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 void kvm_update_cpuid(struct kvm_vcpu *vcpu)
  27 {
  28         struct kvm_cpuid_entry2 *best;
  29         struct kvm_lapic *apic = vcpu->arch.apic;
  30
  31         best = kvm_find_cpuid_entry(vcpu, 1, 0);
  32         if (!best)
  33                 return;
  34
  35         /* Update OSXSAVE bit */
  36         if (cpu_has_xsave && best->function == 0x1) {
  37                 best->ecx &= ~(bit(X86_FEATURE_OSXSAVE));
  38                 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
  39                         best->ecx |= bit(X86_FEATURE_OSXSAVE);
  40         }
  41
  42         if (apic) {
  43                 if (best->ecx & bit(X86_FEATURE_TSC_DEADLINE_TIMER))
  44                         apic->lapic_timer.timer_mode_mask = 3 << 17;
  45                 else
  46                         apic->lapic_timer.timer_mode_mask = 1 << 17;
  47         }
  48
  49         kvm_pmu_cpuid_update(vcpu);
  50 }
  51
  52 static int is_efer_nx(void)
  53 {
  54         unsigned long long efer = 0;
  55
  56         rdmsrl_safe(MSR_EFER, &efer);
  57         return efer & EFER_NX;
  58 }
  59
  60 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
  61 {
  62         int i;
  63         struct kvm_cpuid_entry2 *e, *entry;
  64
  65         entry = NULL;
  66         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
  67                 e = &vcpu->arch.cpuid_entries[i];
  68                 if (e->function == 0x80000001) {
  69                         entry = e;
  70                         break;
  71                 }
  72         }
  73         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
  74                 entry->edx &= ~(1 << 20);
  75                 printk(KERN_INFO "kvm: guest NX capability removed\n");
  76         }
  77 }
  78
  79 /* when an old userspace process fills a new kernel module */
  80 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
  81                              struct kvm_cpuid *cpuid,
  82                              struct kvm_cpuid_entry __user *entries)
  83 {
  84         int r, i;
  85         struct kvm_cpuid_entry *cpuid_entries;
  86
  87         r = -E2BIG;
  88         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
  89                 goto out;
  90         r = -ENOMEM;
  91         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
  92         if (!cpuid_entries)
  93                 goto out;
  94         r = -EFAULT;
  95         if (copy_from_user(cpuid_entries, entries,
  96                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
  97                 goto out_free;
  98         for (i = 0; i < cpuid->nent; i++) {
  99                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
 100                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
 101                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
 102                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
 103                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
 104                 vcpu->arch.cpuid_entries[i].index = 0;
 105                 vcpu->arch.cpuid_entries[i].flags = 0;
 106                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
 107                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
 108                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
 109         }
 110         vcpu->arch.cpuid_nent = cpuid->nent;
 111         cpuid_fix_nx_cap(vcpu);
 112         r = 0;
 113         kvm_apic_set_version(vcpu);
 114         kvm_x86_ops->cpuid_update(vcpu);
 115         kvm_update_cpuid(vcpu);
 116
 117 out_free:
 118         vfree(cpuid_entries);
 119 out:
 120         return r;
 121 }
 122
 123 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
 124                               struct kvm_cpuid2 *cpuid,
 125                               struct kvm_cpuid_entry2 __user *entries)
 126 {
 127         int r;
 128
 129         r = -E2BIG;
 130         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 131                 goto out;
 132         r = -EFAULT;
 133         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
 134                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
 135                 goto out;
 136         vcpu->arch.cpuid_nent = cpuid->nent;
 137         kvm_apic_set_version(vcpu);
 138         kvm_x86_ops->cpuid_update(vcpu);
 139         kvm_update_cpuid(vcpu);
 140         return 0;
 141
 142 out:
 143         return r;
 144 }
 145
 146 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
 147                               struct kvm_cpuid2 *cpuid,
 148                               struct kvm_cpuid_entry2 __user *entries)
 149 {
 150         int r;
 151
 152         r = -E2BIG;
 153         if (cpuid->nent < vcpu->arch.cpuid_nent)
 154                 goto out;
 155         r = -EFAULT;
 156         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
 157                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
 158                 goto out;
 159         return 0;
 160
 161 out:
 162         cpuid->nent = vcpu->arch.cpuid_nent;
 163         return r;
 164 }
 165
 166 static void cpuid_mask(u32 *word, int wordnum)
 167 {
 168         *word &= boot_cpu_data.x86_capability[wordnum];
 169 }
 170
 171 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 172                            u32 index)
 173 {
 174         entry->function = function;
 175         entry->index = index;
 176         cpuid_count(entry->function, entry->index,
 177                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
 178         entry->flags = 0;
 179 }
 180
 181 static bool supported_xcr0_bit(unsigned bit)
 182 {
 183         u64 mask = ((u64)1 << bit);
 184
 185         return mask & (XSTATE_FP | XSTATE_SSE | XSTATE_YMM) & host_xcr0;
 186 }
 187
 188 #define F(x) bit(X86_FEATURE_##x)
 189
 190 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry,
 191                                    u32 func, u32 index, int *nent, int maxnent)
 192 {
 193         return 0;
 194 }
 195
 196 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 197                                  u32 index, int *nent, int maxnent)
 198 {
 199         int r;
 200         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
 201 #ifdef CONFIG_X86_64
 202         unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
 203                                 ? F(GBPAGES) : 0;
 204         unsigned f_lm = F(LM);
 205 #else
 206         unsigned f_gbpages = 0;
 207         unsigned f_lm = 0;
 208 #endif
 209         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
 210         unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
 211
 212         /* cpuid 1.edx */
 213         const u32 kvm_supported_word0_x86_features =
 214                 F(FPU) | F(VME) | F(DE) | F(PSE) |
 215                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
 216                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
 217                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
 218                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
 219                 0 /* Reserved, DS, ACPI */ | F(MMX) |
 220                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
 221                 0 /* HTT, TM, Reserved, PBE */;
 222         /* cpuid 0x80000001.edx */
 223         const u32 kvm_supported_word1_x86_features =
 224                 F(FPU) | F(VME) | F(DE) | F(PSE) |
 225                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
 226                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
 227                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
 228                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
 229                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
 230                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
 231                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
 232         /* cpuid 1.ecx */
 233         const u32 kvm_supported_word4_x86_features =
 234                 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
 235                 0 /* DS-CPL, VMX, SMX, EST */ |
 236                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
 237                 F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
 238                 F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
 239                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
 240                 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
 241                 F(F16C) | F(RDRAND);
 242         /* cpuid 0x80000001.ecx */
 243         const u32 kvm_supported_word6_x86_features =
 244                 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
 245                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
 246                 F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
 247                 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
 248
 249         /* cpuid 0xC0000001.edx */
 250         const u32 kvm_supported_word5_x86_features =
 251                 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
 252                 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
 253                 F(PMM) | F(PMM_EN);
 254
 255         /* cpuid 7.0.ebx */
 256         const u32 kvm_supported_word9_x86_features =
 257                 F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
 258                 F(BMI2) | F(ERMS) | f_invpcid | F(RTM);
 259
 260         /* all calls to cpuid_count() should be made on the same cpu */
 261         get_cpu();
 262
 263         r = -E2BIG;
 264
 265         if (*nent >= maxnent)
 266                 goto out;
 267
 268         do_cpuid_1_ent(entry, function, index);
 269         ++*nent;
 270
 271         switch (function) {
 272         case 0:
 273                 entry->eax = min(entry->eax, (u32)0xd);
 274                 break;
 275         case 1:
 276                 entry->edx &= kvm_supported_word0_x86_features;
 277                 cpuid_mask(&entry->edx, 0);
 278                 entry->ecx &= kvm_supported_word4_x86_features;
