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 <linux/limits.h> 32 #include <asm/irq_remapping.h> 33 #include <asm/iommu_table.h> 34 #include <trace/events/intel_iommu.h> 35 36 #include "../irq_remapping.h" 37 38 typedef int (*dmar_res_handler_t)(struct acpi_dmar_header *, void *); 39 struct dmar_res_callback { 40 dmar_res_handler_t cb[ACPI_DMAR_TYPE_RESERVED]; 41 void *arg[ACPI_DMAR_TYPE_RESERVED]; 42 bool ignore_unhandled; 43 bool print_entry; 44 }; 45 46 /* 47 * Assumptions: 48 * 1) The hotplug framework guarentees that DMAR unit will be hot-added 49 * before IO devices managed by that unit. 50 * 2) The hotplug framework guarantees that DMAR unit will be hot-removed 51 * after IO devices managed by that unit. 52 * 3) Hotplug events are rare. 53 * 54 * Locking rules for DMA and interrupt remapping related global data structures: 55 * 1) Use dmar_global_lock in process context 56 * 2) Use RCU in interrupt context 57 */ 58 DECLARE_RWSEM(dmar_global_lock); 59 LIST_HEAD(dmar_drhd_units); 60 61 struct acpi_table_header * __initdata dmar_tbl; 62 static int dmar_dev_scope_status = 1; 63 static unsigned long dmar_seq_ids[BITS_TO_LONGS(DMAR_UNITS_SUPPORTED)]; 64 65 static int alloc_iommu(struct dmar_drhd_unit *drhd); 66 static void free_iommu(struct intel_iommu *iommu); 67 68 extern const struct iommu_ops intel_iommu_ops; 69 70 static void dmar_register_drhd_unit(struct dmar_drhd_unit *drhd) 71 { 72 /* 73 * add INCLUDE_ALL at the tail, so scan the list will find it at 74 * the very end. 75 */ 76 if (drhd->include_all) 77 list_add_tail_rcu(&drhd->list, &dmar_drhd_units); 78 else 79 list_add_rcu(&drhd->list, &dmar_drhd_units); 80 } 81 82 void *dmar_alloc_dev_scope(void *start, void *end, int *cnt) 83 { 84 struct acpi_dmar_device_scope *scope; 85 86 *cnt = 0; 87 while (start < end) { 88 scope = start; 89 if (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_NAMESPACE || 90 scope->entry_type == ACPI_DMAR_SCOPE_TYPE_ENDPOINT || 91 scope->entry_type == ACPI_DMAR_SCOPE_TYPE_BRIDGE) 92 (*cnt)++; 93 else if (scope->entry_type != ACPI_DMAR_SCOPE_TYPE_IOAPIC && 94 scope->entry_type != ACPI_DMAR_SCOPE_TYPE_HPET) { 95 pr_warn("Unsupported device scope\n"); 96 } 97 start += scope->length; 98 } 99 if (*cnt == 0) 100 return NULL; 101 102 return kcalloc(*cnt, sizeof(struct dmar_dev_scope), GFP_KERNEL); 103 } 104 105 void dmar_free_dev_scope(struct dmar_dev_scope **devices, int *cnt) 106 { 107 int i; 108 struct device *tmp_dev; 109 110 if (*devices && *cnt) { 111 for_each_active_dev_scope(*devices, *cnt, i, tmp_dev) 112 put_device(tmp_dev); 113 kfree(*devices); 114 } 115 116 *devices = NULL; 117 *cnt = 0; 118 } 119 120 /* Optimize out kzalloc()/kfree() for normal cases */ 121 static char dmar_pci_notify_info_buf[64]; 122 123 static struct dmar_pci_notify_info * 124 dmar_alloc_pci_notify_info(struct pci_dev *dev, unsigned long event) 125 { 126 int level = 0; 127 size_t size; 128 struct pci_dev *tmp; 129 struct dmar_pci_notify_info *info; 130 131 BUG_ON(dev->is_virtfn); 132 133 /* 134 * Ignore devices that have a domain number higher than what can 135 * be looked up in DMAR, e.g. VMD subdevices with domain 0x10000 136 */ 137 if (pci_domain_nr(dev->bus) > U16_MAX) 138 return NULL; 139 140 /* Only generate path[] for device addition event */ 141 if (event == BUS_NOTIFY_ADD_DEVICE) 142 for (tmp = dev; tmp; tmp = tmp->bus->self) 143 level++; 144 145 size = struct_size(info, path, level); 146 if (size <= sizeof(dmar_pci_notify_info_buf)) { 147 info = (struct dmar_pci_notify_info *)dmar_pci_notify_info_buf; 148 } else { 149 info = kzalloc(size, GFP_KERNEL); 150 if (!info) { 151 pr_warn("Out of memory when allocating notify_info " 152 "for %s.\n", pci_name(dev)); 153 if (dmar_dev_scope_status == 0) 154 dmar_dev_scope_status = -ENOMEM; 155 return NULL; 156 } 157 } 158 159 info->event = event; 160 info->dev = dev; 161 info->seg = pci_domain_nr(dev->bus); 162 info->level = level; 163 if (event == BUS_NOTIFY_ADD_DEVICE) { 164 for (tmp = dev; tmp; tmp = tmp->bus->self) { 165 level--; 166 info->path[level].bus = tmp->bus->number; 167 info->path[level].device = PCI_SLOT(tmp->devfn); 168 info->path[level].function = PCI_FUNC(tmp->devfn); 169 if (pci_is_root_bus(tmp->bus)) 170 info->bus = tmp->bus->number; 171 } 172 } 173 174 return info; 175 } 176 177 static inline void dmar_free_pci_notify_info(struct dmar_pci_notify_info *info) 178 { 179 if ((void *)info != dmar_pci_notify_info_buf) 180 kfree(info); 181 } 182 183 static bool dmar_match_pci_path(struct dmar_pci_notify_info *info, int bus, 184 struct acpi_dmar_pci_path *path, int count) 185 { 186 int i; 187 188 if (info->bus != bus) 189 goto fallback; 190 if (info->level != count) 191 goto fallback; 192 193 for (i = 0; i < count; i++) { 194 if (path[i].device != info->path[i].device || 195 path[i].function != info->path[i].function) 196 goto fallback; 197 } 198 199 return true; 200 201 fallback: 202 203 if (count != 1) 204 return false; 205 206 i = info->level - 1; 207 if (bus == info->path[i].bus && 208 path[0].device == info->path[i].device && 209 path[0].function == info->path[i].function) { 210 pr_info(FW_BUG "RMRR entry for device %02x:%02x.%x is broken - applying workaround\n", 211 bus, path[0].device, path[0].function); 212 return true; 213 } 214 215 return false; 216 } 217 218 /* Return: > 0 if match found, 0 if no match found, < 0 if error happens */ 219 int dmar_insert_dev_scope(struct dmar_pci_notify_info *info, 220 void *start, void*end, u16 segment, 221 struct dmar_dev_scope *devices, 222 int devices_cnt) 223 { 224 int i, level; 225 struct device *tmp, *dev = &info->dev->dev; 226 struct acpi_dmar_device_scope *scope; 227 struct acpi_dmar_pci_path *path; 228 229 if (segment != info->seg) 230 return 0; 231 232 for (; start < end; start += scope->length) { 233 scope = start; 234 if (scope->entry_type != ACPI_DMAR_SCOPE_TYPE_ENDPOINT && 235 scope->entry_type != ACPI_DMAR_SCOPE_TYPE_BRIDGE) 236 continue; 237 238 path = (struct acpi_dmar_pci_path *)(scope + 1); 239 level = (scope->length - sizeof(*scope)) / sizeof(*path); 240 if (!dmar_match_pci_path(info, scope->bus, path, level)) 241 continue; 242 243 /* 244 * We expect devices with endpoint scope to have normal PCI 245 * headers, and devices with bridge scope to have bridge PCI 246 * headers. However PCI NTB devices may be listed in the 247 * DMAR table with bridge scope, even though they have a 248 * normal PCI header. NTB devices are identified by class 249 * "BRIDGE_OTHER" (0680h) - we don't declare a socpe mismatch 250 * for this special case. 251 */ 252 if ((scope->entry_type == ACPI_DMAR_SCOPE_TYPE_ENDPOINT && 253 info->dev->hdr_type != PCI_HEADER_TYPE_NORMAL) || 254 (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_BRIDGE && 255 (info->dev->hdr_type == PCI_HEADER_TYPE_NORMAL && 256 info->dev->class >> 16 != PCI_BASE_CLASS_BRIDGE))) { 257 pr_warn("Device scope type does not match for %s\n", 258 pci_name(info->dev)); 259 return -EINVAL; 260 } 261 262 for_each_dev_scope(devices, devices_cnt, i, tmp) 263 if (tmp == NULL) { 264 devices[i].bus = info->dev->bus->number; 265 devices[i].devfn = info->dev->devfn; 266 rcu_assign_pointer(devices[i].dev, 267 get_device(dev)); 268 return 1; 269 } 270 BUG_ON(i >= devices_cnt); 271 } 272 273 return 0; 274 } 275 276 int dmar_remove_dev_scope(struct dmar_pci_notify_info *info, u16 segment, 277 struct dmar_dev_scope *devices, int count) 278 { 279 int index; 280 struct device *tmp; 281 282 if (info->seg != segment) 283 return 0; 284 285 for_each_active_dev_scope(devices, count, index, tmp) 286 if (tmp == &info->dev->dev) { 287 RCU_INIT_POINTER(devices[index].dev, NULL); 288 synchronize_rcu(); 289 put_device(tmp); 290 return 1; 291 } 292 293 return 0; 294 } 295 296 static int dmar_pci_bus_add_dev(struct dmar_pci_notify_info *info) 297 { 298 int ret = 0; 299 struct dmar_drhd_unit *dmaru; 300 struct acpi_dmar_hardware_unit *drhd; 301 302 for_each_drhd_unit(dmaru) { 303 if (dmaru->include_all) 304 continue; 305 306 drhd = container_of(dmaru->hdr, 307 struct acpi_dmar_hardware_unit, header); 308 ret = dmar_insert_dev_scope(info, (void *)(drhd + 1), 309 ((void *)drhd) + drhd->header.length, 310 dmaru->segment, 311 dmaru->devices, dmaru->devices_cnt); 312 if (ret) 313 break; 314 } 315 if (ret >= 0) 316 ret = dmar_iommu_notify_scope_dev(info); 317 if (ret < 0 && dmar_dev_scope_status == 0) 318 dmar_dev_scope_status = ret; 319 320 if (ret >= 0) 321 intel_irq_remap_add_device(info); 322 323 return ret; 324 } 325 326 static void dmar_pci_bus_del_dev(struct dmar_pci_notify_info *info) 327 { 328 struct dmar_drhd_unit *dmaru; 329 330 for_each_drhd_unit(dmaru) 331 if (dmar_remove_dev_scope(info, dmaru->segment, 332 dmaru->devices, dmaru->devices_cnt)) 333 break; 334 dmar_iommu_notify_scope_dev(info); 335 } 336 337 static inline void vf_inherit_msi_domain(struct pci_dev *pdev) 338 { 339 struct pci_dev *physfn = pci_physfn(pdev); 340 341 dev_set_msi_domain(&pdev->dev, dev_get_msi_domain(&physfn->dev)); 342 } 343 344 static int dmar_pci_bus_notifier(struct notifier_block *nb, 345 unsigned long action, void *data) 346 { 347 struct pci_dev *pdev = to_pci_dev(data); 348 struct dmar_pci_notify_info *info; 349 350 /* Only care about add/remove events for physical functions. 