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 #include <linux/efi.h> 30 31 #include <asm/alternative.h> 32 #include <asm/cmdline.h> 33 #include <asm/cpuid/api.h> 34 #include <asm/perf_event.h> 35 #include <asm/mmu_context.h> 36 #include <asm/doublefault.h> 37 #include <asm/archrandom.h> 38 #include <asm/hypervisor.h> 39 #include <asm/processor.h> 40 #include <asm/tlbflush.h> 41 #include <asm/debugreg.h> 42 #include <asm/sections.h> 43 #include <asm/vsyscall.h> 44 #include <linux/topology.h> 45 #include <linux/cpumask.h> 46 #include <linux/atomic.h> 47 #include <asm/proto.h> 48 #include <asm/setup.h> 49 #include <asm/apic.h> 50 #include <asm/desc.h> 51 #include <asm/fpu/api.h> 52 #include <asm/mtrr.h> 53 #include <asm/hwcap2.h> 54 #include <linux/numa.h> 55 #include <asm/numa.h> 56 #include <asm/asm.h> 57 #include <asm/bugs.h> 58 #include <asm/cpu.h> 59 #include <asm/mce.h> 60 #include <asm/msr.h> 61 #include <asm/cacheinfo.h> 62 #include <asm/memtype.h> 63 #include <asm/microcode.h> 64 #include <asm/intel-family.h> 65 #include <asm/cpu_device_id.h> 66 #include <asm/fred.h> 67 #include <asm/uv/uv.h> 68 #include <asm/ia32.h> 69 #include <asm/set_memory.h> 70 #include <asm/traps.h> 71 #include <asm/sev.h> 72 #include <asm/tdx.h> 73 #include <asm/posted_intr.h> 74 #include <asm/runtime-const.h> 75 76 #include "cpu.h" 77 78 DEFINE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info); 79 EXPORT_PER_CPU_SYMBOL(cpu_info); 80 81 u32 elf_hwcap2 __read_mostly; 82 83 /* Number of siblings per CPU package */ 84 unsigned int __max_threads_per_core __ro_after_init = 1; 85 EXPORT_SYMBOL(__max_threads_per_core); 86 87 unsigned int __max_dies_per_package __ro_after_init = 1; 88 EXPORT_SYMBOL(__max_dies_per_package); 89 90 unsigned int __max_logical_packages __ro_after_init = 1; 91 EXPORT_SYMBOL(__max_logical_packages); 92 93 unsigned int __num_cores_per_package __ro_after_init = 1; 94 EXPORT_SYMBOL(__num_cores_per_package); 95 96 unsigned int __num_threads_per_package __ro_after_init = 1; 97 EXPORT_SYMBOL(__num_threads_per_package); 98 99 static struct ppin_info { 100 int feature; 101 int msr_ppin_ctl; 102 int msr_ppin; 103 } ppin_info[] = { 104 [X86_VENDOR_INTEL] = { 105 .feature = X86_FEATURE_INTEL_PPIN, 106 .msr_ppin_ctl = MSR_PPIN_CTL, 107 .msr_ppin = MSR_PPIN 108 }, 109 [X86_VENDOR_AMD] = { 110 .feature = X86_FEATURE_AMD_PPIN, 111 .msr_ppin_ctl = MSR_AMD_PPIN_CTL, 112 .msr_ppin = MSR_AMD_PPIN 113 }, 114 }; 115 116 static const struct x86_cpu_id ppin_cpuids[] = { 117 X86_MATCH_FEATURE(X86_FEATURE_AMD_PPIN, &ppin_info[X86_VENDOR_AMD]), 118 X86_MATCH_FEATURE(X86_FEATURE_INTEL_PPIN, &ppin_info[X86_VENDOR_INTEL]), 119 120 /* Legacy models without CPUID enumeration */ 121 X86_MATCH_VFM(INTEL_IVYBRIDGE_X, &ppin_info[X86_VENDOR_INTEL]), 122 X86_MATCH_VFM(INTEL_HASWELL_X, &ppin_info[X86_VENDOR_INTEL]), 123 X86_MATCH_VFM(INTEL_BROADWELL_D, &ppin_info[X86_VENDOR_INTEL]), 124 X86_MATCH_VFM(INTEL_BROADWELL_X, &ppin_info[X86_VENDOR_INTEL]), 125 X86_MATCH_VFM(INTEL_SKYLAKE_X, &ppin_info[X86_VENDOR_INTEL]), 126 X86_MATCH_VFM(INTEL_ICELAKE_X, &ppin_info[X86_VENDOR_INTEL]), 127 X86_MATCH_VFM(INTEL_ICELAKE_D, &ppin_info[X86_VENDOR_INTEL]), 128 X86_MATCH_VFM(INTEL_SAPPHIRERAPIDS_X, &ppin_info[X86_VENDOR_INTEL]), 129 X86_MATCH_VFM(INTEL_EMERALDRAPIDS_X, &ppin_info[X86_VENDOR_INTEL]), 130 X86_MATCH_VFM(INTEL_XEON_PHI_KNL, &ppin_info[X86_VENDOR_INTEL]), 131 X86_MATCH_VFM(INTEL_XEON_PHI_KNM, &ppin_info[X86_VENDOR_INTEL]), 132 133 {} 134 }; 135 136 static void ppin_init(struct cpuinfo_x86 *c) 137 { 138 const struct x86_cpu_id *id; 139 unsigned long long val; 140 struct ppin_info *info; 141 142 id = x86_match_cpu(ppin_cpuids); 143 if (!id) 144 return; 145 146 /* 147 * Testing the presence of the MSR is not enough. Need to check 148 * that the PPIN_CTL allows reading of the PPIN. 149 */ 150 info = (struct ppin_info *)id->driver_data; 151 152 if (rdmsrq_safe(info->msr_ppin_ctl, &val)) 153 goto clear_ppin; 154 155 if ((val & 3UL) == 1UL) { 156 /* PPIN locked in disabled mode */ 157 goto clear_ppin; 158 } 159 160 /* If PPIN is disabled, try to enable */ 161 if (!(val & 2UL)) { 162 wrmsrq_safe(info->msr_ppin_ctl, val | 2UL); 163 rdmsrq_safe(info->msr_ppin_ctl, &val); 164 } 165 166 /* Is the enable bit set? */ 167 if (val & 2UL) { 168 c->ppin = native_rdmsrq(info->msr_ppin); 169 set_cpu_cap(c, info->feature); 170 return; 171 } 172 173 clear_ppin: 174 setup_clear_cpu_cap(info->feature); 175 } 176 177 static void default_init(struct cpuinfo_x86 *c) 178 { 179 #ifdef CONFIG_X86_64 180 cpu_detect_cache_sizes(c); 181 #else 182 /* Not much we can do here... */ 183 /* Check if at least it has cpuid */ 184 if (c->cpuid_level == -1) { 185 /* No cpuid. It must be an ancient CPU */ 186 if (c->x86 == 4) 187 strcpy(c->x86_model_id, "486"); 188 else if (c->x86 == 3) 189 strcpy(c->x86_model_id, "386"); 190 } 191 #endif 192 } 193 194 static const struct cpu_dev default_cpu = { 195 .c_init = default_init, 196 .c_vendor = "Unknown", 197 .c_x86_vendor = X86_VENDOR_UNKNOWN, 198 }; 199 200 static const struct cpu_dev *this_cpu = &default_cpu; 201 202 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = { 203 #ifdef CONFIG_X86_64 204 /* 205 * We need valid kernel segments for data and code in long mode too 206 * IRET will check the segment types kkeil 2000/10/28 207 * Also sysret mandates a special GDT layout 208 * 209 * TLS descriptors are currently at a different place compared to i386. 210 * Hopefully nobody expects them at a fixed place (Wine?) 211 */ 212 [GDT_ENTRY_KERNEL32_CS] = GDT_ENTRY_INIT(DESC_CODE32, 0, 0xfffff), 213 [GDT_ENTRY_KERNEL_CS] = GDT_ENTRY_INIT(DESC_CODE64, 0, 0xfffff), 214 [GDT_ENTRY_KERNEL_DS] = GDT_ENTRY_INIT(DESC_DATA64, 0, 0xfffff), 215 [GDT_ENTRY_DEFAULT_USER32_CS] = GDT_ENTRY_INIT(DESC_CODE32 | DESC_USER, 0, 0xfffff), 216 [GDT_ENTRY_DEFAULT_USER_DS] = GDT_ENTRY_INIT(DESC_DATA64 | DESC_USER, 0, 0xfffff), 217 [GDT_ENTRY_DEFAULT_USER_CS] = GDT_ENTRY_INIT(DESC_CODE64 | DESC_USER, 0, 0xfffff), 218 #else 219 [GDT_ENTRY_KERNEL_CS] = GDT_ENTRY_INIT(DESC_CODE32, 0, 0xfffff), 220 [GDT_ENTRY_KERNEL_DS] = GDT_ENTRY_INIT(DESC_DATA32, 0, 0xfffff), 221 [GDT_ENTRY_DEFAULT_USER_CS] = GDT_ENTRY_INIT(DESC_CODE32 | DESC_USER, 0, 0xfffff), 222 [GDT_ENTRY_DEFAULT_USER_DS] = GDT_ENTRY_INIT(DESC_DATA32 | DESC_USER, 0, 0xfffff), 223 /* 224 * Segments used for calling PnP BIOS have byte granularity. 225 * They code segments and data segments have fixed 64k limits, 226 * the transfer segment sizes are set at run time. 227 */ 228 [GDT_ENTRY_PNPBIOS_CS32] = GDT_ENTRY_INIT(DESC_CODE32_BIOS, 0, 0xffff), 229 [GDT_ENTRY_PNPBIOS_CS16] = GDT_ENTRY_INIT(DESC_CODE16, 0, 0xffff), 230 [GDT_ENTRY_PNPBIOS_DS] = GDT_ENTRY_INIT(DESC_DATA16, 0, 0xffff), 231 [GDT_ENTRY_PNPBIOS_TS1] = GDT_ENTRY_INIT(DESC_DATA16, 0, 0), 232 [GDT_ENTRY_PNPBIOS_TS2] = GDT_ENTRY_INIT(DESC_DATA16, 0, 0), 233 /* 234 * The APM segments have byte granularity and their bases 235 * are set at run time. All have 64k limits. 236 */ 237 [GDT_ENTRY_APMBIOS_BASE] = GDT_ENTRY_INIT(DESC_CODE32_BIOS, 0, 0xffff), 238 [GDT_ENTRY_APMBIOS_BASE+1] = GDT_ENTRY_INIT(DESC_CODE16, 0, 0xffff), 239 [GDT_ENTRY_APMBIOS_BASE+2] = GDT_ENTRY_INIT(DESC_DATA32_BIOS, 0, 0xffff), 240 241 [GDT_ENTRY_ESPFIX_SS] = GDT_ENTRY_INIT(DESC_DATA32, 0, 0xfffff), 242 [GDT_ENTRY_PERCPU] = GDT_ENTRY_INIT(DESC_DATA32, 0, 0xfffff), 243 #endif 244 } }; 245 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page); 246 SYM_PIC_ALIAS(gdt_page); 247 248 #ifdef CONFIG_X86_64 249 static int __init x86_nopcid_setup(char *s) 250 { 251 /* nopcid doesn't accept parameters */ 252 if (s) 253 return -EINVAL; 254 255 /* do not emit a message if the feature is not present */ 256 if (!boot_cpu_has(X86_FEATURE_PCID)) 257 return 0; 258 259 setup_clear_cpu_cap(X86_FEATURE_PCID); 260 pr_info("nopcid: PCID feature disabled\n"); 261 return 0; 262 } 263 early_param("nopcid", x86_nopcid_setup); 264 #endif 265 266 static int __init x86_noinvpcid_setup(char *s) 267 { 268 /* noinvpcid doesn't accept parameters */ 269 if (s) 270 return -EINVAL; 271 272 /* do not emit a message if the feature is not present */ 273 if (!boot_cpu_has(X86_FEATURE_INVPCID)) 274 return 0; 275 276 setup_clear_cpu_cap(X86_FEATURE_INVPCID); 277 pr_info("noinvpcid: INVPCID feature disabled\n"); 278 return 0; 279 } 280 early_param("noinvpcid", x86_noinvpcid_setup); 281 282 /* Standard macro to see if a specific flag is changeable */ 283 static inline bool flag_is_changeable_p(unsigned long flag) 284 { 285 unsigned long f1, f2; 286 287 if (!IS_ENABLED(CONFIG_X86_32)) 288 return true; 289 290 /* 291 * Cyrix and IDT cpus allow disabling of CPUID 292 * so the code below may return different results 293 * when it is executed before and after enabling 294 * the CPUID. Add "volatile" to not allow gcc to 295 * optimize the subsequent calls to this function. 296 */ 297 asm volatile ("pushfl \n\t" 298 "pushfl \n\t" 299 "popl %0 \n\t" 300 "movl %0, %1 \n\t" 301 "xorl %2, %0 \n\t" 302 "pushl %0 \n\t" 303 "popfl \n\t" 304 "pushfl \n\t" 305 "popl %0 \n\t" 306 "popfl \n\t" 307 308 : "=&r" (f1), "=&r" (f2) 309 : "ir" (flag)); 310 311 return (f1 ^ f2) & flag; 312 } 313 314 #ifdef CONFIG_X86_32 315 static int cachesize_override = -1; 316 static int disable_x86_serial_nr = 1; 317 318 static int __init cachesize_setup(char *str) 319 { 320 get_option(&str, &cachesize_override); 321 return 1; 322 } 323 __setup("cachesize=", cachesize_setup); 324 325 /* Probe for the CPUID instruction */ 326 bool cpuid_feature(void) 327 { 328 return flag_is_changeable_p(X86_EFLAGS_ID); 329 } 330 331 static void squash_the_stupid_serial_number(struct cpuinfo_x86 *c) 332 { 333 unsigned long lo, hi; 334 335 if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr) 336 return; 337 338 /* Disable processor serial number: */ 339 340 rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi); 341 lo |= 0x200000; 342 wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi); 343 344 pr_notice("CPU serial number disabled.