xref: /openbmc/linux/arch/x86/kvm/cpuid.c (revision 63705da3)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Kernel-based Virtual Machine driver for Linux
4  * cpuid support routines
5  *
6  * derived from arch/x86/kvm/x86.c
7  *
8  * Copyright 2011 Red Hat, Inc. and/or its affiliates.
9  * Copyright IBM Corporation, 2008
10  */
11 
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>
17 
18 #include <asm/processor.h>
19 #include <asm/user.h>
20 #include <asm/fpu/xstate.h>
21 #include <asm/sgx.h>
22 #include "cpuid.h"
23 #include "lapic.h"
24 #include "mmu.h"
25 #include "trace.h"
26 #include "pmu.h"
27 
28 /*
29  * Unlike "struct cpuinfo_x86.x86_capability", kvm_cpu_caps doesn't need to be
30  * aligned to sizeof(unsigned long) because it's not accessed via bitops.
31  */
32 u32 kvm_cpu_caps[NR_KVM_CPU_CAPS] __read_mostly;
33 EXPORT_SYMBOL_GPL(kvm_cpu_caps);
34 
35 static u32 xstate_required_size(u64 xstate_bv, bool compacted)
36 {
37 	int feature_bit = 0;
38 	u32 ret = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
39 
40 	xstate_bv &= XFEATURE_MASK_EXTEND;
41 	while (xstate_bv) {
42 		if (xstate_bv & 0x1) {
43 		        u32 eax, ebx, ecx, edx, offset;
44 		        cpuid_count(0xD, feature_bit, &eax, &ebx, &ecx, &edx);
45 			offset = compacted ? ret : ebx;
46 			ret = max(ret, offset + eax);
47 		}
48 
49 		xstate_bv >>= 1;
50 		feature_bit++;
51 	}
52 
53 	return ret;
54 }
55 
56 /*
57  * This one is tied to SSB in the user API, and not
58  * visible in /proc/cpuinfo.
59  */
60 #define KVM_X86_FEATURE_PSFD		(13*32+28) /* Predictive Store Forwarding Disable */
61 
62 #define F feature_bit
63 #define SF(name) (boot_cpu_has(X86_FEATURE_##name) ? F(name) : 0)
64 
65 
66 static inline struct kvm_cpuid_entry2 *cpuid_entry2_find(
67 	struct kvm_cpuid_entry2 *entries, int nent, u32 function, u32 index)
68 {
69 	struct kvm_cpuid_entry2 *e;
70 	int i;
71 
72 	for (i = 0; i < nent; i++) {
73 		e = &entries[i];
74 
75 		if (e->function == function &&
76 		    (!(e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) || e->index == index))
77 			return e;
78 	}
79 
80 	return NULL;
81 }
82 
83 static int kvm_check_cpuid(struct kvm_cpuid_entry2 *entries, int nent)
84 {
85 	struct kvm_cpuid_entry2 *best;
86 
87 	/*
88 	 * The existing code assumes virtual address is 48-bit or 57-bit in the
89 	 * canonical address checks; exit if it is ever changed.
90 	 */
91 	best = cpuid_entry2_find(entries, nent, 0x80000008, 0);
92 	if (best) {
93 		int vaddr_bits = (best->eax & 0xff00) >> 8;
94 
95 		if (vaddr_bits != 48 && vaddr_bits != 57 && vaddr_bits != 0)
96 			return -EINVAL;
97 	}
98 
99 	return 0;
100 }
101 
102 static void kvm_update_kvm_cpuid_base(struct kvm_vcpu *vcpu)
103 {
104 	u32 function;
105 	struct kvm_cpuid_entry2 *entry;
106 
107 	vcpu->arch.kvm_cpuid_base = 0;
108 
109 	for_each_possible_hypervisor_cpuid_base(function) {
110 		entry = kvm_find_cpuid_entry(vcpu, function, 0);
111 
112 		if (entry) {
113 			u32 signature[3];
114 
115 			signature[0] = entry->ebx;
116 			signature[1] = entry->ecx;
117 			signature[2] = entry->edx;
118 
119 			BUILD_BUG_ON(sizeof(signature) > sizeof(KVM_SIGNATURE));
120 			if (!memcmp(signature, KVM_SIGNATURE, sizeof(signature))) {
121 				vcpu->arch.kvm_cpuid_base = function;
122 				break;
123 			}
124 		}
125 	}
126 }
127 
128 struct kvm_cpuid_entry2 *kvm_find_kvm_cpuid_features(struct kvm_vcpu *vcpu)
129 {
130 	u32 base = vcpu->arch.kvm_cpuid_base;
131 
132 	if (!base)
133 		return NULL;
134 
135 	return kvm_find_cpuid_entry(vcpu, base | KVM_CPUID_FEATURES, 0);
136 }
137 
138 void kvm_update_pv_runtime(struct kvm_vcpu *vcpu)
139 {
140 	struct kvm_cpuid_entry2 *best = kvm_find_kvm_cpuid_features(vcpu);
141 
142 	/*
143 	 * save the feature bitmap to avoid cpuid lookup for every PV
144 	 * operation
145 	 */
146 	if (best)
147 		vcpu->arch.pv_cpuid.features = best->eax;
148 }
149 
150 void kvm_update_cpuid_runtime(struct kvm_vcpu *vcpu)
151 {
152 	struct kvm_cpuid_entry2 *best;
153 
154 	best = kvm_find_cpuid_entry(vcpu, 1, 0);
155 	if (best) {
156 		/* Update OSXSAVE bit */
157 		if (boot_cpu_has(X86_FEATURE_XSAVE))
158 			cpuid_entry_change(best, X86_FEATURE_OSXSAVE,
159 				   kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE));
160 
161 		cpuid_entry_change(best, X86_FEATURE_APIC,
162 			   vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE);
163 	}
164 
165 	best = kvm_find_cpuid_entry(vcpu, 7, 0);
166 	if (best && boot_cpu_has(X86_FEATURE_PKU) && best->function == 0x7)
167 		cpuid_entry_change(best, X86_FEATURE_OSPKE,
168 				   kvm_read_cr4_bits(vcpu, X86_CR4_PKE));
169 
170 	best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
171 	if (best)
172 		best->ebx = xstate_required_size(vcpu->arch.xcr0, false);
173 
174 	best = kvm_find_cpuid_entry(vcpu, 0xD, 1);
