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 /* For scattered features from cpufeatures.h; we currently expose none */
  71 #define KF(x) bit(KVM_CPUID_BIT_##x)
  72
  73 int kvm_update_cpuid(struct kvm_vcpu *vcpu)
  74 {
  75         struct kvm_cpuid_entry2 *best;
  76         struct kvm_lapic *apic = vcpu->arch.apic;
  77
  78         best = kvm_find_cpuid_entry(vcpu, 1, 0);
  79         if (!best)
  80                 return 0;
  81
  82         /* Update OSXSAVE bit */
  83         if (boot_cpu_has(X86_FEATURE_XSAVE) && best->function == 0x1) {
  84                 best->ecx &= ~F(OSXSAVE);
  85                 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
  86                         best->ecx |= F(OSXSAVE);
  87         }
  88
  89         best->edx &= ~F(APIC);
  90         if (vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE)
  91                 best->edx |= F(APIC);
  92
  93         if (apic) {
  94                 if (best->ecx & F(TSC_DEADLINE_TIMER))
  95                         apic->lapic_timer.timer_mode_mask = 3 << 17;
  96                 else
  97                         apic->lapic_timer.timer_mode_mask = 1 << 17;
  98         }
  99
 100         best = kvm_find_cpuid_entry(vcpu, 7, 0);
 101         if (best) {
 102                 /* Update OSPKE bit */
 103                 if (boot_cpu_has(X86_FEATURE_PKU) && best->function == 0x7) {
 104                         best->ecx &= ~F(OSPKE);
 105                         if (kvm_read_cr4_bits(vcpu, X86_CR4_PKE))
 106                                 best->ecx |= F(OSPKE);
 107                 }
 108         }
 109
 110         best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
 111         if (!best) {
 112                 vcpu->arch.guest_supported_xcr0 = 0;
 113                 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
 114         } else {
 115                 vcpu->arch.guest_supported_xcr0 =
 116                         (best->eax | ((u64)best->edx << 32)) &
 117                         kvm_supported_xcr0();
 118                 vcpu->arch.guest_xstate_size = best->ebx =
 119                         xstate_required_size(vcpu->arch.xcr0, false);
 120         }
 121
 122         best = kvm_find_cpuid_entry(vcpu, 0xD, 1);
 123         if (best && (best->eax & (F(XSAVES) | F(XSAVEC))))
 124                 best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
 125
 126         /*
 127          * The existing code assumes virtual address is 48-bit or 57-bit in the
 128          * canonical address checks; exit if it is ever changed.
 129          */
 130         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
 131         if (best) {
 132                 int vaddr_bits = (best->eax & 0xff00) >> 8;
 133
 134                 if (vaddr_bits != 48 && vaddr_bits != 57 && vaddr_bits != 0)
 135                         return -EINVAL;
 136         }
 137
 138         best = kvm_find_cpuid_entry(vcpu, KVM_CPUID_FEATURES, 0);
 139         if (kvm_hlt_in_guest(vcpu->kvm) && best &&
 140                 (best->eax & (1 << KVM_FEATURE_PV_UNHALT)))
 141                 best->eax &= ~(1 << KVM_FEATURE_PV_UNHALT);
 142
 143         /* Update physical-address width */
 144         vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
 145         kvm_mmu_reset_context(vcpu);
 146
 147         kvm_pmu_refresh(vcpu);
 148         return 0;
 149 }
 150
 151 static int is_efer_nx(void)
 152 {
 153         unsigned long long efer = 0;
 154
 155         rdmsrl_safe(MSR_EFER, &efer);
 156         return efer & EFER_NX;
 157 }
 158
 159 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
 160 {
 161         int i;
 162         struct kvm_cpuid_entry2 *e, *entry;
 163
 164         entry = NULL;
 165         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 166                 e = &vcpu->arch.cpuid_entries[i];
 167                 if (e->function == 0x80000001) {
 168                         entry = e;
 169                         break;
 170                 }
 171         }
 172         if (entry && (entry->edx & F(NX)) && !is_efer_nx()) {
 173                 entry->edx &= ~F(NX);
 174                 printk(KERN_INFO "kvm: guest NX capability removed\n");
 175         }
 176 }
 177
 178 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
 179 {
 180         struct kvm_cpuid_entry2 *best;
 181
 182         best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
 183         if (!best || best->eax < 0x80000008)
 184                 goto not_found;
 185         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
 186         if (best)
 187                 return best->eax & 0xff;
 188 not_found:
 189         return 36;
 190 }
 191 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr);
 192
