drivers/iommu/dmar.c

   1 /*
   2  * Copyright (c) 2006, Intel 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  * Copyright (C) 2006-2008 Intel Corporation
  18  * Author: Ashok Raj <ashok.raj@intel.com>
  19  * Author: Shaohua Li <shaohua.li@intel.com>
  20  * Author: Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com>
  21  *
  22  * This file implements early detection/parsing of Remapping Devices
  23  * reported to OS through BIOS via DMA remapping reporting (DMAR) ACPI
  24  * tables.
  25  *
  26  * These routines are used by both DMA-remapping and Interrupt-remapping
  27  */
  28
  29 #define pr_fmt(fmt)     "DMAR: " fmt
  30
  31 #include <linux/pci.h>
  32 #include <linux/dmar.h>
  33 #include <linux/iova.h>
  34 #include <linux/intel-iommu.h>
  35 #include <linux/timer.h>
  36 #include <linux/irq.h>
  37 #include <linux/interrupt.h>
  38 #include <linux/tboot.h>
  39 #include <linux/dmi.h>
  40 #include <linux/slab.h>
  41 #include <linux/iommu.h>
  42 #include <asm/irq_remapping.h>
  43 #include <asm/iommu_table.h>
  44
  45 #include "irq_remapping.h"
  46
  47 typedef int (*dmar_res_handler_t)(struct acpi_dmar_header *, void *);
  48 struct dmar_res_callback {
  49         dmar_res_handler_t      cb[ACPI_DMAR_TYPE_RESERVED];
  50         void                    *arg[ACPI_DMAR_TYPE_RESERVED];
  51         bool                    ignore_unhandled;
  52         bool                    print_entry;
  53 };
  54
  55 /*
  56  * Assumptions:
  57  * 1) The hotplug framework guarentees that DMAR unit will be hot-added
  58  *    before IO devices managed by that unit.
  59  * 2) The hotplug framework guarantees that DMAR unit will be hot-removed
  60  *    after IO devices managed by that unit.
  61  * 3) Hotplug events are rare.
  62  *
  63  * Locking rules for DMA and interrupt remapping related global data structures:
  64  * 1) Use dmar_global_lock in process context
  65  * 2) Use RCU in interrupt context
  66  */
  67 DECLARE_RWSEM(dmar_global_lock);
  68 LIST_HEAD(dmar_drhd_units);
  69
  70 struct acpi_table_header * __initdata dmar_tbl;
  71 static int dmar_dev_scope_status = 1;
  72 static unsigned long dmar_seq_ids[BITS_TO_LONGS(DMAR_UNITS_SUPPORTED)];
  73
  74 static int alloc_iommu(struct dmar_drhd_unit *drhd);
  75 static void free_iommu(struct intel_iommu *iommu);
  76
  77 extern const struct iommu_ops intel_iommu_ops;
  78
  79 static void dmar_register_drhd_unit(struct dmar_drhd_unit *drhd)
  80 {
  81         /*
  82          * add INCLUDE_ALL at the tail, so scan the list will find it at
  83          * the very end.
  84          */
  85         if (drhd->include_all)
  86                 list_add_tail_rcu(&drhd->list, &dmar_drhd_units);
  87         else
  88                 list_add_rcu(&drhd->list, &dmar_drhd_units);
  89 }
  90
  91 void *dmar_alloc_dev_scope(void *start, void *end, int *cnt)
  92 {
  93         struct acpi_dmar_device_scope *scope;
  94
  95         *cnt = 0;
  96         while (start < end) {
  97                 scope = start;
  98                 if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_NAMESPACE ||
  99                     scope->entry_type == ACPI_DMAR_SCOPE_TYPE_ENDPOINT ||
 100                     scope->entry_type == ACPI_DMAR_SCOPE_TYPE_BRIDGE)
 101                         (*cnt)++;
 102                 else if (scope->entry_type != ACPI_DMAR_SCOPE_TYPE_IOAPIC &&
 103                         scope->entry_type != ACPI_DMAR_SCOPE_TYPE_HPET) {
 104                         pr_warn("Unsupported device scope\n");
 105                 }
 106                 start += scope->length;
 107         }
 108         if (*cnt == 0)
 109                 return NULL;
 110
 111         return kcalloc(*cnt, sizeof(struct dmar_dev_scope), GFP_KERNEL);
 112 }
 113
 114 void dmar_free_dev_scope(struct dmar_dev_scope **devices, int *cnt)
 115 {
 116         int i;
 117         struct device *tmp_dev;
 118
 119         if (*devices && *cnt) {
 120                 for_each_active_dev_scope(*devices, *cnt, i, tmp_dev)
 121                         put_device(tmp_dev);
 122                 kfree(*devices);
 123         }
 124
 125         *devices = NULL;
 126         *cnt = 0;
 127 }
 128
 129 /* Optimize out kzalloc()/kfree() for normal cases */
 130 static char dmar_pci_notify_info_buf[64];
 131
 132 static struct dmar_pci_notify_info *
 133 dmar_alloc_pci_notify_info(struct pci_dev *dev, unsigned long event)
 134 {
 135         int level = 0;
 136         size_t size;
 137         struct pci_dev *tmp;
 138         struct dmar_pci_notify_info *info;
 139
 140         BUG_ON(dev->is_virtfn);
 141
 142         /* Only generate path[] for device addition event */
 143         if (event == BUS_NOTIFY_ADD_DEVICE)
 144                 for (tmp = dev; tmp; tmp = tmp->bus->self)
 145                         level++;
 146
 147         size = sizeof(*info) + level * sizeof(struct acpi_dmar_pci_path);
 148         if (size <= sizeof(dmar_pci_notify_info_buf)) {
 149                 info = (struct dmar_pci_notify_info *)dmar_pci_notify_info_buf;
 150         } else {
 151                 info = kzalloc(size, GFP_KERNEL);
 152                 if (!info) {
 153                         pr_warn("Out of memory when allocating notify_info "
 154                                 "for %s.\n", pci_name(dev));
 155                         if (dmar_dev_scope_status == 0)
 156                                 dmar_dev_scope_status = -ENOMEM;
 157                         return NULL;
 158                 }
 159         }
 160
 161         info->event = event;
 162         info->dev = dev;
 163         info->seg = pci_domain_nr(dev->bus);
 164         info->level = level;
 165         if (event == BUS_NOTIFY_ADD_DEVICE) {
 166                 for (tmp = dev; tmp; tmp = tmp->bus->self) {
 167                         level--;
 168                         info->path[level].bus = tmp->bus->number;
 169                         info->path[level].device = PCI_SLOT(tmp->devfn);
 170                         info->path[level].function = PCI_FUNC(tmp->devfn);
 171                         if (pci_is_root_bus(tmp->bus))
 172                                 info->bus = tmp->bus->number;
 173                 }
 174         }
 175
 176         return info;
 177 }
 178
 179 static inline void dmar_free_pci_notify_info(struct dmar_pci_notify_info *info)
 180 {
 181         if ((void *)info != dmar_pci_notify_info_buf)
 182                 kfree(info);
 183 }
 184
 185 static bool dmar_match_pci_path(struct dmar_pci_notify_info *info, int bus,
 186                                 struct acpi_dmar_pci_path *path, int count)
 187 {
 188         int i;
 189
 190         if (info->bus != bus)
 191                 goto fallback;
 192         if (info->level != count)
 193                 goto fallback;
 194
 195         for (i = 0; i < count; i++) {
 196                 if (path[i].device != info->path[i].device ||
 197                     path[i].function != info->path[i].function)
 198                         goto fallback;
 199         }
 200
 201         return true;
 202
 203 fallback:
 204
 205         if (count != 1)
 206                 return false;
 207
 208         i = info->level - 1;
 209         if (bus              == info->path[i].bus &&
 210             path[0].device   == info->path[i].device &&
 211             path[0].function == info->path[i].function) {
 212                 pr_info(FW_BUG "RMRR entry for device %02x:%02x.%x is broken - applying workaround\n",
 213                         bus, path[0].device, path[0].function);
 214                 return true;
 215         }
 216
 217         return false;
 218 }
 219
 220 /* Return: > 0 if match found, 0 if no match found, < 0 if error happens */
 221 int dmar_insert_dev_scope(struct dmar_pci_notify_info *info,
 222                           void *start, void*end, u16 segment,
 223                           struct dmar_dev_scope *devices,
 224                           int devices_cnt)
 225 {
 226         int i, level;
 227         struct device *tmp, *dev = &info->dev->dev;
 228         struct acpi_dmar_device_scope *scope;
 229         struct acpi_dmar_pci_path *path;
 230
 231         if (segment != info->seg)
 232                 return 0;
 233
 234         for (; start < end; start += scope->length) {
 235                 scope = start;
 236                 if (scope->entry_type != ACPI_DMAR_SCOPE_TYPE_ENDPOINT &&
 237                     scope->entry_type != ACPI_DMAR_SCOPE_TYPE_BRIDGE)
 238                         continue;
 239
 240                 path = (struct acpi_dmar_pci_path *)(scope + 1);
 241                 level = (scope->length - sizeof(*scope)) / sizeof(*path);
 242                 if (!dmar_match_pci_path(info, scope->bus, path, level))
 243                         continue;
 244
 245                 /*
 246                  * We expect devices with endpoint scope to have normal PCI
 247                  * headers, and devices with bridge scope to have bridge PCI
 248                  * headers.  However PCI NTB devices may be listed in the
 249                  * DMAR table with bridge scope, even though they have a
 250                  * normal PCI header.  NTB devices are identified by class
 251                  * "BRIDGE_OTHER" (0680h) - we don't declare a socpe mismatch
 252                  * for this special case.
