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 <asm/processor.h> 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 /* These are scattered features in cpufeatures.h. */ 69 #define KVM_CPUID_BIT_AVX512_4VNNIW 2 70 #define KVM_CPUID_BIT_AVX512_4FMAPS 3 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 kvm_x86_ops->fpu_activate(vcpu); 127 128 /* 129 * The existing code assumes virtual address is 48-bit in the canonical 130 * address checks; exit if it is ever changed. 131 */ 132 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0); 133 if (best && ((best->eax & 0xff00) >> 8) != 48 && 134 ((best->eax & 0xff00) >> 8) != 0) 135 return -EINVAL; 136 137 /* Update physical-address width */ 138 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu); 139 140 kvm_pmu_refresh(vcpu); 141 return 0; 142 } 143 144 static int is_efer_nx(void) 145 { 146 unsigned long long efer = 0; 147 148 rdmsrl_safe(MSR_EFER, &efer); 149 return efer & EFER_NX; 150 } 151 152 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu) 153 { 154 int i; 155 struct kvm_cpuid_entry2 *e, *entry; 156 157 entry = NULL; 158 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) { 159 e = &vcpu->arch.cpuid_entries[i]; 160 if (e->function == 0x80000001) { 161 entry = e; 162 break; 163 } 164 } 165 if (entry && (entry->edx & F(NX)) && !is_efer_nx()) { 166 entry->edx &= ~F(NX); 167 printk(KERN_INFO "kvm: guest NX capability removed\n"); 168 } 169 } 170 171 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu) 172 { 173 struct kvm_cpuid_entry2 *best; 174 175 best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0); 176 if (!best || best->eax < 0x80000008) 177 goto not_found; 178 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0); 179 if (best) 180 return best->eax & 0xff; 181 not_found: 182 return 36; 183 } 184 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr); 185 186 /* when an old userspace process fills a new kernel module */ 187 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu, 188 struct kvm_cpuid *cpuid, 189 struct kvm_cpuid_entry __user *entries) 190 { 191 int r, i; 192 struct kvm_cpuid_entry *cpuid_entries = NULL; 193 194 r = -E2BIG; 195 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES) 196 goto out; 197 r = -ENOMEM; 198 if (cpuid->nent) { 199 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * 200 cpuid->nent); 201 if (!cpuid_entries) 202 goto out; 203 r = -EFAULT; 204 if (copy_from_user(cpuid_entries, entries, 205 cpuid->nent * sizeof(struct kvm_cpuid_entry))) 206 goto out; 207 } 208 for (i = 0; i < cpuid->nent; i++) { 209 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function; 210 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax; 211 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx; 212 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx; 213 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx; 214 vcpu->arch.cpuid_entries[i].index = 0; 215 vcpu->arch.cpuid_entries[i].flags = 0; 216 vcpu->arch.cpuid_entries[i].padding[0] = 0; 217 vcpu->arch.cpuid_entries[i].padding[1] = 0; 218 vcpu->arch.cpuid_entries[i].padding[2] = 0; 219 } 220 vcpu->arch.cpuid_nent = cpuid->nent; 221 cpuid_fix_nx_cap(vcpu); 222 kvm_apic_set_version(vcpu); 223 kvm_x86_ops->cpuid_update(vcpu); 224 r = kvm_update_cpuid(vcpu); 225 226 out: 227 vfree(cpuid_entries); 228 return r; 229 } 230 231 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu, 232 struct kvm_cpuid2 *cpuid, 233 struct kvm_cpuid_entry2 __user *entries) 234 { 235 int r; 236 237 r = -E2BIG; 238 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES) 239 goto out; 240 r = -EFAULT; 241 if (copy_from_user(&vcpu->arch.cpuid_entries, entries, 242 cpuid->nent * sizeof(struct kvm_cpuid_entry2))) 243 goto out; 244 vcpu->arch.cpuid_nent = cpuid->nent; 245 kvm_apic_set_version(vcpu); 246 