drivers/iommu/dmar.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (c) 2006, Intel Corporation.
   4  *
   5  * Copyright (C) 2006-2008 Intel Corporation
   6  * Author: Ashok Raj <ashok.raj@intel.com>
   7  * Author: Shaohua Li <shaohua.li@intel.com>
   8  * Author: Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com>
   9  *
  10  * This file implements early detection/parsing of Remapping Devices
  11  * reported to OS through BIOS via DMA remapping reporting (DMAR) ACPI
  12  * tables.
  13  *
  14  * These routines are used by both DMA-remapping and Interrupt-remapping
  15  */
  16
  17 #define pr_fmt(fmt)     "DMAR: " fmt
  18
  19 #include <linux/pci.h>
  20 #include <linux/dmar.h>
  21 #include <linux/iova.h>
  22 #include <linux/intel-iommu.h>
  23 #include <linux/timer.h>
  24 #include <linux/irq.h>
  25 #include <linux/interrupt.h>
  26 #include <linux/tboot.h>
  27 #include <linux/dmi.h>
  28 #include <linux/slab.h>
  29 #include <linux/iommu.h>
  30 #include <linux/numa.h>
  31 #include <asm/irq_remapping.h>
  32 #include <asm/iommu_table.h>
  33
  34 #include "irq_remapping.h"
  35
  36 typedef int (*dmar_res_handler_t)(struct acpi_dmar_header *, void *);
  37 struct dmar_res_callback {
  38         dmar_res_handler_t      cb[ACPI_DMAR_TYPE_RESERVED];
  39         void                    *arg[ACPI_DMAR_TYPE_RESERVED];
  40         bool                    ignore_unhandled;
  41         bool                    print_entry;
  42 };
  43
  44 /*
  45  * Assumptions:
  46  * 1) The hotplug framework guarentees that DMAR unit will be hot-added
  47  *    before IO devices managed by that unit.
  48  * 2) The hotplug framework guarantees that DMAR unit will be hot-removed
  49  *    after IO devices managed by that unit.
  50  * 3) Hotplug events are rare.
  51  *
  52  * Locking rules for DMA and interrupt remapping related global data structures:
  53  * 1) Use dmar_global_lock in process context
  54  * 2) Use RCU in interrupt context
  55  */
  56 DECLARE_RWSEM(dmar_global_lock);
  57 LIST_HEAD(dmar_drhd_units);
  58
  59 struct acpi_table_header * __initdata dmar_tbl;
  60 static int dmar_dev_scope_status = 1;
  61 static unsigned long dmar_seq_ids[BITS_TO_LONGS(DMAR_UNITS_SUPPORTED)];
  62
  63 static int alloc_iommu(struct dmar_drhd_unit *drhd);
  64 static void free_iommu(struct intel_iommu *iommu);
  65
  66 extern const struct iommu_ops intel_iommu_ops;
  67
  68 static void dmar_register_drhd_unit(struct dmar_drhd_unit *drhd)
  69 {
  70         /*
  71          * add INCLUDE_ALL at the tail, so scan the list will find it at
  72          * the very end.
  73          */
  74         if (drhd->include_all)
  75                 list_add_tail_rcu(&drhd->list, &dmar_drhd_units);
  76         else
  77                 list_add_rcu(&drhd->list, &dmar_drhd_units);
  78 }
  79
  80 void *dmar_alloc_dev_scope(void *start, void *end, int *cnt)
  81 {
  82         struct acpi_dmar_device_scope *scope;
  83
  84         *cnt = 0;
  85         while (start < end) {
  86                 scope = start;
  87                 if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_NAMESPACE ||
  88                     scope->entry_type == ACPI_DMAR_SCOPE_TYPE_ENDPOINT ||
  89                     scope->entry_type == ACPI_DMAR_SCOPE_TYPE_BRIDGE)
  90                         (*cnt)++;
  91                 else if (scope->entry_type != ACPI_DMAR_SCOPE_TYPE_IOAPIC &&
  92                         scope->entry_type != ACPI_DMAR_SCOPE_TYPE_HPET) {
  93                         pr_warn("Unsupported device scope\n");
  94                 }
  95                 start += scope->length;
  96         }
  97         if (*cnt == 0)
  98                 return NULL;
  99
 100         return kcalloc(*cnt, sizeof(struct dmar_dev_scope), GFP_KERNEL);
 101 }
 102
 103 void dmar_free_dev_scope(struct dmar_dev_scope **devices, int *cnt)
 104 {
 105         int i;
 106         struct device *tmp_dev;
 107
 108         if (*devices && *cnt) {
 109                 for_each_active_dev_scope(*devices, *cnt, i, tmp_dev)
 110                         put_device(tmp_dev);
 111                 kfree(*devices);
 112         }
 113
 114         *devices = NULL;
 115         *cnt = 0;
 116 }
 117
 118 /* Optimize out kzalloc()/kfree() for normal cases */
 119 static char dmar_pci_notify_info_buf[64];
 120
 121 static struct dmar_pci_notify_info *
 122 dmar_alloc_pci_notify_info(struct pci_dev *dev, unsigned long event)
 123 {
 124         int level = 0;
 125         size_t size;
 126         struct pci_dev *tmp;
 127         struct dmar_pci_notify_info *info;
 128
 129         BUG_ON(dev->is_virtfn);
 130
 131         /* Only generate path[] for device addition event */
 132         if (event == BUS_NOTIFY_ADD_DEVICE)
 133                 for (tmp = dev; tmp; tmp = tmp->bus->self)
 134                         level++;
 135
 136         size = struct_size(info, path, level);
 137         if (size <= sizeof(dmar_pci_notify_info_buf)) {
 138                 info = (struct dmar_pci_notify_info *)dmar_pci_notify_info_buf;
 139         } else {
 140                 info = kzalloc(size, GFP_KERNEL);
 141                 if (!info) {
 142                         pr_warn("Out of memory when allocating notify_info "
 143                                 "for %s.\n", pci_name(dev));
 144                         if (dmar_dev_scope_status == 0)
 145                                 dmar_dev_scope_status = -ENOMEM;
 146                         return NULL;
 147                 }
 148         }
 149
 150         info->event = event;
 151         info->dev = dev;
 152         info->seg = pci_domain_nr(dev->bus);
 153         info->level = level;
 154         if (event == BUS_NOTIFY_ADD_DEVICE) {
 155                 for (tmp = dev; tmp; tmp = tmp->bus->self) {
 156                         level--;
 157                         info->path[level].bus = tmp->bus->number;
 158                         info->path[level].device = PCI_SLOT(tmp->devfn);
 159                         info->path[level].function = PCI_FUNC(tmp->devfn);
 160                         if (pci_is_root_bus(tmp->bus))
 161                                 info->bus = tmp->bus->number;
 162                 }
 163         }
 164
 165         return info;
 166 }
 167
 168 static inline void dmar_free_pci_notify_info(struct dmar_pci_notify_info *info)
 169 {
 170         if ((void *)info != dmar_pci_notify_info_buf)
 171                 kfree(info);
 172 }
 173
 174 static bool dmar_match_pci_path(struct dmar_pci_notify_info *info, int bus,
 175                                 struct acpi_dmar_pci_path *path, int count)
 176 {
 177         int i;
 178
 179         if (info->bus != bus)
 180                 goto fallback;
 181         if (info->level != count)
 182                 goto fallback;
 183
 184         for (i = 0; i < count; i++) {
 185                 if (path[i].device != info->path[i].device ||
 186                     path[i].function != info->path[i].function)
 187                         goto fallback;
 188         }
 189
 190         return true;
 191
 192 fallback:
 193
 194         if (count != 1)
 195                 return false;
 196
 197         i = info->level - 1;
 198         if (bus              == info->path[i].bus &&
 199             path[0].device   == info->path[i].device &&
 200             path[0].function == info->path[i].function) {
 201                 pr_info(FW_BUG "RMRR entry for device %02x:%02x.%x is broken - applying workaround\n",
 202                         bus, path[0].device, path[0].function);
 203                 return true;
 204         }
 205
 206         return false;
 207 }
 208
 209 /* Return: > 0 if match found, 0 if no match found, < 0 if error happens */
 210 int dmar_insert_dev_scope(struct dmar_pci_notify_info *info,
 211                           void *start, void*end, u16 segment,
 212                           struct dmar_dev_scope *devices,
 213                           int devices_cnt)
 214 {
 215         int i, level;
 216         struct device *tmp, *dev = &info->dev->dev;
 217         struct acpi_dmar_device_scope *scope;
 218         struct acpi_dmar_pci_path *path;
 219
 220         if (segment != info->seg)
 221                 return 0;
 222
 223         for (; start < end; start += scope->length) {
 224                 scope = start;
 225                 if (scope->entry_type != ACPI_DMAR_SCOPE_TYPE_ENDPOINT &&
 226                     scope->entry_type != ACPI_DMAR_SCOPE_TYPE_BRIDGE)
 227                         continue;
 228
 229                 path = (struct acpi_dmar_pci_path *)(scope + 1);
 230                 level = (scope->length - sizeof(*scope)) / sizeof(*path);
 231                 if (!dmar_match_pci_path(info, scope->bus, path, level))
 232                         continue;
 233
 234                 /*
 235                  * We expect devices with endpoint scope to have normal PCI
 236                  * headers, and devices with bridge scope to have bridge PCI
 237                  * headers.  However PCI NTB devices may be listed in the
 238                  * DMAR table with bridge scope, even though they have a
 239                  * normal PCI header.  NTB devices are identified by class
 240                  * "BRIDGE_OTHER" (0680h) - we don't declare a socpe mismatch
 241                  * for this special case.
