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 in the canonical
 130          * address checks; exit if it is ever changed.
 131          */
 132         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
 133         if (best && ((best->eax & 0xff00) >> 8) != 48 &&
 134                 ((best->eax & 0xff00) >> 8) != 0)
 135                 return -EINVAL;
 136
 137         /* Update physical-address width */
 138         vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
 139
 140         kvm_pmu_refresh(vcpu);
 141         return 0;
 142 }
 143
 144 static int is_efer_nx(void)
 145 {
 146         unsigned long long efer = 0;
 147
 148         rdmsrl_safe(MSR_EFER, &efer);
 149         return efer & EFER_NX;
 150 }
 151
 152 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
 153 {
 154         int i;
 155         struct kvm_cpuid_entry2 *e, *entry;
 156
 157         entry = NULL;
 158         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 159                 e = &vcpu->arch.cpuid_entries[i];
 160                 if (e->function == 0x80000001) {
 161                         entry = e;
 162                         break;
 163                 }
 164         }
 165         if (entry && (entry->edx & F(NX)) && !is_efer_nx()) {
 166                 entry->edx &= ~F(NX);
 167                 printk(KERN_INFO "kvm: guest NX capability removed\n");
 168         }
 169 }
 170
 171 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
 172 {
 173         struct kvm_cpuid_entry2 *best;
 174
 175         best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
 176         if (!best || best->eax < 0x80000008)
 177                 goto not_found;
 178         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
 179         if (best)
 180                 return best->eax & 0xff;
 181 not_found:
 182         return 36;
 183 }
 184 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr);
 185
 186 /* when an old userspace process fills a new kernel module */
 187 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
 188                              struct kvm_cpuid *cpuid,
 189                              struct kvm_cpuid_entry __user *entries)
 190 {
 191         int r, i;
 192         struct kvm_cpuid_entry *cpuid_entries = NULL;
 193
 194         r = -E2BIG;
 195         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 196                 goto out;
 197         r = -ENOMEM;
 198         if (cpuid->nent) {
 199                 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) *
 200                                         cpuid->nent);
 201                 if (!cpuid_entries)
 202                         goto out;
 203                 r = -EFAULT;
 204                 if (copy_from_user(cpuid_entries, entries,
 205                                    cpuid->nent * sizeof(struct kvm_cpuid_entry)))
 206                         goto out;
 207         }
 208         for (i = 0; i < cpuid->nent; i++) {
 209                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
 210                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
 211                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
 212                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
 213                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
 214                 vcpu->arch.cpuid_entries[i].index = 0;
 215                 vcpu->arch.cpuid_entries[i].flags = 0;
 216                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
 217                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
 218                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
 219         }
 220         vcpu->arch.cpuid_nent = cpuid->nent;
 221         cpuid_fix_nx_cap(vcpu);
 222         kvm_apic_set_version(vcpu);
 223         kvm_x86_ops->cpuid_update(vcpu);
 224         r = kvm_update_cpuid(vcpu);
 225
 226 out:
 227         vfree(cpuid_entries);
 228         return r;
 229 }
 230
 231 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
 232                               struct kvm_cpuid2 *cpuid,
 233                               struct kvm_cpuid_entry2 __user *entries)
 234 {
 235         int r;
 236
 237         r = -E2BIG;
 238         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 239                 goto out;
 240         r = -EFAULT;
 241         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
 242                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
 243                 goto out;
 244         vcpu->arch.cpuid_nent = cpuid->nent;
 245         kvm_apic_set_version(vcpu);
 246         kvm_x86_ops->cpuid_update(vcpu);
 247         r = kvm_update_cpuid(vcpu);
 248 out:
 249         return r;
 250 }
 251
 252 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
 253                               struct kvm_cpuid2 *cpuid,
 254                               struct kvm_cpuid_entry2 __user *entries)
 255 {
 256         int r;
 257
 258         r = -E2BIG;
 259         if (cpuid->nent < vcpu->arch.cpuid_nent)
 260                 goto out;
 261         r = -EFAULT;
 262         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
