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 <asm/processor.h>
21 #include <asm/fpu/xstate.h>
28 static u32 xstate_required_size(u64 xstate_bv, bool compacted)
31 u32 ret = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
33 xstate_bv &= XFEATURE_MASK_EXTEND;
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);
49 bool kvm_mpx_supported(void)
51 return ((host_xcr0 & (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR))
52 && kvm_x86_ops->mpx_supported());
54 EXPORT_SYMBOL_GPL(kvm_mpx_supported);
56 u64 kvm_supported_xcr0(void)
58 u64 xcr0 = KVM_SUPPORTED_XCR0 & host_xcr0;
60 if (!kvm_mpx_supported())
61 xcr0 &= ~(XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR);
66 #define F(x) bit(X86_FEATURE_##x)
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)
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 in the canonical
128 * address checks; exit if it is ever changed.
130 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
131 if (best && ((best->eax & 0xff00) >> 8) != 48 &&
132 ((best->eax & 0xff00) >> 8) != 0)
135 /* Update physical-address width */
136 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
138 kvm_pmu_refresh(vcpu);
142 static int is_efer_nx(void)
144 unsigned long long efer = 0;
146 rdmsrl_safe(MSR_EFER, &efer);
147 return efer & EFER_NX;
150 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
153 struct kvm_cpuid_entry2 *e, *entry;
156 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
157 e = &vcpu->arch.cpuid_entries[i];
158 if (e->function == 0x80000001) {
163 if (entry && (entry->edx & F(NX)) && !is_efer_nx()) {
164 entry->edx &= ~F(NX);
165 printk(KERN_INFO "kvm: guest NX capability removed\n");
169 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
171 struct kvm_cpuid_entry2 *best;
173 best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
174 if (!best || best->eax < 0x80000008)
176 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
178 return best->eax & 0xff;
182 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr);
184 /* when an old userspace process fills a new kernel module */
185 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
186 struct kvm_cpuid *cpuid,
187 struct kvm_cpuid_entry __user *entries)
190 struct kvm_cpuid_entry *cpuid_entries = NULL;
193 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
197 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) *
202 if (copy_from_user(cpuid_entries, entries,
203 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
206 for (i = 0; i < cpuid->nent; i++) {
207 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
208 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
209 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
210 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
211 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
212 vcpu->arch.cpuid_entries[i].index = 0;
213 vcpu->arch.cpuid_entries[i].flags = 0;
214 vcpu->arch.cpuid_entries[i].padding[0] = 0;
215 vcpu->arch.cpuid_entries[i].padding[1] = 0;
216 vcpu->arch.cpuid_entries[i].padding[2] = 0;
218 vcpu->arch.cpuid_nent = cpuid->nent;
219 cpuid_fix_nx_cap(vcpu);
220 kvm_apic_set_version(vcpu);
221 kvm_x86_ops->cpuid_update(vcpu);
222 r = kvm_update_cpuid(vcpu);
225 vfree(cpuid_entries);
229 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
230 struct kvm_cpuid2 *cpuid,
231 struct kvm_cpuid_entry2 __user *entries)
236 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
239 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
240 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
242 vcpu->arch.cpuid_nent = cpuid->nent;
243 kvm_apic_set_version(vcpu);
244 kvm_x86_ops->cpuid_update(vcpu);
245 r = kvm_update_cpuid(vcpu);
250 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
251 struct kvm_cpuid2 *cpuid,
252 struct kvm_cpuid_entry2 __user *entries)
257 if (cpuid->nent < vcpu->arch.cpuid_nent)
260 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
261 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
266 cpuid->nent = vcpu->arch.cpuid_nent;
270 static void cpuid_mask(u32 *word, int wordnum)
272 *word &= boot_cpu_data.x86_capability[wordnum];
275 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
278 entry->function = function;
279 entry->index = index;
280 cpuid_count(entry->function, entry->index,
281 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
285 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry,
286 u32 func, u32 index, int *nent, int maxnent)
290 entry->eax = 1; /* only one leaf currently */
294 entry->ecx = F(MOVBE);
301 entry->function = func;
302 entry->index = index;
307 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
308 u32 index, int *nent, int maxnent)
311 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
313 unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
315 unsigned f_lm = F(LM);
317 unsigned f_gbpages = 0;
320 unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
321 unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
322 unsigned f_mpx = kvm_mpx_supported() ? F(MPX) : 0;