 242                  */
 243                 if ((scope->entry_type == ACPI_DMAR_SCOPE_TYPE_ENDPOINT &&
 244                      info->dev->hdr_type != PCI_HEADER_TYPE_NORMAL) ||
 245                     (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_BRIDGE &&
 246                      (info->dev->hdr_type == PCI_HEADER_TYPE_NORMAL &&
 247                       info->dev->class >> 16 != PCI_BASE_CLASS_BRIDGE))) {
 248                         pr_warn("Device scope type does not match for %s\n",
 249                                 pci_name(info->dev));
 250                         return -EINVAL;
 251                 }
 252
 253                 for_each_dev_scope(devices, devices_cnt, i, tmp)
 254                         if (tmp == NULL) {
 255                                 devices[i].bus = info->dev->bus->number;
 256                                 devices[i].devfn = info->dev->devfn;
 257                                 rcu_assign_pointer(devices[i].dev,
 258                                                    get_device(dev));
 259                                 return 1;
 260                         }
 261                 BUG_ON(i >= devices_cnt);
 262         }
 263
 264         return 0;
 265 }
 266
 267 int dmar_remove_dev_scope(struct dmar_pci_notify_info *info, u16 segment,
 268                           struct dmar_dev_scope *devices, int count)
 269 {
 270         int index;
 271         struct device *tmp;
 272
 273         if (info->seg != segment)
 274                 return 0;
 275
 276         for_each_active_dev_scope(devices, count, index, tmp)
 277                 if (tmp == &info->dev->dev) {
 278                         RCU_INIT_POINTER(devices[index].dev, NULL);
 279                         synchronize_rcu();
 280                         put_device(tmp);
 281                         return 1;
 282                 }
 283
 284         return 0;
 285 }
 286
 287 static int dmar_pci_bus_add_dev(struct dmar_pci_notify_info *info)
 288 {
 289         int ret = 0;
 290         struct dmar_drhd_unit *dmaru;
 291         struct acpi_dmar_hardware_unit *drhd;
 292
 293         for_each_drhd_unit(dmaru) {
 294                 if (dmaru->include_all)
 295                         continue;
 296
 297                 drhd = container_of(dmaru->hdr,
 298                                     struct acpi_dmar_hardware_unit, header);
 299                 ret = dmar_insert_dev_scope(info, (void *)(drhd + 1),
 300                                 ((void *)drhd) + drhd->header.length,
 301                                 dmaru->segment,
 302                                 dmaru->devices, dmaru->devices_cnt);
 303                 if (ret)
 304                         break;
 305         }
 306         if (ret >= 0)
 307                 ret = dmar_iommu_notify_scope_dev(info);
 308         if (ret < 0 && dmar_dev_scope_status == 0)
 309                 dmar_dev_scope_status = ret;
 310
 311         return ret;
 312 }
 313
 314 static void  dmar_pci_bus_del_dev(struct dmar_pci_notify_info *info)
 315 {
 316         struct dmar_drhd_unit *dmaru;
 317
 318         for_each_drhd_unit(dmaru)
 319                 if (dmar_remove_dev_scope(info, dmaru->segment,
 320                         dmaru->devices, dmaru->devices_cnt))
 321                         break;
 322         dmar_iommu_notify_scope_dev(info);
 323 }
 324
 325 static int dmar_pci_bus_notifier(struct notifier_block *nb,
 326                                  unsigned long action, void *data)
 327 {
 328         struct pci_dev *pdev = to_pci_dev(data);
 329         struct dmar_pci_notify_info *info;
 330
 331         /* Only care about add/remove events for physical functions.
 332          * For VFs we actually do the lookup based on the corresponding
 333          * PF in device_to_iommu() anyway. */
 334         if (pdev->is_virtfn)
 335                 return NOTIFY_DONE;
 336         if (action != BUS_NOTIFY_ADD_DEVICE &&
 337             action != BUS_NOTIFY_REMOVED_DEVICE)
 338                 return NOTIFY_DONE;
 339
 340         info = dmar_alloc_pci_notify_info(pdev, action);
 341         if (!info)
 342                 return NOTIFY_DONE;
 343
 344         down_write(&dmar_global_lock);
 345         if (action == BUS_NOTIFY_ADD_DEVICE)
 346                 dmar_pci_bus_add_dev(info);
 347         else if (action == BUS_NOTIFY_REMOVED_DEVICE)
 348                 dmar_pci_bus_del_dev(info);
 349         up_write(&dmar_global_lock);
 350
 351         dmar_free_pci_notify_info(info);
 352
 353         return NOTIFY_OK;
 354 }
 355
 356 static struct notifier_block dmar_pci_bus_nb = {
 357         .notifier_call = dmar_pci_bus_notifier,
 358         .priority = INT_MIN,
 359 };
 360
 361 static struct dmar_drhd_unit *
 362 dmar_find_dmaru(struct acpi_dmar_hardware_unit *drhd)
 363 {
 364         struct dmar_drhd_unit *dmaru;
 365
 366         list_for_each_entry_rcu(dmaru, &dmar_drhd_units, list)
 367                 if (dmaru->segment == drhd->segment &&
 368                     dmaru->reg_base_addr == drhd->address)
 369                         return dmaru;
 370
 371         return NULL;
 372 }
 373
 374 /**
 375  * dmar_parse_one_drhd - parses exactly one DMA remapping hardware definition
 376  * structure which uniquely represent one DMA remapping hardware unit
 377  * present in the platform
 378  */
 379 static int dmar_parse_one_drhd(struct acpi_dmar_header *header, void *arg)
 380 {
 381         struct acpi_dmar_hardware_unit *drhd;
 382         struct dmar_drhd_unit *dmaru;
 383         int ret;
 384
 385         drhd = (struct acpi_dmar_hardware_unit *)header;
 386         dmaru = dmar_find_dmaru(drhd);
 387         if (dmaru)
 388                 goto out;
 389
 390         dmaru = kzalloc(sizeof(*dmaru) + header->length, GFP_KERNEL);
 391         if (!dmaru)
 392                 return -ENOMEM;
 393
 394         /*
 395          * If header is allocated from slab by ACPI _DSM method, we need to
 396          * copy the content because the memory buffer will be freed on return.
 397          */
 398         dmaru->hdr = (void *)(dmaru + 1);
 399         memcpy(dmaru->hdr, header, header->length);
 400         dmaru->reg_base_addr = drhd->address;
 401         dmaru->segment = drhd->segment;
 402         dmaru->include_all = drhd->flags & 0x1; /* BIT0: INCLUDE_ALL */
 403         dmaru->devices = dmar_alloc_dev_scope((void *)(drhd + 1),
 404                                               ((void *)drhd) + drhd->header.length,
 405                                               &dmaru->devices_cnt);
 406         if (dmaru->devices_cnt && dmaru->devices == NULL) {
 407                 kfree(dmaru);
 408                 return -ENOMEM;
 409         }
 410
 411         ret = alloc_iommu(dmaru);
 412         if (ret) {
 413                 dmar_free_dev_scope(&dmaru->devices,
 414                                     &dmaru->devices_cnt);
 415                 kfree(dmaru);
 416                 return ret;
 417         }
 418         dmar_register_drhd_unit(dmaru);
 419
 420 out:
 421         if (arg)
 422                 (*(int *)arg)++;
 423
 424         return 0;
 425 }
 426
 427 static void dmar_free_drhd(struct dmar_drhd_unit *dmaru)
 428 {
 429         if (dmaru->devices && dmaru->devices_cnt)
 430                 dmar_free_dev_scope(&dmaru->devices, &dmaru->devices_cnt);
 431         if (dmaru->iommu)
 432                 free_iommu(dmaru->iommu);
 433         kfree(dmaru);
 434 }
 435
 436 static int __init dmar_parse_one_andd(struct acpi_dmar_header *header,
 437                                       void *arg)
 438 {
 439         struct acpi_dmar_andd *andd = (void *)header;
 440
 441         /* Check for NUL termination within the designated length */
 442         if (strnlen(andd->device_name, header->length - 8) == header->length - 8) {
 443                 WARN_TAINT(1, TAINT_FIRMWARE_WORKAROUND,
 444                            "Your BIOS is broken; ANDD object name is not NUL-terminated\n"
 445                            "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
 446                            dmi_get_system_info(DMI_BIOS_VENDOR),
 447                            dmi_get_system_info(DMI_BIOS_VERSION),
 448                            dmi_get_system_info(DMI_PRODUCT_VERSION));
 449                 return -EINVAL;
 450         }
 451         pr_info("ANDD device: %x name: %s\n", andd->device_number,
 452                 andd->device_name);
 453
 454         return 0;
 455 }
 456
 457 #ifdef CONFIG_ACPI_NUMA
 458 static int dmar_parse_one_rhsa(struct acpi_dmar_header *header, void *arg)
 459 {
 460         struct acpi_dmar_rhsa *rhsa;
 461         struct dmar_drhd_unit *drhd;
 462
 463         rhsa = (struct acpi_dmar_rhsa *)header;
 464         for_each_drhd_unit(drhd) {
 465                 if (drhd->reg_base_addr == rhsa->base_address) {
 466                         int node = acpi_map_pxm_to_node(rhsa->proximity_domain);
 467
 468                         if (!node_online(node))
 469                                 node = NUMA_NO_NODE;
 470                         drhd->iommu->node = node;
 471                         return 0;
 472                 }
 473         }
 474         WARN_TAINT(
 475                 1, TAINT_FIRMWARE_WORKAROUND,
 476                 "Your BIOS is broken; RHSA refers to non-existent DMAR unit at %llx\n"
 477                 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
 478                 drhd->reg_base_addr,
 479                 dmi_get_system_info(DMI_BIOS_VENDOR),
 480                 dmi_get_system_info(DMI_BIOS_VERSION),
 481                 dmi_get_system_info(DMI_PRODUCT_VERSION));
 482
 483         return 0;
 484 }
 485 #else
 486 #define dmar_parse_one_rhsa             dmar_res_noop
 487 #endif
 488
 489 static void
 490 dmar_table_print_dmar_entry(struct acpi_dmar_header *header)
 491 {
 492         struct acpi_dmar_hardware_unit *drhd;
 493         struct acpi_dmar_reserved_memory *rmrr;
 494         struct acpi_dmar_atsr *atsr;
 495         struct acpi_dmar_rhsa *rhsa;
 496
 497         switch (header->type) {
 498         case ACPI_DMAR_TYPE_HARDWARE_UNIT:
 499                 drhd = container_of(header, struct acpi_dmar_hardware_unit,
 500                                     header);
 501                 pr_info("DRHD base: %#016Lx flags: %#x\n",
 502                         (unsigned long long)drhd->address, drhd->flags);
 503                 break;
 504         case ACPI_DMAR_TYPE_RESERVED_MEMORY:
 505                 rmrr = container_of(header, struct acpi_dmar_reserved_memory,
 506                                     header);
 507                 pr_info("RMRR base: %#016Lx end: %#016Lx\n",
 508                         (unsigned long long)rmrr->base_address,
 509                         (unsigned long long)rmrr->end_address);
 510                 break;
 511         case ACPI_DMAR_TYPE_ROOT_ATS:
 512                 atsr = container_of(header, struct acpi_dmar_atsr, header);
 513                 pr_info("ATSR flags: %#x\n", atsr->flags);
 514                 break;
 515         case ACPI_DMAR_TYPE_HARDWARE_AFFINITY:
 516                 rhsa = container_of(header, struct acpi_dmar_rhsa, header);
 517                 pr_info("RHSA base: %#016Lx proximity domain: %#x\n",
 518                        (unsigned long long)rhsa->base_address,
 519                        rhsa->proximity_domain);
 520                 break;
 521         case ACPI_DMAR_TYPE_NAMESPACE:
 522                 /* We don't print this here because we need to sanity-check
