1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
4 * cpuid support routines
6 * derived from arch/x86/kvm/x86.c
8 * Copyright 2011 Red Hat, Inc. and/or its affiliates.
9 * Copyright IBM Corporation, 2008
12 #include <linux/kvm_host.h>
13 #include <linux/export.h>
14 #include <linux/vmalloc.h>
15 #include <linux/uaccess.h>
16 #include <linux/sched/stat.h>
18 #include <asm/processor.h>
20 #include <asm/fpu/xstate.h>
28 * Unlike "struct cpuinfo_x86.x86_capability", kvm_cpu_caps doesn't need to be
29 * aligned to sizeof(unsigned long) because it's not accessed via bitops.
31 u32 kvm_cpu_caps[NCAPINTS] __read_mostly;
32 EXPORT_SYMBOL_GPL(kvm_cpu_caps);
34 static u32 xstate_required_size(u64 xstate_bv, bool compacted)
37 u32 ret = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
39 xstate_bv &= XFEATURE_MASK_EXTEND;
41 if (xstate_bv & 0x1) {
42 u32 eax, ebx, ecx, edx, offset;
43 cpuid_count(0xD, feature_bit, &eax, &ebx, &ecx, &edx);
44 offset = compacted ? ret : ebx;
45 ret = max(ret, offset + eax);
57 int kvm_update_cpuid(struct kvm_vcpu *vcpu)
59 struct kvm_cpuid_entry2 *best;
60 struct kvm_lapic *apic = vcpu->arch.apic;
62 best = kvm_find_cpuid_entry(vcpu, 1, 0);
66 /* Update OSXSAVE bit */
67 if (boot_cpu_has(X86_FEATURE_XSAVE) && best->function == 0x1)
68 cpuid_entry_change(best, X86_FEATURE_OSXSAVE,
69 kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE));
71 cpuid_entry_change(best, X86_FEATURE_APIC,
72 vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE);
75 if (cpuid_entry_has(best, X86_FEATURE_TSC_DEADLINE_TIMER))
76 apic->lapic_timer.timer_mode_mask = 3 << 17;
78 apic->lapic_timer.timer_mode_mask = 1 << 17;
81 best = kvm_find_cpuid_entry(vcpu, 7, 0);
82 if (best && boot_cpu_has(X86_FEATURE_PKU) && best->function == 0x7)
83 cpuid_entry_change(best, X86_FEATURE_OSPKE,
84 kvm_read_cr4_bits(vcpu, X86_CR4_PKE));
86 best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
88 vcpu->arch.guest_supported_xcr0 = 0;
90 vcpu->arch.guest_supported_xcr0 =
91 (best->eax | ((u64)best->edx << 32)) & supported_xcr0;
92 best->ebx = xstate_required_size(vcpu->arch.xcr0, false);
95 best = kvm_find_cpuid_entry(vcpu, 0xD, 1);
96 if (best && (cpuid_entry_has(best, X86_FEATURE_XSAVES) ||
97 cpuid_entry_has(best, X86_FEATURE_XSAVEC)))
98 best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
101 * The existing code assumes virtual address is 48-bit or 57-bit in the
102 * canonical address checks; exit if it is ever changed.
