arch/x86/kvm/cpuid.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Kernel-based Virtual Machine driver for Linux
   4  * cpuid support routines
   5  *
   6  * derived from arch/x86/kvm/x86.c
   7  *
   8  * Copyright 2011 Red Hat, Inc. and/or its affiliates.
   9  * Copyright IBM Corporation, 2008
  10  */
  11
  12 #include <linux/kvm_host.h>
  13 #include <linux/export.h>
  14 #include <linux/vmalloc.h>
  15 #include <linux/uaccess.h>
  16 #include <linux/sched/stat.h>
  17
  18 #include <asm/processor.h>
  19 #include <asm/user.h>
  20 #include <asm/fpu/xstate.h>
  21 #include "cpuid.h"
  22 #include "lapic.h"
  23 #include "mmu.h"
  24 #include "trace.h"
  25 #include "pmu.h"
  26
  27 static u32 xstate_required_size(u64 xstate_bv, bool compacted)
  28 {
  29         int feature_bit = 0;
  30         u32 ret = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
  31
  32         xstate_bv &= XFEATURE_MASK_EXTEND;
  33         while (xstate_bv) {
  34                 if (xstate_bv & 0x1) {
  35                         u32 eax, ebx, ecx, edx, offset;
  36                         cpuid_count(0xD, feature_bit, &eax, &ebx, &ecx, &edx);
  37                         offset = compacted ? ret : ebx;
  38                         ret = max(ret, offset + eax);
  39                 }
  40
  41                 xstate_bv >>= 1;
  42                 feature_bit++;
  43         }
  44
  45         return ret;
  46 }
  47
  48 bool kvm_mpx_supported(void)
  49 {
  50         return ((host_xcr0 & (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR))
  51                  && kvm_x86_ops->mpx_supported());
  52 }
  53 EXPORT_SYMBOL_GPL(kvm_mpx_supported);
  54
  55 u64 kvm_supported_xcr0(void)
  56 {
  57         u64 xcr0 = KVM_SUPPORTED_XCR0 & host_xcr0;
  58
  59         if (!kvm_mpx_supported())
  60                 xcr0 &= ~(XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR);
  61
  62         return xcr0;
  63 }
  64
  65 #define F(x) bit(X86_FEATURE_##x)
  66
  67 int kvm_update_cpuid(struct kvm_vcpu *vcpu)
  68 {
  69         struct kvm_cpuid_entry2 *best;
  70         struct kvm_lapic *apic = vcpu->arch.apic;
  71
  72         best = kvm_find_cpuid_entry(vcpu, 1, 0);
  73         if (!best)
  74                 return 0;
  75
  76         /* Update OSXSAVE bit */
  77         if (boot_cpu_has(X86_FEATURE_XSAVE) && best->function == 0x1) {
  78                 best->ecx &= ~F(OSXSAVE);
  79                 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
  80                         best->ecx |= F(OSXSAVE);
  81         }
  82
  83         best->edx &= ~F(APIC);
  84         if (vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE)
  85                 best->edx |= F(APIC);
  86
  87         if (apic) {
  88                 if (best->ecx & F(TSC_DEADLINE_TIMER))
  89                         apic->lapic_timer.timer_mode_mask = 3 << 17;
  90                 else
  91                         apic->lapic_timer.timer_mode_mask = 1 << 17;
  92         }
  93
  94         best = kvm_find_cpuid_entry(vcpu, 7, 0);
  95         if (best) {
  96                 /* Update OSPKE bit */
  97                 if (boot_cpu_has(X86_FEATURE_PKU) && best->function == 0x7) {
  98                         best->ecx &= ~F(OSPKE);
  99                         if (kvm_read_cr4_bits(vcpu, X86_CR4_PKE))
 100                                 best->ecx |= F(OSPKE);
 101                 }
 102         }
 103
 104         best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
 105         if (!best) {
 106                 vcpu->arch.guest_supported_xcr0 = 0;
 107                 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
 108         } else {
 109                 vcpu->arch.guest_supported_xcr0 =
 110                         (best->eax | ((u64)best->edx << 32)) &
 111                         kvm_supported_xcr0();
 112                 vcpu->arch.guest_xstate_size = best->ebx =
 113                         xstate_required_size(vcpu->arch.xcr0, false);
 114         }
 115
 116         best = kvm_find_cpuid_entry(vcpu, 0xD, 1);
 117         if (best && (best->eax & (F(XSAVES) | F(XSAVEC))))
 118                 best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
 119
 120         /*
 121          * The existing code assumes virtual address is 48-bit or 57-bit in the
 122          * canonical address checks; exit if it is ever changed.
