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 <trace/events/intel_iommu.h> 34 35 #include "../irq_remapping.h" 36 #include "perf.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 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 /* 132 * Ignore devices that have a domain number higher than what can 133 * be looked up in DMAR, e.g. VMD subdevices with domain 0x10000 134 */ 135 if (pci_domain_nr(dev->bus) > U16_MAX) 136 return NULL; 137 138 /* Only generate path[] for device addition event */ 139 if (event == BUS_NOTIFY_ADD_DEVICE) 140 for (tmp = dev; tmp; tmp = tmp->bus->self) 141 level++; 142 143 size = struct_size(info, path, level); 144 if (size <= sizeof(dmar_pci_notify_info_buf)) { 145 info = (struct dmar_pci_notify_info *)dmar_pci_notify_info_buf; 146 } else { 147 info = kzalloc(size, GFP_KERNEL); 148 if (!info) { 149 if (dmar_dev_scope_status == 0) 150 dmar_dev_scope_status = -ENOMEM; 151 return NULL; 152 } 153 } 154 155 info->event = event; 156 info->dev = dev; 157 info->seg = pci_domain_nr(dev->bus); 158 info->level = level; 159 if (event == BUS_NOTIFY_ADD_DEVICE) { 160 for (tmp = dev; tmp; tmp = tmp->bus->self) { 161 level--; 162 info->path[level].bus = tmp->bus->number; 163 info->path[level].device = PCI_SLOT(tmp->devfn); 164 info->path[level].function = PCI_FUNC(tmp->devfn); 165 if (pci_is_root_bus(tmp->bus)) 166 info->bus = tmp->bus->number; 167 } 168 } 169 170 return info; 171 } 172 173 static inline void dmar_free_pci_notify_info(struct dmar_pci_notify_info *info) 174 { 175 if ((void *)info != dmar_pci_notify_info_buf) 176 kfree(info); 177 } 178 179 static bool dmar_match_pci_path(struct dmar_pci_notify_info *info, int bus, 180 struct acpi_dmar_pci_path *path, int count) 181 { 182 int i; 183 184 if (info->bus != bus) 185 goto fallback; 186 if (info->level != count) 187 goto fallback; 188 189 for (i = 0; i < count; i++) { 190 if (path[i].device != info->path[i].device || 191 path[i].function != info->path[i].function) 192 goto fallback; 193 } 194 195 return true; 196 197 fallback: 198 199 if (count != 1) 200 return false; 201 202 i = info->level - 1; 203 if (bus == info->path[i].bus && 204 path[0].device == info->path[i].device && 205 path[0].function == info->path[i].function) { 206 pr_info(FW_BUG "RMRR entry for device %02x:%02x.%x is broken - applying workaround\n", 207 bus, path[0].device, path[0].function); 208 return true; 209 } 210 211 return false; 212 } 213 214 /* Return: > 0 if match found, 0 if no match found, < 0 if error happens */ 215 int dmar_insert_dev_scope(struct dmar_pci_notify_info *info, 216 void *start, void*end, u16 segment, 217 struct dmar_dev_scope *devices, 218 int devices_cnt) 219 { 220 int i, level; 221 struct device *tmp, *dev = &info->dev->dev; 222 struct acpi_dmar_device_scope *scope; 223 struct acpi_dmar_pci_path *path; 224 225 if (segment != info->seg) 226 return 0; 227 228 for (; start < end; start += scope->length) { 229 scope = start; 230 if (scope->entry_type != ACPI_DMAR_SCOPE_TYPE_ENDPOINT && 231 scope->entry_type != ACPI_DMAR_SCOPE_TYPE_BRIDGE) 232 continue; 233 234 path = (struct acpi_dmar_pci_path *)(scope + 1); 235 level = (scope->length - sizeof(*scope)) / sizeof(*path); 236 if (!dmar_match_pci_path(info, scope->bus, path, level)) 237 continue; 238 239 /* 240 * We expect devices with endpoint scope to have normal PCI 241 * headers, and devices with bridge scope to have bridge PCI 242 * headers. However PCI NTB devices may be listed in the 243 * DMAR table with bridge scope, even though they have a 244 * normal PCI header. NTB devices are identified by class 245 * "BRIDGE_OTHER" (0680h) - we don't declare a socpe mismatch 246 * for this special case. 247 */ 248 if ((scope->entry_type == ACPI_DMAR_SCOPE_TYPE_ENDPOINT && 249 info->dev->hdr_type != PCI_HEADER_TYPE_NORMAL) || 250 (scope->entry_type == ACPI_DMAR_SCOPE_TYPE_BRIDGE && 251 (info->dev->hdr_type == PCI_HEADER_TYPE_NORMAL && 252 info->dev->class >> 16 != PCI_BASE_CLASS_BRIDGE))) { 253 pr_warn("Device scope type does not match for %s\n", 254 pci_name(info->dev)); 255 return -EINVAL; 256 } 257 258 for_each_dev_scope(devices, devices_cnt, i, tmp) 259 if (tmp == NULL) { 260 devices[i].bus = info->dev->bus->number; 261 devices[i].devfn = info->dev->devfn; 262 rcu_assign_pointer(devices[i].dev, 263 get_device(dev)); 264 return 1; 265 } 266 BUG_ON(i >= devices_cnt); 267 } 268 269 return 0; 270 } 271 272 int dmar_remove_dev_scope(struct dmar_pci_notify_info *info, u16 segment, 273 struct dmar_dev_scope *devices, int count) 274 { 275 int index; 276 struct device *tmp; 277 278 if (info->seg != segment) 279 return 0; 280 281 for_each_active_dev_scope(devices, count, index, tmp) 282 if (tmp == &info->dev->dev) { 283 RCU_INIT_POINTER(devices[index].dev, NULL); 284 synchronize_rcu(); 285 put_device(tmp); 286 return 1; 287 } 288 289 return 0; 290 } 291 292 static int dmar_pci_bus_add_dev(struct dmar_pci_notify_info *info) 293 { 294 int ret = 0; 295 struct dmar_drhd_unit *dmaru; 296 struct acpi_dmar_hardware_unit *drhd; 297 298 for_each_drhd_unit(dmaru) { 299 if (dmaru->include_all) 300 continue; 301 302 drhd = container_of(dmaru->hdr, 303 struct acpi_dmar_hardware_unit, header); 304 ret = dmar_insert_dev_scope(info, (void *)(drhd + 1), 305 ((void *)drhd) + drhd->header.length, 306 dmaru->segment, 307 dmaru->devices, dmaru->devices_cnt); 308 if (ret) 309 break; 310 } 311 if (ret >= 0) 312 ret = dmar_iommu_notify_scope_dev(info); 313 if (ret < 0 && dmar_dev_scope_status == 0) 314 dmar_dev_scope_status = ret; 315 316 if (ret >= 0) 317 intel_irq_remap_add_device(info); 318 319 return ret; 320 } 321 322 static void dmar_pci_bus_del_dev(struct dmar_pci_notify_info *info) 323 { 324 struct dmar_drhd_unit *dmaru; 325 326 for_each_drhd_unit(dmaru) 327 if (dmar_remove_dev_scope(info, dmaru->segment, 328 dmaru->devices, dmaru->devices_cnt)) 329 break; 330 dmar_iommu_notify_scope_dev(info); 331 } 332 333 static inline void vf_inherit_msi_domain(struct pci_dev *pdev) 334 { 335 struct pci_dev *physfn = pci_physfn(pdev); 336 337 dev_set_msi_domain(&pdev->dev, dev_get_msi_domain(&physfn->dev)); 338 } 339 340 static int dmar_pci_bus_notifier(struct notifier_block *nb, 341 unsigned long action, void *data) 342 { 343 struct pci_dev *pdev = to_pci_dev(data); 344 struct dmar_pci_notify_info *info; 345 346 /* Only care about add/remove events for physical functions. 347 * For VFs we actually do the lookup based on the corresponding 348 * PF in device_to_iommu() anyway. */ 349 if (pdev->is_virtfn) { 350 /* 351 * Ensure that the VF device inherits the irq domain of the 352 * PF device. Ideally the device would inherit the domain 353 * from the bus, but DMAR can have multiple units per bus 354 * which makes this impossible. The VF 'bus' could inherit 355 * from the PF device, but that's yet another x86'sism to 356 * inflict on everybody else. 357 */ 358 if (action == BUS_NOTIFY_ADD_DEVICE) 359 vf_inherit_msi_domain(pdev); 360 return NOTIFY_DONE; 361 } 362 363 if (action != BUS_NOTIFY_ADD_DEVICE && 364 action != BUS_NOTIFY_REMOVED_DEVICE) 365 return NOTIFY_DONE; 366 367 info = dmar_alloc_pci_notify_info(pdev, action); 368 if (!info) 369 return NOTIFY_DONE; 370 371 down_write(&dmar_global_lock); 372 if (action == BUS_NOTIFY_ADD_DEVICE) 373 dmar_pci_bus_add_dev(info); 374 else if (action == BUS_NOTIFY_REMOVED_DEVICE) 375 dmar_pci_bus_del_dev(info); 376 up_write(&dmar_global_lock); 377 378 dmar_free_pci_notify_info(info); 379 380 return NOTIFY_OK; 381 } 382 383 static struct notifier_block dmar_pci_bus_nb = { 384 .notifier_call = dmar_pci_bus_notifier, 385 .priority = INT_MIN, 386 }; 387 388 static struct dmar_drhd_unit * 389 dmar_find_dmaru(struct acpi_dmar_hardware_unit *drhd) 390 { 391 struct dmar_drhd_unit *dmaru; 392 393 list_for_each_entry_rcu(dmaru, &dmar_drhd_units, list, 394 dmar_rcu_check()) 395 if (dmaru->segment == drhd->segment && 396 dmaru->reg_base_addr == drhd->address) 397 return dmaru; 398 399 return NULL; 400 } 401 402 /* 403 * dmar_parse_one_drhd - parses exactly one DMA remapping hardware definition 404 * structure which uniquely represent one DMA remapping hardware unit 405 * present in the platform 406 */ 407 static int dmar_parse_one_drhd(struct acpi_dmar_header *header, void *arg) 408 { 409 struct acpi_dmar_hardware_unit *drhd; 410 struct dmar_drhd_unit *dmaru; 411 int ret; 412 413 drhd = (struct acpi_dmar_hardware_unit *)header; 414 dmaru = dmar_find_dmaru(drhd); 415 if (dmaru) 416 goto out; 417 418 dmaru = kzalloc(sizeof(*dmaru) + header->length, GFP_KERNEL); 419 if (!dmaru) 420 return -ENOMEM; 421 422 /* 423 * If header is allocated from slab by ACPI _DSM method, we need to 424 * copy the content because the memory buffer will be freed on return. 