104 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
106 int vaddr_bits = (best->eax & 0xff00) >> 8;
108 if (vaddr_bits != 48 && vaddr_bits != 57 && vaddr_bits != 0)
112 best = kvm_find_cpuid_entry(vcpu, KVM_CPUID_FEATURES, 0);
113 if (kvm_hlt_in_guest(vcpu->kvm) && best &&
114 (best->eax & (1 << KVM_FEATURE_PV_UNHALT)))
115 best->eax &= ~(1 << KVM_FEATURE_PV_UNHALT);
117 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT)) {
118 best = kvm_find_cpuid_entry(vcpu, 0x1, 0);
120 cpuid_entry_change(best, X86_FEATURE_MWAIT,
121 vcpu->arch.ia32_misc_enable_msr &
122 MSR_IA32_MISC_ENABLE_MWAIT);
125 /* Note, maxphyaddr must be updated before tdp_level. */
126 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
127 vcpu->arch.tdp_level = kvm_x86_ops.get_tdp_level(vcpu);
128 kvm_mmu_reset_context(vcpu);
130 kvm_pmu_refresh(vcpu);
134 static int is_efer_nx(void)
136 return host_efer & EFER_NX;
139 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
142 struct kvm_cpuid_entry2 *e, *entry;
145 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
146 e = &vcpu->arch.cpuid_entries[i];
147 if (e->function == 0x80000001) {
152 if (entry && cpuid_entry_has(entry, X86_FEATURE_NX) && !is_efer_nx()) {
153 cpuid_entry_clear(entry, X86_FEATURE_NX);
154 printk(KERN_INFO "kvm: guest NX capability removed\n");
158 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
160 struct kvm_cpuid_entry2 *best;
162 best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
163 if (!best || best->eax < 0x80000008)
165 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
167 return best->eax & 0xff;
171 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr);
173 /* when an old userspace process fills a new kernel module */
174 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
175 struct kvm_cpuid *cpuid,
176 struct kvm_cpuid_entry __user *entries)
179 struct kvm_cpuid_entry *cpuid_entries = NULL;
182 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
187 vmalloc(array_size(sizeof(struct kvm_cpuid_entry),
192 if (copy_from_user(cpuid_entries, entries,
193 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
196 for (i = 0; i < cpuid->nent; i++) {
197 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
198 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
199 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
200 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
201 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
202 vcpu->arch.cpuid_entries[i].index = 0;
203 vcpu->arch.cpuid_entries[i].flags = 0;
204 vcpu->arch.cpuid_entries[i].padding[0] = 0;
205 vcpu->arch.cpuid_entries[i].padding[1] = 0;
206 vcpu->arch.cpuid_entries[i].padding[2] = 0;
208 vcpu->arch.cpuid_nent = cpuid->nent;
209 cpuid_fix_nx_cap(vcpu);
210 kvm_apic_set_version(vcpu);
211 kvm_x86_ops.cpuid_update(vcpu);
212 r = kvm_update_cpuid(vcpu);
215 vfree(cpuid_entries);
219 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
220 struct kvm_cpuid2 *cpuid,
221 struct kvm_cpuid_entry2 __user *entries)
226 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
229 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
230 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
232 vcpu->arch.cpuid_nent = cpuid->nent;
233 kvm_apic_set_version(vcpu);
234 kvm_x86_ops.cpuid_update(vcpu);
235 r = kvm_update_cpuid(vcpu);
240 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
241 struct kvm_cpuid2 *cpuid,
242 struct kvm_cpuid_entry2 __user *entries)
247 if (cpuid->nent < vcpu->arch.cpuid_nent)
250 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
251 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
256 cpuid->nent = vcpu->arch.cpuid_nent;
260 static __always_inline void kvm_cpu_cap_mask(enum cpuid_leafs leaf, u32 mask)
262 const struct cpuid_reg cpuid = x86_feature_cpuid(leaf * 32);
263 struct kvm_cpuid_entry2 entry;
265 reverse_cpuid_check(leaf);
266 kvm_cpu_caps[leaf] &= mask;
268 cpuid_count(cpuid.function, cpuid.index,
269 &entry.eax, &entry.ebx, &entry.ecx, &entry.edx);
271 kvm_cpu_caps[leaf] &= *__cpuid_entry_get_reg(&entry, cpuid.reg);
274 void kvm_set_cpu_caps(void)
276 unsigned int f_nx = is_efer_nx() ? F(NX) : 0;
278 unsigned int f_gbpages = F(GBPAGES);
279 unsigned int f_lm = F(LM);
281 unsigned int f_gbpages = 0;
282 unsigned int f_lm = 0;
285 BUILD_BUG_ON(sizeof(kvm_cpu_caps) >
286 sizeof(boot_cpu_data.x86_capability));
288 memcpy(&kvm_cpu_caps, &boot_cpu_data.x86_capability,
289 sizeof(kvm_cpu_caps));
291 kvm_cpu_cap_mask(CPUID_1_ECX,
293 * NOTE: MONITOR (and MWAIT) are emulated as NOP, but *not*
294 * advertised to guests via CPUID!
