arch/x86/kvm/cpuid.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  * cpuid support routines
   4  *
   5  * derived from arch/x86/kvm/x86.c
   6  *
   7  * Copyright 2011 Red Hat, Inc. and/or its affiliates.
   8  * Copyright IBM Corporation, 2008
   9  *
  10  * This work is licensed under the terms of the GNU GPL, version 2.  See
  11  * the COPYING file in the top-level directory.
  12  *
  13  */
  14
  15 #include <linux/kvm_host.h>
  16 #include <linux/export.h>
  17 #include <linux/vmalloc.h>
  18 #include <linux/uaccess.h>
  19 #include <linux/sched/stat.h>
  20
  21 #include <asm/processor.h>
  22 #include <asm/user.h>
  23 #include <asm/fpu/xstate.h>
  24 #include "cpuid.h"
  25 #include "lapic.h"
  26 #include "mmu.h"
  27 #include "trace.h"
  28 #include "pmu.h"
  29
  30 static u32 xstate_required_size(u64 xstate_bv, bool compacted)
  31 {
  32         int feature_bit = 0;
  33         u32 ret = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
  34
  35         xstate_bv &= XFEATURE_MASK_EXTEND;
  36         while (xstate_bv) {
  37                 if (xstate_bv & 0x1) {
  38                         u32 eax, ebx, ecx, edx, offset;
  39                         cpuid_count(0xD, feature_bit, &eax, &ebx, &ecx, &edx);
  40                         offset = compacted ? ret : ebx;
  41                         ret = max(ret, offset + eax);
  42                 }
  43
  44                 xstate_bv >>= 1;
  45                 feature_bit++;
  46         }
  47
  48         return ret;
  49 }
  50
  51 bool kvm_mpx_supported(void)
  52 {
  53         return ((host_xcr0 & (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR))
  54                  && kvm_x86_ops->mpx_supported());
  55 }
  56 EXPORT_SYMBOL_GPL(kvm_mpx_supported);
  57
  58 u64 kvm_supported_xcr0(void)
  59 {
  60         u64 xcr0 = KVM_SUPPORTED_XCR0 & host_xcr0;
  61
  62         if (!kvm_mpx_supported())
  63                 xcr0 &= ~(XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR);
  64
  65         return xcr0;
  66 }
  67
  68 #define F(x) bit(X86_FEATURE_##x)
  69
  70 /* For scattered features from cpufeatures.h; we currently expose none */
  71 #define KF(x) bit(KVM_CPUID_BIT_##x)
  72
  73 int kvm_update_cpuid(struct kvm_vcpu *vcpu)
  74 {
  75         struct kvm_cpuid_entry2 *best;
  76         struct kvm_lapic *apic = vcpu->arch.apic;
  77
  78         best = kvm_find_cpuid_entry(vcpu, 1, 0);
  79         if (!best)
  80                 return 0;
  81
  82         /* Update OSXSAVE bit */
  83         if (boot_cpu_has(X86_FEATURE_XSAVE) && best->function == 0x1) {
  84                 best->ecx &= ~F(OSXSAVE);
  85                 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
  86                         best->ecx |= F(OSXSAVE);
  87         }
  88
  89         best->edx &= ~F(APIC);
  90         if (vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE)
  91                 best->edx |= F(APIC);
  92
  93         if (apic) {
  94                 if (best->ecx & F(TSC_DEADLINE_TIMER))
  95                         apic->lapic_timer.timer_mode_mask = 3 << 17;
  96                 else
  97                         apic->lapic_timer.timer_mode_mask = 1 << 17;
  98         }
  99
 100         best = kvm_find_cpuid_entry(vcpu, 7, 0);
 101         if (best) {
 102                 /* Update OSPKE bit */
 103                 if (boot_cpu_has(X86_FEATURE_PKU) && best->function == 0x7) {
 104                         best->ecx &= ~F(OSPKE);
 105                         if (kvm_read_cr4_bits(vcpu, X86_CR4_PKE))
 106                                 best->ecx |= F(OSPKE);
 107                 }
 108         }
 109
 110         best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
 111         if (!best) {
 112                 vcpu->arch.guest_supported_xcr0 = 0;
 113                 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
 114         } else {
 115                 vcpu->arch.guest_supported_xcr0 =
 116                         (best->eax | ((u64)best->edx << 32)) &
 117                         kvm_supported_xcr0();
 118                 vcpu->arch.guest_xstate_size = best->ebx =
 119                         xstate_required_size(vcpu->arch.xcr0, false);
 120         }
 121
 122         best = kvm_find_cpuid_entry(vcpu, 0xD, 1);
 123         if (best && (best->eax & (F(XSAVES) | F(XSAVEC))))
 124                 best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
 125
 126         /*
 127          * The existing code assumes virtual address is 48-bit or 57-bit in the
 128          * canonical address checks; exit if it is ever changed.
