xref: /openbmc/linux/arch/x86/kvm/cpuid.c (revision d87c25e8)
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].flags != orig->flags ||
137 		    e2[i].eax != orig->eax || e2[i].ebx != orig->ebx ||
138 		    e2[i].ecx != orig->ecx || e2[i].edx != orig->edx)
139 			return -EINVAL;
140 	}
141 
142 	return 0;
143 }
144 
145 static void kvm_update_kvm_cpuid_base(struct kvm_vcpu *vcpu)
146 {
147 	u32 function;
148 	struct kvm_cpuid_entry2 *entry;
149 
150 	vcpu->arch.kvm_cpuid_base = 0;
151 
152 	for_each_possible_hypervisor_cpuid_base(function) {
153 		entry = kvm_find_cpuid_entry(vcpu, function, 0);
154 
155 		if (entry) {
156 			u32 signature[3];
157 
158 			signature[0] = entry->ebx;
159 			signature[1] = entry->ecx;
160 			signature[2] = entry->edx;
161 
162 			BUILD_BUG_ON(sizeof(signature) > sizeof(KVM_SIGNATURE));
163 			if (!memcmp(signature, KVM_SIGNATURE, sizeof(signature))) {
164 				vcpu->arch.kvm_cpuid_base = function;
165 				break;
166 			}
167 		}
168 	}
169 }
170 
171 static struct kvm_cpuid_entry2 *__kvm_find_kvm_cpuid_features(struct kvm_vcpu *vcpu,
172 					      struct kvm_cpuid_entry2 *entries, int nent)
173 {
174 	u32 base = vcpu->arch.kvm_cpuid_base;
175 
176 	if (!base)
177 		return NULL;
178 
179 	return cpuid_entry2_find(entries, nent, base | KVM_CPUID_FEATURES, 0);
180 }
181 
182 static struct kvm_cpuid_entry2 *kvm_find_kvm_cpuid_features(struct kvm_vcpu *vcpu)
183 {
184 	return __kvm_find_kvm_cpuid_features(vcpu, vcpu->arch.cpuid_entries,
185 					     vcpu->arch.cpuid_nent);
186 }
187 
188 void kvm_update_pv_runtime(struct kvm_vcpu *vcpu)
189 {
190 	struct kvm_cpuid_entry2 *best = kvm_find_kvm_cpuid_features(vcpu);
191 
192 	/*
193 	 * save the feature bitmap to avoid cpuid lookup for every PV
194 	 * operation
195 	 */
196 	if (best)
197 		vcpu->arch.pv_cpuid.features = best->eax;
198 }
199 
200 /*
201  * Calculate guest's supported XCR0 taking into account guest CPUID data and
202  * supported_xcr0 (comprised of host configuration and KVM_SUPPORTED_XCR0).
203  */
204 static u64 cpuid_get_supported_xcr0(struct kvm_cpuid_entry2 *entries, int nent)
205 {
206 	struct kvm_cpuid_entry2 *best;
207 
208 	best = cpuid_entry2_find(entries, nent, 0xd, 0);
209 	if (!best)
210 		return 0;
211 
212 	return (best->eax | ((u64)best->edx << 32)) & supported_xcr0;
213 }
214 
215 static void __kvm_update_cpuid_runtime(struct kvm_vcpu *vcpu, struct kvm_cpuid_entry2 *entries,
216 				       int nent)
217 {
218 	struct kvm_cpuid_entry2 *best;
219 	u64 guest_supported_xcr0 = cpuid_get_supported_xcr0(entries, nent);
220 
221 	best = cpuid_entry2_find(entries, nent, 1, 0);
222 	if (best) {
223 		/* Update OSXSAVE bit */
224 		if (boot_cpu_has(X86_FEATURE_XSAVE))
225 			cpuid_entry_change(best, X86_FEATURE_OSXSAVE,
226 				   kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE));
227 
228 		cpuid_entry_change(best, X86_FEATURE_APIC,
229 			   vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE);
230 	}
231 
232 	best = cpuid_entry2_find(entries, nent, 7, 0);
233 	if (best && boot_cpu_has(X86_FEATURE_PKU) && best->function == 0x7)
234 		cpuid_entry_change(best, X86_FEATURE_OSPKE,
235 				   kvm_read_cr4_bits(vcpu, X86_CR4_PKE));
236 
237 	best = cpuid_entry2_find(entries, nent, 0xD, 0);
238 	if (best)
239 		best->ebx = xstate_required_size(vcpu->arch.xcr0, false);
240 
241 	best = cpuid_entry2_find(entries, nent, 0xD, 1);
242 	if (best && (cpuid_entry_has(best, X86_FEATURE_XSAVES) ||
243 		     cpuid_entry_has(best, X86_FEATURE_XSAVEC)))
244 		best->ebx = xstate_required_size(vcpu->arch.xcr0, true);
245 
246 	best = __kvm_find_kvm_cpuid_features(vcpu, entries, nent);
247 	if (kvm_hlt_in_guest(vcpu->kvm) && best &&
248 		(best->eax & (1 << KVM_FEATURE_PV_UNHALT)))
249 		best->eax &= ~(1 << KVM_FEATURE_PV_UNHALT);
250 
251 	if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT)) {
252 		best = cpuid_entry2_find(entries, nent, 0x1, 0);
253 		if (best)
254 			cpuid_entry_change(best, X86_FEATURE_MWAIT,
255 					   vcpu->arch.ia32_misc_enable_msr &
256 					   MSR_IA32_MISC_ENABLE_MWAIT);
257 	}
258 
259 	/*
260 	 * Bits 127:0 of the allowed SECS.ATTRIBUTES (CPUID.0x12.0x1) enumerate
261 	 * the supported XSAVE Feature Request Mask (XFRM), i.e. the enclave's
262 	 * requested XCR0 value.  The enclave's XFRM must be a subset of XCRO
263 	 * at the time of EENTER, thus adjust the allowed XFRM by the guest's
264 	 * supported XCR0.  Similar to XCR0 handling, FP and SSE are forced to
265 	 * '1' even on CPUs that don't support XSAVE.
