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 <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         if (apic) {
  84                 if (best->ecx & F(TSC_DEADLINE_TIMER))
  85                         apic->lapic_timer.timer_mode_mask = 3 << 17;
  86                 else
  87                         apic->lapic_timer.timer_mode_mask = 1 << 17;
  88         }
  89
  90         best = kvm_find_cpuid_entry(vcpu, 7, 0);
  91         if (best) {
  92                 /* Update OSPKE bit */
  93                 if (boot_cpu_has(X86_FEATURE_PKU) && best->function == 0x7) {
  94                         best->ecx &= ~F(OSPKE);
  95                         if (kvm_read_cr4_bits(vcpu, X86_CR4_PKE))
  96                                 best->ecx |= F(OSPKE);
  97                 }
  98         }
  99
 100         best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
 101         if (!best) {
 102                 vcpu->arch.guest_supported_xcr0 = 0;
 103                 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
 104         } else {
 105                 vcpu->arch.guest_supported_xcr0 =
 106                         (best->eax | ((u64)best->edx << 32)) &
 107                         kvm_supported_xcr0();
 108                 vcpu->arch.guest_xstate_size = best->ebx =
 109                         xstate_required_size(vcpu->arch.xcr0, false);
 110         }
 111
 112         best = kvm_find_cpuid_entry(vcpu, 0xD, 1);
 113         if (best && (best->eax & (F(XSAVES) | F(XSAVEC))))
 114                 best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
 115
 116         kvm_x86_ops->fpu_activate(vcpu);
 117
 118         /*
 119          * The existing code assumes virtual address is 48-bit in the canonical
 120          * address checks; exit if it is ever changed.
 121          */
 122         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
 123         if (best && ((best->eax & 0xff00) >> 8) != 48 &&
 124                 ((best->eax & 0xff00) >> 8) != 0)
 125                 return -EINVAL;
 126
 127         /* Update physical-address width */
 128         vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
 129
 130         kvm_pmu_refresh(vcpu);
 131         return 0;
 132 }
 133
 134 static int is_efer_nx(void)
 135 {
 136         unsigned long long efer = 0;
 137
 138         rdmsrl_safe(MSR_EFER, &efer);
 139         return efer & EFER_NX;
 140 }
 141
 142 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
 143 {
 144         int i;
 145         struct kvm_cpuid_entry2 *e, *entry;
 146
 147         entry = NULL;
 148         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 149                 e = &vcpu->arch.cpuid_entries[i];
 150                 if (e->function == 0x80000001) {
 151                         entry = e;
 152                         break;
 153                 }
 154         }
 155         if (entry && (entry->edx & F(NX)) && !is_efer_nx()) {
 156                 entry->edx &= ~F(NX);
 157                 printk(KERN_INFO "kvm: guest NX capability removed\n");
 158         }
 159 }
 160
 161 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
 162 {
 163         struct kvm_cpuid_entry2 *best;
 164
 165         best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
 166         if (!best || best->eax < 0x80000008)
 167                 goto not_found;
 168         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
 169         if (best)
 170                 return best->eax & 0xff;
 171 not_found:
 172         return 36;
 173 }
 174 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr);
 175
 176 /* when an old userspace process fills a new kernel module */
 177 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
 178                              struct kvm_cpuid *cpuid,
 179                              struct kvm_cpuid_entry __user *entries)
 180 {
 181         int r, i;
 182         struct kvm_cpuid_entry *cpuid_entries = NULL;
 183
 184         r = -E2BIG;
 185         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 186                 goto out;
 187         r = -ENOMEM;
 188         if (cpuid->nent) {
 189                 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) *
 190                                         cpuid->nent);
 191                 if (!cpuid_entries)
 192                         goto out;
 193                 r = -EFAULT;
 194                 if (copy_from_user(cpuid_entries, entries,
 195                                    cpuid->nent * sizeof(struct kvm_cpuid_entry)))
 196                         goto out;
 197         }
 198         for (i = 0; i < cpuid->nent; i++) {
 199                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
 200                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
 201                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
 202                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
 203                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
 204                 vcpu->arch.cpuid_entries[i].index = 0;
 205                 vcpu->arch.cpuid_entries[i].flags = 0;
 206                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
