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/user.h>
  20 #include <asm/xsave.h>
  21 #include "cpuid.h"
  22 #include "lapic.h"
  23 #include "mmu.h"
  24 #include "trace.h"
  25
  26 static u32 xstate_required_size(u64 xstate_bv)
  27 {
  28         int feature_bit = 0;
  29         u32 ret = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
  30
  31         xstate_bv &= ~XSTATE_FPSSE;
  32         while (xstate_bv) {
  33                 if (xstate_bv & 0x1) {
  34                         u32 eax, ebx, ecx, edx;
  35                         cpuid_count(0xD, feature_bit, &eax, &ebx, &ecx, &edx);
  36                         ret = max(ret, eax + ebx);
  37                 }
  38
  39                 xstate_bv >>= 1;
  40                 feature_bit++;
  41         }
  42
  43         return ret;
  44 }
  45
  46 void kvm_update_cpuid(struct kvm_vcpu *vcpu)
  47 {
  48         struct kvm_cpuid_entry2 *best;
  49         struct kvm_lapic *apic = vcpu->arch.apic;
  50
  51         best = kvm_find_cpuid_entry(vcpu, 1, 0);
  52         if (!best)
  53                 return;
  54
  55         /* Update OSXSAVE bit */
  56         if (cpu_has_xsave && best->function == 0x1) {
  57                 best->ecx &= ~(bit(X86_FEATURE_OSXSAVE));
  58                 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
  59                         best->ecx |= bit(X86_FEATURE_OSXSAVE);
  60         }
  61
  62         if (apic) {
  63                 if (best->ecx & bit(X86_FEATURE_TSC_DEADLINE_TIMER))
  64                         apic->lapic_timer.timer_mode_mask = 3 << 17;
  65                 else
  66                         apic->lapic_timer.timer_mode_mask = 1 << 17;
  67         }
  68
  69         best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
  70         if (!best) {
  71                 vcpu->arch.guest_supported_xcr0 = 0;
  72                 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
  73         } else {
  74                 vcpu->arch.guest_supported_xcr0 =
  75                         (best->eax | ((u64)best->edx << 32)) &
  76                         host_xcr0 & KVM_SUPPORTED_XCR0;
  77                 vcpu->arch.guest_xstate_size =
  78                         xstate_required_size(vcpu->arch.guest_supported_xcr0);
  79         }
  80
  81         kvm_pmu_cpuid_update(vcpu);
  82 }
  83
  84 static int is_efer_nx(void)
  85 {
  86         unsigned long long efer = 0;
  87
  88         rdmsrl_safe(MSR_EFER, &efer);
  89         return efer & EFER_NX;
  90 }
  91
  92 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
  93 {
  94         int i;
  95         struct kvm_cpuid_entry2 *e, *entry;
  96
  97         entry = NULL;
  98         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
  99                 e = &vcpu->arch.cpuid_entries[i];
 100                 if (e->function == 0x80000001) {
 101                         entry = e;
 102                         break;
 103                 }
 104         }
 105         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
 106                 entry->edx &= ~(1 << 20);
 107                 printk(KERN_INFO "kvm: guest NX capability removed\n");
 108         }
 109 }
 110
 111 /* when an old userspace process fills a new kernel module */
 112 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
 113                              struct kvm_cpuid *cpuid,
 114                              struct kvm_cpuid_entry __user *entries)
 115 {
 116         int r, i;
 117         struct kvm_cpuid_entry *cpuid_entries;
 118
 119         r = -E2BIG;
 120         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 121                 goto out;
 122         r = -ENOMEM;
 123         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
 124         if (!cpuid_entries)
 125                 goto out;
 126         r = -EFAULT;
 127         if (copy_from_user(cpuid_entries, entries,
 128                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
 129                 goto out_free;
 130         for (i = 0; i < cpuid->nent; i++) {
 131                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
 132                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
 133                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
 134                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
 135                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
 136                 vcpu->arch.cpuid_entries[i].index = 0;
 137                 vcpu->arch.cpuid_entries[i].flags = 0;
 138                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
 139                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
 140                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
 141         }
 142         vcpu->arch.cpuid_nent = cpuid->nent;
 143         cpuid_fix_nx_cap(vcpu);
 144         r = 0;
 145         kvm_apic_set_version(vcpu);
 146         kvm_x86_ops->cpuid_update(vcpu);
 147         kvm_update_cpuid(vcpu);
 148
 149 out_free:
 150         vfree(cpuid_entries);
 151 out:
 152         return r;
 153 }
 154
 155 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
 156                               struct kvm_cpuid2 *cpuid,
 157                               struct kvm_cpuid_entry2 __user *entries)
 158 {
 159         int r;
 160
 161         r = -E2BIG;
 162         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 163                 goto out;
 164         r = -EFAULT;
 165         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
 166                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
