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/mshyperv.h>
  41 #include <linux/notifier.h>
  42 #include <linux/ptrace.h>
  43 #include <linux/screen_info.h>
  44 #include <linux/kdebug.h>
  45 #include <linux/efi.h>
  46 #include <linux/random.h>
  47 #include "hyperv_vmbus.h"
  48
  49 struct vmbus_dynid {
  50         struct list_head node;
  51         struct hv_vmbus_device_id id;
  52 };
  53
  54 static struct acpi_device  *hv_acpi_dev;
  55
  56 static struct completion probe_event;
  57
  58 static int hyperv_cpuhp_online;
  59
  60 static int hyperv_panic_event(struct notifier_block *nb, unsigned long val,
  61                               void *args)
  62 {
  63         struct pt_regs *regs;
  64
  65         regs = current_pt_regs();
  66
  67         hyperv_report_panic(regs, val);
  68         return NOTIFY_DONE;
  69 }
  70
  71 static int hyperv_die_event(struct notifier_block *nb, unsigned long val,
  72                             void *args)
  73 {
  74         struct die_args *die = (struct die_args *)args;
  75         struct pt_regs *regs = die->regs;
  76
  77         hyperv_report_panic(regs, val);
  78         return NOTIFY_DONE;
  79 }
  80
  81 static struct notifier_block hyperv_die_block = {
  82         .notifier_call = hyperv_die_event,
  83 };
  84 static struct notifier_block hyperv_panic_block = {
  85         .notifier_call = hyperv_panic_event,
  86 };
  87
  88 static const char *fb_mmio_name = "fb_range";
  89 static struct resource *fb_mmio;
  90 static struct resource *hyperv_mmio;
  91 static DEFINE_SEMAPHORE(hyperv_mmio_lock);
  92
  93 static int vmbus_exists(void)
  94 {
  95         if (hv_acpi_dev == NULL)
  96                 return -ENODEV;
  97
  98         return 0;
  99 }
 100
 101 #define VMBUS_ALIAS_LEN ((sizeof((struct hv_vmbus_device_id *)0)->guid) * 2)
 102 static void print_alias_name(struct hv_device *hv_dev, char *alias_name)
 103 {
 104         int i;
 105         for (i = 0; i < VMBUS_ALIAS_LEN; i += 2)
 106                 sprintf(&alias_name[i], "%02x", hv_dev->dev_type.b[i/2]);
 107 }
 108
 109 static u8 channel_monitor_group(const struct vmbus_channel *channel)
 110 {
 111         return (u8)channel->offermsg.monitorid / 32;
 112 }
 113
 114 static u8 channel_monitor_offset(const struct vmbus_channel *channel)
 115 {
 116         return (u8)channel->offermsg.monitorid % 32;
 117 }
 118
 119 static u32 channel_pending(const struct vmbus_channel *channel,
 120                            const struct hv_monitor_page *monitor_page)
 121 {
 122         u8 monitor_group = channel_monitor_group(channel);
 123
 124         return monitor_page->trigger_group[monitor_group].pending;
 125 }
 126
 127 static u32 channel_latency(const struct vmbus_channel *channel,
 128                            const struct hv_monitor_page *monitor_page)
 129 {
 130         u8 monitor_group = channel_monitor_group(channel);
 131         u8 monitor_offset = channel_monitor_offset(channel);
 132
 133         return monitor_page->latency[monitor_group][monitor_offset];
 134 }
 135
 136 static u32 channel_conn_id(struct vmbus_channel *channel,
 137                            struct hv_monitor_page *monitor_page)
 138 {
 139         u8 monitor_group = channel_monitor_group(channel);
 140         u8 monitor_offset = channel_monitor_offset(channel);
 141         return monitor_page->parameter[monitor_group][monitor_offset].connectionid.u.id;
 142 }
 143
 144 static ssize_t id_show(struct device *dev, struct device_attribute *dev_attr,
 145                        char *buf)
 146 {
 147         struct hv_device *hv_dev = device_to_hv_device(dev);
 148
 149         if (!hv_dev->channel)
 150                 return -ENODEV;
 151         return sprintf(buf, "%d\n", hv_dev->channel->offermsg.child_relid);
 152 }
 153 static DEVICE_ATTR_RO(id);
 154
 155 static ssize_t state_show(struct device *dev, struct device_attribute *dev_attr,
 156                           char *buf)
 157 {
 158         struct hv_device *hv_dev = device_to_hv_device(dev);
 159
 160         if (!hv_dev->channel)
 161                 return -ENODEV;
 162         return sprintf(buf, "%d\n", hv_dev->channel->state);
 163 }
 164 static DEVICE_ATTR_RO(state);
 165
 166 static ssize_t monitor_id_show(struct device *dev,
 167                                struct device_attribute *dev_attr, char *buf)
 168 {
 169         struct hv_device *hv_dev = device_to_hv_device(dev);
 170
 171         if (!hv_dev->channel)
 172                 return -ENODEV;
 173         return sprintf(buf, "%d\n", hv_dev->channel->offermsg.monitorid);
 174 }
 175 static DEVICE_ATTR_RO(monitor_id);
 176
 177 static ssize_t class_id_show(struct device *dev,
 178                                struct device_attribute *dev_attr, char *buf)
 179 {
 180         struct hv_device *hv_dev = device_to_hv_device(dev);
 181
 182         if (!hv_dev->channel)
 183                 return -ENODEV;
 184         return sprintf(buf, "{%pUl}\n",
 185                        hv_dev->channel->offermsg.offer.if_type.b);
 186 }
 187 static DEVICE_ATTR_RO(class_id);
 188
 189 static ssize_t device_id_show(struct device *dev,
 190                               struct device_attribute *dev_attr, char *buf)
 191 {
 192         struct hv_device *hv_dev = device_to_hv_device(dev);
 193
 194         if (!hv_dev->channel)
 195                 return -ENODEV;
 196         return sprintf(buf, "{%pUl}\n",
 197                        hv_dev->channel->offermsg.offer.if_instance.b);
 198 }
 199 static DEVICE_ATTR_RO(device_id);
 200
 201 static ssize_t modalias_show(struct device *dev,
 202                              struct device_attribute *dev_attr, char *buf)
 203 {
 204         struct hv_device *hv_dev = device_to_hv_device(dev);
 205         char alias_name[VMBUS_ALIAS_LEN + 1];
 206
 207         print_alias_name(hv_dev, alias_name);
 208         return sprintf(buf, "vmbus:%s\n", alias_name);
 209 }
 210 static DEVICE_ATTR_RO(modalias);
 211
 212 static ssize_t server_monitor_pending_show(struct device *dev,
 213                                            struct device_attribute *dev_attr,
 214                                            char *buf)
 215 {
 216         struct hv_device *hv_dev = device_to_hv_device(dev);
 217
 218         if (!hv_dev->channel)
 219                 return -ENODEV;
 220         return sprintf(buf, "%d\n",
 221                        channel_pending(hv_dev->channel,
 222                                        vmbus_connection.monitor_pages[1]));
 223 }
 224 static DEVICE_ATTR_RO(server_monitor_pending);
 225
 226 static ssize_t client_monitor_pending_show(struct device *dev,
 227                                            struct device_attribute *dev_attr,
 228                                            char *buf)
 229 {
 230         struct hv_device *hv_dev = device_to_hv_device(dev);
 231
 232         if (!hv_dev->channel)
 233                 return -ENODEV;
 234         return sprintf(buf, "%d\n",
 235                        channel_pending(hv_dev->channel,
 236                                        vmbus_connection.monitor_pages[1]));
 237 }
 238 static DEVICE_ATTR_RO(client_monitor_pending);
 239
 240 static ssize_t server_monitor_latency_show(struct device *dev,
 241                                            struct device_attribute *dev_attr,
 242                                            char *buf)
 243 {
 244         struct hv_device *hv_dev = device_to_hv_device(dev);
 245
 246         if (!hv_dev->channel)
 247                 return -ENODEV;
 248         return sprintf(buf, "%d\n",
 249                        channel_latency(hv_dev->channel,
