arch/x86/kvm/cpuid.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  * cpuid support routines
   4  *
   5  * derived from arch/x86/kvm/x86.c
   6  *
   7  * Copyright 2011 Red Hat, Inc. and/or its affiliates.
   8  * Copyright IBM Corporation, 2008
   9  *
  10  * This work is licensed under the terms of the GNU GPL, version 2.  See
  11  * the COPYING file in the top-level directory.
  12  *
  13  */
  14
  15 #include <linux/kvm_host.h>
  16 #include <linux/export.h>
  17 #include <linux/vmalloc.h>
  18 #include <linux/uaccess.h>
  19 #include <linux/sched/stat.h>
  20
  21 #include <asm/processor.h>
  22 #include <asm/user.h>
  23 #include <asm/fpu/xstate.h>
  24 #include "cpuid.h"
  25 #include "lapic.h"
  26 #include "mmu.h"
  27 #include "trace.h"
  28 #include "pmu.h"
  29
  30 static u32 xstate_required_size(u64 xstate_bv, bool compacted)
  31 {
  32         int feature_bit = 0;
  33         u32 ret = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
  34
  35         xstate_bv &= XFEATURE_MASK_EXTEND;
  36         while (xstate_bv) {
  37                 if (xstate_bv & 0x1) {
  38                         u32 eax, ebx, ecx, edx, offset;
  39                         cpuid_count(0xD, feature_bit, &eax, &ebx, &ecx, &edx);
  40                         offset = compacted ? ret : ebx;
  41                         ret = max(ret, offset + eax);
  42                 }
  43
  44                 xstate_bv >>= 1;
  45                 feature_bit++;
  46         }
  47
  48         return ret;
  49 }
  50
  51 bool kvm_mpx_supported(void)
  52 {
  53         return ((host_xcr0 & (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR))
  54                  && kvm_x86_ops->mpx_supported());
  55 }
  56 EXPORT_SYMBOL_GPL(kvm_mpx_supported);
  57
  58 u64 kvm_supported_xcr0(void)
  59 {
  60         u64 xcr0 = KVM_SUPPORTED_XCR0 & host_xcr0;
  61
  62         if (!kvm_mpx_supported())
  63                 xcr0 &= ~(XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR);
  64
  65         return xcr0;
  66 }
  67
  68 #define F(x) bit(X86_FEATURE_##x)
  69
  70 int kvm_update_cpuid(struct kvm_vcpu *vcpu)
  71 {
  72         struct kvm_cpuid_entry2 *best;
  73         struct kvm_lapic *apic = vcpu->arch.apic;
  74
  75         best = kvm_find_cpuid_entry(vcpu, 1, 0);
  76         if (!best)
  77                 return 0;
  78
  79         /* Update OSXSAVE bit */
  80         if (boot_cpu_has(X86_FEATURE_XSAVE) && best->function == 0x1) {
  81                 best->ecx &= ~F(OSXSAVE);
  82                 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
  83                         best->ecx |= F(OSXSAVE);
  84         }
  85
  86         best->edx &= ~F(APIC);
  87         if (vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE)
  88                 best->edx |= F(APIC);
  89
  90         if (apic) {
  91                 if (best->ecx & F(TSC_DEADLINE_TIMER))
  92                         apic->lapic_timer.timer_mode_mask = 3 << 17;
  93                 else
  94                         apic->lapic_timer.timer_mode_mask = 1 << 17;
  95         }
  96
  97         best = kvm_find_cpuid_entry(vcpu, 7, 0);
  98         if (best) {
  99                 /* Update OSPKE bit */
 100                 if (boot_cpu_has(X86_FEATURE_PKU) && best->function == 0x7) {
 101                         best->ecx &= ~F(OSPKE);
 102                         if (kvm_read_cr4_bits(vcpu, X86_CR4_PKE))
 103                                 best->ecx |= F(OSPKE);
 104                 }
 105         }
 106
 107         best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
 108         if (!best) {
 109                 vcpu->arch.guest_supported_xcr0 = 0;
 110                 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
 111         } else {
 112                 vcpu->arch.guest_supported_xcr0 =
 113                         (best->eax | ((u64)best->edx << 32)) &
 114                         kvm_supported_xcr0();
 115                 vcpu->arch.guest_xstate_size = best->ebx =
 116                         xstate_required_size(vcpu->arch.xcr0, false);
 117         }
 118
 119         best = kvm_find_cpuid_entry(vcpu, 0xD, 1);
 120         if (best && (best->eax & (F(XSAVES) | F(XSAVEC))))
 121                 best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
 122
 123         /*
 124          * The existing code assumes virtual address is 48-bit or 57-bit in the
 125          * canonical address checks; exit if it is ever changed.