323 unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0;
326 const u32 kvm_cpuid_1_edx_x86_features =
327 F(FPU) | F(VME) | F(DE) | F(PSE) |
328 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
329 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
330 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
331 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
332 0 /* Reserved, DS, ACPI */ | F(MMX) |
333 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
334 0 /* HTT, TM, Reserved, PBE */;
335 /* cpuid 0x80000001.edx */
336 const u32 kvm_cpuid_8000_0001_edx_x86_features =
337 F(FPU) | F(VME) | F(DE) | F(PSE) |
338 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
339 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
340 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
341 F(PAT) | F(PSE36) | 0 /* Reserved */ |
342 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
343 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
344 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
346 const u32 kvm_cpuid_1_ecx_x86_features =
347 /* NOTE: MONITOR (and MWAIT) are emulated as NOP,
348 * but *not* advertised to guests via CPUID ! */
349 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
350 0 /* DS-CPL, VMX, SMX, EST */ |
351 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
352 F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
353 F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
354 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
355 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
357 /* cpuid 0x80000001.ecx */
358 const u32 kvm_cpuid_8000_0001_ecx_x86_features =
359 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
360 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
361 F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
362 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
364 /* cpuid 0xC0000001.edx */
365 const u32 kvm_cpuid_C000_0001_edx_x86_features =
366 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
367 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
371 const u32 kvm_cpuid_7_0_ebx_x86_features =
372 F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
373 F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
374 F(ADX) | F(SMAP) | F(AVX512IFMA) | F(AVX512F) | F(AVX512PF) |
375 F(AVX512ER) | F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) |
376 F(SHA_NI) | F(AVX512BW) | F(AVX512VL);
378 /* cpuid 0xD.1.eax */
379 const u32 kvm_cpuid_D_1_eax_x86_features =
380 F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves;
383 const u32 kvm_cpuid_7_0_ecx_x86_features =
384 F(AVX512VBMI) | F(PKU) | 0 /*OSPKE*/ | F(AVX512_VPOPCNTDQ);
387 const u32 kvm_cpuid_7_0_edx_x86_features =
388 KF(AVX512_4VNNIW) | KF(AVX512_4FMAPS);
390 /* all calls to cpuid_count() should be made on the same cpu */
395 if (*nent >= maxnent)
398 do_cpuid_1_ent(entry, function, index);
403 entry->eax = min(entry->eax, (u32)0xd);
406 entry->edx &= kvm_cpuid_1_edx_x86_features;
407 cpuid_mask(&entry->edx, CPUID_1_EDX);
408 entry->ecx &= kvm_cpuid_1_ecx_x86_features;
409 cpuid_mask(&entry->ecx, CPUID_1_ECX);
410 /* we support x2apic emulation even if host does not support
411 * it since we emulate x2apic in software */
412 entry->ecx |= F(X2APIC);
414 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
415 * may return different values. This forces us to get_cpu() before
416 * issuing the first command, and also to emulate this annoying behavior
417 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
419 int t, times = entry->eax & 0xff;
421 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
422 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
423 for (t = 1; t < times; ++t) {
424 if (*nent >= maxnent)
427 do_cpuid_1_ent(&entry[t], function, 0);
428 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
433 /* function 4 has additional index. */
437 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
438 /* read more entries until cache_type is zero */
440 if (*nent >= maxnent)
443 cache_type = entry[i - 1].eax & 0x1f;
446 do_cpuid_1_ent(&entry[i], function, i);
448 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
453 case 6: /* Thermal management */
454 entry->eax = 0x4; /* allow ARAT */
460 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
461 /* Mask ebx against host capability word 9 */
463 entry->ebx &= kvm_cpuid_7_0_ebx_x86_features;
464 cpuid_mask(&entry->ebx, CPUID_7_0_EBX);
465 // TSC_ADJUST is emulated
466 entry->ebx |= F(TSC_ADJUST);
467 entry->ecx &= kvm_cpuid_7_0_ecx_x86_features;
468 cpuid_mask(&entry->ecx, CPUID_7_ECX);
469 /* PKU is not yet implemented for shadow paging. */
471 entry->ecx &= ~F(PKU);
472 entry->edx &= kvm_cpuid_7_0_edx_x86_features;
473 entry->edx &= get_scattered_cpuid_leaf(7, 0, CPUID_EDX);
484 case 0xa: { /* Architectural Performance Monitoring */
485 struct x86_pmu_capability cap;
486 union cpuid10_eax eax;
487 union cpuid10_edx edx;
489 perf_get_x86_pmu_capability(&cap);
492 * Only support guest architectural pmu on a host
493 * with architectural pmu.
