arch/x86/kernel/cpu/bugs.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  Copyright (C) 1994  Linus Torvalds
   4  *
   5  *  Cyrix stuff, June 1998 by:
   6  *      - Rafael R. Reilova (moved everything from head.S),
   7  *        <rreilova@ececs.uc.edu>
   8  *      - Channing Corn (tests & fixes),
   9  *      - Andrew D. Balsa (code cleanup).
  10  */
  11 #include <linux/init.h>
  12 #include <linux/utsname.h>
  13 #include <linux/cpu.h>
  14 #include <linux/module.h>
  15 #include <linux/nospec.h>
  16 #include <linux/prctl.h>
  17 #include <linux/sched/smt.h>
  18 #include <linux/pgtable.h>
  19 #include <linux/bpf.h>
  20
  21 #include <asm/spec-ctrl.h>
  22 #include <asm/cmdline.h>
  23 #include <asm/bugs.h>
  24 #include <asm/processor.h>
  25 #include <asm/processor-flags.h>
  26 #include <asm/fpu/api.h>
  27 #include <asm/msr.h>
  28 #include <asm/vmx.h>
  29 #include <asm/paravirt.h>
  30 #include <asm/alternative.h>
  31 #include <asm/set_memory.h>
  32 #include <asm/intel-family.h>
  33 #include <asm/e820/api.h>
  34 #include <asm/hypervisor.h>
  35 #include <asm/tlbflush.h>
  36
  37 #include "cpu.h"
  38
  39 static void __init spectre_v1_select_mitigation(void);
  40 static void __init spectre_v2_select_mitigation(void);
  41 static void __init retbleed_select_mitigation(void);
  42 static void __init spectre_v2_user_select_mitigation(void);
  43 static void __init ssb_select_mitigation(void);
  44 static void __init l1tf_select_mitigation(void);
  45 static void __init mds_select_mitigation(void);
  46 static void __init md_clear_update_mitigation(void);
  47 static void __init md_clear_select_mitigation(void);
  48 static void __init taa_select_mitigation(void);
  49 static void __init mmio_select_mitigation(void);
  50 static void __init srbds_select_mitigation(void);
  51 static void __init l1d_flush_select_mitigation(void);
  52
  53 /* The base value of the SPEC_CTRL MSR without task-specific bits set */
  54 u64 x86_spec_ctrl_base;
  55 EXPORT_SYMBOL_GPL(x86_spec_ctrl_base);
  56
  57 /* The current value of the SPEC_CTRL MSR with task-specific bits set */
  58 DEFINE_PER_CPU(u64, x86_spec_ctrl_current);
  59 EXPORT_SYMBOL_GPL(x86_spec_ctrl_current);
  60
  61 static DEFINE_MUTEX(spec_ctrl_mutex);
  62
  63 /*
  64  * Keep track of the SPEC_CTRL MSR value for the current task, which may differ
  65  * from x86_spec_ctrl_base due to STIBP/SSB in __speculation_ctrl_update().
  66  */
  67 void write_spec_ctrl_current(u64 val, bool force)
  68 {
  69         if (this_cpu_read(x86_spec_ctrl_current) == val)
  70                 return;
  71
  72         this_cpu_write(x86_spec_ctrl_current, val);
  73
  74         /*
  75          * When KERNEL_IBRS this MSR is written on return-to-user, unless
  76          * forced the update can be delayed until that time.
  77          */
  78         if (force || !cpu_feature_enabled(X86_FEATURE_KERNEL_IBRS))
  79                 wrmsrl(MSR_IA32_SPEC_CTRL, val);
  80 }
  81
  82 u64 spec_ctrl_current(void)
  83 {
  84         return this_cpu_read(x86_spec_ctrl_current);
  85 }
  86 EXPORT_SYMBOL_GPL(spec_ctrl_current);
  87
  88 /*
  89  * AMD specific MSR info for Speculative Store Bypass control.
  90  * x86_amd_ls_cfg_ssbd_mask is initialized in identify_boot_cpu().
  91  */
  92 u64 __ro_after_init x86_amd_ls_cfg_base;
  93 u64 __ro_after_init x86_amd_ls_cfg_ssbd_mask;
  94
  95 /* Control conditional STIBP in switch_to() */
  96 DEFINE_STATIC_KEY_FALSE(switch_to_cond_stibp);
  97 /* Control conditional IBPB in switch_mm() */
  98 DEFINE_STATIC_KEY_FALSE(switch_mm_cond_ibpb);
  99 /* Control unconditional IBPB in switch_mm() */
 100 DEFINE_STATIC_KEY_FALSE(switch_mm_always_ibpb);
 101
 102 /* Control MDS CPU buffer clear before returning to user space */
 103 DEFINE_STATIC_KEY_FALSE(mds_user_clear);
 104 EXPORT_SYMBOL_GPL(mds_user_clear);
 105 /* Control MDS CPU buffer clear before idling (halt, mwait) */
 106 DEFINE_STATIC_KEY_FALSE(mds_idle_clear);
 107 EXPORT_SYMBOL_GPL(mds_idle_clear);
 108
 109 /*
 110  * Controls whether l1d flush based mitigations are enabled,
 111  * based on hw features and admin setting via boot parameter
 112  * defaults to false
 113  */
 114 DEFINE_STATIC_KEY_FALSE(switch_mm_cond_l1d_flush);
 115
 116 /* Controls CPU Fill buffer clear before KVM guest MMIO accesses */
 117 DEFINE_STATIC_KEY_FALSE(mmio_stale_data_clear);
 118 EXPORT_SYMBOL_GPL(mmio_stale_data_clear);
 119
 120 void __init check_bugs(void)
 121 {
 122         identify_boot_cpu();
 123
 124         /*
 125          * identify_boot_cpu() initialized SMT support information, let the
 126          * core code know.
 127          */
 128         cpu_smt_check_topology();
 129
 130         if (!IS_ENABLED(CONFIG_SMP)) {
 131                 pr_info("CPU: ");
 132                 print_cpu_info(&boot_cpu_data);
 133         }
 134
 135         /*
 136          * Read the SPEC_CTRL MSR to account for reserved bits which may
 137          * have unknown values. AMD64_LS_CFG MSR is cached in the early AMD
 138          * init code as it is not enumerated and depends on the family.
 139          */
 140         if (boot_cpu_has(X86_FEATURE_MSR_SPEC_CTRL))
 141                 rdmsrl(MSR_IA32_SPEC_CTRL, x86_spec_ctrl_base);
 142
 143         /* Select the proper CPU mitigations before patching alternatives: */
 144         spectre_v1_select_mitigation();
 145         spectre_v2_select_mitigation();
 146         /*
 147          * retbleed_select_mitigation() relies on the state set by
 148          * spectre_v2_select_mitigation(); specifically it wants to know about
 149          * spectre_v2=ibrs.
 150          */
 151         retbleed_select_mitigation();
 152         /*
 153          * spectre_v2_user_select_mitigation() relies on the state set by
 154          * retbleed_select_mitigation(); specifically the STIBP selection is
 155          * forced for UNRET.
 156          */
 157         spectre_v2_user_select_mitigation();
 158         ssb_select_mitigation();
 159         l1tf_select_mitigation();
 160         md_clear_select_mitigation();
 161         srbds_select_mitigation();
 162         l1d_flush_select_mitigation();
 163
 164         arch_smt_update();
 165
 166 #ifdef CONFIG_X86_32
 167         /*
 168          * Check whether we are able to run this kernel safely on SMP.
 169          *
 170          * - i386 is no longer supported.
 171          * - In order to run on anything without a TSC, we need to be
 172          *   compiled for a i486.
 173          */
 174         if (boot_cpu_data.x86 < 4)
 175                 panic("Kernel requires i486+ for 'invlpg' and other features");
 176
 177         init_utsname()->machine[1] =
 178                 '0' + (boot_cpu_data.x86 > 6 ? 6 : boot_cpu_data.x86);
 179         alternative_instructions();
 180
 181         fpu__init_check_bugs();
 182 #else /* CONFIG_X86_64 */
 183         alternative_instructions();
 184
 185         /*
 186          * Make sure the first 2MB area is not mapped by huge pages
 187          * There are typically fixed size MTRRs in there and overlapping
 188          * MTRRs into large pages causes slow downs.
 189          *
 190          * Right now we don't do that with gbpages because there seems
 191          * very little benefit for that case.
 192          */
 193         if (!direct_gbpages)
 194                 set_memory_4k((unsigned long)__va(0), 1);
 195 #endif
 196 }
 197
 198 /*
 199  * NOTE: This function is *only* called for SVM.  VMX spec_ctrl handling is
 200  * done in vmenter.S.
 201  */
 202 void
 203 x86_virt_spec_ctrl(u64 guest_spec_ctrl, u64 guest_virt_spec_ctrl, bool setguest)
 204 {
 205         u64 msrval, guestval = guest_spec_ctrl, hostval = spec_ctrl_current();
 206         struct thread_info *ti = current_thread_info();
 207
 208         if (static_cpu_has(X86_FEATURE_MSR_SPEC_CTRL)) {
 209                 if (hostval != guestval) {
 210                         msrval = setguest ? guestval : hostval;
 211                         wrmsrl(MSR_IA32_SPEC_CTRL, msrval);
 212                 }
 213         }
 214
 215         /*
 216          * If SSBD is not handled in MSR_SPEC_CTRL on AMD, update
 217          * MSR_AMD64_L2_CFG or MSR_VIRT_SPEC_CTRL if supported.
 218          */
 219         if (!static_cpu_has(X86_FEATURE_LS_CFG_SSBD) &&
 220             !static_cpu_has(X86_FEATURE_VIRT_SSBD))
 221                 return;
 222
 223         /*
 224          * If the host has SSBD mitigation enabled, force it in the host's
 225          * virtual MSR value. If its not permanently enabled, evaluate
 226          * current's TIF_SSBD thread flag.
 227          */
 228         if (static_cpu_has(X86_FEATURE_SPEC_STORE_BYPASS_DISABLE))
 229                 hostval = SPEC_CTRL_SSBD;
 230         else
 231                 hostval = ssbd_tif_to_spec_ctrl(ti->flags);
 232
 233         /* Sanitize the guest value */
 234         guestval = guest_virt_spec_ctrl & SPEC_CTRL_SSBD;
 235
 236         if (hostval != guestval) {
 237                 unsigned long tif;
 238
 239                 tif = setguest ? ssbd_spec_ctrl_to_tif(guestval) :
 240                                  ssbd_spec_ctrl_to_tif(hostval);
 241
 242                 speculation_ctrl_update(tif);
 243         }
 244 }
 245 EXPORT_SYMBOL_GPL(x86_virt_spec_ctrl);
 246
 247 static void x86_amd_ssb_disable(void)
 248 {
 249         u64 msrval = x86_amd_ls_cfg_base | x86_amd_ls_cfg_ssbd_mask;
 250
 251         if (boot_cpu_has(X86_FEATURE_VIRT_SSBD))
 252                 wrmsrl(MSR_AMD64_VIRT_SPEC_CTRL, SPEC_CTRL_SSBD);
 253         else if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD))
 254                 wrmsrl(MSR_AMD64_LS_CFG, msrval);
 255 }
 256
 257 #undef pr_fmt
 258 #define pr_fmt(fmt)     "MDS: " fmt
 259
 260 /* Default mitigation for MDS-affected CPUs */
 261 static enum mds_mitigations mds_mitigation __ro_after_init = MDS_MITIGATION_FULL;
 262 static bool mds_nosmt __ro_after_init = false;
 263
 264 static const char * const mds_strings[] = {
 265         [MDS_MITIGATION_OFF]    = "Vulnerable",
 266         [MDS_MITIGATION_FULL]   = "Mitigation: Clear CPU buffers",
 267         [MDS_MITIGATION_VMWERV] = "Vulnerable: Clear CPU buffers attempted, no microcode",
 268 };
 269
 270 static void __init mds_select_mitigation(void)
 271 {
 272         if (!boot_cpu_has_bug(X86_BUG_MDS) || cpu_mitigations_off()) {
 273                 mds_mitigation = MDS_MITIGATION_OFF;
 274                 return;
 275         }
 276
 277         if (mds_mitigation == MDS_MITIGATION_FULL) {
 278                 if (!boot_cpu_has(X86_FEATURE_MD_CLEAR))
 279                         mds_mitigation = MDS_MITIGATION_VMWERV;
 280
 281                 static_branch_enable(&mds_user_clear);
 282
 283                 if (!boot_cpu_has(X86_BUG_MSBDS_ONLY) &&
 284                     (mds_nosmt || cpu_mitigations_auto_nosmt()))
 285                         cpu_smt_disable(false);
 286         }
 287 }
 288
 289 static int __init mds_cmdline(char *str)
 290 {
 291         if (!boot_cpu_has_bug(X86_BUG_MDS))
 292                 return 0;
 293
 294         if (!str)
 295                 return -EINVAL;
 296
 297         if (!strcmp(str, "off"))
 298                 mds_mitigation = MDS_MITIGATION_OFF;
 299         else if (!strcmp(str, "full"))
 300                 mds_mitigation = MDS_MITIGATION_FULL;
 301         else if (!strcmp(str, "full,nosmt")) {
 302                 mds_mitigation = MDS_MITIGATION_FULL;
 303                 mds_nosmt = true;
 304         }
 305
 306         return 0;
 307 }
 308 early_param("mds", mds_cmdline);
 309
 310 #undef pr_fmt
 311 #define pr_fmt(fmt)     "TAA: " fmt
 312
 313 enum taa_mitigations {
 314         TAA_MITIGATION_OFF,
 315         TAA_MITIGATION_UCODE_NEEDED,
 316         TAA_MITIGATION_VERW,
 317         TAA_MITIGATION_TSX_DISABLED,
 318 };
 319
 320 /* Default mitigation for TAA-affected CPUs */
 321 static enum taa_mitigations taa_mitigation __ro_after_init = TAA_MITIGATION_VERW;
 322 static bool taa_nosmt __ro_after_init;
 323
 324 static const char * const taa_strings[] = {
 325         [TAA_MITIGATION_OFF]            = "Vulnerable",
 326         [TAA_MITIGATION_UCODE_NEEDED]   = "Vulnerable: Clear CPU buffers attempted, no microcode",
 327         [TAA_MITIGATION_VERW]           = "Mitigation: Clear CPU buffers",
 328         [TAA_MITIGATION_TSX_DISABLED]   = "Mitigation: TSX disabled",
 329 };
 330
 331 static void __init taa_select_mitigation(void)
 332 {
 333         u64 ia32_cap;
 334
 335         if (!boot_cpu_has_bug(X86_BUG_TAA)) {
 336                 taa_mitigation = TAA_MITIGATION_OFF;
 337                 return;
 338         }
 339
 340         /* TSX previously disabled by tsx=off */
 341         if (!boot_cpu_has(X86_FEATURE_RTM)) {
 342                 taa_mitigation = TAA_MITIGATION_TSX_DISABLED;
 343                 return;
 344         }
 345
 346         if (cpu_mitigations_off()) {
 347                 taa_mitigation = TAA_MITIGATION_OFF;
 348                 return;
 349         }
 350
 351         /*
 352          * TAA mitigation via VERW is turned off if both
 353          * tsx_async_abort=off and mds=off are specified.
