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/fpu/internal.h> /* For use_eager_fpu.  Ugh! */
  20 #include <asm/user.h>
  21 #include <asm/fpu/xstate.h>
  22 #include "cpuid.h"
  23 #include "lapic.h"
  24 #include "mmu.h"
  25 #include "trace.h"
  26 #include "pmu.h"
  27
  28 static u32 xstate_required_size(u64 xstate_bv, bool compacted)
  29 {
  30         int feature_bit = 0;
  31         u32 ret = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
  32
  33         xstate_bv &= XFEATURE_MASK_EXTEND;
  34         while (xstate_bv) {
  35                 if (xstate_bv & 0x1) {
  36                         u32 eax, ebx, ecx, edx, offset;
  37                         cpuid_count(0xD, feature_bit, &eax, &ebx, &ecx, &edx);
  38                         offset = compacted ? ret : ebx;
  39                         ret = max(ret, offset + eax);
  40                 }
  41
  42                 xstate_bv >>= 1;
  43                 feature_bit++;
  44         }
  45
  46         return ret;
  47 }
  48
  49 bool kvm_mpx_supported(void)
  50 {
  51         return ((host_xcr0 & (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR))
  52                  && kvm_x86_ops->mpx_supported());
  53 }
  54 EXPORT_SYMBOL_GPL(kvm_mpx_supported);
  55
  56 u64 kvm_supported_xcr0(void)
  57 {
  58         u64 xcr0 = KVM_SUPPORTED_XCR0 & host_xcr0;
  59
  60         if (!kvm_mpx_supported())
  61                 xcr0 &= ~(XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR);
  62
  63         return xcr0;
  64 }
  65
  66 #define F(x) bit(X86_FEATURE_##x)
  67
  68 int kvm_update_cpuid(struct kvm_vcpu *vcpu)
  69 {
  70         struct kvm_cpuid_entry2 *best;
  71         struct kvm_lapic *apic = vcpu->arch.apic;
  72
  73         best = kvm_find_cpuid_entry(vcpu, 1, 0);
  74         if (!best)
  75                 return 0;
  76
  77         /* Update OSXSAVE bit */
  78         if (cpu_has_xsave && best->function == 0x1) {
  79                 best->ecx &= ~F(OSXSAVE);
  80                 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
  81                         best->ecx |= F(OSXSAVE);
  82         }
  83
  84         if (apic) {
  85                 if (best->ecx & F(TSC_DEADLINE_TIMER))
  86                         apic->lapic_timer.timer_mode_mask = 3 << 17;
  87                 else
  88                         apic->lapic_timer.timer_mode_mask = 1 << 17;
  89         }
  90
  91         best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
  92         if (!best) {
  93                 vcpu->arch.guest_supported_xcr0 = 0;
  94                 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
  95         } else {
  96                 vcpu->arch.guest_supported_xcr0 =
  97                         (best->eax | ((u64)best->edx << 32)) &
  98                         kvm_supported_xcr0();
  99                 vcpu->arch.guest_xstate_size = best->ebx =
 100                         xstate_required_size(vcpu->arch.xcr0, false);
 101         }
 102
 103         best = kvm_find_cpuid_entry(vcpu, 0xD, 1);
 104         if (best && (best->eax & (F(XSAVES) | F(XSAVEC))))
 105                 best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
 106
 107         vcpu->arch.eager_fpu = use_eager_fpu();
 108         if (vcpu->arch.eager_fpu)
 109                 kvm_x86_ops->fpu_activate(vcpu);
 110
 111         /*
 112          * The existing code assumes virtual address is 48-bit in the canonical
 113          * address checks; exit if it is ever changed.