 129          */
 130         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
 131         if (best) {
 132                 int vaddr_bits = (best->eax & 0xff00) >> 8;
 133
 134                 if (vaddr_bits != 48 && vaddr_bits != 57 && vaddr_bits != 0)
 135                         return -EINVAL;
 136         }
 137
 138         /* Update physical-address width */
 139         vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
 140         kvm_mmu_reset_context(vcpu);
 141
 142         kvm_pmu_refresh(vcpu);
 143         return 0;
 144 }
 145
 146 static int is_efer_nx(void)
 147 {
 148         unsigned long long efer = 0;
 149
 150         rdmsrl_safe(MSR_EFER, &efer);
 151         return efer & EFER_NX;
 152 }
 153
 154 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
 155 {
 156         int i;
 157         struct kvm_cpuid_entry2 *e, *entry;
 158
 159         entry = NULL;
 160         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 161                 e = &vcpu->arch.cpuid_entries[i];
 162                 if (e->function == 0x80000001) {
 163                         entry = e;
 164                         break;
 165                 }
 166         }
 167         if (entry && (entry->edx & F(NX)) && !is_efer_nx()) {
 168                 entry->edx &= ~F(NX);
 169                 printk(KERN_INFO "kvm: guest NX capability removed\n");
 170         }
 171 }
 172
 173 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
 174 {
 175         struct kvm_cpuid_entry2 *best;
 176
 177         best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
 178         if (!best || best->eax < 0x80000008)
 179                 goto not_found;
 180         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
 181         if (best)
 182                 return best->eax & 0xff;
 183 not_found:
 184         return 36;
 185 }
 186 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr);
 187
 188 /* when an old userspace process fills a new kernel module */
 189 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
 190                              struct kvm_cpuid *cpuid,
 191                              struct kvm_cpuid_entry __user *entries)
 192 {
 193         int r, i;
 194         struct kvm_cpuid_entry *cpuid_entries = NULL;
 195
 196         r = -E2BIG;
 197         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 198                 goto out;
 199         r = -ENOMEM;
 200         if (cpuid->nent) {
 201                 cpuid_entries = vmalloc(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
 335         /* cpuid 1.edx */
 336         const u32 kvm_cpuid_1_edx_x86_features =
 337                 F(FPU) | F(VME) | F(DE) | F(PSE) |
 338                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
 339                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
 340                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
 341                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
 342                 0 /* Reserved, DS, ACPI */ | F(MMX) |
 343                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
 344                 0 /* HTT, TM, Reserved, PBE */;
 345         /* cpuid 0x80000001.edx */
 346         const u32 kvm_cpuid_8000_0001_edx_x86_features =
 347                 F(FPU) | F(VME) | F(DE) | F(PSE) |
 348                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
 349                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
 350                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
 351                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
 352                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
 353                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
 354                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
 355         /* cpuid 1.ecx */
 356         const u32 kvm_cpuid_1_ecx_x86_features =
 357                 /* NOTE: MONITOR (and MWAIT) are emulated as NOP,
 358                  * but *not* advertised to guests via CPUID ! */
 359                 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
 360                 0 /* DS-CPL, VMX, SMX, EST */ |
 361                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
 362                 F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
 363                 F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
 364                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
 365                 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
 366                 F(F16C) | F(RDRAND);
 367         /* cpuid 0x80000001.ecx */
 368         const u32 kvm_cpuid_8000_0001_ecx_x86_features =
