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/module.h>
  17 #include <linux/vmalloc.h>
  18 #include <linux/uaccess.h>
  19 #include <asm/fpu/internal.h> /* For use_eager_fpu.  Ugh! */
  20 #include <asm/user.h>
  21 #include <asm/fpu/xstate.h>
  22 #include "cpuid.h"
  23 #include "lapic.h"
  24 #include "mmu.h"
  25 #include "trace.h"
  26 #include "pmu.h"
  27
  28 static u32 xstate_required_size(u64 xstate_bv, bool compacted)
  29 {
  30         int feature_bit = 0;
  31         u32 ret = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
  32
  33         xstate_bv &= XFEATURE_MASK_EXTEND;
  34         while (xstate_bv) {
  35                 if (xstate_bv & 0x1) {
  36                         u32 eax, ebx, ecx, edx, offset;
  37                         cpuid_count(0xD, feature_bit, &eax, &ebx, &ecx, &edx);
  38                         offset = compacted ? ret : ebx;
  39                         ret = max(ret, offset + eax);
  40                 }
  41
  42                 xstate_bv >>= 1;
  43                 feature_bit++;
  44         }
  45
  46         return ret;
  47 }
  48
  49 bool kvm_mpx_supported(void)
  50 {
  51         return ((host_xcr0 & (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR))
  52                  && kvm_x86_ops->mpx_supported());
  53 }
  54 EXPORT_SYMBOL_GPL(kvm_mpx_supported);
  55
  56 u64 kvm_supported_xcr0(void)
  57 {
  58         u64 xcr0 = KVM_SUPPORTED_XCR0 & host_xcr0;
  59
  60         if (!kvm_mpx_supported())
  61                 xcr0 &= ~(XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR);
  62
  63         return xcr0;
  64 }
  65
  66 #define F(x) bit(X86_FEATURE_##x)
  67
  68 int kvm_update_cpuid(struct kvm_vcpu *vcpu)
  69 {
  70         struct kvm_cpuid_entry2 *best;
  71         struct kvm_lapic *apic = vcpu->arch.apic;
  72
  73         best = kvm_find_cpuid_entry(vcpu, 1, 0);
  74         if (!best)
  75                 return 0;
  76
  77         /* Update OSXSAVE bit */
  78         if (boot_cpu_has(X86_FEATURE_XSAVE) && best->function == 0x1) {
  79                 best->ecx &= ~F(OSXSAVE);
  80                 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
  81                         best->ecx |= F(OSXSAVE);
  82         }
  83
  84         if (apic) {
  85                 if (best->ecx & F(TSC_DEADLINE_TIMER))
  86                         apic->lapic_timer.timer_mode_mask = 3 << 17;
  87                 else
  88                         apic->lapic_timer.timer_mode_mask = 1 << 17;
  89         }
  90
  91         best = kvm_find_cpuid_entry(vcpu, 7, 0);
  92         if (best) {
  93                 /* Update OSPKE bit */
  94                 if (boot_cpu_has(X86_FEATURE_PKU) && best->function == 0x7) {
  95                         best->ecx &= ~F(OSPKE);
  96                         if (kvm_read_cr4_bits(vcpu, X86_CR4_PKE))
  97                                 best->ecx |= F(OSPKE);
  98                 }
  99         }
 100
 101         best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
 102         if (!best) {
 103                 vcpu->arch.guest_supported_xcr0 = 0;
 104                 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
 105         } else {
 106                 vcpu->arch.guest_supported_xcr0 =
 107                         (best->eax | ((u64)best->edx << 32)) &
 108                         kvm_supported_xcr0();
 109                 vcpu->arch.guest_xstate_size = best->ebx =
 110                         xstate_required_size(vcpu->arch.xcr0, false);
 111         }
 112
 113         best = kvm_find_cpuid_entry(vcpu, 0xD, 1);
 114         if (best && (best->eax & (F(XSAVES) | F(XSAVEC))))
 115                 best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
 116
 117         if (use_eager_fpu())
 118                 kvm_x86_ops->fpu_activate(vcpu);
 119
 120         /*
 121          * The existing code assumes virtual address is 48-bit in the canonical
 122          * address checks; exit if it is ever changed.
