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 void *hv_panic_page;
  60
  61 static int hyperv_panic_event(struct notifier_block *nb, unsigned long val,
  62                               void *args)
  63 {
  64         struct pt_regs *regs;
  65
  66         regs = current_pt_regs();
  67
  68         hyperv_report_panic(regs, val);
  69         return NOTIFY_DONE;
  70 }
  71
  72 static int hyperv_die_event(struct notifier_block *nb, unsigned long val,
  73                             void *args)
  74 {
  75         struct die_args *die = (struct die_args *)args;
  76         struct pt_regs *regs = die->regs;
  77
  78         hyperv_report_panic(regs, val);
  79         return NOTIFY_DONE;
  80 }
  81
  82 static struct notifier_block hyperv_die_block = {
  83         .notifier_call = hyperv_die_event,
  84 };
  85 static struct notifier_block hyperv_panic_block = {
  86         .notifier_call = hyperv_panic_event,
  87 };
  88
  89 static const char *fb_mmio_name = "fb_range";
  90 static struct resource *fb_mmio;
  91 static struct resource *hyperv_mmio;
  92 static DEFINE_SEMAPHORE(hyperv_mmio_lock);
  93
  94 static int vmbus_exists(void)
  95 {
  96         if (hv_acpi_dev == NULL)
  97                 return -ENODEV;
  98
  99         return 0;
 100 }
 101
 102 #define VMBUS_ALIAS_LEN ((sizeof((struct hv_vmbus_device_id *)0)->guid) * 2)
 103 static void print_alias_name(struct hv_device *hv_dev, char *alias_name)
 104 {
 105         int i;
 106         for (i = 0; i < VMBUS_ALIAS_LEN; i += 2)
 107                 sprintf(&alias_name[i], "%02x", hv_dev->dev_type.b[i/2]);
 108 }
 109
 110 static u8 channel_monitor_group(const struct vmbus_channel *channel)
 111 {
 112         return (u8)channel->offermsg.monitorid / 32;
 113 }
 114
 115 static u8 channel_monitor_offset(const struct vmbus_channel *channel)
 116 {
 117         return (u8)channel->offermsg.monitorid % 32;
 118 }
 119
 120 static u32 channel_pending(const struct vmbus_channel *channel,
 121                            const struct hv_monitor_page *monitor_page)
 122 {
 123         u8 monitor_group = channel_monitor_group(channel);
 124
 125         return monitor_page->trigger_group[monitor_group].pending;
 126 }
 127
 128 static u32 channel_latency(const struct vmbus_channel *channel,
 129                            const struct hv_monitor_page *monitor_page)
 130 {
 131         u8 monitor_group = channel_monitor_group(channel);
 132         u8 monitor_offset = channel_monitor_offset(channel);
 133
 134         return monitor_page->latency[monitor_group][monitor_offset];
 135 }
 136
 137 static u32 channel_conn_id(struct vmbus_channel *channel,
 138                            struct hv_monitor_page *monitor_page)
 139 {
 140         u8 monitor_group = channel_monitor_group(channel);
 141         u8 monitor_offset = channel_monitor_offset(channel);
 142         return monitor_page->parameter[monitor_group][monitor_offset].connectionid.u.id;
 143 }
 144
 145 static ssize_t id_show(struct device *dev, struct device_attribute *dev_attr,
 146                        char *buf)
 147 {
 148         struct hv_device *hv_dev = device_to_hv_device(dev);
 149
 150         if (!hv_dev->channel)
 151                 return -ENODEV;
 152         return sprintf(buf, "%d\n", hv_dev->channel->offermsg.child_relid);
 153 }
 154 static DEVICE_ATTR_RO(id);
 155
 156 static ssize_t state_show(struct device *dev, struct device_attribute *dev_attr,
 157                           char *buf)
 158 {
 159         struct hv_device *hv_dev = device_to_hv_device(dev);
 160
 161         if (!hv_dev->channel)
 162                 return -ENODEV;
 163         return sprintf(buf, "%d\n", hv_dev->channel->state);
 164 }
 165 static DEVICE_ATTR_RO(state);
 166
 167 static ssize_t monitor_id_show(struct device *dev,
 168                                struct device_attribute *dev_attr, char *buf)
 169 {
 170         struct hv_device *hv_dev = device_to_hv_device(dev);
 171
 172         if (!hv_dev->channel)
 173                 return -ENODEV;
 174         return sprintf(buf, "%d\n", hv_dev->channel->offermsg.monitorid);
 175 }
 176 static DEVICE_ATTR_RO(monitor_id);
 177
 178 static ssize_t class_id_show(struct device *dev,
 179                                struct device_attribute *dev_attr, char *buf)
 180 {
 181         struct hv_device *hv_dev = device_to_hv_device(dev);
 182
 183         if (!hv_dev->channel)
 184                 return -ENODEV;
 185         return sprintf(buf, "{%pUl}\n",
 186                        hv_dev->channel->offermsg.offer.if_type.b);
 187 }
 188 static DEVICE_ATTR_RO(class_id);
 189
 190 static ssize_t device_id_show(struct device *dev,
 191                               struct device_attribute *dev_attr, char *buf)
 192 {
 193         struct hv_device *hv_dev = device_to_hv_device(dev);
 194
 195         if (!hv_dev->channel)
 196                 return -ENODEV;
 197         return sprintf(buf, "{%pUl}\n",
 198                        hv_dev->channel->offermsg.offer.if_instance.b);
 199 }
 200 static DEVICE_ATTR_RO(device_id);
 201
 202 static ssize_t modalias_show(struct device *dev,
 203                              struct device_attribute *dev_attr, char *buf)
 204 {
 205         struct hv_device *hv_dev = device_to_hv_device(dev);
 206         char alias_name[VMBUS_ALIAS_LEN + 1];
 207
 208         print_alias_name(hv_dev, alias_name);
 209         return sprintf(buf, "vmbus:%s\n", alias_name);
 210 }
 211 static DEVICE_ATTR_RO(modalias);
 212
 213 #ifdef CONFIG_NUMA
 214 static ssize_t numa_node_show(struct device *dev,
 215                               struct device_attribute *attr, char *buf)
 216 {
 217         struct hv_device *hv_dev = device_to_hv_device(dev);
 218
 219         if (!hv_dev->channel)
 220                 return -ENODEV;
 221
 222         return sprintf(buf, "%d\n", hv_dev->channel->numa_node);
 223 }
 224 static DEVICE_ATTR_RO(numa_node);
 225 #endif
 226
 227 static ssize_t server_monitor_pending_show(struct device *dev,
 228                                            struct device_attribute *dev_attr,
 229                                            char *buf)
 230 {
 231         struct hv_device *hv_dev = device_to_hv_device(dev);
 232
 233         if (!hv_dev->channel)
 234                 return -ENODEV;
 235         return sprintf(buf, "%d\n",
 236                        channel_pending(hv_dev->channel,
 237                                        vmbus_connection.monitor_pages[1]));
 238 }
 239 static DEVICE_ATTR_RO(server_monitor_pending);
 240
 241 static ssize_t client_monitor_pending_show(struct device *dev,
 242                                            struct device_attribute *dev_attr,
 243                                            char *buf)
 244 {
 245         struct hv_device *hv_dev = device_to_hv_device(dev);
 246
 247         if (!hv_dev->channel)
 248                 return -ENODEV;
 249         return sprintf(buf, "%d\n",
 250                        channel_pending(hv_dev->channel,
 251                                        vmbus_connection.monitor_pages[1]));
 252 }
 253 static DEVICE_ATTR_RO(client_monitor_pending);
 254
 255 static ssize_t server_monitor_latency_show(struct device *dev,
 256                                            struct device_attribute *dev_attr,
 257                                            char *buf)
 258 {
 259         struct hv_device *hv_dev = device_to_hv_device(dev);
 260
 261         if (!hv_dev->channel)
 262                 return -ENODEV;
 263         return sprintf(buf, "%d\n",
 264                        channel_latency(hv_dev->channel,
 265                                        vmbus_connection.monitor_pages[0]));
 266 }
 267 static DEVICE_ATTR_RO(server_monitor_latency);
 268
 269 static ssize_t client_monitor_latency_show(struct device *dev,
 270                                            struct device_attribute *dev_attr,
 271                                            char *buf)
 272 {
 273         struct hv_device *hv_dev = device_to_hv_device(dev);
 274
 275         if (!hv_dev->channel)
 276                 return -ENODEV;
 277         return sprintf(buf, "%d\n",
 278                        channel_latency(hv_dev->channel,
 279                                        vmbus_connection.monitor_pages[1]));
 280 }
 281 static DEVICE_ATTR_RO(client_monitor_latency);
 282
 283 static ssize_t server_monitor_conn_id_show(struct device *dev,
 284                                            struct device_attribute *dev_attr,
 285                                            char *buf)
 286 {
 287         struct hv_device *hv_dev = device_to_hv_device(dev);
 288
 289         if (!hv_dev->channel)
 290                 return -ENODEV;
 291         return sprintf(buf, "%d\n",
 292                        channel_conn_id(hv_dev->channel,
 293                                        vmbus_connection.monitor_pages[0]));
 294 }
 295 static DEVICE_ATTR_RO(server_monitor_conn_id);
 296
 297 static ssize_t client_monitor_conn_id_show(struct device *dev,
 298                                            struct device_attribute *dev_attr,
 299                                            char *buf)
 300 {
 301         struct hv_device *hv_dev = device_to_hv_device(dev);
 302
 303         if (!hv_dev->channel)
 304                 return -ENODEV;
 305         return sprintf(buf, "%d\n",
 306                        channel_conn_id(hv_dev->channel,
 307                                        vmbus_connection.monitor_pages[1]));
 308 }
 309 static DEVICE_ATTR_RO(client_monitor_conn_id);
 310
 311 static ssize_t out_intr_mask_show(struct device *dev,
 312                                   struct device_attribute *dev_attr, char *buf)
