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 /* These are scattered features in cpufeatures.h. */
  71 #define KVM_CPUID_BIT_AVX512_4VNNIW     2
  72 #define KVM_CPUID_BIT_AVX512_4FMAPS     3
  73 #define KF(x) bit(KVM_CPUID_BIT_##x)
  74
  75 int kvm_update_cpuid(struct kvm_vcpu *vcpu)
  76 {
  77         struct kvm_cpuid_entry2 *best;
  78         struct kvm_lapic *apic = vcpu->arch.apic;
  79
  80         best = kvm_find_cpuid_entry(vcpu, 1, 0);
  81         if (!best)
  82                 return 0;
  83
  84         /* Update OSXSAVE bit */
  85         if (boot_cpu_has(X86_FEATURE_XSAVE) && best->function == 0x1) {
  86                 best->ecx &= ~F(OSXSAVE);
  87                 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
  88                         best->ecx |= F(OSXSAVE);
  89         }
  90
  91         best->edx &= ~F(APIC);
  92         if (vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE)
  93                 best->edx |= F(APIC);
  94
  95         if (apic) {
  96                 if (best->ecx & F(TSC_DEADLINE_TIMER))
  97                         apic->lapic_timer.timer_mode_mask = 3 << 17;
  98                 else
  99                         apic->lapic_timer.timer_mode_mask = 1 << 17;
 100         }
 101
 102         best = kvm_find_cpuid_entry(vcpu, 7, 0);
 103         if (best) {
 104                 /* Update OSPKE bit */
 105                 if (boot_cpu_has(X86_FEATURE_PKU) && best->function == 0x7) {
 106                         best->ecx &= ~F(OSPKE);
 107                         if (kvm_read_cr4_bits(vcpu, X86_CR4_PKE))
 108                                 best->ecx |= F(OSPKE);
 109                 }
 110         }
 111
 112         best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
 113         if (!best) {
 114                 vcpu->arch.guest_supported_xcr0 = 0;
 115                 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
 116         } else {
 117                 vcpu->arch.guest_supported_xcr0 =
 118                         (best->eax | ((u64)best->edx << 32)) &
 119                         kvm_supported_xcr0();
 120                 vcpu->arch.guest_xstate_size = best->ebx =
 121                         xstate_required_size(vcpu->arch.xcr0, false);
 122         }
 123
 124         best = kvm_find_cpuid_entry(vcpu, 0xD, 1);
 125         if (best && (best->eax & (F(XSAVES) | F(XSAVEC))))
 126                 best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
 127
 128         /*
 129          * The existing code assumes virtual address is 48-bit or 57-bit in the
 130          * canonical address checks; exit if it is ever changed.
 131          */
 132         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
 133         if (best) {
 134                 int vaddr_bits = (best->eax & 0xff00) >> 8;
 135
 136                 if (vaddr_bits != 48 && vaddr_bits != 57 && vaddr_bits != 0)
 137                         return -EINVAL;
 138         }
 139
 140         /* Update physical-address width */
 141         vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
 142         kvm_mmu_reset_context(vcpu);
 143
 144         kvm_pmu_refresh(vcpu);
 145         return 0;
 146 }
 147
 148 static int is_efer_nx(void)
 149 {
 150         unsigned long long efer = 0;
 151
 152         rdmsrl_safe(MSR_EFER, &efer);
 153         return efer & EFER_NX;
 154 }
 155
 156 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
 157 {
 158         int i;
 159         struct kvm_cpuid_entry2 *e, *entry;
 160
 161         entry = NULL;
 162         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 163                 e = &vcpu->arch.cpuid_entries[i];
 164                 if (e->function == 0x80000001) {
 165                         entry = e;
 166                         break;
 167                 }
 168         }
 169         if (entry && (entry->edx & F(NX)) && !is_efer_nx()) {
 170                 entry->edx &= ~F(NX);
 171                 printk(KERN_INFO "kvm: guest NX capability removed\n");
 172         }
 173 }
 174
 175 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
 176 {
 177         struct kvm_cpuid_entry2 *best;
 178
 179         best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
 180         if (!best || best->eax < 0x80000008)
 181                 goto not_found;
 182         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
 183         if (best)
 184                 return best->eax & 0xff;
 185 not_found:
 186         return 36;
 187 }
 188 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr);
 189
