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 <asm/hyperv.h>
  38 #include <asm/hypervisor.h>
  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/random.h>
  45 #include "hyperv_vmbus.h"
  46
  47 static struct acpi_device  *hv_acpi_dev;
  48
  49 static struct tasklet_struct msg_dpc;
  50 static struct completion probe_event;
  51 static int irq;
  52
  53
  54 static void hyperv_report_panic(struct pt_regs *regs)
  55 {
  56         static bool panic_reported;
  57
  58         /*
  59          * We prefer to report panic on 'die' chain as we have proper
  60          * registers to report, but if we miss it (e.g. on BUG()) we need
  61          * to report it on 'panic'.
  62          */
  63         if (panic_reported)
  64                 return;
  65         panic_reported = true;
  66
  67         wrmsrl(HV_X64_MSR_CRASH_P0, regs->ip);
  68         wrmsrl(HV_X64_MSR_CRASH_P1, regs->ax);
  69         wrmsrl(HV_X64_MSR_CRASH_P2, regs->bx);
  70         wrmsrl(HV_X64_MSR_CRASH_P3, regs->cx);
  71         wrmsrl(HV_X64_MSR_CRASH_P4, regs->dx);
  72
  73         /*
  74          * Let Hyper-V know there is crash data available
  75          */
  76         wrmsrl(HV_X64_MSR_CRASH_CTL, HV_CRASH_CTL_CRASH_NOTIFY);
  77 }
  78
  79 static int hyperv_panic_event(struct notifier_block *nb, unsigned long val,
  80                               void *args)
  81 {
  82         struct pt_regs *regs;
  83
  84         regs = current_pt_regs();
  85
  86         hyperv_report_panic(regs);
  87         return NOTIFY_DONE;
  88 }
  89
  90 static int hyperv_die_event(struct notifier_block *nb, unsigned long val,
  91                             void *args)
  92 {
  93         struct die_args *die = (struct die_args *)args;
  94         struct pt_regs *regs = die->regs;
  95
  96         hyperv_report_panic(regs);
  97         return NOTIFY_DONE;
  98 }
  99
 100 static struct notifier_block hyperv_die_block = {
 101         .notifier_call = hyperv_die_event,
 102 };
 103 static struct notifier_block hyperv_panic_block = {
 104         .notifier_call = hyperv_panic_event,
 105 };
 106
 107 struct resource *hyperv_mmio;
 108 DEFINE_SEMAPHORE(hyperv_mmio_lock);
 109
 110 static int vmbus_exists(void)
 111 {
 112         if (hv_acpi_dev == NULL)
 113                 return -ENODEV;
 114
 115         return 0;
 116 }
 117
 118 #define VMBUS_ALIAS_LEN ((sizeof((struct hv_vmbus_device_id *)0)->guid) * 2)
 119 static void print_alias_name(struct hv_device *hv_dev, char *alias_name)
 120 {
 121         int i;
 122         for (i = 0; i < VMBUS_ALIAS_LEN; i += 2)
 123                 sprintf(&alias_name[i], "%02x", hv_dev->dev_type.b[i/2]);
 124 }
 125
 126 static u8 channel_monitor_group(struct vmbus_channel *channel)
 127 {
 128         return (u8)channel->offermsg.monitorid / 32;
 129 }
 130
 131 static u8 channel_monitor_offset(struct vmbus_channel *channel)
 132 {
 133         return (u8)channel->offermsg.monitorid % 32;
 134 }
 135
 136 static u32 channel_pending(struct vmbus_channel *channel,
 137                            struct hv_monitor_page *monitor_page)
 138 {
 139         u8 monitor_group = channel_monitor_group(channel);
 140         return monitor_page->trigger_group[monitor_group].pending;
 141 }
 142
 143 static u32 channel_latency(struct vmbus_channel *channel,
 144                            struct hv_monitor_page *monitor_page)
 145 {
 146         u8 monitor_group = channel_monitor_group(channel);
 147         u8 monitor_offset = channel_monitor_offset(channel);
 148         return monitor_page->latency[monitor_group][monitor_offset];
 149 }
 150
 151 static u32 channel_conn_id(struct vmbus_channel *channel,
 152                            struct hv_monitor_page *monitor_page)
 153 {
 154         u8 monitor_group = channel_monitor_group(channel);
 155         u8 monitor_offset = channel_monitor_offset(channel);
 156         return monitor_page->parameter[monitor_group][monitor_offset].connectionid.u.id;
 157 }
 158
 159 static ssize_t id_show(struct device *dev, struct device_attribute *dev_attr,
 160                        char *buf)
 161 {
 162         struct hv_device *hv_dev = device_to_hv_device(dev);