 313 {
 314         struct hv_device *hv_dev = device_to_hv_device(dev);
 315         struct hv_ring_buffer_debug_info outbound;
 316
 317         if (!hv_dev->channel)
 318                 return -ENODEV;
 319         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
 320         return sprintf(buf, "%d\n", outbound.current_interrupt_mask);
 321 }
 322 static DEVICE_ATTR_RO(out_intr_mask);
 323
 324 static ssize_t out_read_index_show(struct device *dev,
 325                                    struct device_attribute *dev_attr, char *buf)
 326 {
 327         struct hv_device *hv_dev = device_to_hv_device(dev);
 328         struct hv_ring_buffer_debug_info outbound;
 329
 330         if (!hv_dev->channel)
 331                 return -ENODEV;
 332         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
 333         return sprintf(buf, "%d\n", outbound.current_read_index);
 334 }
 335 static DEVICE_ATTR_RO(out_read_index);
 336
 337 static ssize_t out_write_index_show(struct device *dev,
 338                                     struct device_attribute *dev_attr,
 339                                     char *buf)
 340 {
 341         struct hv_device *hv_dev = device_to_hv_device(dev);
 342         struct hv_ring_buffer_debug_info outbound;
 343
 344         if (!hv_dev->channel)
 345                 return -ENODEV;
 346         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
 347         return sprintf(buf, "%d\n", outbound.current_write_index);
 348 }
 349 static DEVICE_ATTR_RO(out_write_index);
 350
 351 static ssize_t out_read_bytes_avail_show(struct device *dev,
 352                                          struct device_attribute *dev_attr,
 353                                          char *buf)
 354 {
 355         struct hv_device *hv_dev = device_to_hv_device(dev);
 356         struct hv_ring_buffer_debug_info outbound;
 357
 358         if (!hv_dev->channel)
 359                 return -ENODEV;
 360         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
 361         return sprintf(buf, "%d\n", outbound.bytes_avail_toread);
 362 }
 363 static DEVICE_ATTR_RO(out_read_bytes_avail);
 364
 365 static ssize_t out_write_bytes_avail_show(struct device *dev,
 366                                           struct device_attribute *dev_attr,
 367                                           char *buf)
 368 {
 369         struct hv_device *hv_dev = device_to_hv_device(dev);
 370         struct hv_ring_buffer_debug_info outbound;
 371
 372         if (!hv_dev->channel)
 373                 return -ENODEV;
 374         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
 375         return sprintf(buf, "%d\n", outbound.bytes_avail_towrite);
 376 }
 377 static DEVICE_ATTR_RO(out_write_bytes_avail);
 378
 379 static ssize_t in_intr_mask_show(struct device *dev,
 380                                  struct device_attribute *dev_attr, char *buf)
 381 {
 382         struct hv_device *hv_dev = device_to_hv_device(dev);
 383         struct hv_ring_buffer_debug_info inbound;
 384
 385         if (!hv_dev->channel)
 386                 return -ENODEV;
 387         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 388         return sprintf(buf, "%d\n", inbound.current_interrupt_mask);
 389 }
 390 static DEVICE_ATTR_RO(in_intr_mask);
 391
 392 static ssize_t in_read_index_show(struct device *dev,
 393                                   struct device_attribute *dev_attr, char *buf)
 394 {
 395         struct hv_device *hv_dev = device_to_hv_device(dev);
 396         struct hv_ring_buffer_debug_info inbound;
 397
 398         if (!hv_dev->channel)
 399                 return -ENODEV;
 400         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 401         return sprintf(buf, "%d\n", inbound.current_read_index);
 402 }
 403 static DEVICE_ATTR_RO(in_read_index);
 404
 405 static ssize_t in_write_index_show(struct device *dev,
 406                                    struct device_attribute *dev_attr, char *buf)
 407 {
 408         struct hv_device *hv_dev = device_to_hv_device(dev);
 409         struct hv_ring_buffer_debug_info inbound;
 410
 411         if (!hv_dev->channel)
 412                 return -ENODEV;
 413         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 414         return sprintf(buf, "%d\n", inbound.current_write_index);
 415 }
 416 static DEVICE_ATTR_RO(in_write_index);
 417
 418 static ssize_t in_read_bytes_avail_show(struct device *dev,
 419                                         struct device_attribute *dev_attr,
 420                                         char *buf)
 421 {
 422         struct hv_device *hv_dev = device_to_hv_device(dev);
 423         struct hv_ring_buffer_debug_info inbound;
 424
 425         if (!hv_dev->channel)
 426                 return -ENODEV;
 427         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 428         return sprintf(buf, "%d\n", inbound.bytes_avail_toread);
 429 }
 430 static DEVICE_ATTR_RO(in_read_bytes_avail);
 431
 432 static ssize_t in_write_bytes_avail_show(struct device *dev,
 433                                          struct device_attribute *dev_attr,
 434                                          char *buf)
 435 {
 436         struct hv_device *hv_dev = device_to_hv_device(dev);
 437         struct hv_ring_buffer_debug_info inbound;
 438
 439         if (!hv_dev->channel)
 440                 return -ENODEV;
 441         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 442         return sprintf(buf, "%d\n", inbound.bytes_avail_towrite);
 443 }
 444 static DEVICE_ATTR_RO(in_write_bytes_avail);
 445
 446 static ssize_t channel_vp_mapping_show(struct device *dev,
 447                                        struct device_attribute *dev_attr,
 448                                        char *buf)
 449 {
 450         struct hv_device *hv_dev = device_to_hv_device(dev);
 451         struct vmbus_channel *channel = hv_dev->channel, *cur_sc;
 452         unsigned long flags;
 453         int buf_size = PAGE_SIZE, n_written, tot_written;
 454         struct list_head *cur;
 455
 456         if (!channel)
 457                 return -ENODEV;
 458
 459         tot_written = snprintf(buf, buf_size, "%u:%u\n",
 460                 channel->offermsg.child_relid, channel->target_cpu);
 461
 462         spin_lock_irqsave(&channel->lock, flags);
 463
 464         list_for_each(cur, &channel->sc_list) {
 465                 if (tot_written >= buf_size - 1)
 466                         break;
 467
 468                 cur_sc = list_entry(cur, struct vmbus_channel, sc_list);
 469                 n_written = scnprintf(buf + tot_written,
 470                                      buf_size - tot_written,
 471                                      "%u:%u\n",
 472                                      cur_sc->offermsg.child_relid,
 473                                      cur_sc->target_cpu);
 474                 tot_written += n_written;
 475         }
 476
 477         spin_unlock_irqrestore(&channel->lock, flags);
 478
 479         return tot_written;
 480 }
 481 static DEVICE_ATTR_RO(channel_vp_mapping);
 482
 483 static ssize_t vendor_show(struct device *dev,
 484                            struct device_attribute *dev_attr,
 485                            char *buf)
 486 {
 487         struct hv_device *hv_dev = device_to_hv_device(dev);
 488         return sprintf(buf, "0x%x\n", hv_dev->vendor_id);
 489 }
 490 static DEVICE_ATTR_RO(vendor);
 491
 492 static ssize_t device_show(struct device *dev,
 493                            struct device_attribute *dev_attr,
 494                            char *buf)
 495 {
 496         struct hv_device *hv_dev = device_to_hv_device(dev);
 497         return sprintf(buf, "0x%x\n", hv_dev->device_id);
 498 }
 499 static DEVICE_ATTR_RO(device);
 500
 501 /* Set up per device attributes in /sys/bus/vmbus/devices/<bus device> */
 502 static struct attribute *vmbus_dev_attrs[] = {
 503         &dev_attr_id.attr,
 504         &dev_attr_state.attr,
 505         &dev_attr_monitor_id.attr,
 506         &dev_attr_class_id.attr,
 507         &dev_attr_device_id.attr,
 508         &dev_attr_modalias.attr,
 509 #ifdef CONFIG_NUMA
 510         &dev_attr_numa_node.attr,
 511 #endif
 512         &dev_attr_server_monitor_pending.attr,
 513         &dev_attr_client_monitor_pending.attr,
 514         &dev_attr_server_monitor_latency.attr,
 515         &dev_attr_client_monitor_latency.attr,
 516         &dev_attr_server_monitor_conn_id.attr,
 517         &dev_attr_client_monitor_conn_id.attr,
 518         &dev_attr_out_intr_mask.attr,
 519         &dev_attr_out_read_index.attr,
 520         &dev_attr_out_write_index.attr,
 521         &dev_attr_out_read_bytes_avail.attr,
 522         &dev_attr_out_write_bytes_avail.attr,
 523         &dev_attr_in_intr_mask.attr,
 524         &dev_attr_in_read_index.attr,
 525         &dev_attr_in_write_index.attr,
 526         &dev_attr_in_read_bytes_avail.attr,
 527         &dev_attr_in_write_bytes_avail.attr,
 528         &dev_attr_channel_vp_mapping.attr,
 529         &dev_attr_vendor.attr,
 530         &dev_attr_device.attr,
 531         NULL,
 532 };
 533 ATTRIBUTE_GROUPS(vmbus_dev);
 534
 535 /*
 536  * vmbus_uevent - add uevent for our device
 537  *
 538  * This routine is invoked when a device is added or removed on the vmbus to
 539  * generate a uevent to udev in the userspace. The udev will then look at its
 540  * rule and the uevent generated here to load the appropriate driver
 541  *
 542  * The alias string will be of the form vmbus:guid where guid is the string
 543  * representation of the device guid (each byte of the guid will be
 544  * represented with two hex characters.