 190 /* when an old userspace process fills a new kernel module */
 191 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
 192                              struct kvm_cpuid *cpuid,
 193                              struct kvm_cpuid_entry __user *entries)
 194 {
 195         int r, i;
 196         struct kvm_cpuid_entry *cpuid_entries = NULL;
 197
 198         r = -E2BIG;
 199         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 200                 goto out;
 201         r = -ENOMEM;
 202         if (cpuid->nent) {
 203                 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) *
 204                                         cpuid->nent);
 205                 if (!cpuid_entries)
 206                         goto out;
 207                 r = -EFAULT;
 208                 if (copy_from_user(cpuid_entries, entries,
 209                                    cpuid->nent * sizeof(struct kvm_cpuid_entry)))
 210                         goto out;
 211         }
 212         for (i = 0; i < cpuid->nent; i++) {
 213                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
 214                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
 215                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
 216                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
 217                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
 218                 vcpu->arch.cpuid_entries[i].index = 0;
 219                 vcpu->arch.cpuid_entries[i].flags = 0;
 220                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
 221                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
 222                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
 223         }
 224         vcpu->arch.cpuid_nent = cpuid->nent;
 225         cpuid_fix_nx_cap(vcpu);
 226         kvm_apic_set_version(vcpu);
 227         kvm_x86_ops->cpuid_update(vcpu);
 228         r = kvm_update_cpuid(vcpu);
 229
 230 out:
 231         vfree(cpuid_entries);
 232         return r;
 233 }
 234
 235 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
 236                               struct kvm_cpuid2 *cpuid,
 237                               struct kvm_cpuid_entry2 __user *entries)
 238 {
 239         int r;
 240
 241         r = -E2BIG;
 242         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 243                 goto out;
 244         r = -EFAULT;
 245         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
 246                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
 247                 goto out;
 248         vcpu->arch.cpuid_nent = cpuid->nent;
 249         kvm_apic_set_version(vcpu);
 250         kvm_x86_ops->cpuid_update(vcpu);
 251         r = kvm_update_cpuid(vcpu);
 252 out:
 253         return r;
 254 }
 255
 256 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
 257                               struct kvm_cpuid2 *cpuid,
 258                               struct kvm_cpuid_entry2 __user *entries)
 259 {
 260         int r;
 261
 262         r = -E2BIG;
 263         if (cpuid->nent < vcpu->arch.cpuid_nent)
 264                 goto out;
 265         r = -EFAULT;
 266         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
 267                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
 268                 goto out;
 269         return 0;
 270
 271 out:
 272         cpuid->nent = vcpu->arch.cpuid_nent;
 273         return r;
 274 }
 275
 276 static void cpuid_mask(u32 *word, int wordnum)
 277 {
 278         *word &= boot_cpu_data.x86_capability[wordnum];
 279 }
 280
 281 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 282                            u32 index)
 283 {
 284         entry->function = function;
 285         entry->index = index;
 286         cpuid_count(entry->function, entry->index,
 287                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
 288         entry->flags = 0;
 289 }
 290
 291 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry,
 292                                    u32 func, u32 index, int *nent, int maxnent)
 293 {
 294         switch (func) {
 295         case 0:
 296                 entry->eax = 1;         /* only one leaf currently */
 297                 ++*nent;
 298                 break;
 299         case 1:
 300                 entry->ecx = F(MOVBE);
 301                 ++*nent;
 302                 break;
 303         default:
 304                 break;
 305         }
 306
 307         entry->function = func;
 308         entry->index = index;
 309
 310         return 0;
 311 }