266 	 */
267 	best = cpuid_entry2_find(entries, nent, 0x12, 0x1);
268 	if (best) {
269 		best->ecx &= guest_supported_xcr0 & 0xffffffff;
270 		best->edx &= guest_supported_xcr0 >> 32;
271 		best->ecx |= XFEATURE_MASK_FPSSE;
272 	}
273 }
274 
275 void kvm_update_cpuid_runtime(struct kvm_vcpu *vcpu)
276 {
277 	__kvm_update_cpuid_runtime(vcpu, vcpu->arch.cpuid_entries, vcpu->arch.cpuid_nent);
278 }
279 EXPORT_SYMBOL_GPL(kvm_update_cpuid_runtime);
280 
281 static void kvm_vcpu_after_set_cpuid(struct kvm_vcpu *vcpu)
282 {
283 	struct kvm_lapic *apic = vcpu->arch.apic;
284 	struct kvm_cpuid_entry2 *best;
285 
286 	best = kvm_find_cpuid_entry(vcpu, 1, 0);
287 	if (best && apic) {
288 		if (cpuid_entry_has(best, X86_FEATURE_TSC_DEADLINE_TIMER))
289 			apic->lapic_timer.timer_mode_mask = 3 << 17;
290 		else
291 			apic->lapic_timer.timer_mode_mask = 1 << 17;
292 
293 		kvm_apic_set_version(vcpu);
294 	}
295 
296 	vcpu->arch.guest_supported_xcr0 =
297 		cpuid_get_supported_xcr0(vcpu->arch.cpuid_entries, vcpu->arch.cpuid_nent);
298 
299 	kvm_update_pv_runtime(vcpu);
300 
301 	vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
302 	vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu);
303 
304 	kvm_pmu_refresh(vcpu);
305 	vcpu->arch.cr4_guest_rsvd_bits =
306 	    __cr4_reserved_bits(guest_cpuid_has, vcpu);
307 
308 	kvm_hv_set_cpuid(vcpu);
309 
310 	/* Invoke the vendor callback only after the above state is updated. */
311 	static_call(kvm_x86_vcpu_after_set_cpuid)(vcpu);
312 
313 	/*
314 	 * Except for the MMU, which needs to do its thing any vendor specific
315 	 * adjustments to the reserved GPA bits.
316 	 */
317 	kvm_mmu_after_set_cpuid(vcpu);
318 }
319 
320 int cpuid_query_maxphyaddr(struct kvm_vcpu *vcpu)
321 {
322 	struct kvm_cpuid_entry2 *best;
323 
324 	best = kvm_find_cpuid_entry(vcpu, 0x80000000, 0);
325 	if (!best || best->eax < 0x80000008)
326 		goto not_found;
327 	best = kvm_find_cpuid_entry(vcpu, 0x80000008, 0);
328 	if (best)
329 		return best->eax & 0xff;
330 not_found:
331 	return 36;
332 }
333 
334 /*
335  * This "raw" version returns the reserved GPA bits without any adjustments for
336  * encryption technologies that usurp bits.  The raw mask should be used if and
337  * only if hardware does _not_ strip the usurped bits, e.g. in virtual MTRRs.
338  */
339 u64 kvm_vcpu_reserved_gpa_bits_raw(struct kvm_vcpu *vcpu)
340 {
341 	return rsvd_bits(cpuid_maxphyaddr(vcpu), 63);
342 }
343 
344 static int kvm_set_cpuid(struct kvm_vcpu *vcpu, struct kvm_cpuid_entry2 *e2,
345                         int nent)
346 {
347 	int r;
348 
349 	__kvm_update_cpuid_runtime(vcpu, e2, nent);
350 
351 	/*
352 	 * KVM does not correctly handle changing guest CPUID after KVM_RUN, as
353 	 * MAXPHYADDR, GBPAGES support, AMD reserved bit behavior, etc.. aren't
354 	 * tracked in kvm_mmu_page_role.  As a result, KVM may miss guest page
355 	 * faults due to reusing SPs/SPTEs. In practice no sane VMM mucks with
356 	 * the core vCPU model on the fly. It would've been better to forbid any
357 	 * KVM_SET_CPUID{,2} calls after KVM_RUN altogether but unfortunately
358 	 * some VMMs (e.g. QEMU) reuse vCPU fds for CPU hotplug/unplug and do
359 	 * KVM_SET_CPUID{,2} again. To support this legacy behavior, check
360 	 * whether the supplied CPUID data is equal to what's already set.
361 	 */
362 	if (vcpu->arch.last_vmentry_cpu != -1) {
363 		r = kvm_cpuid_check_equal(vcpu, e2, nent);
364 		if (r)
365 			return r;
366 
367 		kvfree(e2);
368 		return 0;
369 	}
370 
371 	r = kvm_check_cpuid(vcpu, e2, nent);
372 	if (r)
373 		return r;
374 
375 	kvfree(vcpu->arch.cpuid_entries);
376 	vcpu->arch.cpuid_entries = e2;
377 	vcpu->arch.cpuid_nent = nent;
378 
379 	kvm_update_kvm_cpuid_base(vcpu);
380 	kvm_vcpu_after_set_cpuid(vcpu);
381 
382 	return 0;
383 }
384 
385 /* when an old userspace process fills a new kernel module */
386 int kvm_vcpu_ioctl_set_cpuid(struct kvm_vcpu *vcpu,
387 			     struct kvm_cpuid *cpuid,
388 			     struct kvm_cpuid_entry __user *entries)
389 {
390 	int r, i;
391 	struct kvm_cpuid_entry *e = NULL;
392 	struct kvm_cpuid_entry2 *e2 = NULL;
393 
394 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
395 		return -E2BIG;
396 
397 	if (cpuid->nent) {
398 		e = vmemdup_user(entries, array_size(sizeof(*e), cpuid->nent));
399 		if (IS_ERR(e))
400 			return PTR_ERR(e);
401 
402 		e2 = kvmalloc_array(cpuid->nent, sizeof(*e2), GFP_KERNEL_ACCOUNT);
403 		if (!e2) {
404 			r = -ENOMEM;
405 			goto out_free_cpuid;
406 		}
407 	}
408 	for (i = 0; i < cpuid->nent; i++) {
409 		e2[i].function = e[i].function;
410 		e2[i].eax = e[i].eax;
411 		e2[i].ebx = e[i].ebx;
412 		e2[i].ecx = e[i].ecx;
413 		e2[i].edx = e[i].edx;
414 		e2[i].index = 0;