\n"); 345 clear_cpu_cap(c, X86_FEATURE_PN); 346 347 /* Disabling the serial number may affect the cpuid level */ 348 c->cpuid_level = cpuid_eax(0); 349 } 350 351 static int __init x86_serial_nr_setup(char *s) 352 { 353 disable_x86_serial_nr = 0; 354 return 1; 355 } 356 __setup("serialnumber", x86_serial_nr_setup); 357 #else 358 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c) 359 { 360 } 361 #endif 362 363 static __always_inline void setup_smep(struct cpuinfo_x86 *c) 364 { 365 if (cpu_has(c, X86_FEATURE_SMEP)) 366 cr4_set_bits(X86_CR4_SMEP); 367 } 368 369 static __always_inline void setup_smap(struct cpuinfo_x86 *c) 370 { 371 unsigned long eflags = native_save_fl(); 372 373 /* This should have been cleared long ago */ 374 BUG_ON(eflags & X86_EFLAGS_AC); 375 376 if (cpu_has(c, X86_FEATURE_SMAP)) 377 cr4_set_bits(X86_CR4_SMAP); 378 } 379 380 static __always_inline void setup_umip(struct cpuinfo_x86 *c) 381 { 382 /* Check the boot processor, plus build option for UMIP. */ 383 if (!cpu_feature_enabled(X86_FEATURE_UMIP)) 384 goto out; 385 386 /* Check the current processor's cpuid bits. */ 387 if (!cpu_has(c, X86_FEATURE_UMIP)) 388 goto out; 389 390 cr4_set_bits(X86_CR4_UMIP); 391 392 pr_info_once("x86/cpu: User Mode Instruction Prevention (UMIP) activated\n"); 393 394 return; 395 396 out: 397 /* 398 * Make sure UMIP is disabled in case it was enabled in a 399 * previous boot (e.g., via kexec). 400 */ 401 cr4_clear_bits(X86_CR4_UMIP); 402 } 403 404 /* These bits should not change their value after CPU init is finished. */ 405 static const unsigned long cr4_pinned_mask = X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_UMIP | 406 X86_CR4_FSGSBASE | X86_CR4_CET | X86_CR4_FRED; 407 static DEFINE_STATIC_KEY_FALSE_RO(cr_pinning); 408 static unsigned long cr4_pinned_bits __ro_after_init; 409 410 void native_write_cr0(unsigned long val) 411 { 412 unsigned long bits_missing = 0; 413 414 set_register: 415 asm volatile("mov %0,%%cr0": "+r" (val) : : "memory"); 416 417 if (static_branch_likely(&cr_pinning)) { 418 if (unlikely((val & X86_CR0_WP) != X86_CR0_WP)) { 419 bits_missing = X86_CR0_WP; 420 val |= bits_missing; 421 goto set_register; 422 } 423 /* Warn after we've set the missing bits. */ 424 WARN_ONCE(bits_missing, "CR0 WP bit went missing!?\n"); 425 } 426 } 427 EXPORT_SYMBOL(native_write_cr0); 428 429 void __no_profile native_write_cr4(unsigned long val) 430 { 431 unsigned long bits_changed = 0; 432 433 set_register: 434 asm volatile("mov %0,%%cr4": "+r" (val) : : "memory"); 435 436 if (static_branch_likely(&cr_pinning)) { 437 if (unlikely((val & cr4_pinned_mask) != cr4_pinned_bits)) { 438 bits_changed = (val & cr4_pinned_mask) ^ cr4_pinned_bits; 439 val = (val & ~cr4_pinned_mask) | cr4_pinned_bits; 440 goto set_register; 441 } 442 /* Warn after we've corrected the changed bits. */ 443 WARN_ONCE(bits_changed, "pinned CR4 bits changed: 0x%lx!?\n", 444 bits_changed); 445 } 446 } 447 #if IS_MODULE(CONFIG_LKDTM) 448 EXPORT_SYMBOL_GPL(native_write_cr4); 449 #endif 450 451 void cr4_update_irqsoff(unsigned long set, unsigned long clear) 452 { 453 unsigned long newval, cr4 = this_cpu_read(cpu_tlbstate.cr4); 454 455 lockdep_assert_irqs_disabled(); 456 457 newval = (cr4 & ~clear) | set; 458 if (newval != cr4) { 459 this_cpu_write(cpu_tlbstate.cr4, newval); 460 __write_cr4(newval); 461 } 462 } 463 EXPORT_SYMBOL(cr4_update_irqsoff); 464 465 /* Read the CR4 shadow. */ 466 unsigned long cr4_read_shadow(void) 467 { 468 return this_cpu_read(cpu_tlbstate.cr4); 469 } 470 EXPORT_SYMBOL_GPL(cr4_read_shadow); 471 472 void cr4_init(void) 473 { 474 unsigned long cr4 = __read_cr4(); 475 476 if (boot_cpu_has(X86_FEATURE_PCID)) 477 cr4 |= X86_CR4_PCIDE; 478 if (static_branch_likely(&cr_pinning)) 479 cr4 = (cr4 & ~cr4_pinned_mask) | cr4_pinned_bits; 480 481 __write_cr4(cr4); 482 483 /* Initialize cr4 shadow for this CPU. */ 484 this_cpu_write(cpu_tlbstate.cr4, cr4); 485 } 486 487 /* 488 * Once CPU feature detection is finished (and boot params have been 489 * parsed), record any of the sensitive CR bits that are set, and 490 * enable CR pinning. 491 */ 492 static void __init setup_cr_pinning(void) 493 { 494 cr4_pinned_bits = this_cpu_read(cpu_tlbstate.cr4) & cr4_pinned_mask; 495 static_key_enable(&cr_pinning.key); 496 } 497 498 static __init int x86_nofsgsbase_setup(char *arg) 499 { 500 /* Require an exact match without trailing characters. */ 501 if (strlen(arg)) 502 return 0; 503 504 /* Do not emit a message if the feature is not present. */ 505 if (!boot_cpu_has(X86_FEATURE_FSGSBASE)) 506 return 1; 507 508 setup_clear_cpu_cap(X86_FEATURE_FSGSBASE); 509 pr_info("FSGSBASE disabled via kernel command line\n"); 510 return 1; 511 } 512 __setup("nofsgsbase", x86_nofsgsbase_setup); 513 514 /* 515 * Protection Keys are not available in 32-bit mode. 516 */ 517 static bool pku_disabled; 518 519 static __always_inline void setup_pku(struct cpuinfo_x86 *c) 520 { 521 if (c == &boot_cpu_data) { 522 if (pku_disabled || !cpu_feature_enabled(X86_FEATURE_PKU)) 523 return; 524 /* 525 * Setting CR4.PKE will cause the X86_FEATURE_OSPKE cpuid 526 * bit to be set. Enforce it. 527 */ 528 setup_force_cpu_cap(X86_FEATURE_OSPKE); 529 530 } else if (!cpu_feature_enabled(X86_FEATURE_OSPKE)) { 531 return; 532 } 533 534 cr4_set_bits(X86_CR4_PKE); 535 /* Load the default PKRU value */ 536 pkru_write_default(); 537 } 538 539 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS 540 static __init int setup_disable_pku(char *arg) 541 { 542 /* 543 * Do not clear the X86_FEATURE_PKU bit. All of the 544 * runtime checks are against OSPKE so clearing the 545 * bit does nothing. 546 * 547 * This way, we will see "pku" in cpuinfo, but not 548 * "ospke", which is exactly what we want. It shows 549 * that the CPU has PKU, but the OS has not enabled it. 550 * This happens to be exactly how a system would look 551 * if we disabled the config option. 552 */ 553 pr_info("x86: 'nopku' specified, disabling Memory Protection Keys\n"); 554 pku_disabled = true; 555 return 1; 556 } 557 __setup("nopku", setup_disable_pku); 558 #endif 559 560 #ifdef CONFIG_X86_KERNEL_IBT 561 562 __noendbr u64 ibt_save(bool disable) 563 { 564 u64 msr = 0; 565 566 if (cpu_feature_enabled(X86_FEATURE_IBT)) { 567 rdmsrq(MSR_IA32_S_CET, msr); 568 if (disable) 569 wrmsrq(MSR_IA32_S_CET, msr & ~CET_ENDBR_EN); 570 } 571 572 return msr; 573 } 574 575 __noendbr void ibt_restore(u64 save) 576 { 577 u64 msr; 578 579 if (cpu_feature_enabled(X86_FEATURE_IBT)) { 580 rdmsrq(MSR_IA32_S_CET, msr); 581 msr &= ~CET_ENDBR_EN; 582 msr |= (save & CET_ENDBR_EN); 583 wrmsrq(MSR_IA32_S_CET, msr); 584 } 585 } 586 587 #endif 588 589 static __always_inline void setup_cet(struct cpuinfo_x86 *c) 590 { 591 bool user_shstk, kernel_ibt; 592 593 if (!IS_ENABLED(CONFIG_X86_CET)) 594 return; 595 596 kernel_ibt = HAS_KERNEL_IBT && cpu_feature_enabled(X86_FEATURE_IBT); 597 user_shstk = cpu_feature_enabled(X86_FEATURE_SHSTK) && 598 IS_ENABLED(CONFIG_X86_USER_SHADOW_STACK); 599 600 if (!kernel_ibt && !user_shstk) 601 return; 602 603 if (user_shstk) 604 set_cpu_cap(c, X86_FEATURE_USER_SHSTK); 605 606 if (kernel_ibt) 607 wrmsrq(MSR_IA32_S_CET, CET_ENDBR_EN); 608 else 609 wrmsrq(MSR_IA32_S_CET, 0); 610 611 cr4_set_bits(X86_CR4_CET); 612 613 if (kernel_ibt && ibt_selftest()) { 614 pr_err("IBT selftest: Failed!\n"); 615 wrmsrq(MSR_IA32_S_CET, 0); 616 setup_clear_cpu_cap(X86_FEATURE_IBT); 617 } 618 } 619 620 __noendbr void cet_disable(void) 621 { 622 if (!(cpu_feature_enabled(X86_FEATURE_IBT) || 623 cpu_feature_enabled(X86_FEATURE_SHSTK))) 624 return; 625 626 wrmsrq(MSR_IA32_S_CET, 0); 627 wrmsrq(MSR_IA32_U_CET, 0); 628 } 629 630 /* 631 * Some CPU features depend on higher CPUID levels, which may not always 632 * be available due to CPUID level capping or broken virtualization 633 * software. Add those features to this table to auto-disable them. 634 */ 635 struct cpuid_dependent_feature { 636 u32 feature; 637 u32 level; 638 }; 639 640 static const struct cpuid_dependent_feature 641 cpuid_dependent_features[] = { 642 { X86_FEATURE_MWAIT, CPUID_LEAF_MWAIT }, 643 { X86_FEATURE_DCA, CPUID_LEAF_DCA }, 644 { X86_FEATURE_XSAVE, CPUID_LEAF_XSTATE }, 645 { 0, 0 } 646 }; 647 648 static void filter_cpuid_features(struct cpuinfo_x86 *c, bool warn) 649 { 650 const struct cpuid_dependent_feature *df; 651 652 for (df = cpuid_dependent_features; df->feature; df++) { 653 654 if (!cpu_has(c, df->feature)) 655 continue; 656 /* 657 * Note: cpuid_level is set to -1 if unavailable, but 658 * extended_extended_level is set to 0 if unavailable 659 * and the legitimate extended levels are all negative 660 * when signed; hence the weird messing around with 661 * signs here... 662 */ 663 if (!((s32)df->level < 0 ? 664 (u32)df->level > (u32)c->extended_cpuid_level : 665 (s32)df->level > (s32)c->cpuid_level)) 666 continue; 667 668 clear_cpu_cap(c, df->feature); 669 if (!warn) 670 continue; 671 672 pr_warn("CPU: CPU feature %s disabled, no CPUID level 0x%x\n", 673 x86_cap_flags[df->feature], df->level); 674 } 675 } 676 677 /* 678 * Naming convention should be: <Name> [(<Codename>)] 679 * This table only is used unless init_<vendor>() below doesn't set it; 680 * in particular, if CPUID levels 0x80000002..4 are supported, this 681 * isn't used 682 */ 683 684 /* Look up CPU names by table lookup. */ 685 static const char *table_lookup_model(struct cpuinfo_x86 *c) 686 { 687 #ifdef CONFIG_X86_32 688 const struct legacy_cpu_model_info *info; 689 690 if (c->x86_model >= 16) 691 return NULL; /* Range check */ 692 693 if (!this_cpu) 694 return NULL; 695 696 info = this_cpu->legacy_models; 697 698 while (info->family) { 699 if (info->family == c->x86) 700 return info->model_names[c->x86_model]; 701 info++; 702 } 703 #endif 704 return NULL; /* Not found */ 705 } 706 707 /* Aligned to unsigned long to avoid split lock in atomic bitmap ops */ 708 __u32 cpu_caps_cleared[NCAPINTS + NBUGINTS] __aligned(sizeof(unsigned long)); 709 __u32 cpu_caps_set[NCAPINTS + NBUGINTS] __aligned(sizeof(unsigned long)); 710 711 #ifdef CONFIG_X86_32 712 /* The 32-bit entry code needs to find cpu_entry_area. */ 713 DEFINE_PER_CPU(struct cpu_entry_area *, cpu_entry_area); 714 #endif 715 716 /* Load the original GDT from the per-cpu structure */ 717 void load_direct_gdt(int cpu) 718 { 719 struct desc_ptr gdt_descr; 720 721 gdt_descr.address = (long)get_cpu_gdt_rw(cpu); 722 gdt_descr.size = GDT_SIZE - 1; 723 load_gdt(&gdt_descr); 724 } 725 EXPORT_SYMBOL_GPL(load_direct_gdt); 726 727 /* Load a fixmap remapping of the per-cpu GDT */ 728 void load_fixmap_gdt(int cpu) 729 { 730 struct desc_ptr gdt_descr; 731 732 gdt_descr.address = (long)get_cpu_gdt_ro(cpu); 733 gdt_descr.size = GDT_SIZE - 1; 734 load_gdt(&gdt_descr); 735 } 736 EXPORT_SYMBOL_GPL(load_fixmap_gdt); 737 738 /** 739 * switch_gdt_and_percpu_base - Switch to direct GDT and runtime per CPU base 740 * @cpu: The CPU number for which this is invoked 741 * 742 * Invoked during early boot to switch from early GDT and early per CPU to 743 * the direct GDT and the runtime per CPU area. On 32-bit the percpu base 744 * switch is implicit by loading the direct GDT. On 64bit this requires 745 * to update GSBASE. 