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_MUTEX(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         hv_debug_rm_dev_dir(hv_dev);
 995
 996         mutex_lock(&vmbus_connection.channel_mutex);
 997         hv_process_channel_removal(channel);
 998         mutex_unlock(&vmbus_connection.channel_mutex);
 999         kfree(hv_dev);
1000 }
1001
1002 /*
1003  * Note: we must use the "noirq" ops: see the comment before vmbus_bus_pm.
1004  *
1005  * suspend_noirq/resume_noirq are set to NULL to support Suspend-to-Idle: we
1006  * shouldn't suspend the vmbus devices upon Suspend-to-Idle, otherwise there
1007  * is no way to wake up a Generation-2 VM.
1008  *
1009  * The other 4 ops are for hibernation.
1010  */
1011
1012 static const struct dev_pm_ops vmbus_pm = {
1013         .suspend_noirq  = NULL,
1014         .resume_noirq   = NULL,
1015         .freeze_noirq   = vmbus_suspend,
1016         .thaw_noirq     = vmbus_resume,
1017         .poweroff_noirq = vmbus_suspend,
1018         .restore_noirq  = vmbus_resume,
1019 };
1020
1021 /* The one and only one */
1022 static struct bus_type  hv_bus = {
1023         .name =         "vmbus",
1024         .match =                vmbus_match,
1025         .shutdown =             vmbus_shutdown,
1026         .remove =               vmbus_remove,
1027         .probe =                vmbus_probe,
1028         .uevent =               vmbus_uevent,
1029         .dev_groups =           vmbus_dev_groups,
1030         .drv_groups =           vmbus_drv_groups,
1031         .pm =                   &vmbus_pm,
1032 };
1033
1034 struct onmessage_work_context {
1035         struct work_struct work;
1036         struct hv_message msg;
1037 };
1038
1039 static void vmbus_onmessage_work(struct work_struct *work)
1040 {
1041         struct onmessage_work_context *ctx;
1042
1043         /* Do not process messages if we're in DISCONNECTED state */
1044         if (vmbus_connection.conn_state == DISCONNECTED)
1045                 return;
1046
1047         ctx = container_of(work, struct onmessage_work_context,
1048                            work);
1049         vmbus_onmessage(&ctx->msg);
1050         kfree(ctx);
1051 }
1052
1053 void vmbus_on_msg_dpc(unsigned long data)
1054 {
1055         struct hv_per_cpu_context *hv_cpu = (void *)data;
1056         void *page_addr = hv_cpu->synic_message_page;
1057         struct hv_message *msg = (struct hv_message *)page_addr +
1058                                   VMBUS_MESSAGE_SINT;
1059         struct vmbus_channel_message_header *hdr;
1060         const struct vmbus_channel_message_table_entry *entry;
1061         struct onmessage_work_context *ctx;
1062         u32 message_type = msg->header.message_type;
1063
1064         if (message_type == HVMSG_NONE)
1065                 /* no msg */
1066                 return;
1067
1068         hdr = (struct vmbus_channel_message_header *)msg->u.payload;
1069
1070         trace_vmbus_on_msg_dpc(hdr);
1071
1072         if (hdr->msgtype >= CHANNELMSG_COUNT) {
1073                 WARN_ONCE(1, "unknown msgtype=%d\n", hdr->msgtype);
1074                 goto msg_handled;
1075         }
1076
1077         entry = &channel_message_table[hdr->msgtype];
1078
1079         if (!entry->message_handler)
1080                 goto msg_handled;
1081
1082         if (entry->handler_type == VMHT_BLOCKING) {
1083                 ctx = kmalloc(sizeof(*ctx), GFP_ATOMIC);
1084                 if (ctx == NULL)
1085                         return;
1086
1087                 INIT_WORK(&ctx->work, vmbus_onmessage_work);
1088                 memcpy(&ctx->msg, msg, sizeof(*msg));
1089
1090                 /*
1091                  * The host can generate a rescind message while we
1092                  * may still be handling the original offer. We deal with
1093                  * this condition by ensuring the processing is done on the
1094                  * same CPU.
1095                  */
1096                 switch (hdr->msgtype) {
1097                 case CHANNELMSG_RESCIND_CHANNELOFFER:
1098                         /*
1099                          * If we are handling the rescind message;
1100                          * schedule the work on the global work queue.
1101                          */
1102                         schedule_work_on(vmbus_connection.connect_cpu,
1103                                          &ctx->work);
1104                         break;
1105
1106                 case CHANNELMSG_OFFERCHANNEL:
1107                         atomic_inc(&vmbus_connection.offer_in_progress);
1108                         queue_work_on(vmbus_connection.connect_cpu,
1109                                       vmbus_connection.work_queue,
1110                                       &ctx->work);
1111                         break;
1112
1113                 default:
1114                         queue_work(vmbus_connection.work_queue, &ctx->work);
1115                 }
1116         } else
1117                 entry->message_handler(hdr);
1118
1119 msg_handled:
1120         vmbus_signal_eom(msg, message_type);
1121 }
1122
1123 #ifdef CONFIG_PM_SLEEP
1124 /*
1125  * Fake RESCIND_CHANNEL messages to clean up hv_sock channels by force for
1126  * hibernation, because hv_sock connections can not persist across hibernation.
1127  */
1128 static void vmbus_force_channel_rescinded(struct vmbus_channel *channel)
1129 {
1130         struct onmessage_work_context *ctx;
1131         struct vmbus_channel_rescind_offer *rescind;
1132
1133         WARN_ON(!is_hvsock_channel(channel));
1134
1135         /*
1136          * sizeof(*ctx) is small and the allocation should really not fail,
1137          * otherwise the state of the hv_sock connections ends up in limbo.
1138          */
1139         ctx = kzalloc(sizeof(*ctx), GFP_KERNEL | __GFP_NOFAIL);
1140
1141         /*
1142          * So far, these are not really used by Linux. Just set them to the
1143          * reasonable values conforming to the definitions of the fields.
1144          */
1145         ctx->msg.header.message_type = 1;
1146         ctx->msg.header.payload_size = sizeof(*rescind);
1147
1148         /* These values are actually used by Linux. */
1149         rescind = (struct vmbus_channel_rescind_offer *)ctx->msg.u.payload;
1150         rescind->header.msgtype = CHANNELMSG_RESCIND_CHANNELOFFER;
1151         rescind->child_relid = channel->offermsg.child_relid;
1152
1153         INIT_WORK(&ctx->work, vmbus_onmessage_work);
1154
1155         queue_work_on(vmbus_connection.connect_cpu,
1156                       vmbus_connection.work_queue,
1157                       &ctx->work);
1158 }
1159 #endif /* CONFIG_PM_SLEEP */
1160
1161 /*
1162  * Direct callback for channels using other deferred processing
1163  */
1164 static void vmbus_channel_isr(struct vmbus_channel *channel)
1165 {
1166         void (*callback_fn)(void *);
1167
1168         callback_fn = READ_ONCE(channel->onchannel_callback);
1169         if (likely(callback_fn != NULL))
1170                 (*callback_fn)(channel->channel_callback_context);
1171 }
1172
1173 /*
1174  * Schedule all channels with events pending
1175  */
1176 static void vmbus_chan_sched(struct hv_per_cpu_context *hv_cpu)
1177 {
1178         unsigned long *recv_int_page;
1179         u32 maxbits, relid;
1180
1181         if (vmbus_proto_version < VERSION_WIN8) {
1182                 maxbits = MAX_NUM_CHANNELS_SUPPORTED;
1183                 recv_int_page = vmbus_connection.recv_int_page;
1184         } else {
1185                 /*
1186                  * When the host is win8 and beyond, the event page
1187                  * can be directly checked to get the id of the channel
1188                  * that has the interrupt pending.