 193 /* when an old userspace process fills a new kernel module */
 194 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
 195                              struct kvm_cpuid *cpuid,
 196                              struct kvm_cpuid_entry __user *entries)
 197 {
 198         int r, i;
 199         struct kvm_cpuid_entry *cpuid_entries = NULL;
 200
 201         r = -E2BIG;
 202         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 203                 goto out;
 204         r = -ENOMEM;
 205         if (cpuid->nent) {
 206                 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) *
 207                                         cpuid->nent);
 208                 if (!cpuid_entries)
 209                         goto out;
 210                 r = -EFAULT;
 211                 if (copy_from_user(cpuid_entries, entries,
 212                                    cpuid->nent * sizeof(struct kvm_cpuid_entry)))
 213                         goto out;
 214         }
 215         for (i = 0; i < cpuid->nent; i++) {
 216                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
 217                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
 218                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
 219                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
 220                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
 221                 vcpu->arch.cpuid_entries[i].index = 0;
 222                 vcpu->arch.cpuid_entries[i].flags = 0;
 223                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
 224                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
 225                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
 226         }
 227         vcpu->arch.cpuid_nent = cpuid->nent;
 228         cpuid_fix_nx_cap(vcpu);
 229         kvm_apic_set_version(vcpu);
 230         kvm_x86_ops->cpuid_update(vcpu);
 231         r = kvm_update_cpuid(vcpu);
 232
 233 out:
 234         vfree(cpuid_entries);
 235         return r;
 236 }
 237
 238 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
 239                               struct kvm_cpuid2 *cpuid,
 240                               struct kvm_cpuid_entry2 __user *entries)
 241 {
 242         int r;
 243
 244         r = -E2BIG;
 245         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 246                 goto out;
 247         r = -EFAULT;
 248         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
 249                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
 250                 goto out;
 251         vcpu->arch.cpuid_nent = cpuid->nent;
 252         kvm_apic_set_version(vcpu);
 253         kvm_x86_ops->cpuid_update(vcpu);
 254         r = kvm_update_cpuid(vcpu);
 255 out:
 256         return r;
 257 }
 258
 259 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
 260                               struct kvm_cpuid2 *cpuid,
 261                               struct kvm_cpuid_entry2 __user *entries)
 262 {
 263         int r;
 264
 265         r = -E2BIG;
 266         if (cpuid->nent < vcpu->arch.cpuid_nent)
 267                 goto out;
 268         r = -EFAULT;
 269         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
 270                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
 271                 goto out;
 272         return 0;
 273
 274 out:
 275         cpuid->nent = vcpu->arch.cpuid_nent;
 276         return r;
 277 }
 278
 279 static void cpuid_mask(u32 *word, int wordnum)
 280 {
 281         *word &= boot_cpu_data.x86_capability[wordnum];
 282 }
 283
 284 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 285                            u32 index)
 286 {
 287         entry->function = function;
 288         entry->index = index;
 289         cpuid_count(entry->function, entry->index,
 290                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
 291         entry->flags = 0;
 292 }
 293
 294 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry,
 295                                    u32 func, u32 index, int *nent, int maxnent)
 296 {
 297         switch (func) {
 298         case 0:
 299                 entry->eax = 7;
 300                 ++*nent;
 301                 break;
 302         case 1:
 303                 entry->ecx = F(MOVBE);
 304                 ++*nent;
 305                 break;
 306         case 7:
 307                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 308                 if (index == 0)
 309                         entry->ecx = F(RDPID);
 310                 ++*nent;
 311         default:
 312                 break;
 313         }
 314
 315         entry->function = func;
 316         entry->index = index;
 317
 318         return 0;
 319 }
 320
