1 /* 2 * Kernel-based Virtual Machine driver for Linux 3 * cpuid support routines 4 * 5 * derived from arch/x86/kvm/x86.c 6 * 7 * Copyright 2011 Red Hat, Inc. and/or its affiliates. 8 * Copyright IBM Corporation, 2008 9 * 10 * This work is licensed under the terms of the GNU GPL, version 2. See 11 * the COPYING file in the top-level directory. 12 * 13 */ 14 15 #include <linux/kvm_host.h> 16 #include <linux/export.h> 17 #include <linux/vmalloc.h> 18 #include <linux/uaccess.h> 19 #include <linux/sched/stat.h> 20 21 #include <asm/processor.h> 22 #include <asm/user.h> 23 #include <asm/fpu/xstate.h> 24 #include "cpuid.h" 25 #include "lapic.h" 26 #include "mmu.h" 27 #include "trace.h" 28 #include "pmu.h" 29 30 static u32 xstate_required_size(u64 xstate_bv, bool compacted) 31 { 32 int feature_bit = 0; 33 u32 ret = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET; 34 35 xstate_bv &= XFEATURE_MASK_EXTEND; 36 while (xstate_bv) { 37 if (xstate_bv & 0x1) { 38 u32 eax, ebx, ecx, edx, offset; 39 cpuid_count(0xD, feature_bit, &eax, &ebx, &ecx, &edx); 40 offset = compacted ? ret : ebx; 41 ret = max(ret, offset + eax); 42 } 43 44 xstate_bv >>= 1; 45 feature_bit++; 46 } 47 48 return ret; 49 } 50 51 bool kvm_mpx_supported(void) 52 { 53 return ((host_xcr0 & (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR)) 54 && kvm_x86_ops->mpx_supported()); 55 } 56 EXPORT_SYMBOL_GPL(kvm_mpx_supported); 57 58 u64 kvm_supported_xcr0(void) 59 { 60 u64 xcr0 = KVM_SUPPORTED_XCR0 & host_xcr0; 61 62 if (!kvm_mpx_supported()) 63 xcr0 &= ~(XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR); 64 65 return xcr0; 66 } 67 68 #define F(x) bit(X86_FEATURE_##x) 69 70 /* For scattered features from cpufeatures.h; we currently expose none */ 71 #define KF(x) bit(KVM_CPUID_BIT_##x) 72 73 int kvm_update_cpuid(struct kvm_vcpu *vcpu) 74 { 75 struct kvm_cpuid_entry2 *best; 76 struct kvm_lapic *apic = vcpu->arch.apic; 77 78 best = kvm_find_cpuid_entry(vcpu, 1, 0); 79 if (!best) 80 return 0; 81 82 /* Update OSXSAVE bit */ 83 if (boot_cpu_has(X86_FEATURE_XSAVE) && best->function == 0x1) { 84 best->ecx &= ~F(OSXSAVE); 85 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) 86 best->ecx |= F(OSXSAVE); 87 } 88 89 best->edx &= ~F(APIC); 90 if (vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE) 91 best->edx |= F(APIC); 92 93 if (apic) { 94 if (best->ecx & F(TSC_DEADLINE_TIMER)) 95 apic->lapic_timer.timer_mode_mask = 3 << 17; 96 else 97 apic->lapic_timer.timer_mode_mask = 1 << 17; 98 } 99 100 best = kvm_find_cpuid_entry(vcpu, 7, 0); 101 if (best) { 102 /* Update OSPKE bit */ 103 if (boot_cpu_has(X86_FEATURE_PKU) && best->function == 0x7) { 104 best->ecx &= ~F(OSPKE); 105 if (kvm_read_cr4_bits(vcpu, X86_CR4_PKE)) 106 best->ecx |= F(OSPKE); 107 } 108 } 109 110 best = kvm_find_cpuid_entry(vcpu, 0xD, 0); 111 if (!best) { 112 vcpu->arch.guest_supported_xcr0 = 0; 113 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET; 114 } else { 115 vcpu->arch.guest_supported_xcr0 = 116 (best->eax | ((u64)best->edx << 32)) & 117 kvm_supported_xcr0(); 118 vcpu->arch.guest_xstate_size = best->ebx = 119 xstate_required_size(vcpu->arch.xcr0, false); 120 } 121 122 best = kvm_find_cpuid_entry(vcpu, 0xD, 1); 123 if (best && (best->eax & (F(XSAVES) | F(XSAVEC)))) 124 best->ebx = xstate_required_size(vcpu->arch.xcr0, true); 125 126 /* 127 * The existing code assumes virtual address is 48-bit or 57-bit in the 128 * canonical address checks; exit if it is ever changed. 