351 * For VFs we actually do the lookup based on the corresponding 352 * PF in device_to_iommu() anyway. */ 353 if (pdev->is_virtfn) { 354 /* 355 * Ensure that the VF device inherits the irq domain of the 356 * PF device. Ideally the device would inherit the domain 357 * from the bus, but DMAR can have multiple units per bus 358 * which makes this impossible. The VF 'bus' could inherit 359 * from the PF device, but that's yet another x86'sism to 360 * inflict on everybody else. 361 */ 362 if (action == BUS_NOTIFY_ADD_DEVICE) 363 vf_inherit_msi_domain(pdev); 364 return NOTIFY_DONE; 365 } 366 367 if (action != BUS_NOTIFY_ADD_DEVICE && 368 action != BUS_NOTIFY_REMOVED_DEVICE) 369 return NOTIFY_DONE; 370 371 info = dmar_alloc_pci_notify_info(pdev, action); 372 if (!info) 373 return NOTIFY_DONE; 374 375 down_write(&dmar_global_lock); 376 if (action == BUS_NOTIFY_ADD_DEVICE) 377 dmar_pci_bus_add_dev(info); 378 else if (action == BUS_NOTIFY_REMOVED_DEVICE) 379 dmar_pci_bus_del_dev(info); 380 up_write(&dmar_global_lock); 381 382 dmar_free_pci_notify_info(info); 383 384 return NOTIFY_OK; 385 } 386 387 static struct notifier_block dmar_pci_bus_nb = { 388 .notifier_call = dmar_pci_bus_notifier, 389 .priority = INT_MIN, 390 }; 391 392 static struct dmar_drhd_unit * 393 dmar_find_dmaru(struct acpi_dmar_hardware_unit *drhd) 394 { 395 struct dmar_drhd_unit *dmaru; 396 397 list_for_each_entry_rcu(dmaru, &dmar_drhd_units, list, 398 dmar_rcu_check()) 399 if (dmaru->segment == drhd->segment && 400 dmaru->reg_base_addr == drhd->address) 401 return dmaru; 402 403 return NULL; 404 } 405 406 /* 407 * dmar_parse_one_drhd - parses exactly one DMA remapping hardware definition 408 * structure which uniquely represent one DMA remapping hardware unit 409 * present in the platform 410 */ 411 static int dmar_parse_one_drhd(struct acpi_dmar_header *header, void *arg) 412 { 413 struct acpi_dmar_hardware_unit *drhd; 414 struct dmar_drhd_unit *dmaru; 415 int ret; 416 417 drhd = (struct acpi_dmar_hardware_unit *)header; 418 dmaru = dmar_find_dmaru(drhd); 419 if (dmaru) 420 goto out; 421 422 dmaru = kzalloc(sizeof(*dmaru) + header->length, GFP_KERNEL); 423 if (!dmaru) 424 return -ENOMEM; 425 426 /* 427 * If header is allocated from slab by ACPI _DSM method, we need to 428 * copy the content because the memory buffer will be freed on return. 429 */ 430 dmaru->hdr = (void *)(dmaru + 1); 431 memcpy(dmaru->hdr, header, header->length); 432 dmaru->reg_base_addr = drhd->address; 433 dmaru->segment = drhd->segment; 434 dmaru->include_all = drhd->flags & 0x1; /* BIT0: INCLUDE_ALL */ 435 dmaru->devices = dmar_alloc_dev_scope((void *)(drhd + 1), 436 ((void *)drhd) + drhd->header.length, 437 &dmaru->devices_cnt); 438 if (dmaru->devices_cnt && dmaru->devices == NULL) { 439 kfree(dmaru); 440 return -ENOMEM; 441 } 442 443 ret = alloc_iommu(dmaru); 444 if (ret) { 445 dmar_free_dev_scope(&dmaru->devices, 446 &dmaru->devices_cnt); 447 kfree(dmaru); 448 return ret; 449 } 450 dmar_register_drhd_unit(dmaru); 451 452 out: 453 if (arg) 454 (*(int *)arg)++; 455 456 return 0; 457 } 458 459 static void dmar_free_drhd(struct dmar_drhd_unit *dmaru) 460 { 461 if (dmaru->devices && dmaru->devices_cnt) 462 dmar_free_dev_scope(&dmaru->devices, &dmaru->devices_cnt); 463 if (dmaru->iommu) 464 free_iommu(dmaru->iommu); 465 kfree(dmaru); 466 } 467 468 static int __init dmar_parse_one_andd(struct acpi_dmar_header *header, 469 void *arg) 470 { 471 struct acpi_dmar_andd *andd = (void *)header; 472 473 /* Check for NUL termination within the designated length */ 474 if (strnlen(andd->device_name, header->length - 8) == header->length - 8) { 475 pr_warn(FW_BUG 476 "Your BIOS is broken; ANDD object name is not NUL-terminated\n" 477 "BIOS vendor: %s; Ver: %s; Product Version: %s\n", 478 dmi_get_system_info(DMI_BIOS_VENDOR), 479 dmi_get_system_info(DMI_BIOS_VERSION), 480 dmi_get_system_info(DMI_PRODUCT_VERSION)); 481 add_taint(TAINT_FIRMWARE_WORKAROUND, LOCKDEP_STILL_OK); 482 return -EINVAL; 483 } 484 pr_info("ANDD device: %x name: %s\n", andd->device_number, 485 andd->device_name); 486 487 return 0; 488 } 489 490 #ifdef CONFIG_ACPI_NUMA 491 static int dmar_parse_one_rhsa(struct acpi_dmar_header *header, void *arg) 492 { 493 struct acpi_dmar_rhsa *rhsa; 494 struct dmar_drhd_unit *drhd; 495 496 rhsa = (struct acpi_dmar_rhsa *)header; 497 for_each_drhd_unit(drhd) { 498 if (drhd->reg_base_addr == rhsa->base_address) { 499 int node = pxm_to_node(rhsa->proximity_domain); 500 501 if (!node_online(node)) 502 node = NUMA_NO_NODE; 503 drhd->iommu->node = node; 504 return 0; 505 } 506 } 507 pr_warn(FW_BUG 508 "Your BIOS is broken; RHSA refers to non-existent DMAR unit at %llx\n" 509 "BIOS vendor: %s; Ver: %s; Product Version: %s\n", 510 rhsa->base_address, 511 dmi_get_system_info(DMI_BIOS_VENDOR), 512 dmi_get_system_info(DMI_BIOS_VERSION), 513 dmi_get_system_info(DMI_PRODUCT_VERSION)); 514 add_taint(TAINT_FIRMWARE_WORKAROUND, LOCKDEP_STILL_OK); 515 516 return 0; 517 } 518 #else 519 #define dmar_parse_one_rhsa dmar_res_noop 520 #endif 521 522 static void 523 dmar_table_print_dmar_entry(struct acpi_dmar_header *header) 524 { 525 struct acpi_dmar_hardware_unit *drhd; 526 struct acpi_dmar_reserved_memory *rmrr; 527 struct acpi_dmar_atsr *atsr; 528 struct acpi_dmar_rhsa *rhsa; 529 struct acpi_dmar_satc *satc; 530 531 switch (header->type) { 532 case ACPI_DMAR_TYPE_HARDWARE_UNIT: 533 drhd = container_of(header, struct acpi_dmar_hardware_unit, 534 header); 535 pr_info("DRHD base: %#016Lx flags: %#x\n", 536 (unsigned long long)drhd->address, drhd->flags); 537 break; 538 case ACPI_DMAR_TYPE_RESERVED_MEMORY: 539 rmrr = container_of(header, struct acpi_dmar_reserved_memory, 540 header); 541 pr_info("RMRR base: %#016Lx end: %#016Lx\n", 542 (unsigned long long)rmrr->base_address, 543 (unsigned long long)rmrr->end_address); 544 break; 545 case ACPI_DMAR_TYPE_ROOT_ATS: 546 atsr = container_of(header, struct acpi_dmar_atsr, header); 547 pr_info("ATSR flags: %#x\n", atsr->flags); 548 break; 549 case ACPI_DMAR_TYPE_HARDWARE_AFFINITY: 550 rhsa = container_of(header, struct acpi_dmar_rhsa, header); 551 pr_info("RHSA base: %#016Lx proximity domain: %#x\n", 552 (unsigned long long)rhsa->base_address, 553 rhsa->proximity_domain); 554 break; 555 case ACPI_DMAR_TYPE_NAMESPACE: 556 /* We don't print this here because we need to sanity-check 557 it first. So print it in dmar_parse_one_andd() instead. */ 558 break; 559 case ACPI_DMAR_TYPE_SATC: 560 satc = container_of(header, struct acpi_dmar_satc, header); 561 pr_info("SATC flags: 0x%x\n", satc->flags); 562 break; 563 } 564 } 565 566 /** 567 * dmar_table_detect - checks to see if the platform supports DMAR devices 568 */ 569 static int __init dmar_table_detect(void) 570 { 571 acpi_status status = AE_OK; 572 573 /* if we could find DMAR table, then there are DMAR devices */ 574 status = acpi_get_table(ACPI_SIG_DMAR, 0, &dmar_tbl); 575 576 if (ACPI_SUCCESS(status) && !dmar_tbl) { 577 pr_warn("Unable to map DMAR\n"); 578 status = AE_NOT_FOUND; 579 } 580 581 return ACPI_SUCCESS(status) ? 