 253                  */
 254                 if ((scope->entry_type == ACPI_DMAR_SCOPE_TYPE_ENDPOINT &&
 255                      info->dev->hdr_type != PCI_HEADER_TYPE_NORMAL) ||
 256                     (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_BRIDGE &&
 257                      (info->dev->hdr_type == PCI_HEADER_TYPE_NORMAL &&
 258                       info->dev->class >> 8 != PCI_CLASS_BRIDGE_OTHER))) {
 259                         pr_warn("Device scope type does not match for %s\n",
 260                                 pci_name(info->dev));
 261                         return -EINVAL;
 262                 }
 263
 264                 for_each_dev_scope(devices, devices_cnt, i, tmp)
 265                         if (tmp == NULL) {
 266                                 devices[i].bus = info->dev->bus->number;
 267                                 devices[i].devfn = info->dev->devfn;
 268                                 rcu_assign_pointer(devices[i].dev,
 269                                                    get_device(dev));
 270                                 return 1;
 271                         }
 272                 BUG_ON(i >= devices_cnt);
 273         }
 274
 275         return 0;
 276 }
 277
 278 int dmar_remove_dev_scope(struct dmar_pci_notify_info *info, u16 segment,
 279                           struct dmar_dev_scope *devices, int count)
 280 {
 281         int index;
 282         struct device *tmp;
 283
 284         if (info->seg != segment)
 285                 return 0;
 286
 287         for_each_active_dev_scope(devices, count, index, tmp)
 288                 if (tmp == &info->dev->dev) {
 289                         RCU_INIT_POINTER(devices[index].dev, NULL);
 290                         synchronize_rcu();
 291                         put_device(tmp);
 292                         return 1;
 293                 }
 294
 295         return 0;
 296 }
 297
 298 static int dmar_pci_bus_add_dev(struct dmar_pci_notify_info *info)
 299 {
 300         int ret = 0;
 301         struct dmar_drhd_unit *dmaru;
 302         struct acpi_dmar_hardware_unit *drhd;
 303
 304         for_each_drhd_unit(dmaru) {
 305                 if (dmaru->include_all)
 306                         continue;
 307
 308                 drhd = container_of(dmaru->hdr,
 309                                     struct acpi_dmar_hardware_unit, header);
 310                 ret = dmar_insert_dev_scope(info, (void *)(drhd + 1),
 311                                 ((void *)drhd) + drhd->header.length,
 312                                 dmaru->segment,
 313                                 dmaru->devices, dmaru->devices_cnt);
 314                 if (ret != 0)
 315                         break;
 316         }
 317         if (ret >= 0)
 318                 ret = dmar_iommu_notify_scope_dev(info);
 319         if (ret < 0 && dmar_dev_scope_status == 0)
 320                 dmar_dev_scope_status = ret;
 321
 322         return ret;
 323 }
 324
 325 static void  dmar_pci_bus_del_dev(struct dmar_pci_notify_info *info)
 326 {
 327         struct dmar_drhd_unit *dmaru;
 328
 329         for_each_drhd_unit(dmaru)
 330                 if (dmar_remove_dev_scope(info, dmaru->segment,
 331                         dmaru->devices, dmaru->devices_cnt))
 332                         break;
 333         dmar_iommu_notify_scope_dev(info);
 334 }
 335
 336 static int dmar_pci_bus_notifier(struct notifier_block *nb,
 337                                  unsigned long action, void *data)
 338 {
 339         struct pci_dev *pdev = to_pci_dev(data);
 340         struct dmar_pci_notify_info *info;
 341
 342         /* Only care about add/remove events for physical functions.
 343          * For VFs we actually do the lookup based on the corresponding
 344          * PF in device_to_iommu() anyway. */
 345         if (pdev->is_virtfn)
 346                 return NOTIFY_DONE;
 347         if (action != BUS_NOTIFY_ADD_DEVICE &&
 348             action != BUS_NOTIFY_REMOVED_DEVICE)
 349                 return NOTIFY_DONE;
 350
 351         info = dmar_alloc_pci_notify_info(pdev, action);
 352         if (!info)
 353                 return NOTIFY_DONE;
 354
 355         down_write(&dmar_global_lock);
 356         if (action == BUS_NOTIFY_ADD_DEVICE)
 357                 dmar_pci_bus_add_dev(info);
 358         else if (action == BUS_NOTIFY_REMOVED_DEVICE)
 359                 dmar_pci_bus_del_dev(info);
 360         up_write(&dmar_global_lock);
 361
 362         dmar_free_pci_notify_info(info);
 363
 364         return NOTIFY_OK;
 365 }
 366
 367 static struct notifier_block dmar_pci_bus_nb = {
 368         .notifier_call = dmar_pci_bus_notifier,
 369         .priority = INT_MIN,
 370 };
 371
 372 static struct dmar_drhd_unit *
 373 dmar_find_dmaru(struct acpi_dmar_hardware_unit *drhd)
 374 {
 375         struct dmar_drhd_unit *dmaru;
 376
 377         list_for_each_entry_rcu(dmaru, &dmar_drhd_units, list)
 378                 if (dmaru->segment == drhd->segment &&
 379                     dmaru->reg_base_addr == drhd->address)
 380                         return dmaru;
 381
 382         return NULL;
 383 }
 384
 385 /**
 386  * dmar_parse_one_drhd - parses exactly one DMA remapping hardware definition
 387  * structure which uniquely represent one DMA remapping hardware unit
 388  * present in the platform
 389  */
 390 static int dmar_parse_one_drhd(struct acpi_dmar_header *header, void *arg)
 391 {
 392         struct acpi_dmar_hardware_unit *drhd;
 393         struct dmar_drhd_unit *dmaru;
 394         int ret = 0;
 395
 396         drhd = (struct acpi_dmar_hardware_unit *)header;
 397         dmaru = dmar_find_dmaru(drhd);
 398         if (dmaru)
 399                 goto out;
 400
 401         dmaru = kzalloc(sizeof(*dmaru) + header->length, GFP_KERNEL);
 402         if (!dmaru)
 403                 return -ENOMEM;
 404
 405         /*
 406          * If header is allocated from slab by ACPI _DSM method, we need to
 407          * copy the content because the memory buffer will be freed on return.
 408          */
 409         dmaru->hdr = (void *)(dmaru + 1);
 410         memcpy(dmaru->hdr, header, header->length);
 411         dmaru->reg_base_addr = drhd->address;
 412         dmaru->segment = drhd->segment;
 413         dmaru->include_all = drhd->flags & 0x1; /* BIT0: INCLUDE_ALL */
 414         dmaru->devices = dmar_alloc_dev_scope((void *)(drhd + 1),
 415                                               ((void *)drhd) + drhd->header.length,
 416                                               &dmaru->devices_cnt);
 417         if (dmaru->devices_cnt && dmaru->devices == NULL) {
 418                 kfree(dmaru);
 419                 return -ENOMEM;
 420         }
 421
 422         ret = alloc_iommu(dmaru);
 423         if (ret) {
 424                 dmar_free_dev_scope(&dmaru->devices,
 425                                     &dmaru->devices_cnt);
 426                 kfree(dmaru);
 427                 return ret;
 428         }
 429         dmar_register_drhd_unit(dmaru);
 430
 431 out:
 432         if (arg)
 433                 (*(int *)arg)++;
 434
 435         return 0;
 436 }
 437
 438 static void dmar_free_drhd(struct dmar_drhd_unit *dmaru)
 439 {
 440         if (dmaru->devices && dmaru->devices_cnt)
 441                 dmar_free_dev_scope(&dmaru->devices, &dmaru->devices_cnt);
 442         if (dmaru->iommu)
 443                 free_iommu(dmaru->iommu);
 444         kfree(dmaru);
 445 }
 446
 447 static int __init dmar_parse_one_andd(struct acpi_dmar_header *header,
 448                                       void *arg)
 449 {
 450         struct acpi_dmar_andd *andd = (void *)header;
 451
 452         /* Check for NUL termination within the designated length */
 453         if (strnlen(andd->device_name, header->length - 8) == header->length - 8) {
 454                 WARN_TAINT(1, TAINT_FIRMWARE_WORKAROUND,
 455                            "Your BIOS is broken; ANDD object name is not NUL-terminated\n"
 456                            "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
 457                            dmi_get_system_info(DMI_BIOS_VENDOR),
 458                            dmi_get_system_info(DMI_BIOS_VERSION),
 459                            dmi_get_system_info(DMI_PRODUCT_VERSION));
 460                 return -EINVAL;
 461         }
 462         pr_info("ANDD device: %x name: %s\n", andd->device_number,
 463                 andd->device_name);
 464
 465         return 0;
 466 }
 467
 468 #ifdef CONFIG_ACPI_NUMA
 469 static int dmar_parse_one_rhsa(struct acpi_dmar_header *header, void *arg)
 470 {
 471         struct acpi_dmar_rhsa *rhsa;
 472         struct dmar_drhd_unit *drhd;
 473
 474         rhsa = (struct acpi_dmar_rhsa *)header;
 475         for_each_drhd_unit(drhd) {
 476                 if (drhd->reg_base_addr == rhsa->base_address) {
 477                         int node = acpi_map_pxm_to_node(rhsa->proximity_domain);
 478
 479                         if (!node_online(node))
