1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
4 * cpuid support routines
6 * derived from arch/x86/kvm/x86.c
8 * Copyright 2011 Red Hat, Inc. and/or its affiliates.
9 * Copyright IBM Corporation, 2008
12 #include <linux/kvm_host.h>
13 #include <linux/export.h>
14 #include <linux/vmalloc.h>
15 #include <linux/uaccess.h>
16 #include <linux/sched/stat.h>
18 #include <asm/processor.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 (boot_cpu_has(X86_FEATURE_XSAVE
) && best
->function
== 0x1) {
78 best
->ecx
&= ~F(OSXSAVE
);
79 if (kvm_read_cr4_bits(vcpu
, X86_CR4_OSXSAVE
))
80 best
->ecx
|= F(OSXSAVE
);
83 best
->edx
&= ~F(APIC
);
84 if (vcpu
->arch
.apic_base
& MSR_IA32_APICBASE_ENABLE
)
88 if (best
->ecx
& F(TSC_DEADLINE_TIMER
))
89 apic
->lapic_timer
.timer_mode_mask
= 3 << 17;
91 apic
->lapic_timer
.timer_mode_mask
= 1 << 17;
94 best
= kvm_find_cpuid_entry(vcpu
, 7, 0);
96 /* Update OSPKE bit */
97 if (boot_cpu_has(X86_FEATURE_PKU
) && best
->function
== 0x7) {
98 best
->ecx
&= ~F(OSPKE
);
99 if (kvm_read_cr4_bits(vcpu
, X86_CR4_PKE
))
100 best
->ecx
|= F(OSPKE
);
104 best
= kvm_find_cpuid_entry(vcpu
, 0xD, 0);
106 vcpu
->arch
.guest_supported_xcr0
= 0;
107 vcpu
->arch
.guest_xstate_size
= XSAVE_HDR_SIZE
+ XSAVE_HDR_OFFSET
;
109 vcpu
->arch
.guest_supported_xcr0
=
110 (best
->eax
| ((u64
)best
->edx
<< 32)) &
111 kvm_supported_xcr0();
112 vcpu
->arch
.guest_xstate_size
= best
->ebx
=
113 xstate_required_size(vcpu
->arch
.xcr0
, false);
116 best
= kvm_find_cpuid_entry(vcpu
, 0xD, 1);
117 if (best
&& (best
->eax
& (F(XSAVES
) | F(XSAVEC
))))
118 best
->ebx
= xstate_required_size(vcpu
->arch
.xcr0
, true);
121 * The existing code assumes virtual address is 48-bit or 57-bit in the
122 * canonical address checks; exit if it is ever changed.
124 best
= kvm_find_cpuid_entry(vcpu
, 0x80000008, 0);
126 int vaddr_bits
= (best
->eax
& 0xff00) >> 8;
128 if (vaddr_bits
!= 48 && vaddr_bits
!= 57 && vaddr_bits
!= 0)
132 best
= kvm_find_cpuid_entry(vcpu
, KVM_CPUID_FEATURES
, 0);
133 if (kvm_hlt_in_guest(vcpu
->kvm
) && best
&&
134 (best
->eax
& (1 << KVM_FEATURE_PV_UNHALT
)))
135 best
->eax
&= ~(1 << KVM_FEATURE_PV_UNHALT
);
137 if (!kvm_check_has_quirk(vcpu
->kvm
, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT
)) {
138 best
= kvm_find_cpuid_entry(vcpu
, 0x1, 0);
140 if (vcpu
->arch
.ia32_misc_enable_msr
& MSR_IA32_MISC_ENABLE_MWAIT
)
141 best
->ecx
|= F(MWAIT
);
143 best
->ecx
&= ~F(MWAIT
);
147 /* Update physical-address width */
148 vcpu
->arch
.maxphyaddr
= cpuid_query_maxphyaddr(vcpu
);
149 kvm_mmu_reset_context(vcpu
);
151 kvm_pmu_refresh(vcpu
);
155 static int is_efer_nx(void)
157 unsigned long long efer
= 0;
159 rdmsrl_safe(MSR_EFER
, &efer
);
160 return efer
& EFER_NX
;
163 static void cpuid_fix_nx_cap(struct kvm_vcpu