 263                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
 264                 goto out;
 265         return 0;
 266
 267 out:
 268         cpuid->nent = vcpu->arch.cpuid_nent;
 269         return r;
 270 }
 271
 272 static void cpuid_mask(u32 *word, int wordnum)
 273 {
 274         *word &= boot_cpu_data.x86_capability[wordnum];
 275 }
 276
 277 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 278                            u32 index)
 279 {
 280         entry->function = function;
 281         entry->index = index;
 282         cpuid_count(entry->function, entry->index,
 283                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
 284         entry->flags = 0;
 285 }
 286
 287 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry,
 288                                    u32 func, u32 index, int *nent, int maxnent)
 289 {
 290         switch (func) {
 291         case 0:
 292                 entry->eax = 1;         /* only one leaf currently */
 293                 ++*nent;
 294                 break;
 295         case 1:
 296                 entry->ecx = F(MOVBE);
 297                 ++*nent;
 298                 break;
 299         default:
 300                 break;
 301         }
 302
 303         entry->function = func;
 304         entry->index = index;
 305
 306         return 0;
 307 }
 308
 309 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 310                                  u32 index, int *nent, int maxnent)
 311 {
 312         int r;
 313         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
 314 #ifdef CONFIG_X86_64
 315         unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
 316                                 ? F(GBPAGES) : 0;
 317         unsigned f_lm = F(LM);
 318 #else
 319         unsigned f_gbpages = 0;
 320         unsigned f_lm = 0;
 321 #endif
 322         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
 323         unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
 324         unsigned f_mpx = kvm_mpx_supported() ? F(MPX) : 0;
 325         unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0;
 326
 327         /* cpuid 1.edx */
 328         const u32 kvm_cpuid_1_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(SEP) |
 332                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
 333                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
 334                 0 /* Reserved, DS, ACPI */ | F(MMX) |
 335                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
 336                 0 /* HTT, TM, Reserved, PBE */;
 337         /* cpuid 0x80000001.edx */
 338         const u32 kvm_cpuid_8000_0001_edx_x86_features =
 339                 F(FPU) | F(VME) | F(DE) | F(PSE) |
 340                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
 341                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
 342                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
 343                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
 344                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
 345                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
 346                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
 347         /* cpuid 1.ecx */
 348         const u32 kvm_cpuid_1_ecx_x86_features =
 349                 /* NOTE: MONITOR (and MWAIT) are emulated as NOP,
 350                  * but *not* advertised to guests via CPUID ! */
 351                 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
 352                 0 /* DS-CPL, VMX, SMX, EST */ |
 353                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
 354                 F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
 355                 F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
 356                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
 357                 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
 358                 F(F16C) | F(RDRAND);
 359         /* cpuid 0x80000001.ecx */
 360         const u32 kvm_cpuid_8000_0001_ecx_x86_features =
 361                 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
 362                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
 363                 F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
 364                 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
 365
 366         /* cpuid 0xC0000001.edx */
 367         const u32 kvm_cpuid_C000_0001_edx_x86_features =
 368                 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
 369                 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
 370                 F(PMM) | F(PMM_EN);
 371
 372         /* cpuid 7.0.ebx */
 373         const u32 kvm_cpuid_7_0_ebx_x86_features =
 374                 F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
 375                 F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
 376                 F(ADX) | F(SMAP) | F(AVX512IFMA) | F(AVX512F) | F(AVX512PF) |
 377                 F(AVX512ER) | F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) |
