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, fn) (((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 /* 99 * structure describing one IOMMU in the ACPI table. Typically followed by one 100 * or more ivhd_entrys. 101 */ 102 struct ivhd_header { 103 u8 type; 104 u8 flags; 105 u16 length; 106 u16 devid; 107 u16 cap_ptr; 108 u64 mmio_phys; 109 u16 pci_seg; 110 u16 info; 111 u32 efr_attr; 112 113 /* Following only valid on IVHD type 11h and 40h */ 114 u64 efr_reg; /* Exact copy of MMIO_EXT_FEATURES */ 115 u64 efr_reg2; 116 } __attribute__((packed)); 117 118 /* 119 * A device entry describing which devices a specific IOMMU translates and 120 * which requestor ids they use. 121 */ 122 struct ivhd_entry { 123 u8 type; 124 u16 devid; 125 u8 flags; 126 struct_group(ext_hid, 127 u32 ext; 128 u32 hidh; 129 ); 130 u64 cid; 131 u8 uidf; 132 u8 uidl; 133 u8 uid; 134 } __attribute__((packed)); 135 136 /* 137 * An AMD IOMMU memory definition structure. It defines things like exclusion 138 * ranges for devices and regions that should be unity mapped. 139 */ 140 struct ivmd_header { 141 u8 type; 142 u8 flags; 143 u16 length; 144 u16 devid; 145 u16 aux; 146 u16 pci_seg; 147 u8 resv[6]; 148 u64 range_start; 149 u64 range_length; 150 } __attribute__((packed)); 151 152 bool amd_iommu_dump; 153 bool amd_iommu_irq_remap __read_mostly; 154 155 enum io_pgtable_fmt amd_iommu_pgtable = AMD_IOMMU_V1; 156 /* Guest page table level */ 157 int amd_iommu_gpt_level = PAGE_MODE_4_LEVEL; 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 bool amd_iommu_irtcachedis; 166 static int amd_iommu_target_ivhd_type; 167 168 /* Global EFR and EFR2 registers */ 169 u64 amd_iommu_efr; 170 u64 amd_iommu_efr2; 171 172 /* SNP is enabled on the system? */ 173 bool amd_iommu_snp_en; 174 EXPORT_SYMBOL(amd_iommu_snp_en); 175 176 LIST_HEAD(amd_iommu_pci_seg_list); /* list of all PCI segments */ 177 LIST_HEAD(amd_iommu_list); /* list of all AMD IOMMUs in the 178 system */ 179 180 /* Array to assign indices to IOMMUs*/ 181 struct amd_iommu *amd_iommus[MAX_IOMMUS]; 182 183 /* Number of IOMMUs present in the system */ 184 static int amd_iommus_present; 185 186 /* IOMMUs have a non-present cache? */ 187 bool amd_iommu_np_cache __read_mostly; 188 bool amd_iommu_iotlb_sup __read_mostly = true; 189 190 u32 amd_iommu_max_pasid __read_mostly = ~0; 191 192 bool amd_iommu_v2_present __read_mostly; 193 static bool amd_iommu_pc_present __read_mostly; 194 bool amdr_ivrs_remap_support __read_mostly; 195 196 bool amd_iommu_force_isolation __read_mostly; 197 198 /* 199 * AMD IOMMU allows up to 2^16 different protection domains. This is a bitmap 200 * to know which ones are already in use. 201 */ 202 unsigned long *amd_iommu_pd_alloc_bitmap; 203 204 enum iommu_init_state { 205 IOMMU_START_STATE, 206 IOMMU_IVRS_DETECTED, 207 IOMMU_ACPI_FINISHED, 208 IOMMU_ENABLED, 209 IOMMU_PCI_INIT, 210 IOMMU_INTERRUPTS_EN, 211 IOMMU_INITIALIZED, 212 IOMMU_NOT_FOUND, 213 IOMMU_INIT_ERROR, 214 IOMMU_CMDLINE_DISABLED, 215 }; 216 217 /* Early ioapic and hpet maps from kernel command line */ 218 #define EARLY_MAP_SIZE 4 219 static struct devid_map __initdata early_ioapic_map[EARLY_MAP_SIZE]; 220 static struct devid_map __initdata early_hpet_map[EARLY_MAP_SIZE]; 221 static struct acpihid_map_entry __initdata early_acpihid_map[EARLY_MAP_SIZE]; 222 223 static int __initdata early_ioapic_map_size; 224 static int __initdata early_hpet_map_size; 225 static int __initdata early_acpihid_map_size; 226 227 static bool __initdata cmdline_maps; 228 229 static enum iommu_init_state init_state = IOMMU_START_STATE; 230 231 static int amd_iommu_enable_interrupts(void); 232 static int __init iommu_go_to_state(enum iommu_init_state state); 233 static void init_device_table_dma(struct amd_iommu_pci_seg *pci_seg); 234 235 static bool amd_iommu_pre_enabled = true; 236 237 static u32 amd_iommu_ivinfo __initdata; 238 239 bool translation_pre_enabled(struct amd_iommu *iommu) 240 { 241 return (iommu->flags & AMD_IOMMU_FLAG_TRANS_PRE_ENABLED); 242 } 243 244 static void clear_translation_pre_enabled(struct amd_iommu *iommu) 245 { 246 iommu->flags &= ~AMD_IOMMU_FLAG_TRANS_PRE_ENABLED; 247 } 248 249 static void init_translation_status(struct amd_iommu *iommu) 250 { 251 u64 ctrl; 252 253 ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET); 254 if (ctrl & (1<<CONTROL_IOMMU_EN)) 255 iommu->flags |= AMD_IOMMU_FLAG_TRANS_PRE_ENABLED; 256 } 257 258 static inline unsigned long tbl_size(int entry_size, int last_bdf) 259 { 260 unsigned shift = PAGE_SHIFT + 261 get_order((last_bdf + 1) * entry_size); 262 263 return 1UL << shift; 264 } 265 266 int amd_iommu_get_num_iommus(void) 267 { 268 return amd_iommus_present; 269 } 270 271 /* 272 * Iterate through all the IOMMUs to get common EFR 273 * masks among all IOMMUs and warn if found inconsistency. 274 */ 275 static void get_global_efr(void) 276 { 277 struct amd_iommu *iommu; 278 279 for_each_iommu(iommu) { 280 u64 tmp = iommu->features; 281 u64 tmp2 = iommu->features2; 282 283 if (list_is_first(&iommu->list, &amd_iommu_list)) { 284 amd_iommu_efr = tmp; 285 amd_iommu_efr2 = tmp2; 286 continue; 287 } 288 289 if (amd_iommu_efr == tmp && 290 amd_iommu_efr2 == tmp2) 291 continue; 292 293 pr_err(FW_BUG 294 "Found inconsistent EFR/EFR2 %#llx,%#llx (global %#llx,%#llx) on iommu%d (%04x:%02x:%02x.%01x).\n", 295 tmp, tmp2, amd_iommu_efr, amd_iommu_efr2, 296 iommu->index, iommu->pci_seg->id, 297 PCI_BUS_NUM(iommu->devid), PCI_SLOT(iommu->devid), 298 PCI_FUNC(iommu->devid)); 299 300 amd_iommu_efr &= tmp; 301 amd_iommu_efr2 &= tmp2; 302 } 303 304 pr_info("Using global IVHD EFR:%#llx, EFR2:%#llx\n", amd_iommu_efr, amd_iommu_efr2); 305 } 306 307 static bool check_feature_on_all_iommus(u64 mask) 308 { 309 return !!(amd_iommu_efr & mask); 310 } 311 312 static inline int check_feature_gpt_level(void) 313 { 314 return ((amd_iommu_efr >> FEATURE_GATS_SHIFT) & FEATURE_GATS_MASK); 315 } 316 317 /* 318 * For IVHD type 0x11/0x40, EFR is also available via IVHD. 319 * Default to IVHD EFR since it is available sooner 320 * (i.e. before PCI init). 321 */ 322 static void __init early_iommu_features_init(struct amd_iommu *iommu, 323 struct ivhd_header *h) 324 { 325 if (amd_iommu_ivinfo & IOMMU_IVINFO_EFRSUP) { 326 iommu->features = h->efr_reg; 327 iommu->features2 = h->efr_reg2; 328 } 329 if (amd_iommu_ivinfo & IOMMU_IVINFO_DMA_REMAP) 330 amdr_ivrs_remap_support = true; 331 } 332 333 /* Access to l1 and l2 indexed register spaces */ 334 335 static u32 iommu_read_l1(struct amd_iommu *iommu, u16 l1, u8 address) 336 { 337 u32 val; 338 339 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16)); 340 pci_read_config_dword(iommu->dev, 0xfc, &val); 341 return val; 342 } 343 344 static void iommu_write_l1(struct amd_iommu *iommu, u16 l1, u8 address, u32 val) 345 { 346 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16 | 1 << 31)); 347 pci_write_config_dword(iommu->dev, 0xfc, val); 348 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16)); 349 } 350 351 static u32 iommu_read_l2(struct amd_iommu *iommu, u8 address) 352 { 353 u32 val; 354 355 pci_write_config_dword(iommu->dev, 0xf0, address); 356 pci_read_config_dword(iommu->dev, 0xf4, &val); 357 return val; 358 } 359 360 static void iommu_write_l2(struct amd_iommu *iommu, u8 address, u32 val) 361 { 362 pci_write_config_dword(iommu->dev, 0xf0, (address | 1 << 8)); 363 pci_write_config_dword(iommu->dev, 0xf4, val); 364 } 365 366 /**************************************************************************** 367 * 368 * AMD IOMMU MMIO register space handling functions 369 * 370 * These functions are used to program the IOMMU device registers in 371 * MMIO space required for that driver. 372 * 373 ****************************************************************************/ 374 375 /* 376 * This function set the exclusion range in the IOMMU. DMA accesses to the 377 * exclusion range are passed through untranslated 378 */ 379 static void iommu_set_exclusion_range(struct amd_iommu *iommu) 380 { 381 u64 start = iommu->exclusion_start & PAGE_MASK; 382 u64 limit = (start + iommu->exclusion_length - 1) & PAGE_MASK; 383 u64 entry; 384 385 if (!iommu->exclusion_start) 386 return; 387 388 entry = start | MMIO_EXCL_ENABLE_MASK; 389 memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET, 390 &entry, sizeof(entry)); 391 392 entry = limit; 393 memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET, 394 &entry, sizeof(entry)); 395 } 396 397 static void iommu_set_cwwb_range(struct amd_iommu *iommu) 398 { 399 u64 start = iommu_virt_to_phys((void *)iommu->cmd_sem); 400 u64 entry = start & PM_ADDR_MASK; 401 402 if (!check_feature_on_all_iommus(FEATURE_SNP)) 403 return; 404 405 /* Note: 406 * Re-purpose Exclusion base/limit registers for Completion wait 407 * write-back base/limit. 408 */ 409 memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET, 410 &entry, sizeof(entry)); 411 412 /* Note: 413 * Default to 4 Kbytes, which can be specified by setting base 414 * address equal to the limit address. 415 */ 416 memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET, 417 &entry, sizeof(entry)); 418 } 419 420 /* Programs the physical address of the device table into the IOMMU hardware */ 421 static void iommu_set_device_table(struct amd_iommu *iommu) 422 { 423 u64 entry; 424 u32 dev_table_size = iommu->pci_seg->dev_table_size; 425 void *dev_table = (void *)get_dev_table(iommu); 426 427 BUG_ON(iommu->mmio_base == NULL); 428 429 entry = iommu_virt_to_phys(dev_table); 430 entry |= (dev_table_size >> 12) - 1; 431 memcpy_toio(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET, 432 &entry, sizeof(entry)); 433 } 434 435 /* Generic functions to enable/disable certain features of the IOMMU. */ 436 static void iommu_feature_enable(struct amd_iommu *iommu, u8 bit) 437 { 438 u64 ctrl; 439 440 ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET); 441 ctrl |= (1ULL << bit); 442 writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET); 443 } 444 445 static void iommu_feature_disable(struct amd_iommu *iommu, u8 bit) 446 { 447 u64 ctrl; 448 449 ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET); 450 ctrl &= ~(1ULL << bit); 451 writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET); 452 } 453 454 static void iommu_set_inv_tlb_timeout(struct amd_iommu *iommu, int timeout) 455 { 456 u64 ctrl; 457 458 ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET); 459 ctrl &= ~CTRL_INV_TO_MASK; 460 ctrl |= (timeout << CONTROL_INV_TIMEOUT) & CTRL_INV_TO_MASK; 461 writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET); 462 } 463 464 /* Function to enable the hardware */ 465 static void iommu_enable(struct amd_iommu *iommu) 466 { 467 iommu_feature_enable(iommu, CONTROL_IOMMU_EN); 468 } 469 470 static void iommu_disable(struct amd_iommu *iommu) 471 { 472 if (!iommu->mmio_base) 473 return; 474 475 /* Disable command buffer */ 476 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN); 477 478 /* Disable event logging and event interrupts */ 479 iommu_feature_disable(iommu, CONTROL_EVT_INT_EN); 480 iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN); 481 482 /* Disable IOMMU GA_LOG */ 483 iommu_feature_disable(iommu, CONTROL_GALOG_EN); 484 iommu_feature_disable(iommu, CONTROL_GAINT_EN); 485 486 /* Disable IOMMU PPR logging */ 487 iommu_feature_disable(iommu, CONTROL_PPRLOG_EN); 488 iommu_feature_disable(iommu, CONTROL_PPRINT_EN); 489 490 /* Disable IOMMU hardware itself */ 491 iommu_feature_disable(iommu, CONTROL_IOMMU_EN); 492 493 /* Clear IRTE cache disabling bit */ 494 iommu_feature_disable(iommu, CONTROL_IRTCACHEDIS); 495 } 496 497 /* 498 * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in 499 * the system has one. 500 */ 501 static u8 __iomem * __init iommu_map_mmio_space(u64 address, u64 end) 502 { 503 if (!request_mem_region(address, end, "amd_iommu")) { 504 pr_err("Can not reserve memory region %llx-%llx for mmio\n", 505 address, end); 506 pr_err("This is a BIOS bug. Please contact your hardware vendor\n"); 507 return NULL; 508 } 509 510 return (u8 __iomem *)ioremap(address, end); 511 } 512 513 static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu) 514 { 515 if (iommu->mmio_base) 516 iounmap(iommu->mmio_base); 517 release_mem_region(iommu->mmio_phys, iommu->mmio_phys_end); 518 } 519 520 static inline u32 get_ivhd_header_size(struct ivhd_header *h) 521 { 522 u32 size = 0; 523 524 switch (h->type) { 525 case 0x10: 526 size = 24; 527 break; 528 case 0x11: 529 case 0x40: 530 size = 40; 531 break; 532 } 533 return size; 534 } 535 536 /**************************************************************************** 537 * 538 * The functions below belong to the first pass of AMD IOMMU ACPI table 539 * parsing. In this pass we try to find out the highest device id this 540 * code has to handle. Upon this information the size of the shared data 541 * structures is determined later. 