175 	if (best && (cpuid_entry_has(best, X86_FEATURE_XSAVES) ||
176 		     cpuid_entry_has(best, X86_FEATURE_XSAVEC)))
177 		best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
178 
179 	best = kvm_find_kvm_cpuid_features(vcpu);
180 	if (kvm_hlt_in_guest(vcpu->kvm) && best &&
181 		(best->eax & (1 << KVM_FEATURE_PV_UNHALT)))
182 		best->eax &= ~(1 << KVM_FEATURE_PV_UNHALT);
183 
184 	if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT)) {
185 		best = kvm_find_cpuid_entry(vcpu, 0x1, 0);
186 		if (best)
187 			cpuid_entry_change(best, X86_FEATURE_MWAIT,
188 					   vcpu->arch.ia32_misc_enable_msr &
189 					   MSR_IA32_MISC_ENABLE_MWAIT);
190 	}
191 }
192 EXPORT_SYMBOL_GPL(kvm_update_cpuid_runtime);
193 
194 static void kvm_vcpu_after_set_cpuid(struct kvm_vcpu *vcpu)
195 {
196 	struct kvm_lapic *apic = vcpu->arch.apic;
197 	struct kvm_cpuid_entry2 *best;
198 
199 	best = kvm_find_cpuid_entry(vcpu, 1, 0);
200 	if (best && apic) {
201 		if (cpuid_entry_has(best, X86_FEATURE_TSC_DEADLINE_TIMER))
202 			apic->lapic_timer.timer_mode_mask = 3 << 17;
203 		else
204 			apic->lapic_timer.timer_mode_mask = 1 << 17;
205 
206 		kvm_apic_set_version(vcpu);
207 	}
208 
209 	best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
210 	if (!best)
211 		vcpu->arch.guest_supported_xcr0 = 0;
212 	else
213 		vcpu->arch.guest_supported_xcr0 =
214 			(best->eax | ((u64)best->edx << 32)) & supported_xcr0;
215 
216 	/*
217 	 * Bits 127:0 of the allowed SECS.ATTRIBUTES (CPUID.0x12.0x1) enumerate
218 	 * the supported XSAVE Feature Request Mask (XFRM), i.e. the enclave's
219 	 * requested XCR0 value.  The enclave's XFRM must be a subset of XCRO
220 	 * at the time of EENTER, thus adjust the allowed XFRM by the guest's
221 	 * supported XCR0.  Similar to XCR0 handling, FP and SSE are forced to
222 	 * '1' even on CPUs that don't support XSAVE.
223 	 */
224 	best = kvm_find_cpuid_entry(vcpu, 0x12, 0x1);
225 	if (best) {
226 		best->ecx &= vcpu->arch.guest_supported_xcr0 & 0xffffffff;
227 		best->edx &= vcpu->arch.guest_supported_xcr0 >> 32;
228 		best->ecx |= XFEATURE_MASK_FPSSE;
229 	}
230 
231 	kvm_update_pv_runtime(vcpu);
232 
233 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
234 	vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu);
235 
236 	kvm_pmu_refresh(vcpu);
237 	vcpu->arch.cr4_guest_rsvd_bits =
238 	    __cr4_reserved_bits(guest_cpuid_has, vcpu);
239 
240 	kvm_hv_set_cpuid(vcpu);
241 
242 	/* Invoke the vendor callback only after the above state is updated. */
243 	static_call(kvm_x86_vcpu_after_set_cpuid)(vcpu);
244 
245 	/*
246 	 * Except for the MMU, which needs to do its thing any vendor specific
247 	 * adjustments to the reserved GPA bits.
248 	 */
249 	kvm_mmu_after_set_cpuid(vcpu);
250 }
251 
252 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
253 {
254 	struct kvm_cpuid_entry2 *best;
255 
256 	best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
257 	if (!best || best->eax < 0x80000008)
258 		goto not_found;
259 	best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
260 	if (best)
261 		return best->eax & 0xff;
262 not_found:
263 	return 36;
264 }
265 
266 /*
267  * This "raw" version returns the reserved GPA bits without any adjustments for
268  * encryption technologies that usurp bits.  The raw mask should be used if and
269  * only if hardware does _not_ strip the usurped bits, e.g. in virtual MTRRs.
270  */
271 u64 kvm_vcpu_reserved_gpa_bits_raw(struct kvm_vcpu *vcpu)
272 {
273 	return rsvd_bits(cpuid_maxphyaddr(vcpu), 63);
274 }
275 
276 static int kvm_set_cpuid(struct kvm_vcpu *vcpu, struct kvm_cpuid_entry2 *e2,
277                         int nent)
278 {
279     int r;
280 
281     r = kvm_check_cpuid(e2, nent);
282     if (r)
283         return r;
284 
285     kvfree(vcpu->arch.cpuid_entries);
286     vcpu->arch.cpuid_entries = e2;
287     vcpu->arch.cpuid_nent = nent;
288 
289     kvm_update_kvm_cpuid_base(vcpu);
290     kvm_update_cpuid_runtime(vcpu);
291     kvm_vcpu_after_set_cpuid(vcpu);
292 
293     return 0;
294 }
295 
296 /* when an old userspace process fills a new kernel module */
297 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
298 			     struct kvm_cpuid *cpuid,
299 			     struct kvm_cpuid_entry __user *entries)
300 {
301 	int r, i;
302 	struct kvm_cpuid_entry *e = NULL;
303 	struct kvm_cpuid_entry2 *e2 = NULL;
304 
305 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
306 		return -E2BIG;
307 
308 	if (cpuid->nent) {
309 		e = vmemdup_user(entries, array_size(sizeof(*e), cpuid->nent));
310 		if (IS_ERR(e))
311 			return PTR_ERR(e);
312 
313 		e2 = kvmalloc_array(cpuid->nent, sizeof(*e2), GFP_KERNEL_ACCOUNT);
314 		if (!e2) {
315 			r = -ENOMEM;
316 			goto out_free_cpuid;
317 		}
318 	}
319 	for (i = 0; i < cpuid->nent; i++) {