 545  */
 546 static int vmbus_uevent(struct device *device, struct kobj_uevent_env *env)
 547 {
 548         struct hv_device *dev = device_to_hv_device(device);
 549         int ret;
 550         char alias_name[VMBUS_ALIAS_LEN + 1];
 551
 552         print_alias_name(dev, alias_name);
 553         ret = add_uevent_var(env, "MODALIAS=vmbus:%s", alias_name);
 554         return ret;
 555 }
 556
 557 static const uuid_le null_guid;
 558
 559 static inline bool is_null_guid(const uuid_le *guid)
 560 {
 561         if (uuid_le_cmp(*guid, null_guid))
 562                 return false;
 563         return true;
 564 }
 565
 566 /*
 567  * Return a matching hv_vmbus_device_id pointer.
 568  * If there is no match, return NULL.
 569  */
 570 static const struct hv_vmbus_device_id *hv_vmbus_get_id(struct hv_driver *drv,
 571                                         const uuid_le *guid)
 572 {
 573         const struct hv_vmbus_device_id *id = NULL;
 574         struct vmbus_dynid *dynid;
 575
 576         /* Look at the dynamic ids first, before the static ones */
 577         spin_lock(&drv->dynids.lock);
 578         list_for_each_entry(dynid, &drv->dynids.list, node) {
 579                 if (!uuid_le_cmp(dynid->id.guid, *guid)) {
 580                         id = &dynid->id;
 581                         break;
 582                 }
 583         }
 584         spin_unlock(&drv->dynids.lock);
 585
 586         if (id)
 587                 return id;
 588
 589         id = drv->id_table;
 590         if (id == NULL)
 591                 return NULL; /* empty device table */
 592
 593         for (; !is_null_guid(&id->guid); id++)
 594                 if (!uuid_le_cmp(id->guid, *guid))
 595                         return id;
 596
 597         return NULL;
 598 }
 599
 600 /* vmbus_add_dynid - add a new device ID to this driver and re-probe devices */
 601 static int vmbus_add_dynid(struct hv_driver *drv, uuid_le *guid)
 602 {
 603         struct vmbus_dynid *dynid;
 604
 605         dynid = kzalloc(sizeof(*dynid), GFP_KERNEL);
 606         if (!dynid)
 607                 return -ENOMEM;
 608
 609         dynid->id.guid = *guid;
 610
 611         spin_lock(&drv->dynids.lock);
 612         list_add_tail(&dynid->node, &drv->dynids.list);
 613         spin_unlock(&drv->dynids.lock);
 614
 615         return driver_attach(&drv->driver);
 616 }
 617
 618 static void vmbus_free_dynids(struct hv_driver *drv)
 619 {
 620         struct vmbus_dynid *dynid, *n;
 621
 622         spin_lock(&drv->dynids.lock);
 623         list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
 624                 list_del(&dynid->node);
 625                 kfree(dynid);
 626         }
 627         spin_unlock(&drv->dynids.lock);
 628 }
 629
 630 /*
 631  * store_new_id - sysfs frontend to vmbus_add_dynid()
 632  *
 633  * Allow GUIDs to be added to an existing driver via sysfs.
 634  */
 635 static ssize_t new_id_store(struct device_driver *driver, const char *buf,
 636                             size_t count)
 637 {
 638         struct hv_driver *drv = drv_to_hv_drv(driver);
 639         uuid_le guid;
 640         ssize_t retval;
 641
 642         retval = uuid_le_to_bin(buf, &guid);
 643         if (retval)
 644                 return retval;
 645
 646         if (hv_vmbus_get_id(drv, &guid))
 647                 return -EEXIST;
 648
 649         retval = vmbus_add_dynid(drv, &guid);
 650         if (retval)
 651                 return retval;
 652         return count;
 653 }
 654 static DRIVER_ATTR_WO(new_id);
 655
 656 /*
 657  * store_remove_id - remove a PCI device ID from this driver
 658  *
 659  * Removes a dynamic pci device ID to this driver.