129 */ 130 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0); 131 if (best) { 132 int vaddr_bits = (best->eax & 0xff00) >> 8; 133 134 if (vaddr_bits != 48 && vaddr_bits != 57 && vaddr_bits != 0) 135 return -EINVAL; 136 } 137 138 /* Update physical-address width */ 139 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu); 140 kvm_mmu_reset_context(vcpu); 141 142 kvm_pmu_refresh(vcpu); 143 return 0; 144 } 145 146 static int is_efer_nx(void) 147 { 148 unsigned long long efer = 0; 149 150 rdmsrl_safe(MSR_EFER, &efer); 151 return efer & EFER_NX; 152 } 153 154 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu) 155 { 156 int i; 157 struct kvm_cpuid_entry2 *e, *entry; 158 159 entry = NULL; 160 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) { 161 e = &vcpu->arch.cpuid_entries[i]; 162 if (e->function == 0x80000001) { 163 entry = e; 164 break; 165 } 166 } 167 if (entry && (entry->edx & F(NX)) && !is_efer_nx()) { 168 entry->edx &= ~F(NX); 169 printk(KERN_INFO "kvm: guest NX capability removed\n"); 170 } 171 } 172 173 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu) 174 { 175 struct kvm_cpuid_entry2 *best; 176 177 best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0); 178 if (!best || best->eax < 0x80000008) 179 goto not_found; 180 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0); 181 if (best) 182 return best->eax & 0xff; 183 not_found: 184 return 36; 185 } 186 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr); 187 188 /* when an old userspace process fills a new kernel module */ 189 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu, 190 struct kvm_cpuid *cpuid, 191 struct kvm_cpuid_entry __user *entries) 192 { 193 int r, i; 194 struct kvm_cpuid_entry *cpuid_entries = NULL; 195 196 r = -E2BIG; 197 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES) 198 goto out; 199 r = -ENOMEM; 200 if (cpuid->nent) { 201 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * 202 cpuid->nent); 203 if (!cpuid_entries) 204 goto out; 205 r = -EFAULT; 206 if (copy_from_user(cpuid_entries, entries, 207 cpuid->nent * sizeof(struct kvm_cpuid_entry))) 208 goto out; 209 } 210 for (i = 0; i < cpuid->nent; i++) { 211 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function; 212 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax; 213 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx; 214 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx; 215 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx; 216 vcpu->arch.cpuid_entries[i].index = 0; 217 vcpu->arch.cpuid_entries[i].flags = 0; 218 vcpu->arch.cpuid_entries[i].padding[0] = 0; 219 vcpu->arch.cpuid_entries[i].padding[1] = 0; 220 vcpu->arch.cpuid_entries[i].padding[2] = 0; 221 } 222 vcpu->arch.cpuid_nent = cpuid->nent; 223 cpuid_fix_nx_cap(vcpu); 224 kvm_apic_set_version(vcpu); 225 kvm_x86_ops->cpuid_update(vcpu); 226 r = kvm_update_cpuid(vcpu); 227 228 out: 229 vfree(cpuid_entries); 230 return r; 231 } 232 233 int kvm_vcpu_ioctl_set_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 > KVM_MAX_CPUID_ENTRIES) 241 goto out; 242 r = -EFAULT; 243 if (copy_from_user(&vcpu->arch.cpuid_entries, entries, 244 cpuid->nent * sizeof(struct kvm_cpuid_entry2))) 245 goto out; 246 vcpu->arch.cpuid_nent = cpuid->nent; 247 kvm_apic_set_version(vcpu); 248 kvm_x86_ops->cpuid_update(vcpu); 249 r = kvm_update_cpuid(vcpu); 250 out: 251 return r; 252 } 253 254 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu, 255 struct kvm_cpuid2 *cpuid, 256 struct kvm_cpuid_entry2 __user *entries) 257 { 258 int r; 259 260 r = -E2BIG; 261 if (cpuid->nent < vcpu->arch.cpuid_nent) 262 goto out; 263 r = -EFAULT; 264 if (copy_to_user(entries, &vcpu->arch.cpuid_entries, 265 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2))) 266 goto out; 267 return 0; 268 269 out: 270 cpuid->nent = vcpu->arch.cpuid_nent; 271 return r; 272 } 273 274 static void cpuid_mask(u32 *word, int wordnum) 275 { 276 *word &= boot_cpu_data.x86_capability[wordnum]; 277 } 278 279 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function, 280 u32 index) 281 { 282 entry->function = function; 283 entry->index = index; 284 cpuid_count(entry->function, entry->index, 285 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx); 286 entry->flags = 0; 287 } 288 289 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry, 290 u32 func, u32 index, int *nent, int maxnent) 291 { 292 switch (func) { 293 case 0: 294 entry->eax = 1; /* only one leaf currently */ 295 ++*nent; 296 break; 297 case 1: 298 entry->ecx = F(MOVBE); 299 ++*nent; 300 break; 301 default: 302 break; 303 } 304 305 entry->function = func; 306 entry->index = index; 307 308 return 0; 309 } 310 311 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function, 312 u32 index, int *nent, int maxnent) 313 { 314 int r; 315 unsigned f_nx = is_efer_nx() ? F(NX) : 0; 316 #ifdef CONFIG_X86_64 317 unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL) 318 ? F(GBPAGES) : 0; 319 unsigned f_lm = F(LM); 320 #else 321 unsigned f_gbpages = 0; 322 unsigned f_lm = 0; 323 #endif 324 unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0; 325 unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0; 326 unsigned f_mpx = kvm_mpx_supported() ? F(MPX) : 0; 327 unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0; 328 329 /* cpuid 1.edx */ 330 const u32 kvm_cpuid_1_edx_x86_features = 331 F(FPU) | F(VME) | F(DE) | F(PSE) | 332 F(TSC) | F(MSR) | F(PAE) | F(MCE) | 333 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) | 334 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) | 335 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) | 336 0 /* Reserved, DS, ACPI */ | F(MMX) | 337 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) | 338 0 /* HTT, TM, Reserved, PBE */; 339 /* cpuid 0x80000001.edx */ 340 const u32 kvm_cpuid_8000_0001_edx_x86_features = 341 F(FPU) | F(VME) | F(DE) | F(PSE) | 342 F(TSC) | F(MSR) | F(PAE) | F(MCE) | 343 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) | 344 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) | 345 F(PAT) | F(PSE36) | 0 /* Reserved */ | 346 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) | 347 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp | 348 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW); 349 /* cpuid 1.ecx */ 350 const u32 kvm_cpuid_1_ecx_x86_features = 351 /* NOTE: MONITOR (and MWAIT) are emulated as NOP, 352 * but *not* advertised to guests via CPUID ! */ 353 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ | 354 0 /* DS-CPL, VMX, SMX, EST */ | 355 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ | 356 F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ | 357 F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) | 358 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) | 359 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) | 360 F(F16C) | F(RDRAND); 361 /* cpuid 0x80000001.ecx */ 362 const u32 kvm_cpuid_8000_0001_ecx_x86_features = 363 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ | 364 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) | 365 F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) | 366 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM); 367 368 /* cpuid 0x80000008.ebx */ 369 const u32 kvm_cpuid_8000_0008_ebx_x86_features = 370 F(IBPB) | F(IBRS); 371 372 /* cpuid 0xC0000001.edx */ 373 const u32 kvm_cpuid_C000_0001_edx_x86_features = 374 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) | 375 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) | 376 F(PMM) | F(PMM_EN); 377 378 /* cpuid 7.0.ebx */ 379 const u32 kvm_cpuid_7_0_ebx_x86_features = 380 F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) | 381 F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) | 382 F(ADX) | F(SMAP) | F(AVX512IFMA) | F(AVX512F) | F(AVX512PF) | 383 F(AVX512ER) | F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) | 