1 // SPDX-License-Identifier: GPL-2.0 2 3 #include <linux/bitops.h> 4 #include <linux/init.h> 5 #include <linux/kernel.h> 6 #include <linux/minmax.h> 7 #include <linux/smp.h> 8 #include <linux/string.h> 9 #include <linux/types.h> 10 11 #ifdef CONFIG_X86_64 12 #include <linux/topology.h> 13 #endif 14 15 #include <asm/bugs.h> 16 #include <asm/cpu_device_id.h> 17 #include <asm/cpufeature.h> 18 #include <asm/cpu.h> 19 #include <asm/cpuid.h> 20 #include <asm/hwcap2.h> 21 #include <asm/intel-family.h> 22 #include <asm/microcode.h> 23 #include <asm/msr.h> 24 #include <asm/numa.h> 25 #include <asm/resctrl.h> 26 #include <asm/thermal.h> 27 #include <asm/uaccess.h> 28 29 #include "cpu.h" 30 31 /* 32 * Processors which have self-snooping capability can handle conflicting 33 * memory type across CPUs by snooping its own cache. However, there exists 34 * CPU models in which having conflicting memory types still leads to 35 * unpredictable behavior, machine check errors, or hangs. Clear this 36 * feature to prevent its use on machines with known erratas. 37 */ 38 static void check_memory_type_self_snoop_errata(struct cpuinfo_x86 *c) 39 { 40 switch (c->x86_vfm) { 41 case INTEL_CORE_YONAH: 42 case INTEL_CORE2_MEROM: 43 case INTEL_CORE2_MEROM_L: 44 case INTEL_CORE2_PENRYN: 45 case INTEL_CORE2_DUNNINGTON: 46 case INTEL_NEHALEM: 47 case INTEL_NEHALEM_G: 48 case INTEL_NEHALEM_EP: 49 case INTEL_NEHALEM_EX: 50 case INTEL_WESTMERE: 51 case INTEL_WESTMERE_EP: 52 case INTEL_SANDYBRIDGE: 53 setup_clear_cpu_cap(X86_FEATURE_SELFSNOOP); 54 } 55 } 56 57 static bool ring3mwait_disabled __read_mostly; 58 59 static int __init ring3mwait_disable(char *__unused) 60 { 61 ring3mwait_disabled = true; 62 return 1; 63 } 64 __setup("ring3mwait=disable", ring3mwait_disable); 65 66 static void probe_xeon_phi_r3mwait(struct cpuinfo_x86 *c) 67 { 68 /* 69 * Ring 3 MONITOR/MWAIT feature cannot be detected without 70 * cpu model and family comparison. 71 */ 72 if (c->x86 != 6) 73 return; 74 switch (c->x86_vfm) { 75 case INTEL_XEON_PHI_KNL: 76 case INTEL_XEON_PHI_KNM: 77 break; 78 default: 79 return; 80 } 81 82 if (ring3mwait_disabled) 83 return; 84 85 set_cpu_cap(c, X86_FEATURE_RING3MWAIT); 86 this_cpu_or(msr_misc_features_shadow, 87 1UL << MSR_MISC_FEATURES_ENABLES_RING3MWAIT_BIT); 88 89 if (c == &boot_cpu_data) 90 ELF_HWCAP2 |= HWCAP2_RING3MWAIT; 91 } 92 93 /* 94 * Early microcode releases for the Spectre v2 mitigation were broken. 95 * Information taken from; 96 * - https://newsroom.intel.com/wp-content/uploads/sites/11/2018/03/microcode-update-guidance.pdf 97 * - https://kb.vmware.com/s/article/52345 98 * - Microcode revisions observed in the wild 99 * - Release note from 20180108 microcode release 100 */ 101 struct sku_microcode { 102 u32 vfm; 103 u8 stepping; 104 u32 microcode; 105 }; 106 static const struct sku_microcode spectre_bad_microcodes[] = { 107 { INTEL_KABYLAKE, 0x0B, 0x80 }, 108 { INTEL_KABYLAKE, 0x0A, 0x80 }, 109 { INTEL_KABYLAKE, 0x09, 0x80 }, 110 { INTEL_KABYLAKE_L, 0x0A, 0x80 }, 111 { INTEL_KABYLAKE_L, 0x09, 0x80 }, 112 { INTEL_SKYLAKE_X, 0x03, 0x0100013e }, 113 { INTEL_SKYLAKE_X, 0x04, 0x0200003c }, 114 { INTEL_BROADWELL, 0x04, 0x28 }, 115 { INTEL_BROADWELL_G, 0x01, 0x1b }, 116 { INTEL_BROADWELL_D, 0x02, 0x14 }, 117 { INTEL_BROADWELL_D, 0x03, 0x07000011 }, 118 { INTEL_BROADWELL_X, 0x01, 0x0b000025 }, 119 { INTEL_HASWELL_L, 0x01, 0x21 }, 120 { INTEL_HASWELL_G, 0x01, 0x18 }, 121 { INTEL_HASWELL, 0x03, 0x23 }, 122 { INTEL_HASWELL_X, 0x02, 0x3b }, 123 { INTEL_HASWELL_X, 0x04, 0x10 }, 124 { INTEL_IVYBRIDGE_X, 0x04, 0x42a }, 125 /* Observed in the wild */ 126 { INTEL_SANDYBRIDGE_X, 0x06, 0x61b }, 127 { INTEL_SANDYBRIDGE_X, 0x07, 0x712 }, 128 }; 129 130 static bool bad_spectre_microcode(struct cpuinfo_x86 *c) 131 { 132 int i; 133 134 /* 135 * We know that the hypervisor lie to us on the microcode version so 136 * we may as well hope that it is running the correct version. 