425 */ 426 dmaru->hdr = (void *)(dmaru + 1); 427 memcpy(dmaru->hdr, header, header->length); 428 dmaru->reg_base_addr = drhd->address; 429 dmaru->segment = drhd->segment; 430 dmaru->include_all = drhd->flags & 0x1; /* BIT0: INCLUDE_ALL */ 431 dmaru->devices = dmar_alloc_dev_scope((void *)(drhd + 1), 432 ((void *)drhd) + drhd->header.length, 433 &dmaru->devices_cnt); 434 if (dmaru->devices_cnt && dmaru->devices == NULL) { 435 kfree(dmaru); 436 return -ENOMEM; 437 } 438 439 ret = alloc_iommu(dmaru); 440 if (ret) { 441 dmar_free_dev_scope(&dmaru->devices, 442 &dmaru->devices_cnt); 443 kfree(dmaru); 444 return ret; 445 } 446 dmar_register_drhd_unit(dmaru); 447 448 out: 449 if (arg) 450 (*(int *)arg)++; 451 452 return 0; 453 } 454 455 static void dmar_free_drhd(struct dmar_drhd_unit *dmaru) 456 { 457 if (dmaru->devices && dmaru->devices_cnt) 458 dmar_free_dev_scope(&dmaru->devices, &dmaru->devices_cnt); 459 if (dmaru->iommu) 460 free_iommu(dmaru->iommu); 461 kfree(dmaru); 462 } 463 464 static int __init dmar_parse_one_andd(struct acpi_dmar_header *header, 465 void *arg) 466 { 467 struct acpi_dmar_andd *andd = (void *)header; 468 469 /* Check for NUL termination within the designated length */ 470 if (strnlen(andd->device_name, header->length - 8) == header->length - 8) { 471 pr_warn(FW_BUG 472 "Your BIOS is broken; ANDD object name is not NUL-terminated\n" 473 "BIOS vendor: %s; Ver: %s; Product Version: %s\n", 474 dmi_get_system_info(DMI_BIOS_VENDOR), 475 dmi_get_system_info(DMI_BIOS_VERSION), 476 dmi_get_system_info(DMI_PRODUCT_VERSION)); 477 add_taint(TAINT_FIRMWARE_WORKAROUND, LOCKDEP_STILL_OK); 478 return -EINVAL; 479 } 480 pr_info("ANDD device: %x name: %s\n", andd->device_number, 481 andd->device_name); 482 483 return 0; 484 } 485 486 #ifdef CONFIG_ACPI_NUMA 487 static int dmar_parse_one_rhsa(struct acpi_dmar_header *header, void *arg) 488 { 489 struct acpi_dmar_rhsa *rhsa; 490 struct dmar_drhd_unit *drhd; 491 492 rhsa = (struct acpi_dmar_rhsa *)header; 493 for_each_drhd_unit(drhd) { 494 if (drhd->reg_base_addr == rhsa->base_address) { 495 int node = pxm_to_node(rhsa->proximity_domain); 496 497 if (!node_online(node)) 498 node = NUMA_NO_NODE; 499 drhd->iommu->node = node; 500 return 0; 501 } 502 } 503 pr_warn(FW_BUG 504 "Your BIOS is broken; RHSA refers to non-existent DMAR unit at %llx\n" 505 "BIOS vendor: %s; Ver: %s; Product Version: %s\n", 506 rhsa->base_address, 507 dmi_get_system_info(DMI_BIOS_VENDOR), 508 dmi_get_system_info(DMI_BIOS_VERSION), 509 dmi_get_system_info(DMI_PRODUCT_VERSION)); 510 add_taint(TAINT_FIRMWARE_WORKAROUND, LOCKDEP_STILL_OK); 511 512 return 0; 513 } 514 #else 515 #define dmar_parse_one_rhsa dmar_res_noop 516 #endif 517 518 static void 519 dmar_table_print_dmar_entry(struct acpi_dmar_header *header) 520 { 521 struct acpi_dmar_hardware_unit *drhd; 522 struct acpi_dmar_reserved_memory *rmrr; 523 struct acpi_dmar_atsr *atsr; 524 struct acpi_dmar_rhsa *rhsa; 525 struct acpi_dmar_satc *satc; 526 527 switch (header->type) { 528 case ACPI_DMAR_TYPE_HARDWARE_UNIT: 529 drhd = container_of(header, struct acpi_dmar_hardware_unit, 530 header); 531 pr_info("DRHD base: %#016Lx flags: %#x\n", 532 (unsigned long long)drhd->address, drhd->flags); 533 break; 534 case ACPI_DMAR_TYPE_RESERVED_MEMORY: 535 rmrr = container_of(header, struct acpi_dmar_reserved_memory, 536 header); 537 pr_info("RMRR base: %#016Lx end: %#016Lx\n", 538 (unsigned long long)rmrr->base_address, 539 (unsigned long long)rmrr->end_address); 540 break; 541 case ACPI_DMAR_TYPE_ROOT_ATS: 542 atsr = container_of(header, struct acpi_dmar_atsr, header); 543 pr_info("ATSR flags: %#x\n", atsr->flags); 544 break; 545 case ACPI_DMAR_TYPE_HARDWARE_AFFINITY: 546 rhsa = container_of(header, struct acpi_dmar_rhsa, header); 547 pr_info("RHSA base: %#016Lx proximity domain: %#x\n", 548 (unsigned long long)rhsa->base_address, 549 rhsa->proximity_domain); 550 break; 551 case ACPI_DMAR_TYPE_NAMESPACE: 552 /* We don't print this here because we need to sanity-check 553 it first. So print it in dmar_parse_one_andd() instead. */ 554 break; 555 case ACPI_DMAR_TYPE_SATC: 556 satc = container_of(header, struct acpi_dmar_satc, header); 557 pr_info("SATC flags: 0x%x\n", satc->flags); 558 break; 559 } 560 } 561 562 /** 563 * dmar_table_detect - checks to see if the platform supports DMAR devices 564 */ 565 static int __init dmar_table_detect(void) 566 { 567 acpi_status status = AE_OK; 568 569 /* if we could find DMAR table, then there are DMAR devices */ 570 status = acpi_get_table(ACPI_SIG_DMAR, 0, &dmar_tbl); 571 572 if (ACPI_SUCCESS(status) && !dmar_tbl) { 573 pr_warn("Unable to map DMAR\n"); 574 status = AE_NOT_FOUND; 575 } 576 577 return ACPI_SUCCESS(status) ? 0 : -ENOENT; 578 } 579 580 static int dmar_walk_remapping_entries(struct acpi_dmar_header *start, 581 size_t len, struct dmar_res_callback *cb) 582 { 583 struct acpi_dmar_header *iter, *next; 584 struct acpi_dmar_header *end = ((void *)start) + len; 585 586 for (iter = start; iter < end; iter = next) { 587 next = (void *)iter + iter->length; 588 if (iter->length == 0) { 589 /* Avoid looping forever on bad ACPI tables */ 590 pr_debug(FW_BUG "Invalid 0-length structure\n"); 591 break; 592 } else if (next > end) { 593 /* Avoid passing table end */ 594 pr_warn(FW_BUG "Record passes table end\n"); 595 return -EINVAL; 596 } 597 598 if (cb->print_entry) 599 dmar_table_print_dmar_entry(iter); 600 601 if (iter->type >= ACPI_DMAR_TYPE_RESERVED) { 602 /* continue for forward compatibility */ 603 pr_debug("Unknown DMAR structure type %d\n", 604 iter->type); 605 } else if (cb->cb[iter->type]) { 606 int ret; 607 608 ret = cb->cb[iter->type](iter, cb->arg[iter->type]); 609 if (ret) 610 return ret; 611 } else if (!cb->ignore_unhandled) { 612 pr_warn("No handler for DMAR structure type %d\n", 613 iter->type); 614 return -EINVAL; 615 } 616 } 617 618 return 0; 619 } 620 621 static inline int dmar_walk_dmar_table(struct acpi_table_dmar *dmar, 622 struct dmar_res_callback *cb) 623 { 624 return dmar_walk_remapping_entries((void *)(dmar + 1), 625 dmar->header.length - sizeof(*dmar), cb); 626 } 627 628 /** 629 * parse_dmar_table - parses the DMA reporting table 630 */ 631 static int __init 632 parse_dmar_table(void) 633 { 634 struct acpi_table_dmar *dmar; 635 int drhd_count = 0; 636 int ret; 637 struct dmar_res_callback cb = { 638 .print_entry = true, 639 .ignore_unhandled = true, 640 .arg[ACPI_DMAR_TYPE_HARDWARE_UNIT] = &drhd_count, 641 .cb[ACPI_DMAR_TYPE_HARDWARE_UNIT] = &dmar_parse_one_drhd, 642 .cb[ACPI_DMAR_TYPE_RESERVED_MEMORY] = &dmar_parse_one_rmrr, 643 .cb[ACPI_DMAR_TYPE_ROOT_ATS] = &dmar_parse_one_atsr, 644 .cb[ACPI_DMAR_TYPE_HARDWARE_AFFINITY] = &dmar_parse_one_rhsa, 645 .cb[ACPI_DMAR_TYPE_NAMESPACE] = &dmar_parse_one_andd, 646 .cb[ACPI_DMAR_TYPE_SATC] = &dmar_parse_one_satc, 647 }; 648 649 /* 650 * Do it again, earlier dmar_tbl mapping could be mapped with 651 * fixed map. 652 */ 653 dmar_table_detect(); 654 655 /* 656 * ACPI tables may not be DMA protected by tboot, so use DMAR copy 657 * SINIT saved in SinitMleData in TXT heap (which is DMA protected) 658 */ 659 dmar_tbl = tboot_get_dmar_table(dmar_tbl); 660 661 dmar = (struct acpi_table_dmar *)dmar_tbl; 662 if (!dmar) 663 return -ENODEV; 664 665 if (dmar->width < PAGE_SHIFT - 1) { 666 pr_warn("Invalid DMAR haw\n"); 667 return -EINVAL; 668 } 669 670 pr_info("Host address width %d\n", dmar->width + 1); 671 ret = dmar_walk_dmar_table(dmar, &cb); 672 if (ret == 0 && drhd_count == 0) 673 pr_warn(FW_BUG "No DRHD structure found in DMAR table\n"); 674 675 return ret; 676 } 677 678 static int dmar_pci_device_match(struct dmar_dev_scope devices[], 679 int cnt, struct pci_dev *dev) 680 { 681 int index; 682 struct device *tmp; 683 684 while (dev) { 685 for_each_active_dev_scope(devices, cnt, index, tmp) 686 if (dev_is_pci(tmp) && dev == to_pci_dev(tmp)) 687 return 1; 688 689 /* Check our parent */ 690 dev = dev->bus->self; 691 } 692 693 return 0; 694 } 695 696 struct dmar_drhd_unit * 697 dmar_find_matched_drhd_unit(struct pci_dev *dev) 698 { 699 struct dmar_drhd_unit *dmaru; 700 struct acpi_dmar_hardware_unit *drhd; 701 702 dev = pci_physfn(dev); 703 704 rcu_read_lock(); 705 for_each_drhd_unit(dmaru) { 706 drhd = container_of(dmaru->hdr, 707 struct acpi_dmar_hardware_unit, 708 header); 709 710 if (dmaru->include_all && 711 drhd->segment == pci_domain_nr(dev->bus)) 712 goto out; 713 714 if (dmar_pci_device_match(dmaru->devices, 715 dmaru->devices_cnt, dev)) 716 goto out; 717 } 718 dmaru = NULL; 719 out: 720 rcu_read_unlock(); 721 722 return dmaru; 723 } 724 725 static void __init dmar_acpi_insert_dev_scope(u8 device_number, 726 struct acpi_device *adev) 727 { 728 struct dmar_drhd_unit *dmaru; 729 struct acpi_dmar_hardware_unit *drhd; 730 struct acpi_dmar_device_scope *scope; 731 struct device *tmp; 732 int i; 733 struct acpi_dmar_pci_path *path; 734 735 for_each_drhd_unit(dmaru) { 736 drhd = container_of(dmaru->hdr, 737 struct acpi_dmar_hardware_unit, 738 header); 739 740 for (scope = (void *)(drhd + 1); 741 (unsigned long)scope < ((unsigned long)drhd) + drhd->header.length; 742 scope = ((void *)scope) + scope->length) { 743 if (scope->entry_type != ACPI_DMAR_SCOPE_TYPE_NAMESPACE) 744 continue; 745 if (scope->enumeration_id != device_number) 746 continue; 747 748 path = (void *)(scope + 1); 749 pr_info("ACPI device \"%s\" under DMAR at %llx as %02x:%02x.