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 "hyperv_vmbus.h"
  45
  46 static struct acpi_device  *hv_acpi_dev;
  47
  48 static struct tasklet_struct msg_dpc;
  49 static struct completion probe_event;
  50 static int irq;
  51
  52
  53 static void hyperv_report_panic(struct pt_regs *regs)
  54 {
  55         static bool panic_reported;
  56
  57         /*
  58          * We prefer to report panic on 'die' chain as we have proper
  59          * registers to report, but if we miss it (e.g. on BUG()) we need
  60          * to report it on 'panic'.
  61          */
  62         if (panic_reported)
  63                 return;
  64         panic_reported = true;
  65
  66         wrmsrl(HV_X64_MSR_CRASH_P0, regs->ip);
  67         wrmsrl(HV_X64_MSR_CRASH_P1, regs->ax);
  68         wrmsrl(HV_X64_MSR_CRASH_P2, regs->bx);
  69         wrmsrl(HV_X64_MSR_CRASH_P3, regs->cx);
  70         wrmsrl(HV_X64_MSR_CRASH_P4, regs->dx);
  71
  72         /*
  73          * Let Hyper-V know there is crash data available
  74          */
  75         wrmsrl(HV_X64_MSR_CRASH_CTL, HV_CRASH_CTL_CRASH_NOTIFY);
  76 }
  77
  78 static int hyperv_panic_event(struct notifier_block *nb, unsigned long val,
  79                               void *args)
  80 {
  81         struct pt_regs *regs;
  82
  83         regs = current_pt_regs();
  84
  85         hyperv_report_panic(regs);
  86         return NOTIFY_DONE;
  87 }
  88
  89 static int hyperv_die_event(struct notifier_block *nb, unsigned long val,
  90                             void *args)
  91 {
  92         struct die_args *die = (struct die_args *)args;
  93         struct pt_regs *regs = die->regs;
  94
  95         hyperv_report_panic(regs);
  96         return NOTIFY_DONE;
  97 }
  98
  99 static struct notifier_block hyperv_die_block = {
 100         .notifier_call = hyperv_die_event,
 101 };
 102 static struct notifier_block hyperv_panic_block = {
 103         .notifier_call = hyperv_panic_event,
 104 };
 105
 106 struct resource *hyperv_mmio;
 107
 108 static int vmbus_exists(void)
 109 {
 110         if (hv_acpi_dev == NULL)
 111                 return -ENODEV;
 112
 113         return 0;
 114 }
 115
 116 #define VMBUS_ALIAS_LEN ((sizeof((struct hv_vmbus_device_id *)0)->guid) * 2)
 117 static void print_alias_name(struct hv_device *hv_dev, char *alias_name)
 118 {
 119         int i;
 120         for (i = 0; i < VMBUS_ALIAS_LEN; i += 2)
 121                 sprintf(&alias_name[i], "%02x", hv_dev->dev_type.b[i/2]);
 122 }
 123
 124 static u8 channel_monitor_group(struct vmbus_channel *channel)
 125 {
 126         return (u8)channel->offermsg.monitorid / 32;
 127 }
 128
 129 static u8 channel_monitor_offset(struct vmbus_channel *channel)
 130 {
 131         return (u8)channel->offermsg.monitorid % 32;
 132 }
 133
 134 static u32 channel_pending(struct vmbus_channel *channel,
 135                            struct hv_monitor_page *monitor_page)
 136 {
 137         u8 monitor_group = channel_monitor_group(channel);
 138         return monitor_page->trigger_group[monitor_group].pending;
 139 }
 140
 141 static u32 channel_latency(struct vmbus_channel *channel,
 142                            struct hv_monitor_page *monitor_page)
 143 {
 144         u8 monitor_group = channel_monitor_group(channel);
 145         u8 monitor_offset = channel_monitor_offset(channel);
 146         return monitor_page->latency[monitor_group][monitor_offset];
 147 }
 148
 149 static u32 channel_conn_id(struct vmbus_channel *channel,
 150                            struct hv_monitor_page *monitor_page)
 151 {
 152         u8 monitor_group = channel_monitor_group(channel);
 153         u8 monitor_offset = channel_monitor_offset(channel);
 154         return monitor_page->parameter[monitor_group][monitor_offset].connectionid.u.id;
 155 }
 156
 157 static ssize_t id_show(struct device *dev, struct device_attribute *dev_attr,
 158                        char *buf)
 159 {
 160         struct hv_device *hv_dev = device_to_hv_device(dev);
 161
 162         if (!hv_dev->channel)
 163                 return -ENODEV;
 164         return sprintf(buf, "%d\n", hv_dev->channel->offermsg.child_relid);
 165 }
 166 static DEVICE_ATTR_RO(id);
 167
 168 static ssize_t state_show(struct device *dev, struct device_attribute *dev_attr,
 169                           char *buf)
 170 {
 171         struct hv_device *hv_dev = device_to_hv_device(dev);
 172
 173         if (!hv_dev->channel)
 174                 return -ENODEV;
 175         return sprintf(buf, "%d\n", hv_dev->channel->state);
 176 }
 177 static DEVICE_ATTR_RO(state);
 178
 179 static ssize_t monitor_id_show(struct device *dev,
 180                                struct device_attribute *dev_attr, char *buf)
 181 {
 182         struct hv_device *hv_dev = device_to_hv_device(dev);
 183
 184         if (!hv_dev->channel)
 185                 return -ENODEV;
 186         return sprintf(buf, "%d\n", hv_dev->channel->offermsg.monitorid);
 187 }
 188 static DEVICE_ATTR_RO(monitor_id);
 189
 190 static ssize_t class_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_type.b);
 199 }
 200 static DEVICE_ATTR_RO(class_id);
 201
 202 static ssize_t device_id_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
 207         if (!hv_dev->channel)
 208                 return -ENODEV;
 209         return sprintf(buf, "{%pUl}\n",
 210                        hv_dev->channel->offermsg.offer.if_instance.b);
 211 }
 212 static DEVICE_ATTR_RO(device_id);
 213
 214 static ssize_t modalias_show(struct device *dev,
 215                              struct device_attribute *dev_attr, char *buf)
 216 {
 217         struct hv_device *hv_dev = device_to_hv_device(dev);
 218         char alias_name[VMBUS_ALIAS_LEN + 1];
 219
 220         print_alias_name(hv_dev, alias_name);
 221         return sprintf(buf, "vmbus:%s\n", alias_name);
 222 }
 223 static DEVICE_ATTR_RO(modalias);
 224
 225 static ssize_t server_monitor_pending_show(struct device *dev,
 226                                            struct device_attribute *dev_attr,
 227                                            char *buf)
 228 {
 229         struct hv_device *hv_dev = device_to_hv_device(dev);
