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/mshyperv.h>
  40 #include <linux/notifier.h>
  41 #include <linux/ptrace.h>
  42 #include <linux/screen_info.h>
  43 #include <linux/kdebug.h>
  44 #include <linux/efi.h>
  45 #include <linux/random.h>
  46 #include "hyperv_vmbus.h"
  47
  48 struct vmbus_dynid {
  49         struct list_head node;
  50         struct hv_vmbus_device_id id;
  51 };
  52
  53 static struct acpi_device  *hv_acpi_dev;
  54
  55 static struct completion probe_event;
  56
  57 static int hyperv_cpuhp_online;
  58
  59 static int hyperv_panic_event(struct notifier_block *nb, unsigned long val,
  60                               void *args)
  61 {
  62         struct pt_regs *regs;
  63
  64         regs = current_pt_regs();
  65
  66         hyperv_report_panic(regs, val);
  67         return NOTIFY_DONE;
  68 }
  69
  70 static int hyperv_die_event(struct notifier_block *nb, unsigned long val,
  71                             void *args)
  72 {
  73         struct die_args *die = (struct die_args *)args;
  74         struct pt_regs *regs = die->regs;
  75
  76         hyperv_report_panic(regs, val);
  77         return NOTIFY_DONE;
  78 }
  79
  80 static struct notifier_block hyperv_die_block = {
  81         .notifier_call = hyperv_die_event,
  82 };
  83 static struct notifier_block hyperv_panic_block = {
  84         .notifier_call = hyperv_panic_event,
  85 };
  86
  87 static const char *fb_mmio_name = "fb_range";
  88 static struct resource *fb_mmio;
  89 static struct resource *hyperv_mmio;
  90 static DEFINE_SEMAPHORE(hyperv_mmio_lock);
  91
  92 static int vmbus_exists(void)
  93 {
  94         if (hv_acpi_dev == NULL)
  95                 return -ENODEV;
  96
  97         return 0;
  98 }
  99
 100 #define VMBUS_ALIAS_LEN ((sizeof((struct hv_vmbus_device_id *)0)->guid) * 2)
 101 static void print_alias_name(struct hv_device *hv_dev, char *alias_name)
 102 {
 103         int i;
 104         for (i = 0; i < VMBUS_ALIAS_LEN; i += 2)
 105                 sprintf(&alias_name[i], "%02x", hv_dev->dev_type.b[i/2]);
 106 }
 107
 108 static u8 channel_monitor_group(const struct vmbus_channel *channel)
 109 {
 110         return (u8)channel->offermsg.monitorid / 32;
 111 }
 112
 113 static u8 channel_monitor_offset(const struct vmbus_channel *channel)
 114 {
 115         return (u8)channel->offermsg.monitorid % 32;
 116 }
 117
 118 static u32 channel_pending(const struct vmbus_channel *channel,
 119                            const struct hv_monitor_page *monitor_page)
 120 {
 121         u8 monitor_group = channel_monitor_group(channel);
 122
 123         return monitor_page->trigger_group[monitor_group].pending;
 124 }
 125
 126 static u32 channel_latency(const struct vmbus_channel *channel,
 127                            const struct hv_monitor_page *monitor_page)
 128 {
 129         u8 monitor_group = channel_monitor_group(channel);
 130         u8 monitor_offset = channel_monitor_offset(channel);
 131
 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
 301         if (!hv_dev->channel)
 302                 return -ENODEV;
 303         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
 304         return sprintf(buf, "%d\n", outbound.current_interrupt_mask);
 305 }
 306 static DEVICE_ATTR_RO(out_intr_mask);
 307
 308 static ssize_t out_read_index_show(struct device *dev,
 309                                    struct device_attribute *dev_attr, char *buf)
 310 {
 311         struct hv_device *hv_dev = device_to_hv_device(dev);
 312         struct hv_ring_buffer_debug_info outbound;
 313
 314         if (!hv_dev->channel)
 315                 return -ENODEV;
 316         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
 317         return sprintf(buf, "%d\n", outbound.current_read_index);
 318 }
 319 static DEVICE_ATTR_RO(out_read_index);
 320
 321 static ssize_t out_write_index_show(struct device *dev,
 322                                     struct device_attribute *dev_attr,
 323                                     char *buf)
 324 {
 325         struct hv_device *hv_dev = device_to_hv_device(dev);
 326         struct hv_ring_buffer_debug_info outbound;
 327
 328         if (!hv_dev->channel)
 329                 return -ENODEV;
 330         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
 331         return sprintf(buf, "%d\n", outbound.current_write_index);
 332 }
 333 static DEVICE_ATTR_RO(out_write_index);
 334
 335 static ssize_t out_read_bytes_avail_show(struct device *dev,
 336                                          struct device_attribute *dev_attr,
 337                                          char *buf)
 338 {
 339         struct hv_device *hv_dev = device_to_hv_device(dev);
 340         struct hv_ring_buffer_debug_info outbound;
 341
 342         if (!hv_dev->channel)
 343                 return -ENODEV;
 344         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
 345         return sprintf(buf, "%d\n", outbound.bytes_avail_toread);
 346 }
 347 static DEVICE_ATTR_RO(out_read_bytes_avail);
 348
 349 static ssize_t out_write_bytes_avail_show(struct device *dev,
 350                                           struct device_attribute *dev_attr,
 351                                           char *buf)
 352 {
 353         struct hv_device *hv_dev = device_to_hv_device(dev);
 354         struct hv_ring_buffer_debug_info outbound;
 355
 356         if (!hv_dev->channel)
 357                 return -ENODEV;
 358         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
 359         return sprintf(buf, "%d\n", outbound.bytes_avail_towrite);
 360 }
 361 static DEVICE_ATTR_RO(out_write_bytes_avail);
 362
 363 static ssize_t in_intr_mask_show(struct device *dev,
 364                                  struct device_attribute *dev_attr, char *buf)
 365 {
 366         struct hv_device *hv_dev = device_to_hv_device(dev);
 367         struct hv_ring_buffer_debug_info inbound;
 368
 369         if (!hv_dev->channel)
 370                 return -ENODEV;
 371         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 372         return sprintf(buf, "%d\n", inbound.current_interrupt_mask);
 373 }
 374 static DEVICE_ATTR_RO(in_intr_mask);
 375
 376 static ssize_t in_read_index_show(struct device *dev,
 377                                   struct device_attribute *dev_attr, char *buf)
 378 {
 379         struct hv_device *hv_dev = device_to_hv_device(dev);
 380         struct hv_ring_buffer_debug_info inbound;
 381
 382         if (!hv_dev->channel)
 383                 return -ENODEV;
 384         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 385         return sprintf(buf, "%d\n", inbound.current_read_index);
 386 }
 387 static DEVICE_ATTR_RO(in_read_index);
 388
 389 static ssize_t in_write_index_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
 395         if (!hv_dev->channel)
 396                 return -ENODEV;
 397         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 398         return sprintf(buf, "%d\n", inbound.current_write_index);
 399 }
 400 static DEVICE_ATTR_RO(in_write_index);
 401
 402 static ssize_t in_read_bytes_avail_show(struct device *dev,
 403                                         struct device_attribute *dev_attr,
 404                                         char *buf)
 405 {
 406         struct hv_device *hv_dev = device_to_hv_device(dev);
 407         struct hv_ring_buffer_debug_info inbound;
 408
 409         if (!hv_dev->channel)
 410                 return -ENODEV;
 411         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 412         return sprintf(buf, "%d\n", inbound.bytes_avail_toread);
