1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (C) 2007-2010 Advanced Micro Devices, Inc. 4 * Author: Joerg Roedel <jroedel@suse.de> 5 * Leo Duran <leo.duran@amd.com> 6 */ 7 8 #define pr_fmt(fmt) "AMD-Vi: " fmt 9 #define dev_fmt(fmt) pr_fmt(fmt) 10 11 #include <linux/pci.h> 12 #include <linux/acpi.h> 13 #include <linux/list.h> 14 #include <linux/bitmap.h> 15 #include <linux/slab.h> 16 #include <linux/syscore_ops.h> 17 #include <linux/interrupt.h> 18 #include <linux/msi.h> 19 #include <linux/irq.h> 20 #include <linux/amd-iommu.h> 21 #include <linux/export.h> 22 #include <linux/kmemleak.h> 23 #include <linux/cc_platform.h> 24 #include <linux/iopoll.h> 25 #include <asm/pci-direct.h> 26 #include <asm/iommu.h> 27 #include <asm/apic.h> 28 #include <asm/gart.h> 29 #include <asm/x86_init.h> 30 #include <asm/io_apic.h> 31 #include <asm/irq_remapping.h> 32 #include <asm/set_memory.h> 33 34 #include <linux/crash_dump.h> 35 36 #include "amd_iommu.h" 37 #include "../irq_remapping.h" 38 39 /* 40 * definitions for the ACPI scanning code 41 */ 42 #define IVRS_HEADER_LENGTH 48 43 44 #define ACPI_IVHD_TYPE_MAX_SUPPORTED 0x40 45 #define ACPI_IVMD_TYPE_ALL 0x20 46 #define ACPI_IVMD_TYPE 0x21 47 #define ACPI_IVMD_TYPE_RANGE 0x22 48 49 #define IVHD_DEV_ALL 0x01 50 #define IVHD_DEV_SELECT 0x02 51 #define IVHD_DEV_SELECT_RANGE_START 0x03 52 #define IVHD_DEV_RANGE_END 0x04 53 #define IVHD_DEV_ALIAS 0x42 54 #define IVHD_DEV_ALIAS_RANGE 0x43 55 #define IVHD_DEV_EXT_SELECT 0x46 56 #define IVHD_DEV_EXT_SELECT_RANGE 0x47 57 #define IVHD_DEV_SPECIAL 0x48 58 #define IVHD_DEV_ACPI_HID 0xf0 59 60 #define UID_NOT_PRESENT 0 61 #define UID_IS_INTEGER 1 62 #define UID_IS_CHARACTER 2 63 64 #define IVHD_SPECIAL_IOAPIC 1 65 #define IVHD_SPECIAL_HPET 2 66 67 #define IVHD_FLAG_HT_TUN_EN_MASK 0x01 68 #define IVHD_FLAG_PASSPW_EN_MASK 0x02 69 #define IVHD_FLAG_RESPASSPW_EN_MASK 0x04 70 #define IVHD_FLAG_ISOC_EN_MASK 0x08 71 72 #define IVMD_FLAG_EXCL_RANGE 0x08 73 #define IVMD_FLAG_IW 0x04 74 #define IVMD_FLAG_IR 0x02 75 #define IVMD_FLAG_UNITY_MAP 0x01 76 77 #define ACPI_DEVFLAG_INITPASS 0x01 78 #define ACPI_DEVFLAG_EXTINT 0x02 79 #define ACPI_DEVFLAG_NMI 0x04 80 #define ACPI_DEVFLAG_SYSMGT1 0x10 81 #define ACPI_DEVFLAG_SYSMGT2 0x20 82 #define ACPI_DEVFLAG_LINT0 0x40 83 #define ACPI_DEVFLAG_LINT1 0x80 84 #define ACPI_DEVFLAG_ATSDIS 0x10000000 85 86 #define LOOP_TIMEOUT 2000000 87 88 #define IVRS_GET_SBDF_ID(seg, bus, dev, fd) (((seg & 0xffff) << 16) | ((bus & 0xff) << 8) \ 89 | ((dev & 0x1f) << 3) | (fn & 0x7)) 90 91 /* 92 * ACPI table definitions 93 * 94 * These data structures are laid over the table to parse the important values 95 * out of it. 96 */ 97 98 extern const struct iommu_ops amd_iommu_ops; 99 100 /* 101 * structure describing one IOMMU in the ACPI table. Typically followed by one 102 * or more ivhd_entrys. 103 */ 104 struct ivhd_header { 105 u8 type; 106 u8 flags; 107 u16 length; 108 u16 devid; 109 u16 cap_ptr; 110 u64 mmio_phys; 111 u16 pci_seg; 112 u16 info; 113 u32 efr_attr; 114 115 /* Following only valid on IVHD type 11h and 40h */ 116 u64 efr_reg; /* Exact copy of MMIO_EXT_FEATURES */ 117 u64 efr_reg2; 118 } __attribute__((packed)); 119 120 /* 121 * A device entry describing which devices a specific IOMMU translates and 122 * which requestor ids they use. 123 */ 124 struct ivhd_entry { 125 u8 type; 126 u16 devid; 127 u8 flags; 128 struct_group(ext_hid, 129 u32 ext; 130 u32 hidh; 131 ); 132 u64 cid; 133 u8 uidf; 134 u8 uidl; 135 u8 uid; 136 } __attribute__((packed)); 137 138 /* 139 * An AMD IOMMU memory definition structure. It defines things like exclusion 140 * ranges for devices and regions that should be unity mapped. 141 */ 142 struct ivmd_header { 143 u8 type; 144 u8 flags; 145 u16 length; 146 u16 devid; 147 u16 aux; 148 u16 pci_seg; 149 u8 resv[6]; 150 u64 range_start; 151 u64 range_length; 152 } __attribute__((packed)); 153 154 bool amd_iommu_dump; 155 bool amd_iommu_irq_remap __read_mostly; 156 157 enum io_pgtable_fmt amd_iommu_pgtable = AMD_IOMMU_V1; 158 159 int amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC; 160 static int amd_iommu_xt_mode = IRQ_REMAP_XAPIC_MODE; 161 162 static bool amd_iommu_detected; 163 static bool amd_iommu_disabled __initdata; 164 static bool amd_iommu_force_enable __initdata; 165 static int amd_iommu_target_ivhd_type; 166 167 /* Global EFR and EFR2 registers */ 168 u64 amd_iommu_efr; 169 u64 amd_iommu_efr2; 170 171 /* SNP is enabled on the system? */ 172 bool amd_iommu_snp_en; 173 EXPORT_SYMBOL(amd_iommu_snp_en); 174 175 LIST_HEAD(amd_iommu_pci_seg_list); /* list of all PCI segments */ 176 LIST_HEAD(amd_iommu_list); /* list of all AMD IOMMUs in the 177 system */ 178 179 /* Array to assign indices to IOMMUs*/ 180 struct amd_iommu *amd_iommus[MAX_IOMMUS]; 181 182 /* Number of IOMMUs present in the system */ 183 static int amd_iommus_present; 184 185 /* IOMMUs have a non-present cache? */ 186 bool amd_iommu_np_cache __read_mostly; 187 bool amd_iommu_iotlb_sup __read_mostly = true; 188 189 u32 amd_iommu_max_pasid __read_mostly = ~0; 190 191 bool amd_iommu_v2_present __read_mostly; 192 static bool amd_iommu_pc_present __read_mostly; 193 bool amdr_ivrs_remap_support __read_mostly; 194 195 bool amd_iommu_force_isolation __read_mostly; 196 197 /* 198 * AMD IOMMU allows up to 2^16 different protection domains. This is a bitmap 199 * to know which ones are already in use. 200 */ 201 unsigned long *amd_iommu_pd_alloc_bitmap; 202 203 enum iommu_init_state { 204 IOMMU_START_STATE, 205 IOMMU_IVRS_DETECTED, 206 IOMMU_ACPI_FINISHED, 207 IOMMU_ENABLED, 208 IOMMU_PCI_INIT, 209 IOMMU_INTERRUPTS_EN, 210 IOMMU_INITIALIZED, 211 IOMMU_NOT_FOUND, 212 IOMMU_INIT_ERROR, 213 IOMMU_CMDLINE_DISABLED, 214 }; 215 216 /* Early ioapic and hpet maps from kernel command line */ 217 #define EARLY_MAP_SIZE 4 218 static struct devid_map __initdata early_ioapic_map[EARLY_MAP_SIZE]; 219 static struct devid_map __initdata early_hpet_map[EARLY_MAP_SIZE]; 220 static struct acpihid_map_entry __initdata early_acpihid_map[EARLY_MAP_SIZE]; 221 222 static int __initdata early_ioapic_map_size; 223 static int __initdata early_hpet_map_size; 224 static int __initdata early_acpihid_map_size; 225 226 static bool __initdata cmdline_maps; 227 228 static enum iommu_init_state init_state = IOMMU_START_STATE; 229 230 static int amd_iommu_enable_interrupts(void); 231 static int __init iommu_go_to_state(enum iommu_init_state state); 232 static void init_device_table_dma(struct amd_iommu_pci_seg *pci_seg); 233 234 static bool amd_iommu_pre_enabled = true; 235 236 static u32 amd_iommu_ivinfo __initdata; 237 238 bool translation_pre_enabled(struct amd_iommu *iommu) 239 { 240 return (iommu->flags & AMD_IOMMU_FLAG_TRANS_PRE_ENABLED); 241 } 242 243 static void clear_translation_pre_enabled(struct amd_iommu *iommu) 244 { 245 iommu->flags &= ~AMD_IOMMU_FLAG_TRANS_PRE_ENABLED; 246 } 247 248 static void init_translation_status(struct amd_iommu *iommu) 249 { 250 u64 ctrl; 251 252 ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET); 253 if (ctrl & (1<<CONTROL_IOMMU_EN)) 254 iommu->flags |= AMD_IOMMU_FLAG_TRANS_PRE_ENABLED; 255 } 256 257 static inline unsigned long tbl_size(int entry_size, int last_bdf) 258 { 259 unsigned shift = PAGE_SHIFT + 260 get_order((last_bdf + 1) * entry_size); 261 262 return 1UL << shift; 263 } 264 265 int amd_iommu_get_num_iommus(void) 266 { 267 return amd_iommus_present; 268 } 269 270 /* 271 * Iterate through all the IOMMUs to get common EFR 272 * masks among all IOMMUs and warn if found inconsistency. 273 */ 274 static void get_global_efr(void) 275 { 276 struct amd_iommu *iommu; 277 278 for_each_iommu(iommu) { 279 u64 tmp = iommu->features; 280 u64 tmp2 = iommu->features2; 281 282 if (list_is_first(&iommu->list, &amd_iommu_list)) { 283 amd_iommu_efr = tmp; 284 amd_iommu_efr2 = tmp2; 285 continue; 286 } 287 288 if (amd_iommu_efr == tmp && 289 amd_iommu_efr2 == tmp2) 290 continue; 291 292 pr_err(FW_BUG 293 "Found inconsistent EFR/EFR2 %#llx,%#llx (global %#llx,%#llx) on iommu%d (%04x:%02x:%02x.%01x).\n", 294 tmp, tmp2, amd_iommu_efr, amd_iommu_efr2, 295 iommu->index, iommu->pci_seg->id, 296 PCI_BUS_NUM(iommu->devid), PCI_SLOT(iommu->devid), 297 PCI_FUNC(iommu->devid)); 298 299 amd_iommu_efr &= tmp; 300 amd_iommu_efr2 &= tmp2; 301 } 302 303 pr_info("Using global IVHD EFR:%#llx, EFR2:%#llx\n", amd_iommu_efr, amd_iommu_efr2); 304 } 305 306 static bool check_feature_on_all_iommus(u64 mask) 307 { 308 return !!(amd_iommu_efr & mask); 309 } 310 311 /* 312 * For IVHD type 0x11/0x40, EFR is also available via IVHD. 313 * Default to IVHD EFR since it is available sooner 314 * (i.e. before PCI init). 315 */ 316 static void __init early_iommu_features_init(struct amd_iommu *iommu, 317 struct ivhd_header *h) 318 { 319 if (amd_iommu_ivinfo & IOMMU_IVINFO_EFRSUP) { 320 iommu->features = h->efr_reg; 321 iommu->features2 = h->efr_reg2; 322 } 323 if (amd_iommu_ivinfo & IOMMU_IVINFO_DMA_REMAP) 324 amdr_ivrs_remap_support = true; 325 } 326 327 /* Access to l1 and l2 indexed register spaces */ 328 329 static u32 iommu_read_l1(struct amd_iommu *iommu, u16 l1, u8 address) 330 { 331 u32 val; 332 333 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16)); 334 pci_read_config_dword(iommu->dev, 0xfc, &val); 335 return val; 336 } 337 338 static void iommu_write_l1(struct amd_iommu *iommu, u16 l1, u8 address, u32 val) 339 { 340 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16 | 1 << 31)); 341 pci_write_config_dword(iommu->dev, 0xfc, val); 342 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16)); 343 } 344 345 static u32 iommu_read_l2(struct amd_iommu *iommu, u8 address) 346 { 347 u32 val; 348 349 pci_write_config_dword(iommu->dev, 0xf0, address); 350 pci_read_config_dword(iommu->dev, 0xf4, &val); 351 return val; 352 } 353 354 static void iommu_write_l2(struct amd_iommu *iommu, u8 address, u32 val) 355 { 356 pci_write_config_dword(iommu->dev, 0xf0, (address | 1 << 8)); 357 pci_write_config_dword(iommu->dev, 0xf4, val); 358 } 359 360 /**************************************************************************** 361 * 362 * AMD IOMMU MMIO register space handling functions 363 * 364 * These functions are used to program the IOMMU device registers in 365 * MMIO space required for that driver. 366 * 367 ****************************************************************************/ 368 369 /* 370 * This function set the exclusion range in the IOMMU. DMA accesses to the 371 * exclusion range are passed through untranslated 372 */ 373 static void iommu_set_exclusion_range(struct amd_iommu *iommu) 374 { 375 u64 start = iommu->exclusion_start & PAGE_MASK; 376 u64 limit = (start + iommu->exclusion_length - 1) & PAGE_MASK; 377 u64 entry; 378 379 if (!iommu->exclusion_start) 380 return; 381 382 entry = start | MMIO_EXCL_ENABLE_MASK; 383 memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET, 384 &entry, sizeof(entry)); 385 386 entry = limit; 387 memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET, 388 &entry, sizeof(entry)); 389 } 390 391 static void iommu_set_cwwb_range(struct amd_iommu *iommu) 392 { 393 u64 start = iommu_virt_to_phys((void *)iommu->cmd_sem); 394 u64 entry = start & PM_ADDR_MASK; 395 396 if (!check_feature_on_all_iommus(FEATURE_SNP)) 397 return; 398 399 /* Note: 400 * Re-purpose Exclusion base/limit registers for Completion wait 401 * write-back base/limit. 402 */ 403 memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET, 404 &entry, sizeof(entry)); 405 406 /* Note: 407 * Default to 4 Kbytes, which can be specified by setting base 408 * address equal to the limit address. 