415 		e2[i].flags = 0;
416 		e2[i].padding[0] = 0;
417 		e2[i].padding[1] = 0;
418 		e2[i].padding[2] = 0;
419 	}
420 
421 	r = kvm_set_cpuid(vcpu, e2, cpuid->nent);
422 	if (r)
423 		kvfree(e2);
424 
425 out_free_cpuid:
426 	kvfree(e);
427 
428 	return r;
429 }
430 
431 int kvm_vcpu_ioctl_set_cpuid2(struct kvm_vcpu *vcpu,
432 			      struct kvm_cpuid2 *cpuid,
433 			      struct kvm_cpuid_entry2 __user *entries)
434 {
435 	struct kvm_cpuid_entry2 *e2 = NULL;
436 	int r;
437 
438 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
439 		return -E2BIG;
440 
441 	if (cpuid->nent) {
442 		e2 = vmemdup_user(entries, array_size(sizeof(*e2), cpuid->nent));
443 		if (IS_ERR(e2))
444 			return PTR_ERR(e2);
445 	}
446 
447 	r = kvm_set_cpuid(vcpu, e2, cpuid->nent);
448 	if (r)
449 		kvfree(e2);
450 
451 	return r;
452 }
453 
454 int kvm_vcpu_ioctl_get_cpuid2(struct kvm_vcpu *vcpu,
455 			      struct kvm_cpuid2 *cpuid,
456 			      struct kvm_cpuid_entry2 __user *entries)
457 {
458 	int r;
459 
460 	r = -E2BIG;
461 	if (cpuid->nent < vcpu->arch.cpuid_nent)
462 		goto out;
463 	r = -EFAULT;
464 	if (copy_to_user(entries, vcpu->arch.cpuid_entries,
465 			 vcpu->arch.cpuid_nent * sizeof(struct kvm_cpuid_entry2)))
466 		goto out;
467 	return 0;
468 
469 out:
470 	cpuid->nent = vcpu->arch.cpuid_nent;
471 	return r;
472 }
473 
474 /* Mask kvm_cpu_caps for @leaf with the raw CPUID capabilities of this CPU. */
475 static __always_inline void __kvm_cpu_cap_mask(unsigned int leaf)
476 {
477 	const struct cpuid_reg cpuid = x86_feature_cpuid(leaf * 32);
478 	struct kvm_cpuid_entry2 entry;
479 
480 	reverse_cpuid_check(leaf);
481 
482 	cpuid_count(cpuid.function, cpuid.index,
483 		    &entry.eax, &entry.ebx, &entry.ecx, &entry.edx);
484 
485 	kvm_cpu_caps[leaf] &= *__cpuid_entry_get_reg(&entry, cpuid.reg);
486 }
487 
488 static __always_inline
489 void kvm_cpu_cap_init_scattered(enum kvm_only_cpuid_leafs leaf, u32 mask)
490 {
491 	/* Use kvm_cpu_cap_mask for non-scattered leafs. */
492 	BUILD_BUG_ON(leaf < NCAPINTS);
493 
494 	kvm_cpu_caps[leaf] = mask;
495 
496 	__kvm_cpu_cap_mask(leaf);
497 }
498 
499 static __always_inline void kvm_cpu_cap_mask(enum cpuid_leafs leaf, u32 mask)
500 {
501 	/* Use kvm_cpu_cap_init_scattered for scattered leafs. */
502 	BUILD_BUG_ON(leaf >= NCAPINTS);
503 
504 	kvm_cpu_caps[leaf] &= mask;
505 
506 	__kvm_cpu_cap_mask(leaf);
507 }
508 
509 void kvm_set_cpu_caps(void)
510 {
511 #ifdef CONFIG_X86_64
512 	unsigned int f_gbpages = F(GBPAGES);
513 	unsigned int f_lm = F(LM);
514 	unsigned int f_xfd = F(XFD);
515 #else
516 	unsigned int f_gbpages = 0;
517 	unsigned int f_lm = 0;
518 	unsigned int f_xfd = 0;
519 #endif
520 	memset(kvm_cpu_caps, 0, sizeof(kvm_cpu_caps));
521 
522 	BUILD_BUG_ON(sizeof(kvm_cpu_caps) - (NKVMCAPINTS * sizeof(*kvm_cpu_caps)) >
523 		     sizeof(boot_cpu_data.x86_capability));
524 
525 	memcpy(&kvm_cpu_caps, &boot_cpu_data.x86_capability,
526 	       sizeof(kvm_cpu_caps) - (NKVMCAPINTS * sizeof(*kvm_cpu_caps)));
527 
528 	kvm_cpu_cap_mask(CPUID_1_ECX,
529 		/*
530 		 * NOTE: MONITOR (and MWAIT) are emulated as NOP, but *not*
531 		 * advertised to guests via CPUID!
532 		 */
533 		F(XMM3) | F(PCLMULQDQ) | 0 /* DTES64, MONITOR */ |
534 		0 /* DS-CPL, VMX, SMX, EST */ |
535 		0 /* TM2 */ | F(SSSE3) | 0 /* CNXT-ID */ | 0 /* Reserved */ |
536 		F(FMA) | F(CX16) | 0 /* xTPR Update */ | F(PDCM) |
537 		F(PCID) | 0 /* Reserved, DCA */ | F(XMM4_1) |
538 		F(XMM4_2) | F(X2APIC) | F(MOVBE) | F(POPCNT) |
539 		0 /* Reserved*/ | F(AES) | F(XSAVE) | 0 /* OSXSAVE */ | F(AVX) |
540 		F(F16C) | F(RDRAND)
541 	);
542 	/* KVM emulates x2apic in software irrespective of host support. */
543 	kvm_cpu_cap_set(X86_FEATURE_X2APIC);
544 
545 	kvm_cpu_cap_mask(CPUID_1_EDX,
546 		F(FPU) | F(VME) | F(DE) | F(PSE) |
547 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
548 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SEP) |
549 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
550 		F(PAT) | F(PSE36) | 0 /* PSN */ | F(CLFLUSH) |
551 		0 /* Reserved, DS, ACPI */ | F(MMX) |
552 		F(FXSR) | F(XMM) | F(XMM2) | F(SELFSNOOP) |
553 		0 /* HTT, TM, Reserved, PBE */
554 	);
555 
556 	kvm_cpu_cap_mask(CPUID_7_0_EBX,
557 		F(FSGSBASE) | F(SGX) | F(BMI1) | F(HLE) | F(AVX2) |
558 		F(FDP_EXCPTN_ONLY) | F(SMEP) | F(BMI2) | F(ERMS) | F(INVPCID) |
559 		F(RTM) | F(ZERO_FCS_FDS) | 0 /*MPX*/ | F(AVX512F) |
560 		F(AVX512DQ) | F(RDSEED) | F(ADX) | F(SMAP) | F(AVX512IFMA) |
561 		F(CLFLUSHOPT) | F(CLWB) | 0 /*INTEL_PT*/ | F(AVX512PF) |
562 		F(AVX512ER) | F(AVX512CD) | F(SHA_NI) | F(AVX512BW) |
563 		F(AVX512VL));
564 
565 	kvm_cpu_cap_mask(CPUID_7_ECX,
566 		F(AVX512VBMI) | F(LA57) | F(PKU) | 0 /*OSPKE*/ | F(RDPID) |
567 		F(AVX512_VPOPCNTDQ) | F(UMIP) | F(AVX512_VBMI2) | F(GFNI) |
