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>
19 #include <asm/fpu/internal.h> /* For use_eager_fpu. Ugh! */
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 int kvm_update_cpuid(struct kvm_vcpu
*vcpu
)
70 struct kvm_cpuid_entry2
*best
;
71 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
73 best
= kvm_find_cpuid_entry(vcpu
, 1, 0);
77 /* Update OSXSAVE bit */
78 if (cpu_has_xsave
&& best
->function
== 0x1) {
79 best
->ecx
&= ~F(OSXSAVE
);
80 if (kvm_read_cr4_bits(vcpu
, X86_CR4_OSXSAVE
))
81 best
->ecx
|= F(OSXSAVE
);
85 if (best
->ecx
& F(TSC_DEADLINE_TIMER
))
86 apic
->lapic_timer
.timer_mode_mask
= 3 << 17;
88 apic
->lapic_timer
.timer_mode_mask
= 1 << 17;
91 best
= kvm_find_cpuid_entry(vcpu
, 0xD, 0);
93 vcpu
->arch
.guest_supported_xcr0
= 0;
94 vcpu
->arch
.guest_xstate_size
= XSAVE_HDR_SIZE
+ XSAVE_HDR_OFFSET
;
96 vcpu
->arch
.guest_supported_xcr0
=
97 (best
->eax
| ((u64
)best
->edx
<< 32)) &
99 vcpu
->arch
.guest_xstate_size
= best
->ebx
=
100 xstate_required_size(vcpu
->arch
.xcr0
, false);
103 best
= kvm_find_cpuid_entry(vcpu
, 0xD, 1);
104 if (best
&& (best
->eax
& (F(XSAVES
) | F(XSAVEC
))))
105 best
->ebx
= xstate_required_size(vcpu
->arch
.xcr0
, true);
107 vcpu
->arch
.eager_fpu
= use_eager_fpu();
108 if (vcpu
->arch
.eager_fpu
)
109 kvm_x86_ops
->fpu_activate(vcpu
);
112 * The existing code assumes virtual address is 48-bit in the canonical
113 * address checks; exit if it is ever changed.
115 best
= kvm_find_cpuid_entry(vcpu
, 0x80000008, 0);
116 if (best
&& ((best
->eax
& 0xff00) >> 8) != 48 &&
117 ((best
->eax
& 0xff00) >> 8) != 0)
120 /* Update physical-address width */
121 vcpu
->arch
.maxphyaddr
= cpuid_query_maxphyaddr(vcpu
);
123 kvm_pmu_refresh(vcpu
);
127 static int is_efer_nx(void)
129 unsigned long long efer
= 0;
131 rdmsrl_safe(MSR_EFER
, &efer
);
132 return efer
& EFER_NX
;
135 static void cpuid_fix_nx_cap(struct kvm_vcpu
*vcpu
)
138 struct kvm_cpuid_entry2
*e
, *entry
;
141 for (i
= 0; i
< vcpu
->arch
.cpuid_nent
; ++i
) {
142 e
= &vcpu
->arch
.cpuid_entries
[i
];
143 if (e
->function
== 0x80000001) {
148 if (entry
&& (entry
->edx
& F(NX
)) && !is_efer_nx()) {
149 entry
->edx
&= ~F(NX
);
150 printk(KERN_INFO
"kvm: guest NX capability removed\n");
154 int cpuid_query_maxphyaddr(struct kvm_vcpu
*vcpu
)
156 struct kvm_cpuid_entry2
*best
;
158 best
= kvm_find_cpuid_entry(vcpu
, 0x80000000, 0);
159 if (!best
|| best
->eax
< 0x80000008)
161 best
= kvm_find_cpuid_entry(vcpu
, 0x80000008, 0);
163 return best
->eax
& 0xff;
167 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr
);
169 /* when an old userspace process fills a new kernel module */
