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
 109 static int vmbus_exists(void)
 110 {
 111         if (hv_acpi_dev == NULL)
 112                 return -ENODEV;
 113
 114         return 0;
 115 }
 116
 117 #define VMBUS_ALIAS_LEN ((sizeof((struct hv_vmbus_device_id *)0)->guid) * 2)
 118 static void print_alias_name(struct hv_device *hv_dev, char *alias_name)
 119 {
 120         int i;
 121         for (i = 0; i < VMBUS_ALIAS_LEN; i += 2)
 122                 sprintf(&alias_name[i], "%02x", hv_dev->dev_type.b[i/2]);
 123 }
 124
 125 static u8 channel_monitor_group(struct vmbus_channel *channel)
 126 {
 127         return (u8)channel->offermsg.monitorid / 32;
 128 }
 129
 130 static u8 channel_monitor_offset(struct vmbus_channel *channel)
 131 {
 132         return (u8)channel->offermsg.monitorid % 32;
 133 }
 134
 135 static u32 channel_pending(struct vmbus_channel *channel,
 136                            struct hv_monitor_page *monitor_page)
 137 {
 138         u8 monitor_group = channel_monitor_group(channel);
 139         return monitor_page->trigger_group[monitor_group].pending;
 140 }
 141
 142 static u32 channel_latency(struct vmbus_channel *channel,
 143                            struct hv_monitor_page *monitor_page)
 144 {
 145         u8 monitor_group = channel_monitor_group(channel);
 146         u8 monitor_offset = channel_monitor_offset(channel);
 147         return monitor_page->latency[monitor_group][monitor_offset];
 148 }
 149
 150 static u32 channel_conn_id(struct vmbus_channel *channel,
 151                            struct hv_monitor_page *monitor_page)
 152 {
 153         u8 monitor_group = channel_monitor_group(channel);
 154         u8 monitor_offset = channel_monitor_offset(channel);
 155         return monitor_page->parameter[monitor_group][monitor_offset].connectionid.u.id;
 156 }
 157
 158 static ssize_t id_show(struct device *dev, struct device_attribute *dev_attr,
 159                        char *buf)
 160 {
 161         struct hv_device *hv_dev = device_to_hv_device(dev);
 162
 163         if (!hv_dev->channel)
 164                 return -ENODEV;
 165         return sprintf(buf, "%d\n", hv_dev->channel->offermsg.child_relid);
 166 }
 167 static DEVICE_ATTR_RO(id);
 168
 169 static ssize_t state_show(struct device *dev, struct device_attribute *dev_attr,
 170                           char *buf)
 171 {
 172         struct hv_device *hv_dev = device_to_hv_device(dev);
 173
 174         if (!hv_dev->channel)
 175                 return -ENODEV;
 176         return sprintf(buf, "%d\n", hv_dev->channel->state);
 177 }
 178 static DEVICE_ATTR_RO(state);
 179
 180 static ssize_t monitor_id_show(struct device *dev,
 181                                struct device_attribute *dev_attr, char *buf)
 182 {
 183         struct hv_device *hv_dev = device_to_hv_device(dev);
 184
 185         if (!hv_dev->channel)
 186                 return -ENODEV;
 187         return sprintf(buf, "%d\n", hv_dev->channel->offermsg.monitorid);
 188 }
 189 static DEVICE_ATTR_RO(monitor_id);
 190
 191 static ssize_t class_id_show(struct device *dev,
 192                                struct device_attribute *dev_attr, char *buf)
 193 {
 194         struct hv_device *hv_dev = device_to_hv_device(dev);
 195
 196         if (!hv_dev->channel)
 197                 return -ENODEV;
 198         return sprintf(buf, "{%pUl}\n",
 199                        hv_dev->channel->offermsg.offer.if_type.b);
 200 }
 201 static DEVICE_ATTR_RO(class_id);
 202
 203 static ssize_t device_id_show(struct device *dev,
 204                               struct device_attribute *dev_attr, char *buf)
 205 {
 206         struct hv_device *hv_dev = device_to_hv_device(dev);
 207
 208         if (!hv_dev->channel)
 209                 return -ENODEV;
 210         return sprintf(buf, "{%pUl}\n",
 211                        hv_dev->channel->offermsg.offer.if_instance.b);
 212 }
 213 static DEVICE_ATTR_RO(device_id);
 214
 215 static ssize_t modalias_show(struct device *dev,
 216                              struct device_attribute *dev_attr, char *buf)
 217 {
 218         struct hv_device *hv_dev = device_to_hv_device(dev);
 219         char alias_name[VMBUS_ALIAS_LEN + 1];
 220
 221         print_alias_name(hv_dev, alias_name);
 222         return sprintf(buf, "vmbus:%s\n", alias_name);
 223 }
 224 static DEVICE_ATTR_RO(modalias);
 225
 226 static ssize_t server_monitor_pending_show(struct device *dev,
 227                                            struct device_attribute *dev_attr,
 228                                            char *buf)
 229 {
 230         struct hv_device *hv_dev = device_to_hv_device(dev);
 231
 232         if (!hv_dev->channel)
 233                 return -ENODEV;
 234         return sprintf(buf, "%d\n",
 235                        channel_pending(hv_dev->channel,
 236                                        vmbus_connection.monitor_pages[1]));
 237 }
 238 static DEVICE_ATTR_RO(server_monitor_pending);
 239
 240 static ssize_t client_monitor_pending_show(struct device *dev,
 241                                            struct device_attribute *dev_attr,
 242                                            char *buf)
 243 {
 244         struct hv_device *hv_dev = device_to_hv_device(dev);
 245
 246         if (!hv_dev->channel)
 247                 return -ENODEV;
 248         return sprintf(buf, "%d\n",
 249                        channel_pending(hv_dev->channel,
 250                                        vmbus_connection.monitor_pages[1]));
 251 }
 252 static DEVICE_ATTR_RO(client_monitor_pending);
 253
 254 static ssize_t server_monitor_latency_show(struct device *dev,