 123          */
 124         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
 125         if (best) {
 126                 int vaddr_bits = (best->eax & 0xff00) >> 8;
 127
 128                 if (vaddr_bits != 48 && vaddr_bits != 57 && vaddr_bits != 0)
 129                         return -EINVAL;
 130         }
 131
 132         best = kvm_find_cpuid_entry(vcpu, KVM_CPUID_FEATURES, 0);
 133         if (kvm_hlt_in_guest(vcpu->kvm) && best &&
 134                 (best->eax & (1 << KVM_FEATURE_PV_UNHALT)))
 135                 best->eax &= ~(1 << KVM_FEATURE_PV_UNHALT);
 136
 137         /* Update physical-address width */
 138         vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
 139         kvm_mmu_reset_context(vcpu);
 140
 141         kvm_pmu_refresh(vcpu);
 142         return 0;
 143 }
 144
 145 static int is_efer_nx(void)
 146 {
 147         unsigned long long efer = 0;
 148
 149         rdmsrl_safe(MSR_EFER, &efer);
 150         return efer & EFER_NX;
 151 }
 152
 153 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
 154 {
 155         int i;
 156         struct kvm_cpuid_entry2 *e, *entry;
 157
 158         entry = NULL;
 159         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 160                 e = &vcpu->arch.cpuid_entries[i];
 161                 if (e->function == 0x80000001) {
 162                         entry = e;
 163                         break;
 164                 }
 165         }
 166         if (entry && (entry->edx & F(NX)) && !is_efer_nx()) {
 167                 entry->edx &= ~F(NX);
 168                 printk(KERN_INFO "kvm: guest NX capability removed\n");
 169         }
 170 }
 171
 172 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
 173 {
 174         struct kvm_cpuid_entry2 *best;
 175
 176         best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
 177         if (!best || best->eax < 0x80000008)
 178                 goto not_found;
 179         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
 180         if (best)
 181                 return best->eax & 0xff;
 182 not_found:
 183         return 36;
 184 }
 185 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr);
 186
 187 /* when an old userspace process fills a new kernel module */
 188 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
 189                              struct kvm_cpuid *cpuid,
 190                              struct kvm_cpuid_entry __user *entries)
 191 {
 192         int r, i;
 193         struct kvm_cpuid_entry *cpuid_entries = NULL;
 194
 195         r = -E2BIG;
 196         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 197                 goto out;
 198         r = -ENOMEM;
 199         if (cpuid->nent) {
 200                 cpuid_entries =
 201                         vmalloc(array_size(sizeof(struct kvm_cpuid_entry),
 202                                            cpuid->nent));
 203                 if (!cpuid_entries)
 204                         goto out;
 205                 r = -EFAULT;
 206                 if (copy_from_user(cpuid_entries, entries,
 207                                    cpuid->nent * sizeof(struct kvm_cpuid_entry)))
 208                         goto out;
 209         }
 210         for (i = 0; i < cpuid->nent; i++) {
 211                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
 212                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
 213                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
 214                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
 215                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
 216                 vcpu->arch.cpuid_entries[i].index = 0;
 217                 vcpu->arch.cpuid_entries[i].flags = 0;
 218                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
 219                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
 220                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
 221         }
 222         vcpu->arch.cpuid_nent = cpuid->nent;
 223         cpuid_fix_nx_cap(vcpu);
 224         kvm_apic_set_version(vcpu);
 225         kvm_x86_ops->cpuid_update(vcpu);
 226         r = kvm_update_cpuid(vcpu);
 227
 228 out:
 229         vfree(cpuid_entries);
 230         return r;
 231 }
 232
 233 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
 234                               struct kvm_cpuid2 *cpuid,
 235                               struct kvm_cpuid_entry2 __user *entries)
 236 {
 237         int r;
 238
 239         r = -E2BIG;
 240         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 241                 goto out;
 242         r = -EFAULT;
 243         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
 244                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
 245                 goto out;
 246         vcpu->arch.cpuid_nent = cpuid->nent;
 247         kvm_apic_set_version(vcpu);
 248         kvm_x86_ops->cpuid_update(vcpu);
 249         r = kvm_update_cpuid(vcpu);