137 */ 138 if (cpu_has(c, X86_FEATURE_HYPERVISOR)) 139 return false; 140 141 for (i = 0; i < ARRAY_SIZE(spectre_bad_microcodes); i++) { 142 if (c->x86_vfm == spectre_bad_microcodes[i].vfm && 143 c->x86_stepping == spectre_bad_microcodes[i].stepping) 144 return (c->microcode <= spectre_bad_microcodes[i].microcode); 145 } 146 return false; 147 } 148 149 #define MSR_IA32_TME_ACTIVATE 0x982 150 151 /* Helpers to access TME_ACTIVATE MSR */ 152 #define TME_ACTIVATE_LOCKED(x) (x & 0x1) 153 #define TME_ACTIVATE_ENABLED(x) (x & 0x2) 154 155 #define TME_ACTIVATE_KEYID_BITS(x) ((x >> 32) & 0xf) /* Bits 35:32 */ 156 157 static void detect_tme_early(struct cpuinfo_x86 *c) 158 { 159 u64 tme_activate; 160 int keyid_bits; 161 162 rdmsrl(MSR_IA32_TME_ACTIVATE, tme_activate); 163 164 if (!TME_ACTIVATE_LOCKED(tme_activate) || !TME_ACTIVATE_ENABLED(tme_activate)) { 165 pr_info_once("x86/tme: not enabled by BIOS\n"); 166 clear_cpu_cap(c, X86_FEATURE_TME); 167 return; 168 } 169 pr_info_once("x86/tme: enabled by BIOS\n"); 170 keyid_bits = TME_ACTIVATE_KEYID_BITS(tme_activate); 171 if (!keyid_bits) 172 return; 173 174 /* 175 * KeyID bits are set by BIOS and can be present regardless 176 * of whether the kernel is using them. They effectively lower 177 * the number of physical address bits. 178 * 179 * Update cpuinfo_x86::x86_phys_bits accordingly. 180 */ 181 c->x86_phys_bits -= keyid_bits; 182 pr_info_once("x86/mktme: BIOS enabled: x86_phys_bits reduced by %d\n", 183 keyid_bits); 184 } 185 186 void intel_unlock_cpuid_leafs(struct cpuinfo_x86 *c) 187 { 188 if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL) 189 return; 190 191 if (c->x86_vfm < INTEL_PENTIUM_M_DOTHAN) 192 return; 193 194 /* 195 * The BIOS can have limited CPUID to leaf 2, which breaks feature 196 * enumeration. Unlock it and update the maximum leaf info. 197 */ 198 if (msr_clear_bit(MSR_IA32_MISC_ENABLE, MSR_IA32_MISC_ENABLE_LIMIT_CPUID_BIT) > 0) 199 c->cpuid_level = cpuid_eax(0); 200 } 201 202 static void early_init_intel(struct cpuinfo_x86 *c) 203 { 204 u64 misc_enable; 205 206 if (c->x86 >= 6 && !cpu_has(c, X86_FEATURE_IA64)) 207 c->microcode = intel_get_microcode_revision(); 208 209 /* Now if any of them are set, check the blacklist and clear the lot */ 210 if ((cpu_has(c, X86_FEATURE_SPEC_CTRL) || 211 cpu_has(c, X86_FEATURE_INTEL_STIBP) || 212 cpu_has(c, X86_FEATURE_IBRS) || cpu_has(c, X86_FEATURE_IBPB) || 213 cpu_has(c, X86_FEATURE_STIBP)) && bad_spectre_microcode(c)) { 214 pr_warn("Intel Spectre v2 broken microcode detected; disabling Speculation Control\n"); 215 setup_clear_cpu_cap(X86_FEATURE_IBRS); 216 setup_clear_cpu_cap(X86_FEATURE_IBPB); 217 setup_clear_cpu_cap(X86_FEATURE_STIBP); 218 setup_clear_cpu_cap(X86_FEATURE_SPEC_CTRL); 219 setup_clear_cpu_cap(X86_FEATURE_MSR_SPEC_CTRL); 220 setup_clear_cpu_cap(X86_FEATURE_INTEL_STIBP); 221 setup_clear_cpu_cap(X86_FEATURE_SSBD); 222 setup_clear_cpu_cap(X86_FEATURE_SPEC_CTRL_SSBD); 223 } 224 225 /* 226 * Atom erratum AAE44/AAF40/AAG38/AAH41: 227 * 228 * A race condition between speculative fetches and invalidating 229 * a large page. This is worked around in microcode, but we 230 * need the microcode to have already been loaded... so if it is 231 * not, recommend a BIOS update and disable large pages. 