1189                  */
1190                 void *page_addr = hv_cpu->synic_event_page;
1191                 union hv_synic_event_flags *event
1192                         = (union hv_synic_event_flags *)page_addr +
1193                                                  VMBUS_MESSAGE_SINT;
1194
1195                 maxbits = HV_EVENT_FLAGS_COUNT;
1196                 recv_int_page = event->flags;
1197         }
1198
1199         if (unlikely(!recv_int_page))
1200                 return;
1201
1202         for_each_set_bit(relid, recv_int_page, maxbits) {
1203                 struct vmbus_channel *channel;
1204
1205                 if (!sync_test_and_clear_bit(relid, recv_int_page))
1206                         continue;
1207
1208                 /* Special case - vmbus channel protocol msg */
1209                 if (relid == 0)
1210                         continue;
1211
1212                 rcu_read_lock();
1213
1214                 /* Find channel based on relid */
1215                 list_for_each_entry_rcu(channel, &hv_cpu->chan_list, percpu_list) {
1216                         if (channel->offermsg.child_relid != relid)
1217                                 continue;
1218
1219                         if (channel->rescind)
1220                                 continue;
1221
1222                         trace_vmbus_chan_sched(channel);
1223
1224                         ++channel->interrupts;
1225
1226                         switch (channel->callback_mode) {
1227                         case HV_CALL_ISR:
1228                                 vmbus_channel_isr(channel);
1229                                 break;
1230
1231                         case HV_CALL_BATCHED:
1232                                 hv_begin_read(&channel->inbound);
1233                                 /* fallthrough */
1234                         case HV_CALL_DIRECT:
1235                                 tasklet_schedule(&channel->callback_event);
1236                         }
1237                 }
1238
1239                 rcu_read_unlock();
1240         }
1241 }
1242
1243 static void vmbus_isr(void)
1244 {
1245         struct hv_per_cpu_context *hv_cpu
1246                 = this_cpu_ptr(hv_context.cpu_context);
1247         void *page_addr = hv_cpu->synic_event_page;
1248         struct hv_message *msg;
1249         union hv_synic_event_flags *event;
1250         bool handled = false;
1251
1252         if (unlikely(page_addr == NULL))
1253                 return;
1254
1255         event = (union hv_synic_event_flags *)page_addr +
1256                                          VMBUS_MESSAGE_SINT;
1257         /*
1258          * Check for events before checking for messages. This is the order
1259          * in which events and messages are checked in Windows guests on
1260          * Hyper-V, and the Windows team suggested we do the same.
1261          */
1262
1263         if ((vmbus_proto_version == VERSION_WS2008) ||
1264                 (vmbus_proto_version == VERSION_WIN7)) {
1265
1266                 /* Since we are a child, we only need to check bit 0 */
1267                 if (sync_test_and_clear_bit(0, event->flags))
1268                         handled = true;
1269         } else {
1270                 /*
1271                  * Our host is win8 or above. The signaling mechanism
1272                  * has changed and we can directly look at the event page.
1273                  * If bit n is set then we have an interrup on the channel
1274                  * whose id is n.
1275                  */
1276                 handled = true;
1277         }
1278
1279         if (handled)
1280                 vmbus_chan_sched(hv_cpu);
1281
1282         page_addr = hv_cpu->synic_message_page;
1283         msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
1284
1285         /* Check if there are actual msgs to be processed */
1286         if (msg->header.message_type != HVMSG_NONE) {
1287                 if (msg->header.message_type == HVMSG_TIMER_EXPIRED) {
1288                         hv_stimer0_isr();
1289                         vmbus_signal_eom(msg, HVMSG_TIMER_EXPIRED);
1290                 } else
1291                         tasklet_schedule(&hv_cpu->msg_dpc);
1292         }
1293
1294         add_interrupt_randomness(HYPERVISOR_CALLBACK_VECTOR, 0);
1295 }
1296
1297 /*
1298  * Callback from kmsg_dump. Grab as much as possible from the end of the kmsg
1299  * buffer and call into Hyper-V to transfer the data.
1300  */
1301 static void hv_kmsg_dump(struct kmsg_dumper *dumper,
1302                          enum kmsg_dump_reason reason)
1303 {
1304         size_t bytes_written;
1305         phys_addr_t panic_pa;
1306
1307         /* We are only interested in panics. */
1308         if ((reason != KMSG_DUMP_PANIC) || (!sysctl_record_panic_msg))
1309                 return;
1310
1311         panic_pa = virt_to_phys(hv_panic_page);
1312
1313         /*
1314          * Write dump contents to the page. No need to synchronize; panic should
1315          * be single-threaded.
1316          */
1317         kmsg_dump_get_buffer(dumper, true, hv_panic_page, HV_HYP_PAGE_SIZE,
1318                              &bytes_written);
1319         if (bytes_written)
1320                 hyperv_report_panic_msg(panic_pa, bytes_written);
1321 }
1322
1323 static struct kmsg_dumper hv_kmsg_dumper = {
1324         .dump = hv_kmsg_dump,
1325 };
1326
1327 static struct ctl_table_header *hv_ctl_table_hdr;
1328
1329 /*
1330  * sysctl option to allow the user to control whether kmsg data should be
1331  * reported to Hyper-V on panic.
1332  */
1333 static struct ctl_table hv_ctl_table[] = {
1334         {
1335                 .procname       = "hyperv_record_panic_msg",
1336                 .data           = &sysctl_record_panic_msg,
1337                 .maxlen         = sizeof(int),
1338                 .mode           = 0644,
1339                 .proc_handler   = proc_dointvec_minmax,
1340                 .extra1         = SYSCTL_ZERO,
1341                 .extra2         = SYSCTL_ONE
1342         },
1343         {}
1344 };
1345
1346 static struct ctl_table hv_root_table[] = {
1347         {
1348                 .procname       = "kernel",
1349                 .mode           = 0555,
1350                 .child          = hv_ctl_table
1351         },
1352         {}
1353 };
1354
1355 /*
1356  * vmbus_bus_init -Main vmbus driver initialization routine.