746 */ 747 void __init switch_gdt_and_percpu_base(int cpu) 748 { 749 load_direct_gdt(cpu); 750 751 #ifdef CONFIG_X86_64 752 /* 753 * No need to load %gs. It is already correct. 754 * 755 * Writing %gs on 64bit would zero GSBASE which would make any per 756 * CPU operation up to the point of the wrmsrq() fault. 757 * 758 * Set GSBASE to the new offset. Until the wrmsrq() happens the 759 * early mapping is still valid. That means the GSBASE update will 760 * lose any prior per CPU data which was not copied over in 761 * setup_per_cpu_areas(). 762 * 763 * This works even with stackprotector enabled because the 764 * per CPU stack canary is 0 in both per CPU areas. 765 */ 766 wrmsrq(MSR_GS_BASE, cpu_kernelmode_gs_base(cpu)); 767 #else 768 /* 769 * %fs is already set to __KERNEL_PERCPU, but after switching GDT 770 * it is required to load FS again so that the 'hidden' part is 771 * updated from the new GDT. Up to this point the early per CPU 772 * translation is active. Any content of the early per CPU data 773 * which was not copied over in setup_per_cpu_areas() is lost. 774 */ 775 loadsegment(fs, __KERNEL_PERCPU); 776 #endif 777 } 778 779 static const struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {}; 780 781 static void get_model_name(struct cpuinfo_x86 *c) 782 { 783 unsigned int *v; 784 char *p, *q, *s; 785 786 if (c->extended_cpuid_level < 0x80000004) 787 return; 788 789 v = (unsigned int *)c->x86_model_id; 790 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]); 791 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]); 792 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]); 793 c->x86_model_id[48] = 0; 794 795 /* Trim whitespace */ 796 p = q = s = &c->x86_model_id[0]; 797 798 while (*p == ' ') 799 p++; 800 801 while (*p) { 802 /* Note the last non-whitespace index */ 803 if (!isspace(*p)) 804 s = q; 805 806 *q++ = *p++; 807 } 808 809 *(s + 1) = '\0'; 810 } 811 812 void cpu_detect_cache_sizes(struct cpuinfo_x86 *c) 813 { 814 unsigned int n, dummy, ebx, ecx, edx, l2size; 815 816 n = c->extended_cpuid_level; 817 818 if (n >= 0x80000005) { 819 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx); 820 c->x86_cache_size = (ecx>>24) + (edx>>24); 821 #ifdef CONFIG_X86_64 822 /* On K8 L1 TLB is inclusive, so don't count it */ 823 c->x86_tlbsize = 0; 824 #endif 825 } 826 827 if (n < 0x80000006) /* Some chips just has a large L1. */ 828 return; 829 830 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx); 831 l2size = ecx >> 16; 832 833 #ifdef CONFIG_X86_64 834 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff); 835 #else 836 /* do processor-specific cache resizing */ 837 if (this_cpu->legacy_cache_size) 838 l2size = this_cpu->legacy_cache_size(c, l2size); 839 840 /* Allow user to override all this if necessary. */ 841 if (cachesize_override != -1) 842 l2size = cachesize_override; 843 844 if (l2size == 0) 845 return; /* Again, no L2 cache is possible */ 846 #endif 847 848 c->x86_cache_size = l2size; 849 } 850 851 u16 __read_mostly tlb_lli_4k; 852 u16 __read_mostly tlb_lli_2m; 853 u16 __read_mostly tlb_lli_4m; 854 u16 __read_mostly tlb_lld_4k; 855 u16 __read_mostly tlb_lld_2m; 856 u16 __read_mostly tlb_lld_4m; 857 u16 __read_mostly tlb_lld_1g; 858 859 static void cpu_detect_tlb(struct cpuinfo_x86 *c) 860 { 861 if (this_cpu->c_detect_tlb) 862 this_cpu->c_detect_tlb(c); 863 864 pr_info("Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n", 865 tlb_lli_4k, tlb_lli_2m, tlb_lli_4m); 866 867 pr_info("Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d, 1GB %d\n", 868 tlb_lld_4k, tlb_lld_2m, tlb_lld_4m, tlb_lld_1g); 869 } 870 871 void get_cpu_vendor(struct cpuinfo_x86 *c) 872 { 873 char *v = c->x86_vendor_id; 874 int i; 875 876 for (i = 0; i < X86_VENDOR_NUM; i++) { 877 if (!cpu_devs[i]) 878 break; 879 880 if (!strcmp(v, cpu_devs[i]->c_ident[0]) || 881 (cpu_devs[i]->c_ident[1] && 882 !strcmp(v, cpu_devs[i]->c_ident[1]))) { 883 884 this_cpu = cpu_devs[i]; 885 c->x86_vendor = this_cpu->c_x86_vendor; 886 return; 887 } 888 } 889 890 pr_err_once("CPU: vendor_id '%s' unknown, using generic init.\n" \ 891 "CPU: Your system may be unstable.\n", v); 892 893 c->x86_vendor = X86_VENDOR_UNKNOWN; 894 this_cpu = &default_cpu; 895 } 896 897 void cpu_detect(struct cpuinfo_x86 *c) 898 { 899 /* Get vendor name */ 900 cpuid(0x00000000, (unsigned int *)&c->cpuid_level, 901 (unsigned int *)&c->x86_vendor_id[0], 902 (unsigned int *)&c->x86_vendor_id[8], 903 (unsigned int *)&c->x86_vendor_id[4]); 904 905 c->x86 = 4; 906 /* Intel-defined flags: level 0x00000001 */ 907 if (c->cpuid_level >= 0x00000001) { 908 u32 junk, tfms, cap0, misc; 909 910 cpuid(0x00000001, &tfms, &misc, &junk, &cap0); 911 c->x86 = x86_family(tfms); 912 c->x86_model = x86_model(tfms); 913 c->x86_stepping = x86_stepping(tfms); 914 915 if (cap0 & (1<<19)) { 916 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8; 917 c->x86_cache_alignment = c->x86_clflush_size; 918 } 919 } 920 } 921 922 static void apply_forced_caps(struct cpuinfo_x86 *c) 923 { 924 int i; 925 926 for (i = 0; i < NCAPINTS + NBUGINTS; i++) { 927 c->x86_capability[i] &= ~cpu_caps_cleared[i]; 928 c->x86_capability[i] |= cpu_caps_set[i]; 929 } 930 } 931 932 static void init_speculation_control(struct cpuinfo_x86 *c) 933 { 934 /* 935 * The Intel SPEC_CTRL CPUID bit implies IBRS and IBPB support, 936 * and they also have a different bit for STIBP support. Also, 937 * a hypervisor might have set the individual AMD bits even on 938 * Intel CPUs, for finer-grained selection of what's available. 939 */ 940 if (cpu_has(c, X86_FEATURE_SPEC_CTRL)) { 941 set_cpu_cap(c, X86_FEATURE_IBRS); 942 set_cpu_cap(c, X86_FEATURE_IBPB); 943 set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL); 944 } 945 946 if (cpu_has(c, X86_FEATURE_INTEL_STIBP)) 947 set_cpu_cap(c, X86_FEATURE_STIBP); 948 949 if (cpu_has(c, X86_FEATURE_SPEC_CTRL_SSBD) || 950 cpu_has(c, X86_FEATURE_VIRT_SSBD)) 951 set_cpu_cap(c, X86_FEATURE_SSBD); 952 953 if (cpu_has(c, X86_FEATURE_AMD_IBRS)) { 954 set_cpu_cap(c, X86_FEATURE_IBRS); 955 set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL); 956 } 957 958 if (cpu_has(c, X86_FEATURE_AMD_IBPB)) 959 set_cpu_cap(c, X86_FEATURE_IBPB); 960 961 if (cpu_has(c, X86_FEATURE_AMD_STIBP)) { 962 set_cpu_cap(c, X86_FEATURE_STIBP); 963 set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL); 964 } 965 966 if (cpu_has(c, X86_FEATURE_AMD_SSBD)) { 967 set_cpu_cap(c, X86_FEATURE_SSBD); 968 set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL); 969 clear_cpu_cap(c, X86_FEATURE_VIRT_SSBD); 970 } 971 } 972 973 void get_cpu_cap(struct cpuinfo_x86 *c) 974 { 975 u32 eax, ebx, ecx, edx; 976 977 /* Intel-defined flags: level 0x00000001 */ 978 if (c->cpuid_level >= 0x00000001) { 979 cpuid(0x00000001, &eax, &ebx, &ecx, &edx); 980 981 c->x86_capability[CPUID_1_ECX] = ecx; 982 c->x86_capability[CPUID_1_EDX] = edx; 983 } 984 985 /* Thermal and Power Management Leaf: level 0x00000006 (eax) */ 986 if (c->cpuid_level >= 0x00000006) 987 c->x86_capability[CPUID_6_EAX] = cpuid_eax(0x00000006); 988 989 /* Additional Intel-defined flags: level 0x00000007 */ 990 if (c->cpuid_level >= 0x00000007) { 991 cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx); 992 c->x86_capability[CPUID_7_0_EBX] = ebx; 993 c->x86_capability[CPUID_7_ECX] = ecx; 994 c->x86_capability[CPUID_7_EDX] = edx; 995 996 /* Check valid sub-leaf index before accessing it */ 997 if (eax >= 1) { 998 cpuid_count(0x00000007, 1, &eax, &ebx, &ecx, &edx); 999 c->x86_capability[CPUID_7_1_EAX] = eax; 1000 } 1001 } 1002 1003 /* Extended state features: level 0x0000000d */ 1004 if (c->cpuid_level >= 0x0000000d) { 1005 cpuid_count(0x0000000d, 1, &eax, &ebx, &ecx, &edx); 1006 1007 c->x86_capability[CPUID_D_1_EAX] = eax; 1008 } 1009 1010 /* 1011 * Check if extended CPUID leaves are implemented: Max extended 1012 * CPUID leaf must be in the 0x80000001-0x8000ffff range. 1013 */ 1014 eax = cpuid_eax(0x80000000); 1015 c->extended_cpuid_level = ((eax & 0xffff0000) == 0x80000000) ? eax : 0; 1016 1017 if (c->extended_cpuid_level >= 0x80000001) { 1018 cpuid(0x80000001, &eax, &ebx, &ecx, &edx); 1019 1020 c->x86_capability[CPUID_8000_0001_ECX] = ecx; 1021 c->x86_capability[CPUID_8000_0001_EDX] = edx; 1022 } 1023 1024 if (c->extended_cpuid_level >= 0x80000007) { 1025 cpuid(0x80000007, &eax, &ebx, &ecx, &edx); 1026 1027 c->x86_capability[CPUID_8000_0007_EBX] = ebx; 1028 c->x86_power = edx; 1029 } 1030 1031 if (c->extended_cpuid_level >= 0x80000008) { 1032 cpuid(0x80000008, &eax, &ebx, &ecx, &edx); 1033 c->x86_capability[CPUID_8000_0008_EBX] = ebx; 1034 } 1035 1036 if (c->extended_cpuid_level >= 0x8000000a) 1037 c->x86_capability[CPUID_8000_000A_EDX] = cpuid_edx(0x8000000a); 1038 1039 if (c->extended_cpuid_level >= 0x8000001f) 1040 c->x86_capability[CPUID_8000_001F_EAX] = cpuid_eax(0x8000001f); 1041 1042 if (c->extended_cpuid_level >= 0x80000021) 1043 c->x86_capability[CPUID_8000_0021_EAX] = cpuid_eax(0x80000021); 1044 1045 init_scattered_cpuid_features(c); 1046 init_speculation_control(c); 1047 1048 /* 1049 * Clear/Set all flags overridden by options, after probe. 1050 * This needs to happen each time we re-probe, which may happen 1051 * several times during CPU initialization. 