 250                                        vmbus_connection.monitor_pages[0]));
 251 }
 252 static DEVICE_ATTR_RO(server_monitor_latency);
 253
 254 static ssize_t client_monitor_latency_show(struct device *dev,
 255                                            struct device_attribute *dev_attr,
 256                                            char *buf)
 257 {
 258         struct hv_device *hv_dev = device_to_hv_device(dev);
 259
 260         if (!hv_dev->channel)
 261                 return -ENODEV;
 262         return sprintf(buf, "%d\n",
 263                        channel_latency(hv_dev->channel,
 264                                        vmbus_connection.monitor_pages[1]));
 265 }
 266 static DEVICE_ATTR_RO(client_monitor_latency);
 267
 268 static ssize_t server_monitor_conn_id_show(struct device *dev,
 269                                            struct device_attribute *dev_attr,
 270                                            char *buf)
 271 {
 272         struct hv_device *hv_dev = device_to_hv_device(dev);
 273
 274         if (!hv_dev->channel)
 275                 return -ENODEV;
 276         return sprintf(buf, "%d\n",
 277                        channel_conn_id(hv_dev->channel,
 278                                        vmbus_connection.monitor_pages[0]));
 279 }
 280 static DEVICE_ATTR_RO(server_monitor_conn_id);
 281
 282 static ssize_t client_monitor_conn_id_show(struct device *dev,
 283                                            struct device_attribute *dev_attr,
 284                                            char *buf)
 285 {
 286         struct hv_device *hv_dev = device_to_hv_device(dev);
 287
 288         if (!hv_dev->channel)
 289                 return -ENODEV;
 290         return sprintf(buf, "%d\n",
 291                        channel_conn_id(hv_dev->channel,
 292                                        vmbus_connection.monitor_pages[1]));
 293 }
 294 static DEVICE_ATTR_RO(client_monitor_conn_id);
 295
 296 static ssize_t out_intr_mask_show(struct device *dev,
 297                                   struct device_attribute *dev_attr, char *buf)
 298 {
 299         struct hv_device *hv_dev = device_to_hv_device(dev);
 300         struct hv_ring_buffer_debug_info outbound;
 301
 302         if (!hv_dev->channel)
 303                 return -ENODEV;
 304         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
 305         return sprintf(buf, "%d\n", outbound.current_interrupt_mask);
 306 }
 307 static DEVICE_ATTR_RO(out_intr_mask);
 308
 309 static ssize_t out_read_index_show(struct device *dev,
 310                                    struct device_attribute *dev_attr, char *buf)
 311 {
 312         struct hv_device *hv_dev = device_to_hv_device(dev);
 313         struct hv_ring_buffer_debug_info outbound;
 314
 315         if (!hv_dev->channel)
 316                 return -ENODEV;
 317         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
 318         return sprintf(buf, "%d\n", outbound.current_read_index);
 319 }
 320 static DEVICE_ATTR_RO(out_read_index);
 321
 322 static ssize_t out_write_index_show(struct device *dev,
 323                                     struct device_attribute *dev_attr,
 324                                     char *buf)
 325 {
 326         struct hv_device *hv_dev = device_to_hv_device(dev);
 327         struct hv_ring_buffer_debug_info outbound;
 328
 329         if (!hv_dev->channel)
 330                 return -ENODEV;
 331         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
 332         return sprintf(buf, "%d\n", outbound.current_write_index);
 333 }
 334 static DEVICE_ATTR_RO(out_write_index);
 335
 336 static ssize_t out_read_bytes_avail_show(struct device *dev,
 337                                          struct device_attribute *dev_attr,
 338                                          char *buf)
 339 {
 340         struct hv_device *hv_dev = device_to_hv_device(dev);
 341         struct hv_ring_buffer_debug_info outbound;
 342
 343         if (!hv_dev->channel)
 344                 return -ENODEV;
 345         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
 346         return sprintf(buf, "%d\n", outbound.bytes_avail_toread);
 347 }
 348 static DEVICE_ATTR_RO(out_read_bytes_avail);
 349
 350 static ssize_t out_write_bytes_avail_show(struct device *dev,
 351                                           struct device_attribute *dev_attr,
 352                                           char *buf)
 353 {
 354         struct hv_device *hv_dev = device_to_hv_device(dev);
 355         struct hv_ring_buffer_debug_info outbound;
 356
 357         if (!hv_dev->channel)
 358                 return -ENODEV;
 359         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
 360         return sprintf(buf, "%d\n", outbound.bytes_avail_towrite);
 361 }
 362 static DEVICE_ATTR_RO(out_write_bytes_avail);
 363
 364 static ssize_t in_intr_mask_show(struct device *dev,
 365                                  struct device_attribute *dev_attr, char *buf)
 366 {
 367         struct hv_device *hv_dev = device_to_hv_device(dev);
 368         struct hv_ring_buffer_debug_info inbound;
 369
 370         if (!hv_dev->channel)
 371                 return -ENODEV;
 372         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 373         return sprintf(buf, "%d\n", inbound.current_interrupt_mask);
 374 }
 375 static DEVICE_ATTR_RO(in_intr_mask);
 376
 377 static ssize_t in_read_index_show(struct device *dev,
 378                                   struct device_attribute *dev_attr, char *buf)
 379 {
 380         struct hv_device *hv_dev = device_to_hv_device(dev);
 381         struct hv_ring_buffer_debug_info inbound;
 382
 383         if (!hv_dev->channel)
 384                 return -ENODEV;
 385         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 386         return sprintf(buf, "%d\n", inbound.current_read_index);
 387 }
 388 static DEVICE_ATTR_RO(in_read_index);
 389
 390 static ssize_t in_write_index_show(struct device *dev,
 391                                    struct device_attribute *dev_attr, char *buf)
 392 {
 393         struct hv_device *hv_dev = device_to_hv_device(dev);
 394         struct hv_ring_buffer_debug_info inbound;
 395
 396         if (!hv_dev->channel)
 397                 return -ENODEV;
 398         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 399         return sprintf(buf, "%d\n", inbound.current_write_index);
 400 }
 401 static DEVICE_ATTR_RO(in_write_index);
 402
 403 static ssize_t in_read_bytes_avail_show(struct device *dev,
 404                                         struct device_attribute *dev_attr,
 405                                         char *buf)
 406 {
 407         struct hv_device *hv_dev = device_to_hv_device(dev);
 408         struct hv_ring_buffer_debug_info inbound;
 409
 410         if (!hv_dev->channel)
 411                 return -ENODEV;
 412         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 413         return sprintf(buf, "%d\n", inbound.bytes_avail_toread);
 414 }
 415 static DEVICE_ATTR_RO(in_read_bytes_avail);
 416
 417 static ssize_t in_write_bytes_avail_show(struct device *dev,
 418                                          struct device_attribute *dev_attr,
 419                                          char *buf)
 420 {
 421         struct hv_device *hv_dev = device_to_hv_device(dev);
 422         struct hv_ring_buffer_debug_info inbound;
 423
 424         if (!hv_dev->channel)
 425                 return -ENODEV;
 426         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 427         return sprintf(buf, "%d\n", inbound.bytes_avail_towrite);
 428 }
 429 static DEVICE_ATTR_RO(in_write_bytes_avail);
 430