*vcpu
)
166 struct kvm_cpuid_entry2
*e
, *entry
;
169 for (i
= 0; i
< vcpu
->arch
.cpuid_nent
; ++i
) {
170 e
= &vcpu
->arch
.cpuid_entries
[i
];
171 if (e
->function
== 0x80000001) {
176 if (entry
&& (entry
->edx
& F(NX
)) && !is_efer_nx()) {
177 entry
->edx
&= ~F(NX
);
178 printk(KERN_INFO
"kvm: guest NX capability removed\n");
182 int cpuid_query_maxphyaddr(struct kvm_vcpu
*vcpu
)
184 struct kvm_cpuid_entry2
*best
;
186 best
= kvm_find_cpuid_entry(vcpu
, 0x80000000, 0);
187 if (!best
|| best
->eax
< 0x80000008)
189 best
= kvm_find_cpuid_entry(vcpu
, 0x80000008, 0);
191 return best
->eax
& 0xff;
195 EXPORT_SYMBOL_GPL(cpuid_query_maxphyaddr
);
197 /* when an old userspace process fills a new kernel module */
198 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu
*vcpu
,
199 struct kvm_cpuid
*cpuid
,
200 struct kvm_cpuid_entry __user
*entries
)
203 struct kvm_cpuid_entry
*cpuid_entries
= NULL
;
206 if (cpuid
->nent
> KVM_MAX_CPUID_ENTRIES
)
211 vmalloc(array_size(sizeof(struct kvm_cpuid_entry
),
216 if (copy_from_user(cpuid_entries
, entries
,
217 cpuid
->nent
* sizeof(struct kvm_cpuid_entry
)))
220 for (i
= 0; i
< cpuid
->nent
; i
++) {
221 vcpu
->arch
.cpuid_entries
[i
].function
= cpuid_entries
[i
].function
;
222 vcpu
->arch
.cpuid_entries
[i
].eax
= cpuid_entries
[i
].eax
;
223 vcpu
->arch
.cpuid_entries
[i
].ebx
= cpuid_entries
[i
].ebx
;
224 vcpu
->arch
.cpuid_entries
[i
].ecx
= cpuid_entries
[i
].ecx
;
225 vcpu
->arch
.cpuid_entries
[i
].edx
= cpuid_entries
[i
].edx
;
226 vcpu
->arch
.cpuid_entries
[i
].index
= 0;
227 vcpu
->arch
.cpuid_entries
[i
].flags
= 0;
228 vcpu
->arch
.cpuid_entries
[i
].padding
[0] = 0;
229 vcpu
->arch
.cpuid_entries
[i
].padding
[1] = 0;
230 vcpu
->arch
.cpuid_entries
[i
].padding
[2] = 0;
232 vcpu
->arch
.cpuid_nent
= cpuid
->nent
;
233 cpuid_fix_nx_cap(vcpu
);
234 kvm_apic_set_version(vcpu
);
235 kvm_x86_ops
->cpuid_update(vcpu
);
236 r
= kvm_update_cpuid(vcpu
);
239 vfree(cpuid_entries
);
243 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu
*vcpu
,
244 struct kvm_cpuid2
*cpuid
,
245 struct kvm_cpuid_entry2 __user
*entries
)
250 if (cpuid
->nent
> KVM_MAX_CPUID_ENTRIES
)
253 if (copy_from_user(&vcpu
->arch
.cpuid_entries
, entries
,
254 cpuid
->nent
* sizeof(struct kvm_cpuid_entry2
)))
256 vcpu
->arch
.cpuid_nent
= cpuid
->nent
;
257 kvm_apic_set_version(vcpu
);
258 kvm_x86_ops
->cpuid_update(vcpu
);
259 r
= kvm_update_cpuid(vcpu
);
264 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu
*vcpu
,
265 struct kvm_cpuid2
*cpuid
,
266 struct kvm_cpuid_entry2 __user
*entries
)
271 if (cpuid
->nent
< vcpu
->arch
.cpuid_nent
)
274 if (copy_to_user(entries
, &vcpu
->arch
.cpuid_entries
,
275 vcpu
->arch
.cpuid_nent
* sizeof(struct kvm_cpuid_entry2
)))
280 cpuid
->nent
= vcpu
->arch
.cpuid_nent
;
