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 best = kvm_find_cpuid_entry(vcpu, KVM_CPUID_FEATURES, 0); 139 if (kvm_hlt_in_guest(vcpu->kvm) && best && 140 (best->eax & (1 << KVM_FEATURE_PV_UNHALT))) 141 best->eax &= ~(1 << KVM_FEATURE_PV_UNHALT); 142 143 /* Update physical-address width */ 144 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu); 145 kvm_mmu_reset_context(vcpu); 146 147 kvm_pmu_refresh(vcpu); 148 return 0; 149 } 150 151 static int is_efer_nx(void) 152 { 153 unsigned long long efer = 0; 154 155 rdmsrl_safe(MSR_EFER, &efer); 156 return efer & EFER_NX; 157 } 158 159 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu) 160 { 161 int i; 162 struct kvm_cpuid_entry2 *e, *entry; 163 164 entry = NULL; 165 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) { 166 e = &vcpu->arch.cpuid_entries[i]; 167 if (e->function == 0x80000001) { 168 entry = e; 169 break; 170 } 171 } 172 if (entry && (entry->edx & F(NX)) && !is_efer_nx()) { 173 entry->edx &= ~F(NX); 174 printk(KERN_INFO "kvm: guest NX capability removed\n"); 175 } 176 } 177 178 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu) 179 { 180 struct kvm_cpuid_entry2 *best; 181 182 best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0); 183 if (!best || best->eax < 0x80000008) 184 goto not_found; 185 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0); 186 if (best) 187 return best->eax & 0xff; 188 not_found: 189 return 36; 190 } 191 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr); 192 193 /* when an old userspace process fills a new kernel module */ 194 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu, 195 struct kvm_cpuid *cpuid, 196 struct kvm_cpuid_entry __user *entries) 197 { 198 int r, i; 199 struct kvm_cpuid_entry *cpuid_entries = NULL; 200 201 r = -E2BIG; 202 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES) 203 goto out; 204 r = -ENOMEM; 205 if (cpuid->nent) { 206 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * 207 cpuid->nent); 208 if (!cpuid_entries) 209 goto out; 210 r = -EFAULT; 211 if (copy_from_user(cpuid_entries, entries, 212 cpuid->nent * sizeof(struct kvm_cpuid_entry))) 213 goto out; 214 } 215 for (i = 0; i < cpuid->nent; i++) { 216 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function; 217 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax; 218 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx; 219 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx; 220 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx; 221 vcpu->arch.cpuid_entries[i].index = 0; 222 vcpu->arch.cpuid_entries[i].flags = 0; 223 vcpu->arch.cpuid_entries[i].padding[0] = 0; 224 vcpu->arch.cpuid_entries[i].padding[1] = 0; 225 vcpu->arch.cpuid_entries[i].padding[2] = 0; 226 } 227 vcpu->arch.cpuid_nent = cpuid->nent; 228 cpuid_fix_nx_cap(vcpu); 229 kvm_apic_set_version(vcpu); 230 kvm_x86_ops->cpuid_update(vcpu); 231 r = kvm_update_cpuid(vcpu); 232 233 out: 234 vfree(cpuid_entries); 235 return r; 236 } 237 238 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu, 239 struct kvm_cpuid2 *cpuid, 240 struct kvm_cpuid_entry2 __user *entries) 241 { 242 int r; 243 244 r = -E2BIG; 245 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES) 246 goto out; 247 r = -EFAULT; 248 if (copy_from_user(&vcpu->arch.cpuid_entries, entries, 249 cpuid->nent * sizeof(struct kvm_cpuid_entry2))) 250 goto out; 251 vcpu->arch.cpuid_nent = cpuid->nent; 252 kvm_apic_set_version(vcpu); 253 