 431 static ssize_t channel_vp_mapping_show(struct device *dev,
 432                                        struct device_attribute *dev_attr,
 433                                        char *buf)
 434 {
 435         struct hv_device *hv_dev = device_to_hv_device(dev);
 436         struct vmbus_channel *channel = hv_dev->channel, *cur_sc;
 437         unsigned long flags;
 438         int buf_size = PAGE_SIZE, n_written, tot_written;
 439         struct list_head *cur;
 440
 441         if (!channel)
 442                 return -ENODEV;
 443
 444         tot_written = snprintf(buf, buf_size, "%u:%u\n",
 445                 channel->offermsg.child_relid, channel->target_cpu);
 446
 447         spin_lock_irqsave(&channel->lock, flags);
 448
 449         list_for_each(cur, &channel->sc_list) {
 450                 if (tot_written >= buf_size - 1)
 451                         break;
 452
 453                 cur_sc = list_entry(cur, struct vmbus_channel, sc_list);
 454                 n_written = scnprintf(buf + tot_written,
 455                                      buf_size - tot_written,
 456                                      "%u:%u\n",
 457                                      cur_sc->offermsg.child_relid,
 458                                      cur_sc->target_cpu);
 459                 tot_written += n_written;
 460         }
 461
 462         spin_unlock_irqrestore(&channel->lock, flags);
 463
 464         return tot_written;
 465 }
 466 static DEVICE_ATTR_RO(channel_vp_mapping);
 467
 468 static ssize_t vendor_show(struct device *dev,
 469                            struct device_attribute *dev_attr,
 470                            char *buf)
 471 {
 472         struct hv_device *hv_dev = device_to_hv_device(dev);
 473         return sprintf(buf, "0x%x\n", hv_dev->vendor_id);
 474 }
 475 static DEVICE_ATTR_RO(vendor);
 476
 477 static ssize_t device_show(struct device *dev,
 478                            struct device_attribute *dev_attr,
 479                            char *buf)
 480 {
 481         struct hv_device *hv_dev = device_to_hv_device(dev);
 482         return sprintf(buf, "0x%x\n", hv_dev->device_id);
 483 }
 484 static DEVICE_ATTR_RO(device);
 485
 486 /* Set up per device attributes in /sys/bus/vmbus/devices/<bus device> */
 487 static struct attribute *vmbus_dev_attrs[] = {
 488         &dev_attr_id.attr,
 489         &dev_attr_state.attr,
 490         &dev_attr_monitor_id.attr,
 491         &dev_attr_class_id.attr,
 492         &dev_attr_device_id.attr,
 493         &dev_attr_modalias.attr,
 494         &dev_attr_server_monitor_pending.attr,
 495         &dev_attr_client_monitor_pending.attr,
 496         &dev_attr_server_monitor_latency.attr,
 497         &dev_attr_client_monitor_latency.attr,
 498         &dev_attr_server_monitor_conn_id.attr,
 499         &dev_attr_client_monitor_conn_id.attr,
 500         &dev_attr_out_intr_mask.attr,
 501         &dev_attr_out_read_index.attr,
 502         &dev_attr_out_write_index.attr,
 503         &dev_attr_out_read_bytes_avail.attr,
 504         &dev_attr_out_write_bytes_avail.attr,
 505         &dev_attr_in_intr_mask.attr,
 506         &dev_attr_in_read_index.attr,
 507         &dev_attr_in_write_index.attr,
 508         &dev_attr_in_read_bytes_avail.attr,
 509         &dev_attr_in_write_bytes_avail.attr,
 510         &dev_attr_channel_vp_mapping.attr,
 511         &dev_attr_vendor.attr,
 512         &dev_attr_device.attr,
 513         NULL,
 514 };
 515 ATTRIBUTE_GROUPS(vmbus_dev);
 516
 517 /*
 518  * vmbus_uevent - add uevent for our device
 519  *
 520  * This routine is invoked when a device is added or removed on the vmbus to
 521  * generate a uevent to udev in the userspace. The udev will then look at its
 522  * rule and the uevent generated here to load the appropriate driver
 523  *
 524  * The alias string will be of the form vmbus:guid where guid is the string
 525  * representation of the device guid (each byte of the guid will be
 526  * represented with two hex characters.
 527  */
 528 static int vmbus_uevent(struct device *device, struct kobj_uevent_env *env)
 529 {
 530         struct hv_device *dev = device_to_hv_device(device);
 531         int ret;
 532         char alias_name[VMBUS_ALIAS_LEN + 1];
 533
 534         print_alias_name(dev, alias_name);
 535         ret = add_uevent_var(env, "MODALIAS=vmbus:%s", alias_name);
 536         return ret;
 537 }
 538
 539 static const uuid_le null_guid;
 540
 541 static inline bool is_null_guid(const uuid_le *guid)
 542 {
 543         if (uuid_le_cmp(*guid, null_guid))
 544                 return false;
 545         return true;
 546 }
 547
 548 /*
 549  * Return a matching hv_vmbus_device_id pointer.
 550  * If there is no match, return NULL.
 551  */
 552 static const struct hv_vmbus_device_id *hv_vmbus_get_id(struct hv_driver *drv,
 553                                         const uuid_le *guid)
 554 {
 555         const struct hv_vmbus_device_id *id = NULL;
 556         struct vmbus_dynid *dynid;
 557
 558         /* Look at the dynamic ids first, before the static ones */
 559         spin_lock(&drv->dynids.lock);
 560         list_for_each_entry(dynid, &drv->dynids.list, node) {
 561                 if (!uuid_le_cmp(dynid->id.guid, *guid)) {
 562                         id = &dynid->id;
 563                         break;
 564                 }
 565         }
 566         spin_unlock(&drv->dynids.lock);
 567
 568         if (id)
 569                 return id;
 570
 571         id = drv->id_table;
 572         if (id == NULL)
 573                 return NULL; /* empty device table */
 574
 575         for (; !is_null_guid(&id->guid); id++)
 576                 if (!uuid_le_cmp(id->guid, *guid))
 577                         return id;
 578
 579         return NULL;
 580 }
 581
 582 /* vmbus_add_dynid - add a new device ID to this driver and re-probe devices */
 583 static int vmbus_add_dynid(struct hv_driver *drv, uuid_le *guid)
 584 {
 585         struct vmbus_dynid *dynid;
 586
 587         dynid = kzalloc(sizeof(*dynid), GFP_KERNEL);
 588         if (!dynid)
 589                 return -ENOMEM;
 590
 591         dynid->id.guid = *guid;
 592
 593         spin_lock(&drv->dynids.lock);
 594         list_add_tail(&dynid->node, &drv->dynids.list);
 595         spin_unlock(&drv->dynids.lock);
 596
 597         return driver_attach(&drv->driver);
 598 }
 599
 600 static void vmbus_free_dynids(struct hv_driver *drv)
 601 {
 602         struct vmbus_dynid *dynid, *n;
 603
 604         spin_lock(&drv->dynids.lock);
 605         list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
 606                 list_del(&dynid->node);
 607                 kfree(dynid);
 608         }
 609         spin_unlock(&drv->dynids.lock);
 610 }
 611
 612 /*
 613  * store_new_id - sysfs frontend to vmbus_add_dynid()
 614  *
 615  * Allow GUIDs to be added to an existing driver via sysfs.
 616  */
 617 static ssize_t new_id_store(struct device_driver *driver, const char *buf,
 618                             size_t count)
 619 {
 620         struct hv_driver *drv = drv_to_hv_drv(driver);
 621         uuid_le guid;
 622         ssize_t retval;
 623
 624         retval = uuid_le_to_bin(buf, &guid);
 625         if (retval)
 626                 return retval;
 627
 628         if (hv_vmbus_get_id(drv, &guid))
 629                 return -EEXIST;
 630
 631         retval = vmbus_add_dynid(drv, &guid);
 632         if (retval)
 633                 return retval;
 634         return count;
 635 }
 636 static DRIVER_ATTR_WO(new_id);
 637
 638 /*
 639  * store_remove_id - remove a PCI device ID from this driver
 640  *
 641  * Removes a dynamic pci device ID to this driver.
