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/i387.h> /* For use_eager_fpu.  Ugh! */
  20 #include <asm/fpu-internal.h> /* For use_eager_fpu.  Ugh! */
  21 #include <asm/user.h>
  22 #include <asm/xsave.h>
  23 #include "cpuid.h"
  24 #include "lapic.h"
  25 #include "mmu.h"
  26 #include "trace.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 &= XSTATE_EXTEND_MASK;
  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 u64 kvm_supported_xcr0(void)
  50 {
  51         u64 xcr0 = KVM_SUPPORTED_XCR0 & host_xcr0;
  52
  53         if (!kvm_x86_ops->mpx_supported())
  54                 xcr0 &= ~(XSTATE_BNDREGS | XSTATE_BNDCSR);
  55
  56         return xcr0;
  57 }
  58
  59 #define F(x) bit(X86_FEATURE_##x)
  60
  61 int kvm_update_cpuid(struct kvm_vcpu *vcpu)
  62 {
  63         struct kvm_cpuid_entry2 *best;
  64         struct kvm_lapic *apic = vcpu->arch.apic;
  65
  66         best = kvm_find_cpuid_entry(vcpu, 1, 0);
  67         if (!best)
  68                 return 0;
  69
  70         /* Update OSXSAVE bit */
  71         if (cpu_has_xsave && best->function == 0x1) {
  72                 best->ecx &= ~F(OSXSAVE);
  73                 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
  74                         best->ecx |= F(OSXSAVE);
  75         }
  76
  77         if (apic) {
  78                 if (best->ecx & F(TSC_DEADLINE_TIMER))
  79                         apic->lapic_timer.timer_mode_mask = 3 << 17;
  80                 else
  81                         apic->lapic_timer.timer_mode_mask = 1 << 17;
  82         }
  83
  84         best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
  85         if (!best) {
  86                 vcpu->arch.guest_supported_xcr0 = 0;
  87                 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
  88         } else {
  89                 vcpu->arch.guest_supported_xcr0 =
  90                         (best->eax | ((u64)best->edx << 32)) &
  91                         kvm_supported_xcr0();
  92                 vcpu->arch.guest_xstate_size = best->ebx =
  93                         xstate_required_size(vcpu->arch.xcr0, false);
  94         }
  95
  96         best = kvm_find_cpuid_entry(vcpu, 0xD, 1);
  97         if (best && (best->eax & (F(XSAVES) | F(XSAVEC))))
  98                 best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
  99
 100         vcpu->arch.eager_fpu = guest_cpuid_has_mpx(vcpu);
 101
 102         /*
 103          * The existing code assumes virtual address is 48-bit in the canonical
 104          * address checks; exit if it is ever changed.
 105          */
 106         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
 107         if (best && ((best->eax & 0xff00) >> 8) != 48 &&
 108                 ((best->eax & 0xff00) >> 8) != 0)
 109                 return -EINVAL;
 110
 111         /* Update physical-address width */
 112         vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
 113
 114         kvm_pmu_cpuid_update(vcpu);
 115         return 0;
 116 }
 117
 118 static int is_efer_nx(void)
 119 {
 120         unsigned long long efer = 0;
 121
 122         rdmsrl_safe(MSR_EFER, &efer);
 123         return efer & EFER_NX;
 124 }
 125
 126 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
 127 {
 128         int i;
 129         struct kvm_cpuid_entry2 *e, *entry;
 130
 131         entry = NULL;
 132         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 133                 e = &vcpu->arch.cpuid_entries[i];
 134                 if (e->function == 0x80000001) {
 135                         entry = e;
 136                         break;
 137                 }
 138         }
 139         if (entry && (entry->edx & F(NX)) && !is_efer_nx()) {
 140                 entry->edx &= ~F(NX);
 141                 printk(KERN_INFO "kvm: guest NX capability removed\n");
 142         }
 143 }
 144
 145 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
 146 {
 147         struct kvm_cpuid_entry2 *best;
 148
 149         best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
 150         if (!best || best->eax < 0x80000008)
 151                 goto not_found;
 152         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
 153         if (best)
 154                 return best->eax & 0xff;
 155 not_found:
 156         return 36;
 157 }
 158 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr);
 159
 160 /* when an old userspace process fills a new kernel module */
 161 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
 162                              struct kvm_cpuid *cpuid,
 163                              struct kvm_cpuid_entry __user *entries)
 164 {
 165         int r, i;
 166         struct kvm_cpuid_entry *cpuid_entries;
 167
 168         r = -E2BIG;
 169         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 170                 goto out;
 171         r = -ENOMEM;
 172         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
 173         if (!cpuid_entries)
 174                 goto out;
 175         r = -EFAULT;
 176         if (copy_from_user(cpuid_entries, entries,
 177                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
 178                 goto out_free;
 179         for (i = 0; i < cpuid->nent; i++) {
 180                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
 181                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
 182                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
 183                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