 523                    it first. So print it in dmar_parse_one_andd() instead. */
 524                 break;
 525         }
 526 }
 527
 528 /**
 529  * dmar_table_detect - checks to see if the platform supports DMAR devices
 530  */
 531 static int __init dmar_table_detect(void)
 532 {
 533         acpi_status status = AE_OK;
 534
 535         /* if we could find DMAR table, then there are DMAR devices */
 536         status = acpi_get_table(ACPI_SIG_DMAR, 0, &dmar_tbl);
 537
 538         if (ACPI_SUCCESS(status) && !dmar_tbl) {
 539                 pr_warn("Unable to map DMAR\n");
 540                 status = AE_NOT_FOUND;
 541         }
 542
 543         return ACPI_SUCCESS(status) ? 0 : -ENOENT;
 544 }
 545
 546 static int dmar_walk_remapping_entries(struct acpi_dmar_header *start,
 547                                        size_t len, struct dmar_res_callback *cb)
 548 {
 549         struct acpi_dmar_header *iter, *next;
 550         struct acpi_dmar_header *end = ((void *)start) + len;
 551
 552         for (iter = start; iter < end; iter = next) {
 553                 next = (void *)iter + iter->length;
 554                 if (iter->length == 0) {
 555                         /* Avoid looping forever on bad ACPI tables */
 556                         pr_debug(FW_BUG "Invalid 0-length structure\n");
 557                         break;
 558                 } else if (next > end) {
 559                         /* Avoid passing table end */
 560                         pr_warn(FW_BUG "Record passes table end\n");
 561                         return -EINVAL;
 562                 }
 563
 564                 if (cb->print_entry)
 565                         dmar_table_print_dmar_entry(iter);
 566
 567                 if (iter->type >= ACPI_DMAR_TYPE_RESERVED) {
 568                         /* continue for forward compatibility */
 569                         pr_debug("Unknown DMAR structure type %d\n",
 570                                  iter->type);
 571                 } else if (cb->cb[iter->type]) {
 572                         int ret;
 573
 574                         ret = cb->cb[iter->type](iter, cb->arg[iter->type]);
 575                         if (ret)
 576                                 return ret;
 577                 } else if (!cb->ignore_unhandled) {
 578                         pr_warn("No handler for DMAR structure type %d\n",
 579                                 iter->type);
 580                         return -EINVAL;
 581                 }
 582         }
 583
 584         return 0;
 585 }
 586
 587 static inline int dmar_walk_dmar_table(struct acpi_table_dmar *dmar,
 588                                        struct dmar_res_callback *cb)
 589 {
 590         return dmar_walk_remapping_entries((void *)(dmar + 1),
 591                         dmar->header.length - sizeof(*dmar), cb);
 592 }
 593
 594 /**
 595  * parse_dmar_table - parses the DMA reporting table
 596  */
 597 static int __init
 598 parse_dmar_table(void)
 599 {
 600         struct acpi_table_dmar *dmar;
 601         int drhd_count = 0;
 602         int ret;
 603         struct dmar_res_callback cb = {
 604                 .print_entry = true,
 605                 .ignore_unhandled = true,
 606                 .arg[ACPI_DMAR_TYPE_HARDWARE_UNIT] = &drhd_count,
 607                 .cb[ACPI_DMAR_TYPE_HARDWARE_UNIT] = &dmar_parse_one_drhd,
 608                 .cb[ACPI_DMAR_TYPE_RESERVED_MEMORY] = &dmar_parse_one_rmrr,
 609                 .cb[ACPI_DMAR_TYPE_ROOT_ATS] = &dmar_parse_one_atsr,
 610                 .cb[ACPI_DMAR_TYPE_HARDWARE_AFFINITY] = &dmar_parse_one_rhsa,
 611                 .cb[ACPI_DMAR_TYPE_NAMESPACE] = &dmar_parse_one_andd,
 612         };
 613
 614         /*
 615          * Do it again, earlier dmar_tbl mapping could be mapped with
 616          * fixed map.
 617          */
 618         dmar_table_detect();
 619
 620         /*
 621          * ACPI tables may not be DMA protected by tboot, so use DMAR copy
 622          * SINIT saved in SinitMleData in TXT heap (which is DMA protected)
 623          */
 624         dmar_tbl = tboot_get_dmar_table(dmar_tbl);
 625
 626         dmar = (struct acpi_table_dmar *)dmar_tbl;
 627         if (!dmar)
 628                 return -ENODEV;
 629
 630         if (dmar->width < PAGE_SHIFT - 1) {
 631                 pr_warn("Invalid DMAR haw\n");
 632                 return -EINVAL;
 633         }
 634
 635         pr_info("Host address width %d\n", dmar->width + 1);
 636         ret = dmar_walk_dmar_table(dmar, &cb);
 637         if (ret == 0 && drhd_count == 0)
 638                 pr_warn(FW_BUG "No DRHD structure found in DMAR table\n");
 639
 640         return ret;
 641 }
 642
 643 static int dmar_pci_device_match(struct dmar_dev_scope devices[],
 644                                  int cnt, struct pci_dev *dev)
 645 {
 646         int index;
 647         struct device *tmp;
 648
 649         while (dev) {
 650                 for_each_active_dev_scope(devices, cnt, index, tmp)
 651                         if (dev_is_pci(tmp) && dev == to_pci_dev(tmp))
 652                                 return 1;
 653
 654                 /* Check our parent */
 655                 dev = dev->bus->self;
 656         }
 657
 658         return 0;
 659 }
 660
 661 struct dmar_drhd_unit *
 662 dmar_find_matched_drhd_unit(struct pci_dev *dev)
 663 {
 664         struct dmar_drhd_unit *dmaru;
 665         struct acpi_dmar_hardware_unit *drhd;
 666
 667         dev = pci_physfn(dev);
 668
 669         rcu_read_lock();
 670         for_each_drhd_unit(dmaru) {
 671                 drhd = container_of(dmaru->hdr,
 672                                     struct acpi_dmar_hardware_unit,
 673                                     header);
 674
 675                 if (dmaru->include_all &&
 676                     drhd->segment == pci_domain_nr(dev->bus))
 677                         goto out;
 678
 679                 if (dmar_pci_device_match(dmaru->devices,
 680                                           dmaru->devices_cnt, dev))
 681                         goto out;
 682         }
 683         dmaru = NULL;
 684 out:
 685         rcu_read_unlock();
 686
 687         return dmaru;
 688 }
 689
 690 static void __init dmar_acpi_insert_dev_scope(u8 device_number,
 691                                               struct acpi_device *adev)
 692 {
 693         struct dmar_drhd_unit *dmaru;
 694         struct acpi_dmar_hardware_unit *drhd;
 695         struct acpi_dmar_device_scope *scope;
 696         struct device *tmp;
 697         int i;
 698         struct acpi_dmar_pci_path *path;
 699
 700         for_each_drhd_unit(dmaru) {
 701                 drhd = container_of(dmaru->hdr,
 702                                     struct acpi_dmar_hardware_unit,
 703                                     header);
 704
 705                 for (scope = (void *)(drhd + 1);
 706                      (unsigned long)scope < ((unsigned long)drhd) + drhd->header.length;
 707                      scope = ((void *)scope) + scope->length) {
 708                         if (scope->entry_type != ACPI_DMAR_SCOPE_TYPE_NAMESPACE)
 709                                 continue;
 710                         if (scope->enumeration_id != device_number)
 711                                 continue;
 712
 713                         path = (void *)(scope + 1);
 714                         pr_info("ACPI device \"%s\" under DMAR at %llx as %02x:%02x.%d\n",
 715                                 dev_name(&adev->dev), dmaru->reg_base_addr,
 716                                 scope->bus, path->device, path->function);
 717                         for_each_dev_scope(dmaru->devices, dmaru->devices_cnt, i, tmp)
 718                                 if (tmp == NULL) {
 719                                         dmaru->devices[i].bus = scope->bus;
 720                                         dmaru->devices[i].devfn = PCI_DEVFN(path->device,
 721                                                                             path->function);
 722                                         rcu_assign_pointer(dmaru->devices[i].dev,
 723                                                            get_device(&adev->dev));
 724                                         return;
 725                                 }
 726                         BUG_ON(i >= dmaru->devices_cnt);
 727                 }
 728         }
 729         pr_warn("No IOMMU scope found for ANDD enumeration ID %d (%s)\n",
 730                 device_number, dev_name(&adev->dev));
 731 }
 732
 733 static int __init dmar_acpi_dev_scope_init(void)
 734 {
 735         struct acpi_dmar_andd *andd;
 736
 737         if (dmar_tbl == NULL)
 738                 return -ENODEV;
 739
 740         for (andd = (void *)dmar_tbl + sizeof(struct acpi_table_dmar);
 741              ((unsigned long)andd) < ((unsigned long)dmar_tbl) + dmar_tbl->length;
 742              andd = ((void *)andd) + andd->header.length) {
 743                 if (andd->header.type == ACPI_DMAR_TYPE_NAMESPACE) {
 744                         acpi_handle h;
 745                         struct acpi_device *adev;
 746
 747                         if (!ACPI_SUCCESS(acpi_get_handle(ACPI_ROOT_OBJECT,
 748                                                           andd->device_name,
 749                                                           &h))) {
 750                                 pr_err("Failed to find handle for ACPI object %s\n",
 751                                        andd->device_name);
 752                                 continue;
 753                         }
 754                         if (acpi_bus_get_device(h, &adev)) {
 755                                 pr_err("Failed to get device for ACPI object %s\n",
 756                                        andd->device_name);
 757                                 continue;
 758                         }
 759                         dmar_acpi_insert_dev_scope(andd->device_number, adev);
 760                 }
 761         }
 762         return 0;
 763 }
 764
 765 int __init dmar_dev_scope_init(void)
 766 {
 767         struct pci_dev *dev = NULL;
 768         struct dmar_pci_notify_info *info;
 769
 770         if (dmar_dev_scope_status != 1)
 771                 return dmar_dev_scope_status;
 772
 773         if (list_empty(&dmar_drhd_units)) {
 774                 dmar_dev_scope_status = -ENODEV;
 775         } else {
 776                 dmar_dev_scope_status = 0;
 777
 778                 dmar_acpi_dev_scope_init();
 779
 780                 for_each_pci_dev(dev) {
 781                         if (dev->is_virtfn)
 782                                 continue;
 783
 784                         info = dmar_alloc_pci_notify_info(dev,
 785                                         BUS_NOTIFY_ADD_DEVICE);
 786                         if (!info) {
 787                                 return dmar_dev_scope_status;
 788                         } else {
 789                                 dmar_pci_bus_add_dev(info);
 790                                 dmar_free_pci_notify_info(info);
 791                         }
 792                 }
 793         }
 794
 795         return dmar_dev_scope_status;
 796 }
 797
 798 void __init dmar_register_bus_notifier(void)
 799 {
 800         bus_register_notifier(&pci_bus_type, &dmar_pci_bus_nb);
 801 }
 802
 803
 804 int __init dmar_table_init(void)
 805 {
 806         static int dmar_table_initialized;
 807         int ret;
 808
 809         if (dmar_table_initialized == 0) {
 810                 ret = parse_dmar_table();