 413 }
 414 static DEVICE_ATTR_RO(in_read_bytes_avail);
 415
 416 static ssize_t in_write_bytes_avail_show(struct device *dev,
 417                                          struct device_attribute *dev_attr,
 418                                          char *buf)
 419 {
 420         struct hv_device *hv_dev = device_to_hv_device(dev);
 421         struct hv_ring_buffer_debug_info inbound;
 422
 423         if (!hv_dev->channel)
 424                 return -ENODEV;
 425         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 426         return sprintf(buf, "%d\n", inbound.bytes_avail_towrite);
 427 }
 428 static DEVICE_ATTR_RO(in_write_bytes_avail);
 429
 430 static ssize_t channel_vp_mapping_show(struct device *dev,
 431                                        struct device_attribute *dev_attr,
 432                                        char *buf)
 433 {
 434         struct hv_device *hv_dev = device_to_hv_device(dev);
 435         struct vmbus_channel *channel = hv_dev->channel, *cur_sc;
 436         unsigned long flags;
 437         int buf_size = PAGE_SIZE, n_written, tot_written;
 438         struct list_head *cur;
 439
 440         if (!channel)
 441                 return -ENODEV;
 442
 443         tot_written = snprintf(buf, buf_size, "%u:%u\n",
 444                 channel->offermsg.child_relid, channel->target_cpu);
 445
 446         spin_lock_irqsave(&channel->lock, flags);
 447
 448         list_for_each(cur, &channel->sc_list) {
 449                 if (tot_written >= buf_size - 1)
 450                         break;
 451
 452                 cur_sc = list_entry(cur, struct vmbus_channel, sc_list);
 453                 n_written = scnprintf(buf + tot_written,
 454                                      buf_size - tot_written,
 455                                      "%u:%u\n",
 456                                      cur_sc->offermsg.child_relid,
 457                                      cur_sc->target_cpu);
 458                 tot_written += n_written;
 459         }
 460
 461         spin_unlock_irqrestore(&channel->lock, flags);
 462
 463         return tot_written;
 464 }
 465 static DEVICE_ATTR_RO(channel_vp_mapping);
 466
 467 static ssize_t vendor_show(struct device *dev,
 468                            struct device_attribute *dev_attr,
 469                            char *buf)
 470 {
 471         struct hv_device *hv_dev = device_to_hv_device(dev);
 472         return sprintf(buf, "0x%x\n", hv_dev->vendor_id);
 473 }
 474 static DEVICE_ATTR_RO(vendor);
 475
 476 static ssize_t device_show(struct device *dev,
 477                            struct device_attribute *dev_attr,
 478                            char *buf)
 479 {
 480         struct hv_device *hv_dev = device_to_hv_device(dev);
 481         return sprintf(buf, "0x%x\n", hv_dev->device_id);
 482 }
 483 static DEVICE_ATTR_RO(device);
 484
 485 /* Set up per device attributes in /sys/bus/vmbus/devices/<bus device> */
 486 static struct attribute *vmbus_dev_attrs[] = {
 487         &dev_attr_id.attr,
 488         &dev_attr_state.attr,
 489         &dev_attr_monitor_id.attr,
 490         &dev_attr_class_id.attr,
 491         &dev_attr_device_id.attr,
 492         &dev_attr_modalias.attr,
 493         &dev_attr_server_monitor_pending.attr,
 494         &dev_attr_client_monitor_pending.attr,
 495         &dev_attr_server_monitor_latency.attr,
 496         &dev_attr_client_monitor_latency.attr,
 497         &dev_attr_server_monitor_conn_id.attr,
 498         &dev_attr_client_monitor_conn_id.attr,
 499         &dev_attr_out_intr_mask.attr,
 500         &dev_attr_out_read_index.attr,
 501         &dev_attr_out_write_index.attr,
 502         &dev_attr_out_read_bytes_avail.attr,
 503         &dev_attr_out_write_bytes_avail.attr,
 504         &dev_attr_in_intr_mask.attr,
 505         &dev_attr_in_read_index.attr,
 506         &dev_attr_in_write_index.attr,
 507         &dev_attr_in_read_bytes_avail.attr,
 508         &dev_attr_in_write_bytes_avail.attr,
 509         &dev_attr_channel_vp_mapping.attr,
 510         &dev_attr_vendor.attr,
 511         &dev_attr_device.attr,
 512         NULL,
 513 };
 514 ATTRIBUTE_GROUPS(vmbus_dev);
 515
 516 /*
 517  * vmbus_uevent - add uevent for our device
 518  *
 519  * This routine is invoked when a device is added or removed on the vmbus to
 520  * generate a uevent to udev in the userspace. The udev will then look at its
 521  * rule and the uevent generated here to load the appropriate driver
 522  *
 523  * The alias string will be of the form vmbus:guid where guid is the string
 524  * representation of the device guid (each byte of the guid will be
 525  * represented with two hex characters.
 526  */
 527 static int vmbus_uevent(struct device *device, struct kobj_uevent_env *env)
 528 {
 529         struct hv_device *dev = device_to_hv_device(device);
 530         int ret;
 531         char alias_name[VMBUS_ALIAS_LEN + 1];
 532
 533         print_alias_name(dev, alias_name);
 534         ret = add_uevent_var(env, "MODALIAS=vmbus:%s", alias_name);
 535         return ret;
 536 }
 537
 538 static const uuid_le null_guid;
 539
 540 static inline bool is_null_guid(const uuid_le *guid)
 541 {
 542         if (uuid_le_cmp(*guid, null_guid))
 543                 return false;
 544         return true;
 545 }
 546
 547 /*
 548  * Return a matching hv_vmbus_device_id pointer.
 549  * If there is no match, return NULL.
 550  */
 551 static const struct hv_vmbus_device_id *hv_vmbus_get_id(struct hv_driver *drv,
 552                                         const uuid_le *guid)
 553 {
 554         const struct hv_vmbus_device_id *id = NULL;
 555         struct vmbus_dynid *dynid;
 556
 557         /* Look at the dynamic ids first, before the static ones */
 558         spin_lock(&drv->dynids.lock);
 559         list_for_each_entry(dynid, &drv->dynids.list, node) {
 560                 if (!uuid_le_cmp(dynid->id.guid, *guid)) {
 561                         id = &dynid->id;
 562                         break;
 563                 }
 564         }
 565         spin_unlock(&drv->dynids.lock);
 566
 567         if (id)
 568                 return id;
 569
 570         id = drv->id_table;
 571         if (id == NULL)
 572                 return NULL; /* empty device table */
 573
 574         for (; !is_null_guid(&id->guid); id++)
 575                 if (!uuid_le_cmp(id->guid, *guid))
 576                         return id;
 577
 578         return NULL;
 579 }
 580
 581 /* vmbus_add_dynid - add a new device ID to this driver and re-probe devices */
 582 static int vmbus_add_dynid(struct hv_driver *drv, uuid_le *guid)
 583 {
 584         struct vmbus_dynid *dynid;
 585
 586         dynid = kzalloc(sizeof(*dynid), GFP_KERNEL);
 587         if (!dynid)
 588                 return -ENOMEM;
 589
 590         dynid->id.guid = *guid;
 591
 592         spin_lock(&drv->dynids.lock);
 593         list_add_tail(&dynid->node, &drv->dynids.list);
 594         spin_unlock(&drv->dynids.lock);
 595
 596         return driver_attach(&drv->driver);
 597 }
 598
 599 static void vmbus_free_dynids(struct hv_driver *drv)
 600 {
 601         struct vmbus_dynid *dynid, *n;
 602
 603         spin_lock(&drv->dynids.lock);
 604         list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
 605                 list_del(&dynid->node);
 606                 kfree(dynid);
 607         }
 608         spin_unlock(&drv->dynids.lock);
 609 }
 610
 611 /*
 612  * store_new_id - sysfs frontend to vmbus_add_dynid()
 613  *
 614  * Allow GUIDs to be added to an existing driver via sysfs.
