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 int kvm_update_cpuid(struct kvm_vcpu
*vcpu
)
72 struct kvm_cpuid_entry2
*best
;
73 struct kvm_lapic
*apic
= vcpu
->arch
.apic
;
75 best
= kvm_find_cpuid_entry(vcpu
, 1, 0);
79 /* Update OSXSAVE bit */
80 if (boot_cpu_has(X86_FEATURE_XSAVE
) && best
->function
== 0x1) {
81 best
->ecx
&= ~F(OSXSAVE
);
82 if (kvm_read_cr4_bits(vcpu
, X86_CR4_OSXSAVE
))
83 best
->ecx
|= F(OSXSAVE
);
86 best
->edx
&= ~F(APIC
);
87 if (vcpu
->arch
.apic_base
& MSR_IA32_APICBASE_ENABLE
)
91 if (best
->ecx
& F(TSC_DEADLINE_TIMER
))
92 apic
->lapic_timer
.timer_mode_mask
= 3 << 17;
94 apic
->lapic_timer
.timer_mode_mask
= 1 << 17;
97 best
= kvm_find_cpuid_entry(vcpu
, 7, 0);
99 /* Update OSPKE bit */
100 if (boot_cpu_has(X86_FEATURE_PKU
) && best
->function
== 0x7) {
101 best
->ecx
&= ~F(OSPKE
);
102 if (kvm_read_cr4_bits(vcpu
, X86_CR4_PKE
))
103 best
->ecx
|= F(OSPKE
);
107 best
= kvm_find_cpuid_entry(vcpu
, 0xD, 0);
109 vcpu
->arch
.guest_supported_xcr0
= 0;
110 vcpu
->arch
.guest_xstate_size
= XSAVE_HDR_SIZE
+ XSAVE_HDR_OFFSET
;
112 vcpu
->arch
.guest_supported_xcr0
=
113 (best
->eax
| ((u64
)best
->edx
<< 32)) &
114 kvm_supported_xcr0();
115 vcpu
->arch
.guest_xstate_size
= best
->ebx
=
116 xstate_required_size(vcpu
->arch
.xcr0
, false);
119 best
= kvm_find_cpuid_entry(vcpu
, 0xD, 1);
120 if (best
&& (best
->eax
& (F(XSAVES
) | F(XSAVEC
))))
121 best
->ebx
= xstate_required_size(vcpu
->arch
.xcr0
, true);
124 * The existing code assumes virtual address is 48-bit or 57-bit in the
125 * canonical address checks; exit if it is ever changed.
127 best
= kvm_find_cpuid_entry(vcpu
, 0x80000008, 0);
129 int vaddr_bits
= (best
->eax
& 0xff00) >> 8;
131 if (vaddr_bits
!= 48 && vaddr_bits
!= 57 && vaddr_bits
!= 0)
135 best
= kvm_find_cpuid_entry(vcpu
, KVM_CPUID_FEATURES
, 0);
136 if (kvm_hlt_in_guest(vcpu
->kvm
) && best
&&
137 (best
->eax
& (1 << KVM_FEATURE_PV_UNHALT
)))
138 best
->eax
&= ~(1 << KVM_FEATURE_PV_UNHALT
);
140 /* Update physical-address width */
141 vcpu
->arch
.maxphyaddr
= cpuid_query_maxphyaddr(vcpu
);
142 kvm_mmu_reset_context(vcpu
);
144 kvm_pmu_refresh(vcpu
);
148 static int is_efer_nx(void)
150 unsigned long long efer
= 0;
152 rdmsrl_safe(MSR_EFER
, &efer
);
153 return efer
& EFER_NX
;
156 static void cpuid_fix_nx_cap(struct kvm_vcpu
*vcpu
)
159 struct kvm_cpuid_entry2
*e
, *entry
;
162 for (i
= 0; i
< vcpu
->arch
.cpuid_nent
; ++i
) {
163 e
= &vcpu
->arch
.cpuid_entries
[i
];
164 if (e
->function
== 0x80000001) {
169 if (entry
&& (entry
->edx
& F(NX
)) && !is_efer_nx()) {
170 entry
->edx
&= ~F(NX
);
171 printk(KERN_INFO
"kvm: guest NX capability removed\n");
175 int cpuid_query_maxphyaddr(struct kvm_vcpu
*vcpu
)
177 struct kvm_cpuid_entry2
*best
;
179 best
= kvm_find_cpuid_entry(vcpu
, 0x80000000, 0);