542 * 543 ****************************************************************************/ 544 545 /* 546 * This function calculates the length of a given IVHD entry 547 */ 548 static inline int ivhd_entry_length(u8 *ivhd) 549 { 550 u32 type = ((struct ivhd_entry *)ivhd)->type; 551 552 if (type < 0x80) { 553 return 0x04 << (*ivhd >> 6); 554 } else if (type == IVHD_DEV_ACPI_HID) { 555 /* For ACPI_HID, offset 21 is uid len */ 556 return *((u8 *)ivhd + 21) + 22; 557 } 558 return 0; 559 } 560 561 /* 562 * After reading the highest device id from the IOMMU PCI capability header 563 * this function looks if there is a higher device id defined in the ACPI table 564 */ 565 static int __init find_last_devid_from_ivhd(struct ivhd_header *h) 566 { 567 u8 *p = (void *)h, *end = (void *)h; 568 struct ivhd_entry *dev; 569 int last_devid = -EINVAL; 570 571 u32 ivhd_size = get_ivhd_header_size(h); 572 573 if (!ivhd_size) { 574 pr_err("Unsupported IVHD type %#x\n", h->type); 575 return -EINVAL; 576 } 577 578 p += ivhd_size; 579 end += h->length; 580 581 while (p < end) { 582 dev = (struct ivhd_entry *)p; 583 switch (dev->type) { 584 case IVHD_DEV_ALL: 585 /* Use maximum BDF value for DEV_ALL */ 586 return 0xffff; 587 case IVHD_DEV_SELECT: 588 case IVHD_DEV_RANGE_END: 589 case IVHD_DEV_ALIAS: 590 case IVHD_DEV_EXT_SELECT: 591 /* all the above subfield types refer to device ids */ 592 if (dev->devid > last_devid) 593 last_devid = dev->devid; 594 break; 595 default: 596 break; 597 } 598 p += ivhd_entry_length(p); 599 } 600 601 WARN_ON(p != end); 602 603 return last_devid; 604 } 605 606 static int __init check_ivrs_checksum(struct acpi_table_header *table) 607 { 608 int i; 609 u8 checksum = 0, *p = (u8 *)table; 610 611 for (i = 0; i < table->length; ++i) 612 checksum += p[i]; 613 if (checksum != 0) { 614 /* ACPI table corrupt */ 615 pr_err(FW_BUG "IVRS invalid checksum\n"); 616 return -ENODEV; 617 } 618 619 return 0; 620 } 621 622 /* 623 * Iterate over all IVHD entries in the ACPI table and find the highest device 624 * id which we need to handle. This is the first of three functions which parse 625 * the ACPI table. So we check the checksum here. 626 */ 627 static int __init find_last_devid_acpi(struct acpi_table_header *table, u16 pci_seg) 628 { 629 u8 *p = (u8 *)table, *end = (u8 *)table; 630 struct ivhd_header *h; 631 int last_devid, last_bdf = 0; 632 633 p += IVRS_HEADER_LENGTH; 634 635 end += table->length; 636 while (p < end) { 637 h = (struct ivhd_header *)p; 638 if (h->pci_seg == pci_seg && 639 h->type == amd_iommu_target_ivhd_type) { 640 last_devid = find_last_devid_from_ivhd(h); 641 642 if (last_devid < 0) 643 return -EINVAL; 644 if (last_devid > last_bdf) 645 last_bdf = last_devid; 646 } 647 p += h->length; 648 } 649 WARN_ON(p != end); 650 651 return last_bdf; 652 } 653 654 /**************************************************************************** 655 * 656 * The following functions belong to the code path which parses the ACPI table 657 * the second time. In this ACPI parsing iteration we allocate IOMMU specific 658 * data structures, initialize the per PCI segment device/alias/rlookup table 659 * and also basically initialize the hardware. 660 * 661 ****************************************************************************/ 662 663 /* Allocate per PCI segment device table */ 664 static inline int __init alloc_dev_table(struct amd_iommu_pci_seg *pci_seg) 665 { 666 pci_seg->dev_table = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO | GFP_DMA32, 667 get_order(pci_seg->dev_table_size)); 668 if (!pci_seg->dev_table) 669 return -ENOMEM; 670 671 return 0; 672 } 673 674 static inline void free_dev_table(struct amd_iommu_pci_seg *pci_seg) 675 { 676 free_pages((unsigned long)pci_seg->dev_table, 677 get_order(pci_seg->dev_table_size)); 678 pci_seg->dev_table = NULL; 679 } 680 681 /* Allocate per PCI segment IOMMU rlookup table. */ 682 static inline int __init alloc_rlookup_table(struct amd_iommu_pci_seg *pci_seg) 683 { 684 pci_seg->rlookup_table = (void *)__get_free_pages( 685 GFP_KERNEL | __GFP_ZERO, 686 get_order(pci_seg->rlookup_table_size)); 687 if (pci_seg->rlookup_table == NULL) 688 return -ENOMEM; 689 690 return 0; 691 } 692 693 static inline void free_rlookup_table(struct amd_iommu_pci_seg *pci_seg) 694 { 695 free_pages((unsigned long)pci_seg->rlookup_table, 696 get_order(pci_seg->rlookup_table_size)); 697 pci_seg->rlookup_table = NULL; 698 } 699 700 static inline int __init alloc_irq_lookup_table(struct amd_iommu_pci_seg *pci_seg) 701 { 702 pci_seg->irq_lookup_table = (void *)__get_free_pages( 703 GFP_KERNEL | __GFP_ZERO, 704 get_order(pci_seg->rlookup_table_size)); 705 kmemleak_alloc(pci_seg->irq_lookup_table, 706 pci_seg->rlookup_table_size, 1, GFP_KERNEL); 707 if (pci_seg->irq_lookup_table == NULL) 708 return -ENOMEM; 709 710 return 0; 711 } 712 713 static inline void free_irq_lookup_table(struct amd_iommu_pci_seg *pci_seg) 714 { 715 kmemleak_free(pci_seg->irq_lookup_table); 716 free_pages((unsigned long)pci_seg->irq_lookup_table, 717 get_order(pci_seg->rlookup_table_size)); 718 pci_seg->irq_lookup_table = NULL; 719 } 720 721 static int __init alloc_alias_table(struct amd_iommu_pci_seg *pci_seg) 722 { 723 int i; 724 725 pci_seg->alias_table = (void *)__get_free_pages(GFP_KERNEL, 726 get_order(pci_seg->alias_table_size)); 727 if (!pci_seg->alias_table) 728 return -ENOMEM; 729 730 /* 731 * let all alias entries point to itself 732 */ 733 for (i = 0; i <= pci_seg->last_bdf; ++i) 734 pci_seg->alias_table[i] = i; 735 736 return 0; 737 } 738 739 static void __init free_alias_table(struct amd_iommu_pci_seg *pci_seg) 740 { 741 free_pages((unsigned long)pci_seg->alias_table, 742 get_order(pci_seg->alias_table_size)); 743 pci_seg->alias_table = NULL; 744 } 745 746 /* 747 * Allocates the command buffer. This buffer is per AMD IOMMU. We can 748 * write commands to that buffer later and the IOMMU will execute them 749 * asynchronously 750 */ 751 static int __init alloc_command_buffer(struct amd_iommu *iommu) 752 { 753 iommu->cmd_buf = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 754 get_order(CMD_BUFFER_SIZE)); 755 756 return iommu->cmd_buf ? 0 : -ENOMEM; 757 } 758 759 /* 760 * Interrupt handler has processed all pending events and adjusted head 761 * and tail pointer. Reset overflow mask and restart logging again. 762 */ 763 static void amd_iommu_restart_log(struct amd_iommu *iommu, const char *evt_type, 764 u8 cntrl_intr, u8 cntrl_log, 765 u32 status_run_mask, u32 status_overflow_mask) 766 { 767 u32 status; 768 769 status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET); 770 if (status & status_run_mask) 771 return; 772 773 pr_info_ratelimited("IOMMU %s log restarting\n", evt_type); 774 775 iommu_feature_disable(iommu, cntrl_log); 776 iommu_feature_disable(iommu, cntrl_intr); 777 778 writel(status_overflow_mask, iommu->mmio_base + MMIO_STATUS_OFFSET); 779 780 iommu_feature_enable(iommu, cntrl_intr); 781 iommu_feature_enable(iommu, cntrl_log); 782 } 783 784 /* 785 * This function restarts event logging in case the IOMMU experienced 786 * an event log buffer overflow. 787 */ 788 void amd_iommu_restart_event_logging(struct amd_iommu *iommu) 789 { 790 amd_iommu_restart_log(iommu, "Event", CONTROL_EVT_INT_EN, 791 CONTROL_EVT_LOG_EN, MMIO_STATUS_EVT_RUN_MASK, 792 MMIO_STATUS_EVT_OVERFLOW_MASK); 793 } 794 795 /* 796 * This function restarts event logging in case the IOMMU experienced 797 * GA log overflow. 798 */ 799 void amd_iommu_restart_ga_log(struct amd_iommu *iommu) 800 { 801 amd_iommu_restart_log(iommu, "GA", CONTROL_GAINT_EN, 802 CONTROL_GALOG_EN, MMIO_STATUS_GALOG_RUN_MASK, 803 MMIO_STATUS_GALOG_OVERFLOW_MASK); 804 } 805 806 /* 807 * This function restarts ppr logging in case the IOMMU experienced 808 * PPR log overflow. 809 */ 810 void amd_iommu_restart_ppr_log(struct amd_iommu *iommu) 811 { 812 amd_iommu_restart_log(iommu, "PPR", CONTROL_PPRINT_EN, 813 CONTROL_PPRLOG_EN, MMIO_STATUS_PPR_RUN_MASK, 814 MMIO_STATUS_PPR_OVERFLOW_MASK); 815 } 816 817 /* 818 * This function resets the command buffer if the IOMMU stopped fetching 819 * commands from it. 820 */ 821 static void amd_iommu_reset_cmd_buffer(struct amd_iommu *iommu) 822 { 823 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN); 824 825 writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET); 826 writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET); 827 iommu->cmd_buf_head = 0; 828 iommu->cmd_buf_tail = 0; 829 830 iommu_feature_enable(iommu, CONTROL_CMDBUF_EN); 831 } 832 833 /* 834 * This function writes the command buffer address to the hardware and 835 * enables it. 836 */ 837 static void iommu_enable_command_buffer(struct amd_iommu *iommu) 838 { 839 u64 entry; 840 841 BUG_ON(iommu->cmd_buf == NULL); 842 843 entry = iommu_virt_to_phys(iommu->cmd_buf); 844 entry |= MMIO_CMD_SIZE_512; 845 846 memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET, 847 &entry, sizeof(entry)); 848 849 amd_iommu_reset_cmd_buffer(iommu); 850 } 851 852 /* 853 * This function disables the command buffer 854 */ 855 static void iommu_disable_command_buffer(struct amd_iommu *iommu) 856 { 857 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN); 858 } 859 860 static void __init free_command_buffer(struct amd_iommu *iommu) 861 { 862 free_pages((unsigned long)iommu->cmd_buf, get_order(CMD_BUFFER_SIZE)); 863 } 864 865 static void *__init iommu_alloc_4k_pages(struct amd_iommu *iommu, 866 gfp_t gfp, size_t size) 867 { 868 int order = get_order(size); 869 void *buf = (void *)__get_free_pages(gfp, order); 870 871 if (buf && 872 check_feature_on_all_iommus(FEATURE_SNP) && 873 set_memory_4k((unsigned long)buf, (1 << order))) { 874 free_pages((unsigned long)buf, order); 875 buf = NULL; 876 } 877 878 return buf; 879 } 880 881 /* allocates the memory where the IOMMU will log its events to */ 882 static int __init alloc_event_buffer(struct amd_iommu *iommu) 883 { 884 iommu->evt_buf = iommu_alloc_4k_pages(iommu, GFP_KERNEL | __GFP_ZERO, 885 EVT_BUFFER_SIZE); 886 887 return iommu->evt_buf ? 0 : -ENOMEM; 888 } 889 890 static void iommu_enable_event_buffer(struct amd_iommu *iommu) 891 { 892 u64 entry; 893 894 BUG_ON(iommu->evt_buf == NULL); 895 896 entry = iommu_virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK; 897 898 memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET, 899 &entry, sizeof(entry)); 900 901 /* set head and tail to zero manually */ 902 writel(0x00, iommu->mmio_base + MMIO_EVT_HEAD_OFFSET); 903 writel(0x00, iommu->mmio_base + MMIO_EVT_TAIL_OFFSET); 904 905 iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN); 906 } 907 908 /* 909 * This function disables the event log buffer 910 */ 911 static void iommu_disable_event_buffer(struct amd_iommu *iommu) 912 { 913 iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN); 914 } 915 916 static void __init free_event_buffer(struct amd_iommu *iommu) 917 { 918 free_pages((unsigned long)iommu->evt_buf, get_order(EVT_BUFFER_SIZE)); 919 } 920 921 /* allocates the memory where the IOMMU will log its events to */ 922 static int __init alloc_ppr_log(struct amd_iommu *iommu) 923 { 924 iommu->ppr_log = iommu_alloc_4k_pages(iommu, GFP_KERNEL | __GFP_ZERO, 925 PPR_LOG_SIZE); 926 927 return iommu->ppr_log ? 0 : -ENOMEM; 928 } 929 930 static void iommu_enable_ppr_log(struct amd_iommu *iommu) 931 { 932 u64 entry; 933 934 if (iommu->ppr_log == NULL) 935 return; 936 937 iommu_feature_enable(iommu, CONTROL_PPR_EN); 938 939 entry = iommu_virt_to_phys(iommu->ppr_log) | PPR_LOG_SIZE_512; 940 941 memcpy_toio(iommu->mmio_base + MMIO_PPR_LOG_OFFSET, 942 &entry, sizeof(entry)); 943 944 /* set head and tail to zero manually */ 945 writel(0x00, iommu->mmio_base + MMIO_PPR_HEAD_OFFSET); 946 writel(0x00, iommu->mmio_base + MMIO_PPR_TAIL_OFFSET); 947 948 iommu_feature_enable(iommu, CONTROL_PPRLOG_EN); 949 iommu_feature_enable(iommu, CONTROL_PPRINT_EN); 950 } 951 952 static void __init free_ppr_log(struct amd_iommu *iommu) 953 { 954 free_pages((unsigned long)iommu->ppr_log, get_order(PPR_LOG_SIZE)); 955 } 956 957 static void free_ga_log(struct amd_iommu *iommu) 958 { 959 #ifdef CONFIG_IRQ_REMAP 960 free_pages((unsigned long)iommu->ga_log, get_order(GA_LOG_SIZE)); 961 free_pages((unsigned long)iommu->ga_log_tail, get_order(8)); 962 #endif 963 } 964 965 #ifdef CONFIG_IRQ_REMAP 966 static int iommu_ga_log_enable(struct amd_iommu *iommu) 967 { 968 u32 status, i; 969 u64 entry; 970 971 if (!iommu->ga_log) 972 return -EINVAL; 973 974 entry = iommu_virt_to_phys(iommu->ga_log) | GA_LOG_SIZE_512; 975 memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_BASE_OFFSET, 976 &entry, sizeof(entry)); 977 entry = (iommu_virt_to_phys(iommu->ga_log_tail) & 978 (BIT_ULL(52)-1)) & ~7ULL; 979 memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_TAIL_OFFSET, 980 &entry, sizeof(entry)); 981 writel(0x00, iommu->mmio_base + MMIO_GA_HEAD_OFFSET); 982 writel(0x00, iommu->mmio_base + MMIO_GA_TAIL_OFFSET); 983 984 985 iommu_feature_enable(iommu, CONTROL_GAINT_EN); 986 iommu_feature_enable(iommu, CONTROL_GALOG_EN); 987 988 for (i = 0; i < LOOP_TIMEOUT; ++i) { 989 status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET); 990 if (status & (MMIO_STATUS_GALOG_RUN_MASK)) 991 break; 992 udelay(10); 993 } 994 995 if (WARN_ON(i >= LOOP_TIMEOUT)) 996 return -EINVAL; 997 998 return 0; 999 } 1000 1001 static int iommu_init_ga_log(struct amd_iommu *iommu) 1002 { 1003 if (!AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir)) 1004 return 0; 1005 1006 iommu->ga_log = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 1007 get_order(GA_LOG_SIZE)); 1008 if (!iommu->ga_log) 1009 goto err_out; 1010 1011 iommu->ga_log_tail = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 1012 get_order(8)); 1013 if (!iommu->ga_log_tail) 1014 goto err_out; 1015 1016 return 0; 1017 err_out: 1018 free_ga_log(iommu); 1019 return -EINVAL; 1020 } 1021 #endif /* CONFIG_IRQ_REMAP */ 1022 1023 static int __init alloc_cwwb_sem(struct amd_iommu *iommu) 1024 { 1025 iommu->cmd_sem = iommu_alloc_4k_pages(iommu, GFP_KERNEL | __GFP_ZERO, 1); 1026 1027 return iommu->cmd_sem ? 