drivers/hv/vmbus_drv.c

   1 /*
   2  * Copyright (c) 2009, Microsoft Corporation.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms and conditions of the GNU General Public License,
   6  * version 2, as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  11  * more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along with
  14  * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
  15  * Place - Suite 330, Boston, MA 02111-1307 USA.
  16  *
  17  * Authors:
  18  *   Haiyang Zhang <haiyangz@microsoft.com>
  19  *   Hank Janssen  <hjanssen@microsoft.com>
  20  *   K. Y. Srinivasan <kys@microsoft.com>
  21  *
  22  */
  23 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  24
  25 #include <linux/init.h>
  26 #include <linux/module.h>
  27 #include <linux/device.h>
  28 #include <linux/interrupt.h>
  29 #include <linux/sysctl.h>
  30 #include <linux/slab.h>
  31 #include <linux/acpi.h>
  32 #include <linux/completion.h>
  33 #include <linux/hyperv.h>
  34 #include <linux/kernel_stat.h>
  35 #include <linux/clockchips.h>
  36 #include <linux/cpu.h>
  37 #include <linux/sched/task_stack.h>
  38
  39 #include <asm/hyperv.h>
  40 #include <asm/hypervisor.h>
  41 #include <asm/mshyperv.h>
  42 #include <linux/notifier.h>
  43 #include <linux/ptrace.h>
  44 #include <linux/screen_info.h>
  45 #include <linux/kdebug.h>
  46 #include <linux/efi.h>
  47 #include <linux/random.h>
  48 #include "hyperv_vmbus.h"
  49
  50 struct vmbus_dynid {
  51         struct list_head node;
  52         struct hv_vmbus_device_id id;
  53 };
  54
  55 static struct acpi_device  *hv_acpi_dev;
  56
  57 static struct completion probe_event;
  58
  59 static int hyperv_cpuhp_online;
  60
  61 static int hyperv_panic_event(struct notifier_block *nb, unsigned long val,
  62                               void *args)
  63 {
  64         struct pt_regs *regs;
  65
  66         regs = current_pt_regs();
  67
  68         hyperv_report_panic(regs);
  69         return NOTIFY_DONE;
  70 }
  71
  72 static int hyperv_die_event(struct notifier_block *nb, unsigned long val,
  73                             void *args)
  74 {
  75         struct die_args *die = (struct die_args *)args;
  76         struct pt_regs *regs = die->regs;
  77
  78         hyperv_report_panic(regs);
  79         return NOTIFY_DONE;
  80 }
  81
  82 static struct notifier_block hyperv_die_block = {
  83         .notifier_call = hyperv_die_event,
  84 };
  85 static struct notifier_block hyperv_panic_block = {
  86         .notifier_call = hyperv_panic_event,
  87 };
  88
  89 static const char *fb_mmio_name = "fb_range";
  90 static struct resource *fb_mmio;
  91 static struct resource *hyperv_mmio;
  92 static DEFINE_SEMAPHORE(hyperv_mmio_lock);
  93
  94 static int vmbus_exists(void)
  95 {
  96         if (hv_acpi_dev == NULL)
  97                 return -ENODEV;
  98
  99         return 0;
 100 }
 101
 102 #define VMBUS_ALIAS_LEN ((sizeof((struct hv_vmbus_device_id *)0)->guid) * 2)
 103 static void print_alias_name(struct hv_device *hv_dev, char *alias_name)
 104 {
 105         int i;
 106         for (i = 0; i < VMBUS_ALIAS_LEN; i += 2)
 107                 sprintf(&alias_name[i], "%02x", hv_dev->dev_type.b[i/2]);
 108 }
 109
 110 static u8 channel_monitor_group(struct vmbus_channel *channel)
 111 {
 112         return (u8)channel->offermsg.monitorid / 32;
 113 }
 114
 115 static u8 channel_monitor_offset(struct vmbus_channel *channel)
 116 {
 117         return (u8)channel->offermsg.monitorid % 32;
 118 }
 119
 120 static u32 channel_pending(struct vmbus_channel *channel,
 121                            struct hv_monitor_page *monitor_page)
 122 {
 123         u8 monitor_group = channel_monitor_group(channel);
 124         return monitor_page->trigger_group[monitor_group].pending;
 125 }
 126
 127 static u32 channel_latency(struct vmbus_channel *channel,
 128                            struct hv_monitor_page *monitor_page)
 129 {
 130         u8 monitor_group = channel_monitor_group(channel);
 131         u8 monitor_offset = channel_monitor_offset(channel);
 132         return monitor_page->latency[monitor_group][monitor_offset];
 133 }
 134
 135 static u32 channel_conn_id(struct vmbus_channel *channel,
 136                            struct hv_monitor_page *monitor_page)
 137 {
 138         u8 monitor_group = channel_monitor_group(channel);
 139         u8 monitor_offset = channel_monitor_offset(channel);
 140         return monitor_page->parameter[monitor_group][monitor_offset].connectionid.u.id;
 141 }
 142
 143 static ssize_t id_show(struct device *dev, struct device_attribute *dev_attr,
 144                        char *buf)
 145 {
 146         struct hv_device *hv_dev = device_to_hv_device(dev);
 147
 148         if (!hv_dev->channel)
 149                 return -ENODEV;
 150         return sprintf(buf, "%d\n", hv_dev->channel->offermsg.child_relid);
 151 }
 152 static DEVICE_ATTR_RO(id);
 153
 154 static ssize_t state_show(struct device *dev, struct device_attribute *dev_attr,
 155                           char *buf)
 156 {
 157         struct hv_device *hv_dev = device_to_hv_device(dev);
 158
 159         if (!hv_dev->channel)
 160                 return -ENODEV;
 161         return sprintf(buf, "%d\n", hv_dev->channel->state);
 162 }
 163 static DEVICE_ATTR_RO(state);
 164
 165 static ssize_t monitor_id_show(struct device *dev,
 166                                struct device_attribute *dev_attr, char *buf)
 167 {
 168         struct hv_device *hv_dev = device_to_hv_device(dev);
 169
 170         if (!hv_dev->channel)
 171                 return -ENODEV;
 172         return sprintf(buf, "%d\n", hv_dev->channel->offermsg.monitorid);
 173 }
 174 static DEVICE_ATTR_RO(monitor_id);
 175
 176 static ssize_t class_id_show(struct device *dev,
 177                                struct device_attribute *dev_attr, char *buf)
 178 {
 179         struct hv_device *hv_dev = device_to_hv_device(dev);
 180
 181         if (!hv_dev->channel)
 182                 return -ENODEV;
 183         return sprintf(buf, "{%pUl}\n",
 184                        hv_dev->channel->offermsg.offer.if_type.b);
 185 }
 186 static DEVICE_ATTR_RO(class_id);
 187
 188 static ssize_t device_id_show(struct device *dev,
 189                               struct device_attribute *dev_attr, char *buf)
 190 {
 191         struct hv_device *hv_dev = device_to_hv_device(dev);
 192
 193         if (!hv_dev->channel)
 194                 return -ENODEV;
 195         return sprintf(buf, "{%pUl}\n",
 196                        hv_dev->channel->offermsg.offer.if_instance.b);
 197 }
 198 static DEVICE_ATTR_RO(device_id);
 199
 200 static ssize_t modalias_show(struct device *dev,
 201                              struct device_attribute *dev_attr, char *buf)
 202 {
 203         struct hv_device *hv_dev = device_to_hv_device(dev);
 204         char alias_name[VMBUS_ALIAS_LEN + 1];
 205
 206         print_alias_name(hv_dev, alias_name);
 207         return sprintf(buf, "vmbus:%s\n", alias_name);
 208 }
 209 static DEVICE_ATTR_RO(modalias);
 210
 211 static ssize_t server_monitor_pending_show(struct device *dev,
 212                                            struct device_attribute *dev_attr,
 213                                            char *buf)
 214 {
 215         struct hv_device *hv_dev = device_to_hv_device(dev);
 216
 217         if (!hv_dev->channel)
 218                 return -ENODEV;