 250 out:
 251         return r;
 252 }
 253
 254 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
 255                               struct kvm_cpuid2 *cpuid,
 256                               struct kvm_cpuid_entry2 __user *entries)
 257 {
 258         int r;
 259
 260         r = -E2BIG;
 261         if (cpuid->nent < vcpu->arch.cpuid_nent)
 262                 goto out;
 263         r = -EFAULT;
 264         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
 265                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
 266                 goto out;
 267         return 0;
 268
 269 out:
 270         cpuid->nent = vcpu->arch.cpuid_nent;
 271         return r;
 272 }
 273
 274 static void cpuid_mask(u32 *word, int wordnum)
 275 {
 276         *word &= boot_cpu_data.x86_capability[wordnum];
 277 }
 278
 279 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 280                            u32 index)
 281 {
 282         entry->function = function;
 283         entry->index = index;
 284         cpuid_count(entry->function, entry->index,
 285                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
 286         entry->flags = 0;
 287 }
 288
 289 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry,
 290                                    u32 func, u32 index, int *nent, int maxnent)
 291 {
 292         switch (func) {
 293         case 0:
 294                 entry->eax = 7;
 295                 ++*nent;
 296                 break;
 297         case 1:
 298                 entry->ecx = F(MOVBE);
 299                 ++*nent;
 300                 break;
 301         case 7:
 302                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 303                 if (index == 0)
 304                         entry->ecx = F(RDPID);
 305                 ++*nent;
 306         default:
 307                 break;
 308         }
 309
 310         entry->function = func;
 311         entry->index = index;
 312
 313         return 0;
 314 }
 315
 316 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 317                                  u32 index, int *nent, int maxnent)
 318 {
 319         int r;
 320         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
 321 #ifdef CONFIG_X86_64
 322         unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
 323                                 ? F(GBPAGES) : 0;
 324         unsigned f_lm = F(LM);
 325 #else
 326         unsigned f_gbpages = 0;
 327         unsigned f_lm = 0;
 328 #endif
 329         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
 330         unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
 331         unsigned f_mpx = kvm_mpx_supported() ? F(MPX) : 0;
 332         unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0;
 333         unsigned f_umip = kvm_x86_ops->umip_emulated() ? F(UMIP) : 0;
 334         unsigned f_intel_pt = kvm_x86_ops->pt_supported() ? F(INTEL_PT) : 0;
 335         unsigned f_la57 = 0;
 336
 337         /* cpuid 1.edx */
 338         const u32 kvm_cpuid_1_edx_x86_features =
 339                 F(FPU) | F(VME) | F(DE) | F(PSE) |
 340                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
 341                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
 342                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
 343                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
 344                 0 /* Reserved, DS, ACPI */ | F(MMX) |
 345                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
 346                 0 /* HTT, TM, Reserved, PBE */;
 347         /* cpuid 0x80000001.edx */
 348         const u32 kvm_cpuid_8000_0001_edx_x86_features =
 349                 F(FPU) | F(VME) | F(DE) | F(PSE) |
 350                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
 351                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
 352                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
 353                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
 354                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
 355                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
 356                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
 357         /* cpuid 1.ecx */
 358         const u32 kvm_cpuid_1_ecx_x86_features =
 359                 /* NOTE: MONITOR (and MWAIT) are emulated as NOP,
 360                  * but *not* advertised to guests via CPUID ! */
 361                 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
 362                 0 /* DS-CPL, VMX, SMX, EST */ |
 363                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
 364                 F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
 365                 F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
 366                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
 367                 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