284 static void cpuid_mask(u32
*word
, int wordnum
)
286 *word
&= boot_cpu_data
.x86_capability
[wordnum
];
289 static void do_host_cpuid(struct kvm_cpuid_entry2
*entry
, u32 function
,
292 entry
->function
= function
;
293 entry
->index
= index
;
296 cpuid_count(entry
->function
, entry
->index
,
297 &entry
->eax
, &entry
->ebx
, &entry
->ecx
, &entry
->edx
);
301 entry
->flags
|= KVM_CPUID_FLAG_STATEFUL_FUNC
;
315 entry
->flags
|= KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
320 static int __do_cpuid_func_emulated(struct kvm_cpuid_entry2
*entry
,
321 u32 func
, int *nent
, int maxnent
)
323 entry
->function
= func
;
333 entry
->ecx
= F(MOVBE
);
337 entry
->flags
|= KVM_CPUID_FLAG_SIGNIFCANT_INDEX
;
339 entry
->ecx
= F(RDPID
);
348 static inline void do_cpuid_7_mask(struct kvm_cpuid_entry2
*entry
, int index
)
350 unsigned f_invpcid
= kvm_x86_ops
->invpcid_supported() ? F(INVPCID
) : 0;
351 unsigned f_mpx
= kvm_mpx_supported() ? F(MPX
) : 0;
352 unsigned f_umip
= kvm_x86_ops
->umip_emulated() ? F(UMIP
) : 0;
353 unsigned f_intel_pt
= kvm_x86_ops
->pt_supported() ? F(INTEL_PT
) : 0;
357 const u32 kvm_cpuid_7_0_ebx_x86_features
=
358 F(FSGSBASE
) | F(BMI1
) | F(HLE
) | F(AVX2
) | F(SMEP
) |
359 F(BMI2
) | F(ERMS
) | f_invpcid
| F(RTM
) | f_mpx
| F(RDSEED
) |
360 F(ADX
) | F(SMAP
) | F(AVX512IFMA
) | F(AVX512F
) | F(AVX512PF
) |
361 F(AVX512ER
) | F(AVX512CD
) | F(CLFLUSHOPT
) | F(CLWB
) | F(AVX512DQ
) |
362 F(SHA_NI
) | F(AVX512BW
) | F(AVX512VL
) | f_intel_pt
;
365 const u32 kvm_cpuid_7_0_ecx_x86_features
=
366 F(AVX512VBMI
) | F(LA57
) | F(PKU
) | 0 /*OSPKE*/ |
367 F(AVX512_VPOPCNTDQ
) | F(UMIP
) | F(AVX512_VBMI2
) | F(GFNI
) |
368 F(VAES
) | F(VPCLMULQDQ
) | F(AVX512_VNNI
) | F(AVX512_BITALG
) |
369 F(CLDEMOTE
) | F(MOVDIRI
) | F(MOVDIR64B
);
372 const u32 kvm_cpuid_7_0_edx_x86_features
=
373 F(AVX512_4VNNIW
) | F(AVX512_4FMAPS
) | F(SPEC_CTRL
) |
374 F(SPEC_CTRL_SSBD
) | F(ARCH_CAPABILITIES
) | F(INTEL_STIBP
) |
378 const u32 kvm_cpuid_7_1_eax_x86_features
=
383 entry
->eax
= min(entry
->eax
, 1u);
384 entry
->ebx
&= kvm_cpuid_7_0_ebx_x86_features
;
385 cpuid_mask(&entry
->ebx
, CPUID_7_0_EBX
);
386 /* TSC_ADJUST is emulated */
387 entry
->ebx
|= F(TSC_ADJUST
);
389 entry
->ecx
&= kvm_cpuid_7_0_ecx_x86_features
;
390 f_la57
= entry
->ecx
& F(LA57
);
391 cpuid_mask(&entry
->ecx
, CPUID_7_ECX
);
392 /* Set LA57 based on hardware capability. */
393 entry
->ecx
|= f_la57
;
394 entry
->ecx
|= f_umip
;
395 /* PKU is not yet implemented for shadow paging. */
396 if (!tdp_enabled
|| !boot_cpu_has(X86_FEATURE_OSPKE
))
397 entry
->ecx
&= ~F(PKU
);
399 entry
->edx
&= kvm_cpuid_7_0_edx_x86_features
;
400 cpuid_mask(&entry
->edx
, CPUID_7_EDX
);
402 * We emulate ARCH_CAPABILITIES in software even
403 * if the host doesn't support it.