 378                 F(SHA_NI) | F(AVX512BW) | F(AVX512VL);
 379
 380         /* cpuid 0xD.1.eax */
 381         const u32 kvm_cpuid_D_1_eax_x86_features =
 382                 F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves;
 383
 384         /* cpuid 7.0.ecx*/
 385         const u32 kvm_cpuid_7_0_ecx_x86_features =
 386                 F(AVX512VBMI) | F(PKU) | 0 /*OSPKE*/ | F(AVX512_VPOPCNTDQ);
 387
 388         /* cpuid 7.0.edx*/
 389         const u32 kvm_cpuid_7_0_edx_x86_features =
 390                 KF(AVX512_4VNNIW) | KF(AVX512_4FMAPS);
 391
 392         /* all calls to cpuid_count() should be made on the same cpu */
 393         get_cpu();
 394
 395         r = -E2BIG;
 396
 397         if (*nent >= maxnent)
 398                 goto out;
 399
 400         do_cpuid_1_ent(entry, function, index);
 401         ++*nent;
 402
 403         switch (function) {
 404         case 0:
 405                 entry->eax = min(entry->eax, (u32)0xd);
 406                 break;
 407         case 1:
 408                 entry->edx &= kvm_cpuid_1_edx_x86_features;
 409                 cpuid_mask(&entry->edx, CPUID_1_EDX);
 410                 entry->ecx &= kvm_cpuid_1_ecx_x86_features;
 411                 cpuid_mask(&entry->ecx, CPUID_1_ECX);
 412                 /* we support x2apic emulation even if host does not support
 413                  * it since we emulate x2apic in software */
 414                 entry->ecx |= F(X2APIC);
 415                 break;
 416         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
 417          * may return different values. This forces us to get_cpu() before
 418          * issuing the first command, and also to emulate this annoying behavior
 419          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
 420         case 2: {
 421                 int t, times = entry->eax & 0xff;
 422
 423                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 424                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
 425                 for (t = 1; t < times; ++t) {
 426                         if (*nent >= maxnent)
 427                                 goto out;
 428
 429                         do_cpuid_1_ent(&entry[t], function, 0);
 430                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 431                         ++*nent;
 432                 }
 433                 break;
 434         }
 435         /* function 4 has additional index. */
 436         case 4: {
 437                 int i, cache_type;
 438
 439                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 440                 /* read more entries until cache_type is zero */
 441                 for (i = 1; ; ++i) {
 442                         if (*nent >= maxnent)
 443                                 goto out;
 444
 445                         cache_type = entry[i - 1].eax & 0x1f;
 446                         if (!cache_type)
 447                                 break;
 448                         do_cpuid_1_ent(&entry[i], function, i);
 449                         entry[i].flags |=
 450                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 451                         ++*nent;
 452                 }
 453                 break;
 454         }
 455         case 6: /* Thermal management */
 456                 entry->eax = 0x4; /* allow ARAT */
 457                 entry->ebx = 0;
 458                 entry->ecx = 0;
 459                 entry->edx = 0;
 460                 break;
 461         case 7: {
 462                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 463                 /* Mask ebx against host capability word 9 */
 464                 if (index == 0) {
 465                         entry->ebx &= kvm_cpuid_7_0_ebx_x86_features;
 466                         cpuid_mask(&entry->ebx, CPUID_7_0_EBX);
 467                         // TSC_ADJUST is emulated
 468                         entry->ebx |= F(TSC_ADJUST);
 469                         entry->ecx &= kvm_cpuid_7_0_ecx_x86_features;
 470                         cpuid_mask(&entry->ecx, CPUID_7_ECX);
 471                         /* PKU is not yet implemented for shadow paging. */
 472                         if (!tdp_enabled)
 473                                 entry->ecx &= ~F(PKU);
 474                         entry->edx &= kvm_cpuid_7_0_edx_x86_features;
 475                         entry->edx &= get_scattered_cpuid_leaf(7, 0, CPUID_EDX);
 476                 } else {
 477                         entry->ebx = 0;
 478                         entry->ecx = 0;
 479                         entry->edx = 0;
 480                 }
 481                 entry->eax = 0;
 482                 break;
 483         }
 484         case 9:
 485                 break;
 486         case 0xa: { /* Architectural Performance Monitoring */
 487                 struct x86_pmu_capability cap;
 488                 union cpuid10_eax eax;
 489                 union cpuid10_edx edx;
 490
 491                 perf_get_x86_pmu_capability(&cap);
 492
 493                 /*
 494                  * Only support guest architectural pmu on a host
 495                  * with architectural pmu.