 219         return sprintf(buf, "%d\n",
 220                        channel_pending(hv_dev->channel,
 221                                        vmbus_connection.monitor_pages[1]));
 222 }
 223 static DEVICE_ATTR_RO(server_monitor_pending);
 224
 225 static ssize_t client_monitor_pending_show(struct device *dev,
 226                                            struct device_attribute *dev_attr,
 227                                            char *buf)
 228 {
 229         struct hv_device *hv_dev = device_to_hv_device(dev);
 230
 231         if (!hv_dev->channel)
 232                 return -ENODEV;
 233         return sprintf(buf, "%d\n",
 234                        channel_pending(hv_dev->channel,
 235                                        vmbus_connection.monitor_pages[1]));
 236 }
 237 static DEVICE_ATTR_RO(client_monitor_pending);
 238
 239 static ssize_t server_monitor_latency_show(struct device *dev,
 240                                            struct device_attribute *dev_attr,
 241                                            char *buf)
 242 {
 243         struct hv_device *hv_dev = device_to_hv_device(dev);
 244
 245         if (!hv_dev->channel)
 246                 return -ENODEV;
 247         return sprintf(buf, "%d\n",
 248                        channel_latency(hv_dev->channel,
 249                                        vmbus_connection.monitor_pages[0]));
 250 }
 251 static DEVICE_ATTR_RO(server_monitor_latency);
 252
 253 static ssize_t client_monitor_latency_show(struct device *dev,
 254                                            struct device_attribute *dev_attr,
 255                                            char *buf)
 256 {
 257         struct hv_device *hv_dev = device_to_hv_device(dev);
 258
 259         if (!hv_dev->channel)
 260                 return -ENODEV;
 261         return sprintf(buf, "%d\n",
 262                        channel_latency(hv_dev->channel,
 263                                        vmbus_connection.monitor_pages[1]));
 264 }
 265 static DEVICE_ATTR_RO(client_monitor_latency);
 266
 267 static ssize_t server_monitor_conn_id_show(struct device *dev,
 268                                            struct device_attribute *dev_attr,
 269                                            char *buf)
 270 {
 271         struct hv_device *hv_dev = device_to_hv_device(dev);
 272
 273         if (!hv_dev->channel)
 274                 return -ENODEV;
 275         return sprintf(buf, "%d\n",
 276                        channel_conn_id(hv_dev->channel,
 277                                        vmbus_connection.monitor_pages[0]));
 278 }
 279 static DEVICE_ATTR_RO(server_monitor_conn_id);
 280
 281 static ssize_t client_monitor_conn_id_show(struct device *dev,
 282                                            struct device_attribute *dev_attr,
 283                                            char *buf)
 284 {
 285         struct hv_device *hv_dev = device_to_hv_device(dev);
 286
 287         if (!hv_dev->channel)
 288                 return -ENODEV;
 289         return sprintf(buf, "%d\n",
 290                        channel_conn_id(hv_dev->channel,
 291                                        vmbus_connection.monitor_pages[1]));
 292 }
 293 static DEVICE_ATTR_RO(client_monitor_conn_id);
 294
 295 static ssize_t out_intr_mask_show(struct device *dev,
 296                                   struct device_attribute *dev_attr, char *buf)
 297 {
 298         struct hv_device *hv_dev = device_to_hv_device(dev);
 299         struct hv_ring_buffer_debug_info outbound;
 300         int ret;
 301
 302         if (!hv_dev->channel)
 303                 return -ENODEV;
 304
 305         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 306                                           &outbound);
 307         if (ret < 0)
 308                 return ret;
 309
 310         return sprintf(buf, "%d\n", outbound.current_interrupt_mask);
 311 }
 312 static DEVICE_ATTR_RO(out_intr_mask);
 313
 314 static ssize_t out_read_index_show(struct device *dev,
 315                                    struct device_attribute *dev_attr, char *buf)
 316 {
 317         struct hv_device *hv_dev = device_to_hv_device(dev);
 318         struct hv_ring_buffer_debug_info outbound;
 319         int ret;
 320
 321         if (!hv_dev->channel)
 322                 return -ENODEV;
 323
 324         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 325                                           &outbound);
 326         if (ret < 0)
 327                 return ret;
 328         return sprintf(buf, "%d\n", outbound.current_read_index);
 329 }
 330 static DEVICE_ATTR_RO(out_read_index);
 331
 332 static ssize_t out_write_index_show(struct device *dev,
 333                                     struct device_attribute *dev_attr,
 334                                     char *buf)
 335 {
 336         struct hv_device *hv_dev = device_to_hv_device(dev);
 337         struct hv_ring_buffer_debug_info outbound;
 338         int ret;
 339
 340         if (!hv_dev->channel)
 341                 return -ENODEV;
 342
 343         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 344                                           &outbound);
 345         if (ret < 0)
 346                 return ret;
 347         return sprintf(buf, "%d\n", outbound.current_write_index);
 348 }
 349 static DEVICE_ATTR_RO(out_write_index);
 350
 351 static ssize_t out_read_bytes_avail_show(struct device *dev,
 352                                          struct device_attribute *dev_attr,
 353                                          char *buf)
 354 {
 355         struct hv_device *hv_dev = device_to_hv_device(dev);
 356         struct hv_ring_buffer_debug_info outbound;
 357         int ret;
 358
 359         if (!hv_dev->channel)
 360                 return -ENODEV;
 361
 362         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 363                                           &outbound);
 364         if (ret < 0)
 365                 return ret;
 366         return sprintf(buf, "%d\n", outbound.bytes_avail_toread);
 367 }
 368 static DEVICE_ATTR_RO(out_read_bytes_avail);
 369
 370 static ssize_t out_write_bytes_avail_show(struct device *dev,
 371                                           struct device_attribute *dev_attr,
 372                                           char *buf)
 373 {
 374         struct hv_device *hv_dev = device_to_hv_device(dev);
 375         struct hv_ring_buffer_debug_info outbound;
 376         int ret;
 377
 378         if (!hv_dev->channel)
 379                 return -ENODEV;
 380
 381         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound,
 382                                           &outbound);
 383         if (ret < 0)
 384                 return ret;
 385         return sprintf(buf, "%d\n", outbound.bytes_avail_towrite);
 386 }
 387 static DEVICE_ATTR_RO(out_write_bytes_avail);
 388
 389 static ssize_t in_intr_mask_show(struct device *dev,
 390                                  struct device_attribute *dev_attr, char *buf)
 391 {
 392         struct hv_device *hv_dev = device_to_hv_device(dev);
 393         struct hv_ring_buffer_debug_info inbound;
 394         int ret;
 395
 396         if (!hv_dev->channel)
 397                 return -ENODEV;
 398
 399         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 400         if (ret < 0)
 401                 return ret;
 402
 403         return sprintf(buf, "%d\n", inbound.current_interrupt_mask);