 615  */
 616 static ssize_t new_id_store(struct device_driver *driver, const char *buf,
 617                             size_t count)
 618 {
 619         struct hv_driver *drv = drv_to_hv_drv(driver);
 620         uuid_le guid;
 621         ssize_t retval;
 622
 623         retval = uuid_le_to_bin(buf, &guid);
 624         if (retval)
 625                 return retval;
 626
 627         if (hv_vmbus_get_id(drv, &guid))
 628                 return -EEXIST;
 629
 630         retval = vmbus_add_dynid(drv, &guid);
 631         if (retval)
 632                 return retval;
 633         return count;
 634 }
 635 static DRIVER_ATTR_WO(new_id);
 636
 637 /*
 638  * store_remove_id - remove a PCI device ID from this driver
 639  *
 640  * Removes a dynamic pci device ID to this driver.
 641  */
 642 static ssize_t remove_id_store(struct device_driver *driver, const char *buf,
 643                                size_t count)
 644 {
 645         struct hv_driver *drv = drv_to_hv_drv(driver);
 646         struct vmbus_dynid *dynid, *n;
 647         uuid_le guid;
 648         ssize_t retval;
 649
 650         retval = uuid_le_to_bin(buf, &guid);
 651         if (retval)
 652                 return retval;
 653
 654         retval = -ENODEV;
 655         spin_lock(&drv->dynids.lock);
 656         list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
 657                 struct hv_vmbus_device_id *id = &dynid->id;
 658
 659                 if (!uuid_le_cmp(id->guid, guid)) {
 660                         list_del(&dynid->node);
 661                         kfree(dynid);
 662                         retval = count;
 663                         break;
 664                 }
 665         }
 666         spin_unlock(&drv->dynids.lock);
 667
 668         return retval;
 669 }
 670 static DRIVER_ATTR_WO(remove_id);
 671
 672 static struct attribute *vmbus_drv_attrs[] = {
 673         &driver_attr_new_id.attr,
 674         &driver_attr_remove_id.attr,
 675         NULL,
 676 };
 677 ATTRIBUTE_GROUPS(vmbus_drv);
 678
 679
 680 /*
 681  * vmbus_match - Attempt to match the specified device to the specified driver
 682  */
 683 static int vmbus_match(struct device *device, struct device_driver *driver)
 684 {
 685         struct hv_driver *drv = drv_to_hv_drv(driver);
 686         struct hv_device *hv_dev = device_to_hv_device(device);
 687
 688         /* The hv_sock driver handles all hv_sock offers. */
 689         if (is_hvsock_channel(hv_dev->channel))
 690                 return drv->hvsock;
 691
 692         if (hv_vmbus_get_id(drv, &hv_dev->dev_type))
 693                 return 1;
 694
 695         return 0;
 696 }
 697
 698 /*
 699  * vmbus_probe - Add the new vmbus's child device
 700  */
 701 static int vmbus_probe(struct device *child_device)
 702 {
 703         int ret = 0;
 704         struct hv_driver *drv =
 705                         drv_to_hv_drv(child_device->driver);
 706         struct hv_device *dev = device_to_hv_device(child_device);
 707         const struct hv_vmbus_device_id *dev_id;
 708
 709         dev_id = hv_vmbus_get_id(drv, &dev->dev_type);
 710         if (drv->probe) {
 711                 ret = drv->probe(dev, dev_id);
 712                 if (ret != 0)
 713                         pr_err("probe failed for device %s (%d)\n",
 714                                dev_name(child_device), ret);
 715
 716         } else {
 717                 pr_err("probe not set for driver %s\n",
 718                        dev_name(child_device));
 719                 ret = -ENODEV;
 720         }
 721         return ret;
 722 }
 723
 724 /*
 725  * vmbus_remove - Remove a vmbus device
 726  */
 727 static int vmbus_remove(struct device *child_device)
 728 {
 729         struct hv_driver *drv;
 730         struct hv_device *dev = device_to_hv_device(child_device);
 731
 732         if (child_device->driver) {
 733                 drv = drv_to_hv_drv(child_device->driver);
 734                 if (drv->remove)
 735                         drv->remove(dev);
 736         }
 737
 738         return 0;
 739 }
 740
 741
 742 /*
 743  * vmbus_shutdown - Shutdown a vmbus device
 744  */
 745 static void vmbus_shutdown(struct device *child_device)
 746 {
 747         struct hv_driver *drv;
 748         struct hv_device *dev = device_to_hv_device(child_device);
 749
 750
 751         /* The device may not be attached yet */
 752         if (!child_device->driver)
 753                 return;
 754
 755         drv = drv_to_hv_drv(child_device->driver);
 756
 757         if (drv->shutdown)
 758                 drv->shutdown(dev);
 759 }
 760
 761
 762 /*
 763  * vmbus_device_release - Final callback release of the vmbus child device
 764  */
 765 static void vmbus_device_release(struct device *device)
 766 {
 767         struct hv_device *hv_dev = device_to_hv_device(device);
 768         struct vmbus_channel *channel = hv_dev->channel;
 769
 770         mutex_lock(&vmbus_connection.channel_mutex);
 771         hv_process_channel_removal(channel->offermsg.child_relid);
 772         mutex_unlock(&vmbus_connection.channel_mutex);
 773         kfree(hv_dev);
 774
 775 }
 776
 777 /* The one and only one */
 778 static struct bus_type  hv_bus = {
 779         .name =         "vmbus",
 780         .match =                vmbus_match,
 781         .shutdown =             vmbus_shutdown,
 782         .remove =               vmbus_remove,
 783         .probe =                vmbus_probe,
 784         .uevent =               vmbus_uevent,
 785         .dev_groups =           vmbus_dev_groups,
 786         .drv_groups =           vmbus_drv_groups,
 787 };
 788
 789 struct onmessage_work_context {
 790         struct work_struct work;
 791         struct hv_message msg;
 792 };
 793
 794 static void vmbus_onmessage_work(struct work_struct *work)
 795 {
 796         struct onmessage_work_context *ctx;
 797
 798         /* Do not process messages if we're in DISCONNECTED state */
 799         if (vmbus_connection.conn_state == DISCONNECTED)
 800                 return;
 801
 802         ctx = container_of(work, struct onmessage_work_context,
 803                            work);
 804         vmbus_onmessage(&ctx->msg);
 805         kfree(ctx);
 806 }
 807
 808 static void hv_process_timer_expiration(struct hv_message *msg,
 809                                         struct hv_per_cpu_context *hv_cpu)
 810 {
 811         struct clock_event_device *dev = hv_cpu->clk_evt;
 812
 813         if (dev->event_handler)
 814                 dev->event_handler(dev);
 815
 816         vmbus_signal_eom(msg, HVMSG_TIMER_EXPIRED);
 817 }
 818
 819 void vmbus_on_msg_dpc(unsigned long data)
 820 {
 821         struct hv_per_cpu_context *hv_cpu = (void *)data;
 822         void *page_addr = hv_cpu->synic_message_page;
 823         struct hv_message *msg = (struct hv_message *)page_addr +
 824                                   VMBUS_MESSAGE_SINT;