 114          */
 115         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
 116         if (best && ((best->eax & 0xff00) >> 8) != 48 &&
 117                 ((best->eax & 0xff00) >> 8) != 0)
 118                 return -EINVAL;
 119
 120         /* Update physical-address width */
 121         vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
 122
 123         kvm_pmu_refresh(vcpu);
 124         return 0;
 125 }
 126
 127 static int is_efer_nx(void)
 128 {
 129         unsigned long long efer = 0;
 130
 131         rdmsrl_safe(MSR_EFER, &efer);
 132         return efer & EFER_NX;
 133 }
 134
 135 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
 136 {
 137         int i;
 138         struct kvm_cpuid_entry2 *e, *entry;
 139
 140         entry = NULL;
 141         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 142                 e = &vcpu->arch.cpuid_entries[i];
 143                 if (e->function == 0x80000001) {
 144                         entry = e;
 145                         break;
 146                 }
 147         }
 148         if (entry && (entry->edx & F(NX)) && !is_efer_nx()) {
 149                 entry->edx &= ~F(NX);
 150                 printk(KERN_INFO "kvm: guest NX capability removed\n");
 151         }
 152 }
 153
 154 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
 155 {
 156         struct kvm_cpuid_entry2 *best;
 157
 158         best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
 159         if (!best || best->eax < 0x80000008)
 160                 goto not_found;
 161         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
 162         if (best)
 163                 return best->eax & 0xff;
 164 not_found:
 165         return 36;
 166 }
 167 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr);
 168
 169 /* when an old userspace process fills a new kernel module */
 170 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
 171                              struct kvm_cpuid *cpuid,
 172                              struct kvm_cpuid_entry __user *entries)
 173 {
 174         int r, i;
 175         struct kvm_cpuid_entry *cpuid_entries;
 176
 177         r = -E2BIG;
 178         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 179                 goto out;
 180         r = -ENOMEM;
 181         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
 182         if (!cpuid_entries)
 183                 goto out;
 184         r = -EFAULT;
 185         if (copy_from_user(cpuid_entries, entries,
 186                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
 187                 goto out_free;
 188         for (i = 0; i < cpuid->nent; i++) {
 189                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
 190                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
 191                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
 192                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
 193                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
 194                 vcpu->arch.cpuid_entries[i].index = 0;
 195                 vcpu->arch.cpuid_entries[i].flags = 0;
 196                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
 197                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
 198                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
 199         }
 200         vcpu->arch.cpuid_nent = cpuid->nent;
 201         cpuid_fix_nx_cap(vcpu);
 202         kvm_apic_set_version(vcpu);
 203         kvm_x86_ops->cpuid_update(vcpu);
 204         r = kvm_update_cpuid(vcpu);
 205
 206 out_free:
 207         vfree(cpuid_entries);
 208 out:
 209         return r;
 210 }
 211
 212 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
 213                               struct kvm_cpuid2 *cpuid,
 214                               struct kvm_cpuid_entry2 __user *entries)
 215 {
 216         int r;
 217
 218         r = -E2BIG;
 219         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 220                 goto out;
 221         r = -EFAULT;
 222         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
 223                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
 224                 goto out;
 225         vcpu->arch.cpuid_nent = cpuid->nent;
 226         kvm_apic_set_version(vcpu);
 227         kvm_x86_ops->cpuid_update(vcpu);
 228         r = kvm_update_cpuid(vcpu);
 229 out:
 230         return r;
 231 }
 232
 233 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
 234                               struct kvm_cpuid2 *cpuid,
 235                               struct kvm_cpuid_entry2 __user *entries)
 236 {
 237         int r;
 238
 239         r = -E2BIG;
 240         if (cpuid->nent < vcpu->arch.cpuid_nent)
 241                 goto out;
 242         r = -EFAULT;
 243         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
 244                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
 245                 goto out;
 246         return 0;
 247
 248 out:
 249         cpuid->nent = vcpu->arch.cpuid_nent;
 250         return r;
 251 }
 252
 253 static void cpuid_mask(u32 *word, int wordnum)
 254 {
 255         *word &= boot_cpu_data.x86_capability[wordnum];
 256 }
 257
 258 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 259                            u32 index)
 260 {
 261         entry->function = function;
 262         entry->index = index;
 263         cpuid_count(entry->function, entry->index,
 264                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
 265         entry->flags = 0;
 266 }
 267
 268 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry,
 269                                    u32 func, u32 index, int *nent, int maxnent)
 270 {
 271         switch (func) {
 272         case 0:
 273                 entry->eax = 1;         /* only one leaf currently */
 274                 ++*nent;
 275                 break;
 276         case 1:
 277                 entry->ecx = F(MOVBE);
 278                 ++*nent;
 279                 break;
 280         default:
 281                 break;
 282         }
 283
 284         entry->function = func;
 285         entry->index = index;
 286
 287         return 0;
 288 }
 289
 290 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 291                                  u32 index, int *nent, int maxnent)
 292 {
 293         int r;
 294         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
 295 #ifdef CONFIG_X86_64
 296         unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
 297                                 ? F(GBPAGES) : 0;
 298         unsigned f_lm = F(LM);
 299 #else
 300         unsigned f_gbpages = 0;
 301         unsigned f_lm = 0;
 302 #endif
 303         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
 304         unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
 305         unsigned f_mpx = kvm_mpx_supported() ? F(MPX) : 0;
 306         unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0;
 307
 308         /* cpuid 1.edx */
 309         const u32 kvm_supported_word0_x86_features =
 310                 F(FPU) | F(VME) | F(DE) | F(PSE) |
 311                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