 354          */
 355         if (taa_mitigation == TAA_MITIGATION_OFF &&
 356             mds_mitigation == MDS_MITIGATION_OFF)
 357                 return;
 358
 359         if (boot_cpu_has(X86_FEATURE_MD_CLEAR))
 360                 taa_mitigation = TAA_MITIGATION_VERW;
 361         else
 362                 taa_mitigation = TAA_MITIGATION_UCODE_NEEDED;
 363
 364         /*
 365          * VERW doesn't clear the CPU buffers when MD_CLEAR=1 and MDS_NO=1.
 366          * A microcode update fixes this behavior to clear CPU buffers. It also
 367          * adds support for MSR_IA32_TSX_CTRL which is enumerated by the
 368          * ARCH_CAP_TSX_CTRL_MSR bit.
 369          *
 370          * On MDS_NO=1 CPUs if ARCH_CAP_TSX_CTRL_MSR is not set, microcode
 371          * update is required.
 372          */
 373         ia32_cap = x86_read_arch_cap_msr();
 374         if ( (ia32_cap & ARCH_CAP_MDS_NO) &&
 375             !(ia32_cap & ARCH_CAP_TSX_CTRL_MSR))
 376                 taa_mitigation = TAA_MITIGATION_UCODE_NEEDED;
 377
 378         /*
 379          * TSX is enabled, select alternate mitigation for TAA which is
 380          * the same as MDS. Enable MDS static branch to clear CPU buffers.
 381          *
 382          * For guests that can't determine whether the correct microcode is
 383          * present on host, enable the mitigation for UCODE_NEEDED as well.
 384          */
 385         static_branch_enable(&mds_user_clear);
 386
 387         if (taa_nosmt || cpu_mitigations_auto_nosmt())
 388                 cpu_smt_disable(false);
 389 }
 390
 391 static int __init tsx_async_abort_parse_cmdline(char *str)
 392 {
 393         if (!boot_cpu_has_bug(X86_BUG_TAA))
 394                 return 0;
 395
 396         if (!str)
 397                 return -EINVAL;
 398
 399         if (!strcmp(str, "off")) {
 400                 taa_mitigation = TAA_MITIGATION_OFF;
 401         } else if (!strcmp(str, "full")) {
 402                 taa_mitigation = TAA_MITIGATION_VERW;
 403         } else if (!strcmp(str, "full,nosmt")) {
 404                 taa_mitigation = TAA_MITIGATION_VERW;
 405                 taa_nosmt = true;
 406         }
 407
 408         return 0;
 409 }
 410 early_param("tsx_async_abort", tsx_async_abort_parse_cmdline);
 411
 412 #undef pr_fmt
 413 #define pr_fmt(fmt)     "MMIO Stale Data: " fmt
 414
 415 enum mmio_mitigations {
 416         MMIO_MITIGATION_OFF,
 417         MMIO_MITIGATION_UCODE_NEEDED,
 418         MMIO_MITIGATION_VERW,
 419 };
 420
 421 /* Default mitigation for Processor MMIO Stale Data vulnerabilities */
 422 static enum mmio_mitigations mmio_mitigation __ro_after_init = MMIO_MITIGATION_VERW;
 423 static bool mmio_nosmt __ro_after_init = false;
 424
 425 static const char * const mmio_strings[] = {
 426         [MMIO_MITIGATION_OFF]           = "Vulnerable",
 427         [MMIO_MITIGATION_UCODE_NEEDED]  = "Vulnerable: Clear CPU buffers attempted, no microcode",
 428         [MMIO_MITIGATION_VERW]          = "Mitigation: Clear CPU buffers",
 429 };
 430
 431 static void __init mmio_select_mitigation(void)
 432 {
 433         u64 ia32_cap;
 434
 435         if (!boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA) ||
 436             cpu_mitigations_off()) {
 437                 mmio_mitigation = MMIO_MITIGATION_OFF;
 438                 return;
 439         }
 440
 441         if (mmio_mitigation == MMIO_MITIGATION_OFF)
 442                 return;
 443
 444         ia32_cap = x86_read_arch_cap_msr();
 445
 446         /*
 447          * Enable CPU buffer clear mitigation for host and VMM, if also affected
 448          * by MDS or TAA. Otherwise, enable mitigation for VMM only.
 449          */
 450         if (boot_cpu_has_bug(X86_BUG_MDS) || (boot_cpu_has_bug(X86_BUG_TAA) &&
 451                                               boot_cpu_has(X86_FEATURE_RTM)))
 452                 static_branch_enable(&mds_user_clear);
 453         else
 454                 static_branch_enable(&mmio_stale_data_clear);
 455
 456         /*
 457          * If Processor-MMIO-Stale-Data bug is present and Fill Buffer data can
 458          * be propagated to uncore buffers, clearing the Fill buffers on idle
 459          * is required irrespective of SMT state.
 460          */
 461         if (!(ia32_cap & ARCH_CAP_FBSDP_NO))
 462                 static_branch_enable(&mds_idle_clear);
 463
 464         /*
 465          * Check if the system has the right microcode.
 466          *
 467          * CPU Fill buffer clear mitigation is enumerated by either an explicit
 468          * FB_CLEAR or by the presence of both MD_CLEAR and L1D_FLUSH on MDS
 469          * affected systems.
 470          */
 471         if ((ia32_cap & ARCH_CAP_FB_CLEAR) ||
 472             (boot_cpu_has(X86_FEATURE_MD_CLEAR) &&
 473              boot_cpu_has(X86_FEATURE_FLUSH_L1D) &&
 474              !(ia32_cap & ARCH_CAP_MDS_NO)))
 475                 mmio_mitigation = MMIO_MITIGATION_VERW;
 476         else
 477                 mmio_mitigation = MMIO_MITIGATION_UCODE_NEEDED;
 478
 479         if (mmio_nosmt || cpu_mitigations_auto_nosmt())
 480                 cpu_smt_disable(false);
 481 }
 482
 483 static int __init mmio_stale_data_parse_cmdline(char *str)
 484 {
 485         if (!boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA))
 486                 return 0;
 487
 488         if (!str)
 489                 return -EINVAL;
 490
 491         if (!strcmp(str, "off")) {
 492                 mmio_mitigation = MMIO_MITIGATION_OFF;
 493         } else if (!strcmp(str, "full")) {
 494                 mmio_mitigation = MMIO_MITIGATION_VERW;
 495         } else if (!strcmp(str, "full,nosmt")) {
 496                 mmio_mitigation = MMIO_MITIGATION_VERW;
 497                 mmio_nosmt = true;
 498         }
 499
 500         return 0;
 501 }
 502 early_param("mmio_stale_data", mmio_stale_data_parse_cmdline);
 503
 504 #undef pr_fmt
 505 #define pr_fmt(fmt)     "" fmt
 506
 507 static void __init md_clear_update_mitigation(void)
 508 {
 509         if (cpu_mitigations_off())
 510                 return;
 511
 512         if (!static_key_enabled(&mds_user_clear))
 513                 goto out;
 514
 515         /*
 516          * mds_user_clear is now enabled. Update MDS, TAA and MMIO Stale Data
 517          * mitigation, if necessary.
 518          */
 519         if (mds_mitigation == MDS_MITIGATION_OFF &&
 520             boot_cpu_has_bug(X86_BUG_MDS)) {
 521                 mds_mitigation = MDS_MITIGATION_FULL;
 522                 mds_select_mitigation();
 523         }
 524         if (taa_mitigation == TAA_MITIGATION_OFF &&
 525             boot_cpu_has_bug(X86_BUG_TAA)) {
 526                 taa_mitigation = TAA_MITIGATION_VERW;
 527                 taa_select_mitigation();
 528         }
 529         if (mmio_mitigation == MMIO_MITIGATION_OFF &&
 530             boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA)) {
 531                 mmio_mitigation = MMIO_MITIGATION_VERW;
 532                 mmio_select_mitigation();
 533         }
 534 out:
 535         if (boot_cpu_has_bug(X86_BUG_MDS))
 536                 pr_info("MDS: %s\n", mds_strings[mds_mitigation]);
 537         if (boot_cpu_has_bug(X86_BUG_TAA))
 538                 pr_info("TAA: %s\n", taa_strings[taa_mitigation]);
 539         if (boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA))
 540                 pr_info("MMIO Stale Data: %s\n", mmio_strings[mmio_mitigation]);
 541 }
 542
 543 static void __init md_clear_select_mitigation(void)
 544 {
 545         mds_select_mitigation();
 546         taa_select_mitigation();
 547         mmio_select_mitigation();
 548
 549         /*
 550          * As MDS, TAA and MMIO Stale Data mitigations are inter-related, update
 551          * and print their mitigation after MDS, TAA and MMIO Stale Data
 552          * mitigation selection is done.
 553          */
 554         md_clear_update_mitigation();
 555 }
 556
 557 #undef pr_fmt
 558 #define pr_fmt(fmt)     "SRBDS: " fmt
 559
 560 enum srbds_mitigations {
 561         SRBDS_MITIGATION_OFF,
 562         SRBDS_MITIGATION_UCODE_NEEDED,
 563         SRBDS_MITIGATION_FULL,
 564         SRBDS_MITIGATION_TSX_OFF,
 565         SRBDS_MITIGATION_HYPERVISOR,
 566 };
 567
 568 static enum srbds_mitigations srbds_mitigation __ro_after_init = SRBDS_MITIGATION_FULL;
 569
 570 static const char * const srbds_strings[] = {
 571         [SRBDS_MITIGATION_OFF]          = "Vulnerable",
 572         [SRBDS_MITIGATION_UCODE_NEEDED] = "Vulnerable: No microcode",
 573         [SRBDS_MITIGATION_FULL]         = "Mitigation: Microcode",
 574         [SRBDS_MITIGATION_TSX_OFF]      = "Mitigation: TSX disabled",
 575         [SRBDS_MITIGATION_HYPERVISOR]   = "Unknown: Dependent on hypervisor status",
 576 };
 577
 578 static bool srbds_off;
 579
 580 void update_srbds_msr(void)
 581 {
 582         u64 mcu_ctrl;
 583
 584         if (!boot_cpu_has_bug(X86_BUG_SRBDS))
 585                 return;
 586
 587         if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
 588                 return;
 589
 590         if (srbds_mitigation == SRBDS_MITIGATION_UCODE_NEEDED)
 591                 return;
 592
 593         /*
 594          * A MDS_NO CPU for which SRBDS mitigation is not needed due to TSX
 595          * being disabled and it hasn't received the SRBDS MSR microcode.
 596          */
 597         if (!boot_cpu_has(X86_FEATURE_SRBDS_CTRL))
 598                 return;
 599
 600         rdmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
 601
 602         switch (srbds_mitigation) {
 603         case SRBDS_MITIGATION_OFF:
 604         case SRBDS_MITIGATION_TSX_OFF:
 605                 mcu_ctrl |= RNGDS_MITG_DIS;
 606                 break;
 607         case SRBDS_MITIGATION_FULL:
 608                 mcu_ctrl &= ~RNGDS_MITG_DIS;
 609                 break;
 610         default:
 611                 break;
 612         }
 613
 614         wrmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
 615 }
 616
 617 static void __init srbds_select_mitigation(void)
 618 {
 619         u64 ia32_cap;
 620
 621         if (!boot_cpu_has_bug(X86_BUG_SRBDS))
 622                 return;
 623
 624         /*
 625          * Check to see if this is one of the MDS_NO systems supporting TSX that
 626          * are only exposed to SRBDS when TSX is enabled or when CPU is affected
 627          * by Processor MMIO Stale Data vulnerability.
