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/processor.h>
  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 /* These are scattered features in cpufeatures.h. */
  69 #define KVM_CPUID_BIT_AVX512_4VNNIW     2
  70 #define KVM_CPUID_BIT_AVX512_4FMAPS     3
  71 #define KF(x) bit(KVM_CPUID_BIT_##x)
  72
  73 int kvm_update_cpuid(struct kvm_vcpu *vcpu)
  74 {
  75         struct kvm_cpuid_entry2 *best;
  76         struct kvm_lapic *apic = vcpu->arch.apic;
  77
  78         best = kvm_find_cpuid_entry(vcpu, 1, 0);
  79         if (!best)
  80                 return 0;
  81
  82         /* Update OSXSAVE bit */
  83         if (boot_cpu_has(X86_FEATURE_XSAVE) && best->function == 0x1) {
  84                 best->ecx &= ~F(OSXSAVE);
  85                 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
  86                         best->ecx |= F(OSXSAVE);
  87         }
  88
  89         best->edx &= ~F(APIC);
  90         if (vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE)
  91                 best->edx |= F(APIC);
  92
  93         if (apic) {
  94                 if (best->ecx & F(TSC_DEADLINE_TIMER))
  95                         apic->lapic_timer.timer_mode_mask = 3 << 17;
  96                 else
  97                         apic->lapic_timer.timer_mode_mask = 1 << 17;
  98         }
  99
 100         best = kvm_find_cpuid_entry(vcpu, 7, 0);
 101         if (best) {
 102                 /* Update OSPKE bit */
 103                 if (boot_cpu_has(X86_FEATURE_PKU) && best->function == 0x7) {
 104                         best->ecx &= ~F(OSPKE);
 105                         if (kvm_read_cr4_bits(vcpu, X86_CR4_PKE))
 106                                 best->ecx |= F(OSPKE);
 107                 }
 108         }
 109
 110         best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
 111         if (!best) {
 112                 vcpu->arch.guest_supported_xcr0 = 0;
 113                 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
 114         } else {
 115                 vcpu->arch.guest_supported_xcr0 =
 116                         (best->eax | ((u64)best->edx << 32)) &
 117                         kvm_supported_xcr0();
 118                 vcpu->arch.guest_xstate_size = best->ebx =
 119                         xstate_required_size(vcpu->arch.xcr0, false);
 120         }
 121
 122         best = kvm_find_cpuid_entry(vcpu, 0xD, 1);
 123         if (best && (best->eax & (F(XSAVES) | F(XSAVEC))))
 124                 best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
 125
 126         /*
 127          * The existing code assumes virtual address is 48-bit in the canonical
 128          * address checks; exit if it is ever changed.
 129          */
 130         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
 131         if (best && ((best->eax & 0xff00) >> 8) != 48 &&
 132                 ((best->eax & 0xff00) >> 8) != 0)
 133                 return -EINVAL;
 134
 135         /* Update physical-address width */
 136         vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
 137
 138         kvm_pmu_refresh(vcpu);
 139         return 0;
 140 }
 141
 142 static int is_efer_nx(void)
 143 {
 144         unsigned long long efer = 0;
 145
 146         rdmsrl_safe(MSR_EFER, &efer);
 147         return efer & EFER_NX;
 148 }
 149
 150 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
 151 {
 152         int i;
 153         struct kvm_cpuid_entry2 *e, *entry;
 154
 155         entry = NULL;
 156         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 157                 e = &vcpu->arch.cpuid_entries[i];
 158                 if (e->function == 0x80000001) {
 159                         entry = e;
 160                         break;
 161                 }
 162         }
 163         if (entry && (entry->edx & F(NX)) && !is_efer_nx()) {
 164                 entry->edx &= ~F(NX);
 165                 printk(KERN_INFO "kvm: guest NX capability removed\n");
 166         }
 167 }
 168
 169 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
 170 {
 171         struct kvm_cpuid_entry2 *best;
 172
 173         best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
 174         if (!best || best->eax < 0x80000008)
 175                 goto not_found;
 176         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
 177         if (best)
 178                 return best->eax & 0xff;
 179 not_found:
 180         return 36;
 181 }
 182 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr);
 183
 184 /* when an old userspace process fills a new kernel module */
