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 = 1;         /* only one leaf currently */
 283                 ++*nent;
 284                 break;
 285         case 1:
 286                 entry->ecx = F(MOVBE);
 287                 ++*nent;
 288                 break;
 289         default:
 290                 break;
 291         }
 292
 293         entry->function = func;
 294         entry->index = index;
 295
 296         return 0;
 297 }
 298
 299 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 300                                  u32 index, int *nent, int maxnent)
 301 {
 302         int r;
 303         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
 304 #ifdef CONFIG_X86_64
 305         unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
 306                                 ? F(GBPAGES) : 0;
 307         unsigned f_lm = F(LM);
 308 #else
 309         unsigned f_gbpages = 0;
 310         unsigned f_lm = 0;
 311 #endif
 312         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
 313         unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
 314         unsigned f_mpx = kvm_mpx_supported() ? F(MPX) : 0;
 315         unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0;
 316
 317         /* cpuid 1.edx */
 318         const u32 kvm_cpuid_1_edx_x86_features =
 319                 F(FPU) | F(VME) | F(DE) | F(PSE) |
 320                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
 321                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
 322                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
 323                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
 324                 0 /* Reserved, DS, ACPI */ | F(MMX) |
 325                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
 326                 0 /* HTT, TM, Reserved, PBE */;
 327         /* cpuid 0x80000001.edx */
 328         const u32 kvm_cpuid_8000_0001_edx_x86_features =
 329                 F(FPU) | F(VME) | F(DE) | F(PSE) |
 330                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
 331                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
 332                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
 333                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
 334                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
 335                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
 336                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
 337         /* cpuid 1.ecx */
 338         const u32 kvm_cpuid_1_ecx_x86_features =
 339                 /* NOTE: MONITOR (and MWAIT) are emulated as NOP,
 340                  * but *not* advertised to guests via CPUID ! */
 341                 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
 342                 0 /* DS-CPL, VMX, SMX, EST */ |
 343                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
 344                 F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
 345                 F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
 346                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
 347                 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
 348                 F(F16C) | F(RDRAND);
 349         /* cpuid 0x80000001.ecx */
 350         const u32 kvm_cpuid_8000_0001_ecx_x86_features =
 351                 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
 352                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
 353                 F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
 354                 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
 355
 356         /* cpuid 0xC0000001.edx */
 357         const u32 kvm_cpuid_C000_0001_edx_x86_features =
 358                 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
 359                 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
 360                 F(PMM) | F(PMM_EN);
 361
 362         /* cpuid 7.0.ebx */
 363         const u32 kvm_cpuid_7_0_ebx_x86_features =
 364                 F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
 365                 F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
 366                 F(ADX) | F(SMAP) | F(AVX512F) | F(AVX512PF) | F(AVX512ER) |
 367                 F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) |
 368                 F(AVX512BW) | F(AVX512VL);
 369
 370         /* cpuid 0xD.1.eax */
 371         const u32 kvm_cpuid_D_1_eax_x86_features =
 372                 F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves;