 660  */
 661 static ssize_t remove_id_store(struct device_driver *driver, const char *buf,
 662                                size_t count)
 663 {
 664         struct hv_driver *drv = drv_to_hv_drv(driver);
 665         struct vmbus_dynid *dynid, *n;
 666         uuid_le guid;
 667         ssize_t retval;
 668
 669         retval = uuid_le_to_bin(buf, &guid);
 670         if (retval)
 671                 return retval;
 672
 673         retval = -ENODEV;
 674         spin_lock(&drv->dynids.lock);
 675         list_for_each_entry_safe(dynid, n, &drv->dynids.list, node) {
 676                 struct hv_vmbus_device_id *id = &dynid->id;
 677
 678                 if (!uuid_le_cmp(id->guid, guid)) {
 679                         list_del(&dynid->node);
 680                         kfree(dynid);
 681                         retval = count;
 682                         break;
 683                 }
 684         }
 685         spin_unlock(&drv->dynids.lock);
 686
 687         return retval;
 688 }
 689 static DRIVER_ATTR_WO(remove_id);
 690
 691 static struct attribute *vmbus_drv_attrs[] = {
 692         &driver_attr_new_id.attr,
 693         &driver_attr_remove_id.attr,
 694         NULL,
 695 };
 696 ATTRIBUTE_GROUPS(vmbus_drv);
 697
 698
 699 /*
 700  * vmbus_match - Attempt to match the specified device to the specified driver
 701  */
 702 static int vmbus_match(struct device *device, struct device_driver *driver)
 703 {
 704         struct hv_driver *drv = drv_to_hv_drv(driver);
 705         struct hv_device *hv_dev = device_to_hv_device(device);
 706
 707         /* The hv_sock driver handles all hv_sock offers. */
 708         if (is_hvsock_channel(hv_dev->channel))
 709                 return drv->hvsock;
 710
 711         if (hv_vmbus_get_id(drv, &hv_dev->dev_type))
 712                 return 1;
 713
 714         return 0;
 715 }
 716
 717 /*
 718  * vmbus_probe - Add the new vmbus's child device
 719  */
 720 static int vmbus_probe(struct device *child_device)
 721 {
 722         int ret = 0;
 723         struct hv_driver *drv =
 724                         drv_to_hv_drv(child_device->driver);
 725         struct hv_device *dev = device_to_hv_device(child_device);
 726         const struct hv_vmbus_device_id *dev_id;
 727
 728         dev_id = hv_vmbus_get_id(drv, &dev->dev_type);
 729         if (drv->probe) {
 730                 ret = drv->probe(dev, dev_id);
 731                 if (ret != 0)
 732                         pr_err("probe failed for device %s (%d)\n",
 733                                dev_name(child_device), ret);
 734
 735         } else {
 736                 pr_err("probe not set for driver %s\n",
 737                        dev_name(child_device));
 738                 ret = -ENODEV;
 739         }
 740         return ret;
 741 }
 742
 743 /*
 744  * vmbus_remove - Remove a vmbus device
 745  */
 746 static int vmbus_remove(struct device *child_device)
 747 {
 748         struct hv_driver *drv;
 749         struct hv_device *dev = device_to_hv_device(child_device);
 750
 751         if (child_device->driver) {
 752                 drv = drv_to_hv_drv(child_device->driver);
 753                 if (drv->remove)
 754                         drv->remove(dev);
 755         }
 756
 757         return 0;
 758 }
 759
 760
 761 /*
 762  * vmbus_shutdown - Shutdown a vmbus device
 763  */
 764 static void vmbus_shutdown(struct device *child_device)
 765 {
 766         struct hv_driver *drv;
 767         struct hv_device *dev = device_to_hv_device(child_device);
 768
 769
 770         /* The device may not be attached yet */
 771         if (!child_device->driver)
 772                 return;
 773
 774         drv = drv_to_hv_drv(child_device->driver);
 775
 776         if (drv->shutdown)
 777                 drv->shutdown(dev);
 778 }
 779
 780
 781 /*
 782  * vmbus_device_release - Final callback release of the vmbus child device
 783  */
 784 static void vmbus_device_release(struct device *device)
 785 {
 786         struct hv_device *hv_dev = device_to_hv_device(device);
 787         struct vmbus_channel *channel = hv_dev->channel;
 788
 789         mutex_lock(&vmbus_connection.channel_mutex);
 790         hv_process_channel_removal(channel->offermsg.child_relid);
 791         mutex_unlock(&vmbus_connection.channel_mutex);
 792         kfree(hv_dev);
 793
 794 }
 795
 796 /* The one and only one */
 797 static struct bus_type  hv_bus = {
 798         .name =         "vmbus",
 799         .match =                vmbus_match,
 800         .shutdown =             vmbus_shutdown,
 801         .remove =               vmbus_remove,
 802         .probe =                vmbus_probe,
 803         .uevent =               vmbus_uevent,
 804         .dev_groups =           vmbus_dev_groups,
 805         .drv_groups =           vmbus_drv_groups,
 806 };
 807
 808 struct onmessage_work_context {
 809         struct work_struct work;
 810         struct hv_message msg;
 811 };
 812
 813 static void vmbus_onmessage_work(struct work_struct *work)
 814 {
 815         struct onmessage_work_context *ctx;
 816
 817         /* Do not process messages if we're in DISCONNECTED state */
 818         if (vmbus_connection.conn_state == DISCONNECTED)
 819                 return;
 820
 821         ctx = container_of(work, struct onmessage_work_context,
 822                            work);
 823         vmbus_onmessage(&ctx->msg);
 824         kfree(ctx);
 825 }
 826
 827 static void hv_process_timer_expiration(struct hv_message *msg,
 828                                         struct hv_per_cpu_context *hv_cpu)
 829 {
 830         struct clock_event_device *dev = hv_cpu->clk_evt;
 831
 832         if (dev->event_handler)
 833                 dev->event_handler(dev);
 834
 835         vmbus_signal_eom(msg, HVMSG_TIMER_EXPIRED);
 836 }
 837
 838 void vmbus_on_msg_dpc(unsigned long data)
 839 {
 840         struct hv_per_cpu_context *hv_cpu = (void *)data;
 841         void *page_addr = hv_cpu->synic_message_page;
 842         struct hv_message *msg = (struct hv_message *)page_addr +
 843                                   VMBUS_MESSAGE_SINT;
 844         struct vmbus_channel_message_header *hdr;
 845         const struct vmbus_channel_message_table_entry *entry;
 846         struct onmessage_work_context *ctx;
 847         u32 message_type = msg->header.message_type;
 848
 849         if (message_type == HVMSG_NONE)
 850                 /* no msg */
 851                 return;
 852
 853         hdr = (struct vmbus_channel_message_header *)msg->u.payload;
 854
 855         trace_vmbus_on_msg_dpc(hdr);
 856
 857         if (hdr->msgtype >= CHANNELMSG_COUNT) {
 858                 WARN_ONCE(1, "unknown msgtype=%d\n", hdr->msgtype);
 859                 goto msg_handled;
 860         }
 861
 862         entry = &channel_message_table[hdr->msgtype];
 863         if (entry->handler_type == VMHT_BLOCKING) {
 864                 ctx = kmalloc(sizeof(*ctx), GFP_ATOMIC);
 865                 if (ctx == NULL)
 866                         return;
 867
 868                 INIT_WORK(&ctx->work, vmbus_onmessage_work);
 869                 memcpy(&ctx->msg, msg, sizeof(*msg));
 870
 871                 /*
 872                  * The host can generate a rescind message while we
 873                  * may still be handling the original offer. We deal with
 874                  * this condition by ensuring the processing is done on the
 875                  * same CPU.
 876                  */
 877                 switch (hdr->msgtype) {
 878                 case CHANNELMSG_RESCIND_CHANNELOFFER:
 879                         /*
 880                          * If we are handling the rescind message;
 881                          * schedule the work on the global work queue.
 882                          */
 883                         schedule_work_on(vmbus_connection.connect_cpu,
 884                                          &ctx->work);
 885                         break;
 886
 887                 case CHANNELMSG_OFFERCHANNEL:
 888                         atomic_inc(&vmbus_connection.offer_in_progress);
 889                         queue_work_on(vmbus_connection.connect_cpu,
 890                                       vmbus_connection.work_queue,
 891                                       &ctx->work);
 892                         break;
 893
 894                 default:
 895                         queue_work(vmbus_connection.work_queue, &ctx->work);
 896                 }
 897         } else
 898                 entry->message_handler(hdr);
 899
 900 msg_handled:
 901         vmbus_signal_eom(msg, message_type);
 902 }
 903
 904
 905 /*
 906  * Direct callback for channels using other deferred processing
 907  */
 908 static void vmbus_channel_isr(struct vmbus_channel *channel)
 909 {
 910         void (*callback_fn)(void *);
 911
 912         callback_fn = READ_ONCE(channel->onchannel_callback);
 913         if (likely(callback_fn != NULL))
 914                 (*callback_fn)(channel->channel_callback_context);
 915 }
 916
 917 /*
 918  * Schedule all channels with events pending
 919  */
 920 static void vmbus_chan_sched(struct hv_per_cpu_context *hv_cpu)
 921 {
 922         unsigned long *recv_int_page;
 923         u32 maxbits, relid;
 924
 925         if (vmbus_proto_version < VERSION_WIN8) {
 926                 maxbits = MAX_NUM_CHANNELS_SUPPORTED;
 927                 recv_int_page = vmbus_connection.recv_int_page;
 928         } else {
 929                 /*
 930                  * When the host is win8 and beyond, the event page
 931                  * can be directly checked to get the id of the channel
 932                  * that has the interrupt pending.
 933                  */
 934                 void *page_addr = hv_cpu->synic_event_page;
 935                 union hv_synic_event_flags *event
 936                         = (union hv_synic_event_flags *)page_addr +
 937                                                  VMBUS_MESSAGE_SINT;
 938
 939                 maxbits = HV_EVENT_FLAGS_COUNT;
 940                 recv_int_page = event->flags;
 941         }
 942
 943         if (unlikely(!recv_int_page))
 944                 return;
 945
 946         for_each_set_bit(relid, recv_int_page, maxbits) {
 947                 struct vmbus_channel *channel;
 948
 949                 if (!sync_test_and_clear_bit(relid, recv_int_page))
 950                         continue;
 951
 952                 /* Special case - vmbus channel protocol msg */
 953                 if (relid == 0)
 954                         continue;
 955
 956                 rcu_read_lock();
 957
 958                 /* Find channel based on relid */
 959                 list_for_each_entry_rcu(channel, &hv_cpu->chan_list, percpu_list) {
 960                         if (channel->offermsg.child_relid != relid)
 961                                 continue;
 962
 963                         if (channel->rescind)
 964                                 continue;
 965
 966                         trace_vmbus_chan_sched(channel);
 967
 968                         ++channel->interrupts;
 969
 970                         switch (channel->callback_mode) {
 971                         case HV_CALL_ISR:
 972                                 vmbus_channel_isr(channel);
 973                                 break;
 974
 975                         case HV_CALL_BATCHED:
 976                                 hv_begin_read(&channel->inbound);
 977                                 /* fallthrough */
 978                         case HV_CALL_DIRECT:
 979                                 tasklet_schedule(&channel->callback_event);
 980                         }
 981                 }
 982
 983                 rcu_read_unlock();
 984         }
 985 }
 986
 987 static void vmbus_isr(void)
 988 {
 989         struct hv_per_cpu_context *hv_cpu
 990                 = this_cpu_ptr(hv_context.cpu_context);
 991         void *page_addr = hv_cpu->synic_event_page;
 992         struct hv_message *msg;
 993         union hv_synic_event_flags *event;
 994         bool handled = false;
 995
 996         if (unlikely(page_addr == NULL))
 997                 return;
 998
 999         event = (union hv_synic_event_flags *)page_addr +
1000                                          VMBUS_MESSAGE_SINT;
1001         /*
1002          * Check for events before checking for messages. This is the order
1003          * in which events and messages are checked in Windows guests on
1004          * Hyper-V, and the Windows team suggested we do the same.