 368                 F(F16C) | F(RDRAND);
 369         /* cpuid 0x80000001.ecx */
 370         const u32 kvm_cpuid_8000_0001_ecx_x86_features =
 371                 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
 372                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
 373                 F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
 374                 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM) |
 375                 F(TOPOEXT) | F(PERFCTR_CORE);
 376
 377         /* cpuid 0x80000008.ebx */
 378         const u32 kvm_cpuid_8000_0008_ebx_x86_features =
 379                 F(WBNOINVD) | F(AMD_IBPB) | F(AMD_IBRS) | F(AMD_SSBD) | F(VIRT_SSBD) |
 380                 F(AMD_SSB_NO) | F(AMD_STIBP);
 381
 382         /* cpuid 0xC0000001.edx */
 383         const u32 kvm_cpuid_C000_0001_edx_x86_features =
 384                 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
 385                 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
 386                 F(PMM) | F(PMM_EN);
 387
 388         /* cpuid 7.0.ebx */
 389         const u32 kvm_cpuid_7_0_ebx_x86_features =
 390                 F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
 391                 F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
 392                 F(ADX) | F(SMAP) | F(AVX512IFMA) | F(AVX512F) | F(AVX512PF) |
 393                 F(AVX512ER) | F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) |
 394                 F(SHA_NI) | F(AVX512BW) | F(AVX512VL) | f_intel_pt;
 395
 396         /* cpuid 0xD.1.eax */
 397         const u32 kvm_cpuid_D_1_eax_x86_features =
 398                 F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves;
 399
 400         /* cpuid 7.0.ecx*/
 401         const u32 kvm_cpuid_7_0_ecx_x86_features =
 402                 F(AVX512VBMI) | F(LA57) | F(PKU) | 0 /*OSPKE*/ |
 403                 F(AVX512_VPOPCNTDQ) | F(UMIP) | F(AVX512_VBMI2) | F(GFNI) |
 404                 F(VAES) | F(VPCLMULQDQ) | F(AVX512_VNNI) | F(AVX512_BITALG) |
 405                 F(CLDEMOTE) | F(MOVDIRI) | F(MOVDIR64B);
 406
 407         /* cpuid 7.0.edx*/
 408         const u32 kvm_cpuid_7_0_edx_x86_features =
 409                 F(AVX512_4VNNIW) | F(AVX512_4FMAPS) | F(SPEC_CTRL) |
 410                 F(SPEC_CTRL_SSBD) | F(ARCH_CAPABILITIES) | F(INTEL_STIBP) |
 411                 F(MD_CLEAR);
 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)(f_intel_pt ? 0x14 : 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         /* functions 4 and 0x8000001d have additional index. */
 457         case 4:
 458         case 0x8000001d: {
 459                 int i, cache_type;
 460
 461                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 462                 /* read more entries until cache_type is zero */
 463                 for (i = 1; ; ++i) {
 464                         if (*nent >= maxnent)
 465                                 goto out;
 466
 467                         cache_type = entry[i - 1].eax & 0x1f;
 468                         if (!cache_type)
 469                                 break;
 470                         do_cpuid_1_ent(&entry[i], function, i);
 471                         entry[i].flags |=
 472                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 473                         ++*nent;
 474                 }
 475                 break;
 476         }
 477         case 6: /* Thermal management */
 478                 entry->eax = 0x4; /* allow ARAT */
 479                 entry->ebx = 0;
 480                 entry->ecx = 0;
 481                 entry->edx = 0;
 482                 break;
 483         case 7: {
 484                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 485                 /* Mask ebx against host capability word 9 */
 486                 if (index == 0) {
 487                         entry->ebx &= kvm_cpuid_7_0_ebx_x86_features;
 488                         cpuid_mask(&entry->ebx, CPUID_7_0_EBX);
 489                         // TSC_ADJUST is emulated
 490                         entry->ebx |= F(TSC_ADJUST);
 491                         entry->ecx &= kvm_cpuid_7_0_ecx_x86_features;
 492                         f_la57 = entry->ecx & F(LA57);
 493                         cpuid_mask(&entry->ecx, CPUID_7_ECX);
 494                         /* Set LA57 based on hardware capability. */
 495                         entry->ecx |= f_la57;
 496                         entry->ecx |= f_umip;
 497                         /* PKU is not yet implemented for shadow paging. */
 498                         if (!tdp_enabled || !boot_cpu_has(X86_FEATURE_OSPKE))
 499                                 entry->ecx &= ~F(PKU);
 500                         entry->edx &= kvm_cpuid_7_0_edx_x86_features;
 501                         cpuid_mask(&entry->edx, CPUID_7_EDX);
 502                         /*
 503                          * We emulate ARCH_CAPABILITIES in software even
 504                          * if the host doesn't support it.
