2 * Kernel-based Virtual Machine driver for Linux
3 * cpuid support routines
5 * derived from arch/x86/kvm/x86.c
7 * Copyright 2011 Red Hat, Inc. and/or its affiliates.
8 * Copyright IBM Corporation, 2008
10 * This work is licensed under the terms of the GNU GPL, version 2. See
11 * the COPYING file in the top-level directory.
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>
21 #include <asm/processor.h>
23 #include <asm/fpu/xstate.h>
30 static u32 xstate_required_size(u64 xstate_bv, bool compacted)
33 u32 ret = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
35 xstate_bv &= XFEATURE_MASK_EXTEND;
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);
51 bool kvm_mpx_supported(void)
53 return ((host_xcr0 & (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR))
54 && kvm_x86_ops->mpx_supported());
56 EXPORT_SYMBOL_GPL(kvm_mpx_supported);
58 u64 kvm_supported_xcr0(void)
60 u64 xcr0 = KVM_SUPPORTED_XCR0 & host_xcr0;
62 if (!kvm_mpx_supported())
63 xcr0 &= ~(XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR);
68 #define F(x) bit(X86_FEATURE_##x)
70 /* For scattered features from cpufeatures.h; we currently expose none */
71 #define KF(x) bit(KVM_CPUID_BIT_##x)
73 int kvm_update_cpuid(struct kvm_vcpu *vcpu)
75 struct kvm_cpuid_entry2 *best;
76 struct kvm_lapic *apic = vcpu->arch.apic;
78 best = kvm_find_cpuid_entry(vcpu, 1, 0);
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);
89 best->edx &= ~F(APIC);
90 if (vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE)
94 if (best->ecx & F(TSC_DEADLINE_TIMER))
95 apic->lapic_timer.timer_mode_mask = 3 << 17;
97 apic->lapic_timer.timer_mode_mask = 1 << 17;
100 best = kvm_find_cpuid_entry(vcpu, 7, 0);
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);
110 best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
112 vcpu->arch.guest_supported_xcr0 = 0;
113 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
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);
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);
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.
130 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
132 int vaddr_bits = (best->eax & 0xff00) >> 8;
134 if (vaddr_bits != 48 && vaddr_bits != 57 && vaddr_bits != 0)
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);
143 /* Update physical-address width */
144 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
145 kvm_mmu_reset_context(vcpu);
147 kvm_pmu_refresh(vcpu);
151 static int is_efer_nx(void)
153 unsigned long long efer = 0;
155 rdmsrl_safe(MSR_EFER, &efer);
156 return efer & EFER_NX;
159 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
162 struct kvm_cpuid_entry2 *e, *entry;
165 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
166 e = &vcpu->arch.cpuid_entries[i];
167 if (e->function == 0x80000001) {
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");
178 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
180 struct kvm_cpuid_entry2 *best;
182 best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
183 if (!best || best->eax < 0x80000008)
185 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
187 return best->eax & 0xff;
191 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr);
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)
199 struct kvm_cpuid_entry *cpuid_entries = NULL;
202 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
207 vmalloc(array_size(sizeof(struct kvm_cpuid_entry),
212 if (copy_from_user(cpuid_entries, entries,
213 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
216 for (i = 0; i < cpuid->nent; i++) {
217 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
218 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
219 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
220 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
221 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
222 vcpu->arch.cpuid_entries[i].index = 0;
223 vcpu->arch.cpuid_entries[i].flags = 0;
