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/module.h>
17 #include <linux/vmalloc.h>
18 #include <linux/uaccess.h>
20 #include <asm/fpu/xstate.h>
27 static u32 xstate_required_size(u64 xstate_bv, bool compacted)
30 u32 ret = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
32 xstate_bv &= XFEATURE_MASK_EXTEND;
34 if (xstate_bv & 0x1) {
35 u32 eax, ebx, ecx, edx, offset;
36 cpuid_count(0xD, feature_bit, &eax, &ebx, &ecx, &edx);
37 offset = compacted ? ret : ebx;
38 ret = max(ret, offset + eax);
48 bool kvm_mpx_supported(void)
50 return ((host_xcr0 & (XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR))
51 && kvm_x86_ops->mpx_supported());
53 EXPORT_SYMBOL_GPL(kvm_mpx_supported);
55 u64 kvm_supported_xcr0(void)
57 u64 xcr0 = KVM_SUPPORTED_XCR0 & host_xcr0;
59 if (!kvm_mpx_supported())
60 xcr0 &= ~(XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR);
65 #define F(x) bit(X86_FEATURE_##x)
67 int kvm_update_cpuid(struct kvm_vcpu *vcpu)
69 struct kvm_cpuid_entry2 *best;
70 struct kvm_lapic *apic = vcpu->arch.apic;
72 best = kvm_find_cpuid_entry(vcpu, 1, 0);
76 /* Update OSXSAVE bit */
77 if (cpu_has_xsave && best->function == 0x1) {
78 best->ecx &= ~F(OSXSAVE);
79 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE))
80 best->ecx |= F(OSXSAVE);
84 if (best->ecx & F(TSC_DEADLINE_TIMER))
85 apic->lapic_timer.timer_mode_mask = 3 << 17;
87 apic->lapic_timer.timer_mode_mask = 1 << 17;
90 best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
92 vcpu->arch.guest_supported_xcr0 = 0;
93 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
95 vcpu->arch.guest_supported_xcr0 =
96 (best->eax | ((u64)best->edx << 32)) &
98 vcpu->arch.guest_xstate_size = best->ebx =
99 xstate_required_size(vcpu->arch.xcr0, false);
102 best = kvm_find_cpuid_entry(vcpu, 0xD, 1);
103 if (best && (best->eax & (F(XSAVES) | F(XSAVEC))))
104 best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
106 kvm_x86_ops->fpu_activate(vcpu);
109 * The existing code assumes virtual address is 48-bit in the canonical
110 * address checks; exit if it is ever changed.
112 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
113 if (best && ((best->eax & 0xff00) >> 8) != 48 &&
114 ((best->eax & 0xff00) >> 8) != 0)
117 /* Update physical-address width */
118 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
120 kvm_pmu_refresh(vcpu);
124 static int is_efer_nx(void)
126 unsigned long long efer = 0;
128 rdmsrl_safe(MSR_EFER, &efer);
129 return efer & EFER_NX;
132 static void cpuid_fix_nx_cap(struct kvm_vcpu *vcpu)
135 struct kvm_cpuid_entry2 *e, *entry;
138 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
139 e = &vcpu->arch.cpuid_entries[i];
140 if (e->function == 0x80000001) {
145 if (entry && (entry->edx & F(NX)) && !is_efer_nx()) {
146 entry->edx &= ~F(NX);
147 printk(KERN_INFO "kvm: guest NX capability removed\n");
151 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
153 struct kvm_cpuid_entry2 *best;
155 best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
156 if (!best || best->eax < 0x80000008)
158 best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
160 return best->eax & 0xff;
164 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr);
166 /* when an old userspace process fills a new kernel module */
167 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
168 struct kvm_cpuid *cpuid,
169 struct kvm_cpuid_entry __user *entries)
172 struct kvm_cpuid_entry *cpuid_entries;
175 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
178 cpuid_entries = vmalloc(sizeof(struct kvm_cpuid_entry) * cpuid->nent);
182 if (copy_from_user(cpuid_entries, entries,
183 cpuid->nent * sizeof(struct kvm_cpuid_entry)))
185 for (i = 0; i < cpuid->nent; i++) {
186 vcpu->arch.cpuid_entries[i].function = cpuid_entries[i].function;
187 vcpu->arch.cpuid_entries[i].eax = cpuid_entries[i].eax;