 496                  */
 497                 if (!cap.version)
 498                         memset(&cap, 0, sizeof(cap));
 499
 500                 eax.split.version_id = min(cap.version, 2);
 501                 eax.split.num_counters = cap.num_counters_gp;
 502                 eax.split.bit_width = cap.bit_width_gp;
 503                 eax.split.mask_length = cap.events_mask_len;
 504
 505                 edx.split.num_counters_fixed = cap.num_counters_fixed;
 506                 edx.split.bit_width_fixed = cap.bit_width_fixed;
 507                 edx.split.reserved = 0;
 508
 509                 entry->eax = eax.full;
 510                 entry->ebx = cap.events_mask;
 511                 entry->ecx = 0;
 512                 entry->edx = edx.full;
 513                 break;
 514         }
 515         /* function 0xb has additional index. */
 516         case 0xb: {
 517                 int i, level_type;
 518
 519                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 520                 /* read more entries until level_type is zero */
 521                 for (i = 1; ; ++i) {
 522                         if (*nent >= maxnent)
 523                                 goto out;
 524
 525                         level_type = entry[i - 1].ecx & 0xff00;
 526                         if (!level_type)
 527                                 break;
 528                         do_cpuid_1_ent(&entry[i], function, i);
 529                         entry[i].flags |=
 530                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 531                         ++*nent;
 532                 }
 533                 break;
 534         }
 535         case 0xd: {
 536                 int idx, i;
 537                 u64 supported = kvm_supported_xcr0();
 538
 539                 entry->eax &= supported;
 540                 entry->ebx = xstate_required_size(supported, false);
 541                 entry->ecx = entry->ebx;
 542                 entry->edx &= supported >> 32;
 543                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 544                 if (!supported)
 545                         break;
 546
 547                 for (idx = 1, i = 1; idx < 64; ++idx) {
 548                         u64 mask = ((u64)1 << idx);
 549                         if (*nent >= maxnent)
 550                                 goto out;
 551
 552                         do_cpuid_1_ent(&entry[i], function, idx);
 553                         if (idx == 1) {
 554                                 entry[i].eax &= kvm_cpuid_D_1_eax_x86_features;
 555                                 cpuid_mask(&entry[i].eax, CPUID_D_1_EAX);
 556                                 entry[i].ebx = 0;
 557                                 if (entry[i].eax & (F(XSAVES)|F(XSAVEC)))
 558                                         entry[i].ebx =
 559                                                 xstate_required_size(supported,
 560                                                                      true);
 561                         } else {
 562                                 if (entry[i].eax == 0 || !(supported & mask))
 563                                         continue;
 564                                 if (WARN_ON_ONCE(entry[i].ecx & 1))
 565                                         continue;
 566                         }
 567                         entry[i].ecx = 0;
 568                         entry[i].edx = 0;
 569                         entry[i].flags |=
 570                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 571                         ++*nent;
 572                         ++i;
 573                 }
 574                 break;
 575         }
 576         case KVM_CPUID_SIGNATURE: {
 577                 static const char signature[12] = "KVMKVMKVM\0\0";
 578                 const u32 *sigptr = (const u32 *)signature;
 579                 entry->eax = KVM_CPUID_FEATURES;
 580                 entry->ebx = sigptr[0];
 581                 entry->ecx = sigptr[1];
 582                 entry->edx = sigptr[2];
 583                 break;
 584         }
 585         case KVM_CPUID_FEATURES:
 586                 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
 587                              (1 << KVM_FEATURE_NOP_IO_DELAY) |
 588                              (1 << KVM_FEATURE_CLOCKSOURCE2) |