232 */ 233 if (c->x86_vfm == INTEL_ATOM_BONNELL && c->x86_stepping <= 2 && 234 c->microcode < 0x20e) { 235 pr_warn("Atom PSE erratum detected, BIOS microcode update recommended\n"); 236 clear_cpu_cap(c, X86_FEATURE_PSE); 237 } 238 239 #ifdef CONFIG_X86_64 240 set_cpu_cap(c, X86_FEATURE_SYSENTER32); 241 #else 242 /* Netburst reports 64 bytes clflush size, but does IO in 128 bytes */ 243 if (c->x86 == 15 && c->x86_cache_alignment == 64) 244 c->x86_cache_alignment = 128; 245 #endif 246 247 /* CPUID workaround for 0F33/0F34 CPU */ 248 if (c->x86_vfm == INTEL_P4_PRESCOTT && 249 (c->x86_stepping == 0x3 || c->x86_stepping == 0x4)) 250 c->x86_phys_bits = 36; 251 252 /* 253 * c->x86_power is 8000_0007 edx. Bit 8 is TSC runs at constant rate 254 * with P/T states and does not stop in deep C-states. 255 * 256 * It is also reliable across cores and sockets. (but not across 257 * cabinets - we turn it off in that case explicitly.) 258 * 259 * Use a model-specific check for some older CPUs that have invariant 260 * TSC but may not report it architecturally via 8000_0007. 261 */ 262 if (c->x86_power & (1 << 8)) { 263 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC); 264 set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC); 265 } else if ((c->x86_vfm >= INTEL_P4_PRESCOTT && c->x86_vfm <= INTEL_P4_WILLAMETTE) || 266 (c->x86_vfm >= INTEL_CORE_YONAH && c->x86_vfm <= INTEL_IVYBRIDGE)) { 267 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC); 268 } 269 270 /* Penwell and Cloverview have the TSC which doesn't sleep on S3 */ 271 switch (c->x86_vfm) { 272 case INTEL_ATOM_SALTWELL_MID: 273 case INTEL_ATOM_SALTWELL_TABLET: 274 case INTEL_ATOM_SILVERMONT_MID: 275 case INTEL_ATOM_AIRMONT_NP: 276 set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC_S3); 277 break; 278 } 279 280 /* 281 * PAT is broken on early family 6 CPUs, the last of which 282 * is "Yonah" where the erratum is named "AN7": 283 * 284 * Page with PAT (Page Attribute Table) Set to USWC 285 * (Uncacheable Speculative Write Combine) While 286 * Associated MTRR (Memory Type Range Register) Is UC 287 * (Uncacheable) May Consolidate to UC 288 * 289 * Disable PAT and fall back to MTRR on these CPUs. 290 */ 291 if (c->x86_vfm >= INTEL_PENTIUM_PRO && 292 c->x86_vfm <= INTEL_CORE_YONAH) 293 clear_cpu_cap(c, X86_FEATURE_PAT); 294 295 /* 296 * Modern CPUs are generally expected to have a sane fast string 297 * implementation. However, BIOSes typically have a knob to tweak 298 * the architectural MISC_ENABLE.FAST_STRING enable bit. 299 * 300 * Adhere to the preference and program the Linux-defined fast 301 * string flag and enhanced fast string capabilities accordingly. 302 */ 303 if (c->x86_vfm >= INTEL_PENTIUM_M_DOTHAN) { 304 rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable); 305 if (misc_enable & MSR_IA32_MISC_ENABLE_FAST_STRING) { 306 /* X86_FEATURE_ERMS is set based on CPUID */ 307 set_cpu_cap(c, X86_FEATURE_REP_GOOD); 308 } else { 309 pr_info("Disabled fast string operations\n"); 310 setup_clear_cpu_cap(X86_FEATURE_REP_GOOD); 311 setup_clear_cpu_cap(X86_FEATURE_ERMS); 312 } 313 } 314 315 /* 316 * Intel Quark Core DevMan_001.pdf section 6.4.11 317 * "The operating system also is required to invalidate (i.e., flush) 318 * the TLB when any changes are made to any of the page table entries. 