 123          */
 124         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
 125         if (best && ((best->eax & 0xff00) >> 8) != 48 &&
 126                 ((best->eax & 0xff00) >> 8) != 0)
 127                 return -EINVAL;
 128
 129         /* Update physical-address width */
 130         vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
 131
 132         kvm_pmu_refresh(vcpu);
 133         return 0;
 134 }
 135
 136 static int is_efer_nx(void)
 137 {
 138         unsigned long long efer = 0;
 139
 140         rdmsrl_safe(MSR_EFER, &efer);
 141         return efer & EFER_NX;
 142 }
 143
 144 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
 145 {
 146         int i;
 147         struct kvm_cpuid_entry2 *e, *entry;
 148
 149         entry = NULL;
 150         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 151                 e = &vcpu->arch.cpuid_entries[i];
 152                 if (e->function == 0x80000001) {
 153                         entry = e;
 154                         break;
 155                 }
 156         }
 157         if (entry && (entry->edx & F(NX)) && !is_efer_nx()) {
 158                 entry->edx &= ~F(NX);
 159                 printk(KERN_INFO "kvm: guest NX capability removed\n");
 160         }
 161 }
 162
 163 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
 164 {
 165         struct kvm_cpuid_entry2 *best;
 166
 167         best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
 168         if (!best || best->eax < 0x80000008)
 169                 goto not_found;
 170         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
 171         if (best)
 172                 return best->eax & 0xff;
 173 not_found:
 174         return 36;
 175 }
 176 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr);
 177
 178 /* when an old userspace process fills a new kernel module */
 179 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
 180                              struct kvm_cpuid *cpuid,
 181                              struct kvm_cpuid_entry __user *entries)
 182 {
 183         int r, i;
 184         struct kvm_cpuid_entry *cpuid_entries;
 185
 186         r = -E2BIG;
 187         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 188                 goto out;
 189         r = -ENOMEM;
 190         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
 191         if (!cpuid_entries)
 192                 goto out;
 193         r = -EFAULT;
 194         if (copy_from_user(cpuid_entries, entries,
 195                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
 196                 goto out_free;
 197         for (i = 0; i < cpuid->nent; i++) {
 198                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
 199                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
 200                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
 201                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
 202                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
 203                 vcpu->arch.cpuid_entries[i].index = 0;
 204                 vcpu->arch.cpuid_entries[i].flags = 0;
 205                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
 206                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
 207                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
 208         }
 209         vcpu->arch.cpuid_nent = cpuid->nent;
 210         cpuid_fix_nx_cap(vcpu);
 211         kvm_apic_set_version(vcpu);
 212         kvm_x86_ops->cpuid_update(vcpu);
 213         r = kvm_update_cpuid(vcpu);
 214
 215 out_free:
 216         vfree(cpuid_entries);
 217 out:
 218         return r;
 219 }
 220
 221 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
 222                               struct kvm_cpuid2 *cpuid,
 223                               struct kvm_cpuid_entry2 __user *entries)
 224 {
 225         int r;
 226
 227         r = -E2BIG;
 228         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 229                 goto out;
 230         r = -EFAULT;
 231         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
 232                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
 233                 goto out;
 234         vcpu->arch.cpuid_nent = cpuid->nent;
 235         kvm_apic_set_version(vcpu);
 236         kvm_x86_ops->cpuid_update(vcpu);
 237         r = kvm_update_cpuid(vcpu);
 238 out:
 239         return r;
 240 }
 241
 242 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
 243                               struct kvm_cpuid2 *cpuid,
 244                               struct kvm_cpuid_entry2 __user *entries)
 245 {
 246         int r;
 247
 248         r = -E2BIG;
 249         if (cpuid->nent < vcpu->arch.cpuid_nent)
 250                 goto out;
 251         r = -EFAULT;
 252         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
 253                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