 642  */
 643 static ssize_t remove_id_store(struct device_driver *driver, const char *buf,
 644                                size_t count)
 645 {
 646         struct hv_driver *drv = drv_to_hv_drv(driver);
 647         struct vmbus_dynid *dynid, *n;
 648         uuid_le guid;
 649         ssize_t retval;
 650
 651         retval = uuid_le_to_bin(buf, &guid);
 652         if (retval)
 653                 return retval;
 654
 655         retval = -ENODEV;
 656         spin_lock(&drv->dynids.lock);
 657         list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
 658                 struct hv_vmbus_device_id *id = &dynid->id;
 659
 660                 if (!uuid_le_cmp(id->guid, guid)) {
 661                         list_del(&dynid->node);
 662                         kfree(dynid);
 663                         retval = count;
 664                         break;
 665                 }
 666         }
 667         spin_unlock(&drv->dynids.lock);
 668
 669         return retval;
 670 }
 671 static DRIVER_ATTR_WO(remove_id);
 672
 673 static struct attribute *vmbus_drv_attrs[] = {
 674         &driver_attr_new_id.attr,
 675         &driver_attr_remove_id.attr,
 676         NULL,
 677 };
 678 ATTRIBUTE_GROUPS(vmbus_drv);
 679
 680
 681 /*
 682  * vmbus_match - Attempt to match the specified device to the specified driver
 683  */
 684 static int vmbus_match(struct device *device, struct device_driver *driver)
 685 {
 686         struct hv_driver *drv = drv_to_hv_drv(driver);
 687         struct hv_device *hv_dev = device_to_hv_device(device);
 688
 689         /* The hv_sock driver handles all hv_sock offers. */
 690         if (is_hvsock_channel(hv_dev->channel))
 691                 return drv->hvsock;
 692
 693         if (hv_vmbus_get_id(drv, &hv_dev->dev_type))
 694                 return 1;
 695
 696         return 0;
 697 }
 698
 699 /*
 700  * vmbus_probe - Add the new vmbus's child device
 701  */
 702 static int vmbus_probe(struct device *child_device)
 703 {
 704         int ret = 0;
 705         struct hv_driver *drv =
 706                         drv_to_hv_drv(child_device->driver);
 707         struct hv_device *dev = device_to_hv_device(child_device);
 708         const struct hv_vmbus_device_id *dev_id;
 709
 710         dev_id = hv_vmbus_get_id(drv, &dev->dev_type);
 711         if (drv->probe) {
 712                 ret = drv->probe(dev, dev_id);
 713                 if (ret != 0)
 714                         pr_err("probe failed for device %s (%d)\n",
 715                                dev_name(child_device), ret);
 716
 717         } else {
 718                 pr_err("probe not set for driver %s\n",
 719                        dev_name(child_device));
 720                 ret = -ENODEV;
 721         }
 722         return ret;
 723 }
 724
 725 /*
 726  * vmbus_remove - Remove a vmbus device
 727  */
 728 static int vmbus_remove(struct device *child_device)
 729 {
 730         struct hv_driver *drv;
 731         struct hv_device *dev = device_to_hv_device(child_device);
 732
 733         if (child_device->driver) {
 734                 drv = drv_to_hv_drv(child_device->driver);
 735                 if (drv->remove)
 736                         drv->remove(dev);
 737         }
 738
 739         return 0;
 740 }
 741
 742
 743 /*
 744  * vmbus_shutdown - Shutdown a vmbus device
 745  */
 746 static void vmbus_shutdown(struct device *child_device)
 747 {
 748         struct hv_driver *drv;
 749         struct hv_device *dev = device_to_hv_device(child_device);
 750
 751
 752         /* The device may not be attached yet */
 753         if (!child_device->driver)
 754                 return;
 755
 756         drv = drv_to_hv_drv(child_device->driver);
 757
 758         if (drv->shutdown)
 759                 drv->shutdown(dev);
 760 }
 761
 762
 763 /*
 764  * vmbus_device_release - Final callback release of the vmbus child device
 765  */
 766 static void vmbus_device_release(struct device *device)
 767 {
 768         struct hv_device *hv_dev = device_to_hv_device(device);
 769         struct vmbus_channel *channel = hv_dev->channel;
 770
 771         mutex_lock(&vmbus_connection.channel_mutex);
 772         hv_process_channel_removal(channel->offermsg.child_relid);
 773         mutex_unlock(&vmbus_connection.channel_mutex);
 774         kfree(hv_dev);
 775
 776 }
 777
 778 /* The one and only one */
 779 static struct bus_type  hv_bus = {
 780         .name =         "vmbus",
 781         .match =                vmbus_match,
 782         .shutdown =             vmbus_shutdown,
 783         .remove =               vmbus_remove,
 784         .probe =                vmbus_probe,
 785         .uevent =               vmbus_uevent,
 786         .dev_groups =           vmbus_dev_groups,
 787         .drv_groups =           vmbus_drv_groups,
 788 };
 789
 790 struct onmessage_work_context {
 791         struct work_struct work;
 792         struct hv_message msg;
 793 };
 794
 795 static void vmbus_onmessage_work(struct work_struct *work)
 796 {
 797         struct onmessage_work_context *ctx;
 798
 799         /* Do not process messages if we're in DISCONNECTED state */
 800         if (vmbus_connection.conn_state == DISCONNECTED)
 801                 return;
 802
 803         ctx = container_of(work, struct onmessage_work_context,
 804                            work);
 805         vmbus_onmessage(&ctx->msg);
 806         kfree(ctx);
 807 }
 808
 809 static void hv_process_timer_expiration(struct hv_message *msg,
 810                                         struct hv_per_cpu_context *hv_cpu)
 811 {
 812         struct clock_event_device *dev = hv_cpu->clk_evt;
 813
 814         if (dev->event_handler)
 815                 dev->event_handler(dev);
 816
 817         vmbus_signal_eom(msg, HVMSG_TIMER_EXPIRED);
 818 }
 819
 820 void vmbus_on_msg_dpc(unsigned long data)
 821 {
 822         struct hv_per_cpu_context *hv_cpu = (void *)data;
 823         void *page_addr = hv_cpu->synic_message_page;
 824         struct hv_message *msg = (struct hv_message *)page_addr +
 825                                   VMBUS_MESSAGE_SINT;
 826         struct vmbus_channel_message_header *hdr;
 827         const struct vmbus_channel_message_table_entry *entry;
 828         struct onmessage_work_context *ctx;
 829         u32 message_type = msg->header.message_type;
 830
 831         if (message_type == HVMSG_NONE)
 832                 /* no msg */
 833                 return;
 834
 835         hdr = (struct vmbus_channel_message_header *)msg->u.payload;
 836
 837         trace_vmbus_on_msg_dpc(hdr);
 838
 839         if (hdr->msgtype >= CHANNELMSG_COUNT) {
 840                 WARN_ONCE(1, "unknown msgtype=%d\n", hdr->msgtype);
 841                 goto msg_handled;
 842         }
 843
 844         entry = &channel_message_table[hdr->msgtype];
 845         if (entry->handler_type == VMHT_BLOCKING) {
 846                 ctx = kmalloc(sizeof(*ctx), GFP_ATOMIC);
 847                 if (ctx == NULL)
 848                         return;
 849
 850                 INIT_WORK(&ctx->work, vmbus_onmessage_work);
 851                 memcpy(&ctx->msg, msg, sizeof(*msg));
 852
 853                 /*
 854                  * The host can generate a rescind message while we
 855                  * may still be handling the original offer. We deal with
 856                  * this condition by ensuring the processing is done on the
 857                  * same CPU.
 858                  */
 859                 switch (hdr->msgtype) {
 860                 case CHANNELMSG_RESCIND_CHANNELOFFER:
 861                         /*
 862                          * If we are handling the rescind message;
 863                          * schedule the work on the global work queue.
 864                          */
 865                         schedule_work_on(vmbus_connection.connect_cpu,
 866                                          &ctx->work);
 867                         break;
 868
 869                 case CHANNELMSG_OFFERCHANNEL:
 870                         atomic_inc(&vmbus_connection.offer_in_progress);
 871                         queue_work_on(vmbus_connection.connect_cpu,
 872                                       vmbus_connection.work_queue,
 873                                       &ctx->work);
 874                         break;
 875
 876                 default:
 877                         queue_work(vmbus_connection.work_queue, &ctx->work);
 878                 }
 879         } else
 880                 entry->message_handler(hdr);
 881
 882 msg_handled:
 883         vmbus_signal_eom(msg, message_type);
 884 }
 885
 886
 887 /*
 888  * Direct callback for channels using other deferred processing
 889  */
 890 static void vmbus_channel_isr(struct vmbus_channel *channel)
 891 {
 892         void (*callback_fn)(void *);
 893
 894         callback_fn = READ_ONCE(channel->onchannel_callback);
 895         if (likely(callback_fn != NULL))
 896                 (*callback_fn)(channel->channel_callback_context);
 897 }
 898
 899 /*
 900  * Schedule all channels with events pending
 901  */
 902 static void vmbus_chan_sched(struct hv_per_cpu_context *hv_cpu)
 903 {
 904         unsigned long *recv_int_page;
 905         u32 maxbits, relid;
 906
 907         if (vmbus_proto_version < VERSION_WIN8) {
 908                 maxbits = MAX_NUM_CHANNELS_SUPPORTED;
 909                 recv_int_page = vmbus_connection.recv_int_page;
 910         } else {
 911                 /*
 912                  * When the host is win8 and beyond, the event page
 913                  * can be directly checked to get the id of the channel
 914                  * that has the interrupt pending.