 163
 164         if (!hv_dev->channel)
 165                 return -ENODEV;
 166         return sprintf(buf, "%d\n", hv_dev->channel->offermsg.child_relid);
 167 }
 168 static DEVICE_ATTR_RO(id);
 169
 170 static ssize_t state_show(struct device *dev, struct device_attribute *dev_attr,
 171                           char *buf)
 172 {
 173         struct hv_device *hv_dev = device_to_hv_device(dev);
 174
 175         if (!hv_dev->channel)
 176                 return -ENODEV;
 177         return sprintf(buf, "%d\n", hv_dev->channel->state);
 178 }
 179 static DEVICE_ATTR_RO(state);
 180
 181 static ssize_t monitor_id_show(struct device *dev,
 182                                struct device_attribute *dev_attr, char *buf)
 183 {
 184         struct hv_device *hv_dev = device_to_hv_device(dev);
 185
 186         if (!hv_dev->channel)
 187                 return -ENODEV;
 188         return sprintf(buf, "%d\n", hv_dev->channel->offermsg.monitorid);
 189 }
 190 static DEVICE_ATTR_RO(monitor_id);
 191
 192 static ssize_t class_id_show(struct device *dev,
 193                                struct device_attribute *dev_attr, char *buf)
 194 {
 195         struct hv_device *hv_dev = device_to_hv_device(dev);
 196
 197         if (!hv_dev->channel)
 198                 return -ENODEV;
 199         return sprintf(buf, "{%pUl}\n",
 200                        hv_dev->channel->offermsg.offer.if_type.b);
 201 }
 202 static DEVICE_ATTR_RO(class_id);
 203
 204 static ssize_t device_id_show(struct device *dev,
 205                               struct device_attribute *dev_attr, char *buf)
 206 {
 207         struct hv_device *hv_dev = device_to_hv_device(dev);
 208
 209         if (!hv_dev->channel)
 210                 return -ENODEV;
 211         return sprintf(buf, "{%pUl}\n",
 212                        hv_dev->channel->offermsg.offer.if_instance.b);
 213 }
 214 static DEVICE_ATTR_RO(device_id);
 215
 216 static ssize_t modalias_show(struct device *dev,
 217                              struct device_attribute *dev_attr, char *buf)
 218 {
 219         struct hv_device *hv_dev = device_to_hv_device(dev);
 220         char alias_name[VMBUS_ALIAS_LEN + 1];
 221
 222         print_alias_name(hv_dev, alias_name);
 223         return sprintf(buf, "vmbus:%s\n", alias_name);
 224 }
 225 static DEVICE_ATTR_RO(modalias);
 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 /* Set up per device attributes in /sys/bus/vmbus/devices/<bus device> */
 484 static struct attribute *vmbus_attrs[] = {
 485         &dev_attr_id.attr,
 486         &dev_attr_state.attr,
 487         &dev_attr_monitor_id.attr,
 488         &dev_attr_class_id.attr,
 489         &dev_attr_device_id.attr,
 490         &dev_attr_modalias.attr,
 491         &dev_attr_server_monitor_pending.attr,
 492         &dev_attr_client_monitor_pending.attr,
 493         &dev_attr_server_monitor_latency.attr,
 494         &dev_attr_client_monitor_latency.attr,
 495         &dev_attr_server_monitor_conn_id.attr,
 496         &dev_attr_client_monitor_conn_id.attr,
 497         &dev_attr_out_intr_mask.attr,
 498         &dev_attr_out_read_index.attr,
 499         &dev_attr_out_write_index.attr,
 500         &dev_attr_out_read_bytes_avail.attr,
 501         &dev_attr_out_write_bytes_avail.attr,
 502         &dev_attr_in_intr_mask.attr,
 503         &dev_attr_in_read_index.attr,
 504         &dev_attr_in_write_index.attr,
 505         &dev_attr_in_read_bytes_avail.attr,
 506         &dev_attr_in_write_bytes_avail.attr,
 507         &dev_attr_channel_vp_mapping.attr,
 508         NULL,
 509 };
 510 ATTRIBUTE_GROUPS(vmbus);
 511
 512 /*
 513  * vmbus_uevent - add uevent for our device
 514  *
 515  * This routine is invoked when a device is added or removed on the vmbus to
 516  * generate a uevent to udev in the userspace. The udev will then look at its
 517  * rule and the uevent generated here to load the appropriate driver
 518  *
 519  * The alias string will be of the form vmbus:guid where guid is the string
 520  * representation of the device guid (each byte of the guid will be
 521  * represented with two hex characters.