 230
 231         if (!hv_dev->channel)
 232                 return -ENODEV;
 233         return sprintf(buf, "%d\n",
 234                        channel_pending(hv_dev->channel,
 235                                        vmbus_connection.monitor_pages[1]));
 236 }
 237 static DEVICE_ATTR_RO(server_monitor_pending);
 238
 239 static ssize_t client_monitor_pending_show(struct device *dev,
 240                                            struct device_attribute *dev_attr,
 241                                            char *buf)
 242 {
 243         struct hv_device *hv_dev = device_to_hv_device(dev);
 244
 245         if (!hv_dev->channel)
 246                 return -ENODEV;
 247         return sprintf(buf, "%d\n",
 248                        channel_pending(hv_dev->channel,
 249                                        vmbus_connection.monitor_pages[1]));
 250 }
 251 static DEVICE_ATTR_RO(client_monitor_pending);
 252
 253 static ssize_t server_monitor_latency_show(struct device *dev,
 254                                            struct device_attribute *dev_attr,
 255                                            char *buf)
 256 {
 257         struct hv_device *hv_dev = device_to_hv_device(dev);
 258
 259         if (!hv_dev->channel)
 260                 return -ENODEV;
 261         return sprintf(buf, "%d\n",
 262                        channel_latency(hv_dev->channel,
 263                                        vmbus_connection.monitor_pages[0]));
 264 }
 265 static DEVICE_ATTR_RO(server_monitor_latency);
 266
 267 static ssize_t client_monitor_latency_show(struct device *dev,
 268                                            struct device_attribute *dev_attr,
 269                                            char *buf)
 270 {
 271         struct hv_device *hv_dev = device_to_hv_device(dev);
 272
 273         if (!hv_dev->channel)
 274                 return -ENODEV;
 275         return sprintf(buf, "%d\n",
 276                        channel_latency(hv_dev->channel,
 277                                        vmbus_connection.monitor_pages[1]));
 278 }
 279 static DEVICE_ATTR_RO(client_monitor_latency);
 280
 281 static ssize_t server_monitor_conn_id_show(struct device *dev,
 282                                            struct device_attribute *dev_attr,
 283                                            char *buf)
 284 {
 285         struct hv_device *hv_dev = device_to_hv_device(dev);
 286
 287         if (!hv_dev->channel)
 288                 return -ENODEV;
 289         return sprintf(buf, "%d\n",
 290                        channel_conn_id(hv_dev->channel,
 291                                        vmbus_connection.monitor_pages[0]));
 292 }
 293 static DEVICE_ATTR_RO(server_monitor_conn_id);
 294
 295 static ssize_t client_monitor_conn_id_show(struct device *dev,
 296                                            struct device_attribute *dev_attr,
 297                                            char *buf)
 298 {
 299         struct hv_device *hv_dev = device_to_hv_device(dev);
 300
 301         if (!hv_dev->channel)
 302                 return -ENODEV;
 303         return sprintf(buf, "%d\n",
 304                        channel_conn_id(hv_dev->channel,
 305                                        vmbus_connection.monitor_pages[1]));
 306 }
 307 static DEVICE_ATTR_RO(client_monitor_conn_id);
 308
 309 static ssize_t out_intr_mask_show(struct device *dev,
 310                                   struct device_attribute *dev_attr, char *buf)
 311 {
 312         struct hv_device *hv_dev = device_to_hv_device(dev);
 313         struct hv_ring_buffer_debug_info outbound;
 314
 315         if (!hv_dev->channel)
 316                 return -ENODEV;
 317         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
 318         return sprintf(buf, "%d\n", outbound.current_interrupt_mask);
 319 }
 320 static DEVICE_ATTR_RO(out_intr_mask);
 321
 322 static ssize_t out_read_index_show(struct device *dev,
 323                                    struct device_attribute *dev_attr, char *buf)
 324 {
 325         struct hv_device *hv_dev = device_to_hv_device(dev);
 326         struct hv_ring_buffer_debug_info outbound;
 327
 328         if (!hv_dev->channel)
 329                 return -ENODEV;
 330         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
 331         return sprintf(buf, "%d\n", outbound.current_read_index);
 332 }
 333 static DEVICE_ATTR_RO(out_read_index);
 334
 335 static ssize_t out_write_index_show(struct device *dev,
 336                                     struct device_attribute *dev_attr,
 337                                     char *buf)
 338 {
 339         struct hv_device *hv_dev = device_to_hv_device(dev);
 340         struct hv_ring_buffer_debug_info outbound;
 341
 342         if (!hv_dev->channel)
 343                 return -ENODEV;
 344         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
 345         return sprintf(buf, "%d\n", outbound.current_write_index);
 346 }
 347 static DEVICE_ATTR_RO(out_write_index);
 348
 349 static ssize_t out_read_bytes_avail_show(struct device *dev,
 350                                          struct device_attribute *dev_attr,
 351                                          char *buf)
 352 {
 353         struct hv_device *hv_dev = device_to_hv_device(dev);
 354         struct hv_ring_buffer_debug_info outbound;
 355
 356         if (!hv_dev->channel)
 357                 return -ENODEV;
 358         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
 359         return sprintf(buf, "%d\n", outbound.bytes_avail_toread);
 360 }
 361 static DEVICE_ATTR_RO(out_read_bytes_avail);
 362
 363 static ssize_t out_write_bytes_avail_show(struct device *dev,
 364                                           struct device_attribute *dev_attr,
 365                                           char *buf)
 366 {
 367         struct hv_device *hv_dev = device_to_hv_device(dev);
 368         struct hv_ring_buffer_debug_info outbound;
 369
 370         if (!hv_dev->channel)
 371                 return -ENODEV;
 372         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
 373         return sprintf(buf, "%d\n", outbound.bytes_avail_towrite);