 404 }
 405 static DEVICE_ATTR_RO(in_intr_mask);
 406
 407 static ssize_t in_read_index_show(struct device *dev,
 408                                   struct device_attribute *dev_attr, char *buf)
 409 {
 410         struct hv_device *hv_dev = device_to_hv_device(dev);
 411         struct hv_ring_buffer_debug_info inbound;
 412         int ret;
 413
 414         if (!hv_dev->channel)
 415                 return -ENODEV;
 416
 417         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 418         if (ret < 0)
 419                 return ret;
 420
 421         return sprintf(buf, "%d\n", inbound.current_read_index);
 422 }
 423 static DEVICE_ATTR_RO(in_read_index);
 424
 425 static ssize_t in_write_index_show(struct device *dev,
 426                                    struct device_attribute *dev_attr, char *buf)
 427 {
 428         struct hv_device *hv_dev = device_to_hv_device(dev);
 429         struct hv_ring_buffer_debug_info inbound;
 430         int ret;
 431
 432         if (!hv_dev->channel)
 433                 return -ENODEV;
 434
 435         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 436         if (ret < 0)
 437                 return ret;
 438
 439         return sprintf(buf, "%d\n", inbound.current_write_index);
 440 }
 441 static DEVICE_ATTR_RO(in_write_index);
 442
 443 static ssize_t in_read_bytes_avail_show(struct device *dev,
 444                                         struct device_attribute *dev_attr,
 445                                         char *buf)
 446 {
 447         struct hv_device *hv_dev = device_to_hv_device(dev);
 448         struct hv_ring_buffer_debug_info inbound;
 449         int ret;
 450
 451         if (!hv_dev->channel)
 452                 return -ENODEV;
 453
 454         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 455         if (ret < 0)
 456                 return ret;
 457
 458         return sprintf(buf, "%d\n", inbound.bytes_avail_toread);
 459 }
 460 static DEVICE_ATTR_RO(in_read_bytes_avail);
 461
 462 static ssize_t in_write_bytes_avail_show(struct device *dev,
 463                                          struct device_attribute *dev_attr,
 464                                          char *buf)
 465 {
 466         struct hv_device *hv_dev = device_to_hv_device(dev);
 467         struct hv_ring_buffer_debug_info inbound;
 468         int ret;
 469
 470         if (!hv_dev->channel)
 471                 return -ENODEV;
 472
 473         ret = hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 474         if (ret < 0)
 475                 return ret;
 476
 477         return sprintf(buf, "%d\n", inbound.bytes_avail_towrite);
 478 }
 479 static DEVICE_ATTR_RO(in_write_bytes_avail);
 480
 481 static ssize_t channel_vp_mapping_show(struct device *dev,
 482                                        struct device_attribute *dev_attr,
 483                                        char *buf)
 484 {
 485         struct hv_device *hv_dev = device_to_hv_device(dev);
 486         struct vmbus_channel *channel = hv_dev->channel, *cur_sc;
 487         unsigned long flags;
 488         int buf_size = PAGE_SIZE, n_written, tot_written;
 489         struct list_head *cur;
 490
 491         if (!channel)
 492                 return -ENODEV;
 493
 494         tot_written = snprintf(buf, buf_size, "%u:%u\n",
 495                 channel->offermsg.child_relid, channel->target_cpu);
 496
 497         spin_lock_irqsave(&channel->lock, flags);
 498
 499         list_for_each(cur, &channel->sc_list) {
 500                 if (tot_written >= buf_size - 1)
 501                         break;
 502
 503                 cur_sc = list_entry(cur, struct vmbus_channel, sc_list);
 504                 n_written = scnprintf(buf + tot_written,
 505                                      buf_size - tot_written,
 506                                      "%u:%u\n",
 507                                      cur_sc->offermsg.child_relid,
 508                                      cur_sc->target_cpu);
 509                 tot_written += n_written;
 510         }
 511
 512         spin_unlock_irqrestore(&channel->lock, flags);
 513
 514         return tot_written;
 515 }
 516 static DEVICE_ATTR_RO(channel_vp_mapping);
 517
 518 static ssize_t vendor_show(struct device *dev,
 519                            struct device_attribute *dev_attr,
 520                            char *buf)
 521 {
 522         struct hv_device *hv_dev = device_to_hv_device(dev);
 523         return sprintf(buf, "0x%x\n", hv_dev->vendor_id);
 524 }
 525 static DEVICE_ATTR_RO(vendor);
 526
 527 static ssize_t device_show(struct device *dev,
 528                            struct device_attribute *dev_attr,
 529                            char *buf)
 530 {
 531         struct hv_device *hv_dev = device_to_hv_device(dev);
 532         return sprintf(buf, "0x%x\n", hv_dev->device_id);
 533 }
 534 static DEVICE_ATTR_RO(device);
 535
 536 /* Set up per device attributes in /sys/bus/vmbus/devices/<bus device> */
 537 static struct attribute *vmbus_dev_attrs[] = {
 538         &dev_attr_id.attr,
 539         &dev_attr_state.attr,
 540         &dev_attr_monitor_id.attr,
 541         &dev_attr_class_id.attr,
 542         &dev_attr_device_id.attr,
 543         &dev_attr_modalias.attr,
 544         &dev_attr_server_monitor_pending.attr,
 545         &dev_attr_client_monitor_pending.attr,
 546         &dev_attr_server_monitor_latency.attr,
 547         &dev_attr_client_monitor_latency.attr,
 548         &dev_attr_server_monitor_conn_id.attr,
 549         &dev_attr_client_monitor_conn_id.attr,
 550         &dev_attr_out_intr_mask.attr,
 551         &dev_attr_out_read_index.attr,
 552         &dev_attr_out_write_index.attr,
 553         &dev_attr_out_read_bytes_avail.attr,
 554         &dev_attr_out_write_bytes_avail.attr,
 555         &dev_attr_in_intr_mask.attr,
 556         &dev_attr_in_read_index.attr,
 557         &dev_attr_in_write_index.attr,
 558         &dev_attr_in_read_bytes_avail.attr,
 559         &dev_attr_in_write_bytes_avail.attr,
 560         &dev_attr_channel_vp_mapping.attr,
 561         &dev_attr_vendor.attr,
 562         &dev_attr_device.attr,
 563         NULL,
 564 };
 565 ATTRIBUTE_GROUPS(vmbus_dev);
 566
 567 /*
 568  * vmbus_uevent - add uevent for our device
 569  *
 570  * This routine is invoked when a device is added or removed on the vmbus to
 571  * generate a uevent to udev in the userspace. The udev will then look at its
 572  * rule and the uevent generated here to load the appropriate driver
 573  *
 574  * The alias string will be of the form vmbus:guid where guid is the string
 575  * representation of the device guid (each byte of the guid will be
 576  * represented with two hex characters.
 577  */
 578 static int vmbus_uevent(struct device *device, struct kobj_uevent_env *env)
 579 {
 580         struct hv_device *dev = device_to_hv_device(device);
 581         int ret;
 582         char alias_name[VMBUS_ALIAS_LEN + 1];
 583
 584         print_alias_name(dev, alias_name);
 585         ret = add_uevent_var(env, "MODALIAS=vmbus:%s", alias_name);
 586         return ret;
 587 }
 588
 589 static const uuid_le null_guid;
 590
 591 static inline bool is_null_guid(const uuid_le *guid)
 592 {
 593         if (uuid_le_cmp(*guid, null_guid))
 594                 return false;
 595         return true;
 596 }
 597
 598 /*
 599  * Return a matching hv_vmbus_device_id pointer.