 207                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
 208                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
 209         }
 210         vcpu->arch.cpuid_nent = cpuid->nent;
 211         cpuid_fix_nx_cap(vcpu);
 212         kvm_apic_set_version(vcpu);
 213         kvm_x86_ops->cpuid_update(vcpu);
 214         r = kvm_update_cpuid(vcpu);
 215
 216 out:
 217         vfree(cpuid_entries);
 218         return r;
 219 }
 220
 221 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
 222                               struct kvm_cpuid2 *cpuid,
 223                               struct kvm_cpuid_entry2 __user *entries)
 224 {
 225         int r;
 226
 227         r = -E2BIG;
 228         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 229                 goto out;
 230         r = -EFAULT;
 231         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
 232                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
 233                 goto out;
 234         vcpu->arch.cpuid_nent = cpuid->nent;
 235         kvm_apic_set_version(vcpu);
 236         kvm_x86_ops->cpuid_update(vcpu);
 237         r = kvm_update_cpuid(vcpu);
 238 out:
 239         return r;
 240 }
 241
 242 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
 243                               struct kvm_cpuid2 *cpuid,
 244                               struct kvm_cpuid_entry2 __user *entries)
 245 {
 246         int r;
 247
 248         r = -E2BIG;
 249         if (cpuid->nent < vcpu->arch.cpuid_nent)
 250                 goto out;
 251         r = -EFAULT;
 252         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
 253                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
 254                 goto out;
 255         return 0;
 256
 257 out:
 258         cpuid->nent = vcpu->arch.cpuid_nent;
 259         return r;
 260 }
 261
 262 static void cpuid_mask(u32 *word, int wordnum)
 263 {
 264         *word &= boot_cpu_data.x86_capability[wordnum];
 265 }
 266
 267 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 268                            u32 index)
 269 {
 270         entry->function = function;
 271         entry->index = index;
 272         cpuid_count(entry->function, entry->index,
 273                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
 274         entry->flags = 0;
 275 }
 276
 277 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry,
 278                                    u32 func, u32 index, int *nent, int maxnent)
 279 {
 280         switch (func) {
 281         case 0:
 282                 entry->eax = 7;
 283                 ++*nent;
 284                 break;
 285         case 1:
 286                 entry->ecx = F(MOVBE);
 287                 ++*nent;
 288                 break;
 289         case 7:
 290                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 291                 if (index == 0)
 292                         entry->ecx = F(RDPID);
 293                 ++*nent;
 294         default:
 295                 break;
 296         }
 297
 298         entry->function = func;
 299         entry->index = index;
 300
 301         return 0;
 302 }
 303
 304 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 305                                  u32 index, int *nent, int maxnent)
 306 {
 307         int r;
 308         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
 309 #ifdef CONFIG_X86_64
 310         unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
 311                                 ? F(GBPAGES) : 0;
 312         unsigned f_lm = F(LM);
 313 #else
 314         unsigned f_gbpages = 0;
 315         unsigned f_lm = 0;
 316 #endif
 317         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
 318         unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
 319         unsigned f_mpx = kvm_mpx_supported() ? F(MPX) : 0;
 320         unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0;
 321
 322         /* cpuid 1.edx */
 323         const u32 kvm_cpuid_1_edx_x86_features =
 324                 F(FPU) | F(VME) | F(DE) | F(PSE) |
 325                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
 326                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
 327                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
 328                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
 329                 0 /* Reserved, DS, ACPI */ | F(MMX) |
 330                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
 331                 0 /* HTT, TM, Reserved, PBE */;
 332         /* cpuid 0x80000001.edx */
 333         const u32 kvm_cpuid_8000_0001_edx_x86_features =
 334                 F(FPU) | F(VME) | F(DE) | F(PSE) |
 335                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
 336                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
 337                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
 338                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
 339                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
 340                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