 825         struct vmbus_channel_message_header *hdr;
 826         const struct vmbus_channel_message_table_entry *entry;
 827         struct onmessage_work_context *ctx;
 828         u32 message_type = msg->header.message_type;
 829
 830         if (message_type == HVMSG_NONE)
 831                 /* no msg */
 832                 return;
 833
 834         hdr = (struct vmbus_channel_message_header *)msg->u.payload;
 835
 836         trace_vmbus_on_msg_dpc(hdr);
 837
 838         if (hdr->msgtype >= CHANNELMSG_COUNT) {
 839                 WARN_ONCE(1, "unknown msgtype=%d\n", hdr->msgtype);
 840                 goto msg_handled;
 841         }
 842
 843         entry = &channel_message_table[hdr->msgtype];
 844         if (entry->handler_type == VMHT_BLOCKING) {
 845                 ctx = kmalloc(sizeof(*ctx), GFP_ATOMIC);
 846                 if (ctx == NULL)
 847                         return;
 848
 849                 INIT_WORK(&ctx->work, vmbus_onmessage_work);
 850                 memcpy(&ctx->msg, msg, sizeof(*msg));
 851
 852                 /*
 853                  * The host can generate a rescind message while we
 854                  * may still be handling the original offer. We deal with
 855                  * this condition by ensuring the processing is done on the
 856                  * same CPU.
 857                  */
 858                 switch (hdr->msgtype) {
 859                 case CHANNELMSG_RESCIND_CHANNELOFFER:
 860                         /*
 861                          * If we are handling the rescind message;
 862                          * schedule the work on the global work queue.
 863                          */
 864                         schedule_work_on(vmbus_connection.connect_cpu,
 865                                          &ctx->work);
 866                         break;
 867
 868                 case CHANNELMSG_OFFERCHANNEL:
 869                         atomic_inc(&vmbus_connection.offer_in_progress);
 870                         queue_work_on(vmbus_connection.connect_cpu,
 871                                       vmbus_connection.work_queue,
 872                                       &ctx->work);
 873                         break;
 874
 875                 default:
 876                         queue_work(vmbus_connection.work_queue, &ctx->work);
 877                 }
 878         } else
 879                 entry->message_handler(hdr);
 880
 881 msg_handled:
 882         vmbus_signal_eom(msg, message_type);
 883 }
 884
 885
 886 /*
 887  * Direct callback for channels using other deferred processing
 888  */
 889 static void vmbus_channel_isr(struct vmbus_channel *channel)
 890 {
 891         void (*callback_fn)(void *);
 892
 893         callback_fn = READ_ONCE(channel->onchannel_callback);
 894         if (likely(callback_fn != NULL))
 895                 (*callback_fn)(channel->channel_callback_context);
 896 }
 897
 898 /*
 899  * Schedule all channels with events pending
 900  */
 901 static void vmbus_chan_sched(struct hv_per_cpu_context *hv_cpu)
 902 {
 903         unsigned long *recv_int_page;
 904         u32 maxbits, relid;
 905
 906         if (vmbus_proto_version < VERSION_WIN8) {
 907                 maxbits = MAX_NUM_CHANNELS_SUPPORTED;
 908                 recv_int_page = vmbus_connection.recv_int_page;
 909         } else {
 910                 /*
 911                  * When the host is win8 and beyond, the event page
 912                  * can be directly checked to get the id of the channel
 913                  * that has the interrupt pending.
 914                  */
 915                 void *page_addr = hv_cpu->synic_event_page;
 916                 union hv_synic_event_flags *event
 917                         = (union hv_synic_event_flags *)page_addr +
 918                                                  VMBUS_MESSAGE_SINT;
 919
 920                 maxbits = HV_EVENT_FLAGS_COUNT;
 921                 recv_int_page = event->flags;
 922         }
 923
 924         if (unlikely(!recv_int_page))
 925                 return;
 926
 927         for_each_set_bit(relid, recv_int_page, maxbits) {
 928                 struct vmbus_channel *channel;
 929
 930                 if (!sync_test_and_clear_bit(relid, recv_int_page))
 931                         continue;
 932
 933                 /* Special case - vmbus channel protocol msg */
 934                 if (relid == 0)
 935                         continue;
 936
 937                 rcu_read_lock();
 938
 939                 /* Find channel based on relid */
 940                 list_for_each_entry_rcu(channel, &hv_cpu->chan_list, percpu_list) {
 941                         if (channel->offermsg.child_relid != relid)
 942                                 continue;
 943
 944                         if (channel->rescind)
 945                                 continue;
 946
 947                         trace_vmbus_chan_sched(channel);
 948
 949                         ++channel->interrupts;
 950
 951                         switch (channel->callback_mode) {
 952                         case HV_CALL_ISR:
 953                                 vmbus_channel_isr(channel);
 954                                 break;
 955
 956                         case HV_CALL_BATCHED:
 957                                 hv_begin_read(&channel->inbound);
 958                                 /* fallthrough */
 959                         case HV_CALL_DIRECT:
 960                                 tasklet_schedule(&channel->callback_event);
 961                         }
 962                 }
 963
 964                 rcu_read_unlock();
 965         }
 966 }
 967
 968 static void vmbus_isr(void)
 969 {
 970         struct hv_per_cpu_context *hv_cpu
 971                 = this_cpu_ptr(hv_context.cpu_context);
 972         void *page_addr = hv_cpu->synic_event_page;
 973         struct hv_message *msg;
 974         union hv_synic_event_flags *event;
 975         bool handled = false;
 976
 977         if (unlikely(page_addr == NULL))
 978                 return;
 979
 980         event = (union hv_synic_event_flags *)page_addr +
 981                                          VMBUS_MESSAGE_SINT;
 982         /*
 983          * Check for events before checking for messages. This is the order
 984          * in which events and messages are checked in Windows guests on
 985          * Hyper-V, and the Windows team suggested we do the same.
 986          */
 987
 988         if ((vmbus_proto_version == VERSION_WS2008) ||
 989                 (vmbus_proto_version == VERSION_WIN7)) {
 990
 991                 /* Since we are a child, we only need to check bit 0 */
 992                 if (sync_test_and_clear_bit(0, event->flags))
 993                         handled = true;
 994         } else {
 995                 /*
 996                  * Our host is win8 or above. The signaling mechanism
 997                  * has changed and we can directly look at the event page.