409 */ 410 memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET, 411 &entry, sizeof(entry)); 412 } 413 414 /* Programs the physical address of the device table into the IOMMU hardware */ 415 static void iommu_set_device_table(struct amd_iommu *iommu) 416 { 417 u64 entry; 418 u32 dev_table_size = iommu->pci_seg->dev_table_size; 419 void *dev_table = (void *)get_dev_table(iommu); 420 421 BUG_ON(iommu->mmio_base == NULL); 422 423 entry = iommu_virt_to_phys(dev_table); 424 entry |= (dev_table_size >> 12) - 1; 425 memcpy_toio(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET, 426 &entry, sizeof(entry)); 427 } 428 429 /* Generic functions to enable/disable certain features of the IOMMU. */ 430 static void iommu_feature_enable(struct amd_iommu *iommu, u8 bit) 431 { 432 u64 ctrl; 433 434 ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET); 435 ctrl |= (1ULL << bit); 436 writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET); 437 } 438 439 static void iommu_feature_disable(struct amd_iommu *iommu, u8 bit) 440 { 441 u64 ctrl; 442 443 ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET); 444 ctrl &= ~(1ULL << bit); 445 writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET); 446 } 447 448 static void iommu_set_inv_tlb_timeout(struct amd_iommu *iommu, int timeout) 449 { 450 u64 ctrl; 451 452 ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET); 453 ctrl &= ~CTRL_INV_TO_MASK; 454 ctrl |= (timeout << CONTROL_INV_TIMEOUT) & CTRL_INV_TO_MASK; 455 writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET); 456 } 457 458 /* Function to enable the hardware */ 459 static void iommu_enable(struct amd_iommu *iommu) 460 { 461 iommu_feature_enable(iommu, CONTROL_IOMMU_EN); 462 } 463 464 static void iommu_disable(struct amd_iommu *iommu) 465 { 466 if (!iommu->mmio_base) 467 return; 468 469 /* Disable command buffer */ 470 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN); 471 472 /* Disable event logging and event interrupts */ 473 iommu_feature_disable(iommu, CONTROL_EVT_INT_EN); 474 iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN); 475 476 /* Disable IOMMU GA_LOG */ 477 iommu_feature_disable(iommu, CONTROL_GALOG_EN); 478 iommu_feature_disable(iommu, CONTROL_GAINT_EN); 479 480 /* Disable IOMMU hardware itself */ 481 iommu_feature_disable(iommu, CONTROL_IOMMU_EN); 482 } 483 484 /* 485 * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in 486 * the system has one. 487 */ 488 static u8 __iomem * __init iommu_map_mmio_space(u64 address, u64 end) 489 { 490 if (!request_mem_region(address, end, "amd_iommu")) { 491 pr_err("Can not reserve memory region %llx-%llx for mmio\n", 492 address, end); 493 pr_err("This is a BIOS bug. Please contact your hardware vendor\n"); 494 return NULL; 495 } 496 497 return (u8 __iomem *)ioremap(address, end); 498 } 499 500 static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu) 501 { 502 if (iommu->mmio_base) 503 iounmap(iommu->mmio_base); 504 release_mem_region(iommu->mmio_phys, iommu->mmio_phys_end); 505 } 506 507 static inline u32 get_ivhd_header_size(struct ivhd_header *h) 508 { 509 u32 size = 0; 510 511 switch (h->type) { 512 case 0x10: 513 size = 24; 514 break; 515 case 0x11: 516 case 0x40: 517 size = 40; 518 break; 519 } 520 return size; 521 } 522 523 /**************************************************************************** 524 * 525 * The functions below belong to the first pass of AMD IOMMU ACPI table 526 * parsing. In this pass we try to find out the highest device id this 527 * code has to handle. Upon this information the size of the shared data 528 * structures is determined later. 529 * 530 ****************************************************************************/ 531 532 /* 533 * This function calculates the length of a given IVHD entry 534 */ 535 static inline int ivhd_entry_length(u8 *ivhd) 536 { 537 u32 type = ((struct ivhd_entry *)ivhd)->type; 538 539 if (type < 0x80) { 540 return 0x04 << (*ivhd >> 6); 541 } else if (type == IVHD_DEV_ACPI_HID) { 542 /* For ACPI_HID, offset 21 is uid len */ 543 return *((u8 *)ivhd + 21) + 22; 544 } 545 return 0; 546 } 547 548 /* 549 * After reading the highest device id from the IOMMU PCI capability header 550 * this function looks if there is a higher device id defined in the ACPI table 551 */ 552 static int __init find_last_devid_from_ivhd(struct ivhd_header *h) 553 { 554 u8 *p = (void *)h, *end = (void *)h; 555 struct ivhd_entry *dev; 556 int last_devid = -EINVAL; 557 558 u32 ivhd_size = get_ivhd_header_size(h); 559 560 if (!ivhd_size) { 561 pr_err("Unsupported IVHD type %#x\n", h->type); 562 return -EINVAL; 563 } 564 565 p += ivhd_size; 566 end += h->length; 567 568 while (p < end) { 569 dev = (struct ivhd_entry *)p; 570 switch (dev->type) { 571 case IVHD_DEV_ALL: 572 /* Use maximum BDF value for DEV_ALL */ 573 return 0xffff; 574 case IVHD_DEV_SELECT: 575 case IVHD_DEV_RANGE_END: 576 case IVHD_DEV_ALIAS: 577 case IVHD_DEV_EXT_SELECT: 578 /* all the above subfield types refer to device ids */ 579 if (dev->devid > last_devid) 580 last_devid = dev->devid; 581 break; 582 default: 583 break; 584 } 585 p += ivhd_entry_length(p); 586 } 587 588 WARN_ON(p != end); 589 590 return last_devid; 591 } 592 593 static int __init check_ivrs_checksum(struct acpi_table_header *table) 594 { 595 int i; 596 u8 checksum = 0, *p = (u8 *)table; 597 598 for (i = 0; i < table->length; ++i) 599 checksum += p[i]; 600 if (checksum != 0) { 601 /* ACPI table corrupt */ 602 pr_err(FW_BUG "IVRS invalid checksum\n"); 603 return -ENODEV; 604 } 605 606 return 0; 607 } 608 609 /* 610 * Iterate over all IVHD entries in the ACPI table and find the highest device 611 * id which we need to handle. This is the first of three functions which parse 612 * the ACPI table. So we check the checksum here. 613 */ 614 static int __init find_last_devid_acpi(struct acpi_table_header *table, u16 pci_seg) 615 { 616 u8 *p = (u8 *)table, *end = (u8 *)table; 617 struct ivhd_header *h; 618 int last_devid, last_bdf = 0; 619 620 p += IVRS_HEADER_LENGTH; 621 622 end += table->length; 623 while (p < end) { 624 h = (struct ivhd_header *)p; 625 if (h->pci_seg == pci_seg && 626 h->type == amd_iommu_target_ivhd_type) { 627 last_devid = find_last_devid_from_ivhd(h); 628 629 if (last_devid < 0) 630 return -EINVAL; 631 if (last_devid > last_bdf) 632 last_bdf = last_devid; 633 } 634 p += h->length; 635 } 636 WARN_ON(p != end); 637 638 return last_bdf; 639 } 640 641 /**************************************************************************** 642 * 643 * The following functions belong to the code path which parses the ACPI table 644 * the second time. In this ACPI parsing iteration we allocate IOMMU specific 645 * data structures, initialize the per PCI segment device/alias/rlookup table 646 * and also basically initialize the hardware. 647 * 648 ****************************************************************************/ 649 650 /* Allocate per PCI segment device table */ 651 static inline int __init alloc_dev_table(struct amd_iommu_pci_seg *pci_seg) 652 { 653 pci_seg->dev_table = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO | GFP_DMA32, 654 get_order(pci_seg->dev_table_size)); 655 if (!pci_seg->dev_table) 656 return -ENOMEM; 657 658 return 0; 659 } 660 661 static inline void free_dev_table(struct amd_iommu_pci_seg *pci_seg) 662 { 663 free_pages((unsigned long)pci_seg->dev_table, 664 get_order(pci_seg->dev_table_size)); 665 pci_seg->dev_table = NULL; 666 } 667 668 /* Allocate per PCI segment IOMMU rlookup table. */ 669 static inline int __init alloc_rlookup_table(struct amd_iommu_pci_seg *pci_seg) 670 { 671 pci_seg->rlookup_table = (void *)__get_free_pages( 672 GFP_KERNEL | __GFP_ZERO, 673 get_order(pci_seg->rlookup_table_size)); 674 if (pci_seg->rlookup_table == NULL) 675 return -ENOMEM; 676 677 return 0; 678 } 679 680 static inline void free_rlookup_table(struct amd_iommu_pci_seg *pci_seg) 681 { 682 free_pages((unsigned long)pci_seg->rlookup_table, 683 get_order(pci_seg->rlookup_table_size)); 684 pci_seg->rlookup_table = NULL; 685 } 686 687 static inline int __init alloc_irq_lookup_table(struct amd_iommu_pci_seg *pci_seg) 688 { 689 pci_seg->irq_lookup_table = (void *)__get_free_pages( 690 GFP_KERNEL | __GFP_ZERO, 691 get_order(pci_seg->rlookup_table_size)); 692 kmemleak_alloc(pci_seg->irq_lookup_table, 693 pci_seg->rlookup_table_size, 1, GFP_KERNEL); 694 if (pci_seg->irq_lookup_table == NULL) 695 return -ENOMEM; 696 697 return 0; 698 } 699 700 static inline void free_irq_lookup_table(struct amd_iommu_pci_seg *pci_seg) 701 { 702 kmemleak_free(pci_seg->irq_lookup_table); 703 free_pages((unsigned long)pci_seg->irq_lookup_table, 704 get_order(pci_seg->rlookup_table_size)); 705 pci_seg->irq_lookup_table = NULL; 706 } 707 708 static int __init alloc_alias_table(struct amd_iommu_pci_seg *pci_seg) 709 { 710 int i; 711 712 pci_seg->alias_table = (void *)__get_free_pages(GFP_KERNEL, 713 get_order(pci_seg->alias_table_size)); 714 if (!pci_seg->alias_table) 715 return -ENOMEM; 716 717 /* 718 * let all alias entries point to itself 719 */ 720 for (i = 0; i <= pci_seg->last_bdf; ++i) 721 pci_seg->alias_table[i] = i; 722 723 return 0; 724 } 725 726 static void __init free_alias_table(struct amd_iommu_pci_seg *pci_seg) 727 { 728 free_pages((unsigned long)pci_seg->alias_table, 729 get_order(pci_seg->alias_table_size)); 730 pci_seg->alias_table = NULL; 731 } 732 733 /* 734 * Allocates the command buffer. This buffer is per AMD IOMMU. We can 735 * write commands to that buffer later and the IOMMU will execute them 736 * asynchronously 737 */ 738 static int __init alloc_command_buffer(struct amd_iommu *iommu) 739 { 740 iommu->cmd_buf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 741 get_order(CMD_BUFFER_SIZE)); 742 743 return iommu->cmd_buf ? 0 : -ENOMEM; 744 } 745 746 /* 747 * This function restarts event logging in case the IOMMU experienced 748 * an event log buffer overflow. 749 */ 750 void amd_iommu_restart_event_logging(struct amd_iommu *iommu) 751 { 752 iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN); 753 iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN); 754 } 755 756 /* 757 * This function resets the command buffer if the IOMMU stopped fetching 758 * commands from it. 759 */ 760 static void amd_iommu_reset_cmd_buffer(struct amd_iommu *iommu) 761 { 762 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN); 763 764 writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET); 765 writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET); 766 iommu->cmd_buf_head = 0; 767 iommu->cmd_buf_tail = 0; 768 769 iommu_feature_enable(iommu, CONTROL_CMDBUF_EN); 770 } 771 772 /* 773 * This function writes the command buffer address to the hardware and 774 * enables it. 775 */ 776 static void iommu_enable_command_buffer(struct amd_iommu *iommu) 777 { 778 u64 entry; 779 780 BUG_ON(iommu->cmd_buf == NULL); 781 782 entry = iommu_virt_to_phys(iommu->cmd_buf); 783 entry |= MMIO_CMD_SIZE_512; 784 785 memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET, 786 &entry, sizeof(entry)); 787 788 amd_iommu_reset_cmd_buffer(iommu); 789 } 790 791 /* 792 * This function disables the command buffer 793 */ 794 static void iommu_disable_command_buffer(struct amd_iommu *iommu) 795 { 796 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN); 797 } 798 799 static void __init free_command_buffer(struct amd_iommu *iommu) 800 { 801 free_pages((unsigned long)iommu->cmd_buf, get_order(CMD_BUFFER_SIZE)); 802 } 803 804 static void *__init iommu_alloc_4k_pages(struct amd_iommu *iommu, 805 gfp_t gfp, size_t size) 806 { 807 int order = get_order(size); 808 void *buf = (void *)__get_free_pages(gfp, order); 809 810 if (buf && 811 check_feature_on_all_iommus(FEATURE_SNP) && 812 set_memory_4k((unsigned long)buf, (1 << order))) { 813 free_pages((unsigned long)buf, order); 814 buf = NULL; 815 } 816 817 return buf; 818 } 819 820 /* allocates the memory where the IOMMU will log its events to */ 821 static int __init alloc_event_buffer(struct amd_iommu *iommu) 822 { 823 iommu->evt_buf = iommu_alloc_4k_pages(iommu, GFP_KERNEL | __GFP_ZERO, 824 EVT_BUFFER_SIZE); 825 826 return iommu->evt_buf ? 0 : -ENOMEM; 827 } 828 829 static void iommu_enable_event_buffer(struct amd_iommu *iommu) 830 { 831 u64 entry; 832 833 BUG_ON(iommu->evt_buf == NULL); 834 835 entry = iommu_virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK; 836 837 memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET, 838 &entry, sizeof(entry)); 839 840 /* set head and tail to zero manually */ 841 writel(0x00, iommu->mmio_base + MMIO_EVT_HEAD_OFFSET); 842 writel(0x00, iommu->mmio_base + MMIO_EVT_TAIL_OFFSET); 843 844 iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN); 845 } 846 847 /* 848 * This function disables the event log buffer 849 */ 850 static void iommu_disable_event_buffer(struct amd_iommu *iommu) 851 { 852 iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN); 853 } 854 855 static void __init free_event_buffer(struct amd_iommu *iommu) 856 { 857 free_pages((unsigned long)iommu->evt_buf, get_order(EVT_BUFFER_SIZE)); 858 } 859 860 /* allocates the memory where the IOMMU will log its events to */ 861 static int __init alloc_ppr_log(struct amd_iommu *iommu) 862 { 863 iommu->ppr_log = iommu_alloc_4k_pages(iommu, GFP_KERNEL | __GFP_ZERO, 864 PPR_LOG_SIZE); 865 866 return iommu->ppr_log ? 