 480                                 node = -1;
 481                         drhd->iommu->node = node;
 482                         return 0;
 483                 }
 484         }
 485         WARN_TAINT(
 486                 1, TAINT_FIRMWARE_WORKAROUND,
 487                 "Your BIOS is broken; RHSA refers to non-existent DMAR unit at %llx\n"
 488                 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
 489                 drhd->reg_base_addr,
 490                 dmi_get_system_info(DMI_BIOS_VENDOR),
 491                 dmi_get_system_info(DMI_BIOS_VERSION),
 492                 dmi_get_system_info(DMI_PRODUCT_VERSION));
 493
 494         return 0;
 495 }
 496 #else
 497 #define dmar_parse_one_rhsa             dmar_res_noop
 498 #endif
 499
 500 static void __init
 501 dmar_table_print_dmar_entry(struct acpi_dmar_header *header)
 502 {
 503         struct acpi_dmar_hardware_unit *drhd;
 504         struct acpi_dmar_reserved_memory *rmrr;
 505         struct acpi_dmar_atsr *atsr;
 506         struct acpi_dmar_rhsa *rhsa;
 507
 508         switch (header->type) {
 509         case ACPI_DMAR_TYPE_HARDWARE_UNIT:
 510                 drhd = container_of(header, struct acpi_dmar_hardware_unit,
 511                                     header);
 512                 pr_info("DRHD base: %#016Lx flags: %#x\n",
 513                         (unsigned long long)drhd->address, drhd->flags);
 514                 break;
 515         case ACPI_DMAR_TYPE_RESERVED_MEMORY:
 516                 rmrr = container_of(header, struct acpi_dmar_reserved_memory,
 517                                     header);
 518                 pr_info("RMRR base: %#016Lx end: %#016Lx\n",
 519                         (unsigned long long)rmrr->base_address,
 520                         (unsigned long long)rmrr->end_address);
 521                 break;
 522         case ACPI_DMAR_TYPE_ROOT_ATS:
 523                 atsr = container_of(header, struct acpi_dmar_atsr, header);
 524                 pr_info("ATSR flags: %#x\n", atsr->flags);
 525                 break;
 526         case ACPI_DMAR_TYPE_HARDWARE_AFFINITY:
 527                 rhsa = container_of(header, struct acpi_dmar_rhsa, header);
 528                 pr_info("RHSA base: %#016Lx proximity domain: %#x\n",
 529                        (unsigned long long)rhsa->base_address,
 530                        rhsa->proximity_domain);
 531                 break;
 532         case ACPI_DMAR_TYPE_NAMESPACE:
 533                 /* We don't print this here because we need to sanity-check
 534                    it first. So print it in dmar_parse_one_andd() instead. */
 535                 break;
 536         }
 537 }
 538
 539 /**
 540  * dmar_table_detect - checks to see if the platform supports DMAR devices
 541  */
 542 static int __init dmar_table_detect(void)
 543 {
 544         acpi_status status = AE_OK;
 545
 546         /* if we could find DMAR table, then there are DMAR devices */
 547         status = acpi_get_table(ACPI_SIG_DMAR, 0, &dmar_tbl);
 548
 549         if (ACPI_SUCCESS(status) && !dmar_tbl) {
 550                 pr_warn("Unable to map DMAR\n");
 551                 status = AE_NOT_FOUND;
 552         }
 553
 554         return (ACPI_SUCCESS(status) ? 1 : 0);
 555 }
 556
 557 static int dmar_walk_remapping_entries(struct acpi_dmar_header *start,
 558                                        size_t len, struct dmar_res_callback *cb)
 559 {
 560         int ret = 0;
 561         struct acpi_dmar_header *iter, *next;
 562         struct acpi_dmar_header *end = ((void *)start) + len;
 563
 564         for (iter = start; iter < end && ret == 0; iter = next) {
 565                 next = (void *)iter + iter->length;
 566                 if (iter->length == 0) {
 567                         /* Avoid looping forever on bad ACPI tables */
 568                         pr_debug(FW_BUG "Invalid 0-length structure\n");
 569                         break;
 570                 } else if (next > end) {
 571                         /* Avoid passing table end */
 572                         pr_warn(FW_BUG "Record passes table end\n");
 573                         ret = -EINVAL;
 574                         break;
 575                 }
 576
 577                 if (cb->print_entry)
 578                         dmar_table_print_dmar_entry(iter);
 579
 580                 if (iter->type >= ACPI_DMAR_TYPE_RESERVED) {
 581                         /* continue for forward compatibility */
 582                         pr_debug("Unknown DMAR structure type %d\n",
 583                                  iter->type);
 584                 } else if (cb->cb[iter->type]) {
 585                         ret = cb->cb[iter->type](iter, cb->arg[iter->type]);
 586                 } else if (!cb->ignore_unhandled) {
 587                         pr_warn("No handler for DMAR structure type %d\n",
 588                                 iter->type);
 589                         ret = -EINVAL;
 590                 }
 591         }
 592
 593         return ret;
 594 }
 595
 596 static inline int dmar_walk_dmar_table(struct acpi_table_dmar *dmar,
 597                                        struct dmar_res_callback *cb)
 598 {
 599         return dmar_walk_remapping_entries((void *)(dmar + 1),
 600                         dmar->header.length - sizeof(*dmar), cb);
 601 }
 602
 603 /**
 604  * parse_dmar_table - parses the DMA reporting table
 605  */
 606 static int __init
 607 parse_dmar_table(void)
 608 {
 609         struct acpi_table_dmar *dmar;
 610         int ret = 0;
 611         int drhd_count = 0;
 612         struct dmar_res_callback cb = {
 613                 .print_entry = true,
 614                 .ignore_unhandled = true,
 615                 .arg[ACPI_DMAR_TYPE_HARDWARE_UNIT] = &drhd_count,
 616                 .cb[ACPI_DMAR_TYPE_HARDWARE_UNIT] = &dmar_parse_one_drhd,
 617                 .cb[ACPI_DMAR_TYPE_RESERVED_MEMORY] = &dmar_parse_one_rmrr,
 618                 .cb[ACPI_DMAR_TYPE_ROOT_ATS] = &dmar_parse_one_atsr,
 619                 .cb[ACPI_DMAR_TYPE_HARDWARE_AFFINITY] = &dmar_parse_one_rhsa,
 620                 .cb[ACPI_DMAR_TYPE_NAMESPACE] = &dmar_parse_one_andd,
 621         };
 622
 623         /*
 624          * Do it again, earlier dmar_tbl mapping could be mapped with
 625          * fixed map.
 626          */
 627         dmar_table_detect();
 628
 629         /*
 630          * ACPI tables may not be DMA protected by tboot, so use DMAR copy
 631          * SINIT saved in SinitMleData in TXT heap (which is DMA protected)
 632          */
 633         dmar_tbl = tboot_get_dmar_table(dmar_tbl);
 634
 635         dmar = (struct acpi_table_dmar *)dmar_tbl;
 636         if (!dmar)
 637                 return -ENODEV;
 638
 639         if (dmar->width < PAGE_SHIFT - 1) {
 640                 pr_warn("Invalid DMAR haw\n");
 641                 return -EINVAL;
 642         }
 643
 644         pr_info("Host address width %d\n", dmar->width + 1);
 645         ret = dmar_walk_dmar_table(dmar, &cb);
 646         if (ret == 0 && drhd_count == 0)
 647                 pr_warn(FW_BUG "No DRHD structure found in DMAR table\n");
 648
 649         return ret;
 650 }
 651
 652 static int dmar_pci_device_match(struct dmar_dev_scope devices[],
 653                                  int cnt, struct pci_dev *dev)
 654 {
 655         int index;
 656         struct device *tmp;
 657
 658         while (dev) {
 659                 for_each_active_dev_scope(devices, cnt, index, tmp)
 660                         if (dev_is_pci(tmp) && dev == to_pci_dev(tmp))
 661                                 return 1;
 662
 663                 /* Check our parent */
 664                 dev = dev->bus->self;
 665         }
 666
 667         return 0;
 668 }
 669
 670 struct dmar_drhd_unit *
 671 dmar_find_matched_drhd_unit(struct pci_dev *dev)
 672 {
 673         struct dmar_drhd_unit *dmaru;
 674         struct acpi_dmar_hardware_unit *drhd;
 675
 676         dev = pci_physfn(dev);
 677
 678         rcu_read_lock();
 679         for_each_drhd_unit(dmaru) {
 680                 drhd = container_of(dmaru->hdr,
 681                                     struct acpi_dmar_hardware_unit,
 682                                     header);
 683
 684                 if (dmaru->include_all &&
 685                     drhd->segment == pci_domain_nr(dev->bus))
 686                         goto out;
 687
 688                 if (dmar_pci_device_match(dmaru->devices,
 689                                           dmaru->devices_cnt, dev))
 690                         goto out;
 691         }
 692         dmaru = NULL;
 693 out:
 694         rcu_read_unlock();
 695
 696         return dmaru;
 697 }
 698
 699 static void __init dmar_acpi_insert_dev_scope(u8 device_number,
 700                                               struct acpi_device *adev)
 701 {
 702         struct dmar_drhd_unit *dmaru;
 703         struct acpi_dmar_hardware_unit *drhd;
 704         struct acpi_dmar_device_scope *scope;