 369                 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
 370                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
 371                 F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
 372                 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM) |
 373                 F(TOPOEXT);
 374
 375         /* cpuid 0x80000008.ebx */
 376         const u32 kvm_cpuid_8000_0008_ebx_x86_features =
 377                 F(IBPB) | F(IBRS);
 378
 379         /* cpuid 0xC0000001.edx */
 380         const u32 kvm_cpuid_C000_0001_edx_x86_features =
 381                 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
 382                 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
 383                 F(PMM) | F(PMM_EN);
 384
 385         /* cpuid 7.0.ebx */
 386         const u32 kvm_cpuid_7_0_ebx_x86_features =
 387                 F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
 388                 F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
 389                 F(ADX) | F(SMAP) | F(AVX512IFMA) | F(AVX512F) | F(AVX512PF) |
 390                 F(AVX512ER) | F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) |
 391                 F(SHA_NI) | F(AVX512BW) | F(AVX512VL);
 392
 393         /* cpuid 0xD.1.eax */
 394         const u32 kvm_cpuid_D_1_eax_x86_features =
 395                 F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves;
 396
 397         /* cpuid 7.0.ecx*/
 398         const u32 kvm_cpuid_7_0_ecx_x86_features =
 399                 F(AVX512VBMI) | F(LA57) | F(PKU) | 0 /*OSPKE*/ |
 400                 F(AVX512_VPOPCNTDQ) | F(UMIP) | F(AVX512_VBMI2) | F(GFNI) |
 401                 F(VAES) | F(VPCLMULQDQ) | F(AVX512_VNNI) | F(AVX512_BITALG);
 402
 403         /* cpuid 7.0.edx*/
 404         const u32 kvm_cpuid_7_0_edx_x86_features =
 405                 F(AVX512_4VNNIW) | F(AVX512_4FMAPS) | F(SPEC_CTRL) |
 406                 F(ARCH_CAPABILITIES);
 407
 408         /* all calls to cpuid_count() should be made on the same cpu */
 409         get_cpu();
 410
 411         r = -E2BIG;
 412
 413         if (*nent >= maxnent)
 414                 goto out;
 415
 416         do_cpuid_1_ent(entry, function, index);
 417         ++*nent;
 418
 419         switch (function) {
 420         case 0:
 421                 entry->eax = min(entry->eax, (u32)0xd);
 422                 break;
 423         case 1:
 424                 entry->edx &= kvm_cpuid_1_edx_x86_features;
 425                 cpuid_mask(&entry->edx, CPUID_1_EDX);
 426                 entry->ecx &= kvm_cpuid_1_ecx_x86_features;
 427                 cpuid_mask(&entry->ecx, CPUID_1_ECX);
 428                 /* we support x2apic emulation even if host does not support
 429                  * it since we emulate x2apic in software */
 430                 entry->ecx |= F(X2APIC);
 431                 break;
 432         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
 433          * may return different values. This forces us to get_cpu() before
 434          * issuing the first command, and also to emulate this annoying behavior
 435          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
 436         case 2: {
 437                 int t, times = entry->eax & 0xff;
 438
 439                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 440                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
 441                 for (t = 1; t < times; ++t) {
 442                         if (*nent >= maxnent)
 443                                 goto out;
 444
 445                         do_cpuid_1_ent(&entry[t], function, 0);
 446                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 447                         ++*nent;
 448                 }
 449                 break;
 450         }
 451         /* function 4 has additional index. */
 452         case 4: {
 453                 int i, cache_type;
 454
 455                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 456                 /* read more entries until cache_type is zero */
 457                 for (i = 1; ; ++i) {
 458                         if (*nent >= maxnent)
 459                                 goto out;
 460
 461                         cache_type = entry[i - 1].eax & 0x1f;
 462                         if (!cache_type)
 463                                 break;
 464                         do_cpuid_1_ent(&entry[i], function, i);
 465                         entry[i].flags |=
 466                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 467                         ++*nent;
 468                 }
 469                 break;
 470         }
 471         case 6: /* Thermal management */
