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, bool compacted)
  27 {
  28         int feature_bit = 0;
  29         u32 ret = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
  30
  31         xstate_bv &= XSTATE_EXTEND_MASK;
  32         while (xstate_bv) {
  33                 if (xstate_bv & 0x1) {
  34                         u32 eax, ebx, ecx, edx, offset;
  35                         cpuid_count(0xD, feature_bit, &eax, &ebx, &ecx, &edx);
  36                         offset = compacted ? ret : ebx;
  37                         ret = max(ret, offset + eax);
  38                 }
  39
  40                 xstate_bv >>= 1;
  41                 feature_bit++;
  42         }
  43
  44         return ret;
  45 }
  46
  47 u64 kvm_supported_xcr0(void)
  48 {
  49         u64 xcr0 = KVM_SUPPORTED_XCR0 & host_xcr0;
  50
  51         if (!kvm_x86_ops->mpx_supported())
  52                 xcr0 &= ~(XSTATE_BNDREGS | XSTATE_BNDCSR);
  53
  54         return xcr0;
  55 }
  56
  57 #define F(x) bit(X86_FEATURE_##x)
  58
  59 int kvm_update_cpuid(struct kvm_vcpu *vcpu)
  60 {
  61         struct kvm_cpuid_entry2 *best;
  62         struct kvm_lapic *apic = vcpu->arch.apic;
  63
  64         best = kvm_find_cpuid_entry(vcpu, 1, 0);
  65         if (!best)
  66                 return 0;
  67
  68         /* Update OSXSAVE bit */
  69         if (cpu_has_xsave && best->function == 0x1) {
  70                 best->ecx &= ~F(OSXSAVE);
  71                 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
  72                         best->ecx |= F(OSXSAVE);
  73         }
  74
  75         if (apic) {
  76                 if (best->ecx & F(TSC_DEADLINE_TIMER))
  77                         apic->lapic_timer.timer_mode_mask = 3 << 17;
  78                 else
  79                         apic->lapic_timer.timer_mode_mask = 1 << 17;
  80         }
  81
  82         best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
  83         if (!best) {
  84                 vcpu->arch.guest_supported_xcr0 = 0;
  85                 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
  86         } else {
  87                 vcpu->arch.guest_supported_xcr0 =
  88                         (best->eax | ((u64)best->edx << 32)) &
  89                         kvm_supported_xcr0();
  90                 vcpu->arch.guest_xstate_size = best->ebx =
  91                         xstate_required_size(vcpu->arch.xcr0, false);
  92         }
  93
  94         best = kvm_find_cpuid_entry(vcpu, 0xD, 1);
  95         if (best && (best->eax & (F(XSAVES) | F(XSAVEC))))
  96                 best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
  97
  98         /*
  99          * The existing code assumes virtual address is 48-bit in the canonical
 100          * address checks; exit if it is ever changed.
 101          */
 102         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
 103         if (best && ((best->eax & 0xff00) >> 8) != 48 &&
 104                 ((best->eax & 0xff00) >> 8) != 0)
 105                 return -EINVAL;
 106
 107         /* Update physical-address width */
 108         vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
 109
 110         kvm_pmu_cpuid_update(vcpu);
 111         return 0;
 112 }
 113
 114 static int is_efer_nx(void)
 115 {
 116         unsigned long long efer = 0;
 117
 118         rdmsrl_safe(MSR_EFER, &efer);
 119         return efer & EFER_NX;
 120 }
 121
 122 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
 123 {
 124         int i;
 125         struct kvm_cpuid_entry2 *e, *entry;
 126
 127         entry = NULL;
 128         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 129                 e = &vcpu->arch.cpuid_entries[i];
 130                 if (e->function == 0x80000001) {
 131                         entry = e;
 132                         break;
 133                 }
 134         }
 135         if (entry && (entry->edx & F(NX)) && !is_efer_nx()) {
 136                 entry->edx &= ~F(NX);
 137                 printk(KERN_INFO "kvm: guest NX capability removed\n");
 138         }
 139 }
 140
 141 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
 142 {
 143         struct kvm_cpuid_entry2 *best;
 144
 145         best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
 146         if (!best || best->eax < 0x80000008)
 147                 goto not_found;
 148         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
 149         if (best)
 150                 return best->eax & 0xff;
 151 not_found:
 152         return 36;
 153 }
 154 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr);
 155
 156 /* when an old userspace process fills a new kernel module */
 157 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
 158                              struct kvm_cpuid *cpuid,
 159                              struct kvm_cpuid_entry __user *entries)
 160 {
 161         int r, i;
 162         struct kvm_cpuid_entry *cpuid_entries;
 163
 164         r = -E2BIG;
 165         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 166                 goto out;
 167         r = -ENOMEM;
 168         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
 169         if (!cpuid_entries)
 170                 goto out;
 171         r = -EFAULT;
 172         if (copy_from_user(cpuid_entries, entries,
 173                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
 174                 goto out_free;
 175         for (i = 0; i < cpuid->nent; i++) {
 176                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
 177                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