1052 */ 1053 apply_forced_caps(c); 1054 } 1055 1056 void get_cpu_address_sizes(struct cpuinfo_x86 *c) 1057 { 1058 u32 eax, ebx, ecx, edx; 1059 1060 if (!cpu_has(c, X86_FEATURE_CPUID) || 1061 (c->extended_cpuid_level < 0x80000008)) { 1062 if (IS_ENABLED(CONFIG_X86_64)) { 1063 c->x86_clflush_size = 64; 1064 c->x86_phys_bits = 36; 1065 c->x86_virt_bits = 48; 1066 } else { 1067 c->x86_clflush_size = 32; 1068 c->x86_virt_bits = 32; 1069 c->x86_phys_bits = 32; 1070 1071 if (cpu_has(c, X86_FEATURE_PAE) || 1072 cpu_has(c, X86_FEATURE_PSE36)) 1073 c->x86_phys_bits = 36; 1074 } 1075 } else { 1076 cpuid(0x80000008, &eax, &ebx, &ecx, &edx); 1077 1078 c->x86_virt_bits = (eax >> 8) & 0xff; 1079 c->x86_phys_bits = eax & 0xff; 1080 1081 /* Provide a sane default if not enumerated: */ 1082 if (!c->x86_clflush_size) 1083 c->x86_clflush_size = 32; 1084 } 1085 1086 c->x86_cache_bits = c->x86_phys_bits; 1087 c->x86_cache_alignment = c->x86_clflush_size; 1088 } 1089 1090 static void identify_cpu_without_cpuid(struct cpuinfo_x86 *c) 1091 { 1092 int i; 1093 1094 /* 1095 * First of all, decide if this is a 486 or higher 1096 * It's a 486 if we can modify the AC flag 1097 */ 1098 if (flag_is_changeable_p(X86_EFLAGS_AC)) 1099 c->x86 = 4; 1100 else 1101 c->x86 = 3; 1102 1103 for (i = 0; i < X86_VENDOR_NUM; i++) 1104 if (cpu_devs[i] && cpu_devs[i]->c_identify) { 1105 c->x86_vendor_id[0] = 0; 1106 cpu_devs[i]->c_identify(c); 1107 if (c->x86_vendor_id[0]) { 1108 get_cpu_vendor(c); 1109 break; 1110 } 1111 } 1112 } 1113 1114 #define NO_SPECULATION BIT(0) 1115 #define NO_MELTDOWN BIT(1) 1116 #define NO_SSB BIT(2) 1117 #define NO_L1TF BIT(3) 1118 #define NO_MDS BIT(4) 1119 #define MSBDS_ONLY BIT(5) 1120 #define NO_SWAPGS BIT(6) 1121 #define NO_ITLB_MULTIHIT BIT(7) 1122 #define NO_SPECTRE_V2 BIT(8) 1123 #define NO_MMIO BIT(9) 1124 #define NO_EIBRS_PBRSB BIT(10) 1125 #define NO_BHI BIT(11) 1126 1127 #define VULNWL(vendor, family, model, whitelist) \ 1128 X86_MATCH_VENDOR_FAM_MODEL(vendor, family, model, whitelist) 1129 1130 #define VULNWL_INTEL(vfm, whitelist) \ 1131 X86_MATCH_VFM(vfm, whitelist) 1132 1133 #define VULNWL_AMD(family, whitelist) \ 1134 VULNWL(AMD, family, X86_MODEL_ANY, whitelist) 1135 1136 #define VULNWL_HYGON(family, whitelist) \ 1137 VULNWL(HYGON, family, X86_MODEL_ANY, whitelist) 1138 1139 static const __initconst struct x86_cpu_id cpu_vuln_whitelist[] = { 1140 VULNWL(ANY, 4, X86_MODEL_ANY, NO_SPECULATION), 1141 VULNWL(CENTAUR, 5, X86_MODEL_ANY, NO_SPECULATION), 1142 VULNWL(INTEL, 5, X86_MODEL_ANY, NO_SPECULATION), 1143 VULNWL(NSC, 5, X86_MODEL_ANY, NO_SPECULATION), 1144 VULNWL(VORTEX, 5, X86_MODEL_ANY, NO_SPECULATION), 1145 VULNWL(VORTEX, 6, X86_MODEL_ANY, NO_SPECULATION), 1146 1147 /* Intel Family 6 */ 1148 VULNWL_INTEL(INTEL_TIGERLAKE, NO_MMIO), 1149 VULNWL_INTEL(INTEL_TIGERLAKE_L, NO_MMIO), 1150 VULNWL_INTEL(INTEL_ALDERLAKE, NO_MMIO), 1151 VULNWL_INTEL(INTEL_ALDERLAKE_L, NO_MMIO), 1152 1153 VULNWL_INTEL(INTEL_ATOM_SALTWELL, NO_SPECULATION | NO_ITLB_MULTIHIT), 1154 VULNWL_INTEL(INTEL_ATOM_SALTWELL_TABLET, NO_SPECULATION | NO_ITLB_MULTIHIT), 1155 VULNWL_INTEL(INTEL_ATOM_SALTWELL_MID, NO_SPECULATION | NO_ITLB_MULTIHIT), 1156 VULNWL_INTEL(INTEL_ATOM_BONNELL, NO_SPECULATION | NO_ITLB_MULTIHIT), 1157 VULNWL_INTEL(INTEL_ATOM_BONNELL_MID, NO_SPECULATION | NO_ITLB_MULTIHIT), 1158 1159 VULNWL_INTEL(INTEL_ATOM_SILVERMONT, NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT), 1160 VULNWL_INTEL(INTEL_ATOM_SILVERMONT_D, NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT), 1161 VULNWL_INTEL(INTEL_ATOM_SILVERMONT_MID, NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT), 1162 VULNWL_INTEL(INTEL_ATOM_AIRMONT, NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT), 1163 VULNWL_INTEL(INTEL_XEON_PHI_KNL, NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT), 1164 VULNWL_INTEL(INTEL_XEON_PHI_KNM, NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT), 1165 1166 VULNWL_INTEL(INTEL_CORE_YONAH, NO_SSB), 1167 1168 VULNWL_INTEL(INTEL_ATOM_SILVERMONT_MID2,NO_SSB | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT | MSBDS_ONLY), 1169 VULNWL_INTEL(INTEL_ATOM_AIRMONT_NP, NO_SSB | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT), 1170 1171 VULNWL_INTEL(INTEL_ATOM_GOLDMONT, NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO), 1172 VULNWL_INTEL(INTEL_ATOM_GOLDMONT_D, NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO), 1173 VULNWL_INTEL(INTEL_ATOM_GOLDMONT_PLUS, NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO | NO_EIBRS_PBRSB), 1174 1175 /* 1176 * Technically, swapgs isn't serializing on AMD (despite it previously 1177 * being documented as such in the APM). But according to AMD, %gs is 1178 * updated non-speculatively, and the issuing of %gs-relative memory 1179 * operands will be blocked until the %gs update completes, which is 1180 * good enough for our purposes. 1181 */ 1182 1183 VULNWL_INTEL(INTEL_ATOM_TREMONT, NO_EIBRS_PBRSB), 1184 VULNWL_INTEL(INTEL_ATOM_TREMONT_L, NO_EIBRS_PBRSB), 1185 VULNWL_INTEL(INTEL_ATOM_TREMONT_D, NO_ITLB_MULTIHIT | NO_EIBRS_PBRSB), 1186 1187 /* AMD Family 0xf - 0x12 */ 1188 VULNWL_AMD(0x0f, NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO | NO_BHI), 1189 VULNWL_AMD(0x10, NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO | NO_BHI), 1190 VULNWL_AMD(0x11, NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO | NO_BHI), 1191 VULNWL_AMD(0x12, NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO | NO_BHI), 1192 1193 /* FAMILY_ANY must be last, otherwise 0x0f - 0x12 matches won't work */ 1194 VULNWL_AMD(X86_FAMILY_ANY, NO_MELTDOWN | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO | NO_EIBRS_PBRSB | NO_BHI), 1195 VULNWL_HYGON(X86_FAMILY_ANY, NO_MELTDOWN | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO | NO_EIBRS_PBRSB | NO_BHI), 1196 1197 /* Zhaoxin Family 7 */ 1198 VULNWL(CENTAUR, 7, X86_MODEL_ANY, NO_SPECTRE_V2 | NO_SWAPGS | NO_MMIO | NO_BHI), 1199 VULNWL(ZHAOXIN, 7, X86_MODEL_ANY, NO_SPECTRE_V2 | NO_SWAPGS | NO_MMIO | NO_BHI), 1200 {} 1201 }; 1202 1203 #define VULNBL(vendor, family, model, blacklist) \ 1204 X86_MATCH_VENDOR_FAM_MODEL(vendor, family, model, blacklist) 1205 1206 #define VULNBL_INTEL_STEPS(vfm, max_stepping, issues) \ 1207 X86_MATCH_VFM_STEPS(vfm, X86_STEP_MIN, max_stepping, issues) 1208 1209 #define VULNBL_INTEL_TYPE(vfm, cpu_type, issues) \ 1210 X86_MATCH_VFM_CPU_TYPE(vfm, INTEL_CPU_TYPE_##cpu_type, issues) 1211 1212 #define VULNBL_AMD(family, blacklist) \ 1213 VULNBL(AMD, family, X86_MODEL_ANY, blacklist) 1214 1215 #define VULNBL_HYGON(family, blacklist) \ 1216 VULNBL(HYGON, family, X86_MODEL_ANY, blacklist) 1217 1218 #define SRBDS BIT(0) 1219 /* CPU is affected by X86_BUG_MMIO_STALE_DATA */ 1220 #define MMIO BIT(1) 1221 /* CPU is affected by Shared Buffers Data Sampling (SBDS), a variant of X86_BUG_MMIO_STALE_DATA */ 1222 #define MMIO_SBDS BIT(2) 1223 /* CPU is affected by RETbleed, speculating where you would not expect it */ 1224 #define RETBLEED BIT(3) 1225 /* CPU is affected by SMT (cross-thread) return predictions */ 1226 #define SMT_RSB BIT(4) 1227 /* CPU is affected by SRSO */ 1228 #define SRSO BIT(5) 1229 /* CPU is affected by GDS */ 1230 #define GDS BIT(6) 1231 /* CPU is affected by Register File Data Sampling */ 1232 #define RFDS BIT(7) 1233 /* CPU is affected by Indirect Target Selection */ 1234 #define ITS BIT(8) 1235 /* CPU is affected by Indirect Target Selection, but guest-host isolation is not affected */ 1236 #define ITS_NATIVE_ONLY BIT(9) 1237 /* CPU is affected by Transient Scheduler Attacks */ 1238 #define TSA BIT(10) 1239 1240 static const struct x86_cpu_id cpu_vuln_blacklist[] __initconst = { 1241 VULNBL_INTEL_STEPS(INTEL_IVYBRIDGE, X86_STEP_MAX, SRBDS), 1242 VULNBL_INTEL_STEPS(INTEL_HASWELL, X86_STEP_MAX, SRBDS), 1243 VULNBL_INTEL_STEPS(INTEL_HASWELL_L, X86_STEP_MAX, SRBDS), 1244 VULNBL_INTEL_STEPS(INTEL_HASWELL_G, X86_STEP_MAX, SRBDS), 1245 VULNBL_INTEL_STEPS(INTEL_HASWELL_X, X86_STEP_MAX, MMIO), 1246 VULNBL_INTEL_STEPS(INTEL_BROADWELL_D, X86_STEP_MAX, MMIO), 1247 VULNBL_INTEL_STEPS(INTEL_BROADWELL_G, X86_STEP_MAX, SRBDS), 1248 VULNBL_INTEL_STEPS(INTEL_BROADWELL_X, X86_STEP_MAX, MMIO), 1249 VULNBL_INTEL_STEPS(INTEL_BROADWELL, X86_STEP_MAX, SRBDS), 1250 VULNBL_INTEL_STEPS(INTEL_SKYLAKE_X, 0x5, MMIO | RETBLEED | GDS), 1251 VULNBL_INTEL_STEPS(INTEL_SKYLAKE_X, X86_STEP_MAX, MMIO | RETBLEED | GDS | ITS), 1252 VULNBL_INTEL_STEPS(INTEL_SKYLAKE_L, X86_STEP_MAX, MMIO | RETBLEED | GDS | SRBDS), 1253 VULNBL_INTEL_STEPS(INTEL_SKYLAKE, X86_STEP_MAX, MMIO | RETBLEED | GDS | SRBDS), 1254 VULNBL_INTEL_STEPS(INTEL_KABYLAKE_L, 0xb, MMIO | RETBLEED | GDS | SRBDS), 1255 VULNBL_INTEL_STEPS(INTEL_KABYLAKE_L, X86_STEP_MAX, MMIO | RETBLEED | GDS | SRBDS | ITS), 1256 VULNBL_INTEL_STEPS(INTEL_KABYLAKE, 0xc, MMIO | RETBLEED | GDS | SRBDS), 1257 VULNBL_INTEL_STEPS(INTEL_KABYLAKE, X86_STEP_MAX, MMIO | RETBLEED | GDS | SRBDS | ITS), 1258 VULNBL_INTEL_STEPS(INTEL_CANNONLAKE_L, X86_STEP_MAX, RETBLEED), 1259 VULNBL_INTEL_STEPS(INTEL_ICELAKE_L, X86_STEP_MAX, MMIO | MMIO_SBDS | RETBLEED | GDS | ITS | ITS_NATIVE_ONLY), 1260 VULNBL_INTEL_STEPS(INTEL_ICELAKE_D, X86_STEP_MAX, MMIO | GDS | ITS | ITS_NATIVE_ONLY), 1261 VULNBL_INTEL_STEPS(INTEL_ICELAKE_X, X86_STEP_MAX, MMIO | GDS | ITS | ITS_NATIVE_ONLY), 1262 VULNBL_INTEL_STEPS(INTEL_COMETLAKE, X86_STEP_MAX, MMIO | MMIO_SBDS | RETBLEED | GDS | ITS), 1263 VULNBL_INTEL_STEPS(INTEL_COMETLAKE_L, 0x0, MMIO | RETBLEED | ITS), 1264 VULNBL_INTEL_STEPS(INTEL_COMETLAKE_L, X86_STEP_MAX, MMIO | MMIO_SBDS | RETBLEED | GDS | ITS), 1265 VULNBL_INTEL_STEPS(INTEL_TIGERLAKE_L, X86_STEP_MAX, GDS | ITS | ITS_NATIVE_ONLY), 1266 VULNBL_INTEL_STEPS(INTEL_TIGERLAKE, X86_STEP_MAX, GDS | ITS | ITS_NATIVE_ONLY), 1267 VULNBL_INTEL_STEPS(INTEL_LAKEFIELD, X86_STEP_MAX, MMIO | MMIO_SBDS | RETBLEED), 1268 VULNBL_INTEL_STEPS(INTEL_ROCKETLAKE, X86_STEP_MAX, MMIO | RETBLEED | GDS | ITS | ITS_NATIVE_ONLY), 1269 VULNBL_INTEL_TYPE(INTEL_ALDERLAKE, ATOM, RFDS), 1270 VULNBL_INTEL_STEPS(INTEL_ALDERLAKE_L, X86_STEP_MAX, RFDS), 1271 VULNBL_INTEL_TYPE(INTEL_RAPTORLAKE, ATOM, RFDS), 1272 VULNBL_INTEL_STEPS(INTEL_RAPTORLAKE_P, X86_STEP_MAX, RFDS), 1273 VULNBL_INTEL_STEPS(INTEL_RAPTORLAKE_S, X86_STEP_MAX, RFDS), 1274 VULNBL_INTEL_STEPS(INTEL_ATOM_GRACEMONT, X86_STEP_MAX, RFDS), 1275 VULNBL_INTEL_STEPS(INTEL_ATOM_TREMONT, X86_STEP_MAX, MMIO | MMIO_SBDS | RFDS), 1276 VULNBL_INTEL_STEPS(INTEL_ATOM_TREMONT_D, X86_STEP_MAX, MMIO | RFDS), 1277 VULNBL_INTEL_STEPS(INTEL_ATOM_TREMONT_L, X86_STEP_MAX, MMIO | MMIO_SBDS | RFDS), 1278 VULNBL_INTEL_STEPS(INTEL_ATOM_GOLDMONT, X86_STEP_MAX, RFDS), 1279 VULNBL_INTEL_STEPS(INTEL_ATOM_GOLDMONT_D, X86_STEP_MAX, RFDS), 1280 VULNBL_INTEL_STEPS(INTEL_ATOM_GOLDMONT_PLUS, X86_STEP_MAX, RFDS), 1281 1282 VULNBL_AMD(0x15, RETBLEED), 1283 VULNBL_AMD(0x16, RETBLEED), 1284 VULNBL_AMD(0x17, RETBLEED | SMT_RSB | SRSO), 1285 VULNBL_HYGON(0x18, RETBLEED | SMT_RSB | SRSO), 1286 VULNBL_AMD(0x19, SRSO | TSA), 1287 VULNBL_AMD(0x1a, SRSO), 1288 {} 1289 }; 1290 1291 static bool __init cpu_matches(const struct x86_cpu_id *table, unsigned long which) 1292 { 1293 const struct x86_cpu_id *m = x86_match_cpu(table); 1294 1295 return m && !!