xref: /openbmc/linux/arch/x86/kvm/cpuid.c (revision 34fa67e7)
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 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 			/* ECX[1]: 64B alignment in compacted form */
46 			if (compacted)
47 				offset = (ecx & 0x2) ? ALIGN(ret, 64) : ret;
48 			else
49 				offset = ebx;
50 			ret = max(ret, offset + eax);
51 		}
52 
53 		xstate_bv >>= 1;
54 		feature_bit++;
55 	}
56 
57 	return ret;
58 }
59 
60 /*
61  * This one is tied to SSB in the user API, and not
62  * visible in /proc/cpuinfo.
63  */
64 #define KVM_X86_FEATURE_PSFD		(13*32+28) /* Predictive Store Forwarding Disable */
65 
66 #define F feature_bit
67 #define SF(name) (boot_cpu_has(X86_FEATURE_##name) ? F(name) : 0)
68 
69 
70 static inline struct kvm_cpuid_entry2 *cpuid_entry2_find(
71 	struct kvm_cpuid_entry2 *entries, int nent, u32 function, u32 index)
72 {
73 	struct kvm_cpuid_entry2 *e;
74 	int i;
75 
76 	for (i = 0; i < nent; i++) {
77 		e = &entries[i];
78 
79 		if (e->function == function &&
80 		    (!(e->flags & KVM_CPUID_FLAG_SIGNIFCANT_INDEX) || e->index == index))
81 			return e;
82 	}
83 
84 	return NULL;
85 }
86 
87 static int kvm_check_cpuid(struct kvm_vcpu *vcpu,
88 			   struct kvm_cpuid_entry2 *entries,
89 			   int nent)
90 {
91 	struct kvm_cpuid_entry2 *best;
92 	u64 xfeatures;
93 
94 	/*
95 	 * The existing code assumes virtual address is 48-bit or 57-bit in the
96 	 * canonical address checks; exit if it is ever changed.
97 	 */
98 	best = cpuid_entry2_find(entries, nent, 0x80000008, 0);
99 	if (best) {
100 		int vaddr_bits = (best->eax & 0xff00) >> 8;
101 
102 		if (vaddr_bits != 48 && vaddr_bits != 57 && vaddr_bits != 0)
103 			return -EINVAL;
104 	}
105 
106 	/*
107 	 * Exposing dynamic xfeatures to the guest requires additional
108 	 * enabling in the FPU, e.g. to expand the guest XSAVE state size.
109 	 */
110 	best = cpuid_entry2_find(entries, nent, 0xd, 0);
111 	if (!best)
112 		return 0;
113 
114 	xfeatures = best->eax | ((u64)best->edx << 32);
115 	xfeatures &= XFEATURE_MASK_USER_DYNAMIC;
116 	if (!xfeatures)
117 		return 0;
118 
119 	return fpu_enable_guest_xfd_features(&vcpu->arch.guest_fpu, xfeatures);
120 }
121 
122 static void kvm_update_kvm_cpuid_base(struct kvm_vcpu *vcpu)
123 {
124 	u32 function;
125 	struct kvm_cpuid_entry2 *entry;
126 
127 	vcpu->arch.kvm_cpuid_base = 0;
128 
129 	for_each_possible_hypervisor_cpuid_base(function) {
130 		entry = kvm_find_cpuid_entry(vcpu, function, 0);
131 
132 		if (entry) {
133 			u32 signature[3];
134 
135 			signature[0] = entry->ebx;
136 			signature[1] = entry->ecx;
137 			signature[2] = entry->edx;
138 
139 			BUILD_BUG_ON(sizeof(signature) > sizeof(KVM_SIGNATURE));
140 			if (!memcmp(signature, KVM_SIGNATURE, sizeof(signature))) {
141 				vcpu->arch.kvm_cpuid_base = function;
142 				break;
143 			}
144 		}
145 	}
146 }
147 
148 static struct kvm_cpuid_entry2 *kvm_find_kvm_cpuid_features(struct kvm_vcpu *vcpu)
149 {
150 	u32 base = vcpu->arch.kvm_cpuid_base;
151 
152 	if (!base)
153 		return NULL;
154 
155 	return kvm_find_cpuid_entry(vcpu, base | KVM_CPUID_FEATURES, 0);
156 }
157 
158 void kvm_update_pv_runtime(struct kvm_vcpu *vcpu)
159 {
160 	struct kvm_cpuid_entry2 *best = kvm_find_kvm_cpuid_features(vcpu);
161 
162 	/*
163 	 * save the feature bitmap to avoid cpuid lookup for every PV
164 	 * operation
165 	 */
166 	if (best)
167 		vcpu->arch.pv_cpuid.features = best->eax;
168 }
169 
170 void kvm_update_cpuid_runtime(struct kvm_vcpu *vcpu)
171 {
172 	struct kvm_cpuid_entry2 *best;
173 
174 	best = kvm_find_cpuid_entry(vcpu, 1, 0);
175 	if (best) {
176 		/* Update OSXSAVE bit */
177 		if (boot_cpu_has(X86_FEATURE_XSAVE))
178 			cpuid_entry_change(best, X86_FEATURE_OSXSAVE,
179 				   kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE));
180 
181 		cpuid_entry_change(best, X86_FEATURE_APIC,
182 			   vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE);
183 	}
184 
185 	best = kvm_find_cpuid_entry(vcpu, 7, 0);
186 	if (best && boot_cpu_has(X86_FEATURE_PKU) && best->function == 0x7)
187 		cpuid_entry_change(best, X86_FEATURE_OSPKE,
188 				   kvm_read_cr4_bits(vcpu, X86_CR4_PKE));
189 
190 	best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
191 	if (best)
192 		best->ebx = xstate_required_size(vcpu->arch.xcr0, false);
193 
194 	best = kvm_find_cpuid_entry(vcpu, 0xD, 1);
195 	if (best && (cpuid_entry_has(best, X86_FEATURE_XSAVES) ||
196 		     cpuid_entry_has(best, X86_FEATURE_XSAVEC)))
197 		best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
198 
199 	best = kvm_find_kvm_cpuid_features(vcpu);
200 	if (kvm_hlt_in_guest(vcpu->kvm) && best &&
201 		(best->eax & (1 << KVM_FEATURE_PV_UNHALT)))
202 		best->eax &= ~(1 << KVM_FEATURE_PV_UNHALT);
203 
204 	if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT)) {
205 		best = kvm_find_cpuid_entry(vcpu, 0x1, 0);
206 		if (best)
207 			cpuid_entry_change(best, X86_FEATURE_MWAIT,
208 					   vcpu->arch.ia32_misc_enable_msr &
209 					   MSR_IA32_MISC_ENABLE_MWAIT);
210 	}
211 }
212 EXPORT_SYMBOL_GPL(kvm_update_cpuid_runtime);
213 
214 static void kvm_vcpu_after_set_cpuid(struct kvm_vcpu *vcpu)
215 {
216 	struct kvm_lapic *apic = vcpu->arch.apic;
217 	struct kvm_cpuid_entry2 *best;
218 
219 	best = kvm_find_cpuid_entry(vcpu, 1, 0);
220 	if (best && apic) {
221 		if (cpuid_entry_has(best, X86_FEATURE_TSC_DEADLINE_TIMER))
222 			apic->lapic_timer.timer_mode_mask = 3 << 17;
223 		else
224 			apic->lapic_timer.timer_mode_mask = 1 << 17;
225 
226 		kvm_apic_set_version(vcpu);
227 	}
228 
229 	best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
230 	if (!best)
231 		vcpu->arch.guest_supported_xcr0 = 0;
232 	else
233 		vcpu->arch.guest_supported_xcr0 =
234 			(best->eax | ((u64)best->edx << 32)) & supported_xcr0;
235 
236 	/*
237 	 * Bits 127:0 of the allowed SECS.ATTRIBUTES (CPUID.0x12.0x1) enumerate
238 	 * the supported XSAVE Feature Request Mask (XFRM), i.e. the enclave's
239 	 * requested XCR0 value.  The enclave's XFRM must be a subset of XCRO
240 	 * at the time of EENTER, thus adjust the allowed XFRM by the guest's
241 	 * supported XCR0.  Similar to XCR0 handling, FP and SSE are forced to
242 	 * '1' even on CPUs that don't support XSAVE.