 341                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
 342         /* cpuid 1.ecx */
 343         const u32 kvm_cpuid_1_ecx_x86_features =
 344                 /* NOTE: MONITOR (and MWAIT) are emulated as NOP,
 345                  * but *not* advertised to guests via CPUID ! */
 346                 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
 347                 0 /* DS-CPL, VMX, SMX, EST */ |
 348                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
 349                 F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
 350                 F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
 351                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
 352                 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
 353                 F(F16C) | F(RDRAND);
 354         /* cpuid 0x80000001.ecx */
 355         const u32 kvm_cpuid_8000_0001_ecx_x86_features =
 356                 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
 357                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
 358                 F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
 359                 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
 360
 361         /* cpuid 0x80000008.ebx */
 362         const u32 kvm_cpuid_8000_0008_ebx_x86_features =
 363                 F(AMD_IBPB) | F(AMD_IBRS) | F(AMD_SSBD) | F(VIRT_SSBD) |
 364                 F(AMD_SSB_NO) | F(AMD_STIBP);
 365
 366         /* cpuid 0xC0000001.edx */
 367         const u32 kvm_cpuid_C000_0001_edx_x86_features =
 368                 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
 369                 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
 370                 F(PMM) | F(PMM_EN);
 371
 372         /* cpuid 7.0.ebx */
 373         const u32 kvm_cpuid_7_0_ebx_x86_features =
 374                 F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
 375                 F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
 376                 F(ADX) | F(SMAP) | F(AVX512F) | F(AVX512PF) | F(AVX512ER) |
 377                 F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) |
 378                 F(AVX512BW) | F(AVX512VL);
 379
 380         /* cpuid 0xD.1.eax */
 381         const u32 kvm_cpuid_D_1_eax_x86_features =
 382                 F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves;
 383
 384         /* cpuid 7.0.ecx*/
 385         const u32 kvm_cpuid_7_0_ecx_x86_features = F(PKU) | 0 /*OSPKE*/;
 386
 387         /* cpuid 7.0.edx*/
 388         const u32 kvm_cpuid_7_0_edx_x86_features =
 389                 F(SPEC_CTRL) | F(SPEC_CTRL_SSBD) | F(ARCH_CAPABILITIES) |
 390                 F(INTEL_STIBP) | F(MD_CLEAR);
 391
 392         /* all calls to cpuid_count() should be made on the same cpu */
 393         get_cpu();
 394
 395         r = -E2BIG;
 396
 397         if (WARN_ON(*nent >= maxnent))
 398                 goto out;
 399
 400         do_cpuid_1_ent(entry, function, index);
 401         ++*nent;
 402
 403         switch (function) {
 404         case 0:
 405                 entry->eax = min(entry->eax, (u32)0xd);
 406                 break;
 407         case 1:
 408                 entry->edx &= kvm_cpuid_1_edx_x86_features;
 409                 cpuid_mask(&entry->edx, CPUID_1_EDX);
 410                 entry->ecx &= kvm_cpuid_1_ecx_x86_features;
 411                 cpuid_mask(&entry->ecx, CPUID_1_ECX);
 412                 /* we support x2apic emulation even if host does not support
 413                  * it since we emulate x2apic in software */
 414                 entry->ecx |= F(X2APIC);
 415                 break;
 416         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
 417          * may return different values. This forces us to get_cpu() before
 418          * issuing the first command, and also to emulate this annoying behavior
 419          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
 420         case 2: {
 421                 int t, times = entry->eax & 0xff;
 422
 423                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 424                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
 425                 for (t = 1; t < times; ++t) {
 426                         if (*nent >= maxnent)
 427                                 goto out;
 428
 429                         do_cpuid_1_ent(&entry[t], function, 0);
 430                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 431                         ++*nent;
 432                 }
 433                 break;
 434         }
 435         /* function 4 has additional index. */
 436         case 4: {
 437                 int i, cache_type;
 438
 439                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 440                 /* read more entries until cache_type is zero */
 441                 for (i = 1; ; ++i) {
 442                         if (*nent >= maxnent)
 443                                 goto out;
 444
 445                         cache_type = entry[i - 1].eax & 0x1f;
 446                         if (!cache_type)
 447                                 break;