 373
 374         /* cpuid 7.0.ecx*/
 375         const u32 kvm_cpuid_7_0_ecx_x86_features = F(PKU) | 0 /*OSPKE*/;
 376
 377         /* all calls to cpuid_count() should be made on the same cpu */
 378         get_cpu();
 379
 380         r = -E2BIG;
 381
 382         if (*nent >= maxnent)
 383                 goto out;
 384
 385         do_cpuid_1_ent(entry, function, index);
 386         ++*nent;
 387
 388         switch (function) {
 389         case 0:
 390                 entry->eax = min(entry->eax, (u32)0xd);
 391                 break;
 392         case 1:
 393                 entry->edx &= kvm_cpuid_1_edx_x86_features;
 394                 cpuid_mask(&entry->edx, CPUID_1_EDX);
 395                 entry->ecx &= kvm_cpuid_1_ecx_x86_features;
 396                 cpuid_mask(&entry->ecx, CPUID_1_ECX);
 397                 /* we support x2apic emulation even if host does not support
 398                  * it since we emulate x2apic in software */
 399                 entry->ecx |= F(X2APIC);
 400                 break;
 401         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
 402          * may return different values. This forces us to get_cpu() before
 403          * issuing the first command, and also to emulate this annoying behavior
 404          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
 405         case 2: {
 406                 int t, times = entry->eax & 0xff;
 407
 408                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 409                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
 410                 for (t = 1; t < times; ++t) {
 411                         if (*nent >= maxnent)
 412                                 goto out;
 413
 414                         do_cpuid_1_ent(&entry[t], function, 0);
 415                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 416                         ++*nent;
 417                 }
 418                 break;
 419         }
 420         /* function 4 has additional index. */
 421         case 4: {
 422                 int i, cache_type;
 423
 424                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 425                 /* read more entries until cache_type is zero */
 426                 for (i = 1; ; ++i) {
 427                         if (*nent >= maxnent)
 428                                 goto out;
 429
 430                         cache_type = entry[i - 1].eax & 0x1f;
 431                         if (!cache_type)
 432                                 break;
 433                         do_cpuid_1_ent(&entry[i], function, i);
 434                         entry[i].flags |=
 435                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 436                         ++*nent;
 437                 }
 438                 break;
 439         }
 440         case 6: /* Thermal management */
 441                 entry->eax = 0x4; /* allow ARAT */
 442                 entry->ebx = 0;
 443                 entry->ecx = 0;
 444                 entry->edx = 0;
 445                 break;
 446         case 7: {
 447                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 448                 /* Mask ebx against host capability word 9 */
 449                 if (index == 0) {
 450                         entry->ebx &= kvm_cpuid_7_0_ebx_x86_features;
 451                         cpuid_mask(&entry->ebx, CPUID_7_0_EBX);
 452                         // TSC_ADJUST is emulated
 453                         entry->ebx |= F(TSC_ADJUST);
 454                         entry->ecx &= kvm_cpuid_7_0_ecx_x86_features;
 455                         cpuid_mask(&entry->ecx, CPUID_7_ECX);
 456                         /* PKU is not yet implemented for shadow paging. */
 457                         if (!tdp_enabled)
 458                                 entry->ecx &= ~F(PKU);
 459                 } else {
 460                         entry->ebx = 0;
 461                         entry->ecx = 0;
 462                 }
 463                 entry->eax = 0;
 464                 entry->edx = 0;
 465                 break;
 466         }
 467         case 9:
 468                 break;
 469         case 0xa: { /* Architectural Performance Monitoring */
 470                 struct x86_pmu_capability cap;
 471                 union cpuid10_eax eax;
 472                 union cpuid10_edx edx;
 473
 474                 perf_get_x86_pmu_capability(&cap);
 475
 476                 /*
 477                  * Only support guest architectural pmu on a host
 478                  * with architectural pmu.