 915                  */
 916                 void *page_addr = hv_cpu->synic_event_page;
 917                 union hv_synic_event_flags *event
 918                         = (union hv_synic_event_flags *)page_addr +
 919                                                  VMBUS_MESSAGE_SINT;
 920
 921                 maxbits = HV_EVENT_FLAGS_COUNT;
 922                 recv_int_page = event->flags;
 923         }
 924
 925         if (unlikely(!recv_int_page))
 926                 return;
 927
 928         for_each_set_bit(relid, recv_int_page, maxbits) {
 929                 struct vmbus_channel *channel;
 930
 931                 if (!sync_test_and_clear_bit(relid, recv_int_page))
 932                         continue;
 933
 934                 /* Special case - vmbus channel protocol msg */
 935                 if (relid == 0)
 936                         continue;
 937
 938                 rcu_read_lock();
 939
 940                 /* Find channel based on relid */
 941                 list_for_each_entry_rcu(channel, &hv_cpu->chan_list, percpu_list) {
 942                         if (channel->offermsg.child_relid != relid)
 943                                 continue;
 944
 945                         if (channel->rescind)
 946                                 continue;
 947
 948                         trace_vmbus_chan_sched(channel);
 949
 950                         ++channel->interrupts;
 951
 952                         switch (channel->callback_mode) {
 953                         case HV_CALL_ISR:
 954                                 vmbus_channel_isr(channel);
 955                                 break;
 956
 957                         case HV_CALL_BATCHED:
 958                                 hv_begin_read(&channel->inbound);
 959                                 /* fallthrough */
 960                         case HV_CALL_DIRECT:
 961                                 tasklet_schedule(&channel->callback_event);
 962                         }
 963                 }
 964
 965                 rcu_read_unlock();
 966         }
 967 }
 968
 969 static void vmbus_isr(void)
 970 {
 971         struct hv_per_cpu_context *hv_cpu
 972                 = this_cpu_ptr(hv_context.cpu_context);
 973         void *page_addr = hv_cpu->synic_event_page;
 974         struct hv_message *msg;
 975         union hv_synic_event_flags *event;
 976         bool handled = false;
 977
 978         if (unlikely(page_addr == NULL))
 979                 return;
 980
 981         event = (union hv_synic_event_flags *)page_addr +
 982                                          VMBUS_MESSAGE_SINT;
 983         /*
 984          * Check for events before checking for messages. This is the order
 985          * in which events and messages are checked in Windows guests on
 986          * Hyper-V, and the Windows team suggested we do the same.
 987          */
 988
 989         if ((vmbus_proto_version == VERSION_WS2008) ||
 990                 (vmbus_proto_version == VERSION_WIN7)) {
 991
 992                 /* Since we are a child, we only need to check bit 0 */
 993                 if (sync_test_and_clear_bit(0, event->flags))
 994                         handled = true;
 995         } else {
 996                 /*
 997                  * Our host is win8 or above. The signaling mechanism
 998                  * has changed and we can directly look at the event page.
 999                  * If bit n is set then we have an interrup on the channel
1000                  * whose id is n.
1001                  */
1002                 handled = true;
1003         }
1004
1005         if (handled)
1006                 vmbus_chan_sched(hv_cpu);
1007
1008         page_addr = hv_cpu->synic_message_page;
1009         msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
1010
1011         /* Check if there are actual msgs to be processed */
1012         if (msg->header.message_type != HVMSG_NONE) {
1013                 if (msg->header.message_type == HVMSG_TIMER_EXPIRED)
1014                         hv_process_timer_expiration(msg, hv_cpu);
1015                 else
1016                         tasklet_schedule(&hv_cpu->msg_dpc);
1017         }
1018
1019         add_interrupt_randomness(HYPERVISOR_CALLBACK_VECTOR, 0);
1020 }
1021
1022
1023 /*
1024  * vmbus_bus_init -Main vmbus driver initialization routine.
1025  *
1026  * Here, we
1027  *      - initialize the vmbus driver context
1028  *      - invoke the vmbus hv main init routine
1029  *      - retrieve the channel offers
1030  */
1031 static int vmbus_bus_init(void)
1032 {
1033         int ret;
1034
1035         /* Hypervisor initialization...setup hypercall page..etc */
1036         ret = hv_init();
1037         if (ret != 0) {
1038                 pr_err("Unable to initialize the hypervisor - 0x%x\n", ret);
1039                 return ret;
1040         }
1041
1042         ret = bus_register(&hv_bus);
1043         if (ret)
1044                 return ret;
1045
1046         hv_setup_vmbus_irq(vmbus_isr);
1047
1048         ret = hv_synic_alloc();
1049         if (ret)
1050                 goto err_alloc;
1051         /*
1052          * Initialize the per-cpu interrupt state and
1053          * connect to the host.
1054          */
1055         ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "hyperv/vmbus:online",
1056                                 hv_synic_init, hv_synic_cleanup);
1057         if (ret < 0)
1058                 goto err_alloc;
1059         hyperv_cpuhp_online = ret;
1060
1061         ret = vmbus_connect();
1062         if (ret)
1063                 goto err_connect;
1064
1065         /*
1066          * Only register if the crash MSRs are available
1067          */
1068         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1069                 register_die_notifier(&hyperv_die_block);
1070                 atomic_notifier_chain_register(&panic_notifier_list,
1071                                                &hyperv_panic_block);
1072         }
1073
1074         vmbus_request_offers();
1075
1076         return 0;
1077
1078 err_connect:
1079         cpuhp_remove_state(hyperv_cpuhp_online);
1080 err_alloc:
1081         hv_synic_free();
1082         hv_remove_vmbus_irq();
1083
1084         bus_unregister(&hv_bus);
1085
1086         return ret;
1087 }
1088
1089 /**
1090  * __vmbus_child_driver_register() - Register a vmbus's driver
1091  * @hv_driver: Pointer to driver structure you want to register
1092  * @owner: owner module of the drv
1093  * @mod_name: module name string
1094  *
1095  * Registers the given driver with Linux through the 'driver_register()' call
1096  * and sets up the hyper-v vmbus handling for this driver.
1097  * It will return the state of the 'driver_register()' call.
1098  *
1099  */
1100 int __vmbus_driver_register(struct hv_driver *hv_driver, struct module *owner, const char *mod_name)
1101 {
1102         int ret;
1103
1104         pr_info("registering driver %s\n", hv_driver->name);
1105
1106         ret = vmbus_exists();
1107         if (ret < 0)
1108                 return ret;
1109
1110         hv_driver->driver.name = hv_driver->name;
1111         hv_driver->driver.owner = owner;
1112         hv_driver->driver.mod_name = mod_name;
1113         hv_driver->driver.bus = &hv_bus;
1114
1115         spin_lock_init(&hv_driver->dynids.lock);
1116         INIT_LIST_HEAD(&hv_driver->dynids.list);
1117
1118         ret = driver_register(&hv_driver->driver);
1119
1120         return ret;
1121 }
1122 EXPORT_SYMBOL_GPL(__vmbus_driver_register);
1123
1124 /**
1125  * vmbus_driver_unregister() - Unregister a vmbus's driver
1126  * @hv_driver: Pointer to driver structure you want to
1127  *             un-register
1128  *
1129  * Un-register the given driver that was previous registered with a call to
1130  * vmbus_driver_register()
1131  */
1132 void vmbus_driver_unregister(struct hv_driver *hv_driver)
1133 {
1134         pr_info("unregistering driver %s\n", hv_driver->name);
1135
1136         if (!vmbus_exists()) {
1137                 driver_unregister(&hv_driver->driver);
1138                 vmbus_free_dynids(hv_driver);
1139         }
1140 }
1141 EXPORT_SYMBOL_GPL(vmbus_driver_unregister);
1142
1143
1144 /*
1145  * Called when last reference to channel is gone.
