1 // SPDX-License-Identifier: GPL-2.0-only 2 /* cpu_feature_enabled() cannot be used this early */ 3 #define USE_EARLY_PGTABLE_L5 4 5 #include <linux/memblock.h> 6 #include <linux/linkage.h> 7 #include <linux/bitops.h> 8 #include <linux/kernel.h> 9 #include <linux/export.h> 10 #include <linux/percpu.h> 11 #include <linux/string.h> 12 #include <linux/ctype.h> 13 #include <linux/delay.h> 14 #include <linux/sched/mm.h> 15 #include <linux/sched/clock.h> 16 #include <linux/sched/task.h> 17 #include <linux/sched/smt.h> 18 #include <linux/init.h> 19 #include <linux/kprobes.h> 20 #include <linux/kgdb.h> 21 #include <linux/mem_encrypt.h> 22 #include <linux/smp.h> 23 #include <linux/cpu.h> 24 #include <linux/io.h> 25 #include <linux/syscore_ops.h> 26 #include <linux/pgtable.h> 27 #include <linux/stackprotector.h> 28 #include <linux/utsname.h> 29 30 #include <asm/alternative.h> 31 #include <asm/cmdline.h> 32 #include <asm/perf_event.h> 33 #include <asm/mmu_context.h> 34 #include <asm/doublefault.h> 35 #include <asm/archrandom.h> 36 #include <asm/hypervisor.h> 37 #include <asm/processor.h> 38 #include <asm/tlbflush.h> 39 #include <asm/debugreg.h> 40 #include <asm/sections.h> 41 #include <asm/vsyscall.h> 42 #include <linux/topology.h> 43 #include <linux/cpumask.h> 44 #include <linux/atomic.h> 45 #include <asm/proto.h> 46 #include <asm/setup.h> 47 #include <asm/apic.h> 48 #include <asm/desc.h> 49 #include <asm/fpu/api.h> 50 #include <asm/mtrr.h> 51 #include <asm/hwcap2.h> 52 #include <linux/numa.h> 53 #include <asm/numa.h> 54 #include <asm/asm.h> 55 #include <asm/bugs.h> 56 #include <asm/cpu.h> 57 #include <asm/mce.h> 58 #include <asm/msr.h> 59 #include <asm/cacheinfo.h> 60 #include <asm/memtype.h> 61 #include <asm/microcode.h> 62 #include <asm/intel-family.h> 63 #include <asm/cpu_device_id.h> 64 #include <asm/uv/uv.h> 65 #include <asm/ia32.h> 66 #include <asm/set_memory.h> 67 #include <asm/traps.h> 68 #include <asm/sev.h> 69 #include <asm/tdx.h> 70 71 #include "cpu.h" 72 73 u32 elf_hwcap2 __read_mostly; 74 75 /* Number of siblings per CPU package */ 76 int smp_num_siblings = 1; 77 EXPORT_SYMBOL(smp_num_siblings); 78 79 static struct ppin_info { 80 int feature; 81 int msr_ppin_ctl; 82 int msr_ppin; 83 } ppin_info[] = { 84 [X86_VENDOR_INTEL] = { 85 .feature = X86_FEATURE_INTEL_PPIN, 86 .msr_ppin_ctl = MSR_PPIN_CTL, 87 .msr_ppin = MSR_PPIN 88 }, 89 [X86_VENDOR_AMD] = { 90 .feature = X86_FEATURE_AMD_PPIN, 91 .msr_ppin_ctl = MSR_AMD_PPIN_CTL, 92 .msr_ppin = MSR_AMD_PPIN 93 }, 94 }; 95 96 static const struct x86_cpu_id ppin_cpuids[] = { 97 X86_MATCH_FEATURE(X86_FEATURE_AMD_PPIN, &ppin_info[X86_VENDOR_AMD]), 98 X86_MATCH_FEATURE(X86_FEATURE_INTEL_PPIN, &ppin_info[X86_VENDOR_INTEL]), 99 100 /* Legacy models without CPUID enumeration */ 101 X86_MATCH_INTEL_FAM6_MODEL(IVYBRIDGE_X, &ppin_info[X86_VENDOR_INTEL]), 102 X86_MATCH_INTEL_FAM6_MODEL(HASWELL_X, &ppin_info[X86_VENDOR_INTEL]), 103 X86_MATCH_INTEL_FAM6_MODEL(BROADWELL_D, &ppin_info[X86_VENDOR_INTEL]), 104 X86_MATCH_INTEL_FAM6_MODEL(BROADWELL_X, &ppin_info[X86_VENDOR_INTEL]), 105 X86_MATCH_INTEL_FAM6_MODEL(SKYLAKE_X, &ppin_info[X86_VENDOR_INTEL]), 106 X86_MATCH_INTEL_FAM6_MODEL(ICELAKE_X, &ppin_info[X86_VENDOR_INTEL]), 107 X86_MATCH_INTEL_FAM6_MODEL(ICELAKE_D, &ppin_info[X86_VENDOR_INTEL]), 108 X86_MATCH_INTEL_FAM6_MODEL(SAPPHIRERAPIDS_X, &ppin_info[X86_VENDOR_INTEL]), 109 X86_MATCH_INTEL_FAM6_MODEL(EMERALDRAPIDS_X, &ppin_info[X86_VENDOR_INTEL]), 110 X86_MATCH_INTEL_FAM6_MODEL(XEON_PHI_KNL, &ppin_info[X86_VENDOR_INTEL]), 111 X86_MATCH_INTEL_FAM6_MODEL(XEON_PHI_KNM, &ppin_info[X86_VENDOR_INTEL]), 112 113 {} 114 }; 115 116 static void ppin_init(struct cpuinfo_x86 *c) 117 { 118 const struct x86_cpu_id *id; 119 unsigned long long val; 120 struct ppin_info *info; 121 122 id = x86_match_cpu(ppin_cpuids); 123 if (!id) 124 return; 125 126 /* 127 * Testing the presence of the MSR is not enough. Need to check 128 * that the PPIN_CTL allows reading of the PPIN. 129 */ 130 info = (struct ppin_info *)id->driver_data; 131 132 if (rdmsrl_safe(info->msr_ppin_ctl, &val)) 133 goto clear_ppin; 134 135 if ((val & 3UL) == 1UL) { 136 /* PPIN locked in disabled mode */ 137 goto clear_ppin; 138 } 139 140 /* If PPIN is disabled, try to enable */ 141 if (!(val & 2UL)) { 142 wrmsrl_safe(info->msr_ppin_ctl, val | 2UL); 143 rdmsrl_safe(info->msr_ppin_ctl, &val); 144 } 145 146 /* Is the enable bit set? */ 147 if (val & 2UL) { 148 c->ppin = __rdmsr(info->msr_ppin); 149 set_cpu_cap(c, info->feature); 150 return; 151 } 152 153 clear_ppin: 154 clear_cpu_cap(c, info->feature); 155 } 156 157 static void default_init(struct cpuinfo_x86 *c) 158 { 159 #ifdef CONFIG_X86_64 160 cpu_detect_cache_sizes(c); 161 #else 162 /* Not much we can do here... */ 163 /* Check if at least it has cpuid */ 164 if (c->cpuid_level == -1) { 165 /* No cpuid. It must be an ancient CPU */ 166 if (c->x86 == 4) 167 strcpy(c->x86_model_id, "486"); 168 else if (c->x86 == 3) 169 strcpy(c->x86_model_id, "386"); 170 } 171 #endif 172 } 173 174 static const struct cpu_dev default_cpu = { 175 .c_init = default_init, 176 .c_vendor = "Unknown", 177 .c_x86_vendor = X86_VENDOR_UNKNOWN, 178 }; 179 180 static const struct cpu_dev *this_cpu = &default_cpu; 181 182 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = { 183 #ifdef CONFIG_X86_64 184 /* 185 * We need valid kernel segments for data and code in long mode too 186 * IRET will check the segment types kkeil 2000/10/28 187 * Also sysret mandates a special GDT layout 188 * 189 * TLS descriptors are currently at a different place compared to i386. 190 * Hopefully nobody expects them at a fixed place (Wine?) 191 */ 192 [GDT_ENTRY_KERNEL32_CS] = GDT_ENTRY_INIT(DESC_CODE32, 0, 0xfffff), 193 [GDT_ENTRY_KERNEL_CS] = GDT_ENTRY_INIT(DESC_CODE64, 0, 0xfffff), 194 [GDT_ENTRY_KERNEL_DS] = GDT_ENTRY_INIT(DESC_DATA64, 0, 0xfffff), 195 [GDT_ENTRY_DEFAULT_USER32_CS] = GDT_ENTRY_INIT(DESC_CODE32 | DESC_USER, 0, 0xfffff), 196 [GDT_ENTRY_DEFAULT_USER_DS] = GDT_ENTRY_INIT(DESC_DATA64 | DESC_USER, 0, 0xfffff), 197 [GDT_ENTRY_DEFAULT_USER_CS] = GDT_ENTRY_INIT(DESC_CODE64 | DESC_USER, 0, 0xfffff), 198 #else 199 [GDT_ENTRY_KERNEL_CS] = GDT_ENTRY_INIT(DESC_CODE32, 0, 0xfffff), 200 [GDT_ENTRY_KERNEL_DS] = GDT_ENTRY_INIT(DESC_DATA32, 0, 0xfffff), 201 [GDT_ENTRY_DEFAULT_USER_CS] = GDT_ENTRY_INIT(DESC_CODE32 | DESC_USER, 0, 0xfffff), 202 [GDT_ENTRY_DEFAULT_USER_DS] = GDT_ENTRY_INIT(DESC_DATA32 | DESC_USER, 0, 0xfffff), 203 /* 204 * Segments used for calling PnP BIOS have byte granularity. 205 * They code segments and data segments have fixed 64k limits, 206 * the transfer segment sizes are set at run time. 207 */ 208 [GDT_ENTRY_PNPBIOS_CS32] = GDT_ENTRY_INIT(DESC_CODE32_BIOS, 0, 0xffff), 209 [GDT_ENTRY_PNPBIOS_CS16] = GDT_ENTRY_INIT(DESC_CODE16, 0, 0xffff), 210 [GDT_ENTRY_PNPBIOS_DS] = GDT_ENTRY_INIT(DESC_DATA16, 0, 0xffff), 211 [GDT_ENTRY_PNPBIOS_TS1] = GDT_ENTRY_INIT(DESC_DATA16, 0, 0), 212 [GDT_ENTRY_PNPBIOS_TS2] = GDT_ENTRY_INIT(DESC_DATA16, 0, 0), 213 /* 214 * The APM segments have byte granularity and their bases 215 * are set at run time. All have 64k limits. 216 */ 217 [GDT_ENTRY_APMBIOS_BASE] = GDT_ENTRY_INIT(DESC_CODE32_BIOS, 0, 0xffff), 218 [GDT_ENTRY_APMBIOS_BASE+1] = GDT_ENTRY_INIT(DESC_CODE16, 0, 0xffff), 219 [GDT_ENTRY_APMBIOS_BASE+2] = GDT_ENTRY_INIT(DESC_DATA32_BIOS, 0, 0xffff), 220 221 [GDT_ENTRY_ESPFIX_SS] = GDT_ENTRY_INIT(DESC_DATA32, 0, 0xfffff), 222 [GDT_ENTRY_PERCPU] = GDT_ENTRY_INIT(DESC_DATA32, 0, 0xfffff), 223 #endif 224 } }; 225 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page); 226 227 #ifdef CONFIG_X86_64 228 static int __init x86_nopcid_setup(char *s) 229 { 230 /* nopcid doesn't accept parameters */ 231 if (s) 232 return -EINVAL; 233 234 /* do not emit a message if the feature is not present */ 235 if (!boot_cpu_has(X86_FEATURE_PCID)) 236 return 0; 237 238 setup_clear_cpu_cap(X86_FEATURE_PCID); 239 pr_info("nopcid: PCID feature disabled\n"); 240 return 0; 241 } 242 early_param("nopcid", x86_nopcid_setup); 243 #endif 244 245 static int __init x86_noinvpcid_setup(char *s) 246 { 247 /* noinvpcid doesn't accept parameters */ 248 if (s) 249 return -EINVAL; 250 251 /* do not emit a message if the feature is not present */ 252 if (!boot_cpu_has(X86_FEATURE_INVPCID)) 253 return 0; 254 255 setup_clear_cpu_cap(X86_FEATURE_INVPCID); 256 pr_info("noinvpcid: INVPCID feature disabled\n"); 257 return 0; 258 } 259 early_param("noinvpcid", x86_noinvpcid_setup); 260 261 #ifdef CONFIG_X86_32 262 static int cachesize_override = -1; 263 static int disable_x86_serial_nr = 1; 264 265 static int __init cachesize_setup(char *str) 266 { 267 get_option(&str, &cachesize_override); 268 return 1; 269 } 270 __setup("cachesize=", cachesize_setup); 271 272 /* Standard macro to see if a specific flag is changeable */ 273 static inline int flag_is_changeable_p(u32 flag) 274 { 275 u32 f1, f2; 276 277 /* 278 * Cyrix and IDT cpus allow disabling of CPUID 279 * so the code below may return different results 280 * when it is executed before and after enabling 281 * the CPUID. Add "volatile" to not allow gcc to 282 * optimize the subsequent calls to this function. 