 126          */
 127         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
 128         if (best) {
 129                 int vaddr_bits = (best->eax & 0xff00) >> 8;
 130
 131                 if (vaddr_bits != 48 && vaddr_bits != 57 && vaddr_bits != 0)
 132                         return -EINVAL;
 133         }
 134
 135         best = kvm_find_cpuid_entry(vcpu, KVM_CPUID_FEATURES, 0);
 136         if (kvm_hlt_in_guest(vcpu->kvm) && best &&
 137                 (best->eax & (1 << KVM_FEATURE_PV_UNHALT)))
 138                 best->eax &= ~(1 << KVM_FEATURE_PV_UNHALT);
 139
 140         /* Update physical-address width */
 141         vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
 142         kvm_mmu_reset_context(vcpu);
 143
 144         kvm_pmu_refresh(vcpu);
 145         return 0;
 146 }
 147
 148 static int is_efer_nx(void)
 149 {
 150         unsigned long long efer = 0;
 151
 152         rdmsrl_safe(MSR_EFER, &efer);
 153         return efer & EFER_NX;
 154 }
 155
 156 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
 157 {
 158         int i;
 159         struct kvm_cpuid_entry2 *e, *entry;
 160
 161         entry = NULL;
 162         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 163                 e = &vcpu->arch.cpuid_entries[i];
 164                 if (e->function == 0x80000001) {
 165                         entry = e;
 166                         break;
 167                 }
 168         }
 169         if (entry && (entry->edx & F(NX)) && !is_efer_nx()) {
 170                 entry->edx &= ~F(NX);
 171                 printk(KERN_INFO "kvm: guest NX capability removed\n");
 172         }
 173 }
 174
 175 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
 176 {
 177         struct kvm_cpuid_entry2 *best;
 178
 179         best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
 180         if (!best || best->eax < 0x80000008)
 181                 goto not_found;
 182         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
 183         if (best)
 184                 return best->eax & 0xff;
 185 not_found:
 186         return 36;
 187 }
 188 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr);
 189
 190 /* when an old userspace process fills a new kernel module */
 191 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
 192                              struct kvm_cpuid *cpuid,
 193                              struct kvm_cpuid_entry __user *entries)
 194 {
 195         int r, i;
 196         struct kvm_cpuid_entry *cpuid_entries = NULL;
 197
 198         r = -E2BIG;
 199         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 200                 goto out;
 201         r = -ENOMEM;
 202         if (cpuid->nent) {
 203                 cpuid_entries =
 204                         vmalloc(array_size(sizeof(struct kvm_cpuid_entry),
 205                                            cpuid->nent));
 206                 if (!cpuid_entries)
 207                         goto out;
 208                 r = -EFAULT;
 209                 if (copy_from_user(cpuid_entries, entries,
 210                                    cpuid->nent * sizeof(struct kvm_cpuid_entry)))
 211                         goto out;
 212         }
 213         for (i = 0; i < cpuid->nent; i++) {
 214                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
 215                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
 216                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
 217                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
 218                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
 219                 vcpu->arch.cpuid_entries[i].index = 0;
 220                 vcpu->arch.cpuid_entries[i].flags = 0;
 221                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
 222                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
 223                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
 224         }
 225         vcpu->arch.cpuid_nent = cpuid->nent;
 226         cpuid_fix_nx_cap(vcpu);
 227         kvm_apic_set_version(vcpu);
 228         kvm_x86_ops->cpuid_update(vcpu);
 229         r = kvm_update_cpuid(vcpu);
 230
 231 out:
 232         vfree(cpuid_entries);
 233         return r;
 234 }
 235
 236 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
 237                               struct kvm_cpuid2 *cpuid,
 238                               struct kvm_cpuid_entry2 __user *entries)
 239 {
 240         int r;
 241
 242         r = -E2BIG;
 243         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 244                 goto out;
 245         r = -EFAULT;
 246         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
 247                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
 248                 goto out;
 249         vcpu->arch.cpuid_nent = cpuid->nent;
 250         kvm_apic_set_version(vcpu);
 251         kvm_x86_ops->cpuid_update(vcpu);
 252         r = kvm_update_cpuid(vcpu);
 253 out:
 254         return r;
 255 }
 256
