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/fpu/xstate.h>
  21 #include "cpuid.h"
  22 #include "lapic.h"
  23 #include "mmu.h"
  24 #include "trace.h"
  25 #include "pmu.h"
  26
  27 static u32 xstate_required_size(u64 xstate_bv, bool compacted)
  28 {
  29         int feature_bit = 0;
  30         u32 ret = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
  31
  32         xstate_bv &= XFEATURE_MASK_EXTEND;
  33         while (xstate_bv) {
  34                 if (xstate_bv & 0x1) {
  35                         u32 eax, ebx, ecx, edx, offset;
  36                         cpuid_count(0xD, feature_bit, &eax, &ebx, &ecx, &edx);
  37                         offset = compacted ? ret : ebx;
  38                         ret = max(ret, offset + eax);
  39                 }
  40
  41                 xstate_bv >>= 1;
  42                 feature_bit++;
  43         }
  44
  45         return ret;
  46 }
  47
  48 bool kvm_mpx_supported(void)
  49 {
  50         return ((host_xcr0 & (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR))
  51                  && kvm_x86_ops->mpx_supported());
  52 }
  53 EXPORT_SYMBOL_GPL(kvm_mpx_supported);
  54
  55 u64 kvm_supported_xcr0(void)
  56 {
  57         u64 xcr0 = KVM_SUPPORTED_XCR0 & host_xcr0;
  58
  59         if (!kvm_mpx_supported())
  60                 xcr0 &= ~(XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR);
  61
  62         return xcr0;
  63 }
  64
  65 #define F(x) bit(X86_FEATURE_##x)
  66
  67 int kvm_update_cpuid(struct kvm_vcpu *vcpu)
  68 {
  69         struct kvm_cpuid_entry2 *best;
  70         struct kvm_lapic *apic = vcpu->arch.apic;
  71
  72         best = kvm_find_cpuid_entry(vcpu, 1, 0);
  73         if (!best)
  74                 return 0;
  75
  76         /* Update OSXSAVE bit */
  77         if (cpu_has_xsave && best->function == 0x1) {
  78                 best->ecx &= ~F(OSXSAVE);
  79                 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
  80                         best->ecx |= F(OSXSAVE);
  81         }
  82
  83         if (apic) {
  84                 if (best->ecx & F(TSC_DEADLINE_TIMER))
  85                         apic->lapic_timer.timer_mode_mask = 3 << 17;
  86                 else
  87                         apic->lapic_timer.timer_mode_mask = 1 << 17;
  88         }
  89
  90         best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
  91         if (!best) {
  92                 vcpu->arch.guest_supported_xcr0 = 0;
  93                 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
  94         } else {
  95                 vcpu->arch.guest_supported_xcr0 =
  96                         (best->eax | ((u64)best->edx << 32)) &
  97                         kvm_supported_xcr0();
  98                 vcpu->arch.guest_xstate_size = best->ebx =
  99                         xstate_required_size(vcpu->arch.xcr0, false);
 100         }
 101
 102         best = kvm_find_cpuid_entry(vcpu, 0xD, 1);
 103         if (best && (best->eax & (F(XSAVES) | F(XSAVEC))))
 104                 best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
 105
 106         kvm_x86_ops->fpu_activate(vcpu);
 107
 108         /*
 109          * The existing code assumes virtual address is 48-bit in the canonical
 110          * address checks; exit if it is ever changed.