kvm_x86_ops->cpuid_update(vcpu); 247 r = kvm_update_cpuid(vcpu); 248 out: 249 return r; 250 } 251 252 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu, 253 struct kvm_cpuid2 *cpuid, 254 struct kvm_cpuid_entry2 __user *entries) 255 { 256 int r; 257 258 r = -E2BIG; 259 if (cpuid->nent < vcpu->arch.cpuid_nent) 260 goto out; 261 r = -EFAULT; 262 if (copy_to_user(entries, &vcpu->arch.cpuid_entries, 263 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2))) 264 goto out; 265 return 0; 266 267 out: 268 cpuid->nent = vcpu->arch.cpuid_nent; 269 return r; 270 } 271 272 static void cpuid_mask(u32 *word, int wordnum) 273 { 274 *word &= boot_cpu_data.x86_capability[wordnum]; 275 } 276 277 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function, 278 u32 index) 279 { 280 entry->function = function; 281 entry->index = index; 282 cpuid_count(entry->function, entry->index, 283 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx); 284 entry->flags = 0; 285 } 286 287 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry, 288 u32 func, u32 index, int *nent, int maxnent) 289 { 290 switch (func) { 291 case 0: 292 entry->eax = 1; /* only one leaf currently */ 293 ++*nent; 294 break; 295 case 1: 296 entry->ecx = F(MOVBE); 297 ++*nent; 298 break; 299 default: 300 break; 301 } 302 303 entry->function = func; 304 entry->index = index; 305 306 return 0; 307 } 308 309 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function, 310 u32 index, int *nent, int maxnent) 311 { 312 int r; 313 unsigned f_nx = is_efer_nx() ? F(NX) : 0; 314 #ifdef CONFIG_X86_64 315 unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL) 316 ? F(GBPAGES) : 0; 317 unsigned f_lm = F(LM); 318 #else 319 unsigned f_gbpages = 0; 320 unsigned f_lm = 0; 321 #endif 322 unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0; 323 unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0; 324 unsigned f_mpx = kvm_mpx_supported() ? F(MPX) : 0; 325 unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0; 326 327 /* cpuid 1.edx */ 328 const u32 kvm_cpuid_1_edx_x86_features = 329 F(FPU) | F(VME) | F(DE) | F(PSE) | 330 F(TSC) | F(MSR) | F(PAE) | F(MCE) | 331 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) | 332 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) | 333 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) | 334 0 /* Reserved, DS, ACPI */ | F(MMX) | 335 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) | 336 0 /* HTT, TM, Reserved, PBE */; 337 /* cpuid 0x80000001.edx */ 338 const u32 kvm_cpuid_8000_0001_edx_x86_features = 339 F(FPU) | F(VME) | F(DE) | F(PSE) | 340 F(TSC) | F(MSR) | F(PAE) | F(MCE) | 341 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) | 342 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) | 343 F(PAT) | F(PSE36) | 0 /* Reserved */ | 344 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) | 345 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp | 346 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW); 347 /* cpuid 1.ecx */ 348 const u32 kvm_cpuid_1_ecx_x86_features = 349 /* NOTE: MONITOR (and MWAIT) are emulated as NOP, 350 * but *not* advertised to guests via CPUID ! */ 351 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ | 352 0 /* DS-CPL, VMX, SMX, EST */ | 353 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ | 354 F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ | 355 F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) | 356 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) | 357 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) | 358 F(F16C) | F(RDRAND); 359 /* cpuid 0x80000001.ecx */ 360 const u32 kvm_cpuid_8000_0001_ecx_x86_features = 361 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ | 362 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) | 363 