 321 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 322                                  u32 index, int *nent, int maxnent)
 323 {
 324         int r;
 325         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
 326 #ifdef CONFIG_X86_64
 327         unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
 328                                 ? F(GBPAGES) : 0;
 329         unsigned f_lm = F(LM);
 330 #else
 331         unsigned f_gbpages = 0;
 332         unsigned f_lm = 0;
 333 #endif
 334         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
 335         unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
 336         unsigned f_mpx = kvm_mpx_supported() ? F(MPX) : 0;
 337         unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0;
 338         unsigned f_umip = kvm_x86_ops->umip_emulated() ? F(UMIP) : 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(IBPB) | F(IBRS);
 383
 384         /* cpuid 0xC0000001.edx */
 385         const u32 kvm_cpuid_C000_0001_edx_x86_features =
 386                 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
 387                 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
 388                 F(PMM) | F(PMM_EN);
 389
 390         /* cpuid 7.0.ebx */
 391         const u32 kvm_cpuid_7_0_ebx_x86_features =
 392                 F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
 393                 F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
 394                 F(ADX) | F(SMAP) | F(AVX512IFMA) | F(AVX512F) | F(AVX512PF) |
 395                 F(AVX512ER) | F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) |
 396                 F(SHA_NI) | F(AVX512BW) | F(AVX512VL);
 397
 398         /* cpuid 0xD.1.eax */
 399         const u32 kvm_cpuid_D_1_eax_x86_features =
 400                 F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves;
 401
 402         /* cpuid 7.0.ecx*/
 403         const u32 kvm_cpuid_7_0_ecx_x86_features =
 404                 F(AVX512VBMI) | F(LA57) | F(PKU) | 0 /*OSPKE*/ |
 405                 F(AVX512_VPOPCNTDQ) | F(UMIP) | F(AVX512_VBMI2) | F(GFNI) |
 406                 F(VAES) | F(VPCLMULQDQ) | F(AVX512_VNNI) | F(AVX512_BITALG);
 407
 408         /* cpuid 7.0.edx*/
 409         const u32 kvm_cpuid_7_0_edx_x86_features =
 410                 F(AVX512_4VNNIW) | F(AVX512_4FMAPS) | F(SPEC_CTRL) |
 411                 F(ARCH_CAPABILITIES);
 412
 413         /* all calls to cpuid_count() should be made on the same cpu */
 414         get_cpu();
 415
 416         r = -E2BIG;
 417
 418         if (*nent >= maxnent)
 419                 goto out;
 420
 421         do_cpuid_1_ent(entry, function, index);
 422         ++*nent;
 423
 424         switch (function) {
 425         case 0:
 426                 entry->eax = min(entry->eax, (u32)0xd);
 427                 break;
 428         case 1:
 429                 entry->edx &= kvm_cpuid_1_edx_x86_features;
 430                 cpuid_mask(&entry->edx, CPUID_1_EDX);
 431                 entry->ecx &= kvm_cpuid_1_ecx_x86_features;
 432                 cpuid_mask(&entry->ecx, CPUID_1_ECX);
 433                 /* we support x2apic emulation even if host does not support
 434                  * it since we emulate x2apic in software */
 435                 entry->ecx |= F(X2APIC);
 436                 break;
 437         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
 438          * may return different values. This forces us to get_cpu() before
 439          * issuing the first command, and also to emulate this annoying behavior
 440          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
 441         case 2: {
 442                 int t, times = entry->eax & 0xff;
 443
 444                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 445                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
 446                 for (t = 1; t < times; ++t) {
 447                         if (*nent >= maxnent)
 448                                 goto out;
 449
 450                         do_cpuid_1_ent(&entry[t], function, 0);
 451                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 452                         ++*nent;
 453                 }
 454                 break;
 455         }
 456         /* function 4 has additional index. */
 457         case 4: {
 458                 int i, cache_type;
 459
 460                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 461                 /* read more entries until cache_type is zero */
 462                 for (i = 1; ; ++i) {
 463                         if (*nent >= maxnent)
 464                                 goto out;
 465
 466                         cache_type = entry[i - 1].eax & 0x1f;
 467                         if (!cache_type)
 468                                 break;
 469                         do_cpuid_1_ent(&entry[i], function, i);