 111          */
 112         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
 113         if (best && ((best->eax & 0xff00) >> 8) != 48 &&
 114                 ((best->eax & 0xff00) >> 8) != 0)
 115                 return -EINVAL;
 116
 117         /* Update physical-address width */
 118         vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
 119
 120         kvm_pmu_refresh(vcpu);
 121         return 0;
 122 }
 123
 124 static int is_efer_nx(void)
 125 {
 126         unsigned long long efer = 0;
 127
 128         rdmsrl_safe(MSR_EFER, &efer);
 129         return efer & EFER_NX;
 130 }
 131
 132 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
 133 {
 134         int i;
 135         struct kvm_cpuid_entry2 *e, *entry;
 136
 137         entry = NULL;
 138         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 139                 e = &vcpu->arch.cpuid_entries[i];
 140                 if (e->function == 0x80000001) {
 141                         entry = e;
 142                         break;
 143                 }
 144         }
 145         if (entry && (entry->edx & F(NX)) && !is_efer_nx()) {
 146                 entry->edx &= ~F(NX);
 147                 printk(KERN_INFO "kvm: guest NX capability removed\n");
 148         }
 149 }
 150
 151 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
 152 {
 153         struct kvm_cpuid_entry2 *best;
 154
 155         best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
 156         if (!best || best->eax < 0x80000008)
 157                 goto not_found;
 158         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
 159         if (best)
 160                 return best->eax & 0xff;
 161 not_found:
 162         return 36;
 163 }
 164 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr);
 165
 166 /* when an old userspace process fills a new kernel module */
 167 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
 168                              struct kvm_cpuid *cpuid,
 169                              struct kvm_cpuid_entry __user *entries)
 170 {
 171         int r, i;
 172         struct kvm_cpuid_entry *cpuid_entries;
 173
 174         r = -E2BIG;
 175         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 176                 goto out;
 177         r = -ENOMEM;
 178         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
 179         if (!cpuid_entries)
 180                 goto out;
 181         r = -EFAULT;
 182         if (copy_from_user(cpuid_entries, entries,
 183                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
 184                 goto out_free;
 185         for (i = 0; i < cpuid->nent; i++) {
 186                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
 187                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
 188                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
 189                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
 190                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
 191                 vcpu->arch.cpuid_entries[i].index = 0;
 192                 vcpu->arch.cpuid_entries[i].flags = 0;
 193                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
 194                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
 195                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
 196         }
 197         vcpu->arch.cpuid_nent = cpuid->nent;
 198         cpuid_fix_nx_cap(vcpu);
 199         kvm_apic_set_version(vcpu);
 200         kvm_x86_ops->cpuid_update(vcpu);
 201         r = kvm_update_cpuid(vcpu);
 202
 203 out_free:
 204         vfree(cpuid_entries);
 205 out:
 206         return r;
 207 }
 208
 209 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
 210                               struct kvm_cpuid2 *cpuid,
 211                               struct kvm_cpuid_entry2 __user *entries)
 212 {
 213         int r;
 214
 215         r = -E2BIG;
 216         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 217                 goto out;
 218         r = -EFAULT;
 219         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
 220                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
 221                 goto out;
 222         vcpu->arch.cpuid_nent = cpuid->nent;
 223         kvm_apic_set_version(vcpu);
 224         kvm_x86_ops->cpuid_update(vcpu);
 225         r = kvm_update_cpuid(vcpu);
 226 out:
 227         return r;
 228 }
 229
 230 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
 231                               struct kvm_cpuid2 *cpuid,
 232                               struct kvm_cpuid_entry2 __user *entries)
 233 {
 234         int r;
 235
 236         r = -E2BIG;
 237         if (cpuid->nent < vcpu->arch.cpuid_nent)
 238                 goto out;
 239         r = -EFAULT;
 240         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
 241                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
 242                 goto out;
 243         return 0;
 244
 245 out:
 246         cpuid->nent = vcpu->arch.cpuid_nent;
 247         return r;
 248 }
 249
 250 static void cpuid_mask(u32 *word, int wordnum)
 251 {
 252         *word &= boot_cpu_data.x86_capability[wordnum];
 253 }
 254
 255 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 256                            u32 index)
 257 {
 258         entry->function = function;
 259         entry->index = index;
 260         cpuid_count(entry->function, entry->index,
 261                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