 448                         do_cpuid_1_ent(&entry[i], function, i);
 449                         entry[i].flags |=
 450                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 451                         ++*nent;
 452                 }
 453                 break;
 454         }
 455         case 6: /* Thermal management */
 456                 entry->eax = 0x4; /* allow ARAT */
 457                 entry->ebx = 0;
 458                 entry->ecx = 0;
 459                 entry->edx = 0;
 460                 break;
 461         case 7: {
 462                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 463                 /* Mask ebx against host capability word 9 */
 464                 if (index == 0) {
 465                         entry->ebx &= kvm_cpuid_7_0_ebx_x86_features;
 466                         cpuid_mask(&entry->ebx, CPUID_7_0_EBX);
 467                         // TSC_ADJUST is emulated
 468                         entry->ebx |= F(TSC_ADJUST);
 469                         entry->ecx &= kvm_cpuid_7_0_ecx_x86_features;
 470                         cpuid_mask(&entry->ecx, CPUID_7_ECX);
 471                         /* PKU is not yet implemented for shadow paging. */
 472                         if (!tdp_enabled || !boot_cpu_has(X86_FEATURE_OSPKE))
 473                                 entry->ecx &= ~F(PKU);
 474
 475                         entry->edx &= kvm_cpuid_7_0_edx_x86_features;
 476                         cpuid_mask(&entry->edx, CPUID_7_EDX);
 477                         if (boot_cpu_has(X86_FEATURE_IBPB) &&
 478                             boot_cpu_has(X86_FEATURE_IBRS))
 479                                 entry->edx |= F(SPEC_CTRL);
 480                         if (boot_cpu_has(X86_FEATURE_STIBP))
 481                                 entry->edx |= F(INTEL_STIBP);
 482                         if (boot_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) ||
 483                             boot_cpu_has(X86_FEATURE_AMD_SSBD))
 484                                 entry->edx |= F(SPEC_CTRL_SSBD);
 485                         /*
 486                          * We emulate ARCH_CAPABILITIES in software even
 487                          * if the host doesn't support it.
 488                          */
 489                         entry->edx |= F(ARCH_CAPABILITIES);
 490                 } else {
 491                         entry->ebx = 0;
 492                         entry->ecx = 0;
 493                         entry->edx = 0;
 494                 }
 495                 entry->eax = 0;
 496                 break;
 497         }
 498         case 9:
 499                 break;
 500         case 0xa: { /* Architectural Performance Monitoring */
 501                 struct x86_pmu_capability cap;
 502                 union cpuid10_eax eax;
 503                 union cpuid10_edx edx;
 504
 505                 perf_get_x86_pmu_capability(&cap);
 506
 507                 /*
 508                  * Only support guest architectural pmu on a host
 509                  * with architectural pmu.
 510                  */
 511                 if (!cap.version)
 512                         memset(&cap, 0, sizeof(cap));
 513
 514                 eax.split.version_id = min(cap.version, 2);
 515                 eax.split.num_counters = cap.num_counters_gp;
 516                 eax.split.bit_width = cap.bit_width_gp;
 517                 eax.split.mask_length = cap.events_mask_len;
 518
 519                 edx.split.num_counters_fixed = cap.num_counters_fixed;
 520                 edx.split.bit_width_fixed = cap.bit_width_fixed;
 521                 edx.split.reserved = 0;
 522
 523                 entry->eax = eax.full;
 524                 entry->ebx = cap.events_mask;
 525                 entry->ecx = 0;
 526                 entry->edx = edx.full;
 527                 break;
 528         }
 529         /* function 0xb has additional index. */
 530         case 0xb: {
 531                 int i, level_type;
 532
 533                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 534                 /* read more entries until level_type is zero */
 535                 for (i = 1; ; ++i) {
 536                         if (*nent >= maxnent)
 537                                 goto out;
 538
 539                         level_type = entry[i - 1].ecx & 0xff00;
 540                         if (!level_type)
 541                                 break;
 542                         do_cpuid_1_ent(&entry[i], function, i);
 543                         entry[i].flags |=
 544                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 545                         ++*nent;
 546                 }
 547                 break;
 548         }
 549         case 0xd: {
 550                 int idx, i;
 551                 u64 supported = kvm_supported_xcr0();
 552
 553                 entry->eax &= supported;
 554                 entry->ebx = xstate_required_size(supported, false);
 555                 entry->ecx = entry->ebx;
 556                 entry->edx &= supported >> 32;
 557                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 558                 if (!supported)