283 */ 284 asm volatile ("pushfl \n\t" 285 "pushfl \n\t" 286 "popl %0 \n\t" 287 "movl %0, %1 \n\t" 288 "xorl %2, %0 \n\t" 289 "pushl %0 \n\t" 290 "popfl \n\t" 291 "pushfl \n\t" 292 "popl %0 \n\t" 293 "popfl \n\t" 294 295 : "=&r" (f1), "=&r" (f2) 296 : "ir" (flag)); 297 298 return ((f1^f2) & flag) != 0; 299 } 300 301 /* Probe for the CPUID instruction */ 302 int have_cpuid_p(void) 303 { 304 return flag_is_changeable_p(X86_EFLAGS_ID); 305 } 306 307 static void squash_the_stupid_serial_number(struct cpuinfo_x86 *c) 308 { 309 unsigned long lo, hi; 310 311 if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr) 312 return; 313 314 /* Disable processor serial number: */ 315 316 rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi); 317 lo |= 0x200000; 318 wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi); 319 320 pr_notice("CPU serial number disabled.\n"); 321 clear_cpu_cap(c, X86_FEATURE_PN); 322 323 /* Disabling the serial number may affect the cpuid level */ 324 c->cpuid_level = cpuid_eax(0); 325 } 326 327 static int __init x86_serial_nr_setup(char *s) 328 { 329 disable_x86_serial_nr = 0; 330 return 1; 331 } 332 __setup("serialnumber", x86_serial_nr_setup); 333 #else 334 static inline int flag_is_changeable_p(u32 flag) 335 { 336 return 1; 337 } 338 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c) 339 { 340 } 341 #endif 342 343 static __always_inline void setup_smep(struct cpuinfo_x86 *c) 344 { 345 if (cpu_has(c, X86_FEATURE_SMEP)) 346 cr4_set_bits(X86_CR4_SMEP); 347 } 348 349 static __always_inline void setup_smap(struct cpuinfo_x86 *c) 350 { 351 unsigned long eflags = native_save_fl(); 352 353 /* This should have been cleared long ago */ 354 BUG_ON(eflags & X86_EFLAGS_AC); 355 356 if (cpu_has(c, X86_FEATURE_SMAP)) 357 cr4_set_bits(X86_CR4_SMAP); 358 } 359 360 static __always_inline void setup_umip(struct cpuinfo_x86 *c) 361 { 362 /* Check the boot processor, plus build option for UMIP. */ 363 if (!cpu_feature_enabled(X86_FEATURE_UMIP)) 364 goto out; 365 366 /* Check the current processor's cpuid bits. */ 367 if (!cpu_has(c, X86_FEATURE_UMIP)) 368 goto out; 369 370 cr4_set_bits(X86_CR4_UMIP); 371 372 pr_info_once("x86/cpu: User Mode Instruction Prevention (UMIP) activated\n"); 373 374 return; 375 376 out: 377 /* 378 * Make sure UMIP is disabled in case it was enabled in a 379 * previous boot (e.g., via kexec). 380 */ 381 cr4_clear_bits(X86_CR4_UMIP); 382 } 383 384 /* These bits should not change their value after CPU init is finished. */ 385 static const unsigned long cr4_pinned_mask = 386 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_UMIP | 387 X86_CR4_FSGSBASE | X86_CR4_CET; 388 static DEFINE_STATIC_KEY_FALSE_RO(cr_pinning); 389 static unsigned long cr4_pinned_bits __ro_after_init; 390 391 void native_write_cr0(unsigned long val) 392 { 393 unsigned long bits_missing = 0; 394 395 set_register: 396 asm volatile("mov %0,%%cr0": "+r" (val) : : "memory"); 397 398 if (static_branch_likely(&cr_pinning)) { 399 if (unlikely((val & X86_CR0_WP) != X86_CR0_WP)) { 400 bits_missing = X86_CR0_WP; 401 val |= bits_missing; 402 goto set_register; 403 } 404 /* Warn after we've set the missing bits. */ 405 WARN_ONCE(bits_missing, "CR0 WP bit went missing!?\n"); 406 } 407 } 408 EXPORT_SYMBOL(native_write_cr0); 409 410 void __no_profile native_write_cr4(unsigned long val) 411 { 412 unsigned long bits_changed = 0; 413 414 set_register: 415 asm volatile("mov %0,%%cr4": "+r" (val) : : "memory"); 416 417 if (static_branch_likely(&cr_pinning)) { 418 if (unlikely((val & cr4_pinned_mask) != cr4_pinned_bits)) { 419 bits_changed = (val & cr4_pinned_mask) ^ cr4_pinned_bits; 420 val = (val & ~cr4_pinned_mask) | cr4_pinned_bits; 421 goto set_register; 422 } 423 /* Warn after we've corrected the changed bits. */ 424 WARN_ONCE(bits_changed, "pinned CR4 bits changed: 0x%lx!?\n", 425 bits_changed); 426 } 427 } 428 #if IS_MODULE(CONFIG_LKDTM) 429 EXPORT_SYMBOL_GPL(native_write_cr4); 430 #endif 431 432 void cr4_update_irqsoff(unsigned long set, unsigned long clear) 433 { 434 unsigned long newval, cr4 = this_cpu_read(cpu_tlbstate.cr4); 435 436 lockdep_assert_irqs_disabled(); 437 438 newval = (cr4 & ~clear) | set; 439 if (newval != cr4) { 440 this_cpu_write(cpu_tlbstate.cr4, newval); 441 __write_cr4(newval); 442 } 443 } 444 EXPORT_SYMBOL(cr4_update_irqsoff); 445 446 /* Read the CR4 shadow. */ 447 unsigned long cr4_read_shadow(void) 448 { 449 return this_cpu_read(cpu_tlbstate.cr4); 450 } 451 EXPORT_SYMBOL_GPL(cr4_read_shadow); 452 453 void cr4_init(void) 454 { 455 unsigned long cr4 = __read_cr4(); 456 457 if (boot_cpu_has(X86_FEATURE_PCID)) 458 cr4 |= X86_CR4_PCIDE; 459 if (static_branch_likely(&cr_pinning)) 460 cr4 = (cr4 & ~cr4_pinned_mask) | cr4_pinned_bits; 461 462 __write_cr4(cr4); 463 464 /* Initialize cr4 shadow for this CPU. */ 465 this_cpu_write(cpu_tlbstate.cr4, cr4); 466 } 467 468 /* 469 * Once CPU feature detection is finished (and boot params have been 470 * parsed), record any of the sensitive CR bits that are set, and 471 * enable CR pinning. 472 */ 473 static void __init setup_cr_pinning(void) 474 { 475 cr4_pinned_bits = this_cpu_read(cpu_tlbstate.cr4) & cr4_pinned_mask; 476 static_key_enable(&cr_pinning.key); 477 } 478 479 static __init int x86_nofsgsbase_setup(char *arg) 480 { 481 /* Require an exact match without trailing characters. */ 482 if (strlen(arg)) 483 return 0; 484 485 /* Do not emit a message if the feature is not present. */ 486 if (!boot_cpu_has(X86_FEATURE_FSGSBASE)) 487 return 1; 488 489 setup_clear_cpu_cap(X86_FEATURE_FSGSBASE); 490 pr_info("FSGSBASE disabled via kernel command line\n"); 491 return 1; 492 } 493 __setup("nofsgsbase", x86_nofsgsbase_setup); 494 495 /* 496 * Protection Keys are not available in 32-bit mode. 497 */ 498 static bool pku_disabled; 499 500 static __always_inline void setup_pku(struct cpuinfo_x86 *c) 501 { 502 if (c == &boot_cpu_data) { 503 if (pku_disabled || !cpu_feature_enabled(X86_FEATURE_PKU)) 504 return; 505 /* 506 * Setting CR4.PKE will cause the X86_FEATURE_OSPKE cpuid 507 * bit to be set. Enforce it. 508 */ 509 setup_force_cpu_cap(X86_FEATURE_OSPKE); 510 511 } else if (!cpu_feature_enabled(X86_FEATURE_OSPKE)) { 512 return; 513 } 514 515 cr4_set_bits(X86_CR4_PKE); 516 /* Load the default PKRU value */ 517 pkru_write_default(); 518 } 519 520 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS 521 static __init int setup_disable_pku(char *arg) 522 { 523 /* 524 * Do not clear the X86_FEATURE_PKU bit. All of the 525 * runtime checks are against OSPKE so clearing the 526 * bit does nothing. 527 * 528 * This way, we will see "pku" in cpuinfo, but not 529 * "ospke", which is exactly what we want. It shows 530 * that the CPU has PKU, but the OS has not enabled it. 531 * This happens to be exactly how a system would look 532 * if we disabled the config option. 533 */ 534 pr_info("x86: 'nopku' specified, disabling Memory Protection Keys\n"); 535 pku_disabled = true; 536 return 1; 537 } 538 __setup("nopku", setup_disable_pku); 539 #endif 540 541 #ifdef CONFIG_X86_KERNEL_IBT 542 543 __noendbr u64 ibt_save(bool disable) 544 { 545 u64 msr = 0; 546 547 if (cpu_feature_enabled(X86_FEATURE_IBT)) { 548 rdmsrl(MSR_IA32_S_CET, msr); 549 if (disable) 550 wrmsrl(MSR_IA32_S_CET, msr & ~CET_ENDBR_EN); 551 } 552 553 return msr; 554 } 555 556 __noendbr void ibt_restore(u64 save) 557 { 558 u64 msr; 559 560 if (cpu_feature_enabled(X86_FEATURE_IBT)) { 561 rdmsrl(MSR_IA32_S_CET, msr); 562 msr &= ~CET_ENDBR_EN; 563 msr |= (save & CET_ENDBR_EN); 564 wrmsrl(MSR_IA32_S_CET, msr); 565 } 566 } 567 568 #endif 569 570 static __always_inline void setup_cet(struct cpuinfo_x86 *c) 571 { 572 bool user_shstk, kernel_ibt; 573 574 if (!IS_ENABLED(CONFIG_X86_CET)) 575 return; 576 577 kernel_ibt = HAS_KERNEL_IBT && cpu_feature_enabled(X86_FEATURE_IBT); 578 user_shstk = cpu_feature_enabled(X86_FEATURE_SHSTK) && 579 IS_ENABLED(CONFIG_X86_USER_SHADOW_STACK); 580 581 if (!kernel_ibt && !user_shstk) 582 return; 583 584 if (user_shstk) 585 set_cpu_cap(c, X86_FEATURE_USER_SHSTK); 586 587 if (kernel_ibt) 588 wrmsrl(MSR_IA32_S_CET, CET_ENDBR_EN); 589 else 590 wrmsrl(MSR_IA32_S_CET, 0); 591 592 cr4_set_bits(X86_CR4_CET); 593 594 if (kernel_ibt && ibt_selftest()) { 595 pr_err("IBT selftest: Failed!\n"); 596 wrmsrl(MSR_IA32_S_CET, 0); 597 setup_clear_cpu_cap(X86_FEATURE_IBT); 598 } 599 } 600 601 __noendbr void cet_disable(void) 602 { 603 if (!(cpu_feature_enabled(X86_FEATURE_IBT) || 604 cpu_feature_enabled(X86_FEATURE_SHSTK))) 605 return; 606 607 wrmsrl(MSR_IA32_S_CET, 0); 608 wrmsrl(MSR_IA32_U_CET, 0); 609 } 610 611 /* 612 * Some CPU features depend on higher CPUID levels, which may not always 613 * be available due to CPUID level capping or broken virtualization 614 * software. Add those features to this table to auto-disable them. 615 */ 616 struct cpuid_dependent_feature { 617 u32 feature; 618 u32 level; 619 }; 620 621 static const struct cpuid_dependent_feature 622 cpuid_dependent_features[] = { 623 { X86_FEATURE_MWAIT, 0x00000005 }, 624 { X86_FEATURE_DCA, 0x00000009 }, 625 { X86_FEATURE_XSAVE, 0x0000000d }, 626 { 0, 0 } 627 }; 628 629 static void filter_cpuid_features(struct cpuinfo_x86 *c, bool warn) 630 { 631 const struct cpuid_dependent_feature *df; 632 633 for (df = cpuid_dependent_features; df->feature; df++) { 634 635 if (!cpu_has(c, df->feature)) 636 continue; 637 /* 638 * Note: cpuid_level is set to -1 if unavailable, but 639 * extended_extended_level is set to 0 if unavailable 640 * and the legitimate extended levels are all negative 641 * when signed; hence the weird messing around with 642 * signs here... 643 */ 644 if (!