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/xsave.h>
  21 #include "cpuid.h"
  22 #include "lapic.h"
  23 #include "mmu.h"
  24 #include "trace.h"
  25
  26 static u32 xstate_required_size(u64 xstate_bv)
  27 {
  28         int feature_bit = 0;
  29         u32 ret = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
  30
  31         xstate_bv &= ~XSTATE_FPSSE;
  32         while (xstate_bv) {
  33                 if (xstate_bv & 0x1) {
  34                         u32 eax, ebx, ecx, edx;
  35                         cpuid_count(0xD, feature_bit, &eax, &ebx, &ecx, &edx);
  36                         ret = max(ret, eax + ebx);
  37                 }
  38
  39                 xstate_bv >>= 1;
  40                 feature_bit++;
  41         }
  42
  43         return ret;
  44 }
  45
  46 void kvm_update_cpuid(struct kvm_vcpu *vcpu)
  47 {
  48         struct kvm_cpuid_entry2 *best;
  49         struct kvm_lapic *apic = vcpu->arch.apic;
  50
  51         best = kvm_find_cpuid_entry(vcpu, 1, 0);
  52         if (!best)
  53                 return;
  54
  55         /* Update OSXSAVE bit */
  56         if (cpu_has_xsave && best->function == 0x1) {
  57                 best->ecx &= ~(bit(X86_FEATURE_OSXSAVE));
  58                 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
  59                         best->ecx |= bit(X86_FEATURE_OSXSAVE);
  60         }
  61
  62         if (apic) {
  63                 if (best->ecx & bit(X86_FEATURE_TSC_DEADLINE_TIMER))
  64                         apic->lapic_timer.timer_mode_mask = 3 << 17;
  65                 else
  66                         apic->lapic_timer.timer_mode_mask = 1 << 17;
  67         }
  68
  69         best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
  70         if (!best) {
  71                 vcpu->arch.guest_supported_xcr0 = 0;
  72                 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
  73         } else {
  74                 vcpu->arch.guest_supported_xcr0 =
  75                         (best->eax | ((u64)best->edx << 32)) &
  76                         host_xcr0 & KVM_SUPPORTED_XCR0;
  77                 vcpu->arch.guest_xstate_size =
  78                         xstate_required_size(vcpu->arch.guest_supported_xcr0);
  79         }
  80
  81         kvm_pmu_cpuid_update(vcpu);
  82 }
  83
  84 static int is_efer_nx(void)
  85 {
  86         unsigned long long efer = 0;
  87
  88         rdmsrl_safe(MSR_EFER, &efer);
  89         return efer & EFER_NX;
  90 }
  91
  92 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
  93 {
  94         int i;
  95         struct kvm_cpuid_entry2 *e, *entry;
  96
  97         entry = NULL;
  98         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
  99                 e = &vcpu->arch.cpuid_entries[i];
 100                 if (e->function == 0x80000001) {
 101                         entry = e;
 102                         break;
 103                 }
 104         }
 105         if (entry && (entry->edx & (1 << 20)) && !is_efer_nx()) {
 106                 entry->edx &= ~(1 << 20);
 107                 printk(KERN_INFO "kvm: guest NX capability removed\n");
 108         }
 109 }
 110
 111 /* when an old userspace process fills a new kernel module */
 112 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
 113                              struct kvm_cpuid *cpuid,
 114                              struct kvm_cpuid_entry __user *entries)
 115 {
 116         int r, i;
 117         struct kvm_cpuid_entry *cpuid_entries;
 118
 119         r = -E2BIG;
 120         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 121                 goto out;
 122         r = -ENOMEM;
 123         cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
 124         if (!cpuid_entries)
 125                 goto out;
 126         r = -EFAULT;
 127         if (copy_from_user(cpuid_entries, entries,
 128                            cpuid->nent * sizeof(struct kvm_cpuid_entry)))
 129                 goto out_free;
 130         for (i = 0; i < cpuid->nent; i++) {
 131                 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
 132                 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
 133                 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
 134                 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
 135                 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
 136                 vcpu->arch.cpuid_entries[i].index = 0;
 137                 vcpu->arch.cpuid_entries[i].flags = 0;
 138                 vcpu->arch.cpuid_entries[i].padding[0] = 0;
 139                 vcpu->arch.cpuid_entries[i].padding[1] = 0;
 140                 vcpu->arch.cpuid_entries[i].padding[2] = 0;
 141         }
 142         vcpu->arch.cpuid_nent = cpuid->nent;
 143         cpuid_fix_nx_cap(vcpu);
 144         r = 0;
 145         kvm_apic_set_version(vcpu);
 146         kvm_x86_ops->cpuid_update(vcpu);
 147         kvm_update_cpuid(vcpu);
 148
 149 out_free:
 150         vfree(cpuid_entries);
 151 out:
 152         return r;
 153 }
 154
