857         NULL,
 858 };
 859 ATTRIBUTE_GROUPS(vmbus_drv);
 860
 861
 862 /*
 863  * vmbus_match - Attempt to match the specified device to the specified driver
 864  */
 865 static int vmbus_match(struct device *device, struct device_driver *driver)
 866 {
 867         struct hv_driver *drv = drv_to_hv_drv(driver);
 868         struct hv_device *hv_dev = device_to_hv_device(device);
 869
 870         /* The hv_sock driver handles all hv_sock offers. */
 871         if (is_hvsock_channel(hv_dev->channel))
 872                 return drv->hvsock;
 873
 874         if (hv_vmbus_get_id(drv, hv_dev))
 875                 return 1;
 876
 877         return 0;
 878 }
 879
 880 /*
 881  * vmbus_probe - Add the new vmbus's child device
 882  */
 883 static int vmbus_probe(struct device *child_device)
 884 {
 885         int ret = 0;
 886         struct hv_driver *drv =
 887                         drv_to_hv_drv(child_device->driver);
 888         struct hv_device *dev = device_to_hv_device(child_device);
 889         const struct hv_vmbus_device_id *dev_id;
 890
 891         dev_id = hv_vmbus_get_id(drv, dev);
 892         if (drv->probe) {
 893                 ret = drv->probe(dev, dev_id);
 894                 if (ret != 0)
 895                         pr_err("probe failed for device %s (%d)\n",
 896                                dev_name(child_device), ret);
 897
 898         } else {
 899                 pr_err("probe not set for driver %s\n",
 900                        dev_name(child_device));
 901                 ret = -ENODEV;
 902         }
 903         return ret;
 904 }
 905
 906 /*
 907  * vmbus_remove - Remove a vmbus device
 908  */
 909 static int vmbus_remove(struct device *child_device)
 910 {
 911         struct hv_driver *drv;
 912         struct hv_device *dev = device_to_hv_device(child_device);
 913
 914         if (child_device->driver) {
 915                 drv = drv_to_hv_drv(child_device->driver);
 916                 if (drv->remove)
 917                         drv->remove(dev);
 918         }
 919
 920         return 0;
 921 }
 922
 923
 924 /*
 925  * vmbus_shutdown - Shutdown a vmbus device
 926  */
 927 static void vmbus_shutdown(struct device *child_device)
 928 {
 929         struct hv_driver *drv;
 930         struct hv_device *dev = device_to_hv_device(child_device);
 931
 932
 933         /* The device may not be attached yet */
 934         if (!child_device->driver)
 935                 return;
 936
 937         drv = drv_to_hv_drv(child_device->driver);
 938
 939         if (drv->shutdown)
 940                 drv->shutdown(dev);
 941 }
 942
 943 #ifdef CONFIG_PM_SLEEP
 944 /*
 945  * vmbus_suspend - Suspend a vmbus device
 946  */
 947 static int vmbus_suspend(struct device *child_device)
 948 {
 949         struct hv_driver *drv;
 950         struct hv_device *dev = device_to_hv_device(child_device);
 951
 952         /* The device may not be attached yet */
 953         if (!child_device->driver)
 954                 return 0;
 955
 956         drv = drv_to_hv_drv(child_device->driver);
 957         if (!drv->suspend)
 958                 return -EOPNOTSUPP;
 959
 960         return drv->suspend(dev);
 961 }
 962
 963 /*
 964  * vmbus_resume - Resume a vmbus device
 965  */
 966 static int vmbus_resume(struct device *child_device)
 967 {
 968         struct hv_driver *drv;
 969         struct hv_device *dev = device_to_hv_device(child_device);
 970
 971         /* The device may not be attached yet */
 972         if (!child_device->driver)
 973                 return 0;
 974
 975         drv = drv_to_hv_drv(child_device->driver);
 976         if (!drv->resume)
 977                 return -EOPNOTSUPP;
 978
 979         return drv->resume(dev);
 980 }
 981 #else
 982 #define vmbus_suspend NULL
 983 #define vmbus_resume NULL
 984 #endif /* CONFIG_PM_SLEEP */
 985
 986 /*
 987  * vmbus_device_release - Final callback release of the vmbus child device
 988  */
 989 static void vmbus_device_release(struct device *device)
 990 {
 991         struct hv_device *hv_dev = device_to_hv_device(device);
 992         struct vmbus_channel *channel = hv_dev->channel;
 993
 994         mutex_lock(&vmbus_connection.channel_mutex);
 995         hv_process_channel_removal(channel);
 996         mutex_unlock(&vmbus_connection.channel_mutex);
 997         kfree(hv_dev);
 998 }
 999
1000 /*
1001  * Note: we must use the "noirq" ops: see the comment before vmbus_bus_pm.
1002  *
1003  * suspend_noirq/resume_noirq are set to NULL to support Suspend-to-Idle: we
1004  * shouldn't suspend the vmbus devices upon Suspend-to-Idle, otherwise there
1005  * is no way to wake up a Generation-2 VM.
1006  *
1007  * The other 4 ops are for hibernation.
1008  */
1009
1010 static const struct dev_pm_ops vmbus_pm = {
1011         .suspend_noirq  = NULL,
1012         .resume_noirq   = NULL,
1013         .freeze_noirq   = vmbus_suspend,
1014         .thaw_noirq     = vmbus_resume,
1015         .poweroff_noirq = vmbus_suspend,
1016         .restore_noirq  = vmbus_resume,
1017 };
1018
1019 /* The one and only one */
1020 static struct bus_type  hv_bus = {
1021         .name =         "vmbus",
1022         .match =                vmbus_match,
1023         .shutdown =             vmbus_shutdown,
1024         .remove =               vmbus_remove,
1025         .probe =                vmbus_probe,
1026         .uevent =               vmbus_uevent,
1027         .dev_groups =           vmbus_dev_groups,
1028         .drv_groups =           vmbus_drv_groups,
1029         .pm =                   &vmbus_pm,
1030 };
1031
1032 struct onmessage_work_context {
1033         struct work_struct work;
1034         struct hv_message msg;
1035 };
1036
1037 static void vmbus_onmessage_work(struct work_struct *work)
1038 {
1039         struct onmessage_work_context *ctx;
1040
1041         /* Do not process messages if we're in DISCONNECTED state */
1042         if (vmbus_connection.conn_state == DISCONNECTED)
1043                 return;
1044
1045         ctx = container_of(work, struct onmessage_work_context,
1046                            work);
1047         vmbus_onmessage(&ctx->msg);
1048         kfree(ctx);
1049 }
1050
1051 void vmbus_on_msg_dpc(unsigned long data)
1052 {
1053         struct hv_per_cpu_context *hv_cpu = (void *)data;
1054         void *page_addr = hv_cpu->synic_message_page;
1055         struct hv_message *msg = (struct hv_message *)page_addr +
1056                                   VMBUS_MESSAGE_SINT;
1057         struct vmbus_channel_message_header *hdr;
1058         const struct vmbus_channel_message_table_entry *entry;
1059         struct onmessage_work_context *ctx;
1060         u32 message_type = msg->header.message_type;
1061
1062         if (message_type == HVMSG_NONE)
1063                 /* no msg */
1064                 return;
1065
1066         hdr = (struct vmbus_channel_message_header *)msg->u.payload;
1067
1068         trace_vmbus_on_msg_dpc(hdr);
1069
1070         if (hdr->msgtype >= CHANNELMSG_COUNT) {
1071                 WARN_ONCE(1, "unknown msgtype=%d\n", hdr->msgtype);
1072                 goto msg_handled;
1073         }
1074
1075         entry = &channel_message_table[hdr->msgtype];
1076
1077         if (!entry->message_handler)
1078                 goto msg_handled;
1079
1080         if (entry->handler_type == VMHT_BLOCKING) {
1081                 ctx = kmalloc(sizeof(*ctx), GFP_ATOMIC);
1082                 if (ctx == NULL)
1083                         return;
1084
1085                 INIT_WORK(&ctx->work, vmbus_onmessage_work);
1086                 memcpy(&ctx->msg, msg, sizeof(*msg));
1087
1088                 /*
1089                  * The host can generate a rescind message while we
1090                  * may still be handling the original offer. We deal with
1091                  * this condition by ensuring the processing is done on the
1092                  * same CPU.