((s32)df->level < 0 ? 645 (u32)df->level > (u32)c->extended_cpuid_level : 646 (s32)df->level > (s32)c->cpuid_level)) 647 continue; 648 649 clear_cpu_cap(c, df->feature); 650 if (!warn) 651 continue; 652 653 pr_warn("CPU: CPU feature " X86_CAP_FMT " disabled, no CPUID level 0x%x\n", 654 x86_cap_flag(df->feature), df->level); 655 } 656 } 657 658 /* 659 * Naming convention should be: <Name> [(<Codename>)] 660 * This table only is used unless init_<vendor>() below doesn't set it; 661 * in particular, if CPUID levels 0x80000002..4 are supported, this 662 * isn't used 663 */ 664 665 /* Look up CPU names by table lookup. */ 666 static const char *table_lookup_model(struct cpuinfo_x86 *c) 667 { 668 #ifdef CONFIG_X86_32 669 const struct legacy_cpu_model_info *info; 670 671 if (c->x86_model >= 16) 672 return NULL; /* Range check */ 673 674 if (!this_cpu) 675 return NULL; 676 677 info = this_cpu->legacy_models; 678 679 while (info->family) { 680 if (info->family == c->x86) 681 return info->model_names[c->x86_model]; 682 info++; 683 } 684 #endif 685 return NULL; /* Not found */ 686 } 687 688 /* Aligned to unsigned long to avoid split lock in atomic bitmap ops */ 689 __u32 cpu_caps_cleared[NCAPINTS + NBUGINTS] __aligned(sizeof(unsigned long)); 690 __u32 cpu_caps_set[NCAPINTS + NBUGINTS] __aligned(sizeof(unsigned long)); 691 692 #ifdef CONFIG_X86_32 693 /* The 32-bit entry code needs to find cpu_entry_area. */ 694 DEFINE_PER_CPU(struct cpu_entry_area *, cpu_entry_area); 695 #endif 696 697 /* Load the original GDT from the per-cpu structure */ 698 void load_direct_gdt(int cpu) 699 { 700 struct desc_ptr gdt_descr; 701 702 gdt_descr.address = (long)get_cpu_gdt_rw(cpu); 703 gdt_descr.size = GDT_SIZE - 1; 704 load_gdt(&gdt_descr); 705 } 706 EXPORT_SYMBOL_GPL(load_direct_gdt); 707 708 /* Load a fixmap remapping of the per-cpu GDT */ 709 void load_fixmap_gdt(int cpu) 710 { 711 struct desc_ptr gdt_descr; 712 713 gdt_descr.address = (long)get_cpu_gdt_ro(cpu); 714 gdt_descr.size = GDT_SIZE - 1; 715 load_gdt(&gdt_descr); 716 } 717 EXPORT_SYMBOL_GPL(load_fixmap_gdt); 718 719 /** 720 * switch_gdt_and_percpu_base - Switch to direct GDT and runtime per CPU base 721 * @cpu: The CPU number for which this is invoked 722 * 723 * Invoked during early boot to switch from early GDT and early per CPU to 724 * the direct GDT and the runtime per CPU area. On 32-bit the percpu base 725 * switch is implicit by loading the direct GDT. On 64bit this requires 726 * to update GSBASE. 727 */ 728 void __init switch_gdt_and_percpu_base(int cpu) 729 { 730 load_direct_gdt(cpu); 731 732 #ifdef CONFIG_X86_64 733 /* 734 * No need to load %gs. It is already correct. 735 * 736 * Writing %gs on 64bit would zero GSBASE which would make any per 737 * CPU operation up to the point of the wrmsrl() fault. 738 * 739 * Set GSBASE to the new offset. Until the wrmsrl() happens the 740 * early mapping is still valid. That means the GSBASE update will 741 * lose any prior per CPU data which was not copied over in 742 * setup_per_cpu_areas(). 743 * 744 * This works even with stackprotector enabled because the 745 * per CPU stack canary is 0 in both per CPU areas. 746 */ 747 wrmsrl(MSR_GS_BASE, cpu_kernelmode_gs_base(cpu)); 748 #else 749 /* 750 * %fs is already set to __KERNEL_PERCPU, but after switching GDT 751 * it is required to load FS again so that the 'hidden' part is 752 * updated from the new GDT. Up to this point the early per CPU 753 * translation is active. Any content of the early per CPU data 754 * which was not copied over in setup_per_cpu_areas() is lost. 755 */ 756 loadsegment(fs, __KERNEL_PERCPU); 757 #endif 758 } 759 760 static const struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {}; 761 762 static void get_model_name(struct cpuinfo_x86 *c) 763 { 764 unsigned int *v; 765 char *p, *q, *s; 766 767 if (c->extended_cpuid_level < 0x80000004) 768 return; 769 770 v = (unsigned int *)c->x86_model_id; 771 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]); 772 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]); 773 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]); 774 c->x86_model_id[48] = 0; 775 776 /* Trim whitespace */ 777 p = q = s = &c->x86_model_id[0]; 778 779 while (*p == ' ') 780 p++; 781 782 while (*p) { 783 /* Note the last non-whitespace index */ 784 if (!isspace(*p)) 785 s = q; 786 787 *q++ = *p++; 788 } 789 790 *(s + 1) = '\0'; 791 } 792 793 void detect_num_cpu_cores(struct cpuinfo_x86 *c) 794 { 795 unsigned int eax, ebx, ecx, edx; 796 797 c->x86_max_cores = 1; 798 if (!IS_ENABLED(CONFIG_SMP) || c->cpuid_level < 4) 799 return; 800 801 cpuid_count(4, 0, &eax, &ebx, &ecx, &edx); 802 if (eax & 0x1f) 803 c->x86_max_cores = (eax >> 26) + 1; 804 } 805 806 void cpu_detect_cache_sizes(struct cpuinfo_x86 *c) 807 { 808 unsigned int n, dummy, ebx, ecx, edx, l2size; 809 810 n = c->extended_cpuid_level; 811 812 if (n >= 0x80000005) { 813 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx); 814 c->x86_cache_size = (ecx>>24) + (edx>>24); 815 #ifdef CONFIG_X86_64 816 /* On K8 L1 TLB is inclusive, so don't count it */ 817 c->x86_tlbsize = 0; 818 #endif 819 } 820 821 if (n < 0x80000006) /* Some chips just has a large L1. */ 822 return; 823 824 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx); 825 l2size = ecx >> 16; 826 827 #ifdef CONFIG_X86_64 828 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff); 829 #else 830 /* do processor-specific cache resizing */ 831 if (this_cpu->legacy_cache_size) 832 l2size = this_cpu->legacy_cache_size(c, l2size); 833 834 /* Allow user to override all this if necessary. */ 835 if (cachesize_override != -1) 836 l2size = cachesize_override; 837 838 if (l2size == 0) 839 return; /* Again, no L2 cache is possible */ 840 #endif 841 842 c->x86_cache_size = l2size; 843 } 844 845 u16 __read_mostly tlb_lli_4k[NR_INFO]; 846 u16 __read_mostly tlb_lli_2m[NR_INFO]; 847 u16 __read_mostly tlb_lli_4m[NR_INFO]; 848 u16 __read_mostly tlb_lld_4k[NR_INFO]; 849 u16 __read_mostly tlb_lld_2m[NR_INFO]; 850 u16 __read_mostly tlb_lld_4m[NR_INFO]; 851 u16 __read_mostly tlb_lld_1g[NR_INFO]; 852 853 static void cpu_detect_tlb(struct cpuinfo_x86 *c) 854 { 855 if (this_cpu->c_detect_tlb) 856 this_cpu->c_detect_tlb(c); 857 858 pr_info("Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n", 859 tlb_lli_4k[ENTRIES], tlb_lli_2m[ENTRIES], 860 tlb_lli_4m[ENTRIES]); 861 862 pr_info("Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d, 1GB %d\n", 863 tlb_lld_4k[ENTRIES], tlb_lld_2m[ENTRIES], 864 tlb_lld_4m[ENTRIES], tlb_lld_1g[ENTRIES]); 865 } 866 867 int detect_ht_early(struct cpuinfo_x86 *c) 868 { 869 #ifdef CONFIG_SMP 870 u32 eax, ebx, ecx, edx; 871 872 if (!cpu_has(c, X86_FEATURE_HT)) 873 return -1; 874 875 if (cpu_has(c, X86_FEATURE_CMP_LEGACY)) 876 return -1; 877 878 if (cpu_has(c, X86_FEATURE_XTOPOLOGY)) 879 return -1; 880 881 cpuid(1, &eax, &ebx, &ecx, &edx); 882 883 smp_num_siblings = (ebx & 0xff0000) >> 16; 884 if (smp_num_siblings == 1) 885 pr_info_once("CPU0: Hyper-Threading is disabled\n"); 886 #endif 887 return 0; 888 } 889 890 void detect_ht(struct cpuinfo_x86 *c) 891 { 892 #ifdef CONFIG_SMP 893 int index_msb, core_bits; 894 895 if (detect_ht_early(c) < 0) 896 return; 897 898 index_msb = get_count_order(smp_num_siblings); 899 c->topo.pkg_id = apic->phys_pkg_id(c->topo.initial_apicid, index_msb); 900 901 smp_num_siblings = smp_num_siblings / c->x86_max_cores; 902 903 index_msb = get_count_order(smp_num_siblings); 904 905 core_bits = get_count_order(c->x86_max_cores); 906 907 c->topo.core_id = apic->phys_pkg_id(c->topo.initial_apicid, index_msb) & 908 ((1 << core_bits) - 1); 909 #endif 910 } 911 912 static void get_cpu_vendor(struct cpuinfo_x86 *c) 913 { 914 char *v = c->x86_vendor_id; 915 int i; 916 917 for (i = 0; i < X86_VENDOR_NUM; i++) { 918 if (!cpu_devs[i]) 919 break; 920 921 if (!strcmp(v, cpu_devs[i]->c_ident[0]) || 922 (cpu_devs[i]->c_ident[1] && 923 !strcmp(v, cpu_devs[i]->c_ident[1]))) { 924 925 this_cpu = cpu_devs[i]; 926 c->x86_vendor = this_cpu->c_x86_vendor; 927 return; 928 } 929 } 930 931 pr_err_once("CPU: vendor_id '%s' unknown, using generic init.\n" \ 932 "CPU: Your system may be unstable.\n", v); 933 934 c->x86_vendor = X86_VENDOR_UNKNOWN; 935 this_cpu = &default_cpu; 936 } 937 938 void cpu_detect(struct cpuinfo_x86 *c) 939 { 940 /* Get vendor name */ 941 cpuid(0x00000000, (unsigned int *)&c->cpuid_level, 942 (unsigned int *)&c->x86_vendor_id[0], 943 (unsigned int *)&c->x86_vendor_id[8], 944 (unsigned int *)&c->x86_vendor_id[4]); 945 946 c->x86 = 4; 947 /* Intel-defined flags: level 0x00000001 */ 948 if (c->cpuid_level >= 0x00000001) { 949 u32 junk, tfms, cap0, misc; 950 951 cpuid(0x00000001, &tfms, &misc, &junk, &cap0); 952 c->x86 = x86_family(tfms); 953 c->x86_model = x86_model(tfms); 954 c->x86_stepping = x86_stepping(tfms); 955 956 if (cap0 & (1<<19)) { 957 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8; 958 c->x86_cache_alignment = c->x86_clflush_size; 959 } 960 } 961 } 962 963 static void apply_forced_caps(struct cpuinfo_x86 *c) 964 { 965 int i; 966 967 for (i = 0; i < NCAPINTS + NBUGINTS; i++) { 968 c->x86_capability[i] &= ~cpu_caps_cleared[i]; 969 c->x86_capability[i] |= cpu_caps_set[i]; 970 } 971 } 972 973 static void init_speculation_control(struct cpuinfo_x86 *c) 974 { 975 /* 976 * The Intel SPEC_CTRL CPUID bit implies IBRS and IBPB support, 977 * and they also have a different bit for STIBP support. Also, 978 * a hypervisor might have set the individual AMD bits even on 979 * Intel CPUs, for finer-grained selection of what's available. 