0 : -ENOMEM; 867 } 868 869 static void iommu_enable_ppr_log(struct amd_iommu *iommu) 870 { 871 u64 entry; 872 873 if (iommu->ppr_log == NULL) 874 return; 875 876 entry = iommu_virt_to_phys(iommu->ppr_log) | PPR_LOG_SIZE_512; 877 878 memcpy_toio(iommu->mmio_base + MMIO_PPR_LOG_OFFSET, 879 &entry, sizeof(entry)); 880 881 /* set head and tail to zero manually */ 882 writel(0x00, iommu->mmio_base + MMIO_PPR_HEAD_OFFSET); 883 writel(0x00, iommu->mmio_base + MMIO_PPR_TAIL_OFFSET); 884 885 iommu_feature_enable(iommu, CONTROL_PPRLOG_EN); 886 iommu_feature_enable(iommu, CONTROL_PPR_EN); 887 } 888 889 static void __init free_ppr_log(struct amd_iommu *iommu) 890 { 891 free_pages((unsigned long)iommu->ppr_log, get_order(PPR_LOG_SIZE)); 892 } 893 894 static void free_ga_log(struct amd_iommu *iommu) 895 { 896 #ifdef CONFIG_IRQ_REMAP 897 free_pages((unsigned long)iommu->ga_log, get_order(GA_LOG_SIZE)); 898 free_pages((unsigned long)iommu->ga_log_tail, get_order(8)); 899 #endif 900 } 901 902 #ifdef CONFIG_IRQ_REMAP 903 static int iommu_ga_log_enable(struct amd_iommu *iommu) 904 { 905 u32 status, i; 906 u64 entry; 907 908 if (!iommu->ga_log) 909 return -EINVAL; 910 911 entry = iommu_virt_to_phys(iommu->ga_log) | GA_LOG_SIZE_512; 912 memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_BASE_OFFSET, 913 &entry, sizeof(entry)); 914 entry = (iommu_virt_to_phys(iommu->ga_log_tail) & 915 (BIT_ULL(52)-1)) & ~7ULL; 916 memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_TAIL_OFFSET, 917 &entry, sizeof(entry)); 918 writel(0x00, iommu->mmio_base + MMIO_GA_HEAD_OFFSET); 919 writel(0x00, iommu->mmio_base + MMIO_GA_TAIL_OFFSET); 920 921 922 iommu_feature_enable(iommu, CONTROL_GAINT_EN); 923 iommu_feature_enable(iommu, CONTROL_GALOG_EN); 924 925 for (i = 0; i < LOOP_TIMEOUT; ++i) { 926 status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET); 927 if (status & (MMIO_STATUS_GALOG_RUN_MASK)) 928 break; 929 udelay(10); 930 } 931 932 if (WARN_ON(i >= LOOP_TIMEOUT)) 933 return -EINVAL; 934 935 return 0; 936 } 937 938 static int iommu_init_ga_log(struct amd_iommu *iommu) 939 { 940 if (!AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir)) 941 return 0; 942 943 iommu->ga_log = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 944 get_order(GA_LOG_SIZE)); 945 if (!iommu->ga_log) 946 goto err_out; 947 948 iommu->ga_log_tail = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 949 get_order(8)); 950 if (!iommu->ga_log_tail) 951 goto err_out; 952 953 return 0; 954 err_out: 955 free_ga_log(iommu); 956 return -EINVAL; 957 } 958 #endif /* CONFIG_IRQ_REMAP */ 959 960 static int __init alloc_cwwb_sem(struct amd_iommu *iommu) 961 { 962 iommu->cmd_sem = iommu_alloc_4k_pages(iommu, GFP_KERNEL | __GFP_ZERO, 1); 963 964 return iommu->cmd_sem ? 0 : -ENOMEM; 965 } 966 967 static void __init free_cwwb_sem(struct amd_iommu *iommu) 968 { 969 if (iommu->cmd_sem) 970 free_page((unsigned long)iommu->cmd_sem); 971 } 972 973 static void iommu_enable_xt(struct amd_iommu *iommu) 974 { 975 #ifdef CONFIG_IRQ_REMAP 976 /* 977 * XT mode (32-bit APIC destination ID) requires 978 * GA mode (128-bit IRTE support) as a prerequisite. 979 */ 980 if (AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir) && 981 amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE) 982 iommu_feature_enable(iommu, CONTROL_XT_EN); 983 #endif /* CONFIG_IRQ_REMAP */ 984 } 985 986 static void iommu_enable_gt(struct amd_iommu *iommu) 987 { 988 if (!iommu_feature(iommu, FEATURE_GT)) 989 return; 990 991 iommu_feature_enable(iommu, CONTROL_GT_EN); 992 } 993 994 /* sets a specific bit in the device table entry. */ 995 static void __set_dev_entry_bit(struct dev_table_entry *dev_table, 996 u16 devid, u8 bit) 997 { 998 int i = (bit >> 6) & 0x03; 999 int _bit = bit & 0x3f; 1000 1001 dev_table[devid].data[i] |= (1UL << _bit); 1002 } 1003 1004 static void set_dev_entry_bit(struct amd_iommu *iommu, u16 devid, u8 bit) 1005 { 1006 struct dev_table_entry *dev_table = get_dev_table(iommu); 1007 1008 return __set_dev_entry_bit(dev_table, devid, bit); 1009 } 1010 1011 static int __get_dev_entry_bit(struct dev_table_entry *dev_table, 1012 u16 devid, u8 bit) 1013 { 1014 int i = (bit >> 6) & 0x03; 1015 int _bit = bit & 0x3f; 1016 1017 return (dev_table[devid].data[i] & (1UL << _bit)) >> _bit; 1018 } 1019 1020 static int get_dev_entry_bit(struct amd_iommu *iommu, u16 devid, u8 bit) 1021 { 1022 struct dev_table_entry *dev_table = get_dev_table(iommu); 1023 1024 return __get_dev_entry_bit(dev_table, devid, bit); 1025 } 1026 1027 static bool __copy_device_table(struct amd_iommu *iommu) 1028 { 1029 u64 int_ctl, int_tab_len, entry = 0; 1030 struct amd_iommu_pci_seg *pci_seg = iommu->pci_seg; 1031 struct dev_table_entry *old_devtb = NULL; 1032 u32 lo, hi, devid, old_devtb_size; 1033 phys_addr_t old_devtb_phys; 1034 u16 dom_id, dte_v, irq_v; 1035 gfp_t gfp_flag; 1036 u64 tmp; 1037 1038 /* Each IOMMU use separate device table with the same size */ 1039 lo = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET); 1040 hi = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET + 4); 1041 entry = (((u64) hi) << 32) + lo; 1042 1043 old_devtb_size = ((entry & ~PAGE_MASK) + 1) << 12; 1044 if (old_devtb_size != pci_seg->dev_table_size) { 1045 pr_err("The device table size of IOMMU:%d is not expected!\n", 1046 iommu->index); 1047 return false; 1048 } 1049 1050 /* 1051 * When SME is enabled in the first kernel, the entry includes the 1052 * memory encryption mask(sme_me_mask), we must remove the memory 1053 * encryption mask to obtain the true physical address in kdump kernel. 1054 */ 1055 old_devtb_phys = __sme_clr(entry) & PAGE_MASK; 1056 1057 if (old_devtb_phys >= 0x100000000ULL) { 1058 pr_err("The address of old device table is above 4G, not trustworthy!\n"); 1059 return false; 1060 } 1061 old_devtb = (cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT) && is_kdump_kernel()) 1062 ? (__force void *)ioremap_encrypted(old_devtb_phys, 1063 pci_seg->dev_table_size) 1064 : memremap(old_devtb_phys, pci_seg->dev_table_size, MEMREMAP_WB); 1065 1066 if (!old_devtb) 1067 return false; 1068 1069 gfp_flag = GFP_KERNEL | __GFP_ZERO | GFP_DMA32; 1070 pci_seg->old_dev_tbl_cpy = (void *)__get_free_pages(gfp_flag, 1071 get_order(pci_seg->dev_table_size)); 1072 if (pci_seg->old_dev_tbl_cpy == NULL) { 1073 pr_err("Failed to allocate memory for copying old device table!\n"); 1074 memunmap(old_devtb); 1075 return false; 1076 } 1077 1078 for (devid = 0; devid <= pci_seg->last_bdf; ++devid) { 1079 pci_seg->old_dev_tbl_cpy[devid] = old_devtb[devid]; 1080 dom_id = old_devtb[devid].data[1] & DEV_DOMID_MASK; 1081 dte_v = old_devtb[devid].data[0] & DTE_FLAG_V; 1082 1083 if (dte_v && dom_id) { 1084 pci_seg->old_dev_tbl_cpy[devid].data[0] = old_devtb[devid].data[0]; 1085 pci_seg->old_dev_tbl_cpy[devid].data[1] = old_devtb[devid].data[1]; 1086 __set_bit(dom_id, amd_iommu_pd_alloc_bitmap); 1087 /* If gcr3 table existed, mask it out */ 1088 if (old_devtb[devid].data[0] & DTE_FLAG_GV) { 1089 tmp = DTE_GCR3_VAL_B(~0ULL) << DTE_GCR3_SHIFT_B; 1090 tmp |= DTE_GCR3_VAL_C(~0ULL) << DTE_GCR3_SHIFT_C; 1091 pci_seg->old_dev_tbl_cpy[devid].data[1] &= ~tmp; 1092 tmp = DTE_GCR3_VAL_A(~0ULL) << DTE_GCR3_SHIFT_A; 1093 tmp |= DTE_FLAG_GV; 1094 pci_seg->old_dev_tbl_cpy[devid].data[0] &= ~tmp; 1095 } 1096 } 1097 1098 irq_v = old_devtb[devid].data[2] & DTE_IRQ_REMAP_ENABLE; 1099 int_ctl = old_devtb[devid].data[2] & DTE_IRQ_REMAP_INTCTL_MASK; 1100 int_tab_len = old_devtb[devid].data[2] & DTE_INTTABLEN_MASK; 1101 if (irq_v && (int_ctl || int_tab_len)) { 1102 if ((int_ctl != DTE_IRQ_REMAP_INTCTL) || 1103 (int_tab_len != DTE_INTTABLEN)) { 1104 pr_err("Wrong old irq remapping flag: %#x\n", devid); 1105 memunmap(old_devtb); 1106 return false; 1107 } 1108 1109 pci_seg->old_dev_tbl_cpy[devid].data[2] = old_devtb[devid].data[2]; 1110 } 1111 } 1112 memunmap(old_devtb); 1113 1114 return true; 1115 } 1116 1117 static bool copy_device_table(void) 1118 { 1119 struct amd_iommu *iommu; 1120 struct amd_iommu_pci_seg *pci_seg; 1121 1122 if (!amd_iommu_pre_enabled) 1123 return false; 1124 1125 pr_warn("Translation is already enabled - trying to copy translation structures\n"); 1126 1127 /* 1128 * All IOMMUs within PCI segment shares common device table. 1129 * Hence copy device table only once per PCI segment. 1130 */ 1131 for_each_pci_segment(pci_seg) { 1132 for_each_iommu(iommu) { 1133 if (pci_seg->id != iommu->pci_seg->id) 1134 continue; 1135 if (!__copy_device_table(iommu)) 1136 return false; 1137 break; 1138 } 1139 } 1140 1141 return true; 1142 } 1143 1144 void amd_iommu_apply_erratum_63(struct amd_iommu *iommu, u16 devid) 1145 { 1146 int sysmgt; 1147 1148 sysmgt = get_dev_entry_bit(iommu, devid, DEV_ENTRY_SYSMGT1) | 1149 (get_dev_entry_bit(iommu, devid, DEV_ENTRY_SYSMGT2) << 1); 1150 1151 if (sysmgt == 0x01) 1152 set_dev_entry_bit(iommu, devid, DEV_ENTRY_IW); 1153 } 1154 1155 /* 1156 * This function takes the device specific flags read from the ACPI 1157 * table and sets up the device table entry with that information 1158 */ 1159 static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu, 1160 u16 devid, u32 flags, u32 ext_flags) 1161 { 1162 if (flags & ACPI_DEVFLAG_INITPASS) 1163 set_dev_entry_bit(iommu, devid, DEV_ENTRY_INIT_PASS); 1164 if (flags & ACPI_DEVFLAG_EXTINT) 1165 set_dev_entry_bit(iommu, devid, DEV_ENTRY_EINT_PASS); 1166 if (flags & ACPI_DEVFLAG_NMI) 1167 set_dev_entry_bit(iommu, devid, DEV_ENTRY_NMI_PASS); 1168 if (flags & ACPI_DEVFLAG_SYSMGT1) 1169 set_dev_entry_bit(iommu, devid, DEV_ENTRY_SYSMGT1); 1170 if (flags & ACPI_DEVFLAG_SYSMGT2) 1171 set_dev_entry_bit(iommu, devid, DEV_ENTRY_SYSMGT2); 1172 if (flags & ACPI_DEVFLAG_LINT0) 1173 set_dev_entry_bit(iommu, devid, DEV_ENTRY_LINT0_PASS); 1174 if (flags & ACPI_DEVFLAG_LINT1) 1175 set_dev_entry_bit(iommu, devid, DEV_ENTRY_LINT1_PASS); 1176 1177 amd_iommu_apply_erratum_63(iommu, devid); 1178 1179 amd_iommu_set_rlookup_table(iommu, devid); 1180 } 1181 1182 int __init add_special_device(u8 type, u8 id, u32 *devid, bool cmd_line) 1183 { 1184 struct devid_map *entry; 1185 struct list_head *list; 1186 1187 if (type == IVHD_SPECIAL_IOAPIC) 1188 list = &ioapic_map; 1189 else if (type == IVHD_SPECIAL_HPET) 1190 list = &hpet_map; 1191 else 1192 return -EINVAL; 1193 1194 list_for_each_entry(entry, list, list) { 1195 if (!(entry->id == id && entry->cmd_line)) 1196 continue; 1197 1198 pr_info("Command-line override present for %s id %d - ignoring\n", 1199 type == IVHD_SPECIAL_IOAPIC ? "IOAPIC" : "HPET", id); 1200 1201 *devid = entry->devid; 1202 1203 return 0; 1204 } 1205 1206 entry = kzalloc(sizeof(*entry), GFP_KERNEL); 1207 if (!entry) 1208 return -ENOMEM; 1209 1210 entry->id = id; 1211 entry->devid = *devid; 1212 entry->cmd_line = cmd_line; 1213 1214 list_add_tail(&entry->list, list); 1215 1216 return 0; 1217 } 1218 1219 static int __init add_acpi_hid_device(u8 *hid, u8 *uid, u32 *devid, 1220 bool cmd_line) 1221 { 1222 struct acpihid_map_entry *entry; 1223 struct list_head *list = &acpihid_map; 1224 1225 list_for_each_entry(entry, list, list) { 1226 if (strcmp(entry->hid, hid) || 1227 (*uid && *entry->uid && strcmp(entry->uid, uid)) || 1228 !entry->cmd_line) 1229 continue; 1230 1231 pr_info("Command-line override for hid:%s uid:%s\n", 1232 hid, uid); 1233 *devid = entry->devid; 1234 return 0; 1235 } 1236 1237 entry = kzalloc(sizeof(*entry), GFP_KERNEL); 1238 if (!entry) 1239 return -ENOMEM; 1240 1241 memcpy(entry->uid, uid, strlen(uid)); 1242 memcpy(entry->hid, hid, strlen(hid)); 1243 entry->devid = *devid; 1244 entry->cmd_line = cmd_line; 1245 entry->root_devid = (entry->devid & (~0x7)); 1246 1247 pr_info("%s, add hid:%s, uid:%s, rdevid:%d\n", 1248 entry->cmd_line ? "cmd" : "ivrs", 1249 entry->hid, entry->uid, entry->root_devid); 1250 1251 list_add_tail(&entry->list, list); 1252 return 0; 1253 } 1254 1255 static int __init add_early_maps(void) 1256 { 1257 int i, ret; 1258 1259 for (i = 0; i < early_ioapic_map_size; ++i) { 1260 ret = add_special_device(IVHD_SPECIAL_IOAPIC, 1261 early_ioapic_map[i].id, 1262 &early_ioapic_map[i].devid, 1263 early_ioapic_map[i].cmd_line); 1264 if (ret) 1265 return ret; 1266 } 1267 1268 for (i = 0; i < early_hpet_map_size; ++i) { 1269 ret = add_special_device(IVHD_SPECIAL_HPET, 1270 early_hpet_map[i].id, 1271 &early_hpet_map[i].devid, 1272 early_hpet_map[i].cmd_line); 1273 if (ret) 1274 return ret; 1275 } 1276 1277 for (i = 0; i < early_acpihid_map_size; ++i) { 1278 ret = add_acpi_hid_device(early_acpihid_map[i].hid, 1279 early_acpihid_map[i].uid, 1280 &early_acpihid_map[i].devid, 1281 early_acpihid_map[i].cmd_line); 1282 if (ret) 1283 return ret; 1284 } 1285 1286 return 0; 1287 } 1288 1289 /* 1290 * Takes a pointer to an AMD IOMMU entry in the ACPI table and 1291 * initializes the hardware and our data structures with it. 1292 */ 1293 static int __init init_iommu_from_acpi(struct amd_iommu *iommu, 1294 struct ivhd_header *h) 1295 { 1296 u8 *p = (u8 *)h; 1297 u8 *end = p, flags = 0; 1298 u16 devid = 0, devid_start = 0, devid_to = 0, seg_id; 1299 u32 dev_i, ext_flags = 0; 1300 bool alias = false; 1301 struct ivhd_entry *e; 1302 struct amd_iommu_pci_seg *pci_seg = iommu->pci_seg; 1303 u32 ivhd_size; 1304 int ret; 1305 1306 1307 ret = add_early_maps(); 1308 if (ret) 1309 return ret; 1310 1311 amd_iommu_apply_ivrs_quirks(); 1312 1313 /* 1314 * First save the recommended feature enable bits from ACPI 1315 */ 1316 iommu->acpi_flags = h->flags; 1317 1318 /* 1319 * Done. Now parse the device entries 1320 */ 1321 ivhd_size = get_ivhd_header_size(h); 1322 if (!ivhd_size) { 1323 pr_err("Unsupported IVHD type %#x\n", h->type); 1324 return -EINVAL; 1325 } 1326 1327 p += ivhd_size; 1328 1329 end += h->length; 1330 1331 1332 while (p < end) { 1333 e = (struct ivhd_entry *)p; 1334 seg_id = pci_seg->id; 1335 1336 switch (e->type) { 1337 case IVHD_DEV_ALL: 1338 1339 DUMP_printk(" DEV_ALL\t\t\tflags: %02x\n", e->flags); 1340 1341 for (dev_i = 0; dev_i <= pci_seg->last_bdf; ++dev_i) 1342 set_dev_entry_from_acpi(iommu, dev_i, e->flags, 0); 1343 break; 1344 case IVHD_DEV_SELECT: 1345 1346 DUMP_printk(" DEV_SELECT\t\t\t devid: %04x:%02x:%02x.