F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) | 364 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM); 365 366 /* cpuid 0xC0000001.edx */ 367 const u32 kvm_cpuid_C000_0001_edx_x86_features = 368 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) | 369 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) | 370 F(PMM) | F(PMM_EN); 371 372 /* cpuid 7.0.ebx */ 373 const u32 kvm_cpuid_7_0_ebx_x86_features = 374 F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) | 375 F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) | 376 F(ADX) | F(SMAP) | F(AVX512F) | F(AVX512PF) | F(AVX512ER) | 377 F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) | 378 F(AVX512BW) | F(AVX512VL); 379 380 /* cpuid 0xD.1.eax */ 381 const u32 kvm_cpuid_D_1_eax_x86_features = 382 F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves; 383 384 /* cpuid 7.0.ecx*/ 385 const u32 kvm_cpuid_7_0_ecx_x86_features = F(PKU) | 0 /*OSPKE*/; 386 387 /* cpuid 7.0.edx*/ 388 const u32 kvm_cpuid_7_0_edx_x86_features = 389 KF(AVX512_4VNNIW) | KF(AVX512_4FMAPS); 390 391 /* all calls to cpuid_count() should be made on the same cpu */ 392 get_cpu(); 393 394 r = -E2BIG; 395 396 if (*nent >= maxnent) 397 goto out; 398 399 do_cpuid_1_ent(entry, function, index); 400 ++*nent; 401 402 switch (function) { 403 case 0: 404 entry->eax = min(entry->eax, (u32)0xd); 405 break; 406 case 1: 407 entry->edx &= kvm_cpuid_1_edx_x86_features; 408 cpuid_mask(&entry->edx, CPUID_1_EDX); 409 entry->ecx &= kvm_cpuid_1_ecx_x86_features; 410 cpuid_mask(&entry->ecx, CPUID_1_ECX); 411 /* we support x2apic emulation even if host does not support 412 * it since we emulate x2apic in software */ 413 entry->ecx |= F(X2APIC); 414 break; 415 /* function 2 entries are STATEFUL. That is, repeated cpuid commands 416 * may return different values. This forces us to get_cpu() before 417 * issuing the first command, and also to emulate this annoying behavior 418 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */ 419 case 2: { 420 int t, times = entry->eax & 0xff; 421 422 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC; 423 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT; 424 for (t = 1; t < times; ++t) { 425 if (*nent >= maxnent) 426 goto out; 427 428 do_cpuid_1_ent(&entry[t], function, 0); 429 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC; 430 ++*nent; 431 } 432 break; 433 } 434 /* function 4 has additional index. */ 435 case 4: { 436 int i, cache_type; 437 438 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; 439 /* read more entries until cache_type is zero */ 440 for (i = 1; ; ++i) { 441 if (*nent >= maxnent) 442 goto out; 443 444 cache_type = entry[i - 1].eax & 0x1f; 445 if (!cache_type) 446 break; 447 do_cpuid_1_ent(&entry[i], function, i); 448 entry[i].flags |= 449 KVM_CPUID_FLAG_SIGNIFCANT_INDEX; 450 ++*nent; 451 } 452 break; 453 } 454 case 6: /* Thermal management */ 455 entry->eax = 0x4; /* allow ARAT */ 456 entry->ebx = 0; 457 entry->ecx = 0; 458 entry->edx = 0; 459 break; 460 case 7: { 461 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; 462 /* Mask ebx against host capability word 9 */ 463 if (index == 0) { 464 entry->ebx &= kvm_cpuid_7_0_ebx_x86_features; 465 cpuid_mask(&entry->ebx, CPUID_7_0_EBX); 466 // TSC_ADJUST is emulated 467 entry->ebx |= F(TSC_ADJUST); 468 entry->ecx &= kvm_cpuid_7_0_ecx_x86_features; 469 cpuid_mask(&entry->ecx, CPUID_7_ECX); 470 /* PKU is not yet implemented for shadow paging. */ 471 if (!tdp_enabled) 472 entry->ecx &= ~F(PKU); 473 entry->edx &= kvm_cpuid_7_0_edx_x86_features; 474 entry->edx &= get_scattered_cpuid_leaf(7, 0, CPUID_EDX); 475 } else { 476 entry->ebx = 0; 477 entry->ecx = 0; 478 entry->edx = 0; 479 } 480 entry->eax = 0; 481 break; 482 } 483 case 9: 484 break; 485 case 0xa: { /* Architectural Performance Monitoring */ 486 struct