980 */ 981 if (cpu_has(c, X86_FEATURE_SPEC_CTRL)) { 982 set_cpu_cap(c, X86_FEATURE_IBRS); 983 set_cpu_cap(c, X86_FEATURE_IBPB); 984 set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL); 985 } 986 987 if (cpu_has(c, X86_FEATURE_INTEL_STIBP)) 988 set_cpu_cap(c, X86_FEATURE_STIBP); 989 990 if (cpu_has(c, X86_FEATURE_SPEC_CTRL_SSBD) || 991 cpu_has(c, X86_FEATURE_VIRT_SSBD)) 992 set_cpu_cap(c, X86_FEATURE_SSBD); 993 994 if (cpu_has(c, X86_FEATURE_AMD_IBRS)) { 995 set_cpu_cap(c, X86_FEATURE_IBRS); 996 set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL); 997 } 998 999 if (cpu_has(c, X86_FEATURE_AMD_IBPB)) 1000 set_cpu_cap(c, X86_FEATURE_IBPB); 1001 1002 if (cpu_has(c, X86_FEATURE_AMD_STIBP)) { 1003 set_cpu_cap(c, X86_FEATURE_STIBP); 1004 set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL); 1005 } 1006 1007 if (cpu_has(c, X86_FEATURE_AMD_SSBD)) { 1008 set_cpu_cap(c, X86_FEATURE_SSBD); 1009 set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL); 1010 clear_cpu_cap(c, X86_FEATURE_VIRT_SSBD); 1011 } 1012 } 1013 1014 void get_cpu_cap(struct cpuinfo_x86 *c) 1015 { 1016 u32 eax, ebx, ecx, edx; 1017 1018 /* Intel-defined flags: level 0x00000001 */ 1019 if (c->cpuid_level >= 0x00000001) { 1020 cpuid(0x00000001, &eax, &ebx, &ecx, &edx); 1021 1022 c->x86_capability[CPUID_1_ECX] = ecx; 1023 c->x86_capability[CPUID_1_EDX] = edx; 1024 } 1025 1026 /* Thermal and Power Management Leaf: level 0x00000006 (eax) */ 1027 if (c->cpuid_level >= 0x00000006) 1028 c->x86_capability[CPUID_6_EAX] = cpuid_eax(0x00000006); 1029 1030 /* Additional Intel-defined flags: level 0x00000007 */ 1031 if (c->cpuid_level >= 0x00000007) { 1032 cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx); 1033 c->x86_capability[CPUID_7_0_EBX] = ebx; 1034 c->x86_capability[CPUID_7_ECX] = ecx; 1035 c->x86_capability[CPUID_7_EDX] = edx; 1036 1037 /* Check valid sub-leaf index before accessing it */ 1038 if (eax >= 1) { 1039 cpuid_count(0x00000007, 1, &eax, &ebx, &ecx, &edx); 1040 c->x86_capability[CPUID_7_1_EAX] = eax; 1041 } 1042 } 1043 1044 /* Extended state features: level 0x0000000d */ 1045 if (c->cpuid_level >= 0x0000000d) { 1046 cpuid_count(0x0000000d, 1, &eax, &ebx, &ecx, &edx); 1047 1048 c->x86_capability[CPUID_D_1_EAX] = eax; 1049 } 1050 1051 /* AMD-defined flags: level 0x80000001 */ 1052 eax = cpuid_eax(0x80000000); 1053 c->extended_cpuid_level = eax; 1054 1055 if ((eax & 0xffff0000) == 0x80000000) { 1056 if (eax >= 0x80000001) { 1057 cpuid(0x80000001, &eax, &ebx, &ecx, &edx); 1058 1059 c->x86_capability[CPUID_8000_0001_ECX] = ecx; 1060 c->x86_capability[CPUID_8000_0001_EDX] = edx; 1061 } 1062 } 1063 1064 if (c->extended_cpuid_level >= 0x80000007) { 1065 cpuid(0x80000007, &eax, &ebx, &ecx, &edx); 1066 1067 c->x86_capability[CPUID_8000_0007_EBX] = ebx; 1068 c->x86_power = edx; 1069 } 1070 1071 if (c->extended_cpuid_level >= 0x80000008) { 1072 cpuid(0x80000008, &eax, &ebx, &ecx, &edx); 1073 c->x86_capability[CPUID_8000_0008_EBX] = ebx; 1074 } 1075 1076 if (c->extended_cpuid_level >= 0x8000000a) 1077 c->x86_capability[CPUID_8000_000A_EDX] = cpuid_edx(0x8000000a); 1078 1079 if (c->extended_cpuid_level >= 0x8000001f) 1080 c->x86_capability[CPUID_8000_001F_EAX] = cpuid_eax(0x8000001f); 1081 1082 if (c->extended_cpuid_level >= 0x80000021) 1083 c->x86_capability[CPUID_8000_0021_EAX] = cpuid_eax(0x80000021); 1084 1085 init_scattered_cpuid_features(c); 1086 init_speculation_control(c); 1087 1088 /* 1089 * Clear/Set all flags overridden by options, after probe. 1090 * This needs to happen each time we re-probe, which may happen 1091 * several times during CPU initialization. 1092 */ 1093 apply_forced_caps(c); 1094 } 1095 1096 void get_cpu_address_sizes(struct cpuinfo_x86 *c) 1097 { 1098 u32 eax, ebx, ecx, edx; 1099 bool vp_bits_from_cpuid = true; 1100 1101 if (!cpu_has(c, X86_FEATURE_CPUID) || 1102 (c->extended_cpuid_level < 0x80000008)) 1103 vp_bits_from_cpuid = false; 1104 1105 if (vp_bits_from_cpuid) { 1106 cpuid(0x80000008, &eax, &ebx, &ecx, &edx); 1107 1108 c->x86_virt_bits = (eax >> 8) & 0xff; 1109 c->x86_phys_bits = eax & 0xff; 1110 } else { 1111 if (IS_ENABLED(CONFIG_X86_64)) { 1112 c->x86_clflush_size = 64; 1113 c->x86_phys_bits = 36; 1114 c->x86_virt_bits = 48; 1115 } else { 1116 c->x86_clflush_size = 32; 1117 c->x86_virt_bits = 32; 1118 c->x86_phys_bits = 32; 1119 1120 if (cpu_has(c, X86_FEATURE_PAE) || 1121 cpu_has(c, X86_FEATURE_PSE36)) 1122 c->x86_phys_bits = 36; 1123 } 1124 } 1125 c->x86_cache_bits = c->x86_phys_bits; 1126 c->x86_cache_alignment = c->x86_clflush_size; 1127 } 1128 1129 static void identify_cpu_without_cpuid(struct cpuinfo_x86 *c) 1130 { 1131 #ifdef CONFIG_X86_32 1132 int i; 1133 1134 /* 1135 * First of all, decide if this is a 486 or higher 1136 * It's a 486 if we can modify the AC flag 1137 */ 1138 if (flag_is_changeable_p(X86_EFLAGS_AC)) 1139 c->x86 = 4; 1140 else 1141 c->x86 = 3; 1142 1143 for (i = 0; i < X86_VENDOR_NUM; i++) 1144 if (cpu_devs[i] && cpu_devs[i]->c_identify) { 1145 c->x86_vendor_id[0] = 0; 1146 cpu_devs[i]->c_identify(c); 1147 if (c->x86_vendor_id[0]) { 1148 get_cpu_vendor(c); 1149 break; 1150 } 1151 } 1152 #endif 1153 } 1154 1155 #define NO_SPECULATION BIT(0) 1156 #define NO_MELTDOWN BIT(1) 1157 #define NO_SSB BIT(2) 1158 #define NO_L1TF BIT(3) 1159 #define NO_MDS BIT(4) 1160 #define MSBDS_ONLY BIT(5) 1161 #define NO_SWAPGS BIT(6) 1162 #define NO_ITLB_MULTIHIT BIT(7) 1163 #define NO_SPECTRE_V2 BIT(8) 1164 #define NO_MMIO BIT(9) 1165 #define NO_EIBRS_PBRSB BIT(10) 1166 1167 #define VULNWL(vendor, family, model, whitelist) \ 1168 X86_MATCH_VENDOR_FAM_MODEL(vendor, family, model, whitelist) 1169 1170 #define VULNWL_INTEL(model, whitelist) \ 1171 VULNWL(INTEL, 6, INTEL_FAM6_##model, whitelist) 1172 1173 #define VULNWL_AMD(family, whitelist) \ 1174 VULNWL(AMD, family, X86_MODEL_ANY, whitelist) 1175 1176 #define VULNWL_HYGON(family, whitelist) \ 1177 VULNWL(HYGON, family, X86_MODEL_ANY, whitelist) 1178 1179 static const __initconst struct x86_cpu_id cpu_vuln_whitelist[] = { 1180 VULNWL(ANY, 4, X86_MODEL_ANY, NO_SPECULATION), 1181 VULNWL(CENTAUR, 5, X86_MODEL_ANY, NO_SPECULATION), 1182 VULNWL(INTEL, 5, X86_MODEL_ANY, NO_SPECULATION), 1183 VULNWL(NSC, 5, X86_MODEL_ANY, NO_SPECULATION), 1184 VULNWL(VORTEX, 5, X86_MODEL_ANY, NO_SPECULATION), 1185 VULNWL(VORTEX, 6, X86_MODEL_ANY, NO_SPECULATION), 1186 1187 /* Intel Family 6 */ 1188 VULNWL_INTEL(TIGERLAKE, NO_MMIO), 1189 VULNWL_INTEL(TIGERLAKE_L, NO_MMIO), 1190 VULNWL_INTEL(ALDERLAKE, NO_MMIO), 1191 VULNWL_INTEL(ALDERLAKE_L, NO_MMIO), 1192 1193 VULNWL_INTEL(ATOM_SALTWELL, NO_SPECULATION | NO_ITLB_MULTIHIT), 1194 VULNWL_INTEL(ATOM_SALTWELL_TABLET, NO_SPECULATION | NO_ITLB_MULTIHIT), 1195 VULNWL_INTEL(ATOM_SALTWELL_MID, NO_SPECULATION | NO_ITLB_MULTIHIT), 1196 VULNWL_INTEL(ATOM_BONNELL, NO_SPECULATION | NO_ITLB_MULTIHIT), 1197 VULNWL_INTEL(ATOM_BONNELL_MID, NO_SPECULATION | NO_ITLB_MULTIHIT), 1198 1199 VULNWL_INTEL(ATOM_SILVERMONT, NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT), 1200 VULNWL_INTEL(ATOM_SILVERMONT_D, NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT), 1201 VULNWL_INTEL(ATOM_SILVERMONT_MID, NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT), 1202 VULNWL_INTEL(ATOM_AIRMONT, NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT), 1203 VULNWL_INTEL(XEON_PHI_KNL, NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT), 1204 VULNWL_INTEL(XEON_PHI_KNM, NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT), 1205 1206 VULNWL_INTEL(CORE_YONAH, NO_SSB), 1207 1208 VULNWL_INTEL(ATOM_AIRMONT_MID, NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT), 1209 VULNWL_INTEL(ATOM_AIRMONT_NP, NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT), 1210 1211 VULNWL_INTEL(ATOM_GOLDMONT, NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO), 1212 VULNWL_INTEL(ATOM_GOLDMONT_D, NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO), 1213 VULNWL_INTEL(ATOM_GOLDMONT_PLUS, NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO | NO_EIBRS_PBRSB), 1214 1215 /* 1216 * Technically, swapgs isn't serializing on AMD (despite it previously 1217 * being documented as such in the APM). But according to AMD, %gs is 1218 * updated non-speculatively, and the issuing of %gs-relative memory 1219 * operands will be blocked until the %gs update completes, which is 1220 * good enough for our purposes. 