 312
 313 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 314                                  u32 index, int *nent, int maxnent)
 315 {
 316         int r;
 317         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
 318 #ifdef CONFIG_X86_64
 319         unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
 320                                 ? F(GBPAGES) : 0;
 321         unsigned f_lm = F(LM);
 322 #else
 323         unsigned f_gbpages = 0;
 324         unsigned f_lm = 0;
 325 #endif
 326         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
 327         unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
 328         unsigned f_mpx = kvm_mpx_supported() ? F(MPX) : 0;
 329         unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0;
 330         unsigned f_umip = kvm_x86_ops->umip_emulated() ? F(UMIP) : 0;
 331
 332         /* cpuid 1.edx */
 333         const u32 kvm_cpuid_1_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(SEP) |
 337                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
 338                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
 339                 0 /* Reserved, DS, ACPI */ | F(MMX) |
 340                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
 341                 0 /* HTT, TM, Reserved, PBE */;
 342         /* cpuid 0x80000001.edx */
 343         const u32 kvm_cpuid_8000_0001_edx_x86_features =
 344                 F(FPU) | F(VME) | F(DE) | F(PSE) |
 345                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
 346                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
 347                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
 348                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
 349                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
 350                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
 351                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
 352         /* cpuid 1.ecx */
 353         const u32 kvm_cpuid_1_ecx_x86_features =
 354                 /* NOTE: MONITOR (and MWAIT) are emulated as NOP,
 355                  * but *not* advertised to guests via CPUID ! */
 356                 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
 357                 0 /* DS-CPL, VMX, SMX, EST */ |
 358                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
 359                 F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
 360                 F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
 361                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
 362                 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
 363                 F(F16C) | F(RDRAND);
 364         /* cpuid 0x80000001.ecx */
 365         const u32 kvm_cpuid_8000_0001_ecx_x86_features =
 366                 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
 367                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
 368                 F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
 369                 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
 370
 371         /* cpuid 0xC0000001.edx */
 372         const u32 kvm_cpuid_C000_0001_edx_x86_features =
 373                 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
 374                 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
 375                 F(PMM) | F(PMM_EN);
 376
 377         /* cpuid 7.0.ebx */
 378         const u32 kvm_cpuid_7_0_ebx_x86_features =
 379                 F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
 380                 F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
 381                 F(ADX) | F(SMAP) | F(AVX512IFMA) | F(AVX512F) | F(AVX512PF) |
 382                 F(AVX512ER) | F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) |
 383                 F(SHA_NI) | F(AVX512BW) | F(AVX512VL);
 384
 385         /* cpuid 0xD.1.eax */
 386         const u32 kvm_cpuid_D_1_eax_x86_features =
 387                 F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves;
 388
 389         /* cpuid 7.0.ecx*/
 390         const u32 kvm_cpuid_7_0_ecx_x86_features =
 391                 F(AVX512VBMI) | F(LA57) | F(PKU) | 0 /*OSPKE*/ |
 392                 F(AVX512_VPOPCNTDQ) | F(UMIP);
 393
 394         /* cpuid 7.0.edx*/
 395         const u32 kvm_cpuid_7_0_edx_x86_features =
 396                 KF(AVX512_4VNNIW) | KF(AVX512_4FMAPS);
 397
 398         /* all calls to cpuid_count() should be made on the same cpu */
 399         get_cpu();
 400