 522  */
 523 static int vmbus_uevent(struct device *device, struct kobj_uevent_env *env)
 524 {
 525         struct hv_device *dev = device_to_hv_device(device);
 526         int ret;
 527         char alias_name[VMBUS_ALIAS_LEN + 1];
 528
 529         print_alias_name(dev, alias_name);
 530         ret = add_uevent_var(env, "MODALIAS=vmbus:%s", alias_name);
 531         return ret;
 532 }
 533
 534 static const uuid_le null_guid;
 535
 536 static inline bool is_null_guid(const __u8 *guid)
 537 {
 538         if (memcmp(guid, &null_guid, sizeof(uuid_le)))
 539                 return false;
 540         return true;
 541 }
 542
 543 /*
 544  * Return a matching hv_vmbus_device_id pointer.
 545  * If there is no match, return NULL.
 546  */
 547 static const struct hv_vmbus_device_id *hv_vmbus_get_id(
 548                                         const struct hv_vmbus_device_id *id,
 549                                         const __u8 *guid)
 550 {
 551         for (; !is_null_guid(id->guid); id++)
 552                 if (!memcmp(&id->guid, guid, sizeof(uuid_le)))
 553                         return id;
 554
 555         return NULL;
 556 }
 557
 558
 559
 560 /*
 561  * vmbus_match - Attempt to match the specified device to the specified driver
 562  */
 563 static int vmbus_match(struct device *device, struct device_driver *driver)
 564 {
 565         struct hv_driver *drv = drv_to_hv_drv(driver);
 566         struct hv_device *hv_dev = device_to_hv_device(device);
 567
 568         if (hv_vmbus_get_id(drv->id_table, hv_dev->dev_type.b))
 569                 return 1;
 570
 571         return 0;
 572 }
 573
 574 /*
 575  * vmbus_probe - Add the new vmbus's child device
 576  */
 577 static int vmbus_probe(struct device *child_device)
 578 {
 579         int ret = 0;
 580         struct hv_driver *drv =
 581                         drv_to_hv_drv(child_device->driver);
 582         struct hv_device *dev = device_to_hv_device(child_device);
 583         const struct hv_vmbus_device_id *dev_id;
 584
 585         dev_id = hv_vmbus_get_id(drv->id_table, dev->dev_type.b);
 586         if (drv->probe) {
 587                 ret = drv->probe(dev, dev_id);
 588                 if (ret != 0)
 589                         pr_err("probe failed for device %s (%d)\n",
 590                                dev_name(child_device), ret);
 591
 592         } else {
 593                 pr_err("probe not set for driver %s\n",
 594                        dev_name(child_device));
 595                 ret = -ENODEV;
 596         }
 597         return ret;
 598 }
 599
 600 /*
 601  * vmbus_remove - Remove a vmbus device
 602  */
 603 static int vmbus_remove(struct device *child_device)
 604 {
 605         struct hv_driver *drv;
 606         struct hv_device *dev = device_to_hv_device(child_device);
 607
 608         if (child_device->driver) {
 609                 drv = drv_to_hv_drv(child_device->driver);
 610                 if (drv->remove)
 611                         drv->remove(dev);
 612         }
 613
 614         return 0;
 615 }
 616
 617
 618 /*
 619  * vmbus_shutdown - Shutdown a vmbus device
 620  */
 621 static void vmbus_shutdown(struct device *child_device)
 622 {
 623         struct hv_driver *drv;
 624         struct hv_device *dev = device_to_hv_device(child_device);
 625
 626
 627         /* The device may not be attached yet */
 628         if (!child_device->driver)
 629                 return;
 630
 631         drv = drv_to_hv_drv(child_device->driver);
 632
 633         if (drv->shutdown)
 634                 drv->shutdown(dev);
 635
 636         return;
 637 }
 638
 639
 640 /*
 641  * vmbus_device_release - Final callback release of the vmbus child device
 642  */
 643 static void vmbus_device_release(struct device *device)
 644 {
 645         struct hv_device *hv_dev = device_to_hv_device(device);
 646         struct vmbus_channel *channel = hv_dev->channel;
 647
 648         hv_process_channel_removal(channel,
 649                                    channel->offermsg.child_relid);
 650         kfree(hv_dev);
 651
 652 }
 653
 654 /* The one and only one */
 655 static struct bus_type  hv_bus = {
 656         .name =         "vmbus",
 657         .match =                vmbus_match,
 658         .shutdown =             vmbus_shutdown,
 659         .remove =               vmbus_remove,
 660         .probe =                vmbus_probe,
 661         .uevent =               vmbus_uevent,
 662         .dev_groups =           vmbus_groups,
 663 };
 664
 665 struct onmessage_work_context {
 666         struct work_struct work;
 667         struct hv_message msg;
 668 };
 669
 670 static void vmbus_onmessage_work(struct work_struct *work)
 671 {
 672         struct onmessage_work_context *ctx;
 673
 674         /* Do not process messages if we're in DISCONNECTED state */
 675         if (vmbus_connection.conn_state == DISCONNECTED)
 676                 return;
 677