188 vcpu->arch.cpuid_entries[i].ebx = cpuid_entries[i].ebx;
189 vcpu->arch.cpuid_entries[i].ecx = cpuid_entries[i].ecx;
190 vcpu->arch.cpuid_entries[i].edx = cpuid_entries[i].edx;
191 vcpu->arch.cpuid_entries[i].index = 0;
192 vcpu->arch.cpuid_entries[i].flags = 0;
193 vcpu->arch.cpuid_entries[i].padding[0] = 0;
194 vcpu->arch.cpuid_entries[i].padding[1] = 0;
195 vcpu->arch.cpuid_entries[i].padding[2] = 0;
197 vcpu->arch.cpuid_nent = cpuid->nent;
198 cpuid_fix_nx_cap(vcpu);
199 kvm_apic_set_version(vcpu);
200 kvm_x86_ops->cpuid_update(vcpu);
201 r = kvm_update_cpuid(vcpu);
204 vfree(cpuid_entries);
209 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
210 struct kvm_cpuid2 *cpuid,
211 struct kvm_cpuid_entry2 __user *entries)
216 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
219 if (copy_from_user(&vcpu->arch.cpuid_entries, entries,
220 cpuid->nent * sizeof(struct kvm_cpuid_entry2)))
222 vcpu->arch.cpuid_nent = cpuid->nent;
223 kvm_apic_set_version(vcpu);
224 kvm_x86_ops->cpuid_update(vcpu);
225 r = kvm_update_cpuid(vcpu);
230 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
231 struct kvm_cpuid2 *cpuid,
232 struct kvm_cpuid_entry2 __user *entries)
237 if (cpuid->nent < vcpu->arch.cpuid_nent)
240 if (copy_to_user(entries, &vcpu->arch.cpuid_entries,
241 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
246 cpuid->nent = vcpu->arch.cpuid_nent;
250 static void cpuid_mask(u32 *word, int wordnum)
252 *word &= boot_cpu_data.x86_capability[wordnum];
255 static void do_cpuid_1_ent(struct kvm_cpuid_entry2 *entry, u32 function,
258 entry->function = function;
259 entry->index = index;
260 cpuid_count(entry->function, entry->index,
261 &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
265 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2 *entry,
266 u32 func, u32 index, int *nent, int maxnent)
274 entry->ecx = F(MOVBE);
278 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
280 entry->ecx = F(RDPID);
286 entry->function = func;
287 entry->index = index;
292 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 function,
293 u32 index, int *nent, int maxnent)
296 unsigned f_nx = is_efer_nx() ? F(NX) : 0;
298 unsigned f_gbpages = (kvm_x86_ops->get_lpage_level() == PT_PDPE_LEVEL)
300 unsigned f_lm = F(LM);
302 unsigned f_gbpages = 0;
305 unsigned f_rdtscp = kvm_x86_ops->rdtscp_supported() ? F(RDTSCP) : 0;
306 unsigned f_invpcid = kvm_x86_ops->invpcid_supported() ? F(INVPCID) : 0;
307 unsigned f_mpx = kvm_mpx_supported() ? F(MPX) : 0;
308 unsigned f_xsaves = kvm_x86_ops->xsaves_supported() ? F(XSAVES) : 0;
311 const u32 kvm_supported_word0_x86_features =
312 F(FPU) | F(VME) | F(DE) | F(PSE) |
313 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
314 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
315 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
316 F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
317 0 /* Reserved, DS, ACPI */ | F(MMX) |
318 F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
319 0 /* HTT, TM, Reserved, PBE */;
320 /* cpuid 0x80000001.edx */
321 const u32 kvm_supported_word1_x86_features =
322 F(FPU) | F(VME) | F(DE) | F(PSE) |
323 F(TSC) | F(MSR) | F(PAE) | F(MCE) |
324 F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
325 F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
326 F(PAT) | F(PSE36) | 0 /* Reserved */ |
327 f_nx | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
328 F(FXSR) | F(FXSR_OPT) | f_gbpages | f_rdtscp |
329 0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW);
331 const u32 kvm_supported_word4_x86_features =
332 /* NOTE: MONITOR (and MWAIT) are emulated as NOP,
333 * but *not* advertised to guests via CPUID ! */
334 F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
335 0 /* DS-CPL, VMX, SMX, EST */ |