 312                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
 313                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
 314                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
 315                 0 /* Reserved, DS, ACPI */ | F(MMX) |
 316                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
 317                 0 /* HTT, TM, Reserved, PBE */;
 318         /* cpuid 0x80000001.edx */
 319         const u32 kvm_supported_word1_x86_features =
 320                 F(FPU) | F(VME) | F(DE) | F(PSE) |
 321                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
 322                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
 323                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
 324                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
 325                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
 326                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
 327                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
 328         /* cpuid 1.ecx */
 329         const u32 kvm_supported_word4_x86_features =
 330                 /* NOTE: MONITOR (and MWAIT) are emulated as NOP,
 331                  * but *not* advertised to guests via CPUID ! */
 332                 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
 333                 0 /* DS-CPL, VMX, SMX, EST */ |
 334                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
 335                 F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
 336                 F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
 337                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
 338                 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
 339                 F(F16C) | F(RDRAND);
 340         /* cpuid 0x80000001.ecx */
 341         const u32 kvm_supported_word6_x86_features =
 342                 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
 343                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
 344                 F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
 345                 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
 346
 347         /* cpuid 0xC0000001.edx */
 348         const u32 kvm_supported_word5_x86_features =
 349                 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
 350                 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
 351                 F(PMM) | F(PMM_EN);
 352
 353         /* cpuid 7.0.ebx */
 354         const u32 kvm_supported_word9_x86_features =
 355                 F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
 356                 F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
 357                 F(ADX) | F(SMAP) | F(AVX512F) | F(AVX512PF) | F(AVX512ER) |
 358                 F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(PCOMMIT);
 359
 360         /* cpuid 0xD.1.eax */
 361         const u32 kvm_supported_word10_x86_features =
 362                 F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves;
 363
 364         /* all calls to cpuid_count() should be made on the same cpu */
 365         get_cpu();
 366
 367         r = -E2BIG;
 368
 369         if (*nent >= maxnent)
 370                 goto out;
 371
 372         do_cpuid_1_ent(entry, function, index);
 373         ++*nent;
 374
 375         switch (function) {
 376         case 0:
 377                 entry->eax = min(entry->eax, (u32)0xd);
 378                 break;
 379         case 1:
 380                 entry->edx &= kvm_supported_word0_x86_features;
 381                 cpuid_mask(&entry->edx, 0);
 382                 entry->ecx &= kvm_supported_word4_x86_features;
 383                 cpuid_mask(&entry->ecx, 4);
 384                 /* we support x2apic emulation even if host does not support
 385                  * it since we emulate x2apic in software */
 386                 entry->ecx |= F(X2APIC);
 387                 break;
 388         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
 389          * may return different values. This forces us to get_cpu() before
 390          * issuing the first command, and also to emulate this annoying behavior
 391          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
 392         case 2: {
 393                 int t, times = entry->eax & 0xff;
 394
 395                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 396                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
 397                 for (t = 1; t < times; ++t) {
 398                         if (*nent >= maxnent)
 399                                 goto out;
 400
 401                         do_cpuid_1_ent(&entry[t], function, 0);
 402                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 403                         ++*nent;
 404                 }
 405                 break;
 406         }
 407         /* function 4 has additional index. */
 408         case 4: {
 409                 int i, cache_type;
 410
 411                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 412                 /* read more entries until cache_type is zero */
 413                 for (i = 1; ; ++i) {
 414                         if (*nent >= maxnent)
 415                                 goto out;
 416
 417                         cache_type = entry[i - 1].eax & 0x1f;
 418                         if (!cache_type)
 419                                 break;
 420                         do_cpuid_1_ent(&entry[i], function, i);
 421                         entry[i].flags |=
 422                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 423                         ++*nent;
 424                 }
 425                 break;
 426         }
 427         case 6: /* Thermal management */
 428                 entry->eax = 0x4; /* allow ARAT */
 429                 entry->ebx = 0;
 430                 entry->ecx = 0;
 431                 entry->edx = 0;
 432                 break;
 433         case 7: {
 434                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 435                 /* Mask ebx against host capability word 9 */
 436                 if (index == 0) {
 437                         entry->ebx &= kvm_supported_word9_x86_features;
 438                         cpuid_mask(&entry->ebx, 9);
 439                         // TSC_ADJUST is emulated
 440                         entry->ebx |= F(TSC_ADJUST);
 441                 } else
 442                         entry->ebx = 0;
 443                 entry->eax = 0;
 444                 entry->ecx = 0;
 445                 entry->edx = 0;
 446                 break;
 447         }
 448         case 9:
 449                 break;
 450         case 0xa: { /* Architectural Performance Monitoring */
 451                 struct x86_pmu_capability cap;
 452                 union cpuid10_eax eax;
 453                 union cpuid10_edx edx;
 454
 455                 perf_get_x86_pmu_capability(&cap);
 456
 457                 /*
 458                  * Only support guest architectural pmu on a host
 459                  * with architectural pmu.