1221 */ 1222 1223 VULNWL_INTEL(ATOM_TREMONT, NO_EIBRS_PBRSB), 1224 VULNWL_INTEL(ATOM_TREMONT_L, NO_EIBRS_PBRSB), 1225 VULNWL_INTEL(ATOM_TREMONT_D, NO_ITLB_MULTIHIT | NO_EIBRS_PBRSB), 1226 1227 /* AMD Family 0xf - 0x12 */ 1228 VULNWL_AMD(0x0f, NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO), 1229 VULNWL_AMD(0x10, NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO), 1230 VULNWL_AMD(0x11, NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO), 1231 VULNWL_AMD(0x12, NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO), 1232 1233 /* FAMILY_ANY must be last, otherwise 0x0f - 0x12 matches won't work */ 1234 VULNWL_AMD(X86_FAMILY_ANY, NO_MELTDOWN | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO | NO_EIBRS_PBRSB), 1235 VULNWL_HYGON(X86_FAMILY_ANY, NO_MELTDOWN | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO | NO_EIBRS_PBRSB), 1236 1237 /* Zhaoxin Family 7 */ 1238 VULNWL(CENTAUR, 7, X86_MODEL_ANY, NO_SPECTRE_V2 | NO_SWAPGS | NO_MMIO), 1239 VULNWL(ZHAOXIN, 7, X86_MODEL_ANY, NO_SPECTRE_V2 | NO_SWAPGS | NO_MMIO), 1240 {} 1241 }; 1242 1243 #define VULNBL(vendor, family, model, blacklist) \ 1244 X86_MATCH_VENDOR_FAM_MODEL(vendor, family, model, blacklist) 1245 1246 #define VULNBL_INTEL_STEPPINGS(model, steppings, issues) \ 1247 X86_MATCH_VENDOR_FAM_MODEL_STEPPINGS_FEATURE(INTEL, 6, \ 1248 INTEL_FAM6_##model, steppings, \ 1249 X86_FEATURE_ANY, issues) 1250 1251 #define VULNBL_AMD(family, blacklist) \ 1252 VULNBL(AMD, family, X86_MODEL_ANY, blacklist) 1253 1254 #define VULNBL_HYGON(family, blacklist) \ 1255 VULNBL(HYGON, family, X86_MODEL_ANY, blacklist) 1256 1257 #define SRBDS BIT(0) 1258 /* CPU is affected by X86_BUG_MMIO_STALE_DATA */ 1259 #define MMIO BIT(1) 1260 /* CPU is affected by Shared Buffers Data Sampling (SBDS), a variant of X86_BUG_MMIO_STALE_DATA */ 1261 #define MMIO_SBDS BIT(2) 1262 /* CPU is affected by RETbleed, speculating where you would not expect it */ 1263 #define RETBLEED BIT(3) 1264 /* CPU is affected by SMT (cross-thread) return predictions */ 1265 #define SMT_RSB BIT(4) 1266 /* CPU is affected by SRSO */ 1267 #define SRSO BIT(5) 1268 /* CPU is affected by GDS */ 1269 #define GDS BIT(6) 1270 1271 static const struct x86_cpu_id cpu_vuln_blacklist[] __initconst = { 1272 VULNBL_INTEL_STEPPINGS(IVYBRIDGE, X86_STEPPING_ANY, SRBDS), 1273 VULNBL_INTEL_STEPPINGS(HASWELL, X86_STEPPING_ANY, SRBDS), 1274 VULNBL_INTEL_STEPPINGS(HASWELL_L, X86_STEPPING_ANY, SRBDS), 1275 VULNBL_INTEL_STEPPINGS(HASWELL_G, X86_STEPPING_ANY, SRBDS), 1276 VULNBL_INTEL_STEPPINGS(HASWELL_X, X86_STEPPING_ANY, MMIO), 1277 VULNBL_INTEL_STEPPINGS(BROADWELL_D, X86_STEPPING_ANY, MMIO), 1278 VULNBL_INTEL_STEPPINGS(BROADWELL_G, X86_STEPPING_ANY, SRBDS), 1279 VULNBL_INTEL_STEPPINGS(BROADWELL_X, X86_STEPPING_ANY, MMIO), 1280 VULNBL_INTEL_STEPPINGS(BROADWELL, X86_STEPPING_ANY, SRBDS), 1281 VULNBL_INTEL_STEPPINGS(SKYLAKE_X, X86_STEPPING_ANY, MMIO | RETBLEED | GDS), 1282 VULNBL_INTEL_STEPPINGS(SKYLAKE_L, X86_STEPPING_ANY, MMIO | RETBLEED | GDS | SRBDS), 1283 VULNBL_INTEL_STEPPINGS(SKYLAKE, X86_STEPPING_ANY, MMIO | RETBLEED | GDS | SRBDS), 1284 VULNBL_INTEL_STEPPINGS(KABYLAKE_L, X86_STEPPING_ANY, MMIO | RETBLEED | GDS | SRBDS), 1285 VULNBL_INTEL_STEPPINGS(KABYLAKE, X86_STEPPING_ANY, MMIO | RETBLEED | GDS | SRBDS), 1286 VULNBL_INTEL_STEPPINGS(CANNONLAKE_L, X86_STEPPING_ANY, RETBLEED), 1287 VULNBL_INTEL_STEPPINGS(ICELAKE_L, X86_STEPPING_ANY, MMIO | MMIO_SBDS | RETBLEED | GDS), 1288 VULNBL_INTEL_STEPPINGS(ICELAKE_D, X86_STEPPING_ANY, MMIO | GDS), 1289 VULNBL_INTEL_STEPPINGS(ICELAKE_X, X86_STEPPING_ANY, MMIO | GDS), 1290 VULNBL_INTEL_STEPPINGS(COMETLAKE, X86_STEPPING_ANY, MMIO | MMIO_SBDS | RETBLEED | GDS), 1291 VULNBL_INTEL_STEPPINGS(COMETLAKE_L, X86_STEPPINGS(0x0, 0x0), MMIO | RETBLEED), 1292 VULNBL_INTEL_STEPPINGS(COMETLAKE_L, X86_STEPPING_ANY, MMIO | MMIO_SBDS | RETBLEED | GDS), 1293 VULNBL_INTEL_STEPPINGS(TIGERLAKE_L, X86_STEPPING_ANY, GDS), 1294 VULNBL_INTEL_STEPPINGS(TIGERLAKE, X86_STEPPING_ANY, GDS), 1295 VULNBL_INTEL_STEPPINGS(LAKEFIELD, X86_STEPPING_ANY, MMIO | MMIO_SBDS | RETBLEED), 1296 VULNBL_INTEL_STEPPINGS(ROCKETLAKE, X86_STEPPING_ANY, MMIO | RETBLEED | GDS), 1297 VULNBL_INTEL_STEPPINGS(ATOM_TREMONT, X86_STEPPING_ANY, MMIO | MMIO_SBDS), 1298 VULNBL_INTEL_STEPPINGS(ATOM_TREMONT_D, X86_STEPPING_ANY, MMIO), 1299 VULNBL_INTEL_STEPPINGS(ATOM_TREMONT_L, X86_STEPPING_ANY, MMIO | MMIO_SBDS), 1300 1301 VULNBL_AMD(0x15, RETBLEED), 1302 VULNBL_AMD(0x16, RETBLEED), 1303 VULNBL_AMD(0x17, RETBLEED | SMT_RSB | SRSO), 1304 VULNBL_HYGON(0x18, RETBLEED | SMT_RSB | SRSO), 1305 VULNBL_AMD(0x19, SRSO), 1306 {} 1307 }; 1308 1309 static bool __init cpu_matches(const struct x86_cpu_id *table, unsigned long which) 1310 { 1311 const struct x86_cpu_id *m = x86_match_cpu(table); 1312 1313 return m && !!(m->driver_data & which); 1314 } 1315 1316 u64 x86_read_arch_cap_msr(void) 1317 { 1318 u64 ia32_cap = 0; 1319 1320 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES)) 1321 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, ia32_cap); 1322 1323 return ia32_cap; 1324 } 1325 1326 static bool arch_cap_mmio_immune(u64 ia32_cap) 1327 { 1328 return (ia32_cap & ARCH_CAP_FBSDP_NO && 1329 ia32_cap & ARCH_CAP_PSDP_NO && 1330 ia32_cap & ARCH_CAP_SBDR_SSDP_NO); 1331 } 1332 1333 static void __init cpu_set_bug_bits(struct cpuinfo_x86 *c) 1334 { 1335 u64 ia32_cap = x86_read_arch_cap_msr(); 1336 1337 /* Set ITLB_MULTIHIT bug if cpu is not in the whitelist and not mitigated */ 1338 if (!cpu_matches(cpu_vuln_whitelist, NO_ITLB_MULTIHIT) && 1339 !(ia32_cap & ARCH_CAP_PSCHANGE_MC_NO)) 1340 setup_force_cpu_bug(X86_BUG_ITLB_MULTIHIT); 1341 1342 if (cpu_matches(cpu_vuln_whitelist, NO_SPECULATION)) 1343 return; 1344 1345 setup_force_cpu_bug(X86_BUG_SPECTRE_V1); 1346 1347 if (!cpu_matches(cpu_vuln_whitelist, NO_SPECTRE_V2)) 1348 setup_force_cpu_bug(X86_BUG_SPECTRE_V2); 1349 1350 if (!cpu_matches(cpu_vuln_whitelist, NO_SSB) && 1351 !(ia32_cap & ARCH_CAP_SSB_NO) && 1352 !cpu_has(c, X86_FEATURE_AMD_SSB_NO)) 1353 setup_force_cpu_bug(X86_BUG_SPEC_STORE_BYPASS); 1354 1355 /* 1356 * AMD's AutoIBRS is equivalent to Intel's eIBRS - use the Intel feature 1357 * flag and protect from vendor-specific bugs via the whitelist. 1358 * 1359 * Don't use AutoIBRS when SNP is enabled because it degrades host 1360 * userspace indirect branch performance. 1361 */ 1362 if ((ia32_cap & ARCH_CAP_IBRS_ALL) || 1363 (cpu_has(c, X86_FEATURE_AUTOIBRS) && 1364 !cpu_feature_enabled(X86_FEATURE_SEV_SNP))) { 1365 setup_force_cpu_cap(X86_FEATURE_IBRS_ENHANCED); 1366 if (!cpu_matches(cpu_vuln_whitelist, NO_EIBRS_PBRSB) && 1367 !(ia32_cap & ARCH_CAP_PBRSB_NO)) 1368 setup_force_cpu_bug(X86_BUG_EIBRS_PBRSB); 1369 } 1370 1371 if (!cpu_matches(cpu_vuln_whitelist, NO_MDS) && 1372 !(ia32_cap & ARCH_CAP_MDS_NO)) { 1373 setup_force_cpu_bug(X86_BUG_MDS); 1374 if (cpu_matches(cpu_vuln_whitelist, MSBDS_ONLY)) 1375 setup_force_cpu_bug(X86_BUG_MSBDS_ONLY); 1376 } 1377 1378 if (!cpu_matches(cpu_vuln_whitelist, NO_SWAPGS)) 1379 setup_force_cpu_bug(X86_BUG_SWAPGS); 1380 1381 /* 1382 * When the CPU is not mitigated for TAA (TAA_NO=0) set TAA bug when: 1383 * - TSX is supported or 1384 * - TSX_CTRL is present 1385 * 1386 * TSX_CTRL check is needed for cases when TSX could be disabled before 1387 * the kernel boot e.g. kexec. 1388 * TSX_CTRL check alone is not sufficient for cases when the microcode 1389 * update is not present or running as guest that don't get TSX_CTRL. 1390 */ 1391 if (!(ia32_cap & ARCH_CAP_TAA_NO) && 1392 (cpu_has(c, X86_FEATURE_RTM) || 1393 (ia32_cap & ARCH_CAP_TSX_CTRL_MSR))) 1394 setup_force_cpu_bug(X86_BUG_TAA); 1395 1396 /* 1397 * SRBDS affects CPUs which support RDRAND or RDSEED and are listed 1398 * in the vulnerability blacklist. 1399 * 1400 * Some of the implications and mitigation of Shared Buffers Data 1401 * Sampling (SBDS) are similar to SRBDS. Give SBDS same treatment as 1402 * SRBDS. 1403 */ 1404 if ((cpu_has(c, X86_FEATURE_RDRAND) || 1405 cpu_has(c, X86_FEATURE_RDSEED)) && 1406 cpu_matches(cpu_vuln_blacklist, SRBDS | MMIO_SBDS)) 1407 setup_force_cpu_bug(X86_BUG_SRBDS); 1408 1409 /* 1410 * Processor MMIO Stale Data bug enumeration 1411 * 1412 * Affected CPU list is generally enough to enumerate the vulnerability, 1413 * but for virtualization case check for ARCH_CAP MSR bits also, VMM may 1414 * not want the guest to enumerate the bug. 1415 * 1416 * Set X86_BUG_MMIO_UNKNOWN for CPUs that are neither in the blacklist, 1417 * nor in the whitelist and also don't enumerate MSR ARCH_CAP MMIO bits. 1418 */ 1419 if (!arch_cap_mmio_immune(ia32_cap)) { 1420 if (cpu_matches(cpu_vuln_blacklist, MMIO)) 1421 setup_force_cpu_bug(X86_BUG_MMIO_STALE_DATA); 1422 else if (!cpu_matches(cpu_vuln_whitelist, NO_MMIO)) 1423 setup_force_cpu_bug(X86_BUG_MMIO_UNKNOWN); 1424 } 1425 1426 if (!cpu_has(c, X86_FEATURE_BTC_NO)) { 1427 if (cpu_matches(cpu_vuln_blacklist, RETBLEED) || (ia32_cap & ARCH_CAP_RSBA)) 1428 setup_force_cpu_bug(X86_BUG_RETBLEED); 1429 } 1430 1431 if (cpu_matches(cpu_vuln_blacklist, SMT_RSB)) 1432 setup_force_cpu_bug(X86_BUG_SMT_RSB); 1433 1434 if (!cpu_has(c, X86_FEATURE_SRSO_NO)) { 1435 if (cpu_matches(cpu_vuln_blacklist, SRSO)) 1436 setup_force_cpu_bug(X86_BUG_SRSO); 1437 } 1438 1439 /* 1440 * Check if CPU is vulnerable to GDS. If running in a virtual machine on 1441 * an affected processor, the VMM may have disabled the use of GATHER by 1442 * disabling AVX2. The only way to do this in HW is to clear XCR0[2], 1443 * which means that AVX will be disabled. 1444 */ 1445 if (cpu_matches(cpu_vuln_blacklist, GDS) && !(ia32_cap & ARCH_CAP_GDS_NO) && 1446 boot_cpu_has(X86_FEATURE_AVX)) 1447 setup_force_cpu_bug(X86_BUG_GDS); 1448 1449 if (cpu_matches(cpu_vuln_whitelist, NO_MELTDOWN)) 1450 return; 1451 1452 /* Rogue Data Cache Load? No! */ 1453 if (ia32_cap & ARCH_CAP_RDCL_NO) 1454 return; 1455 1456 setup_force_cpu_bug(X86_BUG_CPU_MELTDOWN); 1457 1458 if (cpu_matches(cpu_vuln_whitelist, NO_L1TF)) 1459 return; 1460 1461 setup_force_cpu_bug(X86_BUG_L1TF); 1462 } 1463 1464 /* 1465 * The NOPL instruction is supposed to exist on all CPUs of family >= 6; 1466 * unfortunately, that's not true in practice because of early VIA 1467 * chips and (more importantly) broken virtualizers that are not easy 1468 * to detect. In the latter case it doesn't even *fail* reliably, so 1469 * probing for it doesn't even work. Disable it completely on 32-bit 1470 * unless we can find a reliable way to detect all the broken cases. 