0 : -ENOMEM; 1028 } 1029 1030 static void __init free_cwwb_sem(struct amd_iommu *iommu) 1031 { 1032 if (iommu->cmd_sem) 1033 free_page((unsigned long)iommu->cmd_sem); 1034 } 1035 1036 static void iommu_enable_xt(struct amd_iommu *iommu) 1037 { 1038 #ifdef CONFIG_IRQ_REMAP 1039 /* 1040 * XT mode (32-bit APIC destination ID) requires 1041 * GA mode (128-bit IRTE support) as a prerequisite. 1042 */ 1043 if (AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir) && 1044 amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE) 1045 iommu_feature_enable(iommu, CONTROL_XT_EN); 1046 #endif /* CONFIG_IRQ_REMAP */ 1047 } 1048 1049 static void iommu_enable_gt(struct amd_iommu *iommu) 1050 { 1051 if (!iommu_feature(iommu, FEATURE_GT)) 1052 return; 1053 1054 iommu_feature_enable(iommu, CONTROL_GT_EN); 1055 } 1056 1057 /* sets a specific bit in the device table entry. */ 1058 static void __set_dev_entry_bit(struct dev_table_entry *dev_table, 1059 u16 devid, u8 bit) 1060 { 1061 int i = (bit >> 6) & 0x03; 1062 int _bit = bit & 0x3f; 1063 1064 dev_table[devid].data[i] |= (1UL << _bit); 1065 } 1066 1067 static void set_dev_entry_bit(struct amd_iommu *iommu, u16 devid, u8 bit) 1068 { 1069 struct dev_table_entry *dev_table = get_dev_table(iommu); 1070 1071 return __set_dev_entry_bit(dev_table, devid, bit); 1072 } 1073 1074 static int __get_dev_entry_bit(struct dev_table_entry *dev_table, 1075 u16 devid, u8 bit) 1076 { 1077 int i = (bit >> 6) & 0x03; 1078 int _bit = bit & 0x3f; 1079 1080 return (dev_table[devid].data[i] & (1UL << _bit)) >> _bit; 1081 } 1082 1083 static int get_dev_entry_bit(struct amd_iommu *iommu, u16 devid, u8 bit) 1084 { 1085 struct dev_table_entry *dev_table = get_dev_table(iommu); 1086 1087 return __get_dev_entry_bit(dev_table, devid, bit); 1088 } 1089 1090 static bool __copy_device_table(struct amd_iommu *iommu) 1091 { 1092 u64 int_ctl, int_tab_len, entry = 0; 1093 struct amd_iommu_pci_seg *pci_seg = iommu->pci_seg; 1094 struct dev_table_entry *old_devtb = NULL; 1095 u32 lo, hi, devid, old_devtb_size; 1096 phys_addr_t old_devtb_phys; 1097 u16 dom_id, dte_v, irq_v; 1098 gfp_t gfp_flag; 1099 u64 tmp; 1100 1101 /* Each IOMMU use separate device table with the same size */ 1102 lo = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET); 1103 hi = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET + 4); 1104 entry = (((u64) hi) << 32) + lo; 1105 1106 old_devtb_size = ((entry & ~PAGE_MASK) + 1) << 12; 1107 if (old_devtb_size != pci_seg->dev_table_size) { 1108 pr_err("The device table size of IOMMU:%d is not expected!\n", 1109 iommu->index); 1110 return false; 1111 } 1112 1113 /* 1114 * When SME is enabled in the first kernel, the entry includes the 1115 * memory encryption mask(sme_me_mask), we must remove the memory 1116 * encryption mask to obtain the true physical address in kdump kernel. 1117 */ 1118 old_devtb_phys = __sme_clr(entry) & PAGE_MASK; 1119 1120 if (old_devtb_phys >= 0x100000000ULL) { 1121 pr_err("The address of old device table is above 4G, not trustworthy!\n"); 1122 return false; 1123 } 1124 old_devtb = (cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT) && is_kdump_kernel()) 1125 ? (__force void *)ioremap_encrypted(old_devtb_phys, 1126 pci_seg->dev_table_size) 1127 : memremap(old_devtb_phys, pci_seg->dev_table_size, MEMREMAP_WB); 1128 1129 if (!old_devtb) 1130 return false; 1131 1132 gfp_flag = GFP_KERNEL | __GFP_ZERO | GFP_DMA32; 1133 pci_seg->old_dev_tbl_cpy = (void *)__get_free_pages(gfp_flag, 1134 get_order(pci_seg->dev_table_size)); 1135 if (pci_seg->old_dev_tbl_cpy == NULL) { 1136 pr_err("Failed to allocate memory for copying old device table!\n"); 1137 memunmap(old_devtb); 1138 return false; 1139 } 1140 1141 for (devid = 0; devid <= pci_seg->last_bdf; ++devid) { 1142 pci_seg->old_dev_tbl_cpy[devid] = old_devtb[devid]; 1143 dom_id = old_devtb[devid].data[1] & DEV_DOMID_MASK; 1144 dte_v = old_devtb[devid].data[0] & DTE_FLAG_V; 1145 1146 if (dte_v && dom_id) { 1147 pci_seg->old_dev_tbl_cpy[devid].data[0] = old_devtb[devid].data[0]; 1148 pci_seg->old_dev_tbl_cpy[devid].data[1] = old_devtb[devid].data[1]; 1149 __set_bit(dom_id, amd_iommu_pd_alloc_bitmap); 1150 /* If gcr3 table existed, mask it out */ 1151 if (old_devtb[devid].data[0] & DTE_FLAG_GV) { 1152 tmp = DTE_GCR3_VAL_B(~0ULL) << DTE_GCR3_SHIFT_B; 1153 tmp |= DTE_GCR3_VAL_C(~0ULL) << DTE_GCR3_SHIFT_C; 1154 pci_seg->old_dev_tbl_cpy[devid].data[1] &= ~tmp; 1155 tmp = DTE_GCR3_VAL_A(~0ULL) << DTE_GCR3_SHIFT_A; 1156 tmp |= DTE_FLAG_GV; 1157 pci_seg->old_dev_tbl_cpy[devid].data[0] &= ~tmp; 1158 } 1159 } 1160 1161 irq_v = old_devtb[devid].data[2] & DTE_IRQ_REMAP_ENABLE; 1162 int_ctl = old_devtb[devid].data[2] & DTE_IRQ_REMAP_INTCTL_MASK; 1163 int_tab_len = old_devtb[devid].data[2] & DTE_INTTABLEN_MASK; 1164 if (irq_v && (int_ctl || int_tab_len)) { 1165 if ((int_ctl != DTE_IRQ_REMAP_INTCTL) || 1166 (int_tab_len != DTE_INTTABLEN)) { 1167 pr_err("Wrong old irq remapping flag: %#x\n", devid); 1168 memunmap(old_devtb); 1169 return false; 1170 } 1171 1172 pci_seg->old_dev_tbl_cpy[devid].data[2] = old_devtb[devid].data[2]; 1173 } 1174 } 1175 memunmap(old_devtb); 1176 1177 return true; 1178 } 1179 1180 static bool copy_device_table(void) 1181 { 1182 struct amd_iommu *iommu; 1183 struct amd_iommu_pci_seg *pci_seg; 1184 1185 if (!amd_iommu_pre_enabled) 1186 return false; 1187 1188 pr_warn("Translation is already enabled - trying to copy translation structures\n"); 1189 1190 /* 1191 * All IOMMUs within PCI segment shares common device table. 1192 * Hence copy device table only once per PCI segment. 1193 */ 1194 for_each_pci_segment(pci_seg) { 1195 for_each_iommu(iommu) { 1196 if (pci_seg->id != iommu->pci_seg->id) 1197 continue; 1198 if (!__copy_device_table(iommu)) 1199 return false; 1200 break; 1201 } 1202 } 1203 1204 return true; 1205 } 1206 1207 void amd_iommu_apply_erratum_63(struct amd_iommu *iommu, u16 devid) 1208 { 1209 int sysmgt; 1210 1211 sysmgt = get_dev_entry_bit(iommu, devid, DEV_ENTRY_SYSMGT1) | 1212 (get_dev_entry_bit(iommu, devid, DEV_ENTRY_SYSMGT2) << 1); 1213 1214 if (sysmgt == 0x01) 1215 set_dev_entry_bit(iommu, devid, DEV_ENTRY_IW); 1216 } 1217 1218 /* 1219 * This function takes the device specific flags read from the ACPI 1220 * table and sets up the device table entry with that information 1221 */ 1222 static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu, 1223 u16 devid, u32 flags, u32 ext_flags) 1224 { 1225 if (flags & ACPI_DEVFLAG_INITPASS) 1226 set_dev_entry_bit(iommu, devid, DEV_ENTRY_INIT_PASS); 1227 if (flags & ACPI_DEVFLAG_EXTINT) 1228 set_dev_entry_bit(iommu, devid, DEV_ENTRY_EINT_PASS); 1229 if (flags & ACPI_DEVFLAG_NMI) 1230 set_dev_entry_bit(iommu, devid, DEV_ENTRY_NMI_PASS); 1231 if (flags & ACPI_DEVFLAG_SYSMGT1) 1232 set_dev_entry_bit(iommu, devid, DEV_ENTRY_SYSMGT1); 1233 if (flags & ACPI_DEVFLAG_SYSMGT2) 1234 set_dev_entry_bit(iommu, devid, DEV_ENTRY_SYSMGT2); 1235 if (flags & ACPI_DEVFLAG_LINT0) 1236 set_dev_entry_bit(iommu, devid, DEV_ENTRY_LINT0_PASS); 1237 if (flags & ACPI_DEVFLAG_LINT1) 1238 set_dev_entry_bit(iommu, devid, DEV_ENTRY_LINT1_PASS); 1239 1240 amd_iommu_apply_erratum_63(iommu, devid); 1241 1242 amd_iommu_set_rlookup_table(iommu, devid); 1243 } 1244 1245 int __init add_special_device(u8 type, u8 id, u32 *devid, bool cmd_line) 1246 { 1247 struct devid_map *entry; 1248 struct list_head *list; 1249 1250 if (type == IVHD_SPECIAL_IOAPIC) 1251 list = &ioapic_map; 1252 else if (type == IVHD_SPECIAL_HPET) 1253 list = &hpet_map; 1254 else 1255 return -EINVAL; 1256 1257 list_for_each_entry(entry, list, list) { 1258 if (!(entry->id == id && entry->cmd_line)) 1259 continue; 1260 1261 pr_info("Command-line override present for %s id %d - ignoring\n", 1262 type == IVHD_SPECIAL_IOAPIC ? "IOAPIC" : "HPET", id); 1263 1264 *devid = entry->devid; 1265 1266 return 0; 1267 } 1268 1269 entry = kzalloc(sizeof(*entry), GFP_KERNEL); 1270 if (!entry) 1271 return -ENOMEM; 1272 1273 entry->id = id; 1274 entry->devid = *devid; 1275 entry->cmd_line = cmd_line; 1276 1277 list_add_tail(&entry->list, list); 1278 1279 return 0; 1280 } 1281 1282 static int __init add_acpi_hid_device(u8 *hid, u8 *uid, u32 *devid, 1283 bool cmd_line) 1284 { 1285 struct acpihid_map_entry *entry; 1286 struct list_head *list = &acpihid_map; 1287 1288 list_for_each_entry(entry, list, list) { 1289 if (strcmp(entry->hid, hid) || 1290 (*uid && *entry->uid && strcmp(entry->uid, uid)) || 1291 !entry->cmd_line) 1292 continue; 1293 1294 pr_info("Command-line override for hid:%s uid:%s\n", 1295 hid, uid); 1296 *devid = entry->devid; 1297 return 0; 1298 } 1299 1300 entry = kzalloc(sizeof(*entry), GFP_KERNEL); 1301 if (!entry) 1302 return -ENOMEM; 1303 1304 memcpy(entry->uid, uid, strlen(uid)); 1305 memcpy(entry->hid, hid, strlen(hid)); 1306 entry->devid = *devid; 1307 entry->cmd_line = cmd_line; 1308 entry->root_devid = (entry->devid & (~0x7)); 1309 1310 pr_info("%s, add hid:%s, uid:%s, rdevid:%d\n", 1311 entry->cmd_line ? "cmd" : "ivrs", 1312 entry->hid, entry->uid, entry->root_devid); 1313 1314 list_add_tail(&entry->list, list); 1315 return 0; 1316 } 1317 1318 static int __init add_early_maps(void) 1319 { 1320 int i, ret; 1321 1322 for (i = 0; i < early_ioapic_map_size; ++i) { 1323 ret = add_special_device(IVHD_SPECIAL_IOAPIC, 1324 early_ioapic_map[i].id, 1325 &early_ioapic_map[i].devid, 1326 early_ioapic_map[i].cmd_line); 1327 if (ret) 1328 return ret; 1329 } 1330 1331 for (i = 0; i < early_hpet_map_size; ++i) { 1332 ret = add_special_device(IVHD_SPECIAL_HPET, 1333 early_hpet_map[i].id, 1334 &early_hpet_map[i].devid, 1335 early_hpet_map[i].cmd_line); 1336 if (ret) 1337 return ret; 1338 } 1339 1340 for (i = 0; i < early_acpihid_map_size; ++i) { 1341 ret = add_acpi_hid_device(early_acpihid_map[i].hid, 1342 early_acpihid_map[i].uid, 1343 &early_acpihid_map[i].devid, 1344 early_acpihid_map[i].cmd_line); 1345 if (ret) 1346 return ret; 1347 } 1348 1349 return 0; 1350 } 1351 1352 /* 1353 * Takes a pointer to an AMD IOMMU entry in the ACPI table and 1354 * initializes the hardware and our data structures with it. 1355 */ 1356 static int __init init_iommu_from_acpi(struct amd_iommu *iommu, 1357 struct ivhd_header *h) 1358 { 1359 u8 *p = (u8 *)h; 1360 u8 *end = p, flags = 0; 1361 u16 devid = 0, devid_start = 0, devid_to = 0, seg_id; 1362 u32 dev_i, ext_flags = 0; 1363 bool alias = false; 1364 struct ivhd_entry *e; 1365 struct amd_iommu_pci_seg *pci_seg = iommu->pci_seg; 1366 u32 ivhd_size; 1367 int ret; 1368 1369 1370 ret = add_early_maps(); 1371 if (ret) 1372 return ret; 1373 1374 amd_iommu_apply_ivrs_quirks(); 1375 1376 /* 1377 * First save the recommended feature enable bits from ACPI 1378 */ 1379 iommu->acpi_flags = h->flags; 1380 1381 /* 1382 * Done. Now parse the device entries 1383 */ 1384 ivhd_size = get_ivhd_header_size(h); 1385 if (!ivhd_size) { 1386 pr_err("Unsupported IVHD type %#x\n", h->type); 1387 return -EINVAL; 1388 } 1389 1390 p += ivhd_size; 1391 1392 end += h->length; 1393 1394 1395 while (p < end) { 1396 e = (struct ivhd_entry *)p; 1397 seg_id = pci_seg->id; 1398 1399 switch (e->type) { 1400 case IVHD_DEV_ALL: 1401 1402 DUMP_printk(" DEV_ALL\t\t\tflags: %02x\n", e->flags); 1403 1404 for (dev_i = 0; dev_i <= pci_seg->last_bdf; ++dev_i) 1405 set_dev_entry_from_acpi(iommu, dev_i, e->flags, 0); 1406 break; 1407 case IVHD_DEV_SELECT: 1408 1409 DUMP_printk(" DEV_SELECT\t\t\t devid: %04x:%02x:%02x.%x " 1410 "flags: %02x\n", 1411 seg_id, PCI_BUS_NUM(e->devid), 1412 PCI_SLOT(e->devid), 1413 PCI_FUNC(e->devid), 1414 e->flags); 1415 1416 devid = e->devid; 1417 set_dev_entry_from_acpi(iommu, devid, e->flags, 0); 1418 break; 1419 case IVHD_DEV_SELECT_RANGE_START: 1420 1421 DUMP_printk(" DEV_SELECT_RANGE_START\t " 1422 "devid: %04x:%02x:%02x.%x flags: %02x\n", 1423 seg_id, PCI_BUS_NUM(e->devid), 1424 PCI_SLOT(e->devid), 1425 PCI_FUNC(e->devid), 1426 e->flags); 1427 1428 devid_start = e->devid; 1429 flags = e->flags; 1430 ext_flags = 0; 1431 alias = false; 1432 break; 1433 case IVHD_DEV_ALIAS: 1434 1435 DUMP_printk(" DEV_ALIAS\t\t\t devid: %04x:%02x:%02x.%x " 1436 "flags: %02x devid_to: %02x:%02x.%x\n", 1437 seg_id, PCI_BUS_NUM(e->devid), 1438 PCI_SLOT(e->devid), 1439 PCI_FUNC(e->devid), 1440 e->flags, 1441 PCI_BUS_NUM(e->ext >> 8), 1442 PCI_SLOT(e->ext >> 8), 1443 PCI_FUNC(e->ext >> 8)); 1444 1445 devid = e->devid; 1446 devid_to = e->ext >> 8; 1447 set_dev_entry_from_acpi(iommu, devid , e->flags, 0); 1448 set_dev_entry_from_acpi(iommu, devid_to, e->flags, 0); 1449 pci_seg->alias_table[devid] = devid_to; 1450 break; 1451 case IVHD_DEV_ALIAS_RANGE: 1452 1453 DUMP_printk(" DEV_ALIAS_RANGE\t\t " 1454 "devid: %04x:%02x:%02x.%x flags: %02x " 1455 "devid_to: %04x:%02x:%02x.%x\n", 1456 seg_id, PCI_BUS_NUM(e->devid), 1457 PCI_SLOT(e->devid), 1458 PCI_FUNC(e->devid), 1459 e->flags, 1460 seg_id, PCI_BUS_NUM(e->ext >> 8), 1461 PCI_SLOT(e->ext >> 8), 1462 PCI_FUNC(e->ext >> 8)); 1463 1464 devid_start = e->devid; 1465 flags = e->flags; 1466 devid_to = e->ext >> 8; 1467 ext_flags = 0; 1468 alias = true; 1469 break; 1470 case IVHD_DEV_EXT_SELECT: 1471 1472 DUMP_printk(" DEV_EXT_SELECT\t\t devid: %04x:%02x:%02x.