xref: /openbmc/linux/arch/x86/kvm/cpuid.c (revision 52beb1fc)
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 /* Check whether the supplied CPUID data is equal to what is already set for the vCPU. */
123 static int kvm_cpuid_check_equal(struct kvm_vcpu *vcpu, struct kvm_cpuid_entry2 *e2,
124 				 int nent)
125 {
126 	struct kvm_cpuid_entry2 *orig;
127 	int i;
128 
129 	if (nent != vcpu->arch.cpuid_nent)
130 		return -EINVAL;
131 
132 	for (i = 0; i < nent; i++) {
133 		orig = &vcpu->arch.cpuid_entries[i];
134 		if (e2[i].function != orig->function ||
135 		    e2[i].index != orig->index ||
136 		    e2[i].eax != orig->eax || e2[i].ebx != orig->ebx ||
137 		    e2[i].ecx != orig->ecx || e2[i].edx != orig->edx)
138 			return -EINVAL;
139 	}
140 
141 	return 0;
142 }
143 
144 static void kvm_update_kvm_cpuid_base(struct kvm_vcpu *vcpu)
145 {
146 	u32 function;
147 	struct kvm_cpuid_entry2 *entry;
148 
149 	vcpu->arch.kvm_cpuid_base = 0;
150 
151 	for_each_possible_hypervisor_cpuid_base(function) {
152 		entry = kvm_find_cpuid_entry(vcpu, function, 0);
153 
154 		if (entry) {
155 			u32 signature[3];
156 
157 			signature[0] = entry->ebx;
158 			signature[1] = entry->ecx;
159 			signature[2] = entry->edx;
160 
161 			BUILD_BUG_ON(sizeof(signature) > sizeof(KVM_SIGNATURE));
162 			if (!memcmp(signature, KVM_SIGNATURE, sizeof(signature))) {
163 				vcpu->arch.kvm_cpuid_base = function;
164 				break;
165 			}
166 		}
167 	}
168 }
169 
170 static struct kvm_cpuid_entry2 *__kvm_find_kvm_cpuid_features(struct kvm_vcpu *vcpu,
171 					      struct kvm_cpuid_entry2 *entries, int nent)
172 {
173 	u32 base = vcpu->arch.kvm_cpuid_base;
174 
175 	if (!base)
176 		return NULL;
177 
178 	return cpuid_entry2_find(entries, nent, base | KVM_CPUID_FEATURES, 0);
179 }
180 
181 static struct kvm_cpuid_entry2 *kvm_find_kvm_cpuid_features(struct kvm_vcpu *vcpu)
182 {
183 	return __kvm_find_kvm_cpuid_features(vcpu, vcpu->arch.cpuid_entries,
184 					     vcpu->arch.cpuid_nent);
185 }
186 
187 void kvm_update_pv_runtime(struct kvm_vcpu *vcpu)
188 {
189 	struct kvm_cpuid_entry2 *best = kvm_find_kvm_cpuid_features(vcpu);
190 
191 	/*
192 	 * save the feature bitmap to avoid cpuid lookup for every PV
193 	 * operation
194 	 */
195 	if (best)
196 		vcpu->arch.pv_cpuid.features = best->eax;
197 }
198 
199 static void __kvm_update_cpuid_runtime(struct kvm_vcpu *vcpu, struct kvm_cpuid_entry2 *entries,
200 				       int nent)
201 {
202 	struct kvm_cpuid_entry2 *best;
203 
204 	best = cpuid_entry2_find(entries, nent, 1, 0);
205 	if (best) {
206 		/* Update OSXSAVE bit */
207 		if (boot_cpu_has(X86_FEATURE_XSAVE))
208 			cpuid_entry_change(best, X86_FEATURE_OSXSAVE,
209 				   kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE));
210 
211 		cpuid_entry_change(best, X86_FEATURE_APIC,
212 			   vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE);
213 	}
214 
215 	best = cpuid_entry2_find(entries, nent, 7, 0);
216 	if (best && boot_cpu_has(X86_FEATURE_PKU) && best->function == 0x7)
217 		cpuid_entry_change(best, X86_FEATURE_OSPKE,
218 				   kvm_read_cr4_bits(vcpu, X86_CR4_PKE));
219 
220 	best = cpuid_entry2_find(entries, nent, 0xD, 0);
221 	if (best)
222 		best->ebx = xstate_required_size(vcpu->arch.xcr0, false);
223 
224 	best = cpuid_entry2_find(entries, nent, 0xD, 1);
225 	if (best && (cpuid_entry_has(best, X86_FEATURE_XSAVES) ||
226 		     cpuid_entry_has(best, X86_FEATURE_XSAVEC)))
227 		best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
228 
229 	best = __kvm_find_kvm_cpuid_features(vcpu, entries, nent);
230 	if (kvm_hlt_in_guest(vcpu->kvm) && best &&
231 		(best->eax & (1 << KVM_FEATURE_PV_UNHALT)))
232 		best->eax &= ~(1 << KVM_FEATURE_PV_UNHALT);
233 
234 	if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT)) {
235 		best = cpuid_entry2_find(entries, nent, 0x1, 0);
236 		if (best)
237 			cpuid_entry_change(best, X86_FEATURE_MWAIT,
238 					   vcpu->arch.ia32_misc_enable_msr &
239 					   MSR_IA32_MISC_ENABLE_MWAIT);
240 	}
241 }
242 
243 void kvm_update_cpuid_runtime(struct kvm_vcpu *vcpu)
244 {
245 	__kvm_update_cpuid_runtime(vcpu, vcpu->arch.cpuid_entries, vcpu->arch.cpuid_nent);
246 }
247 EXPORT_SYMBOL_GPL(kvm_update_cpuid_runtime);
248 
249 static void kvm_vcpu_after_set_cpuid(struct kvm_vcpu *vcpu)
250 {
251 	struct kvm_lapic *apic = vcpu->arch.apic;
252 	struct kvm_cpuid_entry2 *best;
253 
254 	best = kvm_find_cpuid_entry(vcpu, 1, 0);
255 	if (best && apic) {
256 		if (cpuid_entry_has(best, X86_FEATURE_TSC_DEADLINE_TIMER))
257 			apic->lapic_timer.timer_mode_mask = 3 << 17;
258 		else
259 			apic->lapic_timer.timer_mode_mask = 1 << 17;
260 
261 		kvm_apic_set_version(vcpu);
262 	}
263 
264 	best = kvm_find_cpuid_entry(vcpu, 0xD, 0);
265 	if (!best)
266 		vcpu->arch.guest_supported_xcr0 = 0;
267 	else
268 		vcpu->arch.guest_supported_xcr0 =
269 			(best->eax | ((u64)best->edx << 32)) & supported_xcr0;
270 
271 	/*
272 	 * Bits 127:0 of the allowed SECS.ATTRIBUTES (CPUID.0x12.0x1) enumerate
273 	 * the supported XSAVE Feature Request Mask (XFRM), i.e. the enclave's
274 	 * requested XCR0 value.  The enclave's XFRM must be a subset of XCRO
275 	 * at the time of EENTER, thus adjust the allowed XFRM by the guest's
276 	 * supported XCR0.  Similar to XCR0 handling, FP and SSE are forced to
277 	 * '1' even on CPUs that don't support XSAVE.