 255                                            struct device_attribute *dev_attr,
 256                                            char *buf)
 257 {
 258         struct hv_device *hv_dev = device_to_hv_device(dev);
 259
 260         if (!hv_dev->channel)
 261                 return -ENODEV;
 262         return sprintf(buf, "%d\n",
 263                        channel_latency(hv_dev->channel,
 264                                        vmbus_connection.monitor_pages[0]));
 265 }
 266 static DEVICE_ATTR_RO(server_monitor_latency);
 267
 268 static ssize_t client_monitor_latency_show(struct device *dev,
 269                                            struct device_attribute *dev_attr,
 270                                            char *buf)
 271 {
 272         struct hv_device *hv_dev = device_to_hv_device(dev);
 273
 274         if (!hv_dev->channel)
 275                 return -ENODEV;
 276         return sprintf(buf, "%d\n",
 277                        channel_latency(hv_dev->channel,
 278                                        vmbus_connection.monitor_pages[1]));
 279 }
 280 static DEVICE_ATTR_RO(client_monitor_latency);
 281
 282 static ssize_t server_monitor_conn_id_show(struct device *dev,
 283                                            struct device_attribute *dev_attr,
 284                                            char *buf)
 285 {
 286         struct hv_device *hv_dev = device_to_hv_device(dev);
 287
 288         if (!hv_dev->channel)
 289                 return -ENODEV;
 290         return sprintf(buf, "%d\n",
 291                        channel_conn_id(hv_dev->channel,
 292                                        vmbus_connection.monitor_pages[0]));
 293 }
 294 static DEVICE_ATTR_RO(server_monitor_conn_id);
 295
 296 static ssize_t client_monitor_conn_id_show(struct device *dev,
 297                                            struct device_attribute *dev_attr,
 298                                            char *buf)
 299 {
 300         struct hv_device *hv_dev = device_to_hv_device(dev);
 301
 302         if (!hv_dev->channel)
 303                 return -ENODEV;
 304         return sprintf(buf, "%d\n",
 305                        channel_conn_id(hv_dev->channel,
 306                                        vmbus_connection.monitor_pages[1]));
 307 }
 308 static DEVICE_ATTR_RO(client_monitor_conn_id);
 309
 310 static ssize_t out_intr_mask_show(struct device *dev,
 311                                   struct device_attribute *dev_attr, char *buf)
 312 {
 313         struct hv_device *hv_dev = device_to_hv_device(dev);
 314         struct hv_ring_buffer_debug_info outbound;
 315
 316         if (!hv_dev->channel)
 317                 return -ENODEV;
 318         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
 319         return sprintf(buf, "%d\n", outbound.current_interrupt_mask);
 320 }
 321 static DEVICE_ATTR_RO(out_intr_mask);
 322
 323 static ssize_t out_read_index_show(struct device *dev,
 324                                    struct device_attribute *dev_attr, char *buf)
 325 {
 326         struct hv_device *hv_dev = device_to_hv_device(dev);
 327         struct hv_ring_buffer_debug_info outbound;
 328
 329         if (!hv_dev->channel)
 330                 return -ENODEV;
 331         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
 332         return sprintf(buf, "%d\n", outbound.current_read_index);
 333 }
 334 static DEVICE_ATTR_RO(out_read_index);
 335
 336 static ssize_t out_write_index_show(struct device *dev,
 337                                     struct device_attribute *dev_attr,
 338                                     char *buf)
 339 {
 340         struct hv_device *hv_dev = device_to_hv_device(dev);
 341         struct hv_ring_buffer_debug_info outbound;
 342
 343         if (!hv_dev->channel)
 344                 return -ENODEV;
 345         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
 346         return sprintf(buf, "%d\n", outbound.current_write_index);
 347 }
 348 static DEVICE_ATTR_RO(out_write_index);
 349
 350 static ssize_t out_read_bytes_avail_show(struct device *dev,
 351                                          struct device_attribute *dev_attr,
 352                                          char *buf)
 353 {
 354         struct hv_device *hv_dev = device_to_hv_device(dev);
 355         struct hv_ring_buffer_debug_info outbound;
 356
 357         if (!hv_dev->channel)
 358                 return -ENODEV;
 359         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
 360         return sprintf(buf, "%d\n", outbound.bytes_avail_toread);
 361 }
 362 static DEVICE_ATTR_RO(out_read_bytes_avail);
 363
 364 static ssize_t out_write_bytes_avail_show(struct device *dev,
 365                                           struct device_attribute *dev_attr,
 366                                           char *buf)
 367 {
 368         struct hv_device *hv_dev = device_to_hv_device(dev);
 369         struct hv_ring_buffer_debug_info outbound;
 370
 371         if (!hv_dev->channel)
 372                 return -ENODEV;
 373         hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound);
 374         return sprintf(buf, "%d\n", outbound.bytes_avail_towrite);
 375 }
 376 static DEVICE_ATTR_RO(out_write_bytes_avail);
 377
 378 static ssize_t in_intr_mask_show(struct device *dev,
 379                                  struct device_attribute *dev_attr, char *buf)
 380 {
 381         struct hv_device *hv_dev = device_to_hv_device(dev);
 382         struct hv_ring_buffer_debug_info inbound;
 383
 384         if (!hv_dev->channel)
 385                 return -ENODEV;
 386         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 387         return sprintf(buf, "%d\n", inbound.current_interrupt_mask);