 262         entry->flags = 0;
 263 }
 264
 265 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry,
 266                                    u32 func, u32 index, int *nent, int maxnent)
 267 {
 268         switch (func) {
 269         case 0:
 270                 entry->eax = 7;
 271                 ++*nent;
 272                 break;
 273         case 1:
 274                 entry->ecx = F(MOVBE);
 275                 ++*nent;
 276                 break;
 277         case 7:
 278                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 279                 if (index == 0)
 280                         entry->ecx = F(RDPID);
 281                 ++*nent;
 282         default:
 283                 break;
 284         }
 285
 286         entry->function = func;
 287         entry->index = index;
 288
 289         return 0;
 290 }
 291
 292 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 293                                  u32 index, int *nent, int maxnent)
 294 {
 295         int r;
 296         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
 297 #ifdef CONFIG_X86_64
 298         unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
 299                                 ? F(GBPAGES) : 0;
 300         unsigned f_lm = F(LM);
 301 #else
 302         unsigned f_gbpages = 0;
 303         unsigned f_lm = 0;
 304 #endif
 305         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
 306         unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
 307         unsigned f_mpx = kvm_mpx_supported() ? F(MPX) : 0;
 308         unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0;
 309
 310         /* cpuid 1.edx */
 311         const u32 kvm_supported_word0_x86_features =
 312                 F(FPU) | F(VME) | F(DE) | F(PSE) |
 313                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
 314                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
 315                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
 316                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
 317                 0 /* Reserved, DS, ACPI */ | F(MMX) |
 318                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
 319                 0 /* HTT, TM, Reserved, PBE */;
 320         /* cpuid 0x80000001.edx */
 321         const u32 kvm_supported_word1_x86_features =
 322                 F(FPU) | F(VME) | F(DE) | F(PSE) |
 323                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
 324                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
 325                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
 326                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
 327                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
 328                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
 329                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
 330         /* cpuid 1.ecx */
 331         const u32 kvm_supported_word4_x86_features =
 332                 /* NOTE: MONITOR (and MWAIT) are emulated as NOP,
 333                  * but *not* advertised to guests via CPUID ! */
 334                 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
 335                 0 /* DS-CPL, VMX, SMX, EST */ |
 336                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
 337                 F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
 338                 F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
 339                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
 340                 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
 341                 F(F16C) | F(RDRAND);
 342         /* cpuid 0x80000001.ecx */
 343         const u32 kvm_supported_word6_x86_features =
 344                 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
 345                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
 346                 F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
 347                 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
 348
 349         /* cpuid 0x80000008.ebx */
 350         const u32 kvm_cpuid_8000_0008_ebx_x86_features =
 351                 F(AMD_IBPB) | F(AMD_IBRS) | F(AMD_SSBD) | F(VIRT_SSBD) |
 352                 F(AMD_SSB_NO) | F(AMD_STIBP);
 353
 354         /* cpuid 0xC0000001.edx */
 355         const u32 kvm_supported_word5_x86_features =
 356                 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
 357                 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
 358                 F(PMM) | F(PMM_EN);
 359
 360         /* cpuid 7.0.ebx */
 361         const u32 kvm_supported_word9_x86_features =
 362                 F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
 363                 F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
 364                 F(ADX) | F(SMAP) | F(AVX512F) | F(AVX512PF) | F(AVX512ER) |
 365                 F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(PCOMMIT);
 366
 367         /* cpuid 0xD.1.eax */
 368         const u32 kvm_supported_word10_x86_features =
 369                 F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves;
 370
 371         /* cpuid 7.0.edx*/
 372         const u32 kvm_cpuid_7_0_edx_x86_features =
 373                 F(SPEC_CTRL) | F(SPEC_CTRL_SSBD) | F(ARCH_CAPABILITIES) |
 374                 F(INTEL_STIBP) | F(MD_CLEAR);
 375
 376         /* all calls to cpuid_count() should be made on the same cpu */
 377         get_cpu();
 378
 379         r = -E2BIG;
 380