 167                 goto out;
 168         vcpu->arch.cpuid_nent = cpuid->nent;
 169         kvm_apic_set_version(vcpu);
 170         kvm_x86_ops->cpuid_update(vcpu);
 171         kvm_update_cpuid(vcpu);
 172         return 0;
 173
 174 out:
 175         return r;
 176 }
 177
 178 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
 179                               struct kvm_cpuid2 *cpuid,
 180                               struct kvm_cpuid_entry2 __user *entries)
 181 {
 182         int r;
 183
 184         r = -E2BIG;
 185         if (cpuid->nent < vcpu->arch.cpuid_nent)
 186                 goto out;
 187         r = -EFAULT;
 188         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
 189                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
 190                 goto out;
 191         return 0;
 192
 193 out:
 194         cpuid->nent = vcpu->arch.cpuid_nent;
 195         return r;
 196 }
 197
 198 static void cpuid_mask(u32 *word, int wordnum)
 199 {
 200         *word &= boot_cpu_data.x86_capability[wordnum];
 201 }
 202
 203 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 204                            u32 index)
 205 {
 206         entry->function = function;
 207         entry->index = index;
 208         cpuid_count(entry->function, entry->index,
 209                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
 210         entry->flags = 0;
 211 }
 212
 213 static bool supported_xcr0_bit(unsigned bit)
 214 {
 215         u64 mask = ((u64)1 << bit);
 216
 217         return mask & KVM_SUPPORTED_XCR0 & host_xcr0;
 218 }
 219
 220 #define F(x) bit(X86_FEATURE_##x)
 221
 222 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 223                          u32 index, int *nent, int maxnent)
 224 {
 225         int r;
 226         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
 227 #ifdef CONFIG_X86_64
 228         unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
 229                                 ? F(GBPAGES) : 0;
 230         unsigned f_lm = F(LM);
 231 #else
 232         unsigned f_gbpages = 0;
 233         unsigned f_lm = 0;
 234 #endif
 235         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
 236         unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
 237
 238         /* cpuid 1.edx */
 239         const u32 kvm_supported_word0_x86_features =
 240                 F(FPU) | F(VME) | F(DE) | F(PSE) |
 241                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
 242                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
 243                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
 244                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
 245                 0 /* Reserved, DS, ACPI */ | F(MMX) |
 246                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
 247                 0 /* HTT, TM, Reserved, PBE */;
 248         /* cpuid 0x80000001.edx */
 249         const u32 kvm_supported_word1_x86_features =
 250                 F(FPU) | F(VME) | F(DE) | F(PSE) |
 251                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
 252                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
 253                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
 254                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
 255                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
 256                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
 257                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
 258         /* cpuid 1.ecx */
 259         const u32 kvm_supported_word4_x86_features =
 260                 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
 261                 0 /* DS-CPL, VMX, SMX, EST */ |
 262                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
 263                 F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
 264                 F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
 265                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
 266                 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
 267                 F(F16C) | F(RDRAND);
 268         /* cpuid 0x80000001.ecx */
 269         const u32 kvm_supported_word6_x86_features =
 270                 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
 271                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
 272                 F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
 273                 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
 274
 275         /* cpuid 0xC0000001.edx */
 276         const u32 kvm_supported_word5_x86_features =
 277                 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
 278                 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
 279                 F(PMM) | F(PMM_EN);
 280
 281         /* cpuid 7.0.ebx */
 282         const u32 kvm_supported_word9_x86_features =
 283                 F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
 284                 F(BMI2) | F(ERMS) | f_invpcid | F(RTM);
 285
 286         /* all calls to cpuid_count() should be made on the same cpu */
 287         get_cpu();
 288
 289         r = -E2BIG;
 290
 291         if (*nent >= maxnent)
 292                 goto out;
 293
 294         do_cpuid_1_ent(entry, function, index);
 295         ++*nent;
 296
 297         switch (function) {
 298         case 0:
 299                 entry->eax = min(entry->eax, (u32)0xd);
 300                 break;
 301         case 1:
 302                 entry->edx &= kvm_supported_word0_x86_features;