%x " 1473 "flags: %02x ext: %08x\n", 1474 seg_id, PCI_BUS_NUM(e->devid), 1475 PCI_SLOT(e->devid), 1476 PCI_FUNC(e->devid), 1477 e->flags, e->ext); 1478 1479 devid = e->devid; 1480 set_dev_entry_from_acpi(iommu, devid, e->flags, 1481 e->ext); 1482 break; 1483 case IVHD_DEV_EXT_SELECT_RANGE: 1484 1485 DUMP_printk(" DEV_EXT_SELECT_RANGE\t devid: " 1486 "%04x:%02x:%02x.%x flags: %02x ext: %08x\n", 1487 seg_id, PCI_BUS_NUM(e->devid), 1488 PCI_SLOT(e->devid), 1489 PCI_FUNC(e->devid), 1490 e->flags, e->ext); 1491 1492 devid_start = e->devid; 1493 flags = e->flags; 1494 ext_flags = e->ext; 1495 alias = false; 1496 break; 1497 case IVHD_DEV_RANGE_END: 1498 1499 DUMP_printk(" DEV_RANGE_END\t\t devid: %04x:%02x:%02x.%x\n", 1500 seg_id, PCI_BUS_NUM(e->devid), 1501 PCI_SLOT(e->devid), 1502 PCI_FUNC(e->devid)); 1503 1504 devid = e->devid; 1505 for (dev_i = devid_start; dev_i <= devid; ++dev_i) { 1506 if (alias) { 1507 pci_seg->alias_table[dev_i] = devid_to; 1508 set_dev_entry_from_acpi(iommu, 1509 devid_to, flags, ext_flags); 1510 } 1511 set_dev_entry_from_acpi(iommu, dev_i, 1512 flags, ext_flags); 1513 } 1514 break; 1515 case IVHD_DEV_SPECIAL: { 1516 u8 handle, type; 1517 const char *var; 1518 u32 devid; 1519 int ret; 1520 1521 handle = e->ext & 0xff; 1522 devid = PCI_SEG_DEVID_TO_SBDF(seg_id, (e->ext >> 8)); 1523 type = (e->ext >> 24) & 0xff; 1524 1525 if (type == IVHD_SPECIAL_IOAPIC) 1526 var = "IOAPIC"; 1527 else if (type == IVHD_SPECIAL_HPET) 1528 var = "HPET"; 1529 else 1530 var = "UNKNOWN"; 1531 1532 DUMP_printk(" DEV_SPECIAL(%s[%d])\t\tdevid: %04x:%02x:%02x.%x\n", 1533 var, (int)handle, 1534 seg_id, PCI_BUS_NUM(devid), 1535 PCI_SLOT(devid), 1536 PCI_FUNC(devid)); 1537 1538 ret = add_special_device(type, handle, &devid, false); 1539 if (ret) 1540 return ret; 1541 1542 /* 1543 * add_special_device might update the devid in case a 1544 * command-line override is present. So call 1545 * set_dev_entry_from_acpi after add_special_device. 1546 */ 1547 set_dev_entry_from_acpi(iommu, devid, e->flags, 0); 1548 1549 break; 1550 } 1551 case IVHD_DEV_ACPI_HID: { 1552 u32 devid; 1553 u8 hid[ACPIHID_HID_LEN]; 1554 u8 uid[ACPIHID_UID_LEN]; 1555 int ret; 1556 1557 if (h->type != 0x40) { 1558 pr_err(FW_BUG "Invalid IVHD device type %#x\n", 1559 e->type); 1560 break; 1561 } 1562 1563 BUILD_BUG_ON(sizeof(e->ext_hid) != ACPIHID_HID_LEN - 1); 1564 memcpy(hid, &e->ext_hid, ACPIHID_HID_LEN - 1); 1565 hid[ACPIHID_HID_LEN - 1] = '\0'; 1566 1567 if (!(*hid)) { 1568 pr_err(FW_BUG "Invalid HID.\n"); 1569 break; 1570 } 1571 1572 uid[0] = '\0'; 1573 switch (e->uidf) { 1574 case UID_NOT_PRESENT: 1575 1576 if (e->uidl != 0) 1577 pr_warn(FW_BUG "Invalid UID length.\n"); 1578 1579 break; 1580 case UID_IS_INTEGER: 1581 1582 sprintf(uid, "%d", e->uid); 1583 1584 break; 1585 case UID_IS_CHARACTER: 1586 1587 memcpy(uid, &e->uid, e->uidl); 1588 uid[e->uidl] = '\0'; 1589 1590 break; 1591 default: 1592 break; 1593 } 1594 1595 devid = PCI_SEG_DEVID_TO_SBDF(seg_id, e->devid); 1596 DUMP_printk(" DEV_ACPI_HID(%s[%s])\t\tdevid: %04x:%02x:%02x.%x\n", 1597 hid, uid, seg_id, 1598 PCI_BUS_NUM(devid), 1599 PCI_SLOT(devid), 1600 PCI_FUNC(devid)); 1601 1602 flags = e->flags; 1603 1604 ret = add_acpi_hid_device(hid, uid, &devid, false); 1605 if (ret) 1606 return ret; 1607 1608 /* 1609 * add_special_device might update the devid in case a 1610 * command-line override is present. So call 1611 * set_dev_entry_from_acpi after add_special_device. 1612 */ 1613 set_dev_entry_from_acpi(iommu, devid, e->flags, 0); 1614 1615 break; 1616 } 1617 default: 1618 break; 1619 } 1620 1621 p += ivhd_entry_length(p); 1622 } 1623 1624 return 0; 1625 } 1626 1627 /* Allocate PCI segment data structure */ 1628 static struct amd_iommu_pci_seg *__init alloc_pci_segment(u16 id, 1629 struct acpi_table_header *ivrs_base) 1630 { 1631 struct amd_iommu_pci_seg *pci_seg; 1632 int last_bdf; 1633 1634 /* 1635 * First parse ACPI tables to find the largest Bus/Dev/Func we need to 1636 * handle in this PCI segment. Upon this information the shared data 1637 * structures for the PCI segments in the system will be allocated. 1638 */ 1639 last_bdf = find_last_devid_acpi(ivrs_base, id); 1640 if (last_bdf < 0) 1641 return NULL; 1642 1643 pci_seg = kzalloc(sizeof(struct amd_iommu_pci_seg), GFP_KERNEL); 1644 if (pci_seg == NULL) 1645 return NULL; 1646 1647 pci_seg->last_bdf = last_bdf; 1648 DUMP_printk("PCI segment : 0x%0x, last bdf : 0x%04x\n", id, last_bdf); 1649 pci_seg->dev_table_size = tbl_size(DEV_TABLE_ENTRY_SIZE, last_bdf); 1650 pci_seg->alias_table_size = tbl_size(ALIAS_TABLE_ENTRY_SIZE, last_bdf); 1651 pci_seg->rlookup_table_size = tbl_size(RLOOKUP_TABLE_ENTRY_SIZE, last_bdf); 1652 1653 pci_seg->id = id; 1654 init_llist_head(&pci_seg->dev_data_list); 1655 INIT_LIST_HEAD(&pci_seg->unity_map); 1656 list_add_tail(&pci_seg->list, &amd_iommu_pci_seg_list); 1657 1658 if (alloc_dev_table(pci_seg)) 1659 return NULL; 1660 if (alloc_alias_table(pci_seg)) 1661 return NULL; 1662 if (alloc_rlookup_table(pci_seg)) 1663 return NULL; 1664 1665 return pci_seg; 1666 } 1667 1668 static struct amd_iommu_pci_seg *__init get_pci_segment(u16 id, 1669 struct acpi_table_header *ivrs_base) 1670 { 1671 struct amd_iommu_pci_seg *pci_seg; 1672 1673 for_each_pci_segment(pci_seg) { 1674 if (pci_seg->id == id) 1675 return pci_seg; 1676 } 1677 1678 return alloc_pci_segment(id, ivrs_base); 1679 } 1680 1681 static void __init free_pci_segments(void) 1682 { 1683 struct amd_iommu_pci_seg *pci_seg, *next; 1684 1685 for_each_pci_segment_safe(pci_seg, next) { 1686 list_del(&pci_seg->list); 1687 free_irq_lookup_table(pci_seg); 1688 free_rlookup_table(pci_seg); 1689 free_alias_table(pci_seg); 1690 free_dev_table(pci_seg); 1691 kfree(pci_seg); 1692 } 1693 } 1694 1695 static void __init free_iommu_one(struct amd_iommu *iommu) 1696 { 1697 free_cwwb_sem(iommu); 1698 free_command_buffer(iommu); 1699 free_event_buffer(iommu); 1700 free_ppr_log(iommu); 1701 free_ga_log(iommu); 1702 iommu_unmap_mmio_space(iommu); 1703 } 1704 1705 static void __init free_iommu_all(void) 1706 { 1707 struct amd_iommu *iommu, *next; 1708 1709 for_each_iommu_safe(iommu, next) { 1710 list_del(&iommu->list); 1711 free_iommu_one(iommu); 1712 kfree(iommu); 1713 } 1714 } 1715 1716 /* 1717 * Family15h Model 10h-1fh erratum 746 (IOMMU Logging May Stall Translations) 1718 * Workaround: 1719 * BIOS should disable L2B micellaneous clock gating by setting 1720 * L2_L2B_CK_GATE_CONTROL[CKGateL2BMiscDisable](D0F2xF4_x90[2]) = 1b 1721 */ 1722 static void amd_iommu_erratum_746_workaround(struct amd_iommu *iommu) 1723 { 1724 u32 value; 1725 1726 if ((boot_cpu_data.x86 != 0x15) || 1727 (boot_cpu_data.x86_model < 0x10) || 1728 (boot_cpu_data.x86_model > 0x1f)) 1729 return; 1730 1731 pci_write_config_dword(iommu->dev, 0xf0, 0x90); 1732 pci_read_config_dword(iommu->dev, 0xf4, &value); 1733 1734 if (value & BIT(2)) 1735 return; 1736 1737 /* Select NB indirect register 0x90 and enable writing */ 1738 pci_write_config_dword(iommu->dev, 0xf0, 0x90 | (1 << 8)); 1739 1740 pci_write_config_dword(iommu->dev, 0xf4, value | 0x4); 1741 pci_info(iommu->dev, "Applying erratum 746 workaround\n"); 1742 1743 /* Clear the enable writing bit */ 1744 pci_write_config_dword(iommu->dev, 0xf0, 0x90); 1745 } 1746 1747 /* 1748 * Family15h Model 30h-3fh (IOMMU Mishandles ATS Write Permission) 1749 * Workaround: 1750 * BIOS should enable ATS write permission check by setting 1751 * L2_DEBUG_3[AtsIgnoreIWDis](D0F2xF4_x47[0]) = 1b 1752 */ 1753 static void amd_iommu_ats_write_check_workaround(struct amd_iommu *iommu) 1754 { 1755 u32 value; 1756 1757 if ((boot_cpu_data.x86 != 0x15) || 1758 (boot_cpu_data.x86_model < 0x30) || 1759 (boot_cpu_data.x86_model > 0x3f)) 1760 return; 1761 1762 /* Test L2_DEBUG_3[AtsIgnoreIWDis] == 1 */ 1763 value = iommu_read_l2(iommu, 0x47); 1764 1765 if (value & BIT(0)) 1766 return; 1767 1768 /* Set L2_DEBUG_3[AtsIgnoreIWDis] = 1 */ 1769 iommu_write_l2(iommu, 0x47, value | BIT(0)); 1770 1771 pci_info(iommu->dev, "Applying ATS write check workaround\n"); 1772 } 1773 1774 /* 1775 * This function glues the initialization function for one IOMMU 1776 * together and also allocates the command buffer and programs the 1777 * hardware. It does NOT enable the IOMMU. This is done afterwards. 1778 */ 1779 static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h, 1780 struct acpi_table_header *ivrs_base) 1781 { 1782 struct amd_iommu_pci_seg *pci_seg; 1783 1784 pci_seg = get_pci_segment(h->pci_seg, ivrs_base); 1785 if (pci_seg == NULL) 1786 return -ENOMEM; 1787 iommu->pci_seg = pci_seg; 1788 1789 raw_spin_lock_init(&iommu->lock); 1790 atomic64_set(&iommu->cmd_sem_val, 0); 1791 1792 /* Add IOMMU to internal data structures */ 1793 list_add_tail(&iommu->list, &amd_iommu_list); 1794 iommu->index = amd_iommus_present++; 1795 1796 if (unlikely(iommu->index >= MAX_IOMMUS)) { 1797 WARN(1, "System has more IOMMUs than supported by this driver\n"); 1798 return -ENOSYS; 1799 } 1800 1801 /* Index is fine - add IOMMU to the array */ 1802 amd_iommus[iommu->index] = iommu; 1803 1804 /* 1805 * Copy data from ACPI table entry to the iommu struct 1806 */ 1807 iommu->devid = h->devid; 1808 iommu->cap_ptr = h->cap_ptr; 1809 iommu->mmio_phys = h->mmio_phys; 1810 1811 switch (h->type) { 1812 case 0x10: 1813 /* Check if IVHD EFR contains proper max banks/counters */ 1814 if ((h->efr_attr != 0) && 1815 ((h->efr_attr & (0xF << 13)) != 0) && 1816 ((h->efr_attr & (0x3F << 17)) != 0)) 1817 iommu->mmio_phys_end = MMIO_REG_END_OFFSET; 1818 else 1819 iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET; 1820 1821 /* 1822 * Note: GA (128-bit IRTE) mode requires cmpxchg16b supports. 1823 * GAM also requires GA mode. Therefore, we need to 1824 * check cmpxchg16b support before enabling it. 1825 */ 1826 if (!boot_cpu_has(X86_FEATURE_CX16) || 1827 ((h->efr_attr & (0x1 << IOMMU_FEAT_GASUP_SHIFT)) == 0)) 1828 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY; 1829 break; 1830 case 0x11: 1831 case 0x40: 1832 if (h->efr_reg & (1 << 9)) 1833 iommu->mmio_phys_end = MMIO_REG_END_OFFSET; 1834 else 1835 iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET; 1836 1837 /* 1838 * Note: GA (128-bit IRTE) mode requires cmpxchg16b supports. 1839 * XT, GAM also requires GA mode. Therefore, we need to 1840 * check cmpxchg16b support before enabling them. 1841 */ 1842 if (!boot_cpu_has(X86_FEATURE_CX16) || 1843 ((h->efr_reg & (0x1 << IOMMU_EFR_GASUP_SHIFT)) == 0)) { 1844 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY; 1845 break; 1846 } 1847 1848 if (h->efr_reg & BIT(IOMMU_EFR_XTSUP_SHIFT)) 1849 amd_iommu_xt_mode = IRQ_REMAP_X2APIC_MODE; 1850 1851 early_iommu_features_init(iommu, h); 1852 1853 break; 1854 default: 1855 return -EINVAL; 1856 } 1857 1858 iommu->mmio_base = iommu_map_mmio_space(iommu->mmio_phys, 1859 iommu->mmio_phys_end); 1860 if (!iommu->mmio_base) 1861 return -ENOMEM; 1862 1863 return init_iommu_from_acpi(iommu, h); 1864 } 1865 1866 static int __init init_iommu_one_late(struct amd_iommu *iommu) 1867 { 1868 int ret; 1869 1870 if (alloc_cwwb_sem(iommu)) 1871 return -ENOMEM; 1872 1873 if (alloc_command_buffer(iommu)) 1874 return -ENOMEM; 1875 1876 if (alloc_event_buffer(iommu)) 1877 return -ENOMEM; 1878 1879 iommu->int_enabled = false; 1880 1881 init_translation_status(iommu); 1882 if (translation_pre_enabled(iommu) && !is_kdump_kernel()) { 1883 iommu_disable(iommu); 1884 clear_translation_pre_enabled(iommu); 1885 pr_warn("Translation was enabled for IOMMU:%d but we are not in kdump mode\n", 1886 iommu->index); 1887 } 1888 if (amd_iommu_pre_enabled) 1889 amd_iommu_pre_enabled = translation_pre_enabled(iommu); 1890 1891 if (amd_iommu_irq_remap) { 1892 ret = amd_iommu_create_irq_domain(iommu); 1893 if (ret) 1894 return ret; 1895 } 1896 1897 /* 1898 * Make sure IOMMU is not considered to translate itself. The IVRS 1899 * table tells us so, but this is a lie! 1900 */ 1901 iommu->pci_seg->rlookup_table[iommu->devid] = NULL; 1902 1903 return 0; 1904 } 1905 1906 /** 1907 * get_highest_supported_ivhd_type - Look up the appropriate IVHD type 1908 * @ivrs: Pointer to the IVRS header 1909 * 1910 * This function search through all IVDB of the maximum supported IVHD 1911 */ 1912 static u8 get_highest_supported_ivhd_type(struct acpi_table_header *ivrs) 1913 { 1914 u8 *base = (u8 *)ivrs; 1915 struct ivhd_header *ivhd = (struct ivhd_header *) 1916 (base + IVRS_HEADER_LENGTH); 1917 u8 last_type = ivhd->type; 1918 u16 devid = ivhd->devid; 1919 1920 while (((u8 *)ivhd - base < ivrs->length) && 1921 (ivhd->type <= ACPI_IVHD_TYPE_MAX_SUPPORTED)) { 1922 u8 *p = (u8 *) ivhd; 1923 1924 if (ivhd->devid == devid) 1925 last_type = ivhd->type; 1926 ivhd = (struct ivhd_header *)(p + ivhd->length); 1927 } 1928 1929 return last_type; 1930 } 1931 1932 /* 1933 * Iterates over all IOMMU entries in the ACPI table, allocates the 1934 * IOMMU structure and initializes it with init_iommu_one() 1935 */ 1936 static int __init init_iommu_all(struct acpi_table_header *table) 1937 { 1938 u8 *p = (u8 *)table, *end = (u8 *)table; 1939 struct ivhd_header *h; 1940 struct amd_iommu *iommu; 1941 int ret; 1942 1943 end += table->length; 1944 p += IVRS_HEADER_LENGTH; 1945 1946 /* Phase 1: Process all IVHD blocks */ 1947 while (p < end) { 1948 h = (struct ivhd_header *)p; 1949 if (*p == amd_iommu_target_ivhd_type) { 1950 1951 DUMP_printk("device: %04x:%02x:%02x.