224 vcpu->arch.cpuid_entries[i].padding[0] = 0;
225 vcpu->arch.cpuid_entries[i].padding[1] = 0;
226 vcpu->arch.cpuid_entries[i].padding[2] = 0;
228 vcpu->arch.cpuid_nent = cpuid->nent;
229 cpuid_fix_nx_cap(vcpu);
230 kvm_apic_set_version(vcpu);
231 kvm_x86_ops->cpuid_update(vcpu);
232 r = kvm_update_cpuid(vcpu);
235 vfree(cpuid_entries);
239 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
240 struct kvm_cpuid2 *cpuid,
241 struct kvm_cpuid_entry2 __user *entries)
246 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
249 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
250 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
252 vcpu->arch.cpuid_nent = cpuid->nent;
253 kvm_apic_set_version(vcpu);
254 kvm_x86_ops->cpuid_update(vcpu);
255 r = kvm_update_cpuid(vcpu);
260 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
261 struct kvm_cpuid2 *cpuid,
262 struct kvm_cpuid_entry2 __user *entries)
267 if (cpuid->nent < vcpu->arch.cpuid_nent)
270 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
271 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
276 cpuid->nent = vcpu->arch.cpuid_nent;
280 static void cpuid_mask(u32 *word, int wordnum)
282 *word &= boot_cpu_data.x86_capability[wordnum];
285 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
288 entry->function = function;
289 entry->index = index;
290 cpuid_count(entry->function, entry->index,
291 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
295 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry,
296 u32 func, u32 index, int *nent, int maxnent)
304 entry->ecx = F(MOVBE);
308 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
310 entry->ecx = F(RDPID);
316 entry->function = func;
317 entry->index = index;
322 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
323 u32 index, int *nent, int maxnent)
326 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
328 unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
330 unsigned f_lm = F(LM);
332 unsigned f_gbpages = 0;
335 unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
336 unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
337 unsigned f_mpx = kvm_mpx_supported() ? F(MPX) : 0;
338 unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0;
339 unsigned f_umip = kvm_x86_ops->umip_emulated() ? F(UMIP) : 0;
342 const u32 kvm_cpuid_1_edx_x86_features =
343 F(FPU) | F(VME) | F(DE) | F(PSE) |
344 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
345 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
346 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
347 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
348 0 /* Reserved, DS, ACPI */ | F(MMX) |
349 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
350 0 /* HTT, TM, Reserved, PBE */;
351 /* cpuid 0x80000001.edx */
352 const u32 kvm_cpuid_8000_0001_edx_x86_features =
353 F(FPU) | F(VME) | F(DE) | F(PSE) |
354 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
355 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
356 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
357 F(PAT) | F(PSE36) | 0 /* Reserved */ |
358 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
359 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
360 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
362 const u32 kvm_cpuid_1_ecx_x86_features =
363 /* NOTE: MONITOR (and MWAIT) are emulated as NOP,
364 * but *not* advertised to guests via CPUID ! */
365 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
366 0 /* DS-CPL, VMX, SMX, EST */ |
367 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
368 F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
369 F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
370 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
371 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
373 /* cpuid 0x80000001.ecx */
374 const u32 kvm_cpuid_8000_0001_ecx_x86_features =
375 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