 257 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
 258                               struct kvm_cpuid2 *cpuid,
 259                               struct kvm_cpuid_entry2 __user *entries)
 260 {
 261         int r;
 262
 263         r = -E2BIG;
 264         if (cpuid->nent < vcpu->arch.cpuid_nent)
 265                 goto out;
 266         r = -EFAULT;
 267         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
 268                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
 269                 goto out;
 270         return 0;
 271
 272 out:
 273         cpuid->nent = vcpu->arch.cpuid_nent;
 274         return r;
 275 }
 276
 277 static void cpuid_mask(u32 *word, int wordnum)
 278 {
 279         *word &= boot_cpu_data.x86_capability[wordnum];
 280 }
 281
 282 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 283                            u32 index)
 284 {
 285         entry->function = function;
 286         entry->index = index;
 287         cpuid_count(entry->function, entry->index,
 288                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
 289         entry->flags = 0;
 290 }
 291
 292 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry,
 293                                    u32 func, u32 index, int *nent, int maxnent)
 294 {
 295         switch (func) {
 296         case 0:
 297                 entry->eax = 7;
 298                 ++*nent;
 299                 break;
 300         case 1:
 301                 entry->ecx = F(MOVBE);
 302                 ++*nent;
 303                 break;
 304         case 7:
 305                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 306                 if (index == 0)
 307                         entry->ecx = F(RDPID);
 308                 ++*nent;
 309         default:
 310                 break;
 311         }
 312
 313         entry->function = func;
 314         entry->index = index;
 315
 316         return 0;
 317 }
 318
 319 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 320                                  u32 index, int *nent, int maxnent)
 321 {
 322         int r;
 323         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
 324 #ifdef CONFIG_X86_64
 325         unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
 326                                 ? F(GBPAGES) : 0;
 327         unsigned f_lm = F(LM);
 328 #else
 329         unsigned f_gbpages = 0;
 330         unsigned f_lm = 0;
 331 #endif
 332         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
 333         unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
 334         unsigned f_mpx = kvm_mpx_supported() ? F(MPX) : 0;
 335         unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0;
 336         unsigned f_umip = kvm_x86_ops->umip_emulated() ? F(UMIP) : 0;
 337         unsigned f_intel_pt = kvm_x86_ops->pt_supported() ? F(INTEL_PT) : 0;
 338         unsigned f_la57 = 0;
 339
 340         /* cpuid 1.edx */
 341         const u32 kvm_cpuid_1_edx_x86_features =
 342                 F(FPU) | F(VME) | F(DE) | F(PSE) |
 343                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
 344                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
 345                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
 346                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
 347                 0 /* Reserved, DS, ACPI */ | F(MMX) |
 348                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
 349                 0 /* HTT, TM, Reserved, PBE */;
 350         /* cpuid 0x80000001.edx */
 351         const u32 kvm_cpuid_8000_0001_edx_x86_features =
 352                 F(FPU) | F(VME) | F(DE) | F(PSE) |
 353                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
 354                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
 355                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
 356                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
 357                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
 358                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
 359                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
 360         /* cpuid 1.ecx */
 361         const u32 kvm_cpuid_1_ecx_x86_features =
 362                 /* NOTE: MONITOR (and MWAIT) are emulated as NOP,
 363                  * but *not* advertised to guests via CPUID ! */
 364                 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
 365                 0 /* DS-CPL, VMX, SMX, EST */ |
 366                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
 367                 F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
 368                 F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
 369                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
 370                 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
 371                 F(F16C) | F(RDRAND);
 372         /* cpuid 0x80000001.ecx */
 373         const u32 kvm_cpuid_8000_0001_ecx_x86_features =