170 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu
*vcpu
,
171 struct kvm_cpuid
*cpuid
,
172 struct kvm_cpuid_entry __user
*entries
)
175 struct kvm_cpuid_entry
*cpuid_entries
;
178 if (cpuid
->nent
> KVM_MAX_CPUID_ENTRIES
)
181 cpuid_entries
= vmalloc(sizeof(struct kvm_cpuid_entry
) * cpuid
->nent
);
185 if (copy_from_user(cpuid_entries
, entries
,
186 cpuid
->nent
* sizeof(struct kvm_cpuid_entry
)))
188 for (i
= 0; i
< cpuid
->nent
; i
++) {
189 vcpu
->arch
.cpuid_entries
[i
].function
= cpuid_entries
[i
].function
;
190 vcpu
->arch
.cpuid_entries
[i
].eax
= cpuid_entries
[i
].eax
;
191 vcpu
->arch
.cpuid_entries
[i
].ebx
= cpuid_entries
[i
].ebx
;
192 vcpu
->arch
.cpuid_entries
[i
].ecx
= cpuid_entries
[i
].ecx
;
193 vcpu
->arch
.cpuid_entries
[i
].edx
= cpuid_entries
[i
].edx
;
194 vcpu
->arch
.cpuid_entries
[i
].index
= 0;
195 vcpu
->arch
.cpuid_entries
[i
].flags
= 0;
196 vcpu
->arch
.cpuid_entries
[i
].padding
[0] = 0;
197 vcpu
->arch
.cpuid_entries
[i
].padding
[1] = 0;
198 vcpu
->arch
.cpuid_entries
[i
].padding
[2] = 0;
200 vcpu
->arch
.cpuid_nent
= cpuid
->nent
;
201 cpuid_fix_nx_cap(vcpu
);
202 kvm_apic_set_version(vcpu
);
203 kvm_x86_ops
->cpuid_update(vcpu
);
204 r
= kvm_update_cpuid(vcpu
);
207 vfree(cpuid_entries
);
212 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu
*vcpu
,
213 struct kvm_cpuid2
*cpuid
,
214 struct kvm_cpuid_entry2 __user
*entries
)
219 if (cpuid
->nent
> KVM_MAX_CPUID_ENTRIES
)
222 if (copy_from_user(&vcpu
->arch
.cpuid_entries
, entries
,
223 cpuid
->nent
* sizeof(struct kvm_cpuid_entry2
)))
225 vcpu
->arch
.cpuid_nent
= cpuid
->nent
;
226 kvm_apic_set_version(vcpu
);
227 kvm_x86_ops
->cpuid_update(vcpu
);
228 r
= kvm_update_cpuid(vcpu
);
233 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu
*vcpu
,
234 struct kvm_cpuid2
*cpuid
,
235 struct kvm_cpuid_entry2 __user
*entries
)
240 if (cpuid
->nent
< vcpu
->arch
.cpuid_nent
)
243 if (copy_to_user(entries
, &vcpu
->arch
.cpuid_entries
,
244 vcpu
->arch
.cpuid_nent
* sizeof(struct kvm_cpuid_entry2
)))
249 cpuid
->nent
= vcpu
->arch
.cpuid_nent
;
253 static void cpuid_mask(u32
*word
, int wordnum
)
255 *word
&= boot_cpu_data
.x86_capability
[wordnum
];
258 static void do_cpuid_1_ent(struct kvm_cpuid_entry2
*entry
, u32 function
,
261 entry
->function
= function
;
262 entry
->index
= index
;
263 cpuid_count(entry
->function
, entry
->index
,
264 &entry
->eax
, &entry
->ebx
, &entry
->ecx
, &entry
->edx
);
268 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2
*entry
,
269 u32 func
, u32 index
, int *nent
, int maxnent
)
273 entry