 155 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
 156                               struct kvm_cpuid2 *cpuid,
 157                               struct kvm_cpuid_entry2 __user *entries)
 158 {
 159         int r;
 160
 161         r = -E2BIG;
 162         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 163                 goto out;
 164         r = -EFAULT;
 165         if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
 166                            cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
 167                 goto out;
 168         vcpu->arch.cpuid_nent = cpuid->nent;
 169         kvm_apic_set_version(vcpu);
 170         kvm_x86_ops->cpuid_update(vcpu);
 171         kvm_update_cpuid(vcpu);
 172         return 0;
 173
 174 out:
 175         return r;
 176 }
 177
 178 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
 179                               struct kvm_cpuid2 *cpuid,
 180                               struct kvm_cpuid_entry2 __user *entries)
 181 {
 182         int r;
 183
 184         r = -E2BIG;
 185         if (cpuid->nent < vcpu->arch.cpuid_nent)
 186                 goto out;
 187         r = -EFAULT;
 188         if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
 189                          vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
 190                 goto out;
 191         return 0;
 192
 193 out:
 194         cpuid->nent = vcpu->arch.cpuid_nent;
 195         return r;
 196 }
 197
 198 static void cpuid_mask(u32 *word, int wordnum)
 199 {
 200         *word &= boot_cpu_data.x86_capability[wordnum];
 201 }
 202
 203 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 204                            u32 index)
 205 {
 206         entry->function = function;
 207         entry->index = index;
 208         cpuid_count(entry->function, entry->index,
 209                     &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
 210         entry->flags = 0;
 211 }
 212
 213 static bool supported_xcr0_bit(unsigned bit)
 214 {
 215         u64 mask = ((u64)1 << bit);
 216
 217         return mask & KVM_SUPPORTED_XCR0 & host_xcr0;
 218 }
 219
 220 #define F(x) bit(X86_FEATURE_##x)
 221
 222 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry,
 223                                    u32 func, u32 index, int *nent, int maxnent)
 224 {
 225         switch (func) {
 226         case 0:
 227                 entry->eax = 1;         /* only one leaf currently */
 228                 ++*nent;
 229                 break;
 230         case 1:
 231                 entry->ecx = F(MOVBE);
 232                 ++*nent;
 233                 break;
 234         default:
 235                 break;
 236         }
 237
 238         entry->function = func;
 239         entry->index = index;
 240
 241         return 0;
 242 }
 243
 244 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
 245                                  u32 index, int *nent, int maxnent)
 246 {
 247         int r;
 248         unsigned f_nx = is_efer_nx() ? F(NX) : 0;
 249 #ifdef CONFIG_X86_64
 250         unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
 251                                 ? F(GBPAGES) : 0;
 252         unsigned f_lm = F(LM);
 253 #else
 254         unsigned f_gbpages = 0;
 255         unsigned f_lm = 0;
 256 #endif
 257         unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
 258         unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
 259
 260         /* cpuid 1.edx */
 261         const u32 kvm_supported_word0_x86_features =
 262                 F(FPU) | F(VME) | F(DE) | F(PSE) |
 263                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
 264                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
 265                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
 266                 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLSH) |
 267                 0 /* Reserved, DS, ACPI */ | F(MMX) |
 268                 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
 269                 0 /* HTT, TM, Reserved, PBE */;
 270         /* cpuid 0x80000001.edx */
 271         const u32 kvm_supported_word1_x86_features =
 272                 F(FPU) | F(VME) | F(DE) | F(PSE) |
 273                 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
 274                 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
 275                 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
 276                 F(PAT) | F(PSE36) | 0 /* Reserved */ |
 277                 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
 278                 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
 279                 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
 280         /* cpuid 1.ecx */
 281         const u32 kvm_supported_word4_x86_features =
 282                 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
 283                 0 /* DS-CPL, VMX, SMX, EST */ |
 284                 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
 285                 F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