0 : -ENOENT; 582 } 583 584 static int dmar_walk_remapping_entries(struct acpi_dmar_header *start, 585 size_t len, struct dmar_res_callback *cb) 586 { 587 struct acpi_dmar_header *iter, *next; 588 struct acpi_dmar_header *end = ((void *)start) + len; 589 590 for (iter = start; iter < end; iter = next) { 591 next = (void *)iter + iter->length; 592 if (iter->length == 0) { 593 /* Avoid looping forever on bad ACPI tables */ 594 pr_debug(FW_BUG "Invalid 0-length structure\n"); 595 break; 596 } else if (next > end) { 597 /* Avoid passing table end */ 598 pr_warn(FW_BUG "Record passes table end\n"); 599 return -EINVAL; 600 } 601 602 if (cb->print_entry) 603 dmar_table_print_dmar_entry(iter); 604 605 if (iter->type >= ACPI_DMAR_TYPE_RESERVED) { 606 /* continue for forward compatibility */ 607 pr_debug("Unknown DMAR structure type %d\n", 608 iter->type); 609 } else if (cb->cb[iter->type]) { 610 int ret; 611 612 ret = cb->cb[iter->type](iter, cb->arg[iter->type]); 613 if (ret) 614 return ret; 615 } else if (!cb->ignore_unhandled) { 616 pr_warn("No handler for DMAR structure type %d\n", 617 iter->type); 618 return -EINVAL; 619 } 620 } 621 622 return 0; 623 } 624 625 static inline int dmar_walk_dmar_table(struct acpi_table_dmar *dmar, 626 struct dmar_res_callback *cb) 627 { 628 return dmar_walk_remapping_entries((void *)(dmar + 1), 629 dmar->header.length - sizeof(*dmar), cb); 630 } 631 632 /** 633 * parse_dmar_table - parses the DMA reporting table 634 */ 635 static int __init 636 parse_dmar_table(void) 637 { 638 struct acpi_table_dmar *dmar; 639 int drhd_count = 0; 640 int ret; 641 struct dmar_res_callback cb = { 642 .print_entry = true, 643 .ignore_unhandled = true, 644 .arg[ACPI_DMAR_TYPE_HARDWARE_UNIT] = &drhd_count, 645 .cb[ACPI_DMAR_TYPE_HARDWARE_UNIT] = &dmar_parse_one_drhd, 646 .cb[ACPI_DMAR_TYPE_RESERVED_MEMORY] = &dmar_parse_one_rmrr, 647 .cb[ACPI_DMAR_TYPE_ROOT_ATS] = &dmar_parse_one_atsr, 648 .cb[ACPI_DMAR_TYPE_HARDWARE_AFFINITY] = &dmar_parse_one_rhsa, 649 .cb[ACPI_DMAR_TYPE_NAMESPACE] = &dmar_parse_one_andd, 650 .cb[ACPI_DMAR_TYPE_SATC] = &dmar_parse_one_satc, 651 }; 652 653 /* 654 * Do it again, earlier dmar_tbl mapping could be mapped with 655 * fixed map. 656 */ 657 dmar_table_detect(); 658 659 /* 660 * ACPI tables may not be DMA protected by tboot, so use DMAR copy 661 * SINIT saved in SinitMleData in TXT heap (which is DMA protected) 662 */ 663 dmar_tbl = tboot_get_dmar_table(dmar_tbl); 664 665 dmar = (struct acpi_table_dmar *)dmar_tbl; 666 if (!dmar) 667 return -ENODEV; 668 669 if (dmar->width < PAGE_SHIFT - 1) { 670 pr_warn("Invalid DMAR haw\n"); 671 return -EINVAL; 672 } 673 674 pr_info("Host address width %d\n", dmar->width + 1); 675 ret = dmar_walk_dmar_table(dmar, &cb); 676 if (ret == 0 && drhd_count == 0) 677 pr_warn(FW_BUG "No DRHD structure found in DMAR table\n"); 678 679 return ret; 680 } 681 682 static int dmar_pci_device_match(struct dmar_dev_scope devices[], 683 int cnt, struct pci_dev *dev) 684 { 685 int index; 686 struct device *tmp; 687 688 while (dev) { 689 for_each_active_dev_scope(devices, cnt, index, tmp) 690 if (dev_is_pci(tmp) && dev == to_pci_dev(tmp)) 691 return 1; 692 693 /* Check our parent */ 694 dev = dev->bus->self; 695 } 696 697 return 0; 698 } 699 700 struct dmar_drhd_unit * 701 dmar_find_matched_drhd_unit(struct pci_dev *dev) 702 { 703 struct dmar_drhd_unit *dmaru; 704 struct acpi_dmar_hardware_unit *drhd; 705 706 dev = pci_physfn(dev); 707 708 rcu_read_lock(); 709 for_each_drhd_unit(dmaru) { 710 drhd = container_of(dmaru->hdr, 711 struct acpi_dmar_hardware_unit, 712 header); 713 714 if (dmaru->include_all && 715 drhd->segment == pci_domain_nr(dev->bus)) 716 goto out; 717 718 if (dmar_pci_device_match(dmaru->devices, 719 dmaru->devices_cnt, dev)) 720 goto out; 721 } 722 dmaru = NULL; 723 out: 724 rcu_read_unlock(); 725 726 return dmaru; 727 } 728 729 static void __init dmar_acpi_insert_dev_scope(u8 device_number, 730 struct acpi_device *adev) 731 { 732 struct dmar_drhd_unit *dmaru; 733 struct acpi_dmar_hardware_unit *drhd; 734 struct acpi_dmar_device_scope *scope; 735 struct device *tmp; 736 int i; 737 struct acpi_dmar_pci_path *path; 738 739 for_each_drhd_unit(dmaru) { 740 drhd = container_of(dmaru->hdr, 741 struct acpi_dmar_hardware_unit, 742 header); 743 744 for (scope = (void *)(drhd + 1); 745 (unsigned long)scope < ((unsigned long)drhd) + drhd->header.length; 746 scope = ((void *)scope) + scope->length) { 747 if (scope->entry_type != ACPI_DMAR_SCOPE_TYPE_NAMESPACE) 748 continue; 749 if (scope->enumeration_id != device_number) 750 continue; 751 752 path = (void *)(scope + 1); 753 pr_info("ACPI device \"%s\" under DMAR at %llx as %02x:%02x.%d\n", 754 dev_name(&adev->dev), dmaru->reg_base_addr, 755 scope->bus, path->device, path->function); 756 for_each_dev_scope(dmaru->devices, dmaru->devices_cnt, i, tmp) 757 if (tmp == NULL) { 758 dmaru->devices[i].bus = scope->bus; 759 dmaru->devices[i].devfn = PCI_DEVFN(path->device, 760 path->function); 761 rcu_assign_pointer(dmaru->devices[i].dev, 762 get_device(&adev->dev)); 763 return; 764 } 765 BUG_ON(i >= dmaru->devices_cnt); 766 } 767 } 768 pr_warn("No IOMMU scope found for ANDD enumeration ID %d (%s)\n", 769 device_number, dev_name(&adev->dev)); 770 } 771 772 static int __init dmar_acpi_dev_scope_init(void) 773 { 774 struct acpi_dmar_andd *andd; 775 776 if (dmar_tbl == NULL) 777 return -ENODEV; 778 779 for (andd = (void *)dmar_tbl + sizeof(struct acpi_table_dmar); 780 ((unsigned long)andd) < ((unsigned long)dmar_tbl) + dmar_tbl->length; 781 andd = ((void *)andd) + andd->header.length) { 782 if (andd->header.type == ACPI_DMAR_TYPE_NAMESPACE) { 783 acpi_handle h; 784 struct acpi_device *adev; 785 786 if (!ACPI_SUCCESS(acpi_get_handle(ACPI_ROOT_OBJECT, 787 andd->device_name, 788 &h))) { 789 pr_err("Failed to find handle for ACPI object %s\n", 790 andd->device_name); 791 continue; 792 } 793 if (acpi_bus_get_device(h, &adev)) { 794 pr_err("Failed to get device for ACPI object %s\n", 795 andd->device_name); 796 continue; 797 } 798 dmar_acpi_insert_dev_scope(andd->device_number, adev); 799 } 800 } 801 return 0; 802 } 803 804 int __init dmar_dev_scope_init(void) 805 { 806 struct pci_dev *dev = NULL; 807 struct dmar_pci_notify_info *info; 808 809 if (dmar_dev_scope_status != 1) 810 return dmar_dev_scope_status; 811 812 if (list_empty(&dmar_drhd_units)) { 813 dmar_dev_scope_status = -ENODEV; 814 } else { 815 dmar_dev_scope_status = 0; 816 817 dmar_acpi_dev_scope_init(); 818 819 for_each_pci_dev(dev) { 820 if (dev->is_virtfn) 821 continue; 822 823 info = dmar_alloc_pci_notify_info(dev, 824 BUS_NOTIFY_ADD_DEVICE); 825 if (!info) { 826 return dmar_dev_scope_status; 827 } else { 828 dmar_pci_bus_add_dev(info); 829 dmar_free_pci_notify_info(info); 830 } 831 } 832 } 833 834 return dmar_dev_scope_status; 835 } 836 837 void __init dmar_register_bus_notifier(void) 838 { 839 bus_register_notifier(&pci_bus_type, &dmar_pci_bus_nb); 840 } 841 842 843 int __init dmar_table_init(void) 844 { 845 static int dmar_table_initialized; 846 int ret; 847 848 if (dmar_table_initialized == 0) { 849 ret = parse_dmar_table(); 850 if (ret < 0) { 851 if (ret != -ENODEV) 852 pr_info("Parse DMAR table failure.\n"); 853 } else if (list_empty(&dmar_drhd_units)) { 854 pr_info("No DMAR devices found\n"); 855 ret = -ENODEV; 856 } 857 858 if (ret < 0) 859 dmar_table_initialized = ret; 860 else 861 dmar_table_initialized = 1; 862 } 863 864 return dmar_table_initialized < 0 ? dmar_table_initialized : 0; 865 } 866 867 static void warn_invalid_dmar(u64 addr, const char *message) 868 { 869 pr_warn_once(FW_BUG 870 "Your BIOS is broken; DMAR reported at address %llx%s!\n" 871 "BIOS vendor: %s; Ver: %s; Product Version: %s\n", 872 addr, message, 873 dmi_get_system_info(DMI_BIOS_VENDOR), 874 dmi_get_system_info(DMI_BIOS_VERSION), 875 dmi_get_system_info(DMI_PRODUCT_VERSION)); 876 add_taint(TAINT_FIRMWARE_WORKAROUND, LOCKDEP_STILL_OK); 877 } 878 879 static int __ref 880 dmar_validate_one_drhd(struct acpi_dmar_header *entry, void *arg) 881 { 882 struct acpi_dmar_hardware_unit *drhd; 883 void __iomem *addr; 884 u64 cap, ecap; 885 886 drhd = (void *)entry; 887 if (!drhd->address) { 888 warn_invalid_dmar(0, ""); 889 return -EINVAL; 890 } 891 892 if (arg) 893 addr = ioremap(drhd->address, VTD_PAGE_SIZE); 894 else 895 addr = early_ioremap(drhd->address, VTD_PAGE_SIZE); 896 if (!addr) { 897 pr_warn("Can't validate DRHD address: %llx\n", drhd->address); 898 return -EINVAL; 899 } 900 901 cap = dmar_readq(addr + DMAR_CAP_REG); 902 ecap = dmar_readq(addr + DMAR_ECAP_REG); 903 904 if (arg) 905 iounmap(addr); 906 else 907 early_iounmap(addr, VTD_PAGE_SIZE); 908 909 if (cap == (uint64_t)-1 && ecap == (uint64_t)-1) { 910 warn_invalid_dmar(drhd->address, " returns all ones"); 911 return -EINVAL; 912 } 913 914 return 0; 915 } 916 917 int __init detect_intel_iommu(void) 918 { 919 int ret; 920 struct dmar_res_callback validate_drhd_cb = { 921 .cb[ACPI_DMAR_TYPE_HARDWARE_UNIT] = &dmar_validate_one_drhd, 922 .ignore_unhandled = true, 923 }; 924 925 down_write(&dmar_global_lock); 926 ret = dmar_table_detect(); 927 if (!ret) 928 ret = dmar_walk_dmar_table((struct acpi_table_dmar *)dmar_tbl, 929 &validate_drhd_cb); 930 if (!ret && !no_iommu && !iommu_detected && 931 (!dmar_disabled || dmar_platform_optin())) { 932 iommu_detected = 1; 933 /* Make sure ACS will be enabled */ 934 pci_request_acs(); 935 } 936 937 #ifdef CONFIG_X86 938 if (!ret) { 939 x86_init.iommu.iommu_init = intel_iommu_init; 940 x86_platform.iommu_shutdown = intel_iommu_shutdown; 941 } 942 943 #endif 944 945 if (dmar_tbl) { 946 acpi_put_table(dmar_tbl); 947 dmar_tbl = NULL; 948 } 949 up_write(&dmar_global_lock); 950 951 return ret ? ret : 1; 952 } 953 954 static void unmap_iommu(struct intel_iommu *iommu) 955 { 956 iounmap(iommu->reg); 957 release_mem_region(iommu->reg_phys, iommu->reg_size); 958 } 959 960 /** 961 * map_iommu: map the iommu's registers 962 * @iommu: the iommu to map 963 * @phys_addr: the physical address of the base resgister 964 * 965 * Memory map the iommu's registers. Start w/ a single page, and 966 * possibly expand if that turns out to be insufficent. 