496 memset(&cap, 0, sizeof(cap));
498 eax.split.version_id = min(cap.version, 2);
499 eax.split.num_counters = cap.num_counters_gp;
500 eax.split.bit_width = cap.bit_width_gp;
501 eax.split.mask_length = cap.events_mask_len;
503 edx.split.num_counters_fixed = cap.num_counters_fixed;
504 edx.split.bit_width_fixed = cap.bit_width_fixed;
505 edx.split.reserved = 0;
507 entry->eax = eax.full;
508 entry->ebx = cap.events_mask;
510 entry->edx = edx.full;
513 /* function 0xb has additional index. */
517 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
518 /* read more entries until level_type is zero */
520 if (*nent >= maxnent)
523 level_type = entry[i - 1].ecx & 0xff00;
526 do_cpuid_1_ent(&entry[i], function, i);
528 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
535 u64 supported = kvm_supported_xcr0();
537 entry->eax &= supported;
538 entry->ebx = xstate_required_size(supported, false);
539 entry->ecx = entry->ebx;
540 entry->edx &= supported >> 32;
541 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
545 for (idx = 1, i = 1; idx < 64; ++idx) {
546 u64 mask = ((u64)1 << idx);
547 if (*nent >= maxnent)
550 do_cpuid_1_ent(&entry[i], function, idx);
552 entry[i].eax &= kvm_cpuid_D_1_eax_x86_features;
553 cpuid_mask(&entry[i].eax, CPUID_D_1_EAX);
555 if (entry[i].eax & (F(XSAVES)|F(XSAVEC)))
557 xstate_required_size(supported,
560 if (entry[i].eax == 0 || !(supported & mask))
562 if (WARN_ON_ONCE(entry[i].ecx & 1))
568 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
574 case KVM_CPUID_SIGNATURE: {
575 static const char signature[12] = "KVMKVMKVM\0\0";
576 const u32 *sigptr = (const u32 *)signature;
577 entry->eax = KVM_CPUID_FEATURES;
578 entry->ebx = sigptr[0];
579 entry->ecx = sigptr[1];
580 entry->edx = sigptr[2];
583 case KVM_CPUID_FEATURES:
584 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
585 (1 << KVM_FEATURE_NOP_IO_DELAY) |
586 (1 << KVM_FEATURE_CLOCKSOURCE2) |
587 (1 << KVM_FEATURE_ASYNC_PF) |
588 (1 << KVM_FEATURE_PV_EOI) |
589 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
590 (1 << KVM_FEATURE_PV_UNHALT);
593 entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
600 entry->eax = min(entry->eax, 0x8000001a);
603 entry->edx &= kvm_cpuid_8000_0001_edx_x86_features;
604 cpuid_mask(&entry->edx, CPUID_8000_0001_EDX);
605 entry->ecx &= kvm_cpuid_8000_0001_ecx_x86_features;
606 cpuid_mask(&entry->ecx, CPUID_8000_0001_ECX);
608 case 0x80000007: /* Advanced power management */
609 /* invariant TSC is CPUID.80000007H:EDX[8] */
610 entry->edx &= (1 << 8);
611 /* mask against host */
612 entry->edx &= boot_cpu_data.x86_power;
613 entry->eax = entry->ebx = entry->ecx = 0;
616 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
617 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
618 unsigned phys_as = entry->eax & 0xff;
622 entry->eax = g_phys_as | (virt_as << 8);
623 entry->ebx = entry->edx = 0;
627 entry->ecx = entry->edx = 0;
633 /*Add support for Centaur's CPUID instruction*/
635 /*Just support up to 0xC0000004 now*/
636 entry->eax = min(entry->eax, 0xC0000004);
639 entry->edx &= kvm_cpuid_C000_0001_edx_x86_features;
640 cpuid_mask(&entry->edx, CPUID_C000_0001_EDX);
642 case 3: /* Processor serial number */
643 case 5: /* MONITOR/MWAIT */
648 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
652 kvm_x86_ops->set_supported_cpuid(function, entry);
662 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
663 u32 idx, int *nent, int maxnent, unsigned int type)
665 if (type == KVM_GET_EMULATED_CPUID)
666 return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
668 return __do_cpuid_ent(entry, func, idx, nent, maxnent);
673 struct kvm_cpuid_param {
677 bool (*qualifier)(const struct kvm_cpuid_param *param);
680 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
682 return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
685 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
686 __u32 num_entries, unsigned int ioctl_type)
691 if (ioctl_type != KVM_GET_EMULATED_CPUID)
695 * We want to make sure that ->padding is being passed clean from
696 * userspace in case we want to use it for something in the future.