 286                 F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
 287                 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
 288                 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
 289                 F(F16C) | F(RDRAND);
 290         /* cpuid 0x80000001.ecx */
 291         const u32 kvm_supported_word6_x86_features =
 292                 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
 293                 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
 294                 F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
 295                 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
 296
 297         /* cpuid 0xC0000001.edx */
 298         const u32 kvm_supported_word5_x86_features =
 299                 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
 300                 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
 301                 F(PMM) | F(PMM_EN);
 302
 303         /* cpuid 7.0.ebx */
 304         const u32 kvm_supported_word9_x86_features =
 305                 F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
 306                 F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | F(RDSEED) |
 307                 F(ADX);
 308
 309         /* all calls to cpuid_count() should be made on the same cpu */
 310         get_cpu();
 311
 312         r = -E2BIG;
 313
 314         if (*nent >= maxnent)
 315                 goto out;
 316
 317         do_cpuid_1_ent(entry, function, index);
 318         ++*nent;
 319
 320         switch (function) {
 321         case 0:
 322                 entry->eax = min(entry->eax, (u32)0xd);
 323                 break;
 324         case 1:
 325                 entry->edx &= kvm_supported_word0_x86_features;
 326                 cpuid_mask(&entry->edx, 0);
 327                 entry->ecx &= kvm_supported_word4_x86_features;
 328                 cpuid_mask(&entry->ecx, 4);
 329                 /* we support x2apic emulation even if host does not support
 330                  * it since we emulate x2apic in software */
 331                 entry->ecx |= F(X2APIC);
 332                 break;
 333         /* function 2 entries are STATEFUL. That is, repeated cpuid commands
 334          * may return different values. This forces us to get_cpu() before
 335          * issuing the first command, and also to emulate this annoying behavior
 336          * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
 337         case 2: {
 338                 int t, times = entry->eax & 0xff;
 339
 340                 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 341                 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
 342                 for (t = 1; t < times; ++t) {
 343                         if (*nent >= maxnent)
 344                                 goto out;
 345
 346                         do_cpuid_1_ent(&entry[t], function, 0);
 347                         entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
 348                         ++*nent;
 349                 }
 350                 break;
 351         }
 352         /* function 4 has additional index. */
 353         case 4: {
 354                 int i, cache_type;
 355
 356                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 357                 /* read more entries until cache_type is zero */
 358                 for (i = 1; ; ++i) {
 359                         if (*nent >= maxnent)
 360                                 goto out;
 361
 362                         cache_type = entry[i - 1].eax & 0x1f;
 363                         if (!cache_type)
 364                                 break;
 365                         do_cpuid_1_ent(&entry[i], function, i);
 366                         entry[i].flags |=
 367                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 368                         ++*nent;
 369                 }
 370                 break;
 371         }
 372         case 7: {
 373                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 374                 /* Mask ebx against host capability word 9 */
 375                 if (index == 0) {
 376                         entry->ebx &= kvm_supported_word9_x86_features;
 377                         cpuid_mask(&entry->ebx, 9);
 378                         // TSC_ADJUST is emulated
 379                         entry->ebx |= F(TSC_ADJUST);
 380                 } else
 381                         entry->ebx = 0;
 382                 entry->eax = 0;
 383                 entry->ecx = 0;
 384                 entry->edx = 0;
 385                 break;
 386         }
 387         case 9:
 388                 break;
 389         case 0xa: { /* Architectural Performance Monitoring */
 390                 struct x86_pmu_capability cap;
 391                 union cpuid10_eax eax;
 392                 union cpuid10_edx edx;
 393
 394                 perf_get_x86_pmu_capability(&cap);
 395
 396                 /*
 397                  * Only support guest architectural pmu on a host
 398                  * with architectural pmu.