319 * The operating system must reload CR3 to cause the TLB to be flushed" 320 * 321 * As a result, boot_cpu_has(X86_FEATURE_PGE) in arch/x86/include/asm/tlbflush.h 322 * should be false so that __flush_tlb_all() causes CR3 instead of CR4.PGE 323 * to be modified. 324 */ 325 if (c->x86_vfm == INTEL_QUARK_X1000) { 326 pr_info("Disabling PGE capability bit\n"); 327 setup_clear_cpu_cap(X86_FEATURE_PGE); 328 } 329 330 check_memory_type_self_snoop_errata(c); 331 332 /* 333 * Adjust the number of physical bits early because it affects the 334 * valid bits of the MTRR mask registers. 335 */ 336 if (cpu_has(c, X86_FEATURE_TME)) 337 detect_tme_early(c); 338 } 339 340 static void bsp_init_intel(struct cpuinfo_x86 *c) 341 { 342 resctrl_cpu_detect(c); 343 } 344 345 #ifdef CONFIG_X86_32 346 /* 347 * Early probe support logic for ppro memory erratum #50 348 * 349 * This is called before we do cpu ident work 350 */ 351 352 int ppro_with_ram_bug(void) 353 { 354 /* Uses data from early_cpu_detect now */ 355 if (boot_cpu_data.x86_vfm == INTEL_PENTIUM_PRO && 356 boot_cpu_data.x86_stepping < 8) { 357 pr_info("Pentium Pro with Errata#50 detected. Taking evasive action.\n"); 358 return 1; 359 } 360 return 0; 361 } 362 363 static void intel_smp_check(struct cpuinfo_x86 *c) 364 { 365 /* calling is from identify_secondary_cpu() ? */ 366 if (!c->cpu_index) 367 return; 368 369 /* 370 * Mask B, Pentium, but not Pentium MMX 371 */ 372 if (c->x86_vfm >= INTEL_FAM5_START && c->x86_vfm < INTEL_PENTIUM_MMX && 373 c->x86_stepping >= 1 && c->x86_stepping <= 4) { 374 /* 375 * Remember we have B step Pentia with bugs 376 */ 377 WARN_ONCE(1, "WARNING: SMP operation may be unreliable" 378 "with B stepping processors.\n"); 379 } 380 } 381 382 static int forcepae; 383 static int __init forcepae_setup(char *__unused) 384 { 385 forcepae = 1; 386 return 1; 387 } 388 __setup("forcepae", forcepae_setup); 389 390 static void intel_workarounds(struct cpuinfo_x86 *c) 391 { 392 #ifdef CONFIG_X86_F00F_BUG 393 /* 394 * All models of Pentium and Pentium with MMX technology CPUs 395 * have the F0 0F bug, which lets nonprivileged users lock up the 396 * system. Announce that the fault handler will be checking for it. 397 * The Quark is also family 5, but does not have the same bug. 398 */ 399 clear_cpu_bug(c, X86_BUG_F00F); 400 if (c->x86_vfm >= INTEL_FAM5_START && c->x86_vfm < INTEL_QUARK_X1000) { 401 static int f00f_workaround_enabled; 402 403 set_cpu_bug(c, X86_BUG_F00F); 404 if (!f00f_workaround_enabled) { 405 pr_notice("Intel Pentium with F0 0F bug - workaround enabled.\n"); 406 f00f_workaround_enabled = 1; 407 } 408 } 409 #endif 410 411 /* 412 * SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until 413 * model 3 mask 3 414 */ 415 if ((c->x86_vfm == INTEL_PENTIUM_II_KLAMATH && c->x86_stepping < 3) || 416 c->x86_vfm < INTEL_PENTIUM_II_KLAMATH) 417 clear_cpu_cap(c, X86_FEATURE_SEP); 418 419 /* 420 * PAE CPUID issue: many Pentium M report no PAE but may have a 421 * functionally usable PAE implementation. 422 * Forcefully enable PAE if kernel parameter "forcepae" is present. 423 */ 424 if (forcepae) { 425 pr_warn("PAE forced!