 589                              (1 << KVM_FEATURE_ASYNC_PF) |
 590                              (1 << KVM_FEATURE_PV_EOI) |
 591                              (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
 592                              (1 << KVM_FEATURE_PV_UNHALT);
 593
 594                 if (sched_info_on())
 595                         entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
 596
 597                 entry->ebx = 0;
 598                 entry->ecx = 0;
 599                 entry->edx = 0;
 600                 break;
 601         case 0x80000000:
 602                 entry->eax = min(entry->eax, 0x8000001a);
 603                 break;
 604         case 0x80000001:
 605                 entry->edx &= kvm_cpuid_8000_0001_edx_x86_features;
 606                 cpuid_mask(&entry->edx, CPUID_8000_0001_EDX);
 607                 entry->ecx &= kvm_cpuid_8000_0001_ecx_x86_features;
 608                 cpuid_mask(&entry->ecx, CPUID_8000_0001_ECX);
 609                 break;
 610         case 0x80000007: /* Advanced power management */
 611                 /* invariant TSC is CPUID.80000007H:EDX[8] */
 612                 entry->edx &= (1 << 8);
 613                 /* mask against host */
 614                 entry->edx &= boot_cpu_data.x86_power;
 615                 entry->eax = entry->ebx = entry->ecx = 0;
 616                 break;
 617         case 0x80000008: {
 618                 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
 619                 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
 620                 unsigned phys_as = entry->eax & 0xff;
 621
 622                 if (!g_phys_as)
 623                         g_phys_as = phys_as;
 624                 entry->eax = g_phys_as | (virt_as << 8);
 625                 entry->ebx = entry->edx = 0;
 626                 break;
 627         }
 628         case 0x80000019:
 629                 entry->ecx = entry->edx = 0;
 630                 break;
 631         case 0x8000001a:
 632                 break;
 633         case 0x8000001d:
 634                 break;
 635         /*Add support for Centaur's CPUID instruction*/
 636         case 0xC0000000:
 637                 /*Just support up to 0xC0000004 now*/
 638                 entry->eax = min(entry->eax, 0xC0000004);
 639                 break;
 640         case 0xC0000001:
 641                 entry->edx &= kvm_cpuid_C000_0001_edx_x86_features;
 642                 cpuid_mask(&entry->edx, CPUID_C000_0001_EDX);
 643                 break;
 644         case 3: /* Processor serial number */
 645         case 5: /* MONITOR/MWAIT */
 646         case 0xC0000002:
 647         case 0xC0000003:
 648         case 0xC0000004:
 649         default:
 650                 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
 651                 break;
 652         }
 653
 654         kvm_x86_ops->set_supported_cpuid(function, entry);
 655
 656         r = 0;
 657
 658 out:
 659         put_cpu();
 660
 661         return r;
 662 }
 663
 664 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
 665                         u32 idx, int *nent, int maxnent, unsigned int type)
 666 {
 667         if (type == KVM_GET_EMULATED_CPUID)
 668                 return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
 669
 670         return __do_cpuid_ent(entry, func, idx, nent, maxnent);
 671 }
 672
 673 #undef F
 674
 675 struct kvm_cpuid_param {
 676         u32 func;
 677         u32 idx;
 678         bool has_leaf_count;
 679         bool (*qualifier)(const struct kvm_cpuid_param *param);
 680 };
 681
 682 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
 683 {
 684         return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
 685 }
 686
 687 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
 688                                  __u32 num_entries, unsigned int ioctl_type)
 689 {
 690         int i;
 691         __u32 pad[3];
 692
 693         if (ioctl_type != KVM_GET_EMULATED_CPUID)
 694                 return false;
 695
 696         /*
 697          * We want to make sure that ->padding is being passed clean from
 698          * userspace in case we want to use it for something in the future.
 699          *
 700          * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
 701          * have to give ourselves satisfied only with the emulated side. /me
 702          * sheds a tear.