 470                         entry[i].flags |=
 471                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 472                         ++*nent;
 473                 }
 474                 break;
 475         }
 476         case 6: /* Thermal management */
 477                 entry->eax = 0x4; /* allow ARAT */
 478                 entry->ebx = 0;
 479                 entry->ecx = 0;
 480                 entry->edx = 0;
 481                 break;
 482         case 7: {
 483                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 484                 /* Mask ebx against host capability word 9 */
 485                 if (index == 0) {
 486                         entry->ebx &= kvm_cpuid_7_0_ebx_x86_features;
 487                         cpuid_mask(&entry->ebx, CPUID_7_0_EBX);
 488                         // TSC_ADJUST is emulated
 489                         entry->ebx |= F(TSC_ADJUST);
 490                         entry->ecx &= kvm_cpuid_7_0_ecx_x86_features;
 491                         cpuid_mask(&entry->ecx, CPUID_7_ECX);
 492                         entry->ecx |= f_umip;
 493                         /* PKU is not yet implemented for shadow paging. */
 494                         if (!tdp_enabled || !boot_cpu_has(X86_FEATURE_OSPKE))
 495                                 entry->ecx &= ~F(PKU);
 496                         entry->edx &= kvm_cpuid_7_0_edx_x86_features;
 497                         cpuid_mask(&entry->edx, CPUID_7_EDX);
 498                 } else {
 499                         entry->ebx = 0;
 500                         entry->ecx = 0;
 501                         entry->edx = 0;
 502                 }
 503                 entry->eax = 0;
 504                 break;
 505         }
 506         case 9:
 507                 break;
 508         case 0xa: { /* Architectural Performance Monitoring */
 509                 struct x86_pmu_capability cap;
 510                 union cpuid10_eax eax;
 511                 union cpuid10_edx edx;
 512
 513                 perf_get_x86_pmu_capability(&cap);
 514
 515                 /*
 516                  * Only support guest architectural pmu on a host
 517                  * with architectural pmu.
 518                  */
 519                 if (!cap.version)
 520                         memset(&cap, 0, sizeof(cap));
 521
 522                 eax.split.version_id = min(cap.version, 2);
 523                 eax.split.num_counters = cap.num_counters_gp;
 524                 eax.split.bit_width = cap.bit_width_gp;
 525                 eax.split.mask_length = cap.events_mask_len;
 526
 527                 edx.split.num_counters_fixed = cap.num_counters_fixed;
 528                 edx.split.bit_width_fixed = cap.bit_width_fixed;
 529                 edx.split.reserved = 0;
 530
 531                 entry->eax = eax.full;
 532                 entry->ebx = cap.events_mask;
 533                 entry->ecx = 0;
 534                 entry->edx = edx.full;
 535                 break;
 536         }
 537         /* function 0xb has additional index. */
 538         case 0xb: {
 539                 int i, level_type;
 540
 541                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 542                 /* read more entries until level_type is zero */
 543                 for (i = 1; ; ++i) {
 544                         if (*nent >= maxnent)
 545                                 goto out;
 546
 547                         level_type = entry[i - 1].ecx & 0xff00;
 548                         if (!level_type)
 549                                 break;
 550                         do_cpuid_1_ent(&entry[i], function, i);
 551                         entry[i].flags |=
 552                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 553                         ++*nent;
 554                 }
 555                 break;
 556         }
 557         case 0xd: {
 558                 int idx, i;
 559                 u64 supported = kvm_supported_xcr0();
 560
 561                 entry->eax &= supported;
 562                 entry->ebx = xstate_required_size(supported, false);
 563                 entry->ecx = entry->ebx;
 564                 entry->edx &= supported >> 32;
 565                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 566                 if (!supported)
 567                         break;
 568
 569                 for (idx = 1, i = 1; idx < 64; ++idx) {
 570                         u64 mask = ((u64)1 << idx);
 571                         if (*nent >= maxnent)
 572                                 goto out;
 573
 574                         do_cpuid_1_ent(&entry[i], function, idx);
 575                         if (idx == 1) {
 576                                 entry[i].eax &= kvm_cpuid_D_1_eax_x86_features;