967 */ 968 static int map_iommu(struct intel_iommu *iommu, u64 phys_addr) 969 { 970 int map_size, err=0; 971 972 iommu->reg_phys = phys_addr; 973 iommu->reg_size = VTD_PAGE_SIZE; 974 975 if (!request_mem_region(iommu->reg_phys, iommu->reg_size, iommu->name)) { 976 pr_err("Can't reserve memory\n"); 977 err = -EBUSY; 978 goto out; 979 } 980 981 iommu->reg = ioremap(iommu->reg_phys, iommu->reg_size); 982 if (!iommu->reg) { 983 pr_err("Can't map the region\n"); 984 err = -ENOMEM; 985 goto release; 986 } 987 988 iommu->cap = dmar_readq(iommu->reg + DMAR_CAP_REG); 989 iommu->ecap = dmar_readq(iommu->reg + DMAR_ECAP_REG); 990 991 if (iommu->cap == (uint64_t)-1 && iommu->ecap == (uint64_t)-1) { 992 err = -EINVAL; 993 warn_invalid_dmar(phys_addr, " returns all ones"); 994 goto unmap; 995 } 996 if (ecap_vcs(iommu->ecap)) 997 iommu->vccap = dmar_readq(iommu->reg + DMAR_VCCAP_REG); 998 999 /* the registers might be more than one page */ 1000 map_size = max_t(int, ecap_max_iotlb_offset(iommu->ecap), 1001 cap_max_fault_reg_offset(iommu->cap)); 1002 map_size = VTD_PAGE_ALIGN(map_size); 1003 if (map_size > iommu->reg_size) { 1004 iounmap(iommu->reg); 1005 release_mem_region(iommu->reg_phys, iommu->reg_size); 1006 iommu->reg_size = map_size; 1007 if (!request_mem_region(iommu->reg_phys, iommu->reg_size, 1008 iommu->name)) { 1009 pr_err("Can't reserve memory\n"); 1010 err = -EBUSY; 1011 goto out; 1012 } 1013 iommu->reg = ioremap(iommu->reg_phys, iommu->reg_size); 1014 if (!iommu->reg) { 1015 pr_err("Can't map the region\n"); 1016 err = -ENOMEM; 1017 goto release; 1018 } 1019 } 1020 err = 0; 1021 goto out; 1022 1023 unmap: 1024 iounmap(iommu->reg); 1025 release: 1026 release_mem_region(iommu->reg_phys, iommu->reg_size); 1027 out: 1028 return err; 1029 } 1030 1031 static int dmar_alloc_seq_id(struct intel_iommu *iommu) 1032 { 1033 iommu->seq_id = find_first_zero_bit(dmar_seq_ids, 1034 DMAR_UNITS_SUPPORTED); 1035 if (iommu->seq_id >= DMAR_UNITS_SUPPORTED) { 1036 iommu->seq_id = -1; 1037 } else { 1038 set_bit(iommu->seq_id, dmar_seq_ids); 1039 sprintf(iommu->name, "dmar%d", iommu->seq_id); 1040 } 1041 1042 return iommu->seq_id; 1043 } 1044 1045 static void dmar_free_seq_id(struct intel_iommu *iommu) 1046 { 1047 if (iommu->seq_id >= 0) { 1048 clear_bit(iommu->seq_id, dmar_seq_ids); 1049 iommu->seq_id = -1; 1050 } 1051 } 1052 1053 static int alloc_iommu(struct dmar_drhd_unit *drhd) 1054 { 1055 struct intel_iommu *iommu; 1056 u32 ver, sts; 1057 int agaw = -1; 1058 int msagaw = -1; 1059 int err; 1060 1061 if (!drhd->reg_base_addr) { 1062 warn_invalid_dmar(0, ""); 1063 return -EINVAL; 1064 } 1065 1066 iommu = kzalloc(sizeof(*iommu), GFP_KERNEL); 1067 if (!iommu) 1068 return -ENOMEM; 1069 1070 if (dmar_alloc_seq_id(iommu) < 0) { 1071 pr_err("Failed to allocate seq_id\n"); 1072 err = -ENOSPC; 1073 goto error; 1074 } 1075 1076 err = map_iommu(iommu, drhd->reg_base_addr); 1077 if (err) { 1078 pr_err("Failed to map %s\n", iommu->name); 1079 goto error_free_seq_id; 1080 } 1081 1082 err = -EINVAL; 1083 if (cap_sagaw(iommu->cap) == 0) { 1084 pr_info("%s: No supported address widths. Not attempting DMA translation.\n", 1085 iommu->name); 1086 drhd->ignored = 1; 1087 } 1088 1089 if (!drhd->ignored) { 1090 agaw = iommu_calculate_agaw(iommu); 1091 if (agaw < 0) { 1092 pr_err("Cannot get a valid agaw for iommu (seq_id = %d)\n", 1093 iommu->seq_id); 1094 drhd->ignored = 1; 1095 } 1096 } 1097 if (!drhd->ignored) { 1098 msagaw = iommu_calculate_max_sagaw(iommu); 1099 if (msagaw < 0) { 1100 pr_err("Cannot get a valid max agaw for iommu (seq_id = %d)\n", 1101 iommu->seq_id); 1102 drhd->ignored = 1; 1103 agaw = -1; 1104 } 1105 } 1106 iommu->agaw = agaw; 1107 iommu->msagaw = msagaw; 1108 iommu->segment = drhd->segment; 1109 1110 iommu->node = NUMA_NO_NODE; 1111 1112 ver = readl(iommu->reg + DMAR_VER_REG); 1113 pr_info("%s: reg_base_addr %llx ver %d:%d cap %llx ecap %llx\n", 1114 iommu->name, 1115 (unsigned long long)drhd->reg_base_addr, 1116 DMAR_VER_MAJOR(ver), DMAR_VER_MINOR(ver), 1117 (unsigned long long)iommu->cap, 1118 (unsigned long long)iommu->ecap); 1119 1120 /* Reflect status in gcmd */ 1121 sts = readl(iommu->reg + DMAR_GSTS_REG); 1122 if (sts & DMA_GSTS_IRES) 1123 iommu->gcmd |= DMA_GCMD_IRE; 1124 if (sts & DMA_GSTS_TES) 1125 iommu->gcmd |= DMA_GCMD_TE; 1126 if (sts & DMA_GSTS_QIES) 1127 iommu->gcmd |= DMA_GCMD_QIE; 1128 1129 raw_spin_lock_init(&iommu->register_lock); 1130 1131 /* 1132 * This is only for hotplug; at boot time intel_iommu_enabled won't 1133 * be set yet. When intel_iommu_init() runs, it registers the units 1134 * present at boot time, then sets intel_iommu_enabled. 1135 */ 1136 if (intel_iommu_enabled && !drhd->ignored) { 1137 err = iommu_device_sysfs_add(&iommu->iommu, NULL, 1138 intel_iommu_groups, 1139 "%s", iommu->name); 1140 if (err) 1141 goto err_unmap; 1142 1143 err = iommu_device_register(&iommu->iommu, &intel_iommu_ops, NULL); 1144 if (err) 1145 goto err_unmap; 1146 } 1147 1148 drhd->iommu = iommu; 1149 iommu->drhd = drhd; 1150 1151 return 0; 1152 1153 err_unmap: 1154 unmap_iommu(iommu); 1155 error_free_seq_id: 1156 dmar_free_seq_id(iommu); 1157 error: 1158 kfree(iommu); 1159 return err; 1160 } 1161 1162 static void free_iommu(struct intel_iommu *iommu) 1163 { 1164 if (intel_iommu_enabled && !iommu->drhd->ignored) { 1165 iommu_device_unregister(&iommu->iommu); 1166 iommu_device_sysfs_remove(&iommu->iommu); 1167 } 1168 1169 if (iommu->irq) { 1170 if (iommu->pr_irq) { 1171 free_irq(iommu->pr_irq, iommu); 1172 dmar_free_hwirq(iommu->pr_irq); 1173 iommu->pr_irq = 0; 1174 } 1175 free_irq(iommu->irq, iommu); 1176 dmar_free_hwirq(iommu->irq); 1177 iommu->irq = 0; 1178 } 1179 1180 if (iommu->qi) { 1181 free_page((unsigned long)iommu->qi->desc); 1182 kfree(iommu->qi->desc_status); 1183 kfree(iommu->qi); 1184 } 1185 1186 if (iommu->reg) 1187 unmap_iommu(iommu); 1188 1189 dmar_free_seq_id(iommu); 1190 kfree(iommu); 1191 } 1192 1193 /* 1194 * Reclaim all the submitted descriptors which have completed its work. 1195 */ 1196 static inline void reclaim_free_desc(struct q_inval *qi) 1197 { 1198 while (qi->desc_status[qi->free_tail] == QI_DONE || 1199 qi->desc_status[qi->free_tail] == QI_ABORT) { 1200 qi->desc_status[qi->free_tail] = QI_FREE; 1201 qi->free_tail = (qi->free_tail + 1) % QI_LENGTH; 1202 qi->free_cnt++; 1203 } 1204 } 1205 1206 static const char *qi_type_string(u8 type) 1207 { 1208 switch (type) { 1209 case QI_CC_TYPE: 1210 return "Context-cache Invalidation"; 1211 case QI_IOTLB_TYPE: 1212 return "IOTLB Invalidation"; 1213 case QI_DIOTLB_TYPE: 1214 return "Device-TLB Invalidation"; 1215 case QI_IEC_TYPE: 1216 return "Interrupt Entry Cache Invalidation"; 1217 case QI_IWD_TYPE: 1218 return "Invalidation Wait"; 1219 case QI_EIOTLB_TYPE: 1220 return "PASID-based IOTLB Invalidation"; 1221 case QI_PC_TYPE: 1222 return "PASID-cache Invalidation"; 1223 case QI_DEIOTLB_TYPE: 1224 return "PASID-based Device-TLB Invalidation"; 1225 case QI_PGRP_RESP_TYPE: 1226 return "Page Group Response"; 1227 default: 1228 return "UNKNOWN"; 1229 } 1230 } 1231 1232 static void qi_dump_fault(struct intel_iommu *iommu, u32 fault) 1233 { 1234 unsigned int head = dmar_readl(iommu->reg + DMAR_IQH_REG); 1235 u64 iqe_err = dmar_readq(iommu->reg + DMAR_IQER_REG); 1236 struct qi_desc *desc = iommu->qi->desc + head; 1237 1238 if (fault & DMA_FSTS_IQE) 1239 pr_err("VT-d detected Invalidation Queue Error: Reason %llx", 1240 DMAR_IQER_REG_IQEI(iqe_err)); 1241 if (fault & DMA_FSTS_ITE) 1242 pr_err("VT-d detected Invalidation Time-out Error: SID %llx", 1243 DMAR_IQER_REG_ITESID(iqe_err)); 1244 if (fault & DMA_FSTS_ICE) 1245 pr_err("VT-d detected Invalidation Completion Error: SID %llx", 1246 DMAR_IQER_REG_ICESID(iqe_err)); 1247 1248 pr_err("QI HEAD: %s qw0 = 0x%llx, qw1 = 0x%llx\n", 1249 qi_type_string(desc->qw0 & 0xf), 1250 (unsigned long long)desc->qw0, 1251 (unsigned long long)desc->qw1); 1252 1253 head = ((head >> qi_shift(iommu)) + QI_LENGTH - 1) % QI_LENGTH; 1254 head <<= qi_shift(iommu); 1255 desc = iommu->qi->desc + head; 1256 1257 pr_err("QI PRIOR: %s qw0 = 0x%llx, qw1 = 0x%llx\n", 1258 qi_type_string(desc->qw0 & 0xf), 1259 (unsigned long long)desc->qw0, 1260 (unsigned long long)desc->qw1); 1261 } 1262 1263 static int qi_check_fault(struct intel_iommu *iommu, int index, int wait_index) 1264 { 1265 u32 fault; 1266 int head, tail; 1267 struct q_inval *qi = iommu->qi; 1268 int shift = qi_shift(iommu); 1269 1270 if (qi->desc_status[wait_index] == QI_ABORT) 1271 return -EAGAIN; 1272 1273 fault = readl(iommu->reg + DMAR_FSTS_REG); 1274 if (fault & (DMA_FSTS_IQE | DMA_FSTS_ITE | DMA_FSTS_ICE)) 1275 qi_dump_fault(iommu, fault); 1276 1277 /* 1278 * If IQE happens, the head points to the descriptor associated 1279 * with the error. No new descriptors are fetched until the IQE 1280 * is cleared. 