405 entry
->edx
|= F(ARCH_CAPABILITIES
);
408 entry
->eax
&= kvm_cpuid_7_1_eax_x86_features
;
423 static inline int __do_cpuid_func(struct kvm_cpuid_entry2
*entry
, u32 function
,
424 int *nent
, int maxnent
)
427 unsigned f_nx
= is_efer_nx() ? F(NX
) : 0;
429 unsigned f_gbpages
= (kvm_x86_ops
->get_lpage_level() == PT_PDPE_LEVEL
)
431 unsigned f_lm
= F(LM
);
433 unsigned f_gbpages
= 0;
436 unsigned f_rdtscp
= kvm_x86_ops
->rdtscp_supported() ? F(RDTSCP
) : 0;
437 unsigned f_xsaves
= kvm_x86_ops
->xsaves_supported() ? F(XSAVES
) : 0;
438 unsigned f_intel_pt
= kvm_x86_ops
->pt_supported() ? F(INTEL_PT
) : 0;
441 const u32 kvm_cpuid_1_edx_x86_features
=
442 F(FPU
) | F(VME
) | F(DE
) | F(PSE
) |
443 F(TSC
) | F(MSR
) | F(PAE
) | F(MCE
) |
444 F(CX8
) | F(APIC
) | 0 /* Reserved */ | F(SEP
) |
445 F(MTRR
) | F(PGE
) | F(MCA
) | F(CMOV
) |
446 F(PAT
) | F(PSE36
) | 0 /* PSN */ | F(CLFLUSH
) |
447 0 /* Reserved, DS, ACPI */ | F(MMX
) |
448 F(FXSR
) | F(XMM
) | F(XMM2
) | F(SELFSNOOP
) |
449 0 /* HTT, TM, Reserved, PBE */;
450 /* cpuid 0x80000001.edx */
451 const u32 kvm_cpuid_8000_0001_edx_x86_features
=
452 F(FPU
) | F(VME
) | F(DE
) | F(PSE
) |
453 F(TSC
) | F(MSR
) | F(PAE
) | F(MCE
) |
454 F(CX8
) | F(APIC
) | 0 /* Reserved */ | F(SYSCALL
) |
455 F(MTRR
) | F(PGE
) | F(MCA
) | F(CMOV
) |
456 F(PAT
) | F(PSE36
) | 0 /* Reserved */ |
457 f_nx
| 0 /* Reserved */ | F(MMXEXT
) | F(MMX
) |
458 F(FXSR
) | F(FXSR_OPT
) | f_gbpages
| f_rdtscp
|
459 0 /* Reserved */ | f_lm
| F(3DNOWEXT
) | F(3DNOW
);
461 const u32 kvm_cpuid_1_ecx_x86_features
=
462 /* NOTE: MONITOR (and MWAIT) are emulated as NOP,
463 * but *not* advertised to guests via CPUID ! */
464 F(XMM3
) | F(PCLMULQDQ
) | 0 /* DTES64, MONITOR */ |
465 0 /* DS-CPL, VMX, SMX, EST */ |
466 0 /* TM2 */ | F(SSSE3
) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
467 F(FMA
) | F(CX16
) | 0 /* xTPR Update, PDCM */ |
468 F(PCID
) | 0 /* Reserved, DCA */ | F(XMM4_1
) |
469 F(XMM4_2
) | F(X2APIC
) | F(MOVBE
) | F(POPCNT
) |
470 0 /* Reserved*/ | F(AES
) | F(XSAVE
) | 0 /* OSXSAVE */ | F(AVX
) |
472 /* cpuid 0x80000001.ecx */
473 const u32 kvm_cpuid_8000_0001_ecx_x86_features
=
474 F(LAHF_LM
) | F(CMP_LEGACY
) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
475 F(CR8_LEGACY
) | F(ABM
) | F(SSE4A
) | F(MISALIGNSSE
) |
476 F(3DNOWPREFETCH
) | F(OSVW
) | 0 /* IBS */ | F(XOP
) |
477 0 /* SKINIT, WDT, LWP */ | F(FMA4
) | F(TBM
) |
478 F(TOPOEXT
) | F(PERFCTR_CORE
);
480 /* cpuid 0x80000008.ebx */
481 const u32 kvm_cpuid_8000_0008_ebx_x86_features
=
482 F(WBNOINVD
) | F(AMD_IBPB
) | F(AMD_IBRS
) | F(AMD_SSBD
) | F(VIRT_SSBD
) |
483 F(AMD_SSB_NO
) | F(AMD_STIBP
) | F(AMD_STIBP_ALWAYS_ON
);
485 /* cpuid 0xC0000001.edx */