180 if (!best
|| best
->eax
< 0x80000008)
182 best
= kvm_find_cpuid_entry(vcpu
, 0x80000008, 0);
184 return best
->eax
& 0xff;
188 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr
);
190 /* when an old userspace process fills a new kernel module */
191 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu
*vcpu
,
192 struct kvm_cpuid
*cpuid
,
193 struct kvm_cpuid_entry __user
*entries
)
196 struct kvm_cpuid_entry
*cpuid_entries
= NULL
;
199 if (cpuid
->nent
> KVM_MAX_CPUID_ENTRIES
)
204 vmalloc(array_size(sizeof(struct kvm_cpuid_entry
),
209 if (copy_from_user(cpuid_entries
, entries
,
210 cpuid
->nent
* sizeof(struct kvm_cpuid_entry
)))
213 for (i
= 0; i
< cpuid
->nent
; i
++) {
214 vcpu
->arch
.cpuid_entries
[i
].function
= cpuid_entries
[i
].function
;
215 vcpu
->arch
.cpuid_entries
[i
].eax
= cpuid_entries
[i
].eax
;
216 vcpu
->arch
.cpuid_entries
[i
].ebx
= cpuid_entries
[i
].ebx
;
217 vcpu
->arch
.cpuid_entries
[i
].ecx
= cpuid_entries
[i
].ecx
;
218 vcpu
->arch
.cpuid_entries
[i
].edx
= cpuid_entries
[i
].edx
;
219 vcpu
->arch
.cpuid_entries
[i
].index
= 0;
220 vcpu
->arch
.cpuid_entries
[i
].flags
= 0;
221 vcpu
->arch
.cpuid_entries
[i
].padding
[0] = 0;
222 vcpu
->arch
.cpuid_entries
[i
].padding
[1] = 0;
223 vcpu
->arch
.cpuid_entries
[i
].padding
[2] = 0;
225 vcpu
->arch
.cpuid_nent
= cpuid
->nent
;
226 cpuid_fix_nx_cap(vcpu
);
227 kvm_apic_set_version(vcpu
);
228 kvm_x86_ops
->cpuid_update(vcpu
);
229 r
= kvm_update_cpuid(vcpu
);
232 vfree(cpuid_entries
);
236 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu
*vcpu
,
237 struct kvm_cpuid2
*cpuid
,
238 struct kvm_cpuid_entry2 __user
*entries
)
243 if (cpuid
->nent
> KVM_MAX_CPUID_ENTRIES
)
246 if (copy_from_user(&vcpu
->arch
.cpuid_entries
, entries
,
247 cpuid
->nent
* sizeof(struct kvm_cpuid_entry2
)))
249 vcpu
->arch
.cpuid_nent
= cpuid
->nent
;
250 kvm_apic_set_version(vcpu
);
251 kvm_x86_ops
->cpuid_update(vcpu
);
252 r
= kvm_update_cpuid(vcpu
);
257 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu
*vcpu
,
258 struct kvm_cpuid2
*cpuid
,
259 struct kvm_cpuid_entry2 __user
*entries
)
264 if (cpuid
->nent
< vcpu
->arch
.cpuid_nent
)
267 if (copy_to_user(entries
, &vcpu
->arch
.cpuid_entries
,
268 vcpu
->arch
.cpuid_nent
* sizeof(struct kvm_cpuid_entry2
)))
273 cpuid
->nent
= vcpu
->arch
.cpuid_nent
;
277 static void cpuid_mask(u32
*word
, int wordnum
)
279 *word
&= boot_cpu_data
.x86_capability
[wordnum
];
282 static void do_cpuid_1_ent(struct kvm_cpuid_entry2
*entry
, u32 function
,
285 entry
->function
= function
;
286 entry
->index
= index
;
287 cpuid_count(entry
->function
, entry
->index
,
288 &entry
->eax
, &entry
->ebx
, &entry
->ecx
, &entry
->edx
);
292 static int __do_cpuid_ent_emulated(struct kvm_cpuid_entry2
*entry
,
293 u32 func
, u32 index
, int *nent