1093                  */
1094                 switch (hdr->msgtype) {
1095                 case CHANNELMSG_RESCIND_CHANNELOFFER:
1096                         /*
1097                          * If we are handling the rescind message;
1098                          * schedule the work on the global work queue.
1099                          *
1100                          * The OFFER message and the RESCIND message should
1101                          * not be handled by the same serialized work queue,
1102                          * because the OFFER handler may call vmbus_open(),
1103                          * which tries to open the channel by sending an
1104                          * OPEN_CHANNEL message to the host and waits for
1105                          * the host's response; however, if the host has
1106                          * rescinded the channel before it receives the
1107                          * OPEN_CHANNEL message, the host just silently
1108                          * ignores the OPEN_CHANNEL message; as a result,
1109                          * the guest's OFFER handler hangs for ever, if we
1110                          * handle the RESCIND message in the same serialized
1111                          * work queue: the RESCIND handler can not start to
1112                          * run before the OFFER handler finishes.
1113                          */
1114                         schedule_work_on(VMBUS_CONNECT_CPU,
1115                                          &ctx->work);
1116                         break;
1117
1118                 case CHANNELMSG_OFFERCHANNEL:
1119                         atomic_inc(&vmbus_connection.offer_in_progress);
1120                         queue_work_on(VMBUS_CONNECT_CPU,
1121                                       vmbus_connection.work_queue,
1122                                       &ctx->work);
1123                         break;
1124
1125                 default:
1126                         queue_work(vmbus_connection.work_queue, &ctx->work);
1127                 }
1128         } else
1129                 entry->message_handler(hdr);
1130
1131 msg_handled:
1132         vmbus_signal_eom(msg, message_type);
1133 }
1134
1135 #ifdef CONFIG_PM_SLEEP
1136 /*
1137  * Fake RESCIND_CHANNEL messages to clean up hv_sock channels by force for
1138  * hibernation, because hv_sock connections can not persist across hibernation.
1139  */
1140 static void vmbus_force_channel_rescinded(struct vmbus_channel *channel)
1141 {
1142         struct onmessage_work_context *ctx;
1143         struct vmbus_channel_rescind_offer *rescind;
1144
1145         WARN_ON(!is_hvsock_channel(channel));
1146
1147         /*
1148          * sizeof(*ctx) is small and the allocation should really not fail,
1149          * otherwise the state of the hv_sock connections ends up in limbo.
1150          */
1151         ctx = kzalloc(sizeof(*ctx), GFP_KERNEL | __GFP_NOFAIL);
1152
1153         /*
1154          * So far, these are not really used by Linux. Just set them to the
1155          * reasonable values conforming to the definitions of the fields.
1156          */
1157         ctx->msg.header.message_type = 1;
1158         ctx->msg.header.payload_size = sizeof(*rescind);
1159
1160         /* These values are actually used by Linux. */
1161         rescind = (struct vmbus_channel_rescind_offer *)ctx->msg.u.payload;
1162         rescind->header.msgtype = CHANNELMSG_RESCIND_CHANNELOFFER;
1163         rescind->child_relid = channel->offermsg.child_relid;
1164
1165         INIT_WORK(&ctx->work, vmbus_onmessage_work);
1166
1167         queue_work_on(VMBUS_CONNECT_CPU,
1168                       vmbus_connection.work_queue,
1169                       &ctx->work);
1170 }
1171 #endif /* CONFIG_PM_SLEEP */
1172
1173 /*
1174  * Direct callback for channels using other deferred processing
1175  */
1176 static void vmbus_channel_isr(struct vmbus_channel *channel)
1177 {
1178         void (*callback_fn)(void *);
1179
1180         callback_fn = READ_ONCE(channel->onchannel_callback);
1181         if (likely(callback_fn != NULL))
1182                 (*callback_fn)(channel->channel_callback_context);
1183 }
1184
1185 /*
1186  * Schedule all channels with events pending
1187  */
1188 static void vmbus_chan_sched(struct hv_per_cpu_context *hv_cpu)
1189 {
1190         unsigned long *recv_int_page;
1191         u32 maxbits, relid;
1192
1193         if (vmbus_proto_version < VERSION_WIN8) {
1194                 maxbits = MAX_NUM_CHANNELS_SUPPORTED;
1195                 recv_int_page = vmbus_connection.recv_int_page;
1196         } else {
1197                 /*
1198                  * When the host is win8 and beyond, the event page
1199                  * can be directly checked to get the id of the channel
1200                  * that has the interrupt pending.
1201                  */
1202                 void *page_addr = hv_cpu->synic_event_page;
1203                 union hv_synic_event_flags *event
1204                         = (union hv_synic_event_flags *)page_addr +
1205                                                  VMBUS_MESSAGE_SINT;
1206
1207                 maxbits = HV_EVENT_FLAGS_COUNT;
1208                 recv_int_page = event->flags;
1209         }
1210
1211         if (unlikely(!recv_int_page))
1212                 return;
1213
1214         for_each_set_bit(relid, recv_int_page, maxbits) {
1215                 struct vmbus_channel *channel;
1216
1217                 if (!sync_test_and_clear_bit(relid, recv_int_page))
1218                         continue;
1219
1220                 /* Special case - vmbus channel protocol msg */
1221                 if (relid == 0)
1222                         continue;
1223
1224                 rcu_read_lock();
1225
1226                 /* Find channel based on relid */
1227                 list_for_each_entry_rcu(channel, &hv_cpu->chan_list, percpu_list) {
1228                         if (channel->offermsg.child_relid != relid)
1229                                 continue;
1230
1231                         if (channel->rescind)
1232                                 continue;
1233
1234                         trace_vmbus_chan_sched(channel);
1235
1236                         ++channel->interrupts;
1237
1238                         switch (channel->callback_mode) {
1239                         case HV_CALL_ISR:
1240                                 vmbus_channel_isr(channel);
1241                                 break;
1242
1243                         case HV_CALL_BATCHED:
1244                                 hv_begin_read(&channel->inbound);
1245                                 /* fallthrough */
1246                         case HV_CALL_DIRECT:
1247                                 tasklet_schedule(&channel->callback_event);
1248                         }
1249                 }
1250
1251                 rcu_read_unlock();
1252         }
1253 }
1254
1255 static void vmbus_isr(void)
1256 {
1257         struct hv_per_cpu_context *hv_cpu
1258                 = this_cpu_ptr(hv_context.cpu_context);
1259         void *page_addr = hv_cpu->synic_event_page;
1260         struct hv_message *msg;
1261         union hv_synic_event_flags *event;
1262         bool handled = false;
1263
1264         if (unlikely(page_addr == NULL))
1265                 return;
1266
1267         event = (union hv_synic_event_flags *)page_addr +
1268                                          VMBUS_MESSAGE_SINT;
1269         /*
1270          * Check for events before checking for messages. This is the order
1271          * in which events and messages are checked in Windows guests on
1272          * Hyper-V, and the Windows team suggested we do the same.
1273          */
1274
1275         if ((vmbus_proto_version == VERSION_WS2008) ||
1276                 (vmbus_proto_version == VERSION_WIN7)) {
1277
1278                 /* Since we are a child, we only need to check bit 0 */
1279                 if (sync_test_and_clear_bit(0, event->flags))
1280                         handled = true;
1281         } else {
1282                 /*
1283                  * Our host is win8 or above. The signaling mechanism
1284                  * has changed and we can directly look at the event page.