 374                 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
 375                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
 376                 F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
 377                 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM) |
 378                 F(TOPOEXT) | F(PERFCTR_CORE);
 379
 380         /* cpuid 0x80000008.ebx */
 381         const u32 kvm_cpuid_8000_0008_ebx_x86_features =
 382                 F(WBNOINVD) | F(AMD_IBPB) | F(AMD_IBRS) | F(AMD_SSBD) | F(VIRT_SSBD) |
 383                 F(AMD_SSB_NO) | F(AMD_STIBP);
 384
 385         /* cpuid 0xC0000001.edx */
 386         const u32 kvm_cpuid_C000_0001_edx_x86_features =
 387                 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
 388                 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
 389                 F(PMM) | F(PMM_EN);
 390
 391         /* cpuid 7.0.ebx */
 392         const u32 kvm_cpuid_7_0_ebx_x86_features =
 393                 F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
 394                 F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
 395                 F(ADX) | F(SMAP) | F(AVX512IFMA) | F(AVX512F) | F(AVX512PF) |
 396                 F(AVX512ER) | F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) |
 397                 F(SHA_NI) | F(AVX512BW) | F(AVX512VL) | f_intel_pt;
 398
 399         /* cpuid 0xD.1.eax */
 400         const u32 kvm_cpuid_D_1_eax_x86_features =
 401                 F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves;
 402
 403         /* cpuid 7.0.ecx*/
 404         const u32 kvm_cpuid_7_0_ecx_x86_features =
 405                 F(AVX512VBMI) | F(LA57) | F(PKU) | 0 /*OSPKE*/ |
 406                 F(AVX512_VPOPCNTDQ) | F(UMIP) | F(AVX512_VBMI2) | F(GFNI) |
 407                 F(VAES) | F(VPCLMULQDQ) | F(AVX512_VNNI) | F(AVX512_BITALG) |
 408                 F(CLDEMOTE) | F(MOVDIRI) | F(MOVDIR64B);
 409
 410         /* cpuid 7.0.edx*/
 411         const u32 kvm_cpuid_7_0_edx_x86_features =
 412                 F(AVX512_4VNNIW) | F(AVX512_4FMAPS) | F(SPEC_CTRL) |
 413                 F(SPEC_CTRL_SSBD) | F(ARCH_CAPABILITIES) | F(INTEL_STIBP) |
 414                 F(MD_CLEAR);
 415
 416         /* all calls to cpuid_count() should be made on the same cpu */
 417         get_cpu();
 418
 419         r = -E2BIG;
 420
 421         if (*nent >= maxnent)
 422                 goto out;
 423
 424         do_cpuid_1_ent(entry, function, index);
 425         ++*nent;
 426
 427         switch (function) {
 428         case 0:
 429                 entry->eax = min(entry->eax, (u32)(f_intel_pt ? 0x14 : 0xd));
 430                 break;
 431         case 1:
 432                 entry->edx &= kvm_cpuid_1_edx_x86_features;
 433                 cpuid_mask(&entry->edx, CPUID_1_EDX);
 434                 entry->ecx &= kvm_cpuid_1_ecx_x86_features;
 435                 cpuid_mask(&entry->ecx, CPUID_1_ECX);
 436                 /* we support x2apic emulation even if host does not support
 437                  * it since we emulate x2apic in software */
 438                 entry->ecx |= F(X2APIC);
 439                 break;
 440         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
 441          * may return different values. This forces us to get_cpu() before
 442          * issuing the first command, and also to emulate this annoying behavior
 443          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
 444         case 2: {
 445                 int t, times = entry->eax & 0xff;
 446
 447                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 448                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
 449                 for (t = 1; t < times; ++t) {
 450                         if (*nent >= maxnent)
 451                                 goto out;
 452
 453                         do_cpuid_1_ent(&entry[t], function, 0);
 454                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 455                         ++*nent;
 456                 }
 457                 break;
 458         }
 459         /* function 4 has additional index. */
 460         case 4: {
 461                 int i, cache_type;
 462
 463                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 464                 /* read more entries until cache_type is zero */
 465                 for (i = 1; ; ++i) {
 466                         if (*nent >= maxnent)
 467                                 goto out;
 468
 469                         cache_type = entry[i - 1].eax & 0x1f;
 470                         if (!cache_type)
 471                                 break;
 472                         do_cpuid_1_ent(&entry[i], function, i);
 473                         entry[i].flags |=
 474                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 475                         ++*nent;
 476                 }
 477                 break;
 478         }
 479         case 6: /* Thermal management */
 480                 entry->eax = 0x4; /* allow ARAT */
 481                 entry->ebx = 0;