 178                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
 179                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
 180                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
 181                 vcpu->arch.cpuid_entries[i].index = 0;
 182                 vcpu->arch.cpuid_entries[i].flags = 0;
 183                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
 184                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
 185                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
 186         }
 187         vcpu->arch.cpuid_nent = cpuid->nent;
 188         cpuid_fix_nx_cap(vcpu);
 189         kvm_apic_set_version(vcpu);
 190         kvm_x86_ops->cpuid_update(vcpu);
 191         r = kvm_update_cpuid(vcpu);
 192
 193 out_free:
 194         vfree(cpuid_entries);
 195 out:
 196         return r;
 197 }
 198
 199 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
 200                               struct kvm_cpuid2 *cpuid,
 201                               struct kvm_cpuid_entry2 __user *entries)
 202 {
 203         int r;
 204
 205         r = -E2BIG;
 206         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 207                 goto out;
 208         r = -EFAULT;
 209         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
 210                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
 211                 goto out;
 212         vcpu->arch.cpuid_nent = cpuid->nent;
 213         kvm_apic_set_version(vcpu);
 214         kvm_x86_ops->cpuid_update(vcpu);
 215         r = kvm_update_cpuid(vcpu);
 216 out:
 217         return r;
 218 }
 219
 220 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
 221                               struct kvm_cpuid2 *cpuid,
 222                               struct kvm_cpuid_entry2 __user *entries)
 223 {
 224         int r;
 225
 226         r = -E2BIG;
 227         if (cpuid->nent < vcpu->arch.cpuid_nent)
 228                 goto out;
 229         r = -EFAULT;
 230         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
 231                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
 232                 goto out;
 233         return 0;
 234
 235 out:
 236         cpuid->nent = vcpu->arch.cpuid_nent;
 237         return r;
 238 }
 239
 240 static void cpuid_mask(u32 *word, int wordnum)
 241 {
 242         *word &= boot_cpu_data.x86_capability[wordnum];
 243 }
 244
 245 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 246                            u32 index)
 247 {
 248         entry->function = function;
 249         entry->index = index;
 250         cpuid_count(entry->function, entry->index,
 251                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
 252         entry->flags = 0;
 253 }
 254
 255 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry,
 256                                    u32 func, u32 index, int *nent, int maxnent)
 257 {
 258         switch (func) {
 259         case 0:
 260                 entry->eax = 1;         /* only one leaf currently */
 261                 ++*nent;
 262                 break;
 263         case 1:
 264                 entry->ecx = F(MOVBE);
 265                 ++*nent;
 266                 break;
 267         default:
 268                 break;
 269         }
 270
 271         entry->function = func;
 272         entry->index = index;
 273
 274         return 0;
 275 }
 276
 277 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 278                                  u32 index, int *nent, int maxnent)
 279 {
 280         int r;
 281         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
 282 #ifdef CONFIG_X86_64
 283         unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
 284                                 ? F(GBPAGES) : 0;
 285         unsigned f_lm = F(LM);
 286 #else
 287         unsigned f_gbpages = 0;
 288         unsigned f_lm = 0;
 289 #endif
 290         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
 291         unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
 292         unsigned f_mpx = kvm_x86_ops->mpx_supported() ? F(MPX) : 0;
 293         unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0;
 294
 295         /* cpuid 1.edx */
 296         const u32 kvm_supported_word0_x86_features =
 297                 F(FPU) | F(VME) | F(DE) | F(PSE) |
 298                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
 299                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
 300                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
 301                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
 302                 0 /* Reserved, DS, ACPI */ | F(MMX) |
 303                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
 304                 0 /* HTT, TM, Reserved, PBE */;
 305         /* cpuid 0x80000001.edx */
 306         const u32 kvm_supported_word1_x86_features =
 307                 F(FPU) | F(VME) | F(DE) | F(PSE) |
 308                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
 309                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
 310                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
 311                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
 312                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
 313                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
 314                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
 315         /* cpuid 1.ecx */
 316         const u32 kvm_supported_word4_x86_features =
 317                 /* NOTE: MONITOR (and MWAIT) are emulated as NOP,
