arch/x86/kvm/cpuid.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  * cpuid support routines
   4  *
   5  * derived from arch/x86/kvm/x86.c
   6  *
   7  * Copyright 2011 Red Hat, Inc. and/or its affiliates.
   8  * Copyright IBM Corporation, 2008
   9  *
  10  * This work is licensed under the terms of the GNU GPL, version 2.  See
  11  * the COPYING file in the top-level directory.
  12  *
  13  */
  14
  15 #include <linux/kvm_host.h>
  16 #include <linux/export.h>
  17 #include <linux/vmalloc.h>
  18 #include <linux/uaccess.h>
  19 #include <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 = NULL;
 185
 186         r = -E2BIG;
 187         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 188                 goto out;
 189         r = -ENOMEM;
 190         if (cpuid->nent) {
 191                 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) *
 192                                         cpuid->nent);
 193                 if (!cpuid_entries)
 194                         goto out;
 195                 r = -EFAULT;
 196                 if (copy_from_user(cpuid_entries, entries,
 197                                    cpuid->nent * sizeof(struct kvm_cpuid_entry)))
 198                         goto out;
 199         }
 200         for (i = 0; i < cpuid->nent; i++) {
 201                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
 202                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
 203                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
 204                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
 205                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
 206                 vcpu->arch.cpuid_entries[i].index = 0;
 207                 vcpu->arch.cpuid_entries[i].flags = 0;
 208                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
 209                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
 210                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
 211         }
 212         vcpu->arch.cpuid_nent = cpuid->nent;
 213         cpuid_fix_nx_cap(vcpu);
 214         kvm_apic_set_version(vcpu);
 215         kvm_x86_ops->cpuid_update(vcpu);
 216         r = kvm_update_cpuid(vcpu);
 217
 218 out:
 219         vfree(cpuid_entries);
 220         return r;
 221 }
 222
 223 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
 224                               struct kvm_cpuid2 *cpuid,
 225                               struct kvm_cpuid_entry2 __user *entries)
 226 {
 227         int r;
 228
 229         r = -E2BIG;
 230         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 231                 goto out;
 232         r = -EFAULT;
 233         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
 234                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
 235                 goto out;
 236         vcpu->arch.cpuid_nent = cpuid->nent;
 237         kvm_apic_set_version(vcpu);
 238         kvm_x86_ops->cpuid_update(vcpu);
 239         r = kvm_update_cpuid(vcpu);
 240 out:
 241         return r;
 242 }
 243
 244 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
 245                               struct kvm_cpuid2 *cpuid,
 246                               struct kvm_cpuid_entry2 __user *entries)
 247 {
 248         int r;
 249
 250         r = -E2BIG;
 251         if (cpuid->nent < vcpu->arch.cpuid_nent)
 252                 goto out;
 253         r = -EFAULT;
 254         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
 255                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
 256                 goto out;
 257         return 0;
 258
 259 out:
 260         cpuid->nent = vcpu->arch.cpuid_nent;
 261         return r;
 262 }
 263
 264 static void cpuid_mask(u32 *word, int wordnum)
 265 {
 266         *word &= boot_cpu_data.x86_capability[wordnum];
 267 }
 268
 269 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 270                            u32 index)
 271 {
 272         entry->function = function;
 273         entry->index = index;
 274         cpuid_count(entry->function, entry->index,
 275                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
 276         entry->flags = 0;
 277 }
 278
 279 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry,
 280                                    u32 func, u32 index, int *nent, int maxnent)
 281 {
 282         switch (func) {
 283         case 0:
 284                 entry->eax = 1;         /* only one leaf currently */
 285                 ++*nent;
 286                 break;
 287         case 1:
 288                 entry->ecx = F(MOVBE);
 289                 ++*nent;
 290                 break;
 291         default:
 292                 break;
 293         }
 294
 295         entry->function = func;
 296         entry->index = index;
 297
 298         return 0;
 299 }
 300
 301 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 302                                  u32 index, int *nent, int maxnent)