1281 */ 1282 if (fault & DMA_FSTS_IQE) { 1283 head = readl(iommu->reg + DMAR_IQH_REG); 1284 if ((head >> shift) == index) { 1285 struct qi_desc *desc = qi->desc + head; 1286 1287 /* 1288 * desc->qw2 and desc->qw3 are either reserved or 1289 * used by software as private data. We won't print 1290 * out these two qw's for security consideration. 1291 */ 1292 memcpy(desc, qi->desc + (wait_index << shift), 1293 1 << shift); 1294 writel(DMA_FSTS_IQE, iommu->reg + DMAR_FSTS_REG); 1295 pr_info("Invalidation Queue Error (IQE) cleared\n"); 1296 return -EINVAL; 1297 } 1298 } 1299 1300 /* 1301 * If ITE happens, all pending wait_desc commands are aborted. 1302 * No new descriptors are fetched until the ITE is cleared. 1303 */ 1304 if (fault & DMA_FSTS_ITE) { 1305 head = readl(iommu->reg + DMAR_IQH_REG); 1306 head = ((head >> shift) - 1 + QI_LENGTH) % QI_LENGTH; 1307 head |= 1; 1308 tail = readl(iommu->reg + DMAR_IQT_REG); 1309 tail = ((tail >> shift) - 1 + QI_LENGTH) % QI_LENGTH; 1310 1311 writel(DMA_FSTS_ITE, iommu->reg + DMAR_FSTS_REG); 1312 pr_info("Invalidation Time-out Error (ITE) cleared\n"); 1313 1314 do { 1315 if (qi->desc_status[head] == QI_IN_USE) 1316 qi->desc_status[head] = QI_ABORT; 1317 head = (head - 2 + QI_LENGTH) % QI_LENGTH; 1318 } while (head != tail); 1319 1320 if (qi->desc_status[wait_index] == QI_ABORT) 1321 return -EAGAIN; 1322 } 1323 1324 if (fault & DMA_FSTS_ICE) { 1325 writel(DMA_FSTS_ICE, iommu->reg + DMAR_FSTS_REG); 1326 pr_info("Invalidation Completion Error (ICE) cleared\n"); 1327 } 1328 1329 return 0; 1330 } 1331 1332 /* 1333 * Function to submit invalidation descriptors of all types to the queued 1334 * invalidation interface(QI). Multiple descriptors can be submitted at a 1335 * time, a wait descriptor will be appended to each submission to ensure 1336 * hardware has completed the invalidation before return. Wait descriptors 1337 * can be part of the submission but it will not be polled for completion. 1338 */ 1339 int qi_submit_sync(struct intel_iommu *iommu, struct qi_desc *desc, 1340 unsigned int count, unsigned long options) 1341 { 1342 struct q_inval *qi = iommu->qi; 1343 struct qi_desc wait_desc; 1344 int wait_index, index; 1345 unsigned long flags; 1346 int offset, shift; 1347 int rc, i; 1348 1349 if (!qi) 1350 return 0; 1351 1352 restart: 1353 rc = 0; 1354 1355 raw_spin_lock_irqsave(&qi->q_lock, flags); 1356 /* 1357 * Check if we have enough empty slots in the queue to submit, 1358 * the calculation is based on: 1359 * # of desc + 1 wait desc + 1 space between head and tail 1360 */ 1361 while (qi->free_cnt < count + 2) { 1362 raw_spin_unlock_irqrestore(&qi->q_lock, flags); 1363 cpu_relax(); 1364 raw_spin_lock_irqsave(&qi->q_lock, flags); 1365 } 1366 1367 index = qi->free_head; 1368 wait_index = (index + count) % QI_LENGTH; 1369 shift = qi_shift(iommu); 1370 1371 for (i = 0; i < count; i++) { 1372 offset = ((index + i) % QI_LENGTH) << shift; 1373 memcpy(qi->desc + offset, &desc[i], 1 << shift); 1374 qi->desc_status[(index + i) % QI_LENGTH] = QI_IN_USE; 1375 trace_qi_submit(iommu, desc[i].qw0, desc[i].qw1, 1376 desc[i].qw2, desc[i].qw3); 1377 } 1378 qi->desc_status[wait_index] = QI_IN_USE; 1379 1380 wait_desc.qw0 = QI_IWD_STATUS_DATA(QI_DONE) | 1381 QI_IWD_STATUS_WRITE | QI_IWD_TYPE; 1382 if (options & QI_OPT_WAIT_DRAIN) 1383 wait_desc.qw0 |= QI_IWD_PRQ_DRAIN; 1384 wait_desc.qw1 = virt_to_phys(&qi->desc_status[wait_index]); 1385 wait_desc.qw2 = 0; 1386 wait_desc.qw3 = 0; 1387 1388 offset = wait_index << shift; 1389 memcpy(qi->desc + offset, &wait_desc, 1 << shift); 1390 1391 qi->free_head = (qi->free_head + count + 1) % QI_LENGTH; 1392 qi->free_cnt -= count + 1; 1393 1394 /* 1395 * update the HW tail register indicating the presence of 1396 * new descriptors. 1397 */ 1398 writel(qi->free_head << shift, iommu->reg + DMAR_IQT_REG); 1399 1400 while (qi->desc_status[wait_index] != QI_DONE) { 1401 /* 1402 * We will leave the interrupts disabled, to prevent interrupt 1403 * context to queue another cmd while a cmd is already submitted 1404 * and waiting for completion on this cpu. This is to avoid 1405 * a deadlock where the interrupt context can wait indefinitely 1406 * for free slots in the queue. 1407 */ 1408 rc = qi_check_fault(iommu, index, wait_index); 1409 if (rc) 1410 break; 1411 1412 raw_spin_unlock(&qi->q_lock); 1413 cpu_relax(); 1414 raw_spin_lock(&qi->q_lock); 1415 } 1416 1417 for (i = 0; i < count; i++) 1418 qi->desc_status[(index + i) % QI_LENGTH] = QI_DONE; 1419 1420 reclaim_free_desc(qi); 1421 raw_spin_unlock_irqrestore(&qi->q_lock, flags); 1422 1423 if (rc == -EAGAIN) 1424 goto restart; 1425 1426 return rc; 1427 } 1428 1429 /* 1430 * Flush the global interrupt entry cache. 1431 */ 1432 void qi_global_iec(struct intel_iommu *iommu) 1433 { 1434 struct qi_desc desc; 1435 1436 desc.qw0 = QI_IEC_TYPE; 1437 desc.qw1 = 0; 1438 desc.qw2 = 0; 1439 desc.qw3 = 0; 1440 1441 /* should never fail */ 1442 qi_submit_sync(iommu, &desc, 1, 0); 1443 } 1444 1445 void qi_flush_context(struct intel_iommu *iommu, u16 did, u16 sid, u8 fm, 1446 u64 type) 1447 { 1448 struct qi_desc desc; 1449 1450 desc.qw0 = QI_CC_FM(fm) | QI_CC_SID(sid) | QI_CC_DID(did) 1451 | QI_CC_GRAN(type) | QI_CC_TYPE; 1452 desc.qw1 = 0; 1453 desc.qw2 = 0; 1454 desc.qw3 = 0; 1455 1456 qi_submit_sync(iommu, &desc, 1, 0); 1457 } 1458 1459 void qi_flush_iotlb(struct intel_iommu *iommu, u16 did, u64 addr, 1460 unsigned int size_order, u64 type) 1461 { 1462 u8 dw = 0, dr = 0; 1463 1464 struct qi_desc desc; 1465 int ih = 0; 1466 1467 if (cap_write_drain(iommu->cap)) 1468 dw = 1; 1469 1470 if (cap_read_drain(iommu->cap)) 1471 dr = 1; 1472 1473 desc.qw0 = QI_IOTLB_DID(did) | QI_IOTLB_DR(dr) | QI_IOTLB_DW(dw) 1474 | QI_IOTLB_GRAN(type) | QI_IOTLB_TYPE; 1475 desc.qw1 = QI_IOTLB_ADDR(addr) | QI_IOTLB_IH(ih) 1476 | QI_IOTLB_AM(size_order); 1477 desc.qw2 = 0; 1478 desc.qw3 = 0; 1479 1480 qi_submit_sync(iommu, &desc, 1, 0); 1481 } 1482 1483 void qi_flush_dev_iotlb(struct intel_iommu *iommu, u16 sid, u16 pfsid, 1484 u16 qdep, u64 addr, unsigned mask) 1485 { 1486 struct qi_desc desc; 1487 1488 if (mask) { 1489 addr |= (1ULL << (VTD_PAGE_SHIFT + mask - 1)) - 1; 1490 desc.qw1 = QI_DEV_IOTLB_ADDR(addr) | QI_DEV_IOTLB_SIZE; 1491 } else 1492 desc.qw1 = QI_DEV_IOTLB_ADDR(addr); 1493 1494 if (qdep >= QI_DEV_IOTLB_MAX_INVS) 1495 qdep = 0; 1496 1497 desc.qw0 = QI_DEV_IOTLB_SID(sid) | QI_DEV_IOTLB_QDEP(qdep) | 1498 QI_DIOTLB_TYPE | QI_DEV_IOTLB_PFSID(pfsid); 1499 desc.qw2 = 0; 1500 desc.qw3 = 0; 1501 1502 qi_submit_sync(iommu, &desc, 1, 0); 1503 } 1504 1505 /* PASID-based IOTLB invalidation */ 1506 void qi_flush_piotlb(struct intel_iommu *iommu, u16 did, u32 pasid, u64 addr, 1507 unsigned long npages, bool ih) 1508 { 1509 struct qi_desc desc = {.qw2 = 0, .qw3 = 0}; 1510 1511 /* 1512 * npages == -1 means a PASID-selective invalidation, otherwise, 1513 * a positive value for Page-selective-within-PASID invalidation. 1514 * 0 is not a valid input. 1515 */ 1516 if (WARN_ON(!npages)) { 1517 pr_err("Invalid input npages = %ld\n", npages); 1518 return; 1519 } 1520 1521 if (npages == -1) { 1522 desc.qw0 = QI_EIOTLB_PASID(pasid) | 1523 QI_EIOTLB_DID(did) | 1524 QI_EIOTLB_GRAN(QI_GRAN_NONG_PASID) | 1525 QI_EIOTLB_TYPE; 1526 desc.qw1 = 0; 1527 } else { 1528 int mask = ilog2(__roundup_pow_of_two(npages)); 1529 unsigned long align = (1ULL << (VTD_PAGE_SHIFT + mask)); 1530 1531 if (WARN_ON_ONCE(!IS_ALIGNED(addr, align))) 1532 addr = ALIGN_DOWN(addr, align); 1533 1534 desc.qw0 = QI_EIOTLB_PASID(pasid) | 1535 QI_EIOTLB_DID(did) | 1536 QI_EIOTLB_GRAN(QI_GRAN_PSI_PASID) | 1537 QI_EIOTLB_TYPE; 1538 desc.qw1 = QI_EIOTLB_ADDR(addr) | 1539 QI_EIOTLB_IH(ih) | 1540 QI_EIOTLB_AM(mask); 1541 } 1542 1543 qi_submit_sync(iommu, &desc, 1, 0); 1544 } 1545 1546 /* PASID-based device IOTLB Invalidate */ 1547 void qi_flush_dev_iotlb_pasid(struct intel_iommu *iommu, u16 sid, u16 pfsid, 1548 u32 pasid, u16 qdep, u64 addr, unsigned int size_order) 1549 { 1550 unsigned long mask = 1UL << (VTD_PAGE_SHIFT + size_order - 1); 1551 struct qi_desc desc = {.qw1 = 0, .qw2 = 0, .qw3 = 0}; 1552 1553 desc.qw0 = QI_DEV_EIOTLB_PASID(pasid) | QI_DEV_EIOTLB_SID(sid) | 1554 QI_DEV_EIOTLB_QDEP(qdep) | QI_DEIOTLB_TYPE | 1555 QI_DEV_IOTLB_PFSID(pfsid); 1556 1557 /* 1558 * If S bit is 0, we only flush a single page. If S bit is set, 1559 * The least significant zero bit indicates the invalidation address 1560 * range. VT-d spec 6.5.2.6. 