568 		F(VAES) | F(VPCLMULQDQ) | F(AVX512_VNNI) | F(AVX512_BITALG) |
569 		F(CLDEMOTE) | F(MOVDIRI) | F(MOVDIR64B) | 0 /*WAITPKG*/ |
570 		F(SGX_LC) | F(BUS_LOCK_DETECT)
571 	);
572 	/* Set LA57 based on hardware capability. */
573 	if (cpuid_ecx(7) & F(LA57))
574 		kvm_cpu_cap_set(X86_FEATURE_LA57);
575 
576 	/*
577 	 * PKU not yet implemented for shadow paging and requires OSPKE
578 	 * to be set on the host. Clear it if that is not the case
579 	 */
580 	if (!tdp_enabled || !boot_cpu_has(X86_FEATURE_OSPKE))
581 		kvm_cpu_cap_clear(X86_FEATURE_PKU);
582 
583 	kvm_cpu_cap_mask(CPUID_7_EDX,
584 		F(AVX512_4VNNIW) | F(AVX512_4FMAPS) | F(SPEC_CTRL) |
585 		F(SPEC_CTRL_SSBD) | F(ARCH_CAPABILITIES) | F(INTEL_STIBP) |
586 		F(MD_CLEAR) | F(AVX512_VP2INTERSECT) | F(FSRM) |
587 		F(SERIALIZE) | F(TSXLDTRK) | F(AVX512_FP16) |
588 		F(AMX_TILE) | F(AMX_INT8) | F(AMX_BF16)
589 	);
590 
591 	/* TSC_ADJUST and ARCH_CAPABILITIES are emulated in software. */
592 	kvm_cpu_cap_set(X86_FEATURE_TSC_ADJUST);
593 	kvm_cpu_cap_set(X86_FEATURE_ARCH_CAPABILITIES);
594 
595 	if (boot_cpu_has(X86_FEATURE_IBPB) && boot_cpu_has(X86_FEATURE_IBRS))
596 		kvm_cpu_cap_set(X86_FEATURE_SPEC_CTRL);
597 	if (boot_cpu_has(X86_FEATURE_STIBP))
598 		kvm_cpu_cap_set(X86_FEATURE_INTEL_STIBP);
599 	if (boot_cpu_has(X86_FEATURE_AMD_SSBD))
600 		kvm_cpu_cap_set(X86_FEATURE_SPEC_CTRL_SSBD);
601 
602 	kvm_cpu_cap_mask(CPUID_7_1_EAX,
603 		F(AVX_VNNI) | F(AVX512_BF16)
604 	);
605 
606 	kvm_cpu_cap_mask(CPUID_D_1_EAX,
607 		F(XSAVEOPT) | F(XSAVEC) | F(XGETBV1) | F(XSAVES) | f_xfd
608 	);
609 
610 	kvm_cpu_cap_init_scattered(CPUID_12_EAX,
611 		SF(SGX1) | SF(SGX2)
612 	);
613 
614 	kvm_cpu_cap_mask(CPUID_8000_0001_ECX,
615 		F(LAHF_LM) | F(CMP_LEGACY) | 0 /*SVM*/ | 0 /* ExtApicSpace */ |
616 		F(CR8_LEGACY) | F(ABM) | F(SSE4A) | F(MISALIGNSSE) |
617 		F(3DNOWPREFETCH) | F(OSVW) | 0 /* IBS */ | F(XOP) |
618 		0 /* SKINIT, WDT, LWP */ | F(FMA4) | F(TBM) |
619 		F(TOPOEXT) | 0 /* PERFCTR_CORE */
620 	);
621 
622 	kvm_cpu_cap_mask(CPUID_8000_0001_EDX,
623 		F(FPU) | F(VME) | F(DE) | F(PSE) |
624 		F(TSC) | F(MSR) | F(PAE) | F(MCE) |
625 		F(CX8) | F(APIC) | 0 /* Reserved */ | F(SYSCALL) |
626 		F(MTRR) | F(PGE) | F(MCA) | F(CMOV) |
627 		F(PAT) | F(PSE36) | 0 /* Reserved */ |
628 		F(NX) | 0 /* Reserved */ | F(MMXEXT) | F(MMX) |
629 		F(FXSR) | F(FXSR_OPT) | f_gbpages | F(RDTSCP) |
630 		0 /* Reserved */ | f_lm | F(3DNOWEXT) | F(3DNOW)
631 	);
632 
633 	if (!tdp_enabled && IS_ENABLED(CONFIG_X86_64))
634 		kvm_cpu_cap_set(X86_FEATURE_GBPAGES);
635 
636 	kvm_cpu_cap_mask(CPUID_8000_0008_EBX,
637 		F(CLZERO) | F(XSAVEERPTR) |
638 		F(WBNOINVD) | F(AMD_IBPB) | F(AMD_IBRS) | F(AMD_SSBD) | F(VIRT_SSBD) |
639 		F(AMD_SSB_NO) | F(AMD_STIBP) | F(AMD_STIBP_ALWAYS_ON) |
640 		__feature_bit(KVM_X86_FEATURE_PSFD)
641 	);
642 
643 	/*
644 	 * AMD has separate bits for each SPEC_CTRL bit.
645 	 * arch/x86/kernel/cpu/bugs.c is kind enough to
646 	 * record that in cpufeatures so use them.
647 	 */
648 	if (boot_cpu_has(X86_FEATURE_IBPB))
649 		kvm_cpu_cap_set(X86_FEATURE_AMD_IBPB);
650 	if (boot_cpu_has(X86_FEATURE_IBRS))
651 		kvm_cpu_cap_set(X86_FEATURE_AMD_IBRS);
652 	if (boot_cpu_has(X86_FEATURE_STIBP))
653 		kvm_cpu_cap_set(X86_FEATURE_AMD_STIBP);
654 	if (boot_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD))
655 		kvm_cpu_cap_set(X86_FEATURE_AMD_SSBD);
656 	if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
657 		kvm_cpu_cap_set(X86_FEATURE_AMD_SSB_NO);
658 	/*
659 	 * The preference is to use SPEC CTRL MSR instead of the
660 	 * VIRT_SPEC MSR.
661 	 */
662 	if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD) &&
663 	    !boot_cpu_has(X86_FEATURE_AMD_SSBD))
664 		kvm_cpu_cap_set(X86_FEATURE_VIRT_SSBD);
665 
666 	/*
667 	 * Hide all SVM features by default, SVM will set the cap bits for
668 	 * features it emulates and/or exposes for L1.
669 	 */
670 	kvm_cpu_cap_mask(CPUID_8000_000A_EDX, 0);
671 
672 	kvm_cpu_cap_mask(CPUID_8000_001F_EAX,
673 		0 /* SME */ | F(SEV) | 0 /* VM_PAGE_FLUSH */ | F(SEV_ES) |
674 		F(SME_COHERENT));
675 
676 	kvm_cpu_cap_mask(CPUID_C000_0001_EDX,
677 		F(XSTORE) | F(XSTORE_EN) | F(XCRYPT) | F(XCRYPT_EN) |
678 		F(ACE2) | F(ACE2_EN) | F(PHE) | F(PHE_EN) |
679 		F(PMM) | F(PMM_EN)
680 	);
681 
682 	/*
683 	 * Hide RDTSCP and RDPID if either feature is reported as supported but
684 	 * probing MSR_TSC_AUX failed.  This is purely a sanity check and
685 	 * should never happen, but the guest will likely crash if RDTSCP or
686 	 * RDPID is misreported, and KVM has botched MSR_TSC_AUX emulation in
687 	 * the past.  For example, the sanity check may fire if this instance of
688 	 * KVM is running as L1 on top of an older, broken KVM.