 600  * If there is no match, return NULL.
 601  */
 602 static const struct hv_vmbus_device_id *hv_vmbus_get_id(struct hv_driver *drv,
 603                                         const uuid_le *guid)
 604 {
 605         const struct hv_vmbus_device_id *id = NULL;
 606         struct vmbus_dynid *dynid;
 607
 608         /* Look at the dynamic ids first, before the static ones */
 609         spin_lock(&drv->dynids.lock);
 610         list_for_each_entry(dynid, &drv->dynids.list, node) {
 611                 if (!uuid_le_cmp(dynid->id.guid, *guid)) {
 612                         id = &dynid->id;
 613                         break;
 614                 }
 615         }
 616         spin_unlock(&drv->dynids.lock);
 617
 618         if (id)
 619                 return id;
 620
 621         id = drv->id_table;
 622         if (id == NULL)
 623                 return NULL; /* empty device table */
 624
 625         for (; !is_null_guid(&id->guid); id++)
 626                 if (!uuid_le_cmp(id->guid, *guid))
 627                         return id;
 628
 629         return NULL;
 630 }
 631
 632 /* vmbus_add_dynid - add a new device ID to this driver and re-probe devices */
 633 static int vmbus_add_dynid(struct hv_driver *drv, uuid_le *guid)
 634 {
 635         struct vmbus_dynid *dynid;
 636
 637         dynid = kzalloc(sizeof(*dynid), GFP_KERNEL);
 638         if (!dynid)
 639                 return -ENOMEM;
 640
 641         dynid->id.guid = *guid;
 642
 643         spin_lock(&drv->dynids.lock);
 644         list_add_tail(&dynid->node, &drv->dynids.list);
 645         spin_unlock(&drv->dynids.lock);
 646
 647         return driver_attach(&drv->driver);
 648 }
 649
 650 static void vmbus_free_dynids(struct hv_driver *drv)
 651 {
 652         struct vmbus_dynid *dynid, *n;
 653
 654         spin_lock(&drv->dynids.lock);
 655         list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
 656                 list_del(&dynid->node);
 657                 kfree(dynid);
 658         }
 659         spin_unlock(&drv->dynids.lock);
 660 }
 661
 662 /*
 663  * store_new_id - sysfs frontend to vmbus_add_dynid()
 664  *
 665  * Allow GUIDs to be added to an existing driver via sysfs.
 666  */
 667 static ssize_t new_id_store(struct device_driver *driver, const char *buf,
 668                             size_t count)
 669 {
 670         struct hv_driver *drv = drv_to_hv_drv(driver);
 671         uuid_le guid;
 672         ssize_t retval;
 673
 674         retval = uuid_le_to_bin(buf, &guid);
 675         if (retval)
 676                 return retval;
 677
 678         if (hv_vmbus_get_id(drv, &guid))
 679                 return -EEXIST;
 680
 681         retval = vmbus_add_dynid(drv, &guid);
 682         if (retval)
 683                 return retval;
 684         return count;
 685 }
 686 static DRIVER_ATTR_WO(new_id);
 687
 688 /*
 689  * store_remove_id - remove a PCI device ID from this driver
 690  *
 691  * Removes a dynamic pci device ID to this driver.
 692  */
 693 static ssize_t remove_id_store(struct device_driver *driver, const char *buf,
 694                                size_t count)
 695 {
 696         struct hv_driver *drv = drv_to_hv_drv(driver);
 697         struct vmbus_dynid *dynid, *n;
 698         uuid_le guid;
 699         ssize_t retval;
 700
 701         retval = uuid_le_to_bin(buf, &guid);
 702         if (retval)
 703                 return retval;
 704
 705         retval = -ENODEV;
 706         spin_lock(&drv->dynids.lock);
 707         list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
 708                 struct hv_vmbus_device_id *id = &dynid->id;
 709
 710                 if (!uuid_le_cmp(id->guid, guid)) {
 711                         list_del(&dynid->node);
 712                         kfree(dynid);
 713                         retval = count;
 714                         break;
 715                 }
 716         }
 717         spin_unlock(&drv->dynids.lock);
 718
 719         return retval;
 720 }
 721 static DRIVER_ATTR_WO(remove_id);
 722
 723 static struct attribute *vmbus_drv_attrs[] = {
 724         &driver_attr_new_id.attr,
 725         &driver_attr_remove_id.attr,
 726         NULL,
 727 };
 728 ATTRIBUTE_GROUPS(vmbus_drv);
 729
 730
 731 /*
 732  * vmbus_match - Attempt to match the specified device to the specified driver
 733  */
 734 static int vmbus_match(struct device *device, struct device_driver *driver)
 735 {
 736         struct hv_driver *drv = drv_to_hv_drv(driver);
 737         struct hv_device *hv_dev = device_to_hv_device(device);
 738
 739         /* The hv_sock driver handles all hv_sock offers. */
 740         if (is_hvsock_channel(hv_dev->channel))
 741                 return drv->hvsock;
 742
 743         if (hv_vmbus_get_id(drv, &hv_dev->dev_type))
 744                 return 1;
 745
 746         return 0;
 747 }
 748
 749 /*
 750  * vmbus_probe - Add the new vmbus's child device
 751  */
 752 static int vmbus_probe(struct device *child_device)
 753 {
 754         int ret = 0;
 755         struct hv_driver *drv =
 756                         drv_to_hv_drv(child_device->driver);
 757         struct hv_device *dev = device_to_hv_device(child_device);
 758         const struct hv_vmbus_device_id *dev_id;
 759
 760         dev_id = hv_vmbus_get_id(drv, &dev->dev_type);
 761         if (drv->probe) {
 762                 ret = drv->probe(dev, dev_id);
 763                 if (ret != 0)
 764                         pr_err("probe failed for device %s (%d)\n",
 765                                dev_name(child_device), ret);
 766
 767         } else {
 768                 pr_err("probe not set for driver %s\n",
 769                        dev_name(child_device));
 770                 ret = -ENODEV;
 771         }
 772         return ret;
 773 }
 774
 775 /*
 776  * vmbus_remove - Remove a vmbus device
 777  */
 778 static int vmbus_remove(struct device *child_device)
 779 {
 780         struct hv_driver *drv;
 781         struct hv_device *dev = device_to_hv_device(child_device);
 782
 783         if (child_device->driver) {
 784                 drv = drv_to_hv_drv(child_device->driver);
 785                 if (drv->remove)
 786                         drv->remove(dev);
 787         }
 788
 789         return 0;
 790 }
 791
 792
 793 /*
 794  * vmbus_shutdown - Shutdown a vmbus device
 795  */
 796 static void vmbus_shutdown(struct device *child_device)
 797 {
 798         struct hv_driver *drv;
 799         struct hv_device *dev = device_to_hv_device(child_device);
 800
 801
 802         /* The device may not be attached yet */
 803         if (!child_device->driver)
 804                 return;
 805
 806         drv = drv_to_hv_drv(child_device->driver);
 807
 808         if (drv->shutdown)
 809                 drv->shutdown(dev);
 810 }
 811
 812
 813 /*
 814  * vmbus_device_release - Final callback release of the vmbus child device
 815  */
 816 static void vmbus_device_release(struct device *device)
 817 {
 818         struct hv_device *hv_dev = device_to_hv_device(device);
 819         struct vmbus_channel *channel = hv_dev->channel;
 820
 821         mutex_lock(&vmbus_connection.channel_mutex);
 822         hv_process_channel_removal(channel->offermsg.child_relid);
 823         mutex_unlock(&vmbus_connection.channel_mutex);
 824         kfree(hv_dev);
 825
 826 }
 827
 828 /* The one and only one */
 829 static struct bus_type  hv_bus = {
 830         .name =         "vmbus",
 831         .match =                vmbus_match,
 832         .shutdown =             vmbus_shutdown,
 833         .remove =               vmbus_remove,
 834         .probe =                vmbus_probe,
 835         .uevent =               vmbus_uevent,