 185 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
 186                              struct kvm_cpuid *cpuid,
 187                              struct kvm_cpuid_entry __user *entries)
 188 {
 189         int r, i;
 190         struct kvm_cpuid_entry *cpuid_entries = NULL;
 191
 192         r = -E2BIG;
 193         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 194                 goto out;
 195         r = -ENOMEM;
 196         if (cpuid->nent) {
 197                 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) *
 198                                         cpuid->nent);
 199                 if (!cpuid_entries)
 200                         goto out;
 201                 r = -EFAULT;
 202                 if (copy_from_user(cpuid_entries, entries,
 203                                    cpuid->nent * sizeof(struct kvm_cpuid_entry)))
 204                         goto out;
 205         }
 206         for (i = 0; i < cpuid->nent; i++) {
 207                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
 208                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
 209                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
 210                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
 211                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
 212                 vcpu->arch.cpuid_entries[i].index = 0;
 213                 vcpu->arch.cpuid_entries[i].flags = 0;
 214                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
 215                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
 216                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
 217         }
 218         vcpu->arch.cpuid_nent = cpuid->nent;
 219         cpuid_fix_nx_cap(vcpu);
 220         kvm_apic_set_version(vcpu);
 221         kvm_x86_ops->cpuid_update(vcpu);
 222         r = kvm_update_cpuid(vcpu);
 223
 224 out:
 225         vfree(cpuid_entries);
 226         return r;
 227 }
 228
 229 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
 230                               struct kvm_cpuid2 *cpuid,
 231                               struct kvm_cpuid_entry2 __user *entries)
 232 {
 233         int r;
 234
 235         r = -E2BIG;
 236         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 237                 goto out;
 238         r = -EFAULT;
 239         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
 240                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
 241                 goto out;
 242         vcpu->arch.cpuid_nent = cpuid->nent;
 243         kvm_apic_set_version(vcpu);
 244         kvm_x86_ops->cpuid_update(vcpu);
 245         r = kvm_update_cpuid(vcpu);
 246 out:
 247         return r;
 248 }
 249
 250 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
 251                               struct kvm_cpuid2 *cpuid,
 252                               struct kvm_cpuid_entry2 __user *entries)
 253 {
 254         int r;
 255
 256         r = -E2BIG;
 257         if (cpuid->nent < vcpu->arch.cpuid_nent)
 258                 goto out;
 259         r = -EFAULT;
 260         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
 261                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
 262                 goto out;
 263         return 0;
 264
 265 out:
 266         cpuid->nent = vcpu->arch.cpuid_nent;
 267         return r;
 268 }
 269
 270 static void cpuid_mask(u32 *word, int wordnum)
 271 {
 272         *word &= boot_cpu_data.x86_capability[wordnum];
 273 }
 274
 275 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 276                            u32 index)
 277 {
 278         entry->function = function;
 279         entry->index = index;
 280         cpuid_count(entry->function, entry->index,
 281                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
 282         entry->flags = 0;
 283 }
 284
 285 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry,
 286                                    u32 func, u32 index, int *nent, int maxnent)
 287 {
 288         switch (func) {
 289         case 0:
 290                 entry->eax = 1;         /* only one leaf currently */
 291                 ++*nent;
 292                 break;
 293         case 1:
 294                 entry->ecx = F(MOVBE);
 295                 ++*nent;
 296                 break;
 297         default:
 298                 break;
 299         }
 300
 301         entry->function = func;
 302         entry->index = index;
 303
 304         return 0;
 305 }
 306