 318                  * but *not* advertised to guests via CPUID ! */
 319                 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
 320                 0 /* DS-CPL, VMX, SMX, EST */ |
 321                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
 322                 F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
 323                 F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
 324                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
 325                 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
 326                 F(F16C) | F(RDRAND);
 327         /* cpuid 0x80000001.ecx */
 328         const u32 kvm_supported_word6_x86_features =
 329                 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
 330                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
 331                 F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
 332                 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
 333
 334         /* cpuid 0xC0000001.edx */
 335         const u32 kvm_supported_word5_x86_features =
 336                 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
 337                 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
 338                 F(PMM) | F(PMM_EN);
 339
 340         /* cpuid 7.0.ebx */
 341         const u32 kvm_supported_word9_x86_features =
 342                 F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
 343                 F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
 344                 F(ADX) | F(SMAP) | F(AVX512F) | F(AVX512PF) | F(AVX512ER) |
 345                 F(AVX512CD);
 346
 347         /* cpuid 0xD.1.eax */
 348         const u32 kvm_supported_word10_x86_features =
 349                 F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves;
 350
 351         /* all calls to cpuid_count() should be made on the same cpu */
 352         get_cpu();
 353
 354         r = -E2BIG;
 355
 356         if (*nent >= maxnent)
 357                 goto out;
 358
 359         do_cpuid_1_ent(entry, function, index);
 360         ++*nent;
 361
 362         switch (function) {
 363         case 0:
 364                 entry->eax = min(entry->eax, (u32)0xd);
 365                 break;
 366         case 1:
 367                 entry->edx &= kvm_supported_word0_x86_features;
 368                 cpuid_mask(&entry->edx, 0);
 369                 entry->ecx &= kvm_supported_word4_x86_features;
 370                 cpuid_mask(&entry->ecx, 4);
 371                 /* we support x2apic emulation even if host does not support
 372                  * it since we emulate x2apic in software */
 373                 entry->ecx |= F(X2APIC);
 374                 break;
 375         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
 376          * may return different values. This forces us to get_cpu() before
 377          * issuing the first command, and also to emulate this annoying behavior
 378          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
 379         case 2: {
 380                 int t, times = entry->eax & 0xff;
 381
 382                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 383                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
 384                 for (t = 1; t < times; ++t) {
 385                         if (*nent >= maxnent)
 386                                 goto out;
 387
 388                         do_cpuid_1_ent(&entry[t], function, 0);
 389                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 390                         ++*nent;
 391                 }
 392                 break;
 393         }
 394         /* function 4 has additional index. */
 395         case 4: {
 396                 int i, cache_type;
 397
 398                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 399                 /* read more entries until cache_type is zero */
 400                 for (i = 1; ; ++i) {
 401                         if (*nent >= maxnent)
 402                                 goto out;
 403
 404                         cache_type = entry[i - 1].eax & 0x1f;
 405                         if (!cache_type)
 406                                 break;
 407                         do_cpuid_1_ent(&entry[i], function, i);
 408                         entry[i].flags |=
 409                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 410                         ++*nent;
 411                 }
 412                 break;
 413         }
 414         case 7: {
 415                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 416                 /* Mask ebx against host capability word 9 */
 417                 if (index == 0) {
 418                         entry->ebx &= kvm_supported_word9_x86_features;
 419                         cpuid_mask(&entry->ebx, 9);
 420                         // TSC_ADJUST is emulated
 421                         entry->ebx |= F(TSC_ADJUST);
 422                 } else
 423                         entry->ebx = 0;
 424                 entry->eax = 0;
 425                 entry->ecx = 0;
 426                 entry->edx = 0;
 427                 break;
 428         }
 429         case 9:
 430                 break;
 431         case 0xa: { /* Architectural Performance Monitoring */
 432                 struct x86_pmu_capability cap;
 433                 union cpuid10_eax eax;
 434                 union cpuid10_edx edx;
 435
 436                 perf_get_x86_pmu_capability(&cap);
 437
 438                 /*
 439                  * Only support guest architectural pmu on a host
 440                  * with architectural pmu.