278 	 */
279 	best = kvm_find_cpuid_entry(vcpu, 0x12, 0x1);
280 	if (best) {
281 		best->ecx &= vcpu->arch.guest_supported_xcr0 & 0xffffffff;
282 		best->edx &= vcpu->arch.guest_supported_xcr0 >> 32;
283 		best->ecx |= XFEATURE_MASK_FPSSE;
284 	}
285 
286 	kvm_update_pv_runtime(vcpu);
287 
288 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
289 	vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu);
290 
291 	kvm_pmu_refresh(vcpu);
292 	vcpu->arch.cr4_guest_rsvd_bits =
293 	    __cr4_reserved_bits(guest_cpuid_has, vcpu);
294 
295 	kvm_hv_set_cpuid(vcpu);
296 
297 	/* Invoke the vendor callback only after the above state is updated. */
298 	static_call(kvm_x86_vcpu_after_set_cpuid)(vcpu);
299 
300 	/*
301 	 * Except for the MMU, which needs to do its thing any vendor specific
302 	 * adjustments to the reserved GPA bits.
303 	 */
304 	kvm_mmu_after_set_cpuid(vcpu);
305 }
306 
307 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
308 {
309 	struct kvm_cpuid_entry2 *best;
310 
311 	best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
312 	if (!best || best->eax < 0x80000008)
313 		goto not_found;
314 	best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
315 	if (best)
316 		return best->eax & 0xff;
317 not_found:
318 	return 36;
319 }
320 
321 /*
322  * This "raw" version returns the reserved GPA bits without any adjustments for
323  * encryption technologies that usurp bits.  The raw mask should be used if and
324  * only if hardware does _not_ strip the usurped bits, e.g. in virtual MTRRs.
325  */
326 u64 kvm_vcpu_reserved_gpa_bits_raw(struct kvm_vcpu *vcpu)
327 {
328 	return rsvd_bits(cpuid_maxphyaddr(vcpu), 63);
329 }
330 
331 static int kvm_set_cpuid(struct kvm_vcpu *vcpu, struct kvm_cpuid_entry2 *e2,
332                         int nent)
333 {
334 	int r;
335 
336 	__kvm_update_cpuid_runtime(vcpu, e2, nent);
337 
338 	/*
339 	 * KVM does not correctly handle changing guest CPUID after KVM_RUN, as
340 	 * MAXPHYADDR, GBPAGES support, AMD reserved bit behavior, etc.. aren't
341 	 * tracked in kvm_mmu_page_role.  As a result, KVM may miss guest page
342 	 * faults due to reusing SPs/SPTEs. In practice no sane VMM mucks with
343 	 * the core vCPU model on the fly. It would've been better to forbid any
344 	 * KVM_SET_CPUID{,2} calls after KVM_RUN altogether but unfortunately
345 	 * some VMMs (e.g. QEMU) reuse vCPU fds for CPU hotplug/unplug and do
346 	 * KVM_SET_CPUID{,2} again. To support this legacy behavior, check
347 	 * whether the supplied CPUID data is equal to what's already set.
348 	 */
349 	if (vcpu->arch.last_vmentry_cpu != -1)
350 		return kvm_cpuid_check_equal(vcpu, e2, nent);
351 
352 	r = kvm_check_cpuid(vcpu, e2, nent);
353 	if (r)
354 		return r;
355 
356 	kvfree(vcpu->arch.cpuid_entries);
357 	vcpu->arch.cpuid_entries = e2;
358 	vcpu->arch.cpuid_nent = nent;
359 
360 	kvm_update_kvm_cpuid_base(vcpu);
361 	kvm_vcpu_after_set_cpuid(vcpu);
362 
363 	return 0;
364 }
365 
366 /* when an old userspace process fills a new kernel module */
367 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
368 			     struct kvm_cpuid *cpuid,
369 			     struct kvm_cpuid_entry __user *entries)
370 {
371 	int r, i;
372 	struct kvm_cpuid_entry *e = NULL;
373 	struct kvm_cpuid_entry2 *e2 = NULL;
374 
375 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
376 		return -E2BIG;
377 
378 	if (cpuid->nent) {
379 		e = vmemdup_user(entries, array_size(sizeof(*e), cpuid->nent));
380 		if (IS_ERR(e))
381 			return PTR_ERR(e);
382 
383 		e2 = kvmalloc_array(cpuid->nent, sizeof(*e2), GFP_KERNEL_ACCOUNT);
384 		if (!e2) {
385 			r = -ENOMEM;
386 			goto out_free_cpuid;
387 		}
388 	}
389 	for (i = 0; i < cpuid->nent; i++) {
390 		e2[i].function = e[i].function;
391 		e2[i].eax = e[i].eax;
392 		e2[i].ebx = e[i].ebx;
393 		e2[i].ecx = e[i].ecx;
394 		e2[i].edx = e[i].edx;
395 		e2[i].index = 0;
396 		e2[i].flags = 0;
397 		e2[i].padding[0] = 0;
398 		e2[i].padding[1] = 0;
399 		e2[i].padding[2] = 0;
400 	}
401 
402 	r = kvm_set_cpuid(vcpu, e2, cpuid->nent);
403 	if (r)
404 		kvfree(e2);
405 
406 out_free_cpuid:
407 	kvfree(e);
408 
409 	return r;
410 }
411 
412 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
413 			      struct kvm_cpuid2 *cpuid,
414 			      struct kvm_cpuid_entry2 __user *entries)
415 {
416 	struct kvm_cpuid_entry2 *e2 = NULL;
417 	int r;
418 
419 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
420 		return -E2BIG;
421 
422 	if (cpuid->nent) {