 482                 entry->ecx = 0;
 483                 entry->edx = 0;
 484                 break;
 485         case 7: {
 486                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 487                 /* Mask ebx against host capability word 9 */
 488                 if (index == 0) {
 489                         entry->ebx &= kvm_cpuid_7_0_ebx_x86_features;
 490                         cpuid_mask(&entry->ebx, CPUID_7_0_EBX);
 491                         // TSC_ADJUST is emulated
 492                         entry->ebx |= F(TSC_ADJUST);
 493                         entry->ecx &= kvm_cpuid_7_0_ecx_x86_features;
 494                         f_la57 = entry->ecx & F(LA57);
 495                         cpuid_mask(&entry->ecx, CPUID_7_ECX);
 496                         /* Set LA57 based on hardware capability. */
 497                         entry->ecx |= f_la57;
 498                         entry->ecx |= f_umip;
 499                         /* PKU is not yet implemented for shadow paging. */
 500                         if (!tdp_enabled || !boot_cpu_has(X86_FEATURE_OSPKE))
 501                                 entry->ecx &= ~F(PKU);
 502                         entry->edx &= kvm_cpuid_7_0_edx_x86_features;
 503                         cpuid_mask(&entry->edx, CPUID_7_EDX);
 504                         /*
 505                          * We emulate ARCH_CAPABILITIES in software even
 506                          * if the host doesn't support it.
 507                          */
 508                         entry->edx |= F(ARCH_CAPABILITIES);
 509                 } else {
 510                         entry->ebx = 0;
 511                         entry->ecx = 0;
 512                         entry->edx = 0;
 513                 }
 514                 entry->eax = 0;
 515                 break;
 516         }
 517         case 9:
 518                 break;
 519         case 0xa: { /* Architectural Performance Monitoring */
 520                 struct x86_pmu_capability cap;
 521                 union cpuid10_eax eax;
 522                 union cpuid10_edx edx;
 523
 524                 perf_get_x86_pmu_capability(&cap);
 525
 526                 /*
 527                  * Only support guest architectural pmu on a host
 528                  * with architectural pmu.
 529                  */
 530                 if (!cap.version)
 531                         memset(&cap, 0, sizeof(cap));
 532
 533                 eax.split.version_id = min(cap.version, 2);
 534                 eax.split.num_counters = cap.num_counters_gp;
 535                 eax.split.bit_width = cap.bit_width_gp;
 536                 eax.split.mask_length = cap.events_mask_len;
 537
 538                 edx.split.num_counters_fixed = cap.num_counters_fixed;
 539                 edx.split.bit_width_fixed = cap.bit_width_fixed;
 540                 edx.split.reserved = 0;
 541
 542                 entry->eax = eax.full;
 543                 entry->ebx = cap.events_mask;
 544                 entry->ecx = 0;
 545                 entry->edx = edx.full;
 546                 break;
 547         }
 548         /* function 0xb has additional index. */
 549         case 0xb: {
 550                 int i, level_type;
 551
 552                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 553                 /* read more entries until level_type is zero */
 554                 for (i = 1; ; ++i) {
 555                         if (*nent >= maxnent)
 556                                 goto out;
 557
 558                         level_type = entry[i - 1].ecx & 0xff00;
 559                         if (!level_type)
 560                                 break;
 561                         do_cpuid_1_ent(&entry[i], function, i);
 562                         entry[i].flags |=
 563                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 564                         ++*nent;
 565                 }
 566                 break;
 567         }
 568         case 0xd: {
 569                 int idx, i;
 570                 u64 supported = kvm_supported_xcr0();
 571
 572                 entry->eax &= supported;
 573                 entry->ebx = xstate_required_size(supported, false);
 574                 entry->ecx = entry->ebx;
 575                 entry->edx &= supported >> 32;
 576                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 577                 if (!supported)
 578                         break;
 579
 580                 for (idx = 1, i = 1; idx < 64; ++idx) {
 581                         u64 mask = ((u64)1 << idx);
 582                         if (*nent >= maxnent)
 583                                 goto out;
 584
 585                         do_cpuid_1_ent(&entry[i], function, idx);
 586                         if (idx == 1) {
 587                                 entry[i].eax &= kvm_cpuid_D_1_eax_x86_features;
 588                                 cpuid_mask(&entry[i].eax, CPUID_D_1_EAX);
 589                                 entry[i].ebx = 0;
 590                                 if (entry[i].eax & (F(XSAVES)|F(XSAVEC)))