 998                  * If bit n is set then we have an interrup on the channel
 999                  * whose id is n.
1000                  */
1001                 handled = true;
1002         }
1003
1004         if (handled)
1005                 vmbus_chan_sched(hv_cpu);
1006
1007         page_addr = hv_cpu->synic_message_page;
1008         msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
1009
1010         /* Check if there are actual msgs to be processed */
1011         if (msg->header.message_type != HVMSG_NONE) {
1012                 if (msg->header.message_type == HVMSG_TIMER_EXPIRED)
1013                         hv_process_timer_expiration(msg, hv_cpu);
1014                 else
1015                         tasklet_schedule(&hv_cpu->msg_dpc);
1016         }
1017
1018         add_interrupt_randomness(HYPERVISOR_CALLBACK_VECTOR, 0);
1019 }
1020
1021
1022 /*
1023  * vmbus_bus_init -Main vmbus driver initialization routine.
1024  *
1025  * Here, we
1026  *      - initialize the vmbus driver context
1027  *      - invoke the vmbus hv main init routine
1028  *      - retrieve the channel offers
1029  */
1030 static int vmbus_bus_init(void)
1031 {
1032         int ret;
1033
1034         /* Hypervisor initialization...setup hypercall page..etc */
1035         ret = hv_init();
1036         if (ret != 0) {
1037                 pr_err("Unable to initialize the hypervisor - 0x%x\n", ret);
1038                 return ret;
1039         }
1040
1041         ret = bus_register(&hv_bus);
1042         if (ret)
1043                 return ret;
1044
1045         hv_setup_vmbus_irq(vmbus_isr);
1046
1047         ret = hv_synic_alloc();
1048         if (ret)
1049                 goto err_alloc;
1050         /*
1051          * Initialize the per-cpu interrupt state and
1052          * connect to the host.
1053          */
1054         ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "hyperv/vmbus:online",
1055                                 hv_synic_init, hv_synic_cleanup);
1056         if (ret < 0)
1057                 goto err_alloc;
1058         hyperv_cpuhp_online = ret;
1059
1060         ret = vmbus_connect();
1061         if (ret)
1062                 goto err_connect;
1063
1064         /*
1065          * Only register if the crash MSRs are available
1066          */
1067         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1068                 register_die_notifier(&hyperv_die_block);
1069                 atomic_notifier_chain_register(&panic_notifier_list,
1070                                                &hyperv_panic_block);
1071         }
1072
1073         vmbus_request_offers();
1074
1075         return 0;
1076
1077 err_connect:
1078         cpuhp_remove_state(hyperv_cpuhp_online);
1079 err_alloc:
1080         hv_synic_free();
1081         hv_remove_vmbus_irq();
1082
1083         bus_unregister(&hv_bus);
1084
1085         return ret;
1086 }
1087
1088 /**
1089  * __vmbus_child_driver_register() - Register a vmbus's driver
1090  * @hv_driver: Pointer to driver structure you want to register
1091  * @owner: owner module of the drv
1092  * @mod_name: module name string
1093  *
1094  * Registers the given driver with Linux through the 'driver_register()' call
1095  * and sets up the hyper-v vmbus handling for this driver.
1096  * It will return the state of the 'driver_register()' call.
1097  *
1098  */
1099 int __vmbus_driver_register(struct hv_driver *hv_driver, struct module *owner, const char *mod_name)
1100 {
1101         int ret;
1102
1103         pr_info("registering driver %s\n", hv_driver->name);
1104
1105         ret = vmbus_exists();
1106         if (ret < 0)
1107                 return ret;
1108
1109         hv_driver->driver.name = hv_driver->name;
1110         hv_driver->driver.owner = owner;
1111         hv_driver->driver.mod_name = mod_name;
1112         hv_driver->driver.bus = &hv_bus;
1113
1114         spin_lock_init(&hv_driver->dynids.lock);
1115         INIT_LIST_HEAD(&hv_driver->dynids.list);
1116
1117         ret = driver_register(&hv_driver->driver);
1118
1119         return ret;
1120 }
1121 EXPORT_SYMBOL_GPL(__vmbus_driver_register);
1122
1123 /**
1124  * vmbus_driver_unregister() - Unregister a vmbus's driver
1125  * @hv_driver: Pointer to driver structure you want to
1126  *             un-register
1127  *
1128  * Un-register the given driver that was previous registered with a call to
1129  * vmbus_driver_register()
1130  */
1131 void vmbus_driver_unregister(struct hv_driver *hv_driver)
1132 {
1133         pr_info("unregistering driver %s\n", hv_driver->name);
1134
1135         if (!vmbus_exists()) {
1136                 driver_unregister(&hv_driver->driver);
1137                 vmbus_free_dynids(hv_driver);
1138         }
1139 }
1140 EXPORT_SYMBOL_GPL(vmbus_driver_unregister);
1141
1142
1143 /*
1144  * Called when last reference to channel is gone.
1145  */
1146 static void vmbus_chan_release(struct kobject *kobj)
1147 {
1148         struct vmbus_channel *channel
1149                 = container_of(kobj, struct vmbus_channel, kobj);
1150
1151         kfree_rcu(channel, rcu);
1152 }
1153
1154 struct vmbus_chan_attribute {
1155         struct attribute attr;
1156         ssize_t (*show)(const struct vmbus_channel *chan, char *buf);
1157         ssize_t (*store)(struct vmbus_channel *chan,
1158                          const char *buf, size_t count);
1159 };
1160 #define VMBUS_CHAN_ATTR(_name, _mode, _show, _store) \
1161         struct vmbus_chan_attribute chan_attr_##_name \
1162                 = __ATTR(_name, _mode, _show, _store)
1163 #define VMBUS_CHAN_ATTR_RW(_name) \
1164         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RW(_name)
1165 #define VMBUS_CHAN_ATTR_RO(_name) \
1166         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RO(_name)
1167 #define VMBUS_CHAN_ATTR_WO(_name) \
1168         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_WO(_name)
1169
1170 static ssize_t vmbus_chan_attr_show(struct kobject *kobj,
1171                                     struct attribute *attr, char *buf)
1172 {
1173         const struct vmbus_chan_attribute *attribute
1174                 = container_of(attr, struct vmbus_chan_attribute, attr);
1175         const struct vmbus_channel *chan
1176                 = container_of(kobj, struct vmbus_channel, kobj);
1177
1178         if (!attribute->show)
1179                 return -EIO;
1180
1181         return attribute->show(chan, buf);
1182 }
1183
1184 static const struct sysfs_ops vmbus_chan_sysfs_ops = {
1185         .show = vmbus_chan_attr_show,
1186 };
1187
1188 static ssize_t out_mask_show(const struct vmbus_channel *channel, char *buf)
1189 {
1190         const struct hv_ring_buffer_info *rbi = &channel->outbound;
1191
1192         return sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1193 }
1194 static VMBUS_CHAN_ATTR_RO(out_mask);
1195
1196 static ssize_t in_mask_show(const struct vmbus_channel *channel, char *buf)
1197 {
1198         const struct hv_ring_buffer_info *rbi = &channel->inbound;
1199
1200         return sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1201 }