1005          */
1006
1007         if ((vmbus_proto_version == VERSION_WS2008) ||
1008                 (vmbus_proto_version == VERSION_WIN7)) {
1009
1010                 /* Since we are a child, we only need to check bit 0 */
1011                 if (sync_test_and_clear_bit(0, event->flags))
1012                         handled = true;
1013         } else {
1014                 /*
1015                  * Our host is win8 or above. The signaling mechanism
1016                  * has changed and we can directly look at the event page.
1017                  * If bit n is set then we have an interrup on the channel
1018                  * whose id is n.
1019                  */
1020                 handled = true;
1021         }
1022
1023         if (handled)
1024                 vmbus_chan_sched(hv_cpu);
1025
1026         page_addr = hv_cpu->synic_message_page;
1027         msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
1028
1029         /* Check if there are actual msgs to be processed */
1030         if (msg->header.message_type != HVMSG_NONE) {
1031                 if (msg->header.message_type == HVMSG_TIMER_EXPIRED)
1032                         hv_process_timer_expiration(msg, hv_cpu);
1033                 else
1034                         tasklet_schedule(&hv_cpu->msg_dpc);
1035         }
1036
1037         add_interrupt_randomness(HYPERVISOR_CALLBACK_VECTOR, 0);
1038 }
1039
1040 /*
1041  * Boolean to control whether to report panic messages over Hyper-V.
1042  *
1043  * It can be set via /proc/sys/kernel/hyperv/record_panic_msg
1044  */
1045 static int sysctl_record_panic_msg = 1;
1046
1047 /*
1048  * Callback from kmsg_dump. Grab as much as possible from the end of the kmsg
1049  * buffer and call into Hyper-V to transfer the data.
1050  */
1051 static void hv_kmsg_dump(struct kmsg_dumper *dumper,
1052                          enum kmsg_dump_reason reason)
1053 {
1054         size_t bytes_written;
1055         phys_addr_t panic_pa;
1056
1057         /* We are only interested in panics. */
1058         if ((reason != KMSG_DUMP_PANIC) || (!sysctl_record_panic_msg))
1059                 return;
1060
1061         panic_pa = virt_to_phys(hv_panic_page);
1062
1063         /*
1064          * Write dump contents to the page. No need to synchronize; panic should
1065          * be single-threaded.
1066          */
1067         kmsg_dump_get_buffer(dumper, true, hv_panic_page, PAGE_SIZE,
1068                              &bytes_written);
1069         if (bytes_written)
1070                 hyperv_report_panic_msg(panic_pa, bytes_written);
1071 }
1072
1073 static struct kmsg_dumper hv_kmsg_dumper = {
1074         .dump = hv_kmsg_dump,
1075 };
1076
1077 static struct ctl_table_header *hv_ctl_table_hdr;
1078 static int zero;
1079 static int one = 1;
1080
1081 /*
1082  * sysctl option to allow the user to control whether kmsg data should be
1083  * reported to Hyper-V on panic.
1084  */
1085 static struct ctl_table hv_ctl_table[] = {
1086         {
1087                 .procname       = "hyperv_record_panic_msg",
1088                 .data           = &sysctl_record_panic_msg,
1089                 .maxlen         = sizeof(int),
1090                 .mode           = 0644,
1091                 .proc_handler   = proc_dointvec_minmax,
1092                 .extra1         = &zero,
1093                 .extra2         = &one
1094         },
1095         {}
1096 };
1097
1098 static struct ctl_table hv_root_table[] = {
1099         {
1100                 .procname       = "kernel",
1101                 .mode           = 0555,
1102                 .child          = hv_ctl_table
1103         },
1104         {}
1105 };
1106
1107 /*
1108  * vmbus_bus_init -Main vmbus driver initialization routine.
1109  *
1110  * Here, we
1111  *      - initialize the vmbus driver context
1112  *      - invoke the vmbus hv main init routine
1113  *      - retrieve the channel offers
1114  */
1115 static int vmbus_bus_init(void)
1116 {
1117         int ret;
1118
1119         /* Hypervisor initialization...setup hypercall page..etc */
1120         ret = hv_init();
1121         if (ret != 0) {
1122                 pr_err("Unable to initialize the hypervisor - 0x%x\n", ret);
1123                 return ret;
1124         }
1125
1126         ret = bus_register(&hv_bus);
1127         if (ret)
1128                 return ret;
1129
1130         hv_setup_vmbus_irq(vmbus_isr);
1131
1132         ret = hv_synic_alloc();
1133         if (ret)
1134                 goto err_alloc;
1135         /*
1136          * Initialize the per-cpu interrupt state and
1137          * connect to the host.
1138          */
1139         ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "hyperv/vmbus:online",
1140                                 hv_synic_init, hv_synic_cleanup);
1141         if (ret < 0)
1142                 goto err_alloc;
1143         hyperv_cpuhp_online = ret;
1144
1145         ret = vmbus_connect();
1146         if (ret)
1147                 goto err_connect;
1148
1149         /*
1150          * Only register if the crash MSRs are available
1151          */
1152         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1153                 u64 hyperv_crash_ctl;
1154                 /*
1155                  * Sysctl registration is not fatal, since by default
1156                  * reporting is enabled.
1157                  */
1158                 hv_ctl_table_hdr = register_sysctl_table(hv_root_table);
1159                 if (!hv_ctl_table_hdr)
1160                         pr_err("Hyper-V: sysctl table register error");
1161
1162                 /*
1163                  * Register for panic kmsg callback only if the right
1164                  * capability is supported by the hypervisor.
1165                  */
1166                 hv_get_crash_ctl(hyperv_crash_ctl);
1167                 if (hyperv_crash_ctl & HV_CRASH_CTL_CRASH_NOTIFY_MSG) {
1168                         hv_panic_page = (void *)get_zeroed_page(GFP_KERNEL);
1169                         if (hv_panic_page) {
1170                                 ret = kmsg_dump_register(&hv_kmsg_dumper);
1171                                 if (ret)
1172                                         pr_err("Hyper-V: kmsg dump register "
1173                                                 "error 0x%x\n", ret);
1174                         } else
1175                                 pr_err("Hyper-V: panic message page memory "
1176                                         "allocation failed");
1177                 }
1178
1179                 register_die_notifier(&hyperv_die_block);
1180                 atomic_notifier_chain_register(&panic_notifier_list,
1181                                                &hyperv_panic_block);
1182         }
1183
1184         vmbus_request_offers();
1185
1186         return 0;
1187
1188 err_connect:
1189         cpuhp_remove_state(hyperv_cpuhp_online);
1190 err_alloc:
1191         hv_synic_free();
1192         hv_remove_vmbus_irq();
1193
1194         bus_unregister(&hv_bus);
1195         free_page((unsigned long)hv_panic_page);
1196         unregister_sysctl_table(hv_ctl_table_hdr);
1197         hv_ctl_table_hdr = NULL;
1198         return ret;
1199 }
1200
1201 /**
1202  * __vmbus_child_driver_register() - Register a vmbus's driver
1203  * @hv_driver: Pointer to driver structure you want to register
1204  * @owner: owner module of the drv
1205  * @mod_name: module name string
1206  *
1207  * Registers the given driver with Linux through the 'driver_register()' call
1208  * and sets up the hyper-v vmbus handling for this driver.
1209  * It will return the state of the 'driver_register()' call.