1561 * e.g. address bit 12[0] indicates 8KB, 13[0] indicates 16KB. 1562 * size order = 0 is PAGE_SIZE 4KB 1563 * Max Invs Pending (MIP) is set to 0 for now until we have DIT in 1564 * ECAP. 1565 */ 1566 if (!IS_ALIGNED(addr, VTD_PAGE_SIZE << size_order)) 1567 pr_warn_ratelimited("Invalidate non-aligned address %llx, order %d\n", 1568 addr, size_order); 1569 1570 /* Take page address */ 1571 desc.qw1 = QI_DEV_EIOTLB_ADDR(addr); 1572 1573 if (size_order) { 1574 /* 1575 * Existing 0s in address below size_order may be the least 1576 * significant bit, we must set them to 1s to avoid having 1577 * smaller size than desired. 1578 */ 1579 desc.qw1 |= GENMASK_ULL(size_order + VTD_PAGE_SHIFT - 1, 1580 VTD_PAGE_SHIFT); 1581 /* Clear size_order bit to indicate size */ 1582 desc.qw1 &= ~mask; 1583 /* Set the S bit to indicate flushing more than 1 page */ 1584 desc.qw1 |= QI_DEV_EIOTLB_SIZE; 1585 } 1586 1587 qi_submit_sync(iommu, &desc, 1, 0); 1588 } 1589 1590 void qi_flush_pasid_cache(struct intel_iommu *iommu, u16 did, 1591 u64 granu, u32 pasid) 1592 { 1593 struct qi_desc desc = {.qw1 = 0, .qw2 = 0, .qw3 = 0}; 1594 1595 desc.qw0 = QI_PC_PASID(pasid) | QI_PC_DID(did) | 1596 QI_PC_GRAN(granu) | QI_PC_TYPE; 1597 qi_submit_sync(iommu, &desc, 1, 0); 1598 } 1599 1600 /* 1601 * Disable Queued Invalidation interface. 1602 */ 1603 void dmar_disable_qi(struct intel_iommu *iommu) 1604 { 1605 unsigned long flags; 1606 u32 sts; 1607 cycles_t start_time = get_cycles(); 1608 1609 if (!ecap_qis(iommu->ecap)) 1610 return; 1611 1612 raw_spin_lock_irqsave(&iommu->register_lock, flags); 1613 1614 sts = readl(iommu->reg + DMAR_GSTS_REG); 1615 if (!(sts & DMA_GSTS_QIES)) 1616 goto end; 1617 1618 /* 1619 * Give a chance to HW to complete the pending invalidation requests. 1620 */ 1621 while ((readl(iommu->reg + DMAR_IQT_REG) != 1622 readl(iommu->reg + DMAR_IQH_REG)) && 1623 (DMAR_OPERATION_TIMEOUT > (get_cycles() - start_time))) 1624 cpu_relax(); 1625 1626 iommu->gcmd &= ~DMA_GCMD_QIE; 1627 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG); 1628 1629 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, readl, 1630 !(sts & DMA_GSTS_QIES), sts); 1631 end: 1632 raw_spin_unlock_irqrestore(&iommu->register_lock, flags); 1633 } 1634 1635 /* 1636 * Enable queued invalidation. 1637 */ 1638 static void __dmar_enable_qi(struct intel_iommu *iommu) 1639 { 1640 u32 sts; 1641 unsigned long flags; 1642 struct q_inval *qi = iommu->qi; 1643 u64 val = virt_to_phys(qi->desc); 1644 1645 qi->free_head = qi->free_tail = 0; 1646 qi->free_cnt = QI_LENGTH; 1647 1648 /* 1649 * Set DW=1 and QS=1 in IQA_REG when Scalable Mode capability 1650 * is present. 1651 */ 1652 if (ecap_smts(iommu->ecap)) 1653 val |= (1 << 11) | 1; 1654 1655 raw_spin_lock_irqsave(&iommu->register_lock, flags); 1656 1657 /* write zero to the tail reg */ 1658 writel(0, iommu->reg + DMAR_IQT_REG); 1659 1660 dmar_writeq(iommu->reg + DMAR_IQA_REG, val); 1661 1662 iommu->gcmd |= DMA_GCMD_QIE; 1663 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG); 1664 1665 /* Make sure hardware complete it */ 1666 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, readl, (sts & DMA_GSTS_QIES), sts); 1667 1668 raw_spin_unlock_irqrestore(&iommu->register_lock, flags); 1669 } 1670 1671 /* 1672 * Enable Queued Invalidation interface. This is a must to support 1673 * interrupt-remapping. Also used by DMA-remapping, which replaces 1674 * register based IOTLB invalidation. 1675 */ 1676 int dmar_enable_qi(struct intel_iommu *iommu) 1677 { 1678 struct q_inval *qi; 1679 struct page *desc_page; 1680 1681 if (!ecap_qis(iommu->ecap)) 1682 return -ENOENT; 1683 1684 /* 1685 * queued invalidation is already setup and enabled. 1686 */ 1687 if (iommu->qi) 1688 return 0; 1689 1690 iommu->qi = kmalloc(sizeof(*qi), GFP_ATOMIC); 1691 if (!iommu->qi) 1692 return -ENOMEM; 1693 1694 qi = iommu->qi; 1695 1696 /* 1697 * Need two pages to accommodate 256 descriptors of 256 bits each 1698 * if the remapping hardware supports scalable mode translation. 1699 */ 1700 desc_page = alloc_pages_node(iommu->node, GFP_ATOMIC | __GFP_ZERO, 1701 !!ecap_smts(iommu->ecap)); 1702 if (!desc_page) { 1703 kfree(qi); 1704 iommu->qi = NULL; 1705 return -ENOMEM; 1706 } 1707 1708 qi->desc = page_address(desc_page); 1709 1710 qi->desc_status = kcalloc(QI_LENGTH, sizeof(int), GFP_ATOMIC); 1711 if (!qi->desc_status) { 1712 free_page((unsigned long) qi->desc); 1713 kfree(qi); 1714 iommu->qi = NULL; 1715 return -ENOMEM; 1716 } 1717 1718 raw_spin_lock_init(&qi->q_lock); 1719 1720 __dmar_enable_qi(iommu); 1721 1722 return 0; 1723 } 1724 1725 /* iommu interrupt handling. Most stuff are MSI-like. */ 1726 1727 enum faulttype { 1728 DMA_REMAP, 1729 INTR_REMAP, 1730 UNKNOWN, 1731 }; 1732 1733 static const char *dma_remap_fault_reasons[] = 1734 { 1735 "Software", 1736 "Present bit in root entry is clear", 1737 "Present bit in context entry is clear", 1738 "Invalid context entry", 1739 "Access beyond MGAW", 1740 "PTE Write access is not set", 1741 "PTE Read access is not set", 1742 "Next page table ptr is invalid", 1743 "Root table address invalid", 1744 "Context table ptr is invalid", 1745 "non-zero reserved fields in RTP", 1746 "non-zero reserved fields in CTP", 1747 "non-zero reserved fields in PTE", 1748 "PCE for translation request specifies blocking", 1749 }; 1750 1751 static const char * const dma_remap_sm_fault_reasons[] = { 1752 "SM: Invalid Root Table Address", 1753 "SM: TTM 0 for request with PASID", 1754 "SM: TTM 0 for page group request", 1755 "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x33-0x37 */ 1756 "SM: Error attempting to access Root Entry", 1757 "SM: Present bit in Root Entry is clear", 1758 "SM: Non-zero reserved field set in Root Entry", 1759 "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x3B-0x3F */ 1760 "SM: Error attempting to access Context Entry", 1761 "SM: Present bit in Context Entry is clear", 1762 "SM: Non-zero reserved field set in the Context Entry", 1763 "SM: Invalid Context Entry", 1764 "SM: DTE field in Context Entry is clear", 1765 "SM: PASID Enable field in Context Entry is clear", 1766 "SM: PASID is larger than the max in Context Entry", 1767 "SM: PRE field in Context-Entry is clear", 1768 "SM: RID_PASID field error in Context-Entry", 1769 "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x49-0x4F */ 1770 "SM: Error attempting to access the PASID Directory Entry", 1771 "SM: Present bit in Directory Entry is clear", 1772 "SM: Non-zero reserved field set in PASID Directory Entry", 1773 "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x53-0x57 */ 1774 "SM: Error attempting to access PASID Table Entry", 1775 "SM: Present bit in PASID Table Entry is clear", 1776 "SM: Non-zero reserved field set in PASID Table Entry", 1777 "SM: Invalid Scalable-Mode PASID Table Entry", 1778 "SM: ERE field is clear in PASID Table Entry", 1779 "SM: SRE field is clear in PASID Table Entry", 1780 "Unknown", "Unknown",/* 0x5E-0x5F */ 1781 "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x60-0x67 */ 1782 "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x68-0x6F */ 1783 "SM: Error attempting to access first-level paging entry", 1784 "SM: Present bit in first-level paging entry is clear", 1785 "SM: Non-zero reserved field set in first-level paging entry", 1786 "SM: Error attempting to access FL-PML4 entry", 1787 "SM: First-level entry address beyond MGAW in Nested translation", 1788 "SM: Read permission error in FL-PML4 entry in Nested translation", 1789 "SM: Read permission error in first-level paging entry in Nested translation", 1790 "SM: Write permission error in first-level paging entry in Nested translation", 1791 "SM: Error attempting to access second-level paging entry", 1792 "SM: Read/Write permission error in second-level paging entry", 1793 "SM: Non-zero reserved field set in second-level paging entry", 1794 "SM: Invalid second-level page table pointer", 1795 "SM: A/D bit update needed in second-level entry when set up in no snoop", 1796 "Unknown", "Unknown", "Unknown", /* 0x7D-0x7F */ 1797 "SM: Address in first-level translation is not canonical", 1798 "SM: U/S set 0 for first-level translation with user privilege", 1799 "SM: No execute permission for request with PASID and ER=1", 1800 "SM: Address beyond the DMA hardware max", 