 460                  */
 461                 if (!cap.version)
 462                         memset(&cap, 0, sizeof(cap));
 463
 464                 eax.split.version_id = min(cap.version, 2);
 465                 eax.split.num_counters = cap.num_counters_gp;
 466                 eax.split.bit_width = cap.bit_width_gp;
 467                 eax.split.mask_length = cap.events_mask_len;
 468
 469                 edx.split.num_counters_fixed = cap.num_counters_fixed;
 470                 edx.split.bit_width_fixed = cap.bit_width_fixed;
 471                 edx.split.reserved = 0;
 472
 473                 entry->eax = eax.full;
 474                 entry->ebx = cap.events_mask;
 475                 entry->ecx = 0;
 476                 entry->edx = edx.full;
 477                 break;
 478         }
 479         /* function 0xb has additional index. */
 480         case 0xb: {
 481                 int i, level_type;
 482
 483                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 484                 /* read more entries until level_type is zero */
 485                 for (i = 1; ; ++i) {
 486                         if (*nent >= maxnent)
 487                                 goto out;
 488
 489                         level_type = entry[i - 1].ecx & 0xff00;
 490                         if (!level_type)
 491                                 break;
 492                         do_cpuid_1_ent(&entry[i], function, i);
 493                         entry[i].flags |=
 494                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 495                         ++*nent;
 496                 }
 497                 break;
 498         }
 499         case 0xd: {
 500                 int idx, i;
 501                 u64 supported = kvm_supported_xcr0();
 502
 503                 entry->eax &= supported;
 504                 entry->ebx = xstate_required_size(supported, false);
 505                 entry->ecx = entry->ebx;
 506                 entry->edx &= supported >> 32;
 507                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 508                 if (!supported)
 509                         break;
 510
 511                 for (idx = 1, i = 1; idx < 64; ++idx) {
 512                         u64 mask = ((u64)1 << idx);
 513                         if (*nent >= maxnent)
 514                                 goto out;
 515
 516                         do_cpuid_1_ent(&entry[i], function, idx);
 517                         if (idx == 1) {
 518                                 entry[i].eax &= kvm_supported_word10_x86_features;
 519                                 entry[i].ebx = 0;
 520                                 if (entry[i].eax & (F(XSAVES)|F(XSAVEC)))
 521                                         entry[i].ebx =
 522                                                 xstate_required_size(supported,
 523                                                                      true);
 524                         } else {
 525                                 if (entry[i].eax == 0 || !(supported & mask))
 526                                         continue;
 527                                 if (WARN_ON_ONCE(entry[i].ecx & 1))
 528                                         continue;
 529                         }
 530                         entry[i].ecx = 0;
 531                         entry[i].edx = 0;
 532                         entry[i].flags |=
 533                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 534                         ++*nent;
 535                         ++i;
 536                 }
 537                 break;
 538         }
 539         case KVM_CPUID_SIGNATURE: {
 540                 static const char signature[12] = "KVMKVMKVM\0\0";
 541                 const u32 *sigptr = (const u32 *)signature;
 542                 entry->eax = KVM_CPUID_FEATURES;
 543                 entry->ebx = sigptr[0];
 544                 entry->ecx = sigptr[1];
 545                 entry->edx = sigptr[2];
 546                 break;
 547         }
 548         case KVM_CPUID_FEATURES:
 549                 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
 550                              (1 << KVM_FEATURE_NOP_IO_DELAY) |
 551                              (1 << KVM_FEATURE_CLOCKSOURCE2) |