320 		e2[i].function = e[i].function;
321 		e2[i].eax = e[i].eax;
322 		e2[i].ebx = e[i].ebx;
323 		e2[i].ecx = e[i].ecx;
324 		e2[i].edx = e[i].edx;
325 		e2[i].index = 0;
326 		e2[i].flags = 0;
327 		e2[i].padding[0] = 0;
328 		e2[i].padding[1] = 0;
329 		e2[i].padding[2] = 0;
330 	}
331 
332 	r = kvm_set_cpuid(vcpu, e2, cpuid->nent);
333 	if (r)
334 		kvfree(e2);
335 
336 out_free_cpuid:
337 	kvfree(e);
338 
339 	return r;
340 }
341 
342 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
343 			      struct kvm_cpuid2 *cpuid,
344 			      struct kvm_cpuid_entry2 __user *entries)
345 {
346 	struct kvm_cpuid_entry2 *e2 = NULL;
347 	int r;
348 
349 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
350 		return -E2BIG;
351 
352 	if (cpuid->nent) {
353 		e2 = vmemdup_user(entries, array_size(sizeof(*e2), cpuid->nent));
354 		if (IS_ERR(e2))
355 			return PTR_ERR(e2);
356 	}
357 
358 	r = kvm_set_cpuid(vcpu, e2, cpuid->nent);
359 	if (r)
360 		kvfree(e2);
361 
362 	return r;
363 }
364 
365 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
366 			      struct kvm_cpuid2 *cpuid,
367 			      struct kvm_cpuid_entry2 __user *entries)
368 {
369 	int r;
370 
371 	r = -E2BIG;
372 	if (cpuid->nent < vcpu->arch.cpuid_nent)
373 		goto out;
374 	r = -EFAULT;
375 	if (copy_to_user(entries, vcpu->arch.cpuid_entries,
376 			 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
377 		goto out;
378 	return 0;
379 
380 out:
381 	cpuid->nent = vcpu->arch.cpuid_nent;
382 	return r;
383 }
384 
385 /* Mask kvm_cpu_caps for @leaf with the raw CPUID capabilities of this CPU. */
386 static __always_inline void __kvm_cpu_cap_mask(unsigned int leaf)
387 {
388 	const struct cpuid_reg cpuid = x86_feature_cpuid(leaf * 32);
389 	struct kvm_cpuid_entry2 entry;
390 
391 	reverse_cpuid_check(leaf);
392 
393 	cpuid_count(cpuid.function, cpuid.index,
394 		    &entry.eax, &entry.ebx, &entry.ecx, &entry.edx);
395 
396 	kvm_cpu_caps[leaf] &= *__cpuid_entry_get_reg(&entry, cpuid.reg);
397 }
398 
399 static __always_inline
400 void kvm_cpu_cap_init_scattered(enum kvm_only_cpuid_leafs leaf, u32 mask)
401 {
402 	/* Use kvm_cpu_cap_mask for non-scattered leafs. */
403 	BUILD_BUG_ON(leaf < NCAPINTS);
404 
405 	kvm_cpu_caps[leaf] = mask;
406 
407 	__kvm_cpu_cap_mask(leaf);
408 }
409 
410 static __always_inline void kvm_cpu_cap_mask(enum cpuid_leafs leaf, u32 mask)
411 {
412 	/* Use kvm_cpu_cap_init_scattered for scattered leafs. */
413 	BUILD_BUG_ON(leaf >= NCAPINTS);
414 
415 	kvm_cpu_caps[leaf] &= mask;
416 
417 	__kvm_cpu_cap_mask(leaf);
418 }
419 
420 void kvm_set_cpu_caps(void)
421 {
422 #ifdef CONFIG_X86_64
423 	unsigned int f_gbpages = F(GBPAGES);
424 	unsigned int f_lm = F(LM);
425 #else
426 	unsigned int f_gbpages = 0;
427 	unsigned int f_lm = 0;
428 #endif
429 	memset(kvm_cpu_caps, 0, sizeof(kvm_cpu_caps));
430 
431 	BUILD_BUG_ON(sizeof(kvm_cpu_caps) - (NKVMCAPINTS * sizeof(*kvm_cpu_caps)) >
432 		     sizeof(boot_cpu_data.x86_capability));
433 
434 	memcpy(&kvm_cpu_caps, &boot_cpu_data.x86_capability,
435 	       sizeof(kvm_cpu_caps) - (NKVMCAPINTS * sizeof(*kvm_cpu_caps)));
436 
437 	kvm_cpu_cap_mask(CPUID_1_ECX,
438 		/*
439 		 * NOTE: MONITOR (and MWAIT) are emulated as NOP, but *not*
440 		 * advertised to guests via CPUID!
441 		 */
442 		F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
443 		0 /* DS-CPL, VMX, SMX, EST */ |
444 		0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
445 		F(FMA) | F(CX16) | 0 /* xTPR Update */ | F(PDCM) |
446 		F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
447 		F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
448 		0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
449 		F(F16C) | F(RDRAND)
450 	);
451 	/* KVM emulates x2apic in software irrespective of host support. */
452 	kvm_cpu_cap_set(X86_FEATURE_X2APIC);
453 
454 	kvm_cpu_cap_mask(CPUID_1_EDX,
455 		F(FPU) | F(VME) | F(DE) | F(PSE) |
456 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
457 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
458 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
459 		F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
460 		0 /* Reserved, DS, ACPI */ | F(MMX) |
461 		F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
462 		0 /* HTT, TM, Reserved, PBE */
463 	);
464 
465 	kvm_cpu_cap_mask(CPUID_7_0_EBX,
466 		F(FSGSBASE) | F(SGX) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
467 		F(BMI2) | F(ERMS) | F(INVPCID) | F(RTM) | 0 /*MPX*/ | F(RDSEED) |
468 		F(ADX) | F(SMAP) | F(AVX512IFMA) | F(AVX512F) | F(AVX512PF) |
469 		F(AVX512ER) | F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) |
470 		F(SHA_NI) | F(AVX512BW) | F(AVX512VL) | 0 /*INTEL_PT*/
471 	);
472 
473 	kvm_cpu_cap_mask(CPUID_7_ECX,
474 		F(AVX512VBMI) | F(LA57) | F(PKU) | 0 /*OSPKE*/ | F(RDPID) |