 374 }
 375 static DEVICE_ATTR_RO(out_write_bytes_avail);
 376
 377 static ssize_t in_intr_mask_show(struct device *dev,
 378                                  struct device_attribute *dev_attr, char *buf)
 379 {
 380         struct hv_device *hv_dev = device_to_hv_device(dev);
 381         struct hv_ring_buffer_debug_info inbound;
 382
 383         if (!hv_dev->channel)
 384                 return -ENODEV;
 385         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 386         return sprintf(buf, "%d\n", inbound.current_interrupt_mask);
 387 }
 388 static DEVICE_ATTR_RO(in_intr_mask);
 389
 390 static ssize_t in_read_index_show(struct device *dev,
 391                                   struct device_attribute *dev_attr, char *buf)
 392 {
 393         struct hv_device *hv_dev = device_to_hv_device(dev);
 394         struct hv_ring_buffer_debug_info inbound;
 395
 396         if (!hv_dev->channel)
 397                 return -ENODEV;
 398         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 399         return sprintf(buf, "%d\n", inbound.current_read_index);
 400 }
 401 static DEVICE_ATTR_RO(in_read_index);
 402
 403 static ssize_t in_write_index_show(struct device *dev,
 404                                    struct device_attribute *dev_attr, char *buf)
 405 {
 406         struct hv_device *hv_dev = device_to_hv_device(dev);
 407         struct hv_ring_buffer_debug_info inbound;
 408
 409         if (!hv_dev->channel)
 410                 return -ENODEV;
 411         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 412         return sprintf(buf, "%d\n", inbound.current_write_index);
 413 }
 414 static DEVICE_ATTR_RO(in_write_index);
 415
 416 static ssize_t in_read_bytes_avail_show(struct device *dev,
 417                                         struct device_attribute *dev_attr,
 418                                         char *buf)
 419 {
 420         struct hv_device *hv_dev = device_to_hv_device(dev);
 421         struct hv_ring_buffer_debug_info inbound;
 422
 423         if (!hv_dev->channel)
 424                 return -ENODEV;
 425         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 426         return sprintf(buf, "%d\n", inbound.bytes_avail_toread);
 427 }
 428 static DEVICE_ATTR_RO(in_read_bytes_avail);
 429
 430 static ssize_t in_write_bytes_avail_show(struct device *dev,
 431                                          struct device_attribute *dev_attr,
 432                                          char *buf)
 433 {
 434         struct hv_device *hv_dev = device_to_hv_device(dev);
 435         struct hv_ring_buffer_debug_info inbound;
 436
 437         if (!hv_dev->channel)
 438                 return -ENODEV;
 439         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 440         return sprintf(buf, "%d\n", inbound.bytes_avail_towrite);
 441 }
 442 static DEVICE_ATTR_RO(in_write_bytes_avail);
 443
 444 static ssize_t channel_vp_mapping_show(struct device *dev,
 445                                        struct device_attribute *dev_attr,
 446                                        char *buf)
 447 {
 448         struct hv_device *hv_dev = device_to_hv_device(dev);
 449         struct vmbus_channel *channel = hv_dev->channel, *cur_sc;
 450         unsigned long flags;
 451         int buf_size = PAGE_SIZE, n_written, tot_written;
 452         struct list_head *cur;
 453
 454         if (!channel)
 455                 return -ENODEV;
 456
 457         tot_written = snprintf(buf, buf_size, "%u:%u\n",
 458                 channel->offermsg.child_relid, channel->target_cpu);
 459
 460         spin_lock_irqsave(&channel->lock, flags);
 461
 462         list_for_each(cur, &channel->sc_list) {
 463                 if (tot_written >= buf_size - 1)
 464                         break;
 465
 466                 cur_sc = list_entry(cur, struct vmbus_channel, sc_list);
 467                 n_written = scnprintf(buf + tot_written,
 468                                      buf_size - tot_written,
 469                                      "%u:%u\n",
 470                                      cur_sc->offermsg.child_relid,
 471                                      cur_sc->target_cpu);
 472                 tot_written += n_written;
 473         }
 474
 475         spin_unlock_irqrestore(&channel->lock, flags);
 476
 477         return tot_written;
 478 }
 479 static DEVICE_ATTR_RO(channel_vp_mapping);
 480
 481 /* Set up per device attributes in /sys/bus/vmbus/devices/<bus device> */
 482 static struct attribute *vmbus_attrs[] = {
 483         &dev_attr_id.attr,
 484         &dev_attr_state.attr,
 485         &dev_attr_monitor_id.attr,
 486         &dev_attr_class_id.attr,
 487         &dev_attr_device_id.attr,
 488         &dev_attr_modalias.attr,
 489         &dev_attr_server_monitor_pending.attr,
 490         &dev_attr_client_monitor_pending.attr,
 491         &dev_attr_server_monitor_latency.attr,
 492         &dev_attr_client_monitor_latency.attr,
 493         &dev_attr_server_monitor_conn_id.attr,
 494         &dev_attr_client_monitor_conn_id.attr,
 495         &dev_attr_out_intr_mask.attr,
 496         &dev_attr_out_read_index.attr,
 497         &dev_attr_out_write_index.attr,
 498         &dev_attr_out_read_bytes_avail.attr,
 499         &dev_attr_out_write_bytes_avail.attr,
 500         &dev_attr_in_intr_mask.attr,
 501         &dev_attr_in_read_index.attr,
 502         &dev_attr_in_write_index.attr,
 503         &dev_attr_in_read_bytes_avail.attr,
 504         &dev_attr_in_write_bytes_avail.attr,
 505         &dev_attr_channel_vp_mapping.attr,
 506         NULL,
 507 };
 508 ATTRIBUTE_GROUPS(vmbus);
 509
 510 /*
 511  * vmbus_uevent - add uevent for our device
 512  *
 513  * This routine is invoked when a device is added or removed on the vmbus to
 514  * generate a uevent to udev in the userspace. The udev will then look at its
 515  * rule and the uevent generated here to load the appropriate driver
 516  *
 517  * The alias string will be of the form vmbus:guid where guid is the string
 518  * representation of the device guid (each byte of the guid will be
 519  * represented with two hex characters.