 254                 goto out;
 255         return 0;
 256
 257 out:
 258         cpuid->nent = vcpu->arch.cpuid_nent;
 259         return r;
 260 }
 261
 262 static void cpuid_mask(u32 *word, int wordnum)
 263 {
 264         *word &= boot_cpu_data.x86_capability[wordnum];
 265 }
 266
 267 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 268                            u32 index)
 269 {
 270         entry->function = function;
 271         entry->index = index;
 272         cpuid_count(entry->function, entry->index,
 273                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
 274         entry->flags = 0;
 275 }
 276
 277 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry,
 278                                    u32 func, u32 index, int *nent, int maxnent)
 279 {
 280         switch (func) {
 281         case 0:
 282                 entry->eax = 1;         /* only one leaf currently */
 283                 ++*nent;
 284                 break;
 285         case 1:
 286                 entry->ecx = F(MOVBE);
 287                 ++*nent;
 288                 break;
 289         default:
 290                 break;
 291         }
 292
 293         entry->function = func;
 294         entry->index = index;
 295
 296         return 0;
 297 }
 298
 299 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 300                                  u32 index, int *nent, int maxnent)
 301 {
 302         int r;
 303         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
 304 #ifdef CONFIG_X86_64
 305         unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
 306                                 ? F(GBPAGES) : 0;
 307         unsigned f_lm = F(LM);
 308 #else
 309         unsigned f_gbpages = 0;
 310         unsigned f_lm = 0;
 311 #endif
 312         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
 313         unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
 314         unsigned f_mpx = kvm_mpx_supported() ? F(MPX) : 0;
 315         unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0;
 316
 317         /* cpuid 1.edx */
 318         const u32 kvm_cpuid_1_edx_x86_features =
 319                 F(FPU) | F(VME) | F(DE) | F(PSE) |
 320                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
 321                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
 322                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
 323                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
 324                 0 /* Reserved, DS, ACPI */ | F(MMX) |
 325                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
 326                 0 /* HTT, TM, Reserved, PBE */;
 327         /* cpuid 0x80000001.edx */
 328         const u32 kvm_cpuid_8000_0001_edx_x86_features =
 329                 F(FPU) | F(VME) | F(DE) | F(PSE) |
 330                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
 331                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
 332                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
 333                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
 334                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
 335                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
 336                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
 337         /* cpuid 1.ecx */
 338         const u32 kvm_cpuid_1_ecx_x86_features =
 339                 /* NOTE: MONITOR (and MWAIT) are emulated as NOP,
 340                  * but *not* advertised to guests via CPUID ! */
 341                 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
 342                 0 /* DS-CPL, VMX, SMX, EST */ |
 343                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
 344                 F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
 345                 F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
 346                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
 347                 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
 348                 F(F16C) | F(RDRAND);
 349         /* cpuid 0x80000001.ecx */
 350         const u32 kvm_cpuid_8000_0001_ecx_x86_features =
 351                 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
 352                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
 353                 F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
 354                 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
 355
 356         /* cpuid 0xC0000001.edx */
 357         const u32 kvm_cpuid_C000_0001_edx_x86_features =
 358                 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
 359                 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
 360                 F(PMM) | F(PMM_EN);
 361
 362         /* cpuid 7.0.ebx */
 363         const u32 kvm_cpuid_7_0_ebx_x86_features =
 364                 F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