 559                         break;
 560
 561                 for (idx = 1, i = 1; idx < 64; ++idx) {
 562                         u64 mask = ((u64)1 << idx);
 563                         if (*nent >= maxnent)
 564                                 goto out;
 565
 566                         do_cpuid_1_ent(&entry[i], function, idx);
 567                         if (idx == 1) {
 568                                 entry[i].eax &= kvm_cpuid_D_1_eax_x86_features;
 569                                 cpuid_mask(&entry[i].eax, CPUID_D_1_EAX);
 570                                 entry[i].ebx = 0;
 571                                 if (entry[i].eax & (F(XSAVES)|F(XSAVEC)))
 572                                         entry[i].ebx =
 573                                                 xstate_required_size(supported,
 574                                                                      true);
 575                         } else {
 576                                 if (entry[i].eax == 0 || !(supported & mask))
 577                                         continue;
 578                                 if (WARN_ON_ONCE(entry[i].ecx & 1))
 579                                         continue;
 580                         }
 581                         entry[i].ecx = 0;
 582                         entry[i].edx = 0;
 583                         entry[i].flags |=
 584                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 585                         ++*nent;
 586                         ++i;
 587                 }
 588                 break;
 589         }
 590         case KVM_CPUID_SIGNATURE: {
 591                 static const char signature[12] = "KVMKVMKVM\0\0";
 592                 const u32 *sigptr = (const u32 *)signature;
 593                 entry->eax = KVM_CPUID_FEATURES;
 594                 entry->ebx = sigptr[0];
 595                 entry->ecx = sigptr[1];
 596                 entry->edx = sigptr[2];
 597                 break;
 598         }
 599         case KVM_CPUID_FEATURES:
 600                 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
 601                              (1 << KVM_FEATURE_NOP_IO_DELAY) |
 602                              (1 << KVM_FEATURE_CLOCKSOURCE2) |
 603                              (1 << KVM_FEATURE_ASYNC_PF) |
 604                              (1 << KVM_FEATURE_PV_EOI) |
 605                              (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
 606                              (1 << KVM_FEATURE_PV_UNHALT);
 607
 608                 if (sched_info_on())
 609                         entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
 610
 611                 entry->ebx = 0;
 612                 entry->ecx = 0;
 613                 entry->edx = 0;
 614                 break;
 615         case 0x80000000:
 616                 entry->eax = min(entry->eax, 0x8000001a);
 617                 break;
 618         case 0x80000001:
 619                 entry->edx &= kvm_cpuid_8000_0001_edx_x86_features;
 620                 cpuid_mask(&entry->edx, CPUID_8000_0001_EDX);
 621                 entry->ecx &= kvm_cpuid_8000_0001_ecx_x86_features;
 622                 cpuid_mask(&entry->ecx, CPUID_8000_0001_ECX);
 623                 break;
 624         case 0x80000007: /* Advanced power management */
 625                 /* invariant TSC is CPUID.80000007H:EDX[8] */
 626                 entry->edx &= (1 << 8);
 627                 /* mask against host */
 628                 entry->edx &= boot_cpu_data.x86_power;
 629                 entry->eax = entry->ebx = entry->ecx = 0;
 630                 break;
 631         case 0x80000008: {
 632                 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
 633                 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
 634                 unsigned phys_as = entry->eax & 0xff;
 635
 636                 /*
 637                  * Use bare metal's MAXPHADDR if the CPU doesn't report guest
 638                  * MAXPHYADDR separately, or if TDP (NPT) is disabled, as the
 639                  * guest version "applies only to guests using nested paging".
 640                  */
 641                 if (!g_phys_as || !tdp_enabled)
 642                         g_phys_as = phys_as;
 643
 644                 entry->eax = g_phys_as | (virt_as << 8);
 645                 entry->edx = 0;
 646                 /*
 647                  * IBRS, IBPB and VIRT_SSBD aren't necessarily present in
 648                  * hardware cpuid
 649                  */
 650                 if (boot_cpu_has(X86_FEATURE_AMD_IBPB))
 651                         entry->ebx |= F(AMD_IBPB);
 652                 if (boot_cpu_has(X86_FEATURE_AMD_IBRS))
 653                         entry->ebx |= F(AMD_IBRS);
 654                 if (boot_cpu_has(X86_FEATURE_VIRT_SSBD))
 655                         entry->ebx |= F(VIRT_SSBD);
 656                 entry->ebx &= kvm_cpuid_8000_0008_ebx_x86_features;
 657                 cpuid_mask(&entry->ebx, CPUID_8000_0008_EBX);
 658                 /*
 659                  * The preference is to use SPEC CTRL MSR instead of the
 660                  * VIRT_SPEC MSR.