1210  *
1211  */
1212 int __vmbus_driver_register(struct hv_driver *hv_driver, struct module *owner, const char *mod_name)
1213 {
1214         int ret;
1215
1216         pr_info("registering driver %s\n", hv_driver->name);
1217
1218         ret = vmbus_exists();
1219         if (ret < 0)
1220                 return ret;
1221
1222         hv_driver->driver.name = hv_driver->name;
1223         hv_driver->driver.owner = owner;
1224         hv_driver->driver.mod_name = mod_name;
1225         hv_driver->driver.bus = &hv_bus;
1226
1227         spin_lock_init(&hv_driver->dynids.lock);
1228         INIT_LIST_HEAD(&hv_driver->dynids.list);
1229
1230         ret = driver_register(&hv_driver->driver);
1231
1232         return ret;
1233 }
1234 EXPORT_SYMBOL_GPL(__vmbus_driver_register);
1235
1236 /**
1237  * vmbus_driver_unregister() - Unregister a vmbus's driver
1238  * @hv_driver: Pointer to driver structure you want to
1239  *             un-register
1240  *
1241  * Un-register the given driver that was previous registered with a call to
1242  * vmbus_driver_register()
1243  */
1244 void vmbus_driver_unregister(struct hv_driver *hv_driver)
1245 {
1246         pr_info("unregistering driver %s\n", hv_driver->name);
1247
1248         if (!vmbus_exists()) {
1249                 driver_unregister(&hv_driver->driver);
1250                 vmbus_free_dynids(hv_driver);
1251         }
1252 }
1253 EXPORT_SYMBOL_GPL(vmbus_driver_unregister);
1254
1255
1256 /*
1257  * Called when last reference to channel is gone.
1258  */
1259 static void vmbus_chan_release(struct kobject *kobj)
1260 {
1261         struct vmbus_channel *channel
1262                 = container_of(kobj, struct vmbus_channel, kobj);
1263
1264         kfree_rcu(channel, rcu);
1265 }
1266
1267 struct vmbus_chan_attribute {
1268         struct attribute attr;
1269         ssize_t (*show)(const struct vmbus_channel *chan, char *buf);
1270         ssize_t (*store)(struct vmbus_channel *chan,
1271                          const char *buf, size_t count);
1272 };
1273 #define VMBUS_CHAN_ATTR(_name, _mode, _show, _store) \
1274         struct vmbus_chan_attribute chan_attr_##_name \
1275                 = __ATTR(_name, _mode, _show, _store)
1276 #define VMBUS_CHAN_ATTR_RW(_name) \
1277         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RW(_name)
1278 #define VMBUS_CHAN_ATTR_RO(_name) \
1279         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_RO(_name)
1280 #define VMBUS_CHAN_ATTR_WO(_name) \
1281         struct vmbus_chan_attribute chan_attr_##_name = __ATTR_WO(_name)
1282
1283 static ssize_t vmbus_chan_attr_show(struct kobject *kobj,
1284                                     struct attribute *attr, char *buf)
1285 {
1286         const struct vmbus_chan_attribute *attribute
1287                 = container_of(attr, struct vmbus_chan_attribute, attr);
1288         const struct vmbus_channel *chan
1289                 = container_of(kobj, struct vmbus_channel, kobj);
1290
1291         if (!attribute->show)
1292                 return -EIO;
1293
1294         return attribute->show(chan, buf);
1295 }
1296
1297 static const struct sysfs_ops vmbus_chan_sysfs_ops = {
1298         .show = vmbus_chan_attr_show,
1299 };
1300
1301 static ssize_t out_mask_show(const struct vmbus_channel *channel, char *buf)
1302 {
1303         const struct hv_ring_buffer_info *rbi = &channel->outbound;
1304
1305         return sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1306 }
1307 static VMBUS_CHAN_ATTR_RO(out_mask);
1308
1309 static ssize_t in_mask_show(const struct vmbus_channel *channel, char *buf)
1310 {
1311         const struct hv_ring_buffer_info *rbi = &channel->inbound;
1312
1313         return sprintf(buf, "%u\n", rbi->ring_buffer->interrupt_mask);
1314 }
1315 static VMBUS_CHAN_ATTR_RO(in_mask);
1316
1317 static ssize_t read_avail_show(const struct vmbus_channel *channel, char *buf)
1318 {
1319         const struct hv_ring_buffer_info *rbi = &channel->inbound;
1320
1321         return sprintf(buf, "%u\n", hv_get_bytes_to_read(rbi));
1322 }
1323 static VMBUS_CHAN_ATTR_RO(read_avail);
1324
1325 static ssize_t write_avail_show(const struct vmbus_channel *channel, char *buf)
1326 {
1327         const struct hv_ring_buffer_info *rbi = &channel->outbound;
1328
1329         return sprintf(buf, "%u\n", hv_get_bytes_to_write(rbi));
1330 }
1331 static VMBUS_CHAN_ATTR_RO(write_avail);
1332
1333 static ssize_t show_target_cpu(const struct vmbus_channel *channel, char *buf)
1334 {
1335         return sprintf(buf, "%u\n", channel->target_cpu);
1336 }
1337 static VMBUS_CHAN_ATTR(cpu, S_IRUGO, show_target_cpu, NULL);
1338
1339 static ssize_t channel_pending_show(const struct vmbus_channel *channel,
1340                                     char *buf)
1341 {
1342         return sprintf(buf, "%d\n",
1343                        channel_pending(channel,
1344                                        vmbus_connection.monitor_pages[1]));
1345 }
1346 static VMBUS_CHAN_ATTR(pending, S_IRUGO, channel_pending_show, NULL);
1347
1348 static ssize_t channel_latency_show(const struct vmbus_channel *channel,
1349                                     char *buf)
1350 {
1351         return sprintf(buf, "%d\n",
1352                        channel_latency(channel,
1353                                        vmbus_connection.monitor_pages[1]));
1354 }
1355 static VMBUS_CHAN_ATTR(latency, S_IRUGO, channel_latency_show, NULL);
1356
1357 static ssize_t channel_interrupts_show(const struct vmbus_channel *channel, char *buf)
1358 {
1359         return sprintf(buf, "%llu\n", channel->interrupts);
1360 }
1361 static VMBUS_CHAN_ATTR(interrupts, S_IRUGO, channel_interrupts_show, NULL);
1362
1363 static ssize_t channel_events_show(const struct vmbus_channel *channel, char *buf)
1364 {
1365         return sprintf(buf, "%llu\n", channel->sig_events);
1366 }
1367 static VMBUS_CHAN_ATTR(events, S_IRUGO, channel_events_show, NULL);
1368
1369 static ssize_t subchannel_monitor_id_show(const struct vmbus_channel *channel,
1370                                           char *buf)
1371 {
1372         return sprintf(buf, "%u\n", channel->offermsg.monitorid);
1373 }
1374 static VMBUS_CHAN_ATTR(monitor_id, S_IRUGO, subchannel_monitor_id_show, NULL);
1375
1376 static ssize_t subchannel_id_show(const struct vmbus_channel *channel,
1377                                   char *buf)
1378 {
1379         return sprintf(buf, "%u\n",
1380                        channel->offermsg.offer.sub_channel_index);
1381 }
1382 static VMBUS_CHAN_ATTR_RO(subchannel_id);
1383
1384 static struct attribute *vmbus_chan_attrs[] = {
1385         &chan_attr_out_mask.attr,
1386         &chan_attr_in_mask.attr,
1387         &chan_attr_read_avail.attr,
1388         &chan_attr_write_avail.attr,
1389         &chan_attr_cpu.attr,
1390         &chan_attr_pending.attr,
1391         &chan_attr_latency.attr,
1392         &chan_attr_interrupts.attr,
1393         &chan_attr_events.attr,
1394         &chan_attr_monitor_id.attr,
1395         &chan_attr_subchannel_id.attr,
1396         NULL
1397 };
1398
1399 static struct kobj_type vmbus_chan_ktype = {
1400         .sysfs_ops = &vmbus_chan_sysfs_ops,
1401         .release = vmbus_chan_release,
1402         .default_attrs = vmbus_chan_attrs,
1403 };
1404
1405 /*
1406  * vmbus_add_channel_kobj - setup a sub-directory under device/channels
1407  */
1408 int vmbus_add_channel_kobj(struct hv_device *dev, struct vmbus_channel *channel)
1409 {
1410         struct kobject *kobj = &channel->kobj;
1411         u32 relid = channel->offermsg.child_relid;
1412         int ret;
1413
1414         kobj->kset = dev->channels_kset;
1415         ret = kobject_init_and_add(kobj, &vmbus_chan_ktype, NULL,
1416                                    "%u", relid);
1417         if (ret)
1418                 return ret;
1419
1420         kobject_uevent(kobj, KOBJ_ADD);
1421
1422         return 0;
1423 }
1424
1425 /*
1426  * vmbus_device_create - Creates and registers a new child device
1427  * on the vmbus.