1202 static VMBUS_CHAN_ATTR_RO(in_mask);
1203
1204 static ssize_t read_avail_show(const struct vmbus_channel *channel, char *buf)
1205 {
1206         const struct hv_ring_buffer_info *rbi = &channel->inbound;
1207
1208         return sprintf(buf, "%u\n", hv_get_bytes_to_read(rbi));
1209 }
1210 static VMBUS_CHAN_ATTR_RO(read_avail);
1211
1212 static ssize_t write_avail_show(const struct vmbus_channel *channel, char *buf)
1213 {
1214         const struct hv_ring_buffer_info *rbi = &channel->outbound;
1215
1216         return sprintf(buf, "%u\n", hv_get_bytes_to_write(rbi));
1217 }
1218 static VMBUS_CHAN_ATTR_RO(write_avail);
1219
1220 static ssize_t show_target_cpu(const struct vmbus_channel *channel, char *buf)
1221 {
1222         return sprintf(buf, "%u\n", channel->target_cpu);
1223 }
1224 static VMBUS_CHAN_ATTR(cpu, S_IRUGO, show_target_cpu, NULL);
1225
1226 static ssize_t channel_pending_show(const struct vmbus_channel *channel,
1227                                     char *buf)
1228 {
1229         return sprintf(buf, "%d\n",
1230                        channel_pending(channel,
1231                                        vmbus_connection.monitor_pages[1]));
1232 }
1233 static VMBUS_CHAN_ATTR(pending, S_IRUGO, channel_pending_show, NULL);
1234
1235 static ssize_t channel_latency_show(const struct vmbus_channel *channel,
1236                                     char *buf)
1237 {
1238         return sprintf(buf, "%d\n",
1239                        channel_latency(channel,
1240                                        vmbus_connection.monitor_pages[1]));
1241 }
1242 static VMBUS_CHAN_ATTR(latency, S_IRUGO, channel_latency_show, NULL);
1243
1244 static ssize_t channel_interrupts_show(const struct vmbus_channel *channel, char *buf)
1245 {
1246         return sprintf(buf, "%llu\n", channel->interrupts);
1247 }
1248 static VMBUS_CHAN_ATTR(interrupts, S_IRUGO, channel_interrupts_show, NULL);
1249
1250 static ssize_t channel_events_show(const struct vmbus_channel *channel, char *buf)
1251 {
1252         return sprintf(buf, "%llu\n", channel->sig_events);
1253 }
1254 static VMBUS_CHAN_ATTR(events, S_IRUGO, channel_events_show, NULL);
1255
1256 static ssize_t subchannel_monitor_id_show(const struct vmbus_channel *channel,
1257                                           char *buf)
1258 {
1259         return sprintf(buf, "%u\n", channel->offermsg.monitorid);
1260 }
1261 static VMBUS_CHAN_ATTR(monitor_id, S_IRUGO, subchannel_monitor_id_show, NULL);
1262
1263 static ssize_t subchannel_id_show(const struct vmbus_channel *channel,
1264                                   char *buf)
1265 {
1266         return sprintf(buf, "%u\n",
1267                        channel->offermsg.offer.sub_channel_index);
1268 }
1269 static VMBUS_CHAN_ATTR_RO(subchannel_id);
1270
1271 static struct attribute *vmbus_chan_attrs[] = {
1272         &chan_attr_out_mask.attr,
1273         &chan_attr_in_mask.attr,
1274         &chan_attr_read_avail.attr,
1275         &chan_attr_write_avail.attr,
1276         &chan_attr_cpu.attr,
1277         &chan_attr_pending.attr,
1278         &chan_attr_latency.attr,
1279         &chan_attr_interrupts.attr,
1280         &chan_attr_events.attr,
1281         &chan_attr_monitor_id.attr,
1282         &chan_attr_subchannel_id.attr,
1283         NULL
1284 };
1285
1286 static struct kobj_type vmbus_chan_ktype = {
1287         .sysfs_ops = &vmbus_chan_sysfs_ops,
1288         .release = vmbus_chan_release,
1289         .default_attrs = vmbus_chan_attrs,
1290 };
1291
1292 /*
1293  * vmbus_add_channel_kobj - setup a sub-directory under device/channels
1294  */
1295 int vmbus_add_channel_kobj(struct hv_device *dev, struct vmbus_channel *channel)
1296 {
1297         struct kobject *kobj = &channel->kobj;
1298         u32 relid = channel->offermsg.child_relid;
1299         int ret;
1300
1301         kobj->kset = dev->channels_kset;
1302         ret = kobject_init_and_add(kobj, &vmbus_chan_ktype, NULL,
1303                                    "%u", relid);
1304         if (ret)
1305                 return ret;
1306
1307         kobject_uevent(kobj, KOBJ_ADD);
1308
1309         return 0;
1310 }
1311
1312 /*
1313  * vmbus_device_create - Creates and registers a new child device
1314  * on the vmbus.
1315  */
1316 struct hv_device *vmbus_device_create(const uuid_le *type,
1317                                       const uuid_le *instance,
1318                                       struct vmbus_channel *channel)
1319 {
1320         struct hv_device *child_device_obj;
1321
1322         child_device_obj = kzalloc(sizeof(struct hv_device), GFP_KERNEL);
1323         if (!child_device_obj) {
1324                 pr_err("Unable to allocate device object for child device\n");
1325                 return NULL;
1326         }
1327
1328         child_device_obj->channel = channel;
1329         memcpy(&child_device_obj->dev_type, type, sizeof(uuid_le));
1330         memcpy(&child_device_obj->dev_instance, instance,
1331                sizeof(uuid_le));
1332         child_device_obj->vendor_id = 0x1414; /* MSFT vendor ID */
1333
1334
1335         return child_device_obj;
1336 }
1337
1338 /*
1339  * vmbus_device_register - Register the child device
1340  */
1341 int vmbus_device_register(struct hv_device *child_device_obj)
1342 {
1343         struct kobject *kobj = &child_device_obj->device.kobj;
1344         int ret;
1345
1346         dev_set_name(&child_device_obj->device, "%pUl",
1347                      child_device_obj->channel->offermsg.offer.if_instance.b);
1348
1349         child_device_obj->device.bus = &hv_bus;
1350         child_device_obj->device.parent = &hv_acpi_dev->dev;
1351         child_device_obj->device.release = vmbus_device_release;
1352
1353         /*
1354          * Register with the LDM. This will kick off the driver/device
1355          * binding...which will eventually call vmbus_match() and vmbus_probe()
1356          */
1357         ret = device_register(&child_device_obj->device);
1358         if (ret) {
1359                 pr_err("Unable to register child device\n");
1360                 return ret;
1361         }
1362
1363         child_device_obj->channels_kset = kset_create_and_add("channels",
1364                                                               NULL, kobj);
1365         if (!child_device_obj->channels_kset) {
1366                 ret = -ENOMEM;
1367                 goto err_dev_unregister;
1368         }
1369
1370         ret = vmbus_add_channel_kobj(child_device_obj,
1371                                      child_device_obj->channel);
1372         if (ret) {
1373                 pr_err("Unable to register primary channeln");
1374                 goto err_kset_unregister;
1375         }
1376
1377         return 0;
1378
1379 err_kset_unregister:
1380         kset_unregister(child_device_obj->channels_kset);
1381
1382 err_dev_unregister:
1383         device_unregister(&child_device_obj->device);
1384         return ret;
1385 }
1386
1387 /*
1388  * vmbus_device_unregister - Remove the specified child device
1389  * from the vmbus.