%x " 1347 "flags: %02x\n", 1348 seg_id, PCI_BUS_NUM(e->devid), 1349 PCI_SLOT(e->devid), 1350 PCI_FUNC(e->devid), 1351 e->flags); 1352 1353 devid = e->devid; 1354 set_dev_entry_from_acpi(iommu, devid, e->flags, 0); 1355 break; 1356 case IVHD_DEV_SELECT_RANGE_START: 1357 1358 DUMP_printk(" DEV_SELECT_RANGE_START\t " 1359 "devid: %04x:%02x:%02x.%x flags: %02x\n", 1360 seg_id, PCI_BUS_NUM(e->devid), 1361 PCI_SLOT(e->devid), 1362 PCI_FUNC(e->devid), 1363 e->flags); 1364 1365 devid_start = e->devid; 1366 flags = e->flags; 1367 ext_flags = 0; 1368 alias = false; 1369 break; 1370 case IVHD_DEV_ALIAS: 1371 1372 DUMP_printk(" DEV_ALIAS\t\t\t devid: %04x:%02x:%02x.%x " 1373 "flags: %02x devid_to: %02x:%02x.%x\n", 1374 seg_id, PCI_BUS_NUM(e->devid), 1375 PCI_SLOT(e->devid), 1376 PCI_FUNC(e->devid), 1377 e->flags, 1378 PCI_BUS_NUM(e->ext >> 8), 1379 PCI_SLOT(e->ext >> 8), 1380 PCI_FUNC(e->ext >> 8)); 1381 1382 devid = e->devid; 1383 devid_to = e->ext >> 8; 1384 set_dev_entry_from_acpi(iommu, devid , e->flags, 0); 1385 set_dev_entry_from_acpi(iommu, devid_to, e->flags, 0); 1386 pci_seg->alias_table[devid] = devid_to; 1387 break; 1388 case IVHD_DEV_ALIAS_RANGE: 1389 1390 DUMP_printk(" DEV_ALIAS_RANGE\t\t " 1391 "devid: %04x:%02x:%02x.%x flags: %02x " 1392 "devid_to: %04x:%02x:%02x.%x\n", 1393 seg_id, PCI_BUS_NUM(e->devid), 1394 PCI_SLOT(e->devid), 1395 PCI_FUNC(e->devid), 1396 e->flags, 1397 seg_id, PCI_BUS_NUM(e->ext >> 8), 1398 PCI_SLOT(e->ext >> 8), 1399 PCI_FUNC(e->ext >> 8)); 1400 1401 devid_start = e->devid; 1402 flags = e->flags; 1403 devid_to = e->ext >> 8; 1404 ext_flags = 0; 1405 alias = true; 1406 break; 1407 case IVHD_DEV_EXT_SELECT: 1408 1409 DUMP_printk(" DEV_EXT_SELECT\t\t devid: %04x:%02x:%02x.%x " 1410 "flags: %02x ext: %08x\n", 1411 seg_id, PCI_BUS_NUM(e->devid), 1412 PCI_SLOT(e->devid), 1413 PCI_FUNC(e->devid), 1414 e->flags, e->ext); 1415 1416 devid = e->devid; 1417 set_dev_entry_from_acpi(iommu, devid, e->flags, 1418 e->ext); 1419 break; 1420 case IVHD_DEV_EXT_SELECT_RANGE: 1421 1422 DUMP_printk(" DEV_EXT_SELECT_RANGE\t devid: " 1423 "%04x:%02x:%02x.%x flags: %02x ext: %08x\n", 1424 seg_id, PCI_BUS_NUM(e->devid), 1425 PCI_SLOT(e->devid), 1426 PCI_FUNC(e->devid), 1427 e->flags, e->ext); 1428 1429 devid_start = e->devid; 1430 flags = e->flags; 1431 ext_flags = e->ext; 1432 alias = false; 1433 break; 1434 case IVHD_DEV_RANGE_END: 1435 1436 DUMP_printk(" DEV_RANGE_END\t\t devid: %04x:%02x:%02x.%x\n", 1437 seg_id, PCI_BUS_NUM(e->devid), 1438 PCI_SLOT(e->devid), 1439 PCI_FUNC(e->devid)); 1440 1441 devid = e->devid; 1442 for (dev_i = devid_start; dev_i <= devid; ++dev_i) { 1443 if (alias) { 1444 pci_seg->alias_table[dev_i] = devid_to; 1445 set_dev_entry_from_acpi(iommu, 1446 devid_to, flags, ext_flags); 1447 } 1448 set_dev_entry_from_acpi(iommu, dev_i, 1449 flags, ext_flags); 1450 } 1451 break; 1452 case IVHD_DEV_SPECIAL: { 1453 u8 handle, type; 1454 const char *var; 1455 u32 devid; 1456 int ret; 1457 1458 handle = e->ext & 0xff; 1459 devid = PCI_SEG_DEVID_TO_SBDF(seg_id, (e->ext >> 8)); 1460 type = (e->ext >> 24) & 0xff; 1461 1462 if (type == IVHD_SPECIAL_IOAPIC) 1463 var = "IOAPIC"; 1464 else if (type == IVHD_SPECIAL_HPET) 1465 var = "HPET"; 1466 else 1467 var = "UNKNOWN"; 1468 1469 DUMP_printk(" DEV_SPECIAL(%s[%d])\t\tdevid: %04x:%02x:%02x.%x\n", 1470 var, (int)handle, 1471 seg_id, PCI_BUS_NUM(devid), 1472 PCI_SLOT(devid), 1473 PCI_FUNC(devid)); 1474 1475 ret = add_special_device(type, handle, &devid, false); 1476 if (ret) 1477 return ret; 1478 1479 /* 1480 * add_special_device might update the devid in case a 1481 * command-line override is present. So call 1482 * set_dev_entry_from_acpi after add_special_device. 1483 */ 1484 set_dev_entry_from_acpi(iommu, devid, e->flags, 0); 1485 1486 break; 1487 } 1488 case IVHD_DEV_ACPI_HID: { 1489 u32 devid; 1490 u8 hid[ACPIHID_HID_LEN]; 1491 u8 uid[ACPIHID_UID_LEN]; 1492 int ret; 1493 1494 if (h->type != 0x40) { 1495 pr_err(FW_BUG "Invalid IVHD device type %#x\n", 1496 e->type); 1497 break; 1498 } 1499 1500 BUILD_BUG_ON(sizeof(e->ext_hid) != ACPIHID_HID_LEN - 1); 1501 memcpy(hid, &e->ext_hid, ACPIHID_HID_LEN - 1); 1502 hid[ACPIHID_HID_LEN - 1] = '\0'; 1503 1504 if (!(*hid)) { 1505 pr_err(FW_BUG "Invalid HID.\n"); 1506 break; 1507 } 1508 1509 uid[0] = '\0'; 1510 switch (e->uidf) { 1511 case UID_NOT_PRESENT: 1512 1513 if (e->uidl != 0) 1514 pr_warn(FW_BUG "Invalid UID length.\n"); 1515 1516 break; 1517 case UID_IS_INTEGER: 1518 1519 sprintf(uid, "%d", e->uid); 1520 1521 break; 1522 case UID_IS_CHARACTER: 1523 1524 memcpy(uid, &e->uid, e->uidl); 1525 uid[e->uidl] = '\0'; 1526 1527 break; 1528 default: 1529 break; 1530 } 1531 1532 devid = PCI_SEG_DEVID_TO_SBDF(seg_id, e->devid); 1533 DUMP_printk(" DEV_ACPI_HID(%s[%s])\t\tdevid: %04x:%02x:%02x.%x\n", 1534 hid, uid, seg_id, 1535 PCI_BUS_NUM(devid), 1536 PCI_SLOT(devid), 1537 PCI_FUNC(devid)); 1538 1539 flags = e->flags; 1540 1541 ret = add_acpi_hid_device(hid, uid, &devid, false); 1542 if (ret) 1543 return ret; 1544 1545 /* 1546 * add_special_device might update the devid in case a 1547 * command-line override is present. So call 1548 * set_dev_entry_from_acpi after add_special_device. 1549 */ 1550 set_dev_entry_from_acpi(iommu, devid, e->flags, 0); 1551 1552 break; 1553 } 1554 default: 1555 break; 1556 } 1557 1558 p += ivhd_entry_length(p); 1559 } 1560 1561 return 0; 1562 } 1563 1564 /* Allocate PCI segment data structure */ 1565 static struct amd_iommu_pci_seg *__init alloc_pci_segment(u16 id, 1566 struct acpi_table_header *ivrs_base) 1567 { 1568 struct amd_iommu_pci_seg *pci_seg; 1569 int last_bdf; 1570 1571 /* 1572 * First parse ACPI tables to find the largest Bus/Dev/Func we need to 1573 * handle in this PCI segment. Upon this information the shared data 1574 * structures for the PCI segments in the system will be allocated. 1575 */ 1576 last_bdf = find_last_devid_acpi(ivrs_base, id); 1577 if (last_bdf < 0) 1578 return NULL; 1579 1580 pci_seg = kzalloc(sizeof(struct amd_iommu_pci_seg), GFP_KERNEL); 1581 if (pci_seg == NULL) 1582 return NULL; 1583 1584 pci_seg->last_bdf = last_bdf; 1585 DUMP_printk("PCI segment : 0x%0x, last bdf : 0x%04x\n", id, last_bdf); 1586 pci_seg->dev_table_size = tbl_size(DEV_TABLE_ENTRY_SIZE, last_bdf); 1587 pci_seg->alias_table_size = tbl_size(ALIAS_TABLE_ENTRY_SIZE, last_bdf); 1588 pci_seg->rlookup_table_size = tbl_size(RLOOKUP_TABLE_ENTRY_SIZE, last_bdf); 1589 1590 pci_seg->id = id; 1591 init_llist_head(&pci_seg->dev_data_list); 1592 INIT_LIST_HEAD(&pci_seg->unity_map); 1593 list_add_tail(&pci_seg->list, &amd_iommu_pci_seg_list); 1594 1595 if (alloc_dev_table(pci_seg)) 1596 return NULL; 1597 if (alloc_alias_table(pci_seg)) 1598 return NULL; 1599 if (alloc_rlookup_table(pci_seg)) 1600 return NULL; 1601 1602 return pci_seg; 1603 } 1604 1605 static struct amd_iommu_pci_seg *__init get_pci_segment(u16 id, 1606 struct acpi_table_header *ivrs_base) 1607 { 1608 struct amd_iommu_pci_seg *pci_seg; 1609 1610 for_each_pci_segment(pci_seg) { 1611 if (pci_seg->id == id) 1612 return pci_seg; 1613 } 1614 1615 return alloc_pci_segment(id, ivrs_base); 1616 } 1617 1618 static void __init free_pci_segments(void) 1619 { 1620 struct amd_iommu_pci_seg *pci_seg, *next; 1621 1622 for_each_pci_segment_safe(pci_seg, next) { 1623 list_del(&pci_seg->list); 1624 free_irq_lookup_table(pci_seg); 1625 free_rlookup_table(pci_seg); 1626 free_alias_table(pci_seg); 1627 free_dev_table(pci_seg); 1628 kfree(pci_seg); 1629 } 1630 } 1631 1632 static void __init free_iommu_one(struct amd_iommu *iommu) 1633 { 1634 free_cwwb_sem(iommu); 1635 free_command_buffer(iommu); 1636 free_event_buffer(iommu); 1637 free_ppr_log(iommu); 1638 free_ga_log(iommu); 1639 iommu_unmap_mmio_space(iommu); 1640 } 1641 1642 static void __init free_iommu_all(void) 1643 { 1644 struct amd_iommu *iommu, *next; 1645 1646 for_each_iommu_safe(iommu, next) { 1647 list_del(&iommu->list); 1648 free_iommu_one(iommu); 1649 kfree(iommu); 1650 } 1651 } 1652 1653 /* 1654 * Family15h Model 10h-1fh erratum 746 (IOMMU Logging May Stall Translations) 1655 * Workaround: 1656 * BIOS should disable L2B micellaneous clock gating by setting 1657 * L2_L2B_CK_GATE_CONTROL[CKGateL2BMiscDisable](D0F2xF4_x90[2]) = 1b 1658 */ 1659 static void amd_iommu_erratum_746_workaround(struct amd_iommu *iommu) 1660 { 1661 u32 value; 1662 1663 if ((boot_cpu_data.x86 != 0x15) || 1664 (boot_cpu_data.x86_model < 0x10) || 1665 (boot_cpu_data.x86_model > 0x1f)) 1666 return; 1667 1668 pci_write_config_dword(iommu->dev, 0xf0, 0x90); 1669 pci_read_config_dword(iommu->dev, 0xf4, &value); 1670 1671 if (value & BIT(2)) 1672 return; 1673 1674 /* Select NB indirect register 0x90 and enable writing */ 1675 pci_write_config_dword(iommu->dev, 0xf0, 0x90 | (1 << 8)); 1676 1677 pci_write_config_dword(iommu->dev, 0xf4, value | 0x4); 1678 pci_info(iommu->dev, "Applying erratum 746 workaround\n"); 1679 1680 /* Clear the enable writing bit */ 1681 pci_write_config_dword(iommu->dev, 0xf0, 0x90); 1682 } 1683 1684 /* 1685 * Family15h Model 30h-3fh (IOMMU Mishandles ATS Write Permission) 1686 * Workaround: 1687 * BIOS should enable ATS write permission check by setting 1688 * L2_DEBUG_3[AtsIgnoreIWDis](D0F2xF4_x47[0]) = 1b 1689 */ 1690 static void amd_iommu_ats_write_check_workaround(struct amd_iommu *iommu) 1691 { 1692 u32 value; 1693 1694 if ((boot_cpu_data.x86 != 0x15) || 1695 (boot_cpu_data.x86_model < 0x30) || 1696 (boot_cpu_data.x86_model > 0x3f)) 1697 return; 1698 1699 /* Test L2_DEBUG_3[AtsIgnoreIWDis] == 1 */ 1700 value = iommu_read_l2(iommu, 0x47); 1701 1702 if (value & BIT(0)) 1703 return; 1704 1705 /* Set L2_DEBUG_3[AtsIgnoreIWDis] = 1 */ 1706 iommu_write_l2(iommu, 0x47, value | BIT(0)); 1707 1708 pci_info(iommu->dev, "Applying ATS write check workaround\n"); 1709 } 1710 1711 /* 1712 * This function glues the initialization function for one IOMMU 1713 * together and also allocates the command buffer and programs the 1714 * hardware. It does NOT enable the IOMMU. This is done afterwards. 1715 */ 1716 static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h, 1717 struct acpi_table_header *ivrs_base) 1718 { 1719 struct amd_iommu_pci_seg *pci_seg; 1720 1721 pci_seg = get_pci_segment(h->pci_seg, ivrs_base); 1722 if (pci_seg == NULL) 1723 return -ENOMEM; 1724 iommu->pci_seg = pci_seg; 1725 1726 raw_spin_lock_init(&iommu->lock); 1727 iommu->cmd_sem_val = 0; 1728 1729 /* Add IOMMU to internal data structures */ 1730 list_add_tail(&iommu->list, &amd_iommu_list); 1731 iommu->index = amd_iommus_present++; 1732 1733 if (unlikely(iommu->index >= MAX_IOMMUS)) { 1734 WARN(1, "System has more IOMMUs than supported by this driver\n"); 1735 return -ENOSYS; 1736 } 1737 1738 /* Index is fine - add IOMMU to the array */ 1739 amd_iommus[iommu->index] = iommu; 1740 1741 /* 1742 * Copy data from ACPI table entry to the iommu struct 1743 */ 1744 iommu->devid = h->devid; 1745 iommu->cap_ptr = h->cap_ptr; 1746 iommu->mmio_phys = h->mmio_phys; 1747 1748 switch (h->type) { 1749 case 0x10: 1750 /* Check if IVHD EFR contains proper max banks/counters */ 1751 if ((h->efr_attr != 0) && 1752 ((h->efr_attr & (0xF << 13)) != 0) && 1753 ((h->efr_attr & (0x3F << 17)) != 0)) 1754 iommu->mmio_phys_end = MMIO_REG_END_OFFSET; 1755 else 1756 iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET; 1757 1758 /* 1759 * Note: GA (128-bit IRTE) mode requires cmpxchg16b supports. 1760 * GAM also requires GA mode. Therefore, we need to 1761 * check cmpxchg16b support before enabling it. 1762 */ 1763 if (!boot_cpu_has(X86_FEATURE_CX16) || 1764 ((h->efr_attr & (0x1 << IOMMU_FEAT_GASUP_SHIFT)) == 0)) 1765 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY; 1766 break; 1767 case 0x11: 1768 case 0x40: 1769 if (h->efr_reg & (1 << 9)) 1770 iommu->mmio_phys_end = MMIO_REG_END_OFFSET; 1771 else 1772 iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET; 1773 1774 /* 1775 * Note: GA (128-bit IRTE) mode requires cmpxchg16b supports. 