 505                          */
 506                         entry->edx |= F(ARCH_CAPABILITIES);
 507                 } else {
 508                         entry->ebx = 0;
 509                         entry->ecx = 0;
 510                         entry->edx = 0;
 511                 }
 512                 entry->eax = 0;
 513                 break;
 514         }
 515         case 9:
 516                 break;
 517         case 0xa: { /* Architectural Performance Monitoring */
 518                 struct x86_pmu_capability cap;
 519                 union cpuid10_eax eax;
 520                 union cpuid10_edx edx;
 521
 522                 perf_get_x86_pmu_capability(&cap);
 523
 524                 /*
 525                  * Only support guest architectural pmu on a host
 526                  * with architectural pmu.
 527                  */
 528                 if (!cap.version)
 529                         memset(&cap, 0, sizeof(cap));
 530
 531                 eax.split.version_id = min(cap.version, 2);
 532                 eax.split.num_counters = cap.num_counters_gp;
 533                 eax.split.bit_width = cap.bit_width_gp;
 534                 eax.split.mask_length = cap.events_mask_len;
 535
 536                 edx.split.num_counters_fixed = cap.num_counters_fixed;
 537                 edx.split.bit_width_fixed = cap.bit_width_fixed;
 538                 edx.split.reserved = 0;
 539
 540                 entry->eax = eax.full;
 541                 entry->ebx = cap.events_mask;
 542                 entry->ecx = 0;
 543                 entry->edx = edx.full;
 544                 break;
 545         }
 546         /* function 0xb has additional index. */
 547         case 0xb: {
 548                 int i, level_type;
 549
 550                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 551                 /* read more entries until level_type is zero */
 552                 for (i = 1; ; ++i) {
 553                         if (*nent >= maxnent)
 554                                 goto out;
 555
 556                         level_type = entry[i - 1].ecx & 0xff00;
 557                         if (!level_type)
 558                                 break;
 559                         do_cpuid_1_ent(&entry[i], function, i);
 560                         entry[i].flags |=
 561                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 562                         ++*nent;
 563                 }
 564                 break;
 565         }
 566         case 0xd: {
 567                 int idx, i;
 568                 u64 supported = kvm_supported_xcr0();
 569
 570                 entry->eax &= supported;
 571                 entry->ebx = xstate_required_size(supported, false);
 572                 entry->ecx = entry->ebx;
 573                 entry->edx &= supported >> 32;
 574                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 575                 if (!supported)
 576                         break;
 577
 578                 for (idx = 1, i = 1; idx < 64; ++idx) {
 579                         u64 mask = ((u64)1 << idx);
 580                         if (*nent >= maxnent)
 581                                 goto out;
 582
 583                         do_cpuid_1_ent(&entry[i], function, idx);
 584                         if (idx == 1) {
 585                                 entry[i].eax &= kvm_cpuid_D_1_eax_x86_features;
 586                                 cpuid_mask(&entry[i].eax, CPUID_D_1_EAX);
 587                                 entry[i].ebx = 0;
 588                                 if (entry[i].eax & (F(XSAVES)|F(XSAVEC)))
 589                                         entry[i].ebx =
 590                                                 xstate_required_size(supported,
 591                                                                      true);
 592                         } else {
 593                                 if (entry[i].eax == 0 || !(supported & mask))
 594                                         continue;
 595                                 if (WARN_ON_ONCE(entry[i].ecx & 1))
 596                                         continue;
 597                         }
 598                         entry[i].ecx = 0;
 599                         entry[i].edx = 0;
 600                         entry[i].flags |=