 401         r = -E2BIG;
 402
 403         if (*nent >= maxnent)
 404                 goto out;
 405
 406         do_cpuid_1_ent(entry, function, index);
 407         ++*nent;
 408
 409         switch (function) {
 410         case 0:
 411                 entry->eax = min(entry->eax, (u32)0xd);
 412                 break;
 413         case 1:
 414                 entry->edx &= kvm_cpuid_1_edx_x86_features;
 415                 cpuid_mask(&entry->edx, CPUID_1_EDX);
 416                 entry->ecx &= kvm_cpuid_1_ecx_x86_features;
 417                 cpuid_mask(&entry->ecx, CPUID_1_ECX);
 418                 /* we support x2apic emulation even if host does not support
 419                  * it since we emulate x2apic in software */
 420                 entry->ecx |= F(X2APIC);
 421                 break;
 422         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
 423          * may return different values. This forces us to get_cpu() before
 424          * issuing the first command, and also to emulate this annoying behavior
 425          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
 426         case 2: {
 427                 int t, times = entry->eax & 0xff;
 428
 429                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 430                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
 431                 for (t = 1; t < times; ++t) {
 432                         if (*nent >= maxnent)
 433                                 goto out;
 434
 435                         do_cpuid_1_ent(&entry[t], function, 0);
 436                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 437                         ++*nent;
 438                 }
 439                 break;
 440         }
 441         /* function 4 has additional index. */
 442         case 4: {
 443                 int i, cache_type;
 444
 445                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 446                 /* read more entries until cache_type is zero */
 447                 for (i = 1; ; ++i) {
 448                         if (*nent >= maxnent)
 449                                 goto out;
 450
 451                         cache_type = entry[i - 1].eax & 0x1f;
 452                         if (!cache_type)
 453                                 break;
 454                         do_cpuid_1_ent(&entry[i], function, i);
 455                         entry[i].flags |=
 456                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 457                         ++*nent;
 458                 }
 459                 break;
 460         }
 461         case 6: /* Thermal management */
 462                 entry->eax = 0x4; /* allow ARAT */
 463                 entry->ebx = 0;
 464                 entry->ecx = 0;
 465                 entry->edx = 0;
 466                 break;
 467         case 7: {
 468                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 469                 /* Mask ebx against host capability word 9 */
 470                 if (index == 0) {
 471                         entry->ebx &= kvm_cpuid_7_0_ebx_x86_features;
 472                         cpuid_mask(&entry->ebx, CPUID_7_0_EBX);
 473                         // TSC_ADJUST is emulated
 474                         entry->ebx |= F(TSC_ADJUST);
 475                         entry->ecx &= kvm_cpuid_7_0_ecx_x86_features;
 476                         cpuid_mask(&entry->ecx, CPUID_7_ECX);
 477                         entry->ecx |= f_umip;
 478                         /* PKU is not yet implemented for shadow paging. */
 479                         if (!tdp_enabled || !boot_cpu_has(X86_FEATURE_OSPKE))
 480                                 entry->ecx &= ~F(PKU);
 481                         entry->edx &= kvm_cpuid_7_0_edx_x86_features;
 482                         entry->edx &= get_scattered_cpuid_leaf(7, 0, CPUID_EDX);
 483                 } else {
 484                         entry->ebx = 0;
 485                         entry->ecx = 0;
 486                         entry->edx = 0;
 487                 }
 488                 entry->eax = 0;
 489                 break;
 490         }
 491         case 9:
 492                 break;
 493         case 0xa: { /* Architectural Performance Monitoring */
 494                 struct x86_pmu_capability cap;
 495                 union cpuid10_eax eax;
 496                 union cpuid10_edx edx;
 497
 498                 perf_get_x86_pmu_capability(&cap);
 499
 500                 /*
 501                  * Only support guest architectural pmu on a host
 502                  * with architectural pmu.