kvm_x86_ops->cpuid_update(vcpu); 254 r = kvm_update_cpuid(vcpu); 255 out: 256 return r; 257 } 258 259 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu, 260 struct kvm_cpuid2 *cpuid, 261 struct kvm_cpuid_entry2 __user *entries) 262 { 263 int r; 264 265 r = -E2BIG; 266 if (cpuid->nent < vcpu->arch.cpuid_nent) 267 goto out; 268 r = -EFAULT; 269 if (copy_to_user(entries, &vcpu->arch.cpuid_entries, 270 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2))) 271 goto out; 272 return 0; 273 274 out: 275 cpuid->nent = vcpu->arch.cpuid_nent; 276 return r; 277 } 278 279 static void cpuid_mask(u32 *word, int wordnum) 280 { 281 *word &= boot_cpu_data.x86_capability[wordnum]; 282 } 283 284 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function, 285 u32 index) 286 { 287 entry->function = function; 288 entry->index = index; 289 cpuid_count(entry->function, entry->index, 290 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx); 291 entry->flags = 0; 292 } 293 294 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry, 295 u32 func, u32 index, int *nent, int maxnent) 296 { 297 switch (func) { 298 case 0: 299 entry->eax = 7; 300 ++*nent; 301 break; 302 case 1: 303 entry->ecx = F(MOVBE); 304 ++*nent; 305 break; 306 case 7: 307 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; 308 if (index == 0) 309 entry->ecx = F(RDPID); 310 ++*nent; 311 default: 312 break; 313 } 314 315 entry->function = func; 316 entry->index = index; 317 318 return 0; 319 } 320 321 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function, 322 u32 index, int *nent, int maxnent) 323 { 324 int r; 325 unsigned f_nx = is_efer_nx() ? F(NX) : 0; 326 #ifdef CONFIG_X86_64 327 unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL) 328 ? F(GBPAGES) : 0; 329 unsigned f_lm = F(LM); 330 #else 331 unsigned f_gbpages = 0; 332 unsigned f_lm = 0; 333 #endif 334 unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0; 335 unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0; 336 unsigned f_mpx = kvm_mpx_supported() ? F(MPX) : 0; 337 unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0; 338 unsigned f_umip = kvm_x86_ops->umip_emulated() ? F(UMIP) : 0; 339 340 /* cpuid 1.edx */ 341 const u32 kvm_cpuid_1_edx_x86_features = 342 F(FPU) | F(VME) | F(DE) | F(PSE) | 343 F(TSC) | F(MSR) | F(PAE) | F(MCE) | 344 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) | 345 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) | 346 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) | 347 0 /* Reserved, DS, ACPI */ | F(MMX) | 348 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) | 349 0 /* HTT, TM, Reserved, PBE */; 350 /* cpuid 0x80000001.edx */ 351 const u32 kvm_cpuid_8000_0001_edx_x86_features = 352 F(FPU) | F(VME) | F(DE) | F(PSE) | 353 F(TSC) | F(MSR) | F(PAE) | F(MCE) | 354 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) | 355 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) | 356 F(PAT) | F(PSE36) | 0 /* Reserved */ | 357 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) | 358 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp | 359 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW); 360 /* cpuid 1.ecx */ 361 const u32 kvm_cpuid_1_ecx_x86_features = 362 /* NOTE: MONITOR (and MWAIT) are emulated as NOP, 363 * but *not* advertised to guests via CPUID ! */ 364 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ | 365 0 /* DS-CPL, VMX, SMX, EST */ | 366 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ | 367 F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ | 368 F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) | 369 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) | 370 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) | 371 F(F16C) | F(RDRAND); 372 /* cpuid 0x80000001.ecx */ 373 const u32 kvm_cpuid_8000_0001_ecx_x86_features = 374 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ | 375 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) | 376 F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) | 377 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM) | 378 F(TOPOEXT) | F(PERFCTR_CORE); 379 380 /* cpuid 0x80000008.ebx */ 381 const u32 kvm_cpuid_8000_0008_ebx_x86_features = 382 F(AMD_IBPB) | F(AMD_IBRS) | F(AMD_SSBD) | F(VIRT_SSBD) | 383 F(AMD_SSB_NO); 384 385 /* cpuid 0xC0000001.edx */ 386 const u32 kvm_cpuid_C000_0001_edx_x86_features = 387 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) | 388 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) | 389 F(PMM) | F(PMM_EN); 390 391 /* cpuid 7.0.ebx */ 392 const u32 kvm_cpuid_7_0_ebx_x86_features = 393 F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) | 394 F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) | 395 F(ADX) | F(SMAP) | F(AVX512IFMA) | F(AVX512F) | F(AVX512PF) | 396 F(AVX512ER) | F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) | 397 F(SHA_NI) | F(AVX512BW) | F(AVX512VL); 398 399 /* cpuid 0xD.1.eax */ 400 const u32 kvm_cpuid_D_1_eax_x86_features = 401 F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves; 402 403 /* cpuid 7.0.ecx*/ 404 const u32 kvm_cpuid_7_0_ecx_x86_features = 405 F(AVX512VBMI) | F(LA57) | F(PKU) | 0 /*OSPKE*/ | 406 F(AVX512_VPOPCNTDQ) | F(UMIP) | F(AVX512_VBMI2) | F(GFNI) | 407 F(VAES) | F(VPCLMULQDQ) | F(AVX512_VNNI) | F(AVX512_BITALG); 408 409 /* cpuid 7.0.edx*/ 410 const u32 kvm_cpuid_7_0_edx_x86_features = 411 F(AVX512_4VNNIW) | F(AVX512_4FMAPS) | F(SPEC_CTRL) | 412 F(SPEC_CTRL_SSBD) | F(ARCH_CAPABILITIES); 413 414 /* all calls to cpuid_count() should be made on the same cpu */ 415 get_cpu(); 416 417 r = -E2BIG; 418 419 if (*nent >= maxnent) 420 goto out; 421 422 do_cpuid_1_ent(entry, function, index); 423 ++*nent; 424 425 switch (function) { 426 case 0: 427 entry->eax = min(entry->eax, (u32)0xd); 428 break; 429 case 1: 430 entry->edx &= kvm_cpuid_1_edx_x86_features; 431 cpuid_mask(&entry->edx, CPUID_1_EDX); 432 entry->ecx &= kvm_cpuid_1_ecx_x86_features; 433 cpuid_mask(&entry->ecx, CPUID_1_ECX); 434 /* we support x2apic emulation even if host does not support 435 * it since we emulate x2apic in software */ 436 entry->ecx |= F(X2APIC); 437 break; 438 /* function 2 entries are STATEFUL. That is, repeated cpuid commands 439 * may return different values. This forces us to get_cpu() before 440 * issuing the first command, and also to emulate this annoying behavior 441 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */ 442 case 2: { 443 int t, times = entry->eax & 0xff; 444 445 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC; 446 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT; 447 for (t = 1; t < times; ++t) { 448 if (*nent >= maxnent) 449 goto out; 450 451 do_cpuid_1_ent(&entry[t], function, 0); 452 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC; 453 ++*nent; 454 } 455 break; 456 } 457 /* function 4 has additional index. */ 458 case 4: { 459 int i, cache_type; 460 461 