1285                  * If bit n is set then we have an interrup on the channel
1286                  * whose id is n.
1287                  */
1288                 handled = true;
1289         }
1290
1291         if (handled)
1292                 vmbus_chan_sched(hv_cpu);
1293
1294         page_addr = hv_cpu->synic_message_page;
1295         msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
1296
1297         /* Check if there are actual msgs to be processed */
1298         if (msg->header.message_type != HVMSG_NONE) {
1299                 if (msg->header.message_type == HVMSG_TIMER_EXPIRED) {
1300                         hv_stimer0_isr();
1301                         vmbus_signal_eom(msg, HVMSG_TIMER_EXPIRED);
1302                 } else
1303                         tasklet_schedule(&hv_cpu->msg_dpc);
1304         }
1305
1306         add_interrupt_randomness(HYPERVISOR_CALLBACK_VECTOR);
1307 }
1308
1309 /*
1310  * Callback from kmsg_dump. Grab as much as possible from the end of the kmsg
1311  * buffer and call into Hyper-V to transfer the data.
1312  */
1313 static void hv_kmsg_dump(struct kmsg_dumper *dumper,
1314                          enum kmsg_dump_reason reason)
1315 {
1316         size_t bytes_written;
1317         phys_addr_t panic_pa;
1318
1319         /* We are only interested in panics. */
1320         if ((reason != KMSG_DUMP_PANIC) || (!sysctl_record_panic_msg))
1321                 return;
1322
1323         panic_pa = virt_to_phys(hv_panic_page);
1324
1325         /*
1326          * Write dump contents to the page. No need to synchronize; panic should
1327          * be single-threaded.
1328          */
1329         kmsg_dump_get_buffer(dumper, true, hv_panic_page, PAGE_SIZE,
1330                              &bytes_written);
1331         if (bytes_written)
1332                 hyperv_report_panic_msg(panic_pa, bytes_written);
1333 }
1334
1335 static struct kmsg_dumper hv_kmsg_dumper = {
1336         .dump = hv_kmsg_dump,
1337 };
1338
1339 static struct ctl_table_header *hv_ctl_table_hdr;
1340
1341 /*
1342  * sysctl option to allow the user to control whether kmsg data should be
1343  * reported to Hyper-V on panic.
1344  */
1345 static struct ctl_table hv_ctl_table[] = {
1346         {
1347                 .procname       = "hyperv_record_panic_msg",
1348                 .data           = &sysctl_record_panic_msg,
1349                 .maxlen         = sizeof(int),
1350                 .mode           = 0644,
1351                 .proc_handler   = proc_dointvec_minmax,
1352                 .extra1         = SYSCTL_ZERO,
1353                 .extra2         = SYSCTL_ONE
1354         },
1355         {}
1356 };
1357
1358 static struct ctl_table hv_root_table[] = {
1359         {
1360                 .procname       = "kernel",
1361                 .mode           = 0555,
1362                 .child          = hv_ctl_table
1363         },
1364         {}
1365 };
1366
1367 /*
1368  * vmbus_bus_init -Main vmbus driver initialization routine.
1369  *
1370  * Here, we
1371  *      - initialize the vmbus driver context
1372  *      - invoke the vmbus hv main init routine
1373  *      - retrieve the channel offers
1374  */
1375 static int vmbus_bus_init(void)
1376 {
1377         int ret;
1378
1379         /* Hypervisor initialization...setup hypercall page..etc */
1380         ret = hv_init();
1381         if (ret != 0) {
1382                 pr_err("Unable to initialize the hypervisor - 0x%x\n", ret);
1383                 return ret;
1384         }
1385
1386         ret = bus_register(&hv_bus);
1387         if (ret)
1388                 return ret;
1389
1390         hv_setup_vmbus_irq(vmbus_isr);
1391
1392         ret = hv_synic_alloc();
1393         if (ret)
1394                 goto err_alloc;
1395
1396         ret = hv_stimer_alloc(VMBUS_MESSAGE_SINT);
1397         if (ret < 0)
1398                 goto err_alloc;
1399
1400         /*
1401          * Initialize the per-cpu interrupt state and stimer state.
1402          * Then connect to the host.
1403          */
1404         ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "hyperv/vmbus:online",
1405                                 hv_synic_init, hv_synic_cleanup);
1406         if (ret < 0)
1407                 goto err_cpuhp;
1408         hyperv_cpuhp_online = ret;
1409
1410         ret = vmbus_connect();
1411         if (ret)
1412                 goto err_connect;
1413
1414         /*
1415          * Only register if the crash MSRs are available
1416          */
1417         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1418                 u64 hyperv_crash_ctl;
1419                 /*
1420                  * Sysctl registration is not fatal, since by default
1421                  * reporting is enabled.
1422                  */
1423                 hv_ctl_table_hdr = register_sysctl_table(hv_root_table);
1424                 if (!hv_ctl_table_hdr)
1425                         pr_err("Hyper-V: sysctl table register error");
1426
1427                 /*
1428                  * Register for panic kmsg callback only if the right
1429                  * capability is supported by the hypervisor.
1430                  */
1431                 hv_get_crash_ctl(hyperv_crash_ctl);
1432                 if (hyperv_crash_ctl & HV_CRASH_CTL_CRASH_NOTIFY_MSG) {
1433                         hv_panic_page = (void *)get_zeroed_page(GFP_KERNEL);
1434                         if (hv_panic_page) {
1435                                 ret = kmsg_dump_register(&hv_kmsg_dumper);
1436                                 if (ret) {
1437                                         pr_err("Hyper-V: kmsg dump register "
1438                                                 "error 0x%x\n", ret);
1439                                         hv_free_hyperv_page(
1440                                             (unsigned long)hv_panic_page);
1441                                         hv_panic_page = NULL;
1442                                 }
1443                         } else
1444                                 pr_err("Hyper-V: panic message page memory "
1445                                         "allocation failed");
1446                 }
1447
1448                 register_die_notifier(&hyperv_die_block);
1449         }
1450
1451         /*
1452          * Always register the panic notifier because we need to unload
1453          * the VMbus channel connection to prevent any VMbus
1454          * activity after the VM panics.
1455          */
1456         atomic_notifier_chain_register(&panic_notifier_list,
1457                                &hyperv_panic_block);
1458
1459         vmbus_request_offers();
1460
1461         return 0;
1462
1463 err_connect:
1464         cpuhp_remove_state(hyperv_cpuhp_online);
1465 err_cpuhp:
1466         hv_stimer_free();
1467 err_alloc:
1468         hv_synic_free();
1469         hv_remove_vmbus_irq();
1470
1471         bus_unregister(&hv_bus);
1472         unregister_sysctl_table(hv_ctl_table_hdr);
1473         hv_ctl_table_hdr = NULL;
1474         return ret;
1475 }
1476
1477 /**
1478  * __vmbus_child_driver_register() - Register a vmbus's driver
1479  * @hv_driver: Pointer to driver structure you want to register
1480  * @owner: owner module of the drv
1481  * @mod_name: module name string
1482  *
1483  * Registers the given driver with Linux through the 'driver_register()' call
1484  * and sets up the hyper-v vmbus handling for this driver.
1485  * It will return the state of the 'driver_register()' call.