 678         ctx = container_of(work, struct onmessage_work_context,
 679                            work);
 680         vmbus_onmessage(&ctx->msg);
 681         kfree(ctx);
 682 }
 683
 684 static void hv_process_timer_expiration(struct hv_message *msg, int cpu)
 685 {
 686         struct clock_event_device *dev = hv_context.clk_evt[cpu];
 687
 688         if (dev->event_handler)
 689                 dev->event_handler(dev);
 690
 691         msg->header.message_type = HVMSG_NONE;
 692
 693         /*
 694          * Make sure the write to MessageType (ie set to
 695          * HVMSG_NONE) happens before we read the
 696          * MessagePending and EOMing. Otherwise, the EOMing
 697          * will not deliver any more messages since there is
 698          * no empty slot
 699          */
 700         mb();
 701
 702         if (msg->header.message_flags.msg_pending) {
 703                 /*
 704                  * This will cause message queue rescan to
 705                  * possibly deliver another msg from the
 706                  * hypervisor
 707                  */
 708                 wrmsrl(HV_X64_MSR_EOM, 0);
 709         }
 710 }
 711
 712 static void vmbus_on_msg_dpc(unsigned long data)
 713 {
 714         int cpu = smp_processor_id();
 715         void *page_addr = hv_context.synic_message_page[cpu];
 716         struct hv_message *msg = (struct hv_message *)page_addr +
 717                                   VMBUS_MESSAGE_SINT;
 718         struct vmbus_channel_message_header *hdr;
 719         struct vmbus_channel_message_table_entry *entry;
 720         struct onmessage_work_context *ctx;
 721
 722         while (1) {
 723                 if (msg->header.message_type == HVMSG_NONE)
 724                         /* no msg */
 725                         break;
 726
 727                 hdr = (struct vmbus_channel_message_header *)msg->u.payload;
 728
 729                 if (hdr->msgtype >= CHANNELMSG_COUNT) {
 730                         WARN_ONCE(1, "unknown msgtype=%d\n", hdr->msgtype);
 731                         goto msg_handled;
 732                 }
 733
 734                 entry = &channel_message_table[hdr->msgtype];
 735                 if (entry->handler_type == VMHT_BLOCKING) {
 736                         ctx = kmalloc(sizeof(*ctx), GFP_ATOMIC);
 737                         if (ctx == NULL)
 738                                 continue;
 739
 740                         INIT_WORK(&ctx->work, vmbus_onmessage_work);
 741                         memcpy(&ctx->msg, msg, sizeof(*msg));
 742
 743                         queue_work(vmbus_connection.work_queue, &ctx->work);
 744                 } else
 745                         entry->message_handler(hdr);
 746
 747 msg_handled:
 748                 msg->header.message_type = HVMSG_NONE;
 749
 750                 /*
 751                  * Make sure the write to MessageType (ie set to
 752                  * HVMSG_NONE) happens before we read the
 753                  * MessagePending and EOMing. Otherwise, the EOMing
 754                  * will not deliver any more messages since there is
 755                  * no empty slot
 756                  */
 757                 mb();
 758
 759                 if (msg->header.message_flags.msg_pending) {
 760                         /*
 761                          * This will cause message queue rescan to
 762                          * possibly deliver another msg from the
 763                          * hypervisor
 764                          */
 765                         wrmsrl(HV_X64_MSR_EOM, 0);
 766                 }
 767         }
 768 }
 769
 770 static void vmbus_isr(void)
 771 {
 772         int cpu = smp_processor_id();
 773         void *page_addr;
 774         struct hv_message *msg;
 775         union hv_synic_event_flags *event;
 776         bool handled = false;
 777
 778         page_addr = hv_context.synic_event_page[cpu];
 779         if (page_addr == NULL)
 780                 return;
 781
 782         event = (union hv_synic_event_flags *)page_addr +
 783                                          VMBUS_MESSAGE_SINT;
 784         /*
 785          * Check for events before checking for messages. This is the order
 786          * in which events and messages are checked in Windows guests on
 787          * Hyper-V, and the Windows team suggested we do the same.
 788          */
 789
 790         if ((vmbus_proto_version == VERSION_WS2008) ||
 791                 (vmbus_proto_version == VERSION_WIN7)) {
 792
 793                 /* Since we are a child, we only need to check bit 0 */
 794                 if (sync_test_and_clear_bit(0,
 795                         (unsigned long *) &event->flags32[0])) {
 796                         handled = true;
 797                 }
 798         } else {
 799                 /*
 800                  * Our host is win8 or above. The signaling mechanism
 801                  * has changed and we can directly look at the event page.