 628          */
 629         ia32_cap = x86_read_arch_cap_msr();
 630         if ((ia32_cap & ARCH_CAP_MDS_NO) && !boot_cpu_has(X86_FEATURE_RTM) &&
 631             !boot_cpu_has_bug(X86_BUG_MMIO_STALE_DATA))
 632                 srbds_mitigation = SRBDS_MITIGATION_TSX_OFF;
 633         else if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
 634                 srbds_mitigation = SRBDS_MITIGATION_HYPERVISOR;
 635         else if (!boot_cpu_has(X86_FEATURE_SRBDS_CTRL))
 636                 srbds_mitigation = SRBDS_MITIGATION_UCODE_NEEDED;
 637         else if (cpu_mitigations_off() || srbds_off)
 638                 srbds_mitigation = SRBDS_MITIGATION_OFF;
 639
 640         update_srbds_msr();
 641         pr_info("%s\n", srbds_strings[srbds_mitigation]);
 642 }
 643
 644 static int __init srbds_parse_cmdline(char *str)
 645 {
 646         if (!str)
 647                 return -EINVAL;
 648
 649         if (!boot_cpu_has_bug(X86_BUG_SRBDS))
 650                 return 0;
 651
 652         srbds_off = !strcmp(str, "off");
 653         return 0;
 654 }
 655 early_param("srbds", srbds_parse_cmdline);
 656
 657 #undef pr_fmt
 658 #define pr_fmt(fmt)     "L1D Flush : " fmt
 659
 660 enum l1d_flush_mitigations {
 661         L1D_FLUSH_OFF = 0,
 662         L1D_FLUSH_ON,
 663 };
 664
 665 static enum l1d_flush_mitigations l1d_flush_mitigation __initdata = L1D_FLUSH_OFF;
 666
 667 static void __init l1d_flush_select_mitigation(void)
 668 {
 669         if (!l1d_flush_mitigation || !boot_cpu_has(X86_FEATURE_FLUSH_L1D))
 670                 return;
 671
 672         static_branch_enable(&switch_mm_cond_l1d_flush);
 673         pr_info("Conditional flush on switch_mm() enabled\n");
 674 }
 675
 676 static int __init l1d_flush_parse_cmdline(char *str)
 677 {
 678         if (!strcmp(str, "on"))
 679                 l1d_flush_mitigation = L1D_FLUSH_ON;
 680
 681         return 0;
 682 }
 683 early_param("l1d_flush", l1d_flush_parse_cmdline);
 684
 685 #undef pr_fmt
 686 #define pr_fmt(fmt)     "Spectre V1 : " fmt
 687
 688 enum spectre_v1_mitigation {
 689         SPECTRE_V1_MITIGATION_NONE,
 690         SPECTRE_V1_MITIGATION_AUTO,
 691 };
 692
 693 static enum spectre_v1_mitigation spectre_v1_mitigation __ro_after_init =
 694         SPECTRE_V1_MITIGATION_AUTO;
 695
 696 static const char * const spectre_v1_strings[] = {
 697         [SPECTRE_V1_MITIGATION_NONE] = "Vulnerable: __user pointer sanitization and usercopy barriers only; no swapgs barriers",
 698         [SPECTRE_V1_MITIGATION_AUTO] = "Mitigation: usercopy/swapgs barriers and __user pointer sanitization",
 699 };
 700
 701 /*
 702  * Does SMAP provide full mitigation against speculative kernel access to
 703  * userspace?
 704  */
 705 static bool smap_works_speculatively(void)
 706 {
 707         if (!boot_cpu_has(X86_FEATURE_SMAP))
 708                 return false;
 709
 710         /*
 711          * On CPUs which are vulnerable to Meltdown, SMAP does not
 712          * prevent speculative access to user data in the L1 cache.
 713          * Consider SMAP to be non-functional as a mitigation on these
 714          * CPUs.
 715          */
 716         if (boot_cpu_has(X86_BUG_CPU_MELTDOWN))
 717                 return false;
 718
 719         return true;
 720 }
 721
 722 static void __init spectre_v1_select_mitigation(void)
 723 {
 724         if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V1) || cpu_mitigations_off()) {
 725                 spectre_v1_mitigation = SPECTRE_V1_MITIGATION_NONE;
 726                 return;
 727         }
 728
 729         if (spectre_v1_mitigation == SPECTRE_V1_MITIGATION_AUTO) {
 730                 /*
 731                  * With Spectre v1, a user can speculatively control either
 732                  * path of a conditional swapgs with a user-controlled GS
 733                  * value.  The mitigation is to add lfences to both code paths.
 734                  *
 735                  * If FSGSBASE is enabled, the user can put a kernel address in
 736                  * GS, in which case SMAP provides no protection.
 737                  *
 738                  * If FSGSBASE is disabled, the user can only put a user space
 739                  * address in GS.  That makes an attack harder, but still
 740                  * possible if there's no SMAP protection.
 741                  */
 742                 if (boot_cpu_has(X86_FEATURE_FSGSBASE) ||
 743                     !smap_works_speculatively()) {
 744                         /*
 745                          * Mitigation can be provided from SWAPGS itself or
 746                          * PTI as the CR3 write in the Meltdown mitigation
 747                          * is serializing.
 748                          *
 749                          * If neither is there, mitigate with an LFENCE to
 750                          * stop speculation through swapgs.
 751                          */
 752                         if (boot_cpu_has_bug(X86_BUG_SWAPGS) &&
 753                             !boot_cpu_has(X86_FEATURE_PTI))
 754                                 setup_force_cpu_cap(X86_FEATURE_FENCE_SWAPGS_USER);
 755
 756                         /*
 757                          * Enable lfences in the kernel entry (non-swapgs)
 758                          * paths, to prevent user entry from speculatively
 759                          * skipping swapgs.
 760                          */
 761                         setup_force_cpu_cap(X86_FEATURE_FENCE_SWAPGS_KERNEL);
 762                 }
 763         }
 764
 765         pr_info("%s\n", spectre_v1_strings[spectre_v1_mitigation]);
 766 }
 767
 768 static int __init nospectre_v1_cmdline(char *str)
 769 {
 770         spectre_v1_mitigation = SPECTRE_V1_MITIGATION_NONE;
 771         return 0;
 772 }
 773 early_param("nospectre_v1", nospectre_v1_cmdline);
 774
 775 static enum spectre_v2_mitigation spectre_v2_enabled __ro_after_init =
 776         SPECTRE_V2_NONE;
 777
 778 #undef pr_fmt
 779 #define pr_fmt(fmt)     "RETBleed: " fmt
 780
 781 enum retbleed_mitigation {
 782         RETBLEED_MITIGATION_NONE,
 783         RETBLEED_MITIGATION_UNRET,
 784         RETBLEED_MITIGATION_IBPB,
 785         RETBLEED_MITIGATION_IBRS,
 786         RETBLEED_MITIGATION_EIBRS,
 787 };
 788
 789 enum retbleed_mitigation_cmd {
 790         RETBLEED_CMD_OFF,
 791         RETBLEED_CMD_AUTO,
 792         RETBLEED_CMD_UNRET,
 793         RETBLEED_CMD_IBPB,
 794 };
 795
 796 static const char * const retbleed_strings[] = {
 797         [RETBLEED_MITIGATION_NONE]      = "Vulnerable",
 798         [RETBLEED_MITIGATION_UNRET]     = "Mitigation: untrained return thunk",
 799         [RETBLEED_MITIGATION_IBPB]      = "Mitigation: IBPB",
 800         [RETBLEED_MITIGATION_IBRS]      = "Mitigation: IBRS",
 801         [RETBLEED_MITIGATION_EIBRS]     = "Mitigation: Enhanced IBRS",
 802 };
 803
 804 static enum retbleed_mitigation retbleed_mitigation __ro_after_init =
 805         RETBLEED_MITIGATION_NONE;
 806 static enum retbleed_mitigation_cmd retbleed_cmd __ro_after_init =
 807         RETBLEED_CMD_AUTO;
 808
 809 static int __ro_after_init retbleed_nosmt = false;
 810
 811 static int __init retbleed_parse_cmdline(char *str)
 812 {
 813         if (!str)
 814                 return -EINVAL;
 815
 816         while (str) {
 817                 char *next = strchr(str, ',');
 818                 if (next) {
 819                         *next = 0;
 820                         next++;
 821                 }
 822
 823                 if (!strcmp(str, "off")) {
 824                         retbleed_cmd = RETBLEED_CMD_OFF;
 825                 } else if (!strcmp(str, "auto")) {
 826                         retbleed_cmd = RETBLEED_CMD_AUTO;
 827                 } else if (!strcmp(str, "unret")) {
 828                         retbleed_cmd = RETBLEED_CMD_UNRET;
 829                 } else if (!strcmp(str, "ibpb")) {
 830                         retbleed_cmd = RETBLEED_CMD_IBPB;
 831                 } else if (!strcmp(str, "nosmt")) {
 832                         retbleed_nosmt = true;
 833                 } else {
 834                         pr_err("Ignoring unknown retbleed option (%s).", str);
 835                 }
 836
 837                 str = next;
 838         }
 839
 840         return 0;
 841 }
 842 early_param("retbleed", retbleed_parse_cmdline);
 843
 844 #define RETBLEED_UNTRAIN_MSG "WARNING: BTB untrained return thunk mitigation is only effective on AMD/Hygon!\n"
 845 #define RETBLEED_INTEL_MSG "WARNING: Spectre v2 mitigation leaves CPU vulnerable to RETBleed attacks, data leaks possible!\n"
 846
 847 static void __init retbleed_select_mitigation(void)
 848 {
 849         bool mitigate_smt = false;
 850
 851         if (!boot_cpu_has_bug(X86_BUG_RETBLEED) || cpu_mitigations_off())
 852                 return;
 853
 854         switch (retbleed_cmd) {
 855         case RETBLEED_CMD_OFF:
 856                 return;
 857
 858         case RETBLEED_CMD_UNRET:
 859                 if (IS_ENABLED(CONFIG_CPU_UNRET_ENTRY)) {
 860                         retbleed_mitigation = RETBLEED_MITIGATION_UNRET;
 861                 } else {
 862                         pr_err("WARNING: kernel not compiled with CPU_UNRET_ENTRY.\n");
 863                         goto do_cmd_auto;
 864                 }
 865                 break;
 866
 867         case RETBLEED_CMD_IBPB:
 868                 if (!boot_cpu_has(X86_FEATURE_IBPB)) {
 869                         pr_err("WARNING: CPU does not support IBPB.\n");
 870                         goto do_cmd_auto;
 871                 } else if (IS_ENABLED(CONFIG_CPU_IBPB_ENTRY)) {
 872                         retbleed_mitigation = RETBLEED_MITIGATION_IBPB;
 873                 } else {
 874                         pr_err("WARNING: kernel not compiled with CPU_IBPB_ENTRY.\n");
 875                         goto do_cmd_auto;
 876                 }
 877                 break;
 878
 879 do_cmd_auto:
 880         case RETBLEED_CMD_AUTO:
 881         default:
 882                 if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
 883                     boot_cpu_data.x86_vendor == X86_VENDOR_HYGON) {
 884                         if (IS_ENABLED(CONFIG_CPU_UNRET_ENTRY))
 885                                 retbleed_mitigation = RETBLEED_MITIGATION_UNRET;
 886                         else if (IS_ENABLED(CONFIG_CPU_IBPB_ENTRY) && boot_cpu_has(X86_FEATURE_IBPB))
 887                                 retbleed_mitigation = RETBLEED_MITIGATION_IBPB;
 888                 }
 889
 890                 /*
 891                  * The Intel mitigation (IBRS or eIBRS) was already selected in
 892                  * spectre_v2_select_mitigation().  'retbleed_mitigation' will
 893                  * be set accordingly below.
 894                  */
 895
 896                 break;
 897         }
 898
 899         switch (retbleed_mitigation) {
 900         case RETBLEED_MITIGATION_UNRET:
 901                 setup_force_cpu_cap(X86_FEATURE_RETHUNK);
 902                 setup_force_cpu_cap(X86_FEATURE_UNRET);
 903
 904                 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD &&
 905                     boot_cpu_data.x86_vendor != X86_VENDOR_HYGON)
 906                         pr_err(RETBLEED_UNTRAIN_MSG);
 907
 908                 mitigate_smt = true;
 909                 break;
 910
 911         case RETBLEED_MITIGATION_IBPB:
 912                 setup_force_cpu_cap(X86_FEATURE_ENTRY_IBPB);
 913                 mitigate_smt = true;
 914                 break;
 915
 916         default:
 917                 break;
 918         }
 919
 920         if (mitigate_smt && !boot_cpu_has(X86_FEATURE_STIBP) &&
 921             (retbleed_nosmt || cpu_mitigations_auto_nosmt()))
 922                 cpu_smt_disable(false);
 923
 924         /*
 925          * Let IBRS trump all on Intel without affecting the effects of the
 926          * retbleed= cmdline option.