 520  */
 521 static int vmbus_uevent(struct device *device, struct kobj_uevent_env *env)
 522 {
 523         struct hv_device *dev = device_to_hv_device(device);
 524         int ret;
 525         char alias_name[VMBUS_ALIAS_LEN + 1];
 526
 527         print_alias_name(dev, alias_name);
 528         ret = add_uevent_var(env, "MODALIAS=vmbus:%s", alias_name);
 529         return ret;
 530 }
 531
 532 static const uuid_le null_guid;
 533
 534 static inline bool is_null_guid(const __u8 *guid)
 535 {
 536         if (memcmp(guid, &null_guid, sizeof(uuid_le)))
 537                 return false;
 538         return true;
 539 }
 540
 541 /*
 542  * Return a matching hv_vmbus_device_id pointer.
 543  * If there is no match, return NULL.
 544  */
 545 static const struct hv_vmbus_device_id *hv_vmbus_get_id(
 546                                         const struct hv_vmbus_device_id *id,
 547                                         const __u8 *guid)
 548 {
 549         for (; !is_null_guid(id->guid); id++)
 550                 if (!memcmp(&id->guid, guid, sizeof(uuid_le)))
 551                         return id;
 552
 553         return NULL;
 554 }
 555
 556
 557
 558 /*
 559  * vmbus_match - Attempt to match the specified device to the specified driver
 560  */
 561 static int vmbus_match(struct device *device, struct device_driver *driver)
 562 {
 563         struct hv_driver *drv = drv_to_hv_drv(driver);
 564         struct hv_device *hv_dev = device_to_hv_device(device);
 565
 566         if (hv_vmbus_get_id(drv->id_table, hv_dev->dev_type.b))
 567                 return 1;
 568
 569         return 0;
 570 }
 571
 572 /*
 573  * vmbus_probe - Add the new vmbus's child device
 574  */
 575 static int vmbus_probe(struct device *child_device)
 576 {
 577         int ret = 0;
 578         struct hv_driver *drv =
 579                         drv_to_hv_drv(child_device->driver);
 580         struct hv_device *dev = device_to_hv_device(child_device);
 581         const struct hv_vmbus_device_id *dev_id;
 582
 583         dev_id = hv_vmbus_get_id(drv->id_table, dev->dev_type.b);
 584         if (drv->probe) {
 585                 ret = drv->probe(dev, dev_id);
 586                 if (ret != 0)
 587                         pr_err("probe failed for device %s (%d)\n",
 588                                dev_name(child_device), ret);
 589
 590         } else {
 591                 pr_err("probe not set for driver %s\n",
 592                        dev_name(child_device));
 593                 ret = -ENODEV;
 594         }
 595         return ret;
 596 }
 597
 598 /*
 599  * vmbus_remove - Remove a vmbus device
 600  */
 601 static int vmbus_remove(struct device *child_device)
 602 {
 603         struct hv_driver *drv;
 604         struct hv_device *dev = device_to_hv_device(child_device);
 605         u32 relid = dev->channel->offermsg.child_relid;
 606
 607         if (child_device->driver) {
 608                 drv = drv_to_hv_drv(child_device->driver);
 609                 if (drv->remove)
 610                         drv->remove(dev);
 611                 else {
 612                         hv_process_channel_removal(dev->channel, relid);
 613                         pr_err("remove not set for driver %s\n",
 614                                 dev_name(child_device));
 615                 }
 616         } else {
 617                 /*
 618                  * We don't have a driver for this device; deal with the
 619                  * rescind message by removing the channel.