 307 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 308                                  u32 index, int *nent, int maxnent)
 309 {
 310         int r;
 311         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
 312 #ifdef CONFIG_X86_64
 313         unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
 314                                 ? F(GBPAGES) : 0;
 315         unsigned f_lm = F(LM);
 316 #else
 317         unsigned f_gbpages = 0;
 318         unsigned f_lm = 0;
 319 #endif
 320         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
 321         unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
 322         unsigned f_mpx = kvm_mpx_supported() ? F(MPX) : 0;
 323         unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0;
 324
 325         /* cpuid 1.edx */
 326         const u32 kvm_cpuid_1_edx_x86_features =
 327                 F(FPU) | F(VME) | F(DE) | F(PSE) |
 328                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
 329                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
 330                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
 331                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
 332                 0 /* Reserved, DS, ACPI */ | F(MMX) |
 333                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
 334                 0 /* HTT, TM, Reserved, PBE */;
 335         /* cpuid 0x80000001.edx */
 336         const u32 kvm_cpuid_8000_0001_edx_x86_features =
 337                 F(FPU) | F(VME) | F(DE) | F(PSE) |
 338                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
 339                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
 340                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
 341                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
 342                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
 343                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
 344                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
 345         /* cpuid 1.ecx */
 346         const u32 kvm_cpuid_1_ecx_x86_features =
 347                 /* NOTE: MONITOR (and MWAIT) are emulated as NOP,
 348                  * but *not* advertised to guests via CPUID ! */
 349                 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
 350                 0 /* DS-CPL, VMX, SMX, EST */ |
 351                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
 352                 F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
 353                 F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
 354                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
 355                 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
 356                 F(F16C) | F(RDRAND);
 357         /* cpuid 0x80000001.ecx */
 358         const u32 kvm_cpuid_8000_0001_ecx_x86_features =
 359                 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
 360                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
 361                 F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
 362                 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
 363
 364         /* cpuid 0xC0000001.edx */
 365         const u32 kvm_cpuid_C000_0001_edx_x86_features =
 366                 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
 367                 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
 368                 F(PMM) | F(PMM_EN);
 369
 370         /* cpuid 7.0.ebx */
 371         const u32 kvm_cpuid_7_0_ebx_x86_features =
 372                 F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
 373                 F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
 374                 F(ADX) | F(SMAP) | F(AVX512IFMA) | F(AVX512F) | F(AVX512PF) |
 375                 F(AVX512ER) | F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) |
 376                 F(SHA_NI) | F(AVX512BW) | F(AVX512VL);
 377
 378         /* cpuid 0xD.1.eax */
 379         const u32 kvm_cpuid_D_1_eax_x86_features =
 380                 F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves;
 381
 382         /* cpuid 7.0.ecx*/
 383         const u32 kvm_cpuid_7_0_ecx_x86_features =
 384                 F(AVX512VBMI) | F(PKU) | 0 /*OSPKE*/ | F(AVX512_VPOPCNTDQ);
 385
 386         /* cpuid 7.0.edx*/
 387         const u32 kvm_cpuid_7_0_edx_x86_features =
 388                 KF(AVX512_4VNNIW) | KF(AVX512_4FMAPS);
 389
 390         /* all calls to cpuid_count() should be made on the same cpu */