, int maxnent
)
301 entry
->ecx
= F(MOVBE
);
305 entry
->flags
|= KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
307 entry
->ecx
= F(RDPID
);
313 entry
->function
= func
;
314 entry
->index
= index
;
319 static inline int __do_cpuid_ent(struct kvm_cpuid_entry2
*entry
, u32 function
,
320 u32 index
, int *nent
, int maxnent
)
323 unsigned f_nx
= is_efer_nx() ? F(NX
) : 0;
325 unsigned f_gbpages
= (kvm_x86_ops
->get_lpage_level() == PT_PDPE_LEVEL
)
327 unsigned f_lm
= F(LM
);
329 unsigned f_gbpages
= 0;
332 unsigned f_rdtscp
= kvm_x86_ops
->rdtscp_supported() ? F(RDTSCP
) : 0;
333 unsigned f_invpcid
= kvm_x86_ops
->invpcid_supported() ? F(INVPCID
) : 0;
334 unsigned f_mpx
= kvm_mpx_supported() ? F(MPX
) : 0;
335 unsigned f_xsaves
= kvm_x86_ops
->xsaves_supported() ? F(XSAVES
) : 0;
336 unsigned f_umip
= kvm_x86_ops
->umip_emulated() ? F(UMIP
) : 0;
337 unsigned f_intel_pt
= kvm_x86_ops
->pt_supported() ? F(INTEL_PT
) : 0;
341 const u32 kvm_cpuid_1_edx_x86_features
=
342 F(FPU
) | F(VME
) | F(DE
) | F(PSE
) |
343 F(TSC
) | F(MSR
) | F(PAE
) | F(MCE
) |
344 F(CX8
) | F(APIC
) | 0 /* Reserved */ | F(SEP
) |
345 F(MTRR
) | F(PGE
) | F(MCA
) | F(CMOV
) |
346 F(PAT
) | F(PSE36
) | 0 /* PSN */ | F(CLFLUSH
) |
347 0 /* Reserved, DS, ACPI */ | F(MMX
) |
348 F(FXSR
) | F(XMM
) | F(XMM2
) | F(SELFSNOOP
) |
349 0 /* HTT, TM, Reserved, PBE */;
350 /* cpuid 0x80000001.edx */
351 const u32 kvm_cpuid_8000_0001_edx_x86_features
=
352 F(FPU
) | F(VME
) | F(DE
) | F(PSE
) |
353 F(TSC
) | F(MSR
) | F(PAE
) | F(MCE
) |
354 F(CX8
) | F(APIC
) | 0 /* Reserved */ | F(SYSCALL
) |
355 F(MTRR
) | F(PGE
) | F(MCA
) | F(CMOV
) |
356 F(PAT
) | F(PSE36
) | 0 /* Reserved */ |
357 f_nx
| 0 /* Reserved */ | F(MMXEXT
) | F(MMX
) |
358 F(FXSR
) | F(FXSR_OPT
) | f_gbpages
| f_rdtscp
|
359 0 /* Reserved */ | f_lm
| F(3DNOWEXT
) | F(3DNOW
);
361 const u32 kvm_cpuid_1_ecx_x86_features
=
362 /* NOTE: MONITOR (and MWAIT) are emulated as NOP,
363 * but *not* advertised to guests via CPUID ! */
364 F(XMM3
) | F(PCLMULQDQ
) | 0 /* DTES64, MONITOR */ |
365 0 /* DS-CPL, VMX, SMX, EST */ |
366 0 /* TM2 */ | F(SSSE3
) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
367 F(FMA
) | F(CX16
) | 0 /* xTPR Update, PDCM */ |
368 F(PCID
) | 0 /* Reserved, DCA */ | F(XMM4_1
) |
369 F(XMM4_2
) | F(X2APIC
) | F(MOVBE
) | F(POPCNT
) |
370 0 /* Reserved*/ | F(AES
) | F(XSAVE
) | 0 /* OSXSAVE */ | F(AVX
) |
372 /* cpuid 0x80000001.ecx */
373 const u32 kvm_cpuid_8000_0001_ecx_x86_features
=
374 F(LAHF_LM
) | F(CMP_LEGACY
) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
375 F(CR8_LEGACY
) | F(ABM
) | F(SSE4A
) | F(MISALIGNSSE
) |
376 F(3DNOWPREFETCH
) | F(OSVW
) | 0 /* IBS */ | F(XOP
) |
377 0 /* SKINIT, WDT, LWP */ | F(FMA4
) | F(TBM
) |
378 F(TOPOEXT
) | F(PERFCTR_CORE
);
380 /* cpuid 0x80000008.ebx */