384 F(SHA_NI) | F(AVX512BW) | F(AVX512VL); 385 386 /* cpuid 0xD.1.eax */ 387 const u32 kvm_cpuid_D_1_eax_x86_features = 388 F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves; 389 390 /* cpuid 7.0.ecx*/ 391 const u32 kvm_cpuid_7_0_ecx_x86_features = 392 F(AVX512VBMI) | F(LA57) | F(PKU) | 393 0 /*OSPKE*/ | F(AVX512_VPOPCNTDQ); 394 395 /* cpuid 7.0.edx*/ 396 const u32 kvm_cpuid_7_0_edx_x86_features = 397 F(AVX512_4VNNIW) | F(AVX512_4FMAPS) | F(SPEC_CTRL) | 398 F(ARCH_CAPABILITIES); 399 400 /* all calls to cpuid_count() should be made on the same cpu */ 401 get_cpu(); 402 403 r = -E2BIG; 404 405 if (*nent >= maxnent) 406 goto out; 407 408 do_cpuid_1_ent(entry, function, index); 409 ++*nent; 410 411 switch (function) { 412 case 0: 413 entry->eax = min(entry->eax, (u32)0xd); 414 break; 415 case 1: 416 entry->edx &= kvm_cpuid_1_edx_x86_features; 417 cpuid_mask(&entry->edx, CPUID_1_EDX); 418 entry->ecx &= kvm_cpuid_1_ecx_x86_features; 419 cpuid_mask(&entry->ecx, CPUID_1_ECX); 420 /* we support x2apic emulation even if host does not support 421 * it since we emulate x2apic in software */ 422 entry->ecx |= F(X2APIC); 423 break; 424 /* function 2 entries are STATEFUL. That is, repeated cpuid commands 425 * may return different values. This forces us to get_cpu() before 426 * issuing the first command, and also to emulate this annoying behavior 427 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */ 428 case 2: { 429 int t, times = entry->eax & 0xff; 430 431 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC; 432 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT; 433 for (t = 1; t < times; ++t) { 434 if (*nent >= maxnent) 435 goto out; 436 437 do_cpuid_1_ent(&entry[t], function, 0); 438 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC; 439 ++*nent; 440 } 441 break; 442 } 443 /* function 4 has additional index. */ 444 case 4: { 445 int i, cache_type; 446 447 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; 448 /* read more entries until cache_type is zero */ 449 for (i = 1; ; ++i) { 450 if (*nent >= maxnent) 451 goto out; 452 453 cache_type = entry[i - 1].eax & 0x1f; 454 if (!cache_type) 455 break; 456 do_cpuid_1_ent(&entry[i], function, i); 457 entry[i].flags |= 458 KVM_CPUID_FLAG_SIGNIFCANT_INDEX; 459 ++*nent; 460 } 461 break; 462 } 463 case 6: /* Thermal management */ 464 entry->eax = 0x4; /* allow ARAT */ 465 entry->ebx = 0; 466 entry->ecx = 0; 467 entry->edx = 0; 468 break; 469 case 7: { 470 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; 471 /* Mask ebx against host capability word 9 */ 472 if (index == 0) { 473 entry->ebx &= kvm_cpuid_7_0_ebx_x86_features; 474 cpuid_mask(&entry->ebx, CPUID_7_0_EBX); 475 // TSC_ADJUST is emulated 476 entry->ebx |= F(TSC_ADJUST); 477 entry->ecx &= kvm_cpuid_7_0_ecx_x86_features; 478 cpuid_mask(&entry->ecx, CPUID_7_ECX); 479 /* PKU is not yet implemented for shadow paging. */ 480 if (!tdp_enabled || !boot_cpu_has(X86_FEATURE_OSPKE)) 481 entry->ecx &= ~F(PKU); 482 entry->edx &= kvm_cpuid_7_0_edx_x86_features; 483 cpuid_mask(&entry->edx, CPUID_7_EDX); 484 } else { 485 entry->ebx = 0; 486 entry->ecx = 0; 487 entry->edx = 0; 488 } 489 entry->eax = 0; 490 break; 491 } 492 case 9: 493 break; 494 case 0xa: { /* Architectural Performance Monitoring */ 495 struct x86_pmu_capability cap; 496 union cpuid10_eax eax; 497 union cpuid10_edx edx; 498 499 perf_get_x86_pmu_capability(&cap); 500 501 /* 502 * Only support guest architectural pmu on a host 503 * with architectural pmu. 504 */ 505 if (!cap.version) 506 memset(&cap, 0, sizeof(cap)); 507 508 eax.split.version_id = min(cap.version, 2); 509 eax.split.num_counters = cap.num_counters_gp; 510 eax.split.bit_width = cap.bit_width_gp; 511 eax.split.mask_length = cap.events_mask_len; 512 513 