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