 184                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
 185                 vcpu->arch.cpuid_entries[i].index = 0;
 186                 vcpu->arch.cpuid_entries[i].flags = 0;
 187                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
 188                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
 189                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
 190         }
 191         vcpu->arch.cpuid_nent = cpuid->nent;
 192         cpuid_fix_nx_cap(vcpu);
 193         kvm_apic_set_version(vcpu);
 194         kvm_x86_ops->cpuid_update(vcpu);
 195         r = kvm_update_cpuid(vcpu);
 196
 197 out_free:
 198         vfree(cpuid_entries);
 199 out:
 200         return r;
 201 }
 202
 203 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
 204                               struct kvm_cpuid2 *cpuid,
 205                               struct kvm_cpuid_entry2 __user *entries)
 206 {
 207         int r;
 208
 209         r = -E2BIG;
 210         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 211                 goto out;
 212         r = -EFAULT;
 213         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
 214                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
 215                 goto out;
 216         vcpu->arch.cpuid_nent = cpuid->nent;
 217         kvm_apic_set_version(vcpu);
 218         kvm_x86_ops->cpuid_update(vcpu);
 219         r = kvm_update_cpuid(vcpu);
 220 out:
 221         return r;
 222 }
 223
 224 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
 225                               struct kvm_cpuid2 *cpuid,
 226                               struct kvm_cpuid_entry2 __user *entries)
 227 {
 228         int r;
 229
 230         r = -E2BIG;
 231         if (cpuid->nent < vcpu->arch.cpuid_nent)
 232                 goto out;
 233         r = -EFAULT;
 234         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
 235                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
 236                 goto out;
 237         return 0;
 238
 239 out:
 240         cpuid->nent = vcpu->arch.cpuid_nent;
 241         return r;
 242 }
 243
 244 static void cpuid_mask(u32 *word, int wordnum)
 245 {
 246         *word &= boot_cpu_data.x86_capability[wordnum];
 247 }
 248
 249 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 250                            u32 index)
 251 {
 252         entry->function = function;
 253         entry->index = index;
 254         cpuid_count(entry->function, entry->index,
 255                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
 256         entry->flags = 0;
 257 }
 258
 259 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry,
 260                                    u32 func, u32 index, int *nent, int maxnent)
 261 {
 262         switch (func) {
 263         case 0:
 264                 entry->eax = 1;         /* only one leaf currently */
 265                 ++*nent;
 266                 break;
 267         case 1:
 268                 entry->ecx = F(MOVBE);
 269                 ++*nent;
 270                 break;
 271         default:
 272                 break;
 273         }
 274
 275         entry->function = func;
 276         entry->index = index;
 277
 278         return 0;
 279 }
 280
 281 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 282                                  u32 index, int *nent, int maxnent)
 283 {
 284         int r;
 285         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
 286 #ifdef CONFIG_X86_64
 287         unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
 288                                 ? F(GBPAGES) : 0;
 289         unsigned f_lm = F(LM);
 290 #else
 291         unsigned f_gbpages = 0;
 292         unsigned f_lm = 0;
 293 #endif
 294         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
 295         unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
 296         unsigned f_mpx = kvm_x86_ops->mpx_supported() ? F(MPX) : 0;
 297         unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0;
 298
 299         /* cpuid 1.edx */
 300         const u32 kvm_supported_word0_x86_features =
 301                 F(FPU) | F(VME) | F(DE) | F(PSE) |
 302                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
 303                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
 304                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
 305                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
 306                 0 /* Reserved, DS, ACPI */ | F(MMX) |
 307                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
 308                 0 /* HTT, TM, Reserved, PBE */;
 309         /* cpuid 0x80000001.edx */
 310         const u32 kvm_supported_word1_x86_features =
 311                 F(FPU) | F(VME) | F(DE) | F(PSE) |
 312                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
 313                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
 314                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
 315                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
 316                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
 317                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
 318                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
 319         /* cpuid 1.ecx */
 320         const u32 kvm_supported_word4_x86_features =
 321                 /* NOTE: MONITOR (and MWAIT) are emulated as NOP,
 322                  * but *not* advertised to guests via CPUID ! */