1146  */
1147 static void vmbus_chan_release(struct kobject *kobj)
1148 {
1149         struct vmbus_channel *channel
1150                 = container_of(kobj, struct vmbus_channel, kobj);
1151
1152         kfree_rcu(channel, rcu);
1153 }
1154
1155 struct vmbus_chan_attribute {
1156         struct attribute attr;
1157         ssize_t (*show)(const struct vmbus_channel *chan, char *buf);
1158         ssize_t (*store)(struct vmbus_channel *chan,
1159                          const char *buf, size_t count);
1160 };
1161 #define VMBUS_CHAN_ATTR(_name, _mode, _show, _store) \
1162         struct vmbus_chan_attribute chan_attr_##_name \
1163                 = __ATTR(_name, _mode, _show, _store)
1164 #define VMBUS_CHAN_ATTR_RW(_name) \
1165         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RW(_name)
1166 #define VMBUS_CHAN_ATTR_RO(_name) \
1167         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RO(_name)
1168 #define VMBUS_CHAN_ATTR_WO(_name) \
1169         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_WO(_name)
1170
1171 static ssize_t vmbus_chan_attr_show(struct kobject *kobj,
1172                                     struct attribute *attr, char *buf)
1173 {
1174         const struct vmbus_chan_attribute *attribute
1175                 = container_of(attr, struct vmbus_chan_attribute, attr);
1176         const struct vmbus_channel *chan
1177                 = container_of(kobj, struct vmbus_channel, kobj);
1178
1179         if (!attribute->show)
1180                 return -EIO;
1181
1182         return attribute->show(chan, buf);
1183 }
1184
1185 static const struct sysfs_ops vmbus_chan_sysfs_ops = {
1186         .show = vmbus_chan_attr_show,
1187 };
1188
1189 static ssize_t out_mask_show(const struct vmbus_channel *channel, char *buf)
1190 {
1191         const struct hv_ring_buffer_info *rbi = &channel->outbound;
1192
1193         return sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1194 }
1195 static VMBUS_CHAN_ATTR_RO(out_mask);
1196
1197 static ssize_t in_mask_show(const struct vmbus_channel *channel, char *buf)
1198 {
1199         const struct hv_ring_buffer_info *rbi = &channel->inbound;
1200
1201         return sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1202 }
1203 static VMBUS_CHAN_ATTR_RO(in_mask);
1204
1205 static ssize_t read_avail_show(const struct vmbus_channel *channel, char *buf)
1206 {
1207         const struct hv_ring_buffer_info *rbi = &channel->inbound;
1208
1209         return sprintf(buf, "%u\n", hv_get_bytes_to_read(rbi));
1210 }
1211 static VMBUS_CHAN_ATTR_RO(read_avail);
1212
1213 static ssize_t write_avail_show(const struct vmbus_channel *channel, char *buf)
1214 {
1215         const struct hv_ring_buffer_info *rbi = &channel->outbound;
1216
1217         return sprintf(buf, "%u\n", hv_get_bytes_to_write(rbi));
1218 }
1219 static VMBUS_CHAN_ATTR_RO(write_avail);
1220
1221 static ssize_t show_target_cpu(const struct vmbus_channel *channel, char *buf)
1222 {
1223         return sprintf(buf, "%u\n", channel->target_cpu);
1224 }
1225 static VMBUS_CHAN_ATTR(cpu, S_IRUGO, show_target_cpu, NULL);
1226
1227 static ssize_t channel_pending_show(const struct vmbus_channel *channel,
1228                                     char *buf)
1229 {
1230         return sprintf(buf, "%d\n",
1231                        channel_pending(channel,
1232                                        vmbus_connection.monitor_pages[1]));
1233 }
1234 static VMBUS_CHAN_ATTR(pending, S_IRUGO, channel_pending_show, NULL);
1235
1236 static ssize_t channel_latency_show(const struct vmbus_channel *channel,
1237                                     char *buf)
1238 {
1239         return sprintf(buf, "%d\n",
1240                        channel_latency(channel,
1241                                        vmbus_connection.monitor_pages[1]));
1242 }
1243 static VMBUS_CHAN_ATTR(latency, S_IRUGO, channel_latency_show, NULL);
1244
1245 static ssize_t channel_interrupts_show(const struct vmbus_channel *channel, char *buf)
1246 {
1247         return sprintf(buf, "%llu\n", channel->interrupts);
1248 }
1249 static VMBUS_CHAN_ATTR(interrupts, S_IRUGO, channel_interrupts_show, NULL);
1250
1251 static ssize_t channel_events_show(const struct vmbus_channel *channel, char *buf)
1252 {
1253         return sprintf(buf, "%llu\n", channel->sig_events);
1254 }
1255 static VMBUS_CHAN_ATTR(events, S_IRUGO, channel_events_show, NULL);
1256
1257 static ssize_t subchannel_monitor_id_show(const struct vmbus_channel *channel,
1258                                           char *buf)
1259 {
1260         return sprintf(buf, "%u\n", channel->offermsg.monitorid);
1261 }
1262 static VMBUS_CHAN_ATTR(monitor_id, S_IRUGO, subchannel_monitor_id_show, NULL);
1263
1264 static ssize_t subchannel_id_show(const struct vmbus_channel *channel,
1265                                   char *buf)
1266 {
1267         return sprintf(buf, "%u\n",
1268                        channel->offermsg.offer.sub_channel_index);
1269 }
1270 static VMBUS_CHAN_ATTR_RO(subchannel_id);
1271
1272 static struct attribute *vmbus_chan_attrs[] = {
1273         &chan_attr_out_mask.attr,
1274         &chan_attr_in_mask.attr,
1275         &chan_attr_read_avail.attr,
1276         &chan_attr_write_avail.attr,
1277         &chan_attr_cpu.attr,
1278         &chan_attr_pending.attr,
1279         &chan_attr_latency.attr,
1280         &chan_attr_interrupts.attr,
1281         &chan_attr_events.attr,
1282         &chan_attr_monitor_id.attr,
1283         &chan_attr_subchannel_id.attr,
1284         NULL
1285 };
1286
1287 static struct kobj_type vmbus_chan_ktype = {
1288         .sysfs_ops = &vmbus_chan_sysfs_ops,
1289         .release = vmbus_chan_release,
1290         .default_attrs = vmbus_chan_attrs,
1291 };
1292
1293 /*
1294  * vmbus_add_channel_kobj - setup a sub-directory under device/channels
1295  */
1296 int vmbus_add_channel_kobj(struct hv_device *dev, struct vmbus_channel *channel)
1297 {
1298         struct kobject *kobj = &channel->kobj;
1299         u32 relid = channel->offermsg.child_relid;
1300         int ret;
1301
1302         kobj->kset = dev->channels_kset;
1303         ret = kobject_init_and_add(kobj, &vmbus_chan_ktype, NULL,
1304                                    "%u", relid);
1305         if (ret)
1306                 return ret;
1307
1308         kobject_uevent(kobj, KOBJ_ADD);
1309
1310         return 0;
1311 }
1312
1313 /*
1314  * vmbus_device_create - Creates and registers a new child device
1315  * on the vmbus.