486 const u32 kvm_cpuid_C000_0001_edx_x86_features
=
487 F(XSTORE
) | F(XSTORE_EN
) | F(XCRYPT
) | F(XCRYPT_EN
) |
488 F(ACE2
) | F(ACE2_EN
) | F(PHE
) | F(PHE_EN
) |
491 /* cpuid 0xD.1.eax */
492 const u32 kvm_cpuid_D_1_eax_x86_features
=
493 F(XSAVEOPT
) | F(XSAVEC
) | F(XGETBV1
) | f_xsaves
;
495 /* all calls to cpuid_count() should be made on the same cpu */
500 if (*nent
>= maxnent
)
503 do_host_cpuid(entry
, function
, 0);
508 /* Limited to the highest leaf implemented in KVM. */
509 entry
->eax
= min(entry
->eax
, 0x1fU
);
512 entry
->edx
&= kvm_cpuid_1_edx_x86_features
;
513 cpuid_mask(&entry
->edx
, CPUID_1_EDX
);
514 entry
->ecx
&= kvm_cpuid_1_ecx_x86_features
;
515 cpuid_mask(&entry
->ecx
, CPUID_1_ECX
);
516 /* we support x2apic emulation even if host does not support
517 * it since we emulate x2apic in software */
518 entry
->ecx
|= F(X2APIC
);
520 /* function 2 entries are STATEFUL. That is, repeated cpuid commands
521 * may return different values. This forces us to get_cpu() before
522 * issuing the first command, and also to emulate this annoying behavior
523 * in kvm_emulate_cpuid() using KVM_CPUID_FLAG_STATE_READ_NEXT */
525 int t
, times
= entry
->eax
& 0xff;
527 entry
->flags
|= KVM_CPUID_FLAG_STATE_READ_NEXT
;
528 for (t
= 1; t
< times
; ++t
) {
529 if (*nent
>= maxnent
)
532 do_host_cpuid(&entry
[t
], function
, 0);
537 /* functions 4 and 0x8000001d have additional index. */
542 /* read more entries until cache_type is zero */
544 if (*nent
>= maxnent
)
547 cache_type
= entry
[i
- 1].eax
& 0x1f;
550 do_host_cpuid(&entry
[i
], function
, i
);
555 case 6: /* Thermal management */
556 entry
->eax
= 0x4; /* allow ARAT */
561 /* function 7 has additional index. */
566 do_cpuid_7_mask(&entry
[i
], i
);
569 if (*nent
>= maxnent
)
573 do_host_cpuid(&entry
[i
], function
, i
);
580 case 0xa: { /* Architectural Performance Monitoring */
581 struct x86_pmu_capability cap
;
582 union cpuid10_eax eax
;
583 union cpuid10_edx edx
;
585 perf_get_x86_pmu_capability(&cap
);
588 * Only support guest architectural pmu on a host
589 * with architectural pmu.
592 memset(&cap
, 0, sizeof(cap
));
594 eax
.split
.version_id
= min(cap
.version
, 2);
595 eax
.split
.num_counters
= cap
.num_counters_gp
;
596 eax
.split
.bit_width
= cap
.bit_width_gp
;
597 eax
.split
.mask_length
= cap
.events_mask_len
;
599 edx
.split
.num_counters_fixed
= cap
.num_counters_fixed
;
600 edx
.split
.bit_width_fixed
= cap
.bit_width_fixed
;
601 edx
.split
.reserved
= 0;
603 entry
->eax
= eax
.full
;
604 entry
->ebx
= cap
.events_mask
;
606 entry
->edx
= edx
.full
;
610 * Per Intel's SDM, the 0x1f is a superset of 0xb,
611 * thus they can be handled by common code.