475 		F(AVX512_VPOPCNTDQ) | F(UMIP) | F(AVX512_VBMI2) | F(GFNI) |
476 		F(VAES) | F(VPCLMULQDQ) | F(AVX512_VNNI) | F(AVX512_BITALG) |
477 		F(CLDEMOTE) | F(MOVDIRI) | F(MOVDIR64B) | 0 /*WAITPKG*/ |
478 		F(SGX_LC) | F(BUS_LOCK_DETECT)
479 	);
480 	/* Set LA57 based on hardware capability. */
481 	if (cpuid_ecx(7) & F(LA57))
482 		kvm_cpu_cap_set(X86_FEATURE_LA57);
483 
484 	/*
485 	 * PKU not yet implemented for shadow paging and requires OSPKE
486 	 * to be set on the host. Clear it if that is not the case
487 	 */
488 	if (!tdp_enabled || !boot_cpu_has(X86_FEATURE_OSPKE))
489 		kvm_cpu_cap_clear(X86_FEATURE_PKU);
490 
491 	kvm_cpu_cap_mask(CPUID_7_EDX,
492 		F(AVX512_4VNNIW) | F(AVX512_4FMAPS) | F(SPEC_CTRL) |
493 		F(SPEC_CTRL_SSBD) | F(ARCH_CAPABILITIES) | F(INTEL_STIBP) |
494 		F(MD_CLEAR) | F(AVX512_VP2INTERSECT) | F(FSRM) |
495 		F(SERIALIZE) | F(TSXLDTRK) | F(AVX512_FP16)
496 	);
497 
498 	/* TSC_ADJUST and ARCH_CAPABILITIES are emulated in software. */
499 	kvm_cpu_cap_set(X86_FEATURE_TSC_ADJUST);
500 	kvm_cpu_cap_set(X86_FEATURE_ARCH_CAPABILITIES);
501 
502 	if (boot_cpu_has(X86_FEATURE_IBPB) && boot_cpu_has(X86_FEATURE_IBRS))
503 		kvm_cpu_cap_set(X86_FEATURE_SPEC_CTRL);
504 	if (boot_cpu_has(X86_FEATURE_STIBP))
505 		kvm_cpu_cap_set(X86_FEATURE_INTEL_STIBP);
506 	if (boot_cpu_has(X86_FEATURE_AMD_SSBD))
507 		kvm_cpu_cap_set(X86_FEATURE_SPEC_CTRL_SSBD);
508 
509 	kvm_cpu_cap_mask(CPUID_7_1_EAX,
510 		F(AVX_VNNI) | F(AVX512_BF16)
511 	);
512 
513 	kvm_cpu_cap_mask(CPUID_D_1_EAX,
514 		F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | F(XSAVES)
515 	);
516 
517 	kvm_cpu_cap_init_scattered(CPUID_12_EAX,
518 		SF(SGX1) | SF(SGX2)
519 	);
520 
521 	kvm_cpu_cap_mask(CPUID_8000_0001_ECX,
522 		F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
523 		F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
524 		F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
525 		0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM) |
526 		F(TOPOEXT) | F(PERFCTR_CORE)
527 	);
528 
529 	kvm_cpu_cap_mask(CPUID_8000_0001_EDX,
530 		F(FPU) | F(VME) | F(DE) | F(PSE) |
531 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
532 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
533 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
534 		F(PAT) | F(PSE36) | 0 /* Reserved */ |
535 		F(NX) | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
536 		F(FXSR) | F(FXSR_OPT) | f_gbpages | F(RDTSCP) |
537 		0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW)
538 	);
539 
540 	if (!tdp_enabled && IS_ENABLED(CONFIG_X86_64))
541 		kvm_cpu_cap_set(X86_FEATURE_GBPAGES);
542 
543 	kvm_cpu_cap_mask(CPUID_8000_0008_EBX,
544 		F(CLZERO) | F(XSAVEERPTR) |
545 		F(WBNOINVD) | F(AMD_IBPB) | F(AMD_IBRS) | F(AMD_SSBD) | F(VIRT_SSBD) |
546 		F(AMD_SSB_NO) | F(AMD_STIBP) | F(AMD_STIBP_ALWAYS_ON) |
547 		__feature_bit(KVM_X86_FEATURE_PSFD)
548 	);
549 
550 	/*
551 	 * AMD has separate bits for each SPEC_CTRL bit.
552 	 * arch/x86/kernel/cpu/bugs.c is kind enough to
553 	 * record that in cpufeatures so use them.
554 	 */
555 	if (boot_cpu_has(X86_FEATURE_IBPB))
556 		kvm_cpu_cap_set(X86_FEATURE_AMD_IBPB);
557 	if (boot_cpu_has(X86_FEATURE_IBRS))
558 		kvm_cpu_cap_set(X86_FEATURE_AMD_IBRS);
559 	if (boot_cpu_has(X86_FEATURE_STIBP))
560 		kvm_cpu_cap_set(X86_FEATURE_AMD_STIBP);
561 	if (boot_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD))
562 		kvm_cpu_cap_set(X86_FEATURE_AMD_SSBD);
563 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
564 		kvm_cpu_cap_set(X86_FEATURE_AMD_SSB_NO);
565 	/*
566 	 * The preference is to use SPEC CTRL MSR instead of the
567 	 * VIRT_SPEC MSR.
568 	 */
569 	if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD) &&
570 	    !boot_cpu_has(X86_FEATURE_AMD_SSBD))
571 		kvm_cpu_cap_set(X86_FEATURE_VIRT_SSBD);
572 
573 	/*
574 	 * Hide all SVM features by default, SVM will set the cap bits for
575 	 * features it emulates and/or exposes for L1.
576 	 */
577 	kvm_cpu_cap_mask(CPUID_8000_000A_EDX, 0);
578 
579 	kvm_cpu_cap_mask(CPUID_8000_001F_EAX,
580 		0 /* SME */ | F(SEV) | 0 /* VM_PAGE_FLUSH */ | F(SEV_ES) |
581 		F(SME_COHERENT));
582 
583 	kvm_cpu_cap_mask(CPUID_C000_0001_EDX,
584 		F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
585 		F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
586 		F(PMM) | F(PMM_EN)
587 	);
588 
589 	/*
590 	 * Hide RDTSCP and RDPID if either feature is reported as supported but
591 	 * probing MSR_TSC_AUX failed.  This is purely a sanity check and
592 	 * should never happen, but the guest will likely crash if RDTSCP or
593 	 * RDPID is misreported, and KVM has botched MSR_TSC_AUX emulation in
594 	 * the past.  For example, the sanity check may fire if this instance of
595 	 * KVM is running as L1 on top of an older, broken KVM.