 591                                         entry[i].ebx =
 592                                                 xstate_required_size(supported,
 593                                                                      true);
 594                         } else {
 595                                 if (entry[i].eax == 0 || !(supported & mask))
 596                                         continue;
 597                                 if (WARN_ON_ONCE(entry[i].ecx & 1))
 598                                         continue;
 599                         }
 600                         entry[i].ecx = 0;
 601                         entry[i].edx = 0;
 602                         entry[i].flags |=
 603                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 604                         ++*nent;
 605                         ++i;
 606                 }
 607                 break;
 608         }
 609         /* Intel PT */
 610         case 0x14: {
 611                 int t, times = entry->eax;
 612
 613                 if (!f_intel_pt)
 614                         break;
 615
 616                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 617                 for (t = 1; t <= times; ++t) {
 618                         if (*nent >= maxnent)
 619                                 goto out;
 620                         do_cpuid_1_ent(&entry[t], function, t);
 621                         entry[t].flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 622                         ++*nent;
 623                 }
 624                 break;
 625         }
 626         case KVM_CPUID_SIGNATURE: {
 627                 static const char signature[12] = "KVMKVMKVM\0\0";
 628                 const u32 *sigptr = (const u32 *)signature;
 629                 entry->eax = KVM_CPUID_FEATURES;
 630                 entry->ebx = sigptr[0];
 631                 entry->ecx = sigptr[1];
 632                 entry->edx = sigptr[2];
 633                 break;
 634         }
 635         case KVM_CPUID_FEATURES:
 636                 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
 637                              (1 << KVM_FEATURE_NOP_IO_DELAY) |
 638                              (1 << KVM_FEATURE_CLOCKSOURCE2) |
 639                              (1 << KVM_FEATURE_ASYNC_PF) |
 640                              (1 << KVM_FEATURE_PV_EOI) |
 641                              (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
 642                              (1 << KVM_FEATURE_PV_UNHALT) |
 643                              (1 << KVM_FEATURE_PV_TLB_FLUSH) |
 644                              (1 << KVM_FEATURE_ASYNC_PF_VMEXIT) |
 645                              (1 << KVM_FEATURE_PV_SEND_IPI);
 646
 647                 if (sched_info_on())
 648                         entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
 649
 650                 entry->ebx = 0;
 651                 entry->ecx = 0;
 652                 entry->edx = 0;
 653                 break;
 654         case 0x80000000:
 655                 entry->eax = min(entry->eax, 0x8000001f);
 656                 break;
 657         case 0x80000001:
 658                 entry->edx &= kvm_cpuid_8000_0001_edx_x86_features;
 659                 cpuid_mask(&entry->edx, CPUID_8000_0001_EDX);
 660                 entry->ecx &= kvm_cpuid_8000_0001_ecx_x86_features;
 661                 cpuid_mask(&entry->ecx, CPUID_8000_0001_ECX);
 662                 break;
 663         case 0x80000007: /* Advanced power management */
 664                 /* invariant TSC is CPUID.80000007H:EDX[8] */
 665                 entry->edx &= (1 << 8);
 666                 /* mask against host */
 667                 entry->edx &= boot_cpu_data.x86_power;
 668                 entry->eax = entry->ebx = entry->ecx = 0;
 669                 break;
 670         case 0x80000008: {
 671                 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
 672                 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
 673                 unsigned phys_as = entry->eax & 0xff;
 674
 675                 if (!g_phys_as)
 676                         g_phys_as = phys_as;
 677                 entry->eax = g_phys_as | (virt_as << 8);
 678                 entry->edx = 0;
 679                 /*
 680                  * IBRS, IBPB and VIRT_SSBD aren't necessarily present in
 681                  * hardware cpuid
 682                  */
 683                 if (boot_cpu_has(X86_FEATURE_AMD_IBPB))
 684                         entry->ebx |= F(AMD_IBPB);
 685                 if (boot_cpu_has(X86_FEATURE_AMD_IBRS))
 686                         entry->ebx |= F(AMD_IBRS);
 687                 if (boot_cpu_has(X86_FEATURE_VIRT_SSBD))
 688                         entry->ebx |= F(VIRT_SSBD);
 689                 entry->ebx &= kvm_cpuid_8000_0008_ebx_x86_features;
 690                 cpuid_mask(&entry->ebx, CPUID_8000_0008_EBX);
 691                 /*
 692                  * The preference is to use SPEC CTRL MSR instead of the
 693                  * VIRT_SPEC MSR.