 927          */
 928         if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) {
 929                 switch (spectre_v2_enabled) {
 930                 case SPECTRE_V2_IBRS:
 931                         retbleed_mitigation = RETBLEED_MITIGATION_IBRS;
 932                         break;
 933                 case SPECTRE_V2_EIBRS:
 934                 case SPECTRE_V2_EIBRS_RETPOLINE:
 935                 case SPECTRE_V2_EIBRS_LFENCE:
 936                         retbleed_mitigation = RETBLEED_MITIGATION_EIBRS;
 937                         break;
 938                 default:
 939                         pr_err(RETBLEED_INTEL_MSG);
 940                 }
 941         }
 942
 943         pr_info("%s\n", retbleed_strings[retbleed_mitigation]);
 944 }
 945
 946 #undef pr_fmt
 947 #define pr_fmt(fmt)     "Spectre V2 : " fmt
 948
 949 static enum spectre_v2_user_mitigation spectre_v2_user_stibp __ro_after_init =
 950         SPECTRE_V2_USER_NONE;
 951 static enum spectre_v2_user_mitigation spectre_v2_user_ibpb __ro_after_init =
 952         SPECTRE_V2_USER_NONE;
 953
 954 #ifdef CONFIG_RETPOLINE
 955 static bool spectre_v2_bad_module;
 956
 957 bool retpoline_module_ok(bool has_retpoline)
 958 {
 959         if (spectre_v2_enabled == SPECTRE_V2_NONE || has_retpoline)
 960                 return true;
 961
 962         pr_err("System may be vulnerable to spectre v2\n");
 963         spectre_v2_bad_module = true;
 964         return false;
 965 }
 966
 967 static inline const char *spectre_v2_module_string(void)
 968 {
 969         return spectre_v2_bad_module ? " - vulnerable module loaded" : "";
 970 }
 971 #else
 972 static inline const char *spectre_v2_module_string(void) { return ""; }
 973 #endif
 974
 975 #define SPECTRE_V2_LFENCE_MSG "WARNING: LFENCE mitigation is not recommended for this CPU, data leaks possible!\n"
 976 #define SPECTRE_V2_EIBRS_EBPF_MSG "WARNING: Unprivileged eBPF is enabled with eIBRS on, data leaks possible via Spectre v2 BHB attacks!\n"
 977 #define SPECTRE_V2_EIBRS_LFENCE_EBPF_SMT_MSG "WARNING: Unprivileged eBPF is enabled with eIBRS+LFENCE mitigation and SMT, data leaks possible via Spectre v2 BHB attacks!\n"
 978 #define SPECTRE_V2_IBRS_PERF_MSG "WARNING: IBRS mitigation selected on Enhanced IBRS CPU, this may cause unnecessary performance loss\n"
 979
 980 #ifdef CONFIG_BPF_SYSCALL
 981 void unpriv_ebpf_notify(int new_state)
 982 {
 983         if (new_state)
 984                 return;
 985
 986         /* Unprivileged eBPF is enabled */
 987
 988         switch (spectre_v2_enabled) {
 989         case SPECTRE_V2_EIBRS:
 990                 pr_err(SPECTRE_V2_EIBRS_EBPF_MSG);
 991                 break;
 992         case SPECTRE_V2_EIBRS_LFENCE:
 993                 if (sched_smt_active())
 994                         pr_err(SPECTRE_V2_EIBRS_LFENCE_EBPF_SMT_MSG);
 995                 break;
 996         default:
 997                 break;
 998         }
 999 }
1000 #endif
1001
1002 static inline bool match_option(const char *arg, int arglen, const char *opt)
1003 {
1004         int len = strlen(opt);
1005
1006         return len == arglen && !strncmp(arg, opt, len);
1007 }
1008
1009 /* The kernel command line selection for spectre v2 */
1010 enum spectre_v2_mitigation_cmd {
1011         SPECTRE_V2_CMD_NONE,
1012         SPECTRE_V2_CMD_AUTO,
1013         SPECTRE_V2_CMD_FORCE,
1014         SPECTRE_V2_CMD_RETPOLINE,
1015         SPECTRE_V2_CMD_RETPOLINE_GENERIC,
1016         SPECTRE_V2_CMD_RETPOLINE_LFENCE,
1017         SPECTRE_V2_CMD_EIBRS,
1018         SPECTRE_V2_CMD_EIBRS_RETPOLINE,
1019         SPECTRE_V2_CMD_EIBRS_LFENCE,
1020         SPECTRE_V2_CMD_IBRS,
1021 };
1022
1023 enum spectre_v2_user_cmd {
1024         SPECTRE_V2_USER_CMD_NONE,
1025         SPECTRE_V2_USER_CMD_AUTO,
1026         SPECTRE_V2_USER_CMD_FORCE,
1027         SPECTRE_V2_USER_CMD_PRCTL,
1028         SPECTRE_V2_USER_CMD_PRCTL_IBPB,
1029         SPECTRE_V2_USER_CMD_SECCOMP,
1030         SPECTRE_V2_USER_CMD_SECCOMP_IBPB,
1031 };
1032
1033 static const char * const spectre_v2_user_strings[] = {
1034         [SPECTRE_V2_USER_NONE]                  = "User space: Vulnerable",
1035         [SPECTRE_V2_USER_STRICT]                = "User space: Mitigation: STIBP protection",
1036         [SPECTRE_V2_USER_STRICT_PREFERRED]      = "User space: Mitigation: STIBP always-on protection",
1037         [SPECTRE_V2_USER_PRCTL]                 = "User space: Mitigation: STIBP via prctl",
1038         [SPECTRE_V2_USER_SECCOMP]               = "User space: Mitigation: STIBP via seccomp and prctl",
1039 };
1040
1041 static const struct {
1042         const char                      *option;
1043         enum spectre_v2_user_cmd        cmd;
1044         bool                            secure;
1045 } v2_user_options[] __initconst = {
1046         { "auto",               SPECTRE_V2_USER_CMD_AUTO,               false },
1047         { "off",                SPECTRE_V2_USER_CMD_NONE,               false },
1048         { "on",                 SPECTRE_V2_USER_CMD_FORCE,              true  },
1049         { "prctl",              SPECTRE_V2_USER_CMD_PRCTL,              false },
1050         { "prctl,ibpb",         SPECTRE_V2_USER_CMD_PRCTL_IBPB,         false },
1051         { "seccomp",            SPECTRE_V2_USER_CMD_SECCOMP,            false },
1052         { "seccomp,ibpb",       SPECTRE_V2_USER_CMD_SECCOMP_IBPB,       false },
1053 };
1054
1055 static void __init spec_v2_user_print_cond(const char *reason, bool secure)
1056 {
1057         if (boot_cpu_has_bug(X86_BUG_SPECTRE_V2) != secure)
1058                 pr_info("spectre_v2_user=%s forced on command line.\n", reason);
1059 }
1060
1061 static __ro_after_init enum spectre_v2_mitigation_cmd spectre_v2_cmd;
1062
1063 static enum spectre_v2_user_cmd __init
1064 spectre_v2_parse_user_cmdline(void)
1065 {
1066         char arg[20];
1067         int ret, i;
1068
1069         switch (spectre_v2_cmd) {
1070         case SPECTRE_V2_CMD_NONE:
1071                 return SPECTRE_V2_USER_CMD_NONE;
1072         case SPECTRE_V2_CMD_FORCE:
1073                 return SPECTRE_V2_USER_CMD_FORCE;
1074         default:
1075                 break;
1076         }
1077
1078         ret = cmdline_find_option(boot_command_line, "spectre_v2_user",
1079                                   arg, sizeof(arg));
1080         if (ret < 0)
1081                 return SPECTRE_V2_USER_CMD_AUTO;
1082
1083         for (i = 0; i < ARRAY_SIZE(v2_user_options); i++) {
1084                 if (match_option(arg, ret, v2_user_options[i].option)) {
1085                         spec_v2_user_print_cond(v2_user_options[i].option,
1086                                                 v2_user_options[i].secure);
1087                         return v2_user_options[i].cmd;
1088                 }
1089         }
1090
1091         pr_err("Unknown user space protection option (%s). Switching to AUTO select\n", arg);
1092         return SPECTRE_V2_USER_CMD_AUTO;
1093 }
1094
1095 static inline bool spectre_v2_in_ibrs_mode(enum spectre_v2_mitigation mode)
1096 {
1097         return mode == SPECTRE_V2_IBRS ||
1098                mode == SPECTRE_V2_EIBRS ||
1099                mode == SPECTRE_V2_EIBRS_RETPOLINE ||
1100                mode == SPECTRE_V2_EIBRS_LFENCE;
1101 }
1102
1103 static void __init
1104 spectre_v2_user_select_mitigation(void)
1105 {
1106         enum spectre_v2_user_mitigation mode = SPECTRE_V2_USER_NONE;
1107         bool smt_possible = IS_ENABLED(CONFIG_SMP);
1108         enum spectre_v2_user_cmd cmd;
1109
1110         if (!boot_cpu_has(X86_FEATURE_IBPB) && !boot_cpu_has(X86_FEATURE_STIBP))
1111                 return;
1112
1113         if (cpu_smt_control == CPU_SMT_FORCE_DISABLED ||
1114             cpu_smt_control == CPU_SMT_NOT_SUPPORTED)
1115                 smt_possible = false;
1116
1117         cmd = spectre_v2_parse_user_cmdline();
1118         switch (cmd) {
1119         case SPECTRE_V2_USER_CMD_NONE:
1120                 goto set_mode;
1121         case SPECTRE_V2_USER_CMD_FORCE:
1122                 mode = SPECTRE_V2_USER_STRICT;
1123                 break;
1124         case SPECTRE_V2_USER_CMD_AUTO:
1125         case SPECTRE_V2_USER_CMD_PRCTL:
1126         case SPECTRE_V2_USER_CMD_PRCTL_IBPB:
1127                 mode = SPECTRE_V2_USER_PRCTL;
1128                 break;
1129         case SPECTRE_V2_USER_CMD_SECCOMP:
1130         case SPECTRE_V2_USER_CMD_SECCOMP_IBPB:
1131                 if (IS_ENABLED(CONFIG_SECCOMP))
1132                         mode = SPECTRE_V2_USER_SECCOMP;
1133                 else
1134                         mode = SPECTRE_V2_USER_PRCTL;
1135                 break;
1136         }
1137
1138         /* Initialize Indirect Branch Prediction Barrier */
1139         if (boot_cpu_has(X86_FEATURE_IBPB)) {
1140                 setup_force_cpu_cap(X86_FEATURE_USE_IBPB);
1141
1142                 spectre_v2_user_ibpb = mode;
1143                 switch (cmd) {
1144                 case SPECTRE_V2_USER_CMD_FORCE:
1145                 case SPECTRE_V2_USER_CMD_PRCTL_IBPB:
1146                 case SPECTRE_V2_USER_CMD_SECCOMP_IBPB:
1147                         static_branch_enable(&switch_mm_always_ibpb);
1148                         spectre_v2_user_ibpb = SPECTRE_V2_USER_STRICT;
1149                         break;
1150                 case SPECTRE_V2_USER_CMD_PRCTL:
1151                 case SPECTRE_V2_USER_CMD_AUTO:
1152                 case SPECTRE_V2_USER_CMD_SECCOMP:
1153                         static_branch_enable(&switch_mm_cond_ibpb);
1154                         break;
1155                 default:
1156                         break;
1157                 }
1158
1159                 pr_info("mitigation: Enabling %s Indirect Branch Prediction Barrier\n",
1160                         static_key_enabled(&switch_mm_always_ibpb) ?
1161                         "always-on" : "conditional");
1162         }
1163
1164         /*
1165          * If no STIBP, IBRS or enhanced IBRS is enabled, or SMT impossible,
1166          * STIBP is not required.
1167          */
1168         if (!boot_cpu_has(X86_FEATURE_STIBP) ||
1169             !smt_possible ||
1170             spectre_v2_in_ibrs_mode(spectre_v2_enabled))
1171                 return;
1172
1173         /*
1174          * At this point, an STIBP mode other than "off" has been set.
1175          * If STIBP support is not being forced, check if STIBP always-on
1176          * is preferred.