 811                 if (ret < 0) {
 812                         if (ret != -ENODEV)
 813                                 pr_info("Parse DMAR table failure.\n");
 814                 } else  if (list_empty(&dmar_drhd_units)) {
 815                         pr_info("No DMAR devices found\n");
 816                         ret = -ENODEV;
 817                 }
 818
 819                 if (ret < 0)
 820                         dmar_table_initialized = ret;
 821                 else
 822                         dmar_table_initialized = 1;
 823         }
 824
 825         return dmar_table_initialized < 0 ? dmar_table_initialized : 0;
 826 }
 827
 828 static void warn_invalid_dmar(u64 addr, const char *message)
 829 {
 830         WARN_TAINT_ONCE(
 831                 1, TAINT_FIRMWARE_WORKAROUND,
 832                 "Your BIOS is broken; DMAR reported at address %llx%s!\n"
 833                 "BIOS vendor: %s; Ver: %s; Product Version: %s\n",
 834                 addr, message,
 835                 dmi_get_system_info(DMI_BIOS_VENDOR),
 836                 dmi_get_system_info(DMI_BIOS_VERSION),
 837                 dmi_get_system_info(DMI_PRODUCT_VERSION));
 838 }
 839
 840 static int __ref
 841 dmar_validate_one_drhd(struct acpi_dmar_header *entry, void *arg)
 842 {
 843         struct acpi_dmar_hardware_unit *drhd;
 844         void __iomem *addr;
 845         u64 cap, ecap;
 846
 847         drhd = (void *)entry;
 848         if (!drhd->address) {
 849                 warn_invalid_dmar(0, "");
 850                 return -EINVAL;
 851         }
 852
 853         if (arg)
 854                 addr = ioremap(drhd->address, VTD_PAGE_SIZE);
 855         else
 856                 addr = early_ioremap(drhd->address, VTD_PAGE_SIZE);
 857         if (!addr) {
 858                 pr_warn("Can't validate DRHD address: %llx\n", drhd->address);
 859                 return -EINVAL;
 860         }
 861
 862         cap = dmar_readq(addr + DMAR_CAP_REG);
 863         ecap = dmar_readq(addr + DMAR_ECAP_REG);
 864
 865         if (arg)
 866                 iounmap(addr);
 867         else
 868                 early_iounmap(addr, VTD_PAGE_SIZE);
 869
 870         if (cap == (uint64_t)-1 && ecap == (uint64_t)-1) {
 871                 warn_invalid_dmar(drhd->address, " returns all ones");
 872                 return -EINVAL;
 873         }
 874
 875         return 0;
 876 }
 877
 878 int __init detect_intel_iommu(void)
 879 {
 880         int ret;
 881         struct dmar_res_callback validate_drhd_cb = {
 882                 .cb[ACPI_DMAR_TYPE_HARDWARE_UNIT] = &dmar_validate_one_drhd,
 883                 .ignore_unhandled = true,
 884         };
 885
 886         down_write(&dmar_global_lock);
 887         ret = dmar_table_detect();
 888         if (!ret)
 889                 ret = dmar_walk_dmar_table((struct acpi_table_dmar *)dmar_tbl,
 890                                            &validate_drhd_cb);
 891         if (!ret && !no_iommu && !iommu_detected && !dmar_disabled) {
 892                 iommu_detected = 1;
 893                 /* Make sure ACS will be enabled */
 894                 pci_request_acs();
 895         }
 896
 897 #ifdef CONFIG_X86
 898         if (!ret) {
 899                 x86_init.iommu.iommu_init = intel_iommu_init;
 900                 x86_platform.iommu_shutdown = intel_iommu_shutdown;
 901         }
 902
 903 #endif
 904
 905         if (dmar_tbl) {
 906                 acpi_put_table(dmar_tbl);
 907                 dmar_tbl = NULL;
 908         }
 909         up_write(&dmar_global_lock);
 910
 911         return ret ? ret : 1;
 912 }
 913
 914 static void unmap_iommu(struct intel_iommu *iommu)
 915 {
 916         iounmap(iommu->reg);
 917         release_mem_region(iommu->reg_phys, iommu->reg_size);
 918 }
 919
 920 /**
 921  * map_iommu: map the iommu's registers
 922  * @iommu: the iommu to map
 923  * @phys_addr: the physical address of the base resgister
 924  *
 925  * Memory map the iommu's registers.  Start w/ a single page, and
 926  * possibly expand if that turns out to be insufficent.
 927  */
 928 static int map_iommu(struct intel_iommu *iommu, u64 phys_addr)
 929 {
 930         int map_size, err=0;
 931
 932         iommu->reg_phys = phys_addr;
 933         iommu->reg_size = VTD_PAGE_SIZE;
 934
 935         if (!request_mem_region(iommu->reg_phys, iommu->reg_size, iommu->name)) {
 936                 pr_err("Can't reserve memory\n");
 937                 err = -EBUSY;
 938                 goto out;
 939         }
 940
 941         iommu->reg = ioremap(iommu->reg_phys, iommu->reg_size);
 942         if (!iommu->reg) {
 943                 pr_err("Can't map the region\n");
 944                 err = -ENOMEM;
 945                 goto release;
 946         }
 947
 948         iommu->cap = dmar_readq(iommu->reg + DMAR_CAP_REG);
 949         iommu->ecap = dmar_readq(iommu->reg + DMAR_ECAP_REG);
 950
 951         if (iommu->cap == (uint64_t)-1 && iommu->ecap == (uint64_t)-1) {
 952                 err = -EINVAL;
 953                 warn_invalid_dmar(phys_addr, " returns all ones");
 954                 goto unmap;
 955         }
 956
 957         /* the registers might be more than one page */
 958         map_size = max_t(int, ecap_max_iotlb_offset(iommu->ecap),
 959                          cap_max_fault_reg_offset(iommu->cap));
 960         map_size = VTD_PAGE_ALIGN(map_size);
 961         if (map_size > iommu->reg_size) {
 962                 iounmap(iommu->reg);
 963                 release_mem_region(iommu->reg_phys, iommu->reg_size);
 964                 iommu->reg_size = map_size;
 965                 if (!request_mem_region(iommu->reg_phys, iommu->reg_size,
 966                                         iommu->name)) {
 967                         pr_err("Can't reserve memory\n");
 968                         err = -EBUSY;
 969                         goto out;
 970                 }
 971                 iommu->reg = ioremap(iommu->reg_phys, iommu->reg_size);
 972                 if (!iommu->reg) {
 973                         pr_err("Can't map the region\n");
 974                         err = -ENOMEM;
 975                         goto release;
 976                 }
 977         }
 978         err = 0;
 979         goto out;
 980
 981 unmap:
 982         iounmap(iommu->reg);
 983 release:
 984         release_mem_region(iommu->reg_phys, iommu->reg_size);
 985 out:
 986         return err;
 987 }
 988
 989 static int dmar_alloc_seq_id(struct intel_iommu *iommu)
 990 {
 991         iommu->seq_id = find_first_zero_bit(dmar_seq_ids,
 992                                             DMAR_UNITS_SUPPORTED);
 993         if (iommu->seq_id >= DMAR_UNITS_SUPPORTED) {
 994                 iommu->seq_id = -1;
 995         } else {
 996                 set_bit(iommu->seq_id, dmar_seq_ids);
 997                 sprintf(iommu->name, "dmar%d", iommu->seq_id);
 998         }
 999
1000         return iommu->seq_id;
1001 }
1002
1003 static void dmar_free_seq_id(struct intel_iommu *iommu)
1004 {
1005         if (iommu->seq_id >= 0) {
1006                 clear_bit(iommu->seq_id, dmar_seq_ids);
1007                 iommu->seq_id = -1;
1008         }
1009 }
1010
1011 static int alloc_iommu(struct dmar_drhd_unit *drhd)
1012 {
1013         struct intel_iommu *iommu;
1014         u32 ver, sts;
1015         int agaw = 0;
1016         int msagaw = 0;
1017         int err;
1018
1019         if (!drhd->reg_base_addr) {
1020                 warn_invalid_dmar(0, "");
1021                 return -EINVAL;
1022         }
1023
1024         iommu = kzalloc(sizeof(*iommu), GFP_KERNEL);
1025         if (!iommu)
1026                 return -ENOMEM;
1027
1028         if (dmar_alloc_seq_id(iommu) < 0) {
1029                 pr_err("Failed to allocate seq_id\n");
1030                 err = -ENOSPC;
1031                 goto error;
1032         }
1033
1034         err = map_iommu(iommu, drhd->reg_base_addr);
1035         if (err) {
1036                 pr_err("Failed to map %s\n", iommu->name);
1037                 goto error_free_seq_id;
1038         }
1039
1040         err = -EINVAL;
1041         agaw = iommu_calculate_agaw(iommu);
1042         if (agaw < 0) {
1043                 pr_err("Cannot get a valid agaw for iommu (seq_id = %d)\n",
1044                         iommu->seq_id);
1045                 goto err_unmap;
1046         }
1047         msagaw = iommu_calculate_max_sagaw(iommu);
1048         if (msagaw < 0) {
1049                 pr_err("Cannot get a valid max agaw for iommu (seq_id = %d)\n",
1050                         iommu->seq_id);
1051                 goto err_unmap;
1052         }
1053         iommu->agaw = agaw;
1054         iommu->msagaw = msagaw;
1055         iommu->segment = drhd->segment;
1056
1057         iommu->node = NUMA_NO_NODE;
1058
1059         ver = readl(iommu->reg + DMAR_VER_REG);
1060         pr_info("%s: reg_base_addr %llx ver %d:%d cap %llx ecap %llx\n",
1061                 iommu->name,
1062                 (unsigned long long)drhd->reg_base_addr,
1063                 DMAR_VER_MAJOR(ver), DMAR_VER_MINOR(ver),
1064                 (unsigned long long)iommu->cap,
1065                 (unsigned long long)iommu->ecap);
1066
1067         /* Reflect status in gcmd */
1068         sts = readl(iommu->reg + DMAR_GSTS_REG);
1069         if (sts & DMA_GSTS_IRES)
1070                 iommu->gcmd |= DMA_GCMD_IRE;
1071         if (sts & DMA_GSTS_TES)
1072                 iommu->gcmd |= DMA_GCMD_TE;
1073         if (sts & DMA_GSTS_QIES)
1074                 iommu->gcmd |= DMA_GCMD_QIE;
1075
1076         raw_spin_lock_init(&iommu->register_lock);
1077
1078         if (intel_iommu_enabled) {
1079                 err = iommu_device_sysfs_add(&iommu->iommu, NULL,
1080                                              intel_iommu_groups,
1081                                              "%s", iommu->name);
1082                 if (err)
1083                         goto err_unmap;
1084
1085                 iommu_device_set_ops(&iommu->iommu, &intel_iommu_ops);
1086
1087                 err = iommu_device_register(&iommu->iommu);
1088                 if (err)
1089                         goto err_unmap;
1090         }
1091
1092         drhd->iommu = iommu;
1093
1094         return 0;
1095
1096 err_unmap:
1097         unmap_iommu(iommu);
1098 error_free_seq_id:
1099         dmar_free_seq_id(iommu);
1100 error:
1101         kfree(iommu);
1102         return err;
1103 }
1104
1105 static void free_iommu(struct intel_iommu *iommu)
1106 {
1107         if (intel_iommu_enabled) {
1108                 iommu_device_unregister(&iommu->iommu);
1109                 iommu_device_sysfs_remove(&iommu->iommu);
1110         }
1111
1112         if (iommu->irq) {
1113                 if (iommu->pr_irq) {
1114                         free_irq(iommu->pr_irq, iommu);
1115                         dmar_free_hwirq(iommu->pr_irq);
1116                         iommu->pr_irq = 0;
1117                 }
1118                 free_irq(iommu->irq, iommu);
1119                 dmar_free_hwirq(iommu->irq);
1120                 iommu->irq = 0;
1121         }
1122
1123         if (iommu->qi) {
1124                 free_page((unsigned long)iommu->qi->desc);
1125                 kfree(iommu->qi->desc_status);
1126                 kfree(iommu->qi);
1127         }
1128
1129         if (iommu->reg)
1130                 unmap_iommu(iommu);
1131
1132         dmar_free_seq_id(iommu);
1133         kfree(iommu);
1134 }
1135
1136 /*
1137  * Reclaim all the submitted descriptors which have completed its work.