 323                 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
 324                 0 /* DS-CPL, VMX, SMX, EST */ |
 325                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
 326                 F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
 327                 F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
 328                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
 329                 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
 330                 F(F16C) | F(RDRAND);
 331         /* cpuid 0x80000001.ecx */
 332         const u32 kvm_supported_word6_x86_features =
 333                 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
 334                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
 335                 F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
 336                 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
 337
 338         /* cpuid 0xC0000001.edx */
 339         const u32 kvm_supported_word5_x86_features =
 340                 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
 341                 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
 342                 F(PMM) | F(PMM_EN);
 343
 344         /* cpuid 7.0.ebx */
 345         const u32 kvm_supported_word9_x86_features =
 346                 F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
 347                 F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
 348                 F(ADX) | F(SMAP) | F(AVX512F) | F(AVX512PF) | F(AVX512ER) |
 349                 F(AVX512CD);
 350
 351         /* cpuid 0xD.1.eax */
 352         const u32 kvm_supported_word10_x86_features =
 353                 F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves;
 354
 355         /* all calls to cpuid_count() should be made on the same cpu */
 356         get_cpu();
 357
 358         r = -E2BIG;
 359
 360         if (*nent >= maxnent)
 361                 goto out;
 362
 363         do_cpuid_1_ent(entry, function, index);
 364         ++*nent;
 365
 366         switch (function) {
 367         case 0:
 368                 entry->eax = min(entry->eax, (u32)0xd);
 369                 break;
 370         case 1:
 371                 entry->edx &= kvm_supported_word0_x86_features;
 372                 cpuid_mask(&entry->edx, 0);
 373                 entry->ecx &= kvm_supported_word4_x86_features;
 374                 cpuid_mask(&entry->ecx, 4);
 375                 /* we support x2apic emulation even if host does not support
 376                  * it since we emulate x2apic in software */
 377                 entry->ecx |= F(X2APIC);
 378                 break;
 379         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
 380          * may return different values. This forces us to get_cpu() before
 381          * issuing the first command, and also to emulate this annoying behavior
 382          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
 383         case 2: {
 384                 int t, times = entry->eax & 0xff;
 385
 386                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 387                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
 388                 for (t = 1; t < times; ++t) {
 389                         if (*nent >= maxnent)
 390                                 goto out;
 391
 392                         do_cpuid_1_ent(&entry[t], function, 0);
 393                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 394                         ++*nent;
 395                 }
 396                 break;
 397         }
 398         /* function 4 has additional index. */
 399         case 4: {
 400                 int i, cache_type;
 401
 402                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 403                 /* read more entries until cache_type is zero */
 404                 for (i = 1; ; ++i) {
 405                         if (*nent >= maxnent)
 406                                 goto out;
 407
 408                         cache_type = entry[i - 1].eax & 0x1f;
 409                         if (!cache_type)
 410                                 break;
 411                         do_cpuid_1_ent(&entry[i], function, i);
 412                         entry[i].flags |=
 413                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 414                         ++*nent;
 415                 }
 416                 break;
 417         }
 418         case 7: {
 419                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 420                 /* Mask ebx against host capability word 9 */
 421                 if (index == 0) {
 422                         entry->ebx &= kvm_supported_word9_x86_features;
 423                         cpuid_mask(&entry->ebx, 9);
 424                         // TSC_ADJUST is emulated
 425                         entry->ebx |= F(TSC_ADJUST);
 426                 } else
 427                         entry->ebx = 0;
 428                 entry->eax = 0;
 429                 entry->ecx = 0;
 430                 entry->edx = 0;
 431                 break;
 432         }
 433         case 9:
 434                 break;
 435         case 0xa: { /* Architectural Performance Monitoring */
 436                 struct x86_pmu_capability cap;
 437                 union cpuid10_eax eax;
 438                 union cpuid10_edx edx;
 439
 440                 perf_get_x86_pmu_capability(&cap);
 441
 442                 /*
 443                  * Only support guest architectural pmu on a host
 444                  * with architectural pmu.