1776 * XT, GAM also requires GA mode. Therefore, we need to 1777 * check cmpxchg16b support before enabling them. 1778 */ 1779 if (!boot_cpu_has(X86_FEATURE_CX16) || 1780 ((h->efr_reg & (0x1 << IOMMU_EFR_GASUP_SHIFT)) == 0)) { 1781 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY; 1782 break; 1783 } 1784 1785 if (h->efr_reg & BIT(IOMMU_EFR_XTSUP_SHIFT)) 1786 amd_iommu_xt_mode = IRQ_REMAP_X2APIC_MODE; 1787 1788 early_iommu_features_init(iommu, h); 1789 1790 break; 1791 default: 1792 return -EINVAL; 1793 } 1794 1795 iommu->mmio_base = iommu_map_mmio_space(iommu->mmio_phys, 1796 iommu->mmio_phys_end); 1797 if (!iommu->mmio_base) 1798 return -ENOMEM; 1799 1800 return init_iommu_from_acpi(iommu, h); 1801 } 1802 1803 static int __init init_iommu_one_late(struct amd_iommu *iommu) 1804 { 1805 int ret; 1806 1807 if (alloc_cwwb_sem(iommu)) 1808 return -ENOMEM; 1809 1810 if (alloc_command_buffer(iommu)) 1811 return -ENOMEM; 1812 1813 if (alloc_event_buffer(iommu)) 1814 return -ENOMEM; 1815 1816 iommu->int_enabled = false; 1817 1818 init_translation_status(iommu); 1819 if (translation_pre_enabled(iommu) && !is_kdump_kernel()) { 1820 iommu_disable(iommu); 1821 clear_translation_pre_enabled(iommu); 1822 pr_warn("Translation was enabled for IOMMU:%d but we are not in kdump mode\n", 1823 iommu->index); 1824 } 1825 if (amd_iommu_pre_enabled) 1826 amd_iommu_pre_enabled = translation_pre_enabled(iommu); 1827 1828 if (amd_iommu_irq_remap) { 1829 ret = amd_iommu_create_irq_domain(iommu); 1830 if (ret) 1831 return ret; 1832 } 1833 1834 /* 1835 * Make sure IOMMU is not considered to translate itself. The IVRS 1836 * table tells us so, but this is a lie! 1837 */ 1838 iommu->pci_seg->rlookup_table[iommu->devid] = NULL; 1839 1840 return 0; 1841 } 1842 1843 /** 1844 * get_highest_supported_ivhd_type - Look up the appropriate IVHD type 1845 * @ivrs: Pointer to the IVRS header 1846 * 1847 * This function search through all IVDB of the maximum supported IVHD 1848 */ 1849 static u8 get_highest_supported_ivhd_type(struct acpi_table_header *ivrs) 1850 { 1851 u8 *base = (u8 *)ivrs; 1852 struct ivhd_header *ivhd = (struct ivhd_header *) 1853 (base + IVRS_HEADER_LENGTH); 1854 u8 last_type = ivhd->type; 1855 u16 devid = ivhd->devid; 1856 1857 while (((u8 *)ivhd - base < ivrs->length) && 1858 (ivhd->type <= ACPI_IVHD_TYPE_MAX_SUPPORTED)) { 1859 u8 *p = (u8 *) ivhd; 1860 1861 if (ivhd->devid == devid) 1862 last_type = ivhd->type; 1863 ivhd = (struct ivhd_header *)(p + ivhd->length); 1864 } 1865 1866 return last_type; 1867 } 1868 1869 /* 1870 * Iterates over all IOMMU entries in the ACPI table, allocates the 1871 * IOMMU structure and initializes it with init_iommu_one() 1872 */ 1873 static int __init init_iommu_all(struct acpi_table_header *table) 1874 { 1875 u8 *p = (u8 *)table, *end = (u8 *)table; 1876 struct ivhd_header *h; 1877 struct amd_iommu *iommu; 1878 int ret; 1879 1880 end += table->length; 1881 p += IVRS_HEADER_LENGTH; 1882 1883 /* Phase 1: Process all IVHD blocks */ 1884 while (p < end) { 1885 h = (struct ivhd_header *)p; 1886 if (*p == amd_iommu_target_ivhd_type) { 1887 1888 DUMP_printk("device: %04x:%02x:%02x.%01x cap: %04x " 1889 "flags: %01x info %04x\n", 1890 h->pci_seg, PCI_BUS_NUM(h->devid), 1891 PCI_SLOT(h->devid), PCI_FUNC(h->devid), 1892 h->cap_ptr, h->flags, h->info); 1893 DUMP_printk(" mmio-addr: %016llx\n", 1894 h->mmio_phys); 1895 1896 iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL); 1897 if (iommu == NULL) 1898 return -ENOMEM; 1899 1900 ret = init_iommu_one(iommu, h, table); 1901 if (ret) 1902 return ret; 1903 } 1904 p += h->length; 1905 1906 } 1907 WARN_ON(p != end); 1908 1909 /* Phase 2 : Early feature support check */ 1910 get_global_efr(); 1911 1912 /* Phase 3 : Enabling IOMMU features */ 1913 for_each_iommu(iommu) { 1914 ret = init_iommu_one_late(iommu); 1915 if (ret) 1916 return ret; 1917 } 1918 1919 return 0; 1920 } 1921 1922 static void init_iommu_perf_ctr(struct amd_iommu *iommu) 1923 { 1924 u64 val; 1925 struct pci_dev *pdev = iommu->dev; 1926 1927 if (!iommu_feature(iommu, FEATURE_PC)) 1928 return; 1929 1930 amd_iommu_pc_present = true; 1931 1932 pci_info(pdev, "IOMMU performance counters supported\n"); 1933 1934 val = readl(iommu->mmio_base + MMIO_CNTR_CONF_OFFSET); 1935 iommu->max_banks = (u8) ((val >> 12) & 0x3f); 1936 iommu->max_counters = (u8) ((val >> 7) & 0xf); 1937 1938 return; 1939 } 1940 1941 static ssize_t amd_iommu_show_cap(struct device *dev, 1942 struct device_attribute *attr, 1943 char *buf) 1944 { 1945 struct amd_iommu *iommu = dev_to_amd_iommu(dev); 1946 return sprintf(buf, "%x\n", iommu->cap); 1947 } 1948 static DEVICE_ATTR(cap, S_IRUGO, amd_iommu_show_cap, NULL); 1949 1950 static ssize_t amd_iommu_show_features(struct device *dev, 1951 struct device_attribute *attr, 1952 char *buf) 1953 { 1954 struct amd_iommu *iommu = dev_to_amd_iommu(dev); 1955 return sprintf(buf, "%llx:%llx\n", iommu->features2, iommu->features); 1956 } 1957 static DEVICE_ATTR(features, S_IRUGO, amd_iommu_show_features, NULL); 1958 1959 static struct attribute *amd_iommu_attrs[] = { 1960 &dev_attr_cap.attr, 1961 &dev_attr_features.attr, 1962 NULL, 1963 }; 1964 1965 static struct attribute_group amd_iommu_group = { 1966 .name = "amd-iommu", 1967 .attrs = amd_iommu_attrs, 1968 }; 1969 1970 static const struct attribute_group *amd_iommu_groups[] = { 1971 &amd_iommu_group, 1972 NULL, 1973 }; 1974 1975 /* 1976 * Note: IVHD 0x11 and 0x40 also contains exact copy 1977 * of the IOMMU Extended Feature Register [MMIO Offset 0030h]. 1978 * Default to EFR in IVHD since it is available sooner (i.e. before PCI init). 1979 */ 1980 static void __init late_iommu_features_init(struct amd_iommu *iommu) 1981 { 1982 u64 features, features2; 1983 1984 if (!(iommu->cap & (1 << IOMMU_CAP_EFR))) 1985 return; 1986 1987 /* read extended feature bits */ 1988 features = readq(iommu->mmio_base + MMIO_EXT_FEATURES); 1989 features2 = readq(iommu->mmio_base + MMIO_EXT_FEATURES2); 1990 1991 if (!iommu->features) { 1992 iommu->features = features; 1993 iommu->features2 = features2; 1994 return; 1995 } 1996 1997 /* 1998 * Sanity check and warn if EFR values from 1999 * IVHD and MMIO conflict. 2000 */ 2001 if (features != iommu->features || 2002 features2 != iommu->features2) { 2003 pr_warn(FW_WARN 2004 "EFR mismatch. Use IVHD EFR (%#llx : %#llx), EFR2 (%#llx : %#llx).\n", 2005 features, iommu->features, 2006 features2, iommu->features2); 2007 } 2008 } 2009 2010 static int __init iommu_init_pci(struct amd_iommu *iommu) 2011 { 2012 int cap_ptr = iommu->cap_ptr; 2013 int ret; 2014 2015 iommu->dev = pci_get_domain_bus_and_slot(iommu->pci_seg->id, 2016 PCI_BUS_NUM(iommu->devid), 2017 iommu->devid & 0xff); 2018 if (!iommu->dev) 2019 return -ENODEV; 2020 2021 /* Prevent binding other PCI device drivers to IOMMU devices */ 2022 iommu->dev->match_driver = false; 2023 2024 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET, 2025 &iommu->cap); 2026 2027 if (!(iommu->cap & (1 << IOMMU_CAP_IOTLB))) 2028 amd_iommu_iotlb_sup = false; 2029 2030 late_iommu_features_init(iommu); 2031 2032 if (iommu_feature(iommu, FEATURE_GT)) { 2033 int glxval; 2034 u32 max_pasid; 2035 u64 pasmax; 2036 2037 pasmax = iommu->features & FEATURE_PASID_MASK; 2038 pasmax >>= FEATURE_PASID_SHIFT; 2039 max_pasid = (1 << (pasmax + 1)) - 1; 2040 2041 amd_iommu_max_pasid = min(amd_iommu_max_pasid, max_pasid); 2042 2043 BUG_ON(amd_iommu_max_pasid & ~PASID_MASK); 2044 2045 glxval = iommu->features & FEATURE_GLXVAL_MASK; 2046 glxval >>= FEATURE_GLXVAL_SHIFT; 2047 2048 if (amd_iommu_max_glx_val == -1) 2049 amd_iommu_max_glx_val = glxval; 2050 else 2051 amd_iommu_max_glx_val = min(amd_iommu_max_glx_val, glxval); 2052 } 2053 2054 if (iommu_feature(iommu, FEATURE_GT) && 2055 iommu_feature(iommu, FEATURE_PPR)) { 2056 iommu->is_iommu_v2 = true; 2057 amd_iommu_v2_present = true; 2058 } 2059 2060 if (iommu_feature(iommu, FEATURE_PPR) && alloc_ppr_log(iommu)) 2061 return -ENOMEM; 2062 2063 if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE)) { 2064 pr_info("Using strict mode due to virtualization\n"); 2065 iommu_set_dma_strict(); 2066 amd_iommu_np_cache = true; 2067 } 2068 2069 init_iommu_perf_ctr(iommu); 2070 2071 if (amd_iommu_pgtable == AMD_IOMMU_V2) { 2072 if (!iommu_feature(iommu, FEATURE_GIOSUP) || 2073 !iommu_feature(iommu, FEATURE_GT)) { 2074 pr_warn("Cannot enable v2 page table for DMA-API. Fallback to v1.\n"); 2075 amd_iommu_pgtable = AMD_IOMMU_V1; 2076 } else if (iommu_default_passthrough()) { 2077 pr_warn("V2 page table doesn't support passthrough mode. Fallback to v1.\n"); 2078 amd_iommu_pgtable = AMD_IOMMU_V1; 2079 } 2080 } 2081 2082 if (is_rd890_iommu(iommu->dev)) { 2083 int i, j; 2084 2085 iommu->root_pdev = 2086 pci_get_domain_bus_and_slot(iommu->pci_seg->id, 2087 iommu->dev->bus->number, 2088 PCI_DEVFN(0, 0)); 2089 2090 /* 2091 * Some rd890 systems may not be fully reconfigured by the 2092 * BIOS, so it's necessary for us to store this information so 2093 * it can be reprogrammed on resume 2094 */ 2095 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 4, 2096 &iommu->stored_addr_lo); 2097 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 8, 2098 &iommu->stored_addr_hi); 2099 2100 /* Low bit locks writes to configuration space */ 2101 iommu->stored_addr_lo &= ~1; 2102 2103 for (i = 0; i < 6; i++) 2104 for (j = 0; j < 0x12; j++) 2105 iommu->stored_l1[i][j] = iommu_read_l1(iommu, i, j); 2106 2107 for (i = 0; i < 0x83; i++) 2108 iommu->stored_l2[i] = iommu_read_l2(iommu, i); 2109 } 2110 2111 amd_iommu_erratum_746_workaround(iommu); 2112 amd_iommu_ats_write_check_workaround(iommu); 2113 2114 ret = iommu_device_sysfs_add(&iommu->iommu, &iommu->dev->dev, 2115 amd_iommu_groups, "ivhd%d", iommu->index); 2116 if (ret) 2117 return ret; 2118 2119 iommu_device_register(&iommu->iommu, &amd_iommu_ops, NULL); 2120 2121 return pci_enable_device(iommu->dev); 2122 } 2123 2124 static void print_iommu_info(void) 2125 { 2126 static const char * const feat_str[] = { 2127 "PreF", "PPR", "X2APIC", "NX", "GT", "[5]", 2128 "IA", "GA", "HE", "PC" 2129 }; 2130 struct amd_iommu *iommu; 2131 2132 for_each_iommu(iommu) { 2133 struct pci_dev *pdev = iommu->dev; 2134 int i; 2135 2136 pci_info(pdev, "Found IOMMU cap 0x%x\n", iommu->cap_ptr); 2137 2138 if (iommu->cap & (1 << IOMMU_CAP_EFR)) { 2139 pr_info("Extended features (%#llx, %#llx):", iommu->features, iommu->features2); 2140 2141 for (i = 0; i < ARRAY_SIZE(feat_str); ++i) { 2142 if (iommu_feature(iommu, (1ULL << i))) 2143 pr_cont(" %s", feat_str[i]); 2144 } 2145 2146 if (iommu->features & FEATURE_GAM_VAPIC) 2147 pr_cont(" GA_vAPIC"); 2148 2149 if (iommu->features & FEATURE_SNP) 2150 pr_cont(" SNP"); 2151 2152 pr_cont("\n"); 2153 } 2154 } 2155 if (irq_remapping_enabled) { 2156 pr_info("Interrupt remapping enabled\n"); 2157 if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE) 2158 pr_info("X2APIC enabled\n"); 2159 } 2160 if (amd_iommu_pgtable == AMD_IOMMU_V2) 2161 pr_info("V2 page table enabled\n"); 2162 } 2163 2164 static int __init amd_iommu_init_pci(void) 2165 { 2166 struct amd_iommu *iommu; 2167 struct amd_iommu_pci_seg *pci_seg; 2168 int ret; 2169 2170 for_each_iommu(iommu) { 2171 ret = iommu_init_pci(iommu); 2172 if (ret) { 2173 pr_err("IOMMU%d: Failed to initialize IOMMU Hardware (error=%d)!\n", 2174 iommu->index, ret); 2175 goto out; 2176 } 2177 /* Need to setup range after PCI init */ 2178 iommu_set_cwwb_range(iommu); 2179 } 2180 2181 /* 2182 * Order is important here to make sure any unity map requirements are 2183 * fulfilled. The unity mappings are created and written to the device 2184 * table during the amd_iommu_init_api() call. 2185 * 2186 * After that we call init_device_table_dma() to make sure any 2187 * uninitialized DTE will block DMA, and in the end we flush the caches 2188 * of all IOMMUs to make sure the changes to the device table are 2189 * active. 2190 */ 2191 ret = amd_iommu_init_api(); 2192 if (ret) { 2193 pr_err("IOMMU: Failed to initialize IOMMU-API interface (error=%d)!