1357  *
1358  * Here, we
1359  *      - initialize the vmbus driver context
1360  *      - invoke the vmbus hv main init routine
1361  *      - retrieve the channel offers
1362  */
1363 static int vmbus_bus_init(void)
1364 {
1365         int ret;
1366
1367         /* Hypervisor initialization...setup hypercall page..etc */
1368         ret = hv_init();
1369         if (ret != 0) {
1370                 pr_err("Unable to initialize the hypervisor - 0x%x\n", ret);
1371                 return ret;
1372         }
1373
1374         ret = bus_register(&hv_bus);
1375         if (ret)
1376                 return ret;
1377
1378         hv_setup_vmbus_irq(vmbus_isr);
1379
1380         ret = hv_synic_alloc();
1381         if (ret)
1382                 goto err_alloc;
1383
1384         /*
1385          * Initialize the per-cpu interrupt state and stimer state.
1386          * Then connect to the host.
1387          */
1388         ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "hyperv/vmbus:online",
1389                                 hv_synic_init, hv_synic_cleanup);
1390         if (ret < 0)
1391                 goto err_cpuhp;
1392         hyperv_cpuhp_online = ret;
1393
1394         ret = vmbus_connect();
1395         if (ret)
1396                 goto err_connect;
1397
1398         /*
1399          * Only register if the crash MSRs are available
1400          */
1401         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1402                 u64 hyperv_crash_ctl;
1403                 /*
1404                  * Sysctl registration is not fatal, since by default
1405                  * reporting is enabled.
1406                  */
1407                 hv_ctl_table_hdr = register_sysctl_table(hv_root_table);
1408                 if (!hv_ctl_table_hdr)
1409                         pr_err("Hyper-V: sysctl table register error");
1410
1411                 /*
1412                  * Register for panic kmsg callback only if the right
1413                  * capability is supported by the hypervisor.
1414                  */
1415                 hv_get_crash_ctl(hyperv_crash_ctl);
1416                 if (hyperv_crash_ctl & HV_CRASH_CTL_CRASH_NOTIFY_MSG) {
1417                         hv_panic_page = (void *)hv_alloc_hyperv_zeroed_page();
1418                         if (hv_panic_page) {
1419                                 ret = kmsg_dump_register(&hv_kmsg_dumper);
1420                                 if (ret) {
1421                                         pr_err("Hyper-V: kmsg dump register "
1422                                                 "error 0x%x\n", ret);
1423                                         hv_free_hyperv_page(
1424                                             (unsigned long)hv_panic_page);
1425                                         hv_panic_page = NULL;
1426                                 }
1427                         } else
1428                                 pr_err("Hyper-V: panic message page memory "
1429                                         "allocation failed");
1430                 }
1431
1432                 register_die_notifier(&hyperv_die_block);
1433         }
1434
1435         /*
1436          * Always register the panic notifier because we need to unload
1437          * the VMbus channel connection to prevent any VMbus
1438          * activity after the VM panics.
1439          */
1440         atomic_notifier_chain_register(&panic_notifier_list,
1441                                &hyperv_panic_block);
1442
1443         vmbus_request_offers();
1444
1445         return 0;
1446
1447 err_connect:
1448         cpuhp_remove_state(hyperv_cpuhp_online);
1449 err_cpuhp:
1450         hv_synic_free();
1451 err_alloc:
1452         hv_remove_vmbus_irq();
1453
1454         bus_unregister(&hv_bus);
1455         unregister_sysctl_table(hv_ctl_table_hdr);
1456         hv_ctl_table_hdr = NULL;
1457         return ret;
1458 }
1459
1460 /**
1461  * __vmbus_child_driver_register() - Register a vmbus's driver
1462  * @hv_driver: Pointer to driver structure you want to register
1463  * @owner: owner module of the drv
1464  * @mod_name: module name string
1465  *
1466  * Registers the given driver with Linux through the 'driver_register()' call
1467  * and sets up the hyper-v vmbus handling for this driver.
1468  * It will return the state of the 'driver_register()' call.
1469  *
1470  */
1471 int __vmbus_driver_register(struct hv_driver *hv_driver, struct module *owner, const char *mod_name)
1472 {
1473         int ret;
1474
1475         pr_info("registering driver %s\n", hv_driver->name);
1476
1477         ret = vmbus_exists();
1478         if (ret < 0)
1479                 return ret;
1480
1481         hv_driver->driver.name = hv_driver->name;
1482         hv_driver->driver.owner = owner;
1483         hv_driver->driver.mod_name = mod_name;
1484         hv_driver->driver.bus = &hv_bus;
1485
1486         spin_lock_init(&hv_driver->dynids.lock);
1487         INIT_LIST_HEAD(&hv_driver->dynids.list);
1488
1489         ret = driver_register(&hv_driver->driver);
1490
1491         return ret;
1492 }
1493 EXPORT_SYMBOL_GPL(__vmbus_driver_register);
1494
1495 /**
1496  * vmbus_driver_unregister() - Unregister a vmbus's driver
1497  * @hv_driver: Pointer to driver structure you want to
1498  *             un-register
1499  *
1500  * Un-register the given driver that was previous registered with a call to
1501  * vmbus_driver_register()
1502  */
1503 void vmbus_driver_unregister(struct hv_driver *hv_driver)
1504 {
1505         pr_info("unregistering driver %s\n", hv_driver->name);
1506
1507         if (!vmbus_exists()) {
1508                 driver_unregister(&hv_driver->driver);
1509                 vmbus_free_dynids(hv_driver);
1510         }
1511 }
1512 EXPORT_SYMBOL_GPL(vmbus_driver_unregister);
1513
1514
1515 /*
1516  * Called when last reference to channel is gone.