 503                  */
 504                 if (!cap.version)
 505                         memset(&cap, 0, sizeof(cap));
 506
 507                 eax.split.version_id = min(cap.version, 2);
 508                 eax.split.num_counters = cap.num_counters_gp;
 509                 eax.split.bit_width = cap.bit_width_gp;
 510                 eax.split.mask_length = cap.events_mask_len;
 511
 512                 edx.split.num_counters_fixed = cap.num_counters_fixed;
 513                 edx.split.bit_width_fixed = cap.bit_width_fixed;
 514                 edx.split.reserved = 0;
 515
 516                 entry->eax = eax.full;
 517                 entry->ebx = cap.events_mask;
 518                 entry->ecx = 0;
 519                 entry->edx = edx.full;
 520                 break;
 521         }
 522         /* function 0xb has additional index. */
 523         case 0xb: {
 524                 int i, level_type;
 525
 526                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 527                 /* read more entries until level_type is zero */
 528                 for (i = 1; ; ++i) {
 529                         if (*nent >= maxnent)
 530                                 goto out;
 531
 532                         level_type = entry[i - 1].ecx & 0xff00;
 533                         if (!level_type)
 534                                 break;
 535                         do_cpuid_1_ent(&entry[i], function, i);
 536                         entry[i].flags |=
 537                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 538                         ++*nent;
 539                 }
 540                 break;
 541         }
 542         case 0xd: {
 543                 int idx, i;
 544                 u64 supported = kvm_supported_xcr0();
 545
 546                 entry->eax &= supported;
 547                 entry->ebx = xstate_required_size(supported, false);
 548                 entry->ecx = entry->ebx;
 549                 entry->edx &= supported >> 32;
 550                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 551                 if (!supported)
 552                         break;
 553
 554                 for (idx = 1, i = 1; idx < 64; ++idx) {
 555                         u64 mask = ((u64)1 << idx);
 556                         if (*nent >= maxnent)
 557                                 goto out;
 558
 559                         do_cpuid_1_ent(&entry[i], function, idx);
 560                         if (idx == 1) {
 561                                 entry[i].eax &= kvm_cpuid_D_1_eax_x86_features;
 562                                 cpuid_mask(&entry[i].eax, CPUID_D_1_EAX);
 563                                 entry[i].ebx = 0;
 564                                 if (entry[i].eax & (F(XSAVES)|F(XSAVEC)))
 565                                         entry[i].ebx =
 566                                                 xstate_required_size(supported,
 567                                                                      true);
 568                         } else {
 569                                 if (entry[i].eax == 0 || !(supported & mask))
 570                                         continue;
 571                                 if (WARN_ON_ONCE(entry[i].ecx & 1))
 572                                         continue;
 573                         }
 574                         entry[i].ecx = 0;
 575                         entry[i].edx = 0;
 576                         entry[i].flags |=
 577                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 578                         ++*nent;
 579                         ++i;
 580                 }
 581                 break;
 582         }
 583         case KVM_CPUID_SIGNATURE: {
 584                 static const char signature[12] = "KVMKVMKVM\0\0";
 585                 const u32 *sigptr = (const u32 *)signature;
 586                 entry->eax = KVM_CPUID_FEATURES;
 587                 entry->ebx = sigptr[0];
 588                 entry->ecx = sigptr[1];
 589                 entry->edx = sigptr[2];
 590                 break;
 591         }
 592         case KVM_CPUID_FEATURES:
 593                 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
 594                              (1 << KVM_FEATURE_NOP_IO_DELAY) |
 595                              (1 << KVM_FEATURE_CLOCKSOURCE2) |
 596                              (1 << KVM_FEATURE_ASYNC_PF) |
 597                              (1 << KVM_FEATURE_PV_EOI) |
 598                              (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
 599                              (1 << KVM_FEATURE_PV_UNHALT);
 600
 601                 if (sched_info_on())
 602                         entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
 603
 604                 entry->ebx = 0;
 605                 entry->ecx = 0;
 606                 entry->edx = 0;
 607                 break;
 608         case 0x80000000:
 609                 entry->eax = min(entry->eax, 0x8000001a);
 610                 break;
 611         case 0x80000001:
 612                 entry->edx &= kvm_cpuid_8000_0001_edx_x86_features;
 613                 cpuid_mask(&entry->edx, CPUID_8000_0001_EDX);
 614                 entry->ecx &= kvm_cpuid_8000_0001_ecx_x86_features;
 615                 cpuid_mask(&entry->ecx, CPUID_8000_0001_ECX);
 616                 break;
 617         case 0x80000007: /* Advanced power management */
 618                 /* invariant TSC is CPUID.80000007H:EDX[8] */