\n", 2194 ret); 2195 goto out; 2196 } 2197 2198 for_each_pci_segment(pci_seg) 2199 init_device_table_dma(pci_seg); 2200 2201 for_each_iommu(iommu) 2202 iommu_flush_all_caches(iommu); 2203 2204 print_iommu_info(); 2205 2206 out: 2207 return ret; 2208 } 2209 2210 /**************************************************************************** 2211 * 2212 * The following functions initialize the MSI interrupts for all IOMMUs 2213 * in the system. It's a bit challenging because there could be multiple 2214 * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per 2215 * pci_dev. 2216 * 2217 ****************************************************************************/ 2218 2219 static int iommu_setup_msi(struct amd_iommu *iommu) 2220 { 2221 int r; 2222 2223 r = pci_enable_msi(iommu->dev); 2224 if (r) 2225 return r; 2226 2227 r = request_threaded_irq(iommu->dev->irq, 2228 amd_iommu_int_handler, 2229 amd_iommu_int_thread, 2230 0, "AMD-Vi", 2231 iommu); 2232 2233 if (r) { 2234 pci_disable_msi(iommu->dev); 2235 return r; 2236 } 2237 2238 return 0; 2239 } 2240 2241 union intcapxt { 2242 u64 capxt; 2243 struct { 2244 u64 reserved_0 : 2, 2245 dest_mode_logical : 1, 2246 reserved_1 : 5, 2247 destid_0_23 : 24, 2248 vector : 8, 2249 reserved_2 : 16, 2250 destid_24_31 : 8; 2251 }; 2252 } __attribute__ ((packed)); 2253 2254 2255 static struct irq_chip intcapxt_controller; 2256 2257 static int intcapxt_irqdomain_activate(struct irq_domain *domain, 2258 struct irq_data *irqd, bool reserve) 2259 { 2260 return 0; 2261 } 2262 2263 static void intcapxt_irqdomain_deactivate(struct irq_domain *domain, 2264 struct irq_data *irqd) 2265 { 2266 } 2267 2268 2269 static int intcapxt_irqdomain_alloc(struct irq_domain *domain, unsigned int virq, 2270 unsigned int nr_irqs, void *arg) 2271 { 2272 struct irq_alloc_info *info = arg; 2273 int i, ret; 2274 2275 if (!info || info->type != X86_IRQ_ALLOC_TYPE_AMDVI) 2276 return -EINVAL; 2277 2278 ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, arg); 2279 if (ret < 0) 2280 return ret; 2281 2282 for (i = virq; i < virq + nr_irqs; i++) { 2283 struct irq_data *irqd = irq_domain_get_irq_data(domain, i); 2284 2285 irqd->chip = &intcapxt_controller; 2286 irqd->chip_data = info->data; 2287 __irq_set_handler(i, handle_edge_irq, 0, "edge"); 2288 } 2289 2290 return ret; 2291 } 2292 2293 static void intcapxt_irqdomain_free(struct irq_domain *domain, unsigned int virq, 2294 unsigned int nr_irqs) 2295 { 2296 irq_domain_free_irqs_top(domain, virq, nr_irqs); 2297 } 2298 2299 2300 static void intcapxt_unmask_irq(struct irq_data *irqd) 2301 { 2302 struct amd_iommu *iommu = irqd->chip_data; 2303 struct irq_cfg *cfg = irqd_cfg(irqd); 2304 union intcapxt xt; 2305 2306 xt.capxt = 0ULL; 2307 xt.dest_mode_logical = apic->dest_mode_logical; 2308 xt.vector = cfg->vector; 2309 xt.destid_0_23 = cfg->dest_apicid & GENMASK(23, 0); 2310 xt.destid_24_31 = cfg->dest_apicid >> 24; 2311 2312 /** 2313 * Current IOMMU implementation uses the same IRQ for all 2314 * 3 IOMMU interrupts. 2315 */ 2316 writeq(xt.capxt, iommu->mmio_base + MMIO_INTCAPXT_EVT_OFFSET); 2317 writeq(xt.capxt, iommu->mmio_base + MMIO_INTCAPXT_PPR_OFFSET); 2318 writeq(xt.capxt, iommu->mmio_base + MMIO_INTCAPXT_GALOG_OFFSET); 2319 } 2320 2321 static void intcapxt_mask_irq(struct irq_data *irqd) 2322 { 2323 struct amd_iommu *iommu = irqd->chip_data; 2324 2325 writeq(0, iommu->mmio_base + MMIO_INTCAPXT_EVT_OFFSET); 2326 writeq(0, iommu->mmio_base + MMIO_INTCAPXT_PPR_OFFSET); 2327 writeq(0, iommu->mmio_base + MMIO_INTCAPXT_GALOG_OFFSET); 2328 } 2329 2330 2331 static int intcapxt_set_affinity(struct irq_data *irqd, 2332 const struct cpumask *mask, bool force) 2333 { 2334 struct irq_data *parent = irqd->parent_data; 2335 int ret; 2336 2337 ret = parent->chip->irq_set_affinity(parent, mask, force); 2338 if (ret < 0 || ret == IRQ_SET_MASK_OK_DONE) 2339 return ret; 2340 return 0; 2341 } 2342 2343 static int intcapxt_set_wake(struct irq_data *irqd, unsigned int on) 2344 { 2345 return on ? -EOPNOTSUPP : 0; 2346 } 2347 2348 static struct irq_chip intcapxt_controller = { 2349 .name = "IOMMU-MSI", 2350 .irq_unmask = intcapxt_unmask_irq, 2351 .irq_mask = intcapxt_mask_irq, 2352 .irq_ack = irq_chip_ack_parent, 2353 .irq_retrigger = irq_chip_retrigger_hierarchy, 2354 .irq_set_affinity = intcapxt_set_affinity, 2355 .irq_set_wake = intcapxt_set_wake, 2356 .flags = IRQCHIP_MASK_ON_SUSPEND, 2357 }; 2358 2359 static const struct irq_domain_ops intcapxt_domain_ops = { 2360 .alloc = intcapxt_irqdomain_alloc, 2361 .free = intcapxt_irqdomain_free, 2362 .activate = intcapxt_irqdomain_activate, 2363 .deactivate = intcapxt_irqdomain_deactivate, 2364 }; 2365 2366 2367 static struct irq_domain *iommu_irqdomain; 2368 2369 static struct irq_domain *iommu_get_irqdomain(void) 2370 { 2371 struct fwnode_handle *fn; 2372 2373 /* No need for locking here (yet) as the init is single-threaded */ 2374 if (iommu_irqdomain) 2375 return iommu_irqdomain; 2376 2377 fn = irq_domain_alloc_named_fwnode("AMD-Vi-MSI"); 2378 if (!fn) 2379 return NULL; 2380 2381 iommu_irqdomain = irq_domain_create_hierarchy(x86_vector_domain, 0, 0, 2382 fn, &intcapxt_domain_ops, 2383 NULL); 2384 if (!iommu_irqdomain) 2385 irq_domain_free_fwnode(fn); 2386 2387 return iommu_irqdomain; 2388 } 2389 2390 static int iommu_setup_intcapxt(struct amd_iommu *iommu) 2391 { 2392 struct irq_domain *domain; 2393 struct irq_alloc_info info; 2394 int irq, ret; 2395 2396 domain = iommu_get_irqdomain(); 2397 if (!domain) 2398 return -ENXIO; 2399 2400 init_irq_alloc_info(&info, NULL); 2401 info.type = X86_IRQ_ALLOC_TYPE_AMDVI; 2402 info.data = iommu; 2403 2404 irq = irq_domain_alloc_irqs(domain, 1, NUMA_NO_NODE, &info); 2405 if (irq < 0) { 2406 irq_domain_remove(domain); 2407 return irq; 2408 } 2409 2410 ret = request_threaded_irq(irq, amd_iommu_int_handler, 2411 amd_iommu_int_thread, 0, "AMD-Vi", iommu); 2412 if (ret) { 2413 irq_domain_free_irqs(irq, 1); 2414 irq_domain_remove(domain); 2415 return ret; 2416 } 2417 2418 return 0; 2419 } 2420 2421 static int iommu_init_irq(struct amd_iommu *iommu) 2422 { 2423 int ret; 2424 2425 if (iommu->int_enabled) 2426 goto enable_faults; 2427 2428 if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE) 2429 ret = iommu_setup_intcapxt(iommu); 2430 else if (iommu->dev->msi_cap) 2431 ret = iommu_setup_msi(iommu); 2432 else 2433 ret = -ENODEV; 2434 2435 if (ret) 2436 return ret; 2437 2438 iommu->int_enabled = true; 2439 enable_faults: 2440 2441 if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE) 2442 iommu_feature_enable(iommu, CONTROL_INTCAPXT_EN); 2443 2444 iommu_feature_enable(iommu, CONTROL_EVT_INT_EN); 2445 2446 if (iommu->ppr_log != NULL) 2447 iommu_feature_enable(iommu, CONTROL_PPRINT_EN); 2448 return 0; 2449 } 2450 2451 /**************************************************************************** 2452 * 2453 * The next functions belong to the third pass of parsing the ACPI 2454 * table. In this last pass the memory mapping requirements are 2455 * gathered (like exclusion and unity mapping ranges). 2456 * 2457 ****************************************************************************/ 2458 2459 static void __init free_unity_maps(void) 2460 { 2461 struct unity_map_entry *entry, *next; 2462 struct amd_iommu_pci_seg *p, *pci_seg; 2463 2464 for_each_pci_segment_safe(pci_seg, p) { 2465 list_for_each_entry_safe(entry, next, &pci_seg->unity_map, list) { 2466 list_del(&entry->list); 2467 kfree(entry); 2468 } 2469 } 2470 } 2471 2472 /* called for unity map ACPI definition */ 2473 static int __init init_unity_map_range(struct ivmd_header *m, 2474 struct acpi_table_header *ivrs_base) 2475 { 2476 struct unity_map_entry *e = NULL; 2477 struct amd_iommu_pci_seg *pci_seg; 2478 char *s; 2479 2480 pci_seg = get_pci_segment(m->pci_seg, ivrs_base); 2481 if (pci_seg == NULL) 2482 return -ENOMEM; 2483 2484 e = kzalloc(sizeof(*e), GFP_KERNEL); 2485 if (e == NULL) 2486 return -ENOMEM; 2487 2488 switch (m->type) { 2489 default: 2490 kfree(e); 2491 return 0; 2492 case ACPI_IVMD_TYPE: 2493 s = "IVMD_TYPEi\t\t\t"; 2494 e->devid_start = e->devid_end = m->devid; 2495 break; 2496 case ACPI_IVMD_TYPE_ALL: 2497 s = "IVMD_TYPE_ALL\t\t"; 2498 e->devid_start = 0; 2499 e->devid_end = pci_seg->last_bdf; 2500 break; 2501 case ACPI_IVMD_TYPE_RANGE: 2502 s = "IVMD_TYPE_RANGE\t\t"; 2503 e->devid_start = m->devid; 2504 e->devid_end = m->aux; 2505 break; 2506 } 2507 e->address_start = PAGE_ALIGN(m->range_start); 2508 e->address_end = e->address_start + PAGE_ALIGN(m->range_length); 2509 e->prot = m->flags >> 1; 2510 2511 /* 2512 * Treat per-device exclusion ranges as r/w unity-mapped regions 2513 * since some buggy BIOSes might lead to the overwritten exclusion 2514 * range (exclusion_start and exclusion_length members). This 2515 * happens when there are multiple exclusion ranges (IVMD entries) 2516 * defined in ACPI table. 2517 */ 2518 if (m->flags & IVMD_FLAG_EXCL_RANGE) 2519 e->prot = (IVMD_FLAG_IW | IVMD_FLAG_IR) >> 1; 2520 2521 DUMP_printk("%s devid_start: %04x:%02x:%02x.%x devid_end: " 2522 "%04x:%02x:%02x.%x range_start: %016llx range_end: %016llx" 2523 " flags: %x\n", s, m->pci_seg, 2524 PCI_BUS_NUM(e->devid_start), PCI_SLOT(e->devid_start), 2525 PCI_FUNC(e->devid_start), m->pci_seg, 2526 PCI_BUS_NUM(e->devid_end), 2527 PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end), 2528 e->address_start, e->address_end, m->flags); 2529 2530 list_add_tail(&e->list, &pci_seg->unity_map); 2531 2532 return 0; 2533 } 2534 2535 /* iterates over all memory definitions we find in the ACPI table */ 2536 static int __init init_memory_definitions(struct acpi_table_header *table) 2537 { 2538 u8 *p = (u8 *)table, *end = (u8 *)table; 2539 struct ivmd_header *m; 2540 2541 end += table->length; 2542 p += IVRS_HEADER_LENGTH; 2543 2544 while (p < end) { 2545 m = (struct ivmd_header *)p; 2546 if (m->flags & (IVMD_FLAG_UNITY_MAP | IVMD_FLAG_EXCL_RANGE)) 2547 init_unity_map_range(m, table); 2548 2549 p += m->length; 2550 } 2551 2552 return 0; 2553 } 2554 2555 /* 2556 * Init the device table to not allow DMA access for devices 2557 */ 2558 static void init_device_table_dma(struct amd_iommu_pci_seg *pci_seg) 2559 { 2560 u32 devid; 2561 struct dev_table_entry *dev_table = pci_seg->dev_table; 2562 2563 if (dev_table == NULL) 2564 return; 2565 2566 for (devid = 0; devid <= pci_seg->last_bdf; ++devid) { 2567 __set_dev_entry_bit(dev_table, devid, DEV_ENTRY_VALID); 2568 if (!amd_iommu_snp_en) 2569 __set_dev_entry_bit(dev_table, devid, DEV_ENTRY_TRANSLATION); 2570 } 2571 } 2572 2573 static void __init uninit_device_table_dma(struct amd_iommu_pci_seg *pci_seg) 2574 { 2575 u32 devid; 2576 struct dev_table_entry *dev_table = pci_seg->dev_table; 2577 2578 if (dev_table == NULL) 2579 return; 2580 2581 for (devid = 0; devid <= pci_seg->last_bdf; ++devid) { 2582 dev_table[devid].data[0] = 0ULL; 2583 dev_table[devid].data[1] = 0ULL; 2584 } 2585 } 2586 2587 static void init_device_table(void) 2588 { 2589 struct amd_iommu_pci_seg *pci_seg; 2590 u32 devid; 2591 2592 if (!amd_iommu_irq_remap) 2593 return; 2594 2595 for_each_pci_segment(pci_seg) { 2596 for (devid = 0; devid <= pci_seg->last_bdf; ++devid) 2597 __set_dev_entry_bit(pci_seg->dev_table, 2598 devid, DEV_ENTRY_IRQ_TBL_EN); 2599 } 2600 } 2601 2602 static void iommu_init_flags(struct amd_iommu *iommu) 2603 { 2604 iommu->acpi_flags & IVHD_FLAG_HT_TUN_EN_MASK ? 2605 iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) : 2606 iommu_feature_disable(iommu, CONTROL_HT_TUN_EN); 2607 2608 iommu->acpi_flags & IVHD_FLAG_PASSPW_EN_MASK ? 2609 iommu_feature_enable(iommu, CONTROL_PASSPW_EN) : 2610 iommu_feature_disable(iommu, CONTROL_PASSPW_EN); 2611 2612 iommu->acpi_flags & IVHD_FLAG_RESPASSPW_EN_MASK ? 2613 iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) : 2614 iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN); 2615 2616 iommu->acpi_flags & IVHD_FLAG_ISOC_EN_MASK ? 2617 iommu_feature_enable(iommu, CONTROL_ISOC_EN) : 2618 iommu_feature_disable(iommu, CONTROL_ISOC_EN); 2619 2620 /* 2621 * make IOMMU memory accesses cache coherent 2622 */ 2623 iommu_feature_enable(iommu, CONTROL_COHERENT_EN); 2624 2625 /* Set IOTLB invalidation timeout to 1s */ 2626 iommu_set_inv_tlb_timeout(iommu, CTRL_INV_TO_1S); 2627 } 2628 2629 static void iommu_apply_resume_quirks(struct amd_iommu *iommu) 2630 { 2631 int i, j; 2632 u32 ioc_feature_control; 2633 struct pci_dev *pdev = iommu->root_pdev; 2634 2635 /* RD890 BIOSes may not have completely reconfigured the iommu */ 2636 if (!is_rd890_iommu(iommu->dev) || !pdev) 2637 return; 2638 2639 /* 2640 * First, we need to ensure that the iommu is enabled. This is 2641 * controlled by a register in the northbridge 2642 */ 2643 2644 /* Select Northbridge indirect register 0x75 and enable writing */ 2645 pci_write_config_dword(pdev, 0x60, 0x75 | (1 << 7)); 2646 pci_read_config_dword(pdev, 0x64, &ioc_feature_control); 2647 2648 /* Enable the iommu */ 2649 if (!(ioc_feature_control & 0x1)) 2650 pci_write_config_dword(pdev, 0x64, ioc_feature_control | 1); 2651 2652 /* Restore the iommu BAR */ 2653 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4, 2654 iommu->stored_addr_lo); 2655 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 8, 2656 iommu->stored_addr_hi); 2657 2658 /* Restore the l1 indirect regs for each of the 6 l1s */ 2659 for (i = 0; i < 6; i++) 2660 for (j = 0; j < 0x12; j++) 2661 iommu_write_l1(iommu, i, j, iommu->stored_l1[i][j]); 2662 2663 /* Restore the l2 indirect regs */ 2664 for (i = 0; i < 0x83; i++) 2665 iommu_write_l2(iommu, i, iommu->stored_l2[i]); 2666 2667 /* Lock PCI setup registers */ 2668 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4, 2669 iommu->stored_addr_lo | 1); 2670 } 2671 2672 static void iommu_enable_ga(struct amd_iommu *iommu) 2673 { 2674 #ifdef CONFIG_IRQ_REMAP 2675 switch (amd_iommu_guest_ir) { 2676 case AMD_IOMMU_GUEST_IR_VAPIC: 2677 case AMD_IOMMU_GUEST_IR_LEGACY_GA: 2678 iommu_feature_enable(iommu, CONTROL_GA_EN); 2679 iommu->irte_ops = &irte_128_ops; 2680 break; 2681 default: 2682 iommu->irte_ops = &irte_32_ops; 2683 break; 2684 } 2685 #endif 2686 } 2687 2688 static void early_enable_iommu(struct amd_iommu *iommu) 2689 { 2690 iommu_disable(iommu); 2691 iommu_init_flags(iommu); 2692 iommu_set_device_table(iommu); 2693 iommu_enable_command_buffer(iommu); 2694 iommu_enable_event_buffer(iommu); 2695 iommu_set_exclusion_range(iommu); 2696 iommu_enable_ga(iommu); 2697 iommu_enable_xt(iommu); 2698 iommu_enable(iommu); 2699 iommu_flush_all_caches(iommu); 2700 } 2701 2702 /* 2703 * This function finally enables all IOMMUs found in the system after 2704 * they have been initialized. 