423 		e2 = vmemdup_user(entries, array_size(sizeof(*e2), cpuid->nent));
424 		if (IS_ERR(e2))
425 			return PTR_ERR(e2);
426 	}
427 
428 	r = kvm_set_cpuid(vcpu, e2, cpuid->nent);
429 	if (r)
430 		kvfree(e2);
431 
432 	return r;
433 }
434 
435 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
436 			      struct kvm_cpuid2 *cpuid,
437 			      struct kvm_cpuid_entry2 __user *entries)
438 {
439 	int r;
440 
441 	r = -E2BIG;
442 	if (cpuid->nent < vcpu->arch.cpuid_nent)
443 		goto out;
444 	r = -EFAULT;
445 	if (copy_to_user(entries, vcpu->arch.cpuid_entries,
446 			 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
447 		goto out;
448 	return 0;
449 
450 out:
451 	cpuid->nent = vcpu->arch.cpuid_nent;
452 	return r;
453 }
454 
455 /* Mask kvm_cpu_caps for @leaf with the raw CPUID capabilities of this CPU. */
456 static __always_inline void __kvm_cpu_cap_mask(unsigned int leaf)
457 {
458 	const struct cpuid_reg cpuid = x86_feature_cpuid(leaf * 32);
459 	struct kvm_cpuid_entry2 entry;
460 
461 	reverse_cpuid_check(leaf);
462 
463 	cpuid_count(cpuid.function, cpuid.index,
464 		    &entry.eax, &entry.ebx, &entry.ecx, &entry.edx);
465 
466 	kvm_cpu_caps[leaf] &= *__cpuid_entry_get_reg(&entry, cpuid.reg);
467 }
468 
469 static __always_inline
470 void kvm_cpu_cap_init_scattered(enum kvm_only_cpuid_leafs leaf, u32 mask)
471 {
472 	/* Use kvm_cpu_cap_mask for non-scattered leafs. */
473 	BUILD_BUG_ON(leaf < NCAPINTS);
474 
475 	kvm_cpu_caps[leaf] = mask;
476 
477 	__kvm_cpu_cap_mask(leaf);
478 }
479 
480 static __always_inline void kvm_cpu_cap_mask(enum cpuid_leafs leaf, u32 mask)
481 {
482 	/* Use kvm_cpu_cap_init_scattered for scattered leafs. */
483 	BUILD_BUG_ON(leaf >= NCAPINTS);
484 
485 	kvm_cpu_caps[leaf] &= mask;
486 
487 	__kvm_cpu_cap_mask(leaf);
488 }
489 
490 void kvm_set_cpu_caps(void)
491 {
492 #ifdef CONFIG_X86_64
493 	unsigned int f_gbpages = F(GBPAGES);
494 	unsigned int f_lm = F(LM);
495 	unsigned int f_xfd = F(XFD);
496 #else
497 	unsigned int f_gbpages = 0;
498 	unsigned int f_lm = 0;
499 	unsigned int f_xfd = 0;
500 #endif
501 	memset(kvm_cpu_caps, 0, sizeof(kvm_cpu_caps));
502 
503 	BUILD_BUG_ON(sizeof(kvm_cpu_caps) - (NKVMCAPINTS * sizeof(*kvm_cpu_caps)) >
504 		     sizeof(boot_cpu_data.x86_capability));
505 
506 	memcpy(&kvm_cpu_caps, &boot_cpu_data.x86_capability,
507 	       sizeof(kvm_cpu_caps) - (NKVMCAPINTS * sizeof(*kvm_cpu_caps)));
508 
509 	kvm_cpu_cap_mask(CPUID_1_ECX,
510 		/*
511 		 * NOTE: MONITOR (and MWAIT) are emulated as NOP, but *not*
512 		 * advertised to guests via CPUID!
513 		 */
514 		F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
515 		0 /* DS-CPL, VMX, SMX, EST */ |
516 		0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
517 		F(FMA) | F(CX16) | 0 /* xTPR Update */ | F(PDCM) |
518 		F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
519 		F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
520 		0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
521 		F(F16C) | F(RDRAND)
522 	);
523 	/* KVM emulates x2apic in software irrespective of host support. */
524 	kvm_cpu_cap_set(X86_FEATURE_X2APIC);
525 
526 	kvm_cpu_cap_mask(CPUID_1_EDX,
527 		F(FPU) | F(VME) | F(DE) | F(PSE) |
528 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
529 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
530 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
531 		F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
532 		0 /* Reserved, DS, ACPI */ | F(MMX) |
533 		F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
534 		0 /* HTT, TM, Reserved, PBE */
535 	);
536 
537 	kvm_cpu_cap_mask(CPUID_7_0_EBX,
538 		F(FSGSBASE) | F(SGX) | F(BMI1) | F(HLE) | F(AVX2) | F(SMEP) |
539 		F(BMI2) | F(ERMS) | F(INVPCID) | F(RTM) | 0 /*MPX*/ | F(RDSEED) |
540 		F(ADX) | F(SMAP) | F(AVX512IFMA) | F(AVX512F) | F(AVX512PF) |
541 		F(AVX512ER) | F(AVX512CD) | F(CLFLUSHOPT) | F(CLWB) | F(AVX512DQ) |
542 		F(SHA_NI) | F(AVX512BW) | F(AVX512VL) | 0 /*INTEL_PT*/
543 	);
544 
545 	kvm_cpu_cap_mask(CPUID_7_ECX,
546 		F(AVX512VBMI) | F(LA57) | F(PKU) | 0 /*OSPKE*/ | F(RDPID) |
547 		F(AVX512_VPOPCNTDQ) | F(UMIP) | F(AVX512_VBMI2) | F(GFNI) |
548 		F(VAES) | F(VPCLMULQDQ) | F(AVX512_VNNI) | F(AVX512_BITALG) |
549 		F(CLDEMOTE) | F(MOVDIRI) | F(MOVDIR64B) | 0 /*WAITPKG*/ |
550 		F(SGX_LC) | F(BUS_LOCK_DETECT)
551 	);
552 	/* Set LA57 based on hardware capability. */
553 	if (cpuid_ecx(7) & F(LA57))
554 		kvm_cpu_cap_set(X86_FEATURE_LA57);
555 
556 	/*