 620                  */
 621                 hv_process_channel_removal(dev->channel, relid);
 622         }
 623
 624         return 0;
 625 }
 626
 627
 628 /*
 629  * vmbus_shutdown - Shutdown a vmbus device
 630  */
 631 static void vmbus_shutdown(struct device *child_device)
 632 {
 633         struct hv_driver *drv;
 634         struct hv_device *dev = device_to_hv_device(child_device);
 635
 636
 637         /* The device may not be attached yet */
 638         if (!child_device->driver)
 639                 return;
 640
 641         drv = drv_to_hv_drv(child_device->driver);
 642
 643         if (drv->shutdown)
 644                 drv->shutdown(dev);
 645
 646         return;
 647 }
 648
 649
 650 /*
 651  * vmbus_device_release - Final callback release of the vmbus child device
 652  */
 653 static void vmbus_device_release(struct device *device)
 654 {
 655         struct hv_device *hv_dev = device_to_hv_device(device);
 656
 657         kfree(hv_dev);
 658
 659 }
 660
 661 /* The one and only one */
 662 static struct bus_type  hv_bus = {
 663         .name =         "vmbus",
 664         .match =                vmbus_match,
 665         .shutdown =             vmbus_shutdown,
 666         .remove =               vmbus_remove,
 667         .probe =                vmbus_probe,
 668         .uevent =               vmbus_uevent,
 669         .dev_groups =           vmbus_groups,
 670 };
 671
 672 struct onmessage_work_context {
 673         struct work_struct work;
 674         struct hv_message msg;
 675 };
 676
 677 static void vmbus_onmessage_work(struct work_struct *work)
 678 {
 679         struct onmessage_work_context *ctx;
 680
 681         /* Do not process messages if we're in DISCONNECTED state */
 682         if (vmbus_connection.conn_state == DISCONNECTED)
 683                 return;
 684
 685         ctx = container_of(work, struct onmessage_work_context,
 686                            work);
 687         vmbus_onmessage(&ctx->msg);
 688         kfree(ctx);
 689 }
 690
 691 static void hv_process_timer_expiration(struct hv_message *msg, int cpu)
 692 {
 693         struct clock_event_device *dev = hv_context.clk_evt[cpu];
 694
 695         if (dev->event_handler)
 696                 dev->event_handler(dev);
 697
 698         msg->header.message_type = HVMSG_NONE;
 699
 700         /*
 701          * Make sure the write to MessageType (ie set to
 702          * HVMSG_NONE) happens before we read the
 703          * MessagePending and EOMing. Otherwise, the EOMing
 704          * will not deliver any more messages since there is
 705          * no empty slot
 706          */
 707         mb();
 708
 709         if (msg->header.message_flags.msg_pending) {
 710                 /*
 711                  * This will cause message queue rescan to
 712                  * possibly deliver another msg from the
 713                  * hypervisor
 714                  */
 715                 wrmsrl(HV_X64_MSR_EOM, 0);
 716         }
 717 }
 718
 719 static void vmbus_on_msg_dpc(unsigned long data)
 720 {
 721         int cpu = smp_processor_id();
 722         void *page_addr = hv_context.synic_message_page[cpu];
 723         struct hv_message *msg = (struct hv_message *)page_addr +
 724                                   VMBUS_MESSAGE_SINT;
 725         struct vmbus_channel_message_header *hdr;
 726         struct vmbus_channel_message_table_entry *entry;
 727         struct onmessage_work_context *ctx;
 728
 729         while (1) {
 730                 if (msg->header.message_type == HVMSG_NONE)
 731                         /* no msg */
 732                         break;
 733
 734                 hdr = (struct vmbus_channel_message_header *)msg->u.payload;
 735
 736                 if (hdr->msgtype >= CHANNELMSG_COUNT) {
 737                         WARN_ONCE(1, "unknown msgtype=%d\n", hdr->msgtype);
 738                         goto msg_handled;
 739                 }
 740
 741                 entry = &channel_message_table[hdr->msgtype];
 742                 if (entry->handler_type == VMHT_BLOCKING) {
 743                         ctx = kmalloc(sizeof(*ctx), GFP_ATOMIC);
 744                         if (ctx == NULL)
 745                                 continue;
 746
 747                         INIT_WORK(&ctx->work, vmbus_onmessage_work);
 748                         memcpy(&ctx->msg, msg, sizeof(*msg));
 749
 750                         queue_work(vmbus_connection.work_queue, &ctx->work);
 751                 } else
 752                         entry->message_handler(hdr);
 753
 754 msg_handled:
 755                 msg->header.message_type = HVMSG_NONE;
 756
 757                 /*
 758                  * Make sure the write to MessageType (ie set to
 759                  * HVMSG_NONE) happens before we read the
 760                  * MessagePending and EOMing. Otherwise, the EOMing
 761                  * will not deliver any more messages since there is
 762                  * no empty slot
 763                  */
 764                 mb();
 765
 766                 if (msg->header.message_flags.msg_pending) {
 767                         /*
 768                          * This will cause message queue rescan to
 769                          * possibly deliver another msg from the
 770                          * hypervisor
 771                          */
 772                         wrmsrl(HV_X64_MSR_EOM, 0);
 773                 }
 774         }
 775 }
 776
 777 static void vmbus_isr(void)
 778 {
 779         int cpu = smp_processor_id();
 780         void *page_addr;
 781         struct hv_message *msg;
 782         union hv_synic_event_flags *event;
 783         bool handled = false;
 784
 785         page_addr = hv_context.synic_event_page[cpu];
 786         if (page_addr == NULL)
 787                 return;
 788
 789         event = (union hv_synic_event_flags *)page_addr +
 790                                          VMBUS_MESSAGE_SINT;
 791         /*
 792          * Check for events before checking for messages. This is the order
 793          * in which events and messages are checked in Windows guests on
 794          * Hyper-V, and the Windows team suggested we do the same.
 795          */
 796
 797         if ((vmbus_proto_version == VERSION_WS2008) ||
 798                 (vmbus_proto_version == VERSION_WIN7)) {
 799
 800                 /* Since we are a child, we only need to check bit 0 */
 801                 if (sync_test_and_clear_bit(0,
 802                         (unsigned long *) &event->flags32[0])) {
 803                         handled = true;
 804                 }
 805         } else {
 806                 /*
 807                  * Our host is win8 or above. The signaling mechanism
 808                  * has changed and we can directly look at the event page.
 809                  * If bit n is set then we have an interrup on the channel
 810                  * whose id is n.
 811                  */
 812                 handled = true;
 813         }
 814
 815         if (handled)
 816                 tasklet_schedule(hv_context.event_dpc[cpu]);
 817
 818
 819         page_addr = hv_context.synic_message_page[cpu];
 820         msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
 821
 822         /* Check if there are actual msgs to be processed */
 823         if (msg->header.message_type != HVMSG_NONE) {
 824                 if (msg->header.message_type == HVMSG_TIMER_EXPIRED)
 825                         hv_process_timer_expiration(msg, cpu);
 826                 else
 827                         tasklet_schedule(&msg_dpc);
 828         }
 829 }
 830
 831
 832 /*
 833  * vmbus_bus_init -Main vmbus driver initialization routine.