%01x cap: %04x " 1952 "flags: %01x info %04x\n", 1953 h->pci_seg, PCI_BUS_NUM(h->devid), 1954 PCI_SLOT(h->devid), PCI_FUNC(h->devid), 1955 h->cap_ptr, h->flags, h->info); 1956 DUMP_printk(" mmio-addr: %016llx\n", 1957 h->mmio_phys); 1958 1959 iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL); 1960 if (iommu == NULL) 1961 return -ENOMEM; 1962 1963 ret = init_iommu_one(iommu, h, table); 1964 if (ret) 1965 return ret; 1966 } 1967 p += h->length; 1968 1969 } 1970 WARN_ON(p != end); 1971 1972 /* Phase 2 : Early feature support check */ 1973 get_global_efr(); 1974 1975 /* Phase 3 : Enabling IOMMU features */ 1976 for_each_iommu(iommu) { 1977 ret = init_iommu_one_late(iommu); 1978 if (ret) 1979 return ret; 1980 } 1981 1982 return 0; 1983 } 1984 1985 static void init_iommu_perf_ctr(struct amd_iommu *iommu) 1986 { 1987 u64 val; 1988 struct pci_dev *pdev = iommu->dev; 1989 1990 if (!iommu_feature(iommu, FEATURE_PC)) 1991 return; 1992 1993 amd_iommu_pc_present = true; 1994 1995 pci_info(pdev, "IOMMU performance counters supported\n"); 1996 1997 val = readl(iommu->mmio_base + MMIO_CNTR_CONF_OFFSET); 1998 iommu->max_banks = (u8) ((val >> 12) & 0x3f); 1999 iommu->max_counters = (u8) ((val >> 7) & 0xf); 2000 2001 return; 2002 } 2003 2004 static ssize_t amd_iommu_show_cap(struct device *dev, 2005 struct device_attribute *attr, 2006 char *buf) 2007 { 2008 struct amd_iommu *iommu = dev_to_amd_iommu(dev); 2009 return sysfs_emit(buf, "%x\n", iommu->cap); 2010 } 2011 static DEVICE_ATTR(cap, S_IRUGO, amd_iommu_show_cap, NULL); 2012 2013 static ssize_t amd_iommu_show_features(struct device *dev, 2014 struct device_attribute *attr, 2015 char *buf) 2016 { 2017 struct amd_iommu *iommu = dev_to_amd_iommu(dev); 2018 return sysfs_emit(buf, "%llx:%llx\n", iommu->features2, iommu->features); 2019 } 2020 static DEVICE_ATTR(features, S_IRUGO, amd_iommu_show_features, NULL); 2021 2022 static struct attribute *amd_iommu_attrs[] = { 2023 &dev_attr_cap.attr, 2024 &dev_attr_features.attr, 2025 NULL, 2026 }; 2027 2028 static struct attribute_group amd_iommu_group = { 2029 .name = "amd-iommu", 2030 .attrs = amd_iommu_attrs, 2031 }; 2032 2033 static const struct attribute_group *amd_iommu_groups[] = { 2034 &amd_iommu_group, 2035 NULL, 2036 }; 2037 2038 /* 2039 * Note: IVHD 0x11 and 0x40 also contains exact copy 2040 * of the IOMMU Extended Feature Register [MMIO Offset 0030h]. 2041 * Default to EFR in IVHD since it is available sooner (i.e. before PCI init). 2042 */ 2043 static void __init late_iommu_features_init(struct amd_iommu *iommu) 2044 { 2045 u64 features, features2; 2046 2047 if (!(iommu->cap & (1 << IOMMU_CAP_EFR))) 2048 return; 2049 2050 /* read extended feature bits */ 2051 features = readq(iommu->mmio_base + MMIO_EXT_FEATURES); 2052 features2 = readq(iommu->mmio_base + MMIO_EXT_FEATURES2); 2053 2054 if (!iommu->features) { 2055 iommu->features = features; 2056 iommu->features2 = features2; 2057 return; 2058 } 2059 2060 /* 2061 * Sanity check and warn if EFR values from 2062 * IVHD and MMIO conflict. 2063 */ 2064 if (features != iommu->features || 2065 features2 != iommu->features2) { 2066 pr_warn(FW_WARN 2067 "EFR mismatch. Use IVHD EFR (%#llx : %#llx), EFR2 (%#llx : %#llx).\n", 2068 features, iommu->features, 2069 features2, iommu->features2); 2070 } 2071 } 2072 2073 static int __init iommu_init_pci(struct amd_iommu *iommu) 2074 { 2075 int cap_ptr = iommu->cap_ptr; 2076 int ret; 2077 2078 iommu->dev = pci_get_domain_bus_and_slot(iommu->pci_seg->id, 2079 PCI_BUS_NUM(iommu->devid), 2080 iommu->devid & 0xff); 2081 if (!iommu->dev) 2082 return -ENODEV; 2083 2084 /* Prevent binding other PCI device drivers to IOMMU devices */ 2085 iommu->dev->match_driver = false; 2086 2087 /* ACPI _PRT won't have an IRQ for IOMMU */ 2088 iommu->dev->irq_managed = 1; 2089 2090 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET, 2091 &iommu->cap); 2092 2093 if (!(iommu->cap & (1 << IOMMU_CAP_IOTLB))) 2094 amd_iommu_iotlb_sup = false; 2095 2096 late_iommu_features_init(iommu); 2097 2098 if (iommu_feature(iommu, FEATURE_GT)) { 2099 int glxval; 2100 u32 max_pasid; 2101 u64 pasmax; 2102 2103 pasmax = iommu->features & FEATURE_PASID_MASK; 2104 pasmax >>= FEATURE_PASID_SHIFT; 2105 max_pasid = (1 << (pasmax + 1)) - 1; 2106 2107 amd_iommu_max_pasid = min(amd_iommu_max_pasid, max_pasid); 2108 2109 BUG_ON(amd_iommu_max_pasid & ~PASID_MASK); 2110 2111 glxval = iommu->features & FEATURE_GLXVAL_MASK; 2112 glxval >>= FEATURE_GLXVAL_SHIFT; 2113 2114 if (amd_iommu_max_glx_val == -1) 2115 amd_iommu_max_glx_val = glxval; 2116 else 2117 amd_iommu_max_glx_val = min(amd_iommu_max_glx_val, glxval); 2118 } 2119 2120 if (iommu_feature(iommu, FEATURE_GT) && 2121 iommu_feature(iommu, FEATURE_PPR)) { 2122 iommu->is_iommu_v2 = true; 2123 amd_iommu_v2_present = true; 2124 } 2125 2126 if (iommu_feature(iommu, FEATURE_PPR) && alloc_ppr_log(iommu)) 2127 return -ENOMEM; 2128 2129 if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE)) { 2130 pr_info("Using strict mode due to virtualization\n"); 2131 iommu_set_dma_strict(); 2132 amd_iommu_np_cache = true; 2133 } 2134 2135 init_iommu_perf_ctr(iommu); 2136 2137 if (amd_iommu_pgtable == AMD_IOMMU_V2) { 2138 if (!iommu_feature(iommu, FEATURE_GIOSUP) || 2139 !iommu_feature(iommu, FEATURE_GT)) { 2140 pr_warn("Cannot enable v2 page table for DMA-API. Fallback to v1.\n"); 2141 amd_iommu_pgtable = AMD_IOMMU_V1; 2142 } else if (iommu_default_passthrough()) { 2143 pr_warn("V2 page table doesn't support passthrough mode. Fallback to v1.\n"); 2144 amd_iommu_pgtable = AMD_IOMMU_V1; 2145 } 2146 } 2147 2148 if (is_rd890_iommu(iommu->dev)) { 2149 int i, j; 2150 2151 iommu->root_pdev = 2152 pci_get_domain_bus_and_slot(iommu->pci_seg->id, 2153 iommu->dev->bus->number, 2154 PCI_DEVFN(0, 0)); 2155 2156 /* 2157 * Some rd890 systems may not be fully reconfigured by the 2158 * BIOS, so it's necessary for us to store this information so 2159 * it can be reprogrammed on resume 2160 */ 2161 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 4, 2162 &iommu->stored_addr_lo); 2163 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 8, 2164 &iommu->stored_addr_hi); 2165 2166 /* Low bit locks writes to configuration space */ 2167 iommu->stored_addr_lo &= ~1; 2168 2169 for (i = 0; i < 6; i++) 2170 for (j = 0; j < 0x12; j++) 2171 iommu->stored_l1[i][j] = iommu_read_l1(iommu, i, j); 2172 2173 for (i = 0; i < 0x83; i++) 2174 iommu->stored_l2[i] = iommu_read_l2(iommu, i); 2175 } 2176 2177 amd_iommu_erratum_746_workaround(iommu); 2178 amd_iommu_ats_write_check_workaround(iommu); 2179 2180 ret = iommu_device_sysfs_add(&iommu->iommu, &iommu->dev->dev, 2181 amd_iommu_groups, "ivhd%d", iommu->index); 2182 if (ret) 2183 return ret; 2184 2185 iommu_device_register(&iommu->iommu, &amd_iommu_ops, NULL); 2186 2187 return pci_enable_device(iommu->dev); 2188 } 2189 2190 static void print_iommu_info(void) 2191 { 2192 static const char * const feat_str[] = { 2193 "PreF", "PPR", "X2APIC", "NX", "GT", "[5]", 2194 "IA", "GA", "HE", "PC" 2195 }; 2196 struct amd_iommu *iommu; 2197 2198 for_each_iommu(iommu) { 2199 struct pci_dev *pdev = iommu->dev; 2200 int i; 2201 2202 pci_info(pdev, "Found IOMMU cap 0x%x\n", iommu->cap_ptr); 2203 2204 if (iommu->cap & (1 << IOMMU_CAP_EFR)) { 2205 pr_info("Extended features (%#llx, %#llx):", iommu->features, iommu->features2); 2206 2207 for (i = 0; i < ARRAY_SIZE(feat_str); ++i) { 2208 if (iommu_feature(iommu, (1ULL << i))) 2209 pr_cont(" %s", feat_str[i]); 2210 } 2211 2212 if (iommu->features & FEATURE_GAM_VAPIC) 2213 pr_cont(" GA_vAPIC"); 2214 2215 if (iommu->features & FEATURE_SNP) 2216 pr_cont(" SNP"); 2217 2218 pr_cont("\n"); 2219 } 2220 } 2221 if (irq_remapping_enabled) { 2222 pr_info("Interrupt remapping enabled\n"); 2223 if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE) 2224 pr_info("X2APIC enabled\n"); 2225 } 2226 if (amd_iommu_pgtable == AMD_IOMMU_V2) { 2227 pr_info("V2 page table enabled (Paging mode : %d level)\n", 2228 amd_iommu_gpt_level); 2229 } 2230 } 2231 2232 static int __init amd_iommu_init_pci(void) 2233 { 2234 struct amd_iommu *iommu; 2235 struct amd_iommu_pci_seg *pci_seg; 2236 int ret; 2237 2238 for_each_iommu(iommu) { 2239 ret = iommu_init_pci(iommu); 2240 if (ret) { 2241 pr_err("IOMMU%d: Failed to initialize IOMMU Hardware (error=%d)!\n", 2242 iommu->index, ret); 2243 goto out; 2244 } 2245 /* Need to setup range after PCI init */ 2246 iommu_set_cwwb_range(iommu); 2247 } 2248 2249 /* 2250 * Order is important here to make sure any unity map requirements are 2251 * fulfilled. The unity mappings are created and written to the device 2252 * table during the iommu_init_pci() call. 2253 * 2254 * After that we call init_device_table_dma() to make sure any 2255 * uninitialized DTE will block DMA, and in the end we flush the caches 2256 * of all IOMMUs to make sure the changes to the device table are 2257 * active. 2258 */ 2259 for_each_pci_segment(pci_seg) 2260 init_device_table_dma(pci_seg); 2261 2262 for_each_iommu(iommu) 2263 iommu_flush_all_caches(iommu); 2264 2265 print_iommu_info(); 2266 2267 out: 2268 return ret; 2269 } 2270 2271 /**************************************************************************** 2272 * 2273 * The following functions initialize the MSI interrupts for all IOMMUs 2274 * in the system. It's a bit challenging because there could be multiple 2275 * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per 2276 * pci_dev. 2277 * 2278 ****************************************************************************/ 2279 2280 static int iommu_setup_msi(struct amd_iommu *iommu) 2281 { 2282 int r; 2283 2284 r = pci_enable_msi(iommu->dev); 2285 if (r) 2286 return r; 2287 2288 r = request_threaded_irq(iommu->dev->irq, 2289 amd_iommu_int_handler, 2290 amd_iommu_int_thread, 2291 0, "AMD-Vi", 2292 iommu); 2293 2294 if (r) { 2295 pci_disable_msi(iommu->dev); 2296 return r; 2297 } 2298 2299 return 0; 2300 } 2301 2302 union intcapxt { 2303 u64 capxt; 2304 struct { 2305 u64 reserved_0 : 2, 2306 dest_mode_logical : 1, 2307 reserved_1 : 5, 2308 destid_0_23 : 24, 2309 vector : 8, 2310 reserved_2 : 16, 2311 destid_24_31 : 8; 2312 }; 2313 } __attribute__ ((packed)); 2314 2315 2316 static struct irq_chip intcapxt_controller; 2317 2318 static int intcapxt_irqdomain_activate(struct irq_domain *domain, 2319 struct irq_data *irqd, bool reserve) 2320 { 2321 return 0; 2322 } 2323 2324 static void intcapxt_irqdomain_deactivate(struct irq_domain *domain, 2325 struct irq_data *irqd) 2326 { 2327 } 2328 2329 2330 static int intcapxt_irqdomain_alloc(struct irq_domain *domain, unsigned int virq, 2331 unsigned int nr_irqs, void *arg) 2332 { 2333 struct irq_alloc_info *info = arg; 2334 int i, ret; 2335 2336 if (!info || info->type != X86_IRQ_ALLOC_TYPE_AMDVI) 2337 return -EINVAL; 2338 2339 ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, arg); 2340 if (ret < 0) 2341 return ret; 2342 2343 for (i = virq; i < virq + nr_irqs; i++) { 2344 struct irq_data *irqd = irq_domain_get_irq_data(domain, i); 2345 2346 irqd->chip = &intcapxt_controller; 2347 irqd->hwirq = info->hwirq; 2348 irqd->chip_data = info->data; 2349 __irq_set_handler(i, handle_edge_irq, 0, "edge"); 2350 } 2351 2352 return ret; 2353 } 2354 2355 static void intcapxt_irqdomain_free(struct irq_domain *domain, unsigned int virq, 2356 unsigned int nr_irqs) 2357 { 2358 irq_domain_free_irqs_top(domain, virq, nr_irqs); 2359 } 2360 2361 2362 static void intcapxt_unmask_irq(struct irq_data *irqd) 2363 { 2364 struct amd_iommu *iommu = irqd->chip_data; 2365 struct irq_cfg *cfg = irqd_cfg(irqd); 2366 union intcapxt xt; 2367 2368 xt.capxt = 0ULL; 2369 xt.dest_mode_logical = apic->dest_mode_logical; 2370 xt.vector = cfg->vector; 2371 xt.destid_0_23 = cfg->dest_apicid & GENMASK(23, 0); 2372 xt.destid_24_31 = cfg->dest_apicid >> 24; 2373 2374 writeq(xt.capxt, iommu->mmio_base + irqd->hwirq); 2375 } 2376 2377 static void intcapxt_mask_irq(struct irq_data *irqd) 2378 { 2379 struct amd_iommu *iommu = irqd->chip_data; 2380 2381 writeq(0, iommu->mmio_base + irqd->hwirq); 2382 } 2383 2384 2385 static int intcapxt_set_affinity(struct irq_data *irqd, 2386 const struct cpumask *mask, bool force) 2387 { 2388 struct irq_data *parent = irqd->parent_data; 2389 int ret; 2390 2391 ret = parent->chip->irq_set_affinity(parent, mask, force); 2392 if (ret < 0 || ret == IRQ_SET_MASK_OK_DONE) 2393 return ret; 2394 return 0; 2395 } 2396 2397 static int intcapxt_set_wake(struct irq_data *irqd, unsigned int on) 2398 { 2399 return on ? -EOPNOTSUPP : 0; 2400 } 2401 2402 static struct irq_chip intcapxt_controller = { 2403 .name = "IOMMU-MSI", 2404 .irq_unmask = intcapxt_unmask_irq, 2405 .irq_mask = intcapxt_mask_irq, 2406 .irq_ack = irq_chip_ack_parent, 2407 .irq_retrigger = irq_chip_retrigger_hierarchy, 2408 .irq_set_affinity = intcapxt_set_affinity, 2409 .irq_set_wake = intcapxt_set_wake, 2410 .flags = IRQCHIP_MASK_ON_SUSPEND, 2411 }; 2412 2413 static const struct irq_domain_ops intcapxt_domain_ops = { 2414 .alloc = intcapxt_irqdomain_alloc, 2415 .free = intcapxt_irqdomain_free, 