1138  */
1139 static inline void reclaim_free_desc(struct q_inval *qi)
1140 {
1141         while (qi->desc_status[qi->free_tail] == QI_DONE ||
1142                qi->desc_status[qi->free_tail] == QI_ABORT) {
1143                 qi->desc_status[qi->free_tail] = QI_FREE;
1144                 qi->free_tail = (qi->free_tail + 1) % QI_LENGTH;
1145                 qi->free_cnt++;
1146         }
1147 }
1148
1149 static int qi_check_fault(struct intel_iommu *iommu, int index)
1150 {
1151         u32 fault;
1152         int head, tail;
1153         struct q_inval *qi = iommu->qi;
1154         int wait_index = (index + 1) % QI_LENGTH;
1155         int shift = qi_shift(iommu);
1156
1157         if (qi->desc_status[wait_index] == QI_ABORT)
1158                 return -EAGAIN;
1159
1160         fault = readl(iommu->reg + DMAR_FSTS_REG);
1161
1162         /*
1163          * If IQE happens, the head points to the descriptor associated
1164          * with the error. No new descriptors are fetched until the IQE
1165          * is cleared.
1166          */
1167         if (fault & DMA_FSTS_IQE) {
1168                 head = readl(iommu->reg + DMAR_IQH_REG);
1169                 if ((head >> shift) == index) {
1170                         struct qi_desc *desc = qi->desc + head;
1171
1172                         /*
1173                          * desc->qw2 and desc->qw3 are either reserved or
1174                          * used by software as private data. We won't print
1175                          * out these two qw's for security consideration.
1176                          */
1177                         pr_err("VT-d detected invalid descriptor: qw0 = %llx, qw1 = %llx\n",
1178                                (unsigned long long)desc->qw0,
1179                                (unsigned long long)desc->qw1);
1180                         memcpy(desc, qi->desc + (wait_index << shift),
1181                                1 << shift);
1182                         writel(DMA_FSTS_IQE, iommu->reg + DMAR_FSTS_REG);
1183                         return -EINVAL;
1184                 }
1185         }
1186
1187         /*
1188          * If ITE happens, all pending wait_desc commands are aborted.
1189          * No new descriptors are fetched until the ITE is cleared.
1190          */
1191         if (fault & DMA_FSTS_ITE) {
1192                 head = readl(iommu->reg + DMAR_IQH_REG);
1193                 head = ((head >> shift) - 1 + QI_LENGTH) % QI_LENGTH;
1194                 head |= 1;
1195                 tail = readl(iommu->reg + DMAR_IQT_REG);
1196                 tail = ((tail >> shift) - 1 + QI_LENGTH) % QI_LENGTH;
1197
1198                 writel(DMA_FSTS_ITE, iommu->reg + DMAR_FSTS_REG);
1199
1200                 do {
1201                         if (qi->desc_status[head] == QI_IN_USE)
1202                                 qi->desc_status[head] = QI_ABORT;
1203                         head = (head - 2 + QI_LENGTH) % QI_LENGTH;
1204                 } while (head != tail);
1205
1206                 if (qi->desc_status[wait_index] == QI_ABORT)
1207                         return -EAGAIN;
1208         }
1209
1210         if (fault & DMA_FSTS_ICE)
1211                 writel(DMA_FSTS_ICE, iommu->reg + DMAR_FSTS_REG);
1212
1213         return 0;
1214 }
1215
1216 /*
1217  * Submit the queued invalidation descriptor to the remapping
1218  * hardware unit and wait for its completion.
1219  */
1220 int qi_submit_sync(struct qi_desc *desc, struct intel_iommu *iommu)
1221 {
1222         int rc;
1223         struct q_inval *qi = iommu->qi;
1224         int offset, shift, length;
1225         struct qi_desc wait_desc;
1226         int wait_index, index;
1227         unsigned long flags;
1228
1229         if (!qi)
1230                 return 0;
1231
1232 restart:
1233         rc = 0;
1234
1235         raw_spin_lock_irqsave(&qi->q_lock, flags);
1236         while (qi->free_cnt < 3) {
1237                 raw_spin_unlock_irqrestore(&qi->q_lock, flags);
1238                 cpu_relax();
1239                 raw_spin_lock_irqsave(&qi->q_lock, flags);
1240         }
1241
1242         index = qi->free_head;
1243         wait_index = (index + 1) % QI_LENGTH;
1244         shift = qi_shift(iommu);
1245         length = 1 << shift;
1246
1247         qi->desc_status[index] = qi->desc_status[wait_index] = QI_IN_USE;
1248
1249         offset = index << shift;
1250         memcpy(qi->desc + offset, desc, length);
1251         wait_desc.qw0 = QI_IWD_STATUS_DATA(QI_DONE) |
1252                         QI_IWD_STATUS_WRITE | QI_IWD_TYPE;
1253         wait_desc.qw1 = virt_to_phys(&qi->desc_status[wait_index]);
1254         wait_desc.qw2 = 0;
1255         wait_desc.qw3 = 0;
1256
1257         offset = wait_index << shift;
1258         memcpy(qi->desc + offset, &wait_desc, length);
1259
1260         qi->free_head = (qi->free_head + 2) % QI_LENGTH;
1261         qi->free_cnt -= 2;
1262
1263         /*
1264          * update the HW tail register indicating the presence of
1265          * new descriptors.
1266          */
1267         writel(qi->free_head << shift, iommu->reg + DMAR_IQT_REG);
1268
1269         while (qi->desc_status[wait_index] != QI_DONE) {
1270                 /*
1271                  * We will leave the interrupts disabled, to prevent interrupt
1272                  * context to queue another cmd while a cmd is already submitted
1273                  * and waiting for completion on this cpu. This is to avoid
1274                  * a deadlock where the interrupt context can wait indefinitely
1275                  * for free slots in the queue.
1276                  */
1277                 rc = qi_check_fault(iommu, index);
1278                 if (rc)
1279                         break;
1280
1281                 raw_spin_unlock(&qi->q_lock);
1282                 cpu_relax();
1283                 raw_spin_lock(&qi->q_lock);
1284         }
1285
1286         qi->desc_status[index] = QI_DONE;
1287
1288         reclaim_free_desc(qi);
1289         raw_spin_unlock_irqrestore(&qi->q_lock, flags);
1290
1291         if (rc == -EAGAIN)
1292                 goto restart;
1293
1294         return rc;
1295 }
1296
1297 /*
1298  * Flush the global interrupt entry cache.
1299  */
1300 void qi_global_iec(struct intel_iommu *iommu)
1301 {
1302         struct qi_desc desc;
1303
1304         desc.qw0 = QI_IEC_TYPE;
1305         desc.qw1 = 0;
1306         desc.qw2 = 0;
1307         desc.qw3 = 0;
1308
1309         /* should never fail */
1310         qi_submit_sync(&desc, iommu);
1311 }
1312
1313 void qi_flush_context(struct intel_iommu *iommu, u16 did, u16 sid, u8 fm,
1314                       u64 type)
1315 {
1316         struct qi_desc desc;
1317
1318         desc.qw0 = QI_CC_FM(fm) | QI_CC_SID(sid) | QI_CC_DID(did)
1319                         | QI_CC_GRAN(type) | QI_CC_TYPE;
1320         desc.qw1 = 0;
1321         desc.qw2 = 0;
1322         desc.qw3 = 0;
1323
1324         qi_submit_sync(&desc, iommu);
1325 }
1326
1327 void qi_flush_iotlb(struct intel_iommu *iommu, u16 did, u64 addr,
1328                     unsigned int size_order, u64 type)
1329 {
1330         u8 dw = 0, dr = 0;
1331
1332         struct qi_desc desc;
1333         int ih = 0;
1334
1335         if (cap_write_drain(iommu->cap))
1336                 dw = 1;
1337
1338         if (cap_read_drain(iommu->cap))
1339                 dr = 1;
1340
1341         desc.qw0 = QI_IOTLB_DID(did) | QI_IOTLB_DR(dr) | QI_IOTLB_DW(dw)
1342                 | QI_IOTLB_GRAN(type) | QI_IOTLB_TYPE;
1343         desc.qw1 = QI_IOTLB_ADDR(addr) | QI_IOTLB_IH(ih)
1344                 | QI_IOTLB_AM(size_order);
1345         desc.qw2 = 0;
1346         desc.qw3 = 0;
1347
1348         qi_submit_sync(&desc, iommu);
1349 }
1350
1351 void qi_flush_dev_iotlb(struct intel_iommu *iommu, u16 sid, u16 pfsid,
1352                         u16 qdep, u64 addr, unsigned mask)
1353 {
1354         struct qi_desc desc;
1355
1356         if (mask) {
1357                 addr |= (1ULL << (VTD_PAGE_SHIFT + mask - 1)) - 1;
1358                 desc.qw1 = QI_DEV_IOTLB_ADDR(addr) | QI_DEV_IOTLB_SIZE;
1359         } else
1360                 desc.qw1 = QI_DEV_IOTLB_ADDR(addr);
1361
1362         if (qdep >= QI_DEV_IOTLB_MAX_INVS)
1363                 qdep = 0;
1364
1365         desc.qw0 = QI_DEV_IOTLB_SID(sid) | QI_DEV_IOTLB_QDEP(qdep) |
1366                    QI_DIOTLB_TYPE | QI_DEV_IOTLB_PFSID(pfsid);
1367         desc.qw2 = 0;
1368         desc.qw3 = 0;
1369
1370         qi_submit_sync(&desc, iommu);
1371 }
1372
1373 /* PASID-based IOTLB invalidation */
1374 void qi_flush_piotlb(struct intel_iommu *iommu, u16 did, u32 pasid, u64 addr,
1375                      unsigned long npages, bool ih)
1376 {
1377         struct qi_desc desc = {.qw2 = 0, .qw3 = 0};
1378
1379         /*
1380          * npages == -1 means a PASID-selective invalidation, otherwise,
1381          * a positive value for Page-selective-within-PASID invalidation.