1177          */
1178         if (mode != SPECTRE_V2_USER_STRICT &&
1179             boot_cpu_has(X86_FEATURE_AMD_STIBP_ALWAYS_ON))
1180                 mode = SPECTRE_V2_USER_STRICT_PREFERRED;
1181
1182         if (retbleed_mitigation == RETBLEED_MITIGATION_UNRET) {
1183                 if (mode != SPECTRE_V2_USER_STRICT &&
1184                     mode != SPECTRE_V2_USER_STRICT_PREFERRED)
1185                         pr_info("Selecting STIBP always-on mode to complement retbleed mitigation\n");
1186                 mode = SPECTRE_V2_USER_STRICT_PREFERRED;
1187         }
1188
1189         spectre_v2_user_stibp = mode;
1190
1191 set_mode:
1192         pr_info("%s\n", spectre_v2_user_strings[mode]);
1193 }
1194
1195 static const char * const spectre_v2_strings[] = {
1196         [SPECTRE_V2_NONE]                       = "Vulnerable",
1197         [SPECTRE_V2_RETPOLINE]                  = "Mitigation: Retpolines",
1198         [SPECTRE_V2_LFENCE]                     = "Mitigation: LFENCE",
1199         [SPECTRE_V2_EIBRS]                      = "Mitigation: Enhanced IBRS",
1200         [SPECTRE_V2_EIBRS_LFENCE]               = "Mitigation: Enhanced IBRS + LFENCE",
1201         [SPECTRE_V2_EIBRS_RETPOLINE]            = "Mitigation: Enhanced IBRS + Retpolines",
1202         [SPECTRE_V2_IBRS]                       = "Mitigation: IBRS",
1203 };
1204
1205 static const struct {
1206         const char *option;
1207         enum spectre_v2_mitigation_cmd cmd;
1208         bool secure;
1209 } mitigation_options[] __initconst = {
1210         { "off",                SPECTRE_V2_CMD_NONE,              false },
1211         { "on",                 SPECTRE_V2_CMD_FORCE,             true  },
1212         { "retpoline",          SPECTRE_V2_CMD_RETPOLINE,         false },
1213         { "retpoline,amd",      SPECTRE_V2_CMD_RETPOLINE_LFENCE,  false },
1214         { "retpoline,lfence",   SPECTRE_V2_CMD_RETPOLINE_LFENCE,  false },
1215         { "retpoline,generic",  SPECTRE_V2_CMD_RETPOLINE_GENERIC, false },
1216         { "eibrs",              SPECTRE_V2_CMD_EIBRS,             false },
1217         { "eibrs,lfence",       SPECTRE_V2_CMD_EIBRS_LFENCE,      false },
1218         { "eibrs,retpoline",    SPECTRE_V2_CMD_EIBRS_RETPOLINE,   false },
1219         { "auto",               SPECTRE_V2_CMD_AUTO,              false },
1220         { "ibrs",               SPECTRE_V2_CMD_IBRS,              false },
1221 };
1222
1223 static void __init spec_v2_print_cond(const char *reason, bool secure)
1224 {
1225         if (boot_cpu_has_bug(X86_BUG_SPECTRE_V2) != secure)
1226                 pr_info("%s selected on command line.\n", reason);
1227 }
1228
1229 static enum spectre_v2_mitigation_cmd __init spectre_v2_parse_cmdline(void)
1230 {
1231         enum spectre_v2_mitigation_cmd cmd = SPECTRE_V2_CMD_AUTO;
1232         char arg[20];
1233         int ret, i;
1234
1235         if (cmdline_find_option_bool(boot_command_line, "nospectre_v2") ||
1236             cpu_mitigations_off())
1237                 return SPECTRE_V2_CMD_NONE;
1238
1239         ret = cmdline_find_option(boot_command_line, "spectre_v2", arg, sizeof(arg));
1240         if (ret < 0)
1241                 return SPECTRE_V2_CMD_AUTO;
1242
1243         for (i = 0; i < ARRAY_SIZE(mitigation_options); i++) {
1244                 if (!match_option(arg, ret, mitigation_options[i].option))
1245                         continue;
1246                 cmd = mitigation_options[i].cmd;
1247                 break;
1248         }
1249
1250         if (i >= ARRAY_SIZE(mitigation_options)) {
1251                 pr_err("unknown option (%s). Switching to AUTO select\n", arg);
1252                 return SPECTRE_V2_CMD_AUTO;
1253         }
1254
1255         if ((cmd == SPECTRE_V2_CMD_RETPOLINE ||
1256              cmd == SPECTRE_V2_CMD_RETPOLINE_LFENCE ||
1257              cmd == SPECTRE_V2_CMD_RETPOLINE_GENERIC ||
1258              cmd == SPECTRE_V2_CMD_EIBRS_LFENCE ||
1259              cmd == SPECTRE_V2_CMD_EIBRS_RETPOLINE) &&
1260             !IS_ENABLED(CONFIG_RETPOLINE)) {
1261                 pr_err("%s selected but not compiled in. Switching to AUTO select\n",
1262                        mitigation_options[i].option);
1263                 return SPECTRE_V2_CMD_AUTO;
1264         }
1265
1266         if ((cmd == SPECTRE_V2_CMD_EIBRS ||
1267              cmd == SPECTRE_V2_CMD_EIBRS_LFENCE ||
1268              cmd == SPECTRE_V2_CMD_EIBRS_RETPOLINE) &&
1269             !boot_cpu_has(X86_FEATURE_IBRS_ENHANCED)) {
1270                 pr_err("%s selected but CPU doesn't have eIBRS. Switching to AUTO select\n",
1271                        mitigation_options[i].option);
1272                 return SPECTRE_V2_CMD_AUTO;
1273         }
1274
1275         if ((cmd == SPECTRE_V2_CMD_RETPOLINE_LFENCE ||
1276              cmd == SPECTRE_V2_CMD_EIBRS_LFENCE) &&
1277             !boot_cpu_has(X86_FEATURE_LFENCE_RDTSC)) {
1278                 pr_err("%s selected, but CPU doesn't have a serializing LFENCE. Switching to AUTO select\n",
1279                        mitigation_options[i].option);
1280                 return SPECTRE_V2_CMD_AUTO;
1281         }
1282
1283         if (cmd == SPECTRE_V2_CMD_IBRS && !IS_ENABLED(CONFIG_CPU_IBRS_ENTRY)) {
1284                 pr_err("%s selected but not compiled in. Switching to AUTO select\n",
1285                        mitigation_options[i].option);
1286                 return SPECTRE_V2_CMD_AUTO;
1287         }
1288
1289         if (cmd == SPECTRE_V2_CMD_IBRS && boot_cpu_data.x86_vendor != X86_VENDOR_INTEL) {
1290                 pr_err("%s selected but not Intel CPU. Switching to AUTO select\n",
1291                        mitigation_options[i].option);
1292                 return SPECTRE_V2_CMD_AUTO;
1293         }
1294
1295         if (cmd == SPECTRE_V2_CMD_IBRS && !boot_cpu_has(X86_FEATURE_IBRS)) {
1296                 pr_err("%s selected but CPU doesn't have IBRS. Switching to AUTO select\n",
1297                        mitigation_options[i].option);
1298                 return SPECTRE_V2_CMD_AUTO;
1299         }
1300
1301         if (cmd == SPECTRE_V2_CMD_IBRS && boot_cpu_has(X86_FEATURE_XENPV)) {
1302                 pr_err("%s selected but running as XenPV guest. Switching to AUTO select\n",
1303                        mitigation_options[i].option);
1304                 return SPECTRE_V2_CMD_AUTO;
1305         }
1306
1307         spec_v2_print_cond(mitigation_options[i].option,
1308                            mitigation_options[i].secure);
1309         return cmd;
1310 }
1311
1312 static enum spectre_v2_mitigation __init spectre_v2_select_retpoline(void)
1313 {
1314         if (!IS_ENABLED(CONFIG_RETPOLINE)) {
1315                 pr_err("Kernel not compiled with retpoline; no mitigation available!");
1316                 return SPECTRE_V2_NONE;
1317         }
1318
1319         return SPECTRE_V2_RETPOLINE;
1320 }
1321
1322 /* Disable in-kernel use of non-RSB RET predictors */
1323 static void __init spec_ctrl_disable_kernel_rrsba(void)
1324 {
1325         u64 ia32_cap;
1326
1327         if (!boot_cpu_has(X86_FEATURE_RRSBA_CTRL))
1328                 return;
1329
1330         ia32_cap = x86_read_arch_cap_msr();
1331
1332         if (ia32_cap & ARCH_CAP_RRSBA) {
1333                 x86_spec_ctrl_base |= SPEC_CTRL_RRSBA_DIS_S;
1334                 write_spec_ctrl_current(x86_spec_ctrl_base, true);
1335         }
1336 }
1337
1338 static void __init spectre_v2_select_mitigation(void)
1339 {
1340         enum spectre_v2_mitigation_cmd cmd = spectre_v2_parse_cmdline();
1341         enum spectre_v2_mitigation mode = SPECTRE_V2_NONE;
1342
1343         /*
1344          * If the CPU is not affected and the command line mode is NONE or AUTO
1345          * then nothing to do.
1346          */
1347         if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V2) &&
1348             (cmd == SPECTRE_V2_CMD_NONE || cmd == SPECTRE_V2_CMD_AUTO))
1349                 return;
1350
1351         switch (cmd) {
1352         case SPECTRE_V2_CMD_NONE:
1353                 return;
1354
1355         case SPECTRE_V2_CMD_FORCE:
1356         case SPECTRE_V2_CMD_AUTO:
1357                 if (boot_cpu_has(X86_FEATURE_IBRS_ENHANCED)) {
1358                         mode = SPECTRE_V2_EIBRS;
1359                         break;
1360                 }
1361
1362                 if (IS_ENABLED(CONFIG_CPU_IBRS_ENTRY) &&
1363                     boot_cpu_has_bug(X86_BUG_RETBLEED) &&
1364                     retbleed_cmd != RETBLEED_CMD_OFF &&
1365                     boot_cpu_has(X86_FEATURE_IBRS) &&
1366                     boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) {
1367                         mode = SPECTRE_V2_IBRS;
1368                         break;
1369                 }
1370
1371                 mode = spectre_v2_select_retpoline();
1372                 break;
1373
1374         case SPECTRE_V2_CMD_RETPOLINE_LFENCE:
1375                 pr_err(SPECTRE_V2_LFENCE_MSG);
1376                 mode = SPECTRE_V2_LFENCE;
1377                 break;
1378
1379         case SPECTRE_V2_CMD_RETPOLINE_GENERIC:
1380                 mode = SPECTRE_V2_RETPOLINE;
1381                 break;
1382
1383         case SPECTRE_V2_CMD_RETPOLINE:
1384                 mode = spectre_v2_select_retpoline();
1385                 break;
1386
1387         case SPECTRE_V2_CMD_IBRS:
1388                 mode = SPECTRE_V2_IBRS;
1389                 break;
1390
1391         case SPECTRE_V2_CMD_EIBRS:
1392                 mode = SPECTRE_V2_EIBRS;
1393                 break;
1394
1395         case SPECTRE_V2_CMD_EIBRS_LFENCE:
1396                 mode = SPECTRE_V2_EIBRS_LFENCE;
1397                 break;
1398
1399         case SPECTRE_V2_CMD_EIBRS_RETPOLINE:
1400                 mode = SPECTRE_V2_EIBRS_RETPOLINE;
1401                 break;
1402         }
1403
1404         if (mode == SPECTRE_V2_EIBRS && unprivileged_ebpf_enabled())
1405                 pr_err(SPECTRE_V2_EIBRS_EBPF_MSG);
1406
1407         if (spectre_v2_in_ibrs_mode(mode)) {
1408                 x86_spec_ctrl_base |= SPEC_CTRL_IBRS;
1409                 write_spec_ctrl_current(x86_spec_ctrl_base, true);
1410         }
1411
1412         switch (mode) {
1413         case SPECTRE_V2_NONE:
1414         case SPECTRE_V2_EIBRS:
1415                 break;
1416
1417         case SPECTRE_V2_IBRS:
1418                 setup_force_cpu_cap(X86_FEATURE_KERNEL_IBRS);
1419                 if (boot_cpu_has(X86_FEATURE_IBRS_ENHANCED))
1420                         pr_warn(SPECTRE_V2_IBRS_PERF_MSG);
1421                 break;
1422
1423         case SPECTRE_V2_LFENCE:
1424         case SPECTRE_V2_EIBRS_LFENCE:
1425                 setup_force_cpu_cap(X86_FEATURE_RETPOLINE_LFENCE);
1426                 fallthrough;
1427
1428         case SPECTRE_V2_RETPOLINE:
1429         case SPECTRE_V2_EIBRS_RETPOLINE:
1430                 setup_force_cpu_cap(X86_FEATURE_RETPOLINE);
1431                 break;
1432         }
1433
1434         /*
1435          * Disable alternate RSB predictions in kernel when indirect CALLs and
1436          * JMPs gets protection against BHI and Intramode-BTI, but RET
1437          * prediction from a non-RSB predictor is still a risk.
1438          */
1439         if (mode == SPECTRE_V2_EIBRS_LFENCE ||
1440             mode == SPECTRE_V2_EIBRS_RETPOLINE ||
1441             mode == SPECTRE_V2_RETPOLINE)
1442                 spec_ctrl_disable_kernel_rrsba();
1443
1444         spectre_v2_enabled = mode;
1445         pr_info("%s\n", spectre_v2_strings[mode]);
1446
1447         /*
1448          * If Spectre v2 protection has been enabled, fill the RSB during a
1449          * context switch.  In general there are two types of RSB attacks
1450          * across context switches, for which the CALLs/RETs may be unbalanced.
1451          *
1452          * 1) RSB underflow
1453          *
1454          *    Some Intel parts have "bottomless RSB".  When the RSB is empty,
1455          *    speculated return targets may come from the branch predictor,
1456          *    which could have a user-poisoned BTB or BHB entry.
1457          *
1458          *    AMD has it even worse: *all* returns are speculated from the BTB,
1459          *    regardless of the state of the RSB.
1460          *
1461          *    When IBRS or eIBRS is enabled, the "user -> kernel" attack
1462          *    scenario is mitigated by the IBRS branch prediction isolation
1463          *    properties, so the RSB buffer filling wouldn't be necessary to
1464          *    protect against this type of attack.
1465          *
1466          *    The "user -> user" attack scenario is mitigated by RSB filling.
1467          *
1468          * 2) Poisoned RSB entry
1469          *
1470          *    If the 'next' in-kernel return stack is shorter than 'prev',
1471          *    'next' could be tricked into speculating with a user-poisoned RSB
1472          *    entry.
1473          *
1474          *    The "user -> kernel" attack scenario is mitigated by SMEP and
1475          *    eIBRS.
1476          *
1477          *    The "user -> user" scenario, also known as SpectreBHB, requires
1478          *    RSB clearing.
1479          *
1480          * So to mitigate all cases, unconditionally fill RSB on context
1481          * switches.
1482          *
1483          * FIXME: Is this pointless for retbleed-affected AMD?
1484          */
1485         setup_force_cpu_cap(X86_FEATURE_RSB_CTXSW);
1486         pr_info("Spectre v2 / SpectreRSB mitigation: Filling RSB on context switch\n");
1487
1488         /*
1489          * Similar to context switches, there are two types of RSB attacks
1490          * after vmexit:
1491          *
1492          * 1) RSB underflow
1493          *
1494          * 2) Poisoned RSB entry
1495          *
1496          * When retpoline is enabled, both are mitigated by filling/clearing
1497          * the RSB.
1498          *
1499          * When IBRS is enabled, while #1 would be mitigated by the IBRS branch
1500          * prediction isolation protections, RSB still needs to be cleared
1501          * because of #2.  Note that SMEP provides no protection here, unlike
1502          * user-space-poisoned RSB entries.
1503          *
1504          * eIBRS, on the other hand, has RSB-poisoning protections, so it
1505          * doesn't need RSB clearing after vmexit.
1506          */
1507         if (boot_cpu_has(X86_FEATURE_RETPOLINE) ||
1508             boot_cpu_has(X86_FEATURE_KERNEL_IBRS))
1509                 setup_force_cpu_cap(X86_FEATURE_RSB_VMEXIT);
1510
1511         /*
1512          * Retpoline protects the kernel, but doesn't protect firmware.  IBRS
1513          * and Enhanced IBRS protect firmware too, so enable IBRS around
1514          * firmware calls only when IBRS / Enhanced IBRS aren't otherwise
1515          * enabled.
1516          *
1517          * Use "mode" to check Enhanced IBRS instead of boot_cpu_has(), because
1518          * the user might select retpoline on the kernel command line and if
1519          * the CPU supports Enhanced IBRS, kernel might un-intentionally not
1520          * enable IBRS around firmware calls.