2416 .activate = intcapxt_irqdomain_activate, 2417 .deactivate = intcapxt_irqdomain_deactivate, 2418 }; 2419 2420 2421 static struct irq_domain *iommu_irqdomain; 2422 2423 static struct irq_domain *iommu_get_irqdomain(void) 2424 { 2425 struct fwnode_handle *fn; 2426 2427 /* No need for locking here (yet) as the init is single-threaded */ 2428 if (iommu_irqdomain) 2429 return iommu_irqdomain; 2430 2431 fn = irq_domain_alloc_named_fwnode("AMD-Vi-MSI"); 2432 if (!fn) 2433 return NULL; 2434 2435 iommu_irqdomain = irq_domain_create_hierarchy(x86_vector_domain, 0, 0, 2436 fn, &intcapxt_domain_ops, 2437 NULL); 2438 if (!iommu_irqdomain) 2439 irq_domain_free_fwnode(fn); 2440 2441 return iommu_irqdomain; 2442 } 2443 2444 static int __iommu_setup_intcapxt(struct amd_iommu *iommu, const char *devname, 2445 int hwirq, irq_handler_t thread_fn) 2446 { 2447 struct irq_domain *domain; 2448 struct irq_alloc_info info; 2449 int irq, ret; 2450 int node = dev_to_node(&iommu->dev->dev); 2451 2452 domain = iommu_get_irqdomain(); 2453 if (!domain) 2454 return -ENXIO; 2455 2456 init_irq_alloc_info(&info, NULL); 2457 info.type = X86_IRQ_ALLOC_TYPE_AMDVI; 2458 info.data = iommu; 2459 info.hwirq = hwirq; 2460 2461 irq = irq_domain_alloc_irqs(domain, 1, node, &info); 2462 if (irq < 0) { 2463 irq_domain_remove(domain); 2464 return irq; 2465 } 2466 2467 ret = request_threaded_irq(irq, amd_iommu_int_handler, 2468 thread_fn, 0, devname, iommu); 2469 if (ret) { 2470 irq_domain_free_irqs(irq, 1); 2471 irq_domain_remove(domain); 2472 return ret; 2473 } 2474 2475 return 0; 2476 } 2477 2478 static int iommu_setup_intcapxt(struct amd_iommu *iommu) 2479 { 2480 int ret; 2481 2482 snprintf(iommu->evt_irq_name, sizeof(iommu->evt_irq_name), 2483 "AMD-Vi%d-Evt", iommu->index); 2484 ret = __iommu_setup_intcapxt(iommu, iommu->evt_irq_name, 2485 MMIO_INTCAPXT_EVT_OFFSET, 2486 amd_iommu_int_thread_evtlog); 2487 if (ret) 2488 return ret; 2489 2490 snprintf(iommu->ppr_irq_name, sizeof(iommu->ppr_irq_name), 2491 "AMD-Vi%d-PPR", iommu->index); 2492 ret = __iommu_setup_intcapxt(iommu, iommu->ppr_irq_name, 2493 MMIO_INTCAPXT_PPR_OFFSET, 2494 amd_iommu_int_thread_pprlog); 2495 if (ret) 2496 return ret; 2497 2498 #ifdef CONFIG_IRQ_REMAP 2499 snprintf(iommu->ga_irq_name, sizeof(iommu->ga_irq_name), 2500 "AMD-Vi%d-GA", iommu->index); 2501 ret = __iommu_setup_intcapxt(iommu, iommu->ga_irq_name, 2502 MMIO_INTCAPXT_GALOG_OFFSET, 2503 amd_iommu_int_thread_galog); 2504 #endif 2505 2506 return ret; 2507 } 2508 2509 static int iommu_init_irq(struct amd_iommu *iommu) 2510 { 2511 int ret; 2512 2513 if (iommu->int_enabled) 2514 goto enable_faults; 2515 2516 if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE) 2517 ret = iommu_setup_intcapxt(iommu); 2518 else if (iommu->dev->msi_cap) 2519 ret = iommu_setup_msi(iommu); 2520 else 2521 ret = -ENODEV; 2522 2523 if (ret) 2524 return ret; 2525 2526 iommu->int_enabled = true; 2527 enable_faults: 2528 2529 if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE) 2530 iommu_feature_enable(iommu, CONTROL_INTCAPXT_EN); 2531 2532 iommu_feature_enable(iommu, CONTROL_EVT_INT_EN); 2533 2534 return 0; 2535 } 2536 2537 /**************************************************************************** 2538 * 2539 * The next functions belong to the third pass of parsing the ACPI 2540 * table. In this last pass the memory mapping requirements are 2541 * gathered (like exclusion and unity mapping ranges). 2542 * 2543 ****************************************************************************/ 2544 2545 static void __init free_unity_maps(void) 2546 { 2547 struct unity_map_entry *entry, *next; 2548 struct amd_iommu_pci_seg *p, *pci_seg; 2549 2550 for_each_pci_segment_safe(pci_seg, p) { 2551 list_for_each_entry_safe(entry, next, &pci_seg->unity_map, list) { 2552 list_del(&entry->list); 2553 kfree(entry); 2554 } 2555 } 2556 } 2557 2558 /* called for unity map ACPI definition */ 2559 static int __init init_unity_map_range(struct ivmd_header *m, 2560 struct acpi_table_header *ivrs_base) 2561 { 2562 struct unity_map_entry *e = NULL; 2563 struct amd_iommu_pci_seg *pci_seg; 2564 char *s; 2565 2566 pci_seg = get_pci_segment(m->pci_seg, ivrs_base); 2567 if (pci_seg == NULL) 2568 return -ENOMEM; 2569 2570 e = kzalloc(sizeof(*e), GFP_KERNEL); 2571 if (e == NULL) 2572 return -ENOMEM; 2573 2574 switch (m->type) { 2575 default: 2576 kfree(e); 2577 return 0; 2578 case ACPI_IVMD_TYPE: 2579 s = "IVMD_TYPEi\t\t\t"; 2580 e->devid_start = e->devid_end = m->devid; 2581 break; 2582 case ACPI_IVMD_TYPE_ALL: 2583 s = "IVMD_TYPE_ALL\t\t"; 2584 e->devid_start = 0; 2585 e->devid_end = pci_seg->last_bdf; 2586 break; 2587 case ACPI_IVMD_TYPE_RANGE: 2588 s = "IVMD_TYPE_RANGE\t\t"; 2589 e->devid_start = m->devid; 2590 e->devid_end = m->aux; 2591 break; 2592 } 2593 e->address_start = PAGE_ALIGN(m->range_start); 2594 e->address_end = e->address_start + PAGE_ALIGN(m->range_length); 2595 e->prot = m->flags >> 1; 2596 2597 /* 2598 * Treat per-device exclusion ranges as r/w unity-mapped regions 2599 * since some buggy BIOSes might lead to the overwritten exclusion 2600 * range (exclusion_start and exclusion_length members). This 2601 * happens when there are multiple exclusion ranges (IVMD entries) 2602 * defined in ACPI table. 2603 */ 2604 if (m->flags & IVMD_FLAG_EXCL_RANGE) 2605 e->prot = (IVMD_FLAG_IW | IVMD_FLAG_IR) >> 1; 2606 2607 DUMP_printk("%s devid_start: %04x:%02x:%02x.%x devid_end: " 2608 "%04x:%02x:%02x.%x range_start: %016llx range_end: %016llx" 2609 " flags: %x\n", s, m->pci_seg, 2610 PCI_BUS_NUM(e->devid_start), PCI_SLOT(e->devid_start), 2611 PCI_FUNC(e->devid_start), m->pci_seg, 2612 PCI_BUS_NUM(e->devid_end), 2613 PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end), 2614 e->address_start, e->address_end, m->flags); 2615 2616 list_add_tail(&e->list, &pci_seg->unity_map); 2617 2618 return 0; 2619 } 2620 2621 /* iterates over all memory definitions we find in the ACPI table */ 2622 static int __init init_memory_definitions(struct acpi_table_header *table) 2623 { 2624 u8 *p = (u8 *)table, *end = (u8 *)table; 2625 struct ivmd_header *m; 2626 2627 end += table->length; 2628 p += IVRS_HEADER_LENGTH; 2629 2630 while (p < end) { 2631 m = (struct ivmd_header *)p; 2632 if (m->flags & (IVMD_FLAG_UNITY_MAP | IVMD_FLAG_EXCL_RANGE)) 2633 init_unity_map_range(m, table); 2634 2635 p += m->length; 2636 } 2637 2638 return 0; 2639 } 2640 2641 /* 2642 * Init the device table to not allow DMA access for devices 2643 */ 2644 static void init_device_table_dma(struct amd_iommu_pci_seg *pci_seg) 2645 { 2646 u32 devid; 2647 struct dev_table_entry *dev_table = pci_seg->dev_table; 2648 2649 if (dev_table == NULL) 2650 return; 2651 2652 for (devid = 0; devid <= pci_seg->last_bdf; ++devid) { 2653 __set_dev_entry_bit(dev_table, devid, DEV_ENTRY_VALID); 2654 if (!amd_iommu_snp_en) 2655 __set_dev_entry_bit(dev_table, devid, DEV_ENTRY_TRANSLATION); 2656 } 2657 } 2658 2659 static void __init uninit_device_table_dma(struct amd_iommu_pci_seg *pci_seg) 2660 { 2661 u32 devid; 2662 struct dev_table_entry *dev_table = pci_seg->dev_table; 2663 2664 if (dev_table == NULL) 2665 return; 2666 2667 for (devid = 0; devid <= pci_seg->last_bdf; ++devid) { 2668 dev_table[devid].data[0] = 0ULL; 2669 dev_table[devid].data[1] = 0ULL; 2670 } 2671 } 2672 2673 static void init_device_table(void) 2674 { 2675 struct amd_iommu_pci_seg *pci_seg; 2676 u32 devid; 2677 2678 if (!amd_iommu_irq_remap) 2679 return; 2680 2681 for_each_pci_segment(pci_seg) { 2682 for (devid = 0; devid <= pci_seg->last_bdf; ++devid) 2683 __set_dev_entry_bit(pci_seg->dev_table, 2684 devid, DEV_ENTRY_IRQ_TBL_EN); 2685 } 2686 } 2687 2688 static void iommu_init_flags(struct amd_iommu *iommu) 2689 { 2690 iommu->acpi_flags & IVHD_FLAG_HT_TUN_EN_MASK ? 2691 iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) : 2692 iommu_feature_disable(iommu, CONTROL_HT_TUN_EN); 2693 2694 iommu->acpi_flags & IVHD_FLAG_PASSPW_EN_MASK ? 2695 iommu_feature_enable(iommu, CONTROL_PASSPW_EN) : 2696 iommu_feature_disable(iommu, CONTROL_PASSPW_EN); 2697 2698 iommu->acpi_flags & IVHD_FLAG_RESPASSPW_EN_MASK ? 2699 iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) : 2700 iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN); 2701 2702 iommu->acpi_flags & IVHD_FLAG_ISOC_EN_MASK ? 2703 iommu_feature_enable(iommu, CONTROL_ISOC_EN) : 2704 iommu_feature_disable(iommu, CONTROL_ISOC_EN); 2705 2706 /* 2707 * make IOMMU memory accesses cache coherent 2708 */ 2709 iommu_feature_enable(iommu, CONTROL_COHERENT_EN); 2710 2711 /* Set IOTLB invalidation timeout to 1s */ 2712 iommu_set_inv_tlb_timeout(iommu, CTRL_INV_TO_1S); 2713 } 2714 2715 static void iommu_apply_resume_quirks(struct amd_iommu *iommu) 2716 { 2717 int i, j; 2718 u32 ioc_feature_control; 2719 struct pci_dev *pdev = iommu->root_pdev; 2720 2721 /* RD890 BIOSes may not have completely reconfigured the iommu */ 2722 if (!is_rd890_iommu(iommu->dev) || !pdev) 2723 return; 2724 2725 /* 2726 * First, we need to ensure that the iommu is enabled. This is 2727 * controlled by a register in the northbridge 2728 */ 2729 2730 /* Select Northbridge indirect register 0x75 and enable writing */ 2731 pci_write_config_dword(pdev, 0x60, 0x75 | (1 << 7)); 2732 pci_read_config_dword(pdev, 0x64, &ioc_feature_control); 2733 2734 /* Enable the iommu */ 2735 if (!(ioc_feature_control & 0x1)) 2736 pci_write_config_dword(pdev, 0x64, ioc_feature_control | 1); 2737 2738 /* Restore the iommu BAR */ 2739 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4, 2740 iommu->stored_addr_lo); 2741 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 8, 2742 iommu->stored_addr_hi); 2743 2744 /* Restore the l1 indirect regs for each of the 6 l1s */ 2745 for (i = 0; i < 6; i++) 2746 for (j = 0; j < 0x12; j++) 2747 iommu_write_l1(iommu, i, j, iommu->stored_l1[i][j]); 2748 2749 /* Restore the l2 indirect regs */ 2750 for (i = 0; i < 0x83; i++) 2751 iommu_write_l2(iommu, i, iommu->stored_l2[i]); 2752 2753 /* Lock PCI setup registers */ 2754 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4, 2755 iommu->stored_addr_lo | 1); 2756 } 2757 2758 static void iommu_enable_ga(struct amd_iommu *iommu) 2759 { 2760 #ifdef CONFIG_IRQ_REMAP 2761 switch (amd_iommu_guest_ir) { 2762 case AMD_IOMMU_GUEST_IR_VAPIC: 2763 case AMD_IOMMU_GUEST_IR_LEGACY_GA: 2764 iommu_feature_enable(iommu, CONTROL_GA_EN); 2765 iommu->irte_ops = &irte_128_ops; 2766 break; 2767 default: 2768 iommu->irte_ops = &irte_32_ops; 2769 break; 2770 } 2771 #endif 2772 } 2773 2774 static void iommu_disable_irtcachedis(struct amd_iommu *iommu) 2775 { 2776 iommu_feature_disable(iommu, CONTROL_IRTCACHEDIS); 2777 } 2778 2779 static void iommu_enable_irtcachedis(struct amd_iommu *iommu) 2780 { 2781 u64 ctrl; 2782 2783 if (!amd_iommu_irtcachedis) 2784 return; 2785 2786 /* 2787 * Note: 2788 * The support for IRTCacheDis feature is dertermined by 2789 * checking if the bit is writable. 2790 */ 2791 iommu_feature_enable(iommu, CONTROL_IRTCACHEDIS); 2792 ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET); 2793 ctrl &= (1ULL << CONTROL_IRTCACHEDIS); 2794 if (ctrl) 2795 iommu->irtcachedis_enabled = true; 2796 pr_info("iommu%d (%#06x) : IRT cache is %s\n", 2797 iommu->index, iommu->devid, 2798 iommu->irtcachedis_enabled ? "disabled" : "enabled"); 2799 } 2800 2801 static void early_enable_iommu(struct amd_iommu *iommu) 2802 { 2803 iommu_disable(iommu); 2804 iommu_init_flags(iommu); 2805 iommu_set_device_table(iommu); 2806 iommu_enable_command_buffer(iommu); 2807 iommu_enable_event_buffer(iommu); 2808 iommu_set_exclusion_range(iommu); 2809 iommu_enable_ga(iommu); 2810 iommu_enable_xt(iommu); 2811 iommu_enable_irtcachedis(iommu); 2812 iommu_enable(iommu); 2813 iommu_flush_all_caches(iommu); 2814 } 2815 2816 /* 2817 * This function finally enables all IOMMUs found in the system after 2818 * they have been initialized. 2819 * 2820 * Or if in kdump kernel and IOMMUs are all pre-enabled, try to copy 2821 * the old content of device table entries. Not this case or copy failed, 2822 * just continue as normal kernel does. 2823 */ 2824 static void early_enable_iommus(void) 2825 { 2826 struct amd_iommu *iommu; 2827 struct amd_iommu_pci_seg *pci_seg; 2828 2829 if (!copy_device_table()) { 2830 /* 2831 * If come here because of failure in copying device table from old 2832 * kernel with all IOMMUs enabled, print error message and try to 2833 * free allocated old_dev_tbl_cpy. 2834 */ 2835 if (amd_iommu_pre_enabled) 2836 pr_err("Failed to copy DEV table from previous kernel.\n"); 2837 2838 for_each_pci_segment(pci_seg) { 2839 if (pci_seg->old_dev_tbl_cpy != NULL) { 2840 free_pages((unsigned long)pci_seg->old_dev_tbl_cpy, 2841 get_order(pci_seg->dev_table_size)); 2842 pci_seg->old_dev_tbl_cpy = NULL; 2843 } 2844 } 2845 2846 for_each_iommu(iommu) { 2847 clear_translation_pre_enabled(iommu); 2848 early_enable_iommu(iommu); 2849 } 2850 } else { 2851 pr_info("Copied DEV table from previous kernel.