(m->driver_data & which); 1296 } 1297 1298 u64 x86_read_arch_cap_msr(void) 1299 { 1300 u64 x86_arch_cap_msr = 0; 1301 1302 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES)) 1303 rdmsrq(MSR_IA32_ARCH_CAPABILITIES, x86_arch_cap_msr); 1304 1305 return x86_arch_cap_msr; 1306 } 1307 1308 static bool arch_cap_mmio_immune(u64 x86_arch_cap_msr) 1309 { 1310 return (x86_arch_cap_msr & ARCH_CAP_FBSDP_NO && 1311 x86_arch_cap_msr & ARCH_CAP_PSDP_NO && 1312 x86_arch_cap_msr & ARCH_CAP_SBDR_SSDP_NO); 1313 } 1314 1315 static bool __init vulnerable_to_rfds(u64 x86_arch_cap_msr) 1316 { 1317 /* The "immunity" bit trumps everything else: */ 1318 if (x86_arch_cap_msr & ARCH_CAP_RFDS_NO) 1319 return false; 1320 1321 /* 1322 * VMMs set ARCH_CAP_RFDS_CLEAR for processors not in the blacklist to 1323 * indicate that mitigation is needed because guest is running on a 1324 * vulnerable hardware or may migrate to such hardware: 1325 */ 1326 if (x86_arch_cap_msr & ARCH_CAP_RFDS_CLEAR) 1327 return true; 1328 1329 /* Only consult the blacklist when there is no enumeration: */ 1330 return cpu_matches(cpu_vuln_blacklist, RFDS); 1331 } 1332 1333 static bool __init vulnerable_to_its(u64 x86_arch_cap_msr) 1334 { 1335 /* The "immunity" bit trumps everything else: */ 1336 if (x86_arch_cap_msr & ARCH_CAP_ITS_NO) 1337 return false; 1338 if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL) 1339 return false; 1340 1341 /* None of the affected CPUs have BHI_CTRL */ 1342 if (boot_cpu_has(X86_FEATURE_BHI_CTRL)) 1343 return false; 1344 1345 /* 1346 * If a VMM did not expose ITS_NO, assume that a guest could 1347 * be running on a vulnerable hardware or may migrate to such 1348 * hardware. 1349 */ 1350 if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) 1351 return true; 1352 1353 if (cpu_matches(cpu_vuln_blacklist, ITS)) 1354 return true; 1355 1356 return false; 1357 } 1358 1359 static struct x86_cpu_id cpu_latest_microcode[] = { 1360 #include "microcode/intel-ucode-defs.h" 1361 {} 1362 }; 1363 1364 static bool __init cpu_has_old_microcode(void) 1365 { 1366 const struct x86_cpu_id *m = x86_match_cpu(cpu_latest_microcode); 1367 1368 /* Give unknown CPUs a pass: */ 1369 if (!m) { 1370 /* Intel CPUs should be in the list. Warn if not: */ 1371 if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) 1372 pr_info("x86/CPU: Model not found in latest microcode list\n"); 1373 return false; 1374 } 1375 1376 /* 1377 * Hosts usually lie to guests with a super high microcode 1378 * version. Just ignore what hosts tell guests: 1379 */ 1380 if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) 1381 return false; 1382 1383 /* Consider all debug microcode to be old: */ 1384 if (boot_cpu_data.microcode & BIT(31)) 1385 return true; 1386 1387 /* Give new microcode a pass: */ 1388 if (boot_cpu_data.microcode >= m->driver_data) 1389 return false; 1390 1391 /* Uh oh, too old: */ 1392 return true; 1393 } 1394 1395 static void __init cpu_set_bug_bits(struct cpuinfo_x86 *c) 1396 { 1397 u64 x86_arch_cap_msr = x86_read_arch_cap_msr(); 1398 1399 if (cpu_has_old_microcode()) { 1400 pr_warn("x86/CPU: Running old microcode\n"); 1401 setup_force_cpu_bug(X86_BUG_OLD_MICROCODE); 1402 add_taint(TAINT_CPU_OUT_OF_SPEC, LOCKDEP_STILL_OK); 1403 } 1404 1405 /* Set ITLB_MULTIHIT bug if cpu is not in the whitelist and not mitigated */ 1406 if (!cpu_matches(cpu_vuln_whitelist, NO_ITLB_MULTIHIT) && 1407 !(x86_arch_cap_msr & ARCH_CAP_PSCHANGE_MC_NO)) 1408 setup_force_cpu_bug(X86_BUG_ITLB_MULTIHIT); 1409 1410 if (cpu_matches(cpu_vuln_whitelist, NO_SPECULATION)) 1411 return; 1412 1413 setup_force_cpu_bug(X86_BUG_SPECTRE_V1); 1414 1415 if (!cpu_matches(cpu_vuln_whitelist, NO_SPECTRE_V2)) { 1416 setup_force_cpu_bug(X86_BUG_SPECTRE_V2); 1417 setup_force_cpu_bug(X86_BUG_SPECTRE_V2_USER); 1418 } 1419 1420 if (!cpu_matches(cpu_vuln_whitelist, NO_SSB) && 1421 !(x86_arch_cap_msr & ARCH_CAP_SSB_NO) && 1422 !cpu_has(c, X86_FEATURE_AMD_SSB_NO)) 1423 setup_force_cpu_bug(X86_BUG_SPEC_STORE_BYPASS); 1424 1425 /* 1426 * AMD's AutoIBRS is equivalent to Intel's eIBRS - use the Intel feature 1427 * flag and protect from vendor-specific bugs via the whitelist. 1428 * 1429 * Don't use AutoIBRS when SNP is enabled because it degrades host 1430 * userspace indirect branch performance. 1431 */ 1432 if ((x86_arch_cap_msr & ARCH_CAP_IBRS_ALL) || 1433 (cpu_has(c, X86_FEATURE_AUTOIBRS) && 1434 !cpu_feature_enabled(X86_FEATURE_SEV_SNP))) { 1435 setup_force_cpu_cap(X86_FEATURE_IBRS_ENHANCED); 1436 if (!cpu_matches(cpu_vuln_whitelist, NO_EIBRS_PBRSB) && 1437 !(x86_arch_cap_msr & ARCH_CAP_PBRSB_NO)) 1438 setup_force_cpu_bug(X86_BUG_EIBRS_PBRSB); 1439 } 1440 1441 if (!cpu_matches(cpu_vuln_whitelist, NO_MDS) && 1442 !(x86_arch_cap_msr & ARCH_CAP_MDS_NO)) { 1443 setup_force_cpu_bug(X86_BUG_MDS); 1444 if (cpu_matches(cpu_vuln_whitelist, MSBDS_ONLY)) 1445 setup_force_cpu_bug(X86_BUG_MSBDS_ONLY); 1446 } 1447 1448 if (!cpu_matches(cpu_vuln_whitelist, NO_SWAPGS)) 1449 setup_force_cpu_bug(X86_BUG_SWAPGS); 1450 1451 /* 1452 * When the CPU is not mitigated for TAA (TAA_NO=0) set TAA bug when: 1453 * - TSX is supported or 1454 * - TSX_CTRL is present 1455 * 1456 * TSX_CTRL check is needed for cases when TSX could be disabled before 1457 * the kernel boot e.g. kexec. 1458 * TSX_CTRL check alone is not sufficient for cases when the microcode 1459 * update is not present or running as guest that don't get TSX_CTRL. 1460 */ 1461 if (!(x86_arch_cap_msr & ARCH_CAP_TAA_NO) && 1462 (cpu_has(c, X86_FEATURE_RTM) || 1463 (x86_arch_cap_msr & ARCH_CAP_TSX_CTRL_MSR))) 1464 setup_force_cpu_bug(X86_BUG_TAA); 1465 1466 /* 1467 * SRBDS affects CPUs which support RDRAND or RDSEED and are listed 1468 * in the vulnerability blacklist. 1469 * 1470 * Some of the implications and mitigation of Shared Buffers Data 1471 * Sampling (SBDS) are similar to SRBDS. Give SBDS same treatment as 1472 * SRBDS. 1473 */ 1474 if ((cpu_has(c, X86_FEATURE_RDRAND) || 1475 cpu_has(c, X86_FEATURE_RDSEED)) && 1476 cpu_matches(cpu_vuln_blacklist, SRBDS | MMIO_SBDS)) 1477 setup_force_cpu_bug(X86_BUG_SRBDS); 1478 1479 /* 1480 * Processor MMIO Stale Data bug enumeration 1481 * 1482 * Affected CPU list is generally enough to enumerate the vulnerability, 1483 * but for virtualization case check for ARCH_CAP MSR bits also, VMM may 1484 * not want the guest to enumerate the bug. 1485 */ 1486 if (!arch_cap_mmio_immune(x86_arch_cap_msr)) { 1487 if (cpu_matches(cpu_vuln_blacklist, MMIO)) 1488 setup_force_cpu_bug(X86_BUG_MMIO_STALE_DATA); 1489 } 1490 1491 if (!cpu_has(c, X86_FEATURE_BTC_NO)) { 1492 if (cpu_matches(cpu_vuln_blacklist, RETBLEED) || (x86_arch_cap_msr & ARCH_CAP_RSBA)) 1493 setup_force_cpu_bug(X86_BUG_RETBLEED); 1494 } 1495 1496 if (cpu_matches(cpu_vuln_blacklist, SMT_RSB)) 1497 setup_force_cpu_bug(X86_BUG_SMT_RSB); 1498 1499 if (!cpu_has(c, X86_FEATURE_SRSO_NO)) { 1500 if (cpu_matches(cpu_vuln_blacklist, SRSO)) 1501 setup_force_cpu_bug(X86_BUG_SRSO); 1502 } 1503 1504 /* 1505 * Check if CPU is vulnerable to GDS. If running in a virtual machine on 1506 * an affected processor, the VMM may have disabled the use of GATHER by 1507 * disabling AVX2. The only way to do this in HW is to clear XCR0[2], 1508 * which means that AVX will be disabled. 1509 */ 1510 if (cpu_matches(cpu_vuln_blacklist, GDS) && !(x86_arch_cap_msr & ARCH_CAP_GDS_NO) && 1511 boot_cpu_has(X86_FEATURE_AVX)) 1512 setup_force_cpu_bug(X86_BUG_GDS); 1513 1514 if (vulnerable_to_rfds(x86_arch_cap_msr)) 1515 setup_force_cpu_bug(X86_BUG_RFDS); 1516 1517 /* 1518 * Intel parts with eIBRS are vulnerable to BHI attacks. Parts with 1519 * BHI_NO still need to use the BHI mitigation to prevent Intra-mode 1520 * attacks. When virtualized, eIBRS could be hidden, assume vulnerable. 1521 */ 1522 if (!cpu_matches(cpu_vuln_whitelist, NO_BHI) && 1523 (boot_cpu_has(X86_FEATURE_IBRS_ENHANCED) || 1524 boot_cpu_has(X86_FEATURE_HYPERVISOR))) 1525 setup_force_cpu_bug(X86_BUG_BHI); 1526 1527 if (cpu_has(c, X86_FEATURE_AMD_IBPB) && !cpu_has(c, X86_FEATURE_AMD_IBPB_RET)) 1528 setup_force_cpu_bug(X86_BUG_IBPB_NO_RET); 1529 1530 if (vulnerable_to_its(x86_arch_cap_msr)) { 1531 setup_force_cpu_bug(X86_BUG_ITS); 1532 if (cpu_matches(cpu_vuln_blacklist, ITS_NATIVE_ONLY)) 1533 setup_force_cpu_bug(X86_BUG_ITS_NATIVE_ONLY); 1534 } 1535 1536 if (c->x86_vendor == X86_VENDOR_AMD) { 1537 if (!cpu_has(c, X86_FEATURE_TSA_SQ_NO) || 1538 !cpu_has(c, X86_FEATURE_TSA_L1_NO)) { 1539 if (cpu_matches(cpu_vuln_blacklist, TSA) || 1540 /* Enable bug on Zen guests to allow for live migration. */ 1541 (cpu_has(c, X86_FEATURE_HYPERVISOR) && cpu_has(c, X86_FEATURE_ZEN))) 1542 setup_force_cpu_bug(X86_BUG_TSA); 1543 } 1544 } 1545 1546 if (cpu_matches(cpu_vuln_whitelist, NO_MELTDOWN)) 1547 return; 1548 1549 /* Rogue Data Cache Load? No! */ 1550 if (x86_arch_cap_msr & ARCH_CAP_RDCL_NO) 1551 return; 1552 1553 setup_force_cpu_bug(X86_BUG_CPU_MELTDOWN); 1554 1555 if (cpu_matches(cpu_vuln_whitelist, NO_L1TF)) 1556 return; 1557 1558 setup_force_cpu_bug(X86_BUG_L1TF); 1559 } 1560 1561 /* 1562 * The NOPL instruction is supposed to exist on all CPUs of family >= 6; 1563 * unfortunately, that's not true in practice because of early VIA 1564 * chips and (more importantly) broken virtualizers that are not easy 1565 * to detect. In the latter case it doesn't even *fail* reliably, so 1566 * probing for it doesn't even work. Disable it completely on 32-bit 1567 * unless we can find a reliable way to detect all the broken cases. 