296 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
297 0 /* DS-CPL, VMX, SMX, EST */ |
298 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
299 F(FMA) | F(CX16) | 0 /* xTPR Update */ | F(PDCM) |
300 F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
301 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
302 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
305 /* KVM emulates x2apic in software irrespective of host support. */
306 kvm_cpu_cap_set(X86_FEATURE_X2APIC);
308 kvm_cpu_cap_mask(CPUID_1_EDX,
309 F(FPU) | F(VME) | F(DE) | F(PSE) |
310 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
311 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
312 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
313 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
314 0 /* Reserved, DS, ACPI */ | F(MMX) |
315 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
316 0 /* HTT, TM, Reserved, PBE */
319 kvm_cpu_cap_mask(CPUID_7_0_EBX,
320 F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
321 F(BMI2) | F(ERMS) | 0 /*INVPCID*/ | F(RTM) | 0 /*MPX*/ | F(RDSEED) |
322 F(ADX) | F(SMAP) | F(AVX512IFMA) | F(AVX512F) | F(AVX512PF) |
323 F(AVX512ER) | F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) |
324 F(SHA_NI) | F(AVX512BW) | F(AVX512VL) | 0 /*INTEL_PT*/
327 kvm_cpu_cap_mask(CPUID_7_ECX,
328 F(AVX512VBMI) | F(LA57) | F(PKU) | 0 /*OSPKE*/ | F(RDPID) |
329 F(AVX512_VPOPCNTDQ) | F(UMIP) | F(AVX512_VBMI2) | F(GFNI) |
330 F(VAES) | F(VPCLMULQDQ) | F(AVX512_VNNI) | F(AVX512_BITALG) |
331 F(CLDEMOTE) | F(MOVDIRI) | F(MOVDIR64B) | 0 /*WAITPKG*/
333 /* Set LA57 based on hardware capability. */
334 if (cpuid_ecx(7) & F(LA57))
335 kvm_cpu_cap_set(X86_FEATURE_LA57);
338 * PKU not yet implemented for shadow paging and requires OSPKE
339 * to be set on the host. Clear it if that is not the case
341 if (!tdp_enabled || !boot_cpu_has(X86_FEATURE_OSPKE))
342 kvm_cpu_cap_clear(X86_FEATURE_PKU);
344 kvm_cpu_cap_mask(CPUID_7_EDX,
345 F(AVX512_4VNNIW) | F(AVX512_4FMAPS) | F(SPEC_CTRL) |
346 F(SPEC_CTRL_SSBD) | F(ARCH_CAPABILITIES) | F(INTEL_STIBP) |
347 F(MD_CLEAR) | F(AVX512_VP2INTERSECT) | F(FSRM)
350 /* TSC_ADJUST and ARCH_CAPABILITIES are emulated in software. */
351 kvm_cpu_cap_set(X86_FEATURE_TSC_ADJUST);
352 kvm_cpu_cap_set(X86_FEATURE_ARCH_CAPABILITIES);
354 if (boot_cpu_has(X86_FEATURE_IBPB) && boot_cpu_has(X86_FEATURE_IBRS))
355 kvm_cpu_cap_set(X86_FEATURE_SPEC_CTRL);
356 if (boot_cpu_has(X86_FEATURE_STIBP))
357 kvm_cpu_cap_set(X86_FEATURE_INTEL_STIBP);
358 if (boot_cpu_has(X86_FEATURE_AMD_SSBD))
359 kvm_cpu_cap_set(X86_FEATURE_SPEC_CTRL_SSBD);
361 kvm_cpu_cap_mask(CPUID_7_1_EAX,
365 kvm_cpu_cap_mask(CPUID_D_1_EAX,
366 F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | F(XSAVES)
369 kvm_cpu_cap_mask(CPUID_8000_0001_ECX,
370 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
371 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
372 F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
373 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM) |
374 F(TOPOEXT) | F(PERFCTR_CORE)
377 kvm_cpu_cap_mask(CPUID_8000_0001_EDX,
378 F(FPU) | F(VME) | F(DE) | F(PSE) |
379 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
380 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
381 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
382 F(PAT) | F(PSE36) | 0 /* Reserved */ |
383 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
384 F(FXSR) | F(FXSR_OPT) | f_gbpages | F(RDTSCP) |
385 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW)
388 if (!tdp_enabled && IS_ENABLED(CONFIG_X86_64))
389 kvm_cpu_cap_set(X86_FEATURE_GBPAGES);
391 kvm_cpu_cap_mask(CPUID_8000_0008_EBX,
392 F(CLZERO) | F(XSAVEERPTR) |
393 F(WBNOINVD) | F(AMD_IBPB) | F(AMD_IBRS) | F(AMD_SSBD) | F(VIRT_SSBD) |
394 F(AMD_SSB_NO) | F(AMD_STIBP) | F(AMD_STIBP_ALWAYS_ON)
398 * AMD has separate bits for each SPEC_CTRL bit.
