drivers/pci/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 #include <linux/pci.h>
  30 #include <linux/dmar.h>
  31 #include <linux/iova.h>
  32 #include <linux/intel-iommu.h>
  33 #include <linux/timer.h>
  34 #include <linux/irq.h>
  35 #include <linux/interrupt.h>
  36 #include <linux/tboot.h>
  37 #include <linux/dmi.h>
  38
  39 #define PREFIX "DMAR: "
  40
  41 /* No locks are needed as DMA remapping hardware unit
  42  * list is constructed at boot time and hotplug of
  43  * these units are not supported by the architecture.
  44  */
  45 LIST_HEAD(dmar_drhd_units);
  46
  47 static struct acpi_table_header * __initdata dmar_tbl;
  48 static acpi_size dmar_tbl_size;
  49
  50 static void __init dmar_register_drhd_unit(struct dmar_drhd_unit *drhd)
  51 {
  52         /*
  53          * add INCLUDE_ALL at the tail, so scan the list will find it at
  54          * the very end.
  55          */
  56         if (drhd->include_all)
  57                 list_add_tail(&drhd->list, &dmar_drhd_units);
  58         else
  59                 list_add(&drhd->list, &dmar_drhd_units);
  60 }
  61
  62 static int __init dmar_parse_one_dev_scope(struct acpi_dmar_device_scope *scope,
  63                                            struct pci_dev **dev, u16 segment)
  64 {
  65         struct pci_bus *bus;
  66         struct pci_dev *pdev = NULL;
  67         struct acpi_dmar_pci_path *path;
  68         int count;
  69
  70         bus = pci_find_bus(segment, scope->bus);
  71         path = (struct acpi_dmar_pci_path *)(scope + 1);
  72         count = (scope->length - sizeof(struct acpi_dmar_device_scope))
  73                 / sizeof(struct acpi_dmar_pci_path);
  74
  75         while (count) {
  76                 if (pdev)
  77                         pci_dev_put(pdev);
  78                 /*
  79                  * Some BIOSes list non-exist devices in DMAR table, just
  80                  * ignore it
  81                  */
  82                 if (!bus) {
  83                         printk(KERN_WARNING
  84                         PREFIX "Device scope bus [%d] not found\n",
  85                         scope->bus);
  86                         break;
  87                 }
  88                 pdev = pci_get_slot(bus, PCI_DEVFN(path->dev, path->fn));
  89                 if (!pdev) {
  90                         printk(KERN_WARNING PREFIX
  91                         "Device scope device [%04x:%02x:%02x.%02x] not found\n",
  92                                 segment, bus->number, path->dev, path->fn);
  93                         break;
  94                 }
  95                 path ++;
  96                 count --;
  97                 bus = pdev->subordinate;
  98         }
  99         if (!pdev) {
 100                 printk(KERN_WARNING PREFIX
 101                 "Device scope device [%04x:%02x:%02x.%02x] not found\n",
 102                 segment, scope->bus, path->dev, path->fn);
 103                 *dev = NULL;
 104                 return 0;
 105         }
 106         if ((scope->entry_type == ACPI_DMAR_SCOPE_TYPE_ENDPOINT && \
 107                         pdev->subordinate) || (scope->entry_type == \
 108                         ACPI_DMAR_SCOPE_TYPE_BRIDGE && !pdev->subordinate)) {
 109                 pci_dev_put(pdev);
 110                 printk(KERN_WARNING PREFIX
 111                         "Device scope type does not match for %s\n",
 112                          pci_name(pdev));
 113                 return -EINVAL;
 114         }
 115         *dev = pdev;
 116         return 0;
 117 }
 118
 119 static int __init dmar_parse_dev_scope(void *start, void *end, int *cnt,
 120                                        struct pci_dev ***devices, u16 segment)
 121 {
 122         struct acpi_dmar_device_scope *scope;
 123         void * tmp = start;
 124         int index;
 125         int ret;
 126
 127         *cnt = 0;
 128         while (start < end) {
 129                 scope = start;
 130                 if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_ENDPOINT ||
 131                     scope->entry_type == ACPI_DMAR_SCOPE_TYPE_BRIDGE)
 132                         (*cnt)++;
 133                 else if (scope->entry_type != ACPI_DMAR_SCOPE_TYPE_IOAPIC) {
 134                         printk(KERN_WARNING PREFIX
 135                                "Unsupported device scope\n");
 136                 }
 137                 start += scope->length;
 138         }
 139         if (*cnt == 0)
 140                 return 0;
 141
 142         *devices = kcalloc(*cnt, sizeof(struct pci_dev *), GFP_KERNEL);
 143         if (!*devices)
 144                 return -ENOMEM;
 145
 146         start = tmp;
 147         index = 0;
 148         while (start < end) {
 149                 scope = start;
 150                 if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_ENDPOINT ||
 151                     scope->entry_type == ACPI_DMAR_SCOPE_TYPE_BRIDGE) {
 152                         ret = dmar_parse_one_dev_scope(scope,
 153                                 &(*devices)[index], segment);
 154                         if (ret) {
 155                                 kfree(*devices);
 156                                 return ret;
 157                         }
 158                         index ++;
 159                 }
 160                 start += scope->length;
 161         }
 162
 163         return 0;
 164 }
 165
 166 /**
 167  * dmar_parse_one_drhd - parses exactly one DMA remapping hardware definition
 168  * structure which uniquely represent one DMA remapping hardware unit
 169  * present in the platform
 170  */
 171 static int __init
 172 dmar_parse_one_drhd(struct acpi_dmar_header *header)
 173 {
 174         struct acpi_dmar_hardware_unit *drhd;
 175         struct dmar_drhd_unit *dmaru;
 176         int ret = 0;
 177
 178         drhd = (struct acpi_dmar_hardware_unit *)header;
 179         dmaru = kzalloc(sizeof(*dmaru), GFP_KERNEL);
 180         if (!dmaru)
 181                 return -ENOMEM;
 182
 183         dmaru->hdr = header;
 184         dmaru->reg_base_addr = drhd->address;
 185         dmaru->segment = drhd->segment;
 186         dmaru->include_all = drhd->flags & 0x1; /* BIT0: INCLUDE_ALL */