 303 {
 304         int r;
 305         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
 306 #ifdef CONFIG_X86_64
 307         unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
 308                                 ? F(GBPAGES) : 0;
 309         unsigned f_lm = F(LM);
 310 #else
 311         unsigned f_gbpages = 0;
 312         unsigned f_lm = 0;
 313 #endif
 314         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
 315         unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
 316         unsigned f_mpx = kvm_mpx_supported() ? F(MPX) : 0;
 317         unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0;
 318
 319         /* cpuid 1.edx */
 320         const u32 kvm_cpuid_1_edx_x86_features =
 321                 F(FPU) | F(VME) | F(DE) | F(PSE) |
 322                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
 323                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
 324                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
 325                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
 326                 0 /* Reserved, DS, ACPI */ | F(MMX) |
 327                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
 328                 0 /* HTT, TM, Reserved, PBE */;
 329         /* cpuid 0x80000001.edx */
 330         const u32 kvm_cpuid_8000_0001_edx_x86_features =
 331                 F(FPU) | F(VME) | F(DE) | F(PSE) |
 332                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
 333                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
 334                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
 335                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
 336                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
 337                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
 338                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
 339         /* cpuid 1.ecx */
 340         const u32 kvm_cpuid_1_ecx_x86_features =
 341                 /* NOTE: MONITOR (and MWAIT) are emulated as NOP,
 342                  * but *not* advertised to guests via CPUID ! */
 343                 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
 344                 0 /* DS-CPL, VMX, SMX, EST */ |
 345                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
 346                 F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
 347                 F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
 348                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
 349                 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
 350                 F(F16C) | F(RDRAND);
 351         /* cpuid 0x80000001.ecx */
 352         const u32 kvm_cpuid_8000_0001_ecx_x86_features =
 353                 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
 354                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
 355                 F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
 356                 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
 357
 358         /* cpuid 0xC0000001.edx */
 359         const u32 kvm_cpuid_C000_0001_edx_x86_features =
 360                 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
 361                 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
 362                 F(PMM) | F(PMM_EN);
 363
 364         /* cpuid 7.0.ebx */
 365         const u32 kvm_cpuid_7_0_ebx_x86_features =
 366                 F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
 367                 F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
 368                 F(ADX) | F(SMAP) | F(AVX512F) | F(AVX512PF) | F(AVX512ER) |
 369                 F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) |
 370                 F(AVX512BW) | F(AVX512VL);
 371
 372         /* cpuid 0xD.1.eax */
 373         const u32 kvm_cpuid_D_1_eax_x86_features =
 374                 F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves;
 375
 376         /* cpuid 7.0.ecx*/
 377         const u32 kvm_cpuid_7_0_ecx_x86_features = F(PKU) | 0 /*OSPKE*/;
 378
 379         /* all calls to cpuid_count() should be made on the same cpu */
 380         get_cpu();
 381
 382         r = -E2BIG;
 383
 384         if (*nent >= maxnent)
 385                 goto out;
 386
 387         do_cpuid_1_ent(entry, function, index);
 388         ++*nent;
 389
 390         switch (function) {
 391         case 0:
 392                 entry->eax = min(entry->eax, (u32)0xd);
 393                 break;
 394         case 1:
 395                 entry->edx &= kvm_cpuid_1_edx_x86_features;
 396                 cpuid_mask(&entry->edx, CPUID_1_EDX);
 397                 entry->ecx &= kvm_cpuid_1_ecx_x86_features;
 398                 cpuid_mask(&entry->ecx, CPUID_1_ECX);
 399                 /* we support x2apic emulation even if host does not support