->eax
= 1; /* only one leaf currently */
277 entry
->ecx
= F(MOVBE
);
284 entry
->function
= func
;
285 entry
->index
= index
;
290 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2
*entry
, u32 function
,
291 u32 index
, int *nent
, int maxnent
)
294 unsigned f_nx
= is_efer_nx() ? F(NX
) : 0;
296 unsigned f_gbpages
= (kvm_x86_ops
->get_lpage_level() == PT_PDPE_LEVEL
)
298 unsigned f_lm
= F(LM
);
300 unsigned f_gbpages
= 0;
303 unsigned f_rdtscp
= kvm_x86_ops
->rdtscp_supported() ? F(RDTSCP
) : 0;
304 unsigned f_invpcid
= kvm_x86_ops
->invpcid_supported() ? F(INVPCID
) : 0;
305 unsigned f_mpx
= kvm_mpx_supported() ? F(MPX
) : 0;
306 unsigned f_xsaves
= kvm_x86_ops
->xsaves_supported() ? F(XSAVES
) : 0;
309 const u32 kvm_supported_word0_x86_features
=
310 F(FPU
) | F(VME
) | F(DE
) | F(PSE
) |
311 F(TSC
) | F(MSR
) | F(PAE
) | F(MCE
) |
312 F(CX8
) | F(APIC
) | 0 /* Reserved */ | F(SEP
) |
313 F(MTRR
) | F(PGE
) | F(MCA
) | F(CMOV
) |
314 F(PAT
) | F(PSE36
) | 0 /* PSN */ | F(CLFLUSH
) |
315 0 /* Reserved, DS, ACPI */ | F(MMX
) |
316 F(FXSR
) | F(XMM
) | F(XMM2
) | F(SELFSNOOP
) |
317 0 /* HTT, TM, Reserved, PBE */;
318 /* cpuid 0x80000001.edx */
319 const u32 kvm_supported_word1_x86_features
=
320 F(FPU
) | F(VME
) | F(DE
) | F(PSE
) |
321 F(TSC
) | F(MSR
) | F(PAE
) | F(MCE
) |
322 F(CX8
) | F(APIC
) | 0 /* Reserved */ | F(SYSCALL
) |
323 F(MTRR
) | F(PGE
) | F(MCA
) | F(CMOV
) |
324 F(PAT
) | F(PSE36
) | 0 /* Reserved */ |
325 f_nx
| 0 /* Reserved */ | F(MMXEXT
) | F(MMX
) |
326 F(FXSR
) | F(FXSR_OPT
) | f_gbpages
| f_rdtscp
|
327 0 /* Reserved */ | f_lm
| F(3DNOWEXT
) | F(3DNOW
);
329 const u32 kvm_supported_word4_x86_features
=
330 /* NOTE: MONITOR (and MWAIT) are emulated as NOP,
331 * but *not* advertised to guests via CPUID ! */
332 F(XMM3
) | F(PCLMULQDQ
) | 0 /* DTES64, MONITOR */ |
333 0 /* DS-CPL, VMX, SMX, EST */ |
334 0 /* TM2 */ | F(SSSE3
) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
335 F(FMA
) | F(CX16
) | 0 /* xTPR Update, PDCM */ |
336 F(PCID
) | 0 /* Reserved, DCA */ | F(XMM4_1
) |
337 F(XMM4_2
) | F(X2APIC
) | F(MOVBE
) | F(POPCNT
) |
338 0 /* Reserved*/ | F(AES
) | F(XSAVE
) | 0 /* OSXSAVE */ | F(AVX
) |
340 /* cpuid 0x80000001.ecx */
341 const u32 kvm_supported_word6_x86_features
=
342 F(LAHF_LM
) | F(CMP_LEGACY
) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
343 F(CR8_LEGACY
) | F(ABM
) | F(SSE4A
) | F(MISALIGNSSE
) |
344 F(3DNOWPREFETCH
) | F(OSVW
) | 0 /* IBS */ | F(XOP
) |
345 0 /* SKINIT, WDT, LWP */ | F(FMA4
) | F(TBM
);
347 /* cpuid 0xC0000001.edx */
348 const u32 kvm_supported_word5_x86_features
=
349 F(XSTORE