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; 462 /* read more entries until cache_type is zero */ 463 for (i = 1; ; ++i) { 464 if (*nent >= maxnent) 465 goto out; 466 467 cache_type = entry[i - 1].eax & 0x1f; 468 if (!cache_type) 469 break; 470 do_cpuid_1_ent(&entry[i], function, i); 471 entry[i].flags |= 472 KVM_CPUID_FLAG_SIGNIFCANT_INDEX; 473 ++*nent; 474 } 475 break; 476 } 477 case 6: /* Thermal management */ 478 entry->eax = 0x4; /* allow ARAT */ 479 entry->ebx = 0; 480 entry->ecx = 0; 481 entry->edx = 0; 482 break; 483 case 7: { 484 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; 485 /* Mask ebx against host capability word 9 */ 486 if (index == 0) { 487 entry->ebx &= kvm_cpuid_7_0_ebx_x86_features; 488 cpuid_mask(&entry->ebx, CPUID_7_0_EBX); 489 // TSC_ADJUST is emulated 490 entry->ebx |= F(TSC_ADJUST); 491 entry->ecx &= kvm_cpuid_7_0_ecx_x86_features; 492 cpuid_mask(&entry->ecx, CPUID_7_ECX); 493 entry->ecx |= f_umip; 494 /* PKU is not yet implemented for shadow paging. */ 495 if (!tdp_enabled || !boot_cpu_has(X86_FEATURE_OSPKE)) 496 entry->ecx &= ~F(PKU); 497 entry->edx &= kvm_cpuid_7_0_edx_x86_features; 498 cpuid_mask(&entry->edx, CPUID_7_EDX); 499 /* 500 * We emulate ARCH_CAPABILITIES in software even 501 * if the host doesn't support it. 502 */ 503 entry->edx |= F(ARCH_CAPABILITIES); 504 } else { 505 entry->ebx = 0; 506 entry->ecx = 0; 507 entry->edx = 0; 508 } 509 entry->eax = 0; 510 break; 511 } 512 case 9: 513 break; 514 case 0xa: { /* Architectural Performance Monitoring */ 515 struct x86_pmu_capability cap; 516 union cpuid10_eax eax; 517 union cpuid10_edx edx; 518 519 perf_get_x86_pmu_capability(&cap); 520 521 /* 522 * Only support guest architectural pmu on a host 523 * with architectural pmu. 524 */ 525 if (!cap.version) 526 memset(&cap, 0, sizeof(cap)); 527 528 eax.split.version_id = min(cap.version, 2); 529 eax.split.num_counters = cap.num_counters_gp; 530 eax.split.bit_width = cap.bit_width_gp; 531 eax.split.mask_length = cap.events_mask_len; 532 533 edx.split.num_counters_fixed = cap.num_counters_fixed; 534 edx.split.bit_width_fixed = cap.bit_width_fixed; 535 edx.split.reserved = 0; 536 537 entry->eax = eax.full; 538 entry->ebx = cap.events_mask; 539 entry->ecx = 0; 540 entry->edx = edx.full; 541 break; 542 } 543 /* function 0xb has additional index. */ 544 case 0xb: { 545 int i, level_type; 546 547 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; 548 /* read more entries until level_type is zero */ 549 for (i = 1; ; ++i) { 550 if (*nent >= maxnent) 551 goto out; 552 553 level_type = entry[i - 1].ecx & 0xff00; 554 if (!level_type) 555 break; 556 do_cpuid_1_ent(&entry[i], function, i); 557 entry[i].flags |= 558 KVM_CPUID_FLAG_SIGNIFCANT_INDEX; 559 ++*nent; 560 } 561 break; 562 } 563 case 0xd: { 564 int idx, i; 565 u64 supported = kvm_supported_xcr0(); 566 567 entry->eax &= supported; 568 entry->ebx = xstate_required_size(supported, false); 569 entry->ecx = entry->ebx; 570 entry->edx &= supported >> 32; 571 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; 572 if (!supported) 573 break; 574 575 for (idx = 1, i = 1; idx < 64; ++idx) { 576 u64 mask = ((u64)1 << idx); 577 if (*nent >= maxnent) 578 goto out; 579 580 do_cpuid_1_ent(&entry[i], function, idx); 581 if (idx == 1) { 582 entry[i].eax &= kvm_cpuid_D_1_eax_x86_features; 583 cpuid_mask(&entry[i].eax, CPUID_D_1_EAX); 584 entry[i].ebx = 0; 585 if (entry[i].eax & (F(XSAVES)|F(XSAVEC))) 586 entry[i].ebx = 587 xstate_required_size(supported, 588 true); 589 } else { 590 if (entry[i].eax == 0 || !