 303                 cpuid_mask(&entry->edx, 0);
 304                 entry->ecx &= kvm_supported_word4_x86_features;
 305                 cpuid_mask(&entry->ecx, 4);
 306                 /* we support x2apic emulation even if host does not support
 307                  * it since we emulate x2apic in software */
 308                 entry->ecx |= F(X2APIC);
 309                 break;
 310         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
 311          * may return different values. This forces us to get_cpu() before
 312          * issuing the first command, and also to emulate this annoying behavior
 313          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
 314         case 2: {
 315                 int t, times = entry->eax & 0xff;
 316
 317                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 318                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
 319                 for (t = 1; t < times; ++t) {
 320                         if (*nent >= maxnent)
 321                                 goto out;
 322
 323                         do_cpuid_1_ent(&entry[t], function, 0);
 324                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 325                         ++*nent;
 326                 }
 327                 break;
 328         }
 329         /* function 4 has additional index. */
 330         case 4: {
 331                 int i, cache_type;
 332
 333                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 334                 /* read more entries until cache_type is zero */
 335                 for (i = 1; ; ++i) {
 336                         if (*nent >= maxnent)
 337                                 goto out;
 338
 339                         cache_type = entry[i - 1].eax & 0x1f;
 340                         if (!cache_type)
 341                                 break;
 342                         do_cpuid_1_ent(&entry[i], function, i);
 343                         entry[i].flags |=
 344                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 345                         ++*nent;
 346                 }
 347                 break;
 348         }
 349         case 7: {
 350                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 351                 /* Mask ebx against host capability word 9 */
 352                 if (index == 0) {
 353                         entry->ebx &= kvm_supported_word9_x86_features;
 354                         cpuid_mask(&entry->ebx, 9);
 355                         // TSC_ADJUST is emulated
 356                         entry->ebx |= F(TSC_ADJUST);
 357                 } else
 358                         entry->ebx = 0;
 359                 entry->eax = 0;
 360                 entry->ecx = 0;
 361                 entry->edx = 0;
 362                 break;
 363         }
 364         case 9:
 365                 break;
 366         case 0xa: { /* Architectural Performance Monitoring */
 367                 struct x86_pmu_capability cap;
 368                 union cpuid10_eax eax;
 369                 union cpuid10_edx edx;
 370
 371                 perf_get_x86_pmu_capability(&cap);
 372
 373                 /*
 374                  * Only support guest architectural pmu on a host
 375                  * with architectural pmu.
 376                  */
 377                 if (!cap.version)
 378                         memset(&cap, 0, sizeof(cap));
 379
 380                 eax.split.version_id = min(cap.version, 2);
 381                 eax.split.num_counters = cap.num_counters_gp;
 382                 eax.split.bit_width = cap.bit_width_gp;
 383                 eax.split.mask_length = cap.events_mask_len;
 384
 385                 edx.split.num_counters_fixed = cap.num_counters_fixed;
 386                 edx.split.bit_width_fixed = cap.bit_width_fixed;
 387                 edx.split.reserved = 0;
 388
 389                 entry->eax = eax.full;
 390                 entry->ebx = cap.events_mask;
 391                 entry->ecx = 0;
 392                 entry->edx = edx.full;
 393                 break;
 394         }
 395         /* function 0xb has additional index. */
 396         case 0xb: {
 397                 int i, level_type;
 398
 399                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 400                 /* read more entries until level_type is zero */
 401                 for (i = 1; ; ++i) {
 402                         if (*nent >= maxnent)
 403                                 goto out;
 404
 405                         level_type = entry[i - 1].ecx & 0xff00;
 406                         if (!level_type)
 407                                 break;
 408                         do_cpuid_1_ent(&entry[i], function, i);
 409                         entry[i].flags |=
 410                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 411                         ++*nent;
 412                 }
 413                 break;
 414         }
 415         case 0xd: {
 416                 int idx, i;
 417
 418                 entry->eax &= host_xcr0 & KVM_SUPPORTED_XCR0;
 419                 entry->edx &= (host_xcr0 & KVM_SUPPORTED_XCR0) >> 32;
 420                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 421                 for (idx = 1, i = 1; idx < 64; ++idx) {
 422                         if (*nent >= maxnent)
 423                                 goto out;
 424
 425                         do_cpuid_1_ent(&entry[i], function, idx);
 426                         if (entry[i].eax == 0 || !supported_xcr0_bit(idx))
 427                                 continue;