) | F(XSTORE_EN
) | F(XCRYPT
) | F(XCRYPT_EN
) |
350 F(ACE2
) | F(ACE2_EN
) | F(PHE
) | F(PHE_EN
) |
354 const u32 kvm_supported_word9_x86_features
=
355 F(FSGSBASE
) | F(BMI1
) | F(HLE
) | F(AVX2
) | F(SMEP
) |
356 F(BMI2
) | F(ERMS
) | f_invpcid
| F(RTM
) | f_mpx
| F(RDSEED
) |
357 F(ADX
) | F(SMAP
) | F(AVX512F
) | F(AVX512PF
) | F(AVX512ER
) |
358 F(AVX512CD
) | F(CLFLUSHOPT
) | F(CLWB
) | F(PCOMMIT
);
360 /* cpuid 0xD.1.eax */
361 const u32 kvm_supported_word10_x86_features
=
362 F(XSAVEOPT
) | F(XSAVEC
) | F(XGETBV1
) | f_xsaves
;
364 /* all calls to cpuid_count() should be made on the same cpu */
369 if (*nent
>= maxnent
)
372 do_cpuid_1_ent(entry
, function
, index
);
377 entry
->eax
= min(entry
->eax
, (u32
)0xd);
380 entry
->edx
&= kvm_supported_word0_x86_features
;
381 cpuid_mask(&entry
->edx
, 0);
382 entry
->ecx
&= kvm_supported_word4_x86_features
;
383 cpuid_mask(&entry
->ecx
, 4);
384 /* we support x2apic emulation even if host does not support
385 * it since we emulate x2apic in software */
386 entry
->ecx
|= F(X2APIC
);
388 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
389 * may return different values. This forces us to get_cpu() before
390 * issuing the first command, and also to emulate this annoying behavior
391 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
393 int t
, times
= entry
->eax
& 0xff;
395 entry
->flags
|= KVM_CPUID_FLAG_STATEFUL_FUNC
;
396 entry
->flags
|= KVM_CPUID_FLAG_STATE_READ_NEXT
;
397 for (t
= 1; t
< times
; ++t
) {
398 if (*nent
>= maxnent
)
401 do_cpuid_1_ent(&entry
[t
], function
, 0);
402 entry
[t
].flags
|= KVM_CPUID_FLAG_STATEFUL_FUNC
;
407 /* function 4 has additional index. */
411 entry
->flags
|= KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
412 /* read more entries until cache_type is zero */
414 if (*nent
>= maxnent
)
417 cache_type
= entry
[i
- 1].eax
& 0x1f;
420 do_cpuid_1_ent(&entry
[i
], function
, i
);
422 KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
427 case 6: /* Thermal management */
428 entry
->eax
= 0x4; /* allow ARAT */
434 entry
->flags
|= KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
435 /* Mask ebx against host capability word 9 */
437 entry
->ebx
&= kvm_supported_word9_x86_features
;
438 cpuid_mask(&entry
->ebx
, 9);
439 // TSC_ADJUST is emulated
440 entry
->ebx
|= F(TSC_ADJUST
);
450 case 0xa: { /* Architectural Performance Monitoring */
451 struct x86_pmu_capability cap
;
452 union cpuid10_eax eax
;
453 union cpuid10_edx edx
;
455 perf_get_x86_pmu_capability(&cap
);
458 * Only support guest architectural pmu on a host
459 * with architectural pmu.