 399                  */
 400                 if (!cap.version)
 401                         memset(&cap, 0, sizeof(cap));
 402
 403                 eax.split.version_id = min(cap.version, 2);
 404                 eax.split.num_counters = cap.num_counters_gp;
 405                 eax.split.bit_width = cap.bit_width_gp;
 406                 eax.split.mask_length = cap.events_mask_len;
 407
 408                 edx.split.num_counters_fixed = cap.num_counters_fixed;
 409                 edx.split.bit_width_fixed = cap.bit_width_fixed;
 410                 edx.split.reserved = 0;
 411
 412                 entry->eax = eax.full;
 413                 entry->ebx = cap.events_mask;
 414                 entry->ecx = 0;
 415                 entry->edx = edx.full;
 416                 break;
 417         }
 418         /* function 0xb has additional index. */
 419         case 0xb: {
 420                 int i, level_type;
 421
 422                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 423                 /* read more entries until level_type is zero */
 424                 for (i = 1; ; ++i) {
 425                         if (*nent >= maxnent)
 426                                 goto out;
 427
 428                         level_type = entry[i - 1].ecx & 0xff00;
 429                         if (!level_type)
 430                                 break;
 431                         do_cpuid_1_ent(&entry[i], function, i);
 432                         entry[i].flags |=
 433                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 434                         ++*nent;
 435                 }
 436                 break;
 437         }
 438         case 0xd: {
 439                 int idx, i;
 440
 441                 entry->eax &= host_xcr0 & KVM_SUPPORTED_XCR0;
 442                 entry->edx &= (host_xcr0 & KVM_SUPPORTED_XCR0) >> 32;
 443                 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 444                 for (idx = 1, i = 1; idx < 64; ++idx) {
 445                         if (*nent >= maxnent)
 446                                 goto out;
 447
 448                         do_cpuid_1_ent(&entry[i], function, idx);
 449                         if (entry[i].eax == 0 || !supported_xcr0_bit(idx))
 450                                 continue;
 451                         entry[i].flags |=
 452                                KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
 453                         ++*nent;
 454                         ++i;
 455                 }
 456                 break;
 457         }
 458         case KVM_CPUID_SIGNATURE: {
 459                 static const char signature[12] = "KVMKVMKVM\0\0";
 460                 const u32 *sigptr = (const u32 *)signature;
 461                 entry->eax = KVM_CPUID_FEATURES;
 462                 entry->ebx = sigptr[0];
 463                 entry->ecx = sigptr[1];
 464                 entry->edx = sigptr[2];
 465                 break;
 466         }
 467         case KVM_CPUID_FEATURES:
 468                 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
 469                              (1 << KVM_FEATURE_NOP_IO_DELAY) |
 470                              (1 << KVM_FEATURE_CLOCKSOURCE2) |
 471                              (1 << KVM_FEATURE_ASYNC_PF) |
 472                              (1 << KVM_FEATURE_PV_EOI) |
 473                              (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
 474                              (1 << KVM_FEATURE_PV_UNHALT);
 475
 476                 if (sched_info_on())
 477                         entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
 478
 479                 entry->ebx = 0;
 480                 entry->ecx = 0;
 481                 entry->edx = 0;
 482                 break;
 483         case 0x80000000:
 484                 entry->eax = min(entry->eax, 0x8000001a);
 485                 break;
 486         case 0x80000001:
 487                 entry->edx &= kvm_supported_word1_x86_features;
 488                 cpuid_mask(&entry->edx, 1);
 489                 entry->ecx &= kvm_supported_word6_x86_features;
 490                 cpuid_mask(&entry->ecx, 6);
 491                 break;
 492         case 0x80000008: {
 493                 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
 494                 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
 495                 unsigned phys_as = entry->eax & 0xff;
 496
 497                 if (!g_phys_as)
 498                         g_phys_as = phys_as;
 499                 entry->eax = g_phys_as | (virt_as << 8);
 500                 entry->ebx = entry->edx = 0;
 501                 break;
 502         }
 503         case 0x80000019:
 504                 entry->ecx = entry->edx = 0;
 505                 break;
 506         case 0x8000001a:
 507                 break;
 508         case 0x8000001d:
 509                 break;
 510         /*Add support for Centaur's CPUID instruction*/
 511         case 0xC0000000:
 512                 /*Just support up to 0xC0000004 now*/