 388 }
 389 static DEVICE_ATTR_RO(in_intr_mask);
 390
 391 static ssize_t in_read_index_show(struct device *dev,
 392                                   struct device_attribute *dev_attr, char *buf)
 393 {
 394         struct hv_device *hv_dev = device_to_hv_device(dev);
 395         struct hv_ring_buffer_debug_info inbound;
 396
 397         if (!hv_dev->channel)
 398                 return -ENODEV;
 399         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 400         return sprintf(buf, "%d\n", inbound.current_read_index);
 401 }
 402 static DEVICE_ATTR_RO(in_read_index);
 403
 404 static ssize_t in_write_index_show(struct device *dev,
 405                                    struct device_attribute *dev_attr, char *buf)
 406 {
 407         struct hv_device *hv_dev = device_to_hv_device(dev);
 408         struct hv_ring_buffer_debug_info inbound;
 409
 410         if (!hv_dev->channel)
 411                 return -ENODEV;
 412         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 413         return sprintf(buf, "%d\n", inbound.current_write_index);
 414 }
 415 static DEVICE_ATTR_RO(in_write_index);
 416
 417 static ssize_t in_read_bytes_avail_show(struct device *dev,
 418                                         struct device_attribute *dev_attr,
 419                                         char *buf)
 420 {
 421         struct hv_device *hv_dev = device_to_hv_device(dev);
 422         struct hv_ring_buffer_debug_info inbound;
 423
 424         if (!hv_dev->channel)
 425                 return -ENODEV;
 426         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 427         return sprintf(buf, "%d\n", inbound.bytes_avail_toread);
 428 }
 429 static DEVICE_ATTR_RO(in_read_bytes_avail);
 430
 431 static ssize_t in_write_bytes_avail_show(struct device *dev,
 432                                          struct device_attribute *dev_attr,
 433                                          char *buf)
 434 {
 435         struct hv_device *hv_dev = device_to_hv_device(dev);
 436         struct hv_ring_buffer_debug_info inbound;
 437
 438         if (!hv_dev->channel)
 439                 return -ENODEV;
 440         hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound);
 441         return sprintf(buf, "%d\n", inbound.bytes_avail_towrite);
 442 }
 443 static DEVICE_ATTR_RO(in_write_bytes_avail);
 444
 445 static ssize_t channel_vp_mapping_show(struct device *dev,
 446                                        struct device_attribute *dev_attr,
 447                                        char *buf)
 448 {
 449         struct hv_device *hv_dev = device_to_hv_device(dev);
 450         struct vmbus_channel *channel = hv_dev->channel, *cur_sc;
 451         unsigned long flags;
 452         int buf_size = PAGE_SIZE, n_written, tot_written;
 453         struct list_head *cur;
 454
 455         if (!channel)
 456                 return -ENODEV;
 457
 458         tot_written = snprintf(buf, buf_size, "%u:%u\n",
 459                 channel->offermsg.child_relid, channel->target_cpu);
 460
 461         spin_lock_irqsave(&channel->lock, flags);
 462
 463         list_for_each(cur, &channel->sc_list) {
 464                 if (tot_written >= buf_size - 1)
 465                         break;
 466
 467                 cur_sc = list_entry(cur, struct vmbus_channel, sc_list);
 468                 n_written = scnprintf(buf + tot_written,
 469                                      buf_size - tot_written,
 470                                      "%u:%u\n",
 471                                      cur_sc->offermsg.child_relid,
 472                                      cur_sc->target_cpu);
 473                 tot_written += n_written;
 474         }
 475
 476         spin_unlock_irqrestore(&channel->lock, flags);
 477
 478         return tot_written;
 479 }
 480 static DEVICE_ATTR_RO(channel_vp_mapping);
 481
 482 /* Set up per device attributes in /sys/bus/vmbus/devices/<bus device> */
 483 static struct attribute *vmbus_attrs[] = {
 484         &dev_attr_id.attr,
 485         &dev_attr_state.attr,
 486         &dev_attr_monitor_id.attr,
 487         &dev_attr_class_id.attr,
 488         &dev_attr_device_id.attr,
 489         &dev_attr_modalias.attr,
 490         &dev_attr_server_monitor_pending.attr,
 491         &dev_attr_client_monitor_pending.attr,
 492         &dev_attr_server_monitor_latency.attr,
 493         &dev_attr_client_monitor_latency.attr,
 494         &dev_attr_server_monitor_conn_id.attr,
 495         &dev_attr_client_monitor_conn_id.attr,
 496         &dev_attr_out_intr_mask.attr,
 497         &dev_attr_out_read_index.attr,
 498         &dev_attr_out_write_index.attr,
 499         &dev_attr_out_read_bytes_avail.attr,
 500         &dev_attr_out_write_bytes_avail.attr,
 501         &dev_attr_in_intr_mask.attr,
 502         &dev_attr_in_read_index.attr,
 503         &dev_attr_in_write_index.attr,
 504         &dev_attr_in_read_bytes_avail.attr,
 505         &dev_attr_in_write_bytes_avail.attr,
 506         &dev_attr_channel_vp_mapping.attr,
 507         NULL,
 508 };
 509 ATTRIBUTE_GROUPS(vmbus);
 510
 511 /*
 512  * vmbus_uevent - add uevent for our device
 513  *
 514  * This routine is invoked when a device is added or removed on the vmbus to
 515  * generate a uevent to udev in the userspace. The udev will then look at its
 516  * rule and the uevent generated here to load the appropriate driver
 517  *
 518  * The alias string will be of the form vmbus:guid where guid is the string
 519  * representation of the device guid (each byte of the guid will be
 520  * represented with two hex characters.