 441                  */
 442                 if (!cap.version)
 443                         memset(&cap, 0, sizeof(cap));
 444
 445                 eax.split.version_id = min(cap.version, 2);
 446                 eax.split.num_counters = cap.num_counters_gp;
 447                 eax.split.bit_width = cap.bit_width_gp;
 448                 eax.split.mask_length = cap.events_mask_len;
 449
 450                 edx.split.num_counters_fixed = cap.num_counters_fixed;
 451                 edx.split.bit_width_fixed = cap.bit_width_fixed;
 452                 edx.split.reserved = 0;
 453
 454                 entry->eax = eax.full;
 455                 entry->ebx = cap.events_mask;
 456                 entry->ecx = 0;
 457                 entry->edx = edx.full;
 458                 break;
 459         }
 460         /* function 0xb has additional index. */
 461         case 0xb: {
 462                 int i, level_type;
 463
 464                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 465                 /* read more entries until level_type is zero */
 466                 for (i = 1; ; ++i) {
 467                         if (*nent >= maxnent)
 468                                 goto out;
 469
 470                         level_type = entry[i - 1].ecx & 0xff00;
 471                         if (!level_type)
 472                                 break;
 473                         do_cpuid_1_ent(&entry[i], function, i);
 474                         entry[i].flags |=
 475                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 476                         ++*nent;
 477                 }
 478                 break;
 479         }
 480         case 0xd: {
 481                 int idx, i;
 482                 u64 supported = kvm_supported_xcr0();
 483
 484                 entry->eax &= supported;
 485                 entry->ebx = xstate_required_size(supported, false);
 486                 entry->ecx = entry->ebx;
 487                 entry->edx &= supported >> 32;
 488                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 489                 if (!supported)
 490                         break;
 491
 492                 for (idx = 1, i = 1; idx < 64; ++idx) {
 493                         u64 mask = ((u64)1 << idx);
 494                         if (*nent >= maxnent)
 495                                 goto out;
 496
 497                         do_cpuid_1_ent(&entry[i], function, idx);
 498                         if (idx == 1) {
 499                                 entry[i].eax &= kvm_supported_word10_x86_features;
 500                                 entry[i].ebx = 0;
 501                                 if (entry[i].eax & (F(XSAVES)|F(XSAVEC)))
 502                                         entry[i].ebx =
 503                                                 xstate_required_size(supported,
 504                                                                      true);
 505                         } else {
 506                                 if (entry[i].eax == 0 || !(supported & mask))
 507                                         continue;
 508                                 if (WARN_ON_ONCE(entry[i].ecx & 1))
 509                                         continue;
 510                         }
 511                         entry[i].ecx = 0;
 512                         entry[i].edx = 0;
 513                         entry[i].flags |=
 514                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 515                         ++*nent;
 516                         ++i;
 517                 }
 518                 break;
 519         }
 520         case KVM_CPUID_SIGNATURE: {
 521                 static const char signature[12] = "KVMKVMKVM\0\0";
 522                 const u32 *sigptr = (const u32 *)signature;
 523                 entry->eax = KVM_CPUID_FEATURES;
 524                 entry->ebx = sigptr[0];
 525                 entry->ecx = sigptr[1];
 526                 entry->edx = sigptr[2];
 527                 break;
 528         }
 529         case KVM_CPUID_FEATURES:
 530                 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
 531                              (1 << KVM_FEATURE_NOP_IO_DELAY) |
 532                              (1 << KVM_FEATURE_CLOCKSOURCE2) |
 533                              (1 << KVM_FEATURE_ASYNC_PF) |