(supported & mask)) 591 continue; 592 if (WARN_ON_ONCE(entry[i].ecx & 1)) 593 continue; 594 } 595 entry[i].ecx = 0; 596 entry[i].edx = 0; 597 entry[i].flags |= 598 KVM_CPUID_FLAG_SIGNIFCANT_INDEX; 599 ++*nent; 600 ++i; 601 } 602 break; 603 } 604 case KVM_CPUID_SIGNATURE: { 605 static const char signature[12] = "KVMKVMKVM\0\0"; 606 const u32 *sigptr = (const u32 *)signature; 607 entry->eax = KVM_CPUID_FEATURES; 608 entry->ebx = sigptr[0]; 609 entry->ecx = sigptr[1]; 610 entry->edx = sigptr[2]; 611 break; 612 } 613 case KVM_CPUID_FEATURES: 614 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) | 615 (1 << KVM_FEATURE_NOP_IO_DELAY) | 616 (1 << KVM_FEATURE_CLOCKSOURCE2) | 617 (1 << KVM_FEATURE_ASYNC_PF) | 618 (1 << KVM_FEATURE_PV_EOI) | 619 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) | 620 (1 << KVM_FEATURE_PV_UNHALT) | 621 (1 << KVM_FEATURE_PV_TLB_FLUSH) | 622 (1 << KVM_FEATURE_ASYNC_PF_VMEXIT); 623 624 if (sched_info_on()) 625 entry->eax |= (1 << KVM_FEATURE_STEAL_TIME); 626 627 entry->ebx = 0; 628 entry->ecx = 0; 629 entry->edx = 0; 630 break; 631 case 0x80000000: 632 entry->eax = min(entry->eax, 0x8000001f); 633 break; 634 case 0x80000001: 635 entry->edx &= kvm_cpuid_8000_0001_edx_x86_features; 636 cpuid_mask(&entry->edx, CPUID_8000_0001_EDX); 637 entry->ecx &= kvm_cpuid_8000_0001_ecx_x86_features; 638 cpuid_mask(&entry->ecx, CPUID_8000_0001_ECX); 639 break; 640 case 0x80000007: /* Advanced power management */ 641 /* invariant TSC is CPUID.80000007H:EDX[8] */ 642 entry->edx &= (1 << 8); 643 /* mask against host */ 644 entry->edx &= boot_cpu_data.x86_power; 645 entry->eax = entry->ebx = entry->ecx = 0; 646 break; 647 case 0x80000008: { 648 unsigned g_phys_as = (entry->eax >> 16) & 0xff; 649 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U); 650 unsigned phys_as = entry->eax & 0xff; 651 652 if (!g_phys_as) 653 g_phys_as = phys_as; 654 entry->eax = g_phys_as | (virt_as << 8); 655 entry->edx = 0; 656 /* 657 * IBRS, IBPB and VIRT_SSBD aren't necessarily present in 658 * hardware cpuid 659 */ 660 if (boot_cpu_has(X86_FEATURE_AMD_IBPB)) 661 entry->ebx |= F(AMD_IBPB); 662 if (boot_cpu_has(X86_FEATURE_AMD_IBRS)) 663 entry->ebx |= F(AMD_IBRS); 664 if (boot_cpu_has(X86_FEATURE_VIRT_SSBD)) 665 entry->ebx |= F(VIRT_SSBD); 666 entry->ebx &= kvm_cpuid_8000_0008_ebx_x86_features; 667 cpuid_mask(&entry->ebx, CPUID_8000_0008_EBX); 668 /* 669 * The preference is to use SPEC CTRL MSR instead of the 670 * VIRT_SPEC MSR. 671 */ 672 if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD) && 673 !boot_cpu_has(X86_FEATURE_AMD_SSBD)) 674 entry->ebx |= F(VIRT_SSBD); 675 break; 676 } 677 case 0x80000019: 678 entry->ecx = entry->edx = 0; 679 break; 680 case 0x8000001a: 681 break; 682 case 0x8000001d: 683 break; 684 /*Add support for Centaur's CPUID instruction*/ 685 case 0xC0000000: 686 /*Just support up to 0xC0000004 now*/ 687 entry->eax = min(entry->eax, 0xC0000004); 688 break; 689 case 0xC0000001: 690 entry->edx &= kvm_cpuid_C000_0001_edx_x86_features; 691 cpuid_mask(&entry->edx, CPUID_C000_0001_EDX); 692 break; 693 case 3: /* Processor serial number */ 694 case 5: /* MONITOR/MWAIT */ 695 case 0xC0000002: 696 case 0xC0000003: 697 case 0xC0000004: 698 default: 699 entry->eax = entry->ebx = entry->ecx = entry->edx = 0; 700 break; 701 } 702 703 kvm_x86_ops->set_supported_cpuid(function, entry); 