 705         struct device *tmp;
 706         int i;
 707         struct acpi_dmar_pci_path *path;
 708
 709         for_each_drhd_unit(dmaru) {
 710                 drhd = container_of(dmaru->hdr,
 711                                     struct acpi_dmar_hardware_unit,
 712                                     header);
 713
 714                 for (scope = (void *)(drhd + 1);
 715                      (unsigned long)scope < ((unsigned long)drhd) + drhd->header.length;
 716                      scope = ((void *)scope) + scope->length) {
 717                         if (scope->entry_type != ACPI_DMAR_SCOPE_TYPE_NAMESPACE)
 718                                 continue;
 719                         if (scope->enumeration_id != device_number)
 720                                 continue;
 721
 722                         path = (void *)(scope + 1);
 723                         pr_info("ACPI device \"%s\" under DMAR at %llx as %02x:%02x.%d\n",
 724                                 dev_name(&adev->dev), dmaru->reg_base_addr,
 725                                 scope->bus, path->device, path->function);
 726                         for_each_dev_scope(dmaru->devices, dmaru->devices_cnt, i, tmp)
 727                                 if (tmp == NULL) {
 728                                         dmaru->devices[i].bus = scope->bus;
 729                                         dmaru->devices[i].devfn = PCI_DEVFN(path->device,
 730                                                                             path->function);
 731                                         rcu_assign_pointer(dmaru->devices[i].dev,
 732                                                            get_device(&adev->dev));
 733                                         return;
 734                                 }
 735                         BUG_ON(i >= dmaru->devices_cnt);
 736                 }
 737         }
 738         pr_warn("No IOMMU scope found for ANDD enumeration ID %d (%s)\n",
 739                 device_number, dev_name(&adev->dev));
 740 }
 741
 742 static int __init dmar_acpi_dev_scope_init(void)
 743 {
 744         struct acpi_dmar_andd *andd;
 745
 746         if (dmar_tbl == NULL)
 747                 return -ENODEV;
 748
 749         for (andd = (void *)dmar_tbl + sizeof(struct acpi_table_dmar);
 750              ((unsigned long)andd) < ((unsigned long)dmar_tbl) + dmar_tbl->length;
 751              andd = ((void *)andd) + andd->header.length) {
 752                 if (andd->header.type == ACPI_DMAR_TYPE_NAMESPACE) {
 753                         acpi_handle h;
 754                         struct acpi_device *adev;
 755
 756                         if (!ACPI_SUCCESS(acpi_get_handle(ACPI_ROOT_OBJECT,
 757                                                           andd->device_name,
 758                                                           &h))) {
 759                                 pr_err("Failed to find handle for ACPI object %s\n",
 760                                        andd->device_name);
 761                                 continue;
 762                         }
 763                         if (acpi_bus_get_device(h, &adev)) {
 764                                 pr_err("Failed to get device for ACPI object %s\n",
 765                                        andd->device_name);
 766                                 continue;
 767                         }
 768                         dmar_acpi_insert_dev_scope(andd->device_number, adev);
 769                 }
 770         }
 771         return 0;
 772 }
 773
 774 int __init dmar_dev_scope_init(void)
 775 {
 776         struct pci_dev *dev = NULL;
 777         struct dmar_pci_notify_info *info;
 778
 779         if (dmar_dev_scope_status != 1)
 780                 return dmar_dev_scope_status;
 781
 782         if (list_empty(&dmar_drhd_units)) {
 783                 dmar_dev_scope_status = -ENODEV;
 784         } else {
 785                 dmar_dev_scope_status = 0;
 786
 787                 dmar_acpi_dev_scope_init();
 788
 789                 for_each_pci_dev(dev) {
 790                         if (dev->is_virtfn)
 791                                 continue;
 792
 793                         info = dmar_alloc_pci_notify_info(dev,
 794                                         BUS_NOTIFY_ADD_DEVICE);
 795                         if (!info) {
 796                                 return dmar_dev_scope_status;
 797                         } else {
 798                                 dmar_pci_bus_add_dev(info);
 799                                 dmar_free_pci_notify_info(info);
 800                         }
 801                 }
 802
 803                 bus_register_notifier(&pci_bus_type, &dmar_pci_bus_nb);
 804         }
 805
 806         return dmar_dev_scope_status;
 807 }
 808
 809
 810 int __init dmar_table_init(void)
 811 {
 812         static int dmar_table_initialized;
 813         int ret;
 814
 815         if (dmar_table_initialized == 0) {
 816                 ret = parse_dmar_table();
 817                 if (ret < 0) {
 818                         if (ret != -ENODEV)
 819                                 pr_info("Parse DMAR table failure.\n");
 820                 } else  if (list_empty(&dmar_drhd_units)) {
 821                         pr_info("No DMAR devices found\n");
 822                         ret = -ENODEV;
 823                 }
 824
 825                 if (ret < 0)
 826                         dmar_table_initialized = ret;
 827                 else
 828                         dmar_table_initialized = 1;
 829         }
 830
 831         return dmar_table_initialized < 0 ? dmar_table_initialized : 0;
 832 }
 833
 834 static void warn_invalid_dmar(u64 addr, const char *message)
 835 {
 836         WARN_TAINT_ONCE(
 837                 1, TAINT_FIRMWARE_WORKAROUND,
 838                 "Your BIOS is broken; DMAR reported at address %llx%s!\n"
 839                 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
 840                 addr, message,
 841                 dmi_get_system_info(DMI_BIOS_VENDOR),
 842                 dmi_get_system_info(DMI_BIOS_VERSION),
 843                 dmi_get_system_info(DMI_PRODUCT_VERSION));
 844 }
 845
 846 static int __ref
 847 dmar_validate_one_drhd(struct acpi_dmar_header *entry, void *arg)
 848 {
 849         struct acpi_dmar_hardware_unit *drhd;
 850         void __iomem *addr;
 851         u64 cap, ecap;
 852
 853         drhd = (void *)entry;
 854         if (!drhd->address) {
 855                 warn_invalid_dmar(0, "");
 856                 return -EINVAL;
 857         }
 858
 859         if (arg)
 860                 addr = ioremap(drhd->address, VTD_PAGE_SIZE);
 861         else
 862                 addr = early_ioremap(drhd->address, VTD_PAGE_SIZE);
 863         if (!addr) {
 864                 pr_warn("Can't validate DRHD address: %llx\n", drhd->address);
 865                 return -EINVAL;
 866         }
 867
 868         cap = dmar_readq(addr + DMAR_CAP_REG);
 869         ecap = dmar_readq(addr + DMAR_ECAP_REG);
 870
 871         if (arg)
 872                 iounmap(addr);
 873         else
 874                 early_iounmap(addr, VTD_PAGE_SIZE);
 875
 876         if (cap == (uint64_t)-1 && ecap == (uint64_t)-1) {
 877                 warn_invalid_dmar(drhd->address, " returns all ones");
 878                 return -EINVAL;
 879         }
 880
 881         return 0;
 882 }
 883
 884 int __init detect_intel_iommu(void)
 885 {
 886         int ret;
 887         struct dmar_res_callback validate_drhd_cb = {
 888                 .cb[ACPI_DMAR_TYPE_HARDWARE_UNIT] = &dmar_validate_one_drhd,
 889                 .ignore_unhandled = true,
 890         };
 891
 892         down_write(&dmar_global_lock);
 893         ret = dmar_table_detect();
 894         if (ret)
 895                 ret = !dmar_walk_dmar_table((struct acpi_table_dmar *)dmar_tbl,
 896                                             &validate_drhd_cb);
 897         if (ret && !no_iommu && !iommu_detected && !dmar_disabled) {
 898                 iommu_detected = 1;
 899                 /* Make sure ACS will be enabled */
 900                 pci_request_acs();
 901         }
 902
 903 #ifdef CONFIG_X86
 904         if (ret)
 905                 x86_init.iommu.iommu_init = intel_iommu_init;
 906 #endif
 907
 908         if (dmar_tbl) {
 909                 acpi_put_table(dmar_tbl);
 910                 dmar_tbl = NULL;
 911         }
 912         up_write(&dmar_global_lock);
 913
 914         return ret ? 1 : -ENODEV;
 915 }
 916
 917
 918 static void unmap_iommu(struct intel_iommu *iommu)
 919 {
 920         iounmap(iommu->reg);
 921         release_mem_region(iommu->reg_phys, iommu->reg_size);
 922 }
 923
 924 /**
 925  * map_iommu: map the iommu's registers
 926  * @iommu: the iommu to map
 927  * @phys_addr: the physical address of the base resgister
 928  *
 929  * Memory map the iommu's registers.  Start w/ a single page, and
 930  * possibly expand if that turns out to be insufficent.