%d\n", 750 dev_name(&adev->dev), dmaru->reg_base_addr, 751 scope->bus, path->device, path->function); 752 for_each_dev_scope(dmaru->devices, dmaru->devices_cnt, i, tmp) 753 if (tmp == NULL) { 754 dmaru->devices[i].bus = scope->bus; 755 dmaru->devices[i].devfn = PCI_DEVFN(path->device, 756 path->function); 757 rcu_assign_pointer(dmaru->devices[i].dev, 758 get_device(&adev->dev)); 759 return; 760 } 761 BUG_ON(i >= dmaru->devices_cnt); 762 } 763 } 764 pr_warn("No IOMMU scope found for ANDD enumeration ID %d (%s)\n", 765 device_number, dev_name(&adev->dev)); 766 } 767 768 static int __init dmar_acpi_dev_scope_init(void) 769 { 770 struct acpi_dmar_andd *andd; 771 772 if (dmar_tbl == NULL) 773 return -ENODEV; 774 775 for (andd = (void *)dmar_tbl + sizeof(struct acpi_table_dmar); 776 ((unsigned long)andd) < ((unsigned long)dmar_tbl) + dmar_tbl->length; 777 andd = ((void *)andd) + andd->header.length) { 778 if (andd->header.type == ACPI_DMAR_TYPE_NAMESPACE) { 779 acpi_handle h; 780 struct acpi_device *adev; 781 782 if (!ACPI_SUCCESS(acpi_get_handle(ACPI_ROOT_OBJECT, 783 andd->device_name, 784 &h))) { 785 pr_err("Failed to find handle for ACPI object %s\n", 786 andd->device_name); 787 continue; 788 } 789 adev = acpi_fetch_acpi_dev(h); 790 if (!adev) { 791 pr_err("Failed to get device for ACPI object %s\n", 792 andd->device_name); 793 continue; 794 } 795 dmar_acpi_insert_dev_scope(andd->device_number, adev); 796 } 797 } 798 return 0; 799 } 800 801 int __init dmar_dev_scope_init(void) 802 { 803 struct pci_dev *dev = NULL; 804 struct dmar_pci_notify_info *info; 805 806 if (dmar_dev_scope_status != 1) 807 return dmar_dev_scope_status; 808 809 if (list_empty(&dmar_drhd_units)) { 810 dmar_dev_scope_status = -ENODEV; 811 } else { 812 dmar_dev_scope_status = 0; 813 814 dmar_acpi_dev_scope_init(); 815 816 for_each_pci_dev(dev) { 817 if (dev->is_virtfn) 818 continue; 819 820 info = dmar_alloc_pci_notify_info(dev, 821 BUS_NOTIFY_ADD_DEVICE); 822 if (!info) { 823 return dmar_dev_scope_status; 824 } else { 825 dmar_pci_bus_add_dev(info); 826 dmar_free_pci_notify_info(info); 827 } 828 } 829 } 830 831 return dmar_dev_scope_status; 832 } 833 834 void __init dmar_register_bus_notifier(void) 835 { 836 bus_register_notifier(&pci_bus_type, &dmar_pci_bus_nb); 837 } 838 839 840 int __init dmar_table_init(void) 841 { 842 static int dmar_table_initialized; 843 int ret; 844 845 if (dmar_table_initialized == 0) { 846 ret = parse_dmar_table(); 847 if (ret < 0) { 848 if (ret != -ENODEV) 849 pr_info("Parse DMAR table failure.\n"); 850 } else if (list_empty(&dmar_drhd_units)) { 851 pr_info("No DMAR devices found\n"); 852 ret = -ENODEV; 853 } 854 855 if (ret < 0) 856 dmar_table_initialized = ret; 857 else 858 dmar_table_initialized = 1; 859 } 860 861 return dmar_table_initialized < 0 ? dmar_table_initialized : 0; 862 } 863 864 static void warn_invalid_dmar(u64 addr, const char *message) 865 { 866 pr_warn_once(FW_BUG 867 "Your BIOS is broken; DMAR reported at address %llx%s!\n" 868 "BIOS vendor: %s; Ver: %s; Product Version: %s\n", 869 addr, message, 870 dmi_get_system_info(DMI_BIOS_VENDOR), 871 dmi_get_system_info(DMI_BIOS_VERSION), 872 dmi_get_system_info(DMI_PRODUCT_VERSION)); 873 add_taint(TAINT_FIRMWARE_WORKAROUND, LOCKDEP_STILL_OK); 874 } 875 876 static int __ref 877 dmar_validate_one_drhd(struct acpi_dmar_header *entry, void *arg) 878 { 879 struct acpi_dmar_hardware_unit *drhd; 880 void __iomem *addr; 881 u64 cap, ecap; 882 883 drhd = (void *)entry; 884 if (!drhd->address) { 885 warn_invalid_dmar(0, ""); 886 return -EINVAL; 887 } 888 889 if (arg) 890 addr = ioremap(drhd->address, VTD_PAGE_SIZE); 891 else 892 addr = early_ioremap(drhd->address, VTD_PAGE_SIZE); 893 if (!addr) { 894 pr_warn("Can't validate DRHD address: %llx\n", drhd->address); 895 return -EINVAL; 896 } 897 898 cap = dmar_readq(addr + DMAR_CAP_REG); 899 ecap = dmar_readq(addr + DMAR_ECAP_REG); 900 901 if (arg) 902 iounmap(addr); 903 else 904 early_iounmap(addr, VTD_PAGE_SIZE); 905 906 if (cap == (uint64_t)-1 && ecap == (uint64_t)-1) { 907 warn_invalid_dmar(drhd->address, " returns all ones"); 908 return -EINVAL; 909 } 910 911 return 0; 912 } 913 914 void __init detect_intel_iommu(void) 915 { 916 int ret; 917 struct dmar_res_callback validate_drhd_cb = { 918 .cb[ACPI_DMAR_TYPE_HARDWARE_UNIT] = &dmar_validate_one_drhd, 919 .ignore_unhandled = true, 920 }; 921 922 down_write(&dmar_global_lock); 923 ret = dmar_table_detect(); 924 if (!ret) 925 ret = dmar_walk_dmar_table((struct acpi_table_dmar *)dmar_tbl, 926 &validate_drhd_cb); 927 if (!ret && !no_iommu && !iommu_detected && 928 (!dmar_disabled || dmar_platform_optin())) { 929 iommu_detected = 1; 930 /* Make sure ACS will be enabled */ 931 pci_request_acs(); 932 } 933 934 #ifdef CONFIG_X86 935 if (!ret) { 936 x86_init.iommu.iommu_init = intel_iommu_init; 937 x86_platform.iommu_shutdown = intel_iommu_shutdown; 938 } 939 940 #endif 941 942 if (dmar_tbl) { 943 acpi_put_table(dmar_tbl); 944 dmar_tbl = NULL; 945 } 946 up_write(&dmar_global_lock); 947 } 948 949 static void unmap_iommu(struct intel_iommu *iommu) 950 { 951 iounmap(iommu->reg); 952 release_mem_region(iommu->reg_phys, iommu->reg_size); 953 } 954 955 /** 956 * map_iommu: map the iommu's registers 957 * @iommu: the iommu to map 958 * @phys_addr: the physical address of the base resgister 959 * 960 * Memory map the iommu's registers. Start w/ a single page, and 961 * possibly expand if that turns out to be insufficent. 962 */ 963 static int map_iommu(struct intel_iommu *iommu, u64 phys_addr) 964 { 965 int map_size, err=0; 966 967 iommu->reg_phys = phys_addr; 968 iommu->reg_size = VTD_PAGE_SIZE; 969 970 if (!request_mem_region(iommu->reg_phys, iommu->reg_size, iommu->name)) { 971 pr_err("Can't reserve memory\n"); 972 err = -EBUSY; 973 goto out; 974 } 975 976 iommu->reg = ioremap(iommu->reg_phys, iommu->reg_size); 977 if (!iommu->reg) { 978 pr_err("Can't map the region\n"); 979 err = -ENOMEM; 980 goto release; 981 } 982 983 iommu->cap = dmar_readq(iommu->reg + DMAR_CAP_REG); 984 iommu->ecap = dmar_readq(iommu->reg + DMAR_ECAP_REG); 985 986 if (iommu->cap == (uint64_t)-1 && iommu->ecap == (uint64_t)-1) { 987 err = -EINVAL; 988 warn_invalid_dmar(phys_addr, " returns all ones"); 989 goto unmap; 990 } 991 if (ecap_vcs(iommu->ecap)) 992 iommu->vccap = dmar_readq(iommu->reg + DMAR_VCCAP_REG); 993 994 /* the registers might be more than one page */ 995 map_size = max_t(int, ecap_max_iotlb_offset(iommu->ecap), 996 cap_max_fault_reg_offset(iommu->cap)); 997 map_size = VTD_PAGE_ALIGN(map_size); 998 if (map_size > iommu->reg_size) { 999 iounmap(iommu->reg); 1000 release_mem_region(iommu->reg_phys, iommu->reg_size); 1001 iommu->reg_size = map_size; 1002 if (!request_mem_region(iommu->reg_phys, iommu->reg_size, 1003 iommu->name)) { 1004 pr_err("Can't reserve memory\n"); 1005 err = -EBUSY; 1006 goto out; 1007 } 1008 iommu->reg = ioremap(iommu->reg_phys, iommu->reg_size); 1009 if (!iommu->reg) { 1010 pr_err("Can't map the region\n"); 1011 err = -ENOMEM; 1012 goto release; 1013 } 1014 } 1015 err = 0; 1016 goto out; 1017 1018 unmap: 1019 iounmap(iommu->reg); 1020 release: 1021 release_mem_region(iommu->reg_phys, iommu->reg_size); 1022 out: 1023 return err; 1024 } 1025 1026 static int dmar_alloc_seq_id(struct intel_iommu *iommu) 1027 { 1028 iommu->seq_id = find_first_zero_bit(dmar_seq_ids, 1029 DMAR_UNITS_SUPPORTED); 1030 if (iommu->seq_id >= DMAR_UNITS_SUPPORTED) { 1031 iommu->seq_id = -1; 1032 } else { 1033 set_bit(iommu->seq_id, dmar_seq_ids); 1034 sprintf(iommu->name, "dmar%d", iommu->seq_id); 1035 } 1036 1037 return iommu->seq_id; 1038 } 1039 1040 static void dmar_free_seq_id(struct intel_iommu *iommu) 1041 { 1042 if (iommu->seq_id >= 0) { 1043 clear_bit(iommu->seq_id, dmar_seq_ids); 1044 iommu->seq_id = -1; 1045 } 1046 } 1047 1048 static int alloc_iommu(struct dmar_drhd_unit *drhd) 1049 { 1050 struct intel_iommu *iommu; 1051 u32 ver, sts; 1052 int agaw = -1; 1053 int msagaw = -1; 1054 int err; 1055 1056 if (!drhd->reg_base_addr) { 1057 warn_invalid_dmar(0, ""); 1058 return -EINVAL; 1059 } 1060 1061 iommu = kzalloc(sizeof(*iommu), GFP_KERNEL); 1062 if (!iommu) 1063 return -ENOMEM; 1064 1065 if (dmar_alloc_seq_id(iommu) < 0) { 1066 pr_err("Failed to allocate seq_id\n"); 1067 err = -ENOSPC; 1068 goto error; 1069 } 1070 1071 err = map_iommu(iommu, drhd->reg_base_addr); 1072 if (err) { 1073 pr_err("Failed to map %s\n", iommu->name); 1074 goto error_free_seq_id; 1075 } 1076 1077 err = -EINVAL; 1078 if (cap_sagaw(iommu->cap) == 0) { 1079 pr_info("%s: No supported address widths. Not attempting DMA translation.