x86_pmu_capability cap; 487 union cpuid10_eax eax; 488 union cpuid10_edx edx; 489 490 perf_get_x86_pmu_capability(&cap); 491 492 /* 493 * Only support guest architectural pmu on a host 494 * with architectural pmu. 495 */ 496 if (!cap.version) 497 memset(&cap, 0, sizeof(cap)); 498 499 eax.split.version_id = min(cap.version, 2); 500 eax.split.num_counters = cap.num_counters_gp; 501 eax.split.bit_width = cap.bit_width_gp; 502 eax.split.mask_length = cap.events_mask_len; 503 504 edx.split.num_counters_fixed = cap.num_counters_fixed; 505 edx.split.bit_width_fixed = cap.bit_width_fixed; 506 edx.split.reserved = 0; 507 508 entry->eax = eax.full; 509 entry->ebx = cap.events_mask; 510 entry->ecx = 0; 511 entry->edx = edx.full; 512 break; 513 } 514 /* function 0xb has additional index. */ 515 case 0xb: { 516 int i, level_type; 517 518 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; 519 /* read more entries until level_type is zero */ 520 for (i = 1; ; ++i) { 521 if (*nent >= maxnent) 522 goto out; 523 524 level_type = entry[i - 1].ecx & 0xff00; 525 if (!level_type) 526 break; 527 do_cpuid_1_ent(&entry[i], function, i); 528 entry[i].flags |= 529 KVM_CPUID_FLAG_SIGNIFCANT_INDEX; 530 ++*nent; 531 } 532 break; 533 } 534 case 0xd: { 535 int idx, i; 536 u64 supported = kvm_supported_xcr0(); 537 538 entry->eax &= supported; 539 entry->ebx = xstate_required_size(supported, false); 540 entry->ecx = entry->ebx; 541 entry->edx &= supported >> 32; 542 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX; 543 if (!supported) 544 break; 545 546 for (idx = 1, i = 1; idx < 64; ++idx) { 547 u64 mask = ((u64)1 << idx); 548 if (*nent >= maxnent) 549 goto out; 550 551 do_cpuid_1_ent(&entry[i], function, idx); 552 if (idx == 1) { 553 entry[i].eax &= kvm_cpuid_D_1_eax_x86_features; 554 cpuid_mask(&entry[i].eax, CPUID_D_1_EAX); 555 entry[i].ebx = 0; 556 if (entry[i].eax & (F(XSAVES)|F(XSAVEC))) 557 entry[i].ebx = 558 xstate_required_size(supported, 559 true); 560 } else { 561 if (entry[i].eax == 0 || !(supported & mask)) 562 continue; 563 if (WARN_ON_ONCE(entry[i].ecx & 1)) 564 continue; 565 } 566 entry[i].ecx = 0; 567 entry[i].edx = 0; 568 entry[i].flags |= 569 KVM_CPUID_FLAG_SIGNIFCANT_INDEX; 570 ++*nent; 571 ++i; 572 } 573 break; 574 } 575 case KVM_CPUID_SIGNATURE: { 576 static const char signature[12] = "KVMKVMKVM\0\0"; 577 const u32 *sigptr = (const u32 *)signature; 578 entry->eax = KVM_CPUID_FEATURES; 579 entry->ebx = sigptr[0]; 580 entry->ecx = sigptr[1]; 581 entry->edx = sigptr[2]; 582 break; 583 } 584 case KVM_CPUID_FEATURES: 585 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) | 586 (1 << KVM_FEATURE_NOP_IO_DELAY) | 587 (1 << KVM_FEATURE_CLOCKSOURCE2) | 588 (1 << KVM_FEATURE_ASYNC_PF) | 589 (1 << KVM_FEATURE_PV_EOI) | 590 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) | 591 (1 << KVM_FEATURE_PV_UNHALT); 592 593 if (sched_info_on()) 594 entry->eax |= (1 << KVM_FEATURE_STEAL_TIME); 595 596 entry->ebx = 0; 597 entry->ecx = 0; 598 entry->edx = 0; 599 break; 600 case 0x80000000: 601 entry->eax = min(entry->eax, 0x8000001a); 602 break; 603 case 0x80000001: 604 entry->edx &= kvm_cpuid_8000_0001_edx_x86_features; 605 cpuid_mask(&entry->edx, CPUID_8000_0001_EDX); 606 entry->ecx &= kvm_cpuid_8000_0001_ecx_x86_features; 607 cpuid_mask(&entry->ecx, CPUID_8000_0001_ECX); 608 break; 609 case 0x80000007: /* Advanced power management */ 610 /* invariant TSC is CPUID.80000007H:EDX[8] */ 611 entry->edx &= (1 << 8); 612 /* mask against host */ 613 entry->edx &= boot_cpu_data.x86_power; 614 entry->eax = entry->ebx = entry->ecx = 0; 615 break; 616 case 0x80000008: { 617 unsigned g_phys_as = (entry->eax >> 16) & 0xff; 618 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U); 619 unsigned phys_as = entry->eax & 0xff; 620 621 if (!g_phys_as) 622 g_phys_as = phys_as; 623 