 534                              (1 << KVM_FEATURE_PV_EOI) |
 535                              (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
 536                              (1 << KVM_FEATURE_PV_UNHALT);
 537
 538                 if (sched_info_on())
 539                         entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
 540
 541                 entry->ebx = 0;
 542                 entry->ecx = 0;
 543                 entry->edx = 0;
 544                 break;
 545         case 0x80000000:
 546                 entry->eax = min(entry->eax, 0x8000001a);
 547                 break;
 548         case 0x80000001:
 549                 entry->edx &= kvm_supported_word1_x86_features;
 550                 cpuid_mask(&entry->edx, 1);
 551                 entry->ecx &= kvm_supported_word6_x86_features;
 552                 cpuid_mask(&entry->ecx, 6);
 553                 break;
 554         case 0x80000007: /* Advanced power management */
 555                 /* invariant TSC is CPUID.80000007H:EDX[8] */
 556                 entry->edx &= (1 << 8);
 557                 /* mask against host */
 558                 entry->edx &= boot_cpu_data.x86_power;
 559                 entry->eax = entry->ebx = entry->ecx = 0;
 560                 break;
 561         case 0x80000008: {
 562                 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
 563                 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
 564                 unsigned phys_as = entry->eax & 0xff;
 565
 566                 if (!g_phys_as)
 567                         g_phys_as = phys_as;
 568                 entry->eax = g_phys_as | (virt_as << 8);
 569                 entry->ebx = entry->edx = 0;
 570                 break;
 571         }
 572         case 0x80000019:
 573                 entry->ecx = entry->edx = 0;
 574                 break;
 575         case 0x8000001a:
 576                 break;
 577         case 0x8000001d:
 578                 break;
 579         /*Add support for Centaur's CPUID instruction*/
 580         case 0xC0000000:
 581                 /*Just support up to 0xC0000004 now*/
 582                 entry->eax = min(entry->eax, 0xC0000004);
 583                 break;
 584         case 0xC0000001:
 585                 entry->edx &= kvm_supported_word5_x86_features;
 586                 cpuid_mask(&entry->edx, 5);
 587                 break;
 588         case 3: /* Processor serial number */
 589         case 5: /* MONITOR/MWAIT */
 590         case 6: /* Thermal management */
 591         case 0xC0000002:
 592         case 0xC0000003:
 593         case 0xC0000004:
 594         default:
 595                 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
 596                 break;
 597         }
 598
 599         kvm_x86_ops->set_supported_cpuid(function, entry);
 600
 601         r = 0;
 602
 603 out:
 604         put_cpu();
 605
 606         return r;
 607 }
 608
 609 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
 610                         u32 idx, int *nent, int maxnent, unsigned int type)
 611 {
 612         if (type == KVM_GET_EMULATED_CPUID)
 613                 return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
 614
 615         return __do_cpuid_ent(entry, func, idx, nent, maxnent);
 616 }
 617
 618 #undef F
 619
 620 struct kvm_cpuid_param {
 621         u32 func;
 622         u32 idx;
 623         bool has_leaf_count;
 624         bool (*qualifier)(const struct kvm_cpuid_param *param);
 625 };
 626
 627 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
 628 {
 629         return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
 630 }
 631
 632 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
 633                                  __u32 num_entries, unsigned int ioctl_type)
 634 {
 635         int i;
 636         __u32 pad[3];
 637
 638         if (ioctl_type != KVM_GET_EMULATED_CPUID)
 639                 return false;
 640
 641         /*
 642          * We want to make sure that ->padding is being passed clean from
 643          * userspace in case we want to use it for something in the future.
 644          *
 645          * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
 646          * have to give ourselves satisfied only with the emulated side. /me
 647          * sheds a tear.