596 	 */
597 	if (WARN_ON((kvm_cpu_cap_has(X86_FEATURE_RDTSCP) ||
598 		     kvm_cpu_cap_has(X86_FEATURE_RDPID)) &&
599 		     !kvm_is_supported_user_return_msr(MSR_TSC_AUX))) {
600 		kvm_cpu_cap_clear(X86_FEATURE_RDTSCP);
601 		kvm_cpu_cap_clear(X86_FEATURE_RDPID);
602 	}
603 }
604 EXPORT_SYMBOL_GPL(kvm_set_cpu_caps);
605 
606 struct kvm_cpuid_array {
607 	struct kvm_cpuid_entry2 *entries;
608 	int maxnent;
609 	int nent;
610 };
611 
612 static struct kvm_cpuid_entry2 *do_host_cpuid(struct kvm_cpuid_array *array,
613 					      u32 function, u32 index)
614 {
615 	struct kvm_cpuid_entry2 *entry;
616 
617 	if (array->nent >= array->maxnent)
618 		return NULL;
619 
620 	entry = &array->entries[array->nent++];
621 
622 	entry->function = function;
623 	entry->index = index;
624 	entry->flags = 0;
625 
626 	cpuid_count(entry->function, entry->index,
627 		    &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
628 
629 	switch (function) {
630 	case 4:
631 	case 7:
632 	case 0xb:
633 	case 0xd:
634 	case 0xf:
635 	case 0x10:
636 	case 0x12:
637 	case 0x14:
638 	case 0x17:
639 	case 0x18:
640 	case 0x1f:
641 	case 0x8000001d:
642 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
643 		break;
644 	}
645 
646 	return entry;
647 }
648 
649 static int __do_cpuid_func_emulated(struct kvm_cpuid_array *array, u32 func)
650 {
651 	struct kvm_cpuid_entry2 *entry;
652 
653 	if (array->nent >= array->maxnent)
654 		return -E2BIG;
655 
656 	entry = &array->entries[array->nent];
657 	entry->function = func;
658 	entry->index = 0;
659 	entry->flags = 0;
660 
661 	switch (func) {
662 	case 0:
663 		entry->eax = 7;
664 		++array->nent;
665 		break;
666 	case 1:
667 		entry->ecx = F(MOVBE);
668 		++array->nent;
669 		break;
670 	case 7:
671 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
672 		entry->eax = 0;
673 		if (kvm_cpu_cap_has(X86_FEATURE_RDTSCP))
674 			entry->ecx = F(RDPID);
675 		++array->nent;
676 		break;
677 	default:
678 		break;
679 	}
680 
681 	return 0;
682 }
683 
684 static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function)
685 {
686 	struct kvm_cpuid_entry2 *entry;
687 	int r, i, max_idx;
688 
689 	/* all calls to cpuid_count() should be made on the same cpu */
690 	get_cpu();
691 
692 	r = -E2BIG;
693 
694 	entry = do_host_cpuid(array, function, 0);
695 	if (!entry)
696 		goto out;
697 
698 	switch (function) {
699 	case 0:
700 		/* Limited to the highest leaf implemented in KVM. */
701 		entry->eax = min(entry->eax, 0x1fU);
702 		break;
703 	case 1:
704 		cpuid_entry_override(entry, CPUID_1_EDX);
705 		cpuid_entry_override(entry, CPUID_1_ECX);
706 		break;
707 	case 2:
708 		/*
709 		 * On ancient CPUs, function 2 entries are STATEFUL.  That is,
710 		 * CPUID(function=2, index=0) may return different results each
711 		 * time, with the least-significant byte in EAX enumerating the
712 		 * number of times software should do CPUID(2, 0).
713 		 *
714 		 * Modern CPUs, i.e. every CPU KVM has *ever* run on are less
715 		 * idiotic.  Intel's SDM states that EAX & 0xff "will always
716 		 * return 01H. Software should ignore this value and not
717 		 * interpret it as an informational descriptor", while AMD's
718 		 * APM states that CPUID(2) is reserved.
719 		 *
720 		 * WARN if a frankenstein CPU that supports virtualization and
721 		 * a stateful CPUID.0x2 is encountered.
722 		 */
723 		WARN_ON_ONCE((entry->eax & 0xff) > 1);
724 		break;
725 	/* functions 4 and 0x8000001d have additional index. */
726 	case 4:
727 	case 0x8000001d:
728 		/*
729 		 * Read entries until the cache type in the previous entry is
730 		 * zero, i.e. indicates an invalid entry.
731 		 */
732 		for (i = 1; entry->eax & 0x1f; ++i) {
733 			entry = do_host_cpuid(array, function, i);
734 			if (!entry)
735 				goto out;
736 		}
737 		break;
738 	case 6: /* Thermal management */
739 		entry->eax = 0x4; /* allow ARAT */
740 		entry->ebx = 0;
741 		entry->ecx = 0;
742 		entry->edx = 0;
743 		break;
744 	/* function 7 has additional index. */
745 	case 7:
746 		entry->eax = min(entry->eax, 1u);
747 		cpuid_entry_override(entry, CPUID_7_0_EBX);
748 		cpuid_entry_override(entry, CPUID_7_ECX);
749 		cpuid_entry_override(entry, CPUID_7_EDX);
750 
751 		/* KVM only supports 0x7.0 and 0x7.1, capped above via min(). */
752 		if (entry->eax == 1) {
753 			entry = do_host_cpuid(array, function, 1);
754 			if (!entry)
755 				goto out;
756 
757 			cpuid_entry_override(entry, CPUID_7_1_EAX);
758 			entry->ebx = 0;
759 			entry->ecx = 0;
760 			entry->edx = 0;
761 		}
762 		break;
763 	case 9:
764 		break;
765 	case 0xa: { /* Architectural Performance Monitoring */
766 		struct x86_pmu_capability cap;
767 		union cpuid10_eax eax;
768 		union cpuid10_edx edx;
769 
770 		perf_get_x86_pmu_capability(&cap);
771 
772 		/*
773 		 * Only support guest architectural pmu on a host
774 		 * with architectural pmu.
775 		 */
776 		if (!cap.version)
777 			memset(&cap, 0, sizeof(cap));
778 
779 		eax.split.version_id = min(cap.version, 2);
780 		eax.split.num_counters = cap.num_counters_gp;
781 		eax.split.bit_width = cap.bit_width_gp;
782 		eax.split.mask_length = cap.events_mask_len;
783 
784 		edx.split.num_counters_fixed = min(cap.num_counters_fixed, MAX_FIXED_COUNTERS);
785 		edx.split.bit_width_fixed = cap.bit_width_fixed;
786 		if (cap.version)
787 			edx.split.anythread_deprecated = 1;
788 		edx.split.reserved1 = 0;
789 		edx.split.reserved2 = 0;
790 
791 		entry->eax = eax.full;
792 		entry->ebx = cap.events_mask;
793 		entry->ecx = 0;
794 		entry->edx = edx.full;
795 		break;
796 	}
797 	/*
798 	 * Per Intel's SDM, the 0x1f is a superset of 0xb,
799 	 * thus they can be handled by common code.