 187
 188         ret = alloc_iommu(dmaru);
 189         if (ret) {
 190                 kfree(dmaru);
 191                 return ret;
 192         }
 193         dmar_register_drhd_unit(dmaru);
 194         return 0;
 195 }
 196
 197 static int __init dmar_parse_dev(struct dmar_drhd_unit *dmaru)
 198 {
 199         struct acpi_dmar_hardware_unit *drhd;
 200         int ret = 0;
 201
 202         drhd = (struct acpi_dmar_hardware_unit *) dmaru->hdr;
 203
 204         if (dmaru->include_all)
 205                 return 0;
 206
 207         ret = dmar_parse_dev_scope((void *)(drhd + 1),
 208                                 ((void *)drhd) + drhd->header.length,
 209                                 &dmaru->devices_cnt, &dmaru->devices,
 210                                 drhd->segment);
 211         if (ret) {
 212                 list_del(&dmaru->list);
 213                 kfree(dmaru);
 214         }
 215         return ret;
 216 }
 217
 218 #ifdef CONFIG_DMAR
 219 LIST_HEAD(dmar_rmrr_units);
 220
 221 static void __init dmar_register_rmrr_unit(struct dmar_rmrr_unit *rmrr)
 222 {
 223         list_add(&rmrr->list, &dmar_rmrr_units);
 224 }
 225
 226
 227 static int __init
 228 dmar_parse_one_rmrr(struct acpi_dmar_header *header)
 229 {
 230         struct acpi_dmar_reserved_memory *rmrr;
 231         struct dmar_rmrr_unit *rmrru;
 232
 233         rmrru = kzalloc(sizeof(*rmrru), GFP_KERNEL);
 234         if (!rmrru)
 235                 return -ENOMEM;
 236
 237         rmrru->hdr = header;
 238         rmrr = (struct acpi_dmar_reserved_memory *)header;
 239         rmrru->base_address = rmrr->base_address;
 240         rmrru->end_address = rmrr->end_address;
 241
 242         dmar_register_rmrr_unit(rmrru);
 243         return 0;
 244 }
 245
 246 static int __init
 247 rmrr_parse_dev(struct dmar_rmrr_unit *rmrru)
 248 {
 249         struct acpi_dmar_reserved_memory *rmrr;
 250         int ret;
 251
 252         rmrr = (struct acpi_dmar_reserved_memory *) rmrru->hdr;
 253         ret = dmar_parse_dev_scope((void *)(rmrr + 1),
 254                 ((void *)rmrr) + rmrr->header.length,
 255                 &rmrru->devices_cnt, &rmrru->devices, rmrr->segment);
 256
 257         if (ret || (rmrru->devices_cnt == 0)) {
 258                 list_del(&rmrru->list);
 259                 kfree(rmrru);
 260         }
 261         return ret;
 262 }
 263
 264 static LIST_HEAD(dmar_atsr_units);
 265
 266 static int __init dmar_parse_one_atsr(struct acpi_dmar_header *hdr)
 267 {
 268         struct acpi_dmar_atsr *atsr;
 269         struct dmar_atsr_unit *atsru;
 270
 271         atsr = container_of(hdr, struct acpi_dmar_atsr, header);
 272         atsru = kzalloc(sizeof(*atsru), GFP_KERNEL);
 273         if (!atsru)
 274                 return -ENOMEM;
 275
 276         atsru->hdr = hdr;
 277         atsru->include_all = atsr->flags & 0x1;
 278
 279         list_add(&atsru->list, &dmar_atsr_units);
 280
 281         return 0;
 282 }
 283
 284 static int __init atsr_parse_dev(struct dmar_atsr_unit *atsru)
 285 {
 286         int rc;
 287         struct acpi_dmar_atsr *atsr;
 288
 289         if (atsru->include_all)
 290                 return 0;
 291
 292         atsr = container_of(atsru->hdr, struct acpi_dmar_atsr, header);
 293         rc = dmar_parse_dev_scope((void *)(atsr + 1),
 294                                 (void *)atsr + atsr->header.length,
 295                                 &atsru->devices_cnt, &atsru->devices,
 296                                 atsr->segment);
 297         if (rc || !atsru->devices_cnt) {
 298                 list_del(&atsru->list);
 299                 kfree(atsru);
 300         }
 301
 302         return rc;
 303 }
 304
 305 int dmar_find_matched_atsr_unit(struct pci_dev *dev)
 306 {
 307         int i;
 308         struct pci_bus *bus;
 309         struct acpi_dmar_atsr *atsr;
 310         struct dmar_atsr_unit *atsru;
 311
 312         dev = pci_physfn(dev);
 313
 314         list_for_each_entry(atsru, &dmar_atsr_units, list) {
 315                 atsr = container_of(atsru->hdr, struct acpi_dmar_atsr, header);
 316                 if (atsr->segment == pci_domain_nr(dev->bus))
 317                         goto found;
 318         }
 319
 320         return 0;
 321
 322 found:
 323         for (bus = dev->bus; bus; bus = bus->parent) {
 324                 struct pci_dev *bridge = bus->self;
 325
 326                 if (!bridge || !pci_is_pcie(bridge) ||
 327                     bridge->pcie_type == PCI_EXP_TYPE_PCI_BRIDGE)
 328                         return 0;
 329
 330                 if (bridge->pcie_type == PCI_EXP_TYPE_ROOT_PORT) {
 331                         for (i = 0; i < atsru->devices_cnt; i++)
 332                                 if (atsru->devices[i] == bridge)
 333                                         return 1;
 334                         break;
 335                 }
 336         }
 337
 338         if (atsru->include_all)
 339                 return 1;
 340
 341         return 0;
 342 }
 343 #endif
 344
 345 #ifdef CONFIG_ACPI_NUMA
 346 static int __init
 347 dmar_parse_one_rhsa(struct acpi_dmar_header *header)
 348 {
 349         struct acpi_dmar_rhsa *rhsa;
 350         struct dmar_drhd_unit *drhd;
 351
 352         rhsa = (struct acpi_dmar_rhsa *)header;
 353         for_each_drhd_unit(drhd) {
 354                 if (drhd->reg_base_addr == rhsa->base_address) {
 355                         int node = acpi_map_pxm_to_node(rhsa->proximity_domain);
 356
 357                         if (!node_online(node))
 358                                 node = -1;
 359                         drhd->iommu->node = node;
 360                         return 0;
 361                 }
 362         }
 363         WARN_TAINT(
 364                 1, TAINT_FIRMWARE_WORKAROUND,
 365                 "Your BIOS is broken; RHSA refers to non-existent DMAR unit at %llx\n"
 366                 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
 367                 drhd->reg_base_addr,
 368                 dmi_get_system_info(DMI_BIOS_VENDOR),
 369                 dmi_get_system_info(DMI_BIOS_VERSION),
 370                 dmi_get_system_info(DMI_PRODUCT_VERSION));
 371
 372         return 0;