 577                                 cpuid_mask(&entry[i].eax, CPUID_D_1_EAX);
 578                                 entry[i].ebx = 0;
 579                                 if (entry[i].eax & (F(XSAVES)|F(XSAVEC)))
 580                                         entry[i].ebx =
 581                                                 xstate_required_size(supported,
 582                                                                      true);
 583                         } else {
 584                                 if (entry[i].eax == 0 || !(supported & mask))
 585                                         continue;
 586                                 if (WARN_ON_ONCE(entry[i].ecx & 1))
 587                                         continue;
 588                         }
 589                         entry[i].ecx = 0;
 590                         entry[i].edx = 0;
 591                         entry[i].flags |=
 592                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 593                         ++*nent;
 594                         ++i;
 595                 }
 596                 break;
 597         }
 598         case KVM_CPUID_SIGNATURE: {
 599                 static const char signature[12] = "KVMKVMKVM\0\0";
 600                 const u32 *sigptr = (const u32 *)signature;
 601                 entry->eax = KVM_CPUID_FEATURES;
 602                 entry->ebx = sigptr[0];
 603                 entry->ecx = sigptr[1];
 604                 entry->edx = sigptr[2];
 605                 break;
 606         }
 607         case KVM_CPUID_FEATURES:
 608                 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
 609                              (1 << KVM_FEATURE_NOP_IO_DELAY) |
 610                              (1 << KVM_FEATURE_CLOCKSOURCE2) |
 611                              (1 << KVM_FEATURE_ASYNC_PF) |
 612                              (1 << KVM_FEATURE_PV_EOI) |
 613                              (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
 614                              (1 << KVM_FEATURE_PV_UNHALT) |
 615                              (1 << KVM_FEATURE_PV_TLB_FLUSH) |
 616                              (1 << KVM_FEATURE_ASYNC_PF_VMEXIT);
 617
 618                 if (sched_info_on())
 619                         entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
 620
 621                 entry->ebx = 0;
 622                 entry->ecx = 0;
 623                 entry->edx = 0;
 624                 break;
 625         case 0x80000000:
 626                 entry->eax = min(entry->eax, 0x8000001f);
 627                 break;
 628         case 0x80000001:
 629                 entry->edx &= kvm_cpuid_8000_0001_edx_x86_features;
 630                 cpuid_mask(&entry->edx, CPUID_8000_0001_EDX);
 631                 entry->ecx &= kvm_cpuid_8000_0001_ecx_x86_features;
 632                 cpuid_mask(&entry->ecx, CPUID_8000_0001_ECX);
 633                 break;
 634         case 0x80000007: /* Advanced power management */
 635                 /* invariant TSC is CPUID.80000007H:EDX[8] */
 636                 entry->edx &= (1 << 8);
 637                 /* mask against host */
 638                 entry->edx &= boot_cpu_data.x86_power;
 639                 entry->eax = entry->ebx = entry->ecx = 0;
 640                 break;
 641         case 0x80000008: {
 642                 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
 643                 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
 644                 unsigned phys_as = entry->eax & 0xff;
 645
 646                 if (!g_phys_as)
 647                         g_phys_as = phys_as;
 648                 entry->eax = g_phys_as | (virt_as << 8);
 649                 entry->edx = 0;
 650                 /* IBRS and IBPB aren't necessarily present in hardware cpuid */
 651                 if (boot_cpu_has(X86_FEATURE_IBPB))
 652                         entry->ebx |= F(IBPB);
 653                 if (boot_cpu_has(X86_FEATURE_IBRS))
 654                         entry->ebx |= F(IBRS);
 655                 entry->ebx &= kvm_cpuid_8000_0008_ebx_x86_features;
 656                 cpuid_mask(&entry->ebx, CPUID_8000_0008_EBX);
 657                 break;
 658         }
 659         case 0x80000019:
 660                 entry->ecx = entry->edx = 0;
 661                 break;
 662         case 0x8000001a:
 663                 break;
 664         case 0x8000001d:
 665                 break;
 666         /*Add support for Centaur's CPUID instruction*/
 667         case 0xC0000000:
 668                 /*Just support up to 0xC0000004 now*/
 669                 entry->eax = min(entry->eax, 0xC0000004);
 670                 break;
 671         case 0xC0000001:
 672                 entry->edx &= kvm_cpuid_C000_0001_edx_x86_features;