 834  *
 835  * Here, we
 836  *      - initialize the vmbus driver context
 837  *      - invoke the vmbus hv main init routine
 838  *      - get the irq resource
 839  *      - retrieve the channel offers
 840  */
 841 static int vmbus_bus_init(int irq)
 842 {
 843         int ret;
 844
 845         /* Hypervisor initialization...setup hypercall page..etc */
 846         ret = hv_init();
 847         if (ret != 0) {
 848                 pr_err("Unable to initialize the hypervisor - 0x%x\n", ret);
 849                 return ret;
 850         }
 851
 852         tasklet_init(&msg_dpc, vmbus_on_msg_dpc, 0);
 853
 854         ret = bus_register(&hv_bus);
 855         if (ret)
 856                 goto err_cleanup;
 857
 858         hv_setup_vmbus_irq(vmbus_isr);
 859
 860         ret = hv_synic_alloc();
 861         if (ret)
 862                 goto err_alloc;
 863         /*
 864          * Initialize the per-cpu interrupt state and
 865          * connect to the host.
 866          */
 867         on_each_cpu(hv_synic_init, NULL, 1);
 868         ret = vmbus_connect();
 869         if (ret)
 870                 goto err_connect;
 871
 872         if (vmbus_proto_version > VERSION_WIN7)
 873                 cpu_hotplug_disable();
 874
 875         /*
 876          * Only register if the crash MSRs are available
 877          */
 878         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
 879                 register_die_notifier(&hyperv_die_block);
 880                 atomic_notifier_chain_register(&panic_notifier_list,
 881                                                &hyperv_panic_block);
 882         }
 883
 884         vmbus_request_offers();
 885
 886         return 0;
 887
 888 err_connect:
 889         on_each_cpu(hv_synic_cleanup, NULL, 1);
 890 err_alloc:
 891         hv_synic_free();
 892         hv_remove_vmbus_irq();
 893
 894         bus_unregister(&hv_bus);
 895
 896 err_cleanup:
 897         hv_cleanup();
 898
 899         return ret;
 900 }
 901
 902 /**
 903  * __vmbus_child_driver_register() - Register a vmbus's driver
 904  * @hv_driver: Pointer to driver structure you want to register
 905  * @owner: owner module of the drv
 906  * @mod_name: module name string
 907  *
 908  * Registers the given driver with Linux through the 'driver_register()' call
 909  * and sets up the hyper-v vmbus handling for this driver.
 910  * It will return the state of the 'driver_register()' call.
 911  *
 912  */
 913 int __vmbus_driver_register(struct hv_driver *hv_driver, struct module *owner, const char *mod_name)
 914 {
 915         int ret;
 916
 917         pr_info("registering driver %s\n", hv_driver->name);
 918
 919         ret = vmbus_exists();
 920         if (ret < 0)
 921                 return ret;
 922
 923         hv_driver->driver.name = hv_driver->name;
 924         hv_driver->driver.owner = owner;
 925         hv_driver->driver.mod_name = mod_name;
 926         hv_driver->driver.bus = &hv_bus;
 927
 928         ret = driver_register(&hv_driver->driver);
 929
 930         return ret;
 931 }
 932 EXPORT_SYMBOL_GPL(__vmbus_driver_register);
 933
 934 /**
 935  * vmbus_driver_unregister() - Unregister a vmbus's driver
 936  * @hv_driver: Pointer to driver structure you want to
 937  *             un-register
 938  *
 939  * Un-register the given driver that was previous registered with a call to
 940  * vmbus_driver_register()
 941  */
 942 void vmbus_driver_unregister(struct hv_driver *hv_driver)
 943 {
 944         pr_info("unregistering driver %s\n", hv_driver->name);
 945
 946         if (!vmbus_exists())
 947                 driver_unregister(&hv_driver->driver);
 948 }
 949 EXPORT_SYMBOL_GPL(vmbus_driver_unregister);
 950
 951 /*
 952  * vmbus_device_create - Creates and registers a new child device
 953  * on the vmbus.
 954  */
 955 struct hv_device *vmbus_device_create(const uuid_le *type,
 956                                       const uuid_le *instance,
 957                                       struct vmbus_channel *channel)
 958 {
 959         struct hv_device *child_device_obj;
 960
 961         child_device_obj = kzalloc(sizeof(struct hv_device), GFP_KERNEL);
 962         if (!child_device_obj) {
 963                 pr_err("Unable to allocate device object for child device\n");
 964                 return NULL;
 965         }
 966
 967         child_device_obj->channel = channel;
 968         memcpy(&child_device_obj->dev_type, type, sizeof(uuid_le));
 969         memcpy(&child_device_obj->dev_instance, instance,
 970                sizeof(uuid_le));
 971
 972
 973         return child_device_obj;
 974 }
 975
 976 /*
 977  * vmbus_device_register - Register the child device
 978  */
 979 int vmbus_device_register(struct hv_device *child_device_obj)
 980 {
 981         int ret = 0;
 982
 983         dev_set_name(&child_device_obj->device, "vmbus_%d",
 984                      child_device_obj->channel->id);
 985
 986         child_device_obj->device.bus = &hv_bus;
 987         child_device_obj->device.parent = &hv_acpi_dev->dev;
 988         child_device_obj->device.release = vmbus_device_release;
 989
 990         /*
 991          * Register with the LDM. This will kick off the driver/device
 992          * binding...which will eventually call vmbus_match() and vmbus_probe()
 993          */
 994         ret = device_register(&child_device_obj->device);
 995
 996         if (ret)
 997                 pr_err("Unable to register child device\n");
 998         else
 999                 pr_debug("child device %s registered\n",
1000                         dev_name(&child_device_obj->device));
1001
1002         return ret;
1003 }
1004
1005 /*
1006  * vmbus_device_unregister - Remove the specified child device
1007  * from the vmbus.