1801 "SM: Second-level entry address beyond the max", 1802 "SM: No write permission for Write/AtomicOp request", 1803 "SM: No read permission for Read/AtomicOp request", 1804 "SM: Invalid address-interrupt address", 1805 "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x88-0x8F */ 1806 "SM: A/D bit update needed in first-level entry when set up in no snoop", 1807 }; 1808 1809 static const char *irq_remap_fault_reasons[] = 1810 { 1811 "Detected reserved fields in the decoded interrupt-remapped request", 1812 "Interrupt index exceeded the interrupt-remapping table size", 1813 "Present field in the IRTE entry is clear", 1814 "Error accessing interrupt-remapping table pointed by IRTA_REG", 1815 "Detected reserved fields in the IRTE entry", 1816 "Blocked a compatibility format interrupt request", 1817 "Blocked an interrupt request due to source-id verification failure", 1818 }; 1819 1820 static const char *dmar_get_fault_reason(u8 fault_reason, int *fault_type) 1821 { 1822 if (fault_reason >= 0x20 && (fault_reason - 0x20 < 1823 ARRAY_SIZE(irq_remap_fault_reasons))) { 1824 *fault_type = INTR_REMAP; 1825 return irq_remap_fault_reasons[fault_reason - 0x20]; 1826 } else if (fault_reason >= 0x30 && (fault_reason - 0x30 < 1827 ARRAY_SIZE(dma_remap_sm_fault_reasons))) { 1828 *fault_type = DMA_REMAP; 1829 return dma_remap_sm_fault_reasons[fault_reason - 0x30]; 1830 } else if (fault_reason < ARRAY_SIZE(dma_remap_fault_reasons)) { 1831 *fault_type = DMA_REMAP; 1832 return dma_remap_fault_reasons[fault_reason]; 1833 } else { 1834 *fault_type = UNKNOWN; 1835 return "Unknown"; 1836 } 1837 } 1838 1839 1840 static inline int dmar_msi_reg(struct intel_iommu *iommu, int irq) 1841 { 1842 if (iommu->irq == irq) 1843 return DMAR_FECTL_REG; 1844 else if (iommu->pr_irq == irq) 1845 return DMAR_PECTL_REG; 1846 else 1847 BUG(); 1848 } 1849 1850 void dmar_msi_unmask(struct irq_data *data) 1851 { 1852 struct intel_iommu *iommu = irq_data_get_irq_handler_data(data); 1853 int reg = dmar_msi_reg(iommu, data->irq); 1854 unsigned long flag; 1855 1856 /* unmask it */ 1857 raw_spin_lock_irqsave(&iommu->register_lock, flag); 1858 writel(0, iommu->reg + reg); 1859 /* Read a reg to force flush the post write */ 1860 readl(iommu->reg + reg); 1861 raw_spin_unlock_irqrestore(&iommu->register_lock, flag); 1862 } 1863 1864 void dmar_msi_mask(struct irq_data *data) 1865 { 1866 struct intel_iommu *iommu = irq_data_get_irq_handler_data(data); 1867 int reg = dmar_msi_reg(iommu, data->irq); 1868 unsigned long flag; 1869 1870 /* mask it */ 1871 raw_spin_lock_irqsave(&iommu->register_lock, flag); 1872 writel(DMA_FECTL_IM, iommu->reg + reg); 1873 /* Read a reg to force flush the post write */ 1874 readl(iommu->reg + reg); 1875 raw_spin_unlock_irqrestore(&iommu->register_lock, flag); 1876 } 1877 1878 void dmar_msi_write(int irq, struct msi_msg *msg) 1879 { 1880 struct intel_iommu *iommu = irq_get_handler_data(irq); 1881 int reg = dmar_msi_reg(iommu, irq); 1882 unsigned long flag; 1883 1884 raw_spin_lock_irqsave(&iommu->register_lock, flag); 1885 writel(msg->data, iommu->reg + reg + 4); 1886 writel(msg->address_lo, iommu->reg + reg + 8); 1887 writel(msg->address_hi, iommu->reg + reg + 12); 1888 raw_spin_unlock_irqrestore(&iommu->register_lock, flag); 1889 } 1890 1891 void dmar_msi_read(int irq, struct msi_msg *msg) 1892 { 1893 struct intel_iommu *iommu = irq_get_handler_data(irq); 1894 int reg = dmar_msi_reg(iommu, irq); 1895 unsigned long flag; 1896 1897 raw_spin_lock_irqsave(&iommu->register_lock, flag); 1898 msg->data = readl(iommu->reg + reg + 4); 1899 msg->address_lo = readl(iommu->reg + reg + 8); 1900 msg->address_hi = readl(iommu->reg + reg + 12); 1901 raw_spin_unlock_irqrestore(&iommu->register_lock, flag); 1902 } 1903 1904 static int dmar_fault_do_one(struct intel_iommu *iommu, int type, 1905 u8 fault_reason, u32 pasid, u16 source_id, 1906 unsigned long long addr) 1907 { 1908 const char *reason; 1909 int fault_type; 1910 1911 reason = dmar_get_fault_reason(fault_reason, &fault_type); 1912 1913 if (fault_type == INTR_REMAP) 1914 pr_err("[INTR-REMAP] Request device [%02x:%02x.%d] fault index %llx [fault reason %02d] %s\n", 1915 source_id >> 8, PCI_SLOT(source_id & 0xFF), 1916 PCI_FUNC(source_id & 0xFF), addr >> 48, 1917 fault_reason, reason); 1918 else 1919 pr_err("[%s] Request device [%02x:%02x.%d] PASID %x fault addr %llx [fault reason %02d] %s\n", 1920 type ? "DMA Read" : "DMA Write", 1921 source_id >> 8, PCI_SLOT(source_id & 0xFF), 1922 PCI_FUNC(source_id & 0xFF), pasid, addr, 1923 fault_reason, reason); 1924 return 0; 1925 } 1926 1927 #define PRIMARY_FAULT_REG_LEN (16) 1928 irqreturn_t dmar_fault(int irq, void *dev_id) 1929 { 1930 struct intel_iommu *iommu = dev_id; 1931 int reg, fault_index; 1932 u32 fault_status; 1933 unsigned long flag; 1934 static DEFINE_RATELIMIT_STATE(rs, 1935 DEFAULT_RATELIMIT_INTERVAL, 1936 DEFAULT_RATELIMIT_BURST); 1937 1938 raw_spin_lock_irqsave(&iommu->register_lock, flag); 1939 fault_status = readl(iommu->reg + DMAR_FSTS_REG); 1940 if (fault_status && __ratelimit(&rs)) 1941 pr_err("DRHD: handling fault status reg %x\n", fault_status); 1942 1943 /* TBD: ignore advanced fault log currently */ 1944 if (!(fault_status & DMA_FSTS_PPF)) 1945 goto unlock_exit; 1946 1947 fault_index = dma_fsts_fault_record_index(fault_status); 1948 reg = cap_fault_reg_offset(iommu->cap); 1949 while (1) { 1950 /* Disable printing, simply clear the fault when ratelimited */ 1951 bool ratelimited = !__ratelimit(&rs); 1952 u8 fault_reason; 1953 u16 source_id; 1954 u64 guest_addr; 1955 u32 pasid; 1956 int type; 1957 u32 data; 1958 bool pasid_present; 1959 1960 /* highest 32 bits */ 1961 data = readl(iommu->reg + reg + 1962 fault_index * PRIMARY_FAULT_REG_LEN + 12); 1963 if (!(data & DMA_FRCD_F)) 1964 break; 1965 1966 if (!ratelimited) { 1967 fault_reason = dma_frcd_fault_reason(data); 1968 type = dma_frcd_type(data); 1969 1970 pasid = dma_frcd_pasid_value(data); 1971 data = readl(iommu->reg + reg + 1972 fault_index * PRIMARY_FAULT_REG_LEN + 8); 1973 source_id = dma_frcd_source_id(data); 1974 1975 pasid_present = dma_frcd_pasid_present(data); 1976 guest_addr = dmar_readq(iommu->reg + reg + 1977 fault_index * PRIMARY_FAULT_REG_LEN); 1978 guest_addr = dma_frcd_page_addr(guest_addr); 1979 } 1980 1981 /* clear the fault */ 1982 writel(DMA_FRCD_F, iommu->reg + reg + 1983 fault_index * PRIMARY_FAULT_REG_LEN + 12); 1984 1985 raw_spin_unlock_irqrestore(&iommu->register_lock, flag); 1986 1987 if (!ratelimited) 1988 /* Using pasid -1 if pasid is not present */ 1989 dmar_fault_do_one(iommu, type, fault_reason, 1990 pasid_present ? pasid : -1, 1991 source_id, guest_addr); 1992 1993 fault_index++; 1994 if (fault_index >= cap_num_fault_regs(iommu->cap)) 1995 fault_index = 0; 1996 raw_spin_lock_irqsave(&iommu->register_lock, flag); 1997 } 1998 1999 writel(DMA_FSTS_PFO | DMA_FSTS_PPF | DMA_FSTS_PRO, 2000 iommu->reg + DMAR_FSTS_REG); 2001 2002 unlock_exit: 2003 raw_spin_unlock_irqrestore(&iommu->register_lock, flag); 2004 return IRQ_HANDLED; 2005 } 2006 2007 int dmar_set_interrupt(struct intel_iommu *iommu) 2008 { 2009 int irq, ret; 2010 2011 /* 2012 * Check if the fault interrupt is already initialized. 2013 */ 2014 if (iommu->irq) 2015 return 0; 2016 2017 irq = dmar_alloc_hwirq(iommu->seq_id, iommu->node, iommu); 2018 if (irq > 0) { 2019 iommu->irq = irq; 2020 } else { 2021 pr_err("No free IRQ vectors\n"); 2022 return -EINVAL; 2023 } 2024 2025 ret = request_irq(irq, dmar_fault, IRQF_NO_THREAD, iommu->name, iommu); 2026 if (ret) 2027 pr_err("Can't request irq\n"); 2028 return ret; 2029 } 2030 2031 int __init enable_drhd_fault_handling(void) 2032 { 2033 struct dmar_drhd_unit *drhd; 2034 struct intel_iommu *iommu; 2035 2036 /* 2037 * Enable fault control interrupt. 2038 */ 2039 for_each_iommu(iommu, drhd) { 2040 u32 fault_status; 2041 int ret = dmar_set_interrupt(iommu); 2042 2043 if (ret) { 2044 pr_err("DRHD %Lx: failed to enable fault, interrupt, ret %d\n", 2045 (unsigned long long)drhd->reg_base_addr, ret); 2046 return -1; 2047 } 2048 2049 /* 2050 * Clear any previous faults. 2051 */ 2052 dmar_fault(iommu->irq, iommu); 2053 fault_status = readl(iommu->reg + DMAR_FSTS_REG); 2054 writel(fault_status, iommu->reg + DMAR_FSTS_REG); 2055 } 2056 2057 return 0; 2058 } 2059 2060 /* 2061 * Re-enable Queued Invalidation interface. 2062 */ 2063 int dmar_reenable_qi(struct intel_iommu *iommu) 2064 { 2065 if (!ecap_qis(iommu->ecap)) 2066 return -ENOENT; 2067 2068 if (!iommu->qi) 2069 return -ENOENT; 2070 2071 /* 2072 * First disable queued invalidation. 2073 */ 2074 dmar_disable_qi(iommu); 2075 /* 2076 * Then enable queued invalidation again. Since there is no pending 2077 * invalidation requests now, it's safe to re-enable queued 2078 * invalidation. 