1390  */
1391 void vmbus_device_unregister(struct hv_device *device_obj)
1392 {
1393         pr_debug("child device %s unregistered\n",
1394                 dev_name(&device_obj->device));
1395
1396         kset_unregister(device_obj->channels_kset);
1397
1398         /*
1399          * Kick off the process of unregistering the device.
1400          * This will call vmbus_remove() and eventually vmbus_device_release()
1401          */
1402         device_unregister(&device_obj->device);
1403 }
1404
1405
1406 /*
1407  * VMBUS is an acpi enumerated device. Get the information we
1408  * need from DSDT.
1409  */
1410 #define VTPM_BASE_ADDRESS 0xfed40000
1411 static acpi_status vmbus_walk_resources(struct acpi_resource *res, void *ctx)
1412 {
1413         resource_size_t start = 0;
1414         resource_size_t end = 0;
1415         struct resource *new_res;
1416         struct resource **old_res = &hyperv_mmio;
1417         struct resource **prev_res = NULL;
1418
1419         switch (res->type) {
1420
1421         /*
1422          * "Address" descriptors are for bus windows. Ignore
1423          * "memory" descriptors, which are for registers on
1424          * devices.
1425          */
1426         case ACPI_RESOURCE_TYPE_ADDRESS32:
1427                 start = res->data.address32.address.minimum;
1428                 end = res->data.address32.address.maximum;
1429                 break;
1430
1431         case ACPI_RESOURCE_TYPE_ADDRESS64:
1432                 start = res->data.address64.address.minimum;
1433                 end = res->data.address64.address.maximum;
1434                 break;
1435
1436         default:
1437                 /* Unused resource type */
1438                 return AE_OK;
1439
1440         }
1441         /*
1442          * Ignore ranges that are below 1MB, as they're not
1443          * necessary or useful here.
1444          */
1445         if (end < 0x100000)
1446                 return AE_OK;
1447
1448         new_res = kzalloc(sizeof(*new_res), GFP_ATOMIC);
1449         if (!new_res)
1450                 return AE_NO_MEMORY;
1451
1452         /* If this range overlaps the virtual TPM, truncate it. */
1453         if (end > VTPM_BASE_ADDRESS && start < VTPM_BASE_ADDRESS)
1454                 end = VTPM_BASE_ADDRESS;
1455
1456         new_res->name = "hyperv mmio";
1457         new_res->flags = IORESOURCE_MEM;
1458         new_res->start = start;
1459         new_res->end = end;
1460
1461         /*
1462          * If two ranges are adjacent, merge them.
1463          */
1464         do {
1465                 if (!*old_res) {
1466                         *old_res = new_res;
1467                         break;
1468                 }
1469
1470                 if (((*old_res)->end + 1) == new_res->start) {
1471                         (*old_res)->end = new_res->end;
1472                         kfree(new_res);
1473                         break;
1474                 }
1475
1476                 if ((*old_res)->start == new_res->end + 1) {
1477                         (*old_res)->start = new_res->start;
1478                         kfree(new_res);
1479                         break;
1480                 }
1481
1482                 if ((*old_res)->start > new_res->end) {
1483                         new_res->sibling = *old_res;
1484                         if (prev_res)
1485                                 (*prev_res)->sibling = new_res;
1486                         *old_res = new_res;
1487                         break;
1488                 }
1489
1490                 prev_res = old_res;
1491                 old_res = &(*old_res)->sibling;
1492
1493         } while (1);
1494
1495         return AE_OK;
1496 }
1497
1498 static int vmbus_acpi_remove(struct acpi_device *device)
1499 {
1500         struct resource *cur_res;
1501         struct resource *next_res;
1502
1503         if (hyperv_mmio) {
1504                 if (fb_mmio) {
1505                         __release_region(hyperv_mmio, fb_mmio->start,
1506                                          resource_size(fb_mmio));
1507                         fb_mmio = NULL;
1508                 }
1509
1510                 for (cur_res = hyperv_mmio; cur_res; cur_res = next_res) {
1511                         next_res = cur_res->sibling;
1512                         kfree(cur_res);
1513                 }
1514         }
1515
1516         return 0;
1517 }
1518
1519 static void vmbus_reserve_fb(void)
1520 {
1521         int size;
1522         /*
1523          * Make a claim for the frame buffer in the resource tree under the
1524          * first node, which will be the one below 4GB.  The length seems to
1525          * be underreported, particularly in a Generation 1 VM.  So start out
1526          * reserving a larger area and make it smaller until it succeeds.
1527          */
1528
1529         if (screen_info.lfb_base) {
1530                 if (efi_enabled(EFI_BOOT))
1531                         size = max_t(__u32, screen_info.lfb_size, 0x800000);
1532                 else
1533                         size = max_t(__u32, screen_info.lfb_size, 0x4000000);
1534
1535                 for (; !fb_mmio && (size >= 0x100000); size >>= 1) {
1536                         fb_mmio = __request_region(hyperv_mmio,
1537                                                    screen_info.lfb_base, size,
1538                                                    fb_mmio_name, 0);
1539                 }
1540         }
1541 }
1542
1543 /**
1544  * vmbus_allocate_mmio() - Pick a memory-mapped I/O range.
1545  * @new:                If successful, supplied a pointer to the
1546  *                      allocated MMIO space.
1547  * @device_obj:         Identifies the caller
1548  * @min:                Minimum guest physical address of the
1549  *                      allocation
1550  * @max:                Maximum guest physical address
1551  * @size:               Size of the range to be allocated
1552  * @align:              Alignment of the range to be allocated
1553  * @fb_overlap_ok:      Whether this allocation can be allowed
1554  *                      to overlap the video frame buffer.
1555  *
1556  * This function walks the resources granted to VMBus by the
1557  * _CRS object in the ACPI namespace underneath the parent
1558  * "bridge" whether that's a root PCI bus in the Generation 1
1559  * case or a Module Device in the Generation 2 case.  It then
1560  * attempts to allocate from the global MMIO pool in a way that
1561  * matches the constraints supplied in these parameters and by
1562  * that _CRS.
1563  *
1564  * Return: 0 on success, -errno on failure
1565  */
1566 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
1567                         resource_size_t min, resource_size_t max,
1568                         resource_size_t size, resource_size_t align,
1569                         bool fb_overlap_ok)
1570 {
1571         struct resource *iter, *shadow;
1572         resource_size_t range_min, range_max, start;
1573         const char *dev_n = dev_name(&device_obj->device);
1574         int retval;
1575
1576         retval = -ENXIO;
1577         down(&hyperv_mmio_lock);
1578
1579         /*
1580          * If overlaps with frame buffers are allowed, then first attempt to
1581          * make the allocation from within the reserved region.  Because it
1582          * is already reserved, no shadow allocation is necessary.