1517  */
1518 static void vmbus_chan_release(struct kobject *kobj)
1519 {
1520         struct vmbus_channel *channel
1521                 = container_of(kobj, struct vmbus_channel, kobj);
1522
1523         kfree_rcu(channel, rcu);
1524 }
1525
1526 struct vmbus_chan_attribute {
1527         struct attribute attr;
1528         ssize_t (*show)(struct vmbus_channel *chan, char *buf);
1529         ssize_t (*store)(struct vmbus_channel *chan,
1530                          const char *buf, size_t count);
1531 };
1532 #define VMBUS_CHAN_ATTR(_name, _mode, _show, _store) \
1533         struct vmbus_chan_attribute chan_attr_##_name \
1534                 = __ATTR(_name, _mode, _show, _store)
1535 #define VMBUS_CHAN_ATTR_RW(_name) \
1536         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RW(_name)
1537 #define VMBUS_CHAN_ATTR_RO(_name) \
1538         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RO(_name)
1539 #define VMBUS_CHAN_ATTR_WO(_name) \
1540         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_WO(_name)
1541
1542 static ssize_t vmbus_chan_attr_show(struct kobject *kobj,
1543                                     struct attribute *attr, char *buf)
1544 {
1545         const struct vmbus_chan_attribute *attribute
1546                 = container_of(attr, struct vmbus_chan_attribute, attr);
1547         struct vmbus_channel *chan
1548                 = container_of(kobj, struct vmbus_channel, kobj);
1549
1550         if (!attribute->show)
1551                 return -EIO;
1552
1553         return attribute->show(chan, buf);
1554 }
1555
1556 static const struct sysfs_ops vmbus_chan_sysfs_ops = {
1557         .show = vmbus_chan_attr_show,
1558 };
1559
1560 static ssize_t out_mask_show(struct vmbus_channel *channel, char *buf)
1561 {
1562         struct hv_ring_buffer_info *rbi = &channel->outbound;
1563         ssize_t ret;
1564
1565         mutex_lock(&rbi->ring_buffer_mutex);
1566         if (!rbi->ring_buffer) {
1567                 mutex_unlock(&rbi->ring_buffer_mutex);
1568                 return -EINVAL;
1569         }
1570
1571         ret = sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1572         mutex_unlock(&rbi->ring_buffer_mutex);
1573         return ret;
1574 }
1575 static VMBUS_CHAN_ATTR_RO(out_mask);
1576
1577 static ssize_t in_mask_show(struct vmbus_channel *channel, char *buf)
1578 {
1579         struct hv_ring_buffer_info *rbi = &channel->inbound;
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(in_mask);
1593
1594 static ssize_t read_avail_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", hv_get_bytes_to_read(rbi));
1606         mutex_unlock(&rbi->ring_buffer_mutex);
1607         return ret;
1608 }
1609 static VMBUS_CHAN_ATTR_RO(read_avail);
1610
1611 static ssize_t write_avail_show(struct vmbus_channel *channel, char *buf)
1612 {
1613         struct hv_ring_buffer_info *rbi = &channel->outbound;
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_write(rbi));
1623         mutex_unlock(&rbi->ring_buffer_mutex);
1624         return ret;
1625 }
1626 static VMBUS_CHAN_ATTR_RO(write_avail);
1627
1628 static ssize_t show_target_cpu(struct vmbus_channel *channel, char *buf)
1629 {
1630         return sprintf(buf, "%u\n", channel->target_cpu);
1631 }
1632 static VMBUS_CHAN_ATTR(cpu, S_IRUGO, show_target_cpu, NULL);
1633
1634 static ssize_t channel_pending_show(struct vmbus_channel *channel,
1635                                     char *buf)
1636 {
1637         return sprintf(buf, "%d\n",
1638                        channel_pending(channel,
1639                                        vmbus_connection.monitor_pages[1]));
1640 }
1641 static VMBUS_CHAN_ATTR(pending, S_IRUGO, channel_pending_show, NULL);
1642
1643 static ssize_t channel_latency_show(struct vmbus_channel *channel,
1644                                     char *buf)
1645 {
1646         return sprintf(buf, "%d\n",
1647                        channel_latency(channel,
1648                                        vmbus_connection.monitor_pages[1]));
1649 }
1650 static VMBUS_CHAN_ATTR(latency, S_IRUGO, channel_latency_show, NULL);
1651
1652 static ssize_t channel_interrupts_show(struct vmbus_channel *channel, char *buf)
1653 {
1654         return sprintf(buf, "%llu\n", channel->interrupts);
1655 }
1656 static VMBUS_CHAN_ATTR(interrupts, S_IRUGO, channel_interrupts_show, NULL);
1657
1658 static ssize_t channel_events_show(struct vmbus_channel *channel, char *buf)
1659 {
1660         return sprintf(buf, "%llu\n", channel->sig_events);
1661 }
1662 static VMBUS_CHAN_ATTR(events, S_IRUGO, channel_events_show, NULL);
1663
1664 static ssize_t channel_intr_in_full_show(struct vmbus_channel *channel,
1665                                          char *buf)
1666 {
1667         return sprintf(buf, "%llu\n",
1668                        (unsigned long long)channel->intr_in_full);
1669 }
1670 static VMBUS_CHAN_ATTR(intr_in_full, 0444, channel_intr_in_full_show, NULL);
1671
1672 static ssize_t channel_intr_out_empty_show(struct vmbus_channel *channel,
1673                                            char *buf)
1674 {
1675         return sprintf(buf, "%llu\n",
1676                        (unsigned long long)channel->intr_out_empty);
1677 }
1678 static VMBUS_CHAN_ATTR(intr_out_empty, 0444, channel_intr_out_empty_show, NULL);
1679
1680 static ssize_t channel_out_full_first_show(struct vmbus_channel *channel,
1681                                            char *buf)
1682 {
1683         return sprintf(buf, "%llu\n",
1684                        (unsigned long long)channel->out_full_first);
1685 }
1686 static VMBUS_CHAN_ATTR(out_full_first, 0444, channel_out_full_first_show, NULL);
1687
1688 static ssize_t channel_out_full_total_show(struct vmbus_channel *channel,
1689                                            char *buf)
1690 {
1691         return sprintf(buf, "%llu\n",
1692                        (unsigned long long)channel->out_full_total);
1693 }
1694 static VMBUS_CHAN_ATTR(out_full_total, 0444, channel_out_full_total_show, NULL);
1695
1696 static ssize_t subchannel_monitor_id_show(struct vmbus_channel *channel,
1697                                           char *buf)
1698 {
1699         return sprintf(buf, "%u\n", channel->offermsg.monitorid);
1700 }
1701 static VMBUS_CHAN_ATTR(monitor_id, S_IRUGO, subchannel_monitor_id_show, NULL);
1702
1703 static ssize_t subchannel_id_show(struct vmbus_channel *channel,
1704                                   char *buf)
1705 {
1706         return sprintf(buf, "%u\n",
1707                        channel->offermsg.offer.sub_channel_index);
1708 }
1709 static VMBUS_CHAN_ATTR_RO(subchannel_id);
1710
1711 static struct attribute *vmbus_chan_attrs[] = {
1712         &chan_attr_out_mask.attr,
1713         &chan_attr_in_mask.attr,
1714         &chan_attr_read_avail.attr,
1715         &chan_attr_write_avail.attr,
1716         &chan_attr_cpu.attr,
1717         &chan_attr_pending.attr,
1718         &chan_attr_latency.attr,
1719         &chan_attr_interrupts.attr,
1720         &chan_attr_events.attr,
1721         &chan_attr_intr_in_full.attr,
1722         &chan_attr_intr_out_empty.attr,
1723         &chan_attr_out_full_first.attr,
1724         &chan_attr_out_full_total.attr,
1725         &chan_attr_monitor_id.attr,
1726         &chan_attr_subchannel_id.attr,
1727         NULL
1728 };
1729
1730 /*
1731  * Channel-level attribute_group callback function. Returns the permission for
1732  * each attribute, and returns 0 if an attribute is not visible.