 661                  */
 662                 if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD) &&
 663                     !boot_cpu_has(X86_FEATURE_AMD_SSBD))
 664                         entry->ebx |= F(VIRT_SSBD);
 665                 break;
 666         }
 667         case 0x80000019:
 668                 entry->ecx = entry->edx = 0;
 669                 break;
 670         case 0x8000001a:
 671                 break;
 672         case 0x8000001d:
 673                 break;
 674         /*Add support for Centaur's CPUID instruction*/
 675         case 0xC0000000:
 676                 /*Just support up to 0xC0000004 now*/
 677                 entry->eax = min(entry->eax, 0xC0000004);
 678                 break;
 679         case 0xC0000001:
 680                 entry->edx &= kvm_cpuid_C000_0001_edx_x86_features;
 681                 cpuid_mask(&entry->edx, CPUID_C000_0001_EDX);
 682                 break;
 683         case 3: /* Processor serial number */
 684         case 5: /* MONITOR/MWAIT */
 685         case 0xC0000002:
 686         case 0xC0000003:
 687         case 0xC0000004:
 688         default:
 689                 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
 690                 break;
 691         }
 692
 693         kvm_x86_ops->set_supported_cpuid(function, entry);
 694
 695         r = 0;
 696
 697 out:
 698         put_cpu();
 699
 700         return r;
 701 }
 702
 703 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
 704                         u32 idx, int *nent, int maxnent, unsigned int type)
 705 {
 706         if (*nent >= maxnent)
 707                 return -E2BIG;
 708
 709         if (type == KVM_GET_EMULATED_CPUID)
 710                 return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
 711
 712         return __do_cpuid_ent(entry, func, idx, nent, maxnent);
 713 }
 714
 715 #undef F
 716
 717 struct kvm_cpuid_param {
 718         u32 func;
 719         u32 idx;
 720         bool has_leaf_count;
 721         bool (*qualifier)(const struct kvm_cpuid_param *param);
 722 };
 723
 724 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
 725 {
 726         return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
 727 }
 728
 729 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
 730                                  __u32 num_entries, unsigned int ioctl_type)
 731 {
 732         int i;
 733         __u32 pad[3];
 734
 735         if (ioctl_type != KVM_GET_EMULATED_CPUID)
 736                 return false;
 737
 738         /*
 739          * We want to make sure that ->padding is being passed clean from
 740          * userspace in case we want to use it for something in the future.
 741          *
 742          * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
 743          * have to give ourselves satisfied only with the emulated side. /me
 744          * sheds a tear.