1583          */
1584         if (fb_overlap_ok && fb_mmio && !(min > fb_mmio->end) &&
1585             !(max < fb_mmio->start)) {
1586
1587                 range_min = fb_mmio->start;
1588                 range_max = fb_mmio->end;
1589                 start = (range_min + align - 1) & ~(align - 1);
1590                 for (; start + size - 1 <= range_max; start += align) {
1591                         *new = request_mem_region_exclusive(start, size, dev_n);
1592                         if (*new) {
1593                                 retval = 0;
1594                                 goto exit;
1595                         }
1596                 }
1597         }
1598
1599         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
1600                 if ((iter->start >= max) || (iter->end <= min))
1601                         continue;
1602
1603                 range_min = iter->start;
1604                 range_max = iter->end;
1605                 start = (range_min + align - 1) & ~(align - 1);
1606                 for (; start + size - 1 <= range_max; start += align) {
1607                         shadow = __request_region(iter, start, size, NULL,
1608                                                   IORESOURCE_BUSY);
1609                         if (!shadow)
1610                                 continue;
1611
1612                         *new = request_mem_region_exclusive(start, size, dev_n);
1613                         if (*new) {
1614                                 shadow->name = (char *)*new;
1615                                 retval = 0;
1616                                 goto exit;
1617                         }
1618
1619                         __release_region(iter, start, size);
1620                 }
1621         }
1622
1623 exit:
1624         up(&hyperv_mmio_lock);
1625         return retval;
1626 }
1627 EXPORT_SYMBOL_GPL(vmbus_allocate_mmio);
1628
1629 /**
1630  * vmbus_free_mmio() - Free a memory-mapped I/O range.
1631  * @start:              Base address of region to release.
1632  * @size:               Size of the range to be allocated
1633  *
1634  * This function releases anything requested by
1635  * vmbus_mmio_allocate().
1636  */
1637 void vmbus_free_mmio(resource_size_t start, resource_size_t size)
1638 {
1639         struct resource *iter;
1640
1641         down(&hyperv_mmio_lock);
1642         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
1643                 if ((iter->start >= start + size) || (iter->end <= start))
1644                         continue;
1645
1646                 __release_region(iter, start, size);
1647         }
1648         release_mem_region(start, size);
1649         up(&hyperv_mmio_lock);
1650
1651 }
1652 EXPORT_SYMBOL_GPL(vmbus_free_mmio);
1653
1654 static int vmbus_acpi_add(struct acpi_device *device)
1655 {
1656         acpi_status result;
1657         int ret_val = -ENODEV;
1658         struct acpi_device *ancestor;
1659
1660         hv_acpi_dev = device;
1661
1662         result = acpi_walk_resources(device->handle, METHOD_NAME__CRS,
1663                                         vmbus_walk_resources, NULL);
1664
1665         if (ACPI_FAILURE(result))
1666                 goto acpi_walk_err;
1667         /*
1668          * Some ancestor of the vmbus acpi device (Gen1 or Gen2
1669          * firmware) is the VMOD that has the mmio ranges. Get that.
1670          */
1671         for (ancestor = device->parent; ancestor; ancestor = ancestor->parent) {
1672                 result = acpi_walk_resources(ancestor->handle, METHOD_NAME__CRS,
1673                                              vmbus_walk_resources, NULL);
1674
1675                 if (ACPI_FAILURE(result))
1676                         continue;
1677                 if (hyperv_mmio) {
1678                         vmbus_reserve_fb();
1679                         break;
1680                 }
1681         }
1682         ret_val = 0;
1683
1684 acpi_walk_err:
1685         complete(&probe_event);
1686         if (ret_val)
1687                 vmbus_acpi_remove(device);
1688         return ret_val;
1689 }
1690
1691 static const struct acpi_device_id vmbus_acpi_device_ids[] = {
1692         {"VMBUS", 0},
1693         {"VMBus", 0},
1694         {"", 0},
1695 };
1696 MODULE_DEVICE_TABLE(acpi, vmbus_acpi_device_ids);
1697
1698 static struct acpi_driver vmbus_acpi_driver = {
1699         .name = "vmbus",
1700         .ids = vmbus_acpi_device_ids,
1701         .ops = {
1702                 .add = vmbus_acpi_add,
1703                 .remove = vmbus_acpi_remove,
1704         },
1705 };
1706
1707 static void hv_kexec_handler(void)
1708 {
1709         hv_synic_clockevents_cleanup();
1710         vmbus_initiate_unload(false);
1711         vmbus_connection.conn_state = DISCONNECTED;
1712         /* Make sure conn_state is set as hv_synic_cleanup checks for it */
1713         mb();
1714         cpuhp_remove_state(hyperv_cpuhp_online);
1715         hyperv_cleanup();
1716 };
1717
1718 static void hv_crash_handler(struct pt_regs *regs)
1719 {
1720         vmbus_initiate_unload(true);
1721         /*
1722          * In crash handler we can't schedule synic cleanup for all CPUs,
1723          * doing the cleanup for current CPU only. This should be sufficient
1724          * for kdump.
1725          */
1726         vmbus_connection.conn_state = DISCONNECTED;
1727         hv_synic_cleanup(smp_processor_id());
1728         hyperv_cleanup();
1729 };
1730
1731 static int __init hv_acpi_init(void)
1732 {
1733         int ret, t;
1734
1735         if (!hv_is_hyperv_initialized())
1736                 return -ENODEV;
1737
1738         init_completion(&probe_event);
1739
1740         /*
1741          * Get ACPI resources first.
1742          */
1743         ret = acpi_bus_register_driver(&vmbus_acpi_driver);
1744
1745         if (ret)
1746                 return ret;
1747
1748         t = wait_for_completion_timeout(&probe_event, 5*HZ);
1749         if (t == 0) {
1750                 ret = -ETIMEDOUT;
1751                 goto cleanup;
1752         }
1753
1754         ret = vmbus_bus_init();
1755         if (ret)
1756                 goto cleanup;
1757
1758         hv_setup_kexec_handler(hv_kexec_handler);
1759         hv_setup_crash_handler(hv_crash_handler);
1760
1761         return 0;
1762
1763 cleanup:
1764         acpi_bus_unregister_driver(&vmbus_acpi_driver);
1765         hv_acpi_dev = NULL;
1766         return ret;
1767 }
1768
1769 static void __exit vmbus_exit(void)
1770 {
1771         int cpu;
1772
1773         hv_remove_kexec_handler();
1774         hv_remove_crash_handler();
1775         vmbus_connection.conn_state = DISCONNECTED;
1776         hv_synic_clockevents_cleanup();
1777         vmbus_disconnect();
1778         hv_remove_vmbus_irq();
1779         for_each_online_cpu(cpu) {
1780                 struct hv_per_cpu_context *hv_cpu
1781                         = per_cpu_ptr(hv_context.cpu_context, cpu);
1782
1783                 tasklet_kill(&hv_cpu->msg_dpc);
1784         }
1785         vmbus_free_channels();
1786
1787         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1788                 unregister_die_notifier(&hyperv_die_block);
1789                 atomic_notifier_chain_unregister(&panic_notifier_list,
1790                                                  &hyperv_panic_block);
1791         }
1792         bus_unregister(&hv_bus);
1793
1794         cpuhp_remove_state(hyperv_cpuhp_online);
1795         hv_synic_free();
1796         acpi_bus_unregister_driver(&vmbus_acpi_driver);
1797 }
1798
1799
1800 MODULE_LICENSE("GPL");
1801
1802 subsys_initcall(hv_acpi_init);
1803 module_exit(vmbus_exit);