1316  */
1317 struct hv_device *vmbus_device_create(const uuid_le *type,
1318                                       const uuid_le *instance,
1319                                       struct vmbus_channel *channel)
1320 {
1321         struct hv_device *child_device_obj;
1322
1323         child_device_obj = kzalloc(sizeof(struct hv_device), GFP_KERNEL);
1324         if (!child_device_obj) {
1325                 pr_err("Unable to allocate device object for child device\n");
1326                 return NULL;
1327         }
1328
1329         child_device_obj->channel = channel;
1330         memcpy(&child_device_obj->dev_type, type, sizeof(uuid_le));
1331         memcpy(&child_device_obj->dev_instance, instance,
1332                sizeof(uuid_le));
1333         child_device_obj->vendor_id = 0x1414; /* MSFT vendor ID */
1334
1335
1336         return child_device_obj;
1337 }
1338
1339 /*
1340  * vmbus_device_register - Register the child device
1341  */
1342 int vmbus_device_register(struct hv_device *child_device_obj)
1343 {
1344         struct kobject *kobj = &child_device_obj->device.kobj;
1345         int ret;
1346
1347         dev_set_name(&child_device_obj->device, "%pUl",
1348                      child_device_obj->channel->offermsg.offer.if_instance.b);
1349
1350         child_device_obj->device.bus = &hv_bus;
1351         child_device_obj->device.parent = &hv_acpi_dev->dev;
1352         child_device_obj->device.release = vmbus_device_release;
1353
1354         /*
1355          * Register with the LDM. This will kick off the driver/device
1356          * binding...which will eventually call vmbus_match() and vmbus_probe()
1357          */
1358         ret = device_register(&child_device_obj->device);
1359         if (ret) {
1360                 pr_err("Unable to register child device\n");
1361                 return ret;
1362         }
1363
1364         child_device_obj->channels_kset = kset_create_and_add("channels",
1365                                                               NULL, kobj);
1366         if (!child_device_obj->channels_kset) {
1367                 ret = -ENOMEM;
1368                 goto err_dev_unregister;
1369         }
1370
1371         ret = vmbus_add_channel_kobj(child_device_obj,
1372                                      child_device_obj->channel);
1373         if (ret) {
1374                 pr_err("Unable to register primary channeln");
1375                 goto err_kset_unregister;
1376         }
1377
1378         return 0;
1379
1380 err_kset_unregister:
1381         kset_unregister(child_device_obj->channels_kset);
1382
1383 err_dev_unregister:
1384         device_unregister(&child_device_obj->device);
1385         return ret;
1386 }
1387
1388 /*
1389  * vmbus_device_unregister - Remove the specified child device
1390  * from the vmbus.
1391  */
1392 void vmbus_device_unregister(struct hv_device *device_obj)
1393 {
1394         pr_debug("child device %s unregistered\n",
1395                 dev_name(&device_obj->device));
1396
1397         kset_unregister(device_obj->channels_kset);
1398
1399         /*
1400          * Kick off the process of unregistering the device.
1401          * This will call vmbus_remove() and eventually vmbus_device_release()
1402          */
1403         device_unregister(&device_obj->device);
1404 }
1405
1406
1407 /*
1408  * VMBUS is an acpi enumerated device. Get the information we
1409  * need from DSDT.
1410  */
1411 #define VTPM_BASE_ADDRESS 0xfed40000
1412 static acpi_status vmbus_walk_resources(struct acpi_resource *res, void *ctx)
1413 {
1414         resource_size_t start = 0;
1415         resource_size_t end = 0;
1416         struct resource *new_res;
1417         struct resource **old_res = &hyperv_mmio;
1418         struct resource **prev_res = NULL;
1419
1420         switch (res->type) {
1421
1422         /*
1423          * "Address" descriptors are for bus windows. Ignore
1424          * "memory" descriptors, which are for registers on
1425          * devices.
1426          */
1427         case ACPI_RESOURCE_TYPE_ADDRESS32:
1428                 start = res->data.address32.address.minimum;
1429                 end = res->data.address32.address.maximum;
1430                 break;
1431
1432         case ACPI_RESOURCE_TYPE_ADDRESS64:
1433                 start = res->data.address64.address.minimum;
1434                 end = res->data.address64.address.maximum;
1435                 break;
1436
1437         default:
1438                 /* Unused resource type */
1439                 return AE_OK;
1440
1441         }
1442         /*
1443          * Ignore ranges that are below 1MB, as they're not
1444          * necessary or useful here.
1445          */
1446         if (end < 0x100000)
1447                 return AE_OK;
1448
1449         new_res = kzalloc(sizeof(*new_res), GFP_ATOMIC);
1450         if (!new_res)
1451                 return AE_NO_MEMORY;
1452
1453         /* If this range overlaps the virtual TPM, truncate it. */
1454         if (end > VTPM_BASE_ADDRESS && start < VTPM_BASE_ADDRESS)
1455                 end = VTPM_BASE_ADDRESS;
1456
1457         new_res->name = "hyperv mmio";
1458         new_res->flags = IORESOURCE_MEM;
1459         new_res->start = start;
1460         new_res->end = end;
1461
1462         /*
1463          * If two ranges are adjacent, merge them.
1464          */
1465         do {
1466                 if (!*old_res) {
1467                         *old_res = new_res;
1468                         break;
1469                 }
1470
1471                 if (((*old_res)->end + 1) == new_res->start) {
1472                         (*old_res)->end = new_res->end;
1473                         kfree(new_res);
1474                         break;
1475                 }
1476
1477                 if ((*old_res)->start == new_res->end + 1) {
1478                         (*old_res)->start = new_res->start;
1479                         kfree(new_res);
1480                         break;
1481                 }
1482
1483                 if ((*old_res)->start > new_res->end) {
1484                         new_res->sibling = *old_res;
1485                         if (prev_res)
1486                                 (*prev_res)->sibling = new_res;
1487                         *old_res = new_res;
1488                         break;
1489                 }
1490
1491                 prev_res = old_res;
1492                 old_res = &(*old_res)->sibling;
1493
1494         } while (1);
1495
1496         return AE_OK;
1497 }
1498
1499 static int vmbus_acpi_remove(struct acpi_device *device)
1500 {
1501         struct resource *cur_res;
1502         struct resource *next_res;
1503
1504         if (hyperv_mmio) {
1505                 if (fb_mmio) {
1506                         __release_region(hyperv_mmio, fb_mmio->start,
1507                                          resource_size(fb_mmio));
1508                         fb_mmio = NULL;
1509                 }
1510
1511                 for (cur_res = hyperv_mmio; cur_res; cur_res = next_res) {
1512                         next_res = cur_res->sibling;
1513                         kfree(cur_res);
1514                 }
1515         }
1516
1517         return 0;
1518 }
1519
1520 static void vmbus_reserve_fb(void)
1521 {
1522         int size;
1523         /*
1524          * Make a claim for the frame buffer in the resource tree under the
1525          * first node, which will be the one below 4GB.  The length seems to
1526          * be underreported, particularly in a Generation 1 VM.  So start out
1527          * reserving a larger area and make it smaller until it succeeds.
1528          */
1529
1530         if (screen_info.lfb_base) {
1531                 if (efi_enabled(EFI_BOOT))
1532                         size = max_t(__u32, screen_info.lfb_size, 0x800000);
1533                 else
1534                         size = max_t(__u32, screen_info.lfb_size, 0x4000000);
1535
1536                 for (; !fb_mmio && (size >= 0x100000); size >>= 1) {
1537                         fb_mmio = __request_region(hyperv_mmio,
1538                                                    screen_info.lfb_base, size,
1539                                                    fb_mmio_name, 0);
1540                 }
1541         }
1542 }
1543
1544 /**
1545  * vmbus_allocate_mmio() - Pick a memory-mapped I/O range.
1546  * @new:                If successful, supplied a pointer to the
1547  *                      allocated MMIO space.
1548  * @device_obj:         Identifies the caller
1549  * @min:                Minimum guest physical address of the
1550  *                      allocation
1551  * @max:                Maximum guest physical address
1552  * @size:               Size of the range to be allocated
1553  * @align:              Alignment of the range to be allocated
1554  * @fb_overlap_ok:      Whether this allocation can be allowed
1555  *                      to overlap the video frame buffer.
1556  *
1557  * This function walks the resources granted to VMBus by the
1558  * _CRS object in the ACPI namespace underneath the parent
1559  * "bridge" whether that's a root PCI bus in the Generation 1
1560  * case or a Module Device in the Generation 2 case.  It then
1561  * attempts to allocate from the global MMIO pool in a way that
1562  * matches the constraints supplied in these parameters and by
1563  * that _CRS.