 381         if (WARN_ON(*nent >= maxnent))
 382                 goto out;
 383
 384         do_cpuid_1_ent(entry, function, index);
 385         ++*nent;
 386
 387         switch (function) {
 388         case 0:
 389                 entry->eax = min(entry->eax, (u32)0xd);
 390                 break;
 391         case 1:
 392                 entry->edx &= kvm_supported_word0_x86_features;
 393                 cpuid_mask(&entry->edx, 0);
 394                 entry->ecx &= kvm_supported_word4_x86_features;
 395                 cpuid_mask(&entry->ecx, 4);
 396                 /* we support x2apic emulation even if host does not support
 397                  * it since we emulate x2apic in software */
 398                 entry->ecx |= F(X2APIC);
 399                 break;
 400         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
 401          * may return different values. This forces us to get_cpu() before
 402          * issuing the first command, and also to emulate this annoying behavior
 403          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
 404         case 2: {
 405                 int t, times = entry->eax & 0xff;
 406
 407                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 408                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
 409                 for (t = 1; t < times; ++t) {
 410                         if (*nent >= maxnent)
 411                                 goto out;
 412
 413                         do_cpuid_1_ent(&entry[t], function, 0);
 414                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 415                         ++*nent;
 416                 }
 417                 break;
 418         }
 419         /* function 4 has additional index. */
 420         case 4: {
 421                 int i, cache_type;
 422
 423                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 424                 /* read more entries until cache_type is zero */
 425                 for (i = 1; ; ++i) {
 426                         if (*nent >= maxnent)
 427                                 goto out;
 428
 429                         cache_type = entry[i - 1].eax & 0x1f;
 430                         if (!cache_type)
 431                                 break;
 432                         do_cpuid_1_ent(&entry[i], function, i);
 433                         entry[i].flags |=
 434                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 435                         ++*nent;
 436                 }
 437                 break;
 438         }
 439         case 6: /* Thermal management */
 440                 entry->eax = 0x4; /* allow ARAT */
 441                 entry->ebx = 0;
 442                 entry->ecx = 0;
 443                 entry->edx = 0;
 444                 break;
 445         case 7: {
 446                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 447                 /* Mask ebx against host capability word 9 */
 448                 if (index == 0) {
 449                         entry->ebx &= kvm_supported_word9_x86_features;
 450                         cpuid_mask(&entry->ebx, 9);
 451                         // TSC_ADJUST is emulated
 452                         entry->ebx |= F(TSC_ADJUST);
 453                         entry->edx &= kvm_cpuid_7_0_edx_x86_features;
 454                         cpuid_mask(&entry->edx, CPUID_7_EDX);
 455                         if (boot_cpu_has(X86_FEATURE_IBPB) &&
 456                             boot_cpu_has(X86_FEATURE_IBRS))
 457                                 entry->edx |= F(SPEC_CTRL);
 458                         if (boot_cpu_has(X86_FEATURE_STIBP))
 459                                 entry->edx |= F(INTEL_STIBP);
 460                         if (boot_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) ||
 461                             boot_cpu_has(X86_FEATURE_AMD_SSBD))
 462                                 entry->edx |= F(SPEC_CTRL_SSBD);
 463                         /*
 464                          * We emulate ARCH_CAPABILITIES in software even
 465                          * if the host doesn't support it.
 466                          */
 467                         entry->edx |= F(ARCH_CAPABILITIES);
 468                 } else {
 469                         entry->ebx = 0;
 470                         entry->edx = 0;
 471                 }
 472                 entry->eax = 0;
 473                 entry->ecx = 0;
 474                 break;
 475         }
 476         case 9:
 477                 break;
 478         case 0xa: { /* Architectural Performance Monitoring */
 479                 struct x86_pmu_capability cap;
 480                 union cpuid10_eax eax;
 481                 union cpuid10_edx edx;
 482
 483                 perf_get_x86_pmu_capability(&cap);
 484
 485                 /*
 486                  * Only support guest architectural pmu on a host
 487                  * with architectural pmu.
 488                  */
 489                 if (!cap.version)
 490                         memset(&cap, 0, sizeof(cap));
 491
 492                 eax.split.version_id = min(cap.version, 2);
 493                 eax.split.num_counters = cap.num_counters_gp;
 494                 eax.split.bit_width = cap.bit_width_gp;
 495                 eax.split.mask_length = cap.events_mask_len;
 496
 497                 edx.split.num_counters_fixed = cap.num_counters_fixed;
 498                 edx.split.bit_width_fixed = cap.bit_width_fixed;
 499                 edx.split.reserved = 0;
 500
 501                 entry->eax = eax.full;