 472                 entry->eax = 0x4; /* allow ARAT */
 473                 entry->ebx = 0;
 474                 entry->ecx = 0;
 475                 entry->edx = 0;
 476                 break;
 477         case 7: {
 478                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 479                 /* Mask ebx against host capability word 9 */
 480                 if (index == 0) {
 481                         entry->ebx &= kvm_cpuid_7_0_ebx_x86_features;
 482                         cpuid_mask(&entry->ebx, CPUID_7_0_EBX);
 483                         // TSC_ADJUST is emulated
 484                         entry->ebx |= F(TSC_ADJUST);
 485                         entry->ecx &= kvm_cpuid_7_0_ecx_x86_features;
 486                         cpuid_mask(&entry->ecx, CPUID_7_ECX);
 487                         entry->ecx |= f_umip;
 488                         /* PKU is not yet implemented for shadow paging. */
 489                         if (!tdp_enabled || !boot_cpu_has(X86_FEATURE_OSPKE))
 490                                 entry->ecx &= ~F(PKU);
 491                         entry->edx &= kvm_cpuid_7_0_edx_x86_features;
 492                         cpuid_mask(&entry->edx, CPUID_7_EDX);
 493                 } else {
 494                         entry->ebx = 0;
 495                         entry->ecx = 0;
 496                         entry->edx = 0;
 497                 }
 498                 entry->eax = 0;
 499                 break;
 500         }
 501         case 9:
 502                 break;
 503         case 0xa: { /* Architectural Performance Monitoring */
 504                 struct x86_pmu_capability cap;
 505                 union cpuid10_eax eax;
 506                 union cpuid10_edx edx;
 507
 508                 perf_get_x86_pmu_capability(&cap);
 509
 510                 /*
 511                  * Only support guest architectural pmu on a host
 512                  * with architectural pmu.
 513                  */
 514                 if (!cap.version)
 515                         memset(&cap, 0, sizeof(cap));
 516
 517                 eax.split.version_id = min(cap.version, 2);
 518                 eax.split.num_counters = cap.num_counters_gp;
 519                 eax.split.bit_width = cap.bit_width_gp;
 520                 eax.split.mask_length = cap.events_mask_len;
 521
 522                 edx.split.num_counters_fixed = cap.num_counters_fixed;
 523                 edx.split.bit_width_fixed = cap.bit_width_fixed;
 524                 edx.split.reserved = 0;
 525
 526                 entry->eax = eax.full;
 527                 entry->ebx = cap.events_mask;
 528                 entry->ecx = 0;
 529                 entry->edx = edx.full;
 530                 break;
 531         }
 532         /* function 0xb has additional index. */
 533         case 0xb: {
 534                 int i, level_type;
 535
 536                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 537                 /* read more entries until level_type is zero */
 538                 for (i = 1; ; ++i) {
 539                         if (*nent >= maxnent)
 540                                 goto out;
 541
 542                         level_type = entry[i - 1].ecx & 0xff00;
 543                         if (!level_type)
 544                                 break;
 545                         do_cpuid_1_ent(&entry[i], function, i);
 546                         entry[i].flags |=
 547                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 548                         ++*nent;
 549                 }
 550                 break;
 551         }
 552         case 0xd: {
 553                 int idx, i;
 554                 u64 supported = kvm_supported_xcr0();
 555
 556                 entry->eax &= supported;
 557                 entry->ebx = xstate_required_size(supported, false);
 558                 entry->ecx = entry->ebx;
 559                 entry->edx &= supported >> 32;
 560                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 561                 if (!supported)
 562                         break;
 563
 564                 for (idx = 1, i = 1; idx < 64; ++idx) {
 565                         u64 mask = ((u64)1 << idx);
 566                         if (*nent >= maxnent)
 567                                 goto out;
 568
 569                         do_cpuid_1_ent(&entry[i], function, idx);
 570                         if (idx == 1) {
 571                                 entry[i].eax &= kvm_cpuid_D_1_eax_x86_features;
 572                                 cpuid_mask(&entry[i].eax, CPUID_D_1_EAX);
 573                                 entry[i].ebx = 0;
 574                                 if (entry[i].eax & (F(XSAVES)|F(XSAVEC)))