edx.split.num_counters_fixed = cap.num_counters_fixed; 514 edx.split.bit_width_fixed = cap.bit_width_fixed; 515 edx.split.reserved = 0; 516 517 entry->eax = eax.full; 518 entry->ebx = cap.events_mask; 519 entry->ecx = 0; 520 entry->edx = edx.full; 521 break; 522 } 523 /* function 0xb has additional index. */ 524 case 0xb: { 525 int i, level_type; 526 527 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; 528 /* read more entries until level_type is zero */ 529 for (i = 1; ; ++i) { 530 if (*nent >= maxnent) 531 goto out; 532 533 level_type = entry[i - 1].ecx & 0xff00; 534 if (!level_type) 535 break; 536 do_cpuid_1_ent(&entry[i], function, i); 537 entry[i].flags |= 538 KVM_CPUID_FLAG_SIGNIFCANT_INDEX; 539 ++*nent; 540 } 541 break; 542 } 543 case 0xd: { 544 int idx, i; 545 u64 supported = kvm_supported_xcr0(); 546 547 entry->eax &= supported; 548 entry->ebx = xstate_required_size(supported, false); 549 entry->ecx = entry->ebx; 550 entry->edx &= supported >> 32; 551 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; 552 if (!supported) 553 break; 554 555 for (idx = 1, i = 1; idx < 64; ++idx) { 556 u64 mask = ((u64)1 << idx); 557 if (*nent >= maxnent) 558 goto out; 559 560 do_cpuid_1_ent(&entry[i], function, idx); 561 if (idx == 1) { 562 entry[i].eax &= kvm_cpuid_D_1_eax_x86_features; 563 cpuid_mask(&entry[i].eax, CPUID_D_1_EAX); 564 entry[i].ebx = 0; 565 if (entry[i].eax & (F(XSAVES)|F(XSAVEC))) 566 entry[i].ebx = 567 xstate_required_size(supported, 568 true); 569 } else { 570 if (entry[i].eax == 0 || !(supported & mask)) 571 continue; 572 if (WARN_ON_ONCE(entry[i].ecx & 1)) 573 continue; 574 } 575 entry[i].ecx = 0; 576 entry[i].edx = 0; 577 entry[i].flags |= 578 KVM_CPUID_FLAG_SIGNIFCANT_INDEX; 579 ++*nent; 580 ++i; 581 } 582 break; 583 } 584 case KVM_CPUID_SIGNATURE: { 585 static const char signature[12] = "KVMKVMKVM\0\0"; 586 const u32 *sigptr = (const u32 *)signature; 587 entry->eax = KVM_CPUID_FEATURES; 588 entry->ebx = sigptr[0]; 589 entry->ecx = sigptr[1]; 590 entry->edx = sigptr[2]; 591 break; 592 } 593 case KVM_CPUID_FEATURES: 594 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) | 595 (1 << KVM_FEATURE_NOP_IO_DELAY) | 596 (1 << KVM_FEATURE_CLOCKSOURCE2) | 597 (1 << KVM_FEATURE_ASYNC_PF) | 598 (1 << KVM_FEATURE_PV_EOI) | 599 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) | 600 (1 << KVM_FEATURE_PV_UNHALT); 601 602 if (sched_info_on()) 603 entry->eax |= (1 << KVM_FEATURE_STEAL_TIME); 604 605 entry->ebx = 0; 606 entry->ecx = 0; 607 entry->edx = 0; 608 break; 609 case 0x80000000: 610 entry->eax = min(entry->eax, 0x8000001a); 611 break; 612 case 0x80000001: 613 entry->edx &= kvm_cpuid_8000_0001_edx_x86_features; 614 cpuid_mask(&entry->edx, CPUID_8000_0001_EDX); 615 entry->ecx &= kvm_cpuid_8000_0001_ecx_x86_features; 616 cpuid_mask(&entry->ecx, CPUID_8000_0001_ECX); 617 break; 618 case 0x80000007: /* Advanced power management */ 619 /* invariant TSC is CPUID.80000007H:EDX[8] */ 620 entry->edx &= (1 << 8); 621 /* mask against host */ 622 entry->edx &= boot_cpu_data.x86_power; 623 entry->eax = entry->ebx = entry->ecx = 0; 624 break; 625 case 0x80000008: { 626 unsigned g_phys_as = (entry->eax >> 16) & 0xff; 627 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U); 628 unsigned phys_as = entry->eax & 0xff; 629 630 if (!g_phys_as) 631 g_phys_as = phys_as; 632 entry->eax = g_phys_as | (virt_as << 8); 633 entry->edx = 0; 634 /* IBRS and IBPB aren't necessarily present in hardware cpuid */ 635 if (boot_cpu_has(X86_FEATURE_IBPB)) 636 entry->ebx |= F(IBPB); 637 if (boot_cpu_has(X86_FEATURE_IBRS)) 638 entry->ebx |= F(IBRS); 639 entry->ebx &= kvm_cpuid_8000_0008_ebx_x86_features; 640 cpuid_mask(&entry->ebx, CPUID_8000_0008_EBX); 641 break; 642 } 643 case 0x80000019: 644 entry->ecx = entry->edx = 0; 