 552                              (1 << KVM_FEATURE_ASYNC_PF) |
 553                              (1 << KVM_FEATURE_PV_EOI) |
 554                              (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
 555                              (1 << KVM_FEATURE_PV_UNHALT);
 556
 557                 if (sched_info_on())
 558                         entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
 559
 560                 entry->ebx = 0;
 561                 entry->ecx = 0;
 562                 entry->edx = 0;
 563                 break;
 564         case 0x80000000:
 565                 entry->eax = min(entry->eax, 0x8000001a);
 566                 break;
 567         case 0x80000001:
 568                 entry->edx &= kvm_supported_word1_x86_features;
 569                 cpuid_mask(&entry->edx, 1);
 570                 entry->ecx &= kvm_supported_word6_x86_features;
 571                 cpuid_mask(&entry->ecx, 6);
 572                 break;
 573         case 0x80000007: /* Advanced power management */
 574                 /* invariant TSC is CPUID.80000007H:EDX[8] */
 575                 entry->edx &= (1 << 8);
 576                 /* mask against host */
 577                 entry->edx &= boot_cpu_data.x86_power;
 578                 entry->eax = entry->ebx = entry->ecx = 0;
 579                 break;
 580         case 0x80000008: {
 581                 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
 582                 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
 583                 unsigned phys_as = entry->eax & 0xff;
 584
 585                 if (!g_phys_as)
 586                         g_phys_as = phys_as;
 587                 entry->eax = g_phys_as | (virt_as << 8);
 588                 entry->ebx = entry->edx = 0;
 589                 break;
 590         }
 591         case 0x80000019:
 592                 entry->ecx = entry->edx = 0;
 593                 break;
 594         case 0x8000001a:
 595                 break;
 596         case 0x8000001d:
 597                 break;
 598         /*Add support for Centaur's CPUID instruction*/
 599         case 0xC0000000:
 600                 /*Just support up to 0xC0000004 now*/
 601                 entry->eax = min(entry->eax, 0xC0000004);
 602                 break;
 603         case 0xC0000001:
 604                 entry->edx &= kvm_supported_word5_x86_features;
 605                 cpuid_mask(&entry->edx, 5);
 606                 break;
 607         case 3: /* Processor serial number */
 608         case 5: /* MONITOR/MWAIT */
 609         case 0xC0000002:
 610         case 0xC0000003:
 611         case 0xC0000004:
 612         default:
 613                 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
 614                 break;
 615         }
 616
 617         kvm_x86_ops->set_supported_cpuid(function, entry);
 618
 619         r = 0;
 620
 621 out:
 622         put_cpu();
 623
 624         return r;
 625 }
 626
 627 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
 628                         u32 idx, int *nent, int maxnent, unsigned int type)
 629 {
 630         if (type == KVM_GET_EMULATED_CPUID)
 631                 return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
 632
 633         return __do_cpuid_ent(entry, func, idx, nent, maxnent);
 634 }
 635
 636 #undef F
 637
 638 struct kvm_cpuid_param {
 639         u32 func;
 640         u32 idx;
 641         bool has_leaf_count;
 642         bool (*qualifier)(const struct kvm_cpuid_param *param);
 643 };
 644
 645 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
 646 {
 647         return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
 648 }
 649
 650 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
 651                                  __u32 num_entries, unsigned int ioctl_type)
 652 {
 653         int i;
 654         __u32 pad[3];
 655
 656         if (ioctl_type != KVM_GET_EMULATED_CPUID)
 657                 return false;
 658
 659         /*
 660          * We want to make sure that ->padding is being passed clean from
 661          * userspace in case we want to use it for something in the future.
 662          *
 663          * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
 664          * have to give ourselves satisfied only with the emulated side. /me
 665          * sheds a tear.