1008  */
1009 void vmbus_device_unregister(struct hv_device *device_obj)
1010 {
1011         pr_debug("child device %s unregistered\n",
1012                 dev_name(&device_obj->device));
1013
1014         /*
1015          * Kick off the process of unregistering the device.
1016          * This will call vmbus_remove() and eventually vmbus_device_release()
1017          */
1018         device_unregister(&device_obj->device);
1019 }
1020
1021
1022 /*
1023  * VMBUS is an acpi enumerated device. Get the information we
1024  * need from DSDT.
1025  */
1026 #define VTPM_BASE_ADDRESS 0xfed40000
1027 static acpi_status vmbus_walk_resources(struct acpi_resource *res, void *ctx)
1028 {
1029         resource_size_t start = 0;
1030         resource_size_t end = 0;
1031         struct resource *new_res;
1032         struct resource **old_res = &hyperv_mmio;
1033         struct resource **prev_res = NULL;
1034
1035         switch (res->type) {
1036         case ACPI_RESOURCE_TYPE_IRQ:
1037                 irq = res->data.irq.interrupts[0];
1038                 return AE_OK;
1039
1040         /*
1041          * "Address" descriptors are for bus windows. Ignore
1042          * "memory" descriptors, which are for registers on
1043          * devices.
1044          */
1045         case ACPI_RESOURCE_TYPE_ADDRESS32:
1046                 start = res->data.address32.address.minimum;
1047                 end = res->data.address32.address.maximum;
1048                 break;
1049
1050         case ACPI_RESOURCE_TYPE_ADDRESS64:
1051                 start = res->data.address64.address.minimum;
1052                 end = res->data.address64.address.maximum;
1053                 break;
1054
1055         default:
1056                 /* Unused resource type */
1057                 return AE_OK;
1058
1059         }
1060         /*
1061          * Ignore ranges that are below 1MB, as they're not
1062          * necessary or useful here.
1063          */
1064         if (end < 0x100000)
1065                 return AE_OK;
1066
1067         new_res = kzalloc(sizeof(*new_res), GFP_ATOMIC);
1068         if (!new_res)
1069                 return AE_NO_MEMORY;
1070
1071         /* If this range overlaps the virtual TPM, truncate it. */
1072         if (end > VTPM_BASE_ADDRESS && start < VTPM_BASE_ADDRESS)
1073                 end = VTPM_BASE_ADDRESS;
1074
1075         new_res->name = "hyperv mmio";
1076         new_res->flags = IORESOURCE_MEM;
1077         new_res->start = start;
1078         new_res->end = end;
1079
1080         do {
1081                 if (!*old_res) {
1082                         *old_res = new_res;
1083                         break;
1084                 }
1085
1086                 if ((*old_res)->end < new_res->start) {
1087                         new_res->sibling = *old_res;
1088                         if (prev_res)
1089                                 (*prev_res)->sibling = new_res;
1090                         *old_res = new_res;
1091                         break;
1092                 }
1093
1094                 prev_res = old_res;
1095                 old_res = &(*old_res)->sibling;
1096
1097         } while (1);
1098
1099         return AE_OK;
1100 }
1101
1102 static int vmbus_acpi_remove(struct acpi_device *device)
1103 {
1104         struct resource *cur_res;
1105         struct resource *next_res;
1106
1107         if (hyperv_mmio) {
1108                 for (cur_res = hyperv_mmio; cur_res; cur_res = next_res) {
1109                         next_res = cur_res->sibling;
1110                         kfree(cur_res);
1111                 }
1112         }
1113
1114         return 0;
1115 }
1116
1117 /**
1118  * vmbus_allocate_mmio() - Pick a memory-mapped I/O range.
1119  * @new:                If successful, supplied a pointer to the
1120  *                      allocated MMIO space.
1121  * @device_obj:         Identifies the caller
1122  * @min:                Minimum guest physical address of the
1123  *                      allocation
1124  * @max:                Maximum guest physical address
1125  * @size:               Size of the range to be allocated
1126  * @align:              Alignment of the range to be allocated
1127  * @fb_overlap_ok:      Whether this allocation can be allowed
1128  *                      to overlap the video frame buffer.
1129  *
1130  * This function walks the resources granted to VMBus by the
1131  * _CRS object in the ACPI namespace underneath the parent
1132  * "bridge" whether that's a root PCI bus in the Generation 1
1133  * case or a Module Device in the Generation 2 case.  It then
1134  * attempts to allocate from the global MMIO pool in a way that
1135  * matches the constraints supplied in these parameters and by
1136  * that _CRS.
1137  *
1138  * Return: 0 on success, -errno on failure
1139  */
1140 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
1141                         resource_size_t min, resource_size_t max,
1142                         resource_size_t size, resource_size_t align,
1143                         bool fb_overlap_ok)
1144 {
1145         struct resource *iter;
1146         resource_size_t range_min, range_max, start, local_min, local_max;
1147         const char *dev_n = dev_name(&device_obj->device);
1148         u32 fb_end = screen_info.lfb_base + (screen_info.lfb_size << 1);
1149         int i;
1150
1151         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
1152                 if ((iter->start >= max) || (iter->end <= min))
1153                         continue;
1154
1155                 range_min = iter->start;
1156                 range_max = iter->end;
1157
1158                 /* If this range overlaps the frame buffer, split it into
1159                    two tries. */
1160                 for (i = 0; i < 2; i++) {
1161                         local_min = range_min;
1162                         local_max = range_max;
1163                         if (fb_overlap_ok || (range_min >= fb_end) ||
1164                             (range_max <= screen_info.lfb_base)) {
1165                                 i++;
1166                         } else {
1167                                 if ((range_min <= screen_info.lfb_base) &&
1168                                     (range_max >= screen_info.lfb_base)) {
1169                                         /*
1170                                          * The frame buffer is in this window,
1171                                          * so trim this into the part that
1172                                          * preceeds the frame buffer.