\n", 1080 iommu->name); 1081 drhd->ignored = 1; 1082 } 1083 1084 if (!drhd->ignored) { 1085 agaw = iommu_calculate_agaw(iommu); 1086 if (agaw < 0) { 1087 pr_err("Cannot get a valid agaw for iommu (seq_id = %d)\n", 1088 iommu->seq_id); 1089 drhd->ignored = 1; 1090 } 1091 } 1092 if (!drhd->ignored) { 1093 msagaw = iommu_calculate_max_sagaw(iommu); 1094 if (msagaw < 0) { 1095 pr_err("Cannot get a valid max agaw for iommu (seq_id = %d)\n", 1096 iommu->seq_id); 1097 drhd->ignored = 1; 1098 agaw = -1; 1099 } 1100 } 1101 iommu->agaw = agaw; 1102 iommu->msagaw = msagaw; 1103 iommu->segment = drhd->segment; 1104 1105 iommu->node = NUMA_NO_NODE; 1106 1107 ver = readl(iommu->reg + DMAR_VER_REG); 1108 pr_info("%s: reg_base_addr %llx ver %d:%d cap %llx ecap %llx\n", 1109 iommu->name, 1110 (unsigned long long)drhd->reg_base_addr, 1111 DMAR_VER_MAJOR(ver), DMAR_VER_MINOR(ver), 1112 (unsigned long long)iommu->cap, 1113 (unsigned long long)iommu->ecap); 1114 1115 /* Reflect status in gcmd */ 1116 sts = readl(iommu->reg + DMAR_GSTS_REG); 1117 if (sts & DMA_GSTS_IRES) 1118 iommu->gcmd |= DMA_GCMD_IRE; 1119 if (sts & DMA_GSTS_TES) 1120 iommu->gcmd |= DMA_GCMD_TE; 1121 if (sts & DMA_GSTS_QIES) 1122 iommu->gcmd |= DMA_GCMD_QIE; 1123 1124 raw_spin_lock_init(&iommu->register_lock); 1125 1126 /* 1127 * This is only for hotplug; at boot time intel_iommu_enabled won't 1128 * be set yet. When intel_iommu_init() runs, it registers the units 1129 * present at boot time, then sets intel_iommu_enabled. 1130 */ 1131 if (intel_iommu_enabled && !drhd->ignored) { 1132 err = iommu_device_sysfs_add(&iommu->iommu, NULL, 1133 intel_iommu_groups, 1134 "%s", iommu->name); 1135 if (err) 1136 goto err_unmap; 1137 1138 err = iommu_device_register(&iommu->iommu, &intel_iommu_ops, NULL); 1139 if (err) 1140 goto err_sysfs; 1141 } 1142 1143 drhd->iommu = iommu; 1144 iommu->drhd = drhd; 1145 1146 return 0; 1147 1148 err_sysfs: 1149 iommu_device_sysfs_remove(&iommu->iommu); 1150 err_unmap: 1151 unmap_iommu(iommu); 1152 error_free_seq_id: 1153 dmar_free_seq_id(iommu); 1154 error: 1155 kfree(iommu); 1156 return err; 1157 } 1158 1159 static void free_iommu(struct intel_iommu *iommu) 1160 { 1161 if (intel_iommu_enabled && !iommu->drhd->ignored) { 1162 iommu_device_unregister(&iommu->iommu); 1163 iommu_device_sysfs_remove(&iommu->iommu); 1164 } 1165 1166 if (iommu->irq) { 1167 if (iommu->pr_irq) { 1168 free_irq(iommu->pr_irq, iommu); 1169 dmar_free_hwirq(iommu->pr_irq); 1170 iommu->pr_irq = 0; 1171 } 1172 free_irq(iommu->irq, iommu); 1173 dmar_free_hwirq(iommu->irq); 1174 iommu->irq = 0; 1175 } 1176 1177 if (iommu->qi) { 1178 free_page((unsigned long)iommu->qi->desc); 1179 kfree(iommu->qi->desc_status); 1180 kfree(iommu->qi); 1181 } 1182 1183 if (iommu->reg) 1184 unmap_iommu(iommu); 1185 1186 dmar_free_seq_id(iommu); 1187 kfree(iommu); 1188 } 1189 1190 /* 1191 * Reclaim all the submitted descriptors which have completed its work. 1192 */ 1193 static inline void reclaim_free_desc(struct q_inval *qi) 1194 { 1195 while (qi->desc_status[qi->free_tail] == QI_DONE || 1196 qi->desc_status[qi->free_tail] == QI_ABORT) { 1197 qi->desc_status[qi->free_tail] = QI_FREE; 1198 qi->free_tail = (qi->free_tail + 1) % QI_LENGTH; 1199 qi->free_cnt++; 1200 } 1201 } 1202 1203 static const char *qi_type_string(u8 type) 1204 { 1205 switch (type) { 1206 case QI_CC_TYPE: 1207 return "Context-cache Invalidation"; 1208 case QI_IOTLB_TYPE: 1209 return "IOTLB Invalidation"; 1210 case QI_DIOTLB_TYPE: 1211 return "Device-TLB Invalidation"; 1212 case QI_IEC_TYPE: 1213 return "Interrupt Entry Cache Invalidation"; 1214 case QI_IWD_TYPE: 1215 return "Invalidation Wait"; 1216 case QI_EIOTLB_TYPE: 1217 return "PASID-based IOTLB Invalidation"; 1218 case QI_PC_TYPE: 1219 return "PASID-cache Invalidation"; 1220 case QI_DEIOTLB_TYPE: 1221 return "PASID-based Device-TLB Invalidation"; 1222 case QI_PGRP_RESP_TYPE: 1223 return "Page Group Response"; 1224 default: 1225 return "UNKNOWN"; 1226 } 1227 } 1228 1229 static void qi_dump_fault(struct intel_iommu *iommu, u32 fault) 1230 { 1231 unsigned int head = dmar_readl(iommu->reg + DMAR_IQH_REG); 1232 u64 iqe_err = dmar_readq(iommu->reg + DMAR_IQER_REG); 1233 struct qi_desc *desc = iommu->qi->desc + head; 1234 1235 if (fault & DMA_FSTS_IQE) 1236 pr_err("VT-d detected Invalidation Queue Error: Reason %llx", 1237 DMAR_IQER_REG_IQEI(iqe_err)); 1238 if (fault & DMA_FSTS_ITE) 1239 pr_err("VT-d detected Invalidation Time-out Error: SID %llx", 1240 DMAR_IQER_REG_ITESID(iqe_err)); 1241 if (fault & DMA_FSTS_ICE) 1242 pr_err("VT-d detected Invalidation Completion Error: SID %llx", 1243 DMAR_IQER_REG_ICESID(iqe_err)); 1244 1245 pr_err("QI HEAD: %s qw0 = 0x%llx, qw1 = 0x%llx\n", 1246 qi_type_string(desc->qw0 & 0xf), 1247 (unsigned long long)desc->qw0, 1248 (unsigned long long)desc->qw1); 1249 1250 head = ((head >> qi_shift(iommu)) + QI_LENGTH - 1) % QI_LENGTH; 1251 head <<= qi_shift(iommu); 1252 desc = iommu->qi->desc + head; 1253 1254 pr_err("QI PRIOR: %s qw0 = 0x%llx, qw1 = 0x%llx\n", 1255 qi_type_string(desc->qw0 & 0xf), 1256 (unsigned long long)desc->qw0, 1257 (unsigned long long)desc->qw1); 1258 } 1259 1260 static int qi_check_fault(struct intel_iommu *iommu, int index, int wait_index) 1261 { 1262 u32 fault; 1263 int head, tail; 1264 struct q_inval *qi = iommu->qi; 1265 int shift = qi_shift(iommu); 1266 1267 if (qi->desc_status[wait_index] == QI_ABORT) 1268 return -EAGAIN; 1269 1270 fault = readl(iommu->reg + DMAR_FSTS_REG); 1271 if (fault & (DMA_FSTS_IQE | DMA_FSTS_ITE | DMA_FSTS_ICE)) 1272 qi_dump_fault(iommu, fault); 1273 1274 /* 1275 * If IQE happens, the head points to the descriptor associated 1276 * with the error. No new descriptors are fetched until the IQE 1277 * is cleared. 1278 */ 1279 if (fault & DMA_FSTS_IQE) { 1280 head = readl(iommu->reg + DMAR_IQH_REG); 1281 if ((head >> shift) == index) { 1282 struct qi_desc *desc = qi->desc + head; 1283 1284 /* 1285 * desc->qw2 and desc->qw3 are either reserved or 1286 * used by software as private data. We won't print 1287 * out these two qw's for security consideration. 1288 */ 1289 memcpy(desc, qi->desc + (wait_index << shift), 1290 1 << shift); 1291 writel(DMA_FSTS_IQE, iommu->reg + DMAR_FSTS_REG); 1292 pr_info("Invalidation Queue Error (IQE) cleared\n"); 1293 return -EINVAL; 1294 } 1295 } 1296 1297 /* 1298 * If ITE happens, all pending wait_desc commands are aborted. 1299 * No new descriptors are fetched until the ITE is cleared. 1300 */ 1301 if (fault & DMA_FSTS_ITE) { 1302 head = readl(iommu->reg + DMAR_IQH_REG); 1303 head = ((head >> shift) - 1 + QI_LENGTH) % QI_LENGTH; 1304 head |= 1; 1305 tail = readl(iommu->reg + DMAR_IQT_REG); 1306 tail = ((tail >> shift) - 1 + QI_LENGTH) % QI_LENGTH; 1307 1308 writel(DMA_FSTS_ITE, iommu->reg + DMAR_FSTS_REG); 1309 pr_info("Invalidation Time-out Error (ITE) cleared\n"); 1310 1311 do { 1312 if (qi->desc_status[head] == QI_IN_USE) 1313 qi->desc_status[head] = QI_ABORT; 1314 head = (head - 2 + QI_LENGTH) % QI_LENGTH; 1315 } while (head != tail); 1316 1317 if (qi->desc_status[wait_index] == QI_ABORT) 1318 return -EAGAIN; 1319 } 1320 1321 if (fault & DMA_FSTS_ICE) { 1322 writel(DMA_FSTS_ICE, iommu->reg + DMAR_FSTS_REG); 1323 pr_info("Invalidation Completion Error (ICE) cleared\n"); 1324 } 1325 1326 return 0; 1327 } 1328 1329 /* 1330 * Function to submit invalidation descriptors of all types to the queued 1331 * invalidation interface(QI). Multiple descriptors can be submitted at a 1332 * time, a wait descriptor will be appended to each submission to ensure 1333 * hardware has completed the invalidation before return. Wait descriptors 1334 * can be part of the submission but it will not be polled for completion. 