 365                 F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
 366                 F(ADX) | F(SMAP) | F(AVX512F) | F(AVX512PF) | F(AVX512ER) |
 367                 F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(PCOMMIT);
 368
 369         /* cpuid 0xD.1.eax */
 370         const u32 kvm_cpuid_D_1_eax_x86_features =
 371                 F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves;
 372
 373         /* cpuid 7.0.ecx*/
 374         const u32 kvm_cpuid_7_0_ecx_x86_features = F(PKU) | 0 /*OSPKE*/;
 375
 376         /* all calls to cpuid_count() should be made on the same cpu */
 377         get_cpu();
 378
 379         r = -E2BIG;
 380
 381         if (*nent >= maxnent)
 382                 goto out;
 383
 384         do_cpuid_1_ent(entry, function, index);
 385         ++*nent;
 386
 387         switch (function) {
 388         case 0:
 389                 entry->eax = min(entry->eax, (u32)0xd);
 390                 break;
 391         case 1:
 392                 entry->edx &= kvm_cpuid_1_edx_x86_features;
 393                 cpuid_mask(&entry->edx, CPUID_1_EDX);
 394                 entry->ecx &= kvm_cpuid_1_ecx_x86_features;
 395                 cpuid_mask(&entry->ecx, CPUID_1_ECX);
 396                 /* we support x2apic emulation even if host does not support
 397                  * it since we emulate x2apic in software */
 398                 entry->ecx |= F(X2APIC);
 399                 break;
 400         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
 401          * may return different values. This forces us to get_cpu() before
 402          * issuing the first command, and also to emulate this annoying behavior
 403          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
 404         case 2: {
 405                 int t, times = entry->eax & 0xff;
 406
 407                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 408                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
 409                 for (t = 1; t < times; ++t) {
 410                         if (*nent >= maxnent)
 411                                 goto out;
 412
 413                         do_cpuid_1_ent(&entry[t], function, 0);
 414                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 415                         ++*nent;
 416                 }
 417                 break;
 418         }
 419         /* function 4 has additional index. */
 420         case 4: {
 421                 int i, cache_type;
 422
 423                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 424                 /* read more entries until cache_type is zero */
 425                 for (i = 1; ; ++i) {
 426                         if (*nent >= maxnent)
 427                                 goto out;
 428
 429                         cache_type = entry[i - 1].eax & 0x1f;
 430                         if (!cache_type)
 431                                 break;
 432                         do_cpuid_1_ent(&entry[i], function, i);
 433                         entry[i].flags |=
 434                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 435                         ++*nent;
 436                 }
 437                 break;
 438         }
 439         case 6: /* Thermal management */
 440                 entry->eax = 0x4; /* allow ARAT */
 441                 entry->ebx = 0;
 442                 entry->ecx = 0;
 443                 entry->edx = 0;
 444                 break;
 445         case 7: {
 446                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 447                 /* Mask ebx against host capability word 9 */
 448                 if (index == 0) {
 449                         entry->ebx &= kvm_cpuid_7_0_ebx_x86_features;
 450                         cpuid_mask(&entry->ebx, CPUID_7_0_EBX);
 451                         // TSC_ADJUST is emulated
 452                         entry->ebx |= F(TSC_ADJUST);
 453                         entry->ecx &= kvm_cpuid_7_0_ecx_x86_features;
 454                         cpuid_mask(&entry->ecx, CPUID_7_ECX);
 455                         /* PKU is not yet implemented for shadow paging. */
 456                         if (!tdp_enabled)
 457                                 entry->ecx &= ~F(PKU);
 458                 } else {
 459                         entry->ebx = 0;
 460                         entry->ecx = 0;
 461                 }
 462                 entry->eax = 0;
 463                 entry->edx = 0;
 464                 break;
 465         }
 466         case 9:
 467                 break;
 468         case 0xa: { /* Architectural Performance Monitoring */
 469                 struct x86_pmu_capability cap;
 470                 union cpuid10_eax eax;
 471                 union cpuid10_edx edx;
 472
 473                 perf_get_x86_pmu_capability(&cap);
 474
 475                 /*
 476                  * Only support guest architectural pmu on a host
 477                  * with architectural pmu.