1733  */
1734 static umode_t vmbus_chan_attr_is_visible(struct kobject *kobj,
1735                                           struct attribute *attr, int idx)
1736 {
1737         const struct vmbus_channel *channel =
1738                 container_of(kobj, struct vmbus_channel, kobj);
1739
1740         /* Hide the monitor attributes if the monitor mechanism is not used. */
1741         if (!channel->offermsg.monitor_allocated &&
1742             (attr == &chan_attr_pending.attr ||
1743              attr == &chan_attr_latency.attr ||
1744              attr == &chan_attr_monitor_id.attr))
1745                 return 0;
1746
1747         return attr->mode;
1748 }
1749
1750 static struct attribute_group vmbus_chan_group = {
1751         .attrs = vmbus_chan_attrs,
1752         .is_visible = vmbus_chan_attr_is_visible
1753 };
1754
1755 static struct kobj_type vmbus_chan_ktype = {
1756         .sysfs_ops = &vmbus_chan_sysfs_ops,
1757         .release = vmbus_chan_release,
1758 };
1759
1760 /*
1761  * vmbus_add_channel_kobj - setup a sub-directory under device/channels
1762  */
1763 int vmbus_add_channel_kobj(struct hv_device *dev, struct vmbus_channel *channel)
1764 {
1765         const struct device *device = &dev->device;
1766         struct kobject *kobj = &channel->kobj;
1767         u32 relid = channel->offermsg.child_relid;
1768         int ret;
1769
1770         kobj->kset = dev->channels_kset;
1771         ret = kobject_init_and_add(kobj, &vmbus_chan_ktype, NULL,
1772                                    "%u", relid);
1773         if (ret)
1774                 return ret;
1775
1776         ret = sysfs_create_group(kobj, &vmbus_chan_group);
1777
1778         if (ret) {
1779                 /*
1780                  * The calling functions' error handling paths will cleanup the
1781                  * empty channel directory.
1782                  */
1783                 dev_err(device, "Unable to set up channel sysfs files\n");
1784                 return ret;
1785         }
1786
1787         kobject_uevent(kobj, KOBJ_ADD);
1788
1789         return 0;
1790 }
1791
1792 /*
1793  * vmbus_remove_channel_attr_group - remove the channel's attribute group
1794  */
1795 void vmbus_remove_channel_attr_group(struct vmbus_channel *channel)
1796 {
1797         sysfs_remove_group(&channel->kobj, &vmbus_chan_group);
1798 }
1799
1800 /*
1801  * vmbus_device_create - Creates and registers a new child device
1802  * on the vmbus.
1803  */
1804 struct hv_device *vmbus_device_create(const guid_t *type,
1805                                       const guid_t *instance,
1806                                       struct vmbus_channel *channel)
1807 {
1808         struct hv_device *child_device_obj;
1809
1810         child_device_obj = kzalloc(sizeof(struct hv_device), GFP_KERNEL);
1811         if (!child_device_obj) {
1812                 pr_err("Unable to allocate device object for child device\n");
1813                 return NULL;
1814         }
1815
1816         child_device_obj->channel = channel;
1817         guid_copy(&child_device_obj->dev_type, type);
1818         guid_copy(&child_device_obj->dev_instance, instance);
1819         child_device_obj->vendor_id = 0x1414; /* MSFT vendor ID */
1820
1821         return child_device_obj;
1822 }
1823
1824 /*
1825  * vmbus_device_register - Register the child device
1826  */
1827 int vmbus_device_register(struct hv_device *child_device_obj)
1828 {
1829         struct kobject *kobj = &child_device_obj->device.kobj;
1830         int ret;
1831
1832         dev_set_name(&child_device_obj->device, "%pUl",
1833                      child_device_obj->channel->offermsg.offer.if_instance.b);
1834
1835         child_device_obj->device.bus = &hv_bus;
1836         child_device_obj->device.parent = &hv_acpi_dev->dev;
1837         child_device_obj->device.release = vmbus_device_release;
1838
1839         /*
1840          * Register with the LDM. This will kick off the driver/device
1841          * binding...which will eventually call vmbus_match() and vmbus_probe()
1842          */
1843         ret = device_register(&child_device_obj->device);
1844         if (ret) {
1845                 pr_err("Unable to register child device\n");
1846                 return ret;
1847         }
1848
1849         child_device_obj->channels_kset = kset_create_and_add("channels",
1850                                                               NULL, kobj);
1851         if (!child_device_obj->channels_kset) {
1852                 ret = -ENOMEM;
1853                 goto err_dev_unregister;
1854         }
1855
1856         ret = vmbus_add_channel_kobj(child_device_obj,
1857                                      child_device_obj->channel);
1858         if (ret) {
1859                 pr_err("Unable to register primary channeln");
1860                 goto err_kset_unregister;
1861         }
1862         hv_debug_add_dev_dir(child_device_obj);
1863
1864         return 0;
1865
1866 err_kset_unregister:
1867         kset_unregister(child_device_obj->channels_kset);
1868
1869 err_dev_unregister:
1870         device_unregister(&child_device_obj->device);
1871         return ret;
1872 }
1873
1874 /*
1875  * vmbus_device_unregister - Remove the specified child device
1876  * from the vmbus.
1877  */
1878 void vmbus_device_unregister(struct hv_device *device_obj)
1879 {
1880         pr_debug("child device %s unregistered\n",
1881                 dev_name(&device_obj->device));
1882
1883         kset_unregister(device_obj->channels_kset);
1884
1885         /*
1886          * Kick off the process of unregistering the device.
1887          * This will call vmbus_remove() and eventually vmbus_device_release()
1888          */
1889         device_unregister(&device_obj->device);
1890 }
1891
1892
1893 /*
1894  * VMBUS is an acpi enumerated device. Get the information we
1895  * need from DSDT.
1896  */
1897 #define VTPM_BASE_ADDRESS 0xfed40000
1898 static acpi_status vmbus_walk_resources(struct acpi_resource *res, void *ctx)
1899 {
1900         resource_size_t start = 0;
1901         resource_size_t end = 0;
1902         struct resource *new_res;
1903         struct resource **old_res = &hyperv_mmio;
1904         struct resource **prev_res = NULL;
1905
1906         switch (res->type) {
1907
1908         /*
1909          * "Address" descriptors are for bus windows. Ignore
1910          * "memory" descriptors, which are for registers on
1911          * devices.
1912          */
1913         case ACPI_RESOURCE_TYPE_ADDRESS32:
1914                 start = res->data.address32.address.minimum;
1915                 end = res->data.address32.address.maximum;
1916                 break;
1917
1918         case ACPI_RESOURCE_TYPE_ADDRESS64:
1919                 start = res->data.address64.address.minimum;
1920                 end = res->data.address64.address.maximum;
1921                 break;
1922
1923         default:
1924                 /* Unused resource type */
1925                 return AE_OK;
1926
1927         }
1928         /*
1929          * Ignore ranges that are below 1MB, as they're not
1930          * necessary or useful here.