 673                 cpuid_mask(&entry->edx, CPUID_C000_0001_EDX);
 674                 break;
 675         case 3: /* Processor serial number */
 676         case 5: /* MONITOR/MWAIT */
 677         case 0xC0000002:
 678         case 0xC0000003:
 679         case 0xC0000004:
 680         default:
 681                 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
 682                 break;
 683         }
 684
 685         kvm_x86_ops->set_supported_cpuid(function, entry);
 686
 687         r = 0;
 688
 689 out:
 690         put_cpu();
 691
 692         return r;
 693 }
 694
 695 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
 696                         u32 idx, int *nent, int maxnent, unsigned int type)
 697 {
 698         if (type == KVM_GET_EMULATED_CPUID)
 699                 return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
 700
 701         return __do_cpuid_ent(entry, func, idx, nent, maxnent);
 702 }
 703
 704 #undef F
 705
 706 struct kvm_cpuid_param {
 707         u32 func;
 708         u32 idx;
 709         bool has_leaf_count;
 710         bool (*qualifier)(const struct kvm_cpuid_param *param);
 711 };
 712
 713 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
 714 {
 715         return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
 716 }
 717
 718 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
 719                                  __u32 num_entries, unsigned int ioctl_type)
 720 {
 721         int i;
 722         __u32 pad[3];
 723
 724         if (ioctl_type != KVM_GET_EMULATED_CPUID)
 725                 return false;
 726
 727         /*
 728          * We want to make sure that ->padding is being passed clean from
 729          * userspace in case we want to use it for something in the future.
 730          *
 731          * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
 732          * have to give ourselves satisfied only with the emulated side. /me
 733          * sheds a tear.
 734          */
 735         for (i = 0; i < num_entries; i++) {
 736                 if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
 737                         return true;
 738
 739                 if (pad[0] || pad[1] || pad[2])
 740                         return true;
 741         }
 742         return false;
 743 }
 744
 745 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
 746                             struct kvm_cpuid_entry2 __user *entries,
 747                             unsigned int type)
 748 {
 749         struct kvm_cpuid_entry2 *cpuid_entries;
 750         int limit, nent = 0, r = -E2BIG, i;
 751         u32 func;
 752         static const struct kvm_cpuid_param param[] = {
 753                 { .func = 0, .has_leaf_count = true },
 754                 { .func = 0x80000000, .has_leaf_count = true },
 755                 { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
 756                 { .func = KVM_CPUID_SIGNATURE },
 757                 { .func = KVM_CPUID_FEATURES },
 758         };
 759
 760         if (cpuid->nent < 1)
 761                 goto out;
 762         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 763                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
 764
 765         if (sanity_check_entries(entries, cpuid->nent, type))
 766                 return -EINVAL;
 767
 768         r = -ENOMEM;
 769         cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
 770         if (!cpuid_entries)
 771                 goto out;
 772
 773         r = 0;
 774         for (i = 0; i < ARRAY_SIZE(param); i++) {
 775                 const struct kvm_cpuid_param *ent = &param[i];
 776
 777                 if (ent->qualifier && !ent->qualifier(ent))
 778                         continue;
 779
 780                 r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
 781                                 &nent, cpuid->nent, type);
 782
 783                 if (r)
 784                         goto out_free;
 785
 786                 if (!ent->has_leaf_count)
 787                         continue;
 788
 789                 limit = cpuid_entries[nent - 1].eax;
 790                 for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
 791                         r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
 792                                      &nent, cpuid->nent, type);
 793
 794                 if (r)
 795                         goto out_free;
 796         }
 797
 798         r = -EFAULT;
 799         if (copy_to_user(entries, cpuid_entries,
 800                          nent * sizeof(struct kvm_cpuid_entry2)))
 801                 goto out_free;
 802         cpuid->nent = nent;
 803         r = 0;
 804
 805 out_free:
 806         vfree(cpuid_entries);
 807 out:
 808         return r;
 809 }
 810