1486  *
1487  */
1488 int __vmbus_driver_register(struct hv_driver *hv_driver, struct module *owner, const char *mod_name)
1489 {
1490         int ret;
1491
1492         pr_info("registering driver %s\n", hv_driver->name);
1493
1494         ret = vmbus_exists();
1495         if (ret < 0)
1496                 return ret;
1497
1498         hv_driver->driver.name = hv_driver->name;
1499         hv_driver->driver.owner = owner;
1500         hv_driver->driver.mod_name = mod_name;
1501         hv_driver->driver.bus = &hv_bus;
1502
1503         spin_lock_init(&hv_driver->dynids.lock);
1504         INIT_LIST_HEAD(&hv_driver->dynids.list);
1505
1506         ret = driver_register(&hv_driver->driver);
1507
1508         return ret;
1509 }
1510 EXPORT_SYMBOL_GPL(__vmbus_driver_register);
1511
1512 /**
1513  * vmbus_driver_unregister() - Unregister a vmbus's driver
1514  * @hv_driver: Pointer to driver structure you want to
1515  *             un-register
1516  *
1517  * Un-register the given driver that was previous registered with a call to
1518  * vmbus_driver_register()
1519  */
1520 void vmbus_driver_unregister(struct hv_driver *hv_driver)
1521 {
1522         pr_info("unregistering driver %s\n", hv_driver->name);
1523
1524         if (!vmbus_exists()) {
1525                 driver_unregister(&hv_driver->driver);
1526                 vmbus_free_dynids(hv_driver);
1527         }
1528 }
1529 EXPORT_SYMBOL_GPL(vmbus_driver_unregister);
1530
1531
1532 /*
1533  * Called when last reference to channel is gone.
1534  */
1535 static void vmbus_chan_release(struct kobject *kobj)
1536 {
1537         struct vmbus_channel *channel
1538                 = container_of(kobj, struct vmbus_channel, kobj);
1539
1540         kfree_rcu(channel, rcu);
1541 }
1542
1543 struct vmbus_chan_attribute {
1544         struct attribute attr;
1545         ssize_t (*show)(struct vmbus_channel *chan, char *buf);
1546         ssize_t (*store)(struct vmbus_channel *chan,
1547                          const char *buf, size_t count);
1548 };
1549 #define VMBUS_CHAN_ATTR(_name, _mode, _show, _store) \
1550         struct vmbus_chan_attribute chan_attr_##_name \
1551                 = __ATTR(_name, _mode, _show, _store)
1552 #define VMBUS_CHAN_ATTR_RW(_name) \
1553         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RW(_name)
1554 #define VMBUS_CHAN_ATTR_RO(_name) \
1555         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RO(_name)
1556 #define VMBUS_CHAN_ATTR_WO(_name) \
1557         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_WO(_name)
1558
1559 static ssize_t vmbus_chan_attr_show(struct kobject *kobj,
1560                                     struct attribute *attr, char *buf)
1561 {
1562         const struct vmbus_chan_attribute *attribute
1563                 = container_of(attr, struct vmbus_chan_attribute, attr);
1564         struct vmbus_channel *chan
1565                 = container_of(kobj, struct vmbus_channel, kobj);
1566
1567         if (!attribute->show)
1568                 return -EIO;
1569
1570         return attribute->show(chan, buf);
1571 }
1572
1573 static const struct sysfs_ops vmbus_chan_sysfs_ops = {
1574         .show = vmbus_chan_attr_show,
1575 };
1576
1577 static ssize_t out_mask_show(struct vmbus_channel *channel, char *buf)
1578 {
1579         struct hv_ring_buffer_info *rbi = &channel->outbound;
1580         ssize_t ret;
1581
1582         mutex_lock(&rbi->ring_buffer_mutex);
1583         if (!rbi->ring_buffer) {
1584                 mutex_unlock(&rbi->ring_buffer_mutex);
1585                 return -EINVAL;
1586         }
1587
1588         ret = sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1589         mutex_unlock(&rbi->ring_buffer_mutex);
1590         return ret;
1591 }
1592 static VMBUS_CHAN_ATTR_RO(out_mask);
1593
1594 static ssize_t in_mask_show(struct vmbus_channel *channel, char *buf)
1595 {
1596         struct hv_ring_buffer_info *rbi = &channel->inbound;
1597         ssize_t ret;
1598
1599         mutex_lock(&rbi->ring_buffer_mutex);
1600         if (!rbi->ring_buffer) {
1601                 mutex_unlock(&rbi->ring_buffer_mutex);
1602                 return -EINVAL;
1603         }
1604
1605         ret = sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1606         mutex_unlock(&rbi->ring_buffer_mutex);
1607         return ret;
1608 }
1609 static VMBUS_CHAN_ATTR_RO(in_mask);
1610
1611 static ssize_t read_avail_show(struct vmbus_channel *channel, char *buf)
1612 {
1613         struct hv_ring_buffer_info *rbi = &channel->inbound;
1614         ssize_t ret;
1615
1616         mutex_lock(&rbi->ring_buffer_mutex);
1617         if (!rbi->ring_buffer) {
1618                 mutex_unlock(&rbi->ring_buffer_mutex);
1619                 return -EINVAL;
1620         }
1621
1622         ret = sprintf(buf, "%u\n", hv_get_bytes_to_read(rbi));
1623         mutex_unlock(&rbi->ring_buffer_mutex);
1624         return ret;
1625 }
1626 static VMBUS_CHAN_ATTR_RO(read_avail);
1627
1628 static ssize_t write_avail_show(struct vmbus_channel *channel, char *buf)
1629 {
1630         struct hv_ring_buffer_info *rbi = &channel->outbound;
1631         ssize_t ret;
1632
1633         mutex_lock(&rbi->ring_buffer_mutex);
1634         if (!rbi->ring_buffer) {
1635                 mutex_unlock(&rbi->ring_buffer_mutex);
1636                 return -EINVAL;
1637         }
1638
1639         ret = sprintf(buf, "%u\n", hv_get_bytes_to_write(rbi));
1640         mutex_unlock(&rbi->ring_buffer_mutex);
1641         return ret;
1642 }
1643 static VMBUS_CHAN_ATTR_RO(write_avail);
1644
1645 static ssize_t show_target_cpu(struct vmbus_channel *channel, char *buf)
1646 {
1647         return sprintf(buf, "%u\n", channel->target_cpu);
1648 }
1649 static VMBUS_CHAN_ATTR(cpu, S_IRUGO, show_target_cpu, NULL);
1650
1651 static ssize_t channel_pending_show(struct vmbus_channel *channel,
1652                                     char *buf)
1653 {
1654         return sprintf(buf, "%d\n",
1655                        channel_pending(channel,
1656                                        vmbus_connection.monitor_pages[1]));
1657 }
1658 static VMBUS_CHAN_ATTR(pending, S_IRUGO, channel_pending_show, NULL);
1659
1660 static ssize_t channel_latency_show(struct vmbus_channel *channel,
1661                                     char *buf)
1662 {
1663         return sprintf(buf, "%d\n",
1664                        channel_latency(channel,
1665                                        vmbus_connection.monitor_pages[1]));
1666 }
1667 static VMBUS_CHAN_ATTR(latency, S_IRUGO, channel_latency_show, NULL);
1668
1669 static ssize_t channel_interrupts_show(struct vmbus_channel *channel, char *buf)
1670 {
1671         return sprintf(buf, "%llu\n", channel->interrupts);
1672 }
1673 static VMBUS_CHAN_ATTR(interrupts, S_IRUGO, channel_interrupts_show, NULL);
1674
1675 static ssize_t channel_events_show(struct vmbus_channel *channel, char *buf)
1676 {
1677         return sprintf(buf, "%llu\n", channel->sig_events);
1678 }
1679 static VMBUS_CHAN_ATTR(events, S_IRUGO, channel_events_show, NULL);
1680
1681 static ssize_t channel_intr_in_full_show(struct vmbus_channel *channel,
1682                                          char *buf)
1683 {
1684         return sprintf(buf, "%llu\n",
1685                        (unsigned long long)channel->intr_in_full);
1686 }
1687 static VMBUS_CHAN_ATTR(intr_in_full, 0444, channel_intr_in_full_show, NULL);
1688
1689 static ssize_t channel_intr_out_empty_show(struct vmbus_channel *channel,
1690                                            char *buf)
1691 {
1692         return sprintf(buf, "%llu\n",