376 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
377 F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
378 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM) |
379 F(TOPOEXT) | F(PERFCTR_CORE);
381 /* cpuid 0x80000008.ebx */
382 const u32 kvm_cpuid_8000_0008_ebx_x86_features =
383 F(AMD_IBPB) | F(AMD_IBRS) | F(AMD_SSBD) | F(VIRT_SSBD) |
386 /* cpuid 0xC0000001.edx */
387 const u32 kvm_cpuid_C000_0001_edx_x86_features =
388 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
389 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
393 const u32 kvm_cpuid_7_0_ebx_x86_features =
394 F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
395 F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
396 F(ADX) | F(SMAP) | F(AVX512IFMA) | F(AVX512F) | F(AVX512PF) |
397 F(AVX512ER) | F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) |
398 F(SHA_NI) | F(AVX512BW) | F(AVX512VL);
400 /* cpuid 0xD.1.eax */
401 const u32 kvm_cpuid_D_1_eax_x86_features =
402 F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves;
405 const u32 kvm_cpuid_7_0_ecx_x86_features =
406 F(AVX512VBMI) | F(LA57) | F(PKU) | 0 /*OSPKE*/ |
407 F(AVX512_VPOPCNTDQ) | F(UMIP) | F(AVX512_VBMI2) | F(GFNI) |
408 F(VAES) | F(VPCLMULQDQ) | F(AVX512_VNNI) | F(AVX512_BITALG);
411 const u32 kvm_cpuid_7_0_edx_x86_features =
412 F(AVX512_4VNNIW) | F(AVX512_4FMAPS) | F(SPEC_CTRL) |
413 F(SPEC_CTRL_SSBD) | F(ARCH_CAPABILITIES);
415 /* all calls to cpuid_count() should be made on the same cpu */
420 if (*nent >= maxnent)
423 do_cpuid_1_ent(entry, function, index);
428 entry->eax = min(entry->eax, (u32)0xd);
431 entry->edx &= kvm_cpuid_1_edx_x86_features;
432 cpuid_mask(&entry->edx, CPUID_1_EDX);
433 entry->ecx &= kvm_cpuid_1_ecx_x86_features;
434 cpuid_mask(&entry->ecx, CPUID_1_ECX);
435 /* we support x2apic emulation even if host does not support
436 * it since we emulate x2apic in software */
437 entry->ecx |= F(X2APIC);
439 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
440 * may return different values. This forces us to get_cpu() before
441 * issuing the first command, and also to emulate this annoying behavior
442 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
444 int t, times = entry->eax & 0xff;
446 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
447 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
448 for (t = 1; t < times; ++t) {
449 if (*nent >= maxnent)
452 do_cpuid_1_ent(&entry[t], function, 0);
453 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
458 /* function 4 has additional index. */
462 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
463 /* read more entries until cache_type is zero */
465 if (*nent >= maxnent)
468 cache_type = entry[i - 1].eax & 0x1f;
471 do_cpuid_1_ent(&entry[i], function, i);
473 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
478 case 6: /* Thermal management */
479 entry->eax = 0x4; /* allow ARAT */
485 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
486 /* Mask ebx against host capability word 9 */
488 entry->ebx &= kvm_cpuid_7_0_ebx_x86_features;
489 cpuid_mask(&entry->ebx, CPUID_7_0_EBX);
490 // TSC_ADJUST is emulated
491 entry->ebx |= F(TSC_ADJUST);
492 entry->ecx &= kvm_cpuid_7_0_ecx_x86_features;
493 cpuid_mask(&entry->ecx, CPUID_7_ECX);
494 entry->ecx |= f_umip;
495 /* PKU is not yet implemented for shadow paging. */
496 if (!tdp_enabled || !boot_cpu_has(X86_FEATURE_OSPKE))
497 entry->ecx &= ~F(PKU);
498 entry->edx &= kvm_cpuid_7_0_edx_x86_features;
499 cpuid_mask(&entry->edx, CPUID_7_EDX);
501 * We emulate ARCH_CAPABILITIES in software even
502 * if the host doesn't support it.
504 entry->edx |= F(ARCH_CAPABILITIES);
515 case 0xa: { /* Architectural Performance Monitoring */
516 struct x86_pmu_capability cap;
517 union cpuid10_eax eax;
518 union cpuid10_edx edx;
520 perf_get_x86_pmu_capability(&cap);
523 * Only support guest architectural pmu on a host
524 * with architectural pmu.