 802                  * If bit n is set then we have an interrup on the channel
 803                  * whose id is n.
 804                  */
 805                 handled = true;
 806         }
 807
 808         if (handled)
 809                 tasklet_schedule(hv_context.event_dpc[cpu]);
 810
 811
 812         page_addr = hv_context.synic_message_page[cpu];
 813         msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
 814
 815         /* Check if there are actual msgs to be processed */
 816         if (msg->header.message_type != HVMSG_NONE) {
 817                 if (msg->header.message_type == HVMSG_TIMER_EXPIRED)
 818                         hv_process_timer_expiration(msg, cpu);
 819                 else
 820                         tasklet_schedule(&msg_dpc);
 821         }
 822
 823         add_interrupt_randomness(HYPERVISOR_CALLBACK_VECTOR, 0);
 824 }
 825
 826
 827 /*
 828  * vmbus_bus_init -Main vmbus driver initialization routine.
 829  *
 830  * Here, we
 831  *      - initialize the vmbus driver context
 832  *      - invoke the vmbus hv main init routine
 833  *      - get the irq resource
 834  *      - retrieve the channel offers
 835  */
 836 static int vmbus_bus_init(int irq)
 837 {
 838         int ret;
 839
 840         /* Hypervisor initialization...setup hypercall page..etc */
 841         ret = hv_init();
 842         if (ret != 0) {
 843                 pr_err("Unable to initialize the hypervisor - 0x%x\n", ret);
 844                 return ret;
 845         }
 846
 847         tasklet_init(&msg_dpc, vmbus_on_msg_dpc, 0);
 848
 849         ret = bus_register(&hv_bus);
 850         if (ret)
 851                 goto err_cleanup;
 852
 853         hv_setup_vmbus_irq(vmbus_isr);
 854
 855         ret = hv_synic_alloc();
 856         if (ret)
 857                 goto err_alloc;
 858         /*
 859          * Initialize the per-cpu interrupt state and
 860          * connect to the host.
 861          */
 862         on_each_cpu(hv_synic_init, NULL, 1);
 863         ret = vmbus_connect();
 864         if (ret)
 865                 goto err_connect;
 866
 867         if (vmbus_proto_version > VERSION_WIN7)
 868                 cpu_hotplug_disable();
 869
 870         /*
 871          * Only register if the crash MSRs are available
 872          */
 873         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
 874                 register_die_notifier(&hyperv_die_block);
 875                 atomic_notifier_chain_register(&panic_notifier_list,
 876                                                &hyperv_panic_block);
 877         }
 878
 879         vmbus_request_offers();
 880
 881         return 0;
 882
 883 err_connect:
 884         on_each_cpu(hv_synic_cleanup, NULL, 1);
 885 err_alloc:
 886         hv_synic_free();
 887         hv_remove_vmbus_irq();
 888
 889         bus_unregister(&hv_bus);
 890
 891 err_cleanup:
 892         hv_cleanup();
 893
 894         return ret;
 895 }
 896
 897 /**
 898  * __vmbus_child_driver_register() - Register a vmbus's driver
 899  * @hv_driver: Pointer to driver structure you want to register
 900  * @owner: owner module of the drv
 901  * @mod_name: module name string
 902  *
 903  * Registers the given driver with Linux through the 'driver_register()' call
 904  * and sets up the hyper-v vmbus handling for this driver.
 905  * It will return the state of the 'driver_register()' call.
 906  *
 907  */
 908 int __vmbus_driver_register(struct hv_driver *hv_driver, struct module *owner, const char *mod_name)
 909 {
 910         int ret;
 911
 912         pr_info("registering driver %s\n", hv_driver->name);
 913
 914         ret = vmbus_exists();
 915         if (ret < 0)
 916                 return ret;
 917
 918         hv_driver->driver.name = hv_driver->name;
 919         hv_driver->driver.owner = owner;
 920         hv_driver->driver.mod_name = mod_name;
 921         hv_driver->driver.bus = &hv_bus;
 922
 923         ret = driver_register(&hv_driver->driver);
 924
 925         return ret;
 926 }
 927 EXPORT_SYMBOL_GPL(__vmbus_driver_register);
 928
 929 /**
 930  * vmbus_driver_unregister() - Unregister a vmbus's driver
 931  * @hv_driver: Pointer to driver structure you want to
 932  *             un-register
 933  *
 934  * Un-register the given driver that was previous registered with a call to
 935  * vmbus_driver_register()
 936  */
 937 void vmbus_driver_unregister(struct hv_driver *hv_driver)
 938 {
 939         pr_info("unregistering driver %s\n", hv_driver->name);
 940
 941         if (!vmbus_exists())
 942                 driver_unregister(&hv_driver->driver);
 943 }
 944 EXPORT_SYMBOL_GPL(vmbus_driver_unregister);
 945
 946 /*
 947  * vmbus_device_create - Creates and registers a new child device
 948  * on the vmbus.