 931  */
 932 static int map_iommu(struct intel_iommu *iommu, u64 phys_addr)
 933 {
 934         int map_size, err=0;
 935
 936         iommu->reg_phys = phys_addr;
 937         iommu->reg_size = VTD_PAGE_SIZE;
 938
 939         if (!request_mem_region(iommu->reg_phys, iommu->reg_size, iommu->name)) {
 940                 pr_err("Can't reserve memory\n");
 941                 err = -EBUSY;
 942                 goto out;
 943         }
 944
 945         iommu->reg = ioremap(iommu->reg_phys, iommu->reg_size);
 946         if (!iommu->reg) {
 947                 pr_err("Can't map the region\n");
 948                 err = -ENOMEM;
 949                 goto release;
 950         }
 951
 952         iommu->cap = dmar_readq(iommu->reg + DMAR_CAP_REG);
 953         iommu->ecap = dmar_readq(iommu->reg + DMAR_ECAP_REG);
 954
 955         if (iommu->cap == (uint64_t)-1 && iommu->ecap == (uint64_t)-1) {
 956                 err = -EINVAL;
 957                 warn_invalid_dmar(phys_addr, " returns all ones");
 958                 goto unmap;
 959         }
 960
 961         /* the registers might be more than one page */
 962         map_size = max_t(int, ecap_max_iotlb_offset(iommu->ecap),
 963                          cap_max_fault_reg_offset(iommu->cap));
 964         map_size = VTD_PAGE_ALIGN(map_size);
 965         if (map_size > iommu->reg_size) {
 966                 iounmap(iommu->reg);
 967                 release_mem_region(iommu->reg_phys, iommu->reg_size);
 968                 iommu->reg_size = map_size;
 969                 if (!request_mem_region(iommu->reg_phys, iommu->reg_size,
 970                                         iommu->name)) {
 971                         pr_err("Can't reserve memory\n");
 972                         err = -EBUSY;
 973                         goto out;
 974                 }
 975                 iommu->reg = ioremap(iommu->reg_phys, iommu->reg_size);
 976                 if (!iommu->reg) {
 977                         pr_err("Can't map the region\n");
 978                         err = -ENOMEM;
 979                         goto release;
 980                 }
 981         }
 982         err = 0;
 983         goto out;
 984
 985 unmap:
 986         iounmap(iommu->reg);
 987 release:
 988         release_mem_region(iommu->reg_phys, iommu->reg_size);
 989 out:
 990         return err;
 991 }
 992
 993 static int dmar_alloc_seq_id(struct intel_iommu *iommu)
 994 {
 995         iommu->seq_id = find_first_zero_bit(dmar_seq_ids,
 996                                             DMAR_UNITS_SUPPORTED);
 997         if (iommu->seq_id >= DMAR_UNITS_SUPPORTED) {
 998                 iommu->seq_id = -1;
 999         } else {
1000                 set_bit(iommu->seq_id, dmar_seq_ids);
1001                 sprintf(iommu->name, "dmar%d", iommu->seq_id);
1002         }
1003
1004         return iommu->seq_id;
1005 }
1006
1007 static void dmar_free_seq_id(struct intel_iommu *iommu)
1008 {
1009         if (iommu->seq_id >= 0) {
1010                 clear_bit(iommu->seq_id, dmar_seq_ids);
1011                 iommu->seq_id = -1;
1012         }
1013 }
1014
1015 static int alloc_iommu(struct dmar_drhd_unit *drhd)
1016 {
1017         struct intel_iommu *iommu;
1018         u32 ver, sts;
1019         int agaw = 0;
1020         int msagaw = 0;
1021         int err;
1022
1023         if (!drhd->reg_base_addr) {
1024                 warn_invalid_dmar(0, "");
1025                 return -EINVAL;
1026         }
1027
1028         iommu = kzalloc(sizeof(*iommu), GFP_KERNEL);
1029         if (!iommu)
1030                 return -ENOMEM;
1031
1032         if (dmar_alloc_seq_id(iommu) < 0) {
1033                 pr_err("Failed to allocate seq_id\n");
1034                 err = -ENOSPC;
1035                 goto error;
1036         }
1037
1038         err = map_iommu(iommu, drhd->reg_base_addr);
1039         if (err) {
1040                 pr_err("Failed to map %s\n", iommu->name);
1041                 goto error_free_seq_id;
1042         }
1043
1044         err = -EINVAL;
1045         agaw = iommu_calculate_agaw(iommu);
1046         if (agaw < 0) {
1047                 pr_err("Cannot get a valid agaw for iommu (seq_id = %d)\n",
1048                         iommu->seq_id);
1049                 goto err_unmap;
1050         }
1051         msagaw = iommu_calculate_max_sagaw(iommu);
1052         if (msagaw < 0) {
1053                 pr_err("Cannot get a valid max agaw for iommu (seq_id = %d)\n",
1054                         iommu->seq_id);
1055                 goto err_unmap;
1056         }
1057         iommu->agaw = agaw;
1058         iommu->msagaw = msagaw;
1059         iommu->segment = drhd->segment;
1060
1061         iommu->node = -1;
1062
1063         ver = readl(iommu->reg + DMAR_VER_REG);
1064         pr_info("%s: reg_base_addr %llx ver %d:%d cap %llx ecap %llx\n",
1065                 iommu->name,
1066                 (unsigned long long)drhd->reg_base_addr,
1067                 DMAR_VER_MAJOR(ver), DMAR_VER_MINOR(ver),
1068                 (unsigned long long)iommu->cap,
1069                 (unsigned long long)iommu->ecap);
1070
1071         /* Reflect status in gcmd */
1072         sts = readl(iommu->reg + DMAR_GSTS_REG);
1073         if (sts & DMA_GSTS_IRES)
1074                 iommu->gcmd |= DMA_GCMD_IRE;
1075         if (sts & DMA_GSTS_TES)
1076                 iommu->gcmd |= DMA_GCMD_TE;
1077         if (sts & DMA_GSTS_QIES)
1078                 iommu->gcmd |= DMA_GCMD_QIE;
1079
1080         raw_spin_lock_init(&iommu->register_lock);
1081
1082         if (intel_iommu_enabled) {
1083                 err = iommu_device_sysfs_add(&iommu->iommu, NULL,
1084                                              intel_iommu_groups,
1085                                              "%s", iommu->name);
1086                 if (err)
1087                         goto err_unmap;
1088
1089                 iommu_device_set_ops(&iommu->iommu, &intel_iommu_ops);
1090
1091                 err = iommu_device_register(&iommu->iommu);
1092                 if (err)
1093                         goto err_unmap;
1094         }
1095
1096         drhd->iommu = iommu;
1097
1098         return 0;
1099
1100 err_unmap:
1101         unmap_iommu(iommu);
1102 error_free_seq_id:
1103         dmar_free_seq_id(iommu);
1104 error:
1105         kfree(iommu);
1106         return err;
1107 }
1108
1109 static void free_iommu(struct intel_iommu *iommu)
1110 {
1111         iommu_device_sysfs_remove(&iommu->iommu);
1112         iommu_device_unregister(&iommu->iommu);
1113
1114         if (iommu->irq) {
1115                 if (iommu->pr_irq) {
1116                         free_irq(iommu->pr_irq, iommu);
1117                         dmar_free_hwirq(iommu->pr_irq);
1118                         iommu->pr_irq = 0;
1119                 }
1120                 free_irq(iommu->irq, iommu);
1121                 dmar_free_hwirq(iommu->irq);
1122                 iommu->irq = 0;
1123         }
1124
1125         if (iommu->qi) {
1126                 free_page((unsigned long)iommu->qi->desc);
1127                 kfree(iommu->qi->desc_status);
1128                 kfree(iommu->qi);
1129         }
1130
1131         if (iommu->reg)
1132                 unmap_iommu(iommu);
1133
1134         dmar_free_seq_id(iommu);
1135         kfree(iommu);
1136 }
1137
1138 /*
1139  * Reclaim all the submitted descriptors which have completed its work.
1140  */
1141 static inline void reclaim_free_desc(struct q_inval *qi)
1142 {
1143         while (qi->desc_status[qi->free_tail] == QI_DONE ||
1144                qi->desc_status[qi->free_tail] == QI_ABORT) {
1145                 qi->desc_status[qi->free_tail] = QI_FREE;
1146                 qi->free_tail = (qi->free_tail + 1) % QI_LENGTH;
1147                 qi->free_cnt++;
1148         }
1149 }
1150
1151 static int qi_check_fault(struct intel_iommu *iommu, int index)
1152 {
1153         u32 fault;
1154         int head, tail;
1155         struct q_inval *qi = iommu->qi;
1156         int wait_index = (index + 1) % QI_LENGTH;
1157
1158         if (qi->desc_status[wait_index] == QI_ABORT)
1159                 return -EAGAIN;
1160
1161         fault = readl(iommu->reg + DMAR_FSTS_REG);
1162
1163         /*
1164          * If IQE happens, the head points to the descriptor associated
1165          * with the error. No new descriptors are fetched until the IQE
1166          * is cleared.
1167          */
1168         if (fault & DMA_FSTS_IQE) {
1169                 head = readl(iommu->reg + DMAR_IQH_REG);
1170                 if ((head >> DMAR_IQ_SHIFT) == index) {
1171                         pr_err("VT-d detected invalid descriptor: "
1172                                 "low=%llx, high=%llx\n",
1173                                 (unsigned long long)qi->desc[index].low,
1174                                 (unsigned long long)qi->desc[index].high);
1175                         memcpy(&qi->desc[index], &qi->desc[wait_index],
1176                                         sizeof(struct qi_desc));
1177                         writel(DMA_FSTS_IQE, iommu->reg + DMAR_FSTS_REG);
1178                         return -EINVAL;
1179                 }
1180         }
1181
1182         /*
1183          * If ITE happens, all pending wait_desc commands are aborted.
1184          * No new descriptors are fetched until the ITE is cleared.
1185          */
1186         if (fault & DMA_FSTS_ITE) {
1187                 head = readl(iommu->reg + DMAR_IQH_REG);
1188                 head = ((head >> DMAR_IQ_SHIFT) - 1 + QI_LENGTH) % QI_LENGTH;
1189                 head |= 1;
1190                 tail = readl(iommu->reg + DMAR_IQT_REG);
1191                 tail = ((tail >> DMAR_IQ_SHIFT) - 1 + QI_LENGTH) % QI_LENGTH;
1192
1193                 writel(DMA_FSTS_ITE, iommu->reg + DMAR_FSTS_REG);
1194
1195                 do {
1196                         if (qi->desc_status[head] == QI_IN_USE)
1197                                 qi->desc_status[head] = QI_ABORT;
1198                         head = (head - 2 + QI_LENGTH) % QI_LENGTH;
1199                 } while (head != tail);
1200
1201                 if (qi->desc_status[wait_index] == QI_ABORT)
1202                         return -EAGAIN;
1203         }
1204
1205         if (fault & DMA_FSTS_ICE)
1206                 writel(DMA_FSTS_ICE, iommu->reg + DMAR_FSTS_REG);
1207
1208         return 0;
1209 }
1210
1211 /*
1212  * Submit the queued invalidation descriptor to the remapping
1213  * hardware unit and wait for its completion.
1214  */
1215 int qi_submit_sync(struct qi_desc *desc, struct intel_iommu *iommu)
1216 {
1217         int rc;
1218         struct q_inval *qi = iommu->qi;
1219         struct qi_desc *hw, wait_desc;
1220         int wait_index, index;
1221         unsigned long flags;
1222
1223         if (!qi)
1224                 return 0;
1225
1226         hw = qi->desc;
1227
1228 restart:
1229         rc = 0;
1230
1231         raw_spin_lock_irqsave(&qi->q_lock, flags);
1232         while (qi->free_cnt < 3) {
1233                 raw_spin_unlock_irqrestore(&qi->q_lock, flags);
1234                 cpu_relax();
1235                 raw_spin_lock_irqsave(&qi->q_lock, flags);
1236         }
1237
1238         index = qi->free_head;
1239         wait_index = (index + 1) % QI_LENGTH;
1240
1241         qi->desc_status[index] = qi->desc_status[wait_index] = QI_IN_USE;
1242
1243         hw[index] = *desc;
1244
1245         wait_desc.low = QI_IWD_STATUS_DATA(QI_DONE) |
1246                         QI_IWD_STATUS_WRITE | QI_IWD_TYPE;
1247         wait_desc.high = virt_to_phys(&qi->desc_status[wait_index]);
1248
1249         hw[wait_index] = wait_desc;
1250
1251         qi->free_head = (qi->free_head + 2) % QI_LENGTH;
1252         qi->free_cnt -= 2;
1253
1254         /*
1255          * update the HW tail register indicating the presence of
1256          * new descriptors.