 513                 entry->eax = min(entry->eax, 0xC0000004);
 514                 break;
 515         case 0xC0000001:
 516                 entry->edx &= kvm_supported_word5_x86_features;
 517                 cpuid_mask(&entry->edx, 5);
 518                 break;
 519         case 3: /* Processor serial number */
 520         case 5: /* MONITOR/MWAIT */
 521         case 6: /* Thermal management */
 522         case 0x80000007: /* Advanced power management */
 523         case 0xC0000002:
 524         case 0xC0000003:
 525         case 0xC0000004:
 526         default:
 527                 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
 528                 break;
 529         }
 530
 531         kvm_x86_ops->set_supported_cpuid(function, entry);
 532
 533         r = 0;
 534
 535 out:
 536         put_cpu();
 537
 538         return r;
 539 }
 540
 541 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
 542                         u32 idx, int *nent, int maxnent, unsigned int type)
 543 {
 544         if (type == KVM_GET_EMULATED_CPUID)
 545                 return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
 546
 547         return __do_cpuid_ent(entry, func, idx, nent, maxnent);
 548 }
 549
 550 #undef F
 551
 552 struct kvm_cpuid_param {
 553         u32 func;
 554         u32 idx;
 555         bool has_leaf_count;
 556         bool (*qualifier)(const struct kvm_cpuid_param *param);
 557 };
 558
 559 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
 560 {
 561         return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
 562 }
 563
 564 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
 565                                  __u32 num_entries, unsigned int ioctl_type)
 566 {
 567         int i;
 568         __u32 pad[3];
 569
 570         if (ioctl_type != KVM_GET_EMULATED_CPUID)
 571                 return false;
 572
 573         /*
 574          * We want to make sure that ->padding is being passed clean from
 575          * userspace in case we want to use it for something in the future.
 576          *
 577          * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
 578          * have to give ourselves satisfied only with the emulated side. /me
 579          * sheds a tear.
 580          */
 581         for (i = 0; i < num_entries; i++) {
 582                 if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
 583                         return true;
 584
 585                 if (pad[0] || pad[1] || pad[2])
 586                         return true;
 587         }
 588         return false;
 589 }
 590
 591 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
 592                             struct kvm_cpuid_entry2 __user *entries,
 593                             unsigned int type)
 594 {
 595         struct kvm_cpuid_entry2 *cpuid_entries;
 596         int limit, nent = 0, r = -E2BIG, i;
 597         u32 func;
 598         static const struct kvm_cpuid_param param[] = {
 599                 { .func = 0, .has_leaf_count = true },
 600                 { .func = 0x80000000, .has_leaf_count = true },
 601                 { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
 602                 { .func = KVM_CPUID_SIGNATURE },
 603                 { .func = KVM_CPUID_FEATURES },
 604         };
 605
 606         if (cpuid->nent < 1)
 607                 goto out;
 608         if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
 609                 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
 610
 611         if (sanity_check_entries(entries, cpuid->nent, type))
 612                 return -EINVAL;
 613
 614         r = -ENOMEM;
 615         cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
 616         if (!cpuid_entries)
 617                 goto out;
 618
 619         r = 0;
 620         for (i = 0; i < ARRAY_SIZE(param); i++) {
 621                 const struct kvm_cpuid_param *ent = &param[i];
 622
 623                 if (ent->qualifier && !ent->qualifier(ent))
 624                         continue;
 625
 626                 r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
 627                                 &nent, cpuid->nent, type);
 628
 629                 if (r)
 630                         goto out_free;
 631
 632                 if (!ent->has_leaf_count)
 633                         continue;
 634
 635                 limit = cpuid_entries[nent - 1].eax;
 636                 for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
 637                         r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
 638                                      &nent, cpuid->nent, type);
 639
 640                 if (r)
 641                         goto out_free;
 642         }
 643
 644         r = -EFAULT;
 645         if (copy_to_user(entries, cpuid_entries,
 646                          nent * sizeof(struct kvm_cpuid_entry2)))
 647                 goto out_free;
 648         cpuid->nent = nent;
 649         r = 0;
 650
 651 out_free:
 652         vfree(cpuid_entries);