381 const u32 kvm_cpuid_8000_0008_ebx_x86_features
=
382 F(WBNOINVD
) | F(AMD_IBPB
) | F(AMD_IBRS
) | F(AMD_SSBD
) | F(VIRT_SSBD
) |
383 F(AMD_SSB_NO
) | F(AMD_STIBP
);
385 /* cpuid 0xC0000001.edx */
386 const u32 kvm_cpuid_C000_0001_edx_x86_features
=
387 F(XSTORE
) | F(XSTORE_EN
) | F(XCRYPT
) | F(XCRYPT_EN
) |
388 F(ACE2
) | F(ACE2_EN
) | F(PHE
) | F(PHE_EN
) |
392 const u32 kvm_cpuid_7_0_ebx_x86_features
=
393 F(FSGSBASE
) | F(BMI1
) | F(HLE
) | F(AVX2
) | F(SMEP
) |
394 F(BMI2
) | F(ERMS
) | f_invpcid
| F(RTM
) | f_mpx
| F(RDSEED
) |
395 F(ADX
) | F(SMAP
) | F(AVX512IFMA
) | F(AVX512F
) | F(AVX512PF
) |
396 F(AVX512ER
) | F(AVX512CD
) | F(CLFLUSHOPT
) | F(CLWB
) | F(AVX512DQ
) |
397 F(SHA_NI
) | F(AVX512BW
) | F(AVX512VL
) | f_intel_pt
;
399 /* cpuid 0xD.1.eax */
400 const u32 kvm_cpuid_D_1_eax_x86_features
=
401 F(XSAVEOPT
) | F(XSAVEC
) | F(XGETBV1
) | f_xsaves
;
404 const u32 kvm_cpuid_7_0_ecx_x86_features
=
405 F(AVX512VBMI
) | F(LA57
) | F(PKU
) | 0 /*OSPKE*/ |
406 F(AVX512_VPOPCNTDQ
) | F(UMIP
) | F(AVX512_VBMI2
) | F(GFNI
) |
407 F(VAES
) | F(VPCLMULQDQ
) | F(AVX512_VNNI
) | F(AVX512_BITALG
) |
408 F(CLDEMOTE
) | F(MOVDIRI
) | F(MOVDIR64B
);
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
) | F(INTEL_STIBP
) |
416 /* all calls to cpuid_count() should be made on the same cpu */
421 if (*nent
>= maxnent
)
424 do_cpuid_1_ent(entry
, function
, index
);
429 entry
->eax
= min(entry
->eax
, (u32
)(f_intel_pt
? 0x14 : 0xd));
432 entry
->edx
&= kvm_cpuid_1_edx_x86_features
;
433 cpuid_mask(&entry
->edx
, CPUID_1_EDX
);
434 entry
->ecx
&= kvm_cpuid_1_ecx_x86_features
;
435 cpuid_mask(&entry
->ecx
, CPUID_1_ECX
);
436 /* we support x2apic emulation even if host does not support
437 * it since we emulate x2apic in software */
438 entry
->ecx
|= F(X2APIC
);
440 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
441 * may return different values. This forces us to get_cpu() before
442 * issuing the first command, and also to emulate this annoying behavior
443 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
445 int t
, times
= entry
->eax
& 0xff;
447 entry
->flags
|= KVM_CPUID_FLAG_STATEFUL_FUNC
;
448 entry
->flags
|= KVM_CPUID_FLAG_STATE_READ_NEXT
;
449 for (t
= 1; t
< times
; ++t
) {
450 if (*nent
>= maxnent
)
453 do_cpuid_1_ent(&entry
[t
], function
, 0);
454 entry
[t
].flags
|= KVM_CPUID_FLAG_STATEFUL_FUNC
;
459 /* function 4 has additional index. */
463 entry
->flags
|= KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
464 /* read more entries until cache_type is zero */
466 if (*nent
>= maxnent
)
469 cache_type
= entry
[i
- 1].eax
& 0x1f;
472 do_cpuid_1_ent(&entry
[i
], function
, i
);
474 KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
479 case 6: /* Thermal management */
480 entry
->eax
= 0x4; /* allow ARAT */