1257          */
1258         writel(qi->free_head << DMAR_IQ_SHIFT, iommu->reg + DMAR_IQT_REG);
1259
1260         while (qi->desc_status[wait_index] != QI_DONE) {
1261                 /*
1262                  * We will leave the interrupts disabled, to prevent interrupt
1263                  * context to queue another cmd while a cmd is already submitted
1264                  * and waiting for completion on this cpu. This is to avoid
1265                  * a deadlock where the interrupt context can wait indefinitely
1266                  * for free slots in the queue.
1267                  */
1268                 rc = qi_check_fault(iommu, index);
1269                 if (rc)
1270                         break;
1271
1272                 raw_spin_unlock(&qi->q_lock);
1273                 cpu_relax();
1274                 raw_spin_lock(&qi->q_lock);
1275         }
1276
1277         qi->desc_status[index] = QI_DONE;
1278
1279         reclaim_free_desc(qi);
1280         raw_spin_unlock_irqrestore(&qi->q_lock, flags);
1281
1282         if (rc == -EAGAIN)
1283                 goto restart;
1284
1285         return rc;
1286 }
1287
1288 /*
1289  * Flush the global interrupt entry cache.
1290  */
1291 void qi_global_iec(struct intel_iommu *iommu)
1292 {
1293         struct qi_desc desc;
1294
1295         desc.low = QI_IEC_TYPE;
1296         desc.high = 0;
1297
1298         /* should never fail */
1299         qi_submit_sync(&desc, iommu);
1300 }
1301
1302 void qi_flush_context(struct intel_iommu *iommu, u16 did, u16 sid, u8 fm,
1303                       u64 type)
1304 {
1305         struct qi_desc desc;
1306
1307         desc.low = QI_CC_FM(fm) | QI_CC_SID(sid) | QI_CC_DID(did)
1308                         | QI_CC_GRAN(type) | QI_CC_TYPE;
1309         desc.high = 0;
1310
1311         qi_submit_sync(&desc, iommu);
1312 }
1313
1314 void qi_flush_iotlb(struct intel_iommu *iommu, u16 did, u64 addr,
1315                     unsigned int size_order, u64 type)
1316 {
1317         u8 dw = 0, dr = 0;
1318
1319         struct qi_desc desc;
1320         int ih = 0;
1321
1322         if (cap_write_drain(iommu->cap))
1323                 dw = 1;
1324
1325         if (cap_read_drain(iommu->cap))
1326                 dr = 1;
1327
1328         desc.low = QI_IOTLB_DID(did) | QI_IOTLB_DR(dr) | QI_IOTLB_DW(dw)
1329                 | QI_IOTLB_GRAN(type) | QI_IOTLB_TYPE;
1330         desc.high = QI_IOTLB_ADDR(addr) | QI_IOTLB_IH(ih)
1331                 | QI_IOTLB_AM(size_order);
1332
1333         qi_submit_sync(&desc, iommu);
1334 }
1335
1336 void qi_flush_dev_iotlb(struct intel_iommu *iommu, u16 sid, u16 qdep,
1337                         u64 addr, unsigned mask)
1338 {
1339         struct qi_desc desc;
1340
1341         if (mask) {
1342                 BUG_ON(addr & ((1 << (VTD_PAGE_SHIFT + mask)) - 1));
1343                 addr |= (1 << (VTD_PAGE_SHIFT + mask - 1)) - 1;
1344                 desc.high = QI_DEV_IOTLB_ADDR(addr) | QI_DEV_IOTLB_SIZE;
1345         } else
1346                 desc.high = QI_DEV_IOTLB_ADDR(addr);
1347
1348         if (qdep >= QI_DEV_IOTLB_MAX_INVS)
1349                 qdep = 0;
1350
1351         desc.low = QI_DEV_IOTLB_SID(sid) | QI_DEV_IOTLB_QDEP(qdep) |
1352                    QI_DIOTLB_TYPE;
1353
1354         qi_submit_sync(&desc, iommu);
1355 }
1356
1357 /*
1358  * Disable Queued Invalidation interface.
1359  */
1360 void dmar_disable_qi(struct intel_iommu *iommu)
1361 {
1362         unsigned long flags;
1363         u32 sts;
1364         cycles_t start_time = get_cycles();
1365
1366         if (!ecap_qis(iommu->ecap))
1367                 return;
1368
1369         raw_spin_lock_irqsave(&iommu->register_lock, flags);
1370
1371         sts =  readl(iommu->reg + DMAR_GSTS_REG);
1372         if (!(sts & DMA_GSTS_QIES))
1373                 goto end;
1374
1375         /*
1376          * Give a chance to HW to complete the pending invalidation requests.
1377          */
1378         while ((readl(iommu->reg + DMAR_IQT_REG) !=
1379                 readl(iommu->reg + DMAR_IQH_REG)) &&
1380                 (DMAR_OPERATION_TIMEOUT > (get_cycles() - start_time)))
1381                 cpu_relax();
1382
1383         iommu->gcmd &= ~DMA_GCMD_QIE;
1384         writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
1385
1386         IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, readl,
1387                       !(sts & DMA_GSTS_QIES), sts);
1388 end:
1389         raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
1390 }
1391
1392 /*
1393  * Enable queued invalidation.
1394  */
1395 static void __dmar_enable_qi(struct intel_iommu *iommu)
1396 {
1397         u32 sts;
1398         unsigned long flags;
1399         struct q_inval *qi = iommu->qi;
1400
1401         qi->free_head = qi->free_tail = 0;
1402         qi->free_cnt = QI_LENGTH;
1403
1404         raw_spin_lock_irqsave(&iommu->register_lock, flags);
1405
1406         /* write zero to the tail reg */
1407         writel(0, iommu->reg + DMAR_IQT_REG);
1408
1409         dmar_writeq(iommu->reg + DMAR_IQA_REG, virt_to_phys(qi->desc));
1410
1411         iommu->gcmd |= DMA_GCMD_QIE;
1412         writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
1413
1414         /* Make sure hardware complete it */
1415         IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, readl, (sts & DMA_GSTS_QIES), sts);
1416
1417         raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
1418 }
1419
1420 /*
1421  * Enable Queued Invalidation interface. This is a must to support
1422  * interrupt-remapping. Also used by DMA-remapping, which replaces
1423  * register based IOTLB invalidation.
1424  */
1425 int dmar_enable_qi(struct intel_iommu *iommu)
1426 {
1427         struct q_inval *qi;
1428         struct page *desc_page;
1429
1430         if (!ecap_qis(iommu->ecap))
1431                 return -ENOENT;
1432
1433         /*
1434          * queued invalidation is already setup and enabled.