704 705 r = 0; 706 707 out: 708 put_cpu(); 709 710 return r; 711 } 712 713 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func, 714 u32 idx, int *nent, int maxnent, unsigned int type) 715 { 716 if (type == KVM_GET_EMULATED_CPUID) 717 return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent); 718 719 return __do_cpuid_ent(entry, func, idx, nent, maxnent); 720 } 721 722 #undef F 723 724 struct kvm_cpuid_param { 725 u32 func; 726 u32 idx; 727 bool has_leaf_count; 728 bool (*qualifier)(const struct kvm_cpuid_param *param); 729 }; 730 731 static bool is_centaur_cpu(const struct kvm_cpuid_param *param) 732 { 733 return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR; 734 } 735 736 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries, 737 __u32 num_entries, unsigned int ioctl_type) 738 { 739 int i; 740 __u32 pad[3]; 741 742 if (ioctl_type != KVM_GET_EMULATED_CPUID) 743 return false; 744 745 /* 746 * We want to make sure that ->padding is being passed clean from 747 * userspace in case we want to use it for something in the future. 748 * 749 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we 750 * have to give ourselves satisfied only with the emulated side. /me 751 * sheds a tear. 752 */ 753 for (i = 0; i < num_entries; i++) { 754 if (copy_from_user(pad, entries[i].padding, sizeof(pad))) 755 return true; 756 757 if (pad[0] || pad[1] || pad[2]) 758 return true; 759 } 760 return false; 761 } 762 763 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid, 764 struct kvm_cpuid_entry2 __user *entries, 765 unsigned int type) 766 { 767 struct kvm_cpuid_entry2 *cpuid_entries; 768 int limit, nent = 0, r = -E2BIG, i; 769 u32 func; 770 static const struct kvm_cpuid_param param[] = { 771 { .func = 0, .has_leaf_count = true }, 772 { .func = 0x80000000, .has_leaf_count = true }, 773 { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true }, 774 { .func = KVM_CPUID_SIGNATURE }, 775 { .func = KVM_CPUID_FEATURES }, 776 }; 777 778 if (cpuid->nent < 1) 779 goto out; 780 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES) 781 cpuid->nent = KVM_MAX_CPUID_ENTRIES; 782 783 if (sanity_check_entries(entries, cpuid->nent, type)) 784 return -EINVAL; 785 786 r = -ENOMEM; 787 cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent); 788 if (!cpuid_entries) 789 goto out; 790 791 r = 0; 792 for (i = 0; i < ARRAY_SIZE(param); i++) { 793 const struct kvm_cpuid_param *ent = ¶m[i]; 794 795 if (ent->qualifier && !ent->qualifier(ent)) 796 continue; 797 798 r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx, 799 &nent, cpuid->nent, type); 800 801 if (r) 802 goto out_free; 803 804 if (!ent->has_leaf_count) 805 continue; 806 807 limit = cpuid_entries[nent - 1].eax; 808 for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func) 809 r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx, 810 &nent, cpuid->nent, type); 811 812 if (r) 813 goto out_free; 814 } 815 816 r = -EFAULT; 817 if (copy_to_user(entries, cpuid_entries, 818 nent * sizeof(struct kvm_cpuid_entry2))) 819 goto out_free; 820 cpuid->nent = nent; 821 r = 0; 822 823 out_free: 824 vfree(cpuid_entries); 825 out: 826 return r; 827 } 828 829 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i) 830 { 831 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i]; 832 struct kvm_cpuid_entry2 *ej; 833 int j = i; 834 int nent = vcpu->arch.cpuid_nent; 835 836 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT; 837 /* when