486 entry
->flags
|= KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
487 /* Mask ebx against host capability word 9 */
489 entry
->ebx
&= kvm_cpuid_7_0_ebx_x86_features
;
490 cpuid_mask(&entry
->ebx
, CPUID_7_0_EBX
);
491 // TSC_ADJUST is emulated
492 entry
->ebx
|= F(TSC_ADJUST
);
493 entry
->ecx
&= kvm_cpuid_7_0_ecx_x86_features
;
494 f_la57
= entry
->ecx
& F(LA57
);
495 cpuid_mask(&entry
->ecx
, CPUID_7_ECX
);
496 /* Set LA57 based on hardware capability. */
497 entry
->ecx
|= f_la57
;
498 entry
->ecx
|= f_umip
;
499 /* PKU is not yet implemented for shadow paging. */
500 if (!tdp_enabled
|| !boot_cpu_has(X86_FEATURE_OSPKE
))
501 entry
->ecx
&= ~F(PKU
);
502 entry
->edx
&= kvm_cpuid_7_0_edx_x86_features
;
503 cpuid_mask(&entry
->edx
, CPUID_7_EDX
);
505 * We emulate ARCH_CAPABILITIES in software even
506 * if the host doesn't support it.
508 entry
->edx
|= F(ARCH_CAPABILITIES
);
519 case 0xa: { /* Architectural Performance Monitoring */
520 struct x86_pmu_capability cap
;
521 union cpuid10_eax eax
;
522 union cpuid10_edx edx
;
524 perf_get_x86_pmu_capability(&cap
);
527 * Only support guest architectural pmu on a host
528 * with architectural pmu.
531 memset(&cap
, 0, sizeof(cap
));
533 eax
.split
.version_id
= min(cap
.version
, 2);
534 eax
.split
.num_counters
= cap
.num_counters_gp
;
535 eax
.split
.bit_width
= cap
.bit_width_gp
;
536 eax
.split
.mask_length
= cap
.events_mask_len
;
538 edx
.split
.num_counters_fixed
= cap
.num_counters_fixed
;
539 edx
.split
.bit_width_fixed
= cap
.bit_width_fixed
;
540 edx
.split
.reserved
= 0;
542 entry
->eax
= eax
.full
;
543 entry
->ebx
= cap
.events_mask
;
545 entry
->edx
= edx
.full
;
548 /* function 0xb has additional index. */
552 entry
->flags
|= KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
553 /* read more entries until level_type is zero */
555 if (*nent
>= maxnent
)
558 level_type
= entry
[i
- 1].ecx
& 0xff00;
561 do_cpuid_1_ent(&entry
[i
], function
, i
);
563 KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
570 u64 supported
= kvm_supported_xcr0();
572 entry
->eax
&= supported
;
573 entry
->ebx
= xstate_required_size(supported
, false);
574 entry
->ecx
= entry
->ebx
;
575 entry
->edx
&= supported
>> 32;
576 entry
->flags
|= KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
580 for (idx
= 1, i
= 1; idx
< 64; ++idx
) {
581 u64 mask
= ((u64
)1 << idx
);
582 if (*nent
>= maxnent
)
585 do_cpuid_1_ent(&entry
[i
], function
, idx
);
587 entry
[i
].eax
&= kvm_cpuid_D_1_eax_x86_features
;
588 cpuid_mask(&entry
[i
].eax
, CPUID_D_1_EAX
);
590 if (entry
[i
].eax
& (F(XSAVES
)|F(XSAVEC
)))
592 xstate_required_size(supported
,
595 if (entry
[i
].eax
== 0 || !(supported
& mask
))
597 if (WARN_ON_ONCE(entry
[i
].ecx
& 1))
603 KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