243 	 */
244 	best = kvm_find_cpuid_entry(vcpu, 0x12, 0x1);
245 	if (best) {
246 		best->ecx &= vcpu->arch.guest_supported_xcr0 & 0xffffffff;
247 		best->edx &= vcpu->arch.guest_supported_xcr0 >> 32;
248 		best->ecx |= XFEATURE_MASK_FPSSE;
249 	}
250 
251 	kvm_update_pv_runtime(vcpu);
252 
253 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
254 	vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu);
255 
256 	kvm_pmu_refresh(vcpu);
257 	vcpu->arch.cr4_guest_rsvd_bits =
258 	    __cr4_reserved_bits(guest_cpuid_has, vcpu);
259 
260 	kvm_hv_set_cpuid(vcpu);
261 
262 	/* Invoke the vendor callback only after the above state is updated. */
263 	static_call(kvm_x86_vcpu_after_set_cpuid)(vcpu);
264 
265 	/*
266 	 * Except for the MMU, which needs to do its thing any vendor specific
267 	 * adjustments to the reserved GPA bits.
268 	 */
269 	kvm_mmu_after_set_cpuid(vcpu);
270 }
271 
272 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
273 {
274 	struct kvm_cpuid_entry2 *best;
275 
276 	best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
277 	if (!best || best->eax < 0x80000008)
278 		goto not_found;
279 	best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
280 	if (best)
281 		return best->eax & 0xff;
282 not_found:
283 	return 36;
284 }
285 
286 /*
287  * This "raw" version returns the reserved GPA bits without any adjustments for
288  * encryption technologies that usurp bits.  The raw mask should be used if and
289  * only if hardware does _not_ strip the usurped bits, e.g. in virtual MTRRs.
290  */
291 u64 kvm_vcpu_reserved_gpa_bits_raw(struct kvm_vcpu *vcpu)
292 {
293 	return rsvd_bits(cpuid_maxphyaddr(vcpu), 63);
294 }
295 
296 static int kvm_set_cpuid(struct kvm_vcpu *vcpu, struct kvm_cpuid_entry2 *e2,
297                         int nent)
298 {
299 	int r;
300 
301 	r = kvm_check_cpuid(vcpu, e2, nent);
302 	if (r)
303 		return r;
304 
305 	kvfree(vcpu->arch.cpuid_entries);
306 	vcpu->arch.cpuid_entries = e2;
307 	vcpu->arch.cpuid_nent = nent;
308 
309 	kvm_update_kvm_cpuid_base(vcpu);
310 	kvm_update_cpuid_runtime(vcpu);
311 	kvm_vcpu_after_set_cpuid(vcpu);
312 
313 	return 0;
314 }
315 
316 /* when an old userspace process fills a new kernel module */
317 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
318 			     struct kvm_cpuid *cpuid,
319 			     struct kvm_cpuid_entry __user *entries)
320 {
321 	int r, i;
322 	struct kvm_cpuid_entry *e = NULL;
323 	struct kvm_cpuid_entry2 *e2 = NULL;
324 
325 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
326 		return -E2BIG;
327 
328 	if (cpuid->nent) {
329 		e = vmemdup_user(entries, array_size(sizeof(*e), cpuid->nent));
330 		if (IS_ERR(e))
331 			return PTR_ERR(e);
332 
333 		e2 = kvmalloc_array(cpuid->nent, sizeof(*e2), GFP_KERNEL_ACCOUNT);
334 		if (!e2) {
335 			r = -ENOMEM;
336 			goto out_free_cpuid;
337 		}
338 	}
339 	for (i = 0; i < cpuid->nent; i++) {
340 		e2[i].function = e[i].function;
341 		e2[i].eax = e[i].eax;
342 		e2[i].ebx = e[i].ebx;
343 		e2[i].ecx = e[i].ecx;
344 		e2[i].edx = e[i].edx;
345 		e2[i].index = 0;
346 		e2[i].flags = 0;
347 		e2[i].padding[0] = 0;
348 		e2[i].padding[1] = 0;
349 		e2[i].padding[2] = 0;
350 	}
351 
352 	r = kvm_set_cpuid(vcpu, e2, cpuid->nent);
353 	if (r)
354 		kvfree(e2);
355 
356 out_free_cpuid:
357 	kvfree(e);
358 
359 	return r;
360 }
361 
362 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
363 			      struct kvm_cpuid2 *cpuid,
364 			      struct kvm_cpuid_entry2 __user *entries)
365 {
366 	struct kvm_cpuid_entry2 *e2 = NULL;
367 	int r;
368 
369 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
370 		return -E2BIG;
371 
372 	if (cpuid->nent) {
373 		e2 = vmemdup_user(entries, array_size(sizeof(*e2), cpuid->nent));
374 		if (IS_ERR(e2))
375 			return PTR_ERR(e2);
376 	}
377 
378 	r = kvm_set_cpuid(vcpu, e2, cpuid->nent);
379 	if (r)
380 		kvfree(e2);
381 
382 	return r;
383 }
384 
385 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
386 			      struct kvm_cpuid2 *cpuid,
387 			      struct kvm_cpuid_entry2 __user *entries)
388 {
389 	int r;
390 
391 	r = -E2BIG;