689 	 */
690 	if (WARN_ON((kvm_cpu_cap_has(X86_FEATURE_RDTSCP) ||
691 		     kvm_cpu_cap_has(X86_FEATURE_RDPID)) &&
692 		     !kvm_is_supported_user_return_msr(MSR_TSC_AUX))) {
693 		kvm_cpu_cap_clear(X86_FEATURE_RDTSCP);
694 		kvm_cpu_cap_clear(X86_FEATURE_RDPID);
695 	}
696 }
697 EXPORT_SYMBOL_GPL(kvm_set_cpu_caps);
698 
699 struct kvm_cpuid_array {
700 	struct kvm_cpuid_entry2 *entries;
701 	int maxnent;
702 	int nent;
703 };
704 
705 static struct kvm_cpuid_entry2 *do_host_cpuid(struct kvm_cpuid_array *array,
706 					      u32 function, u32 index)
707 {
708 	struct kvm_cpuid_entry2 *entry;
709 
710 	if (array->nent >= array->maxnent)
711 		return NULL;
712 
713 	entry = &array->entries[array->nent++];
714 
715 	entry->function = function;
716 	entry->index = index;
717 	entry->flags = 0;
718 
719 	cpuid_count(entry->function, entry->index,
720 		    &entry->eax, &entry->ebx, &entry->ecx, &entry->edx);
721 
722 	switch (function) {
723 	case 4:
724 	case 7:
725 	case 0xb:
726 	case 0xd:
727 	case 0xf:
728 	case 0x10:
729 	case 0x12:
730 	case 0x14:
731 	case 0x17:
732 	case 0x18:
733 	case 0x1d:
734 	case 0x1e:
735 	case 0x1f:
736 	case 0x8000001d:
737 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
738 		break;
739 	}
740 
741 	return entry;
742 }
743 
744 static int __do_cpuid_func_emulated(struct kvm_cpuid_array *array, u32 func)
745 {
746 	struct kvm_cpuid_entry2 *entry;
747 
748 	if (array->nent >= array->maxnent)
749 		return -E2BIG;
750 
751 	entry = &array->entries[array->nent];
752 	entry->function = func;
753 	entry->index = 0;
754 	entry->flags = 0;
755 
756 	switch (func) {
757 	case 0:
758 		entry->eax = 7;
759 		++array->nent;
760 		break;
761 	case 1:
762 		entry->ecx = F(MOVBE);
763 		++array->nent;
764 		break;
765 	case 7:
766 		entry->flags |= KVM_CPUID_FLAG_SIGNIFCANT_INDEX;
767 		entry->eax = 0;
768 		if (kvm_cpu_cap_has(X86_FEATURE_RDTSCP))
769 			entry->ecx = F(RDPID);
770 		++array->nent;
771 		break;
772 	default:
773 		break;
774 	}
775 
776 	return 0;
777 }
778 
779 static inline int __do_cpuid_func(struct kvm_cpuid_array *array, u32 function)
780 {
781 	struct kvm_cpuid_entry2 *entry;
782 	int r, i, max_idx;
783 
784 	/* all calls to cpuid_count() should be made on the same cpu */
785 	get_cpu();
786 
787 	r = -E2BIG;
788 
789 	entry = do_host_cpuid(array, function, 0);
790 	if (!entry)
791 		goto out;
792 
793 	switch (function) {
794 	case 0:
795 		/* Limited to the highest leaf implemented in KVM. */
796 		entry->eax = min(entry->eax, 0x1fU);
797 		break;
798 	case 1:
799 		cpuid_entry_override(entry, CPUID_1_EDX);
800 		cpuid_entry_override(entry, CPUID_1_ECX);
801 		break;
802 	case 2:
803 		/*
804 		 * On ancient CPUs, function 2 entries are STATEFUL.  That is,
805 		 * CPUID(function=2, index=0) may return different results each
806 		 * time, with the least-significant byte in EAX enumerating the
807 		 * number of times software should do CPUID(2, 0).
808 		 *
809 		 * Modern CPUs, i.e. every CPU KVM has *ever* run on are less
810 		 * idiotic.  Intel's SDM states that EAX & 0xff "will always
811 		 * return 01H. Software should ignore this value and not
812 		 * interpret it as an informational descriptor", while AMD's
813 		 * APM states that CPUID(2) is reserved.
814 		 *
815 		 * WARN if a frankenstein CPU that supports virtualization and
816 		 * a stateful CPUID.0x2 is encountered.
817 		 */
818 		WARN_ON_ONCE((entry->eax & 0xff) > 1);
819 		break;
820 	/* functions 4 and 0x8000001d have additional index. */
821 	case 4:
822 	case 0x8000001d:
823 		/*
824 		 * Read entries until the cache type in the previous entry is
825 		 * zero, i.e. indicates an invalid entry.
826 		 */
827 		for (i = 1; entry->eax & 0x1f; ++i) {
828 			entry = do_host_cpuid(array, function, i);
829 			if (!entry)
830 				goto out;
831 		}
832 		break;
833 	case 6: /* Thermal management */
834 		entry->eax = 0x4; /* allow ARAT */
835 		entry->ebx = 0;
836 		entry->ecx = 0;
837 		entry->edx = 0;
838 		break;
839 	/* function 7 has additional index. */
840 	case 7:
841 		entry->eax = min(entry->eax, 1u);
842 		cpuid_entry_override(entry, CPUID_7_0_EBX);
843 		cpuid_entry_override(entry, CPUID_7_ECX);
844 		cpuid_entry_override(entry, CPUID_7_EDX);
845 
846 		/* KVM only supports 0x7.0 and 0x7.1, capped above via min(). */
847 		if (entry->eax == 1) {
848 			entry = do_host_cpuid(array, function, 1);
849 			if (!entry)
850 				goto out;
851 
852 			cpuid_entry_override(entry, CPUID_7_1_EAX);
853 			entry->ebx = 0;
854 			entry->ecx = 0;
855 			entry->edx = 0;
856 		}
857 		break;
858 	case 9:
859 		break;
860 	case 0xa: { /* Architectural Performance Monitoring */
861 		struct x86_pmu_capability cap;
862 		union cpuid10_eax eax;
863 		union cpuid10_edx edx;
864 
865 		perf_get_x86_pmu_capability(&cap);
866 
867 		/*
868 		 * The guest architecture pmu is only supported if the architecture
869 		 * pmu exists on the host and the module parameters allow it.
870 		 */
871 		if (!cap.version || !enable_pmu)
872 			memset(&cap, 0, sizeof(cap));
873 
874 		eax.split.version_id = min(cap.version, 2);
875 		eax.split.num_counters = cap.num_counters_gp;
876 		eax.split.bit_width = cap.bit_width_gp;
877 		eax.split.mask_length = cap.events_mask_len;
878 
879 		edx.split.num_counters_fixed = min(cap.num_counters_fixed, MAX_FIXED_COUNTERS);
880 		edx.split.bit_width_fixed = cap.bit_width_fixed;
881 		if (cap.version)
882 			edx.split.anythread_deprecated = 1;
883 		edx.split.reserved1 = 0;
884 		edx.split.reserved2 = 0;
885 
886 		entry->eax = eax.full;
887 		entry->ebx = cap.events_mask;
888 		entry->ecx = 0;
889 		entry->edx = edx.full;
890 		break;
891 	}
892 	/*
893 	 * Per Intel's SDM, the 0x1f is a superset of 0xb,
894 	 * thus they can be handled by common code.
895 	 */
896 	case 0x1f:
897 	case 0xb:
898 		/*
899 		 * Populate entries until the level type (ECX[15:8]) of the
900 		 * previous entry is zero.  Note, CPUID EAX.{0x1f,0xb}.0 is
901 		 * the starting entry, filled by the primary do_host_cpuid().