no next entry is found, the current entry[i] is reselected */ 838 do { 839 j = (j + 1) % nent; 840 ej = &vcpu->arch.cpuid_entries[j]; 841 } while (ej->function != e->function); 842 843 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT; 844 845 return j; 846 } 847 848 /* find an entry with matching function, matching index (if needed), and that 849 * should be read next (if it's stateful) */ 850 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e, 851 u32 function, u32 index) 852 { 853 if (e->function != function) 854 return 0; 855 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index) 856 return 0; 857 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) && 858 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT)) 859 return 0; 860 return 1; 861 } 862 863 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu, 864 u32 function, u32 index) 865 { 866 int i; 867 struct kvm_cpuid_entry2 *best = NULL; 868 869 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) { 870 struct kvm_cpuid_entry2 *e; 871 872 e = &vcpu->arch.cpuid_entries[i]; 873 if (is_matching_cpuid_entry(e, function, index)) { 874 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) 875 move_to_next_stateful_cpuid_entry(vcpu, i); 876 best = e; 877 break; 878 } 879 } 880 return best; 881 } 882 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry); 883 884 /* 885 * If no match is found, check whether we exceed the vCPU's limit 886 * and return the content of the highest valid _standard_ leaf instead. 887 * This is to satisfy the CPUID specification. 888 */ 889 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu, 890 u32 function, u32 index) 891 { 892 struct kvm_cpuid_entry2 *maxlevel; 893 894 maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0); 895 if (!maxlevel || maxlevel->eax >= function) 896 return NULL; 897 if (function & 0x80000000) { 898 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0); 899 if (!maxlevel) 900 return NULL; 901 } 902 return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index); 903 } 904 905 bool kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, 906 u32 *ecx, u32 *edx, bool check_limit) 907 { 908 u32 function = *eax, index = *ecx; 909 struct kvm_cpuid_entry2 *best; 910 bool entry_found = true; 911 912 best = kvm_find_cpuid_entry(vcpu, function, index); 913 914 if (!best) { 915 entry_found = false; 916 if (!check_limit) 917 goto out; 918 919 best = check_cpuid_limit(vcpu, function, index); 920 } 921 922 out: 923 if (best) { 924 *eax = best->eax; 925 *ebx = best->ebx; 926 *ecx = best->ecx; 927 *edx = best->edx; 928 } else 929 *eax = *ebx = *ecx = *edx = 0; 930 trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx, entry_found); 931 return entry_found; 932 } 933 EXPORT_SYMBOL_GPL(kvm_cpuid); 934 935 int kvm_emulate_cpuid(struct kvm_vcpu *vcpu) 936 { 937 u32 eax, ebx, ecx, edx; 938 939 if (cpuid_fault_enabled(vcpu) && !kvm_require_cpl(vcpu, 0)) 940 return 1; 941 942 eax = kvm_register_read(vcpu, VCPU_REGS_RAX); 943 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX); 944 kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx, true); 945 kvm_register_write(vcpu, VCPU_REGS_RAX, eax); 946 kvm_register_write(vcpu, VCPU_REGS_RBX, ebx); 947 kvm_register_write(vcpu, VCPU_REGS_RCX, ecx); 948 kvm_register_write(vcpu, VCPU_REGS_RDX, edx); 949 return kvm_skip_emulated_instruction(vcpu); 950 } 951 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid); 952