392 	if (cpuid->nent < vcpu->arch.cpuid_nent)
393 		goto out;
394 	r = -EFAULT;
395 	if (copy_to_user(entries, vcpu->arch.cpuid_entries,
396 			 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
397 		goto out;
398 	return 0;
399 
400 out:
401 	cpuid->nent = vcpu->arch.cpuid_nent;
402 	return r;
403 }
404 
405 /* Mask kvm_cpu_caps for @leaf with the raw CPUID capabilities of this CPU. */
406 static __always_inline void __kvm_cpu_cap_mask(unsigned int leaf)
407 {
408 	const struct cpuid_reg cpuid = x86_feature_cpuid(leaf * 32);
409 	struct kvm_cpuid_entry2 entry;
410 
411 	reverse_cpuid_check(leaf);
412 
413 	cpuid_count(cpuid.function, cpuid.index,
414 		    &entry.eax, &entry.ebx, &entry.ecx, &entry.edx);
415 
416 	kvm_cpu_caps[leaf] &= *__cpuid_entry_get_reg(&entry, cpuid.reg);
417 }
418 
419 static __always_inline
420 void kvm_cpu_cap_init_scattered(enum kvm_only_cpuid_leafs leaf, u32 mask)
421 {
422 	/* Use kvm_cpu_cap_mask for non-scattered leafs. */
423 	BUILD_BUG_ON(leaf < NCAPINTS);
424 
425 	kvm_cpu_caps[leaf] = mask;
426 
427 	__kvm_cpu_cap_mask(leaf);
428 }
429 
430 static __always_inline void kvm_cpu_cap_mask(enum cpuid_leafs leaf, u32 mask)
431 {
432 	/* Use kvm_cpu_cap_init_scattered for scattered leafs. */
433 	BUILD_BUG_ON(leaf >= NCAPINTS);
434 
435 	kvm_cpu_caps[leaf] &= mask;
436 
437 	__kvm_cpu_cap_mask(leaf);
438 }
439 
440 void kvm_set_cpu_caps(void)
441 {
442 #ifdef CONFIG_X86_64
443 	unsigned int f_gbpages = F(GBPAGES);
444 	unsigned int f_lm = F(LM);
445 	unsigned int f_xfd = F(XFD);
446 #else
447 	unsigned int f_gbpages = 0;
448 	unsigned int f_lm = 0;
449 	unsigned int f_xfd = 0;
450 #endif
451 	memset(kvm_cpu_caps, 0, sizeof(kvm_cpu_caps));
452 
453 	BUILD_BUG_ON(sizeof(kvm_cpu_caps) - (NKVMCAPINTS * sizeof(*kvm_cpu_caps)) >
454 		     sizeof(boot_cpu_data.x86_capability));
455 
456 	memcpy(&kvm_cpu_caps, &boot_cpu_data.x86_capability,
457 	       sizeof(kvm_cpu_caps) - (NKVMCAPINTS * sizeof(*kvm_cpu_caps)));
458 
459 	kvm_cpu_cap_mask(CPUID_1_ECX,
460 		/*
461 		 * NOTE: MONITOR (and MWAIT) are emulated as NOP, but *not*
462 		 * advertised to guests via CPUID!
463 		 */
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 */ | F(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) |
471 		F(F16C) | F(RDRAND)
472 	);
473 	/* KVM emulates x2apic in software irrespective of host support. */
474 	kvm_cpu_cap_set(X86_FEATURE_X2APIC);
475 
476 	kvm_cpu_cap_mask(CPUID_1_EDX,
477 		F(FPU) | F(VME) | F(DE) | F(PSE) |
478 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
479 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
480 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
481 		F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
482 		0 /* Reserved, DS, ACPI */ | F(MMX) |
483 		F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
484 		0 /* HTT, TM, Reserved, PBE */
485 	);
486 
487 	kvm_cpu_cap_mask(CPUID_7_0_EBX,
488 		F(FSGSBASE) | F(SGX) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
489 		F(BMI2) | F(ERMS) | F(INVPCID) | F(RTM) | 0 /*MPX*/ | F(RDSEED) |
490 		F(ADX) | F(SMAP) | F(AVX512IFMA) | F(AVX512F) | F(AVX512PF) |
491 		F(AVX512ER) | F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) |
492 		F(SHA_NI) | F(AVX512BW) | F(AVX512VL) | 0 /*INTEL_PT*/
493 	);
494 
495 	kvm_cpu_cap_mask(CPUID_7_ECX,
496 		F(AVX512VBMI) | F(LA57) | F(PKU) | 0 /*OSPKE*/ | F(RDPID) |
497 		F(AVX512_VPOPCNTDQ) | F(UMIP) | F(AVX512_VBMI2) | F(GFNI) |
498 		F(VAES) | F(VPCLMULQDQ) | F(AVX512_VNNI) | F(AVX512_BITALG) |
499 		F(CLDEMOTE) | F(MOVDIRI) | F(MOVDIR64B) | 0 /*WAITPKG*/ |
500 		F(SGX_LC) | F(BUS_LOCK_DETECT)
501 	);
502 	/* Set LA57 based on hardware capability. */
503 	if (cpuid_ecx(7) & F(LA57))
504 		kvm_cpu_cap_set(X86_FEATURE_LA57);
505 
506 	/*
507 	 * PKU not yet implemented for shadow paging and requires OSPKE
508 	 * to be set on the host. Clear it if that is not the case
509 	 */
510 	if (!tdp_enabled || !boot_cpu_has(X86_FEATURE_OSPKE))
511 		kvm_cpu_cap_clear(X86_FEATURE_PKU);
512 
513 	kvm_cpu_cap_mask(CPUID_7_EDX,
514 		F(AVX512_4VNNIW) | F(AVX512_4FMAPS) | F(SPEC_CTRL) |