\n"); 426 set_cpu_cap(c, X86_FEATURE_PAE); 427 add_taint(TAINT_CPU_OUT_OF_SPEC, LOCKDEP_NOW_UNRELIABLE); 428 } 429 430 /* 431 * P4 Xeon erratum 037 workaround. 432 * Hardware prefetcher may cause stale data to be loaded into the cache. 433 */ 434 if (c->x86_vfm == INTEL_P4_WILLAMETTE && c->x86_stepping == 1) { 435 if (msr_set_bit(MSR_IA32_MISC_ENABLE, 436 MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE_BIT) > 0) { 437 pr_info("CPU: C0 stepping P4 Xeon detected.\n"); 438 pr_info("CPU: Disabling hardware prefetching (Erratum 037)\n"); 439 } 440 } 441 442 /* 443 * See if we have a good local APIC by checking for buggy Pentia, 444 * i.e. all B steppings and the C2 stepping of P54C when using their 445 * integrated APIC (see 11AP erratum in "Pentium Processor 446 * Specification Update"). 447 */ 448 if (boot_cpu_has(X86_FEATURE_APIC) && c->x86_vfm == INTEL_PENTIUM_75 && 449 (c->x86_stepping < 0x6 || c->x86_stepping == 0xb)) 450 set_cpu_bug(c, X86_BUG_11AP); 451 452 #ifdef CONFIG_X86_INTEL_USERCOPY 453 /* 454 * MOVSL bulk memory moves can be slow when source and dest are not 455 * both 8-byte aligned. PII/PIII only like MOVSL with 8-byte alignment. 456 * 457 * Set the preferred alignment for Pentium Pro and newer processors, as 458 * it has only been tested on these. 459 */ 460 if (c->x86_vfm >= INTEL_PENTIUM_PRO) 461 movsl_mask.mask = 7; 462 #endif 463 464 intel_smp_check(c); 465 } 466 #else 467 static void intel_workarounds(struct cpuinfo_x86 *c) 468 { 469 } 470 #endif 471 472 static void srat_detect_node(struct cpuinfo_x86 *c) 473 { 474 #ifdef CONFIG_NUMA 475 unsigned node; 476 int cpu = smp_processor_id(); 477 478 /* Don't do the funky fallback heuristics the AMD version employs 479 for now. */ 480 node = numa_cpu_node(cpu); 481 if (node == NUMA_NO_NODE || !node_online(node)) { 482 /* reuse the value from init_cpu_to_node() */ 483 node = cpu_to_node(cpu); 484 } 485 numa_set_node(cpu, node); 486 #endif 487 } 488 489 static void init_cpuid_fault(struct cpuinfo_x86 *c) 490 { 491 u64 msr; 492 493 if (!rdmsrl_safe(MSR_PLATFORM_INFO, &msr)) { 494 if (msr & MSR_PLATFORM_INFO_CPUID_FAULT) 495 set_cpu_cap(c, X86_FEATURE_CPUID_FAULT); 496 } 497 } 498 499 static void init_intel_misc_features(struct cpuinfo_x86 *c) 500 { 501 u64 msr; 502 503 if (rdmsrl_safe(MSR_MISC_FEATURES_ENABLES, &msr)) 504 return; 505 506 /* Clear all MISC features */ 507 this_cpu_write(msr_misc_features_shadow, 0); 508 509 /* Check features and update capabilities and shadow control bits */ 510 init_cpuid_fault(c); 511 probe_xeon_phi_r3mwait(c); 512 513 msr = this_cpu_read(msr_misc_features_shadow); 514 wrmsrl(MSR_MISC_FEATURES_ENABLES, msr); 515 } 516 517 static void init_intel(struct cpuinfo_x86 *c) 518 { 519 early_init_intel(c); 520 521 intel_workarounds(c); 522 523 init_intel_cacheinfo(c); 524 525 if (c->cpuid_level > 9) { 526 unsigned eax = cpuid_eax(10); 527 /* Check for version and the number of counters */ 528 if ((eax & 0xff) && (((eax>>8) & 0xff) > 1)) 529 set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON); 530 } 531 532 if (cpu_has(c, X86_FEATURE_XMM2)) 533 set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC); 534 535 if (boot_cpu_has(X86_FEATURE_DS)) { 536 unsigned int l1, l2; 537 538 rdmsr(MSR_IA32_MISC_ENABLE, l1, l2); 539 if (!(l1 & MSR_IA32_MISC_ENABLE_BTS_UNAVAIL)) 540 set_cpu_cap(c, X86_FEATURE_BTS); 541 if (!