1931          */
1932         if (end < 0x100000)
1933                 return AE_OK;
1934
1935         new_res = kzalloc(sizeof(*new_res), GFP_ATOMIC);
1936         if (!new_res)
1937                 return AE_NO_MEMORY;
1938
1939         /* If this range overlaps the virtual TPM, truncate it. */
1940         if (end > VTPM_BASE_ADDRESS && start < VTPM_BASE_ADDRESS)
1941                 end = VTPM_BASE_ADDRESS;
1942
1943         new_res->name = "hyperv mmio";
1944         new_res->flags = IORESOURCE_MEM;
1945         new_res->start = start;
1946         new_res->end = end;
1947
1948         /*
1949          * If two ranges are adjacent, merge them.
1950          */
1951         do {
1952                 if (!*old_res) {
1953                         *old_res = new_res;
1954                         break;
1955                 }
1956
1957                 if (((*old_res)->end + 1) == new_res->start) {
1958                         (*old_res)->end = new_res->end;
1959                         kfree(new_res);
1960                         break;
1961                 }
1962
1963                 if ((*old_res)->start == new_res->end + 1) {
1964                         (*old_res)->start = new_res->start;
1965                         kfree(new_res);
1966                         break;
1967                 }
1968
1969                 if ((*old_res)->start > new_res->end) {
1970                         new_res->sibling = *old_res;
1971                         if (prev_res)
1972                                 (*prev_res)->sibling = new_res;
1973                         *old_res = new_res;
1974                         break;
1975                 }
1976
1977                 prev_res = old_res;
1978                 old_res = &(*old_res)->sibling;
1979
1980         } while (1);
1981
1982         return AE_OK;
1983 }
1984
1985 static int vmbus_acpi_remove(struct acpi_device *device)
1986 {
1987         struct resource *cur_res;
1988         struct resource *next_res;
1989
1990         if (hyperv_mmio) {
1991                 if (fb_mmio) {
1992                         __release_region(hyperv_mmio, fb_mmio->start,
1993                                          resource_size(fb_mmio));
1994                         fb_mmio = NULL;
1995                 }
1996
1997                 for (cur_res = hyperv_mmio; cur_res; cur_res = next_res) {
1998                         next_res = cur_res->sibling;
1999                         kfree(cur_res);
2000                 }
2001         }
2002
2003         return 0;
2004 }
2005
2006 static void vmbus_reserve_fb(void)
2007 {
2008         int size;
2009         /*
2010          * Make a claim for the frame buffer in the resource tree under the
2011          * first node, which will be the one below 4GB.  The length seems to
2012          * be underreported, particularly in a Generation 1 VM.  So start out
2013          * reserving a larger area and make it smaller until it succeeds.
2014          */
2015
2016         if (screen_info.lfb_base) {
2017                 if (efi_enabled(EFI_BOOT))
2018                         size = max_t(__u32, screen_info.lfb_size, 0x800000);
2019                 else
2020                         size = max_t(__u32, screen_info.lfb_size, 0x4000000);
2021
2022                 for (; !fb_mmio && (size >= 0x100000); size >>= 1) {
2023                         fb_mmio = __request_region(hyperv_mmio,
2024                                                    screen_info.lfb_base, size,
2025                                                    fb_mmio_name, 0);
2026                 }
2027         }
2028 }
2029
2030 /**
2031  * vmbus_allocate_mmio() - Pick a memory-mapped I/O range.
2032  * @new:                If successful, supplied a pointer to the
2033  *                      allocated MMIO space.
2034  * @device_obj:         Identifies the caller
2035  * @min:                Minimum guest physical address of the
2036  *                      allocation
2037  * @max:                Maximum guest physical address
2038  * @size:               Size of the range to be allocated
2039  * @align:              Alignment of the range to be allocated
2040  * @fb_overlap_ok:      Whether this allocation can be allowed
2041  *                      to overlap the video frame buffer.
2042  *
2043  * This function walks the resources granted to VMBus by the
2044  * _CRS object in the ACPI namespace underneath the parent
2045  * "bridge" whether that's a root PCI bus in the Generation 1
2046  * case or a Module Device in the Generation 2 case.  It then
2047  * attempts to allocate from the global MMIO pool in a way that
2048  * matches the constraints supplied in these parameters and by
2049  * that _CRS.
2050  *
2051  * Return: 0 on success, -errno on failure
2052  */
2053 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
2054                         resource_size_t min, resource_size_t max,
2055                         resource_size_t size, resource_size_t align,
2056                         bool fb_overlap_ok)
2057 {
2058         struct resource *iter, *shadow;
2059         resource_size_t range_min, range_max, start;
2060         const char *dev_n = dev_name(&device_obj->device);
2061         int retval;
2062
2063         retval = -ENXIO;
2064         mutex_lock(&hyperv_mmio_lock);
2065
2066         /*
2067          * If overlaps with frame buffers are allowed, then first attempt to
2068          * make the allocation from within the reserved region.  Because it
2069          * is already reserved, no shadow allocation is necessary.
2070          */
2071         if (fb_overlap_ok && fb_mmio && !(min > fb_mmio->end) &&
2072             !(max < fb_mmio->start)) {
2073
2074                 range_min = fb_mmio->start;
2075                 range_max = fb_mmio->end;
2076                 start = (range_min + align - 1) & ~(align - 1);
2077                 for (; start + size - 1 <= range_max; start += align) {
2078                         *new = request_mem_region_exclusive(start, size, dev_n);
2079                         if (*new) {
2080                                 retval = 0;
2081                                 goto exit;
2082                         }
2083                 }
2084         }
2085
2086         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
2087                 if ((iter->start >= max) || (iter->end <= min))
2088                         continue;
2089
2090                 range_min = iter->start;
2091                 range_max = iter->end;
2092                 start = (range_min + align - 1) & ~(align - 1);
2093                 for (; start + size - 1 <= range_max; start += align) {
2094                         shadow = __request_region(iter, start, size, NULL,
2095                                                   IORESOURCE_BUSY);
2096                         if (!shadow)
2097                                 continue;
2098
2099                         *new = request_mem_region_exclusive(start, size, dev_n);
2100                         if (*new) {
2101                                 shadow->name = (char *)*new;
2102                                 retval = 0;
2103                                 goto exit;
2104                         }
2105
2106                         __release_region(iter, start, size);
2107                 }
2108         }
2109
2110 exit:
2111         mutex_unlock(&hyperv_mmio_lock);
2112         return retval;
2113 }
2114 EXPORT_SYMBOL_GPL(vmbus_allocate_mmio);
2115
2116 /**
2117  * vmbus_free_mmio() - Free a memory-mapped I/O range.
2118  * @start:              Base address of region to release.
2119  * @size:               Size of the range to be allocated
2120  *
2121  * This function releases anything requested by
2122  * vmbus_mmio_allocate().