1568 * Enable it explicitly on 64-bit for non-constant inputs of cpu_has(). 1569 */ 1570 static void detect_nopl(void) 1571 { 1572 #ifdef CONFIG_X86_32 1573 setup_clear_cpu_cap(X86_FEATURE_NOPL); 1574 #else 1575 setup_force_cpu_cap(X86_FEATURE_NOPL); 1576 #endif 1577 } 1578 1579 static inline bool parse_set_clear_cpuid(char *arg, bool set) 1580 { 1581 char *opt; 1582 int taint = 0; 1583 1584 while (arg) { 1585 bool found __maybe_unused = false; 1586 unsigned int bit; 1587 1588 opt = strsep(&arg, ","); 1589 1590 /* 1591 * Handle naked numbers first for feature flags which don't 1592 * have names. It doesn't make sense for a bug not to have a 1593 * name so don't handle bug flags here. 1594 */ 1595 if (!kstrtouint(opt, 10, &bit)) { 1596 if (bit < NCAPINTS * 32) { 1597 1598 if (set) { 1599 pr_warn("setcpuid: force-enabling CPU feature flag:"); 1600 setup_force_cpu_cap(bit); 1601 } else { 1602 pr_warn("clearcpuid: force-disabling CPU feature flag:"); 1603 setup_clear_cpu_cap(bit); 1604 } 1605 /* empty-string, i.e., ""-defined feature flags */ 1606 if (!x86_cap_flags[bit]) 1607 pr_cont(" %d:%d\n", bit >> 5, bit & 31); 1608 else 1609 pr_cont(" %s\n", x86_cap_flags[bit]); 1610 1611 taint++; 1612 } 1613 /* 1614 * The assumption is that there are no feature names with only 1615 * numbers in the name thus go to the next argument. 1616 */ 1617 continue; 1618 } 1619 1620 for (bit = 0; bit < 32 * (NCAPINTS + NBUGINTS); bit++) { 1621 const char *flag; 1622 const char *kind; 1623 1624 if (bit < 32 * NCAPINTS) { 1625 flag = x86_cap_flags[bit]; 1626 kind = "feature"; 1627 } else { 1628 kind = "bug"; 1629 flag = x86_bug_flags[bit - (32 * NCAPINTS)]; 1630 } 1631 1632 if (!flag) 1633 continue; 1634 1635 if (strcmp(flag, opt)) 1636 continue; 1637 1638 if (set) { 1639 pr_warn("setcpuid: force-enabling CPU %s flag: %s\n", 1640 kind, flag); 1641 setup_force_cpu_cap(bit); 1642 } else { 1643 pr_warn("clearcpuid: force-disabling CPU %s flag: %s\n", 1644 kind, flag); 1645 setup_clear_cpu_cap(bit); 1646 } 1647 taint++; 1648 found = true; 1649 break; 1650 } 1651 1652 if (!found) 1653 pr_warn("%s: unknown CPU flag: %s", set ? "setcpuid" : "clearcpuid", opt); 1654 } 1655 1656 return taint; 1657 } 1658 1659 1660 /* 1661 * We parse cpu parameters early because fpu__init_system() is executed 1662 * before parse_early_param(). 1663 */ 1664 static void __init cpu_parse_early_param(void) 1665 { 1666 bool cpuid_taint = false; 1667 char arg[128]; 1668 int arglen; 1669 1670 #ifdef CONFIG_X86_32 1671 if (cmdline_find_option_bool(boot_command_line, "no387")) 1672 #ifdef CONFIG_MATH_EMULATION 1673 setup_clear_cpu_cap(X86_FEATURE_FPU); 1674 #else 1675 pr_err("Option 'no387' required CONFIG_MATH_EMULATION enabled.\n"); 1676 #endif 1677 1678 if (cmdline_find_option_bool(boot_command_line, "nofxsr")) 1679 setup_clear_cpu_cap(X86_FEATURE_FXSR); 1680 #endif 1681 1682 if (cmdline_find_option_bool(boot_command_line, "noxsave")) 1683 setup_clear_cpu_cap(X86_FEATURE_XSAVE); 1684 1685 if (cmdline_find_option_bool(boot_command_line, "noxsaveopt")) 1686 setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT); 1687 1688 if (cmdline_find_option_bool(boot_command_line, "noxsaves")) 1689 setup_clear_cpu_cap(X86_FEATURE_XSAVES); 1690 1691 if (cmdline_find_option_bool(boot_command_line, "nousershstk")) 1692 setup_clear_cpu_cap(X86_FEATURE_USER_SHSTK); 1693 1694 /* Minimize the gap between FRED is available and available but disabled. */ 1695 arglen = cmdline_find_option(boot_command_line, "fred", arg, sizeof(arg)); 1696 if (arglen != 2 || strncmp(arg, "on", 2)) 1697 setup_clear_cpu_cap(X86_FEATURE_FRED); 1698 1699 arglen = cmdline_find_option(boot_command_line, "clearcpuid", arg, sizeof(arg)); 1700 if (arglen > 0) 1701 cpuid_taint |= parse_set_clear_cpuid(arg, false); 1702 1703 arglen = cmdline_find_option(boot_command_line, "setcpuid", arg, sizeof(arg)); 1704 if (arglen > 0) 1705 cpuid_taint |= parse_set_clear_cpuid(arg, true); 1706 1707 if (cpuid_taint) { 1708 pr_warn("!!! setcpuid=/clearcpuid= in use, this is for TESTING ONLY, may break things horribly. Tainting kernel.\n"); 1709 add_taint(TAINT_CPU_OUT_OF_SPEC, LOCKDEP_STILL_OK); 1710 } 1711 } 1712 1713 /* 1714 * Do minimum CPU detection early. 1715 * Fields really needed: vendor, cpuid_level, family, model, mask, 1716 * cache alignment. 1717 * The others are not touched to avoid unwanted side effects. 1718 * 1719 * WARNING: this function is only called on the boot CPU. Don't add code 1720 * here that is supposed to run on all CPUs. 1721 */ 1722 static void __init early_identify_cpu(struct cpuinfo_x86 *c) 1723 { 1724 memset(&c->x86_capability, 0, sizeof(c->x86_capability)); 1725 c->extended_cpuid_level = 0; 1726 1727 if (!cpuid_feature()) 1728 identify_cpu_without_cpuid(c); 1729 1730 /* cyrix could have cpuid enabled via c_identify()*/ 1731 if (cpuid_feature()) { 1732 cpu_detect(c); 1733 get_cpu_vendor(c); 1734 intel_unlock_cpuid_leafs(c); 1735 get_cpu_cap(c); 1736 setup_force_cpu_cap(X86_FEATURE_CPUID); 1737 get_cpu_address_sizes(c); 1738 cpu_parse_early_param(); 1739 1740 cpu_init_topology(c); 1741 1742 if (this_cpu->c_early_init) 1743 this_cpu->c_early_init(c); 1744 1745 c->cpu_index = 0; 1746 filter_cpuid_features(c, false); 1747 check_cpufeature_deps(c); 1748 1749 if (this_cpu->c_bsp_init) 1750 this_cpu->c_bsp_init(c); 1751 } else { 1752 setup_clear_cpu_cap(X86_FEATURE_CPUID); 1753 get_cpu_address_sizes(c); 1754 cpu_init_topology(c); 1755 } 1756 1757 setup_force_cpu_cap(X86_FEATURE_ALWAYS); 1758 1759 cpu_set_bug_bits(c); 1760 1761 sld_setup(c); 1762 1763 #ifdef CONFIG_X86_32 1764 /* 1765 * Regardless of whether PCID is enumerated, the SDM says 1766 * that it can't be enabled in 32-bit mode. 1767 */ 1768 setup_clear_cpu_cap(X86_FEATURE_PCID); 1769 #endif 1770 1771 /* 1772 * Later in the boot process pgtable_l5_enabled() relies on 1773 * cpu_feature_enabled(X86_FEATURE_LA57). If 5-level paging is not 1774 * enabled by this point we need to clear the feature bit to avoid 1775 * false-positives at the later stage. 1776 * 1777 * pgtable_l5_enabled() can be false here for several reasons: 1778 * - 5-level paging is disabled compile-time; 1779 * - it's 32-bit kernel; 1780 * - machine doesn't support 5-level paging; 1781 * - user specified 'no5lvl' in kernel command line. 1782 */ 1783 if (!pgtable_l5_enabled()) 1784 setup_clear_cpu_cap(X86_FEATURE_LA57); 1785 1786 detect_nopl(); 1787 } 1788 1789 void __init init_cpu_devs(void) 1790 { 1791 const struct cpu_dev *const *cdev; 1792 int count = 0; 1793 1794 for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) { 1795 const struct cpu_dev *cpudev = *cdev; 1796 1797 if (count >= X86_VENDOR_NUM) 1798 break; 1799 cpu_devs[count] = cpudev; 1800 count++; 1801 } 1802 } 1803 1804 void __init early_cpu_init(void) 1805 { 1806 #ifdef CONFIG_PROCESSOR_SELECT 1807 unsigned int i, j; 1808 1809 pr_info("KERNEL supported cpus:\n"); 1810 #endif 1811 1812 init_cpu_devs(); 1813 1814 #ifdef CONFIG_PROCESSOR_SELECT 1815 for (i = 0; i < X86_VENDOR_NUM && cpu_devs[i]; i++) { 1816 for (j = 0; j < 2; j++) { 1817 if (!cpu_devs[i]->c_ident[j]) 1818 continue; 1819 pr_info(" %s %s\n", cpu_devs[i]->c_vendor, 1820 cpu_devs[i]->c_ident[j]); 1821 } 1822 } 1823 #endif 1824 1825 early_identify_cpu(&boot_cpu_data); 1826 } 1827 1828 static bool detect_null_seg_behavior(void) 1829 { 1830 /* 1831 * Empirically, writing zero to a segment selector on AMD does 1832 * not clear the base, whereas writing zero to a segment 1833 * selector on Intel does clear the base. Intel's behavior 1834 * allows slightly faster context switches in the common case 1835 * where GS is unused by the prev and next threads. 1836 * 1837 * Since neither vendor documents this anywhere that I can see, 1838 * detect it directly instead of hard-coding the choice by 1839 * vendor. 1840 * 1841 * I've designated AMD's behavior as the "bug" because it's 1842 * counterintuitive and less friendly. 1843 */ 1844 1845 unsigned long old_base, tmp; 1846 rdmsrq(MSR_FS_BASE, old_base); 1847 wrmsrq(MSR_FS_BASE, 1); 1848 loadsegment(fs, 0); 1849 rdmsrq(MSR_FS_BASE, tmp); 1850 wrmsrq(MSR_FS_BASE, old_base); 1851 return tmp == 0; 1852 } 1853 1854 void check_null_seg_clears_base(struct cpuinfo_x86 *c) 1855 { 1856 /* BUG_NULL_SEG is only relevant with 64bit userspace */ 1857 if (!IS_ENABLED(CONFIG_X86_64)) 1858 return; 1859 1860 if (cpu_has(c, X86_FEATURE_NULL_SEL_CLR_BASE)) 1861 return; 1862 1863 /* 1864 * CPUID bit above wasn't set. If this kernel is still running 1865 * as a HV guest, then the HV has decided not to advertize 1866 * that CPUID bit for whatever reason. For example, one 1867 * member of the migration pool might be vulnerable. Which 1868 * means, the bug is present: set the BUG flag and return. 1869 */ 1870 if (cpu_has(c, X86_FEATURE_HYPERVISOR)) { 1871 set_cpu_bug(c, X86_BUG_NULL_SEG); 1872 return; 1873 } 1874 1875 /* 1876 * Zen2 CPUs also have this behaviour, but no CPUID bit. 1877 * 0x18 is the respective family for Hygon. 1878 */ 1879 if ((c->x86 == 0x17 || c->x86 == 0x18) && 1880 detect_null_seg_behavior()) 1881 return; 1882 1883 /* All the remaining ones are affected */ 1884 set_cpu_bug(c, X86_BUG_NULL_SEG); 1885 } 1886 1887 static void generic_identify(struct cpuinfo_x86 *c) 1888 { 1889 c->extended_cpuid_level = 0; 1890 1891 if (!cpuid_feature()) 1892 identify_cpu_without_cpuid(c); 1893 1894 /* cyrix could have cpuid enabled via c_identify()*/ 1895 if (!cpuid_feature()) 1896 return; 1897 1898 cpu_detect(c); 1899 1900 get_cpu_vendor(c); 1901 intel_unlock_cpuid_leafs(c); 1902 get_cpu_cap(c); 1903 1904 get_cpu_address_sizes(c); 1905 1906 get_model_name(c); /* Default name */ 1907 1908 /* 1909 * ESPFIX is a strange bug. All real CPUs have it. Paravirt 1910 * systems that run Linux at CPL > 0 may or may not have the 1911 * issue, but, even if they have the issue, there's absolutely 1912 * nothing we can do about it because we can't use the real IRET 1913 * instruction. 