1564  *
1565  * Return: 0 on success, -errno on failure
1566  */
1567 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
1568                         resource_size_t min, resource_size_t max,
1569                         resource_size_t size, resource_size_t align,
1570                         bool fb_overlap_ok)
1571 {
1572         struct resource *iter, *shadow;
1573         resource_size_t range_min, range_max, start;
1574         const char *dev_n = dev_name(&device_obj->device);
1575         int retval;
1576
1577         retval = -ENXIO;
1578         down(&hyperv_mmio_lock);
1579
1580         /*
1581          * If overlaps with frame buffers are allowed, then first attempt to
1582          * make the allocation from within the reserved region.  Because it
1583          * is already reserved, no shadow allocation is necessary.
1584          */
1585         if (fb_overlap_ok && fb_mmio && !(min > fb_mmio->end) &&
1586             !(max < fb_mmio->start)) {
1587
1588                 range_min = fb_mmio->start;
1589                 range_max = fb_mmio->end;
1590                 start = (range_min + align - 1) & ~(align - 1);
1591                 for (; start + size - 1 <= range_max; start += align) {
1592                         *new = request_mem_region_exclusive(start, size, dev_n);
1593                         if (*new) {
1594                                 retval = 0;
1595                                 goto exit;
1596                         }
1597                 }
1598         }
1599
1600         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
1601                 if ((iter->start >= max) || (iter->end <= min))
1602                         continue;
1603
1604                 range_min = iter->start;
1605                 range_max = iter->end;
1606                 start = (range_min + align - 1) & ~(align - 1);
1607                 for (; start + size - 1 <= range_max; start += align) {
1608                         shadow = __request_region(iter, start, size, NULL,
1609                                                   IORESOURCE_BUSY);
1610                         if (!shadow)
1611                                 continue;
1612
1613                         *new = request_mem_region_exclusive(start, size, dev_n);
1614                         if (*new) {
1615                                 shadow->name = (char *)*new;
1616                                 retval = 0;
1617                                 goto exit;
1618                         }
1619
1620                         __release_region(iter, start, size);
1621                 }
1622         }
1623
1624 exit:
1625         up(&hyperv_mmio_lock);
1626         return retval;
1627 }
1628 EXPORT_SYMBOL_GPL(vmbus_allocate_mmio);
1629
1630 /**
1631  * vmbus_free_mmio() - Free a memory-mapped I/O range.
1632  * @start:              Base address of region to release.
1633  * @size:               Size of the range to be allocated
1634  *
1635  * This function releases anything requested by
1636  * vmbus_mmio_allocate().
1637  */
1638 void vmbus_free_mmio(resource_size_t start, resource_size_t size)
1639 {
1640         struct resource *iter;
1641
1642         down(&hyperv_mmio_lock);
1643         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
1644                 if ((iter->start >= start + size) || (iter->end <= start))
1645                         continue;
1646
1647                 __release_region(iter, start, size);
1648         }
1649         release_mem_region(start, size);
1650         up(&hyperv_mmio_lock);
1651
1652 }
1653 EXPORT_SYMBOL_GPL(vmbus_free_mmio);
1654
1655 static int vmbus_acpi_add(struct acpi_device *device)
1656 {
1657         acpi_status result;
1658         int ret_val = -ENODEV;
1659         struct acpi_device *ancestor;
1660
1661         hv_acpi_dev = device;
1662
1663         result = acpi_walk_resources(device->handle, METHOD_NAME__CRS,
1664                                         vmbus_walk_resources, NULL);
1665
1666         if (ACPI_FAILURE(result))
1667                 goto acpi_walk_err;
1668         /*
1669          * Some ancestor of the vmbus acpi device (Gen1 or Gen2
1670          * firmware) is the VMOD that has the mmio ranges. Get that.
1671          */
1672         for (ancestor = device->parent; ancestor; ancestor = ancestor->parent) {
1673                 result = acpi_walk_resources(ancestor->handle, METHOD_NAME__CRS,
1674                                              vmbus_walk_resources, NULL);
1675
1676                 if (ACPI_FAILURE(result))
1677                         continue;
1678                 if (hyperv_mmio) {
1679                         vmbus_reserve_fb();
1680                         break;
1681                 }
1682         }
1683         ret_val = 0;
1684
1685 acpi_walk_err:
1686         complete(&probe_event);
1687         if (ret_val)
1688                 vmbus_acpi_remove(device);
1689         return ret_val;
1690 }
1691
1692 static const struct acpi_device_id vmbus_acpi_device_ids[] = {
1693         {"VMBUS", 0},
1694         {"VMBus", 0},
1695         {"", 0},
1696 };
1697 MODULE_DEVICE_TABLE(acpi, vmbus_acpi_device_ids);
1698
1699 static struct acpi_driver vmbus_acpi_driver = {
1700         .name = "vmbus",
1701         .ids = vmbus_acpi_device_ids,
1702         .ops = {
1703                 .add = vmbus_acpi_add,
1704                 .remove = vmbus_acpi_remove,
1705         },
1706 };
1707
1708 static void hv_kexec_handler(void)
1709 {
1710         hv_synic_clockevents_cleanup();
1711         vmbus_initiate_unload(false);
1712         vmbus_connection.conn_state = DISCONNECTED;
1713         /* Make sure conn_state is set as hv_synic_cleanup checks for it */
1714         mb();
1715         cpuhp_remove_state(hyperv_cpuhp_online);
1716         hyperv_cleanup();
1717 };
1718
1719 static void hv_crash_handler(struct pt_regs *regs)
1720 {
1721         vmbus_initiate_unload(true);
1722         /*
1723          * In crash handler we can't schedule synic cleanup for all CPUs,
1724          * doing the cleanup for current CPU only. This should be sufficient
1725          * for kdump.
1726          */
1727         vmbus_connection.conn_state = DISCONNECTED;
1728         hv_synic_cleanup(smp_processor_id());
1729         hyperv_cleanup();
1730 };
1731
1732 static int __init hv_acpi_init(void)
1733 {
1734         int ret, t;
1735
1736         if (!hv_is_hyperv_initialized())
1737                 return -ENODEV;
1738
1739         init_completion(&probe_event);
1740
1741         /*
1742          * Get ACPI resources first.
1743          */
1744         ret = acpi_bus_register_driver(&vmbus_acpi_driver);
1745
1746         if (ret)
1747                 return ret;
1748
1749         t = wait_for_completion_timeout(&probe_event, 5*HZ);
1750         if (t == 0) {
1751                 ret = -ETIMEDOUT;
1752                 goto cleanup;
1753         }
1754
1755         ret = vmbus_bus_init();
1756         if (ret)
1757                 goto cleanup;
1758
1759         hv_setup_kexec_handler(hv_kexec_handler);
1760         hv_setup_crash_handler(hv_crash_handler);
1761
1762         return 0;
1763
1764 cleanup:
1765         acpi_bus_unregister_driver(&vmbus_acpi_driver);
1766         hv_acpi_dev = NULL;
1767         return ret;
1768 }
1769
1770 static void __exit vmbus_exit(void)
1771 {
1772         int cpu;
1773
1774         hv_remove_kexec_handler();
1775         hv_remove_crash_handler();
1776         vmbus_connection.conn_state = DISCONNECTED;
1777         hv_synic_clockevents_cleanup();
1778         vmbus_disconnect();
1779         hv_remove_vmbus_irq();
1780         for_each_online_cpu(cpu) {
1781                 struct hv_per_cpu_context *hv_cpu
1782                         = per_cpu_ptr(hv_context.cpu_context, cpu);
1783
1784                 tasklet_kill(&hv_cpu->msg_dpc);
1785         }
1786         vmbus_free_channels();
1787
1788         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1789                 unregister_die_notifier(&hyperv_die_block);
1790                 atomic_notifier_chain_unregister(&panic_notifier_list,
1791                                                  &hyperv_panic_block);
1792         }
1793         bus_unregister(&hv_bus);
1794
1795         cpuhp_remove_state(hyperv_cpuhp_online);
1796         hv_synic_free();
1797         acpi_bus_unregister_driver(&vmbus_acpi_driver);
1798 }
1799
1800
1801 MODULE_LICENSE("GPL");
1802
1803 subsys_initcall(hv_acpi_init);
1804 module_exit(vmbus_exit);