 279                 cpuid_mask(&entry->ecx, 4);
 280                 /* we support x2apic emulation even if host does not support
 281                  * it since we emulate x2apic in software */
 282                 entry->ecx |= F(X2APIC);
 283                 break;
 284         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
 285          * may return different values. This forces us to get_cpu() before
 286          * issuing the first command, and also to emulate this annoying behavior
 287          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
 288         case 2: {
 289                 int t, times = entry->eax & 0xff;
 290
 291                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 292                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
 293                 for (t = 1; t < times; ++t) {
 294                         if (*nent >= maxnent)
 295                                 goto out;
 296
 297                         do_cpuid_1_ent(&entry[t], function, 0);
 298                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 299                         ++*nent;
 300                 }
 301                 break;
 302         }
 303         /* function 4 has additional index. */
 304         case 4: {
 305                 int i, cache_type;
 306
 307                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 308                 /* read more entries until cache_type is zero */
 309                 for (i = 1; ; ++i) {
 310                         if (*nent >= maxnent)
 311                                 goto out;
 312
 313                         cache_type = entry[i - 1].eax & 0x1f;
 314                         if (!cache_type)
 315                                 break;
 316                         do_cpuid_1_ent(&entry[i], function, i);
 317                         entry[i].flags |=
 318                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 319                         ++*nent;
 320                 }
 321                 break;
 322         }
 323         case 7: {
 324                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 325                 /* Mask ebx against host capability word 9 */
 326                 if (index == 0) {
 327                         entry->ebx &= kvm_supported_word9_x86_features;
 328                         cpuid_mask(&entry->ebx, 9);
 329                         // TSC_ADJUST is emulated
 330                         entry->ebx |= F(TSC_ADJUST);
 331                 } else
 332                         entry->ebx = 0;
 333                 entry->eax = 0;
 334                 entry->ecx = 0;
 335                 entry->edx = 0;
 336                 break;
 337         }
 338         case 9:
 339                 break;
 340         case 0xa: { /* Architectural Performance Monitoring */
 341                 struct x86_pmu_capability cap;
 342                 union cpuid10_eax eax;
 343                 union cpuid10_edx edx;
 344
 345                 perf_get_x86_pmu_capability(&cap);
 346
 347                 /*
 348                  * Only support guest architectural pmu on a host
 349                  * with architectural pmu.
 350                  */
 351                 if (!cap.version)
 352                         memset(&cap, 0, sizeof(cap));
 353
 354                 eax.split.version_id = min(cap.version, 2);
 355                 eax.split.num_counters = cap.num_counters_gp;
 356                 eax.split.bit_width = cap.bit_width_gp;
 357                 eax.split.mask_length = cap.events_mask_len;
 358
 359                 edx.split.num_counters_fixed = cap.num_counters_fixed;
 360                 edx.split.bit_width_fixed = cap.bit_width_fixed;
 361                 edx.split.reserved = 0;
 362
 363                 entry->eax = eax.full;
 364                 entry->ebx = cap.events_mask;
 365                 entry->ecx = 0;
 366                 entry->edx = edx.full;
 367                 break;
 368         }
 369         /* function 0xb has additional index. */
 370         case 0xb: {
 371                 int i, level_type;
 372
 373                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 374                 /* read more entries until level_type is zero */
 375                 for (i = 1; ; ++i) {
 376                         if (*nent >= maxnent)
 377                                 goto out;
 378
 379                         level_type = entry[i - 1].ecx & 0xff00;
 380                         if (!level_type)
 381                                 break;
 382                         do_cpuid_1_ent(&entry[i], function, i);
 383                         entry[i].flags |=
 384                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 385                         ++*nent;
 386                 }
 387                 break;
 388         }
 389         case 0xd: {
 390                 int idx, i;
 391
 392                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 393                 for (idx = 1, i = 1; idx < 64; ++idx) {
 394                         if (*nent >= maxnent)
 395                                 goto out;