515 		F(SPEC_CTRL_SSBD) | F(ARCH_CAPABILITIES) | F(INTEL_STIBP) |
516 		F(MD_CLEAR) | F(AVX512_VP2INTERSECT) | F(FSRM) |
517 		F(SERIALIZE) | F(TSXLDTRK) | F(AVX512_FP16) |
518 		F(AMX_TILE) | F(AMX_INT8) | F(AMX_BF16)
519 	);
520 
521 	/* TSC_ADJUST and ARCH_CAPABILITIES are emulated in software. */
522 	kvm_cpu_cap_set(X86_FEATURE_TSC_ADJUST);
523 	kvm_cpu_cap_set(X86_FEATURE_ARCH_CAPABILITIES);
524 
525 	if (boot_cpu_has(X86_FEATURE_IBPB) && boot_cpu_has(X86_FEATURE_IBRS))
526 		kvm_cpu_cap_set(X86_FEATURE_SPEC_CTRL);
527 	if (boot_cpu_has(X86_FEATURE_STIBP))
528 		kvm_cpu_cap_set(X86_FEATURE_INTEL_STIBP);
529 	if (boot_cpu_has(X86_FEATURE_AMD_SSBD))
530 		kvm_cpu_cap_set(X86_FEATURE_SPEC_CTRL_SSBD);
531 
532 	kvm_cpu_cap_mask(CPUID_7_1_EAX,
533 		F(AVX_VNNI) | F(AVX512_BF16)
534 	);
535 
536 	kvm_cpu_cap_mask(CPUID_D_1_EAX,
537 		F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | F(XSAVES) | f_xfd
538 	);
539 
540 	kvm_cpu_cap_init_scattered(CPUID_12_EAX,
541 		SF(SGX1) | SF(SGX2)
542 	);
543 
544 	kvm_cpu_cap_mask(CPUID_8000_0001_ECX,
545 		F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
546 		F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
547 		F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
548 		0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM) |
549 		F(TOPOEXT) | 0 /* PERFCTR_CORE */
550 	);
551 
552 	kvm_cpu_cap_mask(CPUID_8000_0001_EDX,
553 		F(FPU) | F(VME) | F(DE) | F(PSE) |
554 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
555 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
556 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
557 		F(PAT) | F(PSE36) | 0 /* Reserved */ |
558 		F(NX) | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
559 		F(FXSR) | F(FXSR_OPT) | f_gbpages | F(RDTSCP) |
560 		0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW)
561 	);
562 
563 	if (!tdp_enabled && IS_ENABLED(CONFIG_X86_64))
564 		kvm_cpu_cap_set(X86_FEATURE_GBPAGES);
565 
566 	kvm_cpu_cap_mask(CPUID_8000_0008_EBX,
567 		F(CLZERO) | F(XSAVEERPTR) |
568 		F(WBNOINVD) | F(AMD_IBPB) | F(AMD_IBRS) | F(AMD_SSBD) | F(VIRT_SSBD) |
569 		F(AMD_SSB_NO) | F(AMD_STIBP) | F(AMD_STIBP_ALWAYS_ON) |
570 		__feature_bit(KVM_X86_FEATURE_PSFD)
571 	);
572 
573 	/*
574 	 * AMD has separate bits for each SPEC_CTRL bit.
575 	 * arch/x86/kernel/cpu/bugs.c is kind enough to
576 	 * record that in cpufeatures so use them.
577 	 */
578 	if (boot_cpu_has(X86_FEATURE_IBPB))
579 		kvm_cpu_cap_set(X86_FEATURE_AMD_IBPB);
580 	if (boot_cpu_has(X86_FEATURE_IBRS))
581 		kvm_cpu_cap_set(X86_FEATURE_AMD_IBRS);
582 	if (boot_cpu_has(X86_FEATURE_STIBP))
583 		kvm_cpu_cap_set(X86_FEATURE_AMD_STIBP);
584 	if (boot_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD))
585 		kvm_cpu_cap_set(X86_FEATURE_AMD_SSBD);
586 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
587 		kvm_cpu_cap_set(X86_FEATURE_AMD_SSB_NO);
588 	/*
589 	 * The preference is to use SPEC CTRL MSR instead of the
590 	 * VIRT_SPEC MSR.
591 	 */
592 	if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD) &&
593 	    !boot_cpu_has(X86_FEATURE_AMD_SSBD))
594 		kvm_cpu_cap_set(X86_FEATURE_VIRT_SSBD);
595 
596 	/*
597 	 * Hide all SVM features by default, SVM will set the cap bits for
598 	 * features it emulates and/or exposes for L1.
599 	 */
600 	kvm_cpu_cap_mask(CPUID_8000_000A_EDX, 0);
601 
602 	kvm_cpu_cap_mask(CPUID_8000_001F_EAX,
603 		0 /* SME */ | F(SEV) | 0 /* VM_PAGE_FLUSH */ | F(SEV_ES) |
604 		F(SME_COHERENT));
605 
606 	kvm_cpu_cap_mask(CPUID_C000_0001_EDX,
607 		F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
608 		F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
609 		F(PMM) | F(PMM_EN)
610 	);
611 
612 	/*
613 	 * Hide RDTSCP and RDPID if either feature is reported as supported but
614 	 * probing MSR_TSC_AUX failed.  This is purely a sanity check and
615 	 * should never happen, but the guest will likely crash if RDTSCP or
616 	 * RDPID is misreported, and KVM has botched MSR_TSC_AUX emulation in
617 	 * the past.  For example, the sanity check may fire if this instance of
618 	 * KVM is running as L1 on top of an older, broken KVM.