462 memset(&cap
, 0, sizeof(cap
));
464 eax
.split
.version_id
= min(cap
.version
, 2);
465 eax
.split
.num_counters
= cap
.num_counters_gp
;
466 eax
.split
.bit_width
= cap
.bit_width_gp
;
467 eax
.split
.mask_length
= cap
.events_mask_len
;
469 edx
.split
.num_counters_fixed
= cap
.num_counters_fixed
;
470 edx
.split
.bit_width_fixed
= cap
.bit_width_fixed
;
471 edx
.split
.reserved
= 0;
473 entry
->eax
= eax
.full
;
474 entry
->ebx
= cap
.events_mask
;
476 entry
->edx
= edx
.full
;
479 /* function 0xb has additional index. */
483 entry
->flags
|= KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
484 /* read more entries until level_type is zero */
486 if (*nent
>= maxnent
)
489 level_type
= entry
[i
- 1].ecx
& 0xff00;
492 do_cpuid_1_ent(&entry
[i
], function
, i
);
494 KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
501 u64 supported
= kvm_supported_xcr0();
503 entry
->eax
&= supported
;
504 entry
->ebx
= xstate_required_size(supported
, false);
505 entry
->ecx
= entry
->ebx
;
506 entry
->edx
&= supported
>> 32;
507 entry
->flags
|= KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
511 for (idx
= 1, i
= 1; idx
< 64; ++idx
) {
512 u64 mask
= ((u64
)1 << idx
);
513 if (*nent
>= maxnent
)
516 do_cpuid_1_ent(&entry
[i
], function
, idx
);
518 entry
[i
].eax
&= kvm_supported_word10_x86_features
;
520 if (entry
[i
].eax
& (F(XSAVES
)|F(XSAVEC
)))
522 xstate_required_size(supported
,
525 if (entry
[i
].eax
== 0 || !(supported
& mask
))
527 if (WARN_ON_ONCE(entry
[i
].ecx
& 1))
533 KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
539 case KVM_CPUID_SIGNATURE
: {
540 static const char signature
[12] = "KVMKVMKVM\0\0";
541 const u32
*sigptr
= (const u32
*)signature
;
542 entry
->eax
= KVM_CPUID_FEATURES
;
543 entry
->ebx
= sigptr
[0];
544 entry
->ecx
= sigptr
[1];
545 entry
->edx
= sigptr
[2];
548 case KVM_CPUID_FEATURES
:
549 entry
->eax
= (1 << KVM_FEATURE_CLOCKSOURCE
) |
550 (1 << KVM_FEATURE_NOP_IO_DELAY
) |
551 (1 << KVM_FEATURE_CLOCKSOURCE2
) |
552 (1 << KVM_FEATURE_ASYNC_PF
) |
553 (1 << KVM_FEATURE_PV_EOI
) |
554 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT
) |
555 (1 << KVM_FEATURE_PV_UNHALT
);
558 entry
->eax
|= (1 << KVM_FEATURE_STEAL_TIME
);
565 entry
->eax
= min(entry
->eax
, 0x8000001a);
568 entry
->edx
&= kvm_supported_word1_x86_features
;
569 cpuid_mask(&entry
->edx
, 1);
570 entry
->ecx
&= kvm_supported_word6_x86_features
;
571 cpuid_mask(&entry
->ecx
, 6);
573 case 0x80000007: /* Advanced power management */
574 /* invariant TSC is CPUID.80000007H:EDX[8] */
575 entry
->edx
&= (1 << 8);
576 /* mask against host */
577 entry
->edx
&= boot_cpu_data
.x86_power
;
578 entry
->eax
= entry
->ebx
= entry
->ecx
= 0;
581 unsigned g_phys_as
= (entry
->eax
>> 16) & 0xff;
582 unsigned virt_as
= max((entry
->eax
>> 8) & 0xff, 48U);
583 unsigned phys_as
= entry
->eax
& 0xff;
587 entry
->eax
= g_phys_as
| (virt_as
<< 8);
588 entry
->ebx
= entry
->edx
= 0;
592 entry
->ecx
= entry
->edx
= 0;
598 /*Add support for Centaur's CPUID instruction*/
600 /*Just support up to 0xC0000004 now*/
601 entry
->eax
= min(entry
->eax
, 0xC0000004);
604 entry
->edx
&= kvm_supported_word5_x86_features
;
605 cpuid_mask(&entry
->edx
, 5);
607 case 3: /* Processor serial number */
608 case 5: /* MONITOR/MWAIT */
613 entry
->eax
= entry
->ebx
= entry
->ecx
= entry
->edx
= 0;
617 kvm_x86_ops
->set_supported_cpuid(function
, entry
);
627 static int do_cpuid_ent(struct kvm_cpuid_entry2
*entry
, u32 func
,
628 u32 idx
, int *nent
, int maxnent
, unsigned int type
)
630 if (type
== KVM_GET_EMULATED_CPUID
)
631 return __do_cpuid_ent_emulated(entry
, func
, idx
, nent
, maxnent
);
633 return __do_cpuid_ent(entry
, func
, idx
, nent
, maxnent
);
638 struct kvm_cpuid_param
{
642 bool (*qualifier
)(const struct kvm_cpuid_param
*param
);
645 static bool is_centaur_cpu(const struct kvm_cpuid_param
*param
)
647 return boot_cpu_data
.x86_vendor
== X86_VENDOR_CENTAUR
;
650 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user
*entries
,
651 __u32 num_entries
, unsigned int ioctl_type
)
656 if (ioctl_type
!= KVM_GET_EMULATED_CPUID
)
660 * We want to make sure that ->padding is being passed clean from
661 * userspace in case we want to use it for something in the future.