 396
 397                         do_cpuid_1_ent(&entry[i], function, idx);
 398                         if (entry[i].eax == 0 || !supported_xcr0_bit(idx))
 399                                 continue;
 400                         entry[i].flags |=
 401                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 402                         ++*nent;
 403                         ++i;
 404                 }
 405                 break;
 406         }
 407         case KVM_CPUID_SIGNATURE: {
 408                 static const char signature[12] = "KVMKVMKVM\0\0";
 409                 const u32 *sigptr = (const u32 *)signature;
 410                 entry->eax = KVM_CPUID_FEATURES;
 411                 entry->ebx = sigptr[0];
 412                 entry->ecx = sigptr[1];
 413                 entry->edx = sigptr[2];
 414                 break;
 415         }
 416         case KVM_CPUID_FEATURES:
 417                 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
 418                              (1 << KVM_FEATURE_NOP_IO_DELAY) |
 419                              (1 << KVM_FEATURE_CLOCKSOURCE2) |
 420                              (1 << KVM_FEATURE_ASYNC_PF) |
 421                              (1 << KVM_FEATURE_PV_EOI) |
 422                              (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT);
 423
 424                 if (sched_info_on())
 425                         entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
 426
 427                 entry->ebx = 0;
 428                 entry->ecx = 0;
 429                 entry->edx = 0;
 430                 break;
 431         case 0x80000000:
 432                 entry->eax = min(entry->eax, 0x8000001a);
 433                 break;
 434         case 0x80000001:
 435                 entry->edx &= kvm_supported_word1_x86_features;
 436                 cpuid_mask(&entry->edx, 1);
 437                 entry->ecx &= kvm_supported_word6_x86_features;
 438                 cpuid_mask(&entry->ecx, 6);
 439                 break;
 440         case 0x80000008: {
 441                 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
 442                 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
 443                 unsigned phys_as = entry->eax & 0xff;
 444
 445                 if (!g_phys_as)
 446                         g_phys_as = phys_as;
 447                 entry->eax = g_phys_as | (virt_as << 8);
 448                 entry->ebx = entry->edx = 0;
 449                 break;
 450         }
 451         case 0x80000019:
 452                 entry->ecx = entry->edx = 0;
 453                 break;
 454         case 0x8000001a:
 455                 break;
 456         case 0x8000001d:
 457                 break;
 458         /*Add support for Centaur's CPUID instruction*/
 459         case 0xC0000000:
 460                 /*Just support up to 0xC0000004 now*/
 461                 entry->eax = min(entry->eax, 0xC0000004);
 462                 break;
 463         case 0xC0000001:
 464                 entry->edx &= kvm_supported_word5_x86_features;
 465                 cpuid_mask(&entry->edx, 5);
 466                 break;
 467         case 3: /* Processor serial number */
 468         case 5: /* MONITOR/MWAIT */
 469         case 6: /* Thermal management */
 470         case 0x80000007: /* Advanced power management */
 471         case 0xC0000002:
 472         case 0xC0000003:
 473         case 0xC0000004:
 474         default:
 475                 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
 476                 break;
 477         }
 478
 479         kvm_x86_ops->set_supported_cpuid(function, entry);
 480
 481         r = 0;
 482
 483 out:
 484         put_cpu();
 485
 486         return r;
 487 }
 488
 489 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
 490                         u32 idx, int *nent, int maxnent, unsigned int type)
 491 {
 492         if (type == KVM_GET_EMULATED_CPUID)
 493                 return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
 494
 495         return __do_cpuid_ent(entry, func, idx, nent, maxnent);
 496 }
 497
 498 #undef F
 499
 500 struct kvm_cpuid_param {
 501         u32 func;
 502         u32 idx;
 503         bool has_leaf_count;
 504         bool (*qualifier)(const struct kvm_cpuid_param *param);
 505 };
 506
 507 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
 508 {
 509         return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
 510 }
 511
 512 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
 513                                  __u32 num_entries, unsigned int ioctl_type)
 514 {
 515         int i;
 516
 517         if (ioctl_type != KVM_GET_EMULATED_CPUID)
 518                 return false;
 519
 520         /*
 521          * We want to make sure that ->padding is being passed clean from
 522          * userspace in case we want to use it for something in the future.
 523          *
 524          * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
 525          * have to give ourselves satisfied only with the emulated side. /me
 526          * sheds a tear.