800 	 */
801 	case 0x1f:
802 	case 0xb:
803 		/*
804 		 * Populate entries until the level type (ECX[15:8]) of the
805 		 * previous entry is zero.  Note, CPUID EAX.{0x1f,0xb}.0 is
806 		 * the starting entry, filled by the primary do_host_cpuid().
807 		 */
808 		for (i = 1; entry->ecx & 0xff00; ++i) {
809 			entry = do_host_cpuid(array, function, i);
810 			if (!entry)
811 				goto out;
812 		}
813 		break;
814 	case 0xd:
815 		entry->eax &= supported_xcr0;
816 		entry->ebx = xstate_required_size(supported_xcr0, false);
817 		entry->ecx = entry->ebx;
818 		entry->edx &= supported_xcr0 >> 32;
819 		if (!supported_xcr0)
820 			break;
821 
822 		entry = do_host_cpuid(array, function, 1);
823 		if (!entry)
824 			goto out;
825 
826 		cpuid_entry_override(entry, CPUID_D_1_EAX);
827 		if (entry->eax & (F(XSAVES)|F(XSAVEC)))
828 			entry->ebx = xstate_required_size(supported_xcr0 | supported_xss,
829 							  true);
830 		else {
831 			WARN_ON_ONCE(supported_xss != 0);
832 			entry->ebx = 0;
833 		}
834 		entry->ecx &= supported_xss;
835 		entry->edx &= supported_xss >> 32;
836 
837 		for (i = 2; i < 64; ++i) {
838 			bool s_state;
839 			if (supported_xcr0 & BIT_ULL(i))
840 				s_state = false;
841 			else if (supported_xss & BIT_ULL(i))
842 				s_state = true;
843 			else
844 				continue;
845 
846 			entry = do_host_cpuid(array, function, i);
847 			if (!entry)
848 				goto out;
849 
850 			/*
851 			 * The supported check above should have filtered out
852 			 * invalid sub-leafs.  Only valid sub-leafs should
853 			 * reach this point, and they should have a non-zero
854 			 * save state size.  Furthermore, check whether the
855 			 * processor agrees with supported_xcr0/supported_xss
856 			 * on whether this is an XCR0- or IA32_XSS-managed area.
857 			 */
858 			if (WARN_ON_ONCE(!entry->eax || (entry->ecx & 0x1) != s_state)) {
859 				--array->nent;
860 				continue;
861 			}
862 			entry->edx = 0;
863 		}
864 		break;
865 	case 0x12:
866 		/* Intel SGX */
867 		if (!kvm_cpu_cap_has(X86_FEATURE_SGX)) {
868 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
869 			break;
870 		}
871 
872 		/*
873 		 * Index 0: Sub-features, MISCSELECT (a.k.a extended features)
874 		 * and max enclave sizes.   The SGX sub-features and MISCSELECT
875 		 * are restricted by kernel and KVM capabilities (like most
876 		 * feature flags), while enclave size is unrestricted.
877 		 */
878 		cpuid_entry_override(entry, CPUID_12_EAX);
879 		entry->ebx &= SGX_MISC_EXINFO;
880 
881 		entry = do_host_cpuid(array, function, 1);
882 		if (!entry)
883 			goto out;
884 
885 		/*
886 		 * Index 1: SECS.ATTRIBUTES.  ATTRIBUTES are restricted a la
887 		 * feature flags.  Advertise all supported flags, including
888 		 * privileged attributes that require explicit opt-in from
889 		 * userspace.  ATTRIBUTES.XFRM is not adjusted as userspace is
890 		 * expected to derive it from supported XCR0.
891 		 */
892 		entry->eax &= SGX_ATTR_DEBUG | SGX_ATTR_MODE64BIT |
893 			      SGX_ATTR_PROVISIONKEY | SGX_ATTR_EINITTOKENKEY |
894 			      SGX_ATTR_KSS;
895 		entry->ebx &= 0;
896 		break;
897 	/* Intel PT */
898 	case 0x14:
899 		if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT)) {
900 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
901 			break;
902 		}
903 
904 		for (i = 1, max_idx = entry->eax; i <= max_idx; ++i) {
905 			if (!do_host_cpuid(array, function, i))
906 				goto out;
907 		}
908 		break;
909 	case KVM_CPUID_SIGNATURE: {
910 		const u32 *sigptr = (const u32 *)KVM_SIGNATURE;
911 		entry->eax = KVM_CPUID_FEATURES;
912 		entry->ebx = sigptr[0];
913 		entry->ecx = sigptr[1];
914 		entry->edx = sigptr[2];
915 		break;
916 	}
917 	case KVM_CPUID_FEATURES:
918 		entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
919 			     (1 << KVM_FEATURE_NOP_IO_DELAY) |
920 			     (1 << KVM_FEATURE_CLOCKSOURCE2) |
921 			     (1 << KVM_FEATURE_ASYNC_PF) |
922 			     (1 << KVM_FEATURE_PV_EOI) |
923 			     (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
924 			     (1 << KVM_FEATURE_PV_UNHALT) |
925 			     (1 << KVM_FEATURE_PV_TLB_FLUSH) |
926 			     (1 << KVM_FEATURE_ASYNC_PF_VMEXIT) |
927 			     (1 << KVM_FEATURE_PV_SEND_IPI) |
928 			     (1 << KVM_FEATURE_POLL_CONTROL) |
929 			     (1 << KVM_FEATURE_PV_SCHED_YIELD) |
930 			     (1 << KVM_FEATURE_ASYNC_PF_INT);
931 
932 		if (sched_info_on())
933 			entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
934 
935 		entry->ebx = 0;
936 		entry->ecx = 0;
937 		entry->edx = 0;
938 		break;
939 	case 0x80000000:
940 		entry->eax = min(entry->eax, 0x8000001f);
941 		break;
942 	case 0x80000001:
943 		cpuid_entry_override(entry, CPUID_8000_0001_EDX);
944 		cpuid_entry_override(entry, CPUID_8000_0001_ECX);
945 		break;
946 	case 0x80000006:
947 		/* L2 cache and TLB: pass through host info. */
948 		break;
949 	case 0x80000007: /* Advanced power management */
950 		/* invariant TSC is CPUID.80000007H:EDX[8] */
951 		entry->edx &= (1 << 8);
952 		/* mask against host */
953 		entry->edx &= boot_cpu_data.x86_power;
954 		entry->eax = entry->ebx = entry->ecx = 0;
955 		break;
956 	case 0x80000008: {
957 		unsigned g_phys_as = (entry->eax >> 16) & 0xff;
958 		unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
959 		unsigned phys_as = entry->eax & 0xff;
960 
961 		/*
962 		 * If TDP (NPT) is disabled use the adjusted host MAXPHYADDR as
963 		 * the guest operates in the same PA space as the host, i.e.