663 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
664 * have to give ourselves satisfied only with the emulated side. /me
667 for (i
= 0; i
< num_entries
; i
++) {
668 if (copy_from_user(pad
, entries
[i
].padding
, sizeof(pad
)))
671 if (pad
[0] || pad
[1] || pad
[2])
677 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2
*cpuid
,
678 struct kvm_cpuid_entry2 __user
*entries
,
681 struct kvm_cpuid_entry2
*cpuid_entries
;
682 int limit
, nent
= 0, r
= -E2BIG
, i
;
684 static const struct kvm_cpuid_param param
[] = {
685 { .func
= 0, .has_leaf_count
= true },
686 { .func
= 0x80000000, .has_leaf_count
= true },
687 { .func
= 0xC0000000, .qualifier
= is_centaur_cpu
, .has_leaf_count
= true },
688 { .func
= KVM_CPUID_SIGNATURE
},
689 { .func
= KVM_CPUID_FEATURES
},
694 if (cpuid
->nent
> KVM_MAX_CPUID_ENTRIES
)
695 cpuid
->nent
= KVM_MAX_CPUID_ENTRIES
;
697 if (sanity_check_entries(entries
, cpuid
->nent
, type
))
701 cpuid_entries
= vzalloc(sizeof(struct kvm_cpuid_entry2
) * cpuid
->nent
);
706 for (i
= 0; i
< ARRAY_SIZE(param
); i
++) {
707 const struct kvm_cpuid_param
*ent
= ¶m
[i
];
709 if (ent
->qualifier
&& !ent
->qualifier(ent
))
712 r
= do_cpuid_ent(&cpuid_entries
[nent
], ent
->func
, ent
->idx
,
713 &nent
, cpuid
->nent
, type
);
718 if (!ent
->has_leaf_count
)
721 limit
= cpuid_entries
[nent
- 1].eax
;
722 for (func
= ent
->func
+ 1; func
<= limit
&& nent
< cpuid
->nent
&& r
== 0; ++func
)
723 r
= do_cpuid_ent(&cpuid_entries
[nent
], func
, ent
->idx
,
724 &nent
, cpuid
->nent
, type
);
731 if (copy_to_user(entries
, cpuid_entries
,
732 nent
* sizeof(struct kvm_cpuid_entry2
)))
738 vfree(cpuid_entries
);
743 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu
*vcpu
, int i
)
745 struct kvm_cpuid_entry2
*e
= &vcpu
->arch
.cpuid_entries
[i
];
746 int j
, nent
= vcpu
->arch
.cpuid_nent
;
748 e
->flags
&= ~KVM_CPUID_FLAG_STATE_READ_NEXT
;
749 /* when no next entry is found, the current entry[i] is reselected */
750 for (j
= i
+ 1; ; j
= (j
+ 1) % nent
) {
751 struct kvm_cpuid_entry2
*ej
= &vcpu
->arch
.cpuid_entries
[j
];
752 if (ej
->function
== e
->function
) {
753 ej
->flags
|= KVM_CPUID_FLAG_STATE_READ_NEXT
;
757 return 0; /* silence gcc, even though control never reaches here */
760 /* find an entry with matching function, matching index (if needed), and that
761 * should be read next (if it's stateful) */
762 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2
*e
,
763 u32 function