 428                         entry[i].flags |=
 429                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 430                         ++*nent;
 431                         ++i;
 432                 }
 433                 break;
 434         }
 435         case KVM_CPUID_SIGNATURE: {
 436                 static const char signature[12] = "KVMKVMKVM\0\0";
 437                 const u32 *sigptr = (const u32 *)signature;
 438                 entry->eax = KVM_CPUID_FEATURES;
 439                 entry->ebx = sigptr[0];
 440                 entry->ecx = sigptr[1];
 441                 entry->edx = sigptr[2];
 442                 break;
 443         }
 444         case KVM_CPUID_FEATURES:
 445                 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
 446                              (1 << KVM_FEATURE_NOP_IO_DELAY) |
 447                              (1 << KVM_FEATURE_CLOCKSOURCE2) |
 448                              (1 << KVM_FEATURE_ASYNC_PF) |
 449                              (1 << KVM_FEATURE_PV_EOI) |
 450                              (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
 451                              (1 << KVM_FEATURE_PV_UNHALT);
 452
 453                 if (sched_info_on())
 454                         entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
 455
 456                 entry->ebx = 0;
 457                 entry->ecx = 0;
 458                 entry->edx = 0;
 459                 break;
 460         case 0x80000000:
 461                 entry->eax = min(entry->eax, 0x8000001a);
 462                 break;
 463         case 0x80000001:
 464                 entry->edx &= kvm_supported_word1_x86_features;
 465                 cpuid_mask(&entry->edx, 1);
 466                 entry->ecx &= kvm_supported_word6_x86_features;
 467                 cpuid_mask(&entry->ecx, 6);
 468                 break;
 469         case 0x80000008: {
 470                 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
 471                 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
 472                 unsigned phys_as = entry->eax & 0xff;
 473
 474                 if (!g_phys_as)
 475                         g_phys_as = phys_as;
 476                 entry->eax = g_phys_as | (virt_as << 8);
 477                 entry->ebx = entry->edx = 0;
 478                 break;
 479         }
 480         case 0x80000019:
 481                 entry->ecx = entry->edx = 0;
 482                 break;
 483         case 0x8000001a:
 484                 break;
 485         case 0x8000001d:
 486                 break;
 487         /*Add support for Centaur's CPUID instruction*/
 488         case 0xC0000000:
 489                 /*Just support up to 0xC0000004 now*/
 490                 entry->eax = min(entry->eax, 0xC0000004);
 491                 break;
 492         case 0xC0000001:
 493                 entry->edx &= kvm_supported_word5_x86_features;
 494                 cpuid_mask(&entry->edx, 5);
 495                 break;
 496         case 3: /* Processor serial number */
 497         case 5: /* MONITOR/MWAIT */
 498         case 6: /* Thermal management */
 499         case 0x80000007: /* Advanced power management */
 500         case 0xC0000002:
 501         case 0xC0000003:
 502         case 0xC0000004:
 503         default:
 504                 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
 505                 break;
 506         }
 507
 508         kvm_x86_ops->set_supported_cpuid(function, entry);
 509
 510         r = 0;
 511
 512 out:
 513         put_cpu();
 514
 515         return r;
 516 }
 517
 518 #undef F
 519
 520 struct kvm_cpuid_param {
 521         u32 func;
 522         u32 idx;
 523         bool has_leaf_count;
 524         bool (*qualifier)(const struct kvm_cpuid_param *param);
 525 };
 526
 527 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
 528 {
 529         return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
 530 }
 531
 532 int kvm_dev_ioctl_get_supported_cpuid(struct kvm_cpuid2 *cpuid,
 533                                       struct kvm_cpuid_entry2 __user *entries)
 534 {
 535         struct kvm_cpuid_entry2 *cpuid_entries;
 536         int limit, nent = 0, r = -E2BIG, i;
 537         u32 func;
 538         static const struct kvm_cpuid_param param[] = {
 539                 { .func = 0, .has_leaf_count = true },
 540                 { .func = 0x80000000, .has_leaf_count = true },
 541                 { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
 542                 { .func = KVM_CPUID_SIGNATURE },
 543                 { .func = KVM_CPUID_FEATURES },
 544         };
 545
 546         if (cpuid->nent < 1)
 547                 goto out;
 548         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 549                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
 550         r = -ENOMEM;
 551         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
 552         if (!cpuid_entries)
 553                 goto out;
 554
 555         r = 0;
 556         for (i = 0; i < ARRAY_SIZE(param); i++) {
 557                 const struct kvm_cpuid_param *ent = &param[i];
 558
 559                 if (ent->qualifier && !ent->qualifier(ent))
 560                         continue;
 561
 562                 r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
 563                                 &nent, cpuid->nent);
 564
 565                 if (r)
 566                         goto out_free;
 567
 568                 if (!ent->has_leaf_count)
 569                         continue;
 570
 571                 limit = cpuid_entries[nent - 1].eax;