 373 }
 374 #endif
 375
 376 static void __init
 377 dmar_table_print_dmar_entry(struct acpi_dmar_header *header)
 378 {
 379         struct acpi_dmar_hardware_unit *drhd;
 380         struct acpi_dmar_reserved_memory *rmrr;
 381         struct acpi_dmar_atsr *atsr;
 382         struct acpi_dmar_rhsa *rhsa;
 383
 384         switch (header->type) {
 385         case ACPI_DMAR_TYPE_HARDWARE_UNIT:
 386                 drhd = container_of(header, struct acpi_dmar_hardware_unit,
 387                                     header);
 388                 printk (KERN_INFO PREFIX
 389                         "DRHD base: %#016Lx flags: %#x\n",
 390                         (unsigned long long)drhd->address, drhd->flags);
 391                 break;
 392         case ACPI_DMAR_TYPE_RESERVED_MEMORY:
 393                 rmrr = container_of(header, struct acpi_dmar_reserved_memory,
 394                                     header);
 395                 printk (KERN_INFO PREFIX
 396                         "RMRR base: %#016Lx end: %#016Lx\n",
 397                         (unsigned long long)rmrr->base_address,
 398                         (unsigned long long)rmrr->end_address);
 399                 break;
 400         case ACPI_DMAR_TYPE_ATSR:
 401                 atsr = container_of(header, struct acpi_dmar_atsr, header);
 402                 printk(KERN_INFO PREFIX "ATSR flags: %#x\n", atsr->flags);
 403                 break;
 404         case ACPI_DMAR_HARDWARE_AFFINITY:
 405                 rhsa = container_of(header, struct acpi_dmar_rhsa, header);
 406                 printk(KERN_INFO PREFIX "RHSA base: %#016Lx proximity domain: %#x\n",
 407                        (unsigned long long)rhsa->base_address,
 408                        rhsa->proximity_domain);
 409                 break;
 410         }
 411 }
 412
 413 /**
 414  * dmar_table_detect - checks to see if the platform supports DMAR devices
 415  */
 416 static int __init dmar_table_detect(void)
 417 {
 418         acpi_status status = AE_OK;
 419
 420         /* if we could find DMAR table, then there are DMAR devices */
 421         status = acpi_get_table_with_size(ACPI_SIG_DMAR, 0,
 422                                 (struct acpi_table_header **)&dmar_tbl,
 423                                 &dmar_tbl_size);
 424
 425         if (ACPI_SUCCESS(status) && !dmar_tbl) {
 426                 printk (KERN_WARNING PREFIX "Unable to map DMAR\n");
 427                 status = AE_NOT_FOUND;
 428         }
 429
 430         return (ACPI_SUCCESS(status) ? 1 : 0);
 431 }
 432
 433 /**
 434  * parse_dmar_table - parses the DMA reporting table
 435  */
 436 static int __init
 437 parse_dmar_table(void)
 438 {
 439         struct acpi_table_dmar *dmar;
 440         struct acpi_dmar_header *entry_header;
 441         int ret = 0;
 442
 443         /*
 444          * Do it again, earlier dmar_tbl mapping could be mapped with
 445          * fixed map.
 446          */
 447         dmar_table_detect();
 448
 449         /*
 450          * ACPI tables may not be DMA protected by tboot, so use DMAR copy
 451          * SINIT saved in SinitMleData in TXT heap (which is DMA protected)
 452          */
 453         dmar_tbl = tboot_get_dmar_table(dmar_tbl);
 454
 455         dmar = (struct acpi_table_dmar *)dmar_tbl;
 456         if (!dmar)
 457                 return -ENODEV;
 458
 459         if (dmar->width < PAGE_SHIFT - 1) {
 460                 printk(KERN_WARNING PREFIX "Invalid DMAR haw\n");
 461                 return -EINVAL;
 462         }
 463
 464         printk (KERN_INFO PREFIX "Host address width %d\n",
 465                 dmar->width + 1);
 466
 467         entry_header = (struct acpi_dmar_header *)(dmar + 1);
 468         while (((unsigned long)entry_header) <
 469                         (((unsigned long)dmar) + dmar_tbl->length)) {
 470                 /* Avoid looping forever on bad ACPI tables */
 471                 if (entry_header->length == 0) {
 472                         printk(KERN_WARNING PREFIX
 473                                 "Invalid 0-length structure\n");
 474                         ret = -EINVAL;
 475                         break;
 476                 }
 477
 478                 dmar_table_print_dmar_entry(entry_header);
 479
 480                 switch (entry_header->type) {
 481                 case ACPI_DMAR_TYPE_HARDWARE_UNIT:
 482                         ret = dmar_parse_one_drhd(entry_header);
 483                         break;
 484                 case ACPI_DMAR_TYPE_RESERVED_MEMORY:
 485 #ifdef CONFIG_DMAR
 486                         ret = dmar_parse_one_rmrr(entry_header);
 487 #endif
 488                         break;
 489                 case ACPI_DMAR_TYPE_ATSR:
 490 #ifdef CONFIG_DMAR
 491                         ret = dmar_parse_one_atsr(entry_header);
 492 #endif
 493                         break;
 494                 case ACPI_DMAR_HARDWARE_AFFINITY:
 495 #ifdef CONFIG_ACPI_NUMA
 496                         ret = dmar_parse_one_rhsa(entry_header);
 497 #endif
 498                         break;
 499                 default:
 500                         printk(KERN_WARNING PREFIX
 501                                 "Unknown DMAR structure type %d\n",
 502                                 entry_header->type);
 503                         ret = 0; /* for forward compatibility */
 504                         break;
 505                 }
 506                 if (ret)
 507                         break;
 508
 509                 entry_header = ((void *)entry_header + entry_header->length);
 510         }
 511         return ret;
 512 }
 513
 514 static int dmar_pci_device_match(struct pci_dev *devices[], int cnt,
 515                           struct pci_dev *dev)
 516 {
 517         int index;
 518
 519         while (dev) {
 520                 for (index = 0; index < cnt; index++)
 521                         if (dev == devices[index])
 522                                 return 1;
 523
 524                 /* Check our parent */
 525                 dev = dev->bus->self;
 526         }
 527
 528         return 0;
 529 }
 530
 531 struct dmar_drhd_unit *
 532 dmar_find_matched_drhd_unit(struct pci_dev *dev)
 533 {