\n"); 2852 2853 for_each_pci_segment(pci_seg) { 2854 free_pages((unsigned long)pci_seg->dev_table, 2855 get_order(pci_seg->dev_table_size)); 2856 pci_seg->dev_table = pci_seg->old_dev_tbl_cpy; 2857 } 2858 2859 for_each_iommu(iommu) { 2860 iommu_disable_command_buffer(iommu); 2861 iommu_disable_event_buffer(iommu); 2862 iommu_disable_irtcachedis(iommu); 2863 iommu_enable_command_buffer(iommu); 2864 iommu_enable_event_buffer(iommu); 2865 iommu_enable_ga(iommu); 2866 iommu_enable_xt(iommu); 2867 iommu_enable_irtcachedis(iommu); 2868 iommu_set_device_table(iommu); 2869 iommu_flush_all_caches(iommu); 2870 } 2871 } 2872 } 2873 2874 static void enable_iommus_v2(void) 2875 { 2876 struct amd_iommu *iommu; 2877 2878 for_each_iommu(iommu) { 2879 iommu_enable_ppr_log(iommu); 2880 iommu_enable_gt(iommu); 2881 } 2882 } 2883 2884 static void enable_iommus_vapic(void) 2885 { 2886 #ifdef CONFIG_IRQ_REMAP 2887 u32 status, i; 2888 struct amd_iommu *iommu; 2889 2890 for_each_iommu(iommu) { 2891 /* 2892 * Disable GALog if already running. It could have been enabled 2893 * in the previous boot before kdump. 2894 */ 2895 status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET); 2896 if (!(status & MMIO_STATUS_GALOG_RUN_MASK)) 2897 continue; 2898 2899 iommu_feature_disable(iommu, CONTROL_GALOG_EN); 2900 iommu_feature_disable(iommu, CONTROL_GAINT_EN); 2901 2902 /* 2903 * Need to set and poll check the GALOGRun bit to zero before 2904 * we can set/ modify GA Log registers safely. 2905 */ 2906 for (i = 0; i < LOOP_TIMEOUT; ++i) { 2907 status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET); 2908 if (!(status & MMIO_STATUS_GALOG_RUN_MASK)) 2909 break; 2910 udelay(10); 2911 } 2912 2913 if (WARN_ON(i >= LOOP_TIMEOUT)) 2914 return; 2915 } 2916 2917 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir) && 2918 !check_feature_on_all_iommus(FEATURE_GAM_VAPIC)) { 2919 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA; 2920 return; 2921 } 2922 2923 if (amd_iommu_snp_en && 2924 !FEATURE_SNPAVICSUP_GAM(amd_iommu_efr2)) { 2925 pr_warn("Force to disable Virtual APIC due to SNP\n"); 2926 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA; 2927 return; 2928 } 2929 2930 /* Enabling GAM and SNPAVIC support */ 2931 for_each_iommu(iommu) { 2932 if (iommu_init_ga_log(iommu) || 2933 iommu_ga_log_enable(iommu)) 2934 return; 2935 2936 iommu_feature_enable(iommu, CONTROL_GAM_EN); 2937 if (amd_iommu_snp_en) 2938 iommu_feature_enable(iommu, CONTROL_SNPAVIC_EN); 2939 } 2940 2941 amd_iommu_irq_ops.capability |= (1 << IRQ_POSTING_CAP); 2942 pr_info("Virtual APIC enabled\n"); 2943 #endif 2944 } 2945 2946 static void enable_iommus(void) 2947 { 2948 early_enable_iommus(); 2949 } 2950 2951 static void disable_iommus(void) 2952 { 2953 struct amd_iommu *iommu; 2954 2955 for_each_iommu(iommu) 2956 iommu_disable(iommu); 2957 2958 #ifdef CONFIG_IRQ_REMAP 2959 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir)) 2960 amd_iommu_irq_ops.capability &= ~(1 << IRQ_POSTING_CAP); 2961 #endif 2962 } 2963 2964 /* 2965 * Suspend/Resume support 2966 * disable suspend until real resume implemented 2967 */ 2968 2969 static void amd_iommu_resume(void) 2970 { 2971 struct amd_iommu *iommu; 2972 2973 for_each_iommu(iommu) 2974 iommu_apply_resume_quirks(iommu); 2975 2976 /* re-load the hardware */ 2977 enable_iommus(); 2978 2979 amd_iommu_enable_interrupts(); 2980 } 2981 2982 static int amd_iommu_suspend(void) 2983 { 2984 /* disable IOMMUs to go out of the way for BIOS */ 2985 disable_iommus(); 2986 2987 return 0; 2988 } 2989 2990 static struct syscore_ops amd_iommu_syscore_ops = { 2991 .suspend = amd_iommu_suspend, 2992 .resume = amd_iommu_resume, 2993 }; 2994 2995 static void __init free_iommu_resources(void) 2996 { 2997 kmem_cache_destroy(amd_iommu_irq_cache); 2998 amd_iommu_irq_cache = NULL; 2999 3000 free_iommu_all(); 3001 free_pci_segments(); 3002 } 3003 3004 /* SB IOAPIC is always on this device in AMD systems */ 3005 #define IOAPIC_SB_DEVID ((0x00 << 8) | PCI_DEVFN(0x14, 0)) 3006 3007 static bool __init check_ioapic_information(void) 3008 { 3009 const char *fw_bug = FW_BUG; 3010 bool ret, has_sb_ioapic; 3011 int idx; 3012 3013 has_sb_ioapic = false; 3014 ret = false; 3015 3016 /* 3017 * If we have map overrides on the kernel command line the 3018 * messages in this function might not describe firmware bugs 3019 * anymore - so be careful 3020 */ 3021 if (cmdline_maps) 3022 fw_bug = ""; 3023 3024 for (idx = 0; idx < nr_ioapics; idx++) { 3025 int devid, id = mpc_ioapic_id(idx); 3026 3027 devid = get_ioapic_devid(id); 3028 if (devid < 0) { 3029 pr_err("%s: IOAPIC[%d] not in IVRS table\n", 3030 fw_bug, id); 3031 ret = false; 3032 } else if (devid == IOAPIC_SB_DEVID) { 3033 has_sb_ioapic = true; 3034 ret = true; 3035 } 3036 } 3037 3038 if (!has_sb_ioapic) { 3039 /* 3040 * We expect the SB IOAPIC to be listed in the IVRS 3041 * table. The system timer is connected to the SB IOAPIC 3042 * and if we don't have it in the list the system will 3043 * panic at boot time. This situation usually happens 3044 * when the BIOS is buggy and provides us the wrong 3045 * device id for the IOAPIC in the system. 3046 */ 3047 pr_err("%s: No southbridge IOAPIC found\n", fw_bug); 3048 } 3049 3050 if (!ret) 3051 pr_err("Disabling interrupt remapping\n"); 3052 3053 return ret; 3054 } 3055 3056 static void __init free_dma_resources(void) 3057 { 3058 free_pages((unsigned long)amd_iommu_pd_alloc_bitmap, 3059 get_order(MAX_DOMAIN_ID/8)); 3060 amd_iommu_pd_alloc_bitmap = NULL; 3061 3062 free_unity_maps(); 3063 } 3064 3065 static void __init ivinfo_init(void *ivrs) 3066 { 3067 amd_iommu_ivinfo = *((u32 *)(ivrs + IOMMU_IVINFO_OFFSET)); 3068 } 3069 3070 /* 3071 * This is the hardware init function for AMD IOMMU in the system. 3072 * This function is called either from amd_iommu_init or from the interrupt 3073 * remapping setup code. 3074 * 3075 * This function basically parses the ACPI table for AMD IOMMU (IVRS) 3076 * four times: 3077 * 3078 * 1 pass) Discover the most comprehensive IVHD type to use. 3079 * 3080 * 2 pass) Find the highest PCI device id the driver has to handle. 3081 * Upon this information the size of the data structures is 3082 * determined that needs to be allocated. 3083 * 3084 * 3 pass) Initialize the data structures just allocated with the 3085 * information in the ACPI table about available AMD IOMMUs 3086 * in the system. It also maps the PCI devices in the 3087 * system to specific IOMMUs 3088 * 3089 * 4 pass) After the basic data structures are allocated and 3090 * initialized we update them with information about memory 3091 * remapping requirements parsed out of the ACPI table in 3092 * this last pass. 3093 * 3094 * After everything is set up the IOMMUs are enabled and the necessary 3095 * hotplug and suspend notifiers are registered. 3096 */ 3097 static int __init early_amd_iommu_init(void) 3098 { 3099 struct acpi_table_header *ivrs_base; 3100 int remap_cache_sz, ret; 3101 acpi_status status; 3102 3103 if (!amd_iommu_detected) 3104 return -ENODEV; 3105 3106 status = acpi_get_table("IVRS", 0, &ivrs_base); 3107 if (status == AE_NOT_FOUND) 3108 return -ENODEV; 3109 else if (ACPI_FAILURE(status)) { 3110 const char *err = acpi_format_exception(status); 3111 pr_err("IVRS table error: %s\n", err); 3112 return -EINVAL; 3113 } 3114 3115 /* 3116 * Validate checksum here so we don't need to do it when 3117 * we actually parse the table 3118 */ 3119 ret = check_ivrs_checksum(ivrs_base); 3120 if (ret) 3121 goto out; 3122 3123 ivinfo_init(ivrs_base); 3124 3125 amd_iommu_target_ivhd_type = get_highest_supported_ivhd_type(ivrs_base); 3126 DUMP_printk("Using IVHD type %#x\n", amd_iommu_target_ivhd_type); 3127 3128 /* Device table - directly used by all IOMMUs */ 3129 ret = -ENOMEM; 3130 3131 amd_iommu_pd_alloc_bitmap = (void *)__get_free_pages( 3132 GFP_KERNEL | __GFP_ZERO, 3133 get_order(MAX_DOMAIN_ID/8)); 3134 if (amd_iommu_pd_alloc_bitmap == NULL) 3135 goto out; 3136 3137 /* 3138 * never allocate domain 0 because its used as the non-allocated and 3139 * error value placeholder 3140 */ 3141 __set_bit(0, amd_iommu_pd_alloc_bitmap); 3142 3143 /* 3144 * now the data structures are allocated and basically initialized 3145 * start the real acpi table scan 3146 */ 3147 ret = init_iommu_all(ivrs_base); 3148 if (ret) 3149 goto out; 3150 3151 /* 5 level guest page table */ 3152 if (cpu_feature_enabled(X86_FEATURE_LA57) && 3153 check_feature_gpt_level() == GUEST_PGTABLE_5_LEVEL) 3154 amd_iommu_gpt_level = PAGE_MODE_5_LEVEL; 3155 3156 /* Disable any previously enabled IOMMUs */ 3157 if (!is_kdump_kernel() || amd_iommu_disabled) 3158 disable_iommus(); 3159 3160 if (amd_iommu_irq_remap) 3161 amd_iommu_irq_remap = check_ioapic_information(); 3162 3163 if (amd_iommu_irq_remap) { 3164 struct amd_iommu_pci_seg *pci_seg; 3165 /* 3166 * Interrupt remapping enabled, create kmem_cache for the 3167 * remapping tables. 3168 */ 3169 ret = -ENOMEM; 3170 if (!AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir)) 3171 remap_cache_sz = MAX_IRQS_PER_TABLE * sizeof(u32); 3172 else 3173 remap_cache_sz = MAX_IRQS_PER_TABLE * (sizeof(u64) * 2); 3174 amd_iommu_irq_cache = kmem_cache_create("irq_remap_cache", 3175 remap_cache_sz, 3176 DTE_INTTAB_ALIGNMENT, 3177 0, NULL); 3178 if (!amd_iommu_irq_cache) 3179 goto out; 3180 3181 for_each_pci_segment(pci_seg) { 3182 if (alloc_irq_lookup_table(pci_seg)) 3183 goto out; 3184 } 3185 } 3186 3187 ret = init_memory_definitions(ivrs_base); 3188 if (ret) 3189 goto out; 3190 3191 /* init the device table */ 3192 init_device_table(); 3193 3194 out: 3195 /* Don't leak any ACPI memory */ 3196 acpi_put_table(ivrs_base); 3197 3198 return ret; 3199 } 3200 3201 static int amd_iommu_enable_interrupts(void) 3202 { 3203 struct amd_iommu *iommu; 3204 int ret = 0; 3205 3206 for_each_iommu(iommu) { 3207 ret = iommu_init_irq(iommu); 3208 if (ret) 3209 goto out; 3210 } 3211 3212 /* 3213 * Interrupt handler is ready to process interrupts. Enable 3214 * PPR and GA log interrupt for all IOMMUs. 3215 */ 3216 enable_iommus_vapic(); 3217 enable_iommus_v2(); 3218 3219 out: 3220 return ret; 3221 } 3222 3223 static bool __init detect_ivrs(void) 3224 { 3225 struct acpi_table_header *ivrs_base; 3226 acpi_status status; 3227 int i; 3228 3229 status = acpi_get_table("IVRS", 0, &ivrs_base); 3230 if (status == AE_NOT_FOUND) 3231 return false; 3232 else if (ACPI_FAILURE(status)) { 3233 const char *err = acpi_format_exception(status); 3234 pr_err("IVRS table error: %s\n", err); 3235 return false; 3236 } 3237 3238 acpi_put_table(ivrs_base); 3239 3240 if (amd_iommu_force_enable) 3241 goto out; 3242 3243 /* Don't use IOMMU if there is Stoney Ridge graphics */ 3244 for (i = 0; i < 32; i++) { 3245 u32 pci_id; 3246 3247 pci_id = read_pci_config(0, i, 0, 0); 3248 if ((pci_id & 0xffff) == 0x1002 && (pci_id >> 16) == 0x98e4) { 3249 pr_info("Disable IOMMU on Stoney Ridge\n"); 3250 return false; 3251 } 3252 } 3253 3254 out: 3255 /* Make sure ACS will be enabled during PCI probe */ 3256 pci_request_acs(); 3257 3258 return true; 3259 } 3260 3261 /**************************************************************************** 3262 * 3263 * AMD IOMMU Initialization State Machine 3264 * 3265 ****************************************************************************/ 3266 3267 static int __init state_next(void) 3268 { 3269 int ret = 0; 3270 3271 switch (init_state) { 3272 case IOMMU_START_STATE: 3273 if (!detect_ivrs()) { 3274 init_state = IOMMU_NOT_FOUND; 3275 ret = -ENODEV; 3276 } else { 3277 init_state = IOMMU_IVRS_DETECTED; 3278 } 3279 break; 3280 case IOMMU_IVRS_DETECTED: 3281 if (amd_iommu_disabled) { 3282 init_state = IOMMU_CMDLINE_DISABLED; 3283 ret = -EINVAL; 3284 } else { 3285 ret = early_amd_iommu_init(); 3286 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_ACPI_FINISHED; 3287 } 3288 break; 3289 case IOMMU_ACPI_FINISHED: 3290 early_enable_iommus(); 3291 x86_platform.iommu_shutdown = disable_iommus; 3292 init_state = IOMMU_ENABLED; 3293 break; 3294 case IOMMU_ENABLED: 3295 register_syscore_ops(&amd_iommu_syscore_ops); 3296 ret = amd_iommu_init_pci(); 3297 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_PCI_INIT; 3298 break; 3299 case IOMMU_PCI_INIT: 3300 ret = amd_iommu_enable_interrupts(); 3301 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_INTERRUPTS_EN; 3302 break; 3303 case IOMMU_INTERRUPTS_EN: 3304 init_state = IOMMU_INITIALIZED; 3305 break; 3306 case IOMMU_INITIALIZED: 3307 /* Nothing to do */ 3308 break; 3309 case IOMMU_NOT_FOUND: 3310 case IOMMU_INIT_ERROR: 3311 case IOMMU_CMDLINE_DISABLED: 3312 /* Error states => do nothing */ 3313 ret = -EINVAL; 3314 break; 3315 default: 3316 /* Unknown state */ 3317 BUG(); 3318 } 3319 3320 if (ret) { 3321 free_dma_resources(); 3322 if (!irq_remapping_enabled) { 3323 disable_iommus(); 3324 free_iommu_resources(); 3325 } else { 3326 struct amd_iommu *iommu; 3327 struct amd_iommu_pci_seg *pci_seg; 3328 3329 for_each_pci_segment(pci_seg) 3330 uninit_device_table_dma(pci_seg); 3331 3332 for_each_iommu(iommu) 3333 iommu_flush_all_caches(iommu); 3334 } 3335 } 3336 return ret; 3337 } 3338 3339 static int __init iommu_go_to_state(enum iommu_init_state state) 3340 { 3341 int ret = -EINVAL; 3342 3343 while (init_state != state) { 3344 if (init_state == IOMMU_NOT_FOUND || 3345 init_state == IOMMU_INIT_ERROR || 3346 init_state == IOMMU_CMDLINE_DISABLED) 3347 break; 3348 ret = state_next(); 3349 } 3350 3351 return ret; 3352 } 3353 3354 #ifdef CONFIG_IRQ_REMAP 3355 int __init amd_iommu_prepare(void) 3356 { 3357 int ret; 3358 3359 amd_iommu_irq_remap = true; 3360 3361 ret = iommu_go_to_state(IOMMU_ACPI_FINISHED); 3362 if (ret) { 3363 amd_iommu_irq_remap = false; 3364 return ret; 3365 } 3366 3367 return amd_iommu_irq_remap ? 