 949  */
 950 struct hv_device *vmbus_device_create(const uuid_le *type,
 951                                       const uuid_le *instance,
 952                                       struct vmbus_channel *channel)
 953 {
 954         struct hv_device *child_device_obj;
 955
 956         child_device_obj = kzalloc(sizeof(struct hv_device), GFP_KERNEL);
 957         if (!child_device_obj) {
 958                 pr_err("Unable to allocate device object for child device\n");
 959                 return NULL;
 960         }
 961
 962         child_device_obj->channel = channel;
 963         memcpy(&child_device_obj->dev_type, type, sizeof(uuid_le));
 964         memcpy(&child_device_obj->dev_instance, instance,
 965                sizeof(uuid_le));
 966
 967
 968         return child_device_obj;
 969 }
 970
 971 /*
 972  * vmbus_device_register - Register the child device
 973  */
 974 int vmbus_device_register(struct hv_device *child_device_obj)
 975 {
 976         int ret = 0;
 977
 978         dev_set_name(&child_device_obj->device, "vmbus_%d",
 979                      child_device_obj->channel->id);
 980
 981         child_device_obj->device.bus = &hv_bus;
 982         child_device_obj->device.parent = &hv_acpi_dev->dev;
 983         child_device_obj->device.release = vmbus_device_release;
 984
 985         /*
 986          * Register with the LDM. This will kick off the driver/device
 987          * binding...which will eventually call vmbus_match() and vmbus_probe()
 988          */
 989         ret = device_register(&child_device_obj->device);
 990
 991         if (ret)
 992                 pr_err("Unable to register child device\n");
 993         else
 994                 pr_debug("child device %s registered\n",
 995                         dev_name(&child_device_obj->device));
 996
 997         return ret;
 998 }
 999
1000 /*
1001  * vmbus_device_unregister - Remove the specified child device
1002  * from the vmbus.
1003  */
1004 void vmbus_device_unregister(struct hv_device *device_obj)
1005 {
1006         pr_debug("child device %s unregistered\n",
1007                 dev_name(&device_obj->device));
1008
1009         /*
1010          * Kick off the process of unregistering the device.
1011          * This will call vmbus_remove() and eventually vmbus_device_release()
1012          */
1013         device_unregister(&device_obj->device);
1014 }
1015
1016
1017 /*
1018  * VMBUS is an acpi enumerated device. Get the information we
1019  * need from DSDT.
1020  */
1021 #define VTPM_BASE_ADDRESS 0xfed40000
1022 static acpi_status vmbus_walk_resources(struct acpi_resource *res, void *ctx)
1023 {
1024         resource_size_t start = 0;
1025         resource_size_t end = 0;
1026         struct resource *new_res;
1027         struct resource **old_res = &hyperv_mmio;
1028         struct resource **prev_res = NULL;
1029
1030         switch (res->type) {
1031         case ACPI_RESOURCE_TYPE_IRQ:
1032                 irq = res->data.irq.interrupts[0];
1033                 return AE_OK;
1034
1035         /*
1036          * "Address" descriptors are for bus windows. Ignore
1037          * "memory" descriptors, which are for registers on
1038          * devices.
1039          */
1040         case ACPI_RESOURCE_TYPE_ADDRESS32:
1041                 start = res->data.address32.address.minimum;
1042                 end = res->data.address32.address.maximum;
1043                 break;
1044
1045         case ACPI_RESOURCE_TYPE_ADDRESS64:
1046                 start = res->data.address64.address.minimum;
1047                 end = res->data.address64.address.maximum;
1048                 break;
1049
1050         default:
1051                 /* Unused resource type */
1052                 return AE_OK;
1053
1054         }
1055         /*
1056          * Ignore ranges that are below 1MB, as they're not
1057          * necessary or useful here.