2079 */ 2080 __dmar_enable_qi(iommu); 2081 2082 return 0; 2083 } 2084 2085 /* 2086 * Check interrupt remapping support in DMAR table description. 2087 */ 2088 int __init dmar_ir_support(void) 2089 { 2090 struct acpi_table_dmar *dmar; 2091 dmar = (struct acpi_table_dmar *)dmar_tbl; 2092 if (!dmar) 2093 return 0; 2094 return dmar->flags & 0x1; 2095 } 2096 2097 /* Check whether DMAR units are in use */ 2098 static inline bool dmar_in_use(void) 2099 { 2100 return irq_remapping_enabled || intel_iommu_enabled; 2101 } 2102 2103 static int __init dmar_free_unused_resources(void) 2104 { 2105 struct dmar_drhd_unit *dmaru, *dmaru_n; 2106 2107 if (dmar_in_use()) 2108 return 0; 2109 2110 if (dmar_dev_scope_status != 1 && !list_empty(&dmar_drhd_units)) 2111 bus_unregister_notifier(&pci_bus_type, &dmar_pci_bus_nb); 2112 2113 down_write(&dmar_global_lock); 2114 list_for_each_entry_safe(dmaru, dmaru_n, &dmar_drhd_units, list) { 2115 list_del(&dmaru->list); 2116 dmar_free_drhd(dmaru); 2117 } 2118 up_write(&dmar_global_lock); 2119 2120 return 0; 2121 } 2122 2123 late_initcall(dmar_free_unused_resources); 2124 IOMMU_INIT_POST(detect_intel_iommu); 2125 2126 /* 2127 * DMAR Hotplug Support 2128 * For more details, please refer to Intel(R) Virtualization Technology 2129 * for Directed-IO Architecture Specifiction, Rev 2.2, Section 8.8 2130 * "Remapping Hardware Unit Hot Plug". 2131 */ 2132 static guid_t dmar_hp_guid = 2133 GUID_INIT(0xD8C1A3A6, 0xBE9B, 0x4C9B, 2134 0x91, 0xBF, 0xC3, 0xCB, 0x81, 0xFC, 0x5D, 0xAF); 2135 2136 /* 2137 * Currently there's only one revision and BIOS will not check the revision id, 2138 * so use 0 for safety. 2139 */ 2140 #define DMAR_DSM_REV_ID 0 2141 #define DMAR_DSM_FUNC_DRHD 1 2142 #define DMAR_DSM_FUNC_ATSR 2 2143 #define DMAR_DSM_FUNC_RHSA 3 2144 #define DMAR_DSM_FUNC_SATC 4 2145 2146 static inline bool dmar_detect_dsm(acpi_handle handle, int func) 2147 { 2148 return acpi_check_dsm(handle, &dmar_hp_guid, DMAR_DSM_REV_ID, 1 << func); 2149 } 2150 2151 static int dmar_walk_dsm_resource(acpi_handle handle, int func, 2152 dmar_res_handler_t handler, void *arg) 2153 { 2154 int ret = -ENODEV; 2155 union acpi_object *obj; 2156 struct acpi_dmar_header *start; 2157 struct dmar_res_callback callback; 2158 static int res_type[] = { 2159 [DMAR_DSM_FUNC_DRHD] = ACPI_DMAR_TYPE_HARDWARE_UNIT, 2160 [DMAR_DSM_FUNC_ATSR] = ACPI_DMAR_TYPE_ROOT_ATS, 2161 [DMAR_DSM_FUNC_RHSA] = ACPI_DMAR_TYPE_HARDWARE_AFFINITY, 2162 [DMAR_DSM_FUNC_SATC] = ACPI_DMAR_TYPE_SATC, 2163 }; 2164 2165 if (!dmar_detect_dsm(handle, func)) 2166 return 0; 2167 2168 obj = acpi_evaluate_dsm_typed(handle, &dmar_hp_guid, DMAR_DSM_REV_ID, 2169 func, NULL, ACPI_TYPE_BUFFER); 2170 if (!obj) 2171 return -ENODEV; 2172 2173 memset(&callback, 0, sizeof(callback)); 2174 callback.cb[res_type[func]] = handler; 2175 callback.arg[res_type[func]] = arg; 2176 start = (struct acpi_dmar_header *)obj->buffer.pointer; 2177 ret = dmar_walk_remapping_entries(start, obj->buffer.length, &callback); 2178 2179 ACPI_FREE(obj); 2180 2181 return ret; 2182 } 2183 2184 static int dmar_hp_add_drhd(struct acpi_dmar_header *header, void *arg) 2185 { 2186 int ret; 2187 struct dmar_drhd_unit *dmaru; 2188 2189 dmaru = dmar_find_dmaru((struct acpi_dmar_hardware_unit *)header); 2190 if (!dmaru) 2191 return -ENODEV; 2192 2193 ret = dmar_ir_hotplug(dmaru, true); 2194 if (ret == 0) 2195 ret = dmar_iommu_hotplug(dmaru, true); 2196 2197 return ret; 2198 } 2199 2200 static int dmar_hp_remove_drhd(struct acpi_dmar_header *header, void *arg) 2201 { 2202 int i, ret; 2203 struct device *dev; 2204 struct dmar_drhd_unit *dmaru; 2205 2206 dmaru = dmar_find_dmaru((struct acpi_dmar_hardware_unit *)header); 2207 if (!dmaru) 2208 return 0; 2209 2210 /* 2211 * All PCI devices managed by this unit should have been destroyed. 2212 */ 2213 if (!dmaru->include_all && dmaru->devices && dmaru->devices_cnt) { 2214 for_each_active_dev_scope(dmaru->devices, 2215 dmaru->devices_cnt, i, dev) 2216 return -EBUSY; 2217 } 2218 2219 ret = dmar_ir_hotplug(dmaru, false); 2220 if (ret == 0) 2221 ret = dmar_iommu_hotplug(dmaru, false); 2222 2223 return ret; 2224 } 2225 2226 static int dmar_hp_release_drhd(struct acpi_dmar_header *header, void *arg) 2227 { 2228 struct dmar_drhd_unit *dmaru; 2229 2230 dmaru = dmar_find_dmaru((struct acpi_dmar_hardware_unit *)header); 2231 if (dmaru) { 2232 list_del_rcu(&dmaru->list); 2233 synchronize_rcu(); 2234 dmar_free_drhd(dmaru); 2235 } 2236 2237 return 0; 2238 } 2239 2240 static int dmar_hotplug_insert(acpi_handle handle) 2241 { 2242 int ret; 2243 int drhd_count = 0; 2244 2245 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD, 2246 &dmar_validate_one_drhd, (void *)1); 2247 if (ret) 2248 goto out; 2249 2250 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD, 2251 &dmar_parse_one_drhd, (void *)&drhd_count); 2252 if (ret == 0 && drhd_count == 0) { 2253 pr_warn(FW_BUG "No DRHD structures in buffer returned by _DSM method\n"); 2254 goto out; 2255 } else if (ret) { 2256 goto release_drhd; 2257 } 2258 2259 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_RHSA, 2260 &dmar_parse_one_rhsa, NULL); 2261 if (ret) 2262 goto release_drhd; 2263 2264 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR, 2265 &dmar_parse_one_atsr, NULL); 2266 if (ret) 2267 goto release_atsr; 2268 2269 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD, 2270 &dmar_hp_add_drhd, NULL); 2271 if (!ret) 2272 return 0; 2273 2274 dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD, 2275 &dmar_hp_remove_drhd, NULL); 2276 release_atsr: 2277 dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR, 2278 &dmar_release_one_atsr, NULL); 2279 release_drhd: 2280 dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD, 2281 &dmar_hp_release_drhd, NULL); 2282 out: 2283 return ret; 2284 } 2285 2286 static int dmar_hotplug_remove(acpi_handle handle) 2287 { 2288 int ret; 2289 2290 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR, 2291 &dmar_check_one_atsr, NULL); 2292 if (ret) 2293 return ret; 2294 2295 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD, 2296 &dmar_hp_remove_drhd, NULL); 2297 if (ret == 0) { 2298 WARN_ON(dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR, 2299 &dmar_release_one_atsr, NULL)); 2300 WARN_ON(dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD, 2301 &dmar_hp_release_drhd, NULL)); 2302 } else { 2303 dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD, 2304 &dmar_hp_add_drhd, NULL); 2305 } 2306 2307 return ret; 2308 } 2309 2310 static acpi_status dmar_get_dsm_handle(acpi_handle handle, u32 lvl, 2311 void *context, void **retval) 2312 { 2313 acpi_handle *phdl = retval; 2314 2315 if (dmar_detect_dsm(handle, DMAR_DSM_FUNC_DRHD)) { 2316 *phdl = handle; 2317 return AE_CTRL_TERMINATE; 2318 } 2319 2320 return AE_OK; 2321 } 2322 2323 static int dmar_device_hotplug(acpi_handle handle, bool insert) 2324 { 2325 int ret; 2326 acpi_handle tmp = NULL; 2327 acpi_status status; 2328 2329 if (!dmar_in_use()) 2330 return 0; 2331 2332 if (dmar_detect_dsm(handle, DMAR_DSM_FUNC_DRHD)) { 2333 tmp = handle; 2334 } else { 2335 status = acpi_walk_namespace(ACPI_TYPE_DEVICE, handle, 2336 ACPI_UINT32_MAX, 2337 dmar_get_dsm_handle, 2338 NULL, NULL, &tmp); 2339 if (ACPI_FAILURE(status)) { 2340 pr_warn("Failed to locate _DSM method.\n"); 2341 return -ENXIO; 2342 } 2343 } 2344 if (tmp == NULL) 2345 return 0; 2346 2347 down_write(&dmar_global_lock); 2348 if (insert) 2349 ret = dmar_hotplug_insert(tmp); 2350 else 2351 ret = dmar_hotplug_remove(tmp); 2352 up_write(&dmar_global_lock); 2353 2354 return ret; 2355 } 2356 2357 int dmar_device_add(acpi_handle handle) 2358 { 2359 return dmar_device_hotplug(handle, true); 2360 } 2361 2362 int dmar_device_remove(acpi_handle handle) 2363 { 2364 return dmar_device_hotplug(handle, false); 2365 } 2366 2367 /* 2368 * dmar_platform_optin - Is %DMA_CTRL_PLATFORM_OPT_IN_FLAG set in DMAR table 2369 * 2370 * Returns true if the platform has %DMA_CTRL_PLATFORM_OPT_IN_FLAG set in 2371 * the ACPI DMAR table. This means that the platform boot firmware has made 2372 * sure no device can issue DMA outside of RMRR regions. 2373 */ 2374 bool dmar_platform_optin(void) 2375 { 2376 struct acpi_table_dmar *dmar; 2377 acpi_status status; 2378 bool ret; 2379 2380 status = acpi_get_table(ACPI_SIG_DMAR, 0, 2381 (struct acpi_table_header **)&dmar); 2382 if (ACPI_FAILURE(status)) 2383 return false; 2384 2385 ret = !!(dmar->flags & DMAR_PLATFORM_OPT_IN); 2386 acpi_put_table((struct acpi_table_header *)dmar); 2387 2388 return ret; 2389 } 2390 EXPORT_SYMBOL_GPL(dmar_platform_optin); 2391