 534         struct dmar_drhd_unit *dmaru = NULL;
 535         struct acpi_dmar_hardware_unit *drhd;
 536
 537         dev = pci_physfn(dev);
 538
 539         list_for_each_entry(dmaru, &dmar_drhd_units, list) {
 540                 drhd = container_of(dmaru->hdr,
 541                                     struct acpi_dmar_hardware_unit,
 542                                     header);
 543
 544                 if (dmaru->include_all &&
 545                     drhd->segment == pci_domain_nr(dev->bus))
 546                         return dmaru;
 547
 548                 if (dmar_pci_device_match(dmaru->devices,
 549                                           dmaru->devices_cnt, dev))
 550                         return dmaru;
 551         }
 552
 553         return NULL;
 554 }
 555
 556 int __init dmar_dev_scope_init(void)
 557 {
 558         struct dmar_drhd_unit *drhd, *drhd_n;
 559         int ret = -ENODEV;
 560
 561         list_for_each_entry_safe(drhd, drhd_n, &dmar_drhd_units, list) {
 562                 ret = dmar_parse_dev(drhd);
 563                 if (ret)
 564                         return ret;
 565         }
 566
 567 #ifdef CONFIG_DMAR
 568         {
 569                 struct dmar_rmrr_unit *rmrr, *rmrr_n;
 570                 struct dmar_atsr_unit *atsr, *atsr_n;
 571
 572                 list_for_each_entry_safe(rmrr, rmrr_n, &dmar_rmrr_units, list) {
 573                         ret = rmrr_parse_dev(rmrr);
 574                         if (ret)
 575                                 return ret;
 576                 }
 577
 578                 list_for_each_entry_safe(atsr, atsr_n, &dmar_atsr_units, list) {
 579                         ret = atsr_parse_dev(atsr);
 580                         if (ret)
 581                                 return ret;
 582                 }
 583         }
 584 #endif
 585
 586         return ret;
 587 }
 588
 589
 590 int __init dmar_table_init(void)
 591 {
 592         static int dmar_table_initialized;
 593         int ret;
 594
 595         if (dmar_table_initialized)
 596                 return 0;
 597
 598         dmar_table_initialized = 1;
 599
 600         ret = parse_dmar_table();
 601         if (ret) {
 602                 if (ret != -ENODEV)
 603                         printk(KERN_INFO PREFIX "parse DMAR table failure.\n");
 604                 return ret;
 605         }
 606
 607         if (list_empty(&dmar_drhd_units)) {
 608                 printk(KERN_INFO PREFIX "No DMAR devices found\n");
 609                 return -ENODEV;
 610         }
 611
 612 #ifdef CONFIG_DMAR
 613         if (list_empty(&dmar_rmrr_units))
 614                 printk(KERN_INFO PREFIX "No RMRR found\n");
 615
 616         if (list_empty(&dmar_atsr_units))
 617                 printk(KERN_INFO PREFIX "No ATSR found\n");
 618 #endif
 619
 620         return 0;
 621 }
 622
 623 static void warn_invalid_dmar(u64 addr, const char *message)
 624 {
 625         WARN_TAINT_ONCE(
 626                 1, TAINT_FIRMWARE_WORKAROUND,
 627                 "Your BIOS is broken; DMAR reported at address %llx%s!\n"
 628                 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
 629                 addr, message,
 630                 dmi_get_system_info(DMI_BIOS_VENDOR),
 631                 dmi_get_system_info(DMI_BIOS_VERSION),
 632                 dmi_get_system_info(DMI_PRODUCT_VERSION));
 633 }
 634
 635 int __init check_zero_address(void)
 636 {
 637         struct acpi_table_dmar *dmar;
 638         struct acpi_dmar_header *entry_header;
 639         struct acpi_dmar_hardware_unit *drhd;
 640
 641         dmar = (struct acpi_table_dmar *)dmar_tbl;
 642         entry_header = (struct acpi_dmar_header *)(dmar + 1);
 643
 644         while (((unsigned long)entry_header) <
 645                         (((unsigned long)dmar) + dmar_tbl->length)) {
 646                 /* Avoid looping forever on bad ACPI tables */
 647                 if (entry_header->length == 0) {
 648                         printk(KERN_WARNING PREFIX
 649                                 "Invalid 0-length structure\n");
 650                         return 0;
 651                 }
 652
 653                 if (entry_header->type == ACPI_DMAR_TYPE_HARDWARE_UNIT) {
 654                         void __iomem *addr;
 655                         u64 cap, ecap;
 656
 657                         drhd = (void *)entry_header;
 658                         if (!drhd->address) {
 659                                 warn_invalid_dmar(0, "");
 660                                 goto failed;
 661                         }
 662
 663                         addr = early_ioremap(drhd->address, VTD_PAGE_SIZE);
 664                         if (!addr ) {
 665                                 printk("IOMMU: can't validate: %llx\n", drhd->address);
 666                                 goto failed;
 667                         }
 668                         cap = dmar_readq(addr + DMAR_CAP_REG);
 669                         ecap = dmar_readq(addr + DMAR_ECAP_REG);
 670                         early_iounmap(addr, VTD_PAGE_SIZE);
 671                         if (cap == (uint64_t)-1 && ecap == (uint64_t)-1) {
 672                                 warn_invalid_dmar(drhd->address,
 673                                                   " returns all ones");
 674                                 goto failed;
 675                         }
 676                 }
 677
 678                 entry_header = ((void *)entry_header + entry_header->length);
 679         }
 680         return 1;
 681
 682 failed:
 683 #ifdef CONFIG_DMAR
 684         dmar_disabled = 1;
 685 #endif
 686         return 0;
 687 }
 688
 689 void __init detect_intel_iommu(void)
 690 {
 691         int ret;
 692
 693         ret = dmar_table_detect();
 694         if (ret)
 695                 ret = check_zero_address();
 696         {
 697 #ifdef CONFIG_INTR_REMAP
 698                 struct acpi_table_dmar *dmar;
 699                 /*
 700                  * for now we will disable dma-remapping when interrupt
 701                  * remapping is enabled.
 702                  * When support for queued invalidation for IOTLB invalidation
 703                  * is added, we will not need this any more.