2123  */
2124 void vmbus_free_mmio(resource_size_t start, resource_size_t size)
2125 {
2126         struct resource *iter;
2127
2128         mutex_lock(&hyperv_mmio_lock);
2129         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
2130                 if ((iter->start >= start + size) || (iter->end <= start))
2131                         continue;
2132
2133                 __release_region(iter, start, size);
2134         }
2135         release_mem_region(start, size);
2136         mutex_unlock(&hyperv_mmio_lock);
2137
2138 }
2139 EXPORT_SYMBOL_GPL(vmbus_free_mmio);
2140
2141 static int vmbus_acpi_add(struct acpi_device *device)
2142 {
2143         acpi_status result;
2144         int ret_val = -ENODEV;
2145         struct acpi_device *ancestor;
2146
2147         hv_acpi_dev = device;
2148
2149         result = acpi_walk_resources(device->handle, METHOD_NAME__CRS,
2150                                         vmbus_walk_resources, NULL);
2151
2152         if (ACPI_FAILURE(result))
2153                 goto acpi_walk_err;
2154         /*
2155          * Some ancestor of the vmbus acpi device (Gen1 or Gen2
2156          * firmware) is the VMOD that has the mmio ranges. Get that.
2157          */
2158         for (ancestor = device->parent; ancestor; ancestor = ancestor->parent) {
2159                 result = acpi_walk_resources(ancestor->handle, METHOD_NAME__CRS,
2160                                              vmbus_walk_resources, NULL);
2161
2162                 if (ACPI_FAILURE(result))
2163                         continue;
2164                 if (hyperv_mmio) {
2165                         vmbus_reserve_fb();
2166                         break;
2167                 }
2168         }
2169         ret_val = 0;
2170
2171 acpi_walk_err:
2172         complete(&probe_event);
2173         if (ret_val)
2174                 vmbus_acpi_remove(device);
2175         return ret_val;
2176 }
2177
2178 #ifdef CONFIG_PM_SLEEP
2179 static int vmbus_bus_suspend(struct device *dev)
2180 {
2181         struct vmbus_channel *channel, *sc;
2182         unsigned long flags;
2183
2184         while (atomic_read(&vmbus_connection.offer_in_progress) != 0) {
2185                 /*
2186                  * We wait here until the completion of any channel
2187                  * offers that are currently in progress.
2188                  */
2189                 msleep(1);
2190         }
2191
2192         mutex_lock(&vmbus_connection.channel_mutex);
2193         list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
2194                 if (!is_hvsock_channel(channel))
2195                         continue;
2196
2197                 vmbus_force_channel_rescinded(channel);
2198         }
2199         mutex_unlock(&vmbus_connection.channel_mutex);
2200
2201         /*
2202          * Wait until all the sub-channels and hv_sock channels have been
2203          * cleaned up. Sub-channels should be destroyed upon suspend, otherwise
2204          * they would conflict with the new sub-channels that will be created
2205          * in the resume path. hv_sock channels should also be destroyed, but
2206          * a hv_sock channel of an established hv_sock connection can not be
2207          * really destroyed since it may still be referenced by the userspace
2208          * application, so we just force the hv_sock channel to be rescinded
2209          * by vmbus_force_channel_rescinded(), and the userspace application
2210          * will thoroughly destroy the channel after hibernation.
2211          *
2212          * Note: the counter nr_chan_close_on_suspend may never go above 0 if
2213          * the VM has no sub-channel and hv_sock channel, e.g. a 1-vCPU VM.
2214          */
2215         if (atomic_read(&vmbus_connection.nr_chan_close_on_suspend) > 0)
2216                 wait_for_completion(&vmbus_connection.ready_for_suspend_event);
2217
2218         WARN_ON(atomic_read(&vmbus_connection.nr_chan_fixup_on_resume) != 0);
2219
2220         mutex_lock(&vmbus_connection.channel_mutex);
2221
2222         list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
2223                 /*
2224                  * Invalidate the field. Upon resume, vmbus_onoffer() will fix
2225                  * up the field, and the other fields (if necessary).
2226                  */
2227                 channel->offermsg.child_relid = INVALID_RELID;
2228
2229                 if (is_hvsock_channel(channel)) {
2230                         if (!channel->rescind) {
2231                                 pr_err("hv_sock channel not rescinded!\n");
2232                                 WARN_ON_ONCE(1);
2233                         }
2234                         continue;
2235                 }
2236
2237                 spin_lock_irqsave(&channel->lock, flags);
2238                 list_for_each_entry(sc, &channel->sc_list, sc_list) {
2239                         pr_err("Sub-channel not deleted!\n");
2240                         WARN_ON_ONCE(1);
2241                 }
2242                 spin_unlock_irqrestore(&channel->lock, flags);
2243
2244                 atomic_inc(&vmbus_connection.nr_chan_fixup_on_resume);
2245         }
2246
2247         mutex_unlock(&vmbus_connection.channel_mutex);
2248
2249         vmbus_initiate_unload(false);
2250
2251         /* Reset the event for the next resume. */
2252         reinit_completion(&vmbus_connection.ready_for_resume_event);
2253
2254         return 0;
2255 }
2256
2257 static int vmbus_bus_resume(struct device *dev)
2258 {
2259         struct vmbus_channel_msginfo *msginfo;
2260         size_t msgsize;
2261         int ret;
2262
2263         /*
2264          * We only use the 'vmbus_proto_version', which was in use before
2265          * hibernation, to re-negotiate with the host.
2266          */
2267         if (!vmbus_proto_version) {
2268                 pr_err("Invalid proto version = 0x%x\n", vmbus_proto_version);
2269                 return -EINVAL;
2270         }
2271
2272         msgsize = sizeof(*msginfo) +
2273                   sizeof(struct vmbus_channel_initiate_contact);
2274
2275         msginfo = kzalloc(msgsize, GFP_KERNEL);
2276
2277         if (msginfo == NULL)
2278                 return -ENOMEM;
2279
2280         ret = vmbus_negotiate_version(msginfo, vmbus_proto_version);
2281
2282         kfree(msginfo);
2283
2284         if (ret != 0)
2285                 return ret;
2286
2287         WARN_ON(atomic_read(&vmbus_connection.nr_chan_fixup_on_resume) == 0);
2288
2289         vmbus_request_offers();
2290
2291         wait_for_completion(&vmbus_connection.ready_for_resume_event);
2292
2293         /* Reset the event for the next suspend. */
2294         reinit_completion(&vmbus_connection.ready_for_suspend_event);
2295
2296         return 0;
2297 }
2298 #else
2299 #define vmbus_bus_suspend NULL
2300 #define vmbus_bus_resume NULL
2301 #endif /* CONFIG_PM_SLEEP */
2302
2303 static const struct acpi_device_id vmbus_acpi_device_ids[] = {
2304         {"VMBUS", 0},
2305         {"VMBus", 0},
2306         {"", 0},
2307 };
2308 MODULE_DEVICE_TABLE(acpi, vmbus_acpi_device_ids);
2309
2310 /*
2311  * Note: we must use the "no_irq" ops, otherwise hibernation can not work with
2312  * PCI device assignment, because "pci_dev_pm_ops" uses the "noirq" ops: in
2313  * the resume path, the pci "noirq" restore op runs before "non-noirq" op (see
2314  * resume_target_kernel() -> dpm_resume_start(), and hibernation_restore() ->
2315  * dpm_resume_end()). This means vmbus_bus_resume() and the pci-hyperv's
2316  * resume callback must also run via the "noirq" ops.