 601                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 602                         ++*nent;
 603                         ++i;
 604                 }
 605                 break;
 606         }
 607         /* Intel PT */
 608         case 0x14: {
 609                 int t, times = entry->eax;
 610
 611                 if (!f_intel_pt)
 612                         break;
 613
 614                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 615                 for (t = 1; t <= times; ++t) {
 616                         if (*nent >= maxnent)
 617                                 goto out;
 618                         do_cpuid_1_ent(&entry[t], function, t);
 619                         entry[t].flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 620                         ++*nent;
 621                 }
 622                 break;
 623         }
 624         case KVM_CPUID_SIGNATURE: {
 625                 static const char signature[12] = "KVMKVMKVM\0\0";
 626                 const u32 *sigptr = (const u32 *)signature;
 627                 entry->eax = KVM_CPUID_FEATURES;
 628                 entry->ebx = sigptr[0];
 629                 entry->ecx = sigptr[1];
 630                 entry->edx = sigptr[2];
 631                 break;
 632         }
 633         case KVM_CPUID_FEATURES:
 634                 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
 635                              (1 << KVM_FEATURE_NOP_IO_DELAY) |
 636                              (1 << KVM_FEATURE_CLOCKSOURCE2) |
 637                              (1 << KVM_FEATURE_ASYNC_PF) |
 638                              (1 << KVM_FEATURE_PV_EOI) |
 639                              (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
 640                              (1 << KVM_FEATURE_PV_UNHALT) |
 641                              (1 << KVM_FEATURE_PV_TLB_FLUSH) |
 642                              (1 << KVM_FEATURE_ASYNC_PF_VMEXIT) |
 643                              (1 << KVM_FEATURE_PV_SEND_IPI);
 644
 645                 if (sched_info_on())
 646                         entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
 647
 648                 entry->ebx = 0;
 649                 entry->ecx = 0;
 650                 entry->edx = 0;
 651                 break;
 652         case 0x80000000:
 653                 entry->eax = min(entry->eax, 0x8000001f);
 654                 break;
 655         case 0x80000001:
 656                 entry->edx &= kvm_cpuid_8000_0001_edx_x86_features;
 657                 cpuid_mask(&entry->edx, CPUID_8000_0001_EDX);
 658                 entry->ecx &= kvm_cpuid_8000_0001_ecx_x86_features;
 659                 cpuid_mask(&entry->ecx, CPUID_8000_0001_ECX);
 660                 break;
 661         case 0x80000007: /* Advanced power management */
 662                 /* invariant TSC is CPUID.80000007H:EDX[8] */
 663                 entry->edx &= (1 << 8);
 664                 /* mask against host */
 665                 entry->edx &= boot_cpu_data.x86_power;
 666                 entry->eax = entry->ebx = entry->ecx = 0;
 667                 break;
 668         case 0x80000008: {
 669                 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
 670                 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
 671                 unsigned phys_as = entry->eax & 0xff;
 672
 673                 if (!g_phys_as)
 674                         g_phys_as = phys_as;
 675                 entry->eax = g_phys_as | (virt_as << 8);
 676                 entry->edx = 0;
 677                 /*
 678                  * IBRS, IBPB and VIRT_SSBD aren't necessarily present in
 679                  * hardware cpuid
 680                  */
 681                 if (boot_cpu_has(X86_FEATURE_AMD_IBPB))
 682                         entry->ebx |= F(AMD_IBPB);
 683                 if (boot_cpu_has(X86_FEATURE_AMD_IBRS))
 684                         entry->ebx |= F(AMD_IBRS);
 685                 if (boot_cpu_has(X86_FEATURE_VIRT_SSBD))
 686                         entry->ebx |= F(VIRT_SSBD);
 687                 entry->ebx &= kvm_cpuid_8000_0008_ebx_x86_features;
 688                 cpuid_mask(&entry->ebx, CPUID_8000_0008_EBX);
 689                 /*
 690                  * The preference is to use SPEC CTRL MSR instead of the
 691                  * VIRT_SPEC MSR.