645 break; 646 case 0x8000001a: 647 break; 648 case 0x8000001d: 649 break; 650 /*Add support for Centaur's CPUID instruction*/ 651 case 0xC0000000: 652 /*Just support up to 0xC0000004 now*/ 653 entry->eax = min(entry->eax, 0xC0000004); 654 break; 655 case 0xC0000001: 656 entry->edx &= kvm_cpuid_C000_0001_edx_x86_features; 657 cpuid_mask(&entry->edx, CPUID_C000_0001_EDX); 658 break; 659 case 3: /* Processor serial number */ 660 case 5: /* MONITOR/MWAIT */ 661 case 0xC0000002: 662 case 0xC0000003: 663 case 0xC0000004: 664 default: 665 entry->eax = entry->ebx = entry->ecx = entry->edx = 0; 666 break; 667 } 668 669 kvm_x86_ops->set_supported_cpuid(function, entry); 670 671 r = 0; 672 673 out: 674 put_cpu(); 675 676 return r; 677 } 678 679 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func, 680 u32 idx, int *nent, int maxnent, unsigned int type) 681 { 682 if (type == KVM_GET_EMULATED_CPUID) 683 return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent); 684 685 return __do_cpuid_ent(entry, func, idx, nent, maxnent); 686 } 687 688 #undef F 689 690 struct kvm_cpuid_param { 691 u32 func; 692 u32 idx; 693 bool has_leaf_count; 694 bool (*qualifier)(const struct kvm_cpuid_param *param); 695 }; 696 697 static bool is_centaur_cpu(const struct kvm_cpuid_param *param) 698 { 699 return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR; 700 } 701 702 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries, 703 __u32 num_entries, unsigned int ioctl_type) 704 { 705 int i; 706 __u32 pad[3]; 707 708 if (ioctl_type != KVM_GET_EMULATED_CPUID) 709 return false; 710 711 /* 712 * We want to make sure that ->padding is being passed clean from 713 * userspace in case we want to use it for something in the future. 714 * 715 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we 716 * have to give ourselves satisfied only with the emulated side. /me 717 * sheds a tear. 718 */ 719 for (i = 0; i < num_entries; i++) { 720 if (copy_from_user(pad, entries[i].padding, sizeof(pad))) 721 return true; 722 723 if (pad[0] || pad[1] || pad[2]) 724 return true; 725 } 726 return false; 727 } 728 729 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid, 730 struct kvm_cpuid_entry2 __user *entries, 731 unsigned int type) 732 { 733 struct kvm_cpuid_entry2 *cpuid_entries; 734 int limit, nent = 0, r = -E2BIG, i; 735 u32 func; 736 static const struct kvm_cpuid_param param[] = { 737 { .func = 0, .has_leaf_count = true }, 738 { .func = 0x80000000, .has_leaf_count = true }, 739 { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true }, 740 { .func = KVM_CPUID_SIGNATURE }, 741 { .func = KVM_CPUID_FEATURES }, 742 }; 743 744 if (cpuid->nent < 1) 745 goto out; 746 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES) 747 cpuid->nent = KVM_MAX_CPUID_ENTRIES; 748 749 if (sanity_check_entries(entries, cpuid->nent, type)) 750 return -EINVAL; 751 752 r = -ENOMEM; 753 cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent); 754 if (!cpuid_entries) 755 goto out; 756 757 r = 0; 758 for (i = 0; i < ARRAY_SIZE(param); i++) { 759 const struct kvm_cpuid_param *ent = ¶m[i]; 760 761 if (ent->qualifier && !ent->qualifier(ent)) 762 continue; 763 764 r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx, 765 &nent, cpuid->nent, type); 766 767 if (r) 768 goto out_free; 769 770 if (!ent->has_leaf_count) 771 continue; 772 773 limit = cpuid_entries[nent - 1].eax; 774 for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func) 775 r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx, 776 &nent, cpuid->nent, type); 777 778 if (r) 779 goto out_free; 780 } 781 782 r = -EFAULT; 783 if (copy_to_user(entries, cpuid_entries, 784 nent * sizeof(struct kvm_cpuid_entry2))) 785 goto out_free; 786 cpuid->nent = nent; 787 r = 0; 788 789 out_free: 790 