 478                  */
 479                 if (!cap.version)
 480                         memset(&cap, 0, sizeof(cap));
 481
 482                 eax.split.version_id = min(cap.version, 2);
 483                 eax.split.num_counters = cap.num_counters_gp;
 484                 eax.split.bit_width = cap.bit_width_gp;
 485                 eax.split.mask_length = cap.events_mask_len;
 486
 487                 edx.split.num_counters_fixed = cap.num_counters_fixed;
 488                 edx.split.bit_width_fixed = cap.bit_width_fixed;
 489                 edx.split.reserved = 0;
 490
 491                 entry->eax = eax.full;
 492                 entry->ebx = cap.events_mask;
 493                 entry->ecx = 0;
 494                 entry->edx = edx.full;
 495                 break;
 496         }
 497         /* function 0xb has additional index. */
 498         case 0xb: {
 499                 int i, level_type;
 500
 501                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 502                 /* read more entries until level_type is zero */
 503                 for (i = 1; ; ++i) {
 504                         if (*nent >= maxnent)
 505                                 goto out;
 506
 507                         level_type = entry[i - 1].ecx & 0xff00;
 508                         if (!level_type)
 509                                 break;
 510                         do_cpuid_1_ent(&entry[i], function, i);
 511                         entry[i].flags |=
 512                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 513                         ++*nent;
 514                 }
 515                 break;
 516         }
 517         case 0xd: {
 518                 int idx, i;
 519                 u64 supported = kvm_supported_xcr0();
 520
 521                 entry->eax &= supported;
 522                 entry->ebx = xstate_required_size(supported, false);
 523                 entry->ecx = entry->ebx;
 524                 entry->edx &= supported >> 32;
 525                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 526                 if (!supported)
 527                         break;
 528
 529                 for (idx = 1, i = 1; idx < 64; ++idx) {
 530                         u64 mask = ((u64)1 << idx);
 531                         if (*nent >= maxnent)
 532                                 goto out;
 533
 534                         do_cpuid_1_ent(&entry[i], function, idx);
 535                         if (idx == 1) {
 536                                 entry[i].eax &= kvm_cpuid_D_1_eax_x86_features;
 537                                 cpuid_mask(&entry[i].eax, CPUID_D_1_EAX);
 538                                 entry[i].ebx = 0;
 539                                 if (entry[i].eax & (F(XSAVES)|F(XSAVEC)))
 540                                         entry[i].ebx =
 541                                                 xstate_required_size(supported,
 542                                                                      true);
 543                         } else {
 544                                 if (entry[i].eax == 0 || !(supported & mask))
 545                                         continue;
 546                                 if (WARN_ON_ONCE(entry[i].ecx & 1))
 547                                         continue;
 548                         }
 549                         entry[i].ecx = 0;
 550                         entry[i].edx = 0;
 551                         entry[i].flags |=
 552                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 553                         ++*nent;
 554                         ++i;
 555                 }
 556                 break;
 557         }
 558         case KVM_CPUID_SIGNATURE: {
 559                 static const char signature[12] = "KVMKVMKVM\0\0";
 560                 const u32 *sigptr = (const u32 *)signature;
 561                 entry->eax = KVM_CPUID_FEATURES;
 562                 entry->ebx = sigptr[0];
 563                 entry->ecx = sigptr[1];
 564                 entry->edx = sigptr[2];
 565                 break;
 566         }
 567         case KVM_CPUID_FEATURES:
 568                 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
 569                              (1 << KVM_FEATURE_NOP_IO_DELAY) |
 570                              (1 << KVM_FEATURE_CLOCKSOURCE2) |
 571                              (1 << KVM_FEATURE_ASYNC_PF) |
 572                              (1 << KVM_FEATURE_PV_EOI) |
 573                              (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
 574                              (1 << KVM_FEATURE_PV_UNHALT);
 575
 576                 if (sched_info_on())
 577                         entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
 578
 579                 entry->ebx = 0;
 580                 entry->ecx = 0;
 581                 entry->edx = 0;
 582                 break;
 583         case 0x80000000:
 584                 entry->eax = min(entry->eax, 0x8000001a);
 585                 break;
 586         case 0x80000001:
 587                 entry->edx &= kvm_cpuid_8000_0001_edx_x86_features;
 588                 cpuid_mask(&entry->edx, CPUID_8000_0001_EDX);
 589                 entry->ecx &= kvm_cpuid_8000_0001_ecx_x86_features;
 590                 cpuid_mask(&entry->ecx, CPUID_8000_0001_ECX);
 591                 break;
 592         case 0x80000007: /* Advanced power management */