 391         get_cpu();
 392
 393         r = -E2BIG;
 394
 395         if (*nent >= maxnent)
 396                 goto out;
 397
 398         do_cpuid_1_ent(entry, function, index);
 399         ++*nent;
 400
 401         switch (function) {
 402         case 0:
 403                 entry->eax = min(entry->eax, (u32)0xd);
 404                 break;
 405         case 1:
 406                 entry->edx &= kvm_cpuid_1_edx_x86_features;
 407                 cpuid_mask(&entry->edx, CPUID_1_EDX);
 408                 entry->ecx &= kvm_cpuid_1_ecx_x86_features;
 409                 cpuid_mask(&entry->ecx, CPUID_1_ECX);
 410                 /* we support x2apic emulation even if host does not support
 411                  * it since we emulate x2apic in software */
 412                 entry->ecx |= F(X2APIC);
 413                 break;
 414         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
 415          * may return different values. This forces us to get_cpu() before
 416          * issuing the first command, and also to emulate this annoying behavior
 417          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
 418         case 2: {
 419                 int t, times = entry->eax & 0xff;
 420
 421                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 422                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
 423                 for (t = 1; t < times; ++t) {
 424                         if (*nent >= maxnent)
 425                                 goto out;
 426
 427                         do_cpuid_1_ent(&entry[t], function, 0);
 428                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 429                         ++*nent;
 430                 }
 431                 break;
 432         }
 433         /* function 4 has additional index. */
 434         case 4: {
 435                 int i, cache_type;
 436
 437                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 438                 /* read more entries until cache_type is zero */
 439                 for (i = 1; ; ++i) {
 440                         if (*nent >= maxnent)
 441                                 goto out;
 442
 443                         cache_type = entry[i - 1].eax & 0x1f;
 444                         if (!cache_type)
 445                                 break;
 446                         do_cpuid_1_ent(&entry[i], function, i);
 447                         entry[i].flags |=
 448                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 449                         ++*nent;
 450                 }
 451                 break;
 452         }
 453         case 6: /* Thermal management */
 454                 entry->eax = 0x4; /* allow ARAT */
 455                 entry->ebx = 0;
 456                 entry->ecx = 0;
 457                 entry->edx = 0;
 458                 break;
 459         case 7: {
 460                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 461                 /* Mask ebx against host capability word 9 */
 462                 if (index == 0) {
 463                         entry->ebx &= kvm_cpuid_7_0_ebx_x86_features;
 464                         cpuid_mask(&entry->ebx, CPUID_7_0_EBX);
 465                         // TSC_ADJUST is emulated
 466                         entry->ebx |= F(TSC_ADJUST);
 467                         entry->ecx &= kvm_cpuid_7_0_ecx_x86_features;
 468                         cpuid_mask(&entry->ecx, CPUID_7_ECX);
 469                         /* PKU is not yet implemented for shadow paging. */
 470                         if (!tdp_enabled)
 471                                 entry->ecx &= ~F(PKU);
 472                         entry->edx &= kvm_cpuid_7_0_edx_x86_features;
 473                         entry->edx &= get_scattered_cpuid_leaf(7, 0, CPUID_EDX);
 474                 } else {
 475                         entry->ebx = 0;
 476                         entry->ecx = 0;
 477                         entry->edx = 0;
 478                 }
 479                 entry->eax = 0;
 480                 break;
 481         }
 482         case 9:
 483                 break;
 484         case 0xa: { /* Architectural Performance Monitoring */
 485                 struct x86_pmu_capability cap;
 486                 union cpuid10_eax eax;
 487                 union cpuid10_edx edx;
 488
 489                 perf_get_x86_pmu_capability(&cap);
 490
 491                 /*
 492                  * Only support guest architectural pmu on a host
 493                  * with architectural pmu.