1693                        (unsigned long long)channel->intr_out_empty);
1694 }
1695 static VMBUS_CHAN_ATTR(intr_out_empty, 0444, channel_intr_out_empty_show, NULL);
1696
1697 static ssize_t channel_out_full_first_show(struct vmbus_channel *channel,
1698                                            char *buf)
1699 {
1700         return sprintf(buf, "%llu\n",
1701                        (unsigned long long)channel->out_full_first);
1702 }
1703 static VMBUS_CHAN_ATTR(out_full_first, 0444, channel_out_full_first_show, NULL);
1704
1705 static ssize_t channel_out_full_total_show(struct vmbus_channel *channel,
1706                                            char *buf)
1707 {
1708         return sprintf(buf, "%llu\n",
1709                        (unsigned long long)channel->out_full_total);
1710 }
1711 static VMBUS_CHAN_ATTR(out_full_total, 0444, channel_out_full_total_show, NULL);
1712
1713 static ssize_t subchannel_monitor_id_show(struct vmbus_channel *channel,
1714                                           char *buf)
1715 {
1716         return sprintf(buf, "%u\n", channel->offermsg.monitorid);
1717 }
1718 static VMBUS_CHAN_ATTR(monitor_id, S_IRUGO, subchannel_monitor_id_show, NULL);
1719
1720 static ssize_t subchannel_id_show(struct vmbus_channel *channel,
1721                                   char *buf)
1722 {
1723         return sprintf(buf, "%u\n",
1724                        channel->offermsg.offer.sub_channel_index);
1725 }
1726 static VMBUS_CHAN_ATTR_RO(subchannel_id);
1727
1728 static struct attribute *vmbus_chan_attrs[] = {
1729         &chan_attr_out_mask.attr,
1730         &chan_attr_in_mask.attr,
1731         &chan_attr_read_avail.attr,
1732         &chan_attr_write_avail.attr,
1733         &chan_attr_cpu.attr,
1734         &chan_attr_pending.attr,
1735         &chan_attr_latency.attr,
1736         &chan_attr_interrupts.attr,
1737         &chan_attr_events.attr,
1738         &chan_attr_intr_in_full.attr,
1739         &chan_attr_intr_out_empty.attr,
1740         &chan_attr_out_full_first.attr,
1741         &chan_attr_out_full_total.attr,
1742         &chan_attr_monitor_id.attr,
1743         &chan_attr_subchannel_id.attr,
1744         NULL
1745 };
1746
1747 /*
1748  * Channel-level attribute_group callback function. Returns the permission for
1749  * each attribute, and returns 0 if an attribute is not visible.
1750  */
1751 static umode_t vmbus_chan_attr_is_visible(struct kobject *kobj,
1752                                           struct attribute *attr, int idx)
1753 {
1754         const struct vmbus_channel *channel =
1755                 container_of(kobj, struct vmbus_channel, kobj);
1756
1757         /* Hide the monitor attributes if the monitor mechanism is not used. */
1758         if (!channel->offermsg.monitor_allocated &&
1759             (attr == &chan_attr_pending.attr ||
1760              attr == &chan_attr_latency.attr ||
1761              attr == &chan_attr_monitor_id.attr))
1762                 return 0;
1763
1764         return attr->mode;
1765 }
1766
1767 static struct attribute_group vmbus_chan_group = {
1768         .attrs = vmbus_chan_attrs,
1769         .is_visible = vmbus_chan_attr_is_visible
1770 };
1771
1772 static struct kobj_type vmbus_chan_ktype = {
1773         .sysfs_ops = &vmbus_chan_sysfs_ops,
1774         .release = vmbus_chan_release,
1775 };
1776
1777 /*
1778  * vmbus_add_channel_kobj - setup a sub-directory under device/channels
1779  */
1780 int vmbus_add_channel_kobj(struct hv_device *dev, struct vmbus_channel *channel)
1781 {
1782         const struct device *device = &dev->device;
1783         struct kobject *kobj = &channel->kobj;
1784         u32 relid = channel->offermsg.child_relid;
1785         int ret;
1786
1787         kobj->kset = dev->channels_kset;
1788         ret = kobject_init_and_add(kobj, &vmbus_chan_ktype, NULL,
1789                                    "%u", relid);
1790         if (ret) {
1791                 kobject_put(kobj);
1792                 return ret;
1793         }
1794
1795         ret = sysfs_create_group(kobj, &vmbus_chan_group);
1796
1797         if (ret) {
1798                 /*
1799                  * The calling functions' error handling paths will cleanup the
1800                  * empty channel directory.
1801                  */
1802                 kobject_put(kobj);
1803                 dev_err(device, "Unable to set up channel sysfs files\n");
1804                 return ret;
1805         }
1806
1807         kobject_uevent(kobj, KOBJ_ADD);
1808
1809         return 0;
1810 }
1811
1812 /*
1813  * vmbus_remove_channel_attr_group - remove the channel's attribute group
1814  */
1815 void vmbus_remove_channel_attr_group(struct vmbus_channel *channel)
1816 {
1817         sysfs_remove_group(&channel->kobj, &vmbus_chan_group);
1818 }
1819
1820 /*
1821  * vmbus_device_create - Creates and registers a new child device
1822  * on the vmbus.
1823  */
1824 struct hv_device *vmbus_device_create(const guid_t *type,
1825                                       const guid_t *instance,
1826                                       struct vmbus_channel *channel)
1827 {
1828         struct hv_device *child_device_obj;
1829
1830         child_device_obj = kzalloc(sizeof(struct hv_device), GFP_KERNEL);
1831         if (!child_device_obj) {
1832                 pr_err("Unable to allocate device object for child device\n");
1833                 return NULL;
1834         }
1835
1836         child_device_obj->channel = channel;
1837         guid_copy(&child_device_obj->dev_type, type);
1838         guid_copy(&child_device_obj->dev_instance, instance);
1839         child_device_obj->vendor_id = 0x1414; /* MSFT vendor ID */
1840
1841         return child_device_obj;
1842 }
1843
1844 /*
1845  * vmbus_device_register - Register the child device
1846  */
1847 int vmbus_device_register(struct hv_device *child_device_obj)
1848 {
1849         struct kobject *kobj = &child_device_obj->device.kobj;
1850         int ret;
1851
1852         dev_set_name(&child_device_obj->device, "%pUl",
1853                      child_device_obj->channel->offermsg.offer.if_instance.b);
1854
1855         child_device_obj->device.bus = &hv_bus;
1856         child_device_obj->device.parent = &hv_acpi_dev->dev;
1857         child_device_obj->device.release = vmbus_device_release;
1858
1859         /*
1860          * Register with the LDM. This will kick off the driver/device
1861          * binding...which will eventually call vmbus_match() and vmbus_probe()
1862          */
1863         ret = device_register(&child_device_obj->device);
1864         if (ret) {
1865                 pr_err("Unable to register child device\n");
1866                 return ret;
1867         }
1868
1869         child_device_obj->channels_kset = kset_create_and_add("channels",
1870                                                               NULL, kobj);
1871         if (!child_device_obj->channels_kset) {
1872                 ret = -ENOMEM;
1873                 goto err_dev_unregister;
1874         }
1875
1876         ret = vmbus_add_channel_kobj(child_device_obj,
1877                                      child_device_obj->channel);
1878         if (ret) {
1879                 pr_err("Unable to register primary channeln");
1880                 goto err_kset_unregister;
1881         }
1882
1883         return 0;
1884
1885 err_kset_unregister:
1886         kset_unregister(child_device_obj->channels_kset);
1887
1888 err_dev_unregister:
1889         device_unregister(&child_device_obj->device);
1890         return ret;
1891 }
1892
1893 /*
1894  * vmbus_device_unregister - Remove the specified child device
1895  * from the vmbus.