 479                  */
 480                 if (!cap.version)
 481                         memset(&cap, 0, sizeof(cap));
 482
 483                 eax.split.version_id = min(cap.version, 2);
 484                 eax.split.num_counters = cap.num_counters_gp;
 485                 eax.split.bit_width = cap.bit_width_gp;
 486                 eax.split.mask_length = cap.events_mask_len;
 487
 488                 edx.split.num_counters_fixed = cap.num_counters_fixed;
 489                 edx.split.bit_width_fixed = cap.bit_width_fixed;
 490                 edx.split.reserved = 0;
 491
 492                 entry->eax = eax.full;
 493                 entry->ebx = cap.events_mask;
 494                 entry->ecx = 0;
 495                 entry->edx = edx.full;
 496                 break;
 497         }
 498         /* function 0xb has additional index. */
 499         case 0xb: {
 500                 int i, level_type;
 501
 502                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 503                 /* read more entries until level_type is zero */
 504                 for (i = 1; ; ++i) {
 505                         if (*nent >= maxnent)
 506                                 goto out;
 507
 508                         level_type = entry[i - 1].ecx & 0xff00;
 509                         if (!level_type)
 510                                 break;
 511                         do_cpuid_1_ent(&entry[i], function, i);
 512                         entry[i].flags |=
 513                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 514                         ++*nent;
 515                 }
 516                 break;
 517         }
 518         case 0xd: {
 519                 int idx, i;
 520                 u64 supported = kvm_supported_xcr0();
 521
 522                 entry->eax &= supported;
 523                 entry->ebx = xstate_required_size(supported, false);
 524                 entry->ecx = entry->ebx;
 525                 entry->edx &= supported >> 32;
 526                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 527                 if (!supported)
 528                         break;
 529
 530                 for (idx = 1, i = 1; idx < 64; ++idx) {
 531                         u64 mask = ((u64)1 << idx);
 532                         if (*nent >= maxnent)
 533                                 goto out;
 534
 535                         do_cpuid_1_ent(&entry[i], function, idx);
 536                         if (idx == 1) {
 537                                 entry[i].eax &= kvm_cpuid_D_1_eax_x86_features;
 538                                 cpuid_mask(&entry[i].eax, CPUID_D_1_EAX);
 539                                 entry[i].ebx = 0;
 540                                 if (entry[i].eax & (F(XSAVES)|F(XSAVEC)))
 541                                         entry[i].ebx =
 542                                                 xstate_required_size(supported,
 543                                                                      true);
 544                         } else {
 545                                 if (entry[i].eax == 0 || !(supported & mask))
 546                                         continue;
 547                                 if (WARN_ON_ONCE(entry[i].ecx & 1))
 548                                         continue;
 549                         }
 550                         entry[i].ecx = 0;
 551                         entry[i].edx = 0;
 552                         entry[i].flags |=
 553                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 554                         ++*nent;
 555                         ++i;
 556                 }
 557                 break;
 558         }
 559         case KVM_CPUID_SIGNATURE: {
 560                 static const char signature[12] = "KVMKVMKVM\0\0";
 561                 const u32 *sigptr = (const u32 *)signature;
 562                 entry->eax = KVM_CPUID_FEATURES;
 563                 entry->ebx = sigptr[0];
 564                 entry->ecx = sigptr[1];
 565                 entry->edx = sigptr[2];
 566                 break;
 567         }
 568         case KVM_CPUID_FEATURES:
 569                 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
 570                              (1 << KVM_FEATURE_NOP_IO_DELAY) |
 571                              (1 << KVM_FEATURE_CLOCKSOURCE2) |
 572                              (1 << KVM_FEATURE_ASYNC_PF) |
 573                              (1 << KVM_FEATURE_PV_EOI) |
 574                              (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
 575                              (1 << KVM_FEATURE_PV_UNHALT);
 576
 577                 if (sched_info_on())
 578                         entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
 579
 580                 entry->ebx = 0;
 581                 entry->ecx = 0;
 582                 entry->edx = 0;
 583                 break;
 584         case 0x80000000:
 585                 entry->eax = min(entry->eax, 0x8000001a);
 586                 break;
 587         case 0x80000001:
 588                 entry->edx &= kvm_cpuid_8000_0001_edx_x86_features;
 589                 cpuid_mask(&entry->edx, CPUID_8000_0001_EDX);
 590                 entry->ecx &= kvm_cpuid_8000_0001_ecx_x86_features;
 591                 cpuid_mask(&entry->ecx, CPUID_8000_0001_ECX);
 592                 break;
 593         case 0x80000007: /* Advanced power management */