2317  *
2318  * Set suspend_noirq/resume_noirq to NULL for Suspend-to-Idle: see the comment
2319  * earlier in this file before vmbus_pm.
2320  */
2321
2322 static const struct dev_pm_ops vmbus_bus_pm = {
2323         .suspend_noirq  = NULL,
2324         .resume_noirq   = NULL,
2325         .freeze_noirq   = vmbus_bus_suspend,
2326         .thaw_noirq     = vmbus_bus_resume,
2327         .poweroff_noirq = vmbus_bus_suspend,
2328         .restore_noirq  = vmbus_bus_resume
2329 };
2330
2331 static struct acpi_driver vmbus_acpi_driver = {
2332         .name = "vmbus",
2333         .ids = vmbus_acpi_device_ids,
2334         .ops = {
2335                 .add = vmbus_acpi_add,
2336                 .remove = vmbus_acpi_remove,
2337         },
2338         .drv.pm = &vmbus_bus_pm,
2339 };
2340
2341 static void hv_kexec_handler(void)
2342 {
2343         hv_stimer_global_cleanup();
2344         vmbus_initiate_unload(false);
2345         /* Make sure conn_state is set as hv_synic_cleanup checks for it */
2346         mb();
2347         cpuhp_remove_state(hyperv_cpuhp_online);
2348         hyperv_cleanup();
2349 };
2350
2351 static void hv_crash_handler(struct pt_regs *regs)
2352 {
2353         int cpu;
2354
2355         vmbus_initiate_unload(true);
2356         /*
2357          * In crash handler we can't schedule synic cleanup for all CPUs,
2358          * doing the cleanup for current CPU only. This should be sufficient
2359          * for kdump.
2360          */
2361         cpu = smp_processor_id();
2362         hv_stimer_cleanup(cpu);
2363         hv_synic_disable_regs(cpu);
2364         hyperv_cleanup();
2365 };
2366
2367 static int hv_synic_suspend(void)
2368 {
2369         /*
2370          * When we reach here, all the non-boot CPUs have been offlined.
2371          * If we're in a legacy configuration where stimer Direct Mode is
2372          * not enabled, the stimers on the non-boot CPUs have been unbound
2373          * in hv_synic_cleanup() -> hv_stimer_legacy_cleanup() ->
2374          * hv_stimer_cleanup() -> clockevents_unbind_device().
2375          *
2376          * hv_synic_suspend() only runs on CPU0 with interrupts disabled.
2377          * Here we do not call hv_stimer_legacy_cleanup() on CPU0 because:
2378          * 1) it's unnecessary as interrupts remain disabled between
2379          * syscore_suspend() and syscore_resume(): see create_image() and
2380          * resume_target_kernel()
2381          * 2) the stimer on CPU0 is automatically disabled later by
2382          * syscore_suspend() -> timekeeping_suspend() -> tick_suspend() -> ...
2383          * -> clockevents_shutdown() -> ... -> hv_ce_shutdown()
2384          * 3) a warning would be triggered if we call
2385          * clockevents_unbind_device(), which may sleep, in an
2386          * interrupts-disabled context.
2387          */
2388
2389         hv_synic_disable_regs(0);
2390
2391         return 0;
2392 }
2393
2394 static void hv_synic_resume(void)
2395 {
2396         hv_synic_enable_regs(0);
2397
2398         /*
2399          * Note: we don't need to call hv_stimer_init(0), because the timer
2400          * on CPU0 is not unbound in hv_synic_suspend(), and the timer is
2401          * automatically re-enabled in timekeeping_resume().
2402          */
2403 }
2404
2405 /* The callbacks run only on CPU0, with irqs_disabled. */
2406 static struct syscore_ops hv_synic_syscore_ops = {
2407         .suspend = hv_synic_suspend,
2408         .resume = hv_synic_resume,
2409 };
2410
2411 static int __init hv_acpi_init(void)
2412 {
2413         int ret, t;
2414
2415         if (!hv_is_hyperv_initialized())
2416                 return -ENODEV;
2417
2418         init_completion(&probe_event);
2419
2420         /*
2421          * Get ACPI resources first.
2422          */
2423         ret = acpi_bus_register_driver(&vmbus_acpi_driver);
2424
2425         if (ret)
2426                 return ret;
2427
2428         t = wait_for_completion_timeout(&probe_event, 5*HZ);
2429         if (t == 0) {
2430                 ret = -ETIMEDOUT;
2431                 goto cleanup;
2432         }
2433         hv_debug_init();
2434
2435         ret = vmbus_bus_init();
2436         if (ret)
2437                 goto cleanup;
2438
2439         hv_setup_kexec_handler(hv_kexec_handler);
2440         hv_setup_crash_handler(hv_crash_handler);
2441
2442         register_syscore_ops(&hv_synic_syscore_ops);
2443
2444         return 0;
2445
2446 cleanup:
2447         acpi_bus_unregister_driver(&vmbus_acpi_driver);
2448         hv_acpi_dev = NULL;
2449         return ret;
2450 }
2451
2452 static void __exit vmbus_exit(void)
2453 {
2454         int cpu;
2455
2456         unregister_syscore_ops(&hv_synic_syscore_ops);
2457
2458         hv_remove_kexec_handler();
2459         hv_remove_crash_handler();
2460         vmbus_connection.conn_state = DISCONNECTED;
2461         hv_stimer_global_cleanup();
2462         vmbus_disconnect();
2463         hv_remove_vmbus_irq();
2464         for_each_online_cpu(cpu) {
2465                 struct hv_per_cpu_context *hv_cpu
2466                         = per_cpu_ptr(hv_context.cpu_context, cpu);
2467
2468                 tasklet_kill(&hv_cpu->msg_dpc);
2469         }
2470         hv_debug_rm_all_dir();
2471
2472         vmbus_free_channels();
2473
2474         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
2475                 kmsg_dump_unregister(&hv_kmsg_dumper);
2476                 unregister_die_notifier(&hyperv_die_block);
2477                 atomic_notifier_chain_unregister(&panic_notifier_list,
2478                                                  &hyperv_panic_block);
2479         }
2480
2481         free_page((unsigned long)hv_panic_page);
2482         unregister_sysctl_table(hv_ctl_table_hdr);
2483         hv_ctl_table_hdr = NULL;
2484         bus_unregister(&hv_bus);
2485
2486         cpuhp_remove_state(hyperv_cpuhp_online);
2487         hv_synic_free();
2488         acpi_bus_unregister_driver(&vmbus_acpi_driver);
2489 }
2490
2491
2492 MODULE_LICENSE("GPL");
2493 MODULE_DESCRIPTION("Microsoft Hyper-V VMBus Driver");
2494
2495 subsys_initcall(hv_acpi_init);
2496 module_exit(vmbus_exit);