1896  */
1897 void vmbus_device_unregister(struct hv_device *device_obj)
1898 {
1899         pr_debug("child device %s unregistered\n",
1900                 dev_name(&device_obj->device));
1901
1902         kset_unregister(device_obj->channels_kset);
1903
1904         /*
1905          * Kick off the process of unregistering the device.
1906          * This will call vmbus_remove() and eventually vmbus_device_release()
1907          */
1908         device_unregister(&device_obj->device);
1909 }
1910
1911
1912 /*
1913  * VMBUS is an acpi enumerated device. Get the information we
1914  * need from DSDT.
1915  */
1916 #define VTPM_BASE_ADDRESS 0xfed40000
1917 static acpi_status vmbus_walk_resources(struct acpi_resource *res, void *ctx)
1918 {
1919         resource_size_t start = 0;
1920         resource_size_t end = 0;
1921         struct resource *new_res;
1922         struct resource **old_res = &hyperv_mmio;
1923         struct resource **prev_res = NULL;
1924
1925         switch (res->type) {
1926
1927         /*
1928          * "Address" descriptors are for bus windows. Ignore
1929          * "memory" descriptors, which are for registers on
1930          * devices.
1931          */
1932         case ACPI_RESOURCE_TYPE_ADDRESS32:
1933                 start = res->data.address32.address.minimum;
1934                 end = res->data.address32.address.maximum;
1935                 break;
1936
1937         case ACPI_RESOURCE_TYPE_ADDRESS64:
1938                 start = res->data.address64.address.minimum;
1939                 end = res->data.address64.address.maximum;
1940                 break;
1941
1942         default:
1943                 /* Unused resource type */
1944                 return AE_OK;
1945
1946         }
1947         /*
1948          * Ignore ranges that are below 1MB, as they're not
1949          * necessary or useful here.
1950          */
1951         if (end < 0x100000)
1952                 return AE_OK;
1953
1954         new_res = kzalloc(sizeof(*new_res), GFP_ATOMIC);
1955         if (!new_res)
1956                 return AE_NO_MEMORY;
1957
1958         /* If this range overlaps the virtual TPM, truncate it. */
1959         if (end > VTPM_BASE_ADDRESS && start < VTPM_BASE_ADDRESS)
1960                 end = VTPM_BASE_ADDRESS;
1961
1962         new_res->name = "hyperv mmio";
1963         new_res->flags = IORESOURCE_MEM;
1964         new_res->start = start;
1965         new_res->end = end;
1966
1967         /*
1968          * If two ranges are adjacent, merge them.
1969          */
1970         do {
1971                 if (!*old_res) {
1972                         *old_res = new_res;
1973                         break;
1974                 }
1975
1976                 if (((*old_res)->end + 1) == new_res->start) {
1977                         (*old_res)->end = new_res->end;
1978                         kfree(new_res);
1979                         break;
1980                 }
1981
1982                 if ((*old_res)->start == new_res->end + 1) {
1983                         (*old_res)->start = new_res->start;
1984                         kfree(new_res);
1985                         break;
1986                 }
1987
1988                 if ((*old_res)->start > new_res->end) {
1989                         new_res->sibling = *old_res;
1990                         if (prev_res)
1991                                 (*prev_res)->sibling = new_res;
1992                         *old_res = new_res;
1993                         break;
1994                 }
1995
1996                 prev_res = old_res;
1997                 old_res = &(*old_res)->sibling;
1998
1999         } while (1);
2000
2001         return AE_OK;
2002 }
2003
2004 static int vmbus_acpi_remove(struct acpi_device *device)
2005 {
2006         struct resource *cur_res;
2007         struct resource *next_res;
2008
2009         if (hyperv_mmio) {
2010                 if (fb_mmio) {
2011                         __release_region(hyperv_mmio, fb_mmio->start,
2012                                          resource_size(fb_mmio));
2013                         fb_mmio = NULL;
2014                 }
2015
2016                 for (cur_res = hyperv_mmio; cur_res; cur_res = next_res) {
2017                         next_res = cur_res->sibling;
2018                         kfree(cur_res);
2019                 }
2020         }
2021
2022         return 0;
2023 }
2024
2025 static void vmbus_reserve_fb(void)
2026 {
2027         int size;
2028         /*
2029          * Make a claim for the frame buffer in the resource tree under the
2030          * first node, which will be the one below 4GB.  The length seems to
2031          * be underreported, particularly in a Generation 1 VM.  So start out
2032          * reserving a larger area and make it smaller until it succeeds.
2033          */
2034
2035         if (screen_info.lfb_base) {
2036                 if (efi_enabled(EFI_BOOT))
2037                         size = max_t(__u32, screen_info.lfb_size, 0x800000);
2038                 else
2039                         size = max_t(__u32, screen_info.lfb_size, 0x4000000);
2040
2041                 for (; !fb_mmio && (size >= 0x100000); size >>= 1) {
2042                         fb_mmio = __request_region(hyperv_mmio,
2043                                                    screen_info.lfb_base, size,
2044                                                    fb_mmio_name, 0);
2045                 }
2046         }
2047 }
2048
2049 /**
2050  * vmbus_allocate_mmio() - Pick a memory-mapped I/O range.
2051  * @new:                If successful, supplied a pointer to the
2052  *                      allocated MMIO space.
2053  * @device_obj:         Identifies the caller
2054  * @min:                Minimum guest physical address of the
2055  *                      allocation
2056  * @max:                Maximum guest physical address
2057  * @size:               Size of the range to be allocated
2058  * @align:              Alignment of the range to be allocated
2059  * @fb_overlap_ok:      Whether this allocation can be allowed
2060  *                      to overlap the video frame buffer.
2061  *
2062  * This function walks the resources granted to VMBus by the
2063  * _CRS object in the ACPI namespace underneath the parent
2064  * "bridge" whether that's a root PCI bus in the Generation 1
2065  * case or a Module Device in the Generation 2 case.  It then
2066  * attempts to allocate from the global MMIO pool in a way that
2067  * matches the constraints supplied in these parameters and by
2068  * that _CRS.
2069  *
2070  * Return: 0 on success, -errno on failure
2071  */
2072 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
2073                         resource_size_t min, resource_size_t max,
2074                         resource_size_t size, resource_size_t align,
2075                         bool fb_overlap_ok)
2076 {
2077         struct resource *iter, *shadow;
2078         resource_size_t range_min, range_max, start;
2079         const char *dev_n = dev_name(&device_obj->device);
2080         int retval;
2081
2082         retval = -ENXIO;
2083         down(&hyperv_mmio_lock);
2084
2085         /*
2086          * If overlaps with frame buffers are allowed, then first attempt to
2087          * make the allocation from within the reserved region.  Because it
2088          * is already reserved, no shadow allocation is necessary.