 521  */
 522 static int vmbus_uevent(struct device *device, struct kobj_uevent_env *env)
 523 {
 524         struct hv_device *dev = device_to_hv_device(device);
 525         int ret;
 526         char alias_name[VMBUS_ALIAS_LEN + 1];
 527
 528         print_alias_name(dev, alias_name);
 529         ret = add_uevent_var(env, "MODALIAS=vmbus:%s", alias_name);
 530         return ret;
 531 }
 532
 533 static const uuid_le null_guid;
 534
 535 static inline bool is_null_guid(const __u8 *guid)
 536 {
 537         if (memcmp(guid, &null_guid, sizeof(uuid_le)))
 538                 return false;
 539         return true;
 540 }
 541
 542 /*
 543  * Return a matching hv_vmbus_device_id pointer.
 544  * If there is no match, return NULL.
 545  */
 546 static const struct hv_vmbus_device_id *hv_vmbus_get_id(
 547                                         const struct hv_vmbus_device_id *id,
 548                                         const __u8 *guid)
 549 {
 550         for (; !is_null_guid(id->guid); id++)
 551                 if (!memcmp(&id->guid, guid, sizeof(uuid_le)))
 552                         return id;
 553
 554         return NULL;
 555 }
 556
 557
 558
 559 /*
 560  * vmbus_match - Attempt to match the specified device to the specified driver
 561  */
 562 static int vmbus_match(struct device *device, struct device_driver *driver)
 563 {
 564         struct hv_driver *drv = drv_to_hv_drv(driver);
 565         struct hv_device *hv_dev = device_to_hv_device(device);
 566
 567         if (hv_vmbus_get_id(drv->id_table, hv_dev->dev_type.b))
 568                 return 1;
 569
 570         return 0;
 571 }
 572
 573 /*
 574  * vmbus_probe - Add the new vmbus's child device
 575  */
 576 static int vmbus_probe(struct device *child_device)
 577 {
 578         int ret = 0;
 579         struct hv_driver *drv =
 580                         drv_to_hv_drv(child_device->driver);
 581         struct hv_device *dev = device_to_hv_device(child_device);
 582         const struct hv_vmbus_device_id *dev_id;
 583
 584         dev_id = hv_vmbus_get_id(drv->id_table, dev->dev_type.b);
 585         if (drv->probe) {
 586                 ret = drv->probe(dev, dev_id);
 587                 if (ret != 0)
 588                         pr_err("probe failed for device %s (%d)\n",
 589                                dev_name(child_device), ret);
 590
 591         } else {
 592                 pr_err("probe not set for driver %s\n",
 593                        dev_name(child_device));
 594                 ret = -ENODEV;
 595         }
 596         return ret;
 597 }
 598
 599 /*
 600  * vmbus_remove - Remove a vmbus device
 601  */
 602 static int vmbus_remove(struct device *child_device)
 603 {
 604         struct hv_driver *drv;
 605         struct hv_device *dev = device_to_hv_device(child_device);
 606         u32 relid = dev->channel->offermsg.child_relid;
 607
 608         if (child_device->driver) {
 609                 drv = drv_to_hv_drv(child_device->driver);
 610                 if (drv->remove)
 611                         drv->remove(dev);
 612                 else {
 613                         hv_process_channel_removal(dev->channel, relid);
 614                         pr_err("remove not set for driver %s\n",
 615                                 dev_name(child_device));
 616                 }
 617         } else {
 618                 /*
 619                  * We don't have a driver for this device; deal with the
 620                  * rescind message by removing the channel.
 621                  */
 622                 hv_process_channel_removal(dev->channel, relid);
 623         }
 624
 625         return 0;
 626 }
 627
 628
 629 /*
 630  * vmbus_shutdown - Shutdown a vmbus device
 631  */
 632 static void vmbus_shutdown(struct device *child_device)
 633 {
 634         struct hv_driver *drv;
 635         struct hv_device *dev = device_to_hv_device(child_device);
 636
 637
 638         /* The device may not be attached yet */
 639         if (!child_device->driver)
 640                 return;
 641
 642         drv = drv_to_hv_drv(child_device->driver);
 643
 644         if (drv->shutdown)
 645                 drv->shutdown(dev);
 646
 647         return;
 648 }
 649
 650
 651 /*
 652  * vmbus_device_release - Final callback release of the vmbus child device
 653  */
 654 static void vmbus_device_release(struct device *device)
 655 {
 656         struct hv_device *hv_dev = device_to_hv_device(device);
 657
 658         kfree(hv_dev);
 659
 660 }
 661
 662 /* The one and only one */
 663 static struct bus_type  hv_bus = {
 664         .name =         "vmbus",
 665         .match =                vmbus_match,
 666         .shutdown =             vmbus_shutdown,
 667         .remove =               vmbus_remove,
 668         .probe =                vmbus_probe,
 669         .uevent =               vmbus_uevent,
 670         .dev_groups =           vmbus_groups,
 671 };
 672
 673 struct onmessage_work_context {
 674         struct work_struct work;
 675         struct hv_message msg;
 676 };
 677
 678 static void vmbus_onmessage_work(struct work_struct *work)
 679 {
 680         struct onmessage_work_context *ctx;
 681
 682         /* Do not process messages if we're in DISCONNECTED state */
 683         if (vmbus_connection.conn_state == DISCONNECTED)
 684                 return;
 685
 686         ctx = container_of(work, struct onmessage_work_context,
 687                            work);
 688         vmbus_onmessage(&ctx->msg);
 689         kfree(ctx);
 690 }
 691
 692 static void hv_process_timer_expiration(struct hv_message *msg, int cpu)
 693 {
 694         struct clock_event_device *dev = hv_context.clk_evt[cpu];
 695
 696         if (dev->event_handler)
 697                 dev->event_handler(dev);
 698
 699         msg->header.message_type = HVMSG_NONE;
 700
 701         /*
 702          * Make sure the write to MessageType (ie set to
 703          * HVMSG_NONE) happens before we read the
 704          * MessagePending and EOMing. Otherwise, the EOMing
 705          * will not deliver any more messages since there is
 706          * no empty slot