 594                 /* invariant TSC is CPUID.80000007H:EDX[8] */
 595                 entry->edx &= (1 << 8);
 596                 /* mask against host */
 597                 entry->edx &= boot_cpu_data.x86_power;
 598                 entry->eax = entry->ebx = entry->ecx = 0;
 599                 break;
 600         case 0x80000008: {
 601                 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
 602                 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
 603                 unsigned phys_as = entry->eax & 0xff;
 604
 605                 if (!g_phys_as)
 606                         g_phys_as = phys_as;
 607                 entry->eax = g_phys_as | (virt_as << 8);
 608                 entry->ebx = entry->edx = 0;
 609                 break;
 610         }
 611         case 0x80000019:
 612                 entry->ecx = entry->edx = 0;
 613                 break;
 614         case 0x8000001a:
 615                 break;
 616         case 0x8000001d:
 617                 break;
 618         /*Add support for Centaur's CPUID instruction*/
 619         case 0xC0000000:
 620                 /*Just support up to 0xC0000004 now*/
 621                 entry->eax = min(entry->eax, 0xC0000004);
 622                 break;
 623         case 0xC0000001:
 624                 entry->edx &= kvm_cpuid_C000_0001_edx_x86_features;
 625                 cpuid_mask(&entry->edx, CPUID_C000_0001_EDX);
 626                 break;
 627         case 3: /* Processor serial number */
 628         case 5: /* MONITOR/MWAIT */
 629         case 0xC0000002:
 630         case 0xC0000003:
 631         case 0xC0000004:
 632         default:
 633                 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
 634                 break;
 635         }
 636
 637         kvm_x86_ops->set_supported_cpuid(function, entry);
 638
 639         r = 0;
 640
 641 out:
 642         put_cpu();
 643
 644         return r;
 645 }
 646
 647 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
 648                         u32 idx, int *nent, int maxnent, unsigned int type)
 649 {
 650         if (type == KVM_GET_EMULATED_CPUID)
 651                 return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
 652
 653         return __do_cpuid_ent(entry, func, idx, nent, maxnent);
 654 }
 655
 656 #undef F
 657
 658 struct kvm_cpuid_param {
 659         u32 func;
 660         u32 idx;
 661         bool has_leaf_count;
 662         bool (*qualifier)(const struct kvm_cpuid_param *param);
 663 };
 664
 665 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
 666 {
 667         return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
 668 }
 669
 670 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
 671                                  __u32 num_entries, unsigned int ioctl_type)
 672 {
 673         int i;
 674         __u32 pad[3];
 675
 676         if (ioctl_type != KVM_GET_EMULATED_CPUID)
 677                 return false;
 678
 679         /*
 680          * We want to make sure that ->padding is being passed clean from
 681          * userspace in case we want to use it for something in the future.
 682          *
 683          * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
 684          * have to give ourselves satisfied only with the emulated side. /me
 685          * sheds a tear.
 686          */
 687         for (i = 0; i < num_entries; i++) {
 688                 if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
 689                         return true;
 690
 691                 if (pad[0] || pad[1] || pad[2])
 692                         return true;
 693         }
 694         return false;
 695 }
 696
 697 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
 698                             struct kvm_cpuid_entry2 __user *entries,
 699                             unsigned int type)
 700 {
 701         struct kvm_cpuid_entry2 *cpuid_entries;
 702         int limit, nent = 0, r = -E2BIG, i;
 703         u32 func;
 704         static const struct kvm_cpuid_param param[] = {
 705                 { .func = 0, .has_leaf_count = true },
 706                 { .func = 0x80000000, .has_leaf_count = true },
 707                 { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
 708                 { .func = KVM_CPUID_SIGNATURE },
 709                 { .func = KVM_CPUID_FEATURES },
 710         };
 711
 712         if (cpuid->nent < 1)
 713                 goto out;
 714         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 715                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
 716
 717         if (sanity_check_entries(entries, cpuid->nent, type))
 718                 return -EINVAL;
 719
 720         r = -ENOMEM;
 721         cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
 722         if (!cpuid_entries)
 723                 goto out;
 724
 725         r = 0;
 726         for (i = 0; i < ARRAY_SIZE(param); i++) {
 727                 const struct kvm_cpuid_param *ent = &param[i];
 728
 729                 if (ent->qualifier && !ent->qualifier(ent))
 730                         continue;
 731
 732                 r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
 733                                 &nent, cpuid->nent, type);
 734
 735                 if (r)
 736                         goto out_free;
 737
 738                 if (!ent->has_leaf_count)
 739                         continue;