vfree(cpuid_entries); 791 out: 792 return r; 793 } 794 795 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i) 796 { 797 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i]; 798 struct kvm_cpuid_entry2 *ej; 799 int j = i; 800 int nent = vcpu->arch.cpuid_nent; 801 802 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT; 803 /* when no next entry is found, the current entry[i] is reselected */ 804 do { 805 j = (j + 1) % nent; 806 ej = &vcpu->arch.cpuid_entries[j]; 807 } while (ej->function != e->function); 808 809 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT; 810 811 return j; 812 } 813 814 /* find an entry with matching function, matching index (if needed), and that 815 * should be read next (if it's stateful) */ 816 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e, 817 u32 function, u32 index) 818 { 819 if (e->function != function) 820 return 0; 821 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index) 822 return 0; 823 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) && 824 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT)) 825 return 0; 826 return 1; 827 } 828 829 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu, 830 u32 function, u32 index) 831 { 832 int i; 833 struct kvm_cpuid_entry2 *best = NULL; 834 835 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) { 836 struct kvm_cpuid_entry2 *e; 837 838 e = &vcpu->arch.cpuid_entries[i]; 839 if (is_matching_cpuid_entry(e, function, index)) { 840 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) 841 move_to_next_stateful_cpuid_entry(vcpu, i); 842 best = e; 843 break; 844 } 845 } 846 return best; 847 } 848 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry); 849 850 /* 851 * If no match is found, check whether we exceed the vCPU's limit 852 * and return the content of the highest valid _standard_ leaf instead. 853 * This is to satisfy the CPUID specification. 854 */ 855 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu, 856 u32 function, u32 index) 857 { 858 struct kvm_cpuid_entry2 *maxlevel; 859 860 maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0); 861 if (!maxlevel || maxlevel->eax >= function) 862 return NULL; 863 if (function & 0x80000000) { 864 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0); 865 if (!maxlevel) 866 return NULL; 867 } 868 return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index); 869 } 870 871 bool kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, 872 u32 *ecx, u32 *edx, bool check_limit) 873 { 874 u32 function = *eax, index = *ecx; 875 struct kvm_cpuid_entry2 *best; 876 bool entry_found = true; 877 878 best = kvm_find_cpuid_entry(vcpu, function, index); 879 880 if (!best) { 881 entry_found = false; 882 if (!check_limit) 883 goto out; 884 885 best = check_cpuid_limit(vcpu, function, index); 886 } 887 888 out: 889 if (best) { 890 *eax = best->eax; 891 *ebx = best->ebx; 892 *ecx = best->ecx; 893 *edx = best->edx; 894 } else 895 *eax = *ebx = *ecx = *edx = 0; 896 trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx, entry_found); 897 return entry_found; 898 } 899 EXPORT_SYMBOL_GPL(kvm_cpuid); 900 901 int kvm_emulate_cpuid(struct kvm_vcpu *vcpu) 902 { 903 u32 eax, ebx, ecx, edx; 904 905 if (cpuid_fault_enabled(vcpu) && !kvm_require_cpl(vcpu, 0)) 906 return 1; 907 908 eax = kvm_register_read(vcpu, VCPU_REGS_RAX); 909 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX); 910 kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx, true); 911 kvm_register_write(vcpu, VCPU_REGS_RAX, eax); 912 kvm_register_write(vcpu, VCPU_REGS_RBX, ebx); 913 kvm_register_write(vcpu, VCPU_REGS_RCX, ecx); 914 kvm_register_write(vcpu, VCPU_REGS_RDX, edx); 915 return kvm_skip_emulated_instruction(vcpu); 916 } 917 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid); 918