 593                 /* invariant TSC is CPUID.80000007H:EDX[8] */
 594                 entry->edx &= (1 << 8);
 595                 /* mask against host */
 596                 entry->edx &= boot_cpu_data.x86_power;
 597                 entry->eax = entry->ebx = entry->ecx = 0;
 598                 break;
 599         case 0x80000008: {
 600                 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
 601                 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
 602                 unsigned phys_as = entry->eax & 0xff;
 603
 604                 if (!g_phys_as)
 605                         g_phys_as = phys_as;
 606                 entry->eax = g_phys_as | (virt_as << 8);
 607                 entry->ebx = entry->edx = 0;
 608                 break;
 609         }
 610         case 0x80000019:
 611                 entry->ecx = entry->edx = 0;
 612                 break;
 613         case 0x8000001a:
 614                 break;
 615         case 0x8000001d:
 616                 break;
 617         /*Add support for Centaur's CPUID instruction*/
 618         case 0xC0000000:
 619                 /*Just support up to 0xC0000004 now*/
 620                 entry->eax = min(entry->eax, 0xC0000004);
 621                 break;
 622         case 0xC0000001:
 623                 entry->edx &= kvm_cpuid_C000_0001_edx_x86_features;
 624                 cpuid_mask(&entry->edx, CPUID_C000_0001_EDX);
 625                 break;
 626         case 3: /* Processor serial number */
 627         case 5: /* MONITOR/MWAIT */
 628         case 0xC0000002:
 629         case 0xC0000003:
 630         case 0xC0000004:
 631         default:
 632                 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
 633                 break;
 634         }
 635
 636         kvm_x86_ops->set_supported_cpuid(function, entry);
 637
 638         r = 0;
 639
 640 out:
 641         put_cpu();
 642
 643         return r;
 644 }
 645
 646 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
 647                         u32 idx, int *nent, int maxnent, unsigned int type)
 648 {
 649         if (type == KVM_GET_EMULATED_CPUID)
 650                 return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
 651
 652         return __do_cpuid_ent(entry, func, idx, nent, maxnent);
 653 }
 654
 655 #undef F
 656
 657 struct kvm_cpuid_param {
 658         u32 func;
 659         u32 idx;
 660         bool has_leaf_count;
 661         bool (*qualifier)(const struct kvm_cpuid_param *param);
 662 };
 663
 664 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
 665 {
 666         return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
 667 }
 668
 669 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
 670                                  __u32 num_entries, unsigned int ioctl_type)
 671 {
 672         int i;
 673         __u32 pad[3];
 674
 675         if (ioctl_type != KVM_GET_EMULATED_CPUID)
 676                 return false;
 677
 678         /*
 679          * We want to make sure that ->padding is being passed clean from
 680          * userspace in case we want to use it for something in the future.
 681          *
 682          * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
 683          * have to give ourselves satisfied only with the emulated side. /me
 684          * sheds a tear.
 685          */
 686         for (i = 0; i < num_entries; i++) {
 687                 if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
 688                         return true;
 689
 690                 if (pad[0] || pad[1] || pad[2])
 691                         return true;
 692         }
 693         return false;
 694 }
 695
 696 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
 697                             struct kvm_cpuid_entry2 __user *entries,
 698                             unsigned int type)
 699 {
 700         struct kvm_cpuid_entry2 *cpuid_entries;
 701         int limit, nent = 0, r = -E2BIG, i;
 702         u32 func;
 703         static const struct kvm_cpuid_param param[] = {
 704                 { .func = 0, .has_leaf_count = true },
 705                 { .func = 0x80000000, .has_leaf_count = true },
 706                 { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
 707                 { .func = KVM_CPUID_SIGNATURE },
 708                 { .func = KVM_CPUID_FEATURES },
 709         };
 710
 711         if (cpuid->nent < 1)
 712                 goto out;
 713         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 714                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
 715
 716         if (sanity_check_entries(entries, cpuid->nent, type))
 717                 return -EINVAL;
 718
 719         r = -ENOMEM;
 720         cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
 721         if (!cpuid_entries)
 722                 goto out;
 723
 724         r = 0;
 725         for (i = 0; i < ARRAY_SIZE(param); i++) {
 726                 const struct kvm_cpuid_param *ent = &param[i];
 727
 728                 if (ent->qualifier && !ent->qualifier(ent))
 729                         continue;
 730
 731                 r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
 732                                 &nent, cpuid->nent, type);
 733
 734                 if (r)
 735                         goto out_free;
 736
 737                 if (!ent->has_leaf_count)
 738                         continue;