1521          */
1522         if (boot_cpu_has_bug(X86_BUG_RETBLEED) &&
1523             boot_cpu_has(X86_FEATURE_IBPB) &&
1524             (boot_cpu_data.x86_vendor == X86_VENDOR_AMD ||
1525              boot_cpu_data.x86_vendor == X86_VENDOR_HYGON)) {
1526
1527                 if (retbleed_cmd != RETBLEED_CMD_IBPB) {
1528                         setup_force_cpu_cap(X86_FEATURE_USE_IBPB_FW);
1529                         pr_info("Enabling Speculation Barrier for firmware calls\n");
1530                 }
1531
1532         } else if (boot_cpu_has(X86_FEATURE_IBRS) && !spectre_v2_in_ibrs_mode(mode)) {
1533                 setup_force_cpu_cap(X86_FEATURE_USE_IBRS_FW);
1534                 pr_info("Enabling Restricted Speculation for firmware calls\n");
1535         }
1536
1537         /* Set up IBPB and STIBP depending on the general spectre V2 command */
1538         spectre_v2_cmd = cmd;
1539 }
1540
1541 static void update_stibp_msr(void * __unused)
1542 {
1543         u64 val = spec_ctrl_current() | (x86_spec_ctrl_base & SPEC_CTRL_STIBP);
1544         write_spec_ctrl_current(val, true);
1545 }
1546
1547 /* Update x86_spec_ctrl_base in case SMT state changed. */
1548 static void update_stibp_strict(void)
1549 {
1550         u64 mask = x86_spec_ctrl_base & ~SPEC_CTRL_STIBP;
1551
1552         if (sched_smt_active())
1553                 mask |= SPEC_CTRL_STIBP;
1554
1555         if (mask == x86_spec_ctrl_base)
1556                 return;
1557
1558         pr_info("Update user space SMT mitigation: STIBP %s\n",
1559                 mask & SPEC_CTRL_STIBP ? "always-on" : "off");
1560         x86_spec_ctrl_base = mask;
1561         on_each_cpu(update_stibp_msr, NULL, 1);
1562 }
1563
1564 /* Update the static key controlling the evaluation of TIF_SPEC_IB */
1565 static void update_indir_branch_cond(void)
1566 {
1567         if (sched_smt_active())
1568                 static_branch_enable(&switch_to_cond_stibp);
1569         else
1570                 static_branch_disable(&switch_to_cond_stibp);
1571 }
1572
1573 #undef pr_fmt
1574 #define pr_fmt(fmt) fmt
1575
1576 /* Update the static key controlling the MDS CPU buffer clear in idle */
1577 static void update_mds_branch_idle(void)
1578 {
1579         u64 ia32_cap = x86_read_arch_cap_msr();
1580
1581         /*
1582          * Enable the idle clearing if SMT is active on CPUs which are
1583          * affected only by MSBDS and not any other MDS variant.
1584          *
1585          * The other variants cannot be mitigated when SMT is enabled, so
1586          * clearing the buffers on idle just to prevent the Store Buffer
1587          * repartitioning leak would be a window dressing exercise.
1588          */
1589         if (!boot_cpu_has_bug(X86_BUG_MSBDS_ONLY))
1590                 return;
1591
1592         if (sched_smt_active()) {
1593                 static_branch_enable(&mds_idle_clear);
1594         } else if (mmio_mitigation == MMIO_MITIGATION_OFF ||
1595                    (ia32_cap & ARCH_CAP_FBSDP_NO)) {
1596                 static_branch_disable(&mds_idle_clear);
1597         }
1598 }
1599
1600 #define MDS_MSG_SMT "MDS CPU bug present and SMT on, data leak possible. See https://www.kernel.org/doc/html/latest/admin-guide/hw-vuln/mds.html for more details.\n"
1601 #define TAA_MSG_SMT "TAA CPU bug present and SMT on, data leak possible. See https://www.kernel.org/doc/html/latest/admin-guide/hw-vuln/tsx_async_abort.html for more details.\n"
1602 #define MMIO_MSG_SMT "MMIO Stale Data CPU bug present and SMT on, data leak possible. See https://www.kernel.org/doc/html/latest/admin-guide/hw-vuln/processor_mmio_stale_data.html for more details.\n"
1603
1604 void cpu_bugs_smt_update(void)
1605 {
1606         mutex_lock(&spec_ctrl_mutex);
1607
1608         if (sched_smt_active() && unprivileged_ebpf_enabled() &&
1609             spectre_v2_enabled == SPECTRE_V2_EIBRS_LFENCE)
1610                 pr_warn_once(SPECTRE_V2_EIBRS_LFENCE_EBPF_SMT_MSG);
1611
1612         switch (spectre_v2_user_stibp) {
1613         case SPECTRE_V2_USER_NONE:
1614                 break;
1615         case SPECTRE_V2_USER_STRICT:
1616         case SPECTRE_V2_USER_STRICT_PREFERRED:
1617                 update_stibp_strict();
1618                 break;
1619         case SPECTRE_V2_USER_PRCTL:
1620         case SPECTRE_V2_USER_SECCOMP:
1621                 update_indir_branch_cond();
1622                 break;
1623         }
1624
1625         switch (mds_mitigation) {
1626         case MDS_MITIGATION_FULL:
1627         case MDS_MITIGATION_VMWERV:
1628                 if (sched_smt_active() && !boot_cpu_has(X86_BUG_MSBDS_ONLY))
1629                         pr_warn_once(MDS_MSG_SMT);
1630                 update_mds_branch_idle();
1631                 break;
1632         case MDS_MITIGATION_OFF:
1633                 break;
1634         }
1635
1636         switch (taa_mitigation) {
1637         case TAA_MITIGATION_VERW:
1638         case TAA_MITIGATION_UCODE_NEEDED:
1639                 if (sched_smt_active())
1640                         pr_warn_once(TAA_MSG_SMT);
1641                 break;
1642         case TAA_MITIGATION_TSX_DISABLED:
1643         case TAA_MITIGATION_OFF:
1644                 break;
1645         }
1646
1647         switch (mmio_mitigation) {
1648         case MMIO_MITIGATION_VERW:
1649         case MMIO_MITIGATION_UCODE_NEEDED:
1650                 if (sched_smt_active())
1651                         pr_warn_once(MMIO_MSG_SMT);
1652                 break;
1653         case MMIO_MITIGATION_OFF:
1654                 break;
1655         }
1656
1657         mutex_unlock(&spec_ctrl_mutex);
1658 }
1659
1660 #undef pr_fmt
1661 #define pr_fmt(fmt)     "Speculative Store Bypass: " fmt
1662
1663 static enum ssb_mitigation ssb_mode __ro_after_init = SPEC_STORE_BYPASS_NONE;
1664
1665 /* The kernel command line selection */
1666 enum ssb_mitigation_cmd {
1667         SPEC_STORE_BYPASS_CMD_NONE,
1668         SPEC_STORE_BYPASS_CMD_AUTO,
1669         SPEC_STORE_BYPASS_CMD_ON,
1670         SPEC_STORE_BYPASS_CMD_PRCTL,
1671         SPEC_STORE_BYPASS_CMD_SECCOMP,
1672 };
1673
1674 static const char * const ssb_strings[] = {
1675         [SPEC_STORE_BYPASS_NONE]        = "Vulnerable",
1676         [SPEC_STORE_BYPASS_DISABLE]     = "Mitigation: Speculative Store Bypass disabled",
1677         [SPEC_STORE_BYPASS_PRCTL]       = "Mitigation: Speculative Store Bypass disabled via prctl",
1678         [SPEC_STORE_BYPASS_SECCOMP]     = "Mitigation: Speculative Store Bypass disabled via prctl and seccomp",
1679 };
1680
1681 static const struct {
1682         const char *option;
1683         enum ssb_mitigation_cmd cmd;
1684 } ssb_mitigation_options[]  __initconst = {
1685         { "auto",       SPEC_STORE_BYPASS_CMD_AUTO },    /* Platform decides */
1686         { "on",         SPEC_STORE_BYPASS_CMD_ON },      /* Disable Speculative Store Bypass */
1687         { "off",        SPEC_STORE_BYPASS_CMD_NONE },    /* Don't touch Speculative Store Bypass */
1688         { "prctl",      SPEC_STORE_BYPASS_CMD_PRCTL },   /* Disable Speculative Store Bypass via prctl */
1689         { "seccomp",    SPEC_STORE_BYPASS_CMD_SECCOMP }, /* Disable Speculative Store Bypass via prctl and seccomp */
1690 };
1691
1692 static enum ssb_mitigation_cmd __init ssb_parse_cmdline(void)
1693 {
1694         enum ssb_mitigation_cmd cmd = SPEC_STORE_BYPASS_CMD_AUTO;
1695         char arg[20];
1696         int ret, i;
1697
1698         if (cmdline_find_option_bool(boot_command_line, "nospec_store_bypass_disable") ||
1699             cpu_mitigations_off()) {
1700                 return SPEC_STORE_BYPASS_CMD_NONE;
1701         } else {
1702                 ret = cmdline_find_option(boot_command_line, "spec_store_bypass_disable",
1703                                           arg, sizeof(arg));
1704                 if (ret < 0)
1705                         return SPEC_STORE_BYPASS_CMD_AUTO;
1706
1707                 for (i = 0; i < ARRAY_SIZE(ssb_mitigation_options); i++) {
1708                         if (!match_option(arg, ret, ssb_mitigation_options[i].option))
1709                                 continue;
1710
1711                         cmd = ssb_mitigation_options[i].cmd;
1712                         break;
1713                 }
1714
1715                 if (i >= ARRAY_SIZE(ssb_mitigation_options)) {
1716                         pr_err("unknown option (%s). Switching to AUTO select\n", arg);
1717                         return SPEC_STORE_BYPASS_CMD_AUTO;
1718                 }
1719         }
1720
1721         return cmd;
1722 }
1723
1724 static enum ssb_mitigation __init __ssb_select_mitigation(void)
1725 {
1726         enum ssb_mitigation mode = SPEC_STORE_BYPASS_NONE;
1727         enum ssb_mitigation_cmd cmd;
1728
1729         if (!boot_cpu_has(X86_FEATURE_SSBD))
1730                 return mode;
1731
1732         cmd = ssb_parse_cmdline();
1733         if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS) &&
1734             (cmd == SPEC_STORE_BYPASS_CMD_NONE ||
1735              cmd == SPEC_STORE_BYPASS_CMD_AUTO))
1736                 return mode;
1737
1738         switch (cmd) {
1739         case SPEC_STORE_BYPASS_CMD_SECCOMP:
1740                 /*
1741                  * Choose prctl+seccomp as the default mode if seccomp is
1742                  * enabled.
1743                  */
1744                 if (IS_ENABLED(CONFIG_SECCOMP))
1745                         mode = SPEC_STORE_BYPASS_SECCOMP;
1746                 else
1747                         mode = SPEC_STORE_BYPASS_PRCTL;
1748                 break;
1749         case SPEC_STORE_BYPASS_CMD_ON:
1750                 mode = SPEC_STORE_BYPASS_DISABLE;
1751                 break;
1752         case SPEC_STORE_BYPASS_CMD_AUTO:
1753         case SPEC_STORE_BYPASS_CMD_PRCTL:
1754                 mode = SPEC_STORE_BYPASS_PRCTL;
1755                 break;
1756         case SPEC_STORE_BYPASS_CMD_NONE:
1757                 break;
1758         }
1759
1760         /*
1761          * We have three CPU feature flags that are in play here:
1762          *  - X86_BUG_SPEC_STORE_BYPASS - CPU is susceptible.
1763          *  - X86_FEATURE_SSBD - CPU is able to turn off speculative store bypass
1764          *  - X86_FEATURE_SPEC_STORE_BYPASS_DISABLE - engage the mitigation
1765          */
1766         if (mode == SPEC_STORE_BYPASS_DISABLE) {
1767                 setup_force_cpu_cap(X86_FEATURE_SPEC_STORE_BYPASS_DISABLE);
1768                 /*
1769                  * Intel uses the SPEC CTRL MSR Bit(2) for this, while AMD may
1770                  * use a completely different MSR and bit dependent on family.
1771                  */
1772                 if (!static_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) &&
1773                     !static_cpu_has(X86_FEATURE_AMD_SSBD)) {
1774                         x86_amd_ssb_disable();
1775                 } else {
1776                         x86_spec_ctrl_base |= SPEC_CTRL_SSBD;
1777                         write_spec_ctrl_current(x86_spec_ctrl_base, true);
1778                 }
1779         }
1780
1781         return mode;
1782 }
1783
1784 static void ssb_select_mitigation(void)
1785 {
1786         ssb_mode = __ssb_select_mitigation();
1787
1788         if (boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1789                 pr_info("%s\n", ssb_strings[ssb_mode]);
1790 }
1791
1792 #undef pr_fmt
1793 #define pr_fmt(fmt)     "Speculation prctl: " fmt
1794
1795 static void task_update_spec_tif(struct task_struct *tsk)
1796 {
1797         /* Force the update of the real TIF bits */
1798         set_tsk_thread_flag(tsk, TIF_SPEC_FORCE_UPDATE);
1799
1800         /*
1801          * Immediately update the speculation control MSRs for the current
1802          * task, but for a non-current task delay setting the CPU
1803          * mitigation until it is scheduled next.
1804          *
1805          * This can only happen for SECCOMP mitigation. For PRCTL it's
1806          * always the current task.