 811 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
 812 {
 813         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
 814         struct kvm_cpuid_entry2 *ej;
 815         int j = i;
 816         int nent = vcpu->arch.cpuid_nent;
 817
 818         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
 819         /* when no next entry is found, the current entry[i] is reselected */
 820         do {
 821                 j = (j + 1) % nent;
 822                 ej = &vcpu->arch.cpuid_entries[j];
 823         } while (ej->function != e->function);
 824
 825         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
 826
 827         return j;
 828 }
 829
 830 /* find an entry with matching function, matching index (if needed), and that
 831  * should be read next (if it's stateful) */
 832 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
 833         u32 function, u32 index)
 834 {
 835         if (e->function != function)
 836                 return 0;
 837         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
 838                 return 0;
 839         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
 840             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
 841                 return 0;
 842         return 1;
 843 }
 844
 845 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
 846                                               u32 function, u32 index)
 847 {
 848         int i;
 849         struct kvm_cpuid_entry2 *best = NULL;
 850
 851         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 852                 struct kvm_cpuid_entry2 *e;
 853
 854                 e = &vcpu->arch.cpuid_entries[i];
 855                 if (is_matching_cpuid_entry(e, function, index)) {
 856                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
 857                                 move_to_next_stateful_cpuid_entry(vcpu, i);
 858                         best = e;
 859                         break;
 860                 }
 861         }
 862         return best;
 863 }
 864 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
 865
 866 /*
 867  * If no match is found, check whether we exceed the vCPU's limit
 868  * and return the content of the highest valid _standard_ leaf instead.
 869  * This is to satisfy the CPUID specification.
 870  */
 871 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
 872                                                   u32 function, u32 index)
 873 {
 874         struct kvm_cpuid_entry2 *maxlevel;
 875
 876         maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
 877         if (!maxlevel || maxlevel->eax >= function)
 878                 return NULL;
 879         if (function & 0x80000000) {
 880                 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
 881                 if (!maxlevel)
 882                         return NULL;
 883         }
 884         return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
 885 }
 886
 887 bool kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx,
 888                u32 *ecx, u32 *edx, bool check_limit)
 889 {
 890         u32 function = *eax, index = *ecx;
 891         struct kvm_cpuid_entry2 *best;
 892         bool entry_found = true;
 893
 894         best = kvm_find_cpuid_entry(vcpu, function, index);
 895
 896         if (!best) {
 897                 entry_found = false;
 898                 if (!check_limit)
 899                         goto out;
 900
 901                 best = check_cpuid_limit(vcpu, function, index);
 902         }
 903
 904 out:
 905         if (best) {
 906                 *eax = best->eax;
 907                 *ebx = best->ebx;
 908                 *ecx = best->ecx;
 909                 *edx = best->edx;
 910         } else
 911                 *eax = *ebx = *ecx = *edx = 0;
 912         trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx, entry_found);
 913         return entry_found;
 914 }
 915 EXPORT_SYMBOL_GPL(kvm_cpuid);
 916
 917 int kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
 918 {
 919         u32 eax, ebx, ecx, edx;
 920
 921         if (cpuid_fault_enabled(vcpu) && !kvm_require_cpl(vcpu, 0))
 922                 return 1;
 923
 924         eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
 925         ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
 926         kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx, true);
 927         kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
 928         kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
 929         kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
 930         kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
 931         return kvm_skip_emulated_instruction(vcpu);
 932 }
 933 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);