 400                  * it since we emulate x2apic in software */
 401                 entry->ecx |= F(X2APIC);
 402                 break;
 403         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
 404          * may return different values. This forces us to get_cpu() before
 405          * issuing the first command, and also to emulate this annoying behavior
 406          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
 407         case 2: {
 408                 int t, times = entry->eax & 0xff;
 409
 410                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 411                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
 412                 for (t = 1; t < times; ++t) {
 413                         if (*nent >= maxnent)
 414                                 goto out;
 415
 416                         do_cpuid_1_ent(&entry[t], function, 0);
 417                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 418                         ++*nent;
 419                 }
 420                 break;
 421         }
 422         /* function 4 has additional index. */
 423         case 4: {
 424                 int i, cache_type;
 425
 426                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 427                 /* read more entries until cache_type is zero */
 428                 for (i = 1; ; ++i) {
 429                         if (*nent >= maxnent)
 430                                 goto out;
 431
 432                         cache_type = entry[i - 1].eax & 0x1f;
 433                         if (!cache_type)
 434                                 break;
 435                         do_cpuid_1_ent(&entry[i], function, i);
 436                         entry[i].flags |=
 437                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 438                         ++*nent;
 439                 }
 440                 break;
 441         }
 442         case 6: /* Thermal management */
 443                 entry->eax = 0x4; /* allow ARAT */
 444                 entry->ebx = 0;
 445                 entry->ecx = 0;
 446                 entry->edx = 0;
 447                 break;
 448         case 7: {
 449                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 450                 /* Mask ebx against host capability word 9 */
 451                 if (index == 0) {
 452                         entry->ebx &= kvm_cpuid_7_0_ebx_x86_features;
 453                         cpuid_mask(&entry->ebx, CPUID_7_0_EBX);
 454                         // TSC_ADJUST is emulated
 455                         entry->ebx |= F(TSC_ADJUST);
 456                         entry->ecx &= kvm_cpuid_7_0_ecx_x86_features;
 457                         cpuid_mask(&entry->ecx, CPUID_7_ECX);
 458                         /* PKU is not yet implemented for shadow paging. */
 459                         if (!tdp_enabled)
 460                                 entry->ecx &= ~F(PKU);
 461                 } else {
 462                         entry->ebx = 0;
 463                         entry->ecx = 0;
 464                 }
 465                 entry->eax = 0;
 466                 entry->edx = 0;
 467                 break;
 468         }
 469         case 9:
 470                 break;
 471         case 0xa: { /* Architectural Performance Monitoring */
 472                 struct x86_pmu_capability cap;
 473                 union cpuid10_eax eax;
 474                 union cpuid10_edx edx;
 475
 476                 perf_get_x86_pmu_capability(&cap);
 477
 478                 /*
 479                  * Only support guest architectural pmu on a host
 480                  * with architectural pmu.
 481                  */
 482                 if (!cap.version)
 483                         memset(&cap, 0, sizeof(cap));
 484
 485                 eax.split.version_id = min(cap.version, 2);
 486                 eax.split.num_counters = cap.num_counters_gp;
 487                 eax.split.bit_width = cap.bit_width_gp;
 488                 eax.split.mask_length = cap.events_mask_len;
 489
 490                 edx.split.num_counters_fixed = cap.num_counters_fixed;
 491                 edx.split.bit_width_fixed = cap.bit_width_fixed;
 492                 edx.split.reserved = 0;
 493
 494                 entry->eax = eax.full;
 495                 entry->ebx = cap.events_mask;
 496                 entry->ecx = 0;
 497                 entry->edx = edx.full;
 498                 break;
 499         }
 500         /* function 0xb has additional index. */
 501         case 0xb: {
 502                 int i, level_type;
 503
 504                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 505                 /* read more entries until level_type is zero */
 506                 for (i = 1; ; ++i) {
 507                         if (*nent >= maxnent)
 508                                 goto out;
 509
 510                         level_type = entry[i - 1].ecx & 0xff00;
 511                         if (!level_type)