557 	 * PKU not yet implemented for shadow paging and requires OSPKE
558 	 * to be set on the host. Clear it if that is not the case
559 	 */
560 	if (!tdp_enabled || !boot_cpu_has(X86_FEATURE_OSPKE))
561 		kvm_cpu_cap_clear(X86_FEATURE_PKU);
562 
563 	kvm_cpu_cap_mask(CPUID_7_EDX,
564 		F(AVX512_4VNNIW) | F(AVX512_4FMAPS) | F(SPEC_CTRL) |
565 		F(SPEC_CTRL_SSBD) | F(ARCH_CAPABILITIES) | F(INTEL_STIBP) |
566 		F(MD_CLEAR) | F(AVX512_VP2INTERSECT) | F(FSRM) |
567 		F(SERIALIZE) | F(TSXLDTRK) | F(AVX512_FP16) |
568 		F(AMX_TILE) | F(AMX_INT8) | F(AMX_BF16)
569 	);
570 
571 	/* TSC_ADJUST and ARCH_CAPABILITIES are emulated in software. */
572 	kvm_cpu_cap_set(X86_FEATURE_TSC_ADJUST);
573 	kvm_cpu_cap_set(X86_FEATURE_ARCH_CAPABILITIES);
574 
575 	if (boot_cpu_has(X86_FEATURE_IBPB) && boot_cpu_has(X86_FEATURE_IBRS))
576 		kvm_cpu_cap_set(X86_FEATURE_SPEC_CTRL);
577 	if (boot_cpu_has(X86_FEATURE_STIBP))
578 		kvm_cpu_cap_set(X86_FEATURE_INTEL_STIBP);
579 	if (boot_cpu_has(X86_FEATURE_AMD_SSBD))
580 		kvm_cpu_cap_set(X86_FEATURE_SPEC_CTRL_SSBD);
581 
582 	kvm_cpu_cap_mask(CPUID_7_1_EAX,
583 		F(AVX_VNNI) | F(AVX512_BF16)
584 	);
585 
586 	kvm_cpu_cap_mask(CPUID_D_1_EAX,
587 		F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | F(XSAVES) | f_xfd
588 	);
589 
590 	kvm_cpu_cap_init_scattered(CPUID_12_EAX,
591 		SF(SGX1) | SF(SGX2)
592 	);
593 
594 	kvm_cpu_cap_mask(CPUID_8000_0001_ECX,
595 		F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
596 		F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
597 		F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
598 		0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM) |
599 		F(TOPOEXT) | 0 /* PERFCTR_CORE */
600 	);
601 
602 	kvm_cpu_cap_mask(CPUID_8000_0001_EDX,
603 		F(FPU) | F(VME) | F(DE) | F(PSE) |
604 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
605 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
606 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
607 		F(PAT) | F(PSE36) | 0 /* Reserved */ |
608 		F(NX) | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
609 		F(FXSR) | F(FXSR_OPT) | f_gbpages | F(RDTSCP) |
610 		0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW)
611 	);
612 
613 	if (!tdp_enabled && IS_ENABLED(CONFIG_X86_64))
614 		kvm_cpu_cap_set(X86_FEATURE_GBPAGES);
615 
616 	kvm_cpu_cap_mask(CPUID_8000_0008_EBX,
617 		F(CLZERO) | F(XSAVEERPTR) |
618 		F(WBNOINVD) | F(AMD_IBPB) | F(AMD_IBRS) | F(AMD_SSBD) | F(VIRT_SSBD) |
619 		F(AMD_SSB_NO) | F(AMD_STIBP) | F(AMD_STIBP_ALWAYS_ON) |
620 		__feature_bit(KVM_X86_FEATURE_PSFD)
621 	);
622 
623 	/*
624 	 * AMD has separate bits for each SPEC_CTRL bit.
625 	 * arch/x86/kernel/cpu/bugs.c is kind enough to
626 	 * record that in cpufeatures so use them.
627 	 */
628 	if (boot_cpu_has(X86_FEATURE_IBPB))
629 		kvm_cpu_cap_set(X86_FEATURE_AMD_IBPB);
630 	if (boot_cpu_has(X86_FEATURE_IBRS))
631 		kvm_cpu_cap_set(X86_FEATURE_AMD_IBRS);
632 	if (boot_cpu_has(X86_FEATURE_STIBP))
633 		kvm_cpu_cap_set(X86_FEATURE_AMD_STIBP);
634 	if (boot_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD))
635 		kvm_cpu_cap_set(X86_FEATURE_AMD_SSBD);
636 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
637 		kvm_cpu_cap_set(X86_FEATURE_AMD_SSB_NO);
638 	/*
639 	 * The preference is to use SPEC CTRL MSR instead of the
640 	 * VIRT_SPEC MSR.
641 	 */
642 	if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD) &&
643 	    !boot_cpu_has(X86_FEATURE_AMD_SSBD))
644 		kvm_cpu_cap_set(X86_FEATURE_VIRT_SSBD);
645 
646 	/*
647 	 * Hide all SVM features by default, SVM will set the cap bits for
648 	 * features it emulates and/or exposes for L1.
649 	 */
650 	kvm_cpu_cap_mask(CPUID_8000_000A_EDX, 0);
651 
652 	kvm_cpu_cap_mask(CPUID_8000_001F_EAX,
653 		0 /* SME */ | F(SEV) | 0 /* VM_PAGE_FLUSH */ | F(SEV_ES) |
654 		F(SME_COHERENT));
655 
656 	kvm_cpu_cap_mask(CPUID_C000_0001_EDX,
657 		F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
658 		F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
659 		F(PMM) | F(PMM_EN)
660 	);
661 
662 	/*
663 	 * Hide RDTSCP and RDPID if either feature is reported as supported but
664 	 * probing MSR_TSC_AUX failed.  This is purely a sanity check and
665 	 * should never happen, but the guest will likely crash if RDTSCP or
666 	 * RDPID is misreported, and KVM has botched MSR_TSC_AUX emulation in
667 	 * the past.  For example, the sanity check may fire if this instance of
668 	 * KVM is running as L1 on top of an older, broken KVM.