1335 */ 1336 int qi_submit_sync(struct intel_iommu *iommu, struct qi_desc *desc, 1337 unsigned int count, unsigned long options) 1338 { 1339 struct q_inval *qi = iommu->qi; 1340 s64 devtlb_start_ktime = 0; 1341 s64 iotlb_start_ktime = 0; 1342 s64 iec_start_ktime = 0; 1343 struct qi_desc wait_desc; 1344 int wait_index, index; 1345 unsigned long flags; 1346 int offset, shift; 1347 int rc, i; 1348 u64 type; 1349 1350 if (!qi) 1351 return 0; 1352 1353 type = desc->qw0 & GENMASK_ULL(3, 0); 1354 1355 if ((type == QI_IOTLB_TYPE || type == QI_EIOTLB_TYPE) && 1356 dmar_latency_enabled(iommu, DMAR_LATENCY_INV_IOTLB)) 1357 iotlb_start_ktime = ktime_to_ns(ktime_get()); 1358 1359 if ((type == QI_DIOTLB_TYPE || type == QI_DEIOTLB_TYPE) && 1360 dmar_latency_enabled(iommu, DMAR_LATENCY_INV_DEVTLB)) 1361 devtlb_start_ktime = ktime_to_ns(ktime_get()); 1362 1363 if (type == QI_IEC_TYPE && 1364 dmar_latency_enabled(iommu, DMAR_LATENCY_INV_IEC)) 1365 iec_start_ktime = ktime_to_ns(ktime_get()); 1366 1367 restart: 1368 rc = 0; 1369 1370 raw_spin_lock_irqsave(&qi->q_lock, flags); 1371 /* 1372 * Check if we have enough empty slots in the queue to submit, 1373 * the calculation is based on: 1374 * # of desc + 1 wait desc + 1 space between head and tail 1375 */ 1376 while (qi->free_cnt < count + 2) { 1377 raw_spin_unlock_irqrestore(&qi->q_lock, flags); 1378 cpu_relax(); 1379 raw_spin_lock_irqsave(&qi->q_lock, flags); 1380 } 1381 1382 index = qi->free_head; 1383 wait_index = (index + count) % QI_LENGTH; 1384 shift = qi_shift(iommu); 1385 1386 for (i = 0; i < count; i++) { 1387 offset = ((index + i) % QI_LENGTH) << shift; 1388 memcpy(qi->desc + offset, &desc[i], 1 << shift); 1389 qi->desc_status[(index + i) % QI_LENGTH] = QI_IN_USE; 1390 trace_qi_submit(iommu, desc[i].qw0, desc[i].qw1, 1391 desc[i].qw2, desc[i].qw3); 1392 } 1393 qi->desc_status[wait_index] = QI_IN_USE; 1394 1395 wait_desc.qw0 = QI_IWD_STATUS_DATA(QI_DONE) | 1396 QI_IWD_STATUS_WRITE | QI_IWD_TYPE; 1397 if (options & QI_OPT_WAIT_DRAIN) 1398 wait_desc.qw0 |= QI_IWD_PRQ_DRAIN; 1399 wait_desc.qw1 = virt_to_phys(&qi->desc_status[wait_index]); 1400 wait_desc.qw2 = 0; 1401 wait_desc.qw3 = 0; 1402 1403 offset = wait_index << shift; 1404 memcpy(qi->desc + offset, &wait_desc, 1 << shift); 1405 1406 qi->free_head = (qi->free_head + count + 1) % QI_LENGTH; 1407 qi->free_cnt -= count + 1; 1408 1409 /* 1410 * update the HW tail register indicating the presence of 1411 * new descriptors. 1412 */ 1413 writel(qi->free_head << shift, iommu->reg + DMAR_IQT_REG); 1414 1415 while (qi->desc_status[wait_index] != QI_DONE) { 1416 /* 1417 * We will leave the interrupts disabled, to prevent interrupt 1418 * context to queue another cmd while a cmd is already submitted 1419 * and waiting for completion on this cpu. This is to avoid 1420 * a deadlock where the interrupt context can wait indefinitely 1421 * for free slots in the queue. 1422 */ 1423 rc = qi_check_fault(iommu, index, wait_index); 1424 if (rc) 1425 break; 1426 1427 raw_spin_unlock(&qi->q_lock); 1428 cpu_relax(); 1429 raw_spin_lock(&qi->q_lock); 1430 } 1431 1432 for (i = 0; i < count; i++) 1433 qi->desc_status[(index + i) % QI_LENGTH] = QI_DONE; 1434 1435 reclaim_free_desc(qi); 1436 raw_spin_unlock_irqrestore(&qi->q_lock, flags); 1437 1438 if (rc == -EAGAIN) 1439 goto restart; 1440 1441 if (iotlb_start_ktime) 1442 dmar_latency_update(iommu, DMAR_LATENCY_INV_IOTLB, 1443 ktime_to_ns(ktime_get()) - iotlb_start_ktime); 1444 1445 if (devtlb_start_ktime) 1446 dmar_latency_update(iommu, DMAR_LATENCY_INV_DEVTLB, 1447 ktime_to_ns(ktime_get()) - devtlb_start_ktime); 1448 1449 if (iec_start_ktime) 1450 dmar_latency_update(iommu, DMAR_LATENCY_INV_IEC, 1451 ktime_to_ns(ktime_get()) - iec_start_ktime); 1452 1453 return rc; 1454 } 1455 1456 /* 1457 * Flush the global interrupt entry cache. 1458 */ 1459 void qi_global_iec(struct intel_iommu *iommu) 1460 { 1461 struct qi_desc desc; 1462 1463 desc.qw0 = QI_IEC_TYPE; 1464 desc.qw1 = 0; 1465 desc.qw2 = 0; 1466 desc.qw3 = 0; 1467 1468 /* should never fail */ 1469 qi_submit_sync(iommu, &desc, 1, 0); 1470 } 1471 1472 void qi_flush_context(struct intel_iommu *iommu, u16 did, u16 sid, u8 fm, 1473 u64 type) 1474 { 1475 struct qi_desc desc; 1476 1477 desc.qw0 = QI_CC_FM(fm) | QI_CC_SID(sid) | QI_CC_DID(did) 1478 | QI_CC_GRAN(type) | QI_CC_TYPE; 1479 desc.qw1 = 0; 1480 desc.qw2 = 0; 1481 desc.qw3 = 0; 1482 1483 qi_submit_sync(iommu, &desc, 1, 0); 1484 } 1485 1486 void qi_flush_iotlb(struct intel_iommu *iommu, u16 did, u64 addr, 1487 unsigned int size_order, u64 type) 1488 { 1489 u8 dw = 0, dr = 0; 1490 1491 struct qi_desc desc; 1492 int ih = 0; 1493 1494 if (cap_write_drain(iommu->cap)) 1495 dw = 1; 1496 1497 if (cap_read_drain(iommu->cap)) 1498 dr = 1; 1499 1500 desc.qw0 = QI_IOTLB_DID(did) | QI_IOTLB_DR(dr) | QI_IOTLB_DW(dw) 1501 | QI_IOTLB_GRAN(type) | QI_IOTLB_TYPE; 1502 desc.qw1 = QI_IOTLB_ADDR(addr) | QI_IOTLB_IH(ih) 1503 | QI_IOTLB_AM(size_order); 1504 desc.qw2 = 0; 1505 desc.qw3 = 0; 1506 1507 qi_submit_sync(iommu, &desc, 1, 0); 1508 } 1509 1510 void qi_flush_dev_iotlb(struct intel_iommu *iommu, u16 sid, u16 pfsid, 1511 u16 qdep, u64 addr, unsigned mask) 1512 { 1513 struct qi_desc desc; 1514 1515 if (mask) { 1516 addr |= (1ULL << (VTD_PAGE_SHIFT + mask - 1)) - 1; 1517 desc.qw1 = QI_DEV_IOTLB_ADDR(addr) | QI_DEV_IOTLB_SIZE; 1518 } else 1519 desc.qw1 = QI_DEV_IOTLB_ADDR(addr); 1520 1521 if (qdep >= QI_DEV_IOTLB_MAX_INVS) 1522 qdep = 0; 1523 1524 desc.qw0 = QI_DEV_IOTLB_SID(sid) | QI_DEV_IOTLB_QDEP(qdep) | 1525 QI_DIOTLB_TYPE | QI_DEV_IOTLB_PFSID(pfsid); 1526 desc.qw2 = 0; 1527 desc.qw3 = 0; 1528 1529 qi_submit_sync(iommu, &desc, 1, 0); 1530 } 1531 1532 /* PASID-based IOTLB invalidation */ 1533 void qi_flush_piotlb(struct intel_iommu *iommu, u16 did, u32 pasid, u64 addr, 1534 unsigned long npages, bool ih) 1535 { 1536 struct qi_desc desc = {.qw2 = 0, .qw3 = 0}; 1537 1538 /* 1539 * npages == -1 means a PASID-selective invalidation, otherwise, 1540 * a positive value for Page-selective-within-PASID invalidation. 1541 * 0 is not a valid input. 1542 */ 1543 if (WARN_ON(!npages)) { 1544 pr_err("Invalid input npages = %ld\n", npages); 1545 return; 1546 } 1547 1548 if (npages == -1) { 1549 desc.qw0 = QI_EIOTLB_PASID(pasid) | 1550 QI_EIOTLB_DID(did) | 1551 QI_EIOTLB_GRAN(QI_GRAN_NONG_PASID) | 1552 QI_EIOTLB_TYPE; 1553 desc.qw1 = 0; 1554 } else { 1555 int mask = ilog2(__roundup_pow_of_two(npages)); 1556 unsigned long align = (1ULL << (VTD_PAGE_SHIFT + mask)); 1557 1558 if (WARN_ON_ONCE(!IS_ALIGNED(addr, align))) 1559 addr = ALIGN_DOWN(addr, align); 1560 1561 desc.qw0 = QI_EIOTLB_PASID(pasid) | 1562 QI_EIOTLB_DID(did) | 1563 QI_EIOTLB_GRAN(QI_GRAN_PSI_PASID) | 1564 QI_EIOTLB_TYPE; 1565 desc.qw1 = QI_EIOTLB_ADDR(addr) | 1566 QI_EIOTLB_IH(ih) | 1567 QI_EIOTLB_AM(mask); 1568 } 1569 1570 qi_submit_sync(iommu, &desc, 1, 0); 1571 } 1572 1573 /* PASID-based device IOTLB Invalidate */ 1574 void qi_flush_dev_iotlb_pasid(struct intel_iommu *iommu, u16 sid, u16 pfsid, 1575 u32 pasid, u16 qdep, u64 addr, unsigned int size_order) 1576 { 1577 unsigned long mask = 1UL << (VTD_PAGE_SHIFT + size_order - 1); 1578 struct qi_desc desc = {.qw1 = 0, .qw2 = 0, .qw3 = 0}; 1579 1580 desc.qw0 = QI_DEV_EIOTLB_PASID(pasid) | QI_DEV_EIOTLB_SID(sid) | 1581 QI_DEV_EIOTLB_QDEP(qdep) | QI_DEIOTLB_TYPE | 1582 QI_DEV_IOTLB_PFSID(pfsid); 1583 1584 /* 1585 * If S bit is 0, we only flush a single page. If S bit is set, 1586 * The least significant zero bit indicates the invalidation address 1587 * range. VT-d spec 6.5.2.6. 1588 * e.g. address bit 12[0] indicates 8KB, 13[0] indicates 16KB. 1589 * size order = 0 is PAGE_SIZE 4KB 1590 * Max Invs Pending (MIP) is set to 0 for now until we have DIT in 1591 * ECAP. 1592 */ 1593 if (!IS_ALIGNED(addr, VTD_PAGE_SIZE << size_order)) 1594 pr_warn_ratelimited("Invalidate non-aligned address %llx, order %d\n", 1595 addr, size_order); 1596 1597 /* Take page address */ 1598 desc.qw1 = QI_DEV_EIOTLB_ADDR(addr); 1599 1600 if (size_order) { 1601 /* 1602 * Existing 0s in address below size_order may be the least 1603 * significant bit, we must set them to 1s to avoid having 1604 * smaller size than desired. 