2705 * 2706 * Or if in kdump kernel and IOMMUs are all pre-enabled, try to copy 2707 * the old content of device table entries. Not this case or copy failed, 2708 * just continue as normal kernel does. 2709 */ 2710 static void early_enable_iommus(void) 2711 { 2712 struct amd_iommu *iommu; 2713 struct amd_iommu_pci_seg *pci_seg; 2714 2715 if (!copy_device_table()) { 2716 /* 2717 * If come here because of failure in copying device table from old 2718 * kernel with all IOMMUs enabled, print error message and try to 2719 * free allocated old_dev_tbl_cpy. 2720 */ 2721 if (amd_iommu_pre_enabled) 2722 pr_err("Failed to copy DEV table from previous kernel.\n"); 2723 2724 for_each_pci_segment(pci_seg) { 2725 if (pci_seg->old_dev_tbl_cpy != NULL) { 2726 free_pages((unsigned long)pci_seg->old_dev_tbl_cpy, 2727 get_order(pci_seg->dev_table_size)); 2728 pci_seg->old_dev_tbl_cpy = NULL; 2729 } 2730 } 2731 2732 for_each_iommu(iommu) { 2733 clear_translation_pre_enabled(iommu); 2734 early_enable_iommu(iommu); 2735 } 2736 } else { 2737 pr_info("Copied DEV table from previous kernel.\n"); 2738 2739 for_each_pci_segment(pci_seg) { 2740 free_pages((unsigned long)pci_seg->dev_table, 2741 get_order(pci_seg->dev_table_size)); 2742 pci_seg->dev_table = pci_seg->old_dev_tbl_cpy; 2743 } 2744 2745 for_each_iommu(iommu) { 2746 iommu_disable_command_buffer(iommu); 2747 iommu_disable_event_buffer(iommu); 2748 iommu_enable_command_buffer(iommu); 2749 iommu_enable_event_buffer(iommu); 2750 iommu_enable_ga(iommu); 2751 iommu_enable_xt(iommu); 2752 iommu_set_device_table(iommu); 2753 iommu_flush_all_caches(iommu); 2754 } 2755 } 2756 } 2757 2758 static void enable_iommus_v2(void) 2759 { 2760 struct amd_iommu *iommu; 2761 2762 for_each_iommu(iommu) { 2763 iommu_enable_ppr_log(iommu); 2764 iommu_enable_gt(iommu); 2765 } 2766 } 2767 2768 static void enable_iommus_vapic(void) 2769 { 2770 #ifdef CONFIG_IRQ_REMAP 2771 u32 status, i; 2772 struct amd_iommu *iommu; 2773 2774 for_each_iommu(iommu) { 2775 /* 2776 * Disable GALog if already running. It could have been enabled 2777 * in the previous boot before kdump. 2778 */ 2779 status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET); 2780 if (!(status & MMIO_STATUS_GALOG_RUN_MASK)) 2781 continue; 2782 2783 iommu_feature_disable(iommu, CONTROL_GALOG_EN); 2784 iommu_feature_disable(iommu, CONTROL_GAINT_EN); 2785 2786 /* 2787 * Need to set and poll check the GALOGRun bit to zero before 2788 * we can set/ modify GA Log registers safely. 2789 */ 2790 for (i = 0; i < LOOP_TIMEOUT; ++i) { 2791 status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET); 2792 if (!(status & MMIO_STATUS_GALOG_RUN_MASK)) 2793 break; 2794 udelay(10); 2795 } 2796 2797 if (WARN_ON(i >= LOOP_TIMEOUT)) 2798 return; 2799 } 2800 2801 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir) && 2802 !check_feature_on_all_iommus(FEATURE_GAM_VAPIC)) { 2803 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA; 2804 return; 2805 } 2806 2807 if (amd_iommu_snp_en && 2808 !FEATURE_SNPAVICSUP_GAM(amd_iommu_efr2)) { 2809 pr_warn("Force to disable Virtual APIC due to SNP\n"); 2810 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA; 2811 return; 2812 } 2813 2814 /* Enabling GAM and SNPAVIC support */ 2815 for_each_iommu(iommu) { 2816 if (iommu_init_ga_log(iommu) || 2817 iommu_ga_log_enable(iommu)) 2818 return; 2819 2820 iommu_feature_enable(iommu, CONTROL_GAM_EN); 2821 if (amd_iommu_snp_en) 2822 iommu_feature_enable(iommu, CONTROL_SNPAVIC_EN); 2823 } 2824 2825 amd_iommu_irq_ops.capability |= (1 << IRQ_POSTING_CAP); 2826 pr_info("Virtual APIC enabled\n"); 2827 #endif 2828 } 2829 2830 static void enable_iommus(void) 2831 { 2832 early_enable_iommus(); 2833 enable_iommus_vapic(); 2834 enable_iommus_v2(); 2835 } 2836 2837 static void disable_iommus(void) 2838 { 2839 struct amd_iommu *iommu; 2840 2841 for_each_iommu(iommu) 2842 iommu_disable(iommu); 2843 2844 #ifdef CONFIG_IRQ_REMAP 2845 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir)) 2846 amd_iommu_irq_ops.capability &= ~(1 << IRQ_POSTING_CAP); 2847 #endif 2848 } 2849 2850 /* 2851 * Suspend/Resume support 2852 * disable suspend until real resume implemented 2853 */ 2854 2855 static void amd_iommu_resume(void) 2856 { 2857 struct amd_iommu *iommu; 2858 2859 for_each_iommu(iommu) 2860 iommu_apply_resume_quirks(iommu); 2861 2862 /* re-load the hardware */ 2863 enable_iommus(); 2864 2865 amd_iommu_enable_interrupts(); 2866 } 2867 2868 static int amd_iommu_suspend(void) 2869 { 2870 /* disable IOMMUs to go out of the way for BIOS */ 2871 disable_iommus(); 2872 2873 return 0; 2874 } 2875 2876 static struct syscore_ops amd_iommu_syscore_ops = { 2877 .suspend = amd_iommu_suspend, 2878 .resume = amd_iommu_resume, 2879 }; 2880 2881 static void __init free_iommu_resources(void) 2882 { 2883 kmem_cache_destroy(amd_iommu_irq_cache); 2884 amd_iommu_irq_cache = NULL; 2885 2886 free_iommu_all(); 2887 free_pci_segments(); 2888 } 2889 2890 /* SB IOAPIC is always on this device in AMD systems */ 2891 #define IOAPIC_SB_DEVID ((0x00 << 8) | PCI_DEVFN(0x14, 0)) 2892 2893 static bool __init check_ioapic_information(void) 2894 { 2895 const char *fw_bug = FW_BUG; 2896 bool ret, has_sb_ioapic; 2897 int idx; 2898 2899 has_sb_ioapic = false; 2900 ret = false; 2901 2902 /* 2903 * If we have map overrides on the kernel command line the 2904 * messages in this function might not describe firmware bugs 2905 * anymore - so be careful 2906 */ 2907 if (cmdline_maps) 2908 fw_bug = ""; 2909 2910 for (idx = 0; idx < nr_ioapics; idx++) { 2911 int devid, id = mpc_ioapic_id(idx); 2912 2913 devid = get_ioapic_devid(id); 2914 if (devid < 0) { 2915 pr_err("%s: IOAPIC[%d] not in IVRS table\n", 2916 fw_bug, id); 2917 ret = false; 2918 } else if (devid == IOAPIC_SB_DEVID) { 2919 has_sb_ioapic = true; 2920 ret = true; 2921 } 2922 } 2923 2924 if (!has_sb_ioapic) { 2925 /* 2926 * We expect the SB IOAPIC to be listed in the IVRS 2927 * table. The system timer is connected to the SB IOAPIC 2928 * and if we don't have it in the list the system will 2929 * panic at boot time. This situation usually happens 2930 * when the BIOS is buggy and provides us the wrong 2931 * device id for the IOAPIC in the system. 2932 */ 2933 pr_err("%s: No southbridge IOAPIC found\n", fw_bug); 2934 } 2935 2936 if (!ret) 2937 pr_err("Disabling interrupt remapping\n"); 2938 2939 return ret; 2940 } 2941 2942 static void __init free_dma_resources(void) 2943 { 2944 free_pages((unsigned long)amd_iommu_pd_alloc_bitmap, 2945 get_order(MAX_DOMAIN_ID/8)); 2946 amd_iommu_pd_alloc_bitmap = NULL; 2947 2948 free_unity_maps(); 2949 } 2950 2951 static void __init ivinfo_init(void *ivrs) 2952 { 2953 amd_iommu_ivinfo = *((u32 *)(ivrs + IOMMU_IVINFO_OFFSET)); 2954 } 2955 2956 /* 2957 * This is the hardware init function for AMD IOMMU in the system. 2958 * This function is called either from amd_iommu_init or from the interrupt 2959 * remapping setup code. 2960 * 2961 * This function basically parses the ACPI table for AMD IOMMU (IVRS) 2962 * four times: 2963 * 2964 * 1 pass) Discover the most comprehensive IVHD type to use. 2965 * 2966 * 2 pass) Find the highest PCI device id the driver has to handle. 2967 * Upon this information the size of the data structures is 2968 * determined that needs to be allocated. 2969 * 2970 * 3 pass) Initialize the data structures just allocated with the 2971 * information in the ACPI table about available AMD IOMMUs 2972 * in the system. It also maps the PCI devices in the 2973 * system to specific IOMMUs 2974 * 2975 * 4 pass) After the basic data structures are allocated and 2976 * initialized we update them with information about memory 2977 * remapping requirements parsed out of the ACPI table in 2978 * this last pass. 2979 * 2980 * After everything is set up the IOMMUs are enabled and the necessary 2981 * hotplug and suspend notifiers are registered. 2982 */ 2983 static int __init early_amd_iommu_init(void) 2984 { 2985 struct acpi_table_header *ivrs_base; 2986 int remap_cache_sz, ret; 2987 acpi_status status; 2988 2989 if (!amd_iommu_detected) 2990 return -ENODEV; 2991 2992 status = acpi_get_table("IVRS", 0, &ivrs_base); 2993 if (status == AE_NOT_FOUND) 2994 return -ENODEV; 2995 else if (ACPI_FAILURE(status)) { 2996 const char *err = acpi_format_exception(status); 2997 pr_err("IVRS table error: %s\n", err); 2998 return -EINVAL; 2999 } 3000 3001 /* 3002 * Validate checksum here so we don't need to do it when 3003 * we actually parse the table 3004 */ 3005 ret = check_ivrs_checksum(ivrs_base); 3006 if (ret) 3007 goto out; 3008 3009 ivinfo_init(ivrs_base); 3010 3011 amd_iommu_target_ivhd_type = get_highest_supported_ivhd_type(ivrs_base); 3012 DUMP_printk("Using IVHD type %#x\n", amd_iommu_target_ivhd_type); 3013 3014 /* Device table - directly used by all IOMMUs */ 3015 ret = -ENOMEM; 3016 3017 amd_iommu_pd_alloc_bitmap = (void *)__get_free_pages( 3018 GFP_KERNEL | __GFP_ZERO, 3019 get_order(MAX_DOMAIN_ID/8)); 3020 if (amd_iommu_pd_alloc_bitmap == NULL) 3021 goto out; 3022 3023 /* 3024 * never allocate domain 0 because its used as the non-allocated and 3025 * error value placeholder 3026 */ 3027 __set_bit(0, amd_iommu_pd_alloc_bitmap); 3028 3029 /* 3030 * now the data structures are allocated and basically initialized 3031 * start the real acpi table scan 3032 */ 3033 ret = init_iommu_all(ivrs_base); 3034 if (ret) 3035 goto out; 3036 3037 /* Disable any previously enabled IOMMUs */ 3038 if (!is_kdump_kernel() || amd_iommu_disabled) 3039 disable_iommus(); 3040 3041 if (amd_iommu_irq_remap) 3042 amd_iommu_irq_remap = check_ioapic_information(); 3043 3044 if (amd_iommu_irq_remap) { 3045 struct amd_iommu_pci_seg *pci_seg; 3046 /* 3047 * Interrupt remapping enabled, create kmem_cache for the 3048 * remapping tables. 3049 */ 3050 ret = -ENOMEM; 3051 if (!AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir)) 3052 remap_cache_sz = MAX_IRQS_PER_TABLE * sizeof(u32); 3053 else 3054 remap_cache_sz = MAX_IRQS_PER_TABLE * (sizeof(u64) * 2); 3055 amd_iommu_irq_cache = kmem_cache_create("irq_remap_cache", 3056 remap_cache_sz, 3057 DTE_INTTAB_ALIGNMENT, 3058 0, NULL); 3059 if (!amd_iommu_irq_cache) 3060 goto out; 3061 3062 for_each_pci_segment(pci_seg) { 3063 if (alloc_irq_lookup_table(pci_seg)) 3064 goto out; 3065 } 3066 } 3067 3068 ret = init_memory_definitions(ivrs_base); 3069 if (ret) 3070 goto out; 3071 3072 /* init the device table */ 3073 init_device_table(); 3074 3075 out: 3076 /* Don't leak any ACPI memory */ 3077 acpi_put_table(ivrs_base); 3078 3079 return ret; 3080 } 3081 3082 static int amd_iommu_enable_interrupts(void) 3083 { 3084 struct amd_iommu *iommu; 3085 int ret = 0; 3086 3087 for_each_iommu(iommu) { 3088 ret = iommu_init_irq(iommu); 3089 if (ret) 3090 goto out; 3091 } 3092 3093 out: 3094 return ret; 3095 } 3096 3097 static bool __init detect_ivrs(void) 3098 { 3099 struct acpi_table_header *ivrs_base; 3100 acpi_status status; 3101 int i; 3102 3103 status = acpi_get_table("IVRS", 0, &ivrs_base); 3104 if (status == AE_NOT_FOUND) 3105 return false; 3106 else if (ACPI_FAILURE(status)) { 3107 const char *err = acpi_format_exception(status); 3108 pr_err("IVRS table error: %s\n", err); 3109 return false; 3110 } 3111 3112 acpi_put_table(ivrs_base); 3113 3114 if (amd_iommu_force_enable) 3115 goto out; 3116 3117 /* Don't use IOMMU if there is Stoney Ridge graphics */ 3118 for (i = 0; i < 32; i++) { 3119 u32 pci_id; 3120 3121 pci_id = read_pci_config(0, i, 0, 0); 3122 if ((pci_id & 0xffff) == 0x1002 && (pci_id >> 16) == 0x98e4) { 3123 pr_info("Disable IOMMU on Stoney Ridge\n"); 3124 return false; 3125 } 3126 } 3127 3128 out: 3129 /* Make sure ACS will be enabled during PCI probe */ 3130 pci_request_acs(); 3131 3132 return true; 3133 } 3134 3135 /**************************************************************************** 3136 * 3137 * AMD IOMMU Initialization State Machine 3138 * 3139 ****************************************************************************/ 3140 3141 static int __init state_next(void) 3142 { 3143 int ret = 0; 3144 3145 switch (init_state) { 3146 case IOMMU_START_STATE: 3147 if (!detect_ivrs()) { 3148 init_state = IOMMU_NOT_FOUND; 3149 ret = -ENODEV; 3150 } else { 3151 init_state = IOMMU_IVRS_DETECTED; 3152 } 3153 break; 3154 case IOMMU_IVRS_DETECTED: 3155 if (amd_iommu_disabled) { 3156 init_state = IOMMU_CMDLINE_DISABLED; 3157 ret = -EINVAL; 3158 } else { 3159 ret = early_amd_iommu_init(); 3160 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_ACPI_FINISHED; 3161 } 3162 break; 3163 case IOMMU_ACPI_FINISHED: 3164 early_enable_iommus(); 3165 x86_platform.iommu_shutdown = disable_iommus; 3166 init_state = IOMMU_ENABLED; 3167 break; 3168 case IOMMU_ENABLED: 3169 register_syscore_ops(&amd_iommu_syscore_ops); 3170 ret = amd_iommu_init_pci(); 3171 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_PCI_INIT; 3172 enable_iommus_vapic(); 3173 enable_iommus_v2(); 3174 break; 3175 case IOMMU_PCI_INIT: 3176 ret = amd_iommu_enable_interrupts(); 3177 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_INTERRUPTS_EN; 3178 break; 3179 case IOMMU_INTERRUPTS_EN: 3180 init_state = IOMMU_INITIALIZED; 3181 break; 3182 case IOMMU_INITIALIZED: 3183 /* Nothing to do */ 3184 break; 3185 case IOMMU_NOT_FOUND: 3186 case IOMMU_INIT_ERROR: 3187 case IOMMU_CMDLINE_DISABLED: 3188 /* Error states => do nothing */ 3189 ret = -EINVAL; 3190 break; 3191 default: 3192 /* Unknown state */ 3193 BUG(); 3194 } 3195 3196 if (ret) { 3197 free_dma_resources(); 3198 if (!irq_remapping_enabled) { 3199 disable_iommus(); 3200 free_iommu_resources(); 3201 } else { 3202 struct amd_iommu *iommu; 3203 struct amd_iommu_pci_seg *pci_seg; 3204 3205 for_each_pci_segment(pci_seg) 3206 uninit_device_table_dma(pci_seg); 3207 3208 for_each_iommu(iommu) 3209 iommu_flush_all_caches(iommu); 3210 } 3211 } 3212 return ret; 3213 } 3214 3215 static int __init iommu_go_to_state(enum iommu_init_state state) 3216 { 3217 int ret = -EINVAL; 3218 3219 while (init_state != state) { 3220 if (init_state == IOMMU_NOT_FOUND || 3221 init_state == IOMMU_INIT_ERROR || 3222 init_state == IOMMU_CMDLINE_DISABLED) 3223 break; 3224 ret = state_next(); 3225 } 3226 3227 return ret; 3228 } 3229 3230 #ifdef CONFIG_IRQ_REMAP 3231 int __init amd_iommu_prepare(void) 3232 { 3233 int ret; 3234 3235 amd_iommu_irq_remap = true; 3236 3237 ret = iommu_go_to_state(IOMMU_ACPI_FINISHED); 3238 if (ret) { 3239 amd_iommu_irq_remap = false; 3240 return ret; 3241 } 3242 3243 return amd_iommu_irq_remap ? 