902 		 */
903 		for (i = 1; entry->ecx & 0xff00; ++i) {
904 			entry = do_host_cpuid(array, function, i);
905 			if (!entry)
906 				goto out;
907 		}
908 		break;
909 	case 0xd: {
910 		u64 permitted_xcr0 = supported_xcr0 & xstate_get_guest_group_perm();
911 		u64 permitted_xss = supported_xss;
912 
913 		entry->eax &= permitted_xcr0;
914 		entry->ebx = xstate_required_size(permitted_xcr0, false);
915 		entry->ecx = entry->ebx;
916 		entry->edx &= permitted_xcr0 >> 32;
917 		if (!permitted_xcr0)
918 			break;
919 
920 		entry = do_host_cpuid(array, function, 1);
921 		if (!entry)
922 			goto out;
923 
924 		cpuid_entry_override(entry, CPUID_D_1_EAX);
925 		if (entry->eax & (F(XSAVES)|F(XSAVEC)))
926 			entry->ebx = xstate_required_size(permitted_xcr0 | permitted_xss,
927 							  true);
928 		else {
929 			WARN_ON_ONCE(permitted_xss != 0);
930 			entry->ebx = 0;
931 		}
932 		entry->ecx &= permitted_xss;
933 		entry->edx &= permitted_xss >> 32;
934 
935 		for (i = 2; i < 64; ++i) {
936 			bool s_state;
937 			if (permitted_xcr0 & BIT_ULL(i))
938 				s_state = false;
939 			else if (permitted_xss & BIT_ULL(i))
940 				s_state = true;
941 			else
942 				continue;
943 
944 			entry = do_host_cpuid(array, function, i);
945 			if (!entry)
946 				goto out;
947 
948 			/*
949 			 * The supported check above should have filtered out
950 			 * invalid sub-leafs.  Only valid sub-leafs should
951 			 * reach this point, and they should have a non-zero
952 			 * save state size.  Furthermore, check whether the
953 			 * processor agrees with permitted_xcr0/permitted_xss
954 			 * on whether this is an XCR0- or IA32_XSS-managed area.
955 			 */
956 			if (WARN_ON_ONCE(!entry->eax || (entry->ecx & 0x1) != s_state)) {
957 				--array->nent;
958 				continue;
959 			}
960 
961 			if (!kvm_cpu_cap_has(X86_FEATURE_XFD))
962 				entry->ecx &= ~BIT_ULL(2);
963 			entry->edx = 0;
964 		}
965 		break;
966 	}
967 	case 0x12:
968 		/* Intel SGX */
969 		if (!kvm_cpu_cap_has(X86_FEATURE_SGX)) {
970 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
971 			break;
972 		}
973 
974 		/*
975 		 * Index 0: Sub-features, MISCSELECT (a.k.a extended features)
976 		 * and max enclave sizes.   The SGX sub-features and MISCSELECT
977 		 * are restricted by kernel and KVM capabilities (like most
978 		 * feature flags), while enclave size is unrestricted.
979 		 */
980 		cpuid_entry_override(entry, CPUID_12_EAX);
981 		entry->ebx &= SGX_MISC_EXINFO;
982 
983 		entry = do_host_cpuid(array, function, 1);
984 		if (!entry)
985 			goto out;
986 
987 		/*
988 		 * Index 1: SECS.ATTRIBUTES.  ATTRIBUTES are restricted a la
989 		 * feature flags.  Advertise all supported flags, including
990 		 * privileged attributes that require explicit opt-in from
991 		 * userspace.  ATTRIBUTES.XFRM is not adjusted as userspace is
992 		 * expected to derive it from supported XCR0.
993 		 */
994 		entry->eax &= SGX_ATTR_DEBUG | SGX_ATTR_MODE64BIT |
995 			      SGX_ATTR_PROVISIONKEY | SGX_ATTR_EINITTOKENKEY |
996 			      SGX_ATTR_KSS;
997 		entry->ebx &= 0;
998 		break;
999 	/* Intel PT */
1000 	case 0x14:
1001 		if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT)) {
1002 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
1003 			break;
1004 		}
1005 
1006 		for (i = 1, max_idx = entry->eax; i <= max_idx; ++i) {
1007 			if (!do_host_cpuid(array, function, i))
1008 				goto out;
1009 		}
1010 		break;
1011 	/* Intel AMX TILE */
1012 	case 0x1d:
1013 		if (!kvm_cpu_cap_has(X86_FEATURE_AMX_TILE)) {
1014 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
1015 			break;
1016 		}
1017 
1018 		for (i = 1, max_idx = entry->eax; i <= max_idx; ++i) {
1019 			if (!do_host_cpuid(array, function, i))
1020 				goto out;
1021 		}
1022 		break;
1023 	case 0x1e: /* TMUL information */
1024 		if (!kvm_cpu_cap_has(X86_FEATURE_AMX_TILE)) {
1025 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
1026 			break;
1027 		}
1028 		break;
1029 	case KVM_CPUID_SIGNATURE: {
1030 		const u32 *sigptr = (const u32 *)KVM_SIGNATURE;
1031 		entry->eax = KVM_CPUID_FEATURES;
1032 		entry->ebx = sigptr[0];
1033 		entry->ecx = sigptr[1];
1034 		entry->edx = sigptr[2];
1035 		break;
1036 	}
1037 	case KVM_CPUID_FEATURES:
1038 		entry->eax = (1 << KVM_FEATURE_CLOCKSOURCE) |
1039 			     (1 << KVM_FEATURE_NOP_IO_DELAY) |
1040 			     (1 << KVM_FEATURE_CLOCKSOURCE2) |
1041 			     (1 << KVM_FEATURE_ASYNC_PF) |
1042 			     (1 << KVM_FEATURE_PV_EOI) |
1043 			     (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT) |
1044 			     (1 << KVM_FEATURE_PV_UNHALT) |
1045 			     (1 << KVM_FEATURE_PV_TLB_FLUSH) |
1046 			     (1 << KVM_FEATURE_ASYNC_PF_VMEXIT) |
1047 			     (1 << KVM_FEATURE_PV_SEND_IPI) |
1048 			     (1 << KVM_FEATURE_POLL_CONTROL) |
1049 			     (1 << KVM_FEATURE_PV_SCHED_YIELD) |
1050 			     (1 << KVM_FEATURE_ASYNC_PF_INT);
1051 
1052 		if (sched_info_on())
1053 			entry->eax |= (1 << KVM_FEATURE_STEAL_TIME);
1054 
1055 		entry->ebx = 0;
1056 		entry->ecx = 0;
1057 		entry->edx = 0;
1058 		break;
1059 	case 0x80000000:
1060 		entry->eax = min(entry->eax, 0x8000001f);
1061 		break;
1062 	case 0x80000001:
1063 		cpuid_entry_override(entry, CPUID_8000_0001_EDX);
1064 		cpuid_entry_override(entry, CPUID_8000_0001_ECX);
1065 		break;
1066 	case 0x80000006:
1067 		/* L2 cache and TLB: pass through host info. */
1068 		break;
1069 	case 0x80000007: /* Advanced power management */
1070 		/* invariant TSC is CPUID.80000007H:EDX[8] */
1071 		entry->edx &= (1 << 8);
1072 		/* mask against host */
1073 		entry->edx &= boot_cpu_data.x86_power;
1074 		entry->eax = entry->ebx = entry->ecx = 0;
1075 		break;
1076 	case 0x80000008: {
1077 		unsigned g_phys_as = (entry->eax >> 16) & 0xff;
1078 		unsigned virt_as = max((entry->eax >> 8) & 0xff, 48U);
1079 		unsigned phys_as = entry->eax & 0xff;
1080 
1081 		/*
1082 		 * If TDP (NPT) is disabled use the adjusted host MAXPHYADDR as
1083 		 * the guest operates in the same PA space as the host, i.e.