 445                  */
 446                 if (!cap.version)
 447                         memset(&cap, 0, sizeof(cap));
 448
 449                 eax.split.version_id = min(cap.version, 2);
 450                 eax.split.num_counters = cap.num_counters_gp;
 451                 eax.split.bit_width = cap.bit_width_gp;
 452                 eax.split.mask_length = cap.events_mask_len;
 453
 454                 edx.split.num_counters_fixed = cap.num_counters_fixed;
 455                 edx.split.bit_width_fixed = cap.bit_width_fixed;
 456                 edx.split.reserved = 0;
 457
 458                 entry->eax = eax.full;
 459                 entry->ebx = cap.events_mask;
 460                 entry->ecx = 0;
 461                 entry->edx = edx.full;
 462                 break;
 463         }
 464         /* function 0xb has additional index. */
 465         case 0xb: {
 466                 int i, level_type;
 467
 468                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 469                 /* read more entries until level_type is zero */
 470                 for (i = 1; ; ++i) {
 471                         if (*nent >= maxnent)
 472                                 goto out;
 473
 474                         level_type = entry[i - 1].ecx & 0xff00;
 475                         if (!level_type)
 476                                 break;
 477                         do_cpuid_1_ent(&entry[i], function, i);
 478                         entry[i].flags |=
 479                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 480                         ++*nent;
 481                 }
 482                 break;
 483         }
 484         case 0xd: {
 485                 int idx, i;
 486                 u64 supported = kvm_supported_xcr0();
 487
 488                 entry->eax &= supported;
 489                 entry->ebx = xstate_required_size(supported, false);
 490                 entry->ecx = entry->ebx;
 491                 entry->edx &= supported >> 32;
 492                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 493                 if (!supported)
 494                         break;
 495
 496                 for (idx = 1, i = 1; idx < 64; ++idx) {
 497                         u64 mask = ((u64)1 << idx);
 498                         if (*nent >= maxnent)
 499                                 goto out;
 500
 501                         do_cpuid_1_ent(&entry[i], function, idx);
 502                         if (idx == 1) {
 503                                 entry[i].eax &= kvm_supported_word10_x86_features;
 504                                 entry[i].ebx = 0;
 505                                 if (entry[i].eax & (F(XSAVES)|F(XSAVEC)))
 506                                         entry[i].ebx =
 507                                                 xstate_required_size(supported,
 508                                                                      true);
 509                         } else {
 510                                 if (entry[i].eax == 0 || !(supported & mask))
 511                                         continue;
 512                                 if (WARN_ON_ONCE(entry[i].ecx & 1))
 513                                         continue;
 514                         }
 515                         entry[i].ecx = 0;
 516                         entry[i].edx = 0;
 517                         entry[i].flags |=
 518                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 519                         ++*nent;
 520                         ++i;
 521                 }
 522                 break;
 523         }
 524         case KVM_CPUID_SIGNATURE: {
 525                 static const char signature[12] = "KVMKVMKVM\0\0";
 526                 const u32 *sigptr = (const u32 *)signature;
 527                 entry->eax = KVM_CPUID_FEATURES;
 528                 entry->ebx = sigptr[0];
 529                 entry->ecx = sigptr[1];
 530                 entry->edx = sigptr[2];
 531                 break;
 532         }
 533         case KVM_CPUID_FEATURES:
 534                 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
 535                              (1 << KVM_FEATURE_NOP_IO_DELAY) |
 536                              (1 << KVM_FEATURE_CLOCKSOURCE2) |
 537                              (1 << KVM_FEATURE_ASYNC_PF) |