964 		 * reductions in MAXPHYADDR for memory encryption affect shadow
965 		 * paging, too.
966 		 *
967 		 * If TDP is enabled but an explicit guest MAXPHYADDR is not
968 		 * provided, use the raw bare metal MAXPHYADDR as reductions to
969 		 * the HPAs do not affect GPAs.
970 		 */
971 		if (!tdp_enabled)
972 			g_phys_as = boot_cpu_data.x86_phys_bits;
973 		else if (!g_phys_as)
974 			g_phys_as = phys_as;
975 
976 		entry->eax = g_phys_as | (virt_as << 8);
977 		entry->edx = 0;
978 		cpuid_entry_override(entry, CPUID_8000_0008_EBX);
979 		break;
980 	}
981 	case 0x8000000A:
982 		if (!kvm_cpu_cap_has(X86_FEATURE_SVM)) {
983 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
984 			break;
985 		}
986 		entry->eax = 1; /* SVM revision 1 */
987 		entry->ebx = 8; /* Lets support 8 ASIDs in case we add proper
988 				   ASID emulation to nested SVM */
989 		entry->ecx = 0; /* Reserved */
990 		cpuid_entry_override(entry, CPUID_8000_000A_EDX);
991 		break;
992 	case 0x80000019:
993 		entry->ecx = entry->edx = 0;
994 		break;
995 	case 0x8000001a:
996 	case 0x8000001e:
997 		break;
998 	case 0x8000001F:
999 		if (!kvm_cpu_cap_has(X86_FEATURE_SEV)) {
1000 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
1001 		} else {
1002 			cpuid_entry_override(entry, CPUID_8000_001F_EAX);
1003 
1004 			/*
1005 			 * Enumerate '0' for "PA bits reduction", the adjusted
1006 			 * MAXPHYADDR is enumerated directly (see 0x80000008).
1007 			 */
1008 			entry->ebx &= ~GENMASK(11, 6);
1009 		}
1010 		break;
1011 	/*Add support for Centaur's CPUID instruction*/
1012 	case 0xC0000000:
1013 		/*Just support up to 0xC0000004 now*/
1014 		entry->eax = min(entry->eax, 0xC0000004);
1015 		break;
1016 	case 0xC0000001:
1017 		cpuid_entry_override(entry, CPUID_C000_0001_EDX);
1018 		break;
1019 	case 3: /* Processor serial number */
1020 	case 5: /* MONITOR/MWAIT */
1021 	case 0xC0000002:
1022 	case 0xC0000003:
1023 	case 0xC0000004:
1024 	default:
1025 		entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
1026 		break;
1027 	}
1028 
1029 	r = 0;
1030 
1031 out:
1032 	put_cpu();
1033 
1034 	return r;
1035 }
1036 
1037 static int do_cpuid_func(struct kvm_cpuid_array *array, u32 func,
1038 			 unsigned int type)
1039 {
1040 	if (type == KVM_GET_EMULATED_CPUID)
1041 		return __do_cpuid_func_emulated(array, func);
1042 
1043 	return __do_cpuid_func(array, func);
1044 }
1045 
1046 #define CENTAUR_CPUID_SIGNATURE 0xC0000000
1047 
1048 static int get_cpuid_func(struct kvm_cpuid_array *array, u32 func,
1049 			  unsigned int type)
1050 {
1051 	u32 limit;
1052 	int r;
1053 
1054 	if (func == CENTAUR_CPUID_SIGNATURE &&
1055 	    boot_cpu_data.x86_vendor != X86_VENDOR_CENTAUR)
1056 		return 0;
1057 
1058 	r = do_cpuid_func(array, func, type);
1059 	if (r)
1060 		return r;
1061 
1062 	limit = array->entries[array->nent - 1].eax;
1063 	for (func = func + 1; func <= limit; ++func) {
1064 		r = do_cpuid_func(array, func, type);
1065 		if (r)
1066 			break;
1067 	}
1068 
1069 	return r;
1070 }
1071 
1072 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
1073 				 __u32 num_entries, unsigned int ioctl_type)
1074 {
1075 	int i;
1076 	__u32 pad[3];
1077 
1078 	if (ioctl_type != KVM_GET_EMULATED_CPUID)
1079 		return false;
1080 
1081 	/*
1082 	 * We want to make sure that ->padding is being passed clean from
1083 	 * userspace in case we want to use it for something in the future.
1084 	 *
1085 	 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
1086 	 * have to give ourselves satisfied only with the emulated side. /me
1087 	 * sheds a tear.
1088 	 */
1089 	for (i = 0; i < num_entries; i++) {
1090 		if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
1091 			return true;
1092 
1093 		if (pad[0] || pad[1] || pad[2])
1094 			return true;
1095 	}
1096 	return false;
1097 }
1098 
1099 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
1100 			    struct kvm_cpuid_entry2 __user *entries,
1101 			    unsigned int type)
1102 {
1103 	static const u32 funcs[] = {
1104 		0, 0x80000000, CENTAUR_CPUID_SIGNATURE, KVM_CPUID_SIGNATURE,
1105 	};
1106 
1107 	struct kvm_cpuid_array array = {
1108 		.nent = 0,
1109 	};
1110 	int r, i;
1111 
1112 	if (cpuid->nent < 1)
1113 		return -E2BIG;
1114 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1115 		cpuid->nent = KVM_MAX_CPUID_ENTRIES;
1116 
1117 	if (sanity_check_entries(entries, cpuid->nent, type))
1118 		return -EINVAL;
1119 
1120 	array.entries = vzalloc(array_size(sizeof(struct kvm_cpuid_entry2),
1121 					   cpuid->nent));
1122 	if (!array.entries)
1123 		return -ENOMEM;
1124 
1125 	array.maxnent = cpuid->nent;
1126 
1127 	for (i = 0; i < ARRAY_SIZE(funcs); i++) {
1128 		r = get_cpuid_func(&array, funcs[i], type);
1129 		if (r)
1130 			goto out_free;
1131 	}
1132 	cpuid->nent = array.nent;
1133 
1134 	if (copy_to_user(entries, array.entries,
1135 			 array.nent * sizeof(struct kvm_cpuid_entry2)))
1136 		r = -EFAULT;
1137 
1138 out_free:
1139 	vfree(array.entries);
1140 	return r;
1141 }
1142 
1143 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
1144 					      u32 function, u32 index)
1145 {
1146 	return cpuid_entry2_find(vcpu->arch.cpuid_entries, vcpu->arch.cpuid_nent,
1147 				 function, index);
1148 }
1149 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
1150 
1151 /*
1152  * Intel CPUID semantics treats any query for an out-of-range leaf as if the
1153  * highest basic leaf (i.e. CPUID.0H:EAX) were requested.  AMD CPUID semantics
1154  * returns all zeroes for any undefined leaf, whether or not the leaf is in
1155  * range.  Centaur/VIA follows Intel semantics.