 575                                         entry[i].ebx =
 576                                                 xstate_required_size(supported,
 577                                                                      true);
 578                         } else {
 579                                 if (entry[i].eax == 0 || !(supported & mask))
 580                                         continue;
 581                                 if (WARN_ON_ONCE(entry[i].ecx & 1))
 582                                         continue;
 583                         }
 584                         entry[i].ecx = 0;
 585                         entry[i].edx = 0;
 586                         entry[i].flags |=
 587                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 588                         ++*nent;
 589                         ++i;
 590                 }
 591                 break;
 592         }
 593         case KVM_CPUID_SIGNATURE: {
 594                 static const char signature[12] = "KVMKVMKVM\0\0";
 595                 const u32 *sigptr = (const u32 *)signature;
 596                 entry->eax = KVM_CPUID_FEATURES;
 597                 entry->ebx = sigptr[0];
 598                 entry->ecx = sigptr[1];
 599                 entry->edx = sigptr[2];
 600                 break;
 601         }
 602         case KVM_CPUID_FEATURES:
 603                 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
 604                              (1 << KVM_FEATURE_NOP_IO_DELAY) |
 605                              (1 << KVM_FEATURE_CLOCKSOURCE2) |
 606                              (1 << KVM_FEATURE_ASYNC_PF) |
 607                              (1 << KVM_FEATURE_PV_EOI) |
 608                              (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
 609                              (1 << KVM_FEATURE_PV_UNHALT) |
 610                              (1 << KVM_FEATURE_PV_TLB_FLUSH) |
 611                              (1 << KVM_FEATURE_ASYNC_PF_VMEXIT);
 612
 613                 if (sched_info_on())
 614                         entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
 615
 616                 entry->ebx = 0;
 617                 entry->ecx = 0;
 618                 entry->edx = 0;
 619                 break;
 620         case 0x80000000:
 621                 entry->eax = min(entry->eax, 0x8000001f);
 622                 break;
 623         case 0x80000001:
 624                 entry->edx &= kvm_cpuid_8000_0001_edx_x86_features;
 625                 cpuid_mask(&entry->edx, CPUID_8000_0001_EDX);
 626                 entry->ecx &= kvm_cpuid_8000_0001_ecx_x86_features;
 627                 cpuid_mask(&entry->ecx, CPUID_8000_0001_ECX);
 628                 break;
 629         case 0x80000007: /* Advanced power management */
 630                 /* invariant TSC is CPUID.80000007H:EDX[8] */
 631                 entry->edx &= (1 << 8);
 632                 /* mask against host */
 633                 entry->edx &= boot_cpu_data.x86_power;
 634                 entry->eax = entry->ebx = entry->ecx = 0;
 635                 break;
 636         case 0x80000008: {
 637                 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
 638                 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
 639                 unsigned phys_as = entry->eax & 0xff;
 640
 641                 if (!g_phys_as)
 642                         g_phys_as = phys_as;
 643                 entry->eax = g_phys_as | (virt_as << 8);
 644                 entry->edx = 0;
 645                 /* IBRS and IBPB aren't necessarily present in hardware cpuid */
 646                 if (boot_cpu_has(X86_FEATURE_IBPB))
 647                         entry->ebx |= F(IBPB);
 648                 if (boot_cpu_has(X86_FEATURE_IBRS))
 649                         entry->ebx |= F(IBRS);
 650                 entry->ebx &= kvm_cpuid_8000_0008_ebx_x86_features;
 651                 cpuid_mask(&entry->ebx, CPUID_8000_0008_EBX);
 652                 break;
 653         }
 654         case 0x80000019:
 655                 entry->ecx = entry->edx = 0;
 656                 break;
 657         case 0x8000001a:
 658                 break;
 659         case 0x8000001d:
 660                 break;
 661         /*Add support for Centaur's CPUID instruction*/
 662         case 0xC0000000:
 663                 /*Just support up to 0xC0000004 now*/
 664                 entry->eax = min(entry->eax, 0xC0000004);
 665                 break;
 666         case 0xC0000001:
 667                 entry->edx &= kvm_cpuid_C000_0001_edx_x86_features;
 668                 cpuid_mask(&entry->edx, CPUID_C000_0001_EDX);
 669                 break;
 670         case 3: /* Processor serial number */