1807          */
1808         if (tsk == current)
1809                 speculation_ctrl_update_current();
1810 }
1811
1812 static int l1d_flush_prctl_set(struct task_struct *task, unsigned long ctrl)
1813 {
1814
1815         if (!static_branch_unlikely(&switch_mm_cond_l1d_flush))
1816                 return -EPERM;
1817
1818         switch (ctrl) {
1819         case PR_SPEC_ENABLE:
1820                 set_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH);
1821                 return 0;
1822         case PR_SPEC_DISABLE:
1823                 clear_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH);
1824                 return 0;
1825         default:
1826                 return -ERANGE;
1827         }
1828 }
1829
1830 static int ssb_prctl_set(struct task_struct *task, unsigned long ctrl)
1831 {
1832         if (ssb_mode != SPEC_STORE_BYPASS_PRCTL &&
1833             ssb_mode != SPEC_STORE_BYPASS_SECCOMP)
1834                 return -ENXIO;
1835
1836         switch (ctrl) {
1837         case PR_SPEC_ENABLE:
1838                 /* If speculation is force disabled, enable is not allowed */
1839                 if (task_spec_ssb_force_disable(task))
1840                         return -EPERM;
1841                 task_clear_spec_ssb_disable(task);
1842                 task_clear_spec_ssb_noexec(task);
1843                 task_update_spec_tif(task);
1844                 break;
1845         case PR_SPEC_DISABLE:
1846                 task_set_spec_ssb_disable(task);
1847                 task_clear_spec_ssb_noexec(task);
1848                 task_update_spec_tif(task);
1849                 break;
1850         case PR_SPEC_FORCE_DISABLE:
1851                 task_set_spec_ssb_disable(task);
1852                 task_set_spec_ssb_force_disable(task);
1853                 task_clear_spec_ssb_noexec(task);
1854                 task_update_spec_tif(task);
1855                 break;
1856         case PR_SPEC_DISABLE_NOEXEC:
1857                 if (task_spec_ssb_force_disable(task))
1858                         return -EPERM;
1859                 task_set_spec_ssb_disable(task);
1860                 task_set_spec_ssb_noexec(task);
1861                 task_update_spec_tif(task);
1862                 break;
1863         default:
1864                 return -ERANGE;
1865         }
1866         return 0;
1867 }
1868
1869 static bool is_spec_ib_user_controlled(void)
1870 {
1871         return spectre_v2_user_ibpb == SPECTRE_V2_USER_PRCTL ||
1872                 spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP ||
1873                 spectre_v2_user_stibp == SPECTRE_V2_USER_PRCTL ||
1874                 spectre_v2_user_stibp == SPECTRE_V2_USER_SECCOMP;
1875 }
1876
1877 static int ib_prctl_set(struct task_struct *task, unsigned long ctrl)
1878 {
1879         switch (ctrl) {
1880         case PR_SPEC_ENABLE:
1881                 if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
1882                     spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
1883                         return 0;
1884
1885                 /*
1886                  * With strict mode for both IBPB and STIBP, the instruction
1887                  * code paths avoid checking this task flag and instead,
1888                  * unconditionally run the instruction. However, STIBP and IBPB
1889                  * are independent and either can be set to conditionally
1890                  * enabled regardless of the mode of the other.
1891                  *
1892                  * If either is set to conditional, allow the task flag to be
1893                  * updated, unless it was force-disabled by a previous prctl
1894                  * call. Currently, this is possible on an AMD CPU which has the
1895                  * feature X86_FEATURE_AMD_STIBP_ALWAYS_ON. In this case, if the
1896                  * kernel is booted with 'spectre_v2_user=seccomp', then
1897                  * spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP and
1898                  * spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED.
1899                  */
1900                 if (!is_spec_ib_user_controlled() ||
1901                     task_spec_ib_force_disable(task))
1902                         return -EPERM;
1903
1904                 task_clear_spec_ib_disable(task);
1905                 task_update_spec_tif(task);
1906                 break;
1907         case PR_SPEC_DISABLE:
1908         case PR_SPEC_FORCE_DISABLE:
1909                 /*
1910                  * Indirect branch speculation is always allowed when
1911                  * mitigation is force disabled.
1912                  */
1913                 if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
1914                     spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
1915                         return -EPERM;
1916
1917                 if (!is_spec_ib_user_controlled())
1918                         return 0;
1919
1920                 task_set_spec_ib_disable(task);
1921                 if (ctrl == PR_SPEC_FORCE_DISABLE)
1922                         task_set_spec_ib_force_disable(task);
1923                 task_update_spec_tif(task);
1924                 break;
1925         default:
1926                 return -ERANGE;
1927         }
1928         return 0;
1929 }
1930
1931 int arch_prctl_spec_ctrl_set(struct task_struct *task, unsigned long which,
1932                              unsigned long ctrl)
1933 {
1934         switch (which) {
1935         case PR_SPEC_STORE_BYPASS:
1936                 return ssb_prctl_set(task, ctrl);
1937         case PR_SPEC_INDIRECT_BRANCH:
1938                 return ib_prctl_set(task, ctrl);
1939         case PR_SPEC_L1D_FLUSH:
1940                 return l1d_flush_prctl_set(task, ctrl);
1941         default:
1942                 return -ENODEV;
1943         }
1944 }
1945
1946 #ifdef CONFIG_SECCOMP
1947 void arch_seccomp_spec_mitigate(struct task_struct *task)
1948 {
1949         if (ssb_mode == SPEC_STORE_BYPASS_SECCOMP)
1950                 ssb_prctl_set(task, PR_SPEC_FORCE_DISABLE);
1951         if (spectre_v2_user_ibpb == SPECTRE_V2_USER_SECCOMP ||
1952             spectre_v2_user_stibp == SPECTRE_V2_USER_SECCOMP)
1953                 ib_prctl_set(task, PR_SPEC_FORCE_DISABLE);
1954 }
1955 #endif
1956
1957 static int l1d_flush_prctl_get(struct task_struct *task)
1958 {
1959         if (!static_branch_unlikely(&switch_mm_cond_l1d_flush))
1960                 return PR_SPEC_FORCE_DISABLE;
1961
1962         if (test_ti_thread_flag(&task->thread_info, TIF_SPEC_L1D_FLUSH))
1963                 return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
1964         else
1965                 return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
1966 }
1967
1968 static int ssb_prctl_get(struct task_struct *task)
1969 {
1970         switch (ssb_mode) {
1971         case SPEC_STORE_BYPASS_DISABLE:
1972                 return PR_SPEC_DISABLE;
1973         case SPEC_STORE_BYPASS_SECCOMP:
1974         case SPEC_STORE_BYPASS_PRCTL:
1975                 if (task_spec_ssb_force_disable(task))
1976                         return PR_SPEC_PRCTL | PR_SPEC_FORCE_DISABLE;
1977                 if (task_spec_ssb_noexec(task))
1978                         return PR_SPEC_PRCTL | PR_SPEC_DISABLE_NOEXEC;
1979                 if (task_spec_ssb_disable(task))
1980                         return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
1981                 return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
1982         default:
1983                 if (boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1984                         return PR_SPEC_ENABLE;
1985                 return PR_SPEC_NOT_AFFECTED;
1986         }
1987 }
1988
1989 static int ib_prctl_get(struct task_struct *task)
1990 {
1991         if (!boot_cpu_has_bug(X86_BUG_SPECTRE_V2))
1992                 return PR_SPEC_NOT_AFFECTED;
1993
1994         if (spectre_v2_user_ibpb == SPECTRE_V2_USER_NONE &&
1995             spectre_v2_user_stibp == SPECTRE_V2_USER_NONE)
1996                 return PR_SPEC_ENABLE;
1997         else if (is_spec_ib_user_controlled()) {
1998                 if (task_spec_ib_force_disable(task))
1999                         return PR_SPEC_PRCTL | PR_SPEC_FORCE_DISABLE;
2000                 if (task_spec_ib_disable(task))
2001                         return PR_SPEC_PRCTL | PR_SPEC_DISABLE;
2002                 return PR_SPEC_PRCTL | PR_SPEC_ENABLE;
2003         } else if (spectre_v2_user_ibpb == SPECTRE_V2_USER_STRICT ||
2004             spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT ||
2005             spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED)
2006                 return PR_SPEC_DISABLE;
2007         else
2008                 return PR_SPEC_NOT_AFFECTED;
2009 }
2010
2011 int arch_prctl_spec_ctrl_get(struct task_struct *task, unsigned long which)
2012 {
2013         switch (which) {
2014         case PR_SPEC_STORE_BYPASS:
2015                 return ssb_prctl_get(task);
2016         case PR_SPEC_INDIRECT_BRANCH:
2017                 return ib_prctl_get(task);
2018         case PR_SPEC_L1D_FLUSH:
2019                 return l1d_flush_prctl_get(task);
2020         default:
2021                 return -ENODEV;
2022         }
2023 }
2024
2025 void x86_spec_ctrl_setup_ap(void)
2026 {
2027         if (boot_cpu_has(X86_FEATURE_MSR_SPEC_CTRL))
2028                 write_spec_ctrl_current(x86_spec_ctrl_base, true);
2029
2030         if (ssb_mode == SPEC_STORE_BYPASS_DISABLE)
2031                 x86_amd_ssb_disable();
2032 }
2033
2034 bool itlb_multihit_kvm_mitigation;
2035 EXPORT_SYMBOL_GPL(itlb_multihit_kvm_mitigation);
2036
2037 #undef pr_fmt
2038 #define pr_fmt(fmt)     "L1TF: " fmt
2039
2040 /* Default mitigation for L1TF-affected CPUs */
2041 enum l1tf_mitigations l1tf_mitigation __ro_after_init = L1TF_MITIGATION_FLUSH;
2042 #if IS_ENABLED(CONFIG_KVM_INTEL)
2043 EXPORT_SYMBOL_GPL(l1tf_mitigation);
2044 #endif
2045 enum vmx_l1d_flush_state l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
2046 EXPORT_SYMBOL_GPL(l1tf_vmx_mitigation);
2047
2048 /*
2049  * These CPUs all support 44bits physical address space internally in the
2050  * cache but CPUID can report a smaller number of physical address bits.
2051  *
2052  * The L1TF mitigation uses the top most address bit for the inversion of
2053  * non present PTEs. When the installed memory reaches into the top most
2054  * address bit due to memory holes, which has been observed on machines
2055  * which report 36bits physical address bits and have 32G RAM installed,
2056  * then the mitigation range check in l1tf_select_mitigation() triggers.
2057  * This is a false positive because the mitigation is still possible due to
2058  * the fact that the cache uses 44bit internally. Use the cache bits
2059  * instead of the reported physical bits and adjust them on the affected
2060  * machines to 44bit if the reported bits are less than 44.