 704                  */
 705                 dmar = (struct acpi_table_dmar *) dmar_tbl;
 706                 if (ret && cpu_has_x2apic && dmar->flags & 0x1)
 707                         printk(KERN_INFO
 708                                "Queued invalidation will be enabled to support "
 709                                "x2apic and Intr-remapping.\n");
 710 #endif
 711 #ifdef CONFIG_DMAR
 712                 if (ret && !no_iommu && !iommu_detected && !dmar_disabled) {
 713                         iommu_detected = 1;
 714                         /* Make sure ACS will be enabled */
 715                         pci_request_acs();
 716                 }
 717 #endif
 718 #ifdef CONFIG_X86
 719                 if (ret)
 720                         x86_init.iommu.iommu_init = intel_iommu_init;
 721 #endif
 722         }
 723         early_acpi_os_unmap_memory(dmar_tbl, dmar_tbl_size);
 724         dmar_tbl = NULL;
 725 }
 726
 727
 728 int alloc_iommu(struct dmar_drhd_unit *drhd)
 729 {
 730         struct intel_iommu *iommu;
 731         int map_size;
 732         u32 ver;
 733         static int iommu_allocated = 0;
 734         int agaw = 0;
 735         int msagaw = 0;
 736
 737         if (!drhd->reg_base_addr) {
 738                 warn_invalid_dmar(0, "");
 739                 return -EINVAL;
 740         }
 741
 742         iommu = kzalloc(sizeof(*iommu), GFP_KERNEL);
 743         if (!iommu)
 744                 return -ENOMEM;
 745
 746         iommu->seq_id = iommu_allocated++;
 747         sprintf (iommu->name, "dmar%d", iommu->seq_id);
 748
 749         iommu->reg = ioremap(drhd->reg_base_addr, VTD_PAGE_SIZE);
 750         if (!iommu->reg) {
 751                 printk(KERN_ERR "IOMMU: can't map the region\n");
 752                 goto error;
 753         }
 754         iommu->cap = dmar_readq(iommu->reg + DMAR_CAP_REG);
 755         iommu->ecap = dmar_readq(iommu->reg + DMAR_ECAP_REG);
 756
 757         if (iommu->cap == (uint64_t)-1 && iommu->ecap == (uint64_t)-1) {
 758                 warn_invalid_dmar(drhd->reg_base_addr, " returns all ones");
 759                 goto err_unmap;
 760         }
 761
 762 #ifdef CONFIG_DMAR
 763         agaw = iommu_calculate_agaw(iommu);
 764         if (agaw < 0) {
 765                 printk(KERN_ERR
 766                        "Cannot get a valid agaw for iommu (seq_id = %d)\n",
 767                        iommu->seq_id);
 768                 goto err_unmap;
 769         }
 770         msagaw = iommu_calculate_max_sagaw(iommu);
 771         if (msagaw < 0) {
 772                 printk(KERN_ERR
 773                         "Cannot get a valid max agaw for iommu (seq_id = %d)\n",
 774                         iommu->seq_id);
 775                 goto err_unmap;
 776         }
 777 #endif
 778         iommu->agaw = agaw;
 779         iommu->msagaw = msagaw;
 780
 781         iommu->node = -1;
 782
 783         /* the registers might be more than one page */
 784         map_size = max_t(int, ecap_max_iotlb_offset(iommu->ecap),
 785                 cap_max_fault_reg_offset(iommu->cap));
 786         map_size = VTD_PAGE_ALIGN(map_size);
 787         if (map_size > VTD_PAGE_SIZE) {
 788                 iounmap(iommu->reg);
 789                 iommu->reg = ioremap(drhd->reg_base_addr, map_size);
 790                 if (!iommu->reg) {
 791                         printk(KERN_ERR "IOMMU: can't map the region\n");
 792                         goto error;
 793                 }
 794         }
 795
 796         ver = readl(iommu->reg + DMAR_VER_REG);
 797         pr_info("IOMMU %d: reg_base_addr %llx ver %d:%d cap %llx ecap %llx\n",
 798                 iommu->seq_id,
 799                 (unsigned long long)drhd->reg_base_addr,
 800                 DMAR_VER_MAJOR(ver), DMAR_VER_MINOR(ver),
 801                 (unsigned long long)iommu->cap,
 802                 (unsigned long long)iommu->ecap);
 803
 804         spin_lock_init(&iommu->register_lock);
 805
 806         drhd->iommu = iommu;
 807         return 0;
 808
 809  err_unmap:
 810         iounmap(iommu->reg);
 811  error:
 812         kfree(iommu);
 813         return -1;
 814 }
 815
 816 void free_iommu(struct intel_iommu *iommu)
 817 {
 818         if (!iommu)
 819                 return;
 820
 821 #ifdef CONFIG_DMAR
 822         free_dmar_iommu(iommu);
 823 #endif
 824
 825         if (iommu->reg)
 826                 iounmap(iommu->reg);
 827         kfree(iommu);
 828 }
 829
 830 /*
 831  * Reclaim all the submitted descriptors which have completed its work.
 832  */
 833 static inline void reclaim_free_desc(struct q_inval *qi)
 834 {
 835         while (qi->desc_status[qi->free_tail] == QI_DONE ||
 836                qi->desc_status[qi->free_tail] == QI_ABORT) {
 837                 qi->desc_status[qi->free_tail] = QI_FREE;
 838                 qi->free_tail = (qi->free_tail + 1) % QI_LENGTH;
 839                 qi->free_cnt++;
 840         }
 841 }
 842
 843 static int qi_check_fault(struct intel_iommu *iommu, int index)
 844 {
 845         u32 fault;
 846         int head, tail;
 847         struct q_inval *qi = iommu->qi;
 848         int wait_index = (index + 1) % QI_LENGTH;
 849
 850         if (qi->desc_status[wait_index] == QI_ABORT)
 851                 return -EAGAIN;
 852
 853         fault = readl(iommu->reg + DMAR_FSTS_REG);
 854
 855         /*
 856          * If IQE happens, the head points to the descriptor associated
 857          * with the error. No new descriptors are fetched until the IQE
 858          * is cleared.
 859          */
 860         if (fault & DMA_FSTS_IQE) {
 861                 head = readl(iommu->reg + DMAR_IQH_REG);
 862                 if ((head >> DMAR_IQ_SHIFT) == index) {
 863                         printk(KERN_ERR "VT-d detected invalid descriptor: "
 864                                 "low=%llx, high=%llx\n",
 865                                 (unsigned long long)qi->desc[index].low,
 866                                 (unsigned long long)qi->desc[index].high);
 867                         memcpy(&qi->desc[index], &qi->desc[wait_index],
 868                                         sizeof(struct qi_desc));
 869                         __iommu_flush_cache(iommu, &qi->desc[index],
 870                                         sizeof(struct qi_desc));
 871                         writel(DMA_FSTS_IQE, iommu->reg + DMAR_FSTS_REG);
 872                         return -EINVAL;
 873                 }
 874         }
 875
 876         /*
 877          * If ITE happens, all pending wait_desc commands are aborted.
 878          * No new descriptors are fetched until the ITE is cleared.
 879          */
 880         if (fault & DMA_FSTS_ITE) {
 881                 head = readl(iommu->reg + DMAR_IQH_REG);
 882                 head = ((head >> DMAR_IQ_SHIFT) - 1 + QI_LENGTH) % QI_LENGTH;
 883                 head |= 1;
 884                 tail = readl(iommu->reg + DMAR_IQT_REG);
 885                 tail = ((tail >> DMAR_IQ_SHIFT) - 1 + QI_LENGTH) % QI_LENGTH;
 886
 887                 writel(DMA_FSTS_ITE, iommu->reg + DMAR_FSTS_REG);
 888
 889                 do {
 890                         if (qi->desc_status[head] == QI_IN_USE)
 891                                 qi->desc_status[head] = QI_ABORT;
 892                         head = (head - 2 + QI_LENGTH) % QI_LENGTH;
 893                 } while (head != tail);
 894
 895                 if (qi->desc_status[wait_index] == QI_ABORT)
 896                         return -EAGAIN;
 897         }
 898
 899         if (fault & DMA_FSTS_ICE)
 900                 writel(DMA_FSTS_ICE, iommu->reg + DMAR_FSTS_REG);
 901
 902         return 0;
 903 }
 904
 905 /*
 906  * Submit the queued invalidation descriptor to the remapping
 907  * hardware unit and wait for its completion.