336 0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
337 F(FMA) | F(CX16) | 0 /* xTPR Update, PDCM */ |
338 F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
339 F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
340 0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
342 /* cpuid 0x80000001.ecx */
343 const u32 kvm_supported_word6_x86_features =
344 F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
345 F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
346 F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
347 0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM);
349 /* cpuid 0x80000008.ebx */
350 const u32 kvm_cpuid_8000_0008_ebx_x86_features =
351 F(AMD_IBPB) | F(AMD_IBRS) | F(AMD_SSBD) | F(VIRT_SSBD) |
352 F(AMD_SSB_NO) | F(AMD_STIBP);
354 /* cpuid 0xC0000001.edx */
355 const u32 kvm_supported_word5_x86_features =
356 F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
357 F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
361 const u32 kvm_supported_word9_x86_features =
362 F(FSGSBASE) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
363 F(BMI2) | F(ERMS) | f_invpcid | F(RTM) | f_mpx | F(RDSEED) |
364 F(ADX) | F(SMAP) | F(AVX512F) | F(AVX512PF) | F(AVX512ER) |
365 F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(PCOMMIT);
367 /* cpuid 0xD.1.eax */
368 const u32 kvm_supported_word10_x86_features =
369 F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | f_xsaves;
372 const u32 kvm_cpuid_7_0_edx_x86_features =
373 F(SPEC_CTRL) | F(SPEC_CTRL_SSBD) | F(ARCH_CAPABILITIES) |
374 F(INTEL_STIBP) | F(MD_CLEAR);
376 /* all calls to cpuid_count() should be made on the same cpu */
381 if (WARN_ON(*nent >= maxnent))
384 do_cpuid_1_ent(entry, function, index);
389 entry->eax = min(entry->eax, (u32)0xd);
392 entry->edx &= kvm_supported_word0_x86_features;
393 cpuid_mask(&entry->edx, 0);
394 entry->ecx &= kvm_supported_word4_x86_features;
395 cpuid_mask(&entry->ecx, 4);
396 /* we support x2apic emulation even if host does not support
397 * it since we emulate x2apic in software */
398 entry->ecx |= F(X2APIC);
400 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
401 * may return different values. This forces us to get_cpu() before
402 * issuing the first command, and also to emulate this annoying behavior
403 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
405 int t, times = entry->eax & 0xff;
407 entry->flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
408 entry->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
409 for (t = 1; t < times; ++t) {
410 if (*nent >= maxnent)
413 do_cpuid_1_ent(&entry[t], function, 0);
414 entry[t].flags |= KVM_CPUID_FLAG_STATEFUL_FUNC;
419 /* function 4 has additional index. */
423 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
424 /* read more entries until cache_type is zero */
426 if (*nent >= maxnent)
429 cache_type = entry[i - 1].eax & 0x1f;
432 do_cpuid_1_ent(&entry[i], function, i);
434 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
439 case 6: /* Thermal management */
440 entry->eax = 0x4; /* allow ARAT */
446 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
447 /* Mask ebx against host capability word 9 */
449 entry->ebx &= kvm_supported_word9_x86_features;
450 cpuid_mask(&entry->ebx, 9);
451 // TSC_ADJUST is emulated
452 entry->ebx |= F(TSC_ADJUST);
453 entry->edx &= kvm_cpuid_7_0_edx_x86_features;
454 cpuid_mask(&entry->edx, CPUID_7_EDX);
455 if (boot_cpu_has(X86_FEATURE_IBPB) &&
456 boot_cpu_has(X86_FEATURE_IBRS))
457 entry->edx |= F(SPEC_CTRL);
458 if (boot_cpu_has(X86_FEATURE_STIBP))
459 entry->edx |= F(INTEL_STIBP);
460 if (boot_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) ||
461 boot_cpu_has(X86_FEATURE_AMD_SSBD))
462 entry->edx |= F(SPEC_CTRL_SSBD);
464 * We emulate ARCH_CAPABILITIES in software even
465 * if the host doesn't support it.