 739
 740                 limit = cpuid_entries[nent - 1].eax;
 741                 for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
 742                         r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
 743                                      &nent, cpuid->nent, type);
 744
 745                 if (r)
 746                         goto out_free;
 747         }
 748
 749         r = -EFAULT;
 750         if (copy_to_user(entries, cpuid_entries,
 751                          nent * sizeof(struct kvm_cpuid_entry2)))
 752                 goto out_free;
 753         cpuid->nent = nent;
 754         r = 0;
 755
 756 out_free:
 757         vfree(cpuid_entries);
 758 out:
 759         return r;
 760 }
 761
 762 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
 763 {
 764         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
 765         int j, nent = vcpu->arch.cpuid_nent;
 766
 767         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
 768         /* when no next entry is found, the current entry[i] is reselected */
 769         for (j = i + 1; ; j = (j + 1) % nent) {
 770                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
 771                 if (ej->function == e->function) {
 772                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
 773                         return j;
 774                 }
 775         }
 776         return 0; /* silence gcc, even though control never reaches here */
 777 }
 778
 779 /* find an entry with matching function, matching index (if needed), and that
 780  * should be read next (if it's stateful) */
 781 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
 782         u32 function, u32 index)
 783 {
 784         if (e->function != function)
 785                 return 0;
 786         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
 787                 return 0;
 788         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
 789             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
 790                 return 0;
 791         return 1;
 792 }
 793
 794 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
 795                                               u32 function, u32 index)
 796 {
 797         int i;
 798         struct kvm_cpuid_entry2 *best = NULL;
 799
 800         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 801                 struct kvm_cpuid_entry2 *e;
 802
 803                 e = &vcpu->arch.cpuid_entries[i];
 804                 if (is_matching_cpuid_entry(e, function, index)) {
 805                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
 806                                 move_to_next_stateful_cpuid_entry(vcpu, i);
 807                         best = e;
 808                         break;
 809                 }
 810         }
 811         return best;
 812 }
 813 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
 814
 815 /*
 816  * If no match is found, check whether we exceed the vCPU's limit
 817  * and return the content of the highest valid _standard_ leaf instead.
 818  * This is to satisfy the CPUID specification.
 819  */
 820 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
 821                                                   u32 function, u32 index)
 822 {
 823         struct kvm_cpuid_entry2 *maxlevel;
 824
 825         maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
 826         if (!maxlevel || maxlevel->eax >= function)
 827                 return NULL;
 828         if (function & 0x80000000) {
 829                 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
 830                 if (!maxlevel)
 831                         return NULL;
 832         }
 833         return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
 834 }
 835
 836 void kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
 837 {
 838         u32 function = *eax, index = *ecx;
 839         struct kvm_cpuid_entry2 *best;
 840
 841         best = kvm_find_cpuid_entry(vcpu, function, index);
 842
 843         if (!best)
 844                 best = check_cpuid_limit(vcpu, function, index);
 845
 846         /*
 847          * Perfmon not yet supported for L2 guest.
 848          */
 849         if (is_guest_mode(vcpu) && function == 0xa)
 850                 best = NULL;
 851
 852         if (best) {
 853                 *eax = best->eax;
 854                 *ebx = best->ebx;
 855                 *ecx = best->ecx;
 856                 *edx = best->edx;
 857         } else
 858                 *eax = *ebx = *ecx = *edx = 0;
 859         trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx);
 860 }
 861 EXPORT_SYMBOL_GPL(kvm_cpuid);
 862
 863 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
 864 {
 865         u32 function, eax, ebx, ecx, edx;
 866
 867         function = eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
 868         ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
 869         kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx);
 870         kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
 871         kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
 872         kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
 873         kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
 874         kvm_x86_ops->skip_emulated_instruction(vcpu);
 875 }
 876 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);