0 : -ENODEV; 3368 } 3369 3370 int __init amd_iommu_enable(void) 3371 { 3372 int ret; 3373 3374 ret = iommu_go_to_state(IOMMU_ENABLED); 3375 if (ret) 3376 return ret; 3377 3378 irq_remapping_enabled = 1; 3379 return amd_iommu_xt_mode; 3380 } 3381 3382 void amd_iommu_disable(void) 3383 { 3384 amd_iommu_suspend(); 3385 } 3386 3387 int amd_iommu_reenable(int mode) 3388 { 3389 amd_iommu_resume(); 3390 3391 return 0; 3392 } 3393 3394 int __init amd_iommu_enable_faulting(void) 3395 { 3396 /* We enable MSI later when PCI is initialized */ 3397 return 0; 3398 } 3399 #endif 3400 3401 /* 3402 * This is the core init function for AMD IOMMU hardware in the system. 3403 * This function is called from the generic x86 DMA layer initialization 3404 * code. 3405 */ 3406 static int __init amd_iommu_init(void) 3407 { 3408 struct amd_iommu *iommu; 3409 int ret; 3410 3411 ret = iommu_go_to_state(IOMMU_INITIALIZED); 3412 #ifdef CONFIG_GART_IOMMU 3413 if (ret && list_empty(&amd_iommu_list)) { 3414 /* 3415 * We failed to initialize the AMD IOMMU - try fallback 3416 * to GART if possible. 3417 */ 3418 gart_iommu_init(); 3419 } 3420 #endif 3421 3422 for_each_iommu(iommu) 3423 amd_iommu_debugfs_setup(iommu); 3424 3425 return ret; 3426 } 3427 3428 static bool amd_iommu_sme_check(void) 3429 { 3430 if (!cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT) || 3431 (boot_cpu_data.x86 != 0x17)) 3432 return true; 3433 3434 /* For Fam17h, a specific level of support is required */ 3435 if (boot_cpu_data.microcode >= 0x08001205) 3436 return true; 3437 3438 if ((boot_cpu_data.microcode >= 0x08001126) && 3439 (boot_cpu_data.microcode <= 0x080011ff)) 3440 return true; 3441 3442 pr_notice("IOMMU not currently supported when SME is active\n"); 3443 3444 return false; 3445 } 3446 3447 /**************************************************************************** 3448 * 3449 * Early detect code. This code runs at IOMMU detection time in the DMA 3450 * layer. It just looks if there is an IVRS ACPI table to detect AMD 3451 * IOMMUs 3452 * 3453 ****************************************************************************/ 3454 int __init amd_iommu_detect(void) 3455 { 3456 int ret; 3457 3458 if (no_iommu || (iommu_detected && !gart_iommu_aperture)) 3459 return -ENODEV; 3460 3461 if (!amd_iommu_sme_check()) 3462 return -ENODEV; 3463 3464 ret = iommu_go_to_state(IOMMU_IVRS_DETECTED); 3465 if (ret) 3466 return ret; 3467 3468 amd_iommu_detected = true; 3469 iommu_detected = 1; 3470 x86_init.iommu.iommu_init = amd_iommu_init; 3471 3472 return 1; 3473 } 3474 3475 /**************************************************************************** 3476 * 3477 * Parsing functions for the AMD IOMMU specific kernel command line 3478 * options. 3479 * 3480 ****************************************************************************/ 3481 3482 static int __init parse_amd_iommu_dump(char *str) 3483 { 3484 amd_iommu_dump = true; 3485 3486 return 1; 3487 } 3488 3489 static int __init parse_amd_iommu_intr(char *str) 3490 { 3491 for (; *str; ++str) { 3492 if (strncmp(str, "legacy", 6) == 0) { 3493 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA; 3494 break; 3495 } 3496 if (strncmp(str, "vapic", 5) == 0) { 3497 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC; 3498 break; 3499 } 3500 } 3501 return 1; 3502 } 3503 3504 static int __init parse_amd_iommu_options(char *str) 3505 { 3506 if (!str) 3507 return -EINVAL; 3508 3509 while (*str) { 3510 if (strncmp(str, "fullflush", 9) == 0) { 3511 pr_warn("amd_iommu=fullflush deprecated; use iommu.strict=1 instead\n"); 3512 iommu_set_dma_strict(); 3513 } else if (strncmp(str, "force_enable", 12) == 0) { 3514 amd_iommu_force_enable = true; 3515 } else if (strncmp(str, "off", 3) == 0) { 3516 amd_iommu_disabled = true; 3517 } else if (strncmp(str, "force_isolation", 15) == 0) { 3518 amd_iommu_force_isolation = true; 3519 } else if (strncmp(str, "pgtbl_v1", 8) == 0) { 3520 amd_iommu_pgtable = AMD_IOMMU_V1; 3521 } else if (strncmp(str, "pgtbl_v2", 8) == 0) { 3522 amd_iommu_pgtable = AMD_IOMMU_V2; 3523 } else if (strncmp(str, "irtcachedis", 11) == 0) { 3524 amd_iommu_irtcachedis = true; 3525 } else { 3526 pr_notice("Unknown option - '%s'\n", str); 3527 } 3528 3529 str += strcspn(str, ","); 3530 while (*str == ',') 3531 str++; 3532 } 3533 3534 return 1; 3535 } 3536 3537 static int __init parse_ivrs_ioapic(char *str) 3538 { 3539 u32 seg = 0, bus, dev, fn; 3540 int id, i; 3541 u32 devid; 3542 3543 if (sscanf(str, "=%d@%x:%x.%x", &id, &bus, &dev, &fn) == 4 || 3544 sscanf(str, "=%d@%x:%x:%x.%x", &id, &seg, &bus, &dev, &fn) == 5) 3545 goto found; 3546 3547 if (sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn) == 4 || 3548 sscanf(str, "[%d]=%x:%x:%x.%x", &id, &seg, &bus, &dev, &fn) == 5) { 3549 pr_warn("ivrs_ioapic%s option format deprecated; use ivrs_ioapic=%d@%04x:%02x:%02x.%d instead\n", 3550 str, id, seg, bus, dev, fn); 3551 goto found; 3552 } 3553 3554 pr_err("Invalid command line: ivrs_ioapic%s\n", str); 3555 return 1; 3556 3557 found: 3558 if (early_ioapic_map_size == EARLY_MAP_SIZE) { 3559 pr_err("Early IOAPIC map overflow - ignoring ivrs_ioapic%s\n", 3560 str); 3561 return 1; 3562 } 3563 3564 devid = IVRS_GET_SBDF_ID(seg, bus, dev, fn); 3565 3566 cmdline_maps = true; 3567 i = early_ioapic_map_size++; 3568 early_ioapic_map[i].id = id; 3569 early_ioapic_map[i].devid = devid; 3570 early_ioapic_map[i].cmd_line = true; 3571 3572 return 1; 3573 } 3574 3575 static int __init parse_ivrs_hpet(char *str) 3576 { 3577 u32 seg = 0, bus, dev, fn; 3578 int id, i; 3579 u32 devid; 3580 3581 if (sscanf(str, "=%d@%x:%x.%x", &id, &bus, &dev, &fn) == 4 || 3582 sscanf(str, "=%d@%x:%x:%x.%x", &id, &seg, &bus, &dev, &fn) == 5) 3583 goto found; 3584 3585 if (sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn) == 4 || 3586 sscanf(str, "[%d]=%x:%x:%x.%x", &id, &seg, &bus, &dev, &fn) == 5) { 3587 pr_warn("ivrs_hpet%s option format deprecated; use ivrs_hpet=%d@%04x:%02x:%02x.%d instead\n", 3588 str, id, seg, bus, dev, fn); 3589 goto found; 3590 } 3591 3592 pr_err("Invalid command line: ivrs_hpet%s\n", str); 3593 return 1; 3594 3595 found: 3596 if (early_hpet_map_size == EARLY_MAP_SIZE) { 3597 pr_err("Early HPET map overflow - ignoring ivrs_hpet%s\n", 3598 str); 3599 return 1; 3600 } 3601 3602 devid = IVRS_GET_SBDF_ID(seg, bus, dev, fn); 3603 3604 cmdline_maps = true; 3605 i = early_hpet_map_size++; 3606 early_hpet_map[i].id = id; 3607 early_hpet_map[i].devid = devid; 3608 early_hpet_map[i].cmd_line = true; 3609 3610 return 1; 3611 } 3612 3613 #define ACPIID_LEN (ACPIHID_UID_LEN + ACPIHID_HID_LEN) 3614 3615 static int __init parse_ivrs_acpihid(char *str) 3616 { 3617 u32 seg = 0, bus, dev, fn; 3618 char *hid, *uid, *p, *addr; 3619 char acpiid[ACPIID_LEN] = {0}; 3620 int i; 3621 3622 addr = strchr(str, '@'); 3623 if (!addr) { 3624 addr = strchr(str, '='); 3625 if (!addr) 3626 goto not_found; 3627 3628 ++addr; 3629 3630 if (strlen(addr) > ACPIID_LEN) 3631 goto not_found; 3632 3633 if (sscanf(str, "[%x:%x.%x]=%s", &bus, &dev, &fn, acpiid) == 4 || 3634 sscanf(str, "[%x:%x:%x.%x]=%s", &seg, &bus, &dev, &fn, acpiid) == 5) { 3635 pr_warn("ivrs_acpihid%s option format deprecated; use ivrs_acpihid=%s@%04x:%02x:%02x.%d instead\n", 3636 str, acpiid, seg, bus, dev, fn); 3637 goto found; 3638 } 3639 goto not_found; 3640 } 3641 3642 /* We have the '@', make it the terminator to get just the acpiid */ 3643 *addr++ = 0; 3644 3645 if (strlen(str) > ACPIID_LEN + 1) 3646 goto not_found; 3647 3648 if (sscanf(str, "=%s", acpiid) != 1) 3649 goto not_found; 3650 3651 if (sscanf(addr, "%x:%x.%x", &bus, &dev, &fn) == 3 || 3652 sscanf(addr, "%x:%x:%x.%x", &seg, &bus, &dev, &fn) == 4) 3653 goto found; 3654 3655 not_found: 3656 pr_err("Invalid command line: ivrs_acpihid%s\n", str); 3657 return 1; 3658 3659 found: 3660 p = acpiid; 3661 hid = strsep(&p, ":"); 3662 uid = p; 3663 3664 if (!hid || !(*hid) || !uid) { 3665 pr_err("Invalid command line: hid or uid\n"); 3666 return 1; 3667 } 3668 3669 /* 3670 * Ignore leading zeroes after ':', so e.g., AMDI0095:00 3671 * will match AMDI0095:0 in the second strcmp in acpi_dev_hid_uid_match 3672 */ 3673 while (*uid == '0' && *(uid + 1)) 3674 uid++; 3675 3676 i = early_acpihid_map_size++; 3677 memcpy(early_acpihid_map[i].hid, hid, strlen(hid)); 3678 memcpy(early_acpihid_map[i].uid, uid, strlen(uid)); 3679 early_acpihid_map[i].devid = IVRS_GET_SBDF_ID(seg, bus, dev, fn); 3680 early_acpihid_map[i].cmd_line = true; 3681 3682 return 1; 3683 } 3684 3685 __setup("amd_iommu_dump", parse_amd_iommu_dump); 3686 __setup("amd_iommu=", parse_amd_iommu_options); 3687 __setup("amd_iommu_intr=", parse_amd_iommu_intr); 3688 __setup("ivrs_ioapic", parse_ivrs_ioapic); 3689 __setup("ivrs_hpet", parse_ivrs_hpet); 3690 __setup("ivrs_acpihid", parse_ivrs_acpihid); 3691 3692 bool amd_iommu_v2_supported(void) 3693 { 3694 /* CPU page table size should match IOMMU guest page table size */ 3695 if (cpu_feature_enabled(X86_FEATURE_LA57) && 3696 amd_iommu_gpt_level != PAGE_MODE_5_LEVEL) 3697 return false; 3698 3699 /* 3700 * Since DTE[Mode]=0 is prohibited on SNP-enabled system 3701 * (i.e. EFR[SNPSup]=1), IOMMUv2 page table cannot be used without 3702 * setting up IOMMUv1 page table. 3703 */ 3704 return amd_iommu_v2_present && !amd_iommu_snp_en; 3705 } 3706 EXPORT_SYMBOL(amd_iommu_v2_supported); 3707 3708 struct amd_iommu *get_amd_iommu(unsigned int idx) 3709 { 3710 unsigned int i = 0; 3711 struct amd_iommu *iommu; 3712 3713 for_each_iommu(iommu) 3714 if (i++ == idx) 3715 return iommu; 3716 return NULL; 3717 } 3718 3719 /**************************************************************************** 3720 * 3721 * IOMMU EFR Performance Counter support functionality. This code allows 3722 * access to the IOMMU PC functionality. 3723 * 3724 ****************************************************************************/ 3725 3726 u8 amd_iommu_pc_get_max_banks(unsigned int idx) 3727 { 3728 struct amd_iommu *iommu = get_amd_iommu(idx); 3729 3730 if (iommu) 3731 return iommu->max_banks; 3732 3733 return 0; 3734 } 3735 EXPORT_SYMBOL(amd_iommu_pc_get_max_banks); 3736 3737 bool amd_iommu_pc_supported(void) 3738 { 3739 return amd_iommu_pc_present; 3740 } 3741 EXPORT_SYMBOL(amd_iommu_pc_supported); 3742 3743 u8 amd_iommu_pc_get_max_counters(unsigned int idx) 3744 { 3745 struct amd_iommu *iommu = get_amd_iommu(idx); 3746 3747 if (iommu) 3748 return iommu->max_counters; 3749 3750 return 0; 3751 } 3752 EXPORT_SYMBOL(amd_iommu_pc_get_max_counters); 3753 3754 static int iommu_pc_get_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, 3755 u8 fxn, u64 *value, bool is_write) 3756 { 3757 u32 offset; 3758 u32 max_offset_lim; 3759 3760 /* Make sure the IOMMU PC resource is available */ 3761 if (!amd_iommu_pc_present) 3762 return -ENODEV; 3763 3764 /* Check for valid iommu and pc register indexing */ 3765 if (WARN_ON(!iommu || (fxn > 0x28) || (fxn & 7))) 3766 return -ENODEV; 3767 3768 offset = (u32)(((0x40 | bank) << 12) | (cntr << 8) | fxn); 3769 3770 /* Limit the offset to the hw defined mmio region aperture */ 3771 max_offset_lim = (u32)(((0x40 | iommu->max_banks) << 12) | 3772 (iommu->max_counters << 8) | 0x28); 3773 if ((offset < MMIO_CNTR_REG_OFFSET) || 3774 (offset > max_offset_lim)) 3775 return -EINVAL; 3776 3777 if (is_write) { 3778 u64 val = *value & GENMASK_ULL(47, 0); 3779 3780 writel((u32)val, iommu->mmio_base + offset); 3781 writel((val >> 32), iommu->mmio_base + offset + 4); 3782 } else { 3783 *value = readl(iommu->mmio_base + offset + 4); 3784 *value <<= 32; 3785 *value |= readl(iommu->mmio_base + offset); 3786 *value &= GENMASK_ULL(47, 0); 3787 } 3788 3789 return 0; 3790 } 3791 3792 int amd_iommu_pc_get_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value) 3793 { 3794 if (!iommu) 3795 return -EINVAL; 3796 3797 return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, false); 3798 } 3799 3800 int amd_iommu_pc_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value) 3801 { 3802 if (!iommu) 3803 return -EINVAL; 3804 3805 return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, true); 3806 } 3807 3808 #ifdef CONFIG_AMD_MEM_ENCRYPT 3809 int amd_iommu_snp_enable(void) 3810 { 3811 /* 3812 * The SNP support requires that IOMMU must be enabled, and is 3813 * not configured in the passthrough mode. 3814 */ 3815 if (no_iommu || iommu_default_passthrough()) { 3816 pr_err("SNP: IOMMU is disabled or configured in passthrough mode, SNP cannot be supported"); 3817 return -EINVAL; 3818 } 3819 3820 /* 3821 * Prevent enabling SNP after IOMMU_ENABLED state because this process 3822 * affect how IOMMU driver sets up data structures and configures 3823 * IOMMU hardware. 3824 */ 3825 if (init_state > IOMMU_ENABLED) { 3826 pr_err("SNP: Too late to enable SNP for IOMMU.\n"); 3827 return -EINVAL; 3828 } 3829 3830 amd_iommu_snp_en = check_feature_on_all_iommus(FEATURE_SNP); 3831 if (!amd_iommu_snp_en) 3832 return -EINVAL; 3833 3834 pr_info("SNP enabled\n"); 3835 3836 /* Enforce IOMMU v1 pagetable when SNP is enabled. */ 3837 if (amd_iommu_pgtable != AMD_IOMMU_V1) { 3838 pr_warn("Force to using AMD IOMMU v1 page table due to SNP\n"); 3839 amd_iommu_pgtable = AMD_IOMMU_V1; 3840 } 3841 3842 return 0; 3843 } 3844 #endif 3845