 494                  */
 495                 if (!cap.version)
 496                         memset(&cap, 0, sizeof(cap));
 497
 498                 eax.split.version_id = min(cap.version, 2);
 499                 eax.split.num_counters = cap.num_counters_gp;
 500                 eax.split.bit_width = cap.bit_width_gp;
 501                 eax.split.mask_length = cap.events_mask_len;
 502
 503                 edx.split.num_counters_fixed = cap.num_counters_fixed;
 504                 edx.split.bit_width_fixed = cap.bit_width_fixed;
 505                 edx.split.reserved = 0;
 506
 507                 entry->eax = eax.full;
 508                 entry->ebx = cap.events_mask;
 509                 entry->ecx = 0;
 510                 entry->edx = edx.full;
 511                 break;
 512         }
 513         /* function 0xb has additional index. */
 514         case 0xb: {
 515                 int i, level_type;
 516
 517                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 518                 /* read more entries until level_type is zero */
 519                 for (i = 1; ; ++i) {
 520                         if (*nent >= maxnent)
 521                                 goto out;
 522
 523                         level_type = entry[i - 1].ecx & 0xff00;
 524                         if (!level_type)
 525                                 break;
 526                         do_cpuid_1_ent(&entry[i], function, i);
 527                         entry[i].flags |=
 528                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 529                         ++*nent;
 530                 }
 531                 break;
 532         }
 533         case 0xd: {
 534                 int idx, i;
 535                 u64 supported = kvm_supported_xcr0();
 536
 537                 entry->eax &= supported;
 538                 entry->ebx = xstate_required_size(supported, false);
 539                 entry->ecx = entry->ebx;
 540                 entry->edx &= supported >> 32;
 541                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 542                 if (!supported)
 543                         break;
 544
 545                 for (idx = 1, i = 1; idx < 64; ++idx) {
 546                         u64 mask = ((u64)1 << idx);
 547                         if (*nent >= maxnent)
 548                                 goto out;
 549
 550                         do_cpuid_1_ent(&entry[i], function, idx);
 551                         if (idx == 1) {
 552                                 entry[i].eax &= kvm_cpuid_D_1_eax_x86_features;
 553                                 cpuid_mask(&entry[i].eax, CPUID_D_1_EAX);
 554                                 entry[i].ebx = 0;
 555                                 if (entry[i].eax & (F(XSAVES)|F(XSAVEC)))
 556                                         entry[i].ebx =
 557                                                 xstate_required_size(supported,
 558                                                                      true);
 559                         } else {
 560                                 if (entry[i].eax == 0 || !(supported & mask))
 561                                         continue;
 562                                 if (WARN_ON_ONCE(entry[i].ecx & 1))
 563                                         continue;
 564                         }
 565                         entry[i].ecx = 0;
 566                         entry[i].edx = 0;
 567                         entry[i].flags |=
 568                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 569                         ++*nent;
 570                         ++i;
 571                 }
 572                 break;
 573         }
 574         case KVM_CPUID_SIGNATURE: {
 575                 static const char signature[12] = "KVMKVMKVM\0\0";
 576                 const u32 *sigptr = (const u32 *)signature;
 577                 entry->eax = KVM_CPUID_FEATURES;
 578                 entry->ebx = sigptr[0];
 579                 entry->ecx = sigptr[1];
 580                 entry->edx = sigptr[2];
 581                 break;
 582         }
 583         case KVM_CPUID_FEATURES:
 584                 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
 585                              (1 << KVM_FEATURE_NOP_IO_DELAY) |
 586                              (1 << KVM_FEATURE_CLOCKSOURCE2) |
 587                              (1 << KVM_FEATURE_ASYNC_PF) |
 588                              (1 << KVM_FEATURE_PV_EOI) |
 589                              (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
 590                              (1 << KVM_FEATURE_PV_UNHALT);
 591
 592                 if (sched_info_on())
 593                         entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
 594
 595                 entry->ebx = 0;
 596                 entry->ecx = 0;
 597                 entry->edx = 0;
 598                 break;
 599         case 0x80000000:
 600                 entry->eax = min(entry->eax, 0x8000001a);
 601                 break;
 602         case 0x80000001:
 603                 entry->edx &= kvm_cpuid_8000_0001_edx_x86_features;
 604                 cpuid_mask(&entry->edx, CPUID_8000_0001_EDX);
 605                 entry->ecx &= kvm_cpuid_8000_0001_ecx_x86_features;
 606                 cpuid_mask(&entry->ecx, CPUID_8000_0001_ECX);
 607                 break;