527 memset(&cap, 0, sizeof(cap));
529 eax.split.version_id = min(cap.version, 2);
530 eax.split.num_counters = cap.num_counters_gp;
531 eax.split.bit_width = cap.bit_width_gp;
532 eax.split.mask_length = cap.events_mask_len;
534 edx.split.num_counters_fixed = cap.num_counters_fixed;
535 edx.split.bit_width_fixed = cap.bit_width_fixed;
536 edx.split.reserved = 0;
538 entry->eax = eax.full;
539 entry->ebx = cap.events_mask;
541 entry->edx = edx.full;
544 /* function 0xb has additional index. */
548 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
549 /* read more entries until level_type is zero */
551 if (*nent >= maxnent)
554 level_type = entry[i - 1].ecx & 0xff00;
557 do_cpuid_1_ent(&entry[i], function, i);
559 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
566 u64 supported = kvm_supported_xcr0();
568 entry->eax &= supported;
569 entry->ebx = xstate_required_size(supported, false);
570 entry->ecx = entry->ebx;
571 entry->edx &= supported >> 32;
572 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
576 for (idx = 1, i = 1; idx < 64; ++idx) {
577 u64 mask = ((u64)1 << idx);
578 if (*nent >= maxnent)
581 do_cpuid_1_ent(&entry[i], function, idx);
583 entry[i].eax &= kvm_cpuid_D_1_eax_x86_features;
584 cpuid_mask(&entry[i].eax, CPUID_D_1_EAX);
586 if (entry[i].eax & (F(XSAVES)|F(XSAVEC)))
588 xstate_required_size(supported,
591 if (entry[i].eax == 0 || !(supported & mask))
593 if (WARN_ON_ONCE(entry[i].ecx & 1))
599 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
605 case KVM_CPUID_SIGNATURE: {
606 static const char signature[12] = "KVMKVMKVM\0\0";
607 const u32 *sigptr = (const u32 *)signature;
608 entry->eax = KVM_CPUID_FEATURES;
609 entry->ebx = sigptr[0];
610 entry->ecx = sigptr[1];
611 entry->edx = sigptr[2];
614 case KVM_CPUID_FEATURES:
615 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
616 (1 << KVM_FEATURE_NOP_IO_DELAY) |
617 (1 << KVM_FEATURE_CLOCKSOURCE2) |
618 (1 << KVM_FEATURE_ASYNC_PF) |
619 (1 << KVM_FEATURE_PV_EOI) |
620 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
621 (1 << KVM_FEATURE_PV_UNHALT) |
622 (1 << KVM_FEATURE_PV_TLB_FLUSH) |
623 (1 << KVM_FEATURE_ASYNC_PF_VMEXIT);
626 entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
633 entry->eax = min(entry->eax, 0x8000001f);
636 entry->edx &= kvm_cpuid_8000_0001_edx_x86_features;
637 cpuid_mask(&entry->edx, CPUID_8000_0001_EDX);
638 entry->ecx &= kvm_cpuid_8000_0001_ecx_x86_features;
639 cpuid_mask(&entry->ecx, CPUID_8000_0001_ECX);
641 case 0x80000007: /* Advanced power management */
642 /* invariant TSC is CPUID.80000007H:EDX[8] */
643 entry->edx &= (1 << 8);
644 /* mask against host */
645 entry->edx &= boot_cpu_data.x86_power;
646 entry->eax = entry->ebx = entry->ecx = 0;
649 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
650 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
651 unsigned phys_as = entry->eax & 0xff;
655 entry->eax = g_phys_as | (virt_as << 8);
658 * IBRS, IBPB and VIRT_SSBD aren't necessarily present in
661 if (boot_cpu_has(X86_FEATURE_AMD_IBPB))
662 entry->ebx |= F(AMD_IBPB);
663 if (boot_cpu_has(X86_FEATURE_AMD_IBRS))
664 entry->ebx |= F(AMD_IBRS);
665 if (boot_cpu_has(X86_FEATURE_VIRT_SSBD))
666 entry->ebx |= F(VIRT_SSBD);
667 entry->ebx &= kvm_cpuid_8000_0008_ebx_x86_features;
668 cpuid_mask(&entry->ebx, CPUID_8000_0008_EBX);
670 * The preference is to use SPEC CTRL MSR instead of the
673 if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD) &&
674 !boot_cpu_has(X86_FEATURE_AMD_SSBD))
675 entry->ebx |= F(VIRT_SSBD);
679 entry->ecx = entry->edx = 0;
685 /*Add support for Centaur's CPUID instruction*/
687 /*Just support up to 0xC0000004 now*/
688 entry->eax = min(entry->eax, 0xC0000004);
691 entry->edx &= kvm_cpuid_C000_0001_edx_x86_features;
692 cpuid_mask(&entry->edx, CPUID_C000_0001_EDX);
694 case 3: /* Processor serial number */
695 case 5: /* MONITOR/MWAIT */
700 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
704 kvm_x86_ops->set_supported_cpuid(function, entry);
714 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
715 u32 idx, int *nent, int maxnent, unsigned int type)
717 if (type == KVM_GET_EMULATED_CPUID)
718 return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
720 return __do_cpuid_ent(entry, func, idx, nent, maxnent);
725 struct kvm_cpuid_param {
729 bool (*qualifier)(const struct kvm_cpuid_param *param);
732 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
734 return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
737 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
738 __u32 num_entries, unsigned int ioctl_type)
743 if (ioctl_type != KVM_GET_EMULATED_CPUID)
747 * We want to make sure that ->padding is being passed clean from
748 * userspace in case we want to use it for something in the future.