1058          */
1059         if (end < 0x100000)
1060                 return AE_OK;
1061
1062         new_res = kzalloc(sizeof(*new_res), GFP_ATOMIC);
1063         if (!new_res)
1064                 return AE_NO_MEMORY;
1065
1066         /* If this range overlaps the virtual TPM, truncate it. */
1067         if (end > VTPM_BASE_ADDRESS && start < VTPM_BASE_ADDRESS)
1068                 end = VTPM_BASE_ADDRESS;
1069
1070         new_res->name = "hyperv mmio";
1071         new_res->flags = IORESOURCE_MEM;
1072         new_res->start = start;
1073         new_res->end = end;
1074
1075         do {
1076                 if (!*old_res) {
1077                         *old_res = new_res;
1078                         break;
1079                 }
1080
1081                 if ((*old_res)->end < new_res->start) {
1082                         new_res->sibling = *old_res;
1083                         if (prev_res)
1084                                 (*prev_res)->sibling = new_res;
1085                         *old_res = new_res;
1086                         break;
1087                 }
1088
1089                 prev_res = old_res;
1090                 old_res = &(*old_res)->sibling;
1091
1092         } while (1);
1093
1094         return AE_OK;
1095 }
1096
1097 static int vmbus_acpi_remove(struct acpi_device *device)
1098 {
1099         struct resource *cur_res;
1100         struct resource *next_res;
1101
1102         if (hyperv_mmio) {
1103                 for (cur_res = hyperv_mmio; cur_res; cur_res = next_res) {
1104                         next_res = cur_res->sibling;
1105                         kfree(cur_res);
1106                 }
1107         }
1108
1109         return 0;
1110 }
1111
1112 /**
1113  * vmbus_allocate_mmio() - Pick a memory-mapped I/O range.
1114  * @new:                If successful, supplied a pointer to the
1115  *                      allocated MMIO space.
1116  * @device_obj:         Identifies the caller
1117  * @min:                Minimum guest physical address of the
1118  *                      allocation
1119  * @max:                Maximum guest physical address
1120  * @size:               Size of the range to be allocated
1121  * @align:              Alignment of the range to be allocated
1122  * @fb_overlap_ok:      Whether this allocation can be allowed
1123  *                      to overlap the video frame buffer.
1124  *
1125  * This function walks the resources granted to VMBus by the
1126  * _CRS object in the ACPI namespace underneath the parent
1127  * "bridge" whether that's a root PCI bus in the Generation 1
1128  * case or a Module Device in the Generation 2 case.  It then
1129  * attempts to allocate from the global MMIO pool in a way that
1130  * matches the constraints supplied in these parameters and by
1131  * that _CRS.
1132  *
1133  * Return: 0 on success, -errno on failure
1134  */
1135 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
1136                         resource_size_t min, resource_size_t max,
1137                         resource_size_t size, resource_size_t align,
1138                         bool fb_overlap_ok)
1139 {
1140         struct resource *iter;
1141         resource_size_t range_min, range_max, start, local_min, local_max;
1142         const char *dev_n = dev_name(&device_obj->device);
1143         u32 fb_end = screen_info.lfb_base + (screen_info.lfb_size << 1);
1144         int i, retval;
1145
1146         retval = -ENXIO;
1147         down(&hyperv_mmio_lock);
1148
1149         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
1150                 if ((iter->start >= max) || (iter->end <= min))
1151                         continue;
1152
1153                 range_min = iter->start;
1154                 range_max = iter->end;
1155
1156                 /* If this range overlaps the frame buffer, split it into
1157                    two tries. */
1158                 for (i = 0; i < 2; i++) {
1159                         local_min = range_min;
1160                         local_max = range_max;
1161                         if (fb_overlap_ok || (range_min >= fb_end) ||
1162                             (range_max <= screen_info.lfb_base)) {
1163                                 i++;
1164                         } else {
1165                                 if ((range_min <= screen_info.lfb_base) &&
1166                                     (range_max >= screen_info.lfb_base)) {
1167                                         /*
1168                                          * The frame buffer is in this window,
1169                                          * so trim this into the part that
1170                                          * preceeds the frame buffer.
1171                                          */
1172                                         local_max = screen_info.lfb_base - 1;
1173                                         range_min = fb_end;
1174                                 } else {
1175                                         range_min = fb_end;
1176                                         continue;
1177                                 }
1178                         }
1179
1180                         start = (local_min + align - 1) & ~(align - 1);
1181                         for (; start + size - 1 <= local_max; start += align) {
1182                                 *new = request_mem_region_exclusive(start, size,
1183                                                                     dev_n);
1184                                 if (*new) {
1185                                         retval = 0;
1186                                         goto exit;
1187                                 }
1188                         }
1189                 }
1190         }
1191
1192 exit:
1193         up(&hyperv_mmio_lock);
1194         return retval;
1195 }
1196 EXPORT_SYMBOL_GPL(vmbus_allocate_mmio);
1197
1198 static int vmbus_acpi_add(struct acpi_device *device)
1199 {
1200         acpi_status result;
1201         int ret_val = -ENODEV;
1202         struct acpi_device *ancestor;
1203
1204         hv_acpi_dev = device;
1205
1206         result = acpi_walk_resources(device->handle, METHOD_NAME__CRS,
1207                                         vmbus_walk_resources, NULL);
1208
1209         if (ACPI_FAILURE(result))
1210                 goto acpi_walk_err;
1211         /*
1212          * Some ancestor of the vmbus acpi device (Gen1 or Gen2
1213          * firmware) is the VMOD that has the mmio ranges. Get that.