 608         case 0x80000007: /* Advanced power management */
 609                 /* invariant TSC is CPUID.80000007H:EDX[8] */
 610                 entry->edx &= (1 << 8);
 611                 /* mask against host */
 612                 entry->edx &= boot_cpu_data.x86_power;
 613                 entry->eax = entry->ebx = entry->ecx = 0;
 614                 break;
 615         case 0x80000008: {
 616                 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
 617                 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
 618                 unsigned phys_as = entry->eax & 0xff;
 619
 620                 if (!g_phys_as)
 621                         g_phys_as = phys_as;
 622                 entry->eax = g_phys_as | (virt_as << 8);
 623                 entry->ebx = entry->edx = 0;
 624                 break;
 625         }
 626         case 0x80000019:
 627                 entry->ecx = entry->edx = 0;
 628                 break;
 629         case 0x8000001a:
 630                 break;
 631         case 0x8000001d:
 632                 break;
 633         /*Add support for Centaur's CPUID instruction*/
 634         case 0xC0000000:
 635                 /*Just support up to 0xC0000004 now*/
 636                 entry->eax = min(entry->eax, 0xC0000004);
 637                 break;
 638         case 0xC0000001:
 639                 entry->edx &= kvm_cpuid_C000_0001_edx_x86_features;
 640                 cpuid_mask(&entry->edx, CPUID_C000_0001_EDX);
 641                 break;
 642         case 3: /* Processor serial number */
 643         case 5: /* MONITOR/MWAIT */
 644         case 0xC0000002:
 645         case 0xC0000003:
 646         case 0xC0000004:
 647         default:
 648                 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
 649                 break;
 650         }
 651
 652         kvm_x86_ops->set_supported_cpuid(function, entry);
 653
 654         r = 0;
 655
 656 out:
 657         put_cpu();
 658
 659         return r;
 660 }
 661
 662 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
 663                         u32 idx, int *nent, int maxnent, unsigned int type)
 664 {
 665         if (type == KVM_GET_EMULATED_CPUID)
 666                 return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
 667
 668         return __do_cpuid_ent(entry, func, idx, nent, maxnent);
 669 }
 670
 671 #undef F
 672
 673 struct kvm_cpuid_param {
 674         u32 func;
 675         u32 idx;
 676         bool has_leaf_count;
 677         bool (*qualifier)(const struct kvm_cpuid_param *param);
 678 };
 679
 680 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
 681 {
 682         return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
 683 }
 684
 685 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
 686                                  __u32 num_entries, unsigned int ioctl_type)
 687 {
 688         int i;
 689         __u32 pad[3];
 690
 691         if (ioctl_type != KVM_GET_EMULATED_CPUID)
 692                 return false;
 693
 694         /*
 695          * We want to make sure that ->padding is being passed clean from
 696          * userspace in case we want to use it for something in the future.
 697          *
 698          * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
 699          * have to give ourselves satisfied only with the emulated side. /me
 700          * sheds a tear.
 701          */
 702         for (i = 0; i < num_entries; i++) {
 703                 if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
 704                         return true;
 705
 706                 if (pad[0] || pad[1] || pad[2])
 707                         return true;
 708         }
 709         return false;
 710 }
 711
 712 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
 713                             struct kvm_cpuid_entry2 __user *entries,
 714                             unsigned int type)
 715 {
 716         struct kvm_cpuid_entry2 *cpuid_entries;
 717         int limit, nent = 0, r = -E2BIG, i;
 718         u32 func;
 719         static const struct kvm_cpuid_param param[] = {
 720                 { .func = 0, .has_leaf_count = true },
 721                 { .func = 0x80000000, .has_leaf_count = true },
 722                 { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
 723                 { .func = KVM_CPUID_SIGNATURE },
 724                 { .func = KVM_CPUID_FEATURES },
 725         };
 726
 727         if (cpuid->nent < 1)
 728                 goto out;
 729         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 730                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
 731
 732         if (sanity_check_entries(entries, cpuid->nent, type))
 733                 return -EINVAL;
 734
 735         r = -ENOMEM;
 736         cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
 737         if (!cpuid_entries)
 738                 goto out;
 739
 740         r = 0;
 741         for (i = 0; i < ARRAY_SIZE(param); i++) {
 742                 const struct kvm_cpuid_param *ent = &param[i];
 743
 744                 if (ent->qualifier && !ent->qualifier(ent))
 745                         continue;
 746
 747                 r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
 748                                 &nent, cpuid->nent, type);
 749
 750                 if (r)