 502                 entry->ebx = cap.events_mask;
 503                 entry->ecx = 0;
 504                 entry->edx = edx.full;
 505                 break;
 506         }
 507         /* function 0xb has additional index. */
 508         case 0xb: {
 509                 int i, level_type;
 510
 511                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 512                 /* read more entries until level_type is zero */
 513                 for (i = 1; ; ++i) {
 514                         if (*nent >= maxnent)
 515                                 goto out;
 516
 517                         level_type = entry[i - 1].ecx & 0xff00;
 518                         if (!level_type)
 519                                 break;
 520                         do_cpuid_1_ent(&entry[i], function, i);
 521                         entry[i].flags |=
 522                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 523                         ++*nent;
 524                 }
 525                 break;
 526         }
 527         case 0xd: {
 528                 int idx, i;
 529                 u64 supported = kvm_supported_xcr0();
 530
 531                 entry->eax &= supported;
 532                 entry->ebx = xstate_required_size(supported, false);
 533                 entry->ecx = entry->ebx;
 534                 entry->edx &= supported >> 32;
 535                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 536                 if (!supported)
 537                         break;
 538
 539                 for (idx = 1, i = 1; idx < 64; ++idx) {
 540                         u64 mask = ((u64)1 << idx);
 541                         if (*nent >= maxnent)
 542                                 goto out;
 543
 544                         do_cpuid_1_ent(&entry[i], function, idx);
 545                         if (idx == 1) {
 546                                 entry[i].eax &= kvm_supported_word10_x86_features;
 547                                 cpuid_mask(&entry[i].eax, 10);
 548                                 entry[i].ebx = 0;
 549                                 if (entry[i].eax & (F(XSAVES)|F(XSAVEC)))
 550                                         entry[i].ebx =
 551                                                 xstate_required_size(supported,
 552                                                                      true);
 553                         } else {
 554                                 if (entry[i].eax == 0 || !(supported & mask))
 555                                         continue;
 556                                 if (WARN_ON_ONCE(entry[i].ecx & 1))
 557                                         continue;
 558                         }
 559                         entry[i].ecx = 0;
 560                         entry[i].edx = 0;
 561                         entry[i].flags |=
 562                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 563                         ++*nent;
 564                         ++i;
 565                 }
 566                 break;
 567         }
 568         case KVM_CPUID_SIGNATURE: {
 569                 static const char signature[12] = "KVMKVMKVM\0\0";
 570                 const u32 *sigptr = (const u32 *)signature;
 571                 entry->eax = KVM_CPUID_FEATURES;
 572                 entry->ebx = sigptr[0];
 573                 entry->ecx = sigptr[1];
 574                 entry->edx = sigptr[2];
 575                 break;
 576         }
 577         case KVM_CPUID_FEATURES:
 578                 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
 579                              (1 << KVM_FEATURE_NOP_IO_DELAY) |
 580                              (1 << KVM_FEATURE_CLOCKSOURCE2) |
 581                              (1 << KVM_FEATURE_ASYNC_PF) |
 582                              (1 << KVM_FEATURE_PV_EOI) |
 583                              (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
 584                              (1 << KVM_FEATURE_PV_UNHALT);
 585
 586                 if (sched_info_on())
 587                         entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
 588
 589                 entry->ebx = 0;
 590                 entry->ecx = 0;
 591                 entry->edx = 0;
 592                 break;
 593         case 0x80000000:
 594                 entry->eax = min(entry->eax, 0x8000001a);
 595                 break;
 596         case 0x80000001:
 597                 entry->edx &= kvm_supported_word1_x86_features;
 598                 cpuid_mask(&entry->edx, 1);
 599                 entry->ecx &= kvm_supported_word6_x86_features;
 600                 cpuid_mask(&entry->ecx, 6);
 601                 break;
 602         case 0x80000007: /* Advanced power management */
 603                 /* invariant TSC is CPUID.80000007H:EDX[8] */
 604                 entry->edx &= (1 << 8);
 605                 /* mask against host */
 606                 entry->edx &= boot_cpu_data.x86_power;
 607                 entry->eax = entry->ebx = entry->ecx = 0;
 608                 break;
 609         case 0x80000008: {
 610                 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
 611                 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
 612                 unsigned phys_as = entry->eax & 0xff;
 613
 614                 /*
 615                  * Use bare metal's MAXPHADDR if the CPU doesn't report guest
 616                  * MAXPHYADDR separately, or if TDP (NPT) is disabled, as the
 617                  * guest version "applies only to guests using nested paging".