399 * arch/x86/kernel/cpu/bugs.c is kind enough to
400 * record that in cpufeatures so use them.
402 if (boot_cpu_has(X86_FEATURE_IBPB))
403 kvm_cpu_cap_set(X86_FEATURE_AMD_IBPB);
404 if (boot_cpu_has(X86_FEATURE_IBRS))
405 kvm_cpu_cap_set(X86_FEATURE_AMD_IBRS);
406 if (boot_cpu_has(X86_FEATURE_STIBP))
407 kvm_cpu_cap_set(X86_FEATURE_AMD_STIBP);
408 if (boot_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD))
409 kvm_cpu_cap_set(X86_FEATURE_AMD_SSBD);
410 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
411 kvm_cpu_cap_set(X86_FEATURE_AMD_SSB_NO);
413 * The preference is to use SPEC CTRL MSR instead of the
416 if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD) &&
417 !boot_cpu_has(X86_FEATURE_AMD_SSBD))
418 kvm_cpu_cap_set(X86_FEATURE_VIRT_SSBD);
421 * Hide all SVM features by default, SVM will set the cap bits for
422 * features it emulates and/or exposes for L1.
424 kvm_cpu_cap_mask(CPUID_8000_000A_EDX, 0);
426 kvm_cpu_cap_mask(CPUID_C000_0001_EDX,
427 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
428 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
432 EXPORT_SYMBOL_GPL(kvm_set_cpu_caps);
434 struct kvm_cpuid_array {
435 struct kvm_cpuid_entry2 *entries;
440 static struct kvm_cpuid_entry2 *do_host_cpuid(struct kvm_cpuid_array *array,
441 u32 function, u32 index)
443 struct kvm_cpuid_entry2 *entry;
445 if (array->nent >= array->maxnent)
448 entry = &array->entries[array->nent++];
450 entry->function = function;
451 entry->index = index;
454 cpuid_count(entry->function, entry->index,
455 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
470 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
477 static int __do_cpuid_func_emulated(struct kvm_cpuid_array *array, u32 func)
479 struct kvm_cpuid_entry2 *entry;
481 if (array->nent >= array->maxnent)
484 entry = &array->entries[array->nent];
485 entry->function = func;
495 entry->ecx = F(MOVBE);
499 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
501 entry->ecx = F(RDPID);
510 static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function)
512 struct kvm_cpuid_entry2 *entry;
515 /* all calls to cpuid_count() should be made on the same cpu */
520 entry = do_host_cpuid(array, function, 0);
526 /* Limited to the highest leaf implemented in KVM. */
527 entry->eax = min(entry->eax, 0x1fU);
530 cpuid_entry_override(entry, CPUID_1_EDX);
531 cpuid_entry_override(entry, CPUID_1_ECX);
535 * On ancient CPUs, function 2 entries are STATEFUL. That is,
536 * CPUID(function=2, index=0) may return different results each
537 * time, with the least-significant byte in EAX enumerating the
538 * number of times software should do CPUID(2, 0).
540 * Modern CPUs, i.e. every CPU KVM has *ever* run on are less
541 * idiotic. Intel's SDM states that EAX & 0xff "will always
542 * return 01H. Software should ignore this value and not
543 * interpret it as an informational descriptor", while AMD's
544 * APM states that CPUID(2) is reserved.
546 * WARN if a frankenstein CPU that supports virtualization and
547 * a stateful CPUID.0x2 is encountered.
549 WARN_ON_ONCE((entry->eax & 0xff) > 1);
551 /* functions 4 and 0x8000001d have additional index. */
555 * Read entries until the cache type in the previous entry is
556 * zero, i.e. indicates an invalid entry.
558 for (i = 1; entry->eax & 0x1f; ++i) {
559 entry = do_host_cpuid(array, function, i);
564 case 6: /* Thermal management */
565 entry->eax = 0x4; /* allow ARAT */
570 /* function 7 has additional index. */
572 entry->eax = min(entry->eax, 1u);
573 cpuid_entry_override(entry, CPUID_7_0_EBX);
574 cpuid_entry_override(entry, CPUID_7_ECX);
575 cpuid_entry_override(entry, CPUID_7_EDX);
577 /* KVM only supports 0x7.0 and 0x7.1, capped above via min(). */
578 if (entry->eax == 1) {
579 entry = do_host_cpuid(array, function, 1);
583 cpuid_entry_override(entry, CPUID_7_1_EAX);
591 case 0xa: { /* Architectural Performance Monitoring */
592 struct x86_pmu_capability cap;
593 union cpuid10_eax eax;
594 union cpuid10_edx edx;
596 perf_get_x86_pmu_capability(&cap);
599 * Only support guest architectural pmu on a host
600 * with architectural pmu.