2089          */
2090         if (fb_overlap_ok && fb_mmio && !(min > fb_mmio->end) &&
2091             !(max < fb_mmio->start)) {
2092
2093                 range_min = fb_mmio->start;
2094                 range_max = fb_mmio->end;
2095                 start = (range_min + align - 1) & ~(align - 1);
2096                 for (; start + size - 1 <= range_max; start += align) {
2097                         *new = request_mem_region_exclusive(start, size, dev_n);
2098                         if (*new) {
2099                                 retval = 0;
2100                                 goto exit;
2101                         }
2102                 }
2103         }
2104
2105         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
2106                 if ((iter->start >= max) || (iter->end <= min))
2107                         continue;
2108
2109                 range_min = iter->start;
2110                 range_max = iter->end;
2111                 start = (range_min + align - 1) & ~(align - 1);
2112                 for (; start + size - 1 <= range_max; start += align) {
2113                         shadow = __request_region(iter, start, size, NULL,
2114                                                   IORESOURCE_BUSY);
2115                         if (!shadow)
2116                                 continue;
2117
2118                         *new = request_mem_region_exclusive(start, size, dev_n);
2119                         if (*new) {
2120                                 shadow->name = (char *)*new;
2121                                 retval = 0;
2122                                 goto exit;
2123                         }
2124
2125                         __release_region(iter, start, size);
2126                 }
2127         }
2128
2129 exit:
2130         up(&hyperv_mmio_lock);
2131         return retval;
2132 }
2133 EXPORT_SYMBOL_GPL(vmbus_allocate_mmio);
2134
2135 /**
2136  * vmbus_free_mmio() - Free a memory-mapped I/O range.
2137  * @start:              Base address of region to release.
2138  * @size:               Size of the range to be allocated
2139  *
2140  * This function releases anything requested by
2141  * vmbus_mmio_allocate().
2142  */
2143 void vmbus_free_mmio(resource_size_t start, resource_size_t size)
2144 {
2145         struct resource *iter;
2146
2147         down(&hyperv_mmio_lock);
2148         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
2149                 if ((iter->start >= start + size) || (iter->end <= start))
2150                         continue;
2151
2152                 __release_region(iter, start, size);
2153         }
2154         release_mem_region(start, size);
2155         up(&hyperv_mmio_lock);
2156
2157 }
2158 EXPORT_SYMBOL_GPL(vmbus_free_mmio);
2159
2160 static int vmbus_acpi_add(struct acpi_device *device)
2161 {
2162         acpi_status result;
2163         int ret_val = -ENODEV;
2164         struct acpi_device *ancestor;
2165
2166         hv_acpi_dev = device;
2167
2168         result = acpi_walk_resources(device->handle, METHOD_NAME__CRS,
2169                                         vmbus_walk_resources, NULL);
2170
2171         if (ACPI_FAILURE(result))
2172                 goto acpi_walk_err;
2173         /*
2174          * Some ancestor of the vmbus acpi device (Gen1 or Gen2
2175          * firmware) is the VMOD that has the mmio ranges. Get that.
2176          */
2177         for (ancestor = device->parent; ancestor; ancestor = ancestor->parent) {
2178                 result = acpi_walk_resources(ancestor->handle, METHOD_NAME__CRS,
2179                                              vmbus_walk_resources, NULL);
2180
2181                 if (ACPI_FAILURE(result))
2182                         continue;
2183                 if (hyperv_mmio) {
2184                         vmbus_reserve_fb();
2185                         break;
2186                 }
2187         }
2188         ret_val = 0;
2189
2190 acpi_walk_err:
2191         complete(&probe_event);
2192         if (ret_val)
2193                 vmbus_acpi_remove(device);
2194         return ret_val;
2195 }
2196
2197 #ifdef CONFIG_PM_SLEEP
2198 static int vmbus_bus_suspend(struct device *dev)
2199 {
2200         struct vmbus_channel *channel, *sc;
2201         unsigned long flags;
2202
2203         while (atomic_read(&vmbus_connection.offer_in_progress) != 0) {
2204                 /*
2205                  * We wait here until the completion of any channel
2206                  * offers that are currently in progress.
2207                  */
2208                 msleep(1);
2209         }
2210
2211         mutex_lock(&vmbus_connection.channel_mutex);
2212         list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
2213                 if (!is_hvsock_channel(channel))
2214                         continue;
2215
2216                 vmbus_force_channel_rescinded(channel);
2217         }
2218         mutex_unlock(&vmbus_connection.channel_mutex);
2219
2220         /*
2221          * Wait until all the sub-channels and hv_sock channels have been
2222          * cleaned up. Sub-channels should be destroyed upon suspend, otherwise
2223          * they would conflict with the new sub-channels that will be created
2224          * in the resume path. hv_sock channels should also be destroyed, but
2225          * a hv_sock channel of an established hv_sock connection can not be
2226          * really destroyed since it may still be referenced by the userspace
2227          * application, so we just force the hv_sock channel to be rescinded
2228          * by vmbus_force_channel_rescinded(), and the userspace application
2229          * will thoroughly destroy the channel after hibernation.
2230          *
2231          * Note: the counter nr_chan_close_on_suspend may never go above 0 if
2232          * the VM has no sub-channel and hv_sock channel, e.g. a 1-vCPU VM.
2233          */
2234         if (atomic_read(&vmbus_connection.nr_chan_close_on_suspend) > 0)
2235                 wait_for_completion(&vmbus_connection.ready_for_suspend_event);
2236
2237         if (atomic_read(&vmbus_connection.nr_chan_fixup_on_resume) != 0) {
2238                 pr_err("Can not suspend due to a previous failed resuming\n");
2239                 return -EBUSY;
2240         }
2241
2242         mutex_lock(&vmbus_connection.channel_mutex);
2243
2244         list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
2245                 /*
2246                  * Invalidate the field. Upon resume, vmbus_onoffer() will fix
2247                  * up the field, and the other fields (if necessary).
2248                  */
2249                 channel->offermsg.child_relid = INVALID_RELID;
2250
2251                 if (is_hvsock_channel(channel)) {
2252                         if (!channel->rescind) {
2253                                 pr_err("hv_sock channel not rescinded!\n");
2254                                 WARN_ON_ONCE(1);
2255                         }
2256                         continue;
2257                 }
2258
2259                 spin_lock_irqsave(&channel->lock, flags);
2260                 list_for_each_entry(sc, &channel->sc_list, sc_list) {
2261                         pr_err("Sub-channel not deleted!\n");
2262                         WARN_ON_ONCE(1);
2263                 }
2264                 spin_unlock_irqrestore(&channel->lock, flags);
2265
2266                 atomic_inc(&vmbus_connection.nr_chan_fixup_on_resume);
2267         }
2268
2269         mutex_unlock(&vmbus_connection.channel_mutex);
2270
2271         vmbus_initiate_unload(false);
2272
2273         /* Reset the event for the next resume. */
2274         reinit_completion(&vmbus_connection.ready_for_resume_event);
2275
2276         return 0;
2277 }
2278
2279 static int vmbus_bus_resume(struct device *dev)
2280 {
2281         struct vmbus_channel_msginfo *msginfo;
2282         size_t msgsize;
2283         int ret;
2284
2285         /*
2286          * We only use the 'vmbus_proto_version', which was in use before
2287          * hibernation, to re-negotiate with the host.
2288          */
2289         if (vmbus_proto_version == VERSION_INVAL ||
2290             vmbus_proto_version == 0) {
2291                 pr_err("Invalid proto version = 0x%x\n", vmbus_proto_version);
2292                 return -EINVAL;
2293         }
2294
2295         msgsize = sizeof(*msginfo) +
2296                   sizeof(struct vmbus_channel_initiate_contact);
2297
2298         msginfo = kzalloc(msgsize, GFP_KERNEL);
2299
2300         if (msginfo == NULL)
2301                 return -ENOMEM;
2302
2303         ret = vmbus_negotiate_version(msginfo, vmbus_proto_version);
2304
2305         kfree(msginfo);
2306
2307         if (ret != 0)
2308                 return ret;
2309
2310         WARN_ON(atomic_read(&vmbus_connection.nr_chan_fixup_on_resume) == 0);
2311
2312         vmbus_request_offers();
2313
2314         if (wait_for_completion_timeout(
2315                 &vmbus_connection.ready_for_resume_event, 10 * HZ) == 0)
2316                 pr_err("Some vmbus device is missing after suspending?\n");
2317
2318         /* Reset the event for the next suspend. */
2319         reinit_completion(&vmbus_connection.ready_for_suspend_event);
2320
2321         return 0;
2322 }
2323 #else
2324 #define vmbus_bus_suspend NULL
2325 #define vmbus_bus_resume NULL
2326 #endif /* CONFIG_PM_SLEEP */
2327
2328 static const struct acpi_device_id vmbus_acpi_device_ids[] = {
2329         {"VMBUS", 0},
2330         {"VMBus", 0},
2331         {"", 0},
2332 };
2333 MODULE_DEVICE_TABLE(acpi, vmbus_acpi_device_ids);
2334
2335 /*
2336  * Note: we must use the "no_irq" ops, otherwise hibernation can not work with
2337  * PCI device assignment, because "pci_dev_pm_ops" uses the "noirq" ops: in
2338  * the resume path, the pci "noirq" restore op runs before "non-noirq" op (see
2339  * resume_target_kernel() -> dpm_resume_start(), and hibernation_restore() ->
2340  * dpm_resume_end()). This means vmbus_bus_resume() and the pci-hyperv's
2341  * resume callback must also run via the "noirq" ops.