611 int t
, times
= entry
->eax
;
616 entry
->flags
|= KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
617 for (t
= 1; t
<= times
; ++t
) {
618 if (*nent
>= maxnent
)
620 do_cpuid_1_ent(&entry
[t
], function
, t
);
621 entry
[t
].flags
|= KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
626 case KVM_CPUID_SIGNATURE
: {
627 static const char signature
[12] = "KVMKVMKVM\0\0";
628 const u32
*sigptr
= (const u32
*)signature
;
629 entry
->eax
= KVM_CPUID_FEATURES
;
630 entry
->ebx
= sigptr
[0];
631 entry
->ecx
= sigptr
[1];
632 entry
->edx
= sigptr
[2];
635 case KVM_CPUID_FEATURES
:
636 entry
->eax
= (1 << KVM_FEATURE_CLOCKSOURCE
) |
637 (1 << KVM_FEATURE_NOP_IO_DELAY
) |
638 (1 << KVM_FEATURE_CLOCKSOURCE2
) |
639 (1 << KVM_FEATURE_ASYNC_PF
) |
640 (1 << KVM_FEATURE_PV_EOI
) |
641 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT
) |
642 (1 << KVM_FEATURE_PV_UNHALT
) |
643 (1 << KVM_FEATURE_PV_TLB_FLUSH
) |
644 (1 << KVM_FEATURE_ASYNC_PF_VMEXIT
) |
645 (1 << KVM_FEATURE_PV_SEND_IPI
);
648 entry
->eax
|= (1 << KVM_FEATURE_STEAL_TIME
);
655 entry
->eax
= min(entry
->eax
, 0x8000001f);
658 entry
->edx
&= kvm_cpuid_8000_0001_edx_x86_features
;
659 cpuid_mask(&entry
->edx
, CPUID_8000_0001_EDX
);
660 entry
->ecx
&= kvm_cpuid_8000_0001_ecx_x86_features
;
661 cpuid_mask(&entry
->ecx
, CPUID_8000_0001_ECX
);
663 case 0x80000007: /* Advanced power management */
664 /* invariant TSC is CPUID.80000007H:EDX[8] */
665 entry
->edx
&= (1 << 8);
666 /* mask against host */
667 entry
->edx
&= boot_cpu_data
.x86_power
;
668 entry
->eax
= entry
->ebx
= entry
->ecx
= 0;
671 unsigned g_phys_as
= (entry
->eax
>> 16) & 0xff;
672 unsigned virt_as
= max((entry
->eax
>> 8) & 0xff, 48U);
673 unsigned phys_as
= entry
->eax
& 0xff;
677 entry
->eax
= g_phys_as
| (virt_as
<< 8);
680 * IBRS, IBPB and VIRT_SSBD aren't necessarily present in
683 if (boot_cpu_has(X86_FEATURE_AMD_IBPB
))
684 entry
->ebx
|= F(AMD_IBPB
);
685 if (boot_cpu_has(X86_FEATURE_AMD_IBRS
))
686 entry
->ebx
|= F(AMD_IBRS
);
687 if (boot_cpu_has(X86_FEATURE_VIRT_SSBD
))
688 entry
->ebx
|= F(VIRT_SSBD
);
689 entry
->ebx
&= kvm_cpuid_8000_0008_ebx_x86_features
;
690 cpuid_mask(&entry
->ebx
, CPUID_8000_0008_EBX
);
692 * The preference is to use SPEC CTRL MSR instead of the
695 if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD
) &&
696 !boot_cpu_has(X86_FEATURE_AMD_SSBD
))
697 entry
->ebx
|= F(VIRT_SSBD
);
701 entry
->ecx
= entry
->edx
= 0;
707 /*Add support for Centaur's CPUID instruction*/
709 /*Just support up to 0xC0000004 now*/
710 entry
->eax
= min(entry
->eax
, 0xC0000004);
713 entry
->edx
&= kvm_cpuid_C000_0001_edx_x86_features
;
714 cpuid_mask(&entry
->edx
, CPUID_C000_0001_EDX
);
716 case 3: /* Processor serial number */
717 case 5: /* MONITOR/MWAIT */
722 entry
->eax
= entry
->ebx