 692                  */
 693                 if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD) &&
 694                     !boot_cpu_has(X86_FEATURE_AMD_SSBD))
 695                         entry->ebx |= F(VIRT_SSBD);
 696                 break;
 697         }
 698         case 0x80000019:
 699                 entry->ecx = entry->edx = 0;
 700                 break;
 701         case 0x8000001a:
 702         case 0x8000001e:
 703                 break;
 704         /*Add support for Centaur's CPUID instruction*/
 705         case 0xC0000000:
 706                 /*Just support up to 0xC0000004 now*/
 707                 entry->eax = min(entry->eax, 0xC0000004);
 708                 break;
 709         case 0xC0000001:
 710                 entry->edx &= kvm_cpuid_C000_0001_edx_x86_features;
 711                 cpuid_mask(&entry->edx, CPUID_C000_0001_EDX);
 712                 break;
 713         case 3: /* Processor serial number */
 714         case 5: /* MONITOR/MWAIT */
 715         case 0xC0000002:
 716         case 0xC0000003:
 717         case 0xC0000004:
 718         default:
 719                 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
 720                 break;
 721         }
 722
 723         kvm_x86_ops->set_supported_cpuid(function, entry);
 724
 725         r = 0;
 726
 727 out:
 728         put_cpu();
 729
 730         return r;
 731 }
 732
 733 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
 734                         u32 idx, int *nent, int maxnent, unsigned int type)
 735 {
 736         if (type == KVM_GET_EMULATED_CPUID)
 737                 return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
 738
 739         return __do_cpuid_ent(entry, func, idx, nent, maxnent);
 740 }
 741
 742 #undef F
 743
 744 struct kvm_cpuid_param {
 745         u32 func;
 746         u32 idx;
 747         bool has_leaf_count;
 748         bool (*qualifier)(const struct kvm_cpuid_param *param);
 749 };
 750
 751 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
 752 {
 753         return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
 754 }
 755
 756 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
 757                                  __u32 num_entries, unsigned int ioctl_type)
 758 {
 759         int i;
 760         __u32 pad[3];
 761
 762         if (ioctl_type != KVM_GET_EMULATED_CPUID)
 763                 return false;
 764
 765         /*
 766          * We want to make sure that ->padding is being passed clean from
 767          * userspace in case we want to use it for something in the future.
 768          *
 769          * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
 770          * have to give ourselves satisfied only with the emulated side. /me
 771          * sheds a tear.
 772          */
 773         for (i = 0; i < num_entries; i++) {
 774                 if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
 775                         return true;
 776
 777                 if (pad[0] || pad[1] || pad[2])
 778                         return true;
 779         }
 780         return false;
 781 }
 782
 783 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
 784                             struct kvm_cpuid_entry2 __user *entries,
 785                             unsigned int type)
 786 {
 787         struct kvm_cpuid_entry2 *cpuid_entries;
 788         int limit, nent = 0, r = -E2BIG, i;
 789         u32 func;
 790         static const struct kvm_cpuid_param param[] = {
 791                 { .func = 0, .has_leaf_count = true },
 792                 { .func = 0x80000000, .has_leaf_count = true },
 793                 { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
 794                 { .func = KVM_CPUID_SIGNATURE },
 795                 { .func = KVM_CPUID_FEATURES },
 796         };
 797
 798         if (cpuid->nent < 1)
 799                 goto out;
 800         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 801                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
 802
 803         if (sanity_check_entries(entries, cpuid->nent, type))
 804                 return -EINVAL;
 805
 806         r = -ENOMEM;
 807         cpuid_entries = vzalloc(array_size(sizeof(struct kvm_cpuid_entry2),
 808                                            cpuid->nent));
 809         if (!cpuid_entries)
 810                 goto out;
 811
 812         r = 0;
 813         for (i = 0; i < ARRAY_SIZE(param); i++) {
 814                 const struct kvm_cpuid_param *ent = &param[i];
 815
 816                 if (ent->qualifier && !ent->qualifier(ent))
 817                         continue;
 818
 819                 r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
 820                                 &nent, cpuid->nent, type);
 821
 822                 if (r)
 823                         goto out_free;
 824
 825                 if (!ent->has_leaf_count)
 826                         continue;
 827
 828                 limit = cpuid_entries[nent - 1].eax;
 829                 for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
 830                         r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