1382          * 0 is not a valid input.
1383          */
1384         if (WARN_ON(!npages)) {
1385                 pr_err("Invalid input npages = %ld\n", npages);
1386                 return;
1387         }
1388
1389         if (npages == -1) {
1390                 desc.qw0 = QI_EIOTLB_PASID(pasid) |
1391                                 QI_EIOTLB_DID(did) |
1392                                 QI_EIOTLB_GRAN(QI_GRAN_NONG_PASID) |
1393                                 QI_EIOTLB_TYPE;
1394                 desc.qw1 = 0;
1395         } else {
1396                 int mask = ilog2(__roundup_pow_of_two(npages));
1397                 unsigned long align = (1ULL << (VTD_PAGE_SHIFT + mask));
1398
1399                 if (WARN_ON_ONCE(!ALIGN(addr, align)))
1400                         addr &= ~(align - 1);
1401
1402                 desc.qw0 = QI_EIOTLB_PASID(pasid) |
1403                                 QI_EIOTLB_DID(did) |
1404                                 QI_EIOTLB_GRAN(QI_GRAN_PSI_PASID) |
1405                                 QI_EIOTLB_TYPE;
1406                 desc.qw1 = QI_EIOTLB_ADDR(addr) |
1407                                 QI_EIOTLB_IH(ih) |
1408                                 QI_EIOTLB_AM(mask);
1409         }
1410
1411         qi_submit_sync(&desc, iommu);
1412 }
1413
1414 /*
1415  * Disable Queued Invalidation interface.
1416  */
1417 void dmar_disable_qi(struct intel_iommu *iommu)
1418 {
1419         unsigned long flags;
1420         u32 sts;
1421         cycles_t start_time = get_cycles();
1422
1423         if (!ecap_qis(iommu->ecap))
1424                 return;
1425
1426         raw_spin_lock_irqsave(&iommu->register_lock, flags);
1427
1428         sts =  readl(iommu->reg + DMAR_GSTS_REG);
1429         if (!(sts & DMA_GSTS_QIES))
1430                 goto end;
1431
1432         /*
1433          * Give a chance to HW to complete the pending invalidation requests.
1434          */
1435         while ((readl(iommu->reg + DMAR_IQT_REG) !=
1436                 readl(iommu->reg + DMAR_IQH_REG)) &&
1437                 (DMAR_OPERATION_TIMEOUT > (get_cycles() - start_time)))
1438                 cpu_relax();
1439
1440         iommu->gcmd &= ~DMA_GCMD_QIE;
1441         writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
1442
1443         IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, readl,
1444                       !(sts & DMA_GSTS_QIES), sts);
1445 end:
1446         raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
1447 }
1448
1449 /*
1450  * Enable queued invalidation.
1451  */
1452 static void __dmar_enable_qi(struct intel_iommu *iommu)
1453 {
1454         u32 sts;
1455         unsigned long flags;
1456         struct q_inval *qi = iommu->qi;
1457         u64 val = virt_to_phys(qi->desc);
1458
1459         qi->free_head = qi->free_tail = 0;
1460         qi->free_cnt = QI_LENGTH;
1461
1462         /*
1463          * Set DW=1 and QS=1 in IQA_REG when Scalable Mode capability
1464          * is present.
1465          */
1466         if (ecap_smts(iommu->ecap))
1467                 val |= (1 << 11) | 1;
1468
1469         raw_spin_lock_irqsave(&iommu->register_lock, flags);
1470
1471         /* write zero to the tail reg */
1472         writel(0, iommu->reg + DMAR_IQT_REG);
1473
1474         dmar_writeq(iommu->reg + DMAR_IQA_REG, val);
1475
1476         iommu->gcmd |= DMA_GCMD_QIE;
1477         writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG);
1478
1479         /* Make sure hardware complete it */
1480         IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, readl, (sts & DMA_GSTS_QIES), sts);
1481
1482         raw_spin_unlock_irqrestore(&iommu->register_lock, flags);
1483 }
1484
1485 /*
1486  * Enable Queued Invalidation interface. This is a must to support
1487  * interrupt-remapping. Also used by DMA-remapping, which replaces
1488  * register based IOTLB invalidation.
1489  */
1490 int dmar_enable_qi(struct intel_iommu *iommu)
1491 {
1492         struct q_inval *qi;
1493         struct page *desc_page;
1494
1495         if (!ecap_qis(iommu->ecap))
1496                 return -ENOENT;
1497
1498         /*
1499          * queued invalidation is already setup and enabled.
1500          */
1501         if (iommu->qi)
1502                 return 0;
1503
1504         iommu->qi = kmalloc(sizeof(*qi), GFP_ATOMIC);
1505         if (!iommu->qi)
1506                 return -ENOMEM;
1507
1508         qi = iommu->qi;
1509
1510         /*
1511          * Need two pages to accommodate 256 descriptors of 256 bits each
1512          * if the remapping hardware supports scalable mode translation.
1513          */
1514         desc_page = alloc_pages_node(iommu->node, GFP_ATOMIC | __GFP_ZERO,
1515                                      !!ecap_smts(iommu->ecap));
1516         if (!desc_page) {
1517                 kfree(qi);
1518                 iommu->qi = NULL;
1519                 return -ENOMEM;
1520         }
1521
1522         qi->desc = page_address(desc_page);
1523
1524         qi->desc_status = kcalloc(QI_LENGTH, sizeof(int), GFP_ATOMIC);
1525         if (!qi->desc_status) {
1526                 free_page((unsigned long) qi->desc);
1527                 kfree(qi);
1528                 iommu->qi = NULL;
1529                 return -ENOMEM;
1530         }
1531
1532         raw_spin_lock_init(&qi->q_lock);
1533
1534         __dmar_enable_qi(iommu);
1535
1536         return 0;
1537 }
1538
1539 /* iommu interrupt handling. Most stuff are MSI-like. */
1540
1541 enum faulttype {
1542         DMA_REMAP,
1543         INTR_REMAP,
1544         UNKNOWN,
1545 };
1546
1547 static const char *dma_remap_fault_reasons[] =
1548 {
1549         "Software",
1550         "Present bit in root entry is clear",
1551         "Present bit in context entry is clear",
1552         "Invalid context entry",
1553         "Access beyond MGAW",
1554         "PTE Write access is not set",
1555         "PTE Read access is not set",
1556         "Next page table ptr is invalid",
1557         "Root table address invalid",
1558         "Context table ptr is invalid",
1559         "non-zero reserved fields in RTP",
1560         "non-zero reserved fields in CTP",
1561         "non-zero reserved fields in PTE",
1562         "PCE for translation request specifies blocking",
1563 };
1564
1565 static const char * const dma_remap_sm_fault_reasons[] = {
1566         "SM: Invalid Root Table Address",
1567         "SM: TTM 0 for request with PASID",
1568         "SM: TTM 0 for page group request",
1569         "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x33-0x37 */
1570         "SM: Error attempting to access Root Entry",
1571         "SM: Present bit in Root Entry is clear",
1572         "SM: Non-zero reserved field set in Root Entry",
1573         "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x3B-0x3F */
1574         "SM: Error attempting to access Context Entry",
1575         "SM: Present bit in Context Entry is clear",
1576         "SM: Non-zero reserved field set in the Context Entry",
1577         "SM: Invalid Context Entry",
1578         "SM: DTE field in Context Entry is clear",
1579         "SM: PASID Enable field in Context Entry is clear",
1580         "SM: PASID is larger than the max in Context Entry",
1581         "SM: PRE field in Context-Entry is clear",
1582         "SM: RID_PASID field error in Context-Entry",
1583         "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x49-0x4F */
1584         "SM: Error attempting to access the PASID Directory Entry",
1585         "SM: Present bit in Directory Entry is clear",
1586         "SM: Non-zero reserved field set in PASID Directory Entry",
1587         "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x53-0x57 */
1588         "SM: Error attempting to access PASID Table Entry",
1589         "SM: Present bit in PASID Table Entry is clear",
1590         "SM: Non-zero reserved field set in PASID Table Entry",
1591         "SM: Invalid Scalable-Mode PASID Table Entry",
1592         "SM: ERE field is clear in PASID Table Entry",
1593         "SM: SRE field is clear in PASID Table Entry",
1594         "Unknown", "Unknown",/* 0x5E-0x5F */
1595         "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x60-0x67 */
1596         "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x68-0x6F */
1597         "SM: Error attempting to access first-level paging entry",
1598         "SM: Present bit in first-level paging entry is clear",
1599         "SM: Non-zero reserved field set in first-level paging entry",
1600         "SM: Error attempting to access FL-PML4 entry",
1601         "SM: First-level entry address beyond MGAW in Nested translation",
1602         "SM: Read permission error in FL-PML4 entry in Nested translation",
1603         "SM: Read permission error in first-level paging entry in Nested translation",
1604         "SM: Write permission error in first-level paging entry in Nested translation",
1605         "SM: Error attempting to access second-level paging entry",
1606         "SM: Read/Write permission error in second-level paging entry",
1607         "SM: Non-zero reserved field set in second-level paging entry",
1608         "SM: Invalid second-level page table pointer",
1609         "SM: A/D bit update needed in second-level entry when set up in no snoop",
1610         "Unknown", "Unknown", "Unknown", /* 0x7D-0x7F */
1611         "SM: Address in first-level translation is not canonical",
1612         "SM: U/S set 0 for first-level translation with user privilege",
1613         "SM: No execute permission for request with PASID and ER=1",
1614         "SM: Address beyond the DMA hardware max",
1615         "SM: Second-level entry address beyond the max",
1616         "SM: No write permission for Write/AtomicOp request",
1617         "SM: No read permission for Read/AtomicOp request",
1618         "SM: Invalid address-interrupt address",
1619         "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x88-0x8F */
1620         "SM: A/D bit update needed in first-level entry when set up in no snoop",
1621 };
1622
1623 static const char *irq_remap_fault_reasons[] =
1624 {
1625         "Detected reserved fields in the decoded interrupt-remapped request",
1626         "Interrupt index exceeded the interrupt-remapping table size",
1627         "Present field in the IRTE entry is clear",
1628         "Error accessing interrupt-remapping table pointed by IRTA_REG",
1629         "Detected reserved fields in the IRTE entry",
1630         "Blocked a compatibility format interrupt request",