 653 out:
 654         return r;
 655 }
 656
 657 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
 658 {
 659         struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
 660         int j, nent = vcpu->arch.cpuid_nent;
 661
 662         e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
 663         /* when no next entry is found, the current entry[i] is reselected */
 664         for (j = i + 1; ; j = (j + 1) % nent) {
 665                 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
 666                 if (ej->function == e->function) {
 667                         ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
 668                         return j;
 669                 }
 670         }
 671         return 0; /* silence gcc, even though control never reaches here */
 672 }
 673
 674 /* find an entry with matching function, matching index (if needed), and that
 675  * should be read next (if it's stateful) */
 676 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
 677         u32 function, u32 index)
 678 {
 679         if (e->function != function)
 680                 return 0;
 681         if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
 682                 return 0;
 683         if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
 684             !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
 685                 return 0;
 686         return 1;
 687 }
 688
 689 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
 690                                               u32 function, u32 index)
 691 {
 692         int i;
 693         struct kvm_cpuid_entry2 *best = NULL;
 694
 695         for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
 696                 struct kvm_cpuid_entry2 *e;
 697
 698                 e = &vcpu->arch.cpuid_entries[i];
 699                 if (is_matching_cpuid_entry(e, function, index)) {
 700                         if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
 701                                 move_to_next_stateful_cpuid_entry(vcpu, i);
 702                         best = e;
 703                         break;
 704                 }
 705         }
 706         return best;
 707 }
 708 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
 709
 710 int cpuid_maxphyaddr(struct kvm_vcpu *vcpu)
 711 {
 712         struct kvm_cpuid_entry2 *best;
 713
 714         best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
 715         if (!best || best->eax < 0x80000008)
 716                 goto not_found;
 717         best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
 718         if (best)
 719                 return best->eax & 0xff;
 720 not_found:
 721         return 36;
 722 }
 723
 724 /*
 725  * If no match is found, check whether we exceed the vCPU's limit
 726  * and return the content of the highest valid _standard_ leaf instead.
 727  * This is to satisfy the CPUID specification.
 728  */
 729 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
 730                                                   u32 function, u32 index)
 731 {
 732         struct kvm_cpuid_entry2 *maxlevel;
 733
 734         maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
 735         if (!maxlevel || maxlevel->eax >= function)
 736                 return NULL;
 737         if (function & 0x80000000) {
 738                 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
 739                 if (!maxlevel)
 740                         return NULL;
 741         }
 742         return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
 743 }
 744
 745 void kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
 746 {
 747         u32 function = *eax, index = *ecx;
 748         struct kvm_cpuid_entry2 *best;
 749
 750         best = kvm_find_cpuid_entry(vcpu, function, index);
 751
 752         if (!best)
 753                 best = check_cpuid_limit(vcpu, function, index);
 754
 755         if (best) {
 756                 *eax = best->eax;
 757                 *ebx = best->ebx;
 758                 *ecx = best->ecx;
 759                 *edx = best->edx;
 760         } else
 761                 *eax = *ebx = *ecx = *edx = 0;
 762         trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx);
 763 }
 764 EXPORT_SYMBOL_GPL(kvm_cpuid);
 765
 766 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
 767 {
 768         u32 function, eax, ebx, ecx, edx;
 769
 770         function = eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
 771         ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
 772         kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx);
 773         kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
 774         kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
 775         kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
 776         kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
 777         kvm_x86_ops->skip_emulated_instruction(vcpu);
 778 }
 779 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);