619 	 */
620 	if (WARN_ON((kvm_cpu_cap_has(X86_FEATURE_RDTSCP) ||
621 		     kvm_cpu_cap_has(X86_FEATURE_RDPID)) &&
622 		     !kvm_is_supported_user_return_msr(MSR_TSC_AUX))) {
623 		kvm_cpu_cap_clear(X86_FEATURE_RDTSCP);
624 		kvm_cpu_cap_clear(X86_FEATURE_RDPID);
625 	}
626 }
627 EXPORT_SYMBOL_GPL(kvm_set_cpu_caps);
628 
629 struct kvm_cpuid_array {
630 	struct kvm_cpuid_entry2 *entries;
631 	int maxnent;
632 	int nent;
633 };
634 
635 static struct kvm_cpuid_entry2 *do_host_cpuid(struct kvm_cpuid_array *array,
636 					      u32 function, u32 index)
637 {
638 	struct kvm_cpuid_entry2 *entry;
639 
640 	if (array->nent >= array->maxnent)
641 		return NULL;
642 
643 	entry = &array->entries[array->nent++];
644 
645 	entry->function = function;
646 	entry->index = index;
647 	entry->flags = 0;
648 
649 	cpuid_count(entry->function, entry->index,
650 		    &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
651 
652 	switch (function) {
653 	case 4:
654 	case 7:
655 	case 0xb:
656 	case 0xd:
657 	case 0xf:
658 	case 0x10:
659 	case 0x12:
660 	case 0x14:
661 	case 0x17:
662 	case 0x18:
663 	case 0x1d:
664 	case 0x1e:
665 	case 0x1f:
666 	case 0x8000001d:
667 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
668 		break;
669 	}
670 
671 	return entry;
672 }
673 
674 static int __do_cpuid_func_emulated(struct kvm_cpuid_array *array, u32 func)
675 {
676 	struct kvm_cpuid_entry2 *entry;
677 
678 	if (array->nent >= array->maxnent)
679 		return -E2BIG;
680 
681 	entry = &array->entries[array->nent];
682 	entry->function = func;
683 	entry->index = 0;
684 	entry->flags = 0;
685 
686 	switch (func) {
687 	case 0:
688 		entry->eax = 7;
689 		++array->nent;
690 		break;
691 	case 1:
692 		entry->ecx = F(MOVBE);
693 		++array->nent;
694 		break;
695 	case 7:
696 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
697 		entry->eax = 0;
698 		if (kvm_cpu_cap_has(X86_FEATURE_RDTSCP))
699 			entry->ecx = F(RDPID);
700 		++array->nent;
701 		break;
702 	default:
703 		break;
704 	}
705 
706 	return 0;
707 }
708 
709 static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function)
710 {
711 	struct kvm_cpuid_entry2 *entry;
712 	int r, i, max_idx;
713 
714 	/* all calls to cpuid_count() should be made on the same cpu */
715 	get_cpu();
716 
717 	r = -E2BIG;
718 
719 	entry = do_host_cpuid(array, function, 0);
720 	if (!entry)
721 		goto out;
722 
723 	switch (function) {
724 	case 0:
725 		/* Limited to the highest leaf implemented in KVM. */
726 		entry->eax = min(entry->eax, 0x1fU);
727 		break;
728 	case 1:
729 		cpuid_entry_override(entry, CPUID_1_EDX);
730 		cpuid_entry_override(entry, CPUID_1_ECX);
731 		break;
732 	case 2:
733 		/*
734 		 * On ancient CPUs, function 2 entries are STATEFUL.  That is,
735 		 * CPUID(function=2, index=0) may return different results each
736 		 * time, with the least-significant byte in EAX enumerating the
737 		 * number of times software should do CPUID(2, 0).
738 		 *
739 		 * Modern CPUs, i.e. every CPU KVM has *ever* run on are less
740 		 * idiotic.  Intel's SDM states that EAX & 0xff "will always
741 		 * return 01H. Software should ignore this value and not
742 		 * interpret it as an informational descriptor", while AMD's
743 		 * APM states that CPUID(2) is reserved.
744 		 *
745 		 * WARN if a frankenstein CPU that supports virtualization and
746 		 * a stateful CPUID.0x2 is encountered.
747 		 */
748 		WARN_ON_ONCE((entry->eax & 0xff) > 1);
749 		break;
750 	/* functions 4 and 0x8000001d have additional index. */
751 	case 4:
752 	case 0x8000001d:
753 		/*
754 		 * Read entries until the cache type in the previous entry is
755 		 * zero, i.e. indicates an invalid entry.
756 		 */
757 		for (i = 1; entry->eax & 0x1f; ++i) {
758 			entry = do_host_cpuid(array, function, i);
759 			if (!entry)
760 				goto out;
761 		}
762 		break;
763 	case 6: /* Thermal management */
764 		entry->eax = 0x4; /* allow ARAT */
765 		entry->ebx = 0;
766 		entry->ecx = 0;
767 		entry->edx = 0;
768 		break;
769 	/* function 7 has additional index. */
770 	case 7:
771 		entry->eax = min(entry->eax, 1u);
772 		cpuid_entry_override(entry, CPUID_7_0_EBX);
773 		cpuid_entry_override(entry, CPUID_7_ECX);
774 		cpuid_entry_override(entry, CPUID_7_EDX);
775 
776 		/* KVM only supports 0x7.0 and 0x7.1, capped above via min(). */
777 		if (entry->eax == 1) {
778 			entry = do_host_cpuid(array, function, 1);
779 			if (!entry)
780 				goto out;
781 
782 			cpuid_entry_override(entry, CPUID_7_1_EAX);
783 			entry->ebx = 0;
784 			entry->ecx = 0;
785 			entry->edx = 0;
786 		}
787 		break;
788 	case 9:
789 		break;
790 	case 0xa: { /* Architectural Performance Monitoring */
791 		struct x86_pmu_capability cap;
792 		union cpuid10_eax eax;
793 		union cpuid10_edx edx;
794 
795 		perf_get_x86_pmu_capability(&cap);
796 
797 		/*
798 		 * Only support guest architectural pmu on a host
799 		 * with architectural pmu.
800 		 */
801 		if (!cap.version)
802 			memset(&cap, 0, sizeof(cap));
803 
804 		eax.split.version_id = min(cap.version, 2);
805 		eax.split.num_counters = cap.num_counters_gp;
806 		eax.split.bit_width = cap.bit_width_gp;
807 		eax.split.mask_length = cap.events_mask_len;
808 
809 		edx.split.num_counters_fixed = min(cap.num_counters_fixed, MAX_FIXED_COUNTERS);
810 		edx.split.bit_width_fixed = cap.bit_width_fixed;
811 		if (cap.version)
812 			edx.split.anythread_deprecated = 1;
813 		edx.split.reserved1 = 0;
814 		edx.split.reserved2 = 0;
815 
816 		entry->eax = eax.full;
817 		entry->ebx = cap.events_mask;
818 		entry->ecx = 0;
819 		entry->edx = edx.full;
820 		break;
821 	}
822 	/*
823 	 * Per Intel's SDM, the 0x1f is a superset of 0xb,
824 	 * thus they can be handled by common code.
825 	 */
826 	case 0x1f:
827 	case 0xb:
828 		/*
829 		 * Populate entries until the level type (ECX[15:8]) of the
830 		 * previous entry is zero.  Note, CPUID EAX.{0x1f,0xb}.0 is
831 		 * the starting entry, filled by the primary do_host_cpuid().