 836         .dev_groups =           vmbus_dev_groups,
 837         .drv_groups =           vmbus_drv_groups,
 838 };
 839
 840 struct onmessage_work_context {
 841         struct work_struct work;
 842         struct hv_message msg;
 843 };
 844
 845 static void vmbus_onmessage_work(struct work_struct *work)
 846 {
 847         struct onmessage_work_context *ctx;
 848
 849         /* Do not process messages if we're in DISCONNECTED state */
 850         if (vmbus_connection.conn_state == DISCONNECTED)
 851                 return;
 852
 853         ctx = container_of(work, struct onmessage_work_context,
 854                            work);
 855         vmbus_onmessage(&ctx->msg);
 856         kfree(ctx);
 857 }
 858
 859 static void hv_process_timer_expiration(struct hv_message *msg,
 860                                         struct hv_per_cpu_context *hv_cpu)
 861 {
 862         struct clock_event_device *dev = hv_cpu->clk_evt;
 863
 864         if (dev->event_handler)
 865                 dev->event_handler(dev);
 866
 867         vmbus_signal_eom(msg, HVMSG_TIMER_EXPIRED);
 868 }
 869
 870 void vmbus_on_msg_dpc(unsigned long data)
 871 {
 872         struct hv_per_cpu_context *hv_cpu = (void *)data;
 873         void *page_addr = hv_cpu->synic_message_page;
 874         struct hv_message *msg = (struct hv_message *)page_addr +
 875                                   VMBUS_MESSAGE_SINT;
 876         struct vmbus_channel_message_header *hdr;
 877         const struct vmbus_channel_message_table_entry *entry;
 878         struct onmessage_work_context *ctx;
 879         u32 message_type = msg->header.message_type;
 880
 881         if (message_type == HVMSG_NONE)
 882                 /* no msg */
 883                 return;
 884
 885         hdr = (struct vmbus_channel_message_header *)msg->u.payload;
 886
 887         if (hdr->msgtype >= CHANNELMSG_COUNT) {
 888                 WARN_ONCE(1, "unknown msgtype=%d\n", hdr->msgtype);
 889                 goto msg_handled;
 890         }
 891
 892         entry = &channel_message_table[hdr->msgtype];
 893
 894         if (!entry->message_handler)
 895                 goto msg_handled;
 896
 897         if (entry->handler_type == VMHT_BLOCKING) {
 898                 ctx = kmalloc(sizeof(*ctx), GFP_ATOMIC);
 899                 if (ctx == NULL)
 900                         return;
 901
 902                 INIT_WORK(&ctx->work, vmbus_onmessage_work);
 903                 memcpy(&ctx->msg, msg, sizeof(*msg));
 904
 905                 /*
 906                  * The host can generate a rescind message while we
 907                  * may still be handling the original offer. We deal with
 908                  * this condition by ensuring the processing is done on the
 909                  * same CPU.
 910                  */
 911                 switch (hdr->msgtype) {
 912                 case CHANNELMSG_RESCIND_CHANNELOFFER:
 913                         /*
 914                          * If we are handling the rescind message;
 915                          * schedule the work on the global work queue.
 916                          */
 917                         schedule_work_on(vmbus_connection.connect_cpu,
 918                                          &ctx->work);
 919                         break;
 920
 921                 case CHANNELMSG_OFFERCHANNEL:
 922                         atomic_inc(&vmbus_connection.offer_in_progress);
 923                         queue_work_on(vmbus_connection.connect_cpu,
 924                                       vmbus_connection.work_queue,
 925                                       &ctx->work);
 926                         break;
 927
 928                 default:
 929                         queue_work(vmbus_connection.work_queue, &ctx->work);
 930                 }
 931         } else
 932                 entry->message_handler(hdr);
 933
 934 msg_handled:
 935         vmbus_signal_eom(msg, message_type);
 936 }
 937
 938
 939 /*
 940  * Direct callback for channels using other deferred processing
 941  */
 942 static void vmbus_channel_isr(struct vmbus_channel *channel)
 943 {
 944         void (*callback_fn)(void *);
 945
 946         callback_fn = READ_ONCE(channel->onchannel_callback);
 947         if (likely(callback_fn != NULL))
 948                 (*callback_fn)(channel->channel_callback_context);
 949 }
 950
 951 /*
 952  * Schedule all channels with events pending
 953  */
 954 static void vmbus_chan_sched(struct hv_per_cpu_context *hv_cpu)
 955 {
 956         unsigned long *recv_int_page;
 957         u32 maxbits, relid;
 958
 959         if (vmbus_proto_version < VERSION_WIN8) {
 960                 maxbits = MAX_NUM_CHANNELS_SUPPORTED;
 961                 recv_int_page = vmbus_connection.recv_int_page;
 962         } else {
 963                 /*
 964                  * When the host is win8 and beyond, the event page
 965                  * can be directly checked to get the id of the channel
 966                  * that has the interrupt pending.
 967                  */
 968                 void *page_addr = hv_cpu->synic_event_page;
 969                 union hv_synic_event_flags *event
 970                         = (union hv_synic_event_flags *)page_addr +
 971                                                  VMBUS_MESSAGE_SINT;
 972
 973                 maxbits = HV_EVENT_FLAGS_COUNT;
 974                 recv_int_page = event->flags;
 975         }
 976
 977         if (unlikely(!recv_int_page))
 978                 return;
 979
 980         for_each_set_bit(relid, recv_int_page, maxbits) {
 981                 struct vmbus_channel *channel;
 982
 983                 if (!sync_test_and_clear_bit(relid, recv_int_page))
 984                         continue;
 985
 986                 /* Special case - vmbus channel protocol msg */
 987                 if (relid == 0)
 988                         continue;
 989
 990                 rcu_read_lock();
 991
 992                 /* Find channel based on relid */
 993                 list_for_each_entry_rcu(channel, &hv_cpu->chan_list, percpu_list) {
 994                         if (channel->offermsg.child_relid != relid)
 995                                 continue;
 996
 997                         if (channel->rescind)
 998                                 continue;
 999
1000                         switch (channel->callback_mode) {
1001                         case HV_CALL_ISR:
1002                                 vmbus_channel_isr(channel);
1003                                 break;
1004
1005                         case HV_CALL_BATCHED:
1006                                 hv_begin_read(&channel->inbound);
1007                                 /* fallthrough */
1008                         case HV_CALL_DIRECT:
1009                                 tasklet_schedule(&channel->callback_event);
1010                         }
1011                 }
1012
1013                 rcu_read_unlock();
1014         }
1015 }
1016
1017 static void vmbus_isr(void)
1018 {
1019         struct hv_per_cpu_context *hv_cpu
1020                 = this_cpu_ptr(hv_context.cpu_context);
1021         void *page_addr = hv_cpu->synic_event_page;
1022         struct hv_message *msg;
1023         union hv_synic_event_flags *event;
1024         bool handled = false;
1025
1026         if (unlikely(page_addr == NULL))
1027                 return;
1028
1029         event = (union hv_synic_event_flags *)page_addr +
1030                                          VMBUS_MESSAGE_SINT;
1031         /*
1032          * Check for events before checking for messages. This is the order
1033          * in which events and messages are checked in Windows guests on
1034          * Hyper-V, and the Windows team suggested we do the same.