 671         case 5: /* MONITOR/MWAIT */
 672         case 0xC0000002:
 673         case 0xC0000003:
 674         case 0xC0000004:
 675         default:
 676                 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
 677                 break;
 678         }
 679
 680         kvm_x86_ops->set_supported_cpuid(function, entry);
 681
 682         r = 0;
 683
 684 out:
 685         put_cpu();
 686
 687         return r;
 688 }
 689
 690 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
 691                         u32 idx, int *nent, int maxnent, unsigned int type)
 692 {
 693         if (type == KVM_GET_EMULATED_CPUID)
 694                 return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
 695
 696         return __do_cpuid_ent(entry, func, idx, nent, maxnent);
 697 }
 698
 699 #undef F
 700
 701 struct kvm_cpuid_param {
 702         u32 func;
 703         u32 idx;
 704         bool has_leaf_count;
 705         bool (*qualifier)(const struct kvm_cpuid_param *param);
 706 };
 707
 708 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
 709 {
 710         return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
 711 }
 712
 713 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
 714                                  __u32 num_entries, unsigned int ioctl_type)
 715 {
 716         int i;
 717         __u32 pad[3];
 718
 719         if (ioctl_type != KVM_GET_EMULATED_CPUID)
 720                 return false;
 721
 722         /*
 723          * We want to make sure that ->padding is being passed clean from
 724          * userspace in case we want to use it for something in the future.
 725          *
 726          * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
 727          * have to give ourselves satisfied only with the emulated side. /me
 728          * sheds a tear.
 729          */
 730         for (i = 0; i < num_entries; i++) {
 731                 if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
 732                         return true;
 733
 734                 if (pad[0] || pad[1] || pad[2])
 735                         return true;
 736         }
 737         return false;
 738 }
 739
 740 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
 741                             struct kvm_cpuid_entry2 __user *entries,
 742                             unsigned int type)
 743 {
 744         struct kvm_cpuid_entry2 *cpuid_entries;
 745         int limit, nent = 0, r = -E2BIG, i;
 746         u32 func;
 747         static const struct kvm_cpuid_param param[] = {
 748                 { .func = 0, .has_leaf_count = true },
 749                 { .func = 0x80000000, .has_leaf_count = true },
 750                 { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
 751                 { .func = KVM_CPUID_SIGNATURE },
 752                 { .func = KVM_CPUID_FEATURES },
 753         };
 754
 755         if (cpuid->nent < 1)
 756                 goto out;
 757         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 758                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
 759
 760         if (sanity_check_entries(entries, cpuid->nent, type))
 761                 return -EINVAL;
 762
 763         r = -ENOMEM;
 764         cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
 765         if (!cpuid_entries)
 766                 goto out;
 767
 768         r = 0;
 769         for (i = 0; i < ARRAY_SIZE(param); i++) {
 770                 const struct kvm_cpuid_param *ent = &param[i];
 771
 772                 if (ent->qualifier && !ent->qualifier(ent))
 773                         continue;
 774
 775                 r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
 776                                 &nent, cpuid->nent, type);
 777
 778                 if (r)
 779                         goto out_free;
 780
 781                 if (!ent->has_leaf_count)
 782                         continue;
 783
 784                 limit = cpuid_entries[nent - 1].eax;
 785                 for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
 786                         r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
 787                                      &nent, cpuid->nent, type);
 788
 789                 if (r)
 790                         goto out_free;
 791         }
 792
 793         r = -EFAULT;
 794         if (copy_to_user(entries, cpuid_entries,
 795                          nent * sizeof(struct kvm_cpuid_entry2)))
 796                 goto out_free;
 797         cpuid->nent = nent;
 798         r = 0;
 799
 800 out_free:
 801         vfree(cpuid_entries);
 802 out:
 803         return r;
 804 }
 805
 806 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
 807 {