1428  */
1429 struct hv_device *vmbus_device_create(const uuid_le *type,
1430                                       const uuid_le *instance,
1431                                       struct vmbus_channel *channel)
1432 {
1433         struct hv_device *child_device_obj;
1434
1435         child_device_obj = kzalloc(sizeof(struct hv_device), GFP_KERNEL);
1436         if (!child_device_obj) {
1437                 pr_err("Unable to allocate device object for child device\n");
1438                 return NULL;
1439         }
1440
1441         child_device_obj->channel = channel;
1442         memcpy(&child_device_obj->dev_type, type, sizeof(uuid_le));
1443         memcpy(&child_device_obj->dev_instance, instance,
1444                sizeof(uuid_le));
1445         child_device_obj->vendor_id = 0x1414; /* MSFT vendor ID */
1446
1447
1448         return child_device_obj;
1449 }
1450
1451 /*
1452  * vmbus_device_register - Register the child device
1453  */
1454 int vmbus_device_register(struct hv_device *child_device_obj)
1455 {
1456         struct kobject *kobj = &child_device_obj->device.kobj;
1457         int ret;
1458
1459         dev_set_name(&child_device_obj->device, "%pUl",
1460                      child_device_obj->channel->offermsg.offer.if_instance.b);
1461
1462         child_device_obj->device.bus = &hv_bus;
1463         child_device_obj->device.parent = &hv_acpi_dev->dev;
1464         child_device_obj->device.release = vmbus_device_release;
1465
1466         /*
1467          * Register with the LDM. This will kick off the driver/device
1468          * binding...which will eventually call vmbus_match() and vmbus_probe()
1469          */
1470         ret = device_register(&child_device_obj->device);
1471         if (ret) {
1472                 pr_err("Unable to register child device\n");
1473                 return ret;
1474         }
1475
1476         child_device_obj->channels_kset = kset_create_and_add("channels",
1477                                                               NULL, kobj);
1478         if (!child_device_obj->channels_kset) {
1479                 ret = -ENOMEM;
1480                 goto err_dev_unregister;
1481         }
1482
1483         ret = vmbus_add_channel_kobj(child_device_obj,
1484                                      child_device_obj->channel);
1485         if (ret) {
1486                 pr_err("Unable to register primary channeln");
1487                 goto err_kset_unregister;
1488         }
1489
1490         return 0;
1491
1492 err_kset_unregister:
1493         kset_unregister(child_device_obj->channels_kset);
1494
1495 err_dev_unregister:
1496         device_unregister(&child_device_obj->device);
1497         return ret;
1498 }
1499
1500 /*
1501  * vmbus_device_unregister - Remove the specified child device
1502  * from the vmbus.
1503  */
1504 void vmbus_device_unregister(struct hv_device *device_obj)
1505 {
1506         pr_debug("child device %s unregistered\n",
1507                 dev_name(&device_obj->device));
1508
1509         kset_unregister(device_obj->channels_kset);
1510
1511         /*
1512          * Kick off the process of unregistering the device.
1513          * This will call vmbus_remove() and eventually vmbus_device_release()
1514          */
1515         device_unregister(&device_obj->device);
1516 }
1517
1518
1519 /*
1520  * VMBUS is an acpi enumerated device. Get the information we
1521  * need from DSDT.
1522  */
1523 #define VTPM_BASE_ADDRESS 0xfed40000
1524 static acpi_status vmbus_walk_resources(struct acpi_resource *res, void *ctx)
1525 {
1526         resource_size_t start = 0;
1527         resource_size_t end = 0;
1528         struct resource *new_res;
1529         struct resource **old_res = &hyperv_mmio;
1530         struct resource **prev_res = NULL;
1531
1532         switch (res->type) {
1533
1534         /*
1535          * "Address" descriptors are for bus windows. Ignore
1536          * "memory" descriptors, which are for registers on
1537          * devices.
1538          */
1539         case ACPI_RESOURCE_TYPE_ADDRESS32:
1540                 start = res->data.address32.address.minimum;
1541                 end = res->data.address32.address.maximum;
1542                 break;
1543
1544         case ACPI_RESOURCE_TYPE_ADDRESS64:
1545                 start = res->data.address64.address.minimum;
1546                 end = res->data.address64.address.maximum;
1547                 break;
1548
1549         default:
1550                 /* Unused resource type */
1551                 return AE_OK;
1552
1553         }
1554         /*
1555          * Ignore ranges that are below 1MB, as they're not
1556          * necessary or useful here.
1557          */
1558         if (end < 0x100000)
1559                 return AE_OK;
1560
1561         new_res = kzalloc(sizeof(*new_res), GFP_ATOMIC);
1562         if (!new_res)
1563                 return AE_NO_MEMORY;
1564
1565         /* If this range overlaps the virtual TPM, truncate it. */
1566         if (end > VTPM_BASE_ADDRESS && start < VTPM_BASE_ADDRESS)
1567                 end = VTPM_BASE_ADDRESS;
1568
1569         new_res->name = "hyperv mmio";
1570         new_res->flags = IORESOURCE_MEM;
1571         new_res->start = start;
1572         new_res->end = end;
1573
1574         /*
1575          * If two ranges are adjacent, merge them.
1576          */
1577         do {
1578                 if (!*old_res) {
1579                         *old_res = new_res;
1580                         break;
1581                 }
1582
1583                 if (((*old_res)->end + 1) == new_res->start) {
1584                         (*old_res)->end = new_res->end;
1585                         kfree(new_res);
1586                         break;
1587                 }
1588
1589                 if ((*old_res)->start == new_res->end + 1) {
1590                         (*old_res)->start = new_res->start;
1591                         kfree(new_res);
1592                         break;
1593                 }
1594
1595                 if ((*old_res)->start > new_res->end) {
1596                         new_res->sibling = *old_res;
1597                         if (prev_res)
1598                                 (*prev_res)->sibling = new_res;
1599                         *old_res = new_res;
1600                         break;
1601                 }
1602
1603                 prev_res = old_res;
1604                 old_res = &(*old_res)->sibling;
1605
1606         } while (1);
1607
1608         return AE_OK;
1609 }
1610
1611 static int vmbus_acpi_remove(struct acpi_device *device)
1612 {
1613         struct resource *cur_res;
1614         struct resource *next_res;
1615
1616         if (hyperv_mmio) {
1617                 if (fb_mmio) {
1618                         __release_region(hyperv_mmio, fb_mmio->start,
1619                                          resource_size(fb_mmio));
1620                         fb_mmio = NULL;
1621                 }
1622
1623                 for (cur_res = hyperv_mmio; cur_res; cur_res = next_res) {
1624                         next_res = cur_res->sibling;
1625                         kfree(cur_res);
1626                 }
1627         }
1628
1629         return 0;
1630 }
1631
1632 static void vmbus_reserve_fb(void)
1633 {
1634         int size;
1635         /*
1636          * Make a claim for the frame buffer in the resource tree under the
1637          * first node, which will be the one below 4GB.  The length seems to
1638          * be underreported, particularly in a Generation 1 VM.  So start out
1639          * reserving a larger area and make it smaller until it succeeds.
1640          */
1641
1642         if (screen_info.lfb_base) {
1643                 if (efi_enabled(EFI_BOOT))
1644                         size = max_t(__u32, screen_info.lfb_size, 0x800000);
1645                 else
1646                         size = max_t(__u32, screen_info.lfb_size, 0x4000000);
1647
1648                 for (; !fb_mmio && (size >= 0x100000); size >>= 1) {
1649                         fb_mmio = __request_region(hyperv_mmio,
1650                                                    screen_info.lfb_base, size,
1651                                                    fb_mmio_name, 0);
1652                 }
1653         }
1654 }
1655
1656 /**
1657  * vmbus_allocate_mmio() - Pick a memory-mapped I/O range.
1658  * @new:                If successful, supplied a pointer to the
1659  *                      allocated MMIO space.
1660  * @device_obj:         Identifies the caller
1661  * @min:                Minimum guest physical address of the
1662  *                      allocation
1663  * @max:                Maximum guest physical address
1664  * @size:               Size of the range to be allocated
1665  * @align:              Alignment of the range to be allocated
1666  * @fb_overlap_ok:      Whether this allocation can be allowed
1667  *                      to overlap the video frame buffer.
1668  *
1669  * This function walks the resources granted to VMBus by the
1670  * _CRS object in the ACPI namespace underneath the parent
1671  * "bridge" whether that's a root PCI bus in the Generation 1
1672  * case or a Module Device in the Generation 2 case.  It then
1673  * attempts to allocate from the global MMIO pool in a way that
1674  * matches the constraints supplied in these parameters and by
1675  * that _CRS.