 512                                 break;
 513                         do_cpuid_1_ent(&entry[i], function, i);
 514                         entry[i].flags |=
 515                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 516                         ++*nent;
 517                 }
 518                 break;
 519         }
 520         case 0xd: {
 521                 int idx, i;
 522                 u64 supported = kvm_supported_xcr0();
 523
 524                 entry->eax &= supported;
 525                 entry->ebx = xstate_required_size(supported, false);
 526                 entry->ecx = entry->ebx;
 527                 entry->edx &= supported >> 32;
 528                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 529                 if (!supported)
 530                         break;
 531
 532                 for (idx = 1, i = 1; idx < 64; ++idx) {
 533                         u64 mask = ((u64)1 << idx);
 534                         if (*nent >= maxnent)
 535                                 goto out;
 536
 537                         do_cpuid_1_ent(&entry[i], function, idx);
 538                         if (idx == 1) {
 539                                 entry[i].eax &= kvm_cpuid_D_1_eax_x86_features;
 540                                 cpuid_mask(&entry[i].eax, CPUID_D_1_EAX);
 541                                 entry[i].ebx = 0;
 542                                 if (entry[i].eax & (F(XSAVES)|F(XSAVEC)))
 543                                         entry[i].ebx =
 544                                                 xstate_required_size(supported,
 545                                                                      true);
 546                         } else {
 547                                 if (entry[i].eax == 0 || !(supported & mask))
 548                                         continue;
 549                                 if (WARN_ON_ONCE(entry[i].ecx & 1))
 550                                         continue;
 551                         }
 552                         entry[i].ecx = 0;
 553                         entry[i].edx = 0;
 554                         entry[i].flags |=
 555                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 556                         ++*nent;
 557                         ++i;
 558                 }
 559                 break;
 560         }
 561         case KVM_CPUID_SIGNATURE: {
 562                 static const char signature[12] = "KVMKVMKVM\0\0";
 563                 const u32 *sigptr = (const u32 *)signature;
 564                 entry->eax = KVM_CPUID_FEATURES;
 565                 entry->ebx = sigptr[0];
 566                 entry->ecx = sigptr[1];
 567                 entry->edx = sigptr[2];
 568                 break;
 569         }
 570         case KVM_CPUID_FEATURES:
 571                 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
 572                              (1 << KVM_FEATURE_NOP_IO_DELAY) |
 573                              (1 << KVM_FEATURE_CLOCKSOURCE2) |
 574                              (1 << KVM_FEATURE_ASYNC_PF) |
 575                              (1 << KVM_FEATURE_PV_EOI) |
 576                              (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
 577                              (1 << KVM_FEATURE_PV_UNHALT);
 578
 579                 if (sched_info_on())
 580                         entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
 581
 582                 entry->ebx = 0;
 583                 entry->ecx = 0;
 584                 entry->edx = 0;
 585                 break;
 586         case 0x80000000:
 587                 entry->eax = min(entry->eax, 0x8000001a);
 588                 break;
 589         case 0x80000001:
 590                 entry->edx &= kvm_cpuid_8000_0001_edx_x86_features;
 591                 cpuid_mask(&entry->edx, CPUID_8000_0001_EDX);
 592                 entry->ecx &= kvm_cpuid_8000_0001_ecx_x86_features;
 593                 cpuid_mask(&entry->ecx, CPUID_8000_0001_ECX);
 594                 break;
 595         case 0x80000007: /* Advanced power management */
 596                 /* invariant TSC is CPUID.80000007H:EDX[8] */
 597                 entry->edx &= (1 << 8);
 598                 /* mask against host */
 599                 entry->edx &= boot_cpu_data.x86_power;
 600                 entry->eax = entry->ebx = entry->ecx = 0;
 601                 break;
 602         case 0x80000008: {
 603                 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
 604                 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
 605                 unsigned phys_as = entry->eax & 0xff;
 606
 607                 if (!g_phys_as)
 608                         g_phys_as = phys_as;
 609                 entry->eax = g_phys_as | (virt_as << 8);
 610                 entry->ebx = entry->edx = 0;
 611                 break;
 612         }
 613         case 0x80000019:
 614                 entry->ecx = entry->edx = 0;