603 memset(&cap, 0, sizeof(cap));
605 eax.split.version_id = min(cap.version, 2);
606 eax.split.num_counters = cap.num_counters_gp;
607 eax.split.bit_width = cap.bit_width_gp;
608 eax.split.mask_length = cap.events_mask_len;
610 edx.split.num_counters_fixed = cap.num_counters_fixed;
611 edx.split.bit_width_fixed = cap.bit_width_fixed;
612 edx.split.reserved = 0;
614 entry->eax = eax.full;
615 entry->ebx = cap.events_mask;
617 entry->edx = edx.full;
621 * Per Intel's SDM, the 0x1f is a superset of 0xb,
622 * thus they can be handled by common code.
627 * Populate entries until the level type (ECX[15:8]) of the
628 * previous entry is zero. Note, CPUID EAX.{0x1f,0xb}.0 is
629 * the starting entry, filled by the primary do_host_cpuid().
631 for (i = 1; entry->ecx & 0xff00; ++i) {
632 entry = do_host_cpuid(array, function, i);
638 entry->eax &= supported_xcr0;
639 entry->ebx = xstate_required_size(supported_xcr0, false);
640 entry->ecx = entry->ebx;
641 entry->edx &= supported_xcr0 >> 32;
645 entry = do_host_cpuid(array, function, 1);
649 cpuid_entry_override(entry, CPUID_D_1_EAX);
650 if (entry->eax & (F(XSAVES)|F(XSAVEC)))
651 entry->ebx = xstate_required_size(supported_xcr0 | supported_xss,
654 WARN_ON_ONCE(supported_xss != 0);
657 entry->ecx &= supported_xss;
658 entry->edx &= supported_xss >> 32;
660 for (i = 2; i < 64; ++i) {
662 if (supported_xcr0 & BIT_ULL(i))
664 else if (supported_xss & BIT_ULL(i))
669 entry = do_host_cpuid(array, function, i);
674 * The supported check above should have filtered out
675 * invalid sub-leafs. Only valid sub-leafs should
676 * reach this point, and they should have a non-zero
677 * save state size. Furthermore, check whether the
678 * processor agrees with supported_xcr0/supported_xss
679 * on whether this is an XCR0- or IA32_XSS-managed area.
681 if (WARN_ON_ONCE(!entry->eax || (entry->ecx & 0x1) != s_state)) {
690 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT)) {
691 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
695 for (i = 1, max_idx = entry->eax; i <= max_idx; ++i) {
696 if (!do_host_cpuid(array, function, i))
700 case KVM_CPUID_SIGNATURE: {
701 static const char signature[12] = "KVMKVMKVM\0\0";
702 const u32 *sigptr = (const u32 *)signature;
703 entry->eax = KVM_CPUID_FEATURES;
704 entry->ebx = sigptr[0];
705 entry->ecx = sigptr[1];
706 entry->edx = sigptr[2];
709 case KVM_CPUID_FEATURES:
710 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
711 (1 << KVM_FEATURE_NOP_IO_DELAY) |
712 (1 << KVM_FEATURE_CLOCKSOURCE2) |
713 (1 << KVM_FEATURE_ASYNC_PF) |
714 (1 << KVM_FEATURE_PV_EOI) |
715 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
716 (1 << KVM_FEATURE_PV_UNHALT) |
717 (1 << KVM_FEATURE_PV_TLB_FLUSH) |
718 (1 << KVM_FEATURE_ASYNC_PF_VMEXIT) |
719 (1 << KVM_FEATURE_PV_SEND_IPI) |
720 (1 << KVM_FEATURE_POLL_CONTROL) |
721 (1 << KVM_FEATURE_PV_SCHED_YIELD) |
722 (1 << KVM_FEATURE_ASYNC_PF_INT);
725 entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
732 entry->eax = min(entry->eax, 0x8000001f);
735 cpuid_entry_override(entry, CPUID_8000_0001_EDX);
736 cpuid_entry_override(entry, CPUID_8000_0001_ECX);
739 /* L2 cache and TLB: pass through host info. */
741 case 0x80000007: /* Advanced power management */
742 /* invariant TSC is CPUID.80000007H:EDX[8] */
743 entry->edx &= (1 << 8);
744 /* mask against host */
745 entry->edx &= boot_cpu_data.x86_power;
746 entry->eax = entry->ebx = entry->ecx = 0;
749 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
750 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
751 unsigned phys_as = entry->eax & 0xff;
755 entry->eax = g_phys_as | (virt_as << 8);
757 cpuid_entry_override(entry, CPUID_8000_0008_EBX);
761 if (!kvm_cpu_cap_has(X86_FEATURE_SVM)) {
762 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
765 entry->eax = 1; /* SVM revision 1 */