2342  *
2343  * Set suspend_noirq/resume_noirq to NULL for Suspend-to-Idle: see the comment
2344  * earlier in this file before vmbus_pm.
2345  */
2346
2347 static const struct dev_pm_ops vmbus_bus_pm = {
2348         .suspend_noirq  = NULL,
2349         .resume_noirq   = NULL,
2350         .freeze_noirq   = vmbus_bus_suspend,
2351         .thaw_noirq     = vmbus_bus_resume,
2352         .poweroff_noirq = vmbus_bus_suspend,
2353         .restore_noirq  = vmbus_bus_resume
2354 };
2355
2356 static struct acpi_driver vmbus_acpi_driver = {
2357         .name = "vmbus",
2358         .ids = vmbus_acpi_device_ids,
2359         .ops = {
2360                 .add = vmbus_acpi_add,
2361                 .remove = vmbus_acpi_remove,
2362         },
2363         .drv.pm = &vmbus_bus_pm,
2364 };
2365
2366 static void hv_kexec_handler(void)
2367 {
2368         hv_stimer_global_cleanup();
2369         vmbus_initiate_unload(false);
2370         /* Make sure conn_state is set as hv_synic_cleanup checks for it */
2371         mb();
2372         cpuhp_remove_state(hyperv_cpuhp_online);
2373         hyperv_cleanup();
2374 };
2375
2376 static void hv_crash_handler(struct pt_regs *regs)
2377 {
2378         int cpu;
2379
2380         vmbus_initiate_unload(true);
2381         /*
2382          * In crash handler we can't schedule synic cleanup for all CPUs,
2383          * doing the cleanup for current CPU only. This should be sufficient
2384          * for kdump.
2385          */
2386         cpu = smp_processor_id();
2387         hv_stimer_cleanup(cpu);
2388         hv_synic_disable_regs(cpu);
2389         hyperv_cleanup();
2390 };
2391
2392 static int hv_synic_suspend(void)
2393 {
2394         /*
2395          * When we reach here, all the non-boot CPUs have been offlined, and
2396          * the stimers on them have been unbound in hv_synic_cleanup() ->
2397          * hv_stimer_cleanup() -> clockevents_unbind_device().
2398          *
2399          * hv_synic_suspend() only runs on CPU0 with interrupts disabled. Here
2400          * we do not unbind the stimer on CPU0 because: 1) it's unnecessary
2401          * because the interrupts remain disabled between syscore_suspend()
2402          * and syscore_resume(): see create_image() and resume_target_kernel();
2403          * 2) the stimer on CPU0 is automatically disabled later by
2404          * syscore_suspend() -> timekeeping_suspend() -> tick_suspend() -> ...
2405          * -> clockevents_shutdown() -> ... -> hv_ce_shutdown(); 3) a warning
2406          * would be triggered if we call clockevents_unbind_device(), which
2407          * may sleep, in an interrupts-disabled context. So, we intentionally
2408          * don't call hv_stimer_cleanup(0) here.
2409          */
2410
2411         hv_synic_disable_regs(0);
2412
2413         return 0;
2414 }
2415
2416 static void hv_synic_resume(void)
2417 {
2418         hv_synic_enable_regs(0);
2419
2420         /*
2421          * Note: we don't need to call hv_stimer_init(0), because the timer
2422          * on CPU0 is not unbound in hv_synic_suspend(), and the timer is
2423          * automatically re-enabled in timekeeping_resume().
2424          */
2425 }
2426
2427 /* The callbacks run only on CPU0, with irqs_disabled. */
2428 static struct syscore_ops hv_synic_syscore_ops = {
2429         .suspend = hv_synic_suspend,
2430         .resume = hv_synic_resume,
2431 };
2432
2433 static int __init hv_acpi_init(void)
2434 {
2435         int ret, t;
2436
2437         if (!hv_is_hyperv_initialized())
2438                 return -ENODEV;
2439
2440         init_completion(&probe_event);
2441
2442         /*
2443          * Get ACPI resources first.
2444          */
2445         ret = acpi_bus_register_driver(&vmbus_acpi_driver);
2446
2447         if (ret)
2448                 return ret;
2449
2450         t = wait_for_completion_timeout(&probe_event, 5*HZ);
2451         if (t == 0) {
2452                 ret = -ETIMEDOUT;
2453                 goto cleanup;
2454         }
2455
2456         ret = vmbus_bus_init();
2457         if (ret)
2458                 goto cleanup;
2459
2460         hv_setup_kexec_handler(hv_kexec_handler);
2461         hv_setup_crash_handler(hv_crash_handler);
2462
2463         register_syscore_ops(&hv_synic_syscore_ops);
2464
2465         return 0;
2466
2467 cleanup:
2468         acpi_bus_unregister_driver(&vmbus_acpi_driver);
2469         hv_acpi_dev = NULL;
2470         return ret;
2471 }
2472
2473 static void __exit vmbus_exit(void)
2474 {
2475         int cpu;
2476
2477         unregister_syscore_ops(&hv_synic_syscore_ops);
2478
2479         hv_remove_kexec_handler();
2480         hv_remove_crash_handler();
2481         vmbus_connection.conn_state = DISCONNECTED;
2482         hv_stimer_global_cleanup();
2483         vmbus_disconnect();
2484         hv_remove_vmbus_irq();
2485         for_each_online_cpu(cpu) {
2486                 struct hv_per_cpu_context *hv_cpu
2487                         = per_cpu_ptr(hv_context.cpu_context, cpu);
2488
2489                 tasklet_kill(&hv_cpu->msg_dpc);
2490         }
2491         vmbus_free_channels();
2492
2493         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
2494                 kmsg_dump_unregister(&hv_kmsg_dumper);
2495                 unregister_die_notifier(&hyperv_die_block);
2496         }
2497
2498         /*
2499          * The panic notifier is always registered, hence we should
2500          * also unconditionally unregister it here as well.
2501          */
2502         atomic_notifier_chain_unregister(&panic_notifier_list,
2503                                          &hyperv_panic_block);
2504
2505         free_page((unsigned long)hv_panic_page);
2506         unregister_sysctl_table(hv_ctl_table_hdr);
2507         hv_ctl_table_hdr = NULL;
2508         bus_unregister(&hv_bus);
2509
2510         cpuhp_remove_state(hyperv_cpuhp_online);
2511         hv_synic_free();
2512         acpi_bus_unregister_driver(&vmbus_acpi_driver);
2513 }
2514
2515
2516 MODULE_LICENSE("GPL");
2517 MODULE_DESCRIPTION("Microsoft Hyper-V VMBus Driver");
2518
2519 subsys_initcall(hv_acpi_init);
2520 module_exit(vmbus_exit);