 619                 entry->edx &= (1 << 8);
 620                 /* mask against host */
 621                 entry->edx &= boot_cpu_data.x86_power;
 622                 entry->eax = entry->ebx = entry->ecx = 0;
 623                 break;
 624         case 0x80000008: {
 625                 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
 626                 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
 627                 unsigned phys_as = entry->eax & 0xff;
 628
 629                 if (!g_phys_as)
 630                         g_phys_as = phys_as;
 631                 entry->eax = g_phys_as | (virt_as << 8);
 632                 entry->ebx = entry->edx = 0;
 633                 break;
 634         }
 635         case 0x80000019:
 636                 entry->ecx = entry->edx = 0;
 637                 break;
 638         case 0x8000001a:
 639                 break;
 640         case 0x8000001d:
 641                 break;
 642         /*Add support for Centaur's CPUID instruction*/
 643         case 0xC0000000:
 644                 /*Just support up to 0xC0000004 now*/
 645                 entry->eax = min(entry->eax, 0xC0000004);
 646                 break;
 647         case 0xC0000001:
 648                 entry->edx &= kvm_cpuid_C000_0001_edx_x86_features;
 649                 cpuid_mask(&entry->edx, CPUID_C000_0001_EDX);
 650                 break;
 651         case 3: /* Processor serial number */
 652         case 5: /* MONITOR/MWAIT */
 653         case 0xC0000002:
 654         case 0xC0000003:
 655         case 0xC0000004:
 656         default:
 657                 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
 658                 break;
 659         }
 660
 661         kvm_x86_ops->set_supported_cpuid(function, entry);
 662
 663         r = 0;
 664
 665 out:
 666         put_cpu();
 667
 668         return r;
 669 }
 670
 671 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
 672                         u32 idx, int *nent, int maxnent, unsigned int type)
 673 {
 674         if (type == KVM_GET_EMULATED_CPUID)
 675                 return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
 676
 677         return __do_cpuid_ent(entry, func, idx, nent, maxnent);
 678 }
 679
 680 #undef F
 681
 682 struct kvm_cpuid_param {
 683         u32 func;
 684         u32 idx;
 685         bool has_leaf_count;
 686         bool (*qualifier)(const struct kvm_cpuid_param *param);
 687 };
 688
 689 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
 690 {
 691         return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
 692 }
 693
 694 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
 695                                  __u32 num_entries, unsigned int ioctl_type)
 696 {
 697         int i;
 698         __u32 pad[3];
 699
 700         if (ioctl_type != KVM_GET_EMULATED_CPUID)
 701                 return false;
 702
 703         /*
 704          * We want to make sure that ->padding is being passed clean from
 705          * userspace in case we want to use it for something in the future.
 706          *
 707          * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
 708          * have to give ourselves satisfied only with the emulated side. /me
 709          * sheds a tear.
 710          */
 711         for (i = 0; i < num_entries; i++) {
 712                 if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
 713                         return true;
 714
 715                 if (pad[0] || pad[1] || pad[2])
 716                         return true;
 717         }
 718         return false;
 719 }
 720
 721 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
 722                             struct kvm_cpuid_entry2 __user *entries,
 723                             unsigned int type)
 724 {
 725         struct kvm_cpuid_entry2 *cpuid_entries;
 726         int limit, nent = 0, r = -E2BIG, i;
 727         u32 func;
 728         static const struct kvm_cpuid_param param[] = {
 729                 { .func = 0, .has_leaf_count = true },
 730                 { .func = 0x80000000, .has_leaf_count = true },
 731                 { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
 732                 { .func = KVM_CPUID_SIGNATURE },
 733                 { .func = KVM_CPUID_FEATURES },
 734         };
 735
 736         if (cpuid->nent < 1)
 737                 goto out;
 738         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 739                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
 740
 741         if (sanity_check_entries(entries, cpuid->nent, type))
 742                 return -EINVAL;
 743
 744         r = -ENOMEM;
 745         cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
 746         if (!cpuid_entries)
 747                 goto out;
 748
 749         r = 0;
 750         for (i = 0; i < ARRAY_SIZE(param); i++) {
 751                 const struct kvm_cpuid_param *ent = &param[i];
 752
 753                 if (ent->qualifier && !ent->qualifier(ent))
 754                         continue;
 755
 756                 r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
 757                                 &nent, cpuid->nent, type);
 758
 759                 if (r)
 760                         goto out_free;
 761
 762                 if (!ent->has_leaf_count)
 763                         continue;
 764
 765                 limit = cpuid_entries[nent - 1].eax;