entry->eax = g_phys_as | (virt_as << 8); 624 entry->ebx = entry->edx = 0; 625 break; 626 } 627 case 0x80000019: 628 entry->ecx = entry->edx = 0; 629 break; 630 case 0x8000001a: 631 break; 632 case 0x8000001d: 633 break; 634 /*Add support for Centaur's CPUID instruction*/ 635 case 0xC0000000: 636 /*Just support up to 0xC0000004 now*/ 637 entry->eax = min(entry->eax, 0xC0000004); 638 break; 639 case 0xC0000001: 640 entry->edx &= kvm_cpuid_C000_0001_edx_x86_features; 641 cpuid_mask(&entry->edx, CPUID_C000_0001_EDX); 642 break; 643 case 3: /* Processor serial number */ 644 case 5: /* MONITOR/MWAIT */ 645 case 0xC0000002: 646 case 0xC0000003: 647 case 0xC0000004: 648 default: 649 entry->eax = entry->ebx = entry->ecx = entry->edx = 0; 650 break; 651 } 652 653 kvm_x86_ops->set_supported_cpuid(function, entry); 654 655 r = 0; 656 657 out: 658 put_cpu(); 659 660 return r; 661 } 662 663 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func, 664 u32 idx, int *nent, int maxnent, unsigned int type) 665 { 666 if (type == KVM_GET_EMULATED_CPUID) 667 return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent); 668 669 return __do_cpuid_ent(entry, func, idx, nent, maxnent); 670 } 671 672 #undef F 673 674 struct kvm_cpuid_param { 675 u32 func; 676 u32 idx; 677 bool has_leaf_count; 678 bool (*qualifier)(const struct kvm_cpuid_param *param); 679 }; 680 681 static bool is_centaur_cpu(const struct kvm_cpuid_param *param) 682 { 683 return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR; 684 } 685 686 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries, 687 __u32 num_entries, unsigned int ioctl_type) 688 { 689 int i; 690 __u32 pad[3]; 691 692 if (ioctl_type != KVM_GET_EMULATED_CPUID) 693 return false; 694 695 /* 696 * We want to make sure that ->padding is being passed clean from 697 * userspace in case we want to use it for something in the future. 698 * 699 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we 700 * have to give ourselves satisfied only with the emulated side. /me 701 * sheds a tear. 702 */ 703 for (i = 0; i < num_entries; i++) { 704 if (copy_from_user(pad, entries[i].padding, sizeof(pad))) 705 return true; 706 707 if (pad[0] || pad[1] || pad[2]) 708 return true; 709 } 710 return false; 711 } 712 713 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid, 714 struct kvm_cpuid_entry2 __user *entries, 715 unsigned int type) 716 { 717 struct kvm_cpuid_entry2 *cpuid_entries; 718 int limit, nent = 0, r = -E2BIG, i; 719 u32 func; 720 static const struct kvm_cpuid_param param[] = { 721 { .func = 0, .has_leaf_count = true }, 722 { .func = 0x80000000, .has_leaf_count = true }, 723 { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true }, 724 { .func = KVM_CPUID_SIGNATURE }, 725 { .func = KVM_CPUID_FEATURES }, 726 }; 727 728 if (cpuid->nent < 1) 729 goto out; 730 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES) 731 cpuid->nent = KVM_MAX_CPUID_ENTRIES; 732 733 if (sanity_check_entries(entries, cpuid->nent, type)) 734 return -EINVAL; 735 736 r = -ENOMEM; 737 cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent); 738 if (!cpuid_entries) 739 goto out; 740 741 r = 0; 742 for (i = 0; i < ARRAY_SIZE(param); i++) { 743 const struct kvm_cpuid_param *ent = ¶m[i]; 744 745 if (ent->qualifier && !ent->qualifier(ent)) 746 continue; 747 748 r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx, 749 &nent, cpuid->nent, type); 750 751 if (r) 752 goto out_free; 753 754 if (!ent->has_leaf_count) 755 continue; 756 757 limit = cpuid_entries[nent - 1].eax; 758 for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func) 759 r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx, 760 &nent, cpuid->nent, type); 761 762 if (r) 763 goto out_free; 764 } 765 766 r = -EFAULT; 767 if (copy_to_user(entries, cpuid_entries, 768 nent * sizeof(struct