 572                 for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
 573                         r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
 574                                      &nent, cpuid->nent);
 575
 576                 if (r)
 577                         goto out_free;
 578         }
 579
 580         r = -EFAULT;
 581         if (copy_to_user(entries, cpuid_entries,
 582                          nent * sizeof(struct kvm_cpuid_entry2)))
 583                 goto out_free;
 584         cpuid->nent = nent;
 585         r = 0;
 586
 587 out_free:
 588         vfree(cpuid_entries);
 589 out:
 590         return r;
 591 }
 592
 593 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
 594 {
 595         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
 596         int j, nent = vcpu->arch.cpuid_nent;
 597
 598         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
 599         /* when no next entry is found, the current entry[i] is reselected */
 600         for (j = i + 1; ; j = (j + 1) % nent) {
 601                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
 602                 if (ej->function == e->function) {
 603                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
 604                         return j;
 605                 }
 606         }
 607         return 0; /* silence gcc, even though control never reaches here */
 608 }
 609
 610 /* find an entry with matching function, matching index (if needed), and that
 611  * should be read next (if it's stateful) */
 612 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
 613         u32 function, u32 index)
 614 {
 615         if (e->function != function)
 616                 return 0;
 617         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
 618                 return 0;
 619         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
 620             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
 621                 return 0;
 622         return 1;
 623 }
 624
 625 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
 626                                               u32 function, u32 index)
 627 {
 628         int i;
 629         struct kvm_cpuid_entry2 *best = NULL;
 630
 631         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 632                 struct kvm_cpuid_entry2 *e;
 633
 634                 e = &vcpu->arch.cpuid_entries[i];
 635                 if (is_matching_cpuid_entry(e, function, index)) {
 636                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
 637                                 move_to_next_stateful_cpuid_entry(vcpu, i);
 638                         best = e;
 639                         break;
 640                 }
 641         }
 642         return best;
 643 }
 644 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
 645
 646 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
 647 {
 648         struct kvm_cpuid_entry2 *best;
 649
 650         best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
 651         if (!best || best->eax < 0x80000008)
 652                 goto not_found;
 653         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
 654         if (best)
 655                 return best->eax & 0xff;
 656 not_found:
 657         return 36;
 658 }
 659
 660 /*
 661  * If no match is found, check whether we exceed the vCPU's limit
 662  * and return the content of the highest valid _standard_ leaf instead.
 663  * This is to satisfy the CPUID specification.
 664  */
 665 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
 666                                                   u32 function, u32 index)
 667 {
 668         struct kvm_cpuid_entry2 *maxlevel;
 669
 670         maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
 671         if (!maxlevel || maxlevel->eax >= function)
 672                 return NULL;
 673         if (function & 0x80000000) {
 674                 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
 675                 if (!maxlevel)
 676                         return NULL;
 677         }
 678         return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
 679 }
 680
 681 void kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
 682 {
 683         u32 function = *eax, index = *ecx;
 684         struct kvm_cpuid_entry2 *best;
 685
 686         best = kvm_find_cpuid_entry(vcpu, function, index);
 687
 688         if (!best)
 689                 best = check_cpuid_limit(vcpu, function, index);
 690
 691         if (best) {
 692                 *eax = best->eax;
 693                 *ebx = best->ebx;
 694                 *ecx = best->ecx;
 695                 *edx = best->edx;
 696         } else
 697                 *eax = *ebx = *ecx = *edx = 0;
 698 }
 699 EXPORT_SYMBOL_GPL(kvm_cpuid);
 700
 701 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
 702 {
 703         u32 function, eax, ebx, ecx, edx;
 704
 705         function = eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
 706         ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
 707         kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx);
 708         kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
 709         kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
 710         kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
 711         kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
 712         kvm_x86_ops->skip_emulated_instruction(vcpu);
 713         trace_kvm_cpuid(function, eax, ebx, ecx, edx);
 714 }
 715 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);