2061  */
2062 static void override_cache_bits(struct cpuinfo_x86 *c)
2063 {
2064         if (c->x86 != 6)
2065                 return;
2066
2067         switch (c->x86_model) {
2068         case INTEL_FAM6_NEHALEM:
2069         case INTEL_FAM6_WESTMERE:
2070         case INTEL_FAM6_SANDYBRIDGE:
2071         case INTEL_FAM6_IVYBRIDGE:
2072         case INTEL_FAM6_HASWELL:
2073         case INTEL_FAM6_HASWELL_L:
2074         case INTEL_FAM6_HASWELL_G:
2075         case INTEL_FAM6_BROADWELL:
2076         case INTEL_FAM6_BROADWELL_G:
2077         case INTEL_FAM6_SKYLAKE_L:
2078         case INTEL_FAM6_SKYLAKE:
2079         case INTEL_FAM6_KABYLAKE_L:
2080         case INTEL_FAM6_KABYLAKE:
2081                 if (c->x86_cache_bits < 44)
2082                         c->x86_cache_bits = 44;
2083                 break;
2084         }
2085 }
2086
2087 static void __init l1tf_select_mitigation(void)
2088 {
2089         u64 half_pa;
2090
2091         if (!boot_cpu_has_bug(X86_BUG_L1TF))
2092                 return;
2093
2094         if (cpu_mitigations_off())
2095                 l1tf_mitigation = L1TF_MITIGATION_OFF;
2096         else if (cpu_mitigations_auto_nosmt())
2097                 l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOSMT;
2098
2099         override_cache_bits(&boot_cpu_data);
2100
2101         switch (l1tf_mitigation) {
2102         case L1TF_MITIGATION_OFF:
2103         case L1TF_MITIGATION_FLUSH_NOWARN:
2104         case L1TF_MITIGATION_FLUSH:
2105                 break;
2106         case L1TF_MITIGATION_FLUSH_NOSMT:
2107         case L1TF_MITIGATION_FULL:
2108                 cpu_smt_disable(false);
2109                 break;
2110         case L1TF_MITIGATION_FULL_FORCE:
2111                 cpu_smt_disable(true);
2112                 break;
2113         }
2114
2115 #if CONFIG_PGTABLE_LEVELS == 2
2116         pr_warn("Kernel not compiled for PAE. No mitigation for L1TF\n");
2117         return;
2118 #endif
2119
2120         half_pa = (u64)l1tf_pfn_limit() << PAGE_SHIFT;
2121         if (l1tf_mitigation != L1TF_MITIGATION_OFF &&
2122                         e820__mapped_any(half_pa, ULLONG_MAX - half_pa, E820_TYPE_RAM)) {
2123                 pr_warn("System has more than MAX_PA/2 memory. L1TF mitigation not effective.\n");
2124                 pr_info("You may make it effective by booting the kernel with mem=%llu parameter.\n",
2125                                 half_pa);
2126                 pr_info("However, doing so will make a part of your RAM unusable.\n");
2127                 pr_info("Reading https://www.kernel.org/doc/html/latest/admin-guide/hw-vuln/l1tf.html might help you decide.\n");
2128                 return;
2129         }
2130
2131         setup_force_cpu_cap(X86_FEATURE_L1TF_PTEINV);
2132 }
2133
2134 static int __init l1tf_cmdline(char *str)
2135 {
2136         if (!boot_cpu_has_bug(X86_BUG_L1TF))
2137                 return 0;
2138
2139         if (!str)
2140                 return -EINVAL;
2141
2142         if (!strcmp(str, "off"))
2143                 l1tf_mitigation = L1TF_MITIGATION_OFF;
2144         else if (!strcmp(str, "flush,nowarn"))
2145                 l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOWARN;
2146         else if (!strcmp(str, "flush"))
2147                 l1tf_mitigation = L1TF_MITIGATION_FLUSH;
2148         else if (!strcmp(str, "flush,nosmt"))
2149                 l1tf_mitigation = L1TF_MITIGATION_FLUSH_NOSMT;
2150         else if (!strcmp(str, "full"))
2151                 l1tf_mitigation = L1TF_MITIGATION_FULL;
2152         else if (!strcmp(str, "full,force"))
2153                 l1tf_mitigation = L1TF_MITIGATION_FULL_FORCE;
2154
2155         return 0;
2156 }
2157 early_param("l1tf", l1tf_cmdline);
2158
2159 #undef pr_fmt
2160 #define pr_fmt(fmt) fmt
2161
2162 #ifdef CONFIG_SYSFS
2163
2164 #define L1TF_DEFAULT_MSG "Mitigation: PTE Inversion"
2165
2166 #if IS_ENABLED(CONFIG_KVM_INTEL)
2167 static const char * const l1tf_vmx_states[] = {
2168         [VMENTER_L1D_FLUSH_AUTO]                = "auto",
2169         [VMENTER_L1D_FLUSH_NEVER]               = "vulnerable",
2170         [VMENTER_L1D_FLUSH_COND]                = "conditional cache flushes",
2171         [VMENTER_L1D_FLUSH_ALWAYS]              = "cache flushes",
2172         [VMENTER_L1D_FLUSH_EPT_DISABLED]        = "EPT disabled",
2173         [VMENTER_L1D_FLUSH_NOT_REQUIRED]        = "flush not necessary"
2174 };
2175
2176 static ssize_t l1tf_show_state(char *buf)
2177 {
2178         if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_AUTO)
2179                 return sprintf(buf, "%s\n", L1TF_DEFAULT_MSG);
2180
2181         if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_EPT_DISABLED ||
2182             (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER &&
2183              sched_smt_active())) {
2184                 return sprintf(buf, "%s; VMX: %s\n", L1TF_DEFAULT_MSG,
2185                                l1tf_vmx_states[l1tf_vmx_mitigation]);
2186         }
2187
2188         return sprintf(buf, "%s; VMX: %s, SMT %s\n", L1TF_DEFAULT_MSG,
2189                        l1tf_vmx_states[l1tf_vmx_mitigation],
2190                        sched_smt_active() ? "vulnerable" : "disabled");
2191 }
2192
2193 static ssize_t itlb_multihit_show_state(char *buf)
2194 {
2195         if (!boot_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
2196             !boot_cpu_has(X86_FEATURE_VMX))
2197                 return sprintf(buf, "KVM: Mitigation: VMX unsupported\n");
2198         else if (!(cr4_read_shadow() & X86_CR4_VMXE))
2199                 return sprintf(buf, "KVM: Mitigation: VMX disabled\n");
2200         else if (itlb_multihit_kvm_mitigation)
2201                 return sprintf(buf, "KVM: Mitigation: Split huge pages\n");
2202         else
2203                 return sprintf(buf, "KVM: Vulnerable\n");
2204 }
2205 #else
2206 static ssize_t l1tf_show_state(char *buf)
2207 {
2208         return sprintf(buf, "%s\n", L1TF_DEFAULT_MSG);
2209 }
2210
2211 static ssize_t itlb_multihit_show_state(char *buf)
2212 {
2213         return sprintf(buf, "Processor vulnerable\n");
2214 }
2215 #endif
2216
2217 static ssize_t mds_show_state(char *buf)
2218 {
2219         if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
2220                 return sprintf(buf, "%s; SMT Host state unknown\n",
2221                                mds_strings[mds_mitigation]);
2222         }
2223
2224         if (boot_cpu_has(X86_BUG_MSBDS_ONLY)) {
2225                 return sprintf(buf, "%s; SMT %s\n", mds_strings[mds_mitigation],
2226                                (mds_mitigation == MDS_MITIGATION_OFF ? "vulnerable" :
2227                                 sched_smt_active() ? "mitigated" : "disabled"));
2228         }
2229
2230         return sprintf(buf, "%s; SMT %s\n", mds_strings[mds_mitigation],
2231                        sched_smt_active() ? "vulnerable" : "disabled");
2232 }
2233
2234 static ssize_t tsx_async_abort_show_state(char *buf)
2235 {
2236         if ((taa_mitigation == TAA_MITIGATION_TSX_DISABLED) ||
2237             (taa_mitigation == TAA_MITIGATION_OFF))
2238                 return sprintf(buf, "%s\n", taa_strings[taa_mitigation]);
2239
2240         if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
2241                 return sprintf(buf, "%s; SMT Host state unknown\n",
2242                                taa_strings[taa_mitigation]);
2243         }
2244
2245         return sprintf(buf, "%s; SMT %s\n", taa_strings[taa_mitigation],
2246                        sched_smt_active() ? "vulnerable" : "disabled");
2247 }
2248
2249 static ssize_t mmio_stale_data_show_state(char *buf)
2250 {
2251         if (mmio_mitigation == MMIO_MITIGATION_OFF)
2252                 return sysfs_emit(buf, "%s\n", mmio_strings[mmio_mitigation]);
2253
2254         if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
2255                 return sysfs_emit(buf, "%s; SMT Host state unknown\n",
2256                                   mmio_strings[mmio_mitigation]);
2257         }
2258
2259         return sysfs_emit(buf, "%s; SMT %s\n", mmio_strings[mmio_mitigation],
2260                           sched_smt_active() ? "vulnerable" : "disabled");
2261 }
2262
2263 static char *stibp_state(void)
2264 {
2265         if (spectre_v2_in_ibrs_mode(spectre_v2_enabled))
2266                 return "";
2267
2268         switch (spectre_v2_user_stibp) {
2269         case SPECTRE_V2_USER_NONE:
2270                 return ", STIBP: disabled";
2271         case SPECTRE_V2_USER_STRICT:
2272                 return ", STIBP: forced";
2273         case SPECTRE_V2_USER_STRICT_PREFERRED:
2274                 return ", STIBP: always-on";
2275         case SPECTRE_V2_USER_PRCTL:
2276         case SPECTRE_V2_USER_SECCOMP:
2277                 if (static_key_enabled(&switch_to_cond_stibp))
2278                         return ", STIBP: conditional";
2279         }
2280         return "";
2281 }
2282
2283 static char *ibpb_state(void)
2284 {
2285         if (boot_cpu_has(X86_FEATURE_IBPB)) {
2286                 if (static_key_enabled(&switch_mm_always_ibpb))
2287                         return ", IBPB: always-on";
2288                 if (static_key_enabled(&switch_mm_cond_ibpb))
2289                         return ", IBPB: conditional";
2290                 return ", IBPB: disabled";
2291         }
2292         return "";
2293 }
2294
2295 static ssize_t spectre_v2_show_state(char *buf)
2296 {
2297         if (spectre_v2_enabled == SPECTRE_V2_LFENCE)
2298                 return sprintf(buf, "Vulnerable: LFENCE\n");
2299
2300         if (spectre_v2_enabled == SPECTRE_V2_EIBRS && unprivileged_ebpf_enabled())
2301                 return sprintf(buf, "Vulnerable: eIBRS with unprivileged eBPF\n");
2302
2303         if (sched_smt_active() && unprivileged_ebpf_enabled() &&
2304             spectre_v2_enabled == SPECTRE_V2_EIBRS_LFENCE)
2305                 return sprintf(buf, "Vulnerable: eIBRS+LFENCE with unprivileged eBPF and SMT\n");
2306
2307         return sprintf(buf, "%s%s%s%s%s%s\n",
2308                        spectre_v2_strings[spectre_v2_enabled],
2309                        ibpb_state(),
2310                        boot_cpu_has(X86_FEATURE_USE_IBRS_FW) ? ", IBRS_FW" : "",
2311                        stibp_state(),
2312                        boot_cpu_has(X86_FEATURE_RSB_CTXSW) ? ", RSB filling" : "",
2313                        spectre_v2_module_string());
2314 }
2315
2316 static ssize_t srbds_show_state(char *buf)
2317 {
2318         return sprintf(buf, "%s\n", srbds_strings[srbds_mitigation]);
2319 }
2320
2321 static ssize_t retbleed_show_state(char *buf)
2322 {
2323         if (retbleed_mitigation == RETBLEED_MITIGATION_UNRET) {
2324             if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD &&
2325                 boot_cpu_data.x86_vendor != X86_VENDOR_HYGON)
2326                     return sprintf(buf, "Vulnerable: untrained return thunk on non-Zen uarch\n");
2327
2328             return sprintf(buf, "%s; SMT %s\n",
2329                            retbleed_strings[retbleed_mitigation],
2330                            !sched_smt_active() ? "disabled" :
2331                            spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT ||
2332                            spectre_v2_user_stibp == SPECTRE_V2_USER_STRICT_PREFERRED ?
2333                            "enabled with STIBP protection" : "vulnerable");
2334         }
2335
2336         return sprintf(buf, "%s\n", retbleed_strings[retbleed_mitigation]);
2337 }
2338
2339 static ssize_t cpu_show_common(struct device *dev, struct device_attribute *attr,
2340                                char *buf, unsigned int bug)
2341 {
2342         if (!boot_cpu_has_bug(bug))
2343                 return sprintf(buf, "Not affected\n");
2344
2345         switch (bug) {
2346         case X86_BUG_CPU_MELTDOWN:
2347                 if (boot_cpu_has(X86_FEATURE_PTI))
2348                         return sprintf(buf, "Mitigation: PTI\n");
2349
2350                 if (hypervisor_is_type(X86_HYPER_XEN_PV))
2351                         return sprintf(buf, "Unknown (XEN PV detected, hypervisor mitigation required)\n");
2352
2353                 break;
2354
2355         case X86_BUG_SPECTRE_V1:
2356                 return sprintf(buf, "%s\n", spectre_v1_strings[spectre_v1_mitigation]);
2357
2358         case X86_BUG_SPECTRE_V2:
2359                 return spectre_v2_show_state(buf);
2360
2361         case X86_BUG_SPEC_STORE_BYPASS:
2362                 return sprintf(buf, "%s\n", ssb_strings[ssb_mode]);
2363
2364         case X86_BUG_L1TF:
2365                 if (boot_cpu_has(X86_FEATURE_L1TF_PTEINV))
2366                         return l1tf_show_state(buf);
2367                 break;
2368
2369         case X86_BUG_MDS:
2370                 return mds_show_state(buf);
2371
2372         case X86_BUG_TAA:
2373                 return tsx_async_abort_show_state(buf);
2374
2375         case X86_BUG_ITLB_MULTIHIT:
2376                 return itlb_multihit_show_state(buf);
2377
2378         case X86_BUG_SRBDS:
2379                 return srbds_show_state(buf);
2380
2381         case X86_BUG_MMIO_STALE_DATA:
2382                 return mmio_stale_data_show_state(buf);
2383
2384         case X86_BUG_RETBLEED:
2385                 return retbleed_show_state(buf);
2386
2387         default:
2388                 break;
2389         }
2390
2391         return sprintf(buf, "Vulnerable\n");
2392 }
2393
2394 ssize_t cpu_show_meltdown(struct device *dev, struct device_attribute *attr, char *buf)
2395 {
2396         return cpu_show_common(dev, attr, buf, X86_BUG_CPU_MELTDOWN);
2397 }
2398
2399 ssize_t cpu_show_spectre_v1(struct device *dev, struct device_attribute *attr, char *buf)
2400 {
2401         return cpu_show_common(dev, attr, buf, X86_BUG_SPECTRE_V1);
2402 }
2403
2404 ssize_t cpu_show_spectre_v2(struct device *dev, struct device_attribute *attr, char *buf)
2405 {
2406         return cpu_show_common(dev, attr, buf, X86_BUG_SPECTRE_V2);
2407 }
2408
2409 ssize_t cpu_show_spec_store_bypass(struct device *dev, struct device_attribute *attr, char *buf)
2410 {
2411         return cpu_show_common(dev, attr, buf, X86_BUG_SPEC_STORE_BYPASS);
2412 }
2413
2414 ssize_t cpu_show_l1tf(struct device *dev, struct device_attribute *attr, char *buf)
2415 {
2416         return cpu_show_common(dev, attr, buf, X86_BUG_L1TF);
2417 }
2418
2419 ssize_t cpu_show_mds(struct device *dev, struct device_attribute *attr, char *buf)
2420 {
2421         return cpu_show_common(dev, attr, buf, X86_BUG_MDS);
2422 }
2423
2424 ssize_t cpu_show_tsx_async_abort(struct device *dev, struct device_attribute *attr, char *buf)
2425 {
2426         return cpu_show_common(dev, attr, buf, X86_BUG_TAA);
2427 }
2428
2429 ssize_t cpu_show_itlb_multihit(struct device *dev, struct device_attribute *attr, char *buf)
2430 {
2431         return cpu_show_common(dev, attr, buf, X86_BUG_ITLB_MULTIHIT);
2432 }
2433
2434 ssize_t cpu_show_srbds(struct device *dev, struct device_attribute *attr, char *buf)
2435 {
2436         return cpu_show_common(dev, attr, buf, X86_BUG_SRBDS);
2437 }
2438
2439 ssize_t cpu_show_mmio_stale_data(struct device *dev, struct device_attribute *attr, char *buf)
2440 {
2441         return cpu_show_common(dev, attr, buf, X86_BUG_MMIO_STALE_DATA);
2442 }
2443
2444 ssize_t cpu_show_retbleed(struct device *dev, struct device_attribute *attr, char *buf)
2445 {
2446         return cpu_show_common(dev, attr, buf, X86_BUG_RETBLEED);
2447 }
2448 #endif