467 entry->edx |= F(ARCH_CAPABILITIES);
478 case 0xa: { /* Architectural Performance Monitoring */
479 struct x86_pmu_capability cap;
480 union cpuid10_eax eax;
481 union cpuid10_edx edx;
483 perf_get_x86_pmu_capability(&cap);
486 * Only support guest architectural pmu on a host
487 * with architectural pmu.
490 memset(&cap, 0, sizeof(cap));
492 eax.split.version_id = min(cap.version, 2);
493 eax.split.num_counters = cap.num_counters_gp;
494 eax.split.bit_width = cap.bit_width_gp;
495 eax.split.mask_length = cap.events_mask_len;
497 edx.split.num_counters_fixed = cap.num_counters_fixed;
498 edx.split.bit_width_fixed = cap.bit_width_fixed;
499 edx.split.reserved = 0;
501 entry->eax = eax.full;
502 entry->ebx = cap.events_mask;
504 entry->edx = edx.full;
507 /* function 0xb has additional index. */
511 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
512 /* read more entries until level_type is zero */
514 if (*nent >= maxnent)
517 level_type = entry[i - 1].ecx & 0xff00;
520 do_cpuid_1_ent(&entry[i], function, i);
522 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
529 u64 supported = kvm_supported_xcr0();
531 entry->eax &= supported;
532 entry->ebx = xstate_required_size(supported, false);
533 entry->ecx = entry->ebx;
534 entry->edx &= supported >> 32;
535 entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
539 for (idx = 1, i = 1; idx < 64; ++idx) {
540 u64 mask = ((u64)1 << idx);
541 if (*nent >= maxnent)
544 do_cpuid_1_ent(&entry[i], function, idx);
546 entry[i].eax &= kvm_supported_word10_x86_features;
547 cpuid_mask(&entry[i].eax, 10);
549 if (entry[i].eax & (F(XSAVES)|F(XSAVEC)))
551 xstate_required_size(supported,
554 if (entry[i].eax == 0 || !(supported & mask))
556 if (WARN_ON_ONCE(entry[i].ecx & 1))
562 KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
568 case KVM_CPUID_SIGNATURE: {
569 static const char signature[12] = "KVMKVMKVM\0\0";
570 const u32 *sigptr = (const u32 *)signature;
571 entry->eax = KVM_CPUID_FEATURES;
572 entry->ebx = sigptr[0];
573 entry->ecx = sigptr[1];
574 entry->edx = sigptr[2];
577 case KVM_CPUID_FEATURES:
578 entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
579 (1 << KVM_FEATURE_NOP_IO_DELAY) |
580 (1 << KVM_FEATURE_CLOCKSOURCE2) |
581 (1 << KVM_FEATURE_ASYNC_PF) |
582 (1 << KVM_FEATURE_PV_EOI) |
583 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
584 (1 << KVM_FEATURE_PV_UNHALT);
587 entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
594 entry->eax = min(entry->eax, 0x8000001a);
597 entry->edx &= kvm_supported_word1_x86_features;
598 cpuid_mask(&entry->edx, 1);
599 entry->ecx &= kvm_supported_word6_x86_features;
600 cpuid_mask(&entry->ecx, 6);
602 case 0x80000007: /* Advanced power management */
603 /* invariant TSC is CPUID.80000007H:EDX[8] */
604 entry->edx &= (1 << 8);
605 /* mask against host */
606 entry->edx &= boot_cpu_data.x86_power;
607 entry->eax = entry->ebx = entry->ecx = 0;
610 unsigned g_phys_as = (entry->eax >> 16) & 0xff;
611 unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
612 unsigned phys_as = entry->eax & 0xff;
615 * Use bare metal's MAXPHADDR if the CPU doesn't report guest
616 * MAXPHYADDR separately, or if TDP (NPT) is disabled, as the
617 * guest version "applies only to guests using nested paging".