1471 * Enable it explicitly on 64-bit for non-constant inputs of cpu_has(). 1472 */ 1473 static void detect_nopl(void) 1474 { 1475 #ifdef CONFIG_X86_32 1476 setup_clear_cpu_cap(X86_FEATURE_NOPL); 1477 #else 1478 setup_force_cpu_cap(X86_FEATURE_NOPL); 1479 #endif 1480 } 1481 1482 /* 1483 * We parse cpu parameters early because fpu__init_system() is executed 1484 * before parse_early_param(). 1485 */ 1486 static void __init cpu_parse_early_param(void) 1487 { 1488 char arg[128]; 1489 char *argptr = arg, *opt; 1490 int arglen, taint = 0; 1491 1492 #ifdef CONFIG_X86_32 1493 if (cmdline_find_option_bool(boot_command_line, "no387")) 1494 #ifdef CONFIG_MATH_EMULATION 1495 setup_clear_cpu_cap(X86_FEATURE_FPU); 1496 #else 1497 pr_err("Option 'no387' required CONFIG_MATH_EMULATION enabled.\n"); 1498 #endif 1499 1500 if (cmdline_find_option_bool(boot_command_line, "nofxsr")) 1501 setup_clear_cpu_cap(X86_FEATURE_FXSR); 1502 #endif 1503 1504 if (cmdline_find_option_bool(boot_command_line, "noxsave")) 1505 setup_clear_cpu_cap(X86_FEATURE_XSAVE); 1506 1507 if (cmdline_find_option_bool(boot_command_line, "noxsaveopt")) 1508 setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT); 1509 1510 if (cmdline_find_option_bool(boot_command_line, "noxsaves")) 1511 setup_clear_cpu_cap(X86_FEATURE_XSAVES); 1512 1513 if (cmdline_find_option_bool(boot_command_line, "nousershstk")) 1514 setup_clear_cpu_cap(X86_FEATURE_USER_SHSTK); 1515 1516 arglen = cmdline_find_option(boot_command_line, "clearcpuid", arg, sizeof(arg)); 1517 if (arglen <= 0) 1518 return; 1519 1520 pr_info("Clearing CPUID bits:"); 1521 1522 while (argptr) { 1523 bool found __maybe_unused = false; 1524 unsigned int bit; 1525 1526 opt = strsep(&argptr, ","); 1527 1528 /* 1529 * Handle naked numbers first for feature flags which don't 1530 * have names. 1531 */ 1532 if (!kstrtouint(opt, 10, &bit)) { 1533 if (bit < NCAPINTS * 32) { 1534 1535 /* empty-string, i.e., ""-defined feature flags */ 1536 if (!x86_cap_flags[bit]) 1537 pr_cont(" " X86_CAP_FMT_NUM, x86_cap_flag_num(bit)); 1538 else 1539 pr_cont(" " X86_CAP_FMT, x86_cap_flag(bit)); 1540 1541 setup_clear_cpu_cap(bit); 1542 taint++; 1543 } 1544 /* 1545 * The assumption is that there are no feature names with only 1546 * numbers in the name thus go to the next argument. 1547 */ 1548 continue; 1549 } 1550 1551 for (bit = 0; bit < 32 * NCAPINTS; bit++) { 1552 if (!x86_cap_flag(bit)) 1553 continue; 1554 1555 if (strcmp(x86_cap_flag(bit), opt)) 1556 continue; 1557 1558 pr_cont(" %s", opt); 1559 setup_clear_cpu_cap(bit); 1560 taint++; 1561 found = true; 1562 break; 1563 } 1564 1565 if (!found) 1566 pr_cont(" (unknown: %s)", opt); 1567 } 1568 pr_cont("\n"); 1569 1570 if (taint) 1571 add_taint(TAINT_CPU_OUT_OF_SPEC, LOCKDEP_STILL_OK); 1572 } 1573 1574 /* 1575 * Do minimum CPU detection early. 1576 * Fields really needed: vendor, cpuid_level, family, model, mask, 1577 * cache alignment. 1578 * The others are not touched to avoid unwanted side effects. 1579 * 1580 * WARNING: this function is only called on the boot CPU. Don't add code 1581 * here that is supposed to run on all CPUs. 1582 */ 1583 static void __init early_identify_cpu(struct cpuinfo_x86 *c) 1584 { 1585 memset(&c->x86_capability, 0, sizeof(c->x86_capability)); 1586 c->extended_cpuid_level = 0; 1587 1588 if (!have_cpuid_p()) 1589 identify_cpu_without_cpuid(c); 1590 1591 /* cyrix could have cpuid enabled via c_identify()*/ 1592 if (have_cpuid_p()) { 1593 cpu_detect(c); 1594 get_cpu_vendor(c); 1595 get_cpu_cap(c); 1596 setup_force_cpu_cap(X86_FEATURE_CPUID); 1597 cpu_parse_early_param(); 1598 1599 if (this_cpu->c_early_init) 1600 this_cpu->c_early_init(c); 1601 1602 c->cpu_index = 0; 1603 filter_cpuid_features(c, false); 1604 1605 if (this_cpu->c_bsp_init) 1606 this_cpu->c_bsp_init(c); 1607 } else { 1608 setup_clear_cpu_cap(X86_FEATURE_CPUID); 1609 } 1610 1611 get_cpu_address_sizes(c); 1612 1613 setup_force_cpu_cap(X86_FEATURE_ALWAYS); 1614 1615 cpu_set_bug_bits(c); 1616 1617 sld_setup(c); 1618 1619 #ifdef CONFIG_X86_32 1620 /* 1621 * Regardless of whether PCID is enumerated, the SDM says 1622 * that it can't be enabled in 32-bit mode. 1623 */ 1624 setup_clear_cpu_cap(X86_FEATURE_PCID); 1625 #endif 1626 1627 /* 1628 * Later in the boot process pgtable_l5_enabled() relies on 1629 * cpu_feature_enabled(X86_FEATURE_LA57). If 5-level paging is not 1630 * enabled by this point we need to clear the feature bit to avoid 1631 * false-positives at the later stage. 1632 * 1633 * pgtable_l5_enabled() can be false here for several reasons: 1634 * - 5-level paging is disabled compile-time; 1635 * - it's 32-bit kernel; 1636 * - machine doesn't support 5-level paging; 1637 * - user specified 'no5lvl' in kernel command line. 1638 */ 1639 if (!pgtable_l5_enabled()) 1640 setup_clear_cpu_cap(X86_FEATURE_LA57); 1641 1642 detect_nopl(); 1643 } 1644 1645 void __init early_cpu_init(void) 1646 { 1647 const struct cpu_dev *const *cdev; 1648 int count = 0; 1649 1650 #ifdef CONFIG_PROCESSOR_SELECT 1651 pr_info("KERNEL supported cpus:\n"); 1652 #endif 1653 1654 for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) { 1655 const struct cpu_dev *cpudev = *cdev; 1656 1657 if (count >= X86_VENDOR_NUM) 1658 break; 1659 cpu_devs[count] = cpudev; 1660 count++; 1661 1662 #ifdef CONFIG_PROCESSOR_SELECT 1663 { 1664 unsigned int j; 1665 1666 for (j = 0; j < 2; j++) { 1667 if (!cpudev->c_ident[j]) 1668 continue; 1669 pr_info(" %s %s\n", cpudev->c_vendor, 1670 cpudev->c_ident[j]); 1671 } 1672 } 1673 #endif 1674 } 1675 early_identify_cpu(&boot_cpu_data); 1676 } 1677 1678 static bool detect_null_seg_behavior(void) 1679 { 1680 /* 1681 * Empirically, writing zero to a segment selector on AMD does 1682 * not clear the base, whereas writing zero to a segment 1683 * selector on Intel does clear the base. Intel's behavior 1684 * allows slightly faster context switches in the common case 1685 * where GS is unused by the prev and next threads. 1686 * 1687 * Since neither vendor documents this anywhere that I can see, 1688 * detect it directly instead of hard-coding the choice by 1689 * vendor. 1690 * 1691 * I've designated AMD's behavior as the "bug" because it's 1692 * counterintuitive and less friendly. 1693 */ 1694 1695 unsigned long old_base, tmp; 1696 rdmsrl(MSR_FS_BASE, old_base); 1697 wrmsrl(MSR_FS_BASE, 1); 1698 loadsegment(fs, 0); 1699 rdmsrl(MSR_FS_BASE, tmp); 1700 wrmsrl(MSR_FS_BASE, old_base); 1701 return tmp == 0; 1702 } 1703 1704 void check_null_seg_clears_base(struct cpuinfo_x86 *c) 1705 { 1706 /* BUG_NULL_SEG is only relevant with 64bit userspace */ 1707 if (!IS_ENABLED(CONFIG_X86_64)) 1708 return; 1709 1710 if (cpu_has(c, X86_FEATURE_NULL_SEL_CLR_BASE)) 1711 return; 1712 1713 /* 1714 * CPUID bit above wasn't set. If this kernel is still running 1715 * as a HV guest, then the HV has decided not to advertize 1716 * that CPUID bit for whatever reason. For example, one 1717 * member of the migration pool might be vulnerable. Which 1718 * means, the bug is present: set the BUG flag and return. 1719 */ 1720 if (cpu_has(c, X86_FEATURE_HYPERVISOR)) { 1721 set_cpu_bug(c, X86_BUG_NULL_SEG); 1722 return; 1723 } 1724 1725 /* 1726 * Zen2 CPUs also have this behaviour, but no CPUID bit. 1727 * 0x18 is the respective family for Hygon. 1728 */ 1729 if ((c->x86 == 0x17 || c->x86 == 0x18) && 1730 detect_null_seg_behavior()) 1731 return; 1732 1733 /* All the remaining ones are affected */ 1734 set_cpu_bug(c, X86_BUG_NULL_SEG); 1735 } 1736 1737 static void generic_identify(struct cpuinfo_x86 *c) 1738 { 1739 c->extended_cpuid_level = 0; 1740 1741 if (!have_cpuid_p()) 1742 identify_cpu_without_cpuid(c); 1743 1744 /* cyrix could have cpuid enabled via c_identify()*/ 1745 if (!have_cpuid_p()) 1746 return; 1747 1748 cpu_detect(c); 1749 1750 get_cpu_vendor(c); 1751 1752 get_cpu_cap(c); 1753 1754 get_cpu_address_sizes(c); 1755 1756 if (c->cpuid_level >= 0x00000001) { 1757 c->topo.initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF; 1758 #ifdef CONFIG_X86_32 1759 # ifdef CONFIG_SMP 1760 c->topo.apicid = apic->phys_pkg_id(c->topo.initial_apicid, 0); 1761 # else 1762 c->topo.apicid = c->topo.initial_apicid; 1763 # endif 1764 #endif 1765 c->topo.pkg_id = c->topo.initial_apicid; 1766 } 1767 1768 get_model_name(c); /* Default name */ 1769 1770 /* 1771 * ESPFIX is a strange bug. All real CPUs have it. Paravirt 1772 * systems that run Linux at CPL > 0 may or may not have the 1773 * issue, but, even if they have the issue, there's absolutely 1774 * nothing we can do about it because we can't use the real IRET 1775 * instruction. 