= entry
->ecx
= entry
->edx
= 0;
726 kvm_x86_ops
->set_supported_cpuid(function
, entry
);
736 static int do_cpuid_ent(struct kvm_cpuid_entry2
*entry
, u32 func
,
737 u32 idx
, int *nent
, int maxnent
, unsigned int type
)
739 if (type
== KVM_GET_EMULATED_CPUID
)
740 return __do_cpuid_ent_emulated(entry
, func
, idx
, nent
, maxnent
);
742 return __do_cpuid_ent(entry
, func
, idx
, nent
, maxnent
);
747 struct kvm_cpuid_param
{
751 bool (*qualifier
)(const struct kvm_cpuid_param
*param
);
754 static bool is_centaur_cpu(const struct kvm_cpuid_param
*param
)
756 return boot_cpu_data
.x86_vendor
== X86_VENDOR_CENTAUR
;
759 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user
*entries
,
760 __u32 num_entries
, unsigned int ioctl_type
)
765 if (ioctl_type
!= KVM_GET_EMULATED_CPUID
)
769 * We want to make sure that ->padding is being passed clean from
770 * userspace in case we want to use it for something in the future.
772 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
773 * have to give ourselves satisfied only with the emulated side. /me
776 for (i
= 0; i
< num_entries
; i
++) {
777 if (copy_from_user(pad
, entries
[i
].padding
, sizeof(pad
)))
780 if (pad
[0] || pad
[1] || pad
[2])
786 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2
*cpuid
,
787 struct kvm_cpuid_entry2 __user
*entries
,
790 struct kvm_cpuid_entry2
*cpuid_entries
;
791 int limit
, nent
= 0, r
= -E2BIG
, i
;
793 static const struct kvm_cpuid_param param
[] = {
794 { .func
= 0, .has_leaf_count
= true },
795 { .func
= 0x80000000, .has_leaf_count
= true },
796 { .func
= 0xC0000000, .qualifier
= is_centaur_cpu
, .has_leaf_count
= true },
797 { .func
= KVM_CPUID_SIGNATURE
},
798 { .func
= KVM_CPUID_FEATURES
},
803 if (cpuid
->nent
> KVM_MAX_CPUID_ENTRIES
)
804 cpuid
->nent
= KVM_MAX_CPUID_ENTRIES
;
806 if (sanity_check_entries(entries
, cpuid
->nent
, type
))
810 cpuid_entries
= vzalloc(array_size(sizeof(struct kvm_cpuid_entry2
),
816 for (i
= 0; i
< ARRAY_SIZE(param
); i
++) {
817 const struct kvm_cpuid_param
*ent
= ¶m
[i
];
819 if (ent
->qualifier
&& !ent
->qualifier(ent
))
822 r
= do_cpuid_ent(&cpuid_entries
[nent
], ent
->func
, ent
->idx
,
823 &nent
, cpuid
->nent
, type
);
828 if (!ent
->has_leaf_count
)
831 limit
= cpuid_entries
[nent
- 1].eax
;
832 for (func
= ent
->func
+ 1; func
<= limit
&& nent
< cpuid
->nent
&& r
== 0; ++func
)
833 r
= do_cpuid_ent(&cpuid_entries
[nent
], func
, ent
->idx
,
834 &nent
, cpuid
->nent
, type
);
841 if (copy_to_user(entries
, cpuid_entries
,
842 nent
* sizeof(struct kvm_cpuid_entry2
)))
848 vfree(cpuid_entries
);
853 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu
*vcpu
, int i
)
855 struct kvm_cpuid_entry2
*e
= &vcpu
->arch
.cpuid_entries
[i
];
856 struct kvm_cpuid_entry2
*ej
;
858 int nent
= vcpu
->arch
.cpuid_nent
;
860 e