(l1 & MSR_IA32_MISC_ENABLE_PEBS_UNAVAIL)) 542 set_cpu_cap(c, X86_FEATURE_PEBS); 543 } 544 545 if (boot_cpu_has(X86_FEATURE_CLFLUSH) && 546 (c->x86_vfm == INTEL_CORE2_DUNNINGTON || 547 c->x86_vfm == INTEL_NEHALEM_EX || 548 c->x86_vfm == INTEL_WESTMERE_EX)) 549 set_cpu_bug(c, X86_BUG_CLFLUSH_MONITOR); 550 551 if (boot_cpu_has(X86_FEATURE_MWAIT) && 552 (c->x86_vfm == INTEL_ATOM_GOLDMONT || 553 c->x86_vfm == INTEL_LUNARLAKE_M)) 554 set_cpu_bug(c, X86_BUG_MONITOR); 555 556 #ifdef CONFIG_X86_64 557 if (c->x86 == 15) 558 c->x86_cache_alignment = c->x86_clflush_size * 2; 559 #else 560 /* 561 * Names for the Pentium II/Celeron processors 562 * detectable only by also checking the cache size. 563 * Dixon is NOT a Celeron. 564 */ 565 if (c->x86 == 6) { 566 unsigned int l2 = c->x86_cache_size; 567 char *p = NULL; 568 569 switch (c->x86_model) { 570 case 5: 571 if (l2 == 0) 572 p = "Celeron (Covington)"; 573 else if (l2 == 256) 574 p = "Mobile Pentium II (Dixon)"; 575 break; 576 577 case 6: 578 if (l2 == 128) 579 p = "Celeron (Mendocino)"; 580 else if (c->x86_stepping == 0 || c->x86_stepping == 5) 581 p = "Celeron-A"; 582 break; 583 584 case 8: 585 if (l2 == 128) 586 p = "Celeron (Coppermine)"; 587 break; 588 } 589 590 if (p) 591 strcpy(c->x86_model_id, p); 592 } 593 #endif 594 595 /* Work around errata */ 596 srat_detect_node(c); 597 598 init_ia32_feat_ctl(c); 599 600 init_intel_misc_features(c); 601 602 split_lock_init(); 603 604 intel_init_thermal(c); 605 } 606 607 #ifdef CONFIG_X86_32 608 static unsigned int intel_size_cache(struct cpuinfo_x86 *c, unsigned int size) 609 { 610 /* 611 * Intel PIII Tualatin. This comes in two flavours. 612 * One has 256kb of cache, the other 512. We have no way 613 * to determine which, so we use a boottime override 614 * for the 512kb model, and assume 256 otherwise. 615 */ 616 if (c->x86_vfm == INTEL_PENTIUM_III_TUALATIN && size == 0) 617 size = 256; 618 619 /* 620 * Intel Quark SoC X1000 contains a 4-way set associative 621 * 16K cache with a 16 byte cache line and 256 lines per tag 622 */ 623 if (c->x86_vfm == INTEL_QUARK_X1000) 624 size = 16; 625 return size; 626 } 627 #endif 628 629 #define TLB_INST_4K 0x01 630 #define TLB_INST_4M 0x02 631 #define TLB_INST_2M_4M 0x03 632 633 #define TLB_INST_ALL 0x05 634 #define TLB_INST_1G 0x06 635 636 #define TLB_DATA_4K 0x11 637 #define TLB_DATA_4M 0x12 638 #define TLB_DATA_2M_4M 0x13 639 #define TLB_DATA_4K_4M 0x14 640 641 #define TLB_DATA_1G 0x16 642 #define TLB_DATA_1G_2M_4M 0x17 643 644 #define TLB_DATA0_4K 0x21 645 #define TLB_DATA0_4M 0x22 646 #define TLB_DATA0_2M_4M 0x23 647 648 #define STLB_4K 0x41 649 #define STLB_4K_2M 0x42 650 651 /* 652 * All of leaf 0x2's one-byte TLB descriptors implies the same number of 653 * entries for their respective TLB types. The 0x63 descriptor is an 654 * exception: it implies 4 dTLB entries for 1GB pages 32 dTLB entries 655 * for 2MB or 4MB pages. Encode descriptor 0x63 dTLB entry count for 656 * 2MB/4MB pages here, as its count for dTLB 1GB pages is already at the 657 * intel_tlb_table[] mapping. 