832 		 */
833 		for (i = 1; entry->ecx & 0xff00; ++i) {
834 			entry = do_host_cpuid(array, function, i);
835 			if (!entry)
836 				goto out;
837 		}
838 		break;
839 	case 0xd: {
840 		u64 guest_perm = xstate_get_guest_group_perm();
841 
842 		entry->eax &= supported_xcr0 & guest_perm;
843 		entry->ebx = xstate_required_size(supported_xcr0, false);
844 		entry->ecx = entry->ebx;
845 		entry->edx &= (supported_xcr0 & guest_perm) >> 32;
846 		if (!supported_xcr0)
847 			break;
848 
849 		entry = do_host_cpuid(array, function, 1);
850 		if (!entry)
851 			goto out;
852 
853 		cpuid_entry_override(entry, CPUID_D_1_EAX);
854 		if (entry->eax & (F(XSAVES)|F(XSAVEC)))
855 			entry->ebx = xstate_required_size(supported_xcr0 | supported_xss,
856 							  true);
857 		else {
858 			WARN_ON_ONCE(supported_xss != 0);
859 			entry->ebx = 0;
860 		}
861 		entry->ecx &= supported_xss;
862 		entry->edx &= supported_xss >> 32;
863 
864 		for (i = 2; i < 64; ++i) {
865 			bool s_state;
866 			if (supported_xcr0 & BIT_ULL(i))
867 				s_state = false;
868 			else if (supported_xss & BIT_ULL(i))
869 				s_state = true;
870 			else
871 				continue;
872 
873 			entry = do_host_cpuid(array, function, i);
874 			if (!entry)
875 				goto out;
876 
877 			/*
878 			 * The supported check above should have filtered out
879 			 * invalid sub-leafs.  Only valid sub-leafs should
880 			 * reach this point, and they should have a non-zero
881 			 * save state size.  Furthermore, check whether the
882 			 * processor agrees with supported_xcr0/supported_xss
883 			 * on whether this is an XCR0- or IA32_XSS-managed area.
884 			 */
885 			if (WARN_ON_ONCE(!entry->eax || (entry->ecx & 0x1) != s_state)) {
886 				--array->nent;
887 				continue;
888 			}
889 			entry->edx = 0;
890 		}
891 		break;
892 	}
893 	case 0x12:
894 		/* Intel SGX */
895 		if (!kvm_cpu_cap_has(X86_FEATURE_SGX)) {
896 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
897 			break;
898 		}
899 
900 		/*
901 		 * Index 0: Sub-features, MISCSELECT (a.k.a extended features)
902 		 * and max enclave sizes.   The SGX sub-features and MISCSELECT
903 		 * are restricted by kernel and KVM capabilities (like most
904 		 * feature flags), while enclave size is unrestricted.
905 		 */
906 		cpuid_entry_override(entry, CPUID_12_EAX);
907 		entry->ebx &= SGX_MISC_EXINFO;
908 
909 		entry = do_host_cpuid(array, function, 1);
910 		if (!entry)
911 			goto out;
912 
913 		/*
914 		 * Index 1: SECS.ATTRIBUTES.  ATTRIBUTES are restricted a la
915 		 * feature flags.  Advertise all supported flags, including
916 		 * privileged attributes that require explicit opt-in from
917 		 * userspace.  ATTRIBUTES.XFRM is not adjusted as userspace is
918 		 * expected to derive it from supported XCR0.
919 		 */
920 		entry->eax &= SGX_ATTR_DEBUG | SGX_ATTR_MODE64BIT |
921 			      SGX_ATTR_PROVISIONKEY | SGX_ATTR_EINITTOKENKEY |
922 			      SGX_ATTR_KSS;
923 		entry->ebx &= 0;
924 		break;
925 	/* Intel PT */
926 	case 0x14:
927 		if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT)) {
928 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
929 			break;
930 		}
931 
932 		for (i = 1, max_idx = entry->eax; i <= max_idx; ++i) {
933 			if (!do_host_cpuid(array, function, i))
934 				goto out;
935 		}
936 		break;
937 	/* Intel AMX TILE */
938 	case 0x1d:
939 		if (!kvm_cpu_cap_has(X86_FEATURE_AMX_TILE)) {
940 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
941 			break;
942 		}
943 
944 		for (i = 1, max_idx = entry->eax; i <= max_idx; ++i) {
945 			if (!do_host_cpuid(array, function, i))
946 				goto out;
947 		}
948 		break;
949 	case 0x1e: /* TMUL information */
950 		if (!kvm_cpu_cap_has(X86_FEATURE_AMX_TILE)) {
951 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
952 			break;
953 		}
954 		break;
955 	case KVM_CPUID_SIGNATURE: {
956 		const u32 *sigptr = (const u32 *)KVM_SIGNATURE;
957 		entry->eax = KVM_CPUID_FEATURES;
958 		entry->ebx = sigptr[0];
959 		entry->ecx = sigptr[1];
960 		entry->edx = sigptr[2];
961 		break;
962 	}
963 	case KVM_CPUID_FEATURES:
964 		entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
965 			     (1 << KVM_FEATURE_NOP_IO_DELAY) |
966 			     (1 << KVM_FEATURE_CLOCKSOURCE2) |
967 			     (1 << KVM_FEATURE_ASYNC_PF) |
968 			     (1 << KVM_FEATURE_PV_EOI) |
969 			     (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
970 			     (1 << KVM_FEATURE_PV_UNHALT) |
971 			     (1 << KVM_FEATURE_PV_TLB_FLUSH) |
972 			     (1 << KVM_FEATURE_ASYNC_PF_VMEXIT) |
973 			     (1 << KVM_FEATURE_PV_SEND_IPI) |
974 			     (1 << KVM_FEATURE_POLL_CONTROL) |
975 			     (1 << KVM_FEATURE_PV_SCHED_YIELD) |
976 			     (1 << KVM_FEATURE_ASYNC_PF_INT);
977 
978 		if (sched_info_on())
979 			entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
980 
981 		entry->ebx = 0;
982 		entry->ecx = 0;
983 		entry->edx = 0;
984 		break;
985 	case 0x80000000:
986 		entry->eax = min(entry->eax, 0x8000001f);
987 		break;
988 	case 0x80000001:
989 		cpuid_entry_override(entry, CPUID_8000_0001_EDX);
990 		cpuid_entry_override(entry, CPUID_8000_0001_ECX);
991 		break;
992 	case 0x80000006:
993 		/* L2 cache and TLB: pass through host info. */
994 		break;
995 	case 0x80000007: /* Advanced power management */
996 		/* invariant TSC is CPUID.80000007H:EDX[8] */
997 		entry->edx &= (1 << 8);
998 		/* mask against host */
999 		entry->edx &= boot_cpu_data.x86_power;
1000 		entry->eax = entry->ebx = entry->ecx = 0;
1001 		break;
1002 	case 0x80000008: {
1003 		unsigned g_phys_as = (entry->eax >> 16) & 0xff;
1004 		unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
1005 		unsigned phys_as = entry->eax & 0xff;
1006 
1007 		/*
1008 		 * If TDP (NPT) is disabled use the adjusted host MAXPHYADDR as
1009 		 * the guest operates in the same PA space as the host, i.e.