 694                  */
 695                 if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD) &&
 696                     !boot_cpu_has(X86_FEATURE_AMD_SSBD))
 697                         entry->ebx |= F(VIRT_SSBD);
 698                 break;
 699         }
 700         case 0x80000019:
 701                 entry->ecx = entry->edx = 0;
 702                 break;
 703         case 0x8000001a:
 704                 break;
 705         case 0x8000001d:
 706                 break;
 707         /*Add support for Centaur's CPUID instruction*/
 708         case 0xC0000000:
 709                 /*Just support up to 0xC0000004 now*/
 710                 entry->eax = min(entry->eax, 0xC0000004);
 711                 break;
 712         case 0xC0000001:
 713                 entry->edx &= kvm_cpuid_C000_0001_edx_x86_features;
 714                 cpuid_mask(&entry->edx, CPUID_C000_0001_EDX);
 715                 break;
 716         case 3: /* Processor serial number */
 717         case 5: /* MONITOR/MWAIT */
 718         case 0xC0000002:
 719         case 0xC0000003:
 720         case 0xC0000004:
 721         default:
 722                 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
 723                 break;
 724         }
 725
 726         kvm_x86_ops->set_supported_cpuid(function, entry);
 727
 728         r = 0;
 729
 730 out:
 731         put_cpu();
 732
 733         return r;
 734 }
 735
 736 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
 737                         u32 idx, int *nent, int maxnent, unsigned int type)
 738 {
 739         if (type == KVM_GET_EMULATED_CPUID)
 740                 return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
 741
 742         return __do_cpuid_ent(entry, func, idx, nent, maxnent);
 743 }
 744
 745 #undef F
 746
 747 struct kvm_cpuid_param {
 748         u32 func;
 749         u32 idx;
 750         bool has_leaf_count;
 751         bool (*qualifier)(const struct kvm_cpuid_param *param);
 752 };
 753
 754 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
 755 {
 756         return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
 757 }
 758
 759 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
 760                                  __u32 num_entries, unsigned int ioctl_type)
 761 {
 762         int i;
 763         __u32 pad[3];
 764
 765         if (ioctl_type != KVM_GET_EMULATED_CPUID)
 766                 return false;
 767
 768         /*
 769          * We want to make sure that ->padding is being passed clean from
 770          * userspace in case we want to use it for something in the future.
 771          *
 772          * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
 773          * have to give ourselves satisfied only with the emulated side. /me
 774          * sheds a tear.
 775          */
 776         for (i = 0; i < num_entries; i++) {
 777                 if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
 778                         return true;
 779
 780                 if (pad[0] || pad[1] || pad[2])
 781                         return true;
 782         }
 783         return false;
 784 }
 785
 786 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
 787                             struct kvm_cpuid_entry2 __user *entries,
 788                             unsigned int type)
 789 {
 790         struct kvm_cpuid_entry2 *cpuid_entries;
 791         int limit, nent = 0, r = -E2BIG, i;
 792         u32 func;
 793         static const struct kvm_cpuid_param param[] = {
 794                 { .func = 0, .has_leaf_count = true },
 795                 { .func = 0x80000000, .has_leaf_count = true },
 796                 { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
 797                 { .func = KVM_CPUID_SIGNATURE },
 798                 { .func = KVM_CPUID_FEATURES },
 799         };
 800
 801         if (cpuid->nent < 1)
 802                 goto out;
 803         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 804                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
 805
 806         if (sanity_check_entries(entries, cpuid->nent, type))
 807                 return -EINVAL;
 808
 809         r = -ENOMEM;
 810         cpuid_entries = vzalloc(array_size(sizeof(struct kvm_cpuid_entry2),
 811                                            cpuid->nent));
 812         if (!cpuid_entries)
 813                 goto out;
 814
 815         r = 0;
 816         for (i = 0; i < ARRAY_SIZE(param); i++) {
 817                 const struct kvm_cpuid_param *ent = &param[i];
 818
 819                 if (ent->qualifier && !ent->qualifier(ent))
 820                         continue;
 821
 822                 r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
 823                                 &nent, cpuid->nent, type);
 824
 825                 if (r)
 826                         goto out_free;
 827
 828                 if (!ent->has_leaf_count)
 829                         continue;
 830
 831                 limit = cpuid_entries[nent - 1].eax;
 832                 for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
 833                         r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
 834                                      &nent, cpuid->nent, type);