 615                 break;
 616         case 0x8000001a:
 617                 break;
 618         case 0x8000001d:
 619                 break;
 620         /*Add support for Centaur's CPUID instruction*/
 621         case 0xC0000000:
 622                 /*Just support up to 0xC0000004 now*/
 623                 entry->eax = min(entry->eax, 0xC0000004);
 624                 break;
 625         case 0xC0000001:
 626                 entry->edx &= kvm_cpuid_C000_0001_edx_x86_features;
 627                 cpuid_mask(&entry->edx, CPUID_C000_0001_EDX);
 628                 break;
 629         case 3: /* Processor serial number */
 630         case 5: /* MONITOR/MWAIT */
 631         case 0xC0000002:
 632         case 0xC0000003:
 633         case 0xC0000004:
 634         default:
 635                 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
 636                 break;
 637         }
 638
 639         kvm_x86_ops->set_supported_cpuid(function, entry);
 640
 641         r = 0;
 642
 643 out:
 644         put_cpu();
 645
 646         return r;
 647 }
 648
 649 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
 650                         u32 idx, int *nent, int maxnent, unsigned int type)
 651 {
 652         if (type == KVM_GET_EMULATED_CPUID)
 653                 return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
 654
 655         return __do_cpuid_ent(entry, func, idx, nent, maxnent);
 656 }
 657
 658 #undef F
 659
 660 struct kvm_cpuid_param {
 661         u32 func;
 662         u32 idx;
 663         bool has_leaf_count;
 664         bool (*qualifier)(const struct kvm_cpuid_param *param);
 665 };
 666
 667 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
 668 {
 669         return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
 670 }
 671
 672 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
 673                                  __u32 num_entries, unsigned int ioctl_type)
 674 {
 675         int i;
 676         __u32 pad[3];
 677
 678         if (ioctl_type != KVM_GET_EMULATED_CPUID)
 679                 return false;
 680
 681         /*
 682          * We want to make sure that ->padding is being passed clean from
 683          * userspace in case we want to use it for something in the future.
 684          *
 685          * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
 686          * have to give ourselves satisfied only with the emulated side. /me
 687          * sheds a tear.
 688          */
 689         for (i = 0; i < num_entries; i++) {
 690                 if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
 691                         return true;
 692
 693                 if (pad[0] || pad[1] || pad[2])
 694                         return true;
 695         }
 696         return false;
 697 }
 698
 699 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
 700                             struct kvm_cpuid_entry2 __user *entries,
 701                             unsigned int type)
 702 {
 703         struct kvm_cpuid_entry2 *cpuid_entries;
 704         int limit, nent = 0, r = -E2BIG, i;
 705         u32 func;
 706         static const struct kvm_cpuid_param param[] = {
 707                 { .func = 0, .has_leaf_count = true },
 708                 { .func = 0x80000000, .has_leaf_count = true },
 709                 { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
 710                 { .func = KVM_CPUID_SIGNATURE },
 711                 { .func = KVM_CPUID_FEATURES },
 712         };
 713
 714         if (cpuid->nent < 1)
 715                 goto out;
 716         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 717                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
 718
 719         if (sanity_check_entries(entries, cpuid->nent, type))
 720                 return -EINVAL;
 721
 722         r = -ENOMEM;
 723         cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
 724         if (!cpuid_entries)
 725                 goto out;
 726
 727         r = 0;
 728         for (i = 0; i < ARRAY_SIZE(param); i++) {
 729                 const struct kvm_cpuid_param *ent = &param[i];
 730
 731                 if (ent->qualifier && !ent->qualifier(ent))
 732                         continue;
 733
 734                 r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
 735                                 &nent, cpuid->nent, type);
 736
 737                 if (r)
 738                         goto out_free;
 739
 740                 if (!ent->has_leaf_count)
 741                         continue;
 742
 743                 limit = cpuid_entries[nent - 1].eax;
 744                 for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
 745                         r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
 746                                      &nent, cpuid->nent, type);
 747
 748                 if (r)
 749                         goto out_free;
 750         }
 751
 752         r = -EFAULT;