1010 		 * reductions in MAXPHYADDR for memory encryption affect shadow
1011 		 * paging, too.
1012 		 *
1013 		 * If TDP is enabled but an explicit guest MAXPHYADDR is not
1014 		 * provided, use the raw bare metal MAXPHYADDR as reductions to
1015 		 * the HPAs do not affect GPAs.
1016 		 */
1017 		if (!tdp_enabled)
1018 			g_phys_as = boot_cpu_data.x86_phys_bits;
1019 		else if (!g_phys_as)
1020 			g_phys_as = phys_as;
1021 
1022 		entry->eax = g_phys_as | (virt_as << 8);
1023 		entry->edx = 0;
1024 		cpuid_entry_override(entry, CPUID_8000_0008_EBX);
1025 		break;
1026 	}
1027 	case 0x8000000A:
1028 		if (!kvm_cpu_cap_has(X86_FEATURE_SVM)) {
1029 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
1030 			break;
1031 		}
1032 		entry->eax = 1; /* SVM revision 1 */
1033 		entry->ebx = 8; /* Lets support 8 ASIDs in case we add proper
1034 				   ASID emulation to nested SVM */
1035 		entry->ecx = 0; /* Reserved */
1036 		cpuid_entry_override(entry, CPUID_8000_000A_EDX);
1037 		break;
1038 	case 0x80000019:
1039 		entry->ecx = entry->edx = 0;
1040 		break;
1041 	case 0x8000001a:
1042 	case 0x8000001e:
1043 		break;
1044 	case 0x8000001F:
1045 		if (!kvm_cpu_cap_has(X86_FEATURE_SEV)) {
1046 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
1047 		} else {
1048 			cpuid_entry_override(entry, CPUID_8000_001F_EAX);
1049 
1050 			/*
1051 			 * Enumerate '0' for "PA bits reduction", the adjusted
1052 			 * MAXPHYADDR is enumerated directly (see 0x80000008).
1053 			 */
1054 			entry->ebx &= ~GENMASK(11, 6);
1055 		}
1056 		break;
1057 	/*Add support for Centaur's CPUID instruction*/
1058 	case 0xC0000000:
1059 		/*Just support up to 0xC0000004 now*/
1060 		entry->eax = min(entry->eax, 0xC0000004);
1061 		break;
1062 	case 0xC0000001:
1063 		cpuid_entry_override(entry, CPUID_C000_0001_EDX);
1064 		break;
1065 	case 3: /* Processor serial number */
1066 	case 5: /* MONITOR/MWAIT */
1067 	case 0xC0000002:
1068 	case 0xC0000003:
1069 	case 0xC0000004:
1070 	default:
1071 		entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
1072 		break;
1073 	}
1074 
1075 	r = 0;
1076 
1077 out:
1078 	put_cpu();
1079 
1080 	return r;
1081 }
1082 
1083 static int do_cpuid_func(struct kvm_cpuid_array *array, u32 func,
1084 			 unsigned int type)
1085 {
1086 	if (type == KVM_GET_EMULATED_CPUID)
1087 		return __do_cpuid_func_emulated(array, func);
1088 
1089 	return __do_cpuid_func(array, func);
1090 }
1091 
1092 #define CENTAUR_CPUID_SIGNATURE 0xC0000000
1093 
1094 static int get_cpuid_func(struct kvm_cpuid_array *array, u32 func,
1095 			  unsigned int type)
1096 {
1097 	u32 limit;
1098 	int r;
1099 
1100 	if (func == CENTAUR_CPUID_SIGNATURE &&
1101 	    boot_cpu_data.x86_vendor != X86_VENDOR_CENTAUR)
1102 		return 0;
1103 
1104 	r = do_cpuid_func(array, func, type);
1105 	if (r)
1106 		return r;
1107 
1108 	limit = array->entries[array->nent - 1].eax;
1109 	for (func = func + 1; func <= limit; ++func) {
1110 		r = do_cpuid_func(array, func, type);
1111 		if (r)
1112 			break;
1113 	}
1114 
1115 	return r;
1116 }
1117 
1118 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
1119 				 __u32 num_entries, unsigned int ioctl_type)
1120 {
1121 	int i;
1122 	__u32 pad[3];
1123 
1124 	if (ioctl_type != KVM_GET_EMULATED_CPUID)
1125 		return false;
1126 
1127 	/*
1128 	 * We want to make sure that ->padding is being passed clean from
1129 	 * userspace in case we want to use it for something in the future.
1130 	 *
1131 	 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
1132 	 * have to give ourselves satisfied only with the emulated side. /me
1133 	 * sheds a tear.
1134 	 */
1135 	for (i = 0; i < num_entries; i++) {
1136 		if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
1137 			return true;
1138 
1139 		if (pad[0] || pad[1] || pad[2])
1140 			return true;
1141 	}
1142 	return false;
1143 }
1144 
1145 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
1146 			    struct kvm_cpuid_entry2 __user *entries,
1147 			    unsigned int type)
1148 {
1149 	static const u32 funcs[] = {
1150 		0, 0x80000000, CENTAUR_CPUID_SIGNATURE, KVM_CPUID_SIGNATURE,
1151 	};
1152 
1153 	struct kvm_cpuid_array array = {
1154 		.nent = 0,
1155 	};
1156 	int r, i;
1157 
1158 	if (cpuid->nent < 1)
1159 		return -E2BIG;
1160 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1161 		cpuid->nent = KVM_MAX_CPUID_ENTRIES;
1162 
1163 	if (sanity_check_entries(entries, cpuid->nent, type))
1164 		return -EINVAL;
1165 
1166 	array.entries = vzalloc(array_size(sizeof(struct kvm_cpuid_entry2),
1167 					   cpuid->nent));
1168 	if (!array.entries)
1169 		return -ENOMEM;
1170 
1171 	array.maxnent = cpuid->nent;
1172 
1173 	for (i = 0; i < ARRAY_SIZE(funcs); i++) {
1174 		r = get_cpuid_func(&array, funcs[i], type);
1175 		if (r)
1176 			goto out_free;
1177 	}
1178 	cpuid->nent = array.nent;
1179 
1180 	if (copy_to_user(entries, array.entries,
1181 			 array.nent * sizeof(struct kvm_cpuid_entry2)))
1182 		r = -EFAULT;
1183 
1184 out_free:
1185 	vfree(array.entries);
1186 	return r;
1187 }
1188 
1189 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
1190 					      u32 function, u32 index)
1191 {
1192 	return cpuid_entry2_find(vcpu->arch.cpuid_entries, vcpu->arch.cpuid_nent,
1193 				 function, index);
1194 }
1195 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
1196 
1197 /*
1198  * Intel CPUID semantics treats any query for an out-of-range leaf as if the
1199  * highest basic leaf (i.e. CPUID.0H:EAX) were requested.  AMD CPUID semantics
1200  * returns all zeroes for any undefined leaf, whether or not the leaf is in
1201  * range.  Centaur/VIA follows Intel semantics.