750 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
751 * have to give ourselves satisfied only with the emulated side. /me
754 for (i = 0; i < num_entries; i++) {
755 if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
758 if (pad[0] || pad[1] || pad[2])
764 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
765 struct kvm_cpuid_entry2 __user *entries,
768 struct kvm_cpuid_entry2 *cpuid_entries;
769 int limit, nent = 0, r = -E2BIG, i;
771 static const struct kvm_cpuid_param param[] = {
772 { .func = 0, .has_leaf_count = true },
773 { .func = 0x80000000, .has_leaf_count = true },
774 { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
775 { .func = KVM_CPUID_SIGNATURE },
776 { .func = KVM_CPUID_FEATURES },
781 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
782 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
784 if (sanity_check_entries(entries, cpuid->nent, type))
788 cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
793 for (i = 0; i < ARRAY_SIZE(param); i++) {
794 const struct kvm_cpuid_param *ent = ¶m[i];
796 if (ent->qualifier && !ent->qualifier(ent))
799 r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
800 &nent, cpuid->nent, type);
805 if (!ent->has_leaf_count)
808 limit = cpuid_entries[nent - 1].eax;
809 for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
810 r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
811 &nent, cpuid->nent, type);
818 if (copy_to_user(entries, cpuid_entries,
819 nent * sizeof(struct kvm_cpuid_entry2)))
825 vfree(cpuid_entries);
830 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
832 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
833 struct kvm_cpuid_entry2 *ej;
835 int nent = vcpu->arch.cpuid_nent;
837 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
838 /* when no next entry is found, the current entry[i] is reselected */
841 ej = &vcpu->arch.cpuid_entries[j];
842 } while (ej->function != e->function);
844 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
849 /* find an entry with matching function, matching index (if needed), and that
850 * should be read next (if it's stateful) */
851 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
852 u32 function, u32 index)
854 if (e->function != function)
856 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
858 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
859 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
864 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
865 u32 function, u32 index)
868 struct kvm_cpuid_entry2 *best = NULL;
870 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
871 struct kvm_cpuid_entry2 *e;
873 e = &vcpu->arch.cpuid_entries[i];
874 if (is_matching_cpuid_entry(e, function, index)) {
875 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
876 move_to_next_stateful_cpuid_entry(vcpu, i);
883 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
886 * If no match is found, check whether we exceed the vCPU's limit
887 * and return the content of the highest valid _standard_ leaf instead.
888 * This is to satisfy the CPUID specification.
890 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
891 u32 function, u32 index)
893 struct kvm_cpuid_entry2 *maxlevel;
895 maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
896 if (!maxlevel || maxlevel->eax >= function)
898 if (function & 0x80000000) {
899 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
903 return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
906 bool kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx,
907 u32 *ecx, u32 *edx, bool check_limit)
909 u32 function = *eax, index = *ecx;
910 struct kvm_cpuid_entry2 *best;
911 bool entry_found = true;
913 best = kvm_find_cpuid_entry(vcpu, function, index);
920 best = check_cpuid_limit(vcpu, function, index);
930 *eax = *ebx = *ecx = *edx = 0;
931 trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx, entry_found);
934 EXPORT_SYMBOL_GPL(kvm_cpuid);
936 int kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
938 u32 eax, ebx, ecx, edx;
940 if (cpuid_fault_enabled(vcpu) && !kvm_require_cpl(vcpu, 0))
943 eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
944 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
945 kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx, true);
946 kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
947 kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
948 kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
949 kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
950 return kvm_skip_emulated_instruction(vcpu);
952 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);