 527          */
 528         for (i = 0; i < num_entries; i++) {
 529                 if (entries[i].padding[0] ||
 530                     entries[i].padding[1] ||
 531                     entries[i].padding[2])
 532                         return true;
 533         }
 534         return false;
 535 }
 536
 537 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
 538                             struct kvm_cpuid_entry2 __user *entries,
 539                             unsigned int type)
 540 {
 541         struct kvm_cpuid_entry2 *cpuid_entries;
 542         int limit, nent = 0, r = -E2BIG, i;
 543         u32 func;
 544         static const struct kvm_cpuid_param param[] = {
 545                 { .func = 0, .has_leaf_count = true },
 546                 { .func = 0x80000000, .has_leaf_count = true },
 547                 { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
 548                 { .func = KVM_CPUID_SIGNATURE },
 549                 { .func = KVM_CPUID_FEATURES },
 550         };
 551
 552         if (cpuid->nent < 1)
 553                 goto out;
 554         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 555                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
 556
 557         if (sanity_check_entries(entries, cpuid->nent, type))
 558                 return -EINVAL;
 559
 560         r = -ENOMEM;
 561         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
 562         if (!cpuid_entries)
 563                 goto out;
 564
 565         r = 0;
 566         for (i = 0; i < ARRAY_SIZE(param); i++) {
 567                 const struct kvm_cpuid_param *ent = &param[i];
 568
 569                 if (ent->qualifier && !ent->qualifier(ent))
 570                         continue;
 571
 572                 r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
 573                                 &nent, cpuid->nent, type);
 574
 575                 if (r)
 576                         goto out_free;
 577
 578                 if (!ent->has_leaf_count)
 579                         continue;
 580
 581                 limit = cpuid_entries[nent - 1].eax;
 582                 for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
 583                         r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
 584                                      &nent, cpuid->nent, type);
 585
 586                 if (r)
 587                         goto out_free;
 588         }
 589
 590         r = -EFAULT;
 591         if (copy_to_user(entries, cpuid_entries,
 592                          nent * sizeof(struct kvm_cpuid_entry2)))
 593                 goto out_free;
 594         cpuid->nent = nent;
 595         r = 0;
 596
 597 out_free:
 598         vfree(cpuid_entries);
 599 out:
 600         return r;
 601 }
 602
 603 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
 604 {
 605         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
 606         int j, nent = vcpu->arch.cpuid_nent;
 607
 608         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
 609         /* when no next entry is found, the current entry[i] is reselected */
 610         for (j = i + 1; ; j = (j + 1) % nent) {
 611                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
 612                 if (ej->function == e->function) {
 613                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
 614                         return j;
 615                 }
 616         }
 617         return 0; /* silence gcc, even though control never reaches here */
 618 }
 619
 620 /* find an entry with matching function, matching index (if needed), and that
 621  * should be read next (if it's stateful) */
 622 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
 623         u32 function, u32 index)
 624 {
 625         if (e->function != function)
 626                 return 0;
 627         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
 628                 return 0;
 629         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
 630             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
 631                 return 0;
 632         return 1;
 633 }
 634
 635 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
 636                                               u32 function, u32 index)
 637 {
 638         int i;
 639         struct kvm_cpuid_entry2 *best = NULL;
 640
 641         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 642                 struct kvm_cpuid_entry2 *e;
 643
 644                 e = &vcpu->arch.cpuid_entries[i];
 645                 if (is_matching_cpuid_entry(e, function, index)) {
 646                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
 647                                 move_to_next_stateful_cpuid_entry(vcpu, i);
 648                         best = e;
 649                         break;
 650                 }
 651         }
 652         return best;
 653 }
 654 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
 655
 656 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
 657 {
 658         struct kvm_cpuid_entry2 *best;
 659
 660         best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
 661         if (!best || best->eax < 0x80000008)
 662                 goto not_found;
 663         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
 664         if (best)
 665                 return best->eax & 0xff;
 666 not_found:
 667         return 36;
 668 }
 669
 670 /*
 671  * If no match is found, check whether we exceed the vCPU's limit
 672  * and return the content of the highest valid _standard_ leaf instead.
 673  * This is to satisfy the CPUID specification.
 674  */
 675 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
 676                                                   u32 function, u32 index)
 677 {
 678         struct kvm_cpuid_entry2 *maxlevel;
 679
 680         maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
 681         if (!maxlevel || maxlevel->eax >= function)
 682                 return NULL;
 683         if (function & 0x80000000) {
 684                 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
 685                 if (!maxlevel)
 686                         return NULL;
 687         }
 688         return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
 689 }
 690
 691 void kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
 692 {
 693         u32 function = *eax, index = *ecx;
 694         struct kvm_cpuid_entry2 *best;
 695
 696         best = kvm_find_cpuid_entry(vcpu, function, index);
 697
 698         if (!best)
 699                 best = check_cpuid_limit(vcpu, function, index);
 700
 701         if (best) {
 702                 *eax = best->eax;
 703                 *ebx = best->ebx;
 704                 *ecx = best->ecx;
 705                 *edx = best->edx;
 706         } else
 707                 *eax = *ebx = *ecx = *edx = 0;
 708 }
 709 EXPORT_SYMBOL_GPL(kvm_cpuid);
 710
 711 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
 712 {
 713         u32 function, eax, ebx, ecx, edx;
 714
 715         function = eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
 716         ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
 717         kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx);
 718         kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
 719         kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
 720         kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
 721         kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
 722         kvm_x86_ops->skip_emulated_instruction(vcpu);
 723         trace_kvm_cpuid(function, eax, ebx, ecx, edx);
 724 }
 725 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);