1676  *
1677  * Return: 0 on success, -errno on failure
1678  */
1679 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
1680                         resource_size_t min, resource_size_t max,
1681                         resource_size_t size, resource_size_t align,
1682                         bool fb_overlap_ok)
1683 {
1684         struct resource *iter, *shadow;
1685         resource_size_t range_min, range_max, start;
1686         const char *dev_n = dev_name(&device_obj->device);
1687         int retval;
1688
1689         retval = -ENXIO;
1690         down(&hyperv_mmio_lock);
1691
1692         /*
1693          * If overlaps with frame buffers are allowed, then first attempt to
1694          * make the allocation from within the reserved region.  Because it
1695          * is already reserved, no shadow allocation is necessary.
1696          */
1697         if (fb_overlap_ok && fb_mmio && !(min > fb_mmio->end) &&
1698             !(max < fb_mmio->start)) {
1699
1700                 range_min = fb_mmio->start;
1701                 range_max = fb_mmio->end;
1702                 start = (range_min + align - 1) & ~(align - 1);
1703                 for (; start + size - 1 <= range_max; start += align) {
1704                         *new = request_mem_region_exclusive(start, size, dev_n);
1705                         if (*new) {
1706                                 retval = 0;
1707                                 goto exit;
1708                         }
1709                 }
1710         }
1711
1712         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
1713                 if ((iter->start >= max) || (iter->end <= min))
1714                         continue;
1715
1716                 range_min = iter->start;
1717                 range_max = iter->end;
1718                 start = (range_min + align - 1) & ~(align - 1);
1719                 for (; start + size - 1 <= range_max; start += align) {
1720                         shadow = __request_region(iter, start, size, NULL,
1721                                                   IORESOURCE_BUSY);
1722                         if (!shadow)
1723                                 continue;
1724
1725                         *new = request_mem_region_exclusive(start, size, dev_n);
1726                         if (*new) {
1727                                 shadow->name = (char *)*new;
1728                                 retval = 0;
1729                                 goto exit;
1730                         }
1731
1732                         __release_region(iter, start, size);
1733                 }
1734         }
1735
1736 exit:
1737         up(&hyperv_mmio_lock);
1738         return retval;
1739 }
1740 EXPORT_SYMBOL_GPL(vmbus_allocate_mmio);
1741
1742 /**
1743  * vmbus_free_mmio() - Free a memory-mapped I/O range.
1744  * @start:              Base address of region to release.
1745  * @size:               Size of the range to be allocated
1746  *
1747  * This function releases anything requested by
1748  * vmbus_mmio_allocate().
1749  */
1750 void vmbus_free_mmio(resource_size_t start, resource_size_t size)
1751 {
1752         struct resource *iter;
1753
1754         down(&hyperv_mmio_lock);
1755         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
1756                 if ((iter->start >= start + size) || (iter->end <= start))
1757                         continue;
1758
1759                 __release_region(iter, start, size);
1760         }
1761         release_mem_region(start, size);
1762         up(&hyperv_mmio_lock);
1763
1764 }
1765 EXPORT_SYMBOL_GPL(vmbus_free_mmio);
1766
1767 static int vmbus_acpi_add(struct acpi_device *device)
1768 {
1769         acpi_status result;
1770         int ret_val = -ENODEV;
1771         struct acpi_device *ancestor;
1772
1773         hv_acpi_dev = device;
1774
1775         result = acpi_walk_resources(device->handle, METHOD_NAME__CRS,
1776                                         vmbus_walk_resources, NULL);
1777
1778         if (ACPI_FAILURE(result))
1779                 goto acpi_walk_err;
1780         /*
1781          * Some ancestor of the vmbus acpi device (Gen1 or Gen2
1782          * firmware) is the VMOD that has the mmio ranges. Get that.
1783          */
1784         for (ancestor = device->parent; ancestor; ancestor = ancestor->parent) {
1785                 result = acpi_walk_resources(ancestor->handle, METHOD_NAME__CRS,
1786                                              vmbus_walk_resources, NULL);
1787
1788                 if (ACPI_FAILURE(result))
1789                         continue;
1790                 if (hyperv_mmio) {
1791                         vmbus_reserve_fb();
1792                         break;
1793                 }
1794         }
1795         ret_val = 0;
1796
1797 acpi_walk_err:
1798         complete(&probe_event);
1799         if (ret_val)
1800                 vmbus_acpi_remove(device);
1801         return ret_val;
1802 }
1803
1804 static const struct acpi_device_id vmbus_acpi_device_ids[] = {
1805         {"VMBUS", 0},
1806         {"VMBus", 0},
1807         {"", 0},
1808 };
1809 MODULE_DEVICE_TABLE(acpi, vmbus_acpi_device_ids);
1810
1811 static struct acpi_driver vmbus_acpi_driver = {
1812         .name = "vmbus",
1813         .ids = vmbus_acpi_device_ids,
1814         .ops = {
1815                 .add = vmbus_acpi_add,
1816                 .remove = vmbus_acpi_remove,
1817         },
1818 };
1819
1820 static void hv_kexec_handler(void)
1821 {
1822         hv_synic_clockevents_cleanup();
1823         vmbus_initiate_unload(false);
1824         vmbus_connection.conn_state = DISCONNECTED;
1825         /* Make sure conn_state is set as hv_synic_cleanup checks for it */
1826         mb();
1827         cpuhp_remove_state(hyperv_cpuhp_online);
1828         hyperv_cleanup();
1829 };
1830
1831 static void hv_crash_handler(struct pt_regs *regs)
1832 {
1833         vmbus_initiate_unload(true);
1834         /*
1835          * In crash handler we can't schedule synic cleanup for all CPUs,
1836          * doing the cleanup for current CPU only. This should be sufficient
1837          * for kdump.
1838          */
1839         vmbus_connection.conn_state = DISCONNECTED;
1840         hv_synic_cleanup(smp_processor_id());
1841         hyperv_cleanup();
1842 };
1843
1844 static int __init hv_acpi_init(void)
1845 {
1846         int ret, t;
1847
1848         if (!hv_is_hyperv_initialized())
1849                 return -ENODEV;
1850
1851         init_completion(&probe_event);
1852
1853         /*
1854          * Get ACPI resources first.
1855          */
1856         ret = acpi_bus_register_driver(&vmbus_acpi_driver);
1857
1858         if (ret)
1859                 return ret;
1860
1861         t = wait_for_completion_timeout(&probe_event, 5*HZ);
1862         if (t == 0) {
1863                 ret = -ETIMEDOUT;
1864                 goto cleanup;
1865         }
1866
1867         ret = vmbus_bus_init();
1868         if (ret)
1869                 goto cleanup;
1870
1871         hv_setup_kexec_handler(hv_kexec_handler);
1872         hv_setup_crash_handler(hv_crash_handler);
1873
1874         return 0;
1875
1876 cleanup:
1877         acpi_bus_unregister_driver(&vmbus_acpi_driver);
1878         hv_acpi_dev = NULL;
1879         return ret;
1880 }
1881
1882 static void __exit vmbus_exit(void)
1883 {
1884         int cpu;
1885
1886         hv_remove_kexec_handler();
1887         hv_remove_crash_handler();
1888         vmbus_connection.conn_state = DISCONNECTED;
1889         hv_synic_clockevents_cleanup();
1890         vmbus_disconnect();
1891         hv_remove_vmbus_irq();
1892         for_each_online_cpu(cpu) {
1893                 struct hv_per_cpu_context *hv_cpu
1894                         = per_cpu_ptr(hv_context.cpu_context, cpu);
1895
1896                 tasklet_kill(&hv_cpu->msg_dpc);
1897         }
1898         vmbus_free_channels();
1899
1900         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1901                 kmsg_dump_unregister(&hv_kmsg_dumper);
1902                 unregister_die_notifier(&hyperv_die_block);
1903                 atomic_notifier_chain_unregister(&panic_notifier_list,
1904                                                  &hyperv_panic_block);
1905         }
1906
1907         free_page((unsigned long)hv_panic_page);
1908         unregister_sysctl_table(hv_ctl_table_hdr);
1909         hv_ctl_table_hdr = NULL;
1910         bus_unregister(&hv_bus);
1911
1912         cpuhp_remove_state(hyperv_cpuhp_online);
1913         hv_synic_free();
1914         acpi_bus_unregister_driver(&vmbus_acpi_driver);
1915 }
1916
1917
1918 MODULE_LICENSE("GPL");
1919
1920 subsys_initcall(hv_acpi_init);
1921 module_exit(vmbus_exit);