617 /* read more entries until level_type is zero */
619 if (*nent
>= maxnent
)
622 level_type
= entry
[i
- 1].ecx
& 0xff00;
625 do_host_cpuid(&entry
[i
], function
, i
);
632 u64 supported
= kvm_supported_xcr0();
634 entry
->eax
&= supported
;
635 entry
->ebx
= xstate_required_size(supported
, false);
636 entry
->ecx
= entry
->ebx
;
637 entry
->edx
&= supported
>> 32;
641 for (idx
= 1, i
= 1; idx
< 64; ++idx
) {
642 u64 mask
= ((u64
)1 << idx
);
643 if (*nent
>= maxnent
)
646 do_host_cpuid(&entry
[i
], function
, idx
);
648 entry
[i
].eax
&= kvm_cpuid_D_1_eax_x86_features
;
649 cpuid_mask(&entry
[i
].eax
, CPUID_D_1_EAX
);
651 if (entry
[i
].eax
& (F(XSAVES
)|F(XSAVEC
)))
653 xstate_required_size(supported
,
656 if (entry
[i
].eax
== 0 || !(supported
& mask
))
658 if (WARN_ON_ONCE(entry
[i
].ecx
& 1))
670 int t
, times
= entry
->eax
;
675 for (t
= 1; t
<= times
; ++t
) {
676 if (*nent
>= maxnent
)
678 do_host_cpuid(&entry
[t
], function
, t
);
683 case KVM_CPUID_SIGNATURE
: {
684 static const char signature
[12] = "KVMKVMKVM\0\0";
685 const u32
*sigptr
= (const u32
*)signature
;
686 entry
->eax
= KVM_CPUID_FEATURES
;
687 entry
->ebx
= sigptr
[0];
688 entry
->ecx
= sigptr
[1];
689 entry
->edx
= sigptr
[2];
692 case KVM_CPUID_FEATURES
:
693 entry
->eax
= (1 << KVM_FEATURE_CLOCKSOURCE
) |
694 (1 << KVM_FEATURE_NOP_IO_DELAY
) |
695 (1 << KVM_FEATURE_CLOCKSOURCE2
) |
696 (1 << KVM_FEATURE_ASYNC_PF
) |
697 (1 << KVM_FEATURE_PV_EOI
) |
698 (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT
) |
699 (1 << KVM_FEATURE_PV_UNHALT
) |
700 (1 << KVM_FEATURE_PV_TLB_FLUSH
) |
701 (1 << KVM_FEATURE_ASYNC_PF_VMEXIT
) |
702 (1 << KVM_FEATURE_PV_SEND_IPI
) |
703 (1 << KVM_FEATURE_POLL_CONTROL
) |
704 (1 << KVM_FEATURE_PV_SCHED_YIELD
);
707 entry
->eax
|= (1 << KVM_FEATURE_STEAL_TIME
);
714 entry
->eax
= min(entry
->eax
, 0x8000001f);
717 entry
->edx
&= kvm_cpuid_8000_0001_edx_x86_features
;
718 cpuid_mask(&entry
->edx
, CPUID_8000_0001_EDX
);
719 entry
->ecx
&= kvm_cpuid_8000_0001_ecx_x86_features
;
720 cpuid_mask(&entry
->ecx
, CPUID_8000_0001_ECX
);
722 case 0x80000007: /* Advanced power management */
723 /* invariant TSC is CPUID.80000007H:EDX[8] */
724 entry
->edx
&= (1 << 8);
725 /* mask against host */
726 entry
->edx
&= boot_cpu_data
.x86_power
;
727 entry
->eax
= entry
->ebx
= entry
->ecx
= 0;
730 unsigned g_phys_as
= (entry
->eax
>> 16) & 0xff;
731 unsigned virt_as
= max((entry
->eax
>> 8) & 0xff, 48U);
732 unsigned phys_as
= entry
->eax
& 0xff;
736 entry
->eax
= g_phys_as
| (virt_as
<< 8);
739 * IBRS, IBPB and VIRT_SSBD aren't necessarily present in
742 if (boot_cpu_has(X86_FEATURE_AMD_IBPB
))
743 entry
->ebx
|= F(AMD_IBPB
);
744 if (boot_cpu_has(X86_FEATURE_AMD_IBRS
))
745 entry
->ebx
|= F(AMD_IBRS
);
746 if (boot_cpu_has(X86_FEATURE_VIRT_SSBD
))
747 entry
->ebx
|= F(VIRT_SSBD
);
748 entry
->ebx
&= kvm_cpuid_8000_0008_ebx_x86_features
;
749 cpuid_mask(&entry
->ebx
, CPUID_8000_0008_EBX
);
751 * The preference is to use SPEC CTRL MSR instead of the