->flags
&= ~KVM_CPUID_FLAG_STATE_READ_NEXT
;
861 /* when no next entry is found, the current entry[i] is reselected */
864 ej
= &vcpu
->arch
.cpuid_entries
[j
];
865 } while (ej
->function
!= e
->function
);
867 ej
->flags
|= KVM_CPUID_FLAG_STATE_READ_NEXT
;
872 /* find an entry with matching function, matching index (if needed), and that
873 * should be read next (if it's stateful) */
874 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2
*e
,
875 u32 function
, u32 index
)
877 if (e
->function
!= function
)
879 if ((e
->flags
& KVM_CPUID_FLAG_SIGNIFCANT_INDEX
) && e
->index
!= index
)
881 if ((e
->flags
& KVM_CPUID_FLAG_STATEFUL_FUNC
) &&
882 !(e
->flags
& KVM_CPUID_FLAG_STATE_READ_NEXT
))
887 struct kvm_cpuid_entry2
*kvm_find_cpuid_entry(struct kvm_vcpu
*vcpu
,
888 u32 function
, u32 index
)
891 struct kvm_cpuid_entry2
*best
= NULL
;
893 for (i
= 0; i
< vcpu
->arch
.cpuid_nent
; ++i
) {
894 struct kvm_cpuid_entry2
*e
;
896 e
= &vcpu
->arch
.cpuid_entries
[i
];
897 if (is_matching_cpuid_entry(e
, function
, index
)) {
898 if (e
->flags
& KVM_CPUID_FLAG_STATEFUL_FUNC
)
899 move_to_next_stateful_cpuid_entry(vcpu
, i
);
906 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry
);
909 * If no match is found, check whether we exceed the vCPU's limit
910 * and return the content of the highest valid _standard_ leaf instead.
911 * This is to satisfy the CPUID specification.
913 static struct kvm_cpuid_entry2
* check_cpuid_limit(struct kvm_vcpu
*vcpu
,
914 u32 function
, u32 index
)
916 struct kvm_cpuid_entry2
*maxlevel
;
918 maxlevel
= kvm_find_cpuid_entry(vcpu
, function
& 0x80000000, 0);
919 if (!maxlevel
|| maxlevel
->eax
>= function
)
921 if (function
& 0x80000000) {
922 maxlevel
= kvm_find_cpuid_entry(vcpu
, 0, 0);
926 return kvm_find_cpuid_entry(vcpu
, maxlevel
->eax
, index
);
929 bool kvm_cpuid(struct kvm_vcpu
*vcpu
, u32
*eax
, u32
*ebx
,
930 u32
*ecx
, u32
*edx
, bool check_limit
)
932 u32 function
= *eax
, index
= *ecx
;
933 struct kvm_cpuid_entry2
*best
;
934 bool entry_found
= true;
936 best
= kvm_find_cpuid_entry(vcpu
, function
, index
);
943 best
= check_cpuid_limit(vcpu
, function
, index
);
953 *eax
= *ebx
= *ecx
= *edx
= 0;
954 trace_kvm_cpuid(function
, *eax
, *ebx
, *ecx
, *edx
, entry_found
);
957 EXPORT_SYMBOL_GPL(kvm_cpuid
);
959 int kvm_emulate_cpuid(struct kvm_vcpu
*vcpu
)
961 u32 eax
, ebx
, ecx
, edx
;
963 if (cpuid_fault_enabled(vcpu
) && !kvm_require_cpl(vcpu
, 0))
966 eax
= kvm_rax_read(vcpu
);
967 ecx
= kvm_rcx_read(vcpu
);
968 kvm_cpuid(vcpu
, &eax
, &ebx
, &ecx
, &edx
, true);
969 kvm_rax_write(vcpu
, eax
);
970 kvm_rbx_write(vcpu
, ebx
);
971 kvm_rcx_write(vcpu
, ecx
);
972 kvm_rdx_write(vcpu
, edx
);
973 return kvm_skip_emulated_instruction(vcpu
);
975 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid
);