 745          */
 746         for (i = 0; i < num_entries; i++) {
 747                 if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
 748                         return true;
 749
 750                 if (pad[0] || pad[1] || pad[2])
 751                         return true;
 752         }
 753         return false;
 754 }
 755
 756 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
 757                             struct kvm_cpuid_entry2 __user *entries,
 758                             unsigned int type)
 759 {
 760         struct kvm_cpuid_entry2 *cpuid_entries;
 761         int limit, nent = 0, r = -E2BIG, i;
 762         u32 func;
 763         static const struct kvm_cpuid_param param[] = {
 764                 { .func = 0, .has_leaf_count = true },
 765                 { .func = 0x80000000, .has_leaf_count = true },
 766                 { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
 767                 { .func = KVM_CPUID_SIGNATURE },
 768                 { .func = KVM_CPUID_FEATURES },
 769         };
 770
 771         if (cpuid->nent < 1)
 772                 goto out;
 773         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 774                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
 775
 776         if (sanity_check_entries(entries, cpuid->nent, type))
 777                 return -EINVAL;
 778
 779         r = -ENOMEM;
 780         cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
 781         if (!cpuid_entries)
 782                 goto out;
 783
 784         r = 0;
 785         for (i = 0; i < ARRAY_SIZE(param); i++) {
 786                 const struct kvm_cpuid_param *ent = &param[i];
 787
 788                 if (ent->qualifier && !ent->qualifier(ent))
 789                         continue;
 790
 791                 r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
 792                                 &nent, cpuid->nent, type);
 793
 794                 if (r)
 795                         goto out_free;
 796
 797                 if (!ent->has_leaf_count)
 798                         continue;
 799
 800                 limit = cpuid_entries[nent - 1].eax;
 801                 for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
 802                         r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
 803                                      &nent, cpuid->nent, type);
 804
 805                 if (r)
 806                         goto out_free;
 807         }
 808
 809         r = -EFAULT;
 810         if (copy_to_user(entries, cpuid_entries,
 811                          nent * sizeof(struct kvm_cpuid_entry2)))
 812                 goto out_free;
 813         cpuid->nent = nent;
 814         r = 0;
 815
 816 out_free:
 817         vfree(cpuid_entries);
 818 out:
 819         return r;
 820 }
 821
 822 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
 823 {
 824         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
 825         struct kvm_cpuid_entry2 *ej;
 826         int j = i;
 827         int nent = vcpu->arch.cpuid_nent;
 828
 829         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
 830         /* when no next entry is found, the current entry[i] is reselected */
 831         do {
 832                 j = (j + 1) % nent;
 833                 ej = &vcpu->arch.cpuid_entries[j];
 834         } while (ej->function != e->function);
 835
 836         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
 837
 838         return j;
 839 }
 840
 841 /* find an entry with matching function, matching index (if needed), and that
 842  * should be read next (if it's stateful) */
 843 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
 844         u32 function, u32 index)
 845 {
 846         if (e->function != function)
 847                 return 0;
 848         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
 849                 return 0;
 850         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
 851             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
 852                 return 0;
 853         return 1;
 854 }
 855
 856 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
 857                                               u32 function, u32 index)
 858 {
 859         int i;
 860         struct kvm_cpuid_entry2 *best = NULL;
 861
 862         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 863                 struct kvm_cpuid_entry2 *e;
 864
 865                 e = &vcpu->arch.cpuid_entries[i];
 866                 if (is_matching_cpuid_entry(e, function, index)) {
 867                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
 868                                 move_to_next_stateful_cpuid_entry(vcpu, i);
 869                         best = e;
 870                         break;
 871                 }
 872         }
 873         return best;
 874 }
 875 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
 876
 877 /*
 878  * If no match is found, check whether we exceed the vCPU's limit
 879  * and return the content of the highest valid _standard_ leaf instead.
 880  * This is to satisfy the CPUID specification.
 881  */
 882 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
 883                                                   u32 function, u32 index)
 884 {
 885         struct kvm_cpuid_entry2 *maxlevel;
 886
 887         maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
 888         if (!maxlevel || maxlevel->eax >= function)
 889                 return NULL;
 890         if (function & 0x80000000) {
 891                 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
 892                 if (!maxlevel)
 893                         return NULL;
 894         }
 895         return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
 896 }
 897
 898 void kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
 899 {
 900         u32 function = *eax, index = *ecx;
 901         struct kvm_cpuid_entry2 *best;
 902
 903         best = kvm_find_cpuid_entry(vcpu, function, index);
 904
 905         if (!best)
 906                 best = check_cpuid_limit(vcpu, function, index);
 907
 908         if (best) {
 909                 *eax = best->eax;
 910                 *ebx = best->ebx;
 911                 *ecx = best->ecx;
 912                 *edx = best->edx;
 913         } else
 914                 *eax = *ebx = *ecx = *edx = 0;
 915         trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx);
 916 }
 917 EXPORT_SYMBOL_GPL(kvm_cpuid);
 918
 919 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
 920 {
 921         u32 function, eax, ebx, ecx, edx;
 922
 923         function = eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
 924         ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
 925         kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx);
 926         kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
 927         kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
 928         kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
 929         kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
 930         kvm_x86_ops->skip_emulated_instruction(vcpu);
 931 }
 932 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);