1173                                          */
1174                                         local_max = screen_info.lfb_base - 1;
1175                                         range_min = fb_end;
1176                                 } else {
1177                                         range_min = fb_end;
1178                                         continue;
1179                                 }
1180                         }
1181
1182                         start = (local_min + align - 1) & ~(align - 1);
1183                         for (; start + size - 1 <= local_max; start += align) {
1184                                 *new = request_mem_region_exclusive(start, size,
1185                                                                     dev_n);
1186                                 if (*new)
1187                                         return 0;
1188                         }
1189                 }
1190         }
1191
1192         return -ENXIO;
1193 }
1194 EXPORT_SYMBOL_GPL(vmbus_allocate_mmio);
1195
1196 static int vmbus_acpi_add(struct acpi_device *device)
1197 {
1198         acpi_status result;
1199         int ret_val = -ENODEV;
1200         struct acpi_device *ancestor;
1201
1202         hv_acpi_dev = device;
1203
1204         result = acpi_walk_resources(device->handle, METHOD_NAME__CRS,
1205                                         vmbus_walk_resources, NULL);
1206
1207         if (ACPI_FAILURE(result))
1208                 goto acpi_walk_err;
1209         /*
1210          * Some ancestor of the vmbus acpi device (Gen1 or Gen2
1211          * firmware) is the VMOD that has the mmio ranges. Get that.
1212          */
1213         for (ancestor = device->parent; ancestor; ancestor = ancestor->parent) {
1214                 result = acpi_walk_resources(ancestor->handle, METHOD_NAME__CRS,
1215                                              vmbus_walk_resources, NULL);
1216
1217                 if (ACPI_FAILURE(result))
1218                         continue;
1219                 if (hyperv_mmio)
1220                         break;
1221         }
1222         ret_val = 0;
1223
1224 acpi_walk_err:
1225         complete(&probe_event);
1226         if (ret_val)
1227                 vmbus_acpi_remove(device);
1228         return ret_val;
1229 }
1230
1231 static const struct acpi_device_id vmbus_acpi_device_ids[] = {
1232         {"VMBUS", 0},
1233         {"VMBus", 0},
1234         {"", 0},
1235 };
1236 MODULE_DEVICE_TABLE(acpi, vmbus_acpi_device_ids);
1237
1238 static struct acpi_driver vmbus_acpi_driver = {
1239         .name = "vmbus",
1240         .ids = vmbus_acpi_device_ids,
1241         .ops = {
1242                 .add = vmbus_acpi_add,
1243                 .remove = vmbus_acpi_remove,
1244         },
1245 };
1246
1247 static void hv_kexec_handler(void)
1248 {
1249         int cpu;
1250
1251         hv_synic_clockevents_cleanup();
1252         vmbus_initiate_unload();
1253         for_each_online_cpu(cpu)
1254                 smp_call_function_single(cpu, hv_synic_cleanup, NULL, 1);
1255         hv_cleanup();
1256 };
1257
1258 static void hv_crash_handler(struct pt_regs *regs)
1259 {
1260         vmbus_initiate_unload();
1261         /*
1262          * In crash handler we can't schedule synic cleanup for all CPUs,
1263          * doing the cleanup for current CPU only. This should be sufficient
1264          * for kdump.
1265          */
1266         hv_synic_cleanup(NULL);
1267         hv_cleanup();
1268 };
1269
1270 static int __init hv_acpi_init(void)
1271 {
1272         int ret, t;
1273
1274         if (x86_hyper != &x86_hyper_ms_hyperv)
1275                 return -ENODEV;
1276
1277         init_completion(&probe_event);
1278
1279         /*
1280          * Get irq resources first.
1281          */
1282         ret = acpi_bus_register_driver(&vmbus_acpi_driver);
1283
1284         if (ret)
1285                 return ret;
1286
1287         t = wait_for_completion_timeout(&probe_event, 5*HZ);
1288         if (t == 0) {
1289                 ret = -ETIMEDOUT;
1290                 goto cleanup;
1291         }
1292
1293         if (irq <= 0) {
1294                 ret = -ENODEV;
1295                 goto cleanup;
1296         }
1297
1298         ret = vmbus_bus_init(irq);
1299         if (ret)
1300                 goto cleanup;
1301
1302         hv_setup_kexec_handler(hv_kexec_handler);
1303         hv_setup_crash_handler(hv_crash_handler);
1304
1305         return 0;
1306
1307 cleanup:
1308         acpi_bus_unregister_driver(&vmbus_acpi_driver);
1309         hv_acpi_dev = NULL;
1310         return ret;
1311 }
1312
1313 static void __exit vmbus_exit(void)
1314 {
1315         int cpu;
1316
1317         hv_remove_kexec_handler();
1318         hv_remove_crash_handler();
1319         vmbus_connection.conn_state = DISCONNECTED;
1320         hv_synic_clockevents_cleanup();
1321         vmbus_disconnect();
1322         hv_remove_vmbus_irq();
1323         tasklet_kill(&msg_dpc);
1324         vmbus_free_channels();
1325         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1326                 unregister_die_notifier(&hyperv_die_block);
1327                 atomic_notifier_chain_unregister(&panic_notifier_list,
1328                                                  &hyperv_panic_block);
1329         }
1330         bus_unregister(&hv_bus);
1331         hv_cleanup();
1332         for_each_online_cpu(cpu) {
1333                 tasklet_kill(hv_context.event_dpc[cpu]);
1334                 smp_call_function_single(cpu, hv_synic_cleanup, NULL, 1);
1335         }
1336         hv_synic_free();
1337         acpi_bus_unregister_driver(&vmbus_acpi_driver);
1338         if (vmbus_proto_version > VERSION_WIN7)
1339                 cpu_hotplug_enable();
1340 }
1341
1342
1343 MODULE_LICENSE("GPL");
1344
1345 subsys_initcall(hv_acpi_init);
1346 module_exit(vmbus_exit);