 740
 741                 limit = cpuid_entries[nent - 1].eax;
 742                 for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
 743                         r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
 744                                      &nent, cpuid->nent, type);
 745
 746                 if (r)
 747                         goto out_free;
 748         }
 749
 750         r = -EFAULT;
 751         if (copy_to_user(entries, cpuid_entries,
 752                          nent * sizeof(struct kvm_cpuid_entry2)))
 753                 goto out_free;
 754         cpuid->nent = nent;
 755         r = 0;
 756
 757 out_free:
 758         vfree(cpuid_entries);
 759 out:
 760         return r;
 761 }
 762
 763 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
 764 {
 765         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
 766         int j, nent = vcpu->arch.cpuid_nent;
 767
 768         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
 769         /* when no next entry is found, the current entry[i] is reselected */
 770         for (j = i + 1; ; j = (j + 1) % nent) {
 771                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
 772                 if (ej->function == e->function) {
 773                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
 774                         return j;
 775                 }
 776         }
 777         return 0; /* silence gcc, even though control never reaches here */
 778 }
 779
 780 /* find an entry with matching function, matching index (if needed), and that
 781  * should be read next (if it's stateful) */
 782 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
 783         u32 function, u32 index)
 784 {
 785         if (e->function != function)
 786                 return 0;
 787         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
 788                 return 0;
 789         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
 790             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
 791                 return 0;
 792         return 1;
 793 }
 794
 795 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
 796                                               u32 function, u32 index)
 797 {
 798         int i;
 799         struct kvm_cpuid_entry2 *best = NULL;
 800
 801         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 802                 struct kvm_cpuid_entry2 *e;
 803
 804                 e = &vcpu->arch.cpuid_entries[i];
 805                 if (is_matching_cpuid_entry(e, function, index)) {
 806                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
 807                                 move_to_next_stateful_cpuid_entry(vcpu, i);
 808                         best = e;
 809                         break;
 810                 }
 811         }
 812         return best;
 813 }
 814 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
 815
 816 /*
 817  * If no match is found, check whether we exceed the vCPU's limit
 818  * and return the content of the highest valid _standard_ leaf instead.
 819  * This is to satisfy the CPUID specification.
 820  */
 821 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
 822                                                   u32 function, u32 index)
 823 {
 824         struct kvm_cpuid_entry2 *maxlevel;
 825
 826         maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
 827         if (!maxlevel || maxlevel->eax >= function)
 828                 return NULL;
 829         if (function & 0x80000000) {
 830                 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
 831                 if (!maxlevel)
 832                         return NULL;
 833         }
 834         return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
 835 }
 836
 837 void kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
 838 {
 839         u32 function = *eax, index = *ecx;
 840         struct kvm_cpuid_entry2 *best;
 841
 842         best = kvm_find_cpuid_entry(vcpu, function, index);
 843
 844         if (!best)
 845                 best = check_cpuid_limit(vcpu, function, index);
 846
 847         /*
 848          * Perfmon not yet supported for L2 guest.
 849          */
 850         if (is_guest_mode(vcpu) && function == 0xa)
 851                 best = NULL;
 852
 853         if (best) {
 854                 *eax = best->eax;
 855                 *ebx = best->ebx;
 856                 *ecx = best->ecx;
 857                 *edx = best->edx;
 858         } else
 859                 *eax = *ebx = *ecx = *edx = 0;
 860         trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx);
 861 }
 862 EXPORT_SYMBOL_GPL(kvm_cpuid);
 863
 864 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
 865 {
 866         u32 function, eax, ebx, ecx, edx;
 867
 868         function = eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
 869         ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
 870         kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx);
 871         kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
 872         kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
 873         kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
 874         kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
 875         kvm_x86_ops->skip_emulated_instruction(vcpu);
 876 }
 877 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);