1035          */
1036
1037         if ((vmbus_proto_version == VERSION_WS2008) ||
1038                 (vmbus_proto_version == VERSION_WIN7)) {
1039
1040                 /* Since we are a child, we only need to check bit 0 */
1041                 if (sync_test_and_clear_bit(0, event->flags))
1042                         handled = true;
1043         } else {
1044                 /*
1045                  * Our host is win8 or above. The signaling mechanism
1046                  * has changed and we can directly look at the event page.
1047                  * If bit n is set then we have an interrup on the channel
1048                  * whose id is n.
1049                  */
1050                 handled = true;
1051         }
1052
1053         if (handled)
1054                 vmbus_chan_sched(hv_cpu);
1055
1056         page_addr = hv_cpu->synic_message_page;
1057         msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
1058
1059         /* Check if there are actual msgs to be processed */
1060         if (msg->header.message_type != HVMSG_NONE) {
1061                 if (msg->header.message_type == HVMSG_TIMER_EXPIRED)
1062                         hv_process_timer_expiration(msg, hv_cpu);
1063                 else
1064                         tasklet_schedule(&hv_cpu->msg_dpc);
1065         }
1066
1067         add_interrupt_randomness(HYPERVISOR_CALLBACK_VECTOR);
1068 }
1069
1070
1071 /*
1072  * vmbus_bus_init -Main vmbus driver initialization routine.
1073  *
1074  * Here, we
1075  *      - initialize the vmbus driver context
1076  *      - invoke the vmbus hv main init routine
1077  *      - retrieve the channel offers
1078  */
1079 static int vmbus_bus_init(void)
1080 {
1081         int ret;
1082
1083         /* Hypervisor initialization...setup hypercall page..etc */
1084         ret = hv_init();
1085         if (ret != 0) {
1086                 pr_err("Unable to initialize the hypervisor - 0x%x\n", ret);
1087                 return ret;
1088         }
1089
1090         ret = bus_register(&hv_bus);
1091         if (ret)
1092                 return ret;
1093
1094         hv_setup_vmbus_irq(vmbus_isr);
1095
1096         ret = hv_synic_alloc();
1097         if (ret)
1098                 goto err_alloc;
1099         /*
1100          * Initialize the per-cpu interrupt state and
1101          * connect to the host.
1102          */
1103         ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "x86/hyperv:online",
1104                                 hv_synic_init, hv_synic_cleanup);
1105         if (ret < 0)
1106                 goto err_alloc;
1107         hyperv_cpuhp_online = ret;
1108
1109         ret = vmbus_connect();
1110         if (ret)
1111                 goto err_connect;
1112
1113         /*
1114          * Only register if the crash MSRs are available
1115          */
1116         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1117                 register_die_notifier(&hyperv_die_block);
1118                 atomic_notifier_chain_register(&panic_notifier_list,
1119                                                &hyperv_panic_block);
1120         }
1121
1122         vmbus_request_offers();
1123
1124         return 0;
1125
1126 err_connect:
1127         cpuhp_remove_state(hyperv_cpuhp_online);
1128 err_alloc:
1129         hv_synic_free();
1130         hv_remove_vmbus_irq();
1131
1132         bus_unregister(&hv_bus);
1133
1134         return ret;
1135 }
1136
1137 /**
1138  * __vmbus_child_driver_register() - Register a vmbus's driver
1139  * @hv_driver: Pointer to driver structure you want to register
1140  * @owner: owner module of the drv
1141  * @mod_name: module name string
1142  *
1143  * Registers the given driver with Linux through the 'driver_register()' call
1144  * and sets up the hyper-v vmbus handling for this driver.
1145  * It will return the state of the 'driver_register()' call.
1146  *
1147  */
1148 int __vmbus_driver_register(struct hv_driver *hv_driver, struct module *owner, const char *mod_name)
1149 {
1150         int ret;
1151
1152         pr_info("registering driver %s\n", hv_driver->name);
1153
1154         ret = vmbus_exists();
1155         if (ret < 0)
1156                 return ret;
1157
1158         hv_driver->driver.name = hv_driver->name;
1159         hv_driver->driver.owner = owner;
1160         hv_driver->driver.mod_name = mod_name;
1161         hv_driver->driver.bus = &hv_bus;
1162
1163         spin_lock_init(&hv_driver->dynids.lock);
1164         INIT_LIST_HEAD(&hv_driver->dynids.list);
1165
1166         ret = driver_register(&hv_driver->driver);
1167
1168         return ret;
1169 }
1170 EXPORT_SYMBOL_GPL(__vmbus_driver_register);
1171
1172 /**
1173  * vmbus_driver_unregister() - Unregister a vmbus's driver
1174  * @hv_driver: Pointer to driver structure you want to
1175  *             un-register
1176  *
1177  * Un-register the given driver that was previous registered with a call to
1178  * vmbus_driver_register()
1179  */
1180 void vmbus_driver_unregister(struct hv_driver *hv_driver)
1181 {
1182         pr_info("unregistering driver %s\n", hv_driver->name);
1183
1184         if (!vmbus_exists()) {
1185                 driver_unregister(&hv_driver->driver);
1186                 vmbus_free_dynids(hv_driver);
1187         }
1188 }
1189 EXPORT_SYMBOL_GPL(vmbus_driver_unregister);
1190
1191 /*
1192  * vmbus_device_create - Creates and registers a new child device
1193  * on the vmbus.
1194  */
1195 struct hv_device *vmbus_device_create(const uuid_le *type,
1196                                       const uuid_le *instance,
1197                                       struct vmbus_channel *channel)
1198 {
1199         struct hv_device *child_device_obj;
1200
1201         child_device_obj = kzalloc(sizeof(struct hv_device), GFP_KERNEL);
1202         if (!child_device_obj) {
1203                 pr_err("Unable to allocate device object for child device\n");
1204                 return NULL;
1205         }
1206
1207         child_device_obj->channel = channel;
1208         memcpy(&child_device_obj->dev_type, type, sizeof(uuid_le));
1209         memcpy(&child_device_obj->dev_instance, instance,
1210                sizeof(uuid_le));
1211         child_device_obj->vendor_id = 0x1414; /* MSFT vendor ID */
1212
1213
1214         return child_device_obj;
1215 }
1216
1217 /*
1218  * vmbus_device_register - Register the child device
1219  */
1220 int vmbus_device_register(struct hv_device *child_device_obj)
1221 {
1222         int ret = 0;
1223
1224         dev_set_name(&child_device_obj->device, "%pUl",
1225                      child_device_obj->channel->offermsg.offer.if_instance.b);
1226
1227         child_device_obj->device.bus = &hv_bus;
1228         child_device_obj->device.parent = &hv_acpi_dev->dev;
1229         child_device_obj->device.release = vmbus_device_release;
1230
1231         /*
1232          * Register with the LDM. This will kick off the driver/device
1233          * binding...which will eventually call vmbus_match() and vmbus_probe()
1234          */
1235         ret = device_register(&child_device_obj->device);
1236
1237         if (ret)
1238                 pr_err("Unable to register child device\n");
1239         else
1240                 pr_debug("child device %s registered\n",
1241                         dev_name(&child_device_obj->device));
1242
1243         return ret;
1244 }
1245
1246 /*
1247  * vmbus_device_unregister - Remove the specified child device
1248  * from the vmbus.