1435          */
1436         if (iommu->qi)
1437                 return 0;
1438
1439         iommu->qi = kmalloc(sizeof(*qi), GFP_ATOMIC);
1440         if (!iommu->qi)
1441                 return -ENOMEM;
1442
1443         qi = iommu->qi;
1444
1445
1446         desc_page = alloc_pages_node(iommu->node, GFP_ATOMIC | __GFP_ZERO, 0);
1447         if (!desc_page) {
1448                 kfree(qi);
1449                 iommu->qi = NULL;
1450                 return -ENOMEM;
1451         }
1452
1453         qi->desc = page_address(desc_page);
1454
1455         qi->desc_status = kzalloc(QI_LENGTH * sizeof(int), GFP_ATOMIC);
1456         if (!qi->desc_status) {
1457                 free_page((unsigned long) qi->desc);
1458                 kfree(qi);
1459                 iommu->qi = NULL;
1460                 return -ENOMEM;
1461         }
1462
1463         raw_spin_lock_init(&qi->q_lock);
1464
1465         __dmar_enable_qi(iommu);
1466
1467         return 0;
1468 }
1469
1470 /* iommu interrupt handling. Most stuff are MSI-like. */
1471
1472 enum faulttype {
1473         DMA_REMAP,
1474         INTR_REMAP,
1475         UNKNOWN,
1476 };
1477
1478 static const char *dma_remap_fault_reasons[] =
1479 {
1480         "Software",
1481         "Present bit in root entry is clear",
1482         "Present bit in context entry is clear",
1483         "Invalid context entry",
1484         "Access beyond MGAW",
1485         "PTE Write access is not set",
1486         "PTE Read access is not set",
1487         "Next page table ptr is invalid",
1488         "Root table address invalid",
1489         "Context table ptr is invalid",
1490         "non-zero reserved fields in RTP",
1491         "non-zero reserved fields in CTP",
1492         "non-zero reserved fields in PTE",
1493         "PCE for translation request specifies blocking",
1494 };
1495
1496 static const char *irq_remap_fault_reasons[] =
1497 {
1498         "Detected reserved fields in the decoded interrupt-remapped request",
1499         "Interrupt index exceeded the interrupt-remapping table size",
1500         "Present field in the IRTE entry is clear",
1501         "Error accessing interrupt-remapping table pointed by IRTA_REG",
1502         "Detected reserved fields in the IRTE entry",
1503         "Blocked a compatibility format interrupt request",
1504         "Blocked an interrupt request due to source-id verification failure",
1505 };
1506
1507 static const char *dmar_get_fault_reason(u8 fault_reason, int *fault_type)
1508 {
1509         if (fault_reason >= 0x20 && (fault_reason - 0x20 <
1510                                         ARRAY_SIZE(irq_remap_fault_reasons))) {
1511                 *fault_type = INTR_REMAP;
1512                 return irq_remap_fault_reasons[fault_reason - 0x20];
1513         } else if (fault_reason < ARRAY_SIZE(dma_remap_fault_reasons)) {
1514                 *fault_type = DMA_REMAP;
1515                 return dma_remap_fault_reasons[fault_reason];
1516         } else {
1517                 *fault_type = UNKNOWN;
1518                 return "Unknown";
1519         }
1520 }
1521
1522
1523 static inline int dmar_msi_reg(struct intel_iommu *iommu, int irq)
1524 {
1525         if (iommu->irq == irq)
1526                 return DMAR_FECTL_REG;
1527         else if (iommu->pr_irq == irq)
1528                 return DMAR_PECTL_REG;
1529         else
1530                 BUG();
1531 }
1532
1533 void dmar_msi_unmask(struct irq_data *data)
1534 {
1535         struct intel_iommu *iommu = irq_data_get_irq_handler_data(data);
1536         int reg = dmar_msi_reg(iommu, data->irq);
1537         unsigned long flag;
1538
1539         /* unmask it */
1540         raw_spin_lock_irqsave(&iommu->register_lock, flag);
1541         writel(0, iommu->reg + reg);
1542         /* Read a reg to force flush the post write */
1543         readl(iommu->reg + reg);
1544         raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1545 }
1546
1547 void dmar_msi_mask(struct irq_data *data)
1548 {
1549         struct intel_iommu *iommu = irq_data_get_irq_handler_data(data);
1550         int reg = dmar_msi_reg(iommu, data->irq);
1551         unsigned long flag;
1552
1553         /* mask it */
1554         raw_spin_lock_irqsave(&iommu->register_lock, flag);
1555         writel(DMA_FECTL_IM, iommu->reg + reg);
1556         /* Read a reg to force flush the post write */
1557         readl(iommu->reg + reg);
1558         raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1559 }
1560
1561 void dmar_msi_write(int irq, struct msi_msg *msg)
1562 {
1563         struct intel_iommu *iommu = irq_get_handler_data(irq);
1564         int reg = dmar_msi_reg(iommu, irq);
1565         unsigned long flag;
1566
1567         raw_spin_lock_irqsave(&iommu->register_lock, flag);
1568         writel(msg->data, iommu->reg + reg + 4);
1569         writel(msg->address_lo, iommu->reg + reg + 8);
1570         writel(msg->address_hi, iommu->reg + reg + 12);
1571         raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1572 }
1573
1574 void dmar_msi_read(int irq, struct msi_msg *msg)
1575 {
1576         struct intel_iommu *iommu = irq_get_handler_data(irq);
1577         int reg = dmar_msi_reg(iommu, irq);
1578         unsigned long flag;
1579
1580         raw_spin_lock_irqsave(&iommu->register_lock, flag);
1581         msg->data = readl(iommu->reg + reg + 4);
1582         msg->address_lo = readl(iommu->reg + reg + 8);
1583         msg->address_hi = readl(iommu->reg + reg + 12);
1584         raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1585 }
1586
1587 static int dmar_fault_do_one(struct intel_iommu *iommu, int type,
1588                 u8 fault_reason, u16 source_id, unsigned long long addr)
1589 {
1590         const char *reason;
1591         int fault_type;
1592
1593         reason = dmar_get_fault_reason(fault_reason, &fault_type);
1594
1595         if (fault_type == INTR_REMAP)
1596                 pr_err("[INTR-REMAP] Request device [%02x:%02x.%d] fault index %llx [fault reason %02d] %s\n",
1597                         source_id >> 8, PCI_SLOT(source_id & 0xFF),
1598                         PCI_FUNC(source_id & 0xFF), addr >> 48,
1599                         fault_reason, reason);
1600         else
1601                 pr_err("[%s] Request device [%02x:%02x.%d] fault addr %llx [fault reason %02d] %s\n",
1602                        type ? "DMA Read" : "DMA Write",
1603                        source_id >> 8, PCI_SLOT(source_id & 0xFF),
1604                        PCI_FUNC(source_id & 0xFF), addr, fault_reason, reason);
1605         return 0;
1606 }
1607
1608 #define PRIMARY_FAULT_REG_LEN (16)
1609 irqreturn_t dmar_fault(int irq, void *dev_id)
1610 {
1611         struct intel_iommu *iommu = dev_id;
1612         int reg, fault_index;
1613         u32 fault_status;
1614         unsigned long flag;
1615         bool ratelimited;
1616         static DEFINE_RATELIMIT_STATE(rs,
1617                                       DEFAULT_RATELIMIT_INTERVAL,
1618                                       DEFAULT_RATELIMIT_BURST);
1619
1620         /* Disable printing, simply clear the fault when ratelimited */
1621         ratelimited = !__ratelimit(&rs);
1622
1623         raw_spin_lock_irqsave(&iommu->register_lock, flag);
1624         fault_status = readl(iommu->reg + DMAR_FSTS_REG);
1625         if (fault_status && !ratelimited)
1626                 pr_err("DRHD: handling fault status reg %x\n", fault_status);
1627
1628         /* TBD: ignore advanced fault log currently */
1629         if (!(fault_status & DMA_FSTS_PPF))
1630                 goto unlock_exit;
1631
1632         fault_index = dma_fsts_fault_record_index(fault_status);
1633         reg = cap_fault_reg_offset(iommu->cap);
1634         while (1) {
1635                 u8 fault_reason;
1636                 u16 source_id;
1637                 u64 guest_addr;
1638                 int type;
1639                 u32 data;
1640
1641                 /* highest 32 bits */
1642                 data = readl(iommu->reg + reg +
1643                                 fault_index * PRIMARY_FAULT_REG_LEN + 12);
1644                 if (!(data & DMA_FRCD_F))
1645                         break;
1646
1647                 if (!ratelimited) {
1648                         fault_reason = dma_frcd_fault_reason(data);
1649                         type = dma_frcd_type(data);
1650
1651                         data = readl(iommu->reg + reg +
1652                                      fault_index * PRIMARY_FAULT_REG_LEN + 8);
1653                         source_id = dma_frcd_source_id(data);
1654
1655                         guest_addr = dmar_readq(iommu->reg + reg +
1656                                         fault_index * PRIMARY_FAULT_REG_LEN);
1657                         guest_addr = dma_frcd_page_addr(guest_addr);
1658                 }
1659
1660                 /* clear the fault */
1661                 writel(DMA_FRCD_F, iommu->reg + reg +
1662                         fault_index * PRIMARY_FAULT_REG_LEN + 12);
1663
1664                 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1665
1666                 if (!ratelimited)
1667                         dmar_fault_do_one(iommu, type, fault_reason,
1668                                           source_id, guest_addr);
1669
1670                 fault_index++;
1671                 if (fault_index >= cap_num_fault_regs(iommu->cap))
1672                         fault_index = 0;
1673                 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1674         }
1675
1676         writel(DMA_FSTS_PFO | DMA_FSTS_PPF, iommu->reg + DMAR_FSTS_REG);
1677
1678 unlock_exit:
1679         raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1680         return IRQ_HANDLED;
1681 }
1682
1683 int dmar_set_interrupt(struct intel_iommu *iommu)
1684 {
1685         int irq, ret;
1686
1687         /*
1688          * Check if the fault interrupt is already initialized.
1689          */
1690         if (iommu->irq)
1691                 return 0;
1692
1693         irq = dmar_alloc_hwirq(iommu->seq_id, iommu->node, iommu);
1694         if (irq > 0) {
1695                 iommu->irq = irq;
1696         } else {
1697                 pr_err("No free IRQ vectors\n");
1698                 return -EINVAL;
1699         }
1700
1701         ret = request_irq(irq, dmar_fault, IRQF_NO_THREAD, iommu->name, iommu);
1702         if (ret)
1703                 pr_err("Can't request irq\n");
1704         return ret;
1705 }
1706
1707 int __init enable_drhd_fault_handling(void)
1708 {
1709         struct dmar_drhd_unit *drhd;
1710         struct intel_iommu *iommu;
1711
1712         /*
1713          * Enable fault control interrupt.
1714          */
1715         for_each_iommu(iommu, drhd) {
1716                 u32 fault_status;
1717                 int ret = dmar_set_interrupt(iommu);
1718
1719                 if (ret) {
1720                         pr_err("DRHD %Lx: failed to enable fault, interrupt, ret %d\n",
1721                                (unsigned long long)drhd->reg_base_addr, ret);
1722                         return -1;
1723                 }
1724
1725                 /*
1726                  * Clear any previous faults.
1727                  */
1728                 dmar_fault(iommu->irq, iommu);
1729                 fault_status = readl(iommu->reg + DMAR_FSTS_REG);
1730                 writel(fault_status, iommu->reg + DMAR_FSTS_REG);
1731         }
1732
1733         return 0;
1734 }
1735
1736 /*
1737  * Re-enable Queued Invalidation interface.
1738  */
1739 int dmar_reenable_qi(struct intel_iommu *iommu)
1740 {
1741         if (!ecap_qis(iommu->ecap))
1742                 return -ENOENT;
1743
1744         if (!iommu->qi)
1745                 return -ENOENT;
1746
1747         /*
1748          * First disable queued invalidation.
1749          */
1750         dmar_disable_qi(iommu);
1751         /*
1752          * Then enable queued invalidation again. Since there is no pending
1753          * invalidation requests now, it's safe to re-enable queued
1754          * invalidation.
1755          */
1756         __dmar_enable_qi(iommu);
1757
1758         return 0;
1759 }
1760
1761 /*
1762  * Check interrupt remapping support in DMAR table description.