 835
 836                 if (r)
 837                         goto out_free;
 838         }
 839
 840         r = -EFAULT;
 841         if (copy_to_user(entries, cpuid_entries,
 842                          nent * sizeof(struct kvm_cpuid_entry2)))
 843                 goto out_free;
 844         cpuid->nent = nent;
 845         r = 0;
 846
 847 out_free:
 848         vfree(cpuid_entries);
 849 out:
 850         return r;
 851 }
 852
 853 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
 854 {
 855         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
 856         struct kvm_cpuid_entry2 *ej;
 857         int j = i;
 858         int nent = vcpu->arch.cpuid_nent;
 859
 860         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
 861         /* when no next entry is found, the current entry[i] is reselected */
 862         do {
 863                 j = (j + 1) % nent;
 864                 ej = &vcpu->arch.cpuid_entries[j];
 865         } while (ej->function != e->function);
 866
 867         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
 868
 869         return j;
 870 }
 871
 872 /* find an entry with matching function, matching index (if needed), and that
 873  * should be read next (if it's stateful) */
 874 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
 875         u32 function, u32 index)
 876 {
 877         if (e->function != function)
 878                 return 0;
 879         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
 880                 return 0;
 881         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
 882             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
 883                 return 0;
 884         return 1;
 885 }
 886
 887 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
 888                                               u32 function, u32 index)
 889 {
 890         int i;
 891         struct kvm_cpuid_entry2 *best = NULL;
 892
 893         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 894                 struct kvm_cpuid_entry2 *e;
 895
 896                 e = &vcpu->arch.cpuid_entries[i];
 897                 if (is_matching_cpuid_entry(e, function, index)) {
 898                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
 899                                 move_to_next_stateful_cpuid_entry(vcpu, i);
 900                         best = e;
 901                         break;
 902                 }
 903         }
 904         return best;
 905 }
 906 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
 907
 908 /*
 909  * If no match is found, check whether we exceed the vCPU's limit
 910  * and return the content of the highest valid _standard_ leaf instead.
 911  * This is to satisfy the CPUID specification.
 912  */
 913 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
 914                                                   u32 function, u32 index)
 915 {
 916         struct kvm_cpuid_entry2 *maxlevel;
 917
 918         maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
 919         if (!maxlevel || maxlevel->eax >= function)
 920                 return NULL;
 921         if (function & 0x80000000) {
 922                 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
 923                 if (!maxlevel)
 924                         return NULL;
 925         }
 926         return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
 927 }
 928
 929 bool kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx,
 930                u32 *ecx, u32 *edx, bool check_limit)
 931 {
 932         u32 function = *eax, index = *ecx;
 933         struct kvm_cpuid_entry2 *best;
 934         bool entry_found = true;
 935
 936         best = kvm_find_cpuid_entry(vcpu, function, index);
 937
 938         if (!best) {
 939                 entry_found = false;
 940                 if (!check_limit)
 941                         goto out;
 942
 943                 best = check_cpuid_limit(vcpu, function, index);
 944         }
 945
 946 out:
 947         if (best) {
 948                 *eax = best->eax;
 949                 *ebx = best->ebx;
 950                 *ecx = best->ecx;
 951                 *edx = best->edx;
 952         } else
 953                 *eax = *ebx = *ecx = *edx = 0;
 954         trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx, entry_found);
 955         return entry_found;
 956 }
 957 EXPORT_SYMBOL_GPL(kvm_cpuid);
 958
 959 int kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
 960 {
 961         u32 eax, ebx, ecx, edx;
 962
 963         if (cpuid_fault_enabled(vcpu) && !kvm_require_cpl(vcpu, 0))
 964                 return 1;
 965
 966         eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
 967         ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
 968         kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx, true);
 969         kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
 970         kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
 971         kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
 972         kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
 973         return kvm_skip_emulated_instruction(vcpu);
 974 }
 975 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);