1249  */
1250 void vmbus_device_unregister(struct hv_device *device_obj)
1251 {
1252         pr_debug("child device %s unregistered\n",
1253                 dev_name(&device_obj->device));
1254
1255         /*
1256          * Kick off the process of unregistering the device.
1257          * This will call vmbus_remove() and eventually vmbus_device_release()
1258          */
1259         device_unregister(&device_obj->device);
1260 }
1261
1262
1263 /*
1264  * VMBUS is an acpi enumerated device. Get the information we
1265  * need from DSDT.
1266  */
1267 #define VTPM_BASE_ADDRESS 0xfed40000
1268 static acpi_status vmbus_walk_resources(struct acpi_resource *res, void *ctx)
1269 {
1270         resource_size_t start = 0;
1271         resource_size_t end = 0;
1272         struct resource *new_res;
1273         struct resource **old_res = &hyperv_mmio;
1274         struct resource **prev_res = NULL;
1275
1276         switch (res->type) {
1277
1278         /*
1279          * "Address" descriptors are for bus windows. Ignore
1280          * "memory" descriptors, which are for registers on
1281          * devices.
1282          */
1283         case ACPI_RESOURCE_TYPE_ADDRESS32:
1284                 start = res->data.address32.address.minimum;
1285                 end = res->data.address32.address.maximum;
1286                 break;
1287
1288         case ACPI_RESOURCE_TYPE_ADDRESS64:
1289                 start = res->data.address64.address.minimum;
1290                 end = res->data.address64.address.maximum;
1291                 break;
1292
1293         default:
1294                 /* Unused resource type */
1295                 return AE_OK;
1296
1297         }
1298         /*
1299          * Ignore ranges that are below 1MB, as they're not
1300          * necessary or useful here.
1301          */
1302         if (end < 0x100000)
1303                 return AE_OK;
1304
1305         new_res = kzalloc(sizeof(*new_res), GFP_ATOMIC);
1306         if (!new_res)
1307                 return AE_NO_MEMORY;
1308
1309         /* If this range overlaps the virtual TPM, truncate it. */
1310         if (end > VTPM_BASE_ADDRESS && start < VTPM_BASE_ADDRESS)
1311                 end = VTPM_BASE_ADDRESS;
1312
1313         new_res->name = "hyperv mmio";
1314         new_res->flags = IORESOURCE_MEM;
1315         new_res->start = start;
1316         new_res->end = end;
1317
1318         /*
1319          * If two ranges are adjacent, merge them.
1320          */
1321         do {
1322                 if (!*old_res) {
1323                         *old_res = new_res;
1324                         break;
1325                 }
1326
1327                 if (((*old_res)->end + 1) == new_res->start) {
1328                         (*old_res)->end = new_res->end;
1329                         kfree(new_res);
1330                         break;
1331                 }
1332
1333                 if ((*old_res)->start == new_res->end + 1) {
1334                         (*old_res)->start = new_res->start;
1335                         kfree(new_res);
1336                         break;
1337                 }
1338
1339                 if ((*old_res)->start > new_res->end) {
1340                         new_res->sibling = *old_res;
1341                         if (prev_res)
1342                                 (*prev_res)->sibling = new_res;
1343                         *old_res = new_res;
1344                         break;
1345                 }
1346
1347                 prev_res = old_res;
1348                 old_res = &(*old_res)->sibling;
1349
1350         } while (1);
1351
1352         return AE_OK;
1353 }
1354
1355 static int vmbus_acpi_remove(struct acpi_device *device)
1356 {
1357         struct resource *cur_res;
1358         struct resource *next_res;
1359
1360         if (hyperv_mmio) {
1361                 if (fb_mmio) {
1362                         __release_region(hyperv_mmio, fb_mmio->start,
1363                                          resource_size(fb_mmio));
1364                         fb_mmio = NULL;
1365                 }
1366
1367                 for (cur_res = hyperv_mmio; cur_res; cur_res = next_res) {
1368                         next_res = cur_res->sibling;
1369                         kfree(cur_res);
1370                 }
1371         }
1372
1373         return 0;
1374 }
1375
1376 static void vmbus_reserve_fb(void)
1377 {
1378         int size;
1379         /*
1380          * Make a claim for the frame buffer in the resource tree under the
1381          * first node, which will be the one below 4GB.  The length seems to
1382          * be underreported, particularly in a Generation 1 VM.  So start out
1383          * reserving a larger area and make it smaller until it succeeds.
1384          */
1385
1386         if (screen_info.lfb_base) {
1387                 if (efi_enabled(EFI_BOOT))
1388                         size = max_t(__u32, screen_info.lfb_size, 0x800000);
1389                 else
1390                         size = max_t(__u32, screen_info.lfb_size, 0x4000000);
1391
1392                 for (; !fb_mmio && (size >= 0x100000); size >>= 1) {
1393                         fb_mmio = __request_region(hyperv_mmio,
1394                                                    screen_info.lfb_base, size,
1395                                                    fb_mmio_name, 0);
1396                 }
1397         }
1398 }
1399
1400 /**
1401  * vmbus_allocate_mmio() - Pick a memory-mapped I/O range.
1402  * @new:                If successful, supplied a pointer to the
1403  *                      allocated MMIO space.
1404  * @device_obj:         Identifies the caller
1405  * @min:                Minimum guest physical address of the
1406  *                      allocation
1407  * @max:                Maximum guest physical address
1408  * @size:               Size of the range to be allocated
1409  * @align:              Alignment of the range to be allocated
1410  * @fb_overlap_ok:      Whether this allocation can be allowed
1411  *                      to overlap the video frame buffer.
1412  *
1413  * This function walks the resources granted to VMBus by the
1414  * _CRS object in the ACPI namespace underneath the parent
1415  * "bridge" whether that's a root PCI bus in the Generation 1
1416  * case or a Module Device in the Generation 2 case.  It then
1417  * attempts to allocate from the global MMIO pool in a way that
1418  * matches the constraints supplied in these parameters and by
1419  * that _CRS.
1420  *
1421  * Return: 0 on success, -errno on failure
1422  */
1423 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
1424                         resource_size_t min, resource_size_t max,
1425                         resource_size_t size, resource_size_t align,
1426                         bool fb_overlap_ok)
1427 {
1428         struct resource *iter, *shadow;
1429         resource_size_t range_min, range_max, start;
1430         const char *dev_n = dev_name(&device_obj->device);
1431         int retval;
1432
1433         retval = -ENXIO;
1434         down(&hyperv_mmio_lock);
1435
1436         /*
1437          * If overlaps with frame buffers are allowed, then first attempt to
1438          * make the allocation from within the reserved region.  Because it
1439          * is already reserved, no shadow allocation is necessary.