kvm_cpuid_entry2))) 769 goto out_free; 770 cpuid->nent = nent; 771 r = 0; 772 773 out_free: 774 vfree(cpuid_entries); 775 out: 776 return r; 777 } 778 779 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i) 780 { 781 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i]; 782 int j, nent = vcpu->arch.cpuid_nent; 783 784 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT; 785 /* when no next entry is found, the current entry[i] is reselected */ 786 for (j = i + 1; ; j = (j + 1) % nent) { 787 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j]; 788 if (ej->function == e->function) { 789 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT; 790 return j; 791 } 792 } 793 return 0; /* silence gcc, even though control never reaches here */ 794 } 795 796 /* find an entry with matching function, matching index (if needed), and that 797 * should be read next (if it's stateful) */ 798 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e, 799 u32 function, u32 index) 800 { 801 if (e->function != function) 802 return 0; 803 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index) 804 return 0; 805 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) && 806 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT)) 807 return 0; 808 return 1; 809 } 810 811 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu, 812 u32 function, u32 index) 813 { 814 int i; 815 struct kvm_cpuid_entry2 *best = NULL; 816 817 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) { 818 struct kvm_cpuid_entry2 *e; 819 820 e = &vcpu->arch.cpuid_entries[i]; 821 if (is_matching_cpuid_entry(e, function, index)) { 822 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) 823 move_to_next_stateful_cpuid_entry(vcpu, i); 824 best = e; 825 break; 826 } 827 } 828 return best; 829 } 830 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry); 831 832 /* 833 * If no match is found, check whether we exceed the vCPU's limit 834 * and return the content of the highest valid _standard_ leaf instead. 835 * This is to satisfy the CPUID specification. 836 */ 837 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu, 838 u32 function, u32 index) 839 { 840 struct kvm_cpuid_entry2 *maxlevel; 841 842 maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0); 843 if (!maxlevel || maxlevel->eax >= function) 844 return NULL; 845 if (function & 0x80000000) { 846 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0); 847 if (!maxlevel) 848 return NULL; 849 } 850 return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index); 851 } 852 853 void kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, u32 *ecx, u32 *edx) 854 { 855 u32 function = *eax, index = *ecx; 856 struct kvm_cpuid_entry2 *best; 857 858 best = kvm_find_cpuid_entry(vcpu, function, index); 859 860 if (!best) 861 best = check_cpuid_limit(vcpu, function, index); 862 863 /* 864 * Perfmon not yet supported for L2 guest. 865 */ 866 if (is_guest_mode(vcpu) && function == 0xa) 867 best = NULL; 868 869 if (best) { 870 *eax = best->eax; 871 *ebx = best->ebx; 872 *ecx = best->ecx; 873 *edx = best->edx; 874 } else 875 *eax = *ebx = *ecx = *edx = 0; 876 trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx); 877 } 878 EXPORT_SYMBOL_GPL(kvm_cpuid); 879 880 int kvm_emulate_cpuid(struct kvm_vcpu *vcpu) 881 { 882 u32 eax, ebx, ecx, edx; 883 884 eax = kvm_register_read(vcpu, VCPU_REGS_RAX); 885 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX); 886 kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx); 887 kvm_register_write(vcpu, VCPU_REGS_RAX, eax); 888 kvm_register_write(vcpu, VCPU_REGS_RBX, ebx); 889 kvm_register_write(vcpu, VCPU_REGS_RCX, ecx); 890 kvm_register_write(vcpu, VCPU_REGS_RDX, edx); 891 return kvm_skip_emulated_instruction(vcpu); 892 } 893 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid); 894