 618                  */
 619                 if (!g_phys_as || !tdp_enabled)
 620                         g_phys_as = phys_as;
 621
 622                 entry->eax = g_phys_as | (virt_as << 8);
 623                 entry->edx = 0;
 624                 /*
 625                  * IBRS, IBPB and VIRT_SSBD aren't necessarily present in
 626                  * hardware cpuid
 627                  */
 628                 if (boot_cpu_has(X86_FEATURE_AMD_IBPB))
 629                         entry->ebx |= F(AMD_IBPB);
 630                 if (boot_cpu_has(X86_FEATURE_AMD_IBRS))
 631                         entry->ebx |= F(AMD_IBRS);
 632                 if (boot_cpu_has(X86_FEATURE_VIRT_SSBD))
 633                         entry->ebx |= F(VIRT_SSBD);
 634                 entry->ebx &= kvm_cpuid_8000_0008_ebx_x86_features;
 635                 cpuid_mask(&entry->ebx, CPUID_8000_0008_EBX);
 636                 /*
 637                  * The preference is to use SPEC CTRL MSR instead of the
 638                  * VIRT_SPEC MSR.
 639                  */
 640                 if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD) &&
 641                     !boot_cpu_has(X86_FEATURE_AMD_SSBD))
 642                         entry->ebx |= F(VIRT_SSBD);
 643                 break;
 644         }
 645         case 0x80000019:
 646                 entry->ecx = entry->edx = 0;
 647                 break;
 648         case 0x8000001a:
 649                 break;
 650         case 0x8000001d:
 651                 break;
 652         /*Add support for Centaur's CPUID instruction*/
 653         case 0xC0000000:
 654                 /*Just support up to 0xC0000004 now*/
 655                 entry->eax = min(entry->eax, 0xC0000004);
 656                 break;
 657         case 0xC0000001:
 658                 entry->edx &= kvm_supported_word5_x86_features;
 659                 cpuid_mask(&entry->edx, 5);
 660                 break;
 661         case 3: /* Processor serial number */
 662         case 5: /* MONITOR/MWAIT */
 663         case 0xC0000002:
 664         case 0xC0000003:
 665         case 0xC0000004:
 666         default:
 667                 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
 668                 break;
 669         }
 670
 671         kvm_x86_ops->set_supported_cpuid(function, entry);
 672
 673         r = 0;
 674
 675 out:
 676         put_cpu();
 677
 678         return r;
 679 }
 680
 681 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
 682                         u32 idx, int *nent, int maxnent, unsigned int type)
 683 {
 684         if (*nent >= maxnent)
 685                 return -E2BIG;
 686
 687         if (type == KVM_GET_EMULATED_CPUID)
 688                 return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
 689
 690         return __do_cpuid_ent(entry, func, idx, nent, maxnent);
 691 }
 692
 693 #undef F
 694
 695 struct kvm_cpuid_param {
 696         u32 func;
 697         u32 idx;
 698         bool has_leaf_count;
 699         bool (*qualifier)(const struct kvm_cpuid_param *param);
 700 };
 701
 702 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
 703 {
 704         return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
 705 }
 706
 707 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
 708                                  __u32 num_entries, unsigned int ioctl_type)
 709 {
 710         int i;
 711         __u32 pad[3];
 712
 713         if (ioctl_type != KVM_GET_EMULATED_CPUID)
 714                 return false;
 715
 716         /*
 717          * We want to make sure that ->padding is being passed clean from
 718          * userspace in case we want to use it for something in the future.
 719          *
 720          * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
 721          * have to give ourselves satisfied only with the emulated side. /me
 722          * sheds a tear.