 908  */
 909 int qi_submit_sync(struct qi_desc *desc, struct intel_iommu *iommu)
 910 {
 911         int rc;
 912         struct q_inval *qi = iommu->qi;
 913         struct qi_desc *hw, wait_desc;
 914         int wait_index, index;
 915         unsigned long flags;
 916
 917         if (!qi)
 918                 return 0;
 919
 920         hw = qi->desc;
 921
 922 restart:
 923         rc = 0;
 924
 925         spin_lock_irqsave(&qi->q_lock, flags);
 926         while (qi->free_cnt < 3) {
 927                 spin_unlock_irqrestore(&qi->q_lock, flags);
 928                 cpu_relax();
 929                 spin_lock_irqsave(&qi->q_lock, flags);
 930         }
 931
 932         index = qi->free_head;
 933         wait_index = (index + 1) % QI_LENGTH;
 934
 935         qi->desc_status[index] = qi->desc_status[wait_index] = QI_IN_USE;
 936
 937         hw[index] = *desc;
 938
 939         wait_desc.low = QI_IWD_STATUS_DATA(QI_DONE) |
 940                         QI_IWD_STATUS_WRITE | QI_IWD_TYPE;
 941         wait_desc.high = virt_to_phys(&qi->desc_status[wait_index]);
 942
 943         hw[wait_index] = wait_desc;
 944
 945         __iommu_flush_cache(iommu, &hw[index], sizeof(struct qi_desc));
 946         __iommu_flush_cache(iommu, &hw[wait_index], sizeof(struct qi_desc));
 947
 948         qi->free_head = (qi->free_head + 2) % QI_LENGTH;
 949         qi->free_cnt -= 2;
 950
 951         /*
 952          * update the HW tail register indicating the presence of
 953          * new descriptors.
 954          */
 955         writel(qi->free_head << DMAR_IQ_SHIFT, iommu->reg + DMAR_IQT_REG);
 956
 957         while (qi->desc_status[wait_index] != QI_DONE) {
 958                 /*
 959                  * We will leave the interrupts disabled, to prevent interrupt
 960                  * context to queue another cmd while a cmd is already submitted
 961                  * and waiting for completion on this cpu. This is to avoid
 962                  * a deadlock where the interrupt context can wait indefinitely
 963                  * for free slots in the queue.
 964                  */
 965                 rc = qi_check_fault(iommu, index);
 966                 if (rc)
 967                         break;
 968
 969                 spin_unlock(&qi->q_lock);
 970                 cpu_relax();
 971                 spin_lock(&qi->q_lock);
 972         }
 973
 974         qi->desc_status[index] = QI_DONE;
 975
 976         reclaim_free_desc(qi);
 977         spin_unlock_irqrestore(&qi->q_lock, flags);
 978
 979         if (rc == -EAGAIN)
 980                 goto restart;
 981
 982         return rc;
 983 }
 984
 985 /*
 986  * Flush the global interrupt entry cache.
 987  */
 988 void qi_global_iec(struct intel_iommu *iommu)
 989 {
 990         struct qi_desc desc;
 991
 992         desc.low = QI_IEC_TYPE;
 993         desc.high = 0;
 994
 995         /* should never fail */
 996         qi_submit_sync(&desc, iommu);
 997 }
 998
 999 void qi_flush_context(struct intel_iommu *iommu, u16 did, u16 sid, u8 fm,
1000                       u64 type)
1001 {
1002         struct qi_desc desc;
1003
1004         desc.low = QI_CC_FM(fm) | QI_CC_SID(sid) | QI_CC_DID(did)
1005                         | QI_CC_GRAN(type) | QI_CC_TYPE;
1006         desc.high = 0;
1007
1008         qi_submit_sync(&desc, iommu);
1009 }
1010
1011 void qi_flush_iotlb(struct intel_iommu *iommu, u16 did, u64 addr,
1012                     unsigned int size_order, u64 type)
1013 {
1014         u8 dw = 0, dr = 0;
1015
1016         struct qi_desc desc;
1017         int ih = 0;
1018
1019         if (cap_write_drain(iommu->cap))
1020                 dw = 1;
1021
1022         if (cap_read_drain(iommu->cap))
1023                 dr = 1;
1024
1025         desc.low = QI_IOTLB_DID(did) | QI_IOTLB_DR(dr) | QI_IOTLB_DW(dw)
1026                 | QI_IOTLB_GRAN(type) | QI_IOTLB_TYPE;
1027         desc.high = QI_IOTLB_ADDR(addr) | QI_IOTLB_IH(ih)
1028                 | QI_IOTLB_AM(size_order);
1029
1030         qi_submit_sync(&desc, iommu);
1031 }
1032
1033 void qi_flush_dev_iotlb(struct intel_iommu *iommu, u16 sid, u16 qdep,
1034                         u64 addr, unsigned mask)
1035 {
1036         struct qi_desc desc;
1037
1038         if (mask) {
1039                 BUG_ON(addr & ((1 << (VTD_PAGE_SHIFT + mask)) - 1));
1040                 addr |= (1 << (VTD_PAGE_SHIFT + mask - 1)) - 1;
1041                 desc.high = QI_DEV_IOTLB_ADDR(addr) | QI_DEV_IOTLB_SIZE;
1042         } else
1043                 desc.high = QI_DEV_IOTLB_ADDR(addr);
1044
1045         if (qdep >= QI_DEV_IOTLB_MAX_INVS)
1046                 qdep = 0;
1047
1048         desc.low = QI_DEV_IOTLB_SID(sid) | QI_DEV_IOTLB_QDEP(qdep) |
1049                    QI_DIOTLB_TYPE;
1050
1051         qi_submit_sync(&desc, iommu);
1052 }
1053
1054 /*
1055  * Disable Queued Invalidation interface.
1056  */
1057 void dmar_disable_qi(struct intel_iommu *iommu)
1058 {
1059         unsigned long flags;
1060         u32 sts;
1061         cycles_t start_time = get_cycles();
1062
1063         if (!ecap_qis(iommu->ecap))
1064                 return;
1065
1066         spin_lock_irqsave(&iommu->register_lock, flags);
1067
1068         sts =  dmar_readq(iommu->reg + DMAR_GSTS_REG);
1069         if (!(sts & DMA_GSTS_QIES))
1070                 goto end;
1071
1072         /*
1073          * Give a chance to HW to complete the pending invalidation requests.
1074          */
1075         while ((readl(iommu->reg + DMAR_IQT_REG) !=
1076                 readl(iommu->reg + DMAR_IQH_REG)) &&
1077                 (DMAR_OPERATION_TIMEOUT > (get_cycles() - start_time)))
1078                 cpu_relax();
1079
1080         iommu->gcmd &= ~DMA_GCMD_QIE;
1081         writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
1082
1083         IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, readl,
1084                       !(sts & DMA_GSTS_QIES), sts);
1085 end:
1086         spin_unlock_irqrestore(&iommu->register_lock, flags);
1087 }
1088
1089 /*
1090  * Enable queued invalidation.