1631         "Blocked an interrupt request due to source-id verification failure",
1632 };
1633
1634 static const char *dmar_get_fault_reason(u8 fault_reason, int *fault_type)
1635 {
1636         if (fault_reason >= 0x20 && (fault_reason - 0x20 <
1637                                         ARRAY_SIZE(irq_remap_fault_reasons))) {
1638                 *fault_type = INTR_REMAP;
1639                 return irq_remap_fault_reasons[fault_reason - 0x20];
1640         } else if (fault_reason >= 0x30 && (fault_reason - 0x30 <
1641                         ARRAY_SIZE(dma_remap_sm_fault_reasons))) {
1642                 *fault_type = DMA_REMAP;
1643                 return dma_remap_sm_fault_reasons[fault_reason - 0x30];
1644         } else if (fault_reason < ARRAY_SIZE(dma_remap_fault_reasons)) {
1645                 *fault_type = DMA_REMAP;
1646                 return dma_remap_fault_reasons[fault_reason];
1647         } else {
1648                 *fault_type = UNKNOWN;
1649                 return "Unknown";
1650         }
1651 }
1652
1653
1654 static inline int dmar_msi_reg(struct intel_iommu *iommu, int irq)
1655 {
1656         if (iommu->irq == irq)
1657                 return DMAR_FECTL_REG;
1658         else if (iommu->pr_irq == irq)
1659                 return DMAR_PECTL_REG;
1660         else
1661                 BUG();
1662 }
1663
1664 void dmar_msi_unmask(struct irq_data *data)
1665 {
1666         struct intel_iommu *iommu = irq_data_get_irq_handler_data(data);
1667         int reg = dmar_msi_reg(iommu, data->irq);
1668         unsigned long flag;
1669
1670         /* unmask it */
1671         raw_spin_lock_irqsave(&iommu->register_lock, flag);
1672         writel(0, iommu->reg + reg);
1673         /* Read a reg to force flush the post write */
1674         readl(iommu->reg + reg);
1675         raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1676 }
1677
1678 void dmar_msi_mask(struct irq_data *data)
1679 {
1680         struct intel_iommu *iommu = irq_data_get_irq_handler_data(data);
1681         int reg = dmar_msi_reg(iommu, data->irq);
1682         unsigned long flag;
1683
1684         /* mask it */
1685         raw_spin_lock_irqsave(&iommu->register_lock, flag);
1686         writel(DMA_FECTL_IM, iommu->reg + reg);
1687         /* Read a reg to force flush the post write */
1688         readl(iommu->reg + reg);
1689         raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1690 }
1691
1692 void dmar_msi_write(int irq, struct msi_msg *msg)
1693 {
1694         struct intel_iommu *iommu = irq_get_handler_data(irq);
1695         int reg = dmar_msi_reg(iommu, irq);
1696         unsigned long flag;
1697
1698         raw_spin_lock_irqsave(&iommu->register_lock, flag);
1699         writel(msg->data, iommu->reg + reg + 4);
1700         writel(msg->address_lo, iommu->reg + reg + 8);
1701         writel(msg->address_hi, iommu->reg + reg + 12);
1702         raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1703 }
1704
1705 void dmar_msi_read(int irq, struct msi_msg *msg)
1706 {
1707         struct intel_iommu *iommu = irq_get_handler_data(irq);
1708         int reg = dmar_msi_reg(iommu, irq);
1709         unsigned long flag;
1710
1711         raw_spin_lock_irqsave(&iommu->register_lock, flag);
1712         msg->data = readl(iommu->reg + reg + 4);
1713         msg->address_lo = readl(iommu->reg + reg + 8);
1714         msg->address_hi = readl(iommu->reg + reg + 12);
1715         raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1716 }
1717
1718 static int dmar_fault_do_one(struct intel_iommu *iommu, int type,
1719                 u8 fault_reason, int pasid, u16 source_id,
1720                 unsigned long long addr)
1721 {
1722         const char *reason;
1723         int fault_type;
1724
1725         reason = dmar_get_fault_reason(fault_reason, &fault_type);
1726
1727         if (fault_type == INTR_REMAP)
1728                 pr_err("[INTR-REMAP] Request device [%02x:%02x.%d] fault index %llx [fault reason %02d] %s\n",
1729                         source_id >> 8, PCI_SLOT(source_id & 0xFF),
1730                         PCI_FUNC(source_id & 0xFF), addr >> 48,
1731                         fault_reason, reason);
1732         else
1733                 pr_err("[%s] Request device [%02x:%02x.%d] PASID %x fault addr %llx [fault reason %02d] %s\n",
1734                        type ? "DMA Read" : "DMA Write",
1735                        source_id >> 8, PCI_SLOT(source_id & 0xFF),
1736                        PCI_FUNC(source_id & 0xFF), pasid, addr,
1737                        fault_reason, reason);
1738         return 0;
1739 }
1740
1741 #define PRIMARY_FAULT_REG_LEN (16)
1742 irqreturn_t dmar_fault(int irq, void *dev_id)
1743 {
1744         struct intel_iommu *iommu = dev_id;
1745         int reg, fault_index;
1746         u32 fault_status;
1747         unsigned long flag;
1748         static DEFINE_RATELIMIT_STATE(rs,
1749                                       DEFAULT_RATELIMIT_INTERVAL,
1750                                       DEFAULT_RATELIMIT_BURST);
1751
1752         raw_spin_lock_irqsave(&iommu->register_lock, flag);
1753         fault_status = readl(iommu->reg + DMAR_FSTS_REG);
1754         if (fault_status && __ratelimit(&rs))
1755                 pr_err("DRHD: handling fault status reg %x\n", fault_status);
1756
1757         /* TBD: ignore advanced fault log currently */
1758         if (!(fault_status & DMA_FSTS_PPF))
1759                 goto unlock_exit;
1760
1761         fault_index = dma_fsts_fault_record_index(fault_status);
1762         reg = cap_fault_reg_offset(iommu->cap);
1763         while (1) {
1764                 /* Disable printing, simply clear the fault when ratelimited */
1765                 bool ratelimited = !__ratelimit(&rs);
1766                 u8 fault_reason;
1767                 u16 source_id;
1768                 u64 guest_addr;
1769                 int type, pasid;
1770                 u32 data;
1771                 bool pasid_present;
1772
1773                 /* highest 32 bits */
1774                 data = readl(iommu->reg + reg +
1775                                 fault_index * PRIMARY_FAULT_REG_LEN + 12);
1776                 if (!(data & DMA_FRCD_F))
1777                         break;
1778
1779                 if (!ratelimited) {
1780                         fault_reason = dma_frcd_fault_reason(data);
1781                         type = dma_frcd_type(data);
1782
1783                         pasid = dma_frcd_pasid_value(data);
1784                         data = readl(iommu->reg + reg +
1785                                      fault_index * PRIMARY_FAULT_REG_LEN + 8);
1786                         source_id = dma_frcd_source_id(data);
1787
1788                         pasid_present = dma_frcd_pasid_present(data);
1789                         guest_addr = dmar_readq(iommu->reg + reg +
1790                                         fault_index * PRIMARY_FAULT_REG_LEN);
1791                         guest_addr = dma_frcd_page_addr(guest_addr);
1792                 }
1793
1794                 /* clear the fault */
1795                 writel(DMA_FRCD_F, iommu->reg + reg +
1796                         fault_index * PRIMARY_FAULT_REG_LEN + 12);
1797
1798                 raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1799
1800                 if (!ratelimited)
1801                         /* Using pasid -1 if pasid is not present */
1802                         dmar_fault_do_one(iommu, type, fault_reason,
1803                                           pasid_present ? pasid : -1,
1804                                           source_id, guest_addr);
1805
1806                 fault_index++;
1807                 if (fault_index >= cap_num_fault_regs(iommu->cap))
1808                         fault_index = 0;
1809                 raw_spin_lock_irqsave(&iommu->register_lock, flag);
1810         }
1811
1812         writel(DMA_FSTS_PFO | DMA_FSTS_PPF | DMA_FSTS_PRO,
1813                iommu->reg + DMAR_FSTS_REG);
1814
1815 unlock_exit:
1816         raw_spin_unlock_irqrestore(&iommu->register_lock, flag);
1817         return IRQ_HANDLED;
1818 }
1819
1820 int dmar_set_interrupt(struct intel_iommu *iommu)
1821 {
1822         int irq, ret;
1823
1824         /*
1825          * Check if the fault interrupt is already initialized.
1826          */
1827         if (iommu->irq)
1828                 return 0;
1829
1830         irq = dmar_alloc_hwirq(iommu->seq_id, iommu->node, iommu);
1831         if (irq > 0) {
1832                 iommu->irq = irq;
1833         } else {
1834                 pr_err("No free IRQ vectors\n");
1835                 return -EINVAL;
1836         }
1837
1838         ret = request_irq(irq, dmar_fault, IRQF_NO_THREAD, iommu->name, iommu);
1839         if (ret)
1840                 pr_err("Can't request irq\n");
1841         return ret;
1842 }
1843
1844 int __init enable_drhd_fault_handling(void)
1845 {
1846         struct dmar_drhd_unit *drhd;
1847         struct intel_iommu *iommu;
1848
1849         /*
1850          * Enable fault control interrupt.
1851          */
1852         for_each_iommu(iommu, drhd) {
1853                 u32 fault_status;
1854                 int ret = dmar_set_interrupt(iommu);
1855
1856                 if (ret) {
1857                         pr_err("DRHD %Lx: failed to enable fault, interrupt, ret %d\n",
1858                                (unsigned long long)drhd->reg_base_addr, ret);
1859                         return -1;
1860                 }
1861
1862                 /*
1863                  * Clear any previous faults.
1864                  */
1865                 dmar_fault(iommu->irq, iommu);
1866                 fault_status = readl(iommu->reg + DMAR_FSTS_REG);
1867                 writel(fault_status, iommu->reg + DMAR_FSTS_REG);
1868         }
1869
1870         return 0;
1871 }
1872
1873 /*
1874  * Re-enable Queued Invalidation interface.
1875  */
1876 int dmar_reenable_qi(struct intel_iommu *iommu)
1877 {
1878         if (!ecap_qis(iommu->ecap))
1879                 return -ENOENT;
1880
1881         if (!iommu->qi)
1882                 return -ENOENT;
1883
1884         /*
1885          * First disable queued invalidation.
1886          */
1887         dmar_disable_qi(iommu);
1888         /*
1889          * Then enable queued invalidation again. Since there is no pending
1890          * invalidation requests now, it's safe to re-enable queued
1891          * invalidation.
1892          */
1893         __dmar_enable_qi(iommu);
1894
1895         return 0;
1896 }
1897
1898 /*
1899  * Check interrupt remapping support in DMAR table description.