766 entry->ebx = 8; /* Lets support 8 ASIDs in case we add proper
767 ASID emulation to nested SVM */
768 entry->ecx = 0; /* Reserved */
769 cpuid_entry_override(entry, CPUID_8000_000A_EDX);
772 entry->ecx = entry->edx = 0;
777 /* Support memory encryption cpuid if host supports it */
779 if (!boot_cpu_has(X86_FEATURE_SEV))
780 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
782 /*Add support for Centaur's CPUID instruction*/
784 /*Just support up to 0xC0000004 now*/
785 entry->eax = min(entry->eax, 0xC0000004);
788 cpuid_entry_override(entry, CPUID_C000_0001_EDX);
790 case 3: /* Processor serial number */
791 case 5: /* MONITOR/MWAIT */
796 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
808 static int do_cpuid_func(struct kvm_cpuid_array *array, u32 func,
811 if (type == KVM_GET_EMULATED_CPUID)
812 return __do_cpuid_func_emulated(array, func);
814 return __do_cpuid_func(array, func);
817 #define CENTAUR_CPUID_SIGNATURE 0xC0000000
819 static int get_cpuid_func(struct kvm_cpuid_array *array, u32 func,
825 if (func == CENTAUR_CPUID_SIGNATURE &&
826 boot_cpu_data.x86_vendor != X86_VENDOR_CENTAUR)
829 r = do_cpuid_func(array, func, type);
833 limit = array->entries[array->nent - 1].eax;
834 for (func = func + 1; func <= limit; ++func) {
835 r = do_cpuid_func(array, func, type);
843 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
844 __u32 num_entries, unsigned int ioctl_type)
849 if (ioctl_type != KVM_GET_EMULATED_CPUID)
853 * We want to make sure that ->padding is being passed clean from
854 * userspace in case we want to use it for something in the future.
856 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
857 * have to give ourselves satisfied only with the emulated side. /me
860 for (i = 0; i < num_entries; i++) {
861 if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
864 if (pad[0] || pad[1] || pad[2])
870 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
871 struct kvm_cpuid_entry2 __user *entries,
874 static const u32 funcs[] = {
875 0, 0x80000000, CENTAUR_CPUID_SIGNATURE, KVM_CPUID_SIGNATURE,
878 struct kvm_cpuid_array array = {
885 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
886 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
888 if (sanity_check_entries(entries, cpuid->nent, type))
891 array.entries = vzalloc(array_size(sizeof(struct kvm_cpuid_entry2),
896 array.maxnent = cpuid->nent;
898 for (i = 0; i < ARRAY_SIZE(funcs); i++) {
899 r = get_cpuid_func(&array, funcs[i], type);
903 cpuid->nent = array.nent;
905 if (copy_to_user(entries, array.entries,
906 array.nent * sizeof(struct kvm_cpuid_entry2)))
910 vfree(array.entries);
914 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
915 u32 function, u32 index)
917 struct kvm_cpuid_entry2 *e;
920 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
921 e = &vcpu->arch.cpuid_entries[i];
923 if (e->function == function && (e->index == index ||
924 !(e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX)))
929 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
932 * Intel CPUID semantics treats any query for an out-of-range leaf as if the
933 * highest basic leaf (i.e. CPUID.0H:EAX) were requested. AMD CPUID semantics
934 * returns all zeroes for any undefined leaf, whether or not the leaf is in
935 * range. Centaur/VIA follows Intel semantics.
937 * A leaf is considered out-of-range if its function is higher than the maximum
938 * supported leaf of its associated class or if its associated class does not
941 * There are three primary classes to be considered, with their respective
942 * ranges described as "<base> - <top>[,<base2> - <top2>] inclusive. A primary
943 * class exists if a guest CPUID entry for its <base> leaf exists. For a given
944 * class, CPUID.<base>.EAX contains the max supported leaf for the class.