 666          */
 667         for (i = 0; i < num_entries; i++) {
 668                 if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
 669                         return true;
 670
 671                 if (pad[0] || pad[1] || pad[2])
 672                         return true;
 673         }
 674         return false;
 675 }
 676
 677 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
 678                             struct kvm_cpuid_entry2 __user *entries,
 679                             unsigned int type)
 680 {
 681         struct kvm_cpuid_entry2 *cpuid_entries;
 682         int limit, nent = 0, r = -E2BIG, i;
 683         u32 func;
 684         static const struct kvm_cpuid_param param[] = {
 685                 { .func = 0, .has_leaf_count = true },
 686                 { .func = 0x80000000, .has_leaf_count = true },
 687                 { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
 688                 { .func = KVM_CPUID_SIGNATURE },
 689                 { .func = KVM_CPUID_FEATURES },
 690         };
 691
 692         if (cpuid->nent < 1)
 693                 goto out;
 694         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 695                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
 696
 697         if (sanity_check_entries(entries, cpuid->nent, type))
 698                 return -EINVAL;
 699
 700         r = -ENOMEM;
 701         cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
 702         if (!cpuid_entries)
 703                 goto out;
 704
 705         r = 0;
 706         for (i = 0; i < ARRAY_SIZE(param); i++) {
 707                 const struct kvm_cpuid_param *ent = &param[i];
 708
 709                 if (ent->qualifier && !ent->qualifier(ent))
 710                         continue;
 711
 712                 r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
 713                                 &nent, cpuid->nent, type);
 714
 715                 if (r)
 716                         goto out_free;
 717
 718                 if (!ent->has_leaf_count)
 719                         continue;
 720
 721                 limit = cpuid_entries[nent - 1].eax;
 722                 for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
 723                         r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
 724                                      &nent, cpuid->nent, type);
 725
 726                 if (r)
 727                         goto out_free;
 728         }
 729
 730         r = -EFAULT;
 731         if (copy_to_user(entries, cpuid_entries,
 732                          nent * sizeof(struct kvm_cpuid_entry2)))
 733                 goto out_free;
 734         cpuid->nent = nent;
 735         r = 0;
 736
 737 out_free:
 738         vfree(cpuid_entries);
 739 out:
 740         return r;
 741 }
 742
 743 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
 744 {
 745         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
 746         int j, nent = vcpu->arch.cpuid_nent;
 747
 748         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
 749         /* when no next entry is found, the current entry[i] is reselected */
 750         for (j = i + 1; ; j = (j + 1) % nent) {
 751                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
 752                 if (ej->function == e->function) {
 753                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
 754                         return j;
 755                 }
 756         }
 757         return 0; /* silence gcc, even though control never reaches here */
 758 }
 759
 760 /* find an entry with matching function, matching index (if needed), and that
 761  * should be read next (if it's stateful) */
 762 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
 763         u32 function, u32 index)
 764 {
 765         if (e->function != function)
 766                 return 0;
 767         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
 768                 return 0;
 769         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
 770             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
 771                 return 0;
 772         return 1;
 773 }
 774
 775 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
 776                                               u32 function, u32 index)
 777 {
 778         int i;
 779         struct kvm_cpuid_entry2 *best = NULL;
 780
 781         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 782                 struct kvm_cpuid_entry2 *e;
 783
 784                 e = &vcpu->arch.cpuid_entries[i];
 785                 if (is_matching_cpuid_entry(e, function, index)) {
 786                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
 787                                 move_to_next_stateful_cpuid_entry(vcpu, i);
 788                         best = e;
 789                         break;
 790                 }
 791         }
 792         return best;
 793 }
 794 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
 795
 796 /*
 797  * If no match is found, check whether we exceed the vCPU's limit
 798  * and return the content of the highest valid _standard_ leaf instead.
 799  * This is to satisfy the CPUID specification.
 800  */
 801 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
 802                                                   u32 function, u32 index)
 803 {
 804         struct kvm_cpuid_entry2 *maxlevel;
 805
 806         maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
 807         if (!maxlevel || maxlevel->eax >= function)
 808                 return NULL;
 809         if (function & 0x80000000) {
 810                 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
 811                 if (!maxlevel)
 812                         return NULL;
 813         }
 814         return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
 815 }
 816
 817 void kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
 818 {
 819         u32 function = *eax, index = *ecx;
 820         struct kvm_cpuid_entry2 *best;
 821
 822         best = kvm_find_cpuid_entry(vcpu, function, index);
 823
 824         if (!best)
 825                 best = check_cpuid_limit(vcpu, function, index);
 826
 827         /*
 828          * Perfmon not yet supported for L2 guest.
 829          */
 830         if (is_guest_mode(vcpu) && function == 0xa)
 831                 best = NULL;
 832
 833         if (best) {
 834                 *eax = best->eax;
 835                 *ebx = best->ebx;
 836                 *ecx = best->ecx;
 837                 *edx = best->edx;
 838         } else
 839                 *eax = *ebx = *ecx = *edx = 0;
 840         trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx);
 841 }
 842 EXPORT_SYMBOL_GPL(kvm_cpuid);
 843
 844 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
 845 {
 846         u32 function, eax, ebx, ecx, edx;
 847
 848         function = eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
 849         ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
 850         kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx);
 851         kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
 852         kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
 853         kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
 854         kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
 855         kvm_x86_ops->skip_emulated_instruction(vcpu);
 856 }
 857 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);