 753         if (copy_to_user(entries, cpuid_entries,
 754                          nent * sizeof(struct kvm_cpuid_entry2)))
 755                 goto out_free;
 756         cpuid->nent = nent;
 757         r = 0;
 758
 759 out_free:
 760         vfree(cpuid_entries);
 761 out:
 762         return r;
 763 }
 764
 765 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
 766 {
 767         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
 768         int j, nent = vcpu->arch.cpuid_nent;
 769
 770         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
 771         /* when no next entry is found, the current entry[i] is reselected */
 772         for (j = i + 1; ; j = (j + 1) % nent) {
 773                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
 774                 if (ej->function == e->function) {
 775                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
 776                         return j;
 777                 }
 778         }
 779         return 0; /* silence gcc, even though control never reaches here */
 780 }
 781
 782 /* find an entry with matching function, matching index (if needed), and that
 783  * should be read next (if it's stateful) */
 784 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
 785         u32 function, u32 index)
 786 {
 787         if (e->function != function)
 788                 return 0;
 789         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
 790                 return 0;
 791         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
 792             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
 793                 return 0;
 794         return 1;
 795 }
 796
 797 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
 798                                               u32 function, u32 index)
 799 {
 800         int i;
 801         struct kvm_cpuid_entry2 *best = NULL;
 802
 803         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 804                 struct kvm_cpuid_entry2 *e;
 805
 806                 e = &vcpu->arch.cpuid_entries[i];
 807                 if (is_matching_cpuid_entry(e, function, index)) {
 808                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
 809                                 move_to_next_stateful_cpuid_entry(vcpu, i);
 810                         best = e;
 811                         break;
 812                 }
 813         }
 814         return best;
 815 }
 816 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
 817
 818 /*
 819  * If no match is found, check whether we exceed the vCPU's limit
 820  * and return the content of the highest valid _standard_ leaf instead.
 821  * This is to satisfy the CPUID specification.
 822  */
 823 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
 824                                                   u32 function, u32 index)
 825 {
 826         struct kvm_cpuid_entry2 *maxlevel;
 827
 828         maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
 829         if (!maxlevel || maxlevel->eax >= function)
 830                 return NULL;
 831         if (function & 0x80000000) {
 832                 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
 833                 if (!maxlevel)
 834                         return NULL;
 835         }
 836         return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
 837 }
 838
 839 void kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
 840 {
 841         u32 function = *eax, index = *ecx;
 842         struct kvm_cpuid_entry2 *best;
 843
 844         best = kvm_find_cpuid_entry(vcpu, function, index);
 845
 846         if (!best)
 847                 best = check_cpuid_limit(vcpu, function, index);
 848
 849         /*
 850          * Perfmon not yet supported for L2 guest.
 851          */
 852         if (is_guest_mode(vcpu) && function == 0xa)
 853                 best = NULL;
 854
 855         if (best) {
 856                 *eax = best->eax;
 857                 *ebx = best->ebx;
 858                 *ecx = best->ecx;
 859                 *edx = best->edx;
 860         } else
 861                 *eax = *ebx = *ecx = *edx = 0;
 862         trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx);
 863 }
 864 EXPORT_SYMBOL_GPL(kvm_cpuid);
 865
 866 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
 867 {
 868         u32 function, eax, ebx, ecx, edx;
 869
 870         function = eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
 871         ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
 872         kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx);
 873         kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
 874         kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
 875         kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
 876         kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
 877         kvm_x86_ops->skip_emulated_instruction(vcpu);
 878 }
 879 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);