946 * - Basic: 0x00000000 - 0x3fffffff, 0x50000000 - 0x7fffffff
947 * - Hypervisor: 0x40000000 - 0x4fffffff
948 * - Extended: 0x80000000 - 0xbfffffff
949 * - Centaur: 0xc0000000 - 0xcfffffff
951 * The Hypervisor class is further subdivided into sub-classes that each act as
952 * their own indepdent class associated with a 0x100 byte range. E.g. if Qemu
953 * is advertising support for both HyperV and KVM, the resulting Hypervisor
954 * CPUID sub-classes are:
956 * - HyperV: 0x40000000 - 0x400000ff
957 * - KVM: 0x40000100 - 0x400001ff
959 static struct kvm_cpuid_entry2 *
960 get_out_of_range_cpuid_entry(struct kvm_vcpu *vcpu, u32 *fn_ptr, u32 index)
962 struct kvm_cpuid_entry2 *basic, *class;
963 u32 function = *fn_ptr;
965 basic = kvm_find_cpuid_entry(vcpu, 0, 0);
969 if (is_guest_vendor_amd(basic->ebx, basic->ecx, basic->edx) ||
970 is_guest_vendor_hygon(basic->ebx, basic->ecx, basic->edx))
973 if (function >= 0x40000000 && function <= 0x4fffffff)
974 class = kvm_find_cpuid_entry(vcpu, function & 0xffffff00, 0);
975 else if (function >= 0xc0000000)
976 class = kvm_find_cpuid_entry(vcpu, 0xc0000000, 0);
978 class = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
980 if (class && function <= class->eax)
984 * Leaf specific adjustments are also applied when redirecting to the
985 * max basic entry, e.g. if the max basic leaf is 0xb but there is no
986 * entry for CPUID.0xb.index (see below), then the output value for EDX
987 * needs to be pulled from CPUID.0xb.1.
989 *fn_ptr = basic->eax;
992 * The class does not exist or the requested function is out of range;
993 * the effective CPUID entry is the max basic leaf. Note, the index of
994 * the original requested leaf is observed!
996 return kvm_find_cpuid_entry(vcpu, basic->eax, index);
999 bool kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx,
1000 u32 *ecx, u32 *edx, bool exact_only)
1002 u32 orig_function = *eax, function = *eax, index = *ecx;
1003 struct kvm_cpuid_entry2 *entry;
1004 bool exact, used_max_basic = false;
1006 entry = kvm_find_cpuid_entry(vcpu, function, index);
1009 if (!entry && !exact_only) {
1010 entry = get_out_of_range_cpuid_entry(vcpu, &function, index);
1011 used_max_basic = !!entry;
1019 if (function == 7 && index == 0) {
1021 if (!__kvm_get_msr(vcpu, MSR_IA32_TSX_CTRL, &data, true) &&
1022 (data & TSX_CTRL_CPUID_CLEAR))
1023 *ebx &= ~(F(RTM) | F(HLE));
1026 *eax = *ebx = *ecx = *edx = 0;
1028 * When leaf 0BH or 1FH is defined, CL is pass-through
1029 * and EDX is always the x2APIC ID, even for undefined
1030 * subleaves. Index 1 will exist iff the leaf is
1031 * implemented, so we pass through CL iff leaf 1
1032 * exists. EDX can be copied from any existing index.
1034 if (function == 0xb || function == 0x1f) {
1035 entry = kvm_find_cpuid_entry(vcpu, function, 1);
1037 *ecx = index & 0xff;
1042 trace_kvm_cpuid(orig_function, index, *eax, *ebx, *ecx, *edx, exact,
1046 EXPORT_SYMBOL_GPL(kvm_cpuid);
1048 int kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1050 u32 eax, ebx, ecx, edx;
1052 if (cpuid_fault_enabled(vcpu) && !kvm_require_cpl(vcpu, 0))
1055 eax = kvm_rax_read(vcpu);
1056 ecx = kvm_rcx_read(vcpu);
1057 kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx, false);
1058 kvm_rax_write(vcpu, eax);
1059 kvm_rbx_write(vcpu, ebx);
1060 kvm_rcx_write(vcpu, ecx);
1061 kvm_rdx_write(vcpu, edx);
1062 return kvm_skip_emulated_instruction(vcpu);
1064 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);