1605 */ 1606 desc.qw1 |= GENMASK_ULL(size_order + VTD_PAGE_SHIFT - 1, 1607 VTD_PAGE_SHIFT); 1608 /* Clear size_order bit to indicate size */ 1609 desc.qw1 &= ~mask; 1610 /* Set the S bit to indicate flushing more than 1 page */ 1611 desc.qw1 |= QI_DEV_EIOTLB_SIZE; 1612 } 1613 1614 qi_submit_sync(iommu, &desc, 1, 0); 1615 } 1616 1617 void qi_flush_pasid_cache(struct intel_iommu *iommu, u16 did, 1618 u64 granu, u32 pasid) 1619 { 1620 struct qi_desc desc = {.qw1 = 0, .qw2 = 0, .qw3 = 0}; 1621 1622 desc.qw0 = QI_PC_PASID(pasid) | QI_PC_DID(did) | 1623 QI_PC_GRAN(granu) | QI_PC_TYPE; 1624 qi_submit_sync(iommu, &desc, 1, 0); 1625 } 1626 1627 /* 1628 * Disable Queued Invalidation interface. 1629 */ 1630 void dmar_disable_qi(struct intel_iommu *iommu) 1631 { 1632 unsigned long flags; 1633 u32 sts; 1634 cycles_t start_time = get_cycles(); 1635 1636 if (!ecap_qis(iommu->ecap)) 1637 return; 1638 1639 raw_spin_lock_irqsave(&iommu->register_lock, flags); 1640 1641 sts = readl(iommu->reg + DMAR_GSTS_REG); 1642 if (!(sts & DMA_GSTS_QIES)) 1643 goto end; 1644 1645 /* 1646 * Give a chance to HW to complete the pending invalidation requests. 1647 */ 1648 while ((readl(iommu->reg + DMAR_IQT_REG) != 1649 readl(iommu->reg + DMAR_IQH_REG)) && 1650 (DMAR_OPERATION_TIMEOUT > (get_cycles() - start_time))) 1651 cpu_relax(); 1652 1653 iommu->gcmd &= ~DMA_GCMD_QIE; 1654 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG); 1655 1656 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, readl, 1657 !(sts & DMA_GSTS_QIES), sts); 1658 end: 1659 raw_spin_unlock_irqrestore(&iommu->register_lock, flags); 1660 } 1661 1662 /* 1663 * Enable queued invalidation. 1664 */ 1665 static void __dmar_enable_qi(struct intel_iommu *iommu) 1666 { 1667 u32 sts; 1668 unsigned long flags; 1669 struct q_inval *qi = iommu->qi; 1670 u64 val = virt_to_phys(qi->desc); 1671 1672 qi->free_head = qi->free_tail = 0; 1673 qi->free_cnt = QI_LENGTH; 1674 1675 /* 1676 * Set DW=1 and QS=1 in IQA_REG when Scalable Mode capability 1677 * is present. 1678 */ 1679 if (ecap_smts(iommu->ecap)) 1680 val |= (1 << 11) | 1; 1681 1682 raw_spin_lock_irqsave(&iommu->register_lock, flags); 1683 1684 /* write zero to the tail reg */ 1685 writel(0, iommu->reg + DMAR_IQT_REG); 1686 1687 dmar_writeq(iommu->reg + DMAR_IQA_REG, val); 1688 1689 iommu->gcmd |= DMA_GCMD_QIE; 1690 writel(iommu->gcmd, iommu->reg + DMAR_GCMD_REG); 1691 1692 /* Make sure hardware complete it */ 1693 IOMMU_WAIT_OP(iommu, DMAR_GSTS_REG, readl, (sts & DMA_GSTS_QIES), sts); 1694 1695 raw_spin_unlock_irqrestore(&iommu->register_lock, flags); 1696 } 1697 1698 /* 1699 * Enable Queued Invalidation interface. This is a must to support 1700 * interrupt-remapping. Also used by DMA-remapping, which replaces 1701 * register based IOTLB invalidation. 1702 */ 1703 int dmar_enable_qi(struct intel_iommu *iommu) 1704 { 1705 struct q_inval *qi; 1706 struct page *desc_page; 1707 1708 if (!ecap_qis(iommu->ecap)) 1709 return -ENOENT; 1710 1711 /* 1712 * queued invalidation is already setup and enabled. 1713 */ 1714 if (iommu->qi) 1715 return 0; 1716 1717 iommu->qi = kmalloc(sizeof(*qi), GFP_ATOMIC); 1718 if (!iommu->qi) 1719 return -ENOMEM; 1720 1721 qi = iommu->qi; 1722 1723 /* 1724 * Need two pages to accommodate 256 descriptors of 256 bits each 1725 * if the remapping hardware supports scalable mode translation. 1726 */ 1727 desc_page = alloc_pages_node(iommu->node, GFP_ATOMIC | __GFP_ZERO, 1728 !!ecap_smts(iommu->ecap)); 1729 if (!desc_page) { 1730 kfree(qi); 1731 iommu->qi = NULL; 1732 return -ENOMEM; 1733 } 1734 1735 qi->desc = page_address(desc_page); 1736 1737 qi->desc_status = kcalloc(QI_LENGTH, sizeof(int), GFP_ATOMIC); 1738 if (!qi->desc_status) { 1739 free_page((unsigned long) qi->desc); 1740 kfree(qi); 1741 iommu->qi = NULL; 1742 return -ENOMEM; 1743 } 1744 1745 raw_spin_lock_init(&qi->q_lock); 1746 1747 __dmar_enable_qi(iommu); 1748 1749 return 0; 1750 } 1751 1752 /* iommu interrupt handling. Most stuff are MSI-like. */ 1753 1754 enum faulttype { 1755 DMA_REMAP, 1756 INTR_REMAP, 1757 UNKNOWN, 1758 }; 1759 1760 static const char *dma_remap_fault_reasons[] = 1761 { 1762 "Software", 1763 "Present bit in root entry is clear", 1764 "Present bit in context entry is clear", 1765 "Invalid context entry", 1766 "Access beyond MGAW", 1767 "PTE Write access is not set", 1768 "PTE Read access is not set", 1769 "Next page table ptr is invalid", 1770 "Root table address invalid", 1771 "Context table ptr is invalid", 1772 "non-zero reserved fields in RTP", 1773 "non-zero reserved fields in CTP", 1774 "non-zero reserved fields in PTE", 1775 "PCE for translation request specifies blocking", 1776 }; 1777 1778 static const char * const dma_remap_sm_fault_reasons[] = { 1779 "SM: Invalid Root Table Address", 1780 "SM: TTM 0 for request with PASID", 1781 "SM: TTM 0 for page group request", 1782 "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x33-0x37 */ 1783 "SM: Error attempting to access Root Entry", 1784 "SM: Present bit in Root Entry is clear", 1785 "SM: Non-zero reserved field set in Root Entry", 1786 "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x3B-0x3F */ 1787 "SM: Error attempting to access Context Entry", 1788 "SM: Present bit in Context Entry is clear", 1789 "SM: Non-zero reserved field set in the Context Entry", 1790 "SM: Invalid Context Entry", 1791 "SM: DTE field in Context Entry is clear", 1792 "SM: PASID Enable field in Context Entry is clear", 1793 "SM: PASID is larger than the max in Context Entry", 1794 "SM: PRE field in Context-Entry is clear", 1795 "SM: RID_PASID field error in Context-Entry", 1796 "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x49-0x4F */ 1797 "SM: Error attempting to access the PASID Directory Entry", 1798 "SM: Present bit in Directory Entry is clear", 1799 "SM: Non-zero reserved field set in PASID Directory Entry", 1800 "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x53-0x57 */ 1801 "SM: Error attempting to access PASID Table Entry", 1802 "SM: Present bit in PASID Table Entry is clear", 1803 "SM: Non-zero reserved field set in PASID Table Entry", 1804 "SM: Invalid Scalable-Mode PASID Table Entry", 1805 "SM: ERE field is clear in PASID Table Entry", 1806 "SM: SRE field is clear in PASID Table Entry", 1807 "Unknown", "Unknown",/* 0x5E-0x5F */ 1808 "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x60-0x67 */ 1809 "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x68-0x6F */ 1810 "SM: Error attempting to access first-level paging entry", 1811 "SM: Present bit in first-level paging entry is clear", 1812 "SM: Non-zero reserved field set in first-level paging entry", 1813 "SM: Error attempting to access FL-PML4 entry", 1814 "SM: First-level entry address beyond MGAW in Nested translation", 1815 "SM: Read permission error in FL-PML4 entry in Nested translation", 1816 "SM: Read permission error in first-level paging entry in Nested translation", 1817 "SM: Write permission error in first-level paging entry in Nested translation", 1818 "SM: Error attempting to access second-level paging entry", 1819 "SM: Read/Write permission error in second-level paging entry", 1820 "SM: Non-zero reserved field set in second-level paging entry", 1821 "SM: Invalid second-level page table pointer", 1822 "SM: A/D bit update needed in second-level entry when set up in no snoop", 1823 "Unknown", "Unknown", "Unknown", /* 0x7D-0x7F */ 1824 "SM: Address in first-level translation is not canonical", 1825 "SM: U/S set 0 for first-level translation with user privilege", 1826 "SM: No execute permission for request with PASID and ER=1", 1827 "SM: Address beyond the DMA hardware max", 1828 "SM: Second-level entry address beyond the max", 1829 "SM: No write permission for Write/AtomicOp request", 1830 "SM: No read permission for Read/AtomicOp request", 1831 "SM: Invalid address-interrupt address", 1832 "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", "Unknown", /* 0x88-0x8F */ 1833 "SM: A/D bit update needed in first-level entry when set up in no snoop", 1834 }; 1835 1836 static const char *irq_remap_fault_reasons[] = 1837 { 1838 "Detected reserved fields in the decoded interrupt-remapped request", 1839 "Interrupt index exceeded the interrupt-remapping table size", 1840 "Present field in the IRTE entry is clear", 1841 "Error accessing interrupt-remapping table pointed by IRTA_REG", 1842 "Detected reserved