 538                              (1 << KVM_FEATURE_PV_EOI) |
 539                              (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
 540                              (1 << KVM_FEATURE_PV_UNHALT);
 541
 542                 if (sched_info_on())
 543                         entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
 544
 545                 entry->ebx = 0;
 546                 entry->ecx = 0;
 547                 entry->edx = 0;
 548                 break;
 549         case 0x80000000:
 550                 entry->eax = min(entry->eax, 0x8000001a);
 551                 break;
 552         case 0x80000001:
 553                 entry->edx &= kvm_supported_word1_x86_features;
 554                 cpuid_mask(&entry->edx, 1);
 555                 entry->ecx &= kvm_supported_word6_x86_features;
 556                 cpuid_mask(&entry->ecx, 6);
 557                 break;
 558         case 0x80000007: /* Advanced power management */
 559                 /* invariant TSC is CPUID.80000007H:EDX[8] */
 560                 entry->edx &= (1 << 8);
 561                 /* mask against host */
 562                 entry->edx &= boot_cpu_data.x86_power;
 563                 entry->eax = entry->ebx = entry->ecx = 0;
 564                 break;
 565         case 0x80000008: {
 566                 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
 567                 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
 568                 unsigned phys_as = entry->eax & 0xff;
 569
 570                 if (!g_phys_as)
 571                         g_phys_as = phys_as;
 572                 entry->eax = g_phys_as | (virt_as << 8);
 573                 entry->ebx = entry->edx = 0;
 574                 break;
 575         }
 576         case 0x80000019:
 577                 entry->ecx = entry->edx = 0;
 578                 break;
 579         case 0x8000001a:
 580                 break;
 581         case 0x8000001d:
 582                 break;
 583         /*Add support for Centaur's CPUID instruction*/
 584         case 0xC0000000:
 585                 /*Just support up to 0xC0000004 now*/
 586                 entry->eax = min(entry->eax, 0xC0000004);
 587                 break;
 588         case 0xC0000001:
 589                 entry->edx &= kvm_supported_word5_x86_features;
 590                 cpuid_mask(&entry->edx, 5);
 591                 break;
 592         case 3: /* Processor serial number */
 593         case 5: /* MONITOR/MWAIT */
 594         case 6: /* Thermal management */
 595         case 0xC0000002:
 596         case 0xC0000003:
 597         case 0xC0000004:
 598         default:
 599                 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
 600                 break;
 601         }
 602
 603         kvm_x86_ops->set_supported_cpuid(function, entry);
 604
 605         r = 0;
 606
 607 out:
 608         put_cpu();
 609
 610         return r;
 611 }
 612
 613 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
 614                         u32 idx, int *nent, int maxnent, unsigned int type)
 615 {
 616         if (type == KVM_GET_EMULATED_CPUID)
 617                 return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
 618
 619         return __do_cpuid_ent(entry, func, idx, nent, maxnent);
 620 }
 621
 622 #undef F
 623
 624 struct kvm_cpuid_param {
 625         u32 func;
 626         u32 idx;
 627         bool has_leaf_count;
 628         bool (*qualifier)(const struct kvm_cpuid_param *param);
 629 };
 630
 631 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
 632 {
 633         return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
 634 }
 635
 636 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
 637                                  __u32 num_entries, unsigned int ioctl_type)
 638 {
 639         int i;
 640         __u32 pad[3];
 641
 642         if (ioctl_type != KVM_GET_EMULATED_CPUID)
 643                 return false;
 644
 645         /*
 646          * We want to make sure that ->padding is being passed clean from
 647          * userspace in case we want to use it for something in the future.
 648          *
 649          * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
 650          * have to give ourselves satisfied only with the emulated side. /me
 651          * sheds a tear.