 707          */
 708         mb();
 709
 710         if (msg->header.message_flags.msg_pending) {
 711                 /*
 712                  * This will cause message queue rescan to
 713                  * possibly deliver another msg from the
 714                  * hypervisor
 715                  */
 716                 wrmsrl(HV_X64_MSR_EOM, 0);
 717         }
 718 }
 719
 720 static void vmbus_on_msg_dpc(unsigned long data)
 721 {
 722         int cpu = smp_processor_id();
 723         void *page_addr = hv_context.synic_message_page[cpu];
 724         struct hv_message *msg = (struct hv_message *)page_addr +
 725                                   VMBUS_MESSAGE_SINT;
 726         struct vmbus_channel_message_header *hdr;
 727         struct vmbus_channel_message_table_entry *entry;
 728         struct onmessage_work_context *ctx;
 729
 730         while (1) {
 731                 if (msg->header.message_type == HVMSG_NONE)
 732                         /* no msg */
 733                         break;
 734
 735                 hdr = (struct vmbus_channel_message_header *)msg->u.payload;
 736
 737                 if (hdr->msgtype >= CHANNELMSG_COUNT) {
 738                         WARN_ONCE(1, "unknown msgtype=%d\n", hdr->msgtype);
 739                         goto msg_handled;
 740                 }
 741
 742                 entry = &channel_message_table[hdr->msgtype];
 743                 if (entry->handler_type == VMHT_BLOCKING) {
 744                         ctx = kmalloc(sizeof(*ctx), GFP_ATOMIC);
 745                         if (ctx == NULL)
 746                                 continue;
 747
 748                         INIT_WORK(&ctx->work, vmbus_onmessage_work);
 749                         memcpy(&ctx->msg, msg, sizeof(*msg));
 750
 751                         queue_work(vmbus_connection.work_queue, &ctx->work);
 752                 } else
 753                         entry->message_handler(hdr);
 754
 755 msg_handled:
 756                 msg->header.message_type = HVMSG_NONE;
 757
 758                 /*
 759                  * Make sure the write to MessageType (ie set to
 760                  * HVMSG_NONE) happens before we read the
 761                  * MessagePending and EOMing. Otherwise, the EOMing
 762                  * will not deliver any more messages since there is
 763                  * no empty slot
 764                  */
 765                 mb();
 766
 767                 if (msg->header.message_flags.msg_pending) {
 768                         /*
 769                          * This will cause message queue rescan to
 770                          * possibly deliver another msg from the
 771                          * hypervisor
 772                          */
 773                         wrmsrl(HV_X64_MSR_EOM, 0);
 774                 }
 775         }
 776 }
 777
 778 static void vmbus_isr(void)
 779 {
 780         int cpu = smp_processor_id();
 781         void *page_addr;
 782         struct hv_message *msg;
 783         union hv_synic_event_flags *event;
 784         bool handled = false;
 785
 786         page_addr = hv_context.synic_event_page[cpu];
 787         if (page_addr == NULL)
 788                 return;
 789
 790         event = (union hv_synic_event_flags *)page_addr +
 791                                          VMBUS_MESSAGE_SINT;
 792         /*
 793          * Check for events before checking for messages. This is the order
 794          * in which events and messages are checked in Windows guests on
 795          * Hyper-V, and the Windows team suggested we do the same.
 796          */
 797
 798         if ((vmbus_proto_version == VERSION_WS2008) ||
 799                 (vmbus_proto_version == VERSION_WIN7)) {
 800
 801                 /* Since we are a child, we only need to check bit 0 */
 802                 if (sync_test_and_clear_bit(0,
 803                         (unsigned long *) &event->flags32[0])) {
 804                         handled = true;
 805                 }
 806         } else {
 807                 /*
 808                  * Our host is win8 or above. The signaling mechanism
 809                  * has changed and we can directly look at the event page.
 810                  * If bit n is set then we have an interrup on the channel
 811                  * whose id is n.
 812                  */
 813                 handled = true;
 814         }
 815
 816         if (handled)
 817                 tasklet_schedule(hv_context.event_dpc[cpu]);
 818
 819
 820         page_addr = hv_context.synic_message_page[cpu];
 821         msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
 822
 823         /* Check if there are actual msgs to be processed */
 824         if (msg->header.message_type != HVMSG_NONE) {
 825                 if (msg->header.message_type == HVMSG_TIMER_EXPIRED)
 826                         hv_process_timer_expiration(msg, cpu);
 827                 else
 828                         tasklet_schedule(&msg_dpc);
 829         }
 830
 831         add_interrupt_randomness(HYPERVISOR_CALLBACK_VECTOR, 0);
 832 }
 833
 834
 835 /*
 836  * vmbus_bus_init -Main vmbus driver initialization routine.
 837  *
 838  * Here, we
 839  *      - initialize the vmbus driver context
 840  *      - invoke the vmbus hv main init routine
 841  *      - get the irq resource
 842  *      - retrieve the channel offers
 843  */
 844 static int vmbus_bus_init(int irq)
 845 {
 846         int ret;
 847
 848         /* Hypervisor initialization...setup hypercall page..etc */
 849         ret = hv_init();
 850         if (ret != 0) {
 851                 pr_err("Unable to initialize the hypervisor - 0x%x\n", ret);
 852                 return ret;
 853         }
 854
 855         tasklet_init(&msg_dpc, vmbus_on_msg_dpc, 0);
 856
 857         ret = bus_register(&hv_bus);
 858         if (ret)
 859                 goto err_cleanup;
 860
 861         hv_setup_vmbus_irq(vmbus_isr);
 862
 863         ret = hv_synic_alloc();
 864         if (ret)
 865                 goto err_alloc;
 866         /*
 867          * Initialize the per-cpu interrupt state and
 868          * connect to the host.