754 if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD
) &&
755 !boot_cpu_has(X86_FEATURE_AMD_SSBD
))
756 entry
->ebx
|= F(VIRT_SSBD
);
760 entry
->ecx
= entry
->edx
= 0;
765 /*Add support for Centaur's CPUID instruction*/
767 /*Just support up to 0xC0000004 now*/
768 entry
->eax
= min(entry
->eax
, 0xC0000004);
771 entry
->edx
&= kvm_cpuid_C000_0001_edx_x86_features
;
772 cpuid_mask(&entry
->edx
, CPUID_C000_0001_EDX
);
774 case 3: /* Processor serial number */
775 case 5: /* MONITOR/MWAIT */
780 entry
->eax
= entry
->ebx
= entry
->ecx
= entry
->edx
= 0;
784 kvm_x86_ops
->set_supported_cpuid(function
, entry
);
794 static int do_cpuid_func(struct kvm_cpuid_entry2
*entry
, u32 func
,
795 int *nent
, int maxnent
, unsigned int type
)
797 if (type
== KVM_GET_EMULATED_CPUID
)
798 return __do_cpuid_func_emulated(entry
, func
, nent
, maxnent
);
800 return __do_cpuid_func(entry
, func
, nent
, maxnent
);
805 struct kvm_cpuid_param
{
807 bool (*qualifier
)(const struct kvm_cpuid_param
*param
);
810 static bool is_centaur_cpu(const struct kvm_cpuid_param
*param
)
812 return boot_cpu_data
.x86_vendor
== X86_VENDOR_CENTAUR
;
815 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user
*entries
,
816 __u32 num_entries
, unsigned int ioctl_type
)
821 if (ioctl_type
!= KVM_GET_EMULATED_CPUID
)
825 * We want to make sure that ->padding is being passed clean from
826 * userspace in case we want to use it for something in the future.
828 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
829 * have to give ourselves satisfied only with the emulated side. /me
832 for (i
= 0; i
< num_entries
; i
++) {
833 if (copy_from_user(pad
, entries
[i
].padding
, sizeof(pad
)))
836 if (pad
[0] || pad
[1] || pad
[2])
842 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2
*cpuid
,
843 struct kvm_cpuid_entry2 __user
*entries
,
846 struct kvm_cpuid_entry2
*cpuid_entries
;
847 int limit
, nent
= 0, r
= -E2BIG
, i
;
849 static const struct kvm_cpuid_param param
[] = {
851 { .func
= 0x80000000 },
852 { .func
= 0xC0000000, .qualifier
= is_centaur_cpu
},
853 { .func
= KVM_CPUID_SIGNATURE
},
858 if (cpuid
->nent
> KVM_MAX_CPUID_ENTRIES
)
859 cpuid
->nent
= KVM_MAX_CPUID_ENTRIES
;
861 if (sanity_check_entries(entries
, cpuid
->nent
, type
))
865 cpuid_entries
= vzalloc(array_size(sizeof(struct kvm_cpuid_entry2
),
871 for (i
= 0; i
< ARRAY_SIZE(param
); i
++) {
872 const struct kvm_cpuid_param
*ent
= ¶m
[i
];
874 if (ent
->qualifier
&& !ent
->qualifier(ent
))
877 r
= do_cpuid_func(&cpuid_entries
[nent
], ent
->func
,
878 &nent
, cpuid
->nent
, type
);
883 limit
= cpuid_entries
[nent
- 1].eax
;
884 for (func
= ent
->func
+ 1; func
<= limit
&& nent
< cpuid
->nent
&& r
== 0; ++func
)
885 r
= do_cpuid_func(&cpuid_entries
[nent
], func
,
886 &nent
, cpuid
->nent
, type
);
893 if (copy_to_user(entries
, cpuid_entries
,
894 nent
* sizeof(struct kvm_cpuid_entry2
)))
900 vfree(cpuid_entries
);
905 static int move_to_next_stateful_cpuid_entry(struct kvm_vcpu