 648          */
 649         for (i = 0; i < num_entries; i++) {
 650                 if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
 651                         return true;
 652
 653                 if (pad[0] || pad[1] || pad[2])
 654                         return true;
 655         }
 656         return false;
 657 }
 658
 659 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
 660                             struct kvm_cpuid_entry2 __user *entries,
 661                             unsigned int type)
 662 {
 663         struct kvm_cpuid_entry2 *cpuid_entries;
 664         int limit, nent = 0, r = -E2BIG, i;
 665         u32 func;
 666         static const struct kvm_cpuid_param param[] = {
 667                 { .func = 0, .has_leaf_count = true },
 668                 { .func = 0x80000000, .has_leaf_count = true },
 669                 { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
 670                 { .func = KVM_CPUID_SIGNATURE },
 671                 { .func = KVM_CPUID_FEATURES },
 672         };
 673
 674         if (cpuid->nent < 1)
 675                 goto out;
 676         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 677                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
 678
 679         if (sanity_check_entries(entries, cpuid->nent, type))
 680                 return -EINVAL;
 681
 682         r = -ENOMEM;
 683         cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
 684         if (!cpuid_entries)
 685                 goto out;
 686
 687         r = 0;
 688         for (i = 0; i < ARRAY_SIZE(param); i++) {
 689                 const struct kvm_cpuid_param *ent = &param[i];
 690
 691                 if (ent->qualifier && !ent->qualifier(ent))
 692                         continue;
 693
 694                 r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
 695                                 &nent, cpuid->nent, type);
 696
 697                 if (r)
 698                         goto out_free;
 699
 700                 if (!ent->has_leaf_count)
 701                         continue;
 702
 703                 limit = cpuid_entries[nent - 1].eax;
 704                 for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
 705                         r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
 706                                      &nent, cpuid->nent, type);
 707
 708                 if (r)
 709                         goto out_free;
 710         }
 711
 712         r = -EFAULT;
 713         if (copy_to_user(entries, cpuid_entries,
 714                          nent * sizeof(struct kvm_cpuid_entry2)))
 715                 goto out_free;
 716         cpuid->nent = nent;
 717         r = 0;
 718
 719 out_free:
 720         vfree(cpuid_entries);
 721 out:
 722         return r;
 723 }
 724
 725 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
 726 {
 727         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
 728         int j, nent = vcpu->arch.cpuid_nent;
 729
 730         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
 731         /* when no next entry is found, the current entry[i] is reselected */
 732         for (j = i + 1; ; j = (j + 1) % nent) {
 733                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
 734                 if (ej->function == e->function) {
 735                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
 736                         return j;
 737                 }
 738         }
 739         return 0; /* silence gcc, even though control never reaches here */
 740 }
 741
 742 /* find an entry with matching function, matching index (if needed), and that
 743  * should be read next (if it's stateful) */
 744 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
 745         u32 function, u32 index)
 746 {
 747         if (e->function != function)
 748                 return 0;
 749         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
 750                 return 0;
 751         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
 752             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
 753                 return 0;
 754         return 1;
 755 }
 756
 757 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
 758                                               u32 function, u32 index)
 759 {
 760         int i;
 761         struct kvm_cpuid_entry2 *best = NULL;
 762
 763         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 764                 struct kvm_cpuid_entry2 *e;
 765
 766                 e = &vcpu->arch.cpuid_entries[i];
 767                 if (is_matching_cpuid_entry(e, function, index)) {
 768                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
 769                                 move_to_next_stateful_cpuid_entry(vcpu, i);
 770                         best = e;
 771                         break;
 772                 }
 773         }
 774         return best;
 775 }
 776 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
 777
 778 /*
 779  * If no match is found, check whether we exceed the vCPU's limit
 780  * and return the content of the highest valid _standard_ leaf instead.
 781  * This is to satisfy the CPUID specification.
 782  */
 783 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
 784                                                   u32 function, u32 index)
 785 {
 786         struct kvm_cpuid_entry2 *maxlevel;
 787
 788         maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
 789         if (!maxlevel || maxlevel->eax >= function)
 790                 return NULL;
 791         if (function & 0x80000000) {
 792                 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
 793                 if (!maxlevel)
 794                         return NULL;
 795         }
 796         return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
 797 }
 798
 799 void kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
 800 {
 801         u32 function = *eax, index = *ecx;
 802         struct kvm_cpuid_entry2 *best;
 803
 804         best = kvm_find_cpuid_entry(vcpu, function, index);
 805
 806         if (!best)
 807                 best = check_cpuid_limit(vcpu, function, index);
 808
 809         /*
 810          * Perfmon not yet supported for L2 guest.
 811          */
 812         if (is_guest_mode(vcpu) && function == 0xa)
 813                 best = NULL;
 814
 815         if (best) {
 816                 *eax = best->eax;
 817                 *ebx = best->ebx;
 818                 *ecx = best->ecx;
 819                 *edx = best->edx;
 820         } else
 821                 *eax = *ebx = *ecx = *edx = 0;
 822         trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx);
 823 }
 824 EXPORT_SYMBOL_GPL(kvm_cpuid);
 825
 826 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
 827 {
 828         u32 function, eax, ebx, ecx, edx;
 829
 830         function = eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
 831         ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
 832         kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx);
 833         kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
 834         kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
 835         kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
 836         kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
 837         kvm_x86_ops->skip_emulated_instruction(vcpu);
 838 }
 839 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);