1900  */
1901 int __init dmar_ir_support(void)
1902 {
1903         struct acpi_table_dmar *dmar;
1904         dmar = (struct acpi_table_dmar *)dmar_tbl;
1905         if (!dmar)
1906                 return 0;
1907         return dmar->flags & 0x1;
1908 }
1909
1910 /* Check whether DMAR units are in use */
1911 static inline bool dmar_in_use(void)
1912 {
1913         return irq_remapping_enabled || intel_iommu_enabled;
1914 }
1915
1916 static int __init dmar_free_unused_resources(void)
1917 {
1918         struct dmar_drhd_unit *dmaru, *dmaru_n;
1919
1920         if (dmar_in_use())
1921                 return 0;
1922
1923         if (dmar_dev_scope_status != 1 && !list_empty(&dmar_drhd_units))
1924                 bus_unregister_notifier(&pci_bus_type, &dmar_pci_bus_nb);
1925
1926         down_write(&dmar_global_lock);
1927         list_for_each_entry_safe(dmaru, dmaru_n, &dmar_drhd_units, list) {
1928                 list_del(&dmaru->list);
1929                 dmar_free_drhd(dmaru);
1930         }
1931         up_write(&dmar_global_lock);
1932
1933         return 0;
1934 }
1935
1936 late_initcall(dmar_free_unused_resources);
1937 IOMMU_INIT_POST(detect_intel_iommu);
1938
1939 /*
1940  * DMAR Hotplug Support
1941  * For more details, please refer to Intel(R) Virtualization Technology
1942  * for Directed-IO Architecture Specifiction, Rev 2.2, Section 8.8
1943  * "Remapping Hardware Unit Hot Plug".
1944  */
1945 static guid_t dmar_hp_guid =
1946         GUID_INIT(0xD8C1A3A6, 0xBE9B, 0x4C9B,
1947                   0x91, 0xBF, 0xC3, 0xCB, 0x81, 0xFC, 0x5D, 0xAF);
1948
1949 /*
1950  * Currently there's only one revision and BIOS will not check the revision id,
1951  * so use 0 for safety.
1952  */
1953 #define DMAR_DSM_REV_ID                 0
1954 #define DMAR_DSM_FUNC_DRHD              1
1955 #define DMAR_DSM_FUNC_ATSR              2
1956 #define DMAR_DSM_FUNC_RHSA              3
1957
1958 static inline bool dmar_detect_dsm(acpi_handle handle, int func)
1959 {
1960         return acpi_check_dsm(handle, &dmar_hp_guid, DMAR_DSM_REV_ID, 1 << func);
1961 }
1962
1963 static int dmar_walk_dsm_resource(acpi_handle handle, int func,
1964                                   dmar_res_handler_t handler, void *arg)
1965 {
1966         int ret = -ENODEV;
1967         union acpi_object *obj;
1968         struct acpi_dmar_header *start;
1969         struct dmar_res_callback callback;
1970         static int res_type[] = {
1971                 [DMAR_DSM_FUNC_DRHD] = ACPI_DMAR_TYPE_HARDWARE_UNIT,
1972                 [DMAR_DSM_FUNC_ATSR] = ACPI_DMAR_TYPE_ROOT_ATS,
1973                 [DMAR_DSM_FUNC_RHSA] = ACPI_DMAR_TYPE_HARDWARE_AFFINITY,
1974         };
1975
1976         if (!dmar_detect_dsm(handle, func))
1977                 return 0;
1978
1979         obj = acpi_evaluate_dsm_typed(handle, &dmar_hp_guid, DMAR_DSM_REV_ID,
1980                                       func, NULL, ACPI_TYPE_BUFFER);
1981         if (!obj)
1982                 return -ENODEV;
1983
1984         memset(&callback, 0, sizeof(callback));
1985         callback.cb[res_type[func]] = handler;
1986         callback.arg[res_type[func]] = arg;
1987         start = (struct acpi_dmar_header *)obj->buffer.pointer;
1988         ret = dmar_walk_remapping_entries(start, obj->buffer.length, &callback);
1989
1990         ACPI_FREE(obj);
1991
1992         return ret;
1993 }
1994
1995 static int dmar_hp_add_drhd(struct acpi_dmar_header *header, void *arg)
1996 {
1997         int ret;
1998         struct dmar_drhd_unit *dmaru;
1999
2000         dmaru = dmar_find_dmaru((struct acpi_dmar_hardware_unit *)header);
2001         if (!dmaru)
2002                 return -ENODEV;
2003
2004         ret = dmar_ir_hotplug(dmaru, true);
2005         if (ret == 0)
2006                 ret = dmar_iommu_hotplug(dmaru, true);
2007
2008         return ret;
2009 }
2010
2011 static int dmar_hp_remove_drhd(struct acpi_dmar_header *header, void *arg)
2012 {
2013         int i, ret;
2014         struct device *dev;
2015         struct dmar_drhd_unit *dmaru;
2016
2017         dmaru = dmar_find_dmaru((struct acpi_dmar_hardware_unit *)header);
2018         if (!dmaru)
2019                 return 0;
2020
2021         /*
2022          * All PCI devices managed by this unit should have been destroyed.
2023          */
2024         if (!dmaru->include_all && dmaru->devices && dmaru->devices_cnt) {
2025                 for_each_active_dev_scope(dmaru->devices,
2026                                           dmaru->devices_cnt, i, dev)
2027                         return -EBUSY;
2028         }
2029
2030         ret = dmar_ir_hotplug(dmaru, false);
2031         if (ret == 0)
2032                 ret = dmar_iommu_hotplug(dmaru, false);
2033
2034         return ret;
2035 }
2036
2037 static int dmar_hp_release_drhd(struct acpi_dmar_header *header, void *arg)
2038 {
2039         struct dmar_drhd_unit *dmaru;
2040
2041         dmaru = dmar_find_dmaru((struct acpi_dmar_hardware_unit *)header);
2042         if (dmaru) {
2043                 list_del_rcu(&dmaru->list);
2044                 synchronize_rcu();
2045                 dmar_free_drhd(dmaru);
2046         }
2047
2048         return 0;
2049 }
2050
2051 static int dmar_hotplug_insert(acpi_handle handle)
2052 {
2053         int ret;
2054         int drhd_count = 0;
2055
2056         ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
2057                                      &dmar_validate_one_drhd, (void *)1);
2058         if (ret)
2059                 goto out;
2060
2061         ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
2062                                      &dmar_parse_one_drhd, (void *)&drhd_count);
2063         if (ret == 0 && drhd_count == 0) {
2064                 pr_warn(FW_BUG "No DRHD structures in buffer returned by _DSM method\n");
2065                 goto out;
2066         } else if (ret) {
2067                 goto release_drhd;
2068         }
2069
2070         ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_RHSA,
2071                                      &dmar_parse_one_rhsa, NULL);
2072         if (ret)
2073                 goto release_drhd;
2074
2075         ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
2076                                      &dmar_parse_one_atsr, NULL);
2077         if (ret)
2078                 goto release_atsr;
2079
2080         ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
2081                                      &dmar_hp_add_drhd, NULL);
2082         if (!ret)
2083                 return 0;
2084
2085         dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
2086                                &dmar_hp_remove_drhd, NULL);
2087 release_atsr:
2088         dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
2089                                &dmar_release_one_atsr, NULL);
2090 release_drhd:
2091         dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
2092                                &dmar_hp_release_drhd, NULL);
2093 out:
2094         return ret;
2095 }
2096
2097 static int dmar_hotplug_remove(acpi_handle handle)
2098 {
2099         int ret;
2100
2101         ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
2102                                      &dmar_check_one_atsr, NULL);
2103         if (ret)
2104                 return ret;
2105
2106         ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
2107                                      &dmar_hp_remove_drhd, NULL);
2108         if (ret == 0) {
2109                 WARN_ON(dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR,
2110                                                &dmar_release_one_atsr, NULL));
2111                 WARN_ON(dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
2112                                                &dmar_hp_release_drhd, NULL));
2113         } else {
2114                 dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD,
2115                                        &dmar_hp_add_drhd, NULL);
2116         }
2117
2118         return ret;
2119 }
2120
2121 static acpi_status dmar_get_dsm_handle(acpi_handle handle, u32 lvl,
2122                                        void *context, void **retval)
2123 {
2124         acpi_handle *phdl = retval;
2125
2126         if (dmar_detect_dsm(handle, DMAR_DSM_FUNC_DRHD)) {
2127                 *phdl = handle;
2128                 return AE_CTRL_TERMINATE;
2129         }
2130
2131         return AE_OK;
2132 }
2133
2134 static int dmar_device_hotplug(acpi_handle handle, bool insert)
2135 {
2136         int ret;
2137         acpi_handle tmp = NULL;
2138         acpi_status status;
2139
2140         if (!dmar_in_use())
2141                 return 0;
2142
2143         if (dmar_detect_dsm(handle, DMAR_DSM_FUNC_DRHD)) {
2144                 tmp = handle;
2145         } else {
2146                 status = acpi_walk_namespace(ACPI_TYPE_DEVICE, handle,
2147                                              ACPI_UINT32_MAX,
2148                                              dmar_get_dsm_handle,
2149                                              NULL, NULL, &tmp);
2150                 if (ACPI_FAILURE(status)) {
2151                         pr_warn("Failed to locate _DSM method.\n");
2152                         return -ENXIO;
2153                 }
2154         }
2155         if (tmp == NULL)
2156                 return 0;
2157
2158         down_write(&dmar_global_lock);
2159         if (insert)
2160                 ret = dmar_hotplug_insert(tmp);
2161         else
2162                 ret = dmar_hotplug_remove(tmp);
2163         up_write(&dmar_global_lock);
2164
2165         return ret;
2166 }
2167
2168 int dmar_device_add(acpi_handle handle)
2169 {
2170         return dmar_device_hotplug(handle, true);
2171 }
2172
2173 int dmar_device_remove(acpi_handle handle)
2174 {
2175         return dmar_device_hotplug(handle, false);
2176 }
2177
2178 /*
2179  * dmar_platform_optin - Is %DMA_CTRL_PLATFORM_OPT_IN_FLAG set in DMAR table
2180  *
2181  * Returns true if the platform has %DMA_CTRL_PLATFORM_OPT_IN_FLAG set in
2182  * the ACPI DMAR table. This means that the platform boot firmware has made
2183  * sure no device can issue DMA outside of RMRR regions.
2184  */
2185 bool dmar_platform_optin(void)
2186 {
2187         struct acpi_table_dmar *dmar;
2188         acpi_status status;
2189         bool ret;
2190
2191         status = acpi_get_table(ACPI_SIG_DMAR, 0,
2192                                 (struct acpi_table_header **)&dmar);
2193         if (ACPI_FAILURE(status))
2194                 return false;
2195
2196         ret = !!(dmar->flags & DMAR_PLATFORM_OPT_IN);
2197         acpi_put_table((struct acpi_table_header *)dmar);
2198
2199         return ret;
2200 }
2201 EXPORT_SYMBOL_GPL(dmar_platform_optin);