698 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
699 * have to give ourselves satisfied only with the emulated side. /me
702 for (i = 0; i < num_entries; i++) {
703 if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
706 if (pad[0] || pad[1] || pad[2])
712 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
713 struct kvm_cpuid_entry2 __user *entries,
716 struct kvm_cpuid_entry2 *cpuid_entries;
717 int limit, nent = 0, r = -E2BIG, i;
719 static const struct kvm_cpuid_param param[] = {
720 { .func = 0, .has_leaf_count = true },
721 { .func = 0x80000000, .has_leaf_count = true },
722 { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
723 { .func = KVM_CPUID_SIGNATURE },
724 { .func = KVM_CPUID_FEATURES },
729 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
730 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
732 if (sanity_check_entries(entries, cpuid->nent, type))
736 cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
741 for (i = 0; i < ARRAY_SIZE(param); i++) {
742 const struct kvm_cpuid_param *ent = ¶m[i];
744 if (ent->qualifier && !ent->qualifier(ent))
747 r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
748 &nent, cpuid->nent, type);
753 if (!ent->has_leaf_count)
756 limit = cpuid_entries[nent - 1].eax;
757 for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
758 r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
759 &nent, cpuid->nent, type);
766 if (copy_to_user(entries, cpuid_entries,
767 nent * sizeof(struct kvm_cpuid_entry2)))
773 vfree(cpuid_entries);
778 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
780 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
781 int j, nent = vcpu->arch.cpuid_nent;
783 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
784 /* when no next entry is found, the current entry[i] is reselected */
785 for (j = i + 1; ; j = (j + 1) % nent) {
786 struct kvm_cpuid_entry2 *ej = &vcpu->arch.cpuid_entries[j];
787 if (ej->function == e->function) {
788 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
792 return 0; /* silence gcc, even though control never reaches here */
795 /* find an entry with matching function, matching index (if needed), and that
796 * should be read next (if it's stateful) */
797 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
798 u32 function, u32 index)
800 if (e->function != function)
802 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
804 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
805 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
810 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
811 u32 function, u32 index)
814 struct kvm_cpuid_entry2 *best = NULL;
816 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
817 struct kvm_cpuid_entry2 *e;
819 e = &vcpu->arch.cpuid_entries[i];
820 if (is_matching_cpuid_entry(e, function, index)) {
821 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
822 move_to_next_stateful_cpuid_entry(vcpu, i);
829 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
832 * If no match is found, check whether we exceed the vCPU's limit
833 * and return the content of the highest valid _standard_ leaf instead.
834 * This is to satisfy the CPUID specification.
836 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
837 u32 function, u32 index)
839 struct kvm_cpuid_entry2 *maxlevel;
841 maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
842 if (!maxlevel || maxlevel->eax >= function)
844 if (function & 0x80000000) {
845 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
849 return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
852 void kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
854 u32 function = *eax, index = *ecx;
855 struct kvm_cpuid_entry2 *best;
857 best = kvm_find_cpuid_entry(vcpu, function, index);
860 best = check_cpuid_limit(vcpu, function, index);
868 *eax = *ebx = *ecx = *edx = 0;
869 trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx);
871 EXPORT_SYMBOL_GPL(kvm_cpuid);
873 int kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
875 u32 eax, ebx, ecx, edx;
877 eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
878 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
879 kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx);
880 kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
881 kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
882 kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
883 kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
884 return kvm_skip_emulated_instruction(vcpu);
886 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);