1084 		 * reductions in MAXPHYADDR for memory encryption affect shadow
1085 		 * paging, too.
1086 		 *
1087 		 * If TDP is enabled but an explicit guest MAXPHYADDR is not
1088 		 * provided, use the raw bare metal MAXPHYADDR as reductions to
1089 		 * the HPAs do not affect GPAs.
1090 		 */
1091 		if (!tdp_enabled)
1092 			g_phys_as = boot_cpu_data.x86_phys_bits;
1093 		else if (!g_phys_as)
1094 			g_phys_as = phys_as;
1095 
1096 		entry->eax = g_phys_as | (virt_as << 8);
1097 		entry->edx = 0;
1098 		cpuid_entry_override(entry, CPUID_8000_0008_EBX);
1099 		break;
1100 	}
1101 	case 0x8000000A:
1102 		if (!kvm_cpu_cap_has(X86_FEATURE_SVM)) {
1103 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
1104 			break;
1105 		}
1106 		entry->eax = 1; /* SVM revision 1 */
1107 		entry->ebx = 8; /* Lets support 8 ASIDs in case we add proper
1108 				   ASID emulation to nested SVM */
1109 		entry->ecx = 0; /* Reserved */
1110 		cpuid_entry_override(entry, CPUID_8000_000A_EDX);
1111 		break;
1112 	case 0x80000019:
1113 		entry->ecx = entry->edx = 0;
1114 		break;
1115 	case 0x8000001a:
1116 	case 0x8000001e:
1117 		break;
1118 	case 0x8000001F:
1119 		if (!kvm_cpu_cap_has(X86_FEATURE_SEV)) {
1120 			entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
1121 		} else {
1122 			cpuid_entry_override(entry, CPUID_8000_001F_EAX);
1123 
1124 			/*
1125 			 * Enumerate '0' for "PA bits reduction", the adjusted
1126 			 * MAXPHYADDR is enumerated directly (see 0x80000008).
1127 			 */
1128 			entry->ebx &= ~GENMASK(11, 6);
1129 		}
1130 		break;
1131 	/*Add support for Centaur's CPUID instruction*/
1132 	case 0xC0000000:
1133 		/*Just support up to 0xC0000004 now*/
1134 		entry->eax = min(entry->eax, 0xC0000004);
1135 		break;
1136 	case 0xC0000001:
1137 		cpuid_entry_override(entry, CPUID_C000_0001_EDX);
1138 		break;
1139 	case 3: /* Processor serial number */
1140 	case 5: /* MONITOR/MWAIT */
1141 	case 0xC0000002:
1142 	case 0xC0000003:
1143 	case 0xC0000004:
1144 	default:
1145 		entry->eax = entry->ebx = entry->ecx = entry->edx = 0;
1146 		break;
1147 	}
1148 
1149 	r = 0;
1150 
1151 out:
1152 	put_cpu();
1153 
1154 	return r;
1155 }
1156 
1157 static int do_cpuid_func(struct kvm_cpuid_array *array, u32 func,
1158 			 unsigned int type)
1159 {
1160 	if (type == KVM_GET_EMULATED_CPUID)
1161 		return __do_cpuid_func_emulated(array, func);
1162 
1163 	return __do_cpuid_func(array, func);
1164 }
1165 
1166 #define CENTAUR_CPUID_SIGNATURE 0xC0000000
1167 
1168 static int get_cpuid_func(struct kvm_cpuid_array *array, u32 func,
1169 			  unsigned int type)
1170 {
1171 	u32 limit;
1172 	int r;
1173 
1174 	if (func == CENTAUR_CPUID_SIGNATURE &&
1175 	    boot_cpu_data.x86_vendor != X86_VENDOR_CENTAUR)
1176 		return 0;
1177 
1178 	r = do_cpuid_func(array, func, type);
1179 	if (r)
1180 		return r;
1181 
1182 	limit = array->entries[array->nent - 1].eax;
1183 	for (func = func + 1; func <= limit; ++func) {
1184 		r = do_cpuid_func(array, func, type);
1185 		if (r)
1186 			break;
1187 	}
1188 
1189 	return r;
1190 }
1191 
1192 static bool sanity_check_entries(struct kvm_cpuid_entry2 __user *entries,
1193 				 __u32 num_entries, unsigned int ioctl_type)
1194 {
1195 	int i;
1196 	__u32 pad[3];
1197 
1198 	if (ioctl_type != KVM_GET_EMULATED_CPUID)
1199 		return false;
1200 
1201 	/*
1202 	 * We want to make sure that ->padding is being passed clean from
1203 	 * userspace in case we want to use it for something in the future.
1204 	 *
1205 	 * Sadly, this wasn't enforced for KVM_GET_SUPPORTED_CPUID and so we
1206 	 * have to give ourselves satisfied only with the emulated side. /me
1207 	 * sheds a tear.
1208 	 */
1209 	for (i = 0; i < num_entries; i++) {
1210 		if (copy_from_user(pad, entries[i].padding, sizeof(pad)))
1211 			return true;
1212 
1213 		if (pad[0] || pad[1] || pad[2])
1214 			return true;
1215 	}
1216 	return false;
1217 }
1218 
1219 int kvm_dev_ioctl_get_cpuid(struct kvm_cpuid2 *cpuid,
1220 			    struct kvm_cpuid_entry2 __user *entries,
1221 			    unsigned int type)
1222 {
1223 	static const u32 funcs[] = {
1224 		0, 0x80000000, CENTAUR_CPUID_SIGNATURE, KVM_CPUID_SIGNATURE,
1225 	};
1226 
1227 	struct kvm_cpuid_array array = {
1228 		.nent = 0,
1229 	};
1230 	int r, i;
1231 
1232 	if (cpuid->nent < 1)
1233 		return -E2BIG;
1234 	if (cpuid->nent > KVM_MAX_CPUID_ENTRIES)
1235 		cpuid->nent = KVM_MAX_CPUID_ENTRIES;
1236 
1237 	if (sanity_check_entries(entries, cpuid->nent, type))
1238 		return -EINVAL;
1239 
1240 	array.entries = vzalloc(array_size(sizeof(struct kvm_cpuid_entry2),
1241 					   cpuid->nent));
1242 	if (!array.entries)
1243 		return -ENOMEM;
1244 
1245 	array.maxnent = cpuid->nent;
1246 
1247 	for (i = 0; i < ARRAY_SIZE(funcs); i++) {
1248 		r = get_cpuid_func(&array, funcs[i], type);
1249 		if (r)
1250 			goto out_free;
1251 	}
1252 	cpuid->nent = array.nent;
1253 
1254 	if (copy_to_user(entries, array.entries,
1255 			 array.nent * sizeof(struct kvm_cpuid_entry2)))
1256 		r = -EFAULT;
1257 
1258 out_free:
1259 	vfree(array.entries);
1260 	return r;
1261 }
1262 
1263 struct kvm_cpuid_entry2 *kvm_find_cpuid_entry(struct kvm_vcpu *vcpu,
1264 					      u32 function, u32 index)
1265 {
1266 	return cpuid_entry2_find(vcpu->arch.cpuid_entries, vcpu->arch.cpuid_nent,
1267 				 function, index);
1268 }
1269 EXPORT_SYMBOL_GPL(kvm_find_cpuid_entry);
1270 
1271 /*
1272  * Intel CPUID semantics treats any query for an out-of-range leaf as if the
1273  * highest basic leaf (i.e. CPUID.0H:EAX) were requested.  AMD CPUID semantics
1274  * returns all zeroes for any undefined leaf, whether or not the leaf is in
1275  * range.  Centaur/VIA follows Intel semantics.