 808         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
 809         struct kvm_cpuid_entry2 *ej;
 810         int j = i;
 811         int nent = vcpu->arch.cpuid_nent;
 812
 813         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
 814         /* when no next entry is found, the current entry[i] is reselected */
 815         do {
 816                 j = (j + 1) % nent;
 817                 ej = &vcpu->arch.cpuid_entries[j];
 818         } while (ej->function != e->function);
 819
 820         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
 821
 822         return j;
 823 }
 824
 825 /* find an entry with matching function, matching index (if needed), and that
 826  * should be read next (if it's stateful) */
 827 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
 828         u32 function, u32 index)
 829 {
 830         if (e->function != function)
 831                 return 0;
 832         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
 833                 return 0;
 834         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
 835             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
 836                 return 0;
 837         return 1;
 838 }
 839
 840 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
 841                                               u32 function, u32 index)
 842 {
 843         int i;
 844         struct kvm_cpuid_entry2 *best = NULL;
 845
 846         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 847                 struct kvm_cpuid_entry2 *e;
 848
 849                 e = &vcpu->arch.cpuid_entries[i];
 850                 if (is_matching_cpuid_entry(e, function, index)) {
 851                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
 852                                 move_to_next_stateful_cpuid_entry(vcpu, i);
 853                         best = e;
 854                         break;
 855                 }
 856         }
 857         return best;
 858 }
 859 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
 860
 861 /*
 862  * If no match is found, check whether we exceed the vCPU's limit
 863  * and return the content of the highest valid _standard_ leaf instead.
 864  * This is to satisfy the CPUID specification.
 865  */
 866 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
 867                                                   u32 function, u32 index)
 868 {
 869         struct kvm_cpuid_entry2 *maxlevel;
 870
 871         maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
 872         if (!maxlevel || maxlevel->eax >= function)
 873                 return NULL;
 874         if (function & 0x80000000) {
 875                 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
 876                 if (!maxlevel)
 877                         return NULL;
 878         }
 879         return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
 880 }
 881
 882 bool kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx,
 883                u32 *ecx, u32 *edx, bool check_limit)
 884 {
 885         u32 function = *eax, index = *ecx;
 886         struct kvm_cpuid_entry2 *best;
 887         bool entry_found = true;
 888
 889         best = kvm_find_cpuid_entry(vcpu, function, index);
 890
 891         if (!best) {
 892                 entry_found = false;
 893                 if (!check_limit)
 894                         goto out;
 895
 896                 best = check_cpuid_limit(vcpu, function, index);
 897         }
 898
 899 out:
 900         if (best) {
 901                 *eax = best->eax;
 902                 *ebx = best->ebx;
 903                 *ecx = best->ecx;
 904                 *edx = best->edx;
 905         } else
 906                 *eax = *ebx = *ecx = *edx = 0;
 907         trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx, entry_found);
 908         return entry_found;
 909 }
 910 EXPORT_SYMBOL_GPL(kvm_cpuid);
 911
 912 int kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
 913 {
 914         u32 eax, ebx, ecx, edx;
 915
 916         if (cpuid_fault_enabled(vcpu) && !kvm_require_cpl(vcpu, 0))
 917                 return 1;
 918
 919         eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
 920         ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
 921         kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx, true);
 922         kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
 923         kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
 924         kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
 925         kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
 926         return kvm_skip_emulated_instruction(vcpu);
 927 }
 928 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);