 869          */
 870         on_each_cpu(hv_synic_init, NULL, 1);
 871         ret = vmbus_connect();
 872         if (ret)
 873                 goto err_alloc;
 874
 875         if (vmbus_proto_version > VERSION_WIN7)
 876                 cpu_hotplug_disable();
 877
 878         /*
 879          * Only register if the crash MSRs are available
 880          */
 881         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
 882                 register_die_notifier(&hyperv_die_block);
 883                 atomic_notifier_chain_register(&panic_notifier_list,
 884                                                &hyperv_panic_block);
 885         }
 886
 887         vmbus_request_offers();
 888
 889         return 0;
 890
 891 err_alloc:
 892         hv_synic_free();
 893         hv_remove_vmbus_irq();
 894
 895         bus_unregister(&hv_bus);
 896
 897 err_cleanup:
 898         hv_cleanup();
 899
 900         return ret;
 901 }
 902
 903 /**
 904  * __vmbus_child_driver_register() - Register a vmbus's driver
 905  * @hv_driver: Pointer to driver structure you want to register
 906  * @owner: owner module of the drv
 907  * @mod_name: module name string
 908  *
 909  * Registers the given driver with Linux through the 'driver_register()' call
 910  * and sets up the hyper-v vmbus handling for this driver.
 911  * It will return the state of the 'driver_register()' call.
 912  *
 913  */
 914 int __vmbus_driver_register(struct hv_driver *hv_driver, struct module *owner, const char *mod_name)
 915 {
 916         int ret;
 917
 918         pr_info("registering driver %s\n", hv_driver->name);
 919
 920         ret = vmbus_exists();
 921         if (ret < 0)
 922                 return ret;
 923
 924         hv_driver->driver.name = hv_driver->name;
 925         hv_driver->driver.owner = owner;
 926         hv_driver->driver.mod_name = mod_name;
 927         hv_driver->driver.bus = &hv_bus;
 928
 929         ret = driver_register(&hv_driver->driver);
 930
 931         return ret;
 932 }
 933 EXPORT_SYMBOL_GPL(__vmbus_driver_register);
 934
 935 /**
 936  * vmbus_driver_unregister() - Unregister a vmbus's driver
 937  * @hv_driver: Pointer to driver structure you want to
 938  *             un-register
 939  *
 940  * Un-register the given driver that was previous registered with a call to
 941  * vmbus_driver_register()
 942  */
 943 void vmbus_driver_unregister(struct hv_driver *hv_driver)
 944 {
 945         pr_info("unregistering driver %s\n", hv_driver->name);
 946
 947         if (!vmbus_exists())
 948                 driver_unregister(&hv_driver->driver);
 949 }
 950 EXPORT_SYMBOL_GPL(vmbus_driver_unregister);
 951
 952 /*
 953  * vmbus_device_create - Creates and registers a new child device
 954  * on the vmbus.
 955  */
 956 struct hv_device *vmbus_device_create(const uuid_le *type,
 957                                       const uuid_le *instance,
 958                                       struct vmbus_channel *channel)
 959 {
 960         struct hv_device *child_device_obj;
 961
 962         child_device_obj = kzalloc(sizeof(struct hv_device), GFP_KERNEL);
 963         if (!child_device_obj) {
 964                 pr_err("Unable to allocate device object for child device\n");
 965                 return NULL;
 966         }
 967
 968         child_device_obj->channel = channel;
 969         memcpy(&child_device_obj->dev_type, type, sizeof(uuid_le));
 970         memcpy(&child_device_obj->dev_instance, instance,
 971                sizeof(uuid_le));
 972
 973
 974         return child_device_obj;
 975 }
 976
 977 /*
 978  * vmbus_device_register - Register the child device
 979  */
 980 int vmbus_device_register(struct hv_device *child_device_obj)
 981 {
 982         int ret = 0;
 983
 984         dev_set_name(&child_device_obj->device, "vmbus_%d",
 985                      child_device_obj->channel->id);
 986
 987         child_device_obj->device.bus = &hv_bus;
 988         child_device_obj->device.parent = &hv_acpi_dev->dev;
 989         child_device_obj->device.release = vmbus_device_release;
 990
 991         /*
 992          * Register with the LDM. This will kick off the driver/device
 993          * binding...which will eventually call vmbus_match() and vmbus_probe()
 994          */
 995         ret = device_register(&child_device_obj->device);
 996
 997         if (ret)
 998                 pr_err("Unable to register child device\n");
 999         else
1000                 pr_debug("child device %s registered\n",
1001                         dev_name(&child_device_obj->device));
1002
1003         return ret;
1004 }
1005
1006 /*
1007  * vmbus_device_unregister - Remove the specified child device
1008  * from the vmbus.
1009  */
1010 void vmbus_device_unregister(struct hv_device *device_obj)
1011 {
1012         pr_debug("child device %s unregistered\n",
1013                 dev_name(&device_obj->device));
1014
1015         /*
1016          * Kick off the process of unregistering the device.
1017          * This will call vmbus_remove() and eventually vmbus_device_release()
1018          */
1019         device_unregister(&device_obj->device);
1020 }
1021
1022
1023 /*
1024  * VMBUS is an acpi enumerated device. Get the information we
1025  * need from DSDT.