1276  *
1277  * A leaf is considered out-of-range if its function is higher than the maximum
1278  * supported leaf of its associated class or if its associated class does not
1279  * exist.
1280  *
1281  * There are three primary classes to be considered, with their respective
1282  * ranges described as "<base> - <top>[,<base2> - <top2>] inclusive.  A primary
1283  * class exists if a guest CPUID entry for its <base> leaf exists.  For a given
1284  * class, CPUID.<base>.EAX contains the max supported leaf for the class.
1285  *
1286  *  - Basic:      0x00000000 - 0x3fffffff, 0x50000000 - 0x7fffffff
1287  *  - Hypervisor: 0x40000000 - 0x4fffffff
1288  *  - Extended:   0x80000000 - 0xbfffffff
1289  *  - Centaur:    0xc0000000 - 0xcfffffff
1290  *
1291  * The Hypervisor class is further subdivided into sub-classes that each act as
1292  * their own independent class associated with a 0x100 byte range.  E.g. if Qemu
1293  * is advertising support for both HyperV and KVM, the resulting Hypervisor
1294  * CPUID sub-classes are:
1295  *
1296  *  - HyperV:     0x40000000 - 0x400000ff
1297  *  - KVM:        0x40000100 - 0x400001ff
1298  */
1299 static struct kvm_cpuid_entry2 *
1300 get_out_of_range_cpuid_entry(struct kvm_vcpu *vcpu, u32 *fn_ptr, u32 index)
1301 {
1302 	struct kvm_cpuid_entry2 *basic, *class;
1303 	u32 function = *fn_ptr;
1304 
1305 	basic = kvm_find_cpuid_entry(vcpu, 0, 0);
1306 	if (!basic)
1307 		return NULL;
1308 
1309 	if (is_guest_vendor_amd(basic->ebx, basic->ecx, basic->edx) ||
1310 	    is_guest_vendor_hygon(basic->ebx, basic->ecx, basic->edx))
1311 		return NULL;
1312 
1313 	if (function >= 0x40000000 && function <= 0x4fffffff)
1314 		class = kvm_find_cpuid_entry(vcpu, function & 0xffffff00, 0);
1315 	else if (function >= 0xc0000000)
1316 		class = kvm_find_cpuid_entry(vcpu, 0xc0000000, 0);
1317 	else
1318 		class = kvm_find_cpuid_entry(vcpu, function & 0x80000000, 0);
1319 
1320 	if (class && function <= class->eax)
1321 		return NULL;
1322 
1323 	/*
1324 	 * Leaf specific adjustments are also applied when redirecting to the
1325 	 * max basic entry, e.g. if the max basic leaf is 0xb but there is no
1326 	 * entry for CPUID.0xb.index (see below), then the output value for EDX
1327 	 * needs to be pulled from CPUID.0xb.1.
1328 	 */
1329 	*fn_ptr = basic->eax;
1330 
1331 	/*
1332 	 * The class does not exist or the requested function is out of range;
1333 	 * the effective CPUID entry is the max basic leaf.  Note, the index of
1334 	 * the original requested leaf is observed!
1335 	 */
1336 	return kvm_find_cpuid_entry(vcpu, basic->eax, index);
1337 }
1338 
1339 bool kvm_cpuid(struct kvm_vcpu *vcpu, u32 *eax, u32 *ebx,
1340 	       u32 *ecx, u32 *edx, bool exact_only)
1341 {
1342 	u32 orig_function = *eax, function = *eax, index = *ecx;
1343 	struct kvm_cpuid_entry2 *entry;
1344 	bool exact, used_max_basic = false;
1345 
1346 	entry = kvm_find_cpuid_entry(vcpu, function, index);
1347 	exact = !!entry;
1348 
1349 	if (!entry && !exact_only) {
1350 		entry = get_out_of_range_cpuid_entry(vcpu, &function, index);
1351 		used_max_basic = !!entry;
1352 	}
1353 
1354 	if (entry) {
1355 		*eax = entry->eax;
1356 		*ebx = entry->ebx;
1357 		*ecx = entry->ecx;
1358 		*edx = entry->edx;
1359 		if (function == 7 && index == 0) {
1360 			u64 data;
1361 		        if (!__kvm_get_msr(vcpu, MSR_IA32_TSX_CTRL, &data, true) &&
1362 			    (data & TSX_CTRL_CPUID_CLEAR))
1363 				*ebx &= ~(F(RTM) | F(HLE));
1364 		}
1365 	} else {
1366 		*eax = *ebx = *ecx = *edx = 0;
1367 		/*
1368 		 * When leaf 0BH or 1FH is defined, CL is pass-through
1369 		 * and EDX is always the x2APIC ID, even for undefined
1370 		 * subleaves. Index 1 will exist iff the leaf is
1371 		 * implemented, so we pass through CL iff leaf 1
1372 		 * exists. EDX can be copied from any existing index.
1373 		 */
1374 		if (function == 0xb || function == 0x1f) {
1375 			entry = kvm_find_cpuid_entry(vcpu, function, 1);
1376 			if (entry) {
1377 				*ecx = index & 0xff;
1378 				*edx = entry->edx;
1379 			}
1380 		}
1381 	}
1382 	trace_kvm_cpuid(orig_function, index, *eax, *ebx, *ecx, *edx, exact,
1383 			used_max_basic);
1384 	return exact;
1385 }
1386 EXPORT_SYMBOL_GPL(kvm_cpuid);
1387 
1388 int kvm_emulate_cpuid(struct kvm_vcpu *vcpu)
1389 {
1390 	u32 eax, ebx, ecx, edx;
1391 
1392 	if (cpuid_fault_enabled(vcpu) && !kvm_require_cpl(vcpu, 0))
1393 		return 1;
1394 
1395 	eax = kvm_rax_read(vcpu);
1396 	ecx = kvm_rcx_read(vcpu);
1397 	kvm_cpuid(vcpu, &eax, &ebx, &ecx, &edx, false);
1398 	kvm_rax_write(vcpu, eax);
1399 	kvm_rbx_write(vcpu, ebx);
1400 	kvm_rcx_write(vcpu, ecx);
1401 	kvm_rdx_write(vcpu, edx);
1402 	return kvm_skip_emulated_instruction(vcpu);
1403 }
1404 EXPORT_SYMBOL_GPL(kvm_emulate_cpuid);
1405