1776 * 1777 * NB: For the time being, only 32-bit kernels support 1778 * X86_BUG_ESPFIX as such. 64-bit kernels directly choose 1779 * whether to apply espfix using paravirt hooks. If any 1780 * non-paravirt system ever shows up that does *not* have the 1781 * ESPFIX issue, we can change this. 1782 */ 1783 #ifdef CONFIG_X86_32 1784 set_cpu_bug(c, X86_BUG_ESPFIX); 1785 #endif 1786 } 1787 1788 /* 1789 * Validate that ACPI/mptables have the same information about the 1790 * effective APIC id and update the package map. 1791 */ 1792 static void validate_apic_and_package_id(struct cpuinfo_x86 *c) 1793 { 1794 #ifdef CONFIG_SMP 1795 unsigned int cpu = smp_processor_id(); 1796 u32 apicid; 1797 1798 apicid = apic->cpu_present_to_apicid(cpu); 1799 1800 if (apicid != c->topo.apicid) { 1801 pr_err(FW_BUG "CPU%u: APIC id mismatch. Firmware: %x APIC: %x\n", 1802 cpu, apicid, c->topo.initial_apicid); 1803 } 1804 BUG_ON(topology_update_package_map(c->topo.pkg_id, cpu)); 1805 BUG_ON(topology_update_die_map(c->topo.die_id, cpu)); 1806 #else 1807 c->topo.logical_pkg_id = 0; 1808 #endif 1809 } 1810 1811 /* 1812 * This does the hard work of actually picking apart the CPU stuff... 1813 */ 1814 static void identify_cpu(struct cpuinfo_x86 *c) 1815 { 1816 int i; 1817 1818 c->loops_per_jiffy = loops_per_jiffy; 1819 c->x86_cache_size = 0; 1820 c->x86_vendor = X86_VENDOR_UNKNOWN; 1821 c->x86_model = c->x86_stepping = 0; /* So far unknown... */ 1822 c->x86_vendor_id[0] = '\0'; /* Unset */ 1823 c->x86_model_id[0] = '\0'; /* Unset */ 1824 c->x86_max_cores = 1; 1825 c->x86_coreid_bits = 0; 1826 c->topo.cu_id = 0xff; 1827 c->topo.llc_id = BAD_APICID; 1828 c->topo.l2c_id = BAD_APICID; 1829 #ifdef CONFIG_X86_64 1830 c->x86_clflush_size = 64; 1831 c->x86_phys_bits = 36; 1832 c->x86_virt_bits = 48; 1833 #else 1834 c->cpuid_level = -1; /* CPUID not detected */ 1835 c->x86_clflush_size = 32; 1836 c->x86_phys_bits = 32; 1837 c->x86_virt_bits = 32; 1838 #endif 1839 c->x86_cache_alignment = c->x86_clflush_size; 1840 memset(&c->x86_capability, 0, sizeof(c->x86_capability)); 1841 #ifdef CONFIG_X86_VMX_FEATURE_NAMES 1842 memset(&c->vmx_capability, 0, sizeof(c->vmx_capability)); 1843 #endif 1844 1845 generic_identify(c); 1846 1847 if (this_cpu->c_identify) 1848 this_cpu->c_identify(c); 1849 1850 /* Clear/Set all flags overridden by options, after probe */ 1851 apply_forced_caps(c); 1852 1853 #ifdef CONFIG_X86_64 1854 c->topo.apicid = apic->phys_pkg_id(c->topo.initial_apicid, 0); 1855 #endif 1856 1857 1858 /* 1859 * Set default APIC and TSC_DEADLINE MSR fencing flag. AMD and 1860 * Hygon will clear it in ->c_init() below. 1861 */ 1862 set_cpu_cap(c, X86_FEATURE_APIC_MSRS_FENCE); 1863 1864 /* 1865 * Vendor-specific initialization. In this section we 1866 * canonicalize the feature flags, meaning if there are 1867 * features a certain CPU supports which CPUID doesn't 1868 * tell us, CPUID claiming incorrect flags, or other bugs, 1869 * we handle them here. 1870 * 1871 * At the end of this section, c->x86_capability better 1872 * indicate the features this CPU genuinely supports! 1873 */ 1874 if (this_cpu->c_init) 1875 this_cpu->c_init(c); 1876 1877 /* Disable the PN if appropriate */ 1878 squash_the_stupid_serial_number(c); 1879 1880 /* Set up SMEP/SMAP/UMIP */ 1881 setup_smep(c); 1882 setup_smap(c); 1883 setup_umip(c); 1884 1885 /* Enable FSGSBASE instructions if available. */ 1886 if (cpu_has(c, X86_FEATURE_FSGSBASE)) { 1887 cr4_set_bits(X86_CR4_FSGSBASE); 1888 elf_hwcap2 |= HWCAP2_FSGSBASE; 1889 } 1890 1891 /* 1892 * The vendor-specific functions might have changed features. 1893 * Now we do "generic changes." 1894 */ 1895 1896 /* Filter out anything that depends on CPUID levels we don't have */ 1897 filter_cpuid_features(c, true); 1898 1899 /* If the model name is still unset, do table lookup. */ 1900 if (!c->x86_model_id[0]) { 1901 const char *p; 1902 p = table_lookup_model(c); 1903 if (p) 1904 strcpy(c->x86_model_id, p); 1905 else 1906 /* Last resort... */ 1907 sprintf(c->x86_model_id, "%02x/%02x", 1908 c->x86, c->x86_model); 1909 } 1910 1911 #ifdef CONFIG_X86_64 1912 detect_ht(c); 1913 #endif 1914 1915 x86_init_rdrand(c); 1916 setup_pku(c); 1917 setup_cet(c); 1918 1919 /* 1920 * Clear/Set all flags overridden by options, need do it 1921 * before following smp all cpus cap AND. 1922 */ 1923 apply_forced_caps(c); 1924 1925 /* 1926 * On SMP, boot_cpu_data holds the common feature set between 1927 * all CPUs; so make sure that we indicate which features are 1928 * common between the CPUs. The first time this routine gets 1929 * executed, c == &boot_cpu_data. 1930 */ 1931 if (c != &boot_cpu_data) { 1932 /* AND the already accumulated flags with these */ 1933 for (i = 0; i < NCAPINTS; i++) 1934 boot_cpu_data.x86_capability[i] &= c->x86_capability[i]; 1935 1936 /* OR, i.e. replicate the bug flags */ 1937 for (i = NCAPINTS; i < NCAPINTS + NBUGINTS; i++) 1938 c->x86_capability[i] |= boot_cpu_data.x86_capability[i]; 1939 } 1940 1941 ppin_init(c); 1942 1943 /* Init Machine Check Exception if available. */ 1944 mcheck_cpu_init(c); 1945 1946 select_idle_routine(c); 1947 1948 #ifdef CONFIG_NUMA 1949 numa_add_cpu(smp_processor_id()); 1950 #endif 1951 } 1952 1953 /* 1954 * Set up the CPU state needed to execute SYSENTER/SYSEXIT instructions 1955 * on 32-bit kernels: 1956 */ 1957 #ifdef CONFIG_X86_32 1958 void enable_sep_cpu(void) 1959 { 1960 struct tss_struct *tss; 1961 int cpu; 1962 1963 if (!boot_cpu_has(X86_FEATURE_SEP)) 1964 return; 1965 1966 cpu = get_cpu(); 1967 tss = &per_cpu(cpu_tss_rw, cpu); 1968 1969 /* 1970 * We cache MSR_IA32_SYSENTER_CS's value in the TSS's ss1 field -- 1971 * see the big comment in struct x86_hw_tss's definition. 1972 */ 1973 1974 tss->x86_tss.ss1 = __KERNEL_CS; 1975 wrmsr(MSR_IA32_SYSENTER_CS, tss->x86_tss.ss1, 0); 1976 wrmsr(MSR_IA32_SYSENTER_ESP, (unsigned long)(cpu_entry_stack(cpu) + 1), 0); 1977 wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long)entry_SYSENTER_32, 0); 1978 1979 put_cpu(); 1980 } 1981 #endif 1982 1983 static __init void identify_boot_cpu(void) 1984 { 1985 identify_cpu(&boot_cpu_data); 1986 if (HAS_KERNEL_IBT && cpu_feature_enabled(X86_FEATURE_IBT)) 1987 pr_info("CET detected: Indirect Branch Tracking enabled\n"); 1988 #ifdef CONFIG_X86_32 1989 enable_sep_cpu(); 1990 #endif 1991 cpu_detect_tlb(&boot_cpu_data); 1992 setup_cr_pinning(); 1993 1994 tsx_init(); 1995 tdx_init(); 1996 lkgs_init(); 1997 } 1998 1999 void identify_secondary_cpu(struct cpuinfo_x86 *c) 2000 { 2001 BUG_ON(c == &boot_cpu_data); 2002 identify_cpu(c); 2003 #ifdef CONFIG_X86_32 2004 enable_sep_cpu(); 2005 #endif 2006 validate_apic_and_package_id(c); 2007 x86_spec_ctrl_setup_ap(); 2008 update_srbds_msr(); 2009 if (boot_cpu_has_bug(X86_BUG_GDS)) 2010 update_gds_msr(); 2011 2012 tsx_ap_init(); 2013 } 2014 2015 void print_cpu_info(struct cpuinfo_x86 *c) 2016 { 2017 const char *vendor = NULL; 2018 2019 if (c->x86_vendor < X86_VENDOR_NUM) { 2020 vendor = this_cpu->c_vendor; 2021 } else { 2022 if (c->cpuid_level >= 0) 2023 vendor = c->x86_vendor_id; 2024 } 2025 2026 if (vendor && !strstr(c->x86_model_id, vendor)) 2027 pr_cont("%s ", vendor); 2028 2029 if (c->x86_model_id[0]) 2030 pr_cont("%s", c->x86_model_id); 2031 else 2032 pr_cont("%d86", c->x86); 2033 2034 pr_cont(" (family: 0x%x, model: 0x%x", c->x86, c->x86_model); 2035 2036 if (c->x86_stepping || c->cpuid_level >= 0) 2037 pr_cont(", stepping: 0x%x)\n", c->x86_stepping); 2038 else 2039 pr_cont(")\n"); 2040 } 2041 2042 /* 2043 * clearcpuid= was already parsed in cpu_parse_early_param(). This dummy 2044 * function prevents it from becoming an environment variable for init. 2045 */ 2046 static __init int setup_clearcpuid(char *arg) 2047 { 2048 return 1; 2049 } 2050 __setup("clearcpuid=", setup_clearcpuid); 2051 2052 DEFINE_PER_CPU_ALIGNED(struct pcpu_hot, pcpu_hot) = { 2053 .current_task = &init_task, 2054 .preempt_count = INIT_PREEMPT_COUNT, 2055 .top_of_stack = TOP_OF_INIT_STACK, 2056 }; 2057 EXPORT_PER_CPU_SYMBOL(pcpu_hot); 2058 2059 #ifdef CONFIG_X86_64 2060 DEFINE_PER_CPU_FIRST(struct fixed_percpu_data, 2061 fixed_percpu_data) __aligned(PAGE_SIZE) __visible; 2062 EXPORT_PER_CPU_SYMBOL_GPL(fixed_percpu_data); 2063 2064 static void wrmsrl_cstar(unsigned long val) 2065 { 2066 /* 2067 * Intel CPUs do not support 32-bit SYSCALL. Writing to MSR_CSTAR 2068 * is so far ignored by the CPU, but raises a #VE trap in a TDX 2069 * guest. Avoid the pointless write on all Intel CPUs. 2070 */ 2071 if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL) 2072 wrmsrl(MSR_CSTAR, val); 2073 } 2074 2075 /* May not be marked __init: used by software suspend */ 2076 void syscall_init(void) 2077 { 2078 wrmsr(MSR_STAR, 0, (__USER32_CS << 16) | __KERNEL_CS); 2079 wrmsrl(MSR_LSTAR, (unsigned long)entry_SYSCALL_64); 2080 2081 if (ia32_enabled()) { 2082 wrmsrl_cstar((unsigned long)entry_SYSCALL_compat); 2083 /* 2084 * This only works on Intel CPUs. 2085 * On AMD CPUs these MSRs are 32-bit, CPU truncates MSR_IA32_SYSENTER_EIP. 2086 * This does not cause SYSENTER to jump to the wrong location, because 2087 * AMD doesn't allow SYSENTER in long mode (either 32- or 64-bit). 2088 */ 2089 wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS); 2090 wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 2091 (unsigned long)(cpu_entry_stack(smp_processor_id()) + 1)); 2092 wrmsrl_safe(MSR_IA32_SYSENTER_EIP, (u64)entry_SYSENTER_compat); 2093 } else { 2094 wrmsrl_cstar((unsigned long)entry_SYSCALL32_ignore); 2095 wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)GDT_ENTRY_INVALID_SEG); 2096 wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL); 2097 wrmsrl_safe(MSR_IA32_SYSENTER_EIP, 0ULL); 2098 } 2099 2100 /* 2101 * Flags to clear on syscall; clear as much as possible 2102 * to minimize user space-kernel interference. 