1026  */
1027 #define VTPM_BASE_ADDRESS 0xfed40000
1028 static acpi_status vmbus_walk_resources(struct acpi_resource *res, void *ctx)
1029 {
1030         resource_size_t start = 0;
1031         resource_size_t end = 0;
1032         struct resource *new_res;
1033         struct resource **old_res = &hyperv_mmio;
1034         struct resource **prev_res = NULL;
1035
1036         switch (res->type) {
1037         case ACPI_RESOURCE_TYPE_IRQ:
1038                 irq = res->data.irq.interrupts[0];
1039                 return AE_OK;
1040
1041         /*
1042          * "Address" descriptors are for bus windows. Ignore
1043          * "memory" descriptors, which are for registers on
1044          * devices.
1045          */
1046         case ACPI_RESOURCE_TYPE_ADDRESS32:
1047                 start = res->data.address32.address.minimum;
1048                 end = res->data.address32.address.maximum;
1049                 break;
1050
1051         case ACPI_RESOURCE_TYPE_ADDRESS64:
1052                 start = res->data.address64.address.minimum;
1053                 end = res->data.address64.address.maximum;
1054                 break;
1055
1056         default:
1057                 /* Unused resource type */
1058                 return AE_OK;
1059
1060         }
1061         /*
1062          * Ignore ranges that are below 1MB, as they're not
1063          * necessary or useful here.
1064          */
1065         if (end < 0x100000)
1066                 return AE_OK;
1067
1068         new_res = kzalloc(sizeof(*new_res), GFP_ATOMIC);
1069         if (!new_res)
1070                 return AE_NO_MEMORY;
1071
1072         /* If this range overlaps the virtual TPM, truncate it. */
1073         if (end > VTPM_BASE_ADDRESS && start < VTPM_BASE_ADDRESS)
1074                 end = VTPM_BASE_ADDRESS;
1075
1076         new_res->name = "hyperv mmio";
1077         new_res->flags = IORESOURCE_MEM;
1078         new_res->start = start;
1079         new_res->end = end;
1080
1081         do {
1082                 if (!*old_res) {
1083                         *old_res = new_res;
1084                         break;
1085                 }
1086
1087                 if ((*old_res)->end < new_res->start) {
1088                         new_res->sibling = *old_res;
1089                         if (prev_res)
1090                                 (*prev_res)->sibling = new_res;
1091                         *old_res = new_res;
1092                         break;
1093                 }
1094
1095                 prev_res = old_res;
1096                 old_res = &(*old_res)->sibling;
1097
1098         } while (1);
1099
1100         return AE_OK;
1101 }
1102
1103 static int vmbus_acpi_remove(struct acpi_device *device)
1104 {
1105         struct resource *cur_res;
1106         struct resource *next_res;
1107
1108         if (hyperv_mmio) {
1109                 for (cur_res = hyperv_mmio; cur_res; cur_res = next_res) {
1110                         next_res = cur_res->sibling;
1111                         kfree(cur_res);
1112                 }
1113         }
1114
1115         return 0;
1116 }
1117
1118 /**
1119  * vmbus_allocate_mmio() - Pick a memory-mapped I/O range.
1120  * @new:                If successful, supplied a pointer to the
1121  *                      allocated MMIO space.
1122  * @device_obj:         Identifies the caller
1123  * @min:                Minimum guest physical address of the
1124  *                      allocation
1125  * @max:                Maximum guest physical address
1126  * @size:               Size of the range to be allocated
1127  * @align:              Alignment of the range to be allocated
1128  * @fb_overlap_ok:      Whether this allocation can be allowed
1129  *                      to overlap the video frame buffer.
1130  *
1131  * This function walks the resources granted to VMBus by the
1132  * _CRS object in the ACPI namespace underneath the parent
1133  * "bridge" whether that's a root PCI bus in the Generation 1
1134  * case or a Module Device in the Generation 2 case.  It then
1135  * attempts to allocate from the global MMIO pool in a way that
1136  * matches the constraints supplied in these parameters and by
1137  * that _CRS.
1138  *
1139  * Return: 0 on success, -errno on failure
1140  */
1141 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj,
1142                         resource_size_t min, resource_size_t max,
1143                         resource_size_t size, resource_size_t align,
1144                         bool fb_overlap_ok)
1145 {
1146         struct resource *iter;
1147         resource_size_t range_min, range_max, start, local_min, local_max;
1148         const char *dev_n = dev_name(&device_obj->device);
1149         u32 fb_end = screen_info.lfb_base + (screen_info.lfb_size << 1);
1150         int i;
1151
1152         for (iter = hyperv_mmio; iter; iter = iter->sibling) {
1153                 if ((iter->start >= max) || (iter->end <= min))
1154                         continue;
1155
1156                 range_min = iter->start;
1157                 range_max = iter->end;
1158
1159                 /* If this range overlaps the frame buffer, split it into
1160                    two tries. */
1161                 for (i = 0; i < 2; i++) {
1162                         local_min = range_min;
1163                         local_max = range_max;
1164                         if (fb_overlap_ok || (range_min >= fb_end) ||
1165                             (range_max <= screen_info.lfb_base)) {
1166                                 i++;
1167                         } else {
1168                                 if ((range_min <= screen_info.lfb_base) &&
1169                                     (range_max >= screen_info.lfb_base)) {
1170                                         /*
1171                                          * The frame buffer is in this window,
1172                                          * so trim this into the part that
1173                                          * preceeds the frame buffer.