1214          */
1215         for (ancestor = device->parent; ancestor; ancestor = ancestor->parent) {
1216                 result = acpi_walk_resources(ancestor->handle, METHOD_NAME__CRS,
1217                                              vmbus_walk_resources, NULL);
1218
1219                 if (ACPI_FAILURE(result))
1220                         continue;
1221                 if (hyperv_mmio)
1222                         break;
1223         }
1224         ret_val = 0;
1225
1226 acpi_walk_err:
1227         complete(&probe_event);
1228         if (ret_val)
1229                 vmbus_acpi_remove(device);
1230         return ret_val;
1231 }
1232
1233 static const struct acpi_device_id vmbus_acpi_device_ids[] = {
1234         {"VMBUS", 0},
1235         {"VMBus", 0},
1236         {"", 0},
1237 };
1238 MODULE_DEVICE_TABLE(acpi, vmbus_acpi_device_ids);
1239
1240 static struct acpi_driver vmbus_acpi_driver = {
1241         .name = "vmbus",
1242         .ids = vmbus_acpi_device_ids,
1243         .ops = {
1244                 .add = vmbus_acpi_add,
1245                 .remove = vmbus_acpi_remove,
1246         },
1247 };
1248
1249 static void hv_kexec_handler(void)
1250 {
1251         int cpu;
1252
1253         hv_synic_clockevents_cleanup();
1254         vmbus_initiate_unload();
1255         for_each_online_cpu(cpu)
1256                 smp_call_function_single(cpu, hv_synic_cleanup, NULL, 1);
1257         hv_cleanup();
1258 };
1259
1260 static void hv_crash_handler(struct pt_regs *regs)
1261 {
1262         vmbus_initiate_unload();
1263         /*
1264          * In crash handler we can't schedule synic cleanup for all CPUs,
1265          * doing the cleanup for current CPU only. This should be sufficient
1266          * for kdump.
1267          */
1268         hv_synic_cleanup(NULL);
1269         hv_cleanup();
1270 };
1271
1272 static int __init hv_acpi_init(void)
1273 {
1274         int ret, t;
1275
1276         if (x86_hyper != &x86_hyper_ms_hyperv)
1277                 return -ENODEV;
1278
1279         init_completion(&probe_event);
1280
1281         /*
1282          * Get irq resources first.
1283          */
1284         ret = acpi_bus_register_driver(&vmbus_acpi_driver);
1285
1286         if (ret)
1287                 return ret;
1288
1289         t = wait_for_completion_timeout(&probe_event, 5*HZ);
1290         if (t == 0) {
1291                 ret = -ETIMEDOUT;
1292                 goto cleanup;
1293         }
1294
1295         if (irq <= 0) {
1296                 ret = -ENODEV;
1297                 goto cleanup;
1298         }
1299
1300         ret = vmbus_bus_init(irq);
1301         if (ret)
1302                 goto cleanup;
1303
1304         hv_setup_kexec_handler(hv_kexec_handler);
1305         hv_setup_crash_handler(hv_crash_handler);
1306
1307         return 0;
1308
1309 cleanup:
1310         acpi_bus_unregister_driver(&vmbus_acpi_driver);
1311         hv_acpi_dev = NULL;
1312         return ret;
1313 }
1314
1315 static void __exit vmbus_exit(void)
1316 {
1317         int cpu;
1318
1319         hv_remove_kexec_handler();
1320         hv_remove_crash_handler();
1321         vmbus_connection.conn_state = DISCONNECTED;
1322         hv_synic_clockevents_cleanup();
1323         vmbus_disconnect();
1324         hv_remove_vmbus_irq();
1325         tasklet_kill(&msg_dpc);
1326         vmbus_free_channels();
1327         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1328                 unregister_die_notifier(&hyperv_die_block);
1329                 atomic_notifier_chain_unregister(&panic_notifier_list,
1330                                                  &hyperv_panic_block);
1331         }
1332         bus_unregister(&hv_bus);
1333         hv_cleanup();
1334         for_each_online_cpu(cpu) {
1335                 tasklet_kill(hv_context.event_dpc[cpu]);
1336                 smp_call_function_single(cpu, hv_synic_cleanup, NULL, 1);
1337         }
1338         hv_synic_free();
1339         acpi_bus_unregister_driver(&vmbus_acpi_driver);
1340         if (vmbus_proto_version > VERSION_WIN7)
1341                 cpu_hotplug_enable();
1342 }
1343
1344
1345 MODULE_LICENSE("GPL");
1346
1347 subsys_initcall(hv_acpi_init);
1348 module_exit(vmbus_exit);