1091  */
1092 static void __dmar_enable_qi(struct intel_iommu *iommu)
1093 {
1094         u32 sts;
1095         unsigned long flags;
1096         struct q_inval *qi = iommu->qi;
1097
1098         qi->free_head = qi->free_tail = 0;
1099         qi->free_cnt = QI_LENGTH;
1100
1101         spin_lock_irqsave(&iommu->register_lock, flags);
1102
1103         /* write zero to the tail reg */
1104         writel(0, iommu->reg + DMAR_IQT_REG);
1105
1106         dmar_writeq(iommu->reg + DMAR_IQA_REG, virt_to_phys(qi->desc));
1107
1108         iommu->gcmd |= DMA_GCMD_QIE;
1109         writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
1110
1111         /* Make sure hardware complete it */
1112         IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, readl, (sts & DMA_GSTS_QIES), sts);
1113
1114         spin_unlock_irqrestore(&iommu->register_lock, flags);
1115 }
1116
1117 /*
1118  * Enable Queued Invalidation interface. This is a must to support
1119  * interrupt-remapping. Also used by DMA-remapping, which replaces
1120  * register based IOTLB invalidation.
1121  */
1122 int dmar_enable_qi(struct intel_iommu *iommu)
1123 {
1124         struct q_inval *qi;
1125         struct page *desc_page;
1126
1127         if (!ecap_qis(iommu->ecap))
1128                 return -ENOENT;
1129
1130         /*
1131          * queued invalidation is already setup and enabled.
1132          */
1133         if (iommu->qi)
1134                 return 0;
1135
1136         iommu->qi = kmalloc(sizeof(*qi), GFP_ATOMIC);
1137         if (!iommu->qi)
1138                 return -ENOMEM;
1139
1140         qi = iommu->qi;
1141
1142
1143         desc_page = alloc_pages_node(iommu->node, GFP_ATOMIC | __GFP_ZERO, 0);
1144         if (!desc_page) {
1145                 kfree(qi);
1146                 iommu->qi = 0;
1147                 return -ENOMEM;
1148         }
1149
1150         qi->desc = page_address(desc_page);
1151
1152         qi->desc_status = kmalloc(QI_LENGTH * sizeof(int), GFP_ATOMIC);
1153         if (!qi->desc_status) {
1154                 free_page((unsigned long) qi->desc);
1155                 kfree(qi);
1156                 iommu->qi = 0;
1157                 return -ENOMEM;
1158         }
1159
1160         qi->free_head = qi->free_tail = 0;
1161         qi->free_cnt = QI_LENGTH;
1162
1163         spin_lock_init(&qi->q_lock);
1164
1165         __dmar_enable_qi(iommu);
1166
1167         return 0;
1168 }
1169
1170 /* iommu interrupt handling. Most stuff are MSI-like. */
1171
1172 enum faulttype {
1173         DMA_REMAP,
1174         INTR_REMAP,
1175         UNKNOWN,
1176 };
1177
1178 static const char *dma_remap_fault_reasons[] =
1179 {
1180         "Software",
1181         "Present bit in root entry is clear",
1182         "Present bit in context entry is clear",
1183         "Invalid context entry",
1184         "Access beyond MGAW",
1185         "PTE Write access is not set",
1186         "PTE Read access is not set",
1187         "Next page table ptr is invalid",
1188         "Root table address invalid",
1189         "Context table ptr is invalid",
1190         "non-zero reserved fields in RTP",
1191         "non-zero reserved fields in CTP",
1192         "non-zero reserved fields in PTE",
1193 };
1194
1195 static const char *intr_remap_fault_reasons[] =
1196 {
1197         "Detected reserved fields in the decoded interrupt-remapped request",
1198         "Interrupt index exceeded the interrupt-remapping table size",
1199         "Present field in the IRTE entry is clear",
1200         "Error accessing interrupt-remapping table pointed by IRTA_REG",
1201         "Detected reserved fields in the IRTE entry",
1202         "Blocked a compatibility format interrupt request",
1203         "Blocked an interrupt request due to source-id verification failure",
1204 };
1205
1206 #define MAX_FAULT_REASON_IDX    (ARRAY_SIZE(fault_reason_strings) - 1)
1207
1208 const char *dmar_get_fault_reason(u8 fault_reason, int *fault_type)
1209 {
1210         if (fault_reason >= 0x20 && (fault_reason <= 0x20 +
1211                                      ARRAY_SIZE(intr_remap_fault_reasons))) {
1212                 *fault_type = INTR_REMAP;
1213                 return intr_remap_fault_reasons[fault_reason - 0x20];
1214         } else if (fault_reason < ARRAY_SIZE(dma_remap_fault_reasons)) {
1215                 *fault_type = DMA_REMAP;
1216                 return dma_remap_fault_reasons[fault_reason];
1217         } else {
1218                 *fault_type = UNKNOWN;
1219                 return "Unknown";
1220         }
1221 }
1222
1223 void dmar_msi_unmask(unsigned int irq)
1224 {
1225         struct intel_iommu *iommu = get_irq_data(irq);
1226         unsigned long flag;
1227
1228         /* unmask it */
1229         spin_lock_irqsave(&iommu->register_lock, flag);
1230         writel(0, iommu->reg + DMAR_FECTL_REG);
1231         /* Read a reg to force flush the post write */
1232         readl(iommu->reg + DMAR_FECTL_REG);
1233         spin_unlock_irqrestore(&iommu->register_lock, flag);
1234 }
1235
1236 void dmar_msi_mask(unsigned int irq)
1237 {
1238         unsigned long flag;
1239         struct intel_iommu *iommu = get_irq_data(irq);
1240
1241         /* mask it */
1242         spin_lock_irqsave(&iommu->register_lock, flag);
1243         writel(DMA_FECTL_IM, iommu->reg + DMAR_FECTL_REG);
1244         /* Read a reg to force flush the post write */
1245         readl(iommu->reg + DMAR_FECTL_REG);
1246         spin_unlock_irqrestore(&iommu->register_lock, flag);
1247 }
1248
1249 void dmar_msi_write(int irq, struct msi_msg *msg)
1250 {
1251         struct intel_iommu *iommu = get_irq_data(irq);
1252         unsigned long flag;
1253
1254         spin_lock_irqsave(&iommu->register_lock, flag);
1255         writel(msg->data, iommu->reg + DMAR_FEDATA_REG);
1256         writel(msg->address_lo, iommu->reg + DMAR_FEADDR_REG);
1257         writel(msg->address_hi, iommu->reg + DMAR_FEUADDR_REG);
1258         spin_unlock_irqrestore(&iommu->register_lock, flag);
1259 }
1260
1261 void dmar_msi_read(int irq, struct msi_msg *msg)
1262 {
1263         struct intel_iommu *iommu = get_irq_data(irq);
1264         unsigned long flag;
1265
1266         spin_lock_irqsave(&iommu->register_lock, flag);
1267         msg->data = readl(iommu->reg + DMAR_FEDATA_REG);
1268         msg->address_lo = readl(iommu->reg + DMAR_FEADDR_REG);
1269         msg->address_hi = readl(iommu->reg + DMAR_FEUADDR_REG);
1270         spin_unlock_irqrestore(&iommu->register_lock, flag);
1271 }
1272
1273 static int dmar_fault_do_one(struct intel_iommu *iommu, int type,