 766                 for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
 767                         r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
 768                                      &nent, cpuid->nent, type);
 769
 770                 if (r)
 771                         goto out_free;
 772         }
 773
 774         r = -EFAULT;
 775         if (copy_to_user(entries, cpuid_entries,
 776                          nent * sizeof(struct kvm_cpuid_entry2)))
 777                 goto out_free;
 778         cpuid->nent = nent;
 779         r = 0;
 780
 781 out_free:
 782         vfree(cpuid_entries);
 783 out:
 784         return r;
 785 }
 786
 787 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
 788 {
 789         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
 790         struct kvm_cpuid_entry2 *ej;
 791         int j = i;
 792         int nent = vcpu->arch.cpuid_nent;
 793
 794         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
 795         /* when no next entry is found, the current entry[i] is reselected */
 796         do {
 797                 j = (j + 1) % nent;
 798                 ej = &vcpu->arch.cpuid_entries[j];
 799         } while (ej->function != e->function);
 800
 801         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
 802
 803         return j;
 804 }
 805
 806 /* find an entry with matching function, matching index (if needed), and that
 807  * should be read next (if it's stateful) */
 808 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
 809         u32 function, u32 index)
 810 {
 811         if (e->function != function)
 812                 return 0;
 813         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
 814                 return 0;
 815         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
 816             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
 817                 return 0;
 818         return 1;
 819 }
 820
 821 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
 822                                               u32 function, u32 index)
 823 {
 824         int i;
 825         struct kvm_cpuid_entry2 *best = NULL;
 826
 827         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 828                 struct kvm_cpuid_entry2 *e;
 829
 830                 e = &vcpu->arch.cpuid_entries[i];
 831                 if (is_matching_cpuid_entry(e, function, index)) {
 832                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
 833                                 move_to_next_stateful_cpuid_entry(vcpu, i);
 834                         best = e;
 835                         break;
 836                 }
 837         }
 838         return best;
 839 }
 840 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
 841
 842 /*
 843  * If no match is found, check whether we exceed the vCPU's limit
 844  * and return the content of the highest valid _standard_ leaf instead.
 845  * This is to satisfy the CPUID specification.
 846  */
 847 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
 848                                                   u32 function, u32 index)
 849 {
 850         struct kvm_cpuid_entry2 *maxlevel;
 851
 852         maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
 853         if (!maxlevel || maxlevel->eax >= function)
 854                 return NULL;
 855         if (function & 0x80000000) {
 856                 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
 857                 if (!maxlevel)
 858                         return NULL;
 859         }
 860         return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
 861 }
 862
 863 bool kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx,
 864                u32 *ecx, u32 *edx, bool check_limit)
 865 {
 866         u32 function = *eax, index = *ecx;
 867         struct kvm_cpuid_entry2 *best;
 868         bool entry_found = true;
 869
 870         best = kvm_find_cpuid_entry(vcpu, function, index);
 871
 872         if (!best) {
 873                 entry_found = false;
 874                 if (!check_limit)
 875                         goto out;
 876
 877                 best = check_cpuid_limit(vcpu, function, index);
 878         }
 879
 880 out:
 881         if (best) {
 882                 *eax = best->eax;
 883                 *ebx = best->ebx;
 884                 *ecx = best->ecx;
 885                 *edx = best->edx;
 886         } else
 887                 *eax = *ebx = *ecx = *edx = 0;
 888         trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx, entry_found);
 889         return entry_found;
 890 }
 891 EXPORT_SYMBOL_GPL(kvm_cpuid);
 892
 893 int kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
 894 {
 895         u32 eax, ebx, ecx, edx;
 896
 897         if (cpuid_fault_enabled(vcpu) && !kvm_require_cpl(vcpu, 0))
 898                 return 1;
 899
 900         eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
 901         ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
 902         kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx, true);
 903         kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
 904         kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
 905         kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
 906         kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
 907         return kvm_skip_emulated_instruction(vcpu);
 908 }
 909 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);