fields in the IRTE entry", 1843 "Blocked a compatibility format interrupt request", 1844 "Blocked an interrupt request due to source-id verification failure", 1845 }; 1846 1847 static const char *dmar_get_fault_reason(u8 fault_reason, int *fault_type) 1848 { 1849 if (fault_reason >= 0x20 && (fault_reason - 0x20 < 1850 ARRAY_SIZE(irq_remap_fault_reasons))) { 1851 *fault_type = INTR_REMAP; 1852 return irq_remap_fault_reasons[fault_reason - 0x20]; 1853 } else if (fault_reason >= 0x30 && (fault_reason - 0x30 < 1854 ARRAY_SIZE(dma_remap_sm_fault_reasons))) { 1855 *fault_type = DMA_REMAP; 1856 return dma_remap_sm_fault_reasons[fault_reason - 0x30]; 1857 } else if (fault_reason < ARRAY_SIZE(dma_remap_fault_reasons)) { 1858 *fault_type = DMA_REMAP; 1859 return dma_remap_fault_reasons[fault_reason]; 1860 } else { 1861 *fault_type = UNKNOWN; 1862 return "Unknown"; 1863 } 1864 } 1865 1866 1867 static inline int dmar_msi_reg(struct intel_iommu *iommu, int irq) 1868 { 1869 if (iommu->irq == irq) 1870 return DMAR_FECTL_REG; 1871 else if (iommu->pr_irq == irq) 1872 return DMAR_PECTL_REG; 1873 else 1874 BUG(); 1875 } 1876 1877 void dmar_msi_unmask(struct irq_data *data) 1878 { 1879 struct intel_iommu *iommu = irq_data_get_irq_handler_data(data); 1880 int reg = dmar_msi_reg(iommu, data->irq); 1881 unsigned long flag; 1882 1883 /* unmask it */ 1884 raw_spin_lock_irqsave(&iommu->register_lock, flag); 1885 writel(0, iommu->reg + reg); 1886 /* Read a reg to force flush the post write */ 1887 readl(iommu->reg + reg); 1888 raw_spin_unlock_irqrestore(&iommu->register_lock, flag); 1889 } 1890 1891 void dmar_msi_mask(struct irq_data *data) 1892 { 1893 struct intel_iommu *iommu = irq_data_get_irq_handler_data(data); 1894 int reg = dmar_msi_reg(iommu, data->irq); 1895 unsigned long flag; 1896 1897 /* mask it */ 1898 raw_spin_lock_irqsave(&iommu->register_lock, flag); 1899 writel(DMA_FECTL_IM, iommu->reg + reg); 1900 /* Read a reg to force flush the post write */ 1901 readl(iommu->reg + reg); 1902 raw_spin_unlock_irqrestore(&iommu->register_lock, flag); 1903 } 1904 1905 void dmar_msi_write(int irq, struct msi_msg *msg) 1906 { 1907 struct intel_iommu *iommu = irq_get_handler_data(irq); 1908 int reg = dmar_msi_reg(iommu, irq); 1909 unsigned long flag; 1910 1911 raw_spin_lock_irqsave(&iommu->register_lock, flag); 1912 writel(msg->data, iommu->reg + reg + 4); 1913 writel(msg->address_lo, iommu->reg + reg + 8); 1914 writel(msg->address_hi, iommu->reg + reg + 12); 1915 raw_spin_unlock_irqrestore(&iommu->register_lock, flag); 1916 } 1917 1918 void dmar_msi_read(int irq, struct msi_msg *msg) 1919 { 1920 struct intel_iommu *iommu = irq_get_handler_data(irq); 1921 int reg = dmar_msi_reg(iommu, irq); 1922 unsigned long flag; 1923 1924 raw_spin_lock_irqsave(&iommu->register_lock, flag); 1925 msg->data = readl(iommu->reg + reg + 4); 1926 msg->address_lo = readl(iommu->reg + reg + 8); 1927 msg->address_hi = readl(iommu->reg + reg + 12); 1928 raw_spin_unlock_irqrestore(&iommu->register_lock, flag); 1929 } 1930 1931 static int dmar_fault_do_one(struct intel_iommu *iommu, int type, 1932 u8 fault_reason, u32 pasid, u16 source_id, 1933 unsigned long long addr) 1934 { 1935 const char *reason; 1936 int fault_type; 1937 1938 reason = dmar_get_fault_reason(fault_reason, &fault_type); 1939 1940 if (fault_type == INTR_REMAP) { 1941 pr_err("[INTR-REMAP] Request device [%02x:%02x.%d] fault index 0x%llx [fault reason 0x%02x] %s\n", 1942 source_id >> 8, PCI_SLOT(source_id & 0xFF), 1943 PCI_FUNC(source_id & 0xFF), addr >> 48, 1944 fault_reason, reason); 1945 1946 return 0; 1947 } 1948 1949 if (pasid == INVALID_IOASID) 1950 pr_err("[%s NO_PASID] Request device [%02x:%02x.%d] fault addr 0x%llx [fault reason 0x%02x] %s\n", 1951 type ? "DMA Read" : "DMA Write", 1952 source_id >> 8, PCI_SLOT(source_id & 0xFF), 1953 PCI_FUNC(source_id & 0xFF), addr, 1954 fault_reason, reason); 1955 else 1956 pr_err("[%s PASID 0x%x] Request device [%02x:%02x.%d] fault addr 0x%llx [fault reason 0x%02x] %s\n", 1957 type ? "DMA Read" : "DMA Write", pasid, 1958 source_id >> 8, PCI_SLOT(source_id & 0xFF), 1959 PCI_FUNC(source_id & 0xFF), addr, 1960 fault_reason, reason); 1961 1962 dmar_fault_dump_ptes(iommu, source_id, addr, pasid); 1963 1964 return 0; 1965 } 1966 1967 #define PRIMARY_FAULT_REG_LEN (16) 1968 irqreturn_t dmar_fault(int irq, void *dev_id) 1969 { 1970 struct intel_iommu *iommu = dev_id; 1971 int reg, fault_index; 1972 u32 fault_status; 1973 unsigned long flag; 1974 static DEFINE_RATELIMIT_STATE(rs, 1975 DEFAULT_RATELIMIT_INTERVAL, 1976 DEFAULT_RATELIMIT_BURST); 1977 1978 raw_spin_lock_irqsave(&iommu->register_lock, flag); 1979 fault_status = readl(iommu->reg + DMAR_FSTS_REG); 1980 if (fault_status && __ratelimit(&rs)) 1981 pr_err("DRHD: handling fault status reg %x\n", fault_status); 1982 1983 /* TBD: ignore advanced fault log currently */ 1984 if (!(fault_status & DMA_FSTS_PPF)) 1985 goto unlock_exit; 1986 1987 fault_index = dma_fsts_fault_record_index(fault_status); 1988 reg = cap_fault_reg_offset(iommu->cap); 1989 while (1) { 1990 /* Disable printing, simply clear the fault when ratelimited */ 1991 bool ratelimited = !__ratelimit(&rs); 1992 u8 fault_reason; 1993 u16 source_id; 1994 u64 guest_addr; 1995 u32 pasid; 1996 int type; 1997 u32 data; 1998 bool pasid_present; 1999 2000 /* highest 32 bits */ 2001 data = readl(iommu->reg + reg + 2002 fault_index * PRIMARY_FAULT_REG_LEN + 12); 2003 if (!(data & DMA_FRCD_F)) 2004 break; 2005 2006 if (!ratelimited) { 2007 fault_reason = dma_frcd_fault_reason(data); 2008 type = dma_frcd_type(data); 2009 2010 pasid = dma_frcd_pasid_value(data); 2011 data = readl(iommu->reg + reg + 2012 fault_index * PRIMARY_FAULT_REG_LEN + 8); 2013 source_id = dma_frcd_source_id(data); 2014 2015 pasid_present = dma_frcd_pasid_present(data); 2016 guest_addr = dmar_readq(iommu->reg + reg + 2017 fault_index * PRIMARY_FAULT_REG_LEN); 2018 guest_addr = dma_frcd_page_addr(guest_addr); 2019 } 2020 2021 /* clear the fault */ 2022 writel(DMA_FRCD_F, iommu->reg + reg + 2023 fault_index * PRIMARY_FAULT_REG_LEN + 12); 2024 2025 raw_spin_unlock_irqrestore(&iommu->register_lock, flag); 2026 2027 if (!ratelimited) 2028 /* Using pasid -1 if pasid is not present */ 2029 dmar_fault_do_one(iommu, type, fault_reason, 2030 pasid_present ? pasid : INVALID_IOASID, 2031 source_id, guest_addr); 2032 2033 fault_index++; 2034 if (fault_index >= cap_num_fault_regs(iommu->cap)) 2035 fault_index = 0; 2036 raw_spin_lock_irqsave(&iommu->register_lock, flag); 2037 } 2038 2039 writel(DMA_FSTS_PFO | DMA_FSTS_PPF | DMA_FSTS_PRO, 2040 iommu->reg + DMAR_FSTS_REG); 2041 2042 unlock_exit: 2043 raw_spin_unlock_irqrestore(&iommu->register_lock, flag); 2044 return IRQ_HANDLED; 2045 } 2046 2047 int dmar_set_interrupt(struct intel_iommu *iommu) 2048 { 2049 int irq, ret; 2050 2051 /* 2052 * Check if the fault interrupt is already initialized. 2053 */ 2054 if (iommu->irq) 2055 return 0; 2056 2057 irq = dmar_alloc_hwirq(iommu->seq_id, iommu->node, iommu); 2058 if (irq > 0) { 2059 iommu->irq = irq; 2060 } else { 2061 pr_err("No free IRQ vectors\n"); 2062 return -EINVAL; 2063 } 2064 2065 ret = request_irq(irq, dmar_fault, IRQF_NO_THREAD, iommu->name, iommu); 2066 if (ret) 2067 pr_err("Can't request irq\n"); 2068 return ret; 2069 } 2070 2071 int __init enable_drhd_fault_handling(void) 2072 { 2073 struct dmar_drhd_unit *drhd; 2074 struct intel_iommu *iommu; 2075 2076 /* 2077 * Enable fault control interrupt. 2078 */ 2079 for_each_iommu(iommu, drhd) { 2080 u32 fault_status; 2081 int ret = dmar_set_interrupt(iommu); 2082 2083 if (ret) { 2084 pr_err("DRHD %Lx: failed to enable fault, interrupt, ret %d\n", 2085 (unsigned long long)drhd->reg_base_addr, ret); 2086 return -1; 2087 } 2088 2089 /* 2090 * Clear any previous faults. 2091 */ 2092 dmar_fault(iommu->irq, iommu); 2093 fault_status = readl(iommu->reg + DMAR_FSTS_REG); 2094 writel(fault_status, iommu->reg + DMAR_FSTS_REG); 2095 } 2096 2097 return 0; 2098 } 2099 2100 /* 2101 * Re-enable Queued Invalidation interface. 2102 */ 2103 int dmar_reenable_qi(struct intel_iommu *iommu) 2104 { 2105 if (!ecap_qis(iommu->ecap)) 2106 return -ENOENT; 2107 2108 if (!iommu->qi) 2109 return -ENOENT; 2110 2111 /* 2112 * First disable queued invalidation. 2113 */ 2114 dmar_disable_qi(iommu); 2115 /* 2116 * Then enable queued invalidation again. Since there is no pending 2117 * invalidation requests now, it's safe to re-enable queued 2118 * invalidation. 