1914 * 1915 * NB: For the time being, only 32-bit kernels support 1916 * X86_BUG_ESPFIX as such. 64-bit kernels directly choose 1917 * whether to apply espfix using paravirt hooks. If any 1918 * non-paravirt system ever shows up that does *not* have the 1919 * ESPFIX issue, we can change this. 1920 */ 1921 #ifdef CONFIG_X86_32 1922 set_cpu_bug(c, X86_BUG_ESPFIX); 1923 #endif 1924 } 1925 1926 /* 1927 * This does the hard work of actually picking apart the CPU stuff... 1928 */ 1929 static void identify_cpu(struct cpuinfo_x86 *c) 1930 { 1931 int i; 1932 1933 c->loops_per_jiffy = loops_per_jiffy; 1934 c->x86_cache_size = 0; 1935 c->x86_vendor = X86_VENDOR_UNKNOWN; 1936 c->x86_model = c->x86_stepping = 0; /* So far unknown... */ 1937 c->x86_vendor_id[0] = '\0'; /* Unset */ 1938 c->x86_model_id[0] = '\0'; /* Unset */ 1939 #ifdef CONFIG_X86_64 1940 c->x86_clflush_size = 64; 1941 c->x86_phys_bits = 36; 1942 c->x86_virt_bits = 48; 1943 #else 1944 c->cpuid_level = -1; /* CPUID not detected */ 1945 c->x86_clflush_size = 32; 1946 c->x86_phys_bits = 32; 1947 c->x86_virt_bits = 32; 1948 #endif 1949 c->x86_cache_alignment = c->x86_clflush_size; 1950 memset(&c->x86_capability, 0, sizeof(c->x86_capability)); 1951 #ifdef CONFIG_X86_VMX_FEATURE_NAMES 1952 memset(&c->vmx_capability, 0, sizeof(c->vmx_capability)); 1953 #endif 1954 1955 generic_identify(c); 1956 1957 cpu_parse_topology(c); 1958 1959 if (this_cpu->c_identify) 1960 this_cpu->c_identify(c); 1961 1962 /* Clear/Set all flags overridden by options, after probe */ 1963 apply_forced_caps(c); 1964 1965 /* 1966 * Set default APIC and TSC_DEADLINE MSR fencing flag. AMD and 1967 * Hygon will clear it in ->c_init() below. 1968 */ 1969 set_cpu_cap(c, X86_FEATURE_APIC_MSRS_FENCE); 1970 1971 /* 1972 * Vendor-specific initialization. In this section we 1973 * canonicalize the feature flags, meaning if there are 1974 * features a certain CPU supports which CPUID doesn't 1975 * tell us, CPUID claiming incorrect flags, or other bugs, 1976 * we handle them here. 1977 * 1978 * At the end of this section, c->x86_capability better 1979 * indicate the features this CPU genuinely supports! 1980 */ 1981 if (this_cpu->c_init) 1982 this_cpu->c_init(c); 1983 1984 bus_lock_init(); 1985 1986 /* Disable the PN if appropriate */ 1987 squash_the_stupid_serial_number(c); 1988 1989 /* Set up SMEP/SMAP/UMIP */ 1990 setup_smep(c); 1991 setup_smap(c); 1992 setup_umip(c); 1993 1994 /* Enable FSGSBASE instructions if available. */ 1995 if (cpu_has(c, X86_FEATURE_FSGSBASE)) { 1996 cr4_set_bits(X86_CR4_FSGSBASE); 1997 elf_hwcap2 |= HWCAP2_FSGSBASE; 1998 } 1999 2000 /* 2001 * The vendor-specific functions might have changed features. 2002 * Now we do "generic changes." 2003 */ 2004 2005 /* Filter out anything that depends on CPUID levels we don't have */ 2006 filter_cpuid_features(c, true); 2007 2008 /* Check for unmet dependencies based on the CPUID dependency table */ 2009 check_cpufeature_deps(c); 2010 2011 /* If the model name is still unset, do table lookup. */ 2012 if (!c->x86_model_id[0]) { 2013 const char *p; 2014 p = table_lookup_model(c); 2015 if (p) 2016 strcpy(c->x86_model_id, p); 2017 else 2018 /* Last resort... */ 2019 sprintf(c->x86_model_id, "%02x/%02x", 2020 c->x86, c->x86_model); 2021 } 2022 2023 x86_init_rdrand(c); 2024 setup_pku(c); 2025 setup_cet(c); 2026 2027 /* 2028 * Clear/Set all flags overridden by options, need do it 2029 * before following smp all cpus cap AND. 2030 */ 2031 apply_forced_caps(c); 2032 2033 /* 2034 * On SMP, boot_cpu_data holds the common feature set between 2035 * all CPUs; so make sure that we indicate which features are 2036 * common between the CPUs. The first time this routine gets 2037 * executed, c == &boot_cpu_data. 2038 */ 2039 if (c != &boot_cpu_data) { 2040 /* AND the already accumulated flags with these */ 2041 for (i = 0; i < NCAPINTS; i++) 2042 boot_cpu_data.x86_capability[i] &= c->x86_capability[i]; 2043 2044 /* OR, i.e. replicate the bug flags */ 2045 for (i = NCAPINTS; i < NCAPINTS + NBUGINTS; i++) 2046 c->x86_capability[i] |= boot_cpu_data.x86_capability[i]; 2047 } 2048 2049 ppin_init(c); 2050 2051 /* Init Machine Check Exception if available. */ 2052 mcheck_cpu_init(c); 2053 2054 numa_add_cpu(smp_processor_id()); 2055 } 2056 2057 /* 2058 * Set up the CPU state needed to execute SYSENTER/SYSEXIT instructions 2059 * on 32-bit kernels: 2060 */ 2061 #ifdef CONFIG_X86_32 2062 void enable_sep_cpu(void) 2063 { 2064 struct tss_struct *tss; 2065 int cpu; 2066 2067 if (!boot_cpu_has(X86_FEATURE_SEP)) 2068 return; 2069 2070 cpu = get_cpu(); 2071 tss = &per_cpu(cpu_tss_rw, cpu); 2072 2073 /* 2074 * We cache MSR_IA32_SYSENTER_CS's value in the TSS's ss1 field -- 2075 * see the big comment in struct x86_hw_tss's definition. 2076 */ 2077 2078 tss->x86_tss.ss1 = __KERNEL_CS; 2079 wrmsrq(MSR_IA32_SYSENTER_CS, tss->x86_tss.ss1); 2080 wrmsrq(MSR_IA32_SYSENTER_ESP, (unsigned long)(cpu_entry_stack(cpu) + 1)); 2081 wrmsrq(MSR_IA32_SYSENTER_EIP, (unsigned long)entry_SYSENTER_32); 2082 2083 put_cpu(); 2084 } 2085 #endif 2086 2087 static __init void identify_boot_cpu(void) 2088 { 2089 identify_cpu(&boot_cpu_data); 2090 if (HAS_KERNEL_IBT && cpu_feature_enabled(X86_FEATURE_IBT)) 2091 pr_info("CET detected: Indirect Branch Tracking enabled\n"); 2092 #ifdef CONFIG_X86_32 2093 enable_sep_cpu(); 2094 #endif 2095 cpu_detect_tlb(&boot_cpu_data); 2096 setup_cr_pinning(); 2097 2098 tsx_init(); 2099 tdx_init(); 2100 lkgs_init(); 2101 } 2102 2103 void identify_secondary_cpu(unsigned int cpu) 2104 { 2105 struct cpuinfo_x86 *c = &cpu_data(cpu); 2106 2107 /* Copy boot_cpu_data only on the first bringup */ 2108 if (!c->initialized) 2109 *c = boot_cpu_data; 2110 c->cpu_index = cpu; 2111 2112 identify_cpu(c); 2113 #ifdef CONFIG_X86_32 2114 enable_sep_cpu(); 2115 #endif 2116 x86_spec_ctrl_setup_ap(); 2117 update_srbds_msr(); 2118 if (boot_cpu_has_bug(X86_BUG_GDS)) 2119 update_gds_msr(); 2120 2121 tsx_ap_init(); 2122 c->initialized = true; 2123 } 2124 2125 void print_cpu_info(struct cpuinfo_x86 *c) 2126 { 2127 const char *vendor = NULL; 2128 2129 if (c->x86_vendor < X86_VENDOR_NUM) { 2130 vendor = this_cpu->c_vendor; 2131 } else { 2132 if (c->cpuid_level >= 0) 2133 vendor = c->x86_vendor_id; 2134 } 2135 2136 if (vendor && !strstr(c->x86_model_id, vendor)) 2137 pr_cont("%s ", vendor); 2138 2139 if (c->x86_model_id[0]) 2140 pr_cont("%s", c->x86_model_id); 2141 else 2142 pr_cont("%d86", c->x86); 2143 2144 pr_cont(" (family: 0x%x, model: 0x%x", c->x86, c->x86_model); 2145 2146 if (c->x86_stepping || c->cpuid_level >= 0) 2147 pr_cont(", stepping: 0x%x)\n", c->x86_stepping); 2148 else 2149 pr_cont(")\n"); 2150 } 2151 2152 /* 2153 * clearcpuid= and setcpuid= were already parsed in cpu_parse_early_param(). 2154 * These dummy functions prevent them from becoming an environment variable for 2155 * init. 2156 */ 2157 2158 static __init int setup_clearcpuid(char *arg) 2159 { 2160 return 1; 2161 } 2162 __setup("clearcpuid=", setup_clearcpuid); 2163 2164 static __init int setup_setcpuid(char *arg) 2165 { 2166 return 1; 2167 } 2168 __setup("setcpuid=", setup_setcpuid); 2169 2170 DEFINE_PER_CPU_CACHE_HOT(struct task_struct *, current_task) = &init_task; 2171 EXPORT_PER_CPU_SYMBOL(current_task); 2172 EXPORT_PER_CPU_SYMBOL(const_current_task); 2173 2174 DEFINE_PER_CPU_CACHE_HOT(int, __preempt_count) = INIT_PREEMPT_COUNT; 2175 EXPORT_PER_CPU_SYMBOL(__preempt_count); 2176 2177 DEFINE_PER_CPU_CACHE_HOT(unsigned long, cpu_current_top_of_stack) = TOP_OF_INIT_STACK; 2178 2179 #ifdef CONFIG_X86_64 2180 /* 2181 * Note: Do not make this dependant on CONFIG_MITIGATION_CALL_DEPTH_TRACKING 2182 * so that this space is reserved in the hot cache section even when the 2183 * mitigation is disabled. 2184 */ 2185 DEFINE_PER_CPU_CACHE_HOT(u64, __x86_call_depth); 2186 EXPORT_PER_CPU_SYMBOL(__x86_call_depth); 2187 2188 static void wrmsrq_cstar(unsigned long val) 2189 { 2190 /* 2191 * Intel CPUs do not support 32-bit SYSCALL. Writing to MSR_CSTAR 2192 * is so far ignored by the CPU, but raises a #VE trap in a TDX 2193 * guest. Avoid the pointless write on all Intel CPUs. 2194 */ 2195 if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL) 2196 wrmsrq(MSR_CSTAR, val); 2197 } 2198 2199 static inline void idt_syscall_init(void) 2200 { 2201 wrmsrq(MSR_LSTAR, (unsigned long)entry_SYSCALL_64); 2202 2203 if (ia32_enabled()) { 2204 wrmsrq_cstar((unsigned long)entry_SYSCALL_compat); 2205 /* 2206 * This only works on Intel CPUs. 2207 * On AMD CPUs these MSRs are 32-bit, CPU truncates MSR_IA32_SYSENTER_EIP. 2208 * This does not cause SYSENTER to jump to the wrong location, because 2209 * AMD doesn't allow SYSENTER in long mode (either 32- or 64-bit). 2210 */ 2211 wrmsrq_safe(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS); 2212 wrmsrq_safe(MSR_IA32_SYSENTER_ESP, 2213 (unsigned long)(cpu_entry_stack(smp_processor_id()) + 1)); 2214 wrmsrq_safe(MSR_IA32_SYSENTER_EIP, (u64)entry_SYSENTER_compat); 2215 } else { 2216 wrmsrq_cstar((unsigned long)entry_SYSCALL32_ignore); 2217 wrmsrq_safe(MSR_IA32_SYSENTER_CS, (u64)GDT_ENTRY_INVALID_SEG); 2218 wrmsrq_safe(MSR_IA32_SYSENTER_ESP, 0ULL); 2219 wrmsrq_safe(MSR_IA32_SYSENTER_EIP, 0ULL); 2220 } 2221 2222 /* 2223 * Flags to clear on syscall; clear as much as possible 2224 * to minimize user space-kernel interference. 2225 */ 2226 wrmsrq(MSR_SYSCALL_MASK, 2227 X86_EFLAGS_CF|X86_EFLAGS_PF|X86_EFLAGS_AF| 2228 X86_EFLAGS_ZF|X86_EFLAGS_SF|X86_EFLAGS_TF| 2229 X86_EFLAGS_IF|X86_EFLAGS_DF|X86_EFLAGS_OF| 2230 X86_EFLAGS_IOPL|X86_EFLAGS_NT|X86_EFLAGS_RF| 2231 X86_EFLAGS_AC|X86_EFLAGS_ID); 2232 } 2233 2234 /* May not be marked __init: used by software suspend */ 2235 void syscall_init(void) 2236 { 2237 /* The default user and kernel segments */ 2238 wrmsr(MSR_STAR, 0, (__USER32_CS << 16) | __KERNEL_CS); 2239 2240 /* 2241 * Except the IA32_STAR MSR, there is NO need to setup SYSCALL and 2242 * SYSENTER MSRs for FRED, because FRED uses the ring 3 FRED 2243 * entrypoint for SYSCALL and SYSENTER, and ERETU is the only legit 2244 * instruction to return to ring 3 (both sysexit and sysret cause 2245 * #UD when FRED is enabled). 