658 */ 659 #define TLB_0x63_2M_4M_ENTRIES 32 660 661 struct _tlb_table { 662 unsigned char descriptor; 663 char tlb_type; 664 unsigned int entries; 665 }; 666 667 static const struct _tlb_table intel_tlb_table[] = { 668 { 0x01, TLB_INST_4K, 32}, /* TLB_INST 4 KByte pages, 4-way set associative */ 669 { 0x02, TLB_INST_4M, 2}, /* TLB_INST 4 MByte pages, full associative */ 670 { 0x03, TLB_DATA_4K, 64}, /* TLB_DATA 4 KByte pages, 4-way set associative */ 671 { 0x04, TLB_DATA_4M, 8}, /* TLB_DATA 4 MByte pages, 4-way set associative */ 672 { 0x05, TLB_DATA_4M, 32}, /* TLB_DATA 4 MByte pages, 4-way set associative */ 673 { 0x0b, TLB_INST_4M, 4}, /* TLB_INST 4 MByte pages, 4-way set associative */ 674 { 0x4f, TLB_INST_4K, 32}, /* TLB_INST 4 KByte pages */ 675 { 0x50, TLB_INST_ALL, 64}, /* TLB_INST 4 KByte and 2-MByte or 4-MByte pages */ 676 { 0x51, TLB_INST_ALL, 128}, /* TLB_INST 4 KByte and 2-MByte or 4-MByte pages */ 677 { 0x52, TLB_INST_ALL, 256}, /* TLB_INST 4 KByte and 2-MByte or 4-MByte pages */ 678 { 0x55, TLB_INST_2M_4M, 7}, /* TLB_INST 2-MByte or 4-MByte pages, fully associative */ 679 { 0x56, TLB_DATA0_4M, 16}, /* TLB_DATA0 4 MByte pages, 4-way set associative */ 680 { 0x57, TLB_DATA0_4K, 16}, /* TLB_DATA0 4 KByte pages, 4-way associative */ 681 { 0x59, TLB_DATA0_4K, 16}, /* TLB_DATA0 4 KByte pages, fully associative */ 682 { 0x5a, TLB_DATA0_2M_4M, 32}, /* TLB_DATA0 2-MByte or 4 MByte pages, 4-way set associative */ 683 { 0x5b, TLB_DATA_4K_4M, 64}, /* TLB_DATA 4 KByte and 4 MByte pages */ 684 { 0x5c, TLB_DATA_4K_4M, 128}, /* TLB_DATA 4 KByte and 4 MByte pages */ 685 { 0x5d, TLB_DATA_4K_4M, 256}, /* TLB_DATA 4 KByte and 4 MByte pages */ 686 { 0x61, TLB_INST_4K, 48}, /* TLB_INST 4 KByte pages, full associative */ 687 { 0x63, TLB_DATA_1G_2M_4M, 4}, /* TLB_DATA 1 GByte pages, 4-way set associative 688 * (plus 32 entries TLB_DATA 2 MByte or 4 MByte pages, not encoded here) */ 689 { 0x6b, TLB_DATA_4K, 256}, /* TLB_DATA 4 KByte pages, 8-way associative */ 690 { 0x6c, TLB_DATA_2M_4M, 128}, /* TLB_DATA 2 MByte or 4 MByte pages, 8-way associative */ 691 { 0x6d, TLB_DATA_1G, 16}, /* TLB_DATA 1 GByte pages, fully associative */ 692 { 0x76, TLB_INST_2M_4M, 8}, /* TLB_INST 2-MByte or 4-MByte pages, fully associative */ 693 { 0xb0, TLB_INST_4K, 128}, /* TLB_INST 4 KByte pages, 4-way set associative */ 694 { 0xb1, TLB_INST_2M_4M, 4}, /* TLB_INST 2M pages, 4-way, 8 entries or 4M pages, 4-way entries */ 695 { 0xb2, TLB_INST_4K, 64}, /* TLB_INST 4KByte pages, 4-way set associative */ 696 { 0xb3, TLB_DATA_4K, 128}, /* TLB_DATA 4 KByte pages, 4-way set associative */ 697 { 0xb4, TLB_DATA_4K, 256}, /* TLB_DATA 4 KByte pages, 4-way associative */ 698 { 0xb5, TLB_INST_4K, 64}, /* TLB_INST 4 KByte pages, 8-way set associative */ 699 { 0xb6, TLB_INST_4K, 128}, /* TLB_INST 4 KByte pages, 8-way set associative */ 700 { 0xba, TLB_DATA_4K, 64}, /* TLB_DATA 4 KByte pages, 4-way associative */ 701 { 0xc0, TLB_DATA_4K_4M, 8}, /* TLB_DATA 4 KByte and 4 MByte pages, 4-way associative */ 702 { 0xc1, STLB_4K_2M, 1024}, /* STLB 4 KByte and 2 MByte pages, 8-way associative */ 703 { 0xc2, TLB_DATA_2M_4M, 16}, /* TLB_DATA 2 MByte/4MByte pages, 4-way associative */ 704 { 0xca, STLB_4K, 512}, /* STLB 4 KByte pages, 4-way associative */ 705 { 0x00, 0, 0 } 706 }; 707 708 static void intel_tlb_lookup(const unsigned char desc) 709 { 710 unsigned int entries; 711 unsigned char k; 712 713 if (desc == 0) 714 return; 715 716 /* look up this descriptor in the table */ 717 for (k = 0; intel_tlb_table[k].descriptor != desc && 718 intel_tlb_table[k].descriptor != 0; k++) 719 ; 720 721 if (intel_tlb_table[k].tlb_type == 0) 722 return; 723 724 entries = intel_tlb_table[k].entries; 725 switch (intel_tlb_table[k].tlb_type) { 726 case STLB_4K: 727 tlb_lli_4k = max(tlb_lli_4k, entries); 728 tlb_lld_4k = max(tlb_lld_4k, entries); 729 break; 730 case STLB_4K_2M: 731 tlb_lli_4k = max(tlb_lli_4k, entries); 732 tlb_lld_4k = max(tlb_lld_4k, entries); 733 tlb_lli_2m = max(tlb_lli_2m, entries); 734 tlb_lld_2m = max(tlb_lld_2m, entries); 735 tlb_lli_4m = max(tlb_lli_4m, entries); 736 tlb_lld_4m = max(tlb_lld_4m, entries); 737 break; 738 case TLB_INST_ALL: 739 tlb_lli_4k = max(tlb_lli_4k, entries); 740 tlb_lli_2m = max(tlb_lli_2m, entries); 741 tlb_lli_4m = max(tlb_lli_4m, entries); 742 break; 743 case TLB_INST_4K: 744 tlb_lli_4k = max(tlb_lli_4k, entries); 745 break; 746 case TLB_INST_4M: 747 tlb_lli_4m = max(tlb_lli_4m, entries); 748 break; 749 case TLB_INST_2M_4M: 750 tlb_lli_2m = max(tlb_lli_2m, entries); 751 tlb_lli_4m = max(tlb_lli_4m, entries); 752 break; 753 case TLB_DATA_4K: 754 case TLB_DATA0_4K: 755 tlb_lld_4k = max(tlb_lld_4k, entries); 756 break; 757 case TLB_DATA_4M: 758 case TLB_DATA0_4M: 759 tlb_lld_4m = max(tlb_lld_4m, entries); 760 break; 761 case TLB_DATA_2M_4M: 762 case TLB_DATA0_2M_4M: 763 tlb_lld_2m = max(tlb_lld_2m, entries); 764 tlb_lld_4m = max(tlb_lld_4m, entries); 765 break; 766 case TLB_DATA_4K_4M: 767 tlb_lld_4k = max(tlb_lld_4k, entries); 768 tlb_lld_4m = max(tlb_lld_4m, entries); 769 break; 770 case TLB_DATA_1G_2M_4M: 771 tlb_lld_2m = max(tlb_lld_2m, TLB_0x63_2M_4M_ENTRIES); 772 tlb_lld_4m = max(tlb_lld_4m, TLB_0x63_2M_4M_ENTRIES); 773 fallthrough; 774 case TLB_DATA_1G: 775 tlb_lld_1g = max(tlb_lld_1g, entries); 776 break; 777 } 778 } 779 780 static void intel_detect_tlb(struct cpuinfo_x86 *c) 781 { 782 union leaf_0x2_regs regs; 783 u8 *desc; 784 785 if (c->cpuid_level < 2) 786 return; 787 788 cpuid_get_leaf_0x2_regs(®s); 789 for_each_leaf_0x2_desc(regs, desc) 790 intel_tlb_lookup(*desc); 791 } 792 793 static const struct cpu_dev intel_cpu_dev = { 794 .c_vendor = "Intel", 795 .c_ident = { "GenuineIntel" }, 796 #ifdef CONFIG_X86_32 797 .legacy_models = { 798 { .family = 4, .model_names = 799 { 800 [0] = "486 DX-25/33", 801 [1] = "486 DX-50", 802 [2] = "486 SX", 803 [3] = "486 DX/2", 804 [4] = "486 SL", 805 [5] = "486 SX/2", 806 [7] = "486 DX/2-WB", 807 [8] = "486 DX/4", 808 [9] = "486 DX/4-WB" 809 } 810 }, 811 { .family = 5, .model_names = 812 { 813 [0] = "Pentium 60/66 A-step", 814 [1] = "Pentium 60/66", 815 [2] = "Pentium 75 - 200", 816 [3] = "OverDrive PODP5V83", 817 [4] = "Pentium MMX", 818 [7] = "Mobile Pentium 75 - 200", 819 [8] = "Mobile Pentium MMX", 820 [9] = "Quark SoC X1000", 821 } 822 }, 823 { .family = 6, .model_names = 824 { 825 [0] = "Pentium Pro A-step", 826 [1] = "Pentium Pro", 827 [3] = "Pentium II (Klamath)", 828 [4] = "Pentium II (Deschutes)", 829 [5] = "Pentium II (Deschutes)", 830 [6] = "Mobile Pentium II", 831 [7] = "Pentium III (Katmai)", 832 [8] = "Pentium III (Coppermine)", 833 [10] = "Pentium III (Cascades)", 834 [11] = "Pentium III (Tualatin)", 835 } 836 }, 837 { .family = 15, .model_names = 838 { 839 [0] = "Pentium 4 (Unknown)", 840 [1] = "Pentium 4 (Willamette)", 841 [2] = "Pentium 4 (Northwood)", 842 [4] = "Pentium 4 (Foster)", 843 [5] = "Pentium 4 (Foster)", 844 } 845 }, 846 }, 847 .legacy_cache_size = intel_size_cache, 848 #endif 849 .c_detect_tlb = intel_detect_tlb, 850 .c_early_init = early_init_intel, 851 .c_bsp_init = bsp_init_intel, 852 .c_init = init_intel, 853 .c_x86_vendor = X86_VENDOR_INTEL, 854 }; 855 856 cpu_dev_register(intel_cpu_dev); 857