619 if (!g_phys_as || !tdp_enabled)
622 entry->eax = g_phys_as | (virt_as << 8);
625 * IBRS, IBPB and VIRT_SSBD aren't necessarily present in
628 if (boot_cpu_has(X86_FEATURE_AMD_IBPB))
629 entry->ebx |= F(AMD_IBPB);
630 if (boot_cpu_has(X86_FEATURE_AMD_IBRS))
631 entry->ebx |= F(AMD_IBRS);
632 if (boot_cpu_has(X86_FEATURE_VIRT_SSBD))
633 entry->ebx |= F(VIRT_SSBD);
634 entry->ebx &= kvm_cpuid_8000_0008_ebx_x86_features;
635 cpuid_mask(&entry->ebx, CPUID_8000_0008_EBX);
637 * The preference is to use SPEC CTRL MSR instead of the
640 if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD) &&
641 !boot_cpu_has(X86_FEATURE_AMD_SSBD))
642 entry->ebx |= F(VIRT_SSBD);
646 entry->ecx = entry->edx = 0;
652 /*Add support for Centaur's CPUID instruction*/
654 /*Just support up to 0xC0000004 now*/
655 entry->eax = min(entry->eax, 0xC0000004);
658 entry->edx &= kvm_supported_word5_x86_features;
659 cpuid_mask(&entry->edx, 5);
661 case 3: /* Processor serial number */
662 case 5: /* MONITOR/MWAIT */
667 entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
671 kvm_x86_ops->set_supported_cpuid(function, entry);
681 static int do_cpuid_ent(struct kvm_cpuid_entry2 *entry, u32 func,
682 u32 idx, int *nent, int maxnent, unsigned int type)
684 if (*nent >= maxnent)
687 if (type == KVM_GET_EMULATED_CPUID)
688 return __do_cpuid_ent_emulated(entry, func, idx, nent, maxnent);
690 return __do_cpuid_ent(entry, func, idx, nent, maxnent);
695 struct kvm_cpuid_param {
699 bool (*qualifier)(const struct kvm_cpuid_param *param);
702 static bool is_centaur_cpu(const struct kvm_cpuid_param *param)
704 return boot_cpu_data.x86_vendor == X86_VENDOR_CENTAUR;
707 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
708 __u32 num_entries, unsigned int ioctl_type)
713 if (ioctl_type != KVM_GET_EMULATED_CPUID)
717 * We want to make sure that ->padding is being passed clean from
718 * userspace in case we want to use it for something in the future.