1156  *
1157  * A leaf is considered out-of-range if its function is higher than the maximum
1158  * supported leaf of its associated class or if its associated class does not
1159  * exist.
1160  *
1161  * There are three primary classes to be considered, with their respective
1162  * ranges described as "<base> - <top>[,<base2> - <top2>] inclusive.  A primary
1163  * class exists if a guest CPUID entry for its <base> leaf exists.  For a given
1164  * class, CPUID.<base>.EAX contains the max supported leaf for the class.
1165  *
1166  *  - Basic:      0x00000000 - 0x3fffffff, 0x50000000 - 0x7fffffff
1167  *  - Hypervisor: 0x40000000 - 0x4fffffff
1168  *  - Extended:   0x80000000 - 0xbfffffff
1169  *  - Centaur:    0xc0000000 - 0xcfffffff
1170  *
1171  * The Hypervisor class is further subdivided into sub-classes that each act as
1172  * their own independent class associated with a 0x100 byte range.  E.g. if Qemu
1173  * is advertising support for both HyperV and KVM, the resulting Hypervisor
1174  * CPUID sub-classes are:
1175  *
1176  *  - HyperV:     0x40000000 - 0x400000ff
1177  *  - KVM:        0x40000100 - 0x400001ff
1178  */
1179 static struct kvm_cpuid_entry2 *
1180 get_out_of_range_cpuid_entry(struct kvm_vcpu *vcpu, u32 *fn_ptr, u32 index)
1181 {
1182 	struct kvm_cpuid_entry2 *basic, *class;
1183 	u32 function = *fn_ptr;
1184 
1185 	basic = kvm_find_cpuid_entry(vcpu, 0, 0);
1186 	if (!basic)
1187 		return NULL;
1188 
1189 	if (is_guest_vendor_amd(basic->ebx, basic->ecx, basic->edx) ||
1190 	    is_guest_vendor_hygon(basic->ebx, basic->ecx, basic->edx))
1191 		return NULL;
1192 
1193 	if (function >= 0x40000000 && function <= 0x4fffffff)
1194 		class = kvm_find_cpuid_entry(vcpu, function & 0xffffff00, 0);
1195 	else if (function >= 0xc0000000)
1196 		class = kvm_find_cpuid_entry(vcpu, 0xc0000000, 0);
1197 	else
1198 		class = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
1199 
1200 	if (class && function <= class->eax)
1201 		return NULL;
1202 
1203 	/*
1204 	 * Leaf specific adjustments are also applied when redirecting to the
1205 	 * max basic entry, e.g. if the max basic leaf is 0xb but there is no
1206 	 * entry for CPUID.0xb.index (see below), then the output value for EDX
1207 	 * needs to be pulled from CPUID.0xb.1.
1208 	 */
1209 	*fn_ptr = basic->eax;
1210 
1211 	/*
1212 	 * The class does not exist or the requested function is out of range;
1213 	 * the effective CPUID entry is the max basic leaf.  Note, the index of
1214 	 * the original requested leaf is observed!
1215 	 */
1216 	return kvm_find_cpuid_entry(vcpu, basic->eax, index);
1217 }
1218 
1219 bool kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx,
1220 	       u32 *ecx, u32 *edx, bool exact_only)
1221 {
1222 	u32 orig_function = *eax, function = *eax, index = *ecx;
1223 	struct kvm_cpuid_entry2 *entry;
1224 	bool exact, used_max_basic = false;
1225 
1226 	entry = kvm_find_cpuid_entry(vcpu, function, index);
1227 	exact = !!entry;
1228 
1229 	if (!entry && !exact_only) {
1230 		entry = get_out_of_range_cpuid_entry(vcpu, &function, index);
1231 		used_max_basic = !!entry;
1232 	}
1233 
1234 	if (entry) {
1235 		*eax = entry->eax;
1236 		*ebx = entry->ebx;
1237 		*ecx = entry->ecx;
1238 		*edx = entry->edx;
1239 		if (function == 7 && index == 0) {
1240 			u64 data;
1241 		        if (!__kvm_get_msr(vcpu, MSR_IA32_TSX_CTRL, &data, true) &&
1242 			    (data & TSX_CTRL_CPUID_CLEAR))
1243 				*ebx &= ~(F(RTM) | F(HLE));
1244 		}
1245 	} else {
1246 		*eax = *ebx = *ecx = *edx = 0;
1247 		/*
1248 		 * When leaf 0BH or 1FH is defined, CL is pass-through
1249 		 * and EDX is always the x2APIC ID, even for undefined
1250 		 * subleaves. Index 1 will exist iff the leaf is
1251 		 * implemented, so we pass through CL iff leaf 1
1252 		 * exists. EDX can be copied from any existing index.
1253 		 */
1254 		if (function == 0xb || function == 0x1f) {
1255 			entry = kvm_find_cpuid_entry(vcpu, function, 1);
1256 			if (entry) {
1257 				*ecx = index & 0xff;
1258 				*edx = entry->edx;
1259 			}
1260 		}
1261 	}
1262 	trace_kvm_cpuid(orig_function, index, *eax, *ebx, *ecx, *edx, exact,
1263 			used_max_basic);
1264 	return exact;
1265 }
1266 EXPORT_SYMBOL_GPL(kvm_cpuid);
1267 
1268 int kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1269 {
1270 	u32 eax, ebx, ecx, edx;
1271 
1272 	if (cpuid_fault_enabled(vcpu) && !kvm_require_cpl(vcpu, 0))
1273 		return 1;
1274 
1275 	eax = kvm_rax_read(vcpu);
1276 	ecx = kvm_rcx_read(vcpu);
1277 	kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx, false);
1278 	kvm_rax_write(vcpu, eax);
1279 	kvm_rbx_write(vcpu, ebx);
1280 	kvm_rcx_write(vcpu, ecx);
1281 	kvm_rdx_write(vcpu, edx);
1282 	return kvm_skip_emulated_instruction(vcpu);
1283 }
1284 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1285