2246 */ 2247 if (!cpu_feature_enabled(X86_FEATURE_FRED)) 2248 idt_syscall_init(); 2249 } 2250 #endif /* CONFIG_X86_64 */ 2251 2252 #ifdef CONFIG_STACKPROTECTOR 2253 DEFINE_PER_CPU_CACHE_HOT(unsigned long, __stack_chk_guard); 2254 #ifndef CONFIG_SMP 2255 EXPORT_PER_CPU_SYMBOL(__stack_chk_guard); 2256 #endif 2257 #endif 2258 2259 static void initialize_debug_regs(void) 2260 { 2261 /* Control register first -- to make sure everything is disabled. */ 2262 set_debugreg(DR7_FIXED_1, 7); 2263 set_debugreg(DR6_RESERVED, 6); 2264 /* dr5 and dr4 don't exist */ 2265 set_debugreg(0, 3); 2266 set_debugreg(0, 2); 2267 set_debugreg(0, 1); 2268 set_debugreg(0, 0); 2269 } 2270 2271 #ifdef CONFIG_KGDB 2272 /* 2273 * Restore debug regs if using kgdbwait and you have a kernel debugger 2274 * connection established. 2275 */ 2276 static void dbg_restore_debug_regs(void) 2277 { 2278 if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break)) 2279 arch_kgdb_ops.correct_hw_break(); 2280 } 2281 #else /* ! CONFIG_KGDB */ 2282 #define dbg_restore_debug_regs() 2283 #endif /* ! CONFIG_KGDB */ 2284 2285 static inline void setup_getcpu(int cpu) 2286 { 2287 unsigned long cpudata = vdso_encode_cpunode(cpu, early_cpu_to_node(cpu)); 2288 struct desc_struct d = { }; 2289 2290 if (boot_cpu_has(X86_FEATURE_RDTSCP) || boot_cpu_has(X86_FEATURE_RDPID)) 2291 wrmsrq(MSR_TSC_AUX, cpudata); 2292 2293 /* Store CPU and node number in limit. */ 2294 d.limit0 = cpudata; 2295 d.limit1 = cpudata >> 16; 2296 2297 d.type = 5; /* RO data, expand down, accessed */ 2298 d.dpl = 3; /* Visible to user code */ 2299 d.s = 1; /* Not a system segment */ 2300 d.p = 1; /* Present */ 2301 d.d = 1; /* 32-bit */ 2302 2303 write_gdt_entry(get_cpu_gdt_rw(cpu), GDT_ENTRY_CPUNODE, &d, DESCTYPE_S); 2304 } 2305 2306 #ifdef CONFIG_X86_64 2307 static inline void tss_setup_ist(struct tss_struct *tss) 2308 { 2309 /* Set up the per-CPU TSS IST stacks */ 2310 tss->x86_tss.ist[IST_INDEX_DF] = __this_cpu_ist_top_va(DF); 2311 tss->x86_tss.ist[IST_INDEX_NMI] = __this_cpu_ist_top_va(NMI); 2312 tss->x86_tss.ist[IST_INDEX_DB] = __this_cpu_ist_top_va(DB); 2313 tss->x86_tss.ist[IST_INDEX_MCE] = __this_cpu_ist_top_va(MCE); 2314 /* Only mapped when SEV-ES is active */ 2315 tss->x86_tss.ist[IST_INDEX_VC] = __this_cpu_ist_top_va(VC); 2316 } 2317 #else /* CONFIG_X86_64 */ 2318 static inline void tss_setup_ist(struct tss_struct *tss) { } 2319 #endif /* !CONFIG_X86_64 */ 2320 2321 static inline void tss_setup_io_bitmap(struct tss_struct *tss) 2322 { 2323 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET_INVALID; 2324 2325 #ifdef CONFIG_X86_IOPL_IOPERM 2326 tss->io_bitmap.prev_max = 0; 2327 tss->io_bitmap.prev_sequence = 0; 2328 memset(tss->io_bitmap.bitmap, 0xff, sizeof(tss->io_bitmap.bitmap)); 2329 /* 2330 * Invalidate the extra array entry past the end of the all 2331 * permission bitmap as required by the hardware. 2332 */ 2333 tss->io_bitmap.mapall[IO_BITMAP_LONGS] = ~0UL; 2334 #endif 2335 } 2336 2337 /* 2338 * Setup everything needed to handle exceptions from the IDT, including the IST 2339 * exceptions which use paranoid_entry(). 2340 */ 2341 void cpu_init_exception_handling(bool boot_cpu) 2342 { 2343 struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw); 2344 int cpu = raw_smp_processor_id(); 2345 2346 /* paranoid_entry() gets the CPU number from the GDT */ 2347 setup_getcpu(cpu); 2348 2349 /* For IDT mode, IST vectors need to be set in TSS. */ 2350 if (!cpu_feature_enabled(X86_FEATURE_FRED)) 2351 tss_setup_ist(tss); 2352 tss_setup_io_bitmap(tss); 2353 set_tss_desc(cpu, &get_cpu_entry_area(cpu)->tss.x86_tss); 2354 2355 load_TR_desc(); 2356 2357 /* GHCB needs to be setup to handle #VC. */ 2358 setup_ghcb(); 2359 2360 if (cpu_feature_enabled(X86_FEATURE_FRED)) { 2361 /* The boot CPU has enabled FRED during early boot */ 2362 if (!boot_cpu) 2363 cpu_init_fred_exceptions(); 2364 2365 cpu_init_fred_rsps(); 2366 } else { 2367 load_current_idt(); 2368 } 2369 } 2370 2371 void __init cpu_init_replace_early_idt(void) 2372 { 2373 if (cpu_feature_enabled(X86_FEATURE_FRED)) 2374 cpu_init_fred_exceptions(); 2375 else 2376 idt_setup_early_pf(); 2377 } 2378 2379 /* 2380 * cpu_init() initializes state that is per-CPU. Some data is already 2381 * initialized (naturally) in the bootstrap process, such as the GDT. We 2382 * reload it nevertheless, this function acts as a 'CPU state barrier', 2383 * nothing should get across. 2384 */ 2385 void cpu_init(void) 2386 { 2387 struct task_struct *cur = current; 2388 int cpu = raw_smp_processor_id(); 2389 2390 #ifdef CONFIG_NUMA 2391 if (this_cpu_read(numa_node) == 0 && 2392 early_cpu_to_node(cpu) != NUMA_NO_NODE) 2393 set_numa_node(early_cpu_to_node(cpu)); 2394 #endif 2395 pr_debug("Initializing CPU#%d\n", cpu); 2396 2397 if (IS_ENABLED(CONFIG_X86_64) || cpu_feature_enabled(X86_FEATURE_VME) || 2398 boot_cpu_has(X86_FEATURE_TSC) || boot_cpu_has(X86_FEATURE_DE)) 2399 cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE); 2400 2401 if (IS_ENABLED(CONFIG_X86_64)) { 2402 loadsegment(fs, 0); 2403 memset(cur->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8); 2404 syscall_init(); 2405 2406 wrmsrq(MSR_FS_BASE, 0); 2407 wrmsrq(MSR_KERNEL_GS_BASE, 0); 2408 barrier(); 2409 2410 x2apic_setup(); 2411 2412 intel_posted_msi_init(); 2413 } 2414 2415 mmgrab(&init_mm); 2416 cur->active_mm = &init_mm; 2417 BUG_ON(cur->mm); 2418 initialize_tlbstate_and_flush(); 2419 enter_lazy_tlb(&init_mm, cur); 2420 2421 /* 2422 * sp0 points to the entry trampoline stack regardless of what task 2423 * is running. 2424 */ 2425 load_sp0((unsigned long)(cpu_entry_stack(cpu) + 1)); 2426 2427 load_mm_ldt(&init_mm); 2428 2429 initialize_debug_regs(); 2430 dbg_restore_debug_regs(); 2431 2432 doublefault_init_cpu_tss(); 2433 2434 if (is_uv_system()) 2435 uv_cpu_init(); 2436 2437 load_fixmap_gdt(cpu); 2438 } 2439 2440 #ifdef CONFIG_MICROCODE_LATE_LOADING 2441 /** 2442 * store_cpu_caps() - Store a snapshot of CPU capabilities 2443 * @curr_info: Pointer where to store it 2444 * 2445 * Returns: None 2446 */ 2447 void store_cpu_caps(struct cpuinfo_x86 *curr_info) 2448 { 2449 /* Reload CPUID max function as it might've changed. */ 2450 curr_info->cpuid_level = cpuid_eax(0); 2451 2452 /* Copy all capability leafs and pick up the synthetic ones. */ 2453 memcpy(&curr_info->x86_capability, &boot_cpu_data.x86_capability, 2454 sizeof(curr_info->x86_capability)); 2455 2456 /* Get the hardware CPUID leafs */ 2457 get_cpu_cap(curr_info); 2458 } 2459 2460 /** 2461 * microcode_check() - Check if any CPU capabilities changed after an update. 2462 * @prev_info: CPU capabilities stored before an update. 2463 * 2464 * The microcode loader calls this upon late microcode load to recheck features, 2465 * only when microcode has been updated. Caller holds and CPU hotplug lock. 2466 * 2467 * Return: None 2468 */ 2469 void microcode_check(struct cpuinfo_x86 *prev_info) 2470 { 2471 struct cpuinfo_x86 curr_info; 2472 2473 perf_check_microcode(); 2474 2475 amd_check_microcode(); 2476 2477 store_cpu_caps(&curr_info); 2478 2479 if (!memcmp(&prev_info->x86_capability, &curr_info.x86_capability, 2480 sizeof(prev_info->x86_capability))) 2481 return; 2482 2483 pr_warn("x86/CPU: CPU features have changed after loading microcode, but might not take effect.\n"); 2484 pr_warn("x86/CPU: Please consider either early loading through initrd/built-in or a potential BIOS update.\n"); 2485 } 2486 #endif 2487 2488 /* 2489 * Invoked from core CPU hotplug code after hotplug operations 2490 */ 2491 void arch_smt_update(void) 2492 { 2493 /* Handle the speculative execution misfeatures */ 2494 cpu_bugs_smt_update(); 2495 /* Check whether IPI broadcasting can be enabled */ 2496 apic_smt_update(); 2497 } 2498 2499 void __init arch_cpu_finalize_init(void) 2500 { 2501 struct cpuinfo_x86 *c = this_cpu_ptr(&cpu_info); 2502 2503 identify_boot_cpu(); 2504 2505 select_idle_routine(); 2506 2507 /* 2508 * identify_boot_cpu() initialized SMT support information, let the 2509 * core code know. 2510 */ 2511 cpu_smt_set_num_threads(__max_threads_per_core, __max_threads_per_core); 2512 2513 if (!IS_ENABLED(CONFIG_SMP)) { 2514 pr_info("CPU: "); 2515 print_cpu_info(&boot_cpu_data); 2516 } 2517 2518 cpu_select_mitigations(); 2519 2520 arch_smt_update(); 2521 2522 if (IS_ENABLED(CONFIG_X86_32)) { 2523 /* 2524 * Check whether this is a real i386 which is not longer 2525 * supported and fixup the utsname. 2526 */ 2527 if (boot_cpu_data.x86 < 4) 2528 panic("Kernel requires i486+ for 'invlpg' and other features"); 2529 2530 init_utsname()->machine[1] = 2531 '0' + (boot_cpu_data.x86 > 6 ? 6 : boot_cpu_data.x86); 2532 } 2533 2534 /* 2535 * Must be before alternatives because it might set or clear 2536 * feature bits. 2537 */ 2538 fpu__init_system(); 2539 fpu__init_cpu(); 2540 2541 /* 2542 * This needs to follow the FPU initializtion, since EFI depends on it. 2543 */ 2544 if (efi_enabled(EFI_RUNTIME_SERVICES)) 2545 efi_enter_virtual_mode(); 2546 2547 /* 2548 * Ensure that access to the per CPU representation has the initial 2549 * boot CPU configuration. 2550 */ 2551 *c = boot_cpu_data; 2552 c->initialized = true; 2553 2554 alternative_instructions(); 2555 2556 if (IS_ENABLED(CONFIG_X86_64)) { 2557 unsigned long USER_PTR_MAX = TASK_SIZE_MAX; 2558 2559 /* 2560 * Enable this when LAM is gated on LASS support 2561 if (cpu_feature_enabled(X86_FEATURE_LAM)) 2562 USER_PTR_MAX = (1ul << 63) - PAGE_SIZE; 2563 */ 2564 runtime_const_init(ptr, USER_PTR_MAX); 2565 2566 /* 2567 * Make sure the first 2MB area is not mapped by huge pages 2568 * There are typically fixed size MTRRs in there and overlapping 2569 * MTRRs into large pages causes slow downs. 2570 * 2571 * Right now we don't do that with gbpages because there seems 2572 * very little benefit for that case. 2573 */ 2574 if (!direct_gbpages) 2575 set_memory_4k((unsigned long)__va(0), 1); 2576 } else { 2577 fpu__init_check_bugs(); 2578 } 2579 2580 /* 2581 * This needs to be called before any devices perform DMA 2582 * operations that might use the SWIOTLB bounce buffers. It will 2583 * mark the bounce buffers as decrypted so that their usage will 2584 * not cause "plain-text" data to be decrypted when accessed. It 2585 * must be called after late_time_init() so that Hyper-V x86/x64 2586 * hypercalls work when the SWIOTLB bounce buffers are decrypted. 2587 */ 2588 mem_encrypt_init(); 2589 } 2590