720 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
721 * have to give ourselves satisfied only with the emulated side. /me
724 for (i = 0; i < num_entries; i++) {
725 if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
728 if (pad[0] || pad[1] || pad[2])
734 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
735 struct kvm_cpuid_entry2 __user *entries,
738 struct kvm_cpuid_entry2 *cpuid_entries;
739 int limit, nent = 0, r = -E2BIG, i;
741 static const struct kvm_cpuid_param param[] = {
742 { .func = 0, .has_leaf_count = true },
743 { .func = 0x80000000, .has_leaf_count = true },
744 { .func = 0xC0000000, .qualifier = is_centaur_cpu, .has_leaf_count = true },
745 { .func = KVM_CPUID_SIGNATURE },
746 { .func = KVM_CPUID_FEATURES },
751 if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
752 cpuid->nent = KVM_MAX_CPUID_ENTRIES;
754 if (sanity_check_entries(entries, cpuid->nent, type))
758 cpuid_entries = vzalloc(sizeof(struct kvm_cpuid_entry2) * cpuid->nent);
763 for (i = 0; i < ARRAY_SIZE(param); i++) {
764 const struct kvm_cpuid_param *ent = ¶m[i];
766 if (ent->qualifier && !ent->qualifier(ent))
769 r = do_cpuid_ent(&cpuid_entries[nent], ent->func, ent->idx,
770 &nent, cpuid->nent, type);
775 if (!ent->has_leaf_count)
778 limit = cpuid_entries[nent - 1].eax;
779 for (func = ent->func + 1; func <= limit && nent < cpuid->nent && r == 0; ++func)
780 r = do_cpuid_ent(&cpuid_entries[nent], func, ent->idx,
781 &nent, cpuid->nent, type);
788 if (copy_to_user(entries, cpuid_entries,
789 nent * sizeof(struct kvm_cpuid_entry2)))
795 vfree(cpuid_entries);
800 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu *vcpu, int i)
802 struct kvm_cpuid_entry2 *e = &vcpu->arch.cpuid_entries[i];
803 struct kvm_cpuid_entry2 *ej;
805 int nent = vcpu->arch.cpuid_nent;
807 e->flags &= ~KVM_CPUID_FLAG_STATE_READ_NEXT;
808 /* when no next entry is found, the current entry[i] is reselected */
811 ej = &vcpu->arch.cpuid_entries[j];
812 } while (ej->function != e->function);
814 ej->flags |= KVM_CPUID_FLAG_STATE_READ_NEXT;
819 /* find an entry with matching function, matching index (if needed), and that
820 * should be read next (if it's stateful) */
821 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2 *e,
822 u32 function, u32 index)
824 if (e->function != function)
826 if ((e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) && e->index != index)
828 if ((e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC) &&
829 !(e->flags & KVM_CPUID_FLAG_STATE_READ_NEXT))
834 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
835 u32 function, u32 index)
838 struct kvm_cpuid_entry2 *best = NULL;
840 for (i = 0; i < vcpu->arch.cpuid_nent; ++i) {
841 struct kvm_cpuid_entry2 *e;
843 e = &vcpu->arch.cpuid_entries[i];
844 if (is_matching_cpuid_entry(e, function, index)) {
845 if (e->flags & KVM_CPUID_FLAG_STATEFUL_FUNC)
846 move_to_next_stateful_cpuid_entry(vcpu, i);
853 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
856 * If no match is found, check whether we exceed the vCPU's limit
857 * and return the content of the highest valid _standard_ leaf instead.
858 * This is to satisfy the CPUID specification.
860 static struct kvm_cpuid_entry2* check_cpuid_limit(struct kvm_vcpu *vcpu,
861 u32 function, u32 index)
863 struct kvm_cpuid_entry2 *maxlevel;
865 maxlevel = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
866 if (!maxlevel || maxlevel->eax >= function)
868 if (function & 0x80000000) {
869 maxlevel = kvm_find_cpuid_entry(vcpu, 0, 0);
873 return kvm_find_cpuid_entry(vcpu, maxlevel->eax, index);
876 void kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx, u32 *ecx, u32 *edx)
878 u32 function = *eax, index = *ecx;
879 struct kvm_cpuid_entry2 *best;
881 best = kvm_find_cpuid_entry(vcpu, function, index);
884 best = check_cpuid_limit(vcpu, function, index);
892 *eax = *ebx = *ecx = *edx = 0;
893 trace_kvm_cpuid(function, *eax, *ebx, *ecx, *edx);
895 EXPORT_SYMBOL_GPL(kvm_cpuid);
897 void kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
899 u32 function, eax, ebx, ecx, edx;
901 function = eax = kvm_register_read(vcpu, VCPU_REGS_RAX);
902 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
903 kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx);
904 kvm_register_write(vcpu, VCPU_REGS_RAX, eax);
905 kvm_register_write(vcpu, VCPU_REGS_RBX, ebx);
906 kvm_register_write(vcpu, VCPU_REGS_RCX, ecx);
907 kvm_register_write(vcpu, VCPU_REGS_RDX, edx);
908 kvm_x86_ops->skip_emulated_instruction(vcpu);
910 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);