669 	 */
670 	if (WARN_ON((kvm_cpu_cap_has(X86_FEATURE_RDTSCP) ||
671 		     kvm_cpu_cap_has(X86_FEATURE_RDPID)) &&
672 		     !kvm_is_supported_user_return_msr(MSR_TSC_AUX))) {
673 		kvm_cpu_cap_clear(X86_FEATURE_RDTSCP);
674 		kvm_cpu_cap_clear(X86_FEATURE_RDPID);
675 	}
676 }
677 EXPORT_SYMBOL_GPL(kvm_set_cpu_caps);
678 
679 struct kvm_cpuid_array {
680 	struct kvm_cpuid_entry2 *entries;
681 	int maxnent;
682 	int nent;
683 };
684 
685 static struct kvm_cpuid_entry2 *do_host_cpuid(struct kvm_cpuid_array *array,
686 					      u32 function, u32 index)
687 {
688 	struct kvm_cpuid_entry2 *entry;
689 
690 	if (array->nent >= array->maxnent)
691 		return NULL;
692 
693 	entry = &array->entries[array->nent++];
694 
695 	entry->function = function;
696 	entry->index = index;
697 	entry->flags = 0;
698 
699 	cpuid_count(entry->function, entry->index,
700 		    &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
701 
702 	switch (function) {
703 	case 4:
704 	case 7:
705 	case 0xb:
706 	case 0xd:
707 	case 0xf:
708 	case 0x10:
709 	case 0x12:
710 	case 0x14:
711 	case 0x17:
712 	case 0x18:
713 	case 0x1d:
714 	case 0x1e:
715 	case 0x1f:
716 	case 0x8000001d:
717 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
718 		break;
719 	}
720 
721 	return entry;
722 }
723 
724 static int __do_cpuid_func_emulated(struct kvm_cpuid_array *array, u32 func)
725 {
726 	struct kvm_cpuid_entry2 *entry;
727 
728 	if (array->nent >= array->maxnent)
729 		return -E2BIG;
730 
731 	entry = &array->entries[array->nent];
732 	entry->function = func;
733 	entry->index = 0;
734 	entry->flags = 0;
735 
736 	switch (func) {
737 	case 0:
738 		entry->eax = 7;
739 		++array->nent;
740 		break;
741 	case 1:
742 		entry->ecx = F(MOVBE);
743 		++array->nent;
744 		break;
745 	case 7:
746 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
747 		entry->eax = 0;
748 		if (kvm_cpu_cap_has(X86_FEATURE_RDTSCP))
749 			entry->ecx = F(RDPID);
750 		++array->nent;
751 		break;
752 	default:
753 		break;
754 	}
755 
756 	return 0;
757 }
758 
759 static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function)
760 {
761 	struct kvm_cpuid_entry2 *entry;
762 	int r, i, max_idx;
763 
764 	/* all calls to cpuid_count() should be made on the same cpu */
765 	get_cpu();
766 
767 	r = -E2BIG;
768 
769 	entry = do_host_cpuid(array, function, 0);
770 	if (!entry)
771 		goto out;
772 
773 	switch (function) {
774 	case 0:
775 		/* Limited to the highest leaf implemented in KVM. */
776 		entry->eax = min(entry->eax, 0x1fU);
777 		break;
778 	case 1:
779 		cpuid_entry_override(entry, CPUID_1_EDX);
780 		cpuid_entry_override(entry, CPUID_1_ECX);
781 		break;
782 	case 2:
783 		/*
784 		 * On ancient CPUs, function 2 entries are STATEFUL.  That is,
785 		 * CPUID(function=2, index=0) may return different results each
786 		 * time, with the least-significant byte in EAX enumerating the
787 		 * number of times software should do CPUID(2, 0).
788 		 *
789 		 * Modern CPUs, i.e. every CPU KVM has *ever* run on are less
790 		 * idiotic.  Intel's SDM states that EAX & 0xff "will always
791 		 * return 01H. Software should ignore this value and not
792 		 * interpret it as an informational descriptor", while AMD's
793 		 * APM states that CPUID(2) is reserved.
794 		 *
795 		 * WARN if a frankenstein CPU that supports virtualization and
796 		 * a stateful CPUID.0x2 is encountered.
797 		 */
798 		WARN_ON_ONCE((entry->eax & 0xff) > 1);
799 		break;
800 	/* functions 4 and 0x8000001d have additional index. */
801 	case 4:
802 	case 0x8000001d:
803 		/*
804 		 * Read entries until the cache type in the previous entry is
805 		 * zero, i.e. indicates an invalid entry.
806 		 */
807 		for (i = 1; entry->eax & 0x1f; ++i) {
808 			entry = do_host_cpuid(array, function, i);
809 			if (!entry)
810 				goto out;
811 		}
812 		break;
813 	case 6: /* Thermal management */
814 		entry->eax = 0x4; /* allow ARAT */
815 		entry->ebx = 0;
816 		entry->ecx = 0;
817 		entry->edx = 0;
818 		break;
819 	/* function 7 has additional index. */
820 	case 7:
821 		entry->eax = min(entry->eax, 1u);
822 		cpuid_entry_override(entry, CPUID_7_0_EBX);
823 		cpuid_entry_override(entry, CPUID_7_ECX);
824 		cpuid_entry_override(entry, CPUID_7_EDX);
825 
826 		/* KVM only supports 0x7.0 and 0x7.1, capped above via min(). */
827 		if (entry->eax == 1) {
828 			entry = do_host_cpuid(array, function, 1);
829 			if (!entry)
830 				goto out;
831 
832 			cpuid_entry_override(entry, CPUID_7_1_EAX);
833 			entry->ebx = 0;
834 			entry->ecx = 0;
835 			entry->edx = 0;
836 		}
837 		break;
838 	case 9:
839 		break;
840 	case 0xa: { /* Architectural Performance Monitoring */
841 		struct x86_pmu_capability cap;
842 		union cpuid10_eax eax;
843 		union cpuid10_edx edx;
844 
845 		perf_get_x86_pmu_capability(&cap);
846 
847 		/*
848 		 * The guest architecture pmu is only supported if the architecture
849 		 * pmu exists on the host and the module parameters allow it.
850 		 */
851 		if (!cap.version || !enable_pmu)
852 			memset(&cap, 0, sizeof(cap));
853 
854 		eax.split.version_id = min(cap.version, 2);
855 		eax.split.num_counters = cap.num_counters_gp;
856 		eax.split.bit_width = cap.bit_width_gp;
857 		eax.split.mask_length = cap.events_mask_len;
858 
859 		edx.split.num_counters_fixed = min(cap.num_counters_fixed, MAX_FIXED_COUNTERS);
860 		edx.split.bit_width_fixed = cap.bit_width_fixed;
861 		if (cap.version)
862 			edx.split.anythread_deprecated = 1;
863 		edx.split.reserved1 = 0;
864 		edx.split.reserved2 = 0;
865 
866 		entry->eax = eax.full;
867 		entry->ebx = cap.events_mask;
868 		entry->ecx = 0;
869 		entry->edx = edx.full;
870 		break;
871 	}
872 	/*
873 	 * Per Intel's SDM, the 0x1f is a superset of 0xb,
874 	 * thus they can be handled by common code.
875 	 */
876 	case 0x1f:
877 	case 0xb:
878 		/*
879 		 * Populate entries until the level type (ECX[15:8]) of the
880 		 * previous entry is zero.  Note, CPUID EAX.{0x1f,0xb}.0 is
881 		 * the starting entry, filled by the primary do_host_cpuid().