 652          */
 653         for (i = 0; i < num_entries; i++) {
 654                 if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
 655                         return true;
 656
 657                 if (pad[0] || pad[1] || pad[2])
 658                         return true;
 659         }
 660         return false;
 661 }
 662
 663 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
 664                             struct kvm_cpuid_entry2 __user *entries,
 665                             unsigned int type)
 666 {
 667         struct kvm_cpuid_entry2 *cpuid_entries;
 668         int limit, nent = 0, r = -E2BIG, i;
 669         u32 func;
 670         static const struct kvm_cpuid_param param[] = {
 671                 { .func = 0, .has_leaf_count = true },
 672                 { .func = 0x80000000, .has_leaf_count = true },
 673                 { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
 674                 { .func = KVM_CPUID_SIGNATURE },
 675                 { .func = KVM_CPUID_FEATURES },
 676         };
 677
 678         if (cpuid->nent < 1)
 679                 goto out;
 680         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 681                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
 682
 683         if (sanity_check_entries(entries, cpuid->nent, type))
 684                 return -EINVAL;
 685
 686         r = -ENOMEM;
 687         cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
 688         if (!cpuid_entries)
 689                 goto out;
 690
 691         r = 0;
 692         for (i = 0; i < ARRAY_SIZE(param); i++) {
 693                 const struct kvm_cpuid_param *ent = &param[i];
 694
 695                 if (ent->qualifier && !ent->qualifier(ent))
 696                         continue;
 697
 698                 r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
 699                                 &nent, cpuid->nent, type);
 700
 701                 if (r)
 702                         goto out_free;
 703
 704                 if (!ent->has_leaf_count)
 705                         continue;
 706
 707                 limit = cpuid_entries[nent - 1].eax;
 708                 for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
 709                         r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
 710                                      &nent, cpuid->nent, type);
 711
 712                 if (r)
 713                         goto out_free;
 714         }
 715
 716         r = -EFAULT;
 717         if (copy_to_user(entries, cpuid_entries,
 718                          nent * sizeof(struct kvm_cpuid_entry2)))
 719                 goto out_free;
 720         cpuid->nent = nent;
 721         r = 0;
 722
 723 out_free:
 724         vfree(cpuid_entries);
 725 out:
 726         return r;
 727 }
 728
 729 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
 730 {
 731         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
 732         int j, nent = vcpu->arch.cpuid_nent;
 733
 734         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
 735         /* when no next entry is found, the current entry[i] is reselected */
 736         for (j = i + 1; ; j = (j + 1) % nent) {
 737                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
 738                 if (ej->function == e->function) {
 739                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
 740                         return j;
 741                 }
 742         }
 743         return 0; /* silence gcc, even though control never reaches here */
 744 }
 745
 746 /* find an entry with matching function, matching index (if needed), and that
 747  * should be read next (if it's stateful) */
 748 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
 749         u32 function, u32 index)
 750 {
 751         if (e->function != function)
 752                 return 0;
 753         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
 754                 return 0;
 755         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
 756             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
 757                 return 0;
 758         return 1;
 759 }
 760
 761 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
 762                                               u32 function, u32 index)
 763 {
 764         int i;
 765         struct kvm_cpuid_entry2 *best = NULL;
 766
 767         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 768                 struct kvm_cpuid_entry2 *e;
 769
 770                 e = &vcpu->arch.cpuid_entries[i];
 771                 if (is_matching_cpuid_entry(e, function, index)) {
 772                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
 773                                 move_to_next_stateful_cpuid_entry(vcpu, i);
 774                         best = e;
 775                         break;
 776                 }
 777         }
 778         return best;
 779 }
 780 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
 781
 782 /*
 783  * If no match is found, check whether we exceed the vCPU's limit
 784  * and return the content of the highest valid _standard_ leaf instead.
 785  * This is to satisfy the CPUID specification.
 786  */
 787 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
 788                                                   u32 function, u32 index)
 789 {
 790         struct kvm_cpuid_entry2 *maxlevel;
 791
 792         maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
 793         if (!maxlevel || maxlevel->eax >= function)
 794                 return NULL;
 795         if (function & 0x80000000) {
 796                 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
 797                 if (!maxlevel)
 798                         return NULL;
 799         }
 800         return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
 801 }
 802
 803 void kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
 804 {
 805         u32 function = *eax, index = *ecx;
 806         struct kvm_cpuid_entry2 *best;
 807
 808         best = kvm_find_cpuid_entry(vcpu, function, index);
 809
 810         if (!best)
 811                 best = check_cpuid_limit(vcpu, function, index);
 812
 813         /*
 814          * Perfmon not yet supported for L2 guest.
 815          */
 816         if (is_guest_mode(vcpu) && function == 0xa)
 817                 best = NULL;
 818
 819         if (best) {
 820                 *eax = best->eax;
 821                 *ebx = best->ebx;
 822                 *ecx = best->ecx;
 823                 *edx = best->edx;
 824         } else
 825                 *eax = *ebx = *ecx = *edx = 0;
 826         trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx);
 827 }
 828 EXPORT_SYMBOL_GPL(kvm_cpuid);
 829
 830 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
 831 {
 832         u32 function, eax, ebx, ecx, edx;
 833
 834         function = eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
 835         ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
 836         kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx);
 837         kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
 838         kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
 839         kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
 840         kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
 841         kvm_x86_ops->skip_emulated_instruction(vcpu);
 842 }
 843 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);