1440          */
1441         if (fb_overlap_ok && fb_mmio && !(min > fb_mmio->end) &&
1442             !(max < fb_mmio->start)) {
1443
1444                 range_min = fb_mmio->start;
1445                 range_max = fb_mmio->end;
1446                 start = (range_min + align - 1) & ~(align - 1);
1447                 for (; start + size - 1 <= range_max; start += align) {
1448                         *new = request_mem_region_exclusive(start, size, dev_n);
1449                         if (*new) {
1450                                 retval = 0;
1451                                 goto exit;
1452                         }
1453                 }
1454         }
1455
1456         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
1457                 if ((iter->start >= max) || (iter->end <= min))
1458                         continue;
1459
1460                 range_min = iter->start;
1461                 range_max = iter->end;
1462                 start = (range_min + align - 1) & ~(align - 1);
1463                 for (; start + size - 1 <= range_max; start += align) {
1464                         shadow = __request_region(iter, start, size, NULL,
1465                                                   IORESOURCE_BUSY);
1466                         if (!shadow)
1467                                 continue;
1468
1469                         *new = request_mem_region_exclusive(start, size, dev_n);
1470                         if (*new) {
1471                                 shadow->name = (char *)*new;
1472                                 retval = 0;
1473                                 goto exit;
1474                         }
1475
1476                         __release_region(iter, start, size);
1477                 }
1478         }
1479
1480 exit:
1481         up(&hyperv_mmio_lock);
1482         return retval;
1483 }
1484 EXPORT_SYMBOL_GPL(vmbus_allocate_mmio);
1485
1486 /**
1487  * vmbus_free_mmio() - Free a memory-mapped I/O range.
1488  * @start:              Base address of region to release.
1489  * @size:               Size of the range to be allocated
1490  *
1491  * This function releases anything requested by
1492  * vmbus_mmio_allocate().
1493  */
1494 void vmbus_free_mmio(resource_size_t start, resource_size_t size)
1495 {
1496         struct resource *iter;
1497
1498         down(&hyperv_mmio_lock);
1499         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
1500                 if ((iter->start >= start + size) || (iter->end <= start))
1501                         continue;
1502
1503                 __release_region(iter, start, size);
1504         }
1505         release_mem_region(start, size);
1506         up(&hyperv_mmio_lock);
1507
1508 }
1509 EXPORT_SYMBOL_GPL(vmbus_free_mmio);
1510
1511 static int vmbus_acpi_add(struct acpi_device *device)
1512 {
1513         acpi_status result;
1514         int ret_val = -ENODEV;
1515         struct acpi_device *ancestor;
1516
1517         hv_acpi_dev = device;
1518
1519         result = acpi_walk_resources(device->handle, METHOD_NAME__CRS,
1520                                         vmbus_walk_resources, NULL);
1521
1522         if (ACPI_FAILURE(result))
1523                 goto acpi_walk_err;
1524         /*
1525          * Some ancestor of the vmbus acpi device (Gen1 or Gen2
1526          * firmware) is the VMOD that has the mmio ranges. Get that.
1527          */
1528         for (ancestor = device->parent; ancestor; ancestor = ancestor->parent) {
1529                 result = acpi_walk_resources(ancestor->handle, METHOD_NAME__CRS,
1530                                              vmbus_walk_resources, NULL);
1531
1532                 if (ACPI_FAILURE(result))
1533                         continue;
1534                 if (hyperv_mmio) {
1535                         vmbus_reserve_fb();
1536                         break;
1537                 }
1538         }
1539         ret_val = 0;
1540
1541 acpi_walk_err:
1542         complete(&probe_event);
1543         if (ret_val)
1544                 vmbus_acpi_remove(device);
1545         return ret_val;
1546 }
1547
1548 static const struct acpi_device_id vmbus_acpi_device_ids[] = {
1549         {"VMBUS", 0},
1550         {"VMBus", 0},
1551         {"", 0},
1552 };
1553 MODULE_DEVICE_TABLE(acpi, vmbus_acpi_device_ids);
1554
1555 static struct acpi_driver vmbus_acpi_driver = {
1556         .name = "vmbus",
1557         .ids = vmbus_acpi_device_ids,
1558         .ops = {
1559                 .add = vmbus_acpi_add,
1560                 .remove = vmbus_acpi_remove,
1561         },
1562 };
1563
1564 static void hv_kexec_handler(void)
1565 {
1566         hv_synic_clockevents_cleanup();
1567         vmbus_initiate_unload(false);
1568         vmbus_connection.conn_state = DISCONNECTED;
1569         /* Make sure conn_state is set as hv_synic_cleanup checks for it */
1570         mb();
1571         cpuhp_remove_state(hyperv_cpuhp_online);
1572         hyperv_cleanup();
1573 };
1574
1575 static void hv_crash_handler(struct pt_regs *regs)
1576 {
1577         vmbus_initiate_unload(true);
1578         /*
1579          * In crash handler we can't schedule synic cleanup for all CPUs,
1580          * doing the cleanup for current CPU only. This should be sufficient
1581          * for kdump.
1582          */
1583         vmbus_connection.conn_state = DISCONNECTED;
1584         hv_synic_cleanup(smp_processor_id());
1585         hyperv_cleanup();
1586 };
1587
1588 static int __init hv_acpi_init(void)
1589 {
1590         int ret, t;
1591
1592         if (x86_hyper_type != X86_HYPER_MS_HYPERV)
1593                 return -ENODEV;
1594
1595         init_completion(&probe_event);
1596
1597         /*
1598          * Get ACPI resources first.
1599          */
1600         ret = acpi_bus_register_driver(&vmbus_acpi_driver);
1601
1602         if (ret)
1603                 return ret;
1604
1605         t = wait_for_completion_timeout(&probe_event, 5*HZ);
1606         if (t == 0) {
1607                 ret = -ETIMEDOUT;
1608                 goto cleanup;
1609         }
1610
1611         ret = vmbus_bus_init();
1612         if (ret)
1613                 goto cleanup;
1614
1615         hv_setup_kexec_handler(hv_kexec_handler);
1616         hv_setup_crash_handler(hv_crash_handler);
1617
1618         return 0;
1619
1620 cleanup:
1621         acpi_bus_unregister_driver(&vmbus_acpi_driver);
1622         hv_acpi_dev = NULL;
1623         return ret;
1624 }
1625
1626 static void __exit vmbus_exit(void)
1627 {
1628         int cpu;
1629
1630         hv_remove_kexec_handler();
1631         hv_remove_crash_handler();
1632         vmbus_connection.conn_state = DISCONNECTED;
1633         hv_synic_clockevents_cleanup();
1634         vmbus_disconnect();
1635         hv_remove_vmbus_irq();
1636         for_each_online_cpu(cpu) {
1637                 struct hv_per_cpu_context *hv_cpu
1638                         = per_cpu_ptr(hv_context.cpu_context, cpu);
1639
1640                 tasklet_kill(&hv_cpu->msg_dpc);
1641         }
1642         vmbus_free_channels();
1643
1644         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1645                 unregister_die_notifier(&hyperv_die_block);
1646                 atomic_notifier_chain_unregister(&panic_notifier_list,
1647                                                  &hyperv_panic_block);
1648         }
1649         bus_unregister(&hv_bus);
1650
1651         cpuhp_remove_state(hyperv_cpuhp_online);
1652         hv_synic_free();
1653         acpi_bus_unregister_driver(&vmbus_acpi_driver);
1654 }
1655
1656
1657 MODULE_LICENSE("GPL");
1658
1659 subsys_initcall(hv_acpi_init);
1660 module_exit(vmbus_exit);