0 : -ENODEV; 3244 } 3245 3246 int __init amd_iommu_enable(void) 3247 { 3248 int ret; 3249 3250 ret = iommu_go_to_state(IOMMU_ENABLED); 3251 if (ret) 3252 return ret; 3253 3254 irq_remapping_enabled = 1; 3255 return amd_iommu_xt_mode; 3256 } 3257 3258 void amd_iommu_disable(void) 3259 { 3260 amd_iommu_suspend(); 3261 } 3262 3263 int amd_iommu_reenable(int mode) 3264 { 3265 amd_iommu_resume(); 3266 3267 return 0; 3268 } 3269 3270 int __init amd_iommu_enable_faulting(void) 3271 { 3272 /* We enable MSI later when PCI is initialized */ 3273 return 0; 3274 } 3275 #endif 3276 3277 /* 3278 * This is the core init function for AMD IOMMU hardware in the system. 3279 * This function is called from the generic x86 DMA layer initialization 3280 * code. 3281 */ 3282 static int __init amd_iommu_init(void) 3283 { 3284 struct amd_iommu *iommu; 3285 int ret; 3286 3287 ret = iommu_go_to_state(IOMMU_INITIALIZED); 3288 #ifdef CONFIG_GART_IOMMU 3289 if (ret && list_empty(&amd_iommu_list)) { 3290 /* 3291 * We failed to initialize the AMD IOMMU - try fallback 3292 * to GART if possible. 3293 */ 3294 gart_iommu_init(); 3295 } 3296 #endif 3297 3298 for_each_iommu(iommu) 3299 amd_iommu_debugfs_setup(iommu); 3300 3301 return ret; 3302 } 3303 3304 static bool amd_iommu_sme_check(void) 3305 { 3306 if (!cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT) || 3307 (boot_cpu_data.x86 != 0x17)) 3308 return true; 3309 3310 /* For Fam17h, a specific level of support is required */ 3311 if (boot_cpu_data.microcode >= 0x08001205) 3312 return true; 3313 3314 if ((boot_cpu_data.microcode >= 0x08001126) && 3315 (boot_cpu_data.microcode <= 0x080011ff)) 3316 return true; 3317 3318 pr_notice("IOMMU not currently supported when SME is active\n"); 3319 3320 return false; 3321 } 3322 3323 /**************************************************************************** 3324 * 3325 * Early detect code. This code runs at IOMMU detection time in the DMA 3326 * layer. It just looks if there is an IVRS ACPI table to detect AMD 3327 * IOMMUs 3328 * 3329 ****************************************************************************/ 3330 int __init amd_iommu_detect(void) 3331 { 3332 int ret; 3333 3334 if (no_iommu || (iommu_detected && !gart_iommu_aperture)) 3335 return -ENODEV; 3336 3337 if (!amd_iommu_sme_check()) 3338 return -ENODEV; 3339 3340 ret = iommu_go_to_state(IOMMU_IVRS_DETECTED); 3341 if (ret) 3342 return ret; 3343 3344 amd_iommu_detected = true; 3345 iommu_detected = 1; 3346 x86_init.iommu.iommu_init = amd_iommu_init; 3347 3348 return 1; 3349 } 3350 3351 /**************************************************************************** 3352 * 3353 * Parsing functions for the AMD IOMMU specific kernel command line 3354 * options. 3355 * 3356 ****************************************************************************/ 3357 3358 static int __init parse_amd_iommu_dump(char *str) 3359 { 3360 amd_iommu_dump = true; 3361 3362 return 1; 3363 } 3364 3365 static int __init parse_amd_iommu_intr(char *str) 3366 { 3367 for (; *str; ++str) { 3368 if (strncmp(str, "legacy", 6) == 0) { 3369 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA; 3370 break; 3371 } 3372 if (strncmp(str, "vapic", 5) == 0) { 3373 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC; 3374 break; 3375 } 3376 } 3377 return 1; 3378 } 3379 3380 static int __init parse_amd_iommu_options(char *str) 3381 { 3382 for (; *str; ++str) { 3383 if (strncmp(str, "fullflush", 9) == 0) { 3384 pr_warn("amd_iommu=fullflush deprecated; use iommu.strict=1 instead\n"); 3385 iommu_set_dma_strict(); 3386 } 3387 if (strncmp(str, "force_enable", 12) == 0) 3388 amd_iommu_force_enable = true; 3389 if (strncmp(str, "off", 3) == 0) 3390 amd_iommu_disabled = true; 3391 if (strncmp(str, "force_isolation", 15) == 0) 3392 amd_iommu_force_isolation = true; 3393 } 3394 3395 return 1; 3396 } 3397 3398 static int __init parse_ivrs_ioapic(char *str) 3399 { 3400 u32 seg = 0, bus, dev, fn; 3401 int ret, id, i; 3402 u32 devid; 3403 3404 ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn); 3405 if (ret != 4) { 3406 ret = sscanf(str, "[%d]=%x:%x:%x.%x", &id, &seg, &bus, &dev, &fn); 3407 if (ret != 5) { 3408 pr_err("Invalid command line: ivrs_ioapic%s\n", str); 3409 return 1; 3410 } 3411 } 3412 3413 if (early_ioapic_map_size == EARLY_MAP_SIZE) { 3414 pr_err("Early IOAPIC map overflow - ignoring ivrs_ioapic%s\n", 3415 str); 3416 return 1; 3417 } 3418 3419 devid = IVRS_GET_SBDF_ID(seg, bus, dev, fn); 3420 3421 cmdline_maps = true; 3422 i = early_ioapic_map_size++; 3423 early_ioapic_map[i].id = id; 3424 early_ioapic_map[i].devid = devid; 3425 early_ioapic_map[i].cmd_line = true; 3426 3427 return 1; 3428 } 3429 3430 static int __init parse_ivrs_hpet(char *str) 3431 { 3432 u32 seg = 0, bus, dev, fn; 3433 int ret, id, i; 3434 u32 devid; 3435 3436 ret = sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn); 3437 if (ret != 4) { 3438 ret = sscanf(str, "[%d]=%x:%x:%x.%x", &id, &seg, &bus, &dev, &fn); 3439 if (ret != 5) { 3440 pr_err("Invalid command line: ivrs_hpet%s\n", str); 3441 return 1; 3442 } 3443 } 3444 3445 if (early_hpet_map_size == EARLY_MAP_SIZE) { 3446 pr_err("Early HPET map overflow - ignoring ivrs_hpet%s\n", 3447 str); 3448 return 1; 3449 } 3450 3451 devid = IVRS_GET_SBDF_ID(seg, bus, dev, fn); 3452 3453 cmdline_maps = true; 3454 i = early_hpet_map_size++; 3455 early_hpet_map[i].id = id; 3456 early_hpet_map[i].devid = devid; 3457 early_hpet_map[i].cmd_line = true; 3458 3459 return 1; 3460 } 3461 3462 static int __init parse_ivrs_acpihid(char *str) 3463 { 3464 u32 seg = 0, bus, dev, fn; 3465 char *hid, *uid, *p; 3466 char acpiid[ACPIHID_UID_LEN + ACPIHID_HID_LEN] = {0}; 3467 int ret, i; 3468 3469 ret = sscanf(str, "[%x:%x.%x]=%s", &bus, &dev, &fn, acpiid); 3470 if (ret != 4) { 3471 ret = sscanf(str, "[%x:%x:%x.%x]=%s", &seg, &bus, &dev, &fn, acpiid); 3472 if (ret != 5) { 3473 pr_err("Invalid command line: ivrs_acpihid(%s)\n", str); 3474 return 1; 3475 } 3476 } 3477 3478 p = acpiid; 3479 hid = strsep(&p, ":"); 3480 uid = p; 3481 3482 if (!hid || !(*hid) || !uid) { 3483 pr_err("Invalid command line: hid or uid\n"); 3484 return 1; 3485 } 3486 3487 i = early_acpihid_map_size++; 3488 memcpy(early_acpihid_map[i].hid, hid, strlen(hid)); 3489 memcpy(early_acpihid_map[i].uid, uid, strlen(uid)); 3490 early_acpihid_map[i].devid = IVRS_GET_SBDF_ID(seg, bus, dev, fn); 3491 early_acpihid_map[i].cmd_line = true; 3492 3493 return 1; 3494 } 3495 3496 __setup("amd_iommu_dump", parse_amd_iommu_dump); 3497 __setup("amd_iommu=", parse_amd_iommu_options); 3498 __setup("amd_iommu_intr=", parse_amd_iommu_intr); 3499 __setup("ivrs_ioapic", parse_ivrs_ioapic); 3500 __setup("ivrs_hpet", parse_ivrs_hpet); 3501 __setup("ivrs_acpihid", parse_ivrs_acpihid); 3502 3503 bool amd_iommu_v2_supported(void) 3504 { 3505 /* 3506 * Since DTE[Mode]=0 is prohibited on SNP-enabled system 3507 * (i.e. EFR[SNPSup]=1), IOMMUv2 page table cannot be used without 3508 * setting up IOMMUv1 page table. 3509 */ 3510 return amd_iommu_v2_present && !amd_iommu_snp_en; 3511 } 3512 EXPORT_SYMBOL(amd_iommu_v2_supported); 3513 3514 struct amd_iommu *get_amd_iommu(unsigned int idx) 3515 { 3516 unsigned int i = 0; 3517 struct amd_iommu *iommu; 3518 3519 for_each_iommu(iommu) 3520 if (i++ == idx) 3521 return iommu; 3522 return NULL; 3523 } 3524 3525 /**************************************************************************** 3526 * 3527 * IOMMU EFR Performance Counter support functionality. This code allows 3528 * access to the IOMMU PC functionality. 3529 * 3530 ****************************************************************************/ 3531 3532 u8 amd_iommu_pc_get_max_banks(unsigned int idx) 3533 { 3534 struct amd_iommu *iommu = get_amd_iommu(idx); 3535 3536 if (iommu) 3537 return iommu->max_banks; 3538 3539 return 0; 3540 } 3541 EXPORT_SYMBOL(amd_iommu_pc_get_max_banks); 3542 3543 bool amd_iommu_pc_supported(void) 3544 { 3545 return amd_iommu_pc_present; 3546 } 3547 EXPORT_SYMBOL(amd_iommu_pc_supported); 3548 3549 u8 amd_iommu_pc_get_max_counters(unsigned int idx) 3550 { 3551 struct amd_iommu *iommu = get_amd_iommu(idx); 3552 3553 if (iommu) 3554 return iommu->max_counters; 3555 3556 return 0; 3557 } 3558 EXPORT_SYMBOL(amd_iommu_pc_get_max_counters); 3559 3560 static int iommu_pc_get_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, 3561 u8 fxn, u64 *value, bool is_write) 3562 { 3563 u32 offset; 3564 u32 max_offset_lim; 3565 3566 /* Make sure the IOMMU PC resource is available */ 3567 if (!amd_iommu_pc_present) 3568 return -ENODEV; 3569 3570 /* Check for valid iommu and pc register indexing */ 3571 if (WARN_ON(!iommu || (fxn > 0x28) || (fxn & 7))) 3572 return -ENODEV; 3573 3574 offset = (u32)(((0x40 | bank) << 12) | (cntr << 8) | fxn); 3575 3576 /* Limit the offset to the hw defined mmio region aperture */ 3577 max_offset_lim = (u32)(((0x40 | iommu->max_banks) << 12) | 3578 (iommu->max_counters << 8) | 0x28); 3579 if ((offset < MMIO_CNTR_REG_OFFSET) || 3580 (offset > max_offset_lim)) 3581 return -EINVAL; 3582 3583 if (is_write) { 3584 u64 val = *value & GENMASK_ULL(47, 0); 3585 3586 writel((u32)val, iommu->mmio_base + offset); 3587 writel((val >> 32), iommu->mmio_base + offset + 4); 3588 } else { 3589 *value = readl(iommu->mmio_base + offset + 4); 3590 *value <<= 32; 3591 *value |= readl(iommu->mmio_base + offset); 3592 *value &= GENMASK_ULL(47, 0); 3593 } 3594 3595 return 0; 3596 } 3597 3598 int amd_iommu_pc_get_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value) 3599 { 3600 if (!iommu) 3601 return -EINVAL; 3602 3603 return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, false); 3604 } 3605 3606 int amd_iommu_pc_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value) 3607 { 3608 if (!iommu) 3609 return -EINVAL; 3610 3611 return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, true); 3612 } 3613 3614 #ifdef CONFIG_AMD_MEM_ENCRYPT 3615 int amd_iommu_snp_enable(void) 3616 { 3617 /* 3618 * The SNP support requires that IOMMU must be enabled, and is 3619 * not configured in the passthrough mode. 3620 */ 3621 if (no_iommu || iommu_default_passthrough()) { 3622 pr_err("SNP: IOMMU is disabled or configured in passthrough mode, SNP cannot be supported"); 3623 return -EINVAL; 3624 } 3625 3626 /* 3627 * Prevent enabling SNP after IOMMU_ENABLED state because this process 3628 * affect how IOMMU driver sets up data structures and configures 3629 * IOMMU hardware. 3630 */ 3631 if (init_state > IOMMU_ENABLED) { 3632 pr_err("SNP: Too late to enable SNP for IOMMU.\n"); 3633 return -EINVAL; 3634 } 3635 3636 amd_iommu_snp_en = check_feature_on_all_iommus(FEATURE_SNP); 3637 if (!amd_iommu_snp_en) 3638 return -EINVAL; 3639 3640 pr_info("SNP enabled\n"); 3641 3642 /* Enforce IOMMU v1 pagetable when SNP is enabled. */ 3643 if (amd_iommu_pgtable != AMD_IOMMU_V1) { 3644 pr_warn("Force to using AMD IOMMU v1 page table due to SNP\n"); 3645 amd_iommu_pgtable = AMD_IOMMU_V1; 3646 } 3647 3648 return 0; 3649 } 3650 #endif 3651