2119 */ 2120 __dmar_enable_qi(iommu); 2121 2122 return 0; 2123 } 2124 2125 /* 2126 * Check interrupt remapping support in DMAR table description. 2127 */ 2128 int __init dmar_ir_support(void) 2129 { 2130 struct acpi_table_dmar *dmar; 2131 dmar = (struct acpi_table_dmar *)dmar_tbl; 2132 if (!dmar) 2133 return 0; 2134 return dmar->flags & 0x1; 2135 } 2136 2137 /* Check whether DMAR units are in use */ 2138 static inline bool dmar_in_use(void) 2139 { 2140 return irq_remapping_enabled || intel_iommu_enabled; 2141 } 2142 2143 static int __init dmar_free_unused_resources(void) 2144 { 2145 struct dmar_drhd_unit *dmaru, *dmaru_n; 2146 2147 if (dmar_in_use()) 2148 return 0; 2149 2150 if (dmar_dev_scope_status != 1 && !list_empty(&dmar_drhd_units)) 2151 bus_unregister_notifier(&pci_bus_type, &dmar_pci_bus_nb); 2152 2153 down_write(&dmar_global_lock); 2154 list_for_each_entry_safe(dmaru, dmaru_n, &dmar_drhd_units, list) { 2155 list_del(&dmaru->list); 2156 dmar_free_drhd(dmaru); 2157 } 2158 up_write(&dmar_global_lock); 2159 2160 return 0; 2161 } 2162 2163 late_initcall(dmar_free_unused_resources); 2164 2165 /* 2166 * DMAR Hotplug Support 2167 * For more details, please refer to Intel(R) Virtualization Technology 2168 * for Directed-IO Architecture Specifiction, Rev 2.2, Section 8.8 2169 * "Remapping Hardware Unit Hot Plug". 2170 */ 2171 static guid_t dmar_hp_guid = 2172 GUID_INIT(0xD8C1A3A6, 0xBE9B, 0x4C9B, 2173 0x91, 0xBF, 0xC3, 0xCB, 0x81, 0xFC, 0x5D, 0xAF); 2174 2175 /* 2176 * Currently there's only one revision and BIOS will not check the revision id, 2177 * so use 0 for safety. 2178 */ 2179 #define DMAR_DSM_REV_ID 0 2180 #define DMAR_DSM_FUNC_DRHD 1 2181 #define DMAR_DSM_FUNC_ATSR 2 2182 #define DMAR_DSM_FUNC_RHSA 3 2183 #define DMAR_DSM_FUNC_SATC 4 2184 2185 static inline bool dmar_detect_dsm(acpi_handle handle, int func) 2186 { 2187 return acpi_check_dsm(handle, &dmar_hp_guid, DMAR_DSM_REV_ID, 1 << func); 2188 } 2189 2190 static int dmar_walk_dsm_resource(acpi_handle handle, int func, 2191 dmar_res_handler_t handler, void *arg) 2192 { 2193 int ret = -ENODEV; 2194 union acpi_object *obj; 2195 struct acpi_dmar_header *start; 2196 struct dmar_res_callback callback; 2197 static int res_type[] = { 2198 [DMAR_DSM_FUNC_DRHD] = ACPI_DMAR_TYPE_HARDWARE_UNIT, 2199 [DMAR_DSM_FUNC_ATSR] = ACPI_DMAR_TYPE_ROOT_ATS, 2200 [DMAR_DSM_FUNC_RHSA] = ACPI_DMAR_TYPE_HARDWARE_AFFINITY, 2201 [DMAR_DSM_FUNC_SATC] = ACPI_DMAR_TYPE_SATC, 2202 }; 2203 2204 if (!dmar_detect_dsm(handle, func)) 2205 return 0; 2206 2207 obj = acpi_evaluate_dsm_typed(handle, &dmar_hp_guid, DMAR_DSM_REV_ID, 2208 func, NULL, ACPI_TYPE_BUFFER); 2209 if (!obj) 2210 return -ENODEV; 2211 2212 memset(&callback, 0, sizeof(callback)); 2213 callback.cb[res_type[func]] = handler; 2214 callback.arg[res_type[func]] = arg; 2215 start = (struct acpi_dmar_header *)obj->buffer.pointer; 2216 ret = dmar_walk_remapping_entries(start, obj->buffer.length, &callback); 2217 2218 ACPI_FREE(obj); 2219 2220 return ret; 2221 } 2222 2223 static int dmar_hp_add_drhd(struct acpi_dmar_header *header, void *arg) 2224 { 2225 int ret; 2226 struct dmar_drhd_unit *dmaru; 2227 2228 dmaru = dmar_find_dmaru((struct acpi_dmar_hardware_unit *)header); 2229 if (!dmaru) 2230 return -ENODEV; 2231 2232 ret = dmar_ir_hotplug(dmaru, true); 2233 if (ret == 0) 2234 ret = dmar_iommu_hotplug(dmaru, true); 2235 2236 return ret; 2237 } 2238 2239 static int dmar_hp_remove_drhd(struct acpi_dmar_header *header, void *arg) 2240 { 2241 int i, ret; 2242 struct device *dev; 2243 struct dmar_drhd_unit *dmaru; 2244 2245 dmaru = dmar_find_dmaru((struct acpi_dmar_hardware_unit *)header); 2246 if (!dmaru) 2247 return 0; 2248 2249 /* 2250 * All PCI devices managed by this unit should have been destroyed. 2251 */ 2252 if (!dmaru->include_all && dmaru->devices && dmaru->devices_cnt) { 2253 for_each_active_dev_scope(dmaru->devices, 2254 dmaru->devices_cnt, i, dev) 2255 return -EBUSY; 2256 } 2257 2258 ret = dmar_ir_hotplug(dmaru, false); 2259 if (ret == 0) 2260 ret = dmar_iommu_hotplug(dmaru, false); 2261 2262 return ret; 2263 } 2264 2265 static int dmar_hp_release_drhd(struct acpi_dmar_header *header, void *arg) 2266 { 2267 struct dmar_drhd_unit *dmaru; 2268 2269 dmaru = dmar_find_dmaru((struct acpi_dmar_hardware_unit *)header); 2270 if (dmaru) { 2271 list_del_rcu(&dmaru->list); 2272 synchronize_rcu(); 2273 dmar_free_drhd(dmaru); 2274 } 2275 2276 return 0; 2277 } 2278 2279 static int dmar_hotplug_insert(acpi_handle handle) 2280 { 2281 int ret; 2282 int drhd_count = 0; 2283 2284 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD, 2285 &dmar_validate_one_drhd, (void *)1); 2286 if (ret) 2287 goto out; 2288 2289 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD, 2290 &dmar_parse_one_drhd, (void *)&drhd_count); 2291 if (ret == 0 && drhd_count == 0) { 2292 pr_warn(FW_BUG "No DRHD structures in buffer returned by _DSM method\n"); 2293 goto out; 2294 } else if (ret) { 2295 goto release_drhd; 2296 } 2297 2298 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_RHSA, 2299 &dmar_parse_one_rhsa, NULL); 2300 if (ret) 2301 goto release_drhd; 2302 2303 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR, 2304 &dmar_parse_one_atsr, NULL); 2305 if (ret) 2306 goto release_atsr; 2307 2308 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD, 2309 &dmar_hp_add_drhd, NULL); 2310 if (!ret) 2311 return 0; 2312 2313 dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD, 2314 &dmar_hp_remove_drhd, NULL); 2315 release_atsr: 2316 dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR, 2317 &dmar_release_one_atsr, NULL); 2318 release_drhd: 2319 dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD, 2320 &dmar_hp_release_drhd, NULL); 2321 out: 2322 return ret; 2323 } 2324 2325 static int dmar_hotplug_remove(acpi_handle handle) 2326 { 2327 int ret; 2328 2329 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR, 2330 &dmar_check_one_atsr, NULL); 2331 if (ret) 2332 return ret; 2333 2334 ret = dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD, 2335 &dmar_hp_remove_drhd, NULL); 2336 if (ret == 0) { 2337 WARN_ON(dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_ATSR, 2338 &dmar_release_one_atsr, NULL)); 2339 WARN_ON(dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD, 2340 &dmar_hp_release_drhd, NULL)); 2341 } else { 2342 dmar_walk_dsm_resource(handle, DMAR_DSM_FUNC_DRHD, 2343 &dmar_hp_add_drhd, NULL); 2344 } 2345 2346 return ret; 2347 } 2348 2349 static acpi_status dmar_get_dsm_handle(acpi_handle handle, u32 lvl, 2350 void *context, void **retval) 2351 { 2352 acpi_handle *phdl = retval; 2353 2354 if (dmar_detect_dsm(handle, DMAR_DSM_FUNC_DRHD)) { 2355 *phdl = handle; 2356 return AE_CTRL_TERMINATE; 2357 } 2358 2359 return AE_OK; 2360 } 2361 2362 static int dmar_device_hotplug(acpi_handle handle, bool insert) 2363 { 2364 int ret; 2365 acpi_handle tmp = NULL; 2366 acpi_status status; 2367 2368 if (!dmar_in_use()) 2369 return 0; 2370 2371 if (dmar_detect_dsm(handle, DMAR_DSM_FUNC_DRHD)) { 2372 tmp = handle; 2373 } else { 2374 status = acpi_walk_namespace(ACPI_TYPE_DEVICE, handle, 2375 ACPI_UINT32_MAX, 2376 dmar_get_dsm_handle, 2377 NULL, NULL, &tmp); 2378 if (ACPI_FAILURE(status)) { 2379 pr_warn("Failed to locate _DSM method.\n"); 2380 return -ENXIO; 2381 } 2382 } 2383 if (tmp == NULL) 2384 return 0; 2385 2386 down_write(&dmar_global_lock); 2387 if (insert) 2388 ret = dmar_hotplug_insert(tmp); 2389 else 2390 ret = dmar_hotplug_remove(tmp); 2391 up_write(&dmar_global_lock); 2392 2393 return ret; 2394 } 2395 2396 int dmar_device_add(acpi_handle handle) 2397 { 2398 return dmar_device_hotplug(handle, true); 2399 } 2400 2401 int dmar_device_remove(acpi_handle handle) 2402 { 2403 return dmar_device_hotplug(handle, false); 2404 } 2405 2406 /* 2407 * dmar_platform_optin - Is %DMA_CTRL_PLATFORM_OPT_IN_FLAG set in DMAR table 2408 * 2409 * Returns true if the platform has %DMA_CTRL_PLATFORM_OPT_IN_FLAG set in 2410 * the ACPI DMAR table. This means that the platform boot firmware has made 2411 * sure no device can issue DMA outside of RMRR regions. 2412 */ 2413 bool dmar_platform_optin(void) 2414 { 2415 struct acpi_table_dmar *dmar; 2416 acpi_status status; 2417 bool ret; 2418 2419 status = acpi_get_table(ACPI_SIG_DMAR, 0, 2420 (struct acpi_table_header **)&dmar); 2421 if (ACPI_FAILURE(status)) 2422 return false; 2423 2424 ret = !!(dmar->flags & DMAR_PLATFORM_OPT_IN); 2425 acpi_put_table((struct acpi_table_header *)dmar); 2426 2427 return ret; 2428 } 2429 EXPORT_SYMBOL_GPL(dmar_platform_optin); 2430