1763  */
1764 int __init dmar_ir_support(void)
1765 {
1766         struct acpi_table_dmar *dmar;
1767         dmar = (struct acpi_table_dmar *)dmar_tbl;
1768         if (!dmar)
1769                 return 0;
1770         return dmar->flags & 0x1;
1771 }
1772
1773 /* Check whether DMAR units are in use */
1774 static inline bool dmar_in_use(void)
1775 {
1776         return irq_remapping_enabled || intel_iommu_enabled;
1777 }
1778
1779 static int __init dmar_free_unused_resources(void)
1780 {
1781         struct dmar_drhd_unit *dmaru, *dmaru_n;
1782
1783         if (dmar_in_use())
1784                 return 0;
1785
1786         if (dmar_dev_scope_status != 1 && !list_empty(&dmar_drhd_units))
1787                 bus_unregister_notifier(&pci_bus_type, &dmar_pci_bus_nb);
1788
1789         down_write(&dmar_global_lock);
1790         list_for_each_entry_safe(dmaru, dmaru_n, &dmar_drhd_units, list) {
1791                 list_del(&dmaru->list);
1792                 dmar_free_drhd(dmaru);
1793         }
1794         up_write(&dmar_global_lock);
1795
1796         return 0;
1797 }
1798
1799 late_initcall(dmar_free_unused_resources);
1800 IOMMU_INIT_POST(detect_intel_iommu);
1801
1802 /*
1803  * DMAR Hotplug Support
1804  * For more details, please refer to Intel(R) Virtualization Technology
1805  * for Directed-IO Architecture Specifiction, Rev 2.2, Section 8.8
1806  * "Remapping Hardware Unit Hot Plug".
1807  */
1808 static u8 dmar_hp_uuid[] = {
1809         /* 0000 */    0xA6, 0xA3, 0xC1, 0xD8, 0x9B, 0xBE, 0x9B, 0x4C,
1810         /* 0008 */    0x91, 0xBF, 0xC3, 0xCB, 0x81, 0xFC, 0x5D, 0xAF
1811 };
1812
1813 /*
1814  * Currently there's only one revision and BIOS will not check the revision id,
1815  * so use 0 for safety.
1816  */
1817 #define DMAR_DSM_REV_ID                 0
1818 #define DMAR_DSM_FUNC_DRHD              1
1819 #define DMAR_DSM_FUNC_ATSR              2
1820 #define DMAR_DSM_FUNC_RHSA              3
1821
1822 static inline bool dmar_detect_dsm(acpi_handle handle, int func)
1823 {
1824         return acpi_check_dsm(handle, dmar_hp_uuid, DMAR_DSM_REV_ID, 1 << func);
1825 }
1826
1827 static int dmar_walk_dsm_resource(acpi_handle handle, int func,
1828                                   dmar_res_handler_t handler, void *arg)
1829 {
1830         int ret = -ENODEV;
1831         union acpi_object *obj;
1832         struct acpi_dmar_header *start;
1833         struct dmar_res_callback callback;
1834         static int res_type[] = {
1835                 [DMAR_DSM_FUNC_DRHD] = ACPI_DMAR_TYPE_HARDWARE_UNIT,
1836                 [DMAR_DSM_FUNC_ATSR] = ACPI_DMAR_TYPE_ROOT_ATS,
1837                 [DMAR_DSM_FUNC_RHSA] = ACPI_DMAR_TYPE_HARDWARE_AFFINITY,
1838         };
1839
1840         if (!dmar_detect_dsm(handle, func))
1841                 return 0;
1842
1843         obj = acpi_evaluate_dsm_typed(handle, dmar_hp_uuid, DMAR_DSM_REV_ID,
1844                                       func, NULL, ACPI_TYPE_BUFFER);
1845         if (!obj)
1846                 return -ENODEV;
1847
1848         memset(&callback, 0, sizeof(callback));
1849         callback.cb[res_type[func]] = handler;
1850         callback.arg[res_type[func]] = arg;
1851         start = (struct acpi_dmar_header *)obj->buffer.pointer;
1852         ret = dmar_walk_remapping_entries(start, obj->buffer.length, &callback);
1853
1854         ACPI_FREE(obj);
1855
1856         return ret;
1857 }
1858
1859 static int dmar_hp_add_drhd(struct acpi_dmar_header *header, void *arg)
1860 {
1861         int ret;
1862         struct dmar_drhd_unit *dmaru;
1863
1864         dmaru = dmar_find_dmaru((struct acpi_dmar_hardware_unit *)header);
1865         if (!dmaru)
1866                 return -ENODEV;
1867
1868         ret = dmar_ir_hotplug(dmaru, true);
1869         if (ret == 0)
1870                 ret = dmar_iommu_hotplug(dmaru, true);
1871
1872         return ret;
1873 }
1874
1875 static int dmar_hp_remove_drhd(struct acpi_dmar_header *header, void *arg)
1876 {
1877         int i, ret;
1878         struct device *dev;
1879         struct dmar_drhd_unit *dmaru;
1880
1881         dmaru = dmar_find_dmaru((struct acpi_dmar_hardware_unit *)header);
1882         if (!dmaru)
1883                 return 0;
1884
1885         /*
1886          * All PCI devices managed by this unit should have been destroyed.
1887          */
1888         if (!dmaru->include_all && dmaru->devices && dmaru->devices_cnt) {
1889                 for_each_active_dev_scope(dmaru->devices,
1890                                           dmaru->devices_cnt, i, dev)
1891                         return -EBUSY;
1892         }
1893
1894         ret = dmar_ir_hotplug(dmaru, false);
1895         if (ret == 0)
1896                 ret = dmar_iommu_hotplug(dmaru, false);
1897
1898         return ret;
1899 }
1900
1901 static int dmar_hp_release_drhd(struct acpi_dmar_header *header, void *arg)
1902 {
1903         struct dmar_drhd_unit *dmaru;
1904
1905         dmaru = dmar_find_dmaru((struct acpi_dmar_hardware_unit *)header);
1906         if (dmaru) {
1907                 list_del_rcu(&dmaru->list);
1908                 synchronize_rcu();
1909                 dmar_free_drhd(dmaru);
1910         }
1911
1912         return 0;
1913 }
1914
1915 static int dmar_hotplug_insert(acpi_handle handle)
1916 {
1917         int ret;
1918         int drhd_count = 0;
1919
1920         ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1921                                      &dmar_validate_one_drhd, (void *)1);
1922         if (ret)
1923                 goto out;
1924
1925         ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1926                                      &dmar_parse_one_drhd, (void *)&drhd_count);
1927         if (ret == 0 && drhd_count == 0) {
1928                 pr_warn(FW_BUG "No DRHD structures in buffer returned by _DSM method\n");
1929                 goto out;
1930         } else if (ret) {
1931                 goto release_drhd;
1932         }
1933
1934         ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_RHSA,
1935                                      &dmar_parse_one_rhsa, NULL);
1936         if (ret)
1937                 goto release_drhd;
1938
1939         ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
1940                                      &dmar_parse_one_atsr, NULL);
1941         if (ret)
1942                 goto release_atsr;
1943
1944         ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1945                                      &dmar_hp_add_drhd, NULL);
1946         if (!ret)
1947                 return 0;
1948
1949         dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1950                                &dmar_hp_remove_drhd, NULL);
1951 release_atsr:
1952         dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
1953                                &dmar_release_one_atsr, NULL);
1954 release_drhd:
1955         dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1956                                &dmar_hp_release_drhd, NULL);
1957 out:
1958         return ret;
1959 }
1960
1961 static int dmar_hotplug_remove(acpi_handle handle)
1962 {
1963         int ret;
1964
1965         ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
1966                                      &dmar_check_one_atsr, NULL);
1967         if (ret)
1968                 return ret;
1969
1970         ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1971                                      &dmar_hp_remove_drhd, NULL);
1972         if (ret == 0) {
1973                 WARN_ON(dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
1974                                                &dmar_release_one_atsr, NULL));
1975                 WARN_ON(dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1976                                                &dmar_hp_release_drhd, NULL));
1977         } else {
1978                 dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
1979                                        &dmar_hp_add_drhd, NULL);
1980         }
1981
1982         return ret;
1983 }
1984
1985 static acpi_status dmar_get_dsm_handle(acpi_handle handle, u32 lvl,
1986                                        void *context, void **retval)
1987 {
1988         acpi_handle *phdl = retval;
1989
1990         if (dmar_detect_dsm(handle, DMAR_DSM_FUNC_DRHD)) {
1991                 *phdl = handle;
1992                 return AE_CTRL_TERMINATE;
1993         }
1994
1995         return AE_OK;
1996 }
1997
1998 static int dmar_device_hotplug(acpi_handle handle, bool insert)
1999 {
2000         int ret;
2001         acpi_handle tmp = NULL;
2002         acpi_status status;
2003
2004         if (!dmar_in_use())
2005                 return 0;
2006
2007         if (dmar_detect_dsm(handle, DMAR_DSM_FUNC_DRHD)) {
2008                 tmp = handle;
2009         } else {
2010                 status = acpi_walk_namespace(ACPI_TYPE_DEVICE, handle,
2011                                              ACPI_UINT32_MAX,
2012                                              dmar_get_dsm_handle,
2013                                              NULL, NULL, &tmp);
2014                 if (ACPI_FAILURE(status)) {
2015                         pr_warn("Failed to locate _DSM method.\n");
2016                         return -ENXIO;
2017                 }
2018         }
2019         if (tmp == NULL)
2020                 return 0;
2021
2022         down_write(&dmar_global_lock);
2023         if (insert)
2024                 ret = dmar_hotplug_insert(tmp);
2025         else
2026                 ret = dmar_hotplug_remove(tmp);
2027         up_write(&dmar_global_lock);
2028
2029         return ret;
2030 }
2031
2032 int dmar_device_add(acpi_handle handle)
2033 {
2034         return dmar_device_hotplug(handle, true);
2035 }
2036
2037 int dmar_device_remove(acpi_handle handle)
2038 {
2039         return dmar_device_hotplug(handle, false);
2040 }