, u32 index
)
765 if (e
->function
!= function
)
767 if ((e
->flags
& KVM_CPUID_FLAG_SIGNIFCANT_INDEX
) && e
->index
!= index
)
769 if ((e
->flags
& KVM_CPUID_FLAG_STATEFUL_FUNC
) &&
770 !(e
->flags
& KVM_CPUID_FLAG_STATE_READ_NEXT
))
775 struct kvm_cpuid_entry2
*kvm_find_cpuid_entry(struct kvm_vcpu
*vcpu
,
776 u32 function
, u32 index
)
779 struct kvm_cpuid_entry2
*best
= NULL
;
781 for (i
= 0; i
< vcpu
->arch
.cpuid_nent
; ++i
) {
782 struct kvm_cpuid_entry2
*e
;
784 e
= &vcpu
->arch
.cpuid_entries
[i
];
785 if (is_matching_cpuid_entry(e
, function
, index
)) {
786 if (e
->flags
& KVM_CPUID_FLAG_STATEFUL_FUNC
)
787 move_to_next_stateful_cpuid_entry(vcpu
, i
);
794 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry
);
797 * If no match is found, check whether we exceed the vCPU's limit
798 * and return the content of the highest valid _standard_ leaf instead.
799 * This is to satisfy the CPUID specification.
801 static struct kvm_cpuid_entry2
* check_cpuid_limit(struct kvm_vcpu
*vcpu
,
802 u32 function
, u32 index
)
804 struct kvm_cpuid_entry2
*maxlevel
;
806 maxlevel
= kvm_find_cpuid_entry(vcpu
, function
& 0x80000000, 0);
807 if (!maxlevel
|| maxlevel
->eax
>= function
)
809 if (function
& 0x80000000) {
810 maxlevel
= kvm_find_cpuid_entry(vcpu
, 0, 0);
814 return kvm_find_cpuid_entry(vcpu
, maxlevel
->eax
, index
);
817 void kvm_cpuid(struct kvm_vcpu
*vcpu
, u32
*eax
, u32
*ebx
, u32
*ecx
, u32
*edx
)
819 u32 function
= *eax
, index
= *ecx
;
820 struct kvm_cpuid_entry2
*best
;
822 best
= kvm_find_cpuid_entry(vcpu
, function
, index
);
825 best
= check_cpuid_limit(vcpu
, function
, index
);
828 * Perfmon not yet supported for L2 guest.
830 if (is_guest_mode(vcpu
) && function
== 0xa)
839 *eax
= *ebx
= *ecx
= *edx
= 0;
840 trace_kvm_cpuid(function
, *eax
, *ebx
, *ecx
, *edx
);
842 EXPORT_SYMBOL_GPL(kvm_cpuid
);
844 void kvm_emulate_cpuid(struct kvm_vcpu
*vcpu
)
846 u32 function
, eax
, ebx
, ecx
, edx
;
848 function
= eax
= kvm_register_read(vcpu
, VCPU_REGS_RAX
);
849 ecx
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
850 kvm_cpuid(vcpu
, &eax
, &ebx
, &ecx
, &edx
);
851 kvm_register_write(vcpu
, VCPU_REGS_RAX
, eax
);
852 kvm_register_write(vcpu
, VCPU_REGS_RBX
, ebx
);
853 kvm_register_write(vcpu
, VCPU_REGS_RCX
, ecx
);
854 kvm_register_write(vcpu
, VCPU_REGS_RDX
, edx
);
855 kvm_x86_ops
->skip_emulated_instruction(vcpu
);
857 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid
);