*vcpu
, int i
)
907 struct kvm_cpuid_entry2
*e
= &vcpu
->arch
.cpuid_entries
[i
];
908 struct kvm_cpuid_entry2
*ej
;
910 int nent
= vcpu
->arch
.cpuid_nent
;
912 e
->flags
&= ~KVM_CPUID_FLAG_STATE_READ_NEXT
;
913 /* when no next entry is found, the current entry[i] is reselected */
916 ej
= &vcpu
->arch
.cpuid_entries
[j
];
917 } while (ej
->function
!= e
->function
);
919 ej
->flags
|= KVM_CPUID_FLAG_STATE_READ_NEXT
;
924 /* find an entry with matching function, matching index (if needed), and that
925 * should be read next (if it's stateful) */
926 static int is_matching_cpuid_entry(struct kvm_cpuid_entry2
*e
,
927 u32 function
, u32 index
)
929 if (e
->function
!= function
)
931 if ((e
->flags
& KVM_CPUID_FLAG_SIGNIFCANT_INDEX
) && e
->index
!= index
)
933 if ((e
->flags
& KVM_CPUID_FLAG_STATEFUL_FUNC
) &&
934 !(e
->flags
& KVM_CPUID_FLAG_STATE_READ_NEXT
))
939 struct kvm_cpuid_entry2
*kvm_find_cpuid_entry(struct kvm_vcpu
*vcpu
,
940 u32 function
, u32 index
)
943 struct kvm_cpuid_entry2
*best
= NULL
;
945 for (i
= 0; i
< vcpu
->arch
.cpuid_nent
; ++i
) {
946 struct kvm_cpuid_entry2
*e
;
948 e
= &vcpu
->arch
.cpuid_entries
[i
];
949 if (is_matching_cpuid_entry(e
, function
, index
)) {
950 if (e
->flags
& KVM_CPUID_FLAG_STATEFUL_FUNC
)
951 move_to_next_stateful_cpuid_entry(vcpu
, i
);
958 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry
);
961 * If no match is found, check whether we exceed the vCPU's limit
962 * and return the content of the highest valid _standard_ leaf instead.
963 * This is to satisfy the CPUID specification.
965 static struct kvm_cpuid_entry2
* check_cpuid_limit(struct kvm_vcpu
*vcpu
,
966 u32 function
, u32 index
)
968 struct kvm_cpuid_entry2
*maxlevel
;
970 maxlevel
= kvm_find_cpuid_entry(vcpu
, function
& 0x80000000, 0);
971 if (!maxlevel
|| maxlevel
->eax
>= function
)
973 if (function
& 0x80000000) {
974 maxlevel
= kvm_find_cpuid_entry(vcpu
, 0, 0);
978 return kvm_find_cpuid_entry(vcpu
, maxlevel
->eax
, index
);
981 bool kvm_cpuid(struct kvm_vcpu
*vcpu
, u32
*eax
, u32
*ebx
,
982 u32
*ecx
, u32
*edx
, bool check_limit
)
984 u32 function
= *eax
, index
= *ecx
;
985 struct kvm_cpuid_entry2
*best
;
986 bool entry_found
= true;
988 best
= kvm_find_cpuid_entry(vcpu
, function
, index
);
995 best
= check_cpuid_limit(vcpu
, function
, index
);
1005 *eax
= *ebx
= *ecx
= *edx
= 0;
1006 trace_kvm_cpuid(function
, *eax
, *ebx
, *ecx
, *edx
, entry_found
);
1009 EXPORT_SYMBOL_GPL(kvm_cpuid
);
1011 int kvm_emulate_cpuid(struct kvm_vcpu
*vcpu
)
1013 u32 eax
, ebx
, ecx
, edx
;
1015 if (cpuid_fault_enabled(vcpu
) && !kvm_require_cpl(vcpu
, 0))
1018 eax
= kvm_rax_read(vcpu
);
1019 ecx
= kvm_rcx_read(vcpu
);
1020 kvm_cpuid(vcpu
, &eax
, &ebx
, &ecx
, &edx
, true);
1021 kvm_rax_write(vcpu
, eax
);
1022 kvm_rbx_write(vcpu
, ebx
);
1023 kvm_rcx_write(vcpu
, ecx
);
1024 kvm_rdx_write(vcpu
, edx
);
1025 return kvm_skip_emulated_instruction(vcpu
);
1027 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid
);