2103 */ 2104 wrmsrl(MSR_SYSCALL_MASK, 2105 X86_EFLAGS_CF|X86_EFLAGS_PF|X86_EFLAGS_AF| 2106 X86_EFLAGS_ZF|X86_EFLAGS_SF|X86_EFLAGS_TF| 2107 X86_EFLAGS_IF|X86_EFLAGS_DF|X86_EFLAGS_OF| 2108 X86_EFLAGS_IOPL|X86_EFLAGS_NT|X86_EFLAGS_RF| 2109 X86_EFLAGS_AC|X86_EFLAGS_ID); 2110 } 2111 2112 #else /* CONFIG_X86_64 */ 2113 2114 #ifdef CONFIG_STACKPROTECTOR 2115 DEFINE_PER_CPU(unsigned long, __stack_chk_guard); 2116 EXPORT_PER_CPU_SYMBOL(__stack_chk_guard); 2117 #endif 2118 2119 #endif /* CONFIG_X86_64 */ 2120 2121 /* 2122 * Clear all 6 debug registers: 2123 */ 2124 static void clear_all_debug_regs(void) 2125 { 2126 int i; 2127 2128 for (i = 0; i < 8; i++) { 2129 /* Ignore db4, db5 */ 2130 if ((i == 4) || (i == 5)) 2131 continue; 2132 2133 set_debugreg(0, i); 2134 } 2135 } 2136 2137 #ifdef CONFIG_KGDB 2138 /* 2139 * Restore debug regs if using kgdbwait and you have a kernel debugger 2140 * connection established. 2141 */ 2142 static void dbg_restore_debug_regs(void) 2143 { 2144 if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break)) 2145 arch_kgdb_ops.correct_hw_break(); 2146 } 2147 #else /* ! CONFIG_KGDB */ 2148 #define dbg_restore_debug_regs() 2149 #endif /* ! CONFIG_KGDB */ 2150 2151 static inline void setup_getcpu(int cpu) 2152 { 2153 unsigned long cpudata = vdso_encode_cpunode(cpu, early_cpu_to_node(cpu)); 2154 struct desc_struct d = { }; 2155 2156 if (boot_cpu_has(X86_FEATURE_RDTSCP) || boot_cpu_has(X86_FEATURE_RDPID)) 2157 wrmsr(MSR_TSC_AUX, cpudata, 0); 2158 2159 /* Store CPU and node number in limit. */ 2160 d.limit0 = cpudata; 2161 d.limit1 = cpudata >> 16; 2162 2163 d.type = 5; /* RO data, expand down, accessed */ 2164 d.dpl = 3; /* Visible to user code */ 2165 d.s = 1; /* Not a system segment */ 2166 d.p = 1; /* Present */ 2167 d.d = 1; /* 32-bit */ 2168 2169 write_gdt_entry(get_cpu_gdt_rw(cpu), GDT_ENTRY_CPUNODE, &d, DESCTYPE_S); 2170 } 2171 2172 #ifdef CONFIG_X86_64 2173 static inline void tss_setup_ist(struct tss_struct *tss) 2174 { 2175 /* Set up the per-CPU TSS IST stacks */ 2176 tss->x86_tss.ist[IST_INDEX_DF] = __this_cpu_ist_top_va(DF); 2177 tss->x86_tss.ist[IST_INDEX_NMI] = __this_cpu_ist_top_va(NMI); 2178 tss->x86_tss.ist[IST_INDEX_DB] = __this_cpu_ist_top_va(DB); 2179 tss->x86_tss.ist[IST_INDEX_MCE] = __this_cpu_ist_top_va(MCE); 2180 /* Only mapped when SEV-ES is active */ 2181 tss->x86_tss.ist[IST_INDEX_VC] = __this_cpu_ist_top_va(VC); 2182 } 2183 #else /* CONFIG_X86_64 */ 2184 static inline void tss_setup_ist(struct tss_struct *tss) { } 2185 #endif /* !CONFIG_X86_64 */ 2186 2187 static inline void tss_setup_io_bitmap(struct tss_struct *tss) 2188 { 2189 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET_INVALID; 2190 2191 #ifdef CONFIG_X86_IOPL_IOPERM 2192 tss->io_bitmap.prev_max = 0; 2193 tss->io_bitmap.prev_sequence = 0; 2194 memset(tss->io_bitmap.bitmap, 0xff, sizeof(tss->io_bitmap.bitmap)); 2195 /* 2196 * Invalidate the extra array entry past the end of the all 2197 * permission bitmap as required by the hardware. 2198 */ 2199 tss->io_bitmap.mapall[IO_BITMAP_LONGS] = ~0UL; 2200 #endif 2201 } 2202 2203 /* 2204 * Setup everything needed to handle exceptions from the IDT, including the IST 2205 * exceptions which use paranoid_entry(). 2206 */ 2207 void cpu_init_exception_handling(void) 2208 { 2209 struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw); 2210 int cpu = raw_smp_processor_id(); 2211 2212 /* paranoid_entry() gets the CPU number from the GDT */ 2213 setup_getcpu(cpu); 2214 2215 /* IST vectors need TSS to be set up. */ 2216 tss_setup_ist(tss); 2217 tss_setup_io_bitmap(tss); 2218 set_tss_desc(cpu, &get_cpu_entry_area(cpu)->tss.x86_tss); 2219 2220 load_TR_desc(); 2221 2222 /* GHCB needs to be setup to handle #VC. */ 2223 setup_ghcb(); 2224 2225 /* Finally load the IDT */ 2226 load_current_idt(); 2227 } 2228 2229 /* 2230 * cpu_init() initializes state that is per-CPU. Some data is already 2231 * initialized (naturally) in the bootstrap process, such as the GDT. We 2232 * reload it nevertheless, this function acts as a 'CPU state barrier', 2233 * nothing should get across. 2234 */ 2235 void cpu_init(void) 2236 { 2237 struct task_struct *cur = current; 2238 int cpu = raw_smp_processor_id(); 2239 2240 #ifdef CONFIG_NUMA 2241 if (this_cpu_read(numa_node) == 0 && 2242 early_cpu_to_node(cpu) != NUMA_NO_NODE) 2243 set_numa_node(early_cpu_to_node(cpu)); 2244 #endif 2245 pr_debug("Initializing CPU#%d\n", cpu); 2246 2247 if (IS_ENABLED(CONFIG_X86_64) || cpu_feature_enabled(X86_FEATURE_VME) || 2248 boot_cpu_has(X86_FEATURE_TSC) || boot_cpu_has(X86_FEATURE_DE)) 2249 cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE); 2250 2251 if (IS_ENABLED(CONFIG_X86_64)) { 2252 loadsegment(fs, 0); 2253 memset(cur->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8); 2254 syscall_init(); 2255 2256 wrmsrl(MSR_FS_BASE, 0); 2257 wrmsrl(MSR_KERNEL_GS_BASE, 0); 2258 barrier(); 2259 2260 x2apic_setup(); 2261 } 2262 2263 mmgrab(&init_mm); 2264 cur->active_mm = &init_mm; 2265 BUG_ON(cur->mm); 2266 initialize_tlbstate_and_flush(); 2267 enter_lazy_tlb(&init_mm, cur); 2268 2269 /* 2270 * sp0 points to the entry trampoline stack regardless of what task 2271 * is running. 2272 */ 2273 load_sp0((unsigned long)(cpu_entry_stack(cpu) + 1)); 2274 2275 load_mm_ldt(&init_mm); 2276 2277 clear_all_debug_regs(); 2278 dbg_restore_debug_regs(); 2279 2280 doublefault_init_cpu_tss(); 2281 2282 if (is_uv_system()) 2283 uv_cpu_init(); 2284 2285 load_fixmap_gdt(cpu); 2286 } 2287 2288 #ifdef CONFIG_MICROCODE_LATE_LOADING 2289 /** 2290 * store_cpu_caps() - Store a snapshot of CPU capabilities 2291 * @curr_info: Pointer where to store it 2292 * 2293 * Returns: None 2294 */ 2295 void store_cpu_caps(struct cpuinfo_x86 *curr_info) 2296 { 2297 /* Reload CPUID max function as it might've changed. */ 2298 curr_info->cpuid_level = cpuid_eax(0); 2299 2300 /* Copy all capability leafs and pick up the synthetic ones. */ 2301 memcpy(&curr_info->x86_capability, &boot_cpu_data.x86_capability, 2302 sizeof(curr_info->x86_capability)); 2303 2304 /* Get the hardware CPUID leafs */ 2305 get_cpu_cap(curr_info); 2306 } 2307 2308 /** 2309 * microcode_check() - Check if any CPU capabilities changed after an update. 2310 * @prev_info: CPU capabilities stored before an update. 2311 * 2312 * The microcode loader calls this upon late microcode load to recheck features, 2313 * only when microcode has been updated. Caller holds and CPU hotplug lock. 2314 * 2315 * Return: None 2316 */ 2317 void microcode_check(struct cpuinfo_x86 *prev_info) 2318 { 2319 struct cpuinfo_x86 curr_info; 2320 2321 perf_check_microcode(); 2322 2323 amd_check_microcode(); 2324 2325 store_cpu_caps(&curr_info); 2326 2327 if (!memcmp(&prev_info->x86_capability, &curr_info.x86_capability, 2328 sizeof(prev_info->x86_capability))) 2329 return; 2330 2331 pr_warn("x86/CPU: CPU features have changed after loading microcode, but might not take effect.\n"); 2332 pr_warn("x86/CPU: Please consider either early loading through initrd/built-in or a potential BIOS update.\n"); 2333 } 2334 #endif 2335 2336 /* 2337 * Invoked from core CPU hotplug code after hotplug operations 2338 */ 2339 void arch_smt_update(void) 2340 { 2341 /* Handle the speculative execution misfeatures */ 2342 cpu_bugs_smt_update(); 2343 /* Check whether IPI broadcasting can be enabled */ 2344 apic_smt_update(); 2345 } 2346 2347 void __init arch_cpu_finalize_init(void) 2348 { 2349 identify_boot_cpu(); 2350 2351 /* 2352 * identify_boot_cpu() initialized SMT support information, let the 2353 * core code know. 2354 */ 2355 cpu_smt_set_num_threads(smp_num_siblings, smp_num_siblings); 2356 2357 if (!IS_ENABLED(CONFIG_SMP)) { 2358 pr_info("CPU: "); 2359 print_cpu_info(&boot_cpu_data); 2360 } 2361 2362 cpu_select_mitigations(); 2363 2364 arch_smt_update(); 2365 2366 if (IS_ENABLED(CONFIG_X86_32)) { 2367 /* 2368 * Check whether this is a real i386 which is not longer 2369 * supported and fixup the utsname. 2370 */ 2371 if (boot_cpu_data.x86 < 4) 2372 panic("Kernel requires i486+ for 'invlpg' and other features"); 2373 2374 init_utsname()->machine[1] = 2375 '0' + (boot_cpu_data.x86 > 6 ? 6 : boot_cpu_data.x86); 2376 } 2377 2378 /* 2379 * Must be before alternatives because it might set or clear 2380 * feature bits. 2381 */ 2382 fpu__init_system(); 2383 fpu__init_cpu(); 2384 2385 alternative_instructions(); 2386 2387 if (IS_ENABLED(CONFIG_X86_64)) { 2388 /* 2389 * Make sure the first 2MB area is not mapped by huge pages 2390 * There are typically fixed size MTRRs in there and overlapping 2391 * MTRRs into large pages causes slow downs. 2392 * 2393 * Right now we don't do that with gbpages because there seems 2394 * very little benefit for that case. 2395 */ 2396 if (!direct_gbpages) 2397 set_memory_4k((unsigned long)__va(0), 1); 2398 } else { 2399 fpu__init_check_bugs(); 2400 } 2401 2402 /* 2403 * This needs to be called before any devices perform DMA 2404 * operations that might use the SWIOTLB bounce buffers. It will 2405 * mark the bounce buffers as decrypted so that their usage will 2406 * not cause "plain-text" data to be decrypted when accessed. It 2407 * must be called after late_time_init() so that Hyper-V x86/x64 2408 * hypercalls work when the SWIOTLB bounce buffers are decrypted. 2409 */ 2410 mem_encrypt_init(); 2411 } 2412