 831                                      &nent, cpuid->nent, type);
 832
 833                 if (r)
 834                         goto out_free;
 835         }
 836
 837         r = -EFAULT;
 838         if (copy_to_user(entries, cpuid_entries,
 839                          nent * sizeof(struct kvm_cpuid_entry2)))
 840                 goto out_free;
 841         cpuid->nent = nent;
 842         r = 0;
 843
 844 out_free:
 845         vfree(cpuid_entries);
 846 out:
 847         return r;
 848 }
 849
 850 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
 851 {
 852         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
 853         struct kvm_cpuid_entry2 *ej;
 854         int j = i;
 855         int nent = vcpu->arch.cpuid_nent;
 856
 857         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
 858         /* when no next entry is found, the current entry[i] is reselected */
 859         do {
 860                 j = (j + 1) % nent;
 861                 ej = &vcpu->arch.cpuid_entries[j];
 862         } while (ej->function != e->function);
 863
 864         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
 865
 866         return j;
 867 }
 868
 869 /* find an entry with matching function, matching index (if needed), and that
 870  * should be read next (if it's stateful) */
 871 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
 872         u32 function, u32 index)
 873 {
 874         if (e->function != function)
 875                 return 0;
 876         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
 877                 return 0;
 878         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
 879             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
 880                 return 0;
 881         return 1;
 882 }
 883
 884 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
 885                                               u32 function, u32 index)
 886 {
 887         int i;
 888         struct kvm_cpuid_entry2 *best = NULL;
 889
 890         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 891                 struct kvm_cpuid_entry2 *e;
 892
 893                 e = &vcpu->arch.cpuid_entries[i];
 894                 if (is_matching_cpuid_entry(e, function, index)) {
 895                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
 896                                 move_to_next_stateful_cpuid_entry(vcpu, i);
 897                         best = e;
 898                         break;
 899                 }
 900         }
 901         return best;
 902 }
 903 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
 904
 905 /*
 906  * If no match is found, check whether we exceed the vCPU's limit
 907  * and return the content of the highest valid _standard_ leaf instead.
 908  * This is to satisfy the CPUID specification.
 909  */
 910 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
 911                                                   u32 function, u32 index)
 912 {
 913         struct kvm_cpuid_entry2 *maxlevel;
 914
 915         maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
 916         if (!maxlevel || maxlevel->eax >= function)
 917                 return NULL;
 918         if (function & 0x80000000) {
 919                 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
 920                 if (!maxlevel)
 921                         return NULL;
 922         }
 923         return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
 924 }
 925
 926 bool kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx,
 927                u32 *ecx, u32 *edx, bool check_limit)
 928 {
 929         u32 function = *eax, index = *ecx;
 930         struct kvm_cpuid_entry2 *best;
 931         bool entry_found = true;
 932
 933         best = kvm_find_cpuid_entry(vcpu, function, index);
 934
 935         if (!best) {
 936                 entry_found = false;
 937                 if (!check_limit)
 938                         goto out;
 939
 940                 best = check_cpuid_limit(vcpu, function, index);
 941         }
 942
 943 out:
 944         if (best) {
 945                 *eax = best->eax;
 946                 *ebx = best->ebx;
 947                 *ecx = best->ecx;
 948                 *edx = best->edx;
 949         } else
 950                 *eax = *ebx = *ecx = *edx = 0;
 951         trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx, entry_found);
 952         return entry_found;
 953 }
 954 EXPORT_SYMBOL_GPL(kvm_cpuid);
 955
 956 int kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
 957 {
 958         u32 eax, ebx, ecx, edx;
 959
 960         if (cpuid_fault_enabled(vcpu) && !kvm_require_cpl(vcpu, 0))
 961                 return 1;
 962
 963         eax = kvm_rax_read(vcpu);
 964         ecx = kvm_rcx_read(vcpu);
 965         kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx, true);
 966         kvm_rax_write(vcpu, eax);
 967         kvm_rbx_write(vcpu, ebx);
 968         kvm_rcx_write(vcpu, ecx);
 969         kvm_rdx_write(vcpu, edx);
 970         return kvm_skip_emulated_instruction(vcpu);
 971 }
 972 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);