882 		 */
883 		for (i = 1; entry->ecx & 0xff00; ++i) {
884 			entry = do_host_cpuid(array, function, i);
885 			if (!entry)
886 				goto out;
887 		}
888 		break;
889 	case 0xd: {
890 		u64 guest_perm = xstate_get_guest_group_perm();
891 
892 		entry->eax &= supported_xcr0 & guest_perm;
893 		entry->ebx = xstate_required_size(supported_xcr0, false);
894 		entry->ecx = entry->ebx;
895 		entry->edx &= (supported_xcr0 & guest_perm) >> 32;
896 		if (!supported_xcr0)
897 			break;
898 
899 		entry = do_host_cpuid(array, function, 1);
900 		if (!entry)
901 			goto out;
902 
903 		cpuid_entry_override(entry, CPUID_D_1_EAX);
904 		if (entry->eax & (F(XSAVES)|F(XSAVEC)))
905 			entry->ebx = xstate_required_size(supported_xcr0 | supported_xss,
906 							  true);
907 		else {
908 			WARN_ON_ONCE(supported_xss != 0);
909 			entry->ebx = 0;
910 		}
911 		entry->ecx &= supported_xss;
912 		entry->edx &= supported_xss >> 32;
913 
914 		for (i = 2; i < 64; ++i) {
915 			bool s_state;
916 			if (supported_xcr0 & BIT_ULL(i))
917 				s_state = false;
918 			else if (supported_xss & BIT_ULL(i))
919 				s_state = true;
920 			else
921 				continue;
922 
923 			entry = do_host_cpuid(array, function, i);
924 			if (!entry)
925 				goto out;
926 
927 			/*
928 			 * The supported check above should have filtered out
929 			 * invalid sub-leafs.  Only valid sub-leafs should
930 			 * reach this point, and they should have a non-zero
931 			 * save state size.  Furthermore, check whether the
932 			 * processor agrees with supported_xcr0/supported_xss
933 			 * on whether this is an XCR0- or IA32_XSS-managed area.
934 			 */
935 			if (WARN_ON_ONCE(!entry->eax || (entry->ecx & 0x1) != s_state)) {
936 				--array->nent;
937 				continue;
938 			}
939 
940 			if (!kvm_cpu_cap_has(X86_FEATURE_XFD))
941 				entry->ecx &= ~BIT_ULL(2);
942 			entry->edx = 0;
943 		}
944 		break;
945 	}
946 	case 0x12:
947 		/* Intel SGX */
948 		if (!kvm_cpu_cap_has(X86_FEATURE_SGX)) {
949 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
950 			break;
951 		}
952 
953 		/*
954 		 * Index 0: Sub-features, MISCSELECT (a.k.a extended features)
955 		 * and max enclave sizes.   The SGX sub-features and MISCSELECT
956 		 * are restricted by kernel and KVM capabilities (like most
957 		 * feature flags), while enclave size is unrestricted.
958 		 */
959 		cpuid_entry_override(entry, CPUID_12_EAX);
960 		entry->ebx &= SGX_MISC_EXINFO;
961 
962 		entry = do_host_cpuid(array, function, 1);
963 		if (!entry)
964 			goto out;
965 
966 		/*
967 		 * Index 1: SECS.ATTRIBUTES.  ATTRIBUTES are restricted a la
968 		 * feature flags.  Advertise all supported flags, including
969 		 * privileged attributes that require explicit opt-in from
970 		 * userspace.  ATTRIBUTES.XFRM is not adjusted as userspace is
971 		 * expected to derive it from supported XCR0.
972 		 */
973 		entry->eax &= SGX_ATTR_DEBUG | SGX_ATTR_MODE64BIT |
974 			      SGX_ATTR_PROVISIONKEY | SGX_ATTR_EINITTOKENKEY |
975 			      SGX_ATTR_KSS;
976 		entry->ebx &= 0;
977 		break;
978 	/* Intel PT */
979 	case 0x14:
980 		if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT)) {
981 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
982 			break;
983 		}
984 
985 		for (i = 1, max_idx = entry->eax; i <= max_idx; ++i) {
986 			if (!do_host_cpuid(array, function, i))
987 				goto out;
988 		}
989 		break;
990 	/* Intel AMX TILE */
991 	case 0x1d:
992 		if (!kvm_cpu_cap_has(X86_FEATURE_AMX_TILE)) {
993 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
994 			break;
995 		}
996 
997 		for (i = 1, max_idx = entry->eax; i <= max_idx; ++i) {
998 			if (!do_host_cpuid(array, function, i))
999 				goto out;
1000 		}
1001 		break;
1002 	case 0x1e: /* TMUL information */
1003 		if (!kvm_cpu_cap_has(X86_FEATURE_AMX_TILE)) {
1004 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
1005 			break;
1006 		}
1007 		break;
1008 	case KVM_CPUID_SIGNATURE: {
1009 		const u32 *sigptr = (const u32 *)KVM_SIGNATURE;
1010 		entry->eax = KVM_CPUID_FEATURES;
1011 		entry->ebx = sigptr[0];
1012 		entry->ecx = sigptr[1];
1013 		entry->edx = sigptr[2];
1014 		break;
1015 	}
1016 	case KVM_CPUID_FEATURES:
1017 		entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
1018 			     (1 << KVM_FEATURE_NOP_IO_DELAY) |
1019 			     (1 << KVM_FEATURE_CLOCKSOURCE2) |
1020 			     (1 << KVM_FEATURE_ASYNC_PF) |
1021 			     (1 << KVM_FEATURE_PV_EOI) |
1022 			     (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
1023 			     (1 << KVM_FEATURE_PV_UNHALT) |
1024 			     (1 << KVM_FEATURE_PV_TLB_FLUSH) |
1025 			     (1 << KVM_FEATURE_ASYNC_PF_VMEXIT) |
1026 			     (1 << KVM_FEATURE_PV_SEND_IPI) |
1027 			     (1 << KVM_FEATURE_POLL_CONTROL) |
1028 			     (1 << KVM_FEATURE_PV_SCHED_YIELD) |
1029 			     (1 << KVM_FEATURE_ASYNC_PF_INT);
1030 
1031 		if (sched_info_on())
1032 			entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
1033 
1034 		entry->ebx = 0;
1035 		entry->ecx = 0;
1036 		entry->edx = 0;
1037 		break;
1038 	case 0x80000000:
1039 		entry->eax = min(entry->eax, 0x8000001f);
1040 		break;
1041 	case 0x80000001:
1042 		cpuid_entry_override(entry, CPUID_8000_0001_EDX);
1043 		cpuid_entry_override(entry, CPUID_8000_0001_ECX);
1044 		break;
1045 	case 0x80000006:
1046 		/* L2 cache and TLB: pass through host info. */
1047 		break;
1048 	case 0x80000007: /* Advanced power management */
1049 		/* invariant TSC is CPUID.80000007H:EDX[8] */
1050 		entry->edx &= (1 << 8);
1051 		/* mask against host */
1052 		entry->edx &= boot_cpu_data.x86_power;
1053 		entry->eax = entry->ebx = entry->ecx = 0;
1054 		break;
1055 	case 0x80000008: {
1056 		unsigned g_phys_as = (entry->eax >> 16) & 0xff;
1057 		unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
1058 		unsigned phys_as = entry->eax & 0xff;
1059 
1060 		/*
1061 		 * If TDP (NPT) is disabled use the adjusted host MAXPHYADDR as
1062 		 * the guest operates in the same PA space as the host, i.e.