 751                         goto out_free;
 752
 753                 if (!ent->has_leaf_count)
 754                         continue;
 755
 756                 limit = cpuid_entries[nent - 1].eax;
 757                 for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
 758                         r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
 759                                      &nent, cpuid->nent, type);
 760
 761                 if (r)
 762                         goto out_free;
 763         }
 764
 765         r = -EFAULT;
 766         if (copy_to_user(entries, cpuid_entries,
 767                          nent * sizeof(struct kvm_cpuid_entry2)))
 768                 goto out_free;
 769         cpuid->nent = nent;
 770         r = 0;
 771
 772 out_free:
 773         vfree(cpuid_entries);
 774 out:
 775         return r;
 776 }
 777
 778 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
 779 {
 780         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
 781         int j, nent = vcpu->arch.cpuid_nent;
 782
 783         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
 784         /* when no next entry is found, the current entry[i] is reselected */
 785         for (j = i + 1; ; j = (j + 1) % nent) {
 786                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
 787                 if (ej->function == e->function) {
 788                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
 789                         return j;
 790                 }
 791         }
 792         return 0; /* silence gcc, even though control never reaches here */
 793 }
 794
 795 /* find an entry with matching function, matching index (if needed), and that
 796  * should be read next (if it's stateful) */
 797 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
 798         u32 function, u32 index)
 799 {
 800         if (e->function != function)
 801                 return 0;
 802         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
 803                 return 0;
 804         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
 805             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
 806                 return 0;
 807         return 1;
 808 }
 809
 810 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
 811                                               u32 function, u32 index)
 812 {
 813         int i;
 814         struct kvm_cpuid_entry2 *best = NULL;
 815
 816         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 817                 struct kvm_cpuid_entry2 *e;
 818
 819                 e = &vcpu->arch.cpuid_entries[i];
 820                 if (is_matching_cpuid_entry(e, function, index)) {
 821                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
 822                                 move_to_next_stateful_cpuid_entry(vcpu, i);
 823                         best = e;
 824                         break;
 825                 }
 826         }
 827         return best;
 828 }
 829 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
 830
 831 /*
 832  * If no match is found, check whether we exceed the vCPU's limit
 833  * and return the content of the highest valid _standard_ leaf instead.
 834  * This is to satisfy the CPUID specification.
 835  */
 836 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
 837                                                   u32 function, u32 index)
 838 {
 839         struct kvm_cpuid_entry2 *maxlevel;
 840
 841         maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
 842         if (!maxlevel || maxlevel->eax >= function)
 843                 return NULL;
 844         if (function & 0x80000000) {
 845                 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
 846                 if (!maxlevel)
 847                         return NULL;
 848         }
 849         return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
 850 }
 851
 852 void kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
 853 {
 854         u32 function = *eax, index = *ecx;
 855         struct kvm_cpuid_entry2 *best;
 856
 857         best = kvm_find_cpuid_entry(vcpu, function, index);
 858
 859         if (!best)
 860                 best = check_cpuid_limit(vcpu, function, index);
 861
 862         if (best) {
 863                 *eax = best->eax;
 864                 *ebx = best->ebx;
 865                 *ecx = best->ecx;
 866                 *edx = best->edx;
 867         } else
 868                 *eax = *ebx = *ecx = *edx = 0;
 869         trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx);
 870 }
 871 EXPORT_SYMBOL_GPL(kvm_cpuid);
 872
 873 int kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
 874 {
 875         u32 eax, ebx, ecx, edx;
 876
 877         eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
 878         ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
 879         kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx);
 880         kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
 881         kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
 882         kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
 883         kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
 884         return kvm_skip_emulated_instruction(vcpu);
 885 }
 886 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);