 703          */
 704         for (i = 0; i < num_entries; i++) {
 705                 if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
 706                         return true;
 707
 708                 if (pad[0] || pad[1] || pad[2])
 709                         return true;
 710         }
 711         return false;
 712 }
 713
 714 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
 715                             struct kvm_cpuid_entry2 __user *entries,
 716                             unsigned int type)
 717 {
 718         struct kvm_cpuid_entry2 *cpuid_entries;
 719         int limit, nent = 0, r = -E2BIG, i;
 720         u32 func;
 721         static const struct kvm_cpuid_param param[] = {
 722                 { .func = 0, .has_leaf_count = true },
 723                 { .func = 0x80000000, .has_leaf_count = true },
 724                 { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
 725                 { .func = KVM_CPUID_SIGNATURE },
 726                 { .func = KVM_CPUID_FEATURES },
 727         };
 728
 729         if (cpuid->nent < 1)
 730                 goto out;
 731         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 732                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
 733
 734         if (sanity_check_entries(entries, cpuid->nent, type))
 735                 return -EINVAL;
 736
 737         r = -ENOMEM;
 738         cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
 739         if (!cpuid_entries)
 740                 goto out;
 741
 742         r = 0;
 743         for (i = 0; i < ARRAY_SIZE(param); i++) {
 744                 const struct kvm_cpuid_param *ent = &param[i];
 745
 746                 if (ent->qualifier && !ent->qualifier(ent))
 747                         continue;
 748
 749                 r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
 750                                 &nent, cpuid->nent, type);
 751
 752                 if (r)
 753                         goto out_free;
 754
 755                 if (!ent->has_leaf_count)
 756                         continue;
 757
 758                 limit = cpuid_entries[nent - 1].eax;
 759                 for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
 760                         r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
 761                                      &nent, cpuid->nent, type);
 762
 763                 if (r)
 764                         goto out_free;
 765         }
 766
 767         r = -EFAULT;
 768         if (copy_to_user(entries, cpuid_entries,
 769                          nent * sizeof(struct kvm_cpuid_entry2)))
 770                 goto out_free;
 771         cpuid->nent = nent;
 772         r = 0;
 773
 774 out_free:
 775         vfree(cpuid_entries);
 776 out:
 777         return r;
 778 }
 779
 780 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
 781 {
 782         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
 783         int j, nent = vcpu->arch.cpuid_nent;
 784
 785         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
 786         /* when no next entry is found, the current entry[i] is reselected */
 787         for (j = i + 1; ; j = (j + 1) % nent) {
 788                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
 789                 if (ej->function == e->function) {
 790                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
 791                         return j;
 792                 }
 793         }
 794         return 0; /* silence gcc, even though control never reaches here */
 795 }
 796
 797 /* find an entry with matching function, matching index (if needed), and that
 798  * should be read next (if it's stateful) */
 799 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
 800         u32 function, u32 index)
 801 {
 802         if (e->function != function)
 803                 return 0;
 804         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
 805                 return 0;
 806         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
 807             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
 808                 return 0;
 809         return 1;
 810 }
 811
 812 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
 813                                               u32 function, u32 index)
 814 {
 815         int i;
 816         struct kvm_cpuid_entry2 *best = NULL;
 817
 818         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 819                 struct kvm_cpuid_entry2 *e;
 820
 821                 e = &vcpu->arch.cpuid_entries[i];
 822                 if (is_matching_cpuid_entry(e, function, index)) {
 823                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
 824                                 move_to_next_stateful_cpuid_entry(vcpu, i);
 825                         best = e;
 826                         break;
 827                 }
 828         }
 829         return best;
 830 }
 831 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
 832
 833 /*
 834  * If no match is found, check whether we exceed the vCPU's limit
 835  * and return the content of the highest valid _standard_ leaf instead.
 836  * This is to satisfy the CPUID specification.
 837  */
 838 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
 839                                                   u32 function, u32 index)
 840 {
 841         struct kvm_cpuid_entry2 *maxlevel;
 842
 843         maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
 844         if (!maxlevel || maxlevel->eax >= function)
 845                 return NULL;
 846         if (function & 0x80000000) {
 847                 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
 848                 if (!maxlevel)
 849                         return NULL;
 850         }
 851         return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
 852 }
 853
 854 void kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
 855 {
 856         u32 function = *eax, index = *ecx;
 857         struct kvm_cpuid_entry2 *best;
 858
 859         best = kvm_find_cpuid_entry(vcpu, function, index);
 860
 861         if (!best)
 862                 best = check_cpuid_limit(vcpu, function, index);
 863
 864         if (best) {
 865                 *eax = best->eax;
 866                 *ebx = best->ebx;
 867                 *ecx = best->ecx;
 868                 *edx = best->edx;
 869         } else
 870                 *eax = *ebx = *ecx = *edx = 0;
 871         trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx);
 872 }
 873 EXPORT_SYMBOL_GPL(kvm_cpuid);
 874
 875 int kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
 876 {
 877         u32 eax, ebx, ecx, edx;
 878
 879         if (cpuid_fault_enabled(vcpu) && !kvm_require_cpl(vcpu, 0))
 880                 return 1;
 881
 882         eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
 883         ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
 884         kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx);
 885         kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
 886         kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
 887         kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
 888         kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
 889         return kvm_skip_emulated_instruction(vcpu);
 890 }
 891 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);