1174                                          */
1175                                         local_max = screen_info.lfb_base - 1;
1176                                         range_min = fb_end;
1177                                 } else {
1178                                         range_min = fb_end;
1179                                         continue;
1180                                 }
1181                         }
1182
1183                         start = (local_min + align - 1) & ~(align - 1);
1184                         for (; start + size - 1 <= local_max; start += align) {
1185                                 *new = request_mem_region_exclusive(start, size,
1186                                                                     dev_n);
1187                                 if (*new)
1188                                         return 0;
1189                         }
1190                 }
1191         }
1192
1193         return -ENXIO;
1194 }
1195 EXPORT_SYMBOL_GPL(vmbus_allocate_mmio);
1196
1197 static int vmbus_acpi_add(struct acpi_device *device)
1198 {
1199         acpi_status result;
1200         int ret_val = -ENODEV;
1201         struct acpi_device *ancestor;
1202
1203         hv_acpi_dev = device;
1204
1205         result = acpi_walk_resources(device->handle, METHOD_NAME__CRS,
1206                                         vmbus_walk_resources, NULL);
1207
1208         if (ACPI_FAILURE(result))
1209                 goto acpi_walk_err;
1210         /*
1211          * Some ancestor of the vmbus acpi device (Gen1 or Gen2
1212          * firmware) is the VMOD that has the mmio ranges. Get that.
1213          */
1214         for (ancestor = device->parent; ancestor; ancestor = ancestor->parent) {
1215                 result = acpi_walk_resources(ancestor->handle, METHOD_NAME__CRS,
1216                                              vmbus_walk_resources, NULL);
1217
1218                 if (ACPI_FAILURE(result))
1219                         continue;
1220                 if (hyperv_mmio)
1221                         break;
1222         }
1223         ret_val = 0;
1224
1225 acpi_walk_err:
1226         complete(&probe_event);
1227         if (ret_val)
1228                 vmbus_acpi_remove(device);
1229         return ret_val;
1230 }
1231
1232 static const struct acpi_device_id vmbus_acpi_device_ids[] = {
1233         {"VMBUS", 0},
1234         {"VMBus", 0},
1235         {"", 0},
1236 };
1237 MODULE_DEVICE_TABLE(acpi, vmbus_acpi_device_ids);
1238
1239 static struct acpi_driver vmbus_acpi_driver = {
1240         .name = "vmbus",
1241         .ids = vmbus_acpi_device_ids,
1242         .ops = {
1243                 .add = vmbus_acpi_add,
1244                 .remove = vmbus_acpi_remove,
1245         },
1246 };
1247
1248 static void hv_kexec_handler(void)
1249 {
1250         int cpu;
1251
1252         hv_synic_clockevents_cleanup();
1253         vmbus_initiate_unload();
1254         for_each_online_cpu(cpu)
1255                 smp_call_function_single(cpu, hv_synic_cleanup, NULL, 1);
1256         hv_cleanup();
1257 };
1258
1259 static void hv_crash_handler(struct pt_regs *regs)
1260 {
1261         vmbus_initiate_unload();
1262         /*
1263          * In crash handler we can't schedule synic cleanup for all CPUs,
1264          * doing the cleanup for current CPU only. This should be sufficient
1265          * for kdump.
1266          */
1267         hv_synic_cleanup(NULL);
1268         hv_cleanup();
1269 };
1270
1271 static int __init hv_acpi_init(void)
1272 {
1273         int ret, t;
1274
1275         if (x86_hyper != &x86_hyper_ms_hyperv)
1276                 return -ENODEV;
1277
1278         init_completion(&probe_event);
1279
1280         /*
1281          * Get irq resources first.
1282          */
1283         ret = acpi_bus_register_driver(&vmbus_acpi_driver);
1284
1285         if (ret)
1286                 return ret;
1287
1288         t = wait_for_completion_timeout(&probe_event, 5*HZ);
1289         if (t == 0) {
1290                 ret = -ETIMEDOUT;
1291                 goto cleanup;
1292         }
1293
1294         if (irq <= 0) {
1295                 ret = -ENODEV;
1296                 goto cleanup;
1297         }
1298
1299         ret = vmbus_bus_init(irq);
1300         if (ret)
1301                 goto cleanup;
1302
1303         hv_setup_kexec_handler(hv_kexec_handler);
1304         hv_setup_crash_handler(hv_crash_handler);
1305
1306         return 0;
1307
1308 cleanup:
1309         acpi_bus_unregister_driver(&vmbus_acpi_driver);
1310         hv_acpi_dev = NULL;
1311         return ret;
1312 }
1313
1314 static void __exit vmbus_exit(void)
1315 {
1316         int cpu;
1317
1318         hv_remove_kexec_handler();
1319         hv_remove_crash_handler();
1320         vmbus_connection.conn_state = DISCONNECTED;
1321         hv_synic_clockevents_cleanup();
1322         vmbus_disconnect();
1323         hv_remove_vmbus_irq();
1324         tasklet_kill(&msg_dpc);
1325         vmbus_free_channels();
1326         if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) {
1327                 unregister_die_notifier(&hyperv_die_block);
1328                 atomic_notifier_chain_unregister(&panic_notifier_list,
1329                                                  &hyperv_panic_block);
1330         }
1331         bus_unregister(&hv_bus);
1332         hv_cleanup();
1333         for_each_online_cpu(cpu) {
1334                 tasklet_kill(hv_context.event_dpc[cpu]);
1335                 smp_call_function_single(cpu, hv_synic_cleanup, NULL, 1);
1336         }
1337         hv_synic_free();
1338         acpi_bus_unregister_driver(&vmbus_acpi_driver);
1339         if (vmbus_proto_version > VERSION_WIN7)
1340                 cpu_hotplug_enable();
1341 }
1342
1343
1344 MODULE_LICENSE("GPL");
1345
1346 subsys_initcall(hv_acpi_init);
1347 module_exit(vmbus_exit);