1063 		 * reductions in MAXPHYADDR for memory encryption affect shadow
1064 		 * paging, too.
1065 		 *
1066 		 * If TDP is enabled but an explicit guest MAXPHYADDR is not
1067 		 * provided, use the raw bare metal MAXPHYADDR as reductions to
1068 		 * the HPAs do not affect GPAs.
1069 		 */
1070 		if (!tdp_enabled)
1071 			g_phys_as = boot_cpu_data.x86_phys_bits;
1072 		else if (!g_phys_as)
1073 			g_phys_as = phys_as;
1074 
1075 		entry->eax = g_phys_as | (virt_as << 8);
1076 		entry->edx = 0;
1077 		cpuid_entry_override(entry, CPUID_8000_0008_EBX);
1078 		break;
1079 	}
1080 	case 0x8000000A:
1081 		if (!kvm_cpu_cap_has(X86_FEATURE_SVM)) {
1082 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
1083 			break;
1084 		}
1085 		entry->eax = 1; /* SVM revision 1 */
1086 		entry->ebx = 8; /* Lets support 8 ASIDs in case we add proper
1087 				   ASID emulation to nested SVM */
1088 		entry->ecx = 0; /* Reserved */
1089 		cpuid_entry_override(entry, CPUID_8000_000A_EDX);
1090 		break;
1091 	case 0x80000019:
1092 		entry->ecx = entry->edx = 0;
1093 		break;
1094 	case 0x8000001a:
1095 	case 0x8000001e:
1096 		break;
1097 	case 0x8000001F:
1098 		if (!kvm_cpu_cap_has(X86_FEATURE_SEV)) {
1099 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
1100 		} else {
1101 			cpuid_entry_override(entry, CPUID_8000_001F_EAX);
1102 
1103 			/*
1104 			 * Enumerate '0' for "PA bits reduction", the adjusted
1105 			 * MAXPHYADDR is enumerated directly (see 0x80000008).
1106 			 */
1107 			entry->ebx &= ~GENMASK(11, 6);
1108 		}
1109 		break;
1110 	/*Add support for Centaur's CPUID instruction*/
1111 	case 0xC0000000:
1112 		/*Just support up to 0xC0000004 now*/
1113 		entry->eax = min(entry->eax, 0xC0000004);
1114 		break;
1115 	case 0xC0000001:
1116 		cpuid_entry_override(entry, CPUID_C000_0001_EDX);
1117 		break;
1118 	case 3: /* Processor serial number */
1119 	case 5: /* MONITOR/MWAIT */
1120 	case 0xC0000002:
1121 	case 0xC0000003:
1122 	case 0xC0000004:
1123 	default:
1124 		entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
1125 		break;
1126 	}
1127 
1128 	r = 0;
1129 
1130 out:
1131 	put_cpu();
1132 
1133 	return r;
1134 }
1135 
1136 static int do_cpuid_func(struct kvm_cpuid_array *array, u32 func,
1137 			 unsigned int type)
1138 {
1139 	if (type == KVM_GET_EMULATED_CPUID)
1140 		return __do_cpuid_func_emulated(array, func);
1141 
1142 	return __do_cpuid_func(array, func);
1143 }
1144 
1145 #define CENTAUR_CPUID_SIGNATURE 0xC0000000
1146 
1147 static int get_cpuid_func(struct kvm_cpuid_array *array, u32 func,
1148 			  unsigned int type)
1149 {
1150 	u32 limit;
1151 	int r;
1152 
1153 	if (func == CENTAUR_CPUID_SIGNATURE &&
1154 	    boot_cpu_data.x86_vendor != X86_VENDOR_CENTAUR)
1155 		return 0;
1156 
1157 	r = do_cpuid_func(array, func, type);
1158 	if (r)
1159 		return r;
1160 
1161 	limit = array->entries[array->nent - 1].eax;
1162 	for (func = func + 1; func <= limit; ++func) {
1163 		r = do_cpuid_func(array, func, type);
1164 		if (r)
1165 			break;
1166 	}
1167 
1168 	return r;
1169 }
1170 
1171 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
1172 				 __u32 num_entries, unsigned int ioctl_type)
1173 {
1174 	int i;
1175 	__u32 pad[3];
1176 
1177 	if (ioctl_type != KVM_GET_EMULATED_CPUID)
1178 		return false;
1179 
1180 	/*
1181 	 * We want to make sure that ->padding is being passed clean from
1182 	 * userspace in case we want to use it for something in the future.
1183 	 *
1184 	 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
1185 	 * have to give ourselves satisfied only with the emulated side. /me
1186 	 * sheds a tear.
1187 	 */
1188 	for (i = 0; i < num_entries; i++) {
1189 		if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
1190 			return true;
1191 
1192 		if (pad[0] || pad[1] || pad[2])
1193 			return true;
1194 	}
1195 	return false;
1196 }
1197 
1198 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
1199 			    struct kvm_cpuid_entry2 __user *entries,
1200 			    unsigned int type)
1201 {
1202 	static const u32 funcs[] = {
1203 		0, 0x80000000, CENTAUR_CPUID_SIGNATURE, KVM_CPUID_SIGNATURE,
1204 	};
1205 
1206 	struct kvm_cpuid_array array = {
1207 		.nent = 0,
1208 	};
1209 	int r, i;
1210 
1211 	if (cpuid->nent < 1)
1212 		return -E2BIG;
1213 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1214 		cpuid->nent = KVM_MAX_CPUID_ENTRIES;
1215 
1216 	if (sanity_check_entries(entries, cpuid->nent, type))
1217 		return -EINVAL;
1218 
1219 	array.entries = vzalloc(array_size(sizeof(struct kvm_cpuid_entry2),
1220 					   cpuid->nent));
1221 	if (!array.entries)
1222 		return -ENOMEM;
1223 
1224 	array.maxnent = cpuid->nent;
1225 
1226 	for (i = 0; i < ARRAY_SIZE(funcs); i++) {
1227 		r = get_cpuid_func(&array, funcs[i], type);
1228 		if (r)
1229 			goto out_free;
1230 	}
1231 	cpuid->nent = array.nent;
1232 
1233 	if (copy_to_user(entries, array.entries,
1234 			 array.nent * sizeof(struct kvm_cpuid_entry2)))
1235 		r = -EFAULT;
1236 
1237 out_free:
1238 	vfree(array.entries);
1239 	return r;
1240 }
1241 
1242 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
1243 					      u32 function, u32 index)
1244 {
1245 	return cpuid_entry2_find(vcpu->arch.cpuid_entries, vcpu->arch.cpuid_nent,
1246 				 function, index);
1247 }
1248 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
1249 
1250 /*
1251  * Intel CPUID semantics treats any query for an out-of-range leaf as if the
1252  * highest basic leaf (i.e. CPUID.0H:EAX) were requested.  AMD CPUID semantics
1253  * returns all zeroes for any undefined leaf, whether or not the leaf is in
1254  * range.  Centaur/VIA follows Intel semantics.