1274                 u8 fault_reason, u16 source_id, unsigned long long addr)
1275 {
1276         const char *reason;
1277         int fault_type;
1278
1279         reason = dmar_get_fault_reason(fault_reason, &fault_type);
1280
1281         if (fault_type == INTR_REMAP)
1282                 printk(KERN_ERR "INTR-REMAP: Request device [[%02x:%02x.%d] "
1283                        "fault index %llx\n"
1284                         "INTR-REMAP:[fault reason %02d] %s\n",
1285                         (source_id >> 8), PCI_SLOT(source_id & 0xFF),
1286                         PCI_FUNC(source_id & 0xFF), addr >> 48,
1287                         fault_reason, reason);
1288         else
1289                 printk(KERN_ERR
1290                        "DMAR:[%s] Request device [%02x:%02x.%d] "
1291                        "fault addr %llx \n"
1292                        "DMAR:[fault reason %02d] %s\n",
1293                        (type ? "DMA Read" : "DMA Write"),
1294                        (source_id >> 8), PCI_SLOT(source_id & 0xFF),
1295                        PCI_FUNC(source_id & 0xFF), addr, fault_reason, reason);
1296         return 0;
1297 }
1298
1299 #define PRIMARY_FAULT_REG_LEN (16)
1300 irqreturn_t dmar_fault(int irq, void *dev_id)
1301 {
1302         struct intel_iommu *iommu = dev_id;
1303         int reg, fault_index;
1304         u32 fault_status;
1305         unsigned long flag;
1306
1307         spin_lock_irqsave(&iommu->register_lock, flag);
1308         fault_status = readl(iommu->reg + DMAR_FSTS_REG);
1309         if (fault_status)
1310                 printk(KERN_ERR "DRHD: handling fault status reg %x\n",
1311                        fault_status);
1312
1313         /* TBD: ignore advanced fault log currently */
1314         if (!(fault_status & DMA_FSTS_PPF))
1315                 goto clear_rest;
1316
1317         fault_index = dma_fsts_fault_record_index(fault_status);
1318         reg = cap_fault_reg_offset(iommu->cap);
1319         while (1) {
1320                 u8 fault_reason;
1321                 u16 source_id;
1322                 u64 guest_addr;
1323                 int type;
1324                 u32 data;
1325
1326                 /* highest 32 bits */
1327                 data = readl(iommu->reg + reg +
1328                                 fault_index * PRIMARY_FAULT_REG_LEN + 12);
1329                 if (!(data & DMA_FRCD_F))
1330                         break;
1331
1332                 fault_reason = dma_frcd_fault_reason(data);
1333                 type = dma_frcd_type(data);
1334
1335                 data = readl(iommu->reg + reg +
1336                                 fault_index * PRIMARY_FAULT_REG_LEN + 8);
1337                 source_id = dma_frcd_source_id(data);
1338
1339                 guest_addr = dmar_readq(iommu->reg + reg +
1340                                 fault_index * PRIMARY_FAULT_REG_LEN);
1341                 guest_addr = dma_frcd_page_addr(guest_addr);
1342                 /* clear the fault */
1343                 writel(DMA_FRCD_F, iommu->reg + reg +
1344                         fault_index * PRIMARY_FAULT_REG_LEN + 12);
1345
1346                 spin_unlock_irqrestore(&iommu->register_lock, flag);
1347
1348                 dmar_fault_do_one(iommu, type, fault_reason,
1349                                 source_id, guest_addr);
1350
1351                 fault_index++;
1352                 if (fault_index >= cap_num_fault_regs(iommu->cap))
1353                         fault_index = 0;
1354                 spin_lock_irqsave(&iommu->register_lock, flag);
1355         }
1356 clear_rest:
1357         /* clear all the other faults */
1358         fault_status = readl(iommu->reg + DMAR_FSTS_REG);
1359         writel(fault_status, iommu->reg + DMAR_FSTS_REG);
1360
1361         spin_unlock_irqrestore(&iommu->register_lock, flag);
1362         return IRQ_HANDLED;
1363 }
1364
1365 int dmar_set_interrupt(struct intel_iommu *iommu)
1366 {
1367         int irq, ret;
1368
1369         /*
1370          * Check if the fault interrupt is already initialized.
1371          */
1372         if (iommu->irq)
1373                 return 0;
1374
1375         irq = create_irq();
1376         if (!irq) {
1377                 printk(KERN_ERR "IOMMU: no free vectors\n");
1378                 return -EINVAL;
1379         }
1380
1381         set_irq_data(irq, iommu);
1382         iommu->irq = irq;
1383
1384         ret = arch_setup_dmar_msi(irq);
1385         if (ret) {
1386                 set_irq_data(irq, NULL);
1387                 iommu->irq = 0;
1388                 destroy_irq(irq);
1389                 return ret;
1390         }
1391
1392         ret = request_irq(irq, dmar_fault, 0, iommu->name, iommu);
1393         if (ret)
1394                 printk(KERN_ERR "IOMMU: can't request irq\n");
1395         return ret;
1396 }
1397
1398 int __init enable_drhd_fault_handling(void)
1399 {
1400         struct dmar_drhd_unit *drhd;
1401
1402         /*
1403          * Enable fault control interrupt.
1404          */
1405         for_each_drhd_unit(drhd) {
1406                 int ret;
1407                 struct intel_iommu *iommu = drhd->iommu;
1408                 ret = dmar_set_interrupt(iommu);
1409
1410                 if (ret) {
1411                         printk(KERN_ERR "DRHD %Lx: failed to enable fault, "
1412                                " interrupt, ret %d\n",
1413                                (unsigned long long)drhd->reg_base_addr, ret);
1414                         return -1;
1415                 }
1416         }
1417
1418         return 0;
1419 }
1420
1421 /*
1422  * Re-enable Queued Invalidation interface.
1423  */
1424 int dmar_reenable_qi(struct intel_iommu *iommu)
1425 {
1426         if (!ecap_qis(iommu->ecap))
1427                 return -ENOENT;
1428
1429         if (!iommu->qi)
1430                 return -ENOENT;
1431
1432         /*
1433          * First disable queued invalidation.
1434          */
1435         dmar_disable_qi(iommu);
1436         /*
1437          * Then enable queued invalidation again. Since there is no pending
1438          * invalidation requests now, it's safe to re-enable queued
1439          * invalidation.
1440          */
1441         __dmar_enable_qi(iommu);
1442
1443         return 0;
1444 }
1445
1446 /*
1447  * Check interrupt remapping support in DMAR table description.
1448  */
1449 int __init dmar_ir_support(void)
1450 {
1451         struct acpi_table_dmar *dmar;
1452         dmar = (struct acpi_table_dmar *)dmar_tbl;
1453         if (!dmar)
1454                 return 0;
1455         return dmar->flags & 0x1;
1456 }