 723          */
 724         for (i = 0; i < num_entries; i++) {
 725                 if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
 726                         return true;
 727
 728                 if (pad[0] || pad[1] || pad[2])
 729                         return true;
 730         }
 731         return false;
 732 }
 733
 734 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
 735                             struct kvm_cpuid_entry2 __user *entries,
 736                             unsigned int type)
 737 {
 738         struct kvm_cpuid_entry2 *cpuid_entries;
 739         int limit, nent = 0, r = -E2BIG, i;
 740         u32 func;
 741         static const struct kvm_cpuid_param param[] = {
 742                 { .func = 0, .has_leaf_count = true },
 743                 { .func = 0x80000000, .has_leaf_count = true },
 744                 { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
 745                 { .func = KVM_CPUID_SIGNATURE },
 746                 { .func = KVM_CPUID_FEATURES },
 747         };
 748
 749         if (cpuid->nent < 1)
 750                 goto out;
 751         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 752                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
 753
 754         if (sanity_check_entries(entries, cpuid->nent, type))
 755                 return -EINVAL;
 756
 757         r = -ENOMEM;
 758         cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
 759         if (!cpuid_entries)
 760                 goto out;
 761
 762         r = 0;
 763         for (i = 0; i < ARRAY_SIZE(param); i++) {
 764                 const struct kvm_cpuid_param *ent = &param[i];
 765
 766                 if (ent->qualifier && !ent->qualifier(ent))
 767                         continue;
 768
 769                 r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
 770                                 &nent, cpuid->nent, type);
 771
 772                 if (r)
 773                         goto out_free;
 774
 775                 if (!ent->has_leaf_count)
 776                         continue;
 777
 778                 limit = cpuid_entries[nent - 1].eax;
 779                 for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
 780                         r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
 781                                      &nent, cpuid->nent, type);
 782
 783                 if (r)
 784                         goto out_free;
 785         }
 786
 787         r = -EFAULT;
 788         if (copy_to_user(entries, cpuid_entries,
 789                          nent * sizeof(struct kvm_cpuid_entry2)))
 790                 goto out_free;
 791         cpuid->nent = nent;
 792         r = 0;
 793
 794 out_free:
 795         vfree(cpuid_entries);
 796 out:
 797         return r;
 798 }
 799
 800 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
 801 {
 802         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
 803         struct kvm_cpuid_entry2 *ej;
 804         int j = i;
 805         int nent = vcpu->arch.cpuid_nent;
 806
 807         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
 808         /* when no next entry is found, the current entry[i] is reselected */
 809         do {
 810                 j = (j + 1) % nent;
 811                 ej = &vcpu->arch.cpuid_entries[j];
 812         } while (ej->function != e->function);
 813
 814         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
 815
 816         return j;
 817 }
 818
 819 /* find an entry with matching function, matching index (if needed), and that
 820  * should be read next (if it's stateful) */
 821 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
 822         u32 function, u32 index)
 823 {
 824         if (e->function != function)
 825                 return 0;
 826         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
 827                 return 0;
 828         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
 829             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
 830                 return 0;
 831         return 1;
 832 }
 833
 834 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
 835                                               u32 function, u32 index)
 836 {
 837         int i;
 838         struct kvm_cpuid_entry2 *best = NULL;
 839
 840         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 841                 struct kvm_cpuid_entry2 *e;
 842
 843                 e = &vcpu->arch.cpuid_entries[i];
 844                 if (is_matching_cpuid_entry(e, function, index)) {
 845                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
 846                                 move_to_next_stateful_cpuid_entry(vcpu, i);
 847                         best = e;
 848                         break;
 849                 }
 850         }
 851         return best;
 852 }
 853 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
 854
 855 /*
 856  * If no match is found, check whether we exceed the vCPU's limit
 857  * and return the content of the highest valid _standard_ leaf instead.
 858  * This is to satisfy the CPUID specification.
 859  */
 860 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
 861                                                   u32 function, u32 index)
 862 {
 863         struct kvm_cpuid_entry2 *maxlevel;
 864
 865         maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
 866         if (!maxlevel || maxlevel->eax >= function)
 867                 return NULL;
 868         if (function & 0x80000000) {
 869                 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
 870                 if (!maxlevel)
 871                         return NULL;
 872         }
 873         return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
 874 }
 875
 876 void kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
 877 {
 878         u32 function = *eax, index = *ecx;
 879         struct kvm_cpuid_entry2 *best;
 880
 881         best = kvm_find_cpuid_entry(vcpu, function, index);
 882
 883         if (!best)
 884                 best = check_cpuid_limit(vcpu, function, index);
 885
 886         if (best) {
 887                 *eax = best->eax;
 888                 *ebx = best->ebx;
 889                 *ecx = best->ecx;
 890                 *edx = best->edx;
 891         } else
 892                 *eax = *ebx = *ecx = *edx = 0;
 893         trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx);
 894 }
 895 EXPORT_SYMBOL_GPL(kvm_cpuid);
 896
 897 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
 898 {
 899         u32 function, eax, ebx, ecx, edx;
 900
 901         function = eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
 902         ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
 903         kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx);
 904         kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
 905         kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
 906         kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
 907         kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
 908         kvm_x86_ops->skip_emulated_instruction(vcpu);
 909 }
 910 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);