1255  *
1256  * A leaf is considered out-of-range if its function is higher than the maximum
1257  * supported leaf of its associated class or if its associated class does not
1258  * exist.
1259  *
1260  * There are three primary classes to be considered, with their respective
1261  * ranges described as "<base> - <top>[,<base2> - <top2>] inclusive.  A primary
1262  * class exists if a guest CPUID entry for its <base> leaf exists.  For a given
1263  * class, CPUID.<base>.EAX contains the max supported leaf for the class.
1264  *
1265  *  - Basic:      0x00000000 - 0x3fffffff, 0x50000000 - 0x7fffffff
1266  *  - Hypervisor: 0x40000000 - 0x4fffffff
1267  *  - Extended:   0x80000000 - 0xbfffffff
1268  *  - Centaur:    0xc0000000 - 0xcfffffff
1269  *
1270  * The Hypervisor class is further subdivided into sub-classes that each act as
1271  * their own independent class associated with a 0x100 byte range.  E.g. if Qemu
1272  * is advertising support for both HyperV and KVM, the resulting Hypervisor
1273  * CPUID sub-classes are:
1274  *
1275  *  - HyperV:     0x40000000 - 0x400000ff
1276  *  - KVM:        0x40000100 - 0x400001ff
1277  */
1278 static struct kvm_cpuid_entry2 *
1279 get_out_of_range_cpuid_entry(struct kvm_vcpu *vcpu, u32 *fn_ptr, u32 index)
1280 {
1281 	struct kvm_cpuid_entry2 *basic, *class;
1282 	u32 function = *fn_ptr;
1283 
1284 	basic = kvm_find_cpuid_entry(vcpu, 0, 0);
1285 	if (!basic)
1286 		return NULL;
1287 
1288 	if (is_guest_vendor_amd(basic->ebx, basic->ecx, basic->edx) ||
1289 	    is_guest_vendor_hygon(basic->ebx, basic->ecx, basic->edx))
1290 		return NULL;
1291 
1292 	if (function >= 0x40000000 && function <= 0x4fffffff)
1293 		class = kvm_find_cpuid_entry(vcpu, function & 0xffffff00, 0);
1294 	else if (function >= 0xc0000000)
1295 		class = kvm_find_cpuid_entry(vcpu, 0xc0000000, 0);
1296 	else
1297 		class = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
1298 
1299 	if (class && function <= class->eax)
1300 		return NULL;
1301 
1302 	/*
1303 	 * Leaf specific adjustments are also applied when redirecting to the
1304 	 * max basic entry, e.g. if the max basic leaf is 0xb but there is no
1305 	 * entry for CPUID.0xb.index (see below), then the output value for EDX
1306 	 * needs to be pulled from CPUID.0xb.1.
1307 	 */
1308 	*fn_ptr = basic->eax;
1309 
1310 	/*
1311 	 * The class does not exist or the requested function is out of range;
1312 	 * the effective CPUID entry is the max basic leaf.  Note, the index of
1313 	 * the original requested leaf is observed!
1314 	 */
1315 	return kvm_find_cpuid_entry(vcpu, basic->eax, index);
1316 }
1317 
1318 bool kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx,
1319 	       u32 *ecx, u32 *edx, bool exact_only)
1320 {
1321 	u32 orig_function = *eax, function = *eax, index = *ecx;
1322 	struct kvm_cpuid_entry2 *entry;
1323 	bool exact, used_max_basic = false;
1324 
1325 	entry = kvm_find_cpuid_entry(vcpu, function, index);
1326 	exact = !!entry;
1327 
1328 	if (!entry && !exact_only) {
1329 		entry = get_out_of_range_cpuid_entry(vcpu, &function, index);
1330 		used_max_basic = !!entry;
1331 	}
1332 
1333 	if (entry) {
1334 		*eax = entry->eax;
1335 		*ebx = entry->ebx;
1336 		*ecx = entry->ecx;
1337 		*edx = entry->edx;
1338 		if (function == 7 && index == 0) {
1339 			u64 data;
1340 		        if (!__kvm_get_msr(vcpu, MSR_IA32_TSX_CTRL, &data, true) &&
1341 			    (data & TSX_CTRL_CPUID_CLEAR))
1342 				*ebx &= ~(F(RTM) | F(HLE));
1343 		}
1344 	} else {
1345 		*eax = *ebx = *ecx = *edx = 0;
1346 		/*
1347 		 * When leaf 0BH or 1FH is defined, CL is pass-through
1348 		 * and EDX is always the x2APIC ID, even for undefined
1349 		 * subleaves. Index 1 will exist iff the leaf is
1350 		 * implemented, so we pass through CL iff leaf 1
1351 		 * exists. EDX can be copied from any existing index.
1352 		 */
1353 		if (function == 0xb || function == 0x1f) {
1354 			entry = kvm_find_cpuid_entry(vcpu, function, 1);
1355 			if (entry) {
1356 				*ecx = index & 0xff;
1357 				*edx = entry->edx;
1358 			}
1359 		}
1360 	}
1361 	trace_kvm_cpuid(orig_function, index, *eax, *ebx, *ecx, *edx, exact,
1362 			used_max_basic);
1363 	return exact;
1364 }
1365 EXPORT_SYMBOL_GPL(kvm_cpuid);
1366 
1367 int kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1368 {
1369 	u32 eax, ebx, ecx, edx;
1370 
1371 	if (cpuid_fault_enabled(vcpu) && !kvm_require_cpl(vcpu, 0))
1372 		return 1;
1373 
1374 	eax = kvm_rax_read(vcpu);
1375 	ecx = kvm_rcx_read(vcpu);
1376 	kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx, false);
1377 	kvm_rax_write(vcpu, eax);
1378 	kvm_rbx_write(vcpu, ebx);
1379 	kvm_rcx_write(vcpu, ecx);
1380 	kvm_rdx_write(vcpu, edx);
1381 	return kvm_skip_emulated_instruction(vcpu);
1382 }
1383 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1384