1202  *
1203  * A leaf is considered out-of-range if its function is higher than the maximum
1204  * supported leaf of its associated class or if its associated class does not
1205  * exist.
1206  *
1207  * There are three primary classes to be considered, with their respective
1208  * ranges described as "<base> - <top>[,<base2> - <top2>] inclusive.  A primary
1209  * class exists if a guest CPUID entry for its <base> leaf exists.  For a given
1210  * class, CPUID.<base>.EAX contains the max supported leaf for the class.
1211  *
1212  *  - Basic:      0x00000000 - 0x3fffffff, 0x50000000 - 0x7fffffff
1213  *  - Hypervisor: 0x40000000 - 0x4fffffff
1214  *  - Extended:   0x80000000 - 0xbfffffff
1215  *  - Centaur:    0xc0000000 - 0xcfffffff
1216  *
1217  * The Hypervisor class is further subdivided into sub-classes that each act as
1218  * their own independent class associated with a 0x100 byte range.  E.g. if Qemu
1219  * is advertising support for both HyperV and KVM, the resulting Hypervisor
1220  * CPUID sub-classes are:
1221  *
1222  *  - HyperV:     0x40000000 - 0x400000ff
1223  *  - KVM:        0x40000100 - 0x400001ff
1224  */
1225 static struct kvm_cpuid_entry2 *
1226 get_out_of_range_cpuid_entry(struct kvm_vcpu *vcpu, u32 *fn_ptr, u32 index)
1227 {
1228 	struct kvm_cpuid_entry2 *basic, *class;
1229 	u32 function = *fn_ptr;
1230 
1231 	basic = kvm_find_cpuid_entry(vcpu, 0, 0);
1232 	if (!basic)
1233 		return NULL;
1234 
1235 	if (is_guest_vendor_amd(basic->ebx, basic->ecx, basic->edx) ||
1236 	    is_guest_vendor_hygon(basic->ebx, basic->ecx, basic->edx))
1237 		return NULL;
1238 
1239 	if (function >= 0x40000000 && function <= 0x4fffffff)
1240 		class = kvm_find_cpuid_entry(vcpu, function & 0xffffff00, 0);
1241 	else if (function >= 0xc0000000)
1242 		class = kvm_find_cpuid_entry(vcpu, 0xc0000000, 0);
1243 	else
1244 		class = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
1245 
1246 	if (class && function <= class->eax)
1247 		return NULL;
1248 
1249 	/*
1250 	 * Leaf specific adjustments are also applied when redirecting to the
1251 	 * max basic entry, e.g. if the max basic leaf is 0xb but there is no
1252 	 * entry for CPUID.0xb.index (see below), then the output value for EDX
1253 	 * needs to be pulled from CPUID.0xb.1.
1254 	 */
1255 	*fn_ptr = basic->eax;
1256 
1257 	/*
1258 	 * The class does not exist or the requested function is out of range;
1259 	 * the effective CPUID entry is the max basic leaf.  Note, the index of
1260 	 * the original requested leaf is observed!
1261 	 */
1262 	return kvm_find_cpuid_entry(vcpu, basic->eax, index);
1263 }
1264 
1265 bool kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx,
1266 	       u32 *ecx, u32 *edx, bool exact_only)
1267 {
1268 	u32 orig_function = *eax, function = *eax, index = *ecx;
1269 	struct kvm_cpuid_entry2 *entry;
1270 	bool exact, used_max_basic = false;
1271 
1272 	entry = kvm_find_cpuid_entry(vcpu, function, index);
1273 	exact = !!entry;
1274 
1275 	if (!entry && !exact_only) {
1276 		entry = get_out_of_range_cpuid_entry(vcpu, &function, index);
1277 		used_max_basic = !!entry;
1278 	}
1279 
1280 	if (entry) {
1281 		*eax = entry->eax;
1282 		*ebx = entry->ebx;
1283 		*ecx = entry->ecx;
1284 		*edx = entry->edx;
1285 		if (function == 7 && index == 0) {
1286 			u64 data;
1287 		        if (!__kvm_get_msr(vcpu, MSR_IA32_TSX_CTRL, &data, true) &&
1288 			    (data & TSX_CTRL_CPUID_CLEAR))
1289 				*ebx &= ~(F(RTM) | F(HLE));
1290 		}
1291 	} else {
1292 		*eax = *ebx = *ecx = *edx = 0;
1293 		/*
1294 		 * When leaf 0BH or 1FH is defined, CL is pass-through
1295 		 * and EDX is always the x2APIC ID, even for undefined
1296 		 * subleaves. Index 1 will exist iff the leaf is
1297 		 * implemented, so we pass through CL iff leaf 1
1298 		 * exists. EDX can be copied from any existing index.
1299 		 */
1300 		if (function == 0xb || function == 0x1f) {
1301 			entry = kvm_find_cpuid_entry(vcpu, function, 1);
1302 			if (entry) {
1303 				*ecx = index & 0xff;
1304 				*edx = entry->edx;
1305 			}
1306 		}
1307 	}
1308 	trace_kvm_cpuid(orig_function, index, *eax, *ebx, *ecx, *edx, exact,
1309 			used_max_basic);
1310 	return exact;
1311 }
1312 EXPORT_SYMBOL_GPL(kvm_cpuid);
1313 
1314 int kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1315 {
1316 	u32 eax, ebx, ecx, edx;
1317 
1318 	if (cpuid_fault_enabled(vcpu) && !kvm_require_cpl(vcpu, 0))
1319 		return 1;
1320 
1321 	eax = kvm_rax_read(vcpu);
1322 	ecx = kvm_rcx_read(vcpu);
1323 	kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx, false);
1324 	kvm_rax_write(vcpu, eax);
1325 	kvm_rbx_write(vcpu, ebx);
1326 	kvm_rcx_write(vcpu, ecx);
1327 	kvm_rdx_write(vcpu, edx);
1328 	return kvm_skip_emulated_instruction(vcpu);
1329 }
1330 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1331